In graph theory, an expander graph is a sparse graph that has strong connectivity properties, quantified using vertex, edge or spectral expansion. Expander constructions have spawned research in pure and applied mathematics, with several applications to complexity theory, design of robust computer networks, and the theory of error-correcting codes.
Introduction to Expander Graphs
Measures of Expansion
First, there are two combinatorial definitions of expansion.
Definition (Edge Expansion): A D-regular graph G is a (K,ϵ) edge expander if for all sets S of at most K vertices, the cut size e(S,Sˉ) is at least ϵ∣S∣D.
In addition, we define h(G) by:
h(G):=S⊂V and ∣S∣≤∣V∣/2mind∣S∣e(S,Sˉ)
Definition (Vertex Expansion): A graph G is a (K,A) vertex expander if for all sets S of at most K vertices, the neighborhood N(S):={u∣∃v∈S s.t. (u,v)∈E} is of size at least A⋅∣S∣.
And then, there is another definition of expansion.
Definition (Spectral Expansion): For γ∈[0,1], we say that a regular graph G has spectral expansion γ if γ(G):=1−λ(G)≥γ, where λ(G) is the second eigenvalue of the normalized adjacency matrix M(G).
Note that by Rayleigh theorem, we have:
λ(G)=π=0max∥π−1∥∥πM−1∥=x⊥1max∥x∥∥xM∥
Some Relations between Different Definitions
Expander Mixing Lemma
We will prove a famous result known as the expander mixing lemma, which gives a relation between a graph’s spectral expansion and a certain regularity or spreadness condition of its edges that is a slightly more general statement than edge-expansion itself.
Theorem (Expander Mixing Lemma): Let G be a D-regular, N vertex graph with spectral expansion 1−λ. Then for all sets of vertices S,T, we have:
Expander mixing lemma also has its converse. We present here without proving it.
Theorem (Converse to Expander Mixing Lemma): Let G be a D-regular, N-vertex graph. Suppose that for all pairs of disjoint vertex sets S,T, we have ∣e(S,T)/(ND)−μ(S)μ(T)∣≤θμ(S)μ(T) for some θ∈[0,1]. Then λ(G)=O(θlog(1/θ)).
Cheeger Inequaltiy
Theorem (Cheeger Inequality): Let G be a D-regular graph, then:
21−λ(G)≤h(G)≤2(1−λ(G))
Proof: (鸽子一会)□
Random Walks on Expanders
Due to the good struction of expanders, we can use the random walk on expanders to reduce the randomness needed by random algorithms.
Hitting Property of Expander Walks: If G is a regular graph with spectral expansion 1−λ, then for any B⊂V(G) of density μ, the probability that a random walk (V1,…,Vt) of t−1 steps in G starting in a uniformly random vertex V1 always remains in B is
Pr[i=1⋀tVi∈B]≤(μ+λ(1−μ))t
Proof: Firstly, we prove a lemma:
Lemma: The probability that the random walk stays entirely within B is precisely ∥1P(MP)t−1∥1, where P is defined as follow:
Pij={[i∈B]0i=ji=j
Proof of Lemma: We can easily prove this lemma by induction on t. ■
And there becomes anther lemma:
Lemma:∥PMP∥≤μ+λ(1−μ), where ∥⋅∥ denotes the spectral norm of matrices.
Proof of Lemma: We decompose M into two parts:
M=γJ+λE
Then we have:
∥PMP∥=∥P(γJ+λE)P∥≤γ∥PJP∥+λ∥PEP∥≤γ∥PJP∥+λ
Noet that the last inequality uses the fact that ∥E∥≤1.
For all x∈RN, let y be xP, then xPJP=yJP=(∑iyi)1P.
To reduce the randomness of a RP algorithm A which uses m random bits, the general approach is to consider an expander graph with vertex set ${0,1}^m, where each vertex is associated with a setting of the random bits. We will choose a uniformly random vertex V1 and then do a random walk of length t−1, visiting additional vertices V2,…,Vt. This requires m random bits for the initial choice, and logD for each of the t−1 steps. For every vertex Vi on the random walk, we will run the algorithm with Vi as the setting of the random coins.
