City University of New York (CUNY)

CUNY Academic Works

Dissertations, Theses, and Capstone Projects CUNY Graduate Center

5-2015

Force to change large cardinal strength

Erin Kathryn Carmody
Graduate Center, City University of New York

How does access to this work benefit you? Let us know!

More information about this work at: https://academicworks.cuny.edu/gc_etds/879
Discover additional works at: https://academicworks.cuny.edu

This work is made publicly available by the City University of New York (CUNY).
Contact: [email protected]
 Force to change large cardinal strength
by

Erin Carmody

A dissertation submitted to the Graduate Faculty in Mathematics in partial

fulfillment of the requirements for the degree of Doctor of Philosophy, The
City University of New York.

2015

i
ii

©2015

Erin Carmody

All Rights Reserved

 iii

This manuscript has been read and accepted for the

Graduate Faculty in Mathematics in satisfaction of the
dissertation requirements for the degree of Doctor of
Philosophy.

Joel David Hamkins

Date Chair of the Examining Committee

Linda Keen

Date Executive Officer

Joel David Hamkins

Arthur W. Apter

Gunter Fuchs

Philipp Rothmaler

Roman Kossak
Supervisory Committee

THE CITY UNIVERSITY OF NEW YORK

iv

Abstract

Force to change large cardinal strength

by

Erin Carmody

Advisor: Joel David Hamkins

This dissertation includes many theorems which show how to change large cardinal prop-

erties with forcing. I consider in detail the degrees of inaccessible cardinals (an analogue of

the classical degrees of Mahlo cardinals) and provide new large cardinal definitions for de-

grees of inaccessible cardinals extending the hyper-inaccessible hierarchy. I showed that for

every cardinal , and ordinal ↵, there is a notion of forcing P such that  is still -inaccessible

in the extension, for every < ↵, but not ↵-inaccessible. I also consider Mahlo cardinals

and degrees of Mahlo cardinals. I showed that for every cardinal , and ordinal ↵, there is

a notion of forcing P such that for every < ↵, the cardinal  is still -Mahlo in the exten-

sion, but not ↵-Mahlo. I also show that a cardinal  which is Mahlo in the ground model

can have every possible inaccessible degree in the forcing extension, but no longer be Mahlo

there. The thesis includes a collection of results which give forcing notions which change

large cardinal strength from weakly compact to weakly measurable, including some earlier

work by others that fit this theme. I consider in detail measurable cardinals and Mitchell
 v

rank. I show how to change a class of measurable cardinals by forcing to an extension where

all measurable cardinals above some fixed ordinal ↵ have Mitchell rank below ↵. Finally,

I consider supercompact cardinals, and a few theorems about strongly compact cardinals.

Here, I show how to change the Mitchell rank for supercompactness for a class of cardinals.
vi

Aknowledgements

First of all, I would like to thank Arthur Apter for helping me see the future. His en-

couragement has inspired me to see how bright our future is together. I have total faith and

trust in Arthur. We work well together. It makes me feel good when we can talk about

interesting problems together, and I look forward to all of our theorems of the future!

From the beginning until the end of this process, Philipp Rothmaler has been a great

mentor to me. He encouraged me to present at seminar, and helped me fall in love with

logic. Thank you Philipp for letting me tell you my ideas and for always being nearby to see

me through until the end. I was your student, and I look forward to our work together in

the future.

Gunter, thank you for being my friend and my mentor through this process. You have

always been there for me, and showed your appreciation of my interests. Your mind is amaz-

ing, and I always learn so much from your talks. I would like to know more, and create

notions of forcing together.

Victoria, dear Victoria! You inspire me and you helped me so much. You took the time

to look at every word of my thesis and applications. You became my sister over time from
 vii

the oral exam until now where we have our work together. I pray that I can find your kind

of focus. Thank you, Victoria.

Dear Roman, thank you for encouraging me in every way possible. Thank you for being

a reliable source of support, and for showing your appreciation. Thank you for everything,

Roman.

Joel is never-waivering and ever-patient. To see him teach is to watch lightning in a bottle,

like an angel defending the line between heaven and hell. He is enthusiastic and gives 100%

all the time. His brain moves fast like a computer and his mind is as gentle as a breeze. He

is kind and aware. He helped me learn how to be a person through his actions and words.

He is so thoughtful and so fun! He is truly-deeply-eternally-vastly, . . . and so on.

I would also like to thank my pre-committee: Kaethe Minden, Miha Habic, and Kameryn

Williams. Thank you for forcing me to come to your wonderful seminar to get ready for the

final examination. I could not have done it without your thoughtful comments and friendship

all through my time as a student. You all are brilliant and I know that you are going to

create beautiful theorems and change the future of set theory and mathematics.
viii

Contents
 1

1. Introduction

Large cardinals are infinite numbers, so big that we cannot prove their existence in ZFC.

If  is a large cardinal, then V |=ZFC, so proving the existence of a large cardinal would

violate Godel’s incompleteness theorem. The theorems in the following chapters assume the

consistency of ZFC with large cardinals. Forcing was developed by Cohen in 1963, and has

been used as a way to prove independence results. Forcing in the following chapters is used

to create an extension where one or many large cardinals has some greatest desired degree

of some large cardinal property. Suppose  2 V is a cardinal with large cardinal property A.

The purpose of this thesis is to find a notion of forcing P such that if G ✓ P is V -generic, the

cardinal  no longer has property A in V [G], but has as many large cardinal properties below

A as possible. The main theorems of this dissertation are positive results like this where

the large cardinal properties of a cardinal change between the ground model and the forcing

extension. A notion of forcing P which can change a large cardinal, like the ones in the main

theorems, does so carefully. That is, P does not just destroy large cardinal properties, it also

preserves large cardinal properties. The project is to pick a desired large cardinal degree

and design a notion of forcing which forces a cardinal to lose all large cardinal properties

above this degree and keep all large cardinal properties below. In other words, we are killing

large cardinals as softly as possible. In particular we are killing large cardinals by just one

degree from the ground model to the forcing extension. Thus, a part of this thesis is also
2

devoted to defining degrees of various large cardinal properties such as inaccessible, Mahlo,

measurable, and supercompact.

The first chapter is about inaccessible cardinals. An inaccessible cardinal  is an uncount-

able cardinal having two properties exhibited by !, the first infinite cardinal. Namely, an

inaccessible cardinal  is regular: cof() = , just like ! since no sequence of finite length

can be cofinal in !, and an inaccessible cardinal  is also a strong limit: 8 < , 2 < 

just like ! since 8n < !, 2n < !. But inaccessible cardinals are uncountable, unlike !. An

inaccessible cardinal might also be a limit of inaccessible cardinals, this is a 1-inaccessible

cardinal. A cardinal  is ↵-inaccessible if and only if  is inaccessible and for every < ↵,

 is a limit of -inaccessible cardinals. The main theorem of the first chapter is that if

 is ↵-inaccessible, there is a forcing extension where  is ↵-inaccessible but not (↵ + 1)-

inaccessible. If a cardinal  is -inaccessible it is called hyper-inaccessible. In the first

chapter, I define ↵-hyper -inaccessible cardinals and beyond this to the richly-inaccessible

cardinals, utterly-inaccessible cardinals, deeply-inaccessible cardinals, truly-inaccessible car-

dinals, eternally-inaccessible cardinals, and vastly-inaccessible cardinals. And, I develop a

notation system for a hierarchy of inaccessibility reaching beyond this.

The second chapter is about forcing to change degrees of Mahlo cardinals. A Mahlo

cardinal  has many inaccessible cardinals below, every closed unbounded (club) subset of 

contains an inaccessible cardinal. Let I(X) be an operator which gives the set of inaccessible

limit points of a set X. A cardinal  is called greatly inaccessible if and only if there is a
 3

uniform, normal filter on , closed under I(X). The first theorem of the second chapter

shows that a greatly inaccessible cardinal is equivalent to a Mahlo cardinal. Since greatly

inaccessible cardinals are every possible inaccessible degree, as defined in chapter 1, Mahlo

cardinals are every possible inaccessible degree defined. One of the main theorems of the

second chapter is to force a Mahlo cardinal  to lose its Mahlo property in the extension

while preserving that  is every inaccessible degree defined. There is also a classical hierarchy

of Mahlo cardinal degrees, like the hierarchy of inaccessible cardinal degrees defined in the

first chapter. A cardinal  is 1-Mahlo if and only if the set of Mahlo cardinals below  is

stationary in  (every club in  contains a Mahlo cardinal). A cardinal  is ↵-Mahlo if and

only if  is Mahlo and for all < ↵ the set of -Mahlo cardinals is stationary in . The

main theorem of the second chapter is that for any  and ↵ where  is ↵-Mahlo in V there

is a forcing extension where  is ↵-Mahlo but not (↵ + 1)-Mahlo.

The third chapter involves many large cardinal notions. A few of the theorems in chapter

three are about weakly compact cardinals. A cardinal  is weakly compact if and only if 

is inaccessible and for every tree of height  whose levels have cardinality less than  has a

branch of length . One theorem shows that weakly compact cardinals are every classical

Mahlo degree. Another shows that one can force a weakly compact cardinal  to lose its weak

compactness in the extension while keeping all of its Mahlo degrees. This chapter includes

theorems by others, especially Victoria Gitman about weakly compact cardinals and the

following large cardinal properties which are defined in chapter three: weakly measurable

cardinals, strongly Ramsey cardinals, and ine↵able cardinals.

4

The fourth chapter is about measurable cardinals, which have all of the large cardinal

properties described so far. A cardinal  is measurable if and only if it is the critical point

of an elementary embedding j : V ! M . The classic hierarchy of degrees for measurable

cardinals is the Mitchell rank, denoted o() for a cardinal , is defined in chapter four. In

this chapter we are able to force to change the Mitchell rank for a class of cardinals in the

ground model. The main theorem is that for any V |= ZFC, any ↵ 2 Ord, and any  > ↵

in V , there is a forcing extension V [G] where o() = min{↵, o()V }. In other words, in the

forcing extension every large cardinal above ↵ has Mitchell rank at most ↵.

The fifth and final chapter of this dissertation is about supercompact and strongly compact

cardinals. A cardinal  is ✓-supercompact if and only if there is an elementary embedding

j : V ! M with critical point  and M ✓ ✓ M . That is, the target of the elementary

embedding is closed under ✓-sequences. Equivalently,  is ✓-supercompact if and only if

there is a normal fine measure on P ✓. A cardinal  is ✓-strongly compact if and only if

there is a -complete fine measure on P ✓. The first main results of the fifth chapter are

due to a key result by Hamkins and Shelah and show how to force a supercompact (or

strongly compact) cardinal to be at most ✓-supercompact (or ✓-strongly compact) for a fixed

✓. Also, Magidor’s result that one can force to separate strongly compact and supercompact

cardinals is in this chapter. In chapter five, Mitchell rank for ✓-supercompact cardinals,

denoted o✓-sc (), for a cardinal  is discussed in detail. The main result shows how to find a

forcing extension where for many , ↵, ✓ in V the Mitchell rank for ✓-supercompactness for

 is at most ↵ in the extension: o✓-sc ()V [G] = min{↵, o✓-sc ()V }.

 5

2. Degrees of Inaccessible Cardinals

In this section, I have found forcing extensions where the degree of an inaccessible cardinal

is reduced to an exact specified amount from its degree in the ground model. In general

terms, for a given cardinal , we find a notion of forcing which simultaneously preserves

large cardinal properties of  up to some level while destroying its large cardinal properties

above that level. Suppose  is a cardinal with large cardinal property A. Thus,  also has

property B for any lower property B which follows from A. The objective is to find a forcing

notion P, with generic object G ✓ P, such that in the forcing extension, V [G], the cardinal

 no longer has property A, but still has property B. The main theorem below shows how

to do this when the di↵erence between A and B is one degree of inaccessibility. An infinite

cardinal  is inaccessible if and only if  is an uncountable, regular, strong limit cardinal. A

cardinal  is 1-inaccessible if and only if  is inaccessible and a limit of inaccessible cardinals.

A cardinal  is ↵-inaccessible if and only if  is inaccessible, and for every < ↵, the cardinal

, is a limit of -inaccessible cardinals.

Theorem 1. If  is ↵-inaccessible, then there is a forcing extension where  is still ↵-

inaccessible, but not (↵ + 1)-inaccessible.

The following Lemma establishes some basic facts about degrees of inaccessible cardinals.
6

Lemma 2. 1. A cardinal  is 0-inaccessible if and only if  is inaccessible.

2. If  is ↵-inaccessible, and < ↵, then  is also -inaccessible.

3. A cardinal  cannot be ⌘-inaccessible, for any ⌘ > .

Proof. 1. If  is 0-inaccessible, then by definition  is inaccessible and a limit of -inaccessible

cardinals for every < 0, hence  is inaccessible. If  is inaccessible then it is vacuously

true that for every < 0, the cardinal  is a limit of -inaccessible cardinals.

2. Suppose  is ↵-inaccessible, and < ↵. Since  is ↵-inaccessible, for every ⌘ < ↵,

the cardinal  is a limit of ⌘-inaccessible cardinals. Since every less than is also less

than ↵, the cardinal  is a limit of -inaccessible cardinals, for every < . Thus,  is also

-inaccessible.

3. By way of contradiction, suppose  is the least cardinal with the property that 

is ( + 1)-inaccessible. It follows from the definition that  is a limit of -inaccessible

cardinals. Thus, there is a cardinal <  for which is -inaccessible. Since  is a limit

ordinal, + 1 < , and so by statement (2), is also ( + 1)-inaccessible. This contradicts

that  is the least cardinal with this property. It follows from the lemma that  cannot

be ⌘-inaccessible for any ⌘  + 1, because being ⌘-inaccessible would imply that  is

( + 1)-inaccessible. ⇤

And now the proof of Theorem ??:

 7

Proof. Let  be ↵-inaccessible. If  is not (↵ + 1)-inaccessible, then trivial forcing will give

the forcing extension where  is ↵-inaccessible, but not (↵ + 1)-inaccessible. Thus, assume

that  is (↵ + 1)-inaccessible, and we will find a forcing extension where it is no longer

(↵ + 1)-inaccessible but still ↵-inaccessible. The idea of the proof is to add a club, C, to 

which contains no ↵-inaccessible cardinals, and then force to change the continuum function

to kill strong limits which are not limit points of C. To change the continuum function, we

will perform Easton forcing.

Easton forcing to change the continuum function works when the GCH holds in the ground

model. For our purposes, we only need the GCH pattern to hold up through , by forcing

with P, which is a -length iteration of Add( , 1) for regular 2 V P . The forcing P, neither

destroys, nor creates, inaccessible cardinals below , and thus is a mild preparatory forcing,

so that we are in a place from which we will eventually perform Easton forcing.

First, see that P does not destroy inaccessible cardinals, by investigating its factors. Let

⌘ <  be inaccessible, and let p 2 P. Let p be the condition, p, restricted to domain [0, ].

And, let p> , be the condition, p, with domain restricted to ( , ]. Let P = {p : p 2 P}

and P> = {p< : p 2 P}, so that P factors as P ⇤ P> . For < ⌘, the first factor,

P , is small relative to ⌘, and thus cannot force 2 ⌘. And, the second factor, P> , is

 -closed, so adds no new subsets to . Thus ⌘ is still a strong limit after forcing with P.

Similar factoring arguments show that ⌘ remains a regular limit after forcing with P [Jech,

232+]. Since P preserves inaccessible cardinals, it follows by induction, that it preserves any
8

degree of inaccessibility defined so far. Further, since inaccessibility is downward absolute,

it follows by induction, that P cannot create -inaccessible cardinals for any < . Let

G ✓ P be V -generic, and force to V [G], where V [G] |= GCH, and where V and V [G] have

the same -inaccessible cardinals for any < .

