Complex Networks

Albert-László Barabási!
Center for Complex Networks Research and Department of Physics!
Northeastern University, Boston!
Central European University, Budapest!
!
Division of Network Medicine!
Harvard Medical School!

www.BarabasiLab.com
INTERNET!

domain2!

domain1!

router!

domain3!
SOCIETY Facebook: The Social Graph!
!

Keith Shepherd's "Sunday Best”. http://baseballart.com/2010/07/shades-of-greatness-a-story-that-needed-to-be-told/!

Video  of  rota+ng  network  
RANDOM NETWORK MODEL!

Pál Erdös! Alfréd Rényi!

(1913-1996)! (1921-1970)!

Erdös-Rényi model (1960)!

!
Connect with probability p!

p=1/6 N=10!
〈k〉 ~ 1.5!
Video  of  random  link  forma+on  
RANDOM NETWORK MODEL!

Pál Erdös! Alfréd Rényi!

(1913-1996)! (1921-1970)!

Erdös-Rényi model (1960)!

!
Connect with probability p! Degree  distribu,on  

p=1/6 N=10!
〈k〉 ~ 1.5!
WORLD WIDE WEB!
Scale-free Network! Random Network! Nodes: WWW documents
Links: URL links!

Expected!
Over 1 Trillion documents
ROBOT: collects all URL s
found in a document and
follows them recursively

P(k) ~ k -ϒ!

Found!
R. Albert, H. Jeong, A-L Barabási, Nature, 401 130 (1999).!
METABOLIC NETWORK ! ! ! ! ! !PROTEIN INTERACTIONS!

Jeong,  Albert,  Oltvai,  and  Barabási,  Nature  407:  651  (2000);    Nature  411:  41  (2001).    
MANY REAL WORLD NETWORKS HAVE A SIMILAR ARCHITECTURE:!

Scale-free networks

WWW, Internet (routers and domains), electronic circuits, computer software,

movie actors, coauthorship networks, sexual web, instant messaging, email
web, citations, phone calls, metabolic, protein interaction, protein domains,
brain function web, linguistic networks, comic book characters, international
trade, bank system, encryption trust net, energy landscapes, earthquakes,
astrophysical network…
ORIGIN OF SF NETWORKS !Growth and preferential attachment!
(1) Networks continuously expand by the GROWTH: !
addition of new nodes! add a new node with m links!

WWW : addition of new documents!

(2) New nodes prefer to link to highly PREFERENTIAL ATTACHMENT: !

connected nodes.! the probability that a node connects to a
node with k links is proportional to k.!
WWW : linking to well known sites!

ki
!(ki ) =
" jkj
Video  of  preferen+al  aHachment  

P(k) ~k-3

Barabási & Albert, Science 286, 509 (1999)!

CONTINUUM THEORY!

∂ki ki ki
∝ Π (ki ) = A = ,  with  ini+al  condi+on   ki (ti ) = m
∂t ∑ j k j 2t
t
ki (t) = m
ti

m 2t m 2t m 2t
P(ki (t ) < k ) = Pt (ti > 2 ) = 1 − Pt (ti ≤ 2 ) = 1 − 2
k k k (m0 + t )
∂P(ki (t ) < k ) 2m 2t 1
∴ P(k ) = = ~ k −3
 γ  =  3  
∂k mo + t k 3
Barabási,  Albert    &  Jeong,  Physica  A  272,  173  (1999).      
Bollobás  et  al.  Rand.    Structures  &  Algorithms  (2001).        
DEGREE DISTRIBUTION!

2m(m +1) "3

P(k) = P(k) ~ k for large k!

k(k +1)(k + 2)
γ=3

 (i)  The  degree  exponent  is  independent  of  m.  

  !
(ii)  The  network  reaches  a  sta+onary  scale-­‐free  state.    
 
(iii)  The  coefficient  of  the  power-­‐law  distribu+on  is  propor+onal  to  m2.  

Krapivsky, Redner, Leyvraz, Phys. Rev. Lett. 2000; Dorogovtsev, Mendes, Samukhin, Phys. Rev.
Lett. 2000; Bollobas et al. Random Struc. Alg. 2001!
ROBUSTNESS OF SCALE-FREE NETWORKS!

node failure!

Failures   Two  videos  of  networks  under  random  and  targeted  aHack   A3acks  

Albert, Jeong, Barabási, Nature 406 378 (2000)!

ROBUSTNESS OF SCALE-FREE NETWORKS!

