betaFeedback

WorksMathematics

Weighted Euler transforms, real-rootedness, and γγ-positivity

Abstract

For a>0a>0 and a polynomial FF of degree mm, let Ha,FH_{a,F} denote the weighted Euler numerator of FF, defined as the numerator of the generating function ∑k≥0(a)kF(k)xk/k!\sum_{k\ge0}(a)_kF(k)x^k/k! with denominator (1−x)a+m(1-x)^{a+m}. For every a>0a>0, we show the following preservation result: if the numerators Ha,FH_{a,F} and Ha,GH_{a,G} have only real nonpositive zeros, then the same holds for Ha,FGH_{a,FG}. The proof relies on an explicit multiplication formula in falling-factorial coordinates and a characterization of multiplier sequences for generalized Laguerre bases due to Brändén and Ottergren. Furthermore, when the symmetric degrees coincide with the ordinary degrees of the input polynomials, the same operation preserves γγ-positivity. For positive integers a=ra=r, the weight is the Ehrhart polynomial of the standard simplex Δr−1Δ_{r-1}. This yields geometric preservation theorems for products with a single simplex factor, as well as applications to Gorenstein polytopes and a Reeve tetrahedron example.

Introduction

In enumerative combinatorics and Ehrhart theory, one frequently evaluates a polynomial FF at nonnegative integers, or equivalently, computes the generating function of the sequence F(k)F(k). Such a function is rational and its numerator often encodes the core combinatorial data. A natural question is how the numerator behaves under multiplication of the input polynomials.

This paper studies a weighted family of such numerators. Fix a>0a>0, and let (a)k=a(a+1)⋯(a+k−1)(a)_k=a(a+1)\cdots(a+k-1) denote the rising factorial (with $(a)_0=1). Given a nonzero polynomial F∈R[t]F\in\mathbb{R}[t] of degree mm, define its weighted Euler numerator Ha,FH_{a,F} via

∑k≥0(a)kk!F(k)xk=Ha,F(x)(1−x)a+m.(1)\sum_{k\geq0}\frac{(a)_k}{k!}F(k)x^k=\frac{H_{a,F}(x)}{(1-x)^{a+m}}. \tag*{(1)}

By Proposition 2.1, Ha,FH_{a,F} is a polynomial of degree at most mm. As usual, the power (1−x)−a(1-x)^{-a} is understood as a formal power series, or analytically for ∣x∣<1|x|<1.

Two special cases explain the role of the parameter aa. For a=1a=1 the weight is simply 11, yielding the ordinary generating function of F(k)F(k). For a=r∈Na=r\in\mathbb{N}, the weight (r)k/k!(r)_k/k! is the Ehrhart polynomial of the standard simplex Δr−1\Delta_{r-1}; this is the source of the geometric applications in § 5.

For the unweighted case a=1a=1, Wagner [18] proved that H1,FGH_{1,FG} has only real nonpositive zeros whenever H1,FH_{1,F} and H1,GH_{1,G} do. Brändén, Ferroni, and Jochemko [5] established further preservation results, including γ\gamma-positivity. For general a>0a>0, Karp and Vishnyakova [10, Open problem 2] asked whether real negative zeros of Ha,FH_{a,F} and Ha,GH_{a,G} imply the same for Ha,FGH_{a,FG}. Vishnyakova [17, Open Problem 1.9] then posed the analogous question for nonpositive zeros. Since Ha,F(0)=F(0)H_{a,F}(0)=F(0), the nonpositive version is more general, as it additionally allows input polynomials with vanishing constant terms. Below, we prove this nonpositive preservation for all a>0a>0 (Theorem 1.1), and deduce the corresponding γ\gamma-positivity result (Theorem 1.2).

2020 Mathematics Subject Classification. Primary 05A15; Secondary 26C10, 52B20, 33C45.

Key words and phrases. Weighted Euler transforms, real-rootedness, γ\gamma-positivity, multiplier sequences, generalized Laguerre polynomials, Ehrhart polynomials.

