Csr2011 june16 12_00_wagner

335
-1

Published on

0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total Views
335
On Slideshare
0
From Embeds
0
Number of Embeds
0
Actions
Shares
0
Downloads
2
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide

Csr2011 june16 12_00_wagner

  1. 1. Graphs of Bounded Treewidth can be Canonized in AC1 Fabian Wagner∗ CSR 2011∗ Institut Theoretische Informatik, Universit¨t Ulm, Germany a fabian.wagner@uni-ulm.de June 14-18, 2011
  2. 2. 1. Graph Isomorphism (GI) Graph: Gi = (V , Ei ), vertices V = {1, . . . , n}, edges Ei ⊆ V × V Isomorphism: G1 ∼ G2 iff ∃φ ∈ Sn such that ∀u, v ∈ V : = (u, v ) ∈ E1 ⇔ (φ(u), φ(v )) ∈ E2 Graph Isomorphism: GI = {(G1 , G2 ) | G1 ∼ G2 } = Complete invariant: f : G → Σ∗ s.t. f (G1 ) = f (G2 ) ⇔ G1 ∼ G2 = Canonical form: f : G → G s.t. f (G1 ) = f (G2 ) ∼ G1 ⇔ G1 ∼ G2 = = G1 G2 v0 v0 v1 v1 v2 v2 v3 v3 v4 v4
  3. 3. 1. Graph Isomorphism (GI) Graph: Gi = (V , Ei ), vertices V = {1, . . . , n}, edges Ei ⊆ V × V Isomorphism: G1 ∼ G2 iff ∃φ ∈ Sn such that ∀u, v ∈ V : = (u, v ) ∈ E1 ⇔ (φ(u), φ(v )) ∈ E2 Graph Isomorphism: GI = {(G1 , G2 ) | G1 ∼ G2 } = Complete invariant: f : G → Σ∗ s.t. f (G1 ) = f (G2 ) ⇔ G1 ∼ G2 = Canonical form: f : G → G s.t. f (G1 ) = f (G2 ) ∼ G1 ⇔ G1 ∼ G2 = = G1 G2 v0 φ v0 v1 v1 v2 v2 v3 v3 v4 v4
  4. 4. 1. Graph Isomorphism (GI) Graph: Gi = (V , Ei ), vertices V = {1, . . . , n}, edges Ei ⊆ V × V Isomorphism: G1 ∼ G2 iff ∃φ ∈ Sn such that ∀u, v ∈ V : = (u, v ) ∈ E1 ⇔ (φ(u), φ(v )) ∈ E2 Graph Isomorphism: GI = {(G1 , G2 ) | G1 ∼ G2 } = Complete invariant: f : G → Σ∗ s.t. f (G1 ) = f (G2 ) ⇔ G1 ∼ G2 = Canonical form: f : G → G s.t. f (G1 ) = f (G2 ) ∼ G1 ⇔ G1 ∼ G2 = = G1 G2 v0 φ v0 v1 v1 v2 v2 v3 v3 v4 v4
  5. 5. 1. Bounded Treewidth Graphs Tree decomposition of width k: D = ({Xi | i ∈ I }, T = (I , F )) ◮ every vertex in a bag Xi ◮ every edge in a bag Xi ◮ all bags which contain vertex v form a connected subtree ◮ treewidth k − 1 ⇔ bags have size ≤ k bag size ≤ k tree structure
  6. 6. 1. Bounded Treewidth graphs k-tree: w u v v k-clique connect new vertex to a k-clique v k-tree partial k-tree, treewidth k graph: subgraph
  7. 7. 1. Results Bounded Treewidth graph (BTW-graph) Known Results: ◮ Isomorphism: BTW-graphs in P [Bod90] ◮ Complete invariant: BTW-graphs in TC1 [GroVer06] ◮ Canonical labeling: BTW-graphs in TC2 [K¨bVer08] o New Results: ◮ Computation: tree decomposition in L [EJT10] ◮ Isomorphism: BTW-graphs in LogCFL [DTW10] ◮ Canonical labeling: BTW-graphs in AC1
  8. 8. 1. Results Restricted classes of BTW-graphs Known Results: ◮ Canonical labeling: BTDW-graphs in L [DTW10] ◮ Canonical labeling: k-trees in L [K¨bKu09] o ◮ Canonical labeling: partial 2-trees in L [ADK08] ◮ Canonical labeling: K3,3 - and K5 -free graphs in L [DNTW09] → partial 3-trees are K5 -free Open: ◮ Canonical labeling: partial 4-trees in L?
  9. 9. 1. Complexity NP P [K¨Ver08] o GI TC2 [GroVer06] [Tor´n04] a TC1BTW-canonical labeling ⊆ AC1 DETBTW-complete invariant SAC1 =LogCFL [DTW10] ⊆ BTW-GI ⊆ k = 3: [DNTW09] k = 2: [ADK08] NL k = 1: [Lin92] [K¨Ku09] o partial k-tree [JKMT03] L [JKMT03] k-tree GI k≤3 NC1 AC0