$ 1 # >3
1 && 3"#
fc = 1 " 2
< k > 2 "#
k2 = K min % N # "1 3 ># >2
"1 <k> 3 "# & 1
k &' N # "1 2 > # >1

γ>3:    <k2>  is  finite;  the  network  will  break  apart  at  a  finite  fc  

!
γ<3:    <k2>    diverges  in  the  Nà  ∞  limit,  so  f c  à  1  
     we  need  to  remove  all  the  nodes  to  break  the  system.  
   
     Finite  systems:   3#$
#
$ #1
f c "1 # CN

Internet:  Router  level  map,  N=228,263;  γ=2.1±0.1;        κ=28  à      fc=0.962  

    !

Cohen  &  Havlin,  Physical  Review  LeIers  (2000)  (2001)  

!COMMUNITY STRUCTURE!

Y.-­‐Y.  Ahn,  J.  P.  Bagrow,  S.  Lehmann,  Nature  (2010)  

COMMUNITY STUCTURE Nodes in the same community are alike!

Busy@Midnight   Sleep@Noon  

Sleep@Midnight   Busy@Noon  
t � 0:00 t � 12:00

Midnight! Noon!

Ahn,  Bagrow,  Lehmann,  Nature  (2010)  

The power of maps

Barabasi Lab!
!

Barabasi Lab!
!

Barabasi Lab!
A system is controllable if it can be driven from any initial state
to any desired final state in finite time.
Video of car components
http://static.ddmcdn.com/gif/mars1.jpg
Video of rotating network
!
LINEAR SYSTEMS! ! Kalmanʼs Controllability Condition!

• Linear Time-Invariant Dynamics

dX
= A ! X(t)+ B! u(t)
dt
A " R N#N : weighted wiring diagram " x1 %
$ '
X(t) " R N#1 : state vector. X= $ x2 ' " u1 %
; u = $ '.
$ x3 ' # u2 &
u(t) " R M#1 : input vector (M $ N). $ '
# x4 &

B " R N#M : input matrix

( % control configuration). !

Ÿ Kalman’s Rank Condition:

A system is controllable iff its rank C = N
controllability matrix has full rank.
C = [B, A" B, A 2 " B,!, A N #1" B]
R. E. Kalman, J.S.I.A.M. Control (1963)
EXAMPLES: Controllable or not controllable? !

Yes No Yes
DIFFICULTIES !

1. Parameters (link weights): usually unknown.

e.g. gene regulatory network, Internet, etc.

2. If brute-force search: (2N-1) combinations.

"N% "N% "N% N

$ ' + $ ' +! + $ ' = 2 (1
#1& #2& #N&

3. Kalman’s rank condition is hard to check for large system.

! rank C = N
C = [B, A" B, A 2 " B,!, A N #1 " B] has dimension N $ NM.
 Matching
Matching : Maximum matching :
a set of edges without a matching of the largest size.
Network common vertices.

Perfect Matching

matched unmatched

Lovász, L. & Plummer, M.D., Matching Theory

MATCHING IN DIRECTED NETWORKS:!

Matching  :    a  set  of  edges  without  common  heads  or  tails.  

Directed  Network   Maximum  matching  

Perfect  Matching  

matched   unmatched  

Minimum  Input  Theorem:    Driver  nodes  =  Unmatched  nodes  

Y.-­‐Y.  Liu,  J.-­‐J.  Slo+ne,  A.-­‐L.  Barabási,  Nature    (2011)  

EXAMPLES: Identifying the driver nodes!

network   Maximum  matching   controlled  network  

Brute-­‐force  search    
O(2N)~1030  for  N=100.  
Hopeless!  

Hopcroh-­‐Karp  Algorithm  
O(N1/2L)  Polynomial!  
Fast  even  for  N~106.  

Y.-­‐Y.  Liu,  J.-­‐J.  Slo+ne,  A.-­‐L.  Barabási,  Nature    (2011)  

 ND of real networks

1. Overall we see no obvious trend in nD (or ND) across these networks.

2. As a group, regulatory networks display very high nD ≈ 0.8.
3. A few social networks display the smallest observed nD values.
 Role of hubs

1. The fraction of driver nodes is significantly higher among low degree 
nodes than among the hubs.
2. Mean degree of driver nodes <kD> is either significantly smaller or 
comparable to <k>.

Driver nodes tend to avoid the hubs.

 ND real vs. ND rand

Complete Degree-preserving
randomization randomization

ND is mainly determined by degree distribution.

