Spectra of Random Graphs - University of South...
Transcript of Spectra of Random Graphs - University of South...
![Page 1: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/1.jpg)
Spectra of RandomGraphs
Linyuan Lu
University of South Carolina
Selected Topics on Spectral Graph Theory (III)Nankai University, Tianjin, May 29, 2014
![Page 2: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/2.jpg)
Five talks
Spectra of Random Graphs Linyuan Lu – 2 / 68
Selected Topics on Spectral Graph Theory
1. Graphs with Small Spectral RadiusTime: Friday (May 16) 4pm.-5:30p.m.
2. Laplacian and Random Walks on GraphsTime: Thursday (May 22) 4pm.-5:30p.m.
3. Spectra of Random GraphsTime: Thursday (May 29) 4pm.-5:30p.m.
4. Hypergraphs with Small Spectral RadiusTime: Friday (June 6) 4pm.-5:30p.m.
5. Laplacian of Random HypergraphsTime: Thursday (June 12) 4pm.-5:30p.m.
![Page 3: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/3.jpg)
Backgrounds
Spectra of Random Graphs Linyuan Lu – 3 / 68
Linear Algebra
I
Graph Theory
II
Probability Theory
III
I: Spectral Graph Theory II: Random Graph TheoryIII: Random Matrix Theory
![Page 4: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/4.jpg)
Outline
Spectra of Random Graphs Linyuan Lu – 4 / 68
■ Classical random theory: Erdos-Renyi model
![Page 5: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/5.jpg)
Outline
Spectra of Random Graphs Linyuan Lu – 4 / 68
■ Classical random theory: Erdos-Renyi model
■ Power law graphs
![Page 6: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/6.jpg)
Outline
Spectra of Random Graphs Linyuan Lu – 4 / 68
■ Classical random theory: Erdos-Renyi model
■ Power law graphs
■ Chung-Lu model
![Page 7: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/7.jpg)
Outline
Spectra of Random Graphs Linyuan Lu – 4 / 68
■ Classical random theory: Erdos-Renyi model
■ Power law graphs
■ Chung-Lu model
■ Edge-independent random graphs
![Page 8: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/8.jpg)
Preliminary
Spectra of Random Graphs Linyuan Lu – 5 / 68
A graph consists of two sets V and E.
- V is the set of vertices (or nodes).- E is the set of edges, where each edge is a pair ofvertices.
The degree of a vertex is the number of edges, which areincident to that vertex.
Diameter: the maximum distance d(u, v), where u and v arein the same connected component.
![Page 9: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/9.jpg)
Preliminary
Spectra of Random Graphs Linyuan Lu – 5 / 68
A graph consists of two sets V and E.
- V is the set of vertices (or nodes).- E is the set of edges, where each edge is a pair ofvertices.
The degree of a vertex is the number of edges, which areincident to that vertex.
Diameter: the maximum distance d(u, v), where u and v arein the same connected component.
Average distance: the average among all distance d(u, v) forpairs of u and v in the same connected component.
![Page 10: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/10.jpg)
Random graphs
Spectra of Random Graphs Linyuan Lu – 6 / 68
A random graph is a set of graphs together with aprobability distribution on that set.
![Page 11: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/11.jpg)
Random graphs
Spectra of Random Graphs Linyuan Lu – 6 / 68
A random graph is a set of graphs together with aprobability distribution on that set.Example: A random graph on 3 vertices and 2 edges withthe uniform distribution on it.
❥ ❥
❥
Probability 13
❥ ❥
❥
�����
Probability 13
❥ ❥
❥
�����
Probability 13
![Page 12: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/12.jpg)
Random graphs
Spectra of Random Graphs Linyuan Lu – 6 / 68
A random graph is a set of graphs together with aprobability distribution on that set.Example: A random graph on 3 vertices and 2 edges withthe uniform distribution on it.
❥ ❥
❥
Probability 13
❥ ❥
❥
�����
Probability 13
❥ ❥
❥
�����
Probability 13
A random graph G almost surely satisfies a property P , if
Pr(G satisfies P ) = 1− on(1).
![Page 13: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/13.jpg)
Erdos-Renyi model G(n, p)
Spectra of Random Graphs Linyuan Lu – 7 / 68
- n nodes
![Page 14: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/14.jpg)
Erdos-Renyi model G(n, p)
Spectra of Random Graphs Linyuan Lu – 7 / 68
- n nodes- For each pair of vertices, create an edge independentlywith probability p.
![Page 15: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/15.jpg)
Erdos-Renyi model G(n, p)
Spectra of Random Graphs Linyuan Lu – 7 / 68
- n nodes- For each pair of vertices, create an edge independentlywith probability p.
- The graph with e edges has the probability pe(1− p)(n2)−e.
![Page 16: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/16.jpg)
Erdos-Renyi model G(n, p)
Spectra of Random Graphs Linyuan Lu – 7 / 68
- n nodes- For each pair of vertices, create an edge independentlywith probability p.
- The graph with e edges has the probability pe(1− p)(n2)−e.
❥ ❥
❥ ❥
❅❅
❅❅
❅
�����
![Page 17: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/17.jpg)
Erdos-Renyi model G(n, p)
Spectra of Random Graphs Linyuan Lu – 7 / 68
- n nodes- For each pair of vertices, create an edge independentlywith probability p.
- The graph with e edges has the probability pe(1− p)(n2)−e.
❥ ❥
❥ ❥
❅❅
❅❅
❅
�����
p
pp
p
![Page 18: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/18.jpg)
Erdos-Renyi model G(n, p)
Spectra of Random Graphs Linyuan Lu – 7 / 68
- n nodes- For each pair of vertices, create an edge independentlywith probability p.
- The graph with e edges has the probability pe(1− p)(n2)−e.
❥ ❥
❥ ❥
❅❅
❅❅
❅
�����
p
pp
p
♣ ♣ ♣ ♣ ♣ ♣ ♣
♣♣♣♣♣♣♣
1− p
1− p
![Page 19: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/19.jpg)
Erdos-Renyi model G(n, p)
Spectra of Random Graphs Linyuan Lu – 7 / 68
- n nodes- For each pair of vertices, create an edge independentlywith probability p.
- The graph with e edges has the probability pe(1− p)(n2)−e.
❥ ❥
❥ ❥
❅❅
❅❅
❅
�����
p
pp
p
♣ ♣ ♣ ♣ ♣ ♣ ♣
♣♣♣♣♣♣♣
1− p
1− p
The probability of thisgraph is
p4(1− p)2.
![Page 20: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/20.jpg)
Evolution of G(n, p)
Spectra of Random Graphs Linyuan Lu – 8 / 68
Erdos-Renyi 1960s:
■ p ∼ c/n for 0 < c < 1: The largest connectedcomponent of Gn,p is a tree and has about1α(log n− 5
2 log log n) vertices, where α = c− 1− log c.
![Page 21: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/21.jpg)
Evolution of G(n, p)
Spectra of Random Graphs Linyuan Lu – 8 / 68
Erdos-Renyi 1960s:
■ p ∼ c/n for 0 < c < 1: The largest connectedcomponent of Gn,p is a tree and has about1α(log n− 5
2 log log n) vertices, where α = c− 1− log c.
