Chromatic Ramsey Number and Circular Chromatic Ramsey Number Xuding Zhu Department of Mathematics Zhejiang Normal University Among 6 people, There are 3 know each other, or 3 do not know each other. Know each other Do not know each other Among 6 people, There are 3 know each other, or 3 do not know each other. Among 6 people, There are 3 know each other, or 3 do not know each other. Among 6 people, There are 3 know each other, or 3 do not know each other. Among 6 people, There are 3 know each other, or 3 do not know each other. Colour the edges of K 6 by red or blue, there is either a red K 3 or a blue K 3 Theorem [Ramsey] For any graphs G and H, there exists a graph F such that if the edges of F are coloured by red and blue, then there is a red copy of G or a blue copy of H Sufficiently large or complicated General Ramsey Type Theorem: For `any’ systems Gi (1 i k) , there exists a system F such that if `elements’ of F are partitioned into k parts, then for some i, the ith part contains Gi as a subsystem. A sufficiently large scale (or complicated) system must contains an interesting sub-system. “Complete disorder is impossible” There are Ramsey type theorems in many branches of mathematics such as combinatorics, number theory, geometry, ergodic theory, topology, combinatorial geometry, set theory, and measure theory. Ramsey Theory has a wide range of applications. Theorem [Ramsey, 1927] t ,n ,k , m For a set M with | M | m, If the k-tuples M are t-colored, then M ' M , | M ' | n all the k-tuples of M’ having the same color. k 1 : pigeonholeprinciple k 2 : colouringtheedges of a completegraph Van der Waeden Theorem For any partition of integers into finitely many parts, one part contains arithematical progression of arbitrary large length. Szemerédi's theorem (1975) Erdos and Turan conjecture (1936) Regularity lemma Every set of integers A with positive density contains arithematical progression of arbitrary length. Timonthy Gowers [2001] gave a proof using both Fourier analysis and combinatorics. Harmonic analysis Ramsey number R(3,k) F (G, H ) means. For any 2-colouring of the edges of F with colours red and blue, there is a red copy of G or a blue copy of H. K 6 ( K3 , K3 ) K 5 ( K3 , K3 ) The Ramsey number of (G,H) is R(G,H) minn : Kn (G, H ) R( Kk , Kl ) R(k , l ) R( 3, k) minn : Kn ( K3 , Kk ) 1933, George Szekeres, Esther Klein, Paul Erdos starting with a geometric problem, Szekeres re-discovered Ramsey theorem, and proved 2 R( 3, k) k Szekere [1933] Erdos [1946] Erdos [1961] R( 3, k) k 2 k 3/2 R( 3, k) lnk k 2 R( 3, k) lnk Graver-Yackel [1968] ln ln k R( 3, k) O k 2 ln k Ajtai-Komlos-Szemeredi [1980] k2 R( 3, k) O ln k Kim [1995] k2 R( 3, k) ln k Many sophisticated probabilistic tools are developed George Szekere and Esther Klein married lived together for 70 year, died on the same day 2005.8.28, within one hour. R(k,l) minn : Kn ( Kk , Kl ) Bounds for R(k,l) k l 3 4 5 6 7 8 3 6 9 14 18 23 28 18 25 36 41 49 61 58 84 43 49 58 87 80 143 101 216 102 165 113 298 132 495 205 540 217 1031 4 5 6 7 8 282 1870 Bounds for R(k,l) k l 3 4 5 6 7 8 3 6 9 14 18 23 28 18 25 36 41 49 61 58 84 43 49 58 87 80 143 101 216 102 165 113 298 132 495 205 540 217 1031 4 5 6 7 8 282 1870 Bounds for R(k,l) k l 3 4 5 6 7 8 3 6 9 14 18 23 28 18 25 36 41 49 61 58 84 43 49 58 87 80 143 101 216 102 165 113 298 132 495 205 540 217 1031 4 5 6 7 8 282 1870 A sufficiently large scale (or complicated) system must contains an interesting sub-system. How to measure a system? What is large scale? What is complicated? How to measure a graph? T heRamsey number of (G, H) is R(G,H) min n : K n (G, H ) T heRamsey number of (G, H) is R(G,H) min | V ( F ) |: F (G, H ) T heSize Ramsey number of (G, H) is RE(G,H) min | E ( F ) |: F (G, H ) T hemax - degree - Ramsey number of (G, H) is R(G,H) min | ( F ) |: F (G, H ) T hechromaticRamsey number of (G, H) is R (G,H) min ( F ) : F (G, H ) T hecircular chromaticRamsey number of (G, H) is Rc (G,H) inf c ( F ) : F (G, H ) Chromatic number Circular chromatic number G=(V,E): a graph k 1 : an integer An k-colouring of