The circular chromatic number of circular-perfect graphs is polytime Arnaud Pêcher joint work with C. Bachoc and A. Thiery Univ. Bordeaux (LaBRI) supported by the ANR/NSC project GraTel ANR-09-blan-0373-01, NSC98-2115-M-002-013-MY3 and NSC99-2923-M-110-001-MY3 2011 Worshop on Graph Theory Taipei, March 11th, 2011 A. Pêcher (Univ. Bordeaux) Bordeaux, 2011 1 / 28 Outline 1 A closed formula. 2 Separating the values. 3 Proving the closed formula. 4 This is the end A. Pêcher (Univ. Bordeaux) A closed formula Bordeaux, 2011 2 / 28 Some properties of Lovasz's Theta function ϑ Lovasz's Theta function is a real function such that, for every graph G : ϑ(G ) is computable in polynomial time with given (polynomial space encoding) accuracy ω(G ) ≤ ϑ(G ) ≤ χf (G ) ≤ χ(G ) if G is homomorphic to A. Pêcher (Univ. Bordeaux) H then ϑ(G ) ≤ ϑ(H ). A closed formula Bordeaux, 2011 3 / 28 Powers of chordless cycles q p = C q th power of the chordless cycle Cp with C9 A. Pêcher (Univ. Bordeaux) 2 C9 A closed formula p vertices 3 C9 Bordeaux, 2011 4 / 28 Complements = circular-cliques Circular-Clique Kp/q : vertices {0, 1, · · · , p − 1} and edges ij , s.t. q −1 = K . q ≤ |i − j | ≤ p − q . Hence Cp p/q K9/1 K9/2 K9/2 A. Pêcher (Univ. Bordeaux) = C9 K9/3 K9/3 A closed formula = C92 K9/4 K9/4 = C93 Bordeaux, 2011 5 / 28 From Perfect to Circular-perfect graphs Circular chromatic number Circular clique number (Vince, 1988) χc (G ) = inf k (Zhu, 2000) /d |G → Kk /d ωc (G ) = sup χ(G ) = dχc (G )e k /d |Kk /d → G ω(G ) = bωc (G )c ω(G ) ≤ ωc (G ) ≤ ωf (G ) = χf (G ) ≤ χc (G ) ≤ χ(G ) Perfect Graph (Berge, 1960) A graph G is perfect if ∀H ⊆ G , χ(H ) = ω(H ). Examples : If G bipartite graphs, chordal graphs, comparability graphs ... is perfect then ϑ(G ) = ϑ A. Pêcher (Univ. Bordeaux) p where χ(G ) = p . K A closed formula Bordeaux, 2011 6 / 28 From Perfect to Circular-perfect graphs Circular chromatic number Circular clique number (Vince, 1988) χc (G ) = inf k (Zhu, 2000) /d |G → Kk /d ωc (G ) = sup χ(G ) = dχc (G )e k /d |Kk /d → G ω(G ) = bωc (G )c ω(G ) ≤ ωc (G ) ≤ ωf (G ) = χf (G ) ≤ χc (G ) ≤ χ(G ) Circular-Perfect Graph (Zhu, 2000) A graph G is circular-perfect if ∀H ⊆ G , χc (H ) = ωc (H ). Examples : If G perfect graphs, circular-cliques, outerplanar graphs... is circular-perfect then ϑ(G ) = ϑ A. Pêcher (Univ. Bordeaux) p/q where χc (G ) = p /q . K A closed formula Bordeaux, 2011 6 / 28 Basics of polytime algorithm to compute χ of perfect graphs For a perfect graph G , χ(G ) is polytime as ϑ(G ) = χ(G ). This may be restated as this polytime algorithm to compute χ(G ) for a perfect graph G with n vertices : (1) compute ϑ(G ) with precision = 0.5 ; (2) for every 1 ≤ p ≤ n, if ϑ(G ) = ϑ(Kp ), return p . Correct as there is a unique A. Pêcher (Univ. Bordeaux) p satisfying (2) and has polyspace encoding. A closed formula Bordeaux, 2011 7 / 28 Core of algorithm to compute χc of circular-perfect graphs For circular √ perfect graphs, ϑ(G ) 6= χc (G ) in