■ p ∼ 1/n+ c/n4/3, the largest connected component isΘ(n2/3). Double jump: Θ(log n) → Θ(n2/3) → Θ(n).
![Page 22: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/22.jpg)
Evolution of G(n, p)
Spectra of Random Graphs Linyuan Lu – 8 / 68
Erdos-Renyi 1960s:
■ p ∼ c/n for 0 < c < 1: The largest connectedcomponent of Gn,p is a tree and has about1α(log n− 5
2 log log n) vertices, where α = c− 1− log c.
■ p ∼ 1/n+ c/n4/3, the largest connected component isΘ(n2/3). Double jump: Θ(log n) → Θ(n2/3) → Θ(n).
■ p ∼ c/n for c > 1: Except for one “giant” component,all the other components are relatively small. The giantcomponent has approximately f(c)n vertices, where
f(c) = 1− 1
c
∞∑
k=1
kk−1
k!(ce−c)k.
![Page 23: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/23.jpg)
Diameter of G(n, p)
Spectra of Random Graphs Linyuan Lu – 9 / 68
Bollobas (1985): (denser graph)
diam(G(n, p)) =
⌊
log n
log np
⌋
or
⌈
log n
log np
⌉
if np ≫ log n.
![Page 24: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/24.jpg)
Diameter of G(n, p)
Spectra of Random Graphs Linyuan Lu – 9 / 68
Bollobas (1985): (denser graph)
diam(G(n, p)) =
⌊
log n
log np
⌋
or
⌈
log n
log np
⌉
if np ≫ log n.
Chung Lu, (2000) (Sparser graph)
diam(G(n, p)) =
{
(1 + o(1)) log nlog np if np → ∞
Θ( log nlog np) if ∞ > np > 1.
![Page 25: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/25.jpg)
Wigner’s semicircle law
Spectra of Random Graphs Linyuan Lu – 10 / 68
(Wigner, 1958)
- A is a real symmetric n× n matrix.- Entries aij are independent random variables.- E(a2k+1
ij ) = 0.
- E(a2ij) = m2.
- E(a2kij ) < M .
The distribution of eigenvalues of A converges into asemicircle distribution of radius 2m
√n.
![Page 26: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/26.jpg)
Spectra of G(n, p)
Spectra of Random Graphs Linyuan Lu – 11 / 68
The eigenvalues of an Erdos-Renyi random graph follow thesemicircle law. ( Furedi and Komlos, 1981)
Laplacian eigenvalues also follow the semicircle law.
![Page 27: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/27.jpg)
Challenge
Spectra of Random Graphs Linyuan Lu – 12 / 68
Erdos-Renyi model G(n, p) is classical, simple, beautiful...,but not suitable to model complex graphs.
![Page 28: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/28.jpg)
Challenge
Spectra of Random Graphs Linyuan Lu – 12 / 68
Erdos-Renyi model G(n, p) is classical, simple, beautiful...,but not suitable to model complex graphs.
■ What are complex graphs?
![Page 29: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/29.jpg)
Challenge
Spectra of Random Graphs Linyuan Lu – 12 / 68
Erdos-Renyi model G(n, p) is classical, simple, beautiful...,but not suitable to model complex graphs.
■ What are complex graphs?
■ How to model these complex graphs by random graphs?
![Page 30: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/30.jpg)
Challenge
Spectra of Random Graphs Linyuan Lu – 12 / 68
Erdos-Renyi model G(n, p) is classical, simple, beautiful...,but not suitable to model complex graphs.
■ What are complex graphs?
■ How to model these complex graphs by random graphs?
■ How to deduce the graph properties of these generalrandom graph models?
![Page 31: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/31.jpg)
Examples of complex graphs
Spectra of Random Graphs Linyuan Lu – 13 / 68
WWW Graphs
Call Graphs
Collaboration Graphs
Gene Regulatory Graphs
Graph of U.S. Power Grid
Costars Graph of Actors
...
![Page 32: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/32.jpg)
A subgraph of the Collaboration Graph
Spectra of Random Graphs Linyuan Lu – 14 / 68
![Page 33: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/33.jpg)
Collaboration Graph at USC
Spectra of Random Graphs Linyuan Lu – 15 / 68
Czabarka Szekely Lu
Griggs
Cooper
West
Milan
Johnston
Chung
Graham
Horn
Tetali
Furedi
Odlyzko
Pralat
Mohr
Yang
LiKleitman
LanMan Peng
Zhao
Martin
P.L. Erdos Katona
Sali
Anstee
Johnson
Biro
Dankelmann
Bokal
Wagner Wang
Howard
Karolyi
Miklos
ToroczkaiSteel
Wormaldde Caen
Shahrokhi Entriger Smith
Chen
Boehnlein Walters Kay
Dutle
Fenner
Chudak
Ho
Jin
Lin
Liu
DoveOuyang
Wu
Yeh
Zhu
Spouge
Chin
Liang
Narayan
SunJonas
SandbergJordan
Faculty, Ph.D. students, Postdocs, and visitors to theCombinatorics Group at the University of South Carolina.
![Page 34: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/34.jpg)
An IP Graph (by Bill Cheswick)
Spectra of Random Graphs Linyuan Lu – 16 / 68
![Page 35: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/35.jpg)
BGP Graph
Spectra of Random Graphs Linyuan Lu – 17 / 68
Vertex: AS(autonomous system)
Edges: AS pairs inBGP routing table.
![Page 36: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/36.jpg)
Large BGP subgraph
Spectra of Random Graphs Linyuan Lu – 18 / 68
Only a portion of 6400 vertices and 13000 edges is drawn.
![Page 37: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/37.jpg)
Hollywood Graph
Spectra of Random Graphs Linyuan Lu – 19 / 68
Vertex: actors andactress
Edges: co-playing inthe same movie
Only 10,000 out of225,000 are shown.
![Page 38: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/38.jpg)
Protein-interaction network
Spectra of Random Graphs Linyuan Lu – 20 / 68
Snel, Bork & Huynen, PNAS 99, 5890 (2002)
![Page 39: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/39.jpg)
A subgraph of the Collaboration Graph
Spectra of Random Graphs Linyuan Lu – 21 / 68
![Page 40: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/40.jpg)
Folklore of Erdos numbers
Spectra of Random Graphs Linyuan Lu – 22 / 68
■ Erdos has Erdos number 0.
■ Erdos’ coauthor has Erdos number 1.
■ Erdos’ coauthor’s coauthor has Erdosnumber 2.
...
![Page 41: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/41.jpg)
Folklore of Erdos numbers
Spectra of Random Graphs Linyuan Lu – 22 / 68
■ Erdos has Erdos number 0.
■ Erdos’ coauthor has Erdos number 1.
■ Erdos’ coauthor’s coauthor has Erdosnumber 2.
...
My Erdos number is 2.
![Page 42: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/42.jpg)
Folklore of Erdos numbers
Spectra of Random Graphs Linyuan Lu – 22 / 68
■ Erdos has Erdos number 0.