Then by the theorem above, in order to reduce the error probability to 2−k, the length of the random walk should be Θ(k). Thus, the number of random bits is m+O(k).
The success of the error-reduction procedure for two-sided error algorithms is obtained by the following theorem.
Chernoff Bound for Expander Walk: Let G be a regular graph with N vertices and spectral expansion 1−λ, and let B∈[N] be a subset of vertices. Consider a random walk V1,…,Vt in G from a uniform start vertex V1. Then for any ϵ>0
Corollary: If a language L has a BPP algorithm with error probability at most 1/3 that uses m(n) random bits on inputs of length n, then for every polynomial k(n), L has a BPP algorithm with error probability at most 2−k(n) that uses m(n)+O(k(n)) random bits.
Optimal Expanders
How good can the expansion of a graph be?
For vertex expansion, the following theorem shows its limitation.
Theorem (Limits on Vertex Expansion):
(1) Show that if a D-regular graph G is a (K,A) expander, then TD is a (K,A) expander, where TD is the infinite D-regular tree.
(2) Show that for every D∈N, there are infinitely many K∈N such that TD is not a (K,D−1+2/K) expander.
(3) Deduce that for constant D∈N and α>0, if a D-regular, N-vertex graph G is an (αN,A) vertex expander, then A≤D−1+o(1), where the o(1) term vanishes as N→∞ (and D,α are held constant).
For spectral expansion, the result is known as Alon–Boppana Bound.
Theorem (Alon–Boppana Bound): Let G be a D-regular graph, then:
λ(G)≥D2D−1−o(1)
where the o(1) term vanishes as N→∞.
Proof: For a graph H, let pl(H) denote the probability that if we choose a random vertex v in H and do a random walk of length 2l, we end back at vertex v. And let TD be the infinite D-regular tree.
Obviously, we have: pl(G)≥pl(TD).
A random walk of length 2l which end back at starting vertex can be seen as a properly parenthesized string of length 2l with at least D−1 types of parentheses.
Thus: pl(TD)≥Cl(D−1)l/D2l, where Cl is the l-th Catalan number.
In addition, we have: tr(M2l)=1+λ22l+⋯+λn2l≤1+(N−1)λ2l(G) and: pl(G)=∑v∈Gpl,v(G)/N=tr(M2l)/N.
Thus we have: Npl(G)≤1+(N−1)λ2l(G).
Notice that Cl=(l2l)/(l+1). Then using the inequality proved above, we have:
The optimal spectral expander is called as the Ramanujan graph.
Definition (Ramanujan Graphs): A d-regular graph G is called as a Ramnujan graph if λ(G)≤D2D−1+o(1).
We will discuss some explicit constructions of Ramnujan graphs in the next section.
Explicit Constructions
A random graph will be a good expander but we will not choose it as an expander. Some reasons we don’t choose it:
Don’t want to tolerate the error probability that the graph isn’t an expander, and checking it is NP-Hard problem.
Defeat the purpose that we want to reduce randomness.
Require exponentially large expander graphs, and thus we can’t even write down a randomly chosen expander.
Mildly Explicit: Construct a complete representation of the graph in time an poly(N).
Fully Explicit: Given a mode u∈[N] and i∈[D], where D is the degree of the expander, compute the ith nerghbor of u in time poly(logN). (more efficiently)
Goal: Devise a fully explicit construction of an infinite family {Gi} of D-regular graphs with spectral expansion at least γ, where D and γ>0 are constants independent of i.
Algebraic Constructions
Discrete torus expanders, p-cycle with inverse chords.