Next, force with C, which will add a club subset to , which contains no ↵-inaccessible

cardinals in the forcing extension. Conditions c 2 C are closed, bounded, subsets of ,

consisting of infinite cardinals and containing no ↵-inaccessible cardinals. The forcing C

is ordered by end-extension: d  c if and only if c = d \ (sup(c) + 1). Let H ✓ C be

V [G]-generic, and let C = [H. Then, it is claimed that C is club in , and contains no

↵-inaccessible cardinals in V [G][H]. First, let us see that C is unbounded. Let 2 , and

let D = {d 2 C : max(d) > }. The set D is dense in C, since, if c 2 P, the set

d = c [ { 0 }, where 0
> and 0
is not ↵-inaccessible, is a condition in D and d is stronger

than c. Hence, there is c 2 D \ G, so that there is a an element of C above . Thus, in

V [G][H], the set C is unbounded in . Suppose \ C is unbounded in < . Since C is

0 0 0
unbounded, there is 2 C, where > , and a condition c 2 C which contains . Since

the conditions are ordered by end-extension, and since there is a sequence (possibly of length

1) of conditions which witness that \ C is unbounded in , which all must be contained in

the condition c (which contains an element above ), it follows that \ c is unbounded in

. Since c is closed, 2 c . Therefore, 2 C, which shows that C is also closed.

 9

The forcing, C, adds a new club, C, to . It remains to show that C contains no ↵-

inaccessible cardinals, C contains unboundedly many -inaccessible cardinals, and that C

preserves cardinals, cofinalities and strong limits. The new club, C, does not contain any

ground model ↵-inaccessible cardinals, since if it did, there would be a condition in G which

contains an ↵-inaccessible, contradicting that no condition in G contains an ↵-inaccessible.

To see that the new club, C, contains unboundedly many ground model -inaccessible car-

dinals, for every < ↵, fix < ↵ and ⌘ < . Let D⌘ be the set of conditions in C which

contain a -inaccessible above ⌘, and which contain a sequence of inaccessible cardinals,

unbounded in ⌘, witnessing that is -inaccessible. Let us see that D⌘ is dense in C. Let

c 2 C. Let be the next -inaccessible above both ⌘ and and the maximal element of c.

Since is the next -inaccessible past ⌘ and < ↵, there are no ↵-inaccessible cardinals

in (⌘, ]. Also, this block, (⌘, ], contains the tails of all sequences of inaccessible cardinals,

which witnesses that is -inaccessible. Let d = c [ {(⌘, ]}\CARD. Then, d extends c,

contains a -inaccessible above ⌘, and a sequence of inaccessible cardinals which witness that

is -inaccessible. Thus, d 2 D⌘ , which shows that D⌘ is dense in C, and thus shows that

C contains unboundedly many -inaccessible cardinals. From this fact also follows, that,

in the final extension,  is still ↵-inaccessible, as we shall soon see. Finally, the forcing C

preserves cardinals and cofinalities greater than or equal to  + 1, forcing over V [G] |= GCH,

since |C| = < = .

For < , the forcing, C, is not  -closed, since if is ↵-inaccessible, there is a -

sequence of conditions unbounded in , but no condition could close the sequence since it
10

would have to include . However, for every < , the set D = {d 2 C : max(d) } is

dense in C and is  -closed. This is true, since for any -sequence of conditions in D , one

can close the sequence by taking unions at limits and adding the top point, which cannot

be inaccessible, because it is not regular, since this top point is above , but has cofinality

. Thus, for every < , the forcing, C, is forcing equivalent to D , which is  -closed.

Thus, C preserves all cardinals, cofinalities, and strong limits. Thus. V [G][H] |= GCH and

we have forced to add C ✓ , club, which contains no ↵-inaccessible cardinals.

The last step of the proof is to force over V [G][H], with E, Easton’s forcing to change the

+
continuum function. Specifically, let E force 2 = +
, where 2 C, for infinite, and

+
is the next element of C past . This forcing preserves all cardinals and cofinalities [Jech,

232+], and also preserves that  is inaccessible since factoring at any <  reveals that the

first factor is too small to force 2 up to , and the second factor adds no new subsets to ,

thus preserving that  is a strong limit cardinal, in addition to preserving that  is a regular

limit cardinal. However, E does not preserve all inaccessible cardinals below . In fact, E

destroys all strong limits which are not limit points of C. Let ⌘ <  be a strong limit which

is not in C 0 , the set of limit points of C. Since C is closed, there is a greatest element of

C below ⌘, call it . Then, the next element of C, call it , is greater than or equal to ⌘.

+
So, when E forces 2 = +
, it destroys that ⌘ is a strong limit since < ⌘ implies +
<⌘

(since ⌘ was a strong limit in V [G][H]). Let K ✓ E be V [G][H]-generic. Since C 0 contains

no ground model ↵-inaccessible cardinals, there are no ↵-inaccessible cardinals below  in

V [G][H][K]. Thus,  is not (↵ + 1)-inaccessible in the final forcing extension.

 11

Finally, see that since C 0 contains unboundedly many ground model -inaccessible cardi-

nals (which is the same as the set of -inaccessible cardinals in the intermediate extensions

since P and C preserve all inaccessible cardinals) for every < ↵. The cardinal  is still

0
-inaccessible in V [G][H][K]. Let < , where 2 C 0 and is inaccessible. Since is a

0
limit point of C, the next element of C, call it , which is above , is also below , and

+ 0 0+ +
< since is a strong limit. Thus, 2  2 = < . So, is still a strong limit. Since

E preserves cardinals and cofinalities, this shows that is still inaccessible in the final ex-

tension. Thus, all inaccessible limit points of C are preserved. The earlier density argument

also shows that for any < ↵ it is dense that C contains an interval containing all cardinals

and also contains a -inaccessible cardinal. Since all limit points of such an interval are limit

points of C, it follows that all inaccessible cardinals in such an interval are preserved. Thus,

all degrees of inaccessible cardinals are preserved. If not, then suppose the least degree of

inaccessibility which is destroyed an ⌘-inaccessible cardinal for some ⌘ < ↵. Then for some

< ⌘, there is no sequence of -inaccessible cardinals witnessing this ⌘-inaccessible cardinal,

which implies that -inaccessible cardinals are destroyed, which contradicts that ⌘ was the

least degree not preserved. Thus, in V [G][H][K], for every < ↵, there are unboundedly

many -inaccessible cardinals below . Thus,  is still ↵-inaccessible in the final extension.

But,  is not (↵ +1)-inaccessible in V [G][H][K] since C 0 contains no ↵-inaccessible cardinals.

⇤
12

Theorem ?? shows how to change the degree of an inaccessible cardinal, in a forcing

extension, to be exactly ↵ when  has degree at least ↵ in the ground model. The following

theorem shows how to force to a universe where there are no inaccessible cardinals, but

where every every ground model weakly inaccessible cardinal is still weakly inaccessible.

Theorem 3. For any V |= ZF C there exists V [G] with no inaccessible cardinals, but where

every ground model weakly inaccessible cardinal is still weakly inaccessible.

Proof. The proof is very similar to the proof of Theorem ??. It is a class forcing, where

the first step is to force that Ord is not Mahlo. We add a club C ✓ Ord such that C

contains no inaccessible cardinals. The second step is to perform Easton forcing to make

+ +
2 = whenever 2 C and where is the next element of C above . The idea of

the proof appears as a footnote in Hamkins[5]. The proof of Theorem ?? shows that the

combined forcing preserves all strong limits of C which are internally strong limits of C, and

destroys all strong limits of C which are not internally strong limits of C. Since C contains

no inaccessible cardinals, there are no inaccessible cardinals in the final extension. However,

as one can see from the proof of Theorem ?? or from Lemma ?? that the forcings preserve all

cardinals and cofinalities, so that all regular limit cardinals are preserved. Hence all weakly

compact inaccessible cardinals are preserved. ⇤

If a cardinal  is -inaccessible, then it is defined to be hyper-inaccessible. Therefore, if

 is hyper-inaccessible, Theorem ?? shows how to force to make  have inaccessible degree

 13

↵ for some ↵ < . One can force to change a hyper-inaccessible cardinal to have maximal

degree by forcing to add a club which avoids all degrees above the desired degree, and then

force with Easton forcing to destroy strong limits which are not limit points of the new club,

as in the proof of Theorem ??. I’ll state it as a corollary below.

Corollary 4. If  is hyper-inaccessible, then for any ↵ < , there exists a forcing extension

where  is ↵-inaccessible, but not hyper-inaccessible.

Lemma ?? shows that the greatest degree of ↵-inaccessibility that  can be is -inaccessible.

Mahlo began the investigation of degrees of inaccessible cardinals [Kan], and fully defined

the analgous notions for Mahlo cardinals, and I shall continue his work by formalizing the

degrees of inaccessible cardinals in the remainder of this section. So, how do we proceed

beyond hyper-inaccessible cardinals in defining degrees of inaccessibility? By repeating the

process: a cardinal  is 1-hyper-inaccessible if and only if  is hyper-inaccessible, and a limit

of hyper-inaccessible cardinals (0-hyper-inaccessible is hyper-inaccessible). That is,  is 1-

hyper-inaccessible if and only if the set { <  : is -inaccessible} is unbounded in . A

cardinal  is ↵-hyper-inaccessible if and only if  is hyper-inaccessible, and for every < ↵,

the cardinal  is a limit of -hyper-inaccessible cardinals. A cardinal  is hyper-hyper-

inaccessible, denoted hyper2 -inaccessible, if and only if  is -hyper-inaccessible (hyper0 -

inaccessible denotes inaccessible). The following theorem shows we have again reached an

apparent roadblock in the hierarchy of inaccessible cardinals.

Theorem 5. If  is ↵-hyper-inaccessible, then ↵  .

14

Proof. Suppose  is the least cardinal with the property that  is ( + 1)-hyper-inaccessible.

Then  is a limit of -hyper-inaccessible cardinals. Let <  be some -hyper-inaccessible

cardinal. Then, for every < , the cardinal is a limit of -hyper inaccessible cardinals.

This implies that for every < + 1 < , the cardinal is a limit of -hyper-inaccessible

cardinals. Thus, by definition, the cardinal is ( + 1)-hyper-inaccessible, which contradicts

that  is the least cardinal with this property.

In order to reach higher degrees of inaccessibility past this limit, repeat the process:

a cardinal  is 1-hyper2 -inaccessible if and only if  inaccessible and a limit of hyper2 -

inaccessible cardinals. A cardinal  is ↵-hyper2 -inaccessible if and only if, for every < ↵,

the cardinal  is inaccessible and a limit of -hyper2 -inaccessible cardinals. By the same

argument from the previous proof, ↵ must be less than of equal to  in this definition. And,

similarly, it is defined that  is hyper3 -inaccessible if and only if  is -hyper2 -inaccessible.

Thus, we arrive at the beginning of the formulation of the continuation of Mahlo’s work with

a general definition of degrees of hyper-inaccessibility:

Definition 6. A cardinal  is ↵-hyper -inaccessible if and only if

1) the cardinal  is inaccessible, and

2) for all ⌘ < , the cardinal  is -hyper⌘ -inaccessible, and

3) for all < ↵, the cardinal  is a limit of -hyper -inaccessible cardinals.

 15

Definition ?? subsumes the previous definitions of hyper-inaccessible cardinals. For ex-

ample, 0-hyper0 -inaccessible is just inaccessible, since the second and third parts of the

definition are not applicable. If  is ↵-hyper0 -inaccessible, then by definition,  is inaccessi-

ble and for all < ↵, the cardinal  is a limit of -hyper0 -inaccessible cardinals, hence  is

↵-inaccessible. And if  is ↵-inaccessible, then it is inaccessible and a limit of -inaccessible

cardinals, for every < ↵, hence  is ↵-hyper0 -inaccessible. And, 0-hyper-inaccessible is just

hyper-inaccessible since, from the general definition, this means that  is inaccessible and

for every ⌘ < 1, the cardinal  is -hyper⌘ -inaccessible, i.e.,  is -hyper0 -inaccessible, hence

hyper-inaccessible. In the previous definition of hyper-inaccessible, it was required that  be

hyper-inaccessible, but this requirement is included in the second part of Definition ?? since

this implies that  is hyper-inaccessible whenever > 0.

Definition ?? gives a general definition for hyper-inaccessible cardinals. The following

theorem shows that a cardinal  can be at most hyper -inaccessible, using this definition.

Theorem 7. A cardinal  cannot be 1-hyper -inaccessible.

Proof. Suppose  is the least cardinal with the property that  is 1-hyper -inaccessible.

Then, by definition,  is a limit of hyper -inaccessible cardinals. Let <  be hyper -

inaccessible. Then, for all < , the cardinal is -hyper -inaccessible. In particular, since

< , the cardinal is -hyper -inaccessible. This implies that is 1-hyper -inaccessible,

contradicting that  is the least cardinal with this property.

16

In order to define degrees of inaccessible cardinals, as in Definition ??, beyond hyper-

inaccessible, we need more words. Here, I will introduce new adjectives to describe higher

degrees of inaccessible cardinals, analogous to hyper-inaccessible degrees. If  is a hyper -

inaccessible cardinal, call it richly-inaccessible. Then, proceed as before:  is 1-richly-

inaccessible if and only if  is inaccessible and a limit of richly-inaccessible cardinals. Then,

follow the same pattern as before to keep finding higher degrees past supposed obstacles:

if  is -richly-inaccessible, then  is called hyper-richly-inaccessible. If  is -hyper-richly-

inaccessible, then  is called hyper2 -richly-inaccessible. If  is hyper -richly-inaccessible, then

 is richly2 -inaccessible. Then, with these words, the greatest  can be is richly -inaccessible.

Thus, define  to be utterly-inaccessible if and only if  is richly -inaccessible, so that the next

degree of inaccessibility is 1-utterly-inaccessible. A cardinal  is richly-utterly-inaccessible

if and only if  is hyper -utterly-inaccessible. A cardinal  is utterly2 -inaccessible if and

only if  is richly -utterly-inaccessible. This goes on forever: if a cardinal  is utterly -

inaccessible it is called deeply-inaccessible; if a cardinal  is deeply -inaccessible, it is called

truly-inaccessible; if a cardinal  is truly -inaccessible, it is called eternally-inaccessible; if a

cardinal  is eternally -inaccessible, it is called vastly-inaccessible, and so on.

I will end this section with an exploration of how to characterize these degrees. Let I

be the class of inaccessible cardinals. Let I be the inaccessible limit point operator, a class

operator which assigns to a class, its inaccessible limit points:

 17

I(X) = {↵ : ↵ is an inaccessible limit point of X}

And, for classes X↵ , let ↵2ORD X↵ ={ : 2 \↵< X↵ } denote the diagonal intersection

of a collection (of classes). All of the degrees of inaccessible cardinals defined so far are also

definable from I, I, and as follows. First, define I ↵ (X) for all ↵:

I 0 (X) = X

I ↵+1 (X) = I(I ↵ (X))

\
I (X) = I (X)
<

Then, I 0 (I) = I, the class of inaccessible cardinals. And, I(I) = I 1 (I) is the class of

1-inaccessible cardinals, I 2 (I) is the class of 2-inaccessible cardinals, and so forth so that

I ↵ (I) is the class of ↵-inaccessible cardinals. Thus, the class of hyper-inaccessible cardinals

is

\
↵
↵2Ord (I (I)) = { :  2 I ↵ (I)}
↵<

↵
Let H = ↵2ORD (I (I)) be the class of hyper-inaccessible cardinals. Let H1 = I(H)

be the class of 1-hyper-inaccessible cardinals. The class of ↵-hyper-inaccessible cardinals is

18

denoted H↵ . If we take the diagonal intersection of these classes we get H 2 = ↵2ORD (H↵ ),

the class of hyper2 -inaccessible cardinals. Then, applying the inaccessible limit point op-

erator to this class gives I(H 2 ) = H12 , the class of 1-hyper2 -inaccessible cardinals. Then

H3 = 2
↵2ORD (H↵ ), the class of hyper3 -inaccessible cardinals. Then, taking the diagonal

intersection of all the H ↵ degrees of hyper-inaccessibility gives R = ↵2ORD (H

↵
), the class

of richly-inaccessible cardinals. Apply I to R to get R1 , R2 , . . . , R↵ , the class of ↵-richly-

inaccessible cardinals. Applying diagonal intersection to this class gives HR = ↵2ORD (R↵ ),

the class of hyper-richly-inaccessible cardinals. Apply I to this class to get HR1 = I(HR),

the class of 1-hyper-richly-inaccessible cardinals. Then H 2 R = ↵2ORD (HR↵ ), the class of

hyper2 -richly-inaccessible cardinals. Proceed in this way to get H ↵ R, the class of hyper↵ -

richly-inaccessible cardinals. Apply the diagonal intersection to these classes to get R2 =

↵
↵2ORD (H R), the class of richly2 inaccessible cardinals. Build up the classes, R↵ , of richly↵ -

↵
inaccessible cardinals and take the diagonal intersection of these to get U = ↵2ORD (R ),

the class of utterly-inaccessible cardinals. Then HRU DT EV is the class of hyper-richly-

utterly-deeply-truly-eternally-vastly-inaccessible cardinals, and so on.