■ Erdos’ coauthor has Erdos number 1.
■ Erdos’ coauthor’s coauthor has Erdosnumber 2.
...
My Erdos number is 2.
Erdos number is the graph distance to Erdos in theCollaboration graph.
![Page 43: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/43.jpg)
Collaboration Graph
Spectra of Random Graphs Linyuan Lu – 23 / 68
![Page 44: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/44.jpg)
Characteristics
Spectra of Random Graphs Linyuan Lu – 24 / 68
■ Large
![Page 45: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/45.jpg)
Characteristics
Spectra of Random Graphs Linyuan Lu – 24 / 68
■ Large
■ Sparse
![Page 46: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/46.jpg)
Characteristics
Spectra of Random Graphs Linyuan Lu – 24 / 68
■ Large
■ Sparse
■ Power law degree distribution
![Page 47: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/47.jpg)
Characteristics
Spectra of Random Graphs Linyuan Lu – 24 / 68
■ Large
■ Sparse
■ Power law degree distribution
■ Small world phenomenon
![Page 48: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/48.jpg)
The power law
Spectra of Random Graphs Linyuan Lu – 25 / 68
The number of vertices of degree k is approximatelyproportional to k−β for some positive β.
![Page 49: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/49.jpg)
The power law
Spectra of Random Graphs Linyuan Lu – 25 / 68
The number of vertices of degree k is approximatelyproportional to k−β for some positive β.
A power law graph is a graph whose degree sequencesatisfies the power law.
![Page 50: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/50.jpg)
Power law distribution
Spectra of Random Graphs Linyuan Lu – 26 / 68
Left: The collaborationgraph follows the powerlaw degree distributionwith exponent β ≈ 3.0
![Page 51: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/51.jpg)
Power law distribution
Spectra of Random Graphs Linyuan Lu – 26 / 68
Left: The collaborationgraph follows the powerlaw degree distributionwith exponent β ≈ 3.0
Right: An IP graphfollows the power law de-gree distribution with ex-ponent β ≈ 2.4
![Page 52: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/52.jpg)
Power law graphs
Spectra of Random Graphs Linyuan Lu – 27 / 68
Left: Part of the collab-oration graph (authorswith Erdos number 2)
Right: An IP graph (by BillCheswick)
![Page 53: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/53.jpg)
Robustness of Power Law
Spectra of Random Graphs Linyuan Lu – 28 / 68
size degree distribution
25,3339
52,186
![Page 54: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/54.jpg)
Basic questions
Spectra of Random Graphs Linyuan Lu – 29 / 68
■ How to model power law graphs?
![Page 55: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/55.jpg)
Basic questions
Spectra of Random Graphs Linyuan Lu – 29 / 68
■ How to model power law graphs?
■ What graph properties can be derived
from the model?
![Page 56: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/56.jpg)
Model G(w1, w2, . . . , wn)
Spectra of Random Graphs Linyuan Lu – 30 / 68
Random graph model with given expected degree sequence(Chung-Lu model)
- n nodes with weights w1, w2, . . . , wn.
![Page 57: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/57.jpg)
Model G(w1, w2, . . . , wn)
Spectra of Random Graphs Linyuan Lu – 30 / 68
Random graph model with given expected degree sequence(Chung-Lu model)
- n nodes with weights w1, w2, . . . , wn.
- For each pair (i, j), create an edge independently withprobability pij = wiwjρ, where ρ = 1
∑ni=1 wi
.
![Page 58: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/58.jpg)
Model G(w1, w2, . . . , wn)
Spectra of Random Graphs Linyuan Lu – 30 / 68
Random graph model with given expected degree sequence(Chung-Lu model)
- n nodes with weights w1, w2, . . . , wn.
- For each pair (i, j), create an edge independently withprobability pij = wiwjρ, where ρ = 1
∑ni=1 wi
.
- The graph H has probability
∏
ij∈E(H)
pij∏
ij 6∈E(H)
(1− pij).
![Page 59: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/59.jpg)
Model G(w1, w2, . . . , wn)
Spectra of Random Graphs Linyuan Lu – 30 / 68
Random graph model with given expected degree sequence(Chung-Lu model)
- n nodes with weights w1, w2, . . . , wn.
- For each pair (i, j), create an edge independently withprobability pij = wiwjρ, where ρ = 1
∑ni=1 wi
.
- The graph H has probability
∏
ij∈E(H)
pij∏
ij 6∈E(H)
(1− pij).
- The expected degree of vertex i is wi.
![Page 60: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/60.jpg)
An example: G(w1, w2, w3, w4)
Spectra of Random Graphs Linyuan Lu – 31 / 68
✒✑✓✏
✒✑✓✏
✒✑✓✏
✒✑✓✏
❅❅
❅❅
❅❅
❅❅❅
���������
![Page 61: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/61.jpg)
An example: G(w1, w2, w3, w4)
Spectra of Random Graphs Linyuan Lu – 31 / 68
✒✑✓✏
✒✑✓✏
✒✑✓✏
✒✑✓✏
❅❅
❅❅
❅❅
❅❅❅
���������
w1 w2
w3 w4
![Page 62: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/62.jpg)
An example: G(w1, w2, w3, w4)
Spectra of Random Graphs Linyuan Lu – 31 / 68
✒✑✓✏
✒✑✓✏
✒✑✓✏
✒✑✓✏
❅❅
❅❅
❅❅
❅❅❅
���������
w1 w2
w3 w4
w1w2ρ
w1w3ρ
w1w4ρ
w2w3ρ
![Page 63: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/63.jpg)
An example: G(w1, w2, w3, w4)
Spectra of Random Graphs Linyuan Lu – 31 / 68
✒✑✓✏
✒✑✓✏
✒✑✓✏
✒✑✓✏
❅❅
❅❅
❅❅
❅❅❅
���������
w1 w2
w3 w4
w1w2ρ
w1w3ρ
w1w4ρ
w2w3ρ
q q q q q q q1− w3w4ρ
1− w2w4ρ
qqqqqqq
![Page 64: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/64.jpg)
An example: G(w1, w2, w3, w4)
Spectra of Random Graphs Linyuan Lu – 31 / 68
✒✑✓✏
✒✑✓✏
✒✑✓✏
✒✑✓✏
❅❅
❅❅
❅❅
❅❅❅
���������
w1 w2
w3 w4
w1w2ρ
w1w3ρ
w1w4ρ
w2w3ρ
q q q q q q q1− w3w4ρ
1− w2w4ρ
qqqqqqq
qqq q qq qqqq
q qqqq
qqq q q
1− w2
1ρ
1− w2
2ρ
1− w2
3ρ
1− w2
4ρ
![Page 65: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/65.jpg)
An example: G(w1, w2, w3, w4)
Spectra of Random Graphs Linyuan Lu – 31 / 68
✒✑✓✏
✒✑✓✏
✒✑✓✏
✒✑✓✏
❅❅
❅❅
❅❅
❅❅❅
���������
w1 w2
w3 w4
w1w2ρ
w1w3ρ
w1w4ρ
w2w3ρ
q q q q q q q1− w3w4ρ
1− w2w4ρ
qqqqqqq
qqq q qq qqqq
q qqqq
qqq q q
1− w2
1ρ
1− w2
2ρ
1− w2
3ρ
1− w2
4ρ
The probability of the graph is
w31w
22w
23w4ρ
4(1− w2w4ρ)× (1− w3w4ρ)4∏
i=1
(1− w2i ρ).
