1 / 32

Hardness of Hyper-Graph C o l o r i n g

Hardness of Hyper-Graph C o l o r i n g. Irit Dinur NEC Joint work with Oded Regev and Cliff Smyth. Question. How many colors does it take to color a 3-colorable graph? 4? 5? 100? log n? The best known algorithm uses n 3/14 colors !!!. Definitions.

lucky
Download Presentation

Hardness of Hyper-Graph C o l o r i n g

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Hardness of Hyper-Graph Coloring Irit Dinur NEC Joint work with Oded Regev and Cliff Smyth

  2. Question • How many colors does it take to color a 3-colorable graph? • 4? • 5? • 100? • log n? • The best known algorithm uses n3/14 colors!!!

  3. Definitions • A hyper-graph, H=(V,E), E  {e V} is • 3-uniform: if each edge contains exactly 3 vertices, |e|=3. • 2Colorable, or has property B: if there exists a red-blue coloring of the vertices, with no monochromatic hyper-edge.

  4. Coloring - Background • Finding the chromatic number: • Approx within n1-eis hard: implies NP  ZPP, [FK] • Approximate Coloring… coloring graphs with tiny chromatic number • Graphs: • Best alg [BK] - O(n3/14) colors • NP-hard to color 3-col graph w/4 colors. [KLS, GK]. • Hypergraphs: • NP hard to decide 3-uniform HG is 2-col, [Lov ’73] • Apx alg …[KNS ‘98] - O(n1/5) colors • 4-uniform 2 vs. const is NP-hard [GHS ‘98] • Maximization variant: different for 3-unif and k>3.

  5. Theorem: Given a 3-uniform hypergraph, deciding whether c=2 or ccis NP-hard Corollary: For any constants k3, c2>c1>1, deciding whether c=c1or cc2in a k-uniform hypergraph is NP-hard Khot ’02: Finding large I.S. in a 3-uniform 3-col graph is NP-hard. Our Result

  6. The Kneser graph PCP, Layered Label-Cover The Hypergraph Construction What’s ahead

  7. [n] n/2-c The Kneser Graph • The Kneser Graph KGn,c: • Vertices: ( ), Edges: disjoint subsets • c(KG)2c+2 : easy • Kneser conj ’55: c(KG) = 2c+2 • c(KG)2c+2: First by Lovasz ’78, using Borsuk-Ulam theorem. Many other proofs, all using topological methods.

  8. Ground set: {1,2,3,4,5} {1,2} {3,5} {3,4} {4,5} {2,5} {1,3} {2,4} {1,4} {2,3} {1,5}

  9. The Kneser Graph Claim:c(KGn,c)  2c+2 Proof: Color #1 - all vertices v, 1v. Color #2 - all remaining vertices v, 2v. … Color #2c+1 - all remaining vertices v, 2c+1  v. Color #2c+2 - all remaining vertices.

  10. {1,2} {3,5} {3,4} {4,5} {2,5} {1,3} {2,4} {1,4} {2,3} {1,5}

  11. The Kneser Graph • What if we allow only c-1 colors ? • A color class is ‘bad’ if it contains a monochromatic edge, • How small can the ‘bad’ color class be? • In the previous example, it is ~2-c,a constant. • Is this the best we can do? • No, 2 colors can already cover 1 – o(1) of vertices Combinatorial Lemma: In any c-1=2c+1 coloring of KGn,c $ ‘bad’ color class whose relative size is >1/n2c

  12. Approximation and Hardness • Optimization: Given a hypergraph H, findc(H). • Approximation: find c’ s.t.c’ < gc • Hardness is proved via a gap-problem: Given H, decide between • [Yes:] if c(H)  m • [No:] if c(H) > mg • A g-approx algorithm can distinguish between the Yes and No cases, based on whether Alg(H) > mg. • m is also a parameter… fixing m=2 makes the problem perhaps easier.

  13. Approximation and Hardness • Thm: Given H, it is NP-hard to decide between • [Yes:] if c(H)  2 • [No:] if c(H)  100 • Hardness is proven via reduction from the PCP theorem: • PCP-Thm: It is NP-hard to dist. bet. • [Yes:] … (next slide) • [No:] … • Reduction: Translate [Yes:] … to c(H)  2 (completeness) and [No:] … to c(H)  100 (soundness)

  14. 1 2 y’ 2 4 1 5 z 3 3 z’ 1 3 y 5 7 y’z’ : yz :   PCP costume: Label-Coverbi-partite Graph G=(Y,Z, E) A:( Y RY , Z RZ ) is a labeling. A covers e=(y,z) E if yz(A(y)) = A(z)). Goal: cover as many edges as possible. A:( Y RY , Z RZ ) is a label cover if every eE is covered . Z Y {1,2,…,Ry} {1,2,…,Rz}

