LINKED The New Science of Networks ALBERT-LASZLO BARABASI PERSEUS PUB earetieval photo of the publishes Pesseus Books i ing isa meter oft “Tent design by Janice Tapia Set in 11-poine Gouly by the Perseus Books Group 6789 10-05 0403 [Rs printing, April 2002 ‘om the World Wide Web at |jwwwperseuspublishing com SzuileimnekeContents 1 ‘THe SECOND Link: TH RANDOM UNIVERSE 9 > Link: Six DEGREES 6 Fiast Link: INTRODUCTION Il PAAMTION, B 1K: SMALL WoaLDs “Tee Excnere Linsk: ERNSTRIN'S THE Niwrit Link: ACHILLES HER! is€S AND Favs THEE: INK: THE AWAKENING INT ‘Tue Twetrri Link: THE FRacMenven Ws 1 Links THE Mar oF Ls ‘Tre Fouxr genre Link: NeTwonk Ec THe LAST LINK: We WITHOUT a SPILER AcknowneDoments ay 2st 27THE FIRST LINK Introduction rwever. They all arrived at the exact samme and not one of them asked for a stock quote or a pecan pie ret less, hundreds of computers in Yahoo's Santa quarters were kept busy responding ns of legitimate cust et, waited. I was one of was frantically busy serving tet three minutes before next day the royals of the Web, Amazon.com, eBay, CNN.com, ETiadk that Yahoo n ghosts. I was patient for about engine. The ed to a more responsive sea chases, were forced t0 wai OF course, getting2 LINKED amounts of data were thrown at this prominent Website, much more it could ever handle. The massive denial-of-service attack Yahoo foam over rd the teen s artack, He was caught braggin ing behind the pseudonym MafiaBoy, this fifteen-year-old suc companies with access the world. Was he a contempo: beat the megact agree on one thing: The a were executed using tools available to a sites. MafiaBoy’s online antics revealed him to be a ran! : Lacking the sites he attacked and clumsy a ly managed to take vulnerable computers from universities and small companics, which he simply instructed to bombard Yahoo with messages ‘One can imagine a fifteen-year-old boy behind his bed the glow of his computer, nding sweet saisfaction in the prod yeard yout!" hurled at Yahoo. He must have screamed that phrase 1a times when Mom or Dad called him to come to dine th brute force wn easy targets, obviou door, dergo circum were excluded ion, had to obey the laws of contemporary Judaism, and 1m the Templ ly Jewish, Christianity. Very few walked! the path, Indeed, reaching them with the $e. In a fragmented and earthbound society wveled by foot, and the distances were long. Chi the historical figure we know toxay as Jesus of Nazaret ing experts would describe his message as was passed down by generat and thodox and pious Jew who never met Jesus. Whi was Saul, he is better known to us by his Roman name, Paul. P tnission was to curb Chri y. He traveled from community to com ns because they put Jesus, condemned by mer, on the sume level as God. He used the authorities as a bla4 LINKED scourging, ban, and excommunication to uphold the traditions and to 19 adhere to Jewish law. Nevertheless, according to fans underwent a force the dev historical accounts, this fierce petsecutor of Cl sudden conversion in the year 34 and became the fiercest supporter of i it possible for a small Jewish sect to become the xe Westen world for t succeed? He understood 1 to spread beyond Judaism, the high barr had to be abolished. Circumcision and the s laxed. He took his message to the original and received the mandate to continue evangelization without demand- the new faith, mal ¢ for Christianity ing circumcision. But Paul understood thar this was not enough: The message had to spread. So he used his firsthand knowledge of the social network of the lized world from Rome to Jerusalem to reach and con- ld. He walked nearly 10,000 miles in the st century ci vert as many people as next twelve years of his life. He did not wander randomly, however; he reached out to the biggest communities of his era, people and places in which the faith could germinate and spread most effectively. He was the first and by far the most effective salesperson of Christian- ity, using theology and social networks equally effectively. So should he, ot Jesus, of the message be credited for Christianity’s success? Could it happen again? 2. There are huge differences between MafiaBoy and Paul: MafiaBoy's was an act of destruction. Paul intentions, became a bridge builder between early Christian comm have something important in common: Both were masters of the net~ work. Though neither of them thought abou key to their success was the existence of a complex network that of- fered an effective medium for their actions. MafiaBoy operated on a Internet is the fastest and most effect jes, But the two in these terms, che network of computers—t Introduction 5 way to reach the largest number of people at the turn of lennium. Paul was a master of first-century social the rvork at the begi nd spread a faith, ‘easy jem in their actions. Bur nearly 7,000 years after Paul we are making the first inroads toward understanding what made Paul and MafiaBoy successful. We now know that the answer lies as much in the steuc- ture and topology of the networks on which they operated as in their abilicy to navigate them, Paul and MafaBoy succeeded because we are all co igi great Argentinea hes everything.” idern er of “There be dragons there!” wrote the ancient mapmakers, marking off the frightening unknown. As adventurous explorers penetrated every region of the globe, these monster-marked patches gradually disap. peared, But there ate 5 map of how the different pa ‘croscopic universe locked w 1. The good news the jentists have been learning iding new light on even be imagined a few years ago. Detailed maps of the Internet have unmasked che Intemet’s vulnerability to hackers. Maps of companies ownership have traced the trail of power and money in Silicon Valley. Maps of interactions between species in ecosystems have offered glimpses of humanity's destructive impact on have pro- teal surprise has come the environment, Maps of genes working toge vid insights into how cancer works ing these maps side by side. Ju ‘ons that are almost indistinguishable, we have learned that these as diverse humans share skel6 LINKED diverse maps follow a common blueprint. A string of recent breath- taking discoveries has forced us to acknowledge that amazingly simple and far-reaching natural laws govern the structure and evolution of all the complex networks that surround us. 4. Have you ever seen a child take apart a favorite toy? Did you chen see the little one cry after realizing he could not put all the pieces back to- gether again? Well, here isa sectet chat never makes the headlines: We idea how to put it back to- 15 to disassemble nature have taken apart the universe and have ns of research d 1 century, we are just now acknowledging that we have no clue ue —except to take it apart further. Reductionismn was the driving force behind much of the twentieth century's scientific research. To comprehend nature, i ells us, we frst must decipher its components. The assumption is that once we under- stand the parts, it will be easy to grasp the whole. Divide and conquer; the devil is in the details. Therefore, for decades we have been forced to see the world through its constituents. We have been trained to study atoms and superstrings to understand the universe; molecules to com- individual genes to understand complex hurnan behavior; prehend prophets to see the origins of fads and religions. Now we are close to knowing just about everything there is co iknow about the pieces. But we are as far as we have ever been from understanding nature as a whole, Indeed, the reassembly turned out to be much harder than scientists anticipated. The reason is simple: Riding reductionism, we run into the hard wall of complexicy. We have learned that nature is not a well-designed puzzle with only one ‘way to put it back ogether. In complex systems the components can fit in so many different ways that it would take billions of years for us to try them all. Yet nature assembles the pieces with a grace and pre- cision honed over tillions of years. [t does so by exploiting the all- encompassing laws of self-organization, whose roots are still largely a mystery to us. Irarduction 7 ‘Today we increasingly recognize that nothing happens in isola- tion. Most events and phenomena are connected, caused by, and teracting with a huge number of puzzle, We have come to see everything is linked to everything else. We are witnes tion in the making as scientists from all different disciplines discover that complexity has a strict architecture. We have come to grasp the importance of networks. With the Internet dominating our life, the word network is on everybody's lips these days, featured in company names and popular journal titles. After September I, witnessing the deadly power of ter- rorist networks, we had to get used to yet another meaning of the term. Very few people realize, however, that the rapidly unfo ence of networks is uncovering phenomena that are far more exciting and revealing than the casual use of the word network could ever con- vey. Some of these discoveries are so fresh that many of the key results still circulate as unpublished papers within the scientific community They open up @ novel perspective on the interconnected world around us, indicating that networks will dominace the new century to a much greater degree than mose people are yet ready to acknowledge. They will drive the fundamental questions that form our view of the world in the coming era This book has a simple aim: to get you about how networks emerge, what they look like, and how they evolve. It shows you a Web-based view of nature, society, and business, a new framework for understanding issues ranging from democracy on the Web to the vulnerability of the Intemet and the spread of deadly Networks are present everywhere. All we need is an eye for them. As Xk to link ky society as a complex social network and to grasp the smallness of this ‘great world in which we live. You will eome 0 understand how and why Paul succeeded and how, despite some obvious differences, his so- milieu was similar to the one we experience today. You will see the enges doctors face when they attempt to cure a disease by focusing we sci- 0 think necworks. It is wu move from sb will learn to see8 LINKED fon a single molecule or gene, disregarding the complex interconnected- Ibe reminded that MafiaBoy is not u will come to appreciate how the Internet, often attacking networks. viewed as entirely human in its creation, has become more akin to an ving the power of the basic laws organism or an ecosystem, demonst that govem all networks. You will see how the emergence of ter also ruled by the laws of network take advantage of the fundamental robustness of nature's webs. You'll ss among such diverse systems as the cone to grasp the opening you to step out of the bo ientific revolution: che new science of networks. THE SECOND LINK The Random Universe (ON Serrewsen 18, 1783, IN St. PETERSEURG Leonhard Euler started the day as usual. He gave a mathematics lesson to one of his grandchil- dren and took up some calculations on the fight of balloons. Just three ‘months earlier, south of Lyon, the Montgolfier brothers had launched ‘an enormous balloon that rose 6,500 feet into the air and landed s about a mile away. Euler was working out the mechanics of the bal- loon’s motion as the Montgolfer brothers were preparing to launch a sheep into the air in front of King Louis XVI in Paris, a flight that place the next ¢: never heard al event, however. After lunch, working with his assistants, he made some calculations on the orbit of the recently discovered planet Uranus. The equations introduced by him, capturing the planet's peculiar of would lead decades later to the discovery of the planet Neptune. Euler did not live to witness that discovery either. About five o'clock in the afternoon, he suffered a brain hemorrhage and before losing consci prolific care: Euler, a Swiss born SSt. Petersburg, had an extraordina 5, physics, and engineering. Not only was the importance of his discoveries unparalleled, their sheer quantity is also overwhelming. Opera Omnia, the incomplete record of Euler’ collected works, currently 9to LINKED es, six hundred pages each. The last seventeen , between his return to St. Petersbut the age of 76, were rather tragedies, about half of his works were wri These rumultuous. Yet, despite many pers pleted while he continued to publish an average of one mathema pper per week in the journal of the St. Petersburg Academy. The amazing thing is that he barely wrote or read a single line during thi ly lost his sight soon after re left completely idressing an amusing problem thar originated in K own not too far from Euler's home in St. Petersburg. Konigsber in easter Prussia, did not suspect in the early eighteenth fewer than seven bridges across the river. Most elegant island Kneiphof, which