Logics for Computer Science: Classical and

Non-Classical

Anita Wasilewska

January 9, 2021
2
For my daughter Agatha
Light of my life
4
Contents

1 Introduction: Paradoxes and Puzzels 9

1.1 Mathematical Paradoxes . . . . . . . . . . . . . . . . . . . . . . . 9
1.2 Computer Science Puzzles . . . . . . . . . . . . . . . . . . . . . . 15
1.2.1 Reasoning about knowledge in distributed systems . . . . 16
1.2.2 Reasoning in Artificial Intelligence . . . . . . . . . . . . . 17
1.3 Homework Problems . . . . . . . . . . . . . . . . . . . . . . . . . 20

2 Introduction to Classical Logic 23

2.1 Propositional Language: Motivation and Description . . . . . . . 24
2.2 Propositional Semantics: Motivation and Description . . . . . . . 29
2.3 Examples of Propositional Tautologies . . . . . . . . . . . . . . . 38
2.4 Predicate Language Description and Application to Artificial In-
telligence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.5 Predicate Semantics: Description and Laws of Quantifiers . . . . 57
2.6 Homework Problems . . . . . . . . . . . . . . . . . . . . . . . . . 66

3 Propositional Semantics: Classical and Many Valued 73

3.1 Formal Propositional Languages . . . . . . . . . . . . . . . . . . 73
3.2 Extensional Semantics M . . . . . . . . . . . . . . . . . . . . . . 83
3.3 Classical Semantics . . . . . . . . . . . . . . . . . . . . . . . . . 88
3.3.1 Tautologies: Decidability and Verification Methods . . . . 97
3.3.2 Sets of Formulas: Consistency and Independence . . . . . 106
3.4 Classical Tautologies and Equivalence of Languages . . . . . . . . 109

5
 3.5 Many Valued Semantics: Lukasiewicz, Heyting, Kleene, Bohvar . 117
3.6 M Tautologies, M Consistency, and M Equivalence of Languages 135
3.6.1 M Tautologies Verification Methods . . . . . . . . . . . . 135
3.6.2 M Consistency . . . . . . . . . . . . . . . . . . . . . . . . 137
3.6.3 M Equivalence of Languages . . . . . . . . . . . . . . . . 144
3.7 Homework Problems . . . . . . . . . . . . . . . . . . . . . . . . . 146

4 General Proof Systems: Syntax and Semantics 153

4.1 Syntax . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
4.1.1 Consequence Operation . . . . . . . . . . . . . . . . . . . 166
4.1.2 Syntactic Consistency . . . . . . . . . . . . . . . . . . . . 171
4.2 Semantics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
4.3 Exercises and Examples . . . . . . . . . . . . . . . . . . . . . . . 179
4.4 Homework Problems . . . . . . . . . . . . . . . . . . . . . . . . . 184

5 Hilbert Proof Systems

Completeness of Classical Propositional Logic 187
5.1 Deduction Theorem . . . . . . . . . . . . . . . . . . . . . . . . . 188
5.1.1 Formal Proofs . . . . . . . . . . . . . . . . . . . . . . . . . 200
5.2 Completeness Theorem: Proof One . . . . . . . . . . . . . . . . . 208
5.2.1 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
5.3 Completeness Theorem: Proof Two . . . . . . . . . . . . . . . . 219
5.4 Some Other Axiomatizations . . . . . . . . . . . . . . . . . . . . 230
5.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
5.6 Homework Problems . . . . . . . . . . . . . . . . . . . . . . . . . 237

6 Automated Proof Systems

Completeness of Classical Propositional Logic 243
6.1 Gentzen Style Proof System RS . . . . . . . . . . . . . . . . . . 243
6.2 Search for Proofs and Decomposition Trees . . . . . . . . . . . . 251
6.3 Strong Soundness and Completeness . . . . . . . . . . . . . . . . 262
6.4 Proof Systems RS1 and RS2 . . . . . . . . . . . . . . . . . . . . 266

6
 6.5 Gentzen Sequent Systems GL, G, LK . . . . . . . . . . . . . . . 273
6.5.1 Gentzen Sequent Systems GL and G . . . . . . . . . . . 273
6.6 GL Soundness and Completeness . . . . . . . . . . . . . . . . . . 287
6.7 Original Gentzen Systems LK, LI
Completeness and Hauptzatz Theorems . . . . . . . . . . . . . . 293
6.8 Homework Problems . . . . . . . . . . . . . . . . . . . . . . . . . 312

7 Introduction to Intuitionistic and Modal Logics 315

7.1 Introduction to Intuitionictic Logic . . . . . . . . . . . . . . . . . 315
7.1.1 Philosophical Motivation . . . . . . . . . . . . . . . . . . 316
7.1.2 Algebraic Intuitionistic Semantics and Completeness The-
orem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 318
7.1.3 Algebraic Semantics and Completeness Theorem . . . . . 319
7.1.4 Connection Between Classical and Intuitionistic Tautologies324
7.2 Gentzen Sequent System LI . . . . . . . . . . . . . . . . . . . . . 327
7.2.1 Decomposition Trees in LI . . . . . . . . . . . . . . . . . . 330
7.2.2 Proof Search Examples . . . . . . . . . . . . . . . . . . . 333
7.2.3 Proof Search Heuristic Method . . . . . . . . . . . . . . . 342
7.3 Introduction to Modal S4 and S5 Logics . . . . . . . . . . . . . . 345
7.3.1 Algebraic Semantics for S4 and S5 . . . . . . . . . . . . . 350
7.3.2 S4 and Intuitionistic Logic, S5 and Classical Logic . . . . 354
7.4 Homework Problems . . . . . . . . . . . . . . . . . . . . . . . . . 356

8 Classical Predicate Semantics and Proof Systems 359

8.1 Formal Predicate Languages . . . . . . . . . . . . . . . . . . . . 359
8.2 Classical Semantics . . . . . . . . . . . . . . . . . . . . . . . . . . 373
8.3 Predicate Tautologies . . . . . . . . . . . . . . . . . . . . . . . . 390
8.3.1 Equational Laws of Quantifiers . . . . . . . . . . . . . . . 394
8.4 Hilbert Proof Systems Soundness and Completeness . . . . . . . 401
8.5 Homework Problems . . . . . . . . . . . . . . . . . . . . . . . . . 406

9 Hilbert Proof Systems

7
 Completeness of Classical Predicate Logic 411
9.1 Reduction Predicate Logic to Propositional Logic . . . . . . . . . 412
9.1.1 Henkin Method . . . . . . . . . . . . . . . . . . . . . . . . 418
9.2 Proof of Completeness Theorem . . . . . . . . . . . . . . . . . . . 429
9.3 Deduction Theorem . . . . . . . . . . . . . . . . . . . . . . . . . 440
9.4 Some other Axiomatizations . . . . . . . . . . . . . . . . . . . . . 445
9.5 Homework Problems . . . . . . . . . . . . . . . . . . . . . . . . . 450

10 Predicate Automated Proof Systems

Completeness of Classical Predicate Logic 453
10.1 QRS Proof System . . . . . . . . . . . . . . . . . . . . . . . . . . 454
10.2 QRS Decomposition Trees . . . . . . . . . . . . . . . . . . . . . . 459
10.2.1 Examples of Decomposition Trees . . . . . . . . . . . . . 461
10.3 Proof of QRS Completeness . . . . . . . . . . . . . . . . . . . . . 468
10.4 Skolemization and Clauses . . . . . . . . . . . . . . . . . . . . . 479
10.4.1 Prenex Normal Forms and Skolemization . . . . . . . . . 482
10.4.2 Clausal Form of Formulas . . . . . . . . . . . . . . . . . . 494
10.5 Homework Problems . . . . . . . . . . . . . . . . . . . . . . . . . 498

11 Formal Theories and Gödel Theorems 501

11.1 Formal Theories: Definition and Examples . . . . . . . . . . . . . 502
11.2 PA: Formal Theory of Natural Numbers . . . . . . . . . . . . . . 512
11.3 Consistency, Completeness, Gödel Theorems . . . . . . . . . . . 526
11.3.1 Hilbert’s Conservation and Consistency Programs . . . . 531
11.3.2 Gödel Incompleteness Theorems . . . . . . . . . . . . . . 533
11.4 Proof of the Incompleteness Theorems . . . . . . . . . . . . . . . 537
11.4.1 The Formalized Completeness Theorem . . . . . . . . . . 543
11.5 Homework Problems . . . . . . . . . . . . . . . . . . . . . . . . . 546

8
Chapter 1

Introduction: Paradoxes
and Puzzels

1.1 Mathematical Paradoxes

Until recently, till the end of the 19th century, mathematical theories used to be
built in an intuitive or axiomatic way. In other words, they were based either
intuitive ideas concerning basic notions of the theory - ideas taken from the
reality - or on the properties of these notions expressed in systems of axioms.
The historical development of mathematics has shown that it is not sufficient
to base theories on an intuitive understanding of their notions only. This fact
became especially obvious in set theory. The basic concept of this theory, set, is
certainly taken from reality, for there we come across many examples of various
sets, all of which are finite. But in mathematics it is also necessary to consider
infinite sets, such as the set of all integers, the set of all rational numbers, the
set of all segments, the set of all triangles.
By a set, we mean intuitively, any collection of objects. For example, the set of
all even integers or the set of all students in a class. The objects that make up a
set are called its members (elements). Sets may themselves be members of sets
for example, the set of all sets of integers has sets as its members. Most sets are
not members of themselves; the set of all students, for example, is not a member
of itself, because the set of all students is not a student. However, there may
be sets that do belong to themselves - for example, the set of all sets. However,
a simple reasoning indicates that it is necessary to impose some limitations on
the concept of a set.

Russell, 1902 Consider the set A of all those sets X such that X is not a
member of X. Clearly, by definition, A is a member of A if and only if

9
 A is not a member of A. So, if A is a member of A, the A is also not a
member of A; and if A is not a member of A, then A is a member of A.
In any case, A is a member of A and A is not a member of A.

This paradox arises because the concept of set was not precisely defined and
was too liberally interpreted. Russell noted the self-reference present in his
paradox (and other paradoxes, two of them stated below) and suggested that
every object must have a definite non-negative integer as its type. Then an
expression x is a member of the set y is meaningful if and only if the type of
y is one greater than the type of x. So, according to the theory of types, it is
meaningless to say that a set belongs to itself, there can not be such set A, as
stated in the Russell paradox.
The paradoxes concerning the notion of a set are called logical paradoxes (anti-
nomies). Two of the most known (besides the Russell’s) logical paradoxes are
Cantor and Burali-Forti antinomies. Both were stated at the end of 19th cen-
tury. The Cantor paradox involves the theory of cardinal numbers, Burali-Forti
paradox is the analogue to Cantor’s in the theory of ordinal numbers. They will
make real sense only to those already familiar with both of the theories, but we
will state them here because they do have an intuitive meaning and had very
important consequences.
The cardinal number cardX of a setX intuitively corresponds, in a case of finite
sets, to a number of elements of the set X. cardX is formally defined to be
the set of all sets Y that are equinumerous with X ( i.e., for which there a
one-to-one correspondence between X and Y ). We define cardX ≤ cardY to
mean that X is equinumerous with a subset of Y ; by cardX < cardY we mean
cardX ≤ cardY and cardX 6= cardY . Cantor proved that if P(X) is the set
of all subsets of X, then cardX < cardP(X). The cardinal numbers behave
similarly to natural numbers in many cases, in particular Schröder- Berstein
proved that if cardX ≤ cardY and cardY ≤ cardX, then cardX = cardY .
The ordinal numbers are the numbers assigned to sets in a similar way as cardinal
numbers but they deal with ordered sets.

Cantor, 1899 Let C be the universal set - that is, the set of all sets. Now, P(C)
is a subset of C, so it follows easily that cardP(C) ≤ cardC. On the other
hand, by Cantor theorem, cardC < cardP(C), so also cardC ≤ cardP(C)
and by Schröder- Berstein theorem we have that cardP(C) = cardC, what
contradicts cardC < cardP(C).
Burali-Forti, 1897 Given any ordinal number, there is a still larger ordinal
number. But the ordinal number determined by the set of all ordinal
numbers is the largest ordinal number.

The approach of eliminating logical paradoxes, known as the theory of types,

was systematized and developed by Whitehead and Russell in 1910 - 1913. It

10
is successful, but difficult in practice and has certain other drawbacks as well.
A different criticism of the logical paradoxes is aimed at their assumption that
the notion of a set is defined in such a way that, for every property P (x),
there exists a corresponding set of all objects x that satisfy P (x). If we reject
this assumption, then the logical paradoxes are no longer derivable. Russell’s
Paradox then simply proves that there is no set A of all sets that do not belong
to themselves; the paradoxes of Cantor and Burali-Forti show that there is no
universal set and no set that contains all ordinal numbers.
It became obvious that the paradoxes described above, as well as many similar
ones occurring in intuitive set theory cannot be avoided by referring to intuition
only. The solution looked for was to characterize the intuitive concept of set by
a suitable set of axioms. If we do so, we obtain an axiomatic set theory without
such antinomies.
The first such axiomatic set theory was invented by Zermello in 1908. However,
in 1922 Fraenkel pointed out some shortcomings of Zermello’s axiomatizations
and proposed improvements. The result is called Zermello-Frankel set theory
ZF, or ZFC, where C stands for axiom of choice. Today ZFC is the standard
form axiomatic set theory. ZFC is considered the most common foundation of
mathematics sufficient to include all actual mathematical theories.
A more radical interpretation of the paradoxes has been advocated by Brouwer
and his intuitionist school. They refuse to accept the universality of certain
basic logical laws, such as the law of excluded middle: A or not A. Such a law,
they claim, is true for finite sets, but it is invalid to extend it to all sets. It
means that the intuitionists’ view of the concept of infinite set differs from that
of most mathematicians. Intuitionists reject the idea of infinite set as a closed
whole. They look upon an infinite set as something which is constantly in a
state of formation. It means that, for example, the set of all positive integers
is not looked upon as a closed whole. It is infinite in a sense that to any given
finite set of positive integers it is always possible to add one more positive
integer. The notion of the set of all subsets of the set of all positive integers is
not regarded meaningful. Obviously, intuitionists’ view-point on the meaning
of the basic logical and set-theoretical concepts used in mathematics is different
from that of most mathematicians in their research. The basic difference lies
in the interpretation of the word exists. For example, let P (n) be a statement
in the arithmetic of positive integers. For the mathematicians the sentence the
exists n, such that P(n) is true if it can be deduced (proved) from the axioms
of arithmetic by means of classical logic. If the mathematician proves it, this
does not mean that he is able to indicate a method of construction of a positive
integer n such that P (n) holds. On the contrary, for the intuitionist the sentence
the exists n, such that P(n) is true only if he is able to construct a number n
such that P (n) is true. In the intuitionists’ universe we are justified in asserting
the existence of an object having a certain property only if we know an effective
method for constructing or finding such an object. The paradoxes are, in this
case, not derivable (or even meaningful), but so are many theorems of everyday

11
mathematics, and for this reason, intuitionism has found few converts among
mathematicians. But, because of its constructive flavor, it has found some
applications in computer science and this is the reason to discuss some of it
here. An exact exposition of the basic ideas of intuitionism is outside the range
of this book, but we will study intuitionists logic, which is a sort of reflection of
intuitionists ideas formulated in formalized deductive system.
As we can see, the axiomatic method was the first step leading to greater pre-
cision in the construction of mathematical theories. In intuitive mathematical
theories the borderline between that which is obvious and that which requires
proof is not exact. In axiomatic theories a system of primitive notions is as-
sumed which are characterized by a set of axioms. Other notions can be defined
by means of the primitive notions. All statements which are consequences of
the axioms are called theorems of the theory. All properties of any notion of
the theory which are not expressed in the axioms, require proof.
For some time this degree of exactness in the construction of theories seemed
sufficient. However, it turned out that the assumption of a consistent set of
axioms does not prevent the occurrence of another kind of paradoxes, called
semantic paradoxes.
For instance, let us consider the arithmetic based on the well known system
of axioms due to Peano (to be discussed in chapter 11) and let’s consider the
following simple reasoning.

Berry, 1906 Let A denote the set of all positive integers which can be defined
in the English language by means of a sentence containing at most 1000
letters. The set A is finite since the set of all sentences containing at most
1000 letters is finite. Hence, there exist positive integers which do not
belong to A. The sentence:
n is the least positive integer which cannot be defined by means of a sen-
tence of the English language containing at most 1000 letters
contains less than 1000 letters and defines a positive integer n. Therefore n
belongs to A. On the other hand, n does not belong to A by the definition
of n. This contradicts the first statement.

It is obvious that the reason for this paradox is that in its construction we used
some notions (e.g the English language, letters, sentences) which do not belong
to pure arithmetic. Usually we do not introduce definitions like the above in
mathematics. The paradox resulted entirely from the fact that we did not say
precisely what notions and sentences belong to the arithmetic and what notions
and sentences concern the arithmetic, examined as a fix and closed deductive
system. Intuitively we conceive the arithmetic to be a set of sentences expressing
certain properties of positive integers and of other notions defined by means of
the notion of integer. For example, a sentence: for every integer n, n2 ≤ 0
certainly belongs to arithmetic.

12
On the other hand we can also talk about the arithmetic. That is, assuming
that all sentences in the arithmetic are formulated in the English language, we
can formulate sentences concerning the problem how many integers we defined
in the arithmetic by means of at most 1000 letters. However, such sentences
about the arithmetic do not belong to arithmetic. They belong to another
theory, which examines the arithmetic as a new subject of investigation. This
theory is usually called meta-arithmetic. In particular, the Berry sentence does
not belong to arithmetic; it belongs to meta-arithmetic and the paradox arises
from the luck of distinction between a theory (language) and its meta-theory
(metalanguage).
For a similar reason in well defined theory the following paradoxes can not
appear.

The Liar Paradox (Greek philosopher Eubulides of Miletus, 400 BC)

A man says: I am lying. If he is lying, then what he says is true, and so
he is not lying. If he is not lying, then what he says is not true, and so he
is lying. In any case, he is lying and he is not lying.

Löb, 1955 Let A be any sentence. Let B be a sentence: If this sentence is true,
then A. So, B asserts: If B is true then A. Now consider the following
argument: Assume B is true. Then, by B, since B is true, A is true. This
argument shows that, if B is true, then A. But this is exactly what B
asserts. Hence, B is true. Therefore, by B, since B is true, A is true. Thus
every sentence is true.

In these cases the paradox arises because the concepts of ” I am true”, ” this
sentence is true, ” I am lying”” should not occur in the language (theory). It
belong to a metalanguage (meta-theory).
The Liar Paradox is a corrected version of a following paradox stated in antiquity
by a Cretan philosopher Epimenides, 600 BC.

Cretan ” Paradox” (The Cretan philosopher Epimenides paradox, 600 BC)

Epimenides, a Cretan said: All Cretans are liars. If what he said is true, then,
since Epimenides is a Cretan, it must be false. Hence, what he said is
false. Thus, there must be some Cretan who is not a liar.

Note that the conclusion that there must be some Cretan who is not a liar is not
logically impossible, so we do not have a genuine paradox. However, the fact
that the utterance by Epimenides of the false sentence could imply the existence
of some Cretan who is not a liar is rather unsettling.
It follows from above semantic paradoxes that in order to exclude them from an
axiomatic theory, it is necessary to describe its language precisely, i.e. the set
of sentences of the theory and the set of signs used to build these sentences. In

13
this way we avoid contradictions caused by a collision between the theory and
meta-theory, that is, by including meta-theoretical statements in the theory.
This inclines us to introduce still greater precision in the construction of math-
ematical (and others) theories and leads to the concept of formalized theories,
in which not only the properties of primitive notions are given in an exact way
by means of a set of axioms, but also the language of the theory is precisely
defined. The formalization of the language also gives the following advantage:
it permits us to describe precisely the logical means assumed in the theory, i.e.
to define the process of deduction precisely.
In formalized mathematical theories, e.g. in formalized arithmetic and formal-
ized set theory, the paradoxes as those mentioned above cannot be constructed.
On the other hand, a mathematician (or a computer scientist) following good in-
tuitions in every -day investigations does not reach a contradiction even though
the language of the theory and the logical means he employs are not precisely
described. This is explained by the fact that his investigations can always in
practice be repeated in a corresponding formalized theory. Thus he avoids
practical difficulties of formalized theories, the formalized language of of which
is complicated and very inconvenient in every day practice. Consequently, in
mathematical practice we build theories axiomatically but always in such a way
that they can be easily formalized, i.e. that all the reasonings can be repeated in
a corresponding formalized theory. However, the formalization of the language
and the logical means are necessary if we intend to develop the meta-theory of
a given practical theory because only in this way such important notions as the
existence of a proof of a given statement or the set of all theorems of the theory
can be precisely defined. In practical, non-formalized axiomatic theories those
notions are far from this precision.
Whatever approach one takes to the paradoxes, it is always necessary first to
examine the language of logic and mathematics or given domain of computer
science, to see what symbols may be used, to determine the way ways in which
these symbols are put together to form terms, formulas, sentences, and proofs,
and to find out what can and cannot be proved if certain axioms and rules of
inference are assumed.
This is the basic task of mathematical logic, and, until it is done, there is no
basis of talking about foundations of logic, mathematics or computer science.
This approach is already almost a century old - the first formalized theory was
built by Frege in 1891. The deep and devastating results of Gödel, Tarski,
Church, Rosser, Kleene and many others followed. They created what is called
a modern mathematical logic and have earned for it its status as an independent
branch of mathematics.

14
1.2 Computer Science Puzzles

Logical and semantical paradoxes have lead the mathematicians to the devel-
opment of a modern classical logic as an independent domain of mathematics.
They have also, as we could see, led to the development of the intuitionistic logic
as rival to the classical one. The classical and intuitionistic logic differ on the
interpretation of the meaning of the word exists, but also, as we will see later,
in the interpretation of logical implication, i.e. the truthfulness of the sentences
of the form if A then B is decided differently in both logics.
In 1918, an American philosopher, C.I. Lewis proposed yet another interpreta-
tion of lasting consequences, of the logical implication. In an attempt to avoid,
what some felt, the paradoxes of implication (a false sentence implies any sen-
tence) he created a modal logic. The idea was to distinguish two sorts of truth:
necessary truth and mere possible (contingent) truth. A possibly true sentence
is one which, though true, could be false.
More recently, modal logic has become much-used in computer science as a tool
for analyzing such notions as knowledge, belief, tense.
The logics other than the classical propositional or predicate logic are usually
called non-standard logics. The use of classical logic in computer science is
known, indisputable, and well established.The existence of PROLOG and Logic
Programming as a separate field of computer science is the best example of
it. But the non-standard logics have been imported into a great many areas
of computer science and, in particular into the research about the specification
and verification of programs, the analysis of behavior of distributed systems and
into almost all areas of artificial intelligence. Even in Logic Programming, once
we start to talk about logic programming programs we run immediately into
some non-standard logics.
Modal logic, for example, has been employed in form of Dynamic logic (Harel
1979) to facilitate the statement and proof of properties of programs.
Temporal Logics were created for the specification and verification of concurrent
programs Harel, Parikh, 1979, 1983), for a specification of hardware circuits
Halpern, Manna and Maszkowski, (1983), to specify and clarify the concept of
causation and its role in commonsense reasoning (Shoham, 1988).
Intuitionistic logic, in the form of Martin-Löf’s theory of types (1982), provides
a complete theory of the process of program specification, construction, and
verification. A similar theme has been developed by Constable (1971) and
Beeson (1983).
The great strength of dynamic and temporal logics relates to their expressive
power. In such systems it is possible to express properties of programs in an
elegant and natural way. This is in large part due to enriched language of such
logics over the classical predicate calculus. In the case of intuitionistic logic

15
the motivation for their employment, as it was mentioned before, is different.
The proponents of intuitionistic logic and mathematics claim that constructive
mathematics is, generally, a more appropriate framework for computer science
than classical logic and mathematics.
Fuzzy logic, Many valued logics were created and developed to reasoning with
incomplete information. Most expert systems are forced to take decisions when
not all the facts pertaining to decision are available. In such context it is natural
to employ logics which, unlike classical logic, are suited to reasoning with such
incomplete information.
The development of different logics and the applications of logic to different
areas of computer science or even artificial intelligence only is beyond the scope
of our investigations.
We present some of the most known motivations (computer science puzzles),
which played a similar role in the development of the reasoning about knowl-
edge in distributed systems and artificial intelligence, as logical and semantical
paradoxes played in the development of logic and foundations of mathematics.

1.2.1 Reasoning about knowledge in distributed systems

The major complexities in designing, understanding and reasoning about dis-

tributed systems arise from the uncertainties inherent in the system, particularly
with regard to message delivery and possible faulty or unexpected behavior of
processors. A protocol must be designed (and proved) to function properly even
if it is possible for messages to be lost, for messages to arrive out of order, or
for some processor to fail. It is difficult (probably impossible) for one node to
know everything about the rest of the network. Yet we are often interested in
situations in which everyone in a group (every processor in the network) knows
a certain fact. But even the state of knowledge in which everyone knows what
everyone knows does not suffice for a number of applications. In some cases
we also need to consider the state in which simultaneously everyone knows a
fact F, everyone knows that everyone knows F, everyone knows that everyone
knows that everyone knows F, and so on. In this case it is said that the group
has common knowledge of F. The relationship between the common knowl-
edge, simultaneous agreement and coordinate action is nicely put together in
the coordinated attack problem, from the distributed system folklore.

Grey, 1978. Halpern, Moses, 1984 Two divisions of an army are camped
on two hilltops overlooking a common valley. In the valley awaits the
enemy. It is clear that if both divisions attack the enemy simultaneously
they will win the battle, whereas if only one division attacks it will be
defeated. The divisions do not initially have plans for launching an attack
on the enemy, and the commanding general of the first division wishes to
coordinate a simultaneous attack (at some time the next day). Neither

16
 general will decide to attack unless he is sure that the other will attack
with him. The generals can only communicate by means of a messenger.
Normally, it takes a messenger one hour to get from one encampment
to the other. However, it is possible that he will get lost in the dark or,
worst yet, be captured by the enemy. Fortunately, on this particular night,
everything goes smoothly. How long will it take them to coordinate an
attack?

Suppose the messenger sent by General A makes it to General B with a message

saying Attack at dawn. Will B attack? No, since A does not know B got the
message, and thus may not attack. So B sends the messenger back with an
acknowledgment. Suppose the messenger makes it. Will A attack? No, because
now A is worried that B does not know A got the message, so that B thinks
A may think that B did not get the original message, and thus not attack.
So A sends the messenger back with an acknowledgment. But of course, this
is not enough either. It is not difficult to be convinced that no amount of
acknowledgments sent back and forth will ever guarantee agreement. Even in
a case that the messenger succeeds in delivering the message every time. All
that is required in this (informal) reasoning is the possibility that the messenger
doesn’t succeed.
This rather convoluted reasoning was expressed formally by Halpern and Moses
in 1985 in terms of a propositional modal logic with m agents. They proved
this logic to be essentially a multi-agent version of the modal logic S5 which we
present in chapter 7.
They also showed that not only is common knowledge (formally defined!) not
attainable in systems where communication is not guaranteed, it is also not at-
tainable in systems where communication is guaranteed, as long as there is some
uncertainty in message delivery time. Thus, in practical distributed systems ,
common knowledge is not attainable. This holds for systems of communicating
humans as well as processors. What is going on here? After all, we often do
reach agreement! It was shown that common knowledge (as formally defined)
is attainable in such models of reality where we assume, for example, events
can be guaranteed to happen simultaneously. It turns also out that even we
can’t always make this assumption in practice, there are some variants of the
definition of common knowledge that are attainable under more reasonable as-
sumptions, and these variants are indistinguishable in certain cases from the
”true” common knowledge, as originally defined.. So, finally, we can prove that
in fact we often do reach agreement!

1.2.2 Reasoning in Artificial Intelligence

A key property of intelligence, one may agree, is flexibility. This flexibility is

intimately connected with the defeasible nature of commonsense inference; we
are all capable of drawing conclusions, acting on them, and then retracting them

17
if necessary in the face of new evidence. If our computer programs are to act
intelligently, they will need to be similarly flexible.
A large portion of the work in artificial intelligence (AI) on reasoning or deduc-
tion involves the development of formal systems that describe this process.
The most usual example of such a flexible inference is the following flying birds
example.

Reiter, 1987 Consider a statement Birds fly. Tweety, we are told, is a bird.
From this, and the fact that birds fly, we conclude that Tweety can fly.
This conclusion, however is defeasible: Tweety may be an ostrich, a pen-
guin, a bird with a broken wing, or a bird whose feet have been set in
concrete.

The inference here is non-monotonic: on learning a new fact (that Tweety has
a broken wing), you are forced to retract your conclusion that he could fly. This
original conclusion didn’t follow logically (in a sense if classical logic) from the
facts that birds typically fly and that Tweety is a bird; it followed from these
facts together with the assumption that Tweety is a typical bird. When we learn
more about Tweety, we may discover that this assumption is unjustified.
It means, by a non-monotonic reasoning (logics) we understand reasonings (log-
ics) in which the introduction of a new information (facts) can invalidate old
theorems.
The inference described here is also called a default reasoning.
It means, by default reasoning we mean the drawing of plausible inferences from
less-then-conclusive evidence in the absence of information to the contrary.
Consider now the following example.

Moore, 1983 Consider my reason for believing that I do not have an older
brother. It is surely not that one of my parents once casually remarked,
You know, you don’t have any older brothers, nor have I pieced it together
by carefully sifting other evidence. I simply believe that if I did have an
older brother I would know about it; therefore, since I don’t know of any
older brothers of mine, I must not have any.

This type of reasoning is not a form of default reasoning at all; it rather seems
to be more like reasoning about one’s own knowledge or belief. Hence it is
called an auto-epistemic reasoning, i.e. the reasoning about one’s own beliefs or
knowledge.
The auto-epistemic reasoning is intended to model the reasoning of an ideally
rational agent reflecting upon his beliefs or knowledge. Logics which describe it
are called auto-epistemic logics.

18
In addition to application to the understanding of common-sense reasoning, non-
monotonic reasoning (non-monotonic logics) has been shown to be important in
other areas. There are applications to logic programming, to planning and rea-
soning about action, and to automated diagnosis. As the formal work matures,
increasing effort is being devoted to applying the improved understanding to
the solution of practical problems.
We end this introduction by McCarthy discussion of a much used in AI puzzle
Missionaries and Cannibals, as a proof of a need of another ”stretch ” from
classical logic.

McCarthy, 1985 Consider the Missionaries and Cannibals puzzle.

Three missionaries and three cannibals come to the river. A rowboat that
seats two is available. If the cannibals ever outnumber the mission-
aries on either bank of the river, the missionaries will be eaten. How
shall they cross the river?

Obviously the puzzler is expected to devise a strategy of rowing the boat

back and forth that gets them all across and avoids the disaster.
Ammarel considered several representations of the problem and discussed
criteria whereby the following representation is preferred for purposes of
AI, because it leads to the smallest state space that must be explored to
find the solution. A state is a triple comprising the number of missionaries,
cannibals and boats on the starting bank of the river. The initial state is
331, the desired state is 000, and one solution is given by the sequence:
331, 220, 321, 300,311, 110, 221, 020, 031, 010, 021, 000.
We are not presently concerned with the heuristic of the problem but
rather with the correctness of the reasoning that goes from the English
statement of the problem to Amerel’s state space representation. A gen-
erally intelligent computer program should be able to carry out this rea-
soning. Of course, there are the well known difficulties in making com-
puters understand English, but suppose the English sentences describing
the problem have already been rather directly translated into first order
logic. The correctness of Amarel’s representation is not an ordinary logical
consequence of these sentences for two further reasons. First, nothing has
been said about the properties of boats or even the fact that rowing across
the river doesn’t change the number of missionaries or cannibals or the
capacity of the boat. Indeed it hasn’t been stated that situations change
as a result of action. These facts follow from common sense knowledge,
so let us imagine that common sense knowledge, or at least the relevant
part of it, is also expressed in first order logic.
The second reason we can’t deduce the propriety of Amarel’s representa-
tion is deeper. Imagine giving someone a problem, and after he puzzles
for a while, he suggests going upstream half a mile and crossing a bridge.

19
 What bridge? you say. No bridge is mentioned in the statement of the
problem. And this dunce replies, Well, they don’t say there isn’t a bridge.
You look at the English and even at the translation of the English into
first order logic, and you must admit that they don’t say there is no bridge.
So you modify the problem to exclude the bridges and pose it again, and
the dunce proposes a helicopter, and after you exclude that, he proposes
a winged horse or that the others hang onto the outside of the boat while
two row.
You now see that while a dunce, he is an inventive dunce. Despairing of
getting him to accept the problem in the proper puzzler’s spirit, you tel
him the solution. To your further annoyance, he attacks your solution on
the grounds that the boat might have a leak or lack oars. After you rectify
that omission from the statement of the problem, he suggests that a sea
monster may swim up the river and may swallow the boat. Again you are
frustrated, and you look for a mode of reasoning that will settle his hash
once and for all.

McCarthy proposes circumscription as a technique for solving his puzzle. He

argues that it is a part of common knowledge that a boat can be used to cross the
river unless there is something with it or something else prevents using it, and if
our facts do not require that there be something that prevents crossing the river,
circumscription will generate the conjecture that there isn’t. Among the various
competing approaches to model the common sense reasoning circumscription
appears to have the most advocates.
One of the serious difficulties is that the circumscription axiom itself involves a
quantification over predicates, and there is therefore a sentence in second-order
logic. Little is known about automated deduction using second-order sentences,
but on the other hand Lifschits has shown in 1987 that in some special cases
the circumscription is equivalent to a first order sentence. In this way we can
go back, in those cases, to our secure and well known classical logic.

1.3 Homework Problems

1. Write definition of logical and semantical paradox.
2. Give an example of a logical paradox.
3. Give an example of a semantical paradox.
4. Describe a difference between logical and semantical paradoxes.
5. Describe a role of paradoxes in the development of classical logic and
foundations of mathematics.
6. Write a definition os a non-standard logic.

20
 7. Give an example of some non-standard logics.
8. Describe a difference between classical and intuitionistic logic.
9. Give two examples of Computer Science Puzzles.
10. What a default reasoning is? Give an example.

11. What a non - monotonic reasoning is? Give an example.

12. What an auto-epistemic reasoning is? Give an example.

21
22
Chapter 2

Introduction to Classical
Logic

.
Logic builds symbolic models of our world. It builds them in such a way as to
be able to describe formally the ways we reason in and about it. It also poses
questions about correctness of such models and develops tools to answer them.
Classical Logic was created to describe the reasoning principles of mathematics
and hence reflects the ”black” and ”white” qualities of mathematics; we expect
from mathematical theorems to be always either true or false and the reasonings
leading to them should guarantee this without any ambiguity. It hence admits
only two logical values and is sometimes called a two-valued logic.
The models we build are based on a principle that the language in which we
reason uses sentences. These sentences are built up from basic assertions about
the world using special words or phrases like ”not”, ”not true” ”and”, ”or”, ”
implies”, ”if ..... then”, ”from the fact that .... we can deduce”, ” if and only
if”, ”equivalent”, ”every”, ”for all”, ”any”, ”some”,” exists”. Basically, it is the
behavior of these words we want to study. Most of these words and phrases
have accepted intuitive meanings and we want our models to formalize these
meanings. To do so we first define a notion of a symbolic language and then
define a formal meaning of its symbols, called semantics.
We use symbols: ¬, for ”not”, ”not true”, ∩ for ”and”, ∪ for ”or”, ⇒ for ”
implies” , ”if ..... then”, ”from the fact that... we can deduce”, and a symbol
⇔ for ” if and only if”, ”equivalent”. We call these symbols propositional con-
nectives. There are other symbols for propositional connectives and there are
other propositional connectives as well that we will introduce later.
We use symbols: a, b, c, p, r, q, . . . , with indices, if necessary to represent the

23
basic assertions, called propositions. Hence we call the symbols a, b, c, p, r, q, . . .
propositional variables.
We use symbols: ∀ for ”every”, ”any”, and ∃ for ”some” ,” exists”. The symbols
∀, ∃ are called quantifiers.
Restricting our attention to the role of propositional connectives yields to what is
called propositional logic with the a propositional language and a propositional
semantics as its basic components. This is a quite simple model to justify,
describe and develop and we will devote first few chapters to it. We do it both
for its own sake, and because it provides a good background for developing and
understanding more difficult logics to follow.
Consideration and study of the role of propositional connectives and quantifiers
leads to what is called a predicate logic with its predicate language and semantics.
This is a much more complicated model and we will develop and study it in full
formality in chapters following the introduction and examination of the formal
propositional logic model.
In this chapter we provide motivation for and description of both propositional
and predicate languages, and discuss their semantics.

2.1 Propositional Language: Motivation and De-

scription
The propositional language is a quite simple symbolic language into which we
can translate (represent) natural language sentences. For example, let’s consider
a natural language sentence ” If 2+2 = 5, then 2+2 = 4”. To translate it into the
propositional language we replace ”2 + 2 = 5” by a propositional variable, let’s
say a, and ”2 + 2 = 4” by a propositional variable b and we write a connective
⇒ for ”if ..... then”. As a result we obtain a propositional language formula
(a ⇒ b). A sentence ”If 2 + 2 6= 4 and 2 + 2 = 5, then 2 + 2 = 4” translates
into a formula ((¬b ∩ a) ⇒ b). A sentence ” fact that it is not true that at the
same time 2 + 2 = 4 and 2 + 2 = 5 implies that 2 + 2 = 4” translates into a
propositional formula (¬(b ∩ a) ⇒ b).
A formal description of symbols and the definition of the set of formulas is
called a syntax of a symbolic language. We use the word syntax to stress that
the formulas carry neither formal meaning nor a logical value. We assign the
meaning and logical value to syntactically defined formulas in a separate step.
This next, separate step is called a semantics. A given symbolic language can
have different semantics and different semantics can define different logics.
We first describe the syntax of the propositional language. The syntax of the
predicate language is much more complex and will be defined later.
The smallest ”building blocks” of a propositional language are propositional

24
variables that represent the the basic assertions called propositions. Histori-
cally, we define propositions as basic, declarative sentences (assertions) that can
always be evaluated as true or false. For example, a statement: ” 2 + 2 = 4”
is a proposition as we assume that it is a well known and agreed upon truth.
A statement: ”2 + 2 = 5” is also a classical proposition (false). A statement:
2 + n = 5 according to the historical definition is not a proposition; it might
be true for some n, for example n=3, false for other n, for example n= 2, and
moreover, we don’t know what n is. Sentences of this kind are called proposi-
tional functions. We treat propositional functions within propositional model
as propositions and represent them by the propositional variables.
Similar examples can be found in natural language rather then in mathematical
language. For example we tend to accept a statement: ”The earth circulates the
sun” as a proposition while a statement: ”Ann is pretty”, even if we accept it
as a proposition by assuming that is always has exactly one logical value, could
also be treated as ambiguous; Ann may be found pretty by some people and
not pretty by others. If we try to improve the situation by saying for example:
”Ann seems to be pretty”, ” I am sure Ann is pretty” or even ”I know that Ann
is pretty” the ambiguity increases rather then decreases.
To deal with these and other ambiguities many non-classical logics were and
are being invented and examined by philosophers, computer scientists, and even
by mathematicians. We will present and study some of them later. Nevertheless
we accept all these and similar statements within classical propositional model
as propositions and represent them by the propositional variables.

Observe that one can think about a natural language as a set W of all words
and sentences based on a given alphabet A. This leads to a simple, abstract
model of a natural language NL as a pair

N L = (A, W).

Some natural languages share the same alphabet, some have different alphabets.
All of them face serious problems with a proper recognition and definitions of
accepted words and complex sentences. We do not want the symbolic languages
to share the same difficulties. We define their components precisely and in such
a way that their recognition and correctness will be easily decided. In order
to distinguish them from natural languages we call their words and sentences
formulas and denote the set of all formulas by F. We call a pair

SL = (A, F). (2.1)

a symbolic language.
We distinguish two categories of symbolic languages: propositional and predi-
cate. We first define the propositional language. The definition of the predicate
language, with its much more complicated structure will follow.

25
Definition 2.1
By a propositional language L we understand a pair
L = (A, F), (2.2)
where A is called propositional alphabet, and F is called a set of all well formed
propositional formulas of L.

Components the language L are defined as follows.

1. Alphabet A
The alphabet A consists of a countably infinite set VAR of propositional vari-
ables, a finite set of propositional connectives, and a set of two parenthesis.
We denote the propositional variables by letters a, b, c, p, q, r, ......., with
indices if necessary. It means that we can also use a1 , a2 , ..., b1 , b2 , ... etc... as
symbols for propositional variables.
Propositional connectives are: ¬, ∩, ∪, ⇒, and ⇔. The connectives have well
established names. We use names negation, conjunction, disjunction, implica-
tion and equivalence or biconditional for the connectives ¬, ∩, ∪, ⇒, and ⇔,
respectively. Parenthesis are (, ).
2. Set F of formulas
Formulas are expressions build by means of elements of the alphabet A. We
denote formulas by capital letters A, B, C, ....., with indices, if necessary.
The set F of all formulas of L is defined recursively as follows.
1. Base step: all propositional variables are formulas. They are called atomic
formulas.
2. Recursive step: for any already defined formulas A, B, the expression:
¬A, (A ∩ B), (A ∪ B), (A ⇒ B), (A ⇔ B) are also formulas.
3. Only those expressions are formulas that are determined to be so by means
of conditions 1. and 2.
We often say that the set F is the set of all well-formed formulas (wff ) to
stress exactness of the definition.
By the definition, any propositional variable is a formula. Let’s take, for example
two variables a and b. They are atomic formulas.
By the recursive step we get that
(a ∩ b), (a ∪ b), (a ⇒ b), (a ⇔ b), ¬a, ¬b
are formulas. Recursive step applied again produces for example formulas
¬(a ∩ b), ((a ⇔ b) ∪ ¬b), ¬¬a, ¬¬(a ∩ b).

26
These are not all formulas we can obtain in the second recursive step. Moreover,
as the recursive process continue we obtain a countably infinite set of all non-
atomic formulas.
Remark that we put parenthesis within the formulas in a way to avoid ambigu-
ity. The expression a ∩ b ∪ a is ambiguous. We don’t know whether it represents
a formula (a ∩ b) ∪ a, or a formula a ∩ (b ∪ a).

Exercise 2.1
Consider a following set
S = {¬a ⇒ (a ∪ b), ((¬a) ⇒ (a ∪ b)), ¬(a ⇒ (a ∪ b)), ¬(a → a)}.

1. Determine which of the elements of S are, and which are not well formed
formulas (wff ) of L = (A, F).
2. For any A 6∈ F re-write it as a correct formula and write in the natural
language what it says.

Solution
The formula ¬a ⇒ (a ∪ b) is not a well formed formula. A correct formula is
(¬a ⇒ (a ∪ b)). The corrected formula says: ”If a is not true , then we have a
or b ”. Another correct formula is ¬(a ⇒ (a ∪ b)). This corrected formula says:
”It is not true that a implies a or b ”.
The formula ((¬a) ⇒ (a ∪ b)) is not correct; (¬a) 6∈ F. The correct formula is
(¬ ⇒ (a∪b)). The formula ¬(a ⇒ (a∪b)) is correct. The formula ¬(a → a) 6∈ F
as the connective → does not belong to the language L. It is a correct formula of
another propositional language; the one that uses a symbol → for implication.

Exercise 2.2
Given a sentence S
”If a natural number a is divisible by 3, then from the fact that a in not divisible
by three we can deduce that a is divisible by 5.”
Write a formula corresponding to the sentence S.

Solution
First we write our sentence in a more ”logical way” as follows:
” If a natural number a is divisible by 3, then (if not(a is divisible by three) then
a is divisible by 5). We denote the sentence: ”a natural number a is divisible
by 3 ” by a, and the sentence ”a is divisible by 5” by b, and we rewrite our
sentence as: ”If a, then (if not a, then b)”.
We replace expressions If ... then and not by symbols ⇒ and ¬, respectively
and we follow the definition of the set of formulas to obtain a formula

27
 (a ⇒ (¬a ⇒ b))

which corresponds to our natural language sentence S.

Observe that for a given logical sentence there is only one schema of a logical
formula corresponding to it. One can replace a by d and b by c and get a formula
(d ⇒ (¬d ⇒ c)), or we get a formula (b ⇒ (¬b ⇒ a)) by replacing a by b and b by
a. We can, in fact, construct as many of those formulas as we wish, but all those
formulas will have the same form as the formula (a ⇒ (¬a ⇒ b)). They will differ
only on a choice of names for the propositional variables assigned corresponding
to logical sentences. The same happens, when we want to do the ”inverse”
transformation from a given formula A to a logical sentence corresponding to
it. There may be as many of them as we can invent, but they all will be built
in the same way; the way described by the formula A.

Exercise 2.3
Write following natural language statement:
”One likes to play bridge or from the fact that the weather is good we conclude
the following: one does not like to play bridge or one likes swimming.”
as a formula of L = (A, F).

Solution
First we identify the needed components of the alphabet A as follows.
Propositional variables: a, b, c.
a denotes statement: One likes to play bridge, b denotes a statement: the weather
is good, c denotes a statement: one likes swimming.
Connectives: ∪, ⇒, ∪.
Then we write the formula of L as (a ∪ (b ⇒ (¬a ∪ c))).

Exercise 2.4
Given a formula (a ∩ (¬a ∪ b)).
Write 2 natural language sentences which correspond to this formula.

Solution
Let propositional variables a, b denote sentences 2+2 = 4 and 2 > 1, respectively.
In this case the corresponding sentence is:
2 + 2 = 4 and we have that 2 + 2 6= 4 or 2 > 1.
If we assume that the propositional variables a, b denote sentences 2 > 1 and
2 + 2 = 4, respectively, then the corresponding natural language statement is:

28
2 > 1 and we have that 2 6> 1 or 2 + 2 = 4.

Symbols for Connectives

The symbols for connectives used in book are not the only used in mathematical,
logical, or computer science literature.
Other symbols employed for these most important propositional connectives are
listed in the table below.

Negation Disjunction Conjunction Implication Equivalence

−A (A ∪ B) (A ∩ B) (A ⇒ B) (A ⇔ B)
NA DAB CAB IAB EAB
A (A ∨ B) (A & B) (A → B) (A ↔ B)
∼A (A ∨ B) (A · B) (A ⊃ B) (A ≡ B)
A0 (A + B) (A · B) (A → B) (A ≡ B)

The first of these systems of notation is the closest to ours and is drawn mainly
from the algebra of sets and lattice theory. The second comes from the Polish
logician J. Lukasiewicz. In this notation the binary connectives precede the
formulas and are not inserted between them; this enables us to dispense with
parenthesis; Lukasiewicz’s notation is usually called the Polish notation and it
is a parenthesis-free notation. The third was used by D. Hilbert. The fourth
comes from Peano and Russell, while the fifth goes back to Schröder and Pierce.

2.2 Propositional Semantics: Motivation and De-

scription

We present here definitions of propositional connectives in terms of two logical

values true or false and discuss their motivations.
The resulting definitions are called a semantics for the classical propositional
connectives. As we consider only two logical values, the semantics is also called
2 valued semantics. The semantics presented here is fairly informal. The formal
definition of classical propositional semantics will be presented in chapter 4.
Classical Connectives
Our language L contains five connectives called conjunction, disjunction, impli-
cation, equivalence, and negation. We divide the connectives into two groups:
one and two argument connectives. Negation is the one argument connective.
Conjunction, disjunction, implication, equivalence are two argument connec-

29
tives. We define their semantics, i.e. their definitions in terms of two logical
values, and give a motivation justifying the definitions as follows.
We denote a statement A is false by A = F , what stands for the logical value
of a formula A is F. We denote a statement A is true by A = T , what stands
for the logical value of a formula A is T.
Negation motivation and definition.
In accordance with the intuition, the negation of a true formula is a false formula,
and the negation of a false formula is a true formula. Moreover, the logical value
of ¬A depends on the logical values of A in a way which can be express in the
form of the following table.
Negation Table

A ¬A
T F (2.3)
F T

Conjuncion motivation and definition.

In accordance with intuition, a conjunction (A ∩ B) is a true formula if both of
its factors are true formulas. If one of the factors, or both, are false formulas,
then the conjunction is a false formula.
The logical value of a conjunction depends on the logical values of its factors in
a way which can be express in the form of the following table.
Conjunction Table

A B (A ∩ B)
T T T
T F F (2.4)
F T F
F F F

Disjunction motivation and definition.

In everyday language the word or is used in two different senses. In the first, a
statement of the form A or B is accepted as true if at least one of the statements
A and B is true; in the other, the compound statement A or B is accepted as true
if one of the statements A and B is true, and the other is false. In mathematics
the word or is used in the former sense.
Hence, we adopt the convention that a disjunction (A∪B) is true if at least one
of the formulas A and B is true. This convention is called a classical semantics
for the disjunction and is expressed in the following table.

30
Disjunction Table
A B (A ∪ B)
T T T
T F T (2.5)
F T T
F F F
As in the case of the other connectives, the logical value of a disjunction depends
only on the logical values of its factors.
Implication motivation and definition.
The symbol ⇒ is used instead of the statements of the form if A, then B, A
implies B, and is called an implication connective. The formula (A ⇒ B) and
is called an implication and A is called its antecedent, B is called its consequent.
The semantics of the implication needs some discussion. In everyday language
the implication statement if A, then B is interpreted to mean that B can be in-
ferred from A. This interpretation differs from that given to it in mathematics,
and hence in classical semantics. The following example explains the meaning
of the statement if A, then B as understood in mathematics. It hence justifies
our semantics for the implication.

Consider the following arithmetical theorem:

For every natural number n,

if 6 DIV IDES n, then 3 DIV IDES n. (2.6)

The above implication (2.6) is true for any natural number, hence, in par-
ticular, for 2,3,6.
Thus the following propositions are true:

If 6 DIV IDES 2, then 3 DIV IDES 2. (2.7)

If 6 DIV IDES 3, then 3 DIV IDES 3. (2.8)
If 6 DIV IDES 6, then 3 DIV IDES 6. (2.9)

It follows from (2.7) that an implication (A ⇒ B) in which both the antecedent

A and the consequent B are false statements is interpreted as a true statement.
It follows from (2.8) that an implication (A ⇒ B) in which false antecedent A
and true consequent B is interpreted as a true statement.
Finally, it follows from (2.9) that an implication (A ⇒ B) in which both the
antecedent A and the consequent B are true statements is interpreted as a
true statement.

31
Thus one case remains to be examined, namely that in which the antecedent of
an implication is a true statement, and the consequent is a false statement.
For example consider the statement:
If 6 DIVIDES 12, then 6 DIVIDES 5.
In accordance with arithmetic of natural numbers, this statement is interpreted
as false.
The above examples justifies adopting the following semantics for the impli-
cation ⇒. An implication (A ⇒ B) is interpreted to be a false statement if
and only if its antecedent A is a true statement and its consequent is a false
statement. In the remaining cases such an implication is interpreted as a true
statement.
We expressed it in a form of the following table.
Implication Table
A B (A ⇒ B)
T T T
T F F (2.10)
F T T
F F T

Equivalence motivation and definition.

An equivalence (A ⇔ B) is, in accordance with intuition, interpreted as true if
both formulas A and B have the same logical value, that is, are either both true
or both false. This is expressed in the following table.
Equivalence Table
A B (A ⇔ B)
T T T
T F F (2.11)
F T F
F F T
We summarize the tables for propositional connectives in the following one table.
We call it a truth table definition of propositional; connectives and hence we
call the semantics defined here a truth tables semantics.

A B ¬A (A ∩ B) (A ∪ B) (A ⇒ B) (A ⇔ B)
T T F T T T T
T F F F T F F (2.12)
F T T F T T F
F F T F F T T

The table (2.12) indicates that the logical value of of propositional connectives
depends only on the logical values of its factors; i.e. it is independent of the

32
formulas A, B. We write the table in a ”formula in depended” form as a set of
the following equations.

¬T = F, ¬F = T ;

(T ∩ T ) = T, (T ∩ F ) = F, (F ∩ T ) = F, (F ∩ F ) = F ;

(T ∪ T ) = T, (T ∪ F ) = T, (F ∪ T ) = T, (F ∪ F ) = F ; (2.13)

(T ⇒ T ) = T, (T ⇒ F ) = F, (F ⇒ T ) = T, (F ⇒ F ) = T ;

(T ⇔ T ) = T, (T ⇔ F ) = F, (F ⇔ T ) = F, (T ⇔ T ) = T.

We use the above set of equations (2.13) to evaluate logical values of formulas.

Example 2.1
Given a formula (A ⇒ (¬A ∩ B)), such that logical values of its basic compo-
nents, i.e. the propositional formulas A, B are: A=T, and B=F. We calculate
the logical value of the formula (A ⇒ (¬A ∩ B)) by substituting the logical values
for the formulas A, B and applying the equations (2.13) as follows.

(T ⇒ (¬T ∩ F )) = (T ⇒ (F ∩ F )) = (T ⇒ F ) = F.

Exercise 2.5
Given a formula A: (((a ∪ b) ∩ ¬c) ⇒ a). Evaluate the logical value of A for the
following sets of logical values of its basic components, i.e. for the propositional
variables a, b: 1. a=T, b=F, c=F, and 2. a=F, b=T, c=T.

Solution
1. Let a=T, b=F, c=F. We evaluate the logical value of A as follows.

(((T ∪ F ) ∩ ¬F ) ⇒ T ) = ((T ∩ ¬F ) ⇒ T ) = ((T ∩ T ) ⇒ T ) = (T ⇒ T ) = T.

2. Let a=F, b=T, c=T. We evaluate the logical value of A as follows.

(((F ∪ T ) ∩ ¬T ) ⇒ F ) = ((T ∩ ¬T ) ⇒ T ) = ((T ∩ F ) ⇒ T ) = (F ⇒ T ) = T.

Extensional Connectives

We observe that our connectives are such that the logical value of a given formula
build by means of its connectives depends only of logical values of its factors.
Connectives with this property are called extensional. We hence adopt the
following definition.

33
Definition 2.2
We call a propositional connective extensional if the logical value of a given
formula build by means of this connective depends only of logical values of its
factors.

Fact 2.1
All connectives ¬, ∪, ∩, ⇒, and ⇔ are extensional.

In everyday language there are expressions which are propositional connectives

but are not extensional. They do not play any role in mathematics and so they
are not discussed in classical logic.

Other Extensional Connectives

The propositional classical connectives ∩, ∪, ⇒, ⇔, ¬ are not the only exten-

sional connectives. We define here all possible unary and binary two valued
extensional connectives.
An extensional unary connective 5 enables us to form from any formula A, a
new formula 5A, whose logical value is defined in terms of the logical value of
A only, i.e. by means of a table of a type (2.3).
Thus there as many unary connectives as there are functions f from the set
{T, F } to the set {T, F }, that is 22 = 4.

All Unary Connectives

A 51 A 52 A ¬A 54 A
T F T F T (2.14)
F F F T T

An extensional binary connective ◦ permits us to form, of any two formulas A

and B, a new formula (A ◦ B), whose logical value is defined from the logical
values A and B only, i.e. by means of a table similar to (2.4), (2.5), (2.48),
(2.11).
So, there are as many binary connectives as many functions f from a set
{T, F } × {T, F } (four elements) to a set {T, F } (two elements) that is, 24 = 16.

34
All Binary Connectives
A B (A◦1 B) (A ∩ B) (A◦3 B) (A◦4 B)
T T F T F F
T F F F T F
F T F F F T
F F F F F F
A B (A ↓ B) (A◦6 B) (A◦7 B) (A ↔ B)
T T F T T T
T F F T F F
F T F F T F
F F T F F T
(2.15)
A B (A◦9 B) (A◦10 B) (A◦11 B) (A ∪ B)
T T F F F T
T F T T F T
F T T F T T
F F F T T F
A B (A◦13 B) (A ⇒ B) (A ↑ B) (A◦16 B)
T T T T F T
T F T F T T
F T F T T T
F F T T T T

Functional Dependency

It can be proved that all propositional connectives, as defined by tables (2.14)

and (2.15), i.e. whether unary or binary, can be defined in terms of disjunction
and negation.
This property of defining a set of connectives in terms of its proper subset is
called a functional dependency of connectives.
There are also two other binary connectives which suffice, each of them sep-
arately, to define all two valued connectives, whether unary or binary. These
connectives play a special role and are denoted in our table (2.15) by ↓ and ↑,
respectively.
The connective ↑ was discovered in 1913 by H.M. Sheffer, who called it alterna-
tive negation. Now it is often called simply as Sheffer’s connective. The formula
(A ↑ B) is read: not both A and B.
The connective ↓ was discovered by J. Lukasiewicz and named joint negation.
The formula (A ↓ B) is read: neither A nor B.
We define formally and examine the functional dependency of connectives in
Chapter 3. We state here some important facts to be proved in Chapter 3.

Fact 2.2 All two-valued propositional connectives and in particular our connec-

35
tives ¬, ∪, ∩, ⇒, and ⇔ are a functionally dependent.

In particular, we prove the following.

Fact 2.3
The alternative negation connective ↑, and the joint negation. ↓ suffice, each
of them separately, to define all propositional connectives , whether unary or
binary.

The following was proved n 1925 by a Polish mathematician E. Żyliński.

Fact 2.4
No propositional connective other than ↑ and ↓ suffices to define all the remain-
ing connectives.

We show now as examples how to define some of our connectives ¬, ∪, ∩, ⇒, and

⇔ in terms of ↑ or ↓ leaving the definability of other connectives as an exercise.

Example 2.2
Definition of negation ¬ in terms of ↑.

This is an interesting example as it shows that one can define a one argument
connective in terms of a two argument connective.
Let’s now look at Sheffer’s alternative negation connective ↑.

Alternative Negation ↑
A B (A ↑ B)
T T F
T F T (2.16)
F T T
F F T

We now write the table (3.3) in the ”formula independed” form of the following
equations.

(T ↑ T ) = F, (T ↑ F ) = T, (F ↑ T ) = T, (F ↑ F ) = T (2.17)

Observe that (T ↑ T ) = F and (F ↑ F ) = T . This means that logical value of

a formula (A ↑ A) is the same as logical value of a formula ¬A, for any logical
value the formula A can take. We write it following our notation as as

¬A = (A ↑ A) (2.18)

36
and call it a definition of ¬ in terms of ↑. We verify its correctness of of by
building the table below.

A ¬A (A ↑ A) computation (A ↑ A)
T F (T ↑ T ) = F F (2.19)
F T (T ↑ T ) = F T
The table shows that the logical value of a formula ¬A is the same as logical
value of a formula (A ↑ A), for any logical value their basic component A can
take, i.e. that our definition (2.18) is correct.

Example 2.3
Definition of conjunction ∩ in terms of ↑.

Observe now that the Sheffer’s connective table (3.3) looks as a negation of the
conjunction table (2.4). It means that the logical value a formula (A ∩ B) is
the same as logical value of a formula ¬(A ↑ B), for all logical values of A and
B. We write it as
(A ∩ B) = ¬(A ↑ B). (2.20)
We have just proved the formula (2.18) to be true for any formula and hence
for the formula ¬(A ↑ B), i.e. we get that ¬(A ↑ B) = (A ↑ B) ↑ (A ↑ B). The
formula (2.24) becomes (A ∩ B) = (A ↑ B) ↑ (A ↑ B).
We call the equality
(A ∩ B) = (A ↑ B) ↑ (A ↑ B) (2.21)
the definition of conjunction in terms of negation and Sheffer’s connective.

Let’s now examine the Lukasiewicz’ s joint negation connective ↓. The formula
A ↓ B is read: neither A nor B. As it is a special connective we re-write its
truth table separately.

Joint Negation ↓

A B (A ↓ B)
T T F
T F F (2.22)
F T F
F F T

We now write the table (3.30) in an ”formula independed” form of the following
equations.

(T ↓ T ) = F, (T ↓ F ) = F, (F ↓ T ) = F, (F ↓ F ) = T (2.23)

37
Observe that T ↓ T = F and F ↓ F = T . This means that logical value of a
formula (A ↓ A) is the same as logical value of a formula ¬A, for any logical
value the formula A can take. We write it as

¬A = (A ↓ A) (2.24)
and call it a definition of ¬ in terms of ↓. We verify its correctness of of by
building the table below.

A ¬A (A ↓ A) computation (A ↓ A)
T F (T ↓ T ) = F F (2.25)
F T (F ↓ F ) = T T

The table shows that the logical value of a formula ¬A is the same as logical
value of a formula (A ↓ A), for any logical value their basic component A can
take, i.e. that our definition (2.24) is correct.

Exercise 2.6
Prove that the equality

(A ∪ B) = ((A ↓ B) ↓ (A ↓ B)) (2.26)

defines ∪ in terms of ↓.

Solution
To prove the correctness of the equation (2.26) we construct a table below.

A B (A ∪ B) ((A ↓ B) ↓ (A ↓ B))
T T T ((T ↓ T ) ↓ (T ↓ T )) = (F ↓ F ) = T
T F T ((T ↓ F ) ↓ (T ↓ F )) = (F ↓ F ) = T (2.27)
F T T ((F ↓ T ) ↓ (F ↓ T )) = (F ↓ F ) = T
F F F ((F ↓ F ) ↓ (F ↓ F )) = (T ↓ T ) = F
The table shows that the logical value of a formula (A∪B) is the same as logical
value of a formula ((A ↓ B) ↓ (A ↓ B)), for any logical value the formulas can
take depending of logical values of their basic components A, B, i.e. that our
definition (2.26) is correct.

2.3 Examples of Propositional Tautologies

Now we connect syntax (formulas of a given language L) with semantics (assign-
ment of truth values to the formulas of L). In logic we are interested in those

38
propositional formulas that must be always true because of their syntactical
structure without reference to the meaning of the propositions they represent.
Such formulas are called propositional tautologies.

Example 2.4
Given a formula (A ⇒ A). Lets now evaluate its logical value for all possible
logical values of its basic component A, i.e. for A=T, and A=F. We put our
calculation in a form of a table below.

A (A ⇒ A) computation (A ⇒ A)
T (T ⇒ T ) = T T (2.28)
F (F ⇒ F ) = T T

The logical value of the formula (A ⇒ A) is always T, what means that it is a

propositional tautology. The table (11.27) is called a truth table for the
formula (A ⇒ A).

Example 2.5
We construct a truth table for a formula (A ⇒ B) as follows.

A B (A ⇒ B) computation (A ⇒ B)
T T (T ⇒ T ) = T T
T F (T ⇒ F ) = F F (2.29)
F T (F ⇒ T ) = T T
F F (F ⇒ F ) = T T

The logical value of the formula (A ⇒ B) is F for A=T and B=F what means
that it is not a propositional tautology. We put these ideas in a form of the
following definition.

Definition 2.3
For any formula A of a propositional language L, we say that A is a propositional
tautology if and only if the logical value of A is T (we write it A=T) for all
possible logical values of its basic components. We write
|= A
to denote that A is a tautology.

Examples of Propositional Tautologies

Given any formula A of L = L{¬, ∩, ∪, ⇒, ⇔} . Here are some basic classical

propositional tautologies, the first of which we have just proved as the example

39
by constructing the table (11.27). We leave the proofs of others as an easy
exercise.

Identity for Implication

|= (A ⇒ A) (2.30)

Identity for Equivalence

|= (A ⇔ A) (2.31)

Excluded Middle
|= (¬A ∪ A) (2.32)

One of the most frequently used classical tautologies are the laws of detachment
for implication and equivalence. The implication law was already known to the
Stoics (3rd century BC) and a rule of inference, based on it is called Modus
Ponens, so we use the same name here.

Modus Ponens
|= ((A ∩ (A ⇒ B)) ⇒ B) (2.33)

Detachment
|= ((A ∩ (A ⇔ B)) ⇒ B) (2.34)
|= ((B ∩ (A ⇔ B)) ⇒ A)

Mathematical and not only mathematical theorems are usually of the form of
an implication, so we will discuss some terminology and more properties of
implication.

Sufficient Given an implication (A ⇒ B), A is called a sufficient condition

for B to hold.

Necessary Given an implication (A ⇒ B), B is called a necessary condition

for A to hold.

Simple The implication (A ⇒ B) is called a simple implication.

Converse Given a simple implication (A ⇒ B), the implication (B ⇒ A) is

called a converse implication to (A ⇒ B).

Opposite Given a simple implication (A ⇒ B), the implication (¬B ⇒ ¬A)

is called an opposite implication to (A ⇒ B). It is also often called a
contrapositive implication.

Contrary Given a simple implication (A ⇒ B), the implication (¬A ⇒ ¬B)

is called a contrary implication to (A ⇒ B).

40
Each of the following pairs of implications: a simple and an opposite, and a
converse and a contrary are equivalent, i.e. the following formulas are tautolo-
gies:

Laws of contraposition (1)

|= ((A ⇒ B) ⇔ (¬B ⇒ ¬A)), (2.35)
|= ((B ⇒ A) ⇔ (¬A ⇒ ¬B)).

The laws of contraposition (2.35) make it possible to replace, in any deductive

argument, a sentence of the form (A ⇒ B) by (¬B ⇒ ¬A), and conversely.
The relationships between all implications involved in the contraposition laws
are usually shown graphically in a following form, which is called the square of
opposition.

converse
(A =⇒ B) (B =⇒ A)

contrapositive
contrary contrary

(¬A =⇒ ¬B) (¬B =⇒ ¬A)

converse

Equivalent implications are situated at the vertices of one and the same diag-
onal. It follows from the contraposition laws that to prove all of the following
implications: (A ⇒ B), (B ⇒ A), (¬A ⇒ ¬B), (¬B ⇒ ¬A), it suffices to prove
any pairs of those implications which are situated at one and the same side of
the square, since the remaining two implications are equivalent to those already
proved to be true.
Consider now the following tautology:
|= ((A ⇔ B)) ⇔ ((A ⇒ B) ∩ (B ⇒ A))). (2.36)

The above tautology (2.36) says that in order to prove a theorem of a form of
(A ⇔ B) it suffices to prove two implications: the simple one (A ⇒ B) and the
converse one (B ⇒ A). Conversely, if a formula (A ⇔ B) is a theorem, then
the implications (A ⇒ B) and (B ⇒ A) are also theorems.
In other words, B is then a necessary condition for A, and at the same time B
is a sufficient condition for A. Accordingly, we say that a theorem of the form

41
of a formula (A ⇔ B) is often formulated as: ” B is necessary and sufficient
condition for A”.
Other laws developed by the Stoics are the hypothetical syllogism and modus
tollendo ponens. We present them here in a form of logical tautology, not as
the rule of reasoning as it was developed. The relationship between those two
approaches is quite obvious and will be discussed in detail in the proof theory
chapter.

Hypothetical syllogism

|= (((A ⇒ B) ∩ (B ⇒ C)) ⇒ (A ⇒ C))

|= ((A ⇒ B) ⇒ ((B ⇒ C) ⇒ (A ⇒ C))) (2.37)

|= ((B ⇒ C) ⇒ ((A ⇒ B) ⇒ (A ⇒ C))).

Modus tollendo ponens

|= (((A ∪ B) ∩ ¬A) ⇒ B) (2.38)

|= (((A ∪ B) ∩ ¬B) ⇒ A)

Here are some other tautologies with a history centuries old. First is called
Duns Scotus Law after an eminent medieval philosopher who lived at the turn
of the 13th century. Second is called Clavius Law , after Clavius, a Euclid
commentator who lived in the late 16th century. The reasonings based on this
law were already known to Euclid, but this type of inference became popular in
scholarly circles owing to Clavius, hence the name. The third is called Frege
Laws after G. Frege who was first to give a formulation of the classical propo-
sitional logic as a formalized axiomatic system in 1879, adopting the second of
them as one of his axioms.

Duns Scotus
|= (¬A ⇒ (A ⇒ B)) (2.39)

Clavius
|= ((¬A ⇒ A) ⇒ A) (2.40)

Frege
|= (((A ⇒ (B ⇒ C)) ∩ (A ⇒ B)) ⇒ (A ⇒ C)) (2.41)

|= ((A ⇒ (B ⇒ C)) ⇒ ((A ⇒ B) ⇒ (A ⇒ C)))

Double Negation
|= (¬¬A ⇔ A) (2.42)

42
Next set of tautologies deal with apagogic proofs which are the proofs by
reductio ad absurdum. The method of apagogic proof consists in negating
the theorem which is to be proved. If the assumption that the theorem is
false yields a contradiction, then we conclude that the theorem is true. The
correctness of this reasoning is based on the following tautology.

Reductio ad Absurdum

|= ((¬A ⇒ (B ∩ ¬B)) ⇒ A) (2.43)

If the theorem to be proved by reductio ad absurdum is of the form of an

implication (A ⇒ B), then the prove often follows a following pattern: it is
assumed that ¬(A ⇒ B) is true, and we try to deduce a contradiction from
this assumption. If we succeed in doing so, then we infer that the implication
(A ⇒ B) is true. The correctness of this reasoning is based on the following
version the reductio ad absurdum tautology (2.43).

|= (((¬(A ⇒ B) ⇒ (C ∩ ¬C)) ⇒ (A ⇒ B)).

Sometimes to prove (A ⇒ B) it is assumed that (A ∩ ¬B) is true and if the

assumption leads to contradiction, then we deduce that the implication (A ⇒ B)
is true. In this case a tautology, which guarantee the correctness this kind of
argument is:

|= (((A ∩ ¬B) ⇒ (C ∩ ¬C)) ⇒ (A ⇒ B)).

Often, when assuming (A ∩ ¬B), we arrive, by deductive reasoning, at the

conclusion ¬A. Then we need the following tautology:

|= (((A ∩ ¬B) ⇒ ¬A) ⇒ (A ⇒ B)).

Sometimes, on assuming (A ∩ ¬B) we arrive by deductive reasoning at the

conclusion B. The following tautology is then applied:

|= (((A ∩ ¬B) ⇒ B) ⇒ (A ⇒ B)).

The proofs based on the application of the laws of contraposition (2.35) are
also classed as apagogic. Instead of proving a simple theorem (A ⇒ B) we
prove the opposite theorem (¬B ⇒ ¬A), which is equivalent to the simple
one. The following two tautologies, also called laws of contraposition, are used,
respectively, when the hypothesis or the thesis of the theorem to be proved is
in the form of a negation.

43
Laws of Contraposition (2)

|= ((¬A ⇒ B) ⇔ (¬B ⇒ A)), (2.44)

|= ((A ⇒ ¬B) ⇔ (B ⇒ ¬A)).

We present now some tautologies characterizing basic properties of conjunction,

disjunction, quivalence, and their interactions.

Conjunction
|= ((A ∩ B) ⇒ A), |= ((A ∩ B) ⇒ B),
|= (((A ⇒ B) ∩ (A ⇒ C)) ⇒ (A ⇒ (B ∩ C))),
|= (((A ⇒ B) ∩ (C ⇒ D)) ⇒ ((A ∩ C) ⇒ (B ∩ D))),
|= (A ⇒ (B ⇒ (A ∩ B))).

Disjunction
|= ((A ⇒ (A ∪ B)), |= ((B ⇒ (A ∪ B)),
|= (((A ⇒ B) ∩ (B ⇒ C)) ⇒ ((A ∪ B) ⇒ C)),
|= (((A ⇒ B) ∩ (C ⇒ D)) ⇒ ((A ∪ C) ⇒ (B ∪ D))).

Here are some more important and frequently used equivalence tautologies,
called also the equivalence laws.

Idempotence
|= ((A ∩ A) ⇔ A), |= ((A ∪ A) ⇔ A),

Associativity
|= (((A ∩ B) ∩ C) ⇔ (A ∩ (B ∩ C))),
|= (((A ∪ B) ∪ C) ⇔ ((A ∪ (B ∪ C))).

Commutativity

|= ((A ∩ B) ⇔ (B ∩ A)), |= ((A ∪ B) ⇔ (B ∪ A)).

Distributivity

|= ((A ∩ (B ∪ C)) ⇔ ((A ∩ B) ∪ (A ∩ C))), (2.45)

|= ((A ∪ (B ∩ C)) ⇔ ((A ∪ B) ∩ (A ∪ C))). (2.46)

De Morgan

|= (¬(A ∪ B) ⇔ (¬A ∩ ¬B)), |= (¬(A ∩ B) ⇔ (¬A ∪ ¬B)). (2.47)

44
Implication
|= ((A ⇒ B) ⇔ (¬A ∪ B)). (2.48)

Negation of Implication

|= (¬(A ⇒ B) ⇔ (A ∩ ¬B)).

Negation of Equivalence

|= (¬(A ⇔ B) ⇔ (A ∩ ¬B) ∪ (B ∩ ¬A)).

Double Negation
|= (¬¬A ⇔ A). (2.49)

Exportation and Importation

|= (((A ∩ B) ⇒ C) ⇔ (A ⇒ (B ⇒ C))).

De Morgan laws (2.47) are named after A. De Morgan (1806 - 1871), an English
logician, who discovered analogous laws for the algebra of sets. They stated that
for any sets A,B the complement of their union is the same as the intersection
of their complements, and vice versa, the complement of the intersection of two
sets is equal to the union of their complements. The laws of the propositional
calculus were formulated later, but they are usually also called De Morgan Laws.

2.4 Predicate Language Description and Appli-

cation to Artificial Intelligence
We define a predicate language L following the pattern established by the sym-
bolic and propositional languages definitions (2.1), (7.44). The predicate lan-
guage L is much more complicated in its structure. Its alphabet A is much
richer. The definition of its set of formulas F is more complicated. In order to
define the set F we introduce an additional set T, called a set of terms of the
predicate language L. We single out this set not only because we need it for the
definition of formulas, but also because of its role in the development of other
notions of predicate logic.

Definition 2.4
By a predicate language L we understand a triple

L = (A, T, F), (2.50)

where A is a predicate alphabet, T, is the set of terms, and F is a set of

formulas.

45
Alphabet A
The components of A are as follows.
1. Propositional connectives: ¬, ∩, ∪, ⇒, ⇔.
2. Quantifiers: we adopt two quantifiers; ∀ (for all, the universal quantifier)
and ∃ (there exists, the existential quantifier).
In a case of the classical logic it is possible to adopt only one quantifier and
to define the other in terms of it and propositional connectives. But the two
quantifiers express better the common intuition, so we assume that we have two
of them.
3. Parenthesis: ( and ).
4. Variabes: we assume that we have, as we did in the propositional case a
countably infinite set VAR of variables. The variables now have a different
meaning than they had in the propositional case. We hence call them variables,
or individual variables to distinguish them from the propositional variables. We
also denote denote them by different symbols, namely by letters x, y, z, ..., with
indices, if necessary. We express it by writing V AR = {x1 , x2 , ....}.
5. Constants: the constants represent in ”real life” concrete elements of sets.
We denote constants by by c, d, e..., with indices, if necessary. We assume that
we have a countably infinite set C = {c1 , c2 , ...} of constants.
6. Predicate symbols: the predicate symbols represent ”real life” relations. We
denote them by P, Q, R, ... with indices, if necessary. We use symbol P for the
set of all predicate symbols. We assume that P it countably infinite.
In ”real life” we write symbolically x < y to express that element x is smaller
then element y according to the two argument order relation <. In our predicate
language L we represent the relation < as a two argument predicate P ∈ P and
write P (x, y), where now x, y are individual variables from the set VAR.
Mathematical statements n < 0, 1 < 2, 0 < m are represented in L by
P (x, c1 ), P (c, c3 ), P (c1 , y), respectively. Here c1 , c2 , c3 are any constants and
x, y any variables.
7. Function symbols: the function symbols represent ”real life” functions. We
denote function symbols by f, g, h, ..., with indices, if necessary. We use symbol
F for the set of all function symbols. We assume that the set F is countably
infinite.
Set T of terms
Terms are expressions built out of function symbols and variables. They describe
how we build compositions of functions. We define the set T of terms recursively
as follows.
1. All variables are terms.

46
2. All constants are terms.
3. For any function symbol f representing a function on n variables, and any
terms t1 , t2 , ..., tn , the expression f (t1 , t2 , ..., tn ) is a term.
4. The set T of terms is the smallest set that fulfills the conditions 1. - 3.
Consider a ”real life” function given by a formula sin(x + y). It is a composition
of two functions defined by formulas sinx and x + y. The sin is one argument
function and we represent it as term f (x) for f ∈ F. The + is a two argument
function and we represent it as a term g(x, y) for g ∈ F. The ”real life” function
sin(x+y) is hence represented by a term f (g(x, y)), where x, y are any individual
variables from the set VAR. Observe that to obtain the predicate language
representation of for example x + y we can first write the real two argument
function formula x + y as +(x, y) and then replace the addition symbol + by
any two argument function symbol g ∈ F and get the term g(x, y).
Here are some more terms of L.

h(c1 ), f (g(c, x)), g(f (f (c)), g(x, y)), f1 (c, g(x, f (c))), g(g(x, y), g(x, h(c))) ....

Set F of formulas
Formulas are now expressions built out of elements of the alphabet A and the
set T of terms. We denote them, as in propositional case by A, B, C, ..... with
indices, if necessary. We build them, as before in recursive steps, the fist of
them says as in the propositional case: all atomic formulas are formulas. The
atomic formulas are the simplest formulas as the propositional variables were
in the case of propositional language. We define them as follows.

Definition 2.5 An atomic formula is any expression of the form R(t1 , t2 , ..., tn )
where R is any predicate R ∈ P and t1 , t2 , ..., tn are terms, i.e. t1 , t2 , ..., t∈ T.

To represent a mathematical statement x + y = 5 in L we first observe that =

as a two argument relation and + is a two argument function, x, y are variables
and 5 is a number. We represent, as before, + by a two argument function
symbol g ∈ F, the relation = by a predicate symbol P ∈ P, the number 5 by a
constant c ∈ C. We re-write x + y = 5 as = (+(x, y), 5), replace mathematical
symbols by corresponding L symbols and get an atomic formula P (g(x, y), c)
representing in L the statement x + y = 5. We have used the same letters x,
y to represent mathematical and and atomic formula variables. We can also
use any other letters for individual variables in the atomic formula re[presenting
x + y = 5. For example P (g(x1 , x2 ), c), P (g(y, x), c).
Here are some more atomic formulas of L.

Q(c), Q(x), Q(g(x1 , x2 )), R(c, d), R(x, f (c)), R(g(x, y), f (g(c, z))), .....

47
Definition 2.6
The set F of formulas of L is the smallest set meeting the following conditions.
1. All atomic formulas (definition 10.2) are formulas;
2. If A, B are formulas, then ¬A, (A ∩ B), (A ∪ B), (A ⇒ B), (A ⇔ B) are
formulas;
3. If A is a formula, then ∀xA, ∃xA are formulas for any variable x ∈ V AR.

Here are some formulas of L.

R(c, d), ∃yR(y, f (c)), R(x, y), (∀xR(x, f (c)) ⇒ ¬R(x, y)),

(R(c, d) ∩ ∀zR(z, f (c))), ∀yR(y, g(c, g(x, f (c)))), ∀y¬∃xR(x, y).

Let’s look now closer at the following formulas.

R(c1 , c2 ), R(x, y), (R(y, d) ⇒ R(a, z)), ∃xR(x, y), ∀yR(x, y), ∃x∀yR(x, y).

Here are some simple observations.

1. Some formulas are without quantifiers.
For example formulas R(c1 , c2 ), R(x, y), (R(y, d) ⇒ R(a, z)). A formula with-
out quantifiers is called an open formula.
Variables x, y in R(x, y) are called free variables. The variables y in R(y, d)
and z in R(a,z) are also free.
2. Quantifiers bind variables within formulas.
The variable x is bounded by ∃x in the formula ∃xR(x, y), the variable y is
free. The variable y is bounded by ∀y in the formula ∀yR(x, y), the variable
y is free.
3. The formula ∃x∀yR(x, y) does not contain any free variables, neither does
the formula R(c1 , c2 ). A formula without any free variables is called a closed
formula or a sentence.
Sometimes in order to distinguish more easily which variable is free and which
is bound in the formula we might use the bold face type for the quantifier
bound variables and write the formulas as follows.

(∀xQ(x, y), ∃yP (y), ∀yR(y, g(c, g(x, f (c)))),

(∀xP (x) ⇒ ∃yQ(x, y)), (∀x(P (x) ⇒ ∃yQ(x, y)))

Observe that the formulas ∃yP (y), (∀x(P (x) ⇒ ∃yQ(x, y))) are closed. We
call a close formula a sentence.

48
Example 2.6
Consider atomic formulas: P (y), Q(x, c), R(z), P1 (g(x, y), z). Here are some
non atomic formulas formed out of them.
1. (P (y) ∪ ¬Q(x, c)) ∈ F. This is an open formula A with two free variables
x,y. We denote A this as formula A(x, y).
2. ∃x(P (y) ∪ ¬Q(x, c)) ∈ F. We write x to denote that x is a bound variable.
The variable y is free. This is a formula B with one free variable y. We denote
B as a formula B(y).
3. ∀y(P (y) ∪ ¬Q(x, c)) ∈ F. The variable y is bound, the variable x is free.
We denote this formula by for example A1 (x).
4. ∀y∃x(P (y) ∪ ¬Q(x, c)) ∈ F has no free variables. It is a closed formula
called also a sentence.

Exercise 2.7
Given the following formulas of L:
P (x, f (c, y)), ∃cP (x, f (c, y)), ∀xf (x, P (c, y)), ∃xP (x, f (c, y)) ⇒ ∀yP (x, f (c, y)).
1. Indicate whether they are, or are not well formed formulas of F. For those
which are not in F write a correct formula. 2. For each correct, or corrected
formula identify all components: connectives, quantifiers, predicate and function
symbols, and list all its terms. 3. For each formula identify its s free and bound
variables. State which are open and which are closed formulas (sentences), if
any.

Solution
Formula A1 = P (x, f (c, y)).
It is a correct atomic formula. P is a 2 argument predicate symbol, f is a
2 argument function symbol, c is a constant. We write it symbolically: P ∈
P, f ∈ F, c ∈ C. It is an open formula with two free variables x,y. We denote
it by A1 (x, y). It has no bound variables.
Formula A2 = ∃cP (x, f (c, y)).
It is a not a correct formula, i.e. ∃cP (x, f (c, y)) 6∈ F. The expression ∃c has no
meaning because c is a constant, not a variable.
The corrected formulas are: B1 = ∃xP (x, f (c, y)), B2 = ∃yP (x, f (c, y)), and
formulas B = ∃zP (z, f (c, y)) for any variable z different then x and y.
None of the correct formulas are open. Variable y is free in B1 = B1 (y), variable
x is free in B2 = B2 (x), both variables x and y are free in all formulas B =
B(x, y). All formulas are nether close, nor open. The terms appearing in any
of them are the same as in A1 = P (x, f (c, y)) and are: x, y, c, f (c, y).

49
Formula A3 = ∀xf (x, P (c, y)).
It is a not a correct formula, i.e. ∀xf (x, P (c, y)) 6∈ F. The function symbol f in
front f (x, P (c, y)) indicate a term and terms are not formulas. Moreover, the
atomic formula P (c, y) can’t be put inside a term!
Formula A4 = ∃xP (x, f (c, y)) ⇒ ∀yP (x, f (c, y))
It is a not a correct formula. The correct formula is A = (∃xP (x, f (c, y)) ⇒
∀yP (x, f (c, y))). It has two free variables x and y and we write it as A = A(x, y).
We often use logic symbols, while writing mathematical statements in a more
symbolic way. For example, mathematicians to say ”all natural numbers are
greater then zero and some integers are equal 1” often write

x ≥ 0, ∀x∈N and ∃y∈Z , y = 1.

Some of them who are more ”logic oriented” would write it as

∀x∈N x ≥ 0 ∩ ∃y∈Z y = 1,

or even as
(∀x∈N x ≥ 0 ∩ ∃y∈Z y = 1).
Observe that none of the above symbolic statements, not even the last one, are
formulas of the predicate language. These are mathematical statements written
with mathematical and logic symbols. They are written with different degrees
of ”logical precision”, the last being, from a logician point of view, the most
precise.
Our goal now is to ”translate ” mathematical and natural language statement
into correct logical formulas, i.e. into formulas of the predicate language L. Let’s
start with some observations about the statements above.
The quantifiers in ∀x∈N and ∃y∈Z used in all of them are not the one used in
logic. In our language L we use only quantifiers ∀x and ∃y, for any variables
x, y ∈ V AR. The quantifiers ∀x∈N , ∃y∈Z are called quantifiers with restricted
domain. The first is restricted to the domain of natural numbers, the second
to the integers. The restriction of the quantifier domain can, and often is given
by more complicated statements. For example we say ”for all x > 2” and
write ∀x>2 , or we say ”exists x > 2 and at same time x + 2 < 8” and write
symbolically ∃(x>2∩x+2<8) . We introduce the quantifiers with restricted domain
into our predicate logic language by expressing them within the language L as
follows.

Definition 2.7
The quantifiers ∀A(x) , ∃A(x) are called quantifiers with restricted domain,
or restricted quantifiers, where A(x) ∈ F is any formula with any free vari-
able x ∈ V AR.

50
A formula ∀A(x) B(x) stands for a formula ∀x(A(x) ⇒ B(x)) ∈ F. We write
it symbolically as
∀A(x) B(x) ≡ ∀x(A(x) ⇒ B(x)). (2.51)

A formula ∃A(x) B(x) stands for a formula ∃x(A(x) ∩ B(x)) ∈ F. We write it

symbolically as
∃A(x) B(x) ≡ ∃x(A(x) ∩ B(x)) (2.52)

The definition 2.7 of restricted quantifiers is obviously faithful to our intuitive

meaning of quantifiers. We use informally a symbol ≡ to stress that we they
are in a sense equivalent. We call (8.18) and (8.19) transformations rules for
restricted quantifiers.
We carry our translations of mathematical statements written with logical sym-
bols into a formula of predicate language L a sequence of steps. Given a math-
ematical statement S written with logical symbols. We obtain a corresponding
formula A that is our translation into L by conducting the following steps.
Step 1. We identify basic statements in S, i.e. mathematical statements that
involve only relations. They will be translated into atomic formulas.
We identify the relations in the basic statements and choose the predicate sym-
bols as their names.
We identify all functions and constants (if any) in the basic statements and
choose the function symbols and constant symbols as their names.
Step 2. We write the basic statements as atomic formulas of L.
Remember that in the predicate language L we write function symbol in front
of the function arguments, not between them as we write in mathematics. The
same applies to relation symbols when we form atomic formulas. For example
a basic mathematical statement x + 2 > y could be re-written as > (+(x, 2), y),
and then we could immediately write it as an atomic formula P (f (x, c), y),
where P ∈ P stands for two argument relation >, f stands for two argument
function +, and c stands for the number (constant) 2.
Step 3. We re-write the statement S a logical formula with restricted domain
quantifiers.
Step 4. We apply equivalences (8.18) and (8.19) to the formula from Step 3
and obtain a formula A of L as a translation, i.e. a representation of the given
mathematical statement.
When we conduct a translation from mathematical statement written without
logical symbols we add a Step 0 to this list to first write the mathematical
statement with logical symbols.
Step 0. We identify logical connectives and quantifiers and write the statement
using them that is as close to the structure of a logical formula as possible.

51
Exercise 2.8
Given a mathematical statement S written with logical symbols

(∀x∈N x ≥ 0 ∩ ∃y∈Z y = 1)

1. Translate it into a proper logical formula with restricted domain quantifiers

i.e. into a formula of L that uses the restricted domain quantifiers. 2. Translate
your restricted domain quantifiers logical formula into a correct logical formula
without restricted domain quantifiers, i.e. into a formula of L.

Solution
Step 1. The basic statements in S are: x ∈ N, x ≥ 0, y ∈ Z, y = 1. The
relations are: ∈ N, ∈ Z, ≥, =. We use one argument predicate symbols N, Z
for ∈ N, ∈ Z, respectively. We use two argument predicate symbols G for ≥,
and E for =. There are no functions. We have two constant symbols c1 , c2 for
numbers 0 and 1, respectively.
Step 2. We write N (x), Z(x) for x ∈ N, x ∈ Z, respectively. G(x, c1 ) for x ≥ 0
and E(y, c2 ) for y = 1. These are all atomic formulas.
Step 3. The statement S becomes a restricted quantifiers formula:

(∀N (x) G(x, c1 ) ∩ ∃Z(y) E(y, c2 )).

Step 4. A formula A ∈ F that corresponds to S is

(∀x (N (x) ⇒ G(x, c1 )) ∩ ∃y (Z(y) ∩ E(y, c2 ))).

Here is a perfectly acceptable short solution to exercise 2.8. We presented

the long solution in order to explain all steps needed to be performed when one
writes the short solution.

Example 2.7
Given a mathematical statement S written with logical symbols

(∀x∈N x ≥ 0 ∩ ∃y∈Z y = 1)

We translate it into a proper formula of L as follows.

The basic statements in S are: x ∈ N, x ≥ 0, y ∈ Z, y = 1. The corresponding
atomic formulas of L are: N (x), G(x, c1 ), Z(y), E(y, c2 ), respectively.
The statement S becomes becomes restricted quantifiers formula
(∀N (x )G(x, c1 ) ∩ ∃Z(y) E(y, c2 )). Applying restricted quantifiers definition 2.7
and transformation rules (8.18), (8.19) we get a following formula A ∈ F

(∀x(N (x) ⇒ G(x, c1 )) ∩ ∃y(Z(y) ∩ E(y, c2 ))).

52
Exercise 2.9
Here is a mathematical statement S:
”For all real numbers x the following holds: If x < 0, then there is a natural
number n, such that x + n < 0.”
1. Re-write S as a symbolic mathematical statement SF that only uses mathe-
matical and logical symbols. 2. Translate the symbolic statement SF into to a
corresponding formula A ∈ F of the predicate language L.

Solution
The symbolic mathematical statement SF is : ∀x∈R (x < 0 ⇒ ∃n∈N x + n < 0).
We write R(x) for x ∈ R , N(y) for n ∈ N , and atomic formula L(x, c) for the
basic statement x < 0. We write f(x,y) for the function +(x, n) and a constant
c for the number 0. We write atomic formula L(f(x,y), c) for x + n < 0. The
symbolic statement SF becomes ∀R(x) (L(x, c) ⇒ ∃N (y) L(f (x, y), c)). The corre-
sponding formula A ∈ F is ∀x(N (x) ⇒ (L(x, c) ⇒ ∃y(N (y) ∩ L(f (x, y), c))).
There are various kinds of non-mathematical statements, that obviously cannot
be justified on the basis of propositional logic. Consider for example a statement
”Any friend of Mary is a friend of John and Peter is not John’s friend. Hence
Peter is not May’s friend. ”
Intuitively, what it says is always true, but translating it it into a propositional
language we get a formula ((a ∩ ¬b) ⇒ ¬c) that can be false. The validity of the
reasoning described by the statement follows from a more complexed structure
provided by the predicate language. We will discuss the notion of validity of
predicate language formulas, i.e. a semantics for predicate logic later. Natural
language statements and reasoning with them also play a special role in creation
of non-classical logics and in Artificial Intelligence research and applications.

Exercise 2.10
Translate a natural language statement S: ”Any friend of Mary is a friend of
John and Peter is not John’s friend. Hence Peter is not May’s friend.” into a
formula A ∈ F of the predicate language L.

Solution
1. We identify the basic relations and functions (if any) and translate them into
atomic formulas.
We have only one relation of ”being a friend”. It is a two argument relation. We
write atomic formula F(x, y) for ”x is a friend of y”. We use constants m, j, p
for Mary, John, and Peter, respectively. We have the following atomic formulas:
F(x, m) for ”x is a friend of Mary”, F(x, j) for ”x is a friend of John”, F(p, j)
for ”Peter is a friend of John”.

53
2. Statement ”Any friend of Mary is a friend of John” translates into a re-
stricted quantifier formula ∀F (x,m) F (x, j). Statement ”Peter is not John’s
friend” translates into ¬F (p, j), and ”Peter is not May’s friend” translates into
¬F (p, m).
3. Restricted quantifiers formula for S is

((∀F (x,m) F (x, j) ∩ ¬F (p, j)) ⇒ ¬F (p, m))

and the formula A ∈ F of L is

((∀x(F (x, m) ⇒ F (x, j)) ∩ ¬F (p, j)) ⇒ ¬F (p, m)).

Here are simple steps we follow in order to perform translations from natural
language to the symbolic predicate language L. They are similar to the steps we
used in the translations of mathematical formulas nevertheless we voice them
separately and call them rules of translation.
Rules of translation to L.
1. Identify the basic relations and functions (if any) and translate them into
atomic formulas.
2. Identify propositional connectives and use symbols ¬, ∪, ∩, ⇒, ⇔ for them.
3. Identify quantifiers. Restricted ∀A(x), ∃A(x) and non-restricted ∀x, ∃x.
4. Use the symbols from 1. - 3. and restricted quantifiers transformation rules
(8.18) and (8.19) to write A ∈ F of the predicate language L.

Example 2.8
Given a natural language statement S: ”For any bird one can find some birds
that are white.” The translation of S into a formula of the predicate language L
is
∀x(B(x) ⇒ ∃x(B(x) ∩ W (x))).

We follow the rules of translation and get the following.

1. Atomic formulas: B(x), W(x). We write one argument predicate B(x) for ”
x is a bird” and one argument predicate W(x) for ” x is white”.
2. There is no propositional connectives in S.
3. Restricted quantifiers: ∀B(x) for ”any bird ” and ∃B(x) for ”one can find
some birds”. A restricted quantifiers formula for S is ∀B(x) ∃B(x) W (x).
4. By the transformation rules we get a required non-restricted formula of the
predicate language L, i.e. the formula ∀x(B(x) ⇒ ∃x(B(x) ∩ W (x))).
Observe that the quantifier ∀x binds the variable x only in the first B(x), even
if its scope covers the second appearance of B(x) as well. It happens because

54
the second appearance of B(x) is bounded by the quantifier ∃x. Let’s re-write
the formula A using x to indicate this fact

∀x(B(x) ⇒ ∃x(B(x) ∩ W (x))).

In this case, and in the similar cases we can apply a predicate logic law of
quantifiers, called Rename Variables Law to our formula A and get a formula
B that is logically equivalent to A. It means that the formula B states exactly
the same what A states but is written in a more comprehensible form:

∀x(B(x) ⇒ ∃y(B(y) ∩ W (y))).

We will discuss and study Laws of Quantifiers in the next section. There is
another important law, one of the Distributivity Laws that allows us to transform
B into a formula ∀x∃y(B(x) ⇒ (B(y) ∩ W (y))). We express it as the following
example.

Example 2.9
Given a natural language statement S: ”For any bird one can find some birds
that white.” The translation of S into a formula of the predicate language L is

∀x∃y(B(x) ⇒ (B(y) ∩ W (y))).

Exercise 2.11
Translate into L a natural language statement
S: ” Some patients like all doctors.”

Solution.
1. Atomic formulas: P(x), D(x), L(x, y). We write one argument predicate
P(x) for ” x is a patient”, one argument predicate D(x) for ” x is a doctor”, and
two argument predicate L(x,y) for ” x likes y”.
2. There is no propositional connectives in S.
3. Restricted quantifiers: ∃P (x) for ”some patients ” and ∀D(x) for ”all doctors”.
Observe that we can’t write L(x, D(y)) for ”x likes doctor y”. D(y) is a predicate,
not a term and hence L(x, D(y)) is not a formula. We have to express the
statement ” x likes all doctors y” in terms of restricted quantifiers and predicate
L(x,y) only. The statement ” x likes all doctors y” means ” all doctors y are
liked by x”, i.e. ”for all doctors y, x likes y”. This translates to ∀D(y) L(x, y)
and the statement S translates to ∃P (x) ∀D(x) L(x, y).
4. By the transformation rules we get the following translation of S into L.

∃x(P (x) ∩ ∀y(D(y) ⇒ L(x, y))).

55
Translations to Logic in Artificial Intelligence

In Artificial Intelligence (AI) we usually deal with what is called an intended

interpretation. It means we use logic symbols to describe, similarly as we do in
mathematics, concrete, specific universes with specific relations and functions,
or constants. In logic we use general symbols without any meaning because
the logic is created to define statements (formulas) and methods of reasoning
that are universally applicable (tautologically true) and hence independent of
any particular domain. In AI we use as symbols intended names for relations,
functions, and constants. The symbolic language we use is still a symbolic
language, even if intended names are used. In the AI language we can write, for
example , an atomic formula Like(John, Mary) instead of a formula L(c1 , c2 )
of L. We write greater(x, y), or > (x, y) instead of R(x, y). We leave it as an
exercise to formally define the AI language you would like to use.

Example 2.10
AI formulas corresponding to a statement
S: ”For every student there is a student that is an elephant.”
are as follows.
1. Restricted quantifiers AI formula:
∀Student(x) ∃Student(x) Elephant(x).
2. Non-restricted quantifiers AI formula :
∀x(Student(x) ⇒ ∃x(Student(x) ∩ Elephant(x))).
3. Re-name variables AI formula:
∀x(Student(x) ⇒ ∃y(Student(y) ∩ Elephant(y))).
4. AI formula after applying the the Distributivity Laws:
∀x∃y(Student(x) ⇒ (Student(y) ∩ Elephant(y))).

Observe that a proper formulas of the predicate language L corresponding

the example 2.10 statement ”For every student there is a student that is an
elephant.” are the same as the formulas corresponding to the natural language
statement ”For any bird one can find some birds that white.” of the example
2.9, namely
1. Restricted quantifiers L formula: ∀P (x) ∃P (x) R(x).
2. Non-restricted quantifiers L formula : ∀x(P (x) ⇒ ∃x(P (x) ∩ Rx))).
3. Re-name variables L formula: ∀x(P (x) ⇒ ∃y(P (y) ∩ R(y))).
4. L formula after applying the the Distributivity Laws
∀x∃y(P (x) ⇒ (P (y) ∩ R(y))).

56
The predicate symbols P, R, Student, Elephant denote in all cases one argument
predicates but AI predicate symbols Student, Elephant (of a slightly different
language than L) impose a particular meaning called the intended interpre-
tation. The predicate symbols P, R and any elements of the set of all predicate
symbols P of L.

Exercise 2.12
Translate a natural language statement ”Any friend of Mary is a friend of
John and Peter is not John’s friend. Hence Peter is not Mary’s friend.” into a
formula A of the predicate AI language (of your choice).

Solution
Statement ”Any friend of Mary is a friend of John” translates into a restricted
quantifier AI formula ∀F riend(x,M ary) F riend(x, John).
Statement ”Peter is not John’s friend” translates into ¬F riend(P eter, John),
and ”Peter is not Mary’s friend” translates into ¬F riend(P eter, M ary).
Restricted quantifiers AI formula for S is ((∀F riend(x,M ary) F riend(x, John) ∩
¬F riend(P eter, John)) ⇒ ¬F riend(P eter, M ary)).
The AI formula is ((∀x(F riend(x, M ary) ⇒ F riend(x, John)) ∩
¬F riend(P eter, John)) ⇒ ¬F riend(P eter, M ary)).
The AI formulas are very useful, as they ”read” as natural language statements
but it is very important to remember that they do not carry any meaning, as the
natural language statements do to the reader. An atomic formula Friend(Peter,
John) is just an atomic formula of a symbolic AI language as P (c, d) is in L.
We assign a meaning to them i.e. their semantics in a separate step as we did
in the propositional case. The first step in this process is an assignment of an
interpretation in a non-empty set U of the predicate, functional and constant
symbols. Each symbol can have many interpretations in a given set and we can
can define the interpretations an many sets. The AI intended interpretation
of the two argument predicate named Friend and constants Peter, John is
to define the set U and a relation Friend. This relation must hold between
elements Peter, John and other elements of U in a way we want to define what
”friendship” means in the set U. This is called in AI a conceptualization.

2.5 Predicate Semantics: Description and Laws

of Quantifiers
The notion of predicate tautology is much more complicated then that of the
propositional. We define it formally in later chapters. Predicate tautologies are
also called valid formulas, or laws of quantifiers to distinguish them from the
propositional case. We provide here a motivation, examples and an intuitive

57
definition of the predicate tautology. We also list and discuss the most used and
useful tautologies and equational laws of quantifiers.
The formulas of the predicate language L have meaning only when an interpre-
tation is given for the symbols. We define the interpretation I in a set U 6= ∅
by interpreting predicate, functional symbols as a concrete relation, function
defined in the universe U, and constants symbols as elements of the set U. The
set U is called the universe of the interpretation I. These two items specify a
model structure for L. We write it as a pair M = (U, I).
Given a formula A of L, and the model structure M = (U, I). Let’s denote by
AI a statement written with logical symbols determined by the formula A and
the interpretation I in the universe U. When A is a closed formula, it means it
is a sentence, formula without free variables, AI represents a proposition that
is true or false. When A is not a sentence it contains free variables and may
be satisfied (i.e. true) for some values in the universe U and not satisfied (i.e.
false) for the others. Lets look at few simple examples.

Example 2.11
Let A be a formula ∃xP (x, c) and consider a model structure M1 = (N, I1 ).
The universe of the interpretation I1 is the set N of natural numbers and we
define I1 as follows: we interpret the predicate P as relation < and the constant
c as number 5, i.e we put PI1 := andcI1 : 5.

The formula A: ∃xP (x, c) under the interpretation I1 becomes a mathematical

statement ∃x x < 0 defined in the set N of natural numbers. We write it for
short
AI1 : ∃x∈N x = 5.
AI1 is obviously a true mathematical statement and say that the formula A:
∃xP (x, c) is true under the interpretation I1 in M1 or that A is true in M1 .
We write it symbolically as

M1 |= ∃xP (x, c)

and say that M1 is a model for the formula A.

Example 2.12
Consider now a model structure M2 = (N, I2 ) and the formula A: ∃xP (x, c).
We interpret now the predicate P as relation < in the set N of natural numbers
and the constant c as number 0, i.e. we put PI2 :< and cI2 : 0.

The formula A: ∃xP (x, c) under the interpretation I2 mathematical statement

∃x x < 0 defined in the set N of natural numbers. We write it for short

AI2 : ∃x∈N x < 0.

58
AI2 is obviously a false mathematical statement. We say that the formula A:
∃xP (x, c) is false under the interpretation I2 in M2 or that A is false in M2 .
We write it symbolically as

M2 6 |= ∃xP (x, c)

and say that M2 is a counter-model for the formula A.

Example 2.13
Consider now a model structure M3 = (Z, I3 ) and the formula A: ∃xP (x, c).
We define an interpretation I3 in the set of all integers Z exactly as the inter-
pretation I1 , i.e. we put PI3 :< and cI3 : 0.

In this case we get AI3 : ∃x∈Z x < 0 and AI3 is obviously a true mathematical
statement. The formula A is true under the interpretation I3 in M3 (A is
satisfied, true in M3 ). We write it symbolically as

M3 |= ∃xP (x, c).

M3 is yet another model for the formula A.

When a formula is not a closed (sentence) thing get more complicated. Given
a model structure M = (U, I) a formula can be satisfied (i.e. true) for some
values in the universe U and not satisfied (i.e. false) for the others.

Example 2.14
Consider the following formulas: 1. A1 : R(x, y), 2. A2 : ∀yR(x, y), 3. A3 :
∃x∀yR(x, y). We define a model structure M = (N, I) where R is interpreted
as a relation ≤ defined in the set N of all natural numbers, i.e. we put RI :≤.

In this case we get the following.

1. A1I : x ≤ y and A1 : R(x, y) is satisfied in M = (N, I) by all n, m ∈ N such
that n ≤ m.
2. A2I : ∀y∈N x ≤ y and A2 : ∀yR(x, y) is satisfied in M = (N, I) only by the
natural number 0.
3. A3I : ∃x∈N ∀y∈N x ≤ y asserts that there is a smallest natural number and
A3 is a true sentence in M = (N, I), i.e. M is a model for A3 .
Observe that changing the universe of M = (N, I) to the set of all integers Z, we
get a different a model structure M1 = (Z, I). In this case A3I : ∃x∈Z ∀y∈Z x ≤ y
asserts that there is a smallest integer and A3 is a false sentence in M1 , i.e. M1
is a counter-model for A3 .
We want predicate language tautologies to have the same property as the propo-
sitional, namely to be always true. In this case, we intuitively agree that it
means that we want predicate tautologies to be formulas that are true under
any interpretation in any possible universe.

59
A rigorous definition of the predicate tautology is provided in a later chapter on
Predicate Logic. We construct it in the following steps.
1. We first define formally the notion of interpretation I of symbols of calL in
a set U 6= ∅ i.e. the model structure M = (U, I) for the predicate language L.
2. Then we define formally a notion ” a formula A of L a is true (valid) in
M = (U, I)”. We write it symbolically
M |= A
and call the model structure M = (U, I) a model for A.
3. We define a notion ”A is a predicate tautology” as follows.

Definition 2.8
For any formula A of predicate language L,
A is a predicate tautology (valid formula) if and only if M |= A for all
model structures M = (U, I) for L.

4. We get immediately from the above definition 2.8 of a following definition of

a notion ” A is not a predicate tautology”.

Definition 2.9
For any formula A of predicate language L,
A is not a predicate tautology if and only if there is a model structure
M = (U, I) for L, such that M 6|= A.
We call such model structure M a counter-model for A.

The definition 2.9 says: to prove that A is not a predicate tautology one has to
show a counter- model. It means one has to show a non-empty set U and define
an interpretation I, such that we can prove that AI is a false.
We use terms predicate tautology or valid formula instead of just saying a tau-
tology in order to distinguish tautologies belonging to two very different lan-
guages. For the same reason we usually reserve the symbol |= for propositional
case. Sometimes symbols |=p or |=f are used to denote predicate tautologies,
where ”p” stands for ”predicate” and ”f” stands ”first order”. The predicate
tautologies are also called laws of quantifiers and we will use both terms for
them.
Here are some examples of predicate tautologies and counter models for formulas
that are not tautologies.
For any formula A(x) with a free variable x:
|=p (∀x A(x) ⇒ ∃x A(x)). (2.53)
Observe that (2.53) represents an infinite number of formulas. It is a tautology
for any formula A(x) of L with a free variable x.
The inverse implication to (2.53) is not a predicate tautology.

60
 6|=p (∃x A(x) ⇒ ∀x A(x)) (2.54)
To prove (2.54) we have to provide an example of a concrete formula A(x) and
construct a counter-model M = (U, I) for the formula F : (∃x A(x) ⇒ ∀x A(x)).
Let A(x) be an atomic formula P (x, c). We take as M = (N, I) for N set of
natural numbers and PI :<, cI : 3. The formula F becomes an obviously false
mathematical statement FI : (∃n∈N n < 3 ⇒ ∀n∈N n < 3).
Observe that we have to be very careful when we deal with quantifiers with
restricted domain. The most basic predicate tautology (2.53) fails when we
use the quantifiers with restricted domain.

Example 2.15
Show that
6|=p (∀B(x) A(x) ⇒ ∃B(x) A(x)). (2.55)

Observe that (2.55) means that corresponding proper formula F of L obtained by

the restricted quantifiers transformations rules (8.18), (8.19) is not a predicate
tautology, i.e.

6|=p (∀x(B(x) ⇒ A(x)) ⇒ ∃x(B(x) ∩ A(x))). (2.56)

We construct a counter-model M for (2.56) as follows. We take M = (N, I)

where N is the set of real numbers, B(x), A(x) are atomic formulas Q(x, c), P (x, c)
and the interpretation I is defined as QI :<, PI :>, cI : 0. The formula F of
(2.56) becomes a mathematical statement

FI : (∀n∈N (n < 0 ⇒ n > 0) ⇒ ∃n∈N (n < 0 ∩ n > 0)).

FI is a false because the statement n < 0 is false for all natural numbers and
F ⇒ B is a true implication for any logical value of B, so ∀n∈N (n < 0 ⇒ n > 0)
is a true statement and ∃n∈N (n < 0 ∩ n > 0) is obviously false.
Restricted quantifiers law corresponding to the predicate tautology (2.53)
is:
|=p (∀B(x) A(x) ⇒ (∃x B(x) ⇒ ∃B(x) A(x))). (2.57)
We remind that (2.57) means that corresponding proper formula of L obtained
by the restricted quantifiers transformations rules (8.18), (8.19) is a predicate
tautology, i.e.

|=p (∀x(B(x) ⇒ A(x)) ⇒ (∃x B(x) ⇒ ∃x (B(x) ∩ A(x)))) (2.58)

.
Another basic predicate tautology called a dictum de omni law is: For any
formulas A(x) with a free variable x ∈ V AR,

61
 |=p (∀x A(x) ⇒ A(y)), (2.59)
where y ∈ V AR and A(y) is a result of substitution of y for all free occurrences
of x in A(x) (if any) and y is free for x in A(x), what means that no occurrence
of a variable y becomes a bound occurrence in A(y). Restricted quantifiers law
corresponding to the dictum de omni law (2.59) is:
|=p (∀B(x) A(x) ⇒ (B(y) ⇒ A(y))), (2.60)
where y ∈ V AR satisfies the same condition as in (2.59).
Observe that we say A is restricted quantifiers law, or A is restricted quantifiers
tautology as a shorthand to formally saying that a formula obtained from A by
the transformations rules (8.18), (8.19) is a predicate tautology.
A more general version of (2.59) is:
|=p (∀x A(x) ⇒ A(t)), (2.61)
where t is a term and A(t) is a result of substitution of t for all free occurrences
of x in A(x) and t is free for x in A(x), what means that no occurrence of a
variable in t becomes a bound occurrence in A(t).
Here is another important tautology, called a generalization law.
|=p (A(x) ⇒ ∀x A(x)). (2.62)
The next important laws are the Distributivity Laws.
1. Distributivity of existential quantifier over conjunction holds only on one
direction, namely the following is a predicate tautology.

|=p (∃x (A(x) ∩ B(x)) ⇒ (∃xA(x) ∩ ∃xB(x))) (2.63)

where A(x), B(x) are any formulas with a free variable x. The inverse implica-
tion is not a predicate tautology, i.e. there are formulas A(x), B(x) with a free
variable x. such that
6|=p ((∃xA(x) ∩ ∃xB(x)) ⇒ ∃x (A(x) ∩ B(x))). (2.64)
To prove (2.64) means that we have to find a concrete formulas A(x), B(x) ∈ F
and a model structure M = (U, I), where the interpretation I is the interpreta-
tion of all predicate, functional, and constant symbols in A(x), B(x), such that
it is a counter- model for the formula
F : ((∃xA(x) ∩ ∃xB(x)) ⇒ ∃x (A(x) ∩ B(x))).
Take M = (R, I) where R is the set of real numbers, and A(x), B(x) be atomic
formulas Q(x, c), P (x, c). We define the interpretation I as QI :>, PI :<, cI : 0.
The formula F becomes an obviously false mathematical statement
FI : ((∃x∈R x > 0 ∩ ∃x∈R x < 0) ⇒ ∃x∈R (x > 0 ∩ x < 0)).

62
2. Distributivity of universal quantifier over disjunction holds only on one di-
rection, namely the following is a predicate tautology for any formulas A(x), B(x)
with a free variable x.

|=p ((∀xA(x) ∪ ∀xB(x)) ⇒ ∀x (A(x) ∪ B(x))). (2.65)

The inverse implication is not a predicate tautology, i.e.there are formulas A(x), B(x)
with a free variable x. such that

6|=p (∀x (A(x) ∪ B(x)) ⇒ (∀xA(x) ∪ ∀xB(x))). (2.66)

It means that we have to find a concrete formula A(x), B(x) ∈ F and a model
structure M = (U, I) that is a counter- model for the formula

F : (∀x (A(x) ∪ B(x)) ⇒ (∀xA(x) ∪ ∀xB(x))).

Take M = (R, I) where R is the set of real numbers, and A(x), B(x) be atomic
formulas Q(x, c), R(x, c). We define QI :≥, RI :<, cI : 0. The formula F
becomes an obviously false mathematical statement

FI : (∀x∈R (x ≥ 0 ∪ x < 0) ⇒ (∀x∈R x ≥ 0 ∪ ∀x∈R x < 0)).

The most frequently used laws of quantifiers have a form of a logical equivalence,
symbolically written as ≡. This not a new logical connective. This is a very
useful symbol. It says that two formulas always have the same logical value,
hence it can be used in the same way we use the equality symbol =. Formally
we define it as follows.

Definition 2.10
For any formulas A, B ∈ F of the predicate language L,

A≡B if and only if |=p (A ⇔ B).

We have also a similar definition for our propositional language L (definition

2.1) and propositional tautology (definition 2.3).
Equational Laws for Quantifiers
De Morgan
For any formula A(x) ∈ F with a free variable x,

¬∀xA(x) ≡ ∃x¬A(x), ¬∃xA(x) ≡ ∀x¬A(x). (2.67)

Definability
For any formula A(x) ∈ F with a free variable x,

∀xA(x) ≡ ¬∃x¬A(x), ∃xA(x) ≡ ¬∀x¬A(x). (2.68)

63
Renaming the Variables
Let A(x) be any formula with a free variable x and let y be a variable that does
not occur in A(x). Let A(x/y) be a result of replacement of each occurrence of
x by y, then the following holds.

∀xA(x) ≡ ∀yA(y), ∃xA(x) ≡ ∃yA(y). (2.69)

Alternations of Quantifiers
Let A(x, y) be any formula with a free variables x and y.

∀x∀y (A(x, y) ≡ ∀y∀x (A(x, y) (2.70)

∃x∃y (A(x, y) ≡ ∃y∃x (A(x, y) (2.71)

Introduction and Elimination Laws

If B is a formula such that B does not contain any free occurrence of x, then
the following logical equivalences hold.

∀x(A(x) ∪ B) ≡ (∀xA(x) ∪ B), (2.72)

∃x(A(x) ∪ B) ≡ (∃xA(x) ∪ B), (2.73)

∀x(A(x) ∩ B) ≡ (∀xA(x) ∩ B), (2.74)
∃x(A(x) ∩ B) ≡ (∃xA(x) ∩ B), (2.75)
∀x(A(x) ⇒ B) ≡ (∃xA(x) ⇒ B), (2.76)
∃x(A(x) ⇒ B) ≡ (∀xA(x) ⇒ B), (2.77)
∀x(B ⇒ A(x)) ≡ (B ⇒ ∀xA(x)), (2.78)
∃x(B ⇒ A(x)) ≡ (B ⇒ ∃xA(x)). (2.79)
Distributivity Laws

Let A(x), B(x) be any formulas with a free variable x.

Distributivity of universal quantifier over conjunction.

∀x (A(x) ∩ B(x)) ≡ (∀xA(x) ∩ ∀xB(x)) (2.80)

Distributivity of existential quantifier over disjunction.

∃x (A(x) ∪ B(x)) ≡ (∃xA(x) ∪ ∃xB(x)) (2.81)

We also define the notion of logical equivalence ≡ for the formulas of the —tex-
titpropositional language (definition 2.1) and its semantics.

Definition 2.11
For any formulas A, B ∈ F of the propositional language L,

A≡B if and only if |= (A ⇔ B).

64
Moreover, we prove that any substitution of propositional tautology by a for-
mulas of the predicate language is a predicate language tautology. The same
holds for the logical equivalence. In particular, we transform the propositional
Implication and Double Negation tautologies (2.48), (2.49) into the following
predicate equivalences.
For any formulas A, B of the predicate language L,
(A ⇒ B) ≡ (¬A ∪ B), (2.82)
¬¬A ≡ A (2.83)
We use (2.82) and (2.83) to prove the following De Morgan Laws for restricted
quantifiers.

Restricted De Morgan
For any formulas A(x), B(x) ∈ F with a free variable x,
¬∀B(x) A(x) ≡ ∃B(x) ¬A(x), ¬∃B(x) A(x) ≡ ∀B(x) ¬A(x). (2.84)

Here is a poof of first equality. The proof of the second one is similar and is left
as an exercise.
¬∀B(x) A(x) ≡ ¬∀x (B(x) ⇒ A(x)) ≡ ¬∀x (¬B(x)∪A(x)) ≡ ∃x ¬(¬B(x)∪A(x))
≡ ∃x (¬¬B(x) ∩ ¬A(x)) ≡ ∃x (B(x) ∩ ¬A(x)) ≡ ∃B(x) ¬A(x)).

We also transform the propositional Distributivity tautologies (2.45), (2.46) into

the following predicate equivalences.
For any formulas A, B of the predicate language L,
(A ∩ (B ∪ C)) ≡ ((A ∩ B) ∪ (A ∩ C)), (2.85)
(A ∪ (B ∩ C)) ≡ ((A ∪ B) ∩ (A ∪ C)) (2.86)
We use (2.85) and (2.86) to prove the following Distributivity Laws for restricted
quantifiers.

Restricted Distributivity Laws

We generalize the Introduction and Elimination Laws (10.32), (11.45), (11.46),
(10.38) to the case of the the restricted quantifiers as folows.

Restricted Introduction and Elimination Laws

If B is a formula such that B does not contain any free occurrence of x, then
the following logical equivalences hold for any formulas A(x), B(x), C(x).
∀C(x) (A(x) ∪ B) ≡ (∀C(x) A(x) ∪ B), (2.87)

65
 ∃C(x) (A(x) ∩ B) ≡ (∃C(x) A(x) ∩ B), (2.88)
∀C(x) (A(x) ⇒ B) ≡ (∃C(x) A(x) ⇒ B), (2.89)
∀C(x) (B ⇒ A(x)) ≡ (B ⇒ ∀C(x) A(x)). (2.90)
The proofs are similar to the proof of the restricted de Morgan Laws.
The similar generalization of the other Introduction and Elimination Laws (11.43),
(11.44), (11.47), (10.39) fails. We can easily follow Example 2.15 and construct
proper counter-models proving the following.

∃C(x) (A(x) ∪ B) 6≡ (∃C(x) A(x) ∪ B),

∀C(x) (A(x) ∩ B) 6≡ (∀C(x) A(x) ∩ B),

∃C(x) (A(x) ⇒ B) 6≡ (∀C(x) A(x) ⇒ B),
∃C(x) (B ⇒ A(x)) 6≡ (B ⇒ ∃xA(x)).

Nevertheless it is possible to correctly generalize them all as to cover quantifiers

with restricted domain. We show it in a case of (11.43) and leave the other
cases to the reader as an exercise.

Example 2.16
The restricted quantifiers version of (11.43) is the following.

∃C(x) (A(x) ∪ B) ≡ (∃C(x) A(x) ∪ (∃x C(x) ∩ B)). (2.91)

We derive (8.74) as follows.

∃C(x) (A(x) ∪ B) ≡ ∃x(C(x) ∩ (A(x) ∪ B)) ≡ ∃x((C(x) ∩ A(x)) ∪ (C(x) ∩ B))

≡ (∃x(C(x) ∩ A(x)) ∪ ∃x(C(x) ∩ B)) ≡ (∃C(x) A(x) ∪ (∃x C(x) ∩ B)).

We leave it as an exercise to specify and write references to transformation or

equational laws used at each step of our computation.

2.6 Homework Problems

Propositional Languages

1. For the following sentences write their corresponding formulas.

(a) If Mr. Smith is happy, Mrs. Smith is not happy, and if If Mr. Smith
is not happy, Mrs. Smith is not happy.

66
 (b) If John doesn’t know logic, then if he knows logic, he was born in the
12th century.
(c) If from the fact that all sides of a triangle ABC are equal we can
deduce that all angles of the triangle ABC are equal and all angles
of the triangle ABC are not equal, then all sides of a triangle ABC
are equal.
(d) If it is not the fact that a line L is parallel to a line M or a line P is
not parallel the line M, then the line L is not parallel to the line M
or the line P is parallel the line M.
(e) If a number a is divisible by 3 and by 5, then from the fact that it is
not divisible by 3, we can deduce that it is also not divisible by 5.
2. For each of the following formulas write 3 corresponding natural language
sentences.

(a) (a ⇒ (¬a ∩ b))

(b) (((p ∪ q) ∩ ¬p) ⇒ q)
(c) ((a ⇒ b) ⇒ (a ⇒ (b ∪ c)))
(d) ¬(p ∩ (¬p ∩ q))
(e) ((a ⇒ ((¬b ∩ b) ⇒ c))

3. Consider a following set S

S = {(a ∩ b) ⇒ ¬(a ∪ b), ((¬a) ⇒ (¬a ⇒ b)), (¬a ⇒ (a ∩ ¬b))}.

1. Determine which of the elements of S are, and which are not well
formed formulas (wff) of L = (A, F).
2. If A ∈ S is not a formula, i.e if A 6∈ F re-write it as a correct formula
and write in the natural language what it says.
4. Write a full definition of a propositional language that uses Hilbert set of
connectives. Give four examples of well form formulas of this language.
List next to them corresponding formulas of our propositional language
L.
5. Write a full definition of a propositional language L that uses Lukasiewicz
set of connectives. Give 4 examples of well form formulas of this language.
Give 4 examples of well form formulas of this language. List next to them
corresponding formulas of our propositional language L.

Propositional Semantics

1. Given a formula A: (((a ∩ b) ∪ ¬c) ⇒ b). Evaluate the logical value of A for
the following sets of logical values of its basic components, i.e. variables
a, b: 1. a=T, b=F, c=F and 2. a=F, b=T, c=T.

67
 2. Given a formula A: (((a ⇒ ¬b) ∪ b) ⇒ a). Evaluate the logical value of A
for all possible logical values of its variables.

3. Given a formula A: (((a ↓ ¬b) ∪ b) ↑ a). Evaluate the logical value of A

for the following sets of logical values of its variables: 1. a=T, b=F and
2. a=F, b=F.

4. Find and prove an equality defining implication in terms of disjunction

and negation.

5. Find and prove an equality defining conjunction in terms of disjunction

and negation.

6. Find and prove an equality and a table defining conjunction in terms of

implication and negation.

7. Prove that ∪ can be defined in terms of ⇒ alone.

8. Find and prove an equality defining ⇒ in terms of ↑.

9. Define ⇒ in terms of ¬ and ∩.

10. Find an equality defining ⇒ in terms of ↓.

11. Define ∩ in terms of ⇒ and ¬.

12. Find an equality defining ∩ in terms of ↓ alone.

Propositional Tautologies

1. Prove 5 propositional tautologies of your choice.

2. Prove that a formula (((¬A ⇒ B) ∩ (B ⇒ C)) ⇒ ((A ∩ B) ⇒ C) is not a

propositional tautology.

3. Show that ”If a number is divisible by 3 and by 5, then from the fact that
it is not divisible by 3, we can deduce that it is also not divisible by 5” is
always a true statement.

4. Determine whether the following arguments logically correct by represent-

ing each sentence as propositional formula and checking whether the con-
clusion is logically implied by the conjunction of the assumptions. To do
this assign logical value T to each formula representing assumption and
F to the formula representing the conclusion, and determine whether a
contradiction results.

(a) If John is a Communist, John is atheist. John is an atheist. Hence

John is a Communist.

68
 (b) If the temperature and air pressure remained constant, there was a
rain.The temperature did remain constant. Therefore, if there was a
rain then the air pressure did not remain constant.
(c) If a = 0 or b = 0, then ab = 0. But ab 6= 0. Hence a 6= 0 or b 6= 0.
(d) If a = 0 and b = 0, then ab = 0. But ab 6= 0. Hence a 6= 0 or b 6= 0.

Predicate Language Description and Application to AI

1. Given the following formulas A1 − A5 of L.

A1 = R(x, y, g(c, x)), A2 = ∃xP (x, f (x, y)), A3 = ∃dR(x, y, g(c, d)),

A4 = ∀z(f (x, P (c, y)), A5 = ∃yP (x, f (c, y)) ∪ ∀yP (x, f (c, y)).

(a) Indicate whether they are, or are not well formed formulas of F. For
those which are not in F write a correct formula.
(b) For each correct, or corrected formula identify all components: con-
nectives, quantifiers, predicate and function symbols, and list all its
terms.
(c) For each formula identify its s free and bound variables. State which
are open and which are closed formulas (sentences), if any.

2. For the following mathematical statements write their corresponding for-

mulas of predicate language L.

(a) ∀n>1 (n + 3 < 8 ∪ ∃x∈R x + n > 8)

(b) ∀x∈R ∃n∈N (x + n > 0 ⇒ ∃m∈N (m = x + n))
(c) If all natural numbers are smaller then zero, then the sum of any two
integers is smaller then zero.
(d) For all natural numbers The following implication holds for all natural
numbers: if n > 0, then there is a real number x, such that n + x = 0
or there is an integer m, such that m > 0.

3. For the following natural language statements write their corresponding

formulas of predicate language L.

(a) Anyone who is persistent can learn logic.

(b) Some people are witty only if they are drunk.
(c) John hates everybody who does not hate himself.
(d) Everybody loves somebody and no one loves everybody.

4. For the following natural language statements write their corresponding

formulas of AI language of your choice.

69
 (a) Anyone who is lazy can’t learn logic.
(b) Some people are happy only if they sing.
(c) John likes everybody who does not like Mary.
(d) Everybody with green eyes likes John.

5. For each of the following formulas (some with restricted quantifiers) write
two corresponding natural language sentences.

(a) ∀x(P (x) ⇒ ∃yQ(x, y)).

(b) ∀x∃y(P (x) ∩ ¬Q(x, y)).
(c) ∀A(x) ∃A(y) B(y).
(d) ∃P (x) ∀N (x) R(x, y).

Predicate Semantics

1. For each of the formulas and each model structure M indicate for what
values the formula is satisfied (if it contains free variables) or whether M
is its model or counter-model (if it is a closed formula. i.e. a sentence).

Formulas are:
(a) P (f (x, y), c)
(b) P (x, y) ⇒ P (y, x)
(c) ∀x∀y∀z((P (x, y) ∩ P (y, z)) ⇒ P (x, z))

Model structures M are:

M1 = (N, I), for N set of natural numbers and PI :≥, fI : multiplication,
and cI : 2
M2 = (Z, I), for Z set of integers and PI :=, fI : +, and cI : 2
M3 = (2Z , I), for 2Z the set of all subsets of Integers, and PI : ⊆, fI : ∩,
and cI : ∅

2. For a given model structure M and corresponding closed formulas deter-

mine for each of them whether M is its model or a counter-model.

(a) Model structure is M = (N, I), for N set of natural numbers and
PI :=, gI : +, fI : multiplication, and cI : 0, dI : 1.
Formulas are:
A1 : ∀x∃y(P (x, g(y, y)) ∪ P (x, g(g(y, y), d)))
A2 : ∀x∀y(P (f (x, y), c) ⇒ (P (x, c) ∪ P (y, c)))
A3 : ∃y P (g(y, y), d)
(b) Model structure is M = (Z, I), for Z set of integers and PI :=, fI : +,

70
 Formulas are:
A1 : ∀x∀y P (f (x, y), f (y, x)))
A2 : ∀x∀y P (f (x, y), y)
3. Prove that the following formulas are not predicate tautologies, i.e. find
for each of them a counter-model M.

(a) (∃x A(x) ⇒ ∀x A(x))

(b) (∀x∃y A(x, y) ⇒ ∃x∀y A(x, y))
(c) (∃x∃y A(x, y) ⇒ ∃y A(y, y))
(b) (∀x∃y A(x, y) ⇒ ∃y A(y, y))
(d) (∃x (A(x) ⇒ B(x)) ⇒ (∃x A(x) ⇒ ∃x B(x)))
(e) (∃x (A(x) ⇒ B(x)) ⇒ (∃x A(x) ⇒ ∃x B(x)))

4. Transform the following formulas with restricted quantifiers into a proper

formulas of the predicate language L.

