Approximating graph coloring of minor-closed graphs Joint Work with Erik Demaine, Mohammad Hajiaghayi, Bojan Mohar, Robin Thomas Partially joint Work with Neil Robertson and Paul Seymour Ken-ichi Kawarabayashi Tohoku University

Contents (Mostly, FOCS paper) Motivation (FOCS paper) 2-approx. of the chromatic number of minor- closed graphs (FOCS paper) Tree-width, Grid-minor, RS-structure. Overview of Algorithm (Robertson-Seymour) Approx. the list-chromatic number of minor- closed graphs. Toward Structural Theorem

Why is it accepted in FOCS? It is building on math deep theory. (although NOT AT ALL practical.) Minor-closed graphs are natural. (a generalization of planar graphs.) It tells how to use RS ’ main structural theorem. It is a bit easier to access (than RS ’ papers) Nice approx. for graph coloring of minor-closed graphs. Lucky.

Motivation Mathematical Motivation 1. Hadwiger ’ s Conjecture. (A far generalization of 4CT) 2. Graph Minor Theory (Robertson-Seymour) Algorithmic Motivation 1. Chromatic number is hard to compute. NP-complete even for deciding 3-colorability of Planar graphs. 2. Even hard to approx. NP-hard to approx. within constant factor. 3. NP-complete to decide the chromatic number of minor closed graphs. (Even for planar graphs) Can you approx. ?

Algorithmic Results Theorem (Demaine, Hajiaghayi, KK, FOCS2005) There exists a 2-approx. algorithm for the chromatic number in minor-closed graphs. (graphs with no K k -minor) The best known result was O(k √logk) approx. Proof uses the whole graph minor papers …. Robertson-Seymour theory consists of 23 papers. Most of them are published in JCTB.

Why is it 2-approx ? The main theorem says that if G is Kk-minor- free graphs, then it can be decomposed into two graphs G 1,G 2 such that both G 1 and G 2 have tree-width at most f(k). If tree-width is bounded, one can compute the chromatic number in the linear time. It remains to give an algorithm for the main theorem …

Proof depends on Robertson-Seymour theorem. It gives a structural theorem for minor-closed graphs. Once we have this structure, the rest of proof is not so hard (but not trivial.) The main challenge is how to obtain RS- structure. It depends on the whole graph minor papers.

delete Minors A graph G has a minor H if H can be formed by removing and contracting edges of G Otherwise, G is H-minor-free For example, planar graphs are both K 3,3 -minor-free and K 5 -minor-free contract H minor of G G *

Graph Minor Theory [Robertson & Seymour 1984 – 2004] Seminal series of ≥ 20 papers Powerful results on excluded minors: Every minor-closed graph property (preserved when taking minors) has a finite set of excluded minors [Wagner ’ s Conjecture] Every minor-closed graph property can be decided in polynomial time For fixed graph H, graphs minor-excluding H have a special structure: drawings on bounded-genus surfaces + “ extra features ”

Highlights of Graph Minor Theory Theorem(The disjoint paths problem) For fixed k, there is a polynomial time algorithm for deciding the disjoint paths problem. Minor testing can be done. Tree-width and grid-minors are discovered. Many mathematical and algorithmic applications.

The disjoint paths problems ………… S1S1 S2S2 S K-1 SKSK T1T1 T K-1 TKTK T2T2

Treewidth [GM2 — Robertson & Seymour 1986] Treewidth of a graph is the smallest possible width of a tree decomposition Tree decomposition spreads out each vertex as a connected subtree of a common tree, such that adjacent vertices have overlapping subtrees Width = maximum overlap − 1 Treewidth 1  tree; 2  series-parallel; … Graph Tree decomposition (width 3)

Tree-Decomposition

Tree-Decomposition of Graph A tree-decomposition of a graph G is (T,W), where T is a tree and W=(W t : t ⊂ V(T)) satisfies ∪ t ⊂ V(T) W t = V(G) If t ’ ⊂ T[t,t ” ], then Wt ∩ Wt ” ⊂ Wt ’ uv ⊂ E(G) for some t ⊂ V(t) s.t. u, v ⊂ Wt. The width is max(|Wt|-1 : t ⊂ V(T)). The tree-width of G is a minimum width.

Tree-width at most 1 G is a forest. Tree-width at most 2 G is series parallel. Tree-width at most 3 G has no minor isomorphic to K ５, Octahedron, 5-prism, V ８. Tree-width of the complete graph of order n is n-1. Tree-width is minor-monotone. The (k ×k)-grid minor has the tree-width k.

