Derivatives and Real Roots of Graph Polynomials XueliangLiandYongtangShi CenterforCombinatorics andLPMC NankaiUniversity, Tianjin300071, China Email: [email protected], [email protected] 6 1 0 2 Abstract n a J Graphpolynomialsarepolynomialsassignedtographs. Interestingly,theyalsoarise 2 in many areas outside graph theory as well. Many properties of graph polynomials 1 have been widely studied. In this paper, we survey some results on the derivative and ] O real roots of graph polynomials, which have applications in chemistry, control theory C and computer science. Related to the derivatives of graph polynomials, polynomial . h reconstruction ofthematchingpolynomial isalsointroduced. t a m Keywords: graphpolynomial; derivatives;realroots;polynomial reconstruction [ AMSSubjectClassification (2010): 05C31,05C90,05C35,05C50 2 v 3 4 1 Introduction 8 1 0 Manykindsofgraphpolynomialshavebeen introducedand extensivelystudied,such as . 1 characteristic polynomial, chromatic polynomial, Tutte polynomial, matching polynomial, 0 6 independencepolynomial,cliquepolynomial,etc. 1 : Let G be a simplegraph with n vertices and m edges, whose vertex set and edge set are v i V(G)andE(G),respectively. ThecomplementGofGisthesimplegraphwhosevertexset X r is V(G) and whose edges are the pairs of nonadjacent vertices of G. For terminology and a notationnotdefined here, werefer to[3]. Denote by A(G) the adjacency matrix of G. The characteristic polynomial of G is de- fined as n φ(G,x) = det(λI −A(G)) = an−i. i i=0 X Therootsλ ,λ ,...,λ ofφ(G,x) = 0arecalledtheeigenvaluesofG. Formoreresultson 1 2 n φ(G,x), werefer to[10, 11]. 1 Denotebym(G,k)thek-thmatchingnumberofGfork ≥ 0. Weassumethatm(G,0) = 1. For k ≥ 1, m(G,k) is defined as the number of ways in which k pairwise independent edges can beselected inG. Thematchingpolynomialis defined as α(G,x) = (−1)km(G,k)xn−2k. k≥0 X There isalsoan auxiliarypolynomialα(G,x,y),which isdefined as α(G,x,y) = (−1)km(G,k)xn−2kyk. k≥0 X Note that α(G,x,y) = yn/2α(G,xy−1/2). In view of this fact, we may define an auxiliary polynomialofφ(G,x,y): φ(G,x,y) = yn/2φ(G,xy−1/2) = a xn−kyk/2. k k≥0 X Notethatφ(G,x,y)isapolynomialiny ifand onlyifG isbipartite. Denote by n(G,k) the k-th independence number of G for k ≥ 0. We assume that n(G,0) = 1. For k ≥ 1, n(G,k) is defined as the number of ways in which k pairwise independentverticescan beselectedin G. Theindependencepolynomialis defined as ω(G,x) = (−1)kn(G,k)xn−k, k≥0 X which is also called independent set polynomial in [35] and stable set polynomial in [58]. Formoreresultsontheindependencepolynomials,werefer thesurveys[42, 61]. Denote by c(G,k) the k-th clique number of G for k ≥ 0. We assume that c(G,0) = 1. For k ≥ 1, c(G,k) is defined as the number of ways in which k pairwise adjacent vertices can be selected in G. Note that c(G,1) = n and c(G,2) = m. The clique polynomial is defined as c(G,x) = (−1)kc(G,k)xn−k. k≥0 X Note that the cliquepolynomialof a graph G is exactly the independence polynomialof the complementG ofG,i.e., c(G,x) = α(G,x). Obviously,we alsohave c(G,x)+c(G,x) = α(G,x)+α(G,x). Thefollowingresultsare easilyobtained. Theorem 1. Let G andG betwo vertex-disjointgraphs. Then wehave 1 1 c(G ∪G ,x) = c(G ,x)+c(G ,x)−1, α(G ∪G ,x) = α(G ,x)·α(G ,x); 1 2 1 2 1 2 1 2 c(G +G ,x) = c(G ,x)·c(G ,x), α(G +G ,x) = α(G ,x)+α(G ,x)−1. 1 2 1 2 1 2 1 2 2 In [35], theauthorsobtainedthefollowingsimilarresult. Theorem 2. Let G and G be two vertex-disjoint graphs with n and n vertices, respec- 1 2 1 2 tively. Then c(G ×G ,x) = n ·c(G ,x)+n ·c(G ,x)−(n +n +n n x)+1. 