(a) (∀A(x) ∃B(x) C(x) ⇒ ¬∃B(x) ¬C(x))

(b) (∃A(x) (∀B(y) C(y) ⇒ ¬C(x))
(c) (∀A(y) ∃B(x) D(x, y) ⇒ ¬∃B(x) C(x))
(d) ∀A(x) (∃B(x) C(x) ∪ ¬∀A(x) C(x))

5. Use proper Equational Laws for Quantifiers to prove that the following
Restricted Introduction and Elimination Laws hold for any formulas A(x),
B(x), C(x), and B, where B does not contain any free occurrence of x.
(a) ∃C(x) (A(x) ∩ B) ≡ (∃C(x) A(x) ∩ B)
(b) ∀C(x) (A(x) ⇒ B) ≡ (∃C(x) A(x) ⇒ B)
(c) ∀C(x) (B ⇒ A(x)) ≡ (B ⇒ ∀C(x) A(x))

71
72
Chapter 3

Propositional Semantics:
Classical and Many Valued

3.1 Formal Propositional Languages

We define here a general notion of a propositional language. We obtain, as spe-
cific cases, various languages for propositional classical logic as well as languages
for many non-classical logics.
We assume that any propositional language contains a countably infinite set
V AR of propositional variables. What distinguishes one propositional language
from the other is the choice of its set CON of propositional connectives. We
adopt a notation LCON for a propositional language with the set CON of logical
connectives. For example, the language L{¬} denotes a propositional language
with only one connective ¬. The language L{¬,⇒} denotes that the language
has only two connectives ¬ and ⇒ adopted as propositional connectives. All
propositional languages share the general way their sets of formulas are formed.
Theoretically one can use any symbols to denote propositional connectives. But
there are some preferences, as connectives have a long history and intuitive
meaning. The formal meaning, i.e. a semantics for a given language is discussed
and defined in the next section.
Different semantics can share the same language. For example, the language
L{¬,∩,∪,⇒} is used as a propositional language for classical logic semantics, in-
tuitionistic logic semantics, and many valued logics semantics. It is also possible
for several languages to share the same semantics. The classical propositional
logic is the best example of such situation. We will prove in the section 3.4 that
the languages:
L{¬⇒} , L{¬∩} , L{¬∪} , L{¬,∩,∪,⇒} , L{¬,∩,∪,⇒,⇔} ,

73
and even two languages with only one binary propositional connectives, denoted
usually by ↑ and ↓, respectively, i.e languages L{↑} , L{↓} all share the same
semantics characteristic for the classical propositional logic.
The connectives have well established symbols and names, even if their semantics
can differ. We use names negation, conjunction, disjunction, implication and
equivalence (biconditional) for ¬, ∩, ∪, ⇒, ⇔, respectively. The connective ↑ is
called alternative negation and A ↑ B reads: not both A and B. The connective
↓ is called joint negation and A ↓ B reads: neither A nor B.
Other most common propositional connectives are probably modal connectives
of possibility and necessity. Standard modal symbols are  for necessity and
♦ for possibility. We will also use symbols C and I for modal connectives of
possibility and necessity, respectively.
A formula CA, or ♦A reads: it is possible that A , A is possible, and a formula
IA, or A reads: it is necessary that A, A is necessary.
A motivation for notation C and I arises from topological semantics for modal
S4 and S5 logics. C becomes equivalent to a set closure operation, and I becomes
equivalent to a set interior operation.
The symbols ♦, C and , I are not the only symbols used for modal connectives.
Other symbols include N for necessity and P for possibility. There is also
a variety of modal logics created by computer scientists, all with their set of
symbols and motivations for their use and their semantics. The modal logics
extend the classical logic and hence their language is for example L{,♦,¬,∩,∪,⇒} .
Knowledge logics also extend the classical logic. Their languages add to the
classical connectives a new knowledge connective, often denoted denoted by K.
The formula KA reads: it is known that A , A is known. The language of a
knowledge logic is for example L{K, ¬,∩,∪,⇒} .
Autoepistemic logics use a believe connective, often denoted by B. The formula
BA reads: it is believed that A. They also extend the classical logic and hence
their language is L{B, ¬,∩,∪,⇒} .
Temporal logics add temporal connectives to the set of classical propositional
connectives. For example some of them use connectives (operators, as they are
often called) F, P, G, and H to denote the following intuitive readings. F A reads
A is true at some future time, P A reads A was true at some past time, GA
reads A will be true at all future times, and HA reads A has always been true
in the past. In order to take account of this variation of truth-values over time,
some formal semantics were created, and many more will be created.
It is possible to create connectives with more then one or two arguments, but
we allow here only one and two argument connectives, as logics which will be
discussed here use only those two kind of connectives.
We adopt the following definition, common to all propositional languages con-

74
sidered in our propositional logics investigations.

Definition 3.1 (Propositional Language)

By a propositional language with a set CON of propositional connectives
we understand a pair
LCON = (A, F). (3.1)
A is a called propositional alphabet and F is called a set of propositional
formulas of the language LCON . The alphabet A, the set CON of propositional
connectives, and the set F of propositional formulas are defined as follows.

1. Alphabet A
The alphabet A = V AR ∪ CON ∪ P AR, where VAR, CON, PAR are all disjoint
sets and VAR, CON are non-empty sets. VAR is countably infinite and is called
a set of propositional variables; we denote elements of VAR by a, b, c, ... etc,
(with indices if necessary).
CON is a finite set of propositional connectives, P AR is a set of auxil-
iary symbols. We assume that P AR 6= ∅ and contains two elements (, ) called
parentheses, i.e. P AR = {(, )}. The set PAR may be empty, for example of a
case of Polish notation, but we assume that it contains two parenthesis as to
make the reading of formulas more natural and uniform.

2. Propositional connectives CON

We assume that the set CON is non empty and finite. We specify it for specific
cases (specific logics). It is possible to consider languages with connectives
which have more then one or two arguments, nevertheless we restrict ourselves
to languages with one or two argument connectives only.
We assume that
CON = C1 ∪ C2
where C1 is a finite set (possibly empty) of unary connectives, C2 is a finite
set (possibly empty) of binary connectives of the language LCON .

2. Set F of formulas
The set F is built recursively from the elements of the alphabet A, i.e. F ⊆ A∗ ,
where A∗ is the set of all finite sequences (strings) form from elements of A and
is defined as follows.
The set F of all formulas of a propositional language LCON is the smallest
set, such that the following conditions hold:

(1) V AR ⊆ F;

75
(2) if A ∈ F, 5 ∈ C1 i.e 5 is an one argument connective, then 5A ∈ F;

(3) if A, B ∈ F, ◦ ∈ C2 i.e ◦ is a two argument connective, then

(A ◦ B) ∈ F.

The elements of the set V AR ⊆ F are called atomic formulas. The set F is
also called a set of all well formed formulas (wff) of the language LCON .

The alphabet A is countably infinite and consequently the set A∗ of all finite
sequences of elements of A is also countably infinite. By definition, F ⊆ A∗ ,
hence the set F is also countably infinite. We state as separate fact.

Fact 3.1 For any propositional language LCON = (A, F), the set F of for-
mulas is countably infinite. We hence consider here only infinitely countable
languages.

Observation 3.1
When defining a language LCON we choose not only the propositional connec-
tives but also the symbols denoting them.
For example, L1 = L{¬} and L2 = L{∼} are two different propositional lan-
guages both with negation as the only connective.

The choice of appropriate well established symbols for logical connectives de-
pends on a personal preferences of books’ authors and creators of different logics.
One can find a variety of them in the literature. We presented some historical
choices in the chapter 2.

Example 3.1
Let L1 = L{¬} and L2 = L{∼} . The formulas of both languages L1 , L2 are
propositional variables or multiple negations of of a propositional variable.

The strings a, ¬b, ¬¬b, ¬¬¬a are well formed formulas of L1 . The corresponding
formulas of L2 are a, ∼ b, ∼∼ b, ∼∼∼ a.
Observe that the strings (¬a), ¬, ¬(¬a), ¬(a), (∼ a), ¬, ∼ (∼ a) ∼ (a) are not
well formed formulas of neither of the languages L1 , L2 .

We adopt the general definition of the set F of formulas of LCON to for example
the language L{∼} as follows.

Example 3.2

76
The set F of all formulas of a propositional language L{∼} is the smallest
set, such that the following conditions hold:
(1) V AR ⊆ F (atomic formulas);
(2) if A ∈ F, then ∼ A ∈ F.

Example 3.3
Consider now LCON for the set of connectives CON = {¬}∪{⇒}, where ¬ ∈ C1
and ⇒ ∈ C2 . It means that we defined a language L = L{¬,⇒} .

By the initial recursive step we get for any a ∈ V AR, a ∈ F. By the recursive
step and its repetition we get for example that ¬a ∈ F, ¬¬a ∈ F, ¬¬¬a ∈ F, ...
etc., i.e. get all formulas from the the example 5.10 language L1 . But also we
also get that (a ⇒ a), (a ⇒ b), ¬(a ⇒ b), (¬a ⇒ b), ¬((a ⇒ a) ⇒ ¬(a ⇒ b))....
etc. are all in F and infinitely many others.
Observe that (¬(a ⇒ b))), a ⇒ b, (a ⇒) are not in F.

Example 3.4
Consider L = LCON for C1 = {¬, P, N }, C2 = {⇒}. If we understand P , N
as a possibility and necessity connectives, the obtained language is called a
modal language with only negation as non-modal connective.

The set of formulas F of L contains all formulas from example 5.11, but also
formulas N a, ¬P a, P ¬a, (N ¬b ⇒ P a), ¬P ¬a, ((N ¬b ⇒ P a) ⇒ b), .... etc.

We adopt the general definition of the set F of formulas of LCON to for example
the modal language L{¬,P,N,⇒} as follows.

Example 3.5
The set F of all formulas of a propositional language L{¬,P,N,⇒} is the
smallest set, such that the following conditions hold:
(1) V AR ⊆ F (atomic formulas);
(2) if A ∈ F, then ¬A, P A, N A ∈ F;
(3) if A, B ∈ F, then (A ⇒ B) ∈ F.

We introduce now formal definitions of basic syntactical notions of a main con-

nective, a sub-formula of a given formula, and of a degree of a given formula.

Definition 3.2 (Main Connective)

Given a language LCON = (A, F).

77
For any connectives 5 ∈ C1 and ◦ ∈ C2 ,
5 is called a main connective of 5A ∈ F and
◦ is a main connective of (B ◦ C) ∈ F.

Observe that it follows directly from the definition of the set of formulas that
for any formula C ∈ F, exactly one of the following holds: C is atomic, or there
is a unique formula A and a unique unary connective 5 ∈ C1 , such that C
is of the form 5A, or here are unique formulas A and B and a unique binary
connective ◦ ∈ C2 , such that C is (A ◦ B). We have hence proved the following.

Observation 3.2
For any formula A ∈ F, A is atomic or has a unique main connective.

Example 3.6
The main connective of (a ⇒ ¬N b) is ⇒. The main connective of N (a ⇒
¬b) is N . The main connective of ¬(a ⇒ ¬b) is ¬ The main connective of
(¬a ∪ ¬(a ⇒ b)) is ∪.

Definition 3.3
We define a notion of direct a direct sub-formula as follows: 1. Atomic
formulas have no direct sub-formulas. 2. A is a direct sub-formula of a formula
5A, where 5 is any unary connective. 3. A, B are direct sub-formulas of a
formula (A ◦ B) where ◦ is any binary connective.

Directly from the definition 3.3 we get the following.

Observation 3.3
For any formula A, A is atomic or has exactly one or two direct sub-formulas
depending on its main connective being unary or binary, respectively.

Example 3.7
The formula (¬a ∪ ¬(a ⇒ b)) has exactly ¬a and ¬(a ⇒ b) as direct sub-
formulas.

Definition 3.4
We define a notion of a sub-formula of a given formula in two steps. 1. For
any formulas A and B, A is a proper sub-formula of B if there is sequence of
formulas, beginning with A, ending with B, and in which each term is a direct
sub-formula of the next. 2. A sub-formula of a given formula A is any proper
sub-formula of A, or A itself.

78
The formula (¬a ∪ ¬(a ⇒ b)) has ¬a and ¬(a ⇒ b) as direct sub-formula. The
formulas ¬a and ¬(a ⇒ b) have a and (a ⇒ b) as their direct sub-formulas,
respectively. The formulas ¬a, ¬(a ⇒ b), a and (a ⇒ b) are all proper sub-
formulas of the formula (¬a ∪ ¬(a ⇒ b)) itself. Atomic formulas a and b are
direct sub-formulas of (a ⇒ b). Atomic formula b is a proper sub-formula of ¬b.

Example 3.8
The set of all sub-formulas of

(¬a ∪ ¬(a ⇒ b))

consists of (¬a ∪ ¬(a ⇒ b)), ¬a, ¬(a ⇒ b), (a ⇒ b), a and b.

Definition 3.5 (Degree of a formula)

By a degree of a formula we mean the number of occurrences of logical connec-
tives in the formula.

The degree of (¬a ∪ ¬(a ⇒ b)) is 4. The degree of ¬(a ⇒ b)) is 2. The degree
of ¬a is 1. The degree of a is 0.
Note that the degree of any proper sub-formula of A must be one less than the
degree of A. This is the central fact upon mathematical induction arguments
are based. Proofs of properties formulas are usually carried by mathematical
induction on their degrees.

Example 3.9
Given a formula A : (¬I¬a ⇒ (¬Ca ∪ (Ia ⇒ ¬Ib))).
1. The language to which A belongs is a modal language L{¬,C,C,∪,∩,⇒} with
the possibility connective C and necessity connective C. Both of them are one
argument connectives.
2. The main connective of A is ⇒, the degree of A is 11.
3. All sub-formulas of A of the degree 0 are the atomic formulas a, b. All
sub-formulas of A of the degree 1 are: ¬a, Ca, Ia, Ib.

Languages with Propositional Constants

A propositional language LCON = (A, F) is called a language with propositional

constants, when we distinguish certain constants, like symbol of truth T or
falsehood F, or other symbols as elements of the alphabet. The propositional
constants are zero-argument connectives. In this case the set CON of logical
connectives contains a a finite, non empty set of zero argument connectives C0 ,
called propositional constants, i.e. we put

CON = C0 ∪ C1 ∪ C2 .

79
The definition of the set F of all formulas of the language LCON contains now
an additional recursive step and goes as follows.
The set F of all formulas of the language LCON with propositional constants is
the smallest set built from the signs of the alphabet A, i.e. F ⊆ A∗ , such that
the following conditions hold:

(1) V AR ⊆ F (atomic formulas),

(2) C0 ⊆ F (atomic formulas),

(3) if A ∈ F, 5 ∈ C1 i.e 5 is an one argument connective, then 5A ∈ F,

(4) if A, B ∈ F, ◦ ∈ C2 i.e ◦ is a two argument connective, then

(A ◦ B) ∈ F.

Example 3.10
Let L = L{T,¬,∩} , i.e. C0 = {V }. Atomic formulas of L are all variables and
the symbol T .

The language admits formulas that involve the symbol T like T, ¬T , (T ∩ a),
(¬a ∩ ¬T ), ¬(b ∩ T ), etc... We might interpret the symbol T as a symbol of truth
(statement that is always true).

Here are some exercises and examples dealing with the formal definition of
propositional languages, syntactical correctness, and their expressiveness.

Exercise 3.1
Given a language L = L{¬, C,I,∪,∩,⇒} and the following set S.

S = {C¬a ⇒ (a ∪ b), (C(¬a ⇒ (a ∪ b))), C¬(a ⇒ (a ∪ b))}

Determine which of the elements of S are, and which are not well formed for-
mulas of L. If A ∈ S is not a correct formula write its corrected version. For
each correct or corrected formula determine its main connective, its degree and
write what it says in the natural language.

Solution
1. C¬a ⇒ (a ∪ b) is not a well formed formula. The corrected formula is
(C¬a ⇒ (a ∪ b)). Its main connective is ⇒ and the degree is 4. The corrected
formula says: If negation of a is possible, then we have a or b.
Another corrected formula is C(¬a ⇒ (a ∪ b)). Its main connective is C, the
degree is 4. The corrected formula says: It is possible that not a implies a or b.

80
2. (C(¬a ⇒ (a ∪ b))) is not a well formed formula. The correct formula is
C(¬a ⇒ (a ∪ b)). The main connective is C, the degree is 4. The formula
C(¬a ⇒ (a ∪ b)) says: It is possible that not a implies a or b .
3. The formula C¬(a ⇒ (a ∪ b)) is a correct formula. The main connective is
C, the degree is 4. The formula says: the negation of the fact that a implies a
or b is possible.

Exercise 3.2
Given a set S of formulas:

S = {((a ⇒ ¬b) ⇒ ¬a), (¬♦a ⇒ ¬a), (a ∪ ¬(a ⇒ b))}.

Define a formal language LCON to which to which all formulas in S belong, i.e.
a language determined by the set S.

Solution
Any propositional language LCON is determined by its set of connectives. The
connectives appearing in the formulas of the set S are: ⇒, ¬b, , ♦ and ∪. Hence
the required language is L{¬,,♦,∪,⇒} .

Exercise 3.3
Write down a set S1 all sub-formulas of the ♦((a ∪ ¬a) ∩ b), a set S2 all proper
sub-formulas of ¬(a ⇒ (b ⇒)).

Solution
The set S1 of all sub-formulas of ♦((a ∪ ¬a) ∩ b) is
S1 = {♦((a ∪ ¬a) ∩ b), ((a ∪ ¬a) ∩ b), (a ∪ ¬a), ¬a, b, a}
a, b are atomic sub-formulas, and ♦((a ∪ ¬a) ∩ b) is not a proper sub-formula.
The set S2 of all proper sub-formulas of ¬(a ⇒ (b ⇒ c)) is
S2 = {(a ⇒ (b ⇒ c)), (b ⇒ c), a, b, c}.

Exercise 3.4
Write the following natural language statement S:
”From the fact that it is possible that Anne is not a boy we deduce that it is not
possible that Anne is not a boy or, if it is possible that Anne is not a boy, then
it is not necessary that Anne is pretty.”
in the following two ways.
1. As a formula A1 ∈ F1 of a language L{¬, , ♦, ∩, ∪, ⇒} .

2. As a formula A2 ∈ F2 of a language L{¬, ∩, ∪, ⇒} .

81
Solution
1. We translate the statement S into a formula A1 of the modal language
L{¬, , ♦, ∩, ∪, ⇒} as follows.
Propositional variables are: a, b. The variable a denotes statement Anne is a
boy and b denotes a statement Anne is pretty.
Propositional modal connectives are: , ♦. The connective ♦ reads it is
possible that, and  reads it is necessary that.
Translation: the formula A1 is (♦¬a ⇒ (¬♦¬a ∪ (♦¬a ⇒ ¬b))).
2. We translate our statement into a formula A2 of the language L{¬, ∩, ∪, ⇒}
as follows.
Propositional variables are: a, b. The variable a denotes statement it is possible
that Anne is not a boy and b denotes a statement it is necessary that Anne is
pretty. Translation: the formula A2 is (a ⇒ (¬a ∪ (a ⇒ ¬b))).

Exercise 3.5
Write the following natural language statement S:
”For all natural numbers n ∈ N the following implication holds: if n < 0, then
there is a natural number m, such that it is possible that n + m < 0, or it is
not possible that there is a natural number m, such that m > 0”
in the following two ways.
1. As a formula A1 of a language L{¬, ∩, ∪, ⇒} .

2. As a formula A2 of a language L{¬, , ♦, ∩, ∪, ⇒} .

Solution
1. We translate the statement S into a formula A1 of the language L{¬, ∩, ∪, ⇒}
as follows.
Propositional variables are: a, b. The variable a denotes statement For all
natural numbers n ∈ N the following implication holds: if n < 0, then there is
a natural number m, such that it is possible that n + m < 0. The variable b
denotes statement it is not possible that there is a natural number m, such that
m > 0. Translation: the formula A1 is (a ∪ ¬b).
2. We translate the statement S into a formula A2 of a language L{¬, , ♦, ∩, ∪, ⇒}
as follows. Propositional variables are: a, b. The variable a denotes statement
For all natural numbers n ∈ N the following implication holds: if n < 0, then
there is a natural number m, such that it is possible that n + m < 0. The
variable b denotes statement there is a natural number m, such that m > 0.
Translation: the formula A2 is (a ∪ ¬♦b).

82
3.2 Extensional Semantics M
Given a propositional language LCON , the symbols for its connectives always
have some intuitive meaning. A formal definition of the meaning of these sym-
bols is called a semantics for the language LCON . A given language can have
different semantics but we always define them in order to single out special for-
mulas of the language, called tautologies, i.e. formulas of the language that is
always true under the given semantics.
We introduced in Chapter 2 a notion of a classical propositional semantics, dis-
cussed its motivation and underlying assumptions. The assumption was that
we consider only two logical values. The other one was that all classical propo-
sitional connectives are extensional. We have also observed that in everyday
language there are expressions such as ”I believe that”, ”it is possible that”,
” certainly”, etc.... and they are represented by some propositional connectives
which are not extensional. Non-extensional connectives do not play any role
in mathematics and so are not discussed in classical logic and will be studied
separately.
The extensional connectives are defined intuitively as such that the logical value
of the formulas form by means of these connectives and certain given formulas
depends only on the logical value(s) of the given formulas. We adopt a following
formal definition of extensional connectives for a propositional language L and
of an extensional semantic for L.

Definition 3.6 (Extensional Connectives)

Let LCON be such that CON = C1 ∪ C2 , where C1 , C2 are the sets of unary and
binary connectives, respectively. Let LV be a non-empty set of logical values. A
connective 5 ∈ C1 or ◦ ∈ C2 is called extensional if it is defined by a respective
function
5 : LV −→ LV or ◦ : LV × LV −→ LV.

A semantics M for a language LCON is called extensional provided all con-

nectives in CON are extensional and its notion of tautology is defined in terms
of connectives and their logical values (see definition 3.7).
A semantics with a set of m-logical values is called a m-valued semantics. The
classical semantics is a special case of a 2-valued extensional semantics. Given a
language, its different semantics define corresponding different logics. Classical
semantics defines classical propositional logic with its set of classical proposi-
tional tautologies. Many of m- valued logics are defined by various extensional
semantics with sets of logical values LV with more then 2 elements. The lan-
guages of many important logics like modal, multi-modal, knowledge, believe,
temporal contain connectives that are not extensional. Consequently they are
defined by the non-extensional semantics. The intuitionistic logic is based on
the same language as the classical one, its Kripke Models semantics is not ex-

83
tensional.
Defining a semantics for a given propositional language means more then defin-
ing propositional connectives. The ultimate goal of any semantics is to define
the notion of its own tautology. In order to define which formulas of LCON we
want to to be tautologies under a given semantics M we assume that the set LV
of logical values of M always has a distinguished logical value, often denoted
by T for ”absolute” truth. We also can distinguish, and often we do, another
special value F representing ”absolute” falsehood. We will use these symbols
T, F. We may also use other symbols like 1, 0 or others. The value T serves to
define a notion of a tautology (as a formula always ”true”).
Extensional semantics share not only the similar pattern of defining their connec-
tives (definition 3.6), but also the method of defining the notion of a tautology.
We hence define a general notion of an extensional semantics (definition 3.7) as
sequence of steps leading to the definition of a tautology. Here are the steps.
Step1: we define all connectives of M as specified by definition 3.6.
Step 2: we define the main component of the definition of a tautology, namely
a function v that assigns to any formula A ∈ F its logical value from VL. It is
often called a truth assignment and we will use this name.
Step 3: given a truth assignment v and a formula A ∈ F, we define what does
it mean that v satisfies A, i.e. that v is a model for A under semantics M.
Step 4: we define a notion of tautology as follows: A is a tautology under seman-
tics M if and only if all truth assignments v satisfy A, i.e. all truth assignments
v are models for A.

We use a notion of a model because it is an important, if not the most important

notion of modern logic. It is usually defined in terms of the notion of satisfaction.
In classical propositional logic these two notions are the same. The use of
expressions ” v satisfies A” and ”v is a model for A” is interchangeable. This
is also a case for the extensional semantics; in particular for some non-classical
semantics, like m-valued semantics discussed in this chapter.
The notions of satisfaction and model are not interchangeable for predicate lan-
guages semantics. We already discussed these notions in chapter 2 and will
define them in full formality in chapter 8 on predicate logic. The use of the
notion of a model also allows us to adopt and discuss the standard predicate
logic definitions of consistency and independence for propositional case.

Given a language LCON and non-empty set LV of logical values. We assume

that the set LV has a special, distinguished logical value which serves to define a
notion of tautology under the semantics M. We denote this distinguished value
as T. We define formally a general notion of an extensional semantics M for
LCON as follows.

84
Definition 3.7 (Extensional Semantics)
A formal definition of an extensional semantics M for a given language LCON
consists of specifying the following steps defining its main components.
Step 1: we define a set LV of logical values and its distinguished value T, and
define all connectives of LCON to be extensional;
Step 2: we define notion of a truth assignment and its extension;
Step 3: we define notions of satisfaction, model, counter model;
Step 4: we define notion of a tautology under the semantics M.

What differs one semantics from the other is the choice of the set LV of logical
values and definition of the the connectives of LCON , i.e. the components
defined in the Step1. The definitions for the Steps 2 and 3, 4 are modification
of the definitions established for the classical case and they are as follows.
Step 1: we follow the definition 3.6 to define the connectives of M.
Step 2 : we define a function called truth assignment and its extension in terms
of the propositional connectives as defined in the Step 1. We use the term M
truth assignment and M truth extension to stress that it is defined relatively to
a given semantics M.

Definition 3.8 (M Truth Assignment)

Let LV be the set of logical values of M and VAR the set of propositional variables
of the language LCON . Any function v : V AR −→ LV , is called a truth
assignment under semantics M, for short M truth assignment.

Definition 3.9 (M Truth Extension)

Given M truth assignment
v : V AR −→ LV . We define its extension v ∗ to the set F of all formulas of
LCON as any function
v ∗ : F −→ LV,
such that the following conditions are satisfied.
(i) for any a ∈ V AR,
v ∗ (a) = v(a);

(ii) For any connectives 5 ∈ C1 , ◦ ∈ C2 , and for any formulas A, B ∈ F,

v ∗ (5A) = 5v ∗ (A), v ∗ ((A ◦ B)) = ◦(v ∗ (A), v ∗ (B)).
We call the v ∗ the M truth extension.

The symbols on the left-hand side of the equations represent connectives in

their natural language meaning and the symbols on the right-hand side repre-
sent connectives in their semantical meaning as defined in the Step1.

85
We use names ”M truth assignment”, ”M truth extension” to stress that we
define them for the set of logical values of M and moreover, that the extension
of v connects the formulas of the language with the connectives as defined by
the semantics M.

Notation Remark For a given function f, we use a symbol f ∗ to denote its

extension to a larger domain. Mathematician often use the same symbol f for
both a function and its extension f ∗ .

Step 3: the notions of satisfaction and model are interchangeable in extensional

semantics. They are not interchangeable in other propositional semantics and
in semantics for predicate languages.We define them as follows.

Definition 3.10 (M Satisfaction, Model)

Given an M truth assignment
v : V AR −→ LV and its M truth extension v ∗ . Let T ∈ LV be the distin-
guished logical value. We say that
the truth assignment v M satisfies a formula A if and only if v ∗ (A) = T .
We write symbolically
v |=M A.

Any truth assignment v, such that v |=M A is called M model for A.

Definition 3.11 (M Counter Model)

Given an M truth assignment
v : V AR −→ LV . Let T ∈ LV be the distinguished logical value. We say that
v does not satisfy a formula A ∈ F if and only if v ∗ (A) 6= T .
We denote it by
v 6 |=M A.

A any v, such that v 6 |=M A is called M counter model for A.

Step 4: we define the notion of a tautology under semantics M, called M

tautology as follows.

Definition 3.12 (M Tautology)

For any formula A ∈ F,
A is M tautology if and only if v |=M A, for all truth assignments v,
v : V AR −→ LV . We denote it as

|=M A.

86
We also say that A is M tautology if and only if all truth assignments v are
M models for A.

Observe that directly from definition 3.11 we get the following equivalent form
of the definition 3.12.

Definition 3.13
For any formula A ∈ F,
A is a M tautology if and only if v ∗ (A) = T , for all truth assignments v,
v : V AR −→ LV .

We denote by MT the set of all tautologies under the semantic M, i.e.

MT = {A ∈ F : |=M A.} (3.2)

Obviously, when we develop a logic by defining its semantics we want the se-
mantics to be such that the logic has a non empty set of its tautologies. We
stress that fact by putting it in a form of the following definition.

Definition 3.14
Given a language LCON and its extensional semantics M (definition 3.7), we
say that the semantics M is well defined if and only if its set MT of all
tautologies (3.2) is non empty, i.e. when

MT 6= ∅ (3.3)

We follow the definitions and pattens established here first in section 3.3. We
use them to define and discuss in details the classical propositional semantics.
Definitions and short discussions of some of the many-valued semantics follow
next in section 3.5. Many valued logics had their beginning in the work of
Lukasiewicz (1920). He was the first to define a 3- valued extensional semantics
for a language L{¬,∩,∪,⇒} of classical logic, and called it a three valued logic
for short. The other logics, now of historical value followed and we will discuss
some of them. In particular we present a Heyting 3-valued semantics as an
introduction to the definition and discussion of first ever semantics for the intu-
itionistic logic and some modal logics. It was proposed by J.C.C McKinsey and
A. Tarski in 1946-48 in a form of cylindrical algebras, now called pseudo-boolean
algebras, or Heyting algebras. The semantics in a form of abstract algebras are
called algebraic models for logics. It became a separate field of modern logic.
The algebraic models are generalization of the extensional semantics, hence the
importance of this section. It can me treated as an introduction to algebraic
models for logics. It will be discussed again in chapter7.

87
3.3 Classical Semantics

We adopt Steps 1- 4 of the general definition 3.7 of extensional semantics to

the case of the classical propositional logic as follows.
Step 1: we define the language, set of logical values, and define all connectives
of the language to be extensional
The language is L{¬, ∪, ∩, ⇒, ⇔} . The set LV of logical values is {T, F }. The
letters T, F stand as symbols of truth and falsehood, respectively. We adopt
T as the distinguished value. There are other notations for logical values, for
example 0,1, but we will use T, F.

Definition of connectives

Negation is a function ¬ : {T, F } −→ {T, F }, such that

¬(T ) = F, ¬(F ) = T. We write it as ¬T = F, ¬F = T .
Notation: we write the name of a two argument function (our connective)
between the arguments, not in front as in function notation, i.e. we write for
example, T ◦ T = T instead of ◦(T, T ) = T .
Conjunction is a function ∩ : {T, F } × {T, F } −→ {T, F }, such that
∩(T, T ) = T, ∩(T, F ) = F, ∩(F, T ) = F, ∩(F, F ) = F . We write it as
T ∩ T = T, T ∩ F = F, F ∩ T = F, F ∩ F = F .
Disjunction is a function ∪ : {T, F } × {T, F } −→ {T, F }, such that
∪(T, T ) = T, ∪(T, F ) = T, ∪(F, T ) = T, ∪(F, F ) = F . We write it as
T ∪ T = T, T ∪ F = T, F ∪ T = T, F ∪ F = F .
Implication is a function ⇒: {T, F } × {T, F } −→ {T, F }, such that
⇒ (T, T ) = T, ⇒ (T, F ) = F, ⇒ (F, T ) = T, ⇒ (F, F ) = T . We write it as
T ⇒ T = T, T ⇒ F = F, F ⇒ T = T, F ⇒ F = T .
Equivalence is a function ⇔: {T, F } × {T, F } −→ {T, F }, such that
⇔ (T, T ) = T, ⇔ (T, F ) = F, ⇔ (F, T ) = F, ⇔ (T, T ) = T . We write it as
T ⇔ T = T, T ⇔ F = F, F ⇔ T = F, T ⇔ T = T .
We write function defining the connectives in a standard form of tables defining
operations in finite sets. We call these tables (3.4) truth tables definition of
propositional connectives, or classical connectives truth tables for short.

Classical Connectives Truth Tables (3.4)

¬ T F ∩ T F ∪ T F
F T T T F T T T
F F F F T F

88
 ⇒ T F ⇔ T F
T T F T T F
F T T F F T

As ultimate goal of our semantics is to define the notion of tautology, a formula

that is always true, we assume that the set {T, F } of our logical values is ordered
and F < T , This makes the symbol T (for truth) the ”greatest” logical value,
what truth supposed to be. We now can write simple formulas defining the
connectives (respective function) as follows.

Classical Connectives Formulas (3.5)

¬ : {F, T } −→ {F, T }, such that ¬F = T, ¬T = F.

∩ : {F, T } × {F, T } −→ {F, T }, such that for any x, y ∈ {F, T },
∩(x, y) = min{x, y}. We write it as x ∪ y = min{x, y}.
∪ : {F, T } × {F, T } −→ {F, T }, such that for any x, y ∈ {F, T },
∪(x, y) = max{x, y}. We write it as x ∪ y = max{x, y}.
⇒: {F, T } × {F, T } −→ {F, T }, such that for any x, y ∈ {F, T },
⇒ (x, y) = ∪(¬x, y). We write it as x ⇒ y = ¬x ∪ y.
⇔: {F, T } × {F, T } −→ {F, T }, such that for any x, y ∈ {F, T },
⇔ (x, y) = ∪(⇒ (x, y), ⇒ (y, x)).
We write it as x ⇔ y = (x ⇒ y) ∩ (y ⇒ x).

Exercise 3.6
Prove that the above connectives formulas are correct, i.e. that they define the
same classical connectives as defined in Step 1.

Solution
This is a problem of proving equality of functions that are given the same names.
We have to show that the use of the same names: ¬, ∪, ∩, ⇒, ⇔ for them is
justified. The equality of functions is defined as follows.

Definition 3.15
Given two sets A, B and functions f, g, such that f : A −→ B and g : A −→ B.
We say that the functions f, g are equal and write it f = g if and only if f(x)
= g(x) for all elements x ∈ A.

The negation definition is the same in both cases. We prove that the two
conjunctions and two disjunctions functions are the equal by comparing both

89
definitions (3.4) and (3.5). We verify now the correctness of the implication
function formula. Consider two functions ⇒: {T, F } × {T, F } −→ {T, F } and
h : {T, F } × {T, F } −→ {T, F }, where ⇒ is the classical implication defined
by definition (3.4) and h is defined by the definition (3.5), i.e. by the for-
mula h(x, y) = ∪(¬x, y). Observe that we have already proved that functions
∪ and ¬ are equal in both cases. We prove that ⇒ = h by evaluating that
⇒ (x, y) = h(x, y) = ∪(¬x, y), for all (x, y) ∈ {T, F } × {T, F } as as follows.
T ⇒ T = T and h(T, T ) = ¬T ∪ T = F ∪ T = T yes.
T ⇒ F = F and h(T, F ) = ¬T ∪ F = F ∪ F = F yes.
F ⇒ F = T and h(F, F ) = ¬F ∪ F = T ∪ F = T yes.
F ⇒ T = T and h(F, T ) = ¬F ∪ T = T ∪ T = T yes.
This proves the correctness of the implication formula ⇒ (x, y) = ∪(¬x, y). We
write it as x ⇒ y = ¬x ∪ y and call it a formula defining implication in
terms of disjunction and negation. We verify the correctness of the equivalence
formula ⇔ (x, y) = ∪(⇒ (x, y), ⇒ (y, x)) in a similar way.

Special Properties of Connectives

Observe that the formulas defining connectives of implication and equivalence

are certain compositions of previously defined connectives. Classical semantics
is a special one, its connectives have strong properties that often do not hold
under other semantics, extensional or not. One of them is a property of defin-
ability of connectives, the other one is a functional dependency. These are basic
properties one asks about any new semantics, and hence a logic, being created.
We generalize these the notion of functional dependency of connectives under a
given extensional semantics M.

Definition 3.16 (Definability of Connectives)

Given a propositional language LCON and its extensional semantics M. A
connective ◦ ∈ CON is definable in terms of some connectives ◦1 , ◦2 , ...◦n ∈
CON if and only if ◦ is a certain function composition of functions ◦1 , ◦2 , ...◦n ,
as they are defined by the semantics M.

We have just proved in Exercise 3.6 that the implication ⇒ is definable in terms
of ∪ and ¬ under classical semantics as it is a composition of ∪ and ¬ defined
by the formula ⇒ (x, y) = ∪(¬x, y). The classical equivalence is definable in
terms of ⇒ and ∩ by the formula ⇔ (x, y) = ∪(⇒ (x, y), ⇒ (y, x)).

Definition 3.17 (Functional Dependency)

Given a propositional language LCON and its extensional semantics M. A
property of defining the set of connectives CON in terms of its proper subset is
called a functional dependency of connectives under the semantics M.

90
Proving the property of functional dependency under a given semantics M con-
sists of identifying a proper subset CON0 of the set CON of connectives, such
that each connective ◦ ∈ CON − CON0 is definable (definition 3.16) in terms
of connectives from CON0 . This is usually a difficult, and often impossible task
for many semantic. We prove now that it holds in the classical case.

Theorem 3.1
The set of connectives of the language L{¬, ∪, ∩, ⇒, ⇔} is functionally depen-
dent under the classical semantics.

Proof
Let’s take a set {¬, ∪}. We have already proved in Exercise 3.6 that the impli-
cation ⇒ and is definable in terms of ∪ and ¬ by the formula x ⇒ y = ¬x ∪ y.
The conjunction is defined by easy verification, similar to the one in Exercise
3.6, by a formula x ∩ y = ¬(¬x ∪ ¬y). By Exercise 3.6, the equivalence formula
is definable in terms of ⇒ and ∩ by the formula x ⇔ y = (x ⇒ y) ∩ (y ⇒ x).
The final formula for for the equivalence is x ⇔ y = (¬x ∪ y) ∩ (¬y ∪ x).

There are many ways to prove this theorem, it means there are many ways to
choose a proper subset CON0 of the set {¬, ∪, ∩, ⇒, ⇔} that defines all other
connectives. Here are the choices.

Theorem 3.2 (Definability of Connectives )

All connectives of the language L{¬, ∪, ∩, ⇒, ⇔} are definable in terms of ¬
and ◦, for any ◦ ∈ {∪, ∩, ⇒}.

Proof
We list all required definability formulas, including the formulas developed in
the proof of Theorem 3.1. An easy verification of their correctness is left as an
exercise.
1. Definability in terms of ⇒ and ¬.
x ∩ y = ¬(x ⇒ ¬y), x ∪ y = ¬x ⇒ y, x ⇔ y = ¬((x ⇒ y) ⇒ ¬(y ⇒ x)).
2. Definability in terms of ∩ and ¬.
x ∪ y = ¬(¬x ∩ ¬y), x ⇒ y = ¬(x ∩ ¬y), x ⇔ y = ¬(x ∩ ¬y) ∩ ¬(y ∩ ¬x).
3. Definability in terms of ∪ and ¬.
x ⇒ y = ¬x ∪ y, x ∩ y = ¬(¬x ∪ ¬y) x ⇔ y = (¬x ∪ y) ∩ (¬y ∪ x).

There are two other important classical binary connectives denoted by ↑ and
↓. The connective ↑ was discovered in 1913 by H.M. Sheffer, who called it
alternative negation. Now it is often called a Sheffer’s connective. A formula
(A ↑ B) reads: not both A and B. The connective ↓ was discovered in 1920 by
J. Lukasiewicz and named joint negation. The formula (A ↓ B) reads: neither
A nor B. They are defined as follows.

91
Alternative Negation is a function ↑: {T, F } × {T, F } −→ {T, F } such that
T ↑ T = F, T ↑ F = T, F ↑ T = T, F ↑ F = T .
Joint Negation is a function ↓: {T, F } × {T, F } −→ {T, F } such that
T ↓ T = F, T ↓ F = F, F ↓ T = F, F ↓ F = T .

Truth Tables for ↑ and ↓

↑ T F ↓ T F
T F T T F F
F T T F F T

We extend our language L{¬, ∪, ∩, ⇒, ⇔} by adding Sheffer and Lukasiewicz

connectives to it. We obtain the language L{¬, ∪, ∩, ⇒, ⇔, ↑, ↓} that contains
now all possible classical connectives.

Theorem 3.3
All connectives of a language L{¬, ∪, ∩, ⇒, ⇔, ↑, ↓} are definable in terms of ↑,
and also separately in terms of ↓.

Proof
Definability formulas of ¬ and ∩ in terms of ↑ are the following.

¬x = x ↑ x, x ∩ y = (x ↑ y) ↑ (x ↑ y) (3.6)

Definability formulas for of the connectives {∪, ⇒, ⇔} in terms of ↑ follow

directly from the formulas in the proof of Theorem 3.2 and the formulas (3.6).
Observe that the x ↑ y = ¬(x ∪ y). The definability of x ↓ y in terms of x ↑ y
follows from (3.6) and definability of ∪ in terms ↑.
Definability formulas of ¬ and ∪ in terms of ↓ are, by simple verification, the
following.
¬x = x ↓ x, x ∪ y = (x ↓ y) ↓ (x ↓ y) (3.7)
Definability formulas for of the connectives {∩, ⇒, ⇔, ↑} in terms of ↓ follow
directly, as in the previous case, from the Theorem 3.2 and the formulas (3.7).

Functional dependency and definability of connectives as expressed in Theorems

3.2, 3.3 are very strong and characteristic properties of the classical semantics.
They hold, for some connectives for some non-classical logics, never in others.
For example, the necessity connective  is definable in terms of the possibility
connectives ♦ and negation ¬ in Modal S4 and S5 logics, but not in majority of
others. The classical implication is definable in terms of negation and disjunc-
tion, but the intuitionistic implication is not. We defined and discussed these
classical properties here as they have to be addressed and examined when one

92
is building semantics for any of a non-classical logic.

Step 2: Truth Assignment, Truth Extension

We define now and examine the components in the Step 2 of the definition 3.7.
We start with the basic notion of the truth assignment. We adopt the exten-
sional semantics M definition 3.8 to the classical case as follows.

Definition 3.18 (Truth Assignment)

Let VAR be the set of all propositional variables of the language L{¬, ∪, ∩, ⇒, ⇔} .
A truth assignment is any function v : V AR −→ {T, F }.

The function v defined above is called the truth assignment because it can be
thought as an assignment to each variable (which represents a logical sentence)
its logical value of T(ruth) of F(alse). Observe that the domain of the truth
assignment is the set of propositional variables, i.e. the truth assignment is
defined only for atomic formulas.
We now extend the truth assignment v from the set of atomic formulas to the
set of all formulas F in order define formally the assignment of a logical value
to any formula A ∈ F.
The definition of the truth extension of the truth assignment v to the set F
follows the definition 3.8 for the extensional semantics M .

Definition 3.19 (Truth Extension) Given the truth assignment v : V AR −→

{T, F }. We define its extension v ∗ to the set F of all formulas as any function
v ∗ : F −→ {T, F }, such that the following conditions are satisfied.

(1) for any a ∈ V AR, v ∗ (a) = v(a);

(2) for any A, B ∈ F,
v ∗ (¬A) = ¬v ∗ (A);
v ∗ ((A ∩ B)) = ∩(v ∗ (A), v ∗ (B));
v ∗ ((A ∪ B)) = ∪(v ∗ (A), v ∗ (B));
v ∗ ((A ⇒ B)) =⇒ (v ∗ (A), v ∗ (B));
v ∗ ((A ⇔ B)) =⇔ (v ∗ (A), v ∗ (B)).

The symbols on the left-hand side of the equations represent the connectives in
their natural language meaning. The symbols on the right-hand side represent

93
connectives in their classical semantics meaning defined by the classical con-
nectives defined by the classical Truth Tables.

For binary connectives (two argument functions) we adopt a convention to write

the symbol of the connective (name of the 2 argument function) between its ar-
guments as we do in a case arithmetic operations. We use this standard notation
and re-write the definition 3.19 as follows.

Definition 3.20 (Standard Notation) Given the truth assignment v : V AR −→

{T, F }. We define its extension v ∗ to the set F of all formulas as any function
v ∗ : F −→ {T, F }, such that the following conditions are satisfied.

(1) for any a ∈ V AR, v ∗ (a) = v(a);

(2) for any A, B ∈ F,
v ∗ (¬A) = ¬v ∗ (A);
v ∗ ((A ∩ B)) = v ∗ (A) ∩ v ∗ (B);
v ∗ ((A ∪ B)) = v ∗ (A) ∪ v ∗ (B);
v ∗ ((A ⇒ B)) = v ∗ (A) ⇒ v ∗ (B);
v ∗ ((A ⇔ B)) = v ∗ (A) ⇔ v ∗ (B).

Given a formula A: ((a ⇒ b) ∪ ¬a)) and a truth assignment v, such that

v(a) = T, v(b) = F . We evaluate the logical value of the formula A using the
standard notation definition 3.20 as follows.
v ∗ (A) = v ∗ (((a ⇒ b)∪¬a))) = ∪(v ∗ ((a ⇒ b), v ∗ (¬a)) = ∪(⇒ (v ∗ (a), v ∗ (b)), ¬v ∗ (a)))
= ∪(⇒ (v(a), v(b)), ¬v(a))) = ∪(⇒ (T, F ), ¬T )) = ∪(F, F ) = F .

Observe that we did not specify v(x) of any x ∈ V AR−{a, b}, as these values do
not influence the computation of the logical value of the formula A. We say: ”v
such that” as we consider its values for the variables a and b only. Nevertheless,
the domain of the truth assignment v is always is the set of all variables VAR
and we have to remember that.

Given a formula A: ((a ⇒ b) ∪ ¬a)) and a truth assignment v, such that

v(a) = F, v(b) = F . We use now the standard notation definition 3.20 to
evaluate the logical value of the formula A. We write is as follows. v ∗ (A) =
v ∗ (((a ⇒ b) ∪ ¬a))) = v ∗ ((a ⇒ b)) ∪ v ∗ (¬a) = (v(a) ⇒ v(b)) ∪ ¬v(a) = (F ⇒
F ) ∪ ¬F = T ∪ T = T .

Step 3: Satisfaction, Model, Counter-Model

94
We define now and examine the components in Step 3 of the definition 3.7.
We adopt the extensional semantics M definitions 3.10, 3.11, and 3.12 to the
classical case as follows.

Definition 3.21 (Satisfaction)

Let v : V AR −→ {T, F }.
We say that v satisfies a formula A ∈ F if and only if v ∗ (A) = T . We denote
it by v |= A.
v does not satisfy a formula A ∈ F if and only if v ∗ (A) 6= T . We denote it
by v 6|= A.

The relation |= is often called a satisfaction relation. Observe, that in the

classical semantics we have that v ∗ (A) 6= T if and only if v ∗ (A) = F . In this
case we say that v falsifies a formula A.

Exercise 3.7
Let A be a formula ((a ⇒ b) ∪ ¬a)) and v be a truth assignment
v : V AR −→ {T, F }, such that v(a) = T, v(b) = F , and v(x) = F for all
x ∈ V AR − {a, b}. Show that v 6|= ((a ⇒ b) ∪ ¬a)).

Proof We evaluate the logical value of the formula A as follows: v ∗ (A) =

v ∗ ((a ⇒ b) ∪ ¬a)) = (v ∗ (a ⇒ b) ∪ v ∗ (¬a)) = ((v(a) ⇒ v(b)) ∪ ¬v(a)) = ((T ⇒
F ) ∪ ¬T ) = (F ∪ F ) = F . It proves tha v 6|= ((a ⇒ b) ∪ ¬a)) and we say that v
falsifies the formula A.

As we remarked before, in practical cases we use a short-hand notation for while

evaluating the logical value of a given formula. Here is a short procedure for
any v and A. We use show it how it works for v and A from the exercise 3.7.

Short-hand Evaluation
Given any formula A ∈ F and any truth assignment v : V AR −→ {T, F }.
1. We write the value of v only for variables appearing in the formula in A.
In our case we write: a = T , b = F for v(a) = T, v(b) = F .
2. Replace all variabes in A by their respective logical values.
In our case we replace a by T and b by F in the formula A ((a ⇒ b) ∪ ¬a)). We
get an equation ((T ⇒ F ) ∪ ¬T ).
3. Use use the connectives definition, in this case the definitionTTables to eval-
uate the logical value of the equation obtained in the step 2.
In our case we evaluate ((T ⇒ F ) ∪ ¬T ) = (F ∪ F ) = F .
4. Write your answer in one of the forms: v |= A, v 6|= A or ”v satisfies A”, ” v
falsifies A”

95
In our case v falsifies A and write v 6|= ((a ⇒ b) ∪ ¬a)).

Example 3.11
Let A be a formula ((a ∩ ¬b) ∪ ¬c) and v be a truth assignment v : V AR −→
{T, F }, such that v(a) = T, v(b) = F, v(c) = T , and v(x) = T for all
x ∈ V AR − {a, b, c}. Using the the short-hand notation we get ((T ∩ ¬F ) ∪
¬T ) = ((T ∩ T ) ∪ F ) = (T ∪ F ) = T . It proves that v satisfies the formula A
and we write v |= ((a ∩ ¬b) ∪ ¬c).

Definition 3.22 (Model, Counter Model)

Given a formula A ∈ F.
Any v : V AR −→ {T, F }, such that v |= A is called a model for A .
Any v, such that v 6|= A is called a counter model for the formula A .

The truth assignment from the Example 3.11 is a model for the formula
((a∩¬b)∪¬c) and the truth assignment from the Exercise 3.7 is a counter-model
for the formula ((a ⇒ b) ∪ ¬a)).

Step 4: Classical Tautology Definition

There are two equivalent ways to define the notion of classical tautology. We
will use them interchangeably. The first uses the notion of truth assignment
and states the following.

Definition 3.23 (Tautology 1) For any formula A ∈ F,

A is tautology if and only if v ∗ (A) = T for all v : V AR −→ {T, F }.

The second uses the notion of satisfaction and model and the fact that in any
extensional semantic the notions ” v satisfies A” and ”v is a model for A” are
interchangeable. It is stated as follows.

Definition 3.24 (Tautology 2)

For any formula A ∈ F,
A is tautology if and only if v |= A for all v : V AR −→ {T, F }, i.e. when
all truth assignments are models for A.

We write symbolically
|= A
for the statement ”A is a tautology”.

96
Remark 3.1
We use the symbol |= A only for classical tautology. For all other extensional
semantics M we must use the symbol |=M A and say ” A is a tautology under
a semantics M, or to say in short ”A is a M semantics tautology”.

We usually use the definition 3.24 to express that a formula in not a tautology,
i.e. we say that a formula is not a tautology if it has a counter model. To stress
it we put it in a form of a following definition.

Definition 3.25
For any formula A ∈ F,
A is not a tautology if and only if A has a counter model;
i.e. when there is a truth assignment v : V AR −→ {T, F }, such that v 6|= A.

We denote the statement ”A is not a tautology” symbolically by

6|= A.

A formula A : ((a ⇒ b) ∪ ¬a)) is not a tautology (6|= ((a ⇒ b) ∪ ¬a))). A truth

assignment v : V AR −→ {T, F }, such that v(a) = T, v(b) = F , and v(x) = F
for all x ∈ V AR − {a, b} is a counter model for A, as we proved Exercise 3.7.

This ends the formal definition of classical semantics that follows the pattern
for extensional semantics established in the definition 3.7.

3.3.1 Tautologies: Decidability and Verification Methods

There is a large number of basic and important tautologies listed and discussed
in Chapter 2. We assume that the reader is familiar, or will familiarize with
them if needed. We will refer to them and use them within our book. Chapter
2 also provides the motivation for classical approach to definition of tautolo-
gies as ways of describing correct rules of our mathematical reasoning. It also
contains an informal definition of classical semantics and discusses a tautology
verification method. We have just defined formally the classical semantics. Our
goal now is to prove formally that the notion of classical tautology is decidable
(Theorem 3.9) and to prove correctness of the tautology verification method
presented in Chapter 2. Moreover we present here other basic tautology verifi-
cation methods and prove their correctness.

We start now a natural question. How do we verify whether a given formula

A ∈ F is or is not a tautology? The answer seems to be very simple. By

97
definition 3.23 we have to examine all truth assignments v : V AR −→ {T, F }.
If they all evaluate to T, we proved that |= A. If at least one evaluates to F, we
found a counter model and proved 6|= A. The verification process is decidable,
if the we have only a finite number of v to consider. So now all we have to
do is to count how many truth assignments there are, i.e. how many there are
functions that map the set V AR of propositional variables into the set {T, F } of
logical values. In order to do so we need to introduce some standard notations
and some known facts. For a given set X, we denote by |X| the cardinality of
X. In a case of a finite set, it is called a number of elements of the set. We
write |X| = n to denote that X has n elements, for n ∈ N . We have a special
names and notations to denote the cardinalities of infinite set. In particular
we write |X| = ℵ0 and say ” cardinality of X is aleph zero,” for any countably
infinite set X. We write |X| = C and say ” cardinality of X is continuum” for
any uncountable set X that has the same cardinality as Real numbers.

Theorem 3.4 (Counting Functions )

For any sets X, Y, there are |Y ||X| functions that map the set X into Y.
In particular, when the set X is countably infinite and the set Y is finite, then
there are nℵ0 = C functions that map the set X into Y.

In our case of counting the truth assignment v : V AR −→ {T, F } we have that

|V AR| = ℵ0 and |{T, F }| = 2. We know that 2ℵ0 = C and hence we get directly
from Theorem 3.4 the following.

Theorem 3.5 (Counting Truth Assignments)

There are uncountably many (exactly as many as real numbers) of all possible
truth assignments v, where v : V AR −→ {T, F }.

Definition 3.26
For any A ∈ F, let V ARA be a set of all propositional variables appearing
in A. Any function vA : V ARA −→ {T, F }, is called a truth assignment
restricted to A.

Example 3.12
Let A = ((a ⇒ ¬b) ∪ ¬c). The set of variables appearing in A is V ARA =
{a, b, c}. The truth assignment restricted to A is any function vA : {a, b, c} −→
{T, F }.

Definition 3.27
Given a formula A ∈ F and a set V ARA of all propositional variables appearing
in A. Any function vA : V ARA −→ {T, F }, such that v |= A ( v 6|= A) is
called a restricted model (counter model) for A.

98
We use the following particular case of Theorem 3.4 to count, for any formula
A, possible truth assignment restricted to A, i.e. all possible restricted models
and counter models for A.

Theorem 3.6 (Counting Functions 1)

For any finite sets X and Y , if X has n elements and Y has m elements, then
there are mn possible functions that map X into Y .

We also can prove it independently by Mathematical Induction over m.

Given a formula A ∈ F, the set V ARA is always finite, and |{T, F }| = 2 , so

directly from Theorem 3.6 we get the following.

Theorem 3.7 (Counting Restricted Truth)

For any A ∈ F, there are 2|V ARA | of possible truth assignments restricted to
A

So there are 23 = 8 possible truth assignment restricted to the formula A =

((a ⇒ ¬b) ∪ ¬c). We usually list them, and their value on the formula A in a
form of an extended truth table below.

vA a b c v ∗ (A) computation v ∗ (A)

v1 T T T (T ⇒ T ) ∪ ¬T = T ∪ F = T T
v2 T T F (T ⇒ T ) ∪ ¬F = T ∪ T = T T
v3 T F F (T ⇒ F ) ∪ ¬F = F ∪ T = T T
v4 F F T (F ⇒ F ) ∪ ¬T = T ∪ F = T T (3.8)
v5 F T T (F ⇒ T ) ∪ ¬T = T ∪ F = T T
v6 F T F (F ⇒ T ) ∪ ¬F = T ∪ T = T T
v7 T F T (T ⇒ F ) ∪ ¬T = F ∪ F = F F
v8 F F F (F ⇒ F ) ∪ ¬F = T ∪ T = T T
v1 , 2 , v3 , v4 , v5 , v6 , v8 are restricted models for A and v7 is a restricted
counter model for A.
Now we are ready to prove the correctness of the well known truth tables tau-
tology verification method. We formulate it as the follows.

Theorem 3.8 (Truth Tables)

For any formula A ∈ F,
|= A if and only if vA |= A for all vA : V ARA −→ {T, F }, i.e.
|= A if and only if all vA : V ARA −→ {T, F } are restricted models for A.

Proof Assume |= A. By definition 3.24 we have that v |= A for all v : V AR −→

{T, F }, hence vA |= A for all vA : V ARA −→ {T, F } as V ARA ⊆ V AR.

99
Assume vA |= A for all vA : V ARA −→ {T, F }. Take any v : V AR −→ {T, F },
as V ARA ⊆ V AR, any v : V AR −→ {T, F } is an extersion of some vA , i.e.
v(a) = vA (a) for all a ∈ V ARA . By Truth Extension Definition 3.19 we get
that v ∗ (A) = vA ∗ (A) = T and v |= A. This ends the proof.

Directly from Theorem 3.7 and the above Theorem 3.8 we get the proof of
the correctness and decidability of the Truth Tables Method, and hence the
decidability of the notion of classical propositional tautology.

Theorem 3.9 (Tautology Decidability)

For any formula A ∈ F, one has to examine at most 2V ARA restricted truth
assignments vA : V ARA −→ {F, T } in order to decide whether
|= A or 6|= A,
i.e. the notion of classical tautology is decidable.

We present now and prove correctness of some basic tautologies verification

methods. We just proved (Theorem 3.9) the correctness of the truth table tau-
tology verification method, so we we start with it.

Truth Table Method

The verification method, called a truth table method consists of examination,

for any formula A, all possible truth assignments restricted to A. By theorem
3.7 we have to perform at most 2|V ARA | steps. If we find a truth assignment
which evaluates A to F , we stop the process and give answer: 6|= A. Otherwise
we continue. If all truth assignments evaluate A to T , we give answer: |= A.
We usually list all restricted truth assignments vA in a form of a truth table
similar to the table3.8, hence the name of the method.
Consider, for example, a formula A: (a ⇒ (a ∪ b)). There are 22 = 4 possible
truth assignment restricted to A. We usually list them, and evaluate their value
on the formula A in a form of an extended truth table as follows.

w a b w∗ (A) computation w∗ (A)

w1 T T T ⇒ (T ∪ T ) = T ⇒ T = T T
w2 T F T ⇒ (T ∪ F ) = T ⇒ T = T T (3.9)
w3 F T F ⇒ (F ∪ T ) = F ⇒ T = T T
w4 F F F ⇒ (F ∪ F ) = F ⇒ F = T T

The table (3.9) shows that all w : V ARA −→ {T, F } are restricted models for
A and hence by Theorem 3.9 we proved that |= (a ⇒ (a ∪ b)) and T 6= ∅.

100
Observe that the table (3.9) proves that the formula 6|= ((a ⇒ ¬b) ∪ ¬c).

Moreover we have proved that the condition (3.3) of the definition 3.14 is fulfilled
and the classical semantics is well defined. We put it as a separate statement.

Fact 3.2
The classical semantics is well defined.

The complexity of the truth table methods grows exponentially. Impossible for
humans to handle formulas with more then few variables, and cumbersome for
computers for formulas with a great number of variables, In practice, if we need,
we use often much shorter and more elegant tautology verification methods pre-
sented below.

Proof by Contradiction Method

In this method, in order to prove that |= A we proceed as follows.
We assume that 6|= A. We work with this assumption. If we get a contradic-
tion, we have proved that 6|= A is impossible. We hence proved |= A. If we do
not get a contradiction, it means that the assumption 6|= A is true, i.e. we have
proved that A is not a tautology.

Exercise 3.8
Follow the Proof by Contradiction Method and examine whether
|= (a ⇒ (a ∪ b)).

Solution
We use a short-hand notation.
Assume that 6|= (a ⇒ (a ∪ b)). It means that (a ⇒ (a ∪ b)) = F for some truth
assignment v. By definition of implication ⇒ we have that
(a ⇒ (a ∪ b)) = F if and only if a = T and (a ∪ b) = F.
From a = T and (a ∪ b) = F we get (T ∪ b) = F . This is a contradiction with
the definition of disjunction ∪. Hence we proved |= (a ⇒ (a ∪ b)).

Exercise 3.9
Use the Proof by Contradiction Method to decide whether
|= ((a ∪ b) ⇒ a).

Solution We do not use short-hand notation.

Assume that 6|= ((a ∪ b) ⇒ a). It means that there is v : V AR −→ {T, F }, such
that v ∗ (((a∪b) ⇒ a)) = F . We evaluate, v ∗ (((a∪b) ⇒ a))) = v ∗ ((a∪b)) ⇒ v(a)

101
and we get that the truth assignment v is such that v ∗ ((a ∪ b)) ⇒ v(a) = F .
By definition implication ⇒ we have that v ∗ ((a ∪ b)) ⇒ v(a) = F if and only
if v(a) ∪ v(b) = T and (a) = F . From (a) = F and v(a) ∪ v(b) = T we get
that F ∪ v(b) = T . This is possible for any v : V AR −→ {T, F }, such that
v(b) = T . This proves that any truth assignment v : V AR −→ {T, F }, such that
(a) = F, v(b) = T is a counter model for ((a∪b) ⇒ a) , i.e. that 6|= ((a∪b) ⇒ a).

Substitution Method
We define and prove the correctness of a method, called Substitution Method
that allows us to obtain new tautologies from formulas already proven to be
tautologies.
We can use the same reasoning as we used in the solution to the Exercise 3.8
that proved |= (a ⇒ (a ∪ b)) to prove that, for example the formulas

((((a ⇒ b) ∩ ¬c) ⇒ ((((a ⇒ b) ∩ ¬c) ∪ ¬d)) (3.10)

((a ⇒ b) ∩ ¬c) ∪ d) ∩ ¬e) ⇒ (((a ⇒ b) ∩ ¬c) ∪ d) ∩ ¬e) ∪ ((a ⇒ ¬e))) (3.11)

are also a tautologies.
Instead of repeating the same argument from Exercise 3.8 for a much more
complicated formulas we make a simple observation that we can obtain (3.10),
(3.11) from the formula (a ⇒ (a ∪ b)) by a proper substitutions (replacements)
of more complicated formulas for the variables a and b in (a ⇒ (a ∪ b)). We use
a notation A(a, b) = (a ⇒ (a ∪ b)) to denote that (a ⇒ (a ∪ b)) is a formula A
with two variables a, b and we denote by

A(a/A1 , b/A2 )

a result of a substitution (replacement) of formulas A1 , A2 on a place of the

variables a, b, respectively, everywhere where they appear in A(a, b).
Theorem 3.10 we are going to prove states that substitutions lead always from
a tautology to a tautology. In particular, making the following substitutions s1
and s2 in A(a, b) = (a ⇒ (a ∪ b)) we get, that the respective formulas (3.10),
(3.11) are tautologies.
By substitution s1: A(a/((a ⇒ b) ∩ ¬c), b/¬d) we get that

|= ((((a ⇒ b) ∩ ¬c) ⇒ ((((a ⇒ b) ∩ ¬c) ∪ ¬d)).

By substitution s2: A(a/((a ⇒ b) ∩ ¬c), b/((a ⇒ ¬e)) we get that

|= (((a ⇒ b) ∩ ¬c) ∪ d) ∩ ¬e) ⇒ (((a ⇒ b) ∩ ¬c) ∪ d) ∩ ¬e) ∪ ((a ⇒ ¬e))).
The theorem 3.10 describes validity of a method of constructing new tautolo-
gies from given tautologies. In order to formulate and prove it we first introduce
needed notations.

102
Let A ∈ F be a formula and V ARA = {a1 , a2 , ...an } be the set of all proposi-
tional variables appearing in A. We will denote it by A(a1 , a2 , ...an ).
Given a formula A(a1 , a2 , ...an ), and A1 , ...An be any formulas. We denote by
A(a1 /A1 , ..., an /An )
the result of simultaneous replacement (substitution) in A(a1 , a2 , ...an ) the vari-
ables a1 , a2 , ...an by formulas A1 , ...An , respectively.

Theorem 3.10
For any formulas A(a1 , a2 , ...an ), A1 , . . . , An ∈ F,
If |= A(a1 , a2 , ...an ) and B = A(a1 /A1 , ..., an /An ), then |= B.

Proof. Let B = A(a1 /A1 , ..., an /An ). Let b1 , b2 , ...bm be all those propositional
variables which occur in A1 , ...An . Given a truth assignment v : V AR −→
{T, F }, any values v(b1 ), v(b2 ), ...v(bm ) defines the logical value of A1 , ...An , i.e.
v ∗ (A1 ), ...v ∗ (An ) and, in turn, v ∗ (B).
Let w : V AR −→ {T, F } be a truth assignment such that w(a1 ) = v ∗ (A1 ), w(a2 ) =
v ∗ (A2 ), ...w(an ) = v ∗ (An ). Obviously, v ∗ (B) = w∗ (A). Since A is a proposi-
tional tautology, w∗ (A) = T , for all possible w, hence v ∗ (B) = w∗ (A) = T for
all truth assignments w and B is also a tautology.

We have proved (Exercise 3.8) that the formula D(a, b) = (a ⇒ (a ∪ b)) is a

tautology. By the above Theorem 3.10 we get that D(a/A, b/B) = ((A ∪ B) ⇒
A) is a tautology. We hence get the following.

Fact 3.3
For any A, B ∈ F, |= ((A ∪ B) ⇒ A).

Generalization Method
Now let’s look at the task of finding whether the formulas (3.10), (3.11) are
tautologies from yet another perspective. This time we observe that both of
them are build in a similar way as a formula (A ⇒ (A ∪ B)), for A = ((a ⇒
b) ∩ ¬c), B = ¬d in (3.10) and for A = ((a ⇒ b) ∩ ¬c), B = ((a ⇒ ¬e)) in
(3.11).
It means we represent, if it is possible, a given formula as a particular case of
some much simpler general formula. Hence the name Generalization Method.
We then use Proof by Contradiction Method or Substitution Method to examine
whether the representation of the given formula is /is not a tautology.
In this case, we prove, for example Proof by Contradiction Method by that
|= (A ⇒ (A ∪ B)), for any formulas A, B ∈ F and get, as a particular cases for
A, B that that both formulas (3.10), (3.11) are tautologies.

103
Let’s assume that there are formulas A, B ∈ F 6|= (A ⇒ (A ∪ B)). This means
that (A ⇒ (A ∪ B)) = F for some truth assignment v. This holds only when
A = T and (A ∪ B) = F , i.e. (T ∪ B) = F . This is a contradiction with the
definition of ∪. So |= (A ⇒ (A ∪ B)) for all A, B ∈ F.

Exercise 3.10
Show that v |= (¬((a ∩ ¬b) ⇒ ((c ⇒ (¬f ∪ d)) ∪ e)) ⇒ ((a ∩ ¬b) ∩ (¬(c ⇒
(¬f ∪ d)) ∩ ¬e))), for all v : V AR −→ {T, F }.

Solution
Observe that we really have to prove that |= (¬((a ∩ ¬b) ⇒ ((c ⇒ (¬f ∪ d)) ∪
e)) ⇒ ((a ∩ ¬b) ∩ (¬(c ⇒ (¬f ∪ d)) ∩ ¬e))). We can hence use any of our
tautology verification methods. In this case V ARA = {a, b, c, d, e, f }, so there
are 26 = 64 restricted truth assignments to consider. Much too many to apply
the Truth Table Method. Our formula is also far too complicated to guess a
simple tautology from which we could obtain it by the Substitution Method.
The Proof by Contradiction Method is less complicated, but before we apply
it let’s look closer at the sub-formulas of our formula and patterns they form
inside the formula it, i.e. we try to apply the Generalization Method first.
Let’s put B = (a ∩ ¬b), C = (c ⇒ (¬f ∪ d)), D = e. We re-write our formula
in a general form as (¬(B ⇒ (C ∪ D)) ⇒ (B ∩ (¬C ∩ ¬D))) and prove that for
all B, C, D ∈ F,

|= (¬(B ⇒ (C ∪ D)) ⇒ (B ∩ (¬C ∩ ¬D)).

We use Proof by Contradiction Method, i.e. we assume that there are formulas
B, C, D ∈ F, such that

6|= (¬(B ⇒ (C ∪ D)) ⇒ (B ∩ (¬C ∩ ¬D)).

This means that there is a truth assignment v, such that (we use short-hand
notation) (¬(B ⇒ (C ∪ D)) ⇒ (B ∩ (¬C ∩ ¬D))) = F . By definition of
implication it is possible if and only if ¬(B ⇒ (C ∪ D)) = T and (B ∩ (¬C ∩
¬D)) = F , i.e. if and only if
(B ⇒ (C ∪D)) = F and (B ∩(¬C ∩¬D)) = F Observe that (B ⇒ (C ∪D)) = F
if and only if B = T, C = F, D = F. We now evaluate the logical value of
(B ∩ (¬C ∩ ¬D)) for B = T, C = F, D = F , i.e. we compute (B ∩ (¬C ∩ ¬D)) =
(T ∩ (¬F ∩ ¬F )) = (T ∩ (T ∩ T )) = T . This contradicts that we must have
(B ∩ (¬C ∩ ¬D)) = F . This proves that for all B, C, D ∈ F

|= (¬(B ⇒ (C ∪ D)) ⇒ (B ∩ (¬C ∩ ¬D))),

and hence is holds for our particular case, i..e.

|= (¬((a ∪ b) ⇒ ((c ⇒ d) ∪ e)) ⇒ ((a ∪ b) ∩ (¬(c ⇒ d) ∩ ¬e)))

104
and that all truth assignments are models for (¬((a ∪ b) ⇒ ((c ⇒ d) ∪ e)) ⇒
((a ∪ b) ∩ (¬(c ⇒ d) ∩ ¬e))).

Sets of Formulas; Tautologies and Contradictions

We distinguish now special sets of formulas and examine their properties. We

define sets of all tautologies, contradictions, consistent sets, inconsistent sets
and discuss a notion of independence of formulas from sets of formulas.

Definition 3.28 (Set of Tautologies)

We denote by T the set of all tautologies, i.e. we put

T = {A ∈ F : |= A.}

We distinguish now another type of formulas, called contradictions.

Definition 3.29 (Contradiction)

A formula A ∈ F is called a contradiction if it does not have a model.

We write symbolically =| A for the statement ”A is a contradiction.”

Directly from the Definition 8.21 we have that

=| A if and only if v 6|= A for all v : V AR −→ {T, F }.

Example 3.13
The following formulas are contradictions

(a ∩ ¬a), (a ∩ ¬(a ∪ b)), ¬(a ⇒ a), ¬(¬(a ∩ b) ∪ b)).

Definition 3.30 (Set of Contradictions)

We denote by C the set of all tautologies, i.e. we put

C = {A ∈ F : =| A.}

Following the proof of Theorem 3.10 we get similar theorem for contradictions,
and hence a a proof of correctness of the Substitution Method of constructing
new contradictions.

Theorem 3.11
For any formulas A(a1 , a2 , ...an ), A1 , . . . , An ∈ F,
If A(a1 , a2 , ...an ) ∈ C and B = A(a1 /A1 , ..., an /An ), then B ∈ C.

105
Directly from the Theorem 3.11 we get the following.

Example 3.14 For any formulas A, B ∈ F, the following formulas are contra-
dictions

(A ∩ ¬A), (A ∩ ¬(A ∪ B)), ¬(A ⇒ A), ¬(¬(A ∩ B) ∪ B)).

Observe, that there are formulas which neither in T nor in C, for example
(a ∪ b). Any truth assignment v, such that v(a) = F, v(b) = F falsifies (a ∪ b)
and it proves that it is not a tautology. Any truth assignment v, such that
v(a) = T, v(b) = T satisfies (a ∪ b), what proves that it is not a contradiction.

3.3.2 Sets of Formulas: Consistency and Independence

Next important notions for any logic are notions of consistency, inconsistency of
the sets of formulas and the independence of a formula from the set of formulas.
We adopt the following definitions.

Definition 3.31
A truth truth assignment v : V AR −→ {T, F } is model for the set G ⊆ F of
formulas if and only if v |= A for all formulas A ∈ G. We denote it by

v |= G.

S
The restriction vG of the model v to the domain V ARG = A∈G V ARA is
called a restricted model for G.

Exercise 3.11
Find a model and a restricted model for a set

G = {((a ∩ b) ⇒ b), (a ∪ b), ¬a}.

Solution
Let v be a truth assignment v : V AR −→ {T, F }. By the defininition 3.31,
v |= {((a ∩ b) ⇒ b), (a ∪ b), ¬a} if and only if v ∗ (((a ∩ b) ⇒ b)) = T, v ∗ ((a ∪ b) =
T ), and v ∗ (¬a) = T . Observe that |= ((a ∩ b) ⇒ b), so we have to find v, such
that v ∗ ((a ∪ b)) = T, v ∗ (¬a) = T . This holds if and only if v(a) = F and
F ∪ v(b) = T , i.e. if and only if v(a) = F and v(b) = T . This proves that
any v such that v(a) = F and v(b) = T is a model for G, and G has only one
restricted model. We put it as a separate fact.

106
Fact 3.4
S
Given G = {((a ∩ b) ⇒ b), (a ∪ b), ¬a}, we have that V ARG = A∈G V ARA =
{a, b} and vG : {a, b} −→ {T, F }, such that vG (a) = F and vG (b) = T is a
unique restricted model for G.

Observation 3.4
For some sets G ⊆ F , V ARG can be infinite. For example, for G = V AR we
have that V ARG = V AR and the notions of model and restricted model are the
same.

Definition 3.32
A set G ⊆ F is called consistent if and only if there is v : V AR −→ {T, F },
such that v |= G.
Otherwise the set G is called inconsistent.

Plainly speaking, a set G is consistent if it has a model, and is inconsistent if it

does not have a model.

Example 3.15
The set G1 = {((a ∩ b) ⇒ b), (a ∪ b), ¬a} is consistent as v : V AR −→ {T, F },
such that v(a) = F and v(b) = T is the model for G1 .

The set G2 = V AR is also consistent, as v : V AR −→ {T, F }, such that

v(a) = T , for all a ∈ V AR is a model for G2 .

Observe that G1 is a finite consistent set. G2 is an infinite consistent set. This

and other examples justify the need of truth assignment domain being the set
VAR of all propositional variables.

Example 3.16
The set G1 = {((a ∩ b) ⇒ b), (a ∩ ¬a), ¬a} is a finite inconsistent set as it
contains a formula (a ∩ ¬a) ∈ C.
The set G2 = V AR ∪ {¬a} for some a ∈ V AR, is an infinite inconsistent set
as it contains a certain variable a and its negation ¬a.

Of course the most obvious example of an infinite consistent set is the set T of
all tautologies, and of an infinite inconsistent consistent set is the set C of all
contradictions.

Definition 3.33

107
A formula A ∈ F is called independent from a set G ⊆ F if and only if
the sets G ∪ {A} and G ∪ {¬A} are both consistent. I.e. when there are truth
assignments v1 , v2 such that

v1 |= G ∪ {A} and v2 |= G ∪ {¬A}.

Exercise 3.12
Show that a formula A = ((a ⇒ b) ∩ c) is independent from the set G =
{((a ∩ b) ⇒ b), (a ∪ b), ¬a}.

Solution
We define two truth assignments v1 , v2 : V AR −→ {T, F } such that v1 |=
G ∪ {(a ⇒ b) ∩ c)} and v2 |= G ∪ {¬(a ⇒ b) ∩ c)} as follows. We have just proved
(Exercise 3.11) that any v : V AR −→ {T, F }, such that v(a) = F, v(b) = T
is a model for G. Take as v1 any truth assignment such that v1 (a) = v(a) =
F, v1 (b) = v(b) = T, v1 (c) = T. We evaluate v1 ∗ (A) = v1 ∗ (((a ⇒ b) ∩ c)) =
(F ⇒ T ) ∩ T = T . This proves that v1 |= G ∪ {A}. Take as v2 any truth
assignment such that, v2 (a) = v(a) = F, v2 (b) = v(b) = T, v2 (c) = F . We
evaluate v2 ∗ (¬A) = v2 ∗ (¬(((a ⇒ b) ∩ c)) = T ∩ T = T . This proves that
v2 |= G ∪ {¬A}. It ends the proof that formula A is independent from G.

Exercise 3.13
Show that a formula A = (¬a ∩ b) is not independent from G = {((a ∩ b) ⇒
b), (a ∪ b), ¬a}.

Solution We have to show that it is impossible to construct v1 , v2 such that

v1 |= G ∪{A} and v2 |= G ∪{¬A}. From Fact 6.1 G has a unique restricted model
v : {a, b} −→ {T, F }, such that v(a) = F, and v(b) = T . and {a, b} = V ARA .
So we have to check now if it is possible v |= A and v |= ¬A. We evaluate
v ∗ (A) = v ∗ ((¬a ∩ b) = ¬v(a) ∩ v(b) = ¬F ∩ T = T ∩ T = T and get v |= A. By
definition v ∗ (¬A) = ¬v ∗ (A) = ¬T = F and v 6|= ¬A. This end the proof that
the formula A = (¬a ∩ b) is not independent from G.

Exercise 3.14
Given a set G = {a, (a ⇒ b)}.
Find a formula A that is independent from G.

Solution
Observe that truth assignment v such that v(a) = T, v(b) = T is the only
restricted model for G. So we have to come up with a formula A such that
there are two different truth assignments, v1 , v2 such that v1 |= G ∪ {A} and
v2 |= G ∪ {¬A}. Let’s think about as simple a formula as it could be, namely

108
let’s consider A = c, where c any propositional variable (atomic formula) dif-
ferent from a and b. G ∪ {A} = {a, (a ⇒ b), c} and any truth assignment v1 ,
such that v1 (a) = T, v1 (b) = T, v1 (c) = T is a model for G ∪ {c}. Likewise for
G ∪ {¬c} = {a, (a ⇒ b), ¬c}. Any v2 such that v2 (a) = T, v2 (b) = T, v2 (c) = F
is a model for G ∪ {¬c}. This proves that we have found the formula A = c that
is independent from G.

Here is a simple generalization of the Exercise 3.14.

Exercise 3.15
Find an infinite number of formulas that are independent from G = {((a ∩
b) ⇒ b), (a ∪ b), ¬a}.

Solution
First we have to find all v : V AR −→ {T, F } such that v |= {((a ∩ b) ⇒
b), (a ∪ b), ¬a}, i.e such that (shorthand notation) ((a ∩ b) ⇒ b) = T, (a ∪
b) = T, ¬a = T . Observe that |= ((a ∩ b) ⇒ b), so we have to consider only
(a ∪ b) = T, ¬a = T . This holds if and only if a = F and (F ∪ b) = T , i.e. if
and only if a = F and b = T. This proves that that vG such that vG (a) = F
and vG (b) = T is the only one restricted model for G. All possible models for
G must be extensions of vG . We define a countably infinite set of formulas (and
their negations) and corresponding extensions v of vG (restricted to to the set
of variables {a, b}) such that v |= G as follows.
Observe that all extensions of v of vG have as domain the infinitely countable set
V AR = {a1 , a2 , . . . , an . . . . }. We take as the infinite set of formulas in which
every formula is to be proved independent of G the set of atomic formulas

F0 = V AR − {a, b} = {a1 , a2 , . . . , an . . . . } − {a, b}.

Let c ∈ F0 . We define truth assignments v1 , v2 : V AR −→ {T, F } as follows

v1 (a) = v(a) = F, v1 (b) = v(b) = T, and v1 (c) = T for all c ∈ F0 .
v2 (a) = v(a) = F, v2 (b) = v(b) = T, and v2 (c) = F for all c ∈ F0 .
Obviously, v1 |= G ∪ {c} and v2 |= G ∪ {¬c} for all c ∈ F0 . What proves that the
set F0 is a countably infinite set of formulas independent from G = {((a∩b) ⇒
b), (a ∪ b), ¬a}.

3.4 Classical Tautologies and Equivalence of Lan-

guages
We first present here a set of most widely used classical propositional tautologies
which we will use, in one form or other, in our investigations in future chapters.

109
Another extended list of tautologies and their discussion is presented in Chapter
2.
As the next step we define notions of a logical equivalence and an equivalence of
languages. We prove that all of the languages

L{¬⇒} , L{¬∩} , L{¬∪} , L{¬,∩,∪,⇒} , L{¬,∩,∪,⇒,⇔} , L{↑} , L{↓}

are equivalent under classical semantics and hence can be used (and are) as
different languages for classical propositional logic.
We generalize these notions to the case of any extensional semantics M in sec-
tion 3.6. We also discuss and examine some particular many valued extensional
semantics and properties of their languages in section 3.5.

Some Tautologies
For any A, B ∈ F, the following formulas are tautologies.

Implication and Negation

(A ⇒ (B ⇒ A)), ((A ⇒ (B ⇒ C)) ⇒ ((A ⇒ B) ⇒ (A ⇒ C))),

((¬B ⇒ ¬A) ⇒ ((¬B ⇒ A) ⇒ B)) (3.12)

(A ⇒ A), (B ⇒ ¬¬B), (¬A ⇒ (A ⇒ B)), (A ⇒ (¬B ⇒ ¬(A ⇒ B))),
(¬¬B ⇒ B), ((A ⇒ B) ⇒ ((¬A ⇒ B) ⇒ B)), ((¬A ⇒ A) ⇒ A).

Disjunction, Conjunction

(A ⇒ (A ∪ B)), (B ⇒ (A ∪ B)), ((A ∩ B) ⇒ A), ((A ∩ B) ⇒ A),

((A ⇒ C) ⇒ ((B ⇒ C) ⇒ ((A ∪ B) ⇒ C))),

(((A ∩ B) ⇒ C) ⇒ (A ⇒ (B ⇒ C)),
(¬(A ∩ B) ⇒ (¬A ∪ ¬B)), ((¬A ∪ ¬B) ⇒ ¬(A ∩ B)), (3.13)
((¬A ∪ B) ⇒ (A ⇒ B)), ((A ⇒ B) ⇒ (¬A ∪ B)),
(A ∪ ¬A).

Contraposition (1)

((A ⇒ B) ⇔ (¬B ⇒ ¬A)), ((B ⇒ A) ⇔ (¬A ⇒ ¬B)). (3.14)

Contraposition (2)

((¬A ⇒ B) ⇔ (¬B ⇒ A)), ((A ⇒ ¬B) ⇔ (B ⇒ ¬A)). (3.15)

110
Double Negation
(¬¬A ⇔ A), (3.16)

Logical Equivalences

Logical equivalence is a very useful notion to use when we want to obtain new
formulas or new tautologies, if needed, on a base of some already known in a
way that guarantee preservation of the logical value of the initial formula. For
any formulas A, B, we say that are logically equivalent if they always have the
same logical value. We write it symbolically as A ≡ B. We have to remember
that the symbol ” ≡” not a logical connective. It is a metalanguage symbol for
saying ”A, B are logically equivalent”. This is a very useful symbol. It says
that two formulas always have the same logical value, hence it can be used in
the same way we use the equality symbol ” =.” Formally we define it as follows.

Definition 3.34 (Logical Equivalence)

For any A, B ∈ F, we say that the formulas A and B are logically equivalent
and denote it as A ≡ B
if and only if v ∗ (A) = v ∗ (B), for all v : V AR → {T, F }.

Observe that the following property follows directly from the definition 3.34.

Property 3.1
For any formulas A, B ∈ F,

A ≡ B if and only if |= (A ⇔ B)

For example we write the laws of contraposition (3.17), (3.18), and the law of
double negation (3.19) as logical equivalences as follows.

E - Contraposition (1), (2)

(A ⇒ B) ≡ (¬B ⇒ ¬A), (B ⇒ A) ≡ (¬A ⇒ ¬B). (3.17)

E - Contraposition (2)

(¬A ⇒ B) ≡ (¬B ⇒ A), (A ⇒ ¬B) ≡ (B ⇒ ¬A). (3.18)

E - Double Negation
¬¬A ≡ A. (3.19)

Logical equivalence is a very useful notion when we want to obtain new formu-
las, or tautologies, if needed, on a base of some already known in a way that

111
guarantee preservation of the logical value of the initial formula.

For example, we easily obtain equivalences for laws of E -Contraposition (3.18)

from equivalences for laws of E- Contraposition (3.17) and the E - Double Nega-
tion equivalence (3.19) as follows. (¬A ⇒ B) ≡(3.17) (¬B ⇒ ¬¬A) ≡(3.19) (¬B ⇒
A). We also have that (A ⇒ ¬B) ≡(3.17) (¬¬B ⇒ ¬A)≡(3.19) (B ⇒ ¬A). This
end the proof of E- Contraposition (3.18).
The correctness of the above procedure of proving new equivalences from the
known ones is established by the following theorem.

Theorem 3.12 (Equivalence Substitution)

Let a formula B1 be obtained from a formula A1 by a substitution of a formula
B for one or more occurrences of a sub-formula A of A1 , what we denote as

B1 = A1 (A/B).

Then the following holds for any formulas A, A1 , B, B1 ∈ F.

If A ≡ B, then A1 ≡ B1 . (3.20)

Proof
By the logical equivalence Definition 3.34 proving our theorem statement 8.44
is equivalent to proving that the implication

If v ∗ (A) = v ∗ (B), then v ∗ (A1 ) = v ∗ (B1 ) (3.21)

holds for all v : V AR → {T, F }.

Consider a truth assignment v. If v ∗ (A) 6= v ∗ (B), then the implication (3.21)
is vacuously true. If v ∗ (A) = v ∗ (B), then so v ∗ (A1 ) = v ∗ (B1 ), since B1 differs
from A1 only in containing B in some places where A1 contains A and the
implication (3.21) holds.

Example 3.17
Let A1 = (C ∪ D) and B = ¬¬C. By E - Double Negation equivalence (3.19)
we have that ¬¬C ≡ C. Let B1 = A1 (C/B) = A1 (C/¬¬C) = (¬¬C ∪ D). By
the Equivalence Substitution Theorem 3.12

(C ∪ D) ≡ (¬¬C ∪ D).

Equivalence of Languages

112
The next set of equivalences, or corresponding tautologies, correspond the notion
of definability of connectives discussed in section 3.3. For example, a tautology

|= ((A ⇒ B) ⇔ (¬A ∪ B))

makes it possible, via Property 3.1, to define implication in terms of disjunction

and negation. We state it in a form of logical equivalence and call it as follows.
Definability of Implication in terms of negation and disjunction:

(A ⇒ B) ≡ (¬A ∪ B) (3.22)

Observation 3.5 The direct proof of this and other Definability of Connectives
Equivalences presented here follow from the definability formulas developed in
the the proof of the Definability of Connectives Theorem 3.2, hence the names.

We are using the logical equivalence notion, instead of the tautology notion, as
it makes the manipulation of formulas much easier.
The equivalence 3.22 allows us, by the force of Theorem 3.12 to replace any
formula of the form (A ⇒ B) placed anywhere in another formula by a formula
(¬A ∪ B) while preserving their logical equivalence. Hence we can use the
equivalence (3.22) to transform a given formula containing implication into an
logically equivalent formula that does contain implication (but contains negation
and disjunction).
We usually use the equation 3.22 to transform any formula A of language con-
taining implication into a formula B of language containing disjunction and
negation and not containing implication at all, such that A ≡ B.

Example 3.18
Let A = ((C ⇒ ¬B) ⇒ (B ∪ C)).
We use equality (3.22) to transform A into a logically equivalent formula not
containing ⇒ as follows.
((C ⇒ ¬B) ⇒ (B ∪ C)) ≡ (¬(C ⇒ ¬B) ∪ (B ∪ C))) ≡ (¬(¬C ∪ ¬B) ∪ (B ∪ C))).

It means that for example that we can, by the Theorem 3.12 transform any
formula A of the language L1 = L{¬,∩,⇒} into a logically formula B of the
language L2 = L{¬,∩,∪} . In general, we say that we can transform a language
L1 into a logically equivalent language L2 if the following condition C1 holds.
C1: for any formula A of L1 , there is a formula B of L2 , such that A ≡ B.

Example 3.19
Let A = (¬A ∪ (¬A ∪ ¬B)). We also can use, in this case, the equivalence 3.22
as follows.

113
(¬A ∪ (¬A ∪ ¬B)) ≡ (¬A ∪ (A ⇒ ¬B)) ≡ (A ⇒ (A ⇒ ¬B)).
It means we eliminated disjunction from A by replacing it by logically equivalent
formula containing implication only.

Observe, that we can’t always use the equivalence (3.22) to eliminate any dis-
junction. For example, we can’t use it for a formula A = ((a ∪ b) ∩ ¬a).
In order to be able to transform any formula of a language containing disjunc-
tion (and some other connectives) into a language with negation and implication
(and some other connectives), but without disjunction we need the following log-
ical equivalence.

Definability of Disjunction in terms of negation and implication:

(A ∪ B) ≡ (¬A ⇒ B) (3.23)

Example 3.20
Consider a formula A = (a ∪ b) ∩ ¬a).
We use equality (3.23) to transform A into its logically equivalent formula not
containing ∪ as follows: ((a ∪ b) ∩ ¬a) ≡ ((¬a ⇒ b) ∩ ¬a).

In general, we use the equality 3.23 and Theorem 3.12 to transform any formula
C of the language L2 = L{¬,∩,∪} into a logically equivalent formula D of the
language L1 = L{¬,∩,⇒} . In general, the following condition hols.
C2: for any formula C of L2 , there is a formula D of L1 , such that C ≡ D.
The languages L1 and L2 for which we the conditions C1, C2 hold are logically
equivalent and denote it by L1 ≡ L2 .
We put it in a general, formal definition as follows.

Definition 3.35 (Equivalence of Languages)

Given two languages:
L1 = LCON1 and L2 = LCON2 , for CON1 6= CON2 .
We say that they are logically equivalent and denote it as L1 ≡ L2 if and
only if the following conditions C1, C2 hold.
C1 For every formula A of L1 , there is a formula B of L2 , such that A ≡ B,
C2 For every formula C of L2 , there is a formula D of L1 , such that C ≡ D.

Example 3.21
To prove the logical equivalence L{¬,∪} ≡ L{¬,⇒} we need two definability equiv-
alences (3.22 ) and (3.23), and the Theorem 3.12.

114
Exercise 3.16
To prove the logical equivalence L{¬,∩,∪,⇒} ≡ L{¬,∩,∪} we needed only the de-
finability equivalence (3.22).

Solution
The equivalence (3.22) proves, by Theorem 3.12 that for any formula A of
L{¬,∩,∪,⇒} there is B of L{¬,∩,∪} that equivalent to A, i.e. condition C1 holds.
Any formula A of language L{¬,∩,∪} is also a formula of L{¬,∩,∪,⇒} and of course
A ≡ A, so both conditions C1 and C2 of definition 3.35 are satisfied.

Exercise 3.17
Show that L{¬,∩} ≡ L{¬,⇒} .

Solution
The equivalence of languages holds by Theorem 3.12, Observation 3.5, and the
following two logical equalities. Definability of Conjunction in terms of
implication and negation and Definability of Implication in terms of con-
junction and negation:
(A ∩ B) ≡ ¬(A ⇒ ¬B) (3.24)
(A ⇒ B) ≡ ¬(A ∩ ¬B). (3.25)

Exercise 3.18
Show that L{¬,∩} ≡ L{¬,∪} .

Solution
Similarly, it is true by Theorem 3.12, Observation 3.5, and the following two log-
ical equalities. Definability of disjunction in terms of negation and conjunction
and definability of conjunction in terms of negation and disjunction:

(A ∪ B) ≡ ¬(¬A ∩ ¬B) (3.26)

(A ∩ B) ≡ ¬(¬A ∪ ¬B). (3.27)

Theorem 3.12, Observation 3.5, and definability of equivalence in terms of im-

plication and conjunction equality

(A ⇔ B) ≡ ((A ⇒ B) ∩ (B ⇒ A)). (3.28)

prove that, for example, L{¬,∩,∪,⇒} ≡ L{¬,∩,∪,⇒,⇔} .

Exercise 3.19
Show that L{¬,∩} ≡ L{↑} and L{¬,∩} ≡ L{↑}

115
Proof
We use the proof of Theorem 3.3 to prove the following definability equivalences
of ¬ and ∩ in terms of ↑:

¬A ≡ (A ↑ A), (A ∩ B) ≡ (A ↑ B) ↑ (A ↑ B) (3.29)

and definability equivalences of ¬ and ∪ in terms of ↓:

¬A ≡ (A ↓ A), (A ∪ B) ≡ (A ↓ B) ↓ (A ↓ B). (3.30)

This proves the condition C1 of definition 3.35.

The definability equivalences for fulfillment of the condition C2 are:

(A ↑ B) = ¬(A ∪ B) and (A ↑ B) = ¬(A ∪ B) (3.31)

Here are some more frequently used, important logical equivalences.

Idempotent
(A ∩ A) ≡ A, (A ∪ A) ≡ A,
Associativity

((A ∩ B) ∩ C) ≡ (A ∩ (B ∩ C)), ((A ∪ B) ∪ C) ≡ (A ∪ (B ∪ C)),

Commutativity

(A ∩ B) ≡ (B ∩ A), (A ∪ B) ≡ (B ∪ A),

Distributivity

(A ∩ (B ∪ C)) ≡ ((A ∩ B) ∪ (A ∩ C)), (A ∪ (B ∩ C)) ≡ ((A ∪ B) ∩ (A ∪ C)),

De Morgan Laws

¬(A ∪ B) ≡ (¬A ∩ ¬B), ¬(A ∩ B) ≡ (¬A ∪ ¬B).

Negation of Implication

¬(A ⇒ B) ≡ (A ∩ ¬B), (3.32)

De Morgan laws are named after A. De Morgan (1806 - 1871), an English

logician, who discovered analogous laws for the algebra of sets. They stated that
for any sets A,B the complement of their union is the same as the intersection
of their complements, and vice versa, the complement of the intersection of two
sets is equal to the union of their complements. The laws of the propositional
calculus were formulated later, but they are usually also called De Morgan Laws.
Observe that De Morgan Laws tell us how to negate disjunction and conjunction,
so the laws stating how to negate other connectives follows them.

116
Consider a tautology A: |= ((¬(A ⇒ B) ⇒ ¬A) ⇒ (A ⇒ B)).
We know by (3.22) that (A ⇒ B) ≡ (¬A ∪ B). By Theorem 3.12, if we replace
(A ⇒ B) by (¬A ∪ B) in A, the logical value of A will remain the same and
((¬(A ⇒ B) ⇒ ¬A) ⇒ (A ⇒ B)) ≡ ((¬(¬A ∪ B) ⇒ ¬A) ⇒ (¬A ∪ B)). Now
we use de Morgan Laws and Double Negation Laws and by Theorem 3.12 we
get ((¬(A ⇒ B) ⇒ ¬A) ⇒ (A ⇒ B)) ≡ ((¬(¬A ∪ B) ⇒ ¬A) ⇒ (¬A ∪ B)) ≡
(((¬¬A ∩ ¬B) ⇒ ¬A) ⇒ (¬A ∪ B)) ≡ (((A ∩ ¬B) ⇒ ¬A) ⇒ (¬A ∪ B)).
This proves that
|= (((A ∩ ¬B) ⇒ ¬A) ⇒ (¬A ∪ B)).

Exercise 3.20
Prove using proper logical equivalences that
(i) ¬(A ⇔ B) ≡ ((A ∩ ¬B) ∪ (¬A ∩ B)),
(ii) ((B ∩ ¬C) ⇒ (¬A ∪ B)) ≡ ((B ⇒ C) ∪ (A ⇒ B)).

Solution (i)
¬(A ⇔ B)≡(3.28) ¬((A ⇒ B) ∩ (B ⇒ A))≡de M organ (¬(A ⇒ B) ∪ ¬(B ⇒
A))≡(3.32) ((A ∩ ¬B) ∪ (B ∩ ¬A))≡commut ((A ∩ ¬B) ∪ (¬A ∩ B)).
Solution (ii)
((B ∩ ¬C) ⇒ (¬A ∪ B))≡(3.23) (¬(B ∩ ¬C) ∪ (¬A ∪ B))≡de M organ ((¬B ∪ ¬¬C) ∪
(¬A ∪ B))≡(3.19) ((¬B ∪ C) ∪ (¬A ∪ B))≡(3.23) ((B ⇒ C) ∪ (A ⇒ B)).

3.5 Many Valued Semantics: Lukasiewicz, Heyt-

ing, Kleene, Bohvar

Many valued logics in general and 3-valued logics in particular is an old object
of study which has its beginning in the work of a Polish mathematician Jan
Leopold Lukasiewicz in 1920. He was the first to define a 3 - valued semantics
for the language L{¬,∩,∪,⇒} of classical logic, and called it a three valued logic
for short. He left the problem of finding a proper axiomatic proof system for it
(i.e. the one that is complete with respect to his semantics) open. The same
happened to all other 3 - valued semantics presented here. They were also first
called 3 valued logics and this terminology is still widely used. Nevertheless, as
these logics were defined only semantically, i.e. defined by providing a semantics
for their languages we call them just semantics (for logics to be developed), not
logics. Existence of a proper axiomatic proof system for a given semantics and
proving its completeness is always a next open question to be answered (when
it is possible). A process of creating a logic (based on a given language) always
is three fold: we define semantics, create axiomatic proof system and prove

117
completeness theorem that established a relationship between semantics and
proof system.
The first of many valued logics invented were first presented in a semantical
form only for other components to be developed later. We can think about the
process of their creation as inverse to the creation of Classical Logic, Modal
Logics, the Intuitionistic Logic which existed as axiomatic systems longtime be-
fore invention of their formal semantics.

Formal definition of many valued extensional semantics M for the language L

we present and discuss here follows the extensional semantics Definition 3.7 in
general and the pattern of presented in detail for the classical case (Section 3.3 )
in particular. It consists of giving definitions of the following main components:
Step 1: given the language L we define a set of logical values and its distinguish
value T, and define all logical connectives of L
Step 2: we define notions of a truth assignment and its extension;
Step 3: we define notions of satisfaction, model, counter model;
Step 4: we define notions tautology under the semantics M.
We present here some of the historically first 3-valued extensional semantics,
called also 3-valued logics. They are named after their authors: Lukasiewicz,
Heyting, Kleene, and Bochvar.
The 3-valued semantics we define here enlist a third logical value, besides clas-
sical T, F . We denote this third value by ⊥, or m in case of Bochvar semantics.
We also assume that the third value is intermediate between truth and falsity,
i.e. that F <⊥< T and F < m < T.

There has been many of proposals relating both to the intuitive interpretation
of this third value ⊥. If T is the only designated value, the third value ⊥ cor-
responds to some notion of incomplete information, like undefined or unknown
and is often denoted by the symbol U or I. If, on the other hand, ⊥ corresponds
to inconsistent information, i.e. its meaning is something like known to be both
true and false then corresponding semantics takes both T and the third logical
value ⊥ as designated. In general, the third logical value denotes a notion of
”unknown”, ”uncertain”, ”undefined”, or even can express that ”we don’t have
a complete information”, depending on the context and motivation for the logic
we plan to develop. In all of presented here semantics we take T as designated
value, i.e. T is the value that defines the notion of satisfiability and tautology.

Lukasiewicz Semantics L

Motivation
Lukasiewicz developed his semantics (called logic) to deal with future contin-

118
gent statements. According to him, such statements are not just neither true
nor false but are indeterminate in some metaphysical sense. It is not only that
we do not know their truth value but rather that they do not possess one. Intu-
itively, ⊥ signifies that the statement cannot be assigned the value true of false;
it is not simply that we do not have sufficient information to decide the truth
value but rather the statement does not have one.

We define all the steps of the Definition3.7 in case of Lukasiewicz’ s semantics

to establish a pattern and proper notation. We leave the detailed steps of other
semantics as an exercise for the reader.

Step 1: L Connectives

The language of the semantics L is L{¬, ∪, ∩, ⇒} . The set LV of logical values

is {T, ⊥, F }. T is the distinguished value. We assume that the set of log0cal
values is ordered, i.e. that
F <⊥< T.

L Negation is a function ¬ : {T, ⊥, F } −→ {T, ⊥, F } such that

¬ ⊥=⊥, ¬T = F, ¬F = T.

L Conjunction is a function ∩ : {T, ⊥, F } × {T, ⊥, F } −→ {T, ⊥, F } such

that for any (x, y) ∈ {T, ⊥, F } × {T, ⊥, F }, we put

x ∩ y = min{x, x}.

L Disjunction is a function ∪ : {T, ⊥, F } × {T, ⊥, F } −→ {T, ⊥, F }, such

that for any (a, b) ∈ {T, ⊥, F } × {T, ⊥, F }, we put

x ∪ y = max{x, y}

L Implication is a function ⇒: {T, ⊥, F } × {T, ⊥, F } −→ {T, ⊥, F } such

that for any (x, y) ∈ {T, ⊥, F } × {T, ⊥, F }, we put

¬x ∪ y if x > y
x⇒y= (3.33)
T otherwise

We write function defining the connectives in a standard form of tables defining

operations in finite sets. We call these tables truth tables definition of proposi-
tional connectives, or L connectives truth tables for short.

L Connectives Truth Tables

119
 ∩ F ⊥ T
¬ F ⊥ T
F F F F
T ⊥ F
⊥ F ⊥ ⊥
T F ⊥ T

∪ F ⊥ T ⇒ F ⊥ T
F F ⊥ T F T T T
⊥ ⊥ ⊥ T ⊥ ⊥ T T
T T T T T F ⊥ T

Step 2: Truth Assignment, Truth Extension

A truth assignment is now any function v : V AR −→ {F, ⊥, T }. We define

its extension to the set F of all formulas as any function v ∗ : F −→ {T, F },
such that the following conditions are satisfied.

(1) for any a ∈ V AR, v ∗ (a) = v(a);

(2) for any A, B ∈ F,

v ∗ (¬A) = ¬v ∗ (A);
v ∗ ((A ∩ B)) = v ∗ (A) ∩ v ∗ (B);
v ∗ ((A ∪ B)) = v ∗ (A) ∪ v ∗ (B);
v ∗ ((A ⇒ B)) = v ∗ (A) ⇒ v ∗ (B).

Step 3: Satisfaction, Model, Counter-Model

We say that a truth assignment v : V AR −→ {F, ⊥, T } L satisfies a formula

A ∈ F if and only if v ∗ (A) = T . We denote it by v |=L A.
Any truth assignment v, v : V AR −→ {F, ⊥, T } such that v |=L A is called a
L model for A.
We say that a truth assignment v does not L satisfy a formula A ∈ F and
denote it by v 6|=L A, if and only if v ∗ (A) =
6 T.
Any truth assignment v, v : V AR −→ {F, ⊥, T } such that v 6 |=L A is called
a L counter- model for A.

Step 4: L Tautology

120
We define, for any A ∈ F, A is a L tautology if and only if v ∗ (A) = T for
all v : V AR −→ {F, ⊥, T }. We also say that A is a L tautology if and only if
all truth assignments v : V AR −→ {F, ⊥, T } are L models for A. We write the
statement ” A is a L tautology” symbolically as

|=L A.

As a next step we define, as we did in the case of classical semantics the notions
of restricted truth assignment and restricted models, (Definitions 3.26, 3.27) i.e.
we have the following.
Any function vA : V ARA −→ {F, ⊥, T }, such that vA |=L A ( vA 6 |=L A)
is called a restricted L model ( L counter model) for A, where V ARA is the set
of all propositional variables appearing in A. We call the function vA , a truth
assignment restricted to A, or restricted truth assignment for short.
We prove, in the same way we proved Theorem 3.8 in Section 3.3, the following
theorem that justifies the correctness of the truth tables L tautologies verifica-
tion method.

Theorem 3.13 (L Truth Tables)

For any formula A ∈ F,
|=L A if and only if vA |=L A for all vA : V ARA −→ {T, ⊥, F }, i.e.
|=L A if and only if all vA are restricted models for A.

Directly from Theorem 3.13 we get that the notion of L propositional tautology
is decidable, i.e. that the following holds.

Theorem 3.14 (Decidability)

For any formula A ∈ F, one has examine at most 3V ARA truth assignments
vA : V ARA −→ {F, ⊥, T } in order to decide whether |=L A, or |6 =L A, i.e.
the notion of L tautology is decidable.

We denote by LT the set of all L tautologies, i.e. we have that

LT = {A ∈ F : |=L A}. (3.34)

We just proved (Theorem 3.14) the correctness of the truth table tautology ver-
ification method for L semantics stated as follows.

L Truth Table Method

The verification method, called a truth table method consists of examination,
for any formula A, all possible truth assignments restricted to A. By Theorem
3.13 we have to perform at most 3|V ARA | steps. If we find a truth assignment
which does not evaluate A to T , i.e. evaluates A to F , or to ⊥, we stop the

121
process and give answer: 6|=L A. Otherwise we continue. If all truth assignments
evaluate A to T , we give answer: |=L A.
Consider, for example, a formula A: (a ⇒ a). There are 31 = 3 possible re-
stricted truth assignment v : {a} −→ {F, ⊥, T }. We list them, and evaluate
their value on the formula A in a form of an extended truth table as follows.

v a v ∗ (A) computation v ∗ (A)

v1 T T ⇒T =T T
v2 ⊥ ⊥⇒⊥= T T
v3 F F ⇒F =T T

This proves that the classical tautology (a ⇒ a) is also a L tautology, i.e.

|= (a ⇒ a) and |=L (a ⇒ a). (3.35)

Moreover (3.35) proves that the condition (3.3) of the definition 3.14 is fulfilled
and the L semantics is well defined. We put it as a separate fact.

Fact 3.5
The Lukasiewicz semantics L is well defined.

As a next step we can adopt all other classical tautology verification methods
from Section 3.3. It is a quite straightforward adaptation and we leave it a san
exercise. Moreover it works for all of many valued semantics presented here, as
does the Decidability Theorem 3.14.

When defining and developing a new logic the first question one asks is how it
relates and compares with the classical case, it means with the classical logic. In
case of new semantics (logics defined semantically) we describe this relationship
in terms of respective sets of tautologies.
Let LT, T denote the sets of all L and classical tautologies, respectively.

Theorem 3.15
The following relationship holds between classical and L tautologies:

LT 6= T and LT ⊂ T. (3.36)

Proof
Consider a formula (¬a ∪ a). It is obviously a classical tautology. Take any
truth assignment v : V AR −→ {F, ⊥, T } such that v(a) =⊥. By definition we
have that v ∗ (¬a ∪ a) = v ∗ (¬a) ∪ v ∗ (a) = ¬v(a) ∪ v(a) = ¬ ⊥ ∪ ⊥=⊥ ∪ ⊥=⊥.

122
This proves that v is a L counter-model for (¬a ∪ a) and hence 6|=L (¬a ∪ a).
This proves LT 6= T.
Observe now that if we restrict the values of functions defining L connectives to
the values T and F only, we get the functions defining the classical connectives.
It is directly visible when we compare the L and classical connectives truth
tables. This means that if v ∗ (A) = T for all v : V AR −→ {F, ⊥, T }, then
v ∗ (A) = T for any v : V AR −→ {F, T } and for any A ∈ F, i.e. LT ⊂ T.

Exercise 3.21
Use the fact that v : V AR −→ {F, ⊥, T } is such that v ∗ ((a ∩ b) ⇒ ¬b) =⊥
under L semantics to evaluate v ∗ (((b ⇒ ¬a) ⇒ (a ⇒ ¬b)) ∪ (a ⇒ b)). Use
shorthand notation.

Solution
Observe that ((a ∩ b) ⇒ ¬b) =⊥ in two cases.
c1: (a ∩ b) =⊥ and ¬b = F .
c12: (a ∩ b) = T and ¬b =⊥ .
Consider c1. We have ¬b = F , i.e. b = T , and hence (a ∩ T ) =⊥ if and only
if a =⊥. We get that v is such that v(a) =⊥ and v(b) = T . We evaluate (in
short hand notation) v ∗ (((b ⇒ ¬a) ⇒ (a ⇒ ¬b)) ∪ (a ⇒ b)) = (((T ⇒ ¬ ⊥) ⇒
(⊥⇒ ¬T )) ∪ (⊥⇒ T )) = ((⊥⇒⊥) ∪ T ) = T .
Consider c2. We have ¬b =⊥, i.e. b =⊥, and hence (a∩ ⊥) = T what is
impossible, hence v from case c1 is the only one, and v ∗ (((b ⇒ ¬a) ⇒ (a ⇒
¬b)) ∪ (a ⇒ b)) = T.

L4 Semantics

We define the semantics L4 as follows. The language is L = L{¬,⇒,∪,∩} .

The logical connectives ¬, ⇒, ∪, ∩ of L4 as the following operations in the set
{F, ⊥1 , ⊥2 , T }, where {F < ⊥1 < ⊥2 < T }.

L4 Negation is a function such that ¬⊥1 = ⊥1 , ¬⊥2 = ⊥2 , ¬F = T, ¬T = F.

L4 Conjunction is a function such that for any x, y ∈ {F, ⊥1 , ⊥2 , T }, x ∩ x =
min{x, y}.
L4 Disjunction is a function such that for any x, y ∈ {F, ⊥1 , ⊥2 , T }, x ∪ y =
min{x, y}.
L4 Implication is a function such that for any x, y ∈ {F, ⊥1 , ⊥2 , T },

¬x ∪ y if x > y
x⇒y= (3.37)
T otherwise

123
Exercise 3.22
Here are 3 simple problems.
1. Write down L4 Connectives Truth Tables.
2. Give an example of a L4 tautology.
3. We know that the formula ((a ⇒ b) ⇒ (¬a ∪ b)) is a classical tautology, i.e.
|= ((a ⇒ b) ⇒ (¬a ∪ b)). Verify whether |=L4 ((a ⇒ b) ⇒ (¬a ∪ b)).

Solution 1.
Here are L4 Connectives Truth Tables.

∩ F ⊥1 ⊥2 T
F F F F F
¬ F ⊥1 ⊥2 T
⊥1 F ⊥1 ⊥1 ⊥1
T ⊥1 ⊥2 F
⊥2 F ⊥1 ⊥2 ⊥2
T F ⊥1 ⊥2 T

∪ F ⊥1 ⊥2 T ⇒ F ⊥1 ⊥2 T
F F ⊥1 ⊥2 T F T T T T
⊥1 ⊥1 ⊥1 ⊥2 T ⊥1 ⊥1 T T T
⊥2 ⊥2 ⊥2 ⊥2 T ⊥2 ⊥2 ⊥2 T T
T T T T T T F ⊥1 ⊥2 T

Solution 2.
Observe that by definition of L4 implication we get x ⇒ x = T for all x ∈
{F, ⊥1 , ⊥2 , T }. Hence v ∗ ((a ⇒ a)) = v(a) ⇒ v(a) = T for all v, what proves
|=L4 (a ⇒ a).
Solution 3.
We use the Proof by Contradiction Method (section 3.3) to verify whether
|=L4 ((a ⇒ b) ⇒ (¬a ∪ b)). Observe that it applied to any situation, as its
correctness is based on our classical reasoning. Assume that 6|=L4 ((a ⇒ b) ⇒
(¬a∪b)). Let v : V AR −→ {F, ⊥1 , ⊥2 , T }, such that v ∗ (((a ⇒ b) ⇒ (¬a∪b))) 6=
T. Observe that in L4 semantics, for any formula A ∈ F, v ∗ (A) 6= T gives us
three possibilities v ∗ (A) = F, v ∗ (A) = ⊥1 , or v ∗ (A) = ⊥2 to consider ( as
opposed to one case in classical case). It is a lot of work, but still less then listing
and evaluating 42 = 16 possibilities of all restricted truth assignment. Moreover,
our formula is a classical tautology, hence we know that it evaluates in T for all
combinations of T and F. A good strategy is to examine first some possibilities

124
of evaluating variables a, b for combination of ⊥1 , ⊥2 with hope of finding a
counter model. So let’s v be a truth assignment such that v(a) = v(b) = ⊥1 .
We evaluate v ∗ ((a ⇒ b) ⇒ (¬a ∪ b)) = ((⊥1 ⇒ ⊥1 ) ⇒ (¬⊥1 ∪ ⊥1 )) = (T ⇒
(⊥1 ∪ ⊥1 )) = (T ⇒ ⊥1 ) = ⊥1 . This proves that v is a counter-model for our
formula. Observe that the v serves also as a L counter model for A when we
put ⊥1 =⊥ and so we get

|= ((a ⇒ b) ⇒ (¬a ∪ b)), 6|=L4 ((a ⇒ b) ⇒ (¬a ∪ b)), 6|=L ((a ⇒ b) ⇒ (¬a ∪ b))

Obviously, any v such that v(a) = v(b) = ⊥2 is also a counter model for A, as
v ∗ ((a ⇒ b) ⇒ (¬a ∪ b)) = ((⊥2 ⇒ ⊥2 ) ⇒ (¬⊥2 ∪ ⊥2 )) = (T ⇒ (⊥2 ∪ ⊥2 )) =
(T ⇒ ⊥2 ) = ⊥2 . We leave it as an exercise to find all possible counter models
for A.

Heyting Semantics H

Motivation
We discuss here the semantics H because of its connection with intuitionistic
logic. The H connectives are such that they represent operations in a cer-
tain 3 element algebra, historically called a 3 element pseudo-boolean algebra.
Pseudo-boolean algebras were created by McKinsey and Tarski in 1948 to pro-
vide semantics for the intuitionistic logic. The intuitionistic logic, the most
important rival to the classical logic was defined and developed by its inventor
Brouwer and his school in 1900s as a proof system only. Heyting provided its
first axiomatization which everybody accepted. McKinsey and Tarski proved
the completeness of the Heyting axiomatization with respect to their pseudo
boolean algebras semantics. The pseudo boolean algebras are also called Heyt-
ing algebras in his honor and so is our semantics H.

We say, that formula A is an intutionistic tautology if and only if it is valid in all

pseudo boolean (Heying) algebras. The pseudo boolean algebras are defined in
a very general and mathematically sophisticated way. Their universe (it means
the set of logical values) can be any non empty set. Their operations that cor-
respond to L = L{¬,⇒,∪,∩} connectives must fulfill a set of special properties
(axioms). But we can prove that the operations defined by H connectives form
a 3-element pseudo boolean algebra with the universe U = {F, ⊥, T }. Hence,
if A is an intuitionistic tautology, then in it is also valid (tautologically true)
for the H semantics, i.e. all intuitionistic propositional tautologies are also the
H semantics tautologies. It means that our H is a good candidate for finding
counter models for the formulas that might not be intuitionistic tautologies.

The other type of models, called Kripke models were defined by Kripke in 1964
and were proved later to be equivalent to the pseudo-boolean models. They are
very general and serve as a method of defining not extensional semantics for

125
various classes of logics. That includes semantics for a great number of modal,
knowledge, belief logics, and many new logics developed and being developed
by computer scientists.

H Connectives

We adopt the same language as in case of classical and Lukasiewicz’s L seman-

tics, i.e. L = L{¬,⇒,∪,∩} . We assume, as before, that {F <⊥< T }.

The connectives ¬, ∪, ∩ of H are defined as in L semantics. They are func-

tions defined by formulas x ∪ y = max{x, y}, x ∩ y = min{x, y}, for any
x, y ∈ {F, ⊥, T }.

The definition of implication and negation for H semantics differs L semantics

and we define them as follows.

H Implication is a function ⇒: {T, ⊥, F } × {T, ⊥, F } −→ {T, ⊥, F } such

that for any (x, y) ∈ {T, ⊥, F } × {T, ⊥, F }, we put

T if x ≤ y
x⇒y= (3.38)
y otherwise
H negation is a function ¬ : {F, ⊥, T } −→ {F, ⊥, T }, such that
¬a = a ⇒ F.
The truth tables for H disjunction and conjunction are hence the same as cor-
responding L tables and the truth tables for H implication and negation are as
follows.

⇒ F ⊥ T
¬ F ⊥ T
F T T T
T F F
⊥ F T T
T F ⊥ T

For Steps 2 - 4 of the definition 3.7 we adopt definitions established for L

semantics. For example, we define the notion of H tautology as follows.

Definition 3.36 ( H Tautology)

For any formula A ∈ F,
A is a H tautology if and only if v ∗ (A) = T, for all v : V AR −→ {F, ⊥, T },
i.e. v |=H A for all v. We write
|=H A

126
to denote that a formula A is an H tautology.

We leave it as an exercise to the reader to prove, in the same way as in case

of classical semantics (section3.3) the following theorems that justify the truth
table method of verification and the decidability theorem for H.

Theorem 3.16 (H Truth Tables)

For any formula A ∈ F,
|=H A if and only if vA |=H A for all vA : V ARA −→ {T, ⊥, F }, i.e.
|=H A if and only if all vA are restricted models for A.

Theorem 3.17 (H Decidability)

For any formula A ∈ F, one has examine at most 3V ARA truth assignments
vA : V ARA −→ {F, ⊥, T } in order to decide whether |=H A, or 6|=H A, i.e.
the notion of H tautology is decidable.

We denote by HT the set of all H tautologies, i.e.

HT = {A ∈ F : |=H A}.

The following fact establishes relationship between classical and H tautologies.

Theorem 3.18
Let HT, LT, T denote the sets of all H, L, and classical tautologies, respectively.
Then the following relationship holds.
HT 6= LT, HT 6= T, and HT ⊂ T. (3.39)

Proof
A formula (¬a ∪ a) a classical tautology and not an H tautology. Take any
truth assignment v : V AR −→ {F, ⊥, T } such that v(a) =⊥. We evaluate
is v ∗ ((¬a ∪ a) = ¬ ⊥ ∪ ⊥= F ∪ ⊥=⊥ This proves that (¬a ∪ a) 6∈ HT and
hence HT 6= T. Directly from the definition of H connectives we get that if
we restrict the values of the functions defining them T and F only, we get the
functions defining the classical connectives. Hence for any formula A ∈ TH we
have that A ∈ TH and LT ⊂ T. A formula (¬¬a ⇒ a) is a L tautology and
not an H tautology by easy evaluation as presented in example 3.23 and (3.40).
This proves HT 6= LT.

Exercise 3.23
We know that v : V AR −→ {F, ⊥, T } is such that v ∗ ((a ∩ b) ⇒ (a ⇒ c)) =⊥
under H semantics.
Evaluate v ∗ (((b ⇒ a) ⇒ (a ⇒ ¬c)) ∪ (a ⇒ b)). You can use a short hand
notation.

127
Solution
By definition of H connectives we have that for any v, v ∗ ((a∩b) ⇒ (a ⇒ c)) =⊥
if and only if a ∩ b) = T and (a ⇒ c) =⊥ if and only if a = T, b = T and
(T ⇒ c) =⊥ if and only if c =⊥. Hence v ∗ ((a ∩ b) ⇒ (a ⇒ c)) =⊥ if and only
if a = T, b = T, c =⊥ . We evaluate v ∗ (((b ⇒ a) ⇒ (a ⇒ ¬c)) ∪ (a ⇒ b)) =
(((T ⇒ T ) ⇒ (T ⇒ ¬ ⊥)) ∪ (T ⇒ T )) = ((T ⇒ (T ⇒ F )) ∪ T ) = T .

Exercise 3.24
We know that the following formulas are basic classical tautologies

|= (a ∪ ¬a), |= (¬¬a ⇒ a), |= ((a ⇒ b) ⇒ (¬a ∪ b)). (3.40)

Use the H semantics to prove that none of them is intuitionostic tautology.

Solution Any v : V AR −→ {F, ⊥, T } such that v(a) = v(b) =⊥ is an H

counter model for all of the formulas. We evaluate (in shorthand notation) it
as follows. ⊥ ∪¬ ⊥=⊥ ∪F =⊥6= T, ¬¬ ⊥⇒⊥= ¬F ⇒⊥= T ⇒⊥=⊥6= T ,
(⊥⇒⊥) ⇒ (¬ ⊥ ∪ ⊥) = T ⇒ (¬ ⊥ ∪ ⊥) = T ⇒ (F ∪ ⊥) = T ⇒⊥=⊥6= T. We
hence proved by the fact ”if a given formula A is not the H semantics tautol-
ogy, it is not intuitionistic tautology” that none of classical tautologies (3.40)
is neither intuitionostic nor H tautology.

The H semantics can serve as a tool of proving that some formulas are not
intutionistic tautologies, but it is not a universal one

Example 3.22
We know that the classical tautology (¬(a ∩ b) ⇒ (¬a ∪ ¬b)) is not intuitionistic
tautology, but nevertheless |=H (¬(a ∩ b) ⇒ (¬a ∪ ¬b)).

Proof
We use the Proof by Contradiction Method (section 3.3.1) and shorthand no-
tation. Assume that 6|=H (¬(a ∩ b) ⇒ (¬a ∪ ¬b)). Let v : V AR −→ {F, ⊥, T }
such that v ∗ ((¬(a ∩ b) ⇒ (¬a ∪ ¬b))) 6= T . We have to consider two cases:
c1 v ∗ ((¬(a ∩ b) ⇒ (¬a ∪ ¬b))) =⊥ and c2 v ∗ ((¬(a ∩ b) ⇒ (¬a ∪ ¬b))) = F .
If we get a contradiction in both cases we have proved |=H (¬(a∩b) ⇒ (¬a∪¬b)).

Consider case c1. By definition of ⇒ we have that v ∗ ((¬(a∩b) ⇒ (¬a∪¬b))) =⊥

if and only if ¬(a∩b) = T and ¬a∪¬b =⊥ if and only if a∩b = F and ¬a∪¬b =⊥.
Let’s look ¬a ∪ ¬b =⊥. This is possible in 3 cases. 1. ¬a =⊥ and ¬b =⊥. Con-
tradiction with the definition of ⊥ as ¬x 6=⊥ for all x ∈ {F, ⊥, T }. 2. ¬a =⊥
and ¬b = F . Contradiction with the definition of ⊥. 3. ¬a = F and ¬b =⊥.

128
Contradiction with the definition of ⊥. This proves that case c1 always leads
to contradiction.

Consider case c2. By definition of ⇒ we have that v ∗ ((¬(a ∩ b) ⇒ (¬a ∪ ¬b))) =

F if and only if 1. ¬(a∩b) =⊥, ¬a∪¬b = F . Contradiction. 2. ¬(a∩b) = T and
¬a ∪ ¬b = F if and only if a ∩ b = F and ¬a ∪ ¬b = F . Observe that a ∩ b = F
in 3 cases. Two involve only T, F and we get a contradiction as in classical case
(our formula is classical tautology). We have hence to consider only the cases
when a =⊥, b = F and a = F, b =⊥. They both lead to the contradiction with
¬a ∪ ¬b = F . This proves that case c2 always leads to contradiction and it ends
the proof.

We can of course also use the Truth Tables Method that involves listing and
evaluating all of 23 = B restricted truth assignments.

Kleene Semantics K

Kleene’s logic semantics was originally conceived to accommodate undecided

mathematical statements.

Motivation
In Kleene’s semantics the third logical value ⊥, intuitively, represents undecided.
Its purpose is to signal a state of partial ignorance. A sentence a is assigned a
value ⊥ just in case it is not known to be either true of false.
For example, imagine a detective trying to solve a murder. He may conjecture
that Jones killed the victim. He cannot, at present, assign a truth value T or
F to his conjecture, so we assign the value ⊥, but it is certainly either true of
false and ⊥ represents our ignorance rather then total unknown.

K Connectives
We adopt the same language as in a case of classical, Lukasiewicz’s L, and Heyt-
ing H semantics, i.e. L = L{¬,⇒,∪,∩} .

We assume, as before, that {F <⊥< T }. The connectives ¬, ∪, ∩ of K

are defined as in L, H semantics. They are functions defined by formulas
x ∪ y = max{x, y}, x ∩ y = min{x, y}, for any x, y ∈ {F, ⊥, T }, and

¬ ⊥=⊥, ¬F = T, ¬T = F.

129
The K implication is defined by the same formula as the classical, i.e.

x ⇒ y = ¬x ∪ y. (3.41)

for any x, y ∈ {F, ⊥, T }.

The connectives truth tables for the K negation, disjunction and conjunction
are the same as the corresponding tables for L, H and K implication table is
as follows.

⇒ F ⊥ T
F T T T
⊥ ⊥ ⊥ T
T F ⊥ T

For Steps 2 - 4 of the definition of K semantics we follow the general M

semantics definition 3.7, or adopt its particular case of L semantics definition.
For example, we define the notion of K tautology as follows.

Definition 3.37 ( K Tautology)

For any formula A ∈ F,
A is a K tautology if and only if v ∗ (A) = T, for all truth assignments
v : V AR −→ {F, ⊥, T }, i.e. v |=K A for all v.

We write
|=K A
to denote that A is a K tautology. We prove, in the same way as in case of L
semantics the following theorems that justify truth table method of verification
and decidability theorem for K.

Theorem 3.19 (K Truth Tables)

For any formula A ∈ F,
|=K A if and only if vA |=K A for all vA : V ARA −→ {T, ⊥, F }, i.e.
|=K A if and only if all vA are restricted models for A.

Directly from Theorem 3.19 we get that the notion of K propositional tautology
is decidable, i.e. that the following holds.

Theorem 3.20 (K Decidability)

For any formula A ∈ F, one has examine at most 3V ARA truth assignments
vA : V ARA −→ {F, ⊥, T } in order to decide whether |=L A, or |6 =K A, i.e.
the notion of K tautology is decidable.

130
We write
KT = {A ∈ F : |=K A}
to denote the set of all K tautologies. The following establishes relationship
between L, K, and classical tautologies.

Theorem 3.21
Let LT, T, KT denote the sets of all L, classical, and K tautologies, respec-
tively. Then the following relationship holds.

LT 6= KT, KT 6= T, and KT ⊂ T. (3.42)

Proof
Obviously |= (a ⇒ a) and also by (3.35) |=L (a ⇒ a). Consider now any
v such that v(a) =⊥. We evaluate in K semantics v ∗ (a ⇒ a) = v(a) ⇒
v(a) =⊥⇒⊥=⊥. This proves that 6 |=K (a ⇒ a) and hence the first two rela-
tionships in (3.42) hold. The third one follows directly from the the fact that,
as in the L case, if we restrict the functions defining K connectives to the values
T and F only, we get the functions defining the classical connectives.

Exercise 3.25
We know that formulas ((a ∩ b) ⇒ a), (a ⇒ (a ∪ b)), (a ⇒ (b ⇒ a)) are classical
tautologies. Show that none of them is K tautology.

Solution Consider any v such that v(a) = v(b) =⊥. We evaluate (in short
hand notation) v ∗ (((a ∩ b) ⇒ a) = (⊥ ∩ ⊥) ⇒⊥=⊥⇒⊥=⊥⇒⊥⊥, v ∗ ((a ⇒
(a ∪ b))) =⊥⇒ (⊥ ∪ ⊥) =⊥⇒⊥=⊥, and v ∗ ((a ⇒ (b ⇒ a))) = (⊥⇒ (⊥⇒⊥
) =⊥⇒⊥=⊥. This proves that v such that v(a) = v(b) =⊥ is a counter model
for all of them. We generalize this example and prove that in fact a similar truth
assignment can serve as a counter model for not only any classical tautology,
but also for any formula A of L{¬,⇒,∪,∩} .

Theorem 3.22
For any formula A ∈ F, 6|=K A, i.e. the set of all K tautologies is empty. We
write it as
KT = ∅.

Proof
We show that a truth assignment v : V AR −→ {F, ⊥, T }, such that v(a) =⊥
for all a ∈ V AR is a counter model for any A ∈ F. We carry the proof the by
mathematical induction over the degree d(A) of the formula A.
Base Case: n=1 i.e. d(A) = 1. In this case we have that A = ¬a for any
a ∈ V AR, or A = (a ◦ b) for a, b ∈ V AR, ◦ ∈ {∪, ∩, ⇒}.