was caught between the two branches of the Peeg nal bridges crossed the two branches of the river (Figure 2.1). The people of amused themselves was: “Can one walk act twice?” No one was to find such a with other parts of the the seven iph theory, Today graph theory is the basis for our works. During the centuries after Euler it grew intoa Pregel River. Solving the Konigsberg problem meant finding a route around the city thas would require a persan 10 eros each bridge only once. In 1736, Leonhard Euler gave birth 2 graph theory by replacing each ofthe four land areas with nodes (Sto D) and each bridge with a link (a 10 g), obtaining a graph with four nodes each the door on the field of networks, let us briefly revi process that led Euler to the introduction of the first graph. reasoning easily understood even by those nut chematies. Nevertheless, he proof cory but rather the step that he took ro solve the prob- Buler’s great insight lay in viewing Konigsberg’s bridges asa graph, cction of nodes connected by links, For this he used nodes to t made2 LINKED represent each of the four land areas separated by the river, distin- guishing them with letters A, B, C, and D. Next he called the bridges links and connected wi bridge between them. He thus obtained a graph wl pieces of land and of links must be either the starting or the end point of the journey. inuous path that goes through all bridges can have only one start- ing and one end point. Thus, such a path cannot exist on a graph that has ‘more than two nodes with an odd number of links. As the Kénigsberg sraph had four such nodes, one could not find the desired path. For our purpose the most aspect of Euler's proof is path does find it. Rs ‘the graph. Given the layout of the Konigsberg bridges, no wwe are, we will never succeed at finding the desired path. The people of Kénigsberg finally agreed with Euler, gave up their fruitless search, and in 1875 builea new bridge between B and C, increas- ing the number of links of these two nodes to four. Now only two nodes (A and D) with an odd number of links remained. [t was then rather rd to find the desired path. Perhaps the creation of their coffeehouse problem. But a change in the layout, the addi 1 of only one extra link, suddenly ine. In many ways Bul ;mbolizes an important message of this book: The construction and structure of graphs or networks is the key to understanding the complex world around us. Small changes in the jons made by . Cayley, Kieehhoff, and “The Rando Universe B Pélya, They uncovered just about everything that is known, but ordered graphs, such as the lattice formed by atoms in a crys the hexagonal lattice made by bees in a bechive. Un eth century the goal of graph theory was simple: and catalogue the properties of the v cluded finding a way 1873, or finding a sequ such that each square is visited only once and the kni stacting point. Some for centuries. ‘Two centuries passed after Euler’s inspiring work before mathemati- cians moved from studying the properties of various graphs to asking the quintessential question of how gr works, came about. Indeed, how laws governing their appearance the 1950s, when two Hung: ians made a revolution in graph theory tered with a weird gait through the elegant shoe shaped fee Knocking on the store's door—an act ‘odd back then as today—he entered, counter, and went up to a fourteen. ye ve me a four digit number,” he s 532," came the wide-eyed boy’ reply as he sta re. The older boy The square of not let him4 LINKED know thirty-seven,” and without taking a breath he conti “Did you know that the point ight line do not form a counta- ble set?" After sl as evidence, his and so he of the cwentieth century. He wrote more than ‘mathematics papers before his death in 1996. TI Allftéd Rényi. These cight papers addressed for the first question pertaining to our under: ww do networks form? Th random networks. Th cegant theory so profoundly determined our thinking about net that we are still struggling to break away from its hold. uundred guests who have been selected and in- ey do no le other person quest list. ‘Offer this group of strangers wine and cheese, and they will immediately cchat, as human beings’ inborn desire to meet and know each inevitably brings them together. Soon you will see thirty to forty the unlabeled dark green bottles is a rare ewenty-yearold vintage por wan that with the red label. But ask that guest to share this in. or her new acquaintances. You know that meet two or three people in become bored talking to the same person join other groups. An outside observer wor cial. Yer there are invisible social links between people who met earl bur now belong ro different groups. As a consequence, subtle paths st t0 cach other For example, not notice anything spe- none of whom initially knows ‘one another, so ing in small groups. At fist, the _groups are isolate from each othe lous lines) becween those in the same group, everyone le of lla sranger As time goes om {right panel), three guests move to dif- ferent groups and a giant cluster emerges. Although not everyone knows everyone ude all the guest By following 1 to css ele, there is there is a path from John to Mary through Mike. If John knew about the wine, chances are that now Mary knows to9, since she could hear it from Mike, who was told by John. As he guests will be increasingly interwoven by such intangi acquaincance: sive wine is increasingly endangered as group of insiders ‘Assuming that each person passes 6 for his new acquaintances, will the reputation of the fine e and more cl16 LINKED ematician is ay, quoting e turned into a much quoted if cach person gets to know at least one other guest, take only thirty minutes to form a single in les all gue ‘ommendation for the wine, you may find yourself ti tle into your expect in the room, Minutes after you igan empty bor. 4. The guests we met at the cockeail party are part of a problem in graph th satis pioneered by Euler. The guests are the web of acquaintances—a graph —emenges,a bu Tinks. Computers biochemical ee cells connected by axons, islands connected by graphs. Whatever the iden mathematician they form the same at Despite its elegance, simplifying body linked by inked by trade, nerve idges are all examples of the nodes and links, for a a graph ora network webs into graphs poses some between human soc ke friends and acquaintances through random encounters and c ; unforgiving laws of chemistry and physics govern all reactions between molecules. There must be a clear difference in the rules that govern the placement of links in the various networks we encounter in nature , where the snge by proposing an elegant matheai “The Rendom Unione 0 complex graphs within a single framework, Since such disparate rules in building their own networ liberately disregarded this diversity and came up wi Ered liked to say, “about whether you cre in other words, are the truths al yet know them?” rds had a clear answer to this are there among question: ical tn truths, and we just rediscover seemed to dom networks playe revealed in the coming chapters. Ends himself created mathem ruths and an alternative view of our world by des hazarded his best guess in assuming that God friend Albert Einstein, at Princeton, was con “God does not play dice with the 7 5. Let's go back to our cocktail party and the exercise is theory. YouF LINKED port of single cluster auch chat, starting from any node, we “any other by navigating along the links between the nodes. Th moment when your expensive wine is in danger, since a rumor can reach everyone who belongs to the giane cluster. Mathematicians call is phenomenon the emergence of a giant component, one that in- cludes a large fraction of all nodes. Physicists call it percolation and witl tell you that we just witnessed a phase transition, similar to the mo- iment in which wacer freezes. Sociologists would tell you that your sub- jects had just formed a community. Though diferes ‘terminology, chey all agree that when we randomly pick \d connect pairs of nodes together in a network, something special happens: The network, after placing a critical number of links, drasti- cally changes, Before, we have a bunch of tiny isolated clusters of nodes, disparate groups of people that communicate only wishin the clusters, After, we have a giant cluster, joined by almost everybody. ines may have different 6. Each of us is part of a large cluster, the worldwide social net, from which no one is left out. We do not know everybody on this globe, but it is guaranteed that there isa path between any to of us in this web of people, Likewise, there isa path between any two neurons in ou brain, the world, between any two chemicals between any two companies in our body, Nothing is ighh Paul Erdés and Alfeéd Rényi cold us why: It requires, intance per person, oné this rerconnected web of ‘one per node to stay connected. One ac least one other neuron for each neuron in the brain, the al participate in at least one reaction for each chemical in our body, trade with at least one other company in the business world. One is the threshold. If nodes have less than one connection on average, then our network breaks inco tiny noncommunicating clusters. If there is more mnncetion per node, chat danger becomes remore zyeedly and excravagantly exceeds the one-link mini- estimate that we know between 200 and 5,000 people neuron is connected to dozens of others, some to thousands. Each company and customers; some of the biggest have links to millions. In out body, most molecules take part in far more than a single reaction—some, like ‘water, in hundreds. Thus, weal networks not only are connected but are well beyond the threshold of one. Random network theory cells us that as the average number of links per node increases beyond the critical cone, the number of nodes left out of the giant cluscer decreases exponen: tially. That is, the more links we ack, the harder itis to find a nace th remains isolated. Nature does not take risks by staying close to the threshold, Ir well surpasses it. Consequently, the networks around us are not just webs. They are very dense networks from which nothing can es- cape and within whieh every node is navigable, This is why there are no islands of people completely isolated from society at large and why all molecules in our body are integrated into a single co This is why the Apostle Pauls message reac why MafiaRoy made headlines: Along t fected millions 7. Ends and Rényi’s discovery of this very special moment when a giant cluster emerges through a phase or percolation tra event in graph theory, but not because tion that only one acquaintance is required to form a society. Rather, it was largely because, before Endéis and Rényi, graph theory had not dealt parties, social networbp, o random graphs, It focused vely on regular graphs, which contain no ambi their structure. But when it comes to such complex systems as th temet or the cell, regular graphs are the exception rather thi norm, Erdés and Rényi acknowledged for the frst rime that real gr from social networks to phone lines, are not nice and regular. They are hopelessly complicated. Humbled by their complexity, the two assumed that these necworks are random. retrospect, itis not surprising that this unlikely pair of mathe- und a respectable field of mathernaties imaticians were the onesness were very es. Though Rényi years younger than Erdés, they knew each other thanks to the friendship between their parents back in Budapest. By the time they started working together, after meeting up in Amsterdam in 1948, both had lived through rather tumultuous times. Subject to the Numenas Clausus laws the number of Jews admitted to university, Rényi had shipyard 1th and Greek competi- tion, he was allowed n 1939. Soon after finishing ed labor, from which he his mathemat somehow with Rényi’ tes activities during the war, deeply admired and respected him. Rényi had boldly disguised himself in the uniform of the Hungarian fascists, Nyilas, to help bis friends escape the concentration camps. According to one story, Rényi enteted the Budapest ghetto dressed as a Nyilas soldier and nanaged t0 escort his parents out, He also lived for years in Nazi- controlled Budapest using false documents. Only those aware of the re- alites of the Nazi terror could cvuly appre 1ge needed to perform thes Noe surprisingly, Rényi’s ability co focus on mathe- matics was highly constrained until the end of the war, when in 1946 e traveled to Leningrad to continue his studies. There his creativity exploded. He not only leamed and absorbed number theory in record time, despite his limited Russian language skills, but also proved some fundamental theorems on one of the notoriously difficult problems of the cox number theory, the Goldbach conjecture. Thus, when he met Erdés 1g young mathematician and fatnily friend but a well-known scientist with an two years later in Amsterdam, he was no longer the aspii international reputati Exdés by then had already developed his eradematk traveling- mathematician lifestyle. He would show up at his colleagues doorstey proclaim, “My brain is open,” an invitation to join in. ireless p fer came from the University of Notre Dame, in South Bend, Indiana ‘Amold Ross, at that rime the chaieman of the suit of mathematical truth. His only permanent job of The Random Universe wu fered Ends a visiting professorship on very generous te come and go as he pleased, since he had an assistant wh up che lectures where he left them off. ‘A Catholic 8 college, Notre Dame was not the prominent university it would become decades later. Nevertheless it offered ErdGs a quiet and comfortable work environment and the opportunity for fre- gues, which Erds, with his quent discussions with his priest col (Once asked about his time there, he remarked tongue-in-cheel are too many plus signs,” a reference to the numerous cnucifines a campus. When Notre Dame event into a permanent one, on the same refused. Perhaps losing the randomness and characterized hi predictability that had if was too much for him to fathom. 