arXiv is now an independent nonprofit! Learn more
License: arXiv.org perpetual non-exclusive license
arXiv:2608.19780v1 [math.CO] 20 Aug 2026

Real-rooted flow polynomials have only integer roots

Meiqiao Zhang Thanks: Email: meiqiaozhang95@163.com and meiqiaozhang@xmu.edu.cn. Affiliation: School of Mathematical Sciences, Xiamen University, China    Fengming Dong Thanks: Email: fengming.dong@nie.edu.sg and donggraph@163.com. Affiliation: National Institute of Education, Nanyang Technological University, Singapore
Abstract

In this article, we show that for any bridgeless graph GG, if its flow polynomial F(G,x)F(G,x) has real zeros only, then GG is the dual of a chordal plane graph and each zero of F(G,x)F(G,x) is an integer in the set {1,2,3}\{1,2,3\}.

1 Preliminaries

All graphs considered in this paper are finite and undirected, and may have loops and parallel edges. For any graph GG, let V(G)V(G) and E(G)E(G) be the vertex set and edge set of GG, respectively.

In 1912, Birkhoff introduced the chromatic polynomial as a tool to attack the Four-Color Conjecture [1] . For any graph GG, the chromatic polynomial P(G,λ)P(G,\lambda) of GG counts the number of proper λ\lambda-colorings of GG for each positive integer λ\lambda. Thus, the Four-Color Conjecture is equivalent to the assertion that P(G,4)>0P(G,4)>0 for all planar graphs GG. It is then well known, from the deletion-contraction formula, that this counting function is in fact a polynomial in λ\lambda with integer coefficients, hence the name. Over the years, the chromatic polynomial has become one of the fundamental graph polynomials and an important object of study, due to the close connections between its algebraic properties and the structural properties of the underlying graph. See [5, 7, 12, 13] for reference.

In particular, the roots of P(G,λ)P(G,\lambda) are called the chromatic roots of GG. Clearly, 00 is a chromatic root of every graph. Also, it can be easily verified that all chromatic roots of a chordal graph are nonnegative integers, where a graph is chordal if it contains no induced cycle of length greater than three. Meanwhile, there exist non-chordal graphs whose chromatic roots are all integers [11, 8, 7, 6, 2], as well as non-planar graphs whose chromatic roots are all real but are not all integers [4]. Motivated by these observations on integral and real chromatic roots, it is natural to consider the following question, which remains widely open.

Problem 1 ([3]).

Is there a planar graph GG that has real chromatic roots only and contains non-integral chromatic roots?

Clearly, for Problem 1, it suffices to consider non-chordal planar graphs.

On the other hand, the flow polynomial was introduced by Tutte [14] in 1950 as a natural counterpart of the chromatic polynomial. For any graph GG, the flow polynomial F(G,λ)F(G,\lambda) of GG counts the number of nowhere-zero Γ\Gamma-flows on any orientation DD of GG, where Γ\Gamma is an additive Abelian group of order λ\lambda, for each positive integer λ\lambda. It is well known that F(G,λ)F(G,\lambda) is also a polynomial in λ\lambda with integer coefficients, as can be seen directly from the following deletion-contraction formula [15]:

F(G,λ)={1,if E(G)=,0,if G has a bridge,F(G1,λ)F(G2,λ),if G=G1G2,(λ1)F(Ge,λ),if e is a loop,F(G/e,λ)F(Ge,λ),if e is neither a loop nor a bridge,F(G,\lambda)=\left\{\begin{array}[]{@{}ll}1,&\text{if }E(G)=\varnothing,\\ 0,&\text{if }G\text{ has a bridge},\\ F(G_{1},\lambda)F(G_{2},\lambda),&\text{if }G=G_{1}\cup G_{2},\\ (\lambda-1)F(G-e,\lambda),&\text{if }e\text{ is a loop},\\ F(G/e,\lambda)-F(G-e,\lambda),&\text{if }e\text{ is neither a loop nor a bridge},\end{array}\right.

where G/eG/e and GeG-e are the graphs obtained from GG by contracting ee and deleting ee respectively, and G1G2G_{1}\cup G_{2} is the disjoint union of graphs G1G_{1} and G2G_{2}.