131
We evaluate: v ∗ (A) = v ∗ (¬a) = ¬v ∗ (a) = ¬ ⊥=⊥, v ∗ (a ◦ b) = v ∗ (a) ◦ v ∗ (b) =⊥
◦ ⊥=⊥ . This proves that the Base Case holds.
Inductive assumption: v ∗ (B) =⊥ for all B such that d(B) = k and 1 ≤ k < n.
Inductive thesis: v ∗ (A) =⊥ for any A such that d(A) = n.
Let A be such that d(A) = n. We have two cases to consider.
Case 1. A = ¬B, so d(B) = n − 1 < n. By inductive assumption v ∗ (B) =⊥.
Hence v ∗ (A) = v ∗ (¬B) = ¬v ∗ (B) = ¬ ⊥=⊥ and inductive thesis holds.
Case 2. A = (B ◦ C) for B, C ∈ F, ◦ ∈ {∪, ∩, ⇒} (and d(A) = n). Let
d(B) = k1 , d(C) = k2 . Hence d(A) = d(B ◦ C) = k1 + k2 + 1 = n. We
get that k1 + k2 = n − 1 < n. From k1 + k2 < n we get that k1 < n and
k2 < n. Hence by inductive assumption v ∗ (B) =⊥ and v ∗ (C) =⊥. We evalu-
ate: v ∗ (A) = v ∗ (B ◦ C) = v ∗ (B) ◦ v ∗ (C) =⊥ ◦ ⊥=⊥. This ends the proof.

Observe that the theorem 3.22 does not invalidate relationships (3.42). They
become now perfectly true statements

LT 6= ∅, T 6= ∅, and ∅ ⊂ T.

But when we develop a logic by defining its semantics we must make sure for
semantics to be such that it has a non empty set of its tautologies. The semantics
K is an example of a correctly and carefully defined semantics that is not well
defined in terms of the definition 3.14. We write is as separate fact.

Fact 3.6
The Kleene semantics K is not well defined.

K semantics also provides a justification for a need of introducing the definition

3.14 as a distinction between correctly and well defined semantics. This is the
main reason why it is included here.

Bochvar semantics B

Bochvar’s 3-valued logic was directly inspired by considerations relating to se-

mantic paradoxes. Here is the motivation for definition of its semantics.

Motivation

Consider a semantic paradox given by a sentence: this sentence is false. If it

is true it must be false, if it is false it must be true. There have been many
proposals relating to how one may deal with semantic paradoxes. Bohvar’s
proposal adopts a strategy of a change of logic. According to Bochvar, such

132
sentences are neither true of false but rather paradoxical or meaningless. The
semantics follows the principle that the third logical value, denoted now by m
is in some sense ”infectious”; if one one component of the formula is assigned
the value m then the formula is also assigned the value m.
Bohvar also adds an one argument assertion operator S that asserts the logical
value of T and F , i.e. S F = F , S T = T and it asserts that meaningfulness is
false, i.e S m = F .

Language LB

The language of B semantics differs from all previous languages in that it con-
tains an extra one argument assertion connective S added to the usual set
{¬, ⇒, ∪, ∩} of the language L = L{¬,S,⇒,∪,∩} of all previous semantics.
LB = L{¬,S,⇒,∪,∩} . (3.43)
The set LV of logical values is {T, m, F }. T is the distinguished value.

B Connectives

We define the connectives of LB the functions defined in the set {F, mT } by the
following truth tables.

B Connectives Truth Tables

∩ F m T ∪ F m T
¬ F m T
F F m F F F m T
T m F
m m m m m m m m
T F m T T T m T

⇒ F m T
S F m T
F T m T
F F T
m m m m
T F m T

For all other steps of definition of B semantics we follow the standard way estab-
lished for extensional M semantics, we did in all previous cases. In particular
we define the notion of B tautology as follows.

Definition 3.38
A formula A of LB is a B tautology if and only if v ∗ (A) = T , for all

133
v : V AR −→ {F, m, T }, i.e. if all variable assignments v are B models for A.

We write
|=B A
to denote that A is an B tautology.
We, prove, in the same way as for all previous logics semantics, the following
theorems that justify the truth table method of verification and decidability for
B tautologies.

Theorem 3.23 (B Truth Tables)

For any formula A of LB ,
|=B A if and only if vA |=B A for all vA : V ARA −→ {F, m, T }.

Theorem 3.24 (Decidability)

For any formula A of LB , one has examine at most 3V ARA truth assignments
v : V ARA −→ {F, m, T } in order to decide whether |=B A, or 6|=B A, i.e. the
notion of B tautology is decidable.

Let denote by FB the set of formulas of the language LB and by BT the set of
all B tautologies:
BT = {A ∈ FB : |=B A}.

Which formulas (if any) are the B tautologies is more complicated to determine
then in the case previous semantics because we have the following Fact 3.7.

Fact 3.7
For any formula A ∈ FB which do not contain a connective S, i.e. for any
formula A of the language L{¬,⇒,∪,∩} , 6|=B A.

Proof We show that a truth assignment v : V AR −→ {F, m, T }, such that

v(a) = m for all a ∈ V AR is a counter model for any A ∈ F. The proof the by
mathematical induction over the degree d(A) of the formula A is similar to the
proof of Theorem 3.22 and is left to the reader as an exercise.

By the Fact 3.7 for a formula to be considered to be a B tautology, it must

contain the connective S. We get by easy evaluation that |=B (Sa ∪ ¬Sa). This
proves that BT 6= ∅ and the B semantics is well defined by definition 3.14.
Of course not all formulas containing the connective S are B tautologies, for
example
6 |=B (a ∪ ¬Sa), 6 |=B (Sa ∪ ¬a), 6 |=B (Sa ∪ S¬a),
as any truth assignment v, such that v(a) = m is a counter model for all of them,
because m∪x = m for all x ∈ {F, m, T } and Sm∪S¬m = F ∪Sm = F ∪F = F .

134
3.6 M Tautologies, M Consistency, and M Equiv-
alence of Languages
The classical truth tables verification method a and classical decidability theo-
rem hold in a proper form in all of L. H, K and B semantics as it was discussed
separately for each of them. We didn’t discuss other classical tautologies veri-
fication methods of substitution and generalization. We do it now in a general
and unifying way for a special case of an extensional M semantics, namely for
any semantics M with a finite set of logical values.

3.6.1 M Tautologies Verification Methods

Given an extensional semantics M defined for a propositional language LCON

with the set F of formulas and a finite, non empty set LV of logical values.
We introduce, as we did in classical, and other cases a notion of a restricted
model (definition 3.26) and prove, in a similar way as we proved theorem 3.8 the
following theorem that justifies the correctness of the M truth tables tautologies
verification method.

Theorem 3.25 (M Truth Tables)

For any formula A ∈ F,
|=M A if and only if vA |=M A for all vA : V ARA −→ LV, i.e.
|=M A if and only if all vA are restricted models for A.

M Truth Table Method

A verification method, called a M truth table method consists of examination,
as in the classical case, for any formula A, all possible M truth assignments
restricted to A. By theorem 3.25 we have to perform at most |LV ||V ARA | steps.
If we find a restricted truth assignment which evaluates A to a value different
then T , we stop the process and give answer: 6|=M A. Otherwise we continue.
If all M truth assignments restricted to A evaluate A to T , we give answer:
|=M A.

Example 3.23
Consider a formula (¬¬a ⇒ a) and H semantics. We evaluate

v a v ∗ (A) computation v ∗ (A)

v1 T ¬¬T ⇒ T = ¬F ⇒ T = F ⇒ T = T T
v2 ⊥ ¬¬ ⊥⇒⊥= ¬F ⇒⊥= T ⇒⊥=⊥ ⊥

It proves that 6|=H (¬¬a ⇒ a).

135
Example 3.24
Consider a formula (¬¬a ⇒ a) and L semantics. We evaluate

v a v ∗ (A) computation v ∗ (A)

v1 T ¬¬T ⇒ T = ¬F ⇒ T = F ⇒ T = T T
v2 ⊥ ¬¬ ⊥⇒⊥= ¬ ⊥⇒⊥=⊥⇒⊥= T T
v3 F ¬¬F ⇒ F = ¬T ⇒ F = F ⇒ F = T T

It proves that |=L (¬¬a ⇒ a).

We also proved that the set HT of all H tautologies is different from the set set
LT of all L tautologies, i.e.
LT 6= HT (3.44)

Directly from Theorem 3.25 and the above we get that the notion of M propo-
sitional tautology is decidable, i.e. that the following holds.

Theorem 3.26 (M Decidability)

For any formula A ∈ F, one has examine at most |LV |V ARA truth assignments
vA : V ARA −→ LV in order to decide whether |=M A, or 6|=M A, i.e. the
notion of M tautology is decidable.

M Proof by Contradiction Method

In this method, in order to prove that |=M A we proceed as follows. We assume
that 6|=M A. We work with this assumption. If we get a contradiction, we
have proved that 6|=M A is impossible. We hence proved |=M A. If we do not
get a contradiction, it means that the assumption 6|=M A is true, i.e. we have
proved that A is not M tautology.
Observe that correctness of his method is based on a correctness of classical
reasoning. Its correctness is based on the Reductio ad Absurdum classical tau-
tology |= ((¬A ⇒ (B ∩ ¬B)) ⇒ A). The contradiction to be obtained follows
from the properties of the M semantics under consideration.

Substitution Method
The Substitution Method allows us to obtain, as in a case of classical semantics
new M tautologies from formulas already proven to be M tautologies. The
theorem 3.27 and its proof is a straightforward modification of the classical
proof (theorem 3.27) and we leave it as an exercise to the reader. It assesses
the validity of the substitution method. In order to formulate and prove it we
first remind of the reader of needed notations.
Let A ∈ F be a formula and V ARA = {a1 , a2 , ...an } be the set of all proposi-
tional variables appearing in A. We will denote it by A(a1 , a2 , ...an ). Given a

136
formula A(a1 , a2 , ...an ), and A1 , ...An be any formulas. We denote by

A(a1 /A1 , ..., an /An )

the result of simultaneous replacement (substitution) in A(a1 , a2 , ...an ) variables

a1 , a2 , ...an by formulas A1 , ...An , respectively.

Theorem 3.27
For any formulas A(a1 , a2 , ...an ), A1 , . . . , An ∈ F,
If |=M A(a1 , a2 , ...an ) and B = A(a1 /A1 , ..., an /An ), then |=M B.

We have proved (exercise 3.24) that the formula D(a) = (¬¬a ⇒ a) is L tau-
tology. By the above theorem 3.27 we get that D(a/A) = (¬¬A ⇒ A) is also L
tautology for any formula A ∈ F. We hence get the following.

Fact 3.8
For any A ∈ F, |=L (¬¬A ⇒ A).

M Generalization Method
In this method we represent, if it is possible, a given formula as a particular
case of some simpler general formula. Hence the name Generalization Method.
We then use other methods to examine the simpler formula thus obtained.

Exercise 3.26
Prove that
|=L (¬¬(¬((a ∩ ¬b) ⇒ ((c ⇒ (¬f ∪ d)) ∪ e)) ⇒ ((a ∩ ¬b) ∩ (¬(c ⇒ (¬f ∪
d)) ∩ ¬e))) ⇒ (¬((a ∩ ¬b) ⇒ ((c ⇒ (¬f ∪ d)) ∪ e)) ⇒ ((a ∩ ¬b) ∩ (¬(c ⇒
(¬f ∪ d)) ∩ ¬e)))).

Solution
Observe that our formula is a particular case of a more general formula (¬¬A ⇒
A) for A = (¬((a ∩ ¬b) ⇒ ((c ⇒ (¬f ∪ d)) ∪ e)) ⇒ ((a ∩ ¬b) ∩ (¬(c ⇒ (¬f ∪
d)) ∩ ¬e))) and by fact 3.8 our formula is proved to be L tautology.

3.6.2 M Consistency

One of the most important notions for any logic are notions of consistency and
inconsistency. We introduced and discussed them in case of classical semantics
in section 3.3. We formulate them now for any M extensional semantics and
examine them in cases of L and H semantics.

137
Consider LCON and let S =
6 ∅ be any non empty set of formulas of LCON . Let
M be an extensional semantics for LCON . We adopt the following definitions.

Definition 3.39 A truth truth assignment v : V AR −→ LV is a M model

for the set G of formulas if and only if v |=M A for all formulas A ∈ G.
We denote it by v |=M G.

Definition 3.40 A set G ⊆ F is called M consistent if and only if there

is v : V AR −→ LV , such that v |=M G. Otherwise the set G is called M
inconsistent.

Observe that the definition of inconsistency can be stated as follows.

Definition 3.41 A set G ⊆ F is called M inconsistent if and only if for

all v : V AR −→ LV there is a formula A ∈ G, such that v ∗ (A) 6= T .

Plainly speaking, a set G is consistent if it has a model, and is inconsistent if it

does not have a model under a semantic M.

Exercise 3.27
Prove that the set
G = {((a ∩ b) ⇒ b), (a ∪ b), ¬a}

is L, H, and K consistent.

Solution Consider a truth assignment v : V AR −→ {T, ⊥, F }. By the defi-

nition 3.40, v must be such that v ∗ (((a ∩ b) ⇒ b)) = T, v ∗ ((a ∪ b) = T ), and
v ∗ (¬a) = T . We want to prove that such v exists. Observe that ((a ∩ b) ⇒
b) is classical tautology, so let’s try to find v : V AR −→ {T, F } such that
v ∗ ((a ∪ b)) = T, v ∗ (¬a) = T . This holds when v(a) = F and hence
F ∪ v(b) = T . This gives us v such that v(a) = F and v(b) = T . We proved
that the connectives of L, H, and K semantics when restricted to the values
T and F become classical connectives. Hence any v such that v(a) = F and
v(b) = T is a L, H, and K model for G.

The same argument prove the following general fact.

Fact 3.9
For any non empty set G of formulas of a language L{¬,⇒,∪,∩} ,
if G is consistent under classical semantics, then it is L, H, and K consistent.

138
Exercise 3.28
Give an example of an infinite set G of formulas of a language LB = L{¬,S,⇒,∪,∩}
that is L, H, K and B consistent.

Solution
Observe that for the set G to be considered to be L, H, K consistent its formulas
must belong to the sub language L{¬,⇒,∪,∩} of the language LB . Let’s take, for
example a set
G = {(a ∪ ¬b) : a, b ∈ V AR}.
G is infinite since the set V AR is infinite. Consider any v : V AR −→ {F, m, T }
or v : V AR −→ {F, ⊥, T } such that v(a) = T, v(b) = F , we have v ∗ (a ∪ b) =
v(a) ∪ v(b) = T ∪ T = T in all semantics L, H, K and B. This proves that G is
L, H, K and B consistent.

Exercise 3.29
Prove that the set
G = {(a ∩ ¬a) : a ∈ V AR}
is L, H, K, and B inconsistent..

Solution
We know that the set G is classically inconsistent, i.e. v ∗ ((a ∩ ¬a)) 6= T for all
v : V AR −→ {F, T } under classical semantics. It also holds for We have to
show that it also holds for L, H, K and B semantics when we restrict the func-
tions defining their connectives to the values T and F only. In order to prove
inconsistency under L, H, K, semantics we have to show that v ∗ ((a ∩ ¬a)) 6= T
for all v : V AR −→ {F, ⊥, T } under the respective semantics, i.e. we have to
evaluate additional case v(a) =⊥ in all of them. Observe that negation ¬ is de-
fined in all of them as ¬ ⊥=⊥, and v ∗ ((a ∩ ¬a)) =⊥ ∩¬ ⊥=⊥ ∩ ⊥=⊥6= T . This
proves that G is L, H, and K inconsistent. The case of B semantics is similar,
except that now we consider all v : V AR −→ {F, m, T } and the additional
case is v(a) = m. By definition ¬m = m and v ∗ ((a ∩ ¬a)) = m ∩ m = m 6= T .

The examples of B consistent, or inconsistent sets G in exercise 3.28 and exercise

3.29 were restricted to formulas from LB = L{¬,S,⇒,∪,∩} that did not include
the connective S. In this sense they were not characteristic to the semantics B.
We pose hence a natural question whether such examples exist.

Exercise 3.30
Give an example of sets G1 , G2 containing some formulas that include the S
connective of the language LB = L{¬,S,⇒,∪,∩} such that G1 is B consistent and
G2 is B inconsistent

139
Solution
There are many such sets G, here are just two simple examples.

G1 = {(Sa ∪ S¬a), (a ⇒ ¬b), S¬(a ⇒ b), (b ⇒ Sa)}

G2 = {Sa, (a ⇒ b), (¬b∪, S¬a}.

Let v : V AR −→ {F, m, T }, be such that v(a) = T, v(b) = F . We evaluate
(ST ∪ S¬T ) = T ∪ T = T, (T ⇒ ¬F ) = T, S¬(T ⇒ F ) = S¬F = T, (F ⇒
ST ) = F ⇒ T = T . This proves that v is a B model for G1 , i.e. G1 is consistent.
Assume now that there is v : V AR −→ {F, m, T }, such that v |=B G2 .
In particular v ∗ (Sa) = T . This is possible if and only if v(a) = T , then
v ∗ (S¬a) = SF = F. This contradicts v |=B G2 . Hence G2 is B inconsistent.

We introduce, as we did in classical case a notion of a contradiction as follows.

Definition 3.42
Let M be an extensional semantics for LCON . We say that a formula A is a
M contradiction if it doesn’t have a M model.

Example 3.25
A formula (Sa ∩ S¬a) of LB = L{¬,S,⇒,∪,∩} is a B contradiction.

Proof
Assume that there is v, such that v |= (Sa ∩ S¬a), i.e. v ∗ ((Sa ∩ S¬a)) = T if
and only if (shorthand notation) Sa = T and S¬a = T . But Sa = T if and
only of a = T . In this case S¬T = SF = F 6= T. This contradiction proves that
such v does not exist, i.e. that for all v, v 6|= (Sa ∩ S¬a).

This also justifies the following.

Example 3.26 The set G = {(Sa ∩ S¬a) : a ∈ V AR} is an countably infinite

B inconsistent set.

Here is a simple problem asking to create your own, specific M semantics ful-
filling certain specifications. This semantics is different from all of previous
semantics defined and examined. We also ask to examine some of its properties,
including M consistency and M inconsistency. We provide an example two dif-
ferent semantics. We encourage the reader to come up with his/hers own and
to write down formally its full definition according to definition 3.7 as it was
done in the case of L semantics.

140
Review Problem

Part 1. Write the following natural language statement:

One likes to play bridge, or from the fact that the weather is good we conclude
the following: one does not like to play bridge or one likes not to play bridge
as a formula of 2 different languages
1. Formula A1 ∈ F1 of a language L{¬, L, ∪, ⇒} , where LA represents
statement ”one likes A”, ”A is liked”.
2. Formula A2 ∈ F2 of a language L{¬, ∪, ⇒} .

Part 2. Define formally, following all steps of the defnition 3.7, a 3 valued
extensional semantics LK for the language L{¬, L, ∪, ⇒} under the following
assumptions.
s1 We assume that the third value is denoted by ⊥ is intermediate between
designated value T and F, i.e. that F <⊥< T.
s2 We model a situation in which one ”likes” only truth, represented by T; i.e.
in which
LT = T, L ⊥= F, LF = F.

s3 The connectives ¬, ∪, ⇒ can be defined as one wishes, but they have to

be defined in such a way to make sure that always ”one likes A or does not like
A”, i.e. it must be assured that |=LK (LA ∪ ¬LA).

Part 3.
1. Verify whether the formulas A1 and A2 from the Part 1. have a model/
counter model under your semantics LK. You can use shorthand notation.
2. Verify whether the following set G is LK consistent. You can use shorthand
notation.
G = {La, (a ∪ ¬Lb), (a ⇒ b), b }.

3. Give an example on an infinite, LK consistent set of formulas of the language

L{¬, L, ∩, ∪, ⇒} . Some formulas must contain the connective L.

Review Problem Solutions

Part 1 Solution
1. We translate the statement into a formula A1 ∈ F1 of a language L{¬, L, ∩, ∪, ⇒}
as follows.

141
Propositional variables: a, b, where a denotes statement: play bridge, b denotes
a statement: the weather is good.

A1 = (La ∪ (b ⇒ (¬Ia ∪ L¬a))).

2. We translate our statement into a formula A2 ∈ F2 of a language L{¬, ∪, ⇒}

as follows.
Propositional Variables: a, b, c, where a denotes statement: One likes to play
bridge, b denotes a statement: the weather is good, and c denotes a statement:
one likes not to play bridge.

A2 = (a ∪ (b ⇒ (¬a ∪ c))).

Part 2 Solution 1
Here is a simple LK semantics. We define the logical connectives by writing
functions defining connectives in form of the truth tables and skipping other
points of the definition 3.7. We leave it to the reader as an exercise to write
down a full definition according to the definition 3.7.

LK Semantics 1

L F ⊥ T ¬ F ⊥ T
F F T T F F

∩ F ⊥ T ∪ F ⊥ T ⇒ F ⊥ T
F F F F F F ⊥ T F T T T
⊥ F ⊥ ⊥ ⊥ ⊥ T T ⊥ T ⊥ T
T F ⊥ T T T T T T F F T

We verify whether the condition s3 is satisfied, i.e. whether |=LK (LA ∪ ¬LA)
by simple evaluation. Let v : V AR −→ {F, ⊥, T } be any truth assignment. For
any formula A, v ∗ (A) ∈ {F, ⊥, T } and LF ∪ ¬LF = LF ∪ ¬LF = F ∪ ¬F ∪ T =
T, L ⊥ ∪¬L ⊥= F ∪ ¬F = F ∪ T = T, LT ∪ ¬LT = T ∪ ¬T = F ∪ T = T.

Part 2 Solution 2
Here is another simple LK semantics. Writing, yet again, a full definition is left
to the reader as an exercise.

142
LK Semantics 2

The logical connectives are the following funcions in the set {F, ⊥, T }, where
{F <⊥< T }. We define ¬F = T, ¬ ⊥= T, ¬T = F and, as by s2, LT =
T, L ⊥= F, LF = F. We define, for any x, y ∈ {F, ⊥, T }

x ∩ y = min{x, y}, x ∪ y = T, x ⇒ y = T if x ≤ y, x ⇒ y = F if x > y.

From the above definition we can see the LK satisfies the requirement s3 that
especially |=LK (LA ∪ ¬LA) since for any truth assignment v, no matter what
values v ∗ (LA) and v ∗ (¬LA) are, the combination of them by ∪ will always be T .

Part 3
1. Verify whether the formulas A1 and A2 from the Part 1. have a model/
counter model under your semantics LK. You can use shorthand notation.
Solution 1
A model for A1 = (La ∪ (b ⇒ (¬La ∪ L¬a))) under LK semantics 1 is any v,
such that v(a) = T . By easy evaluation, A1 does not have no counter model,
i.e. |=LK A1 . Also any v, such that v(a) = T is a model for A1 as we have
v ∗ (A2 ) = T ∪ v ∗ ((b ⇒ (¬a ∪ c))) = T by definition of ∪.

Solution 2 The main connective of A1 and A2 is ∪. By definition of ∪ in LK

semantics 2, x ∪ y = T for all x, y ∈ {F, ⊥, T }, and hence any v is a model for
both A1 and A2 , i.e. they are both tautologies under LK semantics 2.
Part 3
2. Verify whether the following set G is LK consistent. You can use shorthand
notation.
G = {La, (a ∪ ¬Lb), (a ⇒ b), b }.

Solution 1
G is LK consistent under semantics 1 because any v, such that v(a) = T, v(b) =
T is a LK model for G under semantics 1 by straightforward evaluation.
Solution 2
Consider any v, such that v(a) = v(b) = T . We evaluate: v ∗ (La) = LT =
T, v ∗ ((a ∪ ¬Lb)) = T ∪ F = T, v ∗ (a ⇒ b)) = T ⇒ T = T. This proves
v |=LK G, i.e. G is consistent.

Part 3
3. Give an example on an infinite, LK consistent set of formulas of the language
L{¬, L, ∩, ∪, ⇒} . Some formulas must contain the connective L.

143
Solution
The infinite set G = {La : a ∈ V AR} is consistent under both LK semantics,
as any v, such that v(a) = T we get v ∗ (La) = LT = T by s2.
The infinite set G = {(La ∪ (b ∩ L¬c)) : a, b, c ∈ V AR} is consistent under the
semantics 2 by its definition of ∪. Any v, such that v(a) = T is its model.

3.6.3 M Equivalence of Languages

Given an extensional semantics M defined for a propositional language LCON

with the set F of formulas and a set LV 6= ∅ of logical values. We extend now
the classical notions of logical equivalence and equivalence of languages
introduced in section 3.4 to the extensional semantics M.

Definition 3.43 (M Equivalence )

For any formulas A, B ∈ F, we say that A, B are M logically equivalent if
and only if they always have the same logical value assigned by the semantics
M, i.e. when v ∗ (A) = v ∗ (B) for all v : V AR → LV. We write

A ≡M B

to denote that A, B are M logically equivalent.

Remember that ≡M is not a logical connective. It is just a metalanguage

symbol for saying ”formulas A, B are logically equivalent under the semantics
M”. We use symbol ≡ for classical logical equivalence only.

Exercise 3.31
The classical logical equivalence (A ∪ B) ≡ (¬A ⇒ B) holds for all formulas A,
B and is defining ∪ in terms of negation and implication. Show that it does not
hold under L semantics, i.e. that there are formulas A, B, such that

(A ∪ B) 6≡L (¬A ⇒ B)

Solution
Consider a case when A = a and B = b. By definition 3.43 we have to show
v ∗ ((a ∪ b)) 6= v ∗ ((¬a ⇒ b)) for some v : V AR → {F, ⊥, T }. Observe that
v ∗ ((a ∪ b)) = v ∗ ((¬a ⇒ b)) for all v : V AR → {F, T }. So we have to check only
truth assignments that involve ⊥. Let v be any v such that v(a) = v(b) =⊥.
We evaluate v ∗ ((a ∪ b) =⊥ ∪ ⊥=⊥ and v ∗ ((¬a ⇒ b)) = ¬ ⊥⇒⊥= F ⇒⊥= T .
This proves that (a ∪ b) 6≡L (¬a ⇒ b).. and hence we have proved (A ∪ B) 6≡L
(¬A ⇒ B).

144
We proved that the classical equivalence defining disjunction in terms of nega-
tion and implication can’t be used for the same goal in L semantics. It does not
mean that we can’t define L disjunction in terms of L implication. In fact, we
prove by simple evaluation that the following holds.

Fact 3.10
The L disjunction is definable in terms of L implication only, i.e. for any
formulas A, B ∈ F
(A ∪ B) ≡L ((A ⇒ B) ⇒ B).

The classical equivalence substitution theorem 3.12 extends to any semantics

M as follows.

Theorem 3.28 ( M Equivalence)

Let a formula B1 be obtained from a formula A1 by a substitution of a formula
B for one or more occurrences of a sub-formula A of A1 , what we denote as

B1 = A1 (A/B).

Then the following holds for any formulas A, A1 , B, B1 ∈ F.

If A ≡M B, then A1 ≡M B1 .

We leave the proof to the reader as an exercise.

Example 3.27
Let A1 = (a ⇒ (¬a ∪ b)) and consider a sub formula A = (¬a ∪ b) of A1 . By
Fact 3.10, (¬a ∪ b) ≡L ((¬a ⇒ b) ⇒ b). Take B = ((¬a ⇒ b) ⇒ b) and let
B1 = A1 (A/B) = A1 ((¬a ∪ b)/((¬ ⇒ b) ⇒ b)) = (a ⇒ ((¬a ⇒ b) ⇒ b)). By the
M Equivalence Theorem 3.28

(a ⇒ (¬a ∪ b)) ≡L (a ⇒ ((¬ ⇒ b) ⇒ b)).

M Equivalence of Languages

We extend now, in a natural way, the classical notion equivalence of languages

introduced and examined in section 3.4.

Definition 3.44
Given two languages: L1 = LCON1 and L2 = LCON2 , for CON1 6= CON2 . We
say that L1 and L2 are M logically equivalent and denote it by

L 1 ≡M L 2

145
if and only if the following conditions C1, C2 hold.
C1 For any formula A of L1 , there is a formula B of L2 , such that A ≡M B,
C2 For any formula C of L2 , there is a formula D of L1 , such that C ≡M D.

Exercise 3.32 Prove that

L{¬,⇒} ≡L L{¬,⇒,∪}

Solution
Condition C1 holds because any formula of L{¬,⇒} is a formula of L{¬,⇒,∪} .
Condition C2 holds because the Fact 3.10 equivalence (A ∪ B)≡L ((A ⇒ B) ⇒
B) and the Theorem 3.28.

3.7 Homework Problems

Formal Propositional Languages

For the problems below do the following.

(i) Determine which of the formulas is, and which is not a well formed formula.
Determine a formal language of L to which the formula or set of formulas
belong.
(ii) If a formula is correct, write what its main connective is. If it is not correct,
write the corrected formula and then write its main connective. If there
is more then one way to correct the formula, write all possible corrected
formulas.
(iii) If a formula is correct, write what it says. If it is not correct, write the
corrected formula and then write what it says.
(iv) For each of correct formula determine its degree and write down its all
sub-formulas of the degree 0 and 1.
Problems

1. ((a ↑ b) ↑ (a ↑ b) ↑ a)
2. (a ⇒ ¬b) ⇒ ¬a
3. ♦(a ⇒ ¬b) ∪ a, ♦(a ⇒ (¬b ∪ a), ♦a ⇒ ¬b ∪ a
4. (¬♦a ⇒ ¬a), (¬♦a ⇒ ¬a), ¬♦(a ⇒ ¬a)
5. ((a ∪ ¬K¬a)), KK(b ⇒ ¬a), ¬K(a ∪ ¬a)

146
 6. (B(a ∩ b) ⇒ Ka), B((a ∩ b) ⇒ Ka)

7. G(a ⇒ b) ⇒ Ga ⇒ Gb), a ⇒ HF a, F F a ⇒ F a

8. (a ⇒ ((¬b ⇒ (¬a ∪ c)) ⇒ ¬a))

9. ♦((a ∩ ¬a) ⇒ (a ∩ b))

10. ¬♦(a ⇒ ¬a)

11. ♦(♦a ⇒ (¬b ∪ ♦a))

12. (¬(a ∩ b) ∪ a)

13. Write the natural language statement:

From the fact that it is not necessary that an elephant is not a bird we
deduce that:
it is not possible that an elephant is a bird or, if it is possible that an
elephant is a bird, then it is not necessary that a bird flies.
in the following two ways.
1. As a formula A1 ∈ F1 of a language L{¬,C,I,∩,∪,⇒} .
2. As a formula A2 ∈ F2 of a language L{¬,∩,∪,⇒} .

14. Write the natural language statement

If it is not believed that quiz is easy or quiz is not easy, then from the fact
that 2 + 2 = 5 we deduce that it is believed that quiz is easy.
in the following two ways.
1. As a formula A1 of a language L1 = L{¬,B,∩,∪,⇒} , where B is a believe
connective. Statement BA says: It is believed that A.
2. As a formula A2 of a language L2 = L{¬,∩,∪,⇒} .

Formal Classical Semantics

1. Find and prove definability formula defining implication in terms of con-

junction and negation.

2. Find and prove definability formula defining conjunction in terms of dis-

junction and negation.

3. Find and prove definability formula defining conjunction in terms of im-

plication and negation.

4. Prove that ∪ can be defined in terms of ⇒ alone.

5. Find and prove definability formula defining ⇒ in terms of ↑.

147
 6. Find definability formula defining ⇒ in terms of ↓.
7. Define ∩ in terms of ⇒ and ¬.
8. Find definability formula defining ∩ in terms of ↓ alone.
9. Given a formula A: (((a ∩ b) ∪ ¬c) ⇒ b). Evaluate (do not use shorthand
notation) v ∗ (A) for truth assignments v : V AR −→ {T, F } such that
(i)v(a) = T, v(b) = F, v(c) = F, v(x) = T for all x ∈ V AR − {a, b, c},
(ii)v(a) = F, v(b) = T, v(c) = T, v(x) = F for all x ∈ V AR − {a, b, c}.
10. Given a formula A: (((a ⇒ ¬b) ∪ b) ⇒ a). Evaluate (use shorthand
notation) v ∗ (A) for all truth assignments restricted to A.
11. Given a formula A: (((a ↓ ¬b) ∪ b) ↑ a). Evaluate (do not use shorthand
notation) v ∗ (A) for truth assignments v : V AR −→ {T, F } such that
(i) v(a)=T, v(b)=F, v(c) =F for all c ∈ V AR − {a, b},
(ii) v(a)=F, v(b)=T, v(c) =T for all c ∈ V AR − {a, b}.
(iii) List all restricted models and counter-models for A.
Write the following natural language statement From the fact that it is
possible that 2 + 2 6= 4 we deduce that it is not possible that 2 + 2 6= 4 or, if
it is possible that 2 + 2 6= 4, then it is not necessary that you go to school.
as a formula . A ∈ F of a language L = L{¬,∩,∪,⇒} .
(i) Find a restricted model v for the formula A.
(ii) Find 3 models w of A such that v ∗ (A) = w∗ (A) the for v from (i).
How many of such models exist?
(iii) Find all models, counter-models (restricted) for A. Use shorthand
notation.
(iv) Is A ∈C?, is A2 ∈T? Justify your answers.
12. Given v : V AR −→ {T, F } such that v ∗ ((¬a ∪ b) ⇒ (a ⇒ ¬c)) = F .
Evaluate: v ∗ (((b ⇒ a) ⇒ (a ⇒ ¬c)) ∪ (a ⇒ b)).
13. Show that all of the truth assignments v1 , v2 , v3 defined below are models
for the formula A : ((a ∩ ¬b) ∪ ¬c).
v1 : V AR −→ {T, F }, is such that v1 (a) = T, v1 (b) = F, v1 (c) = T,
and v1 (x) = F , for all x ∈ V AR − {a, b, c};
v2 : V AR −→ {T, F } is such that v2 (a) = T, v2 (b) = F, v2 (c) = T,
v2 (d) = T, and v2 (x) = F for all x ∈ V AR − {a, b, c, d};
v3 : V AR −→ {T, F } is such that v3 (a) = T, v3 (b) = F, v3 (c) = T,
v3 (d) = T, v3 (e) = T , and v3 (x) = F , for all x ∈ V AR − {a, b, c, d, e}.
14. Prove that for any formula A ∈ F, if A has a model (counter- model),
then it has uncountably many models (counter-models). More precisely,
as many as there are real numbers. Hint Use the Counting Functions
Theorem 3.4.

148
15. Use Generalization Method to determine whether
|= (¬((a ∪ b) ⇒ ((c ⇒ d) ∪ e)) ⇒ ((a ∪ b) ∩ (¬(c ⇒ d) ∩ ¬e))).

16. Prove |= (¬((a ∪ b) ⇒ (c ⇒ d)) ⇒ (¬((a ∪ b) ⇒ (c ⇒ d)) ⇒ (¬e ∩ a))).

17. Use Proof by Contradiction Method to determine whether

|= (((A ⇒ (B ⇒ C)) ∩ (A ⇒ B)) ⇒ (A ⇒ C)).

18. Use Truth Table and Substitution Methods to prove |= (¬¬A ⇔ A).

19. Use Truth Table and Substitution Methods to prove to prove the Reductio
ad Absurdum tautology ((¬A ⇒ (B ∩ ¬B)) ⇒ A).

20. Use Proof by Contradiction Method to prove the Exportation and Impor-
tation tautology (((A ∩ B) ⇒ C) ⇔ (A ⇒ (B ⇒ C))).

21. For the formulas listed below determine whether they are tautologies or
not. If a formula is not a tautology list its counter-model (restricted). Use
shorthand notation.
(i) A1 = (¬(a ⇒ (b ∩ ¬c)) ⇒ (a ∩ ¬(b ∩ ¬c)))
(ii) A2 = ((a ∩ ¬b) ⇒ ((c ∩ ¬d) ⇒ (a ∩ ¬b)))
(iii) A3 = (¬(A ∩ ¬B) ∪ (A ∩ ¬B))

22. Find all models and a counter-model restricted to G (if exist) for the
following sets G of formulas. Use shorthand notation.
(i) S1 = {a, (a ∩ ¬b), (¬a ⇒ (a ∪ b))}
(ii) S2 = {(a ⇒ b), (c ∩ ¬a), b}
(iii) S3 = {a, (a ∩ ¬b), ¬a, c}

23. Give an example of an infinite set G ⊆ F, such that G =

6 T and G has a
model, i.e. is consistent.

24. Give an example of an infinite consistent set G ⊆ F, such that G ∩ T = ∅.

25. Give an example of an infinite set G ⊆ F, such that G=

6 C and G does
not have a model, i.e.is inconsistent.

26. Give an example of an infinite set G ⊆ F, such that G ∩ C = ∅.

27. Find an infinite number of formulas that are independent from a set
G = {(a ⇒ (a ∪ b)), (a ∪ b), ¬b, (c ⇒ b)}. Use shorthand notation.

28. Given an infinite set G = {(a ∪ ¬a) : a ∈ V AR}. Find 3 formulas A ∈ F

that are independent from G.

29. Give an example of an infinite set G and an infinite set of formulas inde-
pendent from it.

149
Equivalence of Languages

1. Prove that L{∩,¬} ≡ L{∪,¬} .

2. Transform a formula A = ¬(¬(¬a ∩ ¬b) ∩ a) of L{∩,¬} into a logically

equivalent formula B of L{∪,¬} .

3. Transform a formula A = (((¬a ∪ ¬b) ∪ a) ∪ (a ∪ ¬c)) of L{∪,¬} into a

formula B of L{∩,¬} , such that A ≡ B.

4. Prove, using proper logical equivalences (list them at each step) that
(i) ¬(A ⇔ B) ≡ ((A ∩ ¬B) ∪ (¬A ∩ B)).
(ii) ((B ∩ ¬C) ⇒ (¬A ∪ B)) ≡ ((B ⇒ C) ∪ (A ⇒ B)).

5. Prove that L{¬,∩} ≡ L{¬,⇒} .

6. Prove by using proper logical equivalences that

(i) ¬(¬A ∪ ¬(B ⇒ ¬C)) ≡ (A ∩ ¬(B ∩ C)),
(ii) (¬A ∩ (¬A ∪ B)) ≡ (¬A ∪ (¬A ∩ B)).

7. Prove that L{∩,∪.¬} ≡ L{⇒,¬} .

8. Prove that L{∩,∪,⇒,¬} ≡ L{∪,¬} .

9. (i) Transform a formula A = (((a ∪ ¬b) ⇒ a) ∩ (¬a ⇒ ¬b)) of L{∩,∪,⇒,¬}

into a logically equivalent formula B of L{∪,¬} .
(ii) Find all B of L{∪,¬} , such that B ≡ A, for A from (i).

10. (i) Transform a formula A = (((¬a ∪ ¬b) ∪ a) ∪ (a ∪ ¬c)) of L{∪,¬} into

a formula B of L{∩,∪,⇒,¬} , such that A ≡ B.
(ii) Find all B of L{∩,∪,⇒,¬} , such that B ≡ A, for A from (i)

11. Prove that L{∩,∪,⇒,¬} ≡ L{↑} .

12. Prove that L{∩,∪,⇒,¬} ≡ L{ ↓} .

13. Prove that L{↑} = L{ ↓} .

Many Valued Semantics

1. In all 3-valued semantics presented here we chose the language without

the equivalence connective ”⇔”. Extend t L, L4 semantics to a language
containing the equivalence connective. Prove that your semantics is well
defined as by definition 3.14.

2. Extend H, K, semantics to a language containing the equivalence connec-

tive. Are your semantics well defined as by definition 3.14?

150
 3. Extend B, semantics to a language containing the equivalence connective.
Are your semantics well defined as by definition 3.14?
4. Let v : V AR −→ {F, ⊥, T } be any v, such that v ∗ ((a ∪ b) ⇒ (a ⇒ c)) =⊥
under H semantics. Evaluate v ∗ (((b ⇒ a) ⇒ (a ⇒ ¬c)) ∪ (a ⇒ b)).

5. Verify which of the classical tautologies (3.12) are, and which are not L
tautologies.
6. Verify which of the classical tautologies (3.13) are, and which are not L
tautologies.
7. Give an example of 3 formulas

8. For each of 3-valued logic semantics presented in this chapter, find 5 clas-
sical tautologies that are tautologies of that logic.
9. Examine the notion of definability of connectives as defined in section 3.3,
definition 3.16 for L semantics. semantics.

10. Examine the notion of definability of connectives as defined in section 3.3,

definition 3.16 for H semantics. semantics.
11. Given a set G = {((a ∩ b) ⇒ b), (a ∪ b), a}. Verify whether G is consistent
under H semantics.

12. Given a set G = {((a ∩ b) ⇒ b), (a ∪ b), a}. Verify whether G is consistent
under L semantics.
13. Given a language L{¬,⇒,∪,∩} . We define: A formula A ∈ F is called M
independent from a set G ⊆ F if and only if the sets G ∪ {A} and
G ∪ {¬A} are both M consistent. I.e. when there are truth assignments
v1 , v2 such that v1 |=M G ∪ {A} and v2 |=M G ∪ {¬A}.
Given a set G = {((a ∩ b) ⇒ b), (a ∪ b), a}.
(i) Find a formula A that is L independent from a set G.
(ii) Find a formula A that is H independent from a set G.
(iii) Find an infinite number of that are L independent from a set G.
(iv) Find an infinite number of that are H independent from a set G.
14. By exercise 3.31 the classical logical equivalence (A ∪ B) ≡ (¬A ⇒ B)
does not hold under L semantics, i.e. that there are formulas A, B, such
that (A ∪ B) 6≡L (¬A ⇒ B). Show 3 formulas A,B such that it does hold
for L semantics, i.e. such that are formulas A, B, such that (A ∪ B) ≡L
(¬A ⇒ B).

151
152
Chapter 4

General Proof Systems:

Syntax and Semantics

Proof systems are built to prove, construct formal proofs of statements formu-
lated in a given language formulated in a given language. First component of
any proof system is hence its formal language L. Proof systems can be thought
as an inference machine with special statements, called provable statements, or
theorems being its final products. The starting points are called axioms of the
proof system. We distinguish two kinds of axioms: logic LA and specific SA.
When building a proof system for a given language and its semantics i.e. for a
logic defined semantically we choose as a set of logical axioms LA some subset of
tautologies, i.e. statements always true. This is why we call them logical axioms.
A proof system with only logic axioms LA is also called logic proof systems,
or just proof systems for short. If we build a proof system for which there is
no known semantics, as it has happened in the case of classical, intuitionistic,
and modal logics, we think about the logical axioms as statements universally
true. We choose as axioms (finite set) the statements we for sure want to be
universally true, and whatever semantics follows they must be tautologies with
respect to it. Logical axioms are hence not only tautologies under an established
semantics, but they also guide us how to establish a semantics, when it is yet
unknown.
For the set of specific axioms SA we choose these formulas of the language that
describe our knowledge of a universe we want to prove facts about. They are not
universally true, they are true only in the universe we are interested to describe
and investigate. This is why we call them specific axioms. A proof system with
logical axioms LA and specific axioms SA is called a formal theory based on a
proof system with logic axioms LA.
The inference machine is defined by a finite set of rules, called inference rules.

153
The inference rules describe the way we are allowed to transform the information
within the system with axioms as a staring point. The process of this transfor-
mation is called a formal proof. The provable formulas for which we have a
formal proof are called consequences of the axioms, or theorem, or just simple
provable formulas. We use proof systems not only to be able to build formal
proofs in them, but also to search for proofs of given statements of their the
language. We distinguish special proof systems for which it is possible to define
a mechanical method for determining, given any statement of A, but which also
generates a proof, is called syntactically decidable or automatically decidable, or
an automated system
When building a proof system we choose not only axioms of the system, but also
specific rules of inference. The choice of rules is often linked, as was the choice of
axioms, with a given semantics. We want the rules to preserve the truthfulness
of what we are proving from axioms via the rules. Rules with this property
are called sound rules and the system a sound proof system. The notion of
truthfulness is always defined by a given propositional, or predicate language
L semantics M. Rules of inference can be sound under one semantics and not
sound under another. When developing a proof system S the first goal is prove
a theorem, called Soundness Theorem about its relationship with its semantics
M. It states that the following holds for any formula A of the language L of the
system S. If a formula A is provable from logical axioms LA of S only, then A
is a tautology under the semantics M.
A proof system can be sound under one semantics, and not sound under the
other. For example a set of axioms and rules sound under classical logic seman-
tics might not be sound under intuitionistic semantics, H, L, K semantics, or
others. This is why we talk about proof systems for classical logic, intuitionistic
logic, for modal logics etc. In general there are many proof systems that are
sound under a given semantics, i.e. there are many sound proof systems for a
given logic semantically defined. We present some examples at the end of the
chapter. Given a proof system S with logical axioms LA that is sound under a
given semantics M . Let TM be a set of all tautologies defined by the semantics
M , i.e. TM = {A : |=M A}. A natural questions arises: are all tautologies
defined by the semantics M , provable in the system S that is sound under the
semantics M . The positive answer to this question is called a completeness
property of the system S. Because we ask the completeness property question
for sound systems only we put it in a form of a theorem called a Completeness
Theorem for a proof system S, under a semantics M . It states that the following
holds for any formula A of the language L of the system S. A formula A is
provable in S if and only if A is a tautology under the semantics M . We write
it symbolically as: `S A if and only if |=M A. The Completeness Theorem is
composed from two parts: the Soundness Theorem and the completeness part
that proves the completeness property of a sound system.
Proving the Soundness Theorem for S under a semantics M is usually a straight-
forward and not a very difficult task. We first prove that all logical axioms are

154
M tautologies, and then that all inference rules of the system preserve the notion
of the M truth ( M model). Proving the completeness part of the Completeness
Theorem is always a crucial and very difficult task.
We will study two proofs of the Completeness Theorem for classical propositional
Hilbert style proof system in chapter 5, and a constructive proofs for automated
theorem proving systems for classical logic the chapter 6.
Observe that we formulated all these basic theorems linking semantics and syn-
tax (provability) in a general manner. As we first consider propositional lan-
guages (chapters 5, 6, 7) and hence we use proof systems for propositional logics
as examples. The case of predicate logics will be discussed in chapters 8, 9, 10,
11.

4.1 Syntax
In this section we formulate a definition of a proof system S by specifying and
defining and all its components. We define a notion of a formal proof in a
given proof system, and give simple examples of different proof systems. When
defining a proof system S we specify, as the first step, its formal language L.
When It can be a propositional, or a predicate language. It is a first component
of the proof system S. Given a set F of well formed formulas, of the language
L, we often extend this set, and hence the language L to a set E of expressions
build out of the language L, and some additional symbols, if needed. It is a
second component of the proof system S. Proof systems act as an inference
machine, with provable expressions being its final products. This inference
machine is defined by setting, as a starting point a certain non-empty, proper
subset LA of E, called a set of logical axioms of the system S. The production
of provable formulas is to be done by the means of inference rules. The inference
rules transform an expression, or finite string of expressions, called premisses,
into another expression, called conclusion. At this stage the rules don’t carry
any meaning - they define only how to transform strings of symbols of our
language into another string of symbols. This is a reason why investigation of
proof systems is called syntax or syntactic investigation as opposed to semantcal
methods, which deal with semantics of the language and hence of the proof
system. The syntax- semantics connection within proof systems is established
by Soundness and Completeness theorems and will be discussed in detail in the
section 4.2.

Definition 4.1 ( Proof System)

By a proof system we understand a triple

S = (L, E, LA, R),

where L = (A, F) is a formal language, called the language of S with a set F of

155
formulas; E is a set of expressions of S; LA ⊆ E is a non empty set of logical
axioms of the system; R is a finite set of rules of inference.

The components of the proof systems S are defined as follows.

1. The language L of S
In the propositional case, the formal language L consists of two components: an
alphabet A and a set F of formulas. In predicate case the language L consists
of thee components: an alphabet A, a set T of terms and a set F of formulas.
The set T of terms is needed to define properly the set of F of formulas and we
list it as to distinguish it the propositional case. We will denote the language
F of S uniformly as L = (A, F) and specify if it is propositional or a predicate
language accordingly. We assume that the both sets A and F are enumerable,
i.e. we will deal here with enumerable languages only.
Semantical Link. Any semantics M for the language L is called the semantics
for the proof system S.

2. The set E of expressions of S

Given a set F of well formed formulas, of the language L, we often extend this
set (and hence the language L to some set E of expressions build out of the
language L, and some additional symbols, if needed.
Automated theorem proving systems use as their basic components expressions
build out of formulas of the language L. They are for example sequences of
formulas in the proof systems RS and RQ presented in chapter 5.72 and —in
chapter ??, respectively. The first of such systems Gentzen’s systems LK for
classical logic and LK for intuitionistic logic and their followers use expressions
called Gentzen sequents. They are presented and discussed in chapter ??. There
also are resolution based proof systems that use different form of expressions
to represent for clauses and sets of clauses to mention the few. In many proof
system we choose the set of formulas F as expressions, i.e. we put E = F.
Semantical Link. We always have to extend a given semantics M of the lan-
guage L of the system S to the set E of expression.

3. The set LA of logical axioms of S

The logical axioms LA of S form a non-empty subset of the set E of expressions.
In particular, LA is a non-empty subset of formulas, i.e. LA ⊆ F. We assume
here that the set LA of logical axioms is finite, i.e. we consider here only finitely
axiomatizable proof systems.

Semantical Link. Set LA of logical axioms is always a subset of expressions that

are tautologies under the semantics M of the language L of S.

156
4. The set R of rules of inference of S
We assume that the proof system S contains a finite number of inference rules.
We assume that each rule has a finite number of premisses and one conclusion.
We also assume that one can effectively decide, for any inference rule, whether
a given string of expressions form its premisses and conclusion or do not, i.e.
that all rules r ∈ R are primitivvely recursive.
We put it in a formal definition as follows.

Definition 4.2 (Rule of Inference)

Given a non- empty set E of expressions of a proof system S. Each rule of
inference r ∈ R is a relation defined in the set E m , where m ≥ 1 with values in
E, i.e. r ⊆ E m × E.
Elements P1 , P2 , . . . Pm of a tuple (P1 , P2 , . . . Pm , C) ∈ r are called premisses
of the rule r, and C is called its conclusion.

We usually write the inference rules in a following convenient way.

If r is a one premiss rule and (P1 , C) ∈ r, then we write it as
P1
(r) .
C

If r is a two premisses rule and (P1 , P2 , C) ∈ r, then we write it as

P1 ; P2
(r) ,
C
P1 is called a left premiss of r and P2 is called a right premiss.
In general, if r is an m- premisses rule and (P1 , P2 , ...Pm , C) ∈ r, then we will
write it as

P1 ; P2 ; .... ; Pm
(r) .
A
Semantical Link. We want the rules of inference to preserve truthfulness i.e. to
be sound under the semantics M.

Formal Proofs in S

Given a proof system S = (L, E, LA, R). Final products of a single or multiple
use of the inference rules of S, with logical axioms LA taken as a starting point
are called provable expressions of the system S. A single use of an inference rule
is called a direct consequence. A multiple application of rules of inference with
axioms taken as a starting point is called a formal proof. Formal definitions are
as follows.

157
Definition 4.3 (DirectConsequence)
A conclusion of a rule of inference is called a direct consequence of its premisses.
I.e. for any rule of inference r ∈ R, if (P1 , ...Pn , C) ∈ r, then C is called a direct
consequence of P1 , ...Pn by virtue of r.

Definition 4.4 (Formal Proof )

Given a proof system S = (L, E, LA, R). Any sequence E1 , E2 , , En of expres-
sions from E, such that n ≥ 1,

E1 ∈ LA, En = E,

and for each 1 < i ≤ n, either Ei ∈ LA or Ei is a direct consequence of

some of the preceding expressions in E1 , E2 , , En by virtue of one of the rules
of inference r ∈ R is called a formal proof of E in S.
The number n ≥ 1 is the length of the proof E1 , E2 , , En . We write

`S E

to denote that E ∈ E has a formal proof in S. When the proof system S is

fixed we write ` E.

Any expression E such that E has a proof in S, is called a provable expression

of S. The set of all provable expressions of S is denoted by PS and is defined
as follows.
PS = {E ∈ E : `S E}. (4.1)

Consider a simple proof system system S1 with a language L = L{P, ⇒} , where

P is one argument connective. We take E = F, LA = {(A ⇒ A)}, and the set
of rules o inference R = {(r) PBB }. We write our proof system as

B
S1 = (L{P, ⇒} , F, {(A ⇒ A)}, (r)
) (4.2)
PB
where A, B are any formulas. Observe that even the system S1 has only one
axiom, it represents an infinite number of formulas. We call such axiom an
axiom schema.
Consider now a system S2
B
S2 = (L{P,⇒} , F, {(a ⇒ a)}, (r) ), (4.3)
PB
where a ∈ V AR is any variable (atomic formula) and B ∈ F is any formula.
Observe that the system S2 also has only one axiom similar to the axiom of S1 ,
both systems have the same rule of inference but they are very different proof
systems. For example a formula ((P a ⇒ (b ⇒ c)) ⇒ (P a ⇒ (b ⇒ c))) is an

158
axiom of the system S1 for A = (P a ⇒ (b ⇒ c) but is not an axiom of the
system S2 , as this systems permits axioms of the form: (a ⇒ a) for a being a
propositional variable.

A formal proof in a system S carries, as the proof system S does, a semantical

meaning but it is nevertheless purely syntactical in its nature. The rules of
inference of a proof system define only how to transform strings of symbols of
our language into another string of symbols. The definition of a formal proof
says that in order to prove an expression E of a system one has to construct of
s sequence of proper transformations as defined by the rules of inference. Here
some examples of provable formulas and their formal proofs in both S1 and S2
systems. Observe that we do not know the semantics for these systems.

Exercise 4.1 Let S1 , S2 be proof systems (7.2), (7.3), respectively. Show that

`S1 ((P a ⇒ (b ⇒ c)) ⇒ (P a ⇒ (b ⇒ c)))

`S1 P (a ⇒ a), `S2 P (a ⇒ a), `S1 P P (a ⇒ a), `S2 P P (a ⇒ a)

`S1 P P P ((P a ⇒ (b ⇒ c)) ⇒ (P a ⇒ (b ⇒ c)).

Solution Formal proof of ((P a ⇒ (b ⇒ c)) ⇒ (P a ⇒ (b ⇒ c))) in S1 is one

element sequence A1 = ((P a ⇒ (b ⇒ c)) ⇒ (P a ⇒ (b ⇒ c)). It is a proof
because the formula A1 an axiom of S1 . It is not a proof in S2 .
The formulas P (a ⇒ a), and P P (a ⇒ a) are provable formulas of both proof
systems. The formal proofs in both systems of above formulas are identical and
are as follows.
Formal proof of P (a ⇒ a) in S1 and S2 is a sequence A1 , A2 for

A1 = (a ⇒ a), A2 = P (a ⇒ a).
axiom rule (r) application
for B = (a ⇒ a)

Formal proof of P P (a ⇒ a) in S1 and S2 is a sequence A1 , A2 , A3 for

A1 = (a ⇒ a), A2 = P (a ⇒ a), A3 = P P (a ⇒ a).

axiom rule (r) application rule (r) application
for B = (a ⇒ a) for B = P (a ⇒ a)

Formal proof of P P P ((P a ⇒ (b ⇒ c)) ⇒ (P a ⇒ (b ⇒ c))) in S1 is a sequence

A1 , A2 , A3 , A4 for
A1 = ((P a ⇒ (b ⇒ c)) ⇒ (P a ⇒ (b ⇒ c))),
axiom

159
 A2 = P ((P a ⇒ (b ⇒ c)) ⇒ (P a ⇒ (b ⇒ c))),
rule (r) application
A3 = P P ((P a ⇒ (b ⇒ c)) ⇒ (P a ⇒ (b ⇒ c))),
rule (r) application
A4 = P P P ((P a ⇒ (b ⇒ c)) ⇒ (P a ⇒ (b ⇒ c))).
rule (r) application
It is not a proof in S2 . Moreover

6 `S2 P P P ((P a ⇒ (b ⇒ c)) ⇒ (P a ⇒ (b ⇒ c)).

Observe that even if the set of axioms and the inference rules of the proof system
are primitively recursive it doesn’t mean that the notion of ”provable expres-
sion” is also primitively recursive, i.e. that there always will be an effective,
mechanical method (effective procedure) for determining, given any expression
A of the system, whether there is a proof of A. We define the following notions

Definition 4.5 (Decidable system)

A proof system S = (L, E, LA, R) for which there is an effective decision
procedure for determining, for any expression E of the system, whether there is,
or there is no proof of E in S is called a decidable proof system, otherwise
it is called undecidable.

Observe that the above notion of decidability of the system S does not require
us to find a proof, it requires only a mechanical procedure of deciding whether
there is, or there is no such a proof. We hence introduce a following notion.

Definition 4.6 ( Syntactic Decidability)

A proof system S = (L, E, LA, R) for which there is an effective mechanical,
procedure that finds (generates) a formal proof of any E in S, if it exists, is called
syntactically semi- decidable. If additionally there is an effective method of
deciding that if a proof of E in S not found, it does not exist, the system S is
called syntactically decidable. Otherwise S is syntactically undecidable.

The existence of prove systems for classical logic and mathematics that are
syntactically decidable or syntactically semi-decidable was stated (in a differ-
ent form) by German mathematician David Hilbert in early 1900 as a part of
what is called Hilbert’s program. The main goal of Hilbert’s program was to
provide secure foundations for all mathematics. In particular it addressed the
problem of decidability; it said that here should be an algorithm for deciding
the truth or falsity of any mathematical statement. Moreover, it should use
only ”finitistic” reasoning methods. Kurt Gdel showed in 1931 that most of the
goals of Hilbert’s program were impossible to achieve, at least if interpreted in

160
the most obvious way. Nevertheless, Gerhard Gentzen in his work published in
1934/1935 gave a positive answer to existence of syntactical decidability. He
invented proof systems for classical and intiutionistic logics, now called Gentzen
style formalizations. They formed a basis for development of Automated The-
orem Proving area of mathematics and computer science. We will study the
Gentzen style formalizations in chapter ??.

Definition 4.7 (Automated Systems)

A proof system S = (L, E, LA, R) that is proven to be syntactically decidable
or semi-decidable is called an automated proof systems.

Automated proof systems are also called automated theorem proving systems,
Gentzen style formalizations, syntactically decidable systems and and we use all
of these terms interchangeably.

Example 4.1 Any complete Hilbert style proof system for classical proposi-
tional logic is an example of a decidable, but not syntactically decidable proof
system. We conclude its decidability from the Completeness Theorem (to be
proved in next chapter) and the decidability of the notion of classical tautology
(proved in chapter 3).

Example 4.2 The Gentzen style proof systems for classical and intuiionistic
propositional logics presented in chapter ??, are examples of proof systems that
are of both decidable and syntactically decidable.

W are going to prove now, as a simple example t the following

Fact 4.1
The systems proof systems S1 and S2 defined by (7.2) and (7.3), respectively
are syntactically decidable.

Proof Let’s now to think how we can search for a proof in S2 of a formula

P P ((P a ⇒ (b ⇒ c)).

If P P ((P a ⇒ (b ⇒ c)) ⇒ (P a ⇒ (b ⇒ c))) had the proof, the only last step
in this proof would have been the application of the rule (r) to the formula
P P ((P a ⇒ (b ⇒ c)) ⇒ (P a ⇒ (b ⇒ c))). This formula, in turn, if it had
the proof, the only last step in its proof would have been the application of
the rule r to the formula P ((P a ⇒ (b ⇒ c)) ⇒ (P a ⇒ (b ⇒ c))). And again,
this one could be obtained only from the formula ((P a ⇒ (b ⇒ c)) ⇒ (P a ⇒
(b ⇒ c))) by the virtue of the rule r. Here the search process stops; the rule r
puts P in front of the formulas, hence couldn’t be applied here. The formula

161
((P a ⇒ (b ⇒ c)) ⇒ (P a ⇒ (b ⇒ c))) isn’t an axiom of S2 , what means that
the only possible way of finding the proof has failed, i.e. we have proved that
6`S1 P P P ((P a ⇒ (b ⇒ c)) ⇒ (P a ⇒ (b ⇒ c))).
The above example of proof search in S2 defines the following an effective,
automatic Procedure S1 , S2 of searching for a proof of our formula in both
our proof systems. If the search ends with an axiom, we have a proof, if it
doesn’t end with an axiom it means that the proof does not exists. We have
described it, as an example, for one particular formula. It can be easily extended
to any formula A of L{P,⇒} as follows.

Procedure S1 , S2
Step : Check the main connective of A.
If main connective is P , it means that A was obtained by the rule r.
Erase the main connective P .
Repeat until no P left.
If the main connective is ⇒,check if a formula A is an axiom.
If it is an axiom , STOP and YES, we have a proof.
If it is not an axiom , STOP and NO, proof does not exist.

It is an effective, automatic procedure of searching for a proof of our formula in

both our proof systems. This ends the proof .

Observe also, that the systems S1 and S2 are such that we can easily describe
a general form of their provable formulas defined by (4.1) as PS = {E ∈ E :
`S E}. Namely we have the following.

PS1 = {P n (A ⇒ A) : n ∈ N, A ∈ F},

PS2 = {P n (a ⇒ a) : n ∈ N, a ∈ V AR},

where P n denotes n-iteration of P for n ≥ 1 and P 0 denotes absence of P.

Obviously we have that PS1 6= PS2 , and PS2 ⊆ PS1 .

The proof systems S1 and S2 are very simple, indeed. Here is an example of
another two, similar but slightly more complex proof systems.
Consider two proof systems S3 and S4 of the language L{∪,¬} with the set of
expressions E = F and is defined as follows.
(A ∪ ¬A)
S3 = (L{∪,¬} , F, {(A∪¬A)}, (r) , for any A, B ∈ F ). (4.4)
(B ∪ (A ∪ ¬A))

162
 B
S4 = (L{∪,¬} , F, {(A∪¬A)}, (r) , for any A, B ∈ F ), (4.5)
(B ∪ (A ∪ ¬A))

Exercise 4.2 Given proof systems S3 and S4 defined by (4.4), (7.40), respec-
tively.
1. Describe the sets PS3 , PS4 of provable formulas of S3 and S4 .
2. Decide whether is it true/ false that PS3 = PS4 . If yes, prove it, if not, give
an example of a formula A such that A ∈ PS4 and A 6∈ PS3 , or vice versa.

Solution 1.
Let’s first describe the set of provable formulas of both systems. Consider proof
system S3 . Obviously, for any formula A ∈ F, (A ∪ ¬A), as it is the axiom. It
constitutes a proof of length 1 A1 = (A ∪ ¬A)and we have that

`S3 (A ∪ ¬A).

One application of the inference rule (r) to axiom(A ∪ ¬A) gives us a proof
A1 = (A ∪ ¬A), A2 = ((A ∪ ¬A) ∪ (A ∪ ¬A)), and hence

`S3 ((A ∪ ¬A) ∪ (A ∪ ¬A)).

The application of the rule (r) to the already proven above formula A2 give us
the proof A1 = (A ∪ ¬A), A2 = ((A ∪ ¬A) ∪ (A ∪ ¬A)), A3 = (((A ∪ ¬A) ∪ (A ∪
¬A)) ∪ (A ∪ ¬A)), and

`S3 (((A ∪ ¬A) ∪ (A ∪ ¬A)) ∪ (A ∪ ¬A)).

It is easy to see that all provable formulas of S3 will be of the form of the proper
disjunction of the axiom of S3 , what we denote as follows:
[
PS3 = { (A ∪ ¬A)n : A ∈ F}, (4.6)
n∈N

where (A ∪ ¬A)n denotes a disjunction of n formulas of the form (A ∪ ¬A)

defined recursively as follows. (A ∪ ¬A)0 = (A ∪ ¬A), (A ∪ ¬A)n+1 = ((A ∪
¬A)n ∪ (A ∪ ¬A)).
Consider now system S4 . Obviously, as in the case of S4 , `S4 (A ∪ ¬A). One
application of the inference rule to the axiom gives us the proof A1 = (A ∪
¬A), A2 = (B ∪ (A ∪ ¬A)) and we have that

`S4 (B ∪ (A ∪ ¬A)), (4.7)

where B can be any formula from F.

The rule (r) can’t be, by its definition, applied to already proved B ∪ (A ∪ ¬A)).
We can continue with the proof A1 , A2 by constructing for example a proof

163
A1 , A2 , A3 , A4 by inserting axiom (C ∪ ¬C) (or axiom (A ∪ ¬A), if we wish
as A3 step of the proof. We have to remember that the definition 4.4 of the
formal proof allows us to insert an axiom in any place within the proof. A1 =
(A ∪ ¬A), A2 = (B ∪ (A ∪ ¬A)), A3 = (C ∪ ¬C), A4 = (A ∪ (C ∪ ¬C)) and
hence

`S4 (A ∪ (C ∪ ¬C)), `S4 (B ∪ (A ∪ ¬A)), `S4 (C ∪ (B ∪ ¬B)), ......etc...

Multiple application of the rule (r) in S4 means its application to multiple forms
of the axiom. Finally it is clear that we can only construct formal proofs of all
possible formulas of the form (B ∪ (A ∪ ¬A)), and of course of a form of any
axiom (proofs of the length 1) (A ∪ ¬A) for A, B being all possible formulas.
Remark that by saying A, B ∈ F we do not say that A 6= B, that we do not
exclude that case A = B. In particular case we have that

`S4 (A ∪ (A ∪ ¬A)), `S4 (B ∪ (B ∪ ¬B)), `S4 (C ∪ (C ∪ ¬C)), ......etc...

Hence

PS4 = {(B ∪ (A ∪ ¬A)) : A, B ∈ F} ∪ {(A ∪ ¬A) : A ∈ F}. (4.8)

Solution 2.
D = n∈N (A ∪ ¬A)n .
S
We prove now that PS3 ⊆ PS4 . S Let D ∈ PS3 . By (4.6)
D = n∈N (A ∪ ¬A)n = ( n∈N (A ∪ ¬A)n−1 ∪ (A ∪
S
Observe that
S by definition n−1
¬A)) and n∈N (A ∪ ¬A) is a formula of L{∪,¬} . We can denote it by B. We
have proved in (4.7) that for any B ∈ F, `S4 (B ∪ (A ∪ ¬A)). But by definition
D = (B ∪ (A ∪ ¬A)), hence we proved that D ∈ PS4 .This ends the proof.

Consider a formula ((a ∪ ¬b) ∪ (a ∪ ¬a)) of L{∪,¬} . It has a following formal

proof A1 , A2 in S4 .

A1 = (a ∪ ¬a), A2 = ((a ∪ ¬b) ∪ (a ∪ ¬a)).

axiom for A=a rule (r) application
for B = (a ∪ ¬b)

This proves that ((a ∪ ¬b) ∪ (a ∪ ¬a)) ∈ PS4 . Obviously 6 `S3 ((a ∪ ¬b) ∪ (a ∪ ¬a))
and ((a ∪ ¬b) ∪ (a ∪ ¬a)) 6∈ PS3 . We have proved that the proof systems S3 and
S4 defined by (4.4), (7.40) are such that PS3 ⊆ PS4 and PS3 6= PS4 .
Consider now a following proof system S5 .

S5 = ( L{⇒,∪,¬} , F, {(A ⇒ (A ∪ B))}, {(r1), (r2)} ) (4.9)

where the rules of inference are defined as follows.

A ;B A ; (A ∪ B)
(r1) , (r2) .
(A ∪ ¬B) B

164
Exercise 4.3
Given proof systems S5 defined by (7.41).
1. Find a formal proof of a formula ¬(A ⇒ (A ∪ B)) in S5 , i.e. show that
`S5 ¬(A ⇒ (A ∪ B)).
2. Find a formal proof of a formula ¬((a ∪ ¬b) ⇒ ((a ∪ ¬b) ∪ (a ∪ ¬a))).

Solution
1. We construct a formal proof B1 , B2 , B3 , B4 of the formula ¬(A ⇒ (A ∪ B))
as follows. We write comments next to each step of the proof.
B1 = (A ⇒ (A ∪ B)) axiom, B2 = (A ⇒ (A ∪ B)) axiom,
B3 = ((A ⇒ (A ∪ B)) ∪ ¬(A ⇒ (A ∪ B))) rule (r1) application to B1 and B2 ,
B4 = ¬(A ⇒ (A ∪ B)) rule (r2) application to B3
for A = (A ⇒ (A ∪ B)) and B = ¬(A ⇒ (A ∪ B)).

2. We construct a formal proof B1 , B2 , B3 , B4 of the formula ¬((a ∪ ¬b) ⇒

((a ∪ ¬b) ∪ (a ∪ ¬a))) as follows.
B1 = ((a∪¬b) ⇒ ((a∪¬b)∪(a∪¬a))) axiom for A = (a∪¬b) and B = (a∪¬a),
B2 = ((a∪¬b) ⇒ ((a∪¬b)∪(a∪¬a))) axiom for A = (a∪¬b) and B = (a∪¬a),
B3 = (((a ∪ ¬b) ⇒ ((a ∪ ¬b) ∪ (a ∪ ¬a))) ∪ ¬((a ∪ ¬b) ⇒ ((a ∪ ¬b) ∪ (a ∪ ¬a))))
rule (r1) application to B1 and B2 ,
B4 = ¬((a ∪ ¬b) ⇒ ((a ∪ ¬b) ∪ (a ∪ ¬a))) rule (r2) application to B3
for A = ((a ∪ ¬b) ⇒ ((a ∪ ¬b) ∪ (a ∪ ¬a))) and B = ¬((a ∪ ¬b) ⇒ ((a ∪ ¬b) ∪
(a ∪ ¬a))).

Observation 4.1 Observe that the formula ¬((a ∪ ¬b) ⇒ ((a ∪ ¬b) ∪ (a ∪ ¬a)))
is a particular case of the formula ¬(A ⇒ (A ∪ B)) for A = (a ∪ ¬b) and
B = (a ∪ ¬a) and its proof is just a particular case of the proof constructed in
case 1.

We wrote down independently a complicated proof of the particular case to

make reader aware of a need of generalizing particular formulas, if it possible,
and writing simpler proofs for the general case instead of the perticular.

165
4.1.1 Consequence Operation

Given a proof system S = (L, E, LA, R). While proving expressions we often
use some extra information available, besides the axioms of the proof system.
This extra information is called hypotheses in the proof.

Let Γ ⊆ E be a set expressions called hypotheses. A proof from a set Γ of

hypothesis of an expression E ∈ E in S = (L, LA, R) is a formal proof in S,
where the expressions from Γ are treated as additional hypothesis added to the
set LA of the logical axioms of the system S. We define it formally as follows.

Definition 4.8 (Proof from Hypotheses)

Given a proof system S = (L, E, LA, R) and let Γ be any set of expressions of
S, i.e. let Γ ⊆ E.
A proof of an expression E ∈ from the set Γ of expressions is a sequence

E1 , E2 , . . . En

of expressions, such that

E1 ∈ LA ∪ Γ, En = E

and for each i, 1 < i ≤ n, either Ei ∈ LA ∪ Γ or Ei is a direct consequence of

some of the preceding expressions in the sequence E1 , E2 , . . . En by virtue of
one of the rules of inference from R.
We write
Γ `S E
to denote that the expression E has a proof from Γ in S and Γ ` E, when the
system S is fixed.

When the set of hypothesis Γ is a finite set and Γ = {B1 , B2 , ..., Bn }, then we
write
B1 , B2 , ..., Bn `S E

instead of {B1 , B2 , ..., Bn } `S E. The case when Γ is an empty set i.e. when
Γ = ∅ is a special one. By the definition of a proof of E from Γ, ∅ ` E means
that in the proof of E only logical axioms LA of S were used. We hence write
it as we did before
`S E
to denote that E has a proof from the empty set Γ. The set of all expressions
provable from Γ (and logical axioms LA in S is denoted by PS (Γ), i.e.

PS (Γ) = {E ∈ E : Γ `S E}. (4.10)

166
When discussing properties of provability in proof systems we often use a notion
of a consequence operation. In order to follow this tradition we call provable
expressions from Γ in S consequences of Γ. The set of all expressions provable
is then called the set of all consequences from Γ. We observe that when talking
about consequences of Γ in S, we define in fact a function which to every set
Γ ⊆ E assigns a set of all its consequences. We denote this function by CnS
and adopt the following definition.

Definition 4.9 (Consequence)

Given a proof system S = (L, E, LA, R). Any function CnS : 2E −→ 2E
such that for every Γ ∈ 2E ,

CnS (Γ) = {E ∈ E : Γ `S E} (4.11)

is called a consequence determined by S.

Directly from definition 4.14 and (4.10) we have the following.

Fact 4.2 For any proof system S = (L, E, LA, R).

PS (Γ) = CnS (Γ). (4.12)

It proves that the notions of provability from a set Γ in S and consequence

determined by S coincide . It means that we can, and we often use in the
literature the both terms interchangeably.
The definition 4.14 does do more then just re-naming ”provability” by ”con-
sequence”. We are going to prove now that the consequence CnS determined
by S a special (an important) case of a notion a classic consequence operation
as defined by Alfred Tarski in 1930 as a general model of deductive reasoning.
Tarski definition is a formalization of the intuitive concept of the deduction as a
consequence, and therefore it has all the properties which our intuition attribute
to this notion. Here is the definition.

Definition 4.10 (Tarski)

By a consequence operation in a formal language L = (A, F) we under-
stand any mapping C : 2F −→ 2F satisfying the following conditions (t1)
- (t3) expressing properties of reflexivity, monotonicity, and transitivity of the
consequence.

For any sets F, F0 , F1 , F2 , F3 ∈ 2F ,

(t1) F ⊆ C(F ) reflexivity,
(t2) if F1 ⊆ F2 , then C(F1 ) ⊆ C(F2 ), monotonicity.

167
 (t3) if F1 ⊆ C(F2 ) and F2 ⊆ C(F3 ) , then F1 ⊆ C(F3 ), transitivity.
We say that the consequence operation C has a finite character if additionally
it satisfies the following condition t4.
(t4) if a formula B ∈ C(F ), then there exists a finite set F0 ⊆ F , such
that B ∈ C(F0 ). finiteness.

The monotonicity condition (t2) and transitivity condition (t3) are often re-
placed by the following conditions (t2’), (t3’), respectively.
For any formula B ∈ F, any any sets F, F 0 , ∈ 2F ,

(t20 ) if B ∈ C(F ), then B ∈ C(F ∪ F 0 ), (4.13)

(t30 ) C(F ) = C(C(F )). (4.14)

We express the correctness of the replacement conditions (4.13) and (4.14) in a

form of a following theorem.

Theorem 4.1
The Tarski definition 4.10 is equivalent with definitions where one, or both con-
ditions (t2), (t3) are replaced respectively by conditions (t2’), (t3’)given by equa-
tions (4.13) and (4.14).

Proof We prove the equivalency of conditions (t1) - (t3) and (t1) - (t3’). We
leave the proof of the other equivalency to the reader.
Assume (t3). By substituting

F1 = C(C(F )), F2 = C(F ), F3 = F

in (t3) we obtain
C(C(F )) ⊆ C(F ).
On the other hand, it follows from (t1) and (t2)

C(F ) ⊆ C(C(F )),

which with the previous inclusion gives (t3’). Conversely, suppose that (t3’) is
satisfied. If F2 ⊆ C(F3 ), then by (t2) we obtain C(F2 ) ⊆ C(C(F3 )). By (t3’)
C(C(F3 )) = (C(F3 ), hence C(F2 ) ⊆ (C(F3 ) and we proved (t3).

The consequence operation provides a model describing what we intuitively call

a deduction. It formalizes the basic, intuitively obvious and accepted properties
of reasoning by which we obtain (deduce) new facts from already known, or as-
sumed. We hence use it to define, after Tarski, a following notion of a deductive
system.

168
Definition 4.11 (Deductive System)
Given a formal language L = (A, F) and a Tarski consequence C (definition
4.10). A system
D = (L, C)
is called a Tarski deductive system for the language L.

Tarski’s deductive system as a model of reasoning does not provide a method

of actually defining a consequence operation in the language L; it assumes that
it is given. We are going to prove now that our definition 4.14 of consequence
operation CnS determined by a proof system S is a Tarski consequence operation
C in the definition 4.10 sense. It justifies, together with Fact 4.2 the common
use the consequence notion when talking about provability. It means that each
proof system S provides a different example of a consequence operation. They
are all purely syntactic in nature and all defined by the notion of provability.
Hence each proof system can be treated and a syntactic Tarski deductive system
from definition 4.11.

Theorem 4.2
Given a proof system S = (L, E, LA, R). The consequence operation CnS is a
Tarski consequence C in the language L of the system S and the system

DS = (L, CnS )

is Tarski deductive system. We call it a syntactic deductive system deter-

mined by S. Moreover, the consequence operation CnS that has a finite char-
acter.

Proof
By definition 4.14, the consequence operation CnS : 2E −→ 2E is given
by a formula CnS (Γ) = {E ∈ E : Γ `S E}. We have to show that for any
Γ, Γ0 , Γ1 , Γ2 , Γ3 ∈ 2F conditions (t1) - (t4) of the definition 4.14 hold. The
reflexivity condition (t1) becomes Γ ⊆ CnS (Γ). Let E ∈ Γ. The one element
sequence E is a proof of E from Γ, hence we proved that E ∈ C(Γ) and (t1)
holds. To prove the transitivity condition (t2) assume now that Γ1 ⊆ Γ2 . Let
E ∈ CnS (Γ1 ). It means that Γ1 `S E, i.e E has a formal proof from Γ1 , but
Γ1 ⊆ Γ2 , hence this proof also is a proof from Γ2 , and E ∈ CnS (Γ2 ). This
proves that CnS (Γ1 ) ⊆ CnS (Γ2 ) and the condition (t2) holds. Let now E ∈ Γ1
and Γ1 ⊆ CnS (Γ2 ), so E ∈ CnS (Γ) 2. Let E1 , . . . , En be a formal proof of E
from Γ2 . But Γ2 ⊆ CnS (Γ3 ). It means that any expression from Γ2 has a formal
proof from Γ3 . In particular, all expression in the proof E1 , . . . , En that belong
to Γ2 have their formal proofs from Γ3 . Replacing all these expressions by their
proofs from Γ3 we obtain a proof of E from Γ3 . This proves that Γ1 ⊆ CnS Γ3
and the transitivity condition (t3) holds. Let now E ∈ CnS Γ. This means that
E has a proof E1 , . . . , En from Γ. The set Γ0 = {E1 , . . . , En } is obviously a

169
finite subset of Γ and E ∈ CnS Γ0 and (t4) holds.

Non - Monotonic Logics

The Tarski definition 4.10 of a consequence models reasoning which is called
after its condition (t2) or (t2’) a monotonic reasoning. The monotonicity of
reasoning was, since antiquity is the the basic assumption while developing
models for classical and well established non-classical logics. Recently many of
new non- classical logics were developed and are being developed by computer
scientists. For example new modal logics of agents and temporal logics. Tempo-
ral logics are essential in developing theory of programs. Nevertheless they all
are built following the Tarski definition of consequence and were and are called
monotonic logics. A new type of important
Non-Monotonic logics have been proposed at the beginning of the 80s. Histor-
ically the most important proposals are Non-monotonic logic, by McDermott
and Doyle, Default Logic, by Reiter, Circumscription, by McCarthy, and Au-
toepistemic logic, by Moore.
The term non-monotonic logic covers a family of formal frameworks devised
to capture and represent defeasible inference. It is an inference in which it
is possible to draw conclusions tentatively, reserving the right to retract them
in the light of further information. We included most standard examples in
Chapter 1, Introduction.
Non-monotonic logics describe commonsense reasoning which is neither a restric-
tion nor an extension of classical logic. Consequences of premises are drawn as
much due to the absence as to the presence of knowledge. When more knowledge
is acquired, conclusions previously drawn may have to be withdrawn because
the rules of inference that led to them no longer are active. Intelligent decision
makers use this form of commonsense reasoning to infer actions to be performed
from premises which cannot be made by classical logic inference, because they
simply have to make decisions whether or not there is enough information for
a classical logical deduction. Non-monotonic reasoning deals with the problem
of deriving plausible conclusions, but not infallible, from a knowledge base (a
set of formulas). Since the conclusions are not certain, it must be possible to
retract some of them if new information shows that they are wrong. Example:
let the KB contain: Typically birds fly. Penguins do not fly. Tweety is a bird. It
is plausible to conclude that Tweety flies. However if the following information
is added to KB Tweety is a penguin the previous conclusion must be retracted
and, instead, the new conclusion that Tweety does not fly will hold.
The statement ”typically A” can be read as: ”in the absence of information to
the contrary, assume A”. The problem is to define the precise meaning of ”in
the absence of information to the contrary”. The meaning could be: ”there is
nothing in KB that is inconsistent with assumption A”. Other interpretations
are possible Different interpretations give rise to different non-monotonic logics.

170
4.1.2 Syntactic Consistency

Formal theories play crucial role in mathematics and were historically defined
for classical first order logic and consequently for other first and higher order
logics. They are routinely called first order theories. We will discuss them in
more detail in chapter 10 dealing formally with classical predicate logic. First
order theories are hence based on proof systems S with a predicate (first order)
language L. We will call them for short first order proof systems.
We can and we sometimes consider formal theories based on propositional log-
ics, i.e. based on proof systems with language L being propositional. We will
call them propositional theories.

Given a proof system S = (L, E, LA, R). We build (define) a formal theory
based on S as follows.
1. We select a certain finite subset SA of expressions of S, disjoint with the
logical axioms LA of S, i.e. such that LA ∩ SA = ∅.. The set SA is called a set
of specific axioms of the formal theory based on S.

2. We use set SA of specific axioms to define a language

LSA , (4.15)
called a language of the formal theory. Here we have two cases.
c1. S is a first order proof system, i.e. L of S is a predicate language. We
define the language LSA by restricting the sets of constant, functional, predicate
symbols of L to constant, functional, predicate symbols appearing in the set SA
of specific axioms. Both languages LSA and L share the same set of propositional
connectives.
c2. S is a propositional proof system, i.e. L of S is a propositional language.
LSA is defined by restricting L to connectives appearing in the set SA.

Definition 4.12 (Formal Theory)

The system
T = (L, E, LA, SA, R) (4.16)
is called a formal theory based on a proof system S.
The set SA of the set of specific axioms of T . The language LSA defined by
(4.15) is called the language of the theory T .

The set ESA of all expressions of the language LSA provable from the set specific
axioms SA (and logical axioms LA) i.e. the set
T(SA) = {E ∈ ESA : SA `S E } (4.17)

171
is called the set of all theorems of the theory T .
If the set SA of specific axioms of T is empty, then the theory T is, by definition,
identified with the system S, i.e. T = S = (L, E, LA, R).

Definition 4.13 (Consistent Theory)

A theory T = (L, E, LA, SA, R) is consistent if there exists an expression
E ∈ ESA such that E 6∈ T(SA), i.e. such that

SA 6 `S E;

otherwise the theory T is inconsistent.

Observe that the definition 4.13 has purely syntactic meaning. It also reflexes
our intuition what proper provability should mean. it says that a formal the-
ory T based on a proof system S is consistent only when it does not prove all
expressions (formulas in particular cases) of LSA . The theory T such that it
proves everything stated in LSA obviously should be, and its defined as incon-
sistent. In particular, we have the following syntactic definition of consistency-
inconsistency for any proof system S.

Definition 4.14 (Consistent Proof System)

A proof system S = (L, E, LA, R) is consistent if there exists E ∈ E such
that E 6∈ PS , i.e. such that 6 `S E; otherwise S is inconsistent.

4.2 Semantics
We define formally a semantics for a given proof system S = (L, E, LA, R) by
specifying the semantic links of all its components as follows.

Semantic Link1: Language L

The language L of S can be propositional or predicate. Let denote my M a
semantic for L. We call it, for short, a semantics for the proof system S. The
semantics M can be a propositional, a predicate, extensional, or not extensional.
We use M as a general symbol for a semantics.

Semantic Link 2: Set E of Expressons

We always have to extend a given semantics M of the language L to the set
of expressions E of the system S. We often do it by establishing a seman-
tic equivalency under semantics M of E and the set of all formulas F of L.
It means we prove that for a given semantics M under which we build our

172
proof system S, and for any expression E ∈ E there is a formula A ∈ F,
such that E≡M A. For example, in the automated theorem proving system
RS presented in chapter 6 the expressions are finite sequences of formulas of
L = L¬,∩,∪,⇒ . We extend our classical semantics for L to the set F ∗ of all finite
sequences of formulas as follows: for any v : V AR −→ {F, T } and any ∆ ∈ F ∗ ,
∆ = A1 , A2 , ..An , v ∗ (∆) = v ∗ (A1 , A2 , ..An ) = v ∗ (A1 ) ∪ v ∗ (A2 ) ∪ .... ∪ v ∗ (An ),
i.e. ∆ ≡ (A1 ∪ A2 ∪ ... ∪ An ). Sometimes, like in case of Resolution based proof
systems we have also to prove a semantic equivalency of a given formula A of L
with some set EA of expressions (sets of clauses ) representing the formula A.

Semantic Link 3: Logical Axioms LA

Given a semantics M for L and its extension to the set E of all expressions. We
extend the notion of tautology to the set L of expressions and write |=M E to
denote that the expression E ∈ E is a tautology under semantics M. We denote
TM = {E ∈ E : |=M E}
While designing a proof system S we want the logical axioms LA to be a subset
of expressions that are tautologies of under the semantics M, i.e.
LA ⊆ TM .

We can, and we often do, invent proof systems with languages without yet es-
tablished semantics. In this case the logical axioms LA serve as description of
properties of tautologies under a future semantics yet to be built. We want to
choose as logical axioms of a proof system S are not only tautologies under an
already known semantics M, but they can also guide us how to define a seman-
tics when it is yet unknown.

Semantic Link 4: Rules of Inference R

We want the rules of inference r ∈ R to preserve truthfulness. Rules that

preserve the truthfulness are called sound under a given semantics M. We put
it in a general formal definition as follows.

Definition 4.15 (Sound Rule under M)

Given an inference rule r ∈ R of the form
P1 ; P2 ; .... ; Pm
(r) .
C
We say that the rule (r) is sound under a semantics M if the following con-
dition holds for all M models M.
If M |=M {P1 , P2 , .Pm } then M |=M C. (4.18)

173
In case of a propositional language LCON and an extensional semantics M the
M models M are defined in terms of the truth assignment v : V AR −→ LV ,
where LV is the set of logical values with a distinguished value T. The general
definition 4.15 becomes a following definition for a propositional language L and
its extensional semantics M.

Definition 4.16 (Sound Propositional Rule under M)

Given a propositional language LCON and an extensional semantics M, an in-
ference rule of the form
P1 ; P2 ; .... ; Pm
(r)
C
is sound under the semantics M if the following condition holds for any v :
V AR −→ LV .

If v |=M {P1 , P2 , . . . , Pm }, then v |=M C. (4.19)

Observe that we can rewrite the condition (4.19) as follow.

If v ∗ (P1 ) = v ∗ (P2 ) = . . . . = v ∗ (Pm ) = T, then v ∗ (C) = T. (4.20)

A rule of inference be sound under different semantics, but also rules of inference
can be sound under one semantics and not sound under the other.

Example 4.3 Given a propositional language L{¬,∪,⇒} . Consider two rules of

inference:
(A ⇒ B) ¬¬A
(r1) and (r2) .
(B ⇒ (A ⇒ B)) A
The rule (r1) is sound under classical, H and L semantics. The (r2) is sound
under classical and L semantics but is not sound under H semantics.

Consider the rule (r1).

Let v : V AR −→ {F, T } be any truth assignment, such that v ∗ ((A ⇒ B)) = T .
We use condition (4.20) and evaluate logical value of the conclusion under v
as follows. v ∗ ((B ⇒ (A ⇒ B))) = v ∗ (B) ⇒ T = T for any formula B and
any value of v ∗ (B). This proves that v ∗ (B ⇒ (A ⇒ B)) = T and hence the
soundness of (r1). Consider now the H semantics. Let v : V AR −→ {F, ⊥, T } be
any truth assignment, such that v |=H (A ⇒ B), i.e. such that v ∗ ((A ⇒ B)) = T .
We evaluate under H, L semantics v ∗ ((B ⇒ (A ⇒ B))) = v ∗ (B) ⇒ T . Now
v ∗ (B) can be T, F as in classical case, or v ∗ (B) =⊥. The case when v ∗ (B) is
T, F is like in classical semantics, so we have to check the case v ∗ (B) =⊥. But
in both H and L semantics ⊥⇒ T = T . This proves that (r1) is also sound
under H and L semantics.

174
Consider the rule (r2).
The rule (r2) is sound under classical and L by straightforward eveluation.
Assume now v : V AR −→ {F, ⊥, T } be any truth assignment, such that
v |=M ¬¬A, i.e. such that v ∗ (¬¬A) = T under H semantics. We have that
v ∗ (¬¬A) = ¬¬v ∗ (A) = T if and only if ¬v ∗ (A) = F if and only if v ∗ (A) = T or
v ∗ (A) =⊥. This proves that that it is possible to have v |=M ¬¬A and v 6 |=H A,
i.e. that (r2) is not sound.

Definition 4.17 ( Strongly Sound Rule under M)

An inference rule r ∈ R of the form
P1 ; P2 ; .... ; Pm
(r)
C
is strongly sound under a semantics M if the following condition holds for
all M structures M,

M |=M {P1 , P2 , .Pm } if and only if M |=M C. (4.21)

In case of a propositional language LCON and an extensional semantics M the

condition (4.21) is as follows. For for any v : V AR −→ LV ,

v |=M {P1 , P2 , .Pm } if and only if v |=M C. (4.22)

We introduce also a property of a rule being s-strongly sound. We state it

informally as follows. An inference rule r ∈ R is s-strongly sound when the
conjunction of all its premisses is logically equivalent (under a semantics M) to
its conclusion. We denote it informally as

P1 ∩ P2 ∩ . . . ∩ Pm ≡M C. (4.23)

Of course any s-strongly sound rule is strongly sound.

Example 4.4
Given a propositional language L{¬,∪,⇒} . Consider two rules of inference:

A; B A
(r1) and (r2) .
(A ∪ ¬B) ¬¬A

Both rules (r1) and (r2) are sound under classical and H semantics. The rule
(r2) is strongly and s-strongly sound under classical semantics but is not strongly
sound under H semantics. The rule (r1) in not strongly sound under either
semantics.

Consider (r1). Take (in shorthand notation) for A = T and B = T . We evaluate

v ∗ ((A ∪ ¬B)) = T ∪ F = F in both semantics. This proves soundness of (r1)

175
under both semantics.. Take now v such that v(A) = T and v(B) = F , we get
v ∗ ((A ∪ ¬B)) = F ∪ T = T .This proves that v |= (A ∪ ¬B) and v|=H (A ∪ ¬B).
Obviously v 6|= {A, B} and v 6 |=H {A, B}. This proves that (r1) in not strongly
sound under either semantics.
Consider (r2). It is strongly sound under classical semantic. By (8.77) and the
fact that A 𠪪A (r2) is s-strongly sound and so is strongly sound.
(r2) is sound under H semantics. Assume A = T . We evaluate (in shorthand
notation) ¬¬A = ¬¬T = ¬F = T . (r2) is not strongly sound under H seman-
tics. Take v such that v ∗ (A) =⊥, then v ∗ (¬¬A) = ¬¬ ⊥= ¬F = T . This
proves that there is v such that v |= ¬¬A and v 6|= A and (4.22) does not hold
and so (r2) is not strongly sound.
This also proves that A 6≡H ¬¬A, i.e. (r2) is not s-strongly sound (4.22).
Now we are ready to define a notion of a sound and strongly sound proof system.
Strongly sound proof systems play a role in constructive proofs of completeness
theorem. This is why we introduced and singled them out here.

Definition 4.18 (Sound, Strongly Sound)

Given a proof system S = (L, E, LA, R).
We say that the proof system S is sound under a semantics M if the following
conditions hold.
C1. LA ⊆ TM ; C2. Each rule of inference r ∈ R is sound under M.
The proof system S is strongly sound under a semantics M if the condition
C2. is replaced by the following condition
C2’. Each rule of inference r ∈ R is strongly sound under M.

Here is a simple but an important fact about the notion of strong soundness.
The strong soundness of proof systems guarantee a correctness of of construc-
tive proofs of completeness theorems for automated theorem proving systems
introduced and examined in chapter 6 and in chapter 10.

Theorem 4.3 (Counter-Model)

Given a proof system S = (L, E, LA, R) that is strongly sound under a
semantics M. For any r ∈ R, for any M that is a M counter model for one of
its premisses, M is also the M counter model for its conclusion.

Proof
Consider a rule
P1 ; P2 ; .... ; Pm
(r) .
C

176
By definition 4.17, (r) is sound and additionally the following implication holds
for any M structures M,

if M |=M C, then M |=M {P1 , P2 , .Pm }.

It means that inverse implication

if M 6|=M {P1 , P2 , .Pm }, then M 6|=M C

holds for all M structures M. But M 6|=M {P1 , P2 , .Pm } if and only if there
is 1 ≤ i ≤ m such that M 6|=M Pi . This ends the proof.

Example 4.5 The proof system S defined below as follows

S = (L{¬,⇒} , F, {(¬¬A ⇒ A), (A ⇒ (¬A ⇒ B))}, R = {(r)})

where
(A ⇒ B)
(r) .
(B ⇒ (A ⇒ B))
S is sound, but not strongly sound under classical and L semantics. It is not
sound under H semantics.

C1. Both axioms are basic classical tautologies. Hence to prove that first axiom
is L tautology we we have to verify only the case (shorthand notation) A =⊥.
But ¬¬ ⊥⇒⊥= ¬ ⊥⇒⊥=⊥⇒⊥= T and we proved |=L (¬¬A ⇒ A). Observe
that (A ⇒ (¬A ⇒ B)) =⊥ if and only if A = T and (¬A ⇒ B) =⊥ if and
only if (¬T ⇒ B) =⊥ if and only if (F ⇒ B) =⊥, what is impossible undef
L semantics. Hence |=L (A ⇒ (¬A ⇒ B)). We prove, as in example 4.3, that
6|=H (¬¬A ⇒ A), and hence S is not sound under H semantics.
C2. The rule (r) is not strongly sound under neither classical nor L semantic.
Let v : V AR −→ {F, ⊥, T } be any truth assignment, such that v ∗ (B) = F
and v ∗ (A) = F . Obviously v |= (B ⇒ (A ⇒ B)) and v |=L (B ⇒ (A ⇒ B)),
but v 6|= (A ⇒ B) and v 6|=L (A ⇒ B).
Nevertheless, (r) is sound under the both semantics by example 4.3.

Let PS be the set of all provable expressions of S, i.e. PS = {E ∈ E : `S E}.

Let TM be a set of all expressions of S that are tautologies under a semantics
M, TM = {E ∈ E : |=M E}.
When we define (develop) a proof system S our first goal is to make sure that
it a ”sound one”, i.e. that all we prove in it is true (with respect to a given
semantics). Proving the following theorem establishes this goal.

Theorem 4.4 (Soundness Theorem)

177
Given a predicate proof system S and a semantics M.
The following holds.
PS ⊆ TM , (4.24)

i.e. for any E ∈ E, the following implication holds

if `S E then |=M E.

Proof We prove by Mathematical Induction over the length of a proof that if

S is sound as stated in the definition 8.27, the Soundness Theorem holds for S.
It means that in order to prove the Soundness Theorem (under semantics bf M
) for a proof system we have to verify the two conditions: 1. LA ⊆ TM and
2. Each rule of inference r ∈ R is sound under M.

The next step in developing a logic is to answer next necessary and a difficult
question: Given a proof system S, about which we know that all it proves it
true (tautology)with respect to a given semantics. Can we prove all we know to
be true (all tautologies) with respect to the given semantics?

Theorem 4.5 (Completeness Theorem)

Given a predicate proof system S and a semantics M.
The following holds
PS = TM (4.25)

i.e. for any E ∈ E, the following holds

`S E if and only if |=M E.

The Completeness Theorem consists of two parts:

Part 1: Soundness Theorem: PS ⊆ TM .
Part 2: Completeness Part of the Completeness Theorem: TM ⊆ PS .
Proving the Soundness Theorem for S under a semantics M is usually a straight-
forward and not a very difficult task. Proving the Completeness Part of the
Completeness Theorem is always a crucial and very difficult task. There are
many methods and techniques for doing so, even for classical proof systems (log-
ics) alone. Non-classical logics often require new sometimes very sophisticated
methods. We will study two proofs of the Completeness Theorem for classical
propositional Hilbert style proof system in chapter 5, and a constructive proofs
for automated theorem proving systems for classical logic the chapter 6. We
prove provide the proofs of the Completeness Theorem for classical predicate
logic in chapter 9 (Hilbert style) and chapter 10(Gentzen style).

178
4.3 Exercises and Examples
Exercise 4.4
Given a proof system:
(A ⇒ B)
S = (L{¬,⇒} , E = F LA = {(A ⇒ A), (A ⇒ (¬A ⇒ B))}, (r) ).
(B ⇒ (A ⇒ B))
1. Prove that S is sound, but not strongly sound under classical semantics.
2. Prove that S is not sound under K semantics.
3. Write a formal proof in S with 2 applications of the rule (r).

Solution
Parts 1 and 2. In order to prove 1. and 2. we have to verify conditions 1., 2.
and bf 2.’ of definition 8.27. Observe that both axioms of S are basic classical
tautologies. Consider the rule of inference of S.
(A ⇒ B)
(r) .
(B ⇒ (A ⇒ B))
Take any v such that v ∗ ((A ⇒ B))) = T . We evaluate logical value of the
conclusion under the truth assignment v as follows.
v ∗ (B ⇒ (A ⇒ B)) = v ∗ (B) ⇒ T = T
for any B and any value of v ∗ (B). This proves that S is sound under classical
semantics. S is not strongly sound as (A ⇒ B) 6≡ (B ⇒ (A ⇒ B)).
System S is not sound under K semantics because axiom (A ⇒ A) is not a K
semantics tautology.
Part 3. There are many solutions, i.e. one can construct many forrmal proofs.
Here is one of them. For example, one of the formal proofs is a sequence
A1 , A2 , A3 , where
A1 = (A ⇒ A)
(Axiom)
A2 = (A ⇒ (A ⇒ A))
Rule (r) application 1 for A = A, B = A.
A3 = ((A ⇒ A) ⇒ (A ⇒ (A ⇒ A)))
Rule (r) application 2 for A = A, B = (A ⇒ A).

Exercise 4.5
Prove, by constructing a formal proof that
`S ((¬A ⇒ B) ⇒ (A ⇒ (¬A ⇒ B))),
where S is the proof system from Exercise 5.9.

179
Solution
Required formal proof is a sequence A1 , A2 , where
A1 = (A ⇒ (¬A ⇒ B))
Axiom
A2 = ((¬A ⇒ B) ⇒ (A ⇒ (¬A ⇒ B)))
Rule (r) application for A = A, B = (¬A ⇒ B).
Observe that we needed only one application of the rule (r). One more applica-
tion of the rule (r) to A2 gives another solution to Exercise 5.9, namely a proof
A1 , A2 , A3 for A1 , A2 defined above and
A3 = ((A ⇒ (¬A ⇒ B)) ⇒ (¬A ⇒ B) ⇒ (A ⇒ (¬A ⇒ B)))
Rule (r) application for A = (¬A ⇒ B) and B = (A ⇒ (¬A ⇒ B)).

Exercise 4.6
Given a proof system:

(A ⇒ B)
S = (L{∪,⇒} , F, LA = {A1, A2}, (r) ),
(A ⇒ (A ⇒ B))

where A1 = (A ⇒ (A ∪ B)), A2 = (A ⇒ (B ⇒ A)).

1. Prove that S is sound under classical semantics and determine whether S it
is sound or not sound under K semantics.
2. Write a formal proof B1 , B2 , B3 in S with 2 applications of the rule (r) that
starts with axiom A1, i.e such that B1 = A1.
3. Write a formal proof B1 , B2 in S with 1 application of the rule (r) that
starts with axiom A2, i.e such that A1 = A2.

Solution
Part 1. Axioms of S are basic classical tautologies. The proof (in shorthand
notation) of soundness of the rule of inference is the following. Assume (A ⇒
B) = T . Hence the logical value of conclusion is (A ⇒ (A ⇒ B)) = (A ⇒ T ) =
T for all A. S is not sound under K semantics. Let’s take truth assignment
such that A =⊥, B =⊥. The logical value of axiom A1 is as follows.
(A ⇒ (A ∪ B)) = (⊥⇒ (⊥ ∪ ⊥)) =⊥ and 6 |=K (A ⇒ (A ∪ B)). Observe that
the v such that A =⊥, B =⊥ is not the only v that makes A1 6= T , i.e. proves
that 6 |=K A1.
(A ⇒ (A ∪ B)) 6= T if and only if (A ⇒ (A ∪ B)) = F or (A ⇒ (A ∪ B)) =⊥.
The first case is impossible because A1 is a classical tautology. Consider the
second case. (A ⇒ (A ∪ B)) =⊥ in two cases. c1. A =⊥ and (A ∪ B) = F , i.e.
(⊥ ∪B) = F , what is impossible. c2. A = T and (A ∪ B) =⊥, i.e. (T ∪ B) =⊥,
what is impossible. c3. A =⊥ and (A∪B) =⊥, i.e. (⊥ ∪B) =⊥. This is possible
for B =⊥ or B = F , i.e when A =⊥, B =⊥ or A =⊥, B = F . From the above
observation we get a second solution. S is not sound under K semantics. Axiom
A1 is not K semantics tautology. There are exactly two truth assignments v,

180
such that v 6|= A1. One is, as defined in the first solution, namely A =⊥, B =⊥.
The second is A =⊥, B = F .
Part 2. The formal proof B1 , B2 , B3 is as follows.
B1 = (A ⇒ (A ∪ B))
Axiom
B2 = (A ⇒ (A ⇒ (A ∪ B)))
Rule (r) application for A = A and B = (A ∪ B)
B3 = (A ⇒ (A ⇒ (A ⇒ (A ∪ B))))
Rule (r) application for A = A and B = (A ⇒ (A ∪ B)).
Part 3. The formal proof B1 , B2 is as follows.
B1 = (A ⇒ (B ⇒ A))
Axiom
B2 = (A ⇒ (A ⇒ (B ⇒ A))).
Rule (r) application for A = A and B = (B ⇒ A).

Exercise 4.7
Let S be the following proof system:

S = ( L{⇒,∪,¬} , F, A1, (r1), (r2) ),

where the logical axiom A1 is: A1 = (A ⇒ (A ∪ B)),

Rules of inference (r1), (r2) are:

A ;B A ; (A ∪ B)
(r1) , (r2) .
(A ∪ ¬B) B

1. Verify whether S is sound/not sound under classical semantics.

2. Find a formal proof of ¬(A ⇒ (A ∪ B)) in S, ie. show that

`S ¬(A ⇒ (A ∪ B)).

3. Does `S ¬(A ⇒ (A ∪ B)) prove that |= ¬(A ⇒ (A ∪ B))?

Solution
Part 1. The system is not sound. Take any v such that v ∗ (A) = T and
v ∗ (B) = F . The premiss (A∪B of the rule (r2) is T under v, and the conclusion
under v is v ∗ (B) = F .
Part 2. The proof of ¬(A ⇒ (A ∪ B)) is as follows.
B1 : (A ⇒ (A ∪ B)),
axiom
B2 : (A ⇒ (A ∪ B)),
axiom

181
B3 : ((A ⇒ (A ∪ B)) ∪ ¬(A ⇒ (A ∪ B))),
rule (r1) application to B1 and B2
B4 : ¬(A ⇒ (A ∪ B)),
rule (r2) application to B1 and B3 ).
Part 3. System S is not sound, so existence of a proof does not guarantee that
what we proved is a tautology. Moreover, the proof of ¬(A ⇒ (A ∪ B)) used
rule (r2) that is not sound.

Exercise 4.8
Create a 3 valued extensional semantics M for the language
L{¬, L, ∪, ⇒} by defining the connectives ¬, ∪, ⇒ on a set {F, ⊥, T } of logical
values. You must follow the following assumptions a1, a2.
a1 The third logical value value is intermediate between truth and falsity, i.e.
the set of logical values is ordered as follows: F <⊥< T.
a2 The value T is the designated value. The semantics has to model a situation
in which one ”likes” only truth; i.e. the connective L must be such that LT = T ,
L ⊥= F, and LF = F . The connectives ¬, ∪, ⇒ can be defined as you wish,
but you have to define them in such a way to make sure that
|=M (LA ∪ ¬LA).

Solution
Here is a simple M semantics. We define the logical connectives by writing
functions defining connectives in form of the truth tables.
M Semantics

L F ⊥ T ¬ F ⊥ T
F F T T F F

∩ F ⊥ T ∪ F ⊥ T ⇒ F ⊥ T
F F F F F F ⊥ T F T T T
⊥ F ⊥ ⊥ ⊥ ⊥ T T ⊥ T ⊥ T
T F ⊥ T T T T T T F F T

We verify whether the condition s3 is satisfied, i.e. whether |=LK (LA ∪ ¬LA)
by simple evaluation. Let v : V AR −→ {F, ⊥, T } be any truth assignment. For
any formula A, v ∗ (A) ∈ {F, ⊥, T } and LF ∪ ¬LF = LF ∪ ¬LF = F ∪ ¬F ∪ T =
T, L ⊥ ∪¬L ⊥= F ∪ ¬F = F ∪ T = T, LT ∪ ¬LT = T ∪ ¬T = F ∪ T = T.
We verify whether |=M (LA ∪ ¬LA) by simple evaluation. Let v : V AR −→
{F, ⊥, T } be any truth assignment. For any formula A, v ∗ (A) ∈ {F, ⊥, T } and
LF ∪ ¬LF = LF ∪ ¬LF = F ∪ ¬F ∪ T = T, L ⊥ ∪¬L ⊥= F ∪ ¬F = F ∪ T =
T, LT ∪ ¬LT = T ∪ ¬T = F ∪ T = T.

182
Exercise 4.9
Let S be the following proof system

S = ( L{¬,L,∪,⇒} , F, {A1, A2}, {(r1), (r2)} )

where the logical axioms A1, A2 and rules of inference (r1), (r2) defined for any
formulas A, B ∈ F as follows.
A1 (LA ∪ ¬LA),
A2 (A ⇒ LA),

A ;B A
(r1) , (r2) .
(A ∪ B) L(A ⇒ B)

1. Show, by constructing a proper formal proof that

`S ((Lb ∪ ¬Lb) ∪ L((La ∪ ¬La) ⇒ b))).

Please, write comments how each step of the proof was obtained
2. Verify whether the system S is M-sound funder the semantics M you have
developed in Exercise 4.8. You can use shorthand notation.
3. If the system S is not sound/ sound under your semantics M then re-
define the connectives in a way that such obtained new semantics N would make
S sound/not sound

Solution
Part 1. Here is the formal proof B1 , B2 , B3 , B4 .
B1 : (La ∪ ¬La), axiom A1 for A = a,
B2 : L((La ∪ ¬La) ⇒ b), rule (r2) for B = b applied to B1 ,
B3 : (Lb ∪ ¬LAb), axiom A1 for A= b,
B4 : ((Lb ∪ ¬Lb) ∪ L((La ∪ ¬La) ⇒ b)), rule (r1) applied to B3 and B2 .
Part 2. Observe that both logical axioms of S are M tautologies. A1 is M
tautology by definition of the semantics, A2 is M tautology by direct evaluation.
Rule (r1) is sound because when A = T and B = T we get A ∪ B = T ∪ T = T .
Rule (r2) is not sound because when A = T and B = F (or B =⊥ ) we get
L(A ⇒ B) = L(T ⇒ F ) = LF = F (or L(T ⇒⊥) = L ⊥= F ).
Part 3. In order to make the rule (r2) sound while preserving the soundness
of axioms A1, A2 we have to modify only the definition of implication. Here is
the N semantics implication

⇒ F ⊥ T
F T T T
⊥ T ⊥ T
T T T T

183
Observe that it would be hard to convince anybody to use our sound proof
system S, as it would be hard to convince anybody to adopt our N semantics.

4.4 Homework Problems

(A⇒B)
1. Given a proof system S = (L{∪,⇒} , F, LA = {A1, A2}, (r) (A⇒(A⇒B)) ),
where A1 = (A ⇒ (A ∪ B)), A2 = (A ⇒ (B ⇒ A)). Prove, by construct-
ing a formal proof in S that `S (A ⇒ (A ⇒ (A ⇒ (A ⇒ A)))).
Does it prove that |= (A ⇒ (A ⇒ (A ⇒ (A ⇒ A)))).
2. Given a proof system: S = (L{¬,⇒} , E = F LA = {(A ⇒ A), (A ⇒
(A⇒B)
(¬A ⇒ B))}, (r) (B⇒(A⇒B)) ).
(i) Prove that S is sound under classical semantics.
(ii) Prove that S is not sound under K semantics.
(iii) Write a formal proof in S with 3 applications of the rule (r).
(iv) Prove, by constructing a formal proof that `S ((¬A ⇒ B) ⇒ (A ⇒
(¬A ⇒ B))).
3. Given a proof system: S = (L{∪,⇒} , E = F LA = {A1, A2}, R =
{(r)} ), where A1 = (A ⇒ (A ∪ B)), A2 = (A ⇒ (B ⇒ A)) and
(A⇒B)
(r) (A⇒(A⇒B)) .
(i) Prove that S is sound under classical semantics.
(ii) Determine whether S is sound or not sound under L semantics.
(iii) Write a formal proof in S with 3 applications of the rule (r) that
starts with axiom A1.
(iv) Does it prove/ that |=L A for a formula A obtained in (iii).
(v) Prove, by constructing a formal proof in S that `S (A ⇒ (A ⇒ (A ⇒
(A ⇒ A)))). Does it prove (and why) that |= (A ⇒ (A ⇒ (A ⇒ (A ⇒
A)))).
4. S is the following proof system: S = ( L{⇒,∪,¬} , F, LA = {(A ⇒
A ;B
(A ∪ B))} (r1), (r2) ), where (r1) (A∪¬B) , (r2) A ;(A∪B)
B .
(i) Verify whether S is sound/not sound under classical semantics. (ii)
Find a formal proof of ¬(A ⇒ (A ∪ B)) in S, i. e. show that
`S ¬(A ⇒ (A ∪ B)).
(iii) Does above (ii) prove that |= ¬(A ⇒ (A ∪ B))?
5. By a m-valued semantics Sm , for a propositional language L = L{¬,∩,∪,⇒}
we understand any definition of of connectives ¬, ∩, ∪, ⇒ as operations on
a set Lm = {l1 , l2 , ...lm } of logical values (for m ≥ 2).

184
We assume that l1 ≤ l2 ≤ ... ≤ lm , i.e. Lm is totally ordered by a
certain relation ≤ with l1 , lm being smallest and greatest elements, re-
spectively. We denote l1 = F, lm = T and call them (total) False and
Truth, respectively. For example, when m = 2, L2 = {F, T }, F ≤ T .
Semantics S2 is called a classical semantics if the connectives are de-
fined as x ∪ y = max{x, y}, x ∩ y = min{x, y}, ¬T = F, ¬F = T,
and x ⇒ y = ¬x ∪ y, for any x, y ∈ L2 .
Let V AR be a set of propositional variables of L and let Sm be any m-
valued semantics for L. A truth assignment v : V AR −→ Lm is called
a Sm model for a formula A of L if and only if v ∗ (A) = T and logical
value v ∗ (A) is evaluated accordingly to the semantics Sm . We denote is
symbolically as v|=Sm A.
A ;(A⇒B)
Let S = (L, F, {A1, A2, A3}, M P B ) be a proof system with
logical axioms:
A1: (A ⇒ (B ⇒ A)),
A2: ((A ⇒ (B ⇒ C)) ⇒ ((A ⇒ B) ⇒ (A ⇒ C))),
A3: ((¬B ⇒ ¬A) ⇒ ((¬B ⇒ A) ⇒ B)).
(i) Prove that S is sound under S2 classical semantics.
(ii) Define your own S2 semantics under which S is not sound.
(iii) Define your own S4 semantics under which S is sound and other S4
semantics under which S is not sound.
(iv) Define your own Sn semantics such that S is sound for all for 2 ≤
n ≤ m.
(v) Show, by construction a formal proof, that `S (A ⇒ A).

185
186
Chapter 5

Hilbert Proof Systems

Completeness of Classical
Propositional Logic

The Hilbert proof systems are systems based on a language with implication
and contain a Modus Ponens rule as a rule of inference. They are usually called
Hilbert style formalizations. We will call them here Hilbert style proof systems,
or Hilbert systems, for short.
Modus Ponens is probably the oldest of all known rules of inference as it was
already known to the Stoics (3rd century B.C.). It is also considered as the
most natural to our intuitive thinking and the proof systems containing it as
the inference rule play a special role in logic. The Hilbert proof systems put
major emphasis on logical axioms, keeping the rules of inference to minimum,
often in propositional case, admitting only Modus Ponens, as the sole inference
rule.
There are many proof systems that describe classical propositional logic, i.e.
that are complete proof systems with the respect to the classical semantics.
We present here, after Elliott Mendelson’s book Introduction to Mathematical
Logic (1987), a Hilbert proof system for the classical propositional logic and
discuss two ways of proving the Completeness Theorem for it.
Any proof of the Completeness Theorem consists always of two parts. First we
have show that all formulas that have a proof are tautologies. This implication
is also called a Soundness Theorem, or soundness part of the Completeness
Theorem. The second implication says: if a formula is a tautology then it has a
proof. This alone is sometimes called a Completeness Theorem (on assumption
that the system is sound). Traditionally it is called a completeness part of the

187
Completeness Theorem.
The proof of the soundness part is standard. We concentrate here on the com-
pleteness part of the Completeness Theorem and present two proofs of it.
The first proof is based on the one presented in the Mendelson’s book Introduc-
tion to Mathematical Logic (1987). It is is a straightforward constrictive proof
that shows how one can use the assumption that a formula A is a tautology
in order to construct its formal proof. It is hence called a proof - construction
method. It is a beautiful proof
The second proof is non-constrictive. Its strength and importance lies in a fact
that the methods it uses can be applied to the proof of completeness for classical
predicate logic. We will discuss and apply them in Chapter 9.
It proves the completeness part of the Completeness Theorem by proving the
converse implication to it. It shows how one can deduce that a formula A is
not a tautology from the fact that it does not have a proof. It is hence called a
counter-model construction proof.
Both proofs of the Completeness Theorem relay on the Deduction Theorem and
so it is the first theorem we are going to prove.

5.1 Deduction Theorem

We consider first a very simple Hilbert proof system based on a language with
implication as the only connective, with two logical axioms (axiom schemas)
which characterize the implication, and with Modus Ponens as a sole rule of
inference. We call it a Hilbert system H1 and define it as follows.

H1 = ( L{⇒} , F, A1, A2, (M P ) ), (5.1)

where
A1 (A ⇒ (B ⇒ A)),
A2 ((A ⇒ (B ⇒ C)) ⇒ ((A ⇒ B) ⇒ (A ⇒ C))),
(MP) is the following rule of inference, called Modus Ponens

A ; (A ⇒ B)
(M P ) ,
B

and A, B, C ∈ F are any formulas of the propositional language L{⇒} .

Finding formal proofs in this system requires some ingenuity. Let’s construct,
as an example, the formal proof of such a simple formula as A ⇒ A.

188
The formal proof of (A ⇒ A) in H1 is a sequence

B1 , B2 , B3 , B4 , B5 (5.2)

as defined below.
B1 = ((A ⇒ ((A ⇒ A) ⇒ A)) ⇒ ((A ⇒ (A ⇒ A)) ⇒ (A ⇒ A))),
axiom A2 for A = A, B = (A ⇒ A), and C = A
B2 = (A ⇒ ((A ⇒ A) ⇒ A)),
axiom A1 for A = A, B = (A ⇒ A)
B3 = ((A ⇒ (A ⇒ A)) ⇒ (A ⇒ A))),
MP application to B1 and B2
B4 = (A ⇒ (A ⇒ A)),
axiom A1 for A = A, B = A
B5 = (A ⇒ A)
MP application to B3 and B4

We have hence proved the following.

Fact 5.1
For any A ∈ F,
`H1 (A ⇒ A)
and the sequence 5.2 constitutes its formal proof.

It is easy to see that the above proof wasn’t constructed automatically. The
main step in its construction was the choice of a proper form (substitution) of
logical axioms to start with, and to continue the proof with. This choice is far
from obvious for un-experienced prover and impossible for a machine, as the
number of possible substitutions is infinite.
Observe that the systems S1 − S4 from the previous Chapter 4 had inference
rules such that it was possible to ”reverse” their use; to use them in the reverse
manner in order to search for proofs, and we were able to do so in a blind,
fully automatic way. We were able to conduct an argument of the type: if
this formula has a proof the only way to construct it is from such and such
formulas by the means of one of the inference rules, and that formula can be
found automatically. We called proof systems with such property syntactically
decidable and defined them formally as follows.

Definition 5.1 ( Syntactic Decidability)

A proof system S = (L, E, LA, R) for which there is an effective mechanical,
procedure that finds (generates) a formal proof of any E in S, if it exists, is

189
called syntactically semi- decidable. If additionally there is an effective
method of deciding that if a proof of E in S not found, it does not exist, the
system S is called syntactically decidable. Otherwise S is syntactically
undecidable.

We will argue now, that one can’t apply the above argument to the proof search
in Hilbert proof systems as they which contain Modus Ponens as an inference
rule.
A general procedure for searching for proofs in a proof system S can be stated
is as follows. Given an expression B of the system S. If it has a proof, it
must be conclusion of the inference rule. Let’s say it is a rule r. We find its
premisses, with B being the conclusion, i.e. we evaluate r−1 (B). If all premisses
are axioms, the proof is found. Otherwise we repeat the procedure for any non-
axiom premiss.
Search for proof in Hilbert Systems must involve the Modus Ponens. The rule
says: given two formulas A and (A ⇒ B) we can conclude a formula B. Assume
now that we have a formula B and want to find its proof. If it is an axiom, we
have the proof: the formula itself. If it is not an axiom, it had to be obtained
by the application of the Modus Ponens rule, to certain two formulas A and
(A ⇒ B). But there is infinitely many of formulas A and (A ⇒ B). I.e. for any
B, the inverse image of B under the rule M P , M P −1 (B) is countably infinit
Obviously, we have the following.

Fact 5.2
Any Hilbert proof system is not syntactically decidable, in particular, the system
H1 is not syntactically decidable.

Semantic Link 1 System H1 is obviously sound under classical semantics and

is sound under L, H semantics and not sound under K semantics.
We leave the proof of the following theorem (by induction with respect of the
length of the formal proof) as an easy exercise to the reader.

Theorem 5.1 (Soundness of H1 )

For any A ∈ F of H1 , if `H1 A, then |= A.

Semantic Link 2 System H1 is not complete under classical semantics. It

means that not all classical tautologies have a proof in H1 . We have proved
in Chapter 3 that one needs negation and one of the other connectives ∪, ∩, ⇒
to express all classical connectives, and hence all classical tautologies. Our
language contains only implication and one can’t express negation in terms of
implication and hence we can’t provide a proof of any tautology i.e. its logically
equivalent form in our language. It means we have proved the following.

190
Fact 5.3
The proof system H1 is sound, but not complete under the classical semantics.

We have constructed a formal proof (5.2) of (A ⇒ A) in H1 on a base of logical

axioms, as an example of complexity of finding proofs in Hilbert systems.
In order to make the construction of formal proofs easier by the use of previously
proved formulas we use the notions of a formal proof from some hypotheses (and
logical axioms) in any proof system S = (L, E, LA, R) defined as follows in
chapter 4.

Definition 5.2 (Proof from Hypotheses)

Given a proof system S = (L, E, LA, R) and let Γ be any set of expressions of
S, i.e. let Γ ⊆ E.
A proof of an expression E ∈ from the set Γ of expressions is a sequence

E1 , E2 , . . . En

of expressions, such that

E1 ∈ LA ∪ Γ, En = E

and for each i, 1 < i ≤ n, either Ei ∈ LA ∪ Γ or Ei is a direct consequence of

some of the preceding expressions in the sequence E1 , E2 , . . . En by virtue of
one of the rules of inference from R.
We write
Γ `S E

to denote that the expression E has a proof (is provable) from Γ in S and we
write Γ ` E, when the system S is fixed.

When the set of hypothesis Γ is a finite set and Γ = {B1 , B2 , ..., Bn }, then we
write
B1 , B2 , ..., Bn `S E

instead of {B1 , B2 , ..., Bn } `S E. The case when Γ is an empty set i.e. when
Γ = ∅ is a special one. By the definition of a proof of E from Γ, ∅ `S E means
that in the proof of E only logical axioms LA of S were used. We hence write
as we did before
`S E

to denote that E has a proof from the empty set Γ.

191
Definition 5.3 (Consequence in S)
Given a proof system S = (L, F, LA, R) and a set Γ ⊆ F. Any formula A ∈ F
provable from Γ, i.e. such that

Γ `S A

is called a consequence of Γ in S. Formulas from Γ are called hypotheses

or premisses of a proof of A from Γ in S.

The following are simple, but very important properties of the notion of conse-
quence.

Fact 5.4 (Consequence Properties)

Given a proof system S = (L, F, LA, R). For any sets Γ, ∆ ⊆ F the following
holds.
1. If Γ ⊆ ∆ and Γ `S A, then ∆ `S A. monotonicity
2. Γ `S A if and only if
there is a finite subset Γ0 of Γ such that Γ0 `S A. finiteness
3. If ∆ `S A, and, for each B ∈ ∆, Γ `S B, then Γ `S A. transitivity

Proof
The properties follow directly from the definition 5.2 and their proofs are left
to the reader as an exercise.
The monotonicity property represents the fact that if a formula A is provable
from a set Γ of premisses (hypotheses), then if we add still more premisses, A
is still provable. It hence is often stated as follows,

If Γ `S A, then Γ ∪ ∆ `S A, for any set ∆ ⊆ F. (5.3)

The detailed investigation of Tarski general notion of consequence operation,

its relationship with proof systems, and hence with the consequence in S intro-
duced here is included in Chapter 4. Here is an application of the proof from
hypotheses definition 5.2 to the system H1 .

Exercise 5.1
Construct a proof in H1 of a formula (A ⇒ C) from the set of hypotheses
Γ = {(A ⇒ B), (B ⇒ C)}. I.e. show that

(A ⇒ B), (B ⇒ C) `H1 (A ⇒ C).

192
Solution
The required formal proof is a sequence

B1 , B2 , .....B7 (5.4)

such that
B1 = (B ⇒ C),
hypothesis
B2 = (A ⇒ B),
hypothesis
B3 = ((B ⇒ C) ⇒ (A ⇒ (B ⇒ C))),
axiom A1 for A = (B ⇒ C), B = A
B4 = (A ⇒ (B ⇒ C))
B1 , B3 and MP
B5 = ((A ⇒ (B ⇒ C)) ⇒ ((A ⇒ B) ⇒ (A ⇒ C))),
axiom A2
B6 = ((A ⇒ B) ⇒ (A ⇒ C)),
B5 and B4 and MP
B7 = (A ⇒ C).
B2 and B6 and MP

Exercise 5.2
Show, by constructing a formal proof that A `H1 (A ⇒ A).

Solution
The required formal proof is a sequence

B1 , B2 , B3 (5.5)

such that
B1 = A,
hypothesis
B2 = (A ⇒ (A ⇒ A)),
axiom A1 for B = A,
B3 = (A ⇒ A)
B1 , B2 and MP.

We can further simplify the task of constructing formal proofs in H1 by the use
of the following Deduction Theorem.

193
In mathematical arguments, one often assumes a statement A on the assumption
(hypothesis) of some other statement B and then concludes that we have proved
the implication ”if A, then B”. This reasoning is justified by the following
theorem, called a Deduction Theorem. It was first formulated and proved for a
certain Hilbert proof system S for the classical propositional logic by Herbrand
in 1930 in a form stated below.

Theorem 5.2 (Deduction Theorem for S) (Herbrand,1930)

For any formulas A, B of the language of S,

if A `S B, then `S (A ⇒ B).

We are going to prove now that for our system H1 is strong enough to prove
the Herbrand Deduction Theorem for it. In fact we formulate and prove a more
general version of the Theorem 5.2.
To formulate it we introduce the following notation. We write Γ, A `S B for Γ ∪
{A}`S B, and in general we write Γ, A1 , A2 , ..., An `S B for Γ∪{A1 , A2 , ..., An }`S B.
We are now going to prove the following.

Theorem 5.3 (Deduction Theorem for H1 )

For any subset Γ of the set of formulas F of H1 and for any formulas A, B ∈ F,

Γ, A `H1 B if and only if Γ `H1 (A ⇒ B).

In particular,
A `H1 B if and only if `H1 (A ⇒ B).

Proof
We use we use the symbol ` instead of `H1 . for simplicity.
Part 1
We first prove the ”if” part:

If Γ, A ` B then Γ ` (A ⇒ B).

Assume that Γ, A `B, i.e. that we have a formal proof

B1 , B2 , ..., Bn (5.6)

of B from the set of formulas Γ ∪ {A}. In order to prove that Γ ` (A ⇒ B) we

will prove the following a little bit stronger statement S.

S: Γ ` (A ⇒ Bi ) for all Bi (1 ≤ i ≤ n) in the proof (5.6) of B.

194
Hence, in particular case, when i = n, we will obtain that also

Γ ` (A ⇒ B).

The proof of S is conducted by induction on i ( 1 ≤ i ≤ n).

Base Step i = 1.
When i = 1, it means that the formal proof (5.6) contains only one element B1 .
By the definition of the formal proof from Γ ∪ {A}, we have that B1 ∈ LA, or
B1 ∈ Γ, or B1 = A, i.e.

B1 ∈ {A1, A2} ∪ Γ ∪ {A}.

Here we have two cases.

Case 1. B1 ∈ {A1, A2} ∪ Γ.
Observe that (B1 ⇒ (A ⇒ B1 )) is the axiom A1 and by assumption B1 ∈
{A1, A2} ∪ Γ, hence we get the required proof of (A ⇒ B1 ) from Γ by the
following application of the Modus Ponens rule
B1 ; (B1 ⇒ (A ⇒ B1 ))
(M P ) .
(A ⇒ B1 )

Case 2. B1 = A.
When B1 = A, then to prove Γ ` (A ⇒ B) means to prove Γ ` (A ⇒ A). This
holds by the monotonicity of the consequence in H1 (Fact 5.4), and the fact
that we have proved (Fact 5.1) that `(A ⇒ A). The above cases conclude the
proof of the Base case i = 1.
Inductive step
Assume that Γ `(A ⇒ Bk ) for all k < i, we will show that using this fact we
can conclude that also Γ `(A ⇒ Bi ).
Consider a formula Bi in the sequence 5.6. By the definition, Bi ∈ {A1, A2} ∪
Γ ∪ {A} or Bi follows by MP from certain Bj , Bm such that j < m < i. We
have to consider again two cases.
Case 1. Bi ∈ {A1, A2} ∪ Γ ∪ {A}.
The proof of (A ⇒ Bi ) from Γ in this case is obtained from the proof of the
Base Step for i = 1 by replacement B1 by Bi and will be omitted here as a
straightforward repetition.
Case 2. Bi is a conclusion of MP.
If Bi is a conclusion of MP, then we must have two formulas Bj , Bm in the
sequence 5.6 such that j < i, m < i, j 6= m and
Bj ; Bm
(M P ) .
Bi
By the inductive assumption, the formulas Bj , Bm are such that

195
 Γ ` (A ⇒ Bj ) (5.7)
and
Γ ` (A ⇒ Bm ). (5.8)

Moreover, by the definition of the Modus Ponens rule, the formula Bm has to
have a form (Bj ⇒ Bi ), i.e. Bm = (Bj ⇒ Bi ), and the the inductive assumption
(5.8) can be re-written as follows.