8. ‘The Amstentam meeting between Erdés and Rényi was the stact of a very close friendship and collaboration that resulted in over thietyj publications before Rényis early death at the age of forty-nine ‘Among these 8 were the eight legendary papers on theory. The first wed more than a decade after the Amst meeting, addressed for the first time the important q araphs form. Their use of randomness to tackle graph tl is most evident when we look at how many links nodes or network, Regular graphs are unique in that each ni inks, or in a hexagonal lattice exactly three others. Such regularity is clea ndomly; thus all nodes have the same just a5 in Las Vegas, where supposedly we al chance of hitting the jackpot. At the end of the day2 LINKED few of our fellow gamblers walk away richer, Similarly, if we place the links randomly in a graph, some nodes will get more Links than others, Some might even have bad luck and get nothing for a while, The ran- dom world of Erdés and Rényi can be simultaneously unfair and gener- cous: It can make some poor and others rich, Yer a far-reaching predic: tion of Erdés and Rényi’s theory tells us that this only appears to be so. f the network is large, despite the links' completely random placement, mnost all nodes will have approximately the same rumber of links ‘One way to see this isto interview all guests as they leave the cock- tail party, asking them how many acquaintances they made, When ceverybudy leaves, we can draw a histogram by plotting how many of the guests have one, two, or exactly k new acquaintances. For the random work model of Erdés and Rényi the shape of the histogram was de- red and proved exactly in 1982 by one of Erdds’s students, Béla Bol- lobis, professor of mathematics at the University of Memphis in the United States and Trinity College in the United Kingdom. The resule shows that the histogram follows a Poisson distribution, which has re properties that will follow us throughout this book. A Poisson distribution has @ prominent peak, indicating that the majority of nodes have the same number of links as the average node does. On the two sides of the peak the distribution rapidly diminishes, making significant deviations from the average extremely rare, ‘Translated back to a society of 6 billion people, a Poisson distr ton tells us that most of us have roughly the same number of friends and acquaintances. It predicts chat itis exponentially rare to find someone deviates from the average by having considerably more or fewer random graph theory predicts jy, we end up with an extremely wh Jinks than the average person, There that if we assign social links rand democratic society, where all of us are average and very few deviate from the norm to be extremely social or utterly asocial types. We obeain a network with a very uniform fabric in which the mean is the norm. Erdos and Rényi’s random universe is dominated by averages. It pre dlicts that most people have roughly the same number of acquaintances, most neurons connect roughly most companies trade with roughly the same number of other compa- the same number of other neurons; ‘The Random Universe 2B che same number of visitors he long run no node is 9. The random network theory of Erdés and Rényi has dominated scien- tific thinking about networks since its introduction in 1959, Ie created several paradigms that are consciously or unconscioush the iminds of everyone who deals wich randomness. If a network was t simple terms, it urged scribe it as random. Sure enough, so the cell, communication networks, and che economy are alll complex enough to fir the bill You may be thinking that there is something fishy about this ean- dom universe, in which all nodes are equal. Would I be able to write this book if the molecules in my body decided to react to each other randomly? Would there be nations, states, schools, and churches or any other manifes of social order if people interacted with each other fave an economy if companies se- umers randomly, replacing their salespeople with mil- lions of dice? Most of us feel that we do not live in such a random world—that there has to he some order behind these complex systems. Why, then, would wo such unparalleled intellects as Erdés and Rényi choose to model the emergence of networks as a completely ran- down provess? The answer is i works than by the me ity to faithfully caprure the webs nature created around the sure, in their seminal 1959 paper they did mention that tion of certain communication nets (railway, road or systems, ete.).” But, despite this brief journe} work was motivated by a deep curiosity ipths of the problem rather than by its applications. nto tl thisPr LINKED __THE THIRD LINK to a new world, whose mathematical beauty and consistency was the iain driving force behind the subsequent work in graph theory. Until recently we had no alternative for describing our interlinked i if y ; universe. Thus random networks came to dominate our ideas on 1 Six D egrees of S eparation work modeling. Complex real networks were viewed as fundamental random. is holds the record for suggesting good px offered monetary rewards for sol proofs to problems that he found inceresting—$5 for a problem he sidered simple, $500 for a truly difficult one. And he would happi if the proof was delivered. Never mind that often a $1 problem ‘ured out to be more difficult than a $500 one. The lucky mathemat ‘cians who eared one of his rewards never cashed his checks anywa them framed them, The reward was a i ae spiritual value. Let us follow Ends What do real neo would never have s unique answer. And “Thus it could not possibly be from the Transfin depository in Enis’: world of all good mathematical proofs and theo- tems. But though the question might not have won his approv coming chapters, we will see that it makes a huge difference outside world of mather: rinting was also disappearing when the first serious reviews appeared in newspapers around the country. By then é entered the hospital, given birth to Paul, and gone hon iscover that her two older daughters were the vi fever epidemic that was tearing through Budapest. f the city’s many personal tragedies, enthusiasm is too broad. It may not even have a cly we can never offer a rigorous proof. virtually unknown poet and writer, invented what caricature. The volume is a collection of p ics. The authors he most venomously parodied are kn 36 LINKED Igy irtok ti is one of che most read books in Hungarian history. Ie yy an instant celebrity. Never again did he have to wait for com wherever he stopped for him, He wt lows of the Cent they suddenly stopped, twmed, jes in a new aquarium ‘Almost two decades after ey ito, that the seventeen-yearold Erdés was lecturing about che Pychage shoe store a few streets away from the Ca book, Minden masképpen van (Every- theorem i Karinthy published his forty-six ing Is Different), ac recognized as the genius of was st and guarantee his place among cern that Karinthy was sat drew quick bucks. Karinthy, whose incredibly was spent between coffee wg out his unique talen’ fc and waited tome. The short story col- ypest and cannot “Lanestemek,” ot “Chai King Gustav, who in cum is 2 Ste Depees of Seension es of Sep n a tennis champion Korinthy’s character: Remarking that “The worker knows the manager is on friendly terms with the general last year became good friends wi wered in 1967, by Si ley Milgram, a Harvard professor who tumed the concept intoa mu ebrated, groundbreaking study on our interconnect i ner of experimental ly debated experiments prob- obedience to authority and personal conscience. in the United States. The q many acquaintances would ita LINKED vaguely ‘out there,’ on che Great Plains or somewhere.” There was little consensus about how many links it would take to connect people from these remote areas. Milgeam himself pointed out in 1969, “Recently | asked a person of intelligence how many steps he thought it would take, and he said that it would require 100 intermediate persons, or more, to move from Nebraska to Sharon. Milgratn’s experiment entailed sending letters to randomly chosen residents of Wichita and Omaha asking them to participate in a study of social contact in American society. The letter contained a short summary of the study’ purpose, a photograph, and the name and ad- dress of and other information about one of the target persons, with the following four-step instructions: HOW TO TAKE PART IN THIS STUDY 1. ADD YOUR NAME TO THE ROSTER AT THE BOT- TOM OF THIS SHEET, so that the next person who re- ceives this letter will know who it came fro DETACH ONE POSTCARD. FILL IT OUT AND RE- TURN IT TO HARVARD UNIVERSITY. No stamp is important, I allows us to keep needed. The postcard is ver track of the progress of the folder as it moves toward get person. 3. IF YOU KNOW THE TARGET PERSON ON A PER- SONAL BASIS, MAIL THIS FOLDER DIRECTLY TO HIM (HER). Do 1 the target person and know each other on a fist name basis. 4. IF YOU DO NOT KNOW THE TARGET PERSON ON A PERSONAL BASIS, DO NOT TRY TO CONTACT HIM DIRECTLY. INSTEAD, MAIL THIS FOLDER (POST- CARDS AND ALL) TO A PERSONAL ACQUAIN- TANCE WHO IS MORE LIKELY THAN YOU TO KNOW THE TARGET PERSON. You may send the folder ly if you have previoush Six Dope of Sepnotin » twa friend, relative or acquaintance, but it must be some you know on a first name basis. wn had a pressing concem: Would any of the letters mak to the target? Ifthe number of links was indeed around one his friend guessed, then the experiment vr always someone along such a long chain who does not couperate. It therefore a pleasant surprise when wit rived, passing through only two intermediate links! This would turn out to be the shortest path ever recorded, bur eventually 42 of the 160 let: ters made it back, some rei dozen intermediates. These completed chains allowed Milgram to determine the number of people ‘get the letter ro the target. He found that the median num- we of intermediate persons was 5.5, a very small number indeed—and coincidentally, amazingl 6, however, and you get the famous "six degrees of separation As Thomas Blas, a social psychologist who has devoted the teen years to in-depth research on the life and work of Stanley pointed out to me, Milgram himself never used the phras separation.” John Guate originated th is brilliant 1991 ply that on Broadway, the the play, Ousa (pl ‘Stockard Channing in the movie), musing about our interconnectedness, tells her daughter, “Everybody on this planet is separated by only six ‘other people. Six degrees of separation. Between us and everybody else fon this planet. The president of the United States. A gondolier in Venice. ... It not just the big names. It’s anyone. A native in a rain for- est. A Tierra del Fuegan. An Eskimo. I am bound to everyone on this planet by a trail of six people. Ita profound thought. ... How every per- son is a new door opening up into other worlds.” ly fi a few days the first levter ar- required lose to Karinthy’s suggestion. Round it up to degrees of smack ely was in Wichita and Omaha to “over here” however, six myth was bom. Because more people watch movies than reads papers, Guare’s version fas prevailed in popular thought» LINKED Six degrees of separation is intriguing because it suggests that, de- society's enormous size, it can easily be navigated by following lion nodes spite v person to another—a network of six xdes are on average six links from each other. Per- haps we should he surprised that there i a path betwer people. Yet we saw in the previous chapter that being connected re uites very little—barely more than one social link per person. AS we all have many more