![Page 66: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/66.jpg)
Chung-Lu model
Spectra of Random Graphs Linyuan Lu – 32 / 68
For G = G(w1, . . . , wn), let
- d = 1n
∑ni=1wi
- d =∑n
i=1 w2i
∑ni=1 wi
.
- The volume of S: Vol(S) =∑
i∈S wi.
![Page 67: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/67.jpg)
Chung-Lu model
Spectra of Random Graphs Linyuan Lu – 32 / 68
For G = G(w1, . . . , wn), let
- d = 1n
∑ni=1wi
- d =∑n
i=1 w2i
∑ni=1 wi
.
- The volume of S: Vol(S) =∑
i∈S wi.
We haved ≥ d
“=” holds if and only if w1 = · · · = wn.
![Page 68: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/68.jpg)
Chung-Lu model
Spectra of Random Graphs Linyuan Lu – 32 / 68
For G = G(w1, . . . , wn), let
- d = 1n
∑ni=1wi
- d =∑n
i=1 w2i
∑ni=1 wi
.
- The volume of S: Vol(S) =∑
i∈S wi.
We haved ≥ d
“=” holds if and only if w1 = · · · = wn.
A connected component S is called a giant component if
vol(S) = Θ(vol(G)).
![Page 69: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/69.jpg)
Connected components
Spectra of Random Graphs Linyuan Lu – 33 / 68
Chung and Lu (2001) For G = G(w1, . . . , wn),
■ If d < 1− ǫ, then almost surely, all components havevolume at most O(
√n log n).
![Page 70: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/70.jpg)
Connected components
Spectra of Random Graphs Linyuan Lu – 33 / 68
Chung and Lu (2001) For G = G(w1, . . . , wn),
■ If d < 1− ǫ, then almost surely, all components havevolume at most O(
√n log n).
■ If d > 1 + ǫ, then almost surely there is a unique giantcomponent of volume Θ(Vol(G)). All other componentshave size at most
{
log nd−1−log d−ǫd if 1
1−ǫ < d < 21−ǫ
log n1+log d−log 4+2 log(1−ǫ) if d > 4
e(1−ǫ)2 .
![Page 71: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/71.jpg)
Volume of Giant Component
Spectra of Random Graphs Linyuan Lu – 34 / 68
Chung and Lu (2004)If the average degree is strictly greater than 1, then almostsurely the giant component in a graph G in G(w) has
volume (λ0 +O(√
n log3.5 nVol(G) )
)
Vol(G), where λ0 is the unique
positive root of the following equation:
n∑
i=1
wie−wiλ = (1− λ)
n∑
i=1
wi.
![Page 72: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/72.jpg)
A real application
Spectra of Random Graphs Linyuan Lu – 35 / 68
Apply to the Collaboration Graph (2002 data):The size of giant component is predicted to be about177, 400 by our theory. This is rather close to the actualvalue 176, 000, within an error bound of less than 1%.
![Page 73: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/73.jpg)
G(n, p) versus G(w1, . . . , wn)
Spectra of Random Graphs Linyuan Lu – 36 / 68
Question: Does the random graph with equal expecteddegrees generates the smallest giant component among allpossible degree distribution with the same volume?
![Page 74: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/74.jpg)
G(n, p) versus G(w1, . . . , wn)
Spectra of Random Graphs Linyuan Lu – 36 / 68
Question: Does the random graph with equal expecteddegrees generates the smallest giant component among allpossible degree distribution with the same volume?Chung Lu (2004)
■ Yes, for 1 < d ≤ ee−1 .
![Page 75: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/75.jpg)
G(n, p) versus G(w1, . . . , wn)
Spectra of Random Graphs Linyuan Lu – 36 / 68
Question: Does the random graph with equal expecteddegrees generates the smallest giant component among allpossible degree distribution with the same volume?Chung Lu (2004)
■ Yes, for 1 < d ≤ ee−1 .
■ No, for sufficiently large d.
![Page 76: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/76.jpg)
G(n, p) versus G(w1, . . . , wn)
Spectra of Random Graphs Linyuan Lu – 36 / 68
Question: Does the random graph with equal expecteddegrees generates the smallest giant component among allpossible degree distribution with the same volume?Chung Lu (2004)
■ Yes, for 1 < d ≤ ee−1 .
■ No, for sufficiently large d.■ When d ≥ 4
e , almost surely the giant component ofG(w1, . . . , wn) has volume at least
(
1
2
(
1 +
√
1− 4
de
)
+ o(1)
)
Vol(G).
This is asymptotically best possible.
![Page 77: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/77.jpg)
Diameter of G(w1, . . . , wn)
Spectra of Random Graphs Linyuan Lu – 37 / 68
Chung Lu (2002)
■ For a random graph G with admissible expected degreesequence (w1, . . . , wn), the average distance is almostsurely (1 + o(1)) logn
log d.
![Page 78: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/78.jpg)
Diameter of G(w1, . . . , wn)
Spectra of Random Graphs Linyuan Lu – 37 / 68
Chung Lu (2002)
■ For a random graph G with admissible expected degreesequence (w1, . . . , wn), the average distance is almostsurely (1 + o(1)) logn
log d.
■ For a random graph G with strongly admissible expecteddegree sequence (w1, . . . , wn), the diameter is almostsurely Θ( log n
log d).
![Page 79: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/79.jpg)
Diameter of G(w1, . . . , wn)
Spectra of Random Graphs Linyuan Lu – 37 / 68
Chung Lu (2002)
■ For a random graph G with admissible expected degreesequence (w1, . . . , wn), the average distance is almostsurely (1 + o(1)) logn
log d.
■ For a random graph G with strongly admissible expecteddegree sequence (w1, . . . , wn), the diameter is almostsurely Θ( log n
log d).
These results apply to G(n, p) and random power law graphwith β > 3.
![Page 80: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/80.jpg)
Non-admissible graph
versus admissible graph
Spectra of Random Graphs Linyuan Lu – 38 / 68
A random subgraph of the Collabo-
ration Graph.
A Connected component of G(n, p)
with n = 500 and p = 0.002.
![Page 81: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/81.jpg)
Non-admissible graph
versus admissible graph
Spectra of Random Graphs Linyuan Lu – 38 / 68
A random subgraph of the Collabo-
ration Graph.
A Connected component of G(n, p)
with n = 500 and p = 0.002.
- Dense core for non-admissible graphs.- No dense core for admissible graphs.
![Page 82: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/82.jpg)
Power law graphs with β ∈ (2, 3)
Spectra of Random Graphs Linyuan Lu – 39 / 68
Chung, Lu (2002)
- Examples: the WWW graph, Collaboration graph, etc.