Γ ` (A ⇒ (Bj ⇒ Bi )), f or j < i. (5.9)

Observe now that the formula

((A ⇒ (Bj ⇒ Bi )) ⇒ ((A ⇒ Bj ) ⇒ (A ⇒ Bi )))

is a substitution of the axiom schema A2 and hence has a proof in our system.
By the monotonicity of the consequence (5.3), it also has a proof from the set
Γ, i.e.
Γ ` ((A ⇒ (Bj ⇒ Bi )) ⇒ ((A ⇒ Bj ) ⇒ (A ⇒ Bi ))). (5.10)

Applying the rule MP to formulas (5.10) and (5.9,) i.e. performing the following

(A ⇒ (Bj ⇒ Bi )) ; ((A ⇒ (Bj ⇒ Bi )) ⇒ ((A ⇒ Bj ) ⇒ (A ⇒ Bi )))

(M P )
((A ⇒ Bj ) ⇒ (A ⇒ Bi ))

we get that also

Γ `((A ⇒ Bj ) ⇒ (A ⇒ Bi )). (5.11)

Applying again the rule MP to formulas 5.7 and 5.11, i.e. performing the
following

(A ⇒ Bj ) ; ((A ⇒ Bj ) ⇒ (A ⇒ Bi ))
(M P )
(A ⇒ Bi )

we get that

Γ `(A ⇒ Bi )

what ends the proof of the Inductive Step. By the mathematical induction
principle, we hence have proved that Γ `(A ⇒ Bj ) for all i such that 1 ≤ i ≤ n.
In particular it is true for i = n, what means for Bn = B and we have proved
that
Γ ` (A ⇒ B).

196
This ends the proof of the Part 1.

Part 2
The proof of the inverse implication

if Γ ` (A ⇒ B) then Γ, A `B

is straightforward. Assume that Γ ` (A ⇒ B) , hence by the monotonicity of

the consequence (5.3) we have also that Γ, A ` (A ⇒ B). Obviously, Γ, A ` A.
Applying Modus Ponens to the above, we get the proof of B from {Γ, A} i.e.
we have proved that Γ, A ` B. That ends the proof of the deduction theorem
for any set Γ ⊆ F and any formulas A, B ∈ F. The particular case is obtained
from the above by assuming that the set Γ is empty. This ends the proof of the
Deduction Theorem for H1 .

The proof of the following useful lemma provides a good example of multiple
applications of the Deduction Theorem 5.3.

Lemma 5.1
For any A, B, C ∈ F,

(a) (A ⇒ B), (B ⇒ C) `H1 (A ⇒ C),

(b) (A ⇒ (B ⇒ C)) `H1 (B ⇒ (A ⇒ C)).

Proof of (a).
Deduction theorem says:
(A ⇒ B), (B ⇒ C) `H1 (A ⇒ C) if and only if (A ⇒ B), (B ⇒ C), A `H1 C.
We construct a formal proof

B1 , B2 , B3 , B4 , B5

of (A ⇒ B), (B ⇒ C), A `H1 C as follows.

B1 = (A ⇒ B),
hypothesis
B2 = (B ⇒ C),
hypothesis
B3 = A,
hypothesis
B4 = B,
B1 , B3 and MP

197
B5 = C.
B2 , B4 and MP

Thus (A ⇒ B), (B ⇒ C) `H1 (A ⇒ C) by Deduction Theorem.

Proof of (b).
By Deduction Theorem,

(A ⇒ (B ⇒ C)) `H1 (B ⇒ (A ⇒ C)) if and only if (A ⇒ (B ⇒ C)), B `H1 (A ⇒ C).

We construct a formal proof

B1 , B2 , B3 , B4 , B5 , B6 , B7

of (A ⇒ (B ⇒ C)), B `H1 (A ⇒ C). as follows.

B1 = (A ⇒ (B ⇒ C)),
hypothesis
B2 = B,
hypothesis
B3 = ((B ⇒ (A ⇒ B)),
A1 for A = B, B = A
B4 = (A ⇒ B),
B2 , B3 and MP
B5 = ((A ⇒ (B ⇒ C)) ⇒ ((A ⇒ B) ⇒ (A ⇒ C))),
axiomA2
B6 = ((A ⇒ B) ⇒ (A ⇒ C)),
B1 , B5 and MP
B7 = (A ⇒ C).

Thus (A ⇒ (B ⇒ C)) `H1 (B ⇒ (A ⇒ C)) by Deduction Theorem.

Hilbert System H2

The proof system H1 is sound and strong enough to admit the Deduction Theo-
rem, but is not completeas proved in Fact 5.3. We define now a proof system H2
that is complete with respect to classical semantics. The proof of Completeness
Theorem for H2 is to be presented in the next section.

198
H2 is defined as follows.
A ; (A ⇒ B)
H2 = ( L{¬, ⇒} , F, A1, A2, A3, M P ), (5.12)
B
where for any formulas A, B, C ∈ F of L{¬, ⇒} we define
A1 (A ⇒ (B ⇒ A)),
A2 ((A ⇒ (B ⇒ C)) ⇒ ((A ⇒ B) ⇒ (A ⇒ C))),
A3 ((¬B ⇒ ¬A) ⇒ ((¬B ⇒ A) ⇒ B))),

Observation 5.1 Here are some simple, straightforward facts about the proof
system H2 .
1. The language of H2 is obtained from the language of H1 by adding the
connective ¬ to it.
2. H2 is obtained from H1 by adding axiom to it the axiom A3 that characterizes
negation.
3. The use of axioms A1, A2 in the proof of Deduction Theorem 5.3 for H1 is
independent of the negation connective ¬ added to the language of H1 .
4. The proof of Deduction Theorem 5.3 for the system H1 can be repeated as it
is for the system H2 .

Directly from the above Observation 5.1 we get the following.

Theorem 5.4 (Deduction Theorem for H2 )

For any subset Γ of the set of formulas F of H2 and for any formulas A, B ∈ F,

Γ, A `H2 B if and only if Γ `H2 (A ⇒ B).

In particular,
A `H2 B if and only if `H2 (A ⇒ B).

Observe that for the same reason the Lemma 5.1 holds also for H2 . It is a very
i useful lemma for creating proofs in H2 so we re-state it for it here.

Lemma 5.2
For any A, B, C ∈ F,

(a) (A ⇒ B), (B ⇒ C) `H2 (A ⇒ C),

(b) (A ⇒ (B ⇒ C)) `H2 (B ⇒ (A ⇒ C)).

199
We know that the axioms A1, A2 are tautologies and the Modus Ponens rule is
sound. We get by simple verification that |= A3, hence the proof system H2 is
sound, and the following holds.

Theorem 5.5 (Soundness Theorem for H2 )

For every formula A ∈ F, if `H2 A, then |= A.

The soundness theorem proves that the system ”produces” only tautologies. We
show, in the next chapter, that our proof system H2 ”produces” not only tau-
tologies, but that all tautologies are provable in it. This is called a completeness
theorem for classical logic.

Theorem 5.6 (Completeness Theorem for H2 )

For every A ∈ F,
`H2 A if and only if |= A.

The proof of completeness theorem (for a given semantics) is always a main

point in any logic creation. There are many ways (techniques) to prove it,
depending on the proof system, and on the semantics we define for it.
We present in the next sections two proofs of the completeness theorem for
our system H2 . The proofs use very different techniques, hence the reason of
presenting both of them. Both proofs relay heavily on some of the formulas
proved in the next section 5.1.1 and stated in Lemma 5.3.

5.1.1 Formal Proofs

We present here some examples of formal proofs in H2 . There are two reasons for
presenting them. First reason is that all formulas we prove here to be provable
play a crucial role in the proof of Completeness Theorem for H2 , or are needed
to find formal proofs of those needed. The second reason is that they provide a
”training” ground for a reader to learn how to develop formal proofs. For this
second reason we write some proofs in a full detail and we leave some others for
the reader to complete in a way explained in the following example.
We write, were needed ` instead of `H2 .

Example 5.1
We prove that
`H2 (¬¬B ⇒ B) (5.13)
by constructing its formal proof B1 , . . . , B5 , B6 as follows.

200
B1 = ((¬B ⇒ ¬¬B) ⇒ ((¬B ⇒ ¬B) ⇒ B)),
B2 = ((¬B ⇒ ¬B) ⇒ ((¬B ⇒ ¬¬B) ⇒ B)),
B3 = ¬B ⇒ ¬B),
B4 = ((¬B ⇒ ¬¬B) ⇒ B),
B5 = ¬¬B ⇒ (¬B ⇒ ¬¬B)),
B6 = (¬¬B ⇒ B).

Exercise 5.3
Complete the proof B1 , . . . , B5 , B6 of (8.3) by providing comments how each
step of the proof was obtained.

Solution
The proof of (8.3) with comments complementing it is as follows.
B1 = ((¬B ⇒ ¬¬B) ⇒ ((¬B ⇒ ¬B) ⇒ B)),
axiom A3 for A = ¬B, B = B

B2 = ((¬B ⇒ ¬B) ⇒ ((¬B ⇒ ¬¬B) ⇒ B)),

B1 and Lemma 5.2 b for A = (¬B ⇒ ¬¬B), B = (¬B ⇒ ¬B), C = B.
Lemma 5.2 application is: ((¬B ⇒ ¬¬B) ⇒ ((¬B ⇒ ¬B) ⇒ B)) ` ((¬B ⇒
¬B) ⇒ ((¬B ⇒ ¬¬B) ⇒ B))

B3 = (¬B ⇒ ¬B),
Fact 5.1 for A = ¬B

B4 = ((¬B ⇒ ¬¬B) ⇒ B),

B2 , B3 and MP

B5 = (¬¬B ⇒ (¬B ⇒ ¬¬B)),

axiom A1 for A = ¬¬B, B = ¬B

B6 = (¬¬B ⇒ B)
B4 , B5 and Lemma 5.2 a for A = ¬¬B, B = (¬B ⇒ ¬¬B), C = B.
Lemma 5.2 application is:
(¬¬B ⇒ (¬B ⇒ ¬¬B)), ((¬B ⇒ ¬¬B) ⇒ B) ` (¬¬B ⇒ B)

Remark 5.1
Observe that in In step B2 , B3 , B5 , B6 of the proof B1 , . . . , B5 , B6 we call
previously proved results and use their results as a part of our proof. We can

201
insert previously constructed formal proofs of the results we call upon into our
formal proof.

For example we adopt previously constructed proof (5.2) of (A ⇒ A) in H1 to

the proof of (¬B ⇒ ¬B) in H2 by replacing A by ¬B and we insert the proof
of (¬B ⇒ ¬B) after B2 .
The ”old” step B3 becomes now B7 , the ”old” step B4 becomes now B8 , etc.....
Such ”completed” original proof B1 , . . . , B5 , B6 is now B1 , . . . , B9 , B10 looks
now as follows.

B1 = ((¬B ⇒ ¬¬B) ⇒ ((¬B ⇒ ¬B) ⇒ B)), (original B1 ),

axiom A3 for A = ¬B, B = B
B2 = ((¬B ⇒ ¬B) ⇒ ((¬B ⇒ ¬¬B) ⇒ B)), (original B2 )
B1 and Lemma 5.2 b for A = (¬B ⇒ ¬¬B), B = (¬B ⇒ ¬B), C = B,
B3 = ((¬B ⇒ ((¬B ⇒ ¬B) ⇒ ¬B)) ⇒ ((¬B ⇒ (¬B ⇒ ¬B)) ⇒ (¬B ⇒
¬B))), (new proof of B3 inserted )
axiom A2 for A = ¬B, B = (¬B ⇒ ¬B), and C = ¬B
B4 = (¬B ⇒ ((¬B ⇒ ¬B) ⇒ ¬B)),
axiom A1 for A = ¬B, B = (¬B ⇒ ¬B)
B5 = ((¬B ⇒ (¬B ⇒ ¬B)) ⇒ (¬B ⇒ ¬B))),
MP application to B4 and B3
B6 = (¬B ⇒ (¬B ⇒ ¬B)), (end of proof inserted)
axiom A1 for A = ¬B, B = ¬B
B7 = (¬B ⇒ ¬B) (”old” B3 ),
MP application to B5 and B4
B8 = ((¬B ⇒ ¬¬B) ⇒ B), (”old” B4 ) (”old” B4 )
B2 , B3 and MP
B9 = (”old B5 ) (¬¬B ⇒ (¬B ⇒ ¬¬B)), (”old” B5 ) Axiom A1 for
A = ¬¬B, B = ¬B
B10 = (¬¬B ⇒ B). (”old B6 )
B8 , B9 and Lemma 5.2 a for A = ¬¬B, B = (¬B ⇒ ¬¬B), C = B

We repeat our procedure by replacing the step B2 by its formal proof as defined
in the proof of the Lemma 5.1 b, and continue the process for all other steps
which involved application of Lemma 5.2 until we get a full formal proof from
the axioms of H2 only.
Usually we don’t need to do it, but it is important to remember that it always
can be done, if we wished to take time and space to do so.

202
Example 5.2
We prove that
`H2 (B ⇒ ¬¬B) (5.14)
by constructing its formal proof B1 , . . . , B5 as follows.

B1 = ((¬¬¬B ⇒ ¬B) ⇒ ((¬¬¬B ⇒ B) ⇒ ¬¬B)),

B2 = (¬¬¬B ⇒ ¬B),
B3 = ((¬¬¬B ⇒ B) ⇒ ¬¬B),
B4 = (B ⇒ (¬¬¬B ⇒ B)),
B5 = (B ⇒ ¬¬B).

Exercise 5.4
Complete the proof B1 , . . . , B5 of (8.17) by providing comments how each
step of the proof was obtained.

Solution
The proof of (8.17) with comments complementing it is as follows.

B1 = ((¬¬¬B ⇒ ¬B) ⇒ ((¬¬¬B ⇒ B) ⇒ ¬¬B)),

axiom A3 for A = B, B = ¬¬B
B2 = (¬¬¬B ⇒ ¬B),
Example 5.10 for B = ¬B
B3 = ((¬¬¬B ⇒ B) ⇒ ¬¬B),
B1 , B2 and MP, i.e.
(¬¬¬B ⇒ ¬B); ((¬¬¬B ⇒ ¬B) ⇒ ((¬¬¬B ⇒ B) ⇒ ¬¬B))
((¬¬¬B ⇒ B) ⇒ ¬¬B)

B4 = (B ⇒ (¬¬¬B ⇒ B)),
axiom A1 for A = B, B = ¬¬¬B
B5 = (B ⇒ ¬¬B),
B3 , B4 and Lemma 5.2a for A = B, B = (¬¬¬B ⇒ B), C = ¬¬B, i.e.

(B ⇒ (¬¬¬B ⇒ B)), ((¬¬¬B ⇒ B) ⇒ ¬¬B)`H2 (B ⇒ ¬¬B)

Example 5.3
We prove that
`H2 (¬A ⇒ (A ⇒ B)) (5.15)
by constructing its formal proof B1 , . . . , B12 as follows.

203
B1 = ¬A,
B2 = A,
B3 = (A ⇒ (¬B ⇒ A)),
B4 = (¬A ⇒ (¬B ⇒ ¬A)),
B5 = (¬B ⇒ A),
B6 = (¬B ⇒ ¬A),
B7 = ((¬B ⇒ ¬A) ⇒ ((¬B ⇒ A) ⇒ B)),
B8 = ((¬B ⇒ A) ⇒ B),
B9 = B,
B10 = ¬A, A ` B,
B11 = ¬A ` (A ⇒ B),
B12 = (¬A ⇒ (A ⇒ B)).

Example 5.4
We prove that
`H2 ((¬B ⇒ ¬A) ⇒ (A ⇒ B)) (5.16)
by constructing its formal proof B1 , . . . , B7 as follows. Here are consecutive
steps

B1 = (¬B ⇒ ¬A),
B2 = ((¬B ⇒ ¬A) ⇒ ((¬B ⇒ A) ⇒ B)),
B3 = (A ⇒ (¬B ⇒ A)),
B4 = ((¬B ⇒ A) ⇒ B),
B5 = (A ⇒ B),
B6 = (¬B ⇒ ¬A) ` (A ⇒ B),
B7 = ((¬B ⇒ ¬A) ⇒ (A ⇒ B)).

Example 5.5
We prove that
`H2 ((A ⇒ B) ⇒ (¬B ⇒ ¬A)) (5.17)
by constructing its formal proof B1 , . . . , B9 as follows. Here are consecutive
steps

204
B1 = (A ⇒ B),

B2 = (¬¬A ⇒ A),

B3 = (¬¬A ⇒ B),

B4 = (B ⇒ ¬¬B),

B5 = (¬¬A ⇒ ¬¬B),

B6 = ((¬¬A ⇒ ¬¬B) ⇒ (¬B ⇒ ¬A)),

B7 = (¬B ⇒ ¬A),

B8 = (A ⇒ B) ` (¬B ⇒ ¬A),

B9 = ((A ⇒ B) ⇒ (¬B ⇒ ¬A)).

Exercise 5.5
Complete the proof B1 , . . . , B9 of (5.17) by providing comments how each
step of the proof was obtained.

Solution
The proof of (5.17) with comments complementing it is as follows.

B1 = (A ⇒ B),
hypothesis

B2 = (¬¬A ⇒ A),
Example 5.10 for B = A

B3 = (¬¬A ⇒ B),
Lemma 5.2 a for A = ¬¬A, B = A, C = B

B4 = (B ⇒ ¬¬B),
Example 5.11

B5 = (¬¬A ⇒ ¬¬B),
Lemma 5.2 a for A = ¬¬A, B = B, C = ¬¬B

B6 = ((¬¬A ⇒ ¬¬B) ⇒ (¬B ⇒ ¬A)),

Example 5.4 for B = ¬A, A = ¬B

B7 = (¬B ⇒ ¬A),
B5 , B6 and MP

B8 = (A ⇒ B) ` (¬B ⇒ ¬A),
B1 − B7

205
B9 = ((A ⇒ B) ⇒ (¬B ⇒ ¬A)).
Deduction Theorem 5.31

Example 5.6
We prove that
`H2 ((A ⇒ B) ⇒ ((¬A ⇒ B) ⇒ B)) (5.18)
by constructing its formal proof B1 , . . . , B12 as follows. Here are consecutive
steps.

B1 = (A ⇒ B),
B2 = (¬A ⇒ B),
B3 = ((A ⇒ B) ⇒ (¬B ⇒ ¬A)),
B4 = (¬B ⇒ ¬A),
B5 = ((¬A ⇒ B) ⇒ (¬B ⇒ ¬¬A)),
B6 = (¬B ⇒ ¬¬A),,
B7 = ((¬B ⇒ ¬¬A) ⇒ ((¬B ⇒ ¬A) ⇒ B))),
B8 = ((¬B ⇒ ¬A) ⇒ B),
B9 = B,
B10 = (A ⇒ B), (¬A ⇒ B) ` B,
B11 = (A ⇒ B) ` ((¬A ⇒ B) ⇒ B),
B12 = ((A ⇒ B) ⇒ ((¬A ⇒ B) ⇒ B)).

Exercise 5.6
Complete the proof B1 , . . . , B12 of (5.18) by providing comments how each
step of the proof was obtained.

Solution
The proof of (5.18) with comments complementing it is as follows.

B1 = (A ⇒ B),
hypothesis
B2 = (¬A ⇒ B),
hypothesis
B3 = ((A ⇒ B) ⇒ (¬B ⇒ ¬A)),
Example 5.5

206
B4 = (¬B ⇒ ¬A),
B1 , B3 and MP
B5 = ((¬A ⇒ B) ⇒ (¬B ⇒ ¬¬A))
Example 5.5 for A = ¬A, B = B
B6 = (¬B ⇒ ¬¬A),
B2 , B5 and MP
B7 = ((¬B ⇒ ¬¬A) ⇒ ((¬B ⇒ ¬A) ⇒ B))),
axiom A3 for B = B, A = ¬A
B8 = ((¬B ⇒ ¬A) ⇒ B),
B6 , B7 and MP
B9 = B,
B4 , B8 and MP
B10 = (A ⇒ B), (¬A ⇒ B)`H2 B,
B1 − B9
B11 = (A ⇒ B) ` ((¬A ⇒ B) ⇒ B),
Deduction Theorem 5.31
B12 = ((A ⇒ B) ⇒ ((¬A ⇒ B) ⇒ B)). Deduction Theorem 5.31

Example 5.7
We prove that
`H2 ((¬A ⇒ A) ⇒ A) (5.19)
by constructing its formal proof B1 , B2 , B3 as follows. Here are consecutive
steps.

B1 = ((¬A ⇒ ¬A) ⇒ ((¬A ⇒ A) ⇒ A))),

B2 = (¬A ⇒ ¬A),
B3 = ((¬A ⇒ A) ⇒ A)).

Exercise 5.7
Complete the proof B1 , B2 , B3 of (5.19) by providing comments how each
step of the proof was obtained.

Solution
The proof of (5.19) with comments complementing it is as follows.

B1 = ((¬A ⇒ ¬A) ⇒ ((¬A ⇒ A) ⇒ A))),

axiom A3 for B = A

207
B2 = (¬A ⇒ ¬A),
Lemma 5.2 for A = ¬A
B3 = ((¬A ⇒ A) ⇒ A)).
B1 , B2 and MP

The above Examples 5.10 - 5.7 and the Fact 5.1 provide a proof of the following
lemma.

Lemma 5.3
For any formulas A, B, CinF of the system H2 the following holds.

1. `H2 (A ⇒ A);
2. `H2 (¬¬B ⇒ B);
3. `H2 (B ⇒ ¬¬B);
4. `H2 (¬A ⇒ (A ⇒ B));
5. `H2 ((¬B ⇒ ¬A) ⇒ (A ⇒ B));
6. `H2 ((A ⇒ B) ⇒ (¬B ⇒ ¬A));
7. `H2 (A ⇒ (¬B ⇒ (¬(A ⇒ B)));
8. `H2 ((A ⇒ B) ⇒ ((¬A ⇒ B) ⇒ B));
9. `H2 ((¬A ⇒ A) ⇒ A.

The set of provable formulas from the above Lemma 5.3 includes a set of provable
formulas needed, with H2 axioms to execute two proofs of the Completeness
Theorem 5.6 for H2 . These two proofs represent two very different methods of
proving Completeness Theorem.

5.2 Completeness Theorem: Proof One

The Proof One of the Completeness Theorem 5.6 for H2 . presented here is
similar in its structure to the proof of the Deduction Theorem 5.3 and is due
to Kalmar, 1935. It is, as Deduction Theorem was, a constructive proof. It
means it defines a method how one can use the assumption that a formula A is
a tautology in order to construct its formal proof. We hence call it a proof
construction method. It relies heavily on the Deduction Theorem.

Proof One, the first proof of the Completeness Theorem 5.6 presented here is
very elegant and simple, but is applicable only to the classical propositional

208
logic. Methods it uses are specific to a propositional language L{¬, ⇒} and
the proof system H2 . Nevertheless, it can be adopted and extended to other
classical propositional languages L{¬, ∪, ⇒} , L{¬, ∩, ∪,⇒} , L{¬, ∩, ∪,⇒,⇔} , and
proof systems based on them. We do so by adding appropriate new logical
axioms to the logical axioms of H2 (section 5.2.1). Such obtained proof systems
are called extentions of the system H2 . It means that one can think about the
system H2 , i.e. an axiomatization given by set {A1, A2, A3} of logical axioms
of H2 , and its language L{¬, ⇒} as in a sense, a ”minimal one” for classical
propositional logic and its languages that contain implication.
Proof One, i.e. the methods of carrying it, can’t be extended to the classical
predicate logic, not to mention variety of non-classical logics. Hence we present,
in the next section5.3 another, more general proof, called Proof Two, that can.

We have already proved the Soundness Theorem 5.5 for H2 , so in order to prove
the Completeness Theorem 5.6 we need to prove only the completeness part of
the completeness theorem, i.e. the following implication.
For any formula A of H2 ,

if |= A, then `S A. (5.20)

In order to prove (5.20), i.e. to prove that any tautology has a formal proof in
H2 , we need first to present one definition and prove one lemma stated below.
We write ` A instead of `H2 A, as the system H2 is fixed.

Definition 5.4
Let A be a formula and b1 , b2 , ..., bn be all propositional variables that occur in A.
Let v be variable assignment v : V AR −→ {T, F }. We define, for A, b1 , b2 , ..., bn
and v a corresponding formulas A0 , B1 , B2 , ..., Bn as follows:

if v ∗ (A) = T

A
A0 =
¬A if v ∗ (A) = F

bi if v(bi ) = T
Bi =
¬bi if v(bi ) = F

for i = 1, 2, ..., n.

Example 5.8
Let A be a formula
(a ⇒ ¬b) (5.21)
and let v be such that
v(a) = T, v(b) = F. (5.22)

209
In this case b1 = a, b2 = b, and v ∗ (A) = v ∗ (a ⇒ ¬b) = v(a) ⇒ ¬v(b)=
T ⇒ ¬F = T. The corresponding A0 , B1 , B2 are: A0 = A (as v ∗ (A) = T ),
B1 = a (as v(a) = T ), B2 = ¬b (as v(b) = F ).

Here is a simple exercise.

Exercise 5.8
Let A be a formula ((¬a ⇒ ¬b) ⇒ c) and let v be such that v(a) = T, v(b) =
F, v(c) = F.
Evaluate A0 , B1 , ...Bn as defined by the definition 5.4.

Solution
In this case n = 3 and b1 = a, b2 = b, b3 = c, and v ∗ (A) = v ∗ ((¬a ⇒ ¬b) ⇒ c)
=((¬v(a) ⇒ ¬v(b)) ⇒ v(c)) = ((¬T ⇒ ¬F ) ⇒ F ) = (T ⇒ F ) = F . The
corresponding A0 , B1 , B2 , B2 are: A0 = ¬A = ¬((¬a ⇒ ¬b) ⇒ c) (as v ∗ (A) =
F ), B1 = a (as v(a) = T ), B2 = ¬b (as v(b) = F ). B3 = ¬c (as v(c) = F ).
The lemma stated below describes a method of transforming a semantic notion
of a tautology into a syntactic notion of provability. It defines, for any formula
A and a variable assignment v a corresponding deducibility relation `.

Lemma 5.4 (Main Lemma)

0
For any formula A and a variable assignment v, if A , B1 , B2 , ..., Bn are
corresponding formulas defined by 5.4, then

B1 , B2 , ..., Bn ` A0 . (5.23)

Example 5.9 Let A, v be as defined by (5.21) and (5.22), respectively.

1. The Lemma 5.4 asserts that a, ¬b ` (a ⇒ ¬b).
Let A, v be as defined in Exercise 5.8.
2. The Lemma 5.4 asserts that a, ¬b, ¬c ` ¬((¬a ⇒ ¬b) ⇒ c).

Proof of the Main Lemma

The Main Lemma 5.4 states: for any formula A and a variable assignment v, if
0
A , B1 , B2 , ..., Bn are corresponding formulas defined by Definition 5.4, then

B1 , B2 , ..., Bn ` A0 .

Proof We carry the proof by mathematical induction on the degree of A i.e. a

number n of logical connectives in A.

210
Case: n = 0
In the case that n = 0 A is atomic and so consists of a single propositional
variable, say a. We have to cases to consider, v ∗ (A) = T or v ∗ (A) = F .
Clearly, if v ∗ (A) = T then we A0 = A = a, B1 = a, and a ` a holds by the
Deduction Theorem and 11.15. I.e. ` (a ⇒ a) holds by ??). Applying
the the Deduction Theorem we get a ` a.
If v ∗ (A) = F then we A0 = ¬A = ¬a, B1 = ¬a, and ` (¬a ⇒ ¬a) holds
by Lemma 5.3. Applying the the Deduction Theorem we get ¬a ` ¬a. So
the lemma holds for the case n = 0.

Now assume that the lemma holds for any A with j < n logical connectives
(any A of the degree j < n). The goal is to prove that it holds for A with the
degree n.
There are several sub-cases to deal with.

Case: A is ¬A1
If A is of the form ¬A1 then A1 has less then n connectives and by the
0
inductive assumption we have the formulas A1 , B1 , B2 , ..., Bn corre-
sponding to the A1 and the propositional variables b1 , b2 , ..., bn in A1 , as
defined by the definition 5.4, such that
0
B1 , B2 , ..., Bn ` A1 . (5.24)

Observe, that the formulas A and ¬A1 have the same propositional vari-
ables, so the corresponding formulas B1 , B2 , ..., Bn are the same for
both of them. We are going to show that the inductive assumption (5.24)
allows us to prove that the lemma holds for A, ie. that
0
B1 , B2 , ..., Bn ` A .

There two cases to consider.

Case: v ∗ (A1 ) = T
0
If v ∗ (A1 ) = T then by definition 5.4 A1 = A1 and by the inductive
assumption (5.24)
B1 , B2 , ..., Bn ` A1 . (5.25)
0
In this case v ∗ (A) = v ∗ (¬A1 ) = ¬v ∗ (T ) = F and so A = ¬A =
¬¬A1 . We have by Lemma 5.3, ` (A1 ⇒ ¬¬A1 ), By the monotonic-
ity, B1 , B2 , ..., Bn ` (A1 ⇒ ¬¬A1 ). By inductive assumption (5.25)
and Modus Ponens we have that also B1 , B2 , ..., Bn ` ¬¬A1 , that is
0
B1 , B2 , ..., Bn ` ¬A, that is B1 , B2 , ..., Bn ` A .
Case: v ∗ (A1 ) = F
0 0
If v ∗ (A1 ) = F then A1 = ¬A1 and v ∗ (A) = T so A = A. There-
fore the inductive assumption (5.24) B1 , B2 , ..., Bn ` ¬A1 , that is
0
B1 , B2 , ..., Bn ` A .

211
Case: A is (A1 ⇒ A2 )
If A is of the form (A1 ⇒ A2 ) then A1 and A2 have less than n connectives.
A = A(b1 , ... bn ) so there are some subsequences c1 , ..., ck and d1 , ...dm ,
for k, m ≤ n, of the sequence b1 , ..., bn such that A1 = A1 (c1 , ..., ck ) and
A2 = A(d1 , ...dm ). A1 and A2 have less than n connectives and so by
the inductive assumption we have appropriate formulas C1 , ..., Ck and
0 0
D1 , ...Dm such that C1 , C2 , . . . , Ck ` A1 and D1 , D2 , . . . , Dm ` A2 .
The formulas C1 , C2 , ..., Ck and D1 , D2 , ..., Dm are subsequences of for-
mulas B1 , B2 , ..., Bn corresponding to the propositional variables in A.
0
Hence by monotonicity we have also that have B1 , B2 , ..., Bn ` A1 and
0
B1 , B2 , ..., Bn ` A2 , where B1 , B2 , ..., Bn are formulas corresponding to
the propositional variables in A.
Now we have the following sub-cases to consider.

Case: v ∗ (A1 ) = v ∗ (A2 ) = T

0 0
If v ∗ (A1 ) = T then A1 is A1 and if v ∗ (A2 ) = T then A2 is A2 . We
0
also have v ∗ (A1 ⇒ A2 ) = T and so A is (A1 ⇒ A2 ). By the above
and the inductive assumption, therefore, B1 , B2 , ..., Bn ` A2 and by
Lemma 5.3, i.e. ` (A2 ⇒ (A1 ⇒ A2 )). By monotonicity and Modus
0
Ponens, that B1 , B2 , ..., Bn ` (A1 ⇒ A2 ), that is B1 , B2 , ..., Bn ` A .
Case: v ∗ (A1 ) = T, v ∗ (A2 ) = F
0 0
If v ∗ (A1 ) = T then A1 is A1 and if v ∗ (A2 ) = F then A2 is
0
¬A2 . Also we have in this case v ∗ (A1 ⇒ A2 ) = F and so A is
¬(A1 ⇒ A2 ). By the above and the inductive assumption, there-
fore, B1 , B2 , ..., Bn ` ¬A2 . By Lemma 5.3, ` (A1 ⇒ (¬A2 ⇒
¬(A1 ⇒ A2 ))). By monotonicity and Modus Ponens twice, we have
0
that B1 , B2 , ..., Bn ` ¬(A1 ⇒ A2 ), that is B1 , B2 , ..., Bn ` A .
Case: v ∗ (A1 ) = F
0
If v ∗ (A1 ) = F then A1 is ¬A1 and, whatever value v gives A2 ,
0
we have v ∗ (A1 ⇒ A2 ) = T and so A is (A1 ⇒ A2 ). Therefore,
B1 , B2 , ..., Bn ` ¬A1 and by Lemma 5.3, ` (¬A1 ⇒ (A1 ⇒ A2 )).
By monotonicity and Modus Ponens we get that B1 , B2 , ..., Bn `
0
(A1 ⇒ A2 ), that is B1 , B2 , ..., Bn ` A .

With that we have covered all cases and, by induction on n, the proof of the
lemma is complete.

Proof of the Completeness Theorem

Now we use the Main Lemma 5.4 to prove the completeness part of the Com-
pleteness Theorem 5.6, i.e. to prove the implication (5.20):

For any formula A ∈ F, if |= A, then ` A.

212
Proof Assume that |= A. Let b1 , b2 , ..., bn be all propositional variables that
occur in A, i.e. A = A(b1 , b2 , ..., bn ).
Let v : V AR → {T, F } be any variable assignment, and

vA : {b1 , b2 , ...., bn } → {T, F } (5.26)

its restriction to the formula A, i.e. vA = v|{b1 , b2 , ...., bn }. Let

VA = {vA : vA : {b1 , b2 , ...., bn } → {T, F }} (5.27)

By the Main Lemma 5.4 and the assumption that |= A any v ∈ VA defines
formulas B1 , B2 , ..., Bn such that

B1 , B2 , ..., Bn ` A. (5.28)

The proof is based on a method of using all v ∈ VA to define a process of

elimination of all hypothesis B1 , B2 , ..., Bn in (5.28) to finally construct the
proof of A in H2 i.e. to prove that ` A.

Step 1: elimination of Bn .
Observe that by definition 5.4, each Bi is bi or ¬bi depending on the
choice of v ∈ VA . In particular Bn = bn or Bn = ¬bn . We choose two
truth assignments v1 6= v2 ∈ VA such that

v1 |{b1 , ..., bn−1 } = v2 |{b1 , ..., bn−1 } (5.29)

and v1 (bn ) = T and v2 (bn ) = F .

Case 1: v1 (bn ) = T , by definition 5.4 Bn = bn . By the property (5.29),

assumption that |= A, and the Main Lemma 5.4 applied to v1

B1 , B2 , ..., Bn−1 , bn ` A.

By Deduction Theorem 5.3 we have that

B1 , B2 , ..., Bn−1 ` (bn ⇒ A). (5.30)

Case 2: v2 (bn ) = F hence by definition 5.4 Bn = ¬bn . By the property (5.29),

assumption that |= A, and the Main Lemma 5.4 applied to v2

B1 , B2 , ...Bn−1 , ¬bn ` A.

By the Deduction Theorem 5.3 we have that

B1 , B2 , ..., Bn−1 ` (¬bn ⇒ A). (5.31)

213
By Lemma 5.3 of the formula ` ((A ⇒ B) ⇒ ((¬A ⇒ B) ⇒ B)). Hence for for
A = bn , B = A we have that

` ((bn ⇒ A) ⇒ ((¬bn ⇒ A) ⇒ A)).

By monotonicity we have that

B1 , B2 , ..., Bn−1 ` ((bn ⇒ A) ⇒ ((¬bn ⇒ A) ⇒ A)). (5.32)

Applying Modus Ponens twice to the above property (5.32) and properties
(5.30), (5.31) we get that

B1 , B2 , ..., Bn−1 ` A. (5.33)

We have eliminated Bn .

Step 2: elimination of Bn−1 from (5.33). We repeat the Step 1.

As before we have 2 cases to consider: Bn−1 = bn−1 or Bn−1 = ¬bn−1 .
We choose two truth assignments w1 6= w2 ∈ VA such that

w1 |{b1 , ..., bn−2 } = w2 |{b1 , ..., bn−2 } = v1 |{b1 , ..., bn−2 } = v2 |{b1 , ..., bn−2 }
(5.34)
and w1 (bn−1 ) = T and w2 (bn−1 ) = F .
As before we apply Main Lemma, Deduction Theorem, monotonicity,
proper substitutions of the formula ((A ⇒ B) ⇒ ((¬A ⇒ B) ⇒ B)),
and Modus Ponens twice and eliminate Bn−1 just as we eliminated Bn .

After n steps, we finally obtain that

` A.

This ends the proof of Completeness Theorem.

Observe that our proof of the fact that ` A is a constructive one. Moreover, we
have used in it only Main Lemma 5.4 and Deduction Theorem 5.3, and both of
them have fully constructive proofs. So we can always reconstruct all steps in
proofs which use the Main Lemma 5.4and Deduction Theorem 5.3 back to the
original axioms of H2 . The same applies to the proofs that use the formulas
proved in H2 that are stated in Lemma 5.3.

It means that for any A ∈ F, such that |= A, the set VA of all v restricted to
A provides us a method of a construction of the formal proof of A in H2 from
its axioms A1, A2, A3 only. .

214
5.2.1 Examples

Example 5.10
As an example of how the Proof One of the Completeness Theorem works, we
consider a following tautology

|= (a ⇒ (¬a ⇒ b))

and show how to construct its proof, i.e. to show that

` (a ⇒ (¬a ⇒ b)).

We apply the Main Lemma 5.4 to all possible variable assignments v ∈ VA . We

have 4 variable assignments to consider.

Case 1: v(a) = T, v(b) = T .

0
In this case B1 = a, B2 = b and, as in all cases, A = A and by the Main
Lemma 5.4
a, b ` (a ⇒ (¬a ⇒ b)).

Case 2: v(a) = T, v(b) = F .

In this case B1 = a, B2 = ¬b and by the Main Lemma 5.4

a, ¬b ` (a ⇒ (¬a ⇒ b)).

Case 3: v(a) = F, v(b) = T .

In this case B1 = ¬a, B2 = b and by the Main Lemma 5.4

¬a, b ` (a ⇒ (¬a ⇒ b)).

Case 4: v(a) = F, v(b) = F .

In this case B1 = ¬a, B2 = ¬b and by the lemma 5.4

¬a, ¬b ` (a ⇒ (¬a ⇒ b)).

Applying the Deduction Theorem 5.3 to the cases above we have that

D1 (Cases 1 and 2)
a ` (b ⇒ (a ⇒ (¬a ⇒ b))),
a ` (¬b ⇒ (a ⇒ (¬a ⇒ b))),

D2 (Cases 2 and 3)
¬a ` (b ⇒ (a ⇒ (¬a ⇒ b))),
¬a ` (¬b ⇒ (a ⇒ (¬a ⇒ b))).

215
By the monotonicity and the proper substitution of formula
((A ⇒ B) ⇒ ((¬A ⇒ B) ⇒ B))
the provable by Lemma 5.3, we have that
a ` ((b ⇒ (a ⇒ (¬a ⇒ b))) ⇒ ((¬b ⇒ (a ⇒ (¬a ⇒ b))) ⇒ (a ⇒ (¬a ⇒ b))),
¬a ` ((b ⇒ (a ⇒ (¬a ⇒ b))) ⇒ ((¬b ⇒ (a ⇒ (¬a ⇒ b))) ⇒ (a ⇒ (¬a ⇒ b))).

Applying Modus Ponens twice to D1, D2 and these above, respectively, gives
us
a ` (a ⇒ (¬a ⇒ b)) and
¬a ` (a ⇒ (¬a ⇒ b)).
Applying the Deduction Theorem 5.3 to the above we obtain

D3 ` (a ⇒ (a ⇒ (¬a ⇒ b))),
D4 ` (¬a ⇒ (a ⇒ (¬a ⇒ b))).

We form now an appropriate form of the formula

((A ⇒ B) ⇒ ((¬A ⇒ B) ⇒ B)), (5.35)
provable by the Lemma 5.3. The appropriate form is
⇒ ((¬a ⇒ (a ⇒ (¬a ⇒ b))) ⇒ (a ⇒ (¬a ⇒ b)))). (5.36)
We apply Modus Ponens twice to D3 and D4 and (5.58) and get finally the
proof of (a ⇒ (¬a ⇒ b)), i.e. we have proved that
` (a ⇒ (¬a ⇒ b)).

Example 5.11
The Proof One of Completeness Theorem defines a method of efficiently com-
bining v ∈ VA as defined in (5.27), while constructing the proof of A. Let’s
consider the following tautology A = A(a, b, c)
((¬a ⇒ b) ⇒ (¬(¬a ⇒ b) ⇒ c).
We present bellow all steps of Proof One as applied to A.

By the Main Lemma 5.4 and the assumption that |= A(a, b, c) any v ∈ VA
defines formulas Ba , Bb , Bc such that

Ba , Bb , Bc ` A. (5.37)

The proof is based on a method of using all v ∈ VA (there is 16 of them) to

define a process of elimination of all hypothesis Ba , Bb , Bc in (5.37) to construct
the proof of A in H2 i.e. to prove that ` A.

216
Step 1: elimination of Bc .
Observe that by definition 5.4, Bc is c or ¬c depending on the choice of
v ∈ VA . We choose two truth assignments v1 =6 v2 ∈ VA such that

v1 |{a, b} = v2 |{a, b} (5.38)

and v1 (c) = T and v2 (c) = F .

Case 1: v1 (c) = T , by definition 5.4 Bc = c. By the property (5.38), assump-
tion that |= A, and the Main Lemma 5.4 applied to v1

Ba , Bb , c ` A.

By Deduction Theorem 5.3 we have that

Ba , Bb ` (c ⇒ A). (5.39)

Case 2: v2 (c) = F hence by definition 5.4 Bc = ¬c. By the property (5.38),

assumption that |= A, and the Main Lemma 5.4 applied to v2

Ba , Bb , ¬c ` A.

By the Deduction Theorem 5.3 we have that

Ba , Bb ` (¬c ⇒ A). (5.40)

By Lemma 5.3, i.e. provability of the formula (5.35) for A = c, B = A we have

that
` ((c ⇒ A) ⇒ ((¬c ⇒ A) ⇒ A)).
By monotonicity we have that

Ba , Bb ` ((c ⇒ A) ⇒ ((¬c ⇒ A) ⇒ A)). (5.41)

Applying Modus Ponens twice to the above property (5.41) and properties
(5.39), (5.40) we get that
Ba , Bb ` A. (5.42)
and hence we have eliminated Bc .

Step 2: elimination of Bb from (5.42). We repeat the Step 1.

As before we have 2 cases to consider: Bb = b or Bb = ¬b. We choose
from VA two truth assignments w1 6= w2 ∈ VA such that

w1 |{a} = w2 |{a} = v1 |{a} = v2 |{a} (5.43)

and w1 (b) = T and w2 (b) = F .

217
Case 1: w1 (b) = T , by definition 5.4 Bb = b. By the property (5.43), assump-
tion that |= A, and the Main Lemma 5.4 applied to w1

Ba , b ` A.

By Deduction Theorem 5.3 we have that

Ba ` (b ⇒ A). (5.44)

Case 2: w2 (c) = F hence by definition 5.4 Bb = ¬b. By the property (5.3),

assumption that |= A, and the Main Lemma 5.4 applied to w2

Ba , ¬b ` A.

By the Deduction Theorem 5.3 we have that

Ba ` (¬b ⇒ A). (5.45)

By Lemma 5.3, i.e. provability of the formula (5.35) for A = b, B = A we have

that
` ((b ⇒ A) ⇒ ((¬b ⇒ A) ⇒ A)).
By monotonicity we have that

Ba ` ((b ⇒ A) ⇒ ((¬b ⇒ A) ⇒ A)). (5.46)

Applying Modus Ponens twice to the above property (5.46) and properties
(5.44), (5.45) we get that
Ba ` A. (5.47)
and hence we have eliminated Bb .

Step 3: elimination of Ba from (5.47). We repeat the Step 2.

As before we have 2 cases to consider: Ba = a or Ba = ¬a. We choose
from VA two truth assignments g1 6= g2 ∈ VA such that

g1 (a) = T and g2 (a) = F. (5.48)

Case 1: g1 (a) = T , by definition 5.4 Ba = a. By the property (5.48), assump-

tion that |= A, and the Main Lemma 5.4 applied to g1

a ` A.

By Deduction Theorem 5.3 we have that

` (a ⇒ A). (5.49)

218
Case 2: g2 (a) = F hence by definition 5.4 Ba = ¬a. By the property (5.48),
assumption that |= A, and the Main Lemma 5.4 applied to g2

¬a ` A.

By the Deduction Theorem 5.3 we have that

` (¬a ⇒ A). (5.50)

By Lemma 5.3, i.e. provability of the formula (5.35) for A = a, B = A we have

that
` ((a ⇒ A) ⇒ ((¬a ⇒ A) ⇒ A)). (5.51)

Applying Modus Ponens twice to the above property (5.51) and properties
(5.49), (5.50) we get that
` A. (5.52)
and hence we have eliminated Ba , Bb and Bc and constructed the proof of A.

5.3 Completeness Theorem: Proof Two

The Proof Two is much more complicated then the Proof One. Its strength and
importance lies in a fact that the methods it uses can be applied in an extended
version to the proof of completeness for classical predicate logic and even many
of non-classical propositional and predicate logics. The main point of the proof
is a presentation of a general, non- constructive method for proving existence of
a counter-model for any non-provable A. The generality of the method makes
it possible to adopt it for other cases of predicate and some non-classical logics.
We call it a a counter-model existence method.

We prove now the completeness part of the Completeness Theorem 5.6 for H2
by proving that the opposite implication:

if 6` A, then 6|= A (5.53)

to the implication (5.20):

if |= A, then ` A

holds hat for all A ∈ F.

We will show now how one can define of a counter-model for A from the fact
that A is not provable. This means that we deduce that a formula A is not

219
a tautology from the fact that it does not have a proof. We hence call it a a
counter-model existence method.
The definition of the counter-model for any non-provable A is much more general
(and less constructive) then in the case of the Proof One in section 5.2. It can
be generalized to the case of predicate logic, and many of non-classical logics;
propositional and predicate. It is hence a much more general method then the
first one and this is the reason we present it here.
We remind that 6|= A means that there is a truth assignment v : V AR −→
{T, F }, such that v ∗ (A) 6= T , i.e. in classical semantics, such that that v ∗ (A) =
F . Such v is called a counter-model for A, hence the proof provides a counter-
model construction method.
Since we assume in (8.16) that A does not have a proof in H2 (6` A) the method
uses this information in order to show that A is not a tautology, i.e. to define v
such that v ∗ (A) = F . We also have to prove that all steps in that method are
correct. This is done in the following steps.

Step 1: Definition of ∆∗
We use the information 6` A to define a special set ∆∗ ⊆ F, such that
¬A ∈ ∆∗ .
Step 2: Counter - model definition
We define the truth assignment v : V AR −→ {T, F } as follows:

if ∆∗ ` a

T
v(a) =
F if ∆∗ ` ¬a.

Step 3: Prove that v is a counter-model

We first prove a more general property, namely we prove that the set ∆∗
and v defined in the steps 1 and 2, respectively, are such that for every
formula B ∈ F,

if ∆∗ ` B

T
v ∗ (B) =
F if ∆∗ ` ¬B.

Then we use the Step 1 (definition of ∆∗ ) to prove that v ∗ (A) = F .

The definition and the properties of the set ∆∗ , and hence the Step 1, are the
most essential for the proof. The other steps have mainly technical character.
The main notions involved in the Step 1 (definition of ∆∗ ) are: consistent set,
complete set and a consistent complete extension of a set. We are going now to
introduce them and to prove some essential facts about them.

Consistent and Inconsistent Sets

220
There exist two definitions of consistency; semantical and syntactical. The
semantical one uses definition the notion of a model and says, in plain English:

a set of formulas is consistent if it has a model.

The syntactical one uses the notion of provability and says:

a set of formulas is consistent if one can’t prove a contradiction from it.

In our Proof Two of the Completeness Theorem we use assumption that a given
formula A does not have a proof to deduce that A is not a tautology. We hence
use the following syntactical definition of consistency.

Consistent set
We say that a set ∆ ⊆ F of formulas is consistent if and only if there
is no a formula A ∈ F such that

∆ ` A and ∆ ` ¬A. (5.54)

Inconsistent set
A set ∆ ⊆ F is inconsistent if and only if there is a formula A ∈ F such
that ∆ ` A and ∆ ` ¬A.

The notion of consistency, as defined above, is characterized by the following

lemma.

Lemma 5.5 (Consistency Condition)

For every set ∆ ⊆ F of formulas, the following conditions are equivalent:

(i) ∆ is consistent,

(ii) there is a formula A ∈ F such that ∆ 6` A.

Proof The implications: (i) implies (ii) and vice-versa are proved by showing
the corresponding opposite implications. I.e. to establish the equivalence of (i)
and (ii), we first show that not (ii) implies not (i), and then that not (i)
implies not (ii).

Case 1
Assume that not (ii). It means that for all formulas A ∈ F we have that
∆ ` A. In particular it is true for a certain A = B and A = ¬B and
hence proves that ∆ is inconsistent, i.e. not (i) holds.

221
Case 2
Assume that not (i), i.e that ∆ is inconsistent. Then there is a formula
A such that ∆ ` A and ∆ ` ¬A. Let B be any formula. Since
(¬A ⇒ (A ⇒ B)) is provable in H2 by Lemma 5.3, hence by applying
Modus Ponens twice and by detaching from it ¬A first, and A next, we
obtain a formal proof of B from the set ∆, so that ∆ ` B for any formula
B. Thus not (ii).

The inconsistent sets are hence characterized by the following fact.

Lemma 5.6 (Inconsistency Condition)

For every set ∆ ⊆ F of formulas, the following conditions are equivalent:

(i) ∆ is inconsistent,
(ii) for all formulas A ∈ F, ∆ ` A.

We remind here the property of the finiteness of the consequence operation.

Lemma 5.7
For every set ∆ of formulas and for every formula A ∈ F, ∆ ` A if and only
if there is a finite subset ∆0 ⊆ ∆ such that ∆0 ` A.

Proof
If ∆0 ` A for a certain ∆0 ⊆ ∆, then by the monotonicity of the consequence,
also ∆ ` A. Assume now that ∆ ` A and let A1 , A2 , ..., An be a formal
proof of A from ∆. Let ∆0 = {A1 , A2 , ..., An } ∩ ∆. Obviously, ∆0 is finite and
A1 , A2 , ..., An is a formal proof of A from ∆0 .

The following theorem is a simply corollary of the above Lemma 5.7.

Theorem 5.7 (Finite Inconsistency)

If a set ∆ is inconsistent, then there is a finite subset ∆0 ⊆ ∆ which is inconsis-
tent. It follows therefore from that if every finite subset of a set ∆ is consistent,
then the set ∆ is also consistent.

Proof
If ∆ is inconsistent, then for some formula A, ∆ ` A and ∆ ` ¬A. By above
Lemma 5.7, there are finite subsets ∆1 and ∆2 of ∆ such that ∆1 ` A and
∆2 ` ¬A. By monotonicity, the union ∆1 ∪ ∆2 is a finite subset of ∆, such
that ∆1 ∪ ∆2 ` A and ∆1 ∪ ∆2 ` ¬A. Hence ∆1 ∪ ∆2 is a finite inconsistent
subset of ∆. The second implication is the opposite to the one just proved and

222
hence also holds.

The following lemma links the notion of non-provability and consistency. It will
be used as an important step in our proof of the Completeness Theorem.

Lemma 5.8
For any formula A ∈ F, if 6` A, then the set {¬A} is consistent.

Proof
If {¬A} is inconsistent, then by the Inconsistency Condition 5.6 we have {¬A} `
A. This and the Deduction Theorem 5.3 imply ` (¬A ⇒ A). Applying the
Modus Ponens rule to ` (¬A ⇒ A) a formula ((¬A ⇒ A) ⇒ A), provable by
LemmaH2lemma, we get that ` A, contrary to the assumption of the lemma.
Complete and Incomplete Sets
Another important notion, is that of a complete set of formulas. Complete sets,
as defined here are sometimes called maximal, but we use the first name for
them. They are defined as follows.

Complete set
A set ∆ of formulas is called complete if for every formula A ∈ F,

∆ ` A or ∆ ` ¬A. (5.55)

The complete sets are characterized by the following fact.

Lemma 5.9 (Complete set condition)

For every set ∆ ⊆ F of formulas, the following conditions are equivalent:

(i) ∆ is complete,
(ii) for every formula A ∈ F, if ∆ 6` A, then the set ∆ ∪ {A} is inconsistent.

Proof
We consider two cases. We show that (i) implies (ii) and vice-versa, that (ii)
also implies (i).

Case 1
Assume that (i) and that for every formula A ∈ F, ∆ 6` A, we have to
show that in this case ∆ ∪ {A} is inconsistent. But if ∆ 6` A, then from
the definition of complete set and assumption that ∆ is complete set, we
get that ∆ ` ¬A. By the monotonicity of the consequence we have that

223
 ∆ ∪ {A} ` ¬A as well. Since, by formula 11.15 we have ` (A ⇒ A), by
monotonicity ∆ ` (A ⇒ A) and by Deduction Theorem ∆ ∪ {A} ` A.
This proves that ∆ ∪ {A} is inconsistent. Hence (ii) holds.

Case 2
Assume that (ii). Let A be any formula. We want to show that the
condition: ∆ ` A or ∆ ` ¬A is satisfied. If ∆ ` ¬A, then the
condition is obviously satisfied.
If, on other hand, ∆ 6` ¬A, then we are going to show now that it must
be , under the assumption of (ii), that ∆ ` A, i.e. that (i) holds.
Assume that ∆ 6` ¬A, then by (ii), the set ∆ ∪ {¬A} is inconsistent.
It means, by the Consistency Condition 5.5, that ∆ ∪ {¬A} ` A.
By the Deduction Theorem 5.3, this implies that ∆ ` (¬A ⇒ A).
Since ((¬A ⇒ A) ⇒ A) is provable in H2 (Lemma 5.3), by monotonicity
∆ ` ((¬A ⇒ A) ⇒ A). Detaching (¬A ⇒ A), we obtain that ∆ ` A,
what ends the proof that (i) holds.

Incomplete set
A set ∆ of formulas is called incomplete if it is not complete, i.e. if there
exists a formula A ∈ F such that

∆ 6` A and ∆ 6` ¬A. (5.56)

We get as a direct consequence of the lemma 5.9 the following characterization

of incomplete sets.

Lemma 5.10 (Incomplete Set Condition)

For every set ∆ ⊆ F of formulas, the following conditions are equivalent:

(i) ∆ is incomplete,

(ii) there is formula A ∈ F such that ∆ 6` A and the set ∆ ∪ {A} is consistent.

Main Lemma: Complete Consistent Extension

Now we are going to prove a lemma that is essential to the construction of

the special set ∆∗ mentioned in the Step 1 of the proof of the Completeness
Theorem, and hence to the proof of the theorem itself. Let’s first introduce one
more notion.

Extensions

224
 A set ∆∗ of formulas is called an extension of a set ∆ of formulas if the
following condition holds.

{A ∈ F : ∆ ` A} ⊆ {A ∈ F : ∆∗ ` A}. (5.57)

In this case we say also that ∆ extends to the set of formulas ∆∗ .

The Main Lemma states as follows.

Lemma 5.11 ( Complete Consistent Extension)

Every consistent set ∆ of formulas can be extended to a complete consistent set
∆∗ of formulas.

Proof
Assume that the lemma does not hold, i.e. that there is a consistent set ∆,
such that all its consistent extensions are not complete. In particular, as ∆ is
an consistent extension of itself, we have that ∆ is not complete.
The proof consists of a construction of a particular set ∆∗ and proving that it
forms a complete consistent extension of ∆, contrary to the assumption that all
its consistent extensions are not complete.
Construction of ∆∗ .
As we know, the set F of all formulas is enumerable. They can hence be put in
an infinite sequence

A1 , A2 , ...., An , ..... (5.58)

such that every formula of F occurs in that sequence exactly once.
We define now, as the first step in the construction of ∆∗ , an infinite sequence
{∆n }n∈N of consistent subsets of formulas together with a sequence {B}n∈N of
formulas as follows.

Initial Step
In this step we define the sets ∆1 , ∆2 and the formula B1 . We prove that
∆1 and ∆2 are consistent, incomplete extensions of ∆.
We take, as the first set, the set ∆, i.e. we define

∆1 = ∆. (5.59)

Since, by assumption, the set ∆, and hence also ∆1 is not complete, it

follows from the Incomplete Set Condition Lemma 5.10, that there is a
formula B ∈ F such that ∆1 6` B, then and the set ∆1 ∪{B} is consistent.

225
 Let
B1
be the first formula with this property in the sequence (5.58) of all formu-
las; we then define

∆2 = ∆1 ∪ {B1 }. (5.60)

The set ∆2 is consistent and ∆1 = ∆ ⊆ ∆2 , so by the monotonicity, ∆2 is a

consistent extension of ∆. Hence ∆2 cannot be complete.

Inductive Step
Suppose that we have defined a sequence
∆1 , ∆2 , ..., ∆n
of incomplete, consistent extensions of ∆, and a sequence
B1 , B2 , ...Bn−1
of formulas, for n ≥ 2.
Since ∆n is incomplete, it follows from the Incomplete Set Condition
Lemma 5.10, that there is a formula B ∈ F such that ∆n 6` B and
the set ∆n ∪ {B} is consistent.
Let
Bn
be the first formula with this property in the sequence (5.58) of all formu-
las.
We then define

∆n+1 = ∆n ∪ {Bn }. (5.61)

By the definition, ∆ ⊆ ∆n ⊆ ∆n+1 and the set ∆n+1 is consistent. Hence ∆n+1
is an incomplete consistent extension of ∆.
By the principle of mathematical induction we have defined an infinite sequence
∆ = ∆1 ⊆ ∆2 ⊆ ..., ⊆ ∆n ⊆ ∆n+1 ⊆ .... (5.62)
such that for all n ∈ N , ∆n is consistent, and moreover, it is an incomplete
consistent extension of ∆.
Moreover, we have also defined a sequence
B1 , B2 , ..., Bn , .... (5.63)
of formulas, such that for all n ∈ N , ∆n 6` Bn , and the set ∆n ∪ {Bn } is
consistent.
Observe that Bn ∈ ∆n+1 for all n ≥ 1.

226
Definition of ∆∗
Now we are ready to define ∆∗ , i.e. we define:
[
∆∗ = ∆n . (5.64)
n∈N

To complete the proof our theorem we have now to prove that ∆∗ is a complete
consistent extension of ∆. Obviously, by the definition, ∆∗ is an extension of
∆. Now we prove (by contradiction) the following.

Fact 5.5
The set ∆∗ is consistent.

Proof
Assume that ∆∗ is inconsistent. By the Finite Inconsistency Theorem 5.7 there
is a finite subset ∆0 of ∆∗ that is inconsistent. By Definition 5.64 have that
[
∆0 = {C1 , ..., Cn } ⊆ ∆n .
n∈N

By the definition, Ci ∈ ∆ki for certain ∆ki in the sequence (5.62) and 1 ≤ i ≥ n.
Hence ∆0 ⊆ ∆m for m = max{k1 , k2 , ..kn }. But all sets of the sequence (5.62)
are consistent. This contradicts the fact that ∆m is inconsistent, as it contains
an inconsistent subset ∆0 . Hence ∆∗ must be consistent.

Fact 5.6
The set ∆∗ is complete.

Proof
Assume that ∆∗ is not complete. By the Incomplete Set Condition Lemma 5.10,
there is a formula B ∈ F such that ∆∗ 6` B and the set ∆∗ ∪ {B} is consistent.
But, by definition (5.64) of ∆∗ , the above condition means that for every n ∈ N ,
∆n 6` B holds and the set ∆n ∪ {B} is consistent.
Since the formula B is one of the formulas of the sequence (5.58) and it would
have to be one of the formulas of the sequence (5.63), i.e. B = Bj for certain
j. Since Bj ∈ ∆j+1 , it proves that B ∈ ∆∗ = n∈N . But this means that
S
∆∗ ` B, contrary to the assumption. This proves that ∆∗ is a complete
consistent extension of ∆ and ends the proof out our lemma.
Now we are ready to prove the completeness theorem for the system H2 .

Proof of the Completeness Theorem

227
As by assumption our system H2 is sound, we have to prove only the Complete-
ness part of the Completeness Theorem 5.6, i.e we have to show the implication
if |= A, then ` A
for any formula A. We prove it by proving the logically equivalent opposite
implication
if 6` A, then |6 = A.

We remind that 6|= A means that there is a variable assignment v : V AR −→

{T, F }, such that v ∗ (A) 6= T . In classical case it means that v ∗ (A) = F , i.e.
that there is a variable assignment that falsifies A. Such v is also called a
counter-model for A.
Assume that A doesn’t have a proof in S, we want to define a counter-model
for A. But if 6` A, then by the lemma 5.8, the set {¬A} is consistent. By the
Main Lemma 5.11 there is a complete, consistent extension of the set {¬A},
i.e. there is a set set ∆∗ such that {¬A} ⊆ ∆∗ , i.e.
¬A ∈ ∆∗ . (5.65)

Since ∆∗ is a consistent, complete set, it satisfies the following form of the

consistency condition 5.54, which says that for any A, ∆∗ 6` A or ∆∗ 6` ¬A.
It also satisfies the completeness condition (5.55), which says that for any A,
∆∗ ` A or ∆∗ ` ¬A. This means that for any A, exactly one of the
following conditions is satisfied: ∆∗ ` A, ∆∗ ` ¬A. In particular, for
every propositional variable a ∈ V AR exactly one of the following conditions is
satisfied: ∆∗ ` a, ∆∗ ` ¬a. This justifies the correctness of the following
definition.

Definition of v
We define the variable assignment
v : V AR −→ {T, F } (5.66)
as follows:

if ∆∗ ` a

T
v(a) =
F if ∆∗ ` ¬a.

We show, as a separate lemma below, that such defined variable assignment v

has the following property.

Lemma 5.12 (Property of v)

Let v be the variable assignment defined by ( 5.66) and v ∗ its extension to the
set F of all formulas. Then for every formula B ∈ F,
T if ∆∗ ` B

v ∗ (B) = (5.67)
F if ∆∗ ` ¬B.

228
Given the above property (5.67) of v (still to be proven), we prove that the v
is in fact, a counter model for any formula A, such that 6` A as follows. Let
A be such that 6` A. By ( 5.65), ¬A ∈ ∆∗ and obviously, ∆∗ ` ¬A. Hence,
by the property (5.67) of v, v ∗ (A) = F , what proves that v is a counter-model
for A and hence ends the proof of the completeness theorem. In order to really
complete the proof we still have to write a proof of the Lemma 5.12.

Proof of the Lemma 5.12

The proof is conducted by the induction on the degree of the formula A.
If A is a propositional variable, then the lemma is true holds by (5.66), i.e. by
the definition of v.
If A is not a propositional variable, then A is of the form ¬C or (C ⇒ D),
for certain formulas C, D. By the inductive assumption the lemma, i.e. the
property (5.67) holds for the formulas C and D.

Case A = ¬C. We have to consider two possibilities: ∆∗ ` A and ∆∗ ` ¬A.

Assume ∆∗ ` A. It means that ∆∗ ` ¬C. Then from the fact that ∆∗ is

consistent it must be that ∆∗ 6` C. This means, by the inductive assumption,
that v ∗ (C) = F , and accordingly

v ∗ (A) = v ∗ (¬C) = ¬v ∗ (C) = ¬F = T.

Assume now that ∆∗ ` ¬A. Then from the fact that ∆∗ is consistent it must
be that ∆∗ 6` A. I.e. ∆∗ 6` ¬C. If so, then ∆∗ ` C, as the set ∆∗ is
complete. Hence by the inductive assumption, that v ∗ (C) = T , and accordingly

v ∗ (A) = v ∗ (¬C) = ¬v ∗ (C) = ¬T = F.

Thus A satisfies the property (5.67).

Case A = (C ⇒ D). As in the previous case, we assume that the lemma, i.e.
the property (5.67) holds for the formulas C, D and we consider two pos-
sibilities: ∆∗ ` A and ∆∗ ` ¬A.

Assume ∆∗ ` A. It means that ∆∗ ` (C ⇒ D). If at the same time ∆∗ 6` C,

then v ∗ (C) = F , and accordingly

v ∗ (A) = v ∗ (C ⇒ D) = v ∗ (C) ⇒ v ∗ (D) = F ⇒ v ∗ (D) = T.

If at the same time ∆∗ ` C, then, since ∆∗ ` (C ⇒ D), we infer, by Modus

Ponens, that ∆∗ ` D. If so, then

v ∗ (C) = v ∗ (D) = T,

229
and accordingly

v ∗ (A) = v ∗ (C ⇒ D) = v ∗ (C) ⇒ v ∗ (D) = T ⇒ T = T.

Thus, if ∆∗ ` A, then v ∗ (A) = T .

Assume now, as before, that ∆∗ ` ¬A. Then from the fact that ∆∗ is consistent
it must be that ∆∗ 6` A, i.e.,

∆∗ 6` (C ⇒ D).

It follows from this that

∆∗ 6` D,
for if ∆∗ ` D, then, as (D ⇒ (C ⇒ D)) is provable (Lemma 5.4), by mono-
tonicity
∆∗ ` (D ⇒ (C ⇒ D)).
Applying Modus Ponens we obtain ∆∗ ` (C ⇒ D), which is contrary to the
assumption.
Also we must have
∆∗ ` C,
for otherwise, by the fact that ∆∗ we would have

∆∗ ` ¬C.

But this is impossible, since the formula (¬C ⇒ (C ⇒ D)) is provable provable
(Lemma 5.4) and by monotonicity

∆∗ ` (¬C ⇒ (C ⇒ D)).

Applying Modus Ponens we would get ∆∗ ` (C ⇒ D), which is contrary to

the assumption. This ends the proof of the Lemma 5.12 and the Proof Two
of the Completeness Theorem 5.6.

5.4 Some Other Axiomatizations

We present here some of most known, and historically important axiomatiza-
tions of classical propositional logic, i.e. the following Hilbert proof systems.

Lukasiewicz (1929)

L = ( L{¬, ⇒} , F, A1, A2, A3, M P ), (5.68)

where

230
A1 ((¬A ⇒ A) ⇒ A),
A2 (A ⇒ (¬A ⇒ B)),
A3 ((A ⇒ B) ⇒ ((B ⇒ C) ⇒ (A ⇒ C))))
for any A, B, C ∈ F.
2. Hilbert and Ackermann (1928)

HA = ( L{¬,∪} , F, A1 − A4, M P ), (5.69)

where
A1 (¬(A ∪ A) ∪ A),
A2 (¬A ∪ (A ∪ B)),
A3 (¬(A ∪ B) ∪ (B ∪ A)),
A4 (¬(¬B ∪ C) ∪ (¬(A ∪ B) ∪ (A ∪ C))),
for any A, B, C ∈ F.
Modus Ponens rule in the language L{¬,∪} has a form

A ; (¬A ∪ B)
(M P ) .
B

Observe that also the Deduction Theorem is now formulated as follow.

Theorem 5.8 (Deduction Theorem for HA)

For any subset Γ of the set of formulas F of HA and for any formulas A, B ∈ F,

Γ, A `HA B if and only if Γ `HA (¬A ∪ B).

In particular,
A `HA B if and only if `HA (¬A ∪ B).

2. Hilbert (1928)

H = ( L{¬,∪,∩,⇒} , F, A1 − A15, M P ), (5.70)

where
A1 (A ⇒ A),
A2 (A ⇒ (B ⇒ A)),
A3 ((A ⇒ B) ⇒ ((B ⇒ C) ⇒ (A ⇒ C))),
A4 ((A ⇒ (A ⇒ B)) ⇒ (A ⇒ B)),

231
A5 ((A ⇒ (B ⇒ C)) ⇒ (B ⇒ (A ⇒ C))),
A6 ((A ⇒ B) ⇒ ((B ⇒ C) ⇒ (A ⇒ C))),
A7 ((A ∩ B) ⇒ A),
A8 ((A ∩ B) ⇒ B),
A9 ((A ⇒ B) ⇒ ((A ⇒ C) ⇒ (A ⇒ (B ∩ C))),
A10 (A ⇒ (A ∪ B)),
A11 (B ⇒ (A ∪ B)),
A12 ((A ⇒ C) ⇒ ((B ⇒ C) ⇒ ((A ∪ B) ⇒ C))),
A13 ((A ⇒ B) ⇒ ((A ⇒ ¬B) ⇒ ¬A)),
A14 (¬A ⇒ (A ⇒ B)),
A15 (A ∪ ¬A),
for any A, B, C ∈ F.
Kleene (1952)

K = ( L{¬,∪,∩,⇒} , F, A1 − A10, M P ), (5.71)

where
A1 (A ⇒ (B ⇒ A)),
A2 ((A ⇒ (B ⇒ C)) ⇒ (B ⇒ (A ⇒ C))),
A3 ((A ∩ B) ⇒ A),
A4 ((A ∩ B) ⇒ B),
A5 (A ⇒ (B ⇒ (A ∩ B))),
A6 (A ⇒ (A ∪ B)),
A7 (B ⇒ (A ∪ B)),
A8 ((A ⇒ C) ⇒ ((B ⇒ C) ⇒ ((A ∪ B) ⇒ C))),
A9 ((A ⇒ B) ⇒ ((A ⇒ ¬B) ⇒ ¬A)),
A10 (¬¬A ⇒ A)
for any A, B, C ∈ F.
Rasiowa-Sikorski (1950)

RS = ( L{¬,∪,∩,⇒} , F, A1 − A12, M P ), (5.72)

where

232
A1 ((A ⇒ B) ⇒ ((B ⇒ C) ⇒ (A ⇒ C))),
A2 (A ⇒ (A ∪ B)),
A3 (B ⇒ (A ∪ B)),
A4 ((A ⇒ C) ⇒ ((B ⇒ C) ⇒ ((A ∪ B) ⇒ C))),
A5 ((A ∩ B) ⇒ A),
A6 ((A ∩ B) ⇒ B),
A7 ((C ⇒ A) ⇒ ((C ⇒ B) ⇒ (C ⇒ (A ∩ B))),
A8 ((A ⇒ (B ⇒ C)) ⇒ ((A ∩ B) ⇒ C)),
A9 (((A ∩ B) ⇒ C) ⇒ (A ⇒ (B ⇒ C)),
A10 (A ∩ ¬A) ⇒ B),
A11 ((A ⇒ (A ∩ ¬A)) ⇒ ¬A),
A12 (A ∪ ¬A),
for any A, B, C ∈ F.

Here is the shortest axiomatization for the language L{¬, ⇒} . It contains just
one axiom.
Meredith (1953)
L = ( L{¬, ⇒} , F, A1 M P ), (5.73)

where
A1 ((((((A ⇒ B) ⇒ (¬C ⇒ ¬D)) ⇒ C) ⇒ E)) ⇒ ((E ⇒ A) ⇒ (D ⇒ A))).

We have proved in chapter ?? that

L{¬,∪,∩,⇒} ≡ L{↑} ≡ L{↓} .

Here is another axiomatization that uses only one axiom,

Nicod (1917)
N = ( L{↑} , F, A1, (r) ), (5.74)

where
A1 (((A ↑ (B ↑ C)) ↑ ((D ↑ (D ↑ D)) ↑ ((E ↑ B) ↑ ((A ↑ E) ↑ (A ↑ E)))))).
The rule of inference is (r) is expressed in the language L{↑} as

A ↑ (B ↑ C)
.
A

233
5.5 Exercises
Here are few exercises designed to help the readers with understanding the
notions of completeness, monotonicity of the consequence operation, the role of
the deduction theorem and importance of some basic tautologies.
Let S be any Hilbert proof system

A, (A ⇒ B)
S = (L{∩,∪,⇒,¬} , F, LA, (M P ) ) (5.75)
B
with its set LA of logical axioms such that S is complete under classical se-
mantics.
Let X ⊆ F be any subset of the set F of formulas of the language L{∩,∪,⇒,¬}
of S. We define, as we did in chapter 4, a set Cn(X) of all consequences of
the set X as
Cn(X) = {A ∈ F : X `S A}. (5.76)

Plainly speaking, the set Cn(X) of all consequences of the set X is the set of
all formulas that can be proved in S from the set (LA ∪ X).

All exercises 5.9 - 5.13 concern the system S defined by (5.75).

Exercise 5.9
1. Prove that for any subsets X, Y of the set F of formulas the following mono-
tonicity property holds.

If X ⊆ Y , then Cn(X) ⊆ Cn(Y ). (5.77)

2. Do we need the completeness of S to prove that the monotonicity property

holds for S?

Solution
1. Let A ∈ F be any formula such that A ∈ Cn(X). By (5.76), we have that
X `S A. This means that A has a formal proof from the set X ∪ LA. But
X ⊆ Y , hence this proof is also a proof from Y ∪ LA, i.e . Y `S A, and hence
A ∈ Cn(Y ). This proves that Cn(X) ⊆ Cn(Y ).

2. No, we do not need the completeness of S for the monotonicity property to

hold. We have used only the definition of a formal proof from the hypothesis X
and the definition of the consequence operation.

Exercise 5.10

234
Prove that for any set X ⊆ F, the set T ⊆ F of all propositional classical
tautologies of the language L{∩,∪,⇒,¬} of the system S is a subset of Cn(X),
i.e. prove that
T ⊆ Cn(X). (5.78)
2. Do we need the completeness of S to prove that the property (5.78) holds
for S?

Solution
1. The proof system S is complete, so by the completeness theorem we have
that
T = {∈ F : `S A}. (5.79)
By definition (5.76) of the consequence,

{A ∈ F : `S A} = Cn(∅)

and hence Cn(∅) = T. But ∅ ⊆ X for any set X, so by monotonicity property

(5.77),
T ⊆ Cn(X).
2. Yes, the completeness (5.79) of S in the main property used. The next one
is the monotonicity property (5.77).

Exercise 5.11
Prove that for any formulas A, B ∈ F, and for any set X ⊆ F,

(A ∩ B) ∈ Cn(X) if and only if A ∈ Cn(X) and B ∈ Cn(X). (5.80)

List all properties essential to the proof.

1. Proof of the implication:

if (A ∩ B) ∈ Cn(X), then A ∈ Cn(X) and B ∈ Cn(X).

Assume (A ∩ B) ∈ Cn(X), i.e.

X `S (A ∩ B). (5.81)

From monotonicity property proved in exercise 5.9, completeness of S, and the

fact that
|= ((A ∩ B) ⇒ A) and |= ((A ∩ B) ⇒ B) (5.82)
we get that

X `S ((A ∩ B) ⇒ A), and X `S ((A ∩ B) ⇒ B). (5.83)

By the assumption (5.81) we have that X `S (A ∩ B), by (8.3), X`S ((A ∩ B) ⇒

A), and so we get X `S A by Modus Ponens.

235
Similarly, X `S (A ∩ B), by the assumption (5.81), X`S ((A ∩ B) ⇒ B) by
by (??), and so we get X `S B by MP. This proves that A ∈ Cn(X) and
B ∈ Cn(X) and ends the proof of the implication 1.

2. Proof of the implication:

if A ∈ Cn(X) and B ∈ Cn(X), then (A ∩ B) ∈ Cn(X).

Assume now that A ∈ Cn(X) and B ∈ Cn(X), i.e.

X `S A, and X `S B. (5.84)

By the monotonicity property, completeness of S, and a tautology (A ⇒ (B ⇒

(A ∩ B))), we get that

X `S (A ⇒ (B ⇒ (A ∩ B))). (5.85)

By the assumption (5.84) we have that X `S A, X`S B, by (8.17), X `S (A ⇒

(B ⇒ (A ∩ B))), so we get X `S (B ⇒ (A ∩ B)) by Modus Ponens. Applying
Modus Ponens again we obtain X `S (A∩B). This proves that (A∩B) ∈ Cn(X)
and ends the proof and the implication 2, and the completes the proof of (5.80).

Exercise 5.12
Let S be the proof system (5.75). Prove that the Deduction Theorem holds for
S, i.e. prove the following.
For any subset Γ of the set of formulas F of S and for any formulas A, B ∈ F,

Γ, A `S B if and only if Γ `S (A ⇒ B). (5.86)

Solution
The formulas A1 = (A ⇒ (B ⇒ A)) and A2 = ((A ⇒ (B ⇒ C)) ⇒ ((A ⇒
B) ⇒ (A ⇒ C))) are basic propositional tautologies. By the completeness of S
we have that

`S (A ⇒ (B ⇒ A)) and `S ((A ⇒ (B ⇒ C)) ⇒ ((A ⇒ B) ⇒ (A ⇒ C))).

(5.87)
The formulas A1, A2 are axioms of the Hilbert system H1 defined by (5.1). By
(5.87) both axioms A1, A2 of H1 are provable in S. These axioms were sufficient
for the proof of the Deduction Theorem 5.3 for H1 and its proof now can be
repeated for the system S.

Exercise 5.13
Prove that for any A, B ∈ F,

Cn({A, B}) = Cn({(A ∩ B)})

236
1. Proof of the inclusion: Cn({A, B}) ⊆ Cn({(A ∩ B)}).
Assume C ∈ Cn({A, B}), i.e. {A, B}`S C, what we usually write as A, B `S C.
Observe that by exercise 8.7 the Deduction Theorem (theorem 5.3) holds for S.
We apply Deduction Theorem to the assumption A, B `S C twice we get that
the assumption is equivalent to
`S (A ⇒ (B ⇒ C)). (5.88)
We use completeness of S, the fact that the formula (((A ⇒ (B ⇒ C)) ⇒
((A ∩ B) ⇒ C))) is a tautology, and by monotonicity and get that
`S (((A ⇒ (B ⇒ C)) ⇒ ((A ∩ B) ⇒ C))). (5.89)
Applying Modus Ponens to the assumption (5.88) and (8.22) we get `S ((A ∩
B) ⇒ C). This is equivalent to (A ∩ B) `S C by Deduction Theorem. We have
proved that C ∈ Cn({(A ∩ B)}).

2. Proof of the inclusion: Cn({(A ∩ B)}) ⊆ Cn({A, B})}).

Assume that C ∈ Cn({(A ∩ B)}), i.e. (A ∩ B)`S C. By Deduction Theorem,
`S ((A ∩ B) ⇒ C). (5.90)
We want to prove that C ∈ Cn({A, B}). This is equivalent, by the Deduction
Theorem applied twice to proving that
`S (A ⇒ (B ⇒ C)).
The proof as similar to the previous case. We use completeness of S, the fact
that the formula (((A ∩ B) ⇒ C) ⇒ (A ⇒ (B ⇒ C))) is a tautology and by
monotonicity and get that
`S (((A ∩ B) ⇒ C) ⇒ (A ⇒ (B ⇒ C)) (5.91)
Applying Modus Ponens to the assumption (5.88) and (8.22) we get `S (A ⇒
(B ⇒ C)) what ends the proof.

5.6 Homework Problems

0
For the formulas Ai and corresponding truth assignments v find formulas B1 , ..Bk , Ai
as described by the Main Lemma 5.4, i.e. such that
0
B1 , ...Bk ` Ai .

1. A1 = ((¬(b ⇒ a) ⇒ ¬a) ⇒ ((¬b ⇒ (a ⇒ ¬c)) ⇒ c))

v(a) = T, v(b) = F, v(c) = T .

237
 2. A2 = ((a ⇒ (c ⇒ (¬b ⇒ c))) ⇒ ((¬d ⇒ (a ⇒ (¬a ⇒ b))) ⇒ (a ⇒ (¬a ⇒
b))))
v(a) = F, v(b) = F, v(c) = T, v(d) = F

3. A3 = (¬b ⇒ (c ⇒ (¬a ⇒ b)))

v(a) = F, v(b) = F, v(c) = T

4. A4 = (¬a1 ⇒ (a2 ⇒ (¬a3 ⇒ a1 )))

v(a1 ) = F, v(a2 ) = F, v(a3 ) = T

4. A5 = ((b ⇒ (a1 ⇒ (¬c ⇒ b))) ⇒ ((¬b ⇒ (a2 ⇒ (¬a1 ⇒ b))) ⇒ (c ⇒

(¬a ⇒ b)))
v(a) = F, v(b) = T, v(c) = F, v(a1 ) = T, v(a2 ) = F

For any of the formulas listed below construct their formal proofs, as described
in the Proof One of the Completeness Theorem. Follow example 5.10, or exam-
ple 5.11.

1. A1 = (¬¬b ⇒ b)
2. A2 = ((a ⇒ b) ⇒ (¬b ⇒ ¬a))
3. A3 = (¬(a ⇒ b) ⇒ ¬(¬b ⇒ ¬a))
4. A4 = (¬(¬(a ⇒ ¬b) ⇒ ¬c) ⇒ ¬(b ⇒ ¬c))
5. A5 = ((a ⇒ (b ⇒ ¬a)) ⇒ (¬(b ⇒ ¬a) ⇒ ¬a)).

Read carefully proofs of Deduction Theorem 5.3 and Completeness Theo-

rem 5.6 and write careful answers to the following problems.

6. List all formulas that have to be provable in H2 , axioms included, that

are are needed for the proof of Deduction Theorem 5.3. Write down each
part of the proof that uses them.
7. List all formulas that have to be provable in H2 , axioms included, that
are needed for the proof of Main Lemma 5.4.
8. List all formulas that have to be provable in H2 , axioms included, that are
included in the Proof of Completeness Theorem part of the Proof One.
9. List all formulas that have to be provable in H2 , axioms included, that
are needed to carry all of the Proof One of Completeness Theorem ??.

238
 10. We proved the Completeness Theorem for the proof system H2 based on
the language L{¬,⇒} . Extend the H2 proof system to a proof system S1
based on a language L{¬,⇒,∪} by adding new logical axioms, as we did in
a case of H1 and H2 systems. The added logical axioms must be such that
they allow to adopt the Proof One to S1 , i.e. such that it is a complete
proof system with respect to classical semantics.

11. Repeat the same for the language L{¬,⇒,∩} . Call resulting proof system
S2 .

12. Repeat the same for the language L{¬,⇒,∩,∪} , i.e. extends systems S1 or
S2 to a complete proof system S3 based on the language L{¬,⇒,∩,∪} .

13. Prove Completeness Theorem for the system S3 from the previous prob-
lem.

Completeness Proof Two Problems

1. List all formulas that have to be provable in H2 , axioms included, that

are are needed for the Proof Two if the Completeness Theorem 5.6.

2. We proved the Completeness Theorem 5.6 for the proof system H2 based
on the language L{¬,⇒} . Extend the H2 proof system to a proof system
S1 based on a language L{¬,⇒,∪} by adding new logical axioms, as we
did in a case of H1 and H2 systems. The added logical axioms must be
such that they allow to adopt the Proof Two to S1 , i.e. such that it is a
complete proof system with respect to classical semantics.

3. Extend the H2 proof system to a proof system based on a language

L{¬,⇒,∩} by adding new logical axioms. Call resulting proof system S2 .
The added logical axioms must be such that they allow to adopt the Proof
Two to S2 , i.e. such that it is a complete proof system with respect to
classical semantics.

4. Repeat the same for the language L{¬,⇒,∩,∪} , i.e. extends systems S1 or
S2 to a complete proof system S3 based on the language L{¬,⇒,∩,∪} .

5. Conduct appropriate version of Proof Two of the Completeness Theorem

5.6 for the system S3 from the previous problem.

Axiomatizations Problems

1. Let HA be Hilbert and Ackermann proof system (5.69). We use abbrevi-

ation (A ⇒ B) for (6 A ∪ B).
(i) Prove `HA (A ⇒ A), for any A ∈ F.
(ii) Prove `HA (A ⇒ (B ⇒ A)), for any A, B ∈ F.

239
 (iii) Prove `HA ((A ⇒ (B ⇒ C)) ⇒ ((A ⇒ B) ⇒ (A ⇒ C))), for any
A, B, C ∈ F.
(iv) Prove (A ⇒ B), (B ⇒ C) `HA (A ⇒ C), for any A, B, C ∈ F
(v) Prove Deduction Theorem 5.8.
(vi) Prove `HA A if and only if |= A, for any A ∈ F.
2. Let H be Hilbert proof system (??).
(i) Prove `HA ((A ⇒ (B ⇒ C)) ⇒ ((A ⇒ B) ⇒ (A ⇒ C))), for any
A, B, C ∈ F.
(ii) Prove Deduction Theorem for H.
(ii) Prove Completeness Theorem for H.
3. Let K be Kleene proof system (5.71).
(i) Prove `K (A ⇒ A), for any A ∈ F.
(ii) Prove the following.
For any subset Γ of the set of formulas F of K and for any formulas
A, B ∈ F, Γ, A `K B if and only if Γ `K ((A ⇒ B)).

Completeness General Problems

1. Let RS be Rasiowa-Sikorski proof system (5.72).

The set F of formulas of L determines an abstract algebra

F = ( F, ∪, ∩, ⇒, ¬ ), (5.92)

where by performing an operation on a formula (two formulas) means

writing the formula having this operation as a main connective. For ex-
ample ∩(A, B) = (A ∪ B). We define an binary relations ≤ and ≈ in the
algebra F of formulas of L as follows. For any A, B ∈ F,

A ≤ B if and only if `RS (A ⇒ B), (5.93)

A ≈ B if and only if `RS (A ⇒ B) and ` (B ⇒ A). (5.94)

(i) Prove that the relation ≤ defined by (5.93) is a quasi-ordering in F.

(ii) Prove that the relation ≈ defined by (5.93) is an equivalence relation
in F. The equivalence class containing a formula A is denoted by kAk.

(iii) The quasi ordering ≤ in F as defined by (5.93) induces a relation ≤

in F/ ≈ defined as follows:

kAk ≤ kBk if and only if A ≤ B, i.e.

kAk ≤ kBk if and only if `RS (A ⇒ B). (5.95)

240
 Prove that the relation (5.95) is an order relation in F/ ≈
(iv) Prove that the relation ≈ defined by (5.94) is a congruence in the
algebra F of formulas defined by (8.5).
2. The algebra LT = ( F/ ≈, ∪, ∩, ⇒, ¬), where the operations ∪, ∩, ⇒
and ¬ are determined by the congruence relation (5.94) i.e.

kAk ∪ kBk = k(A ∪ B)k,

kAk ∩ kBk = k(A ∩ B)k,

kAk ⇒ kBk = k(A ⇒ B)k,
¬kAk = k¬Ak,
is called a Lindenbaum-Tarski algebra of RS.
Prove that the Lindenbaum-Tarski algebra of RS as defined by (5.72) is a
Boolean algebra. The unit element is the greatest element in (F/ ≈, ≤),
where the order relation ≤ is defined by (5.95).
3. Formulate and prove the Deduction Theorem for Hilbert and Ackermann
system (5.69).

4. Formulate and prove the Deduction Theorem for Lukasiewicz system

(5.68).
5. Formulate and prove the Deduction Theorem Kleene system (5.71).
6. Formulate and prove the Deduction Theorem Rasiowa-Sikorski system
(5.72)
7. Let HS be any Hilbert proof system based on a language LHS . Prove that
if HS is complete under classical semantic, them the Deduction Theorem
appropriately expressed in the language LHS holds for HS.

241
242
Chapter 6

Automated Proof Systems

Completeness of Classical
Propositional Logic

6.1 Gentzen Style Proof System RS

Hilbert style systems are easy to define and admit different proofs of the Com-
pleteness Theorem but they are difficult to use. By humans, not mentioning
computers. Their emphasis is on logical axioms, keeping the rules of inference,
with obligatory Modus Ponens, at a minimum.
Gentzen style proof systems reverse this situation by emphasizing the impor-
tance of inference rules, reducing the role of logical axioms to an absolute mini-
mum. They may be less intuitive then the Hilbert-style systems, but they allow
us to define effective automatic procedures for proof search, what was impossi-
ble in a case of the Hilbert style systems. For this reason they are also called
automated proof systems. They serve as formal models of computing systems
that automate the reasoning process. Building computing systems means pro-
viding an algorithmic description to a formal proof system so that it can be
implemented on a computer to prove theorems in an efficient manner.
The first proof systems of this style was invented by G. Gentzen in 1934, hence
the name. His proof systems for classical and intuitionistic predicate logics intro-
duced special expressions built of formulas called sequents. Hence the Gentzen
style systems using sequents as basic expressions are often called sequent sys-
tems, or Gentzen sequent systems, or simply Gentzen formalizations.

We present here (section 6.5) two Gentzen systems GL and G for classical

243
propositional logic and prove their completeness. We also present a proposi-
tional version of Gentzen original system LK and discuss a proof of Gentzen
Hauptsatz for it. Hauptsatz is literally rendered as the main theorem and is
known as cut-elimination theorem. We prove the equivalency of the cut-free
propositional LK and the complete system G. The Gentzen original formaliza-
tion for intuitionistic propositional logic LI is discussed and presented in chapter
7. The classical and intuitionistic predicate versions are discussed in chapter 9.

The other historically important automated proof system is due to Rasiowa

and Sikorski (1960). Their proof systems for classical propositional and predi-
cate logic use as basic expressions sequences of formulas, less complicated then
Gentzen sequents. As they were inspired Gentzen systems we call them, as we
call many others similarly inspired, Gentzen style proof system, or Gentzen style
formalization. The Rasiowa and Sikorski proof system is simpler and easier to
understand then the Gentzen sequent systems. Hence their system RS is the
first to be presented here in section 6.1.

Historical importance and lasting influence of Rasiowa and Sikorski work lays
in the fact that they were first to use the proof searching capacity of their proof
system to define a constructive method of proving the completeness theorem
for both propositional and predicate classical logic. We introduce and explain
in detail their method and use it prove the completeness of the RS in section
6.3. We also introduce and discuss two other RS style system RS1 and RS2
in in section 6.4. We also generalize the RS completeness proof method to the
Gentzen sequent systems and prove the completeness of GL and G systems in
section 6.5.1. The completeness proof for proof system RSQ for classical pred-
icate logic is presented in chapter 9.

We present here a propositional version of the original Rasiowa and Sikorski

(1960) Gentzen style proof system for classical logic. We call it RS system for
Rasiowa-Sikorski. The RS system extends naturally to predicate logic QRS
system which is presented in chapter 8. Both systems admit a constructive
proof of Completeness Theorem. We prove completeness of RS in section 6.3.
We define components and semantics of the system RS as follows.

Components of the proof system RS

Language L
Let F denote a set of formulas of L = L{¬,⇒,∪,∩} . The rules of inference of our
system RS operate on finite sequences of formulas, i.e. elements of F ∗ , unlike
on plain formulas F in Hilbert style formalizations.
Expressions E

244
We adopt as the set of expressions E of RS the set F ∗ , i.e. E = F ∗ . We will
denote the expressions of RS, i.e. the finite sequences of formulas by Γ, ∆, Σ,
with indices if necessary.

Semantic Link
The intuitive meaning of a sequence Γ ∈ F ∗ is that the truth assignment v makes
it true if and only if it makes the formula of the form of the disjunction of all
formulas of Γ true. As we know, the disjunction in classical logic is associative
and commutative, i.e., for any formulas A, B, C ∈ F, the formulas (A∪(B ∪C)),
((A ∪ B) ∪ C), (A ∪ (C ∪ B)), ((B ∪ A) ∪ C), (C ∪ (B ∪ A)), (C ∪ (A ∪ B)),
((C ∪ A) ∪ B), etc... are logically equivalent. In particular we write

δ{A,B,C} = A ∪ B ∪ C

to denote any disjunction of formulas A, B, C.

In a general case, for any sequence Γ ∈ F ∗ , if Γ is of a form

A1 , A2 , ..., An (6.1)

then by δΓ we will understand any disjunction of all formulas of Γ. We write it

informally
δΓ = A1 ∪ A2 ∪ ... ∪ An .

Formal Semantics for RS

Let v : V AR −→ {T, F } be a truth assignment, v ∗ its extension to the set of
formulas F. We formally extend v to the set |calE of expressions of RS , i.e.
to the set F ∗ of all finite sequences of F as follows. For any sequence Γ ∈ F ∗ ,
if Γ is the sequence (6.1), then we define:

v ∗ (Γ) = v ∗ (δΓ ). (6.2)

Model
A sequence Γ is said to be satisfiable if there is a truth assignment v : V AR −→
{T, F } such that v ∗ (Γ) = T . Such a truth assignment is called a model for Γ.
We denote it as
v |= Γ. (6.3)

Counter- Model
A sequence Γ is said to be falsifiable if there is a truth assignment v, such that
v ∗ (Γ) = F . Such a truth assignment is ] called a counter-model for Γ. We
write it symbolically as
v 6|= Γ. (6.4)

Tautology

245
The sequence Γ is said to be a tautology if v ∗ (Γ) = T for all truth assignments
v : V AR −→ {T, F }. We write it as

|= Γ. (6.5)

Exercise 6.1
Let Γ be a sequence a, (b ∩ a), ¬b, (b ⇒ a).
1. Show that the truth assignment v : V AR −→ {T, F }, such that v(a) = F
and v(b) = T falsifies Γ, i.e. v 6|= Γ.
2. Let Γ be a sequence a, (¬b ∩ a), ¬b, (a ∪ b) and let v be a truth assignment
for which v(a) = T . Prove that v |= Γ.
3. Let Γ be a sequence a, (¬b ∩ a), ¬b, (a ∪ b). Prove that |= Γ.

Solution
1. Γ is the sequence a, (b∩a), ¬b, (b ⇒ a). We eveluate v ∗ (Γ) = v ∗ (δΓ ) = v ∗ (a)∪
v ∗ (b ∩ a) ∪ v ∗ (¬b) ∪ v ∗ (b ⇒ a) = F ∪ (F ∩ T ) ∪ F ∪ (T ⇒ F ) = F ∪ F ∪ F ∪ F = F.
By (6.4) we proved v 6|= Γ.
2. Let Γ be a sequence a, (¬b ∩ a), ¬b, (a ∪ b). We eveluate v ∗ (Γ) = v ∗ (δΓ ) =
v ∗ (a) ∪ v ∗ (¬b ∩ a) ∪ v ∗ (¬b) ∪ v ∗ (a ∪ b) = T ∪ v ∗ (¬b ∩ a) ∪ v ∗ (¬b) ∪ v ∗ (a ∪ b) = T.
By (6.3) we proved v |= Γ.
3. Assume now that Γ is falsifiable i.e. that we have a truth assignment v
for which v ∗ (Γ) = v ∗ (δΓ ) = v ∗ (a) ∪ v ∗ (¬b ∩ a) ∪ v ∗ (¬b) ∪ v ∗ (a ∪ b) = F This is
possible only when (in short-hand notation)

a ∪ (¬b ∩ a) ∪ ¬b ∪ a ∪ b = F,

what is impossible as (¬b ∪ b) = T for all v. This contradiction proves that Γ

that (6.5) holds and Γ is a tautology.

In order to define the axioms LA and the set of rules of inference of RS we need
to introduce some definitions.
Literals
0
We form a special subset F ⊆ F of formulas, called a set of all literals, which
is defined as follows.

LT = V AR ∪ {¬a : a ∈ V AR}. (6.6)

The variables are called positive literals and the elements of the second set of
the above union (6.6) are called negative literals. I.e propositional variables
are called positive literals and the negation of a variable is called a negative

246
literal, a variable or a negation of propositional variable is called a literal.

Indecomposable formulas and sequences

Literals are also called the indecomposable formulas. Now we form finite se-
quences out of formulas (and, as a special case, out of literals). We need to
distinguish the sequences formed out of literals from the sequences formed out
of other formulas, so we adopt the following notation.

We denote by
0 0 0
Γ , ∆ , Σ , . . . with indices if necessary, (6.7)
0 0 0
elements of LT ∗ ⊆ F ∗ , i.e. Γ , ∆ , Σ are finite sequences (empty included)
formed out of literals. We call them indecomposable sequences.
We denote by
Γ, ∆, Σ, . . . with indices if necessary, (6.8)
the elements of F ∗ , i.e. Γ, ∆, Σ denote finite sequences (empty included)
formed out of elements of F.

Logical Axioms LA

As the logical axiom of RS we adopt any sequence of literals which contains

any propositional variable and its negation, i.e any sequence of the form

0 0 0
Γ1 , a, Γ2 , ¬a, Γ3 (6.9)
or of the form
0 0 0
Γ1 , ¬a, Γ2 , a, Γ3 (6.10)
0 0 0
for any variable a ∈ V AR and any sequences Γ1 , Γ2 , Γ3 ∈ LT ∗ of literals.

Semantic Link
Consider axiom (6.9). Directly from the extension of the notion of tautology
to bf RS (6.5), we have that for any truth assignments v : V AR −→ {T, F },
0 0 0 0 0 0 0
v ∗ (Γ1 , ¬a, Γ2 , a, Γ3 ) = v ∗ (Γ1 ) ∪ v ∗ (¬a) ∪ v ∗ (a) ∪ v ∗ (Γ2 , Γ3 ) = v ∗ (Γ1 ) ∪ T ∪
0 0
∗
v (Γ2 , Γ3 ) = T The same applies to the axiom (6.10) We have thus proved the
following.

Fact 6.1
Logical axioms of RS are tautologies.

Rules of inference R

247
All rules of inference of RS are of the form

Γ1 Γ1 ; Γ2
or ,
Γ Γ

where Γ1 , Γ2 , Γ ∈ F ∗ , i.e. Γ1 , Γ2 , Γ are any finite sequences (??) of formulas.

The sequences Γ1 , Γ2 are called premisses and Γ is called a conclusion of the
rule of inference.

Each rule of inference of RS introduces a new logical connective, or a negation

of a logical connective. We denote a rule of inerence that introduces the logical
connective ◦ in the conclusion sequent Γ by (◦). The notation (¬ ◦) means that
the negation of the logical connective ◦ is introduced in the conclusion sequence
Γ. As our language contains the connectives: ∩, ∪, ⇒ and ¬, so we we are going
to define the following seven inference rules:

(∪), (¬ ∪), (∩), (¬ ∩), (⇒), (¬ ⇒), and (¬ ¬). (6.11)

We define formally the inference rules of RS as follows.

Disjunction rules
0 0 0
Γ , A, B, ∆ Γ , ¬A, ∆ : Γ , ¬B, ∆
(∪) 0 , (¬ ∪)
Γ , (A ∪ B), ∆ Γ0 , ¬(A ∪ B), ∆

Conjunction rules
0 0 0
Γ , A, ∆ ; Γ , B, ∆ Γ , ¬A, ¬B, ∆
(∩) , (¬ ∩)
Γ0 , (A ∩ B), ∆ Γ0 , ¬(A ∩ B), ∆

Implication rules
0 0 0
Γ , ¬A, B, ∆ Γ , A, ∆ : Γ , ¬B, ∆
(⇒) 0 , (¬ ⇒)
Γ , (A ⇒ B), ∆ Γ0 , ¬(A ⇒ B), ∆

Negation rule
0
Γ , A, ∆
(¬ ¬)
Γ0 , ¬¬A, ∆
0
where Γ ∈ LT ∗ , ∆ ∈ F ∗ , A, B ∈ F.

The Proof System RS

248
Formally we define the proof system RS as follows.

RS = (L{¬,⇒,∪,∩} , E, LA, R), (6.12)

where E = {Γ : Γ ∈ F ∗ }, LA contains logical axioms of the system defined by

the schemas (6.9) and (6.10), R is the set of rules of inference:

R = {(∪), (¬∪), (∩), (¬∩), (⇒), (¬ ⇒), (¬¬)}

defined by (6.11).

By a formal proof of a sequence Γ in the proof system RS we understand

any sequence
Γ1 , Γ2 , .... Γn (6.13)
of sequences of formulas (elements of F ∗ , such that
Γ1 ∈ LA, Γn = Γ,
and for all i (1 ≤ i ≤ n) Γi ∈ AL, or Γi is a conclusion of one of the inference
rules of RS with all its premisses placed in the sequence Γ1 Γ2 ....Γi−1 .

As the proof system under consideration is fixed, we will write, as usual,

`Γ

instead of `RS Γ to denote that Γ has a formal proof in RS.

As the proofs in RS are sequences (definition of the formal proof) of sequences
of formulas (definition of RS ) we will not use ”,” to separate the steps of the
proof, and write the formal proof as Γ1 ; Γ2 ; .... Γn .
We write, however, the formal proofs in RS in a form of trees rather then in a
form of sequences, ie. in a form of a tree, where leafs of the tree are axioms,
nodes are sequences such that each sequence on the tree follows from the ones
immediately preceding it by one of the rules. The root is a theorem. We
picture, and write our tree-proofs with the node on the top, and leafs on the
very bottom, instead of more common way, where the leafs are on the top and
root is on the bottom of the tree. We adopt hence the following definition.

Definition 6.1 (Proof Tree)

By a proof tree, or RS-proof of Γ we understand a tree TΓ of sequences satisfying
the following conditions:
1. The topmost sequence, i.e the root of TΓ is Γ,
2. all leafs are axioms,
3. the nodes are sequences such that each sequence on the tree follows from the
ones immediately preceding it by one of the rules.

249
We picture, and write our proof trees with the node on the top, and leafs on
the very bottom, instead of more common way, where the leafs are on the top
and root is on the bottom of the tree.
In particular cases we write our proof trees indicating additionally the name
of the inference rule used at each step of the proof. For example, if the tree-
proof of a given formula A from axioms was obtained by the subsequent use of
the rules (∩), (∪), (∪), (∩), (∪), (¬¬), and (⇒), we represent it as the following
proof tree:

A (conclusion of (⇒))

| (⇒)

conclusion of (¬¬)

| (¬¬)

conclusion of (∪)

| (∪)

conclusion of (∩)
^
(∩)

conclusion of (∩) conclusion of (∪)

| (∪) | (∪)

axiom conclusion of (∩)

^
(∩)

axiom axiom

The proof trees are often called derivation trees and we will use this notion
as well. Remark that the proof trees don’t represent a different definition of
a formal proof. Trees represent a certain visualization of the proofs and any
formal proof in any system can be represented in a tree form.

Example 6.1
Here is a proof tree in RS of the de Morgan law (¬(a ∩ b) ⇒ (¬a ∪ ¬b)).

(¬(a ∩ b) ⇒ (¬a ∪ ¬b))

| (⇒)
¬¬(a ∩ b), (¬a ∪ ¬b)
| (¬ ¬)
(a ∩ b), (¬a ∪ ¬b)
^
(∩)

250
 a, (¬a ∪ ¬b) b, (¬a ∪ ¬b)
| (∪) | (∪)
a, ¬a, ¬b b, ¬a, ¬b

To obtain a ”linear ” formal proof of (¬(a ∩ b) ⇒ (¬a ∪ ¬b)) we just write down
the tree as a sequence, starting from the leafs and going up (from left to right)
to the root. The formal proof (with comments) thus obtained is:

a, ¬a, ¬b (axiom)

b, ¬a, ¬b (axiom)
a, (¬a ∪ ¬b) (rule (∪))
b, (¬a ∪ ¬b) (rule (∪))
(a ∩ b), (¬a ∪ ¬b) (rule(∩))
¬¬(a ∩ b), (¬a ∪ ¬b) (rule (¬¬))
(¬(a ∩ b) ⇒ (¬a ∪ ¬b)) (rule (⇒)).

6.2 Search for Proofs and Decomposition Trees

The main advantage of the Gentzen style proof systems lies not in a way we
generate proofs in them, but in the way we can search for proofs in them.
That such proof searches happens to be deterministic and automatic. Before
we describe a general proof search procedure for RS let us look at few simple
examples. Consider now a formula A of the form of another de Morgan law

(¬(a ∪ b) ⇒ (¬a ∩ ¬b)).

Obviously it should have a proof in RS as we want it to be, and will prove later
to be complete. The search for the proof consists of building a certain tree.
We call it a decomposition tree, to be defined formally later. We proceed as
follows.
Observe that the main connective of A is ⇒. So, if A had a proof in RS it
would have come from the only possible rule used in its last step, namely the
rule (⇒) applied to its premiss, namely a sequence ¬¬(a ∪ b), (¬a ∩ ¬b). So the
last step in the proof of A would look as follows.

(¬(a ∪ b) ⇒ (¬a ∩ ¬b))

| (⇒)
¬¬(a ∪ b), (¬a ∩ ¬b)

251
Now, if the sequence ¬¬(a ∪ b), (¬a ∩ ¬b) (and hence also the formula A) had a
proof in RS its only step at this stage would have been the application of the
rule (¬¬) to a sequence (a ∪ b), (¬a ∩ ¬b). So, if A had a proof, its last two steps
would have been:
(¬(a ∪ b) ⇒ (¬a ∩ ¬b))
| (⇒)
¬¬(a ∪ b), (¬a ∩ ¬b)
| (¬¬)
(a ∪ b), (¬a ∩ ¬b)
Again, if the sequence (a ∪ b), (¬a ∩ ¬b) had a proof in RS its only step at this
stage would have been the application of the rule (∪) to a sequence a, b, (¬a∩¬b).
So, if A had a proof, its last three steps would have been as follows.

(¬(a ∪ b) ⇒ (¬a ∩ ¬b))

| (⇒)
¬¬(a ∪ b), (¬a ∩ ¬b)
| (¬ ¬)
(a ∪ b), (¬a ∩ ¬b)
| (∪)
a, b, (¬a ∩ ¬b)
Now, if the sequence a, b, (¬a ∩ ¬b) had a proof in RS its only step at this stage
would have been the application of the rule (∩) to the sequences a, b, ¬a and
a, b, ¬b as its left and right premisses, respectively. Both sequences are axioms
and the following tree is a proof of A in RS.

(¬(a ∪ b) ⇒ (¬a ∩ ¬b))

| (⇒)
¬¬(a ∪ b), (¬a ∩ ¬b)
| (¬¬)
(a ∪ b), (¬a ∩ ¬b)
| (∪)
a, b, (¬a ∩ ¬b)
^
(∩)

a, b, ¬a a, b, ¬b

252
From the above proof tree of A we construct, if we want, its formal proof, written
in a vertical manner, by writing the two axioms, which form the two premisses of
the rule (∩) one above the other. All other sequences remain the same. I.e. the
following sequence of elements of F ∗ is a formal proof of (¬(a∪b) ⇒ (¬a∩¬b))
in RS.

a, b, ¬b

a, b, ¬a
a, b, (¬a ∩ ¬b)
(a ∪ b), (¬a ∩ ¬b)
¬¬(a ∪ b), (¬a ∩ ¬b)
(¬(a ∪ b) ⇒ (¬a ∩ ¬b))

Consider now a formula A of the form

(((a ⇒ b) ∩ ¬c) ∪ (a ⇒ c)).

Observe that the main connective of A is ∪. So, if A had a proof in RS it would

have come from the only possible rule used in its last step, namely the rule (∪)
applied to a sequence ((a ⇒ b) ∩ ¬c), (a ⇒ c). So the last step in the proof of A
would have been:
(((a ⇒ b) ∩ ¬c) ∪ (a ⇒ c))
| (∪)
((a ⇒ b) ∩ ¬c), (a ⇒ c)
Now, if the sequence ((a ⇒ b) ∩ ¬c), (a ⇒ c) (and hence also had our formula )
had a proof in RS its only step at this stage would have been the application
of the rule (∩) to the sequences (a ⇒ b), (a ⇒ c) and ¬c, (a ⇒ c) as its left and
right premisses, respectively. So, if A had a proof, its last two steps would have
been:

(((a ⇒ b) ∩ ¬c) ∪ (a ⇒ c))

| (∪)
((a ⇒ b) ∩ ¬c), (a ⇒ c)
^
(∩)

¬c, (a ⇒ c)
(a ⇒ b), (a ⇒ c)

253
Now, if the sequences (a ⇒ b), (a ⇒ c) and ¬c, (a ⇒ c) had proofs in RS, then
their last, and the only steps would have been the the separate application of
the rule (⇒) to the sequences ¬a, b, (a ⇒ c) and ¬c, ¬a, c, respectively. The
sequence ¬c, ¬a, c is an axiom, so we stop the search on this branch. The
sequence ¬a, b, (a ⇒ c) is not an axiom, so the search continues. In this case
we can go one step further: if ¬a, b, (a ⇒ c) had a proof it would have been
only by the application of the rule (⇒) to a sequence ¬a, b, ¬a, c which is not
an axiom and the search ends. The tree generated by this search is called a
decomposition tree and is the following.

(((a ⇒ b) ∩ ¬c) ∪ (a ⇒ c))

| (∪)
((a ⇒ b) ∩ ¬c), (a ⇒ c)
^
(∩)

¬c, (a ⇒ c)
(a ⇒ b), (a ⇒ c)
| (⇒)
| (⇒) ¬c, ¬a, c
¬a, b, (a ⇒ c)
| (⇒)
¬a, b, ¬a, c

The tree generated by this search contains a non-axiom leaf, so by definition,

it is not a proof.

Decomposition Rules and Trees

The process of searching for the proof of a formula A in RS consists of building
a certain tree, called a decomposition tree whose root is the formula A, nodes
correspond to sequences which are conclusions of certain rules (and those rules
are well defined at each step by the way the node is built), and leafs are axioms or
are sequences of a non- axiom literals. We prove that each formula A generates
its unique, finite decomposition tree, TA such that if all its leafs are axioms,
the tree constitutes the proof of A in RS. If there is a leaf of TA that is not
an axiom, the tree is not a proof, moreover, the proof of A does not exist.
Before we give a proper definition of the proof search procedure by building a
decomposition tree we list few important observations about the structure of
the rules of the system RS.

254
Introduction of Connectives
The rules of RS are defined in such a way that each of them introduces
a new logical connective, or a negation of a connective to a sequence in
its domain (rules (∪), (⇒), (∩)) or a negation of a new logical connective
(rules (¬ ∪), (¬ ∩), (¬ ⇒), (¬ ¬)).
0
The rule (∪) introduces a new connective ∪ to a sequence Γ , A, B, ∆ and
0
it becomes, after the application of the rule, a sequence Γ , (A ∪ B), ∆.
Hence a name for this rule is (∪).
The rule (¬∪) introduces a negation of a connective, ¬∪ by combining
0 0
sequences Γ , ¬A, ∆ and Γ , ¬B, ∆ into one sequence (conclusion of the
0
rule) Γ , ¬(A ∪ B), ∆. Hence a name for this rule is (¬∪).
The same applies to all remaining rules of RS, hence their names say which
connective, or the negation of which connective has been introduced by
the particular rule.
Decomposition Rules
Building a proof search decomposition tree consists of using the inference
rules in an inverse order; we transform them into rules that transform
a conclusion into its premisses. We call such rules the decomposition
rules. Here are all of RS decomposition rules.

Disjunction decomposition rules

0 0
Γ , (A ∪ B), ∆ Γ , ¬(A ∪ B), ∆
(∪) , (¬ ∪)
Γ0 , A, B, ∆ Γ , ¬A, ∆ : Γ0 , ¬B, ∆
0

Conjunction decomposition rules

0 0
Γ , (A ∩ B), ∆ Γ , ¬(A ∩ B), ∆
(∩) , (¬ ∩)
Γ , A, ∆ ; Γ0 , B, ∆
0
Γ0 , ¬A, ¬B, ∆
Implication decomposition rules
0 0
Γ , (A ⇒ B), ∆ Γ , ¬(A ⇒ B), ∆
(⇒) , (¬ ⇒) 0
Γ0 , ¬A, B, ∆ Γ , A, ∆ : Γ0 , ¬B, ∆
Negation decomposition rule
0
Γ , ¬¬A, ∆
(¬ ¬)
Γ0 , A, ∆
0
where Γ ∈ LT ∗ , ∆ ∈ F ∗ , A, B ∈ F.

We write the decomposition rules in a visual tree form as follows.

255
 Tree Decomposition Rules

(∪) rule
0
Γ , (A ∪ B), ∆
| (∪)
0
Γ , A, B, ∆

(¬ ∪) rule

0
Γ , ¬(A ∪ B), ∆
^
(¬ ∪)

0 0
Γ , ¬A, ∆ Γ , ¬B, ∆

(∩) rule:

0
Γ , (A ∩ B), ∆
^
(∩)

0 0
Γ , A, ∆ Γ , B, ∆

(¬ ∩) rule:

0
Γ , ¬(A ∩ B), ∆
| (¬ ∩)
0
Γ , ¬A, ¬B, ∆

(⇒) rule:

0
Γ , (A ⇒ B), ∆

| (∪)
0
Γ , ¬A, B, ∆

256
(¬ ⇒) rule:

0
Γ , ¬(A ⇒ B), ∆
^
(¬ ⇒)

0 0
Γ , A, ∆ Γ , ¬B, ∆

(¬ ¬) rule:
0
Γ , ¬¬A, ∆
| (¬ ¬)
0
Γ , A, ∆

Observe that we use the same names for the inference and decomposition rules,
as once the we have built the decomposition tree (with use of the decomposition
rules) with all leaves being axioms, it constitutes a proof of A in RS with
branches labeled by the proper inference rules.
Now we still need to introduce few useful definitions and observations.

Definition 6.2
1. A sequence Γ is indecomposable if and only if Γ ∈ LT ∗ .
2. A formula A is decomposable if and only if A ∈ F − LT .
3. A sequence Γ is decomposable if and only if it contains a decomposable
formula.

Directly from the definition 6.8 we have three simple, but important observa-
tions.

Fact 6.2
1. For any decomposable sequence Γ, i.e. for any Γ 6∈ LT ∗ there is exactly
one decomposition rule that can be applied to it. This rule is determined by the
first decomposable formula in Γ, and by the main connective of that formula.
2. If the main connective of the first decomposable formula is ∪, ∩, or ⇒, then
the decomposition rule determined by it is (∪), (∩), or (⇒), respectively.
3. If the main connective of the first decomposable formula is ¬, then the de-
composition rule determined by it is determined by the second connective of the
formula. If the second connective is ∪, ∩, ¬, or ⇒, then corresponding decom-
position rule is (¬∪), (¬∩), (¬¬) and (¬ ⇒).

257
Directly from the Fact 6.2 we we have the following lemma.

Lemma 6.1 (Unique Decomposition)

For any sequence Γ ∈ F ∗ ,
Γ ∈ LT ∗ or Γ is in the domain of only one of the RS Decomposition Rules.

Now we define formally, for any formula A ∈ F and Γ ∈ F ∗ their decompositions

trees. The decomposition tree for for the formula A is a particular case (one
element sequence) of the tree for a sequence Γ.

Definition 6.3 (Decomposition Tree TA )

For each formula A ∈ F, its decomposition tree TA is a tree build as follows.
Step 1. The formula A is the root of TA and for any node Γ of the tree we
follow the steps below.
Step 2. If Γ is indecomposable, then Γ becomes a leaf of the tree.
Step 3. If Γ is decomposable, then we traverse Γ from left to right to identify
the first decomposable formula B and identify the decomposition rule deter-
mined by the main connective of B. In case of a one premisses rule we put
is premise as a leaf; in case of a two premisses rule we put its left and right
premisses as the left and right leaves respectively.
Step 4. We repeat Step 2 and Step 3 until we obtain only leaves.

We now prove the following Decomposition Tree Theorem 6.1. This Theorem
provides a crucial step in the proof of the Completeness Theorem for RS.

Theorem 6.1 (Decomposition Tree)

For any sequence Γ ∈ F ∗ the following conditions hold.

1. TΓ is finite and unique.

2. TΓ is a proof of Γ in RS if and only if all its leafs are axioms.

3. 6 `RS if and only if TΓ has a non- axiom leaf.

Proof
The tree TΓ is unique by the Unique Decomposition Lemma 6.1. It is finite
because there is a finite number of logical connectives in Γ and all decomposition
rules diminish the number of connectives. If the tree has a non- axiom leaf it is
not a proof by definition. By the its uniqueness it also means that the proof
does not exist.

258
Exercise 6.2
Construct a decomposition tree TA of the following formula A.

A = ((a ∪ b) ⇒ ¬a) ∪ (¬a ⇒ ¬c))

Solution
The formula A forms a one element decomposable sequence. The first decom-
position rule used is determined by its main connective. We put a box around
it, to make it more visible. The first and only rule applied is (∪) and we can
write the first segment of our decomposition tree TA :

TA

((a ∪ b) ⇒ ¬a) ∪ (¬a ⇒ ¬c))

| (∪)

((a ∪ b) ⇒ ¬a), (¬a ⇒ ¬c)

Now we decompose the sequence ((a ∪ b) ⇒ ¬a), (¬a ⇒ ¬c)). It is a decom-

posable sequence with the first, decomposable formula ((a ∪ b) ⇒ ¬a). The
next step of the construction of our decomposition tree is determined by its
main connective ⇒ (we put the box around it), hence the only rule determined
by the sequence is (⇒). The second stage of the decomposition tree is now as
follows.

TA

((a ∪ b) ⇒ ¬a) ∪ (¬a ⇒ ¬c))

| (∪)

((a ∪ b) ⇒ ¬a), (¬a ⇒ ¬c)

| (⇒)

¬(a ∪ b), ¬a, (¬a ⇒ ¬c)

The next sequence to decompose is the sequence ¬(a ∪ b), ¬a, (¬a ⇒ ¬c).
The first decomposable formula is ¬(a∪b). Its main connective is ¬, so determine
the appropriate decomposition rule we have to examine next connective, which
is ∪.
The rule determine by this stage of decomposition is (¬ ∪) and now the next
stage of the decomposition tree TA is as follows.

259
 TA

((a ∪ b) ⇒ ¬a) ∪ (¬a ⇒ ¬c))

| (∪)
((a ∪ b) ⇒ ¬a), (¬a ⇒ ¬c)
| (⇒)
¬ (a ∪ b), ¬a, (¬a ⇒ ¬c)
^
(¬ ∪)

¬a, ¬a, (¬a ⇒ ¬c) ¬b, ¬a, (¬a ⇒ ¬c)

Now we have two decomposable sequences: ¬a, ¬a, (¬a ⇒ ¬c) and ¬b, ¬a, (¬a ⇒
¬c). They both happen to have the same first decomposable formula (¬a ⇒ ¬c).
We decompose it simultenously and obtain the following:

TA

((a ∪ b) ⇒ ¬a) ∪ (¬a ⇒ ¬c))

| (∪)
((a ∪ b) ⇒ ¬a), (¬a ⇒ ¬c)
| (⇒)
¬ (a ∪ b), ¬a, (¬a ⇒ ¬c)
^
(¬∪)

¬a, ¬a, (¬a ⇒ ¬c) ¬b, ¬a, (¬a ⇒ ¬c)

| (⇒) | (⇒)
¬a, ¬a, ¬¬a, ¬c ¬b, ¬a, ¬¬a, ¬c

It is easy to see that we need only one more step to complete the process of
constructing the unique decomposition tree of TA , namely, by decomposing the
sequences: ¬a, ¬a, ¬¬a, ¬c and ¬b, ¬a, ¬¬a, ¬c.
The complete decomposition tree TA is:

260
 TA
((a ∪ b) ⇒ ¬a) ∪ (¬a ⇒ ¬c))
| (∪)
((a ∪ b) ⇒ ¬a), (¬a ⇒ ¬c)
| (⇒)
¬ (a ∪ b), ¬a, (¬a ⇒ ¬c)
^
(¬∪)

¬a, ¬a, (¬a ⇒ ¬c) ¬b, ¬a, (¬a ⇒ ¬c)

| (⇒) | (⇒)
¬a, ¬a, ¬¬ a, ¬c ¬b, ¬a, ¬¬ a, ¬c
| (¬¬) | (¬¬)
¬a, ¬a, a, ¬c ¬b, ¬a, a, ¬c

All leafs are axioms, the tree represents a proof of A in RS

Exercise 6.3
Prove that the formula A = (((a ⇒ b) ∩ ¬c) ∪ (a ⇒ c)) is not provable in RS,
i.e.
6 `RS (((a ⇒ b) ∩ ¬c) ∪ (a ⇒ c)).

Solution
We construct the formula A decomposition tree as follows.

TA
(((a ⇒ b) ∩ ¬c) ∪ (a ⇒ c))
| (∪)
((a ⇒ b) ∩ ¬c), (a ⇒ c)
^
(∩)

(a ⇒ b), (a ⇒ c) ¬c, (a ⇒ c)
| (⇒) | (⇒)
¬a, b, (a ⇒ c) ¬c, ¬a, c
| (⇒)
¬a, b, ¬a, c

The above tree TA is unique by the Theorem 6.1 and represents the only possible
search for proof of the formula A = ((a ⇒ b) ∩ ¬c) ∪ (a ⇒ c)) in RS. It has a
non-axiom leaf, hence by Theorem 6.1 the proof of A in RS does not exists.

261
6.3 Strong Soundness and Completeness
Our main goal is to prove the Completeness Theorem for RS. The proof of
completeness presented here is due to Rasiowa and Sikorski, as is the proof
system RS. Their proof, and the proof system was inverted for the classical
predicate logic and was published in 1961. We present their predicate logic proof
system QRS together with the proof of its completeness in chapter 10. Both
completeness proofs, for propositional RS and predicate QRS proof systems,
are constructive as they are based on a direct construction of a counter model
for any unprovable formula. The construction of a counter model for a formula
A uses directly its decomposition tree TA . We call such constructed model a
counter model determined by the tree TA . Both proofs relay heavily of the
notion of a strong soundness. We define it now, adopting Chapter 4 general
definition to our semantics.

Definition 6.4 ( Strongly Sound Rules)

Given a proof system S = (L, E, LA, R) An inference rule r ∈ R of the form
P1 ; P2 ; .... ; Pm
(r)
C
is strongly sound (undef classical semantics) if the following condition holds
for all v : V AR −→ {T, F }

v |= {P1 , P2 , .Pm } if and only if v |= C. (6.14)

We say it less formally that a rule (r) is strongly sound if the conjunction of
its premisses is logically equivalent with the conclusion, i.e.

P1 ∩ P2 ∩ . . . ∩ Pm ≡ C. (6.15)

Definition 6.5 (Strongly Sound S)

A proof system S = (L, E, LA, R) is strongly sound (undef classical semantics)
if and only if all logical axioms LA are tautologies and all its rules of inference
r ∈ R are strongly sound.

Theorem 6.2 (Strong Soundness)

The proof system RS (6.18) is strongly sound.

Proof
The logical axioms (6.9), (6.10) are tautologies by Fact ??. We prove as an
example the strong soundness of two of inference rules: (∪) and (¬∪). Proofs
for all other rules follow the same patterns and are left as an exercise. By
definition 6.4 of strong soundness we have to show the condition (8.77). Written

262
formally it says that we have to show that that if P1 , P2 are premisses of a given
rule and C is its conclusion, then for all truth assignments v : V AR −→ {T, F },
v ∗ (P1 ) = v ∗ (C) in case of one premiss rule, and v ∗ (P1 ) ∩ v ∗ (P2 ) = v ∗ (C), in
case of a two premisses rule. Consider the rule (∪).
0
Γ , A, B, ∆
(∪) 0 .
Γ , (A ∪ B), ∆

By the definition:
0 0 0
v ∗ (Γ , A, B, ∆) = v ∗ (δ{Γ0 ,A,B,∆} ) = v ∗ (Γ ) ∪ v ∗ (A) ∪ v ∗ (B) ∪ v ∗ (∆) = v ∗ (Γ ) ∪
0
v ∗ (A ∪ B) ∪ v ∗ (∆) = v ∗ (δ{Γ0 ,(A∪B),∆} ) = v ∗ (Γ , (A ∪ B), ∆).

Consider the rule (¬∪).

0 0
Γ , ¬A, ∆ : Γ , ¬B, ∆
(¬∪) .
Γ0 , ¬(A ∪ B), ∆

By the definition:
0 0 0 0
v ∗ (Γ , ¬A, ∆) ∩ v ∗ (Γ , ¬B, ∆) = (v ∗ (Γ ) ∪ v ∗ (¬A) ∪ v ∗ (∆)) ∩ (v ∗ (Γ ) ∪ v ∗ (¬B) ∪
0 0
v ∗ (∆)) = (v ∗ (Γ , ∆) ∪ v ∗ (¬A)) ∩ (v ∗ (Γ , ∆) ∪ v ∗ (¬B)) = by distributivity =
0 0
(v ∗ (Γ , ∆) ∪ (v ∗ (¬A) ∩ v ∗ (¬B)) = v ∗ (Γ ) ∪ v ∗ (∆) ∪ (v ∗ (¬A ∩ ¬B)) = by the log-
0
ical equivalence of (¬A ∩ ¬B) and ¬(A ∪ B) = v ∗ (δ{Γ0 ,¬(A∪B),∆} = v ∗ (Γ , ¬(A ∪
B), ∆)).

Observe that the strong soundness implies soundness (not only by name!), hence
we have also proved the following.

Theorem 6.3 (Soundness for RS)

For any Γ ∈ F ∗ ,
if `RS Γ, then |= Γ. In particular, for any A ∈ F, if `RS A, then |= A.

We have just proved (Theorem 6.2) that all the rules of inference of RS of are
strongly sound, i.e. C ≡ P and C ≡ P1 ∩ P2 . The strong soundness of the rules
means that if at least one of premisses of a rule is false, so is its conclusion. Hence
given a formula A, such that its TA has a branch ending with a non-axiom leaf.
By Strong Soundness Theorem 6.2, any v that make this non-axiom leaf false
also falsifies all sequences on that branch, and hence falsifies the formula A. This
means that any v, such that it falsifies a non-axiom leaf is a counter-model
for A. We have hence proved the following.

Theorem 6.4 (Counter Model)

263
Given a formula A ∈ F such that its decomposition tree TA contains a non-
axiom leaf LA . Any truth assignment v that falsifies the non-axiom leaf LA is
a counter model for A. We call it a counter-model for A determined by the
decomposition tree TA .

Here is a simple example explaining how the construction of a counter-model

determined by the decomposition tree of a works. Consider a tree

TA

(((a ⇒ b) ∩ ¬c) ∪ (a ⇒ c))

| (∪)
((a ⇒ b) ∩ ¬c), (a ⇒ c)
^
(∩)

(a ⇒ b), (a ⇒ c) ¬c, (a ⇒ c)
| (⇒) | (⇒)
¬a, b, (a ⇒ c) ¬c, ¬a, c
| (⇒)
¬a, b, ¬a, c

The tree TA has a non-axiom leaf LA : ¬a, b, ¬a, c. The truth assignment
v : V AR −→ {T, F } that falsifies the leaf ¬a, b, ¬a, c must be such that
v ∗ (¬a, b, ¬a, c) = v ∗ (¬a)∪v ∗ (b)∪v ∗ (¬a)∪v ∗ (c) = ¬v(a)∪v(b)∪¬v(a)∪v(c) = F,
i.e. v must be such that ¬v(a) ∪ v(b) ∪ ¬v(a) ∪ v(c) = F. We hence get that
v(a) = T, v(b) = F, v(c) = F. By the Counter Model Theorem 6.4, the truth
assignment v determined by the non-axiom leaf also falsifies the formula A, i.e.
we proved that v is a counter model for A and

6|= (((a ⇒ b) ∩ ¬c) ∪ (a ⇒ c)).

The Counter Model Theorem 6.4, says that the logical value F determined by
the evaluation a non-axiom leaf ”climbs” the decomposition tree. We picture it
as follows.

TA

(((a ⇒ b) ∩ ¬c) ∪ (a ⇒ c)) = F

| (∪)
((a ⇒ b) ∩ ¬c), (a ⇒ c) = F
^
(∩)

264
 (a ⇒ b), (a ⇒ c) = F ¬c, (a ⇒ c)
| (⇒) | (⇒)
¬a, b, (a ⇒ c) = F ¬c, ¬a, c
| (⇒) axiom
¬a, b, ¬a, c = F

Observe that the same counter model construction applies to any other non-
axiom leaf of TA , if exists. The other non-axiom leaf of TA defines another
evaluation of the non- axiom leaf to F that also ”climbs the tree” and hence
defines another counter- model for a formula A. By Counter Model 6.4 all
possible restricted counter-models for A are those determined by its all non-
axioms leaves.
In our case the tree TA has only one non-axiom leaf, and hence the formula
(((a ⇒ b) ∩ ¬c) ∪ (a ⇒ c)) only only one restricted counter model.
Our main goal is to prove the Completeness Theorem for RS. We prove first
the Completeness Theorem for formulas A ∈ F and then we generalize it to any
sequences Γ ∈ F ∗ .