The words hyper, richly, utterly, and so on mark places in the process of defining degrees

of inaccessible cardinals where we take a diagonal intersection over ↵ 2 Ord of a collection

of classes. Thus, we have designed a notation system for the classes of inaccessible cardinals

using meta-ordinals. This notation system captures all the classes described so far, and can

go as far as you like. It is a notation system for meta-ordinals which is like Cantor’s normal

form for ordinals, but instead of ! we use ⌦, a symbol for the order-type of Ord. The degree
 19

of any inaccessible cardinal will be denoted by t, a formal syntactic expression for a meta-

ordinal of the form ⌦↵ · +⌦⌘ · +· · ·+⌦· + where ↵ > ⌘ > · · · 2Ord, and , , , 2 Ord.

If s and t are meta-ordinals then the ordering is essentially lexicographical. If the degree on

⌦ of the leading term of t is greater than s, then s < t. If s and t have the same greatest

degree of ⌦, then compare the coefficients of the leading term. If these are the same, compare

the next highest powers of ⌦ in s and t and so on. Now that we have these meta-ordinals,

we can describe the classes of degrees of inaccessible cardinals in a uniform way. If ⌘ and

 are hyper-inaccessible cardinals then ⌘ is in the class of ⌘-inaccessible cardinals and  is

in the class of -inaccessible cardinals. But now I can describe both as being in the class

of ⌦-inaccessible cardinals, which is the diagonal intersection of all classes of ↵-inaccessible

cardinals, for ↵ 2 Ord. Defined in this way, the degree of an inaccessible cardinal  can be

described as t-inaccessible for some meta-ordinal t. The only restriction being that all of the

ordinals in t are less than or equal to . In this way, inaccessible cardinals with the same

degree of inaccessibility can be described with the same meta-ordinal.

Definition 8. If t is a meta-ordinal term with ordinals less than , then a cardinal  is

t-inaccessible if and only if  is inaccessible and for every meta-ordinal term s < t having

ordinals less than , the cardinal  is a limit of s-inaccessible cardinals.

This definition of t-inaccessibility for a cardinal  can be extended to include meta-ordinal

terms t with ordinals less than or equal to  by replacing  by ⌦. For example, a cardinal 

which is -inaccessible is now called ⌦-inaccessible. A cardinal  which is ⌦ · -inaccessible

20

is a hyper -inaccessible cardinal which is better called a richly-inaccessible cardinal, or an

⌦2 -inaccessible cardinal.

Here are a few of the classes described previously with this notation system:

 is ⌦-inaccessible ()  is hyper-inaccessible

 is ⌦2 -inaccessible ()  is richly-inaccessible

 is (⌦2 + ⌦)-inaccessible ()  is hyper-richly-inaccessible

 is ⌦3 -inaccessible ()  is utterly-inaccessible

 is (⌦3 + ⌦2 )-inaccessible ()  is richly-utterly-inaccessible

 is ⌦3 · 2-inaccessible ()  is utterly2 -inaccessible

 is ⌦4 -inaccessible ()  is deeply-inaccessible

 is ⌦5 -inaccessible ()  is truly-inaccessible

 is ⌦6 -inaccessible ()  is eternally-inaccessible

 is ⌦7 -inaccessible ()  is vastly-inaccessible

 is (⌦7 + ⌦6 + ⌦5 + ⌦4 + ⌦3 + ⌦2 + ⌦ + ↵)-inaccessible ()  is ↵-hyper-richly-utterly-

deeply-truly-eternally-vastly-inaccessible

Now that we have this uniform notation, we have the following theorem:

Theorem 9. If  is t-inaccessible, then there is a forcing extension where  is still t-

inaccessible, but not (t + 1)-inaccessible.

 21

Proof. Sketch. The proof is very similar to the proof of Theorem ??. Assume V |= ZFC. Sup-

pose  2 V is t-inaccessible, where t is a meta-ordinal term with only ordinal exponents and

coefficients less than . Force with C. Conditions are closed, bounded subsets of  such that

if c 2 C then c contains no t-inaccessible cardinals. The forcing C preserves cardinals and

cofinalities. Since 8 < , the set D = {d 2 C : max(d) > } is dense, the new club C =

[G ✓ C is unbounded. It follows that C is closed, since it is unbounded and the conditions

are closed. For every s < t, where s is a meta-ordinal term with ordinal parameters less than

, the set Ds = {d 2 C : dcontains a block of cardinals up to an s-inaccessible cardinal}

is dense: if c 2 C, then if ⌘ is the least s-inaccessible cardinal above ⌘, then d = c [

({[max(C), ⌘]} \ CARD) is in Ds .

+
Next force with E: For infinite 2 C, force 2 = +
where is the next element of C

past +
. E destroys strong limits which are not in C 0 : If ⌘ is a strong limit which is not a

+
limit point of C, then there is a maximal element of C below ⌘. Call it . Then < ⌘ and

+ +
the next element of C, call it , is above ⌘. Thus 2 = > ⌘ destroys that ⌘ is a strong

+
limit. If ⌘ 2 C 0 and ⌘ is a strong limit, then +
< ⌘ so 2 < ⌘ for every < ⌘. Thus, ⌘ is

still a strong limit in the final extension.

Since C contains unboundedly many ground model s-inaccessible cardinals for s < t,

with a block of cardinals below them, then in the final extension these cardinals are still
22

s-inaccessible (by a proof by meta-ordinal induction on s), and so  is still t-inaccessible.

However since all t-inaccessible cardinals below  are no longer strong limits,  is not t + 1-

inaccessible. ⇤

Finally, consider the connection between sets of inaccessible cardinals below  and re-

gressive functions on . Consider the set I 2 (I) of 2-inaccessible cardinals. If  2

/ I 2 (I)

this means that there is a tail [ , ) where, for each inaccessible in [ , ), there is, b , the

greatest inaccessible cardinal below . Thus, I can define a regressive function f : [ , ) ! 

with f (↵) = ↵. Thus, for every ↵, there are no inaccessible cardinals between f (↵) and ↵.

/ I 2 (I) if and only if there is a regressive function whose domain

In general, I can say that  2

is a tail of , with the property that there are no inaccessible cardinals between f (↵) and ↵

/ I 3 (I) if and only if there is a regressive

for any ↵ in the domain. Similarly, I can define  2

function f :  !  such that for all inaccessible ↵ < , we have I 2 (I) \ (f (↵), ↵) = ;,

and there exists a regressive function g↵ : ! such that for all 2 (f (↵), ↵), we have

I 1 (I) \ (g↵ ( ), ) = ;. Thus, no ↵ <  is 2-inaccessible, and thus  cannot be a limit of

/ I 3 (I), but  could still be 2-inaccessible.

2-inaccessible cardinals so  2
 23

3. Mahlo Cardinals

This section begins with theorems about Mahlo cardinals, showing that Mahlo cardinals

have all the inaccessible large cardinal properties from the last section. An analogue of the

classical notion of greatly Mahlo is defined for inaccessible cardinals. Also, the classical

degrees of Mahlo cardinals are described, and theorems follow which show how to force a

large cardinal  to have a desired Mahlo degree in the extension. An infinite cardinal  is

Mahlo if and only if the set of inaccessible cardinals below  is a stationary subset of . A

cardinal  is greatly inaccessible if and only if there is a uniform, normal filter on , closed

under the inaccessible limit point operator:

I(X) = {↵ 2 X : ↵ is an inaccessible limit point of X}

A uniform filter on  has for every < , the set [ , ) is in the filter, and normal means

that the filter is closed under diagonal intersections ↵< . The last part of the definition

means that if X is in the filter, so is I(X). The first theorem of this section shows that

greatly inaccessible is equivalent to Mahlo.

Theorem 10. A cardinal  is greatly inaccessible if and only if  is Mahlo.

24

Proof. For the forward implication, suppose  is Mahlo. Let F be the filter generated by

sets of the form C \ I, where C is club in , and I is the set of inaccessible cardinals below

. Note that  is club in itself, and  \ I = I, so I 2 F . Then, the claim is that F is a

uniform, normal filter, closed under I. First, if C, D are club in , then C \ D is club. Thus,

for any clubs C and D, the equation (C \ I) \ (D \ I) = (C \ D) \ I implies that any set A

in the filter, generated by sets of the form C \ I, where C is club, is itself a superset of a set

of the form E \ I, where E is club in . Then, ; 2

/ F , since the empty set has no non-trivial

subsets, hence cannot be a superset of the form C \ I. Next, if A 2 F , and A ✓ B, then

there is a club C 2 F such that C \ I ✓ A ✓ B, thus B 2 F , by construction, since F is

the filter generated by sets of this form. Third, if A and B are elements of F , then there

are clubs C and D such that C \ I ✓ A, and D \ I ✓ B, so A \ B contains (C \ D) \ I.

This is of the form which generated the filter, thus A \ B 2 F . Also, F is uniform since

if b is a bounded subset of , then  \ b contains a club E. Thus, E \ I ✓ E ✓  \ b, so

 \ b 2 F . Since the cardinality of any co-bounded set is , the filter is uniform. It remains

to show that F is normal, and closed under the inaccessible limit point operation. To see

that F is normal, i.e. closed under diagonal intersection, see first that club sets are closed

under diagonal intersection. Let C↵ be club for every ↵ < . Then C↵ is closed, since,

if is a limit point of C↵ , then is a limit point of each C↵ , for ↵ < . To see this, let

↵ < . Then, since is a limit point of C↵ , there exists a such that ↵ < < , with

2 C↵ , i.e., 2 \ < C . Since ↵ < , this implies that 2 C↵ . Hence, that is a

limit point of C↵ implies is a limit point of C↵ , for all ↵ < . Thus, since each C↵ is
 25

closed, 2 C↵ , for all ↵ < . Thus 2 \↵< C↵ . Thus, 2 C↵ . Thus, C↵ is closed.

To see that C↵ is unbounded, observe that for any < , the set \↵< C↵ is club. I’ll

show now that C↵ is unbounded, using this fact. Suppose that < . Since \↵< C↵ is

unbounded, there is 1 2 \↵< C↵ , with 1 > . Then since \↵< 1 C↵ is unbounded, there

is 2 2 \↵< 1 C↵ with 2 > 1. Continue in this way, to get an increasing sequence h n i,

where n+1 2 \↵< n C↵ . Taking the limit of this sequence, limn!1 n = !, and the limit

of the intersection of clubs gives that ! 2 \↵< ! C↵ , and ! > , since it is the limit of

an increasing sequence above . Thus, 2 C↵ , is a witness that C↵ is unbounded.

Finally, for this direction, to show Mahlo implies greatly inaccessible, F is closed under I,

since if A 2 F , then I(A) = I \ A0 \ A, and A0 is club, so I \ A0 2 F , and A 2 F , so that

I \ A0 \ A 2 F , and thus I(A) 2 F .

For the other direction, if  is greatly inaccessible, then the uniform, normal filter, F , on

, contains all the club subsets of . This is true since all tails ( , ) are in F , since  is

regular, and F is uniform, and all clubs can be written as the diagonal intersection of tails.

Let C be club in . Then define

8
>
>
>
>
<[↵, ), if ↵ 2 C
AC
↵ =
>
>
>
>
:[ , ), if ↵ 2
/ C, where next element of C, past ↵

It is claimed that C = AC
↵ . Let x 2 C. Then, for every < x, the set AC contains [x, ),

C C
since the next element of C past is at most x. Thus, x 2 \ <x A so that x 2 <x A .
26

For the other direction, suppose x 2 AC

↵ . Then, for every < x, x 2 AC . Thus, let < x,

then either 2 C or the next element of C past is below x since x 2 AC . Thus, below x,

there is an unbounded subset of x whose elements are in C. Thus, x is a limit point of C.

Since C is closed, x 2 C. Thus C = AC

↵ . Thus, since F is normal, C 2 F . Then, since

 2 F , and I() = I 2 F , and for any club D ✓ , the set D is in F , and their intersection,

D \ I, is in F , hence D \ I is non-empty. Thus I is stationary in , and thus  is Mahlo.

Theorem ?? below shows that a Mahlo cardinal is every degree of inaccessibility defined

previously. However, a cardinal being every degree of inaccessibility is not equivalent to the

cardinal being Mahlo. Theorem ?? separates the two notions with forcing to destroy the

Mahlo property of a cardinal, while preserving that the cardinal is every inaccessible degree.

Theorem 11. If  is Mahlo, then for every meta-ordinal term t having ordinals less than

, the cardinal  is a t-inaccessible cardinal as in Definition ??)

Proof. If  is Mahlo, then  is greatly inaccessible, by Theorem ??. Hence, there is a uniform,

normal filter F on , closed under the inaccessible limit point operation. Since  2 F , and

I() = I 2 F , all sets of ↵-inaccessible cardinals for ↵ < , below , are in F . And, since F

is normal, all sets of hyper inaccessible degrees below , which are all definable from diagonal

intersection, are in F . Hence, every set of inaccessible cardinals definable from I, I, and ,
 27

in the manner of the previous section, by closing under these operations, is unbounded in

. ⇤

Theorem ?? of the last section shows that any cardinal which is at least ↵-inaccessible in

the ground model, for some ↵, can be made to be no more than ↵-inaccessible in a forcing

extension. Therefore, if  is ↵-inaccessible, then there is a forcing extension where  is

↵-inaccessible, but not Mahlo. The following theorem shows that if  is Mahlo, there is a

forcing extension where  is no longer Mahlo but still any degree of inaccessibility defined

in the previous section.

Theorem 12. If  is Mahlo, then there is a forcing extension where  is t-inaccessible for

every meta-ordinal term t with ordinals less than or  , but where  is not Mahlo.

Proof. Let  be a Mahlo cardinal. Let C be the forcing to add a club, C = [G, where

G ✓ C is V -generic, which contains no inaccessible cardinals. The argument in Theorem ??

shows that C preserves cardinals, cofinalities, and all inaccessible cardinals in V . In fact,

V [G]
C did not change V at all, V = V . Since V has all the sets needed to define that  is

↵-inaccessible, for cardinals < , being ↵-inaccessible is absolute to V . Hence C preserves

all inaccessible degrees. Hence  is still every possible inaccessible degree. However,  is no

longer Mahlo in V [G], since C \ I = ;.

⇤
28

Just as there are infinitely many degrees of inaccessible cardinals, there are infinitely many

Mahlo degrees. One might be tempted to define the next degree of Mahlo to be a Mahlo

limit of Mahlo cardinals, exactly as was done with the degrees of inaccessible cardinals, and

indeed there is a hierarchy of Mahlo cardinals that can be defined this way. However, there

is a more powerful and appropriate way to define the degrees of Mahlo cardinals which is

much stronger. The definition of Mahlo cardinal is primarily about stationary sets, and

the degrees of Mahlo cardinals are classically defined by stationary sets. Thus, the classical

degrees of Mahlo cardinals are defined using stationary sets. Namely, an infinite cardinal 

is 1-Mahlo if and only if  is Mahlo, and the set of Mahlo cardinals below  is stationary in

. In general,  is ↵-Mahlo if and only if  is Mahlo, and for every < ↵, the set of -Mahlo

cardinals below  is stationary in . The degrees of Mahlo cardinals go on forever, just as

the degrees of inaccessible cardinals; if  is -Mahlo, then  is hyper-Mahlo, and so on.

The main theorem of this section will show how to force to change degrees of Mahlo

cardinals by adding a club avoiding a stationary set of cardinals of a certain Mahlo degree,

while preserving all stationary subsets of cardinals of a lesser Mahlo degree. A modification

of the forcing C, to add a club, from the proof of Theorem ??, will work. Given two sets

A and B, it is said that A reflects B if for some 2 B, the set A \ is stationary in .

The following lemma shows that if A does not reflect in B, then the forcing to add a club

through B will preserve the stationary subsets of B.