For m≥0m \ge0, let

Γm={∑i=0⌊m/2⌋γixi(1+x)m−2i:γi≥0}.(2)\Gamma_m = \left\{ \sum_{i=0}^{\lfloor m/2 \rfloor} \gamma_i x^i(1+x)^{m-2i} : \gamma_i \ge0 \right\}. \tag*{(2)}

A nonzero polynomial P∈ΓmP \in\Gamma_m is said to be γ\gamma-positive (or nonnegative in some literature) with symmetric degree mm. Such a polynomial satisfies the symmetry P(x)=xmP(1/x)P(x) = x^mP(1/x), although its ordinary degree can be strictly smaller than mm. As usual, nonzero constants are regarded as real-rooted.

We prove two preservation theorems. The first addresses real-rootedness.

Theorem 1.1 (Real-rootedness). Let a>0a > 0, and let F,G∈R[t]F,G \in\mathbb{R}[t] be nonzero polynomials. If every zero of Ha,FH_{a,F} and Ha,GH_{a,G} is real and nonpositive, then the same holds for Ha,FGH_{a,FG}.

The second addresses γ\gamma-positivity, under a subtle degree condition.

Theorem 1.2 (γ\gamma-positivity). Let a>0a > 0, and let F,G∈R[t]F,G \in\mathbb{R}[t] be nonzero polynomials of degrees m,nm,n, respectively. If Ha,F∈ΓmH_{a,F} \in\Gamma_m and Ha,G∈ΓnH_{a,G} \in\Gamma_n, then Ha,FG∈Γm+nH_{a,FG} \in\Gamma_{m+n}.

We begin by passing to falling-factorial coordinates. Let La(ti)=(a)iyi\mathcal{L}_a(t^i) = (a)_i y^i denote the linear change of basis, where tit^i is the falling factorial. This map converts the multiplication of polynomials into the bilinear operation

(f⋆ag)(y)=∑k≥0[y(y+1)]kk!(a)kf(k)(y)g(k)(y).(3)(f \star_a g)(y) = \sum_{k \ge0} \frac{[y(y+1)]^k}{k!(a)_k} f^{(k)}(y)g^{(k)}(y). \tag*{(3)}

Under the substitution y=x/(1−x)y = x/(1-x), the numerator Ha,F(x)H_{a,F}(x) transforms into the polynomial La(F)(y)\mathcal{L}_a(F)(y), and its nonpositive zeros map to the zeros of La(F)\mathcal{L}_a(F) in [−1,0][-1,0]. This polynomial La(F)(y)\mathcal{L}_a(F)(y) coincides exactly with the characterizing polynomial in the deep classification of multiplier sequences for generalized Laguerre bases by Brändén and Ottergren [6]. By invoking their characterization as a black box, we bypass several pages of standard technical estimates and obtain Theorem 1.1 in a few lines.

For Theorem 1.2, bilinearity reduces the problem to the γ\gamma-basis polynomials xi(1+x)m−2ix^i(1+x)^{m-2i}. The γ\gamma-positivity for the product of any two such basis polynomials is proved in two steps: Theorem 1.1 ensures the product has only real nonpositive zeros, and the Brändén–Gal theorem then yields γ\gamma-positivity.

Now we describe two geometric applications of these algebraic results when a=r∈Na = r \in\mathbb{N}. Let PP be a dd-dimensional lattice polytope with Ehrhart polynomial LPL_P, and let Δr−1\Delta_{r-1} be the standard simplex. The identity

Hr,LP(x)=hP×Δr−1∗(x)(4)H_{r,L_P}(x) = h^*_{P \times\Delta_{r-1}}(x) \tag*{(4)}

allows us to restate Theorems 1.1 and 1.2 as follows.

Theorem 1.3 (Real-rootedness, geometric form). Let P,QP,Q be lattice polytopes of dimensions d,ed,e, and let r≥1r \ge1 be an integer. If hP×Δr−1∗h^*_{P \times\Delta_{r-1}} and hQ×Δr−1∗h^*_{Q \times\Delta_{r-1}} are real-rooted, then hP×Q×Δr−1∗h^*_{P \times Q \times\Delta_{r-1}} is real-rooted.