Theorem 6.5 (Completeness Theorem)

For any formula A ∈ F,
1. `RS A if and only if |= A, and for any Γ ∈ F ∗ ,
2. `RS Γ if and only if |= Γ.

Proof
Case 1. We have already proved the Soundness Theorem 6.3, so we need to
prove only the completeness part of it, namely to prove the implication:
if |= A, then `RS A. (6.16)
We prove instead of the opposite implication:
if 6 `RS A then 6|= A. (6.17)

Assume that A is any formula is such that 6 `RS A. By the Decomposition Tree
Theorem 6.1 the tree TA contains a non-axiom leaf LA . We use the non-axiom
leaf LA to define a truth assignment v : V AR −→ {T, F } which falsifies it
as follows:

 F if a appears in LA
v(a) = T if ¬a appears in LA
any value if a does not appear in LA


By the Counter Model Theorem 6.4we have that v also falsifies the formula
A. We proved that
6|= A

265
and it ends the proof of the case 1.
Case 2. Assume that Γ ∈ F ∗ is any sequence such that 6 `RS Γ. But obviously,
`RS Γ if and only if `RS δΓ , where δΓ is any disjunction of all formulas of Γ.
So 6 `RS Γ if and only if 6 `RS δΓ and by already proven Case 1, 6|= δΓ what is
obviously equivalent to 6|= Γ. This ends the proof of Case 2 and Completeness
Theorem.

6.4 Proof Systems RS1 and RS2

We present here a two modifications of the system RS as an exercise of impor-
tance of paying close attention to the syntax. Proof systems might be, as all
presented here RS type systems are, semantically identical, nevertheless they
are very different proof systems.
Language of RS1 is the same as the language of RS, i.e.

L = L{¬,⇒,∪,∩} .

Rules of inference of RS1 operate as rules of RS on finite sequences of for-

mulas and we adopt
E = F∗
as the set of expressions of RS1 . We denote them, as before, by Γ, ∆, Σ . . . ,
with indices if necessary.
The proof system RS1 contains seven inference rules, denoted by the same
symbols as the rules of RS, namely (∪), (¬∪), (∩), (¬∩), (⇒), (¬ ⇒
), (¬¬).
The inference rules of RS1 are quite similar to the rules of RS Look at them
carefully to see where lies the difference.
Reminder Any propositional variable, or a negation of propositional variable
is called a literal. The set LT = V AR ∪ {¬a : a ∈ V AR} is called a set of
all propositional literals. The variables are called positive literals. Negations of
0 0 0
variables are called negative literals. We denote, as before, by Γ , ∆ , Σ , . . .
finite sequences (empty included) formed out of literals. We adopt all logical
axiom of RS as the axioms of RS1, i.e. logical axioms LA of RS1 are:
0 0 0
Γ1 , a, Γ2 , ¬a, Γ3 ,
0 0 0
Γ1 , ¬a, Γ2 , a, Γ3
where a ∈ V AR is any propositional variable.
We define the inference rules of RS1 as follows.

266
Disjunction rules
0 0 0
Γ, A, B, ∆ Γ, ¬A, ∆ : Γ, ¬B, ∆
(∪) , (¬ ∪) ,
Γ, (A ∪ B), ∆0 Γ, ¬ (A ∪ B), ∆0

Conjunction rules
0 0 0
Γ, A, ∆ ; Γ, B, ∆ Γ, ¬A, ¬B, ∆
(∩) , (¬ ∩) 0 ,
Γ, (A ∩ B), ∆0 Γ , ¬(A ∩ B), ∆

Implication rules
0 0 0
Γ, ¬A, B, ∆ Γ, A, ∆ : Γ, ¬B, ∆
(⇒) , (¬ ⇒) ,
Γ, (A ⇒ B), ∆0 Γ, ¬ (A ⇒ B), ∆0

Negation rule
0
Γ, A, ∆
(¬ ¬)
Γ, ¬ ¬ A, ∆0
0
where Γ ∈ F ∗ , ∆ ∈ LT ∗ , A, B ∈ F.

Proof System RS1

Formally we define the proof system RS1 as follows.

RS1 = (L{¬,⇒,∪,∩} , E, LA, R), (6.18)

where E = {Γ : Γ ∈ F ∗ }, LA is the set logical axioms and R is the set of rules

of inference defined above.

Exercise 6.4
Construct a proof in RS1 of a formula

A = (¬(a ∩ b) ⇒ (¬a ∪ ¬b)).

Solution
The decomposition tree below is a proof of A in RS1 as all its leaves are axioms.

TA

(¬(a ∩ b) ⇒ (¬a ∪ ¬b))

| (⇒)
(¬¬(a ∩ b), (¬a ∪ ¬b)

267
 | (∪)
¬¬(a ∩ b), ¬a, ¬b
| (¬¬)
(a ∩ b), ¬a, ¬b
^
(∩)

a, ¬a, ¬b b, ¬a, ¬b

Exercise 6.5
Prove that RS1 is strongly sound.

Solution
0
Observe that the system RS1 is obtained from RS by changing the sequence Γ
0
into Γ and the sequence ∆ into ∆ in all of the rules of inference of RS. These
changes do not influence the essence of proof of strong soundness of the rules
0 0
of RS. One has just to replace the sequence Γ by Γ and the sequence ∆ by ∆
in the proof of strong soundness of each rule of RS to obtain a corresponding
proof of strong soundness of corresponding rule of RS1. We do it, for example
for the rule (∪) of RS1. Consider the rule (∪) of RS1:

0
Γ, A, B, ∆
(∪) .
Γ, (A ∪ B), ∆0
We evaluate:
0 0
v ∗ (Γ, A, B, ∆ ) = v ∗ (δ{Γ,A,B,∆0 } ) = v ∗ (Γ) ∪ v ∗ (A) ∪ v ∗ (B) ∪ v ∗ (∆ )
0 0
= v ∗ (Γ) ∪ v ∗ (A ∪ B) ∪ v ∗ (∆ ) = v ∗ (δ{Γ,(A∪B),∆0 } ) = v ∗ (Γ, (A ∪ B), ∆ ).

Exercise 6.6
Define in your own words, for any formula A ∈ F the decomposition tree TA
in RS1.

Solution
The definition of the decomposition tree TA is again, it its essence similar
to the one for RS except for the changes which reflect the differences in the
corresponding rules of inference. We follow now the following steps.
Step 1 Decompose A using a rule defined by its main connective.
Step 2 Traverse resulting sequence Γ on the new node of the tree from right
to left and find the first decomposable formula.
Step 3 Repeat Step 1 and Step 2 until there is no more decomposable
formulas. End of tree construction.

268
Exercise 6.7
Prove the following Completeness Theorem for RS1.

Theorem 6.6
For any formula A ∈ F,
1. `RS1 A if and only if |= A, and for any Γ ∈ F ∗ ,
2. `RS1 Γ if and only if |= Γ.

Solution Part 1.
Observe that directly from the definition of the uniqueness of the decomposition
tree TA we have that the following holds.

Fact 6.3
The decomposition tree TA is a proof if and only if all leaves are axioms and
the proof does not exist otherwise, i.e. we have that 6 `RS1 A if and only if
there is a non- axiom leaf on TA .

The Fact 6.3 together with strong soundness of the rules of inference of RS1
justify the correctness of construction of a counter-model generated by a the
a non- axiom leaf and hence the correctness of the following proof of the
Completeness Theorem.
We prove, as we did in case of RS the implication

if 6 `RS1 A then 6|= A.

Assume that A is any formula such that 6 `RS1 A. By the Fact 6.3 the decom-
position tree TA contains a non-axiom leaf LA . We use the non-axiom leaf LA
and define a truth assignment v which falsifies A, as follows:

 F if a appears in LA
v(a) = T if ¬a appears in LA
any value if a does not appear in LA .


This proves, by the strong soundness of RS1, that 6|= A.

The proof of Part 2. is identical to the proof in RS case.

Proof System RS2 (6.19)

0
System RS2 is a proof system obtained from RS by changing the sequences Γ
into Γ in all of the rules of inference of RS. The logical axioms LA remain
the same. Observe that now the decomposition tree may not be unique

Exercise 6.8

269
Construct two decomposition trees in RS2 of the formula

A = (¬(¬a ⇒ (a ∩ ¬b)) ⇒ (¬a ∩ (¬a ∪ ¬b))).

Solution
Here are two out of many more decomposition trees.

T1A

(¬(¬a => (a ∩ ¬b)) => (¬a ∩ (¬a ∪ ¬b)))

| (⇒)
¬¬(¬a => (a ∩ ¬b)), (¬a ∩ (¬a ∪ ¬b))
| (¬¬)
(¬a => (a ∩ ¬b)), (¬a ∩ (¬a ∪ ¬b))
| (⇒)
¬¬a, (a ∩ ¬b), (¬a ∩ (¬a ∪ ¬b))
| (¬¬)
a, (a ∩ ¬b), (¬a ∩ (¬a ∪ ¬b))
^
(∩)

a, a, (¬a ∩ (¬a ∪ ¬b)) a, ¬b, (¬a ∩ (¬a ∪ ¬b))

^ ^
(∩) (∩)

a, a.¬a, (¬a ∪ ¬b) a, a, (¬a ∪ ¬b) a, ¬b, ¬a a, ¬b, (¬a∪¬b)

| (∪) | (∪) axiom | (∪)

a, a.¬a, ¬a, ¬b a, a, ¬a, ¬b a, ¬b, ¬a, ¬b
axiom axiom axiom

The other tree is:

T2A

(¬(¬a => (a ∩ ¬b)) => (¬a ∩ (¬a ∪ ¬b)))

| (⇒)
¬¬(¬a => (a ∩ ¬b)), (¬a ∩ (¬a ∪ ¬b))
| (¬¬)
(¬a => (a ∩ ¬b)), (¬a ∩ (¬a ∪ ¬b))
^
(∩)

270
 (¬a => (a ∩ ¬b)), ¬a
(¬a => (a ∩ ¬b)), (¬a ∪ ¬b)
| (⇒)
(¬¬a, (a ∩ ¬b)), ¬a | (∪)
| (¬¬) (¬a => (a ∩ ¬b)), ¬a, ¬b
a, (a ∩ ¬b), ¬a | (⇒)
(¬¬a, (a ∩ ¬b), ¬a, ¬b
^
(∩)
| (¬¬)
a, (a ∩ ¬b), ¬a, ¬b
a, a, ¬a a, ¬b, ¬a ^
(∩)
axiom axiom

a, a, ¬a, ¬b a, ¬b, ¬a, ¬b

axiom axiom

Exercise 6.9
Explain why the system RS2 is strongly sound. You can use the strong
soundness of the system RS.

Solution
The only one difference between RS and RS2 is that in RS2 each inference rule
has at the beginning a sequence of any formulas, not only of literals, as in RS.
So there are many ways to apply the decomposition rules while constructing the
decomposition tree, but it does not affect strong soundness, since for all rules
of RS2 premisses and conclusions are still logically equivalent as they were in
RS .
Consider, for example, RS2 rule
Γ, A, B, ∆
(∪) .
Γ, (A ∪ B), ∆
We evaluate v ∗ (Γ, A, B, ∆) = v ∗ (Γ) ∪ v ∗ (A) ∪ v ∗ (B) ∪ v ∗ (∆) = v ∗ (Γ) ∪ v ∗ (A ∪
B) ∪ v ∗ (∆) = v ∗ (Γ, (A ∪ B), ∆). Similarly, as in RS, we show all other rules of
RS2 to be strongly sound, thus RS2 is sound.

Exercise 6.10
Define shortly, in your own words, for any formula A, its decomposition tree
TA in RS2. Justify why your definition is correct. Show that in RS2 the
decomposition tree for some formula A may not be unique.

Solution
Given a formula A. The decomposition tree TA can be defined as follows. It has

271
A as a root. For each node, if there is a rule of RS2 which conclusion has the
same form as node sequence, i.e. there is a decomposition rule to be applied,
then the node has children that are premises of the rule. If the node consists
only of literals (i.e. there is no decomposition rule to be applied), then it does
not have any children. The last statement define a termination condition for
the tree TA .
This definition defines correctly the decomposition tree TA as it identifies and
uses appropriate decomposition rules. Since all rules of inference of RS2 have
a sequence Γ instead of Γ0 as it was in RS, the choice of the decomposition rule
for a node may not unique. For example consider a node (a => b), (b ∪ a).
Γ in the RS2 rules may be a sequence of formulas, not only literals, so for the
node (a => b), (b ∪ a) we can choose as a decomposition rule either (=>) or
(∪). This leads to a non-unique tree.

Exercise 6.11
Prove the following Completeness Theorem for RS2.

Theorem 6.7
For any formula A ∈ F,
1. `RS2 A if and only if |= A, and for any Γ ∈ F ∗ ,
2. `RS2 Γ if and only if |= Γ.

Solution
We need to prove the completeness part only, as the Soundness has been already
proved, i.e. we have to prove the implication (Part 1): for any formula A,

if 6`RS2 A then 6|= A.

Assume 6`RS2 A. Then every decomposition tree of A has at least one non-
axiom leaf. Otherwise, there would exist a tree with all axiom leaves and it
would be a proof for A. Let TA be a set of all decomposition trees of A. We
choose an arbitrary TA ∈ TA with at least one non-axiom leaf LA . We use the
non-axiom leaf LA to define a truth assignment v which falsifies A, as follows:

 F if a appears in LA
v(a) = T if ¬a appears in LA
any value if a does not appear in LA


The value for a sequence that corresponds to the leaf in is F. Since, because of
the strong soundness F ”climbs” the tree, we found a counter-model for A. This
proves that 6|= A. Part 2. proof is identical to the proof in RS case.

Exercise 6.12

272
Write a procedure T REEA such that for any formula A of RS2 it produces its
unique decomposition tree.

Solution
Here is the procedure.
Procedure T REEA (Formula A, Tree T)
{
B = ChoseLef tM ostF ormula(A) // Choose the left most formula that is
not a literal
c = M ainConnective(B) // Find the main connective of B
R = F indRule(c)// Find the rule which conclusion that has this connective
P = P remises(R)// Get the premises for this rule
AddT oT ree(A, P )// add premises as children of A to the tree
For all p in P // go through all premises
T REEA (p, T ) // build subtrees for each premiss
}

Exercise 6.13
Prove completeness of your Procedure T REEA .

Solution
Procedure T REEA provides a unique tree, since it always chooses the most
left indecomposable formula for a choice of a decomposition rule and there is
only one such rule. This procedure is equivalent to RS system, since with the
decomposition rules of RS the most left decomposable formula is always chosen.
The proof RS system is complete, thus this Procedure T REEA is complete.

6.5 Gentzen Sequent Systems GL, G, LK

Gentzen proof systems GL and G for the classical propositional logic presented
here are inspired by and all are versions of the original (1934) Gentzen system
LK. Their axioms, the rules of inference of the proof system considered here
operate, as the original Gentzen system LK, on expressions called by Gentzen
sequents, hence the name Gentzen sequent proof systems, called also Gentzen
sequent calculus, or sequents calculus. The original system LK is presented and
discussed in detail in section 6.7.

6.5.1 Gentzen Sequent Systems GL and G

The system GL presented here is the most similar in its structure to the system
RS (6.18) and hence is the first to be considered. It admits a constructive

273
proof of the Completeness Theorem that is very similar to the proof of the
Completeness Theorem for the system RS.
Language of GL
We adopt a propositional language L = L{∪,∩,⇒,¬} with the set of formulas
denoted by F and we add a new symbol −→ called a Gentzen arrow, to it.
It means we consider formally a new language L1 = L{∪,∩,⇒,¬} ∪ {−→}. As
the next step we build expressions called sequents out of L1 . The sequents are
built out of finite sequences (empty included) of formulas, i.e. elements of F ∗
of L{∪,∩,⇒,¬} , and the additional sign −→.
We denote , as in the RS system, the finite sequences of formulas of of L{∪,∩,⇒,¬}
by Greek capital letters
Γ, ∆, Σ, . . . ,
with indices if necessary. We define a sequent as follows.

Definition 6.6 (Sequent)

For any Γ, ∆ ∈ F ∗ , the expression

Γ −→ ∆

is called a sequent. Γ is called the antecedent of the sequent, ∆ is called the

succedent, and each formula in Γ and ∆ is called a sequent-formula.

Intuitively, a sequent A1 , ..., An −→ B1 , ..., Bm (where n, m ≥ 1) means: if

A1 ∩ ... ∩ An then B1 ∪ ... ∪ Bm . The sequent A1 , ..., An −→ (where n ≥ 1)
means that A1 ∩ ... ∩ An yields a contradiction. The sequent −→ B1 , ..., Bm
(where m ≥ 1) means that B1 ∪ ... ∪ Bm is true. The empty sequent −→
means a contradiction.
Given non empty sequences Γ, ∆, we denote by σΓ any conjunction of all
formulas of Γ, and by δ∆ any disjunction of all formulas of ∆. The intuitive
semantics for a sequent Γ −→ ∆ (where Γ, ∆ are nonempty) is hence that it is
logically equivalent to the formula (σΓ ⇒ δ∆ ), i.e.

Γ −→ ∆ ≡ (σΓ ⇒ δ∆ ).

Formal semantics

Formally, let v : V AR −→ {T, F } be a truth assignment, v ∗ its extension to the

set of formulas F. We extend v ∗ to the set

SQ = { Γ −→ ∆ : Γ, ∆ ∈ F ∗ } (6.20)

of all sequents as follows.

274
Definition 6.7 For any sequent Γ −→ ∆ ∈ SQ,
v ∗ (Γ −→ ∆) = v ∗ (σΓ ) ⇒ v ∗ (δ∆ ).

In the case when Γ = ∅ we define: v ∗ ( −→ ∆) = T ⇒ v ∗ (δ∆ ). In the case

∆ = ∅ we define v ∗ (Γ −→ ) = v ∗ (σΓ ) ⇒ F.
Model
The sequent Γ −→ ∆ is satisfiable if there is a truth assignment v : V AR −→
{T, F } such that v ∗ (Γ −→ ∆) = T . Such a truth assignment is called a model
for Γ −→ ∆. We write
v |= Γ −→ ∆.

Counter- model
The sequent Γ −→ ∆ is falsifiable if there is a truth assignment v, such that
v ∗ (Γ −→ ∆) = F . In this case v is called a counter-model for Γ −→ ∆ and
we write it as

v 6|= Γ −→ ∆.

Tautology
The sequent Γ −→ ∆ is a tautology if v ∗ (Γ −→ ∆) = T for all truth
assignments v : V AR −→ {T, F } and we write
|= Γ −→ ∆.

Example 6.2
Let Γ −→ ∆ be a sequent
a, (b ∩ a) −→ ¬b, (b ⇒ a).
Any truth assignment v, such that v(a) = T and v(b) = T is a model for
Γ −→ ∆, i.e.
|= a, (b ∩ a) −→ ¬b, (b ⇒ a).

We verify it by performing the following computation.

v ∗ (a, (b ∩ a) −→ ¬b, (b ⇒ a)) = v ∗ (σ{a,(b∩a)} ) ⇒ v ∗ (δ{¬b,(b⇒a)} ) = v(a) ∩
(v(b) ∩ v(a)) ⇒ ¬v(b) ∪ (v(b) ⇒ v(a)) = T ∩ T capT ⇒ ¬T ∪ (T ⇒ T ) = T ⇒
(F ∪ T ) = T ⇒ T = T.
Observe that the only v for which v ∗ (Γ) = v ∗ (a, (b∩a) = T is the above v(a) = T
and v(b) = T that is a model for Γ −→ ∆. Hence it is impossible to find v
which would falsify it, what proves that Γ −→ ∆ is a tautology, i.e.
|= a, (b ∩ a) −→ ¬b, (b ⇒ a).

275
The Proof System GL

The rules of inference of GL are of the form:

P1 P1 ; P2
or ,
C C

where P1 , P2 and C are sequents. P1 , P2 are called premisses and C is called the
conclusion of the rule of inference. Each rule of inference introduces a new logical
connective to the antecedent or to the succedent of the conclusion sequent. We
denote the rule that introduces the logical connective ◦ to the antecedent of the
conclusion sequent P by (◦ →). The notation (→ ◦) means that the logical
connective is introduced to the succedent of the conclusion sequent P .
As our language contains the connectives: ∩, ∪, ⇒ and ¬, we are going to adopt
the following inference rules: (∩ →) and (→ ∩), (∪ →) and (→ ∪), (⇒→) and
(→⇒), and finally, (¬ →) and (→ ¬).

Definition 6.8
Finite sequences formed out of positive literals i.e. out of propositional vari-
ables are called indecomposable. We denote them as before by
0 0
Γ , ∆ , ...

with indices, if necessary.

A sequent is indecomposable if it is formed out of indecomposable sequences,
i.e. is of the form 0 0
Γ −→ ∆
0 0
for any Γ , ∆ ∈ V AR∗ .
0 0
Remark that now the symbols Γ , ∆ , . . . denote sequences of variables (pos-
itive literals), and not sequences of literals as in (RS.

Axioms of GL
As the axioms of GL we adopt any indecomposable sequent sequent which
contains a positive literal a (variable) that appears on both sides of the sequent
arrow −→, i.e any sequent of the form

Γ0 1 , a, Γ0 2 −→ ∆0 1 , a, ∆0 2 , (6.21)

for any a ∈ V AR and any sequences Γ0 1 , Γ0 2 , ∆0 1 , ∆0 2 ∈ V AR∗ .

Semantic Link

276
Consider axiom (6.21). Directly from the Definition 6.7 of semantics for bf GL
we evaluate (in shorthand notation), for any truth assignments v : V AR −→
{T, F }, the following (in shorthand notation).

v ∗ (Γ0 1 , a, Γ0 2 −→ ∆0 1 , a, ∆0 2 ) =

(σΓ0 1 ∩ a ∩ σΓ0 2 ) ⇒ (δ∆0 1 ∪ a ∪ δ∆0 2 ) = T.

The evaluation is correct because |= (((A ∩ a) ∩ B) ⇒ (C ∪ a) ∪ D))). We have
thus proved the following.

Fact 6.4
Logical axioms of GL are tautologies.

Inference Rules of GL (6.22)

We adopt the following rules of inference.

Conjunction rules
0 0 0 0
Γ , A, B, Γ −→ ∆ Γ −→ ∆, A, ∆ ; Γ −→ ∆, B, ∆
(∩ →) , (→ ∩) ,
Γ0 , (A ∩ B), Γ −→ ∆0 Γ −→ ∆, (A ∩ B), ∆0

Disjunction rules
0 0 0 0 0
Γ −→ ∆, A, B, ∆ Γ , A, Γ −→ ∆ ; Γ , B, Γ −→ ∆
(→ ∪) , (∪ →) ,
Γ −→ ∆, (A ∪ B), ∆0 Γ0 , (A ∪ B), Γ −→ ∆0

Implication rules
0 0
Γ , A, Γ −→ ∆, B, ∆
(→⇒) 0 ,
Γ , Γ −→ ∆, (A ⇒ B), ∆0
0 0 0 0
Γ , Γ −→ ∆, A, ∆ ; Γ , B, Γ −→ ∆, ∆
(⇒→) ,
Γ0 , (A ⇒ B), Γ −→ ∆, ∆0
Negation rules
0 0 0 0
Γ , Γ −→ ∆, A, ∆ Γ , A, Γ −→ ∆, ∆
(¬ →) , (→ ¬) .
Γ0 , ¬A, Γ −→ ∆, ∆0 Γ0 , Γ −→ ∆, ¬A, ∆0

Formally we define:

GL = (L, SQ, LA, (∪), (¬∪), (∩), (¬∩), (⇒), (¬ ⇒), (¬¬)), (6.23)

277
where SQ = { Γ −→ ∆ : Γ, ∆ ∈ F ∗ }, (∪), (¬∪), (∩), (¬∩), (⇒), (¬ ⇒), (¬¬)
are the inference rules defined above and AL are the logical axioms of the system
defined by the schema (6.21).
We define the notion of a bf formal proof in GL as in any proof system, i.e.,
by a formal proof of a sequent Γ −→ ∆ in the proof system GL we understand
any sequence
Γ1 −→ ∆1 , Γ2 −→ ∆2 , ...., Γn −→ ∆n
of sequents, such that Γ1 −→ ∆1 ∈ AL, Γn −→ ∆n = Γ −→ ∆, and for all i
(1 < i ≤ n) Γi −→ ∆i ∈ AL, or Γi −→ ∆i is a conclusion of one of the inference
rules of GL with all its premisses placed in the sequence Γ1 −→ ∆1 , ....Γi−1 −→
∆i−1 .
We write, as usual, `GL Γ −→ ∆ to denote that Γ −→ ∆ has a formal proof in
GL, or we write simply ` Γ −→ ∆ when the system GL is fixed.
We say that a formula A ∈ F, has a proof in GL and denote it by `GL A if the
sequent −→ A has a proof in GL, i.e. we define:

`GL A if and only if `GL −→ A. (6.24)

We write, however, the formal proofs in GL in a form of proof trees rather then
in a form of sequences of sequents.

Proof trees
A proof tree Γ −→ ∆ is a tree TΓ−→∆ satisfying the following conditions:
1. The topmost sequent, i.e the root of TΓ−→∆ is Γ −→ ∆.
2. All leaves are axioms.
3. The nodes are sequents such that each sequent on the tree follows from the
ones immediately preceding it by one of the rules.

We picture, and write our proof-trees, as we did in case of RS type systems,

with the node on the top, and leafs on the very bottom, instead of more common
way, where the leaves are on the top and root is on the bottom of the tree. We
also write the proof- trees indicating additionally the name of the inference rule
used at each step of the proof.
Here is a tree- proof of the de Morgan law (¬(a ∩ b) ⇒ (¬a ∪ ¬b)).

−→ (¬(a ∩ b) ⇒ (¬a ∪ ¬b))

| (→⇒)

278
 ¬(a ∩ b) −→ (¬a ∪ ¬b)
| (→ ∪)
¬(a ∩ b) −→ ¬a, ¬b
| (→ ¬)
b, ¬(a ∩ b) −→ ¬a
| (→ ¬)
b, a, ¬(a ∩ b) −→
| (¬ →)
b, a −→ (a ∩ b)
^
(→ ∩)

b, a −→ a b, a −→ b

Remark 6.1
The proof search in GL (to be defined by the decomposition tree) results are not
always unique; one formula (sequent) can have many proofs.

Here is another proof in GL of the de Morgan Law.

−→ (¬(a ∩ b) ⇒ (¬a ∪ ¬b))

| (→⇒)
¬(a ∩ b) −→ (¬a ∪ ¬b)
| (→ ∪)
¬(a ∩ b) −→ ¬a, ¬b
| (→ ¬)
b, ¬(a ∩ b) −→ ¬a
| (¬ −→)
b −→ ¬a, (a ∩ b)
^
(→ ∩)

b −→ ¬a, a b −→ ¬a, b
| (−→ ¬) | (−→ ¬)
b, a −→ a b, a −→ b

279
The process of searching for proofs of a formula A in GL consists, as in the RS
type systems, of building decomposition trees. Their construction is similar to
the one defined for RS type systems and is described intuitiively as follows.
We take a root of a decomposition tree TA a sequent −→ A. For each node, if
there is a rule of GL which conclusion has the same form as the node sequent,
then the node has children that are premises of the rule. If the node consists
only of an indecomposable sequent (built out of variables only), then it does not
have any children. This is a termination condition for the decomposition tree.
We prove that each formula A generates a finite set TA of decomposition trees,
such that the following holds. If there exist a tree TA ∈ TA whose all leaves are
axioms, then tree TA constitutes a proof of A in GL. If all trees in TA have at
least one non-axiom leaf, the proof of A does not exist.
The first step in formally defining a notion of a decomposition tree consists of
transforming the inference rules of GL, as we did in the case of the RS type
systems, into corresponding decomposition rules.

Decomposition rules of GL
Building a proof search decomposition tree consists of using the inference rules
in an inverse order; we transform the inference rules into decomposition rules
by reversing the role of conclusion and its premisses. We call such rules the
decomposition rules. Here are all of GL decomposition rules.
Conjunction decomposition rules

0 0 0
Γ , (A ∩ B), Γ −→ ∆ Γ −→ ∆, (A ∩ B), ∆
(∩ →) , (→ ∩) ,
Γ0 , A, B, Γ −→ ∆0 Γ −→ ∆, A, ∆0 ; Γ −→ ∆, B, ∆0

Disjunction decomposition rules

0 0 0
Γ −→ ∆, (A ∪ B), ∆ Γ , (A ∪ B), Γ −→ ∆
(→ ∪) , (∪ →) 0 ,
Γ −→ ∆, A, B, ∆0 Γ , A, Γ −→ ∆0 ; Γ0 , B, Γ −→ ∆0

Implication decomposition rules

0 0
Γ , Γ −→ ∆, (A ⇒ B), ∆
(→⇒) ,
Γ0 , A, Γ −→ ∆, B, ∆0
0 0
Γ , (A ⇒ B), Γ −→ ∆, ∆
(⇒→) ,
Γ0 , Γ −→ ∆, A, ∆0 ; Γ0 , B, Γ −→ ∆, ∆0

Negation decomposition rules

280
 0 0 0 0
Γ , ¬A, Γ −→ ∆, ∆ Γ , Γ −→ ∆, ¬A, ∆
(¬ →) , (→ ¬) .
Γ0 , Γ −→ ∆, A, ∆0 Γ0 , A, Γ −→ ∆, ∆0

We write the decomposition rules in a visual tree form as follows.

(→ ∪) rule
0
Γ −→ ∆, (A ∪ B), ∆
| (→ ∪)
0
Γ −→ ∆, A, B, ∆
(∪ →) rule

0 0
Γ , (A ∪ B), Γ −→ ∆
^
(∪ →)

0 0 0 0
Γ , A, Γ −→ ∆ Γ , B, Γ −→ ∆

(→ ∩) rule

0
Γ −→ ∆, (A ∩ B), ∆
^
(→ ∩)

0
0
Γ → ∆, B, ∆
Γ −→ ∆, A, ∆

(∩ →) rule
0 0
Γ , (A ∩ B), Γ −→ ∆
| (∩ →)
0 0
Γ , A, B, Γ −→ ∆
(→⇒) rule
0 0
Γ , Γ −→ ∆, (A ⇒ B), ∆
| (→⇒)
0 0
Γ , A, Γ −→ ∆, B, ∆
(⇒→) rule

281
 0 0
Γ , (A ⇒ B), Γ −→ ∆, ∆
^
(⇒→)

0 0 0 0
Γ , Γ −→ ∆, A, ∆ Γ , B, Γ −→ ∆, ∆

(¬ →) rule

0 0
Γ , ¬A, Γ −→ ∆, ∆

| (¬ →)
0 0
Γ , Γ −→ ∆, A, ∆

(−→ ¬) rule
0 0
Γ , Γ −→ ∆, ¬A, ∆
| (¬ →)
0 0
Γ , A, Γ −→ ∆, ∆

Observe that we use the same names for the inference and decomposition rules,
as once the we have built a decomposition tree (with use of the decomposition
rules) with all leaves being axioms, it constitutes a proof of A in GL with
branches labeled by the proper inference rules.
We have already defined (definition 6.8) indecomposable sequence as any se-
0 0 0 0
quence Γ −→ ∆ when Γ , ∆ ∈ V AR∗ . In particular, a formula that is not
a positive literal (propositional variable) is called a decomposable formula,
and a sequent Γ −→ ∆ where either Γ or ∆ contains a decomposable formula
is called a decomposable sequent.

By inspecting the domain of the rules we can see that at most two rules could
apply for any given decomposable sequent Γ −→ ∆.
For any decomposable sequent, at most two decomposition rules can be applied
to it. This rule is determined by the first decomposable formula in Γ when we
traverse it from left to right, and by the main connective of that formula, or by
the first decomposable formula in ∆ when we traverse it from the right to left,
and by the main connective of that formula. We hence are now ready to define
a decomposition tree.

282
Decomposition Tree T→A
For each formula A ∈ F, a decomposition tree T→A is a tree build as follows.
Step 1. The sequent −→ A is the root of T→A and for any node Γ −→ ∆ of
the tree we follow the steps below.
Step 2. If Γ −→ ∆ is indecomposable, then Γ −→ ∆ becomes a leaf of the
tree.
Step 3. If Γ −→ ∆ is decomposable, then we pick a decomposition rule that
applies by matching the sequent of the current node with the domain of the
decomposition rule. To do so we proceed as follows.
1. We traverse Γ from left to right to find the first decomposable formula.
Its main connective ◦ identifies a possible decomposition rule (◦ −→). Then
we check if this decomposition rule applies. If it does we put its conclusions
(conclusion) as leaves (leaf).
2. We traverse ∆ from right to left to find the first decomposable formula.
Its main connective ◦ identifies a possible decomposition rule (−→ ◦). Then
we check if this decomposition rule applies. If it does we put its conclusions
(conclusion) as leaves (leaf). 3. If 1. and 2. applies we choose one of the rules.
Step 4. We repeat steps 2 and 3 until we obtain only leaves.

Observation 6.1
The decomposable Γ −→ ∆ is always in the domain in one of the decomposition
rules (◦ −→), (−→ ◦), or in the domain of both. Hence the tree T→A may not
be unique and all possible choices of 3. give all possible decomposition trees.

We generalize the definition of T→A to the decomposition tree TΣ→Λ of any

sequent Σ −→ Λ ∈ SQ as follows.

Decomposition Tree TΣ−→Λ

Step 1. The sequent Σ −→ Λ is the root of TΣ−→Λ , and for any node Γ −→ ∆
of the tree we follow the steps below.
Step 2. If Γ −→ ∆ is indecomposable, then Γ −→ ∆ becomes a leaf of the
tree. Step 3. and Step 4. are the same as in the above definition of the tree
T→A .

Exercise 6.14
Prove, by constructing a proper decomposition tree that

`GL ((¬a ⇒ b) ⇒ (¬b ⇒ a)).

283
Solution
By definition,we have that

`GL ((¬a ⇒ b) ⇒ (¬b ⇒ a)) if and only if `GL −→ ((¬a ⇒ b) ⇒ (¬b ⇒ a)).

We construct a decomposition tree as follows.

T→A

−→ ((¬a ⇒ b) ⇒ (¬b ⇒ a))

| (→⇒)
(¬a ⇒ b) −→ (¬b ⇒ a)
| (→⇒)
¬b, (¬a ⇒ b) −→ a
| (→ ¬)
(¬a ⇒ b) −→ b, a
^
(⇒−→)

−→ ¬a, b, a b −→ b, a
| (→ ¬) axiom
a −→ b, a
axiom

All leaves of the tree are axioms, hence it constitutes a proof in GL.

Exercise 6.15
Prove, by constructing proper decomposition trees that

6 `GL ((a ⇒ b) ⇒ (¬b ⇒ a)).

Solution
Observe that for some formulas A, their decomposition tree T→A in GL may
not be unique. Hence we have to construct all possible decomposition trees to
see that none of them is a proof, i.e. to see that each of them has a non axiom
leaf. We construct the decomposition trees for −→ A as follows.

T1→A
−→ ((a ⇒ b) ⇒ (¬b ⇒ a))

(one choice)

284
 | (→⇒)
(a ⇒ b) −→ (¬b ⇒ a)
( first of two choices )
| (→⇒)
¬b, (a ⇒ b) −→ a
(one choice)
| (¬ →)
(a ⇒ b) −→ b, a
(one choice)
^
(⇒−→)

−→ a, b, a b −→ b, a
non − axiom axiom

The tree contains a non- axiom leaf, hence it is not a proof. We have one more
tree to construct.

T2→A
−→ ((a ⇒ b) ⇒ (¬b ⇒ a))

| (→⇒)
(a ⇒ b) −→ (¬b ⇒ a)
(second choice)
^
(⇒−→)

−→ (¬b ⇒ a), a b −→ (¬b ⇒ a)

| (−→⇒) | (→⇒)
¬b −→ a, a b, ¬b −→ a
| (¬ →) | (¬ →)
−→ b, a, a b −→ a, b
non − axiom axiom

All possible trees end with a non-axiom leaf. It proves that

6 `GL ((a ⇒ b) ⇒ (¬b ⇒ a)).

Exercise 6.16
Does the tree below constitute a proof in GL?

285
 T→A

−→ ¬¬((¬a ⇒ b) ⇒ (¬b ⇒ a))

| (→ ¬)
¬((¬a ⇒ b) ⇒ (¬b ⇒ a)) −→
| (¬ →)
−→ ((¬a ⇒ b) ⇒ (¬b ⇒ a))
| (→⇒)
(¬a ⇒ b) −→ (¬b ⇒ a)
| (→⇒)
(¬a ⇒ b), ¬b −→ a
| (¬ →)
(¬a ⇒ b) −→ b, a
^
(⇒→)

−→ ¬a, b, a b −→ b, a
| (→ ¬) axiom
a −→ b, a
axiom

Solution
The tree above is not a proof in GL because a decomposition rule used in the
decomposition step below does not exists in GL

(¬a ⇒ b), ¬b −→ a

| (¬ →)

(¬a ⇒ b) −→ b, a.

It is a proof is some system GL1 that has all the rules of GL except its rule
(¬ →)
0 0
Γ , Γ −→ ∆, A, ∆
(¬ →)
Γ0 , ¬A, Γ −→ ∆, ∆0
This rule has to be replaced in by the rule:
0 0
Γ, Γ −→ ∆, A, ∆
(¬ →)1
Γ, ¬A, Γ0 −→ ∆, ∆0

286
6.6 GL Soundness and Completeness
The system GL admits a constructive proof of the Completeness Theorem,
similar to completeness proofs for RS type proof systems (Theorems 9.7, 6.6,
6.7). It also relays on strong soundness property of its inference rules. We are
going to prove that the following holds.

Theorem 6.8 (GL Strong Soundness)

The proof system GL is strongly sound.

Proof We have already proved (Fact 6.4) that logical axioms of GL are tau-
tologies, so we have to prove now that its rules of inference are strongly sound
(definition 6.4). Proofs of strong soundness of rules of inference of GL are more
involved then the proofs for the RS type rules. We prove as an example the
strong soundness of four of inference rules. Proofs for all other rules follows the
same patterns and is left as an exercise.
By definition 6.4 of strong soundness we have to show the condition (8.77).
Written formally it says that we have to show that that if P1 , P2 are premisses
of a given rule and C is its conclusion, then for all truth assignments v :
V AR −→ {T, F },
v ∗ (P1 ) = v ∗ (C) in case of one premiss rule, and
v ∗ (P1 ) ∩ v ∗ (P2 ) = v ∗ (C), in case of a two premisses rule.
In order to prove it we need additional classical equivalencies listed below. You
can fond a list of most basic classical equivalences in Chapter 3.

((A ⇒ B) ∩ (A ⇒ C)) ≡ (A ⇒ (B ∩ C))

((A ⇒ C) ∩ (B ⇒ C)) ≡ ((A ∪ B) ⇒ C)
((A ∩ B) ⇒ C) ≡ (A ⇒ (¬B ∪ C))

0 0
Γ , A, B, Γ −→ ∆
(∩ →)
Γ0 , (A ∩ B), Γ −→ ∆0
0 0 0 0 0
v ∗ (Γ , A, B, Γ −→ ∆ ) = (v ∗ (Γ )∩v ∗ (A)∩v ∗ (B)∩v ∗ (Γ)) ⇒ v ∗ (∆ ) = (v ∗ (Γ )∩
0 0 0
v ∗ (A ∩ B) ∩ v ∗ (Γ)) ⇒ v ∗ (∆ ) = v ∗ (Γ , (A ∩ B), Γ −→ ∆ )

0 0
Γ −→ ∆, A, ∆ ; Γ −→ ∆, B, ∆
(→ ∩)
Γ −→ ∆, (A ∩ B), ∆0
0 0
v ∗ (Γ −→ ∆, A, ∆ ) ∩ v ∗ (Γ −→ ∆, B, ∆ )
0 0
= (v ∗ (Γ) ⇒ v ∗ (∆) ∪ v ∗ (A) ∪ v ∗ (∆ )) ∩ (v ∗ (Γ) ⇒ v ∗ (∆) ∪ v ∗ (B) ∪ v ∗ (∆ ))

287
[we use : ((A ⇒ B) ∩ (A ⇒ C)) ≡ (A ⇒ (B ∩ C))]
0 0
= v ∗ (Γ) ⇒ ((v ∗ (∆) ∪ v ∗ (A) ∪ v ∗ (∆ )) ∩ (v ∗ (∆) ∪ v ∗ (B) ∪ v ∗ (∆ )))
[we use commutativity and distributivity]
0
= v ∗ (Γ) ⇒ (v ∗ (∆) ∪ (v ∗ (A ∩ B)) ∪ v ∗ (∆ ))
0
= v ∗ (Γ −→ ∆, (A ∩ B), ∆ )

0 0 0 0
Γ , A, Γ −→ ∆ ; Γ , B, Γ −→ ∆
(∪ →)
Γ0 , (A ∪ B), Γ −→ ∆0
0 0 0 0
v ∗ (Γ , A, Γ −→ ∆ ) ∩ v ∗ (Γ , B, Γ −→ ∆ )
0 0 0 0
= (v ∗ (Γ ) ∩ v ∗ (A) ∩ v ∗ (Γ)) ⇒ v ∗ (∆ )) ∩ (v ∗ (Γ ) ∩ v ∗ (B) ∩ v ∗ (Γ)) ⇒ v ∗ (∆ ))
[we use: ((A ⇒ C) ∩ (B ⇒ C)) ≡ ((A ∪ B) ⇒ C])
0 0 0
= (v ∗ (Γ ) ∩ v ∗ (A) ∩ v ∗ (Γ)) ∪ (v ∗ (Γ ) ∩ v ∗ (B) ∩ v ∗ (Γ)) ⇒ v ∗ (∆ )
0 0 0
= ((v ∗ (Γ ) ∩ v ∗ (Γ)) ∩ v ∗ (A)) ∪ ((v ∗ (Γ ) ∩ v ∗ (Γ)) ∩ v ∗ (B)) ⇒ v ∗ (∆ )
[we use commutativity and distributivity]
0 0
= ((v ∗ (Γ ) ∩ v ∗ (Γ)) ∩ (v ∗ (A ∪ B)) ⇒ v ∗ (∆ )
0 0
= v ∗ (Γ , (A ∪ B), Γ −→ ∆ )

0 0
Γ , A, Γ −→ ∆, ∆
(→ ¬)
Γ0 , Γ −→ ∆, ¬A, ∆0
0 0 0 0
v ∗ (Γ , A, Γ −→ ∆, ∆ ) = v ∗ (Γ ) ∩ v ∗ (A) ∩ v ∗ (Γ) ⇒ v ∗ (∆) ∪ v ∗ (∆ )
0 0
= (v ∗ (Γ ) ∩ v ∗ (Γ)) ∩ v ∗ (A) ⇒ v ∗ (∆) ∪ v ∗ (∆ )
[we use: ((A ∩ B) ⇒ C) ≡ (A ⇒ (¬B ∪ C))]
0 0 0
= (v ∗ (Γ ) ∩ v ∗ (Γ)) ⇒ ¬v ∗ (A) ∪ v ∗ (∆) ∪ v ∗ (∆ ) = (v ∗ (Γ ) ∩ v ∗ (Γ)) ⇒ v ∗ (∆) ∪
0
v ∗ (¬A) ∪ v ∗ (∆ )
0 0
= v ∗ (Γ , Γ −→ ∆, ¬A, ∆ )

The above shows the premises and conclusions are logically equivalent, therefore
the rules of inference are strongly sound. It ends the proof.
Observe that the strong soundness implies soundness (not only by name!), hence
we have also proved the following

Theorem 6.9 (Soundness for GL)

For any sequent Γ −→ ∆ ∈ SQ,
if `GL Γ −→ ∆, then |= Γ −→ ∆. In particular, for any A ∈ F,
if `GL A, then |= A.

We know by theorem 6.8 that all the rules of inference of GL of are strongly
sound. The strong soundness of the rules means that if at least one of premisses
of a rule is false, so is its conclusion. Hence given a sequent Γ −→ ∆ ∈ SQ, such

288
that its decomposition tree TΓ−→∆ has a branch ending with a non-axiom leaf.
It means that any truth assignment v that make this non-axiom leaf false also
falsifies all sequences on that branch, and hence falsifies the sequent Γ −→ ∆. In
particular, given a sequent −→ A and its tree T−→A , any v, such that falsifies its
a non-axiom leaf is a counter-model for A. We have hence proved the following.

Theorem 6.10 (GL Counter Model)

Given a sequent Γ −→ ∆, such that its decomposition tree TΓ−→∆ contains a
non- axiom leaf LA . Any truth assignment v that falsifies the non-axiom leaf
LA is a counter model for Γ −→ ∆. In particular, given a formula A ∈ F,
and its decomposition tree TA with a non-axiom leaf, this leaf let us define a
counter-model for A determined by the decomposition tree TA .

Here is a simple exercise explaining how the construction of a counter-model

determined by the decomposition tree of a works.

Exercise 6.17
Prove, by constructing a counter-model determined by decomposition tree that

6|= ((b ⇒ a) ⇒ (¬b ⇒ a)).

Solution
We construct the decomposition tree for the formula A : ((b ⇒ a) ⇒ (¬b ⇒ a))
as follows.

T→A

−→ ((b ⇒ a) ⇒ (¬b ⇒ a))

| (→⇒)
(b ⇒ a) −→ (¬b ⇒ a)
| (→⇒)
¬b, (b ⇒ a) −→ a
| (¬ →)
(b ⇒ a) −→ b, a
^
(⇒−→)

−→ b, b, a a −→ b, a
non − axiom axiom

289
The non-axiom leaf LA we want to falsify is −→ b, b, a. Let v : V AR −→ {T, F }
be a truth assignment. By definition 6.7 of semantic for GL we have that
v ∗ (LA ) = v ∗ (−→ b, b, a) = (T ⇒ v(b) ∪ v(b) ∪ v(a)). Hence v ∗ (−→ b, b, a) = F
if and only if (T ⇒ v(b) ∪ v(b) ∪ v(a)) = F if and only if v(b) = v(a) = F .
The Theorem 6.10, says that the logical value F determined by the evaluation
a non-axiom leaf LA ”climbs” the decomposition tree. We picture it as follows.

T→A

−→ ((b ⇒ a) ⇒ (¬b ⇒ a)) F

| (→⇒)

(b ⇒ a) −→ (¬b ⇒ a) F

| (→⇒)

¬b, (b ⇒ a) −→ a F

| (¬ →)

(b ⇒ a) −→ b, a F
^
(⇒−→)

−→ b, b, a F a −→ b, a
non − axiom axiom

So, by theorem 6.10, any truth assignment v : V AR −→ {T, F }, such that

v(b) = v(a) = F falsifies the sequence −→ A, i.e. v ∗ (−→ A) = T ⇒ v ∗ (A) = F .
This is possible only if v ∗ (A) = F . This proves that v is a counter model for A
and we proved that 6|= A.

Our main goal is to prove the Completeness Theorem for RS. We prove first
the Completeness Theorem for formulas A ∈ F and then we generalize it to any
sequences Γ ∈ F ∗ .

Theorem 6.11 (Completeness Theorem)

For any formula A ∈ F,

`GL A if and only if |= A.

For any sequent Γ −→ ∆ ∈ SQ,

`GL Γ −→ ∆ if and only if |= Γ −→ ∆.

290
Proof
We have already proved the Soundness Theorem 6.9, so we need to prove only
the completeness part of it, namely to prove the implication:

if |= A, then `GL A. (6.25)

We prove instead of the logically equivalent opposite implication:

if 6 `GL A then 6|= A. (6.26)

Assume 6 `GL A. By (6.24) it means that 6 `GL −→ A. Let TA be a set of all

decomposition trees of −→ A. As 6 `GL −→ A, each tree T→A in the set TA has
0 0
a non-axiom leaf. We choose an arbitrary T→A from TA . Let LA = Γ −→ ∆ ,
be a non-axiom leaf of T→A . We define a truth assignment v : V AR −→ {T, F }
0 0
which falsifies Γ −→ ∆ as follows.

if a appears in Γ0

 T
v(a) = F if a appears in ∆0
any value if a does not appear in Γ0 → ∆0


By the strong soundness of the rules of inference of GL and Theorem 6.10 it

proves that v ∗ (−→ A) = F , i.e. that 6|= −→ A and hence 6|= A.
Assume that Γ −→ ∆ is any sequence such that 6 `GL Γ −→ ∆. But `GL Γ −→
∆ if and only if `GL (σΓ ⇒ δ∆ ). So 6 `GL Γ −→ ∆ if and only if 6 `GL σΓ ⇒
δ∆ ). By already proven Case 1, 6|= σΓ ⇒ δ∆ ), what is obviously equivalent to
6|= Γ −→ ∆. This ends the proof of Case 2 and Completeness Theorem.

Gentzen Sequent Proof System G (6.27)

The proof system G is in its structure the most similar to the proof system RS2
defined by (6.19).
It is obtained from in the same way is a proof system obtained from GL by
0 0
changing the indecomposable sequences Γ , ∆ into any sequences Σ, Λ ∈ F ∗
in all of the rules of inference of GL.
The logical axioms LA remain the same; i.e. the components of G are as
follows.
Axioms of G
As the axioms of GL we adopt any indecomposable sequent which contains a
positive literal a (variable) that appears on both sides of the sequent arrow −→,
i.e any sequent of the form

Γ0 1 , a, Γ0 2 −→ ∆0 1 , a, ∆0 2 , (6.28)

for any a ∈ V AR and any sequences Γ0 1 , Γ0 2 , ∆0 1 , ∆0 2 ∈ V AR∗ .

291
 Inference Rules of G (6.29)

We adopt the following rules of inference.

Conjunction rules
Σ, A, B, Γ −→ Λ
(∩ →) ,
Σ, (A ∩ B), Γ −→ Λ

Γ −→ ∆, A, Λ ; Γ −→ ∆, B, Λ
(→ ∩) ,
Γ −→ ∆, (A ∩ B), Λ
Disjunction rules
Γ −→ ∆, A, B, Λ
(→ ∪) ,
Γ −→ ∆, (A ∪ B), Λ

Σ, A, Γ −→ Λ ; Σ, B, Γ −→ Λ
(∪ →) ,
Σ, (A ∪ B), Γ −→ Λ

Implication rules
Σ, A, Γ −→ ∆, B, Λ
(→⇒) ,
Σ, Γ −→ ∆, (A ⇒ B), Λ

Σ, Γ −→ ∆, A, Λ ; Σ, B, Γ −→ ∆, Λ
(⇒→) ,
Σ, (A ⇒ B), Γ −→ ∆, Λ
Negation rules
Σ, Γ −→ ∆, A, Λ Σ, A, Γ −→ ∆, Λ
(¬ →) , (→ ¬) ,
Σ, ¬A, Γ −→ ∆, Λ Σ, Γ −→ ∆, ¬A, Λ
where Γ, ∆, Σ. Λ ∈ F ∗ .

Exercise 6.18 Follow the example of the GL system and adopt all needed
definitions and proofs to prove the completeness of the system G.

Solution
We leave it to the reader to fill in details .In particular, one has to accomplish
the steps below.
1. Explain why the system G is strongly sound. You can use the strong
soundness of the system GL .
2. Prove, as an example, a strong soundness of 4 rules of G.
3. Prove the the following Strong Soundness Theorem for G.

292
Theorem 6.12
The proof system G is strongly sound.

4. Define shortly, in your own words, for any formula A ∈ F, its decomposition
tree T→A in G.
5. Extend your definition to a decomposition tree TΓ→∆ .
6. Prove that for any Γ → ∆ ∈ SQ, the decomposition tree TΓ→∆ are finite.
7. Give an example of formulas A, B ∈ F such that that T→A is unigue and
T→B is not.
8. Prove the following Counter Model Theorem for G.

Theorem 6.13
Given a sequent Γ −→ ∆, such that its decomposition tree TΓ−→∆ contains a
non- axiom leaf LA . Any truth assignment v that falsifies the non-axiom leaf
LA is a counter model for Γ −→ ∆.

10. Prove the following Completeness Theorem for G.

Theorem 6.14 For any formula A ∈ F,

1. `G A if and only if |= A,
and for any sequent Γ −→ ∆ ∈ SQ,
2. `G Γ −→ ∆ if and only if |= Γ −→ ∆.

6.7 Original Gentzen Systems LK, LI

Completeness and Hauptzatz Theorems
The original systems LK and LI were created by Gentzen in 1935 for clas-
sical and intuitionistic predicate logics, respectively. The proof system LI for
intuitionistic propositional logic is a particular case of the proof system LK.
Both systems LK and LI have two groups of inference rules and a special rule
called a cut rule. One group consists of a set of rules similar to the rules of
systems GL and G. We call them Logical Rules. The other group contains a
new type of rules, called Structural Rules. The cut rule in Gentzen sequent
systems corresponds to the Modus Ponens rule in Hilbert proof systems as
Modus Ponens is a particular case of the cut rule. The cut rule is needed to
carry the original Gentzen proof of the completeness of the system LK and
proving the adequacy of LI system for intituitionistic logic. Gentzen proof of
completeness of LK was not direct. He used the completeness of already known
Hilbert proof systems H and proved that any formula provable in the systems
H is also provable in LK, respectively. Hence the need of the cut rule.

293
For the system LI he proved only the adequacy of LI system for intituitionistic
logic since the semantics for the intuitionistic logic didn’t yet exist. He used
the acceptance of the Heying intuitionistic axiom system as the definition of the
intuitionistic logic and proved that any formula provable in the Heyting system
is also provable in LI.
Observe that by presence of the cut rule, Gentzen LK, LI systems are also
a Hilbert system. What distinguishes it from all other known Hilbert proof
systems is the fact that the cut rule could be eliminated from it.
This is Gentzen famous Hauptzatz Theorem, also called Cut Elimination The-
orem. The elimination of the cut rule and the structure of other rules makes
it possible to define effective automatic procedures for proof search, what is
impossible in a case of the Hilbert style systems.
Gentzen, in his proof of Hauptzatz Theorem, developed a powerful technique
adaptable to other logics. We present it here in classical propositional case and
show how to adapt it to the intuitionistic case.
Gentzen proof is purely syntactical. It defines a constructive method of trans-
formation of any formal proof (derivation) of a sequent Γ −→ ∆ that uses a
cut rule (and other rules) into its proof without use of the cut rule. Hence the
English name Cut Elimination Theorem.
The completeness (with respect to algebraic semantics defined in chapter 7) of
the cut free system LI follows directly from LI Hauptzatz Theorem 6.22 and the
intuitionistic completeness theorem (chapter 7). The proof is a straightforward
adaptation of the proof of cut free LK Completeness Theorem 6.23 and is left
as a homework exercise in chapter 7.
Rasiowa and Sikorski method of proving completeness theorem by constructing
counter-models on the decomposition trees is a semantical equivalence to purely
syntactical Gentzen proof of cut elimination. It is relatively simple, powerful
and easy to understand. It was the reason it was first to be presented here. But
it is more difficult and sometimes impossible to apply (generalize) to many non-
classical logics then Gentzen cut elimination method. Moreover the Gentzen
method is more flexible and in this sense more general and powerful. This is
why we preset it here.

Components of LK, LI (6.30)

Language L
The language is the same as the in case of GL, namely

L = L{¬,∩,∪,⇒} .

Expressions

294
The set of all expressions E is, as before, the set

SQ = {Γ −→ ∆ : Γ, ∆ ∈ F ∗ } (6.31)

of all sequents.
Logical Axioms
There is only one logical axiom, namely

A −→ A,

where A is any formula of L.

Rules of Inference
There are two groups of rules of inference and they are defined are as follows.

GROUP ONE: STRUCTURAL RULES.

Weakening in the antecedent

Γ −→ ∆
(weak →) ,
A, Γ −→ ∆
Weakening in the succedent
Γ −→ ∆
(→ weak) ,
Γ −→ ∆, A
Contraction in the antecedent
A, A, Γ −→ ∆
(contr →) ,
A, Γ −→ ∆
Contraction in the succedent
Γ −→ ∆, A, A
(→ contr) ,
Γ −→ ∆, A
Exchange in the antecedent
Γ1 , A, B, Γ2 −→ ∆
(exch →) ,
Γ1 , B, A, Γ2 −→ ∆
Exchange in the succedent
∆ −→ Γ1 , A, B, Γ2
(→ exch) ,
∆ −→ Γ1 , B, A, Γ2

Cut Rule

295
 Γ −→ ∆, A ; A, Σ −→ Θ
(cut) .
Γ, Σ −→ ∆, Θ
The formula A is called a cut formula.

GROUP TWO: LOGICAL RULES

Conjunction
A, Γ −→ ∆
(∩ →)1 ,
(A ∩ B), Γ −→ ∆
B, Γ −→ ∆
(∩ →)2 ,
(A ∩ B), Γ −→ ∆
Γ −→ ∆, A ; Γ −→ ∆, B
(→ ∩) .
Γ −→ ∆, (A ∩ B)
Disjunction
Γ −→ ∆, A
(→ ∪)1 ,
Γ −→ ∆, (A ∪ B)
Γ −→ ∆, B
(→ ∪)2 ,
Γ −→ ∆, (A ∪ B)
A, Γ −→ ∆ ; B, Γ −→ ∆
(∪ →) .
(A ∪ B), Γ −→ ∆
Implication
A, Γ −→ ∆, B
(→⇒) ,
Γ −→ ∆, (A ⇒ B)
Γ −→ ∆, A ; B, Γ −→ ∆
(⇒→) .
(A ⇒ B), Γ −→ ∆
Negation
Γ −→ ∆, A
(¬ →) ,
¬A, Γ −→ ∆
A, Γ −→ ∆
(→ ¬) .
Γ −→ ∆, ¬A

Definition 6.9 (Classical System LK)

We define the classical Gentzen system LK as

LK = (L, SQ, AL, Structural Rules, Cut Rule, Logical Rules),

where all the components are defined by (9.24) above.

296
Definition 6.10 (Intuitionistic System LI)
We define the intuitionistic Gentzen system LI as
LK = (L, ISQ, AL, I-Structural Rules, I- Cut Rule, I- Logical Rules),
where ISQ is the following subset of the set SQ of all sequents (6.31)
ISQ = {Γ −→ ∆ : ∆ consists of at most one formula }. (6.32)
The set ISQ is called the set of all intuitionistic sequents.
The I-Structural Rules, I- Cut Rule, I- Logical Rules are the LK rules restricted
to the set ISQ (6.32) of the intuitionistic sequents.

We will study the intuitionistic system LI in chapter 7. We concentrate now on

then classical LK.

Classical System LK

We say that a formula A ∈ F, has a proof in LK and denote it by `LK A if the

sequent −→ A has a proof in GL, i.e. we define:
`LK A if and only if `LK −→ A. (6.33)
Proof Trees
We write formal proofs in LK, as we did for other Gentzen style proof systems
in a form of trees in an ”upside -down” form.
By a proof tree of a sequent Γ −→ ∆ in LK we understand a tree
DΓ−→∆
satisfying the following conditions:
1. The topmost sequent, i.e the root of DΓ−→∆ is Γ −→ ∆.
2. All leaves are axioms.
3. The nodes are sequents such that each sequent on the tree follows from the
ones immediately preceding it by one of the rules.

The proofs are often called derivations. In particular, Gentzen, in his work
used the term derivation we will use this notion as well. This is why we denote
the proof trees by D (for derivation).
Finding derivations D in LK are is a more complex process, as the logical rules
are different, then in GL and G. Proofs rely strongly on use of the Structural
Rules. Even if we find a derivation that does not involve the Cut rule, the
Structural rules are usually present. For example, a derivation of Excluded
Middle (A ∪ ¬A) formula B in LK is as follows.

297
 D
−→ (A ∪ ¬A)
| (→ contr)
−→ (A ∪ ¬A), (A ∪ ¬A)
| (→ ∪)1
−→ (A ∪ ¬A), A
| (→ exch)
−→ A, (A ∪ ¬A)
| (→ ∪)1
−→ A, ¬A
| (→ ¬)
A −→ A
axiom

Here is as yet another example a proof P ( also cut free) of the de Morgan Law
(¬(A ∩ B) ⇒ (¬A ∪ ¬B)).

P
−→ (¬(A ∩ B) ⇒ (¬A ∪ ¬B))
| (→⇒)
(¬(A ∩ B) −→ (¬A ∪ ¬B))
| (→ ¬)
−→ (¬A ∪ ¬B), (A ∩ B)
^
(⇒−→)

−→ (¬A ∪ ¬B), A −→ (¬A ∪ ¬B), B

| (→ exch) | (→ exch)
−→ A, (¬A ∪ ¬B) −→ B, (¬A ∪ ¬B)
| (→ ∪)1 | (→ ∪)1
−→ A, ¬A −→ B, ¬B
| (→ ¬) B −→ B
A −→ A axiom
axiom

298
Observe that the Logical Rules are similar in their structure to the rules of the
system G and hence admit the same proof of their soundness.
The rules (→ ∪)1 , (→ ∪)2 and (→ ∪)1 , (→ ∪)2 are not strongly sound as
A 6≡ (A ∩ B), B 6≡ (A ∩ B) and A 6≡ (A ∩ B), B 6≡ (A ∩ B).
All other Logical Rules are strongly sound.
The Contraction and Exchange structural are also strongly sound as for any
formulas A, B ∈ F, A ≡ (A∩A), A ≡ (A∪A) and (A∩B) ≡ (B ∩A), (A∩B) ≡
(B ∩ A). The Weakening rule is sound because (we use shorthand notation) if
a formula (Γ ⇒ ∆) = T then also ((A ∩ Γ) ⇒ ∆)) = T for any logical value of
the formula A. But obviously (Γ ⇒ ∆) 6≡ ((A ∩ Γ) ⇒ ∆)), i.e. the Weakening
rule is not strongly sound.
The Cut rule is sound as the fact (Γ ⇒ (∆ ∪ A)) = T and ((A ∩ Σ) ⇒ Λ) = T
implies that Γ, Σ −→ ∆, Λ. It is not strongly sound. Any truth assignment
such that Γ = T, ∆ = Σ = Λ = A = F proves that (Γ −→ ∆, A) ∩ (A, Σ −→
Λ) 6≡ (Γ, Σ −→ ∆, Λ). Obviously, |= A −→ A.
We have proved that LK is sound and hence the following theorem holds.

Theorem 6.15 (Soundness for LK)

For any sequent Γ −→ ∆,

if `LK Γ −→ ∆, then |= Γ −→ ∆.

In particular, for any A ∈ F,

if `LK A, then |= A.

We follow now Gentzen way of proving completeness of LK. We choose any

complete Hilbert proof system for the LK language L = L{¬,∩,∪,⇒} and prove
its equivalency with LK.
Gentzen referred to the Hilbert-Ackerman (1920) system (axiomatization) in-
cluded in chapter 5. We choose here the Rasiowa-Sikorski (1952) formalization
R also included in chapter 5.
We do it for two reasons. First, it reflexes a connection between classical and
intuitionistic logics very much in a spirit Gentzen relationship between LK and
LI.
We obtain a complete proof system I (chapter 7) from R by just removing the
last axiom A10. Second, both sets of axioms reflect the best what set of prov-
able formulas is needed to conduct algebraic proofs of completeness of R and I,
respectively.

299
 Axioms of R (6.34)

The set of logical axioms of the Hilbert style proof system RS for classical
propositional logic all formulas of the forms
A1 ((A ⇒ B) ⇒ ((B ⇒ C) ⇒ (A ⇒ C))),
A2 (A ⇒ (A ∪ B)),
A3 (B ⇒ (A ∪ B)),
A4 ((A ⇒ C) ⇒ ((B ⇒ C) ⇒ ((A ∪ B) ⇒ C))),
A5 ((A ∩ B) ⇒ A),
A6 ((A ∩ B) ⇒ B),
A7 ((C ⇒ A) ⇒ ((C ⇒ B) ⇒ (C ⇒ (A ∩ B))),
A8 ((A ⇒ (B ⇒ C)) ⇒ ((A ∩ B) ⇒ C)),
A9 (((A ∩ B) ⇒ C) ⇒ (A ⇒ (B ⇒ C)),
A10 (A ∩ ¬A) ⇒ B),
A11 ((A ⇒ (A ∩ ¬A)) ⇒ ¬A),
A12 (A ∪ ¬A),
where A, B, C ∈ F are any formulas in L = L{¬,∩,∪,⇒} .
We adopt a Modus Ponens

A ; (A ⇒ B)
(M P )
B
as the only inference rule.
We define Hilbert System R as

R = ( L{¬,∩,∪,⇒} , F, A1 − A12, (M P ) ), (6.35)

where A1 - A12 are defined by (6.34).

The system R is complete, i.e. we have the following.

Theorem 6.16
For any formula A ∈ F,

`R A if and only if |= A.

We leave it as an exercise for the reader to show that all axioms A1 - A12 of the
system R are provable in LK. Moreover, the Modus Ponens is a particular case

300
of the cut rule, for Γ, ∆, Σ empty sequences and Θ containing only one element,
a formula B. We call it also MP rule.
−→ A ; A −→ B
(M P ) .
−→ B
This proves the following.

Theorem 6.17
For any formula A ∈ F,

if `R A, then `LK A.

Directly from the above theorem 6.17, soundness of LK (theorem 6.15) and
completeness of R (theorem 6.16) we get the completeness of LK.

Theorem 6.18 (LK Completeness)

For any formula A ∈ F,

`LK A if and only if |= A.

Here is Gentzen original formulation of the Hauptzatz Theorems, which we call

also the Cut Elimination Theorem.

Theorem 6.19 (Hauptzatz) (Classical LK)

Every derivation in LK can be transformed into another LK derivation of the
same sequent, in which no cuts occur.

Theorem 6.20 (Hauptzatz) (Intuitionistic LI)

Every derivation in LI can be transformed into another LI derivation of the
same sequent, in which no cuts occur.

The proof is quite long and involved. We present here its main and most im-
portant steps. To facilitate the proof we introduce a more general form of the
cut rule, called a mix rule defined as follows.
Γ −→ ∆ ; Σ −→ Θ
(mix) , (6.36)
Γ, Σ∗ −→ ∆∗ , Θ

where Σ∗ , ∆∗ are obtained from Σ, ∆ by removing all occurrences of a common

formula A. The formula A is now called a mix formula.

Example 6.3

301
Here are some examples of an applications of the mix rule. Observe that the
mix rule applies, as the cut does, to only one mix formula at the time.
a −→ b, ¬a ; (b ∪ c), b, b, D, b −→
(mix)
a, (b ∪ c), D −→ ¬a

b is the mix formula.

A −→ B, B, ¬A ; (b ∪ c), B, B, D, B −→ ¬B
(mix)
A, (b ∪ c), D −→ ¬A, ¬B
B is the mix formula.
A −→ B, ¬A, ¬A ; ¬A, B, B, ¬A, B −→ ¬B
(mix)
A, B, B −→ B, ¬B
¬A is the mix formula.

Notice, that every derivation with cut may be transformed into a derivation with
mix by means of a number of weakenings (multiple application of the weakening
rules) and interchanges (multiple application of the exchange rules). Conversely,
every mix may be transformed into a cut derivation by means of a certain num-
ber of preceding exchanges and contractions, though we do not use this fact in
the proof. Observe that cut is a particular case of mix.

Proof of Hauptzatz Theorems

The proof for LI is the same as for LK. We must just be careful to add, at
each step, the restriction to the ISQ sequences and the form of the rules. These
restrictions do not alter the flow and validity of the LKproof. We leave it as
homework exercise to the reader to re-write the proof below step by step for LI.
We conduct the proof in three main steps.
Step 1: we consider only derivations in which only mix rule is used.
Step 2: we consider first derivation with a certain Property H (definition 6.11)
and prove lemma 6.2 for them. This lemma is the most crucial for the proof of
the Hauptzatz.

Definition 6.11
We say that a derivation DΓ−→∆ of a sequent Γ −→ ∆ has a Property H if
it satisfies the the following conditions.
1. The root Γ −→ ∆ of the derivation DΓ−→∆ is obtained by direct use of
the mix rule, i.e. the mix rule is the last rule of inference used in the proof
(derivation) of Γ −→ ∆.
2. The derivation DΓ−→∆ does not contain any other application of the mix rule,
i.e. the proof (derivation) of Γ −→ ∆ does not contain any other application of
the mix rule.

302
Lemma 6.2 (H lemma)
Any derivation that fulfills the Property H (definition 6.11) may be trans-
formed into a derivation of the same sequent) in which no mix occurs.

Step 3: we use the H lemma 6.2 and to prove the the Hauptzatz as follows.
Hauptzatz proof from H lemma
Let D be any derivation (tree proof). Let Γ −→ ∆ be any node on D such that
its sub-tree DΓ−→∆ has the PropertyH (definition 6.11). By H lemma 6.2 the
sub-tree DΓ−→∆ can be replaced by a tree D∗ Γ−→∆ in which no mix occurs.
The rest of D remains unchanged. We repeat this procedure for each node N,
such that the sub-tree DN has the Property H until every application of mix
rule has systematically been eliminated. This ends the proof of Hauptzatz pro-
vided the H lemma 6.2 has already been proved.

Step 2: proof of H lemma.

We now consider derivation tree D with the Property H, i.e. such that the
mix rule is the last rule of inference used, and D does not contain any other
application of the mix rule.

We define now two important notions: degree n and rank r of the derivation
D. Observe that D contains only one application of mix rule, and the mix rule,
contains only one mix formula A. Mix rule used may contain many copies of
A, but there always is only one mix formula. We call is a mix formula of D.

Definition 6.12
Given a derivation tree D with the Property H.
Let A ∈ F be the mix formula of D. The degree n ≥ 0 of A is called the degree
of the derivation D. We write it as degD = degA = n.

Definition 6.13
Given a derivation tree D with the Property H. We define the rank r of D as a
sum of its left rank Lr and right rank Rr of D, i.e.
r = Lr + Rr,
where:
1. the left rank Lr of D in the largest number of consecutive nodes on the branch
of D staring with the node containing the left premiss of the mix rule, such that
each sequent on these nodes contains the mix formula in the succedent;
2. the right rank Rr of D in the largest number of consecutive nodes on the
branch of D staring with the node containing the right premiss of the mix

303
rule, such that each sequent on these nodes contains the mix formula in the
antecedent.

The lowest possible rank is evidently 2.

To prove the lemma we carry out two complete inductions, one on the degree n,
the other on the rank r, of the derivation D.
It means we prove the lemma for a derivation of the degree n, assuming it to
hold for derivations of a lower degree (in so far as there are such derivations,
i.e., as long as n 6= 0), supposing, therefore, that derivations of lower degree can
be already transformed into derivations without mix.
Furthermore, we shall begin by considering the case 1when the rank r = 2, and
after that the case 2 when the rank r > 2, where we assume that the lemma
already holds for derivations of the same degree, but a lower rank.

Case 1. Rank of r =2.

We present some cases and leave similar others to the reader as an exercise.
Observe that first group contains cases that are especially simple in that they
allow the mix to be immediately eliminated. The second group contains the
most important cases since their consideration brings out the basic idea behind
the whole proof, Here we use the induction hypothesis with respect do the degree
of the derivation. We reduce each one of the cases to transformed derivations
of a lower degree.

GROUP 1. Axioms and Structural Rules.

1. The left premiss of the mix rule is an axiom A −→ A.

Then the sub-tree of D containing mix is as follows.

A, Σ∗ −→ ∆
^
(mix)

A −→ A Σ −→ ∆

We transform it, and replace it in D by

A, Σ∗ −→ ∆

304
 possibly several exchanges and contractions
Σ −→ ∆

Such obtained D∗ proves the same sequent and contains no mix.

2 . The right premiss of the mix rule is an axiom A −→ A.

This is a case dual to 1. We show here the dial transformation, but will leave
the dual cases to the reader in the future.

Then the sub-tree of D containing mix is as follows.

Σ −→ ∆∗ , A
^
(mix)

Σ −→ ∆ A −→ A

We transform it, and replace it in D by

Σ −→ ∆∗ , A

possibly several exchanges and contractions

Σ −→ ∆

Such obtained D∗ proves the same sequent and contains no mix.

Suppose that neither of premisses of mix is an axiom. As the rank r=2 , the
right and left ranks are equal one. This means that in the sequents on the nodes
directly below left premiss of the mix, the mix formula A does not occur in the
succedent; in the sequents on the nodes directly below right premiss of the mix,
the mix formula A does not occur in the antecedent.
In general, if a formula occurs in the antecedent (succedent) of a conclusion of
a rule of inference, it is either obtained by a logical rule or by a contraction rule.

3. The left premiss of the mix rule is the conclusion of a contraction rule
(→ contr). The sub-tree of D containing mix is:

Γ, Σ∗ −→ ∆, Θ

305
 ^
(mix)

Γ −→ ∆, A Σ −→ Θ
| (→ contr)
Γ −→ ∆

We transform it, and replace it in D by

Γ, Σ∗ −→ ∆, Θ

possibly several weakenings and exchanges

Γ −→ ∆

Observe that the whole branch of D that starts with the node Σ −→ Θ disap-
pears. Such obtained D∗ proves the same sequent and contains no mix.

4. The right premiss of the mix rule is the conclusion of a contraction rule
(→ contr). It is a dual case to 3. and is left to the reader.

GROUP 2. Logical Rules.

1. The main connective of the mix formula is ∩, i.e. the mix formula is (A ∩ B).
The left premiss of the mix rule is the conclusion of a rule (→ ∩). The right
premiss of the mix rule is the conclusion of a rule (∩ →)1 .
The sub-tree T of D containing mix is:

Γ, Σ −→ ∆, Θ
^
(mix)

Γ −→ ∆, (A ∩ B) (A ∩ B), Σ −→ Θ
^
((→ ∩)) | (∩ →)1
A, Σ −→ Θ

Γ −→ ∆, A Γ −→ ∆, B

We transform T into T∗ as follows.

306
 Γ, Σ −→ ∆, Θ

possibly several weakenings and exchanges

Γ, Σ∗ −→ ∆∗ , Θ
^
(mix)

Γ −→ ∆, A A, Σ −→ Θ

We replace T by T∗ in D and obtain D∗ . Now we can apply induction hypothe-

sis with respect to the degree of the mix formula. The mix formula A in D∗ has
a lower degree then the mix formula (A ∩ B) and by the inductive assumption
the derivation D∗ , and hence D may be transformed into one without mix.

2. The case when the left premiss of the mix rule is the conclusion of a rule
(→ ∩) and right premiss of the mix rule is the conclusion of a rule (∩ →)2
3. The main connective of the mix formula is ∪, i.e. the mix formula is (A ∪ B).
The left premiss of the mix rule is the conclusion of a rule (→ ∪)1 or (→ ∪)2 .
The right premiss of the mix rule is the conclusion of a rule (∪ →)1 . This is to
be dealt with symmetrically to the ∩ cases.
4. The main connective of the mix formula is ¬, i.e. the mix formula is ¬A.
The left premiss of the mix rule is the conclusion of a rule (→ ¬). The right
premiss of the mix rule is the conclusion of a rule (¬ →).

Here is the sub-tree T of D containing the application of the mix rule.

Γ, Σ −→ ∆, Θ
^
(mix)

Γ −→ ∆, ¬A ¬A, Σ −→ Θ
| (→ ¬) | (¬ →)
A, Γ −→ ∆ Σ −→ Θ, A

We transform T into T∗ as follows.

Γ, Σ −→ ∆, Θ

possibly several weakenings and exchanges

307
 Σ, Γ∗ −→ Θ∗ , ∆
^
(mix)

Σ −→ Θ, A A, Γ −→ ∆

We replace T by T∗ in D and obtain D∗ . The new mix in D∗ may be eliminated

by virtue of inductive assumption, and so from the derivation D.

5. The main connective of the mix formula is ⇒, i.e. the mix formula is
(A ⇒ B). The left premiss of the mix rule is the conclusion of a rule ((→⇒).
The right premiss of the mix rule is the conclusion of a rule (⇒→).

Here is the sub-tree T of D containing the application of the mix rule.

Γ, Σ −→ ∆, Θ
^
(mix)

Γ −→ ∆, (A ⇒ B) (A ⇒ B), Σ −→ Θ
^
| (→⇒) ((→ ∩))
A, Γ −→ ∆, B

Σ −→ Θ, A B, Σ −→ Θ,

We transform Tinto T∗ as follows.

Γ, Σ −→ ∆, Θ

possibly several weakenings and exchanges

Σ, Γ∗ , Σ∗∗ −→ Θ∗ , ∆∗ , Θ
^
(mix)

Σ −→ Θ, A A, Γ, Σ∗ , −→ ∆∗ , Θ
^
(mix)

A, Γ −→ ∆, B B, Σ −→ Θ,

308
The asteriks are, of course, intended as follows: Σ∗ , ∆∗ results from Σ, ∆ by
the omission of all formulas B; Γ∗ , Σ∗∗ , Θ∗ results from Γ, Σ∗ , Θ by the omission
of all formulas A.
We replace T by T∗ in D and obtain D∗ . Now we have two mixes, but both
mix formulas are of a lower degree then n. We first apply the inductive to the
assumption to the lower mix. Thus it can be eliminated. We can then also
eliminate the upper mix. This ends the proof of the case of rank r=2.

Case r > 2.
In the case r = 2, we generally reduced the derivation to one of lower degree.
Now we shall proceed to reduce the derivation to one of the same degree, but
of a lower rank. This allows us to to be able to carry the induction with respect
to the rank r of the derivation.
We use the inductive assuption in all cases except, as before, a case of an axiom
or structural rules. In these cases the mix can be eliminated immediately, as it
was eliminated in the previous case of rank r = 2.
In a case of logical rules we obtain the reduction of the mix of the lemma to
derivations with mix of a lower ranks which consequently can be eleminated by
the inductive assumption. We carry now proofs for two logical rules: (→ ∩)
and (∪ →. The proof for all other rules is similar and is left to the reader.
Also, we consider a case of left rank Lr= 1 and the right rank Rr = r ¿1. The
symmetrical case left rank Lr = r ¿1 1 and the right rank Rr = 1 is left to the
reader as an exercise.

Case: Lr = 1, Rr = r > 1. The right premiss of the mix is a conclusion of the

rule (→ ∩), i.e. it is of a form Γ −→ ∆, (A∩B) and Γ contains the mix formula
M . The left premiss of the mix is a sequent Θ −→ Σ and Θ contains the mix
formula M . The end of the derivation D, .i.e. the sub-tree T of D containing
mix is:

Θ, Γ∗ −→ Σ∗ , ∆, (A ∩ B)
^
(mix)

Θ −→ Σ Γ −→ ∆, (A ∩ B)
^
(→ ∩)

Γ −→ ∆, A Γ −→ ∆, B

We transform T into T∗ as follows.

309
 Θ, Γ∗ −→ Σ∗ , ∆, (A ∩ B)
^
(→ ∩)

Θ, Γ∗ −→ Σ∗ , ∆, A Θ, Γ∗ −→ Σ∗ , ∆, B
^ ^
(mix) (mix)

Θ −→ Σ Γ −→ ∆, A Θ −→ Σ Γ −→ ∆, A

We replace T by T∗ in D and obtain D∗ . Now we have two mixes, but both

have the right rank Rr = r-1 and both of them can be eliminated by the induc-
tive assumption.

Case: Lr = 1, Rr = r > 1. The right premiss of the mix is a conclusion of the

rule (∪ →, i.e. it is of a form (A ∪ B), Γ −→ ∆ and Γ contains the mix formula
M . The left premiss of the mix is a sequent Θ −→ Σ and Θ contains the mix
formula M . The end of the derivation D, .i.e. the sub-tree T of D containing
mix is:

Θ, (A ∪ B)∗ , Γ∗ −→ Σ∗ , ∆, (A ∩ B)
^
(mix)

Θ −→ Σ (A ∪ B)Γ −→ ∆
^
(∪ →)

A, Γ −→ ∆ B, Γ −→ ∆

∗
(A ∪ B) stands either for or for nothing according as (A ∪ B) is unequal or
equal to the mix formula M . The mix formula M certainly occurs in Γ. For
otherwise M would been equal to (A ∪ B) and the right rank Rr would be equal
to 1 contrary to the assumption.
We transform T into T∗ as follows.

Θ, (A ∪ B), Γ∗ −→ Σ∗ , ∆, (A ∩ B)
^
(∪ →)

310
 A, Θ, Γ∗ −→ Σ∗ , ∆ B, Θ, Γ∗ −→ Σ∗ , ∆
some weakenings, exchanges some weakenings, exchanges
∗ ∗ ∗
Θ, A , Γ −→ Σ , ∆ Θ, B ∗ , Γ∗ −→ Σ∗ , ∆
^ ^
(mix) (mix)

Θ −→ Σ A, Γ −→ ∆ Θ −→ Σ B, Γ −→ ∆

Now we have two mixes, but both have the right rank Rr = r-1 and both of
them can be eliminated by the inductive assumption. We replace T by T∗ in
D and obtain D∗ . This ends the proof of the Hauptzatz lemma and hence the
proof of the Hauptzatz Theorem 6.19 and Hauptzatz Theorem 6.20.
Let’s denote by LK - c and LI - c the systems LK, LI without the cut rule,
i.e. we put
LK − c = LK − {(cut)}. (6.37)

LI − c = LI − {(cut)}. (6.38)

We re-write the Hauptzatz Theorems as follows.

Theorem 6.21 ( LK Hauptzatz)

For every LK sequent Γ −→ ∆,

`LK Γ −→ ∆ if and only if `LK−c Γ −→ ∆.

Theorem 6.22 ( LI Hauptzatz)

For every LI sequent Γ −→ ∆,

`LK Γ −→ ∆ if and only if `LK−c Γ −→ ∆.

This is why the cut-free Gentzen systems LK-c and LI -c are just called Gentzen
LK, LI, respectively.
Directly from the Completeness Theorem 6.18 and the Hauptzatz Theorem 6.19
we get that the following.

Theorem 6.23 (LK-c Completeness)

For any sequent Γ −→ ∆,

`LK−c Γ −→ ∆ if and only if |= Γ −→ ∆.

311
Let G be the Gentzen sequents proof system defined by (6.27). We replace the
logical axiom of G
Γ0 1 , a, Γ0 2 −→ ∆0 1 , a, ∆0 2 ,
where a ∈ V AR is any propositional variable and Γ0 1 , Γ0 2 , ∆0 1 , ∆0 2 ∈ V AR∗
are any indecomposable sequences, by a new logical axiom

Γ1 , A, Γ2 −→ ∆1 , A, ∆2 (6.39)

for any A ∈ F and any sequences Γ1 , Γ2 , ∆1 , ∆2 ∈ SQ. We call a resulting proof

system GK, i.e. we have that

GK = ( L{∪,∩,⇒,¬} , SQ, LA, R ) (6.40)

where LA is the axiom (6.39) and R is the set (6.29) of rules of G.

Observe that the only difference between the systems GK and G is the form
of their logical axioms, both being tautologies. Hence hence get the proof com-
pleteness of GK in the same way as we proved it for G, i.e. we have the
following.

Theorem 6.24
For any formula A ∈ F,

`GK A if and only if |= A.

For any sequent Γ −→ ∆ ∈ SQ,

`GK Γ −→ ∆ if and only if |= Γ −→ ∆.

By the GK the completeness theorem 6.24, LK-c completeness theorem 6.23

we get the equivalency of GK and the cut free LK-c.

Theorem 6.25 (LK, GK Equivalency)

The proof systems GK and the cut freeLK are equivalent, i.e for any sequent
Γ −→ ∆,
`LK Γ −→ ∆ if and only if `GK Γ −→ ∆.

6.8 Homework Problems

1. Write all proofs in GL of (¬(a ∩ b) ⇒ (¬a ∪ ¬b))).
2. Find a formula which has a unique decomposition tree in GL.
3. Define shortly, in your own words, for any formula A ∈ F, its decomposi-
tion tree T→A in G.

312
 4. Extend your definition T→A in G to a decomposition tree TΓ→∆ .

5. Prove that for any Γ → ∆ ∈ SQ, the decomposition tree TΓ→∆ in G are
finite.

6. Write all proofs in G of (¬(a ∩ b) ⇒ (¬a ∪ ¬b))).

7. Find a formula A which has a unique decomposition tree in G.

8. Prove strong soundness of rules (→ ∪), (→⇒) in GL. List all logical equiv-
alences used in the proofs.

9. Prove strong soundness of rules (⇒→), (¬ →) in GL. List all logical equiv-
alences used in the proofs.

10. Prove strong soundness of rules (∪ →), (→ ¬), (∩ →) in G. List all logical
equivalences used in the proofs.

11. Prove strong soundness of rules (⇒→), (→ ∪), (⇒→) in G. List all logical
equivalences used in the proofs.

12. Explain why the system G is strongly sound.

13. Prove the following.

For any sequent Γ −→ ∆ ∈ SQ,
if `G Γ −→ ∆, then |= Γ −→ ∆.

14. Given a formula A = ((b ⇒ (a ∩ c)) ⇒ (¬(a ∪ c) ⇒ (¬b ∪ a))).

(i) Find all counter models determined by the decomposition trees of A
in GL. Explain why the definition of a counter model determined by the
decomposition tree is correct.
(ii) Find all counter models determined by the decomposition trees of A
in G. Explain why the definition of a counter model determined by the
decomposition tree is correct.

15. Prove the following.

Given a sequent Γ −→ ∆, such that its decomposition tree TΓ−→∆ in G
contains a non- axiom leaf LA . Any truth assignment v that falsifies the
non-axiom leaf LA is a counter model for Γ −→ ∆.

16. Prove the following.

For any sequent Γ −→ ∆ ∈ SQ,
`G Γ −→ ∆ if and only if |= Γ −→ ∆.

17. Let LK-c = LK − {(cut)} and GK be proof systems defined as defined

by (6.37) and (6.40), respectively
(i) We know that GK is strongly sound. Prove that LK-c is sound but
not strongly sound.

313
 (ii) Find proofs of axioms A3, A7, and A11 of the R system (6.34) in LK-c
and in GK, i.e. proofs of formulas (B ⇒ (A ∪ B)), ((C ⇒ A) ⇒ ((C ⇒
B) ⇒ (C ⇒ (A ∩ B))), and ((A ⇒ (A ∩ ¬A)) ⇒ ¬A). Compare your
results.
(iii) Find proofs of axioms A1, A8, and A9 of the R system (6.34) in LK-
c and in GK, i.e. proofs of formulas ((A ⇒ B) ⇒ ((B ⇒ C) ⇒ (A ⇒
C))), ((A ⇒ (B ⇒ C)) ⇒ ((A ∩ B) ⇒ C)), and (((A ∩ B) ⇒ C) ⇒ (A ⇒
(B ⇒ C)). Compare your results.
(iv) Find proofs of axioms A1, A5, and A12 of the R system (6.34) in
LK-c and in GK, i.e. proofs of formulas ((A ⇒ B) ⇒ ((B ⇒ C) ⇒ (A ⇒
C))), ((A ∩ B) ⇒ A), and (A ∪ ¬A) . Compare your results.

18. Re- write carefully the proof of the classical Hauptzatz Theorem 6.19 for
the case o the intuitionistic system LI (definition 6.10.
19. Define shortly, in your own words, for any formula A ∈ F, its decomposi-
tion tree TA in LK-c. Is the tree TA always finite?

20. Given a formula A = (¬(a ∩ b) ⇒ (¬a ∪ ¬b))). Construct one infinite and
one infinite decomposition tree for A.
21. Describe major differences in the decomposition trees in LK-c and GK.
22. We have proved that LK-c and GK are equivalent, i.e. that for any
sequent Γ −→ ∆,

`LK−c Γ −→ ∆ if and only if `GK Γ −→ ∆.

The proof was not constructive; it was obtained from that fact that both
systems are complete.
(ii) Describe a constructive procedure of transforming any proof in GK
into a proof in LK-c.
(i) Transform a proof of a formula (A ⇒ (A ∪ B)) in GK into a proof in
LK-c.
(ii) Describe a constructive procedure of transforming any proof in GK
into a proof in LK-c.
(iii) Show that the procedure of elimination of structural rules of LK-c
leads to the rules inference of GK .

314
Chapter 7

Introduction to
Intuitionistic and Modal
Logics

7.1 Introduction to Intuitionictic Logic

Intuitionistic logic has developed as a result of certain philosophical views on

the foundation of mathematics, known as intuitionism. Intuitionism was orig-
inated by L. E. J. Brouwer in 1908. The first Hilbert style formalization of the
intuitionistic logic, formulated as a proof system, is due to A. Heyting (1930).
In this chapter we present a Hilbert style proof system I that is equivalent to
the Heyting’s original formalization and discuss the relationship between intu-
itionistic and classical logic.
There have been, of course, several successful attempts at creating semantics for
the intuitionistic logic,i.e. to define formally a notion of the intuitionistic tau-
tology. The most recent are Kripke models were defined by Kripke in 1964. The
first intuitionistic semantics was defined in a form of pseudo- Boolean algebras
by McKinsey, Tarski in 1944 - 1946. McKinsey, Tarski algebraic approach to
the intuitionostic semantics (and classical) was followed by many authors and
developed into a new field of Algebraic Logic. The pseudo- Boolean algebras are
called also Heyting algebras.
An uniform presentation of algebraic models for classical, intuitionistic and
modal logics was first given in a now classic algebraic logic book: ”Mathematics
of Metamathematics”, Rasiowa, Sikorski (1964).
The goal of this chapter is to give a presentation of the intuitionistic logic formu-
lated as a proof system, discuss its algebraic semantics and the basic theorems

315
that establish the relationship between classical and intuitionistic logics.

7.1.1 Philosophical Motivation

Intuitionists’ view-point on the meaning of the basic logical and set theoretical
concepts used in mathematics is different from that of most mathematicians in
their research.
The basic difference lies in the interpretation of the word exists. For exam-
ple, let A(x) be a statement in the arithmetic of natural numbers. For the
mathematicians the sentence
∃xA(x) (7.1)

is true if it is a theorem of arithmetic, i.e. if it can be deduced from the axioms

of arithmetic by means of classical logic. If a mathematician proves sentence
( 7.1), this does not mean that he is able to indicate a method of construction
of a natural number n such that A(n) holds.
For the intuitionist the sentence (7.1) is true only he is able to provide a con-
structive method of finding a number n such that A(n) is true.
Moreover, the mathematician often obtains the proof of the existential sentence
(7.1), i.e. of the sentence ∃xA(x) by proving first a sentence

¬∀x ¬A(x). (7.2)

Next he makes use of a classical tautology

(¬∀x ¬A(x)) ⇒ ∃xA(x)). (7.3)

By applying Modus Ponens to (7.2) and (7.3) he obtains (7.1).

For the intuitionist such method is not acceptable, for it does not give any
method of constructing a number n such that A(n) holds. For this reason the
intuitionist do not accept the classical tautology (7.3) i.e. (¬∀x ¬A(x)) ⇒
∃xA(x)) as intuitionistic tautology, or as as an intuitionistically provable sen-
tence.
Let us denote by `I A and |=I A the fact that A is intuitionistically provable
and that A is intuitionistic tautology, respectively. The proof system I for the
intuitionistic logic has hence to be such that

6 `I (¬∀x ¬A(x)) ⇒ ∃xA(x)).

The intuitionistic semantics I has to be such that one can prove in that also

6 |=I (¬∀x ¬A(x)) ⇒ ∃xA(x)).

316
The above means also that intuitionists interpret differently the meaning of
propositional connectives.
Intuitionistic implication
The intuitionistic implication (A ⇒ B) is considered by to be true if there exists
a method by which a proof of B can be deduced from the proof of A. In the
case of the implication
(¬∀x ¬A(x)) ⇒ ∃xA(x))
there is no general method which, from a proof of the sentence (¬∀x ¬A(x)),
permits is to obtain an intuitionistic proof of the sentence ∃xA(x), i.e. to
construct a number n such that A(n) holds, hence we can’t accept it as an
intuitionistic theorem or tautology.
Intuitionistic negation
The negation and the disjunction are also understood differently. The sentence
¬A is considered intuitionistically true if the acceptance of the sentence A leads
to absurdity.
As a result of above understanding of negation and implication we have that in
the intuitionistic logic I
`I (A ⇒ ¬¬A)
but
6 `I (¬¬A ⇒ A).

Consequently, the intuitionistic semantics I has to be such that

|=I (A ⇒ ¬¬A)
and
6 |=I (¬¬A ⇒ A).

Intuitionistic disjunction
The intuitionist regards a disjunction (A ∪ B) as true if one of the sentences
A, B is true and there is a method by which it is possible to find out which of
them is true. As a consequence a classical law of excluded middle
(A ∪ ¬A)
is not acceptable by the intuitionists since there is no general method of finding
out, for any given sentence A, whether A or ¬A is true. This means that the
intuitionistic logic must be such that
6 `I (A ∪ ¬A)
and the intuitionistic semantics I has to be such that
6|=I (A ∪ ¬A).

317
Intuitionists’ view of the concept of infinite set also differs from that which is
generally accepted in mathematics. Intuitionists reject the idea of infinite set as
a closed whole. They look upon an infinite set as something which is constantly
in a state of formation. Thus, for example, the set of all natural numbers is
infinite in the sense that to any given finite set of natural numbers it is always
possible to add one more natural number. The notion of the set of all subsets
of the set of all natural numbers is not regarded meaningful. Thus intuitionists
reject the general idea of a set as defined by a modern set theory.
An exact exposition of the basic ideas of intuitionism is outside the range of our
investigations. Our goal is to give a presentation of of the intuitionistic logic,
which is a sort of reflection of intuitionistic ideas formulated as a proof system.

7.1.2 Algebraic Intuitionistic Semantics and Completeness

Theorem

There are many proof systems describing the intuitionistic logic. We define now
a system I with a set of axioms that is due to Rasiowa (1959). We adopted
this axiomatization for two reasons. Firs is that it is the most natural and
appropriate set of axioms to carry the the algebraic proof of the completeness
theorem and the second is that they visibly describe the main difference between
intuitionistic and classical logic. Namely, by adding the only one more axiom
(A ∪ ¬A) we get a (complete) formalization for classical logic. Here are the
components if the proof system I.
Language We adopt a propositional language L = L{∪,∩,⇒,¬} with the set of
formulas denoted by F.
Axioms
A1 ((A ⇒ B) ⇒ ((B ⇒ C) ⇒ (A ⇒ C))),
A2 (A ⇒ (A ∪ B)),
A3 (B ⇒ (A ∪ B)),
A4 ((A ⇒ C) ⇒ ((B ⇒ C) ⇒ ((A ∪ B) ⇒ C))),
A5 ((A ∩ B) ⇒ A),
A6 ((A ∩ B) ⇒ B),
A7 ((C ⇒ A) ⇒ ((C ⇒ B) ⇒ (C ⇒ (A ∩ B))),
A8 ((A ⇒ (B ⇒ C)) ⇒ ((A ∩ B) ⇒ C)),
A9 (((A ∩ B) ⇒ C) ⇒ (A ⇒ (B ⇒ C)),
A10 (A ∩ ¬A) ⇒ B),
A11 ((A ⇒ (A ∩ ¬A)) ⇒ ¬A),

318
where A, B, C are any formulas in L.

Rules of inference
We adopt the Modus Ponens rule

A ; (A ⇒ B)
(M P )
B

as the only inference rule.

A proof system
I = ( L, F A1 − A11, (M P ) ), (7.4)
A1 − A11 defined above, is called a Hilbert style formalization for Intuitionistic
propositional logic.
We introduce, as usual, the notion of a formal proof in I and denote by

`I A

the fact that A has a formal proof in I, or that that A is intuitionistically

provable in I.

7.1.3 Algebraic Semantics and Completeness Theorem

We shortly present here Tarski, Rasiowa, and Sikorski psedo-Boolean algebra

semantics and discuss the algebraic completeness theorem for the intuitionistic
propositional logic. We leave the Kripke semantics for the reader to explore
from other, multiple sources.
Here are some algebraic basic definitions.
Relatively Pseudo-Complemented Lattice
A lattice (B, ∩, ∪) is said to be relatively pseudo-complemented (Birkhoff, 1935)
if for any elements a, b ∈ B, there exists the greatest element c, such that
a ∩ c ≤ b. Such element is denoted by a ⇒ b and called the pseudo-complement
of a relative to b. By definition

x ≤ a ⇒ b if and only if a ∩ x ≤ b for all x, a, b ∈ B. (7.5)

The equation (7.5) can serve as the definition of the relative pseudo-complement
a ⇒ b.

Fact 7.1 Every relatively pseudo-complemented lattice (B, ∩, ∪) has the greatest
element, called a unit element and denoted by 1.

319
Proof Observe that a∩x ≤ a for all x, a ∈ B. By (7.5) we have that x ≤ a ⇒ a
for all x ∈ B, i.e. a ⇒ a = 1.
An abstract algebra
B = (B, 1, ⇒, ∩, ∪, ⇒) (7.6)
is said to be a relatively pseudo-complemented lattice if (B, ∩, ∪) is rel-
atively pseudo-complemented lattice with the relative pseudo-complement ⇒
defined by (7.5) and with the unit element 1 (Fact 7.1 ).
Relatively Pseudo-complemented Set Lattices
Consider a topological space X with an interior operation I. Let G(X) be the
class of all open subsets of X and G ∗ (X) be the class of all both dense and open
subsets of X. Then the algebras

(G(X), X, ∪, ∩, ⇒), (G ∗ (X), X, ∪, ∩, ⇒),

where ∪, ∩ are set-theoretical operations of union, intersection, and ⇒ is defined

by
Y ⇒ Z = I(X − Y ) ∪ Z
are relatively pseudo-complemented lattices.
Clearly, all sub algebras of these algebras are also relatively pseudo-complemented
lattices, called relatively pseudo-complemented set lattices. They are typical ex-
amples of relatively pseudo-complemented lattices
Pseudo - Boolean Algebra (Heyting Algebra)
An algebra
B = (B, 1, 0, ⇒, ∩, ∪, ¬), (7.7)
is said to be a pseudo - Boolean algebra if and only if (B, 1, ⇒, ∩, ∪) it is
a relatively pseudo-complemented lattice (7.6) in which a zero element 0 exists
and ¬ is a one argument operation defined as follows

¬a = a ⇒ 0 (7.8)

The operation ¬ is called a pseudo-complementation.

The pseudo - Boolean algebras are also called Heyting algebras to stress their
connection to the intuitionistic logic.

Let X be topological space with an interior operation I. Let G(X) be the class
of all open subsets of X. Then

(G(X), X, ∅, ∪, ∩, ⇒, ¬), (7.9)

where ∪, ∩ are set-theoretical operations of union, intersection, ⇒ is defined by

Y ⇒ Z = I(X − Y ) ∪ Z

320
and ¬ is defined as

¬Y = Y ⇒ ∅ = I(X − Y ), for all Y ⊆ X

is a pseudo - Boolean algebra.

Every sub algebra of G(X) is also a pseudo-Boolean algebra. They are called
pseudo-fields of sets.
The following Theorem 7.1 states that pseudo-fields are typical examples of
pseudo - Boolean algebras. The theorems of this type are often called Stone
Representation Theorems to remember an American mathematician H.M
Stone. He was one of the first to initiate the investigations between logic and
general topology in the article ”The Theory of Representations for Boolean
Algebras”,Trans. of the Amer.Math, Soc 40, 1036.

Theorem 7.1 (Representation Theorem) (McKinsey, Tarski, 1946)

For every pseudo - Boolean (Heyting) algebra (7.7)

B = (B, 1, 0, ⇒, ∩, ∪, ¬),

there exists a monomorphism h of B into a pseudo-field (7.9) G(X) of all open

subsets of a compact topological T0 space X.

Another typical (and interesting) example of a class of pseudo - Boolean algebras

is the following.
Linear Pseudo - Boolean Algebra
Let (B, ≤) be a chain (linearly ordered set) with the greatest element 1 and the
least element (smallest) 0.
An algebra
H = (B, 1, 0, ⇒, ∩, ∪, ¬) (7.10)
is called a linear pseudo - Boolean algebra if and only if its operations are defined
as follows.
For any a, b ∈ B,

a ∪ b = max{a, b}, a ∩ b = min{a, b},


1 if a ≤ b
a⇒b= (7.11)
b otherwise,

and define the pseudo-complementation ¬ as

¬a = a ⇒ 0.

We leave the proof that (7.10) is a pseudo-Boolean algebra as a homework exer-

cise. Observe that the linear pseudo - Boolean algebra (7.10) is a generalization
of the 3-valued Heyting semantics defined in chapter 3.

321
Algebraic Models
We say that a formula A is an intuitionistoc tautology if and only if any
pseudo-Boolean algebra (7.7) is a model for A. This kind of models because
their connection to abstract algebras are called algebraic models.
We put it formally as follows.

Definition 7.1 (Intuitionistic Algebraic Model)

Let A be a formula of the language L{∪,∩,⇒,¬} and let B = (B, 1, 0, ⇒
, ∩, ∪, ¬) be a pseudo - Boolean topological Boolean algebra (7.7).
We say that the algebra B is a model for the formula A and denote it by

B |= A

if and only if v ∗ (A) = 1 holds for all variables assignments v : V AR −→ B.

Definition 7.2 (Intuitionistic Tautology)

The formula A is an intuitionistic tautology and is denoted by

|=I A if and only if B |= A

for all pseudo-Boolean algebras B.

In Algebraic Logic the notion of tautology is often defined using a notion ”a

formula A is valid in an algebra B ”. It is formally defined as follows.

Definition 7.3
A formula A is valid in a pseudo-Boolean algebra B = (B, 1, 0, ⇒, ∩, ∪, ¬),
if and only if v ∗ (A) = 1 holds for all variables assignments v : V AR −→ B.

Directly from definition 7.2 and definition 7.3 we get the following.

Fact 7.2 For any formula A, |=I A if and only if A is valid in all pseudo-
Boolean algebras.

We write now `I A to denote any proof system for the Intuitionistic proposi-
tional logic, and in particular the Hilbert style formalization for Intuitionistic
propositional logic I defined by (7.4).

Theorem 7.2 (Intuitionistic Completeness Theorem) (Mostowski 1948)

For any formula A of L{∪, ∩,⇒,¬} ,

`I A if and only if |=I A.

322
The intuitionistic completeness theorem 7.2 follows also directly from the fol-
lowing general algebraic completeness theorem 7.3 that combines results of of
Mostowski (1958), Rasiowa (1951) and Rasiowa-Sikorski (1957).

Theorem 7.3 (Algebraic Completeness Theorem)

For any formula A of L= L{∪,∩,⇒,¬} the following conditions are equivalent.
(i) `I A,
(ii) |=I A,
(iii) A is valid in every pseudo-Boolean algebra

(G(X), X, ∅, ∪, ∩, ⇒, ¬)

of open subsets of any topological space X,

r
(iv) A is valid in every pseudo-Boolean algebra B with at most 22 elements,
where r is the number of all sub formulas of A.
Moreover, each of the conditions (i) - (iv) is equivalent to the following one.
(v) A is valid in the pseudo-Boolean algebra (G(X), X, ∅, ∪, ∩, ⇒, ¬) of
open subsets of a dense-in -itself metric space X 6= ∅ (in particular of an n-
dimensional Euclidean space X).

The following theorem follows from the equivalence of conditions (i) and (iv).

Theorem 7.4 (Decidability)

Every proof system for the intuitionistic propositional logic is decidable.

Examples of intuitionistic propositional tautologies

The following classical tautologies are provable in I and hence are also intu-
itionistic tautologies.
(A ⇒ A), (7.12)
(A ⇒ (B ⇒ A)), (7.13)
(A ⇒ (B ⇒ (A ∩ B))), (7.14)
((A ⇒ (B ⇒ C)) ⇒ ((A ⇒ B) ⇒ (A ⇒ C))), (7.15)
(A ⇒ ¬¬A), (7.16)
¬(A ∩ ¬A), (7.17)
((¬A ∪ B) ⇒ (A ⇒ B)), (7.18)
(¬(A ∪ B) ⇒ (¬A ∩ ¬B)), (7.19)

323
 ((¬A ∩ ¬B) ⇒ (¬(A ∪ B)), (7.20)
((¬A ∪ ¬B) ⇒ (¬A ∩ ¬B)), (7.21)
((A ⇒ B) ⇒ (¬B ⇒ ¬A)), (7.22)
((A ⇒ ¬B) ⇒ (B ⇒ ¬A)), (7.23)
(¬¬¬A ⇒ ¬A), (7.24)
(¬A ⇒ ¬¬¬A), (7.25)
(¬¬(A ⇒ B) ⇒ (A ⇒ ¬¬B)), (7.26)
((C ⇒ A) ⇒ ((C ⇒ (A ⇒ B)) ⇒ (C ⇒ B)), (7.27)

Examples of classical tautologies that are not intuitionistic tautologies

The following classical tautologies are not intuitionistic tautologies.

(A ∪ ¬A), (7.28)
(¬¬A ⇒ A), (7.29)
((A ⇒ B) ⇒ (¬A ∪ B)), (7.30)
(¬(A ∩ B) ⇒ (¬A ∪ ¬B)), (7.31)
((¬A ⇒ B) ⇒ (¬B ⇒ A)), (7.32)
((¬A ⇒ ¬B) ⇒ (B ⇒ A)), (7.33)
((A ⇒ B) ⇒ A) ⇒ A), (7.34)

7.1.4 Connection Between Classical and Intuitionistic Tau-

tologies

The intuitionistic logic has been created as a rival to the classical one. So a
question about the relationship between these two is a natural one. We present
here some examples of tautologies and some historic results about the connection
between the classical and intuitionistic logic.
The first connection is quite obvious. It was proved by Rasiowa and Sikorski in
1964 that by adding the axiom
A12 (A ∪ ¬A)
to the set of axioms of our system I defined by (7.4) we obtain a Hilbert proof
system H that is complete with respect to classical semantics.
This proves the following.

324
Theorem 7.5
Every formula that is derivable intuitionistically is classically derivable, i.e.

if `I A, then ` A,

where we use symbol ` for classical (complete classical proof system) provability.

We write
|= A and ] |=I A
to denote that A is a classical and intuitionistic tautology, respectively.
As both proof systems, I and H are complete under respective semantics, we
can state this as the following relationship between classical and intuitionistic
tautologies.

Theorem 7.6
For any formula A ∈ F,

if |=I A, then |= A.

The next relationship shows how to obtain intuitionistic tautologies from the
classical tautologies and vice versa. It has been proved by Glivenko in 1929 and
independently by Tarski in 1938.

Theorem 7.7 (Glivenko, Tarski)

For any formula A ∈ F, A is a classically provable if and only if ¬¬A is an
intuitionistically provable, i.e.

`I A if and only if ` ¬¬A

where we use symbol ` for classical provability.

Theorem 7.8 (McKinsey, Tarski, 1946)

For any formula A ∈ F, A is a classical tautology if and only if ¬¬A is an
intuitionistic tautology, i.e.

|= A if and only if |=I ¬¬A.

The following relationships were proved by Gödel.

325
Theorem 7.9 (Gödel, 1931)
For any A, B ∈ F, a formula (A ⇒ ¬B) is a classically provable if and only if
it is an intuitionistically provable, i.e.

` (A ⇒ ¬B) if and only if `I (A ⇒ ¬B).

Theorem 7.10 (Gödel, 1931)

If a formula A contains no connectives except ∩ and ¬, then A is a classically
provable if and only if it is an intuitionistically provable.

By the completeness of classical and intuitionisctic logics we get the following

equivalent semantic form of theorems 7.9 and 7.10.

Theorem 7.11
A formula (A ⇒ ¬B) is a classical tautology if and only if it is an intuitionistic
tautology, i.e.

|= (A ⇒ ¬B) if and only if |=I (A ⇒ ¬B).

Theorem 7.12
If a formula A contains no connectives except ∩ and ¬, then A is a classical
tautology if and only if it is an intuitionistic tautology.

On intuitionistically derivable disjunction

In a classical logic it is possible for the disjunction (A ∪ B) to be a tautology

when neither A nor B is a tautology. The tautology (A ∪ ¬A) is the simplest
example. This does not hold for the intuitionistic logic.

This fact was stated without the proof by Gödel in 1931 and proved by Gentzen
in 1935 via his proof system LI which is presented in chapter 6 and discussed
in the next section 7.2.

Theorem 7.13 (Gentzen 1935)

A disjunction (A ∪ B) is intuitionistically provable if and only if either A or B
is intuitionistically provable , i.e.

`I (A ∪ B) if and only if `I A or `I B.

326
We obtain, via the Completeness Theorem 7.2 the following equivalent semantic
version of the above.

Theorem 7.14
A disjunction (A ∪ B) is intuitionistic tautology if and only if either A or B is
intuitionistic tautology, i.e.

|=I (A ∪ B) if and only if |=I A or |=I B.

7.2 Gentzen Sequent System LI

In 1935 G. Gentzen formulated a first syntactically decidable formalizations for
classical and intuitionistic logic and proved its equivalence with the Heyting’s
original Hilbert style formalization. He named his classical system LK (K for
Klassisch) and intuitionistic system LI (I for Intuitionistisch). In order to prove
the completeness of the system LK and proving the adequacy of LI he intro-
duced a special rule, called cut rule that corresponds to the Modus Ponens
rule in Hilbert proof systems. Then, as the next step he proved the now famous
Gentzen Hauptzatz, called in English the Cut Elimination Theorem.
The Gentzen original proof system LI is a particular case of his proof system
LK for the classical logic. Both of them were presented in chapter 6 together
with the proof of the Hauptzatz for both, LK and LI systems.
The elimination of the cut rule and the structure of other rules makes it possible
to define effective automatic procedures for proof search, what is impossible in
a case of the Hilbert style systems.
The Gentzen system LI is defined as follows.
Language of LI
Let SQ = { Γ −→ ∆ : Γ, ∆ ∈ F ∗ } be the set of all Gentzen sequents built out
of the formulas of the language

L = L{∪,∩,⇒,¬} (7.35)

and the additional symbol −→.

In order to describe the the intuitionistic logic we deal, after Gentzen, only with
sequents of the form Γ −→ ∆, where ∆ consists of at most one formula. I.e.
we assume that all LI sequents are elements of a following subset ISQ of the
set SQ of all sequents.

ISQ = {Γ −→ ∆ : ∆ consists of at most one formula }. (7.36)

The set ISQ is called the set of all intuitionistic sequents; the LI sequents.

327
Axioms of LI
As the axioms of LI we adopt any sequent from the set ISQ defined by ( 7.36),
which contains a formula that appears on both sides of the sequent arrow −→,
i.e any sequent of the form

Γ1 , A, Γ2 −→ A, (7.37)

for any formula A ∈ F of the language (7.35) and for any sequences Γ1 , Γ2 ∈ F ∗ .
Inference rules of LI
The set inference rules is divided into two groups: the structural rules and the
logical rules. They are defined as follows.
Structural Rules of LI
Weakening

Γ −→
(→ weak) .
Γ −→ A
A is called the weakening formula.
Contraction

A, A, Γ −→ ∆
(contr →) ,
A, Γ −→ ∆
A is called the contraction formula , ∆ contains at most one formula.
Exchange

Γ1 , A, B, Γ2 −→ ∆
(exchange →) ,
Γ1 , B, A, Γ2 −→ ∆

∆ contains at most one formula.

Logical Rules of LI
Conjunction rules

A, B, Γ −→ ∆ Γ −→ A ; Γ −→ B
(∩ →) , (→ ∩) ,
(A ∩ B), Γ −→ ∆ Γ −→ (A ∩ B)

∆ contains at most one formula.

Disjunction rules

Γ −→ A Γ −→ B
(→ ∪)1 , (→ ∪)2 ,
Γ −→ (A ∪ B) Γ −→ (A ∪ B)

328
 A, Γ −→ ∆ ; B, Γ −→ ∆
(∪ →) ,
(A ∪ B), Γ −→ ∆

∆ contains at most one formula.

Implication rules

A, Γ −→ B Γ −→ A ; B, Γ −→ ∆
(→⇒) , (⇒→) ,
Γ −→ (A ⇒ B) (A ⇒ B), Γ −→ ∆

∆ contains at most one formula.

Negation rules

Γ −→ A A, Γ −→
(¬ →) , (→ ¬) .
¬A, Γ −→ Γ −→ ¬A

Formally we define:

LI = (L, ISQ, LA, Structural rules, Logical rules ), (7.38)

where ISQ is defined by (7.36), Structural rules and Logical rules are the infer-
ence rules defined above, and LA is the axiom defined by the schema (7.37).

We write
`LI Γ −→ ∆
to denote that the sequent Γ −→ ∆ has a proof in LI.
We say that a formula A ∈ F has a proof in LI and write it as

`LI A

when the sequent −→ A has a proof in LI, i.e.

`LI A if and only if `LI −→ A.

The completeness of of our cut-free LI follows directly from LI Hauptzatz Theo-

rem proved in chapter 6 and the Intuitionistic Completeness Theorem 7.2. The
proof is a straightforward adaptation of the proof of cut free LK Completeness
Theorem proved in chapter 6 and is left as a homework exercise.

Theorem 7.15 (Completeness of LI)

For any sequent Γ −→ ∆ ∈ ISQ,

`LI Γ −→ ∆ if and only of |=I Γ −→ ∆.

329
In particular, for any formula A,

`LI A if and only of |=I A.

Theorem 7.16 (Intuitionistically Derivable Disjunction)

For any formulas A, B,

`LI (A ∪ B) if and only if `LI A or `LI B.

In particular, a disjunction (A ∪ B) is intuitionistically provable in any proof

system I if and only if either A or B is intuitionistically provable in I.

The particular form the theorem 7.16 was stated without the proof by Gödel in
1931. The theorem proved by Gentzen in 1935 via his Hauptzatz Theorem.
Proof
Assume `LI (A ∪ B). This equivalent to `LI −→ (A ∪ B). The last step in
the proof of −→ (A ∪ B)i LI must be the application of the rule (→ ∪)1 to
the sequent −→ A, or the application of the rule (→ ∪)2 to the sequent −→ B.
There is no other possibilities. We have proved that `LI (A ∪ B) implies `LI A
or `LI B. The inverse is obvious by respective applications of rules (→ ∪)1
(→ ∪)2 to −→ A and −→ B.

7.2.1 Decomposition Trees in LI

Search for proofs in LI is a much more complicated process then the one in
classical systems RS or GL defined in chapter 6.
Here, as in any other Gentzen style proof system, proof search procedure consists
of building the decomposition trees.
In RS the decomposition tree TA of any formula A, and hence of any sequence
Γ is always unique.
In GL the ”blind search” defines, for any formula A a finite number of decom-
position trees, but it can be proved that the search can be reduced to examining
only one of them, due to the absence of structural rules.
In LI the structural rules play a vital role in the proof construction and hence,
in the proof search. We consider here a number of examples to show the com-
plexity of the problem of examining possible decomposition trees for a given
formula A. We are going to see that the fact that a given decomposition tree
ends with an axiom leaf does not always imply that the proof does not exist.
It might only imply that our search strategy was not good. Hence the problem
of deciding whether a given formula A does, or does not have a proof in LI
becomes more complex then in the case of Gentzen system for classical logic.

330
Before we define a heuristic method of searching for proof and deciding whether
such a proof exists or not in LI we make some observations.

Observation 1: the logical rules of LI are similar to those in Gentzen type

classical formalizations we examined in previous chapters in a sense that
each of them introduces a logical connective.
Observation 2: The process of searching for a proof is, as before a decompo-
sition process in which we use the inverse of logical and structural rules
as decomposition rules.
For example the implication rule:

A, Γ −→ B
(→⇒)
Γ −→ (A ⇒ B)

becomes an implication decomposition rule (we use the same name

(→⇒) in both cases)

Γ −→ (A ⇒ B)
(→⇒) .
A, Γ −→ B
Observation 3: we write our proofs in as trees, instead of sequences of expres-
sions, so the proof search process is a process of building a decomposition
tree. To facilitate the process we write, as before, the decomposition rules,
structural rules included in a ”tree ” form.

For example the the above implication decomposition rule is written as follows.

Γ −→ (A ⇒ B)

| (→⇒)
A, Γ −→ B

The two premisses implication rule (⇒→) written as the tree decomposition
rule becomes

(A ⇒ B), Γ −→
^
(⇒→)
Γ −→ A B, Γ −→

331
For example the structural weakening rule is written as the decomposition rule
is written as

Γ −→ A
(→ weak)
Γ −→

We write it in a tree form as follows.

Γ −→ A

| (→ weak)

Γ −→

We define, as before the notion of decomposable and indecomposable formulas

and sequents as follows.

Decomposable formula is any formula of the degree ≥ 1.

Decomposable sequent is any sequent that contains a decomposable for-

mula.

Indecomposable formula is any formula of the degree 0, i.e. any proposi-

tional variable.

Remark: In a case of formulas written with use of capital letters A, B, C, ..

etc, we treat these letters as propositional variables, i.e. as indecomposable
formulas.

Indecomposable sequent is a sequent formed from indecomposable formulas

only.

Decomposition tree construction (1): given a formula A we construct its

decomposition tree TA as follows.

Root of the tree is the sequent −→ A.

Given a node n of the tree we identify a decomposition rule applicable at this

node and write its premisses as the leaves of the node n.

We stop the decomposition process when we obtain an axiom or all leaves of

the tree are indecomposable.

Observation 4: the decomposition tree TA obtained by the construction (1)

most often is not unique.

332
Observation 5: the fact that we find a decomposition tree TA with non-axiom
leaf does not mean that 6 `LI A. This is due to the role of structural rules
in LI and will be discussed later in the chapter.
We illustrate the problems arising with proof search procedures, i.e. de-
composition trees construction in the next section 7.2.2 and give a heuristic
proof searching procedure in the section 7.2.3.

7.2.2 Proof Search Examples

We perform proof search and decide the existence of proofs in LI for a given
formula A ∈ F by constructing its decomposition trees TA . We examine here
some examples to show the complexity of the problem.
Remark
In the following and similar examples when building the decomposition trees for
formulas representing general schemas we treat the capital letters A, B, C, D...
as propositional variables, i.e. as indecomposable formulas.
Example 1
Determine whether `LI −→ ((¬A ∩ ¬B) ⇒ ¬(A ∪ B)).
This means that we have to construct some, or all decomposition trees of

−→ ((¬A ∩ ¬B) ⇒ ¬(A ∪ B)).

If we find a decomposition tree such that all its leaves are axioms, we have a
proof.
If all possible decomposition trees have a non-axiom leaf, proof of A in LI does
not exist.
Consider the following decomposition tree of −→ ((¬A ∩ ¬B) ⇒ ¬(A ∪ B)).

T1

−→ ((¬A ∩ ¬B) ⇒ (¬(A ∪ B))

| (−→⇒)
(¬A ∩ ¬B) −→ ¬(A ∪ B)
| (−→ ¬)
(A ∪ B), (¬A ∩ ¬B) −→
| (exch −→)
(¬A ∩ ¬B), (A ∪ B) −→
| (∩ −→)

333
 ¬A, ¬B, (A ∪ B) −→
| (¬ −→)
¬B, (A ∪ B) −→ A
| (−→ weak)
¬B, (A ∪ B) −→
| (¬ −→)
(A ∪ B) −→ B
^
(∪ −→)

A −→ B B −→ B
non − axiom axiom

The tree T1 has a non-axiom leaf, so it does not constitute a proof in LI. But
this fact does not yet prove that proof doesn’t exist, as the decomposition tree
in LI is not always unique.
Let’s consider now the following tree.

T2

−→ ((¬A ∩ ¬B) ⇒ (¬(A ∪ B))

| (−→⇒)
(¬A ∩ ¬B) −→ ¬(A ∪ B)
| (−→ ¬)
(A ∪ B), (¬A ∩ ¬B) −→
| (exch −→)
(¬A ∩ ¬B), (A ∪ B) −→
| (∩ −→)
¬A, ¬B, (A ∪ B) −→
| (exch −→)
¬A, (A ∪ B), ¬B −→
| (exch −→)
(A ∪ B), ¬A, ¬B −→
^
(∪ −→)

334
 A, ¬A, ¬B −→ B, ¬A, ¬B −→
| (exch −→) | (exch −→)
¬A, A, ¬B −→ B, ¬B, ¬A −→
| (¬ −→) | (exch −→)
A, ¬B −→ A ¬B, B, ¬A −→
axiom | (¬ −→)
B, ¬A −→ B
axiom

All leaves of T2 are axioms, what proves that T2 is a proof of A and hence we
proved that
`LI ((¬A ∩ ¬B) ⇒ ¬(A ∪ B)).

Example 2
Part 1: Prove that
`LI −→ (A ⇒ ¬¬A),

Part 2: Prove that

6 `LI −→ (¬¬A ⇒ A).

Solution of Part 1
To prove that
`LI −→ (A ⇒ ¬¬A)
we have to construct some, or all decomposition trees of

−→ (A ⇒ ¬¬A).

We treat the sub formulas A, B as indecomposable formulas.

Consider the following decomposition tree.

−→ (A ⇒ ¬¬A).

| (−→⇒)
A −→ ¬¬A
| (−→ ¬)
¬A, A −→

335
 | (¬ −→)
A −→ A
axiom

All leaves of T are axioms what proves that T is a proof of −→ (A ⇒ ¬¬A)

and we don’t need to construct other decomposition trees.

Solution of Part 2
To prove that
6 `LI −→ (¬¬A ⇒ A)
we have to construct all decomposition trees of (A ⇒ ¬¬A) and show that each
of them has an non-axiom leaf.
Consider the first decomposition tree defined as follows.

T1

−→ (¬¬A ⇒ A)

f irst of 2 choices : (→⇒), (→ weak)

| (→⇒)
¬¬A −→ A
f irst of 2 choices : (→ weak), (contr →)
| (→ weak)
¬¬A −→
f irst of 2 choices : (¬ →), (contr →)
| (¬ →)
−→ ¬A
f irst of 2 choices : (¬ →), (→ weak)
| (→ ¬)
A −→
indecomposable
non − axiom

We use the first tree created to define all other possible decomposition trees by
exploring the alternative search paths as indicated at the nodes of the tree.

336
 T1

−→ (¬¬A ⇒ A)

| (−→⇒)
one of 2 choices
¬¬A −→ A
]

| (contr −→)

second of 2 choices
¬¬A, ¬¬A −→ A
| (−→ weak)
f irst of 2 choices
¬¬A, ¬¬A −→
| (¬ −→)
f irst of 2 choices
¬¬A −→ ¬A
| (−→ ¬)
the only choice
A, ¬¬A −→
| (exch −→)
the only choice
¬¬A, A −→
| (−→ ¬)
the only choice
A −→ ¬A
| (−→ ¬)
f irst of 2 choices
A, A −→
indecomposable
non − axiom

337
We can see from the above decomposition trees that the ”blind” construction
of all possible trees only leads to more complicated trees, due to the presence of
structural rules. Observe that the ”blind” application of (contr −→) gives an
infinite number of decomposition trees. To decide that none of them will produce
a proof we need some extra knowledge about patterns of their construction, or
just simply about the number useful of application of structural rules within
the proofs.
In this case we can just make an ”external” observation that the our first tree
T1 is in a sense a minimal one; that all other trees would only complicate this
one in an inessential way, i.e. we will never produce a tree with all axioms
leaves.
One can formulate a deterministic procedure giving a finite number of trees,
but the proof of its correctness require some extra knowledge. We are going to
discuss a motivation and an heuristics for the proof search in the next section.
Within the scope of this book we accept the ”external” explanation for the
heuristics we use as a sufficient solution.
As we can see from the above examples structural rules and especially the
(contr →) rule complicates the proof searching task.
The Gentzen type proof systems RS and GL from chapter don’t contain the
structural rules and are complete with respect to classical semantics, as is the
original Gentzen system LK, which does contain the structural rules. As (via
Completeness Theorem) all three classical proof system RS, GL, LK are equiv-
alent we can say that the structural rules can be eliminated from the system
LK.
A natural question of elimination of structural rules from the intutionistic
Gentzen system LI arizes.
The following example illustrates the negative answer.
Example 3
We know, by the theorem about the connection between classical and intuition-
istic logic (theorem 7.6) and corresponding Completeness Theorems that for any
formula A ∈ F,
|= A if and only if `I ¬¬A,

where |= A means that A is a classical tautology, `I means that A is intutionis-

tically provable, i.e. is provable in any intuitionistically complete proof system.
The system LI is intuitionistically complete, so we have that for any formula A,

|= A if and only if `LI ¬¬A.

We have just proved that 6 `LI (¬¬A ⇒ A). Obviously |= (¬¬A ⇒ A), so we
know that ¬¬(¬¬A ⇒ A) must have a proof in LI.

338
We are going to prove that

`LI ¬¬(¬¬A ⇒ A)

and that the structural rule (contr −→) is essential to the existence of its proof,
i.e. that without it the formula ¬¬(¬¬A ⇒ A) is not provable in LI.
The following decomposition tree T is a proof of ¬¬(¬¬A ⇒ A) in LI.

−→ ¬¬(¬¬A ⇒ A)

f irst of 2 choices : (→ ¬), (→ weak)

| (−→ ¬)
¬(¬¬A ⇒ A) −→
f irst of 2 choices : (contr −→), (¬ −→)
| (contr −→)
¬(¬¬A ⇒ A), ¬(¬¬A ⇒ A) −→
one of 2 choices
| (¬ −→)
¬(¬¬A ⇒ A) −→ (¬¬A ⇒ A)
one of 3 choices
| (−→⇒)
¬(¬¬A ⇒ A), ¬¬A −→ A
one of 2 choices
| (−→ weak)
¬(¬¬A ⇒ A), ¬¬A −→
one of 3 choices
| (exch −→)
¬¬A, ¬(¬¬A ⇒ A) −→
one of 3 choices
| (¬ −→)
¬(¬¬A ⇒ A) −→ ¬A
one of 3 choices
| (−→ ¬)

339
 A, ¬(¬¬A ⇒ A) −→
one of 2 choices
| (exch −→)
¬(¬¬A ⇒ A), A −→
one of 3 choices
| (¬ −→)
A −→ (¬¬A ⇒ A)
one of 3 choices
| (−→⇒)
¬¬A, A −→ A
axiom

Assume now that the rule (contr −→) is not available. All possible decomposi-
tion trees are as follows.

T1
−→ ¬¬(¬¬A ⇒ A)
| (−→ ¬)
one of 2 choices
¬(¬¬A ⇒ A) −→
| (¬ −→)
only one choice
−→ (¬¬A ⇒ A)
| (−→⇒)
one of 2 choices
¬¬A −→ A
| (−→ weak)
only one choice
¬¬A −→
| (¬ −→)
only one choice
−→ ¬A
| (−→ ¬)
one of 2 choices
A −→
non − axiom

340
 T2

−→ ¬¬(¬¬A ⇒ A)

| (−→ weak)
second of 2 choices
−→
non − axiom

T3

−→ ¬¬(¬¬A ⇒ A)

| (−→ ¬)
¬(¬¬A ⇒ A) −→
| (¬ −→)
−→ (¬¬A ⇒ A)
| (−→ weak)
second of 2 choices
−→
non − axiom

T4

−→ ¬¬(¬¬A ⇒ A)

| (−→ ¬)
¬(¬¬A ⇒ A) −→
| (¬ −→)
−→ (¬¬A ⇒ A)
| (−→⇒)
]

¬¬A −→ A

| (−→ weak)
only one choice
¬¬A −→
| (¬ −→)

341
 only one choice
−→ ¬A
| (−→ weak)
second of 2 choices
−→
non − axiom

This proves that the formula ¬¬(¬¬A ⇒ A) is not provable in LI without

(contr −→) rule and hence this rule can’t be eliminated.