Ramanujan Graphs:G=(V,E) is a graph with vertex set V=Fq∪{∞}, the finite field of prime order q s.t. q≡1(mod4) plus one extra node representing infinity. The edges in this graph connect each node z with all z′ of the form:
z′=(−a2+ia3)z+(a0−ia1)(a0+ia1)z+(a2+ia3)
for a0,a1,a2,a3∈Z such that a02+a12+a22+a32=p, a0 is odd and positive, and a1,a2,a3 are even, for some fixed prime p=q such that p≡1(mod4), q is a square modulo p, and i∈Fq such that i2=−1(modq).
It’s an optimal spectral expander. (?, omit the proof)
Graph Operations
Definition: An (N,D,γ)-graph is a D-regular graph on N vertices with spectral expansion γ.
Squaring
Definition (Squaring of Graphs): If G=(V,E) is a D-regular graph, then G2=(V,E′) is a D2-regular graph on the same vertex set, where the (i,j)th neighbor of a vertex x is the jth neighbor of the ith neighbor of x.
Lemma: If G is a (N,D,1−λ)-graph, then G2 is a (N,D2,1−λ2)-graph.
Proof: Let M be the random walk matrix for G, and then M2 is the random walk matrix for G2. Thus: ∀x⊥1,∥xM2∥≤λ∥xM∥≤λ2∥x∥. □
Tensor
Definition (Tensor Product of Graphs): Let G1=(V1,E1) be D1-regular and G2=(V2,E2) be D2-regular. Then their tensor product is the D1D2-regular graph G1⊗G2=(V1×V2,E), where the (i1,i2)th neighbor of a vertex (x1,x2) is (y1,y2), where yb is the ibth neighbor of xb in Gb.
A Few Comments on the Tensor of Matrix and Vector:
Lemma: If G1 is an (N1,D1,γ1)-graph and G2 is an (N2,D2,γ2)- graph, then G1⊗G2 is an (N1N2,D1D2,min{γ1,γ2})-graph.
Proof: Let M1,M2 be the random walk matrix for G1 and G2, and then M≜M1⊗M2 is the random walk matrix for G1⊗G2.
∀x(∈RN1N2)⊥1N1N2, decompose x as x=x∥+x⊥, where x∥ is a multiple of 1N2 on each cloud of size N2 and x⊥ is orthogonal to 1N2 on each cloud.
Note that ⟨x⊥,1N1N2⟩=S∈each cloud∑⟨x⊥∣S,1N2⟩=0. Thus, ⟨x∥,1N1N2⟩=⟨x,1N1N2⟩−⟨x⊥,1N1N2⟩=0, i.e. x∥⊥1N1N2. Therefore, if we write x⊥ as x⊥=y⊗1N2, then y⊥1N1.
For the first case, we have: x∥M=(y⊗1N2)(M1⊗M2)=(yM1)⊗1N2. The expansion G1 tells us that M1 shrinks y by a factor of λ1, and thus ∥x∥M∥=N2∥yM∥≤λ1N2∥y∥=λ1∥x∥∥.
For another case, we have: x⊥M=x⊥(IN1⊗M2)(M1⊗IN2). The expansion of G2 tells us that M2 will shrink x⊥ by a factor of λ2 on each cloud, and thus IN1⊗M2 will shrink x⊥ by the same factor. The subsequent application of M1⊗IN2 cannot increase the length. Thus, ∥x⊥M∥≤λ2∥x⊥∥.
Finally, we argue that x∥M and x⊥M are orthogonal. Note that x∥M=(yM1)⊗1N2 is a multiple of 1N2 on every cloud. Thus it suffices to argue that x⊥ remains orthogonal to 1N2 on every cloud after we apply M. Applying IN1⊗M2 retains this property and applying M1⊗IN2 retains this property because it assigns each cloud a linear combination of several other clouds.
Definition (Zig–Zag Product): Let G be a D1-regular digraph on N1 vertices, and H a D2-regular digraph on D1 vertices. Then GⓏH is a graph whose vertices are pairs (u,i)∈[N1]×[D1]. For a,b∈[D2], the (a,b)th neighbor of a vertex (u,i) is the vertex (v,j) computed as follows:
(1) Let i′ be the ath neighbor of i in H.