 29

That is, if we force to add a club which avoids the strong set, then the stationarity of the

weak set and all of its stationary subsets will be preserved if the strong set does not reflect

in the weak set. In the proof of the theorem, the strong set will be Mahlo cardinals of a

fixed degree, and the weak set will be the set of Mahlo cardinals of a lesser degree which are

not of the fixed degree.

Lemma 13. If  is Mahlo, and the sets A and B partition the inaccessible cardinals below

, where A does not reflect in B, then the forcing to add a club avoiding A will preserve all

the stationary subsets of B. Furthermore, the forcing does not add sets to V .

Proof. Suppose S 2 V is a stationary subset of B. Let P be the forcing to add a club through

the complement of A, and let Ė be a name for a club subset of , and p a condition which

forces that E is club in . Let 2 S be such that hV , 2, A \ , B \ , Ė \ V i is an elementary

substructure of hV , 2, A, B, Ėi. We can find such 2 S, since the set of giving rise to

elementary substructures is club and S is stationary. Since 2 B, by the assumption, there is

a club set c ✓ , containing no points from A. Now we construct a pseudo-generic -sequence

of conditions below , deciding more and more about Ė, using the elements of c to guide the

construction. Build a descending sequence of conditions hc↵ : ↵ < i in P \ V , below p, and

given c↵ , choose c↵+1 to force a specific ordinal above ↵ into Ė with sup(c↵+1 ) 2 c, and at

limits , let c = ([↵< c↵ ) [ {sup([↵< c↵ }. Notice that sup([↵< c↵ ) 2 c, since c is closed,

and therefore is not in A, so that c is a condition for limit ordinals , in the construction.

That is, we can get through the limit steps below , precisely because c contains no points
30

from A. Let c⇤ = ([↵< ) [ { }. Then c⇤ ✓ is club in that decides Ė in a way that is

unbounded in . Thus, c⇤ is a condition which forces Ė meets S. Thus, S must still be

stationary in the extension. Finally, this club shooting forcing does not add sets to V by

Lemma ??

The main result of this section is that if  is ↵-Mahlo, then there is a forcing extension

where  is still ↵-Mahlo, but no longer (↵ + 1)-Mahlo. That is,  is forced to maximal Mahlo

degree, ↵-Mahlo. In fact, a very similar argument can be used to show that if  is Mahlo

of any fixed hyper-degree (for example if  is hyper-richly-Mahlo), it can be forced to be

↵-Mahlo (but not hyper-richly Mahlo).

Theorem 14. If  is ↵-Mahlo, then there is a forcing extension where  is ↵-Mahlo, but

not (↵ + 1)-Mahlo.

Proof. Let ↵ <  be fixed. Suppose  is ↵-Mahlo. Let A be the set of ↵-Mahlo cardinals

below . Note that A does not reflect in its complement in the inaccessible cardinals, for if,

for some < , inaccessible, in the complement of A, the set of ↵-Mahlo cardinals below

is stationary, then is (↵ + 1)-Mahlo, hence ↵-Mahlo, hence 2 A. But, was supposed

to be an inaccessible cardinal in the complement of A. Thus, A does not reflect in the set of

inaccessible cardinals in its complement.

 31

If  is not (↵ + 1)-Mahlo, then force trivially to show the result. Let C be the notion

of forcing which adds a club C through the complement of A. Conditions are ordered by

end-extension. Then by Lemma ?? the forcing C preserves truth in V , preserves cardinals,

cofinalities, and adds a club to , disjoint from A. Since A is no longer stationary in , it is

no longer (↵ + 1)-Mahlo. Since A contains no clubs,  \ A is stationary. Thus, by Lemma ??,

since A does not reflect in its complement, all stationary subsets of \A are preserved. Thus,

I only need to show that 8 < ↵ the set T = { <  : is -Mahlo but not ( +1)-Mahlo }

is stationary. Let C ✓  be club. Let be least in C 0 which is -Mahlo. Then C \ is

club in so that (C \ )0 \ = C 0 \ is club in and contains no -Mahlo cardinals by the

minimality of . Thus is not ( + 1)-Mahlo. Thus T is stationary. Thus the stationarity

of T is preserved. Thus  is still ↵-Mahlo.

As in the previous section for inaccessible cardinals (Definition ??), we can define the

higher degrees of Mahlo cardinals:

Definition 15. A cardinal  is ↵-hyper -Mahlo if and only if

1) the cardinal  is Mahlo, and

2) for all ⌘ < , the cardinal  is -hyper⌘ -Mahlo, and

3) for all < ↵, the set of -hyper -Mahlo cardinals below  is stationary in .
32

Also, as in Definition ?? for degrees of inaccessible cardinals, we can define the classes of

degrees of Mahlo cardinals uniformly by using meta-ordinals:

Definition 16. If t is a meta-ordinal term having only ordinals less than , then a cardinal

 is t-Mahlo if and only if for every meta-ordinal term s < t having only ordinals less than

, the set of s-inaccessible cardinals is stationary in .

Theorem 17. If  is t-Mahlo, where t is a meta-ordinal term having parameters less than

, then there is a forcing extension V [G] where  is t-Mahlo, but not (t + 1)-Mahlo.

Proof. Sketch. The proof is a generalization of the proof of Theorem ??. The proof is to

add a club C ✓  which avoids the t-Mahlo cardinals below . ⇤

To end this section, we take a look at a large cardinal property below inaccessible, and

large cardinal properties just above inaccessible. A cardinal  is worldly if and only if V

is a model of ZFC. If  is inaccessible, then it is worldly since V |= ZFC. Therefore, the

first theorem is about singular worldly cardinals, which are not inaccessible. A cardinal  is

⌃n -worldly if and only if V satisfies the ⌃n fragment of ZFC. The following theorem shows

how to force a worldly cardinal to be ⌃n -worldly, but not ⌃n+1 -worldly [Hamkins3].

Theorem 18. (Hamkins) If ✓ is any singular worldly cardinal, then for any n 2 N, there is

a forcing extension where ✓ is not worldly, but still ⌃n -worldly.

 33

The next theorem shows that it is not always possible to force between large cardinal

properties. It shows that if  is inaccessible, there may not be a forcing extension where 

is worldly but not inaccessible.

Theorem 19. (Fuchs) If the existence of an inaccessible cardinals is consistent with ZFC,

then it is consistent with ZFC that a cardinal  is inaccessible and has no forcing extension

where  is worldly and not inaccessible.

Let me say a little about the proof. This proof will show that if  is a regular worldly

cardinal, it may not be possible to force that  is a singular worldly cardinal. Suppose  is a

regular worldly cardinal and there is a forcing extension V [G] where  is a singular worldly

cardinal. Then, by a covering lemma argument with the Dodd-Jensen core model, it follows

that  is measurable in an inner model [Fuchs]. Since not all universes have inner models

with measurable cardinals, for some V and  inaccessible, there is no extension where  is a

singular worldly cardinal.

The next theorem is about large cardinal properties just above inaccessible. A cardinal

 is ⌃n -correct if and only if V ⌃n V . A cardinal  is ⌃n -reflecting if and only if  is

⌃n -correct and inaccessible. The following theorems show how force to change degrees of

reflecting cardinals.

Theorem 20. If  is ⌃1 -reflecting, then there is a forcing extension where  is still ⌃1 -

reflecting, but not ⌃2 -reflecting.

34

Proof. First, let us establish that all inaccessible cardinals are ⌃1 -reflecting. Then, we shall

prove the theorem by exhibiting a forcing notion which does not preserve that a cardinal is

⌃2 -reflecting, but which does preserve that the cardinal is inaccessible (hence ⌃1 -reflecting).

Suppose  is inaccessible. It follows that V = H . The Lévy reflection theorem says that

H ⌃1 V for every uncountable cardinal . Thus, if  is inaccessible, V ⌃1 V , and so it

is ⌃1 -reflecting. Let P be the canonical forcing of the GCH, and let G ✓ P be V -generic.

Next force over V [G] to make 2 = ++ in V [G][g], where g is V [G]-generic. Since the

canonical forcing of the GCH does not change that  is inaccessible, and forcing 2 = ++

also cannot change that  is a strong limit (or regular), it follows that  is inaccessible in

V [G][g]. Thus  is ⌃1 -reflecting in the final extension. However, V [G][g] = V [G] |= GCH

while V [G][g] |= 2 = ++ . Thus,  is not ⌃2 -reflecting in V [G][g], since V [G][g] thinks the

GCH holds everywhere while, V [G][g] sees a violation of GCH at . ⇤

 35

4. Weakly compact cardinals and beyond

In this section, forcing notions are found which distinguish pairs of large cardinals in the

[Mahlo, measurable] interval. The first theorem separates the notions of Mahlo and weakly

compact for a given cardinal with these properties. A cardinal  is weakly compact if and only

if it is uncountable and satisfies the partition party  ! ()2 . That is,  is weakly compact if

and only if for every coloring of pairs of elements of  into two colors, there is a homogeneous

subset of  of order-type . There are other various equivalent characterizations of weakly

compact cardinals. An important characterization is that weakly compact cardinals are

inaccessible and have the tree property. An uncountable cardinal  has the tree property

if every tree of height  whose levels have cardinality less than  has a branch of length

 [Jech p. 120]. For example, !1 does not have the tree property since there exists an

Aronszajn tree, a tree of height !1 with no uncountable branches. A -Aronszajn tree is a

tree of height  with countable levels and no branches of length . And, CH implies there

is an !2 -Aranszajn tree, so in this universe, !2 does not have the tree property. A -Souslin

tree is a tree of height  with no chains or antichains of size . Thus a -Souslin tree is a

-Aronszajn tree, since a chain is a branch. Thus, the existence of a -Souslin tree destroys

the tree property of , thus its weak compactness.

The following theorems are analogues of Theorem ?? and Theorem ?? from the previous

section. First, we see that every weakly compact cardinal is every degree of Mahlo.
36

Theorem 21. If  is weakly compact then for every meta-ordinal term t having only ordinals

less than , the cardinal  is t-Mahlo.

Proof. Suppose  is weakly compact. By induction suppose  is s-Mahlo for all s < t. We

want to show that  is t-Mahlo. Fix any club C ✓ . Put C 2 M |= ZFC where M is

a -model, M < ✓ M . Get j : M ! N with cp(j) = . Without loss of generality, we

can assume V ✓ M . Since degrees of Mahloness are downward absolute the fact that  is

s-Mahlo in V implies  is s-Mahlo in N . Since C is club, it follows that  2 j(C) so that

N |= “the set of s-Mahlo cardinals meets j(C)”. By elementarity, M |= “C has an s-Mahlo

cardinal”. Thus C has an s-Mahlo cardinal since M is a -model. Thus  is t-Mahlo. ⇤

Since the degrees of Mahlo cardinals discussed in the previous section are characterized

by stationary sets, it follows that if a notion of forcing preserves stationary sets, it preserves

any degree of Mahlo. This theorem shows how to force a weakly compact cardinal to lose

this property in a forcing extension while retaining that it is every possible degree of Mahlo.

Theorem 22. If  is weakly compact, then there is a forcing extension where for every

meta-ordinal t having only ordinals less than , the cardinal  is t-Mahlo but not weakly

compact.

Proof. Suppose  is weakly compact. Let S be the forcing to add a -Souslin tree. Conditions

in S are normal trees of height ↵ + 1 for some ↵ < . The conditions in S are ordered by

end-extension: s  t if and only if s is a tree which contains t as an initial segment. Also, it is

 37

required that for all T 2 S, the condition T is a normal tree: every element of T is a -length

binary sequence for some  ↵, and if t ⇢ s and s 2 T then t 2 T , and every node splits in

two at every level, and for every node t and every level there is a node s at level which

extends t, and finally |T | < . Let G ⇢ S be V -generic. Let S = [G be the new tree. The

generic object S is a -Souslin tree [generalization of Jech p. 239]. These are the conditions.

The forcing notion S is -strategically closed since we can play the game along the paths

as not to get stuck at an Aronszajn tree of height less than . That is, the second player

keeps track of the paths which are played and always chooses nodes which continue paths

already chosen, so that at limits player two can continue play. Thus, S preserves stationary

sets. Thus, the forcing preserves that  is any degree of Mahlo since the set of inaccessible

cardinals, set of Mahlo cardinals, set of 1-Mahlo cardinals, etc, below , remain stationary

in . However, since the new tree, S, is a -Aronszajn tree, it does not have a branch of

length . Therefore,  no longer has the tree property, so it is no longer weakly compact.

A cardinal  is measurable if and only if there is a transitive class M , and an elementary

embedding j : V ! M with critical point . A cardinal  is weakly measurable if and only

if for every family A ✓ P (), of size at most + , there is a non-principal -complete filter

on  measuring every set in A. And, if ()< = , then a cardinal  is weakly measurable

if and only if for every transitive set M of size + , with  2 M , there exists a transitive

N and an elementary embedding j : M ! N with critical point . For the next large
38

cardinal definition, we need the definition of a -model: a transitive set M is a -model if

and only if M |= ZF C where M is size  with  2 M and M < ✓ M . Then, a cardinal  is

strongly Ramsey if every A ✓  is contained in a -model, M , for which there exists a weakly

amenable M -ultrafilter on . An M -ultrafilter U is said to be weakly amenable (to M ) if

for every A 2 M of size  in M , the intersection A \ U is an element of M . An uncountable

regular cardinal  is ine↵able if for every sequence hA↵ | ↵ < i, with A↵ ✓ ↵, there is A ✓ 

such that the set S = {↵ <  | A \  = A↵ } is stationary. Measurable cardinals are weakly

measurable; weakly measurable cardinals are strongly Ramsey cardinals; strongly Ramsey

cardinals are not necessarily ine↵able cardinals, but are below strongly Ramsey cardinals in

consistency strength; and ine↵able cardinals are weakly compact. The following theorems

show how to force the di↵erence between these large cardinal properties for a given cardinal.

The following theorem, by Jason Schanker [Schanker], fits exactly in the theme:

Theorem 23. (Schanker) If  is weakly measurable, then the measurability of  can be

destroyed while preserving that it is weakly measurable.

The preservation part of the following theorem, and the forcing used, is due to Victoria

Gitman:

Theorem 24. If  is strongly Ramsey, then there is a forcing extension where  is not

weakly measurable, but is still strongly Ramsey.

 39

Proof. Let  be strongly Ramsey. Make a preparation P to make  indestructible by adding

a slim--Kurepa tree [Gitman]. Let G ✓ P be V -generic. Let K be the forcing to add a

slim--Kurepa tree over V [G]. Let H ✓ K be V [G]-generic. In the extension,  is still

strongly Ramsey [Gitman]. But,  cannot be weakly measurable in V [G][H]. Suppose, by

way of contradiction, that  is weakly measurable in the final extension. Let T 2 V [G][H]

be a slim--Kurepa tree, and let |M | = + , be a transitive set containing , the tree T , and

the + many branches of T . Since  is weakly measurable, there is an N , and an elementary

embedding j : M ! N with cp(j) = . Then, by elementarity, j(T ) is a slim-j()-Kurepa

tree. But, since T has + many paths, and j(T ) is an end-extension of T , each of the +

many paths of T extends to a node on the th level of j(T ). Thus, there are + many nodes

on the th level of j(T ), and M can see all the paths of T . So, N can see that j(T ) is

not a slim-j()-Kurepa tree, which contradicts that j(T ) is a slim-j()-Kurepa tree in N by

elementarity. Thus,  is not weakly measurable in V [G][H]. ⇤

The following theorem shows how to force to remove a large cardinal property while

preserving is greater large cardinal property. If a cardinal  is strongly Ramsey, then it is

consistent that the cardinal is also an ine↵able cardinal. However, the following theorem

shows how to force to an extension where  is not ine↵able, but still strongly Ramsey.

Theorem 25. If  is strongly Ramsey and ine↵able, then there is a forcing extension where

 is not ine↵able, but it is still strongly Ramsey.

40

Proof. Suppose  is ine↵able. The forcing to add a slim--Kurepa tree, as in the previous

theorem, will preserve that  is strongly Ramsey [Gitman]. However, forcing to add such a

tree ensures that  is not ine↵able in the extension since the characterization of  ine↵able

implies there are no slim--Kurepa trees [Jensen/Kunen]. ⇤

 41

5. Measurable cardinals

In this section, degrees of measurable cardinals are changed by forcing. A cardinal  is

measurable if and only if  is uncountable and there is a -complete nonprincipal ultrafilter on

. Suppose  is measurable by measure µ. Then, the corresponding ultrapower embedding

j : V ! Mµ , by measure µ, is an elementary embedding with critical point . Also, if 

is the critical point of a non-trivial elementary embedding, then it is measurable. Thus,

another characterization of measurable cardinals is the following embedding definition:  is

measurable if and only if it is the critical point of an elementary embedding j : V ! M in

V.