7.2.3 Proof Search Heuristic Method

Before we define a heuristic method of searching for proof in LI let’s make some
additional observations to the observations 1-5 from section 7.2.1.

Observation 6: Our goal while constructing the decomposition tree is to ob-

tain axiom or indecomposable leaves. With respect to this goal the use
logical decomposition rules has a priority over the use of the structural
rules and we use this information while describing the proof search heuris-
tic.

Observation 7: all logical decomposition rules (◦ →), where ◦ denotes any

connective, must have a formula we want to decompose as the first formula
at the decomposition node, i.e. if we want to decompose a formula ◦A,
the node must have a form ◦A, Γ −→ ∆. Sometimes it is necessary to
decompose a formula within the sequence Γ first in order to find a proof.

For example, consider two nodes

n1 = ¬¬A, (A ∩ B) −→ B

and
n2 = (A ∩ B), ¬¬A −→ B.
We are going to see that the results of decomposing n1 and n2 differ dramatically.
Let’s decompose the node n1 . Observe that the only way to be able to de-
compose the formula ¬¬A is to use the rule (→ weak) first. The two possible
decomposition trees that starts at the node n1 are as follows.

T1n1

¬¬A, (A ∩ B) −→ B

342
 | (→ weak)
¬¬A, (A ∩ B) −→
| (¬ →)
(A ∩ B) −→ ¬A
| (∩ →)
A, B −→ ¬A
| (→ ¬)
A, A, B −→
non − axiom

T2n1

¬¬A, (A ∩ B) −→ B

| (→ weak)
¬¬A, (A ∩ B) −→
| (¬ →)
(A ∩ B) −→ ¬A
| (→ ¬)
A, (A ∩ B) −→
| (∩ →)
A, A, B −→
non − axiom

Let’s now decompose the node n2 . Observe that following our Observation 6
we start by decomposing the formula (A ∩ B) by the use of the rule (∩ →) first.
A decomposition tree that starts at the node n2 is as follows.

T1n2

(A ∩ B), ¬¬A −→ B

| (∩ →)
A, B, ¬¬A −→ B
axiom

343
This proves that the node n2 is provable in LI, i.e.

`LI (A ∩ B), ¬¬A −→ B.

Of course, we have also that the node n1 is also provable in LI, as one can
obtain the node n2 from it by the use of the rule (exch →).

Observation 8: the use of structural rules are important and necessary while
we search for proofs. Nevertheless we have to use them on the ”must”
basis and set up some guidelines and priorities for their use.

For example, use of weakening rule discharges the weakening formula, and
hence an information that may be essential to the proof. We should use
it only when it is absolutely necessary for the next decomposition steps.
Hence, the use of weakening rule (→ weak) can, and should be restricted
to the cases when it leads to possibility of the use of the negation rule
(¬ →).
This was the case of the decomposition tree T1n1 . We used it as an
necessary step, but still it discharged too much information and we didn’t
get a proof, when proof of the node existed.
In this case the first rule in our search should have been the exchange
rule, followed by the conjunction rule (no information discharge) not the
weakening (discharge of information) followed by negation rule. The full
proof of the node n1 is the following.

T3n1
¬¬A, (A ∩ B) −→ B
| (exch −→)
(A ∩ B), ¬¬A −→ B
| (∩ →)
A, B, ¬¬A −→ B
axiom

As a result of the observations 1- 5 from section 7.2.1 and observations 6 -

8 above we adopt the following.

Heuristic Procedure for Proof Search in LI.

For any A ∈ F we construct the set of decomposition trees T→A following the
rules below.

344
 Rules for Decomposition Tree Generation
1. Use first logical rules where applicable.
2. Use (exch →) rule to decompose, via logical rules, as many formulas on the
left side of −→ as possible.
3. Use (→ weak) only on a ”must” basis in connection with (¬ →) rule.
4. Use (contr →) rule as the last recourse and only to formulas that contain ¬
or ⇒ as connectives.
5. Let’s call a formula A to which we apply (contr →) rule a contraction
formula.
6. The only contraction formulas are formulas containing ¬ between theirs log-
ical connectives.
7. Within the process of construction of all possible trees use (contr →) rule
only to contraction formulas.
8. Let C be a contraction formula appearing on the node n of the decomposi-
tion tree of T→A . For any contraction formula C, any node n, we apply
(contr →) rule the the formula C at most as many times as the number
of sub-formulas of C.

If we find a tree with all axiom leaves we have a proof, i.e. `LI A and if all
(finite number) trees have a non-axiom leaf we have proved that proof of A does
not exist, i.e. 6 `LI A.

7.3 Introduction to Modal S4 and S5 Logics

The non-classical logics can be divided in two groups: those that rival classical
logic and those which extend it. The Lukasiewicz, Kleene, and Intuitionistic
Logics are in the first group. The modal logics are in the second.
The rival logics do not differ from classical logic in terms of the language em-
ployed. Rather, rival logics differ in that certain theorems or tautologies of
classical logic are rendered false, or not provable in them.
Perhaps the most notorious example of this is the law of excluded middle (A ∪
¬A). This is provable in, and is a tautology of classical logic but is not provable
in, and is not tautology of intuitionistic logic, or is not a tautology under any
of the extensional logics semantics we have discussed.
Logics which extend classical logic sanction all the theorems of classical logic
but, generally, supplement it in two ways. Firstly, the languages of these non-
classical logics are extensions of those of classical logic, and secondly, the the-
orems of these non-classical logics supplement those of classical logic. Usually,

345
such supplementation is provided by the enriched language. For example, modal
logics are enriched by the addition of two new connectives that represent the
meaning of it is necessary that and it is possible that. We use the notation I for
it is necessary that and C for it is possible that. Other notations used are: ∇,
N, L for it is necessary that, and ♦ P, M for it is possible that. The symbols N,
L, P, M or alike, are often used in computer science investigations. The symbols
∇ and  were first to be used in modal logic literature, the symbols I, C come
from algebraic and topological interpretation of modal logics. I corresponds to
the interior of the set and C to its closure.
The idea of a modal logic was first formulated by an American philosopher,
C.I. Lewis in 1918. He has proposed yet another interpretation of lasting con-
sequences, of the logical implication. In an attempt to avoid, what some felt,
the paradoxes of semantics for classical implication which accepts as true that a
false sentence implies any sentence he created a notion of a modal truth, which
lead to the notion of modal logic. The idea was to distinguish two sorts of truth:
necessary truth and mere possible (contingent) truth. A possibly true sentence
is one which, though true, could be false. A necessary truth is hence the one
which could not be otherwise; a contingent (possible) truth is one which could.
The distinction between them is a metaphysical one and should not be confused
with the distinction between a priori and a posteriori truths. An a priori
truth is one which can be known independently of experience, and an a poste-
riori truth is one which cannot. Such notions appeal to epistemic considerations
and the whole area of modal logics bristles with philosophical difficulties and
hence the numbers of logics have been created. Unlike the classical connectives,
the modal connectives do not admit of truth-functional interpretation. This was
the reason for which modal logics was first developed as a proof systems, with
intuitive notion of semantics expressed by the set of adopted axioms.
The first semantics, and hence the proofs of the completeness theorems came
some 20 years later. It took yet another 25 years for discovery and development
of the second more general approach to the semantics. These are two established
ways of interpret modal connectives, i.e. to define modal semantics.
The historically first one is due to Mc Kinsey and Tarski (1944, 1946). It is a
topological interpretation that provides a powerful mathematical interpretation
of some of them, namely S4 and S5. It connects the modal notion of necessity
with the topological notion of interior of a set, and the notion of possibility with
the notion of its closure . Our choice of symbols I and C for modal connectives
comes from this interpretation. The topological interpretation powerful as it is,
is less universal in providing models for other modal logics. The most recent
one is due to Kripke (1964). It uses the notion possible world. Roughly, we say
that CA is true if A is true in some possible world, called actual world, and IA
is true if A is true in every possible world.
We present the formal definition later in this chapter, but this intuitive meaning
can be useful in unconvincing ourselves about validity (or sense) of adopted

346
axioms and rules of inference.
As we have already mentioned, modal logics were first developed, as was the
intuitionistic logic, in a form of proof systems only. First Hilbert style modal
proof system was published by Lewis and Langford in 1932. They presented
a formalization for two modal logics, which they called S1 and S2. They also
outlined three other proof systems, called S3, S4, and S5.
In 1933 Gödel worked with Heyting’s ”sentential logic” proof system, what we
are calling now Intuitionistic logic. He considered a particular modal proof sys-
tem and asserted that theorems of Heyting’s ”sentential logic” could be obtained
from it by using a certain translation. His presentation of the discovered proof
system, now known as S4 logic, was particularly elegant.
Since then hundreds of modal logics have been created. There are some standard
texts in the subject. These are, between the others: Hughes and Cresswell
(1969) for philosophical motivation for various modal logics and Intuitionistic
logic, Bowen (1979) for a detailed and uniform study of Kripke models for
modal logics, Segeberg (1971) for excellent classification, and Fitting (1983), for
extended and uniform studies of automated proof methods for classes of modal
logics.
Hilbert Style Modal Proof Systems
We present here Hilbert style formalization for S4 and S5 logics due to Mc
Kinsey and Tarski (1948) and Rasiowa and Sikorski (1964). We also discuss
the relationship between S4 and S5, and between the Intuitionistic logic and S4
modal logic, as first observed by Gödel.

They stress the connection between S4, S5 and topological spaces which consti-
tute models for them. Hence the use of symbols I, C for necessity and possibility,
respectively. The connective I corresponds to the symbol denoting a topological
interior of a set and C to the closure of a set.

Modal Language
We add to the propositional language L{∪,∩,⇒,¬} two extra one argument con-
nectives I and C. I.e. we adopt
L = L{∪,∩,⇒,¬,I,C} (7.39)
as our modal language. We read a formula IA, CA as necessary A and possible
A, respectively.
The language is common to all modal logics. Modal logics differ on a choice of
axioms and rules of inference, when studied as proof systems and on a choice of
semantics.
McKinsey, Tarski (1948)
As modal logics extend the classical logic, any modal logic contains two groups

347
of axioms: classical and modal.
Axioms Group 1: classical axioms
Any modal logic adopts as its classical axioms any complete set of axioms for a
classical propositional logic.
Axioms Group 2: modal axioms
M1 (IA ⇒ A),
M2 (I(A ⇒ B) ⇒ (IA ⇒ IB)),
M3 (IA ⇒ IIA),
M4 (CA ⇒ ICA).
Rules of inference
We adopt the Modus Ponens (M P )

A ; (A ⇒ B)
(M P )
B
and an additional modal rule (I) introduced by Gödel

A
(I)
IA
referred to as necessitation.
We define modal proof systems S4 and S5 as follows.

S4 = ( L, F, classical axioms, M 1 − M 3, (M P ), (I) ), (7.40)

S5 = ( L, F, classical axioms, M 1 − M 4, (M P ), (I) ). (7.41)

Observe that the axioms of S5 extend the axioms of S4 and both system share
the same inference rules, hence we have immediately the following.

Fact 7.3
For any formula A ∈ F, if `S4 A, then `S5 A.

Rasiowa, Sikorski (1964)

It is often the case, and it is in our systems, that modal connectives are express-
ible by each other, i.e. that we can adopt one of them and define the other as
follows.
IA = ¬C¬A, (7.42)
and
CA = ¬I¬A. (7.43)

348
The equality sign in equations (7.42), and (7.43) means that we replace the
formula in left side of the equation by the formula in the right side anywhere
where the left side (or right side) formula is appears as a sub formula of a
formula of L. In modal logics S4 and S5 the connective C is expressible by ¬
and I, as stated above by (7.43), we hence assume now that the language L
contains only one modal connective I.
Language

L = L{∩,∪,⇒,¬,I} . (7.44)

There are, as before, two groups of axioms: classical and modal.

Axioms Group 1: classical axioms
We adopt as classical axioms any complete set of axioms for a classical propo-
sitional logic.
Axioms Group2: modal axioms
R1 ((IA ∩ IB) ⇒ I(A ∩ B)),
R2 (IA ⇒ A),
R3 (IA ⇒ IIA),
R4 I(A ∪ ¬A),
R5 (¬I¬A ⇒ I¬I¬A)
Rules of inference
We adopt the Modus Ponens (M P )
A ; (A ⇒ B)
(M P )
B
and an additional modal rule ( RI)
(A ⇒ B)
(RI) .
(IA ⇒ IB)
We define modal logic proof systems RS4, RS5 as follows.
RS4 = ( L, F, classical axioms, R1 − R4, (M P ), (RI) ). (7.45)
RS5 = ( L, F, classical axioms, R1 − R5, (M P ), (RI) ). (7.46)
Observe that the axioms of RS5 extend, as the axioms of TS5, the axioms of
TS4 and both system share the same inference rules, hence we have immediately
the following.

Fact 7.4
For any formula A ∈ F, if `RS4 A, then `RS5 A.

349
7.3.1 Algebraic Semantics for S4 and S5

The McKinsey, Tarski proof systems (7.40), (7.41), and Rasiowa, Sikorski proof
systems (7.45), (7.45) for modal logics S4, S5 are complete with the respect to
both algebraic topological semantics, and Kripke semantics.
We shortly discuss the topological semantics, and algebraic completeness the-
orems and leave the Kripke semantics for the reader to explore from other,
multiple sources.
The topological semantics was initiated by McKinsey, Tarski’s (1946, 1948) and
consequently developed and examined by many authors into a field of Alge-
braic Logic.They are presented in detail in now classic algebraic logic books:
”Mathematics of Metamathematics”, Rasiowa, Sikorski (1964) and ”An Alge-
braic Approach to Non-Classical Logics”, Rasiowa (1974).
We want to point out that the first idea of a connection between modal propo-
sitional calculus and topology is due to Tang Tsao -Chen, (1938) and Dugunji
(1940).
Here are some basic definitions.
Boolean Algebra
An abstract algebra
B = (B, 1, 0, ⇒, ∩, ∪, ¬) (7.47)
is said to be a Boolean algebra if it is a distributive lattice and every element
a ∈ B has a complement ¬a ∈ B.
Topological Boolean algebra
By a topological Boolean algebra we mean an abstract algebra

B = (B, 1, 0, ⇒, ∩, ∪, ¬, I), (7.48)

where (B, 1, 0, ⇒, ∩, ∪, ¬) is a Boolean algebra (7.47) and, moreover, the

following conditions hold for any a, b ∈ B.

I(a ∩ b) = Ia ∩ Ib, Ia ∩ a = Ia, IIa = Ia, and I1 = 1. (7.49)

The element Ia is called a interior of a. The element ¬I¬a is called a closure
of a and will be denoted by Ca. Thus the operations I and C are such that

Ca = ¬I¬a and Ia = ¬C¬a.

In this case we write the topological Boolean algebra (7.48) as

B = (B, 1, 0, ⇒, ∩, ∪, ¬, I, C). (7.50)

It is easy to prove that in in any topological Boolean algebra (7.50) the following
conditions hold for any a, b ∈ B.

350
 C(a ∪ b) = Ca ∪ Cb, Ca ∪ a = Ca, CCa = Ca, and C0 = 0. (7.51)

If X is a topological space with an interior operation I, then the family P(X)

of all subsets of X is a topological Boolean algebra with 1 = X, the operation
⇒ defined by the formula

Y ⇒ Z = (X − Y ) ∪ Z, for all subsets Y, Zof X,

set-theoretical operations of union, intersection, complementation, and the in-

terior operation I. Obviously, every sub algebra of this algebra is a topological
Boolean algebra, called a topological field of sets or, more precisely, a topological
field of subsets of X.
Given a topological Boolean algebra (7.47) (B, 1, 0, ⇒, ∩, ∪, ¬). The element
a ∈ B is said to be open (closed) if a = Ia (a = Ca).
Clopen Topological Boolean algebra
A topological Boolean algebra (7.50)

B = (B, 1, 0, ⇒, ∩, ∪, ¬, I, C ).

such that every open element is closed and every closed element is open, i.e.
such that for any a ∈ B

CIa = Ia and ICa = Ca (7.52)

is called a clopen topological Boolean algebra.

We loosely say that a formula A of a modal language is a modal S4 tautology
if and only if any topological Boolean algebra (7.50) is a model for A.
We say that A is a modal S5 tautology if and only if any clopen topological
Boolean algebra (7.52) is a model for A.
We put it formally as follows.

Definition 7.4 (Modal Algebraic Model)

For any formula A of a modal language L{∪,∩,⇒,¬,I,C} and for any topological
Boolean algebra B = (B, 1, 0, ⇒, ∩, ∪, ¬, I, C),
the algebra B is a model for the formula A and denote it by

B |= A

if and only if v ∗ (A) = 1 holds for all variables assignments v : V AR −→ B.

Definition 7.5 (S4, S5 Tautology)

351
The formula A is a modal S4 tautology (S5 tautology) and is denoted by

|=S4 A (|=S5 A)

if and only if for all topological Boolean algebras (clopen topological

Boolean algebras) B we have that

B |= A (B |= A).

In Algebraic Logic the notion of tautology is often defined using a notion ”a

formula A is valid in an algebra B ”.It is formally defined in our case as as
follows.

Definition 7.6
A formula A is valid in a topological Boolean algebra B = (B, 1, 0, ⇒
, ∩, ∪, ¬, I, C), if and only if v ∗ (A) = 1 holds for all variables assignments
v : V AR −→ B.

Directly form definitions 7.5, 7.6 we get the following.

Fact 7.5 For any formula A, |=S4 A (|=S5 A) if and only if A is valid in
all topological Boolean algebras (A is valid in all clopen topological Boolean
algebras).

We write `S4 A and `S5 A do denote any proof system for modal S4, S5
logics and in particular the proof systems (7.40), (7.45), and (7.41), (7.46),
respectively.

Theorem 7.17 (Completeness Theorem) For any formula A of the modal

language L{∪,∩,⇒,¬,I,C} ,

`S4 A if and only if |=S4 A,

`S5 A if and only if |=S5 A.

The completeness for S4 follows directly from the Theorem 7.18. The com-
pleteness for S5 follows from the S4 completeness and Embedding Theorems
7.22, 7.23. It also can be easily proved independently by adopting the Algebraic
Completeness Theorem proof for S4 to clopen topological algebras.

Theorem 7.18 (Algebraic Completeness Theorem)

For any formula A of the modal language L= L{∪,∩,⇒,¬,I,C} the following con-
ditions are equivalent.

352
(i) `S4 A,
(ii) |=S4 A,
(iii) A is valid in every topological field of sets B(X),
r
(iv) A is valid in every topological Boolean algebra B with at most 22 elements,
where r is the number of all sub formulas of A,
(iv) v ∗ (A) = X for every variable assignment v in the topological field of sets
B(X) of all subsets of a dense-in -itself metric space X 6= ∅ (in particular of an
n-dimensional Euclidean space X).

On S4 derivable disjunction
In a classical logic it is possible for the disjunction (A ∪ B) to be a tautology
when neither A nor B is a tautology. The tautology (A ∪ ¬A) is the simplest
example. This does not hold for the intuitionistic logic. We have a similar
theorem for modal S4 logic, as proved by McKinsey and Tarski.

Theorem 7.19
A disjunction (IA∪IB) is S4 provable if and only if either A or B is S4 provable,
i.e.
`S4 (IA ∪ IB) if and only if `S4 A or `S4 B.

The proof follows directly from the Completeness Theorem 7.18 and the follow-
ing semantical, proof system independent version of the theorem 7.19.

Theorem 7.20 (McKinsey, Tarski, 1948)

For any A ∈ F,

|=S4 (IA ∪ IB) if and only if |=S4 A or |=S4 B.

The completeness theorem allows us to formulate theorems about logics in terms

independent of the proof system considered. In this sense the notion of tautology
is more general then the notion of provability. This is why often we use rather the
tautology formulation of the known facts about the logic and their relationships
instead of the notion of provability.
Following the Completeness Theorem 7.18 we get a semantical version of the
theorem 7.3.

Theorem 7.21
For any formula A ∈ F,

if |=S4 A, then |=S5 A.

353
Consider a modal language L with both modal connectives, i.e.

L= L{∪,∩,⇒,¬,I,C} .

The above theorem7.21 says that the S4 tautologies form a subset of S5 tau-
tologies. We have even a more powerful relationship, namely the following.

Theorem 7.22 (Embedding 1)

For any formula A ∈ F,

|=S4 A if and only if |=S5 ICA,

`S4 A if and only if `S5 ICA.

Theorem 7.23 (Embedding 2)

For any formula A ∈ F,

|=S5 A if and only if |=S4 ICIA.

`S5 A if and only if `S4 ICIA.

Theorem 7.24 (Embedding 3)

For any formula A ∈ F,

if |=S5 A, then |=S4 ¬I¬A.

The fist proof of the above embedding theorems was given by Matsumoto in
1955. Provability. Fitting semantical 1983 Ohnishi and Matsumoto 1957/59
Gentzen Methods in Modal Calculi Osaka Mathematical Journal 9.113 -130

7.3.2 S4 and Intuitionistic Logic, S5 and Classical Logic

As we have said in the introduction, Gödel was the first to consider the connec-
tion between the intuitionistic logic and a logic which was named later S4. His
proof was purely syntactic in its nature, as semantics for neither intuitionistic
logic nor modal logic S4 had not been invented yet.
The algebraic proof of this fact, was first published by McKinsey and Tarski in
1948. We now define the mapping establishing the connection (definition 7.7)
and refer the reader to Rasiowa and Sikorski book ”Mathematics of Metamath-
ematics” for its proof.
Let L be a propositional language of modal logic, as defined by (7.44), i.e the
language
L = L{∩,∪,⇒,¬,I} .

354
Let L0 be a language obtained from L by elimination of the connective I and by
the replacement the negation connective ¬ by the intuitionistic negation, which
we will denote here by a symbol ∼. Such obtained language

L0 = L{∩,∪,⇒,∼} (7.53)

is a propositional language of the intuitionistic logic.

In order to establish the connection between the languages (7.44) and (7.53),
and hence between modal and intuitionistic logic, we define a mapping f which
to every formula A ∈ F0 of L0 assigns a formula f (A) ∈ F of L.
We define a mapping f as follows.

Definition 7.7 (Gódel - Tarski)

A function f : F0 → F be such that

f (a) = Ia for any a ∈ V AR,

f ((A ⇒ B)) = I(f (A) ⇒ f (B)),

f ((A ∪ B)) = (f (A) ∪ f (B)),
f ((A ∩ B)) = (f (A) ∩ f (B)),
f (∼ A) = I¬f (A),
where A, B denote any formulas in L0 is called a Gödel - Tarski mapping.

Example
Let A be a formula
((∼ A ∩ ∼ B) ⇒∼ (A ∪ B))
and f be the mapping of definition 7.7. We evaluate f (A) as follows

f ((∼ A ∩ ∼ B) ⇒∼ (A ∪ B)) =

I(f (∼ A ∩ ∼ B) ⇒ f (∼ (A ∪ B)) =
I((f (∼ A) ∩ f (∼ B)) ⇒ f (∼ (A ∪ B)) =
I((I¬f A ∩ I¬f B) ⇒ I¬f (A ∪ B)) =
I((I¬A ∩ I¬B) ⇒ I¬(f A ∪ f B)) =
I((I¬A ∩ I¬B) ⇒ I¬(A ∪ B)).

We use notation `I A do denote the fact that A is intuitionistically provable,

i.e. provable in any intuitionistic proof system I.
With these hypotheses and notation the following theorem holds.

355
Theorem 7.25
Let f be the Gödel mapping (definition 7.7). For any formula A of L0 ,

`I A if and only if `S4 f (A),

where I, S4 denote any proof systems for intuitionistic and and S4 logic, re-
spectively.

In order to establish the connection between the modal logic and classical logic
we consider the Gódel - Tarski mapping (definition ??) between the modal
language L{∩,∪,⇒,¬,I} and its classical sub-language L{¬,∩,∪,⇒} .
Now with every classical formula A we associate a modal formula f (A) defined
by induction on the length of A as follows:

f (a) = Ia, f ((A ⇒ B)) = I(f (A) ⇒ f (B)),

f ((A ∪ B)) = (f (A) ∪ f (B)), f ((A ∩ B)) = (f (A) ∩ f (B)), f (¬A) = I¬f (A).

We use notation `H A do denote the fact that A is classically provable, i.e.

provable in any proof system for classical propositional logic.
The following theorem established relationship between classical logic and modal
S5.

Theorem 7.26
Let f be the Gödel mapping (definition 7.7) between L{¬,∩,∪,⇒} and L{∩,∪,⇒,¬,I} .
For any formula A of L{¬,∩,∪,⇒} ,

`H A if and only if `S5 f (A),

where H, S5 denote any proof systems for classical and and S5 modal logic,
respectively.

7.4 Homework Problems

1. The algebraic models for the intuitionistic logic are defined in terms of
Pseudo-Boolean Algebras in the following way. A formula A is said to be
an intuitionistic tautology if and only if v |= A, for all v and all Pseudo-
Boolean Algebras, where v maps V AR into universe of a Pseudo-Boolean
Algebra. I.e. A is an intuitionistic tautology if and only if it is true in all
Pseudo-Boolean Algebras under all possible variable assignments.
A 3 element Heyting algebra as defined in chapter 3 is a 3 element Pseudo-
Boolean Algebra.

356
 (i) Show that the 3 element Heyting algebra is a model for all formulas
(7.12) - (7.27).
(ii) Determine for which of the formulas (7.28) - (7.34) the 3 element
Heyting algebra acts as a counter-model.

2. Find proofs by constructing proper decomposition trees in Gentzen System

LI of axioms A1 - A11 of the proof system I defined by (7.4).

3. The completeness with respect to algebraic semantics of system LI follows

from the Hauptzatz Theorem and the completeness theorem 7.2. The
proof is a straightforward adaptation of the proof of the completeness
theorem for LK included in chapter 6. Write carefully all steps of the
proof of completeness theorem for LI .

4. Find proofs by constructing proper decomposition trees in Gentzen System

LI of the intuitionistic tautologies (7.12) - (7.27).

5. Show that none of the formulas (7.28) - (7.34) is provable in LI.

6. Find proofs by constructing proper decomposition trees in Gentzen System

LI of double negation of all of the formulas (7.28) - (7.34).

7. Give the proof of the Glivenko theorem 7.7, i.e. prove that any formula A
is a classically provable if and only if ¬¬A is an intuitionistically provable.

8. Give few examples of formulas illustrating that the following theorems

hold.
Gödel (1) For any A, B ∈ F, a formula (A ⇒ ¬B) is a classically
provable if and only if it is intuitionistically provable.

9. Give examples of formulas illustrating that the following theorems hold.

Gödel (2) If a formula A contains no connectives except ∩ and ¬, then
A is a classically provable if and only if it is an intuitionistically provable.

10. Use the Completeness Theorem 7.18 to show that the following proof
system CS4 is a complete proof system for the modal logic S4.
We adopt the modal language L{∪,∩,⇒,¬,I,C} . We adopt, as before, two
groups of axioms: classical and modal.
Group 1: we take any complete set of axioms for a classical propositional
logic. Group 2: the following modal axioms.
C1 (C(A ∪ B) ⇒ (CA ∪ CB)),
C2 (A ⇒ CA),
C3 (CCA ⇒ CA),
C4 C(A ∩ ¬A).

357
 Rules of inference: we adopt the Modus Ponens (M P ) and an additional
rule,
(A ⇒ B)
(C) .
(C¬B ⇒ C¬A)
We define the proof system CS4 as follows

CS4 = ( L, F, classical axioms, C1 − C4, (M P ), (C) ).

11. Evaluate f (A), where f is the Gödel- Tarski mapping (definition 7.7), for
all the formulas A listed below.
(i) (¬A ∪ ¬B) ⇒ (¬A ∩ ¬B))
(ii) ((A ⇒ B) ⇒ (¬B ⇒ ¬A))
(iii) ((A ⇒ ¬B) ⇒ (B ⇒ ¬A))
(iv) (¬¬¬A ⇒ ¬A)
(v) (¬A ⇒ ¬¬¬A)
(vi) (¬¬(A ⇒ B) ⇒ (A ⇒ ¬¬B))
(vii) ((C ⇒ A) ⇒ ((C ⇒ (A ⇒ B)) ⇒ (C ⇒ B))

12. Use the Completeness Theorem 7.18 and Embedding Theorems 7.22,
7.23 to show the following.
(i) For any formula A, `RS4 A, if and only if vdashRS5 I¬I¬A,
where RS4, RS5 are proof system (7.45) and (7.46).
(ii) For any formula A, `S5 A, if and only if |=S4 I¬I¬IA,
where S4, S5 are proof system (7.40) and (7.41).

358
Chapter 8

Classical Predicate
Semantics and Proof
Systems

8.1 Formal Predicate Languages

Propositional languages are also called zero order languages, as opposed to pred-
icate languages that are called first order languages. The same applies to the
use of terms propositional and predicate logic; they are often called zero order
and first order logics and we will use both terms equally.
We define a predicate language L following the pattern established by the propo-
sitional languages definitions. The predicate language L is more complicated in
its structure and hence its alphabet A is much richer. The definition of its set
of formulas F is more complicated. In order to define the set F we introduce
an additional set T, called a set of terms of the predicate language L. We
single out this set not only because we need it for the definition of formulas,
but also because of its role in the development of other notions of predicate logic.

We will work with different predicate languages, depending on what applications

we have in mind. All of these languages have some common features, and we
begin with a following general definition.

Definition 8.1
By a predicate language L we understand a triple

L = (A, T, F), (8.1)

359
where A is a predicate alphabet, T, is the set of terms, and F is a set of
formulas.

The components of L are as follows.

1. Alphabet A is the set

A = V AR ∪ CON ∪ P AR ∪ Q ∪ P ∪ F ∪ C, (8.2)

where V AR is set of predicate variables, CON is a set of propositional connec-

tives, P AR a set of parenthesis, Q a set of quantifiers, P, a set of predicate
symbols, F a set of functions symbols, and C a set of constant symbols. We
assume that all of the sets defining the alphabet are disjoint.

Predicate Variables V AR
We assume that we always have a countably infinite set V AR of predicate
variables, called usually variables. We denote variables by x, y, z, ..., with
indices, if necessary, what we often express by writing

V AR = {x1 , x2 , ....}.

Propositional connectives CON

We define the set of propositional connectives CON in the same way as in the
case of the propositional languages. It means that we assume that CON is
non-empty and finite set and that consider only the connectives with one or
two arguments, i.e.
CON = C1 ∪ C2
where C1 is a finite set (possibly empty) of unary connectives, C2 is a finite set
(possibly empty) of binary connectives of the language L.

Parenthesis P AR
As in the propositional case, we adopt the signs ( and ) for our parenthesis., i.e.
we define the set P AR as
P AR = {(, )}.

The set of propositional connectives CON defines a propositional part of the

predicate logic language. What really differ one predicate language from the
other is the choice of additional symbols to the symbols described above. These
are called quantifiers symbols, predicate symbols, function symbols, and con-
stant symbols. I.e. a particular predicate language is determined by specifying
the following sets of symbols.

360
Quantifiers Q
We adopt two quantifiers; ∀ (for all, the universal quantifier) and ∃ (there exists,
the existential quantifier), i.e. we have the following set of quantifiers
Q = {∀, ∃}.
In a case of the classical logic and the logics that extend it, it is possible to adopt
only one quantifier and to define the other in terms of it and propositional con-
nectives. It is impossible in a case of many non-classical logics, for example the
intuitionistic logic. But even in the case of classical logic two quantifiers express
better the common intuition, so we assume that we have two of them.

Predicate symbols P
Predicate symbols represent relations. We assume that we have an non empty,
finite or countably infinite set bf P of predicate, or relation symbols. We denote
predicate symbols by P, Q, R, ..., with indices, if necessary, what we often express
by writing
P = {P1 , P2 , ...}.

Each predicate symbol P ∈ P has a positive integer #P assigned to it; if

#P = n then say P is called an n-ary (n - place) predicate (relation) symbol.

Function symbols F
We assume that we have a finite (may be empty) or countably infinite set F of
function symbols. When the set F is empty we say that we deal with a language
without functional symbols. We denote functional symbols by f, g, h, ..., with
indices, if necessary, what we often express by writing
F = {f1 , f2 , ...}.

Similarly, as in the case of predicate symbols, each function symbol f ∈ F has

a positive integer #f assigned to it; if #f = n then say f is called an n-ary (n
- place) function symbol.

Constant symbols C
We also assume that we have a finite (may be empty) or countably infinite set
C of constant symbols. The elements of C are denoted by c, d, e..., with indices,
if necessary, what we often express by writing
C = {c1 , c2 , ...}.
When the set C is empty we say that we deal with a language without constant
symbols.

361
Sometimes the constant symbols are defined as 0-ary function symbols, i.e.
C ⊆ F. We single them out as a separate set for our convenience.

Observe that what distinguishes now one predicate language L form the pother
is the choice of the components CON , and P, F, C of its alphabet A. We
hence will write

LCON (P, F, C) (8.3)

to denote the predicate language L determined by P, F, C and the set of propo-
sitional connectives CON .

Once the set of propositional connectives is fixed, the predicate language is

determined by the sets P, F and C and we l write

L(P, F, C) (8.4)

for the predicate language L determined by P, F and C (with a fixed set of

propositional connectives). If there is no danger of confusion, we may abbreviate
L(P, F, C) to just L.
We sometimes allow the same symbol to be used as an n-place relation symbol,
and also as an m-place one; no confusion should arise because the different uses
can be told apart easily. Similarly for function symbols.
Having defined the basic elements of syntax, the alphabet, we can now complete
the formal definition of the predicate language by defining two more complex
sets: the set T of all terms and the set F of all well formed formulas of the
language L(P, F, C).

2. Terms T
The set T of terms of a predicate language L(P, F, C) is defined as follows.

Definition 8.2 (Terms)

Given a predicate language L(P, F, C) with an alphabet A. The set T of terms
of L is the smallest set T ⊆ A∗ meeting the conditions:
(i) any variable is a term, i.e. V AR ⊆ T;
(ii) any constant symbol is a term, i.e. C ⊆ T;
(iii) if f is an nplace function symbol, i.e. f ∈ F and #f = n and t1 , t2 , ..., tn ∈
T, then f (t1 , t2 , ..., tn ) ∈ T.

Example 8.1

362
Let f ∈ F, #f = 1, i.e. f is a one place function symbol. Let x, y be predicate
variables, c, d constants, i.e. x, y ∈ V AR, c, d ∈ C. The following expressions
are terms:

x, y, f (x), f (y), f (c), f (d), f (f (x), ) f (f (y)), f f ((c)), f (f (d)), ...etc.

Example 8.2
If F = ∅, C = ∅, then the set T of terms consists of variables only, i.e.

T = V AR = {x1 , x2 , ....}.

From the above we get the following observation.

Remark 8.1
For any predicate language L(P, F, C), the set T of its terms is always non-
empty.

Example 8.3
If f ∈ F, #f = 1, g ∈ F, #g = 2, x, y ∈ V AR, c, d ∈ C, then some of the
terms are the following:

f (g(x, y)), f (g(c, x)), g(f (f (c)), g(x, y)), g(c, g(x, f (c))).

From time to time, the logicians are and we may be informal about how we
write terms. For instance, if we denote a two place function symbol g by +, we
may write x + y instead +(x, y). Because in this case we can think of x + y as
an unofficial way of designating the ”real” term +(x, y), or even g(x, y).

2. Formulas F
Before we define the set of formulas, we need to define one more set; the set of
atomic, or elementary formulas. They are the ”smallest” formulas as were the
propositional variables in the case of propositional languages.

Atomic formulas
An atomic formula of a predicate language L(P, F, C) is any element of the
alphabet A∗ of the form
R(t1 , t2 , ..., tn ),
where R ∈ P, #R = n, i.e. R is n-ary relational symbol and t1 , t2 , ..., tn are
terms. The set of all atomic formulas is denoted by AF and is defines as

AF = {R(t1 , t2 , ..., tn ) ∈ A∗ : R ∈ P, t1 , t2 , ..., tn ∈ T, #R = n, n ≥ 1}. (8.5)

363
Example 8.4
Consider a language
L(∅, {P }, ∅),
for #P = 1, i.e. a language without neither functional, nor constant symbols,
and with one, one-place predicate symbol P . The set of atomic formulas contains
all formulas of the form P (x), for x any variable, i.e.
AF = {P (x) : x ∈ V AR}.

Example 8.5
Let now
L = L({f, g}, {R}, {c, d}),
for #f = 1, #g = 2 , #R = 2, i.e. L has two functional symbols: one -place
symbol f and two-place symbol g; one two-place predicate symbol R, and two
constants: c,d. Some of the atomic formulas in this case are the following.
R(c, d), R(x, f (c)), R(f (g(x, y)), f (g(c, x))), R(y, g(c, g(x, f (c)))).

Given a predicate language

L = LCON (P, F, C),
where CON is non-empty, finite set of propositional connectives such that
CON = C1 ∪ C2 for C1 a finite set (possibly empty) of unary connectives,
C2 a finite set (possibly empty) of binary connectives of the language L. We
define the set F of all well formed formulas of the predicate language L as
follows.

Definition 8.3 (Formulas)

The set F of all well formed formulas, called shortly set of formulas, of the
language LCON P, F, C) is the smallest set meeting the following conditions:

1. any atomic formula of LCON (P, F, C) is a formula, i.e.

AF ⊆ F;

2. if A is a formula of LCON (P, F, C), 5 is an one argument propositional

connective, then 5A is a formula of LCON (P, F, C), i.e. if the following
recursive condition holds
if A ∈ F, 5 ∈ C1 , then 5 A ∈ F;

3. if A, B are formulas of L(P, F, C), ◦ is a two argument propositional

connective, then (A ◦ B) is a formula of L(P, F, C), i.e. if the following
recursive condition holds
if A ∈ F, 5 ∈ C2 , then (A ◦ B) ∈ F;

364
 4. if A is a formula of L(P, F, C) and x is a variable, then ∀xA, ∃xA are
formulas of L(P, F, C), i.e. if the following recursive condition holds

if A ∈ F, x ∈ V AR, ∀, ∃ ∈ Q, then ∀xA, ∃xA ∈ F.

In formulas ∀xA, ∃xA, the formula A is in the scope of the quantifier ∀, ∃,

respectively.

Example 8.6
Let L be a language with with the set {∩, ∪, ⇒, ¬} of connectives and with
two functional symbols: one -place and one two-place, one two-place predicate
symbol, and two constants. We write L as

L = L{¬,∩,∪,⇒} ({R}, {f, g}, {c, d}, )

where #f = 1, #g = 2 , #R = 2. Some of the formulas of L are the following.

R(c, f (d)), ∃xR(x, f (c)), ¬R(x, y), ∀z(∃xR(x, f (c)) ⇒ ¬R(x, y)),

(R(c, d) ∩ ∃xR(x, f (c))), ∀yR(y, g(c, g(x, f (c)))), ∀y¬∃xR(x, y).

The formula R(x, f (c)) is in a scope of the quantifier ∃x in ∃xR(x, f (c)).

The formula (∃xR(x, f (c)) ⇒ ¬R(x, y)) isn’t in a scope of any quantifier.
The formula (∃xR(x, f (c)) ⇒ ¬R(x, y)) is in the scope of ∀ in ∀z(∃xR(x, f (c)) ⇒
¬R(x, y)).

Example 8.7
Let L be a language with with the set {¬, , ♦, ∩, ∪, ⇒} of connectives and P,
F, and C the same as in previous exercise, i.e.

L = L{¬,,♦,∩,∪,⇒} ({R}, {f, g}, {c, d}, )

where #f = 1, #g = 2 , #R = 2.
L is now a language of some first order modal logic. Some of the formulas of
L are the following.

♦¬R(c, f (d)), ♦∃xR(x, f (c)), ¬♦R(x, y), ∀z(∃xR(x, f (c)) ⇒ ¬R(x, y)),

(R(c, d) ∩ ∃xR(x, f (c))), ∀yR(y, g(c, g(x, f (c)))), ∀y¬♦∃xR(x, y).

The formula R(x, f (c)) is in a scope of the quantifier ∃x in ♦∃xR(x, f (c)).

The formula (∃xR(x, f (c)) ⇒ ¬R(x, y)) isn’t in a scope of any quantifier.

365
The formula (∃xR(x, f (c)) ⇒ ¬R(x, y)) is in the scope of ∀z in ∀z(∃xR(x, f (c)) ⇒
¬R(x, y)). Formula ¬♦∃xR(x, y) is in the scope of ∀y in ∀y¬♦∃xR(x, y).

Given a predicate language L = (A, T, F), we must distinguish between formulas

like P (x, y), ∀xP (x, y) and ∀x∃yP (x, y).
This is done by introducing the notion of free and bound variables, open
and closed formulas (sentences). Before we formulate proper definitions, here
are some simple observations.
1. Some formulas are without quantifiers.
For example formulas R(c1 , c2 ), R(x, y), (R(y, d) ⇒ R(a, z)). A formula with-
out quantifiers is called an open formula.
Variables x, y in R(x, y) are called free variables. The variables y in R(y, d)
and z in R(a,z) are also free.
2. Quantifiers bind variables within formulas.
The variable x is bounded by ∃x in the formula ∃xR(x, y), the variable y is
free. The variable y is bounded by ∀y in the formula ∀yR(x, y), the variable
y is free.
3. The formula ∃x∀yR(x, y) does not contain any free variables, neither does
the formula R(c1 , c2 ). A formula without any free variables is called a closed
formula or a sentence.
Sometimes in order to distinguish more easily which variable is free and which
is bound in the formula we might use the bold face type for the quantifier
bound variables and write the formulas as follows.
(∀xQ(x, y), ∃yP (y), ∀yR(y, g(c, g(x, f (c)))),
(∀xP (x) ⇒ ∃yQ(x, y)), (∀x(P (x) ⇒ ∃yQ(x, y)))

Observe that the formulas ∃yP (y), (∀x(P (x) ⇒ ∃yQ(x, y))) are closed. We
call a close formula a sentence.

Example 8.8
Consider atomic formulas: P (y), Q(x, c), R(z), P1 (g(x, y), z). Here are some
non atomic formulas formed out of them.
1. (P (y) ∪ ¬Q(x, c)) ∈ F. This is an open formula A with two free variables
x,y. We denote A this as formula A(x, y).
2. ∃x(P (y) ∪ ¬Q(x, c)) ∈ F. We write x to denote that x is a bound variable.
The variable y is free. This is a formula B with one free variable y. We denote
B as a formula B(y).
3. ∀y(P (y) ∪ ¬Q(x, c)) ∈ F. The variable y is bound, the variable x is free.
We denote this formula by for example A1 (x).

366
4. ∀y∃x(P (y) ∪ ¬Q(x, c)) ∈ F has no free variables. It is a closed formula
called also a sentence.

Exercise 8.1
Given the following formulas of L:
P (x, f (c, y)), ∃cP (x, f (c, y)), ∀xf (x, P (c, y)), ∃xP (x, f (c, y)) ⇒ ∀yP (x, f (c, y)).
1. Indicate whether they are, or are not well formed formulas of F. For those
which are not in F write a correct formula.
2. For each correct, or corrected formula identify all components: connectives,
quantifiers, predicate and function symbols, and list all its terms.
3. For each formula identify its s free and bound variables. State which are open
and which are closed formulas (sentences), if any.

Solution
Formula A1 = P (x, f (c, y)).
It is a correct atomic formula. P is a 2 argument predicate symbol, f is a
2 argument function symbol, c is a constant. We write it symbolically: P ∈
P, f ∈ F, c ∈ C. It is an open formula with two free variables x,y. We denote
it by A1 (x, y). It has no bound variables.
Formula A2 = ∃cP (x, f (c, y)).
It is a not a correct formula, i.e. ∃cP (x, f (c, y)) 6∈ F. The expression ∃c has no
meaning because c is a constant, not a variable.
The corrected formulas are: B1 = ∃xP (x, f (c, y)), B2 = ∃yP (x, f (c, y)), and
formulas B = ∃zP (z, f (c, y)) for any variable z different then x and y.
None of the correct formulas are open. Variable y is free in B1 = B1 (y), variable
x is free in B2 = B2 (x), both variables x and y are free in all formulas B =
B(x, y). All formulas are nether close, nor open. The terms appearing in any
of them are the same as in A1 = P (x, f (c, y)) and are: x, y, c, f (c, y).
Formula A3 = ∀xf (x, P (c, y)).
It is a not a correct formula, i.e. ∀xf (x, P (c, y)) 6∈ F. The function symbol f in
front f (x, P (c, y)) indicate a term and terms are not formulas. Moreover, the
atomic formula P (c, y) can’t be put inside a term!
Formula A4 = ∃xP (x, f (c, y)) ⇒ ∀yP (x, f (c, y)).
It is a not a correct formula. The correct formula is A = (∃xP (x, f (c, y)) ⇒
∀yP (x, f (c, y))). It has two free variables x and y and we write it as A = A(x, y).
Informally, in the formula P (x, y) both variables x and y are called free vari-
ables. They are not in the scope of any quantifier. The formula of that type
(without quantifiers) is an open formula.
The formal definition of the set of free variables of a formula is the following.

367
Definition 8.4 (Free and Bound Variables)
The set F V (A) of free variables of a formula A is defined by the induction of
the degree of the formula as follows.

1. If A is an atomic formula, i.e. A ∈ AF, then F V (A) is just the set of

variables appearing in the expression A;

2. for any unary propositional connective, i.e any 5 ∈ C1 ,

F V (5A)= F V (A),
i.e. the free variables of 5A are the free variables of A;

3. for any binary propositional connective, i.e any ◦ ∈ C2 ,

F V (A ◦ B)= F V (A) ∪ F V (B),
i.e. the free variables of (A ◦ B) are the free variables of A together with
the free variables of B;

4. F V (∀xA) = F V (∃xA) = F V (A) − {x},

i.e. the free variables of ∀xA and ∃xA are the free variables of A, except
for x.

A formula with no free variables is called a sentence.

A variable is called bound if it is not free.
A formula with no bound variables is called an open formula.

Example 8.9 The formulas ∃xQ(c, g(x, d)), ¬∀x(P (x) ⇒ ∃y(R(f (x), y) ∩
¬P (c))) are sentences. The formulas Q(c, g(x, d)), ¬(P (x) ⇒ (R(f (x), y) ∩
¬P (c))) are open formulas. The formulas ∃xQ(c, g(x, y)), ¬(P (x) ⇒ ∃y(R(f (x), y)∩
¬P (c))) are neither sentences nor open formulas. They contain some free and
some bound variables; the variable y is free in the first formula, the variable x
is free in the second.

The definition 8.1 defines a predicate language L = LCON (P, F, C) (8.3) with
its sets of predicate, function and constant symbol possibly countably infinite
sets. We use its most general case with sets of predicate, function and constant
symbol all countably infinite sets for defining all relevant notions concerning
provability and semantics. In particular, we will define in detail the classical
semantics for this most general form of L and prove the completeness theorem
for classical predicate logic based on it.
When we deal with formal theory Th(SA) with a set SA of specific axioms
we restrict the language LCON (P, F, C) to the symbols characteristic for that
theory. We hence introduce the following definition.

368
Definition 8.5
Given a language LCON (P, F, C) = (A, T, F).
Let F0 ⊆ F be a non-empty, finite subset of formulas of L. Denote by P0 , F0 , C0
the sets of all predicate, function, and constant symbols appearing in the formu-
las from the set F0 . The language

LCON (P0 , F0 , C0 )

is called a language defined by the set F0 of formulas.

Example 8.10 Consider a language L = L{¬,⇒,∪,∩} (P, F, C) and a following

set F0 of formulas of L

F0 = {∃xQ(c, g(x, d)), ¬∀x(P (x) ⇒ ∃y(R(f (x), y)∩¬P (e))), ¬(F (a)∩R(y, h(c))}.

A language defined by the set F0 of formulas is

L{¬,⇒,∪,∩} ({P, R, Q, F }, {g, f, h}, {a, c, d, e}),

where # Q = #R = 2, #P =# F = 1, # g = 2, #f = # h = 1.

It is common practice to use the notation

A(x1 , x2 , ..., xn ) (8.6)

to indicate that F V (A) ⊆ {x1 , x2 , ..., xn } without implying that all of x1 , x2 , ..., xn
are actually free in A. This is similar to the practice in algebra of writing
p(x1 , x2 , ..., xn ) for a polynomial p in the variables x1 , x2 , ..., xn without imply-
ing that all of them have nonzero coefficients.

Definition 8.6 (Replacing x by t in A)

If A(x) is a formula, and t is a term then A(x/t) or, more simply, A(t) de-
notes the result of replacing all occurrences of the free variable x by the term t
throughout. When using the notation A(t) we always assume that none of the
variables in t occur as bound variables in A.

The assumption that none of the variables in t occur as bound variables in A

is essential, otherwise by substituting t on the place of x we would distort the
meaning of A(t). Let t = y and A(x) is ∃y(x 6= y), i.e. the variable y in t is
bound in A. The substitution of t for x produces a formula A(t) of the form
∃y(y 6= y), which has a different meaning than ∃y(x 6= y).
But if t = z, i.e. the variable z in t is not bound in A, then A(x/t) = A(t) is
∃y(z 6= y) and express the same meaning as A(x).
Remark that if for example t = f (z, x) we obtain ∃y(f (z, x) 6= y) as a result of
substitution of t = f (z, x) for x in ∃y(x 6= y).

369
This notation is convenient because we can agree to write as

A(t1 , t2 , ..., tn ) or A(x1 /t1 , x2 /t2 , . . . , xn /tn )

a result of substituting in A the terms t1 , t2 , . . . , tn for all free occurrences (if

any) of x1 , x2 , . . . , xn , respectively. But when using this notation we always
assume that none of the variables in t1 , t2 , ..., tn occur as bound variables in A.
The above assumption that none of the variables in t1 , t2 , ..., tn occur as bound
variables in A is often expressed using the notion: t1 , t2 , ,̇tn are free for all theirs
variables in A which is defined formally as follows.

Definition 8.7 (Term t free for y in A)

If A ∈ F and t is a term, then t is said to be free for y if no free occurrence of
y lies within the scope of any quantifier bounding variables in t.

Example 8.11 Let A , B be the formulas

∀yP (f (x, y), y), ∀yP (f (x, z), y),

respectively. The term t = f (x, y) is free for x and is not free for y in A. The
term t = f (x, z) is free for x and z in B. The term t = y is not free neither for
x nor for z in A, B.

Example 8.12
Let A be a formula

(∃xQ(f (x), g(x, z)) ∩ P (h(x, y), y)).

The term t1 = f (x) is not free for x in A; the term t2 = g(x, z) is free for z
only, term t3 = h(x, y) is free for y only because x occurs as a bound variable
in A; term t4 .

Definition 8.8 (Replacement)

If A(x), A(x1 , x2 , ..., xn ) ∈ F and t, t1 , t2 , ..., tn ∈ T , then A(x/t), A(x1 /t1 , x2 /t2 , . . . , xn /tn )
or, more simply just
A(t), A(t1 , t2 , ..., tn )
denotes the result of replacing all occurrences of the free variables x, x1 , x2 , ..., xn ,
by the terms t, t, t1 , t2 , ..., tn , respectively, assuming that t, t1 , t2 , ..., tn are free
for all theirs variables in A.

Classical Restricted Domain Quantifiers

370
We often use logic symbols, while writing mathematical statements. For exam-
ple mathematicians in order to say ”all natural numbers are greater then zero
and some integers are equal 1” often write it as

x ≥ 0, ∀x∈N and ∃y∈Z , y = 1.

Some of them, who are more ”logic oriented”, would also write it as

∀x∈N x ≥ 0 ∩ ∃y∈Z y = 1,

or even as
(∀x∈N x ≥ 0 ∩ ∃y∈Z y = 1).

None of the above symbolic statements are formulas of the predicate language
L. These are mathematical statement written with mathematical and logic
symbols. They are written with different degree of ”logical precision”, the last
being, from a logician point of view the most precise.
Observe that the quantifiers in ∀x∈N and ∃y∈Z used in all of them are not
the one used in the predicate language L, which admits only quantifiers ∀x
and ∃y, for any variables x, y ∈ V AR. The quantifiers ∀x∈N , ∃y∈Z are called
quantifiers with restricted domain. The first is restricted to the domain of
natural numbers, the second to the integers. The restriction of the quantifier
domain can, and often is given by more complicated statements. For example
we say ”for all x > 2” and write ∀x>2 , or we say ”exists x > 2 and at same time
x + 2 < 8” and write symbolically ∃(x>2∩x+2<8) .
Our goal now is to correctly ”translate ” mathematical and natural language
statement into formulas of the predicate language L of the classical predicate
logic with the the set {¬.∩, ∪, ⇒} of propositional connectives. We say ”classical
predicate logic” to express that we define all notions for the classical semantics
to be defined formally in the next section 8.2. One can extend these notions
to non-classical logics, but we describe and will talk only about classical case.
We introduce the quantifiers with restricted domain into the classical predicate
logic language by expressing them within the language L as follows.

Definition 8.9
Given a classical predicate logic language

L = L{¬,∩,∪,⇒,¬} (P, F, C).

The quantifiers ∀A(x) , ∃A(x) are called quantifiers with restricted domain,
or restricted quantifiers, where A(x) ∈ F is any formula with any free vari-
able x ∈ V AR.
A formula ∀A(x) B(x) is an abbreviation of a formula ∀x(A(x) ⇒ B(x)) ∈ F.
We write it symbolically as

∀A(x) B(x) = ∀x(A(x) ⇒ B(x)). (8.7)

371
A formula ∃A(x) B(x) stands for a formula ∃x(A(x) ∩ B(x)) ∈ F. We write it
symbolically as
∃A(x) B(x) = ∃x(A(x) ∩ B(x)) (8.8)

The definition 8.9 of restricted quantifiers is obviously faithful to our intuitive

meaning of quantifiers. We use informally a symbol = to stress that we they
are in a sense equivalent with respect to classical semantics. We call (8.7) and
(8.8) transformations rules for restricted quantifiers.

Exercise 8.2
Given a mathematical statement S written with logical symbols
(∀x∈N x ≥ 0 ∩ ∃y∈Z y = 1)
1. Translate it into a proper logical formula with restricted domain quantifiers
i.e. into a formula of L that uses the restricted domain quantifiers.
2. Translate your restricted domain quantifiers logical formula into a correct
logical formula without restricted domain quantifiers, i.e. into a formula of L.

Solution
We proceed to write this and other similar problems solutions in a sequence of
steps.

STEP 1. We identify basic statements in S, i.e. mathematical statements that

involve only relations. They will be translated into atomic formulas. We iden-
tify the relations in the basic statements and choose the predicate symbols as
their names. We identify all functions and constants (if any) in the basic state-
ments and choose the function symbols and constant symbols as their names.

The basic statements in S are: x ∈ N, x ≥ 0, y ∈ Z, y = 1. The relations are:

∈ N, ∈ Z, ≥, =. We use one argument predicates symbols N, Z for ∈ N, ∈ Z,
respectively. We use two argument predicate symbols G for ≥, and E for =.
There are no functions. We have two constant symbols c1 , c2 for numbers 0 and
1, respectively.

STEP 2. We write the basic statements as atomic formulas of L.

We write N (x), Z(x) for x ∈ N, x ∈ Z, respectively. G(x, c1 ) for x ≥ 0 and

E(y, c2 ) for y = 1. These are all atomic formulas.

STEP 3. We re-write the statement S a logical formula with restricted domain

quantifiers.

372
The statement S becomes a restricted quantifiers formula:

(∀N (x) G(x, c1 ) ∩ ∃Z(y) E(y, c2 )).

STEP 4. We apply (8.7) and (8.8) to the formula from STEP 3. and obtain a
formula A of L as a representation of the given mathematical statement S.

A formula A ∈ F that corresponds to S is

(∀x (N (x) ⇒ G(x, c1 )) ∩ ∃y (Z(y) ∩ E(y, c2 ))).

Here is a perfectly acceptable short solution. We presented the long solution in

order to explain all steps needed to be performed when one writes a solution.

Short Solution
The basic statements in S are: x ∈ N, x ≥ 0, y ∈ Z, y = 1. The corresponding
atomic formulas of L are: N (x), G(x, c1 ), Z(y), E(y, c2 ), respectively.
The statement S becomes becomes a restricted quantifiers formula
(∀N (x )G(x, c1 ) ∩ ∃Z(y) E(y, c2 )). Applying restricted quantifiers definition 8.9
and transformation rules (8.7), (8.8) we get a following formula A ∈ F

(∀x(N (x) ⇒ G(x, c1 )) ∩ ∃y(Z(y) ∩ E(y, c2 ))).

8.2 Classical Semantics

The notion of predicate tautology is much more complicated then that of the
propositional. Predicate tautologies are also called valid formulas, or laws of
quantifiers to distinguish them from the propositional case. The formulas of a
predicate language L have meaning only when an interpretation is given for all
its symbols. We define an interpretation I by interpreting predicate, functional
symbols as a concrete relation, function defined in a certain set U 6= ∅, and
constants symbols as elements of the set U. The set U is called the universe
of the interpretation I. These two items specify a structure M = (U, I) for the
language L.
.
The semantics for a first order language L in general, and for the first order
classical logic in particular, is defined, after Tarski (1936) in terms of the struc-
ture M = [U, I], an assignment s of L, and a satisfaction relation (M, s) |= A
between structures, assignments and formulas of L.

The definition of a structure M = [U, I] and the assignment s of L is common

for different predicate languages and for different semantics and we define them

373
as follows.

Definition 8.10 (Structure)

Given a predicate language L = LCON (P, F, C). A structure for L is a pair
M = [U, I],
where U is a non empty set called a universe and I is an assignment called an
interpretation of the language L(P, F, C) in the universe U defined as follows.

1. I assigns to any predicate symbol P ∈ P a relation PI defined in the

universe U . I.e. for any P ∈ P, if #P = n, then
PI ⊆ U n .

2. I assigns to any functional symbol f ∈ F a function fI defined in the

universe U . I.e. for any f ∈ F, if #f = n, then
fI : U n −→ U.

3. I assigns to any constant symbol c ∈ C an element cI of the universe. I.e

for any c ∈ C,
cI ∈ U.

Example 8.13
Let L be a language with one two-place predicate symbol, two functional symbols:
one -place and one two-place, and two constants, i.e.
L = L({R}, {f, g}, {c, d}, )
where #R = 2, #f = 1, #g = 2, and c, d ∈ C.
We define a structure M = [U, I] as follows. We take as the universe the set
U = {1, 3, 5, 6}. The predicate R is interpreted as ≤, what we write as RI : ≤.
We interpret f as a function fI : {1, 3, 5, 6} −→ {1, 3, 5, 6} such that fI (x) = 5
for all x ∈ {1, 3, 5, 6}, and we put gI : {1, 3, 5, 6} × {1, 3, 5, 6} −→ {1, 3, 5, 6}
such that gI (x, y) = 1 for all x ∈ {1, 3, 5, 6}. The constant c becomes cI = 3,
and dI = 6, what we write as We write the structure M as
M = [{1, 3, 5, 6} ≤, fI , gI , cI = 3, dI = 6]

Exercise 8.3
Given a language
L = L({R}, {g}, ∅, )
where #R = 2, #g = 2 . Define two structures for L, both with infinite universe:
one infinitely countable and one uncountable.

374
Solution
There are many such structures. Here are two of the very simple.
M1 = [N, , ≤, +], where N is the set of natural numbers, and for example
M2 = [R, , ≤, +], where R is the set of real numbers.

Definition 8.11 (Assignment)

Given a first order language L = L(P, F, C) with the set V AR of variables. Let
M = [U, I] be a structure for L with the universe U 6= ∅.
An assignment of L in M = [U, I] is any function

s : V AR −→ U (8.9)

The assignment s is also called an interpretation of variables VAR of L in a

structure M = [U, I].

Let M = [U, I] be a structure for L and s : V AR −→ U be an interpretation

of variables VAR of L in a structure M.
Let T be the set of all terms of L. By definition 10.11, V AR ⊂ T. We use the
interpretation I to extend the assignment s to the set the set T of all terms
of L. Because of that we denote this extension by I rather then by s∗ as we
did before. The extension I of v is hence a mapping from T to U . It associates
with each t ∈ T an element I(t) ∈ U . We denote this element I(t) by tI . We
define the extension I(t) = tI of s by the induction of the length of the term
t ∈ T and call it an interpretation of terms of L in a structure M = [U, I].

Definition 8.12 (Interpretation of Terms)

Given a language L = L(P, F, C) and a structure M = [U, I] for L. Let an
assignment
s : V AR −→ U
be any interpretation of variables VAR of L (assignment) in M.
We extend s to a function
sI : T −→ U (8.10)
called an interpretation sI of terms of L in M.
The function sI is defined by induction on the complexity of terms as follows.

1. For any x ∈ V AR,

sI (x) = s(x);

2. for any c ∈ C,
sI (c) = cI ;

375
3. for any t1 , t2 , . . . , tn ∈ T, n ≥ 1, f ∈ F, such that #f = n, and for any
term t = f (t1 , t2 , . . . , tn ), we put

sI (t) = fI (sI (t1 ), sI (t2 ), . . . , sI (tn )),

i.e. we define

sI (f (t1 , t2 , . . . , tn )) = fI (sI (t1 ), sI (t2 ), . . . , sI (tn )),

for any t1 , t2 , . . . , tn ∈ T, f ∈ T, such that #f = n.

Exercise 8.4
Consider a language

L = L({P, R}, {f, g, h, }, {c, d})

for # P = # R = 2, #f = # g=1, # h= 2.
Let M = [{0, 1}, I], where the interpretation I is defined as follows.
PI = {(0, 0)}, RI = {(0, 0), (1, 1)},
fI (0) = 0, fI (1) = 0, gI (0) = 1, gI (1) = 1, hI is given by a formula
hI (x, y) = x for all (x, y) ∈ {0, 1} × {0, 1}, and cI = 1, dI = 0.

The assignment s : V AR −→ {0, 1} is such that s(x) = s(y) = s(z) = 1 = 1,

for x, y, z ∈ V AR (and any values for all other variables).
Given a set T0 of terms, evaluate sI (t) for all t ∈ T0 .
T0 = {z, y, x, c, f (c), f (x), g(z), f (g(d)), g(f (g(z))), h(c, f (g(d))), h(f (x), g(z))}.

Solution
First we evaluate terms that are variables and constants of L using the formulas
1. and 2. of definition 8.12: sI (x) = s(x), sI (c) = cI , respectively and obtain:
sI (z) = s(z) = 1, sI (y) = s(y) = 1, sI (x) = s(x) = 1, sI (c) = cI = 1, sI (d) =
dI = 2. We use the formula sI (f (t1 , t2 , . . . , tn )) = fI (sI (t1 ), sI (t2 ), . . . , sI (tn )
to evaluate the rest of terms in T0 and obtain:
sI f (c) = fI (sI (c)) = fI (cI )) = fI (1) = 0, sI f (x) = fI (sI (x)) = fI (1) = 0
sI g(z) = gI (sI (c)) = gI (1) = 1, sI (f (g(d)) = fI (sI (f (g(d))) = fI (gI (sI (d))) =
fI (gI (cI )) = fI (gI (1)) = fI (1) = 0,
sI g(f (g(z)) = gI (fI (gI (sI (z)))) = gI (fI (gI (1))) = gI (fI (1)) = gI (0) = 1,
sI (h(c, f (g(d)))) = hI (sI (c), sI (f (g(d)))) = hI (cI , fI (gI (sI (d))))
= hI (1, fI (gI (0))) = hI (1, fI (1)) = hI (1, 0) = 1,

376
sI (h(f (x), g(z))) = hI (fI (sI (x)), gI (sI (x))) = hI (fI (1), gI (1)) = hI (0, 1) = 0.

Observe that the interpretation of predicate symbols is irrelevant when eval-

uating an interpretation of terms, as the terms do not involve the predicate
symbols.

Example 8.14
Consider a language
L = L({P, R}, {f, h}, ∅ )
for # P = # R = 2, #f = 1, # h= 2.
Let M = [Z, I], where Z is the set on integers and the interpretation I for
elements of F and C is as follows.
fI : Z −→ Z is given by formula f (m) = m + 1 for all m ∈ Z.
hI : Z × Z −→ Z is given by formula f (m, n) = m + n for all m, n ∈ Z.
Let s : V AR −→ Z be any assignment such that s(x) = −5, s(y) = 2 and
t1 , t2 ∈ T be t1 = h(y, f (f (x))) and t2 = h(f (x), h(x, f (y)).
We evaluate:
sI (t1 ) = sI (h(y, f (x)) = hI (sI (y), fI (sI (x))) = +(2, fI (−5)) = 2 − 4 = −2,

sI (t2 ) = sI (h(f (x), h(x, f (y))) = +(fI (−5), +(−5, 3)) = −4 + (−5 + 3) = −6.

For any t ∈ T, let x1 , x2 , . . . , xn ∈ V AR be all variables appearing in t, we

write it, in a similar way as we did in (8.6) for variables in formulas, as

t(x1 , x2 , . . . , xn ).

Observation 8.1
For any term t(x1 , x2 , . . . , xn ) ∈ T, any structure M = [U, I] and any assign-
ments s, s0 of L in M, the following holds.
If s(x) = s0 (x) for all x ∈ {x1 , x2 , . . . , xn }, i.e the assignments s, s0 agree on
all variables appearing in t, then, sI (t) = s0 I (t).

Thus for any t ∈ T, the function sI : T −→ U defined by (8.10) depends on

only a finite number of values of s(x) for x ∈ V AR.
Given a structure M = [U, I] and an assignment s : V AR −→ U . We write
a
s(x) (8.11)

377
to denote any assignment s0 : V AR −→ U such that s, s0 agree on all variables
except on x, such that s0 (x) = a, for certain a ∈ U .

Given a first order (predicate) language L = LCON (P, F, C). The satisfaction
relation (M, s) |= A between structures, assignments and formulas of L is de-
fined by induction on the complexity of formulas of L. It is the satisfaction
relation (M, s) |= A that allows us to distinguish one one semantics for a
given L from the other, and consequently one logic from the other. We define
now only a classical satisfaction and the notion of classical predicate tautology.

Definition 8.13 (Classical Satisfaction)

Given a classical predicate (first order) language

L = L{¬,∩,∪,⇒,¬} (P, F, C). (8.12)

Let M = [U, I] be a structure for L, s be any assignment of L in M, i.e.

s : V AR −→ U . Let A ∈ F be any formula of L. We define a relation

(M, s) |= A

that reads: the assignment s satisfies the formula A in M, by induction on the

complexity of A as follows.

(i) A as atomic formula (8.5), i.e. A is P (t1 , t2 , . . . , tn ) for P ∈ P, #P = n.

(M, s) |= P (t1 , t2 , . . . , tn ) if and only if (sI (t1 ), sI (t2 ), . . . , sI (tn )) ∈ PI .

(ii) A as not atomic formula and has one of connectives of L as the main
connective.

(M, s) |= ¬A if and only if (M, s) 6|= A,

(M, s) |= (A ∩ B) if and only if (M, s) |= A and (M, s) |= B,

(M, s) |= (A ∪ B) if and only if (M, s) |= A or (M, s) |= B or both,

(M, s) |= (A ⇒ B) if and only if ether (M, s) 6|= A or else (M, s) |= B or

both.

(iii) A as not atomic formula and begins with one of the quantifiers.

(M, s) |= ∃xA if and only if there is s0 such that s, s0 agree on all variables
except on x, and (M, s0 ) |= A,

(M, s) |= ∀xA if and only if for all s0 such that s, s0 agree on all variables
except on x, and (M, s0 ) |= A.

378
Observe that that the truth or falsity of (M, s) |= A depends only on the values
of s(x) for variables x which are actually free in the formula A. This is why we
often write the condition (iii) as

(iii)’ A(x) as not atomic formula (with a free variable x) and begins with one
of the quantifiers.
(M, s) |= ∃xA(x) if and only if there is s0 such that s(y) = s0 (y) for all
y ∈ V AR − {x}, and (M, s0 ) |= A(x),
(M, s) |= ∀xA if and only if for all s0 such that s(y) = s0 (y) for all
y ∈ V AR − {x}, and (M, s0 ) |= A.

Exercise 8.5
For the structures Mi , find assignments si , s0 i (1 ≤ i ≤ 4), such that

(Mi , si ) |= Q(x, c), (Mi , s0 i ) 6|= Q(x, c) for Q ∈ P, c ∈ C.

The structures Mi are defined as follows (the interpretation I for each of them
is specified only for symbols in the formula Q(x, c), and N denotes the set of
natural numbers.

M1 = [{1}, QI :=, cI : 1], M2 = [{1, 2}, QI :≤, cI : 1],

M3 = [N, QI :≥, cI : 0], and M3 = [N, QI :≥, cI : 1.]

Solution
Consider M1 = [{1}, QI :=, cI : 1]. Observe that all s : V AR −→ {1} must
are defined by a formula s(x) = 1 for all x ∈ V AR. We evaluate (definition
8.12), sI (x) = 1, sI (c) = cI = 1. By definition 8.13, (M1 , s) |= Q(x, c) if and
only if (sI (x), sI (c)) ∈ QI , i.e. (1, 1) ∈= what is true as 1 = 1. We have
proved

(M1 , s) |= Q(x, c) for all assignments s : V AR −→ {1}.

Consider M2 = [{1, 2}, QI :≤, cI : 1]. Let s : V AR −→ {1, 2} be any

assignment, such that s(x) = 1. We evaluate sI (x) = 1, sI (c) = 1 and verify
whether (sI (x), sI (c)) ∈ QI , i.e. whether (1, 1) ∈≤ . This is true as 1 ≤ 1. We
have found s (in fact uncountably many such s) such that

(M2 , s) |= Q(x, c).

Let now s0 be any assignment s0 : V AR −→ {1, 2}, such that s0 (x) = 2. We

evaluate s0 I (x) = 1, s0 I (c) = 1 and verify whether s0 I (x), s0 I (c)) ∈ QI , i.e.
whether (2, 1) ∈≤ . This is not true as 2 6≤ 1. We have found s0 6= s (in fact
uncountably many such s’) such that

(M2 , s0 ) 6|= Q(x, c).

379
Consider M3 = [N, QI :≥, cI : 0]. Let s : V AR −→ N be any assignment, such
that s(x) = 5. We evaluate sI (x) = 5, sI (c) = 0. Observe that the condition
(sI (x), sI (c)) ∈ QI holds as 5 ≥ 0 and

(M3 , s) |= Q(x, c).

Let now s0 be any assignment s0 : V AR −→ N . By definition, s0 (x) = n, for

any n ∈ N , and s0 I (x), 0) ∈ QI holds for any s0 as n ≥ 0 for all n ∈ N . This
proves that there is no s0 , such that (M3 , s0 ) 6|= Q(x, c).

Consider M4 = [N, QI :≥, cI : 1]. Let s : V AR −→ N be any assignment, such

that s(x) = 5. We evaluate sI (x) = 5, sI (c) = 1. Observe that the condition
(sI (x), sI (c)) ∈ QI holds as 5 ≥ 1 and hence

(M4 , s) |= Q(x, c).

Let now s0 be any assignment s0 : V AR −→ N , such that s0 (x) = 0. The the

condition (s0 I (x), s0 I (c)) ∈ QI does not holds as 0 6≥ 1 and

(M4 , s0 ) 6|= Q(x, c).

Directly from the definition8.13 we have that the following holds.

Example 8.15
Let Mi (1 ≤ i ≤ 4) be structures in defined the exercise 8.5 and let corresponding
assignments si be as defined as its solutions.
1. (M1 , s) |= Q(x, c), (M1 , s) |= ∀xQ(x, c), (M1 , s) |= ∃xQ(x, c).
2. (M2 , s) |= Q(x, c), (M2 , s) 6|= ∀xQ(x, c), (M1 , s) |= ∃xQ(x, c).
3. (M3 , s) |= Q(x, c), (M3 , s) |= ∀xQ(x, c), (M3 , s) |= ∃xQ(x, c).
4. (M4 , s) |= Q(x, c), (M4 , s) 6|= ∀xQ(x, c), (M4 , s) |= ∃xQ(x, c).

Definition 8.14 (Model)

Given a language L, a formula A of L, and a structure M = [U, I] for L.
The structure M is a model for the formula A if and only if (M, s) |= A
for all s : V AR −→ U. We denote it as M |= A.
For any set Γ ⊆ F of formulas of L, M is a model for Γ if and only if
M |= A for all A ∈ Γ. We denote it as M |= Γ.

We define now a very important semantic notion. It has different names: logi-
cal consequence, logical implication, semantic consequence, logical (semantical)
entailment. We use a name logical consequence and define it as follows.

380
Definition 8.15 (Logical Consequence)
For any A, B ∈ F and any set Γ ⊆ F of formulas of L, we say that a formula
B is a logical consequence of a set Γ and write it as Γ |= B, if and only if
all models of the set Γ are models of the formula B.

When Γ |= B we also say that Γ logically implies B. When Γ = {A} we write

it as A |= B and say A logically implies B.
We say that A and B are logically equivalent if and only if A |= B and
A |= B.

Directly from the model definition 9.3 we get the following.

Definition 8.16 (Counter Model)

Given a language L = L{¬,∩,∪,⇒,¬} (P, F, C), a formula A of L, and a structure
M = [U, I] for L.
The structure M = [U, I] is a counter model for the formula A if and only
if there is an assignment s : V AR −→ U , such that (M, s) 6|= A.
We denote it as M 6|= A.
For any set Γ ⊆ F of formulas of L, M is a counter model for Γ if and only
if there is A ∈ Γ, such that M 6|= A.
We denote it as M 6|= Γ.

Observe that if A is a sentence (definition 8.4) then the truth or falsity of

(M, s) |= A is completely independent of s. Hence if (M, s) |= A for some s, it
holds for all s and the following holds.

Fact 8.1
For any formula A of L,
If A is a sentence, then if there s such that (M, s) |= A, then M is a model for
A, i.e. M |= A

We transform any formula A of L into a certain sentence by binding all its

free variables. The resulting sentence is called a closure of A and is defined as
follows.

Definition 8.17 (Closure)

Given a formula A of L.

381
By the closure of A we mean the formula obtained from A by prefixing in
universal quantifiers all variables the are free in A. If A does not have free
variables (i.e. is a sentence), the closure if A is defined to be A itself.

Obviously, a closure of any formula is always a sentence. For example, if A, B

are formulas
(P (x1 , x2 ) ⇒ ¬∃x2 Q(x1 , x2 , x3 )), (∀x1 P (x1 , x2 ) ⇒ ¬∃x2 Q(x1 , x2 , x3 )),
their respective closures are
∀x1 ∀x2 ∀x3 ((P (x1 , x2 ) ⇒ ¬∃x2 Q(x1 , x2 , x3 ))),
∀x1 ∀x2 ∀x3 ((∀x1 P (x1 , x2 ) ⇒ ¬∃x2 Q(x1 , x2 , x3 ))).

Example 8.16
Let Q ∈ P, #Q = 2 and c ∈ C Consider formulas
Q(x, c), ∃xQ(x, c), ∀xQ(x, c)
and the structures from exercise 8.5 defined as follows.
M1 = [{1}, QI :=, cI : 1], M2 = [{1, 2}, QI :≤, cI : 1],
M3 = [N, QI :≥, cI : 0], and M4 = [N, QI :≥, cI : 1.]
Directly from example 8.15 and Fact 8.1, we get that:
1. M1 |= Q(x, c), M1 |= ∀xQ(x, c), M1 |= ∃xQ(x, c).
2. M2 6|= Q(x, c), M2 6|= ∀xQ(x, c), M2 |= ∃xQ(x, c).
3. M3 |= Q(x, c), M3 |= ∀xQ(x, c), M3 |= ∃xQ(x, c).
4. M4 6|= Q(x, c), M4 6|= ∀xQ(x, c), M4 |= ∃xQ(x, c).

Definition 8.18 (True, False in M)

Given a structure M = [U, I] for L and a formula A of L. We say that:
A is true in M (written as M |= A) if and only if all assignments s of L in
M satisfy A, i.e. when M is a model for A.
A is false in M (written as M =| A) if and only if no assignment s of L in
M satisfies A.

By the definition 9.3 we have that A is true in M only when the structure M
is a model for A. This is why we use the notation M |= A in both cases.
Obviously, if A is not true in M, then it is false, and vice versa. This proves
correctness of our definition with respect to the intuitive understanding.
We get directly from definition 8.18 and the example 8.16 the following.

382
Example 8.17
Let M1 − M4 be structures defined in example 8.5.
1. Formulas Q(x, c), ∀xQ(x, c), ∃xQ(x, c) are all true in the structures M1
and M3 .
2. Formula ∃xQ(x, c) is also true in M2 and in M3 .
3. Formulas ¬Q(x, c), ¬∀xQ(x, c), ¬∃xQ(x, c) are all false in the structures
M1 and M3 .
4. Formula ¬∃xQ(x, c) is also false in M2 and in M3 .
5. Formulas (Q(x, c) ∩ ¬Q(x, c)), (¬∀xQ(x, c) ∩ ∀xQ(x, c)), and the formula
(∃xQ(x, c) ∩ ¬∃xQ(x, c)) are all false in all structures M1 − M4 .
6. The formula ∀xQ(x, c) is false in a structure M5 = [N, QI :<, cI : 0].

Here are some properties of the notions ”A is true in M, written symbolically as

M |= A, and ”A is false in M, written symbolically as M=| A. They are obvious
under intuitive understanding of the notion of satisfaction. Their formal proofs
are left as exercise for the reader.

Property 8.1 (Truth, Falsity, Satisfaction)

Given a structure M = [U, I] for L and any formulas formula A, B of L. The
following properties hold.
P1. A is false in M if and only if ¬A is true in M, i.e.

M =| A if and only if M |= ¬A.

P2. A is true in M if and only if ¬A is false in M, i.e.

M |= A if and only if M =| ¬A.

P3. It is not the case that both M |= A and M |= ¬A, i.e. no formula of L
can be both true and false in M, i.e. there is no formula A, such that M |= A
and M =| A.
P4. If M |= A and M |= (A ⇒ B), then M |= B.
P5. (A ⇒ B) is false in M if and only if M |= A and M |= ¬B, i.e.

M =| (A ⇒ B) if and only if M |= A and M |= ¬B.

P6. M |= A if and only if M |= ∀xA.

P7. A formula A its true in M if and only if its closure (definition 8.17) is
true in M.

383
Definition 8.19 (Valid, Tautology)
Given a language L = L{¬,∩,∪,⇒,¬} (P, F, C), a formula A of L. .
A formula A is predicate tautology (is valid) if and only if M |= A for
all structures M = [U, I], i.e. when A is true in all structures M for L.
We write
|= A or |=p A,
to denote that a formula A is predicate tautology (is valid).

We write
|=p A
when there is a need to stress a distinction between propositional and predicate
tautologies, otherwise we will use the symbol |= .
Predicate tautologies are also called laws of quantifiers.
Following the notation T for the set of all propositional tautologies (chapter 5)
we denote by Tp the set of all predicate tautologies, i.e.

Tp = {A of L{¬,∩,∪,⇒,¬} (P, F, C) : |=p A}. (8.13)

Directly from the definition 8.18, the tautology definition 8.19 we get the fol-
lowing basic properties of logical consequence as defined by definition 8.15.

Property 8.2
For any A, B ∈ F and any set Γ ⊆ F of formulas of L,
P1. A |= B if and only if |= (A ⇒ B).
P2. If A |= B and A is true in M, then B is true in M.
P2. If Γ |= B and if all formulas in Γ are true in M, then B is true in M.

We get immediately from the above definition 8.19 of a following definition of a

notion ” A is not a predicate tautology”.

Definition 8.20
For any formula A of predicate language L,
A is not a predicate tautology (6|= A) if and only if there is a structure
M = (U, I) for L, such that M 6|= A.
We call such structure M a counter-model for A.

The definition 8.20 says: to prove that a formula A is not a predicate tautology
one has to show a counter- model M = (U, I). It means one has to show a non-
empty set U, define an interpretation I, and an assignment s : V AR −→ U

384
such that (M, s) 6|= A.

We introduce, similarly as in a case of propositional semantic a notion of pred-

icate contradiction.

Definition 8.21 (Contradiction)

For any formula A of predicate a language L,
A is a predicate contradiction if and only if A is false in all structures M.
We denote it as =| A and write symbolically

=| A if and only if M=| A, for all structures M.

When there is a need to distinguish between propositional and predicate contra-

dictions we also use symbol
=|p A,
where ”p” stands for ”predicate”.

Following the notation C for the set of all propositional tautologies (chapter
5) we denote by Cp the set of all predicate contradictions, i.e.

Cp = {A of L{¬,∩,∪,⇒,¬} (P, F, C) : =|p A}. (8.14)

Directly from the definition 8.18 and Property 8.1 we have the folowing duality
property, the same as the one for propositional logic.

Fact 8.2
For any formula A of predicate a language L,

A ∈ Tp if and only if ¬A ∈ Cp ,

A ∈ Cp if and only if ¬A ∈ Tp .

Obviously, the formulas (Q(x, c) ∩ ¬Q(x, c)), (¬∀xQ(x, c) ∩ ∀xQ(x, c)), and the
formula (∃xQ(x, c) ∩ ¬∃xQ(x, c)) defined in example 8.17 are not only false in
the structures M1 − M4 , but are false in all structures M for L. By definition
8.21 they all are predicate contradictions. Observe that they all are substitutions
of propositional contradictions (a ∩ ¬a) or (¬a ∩ a). By the same argument the
formulas (Q(x, c)∪¬Q(x, c)), (¬∀xQ(x, c)∪∀xQ(x, c)), (∃xQ(x, c)∩¬∃xQ(x, c))
are predicate tautologies as they are substitutions of propositional tautologies
(a ∪ ¬a) or (¬a ∪ a).
We put these examples and observations in a following theorems that establish
relationship between propositional and predicate tautologies and contradictions.

385
We write now |=, =| do denote respectively propositional tautologies and con-
tradiction, and |=p , =|p for predicate tautologies and contradictions. We first
formalize and prove (theorem 8.1) the intuitively obvious fact: if a formula A is
a propositional tautology (contradiction), then replacing propositional variables
in A by any formulas of a predicate language we obtain a formula which is a
predicate tautology (contradiction).

Example 8.18
Let consider the following example of a propositional tautology and a proposi-
tional contradiction.

|= ((a ⇒ b) ⇒ (¬a ∪ b)) and =| ((a ∪ ¬a) ⇒ (¬b ∩ b)).

Substituting ∃xP (x, z) for a, and ∀yR(y, z) for b, we obtain, by theorem 8.1,
that

|=p ((∃xP (x, z) ⇒ ∀yR(y, z)) ⇒ (¬∃xP (x, z) ∪ ∀yR(y, z))) and

=| ((∃xP (x, z) ∪ ¬∃xP (x, z)) ⇒ (¬∀yR(y, z) ∩ ∀yR(y, z))).

We put it all in a more formal and more general and precise language as follows.
Given a propositional language L0 = L{¬,∩,∪,⇒,¬} with the set F0 of formulas
and a predicate languageL = L{¬,∩,∪,⇒,¬} (P, F, C) with the set F of formulas.
Let A(a1 , a2 , . . . , an ) ∈ F0 and A1 , A2 , . . . , An ∈ F. We denote by

A(a1 /A1 , a2 /A2 , . . . , an /An ) (8.15)

the result of replacing in A the free variables a1 , a2 , . . . , an by the formulas

A1 , A2 , . . . , An ∈ F. Of course A(a1 /A1 , a2 /A2 , . . . , an /An ) ∈ F.

Theorem 8.1
Given a propositional language L0 with the set F0 of formulas and a predicate
language L with the set F of formulas.
For any A(a1 , a2 , . . . an ) ∈ F0 and any A1 , A2 , . . . An ∈ F the following holds.
1. If |= A(a1 , a2 , . . . , an ), then |=p A(a1 /A1 , a2 /A2 , . . . , an /An ).
2. If =| A(a1 , a2 , . . . , an ), then =|p A(a1 /A1 , a2 /A2 , . . . , an /An ).

Proof 1. follows directly from satisfaction definition 8.13. 2. follows from

definition 8.13, property 8.1, and definition 8.21.
Some predicate tautologies are, by theorem 8.1, substitutions of propositional
formulas. Visibly a predicate formula (∀x A(x) ⇒ ∃x A(x)) can not be obtained
as a substitution of a propositional formula. We prove now that it is a predicate
tautology.

386
Fact 8.3
For any formula A(x) of L,
|= (∀x A(x) ⇒ ∃x A(x)).

Proof
Assume that 6|= (∀x A(x) ⇒ ∃x A(x)). By definition 8.20 there is a structure
M = (U, I) and s : V AR −→ U, such that (M, s) 6|= (∀x A(x) ⇒ ∃x A(x)).
By definition 8.13, (M, s) |= ∀x A(x) and (M, s) 6|= ∃x A(x). It means that
(M, s0 ) |= A(x) for all s0 such that s, s0 agree on all variables except on x,
and it is not true that there is s0 such that s, s0 agree on all variables except
on x, and (M, s0 ) |= A(x). This is impossible and this contradiction proves
|= (∀x A(x) ⇒ ∃x A(x)).
Given a set F of formulas of a predicate language L. We denote by OF set of all
open formulas of L, i.e. formulas without quantifiers. We prove that any open
formula in order to be predicate tautologies must be a substitution definied in
theorem 8.1 of a propositional tautology. I.e. we have the following substitution
theorem.

Theorem 8.2
Any open formula A of a predicate language L is a predicate tautology if and
only if it is a substitution of a propositional tautology as defined in theorem 8.1.

Proof
Observe that every open formula from A ∈ OF is a form
B(a1 /A1 , a2 /A2 , . . . , an /An ) for certain propositional formula B(a1 , a2 , . . . an ),
where A1 , A2 , . . . , An are predicate atomic formulas from the set AF as defined
in (8.5). Theorem 8.2 follows directly from the following.

Lemma 8.1
Let σ be a one to one mapping from the set V0 of propositional variables of
propositional language L0 into the set AF of the atomic formulas of the predicate
language L. For any A(a1 , a2 , . . . an ) ∈ F0 ,
|= A(a1 , a2 , . . . , an ), if and only if |=p A(a1 /σ(a1 ), . . . , an /σ(an )).

Proof of lemma
The implication ”if |= A(a1 , a2 , . . . , an ), then
|=p A(a1 /σ(a1 ), . . . , an /σ(an ))” holds as a particular case of theorem 8.2.
We prove now the converse implication by proving its opposite
if 6|= A(a1 , a2 , . . . , an ), then 6|=p A(a1 /σ(a1 ), . . . , an /σ(an )). (8.16)
Assume 6|= A(a1 , a2 , . . . , an ). There exists a truth assignment v : V0 −→ {T, F }
such that v ∗ (A(a1 , a2 , . . . , an )) = F . We construct a counter model M for

387
A(a1 /σ(a1 ), . . . , an /σ(an )) as follows. Let M = [T, I], where T is the set of all
terms of L, and for any c ∈ C, f ∈ F, P ∈ P we put cI = c, fI (t1 , t2 , . . . , tn ) =
f (t1 , t2 , . . . tn ), PI ⊆ T#P .
Let now the s assignment of L in M be an identity, i.e. s : V AR −→ T is
such that s(x) = x for all x ∈ V AR. We extend s to the interpretation of terms
(definition 8.12) as follows.
sI (x) = s(x) = x, sI (c) = cI = c, sI (f (t1 , t2 , . . . tn )) = fI (sI (t1 ), . . . , sI (tn )) =
f (t1 , t2 , . . . tn ), i.e. we have that sI (t) = t for all t ∈ T.
We have that for every atomic formula P (t1 , t2 , . . . tn ) there is exactly one propo-
sitional variable a, such that P (t1 , t2 , . . . tn ) = σ(a). We define now that PI as
follows.
(t1 , t2 , . . . tn ) ∈ PI if and only if P (t1 , t2 , . . . tn ) = σ(a) and v(a) = T .
(t1 , t2 , . . . tn ) 6∈ PI if and only if P (t1 , t2 , . . . tn ) = σ(a) and v(a) = F .
We assumed that v : V0 −→ {T, F } is such that v ∗ (A(a1 , a2 , . . . , an )) = F .
Directly form definition of the assignment s and the interpretation I we have
that ([T, I], s) 6|= A(a1 /σ(a1 ), . . . , an /σ(an )). It end the roof of lemma 8.1 and
hence the proof of theorem 8.2.

Fact 8.4
The converse implication to (8.3) is not a predicate tautology, i.e. there is a
formula A of L. such that

6|= (∃x A(x) ⇒ ∀x A(x)). (8.17)

Proof
Observe that to prove (8.17) we have to provide an example of an instance of a
formula A(x) and construct a counter-model M = (U, I) for it. Let A(x) be an
atomic formula P (x, c), for any P ∈ P, #P = 2. The instance is

(∃x P (x, c) ⇒ ∀x P (x, c)).

We take as M = (N, PI :<, cI : 3) for N set of natural numbers. Let s be any

assignment s : V AR −→ N. We show now (M, s) |= ∃x P (x, c). Take any s0
such that s0 (x) = 2 and s0 (y) = s(y) for all y ∈ V AR−{x}. We have (2, 3) ∈ PI ,
as 2 < 3 and hence there exists s0 that agrees with s on all variables except on x,
and (M, s0 ) |= P (x, c). But (M, s) 6|= ∀x P (x, c) as for example for s0 such that
s0 (x) = 5 and s0 (y) = s(y) for all y ∈ V AR − {x}, (2, 3) 6∈ PI , as 5 6< 3. This
proves that M = (N, PI :<, cI : 3) is a counter model for ∀x P (x, c). Hence
6|= (∃x A(x) ⇒ ∀x A(x)).
The ”shorthand” solution is: the formula (∃x P (x, c) ⇒ ∀x P (x, c)) becomes
in M = (N, PI :<, cI : 3) a mathematical statement (written with logical
symbols): ∃n n < 3 ⇒ ∀n n < 3. It is an obviously false statement in the set

388
N of natural numbers, as there is n ∈ N , such that n < 3, for example n = 2,
and it is not true that all natural numbers are smaller then 3.
We have to be very careful when we deal with quantifiers with restricted
domain (definition 8.9). We adopt the following definition for restricted domain
quantifiers.

Definition 8.22 (Restricted Quantifiers Tautology)

For any formulas
A(x), B(x) ∈ F with any free variable x ∈ V AR, and for the restricted domain
quantifies ∀B(x) , ∃B(x) we define

|= ∀B(x) A(x) if and only if |= ∀x (B(x) ⇒ A(x)),

|= ∃B(x) A(x) if and only if |= ∃x (B(x) ∩ A(x)).

The most basic predicate tautology (8.3) fails when we use the quantifiers with
restricted domain. We show now that

6|= (∀B(x) A(x) ⇒ ∃B(x) A(x)). (8.18)

By definition 8.22 to prove (8.18) means to prove that corresponding proper

formula of L obtained by the restricted quantifiers transformations rules (8.48),
(8.49) is not a predicate tautology, i.e. to show that

6|= (∀x(B(x) ⇒ A(x)) ⇒ ∃x(B(x) ∩ A(x))). (8.19)

In order to prove (8.19) we have to provide an example of particular formulas

A(x), B(x) and to construct a counter model for these particular formulas. We
take as B(x), A(x) atomic formulas Q(x, c), P (x, c) and we construct a counter
model a corresponding formula

(∀x(Q(x, c) ⇒ P (x, c)) ⇒ ∃x(Q(x, c) ∩ P (x, c))) (8.20)

as follows. We take M = (N, I), where N is the set of real numbers and the
interpretation I is defined as QI :<, PI :>, cI : 0. The ”shorthand” solution is
as follows. The formula 8.19) becomes a mathematical statement

(∀n (n < 0 ⇒ n > 0) ⇒ ∃n∈N (n < 0 ∩ n > 0)).

This statement is a false in the set N of natural numbers because the statement
n < 0 is false for all natural numbers and F ⇒ B is a true implication for
any logical value of B, so ∀n (n < 0 ⇒ n > 0) is a true statement and
∃n (n < 0 ∩ n > 0) is obviously false in the set N of natural numbers.
The restricted quantifiers law corresponding to the predicate tautology (8.3) is:

|= (∀B(x) A(x) ⇒ (∃x B(x) ⇒ ∃B(x) A(x))). (8.21)

389
By definition 8.22 and restricted quantifiers transformations rules (8.7), (8.8)
proving (8.19 ) is means proving

|= (∀x(B(x) ⇒ A(x)) ⇒ (∃x B(x) ⇒ ∃x (B(x) ∩ A(x)))).

We leave the proof and an exercise for the reader.

8.3 Predicate Tautologies

We have already proved in Fact 8.3 the basic predicate tautology

|= (∀x A(x) ⇒ ∃x A(x)).

We are going to prove now the following.

Fact 8.5 (Dictum de Omni)

For any formula A(x) of L,

|= (∀x A(x) ⇒ A(t)), |= (∀x A(x) ⇒ A(x)), (8.22)

|= (A(t) ⇒ ∃x A(x)), (8.23)

where t is a term, A(t) is a result of substitution of t for all free occurrences of
x in A(x), and t is free for x in A(x) (definition 8.7), i.e. no occurrence of a
variable in t becomes a bound occurrence in A(t).

Proof of (8.22) is constructed in a sequence of steps. We leave details to the

reader to complete as an exercise. Here are the steps.
S1 Consider a structure M = [U, I] and s : V AR −→ U. Let t, u be two terms.
Denote by t0 a result of replacing in t all occurrences of a variable x by the term
u, i.e. t0 = t(x/u). Let s0 results from s by replacing s(x) by sI (u). We prove
by induction over the length of t that

sI (t(x/u)) = sI (t0 ) = s0 I (t). (8.24)

S2 Let t be free for x in A(x). A(t) is a results from A(x) by replacing t for
all free occurrences of x in A(x), i.e. A(t) = A(x/t). Let s : V AR −→ U and
s0 be obtained from s by replacing s(x) by sI (t). We use (8.24) and induction
on the number of connectives and quantifiers in A(x) and prove

(M, s) |= A(x/t) if and only if (M, s0 ) |= A(x). (8.25)

S3 Directly from definition 8.13 and (8.25) we get that for any M = (U, I) and
any s : V AR −→ U ,

if (M, s) |= ∀xA(x), then (M, s) |= A(t).

390
This proves that (∀x A(x) ⇒ A(t)) is a predicate tautology. Observe that a
term x is free for x in A(x), so we also get as a particular case of t = x that
|= (∀x A(x) ⇒ A(x)).
Proof of (8.23) follows from (8.22), theorem 8.1, property 8.1, theorem 8.5,
and definability law (8.49). We carry it as follows. First we observe that by
theorem 8.1 we have that |= ((∀x¬A(x) ⇒ ¬A(t)) ⇒ (A(t) ⇒ ¬∀x¬A(x))) as a
substitution of propositional tautology ((a ⇒ ¬b) ⇒ (b ⇒ ¬a)). By just proved
(8.22) we have that |= (∀x¬A(x) ⇒ ¬A(t)) for A(x) being a formula ¬A(x).
By P 4 in property 8.1, we get |= (A(t) ⇒ ¬∀x¬A(x)). We apply the existential
quantifier definability law (8.49) and equivalence substitution theorem 8.5 and
get |= (A(t) ⇒ ∃x A(x)). This ends the proof of (8.22).

Remark the restrictions in (8.22) and (8.23) are essential. Here is a simple
example explaining why they are needed in (8.22). The example for (8.23) is
similar.
Let A(x) be a formula ¬∀y P (x, y), for P ∈ P. Notice that a term t = y is not
free for y in A(x). Consider (8.22) A(x) = ¬∀y P (x, y) and t = y.
(∀x¬∀y P (x, y) ⇒ ¬∀y P (y, y)), (8.26)
Take M = [N, I] for I such that PI : =. Obviously, M |= ∀x¬∀y P (x, y)
as ∀n¬∀n(n = n) is a true mathematical statement in the set N of natural
numbers. M 6|= ¬∀y P (y, y) as ¬∀n (n = n) is a false statement for n ∈ N .
Hence M is a counter model for for (8.26) and we proved that without the
restriction (8.22) does not hold.
Here are some useful and easy to prove properties of the notion ”t free for x in
A(x)” (definition 8.7).

Property 8.3 (t free for x in A(x))

For any formula A ∈ F and any term t ∈ T the following properties hold.
P1. A closed tern t, i.e. term with no variables is free for any variable x in A.
P2. A term t is free for any variable in A if none of the variables in t is bound
in A.
P3. Term t = x is free for x in any formula A.
P4. Any term is free for x in A if A contains no free occurrences of x.

Here are some more important predicate tautologies.

Generalization
For any formulas A(x), B(x), A, B of L, where A, B does not contain any free
occurrences of x,
|= ((B ⇒ A(x)) ⇒ (B ⇒ ∀x A(x))), (8.27)

391
 |= ((B(x) ⇒ A) ⇒ (∃xB(x) ⇒ A)), (8.28)

Distributivity 1
For any formulas A(x), B(x), A, B of L, such that A , B does not contain any
free occurrences of x,

|= (∀x(A ⇒ B(x)) ⇒ (A ⇒ ∀x B(x))), (8.29)

|= ∀x(A(x) ⇒ B) ⇒ (∃xA(x) ⇒ B) (8.30)

|= ∃x(A(x) ⇒ B) ⇒ (∀xA(x) ⇒ B) (8.31)

The restrictions that the formulas A, B do not contain any free occurrences of
x is essential for both Generalization and Distributivity 1 tautologies.
Here is a simple example explaining why they are needed in (8.29). The re-
laxation of the assumption that A, B do not contain any free occurrences of x
would lead to the following disaster. Let A and B be both atomic formula P(x).
Thus x is free in A and we have the following instance of (8.29).

(∀x(P (x) ⇒ P (x)) ⇒ (P (x) ⇒ ∀x P (x))).

Observe that ∀x(P (x) ⇒ P (x)) is a predicate tautology. Take M = [N, I] for
I such that PI = ODD, where ODD ⊆ N is the set of odd numbers. Let
s : V AR −→ N . By definition if I, sI (x) ∈ PI if and only if sI (x) ∈ ODD.
Then obviously (M, s) 6|= ∀x P (x) and M = [N, I] is a counter model for (8.29)
as (M, s) |= ∀x(P (x) ⇒ P (x)).
The examples for (8.30), (8.31), and (8.29) similar.
Distributivity 2
For any formulas A(x), B(x) of L,

|= (∃x (A(x) ∩ B(x)) ⇒ (∃xA(x) ∩ ∃xB(x))), (8.32)

|= ((∀xA(x) ∪ ∀xB(x)) ⇒ ∀x (A(x) ∪ B(x))), (8.33)

|= (∀x(A(x) ⇒ B(x)) ⇒ (∀xA(x) ⇒ ∀xB(x))) (8.34)
The converse mplications to (8.32), (8.33), (8.34) are not a predicate tautologies
, i.e. there are formulas A(x), B(x), such that

6|= ((∃xA(x) ∩ ∃xB(x)) ⇒ ∃x (A(x) ∩ B(x))). (8.35)

6|= (∀x (A(x) ∪ B(x)) ⇒ (∀xA(x) ∪ ∀xB(x))), (8.36)

6|= ((∀xA(x) ⇒ ∀xB(x)) ⇒ ∀x(A(x) ⇒ B(x))) (8.37)
To prove (8.35), (8.36) we have to find particular formulas A(x), B(x) ∈ F and
counter models M = [U, I] for these particular cases.

392
Consider (8.35). We take as A(x), B(x) atomic formulas Q(x, c), P (x, c). The
particular case of (8.35) is now a formula

((∃xP (x, c) ∩ ∃xQ(x, c)) ⇒ ∃x (P (x, c) ∩ Q(x, c))).

Take M = [R, I] where R is the set of real numbers, and the interpretation I is
QI :>, PI :<, cI : 0. The particular case formula becomes an obviously false
mathematical statement

((∃x∈R x > 0 ∩ ∃x∈R x < 0) ⇒ ∃x∈R (x > 0 ∩ x < 0)).

Consider (8.36). We take as Let A(x), B(x) be atomic formulas Q(x, c), R(x, c).
The particular case of (8.36 ) is now a formula

(∀xQ(x, c) ∪ R(x, c)) ⇒ (∀xQ(x, c) ∪ ∀xR(x, c))).

Take M = (R, I) where R is the set of real numbers and QI :≥, RI :<, cI : 0.
The particular formula becomes an obviously false mathematical statement

(∀x∈R (x ≥ 0 ∪ x < 0) ⇒ (∀x∈R x ≥ 0 ∪ ∀x∈R x < 0)).

De Morgan
For any formulas A(x), B(x) of L,

|= (¬∀xA(x) ⇒ ∃x¬A(x)), (8.38)

|= (¬∃xA(x) ⇒ ∀x¬A(x)), (8.39)

|= (∃x¬A(x) ⇒ ¬∀xA(x)), (8.40)
|= (∀x¬A(x)) ⇒ ¬∃xA(x)). (8.41)

We prove (8.38) as an example.The proofs of all other laws are similar. As-
sume that (8.38) does not hold. By definition 8.16 there is M = (U, I) and
s : V AR −→ U, such that (M, s) |= ¬∀x¬A(x)) and (M, s) 6|= ∃x¬A(x).
Consider (M, s) |= ¬∀xA(x). By satisfaction definition 8.13, (M, s) 6|= ∀xA(x).
This holds only if for all s0 , such that s, s0 agree on all variables except on x,
(M, s0 ) 6|= A(x).
Consider (M, s) 6|= ∃x¬A(x). This holds only if there is no s0 , such that
(M, s0 ) |= ¬A(x), i.e. there is no s0 , such that (M, s0 ) 6|= A(x). This means
that for all s0 , (M, s0 ) |= A(x). Contradiction with (M, s0 ) 6|= A(x).
Quantifiers Alternations
For any formula A(x, y) of L,

|= (∃x∀yA(x, y) ⇒ ∀y∃xA(x, y)). (8.42)

393
The converse implications to (8.42) is not a predicate tautology. Take as A(x, y)
an atomic formulas R(x, y). Take M = (R, I) where R is the set of real numbers
and RI :< . The instance of (8.42) particular formula becomes a mathematical
statement
(∀y∃x(x < y) ⇒ ∃x∀y(x < y))
that obviously false in the set of real numbers. We proved

6|= (∀y∃xA(x, y) ⇒ ∃x∀yA(x, y)). (8.43)

8.3.1 Equational Laws of Quantifiers

The most frequently used laws of quantifiers have a form of a logical equivalence,
symbolically written as ≡. This not a new logical connective. This is a very
useful symbol. It has the same properties as the equality = and can be used in
the same way we use the equality symbol =.
Note that we use the same equivalence symbol ≡ and the tautology symbol
|= for propositional and predicate languages and semantics when there is no
confusion. Formally we define the predicate equivalence as follows.

Definition 8.23 (Logical equivalence)

For any formulas A, B ∈ F of the predicate language L,
A ≡ B if and only if |= (A ⇒ B) and |= (B ⇒ A).

Remark that our predicate language L = L{¬,∩,∪,⇒,¬} (P, F, C) we defined the

semantics for (definition 8.13 ) does not include the equivalence connective ⇔. If
it does we extend the satisfaction definition 8.13 in a natural way and can adopt
the following definition 8.24 of logical equivalence that is obviously equivalent
definition to the propositional one and to our definition 8.23.

Definition 8.24
For any formulas A, B ∈ F of the predicate language L,
A ≡ B if and only if |= (A ⇔ B).

We re-write the basic theorem 8.1 establishing relationship between proposi-

tional and some predicate tautologies as follows.

Theorem 8.3 (Tautologies)

If a formula A is a propositional tautology, then by substituting for propositional
variables in A any formula of the predicate language L we obtain a formula which
is a predicate tautology.

394
Directly from the theorem 8.3 and logical equivalence definition 8.23 we get that
the following is true.

Theorem 8.4 (Equivalences)

Given propositional formulas A, B.
If A ≡ B is a propositional equivalence, and A0 , B 0 are formulas of the predicate
language L obtained by a substitution of any formula of L for propositional vari-
ables in A and B, respectively, then A0 ≡ B 0 holds under predicate semantics.

Example 8.19
Consider the following propositional logical equivalence:

(a ⇒ b) ≡ (¬a ∪ b).

Substituting ∃xP (x, z) for a, and ∀yR(y, z) for b, we get from theorem 8.4 that
the following equivalence holds:

(∃xP (x, z) ⇒ ∀yR(y, z)) ≡ (¬∃xP (x, z) ∪ ∀yR(y, z)).

We prove in similar way as in the propositional case (chapter 3) the following.

Theorem 8.5 (Equivalence Substitution)

Let a formula B1 be obtained from a formula A1 by a substitution of a formula
B for one or more occurrences of a sub-formula A of A1 , what we denote as

B1 = A1 (A/B).

Then the following holds for any formulas A, A1 , B, B1 of L.

If A ≡ B, then A1 ≡ B1 . (8.44)

Directly from the Dictum de Omi (8.22) and the Generalization (??) tautologies
we get the proof of the following theorem 8.6 useful for building new logical
equivalences from the old, known ones.

Theorem 8.6
For any formulas A(x), B(x) of L.

if A(x) ≡ B(x), then ∀xA(x) ≡ ∀xB(x),

if A(x) ≡ B(x), then ∃xA(x) ≡ ∃xB(x).

395
Example 8.20
We know from the example 8.19 that the formulas (∃xP (x, z) ⇒ ∀yR(y, z))
and (¬∃xP (x, z) ∪ ∀yR(y, z)) are logically equivalent. We get, as the direct
consequence of the theorem 8.6 the following equivalences:

∀z(∃xP (x, z) ⇒ ∀yR(y, z)) ≡ ∀z(¬∃xP (x, z) ∪ ∀yR(y, z)),

Theorem 8.4 and theorem 8.6 allow us to use propositional tautologies and
predicate formulas to build predicate equivalences. Here is a simple example.

Exercise 8.6
Prove that for any formulas A(x), B(x) of L

¬∀x¬(A(x) ∪ B(x)) ≡ ¬∀x(¬A(x) ∩ ¬B(x)). (8.45)

Solution
By the substituting A(x) for a, and any formula B(x) for b, in the propositional
de Morgan Law: ¬(a ∪ b) ≡ (¬a ∩ ¬b), we get via theorem 8.4 that

¬(A(x) ∪ B(x)) ≡ (¬A(x) ∩ ¬B(x)).

Applying the theorem 8.6 to the above we obtain that

∀x¬(A(x) ∪ B(x)) ≡ ∀x(¬A(x) ∩ ¬B(x)).

We know, from the propositional logic, that for any propositional variables a, b,
a ≡ b if and only if ¬a ≡ ¬b. Substituting ∀x¬(A(x) ∪ B(x)) and ∀x(¬A(x) ∩
¬B(x)) for a and b, respectively, we get that

∀x¬(A(x) ∪ B(x)) ≡ ∀x(¬A(x) ∩ ¬B(x))

if and only if
¬∀x¬(A(x) ∪ B(x)) ≡ ¬∀x(¬A(x) ∩ ¬B).
But we have proved that ∀x¬(A(x)∪B) ≡ ∀x(¬A(x)∩¬B) holds, so we conclude
that the equivalence (8.45) also holds.

Exercise 8.7
Prove that for any formulas A(x), B of L

∀x¬(A(x) ∪ B) ≡ ∀x(¬A(x) ∩ ¬B)

Solution
By the substituting A(x) for a, and any formula B for b, in the propositional
de Morgan law: ¬(a ∪ b) ≡ (¬a ∩ ¬b), we get that

¬(A(x) ∪ B) ≡ (¬A(x) ∩ ¬B).

396
Applying the theorem 8.4 to the above we obtain that
∀x¬(A(x) ∪ B) ≡ ∀x(¬A(x) ∩ ¬B).
As we can see, it is possible to obtain a fair amount of predicate tautologies
from the propositional tautologies and theorems 8.3, 8.4 and 8.6, but as we
have proved will for never obtain for example the most basic law: (∀xA(x) ⇒
∃xA(x)), any many the most important others.
We concentrate now only on these laws which have a form of a logical equiva-
lence.They are called the equational laws for quantifiers.
Directly from the definition 8.23 and the de Morgan tautologies (8.38)-(8.41)
we get one of the most important equational laws, called also De Morgan Laws.
De Morgan Laws
¬∀xA(x) ≡ ∃x¬A(x) (8.46)
¬∃xA(x) ≡ ∀x¬A(x) (8.47)

Now we will apply them to show that the quantifiers can be defined one by the
other i.e. that the following Definability Laws hold.
Definability Laws
∀xA(x) ≡ ¬∃x¬A(x) (8.48)
∃xA(x) ≡ ¬∀x¬A(x) (8.49)
The law (8.48) is often used as a definition of the universal quantifier in terms
of the existential one (and negation), the law (8.49) as a definition of the
existential quantifier in terms of the universal one (and negation).
Proof of (8.48)
Substituting any formula A(x) for a variable a in the propositional equivalence
a 𠪪a we get by theorem 8.4 that A(x) 𠪪A(x). Applying the theorem 8.6
to the above we obtain ∃xA(x) ≡ ∃x¬¬A(x). By the de Morgan Law (8.46)
∃x¬¬A(x) ≡ ¬∀x¬A(x) and hence ∃xA(x) ≡ ¬∀x¬A(x), what ends the proof.
Proof of (8.49)
We obtain ∀xA(x) ≡ ∀¬¬A(x) in a similar way as above. By the de Morgan
Law (8.47), ∀¬¬A(x) ≡ ¬∃¬A(x) and hence ∀xA(x) ≡ ¬∃¬A(x), what ends
the proof.
Other important equational laws are the following introduction and elimination
laws. We prove later the first two of them. We show that the laws (11.44) -
(10.39) can be deduced from laws (10.32) and (11.43), the de Morgan laws (8.46),
(8.47), definability laws (8.48), (8.49), propositional tautologies and theorems
8.3, 8.4, and theorem 8.5.
Introduction and Elimination Laws
If B is a formula such that B does not contain any free occurrence of x,
then the following logical equivalences hold.

397
 ∀x(A(x) ∪ B) ≡ (∀xA(x) ∪ B) (8.50)
∀x(A(x) ∩ B) ≡ (∀xA(x) ∩ B) (8.51)
∃x(A(x) ∪ B) ≡ (∃xA(x) ∪ B) (8.52)
∃x(A(x) ∩ B) ≡ (∃xA(x) ∩ B) (8.53)
∀x(A(x) ⇒ B) ≡ (∃xA(x) ⇒ B) (8.54)
∃x(A(x) ⇒ B) ≡ (∀xA(x) ⇒ B) (8.55)
∀x(B ⇒ A(x)) ≡ (B ⇒ ∀xA(x)) (8.56)
∃x(B ⇒ A(x)) ≡ (B ⇒ ∃xA(x)) (8.57)

The equivalences (10.32)-(11.45) make it possible to introduce a quantifier that

precedes a disjunction or a conjunction into one component on the condition
that the other component does not contain any free occurrence of a variable
which is bound by that quantifier. These equivalences also make possible to
eliminate a quantifier from a component of a disjunction or a conjunction and
to place it before that disjunction or conjunction as a whole, on the condition
that the other component does not contain any free occurrence of a variable
which that quantifier would then bind.
The equivalences (11.46)-(10.39 )make it possible to introduce a quantifier pre-
ceding an implication into the consequent of that implication, on the condition
that that antecedent does not contain any free occurrence of a variable which
is bound by that quantifier; they also make it possible to introduce a univer-
sal quantifier preceding an implication into the consequent of that implication
while changing it into an existential quantifier in the process,on the condition
that the consequent of that implication does not contain any free occurrence of
a variable bound by that quantifier. Equivalences (11.46)-(10.39) further enable
us to eliminate quantifiers from the antecedent of an implication to the position
preceding the whole implication, while changing a universal quantifier into an
existential one, and vice versa, in the process, and also to eliminate quantifiers
from the consequent of an implication to the position preceding the whole impli-
cation; the conditions that the other component of the implication in question
does not contain any free occurrence of a variable which that quantifier would
then bind, must be satisfied, respectively.
As we said before, the equivalences (10.32)-(10.39) are not independent, some
of them are the consequences of the others. Assuming that we have already
proved (10.32) and (11.43), the proof of (11.44) is as follows.
Proof of (11.44) ∃x(A(x) ∪ B) is logically equivalent, by the definability law
(8.49) to ¬∀x¬(A(x) ∪ B). By the reasoning presented in the proof of (8.45 ) for
B instead of B(x), we have that ¬∀x¬(A(x) ∪ B) ≡ ¬∀x(¬A(x) ∩ ¬B). By the
introduction law (11.43), ¬∀x(¬A(x) ∩ ¬B) ≡ ¬(∀x¬A(x) ∩ ¬B). Substituting
∀x¬A(x) for a and ¬B for b in propositional equivalence ¬(a∩¬b) ≡ (¬a∪¬¬b),

398
we get, by the theorem 8.4 that ¬(∀x¬A(x) ∩ ¬B) ≡ (¬∀x¬A(x) ∪ ¬¬B). In a
similar way we prove that ¬¬B ≡ B, by the definability law (8.49) ¬∀x¬A(x) ≡
∃xA(x), hence by theorem 8.5 ¬∀x¬A(x) ∪ ¬¬B ≡ (∃xA(x) ∪ B) and finally,
∃x(A(x) ∪ B) ≡ (∃xA(x) ∪ B), what end the proof.
We can write this proof in a shorter, symbolic way as follows:
law 8.49
∃x(A(x) ∪ B) ≡ ¬∀x¬(A(x) ∪ B)
thm 8.3, 8.4
≡ ¬∀x(¬A(x) ∩ ¬B)
law 11.43
≡ ¬(∀x¬A(x) ∩ ¬B)
(8.46), thm 8.5
≡ (¬∀x¬A(x) ∪ ¬¬B)
thm 8.5
≡ (∃xA(x) ∪ B)

Distributivity Laws
Let A(x), B(x) be any formulas with a free variable x.
Law of distributivity of universal quantifier over conjunction

∀x (A(x) ∩ B(x)) ≡ (∀xA(x) ∩ ∀xB(x)) (8.58)

Law of distributivity of existential quantifier over disjunction.

∃x (A(x) ∪ B(x)) ≡ (∃xA(x) ∪ ∃xB(x)) (8.59)

Alternations of Quantifiers Laws

Let A(x, y) be any formula with a free variables x,y.

∀x∀y (A(x, y) ≡ ∀y∀x (A(x, y) (8.60)

∃x∃y (A(x, y) ≡ ∃y∃x (A(x, y) (8.61)

Renaming the Variables

Let A(x) be any formula with a free variable x and let y be a variable that does
not occur in A(x).
Let A(x/y) be a result of replacement of each occurrence of x by y, then the
following holds.
∀xA(x) ≡ ∀yA(y), (8.62)
∃xA(x) ≡ ∃yA(y). (8.63)

Restricted De Morgan Laws

For any formulas A(x), B(x) ∈ F with a free variable x,

¬∀B(x) A(x) ≡ ∃B(x) ¬A(x), ¬∃B(x) A(x) ≡ ∀B(x) ¬A(x). (8.64)

399
Here is a poof of first equality. The proof of the second one is similar and is left
as an exercise.

¬∀B(x) A(x) ≡ ¬∀x (B(x) ⇒ A(x)) ≡ ¬∀x (¬B(x)∪A(x)) ≡ ∃x ¬(¬B(x)∪A(x))

≡ ∃x (¬¬B(x) ∩ ¬A(x)) ≡ ∃x (B(x) ∩ ¬A(x)) ≡ ∃B(x) ¬A(x)).

Restricted Introduction and Elimination Laws

If B is a formula such that B does not contain any free occurrence of x, then
the following logical equivalences hold for any formulas A(x), B(x), C(x).

∀C(x) (A(x) ∪ B) ≡ (∀C(x) A(x) ∪ B), (8.65)

∃C(x) (A(x) ∩ B) ≡ (∃C(x) A(x) ∩ B), (8.66)

∀C(x) (A(x) ⇒ B) ≡ (∃C(x) A(x) ⇒ B), (8.67)
∀C(x) (B ⇒ A(x)) ≡ (B ⇒ ∀C(x) A(x)). (8.68)
The proofs are similar to the proof of the restricted de Morgan Laws.
The similar generalization of the other Introduction and Elimination Laws (11.43),
(11.44), (11.47), (10.39) for restricted domain quantifiers fails. We can easily
follow the proof of (8.18) and construct proper counter-models proving the fol-
lowing.
∃C(x) (A(x) ∪ B) 6≡ (∃C(x) A(x) ∪ B),
∀C(x) (A(x) ∩ B) 6≡ (∀C(x) A(x) ∩ B),
∃C(x) (A(x) ⇒ B) 6≡ (∀C(x) A(x) ⇒ B),
∃C(x) (B ⇒ A(x)) 6≡ (B ⇒ ∃xA(x)).

Nevertheless it is possible to correctly generalize them all as to cover quantifiers

with restricted domain. We show it in a case of (11.43) and leave the other
cases to the reader as an exercise.

Example 8.21
The restricted quantifiers version of (11.43) is the following.

∃C(x) (A(x) ∪ B) ≡ (∃C(x) A(x) ∪ (∃x C(x) ∩ B)). (8.69)

We derive (8.74) as follows.

∃C(x) (A(x) ∪ B) ≡ ∃x(C(x) ∩ (A(x) ∪ B)) ≡ ∃x((C(x) ∩ A(x)) ∪ (C(x) ∩ B))

≡ (∃x(C(x) ∩ A(x)) ∪ ∃x(C(x) ∩ B)) ≡ (∃C(x) A(x) ∪ (∃x C(x) ∩ B)).

We leave it as an exercise to specify and write references to transformation or

equational laws used at each step of our computation.

400
8.4 Hilbert Proof Systems Soundness and Com-
pleteness
We adopt now general definition from chapter 4 concerning proof systems to
the case of classical first order (predicate) logic.
We refer the reader to chapters 4 and 5 for a great array of example, exercises,
homework problems explaining in a great detail all notions we introduce here
for the predicate case. The examples and exercises we provide here are not
numerous and restricted to the laws of quantifiers.
Given a language L = L{¬,∩,∪,⇒,¬} (P, F, C). Any proof system

S = (L = L{¬,∩,∪,⇒,¬} (P, F, C), F, LA, R) (8.70)

is a predicate (first order) proof system.

The predicate proof system (11.1) is a Hilbert proof system if the set R of its
rules contains the Modus Ponens rule
A ; (A ⇒ B)
(M P ) ,
B
where A, B ∈ F.

Semantic Link: Logical Axioms LA

We want the set LA of logical axioms to be a non-empty set of classical predicate

tautologies (8.13), i.e.
LA ⊆ Tp ,
where
Tp = {A of L{¬,∩,∪,⇒,¬} (P, F, C) : |=p A}.

Remark 8.2
We use symbols |=p , Tp to stress the fact that we talk about predicate lah=nguage
and classical predicate tautologies.

Semantic Link 2: Rules of Inference R

We want the the rules of inference r ∈ R of S to preserve truthfulness. Rules

that do so are called sound. We define it formally as follows.

Definition 8.25 (Sound Rule)

401
Given an inference rule r ∈ R of the form

P1 ; P2 ; .... ; Pm
(r) ,
C
where P1 .P2 , . . . , Pm , C ∈ F.
(i) We say that the rule (r) is sound if and only if the following condition
holds for all structures M = [U, I] for L.

If M |= {P1 , P2 , .Pm } then M |= C. (8.71)

(ii) The rule (r) is not sound if and only if there is a structure M = [U, I],
such that
M |= {P1 , P2 , .Pm } and M |6 = C. (8.72)

In order to prove that the rule (r) is sound we have to show the implication
(8.71). It means, by definitions 9.3, 8.18, we have to show that that if all pre-
misses of the rule (r) are true in M = [U, I], so is its conclusion. This also
justifies correctness of the definition 8.25; sound rules do preserve the truth-
fulness as it is defined in our semantics.

Exercise 8.8
Prove the soundness of the rule

¬∀xA(x)
(r1) . (8.73)
∃x¬A(x)

Proof
Assume that the soundness condition (8.71) does not hold for for all structures
M = [U, I]. It means we assume that there is a structure M = [U, I], such that
M |= ¬∀xA(x) and M 6|= ∃x¬A(x).
Let M |= ¬∀xA(x). By definition 9.3, for all s : V AR −→ U we have
(M, s) |= ¬∀x¬A(x)). Hence by satisfaction definition 8.13, (M, s) 6|= ∀xA(x).
This holds only if for all s0 , such that s, s0 agree on all variables except on x,
(M, s0 ) 6|= A(x).
Observe that (M, s) 6|= ∃x¬A(x) only if there is no s0 , such that (M, s0 ) |=
¬A(x), i.e. there is no s0 , such that (M, s0 ) 6|= A(x). This means that for all s0 ,
(M, s0 ) |= A(x). Contradiction with (M, s0 ) 6|= A(x).

Exercise 8.9
Prove the soundness of the rule

∀xA(x)
(r2) . (8.74)
∃xA(x)

402
Proof
Assume that the soundness condition (8.71) does not hold for for all structures
M = [U, I]. It means we assume that there is a structure M = [U, I], such that
M |= ∀xA(x) and M 6|= ∃xA(x).
Let M |= ∀xA(x). By definition 9.3, for all s : V AR −→ U we have (M, s) |=
∀x¬A(x)).
By definition 8.13, (M, s) |= ∀x A(x) and (M, s) 6|= ∃x A(x). It means that
(M, s0 ) |= A(x) for all s0 such that s, s0 agree on all variables except on x, and
it is not true that there is s0 such that s, s0 agree on all variables except on x,
and (M, s0 ) |= A(x). This is impossible and this contradiction proves soundness
of (r2).

Exercise 8.10
Prove that the rule
∃xA(x)
(r3) . (8.75)
∀xA(x)
is not sound.

Proof
Observe that to prove that the rule (8.75) is not sound we have to provide an
example of an instance of a formula A(x) and construct prove (ii) of definition
8.25 for it.
Let A(x) be an atomic formula P (x, c), for any P ∈ P, #P = 2. We take as
M = (N, PI :<, cI : 3) for N set of natural numbers. Let s be any assignment
s : V AR −→ N. Obviously (M, s) |= ∃x P (x, c).
Take any s0 such that s0 (x) = 2 and s0 (y) = s(y) for all y ∈ V AR − {x}. We
have (2, 3) ∈ PI , as 2 < 3 and hence there exists s0 that agrees with s on all
variables except on x, and (M, s0 ) |= P (x, c). But (M, s) 6|= ∀x P (x, c) as for
example for s0 such that s0 (x) = 5 and s0 (y) = s(y) for all y ∈ V AR − {x},
(2, 3) 6∈ PI , as 5 6< 3.
This proves that M = (N, PI :<, cI : 3) is a model for (∃x P (x, c) and hence
6|= ∀x A(x)).
The ”shorthand” solution is: the formula (∃x P (x, c) becomes in M = (N, PI :<
, cI : 3) a true mathematical statement (written with logical symbols): ∃n n <
3. The formula (∀x P (x, c) becomes a mathematical frmula ∀n n < 3 which is
an obviously false statement in the set N of natural numbers, as there is n ∈ N ,
such that n < 3, for example n = 2, and it is not true that all natural numbers
are smaller then 3. So the rule (r3) is not sound.

Definition 8.26 (Strongly Sound Rule)

403
An inference rule r ∈ R of the form
P1 ; P2 ; .... ; Pm
(r)
C
is strongly sound if the following condition holds for all structures M = [U, I]
for L.

M |= {P1 , P2 , .Pm } if and only if M |=C. (8.76)

We can, and we do state it informally as: ” an inference rule r ∈ R is strongly

sound when the conjunction of all its premisses is logically equivalent to its
conclusion”. We denote it informally as

P1 ∩ P2 ∩ . . . ∩ Pm ≡ C. (8.77)

Example 8.22
The sound rule (8.73)
¬∀xA(x)
(r1)
∃x¬A(x)
is strongly sound by De Morgan Law (8.46).
The sound rule (8.75)
∀xA(x)
(r2)
∃xA(x)
is not strongly sound by exercise 8.10.

Definition 8.27 (Sound Proof System)

Given the predicate (first order) proof system (11.1)

S = (L, F, LA, R).

We say that the proof system S is sound if the following conditions hold.
(1) LA ⊆ Tp ;
(2) Each rule of inference r ∈ R is sound.
The proof system S is strongly sound if the condition (2) is replaced by the
following condition (2’)
(2’) Each rule of inference r ∈ R is strongly sound under M.

The set of all provable expressions of S is denoted by PS and is defined as

follows.
PS = {A ∈ F : `S A}. (8.78)

404
When we define (develop) a proof system S our first goal is to make sure that it
a ”sound” one, i.e. that all we prove in it is true. Proving the following theorem
establishes this goal.

Theorem 8.7 (Soundness Theorem for S)

Given a predicate proof system S.
For any A ∈ F, the following implication holds.

If `S A then |=p A. (8.79)

We write (8.79) it in a more concise form as

PS ⊆ Tp . (8.80)

Proof
Observe that if we have already proven that S is sound as stated in the defi-
nition 8.27, the proof of the implication (8.79) is straightforward mathematical
induction over the length of a proof.
It means that in order to prove the Soundness Theorem 8.7 for a proof system
Sit is enought to verify the two conditions of the definition 8.27 (1) LA ⊆ Tp
and (2) each rule of inference r ∈ R is sound.

We again refer the reader to chapter 4 for detailed examples, exercises and
problems.
As we can see, proving Soundness Theorem 8.7 for any proof system we develop
is indispensable and the proof is quite easy. The next step in developing a logic
(classical predicate logic in our case now) is to answer necessary and a difficult
question: Given a proof system S, about which we know that all it proves it
true (tautology). Can we prove all we know to be true (all tautologies)?
Proving the following theorem establishes this goal.

Theorem 8.8 (Completeness Theorem for S)

Given a predicate proof system S.
For any A ∈ F, the following implication holds.

`S A if and only if |=p A. (8.81)

We write (8.81) it in a more concise form as

PS = Tp . (8.82)

The Completeness Theorem consists of two parts:

Part 1: Soundness Theorem: PS ⊆ Tp .

405
Part 2: Completeness part of the Completeness Theorem: Tp ⊆ PS .
Proving the Soundness Theorem for S is usually a straightforward and not a very
difficult task. Proving the Completeness part of the Completeness Theorem is
always a crucial and very difficult task. There are many methods and techniques
for doing so, even for classical proof systems (logics) alone. Non-classical logics
often require new sometimes very sophisticated methods. We presented two
proofs of the Completeness Theorem for classical propositional Hilbert style
proof system in chapter 5, and a constructive proofs for automated theorem
proving systems for classical propositional logic the chapter 6.
We present a proof of the Completeness Theorem for predicate (first order) logic
in the next chapter 9.

8.5 Homework Problems

Predicate Languages

1. Given the following formulas A1 − A5 of a predicate language L.

A1 = R(x, y, g(c, x)), A2 = ∃xP (x, f (x, y)), A3 = ∃dR(x, y, g(c, d)),

A4 = ∀z(f (x, P (c, y)), A5 = ∃yP (x, f (c, y)) ∪ ∀yP (x, f (c, y)).

(a) Indicate whether they are, or are not well formed formulas of F. For
those which are not in F write a correct formula.
(b) For each correct, or corrected formula identify all components: con-
nectives, quantifiers, predicate and function symbols, and list all its
terms.
(c) For each formula identify its s free and bound variables. State which
are open and which are closed formulas (sentences), if any.
(d) Describe a language defined by the set F0 = {A1 , A2 , . . . A5 } of
formulas that are correct or corrected.

2. For the following mathematical statements write their corresponding for-

mulas of predicate language L.