![Page 83: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/83.jpg)
Power law graphs with β ∈ (2, 3)
Spectra of Random Graphs Linyuan Lu – 39 / 68
Chung, Lu (2002)
- Examples: the WWW graph, Collaboration graph, etc.- Non-admissible.
![Page 84: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/84.jpg)
Power law graphs with β ∈ (2, 3)
Spectra of Random Graphs Linyuan Lu – 39 / 68
Chung, Lu (2002)
- Examples: the WWW graph, Collaboration graph, etc.- Non-admissible.- Containing a dense core, with diameter log log n.
![Page 85: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/85.jpg)
Power law graphs with β ∈ (2, 3)
Spectra of Random Graphs Linyuan Lu – 39 / 68
Chung, Lu (2002)
- Examples: the WWW graph, Collaboration graph, etc.- Non-admissible.- Containing a dense core, with diameter log log n.- Mostly vertices are within the distance of O(log log n)from the core.
![Page 86: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/86.jpg)
Power law graphs with β ∈ (2, 3)
Spectra of Random Graphs Linyuan Lu – 39 / 68
Chung, Lu (2002)
- Examples: the WWW graph, Collaboration graph, etc.- Non-admissible.- Containing a dense core, with diameter log log n.- Mostly vertices are within the distance of O(log log n)from the core.
- There are some vertices at the distance of O(log n).
![Page 87: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/87.jpg)
Power law graphs with β ∈ (2, 3)
Spectra of Random Graphs Linyuan Lu – 39 / 68
Chung, Lu (2002)
- Examples: the WWW graph, Collaboration graph, etc.- Non-admissible.- Containing a dense core, with diameter log log n.- Mostly vertices are within the distance of O(log log n)from the core.
- There are some vertices at the distance of O(log n).
The diameter is Θ(log n), while the average distance isO(log log n).
![Page 88: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/88.jpg)
Experimental results
Spectra of Random Graphs Linyuan Lu – 40 / 68
■ Faloutsos et al. (1999) The eigenvalues of theInternet graph do not follow the semicircle law.
■ Farkas et. al. (2001), Goh et. al. (2001) Thespectrum of a power law graph follows a “triangular-like”distribution.
■ Mihail and Papadimitriou (2002) They showed thatthe large eigenvalues are determined by the largedegrees. Thus, the significant part of the spectrum of apower law graph follows the power law.
µi ≈√
di.
![Page 89: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/89.jpg)
Eigenvalues of G(w1, . . . , wn)
Spectra of Random Graphs Linyuan Lu – 41 / 68
Chung, Vu, and Lu (2003)Suppose w1 ≥ w2 ≥ . . . ≥ wn. Let µi be i-th largesteigenvalue of G(w1, w2, . . . , wn). Let m = w1 andd =
∑ni=1w
2i ρ. Almost surely we have:
■ (1−o(1))max{√m, d} ≤ µ1 ≤ 7√log n ·max{√m, d}.
![Page 90: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/90.jpg)
Eigenvalues of G(w1, . . . , wn)
Spectra of Random Graphs Linyuan Lu – 41 / 68
Chung, Vu, and Lu (2003)Suppose w1 ≥ w2 ≥ . . . ≥ wn. Let µi be i-th largesteigenvalue of G(w1, w2, . . . , wn). Let m = w1 andd =
∑ni=1w
2i ρ. Almost surely we have:
■ (1−o(1))max{√m, d} ≤ µ1 ≤ 7√log n ·max{√m, d}.
■ µ1 = (1 + o(1))d, if d >√m log n.
![Page 91: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/91.jpg)
Eigenvalues of G(w1, . . . , wn)
Spectra of Random Graphs Linyuan Lu – 41 / 68
Chung, Vu, and Lu (2003)Suppose w1 ≥ w2 ≥ . . . ≥ wn. Let µi be i-th largesteigenvalue of G(w1, w2, . . . , wn). Let m = w1 andd =
∑ni=1w
2i ρ. Almost surely we have:
■ (1−o(1))max{√m, d} ≤ µ1 ≤ 7√log n ·max{√m, d}.
■ µ1 = (1 + o(1))d, if d >√m log n.
■ µ1 = (1 + o(1))√m, if
√m > d log2 n.
![Page 92: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/92.jpg)
Eigenvalues of G(w1, . . . , wn)
Spectra of Random Graphs Linyuan Lu – 41 / 68
Chung, Vu, and Lu (2003)Suppose w1 ≥ w2 ≥ . . . ≥ wn. Let µi be i-th largesteigenvalue of G(w1, w2, . . . , wn). Let m = w1 andd =
∑ni=1w
2i ρ. Almost surely we have:
■ (1−o(1))max{√m, d} ≤ µ1 ≤ 7√log n ·max{√m, d}.
■ µ1 = (1 + o(1))d, if d >√m log n.
■ µ1 = (1 + o(1))√m, if
√m > d log2 n.
■ µk ≈√wk and µn+1−k ≈ −√
wk, if√wk > d log2 n.
![Page 93: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/93.jpg)
Random power law graphs
Spectra of Random Graphs Linyuan Lu – 42 / 68
The first k and last k eigenvalues of the random power lawgraph with β > 2.5 follows the power law distribution withexponent 2β − 1. It results a “triangular-like” shape.
![Page 94: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/94.jpg)
Laplacian spectrum
Spectra of Random Graphs Linyuan Lu – 43 / 68
Random walks on a graph G:
πk+1 = AD−1πk.
AD−1 ∼ D−1/2AD−1/2.✍✌✎☞v ✍✌
✎☞
✍✌✎☞
✍✌✎☞
✲��������✒✻
1dv
1dv1
dv
![Page 95: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/95.jpg)
Laplacian spectrum
Spectra of Random Graphs Linyuan Lu – 43 / 68
Random walks on a graph G:
πk+1 = AD−1πk.
AD−1 ∼ D−1/2AD−1/2.✍✌✎☞v ✍✌
✎☞
✍✌✎☞
✍✌✎☞
✲��������✒✻
1dv
1dv1
dv
Laplacian spectrum
0 = λ0 ≤ λ1 ≤ · · · ≤ λn−1 ≤ 2
are the eigenvalues of L = I −D−1/2AD−1/2.
![Page 96: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/96.jpg)
Laplacian spectrum
Spectra of Random Graphs Linyuan Lu – 43 / 68
Random walks on a graph G:
πk+1 = AD−1πk.
AD−1 ∼ D−1/2AD−1/2.✍✌✎☞v ✍✌
✎☞
✍✌✎☞
✍✌✎☞
✲��������✒✻
1dv
1dv1
dv
Laplacian spectrum
0 = λ0 ≤ λ1 ≤ · · · ≤ λn−1 ≤ 2
are the eigenvalues of L = I −D−1/2AD−1/2.The eigenvalues of AD−1 are 1, 1− λ1, . . . , 1− λn−1.
![Page 97: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/97.jpg)
Laplacian Spectral Radius
Spectra of Random Graphs Linyuan Lu – 44 / 68
Let
- wmin = min{w1, . . . , wn},- d = 1
n
∑ni=1wi,
- g(n) — a function tending to infinity arbitrarily slowly.