The theorems in this section are about reducing Mitchell degrees of measurable cardinals in

a forcing extension. The Mitchell order is defined on normal measures; for normal measures

µ and ⌫, the following relation u C ⌫ holds when µ 2 M⌫ , where M⌫ is the ultrapower

by ⌫. If  is measurable, then the Mitchell order on the measures on  is well-founded

(and transitive) [Jech p. 358]. Thus, let m() be the collection of normal measures on .

The definition of rank for µ 2 m() with respect to C is o(µ) = sup{o(⌫) + 1 |⌫ C µ}.

The Mitchell order on , denoted o(), is the height of the Mitchell order on m(), thus

o() = sup{o(µ) + 1 | µ 2 m()}. If  is not measurable, then o() = 0, since m() is empty.

A cardinal  is measurable if and only if o() 1. For o() 2, there is a measure µ on ,

such that if j : V ! Mµ is the corresponding ultrapower embedding, there is a measure on 

in Mµ . Thus  is measurable in Mµ . Thus Mµ |=  2 j({ <  | is measurable }), and so

42

the set { <  | is measurable } is in µ. That is, µ concentrates on measurable cardinals.

Therefore, o() 2 if and only if there is a normal measure on  which concentrates on

measurable cardinals. A cardinal  has Mitchell order o() 3, if and only if there is a

normal measure on  which concentrates on measurable cardinals with Mitchell order 2. A

cardinal  has o() ↵ if and only if for every < ↵, there is a normal measure on  which

concentrates on measurable cardinals of order .

The following theorem shows how to force any measurable cardinal, of any Mitchell rank,

to have order exactly one. In the proof of the following theorem, and the proof of the

main theorems, we will need that the forcing notions do not create large cardinals. In the

previous sections, since being Mahlo or inaccessible is downwards absolute, we knew that

the forcing could not create new Mahlo or inaccessible cardinals. However, it is possible that

forcing create large cardinals in the extension. Therefore, we use the approximation and

cover properties of [Hamkins4] to ensure that no new measurable cardinals are created.

Theorem 26. If  is a measurable cardinal, then there is a forcing extension where o() = 1.

In addition, the forcing will preserve all measurable cardinals and will not create any new

measurable cardinals.

Proof. Suppose  is a measurable cardinal. Assume that 2 = + , or force it to be true since

forcing to have the GCH hold at  preserves the measurability of . Pick j : V ! M , an

elementary ultrapower embedding by µ, a normal measure on  which has minimal Mitchell

rank o(µ) = 0, with cp(j) = , such that M |=  is not measurable. Let P be a -length
 43

iteration, with Easton support, of Q , a P -name for the forcing to add a club to of

cardinals which are not measurable in V , whenever <  is inaccessible, otherwise force

trivially at stage . Let G ✓ P be V -generic, let Q̇G = Q, and let Q̇ be a P-name for a

forcing to add a club of cardinals to  which are not measurable in V [G]. Let g ✓ Q be

V [G]-generic. Let 0 is the first inaccessible cardinal. This is the first non-trivial stage of

P, and at this stage a club is added to 0. Thus the forcing up to and including stage 0

has cardinality at most 0 and the forcing after stage 0 is closed up to the next inaccessible

cardinal. Thus, P has a closure point at 0. Thus by Corollary 22 in [Hamkins4], the forcing

does not create measurable cardinals.

We want to lift j through P and Q to find j : V [G][g] ! M [j(G)][j(g)] which witnesses

that  is measurable in V [G][g]. The main task is to find j(G), an M -generic filter for j(P),

and j(g) an M [j(G)]-generic filter for j(Q). Since the critical point of j is , the forcings P

and j(P) are isomorphic up to stage . Also, since Q̇ is a P-name for a forcing which adds

a club of ground model non-measurable cardinals to  and V [G] and M [G] have the same

bounded subsets of , the forcing at stage  of j(P) is Q̇. Thus, j(P) factors as P ⇤ Q̇ ⇤ Ptail ,

where Ptail is the forcing j(P) past stage . Since G is V -generic, it is also M -generic. Thus,

form the structure M [G]. Similarly, since g is V [G]-generic, it follows that g is M [G]-generic,

so construct the structure M [G][g]. Since both P and Q have size , they both have the + -

chain condition, so the iteration P ⇤ Q has the + -chain condition. Also, since j 00  =  2 M ,

and j is an ultrapower embedding, it follows that M  ✓ M in V . Thus, since G ⇤ g is

V -generic, M [G][g] ✓ M [G][g] [Hamkins1 (Lemma 54)].

44

The goal is to diagonalize to find an M [G][g]-generic filter for Ptail in V [G][g], and so far

we have satisfied one criterion which is that M [G][g] is closed under  sequences in V [G][g].

Next, since |P| = , and 2 = + , it follows that P has at most + dense sets, and so does any


tail P[ ,) . Thus, Ptail has at most |j(+ )|V  + = + dense subsets in M [G][g]. Finally,

since for every < , there is, in M [G][g], a dense subset of P which is  -closed, there

is a dense subset of Ptail which is  -closed. Thus, it is possible to diagonalize to get an

M [G][g]-generic filter Gtail ✓ Ptail , in V [G][g]. Thus, j(G) = G ⇤ g ⇤ Gtail ✓ j(P) is M -generic,

and j 00 G ✓ G ⇤ g ⇤ Gtail . Therefore, the lifting criterion is satisfied, and so the embedding

lifts to j : V [G] ! M [j(G)].

The next goal is to lift the embedding through Q. The forcing j(Q) is the forcing to add

a club of non-measurable cardinals to j(), and we need an M [j(G)]-generic filter for j(Q).

Since Gtail 2 V [G][g], it follows that M [j(G)] ✓ M [j(G)]. Also, since |Q| = , it has at


most + many dense sets in V [G], by elementarity j(Q) has at most |j(+ )|V [G]  + = +

many dense subsets in M [j(G)]. For any < , the forcing Q has a dense subset which

is  -closed. Since  < j(), the forcing j(Q) has a dense subset which is  -closed.

Note that c = [g, which is club in , is in M [j(G)]. Let c̄ = c [ {}. For all < ,

we have 2 c then is not measurable in V . Thus, 2 c implies is not measurable in

M , since M and V have the same P (P ( )) since < . By Corollary 22 in [Hamkins4]

since M ✓ M [j(G)] also satisfies approximation and cover properties, it follows that is not

measurable in M [j(G)]. By choice of j : V ! M , we have that  is not measurable in M .

Thus  is not measurable in M [j(G)], again by [Hamkins4]. Thus c̄ is a closed, bounded

 45

subset of j() which contains no measurable cardinals. Thus c̄ 2 j(Q). Thus, diagonalize to

get a generic filter g ⇤ ✓ j(Q), in V [G], which meets a dense subset which is  -closed, and

which contains c. Thus, let j(g) = g ⇤ , and observe that j 00 g ✓ g ⇤ since g 2 [g ⇤ . Thus, the

embedding lifts to j : V [G][g] ! M [j(G)][j(g)], in V [G][g]. Therefore,  is still measurable

in V [G][g]. Thus o() 1 in V [G][g].

Finally o()V [G][g]  1 will follow from the fact that Q adds a club subset c ✓  which

contains no cardinals which are measurable in V . Since the combined forcing does not create

measurable cardinals, the club c contains no cardinals which are measurable in V [G][g]. Since

the new club has measure one in any normal measure on , the complement of c, which

contains the measurable cardinals of V [G][g] below  has measure zero. Thus, there is no

normal measure on  in V [G][g] which concentrates on the measurable cardinals of V [G][g].

Hence o()V [G][g] ⇧ 1. Thus, o() = 1 in V [G][g]. That the forcing preserves all measurable

cardinals and creates no new measurable cardinals will follow from the proof of the more

general Theorem ??. ⇤

The previous theorem shows how to make the Mitchell order of any measurable cardinal

exactly one, for any measurable cardinal. The next theorem will show how to make the

Mitchell order of any measurable cardinal exactly ↵, for any ↵ < , in a forcing extension,

whenever the order is at least ↵. Preceding the proof are the following lemmas, so that we

may generalize the proof for any ↵ < .

46

Lemma 27. Let  be an infinite cardinal and < = . Let S ✓  be a subset of 

which contains the singular cardinals. Then QS , the forcing to add a club C ⇢ S, preserves

cardinals and cofinalities, and for all < , the forcing QS has a  -closed dense subset.

Proof. Conditions c 2 QS are closed, bounded subsets of S. The ordering on QS is end-

extension: c  d if and only if a 2 c \ d ! a > max(d). Let G ✓ QS be V -generic, and let

C = [G. Then C ✓ S is club in . First, see that C is unbounded. Let 2 , and let

D = {d 2 QS : max(d) > }. Then, D is dense in QS , since if c 2 QS , the following is a

condition: d = c [ { 0 }, where 0
> and 0
is in S. Then, d is an end-extension of c, hence

stronger, and d 2 D , which shows D is dense in QS . Hence, there is c 2 D \ G, so that

there is an element of C above . Thus, C is unbounded in . Next, see that C is closed.

0 0
Suppose \ C is unbounded in < . Since C is unbounded, there is 2 C, where > ,

0
and a condition c 2 C which contains . Since the conditions are ordered by end-extension,

and since there is a sequence (possibly of length 1) of conditions which witness that \ C is

unbounded in C, which will all be contained in the condition c, which contains an element

above , it follows that \ c is unbounded in . Since c is closed, 2 c. Therefore 2 C,

which shows C is closed.

Furthermore, QS preserves cardinals and cofinalities + since |QS | = < = . If

< , then QS is not necessarily  -closed. Let be the first limit of S which is not

in S. Then, the initial segments of S below are in QS , and union up to S \ , so that

no condition can close this -sequence since 2

/ S. However, for every < , the set
 47

D = {d 2 QS : max(d) } is dense in QS and is  -closed since cof() > . Thus,

for every < , the poset QS is forcing equivalent to C , which is  -closed. Thus, QS

preserves all cardinals, cofinalities, and strong limits  .

The following Lemma is about the Mitchell rank of a normal measure µ on a measurable

cardinal . By definition, o() is the height of the well-founded Mitchell relation C on m(),

the collection of normal measures on . Thus o() = {o(µ) + 1 | µ 2 m()}. Similarly the

definition of rank for µ 2 m() with respect to C is o(µ) = sup{o(⌫) + 1 |⌫ C µ}.

Lemma 28. If  is a measurable cardinal, and µ is a normal measure on , and jµ : V ! Mµ

is the ultrapower embedding by µ with critical point , then o(µ) = o()Mµ . Thus, < o()

if and only if there exists j : V ! M elementary embedding with critical point  with

o()M = .

Proof. Suppose  is measurable, µ is a normal ulltrafilter on , and jµ : V ! Mµ is the

ultrapower embedding by µ with critical point . The model Mµ computes C correctly for

⌫ C µ, which means ⌫ 2 Mµ , since M  ✓ M and so Mµ has all the functions f :  ! V

48

needed to check whether ⌫ C µ. Then

o()Mµ = sup{o(⌫)M⌫ + 1 | ⌫ 2 Mµ }

= sup{o(⌫)Mµ + 1 | ⌫ C µ}

= sup{o(⌫) + 1 | ⌫ C µ}

= o(µ)

Thus, the Mitchell rank of  in Mµ is the Mitchell rank of µ. Let < o(). Since C is

well-founded, there is µ a normal measure on  such that o(µ) = . Then if jµ : V ! Mµ is

the ultrapower by µ with critical point , then o()Mµ = o(µ) = , as desired. Conversely,

suppose there is an elementary embedding j : V ! M with cp(j) =  and o()M = . Let

µ be the induced normal measure, using  as a seed (X 2 µ if and only if  2 j(X)) and let

Mµ be the ultrapower by µ. Then M and Mµ have the same normal measures and the same

Mitchell order and ranks, so Mµ |= o() = . Since we have established o()Mµ = o(µ), the

Mitchell rank of µ in m() is , that is o(µ) = . Thus since o() = {o(µ) + 1|µ 2 m()},

it follows that o()V > .

The following lemma generalizes Corollary 22 of [Hamkins4] by showing that if V ✓ V̄

satisfies the approximation and cover properties, then the Mitchell rank of a cardinal  >

cannot increase between V and V̄ .

 49

Lemma 29. Suppose V ✓ V̄ satisfies the approximation and cover properties. If  > ,

then o()V̄  o()V .

Proof. Suppose V ✓ V̄ satisfies the approximation and cover properties. Suppose  > .

We want to show o()V̄  o()V by induction on . Assume inductively that for all 0 < ,

we have o(0 )V̄  o(0 )V . If o()V̄ > o()V then by Lemma ?? get and elementary ultrapower

embedding j : V̄ ! M̄ with critical point  and o()M̄ = o()V . By [Hamkins4], j is a lift of

an ultrapower embedding j V : V ! M for some M , and this embedding is a class in V .

Again by ?? we have established that o()M < o()V . Thus, o()M < o()M̄ . However, this

contradicts that j applied to the induction hypothesis gives o()M̄  o()M since  < j().

Therefore, o()V̄  o()V . ⇤

Lemma 30. If V ✓ V [G] is a forcing extension and every normal ultrapower embedding

j : V ! M in V (with any critical point ) lifts to an embedding j : V [G] ! M [j(G)], then

o()V [G] o()V . In addition, if V ✓ V [G] also has approximation and cover properties,

then o()V [G] = o()V .

Proof. Suppose V ✓ V [G] is a forcing extension where every normal ultrapower embedding

in V lifts to a normal ultrapower embedding in V [G]. Suppose  is a measurable cardinal,

and for every measurable cardinal 0 < , assume o(0 )V [G] o(0 )V . Fix any < o().

By Lemma ?? there exists an elementary embedding j : V ! M with cp(j) =  and

M |= o() = . Then, by the assumption on V ✓ V [G], this j lifts to j : V [G] ! M [j(G)].

50

Then, applying j to the induction hypothesis gives o()M [j(G)] o()M since  < j().

Then, since M |= o() = it follows that o()M [j(G)] . Thus by Lemma ?? o()V [G] > .

Thus o()V [G] o()V .

If V ✓ V [G] also satisfies the approximation and cover properties, then by Lemma ??

the Mitchell rank does not go up between these models. Thus o()V [G] = o()V . ⇤

Theorem 31. For any V |= ZF C + GCH and any ordinal ↵, there is a forcing extension

V [G] where every cardinal  above ↵ in V [G] has o()V [G] = min{↵, o()V }. In particular, if

1  ↵, then the forcing preserves all measurable cardinals of V and creates no new measurable

cardinals.

Proof. Let ↵ 2 Ord. The following notion of forcing will ensure that the Mitchell rank of all

cardinals above ↵ which have Mitchell rank above ↵ will change. Let P be an Easton support

Ord-length iteration, forcing at inaccessible stages , to add a club c ✓ such that 2c

implies o( )V < ↵. Let G ✓ P be V -generic. Let 0 denote the first inaccessible cardinal.

The forcing up to this stage, P 0 , adds nothing and the forcing Q at stage 0 adds a club

to 0, so that |P 0 ⇤ Q̇|  0. The forcing past stage 0 is closed up to the next inaccessible

cardinal. Since P has a closure point at 0, by Lemma 13 in [Hamkins4] it follows that

+
V ✓ V [G] satisfies the 0 approximation and cover properties. Thus, by Lemma ?? the

Mitchell rank does not go up between V and V [G]. Note that any measurable cardinal is

+
above 0 .
 51

Let  be a measurable cardinal above ↵. Fix any 0 < min{↵, o()V }. The induction

hypothesis is that for all 0 < , we have o(0 )V [G] = min{↵, o(0 )V }. By Lemma ??, fix an

elementary embedding j : V ! M with cp(j) =  and o()M = 0. We shall lift j through

P. The forcing above  does not a↵ect o() since the forcing past stage  has a dense subset

which is closed up to the next inaccessible. Allow me to call P ⇤ Q̇ the forcing up to and

including stage  while I am lifting the embedding, first through P and then through Q.