In analogy with chromatic roots, the roots of F(G,λ)F(G,\lambda) are called the flow roots of GG. For any connected plane graph GG, a classical duality due to Tutte [14] shows that

P(G,λ)=λF(G,λ),P(G,\lambda)=\lambda F(G^{*},\lambda),

where GG^{*} is the dual plane graph of GG. Thus, Problem 1 can be equivalently stated in terms of flow roots as follows.

Problem 2 ([4, 3]).

Is there a planar graph GG that has real flow roots only and contains non-integral flow roots?

Problem 2 also remains open. In fact, even the following weaker question is still unsolved.

Problem 3 ([4, 3]).

Is there a graph GG that has real flow roots only and contains non-integral flow roots?

Before introducing our main results, we first review some relevant developments concerning Problems 12, and 3. Note that we restrict our attention to bridgeless graphs when considering flow polynomials as F(G,λ)F(G,\lambda) is trivially zero if GG has a bridge.

In 2011, Kung and Royle have characterized the graphs whose flow roots are all integers [10].

Theorem 4 ([10]).

If GG is a bridgeless graph, then its flow roots are integral if and only if GG is the dual of a chordal and plane graph.

Inspired by Theorem 4, it is natural to ask whether every graph whose flow roots are all real must also be the dual of a chordal plane graph [3]. Moreover, Dong extended Theorem 4 by determining all possible flow roots of graphs with only integral flow roots as follows [4].

Theorem 5 ([4]).

Let GG be a graph with real flow roots only. The following statements are equivalent:

  1. (i)

    GG is the dual of some chordal and plane graph;

  2. (ii)

    each flow root of GG is in the set {1,2,3}\{1,2,3\};

  3. (iii)

    GG has no flow roots in the interval (1,2)(1,2).

In this paper, building on the aforementioned results, we give negative answers to Problems 12, and 3.

Theorem 6.

Let GG be a bridgeless graph. Then GG has only real flow roots if and only if GG has only integral flow roots if and only if GG is the dual of a chordal and plane graph. Moreover, every flow root of such a graph belongs to {1,2,3}\{1,2,3\}.

The following corollary is then direct.

Corollary 7.

Let GG be a loopless planar graph. Then GG has only real chromatic roots if and only if GG has only integral chromatic roots if and only if GG is chordal. Moreover, every chromatic root of such a graph belongs to {0,1,2,3}\{0,1,2,3\}.

We shall present some preliminary results in Section 2 and prove Theorem 6 in Section 3.

2 Preliminaries

In this section, we introduce some notations and properties of flow polynomials that will be used in the proof of Theorem 6.

Let GG be a graph. For any V0V(G)V_{0}\subseteq V(G), let GV0G-V_{0} denote the graph obtained from GG by deleting all the vertices in V0V_{0} and all the edges incident with some vertex in V0V_{0}. For any E0E(G)E_{0}\subseteq E(G), let GE0G-E_{0} denote the spanning subgraph of GG with edge set E(G)E0E(G)\setminus E_{0}. Moreover, we simply write GvG-v for G{v}G-\{v\} and GeG-e for G{e}G-\{e\}. Also, for any u,vV(G)u,v\in V(G), let G+uvG+uv be the graph obtained from GG by adding an edge between uu and vv. GG is said to be nonseparable if either |E(G)||V(G)|=1|E(G)|\leq|V(G)|=1 or GG is connected without loops or cut-vertices, and separable otherwise. A block of GG is a maximal nonseparable subgraph of GG. An edge-cut SS of GG is said to be proper if GSG-S has no isolated vertices. GG is kk-edge-connected if GG is connected with |V(G)|2|V(G)|\geq 2 and every edge-cut of GG has size at least kk.

The following lemma indicates that the flow polynomial F(G,λ)F(G,\lambda) of GG can be factorized or simplified whenever GG contains a loop, GG is disconnected, GG is separable, GG contains a 2-edge-cut, or GG contains a proper 3-edge-cut.

Lemma 8 ([9, 15]).

Let GG be a bridgeless graph.

  1. (i)

    If GG contains a loop ee, then F(G,λ)=(λ1)F(Ge,λ).F(G,\lambda)=(\lambda-1)F(G-e,\lambda).