than one link, each of us isa part of the giant net- in which any pair of n ay OW work that we call society. Stanley Milgram awakened us to the fact that not only are we con nected, but we live in a world in which no one is more than a few shakes from anyone else, That is, we live in a small world. Our world is small because society is a very dense web, We have far more friends than the critical one needed to keep us connected. Yet is six de- es something uniquely human, tied somehow to our desire to forma nks? Or do other kinds of networks look the same? Answers to these questions surfaced only a few years ayo. We now know that social aly small worlds, wetworks are not the 2. “Suppose all the information stored on computers everywhere were All the best informacion in every computer at CERN and Id be available to me and anyone else. There would be a single global information space.” This was the dream of Tim Berners-Lee in 1980 while working as a programmer at the European Organization for Nuclear Research, commonly known by its French acronyin, CERN, in Geneva, Switzerland. To turn bis dream into re- ality, he wrote a program that allowed computers to share tion~to link to each other. By inventing the links, Bemers-Lee re- leased a genie whose existence had been unknown to us. In less than ren years the genie turned into the World Wide Web, one of the le networks. It is a virtual network whose ave itall: news, movies, gossip, maps, pic- and books. If it can be written, drawn, or est ever les are W tures, recipes, Six Degrees of 4 chances are there is already a node on the Web con- taining it in some form, ‘The power of the Web the uniform resou (URLs) that all -k of a mouse from one page to another. They allow us to sur, locate, and string togethe: ‘These links tum the collection of a huge net work spun together by mouse clicks. They are the stitches chat keep the fabri on society together. Remove the links, and! spectacularly vanish, Huge inaecessble darabases w nth locators ts to move with ch formation, be left behit How large is the Web today? How many Web documents and link are out there? Until recently no one knew for sure—there’s no sit organization to keep track of al Lawrence and Lee Giles, wor Pr , che contemporary ruins of an interconnected work! the nodes and links. [t was Steve at the NEC Research Inseituee at \eeton, who took up this unique challenge in 1998. Their measute- rents indicaved thar in 1999 the Web had close to a jents—not bad fot a virtual society born less than a decade earlier. Considering thar ie grows much faster than human society, chances are that by the time this book is published there will be more Web docu ments than people on Earth, But the real issue isn’t the overall size of the Web, Ir’ the distance between any two documents. How many clicks does it take to gee from the home page ofa high-s in Omaha to the Webpage of « m nodes, conld the Web be a stockbroker? Desp The answ uestion is not irrelevant to anybody surfs the Web. If Webpages are thousands of clicks from each is hopeless to find any document without a search engine. t the Web was not a small world would also indicare that the networks behind society and the ly different. If that were the case, 0 fully understand networks we would need to understand why and how this difference emerges. There- fore, atthe end of 1996 I set out with Reka Albert, a Ph.D. student, and pwoong Jeong, a postdoctoral research group ar the physics dep: Dame—to grasp the sie of the wor ion docu e universe were fundamen-2 LINKED ‘Six Depress of Separation 33 ly an inven inforrnat feom me. Yet going from my Webpage to th tnajor would often requi is chat, taken together, lar map for sociery, it we ud personal interests and char could say there are eleven degrees of separation at Notre Dame However, the Webpages within our unive would be a must-use tool for everyone from ps represent only a tiny subset of the World Wide Web. TI epidemiologists. OF course, such a social sea 1999 was atleast 3,000 times larger. Would the distance would take at least a between two randomly selected nodes on the Wide Web was also 3,000 times longer than the eleven clicks out measurements indi- it take a full 33,000 clicks to get from one page to is question we needed a map of the problem was that nobody had one. Even the largest ‘h engines that tirelessly scan the Web with thousands of comput- ers have managed to cover less than 15 percent of Could we determine the sepa Web! To answer random systems w (Our approach had a si fe research group clearly could not compet created a robot to accomplish a more modest gor gave usa map of the nd.edu domain by mapping about 300,000 documen University of Notre Dame, a rather ed ing from philosophi ‘were not concemed about the obtained node-to-node erage separation between is tiny sample. Next we took a sl get piece, wi nodes, and determined the separation again. We repeat gest systems our computer reproducible the separation on the ber of nek he nodes 2b, mol the sraduace students hsns LINKED ed that the diameter of the hr say: nineteen degrees of Web was 18.59, ion. Whi the Web is a small world. Any document is on av icks away from any more fundament the globe. This suspicion was confirmed by which demonstrat a chance to study. Indeed, species in ks away from each other; mol average by three chemical reactions; in different fields of science are separated by four to six coau- the C. elegans worm are every network sci separated by fourteen synaps solute cecord at nin lay a separation between two and fourteen. Nineteen degrees ly far This is nc the case, that huge networks, with hundreds of of nodes, collapse, displaying sepa. rnoxles they have. Our society, a The Web, w a network 10 nodes, has a separation of nineteen, The Intemet, sands of routers, has a separation of ren. Seen f fous chapter, we saw that random ne node to form a giant cluster. The ques large. But as we add more the distance between the nodes suddenly collapses. Consider a network in whi node we can reach k other nodes wit the nodes have on average k links. This means that from a term present in the for number is rather smal The ten-based logat if we have ewo networks, + one. The logarithm shrinks the huge ating the small worlds around us. ngs he had arranged ahead close friend and literary rival wve got to nun home because Karinthy promised and perhaps he forgot that he promised, and he wil estingly, six degrees appears t an average her, the separa I be only two degrees higher than the separa networks, cre people of his generation, Karinthy was time. DesSix Deas of Separation n 36 press and scientific texts J the six degrees concept rated to the United father or uncles, who oft never know, but it ¢ ion of the idea of six de (Or did he hear about it ftom others in the coffee cory suggests? We will perhaps never have an answer, im of events to speculate on the subsequent The sixjnineteen degrees phrase is deeply mi that easy to find in a small wor from document six/nine- In other words, number or a very ins" appeared, had a wnique to have been good friends with many writ ‘ea, Web document is around seven, this we can follow only seven links from the first page, away, in principle we w times more than the tot dict that Karinthy’s j have sheink even gigant mn has an easy resolut back wo pages ic even if it takes only one second take over 300 n ‘years to get s away! Neverthel never know if they amused themselves foofs and theorems. Bu inks do not stop there. Stanley -nts uncovering the 5.5 links in 1967, five-link conjecture and almost a decade af- search engines, The ter Bris andl Rényi seem to have been aware of ry and most likely had nev e heen influenced by the work of Ithe ick, of course, , we avoid having to check all the grees and can zero in on the desired page% LINKED cs. While ehis method seems to be the mest efficient, it is always possible within 2 few c almost always fails to find the shortest path. Indeed, that the wrestler whose Webpage we bypassed balances his tough guy image with a link to the best Picasso site, But mast people looking for Picasso would ignore the wrestler’ fink and eventually follow a longer ion, The computer, having no taste or bias (yet), wrestler, modem art, and inks to all of them, By inevitably locare the shortest one, pages fundamental problem with path co the dest thew through with eq frog's love life pages, pragma ag all the possible paths, ependent of the content of the intermedi Finding Picasso on the Web highlights six degrees: Milgram’s method overestimated the shortest distance be tween two people in the United States. Six degrees is really an upper ber of paths with widely different t. There is an enormous lengths between any two people. Milgeam’s subjects were never aware of the shortest path to their tanger. This is like b where we can see only the corridors and doors next t0 us. Even if we have a compass and we know that the exit is toward the north, finding it could be woefully inefficient and time-consuming. With a map of the ‘maze in hand, we could be out in five minutes. Similarly, Milgram’s let- te1s would have followed the shortest path between Omaha and Boston if all participants had had a map that compiled the social links of sans. Lacking such @ map, they forwarded the message to the tight direc the ng lose in a huge maze all Ame those that they shoughe were most tive, As most of us do not know our senator on a first name basis, we would try to find somebody who does and who would be willing to bro- he president. That would take at least three hand- 1 have no clue that the gentleman kera meeting wi shakes. In the mea ne, yOu mi sat next to a few days earlier at a ty wene ro school with, the yo the president. Thus in reality you are only nwo degrees away president. Similarly, the paths recorded by Milgram’s experi Six Dees of Se x invariably longer than the shortest possible. Thus, the real separation d Te must be shorter than six—per- haps shorter than Karinthy’s five. We don’t have a social search engine, so we may never know the real number with total certainty. 6. Six degrees is the product of our modern society—a result of o tence on keeping in touch. It is aided by our relatively newfound a lage we've grown us The global v humans. The anc left behind in the the gold mines of the Rocky Mountains ic was impossible to reach loved ones separated by oceans and continents. No postcards, no phone calls. In the subtle social network of chose days, it was rather difficult co activate the links that had been broken when people changed in this century as the mai travel demolished barrie grants to America can links to the people they leave behind. We can and do keep in touch. I keep ttack of my relatives and friends even if they are as far away as Korea ot eastem Europe. The world has collapsed irreversibly in the twentieth century. And it is un- dergoing yet another implosion right now, as the Internet reaches 8 of most Americans lost contact with those they every comer of the world. Though we are nineteen ¢ everybody on che Web, we are only one click away f ‘They might have hopped three cities person. But no matter where t ks that would have died out a hundred years ago are kept alive and can be easily activated. ‘The number of social links an individual can actively maintain has increased dramatically, «legroes of separation. Milgram estimated six. Kat be much closer these days raging down the ive, We couldra LINKED are a generic property of networks in general. Short nystery of our society or something peculiar about the their structure—it ber of Web- rent from the separati Web: Most net simply doesn’t pages or friends. The resulting small worlds are rather Euclidean world to which we are accustomed and in which distances are ability to reach people has less and less to do with, the physical distance between us. Discovering common acquaintances with perfect strangers on worldwide trips repeatedly reminds us that some people on the other side of the planet are often closer along the social network than people living next door. Navigating this non-Euclidean ‘measured THE FOURTH LINK Small Worlds Witen MARK GRANOVETTER SUI 2 graduate ‘manuscript. Hat ITTED his first-ever paper fot pul Harvard, but he had high and MIT were the hotbeds of the new ideas. A series of le rison White, a pioneer of the network perspet of Boston, in the late sixties it was a working. neighborhood. Aiming to find out how people “netw: current job. Was not a friend. It was a2 LINKED «it out in August 1969 t0 the American Sociological inber he receive: iped, Granovetter did not In 1972 he submitted a somewhat shortened version of the manuscript American Journal of Sociology. This time he to a different journal, had better luck, and the paper was finally published in May 1973, four years after its first submission. Today