Let G ✓ P be V -generic, and g ✓ Q be V [G ]-generic. First, we need an M -generic filter

for j(P ). Since cp(j) = , the forcings P and j(P ) agree up to stage . Since ↵ < ,

where ↵ is the ordinal in the statement of the theorem, and the forcing Q adds a club c ✓ 

such that 2 c implies o( )V < ↵ which implies o( )M < ↵, it follows that the th stage

of j(P ) is Q. Thus, j(P ) factors as P ⇤ Q̇ ⇤ Ptail where Ptail is the forcing past stage .

Since G ⇤ g is generic for P ⇤ Q̇, we just need an M [G ][g]-generic filter for Ptail which

is in V [G ][g] (since we want the final embedding to be there). Thus, diagonalize to get

an M [G ][g]-generic filter for Ptail after checking we have met the criteria. Namely, since

|P | = |Q̇| = , the forcing P ⇤ Q̇ has the + -chain condition. Since M  ✓ M , it follows

[Hamkins1 (Theorem 54)] that M [G ][g] ✓ M [G ][g]. Also, since 2 = + , the forcing P


has + many dense sets in V . Thus, Ptail has at most |j(+ )|V  + = + many dense sets

in M [G ][g]. Also, since for any < , the forcing P has a dense subset which is  -closed

(by Lemma ??), it follows that there is a dense subset of Ptail which is -closed. Thus, we

may diagonalize to get G⇤ ⇢ Ptail for this dense subset, a generic filter in V [G ][g], which is
52

M [G ][g]-generic. Thus, j(G ) = G ⇤ g ⇤ G⇤ is M -generic for j(P ) and j 00 G ✓ G ⇤ g ⇤ G⇤ ,

so we may lift j to j : V [G ] ! M [j(G )].

Next, we lift j through Q. The forcing j(Q) adds a club C ✓ j() such that 2 C

implies o( )M [j(G )] < ↵. Since G ⇤ g ⇤ G⇤ 2 V [G ][g] and M  ✓ M , it follows [Hamkins1

(Theorem 53)] that M [j(G )] ✓ M [j(G )]. Also, by Lemma ??, there is a dense subset of

j(Q) which is -closed, and since |Q| = , it follows that j(Q) has at most + many dense

subsets in M [j(G )]. Let c = [g, which is club in , and consider the set c̄ = c [ {} 2

M [j(G )]. We have 2 c implies o( )V < ↵. Since M and V agree on P (P ( )) for < ,

it follows that o( )V = o( )M . By Lemma ??, we have o( )M [j(G )]  o( )M . Thus 2c

implies o( )M [j(G )] < ↵. Since o()M < ↵, it follows that o()M [j(G )] < ↵, so that c̄ is

a condition of j(Q), since c̄ is a closed, bounded subset of j() which such that 2 c̄

implies o( )M [j(G )] < ↵. Thus, diagonalize to get an M [j(G)]-generic filter, g ⇤ ✓ j(Q) which

contains the condition c̄. Then j 00 g ✓ j(g) = g ⇤ . Therefore, we may lift the embedding to

j : V [G ][g] ! M [j(G )][j(g)].

From the previous lifting arguments will follow that o()V [G] min{o()V , ↵}. The lifted

embedding j : V ! M was chosen so that o()M = 0 < min{o()V , ↵}. Since 0 was

arbitrary, it follows from Lemma ?? that o()V min{o()V , ↵}. All that remains to see

is o()V [G]  min{o()V , ↵}. By Lemma ??, which shows Mitchell rank does not go up for

V ✓ V [G], we have o()V [G]  o()V . Consider any embedding j : V [G] ! M [j(G)] with

critical point  in V [G], and the new club c ✓ , where 8 < , 2 c implies o( )V < ↵, which
 53

is in any normal measure on . By the induction hypothesis, 2 c implies o( )V [G] < ↵.

Applying j to this statement gives, 8 < j(), 2 j(c) implies o( )M [j(G)] < ↵. Since

 < j() and  2 j(c), this gives o()M [j(G)] < ↵. Thus by Lemma ??, since j was arbitrary,

o()V [G]  ↵. Thus o()V [G]  min{o()V , ↵}. Thus, o()V [G] = min{o()V , ↵}. ⇤

I would like to now consider cases where o() . For this we use the concept of

representing function. Suppose  is measurable, o() = ↵, and ↵ 2 H✓+ . The ordinal ↵ is

represented by f :  ! V if whenever j : V ! M an elementary embedding with cp(j) = 

and M  ✓ M , then j(f )() = ↵ (this can be formalized as a first-order statement using

extenders) [Hamkins2]. Let F : Ord ! Ord. The main theorem shows how to change the

Mitchell rank for all measurable cardinals for which F represents F ( ) (through any

-closed embedding with critical point ).

Theorem 32. For any V |= ZF C + GCH and any F : Ord ! Ord, there is a forcing

extension V [G] where, if  is measurable and F  represents F () in V, then o()V [G] =

min{o()V , F ()}.

Proof. For example, we can do this for numerous functions satisfying this hypothesis, such

2
as F ( ) = , F ( ) = + 3, F ( ) = + ! 2 · 2 + 5, and any other function which can be

defined like this. Suppose V |= ZFC + GCH. Let F : Ord ! Ord. Let P be an Easton

support Ord-length iteration, forcing at inaccessible stages , to add a club C ✓ such

that 2 C implies o( )V < F ( ). Let 0 denote the first inaccessible cardinal. The forcing
54

up to this stage, P 0 , adds nothing and the forcing Q at stage 0 adds a club to 0, so that

|P 0 ⇤ Q̇|  0. The forcing past stage 0 is closed up to the next inaccessible cardinal. Let

G ✓ P be V -generic. Since P has a closure point at 0, by Lemma 13 in [Hamkins4] it follows

+
that V ✓ V [G] satisfies the 0 approximation and cover properties. Thus, by Lemma ?? the

Mitchell rank does not go up between V and V [G].

Let  be a measurable cardinal such that F  represents F () = ↵ in V . Here we will

need that representing functions are still representing functions in the extension. The fact

we will use for this is from [Hamkins4] which says that if V ✓ V [G] satisfies the hypothesis

of approximation and cover theorem, then every -closed embedding j : V [G] ! M [j(G)] is

a lift of an embedding j V : V ! M that is in V (and j V is also a -closed embedding.

Thus, if F  is a representing function for F () in the ground model, (j V )(F )() =

F () = ↵, and thus F  is a representing function for F () with respect to such embeddings

j in V [G].

The induction hypothesis is that for all 0 <  for which F 0 represents F (0 ) in V has

o(0 )V [G] = min{F (0 ), o(0 )V }. By Lemma ?? fix an elementary embedding j : V ! M with

cp(j) =  and o()M = 0 < min{o()V , F ()}. Note that j(F )() = j(F )() = F ().

We shall lift j through P. The forcing above  does not a↵ect o() since the forcing past stage

 is closed up to the next inaccessible. Allow me to call P ⇤ Q̇ the forcing up to and including

stage  while I am lifting the embedding, first through P and then through Q. Let G ✓ P

be V -generic, and g ✓ Q be V [G ]-generic. First, we need an M -generic filter for j(P ).
 55

Since cp(j) = , the forcings P and j(P ) agree up to stage . The forcing Q adds a club

C ✓  such that 2 C implies o( )V < F ( ). Since j(F )() = F () and o()M < F (),

the th stage of j(P ) is Q. Thus, j(P ) factors as P ⇤ Q̇ ⇤ Ptail where Ptail is the forcing

past stage . Since G ⇤ g is generic for P ⇤ Q̇, we just need an M [G ][g]-generic filter for

Ptail which is in V [G ][g] (since we want the final embedding to be there). Thus, we shall

diagonalize to get an M [G ][g]-generic filter for Ptail after checking we have met the criteria.

Namely, since |P | = |Q| = , the forcing P ⇤ Q̇ has the + -chain condition. Since M  ✓ M ,

it follows [Hamkins1 (Theorem 54)] that M [G ][g] ✓ M [G ][g]. Also, since 2 = + , the


forcing P has + many dense sets in V . Thus, Ptail has at most |j(+ )|V  + = +

many dense sets in M [G ][g]. Also, since for any < , the forcing P has a dense subset

which is  -closed (by Lemma ??), it follows that there is a dense subset of Ptail which is

-closed. Thus, we may diagonalize to get G⇤ ✓ Ptail for this dense subset, a generic filter

in V [G ][g], which is M [G ][g]-generic. Thus, j(G ) = G ⇤ g ⇤ G⇤ is M -generic for j(P )

and j 00 G ✓ G ⇤ g ⇤ G⇤ , so we may lift j to j : V [G ] ! M [j(G )].

Next, we lift j through Q. The forcing j(Q) adds a club C ✓ j() such that 2 C

implies o( )M [j(G )] < j(F )( ). Since j(G ) 2 V [G ][g] and M  ✓ M , it follows [Hamkins1

(Theorem 53)] that M [j(G )] ✓ M [j(G )]. Also, by Lemma ??, there is a dense subset of

j(Q) which is -closed, and since |Q| = , it follows that j(Q) has at most + many dense

subsets in M [j(G )]. Let c = [g and consider the set c̄ = c [ {}. By the construction of

the forcing, 8 < , 2 c implies o( )V < F ( ). Thus, 8 < , 2 c implies o( )M < F ( ).

By Lemma ?? it follows that 2 c implies o( )M [j(G)] < F ( ). Since o()M [j(G)]  o()M =
56

0 < min{o()V , F () = ↵} it follows that o()M [j(G)] < ↵ = j(F )(). Thus c̄ is a closed,

bounded subset of j() such that if 2 c̄ then o( )M [j(G)] < j(F )( ). Thus c̄ is a condition

of j(Q). Thus, diagonalize to get an M [j(G)]-generic filter, g ⇤ ✓ j(Q) which contains the

condition c̄, so that j 00 g ✓ j(g) = g ⇤ . Therefore, we may lift the embedding to j : V [G ][g] !

M [j(G )][j(g)].

From the previous lifting arguments will follow that o()V [G] min{o()V , F ()}. The

lifted embedding j : V ! M was chosen so that o()M = 0 < min{o()V , F ()}. Since

0 was arbitrary, by Lemma ?? it follows that o()V [G] min{o()V , F ()}. All that is

left is to see that o()V [G]  min{o()V , F ()}. By Lemma ?? which shows that Mitchell

rank does not go up for V ✓ V [G], we have o()V [G]  o()V . Consider any embedding

j : V [G] ! M [j(G)] in V [G], and the new club c ✓  where 8 < , 2 c implies

o( )V < F ( ). By the induction hypothesis, 2 c implies o( )V [G] < F ( ). Apply j to this

statement to get 8 < j(), 2 j(c) implies o( )M [j(G)] < j(F )() = F (). Note that since

F  represents F () in V [G] and M [j(G)] ✓ M [j(G)], it follows that F  represents

F () in M [j(G)] as well. Since j was arbitrary,  < j(), and  2 j(c) it now follows

by Lemma ?? it follows that o()V [G]  F (). Thus o()V [G]  min{o()V , F ()}. Thus

o()V [G] = min{o()V , F ()}.

⇤
 57

6. Supercompact and strongly compact cardinals.

This section begins with supercompact and strongly compact cardinals. The main theorem

is about forcing to change degrees of supercompact cardinals. An uncountable cardinal  is ✓-

supercompact if and only if there is an elementary embedding j : V ! M , with critical point

 and M ✓ ✓ M , where  < ✓ < j(). Equivalently,  is ✓-supercompact if and only if there is

a normal fine measure on P ✓. A fine measure in the previous sentence means a -complete

ultrafilter such that 8 2 P ✓ the set {X ✓ P ✓ | 2 X} is in the filter. A cardinal  is

supercompact if and only if it is ✓-supercompact for every ✓ > . Notice that  measurable

means that  is -supercompact. An uncountable cardinal  is ✓-strongly compact if and

only if there is an elementary embedding j : V ! M , with critical point , such that for all

t ✓ M , where |t|V  ✓, there is an s 2 M such that t ✓ s and |s|M < j(). Equivalently, a

cardinal  is ✓-strongly compact for  < ✓ if and only if there is a -complete fine measure on

P ✓. An uncountable cardinal  is strongly compact if and only if it is ✓-strongly compact,

for every ✓ > . Equivalently, an uncountable regular cardinal  is strongly compact if

and only if for any set S, every -complete filter on S can be extended to a -complete

ultrafilter on S [Jech p. 365]. The following theorems show how to distinguish, by forcing,

between levels of supercompactness, levels of strong compactness, between measurable and

supercompact cardinals, between measurable and strongly compact cardinals, and between

strongly compact and supercompact cardinals.

58
<
Theorem 33. If  is < ✓-supercompact, for ✓  regular with < ✓ for all < ✓, then

there is a forcing extension where  is < ✓-supercompact, but not ✓-supercompact, and indeed

not even ✓-strongly compact.

Proof. Suppose  is < ✓-supercompact for some regular ✓ > , and let P = Add(!, 1) ⇤

Add(✓, 1). By [Hamkins/Shelah], after forcing with P, the cardinal  is not ✓-supercompact.

However, adding a Cohen real is small relative to , so the first factor of P preserves that 

is < ✓-supercompact. The second factor of P is < ✓-closed, so that a normal fine measure

on P , where < ✓ is still a normal fine measure on P , since no new subsets of were

added. Let G ✓ P be V -generic. Thus, in V [G], the cardinal  is < ✓-supercompact, but

not ✓-supercompact.

Corollary 34. If  is ✓-supercompact where  < ✓ and ✓< = ✓, then there is a forcing

extension where  is ✓-supercompact, but not ✓+ -supercompact.

<
Theorem 35. If  is <✓-strongly compact, for ✓  regular with < ✓ for all < ✓,

then there is a forcing extension where  is <✓-strongly compact, but not ✓-strongly compact.

Proof. Suppose  is < ✓-strongly compact for regular ✓ , and let P = Add(!, 1)⇤Add(✓, 1).

By [Hamkins/Shelah], this forcing destroys the ✓-strong compactness of . However, adding

a Cohen real is small relative to , so the first factor preserves that  is < ✓-strongly compact.

Also, since the second factor is < ✓-closed, any -complete fine measure on P , where < ✓,
 59

is still a -complete fine measure since there are no new subsets of to measure. Thus, P

preserves that  is < ✓-strongly compact. Let G ✓ P be V -generic. Then, in V [G], the

cardinal  is < ✓-strongly compact, but not ✓-strongly compact. ⇤

Corollary 36. If  is ✓-strongly compact, where   ✓ and ✓< = ✓, then there is a forcing

extension where  is ✓-strongly compact, but not ✓+ -strongly compact.

Corollary 37. If  is measurable, then there is a forcing extension where  is measurable,

but not strongly compact.

Proof. Let ✓ =  in the proof of Theorem ?? ⇤

The following theorem is due to Magidor [Magidor]. It shows how to force between strong

compactness and supercompactness.

Theorem 38. (Magidor) If  is strongly compact, then there is a forcing extension where 

is strongly compact, but not supercompact.

A word on the proof of the theorem. Suppose  is a strongly compact cardinal. Then,

there is a forcing extension, V [G], where  is the least strongly compact cardinal, and the

least measurable cardinal [Magidor]. However, the least measurable cardinal can never be

supercompact, since there are many measurable cardinals below any supercompact cardinal.

Thus, in V [G], the cardinal  is strongly compact, but not supercompact.

60

As for measurable cardinals, an analgous rank to Mitchell rank can be assigned to super-

compactness embeddings. Suppose a cardinal  is ✓-supercompact for fixed ✓. If µ and ⌫ are

normal fine measures on P ✓, define the Mitchell relation µC✓-sc ⌫ if and only if µ 2 M⌫ where

j : V ! M⌫ is an ultrapower embedding by ⌫. Since the relation C✓-sc is well-founded for

given  and ✓, the Mitchell rank of µ is its rank with respect to C✓-sc . Thus, for fixed  and

✓, the notation osc () = ↵ means the height of C✓-sc on normal fine measures on P ✓ is ↵. By

definition, o✓-sc () is the height of the well-founded Mitchell relation C✓-sc on m(), the col-

lection of normal fine measures on P ✓. Thus the definition of rank for µ 2 m() with respect

to C✓-sc is o✓-sc (µ) = sup{o✓-sc (⌫) + 1 |⌫ C✓-sc µ}. Thus o✓-sc () = {o✓-sc (µ) + 1 | µ 2 m()}.