(a) ∀n>1 (n + 3 < 8 ∪ ∃x∈R x + n > 8)

(b) ∀x∈R ∃n∈N (x + n > 0 ⇒ ∃m∈N (m = x + n))
(c) If all natural numbers are smaller then zero, then the sum of any two
integers is smaller then zero.
(d) For all natural numbers The following implication holds for all natural
numbers: if n > 0, then there is a real number x, such that n + x = 0
or there is an integer m, such that m > 0.

406
3. For each of the following formulas (some with restricted quantifiers) write
2 corresponding natural language sentences.

(a) ∀x(P (x) ⇒ ∃yQ(x, y)).

(b) ∀x∃y(P (x) ∩ ¬Q(x, y)).
(c) ∀A(x) ∃A(y) B(y).
(d) ∃P (x) ∀N (x) R(x, y).

4. Is the term t = f (x, y) free for x in the following formulas?

(a) (P (x, y) ⇒ ∀yP (x, y).

(b) (∀yP (x, y) ∪ ∃yP (x, y)).
(c) ∀xP (x, y).
(d) ∀yP (x, y).
(e) (∀yQ(y) ⇒ P (x, y)).

5. Justify that for any formula A ∈ F and any term t ∈ T the following facts
hold.

(a) A closed tern t, i.e. term with no variables is free for any variable x
in A.
(b) A term t is free for any variable in A if none of the variables in t is
bound in A.
(c) Term t = x is free for x in any formula A.
(d) Any term is free for x in A if A contains no free occurrences of x.

6. Translate the following formulas in everyday English.

(a) ∀x(P (x) ∩ ∀y(¬L(x, y) ⇒ ¬H(x))), where P (x) means ” x is a per-

son”, L(x, y)means ”x likes y” , and H(x) means ”x is happy”.
(b) ∀x((E(x) ∩ P (x)) ⇒ E(x, b)), where E(x) means ” x is an even inte-
ger”, P(x) means ” x is prime”, Q(x,y) means ” x is equal to x”, and
b denotes 2.
(c) ¬∀y((P (y)∩∀x(P (x)) ⇒ E(x, y))), where P(x) means ”x is an integer,
and L(x, y)means ”x ≤ y”.

7. Use the restricted quantifiers to translate the following natural language

sentences into a proper formulas of a proper formal predicate language
L = L{¬,∩,∪,⇒,¬} (P, F, C). In each case specify the sets P, F, C.

(a) Some politician are honest, some are not.

(b) Any sets that have the same elements are equal.
(c) Somebody hates everyone who does not hate himself.

407
 (d) Birds can fly and if anyone can fly Tweety can.
(e) Anyone who knows logic loves it.

Classical Semantics

1. Given a predicate language L(P, F, C) and a structure M = [U, I] such

that U = N and PI : =, fI : +, gI : ·, aI : 0, bI : 1 for N set on natural
numbers. For each of the following formula A decide whether M |= A
or not. Do so by examining the corresponding mathematical statement
defined by M.
(a) ∀x∃y(P (x, f (y, y)) ∪ P (x, f (f (y, y), b))).
(b) ∀x∃y(P (g(x, y), a) ⇒ (P (x, a) ∪ P (y, a))).
(c) ∃yP (f (y, y), b).
2. Let M = [U, I] be a structure such that U = Z and PI : =, fI : + for Z
set of integers. For each of the following formula A decide whether M |= A
or not. Do so by examining the corresponding mathematical statement
defined by M.
(a) ∀x∀yP (f (x, y), f (y, x)).
(b) ∀x∀y∀zP (f (x, f (y, z)), f (f (x, y), z)).
(c) ∀x∀y∃zP (f (x, z), y)
3. Let M = [U, I] be a structure such that U = N − {0} or N set of natural
numbers and PI : =, fI (x, y) is xy For each of the following formula A
decide whether M |= A or not. Do so by examining the corresponding
mathematical statement defined by M.
(a) ∀x∀yP (f (x, y), f (y, x)).
(b) ∀x∀y∀zP (f (x, f (y, z)), f (f (x, y), z)).
(c) ∀x∀y∃zP (f (x, z), y).
4. For each formula below, where P, Q ∈ P, find a structure M = [U, I] for
L = L{¬,∩,∪,⇒,¬} (P, F, C) that is its counter model. Justify its correct-
ness.

(a) ( (∀xP (x) ⇒ ∀yQ(y)) ⇒ (∀x(P (x) ⇒ Q(x)))).

(b) ( (∀xP (x) ∪ ∀yQ(y)) ⇒ (∀x(P (x) ∪ ∀xQ(x)))).

5. Show that the following formulas are predicate tautologies for any formulas
A, B in L.
(a) (∀x∀yA(x, y) ⇒ ∀y∀xA(x, y)).
(b) (∃x∃yA(x, y) ⇒ ∃y∃xA(x, y)).

408
 (c) (∀x(A(x) ⇒ B(x)) ⇒ (∀xA(x) ⇒ ∀xB(x))).
6. Prove that the following formulas are not predicate tautologies by finding
their proper instances and constructing counter models for them.

(a) ((∃xA(x) ∩ ∃xB(x)) ⇒ ∃x (A(x) ∩ B(x))).

(b) (∀x (A(x) ∪ B(x)) ⇒ (∀xA(x) ∪ ∀xB(x))).
(c) ((∀xA(x) ⇒ ∀xB(x)) ⇒ ∀x(A(x) ⇒ B(x))).

7. Prove that the following formulas are predicate tautologies for any for-
mulas A(x), B(x), A, B of L, such that A , B does not contain any free
occurrences of x.

(a) (∀x(A ⇒ B(x)) ⇒ (A ⇒ ∀x B(x))),

(b) (∃x(A(x) ⇒ B) ⇒ (∀xA(x) ⇒ B)).
(c) (∀x(A(x) ⇒ B) ⇒ (∃xA(x) ⇒ B)).
8. Prove that the restrictions: ”A , B does not contain any free occurrences
of x” are essential for all of the following tautologies, i.e. give examples of
formulas for which the laws without these restrictions fail and construct
counter models for them.

(a) (∀x(A(x) ⇒ B) ⇒ (∃xA(x) ⇒ B)).

(b) (∃x(A(x) ⇒ B) ⇒ (∀xA(x) ⇒ B)).

9. Prove that the converse implication to the formulas listed below are pred-
icate tautologies for any formulas A(x), B(x), A, B of L, such that A, B
does not contain any free occurrences of x.

(a) (∀x(A ⇒ B(x)) ⇒ (A ⇒ ∀x B(x))),

(b) (∃x(A(x) ⇒ B) ⇒ (∀xA(x) ⇒ B)).
(c) (∀x(A(x) ⇒ B) ⇒ (∃xA(x) ⇒ B)).

409
410
Chapter 9

Hilbert Proof Systems

Completeness of Classical
Predicate Logic

There are several quite distinct approaches to the Completeness Theorem, cor-
responding to the ways of thinking about proofs. Within each of the approaches
there are endless variations in exact formulation, corresponding to the choice of
methods we want to use to proof the Completeness Theorem. Different basic
approaches are important, though, for they lead to different applications. We
have presented two of the approaches for the propositional logic: Hilbert style
formalizations (proof systems) in chapter 5, and Gentzen style formalizations
(automated proof systems) in chapter 6. We have also presented for each of the
approaches methods of proving the completeness theorem. Two proofs of com-
pleteness theorem for Hilbert style proof system in chapter 5 and a constructive
proofs for several Gentzen style proof systems in chapter 6.
There are many proofs of the Completeness Theorem for predicate (first order)
logic. We present here in a great detail, a version of Henkin’s proof as included
in a classic Handbook of Mathematical Logic (1977). It contains a method for re-
ducing certain problems of first-order logic back to problems about propositional
logic. We give independent proof of Compactness Theorem 9.1 for propositional
logic. Reduction to Propositional Logic Theorem 9.2, Compactness Theorem 9.3
for first-order logic, Löwenheim-Skolem Theorem 9.4 and Gödel Completeness
Theorem 9.7 fall out of the Henkin method.
We choose this particular proof of completeness of first order logic not only for
it being one of the oldest and most classical, but also for its connection with
the propositional logic. Moreover, the proof of the Compactness Theorem 9.1 is
based on semantical version of syntactical notions and techniques crucial to the

411
second proof of completeness theorem for propositional logic covered in chapter
5 and hence is familiar to the reader.

9.1 Reduction Predicate Logic to Propositional

Logic
Let L = L(P, F, C) be a first order language with equality (definition 9.12). We
assume that the sets P, F, C are infinitely enumerable. We also assume that it
has a full set of propositional connectives, i.e.

L = L{¬,∩,∪,⇒} (P, F, C).

Our goal now is to define a propositional logic within L = L(P, F, C). We

do it in a sequence of steps.
First we define a special subset P F of formulas of L, called a set of all propo-
sitional formulas of L.
Intuitively these are formulas of L which are not direct propositional combina-
tion of simpler formulas, that are atomic formulas or formulas beginning with
quantifiers. Formally, we have the following.

Definition 9.1 (Prime Formulas)

Any formula from that set P defined by (9.1) is called a prime formula of L.

P = AF ∪ {∀xB, B ∈ F} ∪ {∀xB : B ∈ F}, (9.1)

where the set AF is the set of all atomic formulas of L.

The set P ⊆ F, called a set of all prime formulas of L plays in the propositional
logic we define the role

Example 9.1
The following are primitive formulas.
R(t1 , t2 ), ∀x(A(x) ⇒ ¬A(x)), (c = c), ∃x(Q(x, y) ∩ ∀yA(y)).
The following are not primitive formulas.
(R(t1 , t2 ) ⇒ (c = c)), (R(t1 , t2 ) ∪ ∀x(A(x) ⇒ ¬A(x)).

Given a set P of primitive formulas we define in a standard way the set P F of

propositional formulas of L. as follows.

Definition 9.2 (Propositional Formulas of L)

412
Let F, P be sets of all formulas and prime formulas (9.1) of L, respectively.
The smallest set P F ⊆ F such that
(i) P ⊆ P F,
(ii)If A, B ∈ P F, then (A ⇒ B), (A ∪ B), (A ∩ B), and ¬A ∈ P F
is called a set of all propositional formulas of the predicate language L.
The set P is called atomic propositional formulas of L.

Propositional Semantics for L

We define propositional semantics for propositional formulas in P F as follows.

Definition 9.3 (Truth assignment)

Let P be a set of atomic propositional formulas of L and {T, F } be the set
of logical values ”true” and ”false”. Any function

v : P −→ {T, F }

is called a truth assignment in L.

We extend v to the set P F of all propositional formulas of L by defining the

mapping
v ∗ : P F −→ {T, F }
as follows.
v ∗ (A) = v(A) for A ∈ P,
and for any A, B ∈ P F,
v ∗ (A ⇒ B) = v ∗ (A) ⇒ v ∗ (B),
v ∗ (A ∪ B) = v ∗ (A) ∪ v ∗ (B),
v ∗ (A ∩ B) = v ∗ (A) ∩ v ∗ (B),
v ∗ (¬A) = ¬v ∗ (A).

Definition 9.4 (Propositional Model)

A truth assignment v : P −→ {T, F } is called a propositional model for a
formula A ∈ P F if and only if v ∗ (A) = T .

Definition 9.5 (Propositional Tautology)

For any formula A ∈ P F,
A ∈ P F is a propositional tautology of L if and only if
v ∗ (A) = T for all v : P −→ {T, F }.

413
For the sake of simplicity we will often say model, tautology instead propositional
model, propositional tautology for L.

Definition 9.6 (Model for the Set)

Given a set S of propositional formulas. We say that v is a model for the set
S if and only if v is a model for all formulas A ∈ S.

Definition 9.7 (Consistent Set)

A set S ⊆ P F of propositional formulas of L is consistent (in a sense of
propositional logic) if it has a (propositional) model.

Definition 9.8 (Inconsistent Set)

A set S ⊆ P F of propositional formulas of L is inconsistent if it has no model.

Theorem 9.1 (Compactness Theorem for Propositional Logic of L)

A set S ⊆ P F of propositional formulas of L is consistent if and only if every
finite subset of S is consistent.

Proof
Assume that S is a consistent set. By definition 9.7, it has a model. Tts model
is also a model for all its subsets, including all finite subsets, and so all its finite
subsets are consistent.
To prove the converse implication, i.e. the nontrivial half of the Compactness
Theorem we write it in a slightly modified form. To do so, we introduce the
following definition.

Definition 9.9 (Finitely Consistent Set (FC))

Any set S such that all its finite subsets are consistent is called finitely consistent.

We use this definition 9.9 to re-write the Compactness Theorem as follows.

A set S of propositional formulas of L is consistent if and only if it is finitely
consistent.
The nontrivial half of it still to be proved is now stated now as
Every finitely consistent set of propositional formulas of L is consistent.
The proof of the nontrivial half of the Compactness Theorem 9.1, as stated
above, consists of the following four steps.
S1 We introduce the notion of a maximal finitely consistent set.
S2 We show that every maximal finitely consistent set is consistent by con-
structing its model.

414
S3 We show that every finitely consistent set S can be extended to a maximal
finitely consistent set S ∗ . I.e we show that for every finitely consistent set S
there is a set S ∗ , such that S ⊆ S ∗ and S ∗ is maximal finitely consistent.
S4 We use S2 and S3 to justify the following reasoning.
Given a finitely consistent set S. We extend it, via construction to be defined in
the step S3 to a maximal finitely consistent set S ∗ . By the S2, S ∗ is consistent
and hence so is the set S.
This ends the proof of the Compactness Theorem 9.1.
Here are the details and proofs needed for completion of steps S1 - S4.
Step S1
We introduce the following definition.

Definition 9.10 (Maximal Finitely Consistent Set (MFC))

Any set S ⊆ P F is maximal finitely consistent if it is finitely consistent and for
every formula A, either A ∈ S or ¬A ∈ S .

We use notation MFC for maximal finitely consistent set, and FC for the
finitely consistent set.
Step S2
We prove the following MFC lemma 6.3 and the Property 9.1.

Lemma 9.1
Any MFC set is consistent.

Proof
Given a MFC set denoted by S ∗ . We prove its consistency by constructing
model for it, i.e. by constructing a truth assignment v : P −→ {T, F }, such
that for all A ∈ S ∗ , v ∗ (A) = T .
Observe that directly from the definition 9.10 we have the following property of
the the MFC sets.

Property 9.1 (MFC)

For any MFC set S ∗ and for every A ∈ P F, exactly one of the formulas A, ¬A
belongs to S ∗ .

In particular, for any P ∈ P F, we have that exactly one of formulas P, ¬P

belongs to S ∗ . This justifies the correctness of the following definition.

Definition 9.11
For any MFC set S ∗ , mapping v : P −→ {T, F }, such that

415
 if P ∈ S ∗

T
v(P ) =
F 6 S∗
if P ∈
is called a truth assignment defined by S ∗ .
We extend v to v ∗ : P F −→ {T, F } in a usual way.

We prove now that the truth assignment v defined by S ∗ (definition 9.11) is a

model for S ∗ , we show for any A ∈ P F,

if A ∈ S ∗

T
v ∗ (A) =
F 6 S∗
if A ∈

We prove it by induction on the degree of the formula A as follows.

The base case of A ∈ P follows immediately from the definition of v.

Case A = ¬C Assume that A ∈ S ∗ . This means ¬C ∈ S ∗ and by MCF

Property we have that C 6∈ S ∗ . So by the inductive assumption v ∗ (C) =
F and v ∗ (A) = v ∗ (¬C) = ¬v ∗ (C) = ¬F = T.
Assume now that A 6∈ S ∗ . By MCF Property ?? we have that C ∈ S ∗ .
By the inductive assumption v ∗ (C) = T and v ∗ (A) = v ∗ (¬C) = ¬v ∗ (T ) =
¬T = F.

This proves that for any formula A,

if ¬A ∈ S ∗

∗ T
v (¬A) =
F 6 S∗
if ¬A ∈

Case A = (B ∪ C) Let (B ∪ C) ∈ S ∗ . It is enough to prove that in this

case B ∈ S and C ∈ S ∗ , because then from the inductive assumption
∗

v ∗ (C) = v ∗ (D) = T and v ∗ (B ∪ C) = v ∗ (B) ∪ v ∗ (C) = T ∪ T = T .

Assume that (B ∪ C) ∈ S ∗ , B 6∈ S ∗ and C 6∈ S ∗ . Then byMCF Prop-
erty ?? we have that ¬B ∈ S ∗ , ¬C ∈ S ∗ and consequently the set

{(B ∪ C), ¬B, ¬C}

is a finite inconsistent subset of S ∗ , what contradicts the fact that S ∗ is

finitely consistent.
Assume now that (B ∪ C) 6∈ S ∗ . By MCF Property ??, ¬(B ∪ C) ∈ S ∗
and by the A = ¬C we have that v ∗ (¬(B ∪ C)) = T . But v ∗ (¬(B ∪ C)) =
¬v ∗ ((B ∪ C)) = T means that v ∗ ((B ∪ C)) = F , what end the proof of
this case.

416
The remaining cases of A = (B ∩ C), A = (B ⇒ C) are similar to the above and
are left to the reader as an exercise.
This end the proof of lemma 9.1 and completes the step S2.
S3: Maximal finitely consistent extension
Given a finitely consistent set S, we construct its maximal finitely consistent
extension S ∗ as follows.
The set of all formulas of L is countable, so is P F. We assume that all propo-
sitional formulas form a one-to-one sequence

A1 , A2 , ...., An , ..... (9.2)

We define a chain
S0 ⊆ S1 ⊆ S2 .... ⊆ Sn ⊆ .... (9.3)
of extentions of the set S by

S0 = S;


Sn ∪ {An } if Sn ∪ {An } is finitely consistent
Sn+1 =
Sn ∪ {¬An } otherwise.

We take

[
S∗ = Sn . (9.4)
n∈N

Clearly, S ⊆ S ∗ and for every A, either A ∈ S ∗ or ¬A ∈ S ∗ . To finish the proof

that S ∗ is MCF we have to show that it is finitely consistent.
First, let observe that if all sets Sn are finitely consistent, so is S ∗ = n∈N Sn .
S
Namely, let SF = {B1 , ..., Bk } be a finite subset of S ∗ . This means that there
are sets Si1 , ...Sik in the chain ( 9.3) such that Bm ∈ Sim , m = 1, ..k. Let
M = max(i1 , ...ik ). Obviously SF ⊆ SM and SM is finitely consistent as an
element of the chain (9.3). This proves the if all sets Sn are finitely consistent,
so is S ∗ .
Now we have to prove only that all Sn in the chain (9.3) are finitely consistent.
We carry the proof by induction over the length of the chain. S0 = S , so it is
FC by assumption of the Compactness Theorem 9.1. Assume now that Sn is
FC, we prove that so is Sn+1 . We have two cases to consider.

Case 1 Sn+1 = Sn ∪ {An }, then Sn+1 is FC by the definition of the chain

(9.3).

417
Case 2 Sn+1 = Sn ∪ {¬An }. Observe that this can happen only if Sn ∪ {An }
0 0
is not FC, i.e. there is a finite subset Sn ⊆ Sn , such that Sn ∪ {An } is not
consistent.
Suppose now that Sn+1 is not FC. This means that there is a finite subset
00 00
Sn ⊆ Sn , such that Sn ∪ {¬An } is not consistent.
0 00
Take Sn ∪ Sn . It is a finite subset of Sn so is consistent by the inductive
0 00
assumption. Let v be a model of Sn ∪ Sn . Then one of v ∗ (A), v ∗ (¬A)
0
must be T. This contradicts the inconsistency of both Sn ∪ {An } and
0
Sn ∪ {¬An }.

Thus, in ether case, Sn+1 , is after all consistent.

This completes the proof of the step S3 and the proof of the compactness
theorem for propositional logic of L (theorem 9.1) via the argument presented
in the step S4.

9.1.1 Henkin Method

Propositional tautologies within L, as defined here (definition 9.5) barely scratch

the surface of the collection of predicate (first -order) tautologies, i.e. of the
predicate valid formulas, as they are often called. For example the following
first-order formulas are propositional tautologies,

(∃xA(x) ∪ ¬∃xA(x)),

(∀xA(x) ∪ ¬∀xA(x)),

(¬(∃xA(x) ∪ ∀xA(x)) ⇒ (¬∃xA(x) ∩ ¬∀xA(x))),

but the following are predicate (first order) tautologies (valid formulas) that are
not propositional tautologies:

∀x(A(x) ∪ ¬A(x)),

(¬∀xA(x) ⇒ ∃x¬A(x)).

The first formula above is just a prime formula, the second is of the form (¬B ⇒
C), for B and C prime.
To stress the difference between the propositional and predicate (first order)
tautologies some books reserve the word tautology for the propositional tau-
tologies alone, using the notion of valid formula for the predicate (first order)
tautologies. We use here both notions, with the preference to predicate tautology
or tautology for short when there is no room for misunderstanding.

418
To make sure that there is no misunderstandings we remind the following defi-
nitions from chapter 8.
Given a first order language L with the set of variables V AR and the set of
formulas F. Let M = [M, I] be a structure for the language L, with the
universe M and the interpretation I and let s : V AR −→ M be an assignment
of L in M . We bring back some basic definitions from Chapter 8

A is satisfied in M
Given a structure M = [M, I], we say that a formula A is satisfied in M
if there is an assignment s : V AR −→ M such that

(M, s) |= A.

A is true in M
Given a structure M = [M, I], we say that a formula A is true in M if
(M, s) |= A for all assignments s : V AR −→ M .

Model M
If A is true in a structure M = [M, I], then M is called a model for A.
We denote it as
M |= A.

A is predicate tautology (valid)

A formula A is a predicate tautology (valid) if it is true in all structures
M = [M, I], i.e. if all structures are models of A.
We use use the term predicate tautology and and denote it, when there
is no confusion with propositional case as

|= A.

Case: A is a sentence
If A is a sentence, then the truth or falsity of (M, s) |= A is completely
independent of s. Thus we write

M |= A

and read M is a model of A, if for some (hence every) valuation s,

(M, s) |= A.

Model of a set of sentences

M is a model of a set S of sentences if and only if M |= A for all A ∈ S.
We write it
M |= S.

419
Predicate and Propositional Models

The relationship between the predicate models that are defined in terms of
structures M = [M, I] and assignments s : V AR −→ M and propositional
models that are defined in terms of truth assignments v : P −→ {T, F } is
established by the following lemma.

Lemma 9.2
Let M = [M, I] be a structure for the language L and let s : V AR −→ M an
assignment in M. There is a truth assignments v : P −→ {T, F } such that for
all formulas A of L,
(M, s) |= A if and only if v ∗ (A) = T.
In particular, for any set S of sentences of L,
if M |= S then S is consistent in sense of propositional logic.

Proof For any prime formula A ∈ P we define


T if (M, s) |= A
v(A) =
F otherwise.

Since every formula in L is either prime or is built up from prime formulas by

means of propositional connectives, the conclusion is obvious.
Observe, that the converse of the lemma is far from true. Consider a set

S = {∀x(A(x) ⇒ B(x)), ∀xA(x), ∃x¬B(x)}.

All formulas of S are different prime formulas, S is hence consistent in the sense
of propositional logic and obviously has no predicate (first-order) model.

Definition 9.12 (Language with Equality)

A predicate language
L = L(P, F, C)
is called a first order (predicate) language with equality if we one it its predicate
symbols is a two argument symbol E ∈ P representing an identity relation.
We write t = s as the abbreviation of E(t, s) for any terms t, s ∈ T and t 6= s
as the abbreviation of ¬E(t, s) .

Let L = L{¬,∩,∪,⇒} (P, F, C) be a predicate (first order) language with equality.

We adopt a following set of axioms.
Equality Axioms (9.5)

420
 For any free variable or constant of L, i.e for any u, w, ui , wi ∈ (V AR ∪ C),

E1 u = u,

E2 (u = w ⇒ w = u),

E3 ((u1 = u2 ∩ u2 = u3 ) ⇒ u1 = u3 ),

E4 ((u1 = w1 ∩ ... ∩ un = wn ) ⇒ (R(u1 , ..., un ) ⇒ R(w1 , ..., wn ))),

E5 ((u1 = w1 ∩ ... ∩ un = wn ) ⇒ (t(u1 , ..., un ) ⇒ t(w1 , ..., wn ))),

where R ∈ P and t ∈ T, i.e. R is an arbitrary n-ary relation symbol of L and

t ∈ T is an arbitrary n-ary term of L.

Observe that given any structure M = [M, I]. We have by simple verification
that for all s : V AR −→ M , and for all A ∈ {E1, E2, E3, E4, E5},

(M, s) |= A.

This proves the following

Fact 9.1
All equality axioms are predicate tautologies (valid) of L.

This is why we still call logic with equality axioms added to it, a logic.

Henkin’s Witnessing Expansion of L

Now we are going to define notions that are fundamental to the Henkin’s tech-
nique for reducing predicate logic to propositional logic. The first one is that of
witnessing expansion of the language L.
We construct an expansion of the language L by adding a set C of new constants
to it, i.e. by adding a specially constructed the set C to the set C such that
C ∩ C = ∅. The construction of the expansion is described below. The language
such constructed is called witnessing expansion of the language L.

Definition 9.13
For any predicate language L = L(P, F,C), the language

L(C) = L(P, F,C ∪ C))

for the set C defined by (10.11) and L(C) defined by (9.9) and the construction
described below is called a witnessing expansion of L. We write also

L(C) = L ∪ C.

421
Construction of the witnessing expansion of L
We define the set C of new constants by constructing an infinite sequence

C0 , C1 , ..., Cn , .... (9.6)

of sets of constants together with an infinite sequence

L0 , L1 , ..., Ln , .... (9.7)

We define sequences (11.25), (9.7) as follows. Let

C0 = ∅, L0 = L ∪ C0 = L.

We denote by
A[x]
the fact that the formula A has exactly one free variable and for each such a
formula we introduce a distinct new constant denoted by

cA[x] .

We define
C1 = {cA[x] : A[x] ∈ L0 }, L1 = L ∪ C1 .

Assume that we have defined Cn and Ln . We assign distinct new constant

symbol cA[x] to each formula A[x] of Ln which is not already a formula of Ln−1
(i.e., if some constant from Cn appears in A[x]). We write it informally as
A[x] ∈ (Ln − Ln−1 ). We define Cn+1 = Cn ∪ {cA[x] : A[x] ∈ (Ln − Ln−1 )} and
Ln+1 = L ∪ Cn+1 . We put
[
C= Cn (9.8)

and
L(C) = L ∪ C. (9.9)

Definition 9.14 (Henkin Axioms)

The following sentences
H1 (∃xA(x) ⇒ A(cA[x] )),
H2 (A(c¬A[x] ) ⇒ ∀xA(x))
are called Henkin axioms and for any formula A, a constant cA[x] ∈ C as
defined by (10.11) called a witnessing constant.

The informal idea behind the Henkin axioms is the following.

422
The axiom H1 says:
If ∃xA(x) is true in a structure, choose an element a satisfying A(x) and give
it a new name cA[x] .
The axiom H2 says:
If ∀xA(x) is false, choose a counterexample b and call it by a new name c¬A[x] .

Definition 9.15 (QuantifiersAxioms)

The following sentences
Q1 (∀xA(x) ⇒ A(t)), t is a closed term of L(C);
Q2 (A(t) ⇒ ∃xA(x)), t is a closed term of L(C)
are called quantifiers axioms.

Observe that the quantifiers axioms Q1, Q2 obviously are predicate tautologies.

Definition 9.16 (Henkin Set)

Any set of sentences of L(C) which are either Henkin axioms (definition 9.14)
or quantifiers axioms (definition 9.15) is called the Henkin set and denoted
by
SHenkin .

The Henkin is obviously not true in every L(C)-structure, but we are going to
show that every L -structure can be turned into an L(C)-structure which is a
model of SHenkin . Before we do so we need to introduce two new notions.

Reduct and Expansion

0 0 0 0
Given two languages L and L such that L ⊆ L . Let M = [M, I ] be a
0
structure for L . The structure
0
M = [M, I | L]
0 0
is called the reduct of M to the language L and M is called the expansion
0
of M to the language L .

0 0
Thus the reduct and the expansion M and M are the same except that M
0
assigns meanings to the symbols in L − L .

Lemma 9.3
Let M = [M, I] be any structure for the language L and let L(C) be the wit-
0 0
nessing expansion of L. There is an expansion M = [M, I ] of M = [M, I]
0
such that M is a model of the set SHenkin

423
 0
Proof In order to define the expansion of M to M we have to define the
0
interpretation I for the symbols of the language L(C) = L ∪ C, such that
0
I | L = I. This means that we have to define cI 0 for all c ∈ C. By the
definition, cI 0 ∈ M , so this also means that we have to assign the elements of
M to all constants c ∈ C in such a way that the resulting expansion is a model
for all sentences from SHenkin .
The quantifier axioms (definition 9.15) are predicate tautologies so they are
going to be true regardless, so we have to worry only about the Henkin axioms
(definition 9.14). Observe now that if the lemma 9.3 holds for the Henkin axiom
H1, then it must hold for the axiom H2. Namely, let’s consider the axiom H2:
(A(c¬A[x] ) ⇒ ∀xA(x)).

0 0
Assume that A(c¬A[x] ) is true in the expansion M , i.e. that M |= A(c¬A[x] )
0 0
and that M 6|= ∀xA(x). This means that M |= ¬∀xA(x) and by the de Morgan
0 0
Laws, M |= ∃x¬A(x). But we have assumed that M is a model for H1. In
0 0
particular M |= (∃x¬A(x) ⇒ ¬A(c¬A[x] )), and hence M |= ¬A(c¬A[x] ) and
0 0
this contradicts the assumption that M |= A(c¬A[x] ). Thus if M is a model
for all axioms of the type H1, it is also a model for all axioms of the type H2.
S
We define cI 0 for all c ∈ C = Cn by induction on n. Let n = 1 and cA[x] ∈ C1 .
By definition, C1 = {cA[x] : A[x] ∈ L}. In this case we have that ∃xA(x) ∈ L and
hence the notion M |= ∃xA(x) is well defined, as M = [M, I] is the structure
for the language L.
As we consider arbitrary structure M, there are two possibilities: M |= ∃xA(x)
or M 6|= ∃xA(x).
We define cI 0 , for all c ∈ C1 as follows.
If M |= ∃xA(x), then (M, v 0 ) |= A(x) for certain v 0 (x) = a ∈ M . We set
(cA[x]) )I 0 = a. If M 6|= ∃xA(x), we set (cA[x]) )I 0 arbitrarily.
This makes all the positive Henkin axioms about the cA[x] ∈ C1 true, i.e. M =
(M, I) |= (∃xA(x) ⇒ A(cA[x] )). But once cA[x] ∈ C1 are all interpreted in M ,
0
then the notion M |= A is defined for all formulas A ∈ L ∪ C1 . We carry the
same argument and define cI 0 , for all c ∈ C2 and so on. The inductive step in
the exactly the same way as the one above.

Definition 9.17 (Canonical structure )

0 0
Given a structure M = [M, I] for the language L. The expansion M = [M, I ]
of M = [M, I] is called a canonical structure for L(C) if all a ∈ M are
denoted by some c ∈ C. That is,
M = {cI 0 : c ∈ C}.

Now we are ready to state and proof a lemma 9.2 that provides the essential

424
step in the proof of the completeness theorem for predicate logic.

Theorem 9.2 (The reduction to propositional logic)

Let L = L(P, F, C) be a predicate language and let L(C) = L(P, F, C ∪ C) be a
witnessing expansion of L.
For any set S of sentences of L the following conditions are equivalent.

(i) S has a model, i.e. there is a structure M = [M, I] for the language L
such that M |= A for all A ∈ S.
(ii) There is a canonical L(C) structure M = [M, I] which is a model for
S, i.e. such that M |= A for all A ∈ S.
(iii) The set S ∪ SHenkin ∪ EQ is consistent in sense of propositional logic,
where EQ denotes the equality axioms E1 − E5.

Proof The implication (ii) → (i) is immediate. The implication (i) → (iii)
follows from lemma 9.3. We have to prove only the implication (iii) → (ii).
Assume that the set S ∪ SHenkin ∪ EQ is consistent in sense of propositional
logic and let v be a truth assignment to the prime sentences of L(C), such that
v ∗ (A) = T for all A ∈ S ∪ SHenkin ∪ EQ. To prove the lemma, we construct a
canonical L(C) structure M = [M, I] such that, for all sentences A of L(C),

M |= A if and only if v ∗ (A) = T. (9.10)

The truth assignment v is a propositional model for the set SHenkin , so v ∗

satisfies the following conditions:

v ∗ (∃xA(x)) = T if and only if v ∗ (A(cA[x] )) = T, (9.11)

v ∗ (∀xA(x)) = T if and only if v ∗ (A(t)) = T, (9.12)

for all closed terms t of L(C).
The conditions (9.11) and (9.12) allow us to construct the canonical L(C)
model M = [M, I] out of the constants in C in the following way.
To define M = [M, I] we must (1.) specify the universe M of M, (2.) define,
for each n-ary predicate symbol R ∈ P, the interpretation RI as an n-argument
relation in M , (3.) define, for each n-ary function symbol f ∈ F, the interpre-
tation fI : M n → M , and (4.) define, for each constant symbol c of L(C), i.e.
c ∈ C ∪ C, an element cI ∈ M .
The construction of M = [M, I] must be such that the condition (9.10) holds
for for all sentences A of L(C). This condition (9.10) tells us how to construct
the definitions (1.) - (4.) above. Here are the definitions.

425
(1.) Definition of the universe M of M.
In order to define the universe M we first define a relation ≈ on C by

c ≈ d if and only if v(c = d)) = T. (9.13)

The equality axioms axioms guarantee that the relation (9.13) is equivalence re-
lation on C, i.e. is reflexive, symmetric, and transitive. All axioms are predicate
tautologies, so v(c = d)) = T by axiom E1 and c ≈ c holds for any c ∈ C.
Symmetry condition ” if c ≈ d, then d ≈ c ” holds by axiom E2. Assume c ≈ d,
by definition v(c = d)) = T . By axiom E2

v ∗ ((c = d ⇒ d = c)) = v(c = d) ⇒ v(d = c) = T,

i.e. T ⇒ v(d = c) = T . This is possible only if v(d = c) = T . This proves that

d ≈ c.
We prove transitivity in a similar way. Assume now that c ≈ d and d ≈ e. We
check to see that c ≈ e. By the axiom E3 we have that

v ∗ (((c = d ∩ d = e) ⇒ c = e)) = T.

Since v(c = d)) = T and v(d = e)) = T by c ≈ d and d ≈ e,

v ∗ ((c = d ∩ d = e) ⇒ c = e) = (T ∩ T ⇒ c = e) = (T ⇒ c = e) = T,

we get that v(c = e) = T and hence d ≈ e.

We denote by [c] the equivalence class of c and we define the universe M of M
as
M = {[c] : c ∈ C}. (9.14)

(2.) Definition of RI ⊆ M n .

Let M be given by (9.14). We define

([c1 ], [c2 ], . . . , [cn ]) ∈ RI if and only if v(R(c1 , c2 , . . . , cn )) = T. (9.15)

We have to prove now that RI is well defined by the condition (9.15). To do so

we must check

if [c1 ] = [d1 ], [c2 ] = [d2 ], . . . , [cn ] = [dn ] and ([c1 ], [c2 ], . . . , [cn ]) ∈ RI ,

then ([d1 ], [d2 ], . . . , [dn ]) ∈ RI .

We have by the axiom E4 that

v ∗ (((c1 = d1 ∩ ... ∩ cn = dn ) ⇒ (R(c1 , ..., cn ) ⇒ R(d1 , ..., dn )))) = T. (9.16)

426
By the assumption [c1 ] = [d1 ], . . . , [cn ] = [dn ] we have that v(c1 = d1 ) =
T, . . . , v(cn = dn ) = T . By the assumption ([c1 ], [c2 ], . . . , [cn ]) ∈ RI , we have
that v(R(c1 , ..., cn )) = T . Hence the condition (9.16) becomes

(T ⇒ (T ⇒ v(R(d1 , ..., dn )))) = T.

It holds only when v(R(d1 , ..., dn )) = T and by (9.15) we proved that

([d1 ], [d2 ], . . . , [dn ]) ∈ RI .

(3.) Definition of fI : M n → M .
Let c1 , c2 , . . . , cn ∈ C and f ∈ F. We claim that there is c ∈ C such that
f (c1 , c2 , . . . , cn ) = c and v(f (c1 , c2 , . . . , cn ) = c) = T .
For consider the formula A(x) given by f (c1 , c2 , . . . , cn ) = x. If v ∗ (∃xA(x)) =
v ∗ (f (c1 , c2 , . . . , cn ) = x) = T , we want to prove v ∗ (A(cA[x] )) = T , i.e.

v(f (c1 , c2 , . . . , cn ) = cA ) = T.

So suppose that v(f (c1 , c2 , . . . , cn ) = cA ) = F . But one member of he Henkin

set SHenkin (definition 9.16) is the sentence (A(f (c1 , c2 , . . . , cn )) ⇒ ∃xA(x))
so we must have that v ∗ (A(f (c1 , c2 , . . . , cn ))) = F . But this says that v
assigns F to the atomic sentence f (c1 , c2 , . . . , cn ) = f (c1 , c2 , . . . , cn ), i.e. By
the axiom E1 v(ci = ci ) = T for i = 1, 2 . . . n and by E5

(v ∗ (c1 = c1 ∩ . . . cn = cn ) ⇒ v ∗ (f (c1 , . . . , cn ) = f (c1 , . . . , cn ))) = T.

This means that T ⇒ F = T and this contradiction proves there is c ∈ C such

that f (c1 , c2 , . . . , cn ) = c and v(f (c1 , c2 , . . . , cn ) = c) = T . We can hence
define

fI (([c1 ], . . . , [cn ]) = [c] for c such that v(f (c1 , . . . , cn ) = c) = T. (9.17)

The argument similar to the one used in (2.) proves that fI is well defined.

(4.) Definition of cI ∈ M .
For any c ∈ C we take cI = [c]. If d ∈ C, then an argument similar to that used
on (3.) shows that there is c ∈ C such that v(d = c) = T , i.e. d ≈ c, so we put
dI = [c].

This completes the construction of the canonical structure M = [M, I] and guar-
antees that (9.10) holds for for all atomic propositional sentences (definition
9.2), i.e. we proved that

M |= B if and only if v ∗ (B) = T, for sentences B ∈ P.

427
To complete the proof of the Lemma 9.2 we prove that the property (9.10) holds
for the canonical structure M = [M, I] defined above and all other sentences.
We carry the proof by induction on length of formulas. The case of propositional
connectives is trivial. For example, M |= (A ∩ B) if and only if M |= A
and M |= B) ( follows directly from the satisfaction definition) if and only
if v ∗ (A) = T and v ∗ (B) = T (by the induction hypothesis) if and only if
v ∗ (A ∩ B) = T . We proved

M |= (A ∩ B) if and only if v ∗ (A ∩ B) = T,

for all sentences A, B of L(C). The proof for all other connectives is similar.
We prove now the case of a sentence B of the form ∃xA(x), i.e. we want to show
that
M |= ∃xA(x) if and only if v ∗ (∃xA(x)) = T. (9.18)
v ∗ (∃xA(x)) = T . Then there is a c such that v ∗ (A(c) = T , so by induction
hypothesis, M |= A(c) so M |= ∃xA(x).
On the other hand, if v ∗ (∃xA(x)) = F , then by SHenking quantifier axiom Q2
(definition 9.15) we have that v ∗ (A(t)) = F for all closed terms t of L(C). In
particular, for every c ∈ C v ∗ (A(c)) = F . By induction hypothesis, M |=
¬A(c), for all c ∈ C. Since every element of M is denoted by some c ∈ C,
M |= ¬∃xA(x). Thus we proved (9.18).
The proof of the case of a sentence B of the form ∀xA(x) is similar and is left
to the reader.
The Reduction to Propositional Logic Theorem 9.2 provides not only a method
of constructing models of theories out of symbols, but also gives us immediate
proofs of the Compactness Theorem 9.3 for the predicate logic and Lowenheim-
Skolem Theorem 9.4.

Theorem 9.3 (Compactness theorem for the predicate logic)

Let S be an y set of predicate formulas of L.
The set S has a model if and only if any finite subset S0 of S has a model.

Proof
Let S be a set of predicate formulas such that every finite subset S0 of S has
a model. We need to show that S has a model. By the implication (iii) → (i)
of the Theorem 9.2 this is equivalent to proving that S ∪ SHenkin ∪ EQ is
consistent in the sense of propositional logic. By the Compactness Theorem 9.1
for propositional logic of L, it suffices to prove that for every finite subset S0 ⊂ S,
S0 ∪ SHenkin ∪ EQ is consistent, which follows from the hypothesis and the
implication (i) → (iii) of the Reduction to Propositional Logic Theorem 9.2.

Theorem 9.4 (Löwenheim-Skolem Theorem)

428
Let κ be an infinite cardinal and let Γ be a set of at most κ formulas of the first
order language.
If the set S has a model, then there is a model M = [M, I] of S such that
cardM ≤ κ.

Proof Let L be a predicate language with the alphabet A such that card(A) ≤
S card(F) ≤ κ. By the definition of the witnessing expansion L(C)
κ. Obviously,
of L, C = n Cn and for each n, card(Cn ) ≤ κ. So also cardC ≤ κ. Thus any
canonical structure for L(C) has ≤ κ elements. By the implication (i) → (ii) of
the Reduction to Propositional Logic Lemma 9.2 there is a model of S (canonical
structure) with ≤ κ elements.

9.2 Proof of Completeness Theorem

The proof of Gödel’s completeness theorem given by Kurt Gdel in his doctoral
dissertation of 1929 (and a rewritten version of the dissertation, published as an
article in 1930) is not easy to read today; it uses concepts and formalism that are
no longer used and terminology that is often obscure. The version given below
attempts to represent all the steps in the proof and all the important ideas
faithfully, while restating the proof in the modern language of mathematical
logic. This outline should not be considered a rigorous proof of the theorem.
It was first proved by Kurt Gödel in 1929. It was then simplified in 1947, when
Leon Henkin observed in his Ph.D. thesis that the hard part of the proof can be
presented as the Model Existence Theorem (published in 1949). Henkin’s proof
was simplified by Gisbert Hasenjaeger in 1953. Other now classical proofs has
been published by Rasiowa and Sikorski (1951 1952) using Boolean algebraic
methods and by by Beth (1953), using topological methods. Still other proofs
may be found in Hintikka(1955) and in Beth(1959).

Hilbert-style Proof System H

Language L
The language L of the proof system H is a predicate (first order) language
with equality (definition 9.12). We assume that the sets P, F, C are infinitely
enumerable. We also assume that it has a full set of propositional connectives,
i.e.
L = L{¬,∩,∪,⇒} (P, F, C). (9.19)

Logical Axioms LA
The set LA of logical axioms consists of three groups of axioms: propositional
axioms P A, equality axioms EA, and quantifiers axioms QA. we write it sym-

429
bolically as
LA = {P A, EA, QA}.
For the set P A of propositional axioms we choose any complete set of axioms
for propositional logic with a full set {¬, ∩, ∩, ⇒} of propositional connectives.
In some formalizations, including the one in the Handbook of Mathematical
Logic, Barwise, ed. (1977) we base our system H on, the authors just say for this
group of axioms: ”all tautologies”. They of course mean all predicate formulas
of L that are substitutions of propositional tautologies. This is done for the
need of being able to use freely these predicate substitutions of propositional
tautologies in the proof of completeness theorem for the proof system they
formalize this way.
In this case these tautologies are listed as axioms of the system and hence are
provable in it. This is a convenient approach, but also the one that makes such
a proof system not to be finately axiomatizable.
We avoid the infinite axiomatization by choosing a proper finite set of pred-
icate language version of propositional axioms that is known (proved already
for propositional case) to be complete, i.e. the one in which all propositional
tautologies are provable.

We choose, for name H (Hilbert) and historical sake, the set of Hilbert (1928)
axioms from chapter 5.
.For the set EA of equational axioms we choose the same set (9.5) as in sec-
tion 9.1.1 because they were used in the proof of Reduction to Propositional
Logic Theorem 9.2 and we want to be able to carry this proof within the system
H.
For the set QA of quantifiers axioms we choose the axioms such that the Henkin
set SHenkin axioms Q1, Q2 are their particular cases, so again a proof of the
Reduction to Propositional Logic Theorem 9.2 can be carried within H.

Rules of inference R
There are three inference rules: Modus Ponens (M P ) and two quantifiers rules
(G), (G1), (G2), called Generalization Rules.

We define the proof system H as follows.

H = (L{¬,∩,∪,⇒} (P, F, C), F, LA, R = {(M P ), (G), (G1), (G2)}), (9.20)

where
L = L{¬,∩,∪,⇒} (P, F, C) is predicate (first order) language with equality (defi-
nition 9.12). We assume that the sets P, F, C are infinitely enumerable.

430
F is the set of all well formed formulas of L.
LA is the set of logical axioms and

LA = {P A, EA, QA} (9.21)

for P A, EA, QA defined as follows.

P A is the set of propositional axioms (Hilbert, 1928)
A1 (A ⇒ A),
A2 (A ⇒ (B ⇒ A)),
A3 ((A ⇒ B) ⇒ ((B ⇒ C) ⇒ (A ⇒ C))),
A4 ((A ⇒ (A ⇒ B)) ⇒ (A ⇒ B)),
A5 ((A ⇒ (B ⇒ C)) ⇒ (B ⇒ (A ⇒ C))),
A6 ((A ⇒ B) ⇒ ((B ⇒ C) ⇒ (A ⇒ C))),
A7 ((A ∩ B) ⇒ A),
A8 ((A ∩ B) ⇒ B),
A9 ((A ⇒ B) ⇒ ((A ⇒ C) ⇒ (A ⇒ (B ∩ C))),
A10 (A ⇒ (A ∪ B)),
A11 (B ⇒ (A ∪ B)),
A12 ((A ⇒ C) ⇒ ((B ⇒ C) ⇒ ((A ∪ B) ⇒ C))),
A13 ((A ⇒ B) ⇒ ((A ⇒ ¬B) ⇒ ¬A)),
A14 (¬A ⇒ (A ⇒ B)),
A15 (A ∪ ¬A),
for any A, B, C ∈ F.

EA is the set of equality axioms.

E1 u = u,
E2 (u = w ⇒ w = u),
E3 ((u1 = u2 ∩ u2 = u3 ) ⇒ u1 = u3 ),
E4 ((u1 = w1 ∩ ... ∩ un = wn ) ⇒ (R(u1 , ..., un ) ⇒ R(w1 , ..., wn ))),
E5 ((u1 = w1 ∩ ... ∩ un = wn ) ⇒ (t(u1 , ..., un ) ⇒ t(w1 , ..., wn ))),
for any free variable or constant of L, R ∈ P, and t ∈ T, where R is an arbitrary
n-ary relation symbol of L and t ∈ T is an arbitrary n-ary term of L.

431
QA is the set of quantifiers axioms.
Q1 (∀xA(x) ⇒ A(t)),
Q2 (A(t) ⇒ ∃xA(x)),
where where t is a term, A(t) is a result of substitution of t for all free occur-
rences of x in A(x), and t is free for x in A(x), i.e. no occurrence of a variable
in t becomes a bound occurrence in A(t).

R is the set of rules of inference,

R = {(M P ), (G), (G1), (G2)},

where
(M P ) is Modus Ponens rule

A ; (A ⇒ B)
(M P ) ,
B
for any formulas A, B ∈ F.
(G), (G1), (G2) are the following quantifiers generalization rules.

A
(G) ,
∀xA
where A ∈ F. In particular we write
A(x)
(G)
∀xA(x)

for A(x) ∈ F and x ∈ V AR.

(B ⇒ A(x))
(G1) ,
(B ⇒ ∀xA(x))
where A(x), B ∈ F, x ∈ V AR, and B is such that x is not free in B.

(A(x) ⇒ B)
(G2) ,
(∃xA(x) ⇒ B)
where A(x), B ∈ F , x ∈ V AR, and B is such that x is not free in B.

We define, as we do for any proof system, a notion of a proof of a formula A

from a set S of formulas in H as a finite sequence of formulas B1 , B2 , . . . Bn ,
with Bn = A, each of which is either a logical axiom of H, a member of S, or
else follows from earlier formulas in the sequence by one of the inference rules
from R. We write it formally as follows.

432
Definition 9.18 (Proof from Γ)
Let Γ ⊆ F be any set of formulas of L. A proof in H of a formula A ∈ F from
s set Γ of formulas is a sequence

B1 , B2 , . . . Bn

of formulas, such that

B1 ∈ LA ∪ Γ, Bn = A
and for each i, 1 < i ≤ n, either Bi ∈ LA ∪ Γ or Bi is a conclusion of some of
the preceding expressions in the sequence B1 , B2 , . . . Bn by virtue of one of the
rules of inference from R.
We write
Γ `H A
to denote that the formula A has a proof from Γ in H and

Γ ` A,

when the proof system H is fixed.

The case when Γ = ∅ is a special one. By the definition 9.18, ∅ `H A means

that in the proof of A only logical axioms LA are used. We hence write

`H A

to denote that a formula A has a proof in H.

As we work with a fixed (and only one) proof system, we use the notation

Γ ` A and ` A

to denote the proof of a formula A from a set Γ and proof of a formula A in H,

respectively.
Any proof of the completeness theorem for a given proof system consists always
of two parts. First we have show that all formulas that have a proof in the system
are tautologies. This implication is called a soundness theorem, or soundness
part of the completeness theorem.
The second implication says: if a formula is a tautology then it has a proof in
the proof system. This alone is sometimes called a completeness theorem (on
assumption that the system is sound). Traditionally it is called a completeness
part of the completeness theorem.
We know that all axioms of H are predicate tautologies (proved in chapter 8 and
all rules of inference from R are sound as the corresponding formulas were also
proved in chapter 8 to be predicate tautologies and so the system H is sound,
i.e. the following holds for H.

433
Theorem 9.5 (Soundness Theorem)
For every formula A ∈ F of the language L of the proof system H,
if ` A, then |= A.

The soundness theorem proves that the proofs in the system ”produce” only
tautologies. We show here, as the next step that our proof system H ”produces”
not only tautologies, but that all tautologies are provable in it.
This is called a completeness theorem for classical predicate (first order logic,
as it all is proven with respect to classical semantics. This is why it is called a
completeness of predicate logic theorem.
The goal is now to prove the completeness part of the following.

Theorem 9.6 (Gödel Completeness of Predicate Logic)

For any formula A of the language L of the system H,
A is provable in H if and only if A is a predicate tautology (valid).
We write it symbolically as
`A if and only if |= A. (9.22)

We are going to prove the Gödel’ s Theorem 9.6 as a particular case of The-
orem 9.7 that follows. It is its more general, and more modern version. This
version, as well as the method of proving it, was first introduced by Henkin in
1947. It became with its consequent improvements, as classical as the Gödel’s
own. It uses the notion of a logical implication, and some other notions. We
introduce them below.

Definition 9.19 (Sentence, Closure)

Any formula of L without free variables is called a sentence.
For any formula A(x1 , . . . xn ), a sentence
∀x1 ∀x2 . . . ∀xn A(x1 , . . . xn )
is called a closure of A(x1 , . . . xn ).

Directly from the definition 10.16 have that the following hold.

Fact 9.2 For any formula A(x1 , . . . xn ),

|= A(x1 , . . . xn ) if and only if |= ∀x1 ∀x2 . . . ∀xn A(x1 , . . . xn ).

Definition 9.20 (Logical Implication)

For any set Γ ⊆ F of formulas of L and any A ∈ F, we say that the set Γ
logically implies the formula A and write it as Γ |= A if and only if all
models of Γ are models of A.

434
Observe, that in order to prove that Γ |= B we have to show that the implication

if M |= Γ then M |= B

holds for all structures M = [U, I] for L.

Lemma 9.4 Let Γ be a set of sentences of L, for any formula A(x1 , . . . xn )

that is not a sentence,

Γ ` A(x1 , . . . xn ) if and only if Γ |= ∀x1 ∀x2 . . . ∀xn A(x1 , . . . xn ).

Proof
Let B1 , B2 , . . . Bn be a proof of A(x1 , . . . xn ) from Γ and let M be a model of Γ.
We use Fact 9.2 and prove by induction on n, that M |= ∀x1 ∀x2 . . . ∀xn Bn (x1 , . . . xn ),
and hence M |= ∀x1 ∀x2 . . . ∀xn A(x1 , . . . xn ). The converse implication is obvi-
ous.
Fact 9.2 and Lemma 9.4 show that we need to consider only sentences (closed
formulas) of L, since a formula of F is a tautology if and only if its closure is
a tautology and is provable from Γ if and only if its closure is provable from Γ.
This justifies the following generalization of the original Gödel’ s completeness
of predicate logic Theorem 9.6.

Theorem 9.7 (Gödel Completeness Theorem)

Let Γ be any set of sentences and A any sentence of a language L of Hilbert
proof system H.
A sentence A is provable from Γ in H if and only if the set Γ logically
implies A.
We write it in symbols,

Γ`A if and only if Γ |= A. (9.23)

Remark
We want to remind the readers that the Reduction Predicate Logic to Proposi-
tional Logic Section 9.1 is an integral and the first part of the proof the Gödel
Completeness Theorem 9.7. We presented it separately for two reasons.
R1. The reduction method and theorems and their proofs are purely semantical
in their nature and hence are independent of the proof system H.
R2. Because of R1. the reduction method can be used/adapted to a proof of
completeness theorem of any other proof system one needs to prove the classical
completeness theorem for. See section ??.
In order to prove it we must formulate it properly so we need to introduce
few new important and classical notion and prove some lemmas needed for the
proof. The first is the notion of consistency.

435
There are two definitions of consistency; semantical and syntactical. The
semantical one uses definition the notion of a model and says, in plain English:
a set of formulas is consistent if it has a model.
The syntactical one uses the notion of provability and says: a set of formulas
is consistent if one can’t prove a contradiction from it.
We have used, in the Proof Two of the Completeness Theorem for propositional
logic (chapter 5) the syntactical definition of consistency. We use here the
following semantical definition.

Definition 9.21 (Consistent/Inconsistent)

A set Γ ⊆ F of formulas of L is consistent if and only if it has a model,
otherwise, it is inconsistent.

Directly from the above definitions we have the following.

Lemma 9.5
For any set Γ ⊆ F of formulas of L and any A ∈ F,
if Γ |= A, then the set Γ ∪ {¬A} is inconsistent.

Proof
Assume Γ |= A and Γ ∪ {¬A} is consistent. By definition 9.21 there is a struc-
ture M = [U, I], such that M |= Γ and M |= ¬A, i.e. M 6|= A. This is a
contradiction with Γ |= A.

Now we are going to prove the following Lemma 9.6 that is crucial, together with
the Reduction to Propositional Logic Theorem 9.2 and the above Lemma 9.5 to
the proof of the Completeness Theorem 9.7.

Lemma 9.6
Let Γ be any set of sentences of a language L of Hilbert proof system H.
The following conditions hold For any formulas A, B ∈ F of L.
(i) If Γ ` (A ⇒ B) and Γ ` (¬A ⇒ B), then Γ ` B.
(ii) If Γ ` ((A ⇒ C) ⇒ B), then Γ ` (¬A ⇒ B) and Γ ` (C ⇒ B).
(iii) If x does not appear in B and if Γ ` ((∃yA(y) ⇒ A(x)) ⇒ B), then Γ ` B.
(iv) If x does not appear in B and if Γ ` ((A(x) ⇒ ∀yA(y)) ⇒ B), then Γ ` B.

Proof
(i) Notice that the formula ((A ⇒ B) ⇒ ((¬A ⇒ B) ⇒ B)) is a substitution

436
of a propositional tautology, hence by definition of H, is provable in it. By
monotonicity,
Γ ` ((A ⇒ B) ⇒ ((¬A ⇒ B) ⇒ B)).
By assuption Γ ` (A ⇒ B) and Modus Ponens we get

Γ ` ((¬A ⇒ B) ⇒ B).

By assuption SΓ ` (¬A ⇒ B) and Modus Ponens we get Γ ` B.

(ii) The formulas (1) (((A ⇒ B) ⇒ (¬A ⇒ B))) and (2) (((A ⇒ B) ⇒
B) ⇒ (C ⇒ B)) are substitution of a propositional tautologies, hence are
provable in H. Assume Γ ` ((A ⇒ C) ⇒ B). By monotonicity and (1) we get
Γ ` (¬A ⇒ B) and by (2) we get ` (C ⇒ B).
(iii) Assume Γ ` ((∃yA(y) ⇒ A(x)) ⇒ B). Observe that it is a particular case
of assumption Γ ` ((A ⇒ C) ⇒ B) in (ii), for A = ∃yA(y), C = A(x), B = B.
Hence by (ii) we have that Γ ` (¬∃yA(y) ⇒ B) and Γ ` (A(x) ⇒ B).
Apply Generalization Rule G2 to Γ ` (A(x) ⇒ B) and we have Γ ` (∃yA(y) ⇒
B.) Then by (i) applied to Γ ` (∃yA(y) ⇒ B) and Γ ` (¬∃yA(y) ⇒ B) we get
Γ ` B.
The proof of (iv) is similar to (iii), but uses the Generalization Rule G1.This
ends the proof of the lemma.
Now we are ready to conduct the proof of the Completeness Theorem for H.
There are two versions. Theorem 9.7 that is Gödel original formulation and the
one we used in previous chapters of the book. It follows from theorem ??). We
put them both together as follows.

Theorem 9.8 (H Completeness)

Let Γ be any set of sentences and A any sentence of a language L of Hilbert
proof system H.

Γ`A if and only if Γ |= A. (9.24)

In particular, for any formula A of L,

`A if and only if |= A. (9.25)

Proof
We first prove the completeness part (9.24), i.e. we prove the implication

if Γ |= A, then Γ ` A. (9.26)

Suppose that Γ |= A, i.e. we assume that all L models of Γ are models of A.

By Lemma 9.5 the set Γ ∪ {¬A} is inconsistent.

437
Let M |= Γ. We construct, as a next step, a witnessing expansion language L(C)
of L (definition 9.13). By the Reduction to Propositional Logic Theorem 9.2,
the set Γ ∪ SHenkin ∪ EQ is consistent in a sense of propositional logic in L. The
set SHenkin is a Henkin Set (definition 9.16) and EQ are equality axioms (9.5)
that are also the equality axioms EQ of H.
By the Compactness Theorem 9.1 for propositional logic of L there is a finite
set S0 ⊆ Γ ∪ SHenkin ∪ EQ such that S0 ∪ {¬A} is inconsistent in the sense of
propositional logic.
We list all elements of S0 in a sequence

A1 , A2 , . . . , An , B1 , B2 , . . . , Bm (9.27)

as follows. The sequence A1 , A2 , . . . , An consists of those elements of S0

which are either in Γ ∪ EQ or else are quantifiers axioms (definition 9.15) that
are particular cases of the quantifiers axioms QA of H. We list them in any
order.
The sequence B1 , B2 , . . . , Bm consists of elements of S0 which are Henkin Ax-
ioms (definition 9.14) but listed carefully as to be described as follows. Observe
that by definition 9.13,
[
L(C) = Ln , for L = L0 ⊆ L1 ⊆ . . . .
n∈N

We define the rank of A ∈ L(C) to be the least n, such that A ∈ Ln .

Now we choose for B1 a Henkin Axiom in S0 of the maximum rank.
We choose for B1 a Henkin Axiom in S0 − {B1 } of the maximum rank.
We choose for B2 a Henkin Axiom in S0 − {B1 , B2 } of the maximum rank, etc.
The point of choosing the formulas Bi ’s in this way is to make sure that the wit-
nessing constant about which Bi speaks, does not appear in Bi+1 , Bi+2 , . . . , Bm .
For example, if B1 is
(∃xC(x) ⇒ C(cC[x] )),
then C[x] does not appear in any of the other B2 , . . . , Bm , by the maximality
condition on B1 .
We know that that S0 ∪ {¬A} is inconsistent in the sense of propositional logic,
i.e. it does not have a (propositional) model. This means that v ∗ (¬A) 6= T for
all v and so v ∗ (A) = T for all v. Hence a sentence

(A1 ⇒ (A2 ⇒ . . . (An ⇒ (B1 ⇒ . . . (Bm ⇒ A))..)

is a propositional tautology.
We now replace each witnessing constant in this sentence by a distinct new
variable and write the result as

(A1 0 ⇒ (A2 0 ⇒ . . . (An 0 ⇒ (B1 0 ⇒ . . . (Bm 0 ⇒ A))..)

438
. We have A0 = A since A has no witnessing constant in it. The result is still a
tautology and hence is provable in H from propositional axioms P A and Modus
Ponens. By monotonicity

S0 ` (A1 0 ⇒ (A2 0 ⇒ . . . (An 0 ⇒ (B1 0 ⇒ . . . (Bm 0 ⇒ A))..). (9.28)

Each of A1 0 , A2 0 , . . . , An 0 is either a quantifiers axiom from QA of H or else in

S0 , so
S0 ` Ai 0 for all 1 ≤ i ≤ n.
We apply Modus Ponens to the above and (9.28) n times and get

S0 ` (B1 0 ⇒ (B2 0 ⇒ . . . (Bm 0 ⇒ A))..). (9.29)

For example, if B1 0 is (∃xC(x) ⇒ C(x)), we have by (9.29)

S0 ` ((∃xC(x) ⇒ C(x)) ⇒ B). (9.30)

for B = (B2 0 ⇒ . . . (Bm 0 ⇒ A))..). By the Reduction to Propositional Logic

Theorem 9.2 part (iii), we get S0 ` B, i.e.

S0 ` (B2 0 ⇒ . . . (Bm 0 ⇒ A))..). (9.31)

If, for example, B2 0 is (D(x) ⇒ ∀xD(x)), we have by (9.30)

S0 ` ((∃xC(x) ⇒ C(x)) ⇒ D). (9.32)

for D = (B3 0 ⇒ . . . (Bm 0 ⇒ A))..). By the Reduction to Propositional Logic

Theorem 9.2 part (iv), we get S0 ` D, i.e.

S0 ` (B3 0 ⇒ . . . (Bm 0 ⇒ A))..). (9.33)

. We hence apply parts (iii) ad (iv) of Theorem 9.2 to successively remove all
B1 0 , ...., Bm 0 and obtain the proof of A from S0 .
This ends the proof that Γ ` A and hence the proof of the completeness part
of (9.24).

The soundness part of of (9.24), i.e. the implication

if Γ ` A, then Γ |= A,

holds for any sentence A of L directly by Fact 9.2, Lemma 9.4, and Theorem 9.5.

The Theorem 9.6, as expressed by (9.25) follows from Fact 9.2, Lemma 9.4 as a
case of (9.24) for Γ = ∅.
This ends the proof of Theorem 9.8 as well as Theorem 9.7, and the proof of
the original Gödel Completeness of Predicate Logic Theorem 9.6.

439
9.3 Deduction Theorem
In mathematical arguments, one often assumes a statement A on the assumption
(hypothesis) of some other statement B and then concludes that we have proved
the implication ”if A, then B”. This reasoning is justified by the following
theorem, called a Deduction Theorem. It was first formulated and proved for a
certain Hilbert proof system S for the classical propositional logic by Herbrand
in 1930 in a form stated below.

Theorem 9.9 (Deduction Theorem) (Herbrand,1930)

For any formulas A, B of the language of a propositional proof system S,

if A `S B, then `S (A ⇒ B).

In chapter 5 we formulated and proved the following, more general version of

the Herbrand Theorem 9.10 for a very simple (two logical axioms and Modus
Ponens) propositional proof system H1.

Theorem 9.10 (Deduction Theorem)

For any subset Γ of the set of formulas F of H1 and for any formulas A, B ∈ F,

Γ, A `H1 B if and only if Γ `H1 (A ⇒ B).

In particular,
A `H1 B if and only if `H1 (A ⇒ B).

A natural question arises: does deduction theorem holds for the predicate logic
in general and for its proof system H we defined here?.
The Theorem 9.10 cannot be carried directly to the predicate logic, but it nev-
ertheless holds with some modifications. Here is where the problem lays.

Fact 9.3 Given the proof system (9.20), i.e.

H = (L(P, F, C), F, LA, R = {(M P ), (G), (G1), (G2)}).
For any formula A(x) ∈ F,
A(x) ` ∀xA(x), but it is not always the case that ` (A(x) ⇒ ∀xA(x)).

Proof
Obviously, A(x) ` ∀xA(x) by Generalization rule (G). Let now A(x) be an
atomic formula P (x). By the Completeness Theorem 9.6, ` (P (x) ⇒ ∀xP (x))
if and only if |= (P (x) ⇒ ∀xP (x)). Consider a structure M = [M, I], where
M contains at least two elements c and d. We define PI ⊆ M as a property
that holds only for c, i.e. PI = {c}. Take any assignment of L in M, i.e.

440
s : V AR −→ M . Then (M, s) |= P (x) only when s(x) = c for all x ∈ V AR.
M = [M, I] is a counter model for (P (x) ⇒ ∀xP (x)), as we found s such
(M, s) |= P (x) and obviously (M, s) 6|= ∀xP (x). This proves that Deduction
Theorem fails for A(x) being an atomic formula P (x).

The Fact 9.3 shows that the problem is with application of the generalization rule
(G) to the formula A ∈ Γ. To handle this we introduce, after Mendelson(1987)
the following notion.

Definition 9.22
Let A be one of formulas in Γ and let
B1 , B2 , ..., Bn (9.34)
a deduction (proof of Bn from Γ, together with justification at each step.
We say that the formula Bi depends upon A in the proof (9.34)
if and only if
(1) Bi is A and the justification for Bi is Bi ∈ Γ
or
(2) Bi is justified as direct consequence by MP or (G) of some preceding for-
mulas in the sequence (9.34), where at least one of these preceding formulas
depends upon A.

Here is a deduction
B1 , B2 , . . . , B5 (9.35)
showing that
A, (∀xA ⇒ C).
B1 A, Hyp
B2 ∀xA, B1 , (G)
B3 (∀xA ⇒ C), Hyp
B4 C, MP on B2 , B3
B5 ∀xC. (G)
Observe that the formulas A, C may, or may not have x as a free variable.

Example 9.2
In the derivation (9.35)
B1 depends upon A,
B2 depends upon A,
B3 depends upon (∀xA ⇒ C),
B4 depends upon A and (∀xA ⇒ C),
B5 depends upon A and (∀xA ⇒ C).

441
Lemma 9.7
If B does not depend upon A in a deduction showing that Γ, A ` B, then
Γ ` B.

Proof
Let B1 , B2 , . . . , Bn = B be a deduction of B from Γ, A in which B does not
depend upon A. we prove by Induction that Γ ` B. Assume that Lemma 9.7
holds for all deductions of the length less than n. If B ∈ Γ or B ∈ LA, then
Γ ` B. If B is a direct consequence of two preceding formulas, then, since B
does not depend upon A , neither do theses preceding formulas. By inductive
hypothesis, theses preceding formulas have a proof from Γ alone. Hence so does
B.
Now we are ready to formulate and prove the Deduction Theorem 9.11 for
predicate logic.

Theorem 9.11 (Deduction Theorem)

For any formulas A, B of the language of proof system H the following holds.
(1) Assume that in some deduction showing that

Γ, A ` B,

no application of the generalization rule (G) to a formula that depends upon A

has as its quantified variable a free variable of A. Then

Γ ` (A ⇒ B).

(2) If Γ ` (A ⇒ B), then Γ, A `B.

Proof
The proof extends the proof of the Deduction Theorem for propositional logic
from chapter 5. We adopt the propositional proof (for a different proof system)
to the system H and adding the predicate case. For the sake of clarity and
independence we write now the whole proof in all details.
(1) Assume that Γ, A `B, i.e. that we have a formal proof

B1 , B2 , ..., Bn (9.36)

of B from the set of formulas Γ ∪ {A}. In order to prove that Γ ` (A ⇒ B) we

will prove the following a little bit stronger statement S.

S: Γ ` (A ⇒ Bi ) for all Bi (1 ≤ i ≤ n) in the proof (9.36) of B.

Hence, in particular case, when i = n, we will obtain that also

Γ ` (A ⇒ B).

442
The proof of S is conducted by induction on i ( 1 ≤ i ≤ n).
Base Step i = 1.
When i = 1, it means that the formal proof (5.6) contains only one element B1 .
By the definition of the formal proof from Γ ∪ {A}, we have that B1 ∈ LA, or
B1 ∈ Γ, or B1 = A, i.e.
B1 ∈ LA ∪ Γ ∪ {A}.
Here we have two cases.
Case 1. B1 ∈ LA ∪ Γ.
Observe that the formula is a particular case of A2 of H. By assumption B1 ∈
LA ∪ Γ, hence we get the required proof of (A ⇒ B1 ) from Γ by the following
application of the Modus Ponens rule
B1 ; (B1 ⇒ (A ⇒ B1 ))
(M P ) .
(A ⇒ B1 )

Case 2. B1 = A.
When B1 = A, then to prove Γ ` (A ⇒ B) means to prove Γ ` (A ⇒ A). But
(A ⇒ A) ∈ LA (axiom A21of H), i.e. ` (A ⇒ A). By the monotonicity of the
consequence we have that Γ`(A ⇒ A). The above cases conclude the proof of
the Base case i = 1.
Inductive step
Assume that Γ `(A ⇒ Bk ) for all k < i, we will show that using this fact we
can conclude that also Γ `(A ⇒ Bi ).
Consider a formula Bi in the sequence 9.36. By the definition, Bi ∈ LA∪Γ∪{A}
or Bi follows by MP from certain Bj , Bm such that j < m < i. We have to
consider again two cases.
Case 1. Bi ∈ LA ∪ Γ ∪ {A}.
The proof of (A ⇒ Bi ) from Γ in this case is obtained from the proof of the
Base Step for i = 1 by replacement B1 by Bi and will be omitted here as a
straightforward repetition.
Case 2. Bi is a conclusion of MP.
If Bi is a conclusion of MP, then we must have two formulas Bj , Bm in the
sequence 9.36 such that j < i, m < i, j 6= m and
Bj ; Bm
(M P ) .
Bi
By the inductive assumption, the formulas Bj , Bm are such that

Γ ` (A ⇒ Bj ) (9.37)
and
Γ ` (A ⇒ Bm ). (9.38)

443
Moreover, by the definition of the Modus Ponens rule, the formula Bm has to
have a form (Bj ⇒ Bi ), i.e. Bm = (Bj ⇒ Bi ), and the the inductive assumption
(9.38) can be re-written as follows.

Γ ` (A ⇒ (Bj ⇒ Bi )), f or j < i. (9.39)

Observe now that the formula

((A ⇒ (Bj ⇒ Bi )) ⇒ ((A ⇒ Bj ) ⇒ (A ⇒ Bi )))

is a substitution of the axiom A3 of H and hence has a proof in H. By the

monotonicity,

Γ ` ((A ⇒ (Bj ⇒ Bi )) ⇒ ((A ⇒ Bj ) ⇒ (A ⇒ Bi ))). (9.40)

Applying the rule MP to formulas (9.40) and (9.39,) i.e. performing the following

(A ⇒ (Bj ⇒ Bi )) ; ((A ⇒ (Bj ⇒ Bi )) ⇒ ((A ⇒ Bj ) ⇒ (A ⇒ Bi )))

(M P )
((A ⇒ Bj ) ⇒ (A ⇒ Bi ))

we get that also

Γ `((A ⇒ Bj ) ⇒ (A ⇒ Bi )). (9.41)

Applying again the rule MP to formulas 9.37 and 9.41, i.e. performing the
following

(A ⇒ Bj ) ; ((A ⇒ Bj ) ⇒ (A ⇒ Bi ))
(M P )
(A ⇒ Bi )
we get that
Γ `(A ⇒ Bi ).

Finally, suppose that there is some j < i such that Bi is ∀xBj . By hypothesis
Γ ` Bj and either (i) Bj does not depend upon A or (ii) x is not free variable
in A.
We have two cases (i) and (ii) to consider.
(i) If Bj does not depend upon A , then by Lemma 9.7 Γ ` Bj and, consequently,
by the generalization rule (G), Γ ` ∀xBj . Thus Γ ` Bi .
Now, by hypothesis Γ ` Bj and by axiom A2, ` (Bi ⇒ (A ⇒ Bi )). Applying
MP we get Γ ` A ⇒ Bi ).
(ii) If x is not free variable in A, then, by Completeness Theorem 9.6 and
|= (∀x(A ⇒ Bj ) ⇒ (A ⇒ ∀xBj )) we have that ` (∀x(A ⇒ Bj ) ⇒ (A ⇒ ∀xBj ))
.

444
Since Γ ` A ⇒ Bi ), we get by the generalization rule (G), Γ ` ∀x(A ⇒ Bj ),
and so, by MP, Γ ` A ⇒ ∀xBj ); that is Γ ` A ⇒ Bi ).
This completes the induction and the case (1) holds for i = n.
(2) The proof of the implication

if Γ ` (A ⇒ B) then Γ, A `B

is straightforward. Assume that Γ ` (A ⇒ B) , hence by monotonicitywe have

also that Γ, A ` (A ⇒ B). Obviously, Γ, A ` A. Applying Modus Ponens to
the above, we get the proof of B from {Γ, A} i.e. we have proved that Γ, A ` B.
This ends the proof of the Deduction Theorem for H.

9.4 Some other Axiomatizations

We present here some of most known, and historically important axiomatiza-

tions of classical predicate logic, i.e. the following Hilbert style proof systems.

1. Hilbert and Ackermann (1928)

D. Hilbert and W. Ackermann, Grundzügen der Theoretischen Logik (Princi-

ples of Theoretical Logic), Springer - Verlag, 1928. The book grew from the
courses on logic and foundations of mathematics Hilbert gave in years 1917-
1922. He received help in writeup from Barnays and the material was put into
the book by Ackermann and Hilbert. It was conceived as an introduction to
mathematical logic and was followed by D. Hilbert and P. Bernays, Grundzügen
der Mathematik I,II. Springer -Verlag, 1934, 1939.
Hilbert and Ackermann formulated and asked a question of the completeness
for their deductive (proof) system. It was answered affirmatively by Kurt Gödel
in 1929 with proof of his Completeness Theorem 9.6.
We define the Hilbert and Ackermann system HA following a pattern estab-
lished for the H system (9.20). The original language use by Hilbert and Ack-
ermann contained only negation ¬ and disjunction ∪ and so do we.

HA = (L{¬,∪} (P, F, C), F, LA, R = {(M P ), (SB), (G1), (G2)}), (9.42)

where the set LA of logical axioms is as follows.

Propositional Axioms
A1 (¬(A ∪ A) ∪ A), A2 (¬A ∪ (A ∪ B)),
A3 (¬(A ∪ B) ∪ (B ∪ A)), A4 (¬(¬B ∪ C) ∪ (¬(A ∪ B) ∪ (A ∪ C))),

445
for any A, B, C, ∈ F.

Quantifiers Axioms
Q1 (¬∀xA(x) ∪ A(x)), Q2 (¬A(x) ∪ ∃xA(x)),
Q2 (¬A(x) ∪ ∃xA(x)),
for any A(x) ∈ F.

Rules of Inference R

(MP) is the Modus Ponens rule. It has, in the language L{¬,∪} , a form

A ; (¬A ∪ B)
(M P ) .
B

. (SB) is a substitution rule

A(x1 , x2 , . . . xn )
(SB) ,
A(t1 , t2 , . . . tn )
where A(x1 , x2 , . . . xn ) ∈ F and t1 , t2 , . . . tn ∈ T.
(G1), (G2) are quantifiers generalization rules.

(¬B ∪ A(x)) (¬A(x) ∪ B)

(G1) , (G2) ,
(¬B ∪ ∀xA(x)) (¬∃xA(x) ∪ B)
where A(x), B ∈ F and B is such that x is not free in B.

The HA system is usually written now with the use of implication, i.e. as based
on a language L = L{¬,⇒} (P, F, C), i.e. as a proof system

HAI = (L{¬,⇒} (P, F, C), F, LA, R = {(M P ), (SB), (G1), (G2)}), (9.43)

where the set LA of logical axioms is as follows.

Propositional Axioms
A1 ((A ∪ A) ⇒ A), A2 (A ⇒ (A ∪ B)),
A3 ((A ∪ B) ⇒ (B ∪ A)), A4 ((¬B ∪ C) ⇒ ((A ∪ B) ⇒ (A ∪ C))),
for any A, B, C, ∈ F.

446
Quantifiers Axioms
Q1 (∀xA(x) ⇒ A(x)), Q2 (A(x) ⇒ ∃xA(x)),
for any A(x) ∈ F.

Rules of Inference R
(M P ) is Modus Ponens rule
A ; (A ⇒ B)
(M P ) ,
B
for any formulas A, B ∈ F.
(SB) is a substitution rule

A(x1 , x2 , . . . xn )
(SB) ,
A(t1 , t2 , . . . tn )
where A(x1 , x2 , . . . xn ) ∈ F and t1 , t2 , . . . tn ∈ T.
(G1), (G2) are quantifiers generalization rules.

(B ⇒ A(x)) (A(x) ⇒ B)
(G1) , (G2) ,
(B ⇒ ∀xA(x)) (∃xA(x) ⇒ B)
where A(x), B ∈ F and B is such that x is not free in B.

The form of the quantifiers axioms Q1, Q2, and quantifiers generalization rule
(Q2) is due to Bernays.

2. Mendelson (1987)
Here is the first order logic proof system HM as introduced in the Elliott
Mendelson’s book Introduction to Mathematical Logic, hence the name. (1987).
It is an generalization to the predicate language of the proof system H2 for
propositional logic defined and studied in Chapter 5.
HM = (L{¬,∪} (P, F, C), F, LA, R = {(M P ), (G)}). (9.44)

Propositional Axioms
A1 (A ⇒ (B ⇒ A)),
A2 ((A ⇒ (B ⇒ C)) ⇒ ((A ⇒ B) ⇒ (A ⇒ C))),
A3 ((¬B ⇒ ¬A) ⇒ ((¬B ⇒ A) ⇒ B))),
for any A, B, C, ∈ F.

447
Quantifiers Axioms
Q1 (∀xA(x) ⇒ A(t)),
where where t is a term, A(t) is a result of substitution of t for all free occurrences
of x in A(x), and t is free for x in A(x), i.e. no occurrence of a variable in t
becomes a bound occurrence in A(t).
Q2 (∀x(B ⇒ A(x)) ⇒ (B ⇒ ∀xA(x))), where A(x), B ∈ F and B is such
that x is not free in B.

Rules of Inference R
(M P ) is the Modus Ponens rule

A ; (A ⇒ B)
(M P ) ,
B
for any formulas A, B ∈ F.
(G) is the generalization rule

A(x)
(G) ,
∀xA(x)

where A(x) ∈ F and x ∈ V AR.

Rasiowa-Sikorski (1950)
Rasiowa and Sikorski are the authors of the first algebraic proof of the Gödel
completeness theorem ever given in 1950. Other algebraic proofs were later
given by Rieger, Beth, Los in 1951, and Scott in 1954.
Here is their original axiomatization.

RS = (L{¬,∩,∪,⇒} (P, F, C), F, LA, R). (9.45)

Propositional Axioms
A1 ((A ⇒ B) ⇒ ((B ⇒ C) ⇒ (A ⇒ C))),
A2 (A ⇒ (A ∪ B)),
A3 (B ⇒ (A ∪ B)),
A4 ((A ⇒ C) ⇒ ((B ⇒ C) ⇒ ((A ∪ B) ⇒ C))),
A5 ((A ∩ B) ⇒ A),
A6 ((A ∩ B) ⇒ B),
A7 ((C ⇒ A) ⇒ ((C ⇒ B) ⇒ (C ⇒ (A ∩ B))),

448
A8 ((A ⇒ (B ⇒ C)) ⇒ ((A ∩ B) ⇒ C)),
A9 (((A ∩ B) ⇒ C) ⇒ (A ⇒ (B ⇒ C)),
A10 (A ∩ ¬A) ⇒ B),
A11 ((A ⇒ (A ∩ ¬A)) ⇒ ¬A),
A12 (A ∪ ¬A),
for any A, B, C ∈ F.

Rules of Inference R = {(M P ), (SB), (Q1), (Q2), (Q3), (Q4)}.

(M P ) is Modus Ponens rule

A ; (A ⇒ B)
(M P ) ,
B
for any formulas A, B ∈ F.
(SB) is a substitution rule

A(x1 , x2 , . . . xn )
(SB) ,
A(t1 , t2 , . . . tn )
where A(x1 , x2 , . . . xn ) ∈ F and t1 , t2 , . . . tn ∈ T.

(G1), (G2) are the following quantifiers introduction rules.

(B ⇒ A(x)) (A(x) ⇒ B)
(G1) , (G2) ,
(B ⇒ ∀xA(x)) (∃xA(x) ⇒ B)

where A(x), B ∈ F and B is such that x is not free in B.

(G3), (G3) are the following quantifiers elimination rules.

(B ⇒ ∀xA(x)) ∃x(A(x) ⇒ B)
(G3) , (G4) ,
(B ⇒ A(x)) A(x) ⇒ B)
where A(x), B ∈ F and B is such that x is not free in B.

The algebraic logic starts from purely logical considerations, abstracts from
them, places them into a general algebraic contest, and makes use of other
branches of mathematics such as topology, set theory, and functional analysis.
For example, Rasiowa and Sikorski algebraic generalization of the completeness
theorem for classical predicate logic is the following.

449
Theorem 9.12 (Rasiowa, Sikorski 1950)
For every formula A of the classical predicate calculus S = {L, C} the following
conditions are equivalent

i A is derivable in RS;

ii A is valid in every realization of L;

iii A is valid in every realization of L in any complete Boolean algebra;

iv A is valid in every realization of L in the field B(X) of all subsets of any

set X 6= ∅;

v A is valid in every semantic realization of L in any enumerable set;

vi there exists a non-degenerate Boolean algebra A and an infinite set J such

that A is valid in every realization of L in J and A;

vii AR (i) = V for the canonical realization R of L in the Lindenbaum-Tarski

algebra LT of S and the identity valuation i;

viii A is a predicate tautology.

9.5 Homework Problems

1. Prove that for any equality axioms (9.5) A and for every structure M =
[M, I] and every s : V AR −→ M , (M, s) |= A.

2. Let H be the proof system defined (9.20). Prove the following.

(i) QA axioms Q1 and Q2 of H are predicate tautologies.
(ii) The rules of inference (G), (G1), (G2) of H are sound.

3. A proof system S is strongly sound if for any rule of inference r of S, a

conjunction of all premisses of r is logically equivalent with its conclusion.
Show that the proof system H is not strongly sound.

4. Prove soundness theorem for Hilbert Ackerman system HA (9.42).

5. Given two proof systems S and K we say that S and K are equivalent
and write it as S ≡ K if they have the same sets of of theorems.
Prove that HA ≡ HAI for HA defined by (9.42) and HAI defined by
(9.43)

6. We know that the Medndelson proof system HM defined by (9.44) is

complete. Prove that HM ≡ H, where H be the proof system defined
(9.20).

450
7. Let RSE be a proof system obtained from RS system defined by (9.45)
by changing the language L{¬,∩,∪,⇒} (P, F, C) of RS to the language with
equality (definition 9.12) and adding Eguality Axioms (9.5) to the set LA
of logocal axioms of RS. Prove Completeness Theorem 9.7 for RSE.
8. Prove Deduction Theorem 9.11 for Mendelson (1973) formalization.

9. In the proof of Deduction Theorem 9.11 for the proof system H we used
gthe completeness of H. Write a proof of the Deduction Theorem 9.11 for
H without use of its completeness.

451
452
Chapter 10

Predicate Automated Proof

Systems
Completeness of Classical
Predicate Logic

We define and discuss here a Rasiowa and Sikorski Gentzen style proof system
QRS for classical predicate logic. The propositional version of it, the RS proof
system, was studied in detail in chapter 6. These both proof systems admit a
constructive proof of completeness theorem. We adopt Rasiowa, Sikorski (1961)
technique of construction a counter model determined by a decomposition tree
to prove QRS completeness theorem 10.4. The proof, presented in section 10.3,
is a generalization of the completeness proofs of RS and other Gentzen style
propositional systems presented in details in chapter 6. We refer the reader to
this chapter as it provides a good introduction to the subject.
The other Gentzen type predicate proof system, including the original Gentzen
proof systems LK, LI for classical and intuitionistic predicate logics are ob-
tained from their propositional versions discussed in detail in chapter 6. It can
be done in a similar way as a generalization of the propositional RS the predicate
QRS system presented here. We leave these generalizations as an exercises for
the reader. That includes also the predicate language version of Gentzen proof
of cut elimination theorem, Hauptzatz (1935). The Hauptzatz proof for the
predicate classical LK and intuitionistic LI systems is easily obtained from the
propositional proof included in chapter6.
There are of course other types of automated proof systems based on different
methods of deduction.
There is a Natural Deduction mentioned by Gentzen in his Hauptzatz paper in

453
1935 and later fully developed by Dag Prawitz (1965). It is now called Prawitz,
or Gentzen-Prawitz Natural Deduction.
There is a Semantic Tableaux deduction method invented by Evert Beth (1955).
It was conequently simplified and further developed by Raymond Smullyan
(1968). It is now often called Smullyan Semantic Tableaux.
Finally, there is a Resolution. The resolution method can be traced back to Davis
and Putnam (1960). Their work is still known as Davis-Putnam method.The
difficulties of their method were eliminated by John Alan Robinson (1965) and
developed into what we call now Robinson Resolution, or just a Resolution.
There are many excellent textbooks covering each of these methods. We recom-
mend Melvin Fitting book First-order logic and automated theorem proving(2nd
ed.). Springer-Verlag(1996) as the one that not only covers all of them but also
discusses their relationships.
The Resolution proof system for propositional or predicate logic operates on
a set of clauses as a basic expressions and uses a resolution rule as the only
rule of inference. In section 10.4 we define and prove correctness of effective
procedures of converting any formula A into a corresponding set of clauses in
both propositional and predicate cases. The correctness of propositional case
is established by theorem 10.5, of predicate case by theorem 10.6. In the first
step of the predicate procedure we define a process of elimination of quantifiers
from the original language. It is called Skolemization of the language and is
presented in section 10.4.1. The correctness of the Skolemization is established
by Skolem theorem 10.11. In the second step of the procedure we show how
convert a quantifiers free formula into logically equivalent set of clauses. It is
presented with a proof of correctness (theorem 10.13) in section 10.4.2.

10.1 QRS Proof System

We define components and semantics of the proof system QRS as follows.

Language L
We adopt a predicate (first order) language

L = L{∩,∪,⇒,¬} (P, F, C) (10.1)

for P, F, C countably infinite sets of predicate, functional, and constant symbols

respectively.
Let F denote a set of formulas of L. The rules of inference of our system QRS
operate on finite sequences of formulas, i.e. elements of F ∗ so we define the set
of expressions of of QRS as follows. Expressions E
We adopt as the set of expressions E of RS the set F ∗ , i.e.

E = F ∗.

454
We will denote the expressions of QRS, i.e. the finite sequences of formulas by

Γ, ∆, Σ, with indices if necessary.

In order to define the axioms LA and the set of rules of inference of QRS we
need to bring back some notions and to introduce some definitions.
An atomic formula of the predicate language L defined by (10.1) is any
element of A∗ (finite strings over the alphabet of L) of the form

R(t1 , t2 , ..., tn )

where R ∈ P, #R = n and t1 , t2 , ..., tn ∈ T.

The set of all atomic formulas is denoted by AF and is defined as

AF = {R(t1 , t2 , ..., tn ) ∈ A∗ : R ∈ P, t1 , t2 , ..., tn ∈ T, n ≥ 1}. (10.2)

We use symbols R, Q, P, ... with indices if necessary to denote the atomic for-
mulas.
Literals
We form a special subset LT ⊆ F of formulas, called a set of all literals, which
is defined as follows.

LT = {R : R ∈ AF} ∪ {¬R : R ∈ AF}. (10.3)

The atomic formulas (10.2) are called positive literals and the elements of the
second set of the above union (10.3), i.e. the negations of the atomic formulas
are called negative literals.
Indecomposable, Decomposable Formulas
A formula A ∈ F is indecomposable if and only if it is atomic or a negation of
an atomic formula, i.e. an literal. Otherwise A is decomposable.
Now we form finite sequences out of formulas (and, as a special case, out of
literals). We need to distinguish the sequences formed out of literals from the
sequences formed out of other formulas, so we adopt the following definition
and notaions.
Indecomposable, Decomposable Sequences
A sequence Γ is indecomposable if and only if is formed out of indecomposable
formulas only. Otherwise is decomposable.
We denote indecomposable sequences by by
0 0 0
Γ , ∆ , Σ , . . . with indices if necessary. (10.4)
0 0 0
By definition, Γ , ∆ , Σ . . . are finite sequences (empty included) formed out
0 0 0 0 0 0
of literals, i.e Γ , ∆ , Σ Γ , ∆ , Σ ∈ LT ∗ .

455
We denote by
Γ, ∆, Σ, . . . with indices if necessary, (10.5)
the elements of F ∗ , i.e. we denote Γ, ∆, Σ finite sequences (empty included)
formed out of elements of F.

Logical Axioms LA
As the logical axiom of QRS we adopt any sequence of formulas which contains
a and its negation, i.e any sequence of the form

Γ1 , A, Γ2 , ¬A0 Γ3 or Γ1 , ¬A, Γ2 , A, Γ3 (10.6)

for any literal A ∈ LT and any sequences Γ1 , Γ2 , Γ3 ∈ F ∗ of formulas.

Rules of inference R (10.7)

Group 1: Propositional Rules

Disjunction rules
0 0 0
Γ , A, B, ∆ Γ , ¬A, ∆ : Γ , ¬B, ∆
(∪) 0 , (¬ ∪)
Γ , (A ∪ B), ∆ Γ0 , ¬(A ∪ B), ∆

Conjunction rules
0 0 0
Γ , A, ∆ ; Γ , B, ∆ Γ , ¬A, ¬B, ∆
(∩) , (¬ ∩)
Γ0 , (A ∩ B), ∆ Γ0 , ¬(A ∩ B), ∆

Implication rules
0 0 0
Γ , ¬A, B, ∆ Γ , A, ∆ : Γ , ¬B, ∆
(⇒) , (¬ ⇒)
Γ0 , (A ⇒ B), ∆ Γ0 , ¬(A ⇒ B), ∆

Negation rule
0
Γ , A, ∆
(¬ ¬) 0
Γ , ¬¬A, ∆
0
where Γ ∈ LT ∗ , ∆ ∈ F ∗ , A, B ∈ F.

Group 2: Quantifiers Rules

456
 (∃)
0
Γ , A(t), ∆, ∃xA(x)
Γ0 , ∃xA(x), ∆
where t is an arbitrary term.

(∀)
0
Γ , A(y), ∆
Γ0 , ∀xA(x), ∆

where y is a free individual variable which

does not appear in any formula in the con-
0
clusion, i.e. in the sequence Γ , ∀xA(x), ∆.

(¬∀)
0
Γ , ∃x¬A(x), ∆
Γ0 , ¬∀xA(x), ∆

(¬∃)
0
Γ , ∀x¬A(x), ∆
Γ0 , ¬∃xA(x), ∆

0
Γ ∈ LT ∗ , ∆ ∈ F ∗ , A, B ∈ F.
Note that A(t), A(y) denotes a formula obtained from A(x) by writing t, y,
respectively, in place of all occurrences of x in A. The variable y in (∀) is called
the eigenvariable. The condition: where y is a free individual variable which
does not appear in any formula in the conclusion is called the eigenvariable
condition.
All occurrences of y in A(y) of the rule (∀) are fully indicated.
The Proof System QRS
Formally we define the proof system QRS as follows.

QRS = (L{¬,⇒,∪,∩} , E, LA, R), (10.8)

where E = {Γ : Γ ∈ F ∗ }, LA contains logical axioms of the system defined by

(10.6), R is the set of rules of inference:

R = {(∪), (¬∪), (∩), (¬∩), (⇒), (¬ ⇒), (¬¬), (¬∀), (¬∃), (∀), (∃))}

defined by (10.7).

457
By a formal proof of a sequence Γ in the proof system RS we understand
any sequence
Γ1 , Γ2 , .... Γn (10.9)
of sequences of formulas (elements of F ∗ , such that
1. Γ1 ∈ LA, Γn = Γ, and
2. for all i (1 ≤ i ≤ n) Γi ∈ AL, or Γi is a conclusion of one of the inference
rules of QRS with all its premisses placed in the sequence Γ1 , Γ2 , .... Γi−1 .
We write, as usual,
`QRS Γ

to denote that Γ has a formal proof in QRS.

As the proofs in QRS are sequences (definition of the formal proof) of sequences
of formulas (definition of expressions E) we will not use ”,” to separate the steps
of the proof, and write the formal proof as Γ1 ; Γ2 ; .... Γn .
We write, however, the formal proofs in QRS as we did the propositional case
(chapter 6), in a form of trees rather then in a form of sequences, ie. in a form
of a tree, where leafs of the tree are axioms, nodes are sequences such that
each sequence on the tree follows from the ones immediately preceding it by one
of the rules. The root is a theorem. We picture, and write our tree-proofs with
the node on the top, and leafs on the very bottom, instead of more common
way, where the leafs are on the top and root is on the bottom of the tree. We
adopt hence the following definition.

Definition 10.1 (Proof Tree)

By a proof tree, or QRS-proof of Γ we understand a tree TΓ of sequences sat-
isfying the following conditions:
1. The topmost sequence, i.e the root of TΓ is Γ,
2. all leafs are axioms,
3. the nodes are sequences such that each sequence on the tree follows from the
ones immediately preceding it by one of the rules of inference (10.7).

We picture, and write our proof trees with the node on the top, and leafs on
the very bottom, instead of more common way, where the leafs are on the top
and root is on the bottom of the tree.
In particular cases, as in the propositional case, we will write our proof- trees
indicating additionally the name of the inference rule used at each step of the
proof. For example, if in a proof of a formula A from axioms (10.6) we use
subsequently the rules

(∩), (∪), (∀), (∩), (¬¬), (∀), (⇒)

458
we represent the proof as the following tree denoted by TA .

TA

F ormula A

| (⇒)
conclusion of (∀)
| (∀)
conclusion of (¬¬)
| (¬¬)
conclusion of (∩)
^
(∩)

conclusion of (∀) conclusion of (∪)

| (∀) | (∪)
axiom conclusion of (∩)
^
(∩)
axiom axiom

Remark that the derivation trees don’t represent a different definition of a

formal proof. This remains the same in the Gentzen - style systems. Trees
represent a certain visualization for those proofs and any formal proof in any
system can be represented in a tree form.

10.2 QRS Decomposition Trees

The main advantage of the Gentzen proof systems lies not in a way we generate
proofs in them, but in the way we can search for proofs in them. That such proof
searches happens to be deterministic and automatic. We conduct such search
by treating inference rules as decomposition rules (see chapter 6) and building
decomposition trees.A general principle of building a decomposition tree is the
following.
Decomposition Tree TΓ
For each Γ ∈ F ∗ , a decomposition tree TΓ is a tree build as follows.
Step 1. The sequence Γ is the root of TΓ and for any node ∆ of the tree we
follow the steps bellow.

459
Step 2. If ∆ is indecomposable or an axiom, then ∆ becomes a leaf of the
tree.
Step 3. If ∆ is decomposable, then we traverse ∆ from left to right to identify
the first decomposable formula B and identify inference rule treated as de-
composition rule determined uniquely by B. We put its left and right premisses
as the left and right leaves, respectively.
Step 4. We repeat steps 2 and 3 until we obtain only leaves or infinite branch.
In particular case when when Γ has only one element, namely a a formula A ∈ F,
we define we call it a decomposition tree of A and denote by TA .
Here is a detailed definition of the decomposition tree for QRS.

QRS Decomposition Tree Definition (10.10)

Given a formula A ∈ F, we define its decomposition tree TA as follows.

Observe that the inference rules of QRS are divided in two groups: propo-
sitional connectives rules and quantifiers rules. The propositional connectives
rules are: (∪), (¬∪), (∩), (¬∩), (⇒), (¬ ⇒), and (¬¬). The quantifiers rules are:
(∀), (∃), (¬∀) and (¬∃).
We define the decomposition tree in the case of the propositional rules and
the rules (¬∀), (¬∃) in the exactly the same way as in the propositional case
(chapter 6).
The case of the rules (∀) and (∃) is more complicated, as the rules contain the
specific conditions under which they are applicable.
0
To define the way of decomposing the sequences of the form Γ , ∀xA(x), ∆ or
0
Γ , ∃xA(x), ∆, i.e. to deal with the rules (∀) and (∃) in a way which would
preserve the property of the uniqueness of the decomposition tree, we assume
that all terms form a one-to one sequence
t1 , t2 , ...., tn , ...... (10.11)
Observe, that by the definition, all free variables are terms, hence all free vari-
ables appear in the sequence 10.11. Let Γ be a sequence on the tree with ∀ as
0
a main connective, i.e. Γ is of the form Γ , ∀xA(x), ∆. We write a sequence
0
Γ , A(x), ∆ below it on the tree, as its child, where the variable x has to fulfill
the following condition.

Condition 10.1 (∀)

x is the first free variable in the sequence 10.11 such that x does not appear in
0
any formula in Γ , ∀xA(x), ∆.

Observe, that the condition 10.1 corresponds to the restriction put on the ap-
plication of the rule (∀).

460
If the main connective of Γ, i.e. the main connective of the first formula in Γ
0
which is not an literal, is (∃). In this case Γ is of the form Γ , ∃xA(x), ∆, we
0
write a sequence Γ , A(t), ∆, ∃xA(x) as its child, where the term t has to fulfill
the following condition.

Condition 10.2 (∃)

t is the first term in the sequence 10.11 such that the formula A(t) does not
0
appear in any sequence which is placed above Γ , A(t), ∆, ∃xA(x) on the tree.

The fact that the sequence 10.11 is one- to - one and the fact that, by the
conditions 10.1 and 10.2, we always chose the first appropriate term (variable)
from this sequence, guarantee that the decomposition process is also unique in
the case of the quantifiers rules (∀) and (∃).

From all above, and we conclude the following.

Theorem 10.1
For any formula A ∈ F,
(i) the decomposition tree TA is unique.
(ii) Moreover, the following conditions hold.
1. If TA is finite and all its leaves are axioms, then
`QRS A
and TA is a tree-proof of A in QRS.
2. If TA is finite and contains a non-axiom leaf, or TA is infinite, then
6 `QRS A.

10.2.1 Examples of Decomposition Trees

In all the examples below, the formulas A(x), B(x) represent any formula. But
there is no indication about their particular components, so they are treated as
indecomposable formulas.

Example 10.1
The decomposition tree TA of the de Morgan Law
(¬∀xA(x) ⇒ ∃x¬A(x))
is the following.

461
 TA

(¬∀xA(x) ⇒ ∃x¬A(x))

| (⇒)
¬¬∀xA(x), ∃x¬A(x)
| (¬¬)
∀xA(x), ∃x¬A(x)
| (∀)
A(x1 ), ∃x¬A(x)
where x1 is a first free variable in the sequence 10.11 such that x1 does not appear in
∀xA(x), ∃x¬A(x)

| (∃)

A(x1 ), ¬A(x1 ), ∃x¬A(x)

where x1 is the first term (variables are terms) in the sequence 10.11 such that ¬A(x1 ) does
not appear on a tree above A(x1 ), ¬A(x1 ), ∃x¬A(x)
Axiom

The above tree TA ended with an axiom, so it represents a proof of

(¬∀xA(x) ⇒ ∃x¬A(x))

in QRS, i.e. we proved that

`QRS (¬∀xA(x) ⇒ ∃x¬A(x)).

Example 10.2
The decomposition tree TA of

(∀xA(x) ⇒ ∃xA(x))

is the following.

TA

(∀xA(x) ⇒ ∃xA(x))

| (⇒)

462
 ¬∀xA(x), ∃xA(x)
| (¬∀)
¬∀xA(x), ∃xA(x)
∃x¬A(x), ∃xA(x)
| (∃)
¬A(t1 ), ∃xA(x), ∃x¬A(x)
where t1 is the first term in the sequence 10.11, such that ¬A(t1 ) does not appear on the tree
above ¬A(t1 ), ∃xA(x), ∃x¬A(x)

| (∃)
¬A(t1 ), A(t1 ), ∃x¬A(x), ∃xA(x)
where t1 is the first term in the sequence 10.11, such that A(t1 ) does not appear on the tree
above ¬A(t1 ), A(t1 ), ∃x¬A(x), ∃xA(x)
Axiom

The above tree also ended with the axiom, hence we proved that
`( QRS) (∀xA(x) ⇒ ∃xA(x)).

Example 10.3
The decomposition tree TA of
(∃xA(x) ⇒ ∀xA(x))
is the following.

TA
(∃xA(x) ⇒ ∀xA(x))
| (⇒)
¬∃xA(x), ∀xA(x)
| (¬∃)
∀x¬A(x), ∀xA(x)
| (∀)
¬A(x1 ), ∀xA(x)
where x1 is a first free variable in 10.11 such that x1 does not appear in ∀x¬A(x), ∀xA(x)

| (∀)
¬A(x1 ), A(x2 )
where x2 is a first free variable in 10.11 such that x2 does not appear in ¬A(x1 ), ∀xA(x), the
sequence 10.11 is one-to- one, hence x1 6= x2
Non - axiom

463
The decomposition tree, for any formula A is unique, so we conclude from the
fact that the above tree has a non-axiom branch that

6 `QRS (∃xA(x) ⇒ ∀xA(x)).

Remark when constructing the following tree TA for the formula ∃xA(x) in
example 10.4 below we adopt on the right branch of the a tree in the the short-
hand notation instead of the repeating a similar reasoning performed on the left
branch.

Example 10.4
The decomposition tree TA of the formula ∃xA(x) is the following.

TA

∃xA(x)

| (∃)
A(t1 ), ∃xA(x)
where t1 is the first term in the sequence 10.11, such that A(t1 ) does not appear on the tree
above A(t1 ), ∃xA(x)

| (∃)

A(t1 ), A(t2 ), ∃xA(x)

where t2 is the first term in the sequence 10.11, such that A(t2 ) does not appear on the tree
above A(t1 ), A(t2 ), ∃xA(x), i.e. t2 6= t1

| (∃)

A(t1 ), A(t2 ), A(t3 ), ∃xA(x)

where t3 is the first term in the sequence 10.11, such that A(t3 ) does not appear on the tree
above A(t1 ), A(t2 ), A(t3 ), ∃xA(x), i.e. t3 6= t2 6= t1

| (∃)

A(t1 ), A(t2 ), A(t3 ), A(t4 ), ∃xA(x)

| (∃)
.....
| (∃)
.....
inf inite branch

464
Obviously, the above decomposition tree is infinite, what proves that

6` QRS ∃xA(x).

We will find now the proof of the distributivity law

(∃x(A(x) ∩ B(x)) ⇒ (∃xA(x) ∩ ∃xB(x)))

and show that we can’t prove in QRS the inverse implication

((∃xA(x) ∩ ∃xB(x)) ⇒ ∃x(A(x) ∩ B(x))).

Remark when constructing the following trees TA in examples 10.5, 10.6 adopt,
as we did in the previous example 10.4, the shorthand notation when the rea-
soning is similar to the one presented in the example 10.4.

Example 10.5
The decomposition tree A of the first formula

(∃x(A(x) ∩ B(x)) ⇒ (∃xA(x) ∩ ∃xB(x)))

is the following.

TA

(∃x(A(x) ∩ B(x)) ⇒ (∃xA(x) ∩ ∃xB(x)))

| (⇒)
¬∃x(A(x) ∩ B(x)), (∃xA(x) ∩ ∃xB(x))
| (¬∃)
∀x¬(A(x) ∩ B(x)), (∃xA(x) ∩ ∃xB(x))
| (∀)
¬(A(x1 ) ∩ B(x1 )), (∃xA(x) ∩ ∃xB(x))
where x1 is a first free variable in the sequence 10.11 such that x1 does not appear in
∀x¬(A(x) ∩ B(x)), (∃xA(x) ∩ ∃xB(x))

| (¬∩)

¬A(x1 ), ¬B(x1 ), (∃xA(x) ∩ ∃xB(x))

^
(∩)

465
 ¬A(x1 ), ¬B(x1 ), ∃xA(x) ¬A(x1 ), ¬B(x1 ), ∃xB(x)
| (∃) | (∃)
¬A(x1 ), ¬B(x1 ), A(t1 ), ∃xA(x) ¬A(x1 ), ¬B(x1 ), B(t1 ), ∃xB(x)
where t1 is the first term in the sequence 10.11, | (∃)
such that A(t1 ) does not appear on the tree
...
above ¬A(x1 ), ¬B(x1 ), A(t1 ), ∃xA(x) Observe,
that it is possible that t1 = x1 , as A(x1 ) does | (∃)
not appear on the tree above. By the definition
¬A(x1 ), ¬B(x1 ), ...B(x1 ), ∃xB(x)
of the sequence 10.11, x1 is placed somewhere
in it, i.e. x1 = ti , for certain i ≥ 1. It means
that after i applications of the step (∃) in the
decomposition tree, we will get a step:

| (∃)

¬A(x1 ), ¬B(x1 ), ...A(x1 ), ∃xA(x)

All leaves of the above tree TA are axioms, what means that we proved

`QRS (∃x(A(x) ∩ B(x)) ⇒ (∃xA(x) ∩ ∃xB(x))).

We construct now, as the last example, a decomposition tree TA of the formula

((∃xA(x) ∩ ∃xB(x)) ⇒ ∃x(A(x) ∩ B(x))).

Example 10.6
The decomposition tree of the formula

((∃xA(x) ∩ ∃xB(x)) ⇒ ∃x(A(x) ∩ B(x)))

is the following.

TA

((∃xA(x) ∩ ∃xB(x)) ⇒ ∃x(A(x) ∩ B(x)))

| (⇒)
¬(∃xA(x) ∩ ∃xB(x))∃x(A(x) ∩ B(x))
| (¬∩)
¬∃xA(x), ¬∃xB(x), ∃x(A(x) ∩ B(x))
| (¬∃)
∀x¬A(x), ¬∃xB(x), ∃x(A(x) ∩ B(x))

466
 | (∀)

¬A(x1 ), ¬∃xB(x), ∃x(A(x) ∩ B(x))

| (¬∃)

¬A(x1 ), ∀x¬B(x), ∃x(A(x) ∩ B(x))

| (∀)

¬A(x1 ), ¬B(x2 ), ∃x(A(x) ∩ B(x))

By the reasoning similar to the reasonings in the previous examples we get that x1 6= x2

| (∃)

¬A(x1 ), ¬B(x2 ), (A(t1 ) ∩ B(t1 )), ∃x(A(x) ∩ B(x))

where t1 is the first term in the sequence 10.11, such that (A(t1 ) ∩ B(t1 )) does not appear on
the tree above ¬A(x1 ), ¬B(x2 ), (A(t1 ) ∩ B(t1 )), ∃x(A(x) ∩ B(x)) Observe, that it is possible that
t1 = x1 , as (A(x1 ) ∩ B(x1 )) does not appear on the tree above. By the definition of the
sequence 10.11, x1 is placed somewhere in it, i.e. x1 = ti , for certain i ≥ 1. For simplicity, we
assume that t1 = x1 and get the sequence:

¬A(x1 ), ¬B(x2 ), (A(x1 ) ∩ B(x1 )), ∃x(A(x) ∩ B(x))

^
(∩)

467
 ¬A(x1 ), ¬B(x2 ), ¬A(x1 ), ¬B(x2 ),
A(x1 ), ∃x(A(x) ∩ B(x)) B(x1 ), ∃x(A(x) ∩ B(x))
Axiom | (∃)
¬A(x1 ), ¬B(x2 ), B(x1 ),
(A(x2 ) ∩ B(x2 )), ∃x(A(x) ∩ B(x))
where x2 = t2 (x1 6= x2 ) is the
first term in the sequence 10.11, such that
(A(x2 ) ∩ B(x2 )) does not appear on the
tree above ¬A(x1 ), ¬B(x2 ), (B(x1 ), (A(x2 ) ∩
B(x2 )), ∃x(A(x) ∩ B(x)). We assume that t2 =
x2 for the reason of simplicity.
^
(∩)

¬A(x1 ), ¬A(x1 ),
¬B(x2 ), ¬B(x2 ),
B(x1 ), A(x2 ), B(x1 ), B(x2 ),
∃x(A(x) ∩ B(x)) ∃x(A(x) ∩ B(x))
| (∃) Axiom
...
^
(∩)
...
| (∃)
...
| (∃)
inf inite branch

The above decomposition tree TA contains an infinite branch what means that

6`QRS ((∃xA(x) ∩ ∃xB(x)) ⇒ ∃x(A(x) ∩ B(x))).

10.3 Proof of QRS Completeness

Our main goal is to prove the Completeness Theorem for QRS. The proof of
completeness theorem presented here is due to Rasiowa and Sikorski (1961), as

468
is the proof system QRS. We adopted their proof to propositional case in chap-
ter 6.The completeness proofs, in the propositional case and in predicate case,
are constructive as they are based on a direct construction of a counter model
for any unprovable formula. The construction of the counter model for the un-
provable formula A uses the decomposition tree TA . We call such constructed
counter model a counter model determined by the tree TA . Rasiowa-Sikorski
type of constructive proofs of counter models determined by the tree decompo-
sition trees relay heavily of the notion of a strong soundness. We define it here
(definition 10.8), adopting chapter 4 general definition to our case.
Given a first order language L (10.1) with the set V AR of variables and the set
F of formulas. We define, after chapter 8 a notion of a model and a counter-
model of a formula A of L and then extend it to the the set F ∗ establishing the
semantics for QRS.

Definition 10.2 (Model)

A structure M = [M, I] is called a model of A ∈ F if and only if
(M, v) |= A
for all assignments v : V AR −→ M .
We denote it by
M |= A.

M is called the universe of the model, I the interpretation.

Definition 10.3 (Counter - Model)

A structure M = [M, I] is called a counter- model of A ∈ F if and only if there
is v : V AR −→ M , such that
(M, v) 6|= A.
We denote it by
M 6|= A.

The definition of the first order logic tautology is the following.

Definition 10.4 (Tautology)

For any A ∈ F, A is called a (predicate) tautology and denoted by
|= A
if and only if all structures M = [M, I] are models of A, i.e.
|= A if and only if M |= A
for all structures M = [M, I] for L.

469
Directly from the above definition we get the following, simple fact.

Fact 10.1 (Counter Model)

For any A ∈ F, A is not a tautology (6|= A) if and only if there is a counter -
model M = [M, I] of A, i.e. we can define M, I, and v such that ([M, I], v) 6|= A.

Definition 10.5
For any sequence Γ ∈ F ∗ , by
δΓ
we understand any disjunction of all formulas of Γ.

Definition 10.6 (QRS Semantics)

A structure M = [M, I] for L is called a model of a Γ ∈ F ∗ and denoted by

M |= Γ

if and only if
M |= δΓ .
The sequence Γ is a predicate tautology if and only if the formula δΓ is a predicate
tautology, i.e.
|= Γ if and only if |= δΓ .

Our goal now is to prove the completeness theorem for QRS.The correctness of
the proof we present depends on the strong soundness of the rules of inference
of rules of inference defined as follows.

Definition 10.7 ( Strongly Sound Rules)

Given a predicate language (10.1) proof system S = (L, E, LA, R) An inference
rule r ∈ R of the form

P1 ; P2 ; .... ; Pm
(r)
C
is strongly sound if the following condition holds for and structure M = [M, I]
for L.
M |= {P1 , P2 , .Pm } if and only if M |= C. (10.12)

We say it less formally that a rule (r) is strongly sound if the conjunction of
its premisses is logically equivalent with the conclusion, i.e.

P1 ∩ P2 ∩ . . . ∩ Pm ≡ C. (10.13)

470
Definition 10.8 (Strongly Sound System)
A predicate language (10.1) proof system S = (L, E, LA, R) is strongly sound
if and only if all logical axioms LA are tautologies and all its rules of inference
r ∈ R are strongly sound.

Theorem 10.2 (Strong Soundness)

The proof system QRS (10.8) is strongly sound.

Proof
We have already proved in chapter 6 strong soundness of the propositional rules.
The quantifiers rule are strongly sound by straightforward verification and is left
as an exercise.
The strong soundness property is stronger then soundness property, hence also
the following holds.

Theorem 10.3 (Soundness Theorem)

For any Γ ∈ F ∗ ,
if `QRS Γ, then |= Γ.
In particular, for any formula A ∈ F,

if `QRS A, then |= A.

Theorem 10.4 (Completeness Theorem)

For any Γ ∈ F ∗ ,
`QRS Γ if and only if |= Γ.
In particular, for any formula A ∈ F,

`QRS A if and only if |= A.

Proof
We have to prove the inverse implication to the soundness theorem 10.3. We
need to prove the formula A case only because the case of a sequence Γ can be
reduced to the formula case. Namely, the disjunction of all formulas in Γ. I.e.
we prove the implication:

if |= A, then `QRS A.

We do it, as in the propositional case, by proving the opposite implication

if 6`QRS A then 6|= A.

This means that we want prove that for any formula A, unprovability of A in
QRS (6`QRS A ), allows us to define its counter- model. The counter- model

471
is determined, as in the propositional case, by the decomposition tree TA . By
theorem 10.1 each formula A, generates its unique decomposition tree TA and
A has a proof only if this tree is finite and all its end sequences (leaves) are
axioms. Moreover, it says that we have two cases to consider:
(C1) the tree TA is finite and contains a leaf which is not axiom, or
(C2) the tree TA is infinite.
We will show how to construct a counter- model for A in both cases: a counter-
model determined by a non-axiom leaf of the decomposition tree TA , or a
counter- model determined by an infinite branch of TA .
Proof in case (C1): TA is finite and contains a non- axiom leaf.

Before describing a general method of constructing the counter-model deter-

mined by the decomposition tree TA we describe it, as an example, for a case of
a formula
(∃xA(x) ⇒ ∀xA(x)),
and its particular case

(∃x(P (x) ∩ R(x, y)) ⇒ ∀x(P (x) ∩ R(x, y))) (10.14)

for P , R one and two argument predicate symbols, respectively.

We construct the counter model for the formula (10.14) as follows.
First we build its decomposition tree:

TA

(∃x(P (x) ∩ R(x, y)) ⇒ ∀x(P (x) ∩ R(x, y)))

| (⇒)
¬∃x(P (x) ∩ R(x, y)), ∀x(P (x) ∩ R(x, y))
| (¬∃)
∀x¬(P (x) ∩ R(x, y)), ∀x(P (x) ∩ R(x, y))
| (∀)
¬(P (x1 ) ∩ R(x1 , y)), ∀x(P (x) ∩ R(x, y))
where x1 is a first free variable in 10.11 such that x1 does not appear in
∀x¬(P (x) ∩ R(x, y)), ∀x(P (x) ∩ R(x, y))

| (¬∩)

¬P (x1 ), ¬R(x1 , y), ∀x(P (x) ∩ R(x, y))

472
 | (∀)
¬P (x1 ), ¬R(x1 , y), (P (x2 ) ∩ R(x2 , y))
where x2 is a first free variable in the sequence 10.11 such that x2 does not appear in
¬P (x1 ), ¬R(x1 , y), ∀x(P (x) ∩ R(x, y)), the sequence 10.11 is one-to- one, hence x1 6= x2

^
(∩)

¬P (x1 ), ¬R(x1 , y), P (x2 ) ¬P (x1 ), ¬R(x1 , y), R(x2 , y)

x1 6= x2 , Non-axiom x1 6= x2 , Non-axiom

There are two non-axiom leaves. In order to define a counter-model for (10.14)
determined by the tree TA we need to chose only one of them. Let’s choose the
leaf
LA = ¬P (x1 ), ¬R(x1 , y), P (x2 ). (10.15)

We use the non-axiom leaf LA to define a structure M = [M, I] and an

assignment v, such that
(M, v) 6|= A.
Such defined M is called a counter - model determined by the tree TA .
We take a the universe of M the set T of all terms of our language L, i.e. we
put
M = T.
We define the interpretation I as follows. For any predicate symbol Q ∈
P, #Q = n we put that QI (t1 , . . . tn ) is true (holds) for terms t1 , . . . tn if and
only if the negation ¬QI (t1 , . . . tn ) of the formula Q(t1 , . . . tn ) appears on the
leaf LA and QI (t1 , . . . tn ) is false (does not hold) for terms t1 , . . . tn otherwise.
For any functional symbol f ∈ F, #f = n we put fI (t1 , . . . tn ) = f (t1 , . . . tn ).
It is easy to see that in particular case of our non-axiom leaf (10.15)

LA = ¬P (x1 ), ¬R(x1 , y), P (x2 )

PI (x1 ) is true for x1 , and not true for x2 . RI (x1 , y) is true (holds) holds for x1
any for any y ∈ V AR.
We define the assignment v : V AR −→ T as identity, i.e., we put v(x) = x for
any x ∈ V AR.
Obviously, for such defined structure [M, I] and the assignment v we have that

([T, I], v) |= P (x1 ), ([T, I], v) |= R(x1 , y), and ([T, I], v) 6|= P (x2 ).

473
We hence obtain that

([T, I], v) 6|= ¬P (x1 ), ¬R(x1 , y), P (x2 ).

This proves that such defined structure [T, I] is a counter model for a non-axiom
leaf (10.15). By the strong soundness of QRS) (theorem 10.2) the structure
M = [T, I] is also a counter- model for the formula (10.14), i.e. we proved that

6|= (∃x(P (x) ∩ R(x, y)) ⇒ ∀x(P (x) ∩ R(x, y))).

C1: General Method

Let A be any formula such that 6`QRS A.
Let TA be a decomposition tree of A. By the fact that 6`QRS and C1, the tree
TA is finite and has a non axiom leaf

LA ⊆ LT ∗ . (10.16)

By definition, the leaf LA contains only atomic formulas and negations of atomic
formulas.
We use the non-axiom leaf LA (10.16) to define a structure M = [M, I] an
assignment v : V AR −→ M , such that (M, v) 6|= A. Such defined structure M
is called a counter - model determined by the tree TA .

Structure M Definition (10.17)

Given LA . We define a structure

M = [M, I] (10.18)

and an assignment v : V AR −→ M as follows.

1. We take a the universe of M the set T of all terms of our language L’ i.e.
we put
M = T.

2. For any predicate symbol Q ∈ P, #Q = n,

QI ⊆ Tn

is such that QI (t1 , . . . tn ) holds (is true) for terms t1 , . . . tn if and only if the
negation ¬Q(t1 , . . . tn ) of the formula Q(t1 , . . . tn ) appears on the leaf LA and
QI (t1 , . . . tn ) does not hold (is false) for terms t1 , . . . tn otherwise.
3. For any constant c ∈ C, we put cI = c, for any variable x, xI = x.
For any functional symbol f ∈ F, #f = n,

fI : Tn −→ T

474
is identity function, i.e. we put

fI (t1 , . . . tn ) = f (t1 , . . . tn )

for all t1 , . . . tn ∈ T.
4. We define the assignment v : V AR −→ T as identity, i.e. we put for all
x ∈ V AR v(x) = x.
Obviously, for such defined structure [T, I] and the assignment v we have that

([T, I], v) 6|= P if formula P appears in LA ,

([T, I], v) |= P if formula ¬P appears in LA .

This proves that the structure M = [T, I] and assignment v defined by (10.18)
are such that
([T, I], v) 6|= LA .
By the strong soundness (theorem 10.2) of QRS

(([T, I], v) 6|= A.

This proves M 6|= A and we proved that

6|= A.

This ends the proof of the case C1.

Proof in case (C2): TA is infinite.

The case of the infinite tree is similar, even if a little bit more complicated.
Observe first that the rule (∃) is the t rule of inference (decomposition) which
can ”produces” an infinite branch. We first show how to construct the counter-
model in the case of the simplest application of this rule, i.e. in the case of the
formula
∃xP (x)
where P is an one argument relational symbol. All other cases are similar to
this one. The infinite branch BA in this case consists of all elements of the
decomposition tree:

TA

∃xP (x)

| (∃)
P (t1 ), ∃xP (x)

475
 where t1 is the first term in the sequence 10.11, such that P (t1 ) does not appear on the tree
above P (t1 ), ∃xP (x)

| (∃)

P (t1 ), P (t2 ), ∃xP (x)

where t2 is the first term in the sequence 10.11, such that P (t2 ) does not appear on the tree
above P (t1 ), P (t2 ), ∃xP (x), i.e. t2 6= t1

| (∃)

P (t1 ), P (t2 ), P (t3 ), ∃xP (x)

where t3 is the first term in the sequence 10.11, such that P (t3 ) does not appear on the tree
above P (t1 ), P (2 ), P (t3 ), ∃xP (x), i.e. t3 6= t2 6= t1

| (∃)

P (t1 ), P (t2 ), P (t3 ), P (t4 ), ∃xP (x)

| (∃)
.....
| (∃)
.....

The infinite branch of TA , written from the top, in oder of appearance of for-
mulas is
BA = {∃xP (x), P (t1 ), A(t2 ), P (t2 ), P (t4 ), .....}

where t1 , t2 , .... is a one - to one sequence (10.11) of all elements of the set T of
all terms.
This means that the infinite branch B contains with the formula ∃xP (x) all its
instances P (t), for all terms t ∈ T.
We define the structure M = [M, I] and valuation v following the definition
10.17. We take as the universe M the set T of all terms, and now in our case
we define PI as follows: PI (t) holds if ¬P (t) ∈ BA and PI (t) does not hold if
P (t) ∈ BA .
It is easy to see that for any formula P (t) ∈ B,

([T, I], v) 6|= P (t).

But the A(t) ∈ B are all instances ∃xA(x), hence

([T, I], v) 6|= ∃xA(x).

476
C1: General Method
Let A be any formula such that 6`QRS A.
Let TA be an infinite decomposition tree of a formula A. Let BA the infinite
¯
branch of TA , written from the top, in oder of appearance of sequences Γ ∈ F ∗
on it, where Γ0 = A.

BA = {Γ0 , Γ1 , Γ2 , . . . Γi , Γi+1 , . . . } (10.19)

We define a set LF ⊆ F of all indecomposable formulas appearing in at least

one Γi , i ≤ j, i.e.

LF = {B ∈ LT : there is Γi ∈ BA , such that B is in Γi }. (10.20)

Note, that the following holds.

(1) if i ≤ i0 and an indecomposable formula appears in Γi , then it also appears
in Γi0 .
(2) Since none of Γi , is an axiom (10.6), for every atomic formula (10.2) P ∈ AF,
at most one of the formulas P and ¬P is in LF (10.20).
Counter Model Definition
Let T be the set of all terms. We define the structure M = [T, I] and valuation
v in the set T as in the definition 10.17, with the interpretation of predicates
Q ∈ Q defined as follows.
For any predicate symbol Q ∈ P, #Q = n, QI ⊆ Tn is such that:
(1) QI (t1 , . . . tn ) does not hold (is false) for terms t1 , . . . tn if and only if

QI (t1 , . . . tn ) ∈ LF

and
(2) QI (t1 , . . . tn ) does holds (is true) for terms t1 , . . . tn if and only if

QI (t1 , . . . tn ) 6∈ LF.

This proves that the structure M = [T, I] is such that

M 6|= LF. (10.21)

To prove that 6|= A it suffices that

M 6|= A. (10.22)

For this purpose we first introduce, for any formula A ∈ F, an inductive defini-
tion of the order ord A of the formula A.
(1) If A ∈ AF, then ord A = 1.
(2) If ord A = n, then ord ¬A = n + 1. (3) If ord A ≤ n and ord B ≤ n,
then ord (A ∪ B) = ord (A ∩ B) = ord (A ⇒ B) = n + 1.
(4) If ord A(x) = n, then ord ∃xA(x) = ord ∀xA(x) = n + 1.

477
We conduct the proof of (10.23) by contradiction. Suppose that (10.23) does
not hold, i.e. assume that
M |= A. (10.23)

Consider now a set M F of all formulas B appearing in one of the sequences Γi

of the branch BA , such that
M |= B. (10.24)
We write the the set M F formally as follows.

M F = {B ∈ F : for some Γi ∈ BA , B is in Γi and M |= B}. (10.25)

Observe that by assumption (10.23) and the definition (10.25), the formula A
is in M F and hence M F 6= ∅.
Let B 0 be a formula in M F such that ord B 0 ≤ ord B for every B ∈ M F. There
exists Γi ∈∈ BA that is of the form Γ0 , B 0 , ∆ with an indecomposable Γ0 .
We have that B 0 can not be of the form

¬∃xA(x) or ¬∀xA(x) (10.26)

for if (refn-Q) is in M F, then also formula ∀x¬A(x) or ∃x¬A(x) is in M F and

the orders of the two formulas are equal.
We carry the same order argument and show that B 0 can not be of the form

(A ∪ B), ¬(A ∪ B), (A ∩ B), ¬(A ∩ B), (A ⇒ B), ¬(A ⇒ B), ¬¬A, ∀xA(x).
(10.27)
The formula B 0 can’t be of the form

∃xB(x) (10.28)

since then there exists term t and j such that i ≤ j, B 0 (t) appears in Γj and the
formula B(t) satisfies (10.24). Thus B(t) ∈ M F and ordB(t) < ordB 0 . This
contradicts the definition of B 0 .
Since B 0 is not of the form (10.26), (10.27), (10.28), B 0 is indecomposable. Thus
B 0 ∈ LF (10.20), and consequently by (10.21),

M 6|= B 0 .

On the other hand B 0 by definition is in the set M F and hence is one o the
formulas satisfying (10.24), i.e.

M 6|= B 0 .

This contradiction proves that (10.23) M 6|= A and hence we proved

6|= A.

This ends the proof of the Completeness Theorem 10.4 for QRS.

478
10.4 Skolemization and Clauses
The resolution proof system for propositional and predicate logic operates on a
set of clauses as a basic expressions and uses a resolution rule as the only
rule of inference.
The goal of this part is to define an effective process of transformation of
any formula A of a predicate language L = L{¬,∪,∩,⇒} (P, F, C) into a certain
corresponding set of clauses CA . This is done in two stages.
S1. We convert any formula A of L into an open formula A∗ of a language L∗ by
a process of elimination of quantifiers from the original L. The method is due
to T. Skolem (1920) and is called Skolemization. The resulting formula A∗ is
equisatisfiable with A: it is satisfiable if and only if the original one is satisfiable
(Skolem Theorem 10.11).
The stage S1. is performed as the first step in a Resolution based automated
theorem prover and is described in section 10.4.1.
S2. We define a proof system QRS∗ based on the language L∗ and use it
transform any formula A∗ of L∗ into an logically equivalent set of clauses CA∗
(theorem 10.13).
The final result of stages S1. and S1 is the set CA of clauses corresponding to
the formula A, called a clausal form of A (theorem 10.6.
The transformation process for any propositional formula A into its logically
equivalent set CA of clauses follows directly from the use of the propositional
system RS (theorem 10.5).
.

Definition 10.9 (Clauses)

Given a language L, propositional or predicate.
1. A literal as an atomic, or a negation of an atomic formula of L. We denote
by LT the set of all literals of L.
2. A clause C is a finite set of literals.Empty clause is denoted by {}.
3. We denote by C any finite set of all clauses.
C = {C1 , C2 , . . . Cn },
for any n ≥ 0.

Definition 10.10
Given a propositional or predicate language L, and a sequence Γ ∈ LT ∗ . A
clause determined by Γ is a set form out of all elements of the sequence Γ
We we denote it by CΓ .