1 2 2 1 1 2 1 2 1 2 Formorepropertieson c(G,x) and α(G,x),werefer to[35]. Manypropertiesofgraphpolynomialshavebeenwidelystudied. Inthispaper,wesurvey some results on the derivativeand real roots of graph polynomials, which have applications in chemistry,control theory and computerscience. Related to the derivativesof graph poly- nomials,polynomialreconstructionofthematchingpolynomialis alsointroduced. 2 Derivatives of graph polynomials The derivatives of the characteristic polynomial were examined by Clarke [9] and the followingresultwas showed. Theorem 3. Let G be a simple graphs and φ(G,x) be the characteristic polynomial of G. Then dφ(G,x)/dx = φ(G−v,x). v∈V(G) X GutmanandHosoya[28] gotasimilarresultforthematchingpolynomial. Theorem 4. Let G bea simplegraphsandα(G,x) bethematchingpolynomialofG. Then dα(G,x)/dx = α(G−v,x). v∈V(G) X One can get the first derivative of the independence polynomial and clique polynomial, which have similar expressions as the matching polynomial and characteristic polynomial. That is, dω(G,x)/dx = ω(G−v,x), dc(G,x)/dx = c(G−v,x). v∈V(G) v∈V(G) X X Although the four first derivativesobey fully analogous expressions, their proofs existingin the literatures, are quite dissimilar. Li and Gutman [44] provided a unified approach to all theseformulasby introducingageneral graph polynomial. Let f be a complex-valued function defined on the set of graphs G such that G ∼= G 1 2 impliesf(G ) = f(G ). LetGbeagraphonnverticesandS(G)bethesetofallsubgraphs 1 2 ofG. Define S (G) = {H : H ∈ S(G) and |V(H)| = k}, p(G,k) = f(H). k H∈XSk(G) 3 Then, thegeneral graphpolynomialofG isdefined as n P(G,x) = p(G,k)xn−k. k=0 X Actually,let (−1)|V(H)|/2 ifH is 1-regular; f(H) = ( 0 otherwise. Then theresultingpolynomialisthematchingpolynomial. Let (−1)|V(H)| ifH isno edges; f(H) = ( 0 otherwise. Then theresultingpolynomialistheindependencepolynomial. Let (−1)r(H) ·2c(H) ifallcomponentsofH are 1-or2-regular; f(H) = ( 0 otherwise, wherer(H)isthenumberofcomponentsinH andc(H)isthenumberofcyclesinH. Then theresultingpolynomialis thecharacteristicpolynomial. Let (−1)|V(H)| ifH is acompletegraph; f(H) = ( 0 otherwise. Then theresultingpolynomialisthecliquepolynomial. Thefollowingtheorem was obtainedbyLi andGutmanin [44]. Theorem 5. ForthegraphpolynomialP(G,x) ofG, wehave d (P(G,x)) = P(G−v,x). dx v∈V(G) X Furthermore, Gutman[24, 27]gotthefirst derivativeformulaforα(G,x,y): ∂α(G,x,y)/∂y = − α(G−u−v,x,y). uv∈E(G) X To find an expression for ∂φ(G,x,y)/∂y of a bipartite graph was posed by Gutman as a problemin [23]. Asolutionofthisproblemwas offered byLi and Zhang[48]. Theorem 6. Fora bipartitegraphG, ∂φ(G,x,y)/∂y = − φ(G−u−v,x,y)− n(C)yn(C)/2−1φ(G−C,x,y), (1) uv∈E(G) C⊆G X X where C isa cycle, possessingn(C) vertices. 4 Theabovetheorem was proved by using Sachs Theorem for thecoefficients of thechar- acteristicpolynomialand byverifyingtheequalityoftherespectivecoefficientsofthepoly- nomialsontheleft-and right-handsidesofEq. (1). In [29], theauthorsput forward another routetoEq. (1), fromwhich itbecomeevidentthatEq. (1)holdsforan arbitrary graph. Moreover,ifwedefine P(G,x,y) = p(G,k)xiyj, i+j=n X then wecan obtain ∂P(G,x,y) = ny−1P(G,x,y)−xy−1 P(G−v,x,y). ∂y v∈V(G) X Derivativesofothergraph polynomialshavealso been studied,such as thecubepolyno- mial[5], theTuttepolynomial[15], theWienerpolynomial[39], etc. 3 Polynomial reconstruction of the matching polynomial Thederivativeofagraphpolynomialisrelatedtheproblemofpolynomialreconstruction. This section aims to prove that graphs with pendant edges are polynomial reconstructible and,ontheotherhand,todisplaysomeevidencethatarbitrarygraphsarenot,whichisgiven in [47]. The famous (and still unsolved) reconstruction conjecture of Kelly [38] and Ulam [63] states that every graph G with at least three vertices can be reconstructed from (the isomor- phismclassesof)its