  10. 10. 2. BTW-Canonization: ingredients ◮ Theorem: {G | G has treewidth ≤ k} in L [EJT10] ◮ for graph G , treewidth k, approximate tree decomposition: width 4k + 3, depth O(log n) Theorem: approximate tree decomposition in L [EJT10] ◮ number of bags: ≤ n4k+3 bags of size 4k + 3 ◮ pre-computation in L: ◮ all possible bags, ◮ all split components of bags w.r.t. fixed root Xr ◮ all possible child bags w.r.t. fixed root Xr
  11. 11. 2. BTW-Canonization: minimal description Xr X = {u, v , w } σ = (u u v w w v)   0 1 0 G [X ]σ =  1 0 1  0 1 0 X Definition: minimal description: ′ C (G , X , σ) = C0 (G , X , σ) C0 (G , X , σ) [children] (4k+3)2 −|X |2 C (G , X , σ) = 010 101 010 2 uwv 2(4k+3−|X |)⌈log n⌉ [ ]
  12. 12. 2. BTW-Canonization: inductive construction C (G , X , σ) Xr sort ... X C1 C2 C3 min min min ... ... ... X1 X2 X3 C (G , X1 , σ1 ) ′ ′ C (G , X2 , σ2 ) C (G , X3 , σ3 ) C (G , X3 , σ3 ) X ′ X3 minimal description for i -th split component: C i = minXi ,σi C (G , Xi , σi )[...] X1 X2
  13. 13. 2. BTW-Canonization: depth parameter C (G , X , σ, d + 1) Xr d =3 sort d =2 ... X C1 C2 C3 d =1 min min min ... ... ... d =0 X1 X2 X3 ′ ′ C (G , X1 , σ1 , d) C (G , X2 , σ2 , d) C (G , X3 , σ3 , d) C (G , X3 , σ3 , d) minimal description for i -th split component: C i = minXi ,σi C (G , Xi , σi , d )
  14. 14. 2. BTW-Canonization: circuit depth C (G , X , σ, d ) X C (G , X , σ, d − 1) canon ∨ sort ... ... 1 i C C min min ... ... X1 X2 XiC (G , X1 , σ1 , d − i ) C (G , Xi , σi , d − i ) ◮ minimum: first order definable, in AC0 [Imm99] ◮ sort: i children of equal size ≤ n/i : with O(log i )-depth AC1 -circuits
  15. 15. 2. BTW-Canonization: example canon ∨ sort C (G , Xi , σi , d ) ... C1 C2 C3 ... min min min no−canon ... ... ... ...
  16. 16. 2. BTW-Canonization: example computes C (G , Xr , σr , c log n) tree decomposition for: tree of depth log n . . . 3 tree of depth 3 2 tree of depth 2 1 bag with leafs 0 leaf d = c log n
  17. 17. 2. BTW-Canonization: example . . . computes C (G , Xr , σr , c log n) tree decomposition for: d =2 tree of depth log n . . . . . . tree of depth 3 tree of depth 2 d =1 bag with leafs leaf . . .d = c log n d =0 pre-computation
  18. 18. 2. BTW-Canonization: example . . . computes C (G , Xr , σr , c log n) tree decomposition for: d =2 tree of depth log n . . . . . . tree of depth 3 tree of depth 2 d =1 bag with leafs leaf . . .d = c log n d =0 pre-computation
  19. 19. 2. BTW-Canonization: example . . . sort computes . . . C (G , Xr , σr , c log n) tree decomposition for: d =i tree of depth log n . . . tree of depth 3 . . . tree of depth 2 bag with leafs d =1 leaf . . .d = c log n d =0 pre-computation
  20. 20. 2. BTW-Canonization: algorithm input: graph G , treewidth k output: minimal description C 1: for all bags Xr = {v1 , . . . , v4k+3 } ⊆ V in parallel do 2: for all σ ∈ Sym(Xr ) in parallel do 3: C = min{C , C (G , Xr , σ, c log n)} recursively 4: relabel vertices in C in increasing order in L [DLN08] 5: return C → complete invariant return new labels → canonical form ◮ C = 010101010222 uvw 222 [0101010102222 xyw 222 . . . ] → u = 1, v = 2, w = 3, x = 4, y = 5, . . .