G is f : V 0,1,..., k 1 such that x ~ y f ( x) f ( y ) 0 1 1 2 0 k 3 A 3-colouring of C5 The chromatic number of G is (G) mink : G has a k - colouring G=(V,E): a graph kr 1 : a real number an integer A (circular) k-colouring r-colouring of G is of G is An r ) k 1 f : V [[000,,1,k,..., 0 1 1.5 0.5 2 A 2.5-coloring such that x ~ y f ( x) 1f(|yf) ( x) f ( y1) || rf (x1) f ( y ) | k 1 The circular chromatic number of G is c (G) min inf { r: G has a circular r-colouring } f is k-colouring of G f is a circular k-colouring of G Therefore for any graph G, c (G) (G) 0=r p p’ 0 1 r 4 2 3 The distance between p, p’ in the circle is | p - p'|r min | p p'|, r | p p'| f is a circular r-colouring if x~y |f(x)-f(y)|_r ≥ 1 Basic relation between (G ) and c (G ) (G) 1 c (G) (G). Circular chromatic number of a graph is a refinement of its chromatic number. Graph coloring is a model for resource distribution Circular graph coloring is a model for resource distribution of periodic nature. T hechromaticRamsey number of (G, H) is R (G,H) min ( F ) : F (G, H ) R(G) R (G, G) Introduced by Burr-Erdos-Lovasz in 1976 R(G, H) R(G, H ) R(Kk ,Kl ) R(k , l ) 2 ( n 1 ) If F has chromatic number , then there is a 2 edge colouring of F in which each monochromatic subgraph has chromatic number n-1. F (G, G) n4 for any n-chromatic G. 2 ( n 1 ) If F has chromatic number , then there is a 2 edge colouring of F in which each monochromatic subgraph has chromatic number n-1. F (G, G) for any n-chromatic G. Could be much larger Observation : If (G) n, then R (G) (n 1)2 1 Conjecture[Burr - Erdos- Lovasz,1976]: For each n, thereis a graph G with (G) n and R (G) (n 1) 2 1 The conjecture is true for n=3,4 (Burr-Erdos-Lovasz, 1976) The conjecture is true for n=5 (Zhu, 1992) There are some upper bounds on minR (G) : (G) n Attempts by Tardif, West, etc. on non-diagonal cases of chromatic Ramsey numbers of graphs. No more other case of the conjecture were verified, until 2011 The conjecture is true (Zhu, 2011) Conjecture[Burr - Erdos- Lovasz,1976]: For each n, 2 2 thereis amin graph G with ( G ) n and R ( G ) ( n 1 ) 1 R (G) : (G) n (n 1) 1 Lemma[Burr - Erdos- Lovasz] R (G) n K n hom(G) For any 2 edge-colouring of Kn, there is a monochromatic graph which is a homomorphic image of G. Graph homomorphism = edge preserving map G H To prove Burr-Erdos-Lovasz conjecture for n, we need to construct an n-chromatic graph G, so that any 2 edge colouring of K ( n1)2 1 has a monochromatic subgraph which is a homomorphic image of G. The construction of G is easy: Take all 2 edge colourings of K( n1)2 1 c1 , c2 ,, cm For each 2 edge colouring ci of K ( n1)2 1 , one of the monochromatic Gi , has chromatic number at least n. subgraph, say Gi, Conjecture[Burr - Erdos- Lovasz,1976]: For each n, thereis a graph G with (G) n and R (G) (n 1) 2 1 To prove this conjecture for n, we need to construct an n-chromatic graph G, so that any 2 edge colouring of K ( n1)2 1 has a monochromatic subgraph which is a homomorphic image of G. The construction of G is easy: Take all 2 edge colourings of K( n1)2 1 c1 , c2 ,, cm For each 2 edge colouring ci of K ( n1)2 1 , one of the monochromatic Gi , has chromatic number at least n. subgraph, say Gi, G G1 G2 Gm GH H G GxH Projections are homomorphisms To prove this conjecture for n, we need to construct an n-chromatic graph G, so that any 2 edge colouring of K ( n1)2 1 has a monochromatic subgraph which is a homomorphic image of G. ? The construction of G is easy: Take all 2 edge colourings of K( n1)2 1 c1 , c2 ,, cm For each 2 edge colouring ci of K ( n1)2 1 , one of the monochromatic Gi , has chromatic number at least n. subgraph, say Gi, G G1 G2 Gm Each Gi is a homomorphi c image of G H G Conjecture[Hedetniemi, 1966]: (G H ) min (G), ( H ) To prove this conjecture for n, we need to construct an n-chromatic graph G, so that any 2 edge colouring of K ( n1)2 1 has a monochromatic subgraph which is a homomorphic image of G. ? If Hedetniemi’s conjecture is true, then Burr-Erdos-Lovasz conjecture is true. A k-colouring of G partition V(G) into k independent