vertex-deletedsubgraphs. Withrespect to agraph polynomialP(G),thisquestionmaybeadapted as follows: Can P(G) of a graph G = (V,E) be reconstructed from the graph polynomials of the vertex deleted-subgraphs, that is from the collection P(G ) for v ∈ V ? Here, this problem −v is considered for the matching polynomial of a graph. For results about the polynomial reconstruction of other graph polynomials, see the article by Bresˇar, Imrich, and Klavzˇar [4, Section 1] and the references therein. For additional results, see [59, Section 7] [60, Subsection 4.7.3]. The matching polynomial we considered here is the generating function of the number of its matchings with respect to their cardinality, denoted by M(G,x,y), which is different from the above α(G,x) and α(G,x,y). Let G = (V,E) be a graph. A matching in G is an edge subset A ⊆ E, such that no two edges in A have a common vertex. The matching polynomialM(G,x,y)isdefined as M(G,x,y) = xdef(G,A)y|A|, A⊆Eisamatching X 5 wheredef(G,A) = |V|−| e|isthenumberofverticesnotincludedinanyoftheedges e∈A of A. A matching A is a perfect matching, if its edges include all vertices, that means if S def(G,A) = 0. A near-perfect matching A is a matching that includes all vertices except one, that means def(G,A) = 1. For more information about matchings and the matching polynomial,see[17, 22, 51]. For a graph G = (V,E) with a vertex v ∈ V, G is the graph arising from the deletion −v of v, i.e., arising by the removal of v and all the edges incident with v. The multiset of (the isomorphism classes of) the vertex-deleted subgraphs G for v ∈ V is the deck of −v G. The polynomialdeck D (G) with respect to a graph polynomialP(G) is the multiset of P P(G ) for v ∈ V. A graph polynomial P(G) is polynomial reconstructible, if P(G) can −v bedeterminedfrom D (G). P By arguments analogous to those used in Kelly’s Lemma [38], the derivative of the matchingpolynomialofagraphG = (V,E)equalsthesumofthepolynomialsinthecorre- spondingpolynomialdeck. Proposition 1 (Lemma 1 in [18]). Let G = (V,E) be a graph. The matching polynomial M(G,x,y) satisfies δ M(G,x,y) = M(G ,x,y). (2) −v δx v∈V X Inotherwords,allcoefficientsofthematchingpolynomialexcepttheonecorresponding to the number of perfect matchings can be determined from the polynomial deck and thus also fromthedeck: 1 m (G) = m (G ) ∀i ≥ 1, (3) i,j i,j −v i v∈V X where m (G) isthecoefficient ofthemonomialxiyj inM(G,x,y). i,j Consequently, the (polynomial) reconstruction of the matching polynomial reduces to thedeterminationofthenumberofperfect matchings. Proposition 2. The matching polynomial M(G,x,y) of a graph G can be determined from itspolynomialdeckD (G)anditsnumberofperfectmatchings. Inparticular,thematching M polynomial M(G,x,y) of a graph with an odd number of vertices is polynomial recon- structible. Tutte[62,Statement6.9]showedthatthenumberofperfect matchingsofasimplegraph can be determined from its deck of vertex-deleted subgraphs and therefore gave an affirma- tiveansweronthereconstructionproblemforthematchingpolynomial. The matching polynomial of a simple graph can also be reconstructed from the deck of edge-extracted and edge-deleted subgraphs [18, Theorem 4 and 6] and from the polynomial 6 deck of the edge-extracted graphs [24, Corollary 2.3]. For a simple graph G on n vertices, the matching polynomial is reconstructible from the collection of induced subgraphs of G with⌊n⌋+1 vertices[20,Theorem 4.1]. 