For a finite graph GG with positive vertex weights ww, McMullen's clique polynomial Q(t)=∑K(−1)∣K∣tw(K)Q(t) = \sum_K (-1)^{|K|} t^{w(K)}, summed over the cliques KK of GG, packages the weighted growth of the positive semigroup of the right-angled Artin group W(G)W(G): the smallest positive zero of QQ equals 1/λ(G,w)1/\lambda(G,w), where λ(G,w)\lambda(G,w) is the exponential growth rate of the number of semigroup elements of weight at most TT. Minimizing λ(G,w)\lambda(G,w) over admissible weights (those with w(K)≤1w(K) \leq 1 on every clique) defines an invariant λ(G)\lambda(G) that governs which exponential growth rates the group can realize and enters the theory of entropy of graph actions and of Perron-Frobenius integer matrices. At the end of Section 6 of 'Entropy and the clique polynomial' (2015), McMullen asks two open questions under the hypotheses that the complement graph G′G' is connected and ∣V(G)∣>1|V(G)| > 1: (a) is λ(G)≥∣V(G)∣\lambda(G) \geq |V(G)|? and (b) if GG contains a clique of order nn, is λ(G)≥2n\lambda(G) \geq 2^n? Both bounds are sharp: λ(nA1)=n\lambda(nA_1) = n for nn isolated vertices, and λ(Kn+A1)=2n\lambda(K_n + A_1) = 2^n for a complete graph with one added isolated vertex, so neither constant can be improved. Both questions remain open, with no published resolution known to date; the best general lower bound under this hypothesis remains the paper's own Ramsey-type estimate λ(G)≥λn\lambda(G) \geq \lambda_n, where the Ramsey parameter nn satisfies ∣V(G)∣≥(2n−2n−1)|V(G)| \geq \binom{2n-2}{n-1} (so nn is of order log⁡2∣V(G)∣\log_2 |V(G)|) and λn\lambda_n is of order n/log⁡nn/\log n, i.e., a bound of order log⁡∣V(G)∣/log⁡log⁡∣V(G)∣\log|V(G)|/\log\log|V(G)| in the number of vertices. A companion open problem stated in the same paper - that lim inf⁡ηgg>0\liminf \eta_g^g > 0 for the minimal spectral radius ηg\eta_g of symplectic integer matrices, which the paper notes is equivalent to the Schinzel-Zassenhaus conjecture - has since been resolved affirmatively by Dimitrov's proof of that conjecture.

Background

Let GG be a finite, undirected, simple graph. The right-angled Artin group W(G)W(G) is presented by one generator per vertex, with two generators commuting exactly when the corresponding vertices are joined by an edge. Its positive semigroup W(G)+W(G)^+ consists of the elements expressible as products of the vertex-generators; since commuting generators may be reordered, for any assignment w:V(G)→R+w : V(G) \to \mathbb{R}^+ of positive vertex weights the weight w(g)=∑iw(ai)w(g) = \sum_i w(a_i) of such a product is independent of the word chosen to express gg. The weighted growth rate is λ(G,w)=lim⁡T→∞N(T)1/T\lambda(G,w) = \lim_{T \to \infty} N(T)^{1/T}, where N(T)N(T) counts semigroup elements gg with 0<w(g)≤T0 < w(g) \leq T. McMullen's clique polynomial Q(t)=∑K(−1)∣K∣tw(K)Q(t) = \sum_K (-1)^{|K|} t^{w(K)} - the sum runs over all cliques KK of GG (complete subgraphs, including the empty one), with w(K)=∑v∈Kw(v)w(K) = \sum_{v \in K} w(v) - encodes this growth exactly: by Theorem 1.3 of the source paper, the smallest positive zero of QQ equals 1/λ(G,w)1/\lambda(G,w), and h(w)=log⁡λ(G,w)h(w) = \log \lambda(G,w) is convex; when the complement graph