Chung, Vu, and Lu (2003)
■ If wmin ≫ log2 n, then almost surely the Laplacianspectrum λi’s of G(w1, . . . , wn) satisfy
maxi6=0
|1− λi| ≤ (1 + o(1))4√d+
g(n) log2 n
wmin.
![Page 98: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/98.jpg)
Laplacian Spectral Radius
Spectra of Random Graphs Linyuan Lu – 44 / 68
Let
- wmin = min{w1, . . . , wn},- d = 1
n
∑ni=1wi,
- g(n) — a function tending to infinity arbitrarily slowly.
Chung, Vu, and Lu (2003)
■ If wmin ≫ log2 n, then almost surely the Laplacianspectrum λi’s of G(w1, . . . , wn) satisfy
maxi6=0
|1− λi| ≤ (1 + o(1))4√d+
g(n) log2 n
wmin.
■ If wmin ≫√d, the Laplacian spectrum follows the
semi-circle distribution with radius r ≈ 2√d.
![Page 99: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/99.jpg)
General random graphs
Spectra of Random Graphs Linyuan Lu – 45 / 68
General edge-independent random graphs:
■ n: the number of vertices.
■ pij: a probability for ij being an edge.
■ Edges are mutually independent.
Question: What can we say about the spectrum of theadjacency matrix and the Laplacian matrix?
![Page 100: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/100.jpg)
Notation
Spectra of Random Graphs Linyuan Lu – 46 / 68
- A: adjacency matrix
- A := (pij): the expectation of A
- ∆: the maximum expected degree
- δ: the minimum expected degree
- D: the diagonal matrix of degrees
- D: the expectation of D
- L := I −D−1/2AD−1/2: the normalized Laplacian
- L := I − D−1/2AD−1/2: the Laplacian of A
![Page 101: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/101.jpg)
Known results
Spectra of Random Graphs Linyuan Lu – 47 / 68
Oliveira [2010]: For ∆ ≥ C lnn, with high probability wehave
|λi(A)− λi(A)| ≤ 4√∆ lnn.
For δ ≥ C lnn, with high probability we have
λi(L)− λi(L) ≤ 14√
ln(4n)/δ.
![Page 102: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/102.jpg)
Known results
Spectra of Random Graphs Linyuan Lu – 47 / 68
Oliveira [2010]: For ∆ ≥ C lnn, with high probability wehave
|λi(A)− λi(A)| ≤ 4√∆ lnn.
For δ ≥ C lnn, with high probability we have
λi(L)− λi(L) ≤ 14√
ln(4n)/δ.
Chung-Radcliffe [2011] reduces the constant coefficient
using a new matrix Chernoff inequality.
![Page 103: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/103.jpg)
Our results
Spectra of Random Graphs Linyuan Lu – 48 / 68
Lu-Peng [2012+]: If ∆ ≫ ln4 n, then almost surely
|λi(A)− λi(A)| ≤ (2 + o(1))√∆.
![Page 104: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/104.jpg)
Our results
Spectra of Random Graphs Linyuan Lu – 48 / 68
Lu-Peng [2012+]: If ∆ ≫ ln4 n, then almost surely
|λi(A)− λi(A)| ≤ (2 + o(1))√∆.
Lu-Peng [2012+]:
Let Λ := {λi(L) : |1− λi(L)| = ω(1/√lnn)}.
If δ ≫ max{|Λ|, ln4 n}, then almost surely
|λi(L)− λi(L)| ≤
2 +
√
∑
λ∈Λ(1− λ)2 + o(1)
1√δ.
![Page 105: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/105.jpg)
Our results
Spectra of Random Graphs Linyuan Lu – 48 / 68
Lu-Peng [2012+]: If ∆ ≫ ln4 n, then almost surely
|λi(A)− λi(A)| ≤ (2 + o(1))√∆.
Lu-Peng [2012+]:
Let Λ := {λi(L) : |1− λi(L)| = ω(1/√lnn)}.
If δ ≫ max{|Λ|, ln4 n}, then almost surely
|λi(L)− λi(L)| ≤
2 +
√
∑
λ∈Λ(1− λ)2 + o(1)
1√δ.
In both case, we remove the multiplicative factor√lnn.
![Page 106: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/106.jpg)
Random symmetric matrices
Spectra of Random Graphs Linyuan Lu – 49 / 68
B = (bij) is a random symmetric matrix satisfying:
- bij: independent, but not necessary identical,- |bij| ≤ K,- E(bij) = 0,- Var(bij) ≤ σ2.
![Page 107: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/107.jpg)
Random symmetric matrices
Spectra of Random Graphs Linyuan Lu – 49 / 68
B = (bij) is a random symmetric matrix satisfying:
- bij: independent, but not necessary identical,- |bij| ≤ K,- E(bij) = 0,- Var(bij) ≤ σ2.
Furedi-Komlos [1981]:
‖B‖ ≤ 2σ√n+ cn1/3 lnn.
![Page 108: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/108.jpg)
Random symmetric matrices
Spectra of Random Graphs Linyuan Lu – 49 / 68
B = (bij) is a random symmetric matrix satisfying:
- bij: independent, but not necessary identical,- |bij| ≤ K,- E(bij) = 0,- Var(bij) ≤ σ2.
Furedi-Komlos [1981]:
‖B‖ ≤ 2σ√n+ cn1/3 lnn.
Vu [2007]:
‖B‖ ≤ 2σ√n+ c(Kσ)1/2n1/4 lnn.
![Page 109: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/109.jpg)
Our result
Spectra of Random Graphs Linyuan Lu – 50 / 68
Lu-Peng [2012+]: We further assume Var(bij) ≤ σ2ij. Let
∆ := max1≤i≤n
∑nj=1 σ
2ij. If ∆ ≥ C ′K2 ln4 n, then
asymptotically almost surely
‖B‖ ≤ 2√∆+ C
√K∆1/4 lnn.
■ It generalizes Vu’s theorem.
■ This result is asymptotically tight.
![Page 110: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/110.jpg)
Graph percolation
Spectra of Random Graphs Linyuan Lu – 51 / 68
■ G: a connected graph on n vertices
■ p: a probability (0 ≤ p ≤ 1)
Gp: a random spanning subgraph of G, obtained as follows:for each edge f of G, independently,
Pr(f is an edge of Gp) = p.
![Page 111: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/111.jpg)
Graph percolation
Spectra of Random Graphs Linyuan Lu – 51 / 68
■ G: a connected graph on n vertices
■ p: a probability (0 ≤ p ≤ 1)
Gp: a random spanning subgraph of G, obtained as follows:for each edge f of G, independently,
Pr(f is an edge of Gp) = p.
![Page 112: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/112.jpg)
Spectrum of Gp
Spectra of Random Graphs Linyuan Lu – 52 / 68
Lu-Peng [2012+]:
■ If p ≫ ln4 n∆ , then almost surely we have
|λi(A(Gp))− pλi(A(G))| ≤ (2 + o(1))√
p∆.