sets. (G) : familyof independent sets of G (G) min ( X ) X ( G ) : (G) 0,1 ( X ) 1 vX , X (G ) integer linear programming Af k-colouring (G) : fractional chromatic of G sets. of G partition V(G) number into k independent (G) : familyof independent sets of G f (G (G)) min ( X ) X ( G ) : (G) [0,1] ( X ) 1 vX , X (G ) linear programming f (G) (G) Conjecture[Hedetniemi, 1966]: (G H ) min (G), ( H ) Fractional Hedetniemi’s conjecture Conjecture[Z, 2002]: f (G H ) min f (G), f ( H ) To prove this conjecture for n, we need to construct an n-chromatic graph G, so that any 2 edge colouring of K ( n1)2 1 has a monochromatic subgraph which is a homomorphic image of G. If Hedetniemi’s conjecture is true, then Burr-Erdos-Lovasz conjecture is true. Observation: If fractional Hedetniemi’s conjecture is true, then Burr-Erdos-Lovasz conjecture is true. To prove this conjecture for n, we need to construct an n-chromatic graph G, so that any 2 edge colouring of K ( n1)2 1 has a monochromatic subgraph which is a homomorphic image of G. The construction of G is easy: Take all 2 edge colourings of K( n1)2 1 c1 , c2 ,, cm For each 2 edge colouring ci of K ( n1)2 1 , one of the monochromatic Gi , has chromatic subgraph, say Gi, at least n. > n-1 fractionalnumber chromatic number G G1 G2 Gm f (G) n 1 (G) f (G) n 1 Each Gi is a homomorphi c image of G Conjecture[Hedetniemi, 1966]: (G H ) min (G), ( H ) Fractional Hedetniemi’s conjecture Conjecture [Z, 2002]: Theorem [Z, 2011] f (G H ) min f (G), f ( H ) Theorem [Huajun Zhang, 2011] Conjecture : transitive, then If both G and[Z, H 2002] are vertex f (G H ) min f (G), f ( H ) Af k-colouring (G) : fractional clique chromatic number number Gof G sets. of G partition V(G) into kofindependent (G) : familyof independent sets of G f (G (G)) min ( X ) X ( G ) : (G) [0,1] ( X ) 1 vX , X (G ) linear programming f (G) max (v) vV ( G ) : V (G) [0,1] (v) 1 vX , X ( G ) dual problem The fractional chromatic number of G is obtained by solving a linear programming problem The fractional clique number of G is obtained by solving its dual problem f (G) f (G) Fractional Hedetniemi’s conjecture is true Conjecture [Z, 2002]: Theorem [Z, 2010] f (G H ) min f (G), f (H ) Pr oof sketch : f (G H ) min f (G), f (H ) f (G H ) min f (G), f (H ) Easy! Difficult! f (G H ) min f (G), f (H ) suffices to constructa fractionalclique of G H with totalweight min f (G), f ( H ) Easy Easy Difficult suffices to constructa fractionalclique of G H with totalweight min f (G), f ( H ) g : V(G) [ 0,1], a maximumfractionalclique of G h : V ( H ) [0,1], a maximumfractionalclique of H : V (G H ) [0,1], defined as g(x)h(y) (x,y) maxω f (G),ω f (H) is a fractionalclique of G H Difficult! with totalweight min f (G), f ( H ) independent set U of G H , g ( x)h( y) max ( x , y )U f (G), f ( H ) Easy! T hecircular chromaticRamsey number of (G, H) is Rc (G,H) inf c ( F ) : F (G, H ) Rc (G) Rc (G, G) What is the relation between Rc (G) and c(G) ? Basic relation between (G ) and c (G ) (G) 1 c (G) (G). c (G) (G). c (G ) is a refinement of (G ) (G ) is an approximation of c (G ) Circular colouring is a good model for periodical scheduling problems There are many periodical scheduling problems in computer sciences. The reciprocal of c (G ). is studied by computer scientists as efficiency of a certain scheduling method, in 1986. R (n) minR (G) : (G) n 2 Conjecture [Burr Erdos Lovasz, 1976] : R ( n ) ( n 1 ) 1 Theorem [Zhu, 2011] Rc ( z) inf Rc (G) : c (G) z Rc ( z) ? No conjecture yet! f (G) c (G) (G). (G) c (G) 1. (G) f (G) can be arbitrarily large. Using fractional version of Hedetniemi’s conjecture, Jao-Tardif-West-Zhu proved in 2014 If k 2 is an integer, thenRc (k ) k (k 1) Rc (2) 2(2 1) Rc (K3 ) 3(3 1) Rc (3) 3(3 1) ? c (G) can be real rational 2 Whatare thepossible valuesof Rx (G) ? c(G) min inf r : G has a circular r - colouring Rc (G) min inf c ( F ) : F (G, G) ? No ! [ Jao-Tardif-West-Zhu, 2014] Some other results by Jao-Tardif-West-Zhu, 2014 5 If 2 z , then R c ( z ) 4 2 Rc ( K3 ) 6 Rc ( K3 , K4 ) 9 14 R c (C3 , C5 ) 5 3 9 4 R c (C3 , C7 ) 2 謝謝