  15. Z Y Theorem[ALMSS,AS,Raz]: It is NP-hard to distinguish [Yes:]label-cover for the graph. [No:] Any labeling covers <  of the edges. yz :  Label-Coverbi-partiteGraph G=(Y,Z, E) {1,2,…,Ry} {1,2,…,Rz}

  16. Do we really need layers?? The hypergraph is built in the following way: • New vertices are created from Y • New vertices are created from Z • Hyperedges are based on the edges – always between Y and Z • Therefore, without layers, the hypergraph is always 2-colorable !

  17. xy : RX0 RX2 x’y’ : RX0 RX1 x’ Y’ x Y New Layered Label-Cover multi-partite Graph G=(X0,X1,..,XL,E) X0 X1 X2 XL {1,…,R0} {1,…,R1} {1,…,R2} {1,…,RL}

  18. S1 S2 S3 X0 X1 X2 XL-1 XL Layered Label-Cover Theorem: [D., Guruswami, Khot, Regev, ’01] L>0,>0 in an (L+1)-partite graph it is NP-hard to distinguish between the following: [Yes:]label-cover for the graph [No:]For every i,j any label-cover of Xi and Xj covers <  of the edges between them Moreover, for any k>0 layers i1<…<ik, and subsets SjÍ Xij of relative size 2/k, $Sj,Sj’, with 1/k2 of all the edges between Xij and Xij’

  19. p p

  20. Reducing Label-Cover to Hyper Graph Coloring Reduction: Translate multi-partite G into a hyper-graph H s.t. [Yes:]label-cover for G c(H) = 2 [No:]Every labelingcovers <  of the edges c(H)  100

  21. V =  (Xi ( )) Ri Ri/2 - 49 The Hypergraph Construction This is really the “Long-Code” X10000 X0 Xi

  22. 3-uniform Hyper-Edges X10000 X0 Xi

  23. 3-uniform Hyper-Edges

  24. yz :  3-uniform Hyper-Edges {1,2,…,Ry} {1,2,…,Rz} Y z RY Rz {v1 ,v2,u} ÎE iff: v1Ç v2 = f and xy( R\(v1È v2))Íu Note that v1 ,v2, are connected to a constant fraction of the u

  25. Proof Part I – [Yes] maps to [Yes] A label-cover of G translates to a legal 2-coloring of H.

  26. Red(x) = {v | ax v} Blue(x) = {v | ax v} 2 3 1 4 1 2 2 1 1 [Yes] maps to [Yes] Two disjoint vertices cannot both be red X10000 X0 Xi If v1, v2 blue, then axÎ R\(v1È v2) thus, ay=xy(ax)Îu, so u is red If v1, v2 blue, then axÎ R\(v1È v2) thus, ay=xy(ax)Îu, so u is red

  27. Part II – [No] maps to [No] If there’s no –cover for G, then (H)=100. Given a 99-coloring of H, We find in G, 2 layers Xi and Xj and a labeling that covers >  of edges between them

  28. Part II – [No] maps to [No] Combinatorial Lemma: In every 99 coloring of KGR,49 $ `special’ color class whose relative size is >1/R98 Given a 99-coloring of the HG, we find for each block, a ‘special’ color-class that is: Large > 1/R98 Contains two vertices v1Ç v2 = f

  29. S1 S2 S3 X0 X1 X2 XL-1 XL Concentrate on Si , Sj Sj Si Part II – [No] maps to [No] Given a 99-coloring, find “special” colors Assume blue is the prevalent special color. By layered label-cover theorem, $ Xi,Xj with many edges between Si and Sj

  30. (last slide of proof) Left blocks x: Each contain blue v1,v2disjoint – preventing all right hand u, {v1,v2, u}  E from being blue. These are u containing the “hole” Ri\(v1v2) and are a constant fraction of the Kneser block. Key point: the holes must be `aligned’ Define a labeling for Si , Sj as follows: y x Sj Si Define  x  Si : A(x) R R\(v1v2) Define  y  Sj : A(y) so as to maximize cover size We prove: A(y) is “popular” among its neighbors

  31. Summary • Kneser graph • PCP and Layered Label-Cover • Hypergraph Construction • Proof of Reduction

  32. Open Questions • Coloring: • We still can’t color a 2-col 3-uniform H.G. with less than ne colors, or prove a matching hardness • Worse situation for graphs: 3 vs. anything bigger than 5 • Maximization Versions of Coloring (e.g. max-cut)

More Related