general : 5 < χc (C5 ) = 2.5. Hence we can not get χc (G ) directly from ϑ(C5 ) = ϑ(G ). But ... to compute χc (G ) for a circular-perfect graph polynomial time : G with n vertices in (1) compute ϑ(G ) for some precision > 0 ; (2) for every 1 ≤ p , q ≤ n ((p , q ) = 1, 2q ≤ p ), if |ϑ(G ) − ϑ(Kp/q )| < , return p /q . Correct provided there is a unique pair (p , q ) satisfying (2) and has polyspace encoding. Hence, roughly speaking, we need to prove that 0 ∃ > 0 such that for every pq 6= qp0 , we have ϑ(Kp/q ) − ϑ(Kp0 /q0 ) ≥ . A. Pêcher (Univ. Bordeaux) A closed formula Bordeaux, 2011 8 / 28 q = 2, q = 3 and p odd Already solved cases : Theorem - Lovász (1978) - q=2, ϑ p odd p/2 = ϑ(Cp ) = p K cos πp 1 + cos πp Proofs : Lovász : algebraic arguments Knuth (1994) : linear program with two variables Theorem - Brimkov et al (2000) - q=3, p odd p c + 1 2π p 2π − cos b c − cos b p 3 p 3 ϑ Kp/3 = p 1 − cos 2pπ b p3 c − 1 cos 2pπ b p3 c + 1 − 1 Proof : 1 2 linear program with 3 variables + geometrical arguments A. Pêcher (Univ. Bordeaux) A closed formula Bordeaux, 2011 9 / 28 A closed formula ∀0 ≤ k ≤ q − 1, ak = cos kp 2π q p qY −1 q −1 A0 (x ) = 2 (x − ai ) i =1 Theorem (Bachoc, P., Thiery 2010) ϑ(Kp/q ) = A. Pêcher (Univ. Bordeaux) p q 1+ 1 A0 (1) q−1 X i =1 A0 A closed formula cos 2i π ! q Bordeaux, 2011 10 / 28 Outline 1 A closed formula. 2 Separating the values. 3 Proving the closed formula. 4 This is the end A. Pêcher (Univ. Bordeaux) Injectivity Bordeaux, 2011 11 / 28 The main result 0 Let p , p 0 , q , q 0 ≤ n such that qp 6= qp0 . Let ∆ = ϑ(Kp0 /q0 ) − ϑ(Kp/q ). Theorem (Bachoc, P., Thiery (2011) ∆ ≥ c −n for some 5 c >0 Hence, computing χc of circular-perfect graphs is polytime. The proof is in two steps : (1) ∆ 6= 0 5 (2) if ∆ 6= 0 then ∆ ≥ c −n for some A. Pêcher (Univ. Bordeaux) c Injectivity >0 Bordeaux, 2011 12 / 28 ∆ = 0 : taking advantage of monoticity We have p q ≤ p q 0 0 if and only if K p/q → Kp0 /q0 (Bondy & Hell '96) if and only if ϑ(Kp/q ) ≤ ϑ(Kp0 /q0 ) 0 Assume ∆ = 0 : ϑ(Kp/q ) = ϑ(Kp0 /q0 ) = ϑ for some pq < qp0 . i h 0 Hence for every ba ∈ pq , qp0 , ϑ(Ka/b ) = ϑ. 0 Take ba ∈ qp , pq0 such that b is prime ; b is coprime with a and a + 1. h i A. Pêcher (Univ. Bordeaux) Injectivity Bordeaux, 2011 13 / 28 A avour of algebraic number theory (1/2) Notations and denitions : for every k , let ζk = exp2i π/k ; let Φ be the Euler function ; let Q (ζk ) denote the cylotomic eld : the smallest complex eld containing ζk ; for every x ∈ Q (ζk ), let polmin(x) ∈ Q[X ] be the minimal polynomial of x ; x is called an algebraic integer if polmin(x) ∈ Z[X ]. Some basic observations : Q (ζk ) is a vector space over Q whose dimension is Φ(k ) (hence at most k ) ; the set of algebraic integers is a ring. A. Pêcher (Univ. Bordeaux) Injectivity Bordeaux, 2011 14 / 28 A avour of algebraic number theory (2/2) Notations : if K and L are two cyclotomic elds s.t. L is an extension of L/K )=Aut(X L/K ) be the Galois group of L over K ; L