  2. (ii)

    If GG is disconnected and G1,G2,,GkG_{1},G_{2},\dots,G_{k} are the components of GG, then

    F(G,λ)=i=1kF(Gi,λ).F(G,\lambda)=\prod_{i=1}^{k}F(G_{i},\lambda).
  3. (iii)

    If GG is connected and G1,G2,,GkG_{1},G_{2},\dots,G_{k} are the blocks of GG, then

    F(G,λ)=i=1kF(Gi,λ).F(G,\lambda)=\prod_{i=1}^{k}F(G_{i},\lambda).
  4. (iv)

    If ee is an edge in a 2-edge-cut of GG, then F(G,λ)=F(G/e,λ)F(G,\lambda)=F(G/e,\lambda).

  5. (v)

    Assume that GG is bridgeless, SS is a proper 3-edge-cut of GG whose removal separates GG into two subgraphs H1H_{1} and H2H_{2}. Let GiG_{i} be the graph obtained from GG by contracting E(H3i)E(H_{3-i}) for i=1,2i=1,2. Then

    F(G,λ)=F(G1,λ)F(G2,λ)(λ1)(λ2).F(G,\lambda)=\frac{F(G_{1},\lambda)F(G_{2},\lambda)}{(\lambda-1)(\lambda-2)}.

    An example is as shown in Figure 1.

H1H_{1}H2H_{2}GGH1H_{1}G1G_{1}
Figure 1: GG has a 3-edge-cut

Moreover, some standard facts of flow polynomials are as follows.

Proposition 9 ([16, 10]).

Let GG be a bridgeless graph, where |V(G)|=n|V(G)|=n and |E(G)|=m|E(G)|=m.

  1. (i)

    The polynomial F(G,λ)F(G,\lambda) has no real roots in (,1)(-\infty,1).

  2. (ii)

    If m>0m>0 and GG is nonseparable, then 11 is a flow root of GG with multiplicity one.

  3. (iii)

    If GG is 33-edge-connected, then

    F(G,λ)=i=0mn+1biλi,\displaystyle F(G,\lambda)=\sum_{i=0}^{m-n+1}b_{i}\lambda^{i}, (2.1)

    where br=1,br1=mb_{r}=1,b_{r-1}=-m, and br2,br3,,b0b_{r-2},b_{r-3},\dots,b_{0} are integers.

Then we establish our key lemma below.

Lemma 10.

Let GG be a nonseparable 33-edge-connected graph with |V(G)|=n|V(G)|=n and |E(G)|=m|E(G)|=m. If GG has only real flow roots and mn+1m\geq n+1, then m2n1m\leq 2n-1. Moreover, if the equality holds, then F(G,λ)=(λ1)(λ2)n1.F(G,\lambda)=(\lambda-1)(\lambda-2)^{n-1}.

Proof. Note that GG is nonempty as mn+1m\geq n+1 and GG is 33-edge-connected. Let r=mn+1r=m-n+1. Then r2r\geq 2. By Proposition 9 (iii), we assume that

F(G,λ)=i=0rbiλi,F(G,\lambda)=\sum_{i=0}^{r}b_{i}\lambda^{i},

where br=1b_{r}=1, br1=mb_{r-1}=-m, and br2,br3,,b0b_{r-2},b_{r-3},\dots,b_{0} are integers.

Since F(G,λ)F(G,\lambda) is a polynomial of degree r2r\geq 2, let ρ1,ρ2,,ρr\rho_{1},\rho_{2},\ldots,\rho_{r} be all the roots of F(G,x)F(G,x), counted with multiplicity. Then by Proposition 9 (i) and (ii), we can further assume that ρ1=1\rho_{1}=1, and ρi>1\rho_{i}>1 for 2ir2\leq i\leq r. Clearly, i=1rρi=br1=m\sum_{i=1}^{r}\rho_{i}=-b_{r-1}=m, implying that

n1=mr=i=1rρir=i=1r(ρi1)=i=2r(ρi1).\displaystyle n-1=m-r=\sum_{i=1}^{r}\rho_{i}-r=\sum_{i=1}^{r}(\rho_{i}-1)=\sum_{i=2}^{r}(\rho_{i}-1). (2.2)

Let Q(λ)=i=2r(λρi)Q(\lambda)=\prod\limits_{i=2}^{r}(\lambda-\rho_{i}). Then Q(1)0Q(1)\neq 0 and F(G,λ)=(λ1)Q(λ)F(G,\lambda)=(\lambda-1)Q(\lambda). Further,