(2) Let v be the i′th neighbor of u in G, so e=(u,v) is the i′th edge leaving u. Let j′ be such that e is the j′th edge entering v in G.
(3) Let j be the bth neighbor of j′ in H.
Lemma: If G is a (N1,D1,γ1)-graph, and H is a (D1,D2,γ2)-graph then GⓏH is a (N1D1,D22,γ=γ1⋅γ22)-graph. In particular, if γ1=1−λ1 and γ2=1−λ2, then γ=1−λ for λ≤λ1+2λ2.
Proof: (TODO)
Replacement Product
Definition (Replacement Product): Given a D1-regular graph G1 on N1 vertices and a D2-regular graph G2 on D1 vertices, consider the following graph G1ⓡG2 on vertex set [N1]×[D1]: vertex (u,i) is connected to (v,j) iff u=v and (i,j) is an edge in G2, or v is the i’th neighbor of u in G1 and u is the jth neighbor of v.
(1) Prove that there is a function g such that if G1 has spectral expansion γ1>0 and G2 has spectral expansion γ2>0 (and both graphs are undirected), then G1ⓡG2 has spectral expansion g(γ1,γ2,D2)>0.
(2) Show how to convert an explicit construction of constant degree spectral expanders into an explicit construction of degree 3 spectral expanders.
(3) Without using Theorem 4.14, prove an analogue of Part 1 for edge expansion. That is, there is a function h such that if G1 is an (N1/2,ε1) edge expander and G2 is a (D1/2,ε2) edge expander, then G1 $r G2 is a (N1D1/2,h(ε1,ε2,D2)) edge expander, where h(ε1,ε2,D2) > 0 if ε1,ε2 > 0. (Hint: given any set S of vertices of G1 $r G2, partition S into the clouds that are more than “half-full” and those that are not.)
(4) Prove that the functions g(γ1,γ2,D2) and h(ε1,ε2,D2) must depend on D2, by showing that G1ⓡG2 cannot be a (N1D1/2,ε) edge expander if ε>1/(D2+1) and N1≥2.
The Expander Construction
Construction of Mildly Explicit Expanders: Let H be a (D4,D,7/8)-graph, and define:
G1=H2Gt+1=Gt2ⓏH
We can easily prove that for all t, Gt is a (D4t,D2,1/2)-graph.
To bound the time to compute neighbors in Gt, we let time(Gt) be the time to find the Ith neighbor of X in Gt(X∈[D4t],I∈[D2]).
First, to do zig-zag product, we should caculate (u,i) and (a,b). By properly labeling of vertices and edges, we can caculate them in poly(logNt) time (e.g. vertex (u,i) in Gt can be labelled as (u−1)D2+i, and edge (a,b) in Gt can be labelled as (a−1)D+b).
And then, walk on H only takes the constant time, while walk on Gt−12 takes 2time(Gt−1) time (coefficient 2 is due to squaring).
Therefore: time(Gt)=2time(Gt−1)+poly(logNt)=2tpoly(logNt)=NtΘ(1), where the last equality holds because Nt=D4t for a constant D. Thus,
this construction is only mildly explicit.
Construction of Fully Explicit Expanders: Let H be a (D8,D,7/8)-graph, and define:
G1=H2Gt+1=(Gt⊗Gt)2ⓏH
Obviously, we have: Nt=Nt−12D8. Thus: Nt≈c2t.
Therefore: time(Gt)=4time(Gt−1)+poly(logNt)=4tpoly(logNt)=poly(logNt). This construction is fully explicit.
A Better Construction of Fully Explicit Expanders: Let H be a (D8,D,7/8)-graph, and define:
G1=H2Gt+1=(G⌊t/2⌋⊗G⌈t/2⌉)2ⓏH
Now Nt=D8t, and time(Gt)=O(time(G⌊t/2⌋)+time(G⌈t/2⌉))+poly(logNt)=poly(t).