G′G' (same vertex set, two distinct vertices adjacent exactly where GG has no edge) is connected and ∣V(G)∣>1|V(G)| > 1, hh is strictly convex, real-analytic, and tends to infinity at the boundary of the cone of positive weights. A weight is admissible if w(K)≤1w(K) \leq 1 for every clique KK; this normalization fixes the scale of the weights. The minimal growth rate of GG is λ(G)=inf⁡{λ(G,w):w admissible}\lambda(G) = \inf\{\lambda(G,w) : w \text{ admissible}\}. The same paper develops the parallel theory of metrized directed graphs (directed graphs whose edges carry positive real lengths), where the Perron polynomial - the determinant of II minus the length-weighted adjacency matrix - plays the role of QQ and the growth rate counts closed paths; with integral edge-lengths this connects λ\lambda to spectral radii (the largest modulus of an eigenvalue) of adjacency matrices of nonnegative integer matrices, and hence to questions about small eigenvalues of integer matrices and the entropies of surface dynamical systems. The complement-connectivity hypothesis structures the problem. The join G∗HG \ast H of two graphs is their disjoint union with all cross edges added; W(G∗H)≅W(G)×W(H)W(G \ast H) \cong W(G) \times W(H) (a direct product), and growth rates multiply: λ(G∗H)=λ(G)λ(H)\lambda(G \ast H) = \lambda(G)\lambda(H) (equation (6.2) of the paper). Every graph factors canonically as a join G=G1∗⋯∗GkG = G_1 \ast \cdots \ast G_k where the Gi′G_i' are the connected components of G′G'; graphs with G′G' connected are called prime, and only the prime case is essential. The hypothesis is also necessary for any bound in terms of vertex or clique counts: the complete graph KnK_n has W(Kn)≅ZnW(K_n) \cong \mathbb{Z}^n, polynomial growth, and λ(Kn)=1\lambda(K_n) = 1, while the complement of KnK_n is totally disconnected. Two cone subtleties matter for extremal arguments: if G=A1∗HG = A_1 \ast H (one vertex joined to all of HH) with λ(H)>1\lambda(H) > 1, then λ(G)=λ(H)\lambda(G) = \lambda(H) but the infimum is not attained - the optimal weight degenerates, vanishing on the cone vertex. The known sharp values calibrate the questions (throughout, AmA_m denotes the path graph with mm vertices, and nGnG denotes the disjoint union of nn copies of GG). For nA1nA_1 (nn isolated vertices, so WW is the free product of nn copies of the integers, realized as the free group on nn generators at the semigroup level): Q(t)=1−ntQ(t) = 1 - nt and λ(nA1)=n\lambda(nA_1) = n, with connected complement KnK_n. For Kn+A1K_n + A_1 (KnK_n plus one isolated vertex, complement the star K1,nK_{1,n}, hence connected, containing an nn-clique): with clique weights 1/n1/n on the KnK_n factor and weight 11 on the isolated vertex, Q(t)=(1−t1/n)n−tQ(t) = (1 - t^{1/n})^n - t, whose smallest positive zero is t=2−nt = 2^{-n}, so λ(Kn+A1)=2n\lambda(K_n + A_1) = 2^n exactly. Further values: λ(A2)=1\lambda(A_2) = 1 (a single edge, W≅Z2W \cong \mathbb{Z}^2), λ(A3)=2\lambda(A_3) = 2, λ(A2+nA1)=(1+n)2\lambda(A_2 + nA_1) = (1 + \sqrt{n})^2, and λ(2A2)=6+42\lambda(2A_2) = 6 + 4\sqrt{2} (two disjoint edges). What is known in general under the connected-complement hypothesis comes from the source paper: by Theorem 6.1, for fixed MM