Granovert of Weak Ties, is recognized as one of the most in 1986, In The Strength of Weak Ties Gra sounds preposterous at first: When news, launching a restaurant, or spreading the latest fad, our weak so- les are more important than our cherished st comes to finding a job, getting is rather generic, ‘most of whom are in touch with one another—a densely Moreover, Ego will have a collection of of these acquain- own right and bi tances, however, is likely to have close friends in therefore to be enmeshed in a closely knit clump of social cone different from Ego’. Hidden within Granovetter's argument there isan image of a society that is very different from the random universe Erdés and Rényi de- picted. nw society is structured into highly connected clusters, or close-knit circles of friends, in which everybody knows everybody else. A few external links connecting these them from being isolated from the rest of the world. If Grano then the network describing our society has a rather pec It isa collection of complete graphs, tiny clusters in which each node nected to complete graphs are pelos ies, shown as bold lines. Weak Weak ties play an important role in an; spreading rumort to getting 2 job. ‘Weak ties play a crucial role in our a outside world. Often our close friends can offer us job. They move in the same circles we do and are inevi the same information. To get new information, we have to ac weak ties. Indeed, managerial workers are more likely to hear about fough weak ties (27.8 percent of the cases) than thro In the Erdés-Rényi social universe the likelihvod » closest friends knowing each other is“ LINKED the same as the chance African tribal chief, Bur that is not what our society looks like, In most cases two good friends know each other’ friends. They often go to the same parties, frequent the same pubs, and watch the same movies. The stronger the tie between two people, the larger the overlap between, their circles of friends. Though Granovecter’s argument about the im- portance of weak ties at first glance may seem counterintuitive and even paradoxical, it formulates a situple truth about our social organiza- tion, Granovetter’s society, a fragmented wel ugh weak cies, experience picture offered by Erdéis and Rényi. To ace of society, somehow the theory of ran- ters communicating th ver than the completely rand ly understand the scru dom networks had to be reconciled with the clustered reality depicted by Granovetter. It took almost three decades to accomplish this. Inter- estingly, the clue for a possible solution did not come from sociology or graph theory, 1. Across from the Central Café, af descend chr case into one of the elite studio theaters of Budapest. Appropriately named the Kamra, or Closet, since it holds only about ten actors on the stage and a hundred people in the audience, the seats at its perform. ances are highly coveted by those familiar with Budapest's burgeoning theater life, The last performance ! watched in the Kamra did away with the curtain to save space, forcing the audience to guess exactly when the play ended. Ic was hard to miss, though, as suddenly everyone around me burst into tumultuous applause, which was echoed and am- ed by the black walls of the sma vl cavern. In no time Our palms came to- ly the same moment, united by @ mysterious force that paces from Ka foor and narrow gether at prec urged us to clap in phase, as if following the baton of an inv cor. As the actors bowed, disappeared backstage, and reappeared, ced te ble can. ic applause grew even stronger: Lis syncheony dis Small Works 6 porarily as the clapping gathered speed and strength, only to reappear in full force a few seconds Synchronized clapping is hardly unique to the tiny Kamra of Buclapest. Iris @ regular occurrence after theater performances, certs, or sports events in eastemn Europe and is occasionally heard over the world. It spontaneous Square Garden when the audience unconsciously sy ping to honor Wayne Gretzky, the legendary hockey player, before his retirement from the New York Range: terious, synchronized clapping offers a wonderful example of self-organ- ization following strict laws extensively researched by physicists and mathematicians. Some species of fireflies are also subject to these laws. In souchwest Asia they gather by the millions around tall mangrove periodically, Then suddenly, all the fireflies begin to switch their fluorescent tails on and off at exactly the same moment, urning the beacon-shaped tree into a huge pulsing light bulb visible for miles. A subtle urge to synchronize is pervasive in nature. Indee: drives the firing of thousands of pacemaker cells in into synchrony the menstrual cycles of women wi long periods of time. Duncan Watts, workis Comell University in the emerged, for examy 1999, Spontaneous and eys- art and brings ive together for on his Ph.D. in applied machem 1990s, was asked to invest the spotlight by carefully adjusting their chirp to match chat of their neigl them together and from the cacophony i symph often enjoy on the back porch on. Watts does not fit che sterec matician. Possessing an agi rening to the other crickets and ceflect on iced, the cricket study tured him into a student of soe: and eventually a sociologist, a transformation made of when he was offered a professorship in the department of socie Columbia University.6 LINKED ing to grasp how crickets synchronize, Watts was struck by the concept of six degrees of separation, planted in his head by his father versation. People wonder about thi six degrees all the time, but such coffeehouse philosophy rarely leads to serious research. Warts tho yy comprehend synchronize he needed to understand how they pay attention to ea other. Do all crickets listen to every other cricket that is chirping? Or do some pick a favorite one and try 0 synchronize with that one only? ‘What is the structure of the network encoding how crickets, or people, influence each inding himself thinking more and mote about ‘networks, and less and less about crickets, Watts approached his Ph.D. hematics professor at ‘chaos and synchto- lea to her? advisor, Steven Strognts for advice. An applied Cornell with a distinguished record in the stuly nization, Strogatz is not known for allowing an unconventional pass him by. Soon they were of to uncharted terrcoris, taking networks beyond the boundaries set by Ends and Rényj. ‘Watts started his voyage into networks with a simple question: What is che likelihood thar two friends of mine know each other? As has a clear answer in random network jomly, my two best friends wwe have just seen, this q theory. Because the nodes are have the same chance of knowing each other as do a gondolier from Venice and an Eskimo fisherman, Clearly, as Granovetter argued twenty-five years earlier, that is not how society works. We are part of clusters in which everybody knows everybody else, ‘Thus my two best friends will inevitably know each other. To gather evidence about the clustered nature of society in terms that are acceptable to a mathe- physicist we need to be able to measure clustering, To Watts and Swogac: introduced a quantity called the cias- are all friends with each other as wel k, obtaining altogether six friends ks. Chances are, wur friends are not friends with each other. with a however, that some of Thea the real count « of friends is 0.66, ob- thy between your friends Small Words ” (four) by the number of links thar they cou friends with each other (six) The clustering coefficient tells you how closely knit your friends is. A number close to 1.0 means that all your friends are good friends with each other. On the other hand, if the clustering coefficient is zero, then you are che only person who holds your friends together, as they do not seem to enjoy each other’s company. Granoverters vision ay hi of society includes clustering coefficient. To obtain quantitative evidence tha society is indeed full of such clusters, we would need to measure the clustering coefficient for each person on Earth. As there are no maps t who is connected to whom and who is friends with whom, this isan im- possible task. Fortunately, however, a peculiar subset of society regularly publishes their social ties. We can therefore look for . is unusual gr 2. ‘Today Paul Erdés is famous not only for his countless theorems and proofs, but also for a concept he inspired: the Erdés number, Erd6s pub- lished over 1,500 papers with 507 coauthors. leled honor isan unpar: to be counted among his hundreds of coauthors. Short of ¢ bbe only two links from him. To keep track have Erdés number one, Those who wrote a paper with thor have Erdés number two, and so on. A low Erd6s number is matter of pride—so much so that some suspect that counterfeit collaborations may have been concocted after Erdés's death in 1996 to lower someone's Jerry Grossmann, a Rochester, Michigan, i the Ends numbers for thousands46 LIN of mathematicians, allowing any published mathematician to calculate his or her own, Mose mathematicians turn out to have rather small Erdés numbers, being typically ewo to five steps from Ends, But Erdos’ ‘well beyond his immediate field. Economists, physicist scientists also can be easily connected to him. Einstein has Exdés number two, Paul Samuelson, the Nobel prize-winning economist, has five. James D. Watson, the codiscoverer of the double helix, has eight, Noarn Chomsky, the famous linguist, has four. Even: Will founder of Microsof, who hes published litle science, has an Erds num: ber of fout. My Erdis nuraber is also four: Erdés wrote a paper with Jos who had George H. Weiss among his seventeen coat thors, who in tum worked with H. Eugene Stanley, my Ph.D. advisor, with whom I have coauthored a book and over a dozen scientific articles. “The very existence of the Erdés number det entific community forms a highly interconnected network in which a scientists are linked to each other through the papers they have writ- ten. The smallness of most Erdés numbers indicates that this web of si- ence truly isa small world. As it only eazely happens that the authors of personally know each other, coauthorships repte- aks. Consequently, the web of science is @ smal ial network, with the unique feature that its at researchers ca rates that the sci- a publication do w sent strong soci scale prototype of 04 links are regularly publi pees on a certain copi puterized databases; this automat cof the social and professional links between sciemtsts. We c igial record therefore use them to study the structure of the collaboration network ‘This is exactly what a group of us did in the spring of 2000. Tamas Vicsek, a distinguished researcher and chairman of the department of biological physics at Batvés University in Budapest during the aca demic year 1999-2000, organized a year-long program focusing on bio- logical physics at xe of Advanced Study, located in a charm ing medieval Buda castle overlooking the Danube. Zoltén Néda, a nia, was one of the participants, and he had at that time a masters student in Néda’s eo _roup, Also joining the team was Andris Schubert, an expert on metrics working for the Hungarian Academy, who had access fot search purposes to large coauthorship databases. To Ravase, Néda, Schubert, and Hawoong Jeong, we linked cians through papers published between 1991 and 1998, reasse the highly interwoven network of 70.975 1 Eda cient, approximately 10%. However, our measurements indicated th the clustering coefficient for the real collaboration network is a 10,000 times larger chan that, proving that mathematicians do not pick their collaborators randoraly: Rather, they form a highly clustered net lar co the one spotted by Granovetter in Unknown to us, Mark Newman, a physicist at the Santa Fe Insti in particular ing questions computer scientists —ask: lar to the ones we were asking in Budapest whose expertise ranges from random systems to species ext ecosystems, recognized the unique opportunity our comput coffers us to finally understand networks. Before turning. tocol networks, now considered hhad already written several papers on small world jing owt sults, he.posted on the Intemet his first paper on co nan’s paper proved that the day-to-day lassics. As our computer was ch W occasional weak ties, His work, combined with our own, of: fered quantitative evidence for something we had felt to be tru along but that was notoriously difficult to measure before computers: Clustering is indeed present in social systems.0 LINKED its two-to-three-week life span, shining career since Sydney Brenner the Molecular Sciences I a pig” of molecular biology ceeded in figuring out the precise wiring of Is which neurons ate connected chan woul! be the case in a random same pattern when studying the are generators and transformers power does the collaboration network of Hollywood actors, a network 1 will discuss