2 Thefollowingresult isfrom [47]for simplegraphs withpendantedges. Theorem 7. Let G = (V,E) be a simple graph with a vertex of degree 1. Then, G has a perfect matching if and only if each vertex-deleted subgraph G for v ∈ V has a near- −v perfect matching. AsprovedrecentlybyHuangandLih[36],thisstatementcanbegeneralizedtoarbitrary simplegraphs. Corollary 8. Let G = (V,E) bea forest. Then G has a perfect matchingif andonlyif each vertex-deleted subgraphG forv ∈ V hasa near-perfectmatching. −v Forestshaveeithernoneoroneperfectmatching,becauseeverypendantedgemustbein aperfectmatching(inordertocovertheverticesofdegree1)andthesameholdsrecursively for the subforest arising by deleting all the vertices of the pendant edges. Therefore, from Proposition 2 and Corollary 8 the polynomialreconstructibilityof the matching polynomial follows. Corollary9. ThematchingpolynomialM(G,x,y)ofaforestispolynomialreconstructible. On the other hand, arbitrary graphs with pendant edges can have more than one perfect matching. However,Corollary8canbeextendedtoobtainthenumberofperfectmatchings. For a graph G = (V,E), the number of perfect matchings and of near-perfect matchings of G is denotedbyp(G) and np(G), respectively. Theorem 10. Let G = (V,E) be a simple graph with a pendant edge e = {u,w} where w is avertex ofdegree1. Then we have p(G) = np(G ) ≤ np(G ) ∀v ∈ V andparticularly (4) −u −v p(G) = min{np(G ) | v ∈ V}. (5) −v By applying this theorem, the number of perfect matchings of a simple graph with pen- dant edges can be determined from its polynomialdeck and the following result is obtained as acorollary. Corollary 11. The matchingpolynomialM(G,x,y) of a simplegraphwith a pendant edge is polynomialreconstructible. 7 While it is true that the matching polynomialsof graphs with an odd number of vertices orwithapendant edgearepolynomialreconstructible,itdoes notholdforarbitrary graphs. There are graphs which have the same polynomial deck and yet their matching polyno- mialsaredifferent. Althoughtherearealready counterexampleswithaslittleassixvertices, it seems that nothing has been published before in connection with the question addressed here. Remark 1. The matching polynomial M(G,x,y) of an arbitrary graph is not polynomial reconstructible. The minimalcounterexamplefor simplegraphs(with respect to thenumber of vertices andedges)areconstructedin[47]. The question arises here: whether or not there are such counterexamples consisting of graphs with an arbitrary even number of vertices. In [47], we gave an affirmative answer to thisquestion. 4 Roots of beta-polynomials and independence polynomi- als Polynomials whose all zeros are real-valued numbers are said to be real. Several graph polynomials have been known to be real; among them the matching polynomial α(G,x) plays adistinguishedrole[21,34]. Polynomialswithonlyreal rootsariseinvariousapplicationsincontroltheoryandcom- puter science [65], but also admit interestingmathematicalproperties on theirown. Newton noted that the sequence of coefficients of such polynomials form a log-concave (and hence unimodal) sequence. These polynomials also have strong connections to totally positive matrices. Thefactthatforallgraphs,allzerosofthematchingpolynomialarereal-valuedwasfirst establishedby Heilmannand Lieb [34]. LetC beacircuitcontainedinagraphG. IfC isaHamiltoniancycle,thenα(G−C,x) ≡ 1. In certain considerations in theoretical chemistry [1, 40, 54, 55], graph polynomials β(G,C,x)are encountered,defined as β(G,C,x) = α(G,x)−2α(G−C,x) (6) and β(G,C,x) = α(G,x)+2α(G−C,x) (7) Formula (6) is used in the case of so-called Hu¨ckel-type circuits, whereas formula (7) for the so-called Mo¨bius-type circuits. For details, see [54]. These polynomials are also called circuit characteristicpolynomials[1]. 