So that o✓-sc () = 0 means that  is not ✓-supercompact, and o✓-sc () > 0 means that  is

✓-supercompact.

The main theorems of this section will rely on the following Lemmas about Mitchell rank

for supercompactness and are analogues of Lemmas ??, ??, and ?? for Mitchell rank.

Lemma 39. Proof. Suppose  is ✓-supercompact, µ is a normal fine measure on P ✓, and

jµ : V ! Mµ is the ultrapower emebedding by µ with critical point . The model Mµ

computes C✓-sc correctly for ⌫ C✓-sc µ, which means ⌫ 2 Mµ , since M ✓ ✓ M , and so Mµ has
 61

all the functions needed to check whether ⌫ C✓-sc µ. Then

o✓-sc ()Mµ = sup{o✓-sc (⌫)M⌫ + 1 | ⌫ 2 Mµ }

= sup{o✓-sc (⌫)Mµ + 1 | ⌫ C✓-sc µ}

= sup{o✓-sc (⌫) + 1 | ⌫ C✓-sc µ}

= o✓-sc (µ)

Thus, the Mitchell rank for supercompactness of  in Mµ is the Mitchell rank of µ. Let

< o✓-sc (). Since C✓-sc is well-founded, there is µ a normal fine measure on P ✓ such

that o✓-sc (µ) = . Then if jµ : V ! Mµ is the ultrapower by µ with critical point , then

o✓-sc ()Mµ = o✓-sc (µ) = . Conversely, suppose there is an elementary embedding j : V ! M

with cp(j) =  and o✓-sc ()M = . Let µ be the induced normal fine measure, using j 00 ✓

as a seed (X 2 µ if and only if j 00 ✓ 2 j(X)) and let Mµ be the ultrapower by µ. Then

M and Mµ have the same normal fine measures and the same Mitchell order and ranks for

✓-supercompactness, so Mµ |= o✓-sc () = . Since we have established o✓-sc ()Mµ = o✓-sc (µ),

the Mitchell rank for supercompactness of µ in m() is , that is o✓-sc (µ) = . Thus since

o✓-sc () = {o✓-sc (µ) + 1 | µ 2 m()}, it follows that o✓-sc ()V > .

The following Lemma is an analogue of Lemma ?? for degrees of Mitchell rank for super-

compactness. It is a generalization of Corollary 26 in [Hamkins4] which states that if V ✓ V̄

62

satisfies the approximation and cover properties, then no supercompact cardinals are created

from V to V̄ .

Lemma 40. Suppose V ✓ V̄ satisfies the approximation and cover properties. If , ✓ > ,

then o✓-sc ()V̄  o✓-sc ()V .

Proof. Suppose V ✓ V̄ satisfies the approximation and cover properties. Suppose , ✓ > .

We want to show o✓-sc ()V̄  o✓-sc ()V by induction on . Assume inductively that for all

0 < , we have o✓-sc (0 )V̄  o✓-sc (0 )V . If o✓-sc ()V̄ > o✓-sc ()V then by Lemma ?? get a ✓-

supercompactness embedding j : V̄ ! M̄ with critical point  and o✓-sc ()M̄ = o✓-sc ()V . By

[Hamkins4], j is a lift of a ✓-supercompactness embedding j V : V ! M for some M , and

this embedding is a class in V . Again by ?? we have established that o✓-sc ()M < o✓-sc ()V .

Thus, o✓-sc ()M < o✓-sc ()M̄ . However, this contradicts that j applied to the induction

hypothesis gives o✓-sc ()M̄  o✓-sc ()M since  < j(). Therefore, o✓-sc ()V̄  o✓-sc ()V . ⇤

Lemma 41. If V ✓ V [G] is a forcing extension and every ✓-supercompactness embedding

j : V ! M in V (with any critical point ) lifts to an embedding j : V [G] ! M [j(G)],

then o✓-sc ()V [G] o✓-sc ()V . In addition, if V ✓ V [G] also has approximation and cover

properties, then o✓-sc ()V [G] = o✓-sc ()V .

Proof. Suppose V ✓ V [G] is a forcing extension where every normal ultrapower embedding in

V lifts to a normal ultrapower embedding in V [G]. Suppose  is a ✓-supercompact cardinal,

and for every ✓-supercompact cardinal 0 < , assume o✓-sc (0 )V [G] o✓-sc (0 )V . Fix any <
 63

o✓-sc (). By Lemma ?? there exists an elementary embedding j : V ! M with cp(j) =  and

M |= o✓-sc () = . Then, by the assumption on V ✓ V [G], this j lifts to j : V [G] ! M [j(G)].

Then, applying j to the induction hypothesis gives o✓-sc ()M [j(G)] o✓-sc ()M since  < j().

Then, since M |= o✓-sc () = it follows that o✓-sc ()M [j(G)] . Thus by Lemma ?? we have

o✓-sc ()V [G] > . Thus o✓-sc ()V [G] o✓-sc ()V .

If V ✓ V [G] also satisfies the approximation and cover properties, then by Lemma ??

the Mitchell rank does not go up between these models. Thus ✓-sc o()
V [G]
= o✓-sc ()V . ⇤

The following theorem shows how to force a + -supercompact cardinal  to have Mitchell

rank for supercompactness at most 1.

Theorem 42. If V |= ZF C + GCH then there is a forcing extension where every cardinal

 which is + -supercompact has o+ -sc ()V [G] = min{o+ -sc ()V , 1}.

Proof. Suppose V |= ZF C + GCH. Let P be an Easton support Ord-length iteration which

forces at inaccessible stages to add a club c ✓ such that 2 c implies o + -sc ( )V = 0.

Let G ✓ P be V -generic. Let 0 be the first inaccessible cardinal. The forcing before stage

0 is trivial, and the forcing at stage 0 adds a club to 0. Thus the forcing P up to and

including stage 0 has cardinality 0, and the forcing past stage 0 has a dense set which is

closed up to the next inaccessible cardinal. Thus P has a closure point at 0 which implies

+
V ✓ V [G] satisfies the 0 approximation and cover properties by [Hamkins4 (Lemma 13)].

Thus by [Hamkins4 (Corollary 26)], the forcing P does not create ✓-supercompact cardinals
64

for any ✓. Note that any ✓-supercompact cardinal , both ✓ and  are above this closure

point.

Suppose  is + -supercompact. The induction hypothesis is that any 0 <  has o0+ -sc (0 )V [G] =

min{o(0 )V , 1}. Pick j : V ! M a + -supercompactness embedding with critical point 

such that o+ -sc ()M = 0 < min{o+ -sc ()V , 1}. We shall lift j through P. The forcing above

 does not a↵ect the Mitchell rank for + -supercompactness of  since the forcing past stage

 has a dense subset which is closed up to the next inaccessible cardinal past . Call P ⇤ Q̇

the forcing up to and including stage . First we shall lift j through P . Let G ✓ P be

V -generic and g ✓ Q be V [G ]-generic. First, we need an M -generic filter for j(P ). Since

cp(j) = , the forcings P and j(P ) agree up to stage . The forcing Q adds a club c ✓ 

such that 8 < , 2 c implies o + -sc ( )V = 0. Since o + -sc ( )V = 0 implies o + -sc ( )M = 0, it

follows that the th stage of j(P ) is Q. Thus, j(P ) factors as P ⇤ Q̇ ⇤ Ptail where Ptail is the

forcing past stage . Since G ⇤ g is M -generic for P ⇤ Q̇, we only need an M [G ][g]-generic

filter for Ptail which is V [G ][g]. We will diagonalize to obtain such a filter, after checking

that we have met the criteria. Since |P | = |Q| = , and V |= GCH, it follows that the

+
forcing P ⇤ Q̇ has the + -chain condition. Since M  ✓ M , it follows [Hamkins1 (Theorem

+
54)] that M [G ][g] ✓ M [G ][g]. Since |P | = , and has the + -chain condition, it has at
<
+ +
most < = + many maximal antichains. Since |j(+ )|  + = ++ , it follows that

Ptail has at most ++ many maximal antichains in M [G ][g]. Since Ptail has a dense subset

which is closed up to the next inaccessible past , it has a dense subset which is + -closed.

Thus, we can enumerate the maximal antichains of Ptail , and build a descending sequence of
 65

conditions meeting them all through the dense set which has + -closure, using this closure

and the fact that M [G ][g] is closed under + -sequences to build through the limit stages.

Then the upward closure of this sequence Gtail ✓ Ptail is M [G ][g]-generic and is in V [G ][g].

Thus j(G ) = G ⇤ g ⇤ Gtail is M -generic, it is in V [G ][g] and j 00 G ✓ j(G ). Thus we have

the partial lift j : V [G ] ! M [j(G )].

Next we lift through Q by finding an M [j(G )]-generic filter for j(Q). The forcing j(Q)

adds a club C ✓ j() such that 8 < j(), 2 C implies o + -sc ( )M [j(G )] = 0. Since

+ +
j(G ) 2 V [G ][g] and M  ✓ M , it follows that M [j(G )] ✓ M [j(G )] by [Hamkins1

(Theorem 53)]. Since for every < , there is a dense subset of Q which is  -closed, it

follows that there is a dense subset of j(Q) which is + -closed. Also since |Q| =  and Q has

the + -chain condition, the forcing Q has at most + many maximal antichains. Therefore
<
+
j(Q) has at most |j(+ )|  + = ++ many maximal antichains in M [j(G )]. Thus we

can diagonalize to get a generic filter, but we will do so with a master condition as follows.

Let c = [g the new club of  and consider c̄ = c [ {} which is in M [j(G )]. We have 2c

implies o + -sc ( )V = 0. Thus, 2 c implies o + -sc ( )M = 0. By Corollary 26 [Hamkins4], no

+
new -supercompact cardinals are created between M and M [j(G )], so that 2 c implies

o + -sc ( )M [j(G )] = 0. Since o+ -sc ()M = 0, it follows that o+ -sc ()M [j(G )] = 0. Thus c̄

is a closed, bounded subset of j() such that every in c̄ has o + -sc ( )M [j(G )] = 0. Thus

c̄ 2 j(Q). Thus diagonalize to get an M [j(G )]-generic filter g ⇤ ✓ j(Q) which contains

the master condition c̄. Then j 00 g ✓ j(g) = g ⇤ . Therefore we may lift the embedding to

j : V [G ][g] ! M [j(G )][j(g)].

66

The success of the lifting arguments show that o+ -sc ()V [G] min{o+ -sc ()V , 1}. By

Lemma ?? we have o+ -sc ()V [G]  o+ -sc ()V . All that remains is to see that o+ -sc ()V [G] 

1. Consider any + -supercompact embedding j : V [G] ! M [j(G)] with critical point 

in V [G], and the new club c ✓ , where 8 < , 2 c implies o + -sc ( )V = 0. This club

is in any normal fine measure on P + . By the induction hypothesis, 8 < , 2 c im-

plies o + -sc ( )V [G] = 0. Applying j to this statement gives 8 < j(), 2 j(c) implies

o + -sc ( )M [j(G)] = 0. Thus by Lemma ??, since j was arbitrary,  < j(), and  2 j(c)

it follows that o+ -sc ()M [j(G)] = 0, and hence o+ -sc ()V [G]  1. Thus o+ -sc ()V [G] 

min{o+ -sc ()V , 1}. Thus for any  which is + -supercompact, o+ -sc ()V [G] = min{o+ -sc ()V , 1}.

For the following theorem for changing Mitchell rank for supercompactness, we will again

need representing functions because I would like to now consider cases where o+ -sc ()

. Then for the most general theorem I would like to consider the cases where  is ✓-

supercompact for ✓ > + . In [Hamkins2] representing functions are considered for super-

compactness embeddings as well. Let ↵ 2 H✓+ and  be a ✓-supercompact cardinal. The

ordinal ↵ is represented by f :  ! V if whenever j : V ! M an elementary embedding

with cp(j) =  and M ✓ ✓ M , then j(f )() = ↵ (this can be formalized as a first-order

statement using extenders) [Hamkins2]. Let F : Ord ! Ord. The next theorem shows how

+
to change the Mitchell rank for supercompactness for all -supercompact cardinals for

which F represents F ( ).
 67

Theorem 43. For any V |= ZF C + GCH and any F : Ord ! Ord, there is a forc-

ing extension V [G] where, if  is + -supercompact and F  represents F () in V, then

o+ -sc ()V [G] = min{o+ -sc ()V , F ()}.

Proof. Suppose V |= ZF C + GCH. Let F : Ord ! Ord. Let P be an Easton support

Ord-length iteration which forces at inaccessible stages to add a club c ✓ such that

8 < , 2 c implies o + -sc ( )V < F ( ). Let G ✓ P be V -generic. Let 0 be the first

inaccessible cardinal. The forcing before stage 0 is trivial, and the forcing at stage 0 adds a

club to 0. Thus the forcing P up to and including stage 0 has cardinality 0 and the forcing

past stage 0 has a dense set which is closed up to the next inaccessible cardinal. Thus P has

+
a closure point at 0, thus V ✓ V [G] satisfies the 0 approximation and cover properties by

[Hamkins4 (Lemma 13)]. Thus by [Hamkins4 (Corollary 26)], the forcing P does not create

✓-supercompact cardinals for any ✓. Note that for any ✓-supercompact cardinal , both 

and ✓ are above this closure point.

Let  be a + -supercompact cardinal such that F  represents F () = ↵ in V . Here

we will need that representing functions are still representing functions in the extension.

The fact we will use for this is from [Hamkins4] which says that if V ✓ V [G] satisfies the

hypothesis of approximation and cover theorem, then every ✓-closed embedding j : V [G] !

M [j(G)] is a lift of an embedding j V : V ! M that is in V (and j V is also a ✓-

closed embedding). Thus, if F  is a representing function for F () in the ground model,
68

(j V )(F )() = F (), and thus F  is a representing function for F () with respect

to such embeddings j in V [G].

The induction hypothesis is that any 0 < , where F 0 represents F (0 ) in V , has

o0+ -sc (0 )V [G] = min{o0+ -sc (0 )V , F (0 )}. Pick j : V ! M a + -supercompactness embed-

ding with critical point  such that o+ -sc ()M = 0 < min{o+ -sc ()V , F ()}. We shall lift j

through P. The forcing above  does not a↵ect the Mitchell rank for + -supercompactness

of  since the forcing past stage  has a dense subset which is closed up to the next inac-

cessible cardinal past . Call P ⇤ Q̇ the forcing up to and including stage . First we shall

lift j through P . Let G ✓ P be V -generic and g ✓ Q be V [G ]-generic. First, we need

an M -generic filter for j(P ). Since cp(j) = , the forcings P and j(P ) agree up to stage

. The forcing Q adds a club c ✓  such that 8 < , 2 c implies o + -sc ( )V < F ( ).

Since o + -sc ( )V < F ( ) implies o + -sc ( )M < F ( ) = j(F )( ), it follows that the th stage

of j(P ) is Q. Thus, j(P ) factors as P ⇤ Q̇ ⇤ Ptail where Ptail is the forcing past stage

. Since G ⇤ g is M -generic for P ⇤ Q̇, we only need an M [G ][g]-generic filter for Ptail

which is V [G ][g]. We will diagonalize to obtain such a filter, after checking that we have

met the criteria. Since |P | = |Q| = , and V |= GCH, it follows that the forcing P ⇤ Q̇

+
has the + -chain condition. Since M  ✓ M , it follows [Hamkins1 (Theorem 54)] that

+
M [G ][g] ✓ M [G ][g]. Since |P | = , and has the + -chain condition, it has at most
<
+ +
< = + many maximal antichains. Since |j(+ )|  + = ++ , it follows that Ptail

has at most ++ many maximal antichains in M [G ][g]. Since Ptail has a dense subset which

is closed up to the next inaccessible past , it has a dense subset which is + -closed. Thus,
 69

we can enumerate the maximal antichains of Ptail , and build a descending sequence of condi-

tions meeting them all through the dense set which has + -closure, using this closure and

the fact that M [G ][g] is closed under + -sequences to build through the limit stages. Then

the upward closure of this sequence Gtail ✓ Ptail is M [G ][g]-generic and is in V [G ][g]. Thus

j(G ) = G ⇤ g ⇤ Gtail is M -generic, it is in V [G ][g] and j 00 G ✓ j(G ). Thus we have the

partial lift j : V [G ] ! M [j(G )].