in detail in the next chapter. clustering generated by Watts and jentific community has subse- ent on the Web; we have spot computers on the Intemet; economists have decectr Sal Work st (left). To make this world a small one, a few extra links were added, connecting These long-range links offer the erucial short drastically shorte randomly selected nodes (right cuts between ng the average separation be tween all nodes and cell biologists have learned that it characterizes ecules packed within a cell. The di lly ek property of complex necworks and posed the fist serio view that real networks are fundamentally random. from a work model in their 1998 study published! in Nature. They proposed a id clustering with the com- model that for the first time recone pletely haphazard character of rand32 LINKED nected to each other by three links. Thus, clustering coefficient. Indeed, if all to each other, there woukl be six links the clustering coefficient is ans, To see that network four neighbors, the resulting network hi four neighbors were « between them, Since there are only thre 3/6, oF 015, close to the 0.56 we found for indeed represents significant clustering, consider a rand in which a typical node still has four neighbors but is connected ran- domly to any node in the system. The number of links between my four neighbors now depends on the sive of the network. If I have twelve rnodes, as I do in the figure, the clustering coefficient is 0.33. For 1 lion nodes, however, it drops to four over a Clearly the cluster- of 05, which the new model predicts, is gigantic com- cor pared We however, for the high clustering the model offers us. Our small world is gone. In the model society shown in the figure only my immediate and next nearest neighbors are close to me. erally have to go around the circle, shaking innumerable hands along the way. Indeed, it is easy to check that the shortest path connecting the top node to the bottom one is at least three links long. That may not sound like a lot, but if ! had had the patience (and space) to draw 6 billion ng the same ring, each connected to its immediate and next ld re- ‘ety on a circle not very large world, as we have links to distant people around the globe. Each cof us has friends who do not live next door to us. IFL want to find a path alia, Iwill not go door to door, since sooner or later To get to somebody on the other side of the circle, I would the circle wi from high school moved to Sydney a few years ago. Thus all 1 need to dois find a link to my Australian tatget through the increasingly dense fricndship links my friend is creating around him right now. A realistic allow for distant links. We can easily lesctibed above by ad id the circle, That is, nod a few links to some yy eo nodes ly chosen nodes aro Small Worlds 3 neighbors will be a ot closer to each other too. dom links, {can bring all the nodes very close together. ‘The surprising finding of Watts and Strogats is that even a few extr links are suffi tween the nodes. These few tering coefficient, Yet thanks to the long bridges necting nodes on the opposite side of the circle, I nodes spectacularly collapses. The model's ability to severely de- ease the separati while keeping the clustering coeff lent practi hanged indicates that we can afford to be very provinci choosing our friends, as Long as a small fraction of the population has some long-range links. According to the insight provided by this simple ‘model, six degrees are rooted in the fact that a few people have friends and relatives that do not live next door any longer. These offer us short paths to people in very re networks do not need to be full of ran features. A few su inks will do che 5. The publication of the Watts-Strogat: paper on clustering, two years f- cer Erdés’s death, gatnered enormous interest among physicists and mathematicians alike. First, it formalized Granovetter's vision by offer- ing a model that did display significant clustering. Second, it played a in bringing the small-world problem, a much investigated issue within sociology, to the actention of the physics community. For a short time it seemed as if the more general, cluster: unique rol6 LINKED other. Therefore both models depict a deeply egalitarian society, whose links are ruled by the throw of a dice. When the landmark paper of Watts a Ends and Rényi such hubs are extrem¢ inodel could not accounc for our robot's finding. The fare much better: It, t00, forbids nodes average node has. Something im. from both models, limiting our understand: ‘ed us to search for a better ing gether the random w ry unexpected eurn. We had t ed about networks Je up just about everything we had Hubs and Connectors MALCOLM GLADWELL, A STARE WRITER at the New Yorker mag setibes in his recent book, The Tipping Poi how social you are. He gives you a list: Manhattan phone book and ash know anybody with thar name. M College of M ts in their early twenties, In other words, they typical recorded an average score of about twenty-one people with the same surname as somebody on the educated surprising. But what caught college class, th ip of people of similar age, education, and income the range was enormous: The lowest score was 16 while the highest was 108. Gladwell ended up testing about four hundred people altoy ing a few high scorers in every x 5556 LINKED Connectors a network. They create level ; thought that he was unknown to him, he ing together different rac Wg connectors, Gladw seeing some: phenomenon that ‘was puzaling my rescarch group well before the publication of The Tip- ping Poine. Connectors—nodes with an anomalously large number of ks—are present in very diverse complex systems, ranging from the I property of most net- ss as disparate as biology, comy ace, and ecology. Their discovery has turned everything we thought we knew about networks on its head. Clustering, exposed the first crack in the Erdos-Rényi random worldview. The simple model of Watts and Scrogats, discussed in the previous chapter, the day, reconciling the aration. The connectors are the final blow to both models. Accounting for these highly connected nodes requires abandoning once and for .¢ random worldview. sav of friends with six degrees of sep 1 Cyberspace embodies the ultimate freedom of speech. Some may be of- ed, others may love it, but the content of a Webpage is hard to cen: ce posted, itis available to hundreds of millions of people. This sense of expression, coupled with diminished publishing 1e Web the ultimate forum of democracy; everybody's ness magazines. Ife is not. Th ¢ Web. We learned that the topology Hubs and Connectors 37 ‘When it comes to the Web, the key question is no longer wl your views can be published. They can. Once publ instantaneously available to anyone around the world wit connection. Rather, faced with a jungle of abi question ig, if you post informa In order to be read you have w nent on the Web bi Webpage, ina very shore time everyone would know what you say. But the average Webpage only has about five to seven links my homepage, wor there are about forty run across ink poi Each of us has very different The links we add to our Webpages reflecseo LINKED sample of 203 million Webpages uncovered an even wider spectru The vast majo as many as 90 percent of all documents, have ten oF a few, about three, are referenced by Just a5 in society a few connect ‘of people, we found that the archi inated by a few very highly connected nodes, or hubs. These hubs, of links are held toged The hubs are t sive manner, we somehow create hubs, Websites to which every- links. They are very easy to find, no marter where you are on the Web. Compared co these hubs, the rest of the Web is invisible, For all ced by only one or two other documents do m, Even the search engines ng them as they crawl the Web looking are biased against them, for the hottest new sites. 2. Kevin Bacon's movie The Air Up There was (Craig Fass, Brian Turtle, and Mike Gi Albright College in Reading, Pennsylvania, had a re occurred 10 them hat Bacon had played in so many chat you could connect him g on television the students from Habs and Conmsce ~ prove to the Jon S Much to their wart audience, nay Kevin Bacon, and inect Bacon to any actor Kevin Men. Using the yy, Tom Cruise has Bacon number one. Mike Myer connected to Robert Wagner through hagged Me, Wagner has Bacon number one thanks to Wild ven such historical figures as Charlie Chaplin have a path to Bacon: Chaplin played with Barry Norton in Monsieur Verdowx, who as we already know, is only one link away from Bacon. Thus Charlie Chaplin has Ba- con number three. To further tangle » Paul Beds has 2 Bacon 4. Number, 2 documentary about him in cast was Gene Patterson, which he plays himself laying himself, créme de Ia creme of graph theory, many mathematicians have not a small Erdos number but a small Bacon number too. had two computer Glen Wasson and Br diacelyw INKED. sconds it the game. If you type in the uny two actors I provides the shortest path between her, listing the chain of actors and movies through which they are connected. In no time the Website was ng over 20,000 visits per day, eventually landing on Time maga- ist of top ten sites of 1997. Last time I checked, on August 26, 2001, it had hosted over 13,000 visitors that day alone. ‘We can play the Kevin Bacon game because Hollywood forms a densely interconnected network in which the nodes are actors linked by the jes in which they have appeared. An actor has links to all other 2e xe cast. Thus those who have played in several movies acquite ly. As each actor has an average of twenty-seven links, many han the necessary one t0 make the network fully connected, sie degrees is unavoidable: Each actor can be connected to any other actor through three links on average. Yet, as my research group has noticed when analyzing the actor network, averages do not apply here. As many as 41 percent of actors have fewer than ren links. These are the less known actors whose names appear on the movie screen after you have walked out of the theater. A tiny minority of actors, however, have far more than ten links. John Carradine collected 4,000 links to 1 actors during his prolific career, while Robert Mitchum acted with 2,905 colleagues during his decades on the silver screen. These ex- ceptionally well connected actors are the hubs of Hollywood. Remove a few of them, and suddenly the paths from almost any actor to Bacon will drastically lengthen. ‘On the one hand, an educated guess would be that the actors who have played in the most movies are also the most connected, having the shortest distance to everybody else in Hollywood. This turns out to be true on average: The more movies an actor plays in, the shorter his oF het average distance to his of her peers. On the other hand, the list of to give us the most-connected actors bs the most movies fai we surprises, too, Compiled by Hawoong Jeong, the top ten hey played, umber of movies in whi parentheses the looks like this: Mel Blane (759), Tom Byron (679), Mare Wallice (535), Ron Jeremy (500), Peter North (491), T. T. Boy (449), Tom London Randy West (425), Mike Homer (418), and Joey Silvera (410). ce Bugs Bunny, Woody Woodpecker, Daffy Duck, Porky Pig, ‘Tweety Pie, and Sylvester. And those over fifty years old would have seen Tom London, pethaps th B western movie actor, por- fers, and henchmen. The rest of the ac- eluded us. In the end, after some ned them down. They are all porn stars. ‘This list is perhaps the most vivid demonstration that, when it comes to networks, size does not always matter. Despite the record number of movies porn stars make, they fail to be anywhere near the center of Hollywood. As networks are clustered, nodes that are linked only to nodes in their cluster could have a central role in that subeul- ture or genre, Without links connecting them to the outside world, they can be quite far from nodes in other is rather tray countless sheriffs, ranch. ors on the yy to porn stars to the cast of a Martin Scorses Tarkovsky movie, They simply move in very different w. central position in networks is reserved for those neces that se s taneously part of many large clusters. They are the actors who pages that not only reference modem act, but have links of human inquiry. They are the people who regularly co with people from diverse fields and se nets of networks, equally at home in arts and (Of course, Bacon is a prominent Hollywood actor. He hi over forty-six movies, collecting links to more than 1,800 acto wverage separation from everyone else in Hollywood is 2.79-— most actors are within three links of2 LINKED are so good at the Kevin Bacon game, easily connecting any other actor to him. But is Bacon the most connected actor! When Hawoung Jeory prepared the list of the thousand most-connected actors, the real hubs ‘of Hollywood, it took us # while co