8 Alreadyinthefirstpaperdevotedtothistopic[1],Aiharamentionedthatthezerosofthe β-polynomialsare real-valued, but gaveno argument to supporthis claim. In themeantime, for a number of classes of graphs it was shown that β(G,C,x) is indeed a real polynomial [25,26,30,40, 49,55,41]. In additiontothis,bymeansofextensivecomputersearchesnot a single graph with non-real β-polynomial could be detected. The following conjecture has been put forward byGutmanand Mizoguchiin[25, 26, 30]. Conjecture 1. For any circuit C contained in any graph G, the β-polynomials β(G,C,x), Eqs. (6)and (7), arereal. Manyresultshavebeen obtained. In particular,β(G,C,x)has been showntobereal for unicyclic graphs [30], bicyclic graphs [55], graphs in which no edge belongs to more than one circuit [55], graphs without 3-matchings (i.e., m(G,3) = 0) [40], several (but not all) classes ofgraphswithout4-matchings(i.e., m(G,4) = 0)[41]. In [46], Li et al. showed that the conjecture is true for complete graphs. Actually, they provedastrongerresult forcompletegraphs. Theorem 12. ForanycircuitC inthecompletegraphK , thepolynomial n β(K ,C,t;x) = α(K ,x)+tα(K −C,x) n n n is realfor anyreal tsuchthat|t| ≤ n−1. Theproofofferedin[46]reliesonanearlierpublishedtheorembyTura´n. In[45],Liand Gutman presented an elementary self-contained proof for complete graphs. Finally, in [50], Li et al. showed that the conjecture is true for all graphs, and therefore completely solved thisconjecture. Theorem 13. For any circuit C contained in any graph G, all roots of the polynomial β(G,C,x)arereal. Chudnovskyand Seymour[8]provedthefollowingresult forindependencepolynomial. Theorem 14. If Gis clawfree, then allrootsofitsindependencepolynomialarereal. Theorem 14 extends a theorem of [34], answering a question posed by Hamidoune [33] and Stanley [58]. Since all line graphs are clawfree, this extends the result of [34]. Later, LevitandMandrescustudiedtherootsofindependencepolynomialsofalmostallverywell- covered graphs [43]. In [53], Mandrescu showed that starting from a graph G whose inde- pendence polynomial has only real roots, one can build an infinite family of graphs, whose independencepolynomialshaveonlyreal roots. Real roots of other graph polynomialshave also been extensivelystudied, such as edge- cover polynomial [2], the expected independence polynomial [7], domination polynomial 9 [6], sigma-polynomial [67], chromatic polynomial [14, 37, 66], Wiener polynomial [12], flow polynomial[37], Tuttepolynomial[16, 64], etc. Formore resultson theroots of graph polynomials,werefer to[13, 31, 32, 52, 56, 57, 65]. Acknowledgment Theauthorsaresupportedby NSFC andthe“973”program. References [1] J. Aihara, Resonance energies of benzenoid hydrocarbons, J. Amer. Chem. Soc. 99(1977),2048–2053. [2] S. Akbari and M.R. Oboudi, On the edge cover polynomial of a graph, European J. Combin.34(2013),297–321. [3] J.A. Bondy, U.S.R. Murty, Graph Therory, GTM 244, Springer-Verlag, New York, 2008. [4] B. Bresˇar, W. Imrich and S. Klavzˇar, Reconstructing subgraph-counting graph polyno- mialsofincreasing familiesofgraphs, DiscreteMath.297(1-3)(2005),159–166. [5] B. Bresˇar, S. Klavzˇar and R. Sˇkrekovski, The cube polynomial and its derivatives: the caseofmediangraphs,Electron.J.Combin. 10(2003),R3. [6] J.I. Brown and J. Tufts, On the roots of domination polynomials, Graphs Combin. 30(2014),527–547. [7] J.I. Brown, K. Dilcher and D.V. Manna, On the roots of expected independence poly- nomials,J.GraphTheory73(2013),322–326. [8] M. Chudnovsky and P. Seymour, The roots of the independence polynomial of a clawfreegraph, J.Combin.Theory, Ser. B97(2007),350–357. [9] F.H.Clarke,Agraphpolynomialanditsapplications,DiscreteMath.3(1972),305–313. [10] D. Cvetkovic´, M. Doob, I. Gutman and A. Torgasˇev, Recent Results in the Theory of GraphSpectra,North-Holland,Amsterdam,1988. [11] D. Cvetkovic´, M. Doob and H. Sachs, Spectra of Graphs—Theory and Applications, AcademicPress, New York, 1980. 10