only finitely many graphs with connected complement have λ(G)≤M\lambda(G) \leq M, proved via a Ramsey-theoretic argument: by Ramsey theory, if ∣V(G)∣≥(2n−2n−1)|V(G)| \geq \binom{2n-2}{n-1}, then either GG or its complement G′G' contains a clique of order nn. In the branch where G′G' contains the clique (so GG contains nA1nA_1 as an induced subgraph), monotonicity gives λ(G)≥λ(nA1)=n\lambda(G) \geq \lambda(nA_1) = n; in the branch where GG itself contains the clique KK of order nn, the optimal weight satisfies w(a)≤1/nw(a) \leq 1/n for some a∈Ka \in K, and a two-vertex induced subgraph then gives λ(G)≥λn\lambda(G) \geq \lambda_n, where 1/λn1/\lambda_n is the smallest positive root of 1−t1/n−t1 - t^{1/n} - t and λn\lambda_n grows like n/log⁡nn/\log n. Since the Ramsey parameter nn satisfying ∣V(G)∣≥(2n−2n−1)|V(G)| \geq \binom{2n-2}{n-1} is of order log⁡2∣V(G)∣\log_2 |V(G)|, the resulting general lower bound is of order log⁡∣V(G)∣/log⁡log⁡∣V(G)∣\log|V(G)|/\log\log|V(G)|. By Theorem 1.5, only finitely many graphs with connected complement satisfy 1<λ(G)≤M1 < \lambda(G) \leq M; and by Theorem 1.6, the graphs with connected complement and 1<λ(G)<81 < \lambda(G) < 8 are exactly nA1nA_1 for 2≤n≤72 \leq n \leq 7, A2+A1A_2 + A_1, A2+2A1A_2 + 2A_1, A2+3A1A_2 + 3A_1, A3+A1A_3 + A_1, and K1,3+A1K_{1,3} + A_1. The two open questions ask whether the true general lower bounds are the sharp linear and exponential ones rather than this Ramsey-type bound of order log⁡∣V(G)∣/log⁡log⁡∣V(G)∣\log|V(G)|/\log\log|V(G)|. For context on the arithmetic side of the same paper: Theorem 1.1 determines the minimum spectral radius over reciprocal Perron-Frobenius matrices A∈M2g(Z)A \in M_{2g}(\mathbb{Z}) (nonnegative integer matrices, with strongly connected directed graph, whose eigenvalues are invariant under μ↦μ−1\mu \mapsto \mu^{-1}) as the largest root of t2g−tg(1+t+t−1)+1t^{2g} - t^g(1 + t + t^{-1}) + 1, whence ρ(A)g≥(3+5)/2\rho(A)^g \geq (3 + \sqrt{5})/2 - a result motivated by lower bounds for the entropies of pseudo-Anosov mapping classes of surfaces (the irreducible, exponentially stretching homeomorphisms in the Nielsen-Thurston classification of surface maps).

Problem Statement

Let GG be a finite, undirected, simple graph with vertex set V(G)V(G), and let G′G' denote its complement (same vertices; two distinct vertices adjacent in G′G' exactly when they are not adjacent in GG). The right-angled Artin group W(G)W(G) is the group presented by one generator ava_v for each vertex v∈V(G)v \in V(G), with the relations auav=avaua_u a_v = a_v a_u exactly when uu and vv are joined by an edge of GG; its positive semigroup W(G)+W(G)^+ consists of the elements expressible as products of the vertex-generators. Since commuting generators may be reordered, for any assignment w:V(G)→R+w : V(G) \to \mathbb{R}^+ of positive vertex weights, the weight w(g)=∑iw(ai)w(g) = \sum_i w(a_i) of such a product is independent of the word chosen to express gg. Let N(T)N(T) denote the number of semigroup elements g∈W(G)+g \in W(G)^+ with 0<w(g)≤T0 < w(g) \leq T, and let λ(G,w)=lim⁡T→∞N(T)1/T\lambda(G,w) = \lim_{T \to \infty} N(T)^{1/T} be the resulting exponential growth rate. By Theorem 1.3 of McMullen's paper 'Entropy