TraceK (x )= σ(x ) be the trace of any element x Gal( L σ∈Gal(L/K ) x )= NormK ( Y σ∈Gal(L/K ) σ(x ) be the norm of any element x K , let of L; of L. let σi be the automorphism of Q (ζk ) s.t. σi (ζk ) = ζki (i coprime). Gal(Q (ζk )) = {σi , (i , k ) = 1} ; TraceLK is linear and for every element for every x ∈ Q (ζk ), polmin(x ) = x of L, TraceLK (x ) ∈ K ; (X − σ(x )). Y σ∈Gal(Q(ζk )/Q) A. Pêcher (Univ. Bordeaux) Injectivity Bordeaux, 2011 15 / 28 A useful property of ϑ(Kk /d ) Assume that d is prime : Gal(Q (ζkd ) /Q (ζk ))={σi , 1 ≤ i ≤ d − 1} ; for every i , σi (cos(2π/d )) = cos( 2i π/d ) as d −i 1 i cos(2i π/d ) = 2 (ζd + ζd ) = σi 12 (ζd + ζdd −1 ) ; hence (setting ( ) = A0 (X )/A0 (1)) : L0 X ϑ(Kk /d ) = = = ! k 1+ d 1+ d d i =1...d −1 L0 (σi (cos(2π/d ))) ! k k X X i =1...d −1 σi (L0 (cos(2π/d ))) kd ) 1 + TraceQ(ζ Q(ζk ) (L0 (cos(2π/d )) thus ϑ(Kk /d ) ∈ Q(ζk ) A. Pêcher (Univ. Bordeaux) Injectivity Bordeaux, 2011 16 / 28 ∆ = 0 : ϑ(Kp/q ) is rational Recall that we have chosen a and b such that ϑ = ϑ(Ka/b ), (a, b) = 1 and (a + 1, b) = 1. Then ϑ ∈ Q(ζa ) and likewise ϑ ∈ Q(ζa+1 ). Thus ϑ ∈ Q ! b is prime, 0 Now take ba ∈ qp , qp0 such that a/b is not integer, a is prime, (a, b) = 1 , (a, b + 1) = 1 and a/(b + 1) ≥ p /q . ∀i , 2 cos(2i π/a) and ai = 2 cos(2i π/b) are algebraic integers of Q(ζab ) ; hence ∀i , A0 (cos(2i π/b)) Qbis−1an algebraic integer of Q(ζab ) as b − 1 A0 (cos(2i π/b )) = 2 (2i π/b) − ai ) ; i =1 (cos Qb −1 P thus x = b l =1 (2 − 2al )ϑ = a 1 + bi =−11 A0 cos 2bi π is an algebraic integer of Q(ζab ). h A. Pêcher (Univ. Bordeaux) i Injectivity Bordeaux, 2011 17 / 28 ∆ = 0 : ϑ(Kp/q ) is integer, a contradiction Let ϑ = k /d . ab) (x ) = NormQ(ζ ab) Hence NormQ(ζ Q Q Q(ζ ab ) NormQ b bY −1 l =1 (2 − 2al )ϑ ! = Q(ζ ab ) NormQ = Q b−1 (2 − 2a )ϑ ∈ Z b l l =1 b dk Φ(ab) bϑ Q(ζ ab ) NormQ bY −1 l =1 (2 − 2al ) ! a2(b−1)φ(b) Since a is prime, if l is a prime factor of d distinct of a then l divides b. Likewise l divides b + 1. Thus l = 1 and so d is a power of a. Taking a big enough, we must have d = a0 = 1. Therefore ϑ = ϑ(Ka/b ) is an integer, and a result of Grötschel et. al implies that a/b is an integer, a nal contradiction : ∆ 6= 0. A. Pêcher (Univ. Bordeaux) Injectivity Bordeaux, 2011 18 / 28 ∆ ≥ c −n 5 0 0 Recall that ∆j= kϑ(K p0 /q0 ) − ϑ(Kj p/q )kwith p , p , q , q ≤ n. 0 Let ai = cos ipq 2pπ , ai0 = cos ipq0 2pπ0 , and let α = qq 1 2 3 0 qY −1 0 −1 qY i =1 i =1 (2 − 2ai ) (2 − 2ai0 )∆ α is an algebraic integer of Q(ζpp0 qq0 ) deg(polmin(α)) ≤ qq 0 pp 0 ≤ n4 for every σ ∈ Gal(Q(α)), |σ(α)| ≤ 4n n3 Y Since α 6= 0 and α is an algebraic integer, we have σ(α) ≥ 1. σ∈Aut(Q(α)) And the result follows from |α| ≥ 1 Y σ∈Aut(Q(α)),σ6=Id A. Pêcher (Univ. Bordeaux) |σ(α)| Injectivity ≥ 4 1 n 4n n 3 Bordeaux, 2011 19 / 28 Outline 1 A closed formula. 2 Separating the values. 3 Proving the closed formula. 