JOVANA
Library Glossary Getting Started Three Levels Fields How it works Mission
Join the mission
Back to the library
网络科学 1960

随机图的演化

保罗·埃尔德什、阿尔弗雷德·雷尼

随机撒下连线,在某个精确的瞬间,一张巨网骤然成形。

Choose your version
In depth · the introduction

不断给一群陌生人随机牵线,到某个精确的瞬间,散落的小团体便会融成一张几乎囊括所有人的巨网。

把这个想法拆开看

取大量的点,开始一对一对地随机连线,一条接一条。起初你得到许多细小、彼此分离的碎块。很长一段时间里什么戏剧性的事都没发生——碎块只是慢慢变大。然后,在一个精确的临界点上,骤变发生了:一个碎块猛地盖过其余所有,把整幅画里很大一部分的点连成一片。

埃尔德什与雷尼找出了那个临界点究竟在哪。神奇的数字,是平均每个点一条连线。每点平均不到一条线时,网络是一盘小碎屑;越过那条线,一张「巨大」的网便骤然成形。这变化不是渐进的,而是一道阈值——就像水在某个确定的温度结成冰。

它从哪里来

保罗·埃尔德什——那位四海为家的匈牙利人,几乎一无所有,睡在同行的沙发上,合作者之多冠绝古今——与阿尔弗雷德·雷尼联手。雷尼爱说,数学家是把咖啡变成定理的机器。在 1959–1960 年间一连串论文里,他们问了一个貌似简单的问题:一个典型的图,长什么样?通过想象图一条边一条边地生长,他们发现它的性格会经由骤然的跳变而改变,并写下了这些随机网络的第一部「自然史」。

它为何重要

在他们之前,图是一张一张、靠手研究的。埃尔德什与雷尼表明:你可以一次性地对整个随机系综作推理,并在规模很大时以近乎确定的把握预言它的特征。而他们那个临界点的想法,后来证明无处不在:同一套数学,描述着一场病何时倾覆为流行病、一则流言何时疯传、一张电网或互联网何时变得稳健连通、一种材料何时忽然开始导电。他们找到的,是连接本身的一条法则。

一个打比方

想象一锅里散布的爆米花粒,每一粒都能粘住邻居。随机撒下粘性的连线,你先得到几个小团。继续撒,团仍旧小——直到在某个临界的粘度上,接下来的几条线把一个个团搭成一整张、横贯整锅的薄片。前一刻:松散的碎屑。后一刻:一团相连的整体。这跨越,恰发生在每一粒平均只有一条连线之时。

圆盘里的七十个点,由随机的连线相连。滑块设定每个点的平均连线数。每点不到一条线时,只形成各自分离的小碎块。当平均数越过 1,一个巨大的连通群以颜色亮起,迅速纳入大部分点。

它落在哪里

这是网络科学的奠基之作。它与物理学中的渗流理论、生物学中的流行病模型并立——后两者描述的,正是同一个骤然的临界点。后来的研究者——Watts 与 Strogatz 的「小世界」、Barabási 与 Albert 的「枢纽」——造出了更丰富的模型,因为真实的网络(友谊、万维网、神经元)并非完全随机。但它们每一个,都对照着埃尔德什–雷尼随机图来度量:那是纯粹随机的基线。

The original document
Original source text
P. Erdős & A. Rényi · Publ. Math. Inst. Hungar. Acad. Sci. 5 (1960) · pp. 17–61 · dedicated to Professor P. Turán on his 50th birthday
The model
For n labelled vertices there are C(n,2) = n(n−1)/2 possible edges. Erdős and Rényi study the graph Γ(n,N) formed by choosing N of those edges at random, every such graph being equally likely. They let N grow with n and ask how a typical graph looks at each stage — treating the steadily rising N as a kind of time, so that the graph 'evolves'.
Threshold functions
Their central discovery is that almost every property appears not gradually but at a sharp threshold. For many properties A there is a threshold function: if N grows much more slowly than it, almost no graph has A; if N grows much faster, almost every graph has A. The switch is abrupt, and most of the paper computes these thresholds, property by property.
The double jump — the giant component
The most famous threshold governs the largest connected component, near N ≈ n/2 — that is, an average of about one edge per vertex. Below it (N = cn/2 with c < 1) every component is small, of order log n. Exactly at c = 1 the largest is of order n^(2/3). Above it (c > 1) a single 'giant' component of order n emerges and absorbs a finite fraction of all vertices, while every other component stays small. Erdős and Rényi named this sudden change the 'double jump'.
[ … ]
Connectivity and a whole natural history
Pushing N higher, isolated vertices are the last obstacle to connectivity: around N ≈ (n/2)·ln n the last isolated vertex is absorbed and the graph becomes connected — almost exactly when its minimum degree first reaches 1. Isolated trees appear first (the larger the tree, the later it shows up), then the first cycles arrive together with the giant near mean degree one. The paper charts this whole sequence of thresholds, giving the first detailed natural history of the random graph.
Budapest · 1960