and the clique polynomial', λ(G,w)\lambda(G,w) equals the reciprocal of the smallest positive zero of the clique polynomial Q(t)=∑K(−1)∣K∣tw(K)Q(t) = \sum_K (-1)^{|K|} t^{w(K)}, where the sum runs over all cliques KK of GG (complete subgraphs, including the empty one, and w(K)=∑v∈Kw(v)w(K) = \sum_{v \in K} w(v)). A weight ww is admissible if w(K)≤1w(K) \leq 1 for every clique KK; the minimal growth rate of GG is λ(G)=inf⁡{λ(G,w):w admissible}\lambda(G) = \inf\{\lambda(G,w) : w \text{ admissible}\}. McMullen's two questions, posed at the end of Section 6 of that paper, are: (a) If G′G' is connected and ∣V(G)∣>1|V(G)| > 1, is λ(G)≥∣V(G)∣\lambda(G) \geq |V(G)|? (b) If G′G' is connected, ∣V(G)∣>1|V(G)| > 1, and GG contains a clique of order nn, is λ(G)≥2n\lambda(G) \geq 2^n? Both bounds are sharp and therefore optimal up to equality: for G=nA1G = nA_1 (nn isolated vertices) the complement is the complete graph KnK_n, which is connected, and λ(G)=n=∣V(G)∣\lambda(G) = n = |V(G)|; for G=Kn+A1G = K_n + A_1 (KnK_n with one added isolated vertex) the complement is the star K1,nK_{1,n} (one center vertex joined to nn leaves), which is connected, GG contains a clique of order nn, and λ(G)=2n\lambda(G) = 2^n. The two questions share the same hypothesis and the same objective - sharp lower bounds on λ(G)\lambda(G) in terms of elementary graph parameters - and are stated together in the source.

References

  1. Curtis T. McMullen. Entropy and the clique polynomial. Journal of Topology 8 (2015), no. 1, 184-212. DOI: 10.1112/jtopol/jtu022. URL: https://doi.org/10.1112/jtopol/jtu022
  2. Vesselin Dimitrov. A proof of the Schinzel-Zassenhaus conjecture on polynomials. arXiv:1912.12545 (2019). URL: https://arxiv.org/abs/1912.12545
  3. Livio Liechti. On the arithmetic and the geometry of skew-reciprocal polynomials. Proceedings of the American Mathematical Society 147 (2019), no. 12, 5131-5139. DOI: 10.1090/proc/14668. URL: https://doi.org/10.1090/proc/14668 (preprint: arXiv:1812.04918, https://arxiv.org/abs/1812.04918)
  4. Erwan Lanneau, Livio Liechti, and Chi Cheuk Tsang. Minimal stretch factors of orientation-reversing fully-punctured pseudo-Anosov maps. Transactions of the American Mathematical Society 378 (2025), no. 4, 2943-2968. DOI: 10.1090/tran/9350. URL: https://doi.org/10.1090/tran/9350 (preprint: arXiv:2402.15369, https://arxiv.org/abs/2402.15369)
  5. Eriko Hironaka and Chi Cheuk Tsang. Standardly embedded train tracks and pseudo-Anosov maps with minimum expansion factor. Groups, Geometry, and Dynamics 19 (2024), no. 4, 1263-1317. DOI: 10.4171/ggd/805. URL: https://doi.org/10.4171/ggd/805 (preprint: arXiv:2210.13418, https://arxiv.org/abs/2210.13418)
  6. Mehdi Yazdi. Lower bound for dilatations. Journal of Topology 11 (2018), no. 3, 602-614. DOI: 10.1112/topo.12061. URL: https://doi.org/10.1112/topo.12061 (preprint: arXiv:1610.04278, https://arxiv.org/abs/1610.04278)
  7. Vsevolod Gubarev. PC-polynomial of graph. arXiv:1808.03932 (2018). URL: https://arxiv.org/abs/1808.03932
  8. A. Schinzel and H. Zassenhaus. A refinement of two theorems of Kronecker. Michigan Mathematical Journal 12 (1965), no. 1, 81-85. DOI: 10.1307/mmj/1028999247. URL: https://doi.org/10.1307/mmj/1028999247