 Degree Dependence

Construct ER and SF networks using the static model (Goh et al. PRL 2001)
1. ER : n D ( k ) " e # k /2
as k >> 1.
% 1 (
#'1# * k /2
& $ #1)
2. SF : n D ( k ,$ ) " e as k >> 1
(consistent with $ c = 2 :SF : n D ($ ) +1 as $ +$ c = 2. ).
 Degree Heterogeneity

Degree heterogeneity H = 2 × Gini coefficient

" $i $ j i # j P(i)P( j)
H= =
k k
 Results
• Mean degree <k> and degree heterogeneity H are
the two main factors that determine ND.
• Sparse and heterogeneous networks are harder to
control than dense and homogeneous networks.
!
Video  of  individuals  that  
are  NOT  control  nodes  
Video  of  individuals  that  
ARE  control  nodes  
 !
OBSERVABILITY: Reconstruct the state of a complex system!

Observability:  
 Reconstruct  the  state  of  the  
system  using  data  collected  
from  a  small  number  of  
observers.    

Liu,  Slo+ne,  Barabási  PNAS  (2013)  

Network Science
an interactive textbook

barabasilab.com/NetworkScienceBook/  

facebook.com/NetworkScienceBook  
 genes and other cellular components interact with each
other. Most cellular processes, from the processing of food

CHAPTER 1
by our cells to sensing changes in the environment, rely on
molecular networks. The breakdown of these networks is
responsible for most human diseases. This has led to the
emergence of network biology, a new subfield of biology
that aims to understand the behavior of cellular networks.
A parallel movement within medicine, called network
medicine, aims to uncover the role of networks in human
disease (Image 1.7a/b). Networks are particularly import-
ant in drug development. The ultimate goal of network
pharmacology is to develop drugs that can cure diseases
without significant side effects. This goal is pursued at
many levels, from millions of dollars invested to map out
!"#$%&'()#*&'()+ cellular networks to the development of tools and databas-
,%-./012&34&+3/5/$6&#47&"%73834%( es to store, curate, and analyze patient and genetic data.
!"#$%&#'()&*#(+#,-(#.//0&/#(+#!"#$%&'(&)*&+,'-&.&#*/,1#,%&#,(2#*&).&-# Several new companies take advantage of these opportuni-
3(0*4!5#.4#6&4&,.'/7#$%&#'()&*#+*(8#9::;1#+('0/&/#(4#4&,-(*<#=.(5(6>#?@@A1# ties, from GeneGo that aims to collect accurate maps of cel-
,%&#'()&*#+*(8#9:@@#B./'0/&/#4&,-(*<#8&B.'.4&#?@9A7 lular interactions from scientific literature to Genomatica
that uses the predictive power behind metabolic networks
="#$%&#2*(8.4&4,#*(5&#4&,-(*</#25!>#.4#=(,%#'&55#=.(5(6>#!4B#8&B.'!5#
*&/&!*'%#./#.550/,*!,&B#=>#,%&#+!',#,%!,#,%&#9::;#!*,.'5&#(4#4&,-(*<#=.(5(6>#
to identify drug targets in bacteria and humans. Recently
./#,%&#/&'(4B#8(/,#'.,&B#!*,.'5&#.4#,%&#%./,(*>#(+#C!,0*&#D&).&-/#E&4&,.'/7# most major pharmaceutical companies have made signifi-

Introduction

From Saddam Hussein to network

theory

Vulnerability due to interconnec-

tivity

Networks at the heart of complex

systems

Two forces helped the emergence

of network science

The characteristics of network

science

The impact of network science

Scientific impact !"#$%&'(9

:;%&4%-./01&+%;347&#&"353-#06&%4$#$%"%4-(
Summary
$%./#B.!6*!8#-!/#B&/.64&B#B0*.46#,%&#F+6%!4#-!*#,(#2(*,*!>#,%&#F8&*.'!4#/,*!,&6>#.4#F+6%!4./,!47#G%.5&#.,#%!/#=&&4#(''!/.(4!55>#*.B.'05&B#.4#,%&#2*&//1#
Bibliography .,#2(*,*!>/#-&55#,%&#'(825&H.,.&/#!4B#,%&#.4,&*'(44&',&B#4!,0*&#(+#!#8.5.,!*>I/#&46!6&8&4,7##012"3&'4%52'!&+'65%7'8*2&,9
'<&=&,>:?@AB&CD!>,D>

barabasilab.com/NetworkScienceBook/                                  facebook.com/NetworkScienceBook  
 