find Bacon on it. We saw Rod ‘Sreiger in the number-one spot, with an average distance of 2.53 from everybody else, Donald Pleasence, with a separation of 2.54, was right behind hira. Martin Sheen, Christopher Lee, Charlton Heston took up the next four spots, each, less than 2.57. We read hundreds of names, browsing through dozens of pages, without a sign of Bacon, Eventually we discovered himn towards the end of the list, at che 876th spot. Why do we play the Kevin Bacon game then? Bacon's prominence the publicity offered by the Stewart show. nst actors. Bacon is by no means spe~ cial, Not only is he far from the center of the universe, he's far indeed from the center of Hollywood, 4. A random universe does not support connectors. If society were ran: dom, then in Gladwel ample of four hundsed people, with their average of around 39 social links, the most social person should have far fewer acquaintances than the 118 found. If the Web ‘were a random network, the probability of there being a page with five hundred incoming links would be 10-°—that is, practically zero, indi- cating that hubs are forbidden in a randoraly linked Web. Yet the latest vey, covering less than a fifth of the full Web, found four hun- ich pages and one document with over ¢wo million incoming, ks. The chance of finding such a node in a random network is ler than the chance of locating a particular atom in the universe. If Hollywood forms a random network, Rod Steiger does not exist, as the 1 of having such a well connected actor is about 10-1, which number that it’s hard to come up with @ proper c ineredihly small numbers help explain our surprise con the Web and in Hollywood during our yodest social Hb and Connectors ‘6 carly attempts to understand the structure of real networks. There w: nothing to prepare us for them because they were forbidden by both the Erdés-Rényi and the Warts-Strogatz models. They simply were nor sup posed to exist The discovery that on the Web a few hubs grab most of the links untic search for hubs in many areas. The results are star ing: We now know that Hollywood, the Web, and society are not unique by any means. For example, hubs surface in the cell, work of molecules connected by chemical reactions. A few molecu osphate (ATP), are the Rod Steigers of the cell, participating in a huge number of reactions. On the Inter. -s connecting computers worldwide, a ined to play a crucial role in guaranteeing the In- temet’s robustness against failures, Erdés is a major hub of mathemat ics, a 507 mathematicians have ErdGs number one. According to an ATSC study, a few phone numbers are responsible for an extraordinar- ily high fraction of calls placed or received. While those with a teenager out the iden. tity of some of these phone hubs, the cruth is that telemarketing firms and consumer service numbers are probably the real culprits. Hubs ap- pear in most large complex networks that scientists have been able to study so far. They are ubiquitous, a generic building block of our com- plex, interconnected world such as water or ademosine t1i net, the network of physical w hubs were dere: ‘ng in their homes might have st 5 Lately hubs are enjoying exceptional attention. Celebrating the power of connectors, + Rosen spends several chapters in his book The Anatomy of Buzz categorizing social hubs and inspecting their role in spreading news and hype. Every four years the United States inaugu- rates 8 new social hub—the president. indeed, Franklin Del sevelt’ appointment book had about 22,000 names in it, maki bone of the biggest hubs of his era. Threeos INKED understanding the processes behind many forms of cancer at the molec ular level. Ecologists believe that the hubs of food webs are the key- stone species of an ecosystem, paramount in maintaining the ecosys- » hubs is well deserved. Hubs are special. They dominate the struc networks in which they are present, mak- ing them look like small worlds. Indeed, with links to an unusually large number of nodes, hubs create short paths between any two nodes the system. Consequently, while the average separ randomly selected people on Earth is six, the distance between anybody and a connectors often only one or two. Similarly, while two pages o \e Web are nineteen clicks away, Yahoo.com, a giant hub, is reachable fiom most Webpages in two to three clicks. From the perspective of the bubs the world is “The view that networks are random, held for decades under the in fluence of Erdés and Rényi, has lately been questioned on many fronts. Watts and Strogats's model offered a simple explanation of clustering, bringing random networks and clustering under the same roof. Hubs, however, again challenge the status quo. They cannot be explained by either of the models we have seen so far. Therefore, hubs force us to re- consider our knowledge of networks and to ask three fundamental ques- tions: How do hubs appear? How many of them are expected in a given network! Why did all previous models fail to accoune for them? During the last two years we have answered most of these ques- ‘The attentio feof leed very ti tions. Indeed, we have found thac hubs are not rare accidents of our in- terlinked universe. Instead, they follow strict mathematical laws whose ubiquity and reach challenge us to think very differently about net~ works. Uncovering and explaining these laws has been a fascinating + coaster ride during which we have leaned more about our com nterconnected world than was known in the last hundred years. THE SIXTH LINK The 80/20 Rule \VILFREDO PARETO, THE INFLUENTIAL ITALIAN ECONOMIST, wh le giving atl: in the eatly 1900 at an economies confetence in Geneva, nasser Peatedly and noisily interrupted by his powerful Schmoller. Von Schmoller, whe from Berlin ruled the German academic work patronizing tone, “But are there laws in economics __ Despite his aristocratic upbringing Pareto had title eespoct Pearances, reportedly having written his monumental work Trattauy Sociologia verale while owning a single pair of shoes and sui was therefore easy for him to transform himself into a beggar the next ey and approach von Schmoller on the street. “Please, sir." Pareto fornothing?” cell me where I can find a restaurant where you can eat 'y dear man,” replied van Schmoller, “there are no such restaurants, but there isa place around the corner where you can h: good meal very cheaply laugh fay ss there are laws in econon shes 4 Tasco had tumed his attention to economics after working for two gecades as a raway engineer, Deeply influenced hy the mathemat eauty of Newronian physics, he devoted the rest of his life ies into an exact science, descr ind universality to those 65a LINKED the three-volume n for economists and sociologists (Qutside academia Pareto is best known for one of his empirical at 80 percent of his peas L observer of the Murphy’ Law of management: 80 percent by only 20 percent of the employees, 80 percent of ms are created by only 20 percent of re made during 20 percent of me; are largely irrelevant. Let me 80/20 rule: 80 percent ‘non: In most cases four-fifths of our eff ibute a few more items that approximate robot to map the ind the and Rényi, we rach other ran: ‘Web would look like. Gt ‘expected to find that Webpages ate connected domly. As we discussed in Chapter 2, the nu page should follow a peaked distr you are not a physicist or never heard of power laws. Th Your histogram will have a peak ar less you hang our a molecules in a gas, many peopl In the past few decades sci sion nature ge instead of a bell curve. Power laws are very curves describing our heights. have a peak. Rather, a his ously decreasing cu ates qual6 LINKED very large ones. These extraordinarily large events are simply forbid. den in a bell curve.! Each power law is characterized by a unique exponent, telling us, for example, how many very popular Webpages are out there relative networks the power law describes the often called the degree exponent distribution of incoming links degree distribution; the expo Our measurements indicated 1 ‘on Webpages followed a power law with a unique and w close to wo. A si we looked at outgoing links, the degree exponent this time being ely Laeger.” Our tiny cob offered compelling evidence that lions of Web page creators work together in some magic way to generate a complex Web char defies the ra werse. Their collective action forces the degree distribution to evade the bell curve—a signature of random networks—and to turn the Web into a very peculiar network described by a power law, The robot fa ressing question, however. What was it about the Web that prompted it to defy the strict predictions of random networks? Then we realized that there was another way to approach this problem. Could it be that equally simple laws characterize most com- plex networks and we had not seen them because we had not looked m before? This second th more fruitful. Indeed, a few months later, while analyzing the lar power law was prese: ine of questioning turned out to be fo ‘The 80/20 Rule ° | relationship: The numbe exietly k other actors decays following a power law: Later we net, a professor of physics at Boston University, distribution of citations in physics journ Viewing citations as links of a network whose nodes are publications, Redner’s finding w the citation network is also described law degree distribution, Subsequently, in numerous lange at we and many other scientists have had a chance to in. vestigate, an amazingly simple and consistent pattem has emerged: ‘The number of nodes with exactly k links follows a power law, each with a unique degree exponent that for most systems varies between two and three. work and one described by a power-law degree distribution are best sec by comparing a US. roadmap with an airline routing map. On the roadmap cities are the nodes and the highways connecting them the links. This isa fairly uniform network: Each major city has at least one link to che highway system, and there are no cities served by hundreds of highways. Thus most nodes are fairly similay, with roughly the same ‘number of links. As we saw in Chapter 2, such uniformity is an inher. ent property of random networks with a peaked degree distribution The airline routing map differs drastically from the roadmap. The nodes of this network are airports connected by ditect flights between them, Inspecting the maps displayed in the glossy flight mago placed on the back of each airplane seat, we cannot fail to not few hubs, such as Chieayo, Dallas, Denver, Atlanta, from which flights depart to al ports ace majority» LINKED ‘The 80/20 Rule n Power Law Distnbution way map, where most nodes few hubs connect hundreds of sm: wumerous tiny nodes co ly high number of links. The few nodes to each other are not sufficient to 2 is fully connected. This function is secured by the rel bs that keep real networks from f a random network the peak of None of mde it las rion implies that the inks and that nodes de. rare. Therefore, a random net: work has a characteristic scale is e node and fixed by the peak of the absence of a peak in a that in a real network there is no. Figure 6.1 Random and Scale-Free Networks. The degree follows a bell curve number of links, and nade: with a very larg ‘Thus a random network A) lar to a national highway network, in which the the cities, and the links are the major highways connecting ther. Indeed, are served by roughly the same the power law degree of highway: (boro free network predi ed hubs (top gle node which we could pick out and claim in these networks. Thi the realization that most complex networks in gree distribut sm scale-free networks exponentially —a much faster decay than that predi They both told us, in rigorous mathemat ‘The surprising discovery of power modlates such highly linked anomalies in a nat each scale-free network will have several large discovered them in 1999. they The random network theory of jon by Watts and Stro: links should decrease me that we are scale-free gave legitimacy to2 LINKED individuals, wh 80/20 rule truly applies, you can by wws formulate in mathemati events carry most of the action. Power havior. We just needed celebrated discovery was that income distribu ws, implying that most money is eamed majority of d 1 fo a few very ly dominated by a roll they signal a trans might seem that the discovery that networks obey a simple power law would be exciting only to a few mathem: ians or phys s. But power laws are at the heart of some of the most stunning conceptual ad- vances in the second hal of the ewentieth century, emerging in fields like chaos, fractals, and phase trar ns. Spotting rem in networks The 80/20 Rule B the cell, and many other complex systems all obey a power 1s to paraphrase Pareto and claim for the first time that perhaps there were laws behind complex networks. 