![Page 113: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/113.jpg)
Spectrum of Gp
Spectra of Random Graphs Linyuan Lu – 52 / 68
Lu-Peng [2012+]:
■ If p ≫ ln4 n∆ , then almost surely we have
|λi(A(Gp))− pλi(A(G))| ≤ (2 + o(1))√
p∆.
■ Suppose that all but k Laplacian eigenvalues λ of Gsatisfies |1− λ| = o( 1√
lnn). If δ ≫ max{k, ln4 n}, then
for p ≫ max{kδ ,
ln4 nδ }, almost surely we have
|λi(L(Gp))−λi(L(G))|≤(2+√
∑ki=1(1− λi)2+o(1)))
1√pδ.
�
![Page 114: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/114.jpg)
Method
Spectra of Random Graphs Linyuan Lu – 53 / 68
We will illustrate Wigner’s trace method through the sketchproof of the following result.
![Page 115: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/115.jpg)
Method
Spectra of Random Graphs Linyuan Lu – 53 / 68
We will illustrate Wigner’s trace method through the sketchproof of the following result.
Lu-Peng [2012+]: If B = (bij) is a random symmetricmatrix satisfying:
- bij: independent, but not necessary identical,- |bij| ≤ K,- E(bij) = 0,- Var(bij) ≤ σ2
ij.
then almost surely
‖B‖ ≤ 2√∆+ C
√K∆1/4 lnn,
where ∆ := max1≤i≤n
∑nj=1 σ
2ij.
![Page 116: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/116.jpg)
Sketch proof
Spectra of Random Graphs Linyuan Lu – 54 / 68
WLOG, we can assume K = 1 and bii = 0. Using Wigner’strace method, we have
E(
Trace(Bk))
=∑
i1,i2,...,ik
E(bi1i2bi2i3 . . . bik−1ikbiki1)
=
⌊k/2⌋+1∑
p=2
∑
w∈G(n,k,p)
∏
e∈E(w)
E(bqee ).
Here G(n, k, p) is the set of “good” closed walks w in Kn oflength k on p vertices, where each edge in w appears morethan once (qe ≥ 2).
![Page 117: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/117.jpg)
Continue
Spectra of Random Graphs Linyuan Lu – 55 / 68
Let G(k, p) be the set of good closed walks w of length k onthe complete graph Kp where vertices first appear in w inthe order 1, 2, . . . , p.
![Page 118: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/118.jpg)
Continue
Spectra of Random Graphs Linyuan Lu – 55 / 68
Let G(k, p) be the set of good closed walks w of length k onthe complete graph Kp where vertices first appear in w inthe order 1, 2, . . . , p.
All walks in G(n, k, p) can be coded by a walk in G(k, p) plusthe ordered p distinct vertices. Let[n]p := {(v1, v2, . . . , vp) ∈ [n]p : v1, v2, . . . , vp are distinct}.
![Page 119: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/119.jpg)
Continue
Spectra of Random Graphs Linyuan Lu – 55 / 68
Let G(k, p) be the set of good closed walks w of length k onthe complete graph Kp where vertices first appear in w inthe order 1, 2, . . . , p.
All walks in G(n, k, p) can be coded by a walk in G(k, p) plusthe ordered p distinct vertices. Let[n]p := {(v1, v2, . . . , vp) ∈ [n]p : v1, v2, . . . , vp are distinct}.Define a rooted tree T (w) so that the edgeijij+1 ∈ E(T (w)) if it brings in a new vertex ij+1 when itoccurs first time.
![Page 120: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/120.jpg)
continue
Spectra of Random Graphs Linyuan Lu – 56 / 68
∑
w∈G(n,k,p)
∏
e∈E(w)
σ2e =
∑
w∈G(k,p)
∑
(v1,...,vp)∈[n]p
∏
xy∈E(w)
σ2vxvy
≤∑
w∈G(k,p)
n∑
v1=1
n∑
v2=1
· · ·n
∑
vp=1
∏
xy∈E(T )
σ2vxvy
=∑
w∈G(k,p)
n∑
v1=1
n∑
v2=1
· · ·n
∑
vp−1=1
p−1∏
y=2
σ2vη(y)vy
n∑
vp=1
σ2vη(p)vp
≤ ∆∑
w∈G(k,p)
n∑
v1=1
n∑
v2=1
· · ·n
∑
vp−1=1
p−1∏
y=2
σ2vη(y)vy
≤ · · ·≤ n∆p−1
∣
∣
∣G(k, p)
∣
∣
∣.
![Page 121: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/121.jpg)
Continue
Spectra of Random Graphs Linyuan Lu – 57 / 68
Vu [2007] proved
|G(k, p)| ≤(
k
2p− 2
)
22k−2p+3pk−2p+2(k − 2p+ 4)k−2p+2.
![Page 122: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/122.jpg)
Continue
Spectra of Random Graphs Linyuan Lu – 57 / 68
Vu [2007] proved
|G(k, p)| ≤(
k
2p− 2
)
22k−2p+3pk−2p+2(k − 2p+ 4)k−2p+2.
We get
∣
∣E(
Trace(Bk))∣
∣ ≤∑
w∈G(n,k)
∏
e∈E(w)
σ2e ≤
k/2+1∑
p=2
n∆p−1∣
∣
∣G(k, p)
∣
∣
∣
≤ n
k/2+1∑
p=2
∆p−1
(
k
2p− 2
)
22k−2p+3pk−2p+2(k − 2p+ 4)k−2p+2
:= n
k/2+1∑
p=2
S(n, k, p).
![Page 123: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/123.jpg)
Continue
Spectra of Random Graphs Linyuan Lu – 58 / 68
One can show
S(n, k, p− 1) ≤ 16k4
∆S(n, k, p).
![Page 124: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/124.jpg)
Continue
Spectra of Random Graphs Linyuan Lu – 58 / 68
One can show
S(n, k, p− 1) ≤ 16k4
∆S(n, k, p).
For any even integer k such that k4 ≤ ∆32 , we get
∣
∣E(
Trace(Bk))∣
∣ ≤k/2+1∑
p=2
S(n, k, p)
≤ S(n, k, k/2 + 1)
k/2+1∑
p=2
(
1
2
)k/2+1−p
< 2S(n, k, k/2 + 1)
= n2k+2∆k/2.
![Page 125: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/125.jpg)
continue
Spectra of Random Graphs Linyuan Lu – 59 / 68
For even k, we have
Pr(‖B‖ ≥ 2√∆+ C∆1/4 lnn)
= Pr(‖B‖k ≥ (2√∆+ C∆1/4 lnn)k)
≤ Pr(Trace(Bk) ≥ (2√∆+ C∆1/4 lnn)k)
≤ E(Trace(Bk))
(2√∆+ C∆1/4 lnn))k
(Markov’s inequality)
≤ n2k+2∆k/2
(2√∆+ C∆1/4 lnn))k
= 4ne−(1+o(1))C2 k∆−1/4 lnn.