4 This is the end A. Pêcher (Univ. Bordeaux) Proof of the formula Bordeaux, 2011 20 / 28 From a Semi-Denite Program to a Linear Program B(G ) = {B |B is SDP, tr(B ) = 1 and ∀ij ∈ E (G ) bij = 0} nX o ϑ(G ) = max bij |B ∈ B(G ) ϑp/q circulant graph : optimal symmetric circulant matrix B ij = b(i +j ) mod p and ∀i , bi = bp−i , P eigenvalues fi = j bj cos(2ij π/p ) P 1 bi = p j fj cos(2ij π/p ) (inverse discrete Fourier transform) B tr(B ) = 1 ↔ i =1 B SDP ↔ ∀i , fi ≥ 0 and fi = fp−i ∀ij ∈ E (G ) bij = 0 ↔ ∀1 ≤ i ≤ q − 1, bi = 0 X f ↔ ∀1 ≤ i ≤ q − 1, ϑ(G ) = max X A. Pêcher (Univ. Bordeaux) X j cos(2ij π/p ) = 0 f ij ↔ ϑ(G ) = max pf0 b Proof of the formula Bordeaux, 2011 21 / 28 Linear programs Main linear program ϑp/q = sup i ≥0 : pf0 f X i =0..p/2 1≤j ≤q−1 X i =0..p/2 i =1 f i cos (2ij π/p ) = 0 f Dual ϑ∗ (Kp/q ) = inf : pg0 1 ≤ i ≤ p /2 A. Pêcher (Univ. Bordeaux) X j =0..q−1 X j =0..q−1 Proof of the formula g j ≥1 g j cos (2ji π/p ) ≥ 0 Bordeaux, 2011 22 / 28 Linear system of optimal solution suggest that there is an optimal solution of main linear program whose non-zero coecients (f0 , f1∗ , . . . , fq∗−1 ) satisfy : Computer experimentations 1 1 cos j 1 k . . . 1 p q . . . cos j 1 2π (q −1)p q ... p k cos j k1 . . . 2π p A. Pêcher (Univ. Bordeaux) ... cos j p q 2(q −1)π q p . . . (q −1)p k 2(q −1)π p Proof of the formula f0 ∗ f1 = .. . ∗ q−1 f Bordeaux, 2011 1 0 .. . 0 23 / 28 Tchebychev polynomials Tchebychev polynomials Let T0 = 1, T1 Tq = x and ∀q ≥ 2, T q = 2xTq−1 − Tq−2 . ∀x , Tq (cos(x )) = cos(qx ) 1 1 .. . 1 1 1 q−1 (a1 ) .. .. .. . . . 1 T1 (aq−1 ) . . . Tq−1 (aq−1 ) | {z } ( ) T1 a1 ... T f0 ∗ f1 .. . ∗ q −1 f = 1 0 .. . 0 T A. Pêcher (Univ. Bordeaux) Proof of the formula Bordeaux, 2011 24 / 28 Lagrange interpolation Lagrange interpolation polynomials Lk w.r.t. points Qq−1 ai i =0,i 6=k (y − ai ) k (y ) = Qq−1 i =0,i 6=k (ak − ai ) L Let λk ,j be the coecients on the Tchebychev polynomials basis : k (y ) = λk ,0 T0 (y ) + . . . + λk ,q−1 Tq−1 (y ) L Hence (λi ,j )0≤i ,j ,≤q−1 is the inverse of the matrix T . Lemma (key point of the proof) For every i , λi ,0 ≥ 0 and L0 (cos(2i π/p )) ≥ 0. Thus (λ0,0 , λ1,0 , . . . , λq−1,0 ) is a feasible solution (restricted to non-zero entries) of the main linear program ; (λ0,0 , λ0,1 , . . . , λ0,q−1 ) is a feasible solution of the dual linear program. A. Pêcher (Univ. Bordeaux) Proof of the formula Bordeaux, 2011 25 / 28 The closed formula Hence, ϑ(Kp/q ) = p λ0,0 with λ0,0 given by the polynomial identity : ( ) = λ0,0 T0 (y ) + . . . + λ0,q−1 Tq−1 (y ) L0 y Since q−1 X i =0 j cos T 2i π q = 0 if j = 6 0 = Together with q if j = 0 ( ) = A0 (y )/A0 (1), we have nally, L0 y ϑ(Kp/q ) = A. Pêcher (Univ. Bordeaux) p q 1+ 1 A0 (1) q −1 X i =1 A0 Proof of the formula cos 2i π ! q Bordeaux, 2011 26 / 28 Outline 1 A closed formula. 2 Separating the values. 3 Proving the closed formula. 4 This is the end A. Pêcher (Univ. Bordeaux) This is the end Bordeaux, 2011 27 / 28 A. Pêcher (Univ. Bordeaux) This is the end Bordeaux, 2011 28 / 28 A. Pêcher (Univ. Bordeaux) This is the end Bordeaux, 2011 28 / 28