Next we lift through Q by finding an M [j(G )]-generic filter for j(Q). The forcing j(Q)

adds a club C ✓ j() such that 8 < j(), 2 C implies o + -sc ( )M [j(G )] < j(F )( ). Since

+ +
j(G ) 2 V [G ][g] and M  ✓ M , it follows that M [j(G )] ✓ M [j(G )] by [Hamkins1

(Theorem 53)]. Since for every < , there is a dense subset of Q which is  -closed,

it follows that there is a dense subset of j(Q) which is + -closed. Also since |Q| = 

and Q has the + -chain condition, the forcing Q has at most + many maximal antichains.
<
+
Therefore j(Q) has at most |j(+ )|  + = ++ many maximal antichains in M [j(G )].

Thus we can diagonalize to get a generic filter, but we will do so with a master condition as

follows. Let c = [g the new club of  and consider c̄ = c [ {} which is in M [j(G )]. We

have 2 c implies o + -sc ( )V < F ( ). Thus, 2 c implies o + -sc ( )M < F ( ). By Lemma

?? the Mitchell rank for supercompactness does not go up between M and M [j(G )], so

that 2 c implies o + -sc ( )M [j(G )] < F ( ) = j(F )( ). Since o+ -sc ()M [j(G)]  o+ -sc ()M =

0 < min{o+ -sc ()V , F ()} it follows that o+ -sc ()M [j(G)] < F () = j(F )(). Thus c̄ is a

closed, bounded subset of j() such that every in c̄ has o + -sc ( )M [j(G )] < j(F )(). Thus

c̄ 2 j(Q). Thus diagonalize to get an M [j(G )]-generic filter g ⇤ ✓ j(Q) which contains
70

the master condition c̄. Then j 00 g ✓ j(g) = g ⇤ . Therefore we may lift the embedding to

j : V [G ][g] ! M [j(G )][j(g)].

The success of the lifting arguments for o+ -sc ()M = 0 < min{o+ -sc ()V , F ()}, where

0 was arbitrary, show that o+ -sc ()V [G] min{o+ -sc ()V , F ()}. By Lemma ?? we have

o+ -sc ()V [G]  o+ -sc ()V . All that remains is to see that o+ -sc ()V [G]  F (). Consider

any + -supercompact embedding j : V [G] ! M [j(G)] with critical point  in V [G], and the

new club c ✓ , where 8 < , 2 c implies o + -sc ( )V < F ( ). This club is in any normal

fine measure on P + . By the induction hypothesis, 8 , , 2 c implies o + -sc ( )V [G] < F ( ).

Applying j to this statement gives 8 < j(), 2 j(c) implies o + -sc ( )M [j(G)] < j(F )( ).

Since F  represents F (), we have j(F )() = F (). Since  < j() and  2 j(c) it follows

that o+ -sc ()M [j(G)] < F (). Thus by Lemma ??, since j was arbitrary, o+ -sc ()V [G]  F ().

Thus o+ -sc ()V [G]  min{o+ -sc ()V , F ()}. Thus for any  which is + -supercompact for

which F  represents F () in V has o+ -sc ()V [G] = min{o+ -sc ()V , F ()}. ⇤

The most general theorem considers the case of changing Mitchell rank for ✓-supercompactness

for a ✓-supercompact cardinal  when ✓ > + .

Theorem 44. For any V |= ZF C + GCH, any ⇥ : Ord ! Ord and any F : Ord ! Ord,

there is a forcing extension V [G] where, if  is ⇥()-supercompact, ⇥  represents ⇥(),

F  represents F () in V, and ⇥00  ✓ , then o⇥()-sc ()V [g] = min{o⇥()-sc ()V , F ()}.
 71

Proof. Suppose V |= ZF C + GCH. Let ⇥ : Ord ! Ord and F : Ord ! Ord. Let P be an

Easton support Ord-length iteration which forces at to add a club c ✓ such that 2c

implies o⇥( )-sc ( )V < F ( ), whenever is a closure point of ⇥, meaning ⇥00 ✓ , otherwise

force trivially. Let G ✓ P be V -generic. Let 0 be the first inaccessible cardinal. The forcing

before stage 0 is trivial, and the forcing at stage 0 adds a club to 0. Thus the forcing P

up to and including stage 0 has cardinality 0 and the forcing past stage 0 has a dense set

which is closed up to the next inaccessible cardinal. Thus P has a closure point at 0, thus

+
V ✓ V [G] satisfies the 0 approximation and cover properties by [Hamkins4 (Lemma 13)].

Thus by Lemma ??, the forcing P does not increase Mitchell rank for ✓-supercompactness

for cardinals above 0 where ✓ is also above 0.

Let  be a ⇥()-supercompact cardinal such that ⇥00  ✓ , and ⇥  represents ⇥(),

and F  represents F () in V . Here we will need that representing functions are still

representing functions in the extension. The fact we will use for this is from [Hamkins4] which

says that if V ✓ V [G] satisfies the hypothesis of approximation and cover theorem, then every

⇥()-closed embedding j : V [G] ! M [j(G)] is a lift of an embedding j V : V ! M that is

in V (and j V is also a ⇥()-closed embedding). Thus, if F  is a representing function

for F () in the ground model, (j V )(F )() = F (), and thus F  is a representing

function for F () with respect to such embeddings j in V [G].

The induction hypothesis is that any 0 < , where ⇥00 0 ✓ 0 , and ⇥ 0 represents ⇥(),

and F 0 represents F (0 ) in V , has o⇥()-sc (0 )V [G] = min{o⇥()-sc (0 )V , F (0 )}. Pick j :
72

V ! M a ⇥()-supercompactness embedding with critical point  such that o⇥()-sc ()M =

0 < min{o⇥()-sc ()V , F ()}. We shall lift j through P. The forcing above  does not

a↵ect the Mitchell rank for ⇥()-supercompactness of  since the forcing past stage  has

a dense subset which is closed up to the next closure point of ⇥ past . Call P ⇤ Q̇ the

forcing up to and including stage . First we shall lift j through P . Let G ✓ P be

V -generic and g ✓ Q be V [G ]-generic. First, we need an M -generic filter for j(P ). Since

cp(j) = , the forcings P and j(P ) agree up to stage . The forcing Q adds a club

c ✓  such that 8 < , 2 c implies o⇥()-sc ( )V < F ( ). Since o⇥()-sc ( )V < F ( )

implies o⇥()-sc ( )M < F ( ) = j(F )( ), and j(⇥)() = ⇥(), it follows that the th stage

of j(P ) is Q. Thus, j(P ) factors as P ⇤ Q̇ ⇤ Ptail where Ptail is the forcing past stage

. Since G ⇤ g is M -generic for P ⇤ Q̇, we only need an M [G ][g]-generic filter for Ptail

which is V [G ][g]. We will diagonalize to obtain such a filter, after checking that we have

met the criteria. Since |P | = |Q| = , and V |= GCH, it follows that the forcing P ⇤ Q̇

has the + -chain condition. Since M ⇥() ✓ M , it follows [Hamkins1 (Theorem 54)] that

M [G ][g]⇥() ✓ M [G ][g]. Since |P | = , and has the + -chain condition, it has at most

+ ⇥()<
< = + many maximal antichains. Since |j(+ )|  + = ⇥()+ , it follows that Ptail

has at most ⇥()+ many maximal antichains in M [G ][g]. Since Ptail has a dense subset which

is closed up to the next closure point of ⇥ past , it has a dense subset which is ⇥()-closed.

Thus, we can enumerate the maximal antichains of Ptail , and build a descending sequence

of conditions meeting them all through the dense set which has ⇥()-closure, using this

closure and the fact that M [G ][g] is closed under ⇥()-sequences to build through the limit
 73

stages. Then the upward closure of this sequence Gtail ✓ Ptail is M [G ][g]-generic and is in

V [G ][g]. Thus j(G ) = G ⇤ g ⇤ Gtail is M -generic, it is in V [G ][g] and j 00 G ✓ j(G ). Thus

we have the partial lift j : V [G ] ! M [j(G )].

Next we lift through Q by finding an M [j(G )]-generic filter for j(Q). The forcing j(Q)

adds a club C ✓ j() such that 8 < j(), 2 C implies o⇥()-sc ( )M [j(G )] < j(F )( ).

Since j(G ) 2 V [G ][g] and M ⇥() ✓ M , it follows that M [j(G )]⇥() ✓ M [j(G )] by

[Hamkins1 (Theorem 53)]. Since for every < , there is a dense subset of Q which is

 -closed, it follows that there is a dense subset of j(Q) which is ⇥()-closed. Also

since |Q| =  and Q has the + -chain condition, the forcing Q has at most + many

⇥()<
maximal antichains. Therefore j(Q) has at most |j(+ )|  + = ⇥()+ many maximal

antichains in M [j(G )]. Thus we can diagonalize to get a generic filter, but we will do so

with a master condition as follows. Let c = [g the new club of  and consider c̄ = c [ {}

which is in M [j(G )]. We have 2 c implies o⇥()-sc ( )V < F ( ). Thus, 2 c implies

o⇥()-sc ( )M < F ( ). By Lemma ?? it follows 2 c implies o⇥()-sc ( )M [j(G )]  o⇥()-sc ( )M <

F ( ) = j(F )( ). Since o⇥()-sc ()M [j(G)]  o⇥()-sc ()M = 0 < min{o⇥()-sc ()V , F ()}, and

j(⇥)() = ⇥() it follows that oj(⇥)()-sc ()M [j(G)] < F () = j(F )(). Thus c̄ is a closed,

bounded subset of j() such that every in c̄ has oj(⇥)()-sc ( )M [j(G )] < j(F )(). Thus

c̄ 2 j(Q). Thus diagonalize to get an M [j(G )]-generic filter g ⇤ ✓ j(Q) which contains

the master condition c̄. Then j 00 g ✓ j(g) = g ⇤ . Therefore we may lift the embedding to

j : V [G ][g] ! M [j(G )][j(g)].

74

The success of the lifting arguments for o⇥()-sc ()M = 0 < min{o⇥()-sc ()V , F ()},

where 0 was arbitrary, shows that o⇥()-sc ()V [G] min{o⇥()-sc ()V , F ()}. By Lemma ??

we have o⇥()-sc ()V [G]  o⇥()-sc ()V . All that remains is to see that o⇥()-sc ()V [G]  F ().

Consider any ⇥()-supercompactness embedding j : V [G] ! M [j(G)] with critical point

 in V [G], and the new club c ✓ , where 8 < , 2 c implies o⇥( )-sc ( )V < F ( ).

This club is in any normal fine measure on P ⇥(). By the induction hypothesis, 8 < ,

2 c implies o⇥( )-sc ( )V [G] < F ( ). Applying j to this statement gives 8 < j(), 2 j(c)

implies oj(⇥)()-sc ( )M [j(G)] < j(F )( ). Since F  represents F (), we have j(F )() =

F () and since ⇥  represents ⇥(), we have j(⇥)() = ⇥(). Since  < j() and

 2 j(c) it follows that o⇥()-sc ()M [j(G)] < F (). Thus by Lemma ??, since j was arbitrary,

o⇥()-sc ()V [G]  F (). Thus o⇥()-sc ()V [G]  min{o⇥()-sc ()V , F ()}. Thus for any  which

is ⇥()-supercompact for which ⇥00  ✓ , and F  represents F (), and ⇥  represents

⇥() in V has o⇥()-sc ()V [G] = min{o⇥()-sc ()V , F ()}. ⇤

The last theorem is about large cardinal properties close to the top. A cardinal  is huge

with target if there is an elementary embedding j : V ! M , with critical point , such that

j() = and M ✓ M . A cardinal  is superhuge if it is huge with target for unboundedly

many cardinals.

Theorem 45. If  is huge with target , then there is a forcing extension where  is still

huge with target , but  is not superhuge.

 75
+
Proof. Suppose  is huge with target . Force with Add(!, 1)⇤ Add( , 1). Then, by

Hamkins and Shelah [HS], in the extension, the cardinal  is not ⌘-supercompact, for all

+
⌘ > . Thus, in the extension,  is not superhuge, since it is not even -supercompact,

hence not +
-huge. But,  is still huge with target since the forcing is  -closed. ⇤

Open Questions:

1. How high does the hierarchy of inaccessible degrees go? Can we define E0 -inaccessible?

2. If  is a ⌃n+1 -reflecting cardinal, is there V [G] where  is ⌃n -reflecting, but not ⌃n+1 -

reflecting?

3. Is there a forcing like this for n-extendible cardinals?

4. Given any large cardinal degree, is there a forcing to softly kill?

5. If o() = n is there a forcing extension where  is least with o() = n?

6. Is there a forcing extension where the least strongly compact cardinal is least with

o() = 3, the second strongly compact cardinal is least above this with o() = 2 and the

third strongly compact cardinal is the least measurable above this one?

7. Is there a forcing extension where the least strongly compact cardinal is the least measur-

able, the second strongly compact cardinal is least with o() = 2, the third strongly compact

cardinal is least with o() = 3, ...

76

References

[Apter/Gitik] Apter/Gitik, The least measurable can be strongly compact and indestructible. The Journal of

Symbolic Logic, Vol. 63, No. 4, 1404-1412, Dec., 1998.

[Easton] Easton, Powers of regular cardinals. Ann. Math. Logic 1, 139-178, MR 42 #4392, 1970.

[Fuchs] Fuchs, Personal correspondence, May 2014.

[Gaifman] Gaifman, A generalization of Mahlo’s method for obtaining large cardinal numbers. IJM 5, 188-

200, 1967.

[Gitman] Gitman, Ramsey-like cardinals. Journal of Symbolic Logic, vol. 76, iss. 2: 519-540, 2011.

[Hamkins1] Hamkins, Forcing and Large cardinals, based on CUNY set theory lecture 2004.

[Hamkins2] Hamkins, Destruction or preservation: as you like it. Ann. Pure Appl. Logic, vol. 91, iss. 2-3,

pp. 191-229, 1998.

[Hamkins3] Hamkins, Answer mathoverflow question, mathoverflow.net Forcing mildly over a worldly car-

dinal. May 7, 2013.

[Hamkins4] Hamkins, Extensions with the approximation and cover properties have no new large cardinals.

Fundamenta Mathematicae 180(3):257-277, 2003.

[Hamkins5] Hamkins, A multiverse perpective on the axiom of constructibility Infinity and Truth, Vol. 25,

2013.

[Hamkins/Shelah] Hamkins, Shelah. Superdestructibility: a dual to Laver’s indestructibility. Journal of Sym-

bolic Logic, vol. 63, iss. 2, pp. 549-554, 1998.

[Jech] Jech, Set Theory. Third Millenium Edition, Revised and Expanded, Springer Monographs in Mathe-

matics, p. 120, 2002.

[Jensen/Kunen] Jensen, Kunen, Some combinatorial properties of L and V. Unpublished, 1969.

[Kan] Kanamori, The Higher Infinite. Springer Monographs in Mathematics, 2nd edition, 2003.

[Magidor] Magidor, A Study on Identity Crises. Ann. pure appl. logic, vol. 10, pp. 33-57, 1976.
 77

[Mahlo] Mahlo, Uber lineare transfinite Mengen. Berichte Konig. Sachs. Ges. Wiss. Leipzig 63, 187-225,

1911.

[Mitchell] Mitchell, Hypermeasurable cardinals. In ”Logic Colloquium ’78” (M. Bo↵a et al., eds.), Mons,

Stud. Logic Foundations Math. 97, North Holland Publishing Co., Amsterdam-New York, 1979, pp.

303-316. MR 82j:03067; the entire collection MR 81a:03001, 1978.

[Schanker] Schanker, Weakly measurable cardinals. Mathematical Logic Quarterly 57 (3):266-280, 2011.

[Todorcevic] Walks on ordinals and their characteristics. Progress in Mathematics, 263. Birkhuser Verlag,

Basel, vi+324 pp. ISBN: 978-3-7643-8528-6, 2007.