Setting k =(
∆32
)1/4, this probability is o(1) for sufficiently
large C. �
![Page 126: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/126.jpg)
continue
Spectra of Random Graphs Linyuan Lu – 59 / 68
For even k, we have
Pr(‖B‖ ≥ 2√∆+ C∆1/4 lnn)
= Pr(‖B‖k ≥ (2√∆+ C∆1/4 lnn)k)
≤ Pr(Trace(Bk) ≥ (2√∆+ C∆1/4 lnn)k)
≤ E(Trace(Bk))
(2√∆+ C∆1/4 lnn))k
(Markov’s inequality)
≤ n2k+2∆k/2
(2√∆+ C∆1/4 lnn))k
= 4ne−(1+o(1))C2 k∆−1/4 lnn.
Setting k =(
∆32
)1/4, this probability is o(1) for sufficiently
large C. �
![Page 127: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/127.jpg)
Percolation threshold pc
Spectra of Random Graphs Linyuan Lu – 60 / 68
■ For p < pc, almost surely there is no giant component
■ For p > pc, almost surely there is a giant component.
pc
![Page 128: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/128.jpg)
Motivations
Spectra of Random Graphs Linyuan Lu – 61 / 68
■ Graph theory: random graphs
■ Theoretical physics: crystals melting
■ Sociology: the spread of disease on contact networks
![Page 129: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/129.jpg)
Percolation of Zd
Spectra of Random Graphs Linyuan Lu – 62 / 68
![Page 130: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/130.jpg)
Percolation of Zd
Spectra of Random Graphs Linyuan Lu – 63 / 68
![Page 131: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/131.jpg)
Percolation of Zd
Spectra of Random Graphs Linyuan Lu – 63 / 68
Kesten (1980): pc(Z2) = 1
2 .
![Page 132: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/132.jpg)
Percolation of Zd
Spectra of Random Graphs Linyuan Lu – 63 / 68
Kesten (1980): pc(Z2) = 1
2 .
Lorenz and Ziff (1997, simulation):pc(Z
3) ≈ 0.2488126± 0.0000005 if it exists.
![Page 133: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/133.jpg)
Percolation of Zd
Spectra of Random Graphs Linyuan Lu – 63 / 68
Kesten (1980): pc(Z2) = 1
2 .
Lorenz and Ziff (1997, simulation):pc(Z
3) ≈ 0.2488126± 0.0000005 if it exists.
Kesten (1990): pc(Zd) ∼ 1
2d as d → ∞.
![Page 134: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/134.jpg)
d-regular graphs
Spectra of Random Graphs Linyuan Lu – 64 / 68
Alon, Benjamini, Stacey (2004): Suppose d ≥ 2 and let(Gn) be a sequence of d-regular expanders withgirth(Gn) → ∞, then
pc =1
d− 1+ o(1).
![Page 135: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/135.jpg)
Percolation of dense graphs
Spectra of Random Graphs Linyuan Lu – 65 / 68
Bollobas, Borgs, Chayes, and Riordan (2008): Supposethat G is a dense graph (i.e., average degree d = Θ(n)). Letµ be the largest eigenvalue of the adjacency matrix of G.Then
pc ≈1
µ.
![Page 136: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/136.jpg)
Percolation of dense graphs
Spectra of Random Graphs Linyuan Lu – 65 / 68
Bollobas, Borgs, Chayes, and Riordan (2008): Supposethat G is a dense graph (i.e., average degree d = Θ(n)). Letµ be the largest eigenvalue of the adjacency matrix of G.Then
pc ≈1
µ.
Remark: The requirement of “dense graph” is essential.Their methods can not be extended to sparse graphs.
![Page 137: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/137.jpg)
Percolation of sparse graphs
Spectra of Random Graphs Linyuan Lu – 66 / 68
Chung, Lu, Horn [2008]:
■ If p < 1µ , then Gp has no giant component.
![Page 138: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/138.jpg)
Percolation of sparse graphs
Spectra of Random Graphs Linyuan Lu – 66 / 68
Chung, Lu, Horn [2008]:
■ If p < 1µ , then Gp has no giant component.
■ The condition p > 1µ in general does not imply that Gp
has a giant component.
![Page 139: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/139.jpg)
Percolation of sparse graphs
Spectra of Random Graphs Linyuan Lu – 66 / 68
Chung, Lu, Horn [2008]:
■ If p < 1µ , then Gp has no giant component.
■ The condition p > 1µ in general does not imply that Gp
has a giant component.
■ If p > 1µ , ∆ = O(d), and σ = o( 1
log n), then Gp has agiant component.
![Page 140: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/140.jpg)
Percolation of G(w)
Spectra of Random Graphs Linyuan Lu – 67 / 68
Bhamidi-van der Hofstad-van Leeuwaarden [2012]:Consider G(w), where w = (w1, . . . , wn) follows the powerlaw of exponent β. If E(
∑ni=1w
2i ) converges and is bounded,
then the percolation threshed is (1 + o(1))1d.
■ For β > 4, E(∑n
i=1w3i ) converges. The largest
component has the size Θ(n2/3) at the critical window.
![Page 141: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/141.jpg)
Percolation of G(w)
Spectra of Random Graphs Linyuan Lu – 67 / 68
Bhamidi-van der Hofstad-van Leeuwaarden [2012]:Consider G(w), where w = (w1, . . . , wn) follows the powerlaw of exponent β. If E(
∑ni=1w
2i ) converges and is bounded,
then the percolation threshed is (1 + o(1))1d.
■ For β > 4, E(∑n
i=1w3i ) converges. The largest
component has the size Θ(n2/3) at the critical window.
■ For 2 < β < 3, E(∑n
i=1w3i ) diverges. The largest
component has the size Θ(nβ−2β−1 ) at the critical window.
![Page 142: Spectra of Random Graphs - University of South Carolinapeople.math.sc.edu/lu/talks/nankai_2014/spec_nankai_3.pdf · 2014-06-01 · Spectra of Random Graphs Linyuan Lu University of](https://reader033.fdocuments.us/reader033/viewer/2022053014/5f10ffdee5752f61792b3f35/html5/thumbnails/142.jpg)
References
Spectra of Random Graphs Linyuan Lu – 68 / 68
1. Fan Chung, Linyuan Lu, and Van Vu, Eigenvalues of random powerlaw graphs, Annals of Combinatorics, 7 (2003), 21–33.
2. Fan Chung, Linyuan Lu and Van Vu, The spectra of random graphswith given expected degrees, Proceedings of National Academy of
Sciences, 100, No. 11, (2003), 6313-6318.3. Fan Chung, Linyuan Lu, and Van Vu, Eigenvalues of random power
law graphs, Internet Mathematics, 1 No. 3, (2004), 257–275.4. Fan Chung and Linyuan Lu, The volume of the giant component for
a random graph with given expected degrees, SIAM J. Discrete
Math., 20 (2006), No. 2, 395–411.5. Linyuan Lu and Xing Peng, Spectra of edge-independent random
graphs, Electronic Journal of Combinatorics, 20 (4), (2013) P27.
Homepage: http://www.math.sc.edu/∼ lu/
Thank You