{F(G,λ)=i=1ribiλi1.F(G,λ)=Q(λ)+(λ1)Q(λ)=i=2r(λρi)+(λ1)Q(λ).\left\{\begin{aligned} &F^{\prime}(G,\lambda)=\sum_{i=1}^{r}ib_{i}\lambda^{i-1}.\\ &F^{\prime}(G,\lambda)=Q(\lambda)+(\lambda-1)Q^{\prime}(\lambda)=\prod_{i=2}^{r}(\lambda-\rho_{i})+(\lambda-1)Q^{\prime}(\lambda).\\ \end{aligned}\right.

Then it is clear that F(G,1)=i=1ribiF^{\prime}(G,1)=\sum\limits_{i=1}^{r}ib_{i} while F(G,1)=Q(1)0F^{\prime}(G,1)=Q(1)\neq 0, which implies that F(G,1)F^{\prime}(G,1) is a nonzero integer. Moreover, since ρi1>0\rho_{i}-1>0 for 2ir2\leq i\leq r, we have

i=2r(ρi1)=|Q(1)|=|F(G,1)|=|i=1ribi|1.\prod_{i=2}^{r}(\rho_{i}-1)=|Q(1)|=|F^{\prime}(G,1)|=\left|\sum_{i=1}^{r}ib_{i}\right|\geq 1.

Then by (2.2) and the AM-GM inequality,

n1\displaystyle n-1 =i=2r(ρi1)(r1)(i=2r(ρi1))1/(r1)r1,\displaystyle=\sum_{i=2}^{r}(\rho_{i}-1)\geq(r-1)\left(\prod_{i=2}^{r}(\rho_{i}-1)\right)^{1/(r-1)}\geq r-1, (2.3)

where the equality in the first inequality holds if and only if ρi\rho_{i} are all equal for 2ir2\leq i\leq r. Hence rnr\leq n, which is equivalent to m2n1m\leq 2n-1.

Moreover, for the case m=2n1m=2n-1, we have r=nr=n and all the equalities in (2.3) hold. Therefore, ρi\rho_{i} are all equal for 2ir2\leq i\leq r. Then by (2.2),

n1=i=2r(ρi1)=i=2n(ρi1),n-1=\sum_{i=2}^{r}(\rho_{i}-1)=\sum_{i=2}^{n}(\rho_{i}-1),

which implies that ρi=2\rho_{i}=2 for all 2ir2\leq i\leq r. Hence F(G,λ)=(λ1)(λ2)n1F(G,\lambda)=(\lambda-1)(\lambda-2)^{n-1}. ∎

3 The proof of Theorem 6

In this section, we shall prove Theorem 6 by the following lemma established in [4].

Lemma 11 ([4]).

Suppose that GG is a 33-edge-connected graph with |V(G)|=n|V(G)|=n such that GG has only real roots. Let k=|{vV(G):dG(v)>3}|k=\left|\{v\in V(G):d_{G}(v)>3\}\right|. If n3n\geq 3 and GG has no proper 33-edge-cut, then n2k+1n\geq 2k+1, and

m2n+2k3+4(k1)2n2k.m\geq 2n+2k-3+\frac{4(k-1)^{2}}{n-2k}. (3.1)

Proof of Theorem 6. By Theorems 4 and 5, it suffices to show that if GG has only real flow roots, then GG has only integral flow roots. Suppose that GG is a counterexample to the above statement with the minimum number of edges. Then GG has only real flow roots and contains a non-integer flow root xx. Let |V(G)|=n|V(G)|=n, |E(G)|=m|E(G)|=m, and k=|{vV(G):dG(v)>3}|k=\left|\{v\in V(G):d_{G}(v)>3\}\right|.

Claim 1.

GG is loopless.

Proof. Assume that GG contains a loop ee. Then by Lemma 8 (i), GeG-e has only real flow roots and contains the non-integer flow root xx with |E(Ge)|<|E(G)||E(G-e)|<|E(G)|, a contradiction to the assumption of GG. Hence GG is loopless. \natural

Claim 2.

GG is connected.