 #789:(;<=>
$?8@ABCB9D<

CHAPTER 2
H#"#$%&'()"+8-$#8$"&$"$#8'?#%$("4!#9"#$%&'()+"0-+%-#2?-+*$0"=9"+'4$"
$,$4$#%!(9"/('/$(%9"'1"%*$"?#0$(,9-#2"2(!/*5"K$($"&$"+?44!(-G$"%*$"
4'+%"8'44'#,9"$#8'?#%$($0"$,$4$#%!(9"#$%&'()"%9/$+3"%'2$%*$("&-%*"
%*$-("=!+-8"/('/$(%-$+3"!#0"!#"-,,?+%(!%-.$",-+%"'1"($!,"+9+%$4+"%*!%"+*!($"%*$"
/!(%-8?,!("/('/$(%95"L'%$"%*!%"-#"4!#9"($!,"#$%&'()"&$"#$$0"%'"8'4=-#$"
+$.$(!,"'1"%*$+$"$,$4$#%!(9"#$%&'()"8*!(!8%$(-+%-8+5">'("$:!4/,$"%*$"
JJJ"-+"!"0-($8%$0"4?,%-C2(!/*"&-%*"+$,1C-#%$(!8%-'#+5"6*$"4'=-,$"8!,,"
#$%&'()"-+"0-($8%$0"!#0"&$-2*%$03"&-%*'?%"+$,1C,''/+5

5)'#+"3&"'()"&!*+,E"!"#$%&'()"&*'+$",-#)+"0'"#'%"*!.$"!"/($0$1-#$0" ."/01#)&"+23&#*).-"-#"4!#9"#$%&'()+"#'0$+"0'"#'%"-#%$(!8%"&-%*"
0-($8%-'#5"D:!4/,$+E"H#%$(#$%3"/'&$("2(-03"+8-$#8$"8',,!='(!%-'#"#$%&'()+3" %*$4+$,.$+3"+'"%*$"0-!2'#!,"$,$4$#%+"'1"!07!8$#89"4!%(-:"!($"G$('3"!""$%$'("""
/('%$-#"-#%$(!8%-'#+5 "$%)(***(+5"H#"+'4$"+9+%$4+"+$,1C-#%$(!8%-'#+"!($"!,,'&$0I"-#"+?8*"#$%&'()+3"
($/($+$#%-#2"%*$"1!8%"%*!%"#'0$$""*!+"!"+$,1C-#%$(!8%-'#5"D:!4/,$+E"JJJ3"
/('%$-#"-#%$(!8%-'#+5"

The bridges of Königsberg

Networks and graphs

Degree, average degree and degree

distribution

Real networks are sparse '#+"3&"'()"&!*+,E"!"#$%&'()"&*'+$",-#)+"*!.$"+$,$8%$0"0-($8%-'#+5" 45/&#$+26%E"-#"!"4?,%-2(!/*"#'0$+"!($"/$(4-%%$0"%'"*!.$"4?,%-/,$",-#)+"

D:!4/,$+E"JJJ3"4'=-,$"/*'#$"8!,,+3"8-%!%-'#"#$%&'()5 @'("/!(!,,$,",-#)+A"=$%&$$#"%*$45"K$#8$"!""""8!#"*!.$"!#9"/'+-%-.$"-#%$2$(5
Adjacency Matrix

Weighted and unweighted net-

works

Bipartite networks

Paths and distances in networks

Connectedness and Components

Clustering coefficient

Case study and summary

!"#$%&"'()"&!*+,-(!"#$%&'()"&*'+$",-#)+"*!.$"!"/($0$1-#$0"&$-2*%3" 3*46/"&"($+26%-(-#"!"8'4/,$%$"2(!/*"!,,"#'0$+"!($"8'##$8%$0"%'"$!8*"
Appendix A: Global clustering co- +%($#2%*"'("1'&"/!(!4$%$(5"6*$"$,$4$#%+"'1"%*$"!07!8$#89"4!%(-:"!($"!"#";" '%*$(I"#'"+$,1C8'##$8%-'#+5
efficient <"-1"""!#0"#"!($"#'%"8'##$8%$03"'("!"#$%$&"#$-1"%*$($"-+"!",-#)"&-%*"&$-2*%"&-7"
=$%&$$#"%*$45">'("?#&$-2*%$0"@=-#!(9A"#$%&'()+3"%*$"!07!8$#89"4!%(-:"
Bibliography '#,9"-#0-8!%$+"%*$"/($+$#8$"@!"#";"BA"'("%*$"!=+$#8$"@!"#";"<A"'1"!",-#)"=$C
%&$$#"%&'"#'0$+5"D:!4/,$+E"F'=-,$"/*'#$"8!,,+3"$4!-,"#$%&'()5

32."(.&5'E(2)'(.5442+E(F(GH

barabasilab.com/NetworkScienceBook/                                  facebook.com/NetworkScienceBook  
 
WHAT IS “NETWORK SCIENCE”?!

NRC Report on Network Science

An attempt to
understand networks
emerging in nature,
technology and
society using a
unified set of tools
and principles.!

What is new here?!

!
Despite the apparent differences, many
networks emerge and evolve driven by a
fundamental set of laws and mechanism.!
!
www.BarabasiLab.com!