479
Example 10.7
In particular,
1. if Γ1 = a, a, ¬b, c, ¬b, c and Γ2 = ¬b, c, a, then CΓ1 = CΓ2 {a, c, ¬b}.
2. If Γ1 = ¬P (x1 ), ¬R(x1 , y), P (x2 ), ¬P (x1 ), ¬R(x1 , y), P (x2 ) and
Γ2 = ¬P (x1 ), ¬R(x1 , y), P (x2 ), then CΓ1 = CΓ2 {¬P (x1 ), ¬R(x1 , y), P (x2 )}.

The semantics for clauses is basically the same as for the sequences. We define
it as follows.

Definition 10.11 Given a propositional or predicate language L. For any

clause C, write δC for a disjunction of all literals in C.

Definition 10.12 (Clauses Semantics)

Let M = [M, I] be a structure for a predicate language L, or a truth assignment
v in case of L propositional.
M is called a model for a clause C (predicate or propositional), (M |= C) if
and only if
M |= δC .
M is called a model for a set C of clauses (M |= C) if and only if
M |= δC for all clauses C ∈ C.

Definition 10.13 (Equivalence)

A formula A of a language L is equivalent with a set set C of clauses (A ≡ C)
if and only if A ≡ σC , where σC is a conjunction of all formulas δC for all
clauses C ∈ C.

Theorem 10.5 (Formula-Clauses Equivalency)

For any formula A of a propositional language L, there is an effective procedure
of generating a corresponding set CA of clauses such that
A ≡ CA (10.29)

Proof
Let L = L{¬,∪,∩,⇒} . Given A ∈ F, we use the RS system (chapter 6) to build
the decomposition tree TA . We form clauses out of the leaves of the tree TA ,
i.e. for every leaf L we create a clause CL determined by L (definition 10.10).
We put
CA = {CL : L is a leaf of TA }.
Directly from the strong soundness (10.13) of rules of inference of RS and the
definition 10.13 we get A ≡ CA . This ends the proof for the propositional case.
Consider a decomposition tree of a formula (((a ⇒ b) ∩ ¬c) ∪ (a ⇒ c))

480
 TA

(((a ⇒ b) ∩ ¬c) ∪ (a ⇒ c))

| (∪)
((a ⇒ b) ∩ ¬c), (a ⇒ c)
^
(∩)

(a ⇒ b), (a ⇒ c) ¬c, (a ⇒ c)
| (⇒) | (⇒)
¬a, b, (a ⇒ c) ¬c, ¬a, c
| (⇒)
¬a, b, ¬a, c

Example 10.8
For the formula (((a ⇒ b) ∩ ¬c) ∪ (a ⇒ c)) and the tree TA , the leaves are
L1 = ¬a, b, ¬a, c and CL1 = {¬a, b, c} and
L2 = ¬c, ¬a, c and CL2 = {¬c, ¬a, c}. The set of clauses is

CA = {{¬a, b, c}, {¬c, ¬a, c}}.

By theorem 10.5, A ≡ CA . Semantically it means, by definition 10.13,

A ≡ (((¬a ∪ b) ∪ c) ∩ ((¬c ∪ ¬a) ∪ c)).

Theorem 10.6 (Clausal Form)

For any formula A of a predicate language L, there is an effective procedure of
generating an open formula A∗ of a quantifiers free language L∗ and a set CA∗
of clauses such that
A∗ ≡ CA∗ . (10.30)
The set CA∗ of clauses of L∗ with the property (10.30) is called a clausal form
of the formula A of L.

Proof
Given a formula A of a language L. The open formula A∗ of the quantifiers
free language L∗ is obtained by the Skolemization process. The effectiveness
and correctness of the process follows from PNF theorem 10.10 and Skolem
theorem 10.11 described in section 10.4.1.
As the next step, we define (section 10.4.2) a proof system QRS∗ (10.43) based
on the quantifiers free language L∗ . The system QRS∗ is a version of the system
QRS (10.8) restricted to its Propositional Rules. At this point we carry the
proof in a similar way to the proof in the propositional case (theorem 10.5).
Namely, for any formula A∗ of L∗ obtained from A of L we construct its the

481
decomposition tree TA∗ . We form clauses out of the leaves of the tree TA∗ , i.e.
for every leaf L we create a clause CL determined by L and we put

CA∗ = {CL : L is a leaf of TA∗ }.

This is the clausal form of the formula A of L by theorem 10.13 proved in

section 10.4.2. To complete the proof we need now to develop results of the
section 10.4.1 and the section 10.4.2.

10.4.1 Prenex Normal Forms and Skolemization

We remind the following important notion.

Term t is free for x in A(x). Let A(x) ∈ F and t be a term, A(t) be a result
of substituting t for all free occurrences of x in A(x).
We say that t is free for x in A(x), if no occurrence of a variable in t
becomes a bound occurrence in A(t).

In particular, if A(x), A(x1 , x2 , ..., xn ) ∈ F and t, t1 , t2 , ..., tn ∈ T, then

A(x/t), A(x1 /t1 , x2 /t2 , ..., xn /tn )

or, more simply just

A(t), A(t1 , t2 , ..., tn )

denotes the result of replacing all occurrences of the free variables x, x1 , x2 , ..., xn ,
by the terms t, t1 , t2 , ..., tn , respectively, assuming that t, t1 , t2 , ..., tn are free for
x, x1 , x2 , ..., xn , respectively, in A.
The assumption that t is free for x in A(x) while substituting t for x, is
important because otherwise we would distort the meaning of A(t). This is
illustrated by the following example.

Example 10.9
Let t = y and A(x) be
∃y(x 6= y).
Obviously t is not free for y in A. The substitution of t for x produces a formula
A(t) of the form
∃y(y 6= y),
which has a different meaning than ∃y(x 6= y).

482
Here are more examples illustrating the notion: t is free for x in A(x).

Example 10.10
Let A(x) be a formula
(∀yP (x, y) ∩ Q(x, z))
and t be a term f (x, z), i.e. t = f (x, z).
None of the occurrences of the variables x, z of t is bound in A(t), hence we say
that t = f (x, z) is free for x in (∀yP (x, y) ∩ Q(x, z)).

Substituting t on a place of x in A(x) we obtain a formula A(t) of the form

(∀yP (f (x, z), y) ∩ Q(f (x, z), z)).

Example 10.11
Let A(x) be a formula
(∀yP (x, y) ∩ Q(x, z))
The term t = f (y, z) is not free for x in A(x) because substituting t = f (y, z)
on a place of x in A(x) we obtain now a formula A(t) of the form

(∀yP (f (y, z), y) ∩ Q(f (y, z), z))

which contain a bound occurrence of the variable y of t (∀yP (f (y, z), y)).

The other occurrence (Q(f (y, z), z)) of y is free, but it is not sufficient, as for
term to be free for x, all occurrences of its variables has to be free in A(t).
Another important notion we will use here is the following notion of similarity
of formulas.
Intuitively, we say that A(x) and A(y) are similar if and only if A(x) and A(y)
are the same except that A(x) has free occurrences of x in exactly those places
where A(y) has free occurrences of y.

Example 10.12
The formulas ∃z(P (x, z) ⇒ Q(x)) and ∃z(P (y, z) ⇒ Q(y)) are similar.

The formal definition of this notion follows.

Definition 10.14 (Similarity)

Let x and y be two different variables. We say that the formulas A(x) and
A(x/y) are similar and denote it by
A(x) ∼ A(x/y)
if and only if y is free for x in A(x) and A(x) has no free occurrences of y.

483
Example 10.13
The formulas A(x): ∃z(P (x, z) ⇒ Q(x, y)) and A(x/y): ∃z(P (y, z) ⇒ Q(y, y))
are not similar; y is free for x in A(x), but the formula A(x/y) has a free
occurrence of y.

Example 10.14
The formulas A(x): ∃z(P (x, z) ⇒ Q(x, y)) and A(x/w): ∃z(P (w, z) ⇒ Q(w, y)
are similar; w is free for x in A(x) and the formula A(x/w) has no free
occurrence of w.

Directly from the definition we get the following.

Lemma 10.1
For any formula A(x) ∈ F, if A(x) and A(x/y) are similar A(x) ∼ A(y), then

∀xA(x) ≡ ∀yA(y),

∃xA(x) ≡ ∃yA(y).

We prove, by the induction on the number of connectives and quantifiers in a

formula A the following.

Theorem 10.7 (Replacement Theorem)

For any formulas A, B ∈ F, if B is a sub-formula of A, if A∗ is the result of
replacing zero or more occurrences of B in A by a formula C, and B ≡ C, then
A ≡ A∗ .

Directly from lemma 10.1 and replacement theorem 10.7 we get that the follow-
ing theorem holds.

Theorem 10.8 (Change of Bound Variables)

For any formula A(x), A(y), B ∈ F, if A(x) and A(x/y) are similar, i.e. A(x) ∼
A(y), and the formula ∀xA(x) or ∃xA(x) is a sub-formula of B, and B ∗ is the
result of replacing zero or more occurrences of A(x) in B by a formula ∀yA(y)
or ∃yA(y), then B ≡ B ∗ .

Definition 10.15 (Naming Variables Apart)

We say that a formula B has its variables named apart if no two quantifiers
in B bind the same variable and no bound variable is also free.

We can now use theorem 10.8 to prove its more general version.

484
Theorem 10.9 (Naming Variables Apart) Every formula A ∈ F is logi-
cally equivalent to one in which all variables are named apart.

We use the above theorems plus the equational laws for quantifiers (10.31) to
prove, as a next step a so called a Prenex Form Theorem 10.10.
In order to do so we first we define an important notion of prenex normal
form of a formula.

Definition 10.16 (Closure of a Formula)

By a closure of a formula A we mean a closed formula A0 obtained from A
prefixing in universal quantifiers all those variables that a free in A; i.e. if
A(x1 , . . . . , xn ) then A0 ≡ A is

∀x1 ∀x2 ....∀xn A(x1 , x2 , . . . . ., xn )

Example 10.15
Let A be a formula (P (x, y) ⇒ ¬∃z R(x, y, z)), its closure A0 ≡ A is
∀x∀y(P (x, y) ⇒ ¬∃z R(x, y, z)).

Definition 10.17 (Prenex Normal Form)

Any formula A of the form

Q1 x1 Q2 x2 ....Qn xn B

where each Qi is a universal or existential quantifier, i.e. for all 1 ≤ i ≤ n,

Qi ∈ {∃, ∀}, xi 6= xj for i 6= j, and B contains no quantifiers, is said to be
in prenex normal form (PNF).
We include the case n = 0 when there are no quantifiers at all.

We assume that the formula A in PNF is always closed. If it is not closed we

form its closure (definition 10.16) instead. We prove that, for every formula
A, we can effectively construct a formula B that is in the prenex normal form
PNF and A ≡ B.

Theorem 10.10 (PNF Theorem)

There is an effective procedure for transforming any formula A ∈ F into a
logically equivalent formula A0 in the prenex normal form PNF.

Proof
We use theorems 10.7, 10.8, 10.9, theorem 10.15, and the following logical equiv-
alences (10.31) proved in chapter 2.

485
 Equational Laws of Quantifiers (10.31)

∀x(A(x) ∪ B) ≡ (∀xA(x) ∪ B) (10.32)

∀x(A(x) ∩ B) ≡ (∀xA(x) ∩ B) (10.33)

∃x(A(x) ∪ B) ≡ (∃xA(x) ∪ B) (10.34)

∃x(A(x) ∩ B) ≡ (∃xA(x) ∩ B) (10.35)

∀x(A(x) ⇒ B) ≡ (∃xA(x) ⇒ B) (10.36)

∃x(A(x) ⇒ B) ≡ (∀xA(x) ⇒ B) (10.37)

∀x(B ⇒ A(x)) ≡ (B ⇒ ∀xA(x)) (10.38)

∃x(B ⇒ A(x)) ≡ (B ⇒ ∃xA(x)) (10.39)

where B is a formula such that B does not contain any free occurrence of x.
The formal procedure is defined by induction on the number k of occurrences of
connectives and quantifiers in A. We show now how it works in some particular
cases.

Exercise 10.1
Find a prenex normal form PNF of a formula A: (∀x(P (x) ⇒ ∃xQ(x)).

Solution

We find PNF in the following steps.

Step 1: Rename Variables Apart
By the theorem 10.8 we can make all bound variables in A different, i.e. we
transform A into an equivalent formula A0

∀x(P (x) ⇒ ∃yQ(y)).

486
Step 2: Pull out Quantifiers

We apply the equational law

(C ⇒ ∃yQ(y)) ≡ ∃y (C ⇒ Q(y))

to the sub-formula B : (P (x) ⇒ ∃yQ(y)) of A0 for C = P (x), as P(x) does

not contain the variable y. We get its equivalent formula B ∗ : ∃y(P (x) ⇒
Q(y)). We substitute now B ∗ on place of B in A0 and get a formula

A00 : ∀x∃y(P (x) ⇒ Q(y))

such that A00 ≡ A0 ≡ A.

A00 is a required prenex normal form PNF for A.

Exercise 10.2 Find a prenex normal form PNF formula A0 for the formula
A:
(∃x∀y R(x, y) ⇒ ∀y∃x R(x, y))

Solution
Step 1: Rename Variables Apart
Take a sub- formula B(x, y) : ∀y∃x R(x, y) of A, get B(x/z, y/w) : ∀z∃w R(z, w)
and replace B(x,y) by B(x/z, y/w) in A and get

(∃x∀y R(x, y) ⇒ ∀z∃w R(z, w))

Step 2: Pull out quantifiers

We use corresponding equational laws of quantifiers (11.46), 11.47) to pull out
first quantifiers ∃x∀y and get the following

A0 : ∀x∃y((R(x, y) ⇒ ∀z∃w R(z, w))),

such that A0 ≡ A. Now we pull quantifiers ∀z∃w in (R(x, y) ⇒ ∀z∃w R(z, w))
and get the prenex normal form PNF formula

A00 : ∀x∃y∀z∃w ((R(x, y) ⇒ R(z, w))),

such that A00 ≡ A0 ≡ A.

Observe we can also perform a different Step 2 by pulling first the quantifiers
∀z∃w and then quantifiers ∀x∃y and obtain another PNF A000 for A:

A000 : ∀z∃w∀x∃y (R(x, y) ⇒ R(z, w)).

We will show now how any formula A in its prenex normal form PNF we can
transformed it into a corresponding open formula A∗ .

487
The open formula A∗ belongs to a richer language then the initial language to
which the formula A belongs. The transformation process adds new constants,
called Skolem constants, and new function symbols, called Skolem function
symbols to the initial language.
The whole process is called the skolemisation of the initial language L, the such
build extension of the initial language is called a Skolem extension of L,.

Skolem Procedure of Elimination of Quantifiers (10.40)

Given a formula A of the language L = L{¬,∪,∩,⇒} (P, F, C) in its prenex normal

form PNF, i.e.

A = Q1 x1 Q2 x2 . . . Qn xn B(x1 , x2 , . . . xn ) (10.41)

where each Qi is a universal or existential quantifier, i.e. for all 1 ≤ i ≤ n,

Qi ∈ {∃, ∀}, xi 6= xj for i 6= j, and B(x1 , x2 , . . . xn ) contains no quantifiers.

We describe now a procedure of elimination of all quantifiers from the formula

A (10.41) and hence transforming it into a corresponding open formula A∗ .
We assume that the formula A = Q1 x1 Q2 x2 . . . Qn xn B(x1 , x2 , . . . xn ) is closed.
If it is not closed we form its closure instead.
We considerer 3 cases.
Case 1
All quantifiers Qi for 1 ≤ i ≤ n are universal, i.e. the closed formula A is

∀x1 ∀x2 . . . ∀xn B(x1 , x2 , . . . , xn )

We replace the formula A by the open formula A∗ :

B(x1 , x2 , . . . . , xn ).

Case 2
All quantifiers Qi for 1 ≤ i ≤ n are existential, i.e. the closed formula A is

∃x1 ∃x2 ....∃xn B(x1 , x2 , . . . xn )

We replace the formula A by the open formula A∗ :

B(c1 , c2 , . . . . , cn )

where c1 , c2 , . . . . , cn and new individual constants, all different, added to

our original language L. We call such constants added to the language Skolem
constants
Case 3
The quantifiers are mixed . We assume that A is closed. If it is not closed we

488
form its closure instead. We eliminate quantifiers one by one and step by step
depending on first, and consecutive quantifiers.

Given a closed PNF formula A

Q1 x1 Q2 x2 . . . Qn xn B(x1 , x2 , . . . xn )
Step 1 Elimination of Q1 x1
We have two possibilities for the first quantifier Q1 x1 , namely P1 Q1 x1 is uni-
versal or P2 Q1 x1 is existential.

Consider P1
First quantifier in A is universal, i. e. A is
∀x1 Q2 x2 . . . Qn xn B(x1 , x2 , . . . xn )

We replace A by a formula A1 :
Q2 x2 . . . Qn xn B(x1 , x2 , . . . xn )
We have eliminated the quantifier Q1 in this case.

Consider P2
First quantifier in A is existential, i. e. A is
∃x1 Q2 x2 . . . Qn xn B(x1 , x2 , . . . xn )

We replace A by a formula A1 :
Q2 x2 . . . Qn xn B(b1 , x2 , . . . xn )
where b1 is a new constant symbol added to our original language L. We
call such constant symbol added to the language Skolem constant symbol.
We have eliminated the quantifier Q1 in this case. We have covered all cases
and this ends the Step 1.

Step 2 Elimination of Q2 x2 .

Consider now the PNF formula A1 from Step1- P1

Q2 x2 . . . Qn xn B(x1 , x2 , . . . xn )
Remark that the formula A1 might not be closed.
We have again two possibilities for elimination of the quantifier Q2 x2 , namely
P1 Q2 x2 is universal or P2 Q2 x2 is existential.

489
Consider P1
First quantifier in A1 is universal, i.e. A1 is

∀x2 Q3 x3 . . . Qn xn B(x1 , x2 , x3 , . . . xn )

We replace A1 by the following A2

Q3 x3 . . . Qn xn B(x1 , x2 , x3 , . . . xn )

We have eliminated the quantifier Q2 in this case.

Consider P2
First quantifier in A1 is existential, i.e. A1 is

∃x2 Q3 x3 . . . Qn xn B(x1 , x2 , x3 , . . . xn )

Observe that now the variable x1 is a free variable in B(x1 , x2 , x3 , . . . xn ) and

hence in A1 .
We replace A1 by the following A2

Q3 x3 . . . Qn xn B(x1 , f (x1 ), x3 , . . . xn )

where f is a new one argument functional symbol added to our original

language L. We call such functional symbols added to the original language
Skolem functional symbols.
We have eliminated the quantifier Q2 in this case.

Consider now the PNF formula A1 from Step1 - P2

Q2 x2 Q3 x3 . . . Qn xn B(b1 , x2 , . . . xn )

Again we have two cases.

Consider P1
First quantifier in A1 is universal, i.e. A1 is

∀x2 Q3 x3 . . . Qn xn B(b1 , x2 , x3 , . . . xn )

We replace A1 by the following A2

Q3 x3 . . . Qn xn B(b1 , x2 , x3 , . . . xn )

We have eliminated the quantifier Q2 in this case.

Consider P2
First quantifier in A1 is existential, i.e. A1 is

∃x2 Q3 x3 . . . Qn xn B(b1 , x2 , x3 , . . . xn )

490
We replace A1 by A2
Q3 x3 . . . Qn xn B(b1 , b2 , x3 , . . . xn )
where b2 6= b1 is a new Skolem constant symbol added to our original language
L.
We have eliminated the quantifier Q2 in this case. We have covered all cases
and this ends the Step 2. Step 3 Elimination of Q3 x3

Let’s now consider, as an example formula A2 from Step 2; P1 i.e. the formula
Q3 x3 . . . Qn xn B(x1 , x2 , x3 , . . . xn )
We have again 2 choices to consider, but will describe only the following.

P2 First quantifier in A2 is existential, i. e. A2 is

∃x2 Q4 x4 . . . Qn xn B(x1 , x2 , x3 , x4 , . . . xn )
Observe that now the variables x1 , x2 are free variables in B(x1 , x2 , x3 , . . . xn )
and hence in A2 .

We replace A2 by the following A3

Q4 x3 . . . Qn xn B(x1 , x2 , g(x1 , x2 ), x4 . . . xn )
where g is a new two argument functional symbol added to our original
language L.
We have eliminated the quantifier Q3 in this case.

Step i
At each Step i, for 1 ≤ i ≤ n), we build a binary tree of possibilities:
P1 Qi xi is universal or P2 Qi xi is existential and as result we obtain a
formula Ai with one less quantifier. The elimination process builds a sequence
of formulas
A, A1 , A2 , . . . , An = A∗
where the formula A belongs to our original language
L = L{¬,∪,∩,⇒} (P, F, C),
∗
the formula A belongs to its Skolem extension language (10.42) defined as
follows.

Definition 10.18
The language L∗ obtained from L by the quantifiers elimination procedure (10.40)
is is called a Skolem extension of L.
L∗ = L{¬,∪,∩,⇒} (P, F ∪ SF, C ∪ SC). (10.42)

491
Observe that in the elimination process (10.40) a universal quantifier intro-
duces free variables in the formula B(x1 , x2 , . . . xn ). The elimination of an
existential quantifier that follows universal quantifiers introduces a new func-
tional symbol with number of arguments equal the number of universal quanti-
fiers preceding it.
The resulting is an open formula A∗ of Skolem extension language L∗ . By
PNF theorem 10.10, for any formula A of L its PNF formula (10.41) exists
and is logically equivalent with A. We hence introduce the following definition.

Definition 10.19 (Skolemization)

Given a formula A of L.
A formula A∗ of the Skolem extension language L∗ (10.42) obtained from a
PNF form of A by the Skolem Procedure (10.40) is called a Skolem form of
the formula A and the process obtaining it is called a Skolemization of A.

Exercise 10.3 Let A be a PNF formula

∀y1 ∃y2 ∀y3 ∃y4 B(y1 , y2 , y3 , y4 , y4 ).
Find the Skolem form of A (the formula B(y1 , y2 , y3 , y4 , y4 ) is quantifiers
free).

Solution
We eliminate ∀y1 and get a formula A1
∃y2 ∀y3 ∃y4 B(y1 , y2 , y3 , y4 ).
We eliminate ∃y2 by replacing y2 by h(y1 ) where h is a new one argument
functional symbol added to our original language L.
We get a formula A2
∀y3 ∃y4 B(y1 , h(y1 ), y3 , y4 ).
We eliminate ∀y3 and get a formula A3
∃y4 B(y1 , h(y1 ), y3 , y4 ).
We eliminate ∃y4 by replacing y4 by f (y1 , y3 ), where f is a new two argument
functional symbol added to our original language L.
We get a formula A4 that is our resulting open formula A∗
B(y1 , h(y1 ), y3 , f (y1 , y3 )).

Exercise 10.4
Let now A be a PNF formula
∃y1 ∀y2 ∀y3 ∃y4 ∃y5 ∀y6 B(y1 , y2 , y3 , y4 , y4 , y5 , y6 )
Find the Skolem form of A (the formula B(y1 , y2 , y3 , y4 , y4 , y5 ) is quantifiers
free).

492
Solution
We eliminate ∃y1 and get a formula A1
∀y2 ∀y3 ∃y4 ∃y5 ∀y6 B(b1 , y2 , y3 , y4 , y4 , y5 , y6 )
where b1 is a new constant symbol added to our original language L.
We eliminate ∀y2 , f orally3 and get a formulas A2 , A3 ; here is the formula A3
∃y4 ∃y5 ∀y6 B(b1 , y2 , y3 , y4 , y4 , y5 , y6 )
We eliminate ∃y4 and get a formula A4
∃y5 ∀y6 B(b1 , y2 , y3 , g(y2 , y3 ), y5 , y6 )
where g is a new two argument functional symbol added to our original
language L.
We eliminate ∃y5 and get a formula A5
∀y6 B(b1 , y2 , y3 , g(y2 , y3 ), h(y2 , y3 ), y6 )
where h is a new two argument functional symbol added to our original
language L.
We eliminate ∀y6 and get a formula A6 that is the resulting open formula A∗
B(b1 , y2 , y3 , g(y2 , y3 ), h(y2 , y3 ), y6 ).

The correctness of the Skolemization process is established by the Skolem

theorem 10.11. It states informally that the formula A∗ obtained from a formula
A via the Skolemization is satisfiable if and only if the original one is satisfiable.
We define this notion formally as follows.
.

Definition 10.20 (Equisatisfiable)

For any formulas A of L and B of the Skolem extension L∗ (10.42) of L,
we say that A and B are equisatisfiable if and only if the following conditions
are satisfied.
1. Any structure M of L can be extended to a structure M∗ of L∗ and following
implication holds.
If M |= A, then M∗ |= B.

2. Any structure M∗ of L∗ can be restricted to a structure M of L and following

implication holds.
If M∗ |= B, then M |= A.

Theorem 10.11 (Skolem Theorem)

Let L∗ be the Skolem extension (10.42) of a language L.
Any formula A of L and its Skolem form A∗ of L∗ are equisatisfiable.

493
10.4.2 Clausal Form of Formulas

Let L∗ be the Skolem extension of L, i.e. L∗ does not contain quantifiers. We

define a proof system QRS∗ as an open formulas language version of QRS that
includes only its Group 1: Propositional Rules of (10.7).
We denote the set of formulas of L∗ by OF to stress the fact that all its formulas
are open and define QRS∗ formally as follows.

QRS∗ = (L∗ , E, LA, R), (10.43)

where E = {Γ : Γ ∈ OF ∗ }, LA is defined by (10.6), and R contains Group 1:

Propositional Rules (10.7):

0 0 0
Γ , A, B, ∆ Γ , ¬A, ∆ : Γ , ¬B, ∆
(∪) , (¬ ∪)
Γ0 , (A ∪ B), ∆ Γ0 , ¬(A ∪ B), ∆

0 0 0
Γ , A, ∆ ; Γ , B, ∆ Γ , ¬A, ¬B, ∆
(∩) , (¬ ∩)
Γ0 , (A ∩ B), ∆ Γ0 , ¬(A ∩ B), ∆

0 0 0
Γ , ¬A, B, ∆ Γ , A, ∆ : Γ , ¬B, ∆
(⇒) , (¬ ⇒)
Γ0 , (A ⇒ B), ∆ Γ0 , ¬(A ⇒ B), ∆

0
Γ , A, ∆
(¬ ¬) 0
Γ , ¬¬A, ∆
0
where Γ ∈ LT ∗ , ∆ ∈ OF ∗ , A, B ∈ OF.

For any formula A ∈ OF we define, as we did in chapter 6 its decomposition

tree TA as follows.
Decomposition tree TA
Step 1. The formula A is the root of TA and for any node ∆ of the tree we
follow the steps bellow.
Step 2. If ∆ in indecomposable, then ∆ becomes a leaf of the tree.
Step 3. If ∆ is decomposable, then we traverse ∆ from left to right to identify
the first decomposable formula B. In case of a one premisses rule we put is
premise as a leaf; in case of a two premisses rule we put its left and right
premisses as the left and right leaves, respectively.
Step 4. We repeat steps 2 and 3 until we obtain only leaves.
We adopt the definition 10.12 to QRS∗ and the language L∗ .

Definition 10.21 (Semantics)

494
For any sequence Γ of formulas of L∗ , any structure M = [M, I] for L∗ ,
M |= Γ if and only if M |= δΓ ,
where δΓ denotes a disjunction of all formulas in Γ.

The semantics for clauses is basically the same as for the sequences. We define
it, after definition 10.5, as follows.

Definition 10.22 (Clauses Semantics)

For any clause C of the language L∗ (definition 10.9), δC denotes a disjunction
of all literals in C.
For any finite set of clauses C of L∗ , any structure M = [M, I] for L∗ , and any
C ∈ C,
1. M |= C if and only if M |= δC .
2. M |= C if and only if M |= δC for all C ∈ C.
3. (A ≡ C) if and only if A ≡ σC ,
where σC is a conjunction of all formulas δC for all clauses C ∈ C.

Obviously, all rules of QRS∗ are strongly sound (definition 10.7) and theorem
10.2 holds for QRS∗ , i.e. we have the following.

Theorem 10.12 (Strong Soundness)

The proof system QRS∗ is strongly sound.

We are going to prove now that any formula A of L∗ can be transformed into
in logically equivalent set of clauses.

Theorem 10.13 (Formula-Clauses Equivalency)

For any formula A of L∗ , there is an effective procedure of generating a set of
clauses CA of L∗ such that
A ≡ CA (10.44)

Proof
Given A ∈ OF. Here is the two steps procedure. S1. We construct (finite and
unique) decomposition tree TA . S2. We form clauses out of the leaves of the
tree TA , i.e. for every leaf L we create a clause CL determined by L (definition
10.10) and we put
CA = {CL : L is a leaf of TA }.

Directly from the strong soundness of rules of inference of QRS∗ (theorem

10.12) and the semantics for clauses definition 10.22 we get that
A ≡ CA .

495
Exercise 10.5
Find he set CA of clauses for the following formula A.

(((P (b, f (x)) ⇒ Q(x)) ∪ ¬R(z)) ∪ (P (b, f (x)) ∩ R(z))))

Solution
S1. We construct the decomposition tree for A as follows

TA

(((P (b, f (x)) ⇒ Q(x)) ∪ ¬R(z)) ∪ (P (b, f (x)) ∩ R(z)))

| (∪)
(((P (b, f (x)) ⇒ Q(x)) ∪ ¬R(z)), (P (b, f (x)) ∩ R(z))
| (∪)
(P (b, f (x)) ⇒ Q(x)), ¬R(z), (P (b, f (x)) ∩ R(z))
| (⇒)
¬P (b, f (x)), Q(x), ¬R(z), (P (b, f (x)) ∩ R(z))
^
(∩)

¬P (b, f (x)), Q(x), ¬R(z), P (b, f (x)) ¬P (b, f (x)), Q(x), ¬R(z), R(z)

S2. The leaves of TA are

L1 = ¬P (b, f (x)), Q(x), ¬R(z), P (b, f (x)) and
L2 = ¬P (b, f (x)), Q(x), ¬R(z), R(z).
The corresponding clauses are
C1 = {¬P (b, f (x)), Q(x), ¬R(z), P (b, f (x))} and
C2 = {¬P (b, f (x)), Q(x), ¬R(z), R(z)}.
The set of clauses is

CA = {{¬P (b, f (x)), Q(x), ¬R(z), P (b, f (x)}, {¬P (b, f (x)), Q(x), ¬R(z), R(z)}.

Definition 10.23 Clausal Form Given a formula A of the language L and its
Skolem form A∗ of L∗ . The set CA∗ of clauses such that

A∗ ≡ CA∗

s called a clausal form of the formula A of L.

Exercise 10.6 Find the clausal form of a formula A:

(∃x∀y (R(x, y) ∪ ¬P (x)) ⇒ ∀y∃x ¬R(x, y)).

496
Solution
Step 1: We rename variables apart in A and get a formula A0 :

(∃x∀y (R(x, y) ∪ ¬P (x)) ⇒ ∀z∃w ¬R(z, w)).

Step 2: We use Equational Laws of Quantifiers (11.46), (11.47)t o pull out ∃x

and ∀y and get a formula A00 :

(∀x∃y ((R(x, y) ∪ ¬P (x)) ⇒ ∀z∃w ¬R(z, w)).

Step 3: We use Equational Laws of Quantifiers (11.46), (11.47)t o pull out ∃x

and ∀y and get a formula A000 :

(∀x∃y ((R(x, y) ∪ ¬P (x)) ⇒ ∀z∃w ¬R(z, w)).

Step 4: We use Equational Laws of Quantifiers (10.38), (10.39)t o pull out ∃z

and ∀w from the sub formula ((R(x, y) ∪ ¬P (x)) ⇒ ∀z∃w ¬R(z, w)) and get a
formula A0000 This is the prenex normal form PNF of A.

(∀x∃y∀z∃w ((R(x, y) ∪ ¬P (x)) ⇒ ¬R(z, w)). (10.45)

Step 5: We perform the Skolemization Procedure (10.40) to (10.45). Observe

(10.45) that the formula is of the form of the formula of exercise 10.3. We follow
the exercise and eliminate ∀x and get a formula A1

∃y∀z∃w ((R(x, y) ∪ ¬P (x)) ⇒ ¬R(z, w)).

We eliminate ∃y by replacing y by h(x) where h is a new one argument

functional symbol added to our original language L.
We get a formula A2

∀z∃w ((R(x, h(x)) ∪ ¬P (x)) ⇒ ¬R(z, w)).

We eliminate ∀z and get a formula A3

∃w ((R(x, h(x)) ∪ ¬P (x)) ⇒ ¬R(z, w)).

We eliminate ∃w by replacing w by f (x, z), where f is a new two argument

functional symbol added to our original language L.
We get a formula A4 that is our resulting open formula A∗

A∗ : ((R(x, h(x)) ∪ ¬P (x)) ⇒ ¬R(z, (x, z))). (10.46)

Step 6: We build the decomposition tree TA∗ for (10.46).

TA∗

((R(x, h(x)) ∪ ¬P (x)) ⇒ ¬R(z, f (x, z)))

| (⇒)

497
 ¬(R(x, h(x)) ∪ ¬P (x)), ¬R(z, f (x, z))
^
(¬∪)

¬¬P (x), ¬R(z, f (x, z))

¬R(x, h(x)), ¬R(z, f (x, z)
| (¬¬)
P (x), ¬R(z, f (x, z))

Step 7: The leaves of TA∗ are

L1 = ¬R(x, h(x)), ¬R(z, f (x, z) and L2 = P (x), ¬R(z, f (x, z)).
The corresponding clauses are
C1 = {¬R(x, h(x)), ¬R(z, f (x, z)} and
C2 = {P (x), ¬R(z, f (x, z))}.
Step 8: The clausal form of the formula A

(∃x∀y (R(x, y) ∪ ¬P (x)) ⇒ ∀y∃x ¬R(x, y))

is the set of clauses

CA∗ = { {¬R(x, h(x)), ¬R(z, f (x, z)}, {P (x), ¬R(z, f (x, z))} }.

10.5 Homework Problems

1. Given a predicate (first order) language (10.1), i.e. L = L{∩,∪,⇒,¬} (P, F, C).
Let QRS de a proof system (10.8). For any formulas A, B of L, we define:

`QRS A ≡ B it and only if `QRS (A ⇒ B) and `QRS (B ⇒ A).

Show that for any formulas A(x), B(x) with a free variable x the following
holds.
Remider: 1. you treat A(x), B(x) as atomic foprmulas, 2.you must trans-
form formulas with restricted domain quantifiers into proper formulas of
L.
(i) `QRS f orallx (A(x) ∩ B(x)) ≡ (∀xA(x) ∩ ∀xB(x))
(ii) `QRS ∃x (A(x) ∪ B(x)) ≡ (∃xA(x) ∪ ∃xB(x)).
(iii) `QRS ¬∀B(x) A(x) ≡ ∃B(x) .
(iv) `QRS ¬∃B(x) A(x) ≡ ∀B(x) ¬A(x).
(v) `QRS ¬∀xA(x) ≡ ∃x¬A(x).
(vi) `QRS ¬∃xA(x) ≡ ∀x¬A(x).
(vii) `QRS (∀x(B(x) ⇒ A(x)) ⇒ (∃x B(x) ⇒ ∃x (B(x) ∩ A(x))))

498
2. Show that for any formulas A(x), B B where B does not contain any free
occurrence of x the following holds.
(i) `QRS ∀x(A(x) ∩ B) ≡ (∀xA(x) ∩ B).
(ii) `QRS ∀x(A(x) ∪ B) ≡ (∀xA(x) ∪ B).
(iii) `QRS ∃x(A(x) ⇒ B) ≡ (∀xA(x).
(iv) `QRS ∃x(A(x) ⇒ B) ≡ (∀xA(x).
3. Prove that following formulas are not provable in QRS.
Remider: you must transform formulas with restricted domain quantifiers
into proper formulas of L.
(i) ∃C(x) (A(x) ∪ B) 6≡ (∃C(x) A(x) ∪ B).
(ii) ∀C(x) (A(x) ∩ B) 6≡ (∀C(x) A(x) ∩ B).
(iii) ∃C(x) (A(x) ⇒ B) 6≡ (∀C(x) A(x) ⇒ B).
(iv) ∃C(x) (B ⇒ A(x)) 6≡ (B ⇒ ∃xA(x)).
4. Prove that following formulas are not provable in QRS.
(i) (∃x ¬A(x) ⇒ ∀x A(x))
(ii) (∀x∃y A(x, y) ⇒ ∃x∀y A(x, y)).
(iii) (∃x∃y A(x, y) ⇒ ∃y A(y, y)).
(iv) (∀x∃y A(x, y) ⇒ ∃y A(y, y)).
(v) (∀x (A(x) ⇒ B(x)) ⇒ (∀x A(x) ⇒ ∃x B(x))).
5. Prove that following formulas are not provable in QRS.
(i) A1 : ∀x¬∃y(P (x, g(y, y)) ∪ P (x, g(g(y, y), d))).
(ii) A2 : (¬∀yP (f (x, y), c) ⇒ (P (x, c) ∪ P (y, c)))
(iii) A3 : ∀x(P (x) ⇒ ∃yQ(x, y)).
(iv)A4 : ∀x¬∃y(P (x) ∩ ¬Q(x, y)).
6. Find counter-models determined by the decomposition trees TAi for the
following formulas Ai , i = 1, 2, 3, 4.
(i) A1 : ∀x¬∃y(Q(x, g(y)) ∪ R(x, f (x, y), c))).
(ii) A2 : (¬∀yR(f (x, y), c) ⇒ (Q(x, c) ∪ Q(y, c)))
(iii) A3 : ∀x(P (x) ⇒ ∃yQ(x, y)).
(iv)A4 : ∀x¬∃y(P (x) ∩ ¬Q(f (x, y))).
7. Find prenex normal form PNF of the following formulas.
Reminder: We assume that the formula A in PNF is always closed. If it
is not closed we form its closure (definition 10.16) instead.
(i) (∀x(P (x) ⇒ ¬∀yP (y)) ⇒ (∃x R(x, y) ⇒ ∃y (R(x, y) ∩ P (y)))).
(ii) ((∀xQ(x) ⇒ (∃xR(x) ∪ ¬∀xQ(x))) ⇒ (¬∃xQ(x) ∩ R(x))).

499
 (iii) (∀x R(f (x, y), c) ⇒ (∃xR(f (x, y), c))∩¬R(f (x, y), c)) ⇒ (¬∀x R(f (x, y), c) ⇒
∃x R(f (x, y), c))).
(iv) ((∃R(y) P (y) ⇒ Q(x)) ⇒ (P (y) ⇒ ∃xQ(x)))
8. Find a Skolem form of the following formulas (the formula B(y1 , y2 , y3 , y4 , y4 )
is quantifiers free).
(i) ∀y1 ∀y2 ∀y3 ∃y4 B(y1 , y2 , y3 , y4 , y4 ).
(ii) ∃y1 ∃y2 ∀y3 ∃y4 B(y1 , y2 , y3 , y4 , y4 ).
(iii) ∃y1 ∀y2 ∃y3 ∃y4 B(y1 , y2 , y3 , y4 , y4 ).
(iv) ∀y1 ∀y2 ∃y3 ∃y4 B(y1 , y2 , y3 , y4 , y4 ).

9. Find the clausal form of the following formulas.

(i) (∀x(P (x) ⇒ ¬∀yP (y)) ⇒ (∃x R(x, y) ⇒ ∃y (R(x, y) ∩ P (y)))).
(ii) ((∀xQ(x) ⇒ (∃xR(x) ∪ ¬∀xQ(x))) ⇒ (¬∃xQ(x) ∩ R(x))).
(iii) (∀x R(f (x, y), c) ⇒ (∃xR(f (x, y), c))∩¬R(f (x, y), c)) ⇒ (¬∀x R(f (x, y), c) ⇒
∃x R(f (x, y), c))).
(iv) ((∃R(y) P (y) ⇒ Q(x)) ⇒ (P (y) ⇒ ∃xQ(x)))
10. Find the set of clauses logically equivalent to clausal form of the following
formulas.
(i) (((a ⇒ b) ∩ ¬c) ∪ (a ⇒ c)).
(ii) ((a ∪ b) ⇒ ¬a) ∪ (¬a ⇒ ¬c)).
(iii) (¬(¬a => (a ∩ ¬b)) => (¬a ∩ (¬a ∪ ¬b))).
(iv) (¬(a ∩ b) ⇒ (¬a ∪ ¬b)).
(v) (((¬a ⇒ (b ∩ c)) ∩ ¬(c ∪ a)) ∪ (a ⇒ c))

500
Chapter 11

Formal Theories and Gödel

Theorems

Formal theories play crucial role in mathematics and were historically defined
for classical predicate (first order logic) and consequently for other first and
higher order logics, classical and non-classical.
The idea of formalism in mathematics, which resulted in the concept of formal
theories, or formalized theories, as they are also called. Their concept was
developed in connection with the Hilbert Program. One of the main objects
of the program was to construct a formal theory that would cover the whole
mathematics and to prove its consistency by employing the simplest of logical
means. This part of the program was called the Consistency Program, where a
formal theory is said to be consistent if no formal proof can be carried in that
theory for a formula A and at the same time for its negation ¬A.
In 1930, while still in his twenties Kurt Gödel made a historic announcement:
Hilbert Consistency Program could not be carried out. He justified his claim by
proving his Inconsistency Theorem, called also Second Incompleteness Theorem.
Roughly speaking the theorem states that a proof of the consistency of every
formal theory that contains arithmetic of natural numbers can be carried out
only in mathematical theory which is more comprehensive than the one whose
consistency is to be proved. In particular, a proof of the consistency of formal
(elementary, first order) arithmetic can be carried out only in mathematical
theory which contains the whole arithmetic and also other theorems that do
not belong to arithmetic. It applies to a formal theory that would cover the
whole mathematics because it would obviously contain the arithmetic on natural
numbers. Hence the Hilbert Consistency Program fails.
Gödel’s result concerning the proofs of the consistency of formal mathematical
theories has had a decisive impact on research in properties of formal theories.

501
Instead of looking for direct proofs of inconsistency of mathematical theories,
mathematicians concentrated largely on relative proofs that demonstrate that
a theory under consideration is consistent if a certain other theory, for example
a formal theory of natural numbers, is consistent. All those relative proofs are
rooted in a deep conviction that even though it cannot be proved that the theory
of natural numbers is free of inconsistencies, it is consistent. This conviction
is confirmed by centuries of development of mathematics and experiences of
mathematicians.
A formal theory is called complete if for every sentence (formula without free
variables) of the language of that theory there is a formal proof of it or of its
negation. A formal theory which does not have this property is called incom-
plete. Hence a formal theory is incomplete if there is a sentence A of the
language of that theory, such that neither A nor ¬A are provable in it. Such
sentences are called undecidable in the theory in question or independent of the
theory.
It might seem that one should be able to formalize a theory such as the formal
theory of natural numbers in a way to make it complete, i.e. free of undecidable
(independent) sentences. But it is not the case in view of Gödel’s Incomplete-
ness Theorem. It states that every consistent formal theory which contains
the arithmetic of natural numbers is incomplete. The Inconsistency Theorem
follows from it. This is why the Incompleteness and Inconsistency Theorems
are now called Gödel First Incompleteness Theorem (theorems 11.3, 11.6) and
Gödel Second Incompleteness (theorems 11.4, 11.7), respectively.
The third part of the Hilbert Program posed and was concerned with the problem
of decidability of formal mathematical theories. A formal theory is called de-
cidable if there is a method of determining, in a finite number of steps, whether
any given formula in that theory is its theorem or not. If a theory is decidable
and if the decision algorithm is known, then the study of problems expressible
in the language of the theory reduces to a purely mechanical procedure. In
undecidable theories there is no mechanical procedure. Most of mathematical
theories are undecidable. Gödel proved in 1931 that the arithmetic of of natural
numbers is undecidable.
We discuss the Hilbert Program and Gödel’s Theorems in more details in sec-
tions 11.3.1 and 11.3.2, respectively.

11.1 Formal Theories: Definition and Examples

We define here a notion of a formal theory based on a predicate (first order)
language. Formal theories are also routinely called first order theories, elemen-
tary theories, formal axiomatic theories, or just theories, when it is clear from
the context that they are formal theories. We will often use the term theory for
simplicity.

502
Remark 11.1
We consider here only classical formal theories based on a complete classical
Hilbert style proof system. We also assume that its language contains the full
set {¬, ∩, ∪, ⇒} of propositional connectives.

Given a classical Hilbert style proof system

H = (L, F, LA, R) (11.1)

with a predicate (first order) language

L = L{¬,∩,∪,⇒} (P, F, C),

where the sets P, F, C are infinitely enumerable.

A formal theory based on H is a proof system obtained from H by adding a
new special set SA of axioms to it, called the set of specific axioms. The specific
axioms are characteristic descriptions of the universe of the formal theory.This
is why we call them specific axioms and denote by SA. The specific axioms are
to be true only in a certain structure as opposed to logical axioms LA that are
true in all structures.

Definition 11.1 (Specific Axioms)

Let SA be a certain set of formulas of L of H = (L, F, LA, R), such that

SA ⊆ F and LA ∩ SA = ∅. (11.2)

We call the set SA a set of specific axioms of H.

Definition 11.2 (Language LSA )

Given a proof system H = (L, F, LA, R) and a non-empty set SA (11.2) of
specific axioms. We define a language

LSA ⊆ L (11.3)

by restricting the sets P, F, C of predicate, functional, and constant symbols

of L to predicate, functional, and constant symbols appearing in the set SA of
specific axioms. Both languages LSA and L share the same set of propositional
connectives.

Obviously, if SA = ∅, then LSA = L.

Now we are ready to define a formal (first order) classical theory as follows.

Definition 11.3 (Formal Theory)

503
A proof system
T = (L, F, LA, SA, R), (11.4)
is called a formal theory with the set SA of specific axioms.
The language LSA defined by (11.3) is called the language of the theory T .
The theory T (11.4) is based on a complete classical proof system

H = (L, F, LA, R).

Definition 11.4
Given a theory T = (L, F, LA, SA, R). We denote by FSA the set of formulas
of the language LSA of T . We denote by T the set all provable formulas in the
theory T , i.e.
T = {B ∈ FSA : SA ` B.} (11.5)
We also write `T B to denote that B ∈ T.

Definition 11.5 (Theory with Equality)

A theory T is called a theory with equality if and only if its language LSA
has as one of its predicates, a two argument predicate P which we denote by =,
and all Equality Axioms (11.6) are provable in T .

Equality Axioms (11.6)

For any any free variable or constant of LSA , R ∈ P, and t ∈ T, where R is an
arbitrary n-ary relation symbol of LSA and t ∈ T is an arbitrary n-ary term of
LSA the following properties hold.
E1 u = u,
E2 (u = w ⇒ w = u),
E3 ((u1 = u2 ∩ u2 = u3 ) ⇒ u1 = u3 ),
E4 ((u1 = w1 ∩ ... ∩ un = wn ) ⇒ (R(u1 , ..., un ) ⇒ R(w1 , ..., wn ))),
E5 ((u1 = w1 ∩ ... ∩ un = wn ) ⇒ (t(u1 , ..., un ) ⇒ t(w1 , ..., wn ))).

Directly from above definitions we have the following.

Fact 11.1 The Hilbert style proof system H defined in chapter 9 is a theory
with equality with the set of specific axioms SA = ∅.

Some Examples of Formal Theories

Formal theories are abstract models of real mathematical theories we develop
using laws of logic. Hence the theories we present here are based on a complete

504
proof system H for classical predicate logic with a language
L = (L{¬,∩,∪,⇒} (P, F, C).

The first order formal theories are also called Elementary Theories.
T 1. Theory of equality
Language
LT 1 = L{¬,⇒,∪,∩} (P = {P }, F = ∅, C = ∅),
where # P = 2, i.e. P is a two argument predicate. The intended interpretation
of P is equality, so we use the equality symbol = instead of P . We write x = y
instead = (x, y). We write the language of T 1 as follows.
LT 1 = L{¬,⇒,∪,∩} ({=}, ∅, ∅).

Specific Axioms
e1 x = x,
e2 (x = y ⇒ y = x),
e3 (x = y ⇒ (y = z ⇒ x = z)),
for any x, y, z ∈ V AR,

Exercise 11.1
Show that the theory T 1 of equality is a theory with equality of definition 11.5.

Solution
The first to axioms e1, e2 are particular cases of E1, E2. We have only to show
that the axiom E3 is provable in T 1, i.e. that the formula
((x = y ∩ y = z) ⇒ x = z) ∈ T1, (11.7)
where, by (11.5) T1 = {A ∈ F{e1,e2,e3} : {e1, e2, e3} ` A}.
Observe that by definition, T 1 is based on a complete Hilbert style proof system.
A formula
(((A ⇒ (B ⇒ C)) ⇒ ((A ∩ B) ⇒ C))
is a predicate tautology, hence is provable in T 1 for any A, B, C ∈ F{e1,e2,e3} .
In particular its instance for A : x = y, B : y = z, C : x = z is also provable in
T 1 what means that
(((x = y ⇒ (y = z ⇒ x = z)) ⇒ ((x = y ∩ y = z) ⇒ x = z)) ∈ T1. (11.8)
Applying Modus Ponens (MP) to axiom e3 and (11.8), we get that
((x = y ∩ y = z) ⇒ x = z) ∈ T1.
It proves that (11.7) holds and ends the proof.

505
Observation 11.1 We have chosen to write the specific axioms as open for-
mulas. Sometimes it is more convenient to write them as closed formulas (sen-
tences). In this case new axioms will be closures of axioms that were open
formulas.

Taking closures of axioms of T 1 we obtain the following new formalization.

T 2. Theory of equality (2)

We adopt a closure of the axioms e1, e2, e3, i.e. the following new set of axioms.
Specific Axioms
(e1) ∀x(x = x),
(e2 ) ∀x∀y(x = y ⇒ y = x),
(e3) ∀x∀y∀z(x = y ⇒ (y = z ⇒ x = z)).

T 3. Theory of Partial Order

Partial order relation is also called order relation.
Language
LT 1 = L{¬,⇒,∪,∩} (P = {P, Q}, F = ∅, C = ∅),
where # P = 2, i.e. P is a two argument predicate. The intended interpretation
of P is equality, so we use the equality symbol = instead of P . We write x = y
instead = (x, y).
Q is a two argument predicate. The intended interpretation of Q is partial
order, called also order relation, so we use the order symbol ≤ instead of Q. We
write x ≤ y instead ≤ (x, y).
The language is
LT 3 = L{¬,⇒,∪,∩} ({=, ≤}, ∅, ∅).

Specific Axioms
There are two groups of specific axioms: equality and order axioms. We adopt
equality axioms (11.6) to the language LT 3 as follows.
Equality Axioms
For any x, y, z, x1 , x2 , y1 , y2 , ∈ V AR,
e1 x = x,
e2 (x = y ⇒ y = x),
e3 ((x = y ∩ y = z) ⇒ x = z),
e4 ((x1 = y1 ∩ x2 = y2 ) ⇒ (x1 ≤ x2 ⇒ y1 ≤ y2 )).

506
Partial Order Axioms
o1 x ≤ x, (reflexivity)
o2 ((x ≤ y ∩ y ≤ x) ⇒ x = y), (antisymmetry)
o3 ((x ≤ y ∩ y ≤ z) ⇒ x ≤ z), (trasitivity )
where x, y, z ∈ V AR.
The model of T 3 is called a partially ordered structure.

T 4. Theory of Partial Order (2)

Here is another formalization for partial order.
Language
LT 4 = L{¬,⇒,∪,∩} (P = {P }, F = ∅, C = ∅),
where # P = 2, i.e. P is a two argument predicate. The intended interpretation
of P (x, y) is x < y , so we use the ”less” symbol < instead of P . We write x < y
instead < (x, y). We also write x 6< y for ¬(x < y), i.e. ¬ < (x, y).
The language of T 4 is

LT 4 = L{¬,⇒,∪,∩} ({<}, ∅, ∅).

Specific Axioms
For any x, y, z ∈ V AR,
p1 x 6< x, (irreflexivity)
p2 ((x ≤ y ∩ y ≤ z) ⇒ x ≤ z). (trasitivity )

T 5. Theory of Linear Order

Linear order relation is also called total order relation.
Language
LT 5 = L{¬,⇒,∪,∩} ({=, ≤}, ∅, ∅).

Specific Axioms
We adopt all axioms of theory T 3 of partial order and add the following addi-
tional axiom.
o4 (x ≤ y) ∪ (y ≤ x).

This axiom says that in linearly ordered sets each two elements are comparable.

507
T 6. Theory of Dense Order
Language
LT 6 = L{¬,⇒,∪,∩} ({=, ≤}, ∅, ∅).

Specific Axioms
We adopt all axioms of theory T 5 of linear order and add the following additional
axiom. We write x 6= y for ¬(x = y), i.e. for the formula ¬ = (x, y).
o5 ((x ≤ y ∩ x 6= y) ⇒ ∃z((x ≤ z ∩ x 6= z) ∩ (z ≤ y ∩ z 6= y))).
This axiom says that in linearly ordered sets between any two different elements
there is a third element between them, respective to the order.

T 7. Lattice Theory
Language

LT 7 = L{¬,⇒,∪,∩} (P = {P, Q}, F = {f, g}, C = ∅),

where # P = 2, i.e. P is a two argument predicate. The intended interpretation

of P is equality, so we use the equality symbol = instead of P . We write x = y
instead = (x, y).
Q is a two argument predicate. The intended interpretation of Q is partial
order, called also order relation, so we use the order symbol ≤ instead of Q. We
write x ≤ y instead ≤ (x, y).
# f = # g 2, i.e. f, g are a two argument functional symbols. The intended
interpretation of f, g is the lattice intersection ∧ and union ∨, respectively.
We write (x ∧ y) for ∧(x, y) and (x ∨ y) for ∨(x, y).
Observe that (x ∩ y), (x ∪ y) are atomic formulas of LT 7 and (x ∧ y) and
(x ∨ y) are terms of LT 7 .
We write the language as

LT 7 = L{¬,⇒,∪,∩} ({=, ≤}, {∧, ∨}, ∅).

Specific Axioms
There are three groups of specific axioms: equality axioms, order axioms, and
lattice axioms. We adopt equality axioms (11.6) to the language LT 7 as follows.

Equality Axioms
For any x, y, z, x1 , x2 , y1 , y2 , ∈ V AR,
e1 x = x,
e2 (x = y ⇒ y = x),

508
e3 ((x = y ∩ y = z) ⇒ x = z),
e4 ((x1 = y1 ∩ x2 = y2 ) ⇒ (x1 ≤ x2 ⇒ y1 ≤ y2 )),
e5 ((x1 = y1 ∩ x2 = y2 ) ⇒ (x1 ∧ x2 ⇒ y1 ∧ y2 )),
e6 ((x1 = y1 ∩ x2 = y2 ) ⇒ (x1 ∨ x2 ⇒ y1 ∨ y2 )).

Remark 11.2
We write ∧ for the lattice functional symbol of intersection in order to better
distinguish it from the conjunction symbol ∩ in the formula.
The same applies to the next axiom e7 that involves lattice functional symbol ∨
for the union and disjunction symbol ∪ in the formula.

Partial Order Axioms

For any x, y, z ∈ V AR,
o1 x ≤ x, (reflexivity)
o2 ((x ≤ y ∩ y ≤ x) ⇒ x = y), (antisymmetry)
o3 ((x ≤ y ∩ y ≤ z) ⇒ x ≤ z). (trasitivity)

Lattice Axioms
For any x, y, z ∈ V AR,
b1 (x ∧ y) = (y ∧ x), (x ∧ y) = (x ∧ y),
b2 (x ∧ (y ∧ z)) = ((x ∧ y) ∧ z), (x ∨ (y ∨ z)) = ((x ∨ y) ∨ z),
b3 (((x ∧ y) ∨ y) = y), ((x ∧ (x ∨ y)) = x).

T 8. Theory of Distributive Lattices

Language
LT 8 = L{¬,⇒,∪,∩} ({=, ≤}, {∧, ∨}, ∅).

Specific Axioms
We adopt all axioms of theory T 7 of lattice theory and add the following addi-
tional axiom.
b4 (x ∧ (y ∨ z)) = ((x ∧ y) ∨ (x ∧ z)).

T 9. Theory of Boolean Algebras

Language
LT 9 = L{¬,⇒,∪,∩} ({=, ≤}, {∧, ∨, −}, ∅),

509
where − is one argument function symbol representing algebra complement.
Specific Axioms
We adopt all axioms of theory T 8 of distributive lattices theory and add the
following additional axioms that characterize the algebra complement −.
b5 (((x ∧ −x) ∨ y) = y), (((x ∨ −x) ∧ y) = y).

T 10. Theory of Groups

Language
LT 10 = L{¬,⇒,∪,∩} (P = {P }, F = {f, g}, C = {c}),
where # P = 2, i.e. P is a two argument predicate. The intended interpretation
of P is equality, so we use the equality symbol = instead of P . We write x = y
instead = (x, y).
f is a two argument functional symbol. The intended interpretation of f is
group operation ◦. We write (x ◦ y) for the formula ◦(x, y).
g is a one argument functional symbol. g(x) represent a group inverse element
to a given x and we denote it by x−1 . We hence use a symbol −1 for g.
c is a constant symbol representing unit element in the group and we use a
symbol e to denote it

−1
LT 10 = L{¬,⇒,∪,∩} ({=}, {◦, }, {e}).

Specific Axioms
There are two groups of specific axioms: equality axioms and group axioms. We
adopt equality axioms (11.6) to the language LT 10 as follows.

Equality Axioms
For any x, y, z, x1 , x2 , y1 , y2 , ∈ V AR,
e1 x = x,
e2 (x = y ⇒ y = x),
e3 ((x = y ∩ y = z) ⇒ x = z),
e4 (x = y ⇒ x−1 = y −1 ),
e5 ((x1 = y1 ∩ x2 = y2 ) ⇒ (x1 ◦ x2 ⇒ y1 ◦ y2 )).

Group Axioms
g1 (x ◦ (y ◦ z)) = ((x ◦ y) ◦ z),
g2 (x ◦ e) = x,

510
g3 (x ◦ x−1 ) = e.

T 11. Theory of Abelian Groups

Language is the same as LT 11 , i.e.

−1
LT 11 = L{¬,⇒,∪,∩} ({=}, {◦, }, {e})0

Specific Axioms
We adopt all axioms of theory T 11 of groups and add the following additional
axiom.
g4 (x ◦ y) = (y ◦ x).

T 12. Theory of Groups (2)

Here is another formalization for group theory.
Language
LT 12 = L{¬,⇒,∪,∩} (P = {P }, F = {f }, C = {c}),
where # P = 2, i.e. P is a two argument predicate. The intended interpretation
of P is equality, so we use the equality symbol = instead of P . We write x = y
instead = (x, y).
f is a two argument functional symbols. The intended interpretation of f is
group operation ◦. We write (x ◦ y) for the formula ◦(x, y).
c is a constant symbol representing unit element in the group and we use a unit
symbol e to denote it.

LT 12 = L{¬,⇒,∪,∩} ({=}, {◦}, {e}).

Specific Axioms
For any x, y, z, x1 , x2 , y1 , y2 , ∈ V AR,
a1 (x ◦ (y ◦ z)) = ((x ◦ y) ◦ z),
a2 (x ◦ e) = x,
a3 ∀x∃y((x ◦ y) = e),
a4 x = x,
a5 (x = y ⇒ y = x),
a6 (x = y ⇒ (y = z ⇒ x = z)),
a7 (x = y ⇒ (x ◦ z = y ∩ z ◦ x = z ◦ y)).

511
T 13. Theory of Abelian Groups (2)
We adopt the language and all axioms of theory T 12 of groups and add the
following additional axiom.
a8 (x ◦ y) = (y ◦ x).

Observe that what we formally prove in the formal axiomatic theories presented
here represents only fragments of corresponding axiomatic theories developed in
mathematics. For example Group Theory, Boolean Algebras Theory are fields
in mathematics and many theorems developed there, like the Representation
Theorem for Boolean Algebras, and many, many others in other domains can not
be expressed in the languages of respective formal theories. This is a reason why
we also call them elementary theories. For example, we say elementary group
theory to distinguish it from the Group Theory as a lager field of mathematics.

11.2 PA: Formal Theory of Natural Numbers

Next to geometry, the theory of natural numbers is the most intuitive and
intuitively known of all branches of mathematics. This is why the first attempts
to formalize mathematics begin with with arithmetic of natural numbers. The
first attempts of axiomatic formalization of arithmetic of natural numbers was
given by Dedekind in 1879 and by Peano in 1889. The Peano formalization
became known as Peano Postulates (axioms) and can be written as follows.
p1 0 is a natural number.
p2 If nis a natural number, there is another number which we denote by n0 .
We call n0 a successor of n. The intuitive meaning of n0 is n + 1.
p3 0 6= n0 , for any natural number n.
p4 If n0 = m0 , then n = m, for any natural numbers n, m.
p5 If W is is a property that may or may not hold for natural numbers, and
if (i) 0 has the property W and (ii) whenever a natural number n has the
property W, then n0 has the property W,
then all natural numbers have the property W.
p5 is called Principle of Induction.
These axioms, together with a certain amount of set theory, are sufficient to
develop not only theory of natural numbers, but also theory of rational and
even real numbers. But they can’t act as a fully formal theory as they include
intuitive notions like ”property” and ”has a property”.
A formal theory of natural numbers based on Peano Postulates is referred in

512
literature as Peano Arithmetic, or simply PA. We present here formalization by
Mendelson (1973) that is included and worked out in smallest details in his book
Intoduction to Mathematical Logic (1987). We refer the reader to this excellent
book for details and further reading.

We assume, as we did in the previous section ?? that T 14 and other theories

considered here are based on a complete Hilbert style proof system

H = (L, F, LA, R) (11.9)

for classical predicate logic with a language

L = (L{¬,∩,∪,⇒} (P, F, C),

We additionally assume now that the system H has as one of ts inference rules
a a generalization rule
A(x)
(G) . (11.10)
∀xA(x)
We do so to facilitate use the Mendelson’s book as a supplementary reading to
the material included here and for additional reading for material not covered
here.

Remark 11.3
The Deduction Theorem as proved in chapter 9 holds for the proof system system
H defined by (11.9).

T 14. Peano Arithmetic P A (11.11)

Language
LP A = L(P = {P }, F = {f, g, h}, C = {c}),
where # P = 2, i.e. P is a two argument predicate. The intended interpretation
of P is equality, so we use the equality symbol = instead of P . We write x = y
instead = (x, y). We write x 6= y for ¬(x = y).
f is a one argument functional symbol. f (x) represent the successor of a given
x and we denote it by x0 . We hence use a symbol 0 for f .
g, h is are two argument functional symbols. The intended interpretation of f is
addition and the intended interpretation of g is multiplication. We write x + y
for f (x, y) and x · y for g(x, y).
c is a constant symbol representing zero and we use a symbol 0 to denote c.

LP A = L{¬,⇒,∪,∩} ({=}, { 0 , +, ·}, {0}).

Specific Axioms

513
P1 (x = y ⇒ (x = z ⇒ y = z)),
P2 (x = y ⇒ x0 = y 0 ),
P3 0 6= x0 ,
P4 (x0 = y 0 ⇒ x = y),
P5 x + 0 = x,
P6 x + y 0 = (x + y)0
P7 x · 0 = 0,
P8 x · y 0 = (x · y) + x,
P9 (A(0) ⇒ (∀x(A(x) ⇒ A(x0 ) ⇒ ∀xA(x)))),
for all formulas A(x) of LP A and all x, y, z ∈ V AR.

The axiom P9 is called Principle of Mathematical Induction. It does not

fully corresponds to Peano Postulate p5 which refers intuitively to all possible
properties on natural numbers (uncountably many). The P7 axiom applies only
to properties defined by infinitely countably formulas of A(x) of LP A .
Axioms P3, P4 correspond to Peano Postulates p3, p4. The Postulates p1, p2
are fulfilled by presence of 0 and successor function. Axioms P1, P2 deal with
some needed properties of equality that were probably assumed as intuitively
obvious by Peano and Dedekind. Axioms P5 - P8 are the recursion equations
for addition and multiplication. They are not stated in the Peano Postulates
as Dedekind and Peano allowed the use of intuitive set theory within which
the existence of addition and multiplication and their properties P5-P8 can be
proved (Mendelson, 1973).
Observe that while axioms P1 - P6 of theory of Peano Arithmetic PA are par-
ticular formulas of LP A , the axiom P9 as an axiom schema providing an infinite
number of axioms. This means that the set of axioms P1 - P9 does not provide
a finite axiomatization for Peano Arithmetic. But any formalization of Peano
Postulates must include formalization of the Principle of Induction p5 and hence
must contain some form of induction axiom P9. It was proved formally in 1952
by Czeslaw Ryll-Nardzewski and Rabin in 1961.

Theorem 11.1 (Ryll-Nardzewski)

Peano Arithmetic is is not finitely axiomatizable; that is there is no theory K
having inly a finite number of proper axioms, whose theprems are the same as
those of P A.

By definition 11.4, given a theory T , we denote by T the set all provable formulas
in T . In particular, PA denotes the set of all formulas provable in Peano
Arithmetic PA.

514
Theory P A is one of many formalizations of Peano Arithmetic. They all repre-
sent what we call Peano Arithmetic if they have the same set of theorems. We
adopt hence the following definition.

Definition 11.6 Any theory T such that T = PA for PA defined by (11.11)

is called a Peano arithmetic.

Taking closure of axioms P1 - P8 of T 14 we obtain new theory T 15 . The axiom

P9 is a sentence (closed formula) already.
T15. Theory CPA

LT 15 = LT 14 = L{¬,⇒,∪,∩} ({=}, {0 , +, ·}, {0}).

We denote the specific axioms of T 15 by CP A to express that its specific axioms
are closures of specific axioms of P A.
Specific Axioms
C1 ∀x∀y∀z(x = y ⇒ (x = z ⇒ y = z)),
C2 ∀x∀y(x = y ⇒ x0 = y 0 ),
C3 ∀x(0 6= x0 ),
C4 ∀x∀y(x0 = y 0 ⇒ x = y),
C5 ∀x(x + 0 = x),
C6 ∀x∀y(x + y 0 = (x + y)0 )
C7 ∀x(x · 0 = 0),
C8 ∀x∀y(x · y 0 = (x · y) + x),
C9 (A(0) ⇒ (∀x(A(x) ⇒ A(x0 )) ⇒ ∀xA(x))),
for all formulas A(x) of LP A and all x, y, z ∈ V AR.

Here is a very simple exercise.

Fact 11.2
Theory CP A is a Peano Arithmetic.

Proof
By definition 11.6 we have to show that PA = CPA. Observe that LCP A = LP A
, so we have to show that for any formula B of LP A ,

`P A B if and only if `CP A B. (11.12)

515
Both theories are based on the same Hilbert proof system H, so to prove (11.12)
means to prove that
(1) all axioms C1 − C8 of CP A are provable in P A and vice versa,
(2) all axioms P 1 − P 8 of LP A are provable in CP A.
Here are detailed proofs for axioms P1, and C1. The proofs for other axioms
follow the same pattern.
(1) We prove that the axiom C1 ∀x∀y∀z(x = y ⇒ (y = z ⇒ x = z)) is
provable in P A as follows.
Observe that axioms of CPA are closures of respective axioms of P A. Consider
axiom P1: (x = y ⇒ (y = z ⇒ x = z)). As the proof system H has a
generalization rule 11.10
A(x)
(G)
∀xA(x)
as its rule of inference, we obtain a proof B1, B2, B3, B4 of C1 as follows.
B1: (x = y ⇒ (x = z ⇒ y = z)), (axiom P1)
B2: ∀z(x = y ⇒ (x = z ⇒ y = z)), (GA)
B3: ∀y∀z(x = y ⇒ (x = z ⇒ y = z)), (GA)
B4: ∀x∀y∀z(x = y ⇒ (x = z ⇒ y = z)). (axiom C1)
This ends the proof of (1) for axioms P1, and C1.
(2) We prove that the axiom P1 (x = y ⇒ (y = z ⇒ x = z)) of LP A is
provable in CP A as follows.
By H completeness a predicate tautology

(∀xA(x) ⇒ A(t)), (11.13)

where term t is free for x in A(x) is provable in H for any formula A(x) of L and
hence for any formula A(x) of its particular sublanguage LP A . So its particular
case for A(x) = (x = y ⇒ (x = z ⇒ y = z)) and t = x is provable in CP A, i.e.
we have that the formula

(∀x∀y∀z(x = y ⇒ (x = z ⇒ y = z)) ⇒ ∀y∀z(x = y ⇒ (x = z ⇒ y = z)))

is provable in CP A.
We construct a proof B1, B2, B3, B4, B5, B6, B7 of P1 in CP A in as follows.
B1 ∀x∀y∀z(x = y ⇒ (x = z ⇒ y = z)), (axiom C1)
B2 (∀x∀y∀z(x = y ⇒ (x = z ⇒ y = z)) ⇒ ∀y∀z(x = y ⇒ (x = z ⇒ y = z))),
by (11.15)
B3 ∀y∀z(x = y ⇒ (x = z ⇒ y = z)), MP on B1, B2

516
B4 (∀y∀z(x = y ⇒ (x = z ⇒ y = z)) ⇒ ∀z(x = y ⇒ (x = z ⇒ y = z))), by
(11.15)
B5 ∀z(x = y ⇒ (x = z ⇒ y = z)), MP on B3, B4
B6 (∀z(x = y ⇒ (x = z ⇒ y = z)) ⇒ (x = y ⇒ (x = z ⇒ y = z))), by (11.15)
B7 (x = y ⇒ (x = z ⇒ y = z)) MP on B5, B6
This ends the proof of (2) for axioms P1, and C1.
The proofs for other axioms is similar and are left as homework assignment.
Here are some more basic facts about P A.

Fact 11.3
The following formulas are provable in P A for any terms t, s, r of LP A .
P1’ (t = r ⇒ (t = s ⇒ r = s)),
P2’ (t = r ⇒ t0 = r0 ),
P3’ 0 6= t0 ,
P4’ (t0 = r0 ⇒ t = r),
P5’ t + 0 = t,
P6’ t + r0 = (t + r)0
P7’ t · 0 = 0,
P8’ t · r0 = (t · r) + t.

We named the properties as P1’- P8’ to stress the fact that they are generaliza-
tions of axioms P1 - P8 to the set of all terms of LP A .
Proof
We write the proof for P1’ as an example. Proofs of all other formulas follow
the same pattern.
Consider axiom P1: (x = y ⇒ (y = z ⇒ x = z)). By Fact 11.2 its closure
∀x∀y∀z(x = y ⇒ (x = z ⇒ y = z)) is provable in T hP A , i.e.

`P A ∀x∀y∀z(x = y ⇒ (x = z ⇒ y = z)) (11.14)

By H completeness a predicate tautology

(∀xA(x) ⇒ A(t)), (11.15)

where term t is free for x in A(x) is provable in H for any formula A(x) of L and
hence for any formula A(x) of its particular sublanguage LP A . So its particular
case for A(x) = ∀y∀z(x = y ⇒ (x = z ⇒ y = z)) the formula (11.15) is provable
in T hP A . Observe that any term t is free for x in this particular A(x). We get

517
that for any term t,

`P A (∀x∀y∀z(x = y ⇒ (x = z ⇒ y = z)) ⇒ ∀y∀z(t = y ⇒ (t = z ⇒ y = z))),

(11.16)
Applying MP to (11.14) and (11.16) we get that for any t

`P A ∀y∀z(t = y ⇒ (t = z ⇒ y = z)). (11.17)

Observe that any term r is free for for y in ∀z(t = y ⇒ (t = z ⇒ y = z)). so we

have that for all terms r

`P A (∀y∀z(t = y ⇒ (t = z ⇒ y = z)) ⇒ ∀z(t = r ⇒ (t = z ⇒ r = z))),

(11.18)
as a particular case of 11.15). Applying MP to (11.17) and (11.18) we get that
for any terms t, r
`P A ∀z(t = r ⇒ (t = z ⇒ r = z)). (11.19)
Observe that any term s is free for for z in (t = r ⇒ (t = z ⇒ r = z)). so we
have that

`P A (∀z(t = y ⇒ (t = z ⇒ y = z)) ⇒ (t = r ⇒ (t = s ⇒ r = s))), (11.20)

for all terms r, t, s as a particular case of 11.15). Applying MP to (11.19) and

(11.20) we get that for any t, r

`P A (t = r ⇒ (t = s ⇒ r = s)).

This ends the proof of P 0 .

The proofs of properties P2’ - P8’ follow the same pattern and are left as a
homework assignment.

Fact 11.4
The following formulas are provable in P A for any terms t, s, r of LP A .
a1 t = t,
a2 (t = r ⇒ r = t),
a3 (t = r ⇒ (r = s ⇒ t = s)),
a4 (r = t ⇒ (t = s ⇒ r = s)),
a5 (t = r ⇒ (t + s = r + s)),
a6 t = 0 + t.

Proof
We use in the proof Fact 11.2, Fact 11.3, axioms of PA (11.11, and completeness
of the system H. We denote it in the comments. The details of the steps

518
are similar to the proof of Fact 11.3 and is left to the reader as as homework
assignment.
a1 We construct a proof of t = t in CP A in as follows.
B1 t + 0 = t, P5’ in Fact 11.3
B2 (t + 0 = t ⇒ (t + 0 = t ⇒ t = t)), P1’ in Fact 11.3 for t = t + 0, r = t, s = t
B3 (t + 0 = t ⇒ t = t), MP on B1, B2
B4 t = t. MP on B1, B3
a2 We construct a proof of (t = r ⇒ r = t) as follows.
B1 (t = r ⇒ (t = t ⇒ r = t)), P1’ in Fact 11.3 for r = t, s = t
B2 (t = t ⇒ (t = r ⇒ r = t)), B1, tautology
B3 t = r ⇒ r = t. MP on B2, a1
a3 We construct a proof of (t = r ⇒ (r = s ⇒ t = s)) as follows.
B1 (r = t ⇒ (r = s ⇒ t = s)), P1’ in Fact 11.3
B2 t = r ⇒ r = t, a2
B3 (t = r ⇒ r = t). MP on B1, B2
a4 We construct a proof of (r = t ⇒ (t = s ⇒ r = s)) as follows.
B1 (r = t ⇒ (t = s ⇒ r = s)), a3 for t = r, r = t
B2 (t = s ⇒ (r = t ⇒ r = s)), B1, tautology
B3 s = t ⇒ t = s, a2
B4 (s = t ⇒ (r = t ⇒ r = s)), B1, B2, tautology
B5 (r = t ⇒ (t = s ⇒ r = s)). B4, tautology

a5 We prove (t = r ⇒ (t + s = r + s)) by the Principle of Mathematical Induc-

tion P9 (A(0) ⇒ (∀x(A(x) ⇒ A(x0 ) ⇒ ∀xA(x)))).

The proof uses the Deduction Theorem which holds for the proof system H
(Remark 11.3) and so can be use in P A.
We first apply the Induction Rule to A(z) : (x = y ⇒ x + z = y + z) to prove

`P A ∀z(x = y ⇒ x + z = y + z).

(i) We prove that `P A A(0), i.e. `P A (x = y ⇒ x + 0 = y + 0). Here the steps

in the proof.
B1 x + 0 = x, P5’

519
B2 y + 0 = y, P5’
B3 x = y, Hyp
B4 (x + 0 = x ⇒ (x = y ⇒ x + 0 = y), a3 for t = x + 0, r = x, s = y
B5 (x = y ⇒ x + 0 = y), MP on B1, B4
B6 x + 0 = y, MP on B3, B5
B7 (x+0 = y ⇒ (y+0 = y ⇒ x+0 = y+0), a4 for r = x+0, t = y, s = y = 0
B8 (y + 0 = y ⇒ x + 0 = y + 0), MP on B6, B7
B9 x + 0 = y + 0), MP on B2, B8
B10 (x = y ⇒ x + 0 = y + 0). B1- B9, Deduction Theorem
Thus, `P A A(0).
(ii) We prove that `P A ∀z(A(z) ⇒ A(z 0 ), i.e.
∀z((x = y ⇒ x + z = y + z) ⇒ (x = y ⇒ x + z 0 = y + z 0 )). Here the steps in
the proof.
C1 (x = y ⇒ x + z = y + z), Hyp
C2 x = y, Hyp
C3 x + z = (x + z)0 ,
0
P6’
C4 y + z 0 = (y + z)0 , P6’
C5 x + z = y + z), MP on B1, B2
C6 (x + z = y + z ⇒ (x + z)0 = (y + z)0 ) P2’ for t = x + z, r = y + z,
C7 (x + z)0 = (y + z)0 , MP on B5, B6
0 0
C8 x + z = y + z , a3 substitution and MP on B3, B7
C9 ((x = y ⇒ x + z = y + z) ⇒ x + z 0 = y + z 0 ) B1- B8, Deduction Theorem
This proves ` A(z) ⇒ A(z 0 ).
C10 (((x = y ⇒ x + 0 = y + 0) ⇒ ((x = y ⇒ x + z = y + z) ⇒ x + z 0 =
y + z 0 )) ⇒ ∀z(x = y ⇒ x + z = y + z)), P9 for A(z) : (x = y ⇒ x + z = y + z)
C11 ((x = y ⇒ x + z = y + z) ⇒ x + z 0 = y + z 0 )) ⇒ ∀z(x = y ⇒ x + z = y + z),
MP on C10 and B10
C12 ∀z(x = y ⇒ x + z = y + z), MP on C11 and C9
C13 ∀y∀z(x = y ⇒ x + z = y + z), (GA)
C14 ∀x∀y∀z(x = y ⇒ x + z = y + z), (GA)
Now we repeat here the proof of P1’ of Fact 11.3. We apply it step by step to

520
C14. We eliminate the quantifiers ∀x∀y∀z and replace variables x, y, z by terms
t, r, s using the tautology (11.15) (∀xA(x) ⇒ A(t)) and Modus Ponens. Finally,
we obtain the proof of a5, i.e.
`P A (t = r ⇒ (t + s = r + s)).

We go on proving other basic properties of addition and multiplications includ-

ing for example the following.

Fact 11.5
The following formulas are provable in P A for any terms t, s, r of LP A .
(i) t · (r + s) = (t · r) + (t · s), distributivity
(ii) (r + s) · t = (r · t) + (s · t), distributivity
(iii) (r · t) · s = r · (t · s), associativity of ·
(iv) (t + s = r + s ⇒ t = r), canlcellation law for +

Proof
(i) Prove `P A t · (x + z) = (x · y) + (x · z) by induction on z.
(ii) Prove from (i) and property t · r = r · t.
(iii) Prove `P A (x · y) · z = x · (y · z) by induction on z.
(iv) Prove (x + z = y + z ⇒ x = y) by induction on z

Definition 11.7 (Numerals)

The terms 0, 00 , 000 , 0000 , . . . are called numerals and denoted by 0, 1, 2, 3, . . . .
More precisely,
(1) 0 is 0,
(2) for any natural number n, n + 1 is (n)0 .

In general, if n is a natural number, n stands for the corresponding numeral

0” . . .0 , i.e. by 0 followed by n strokes.
The numerals can be defined recursively as follows.
(1) 0 is a numeral,
(2) if u is a numeral, then u0 is also a numeral.
Here are some more of many properties, intuitively obvious, that provable in
T hP A . We give some proofs and an example, and leave the others as an exercise.
Reminder
We use numerals n, m as un abbreviation of the terms r, s they represent.

Fact 11.6 The following formulas are provable in P A for any terms t, s of
LP A .

521
1. t + 1 = t0 ,
2. t · 1 = t,
3. t · 2 = t + t,
4. (t + s = 0 ⇒ (t = 0 ∩ s = 0)),
5. (t 6= 0 ⇒ (s · t = 0 ⇒ s = 0)),

Proof
1. Major steps in the proof of t + 1 = t0 in P A are as follows.
The comments at each step explain how to reconstruct the formal proof from
the properties already proven.
B1 t + 00 = (t + 0)0 , P6’
B2 t + 0 = t, P5’
B3 (t + 0)0 = t0 , B2, P2’, MP
0 0
B4 t + 0 = t , B1, B3, Fact 11.4 a3, MP
B5 t + 1 = t0 . B4, abbreviation
2. Major steps in the proof of t · 1 = t in P A are as follows.
B1 t · 00 = t · 0 + t, P8’
B2 t · 0 = 0, P7’
B3 (t · 0) + t = 0 + t, B1, Fact 11.4 a4, MP
B4 t · 00 = 0 + t, B1, B3, Fact 11.4 a3, MP
B5 0 + t = t, Fact 11.4 a3, a6, MP
B6 t · 00 = t, B4, B5, Fact 11.4 a3, MP
B7 t + 1 = t0 , B6, abbreviation
3. Major steps in the proof of t · 2 = t + t in P A are as follows.
B1 t + 10 = (t · 10 ) + t, P8’
B2 t + 1 = t0 , part 2.
B3 (t · 1) + t = t + t, B2, Fact 11.4 a5, MP
B4 t · 10 = t + t, B1, B3, Fact 11.4 a3, MP
B5 t · 2 = t + t, B4, abbreviation
4. We prove (t + s = 0 ⇒ (t = 0 ∩ s = 0)) by the following steps.
(s1) We apply the Principle of Mathematical Induction to A(y) : (x + y = 0 ⇒

522
(x = 0 ∩ y = 0)) and prove

∀y(x + y = 0 ⇒ (x = 0 ∩ y = 0)). (11.21)

(s2) We apply the generalization rule (G) to (5.24) and get

∀x∀y(x + y = 0 ⇒ (x = 0 ∩ y = 0)). (11.22)

(s3) We repeat here the proof of P1’ of Fact 11.3. We apply it step by step to
(11.22). We eliminate the quantifiers ∀x∀y and replace variables x, y by terms
t, s using the tautology (11.15) (∀xA(x) ⇒ A(t)) and Modus Ponens. Finally,
we obtain the proof of 4., i.e.

`P A (t + s = 0 ⇒ (t = 0 ∩ s = 0)).

We are going to prove now, as an example, the following.

Fact 11.7
Let n, m be any natural numbers.
(1) If m 6= n, then m 6= n.
(2) m + n = m + n and m · mn = m · n are provable in P A .
(3) Any model for P A is infinite.

Proof
Assume m 6= n, then m < n or n < m. Assume m < n. Her are major steps in
the formal proof of n 6= m.
The proof uses the Deduction Theorem which holds for the proof system H
(Remark 11.3) and so can be use in P A.
B1 m = n, Hyp
B2 0000000 = 000000 B2 is abbreviation of B2 for m applications of 0 on left side
of equation and n applications of 0 on the right
B3 0 = t0 , for t = n − m − 1

We apply P4’ m times a=in a row. Then 0 = 00000 with n − m applications of 0

on right side. Let t = n − m − 1. Since n > m, n − m − 1 ≥ 0. Thus, 0 = t0 .

B4 0 6= t0 , P3’
B5 0 = t ∩ 0 6= t0 ,
0
B3, B4, tautology (A ⇒ (B ⇒ (A ∩ B))), MP
B6 (m = n ⇒ 0 = t0 ∩ 0 6= t0 ), B1, B5, Deduction Theorem
B7 m 6= n B6, tautology ((A ⇒ (C ∩ ¬C)) ⇒ ¬A), MP

523
The proof of the case n < m is similar and left to the reader.
(2) We use mathematical induction for natural numbers in the metalanguage
with respect to natural number n. Base case: m + 0 is m.
By P3’ m = m + 0, hence m + 0 = m + 0 and the base step holds.
Inductive step. Assume that m + n = m + n is provable. By P2’ and P6’ we
get (m + n)0 = m + (n)0 . But m + (n + 1) is (m + n)0 and n + 1 is (n)0 . Hence,
m + (n + 1) = m + n + 1 and by mathematical induction m + n = m + n is
provable in T hP A , for all n, m. The proof that m · mn = m · n is provable in
P A for all n, m is similar.
(3) By (2), in a model for PA the objects corresponding to numerals must be
distinct. But the set of numerals is infinitely countable, so universe of any model
for PS must contain infinitely countable subset and hence is infinite.
An order relation can be introduced by in PA as follows.

Definition 11.8 (Order)

For any terms t, s of LP A , we write
t < s for a formula ∃w(w 6= 0 ∩ w + t = s),
where we choose w to be the first variable not in t or s,
t ≤ s for a formula t < s ∪ t = s,
t > s for a formula s < t,
t ≥ s for a formula s ≤ t,
t 6< s for a formula ¬(t < s), and so on...

Then we prove properties of order relation, for example the following.

Fact 11.8
For any terms t, r, s of LP A , the following formulas are provable in P A.
o1 t ≤ t,
o2 (t ≤ s ⇒ (s ≤ r ⇒ t ≤ r)),
o3 ((t ≤ s ∩ s ≤ t) ⇒ t = s),
o4 (t ≤ s ⇒ (t + r ≤ s + r)),
o5 (r > 0 ⇒ (t > 0 ⇒ r · t > 0)).

There are several stronger forms of the the Principle of Mathematical Induction
P9 (A(0) ⇒ (∀x(A(x) ⇒ A(x0 ) ⇒ ∀xA(x)))) that are provable in P A. Here is
one of them.

524
Fact 11.9 (Complete Induction)
The following formula, called Complete Induction Principle is provable in P A.

(∀x∀z(z < x ⇒ A(z)) ⇒ A(x)) ⇒ ∀xA(x)).

In plain English, Complete Induction Principle says:

consider a property P such that , for any x, if P holds for for all natural num-
bers less then x, then P holds for x also. Then P holds for all natural numbers.

We proved and cited only some of the basic properties corresponding to proper-
ties of arithmetic of natural numbers. There are many more of them, developed
in many Classical Logic textbooks. We refer the reader especially to Mendelson
(1997) that we found the most rigorous and complete. The proofs included here
are more precise and complete versions of the few of the Mendelson’s proofs.
We selected and proved some direct consequences Peano Arithmetic axioms not
only because they are needed as the starting point for a strict development of
the formal theory of arithmetic of natural numbers but also because they are
good examples of how one develops any formal theory.
From this point on one can generally translate onto the language LP A and
prove in the P A the results from any text on elementary number theory. Some
standard results of number theory are proved with the aid of theory of complex
variables and it is often not known whether elementary proofs (or proofs in P A
can be given for such theorems. The statements of some other results of number
theory cannot even be formulated in P A.
Hence a natural question about the strength and expressive powers of P A is a
very important one. We will address it shortly in next section with connection of
the formulation and proofs of Gödel Theorems. Gödel, in order to prove them
developed the huge scientific apparatus which grew into new field of Mathe-
matics of Recursion Theory, and into Theory of Computation with input from
Church and Turing.
We know by Ryll Nardzewski Theorem 11.1 that PA is not finitely axioma-
tizable. We want to bring reader’s attention a finitely axiomatizable proper
sub-theory of PA, RR, that has the same expressive power with respect to the
Gödel Theorems. Here it is, as formalized and discussed in detail in Mendelson’s
book.

T 16. Robinson System RR (11.23)

Language
The language of RR is the same as the language of P A, i.e.

LRR = L{¬,⇒,∪,∩} ({=}, {0 , +, ·}, {0}).

525
Specific Axioms
r1 x = x,
r2 (x = y ⇒ y = x),
r3 (x = y ⇒ (y = z ⇒ x = z)),
r4 (x = y ⇒ x0 = y 0 ),
r5 (x = y ⇒ (x + z = y + z ⇒ z + x = z + y)),
r6 (x = y ⇒ (x · z = y · z ⇒ z · x = z · y)),
r7 (x0 = y 0 ⇒ x = y),
r8 0 6= x0 ,
r9 (x 6= 0 ⇒ ∃y x = y 0 ),
r10 x + 0 = x,
r11 x + y 0 = (x + y)0 ,
r12 x · 0 = 0,
r13 x · y 0 = x · y + x,
r14 (y = x · z + p ∩ ((p < x ∩ y < x · q + r) ∩ r < x) ⇒ p = r).
for any x, y, z, p, q, r ∈ V AR,
Axioms r1 - r13 are due to Robinson (1950), hence the name. Axiom r14 is due
to Mendelson (1973). It expresses the uniqueness of remainder. The relation <
is as defined by definition 11.8.
Gödel showed that there are closed formulas of the language LP A of P A that
are neither provable nor disprovable in P A, if P A is consistent. Hence there is
a formula that is true under standard interpretation but is not provable in P A.
We also see that the incompleteness of P A cannot be attributed to omission
of some essential axiom but has deeper underlying causes that apply to other
theories as well. Robinson proved in 1950, that the Gödel Theorems hold his
system RR. In particular RR has the same incompleteness property as P A.

11.3 Consistency, Completeness, Gödel Theo-

rems

Formal theories, because of their precise structure, became themselves an object

of of mathematical research. The mathematical theory concerned with the study
of formalized mathematical theories is called, after Hilbert, metamathematics.
The most important open problems of metamathematics were introduced by

526
Hilbert as a part of the Hilbert Program. They were concerned with notions of
consistency, completeness, and decidability. The answers to Hilbert problems
were given by Gödel in 1930 in a form of his two theorem. They are some of the
most important and influential results in twentieth century mathematics. We
will discuss here these notions and Gödel’s results.

There are two definitions of consistency: semantical and syntactical.

The semantical one is based on the notion of a model and says, in plain English:
a theory is consistent if the set of its specific axioms has a model.
The syntactical one uses the notion of provability and says: a theory is con-
sistent if one can’t prove a contradiction in it.
We have used, in the proof two of the completeness theorem for propositional
logic (chapter 5) the syntactical definition of consistency. In chapter 9, section
about the reduction predicate logic to propositional logic we used the seman-
tical definition. Both were defined for propositional semantics. We extend
now these definitions to the predicate language, predicate semantics, and for-
mal theories. In order to distinguish these two definitions we call the semantic
one model-consistent, and syntactic one just consistent.

Definition 11.9 (Model for a Theory)

Given a first order theory (definition 11.3)
T = (L, F, LA, SA, R).
Any structure M = [M, I] that is a model for the set SA of the specific axioms
of T , i.e. such that M |= SA, is called a model for the theory T .

Definition 11.10 (Model - Consistent Theory)

A first order theory T = (L, F, LA, SA, R) is model - consistent if and
only if it has a model.

Consider the Peano Arithmetics P A and a structure M = [M, I] for its language
LP A = L{¬,⇒,∪,∩} ({=}, {0 , +, ·}, {0}),
such that the universe M is the set N of natural numbers (nonnegative integers)
and the interpretation I is defined as follows
(1) the constant symbol 0 is interpreted as a natural number 0,
(2) the one argument function symbol 0 (successor) is interpreted as successor
operation (addition of 1) on natural numbers; succ(n) = n + 1,
(3) the two argument function symbols +, · are interpreted as ordinary addition
and multiplication in N,

527
(4) the predicate symbol ”=” is interpreted as equality relation in N.

Definition 11.11 (Standard Model)

We denote M = [N, I] for I defined by (1) - (4) as

M = [N, =, succ, +, · ] (11.24)

and call it a standard model for P A. The interpretation I defined by (1) -

(4) is called a standard interpretation.
Any model for P A in which the predicate symbol ”=” is interpreted as equality
relation in N that is not isomorphic to the standard model is called a nonstan-
dard model for P A.

Observe that if we recognize that the set N of natural numbers with the standard
interpretation, i.e. the structure (11.24) to be a model for P A, then, of course,
P A is consistent (model-consistent). However, semantic methods, involving a
fair amount of set-theoretic reasoning, are regarded by many (and were regarded
as such by Gödel) as too precarious to serve as basis of consistency proofs.
Moreover, we have not proved formally that the axioms of P A are true under
standard interpretation; we only have taken it as intuitively obvious. Hence for
this and other reasons it is common practice to take the model-consistency of
P A as un explicit, unproved assumption and to adopt, after Gödel the following
syntactic definition of consistency.

Definition 11.12 (Consistent Theory)

Given a theory T = (L, F LA, SA, R).
Let T be the set (11.5) of all provable formulas in T .
The theory T is consistent if and only if there is no formula A of the
language LSA such that

`T A and `T ¬A. (11.25)

We also write the condition (11.25) as

A∈T and ¬A ∈ T.

Directly from definition 11.12 we get the definition of inconsistency.

We list it separately for its importance to the proof of the Gödel Theorem 11.4.

Definition 11.13 (Inconsistent Theory)

The theory T = (L, F, LA, SA, R) is inconsistent if and only if there is
a formula A of the language LSA such that

`T A and `T ¬A.

528
Observe that the definitions 11.12, 11.13 have purely syntactic meaning. They
express the common intuition what proper provability should mean. They say
that a provability (formal theory) is a good one (consistent) only when one can’t
prove a formula and its negation; and is inconsistent when it is possible to prove
a contradiction in it.
Here is one of basic characterization of consistent theories.

Theorem 11.2 (Consistent)

A theory T = (L, F, LA, SA, R) based on the proof system H = (L, F, LA, R)
defined by (11.1) is consistent if and only if there is a formula A of the
language LSA such that
A 6∈ T.

Proof
Let denote by CC the consistency condition in the definition 11.12 and by CT
consistency condition in the theorem 11.2.
1. We prove implication ” if CC, then CT”.
Assume not CT. This means that A ∈ T for all formulas A,

A∈T and ¬A ∈ T. (11.26)

In particular there is B such that and B ∈ T and ¬B ∈ T and not CC holds.

2. We prove implication ” if CT, then CD”.
Assume not CD. This means that there is A of LSA , such that A ∈ T. By
definition 11.12 all tautologies are derivable in T. Hence

(((A ∩ B) ⇒ C) ⇒ ((A ⇒ (B ⇒ C)))), ((A ∩ ¬A) ⇒ C) ∈ T, (11.27)

for all A, B, C ∈ F. In particular, when B = ¬A we get that

(((A ∩ ¬A) ⇒ C) ⇒ ((A ⇒ (¬A ⇒ C)))) ∈ T. (11.28)

Applying MP (11.27) and (11.28) we get

((A ⇒ (¬A ⇒ C))) ∈ T. (11.29)

Applying MP twice to (11.29) and (11.26) we get that C ∈ T, for all C. We

proved not CT. This ends the proof of 2. and of the theorem.
Theorem 11.2 often serves a following definition of consistency.

Definition 11.14
A theory T is consistent if and only if T 6= FSA , i.e. there is A of LSA , such
that A 6∈ T.

529
The next important characterization of a formal theory is the one of its complete-
ness understood as the ability of proving or disapproving any of its statements,
provided it is correctly formulated in its language.

Definition 11.15 (Complete Theory)

A theory T = (L, F, LA, SA, R) is complete if and only if for any closed
formula (sentence) A of the language LSA ,

`T A or `T ¬A. (11.30)

We also write the condition (11.30) as

A∈T or ¬A ∈ T. (11.31)

Directly from definition 11.15 we get the definition of incompleteness.

We list it separately for its importance to the proof of the Gödel Incompleteness
Theorem 11.3, Theorem 11.6.

Definition 11.16 (Incomplete Theory)

A theory T is incomplete if and only if there is a closed formula (sentence)
A of the language LSA , such that

6`T A and 6`T ¬A. (11.32)

We also write the condition (11.32) as

A 6∈ T and ¬A 6∈ T.

Any sentence A with the property (11.32) is called an independent, or unde-

cidable sentence of the theory T.

By definition 11.16, in order to prove that a given theory T is incomplete we

have to construct a sentence A of LSA and prove that either A nor ¬A has a
proof in it.
We are now almost ready to discuss Gödel Theorems. One of the most (if
not the most) comprehensive development with detailed and strict proofs of all
needed to carry proofs of Gödel Theorems can be found the Mendelson (1984)
book. The Gödel Theorems chapter in Mendelson book is over 50 pages long,
technically sound and beautiful. We are confident that our readers, at this stage
of our book, are ready and able to follow Mendelson’s or other authors’ work.
We present here a short, high level approach adopting style of Smorynski’s chap-
ter in the Handbook of Mathematical Logic, Studies in Logic and Foundations
of Mathematics, Volume 20 (1977). The chapter also is over 40 pages long (it

530
seems to be a norm when one wants to really prove Gödel’s results). It is writ-
ten in a very condensed and general way and concentrates on presentation of
modern results. It assumes that readers are already familiar with the traditional
approach so beautifully presented in Mendelson’s book, but I encourage readers
to reach for it, as it is, in its own style a very interesting work.
We also want to bring to readers attention that the introduction to Smorynski’s
chapter contains an excellent discussion of Hilbert Program and its relationship
to Gödel’s results. It gives an explanation why and how devastating Gödel The-
orems were to the optimism reflected in Hilbert’s Consistency and Conservation
Programs.

11.3.1 Hilbert’s Conservation and Consistency Programs

Hilbert proposed his Conservation Program and Consistency Programs as re-

sponse to L.E.J. Brouwer and Herman Weyl (1920) propagation of their theory
that existence (as early as 1908) of Zermello’s paradoxes free axiomatization
of set theory makes the need for investigations (and proof) into consistency of
mathematics superfluous. Hilbert decided to intervene. He wrote:
” .... they (Brouwer and Weil) would chop and mangle the science. If we would
follow such a reform as the one they suggest, we would run the risk of losing a
great part of our most valuable treasures!”
Hilbert stated his Conservation Program as follows: To justify the use of
abstract techniques he would show - by as simple and concrete a means as
possible - that the use of abstract techniques was conservative - i.e. that any
concrete assertion one could derive by means of such abstract techniques would
be derivable without them.
We follow Smorynski’s clarification of some of Hilbertian jargon whose exact
meaning was never defined by Hilbert. We hence talk, in the domain of mathe-
matics, about finitistically meaningful statements and finitistic means of proof.
By the finitistically meaningful statements we mean for example identities of
the form
∀x(f (x) = g(x)),
where f, g are reasonably simple functions, for example primitive recursive. We
will call them real statements. Finitistic proofs correspond to computations or
combinatorial manipulations.
More complicated statements are called ideal ones and, as such, have no mean-
ing, but can be manipulated abstractly and the use of ideal statements and
abstract reasoning about them would not allow one to derive any new real
statements, i.e. none which were not already derivable To refute Weyl and
Brouwer, Hilbert required that this latter conservation property itself be finitis-
tically provable.

531
Hilbert’s Consistency Program asks to devise a finitistic means of proving
the consistency of various formal systems encoding abstract reasoning with ideal
statements.
The Consistency Program is a natural outgrowth and successor to the Conser-
vation Program. There are two reasons for this.
R1. Consistency is the assertion that some string of symbols is not provable.
Since derivations are simple combinatorial manipulations, this is a finitistically
meaningful and ought to have a finitistic proof.
R2. Proving a consistency of a formal system encoding the abstract concepts
already establishes the conservation result!
Reason R1 is straightforward. We will discuss R2 as it is particularly important.
Let’s denote by R a formal systems encoding real statements with their finitistic
proofs and by I the ideal system with its abstract reasoning.
Let A be a real statement ∀x(f (x) = g(x)).
Assume `I A. Then there is a derivation d of A in I. But, derivations are
concrete objects and, for some real formula P (x, y) encoding derivations in I,

`R P (d, pAq),

where pAq is some code for A.

Now, if A were false, one would have f (a) 6= g(a) for some a and hence

`R P (c, p¬Aq)

for some c. In fact, one would have a stronger assertion

`R (f (x) 6= g(x) ⇒ P (cx , p¬Aq)).

But, if R proves consistency of I, we have

`R ¬(P (d, pAq) ∩ P (c, p¬Aq)),

whence `R f (x) = g(x), with free variable x, i.e. `R ∀x(f (x) = g(x)).
To make the above argument rigorous, one has to define and explain the basics
of encoding, the assumptions on the formula P (x, y) and to deliver the whole
argument in a formal rigorous way, i.e. to develop rigorously the whole appa-
ratus developed originally by Gödel and needed for the proofs of his theorems.
We bring it here because it clearly invited Hilbert to establish his Consistency
Program. Since Consistency Program was as broad as the general Conservation
Program and, since it was more tractable, Hilbert fixed on it asserting:
”if the arbitrary given axioms do not contradict each other through their conse-
quences, then they are true, then the objects defined through the axioms exist.
That, for me, is the criterion of truth and existence”.

532
The Consistency Program had as its goal the proof, by finitistic means of the
consistence of strong systems. The solution would completely justify the use of
abstract concepts and would repudiate Brouwer and Weyl.
Gödel proved that it couldn’t work.

11.3.2 Gödel Incompleteness Theorems

In 1920, while in his twenties, Kurt Gödel announced that Hilbert’s Consistency
Program could not be carried out. He had proved two theorems which gave
a blow to the Hilbert’s Program but on the other hand changed the face of
mathematics establishing mathematical logic as strong and rapidly developing
discipline.
Loosely stated these theorems are:

Theorem 11.3 (First Incompleteness Theorem)

Let T be a formal theory containing arithmetic. Then there is a sentence A in
the language of T which asserts its own unprovability and is such that:
(i) If T is consistent, then 6`T A.
(ii) If T is ω- consistent, then 6`T ¬A.

Theorem 11.4 (Second Incompleteness Theorem)

Let T be a consistent formal theory containing arithmetic. Then

6`T ConT ,

where ConT is the sentence in the language of T asserting the consistency of T.

Observe that the Second Incompleteness Theorem destroys the Consistency Pro-
gram. It states that R can’t prove its own consistency, so obviously it can’t prove
consistency of I.
Smorynski’s argument that the First Incompleteness Theorem destroys the Con-
servation Program is as follows. The the sentence A is real and is easily seen
to be true. It asserts its own unprovability and is indeed unprovable. Thus the
Conservation Program cannot be carried out and, hence, the same must hold
for the Consistency Program.
M. Detlefsen in the Appendix of his book ”Hilbert Program: An Essay on Math-
ematical Instrumentalism”, Springer, 2013, argues that Smorynski’s argument
is ambiguous, as he doesn’t tell us whether it is unprovability in R or unprov-
ability in I. We recommend to the reader interested a philosophical discussion
of Hilbert Program to read this Appendix, if not the whole book.

533
We will now formulate the Incompleteness Theorems in a more precise formal
way and describe the main ideas behind their proofs.

Arithmetization and Encoding (11.33)

Observe that that in order to formalize the Incompleteness Theorems one has
first to ”translate” the sentences A and ConT into the language of T . For the
First Incompleteness Theorems 11.3 one needs to ” translate ” a self-referring
sentence ”I am not provable in a theory T”; for the Second Theorem 11.4 the
self-referring sentence is ”I am consistent”.
The assumption in both theorems is that T contains arithmetic means usually
it contains the Peano Arithmetic PA (11.11), or even its sub-theory RR (11.23),
called Robinson System. In this case the final product of such ”translation”
must be a sentence A or sentence ConT of the language LP A of PA, usually
written as
LP A = L({=}, {0 , +, ·}, {0}).

This ”translation” process into the language of some formal system containing
arithmetic is called arithmetization and encoding, or encoding for short. We
define a notion of arithmetization as follows.
An arithmetization of a theory T is a one-to-one function g from the set of
symbols of the language of T, expressions (formulas) of T, and finite sequences
of expressions of T (proofs) into the set of positive integers. The function g
must satisfy the following conditions.
(1) g is effectively computable;
(2) there is an effective procedure that determines whether any given positive
integer n is in the range of g and, if n is in the range of g, the procedure finds
the object x such that g(x) = m.
Arithmetization, i.e. a method of associating numbers with symbols, expres-
sions, and sequences of expressions was originally devised by Gödel in 1931 in
order to arithmetize Peano Arithmetic PA and encode the arithmetization pro-
cess PA in order to formulate and to prove his Incompleteness Theorems 11.3,
11.4.
Functions and relations whose arguments and values are natural numbers are
called the number-theoretic functions and relations.
In order to arithmetize and encode in a formal system we have to
1. associate numbers with symbols symbols of the language of the system, asso-
ciate numbers with expressions, and sequences of expressions of the language of
the system (arithmetization, encoding of basic syntax, and encoding of syntax)
2. replace assertions about the system by number-theoretic statements, and ex-
press these number-theoretic statements within the formal system itself ( arith-
metization,, encoding).

534
We want the number - theoretic function to be representable P A and the pred-
icates to be expressible in PA, i.e. their characteristic functions to be repre-
sentable in P A.
The study of representability of functions in P A leads to the class of number-
theoretic functions that turn out to be of great importance in mathematical
logic, namely the x primitive recursive and recursive functions. Their definition
and study in a form of a Recursion Theory is an important field of mathe-
matics and of computer science which developed out of the Gödel proof of the
Incompleteness Theorems.
We prove that the class of recursive functions is identical with the class of func-
tions representable in PA, i.e. we prove: every recursive function is representable
in PA and every function representable in PA is recursive.
The representability of primitive recursive and recursive functions in S in gen-
eral and in P A in particular plays crucial role in the encoding process and
consequently in the proof of Gödel Theorems.
The details of arithmetization and encoding are as complicated and tedious as
fascinating but are out of scope of our book. We recommend Mendelson’s book
”Introduction to Mathematical Logic”, 4th ed., Chapman & Hall (1997) as the
one with the most comprehensive and detailed presentation.

Theories T and S (11.34)

We assume at this moment that T is some fixed, but for a moment unspecified
consistent formal theory. We also assume that encoding is done in some fixed
theory S and that T contains S, i.e. the language of T is an extension of the
language of S and
S ⊆ T,
i.e. for any formula A,
if `S A, then `T A. (11.35)
We also assume that T and S contain as constants only numerals (definition
11.7)
0, 1, 2, 3, . . . ,
and T contains infinitely countably many functional and predicate symbols.
Usually S is taken to be a formal theory of arithmetic, but sometimes S can be
a weak set theory. But in any case S always contains numerals.
We also assume that theories T and S as defined by (11.34) are such that the
following Principles of Encoding (11.36) hold.

Principles of Encoding for T and S (11.36)

535
The mechanics, conditions and details of encoding for T and S for S being
Peano Arithmetic P A or its sub-theory Robinson Arithmetic RR (11.23) are
beautifully presented in the smallest detail in Mendelson.
The Smorynski’s approach we discuss here covers a larger class of formal theories
and uses a more general and modern approach. We can’t include all details but
we are convinced that at this stage the reader will be able to follow Smorynski’s
chapter in the Encyclopedia. The chapter is very well and clearly written and
is now classical. We wholeheartedly recommend it as a future reading.
We also follow Smorynski approach explaining what is to be encoded, where it
is to be encoded, and which are the most important encoding and provability
conditions needed for the proofs of the Incompleteness Theorems.
We first encode the syntax of T in S.
Since encoding takes place in S, it has a sufficient supply of constants (countably
infinite set of numerals
0, 1, 2, 3, . . . ,
and closed terms to be used as codes.
We assign to each formula A of the language of T a closed term,

pAq

called the code of A. If A(x) is a formula with a free variable x, then the code
pA(x)q is a closed term encoding the formula A(x), with x viewed as a syntactic
object and not as a parameter.
We do it recursively, first we assign codes (unique closed terms from S) to its
basic syntactic objects, i.e. elements of the alphabet of the language of T.
Terms and formulas are finite sequences of these symbols and derivations (formal
proofs) are also finite sequences of formulas. It means that S have to be able
to encode and manipulate finite sequences. We use for such encoding a class
primitive recursive functions and relations. We assume S admits a representation
of these functions and relations and finish encoding syntax.
S will also have to have certain function symbols and we have to be able to
encode them.
1. S must have we functional symbols, neg, impl, etc., corresponding to the
logical connectives and quantifiers, such that, such that, for all formulas A, B
of the language of T,

`S neg(pAq), `S impl(pA ⇒ Bq), etc.

An operation of substitution of a variable x in a formula A(x) by a term t is of

a special importance in logic, so it must be represented in S, i.e.
2. S must have in a functional symbol sub that represents the substitution
operator, such that for any formula A(x) and term t with codes pA(x)q, ptq,

536
respectively,
`S sub(pA(x)q, ptq) = pA(t)q. (11.37)
Iteratation of sub allows one to define sub3 , sub4 , sub5 , . . . , such that

`S subn (pA(x1 , . . . , xn )q, pt1 q, . . . , ptn q) = pA(t1 , . . . , tn )q.

Finally, we have to encode derivations in S , i.e.

3. S has to have in a binary relation P rovT (x, y), such that for closed terms
t1 , t2 ,
`S P rovT (t1 , t2 ) if and only if t1 is a code of a derivation in T of the formula
with a code t2 .
We read P rovT (x, y) as ”x proves y in T ” or ” x is a proof of y in T”.
It follows that for some closed term t,

`T A if and only if `S P rovT (t, pAq).

We define
P rT (y) ⇔ ∃xP rovT (x, y) (11.38)
and obtain a predicate asserting provability.
However, it is not always true

` T A if and only if `S P rT (pAq),

unless S is fairly sound (to be defined separately).

The encoding can be carried out, however, in such a way that the following
conditions essential to the proofs of the Incompleteness Theorems hold for any
sentence A of T.

Derivability Conditions (Hilbert-Bernays, 1939) (11.39)

D1 `T A implies `S P rT (pAq).

D2 `S ((P rT (pAq) ⇒ P rT (pP rT (pAq)q))).

D3 `S ((P rT (pAq) ∩ P rT (p(A ⇒ B)q)) ⇒ P rT (pBq)).

11.4 Proof of the Incompleteness Theorems

The following theorem 11.5 is essential to the proof of the Incompleteness The-
orems. It is called historically Diagonalization Lemma or Fixed Point Theorem
and both names are used interchangeably. The fist name as is historically older,

537
important for convenience of references and the second name is routinely used
in computer science community.
Mendelson (1977) believes that the central idea was first explicitly mentions by
Carnap who pointed out in 1934 that the result was implicit in the work of
Gödel (1931). Gödel was not aware of Carnap work until 1937.
The theorem 11.5 is called Diagonalization Lemma because the argument used
in its proof has some resemblance to the the diagonal arguments used by Cantor
in 1891. He first used it proving that there are infinite sets that can not be put
in one-to-one correspondence with the set on natural numbers. He then used
its generalization in the proof of his famous Cantor Theorem: for every set X,
its set of all subsets has a larger cardinality than X itself (see chapter 1).
In mathematics, a fixed-point theorem is a name of a theorem saying that a
function f under some conditions, will have a at least one fixed point, i.e. a
point x such that f (x) = x.
The theorem 11.5 says that for any formula A in the language of theory T with
one free variable there is a sentence B such that the formula (B ⇔ A(pBq)) is
provable in T .
Intuitively, B is a self-referential sentence saying that B has property A. The
sentence B can be viewed as a fixed point of the operation assigning to each
formula A the sentence A(pBq). Hence the name Fixed Point Theorem.
Theorem 11.5 proves the existence of self-referential sentences in certain formal
theories of natural numbers. These sentences then, in turn, are to be used to
prove Gödel’s Incompleteness Theorems. Here it is.

Theorem 11.5 (Diagonalization Lemma)

Let T, S be theories defined by (11.34).
Let A(x) be a formula in the language of T with x as the only free variable.
Then there is a sentence B such that

`S (B ⇔ A(pBq)).

NOTE: If A, B are not in the language of S, then by `S (B ⇔ A(pBq)) we

mean that the equivalence is proved in the theoryS 0 in the language of T whose
only non-logical axioms are those of S.
Proof
Given A(x), let (C(x) ⇔ A(sub(x, x)) be a diagonalization of A(x).
Let m = pC(x)q and B = C(m).
Then we claim
`S (B ⇔ A(pBq)).
For, in S, we see that

538
 B ⇔ C(m) ⇔ A(sub(m, m))

⇔ A(sub(pC(x)q, m) (since m = pC(x)q)

⇔ A(pC(m)q) ⇔ A(pBq) by (11.37)) and B = C(m).

This proves (we leave details to the reader as a homework exercise)

`S (B ⇔ A(pBq)).

Theorem 11.6 (First Incompleteness Theorem)

Let T, S be theories defined by (11.34).
Then there is a sentence G in the language of T such that:
(i) 6`T G.
(ii) under an additional assumption, 6`T ¬A.

Proof
Applying Diagonalization Lemma 11.5 for a formula A(x) being ¬P rT (x), where
P rT (x) is defined by (11.38) we get that there is a sentence G such that

`S (G ⇔ ¬P rT (pGq)).

By the assumed property (11.35) in the definition (11.34) of T, S we have that

also
`T (G ⇔ ¬P rT (pGq)). (11.40)

( i) We conduct the proof by contradiction. Assume `T G.

Observe that `T G implies `T P rT (pGq) by D1 and (11.35). This and the
above (11.40) contradicts the consistency of T .
(ii) The additional assumption is assuming that the converse implication to D1
holds, i.e that `T P rT (pGq) implies `T G.
We conduct the proof by contradiction. Assume `T ¬G.
Hence `T ¬¬P rT (pBq)) so we have that `T P rT (pBq)). By the additional
assumption it implies that `T G what contradicts contradicting the consistency
of T .
This ends the proof.
Observe that the sentence G is equivalent in T to an assertion that G is unprov-
able in T. In other words it says ” I am not provable in T” and hence theorem
11.6 is a strict mathematical formalization of the intuitively stated theorem
11.3. We call G the Gödel’s sentence.

Theorem 11.7 (Second Incompleteness Theorem)

539
Let T, S be theories defined by (11.34).
Let ConT be a sentence ¬P rT (pCq)), where is C is any contradictory statement.
Then
6`T ConT .

Proof
Let G the Gödel’s sentence of the First Incompleteness Theorem 11.6.
We prove that
`T (ConT ⇔ G) (11.41)
and use it to prove that 6`T ConT . We conduct the proof by contradiction.
Assume `T ConT . By (11.41) `T (ConT ⇔ G), so `T G what contradicts
the First Incompleteness Theorem 11.6.
To complete the proof we have to to prove now (11.41). We know by Logic 11.1
that

`T (ConT ⇔ G) if and only if `T (ConT ⇒ G) and `T (G ⇒ ConT ).

1. We prove the implication `T (G ⇒ ConT ).

By definition of ConT we have to prove now

`T (G ⇒ ¬P rT (pCq)). (11.42)

The formula C is a contradiction, so (C ⇒ G) is a predicate tautology. Hence

`T (C ⇒ G) and by D1

`S P rT (p(C ⇒ G)q).

We write D3 for A = P rT (pCq) and B = `S P rT (p(C ⇒ G)q) and obtain

that
`S ((P rT (pCq) ∩ P rT (p(C ⇒ G)q)) ⇒ P rT (pGq)). (11.43)
We have by Logic 11.2

`S (P rT (pCq) ⇒ (P rT (pCq) ∩ P rT (p(C ⇒ G)q))). (11.44)

We get from (11.44), (11.43), and Logic 11.3

`S (P rT (pCq) ⇒ P rT (pGq)). (11.45)

We apply Logic 11.4 (contraposition) to the above (11.45) and get

`S (¬P rT (pGq) ⇒ ¬P rT (pCq)). (11.46)

Observe that we by the property (11.40) in the proof of the First Incompleteness
Theorem 11.3 we have
`S (G ⇒ ¬P rT (pGq)). (11.47)

540
We put (11.46) and (11.47) together and get

`S (G ⇒ ¬P rT (pGq)) and `S (¬P rT (pGq) ⇒ ¬P rT (pCq)).

Applying Logic 11.4 to the above we get `S (G ⇒ ¬P rT (pCq)). But by C is

by definition ConT and hence we have proved the `S (G ⇒ ConT ) and hence
also
`T (G ⇒ ConT ).

2 . We prove now `T (ConT ⇒ G), i.e. the implication

`T (¬P rT (pCq) ⇒ G). (11.48)

Here is a concise proof. We leave it to the reader as an exercise to write a

detailed version.
By D2,
`S ((P rT (pGq) ⇒ P rT (pP rT (pGq)q))).
This implies
`S (P rT (pGq) ⇒ P rT (p¬Gq)),
by D1, D3, since `S (G ⇒ ¬P rT (pGq)).
This yields
`S ((P rT (pGq) ⇒ P rT (p(G ∩ ¬G)q)),
by D1, D3, and logic properties, which imples

`S ((P rT (pGq) ⇒ P rT (pCq)),

by D1, D3, and logic properties. By Logic 11.4 (contraposition)

`S (¬P rT (pGq) ⇒ ¬P rT (pCq)),

which is `S (ConT ⇒ G) and hence also

`T (ConT ⇒ G).

This ends the proof.

We prove now, as an exercise and reminder, the steps in the proof of part 1.
that follow the predicate logic properties, hence the name Logic. The discovery
of needed properties and their proofs for the part 2. is left as a homework
exercise.

Remark 11.4
By definition 11.3 the theories T, S are based on a complete proof system for
predicate logic and by the monotonicity of classical consequence everything prov-
able there is provable in T, S. In particular all predicate tautologies are provable
in T and in S.

541
Logic 11.1
Given a complete proof system H, for any formulas A, B of the language of H,

` (A ⇔ B) if and only if ` (A ⇒ B) and ` (B ⇒ A).

Proof
1. We prove implication if ` (A ⇔ B), then ` (A ⇒ B) and ` (B ⇒ A).
Directly from provability of a tautology ((A ⇔ B) ⇒ ((A ⇒ B) ∩ (B ⇒ A))),
assumption ` (A ⇔ B), and MP we get ` ((A ⇒ B) ∩ (B ⇒ A)). Consequently,
from ` ((A ⇒ B) ∩ (B ⇒ A)), provability of tautologies ((A ∩ B) ⇒ A), ((A ∩
B) ⇒ B) and MP applied twice we get ` (A ⇒ B), ` (B ⇒ A).
2. We prove implication if ` (A ⇒ B) and ` (B ⇒ A), then ` (A ⇔ B).
Directly from provability of tautology ((A ⇒ B) ⇒ ((B ⇒ A) ⇒ (A ⇔ B))),
assumption ` (A ⇒ B), ` (B ⇒ A), MP applied twice we get ` (A ⇔ B).

Logic 11.2 Given a complete proof system H, for any formulas A, B of the
language of H,

` (A ⇒ (A ∪ B)) and ` (A ⇒ (B ∪ A)).

Proof Directly from predicate tautologies (A ⇒ (A ∪ B)), (A ⇒ (B ∪ A)) and

completeness.

Logic 11.3
Given a complete proof system H, for any formulas A, B of the language of H,

if ` (A ⇒ B) and ` (B ⇒ C), then ` (A ⇒ C).

Proof From completeness and predicate respective tautology we get

` ((A ⇒ B) ⇒ ((B ⇒ C) ⇒ (A ⇒ C))). (11.49)

Assume (A ⇒ B). Applying MP to (11.49) twice we get the proof of (A ⇒ C).

Logic 11.4
Given a complete proof system H, for any formulas A, B of the language of H,

` (A ⇒ B) if and only if ` (¬B ⇒ ¬A).

Proof Directly from predicate tautology ((A ⇒ B) ⇔ (¬B ⇒ ¬A)), complete-

ness and MP.

542
Observation 11.2
We proved, a part of proof of the Second Incompleteness Theorem 11.7 the equiv-
alence (11.41) which says that the self-referential Gödel sentence G which asserts
its own unprovability is equivalent to the sentence asserting consistency. Hence,
the sentence G is unique up to provable equivalence (11.41) and we can say that
G is the sentence that asserts its own unprovability.

ω-consistency
We used, in the part (ii) of the First Incompleteness Theorem 11.6, an additional
assumption that `T P rT (pGq) implies `T G, instead of a habitual assumption
of ω-consistency.
The concept of ω-consistency was introduced by Gödel for purpose of stating
assumption needed for the proof of his First Incompleteness Theorem 11.3.
The modern researchers proved that the assuption of the ω-consistency can
be replaced, as we did, by other more general better suited for new proofs
conditions.
Informally, we say that T is ω- consistent if the following two conditions are
not satisfied for any formula A:
(i) `T ∃xA(x);
(ii) `T ¬A(n) for every natural number n.
Formally, ω-consistency can be represented (in varying degrees of generality) by
(modification of) the following formula

(P rT (p∃xA(x)q) ⇒ ∃x ¬P rT (p¬A(x)q)). (11.50)

11.4.1 The Formalized Completeness Theorem

Proving completeness of a proof system with respect to a given semantics is the

first and most important goal while developing a logic and was the central focus
of our study. So we now conclude our book with presentation the formalized
completeness theorem 11.8. We discuss its proof and show how to use it to
give new type of proofs, called model-theoretic proofs, of the incompleteness
theorems for Peano Arithmetic PA, i.e. for the case when S = P A.
Formalizing the proof of completeness theorem for classical predicate logic from
chapter 9 within PA we get the following.

Theorem 11.8 (Hilbert-Bernays Completeness Theorem)

Let U be a theory with a primitive recursive set of axioms.
There is a set T rM of formulas such that in P A + ConU one can prove that this

543
set T rM defines a model M of U :

`P A+ConU ∀x(P rU (x) ⇒ T rM (x)) (11.51)

Moreover the set T rM is of type ∆2 .

The Hilbert-Bernays Completeness Theorem 11.8 asserts that modulo ConU ,

one can prove in PA the existence of a model of U whose truth definition is of
type ∆2 . Its proof is just an arithmetization of the Henkin proof presented in
chapter 9. Following the Henkin proof one adds to the language of U an infinite
primitive recursive set of new constants

c0 , c1 , c2 . . . ,

and adds the axiom (Henkin Axiom)

(∃xA(x) ⇒ A(cA[x] )) (11.52)

for each formula A(x). One then enumerates sentences

A0 , A1 , A2 , . . . (11.53)

in this augmented language and defines a complete theory by staring with U and
adding at each step n a sentence An , or ¬An according to whether An is con-
sistent with what has been chosen before or not.
The construction is then described within PA. Assuming ConU one can also
prove that the construction never terminates. The resulting set of sentences
forms a complete theory which by axioms (11.52) forms a model of U. Inspection
shows that the truth definition T rM of type ∆2 .

The Hilbert-Bernays Completeness Theorem 11.8 makes possible to conduct

new type of proofs of the incompleteness theorems, model- theoretic proofs.
Gödel chose as the self-referring sentence a syntactic statement ” I do not
have a proof”. He did not want (and saw difficulties with) to use the sentence
involving the notion of truth, i.e. the sentence ”I am not true”. The new proofs
use exactly this and this is why they are called model-theoretic proofs.
Dana Scott was the first to observe that one can give a model- theoretic
proof of the First Incompleteness Theorem. Here is the theorem and his short
proof.

Theorem 11.9 (First Incompleteness Theorem)

Let PA be a Peano Arithmetic.
There is a sentence G of PA, such that
(i) 6`P A G;
(ii) 6`P A ¬G.

544
Proof
Assume PA is complete. Then, since PA is true, `P A ConP A and we can apply
the completeness theorem 11.8 to obtain a formula T rM which gives a truth
definition for the model of PA. Observe that once P A is complete we have that
P rP A is T rM . We choose G by

`P A (G ⇔ ¬T rM (pGq)). (11.54)

We claim 6`P A G, 6`P A ¬G. For if `P A G, then `P A T rM (pGq)). By (11.54)

`P A ¬G. Contradiction. Similarly, `P A ¬G implies `P A G.
Observe that the sentence G as defined by (11.54) asserts ”I am not true”.
Scott ’s proof differs from the proof of the First Incompleteness Theorem 11.6
not only by the choice of the model- theoretic method, but also by be a choice
of the model- theoretic sentence G.
Let’s compare these two independent sentences G:
the classic syntactic one of theorem 11.6 representing statement ” I do not have
a proof” and
the model- theoretic one of theorem 11.9 representing statement ”I am not
true”.

Property 11.1
The sentence GS of the First Incompleteness Theorem 11.6 asserting its own
provability is
(i) unique up to provable equivalence (Observation 11.2);
(ii) the sentence is Π1 and hence true.
The sentence G of the First Incompleteness Theorem 11.6 asserting its own
falsity in the model constructed is
(iii) not unique - for the following implication holds

if (G ⇔ ¬T rM (pGq)), then (¬G ⇔ ¬T rM (p¬Gq)).

(iv) the sentence is ∆2 (theorem 11.8, and, by (iii) there is no obvious way od
deciding its truth or falsity.

Georg Kreisler was the first to present a model- theoretic proof of the
following.

Theorem 11.10 (Second Incompleteness Theorem)

Let PA be a Peano Arithmetic. 6`P A ConP A .

The proof is uses, as did the proof of Hilbert-Bernays Completeness Theorem

11.8 the arithmetization of Henkin proof of completeness theorem presented in
chapter 9. The proof is carried by contradiction. We assume `P A ConP A .

545
Then we show, for any presentation of the Henkin proof construction (as given
by encoding, the enumeration of sentences (11.53) . . . etc.) there is a number m
such that, for any model N of P A, the sequence of models determined by the
given presentations must stop after fewer then m steps with a model in which
ConP A is false.

11.5 Homework Problems

1. Follow the proof of Fact 11.2 for the case of axioms P1 and C1 to prove
the case of axioms P2 and C2.

2. Prove the case of axioms P2, C2 and axioms P23, C3 of the Fact 11.2.

3. Prove Fact 11.2 in case of axioms P5, C5 and axioms P8, C8 of the Fact
11.2.

4. Complete the proof of Fact 11.2 or all cases.

5. We proved that the property P 10 of Fact 11.3 is a generalization of axiom

P 1 of PA (11.11, i.e. it is provable in PA.
(i) Write detailed proofs of properties P 20 − P 50 in P A.
(i) Write detailed proofs of properties P 60 − P 80 in P A.

6. Follow the definition 11.8 and prove the following formulas pre provable
in PA for ant terms t, r, s.
(i) t 6< t.
(ii) (t < s ⇒ (s < r ⇒ t < r)).
(iii) (0 < 1), (1 < 2), (2 < 3), (3 < 4), . . . .
(iv) 0 ≤ t.
(v) t ≤ t.
(vi) (t ≤ r ∪ r ≤ t).
(vii) (t ≤ r ⇒ (r ≤ t ⇒ t = r)).

7. Follow the definition 11.8 and prove the following formulas pre provable
in PA for ant terms t, r, s.
(i) (t ≤ s ⇒ (s ≤ r ⇒ t ≤ r)),
(ii) (t ≤ s ⇒ (t + r ≤ s + r)),
(ii) (r > 0 ⇒ (t > 0 ⇒ r · t > 0)).

8. Let RR be the Robinson System (11.23). Show that RR is a proper sub-

theory of P A by finding a model of RR that is not a model for P A.

546
 9. Let RR be the Robinson System (11.23). Let n, m be any natural numbers.
Prove the following holds in RR.
(i) If m 6= n, then m 6= n.
(ii) m + n = m + n and m · mn = m · n are provable in RR .
(ii) Any model for RR is infinite.

10. Here us the reasoning we used explaining Hilbert Consistency Program.

” Let A be a real statement ∀x(f (x) = g(x)). Assume `I A. Then there
is a derivation d of A in I. But, derivations are concrete objects and, for
some real formula P (x, y) encoding derivations in I, `R P (d, pAq), where
pAq is some code for A.
Now, if A were false, one would have f (a) 6= g(a) for some a and hence
`R P (c, p¬Aq) for some c. In fact, one would have a stronger assertion
`R (f (x) 6= g(x) ⇒ P (cx , p¬Aq)). But, if R proves consistency of I, we
have
`R ¬(P (d, pAq) ∩ P (c, p¬Aq)),
whence `R f (x) = g(x), with free variable x, i.e. `R ∀x(f (x) = g(x)).”

(i) Write down a detailed proof of correctness of the last part of reasoning:
”But, if R proves consistency of I, we have

`R ¬(P (d, pAq) ∩ P (c, p¬Aq)),

whence `R f (x) = g(x), with free variable x, i.e. `R ∀x(f (x) = g(x)).”

(ii) List, prove and use proper Logic Properties similar to properties Logic
11.1 - Logic 11.4 in the proof of Theorem 11.7.

547