However, these constructions do not reach Alon–Boppana Bound, i.e. they are all not the optimal spectral expanders. Therefore here remains an open problem.
Open Problem: Give an explicit “combinatorial” construction of constant-degree expander graphs G with λ(G)≤2D−1/D+oN(1) (or even λ(G)=O(1/D)), where D is the degree.
Margulis-Gabber-Galil’s Construction
(TODO)
Undirected S-T Connectivity in Deterministic Logspace
Algorithm (Undirected S-T Connectivity in L):
Input: An undirected graph G with N edges and vertices s and t.
(1) Let H be a fixed (D4,D,3/4) graph for some constant D.
(2) Reduce (G,s,t) to (G0,s0,t0), where G0 is a D2-regular graph in which every connected component is nonbipartite and s0 and t0 are connected in G0 iff s and t are connected in G.
(3) For k=1,...,l=O(logN), define:
Let Gk=Gk−12ⓏH.
Let sk and tk be any two vertices in the “clouds” of Gk corresponding to sk−1 and tk−1, respectively.
(4) Try all paths of length O(logN) in Gl from sl and accept if any of them visit tl.
Some Problems
Problem (More Combinatorial Consequences of Spectral Expansion): Let G be a D-regular, connected, undirected, nonbipartite graph on N vertices with spectral expansion γ=1−λ. Prove that:
(1) The independence number α(G) is at most λN/(1+λ).
(2) The chromatic number χ(G) is at least (1+λ)/λ.
(3) The diameter of G is O(log1/λN).
Proof:
(1) Let S be the maximum independent set of G, then e(S,S)=0.
By Expander Mixing Lemma, we have: μ2(S)≤λμ(S)(1−μ(S))⇒μ(S)≤λ/(1+λ).
Thus, α(G)=μ(S)N is at most λN/(1+λ), as desired.
(2) Let c:V(G)→[χ(G)] be the coloring mapping of G.
Notice that for every 1≤i≤χ(G), c−1(i) is an independent set of G.
Thus: N=∑i=1χ(G)∣c−1(i)∣≤χ(G)⋅λN/(1+λ)⇒χ(G)≥(1+λ)/λ, as desired.
(3) Let M be the random walk matrix for G, and it has eigenvalues ∣λ1∣≥∣λ2∣≥…≥∣λn∣, where λ1=1 and ∣λi∣≤λ(i>1).
By using spectral decomposition, we can write M as:
M=i=1∑NλiviviT
where v1,…,vn denote normalized orthonormal column eigenvectors with eigenvalues λ1,…,λn.
For a positive integer m, consider arbitrary entry of matrix Mm:
Thus if we have m>log1/λ(N−1), then ∀1≤r,s≤N,(Mm)r,s>0, i.e. Mm has all entries positive.
Therefore the diameter of G is at most ⌊log1/λ(N−1)+1⌋, as desired. □
Problem (Error Reduction For Free): Show that if a problem has a BPP algorithm with constant error probability, then it has a BPP algorithm with error probability 1/n that uses exactly the same
number of random bits.
Proof: Let A(x;r) be the BPP algorithm of this problem, which r∈{0,1}m.
Then we can design an expander with 2m vertices, and do a random walk on it. However, this method costs the extra ranom bits.
Notice that if we only want to reduce error probability less than 1/n, the length of the random walk is only O(logn). Thus we can directly try all paths of length O(logn) in polynomial time by using some method such as depth first search. By this way, we only need to randomly choose the starting vertex, which only need m random bits. □
[4] Computational Complexity: A Modern Approach. Sanjeev Arora, Boaz Barak, 2009.
[5] Chung, F. R. K. (1989). “Diameters and eigenvalues”. Journal of the American Mathematical Society. 2 (2): 187–196. doi:10.1090/S0894-0347-1989-0965008-X. ISSN 0894-0347.