4. With a big O head and two great H ea ter molecule and its H,O symbol is fa temmal structure are known in miniscule det Water is the most common and most studied substance on E liquid water, the collection of billions of cohesive molecules crowded in a glass, continues to chall Gases are simple: Molecules fly in empty space, taking notice of between these two extremes. The ai molecules together are not strong enc order. Trapped between order and cl 1 majestic dance in which son and somewhat ordered groups, move together, and jain other molecules forming yet other groups. ling a glass of water does not significantly al makes the mot a perfectly ordered ice cn officer's command. But ly learning their precis in contrast, may have never experi- enced ice before, They follow a myst dering lifestyle for a rigid, ordered one. Ice, a familiar symbol of cold and perfect order, emerges spontaneously. 1s, paint us urge to exchange their wan."4 LINKED 1e cold crystalline order of a Exchanging their water dance for solid is one of the best-known examp! iat physicists had sought to understand for decades prior to the common in various materials taking forms the atoms point theie spins randomly in all to some critical temperature, however, orient their spins in mn and form a magnet. wg of a liquid and the emergence of a magnet are both fe to the perfect order of At the freezing take up che highly ordered common orientation once cooled under a critical eemperature. Such sudden transitions hold the key to a about how nature works, of equal interest to scientists and philosophers alike: How does order emerge from disorder? 5. “The ordered and the disordered states of a magnet correspond modynamically distinct phases of matter. Right at the transi poised to choose between these two phases, just ny the mountain. Un- goes back and forth, limber on a crest choosing which side to go decided which way to go, the system frequent! ‘Near the critical point, elements of order and disorder jgnaling that the system explores both sides of the crest mnsition temperature, clusters of atoms develop jon. The closer the metal gets to the “The 80/20 Rule 5 asa rough measure of the approach the critical point this corel pow red by a unig metal gets to the phase transit over which the spins knaw about each other. The strength of the magnet ity ofthe critical temperature, determined by the fraction of law chara in the vie ‘As physicists carefully investigated how in various sy ‘emerges from disorder, more power laws were discovered to opk ing a phase transition. The same laws emerged as once heated, or wl uffcie y chilled. The disorder-order tran amaring degree of mathema body knew why. Why di their identity at some critical point and decide to fo power laws! What is behind the high degree of similarity between such disparate systems? And what do power laws have to do with it? 1 frst major breakthroughs rowan jsorder to order came during the%6 LINKED The 80/20 Rue in the vicinity of the critical point. Kadanoff’s idea wlel that could be used to derive precise power law emer offered an appealing vi ‘mathematical demonstrated that ehe transition from disorder to order did not require rine unknown exponents but could be expressed in terms of any Wo of them. Unknown to him, several other researchers arrived at the same conclusion simultaneously. Ben Widom, a physical chemist from Comel University, and A. Z. Pa physicists from the rough a different route. he pyramid of won him the 1982 Nobel pri laws. In ordinary systems ions decay ra laws. But all that changes if the system transition. Then power laws emerge—na 120s is departing in favor of ox us loud and clear that the road from disorder to 0 the powerful forces of self-organization and is paved by power laws us that power laws are lar scaling relations mique and deep meaning of power laws perhaps ex; ‘excitement when we fist spotted them on the Web. they were unprecedented and unexpected in the context of ne Te was that hoped th 1971 for the final answer. It wasn't only thar ons—the tangible measure of success in academia, was very close to jeopardizing his job at Cornell. scripts from his desk drawer. Two of these, submitted simultaneously on y Physical Review B, turne and ‘Wilson took the 1s developed by Kadanoff and molded them into a power renormalization. The starting point of his approach was scale invariance: He assumed that in the vicinity of the critical pa apply, generating the mysterivus powIN that has similar characteristics in a wide range of systems, “The ubiquity of power laws in systems unclergoing a transition from disorder to order prompted my Ph.D. advisor, H. Eugene Stanley, who at Boston University leads the most active research group investigating phase transitions, to joke that in Boston there is only log-log paper. Stanley, who has been involved in all major discoveries shaping our un- derstanding of phase transitions and universality, was referring to the 1e presence of power laws in exper plo used by scientists to derect mental data, Indeed, wherever physicists, biologists, ecologists, materi jentists, mathematicians, or economists looked in the eighties and f self-organization reigned, power laws and universality em. It appears that networks are no different: Behind the her strict mathematical expression, a power law. Is nineti sreeced hubs chere ‘This brings us co che next puzzl systems in transition from chaos 0 onder, what kl of transition is take ing place in complex networks? If power laws appear in the vicinity of critical point, what tunes real networks to their own critical point, al- lowing them to display a scale-free behavior? We had come to un ists uncovered the mechanisms If power laws are the signature of stand critical phenomena after phy Jeming phase transitions; rigorous theories now allow us to calculate quantities characterizing systems giving birth to «works we had only observed the bubs. We now with high precision ganization and order. To be sure, this was an important breakthrough, swing us to remove networks from the realm of most important questions, pertaining to the mechanisms le for the hubs and the power laws, were still unanswered. Are \etworks in a continuous state of transition from disorder to order? ubs appear in networks of all kinds, ranging from actors to the Why ate they described by power laws? Are there fundamental laws forcing different networks 10 take up the same universal form and ype! How does nature spin its webs? THE SEVENTH LINK Rich Get Richer ONCE 4 PROMINENT MERCHANT PORT of the Portuguese empire, Porto to- ly defensible narrow key. With its magnificent cas- tles overlooking the river and a rich history of wine making, one might expect it to be one of the most visited cities in the world. But hidden as it is in the norchwest comer of the Iberian Peninsula, few tourists make the detour. Thete are apparently £00 few fans of the distinctive full-bodied Porto vintage to awaken this great medieval city from its dreamlike stare 1 visited Porto in the summer of 1999, shortly after my studen he Web. T was sy University of Porto, José Mendes mer of 1999 very few people were networks, and there were no talks on the subject during, this workshop, But networks were very much on my mind, I could not | finished our manuscript on the role of pow attending a workshop on none ized by two professors of physic © Web was the only network 1g €0 understand it, we were sear features. At the ime, we wanted ro learn more about pRich Get Richer at eT LINKED ‘The eight-hour flight from Lisbon to New York seen nity to prepare the frst draft. As soon as the plane took 0 hased before che Port databace. Jay Brockman, a computer gave us data on a man-made network, 1 wiring diagram of a com. 1 Reka Albert and I agreed that my departure, | received a Long e-mail ailing some ongoing acti end of the message looked at the de- aweek there was a sentence gree distribution too, and in f the distributi suddenly made the Web was by no means ing in the conference hall paying no atten- ks, thinking about the implications of this finding, If ewo ferent as the Web and the Hollywood acting communi law oF tion to the networks as both display power-law des must be responsible. If such a law existed, it could poten: networks. During the first break between talks I decided to withdraw quiet of the seminary where we were be however, During the fifteen-minute walk hack to my t life. Second, nodes, we link chem randomly to each other. These assumptions we unquestioned in over forty years of network research, ut the ‘of hubs—and the power laws that describe them—forced us t both assumptions. The manuscript submitted to Science along this path. fax Réka, asking her to verify the idea using the computes. A few hours later she e-mailed me the answer. To my great astonishment, the idea -richer phenomenon, potentially present in he Web 2. There is one thing abo Each day new est hobby or interest; by conporacions ex ‘Web that everybody agrees ot8 LINKED ich Gee Richer 8 the Web to dis: seminate information to citizens; by college professors publishing lecture notes; by nonprofit organizations trying to reach those who 1d by thousands of dot.com compa- hy pages to compete for your wallet. It is estimated years the Web Despite their diver growth, Pick any network you can be true: Starting with a few nodes dition of new nodes, reaching growth forces us to rethink our modeling assumptions. Bo Rényi and Watts-Strogate models assumed that we have a of nodes that are wired toge ated by these models are ther nodes ret smples suggested that priate. Instead, we sl ‘This was che initial insight we. we ended up dethroning the first the random universe—its ware an essential feature: ely through the ad- ost real necworks and services; by governments increasingly of in host abour an exabyte (10 ts, most of which ely sanity lands ontine, are preseni so far there are no signs ofa slowdown, th over a billion documents available toda ye Web emerged one node at a time, But it did. Barely a decade ago it ul only one node, Tim Berners-Lee's famous first Webpage. As physi- ted creating pages of core and keep adding nodes, one after the other. Let us assume that each new node has two links. Thus, if we start with two nodes, our third node will link to both of them. The fourth node has three nodes from which to choose. How do we pick insion is in stark contrast co the assumption of the network models book, which assume the number of nodes in a of the first silent movies IMDb.com database, Hol easing demand adding a few new faces w st boom between 1908 and ing the trade went from under 50 to close to 2,000 a year. A second starting in the 1980s turned moviemal y we know today. From a tiny cluster of nwood had only 53 actors in 1900, Wi core slowly expanded, wood experienced its 14, when the number of actors join- we connect by this simple alg work model of Erddis and Rényi only in its growing nature. This differ- ic network of over a half-million nodes, many as 13,209 names of actors appearing for the ing in a clear advantage for the senior nodes, we wide canvas of the movie screen were added to the some rare statistical fluctuationse LINKED we system, with owo links Model A was The poorest node will be the last one only, because nobody has had ti among our frst predicted that they are too small wre, Model A failed to account for the hubs and the connector alone cannot explain 1999 Super Bowl numerous neverheardof.com companies com, WebEx.com, and Epidemic Marketing ame to millions of blew $2 million pet Americans following the year alone E*Trade 9 itself, AlraVista, one iod on cable television advertising, a medium whose history spans over two decades. ‘What did these companies want to achieve? The answer is simple, jconventional. Startups and established companies alike had been burning venture ¢: 1 a day, to defeat the random universe of Exdés and Rényi. They knew that we do not link randomly on the Web. They wanted to rage of this non randomness by begging us to link to th How do we in fact decide which Websites to link to on the Wo random network models, we would ran- of the nodes. A bit of reflection as to how we make our indicates otherwise, For example, choices of Webpages ead q Rich Get Richer 6 with links to news outlets abound, A quick sea returns about 109,000,000 hits. Yahoo's Ido not think that anybody ever does that. Rather, most of a few major news outlets. Without giving the matter link to one of them. As a longtime reader of the New Yunk Time: a no-brainer for me to choose nytimes.com. Others might prefer CNN.com or MSNBC.com. $ Preferential attachmes whose job it is to make a movie prot ‘Thus casting is determined by two competing facto be known to get good roles, but assumption inherent in random networks—their in the Erdés-Rényi and Watts-Strogats mod sween the nodes of a network; chus all nodes nks. The examples just