Tree-Width Discovered by Robertson-Seymour. NP-hard to determine tree-width. A linear time to decide whether tree-width is k or not for fixed k. Many NP-hard problems can be solved in polynomial time if a given graph has small tree-width. (even linear) It is useful for structural results. It is a key for the proof of RS. It is closely related to grid.

Grid Minors Why important ? r  r grid: r 2 vertices, 2 r (r − 1) edges Treewidth ~ r r  r grid is the canonical planar graph of treewidth Θ(r): every planar graph of treewidth w has an Ω(w)  Ω(w) grid minor [Robertson, Seymour, Thomas 1994] So any planar graph of large treewidth has an r  r grid minor certifying large treewidth What about nonplanar graphs? r r

r r Grid Minors Why important ? For any fixed graph H, every H-minor-free graph of treewidth ≥ w(r) has an r  r grid minor [GM5 — Robertson & Seymour 1986] Re-proved & strengthened [Robertson, Seymour, Thomas 1994; Reed 1997; Diestel, Jensen, Gorbunov, Thomassen 1999] Best bound of these: w(r) = 20 5 |V(H)| 3 r [Robertson, Seymour, Thomas 1994] New optimal bound: w(r) = Θ(r) [Demaine,Hajiaghayi KK 2005] Grids certify large treewidth in H-minor-free graph

Huge Grid is important Routing problem The disjoint paths problem and its generalization. Actually, Robertson-Seymour use this idea.

Structure of H-minor-free Graphs [GM16 — Robertson & Seymour 2003] Main result of RS Every H-minor-free graph can be written as O(1)-clique sums of graphs Each summand is a graph that can be O(1)-almost-embedded into a bounded-genus surface O(1) constants depend only on |V(H)|

Almost-Embeddable Graphs A graph is O(1)-almost-embeddable into a bounded-genus surface if it is A bounded-genus graph + a bounded number of vortices: Vortex = Replace a face in the bounded-genus graph by a graph of bounded pathwidth The interiors of the replaced faces are disjoint + a bounded number of apices: Apex = extra vertex with any incident edges

What do we need ? Crosscaps Handles Genus Vortex Apex (easy)

But There cannot be so many crossings that are far apart. The genus addition process stops quite soon. Otherwise, we would get a desired minor, a contradiction.

We know that Any long jump must be contained in the handle. This tells how to detect a handle. Any crossings and crosscaps are contained in small area. This tells how we can find a crosscap and a vortex. If there is no crosscap in the small area, then it is either vortex or planar graph. There cannot be many non-planar small areas that are far apart. This tells us that there are bounded number of vortices.

In summary 1. Stating with huge grid H. 2. As long as there is a long jump, we shall detect handles. 3. Otherwise the graph is embedded into a surface such that all the non-planar graphs are in small areas. 4. We shall look at each small area, and detect either vortex or crosscap. 5. There are only finitely many vortices and crosscaps. So the process stops.

Almost-Embeddable Graphs A graph is O(1)-almost-embeddable into a bounded-genus surface if it is A bounded-genus graph + a bounded number of vortices: Vortex = Replace a face in the bounded-genus graph by a graph of bounded pathwidth The interiors of the replaced faces are disjoint + a bounded number of apices: Apex = extra vertex with any incident edges

Approx. list coloring Theorem[Mohar and KK] There is an O(k)-approx. for graphs without Kk-minor, I.e., minor-closed graphs. Actually, it is “ almost ” O(√logk)-approx. It is approximating within O(√logk)c + O(k), where c is optimal. The best know appox. was O(k √logk) approx. Open: O(1) ? (Maybe NP-hard.)

Algorithm for List-coloring Theorem[KK & BM] There is an O( ) algorithm for the following: Input : A graph G, vertex set Z with |Z| <= 4k, precoloring of Z and each vertex in G has 16k- colors available in each list. Output : Determine either G has a K k -minor, or Precoloring of Z can be extended to the whole graph G, or G has a subgraph H such that H has no K k -minor and has a vertex set Z ’ with |Z ’ | <=4k such that some precoloring of Z ’ cannot be extended to H.

Algorithm for List Coloring Corollary: There is an O( ) algorithm for deciding the following: (1) G has a K k -minor (2) G has a 16k-list-coloring (3) G has a subgraph H such that H has no K k -minor and no 12k-list-coloring. It is easy to list-color by O(k √logk) colors