Proof. Assume that GG is disconnected with components G1,G2,,GcG_{1},G_{2},\dots,G_{c}, where c2c\geq 2 and |E(Gi)|>0|E(G_{i})|>0 for all 1ic1\leq i\leq c. Then by Lemma 8 (ii), each GiG_{i} has only real flow roots and some GiG_{i} has the non-integer flow root xx, where each GiG_{i} contains less edges than GG, a contradiction to the assumption of GG. Hence GG is connected. \natural

Claim 3.

GG is nonseparable with n2n\geq 2.

Proof. If n=1n=1, then m=0m=0 due to Claim 1. As a result, F(G,λ)=1F(G,\lambda)=1, a contradiction to the assumption of GG. Thus n2n\geq 2. By Claims 1 and 2, it remains to show that GG contains no cut-vertices.

Assume that GG contains cut-vertices, which implies that GG has blocks G1,G2,,GbG_{1},G_{2},\dots,G_{b}, where b2b\geq 2 and |E(Gi)|>0|E(G_{i})|>0 for all 1ib1\leq i\leq b. Then by Lemma 8 (iii), each GiG_{i} has only real flow roots and some GiG_{i} has the non-integer flow root xx, where each GiG_{i} contains less edges than GG, a contradiction. Hence Claim 3 holds. \natural

Claim 4.

GG is 33-edge-connected.

Proof. By Claims 2 and 3, it remains to show that GG contains no edge-cut of size two.

Assume that GG contains a 2-edge-cut SS, where eSe\in S. Then by Lemma 8 (iv), G/eG/e has only real flow roots and contains the non-integer flow root xx with |E(Ge)|<|E(G)||E(G-e)|<|E(G)|, a contradiction. Hence Claim 4 holds. \natural

Claim 5.

GG has no proper 33-edge-cut.

Proof. Assume that GG has a proper 33-edge-cut SS, whose removal separates GG into two subgraphs H1H_{1} and H2H_{2}. Then |E(H1)|,|E(H2)|>0|E(H_{1})|,|E(H_{2})|>0. Let GiG_{i} be the graph obtained from GG by contracting E(H3i)E(H_{3-i}) for i=1,2i=1,2. Then by Lemma 8 (v), G1G_{1}, G2G_{2} have only real flow roots and at least one of G1G_{1}, G2G_{2} has the non-integer flow root xx, where G1,G2G_{1},G_{2} both contain less edges than GG, a contradiction. Hence Claim 5 holds. \natural

Claim 6.

n4n\geq 4.

Proof. Suppose n=2n=2. Since GG is 3-edge-connected by Claim 4, there are at least three edges in GG, implying that m3=n+1m\geq 3=n+1. Then Lemma 10 indicates that m2n1=3m\leq 2n-1=3. Hence m=3=2n1m=3=2n-1. Then again by Lemma 10, F(G,λ)=(λ1)(λ2)F(G,\lambda)=(\lambda-1)(\lambda-2), a contradiction to the assumption of GG.

Suppose n=3n=3. Since GG is 3-edge-connected, the minimum degree of GG is at least three, implying that 2m3n=92m\geq 3n=9. Thus m5>n+1m\geq 5>n+1. Consequently, Lemma 10 indicates that m2n1=5m\leq 2n-1=5. Hence m=5=2n1m=5=2n-1. Then by Lemma 10, F(G,λ)=(λ1)(λ2)2F(G,\lambda)=(\lambda-1)(\lambda-2)^{2}, also a contradiction to the assumption of GG.

Then by Claim 3, the claim holds. \natural

Claim 7.

n2k+1n\geq 2k+1.

Proof. By Claims 45 and 6, we have n4n\geq 4 and GG is a 33-edge connected graph which has no proper 33-edge-cut. Then, Lemma 11 implies that n2k+1n\geq 2k+1. \natural

Claim 8.

k=0k=0.

Proof. Suppose that k2k\geq 2. Since n2k+1n\geq 2k+1 by Claim 7, we have 4(k1)2n2k0\frac{4(k-1)^{2}}{n-2k}\geq 0. Then (3.1) indicates

m2n+2k3+4(k1)2n2k2n+2k3>2n1,m\geq 2n+2k-3+\frac{4(k-1)^{2}}{n-2k}\geq 2n+2k-3>2n-1,

a contradiction to Lemma 10.