  21. 21. 2. BTW-Canonization: algorithm input: graph G , treewidth k output: canonical labeling 1: for all bags Xr = {v1 , . . . , v4k+3 } ⊆ V in parallel do 2: for all σ ∈ Sym(Xr ) in parallel do 3: C = min{C , C (G , Xr , σ, c log n)} recursively 4: relabel vertices in C in increasing order in L [DLN08] 5: return C → complete invariant return new labels → canonical form ◮ C = 010101010222 uvw 222 [0101010102222 xyw 222 . . . ] → u = 1, v = 2, w = 3, x = 4, y = 5, . . .
  22. 22. 6. Tools for Canonization Decomposition Reduction tree decomposition [DTW10,Wag11] BTW-Canon ≤ BTDW-Canon [DTW10] tree distance decomposition [DTW10] 2-connected component tree [DLNTW09,DNTW09] 3-connected component tree [DLNTW09,DNTW09] four-connected component tree [DNTW09] colored tree [KK09] tree representation of inverval graphs [KKLV10] → merge canons → analyse resources [Lin92] AC1 -circuit [Wag11] constant size components [DTW10,Wag11] NAuxPDA [DTW10] 3-connected planar components [DLN08] logspace machine [Lin92] Canonize Pieces Computational models
  23. 23. A. PlanarGI ∈ L - Canonize Biconnected Components Isomorphism order S < T if [Lindell92]: 1. |S| < |T | or 2. |S| = |T | but #s < #t or 3. |S| = |T | and #s = #t = k but (S1 , . . . , Sk ) < (T1 , . . . , Tk ) S s T t canon(S) = ( ((())) (()()) (. . . ) ) S2 S1 S3 T3 T1 T2
  24. 24. A. PlanarGI ∈ L - Canonize Biconnected Components Isomorphism order S < T if [Lindell92]: 1. |S| < |T | or 2. |S| = |T | but #s < #t or 3. |S| = |T | and #s = #t = k but (S1 , . . . , Sk ) < (T1 , . . . , Tk ) S s T t ≤ ≤ ≤ ≤ ≤ ≤ ≤ ≤ canon(S) = ( ((())) (()()) (. . . ) ) S1 S2 S3 T1 T2 T3
  25. 25. BTW-GI in LogCFL ◮ compute tree decomposition for G [Wanke94,GoLeSc02,EJT10] ◮ compute augmented tree Sr for G ◮ guess root r ′ in H, φ : Xr → Xr ′ ◮ if r ′ , φ not consistent then reject ◮ recursion: (s1 ≤T · · · ≤T sl ) ≤T (t1 ≤T · · · ≤T tl ) ◮ return graph G augmented tree Sr graph H ≤log m r s1 s2 ... sl a1,1 a2,1 ... a2,k2 al,kl
  26. 26. BTW-GI in LogCFL ◮ compute tree decomposition for G [Wanke94,GoLeSc02,EJT10] ◮ compute augmented tree Sr for G ◮ guess root r ′ in H, φ : Xr → Xr ′ ◮ if r ′ , φ not consistent then reject ◮ recursion: (s1 ≤T · · · ≤T sl ) ≤T (t1 ≤T · · · ≤T tl ) ◮ return graph G augmented tree Sr Tr′ graph H guess r φ r′ guess s1 s2 ... sl a1,1 a2,1 ... a2,k2 al,kl
  27. 27. BTW-GI in LogCFL ◮ compute tree decomposition for G [Wanke94,GoLeSc02,EJT10] ◮ compute augmented tree Sr for G ◮ guess root r ′ in H, φ : Xr → Xr ′ ◮ if r ′ , φ not consistent then reject ◮ recursion: (s1 ≤T · · · ≤T sl ) ≤T (t1 ≤T · · · ≤T tl ) ◮ return graph G augmented tree Sr Tr′ graph H r r′ guess ≤T ≤T guess s1 s2 ... sl t1 t2 tl ≤T ≤T ≤T ≤T a1,1 a2,1 ... a2,k2 al,kl ′ a2,1 remember bags along computation path depth first search + ≤T guess ≤T + ≤lex guess and extend φ
  1. A particular slide catching your eye?

    Clipping is a handy way to collect important slides you want to go back to later.

×