Sparsest Cut S S G) = min |E(S, S)| |S| S µ V G = (V, E) c- balanced separator G) = min |E(S, S)| |S| S µ V c |S| ¸ c ¢ |V| Both NP-hard
Why these problems are important Arise in analysis of random walks, PRAM simulation, packet routing, clustering, VLSI layout etc. Underlie many divide-and-conquer graph algorithms (surveyed by Shmoys’95) Related to curvature of Riemannian manifolds and 2 nd eigenvalue of Laplacian (Cheeger’70) Graph-theoretic parameters of inherent interest (cf. Lipton-Tarjan planar separator theorem)
Previous approximation algorithms 1)Eigenvalue approaches ( Cheeger’70, Alon’85, Alon-Milman’85 ) 2c(G) ¸ L (G) ¸ c(G) 2 /2 c(G) = min S µ V E(S, S c )/ E(S) 2) O(log n) -approximation via multicommodity flows ( Leighton-Rao 1988 ) Approximate max-flow mincut theorems Region-growing argument 3) Embeddings of finite metric spaces into l 1 (Linial, London, Rabinovich’94) Geometric approach; more general result
Our results 1.O( ) -approximation to sparsest cut and conductance 2.O( )-pseudoapproximation to c-balanced separator (algorithm outputs a c’-balanced separator, c’ < c) 3.Existence of expander flows in every graph (approximate certificates of expansion) log n
LP Relaxations for c-balanced separator Motivation: Every cut (S, S c ) defines a (semi) metric X ij 2 {0,1} i< j X ij ¸ c(1-c)n 2 X ij + X j k ¸ X ik 0 · X ij · 1 Semidefinite There exist unit vectors v 1, v 2, …, v n 2 < n such that X ij = |v i - v j | 2 /4 Min (i, j) 2 E X ij
Semidefinite relaxation (contd) Min (i, j) 2 E |v i –v j | 2 /4 |v i | 2 = 1 |v i –v j | 2 + |v j –v k | 2 ¸ |v i –v k | 2 8 i, j, k i < j |v i –v j | 2 ¸ 4c(1-c)n 2 Unit l 2 2 space
l 2 2 space Unit vectors v 1, v 2,… v n 2 < d |v i –v j | 2 + |v j –v k | 2 ¸ |v i –v k | 2 8 i, j, k ViVi VkVk VjVj Angles are non obtuse Taking r steps of length s only takes you squared distance rs 2 (i.e. distance r s) ss ss
Example of l 2 2 space: hypercube {-1, 1} k |u – v| 2 = i |u i – v i | 2 = 2 i |u i – v i | = 2 |u – v| 1 In fact, every l 1 space is also l 2 2 Conjecture (Goemans, Linial): Every l 2 2 space is l 1 up to distortion O(1)
Our Main Theorem Two subsets S and T are -separated if for every v i 2 S, v j 2 T |v i –v j | 2 ¸ ¸ Thm: If i< j |v i –v j | 2 = (n 2 ) then there exist two sets S, T of size (n) that are -separated for = ( 1 ) <d<d log n
Main thm ) O( )-approximation log n v 1, v 2,…, v n 2 < d is optimum SDP soln; SDP opt = (I, j) 2 E |v i –v j | 2 S, T : –separated sets of size (n) Do BFS from S until you hit T. Take the level of the BFS tree with the fewest edges and output the cut (R, R c ) defined by this level (i, j) 2 E |v i –v j | 2 ¸ |E(R, R c )| £ ) |E(R, R c )| · SDP opt / · O( SDP opt ) log n
Next min: Proof-sketch of Main Thm
Projection onto a random line <d<d v u ?? 1 d 1 d e -t 2 /2 d log n Pr u [ projection exceeds 2 ] < 1/n 2
Algorithm to produce two –separated sets <d<d u SuSu TuTu 0.01 d Check if S u and T u have size (n) If any v i 2 S u and v j 2 T u satisfy |v i –v j | 2 · and repeat until no such v i, v j can be found delete them If S u, T u still have size (n), output them Main difficulty: Show that whp only o(n) points get deleted d “Stretched pair”: v i, v j such that |v i –v j | 2 · and | h v i –v j, u i | ¸ 0.01 Obs: Deleted pairs are stretched and they form a matching.
“Matching is of size o(n) whp” : trivial argument fails d “Stretched pair”: v i, v j such that |v i –v j | 2 · and | h v i –v j, u i | ¸ 0.01 O( 1 ) £ standard deviation ) Pr U [ v i, v j get stretched] = exp( - 1 ) = exp( - ) log n E[# of stretched pairs] = O( n 2 ) £ exp(- ) log n
Suppose with probability (1) there is a matching of (n) stretched pairs ViVi Ball (v i, ) u VjVj 0.01 d
The walk on stretched pairs u ViVi VjVj 0.01 d d r steps 0.01 d r |v final - v i | < r | | ¸ 0.01r d = O( r ) x standard dev. v final Contradiction!!
Measure concentration (P. Levy, Gromov etc.) <d<d A A : measurable set with (A) ¸ 1/4 A : points with distance · to A AA A ) ¸ 1 – exp(- 2 d) Reason: Isoperimetric inequality for spheres
Expander flows: Motivation G = (V, E) S S Idea: Embed a d-regular (weighted) graph such that 8 S w(S, S c ) = (d |S|) Cf. Jerrum-Sinclair, Leighton-Rao (embed a complete graph) “Expander” Graph w satisfies (*) iff L (w) = (1) [Cheeger] (*) Our Thm: If G has expansion , then a d-regular expander flow can be routed in it where d= log n (certifies expansion = (d) )
Example of expander flow n-cycle Take any 3-regular expander on n nodes Put a weight of 1/3n on each edge Embed this into the n-cycle Routing of edges does not exceed any capacity ) expansion = (1/n)
Formal statement : 9 0 >0 such that following LP is feasible for d = (G) log n f p ¸ 0 8 paths p in G 8i j p 2 P ij f p = d (degree) P ij = paths whose endpoints are i, j 8S µ V i 2 S j 2 S c p 2 P ij f p ¸ 0 d |S| (demand graph is an expander) 8e 2 E p 3 e f p · 1 (capacity)
New result (A., Hazan, Kale; 2004) O(n 2 ) time algorithm that given any graph G finds for some d >0 a d-regular expander flow a cut of expansion O( d ) log n Ingredients: Approximate eigenvalue computations; Approximate flow computations (Garg-Konemann; Fleischer) Random sampling (Benczur-Karger + some more) Idea: Define a zero-sum game whose optimum solution is an expander flow; solve approximately using Freund-Schapire approximate solver. ) d) · (G) · O(d ) log n
Open problems Improve approximation ratio to O(1); better rounding?? (our conjectures may be useful…) Extend result to other expansion-like problems (multicut, general sparsest cut; MIN-2CNF deletion) Resolve conjecture about embeddability of l 2 2 into l 1 Any applications of expander flows?
A concrete conjecture (prove or refute) G = (V, E); = (G) For every distribution on n/3 –balanced cuts {z S } (i.e., S z S =1) there exist (n) disjoint pairs ( i 1, j 1 ), ( i 2, j 2 ), ….. such that for each k, distance between i k, j k in G is O(1/ ) i k, j k are across (1) fraction of cuts in {z S } ( i.e., S: i 2 S, j 2 S c z S = (1) ) Conjecture ) existence of d-regular expander flows for d =
log n