Suppose that k=1k=1. (3.1) indicates

m2n+23=2n1.m\geq 2n+2-3=2n-1.

Then by Lemma 10, we have m=2n1m=2n-1 and F(G,x)=(x1)(x2)n1F(G,x)=(x-1)(x-2)^{n-1}, a contradiction to the assumption of GG.

Hence k=0k=0. \natural

Claim 9.

n=4n=4.

Proof. By Claim 4, every vertex in GG has degree at least three, while by Claim 8, every vertex in GG has degree at most three. Hence GG is cubic and m=3n/2m=3n/2. Then (3.1) indicates

3n22n3+4n,\frac{3n}{2}\geq 2n-3+\frac{4}{n},

or equivalently

(n2)(n4)0.(n-2)(n-4)\leq 0.

Thus n=4n=4 follows from Claim 6. \natural

Claim 10.

GG has only integral roots.

Proof. Now GG is a cubic, 33-edge-connected and loopless multigraph on four vertices. We shall show that GG is K4K_{4}. Equivalently, we shall prove that GG contains no parallel edges.

Suppose that GG has parallel edges between vertices u,vu,v in GG. Then there are at most two edges between {u,v}\{u,v\} and V(G){u,v}V(G)\setminus\{u,v\} in GG, which indicates an edge-cut of size at most two in GG, a contradiction to the fact that GG is 33-edge-connected. Consequently, GG is K4K_{4}, where

F(K4,λ)=(λ1)(λ2)(λ3).F(K_{4},\lambda)=(\lambda-1)(\lambda-2)(\lambda-3).

Hence the claim holds. \natural

Claim 10 contradicts the assumption of GG. The theorem is proven. ∎

References

  • [1] G.D. Birkhoff, A determinant formula for the number of ways of coloring a map, Annal. Math. 14 (1912) 42–46.
  • [2] I.G. Dmitriev, Weakly cyclic graphs with integral chromatic number (Russian), Metody Diskret. Analiz., 34 (1980) 3–7.
  • [3] F.M. Dong, A survey on the study of real zeros of flow polynomials, J. Graph Theory, 92 (2019) 361–376.
  • [4] F.M. Dong, On graphs whose flow polynomials have real roots only, Electronic Journal of Combinatorics 25(3) (2018) #P3.26.
  • [5] F.M. Dong and K.M. Koh, Foundations of the chromatic polynomial, in the Handbook on the Tutte Polynomial and Related Topics, Jo Ellis-Monaghan and Iain Moffatt (ed.), pp 232–266, CRC press, 2022.
  • [6] F.M. Dong and K.M. Koh, Non-chordal graphs having integral-root chromatic polynomials, Bulletin of Combin. Applic., 22 (1998) 67–77.
  • [7] F.M. Dong, K.M. Koh and K.L. Teo, Chromatic Polynomials and Chromaticity of Graphs, World Scientific, Singapore, 2005.
  • [8] F.M. Dong, K.L. Teo, K.M. Koh and M. Hendy, Non-chordal graphs having integralroot chromatic polynomials (II), Discrete Math., 245 (2002) 247–253.
  • [9] B. Jackson, A Zero-Free Interval for Chromatic Polynomials of Graphs, Combinatorics, Probability and Computing, 2(3) (1993) 325–336.
  • [10] J.P.S. Kung and G.F. Royle, Graphs whose flow polynomials have only integral roots, European Journal of Combinatorics, 32 (2011) 831–840.
  • [11] R.C. Read. Reviewer’s remarks. MR50:6906, 1975.
  • [12] R.C. Read and W.T. Tutte, Chromatic polynomials, in Selected Topics in Graph Theory 3, Academic Press (1988) 15–42.
  • [13] G. Royle, Recent results on chromatic and flow roots of graphs and matroids, in: Surveys in combinatorics, London Math. Soc. Lecture Note Ser., vol. 365, Cambridge Univ. Press, Cambridge, 2009, pp. 289–327.
  • [14] W.T. Tutte, A ring in graph theory, Proc. Cambridge Philos. Soc., 43 (1947) 26–40.
  • [15] W.T. Tutte, Graph Theory, Addison-Welsey, Reading, Mass., 1984.
  • [16] C.D. Wakelin, Chromatic Polynomials, Ph.D. Thesis, University of Nottingham, 1994.