Theorem 1.4 (γ\gamma-positivity, geometric form). Let P,QP,Q be lattice polytopes of dimensions d,ed,e, and let r≥1r \ge1 be an integer. If hP×Δr−1∗∈Γdh^*_{P \times\Delta_{r-1}} \in\Gamma_d and hQ×Δr−1∗∈Γeh^*_{Q \times\Delta_{r-1}} \in\Gamma_e, then hP×Q×Δr−1∗∈Γd+eh^*_{P \times Q \times\Delta_{r-1}} \in\Gamma_{d+e}.

A key structural feature of these results is that the output retains only a single simplex factor. This connects to the general study of Ehrhart invariants under polytope operations (see Liu [11]). Moreover, Theorem 1.4 naturally applies to Gorenstein polytopes.

The remainder of the paper is organized as follows. Section 2 establishes the multiplication formula, Section 3 proves the preservation of real-rootedness, Section 4 proves γ\gamma-positivity, and Section 5 develops the geometric consequences, including a Reeve tetrahedron example.

The multiplication formula for weighted Euler numerators

This section reduces the product Ha,FGH_{a,FG} to an explicit bilinear formula (14). This formula is the key tool for the preservation arguments in § 3 and 4.

Falling-factorial coordinates

To simplify multiplication, we pass to the basis of falling factorials tj=t(t−1)⋯(t−j+1)t^{j}=t(t-1)\cdots(t-j+1), with t0=1t^{0}=1. If F(t)=∑j=0mcjtjF(t)=\sum_{j=0}^{m}c_{j}t^{j}, define

La(F)(y)=∑j=0mcj(a)jyj.(5)\mathcal{L}_{a}(F)(y)=\sum_{j=0}^{m}c_{j}(a)_{j}y^{j}. \tag*{(5)}

Since (a)j>0(a)_{j}>0, La\mathcal{L}_{a} is a degree-preserving linear isomorphism. For F≠0F\ne0 of degree mm,

lc⁡(La(F))=(a)mlc⁡(F),(6)\operatorname{lc}(\mathcal{L}_{a}(F))=(a)_{m}\operatorname{lc}(F), \tag*{(6)}

where lc⁡\operatorname{lc} extracts the leading coefficient.

Proposition 2.1. For F=∑j=0mcjtjF=\sum_{j=0}^{m}c_{j}t^{j}, we have

∑k≥0(a)kk!F(k)xk=∑j=0mcj(a)jxj(1−x)m−j(1−x)a+m.(7)\sum_{k\geq0}\frac{(a)_{k}}{k!}F(k)x^{k}=\frac{\sum_{j=0}^{m}c_{j}(a)_{j}x^{j}(1-x)^{m-j}}{(1-x)^{a+m}}. \tag*{(7)}

If m=deg⁡Fm=\deg F, the numerator and La(F)\mathcal{L}_{a}(F) are related by the mutually inverse substitutions

Ha,F(x)=(1−x)mLa(F)(x1−x),(8)H_{a,F}(x)=(1-x)^{m}\mathcal{L}_{a}(F)\left(\frac{x}{1-x}\right), \tag*{(8)}

and

La(F)(y)=(1+y)mHa,F(y1+y).(9)\mathcal{L}_{a}(F)(y)=(1+y)^{m}H_{a,F}\left(\frac{y}{1+y}\right). \tag*{(9)}

Furthermore, the endpoint values satisfy

Ha,F(0)=F(0),Ha,F(1)=(a)mlc⁡(F).(10)H_{a,F}(0)=F(0),\qquad H_{a,F}(1)=(a)_{m}\operatorname{lc}(F). \tag*{(10)}

Proof. The binomial series and the identity (a)j+ℓ=(a)j(a+j)ℓ(a)_{j+\ell}=(a)_{j}(a+j)_{\ell} yield

∑k≥0(a)kk!kjxk=∑k≥j(a)k(k−j)!xk=(a)jxj∑ℓ≥0(a+j)ℓℓ!xℓ=(a)jxj(1−x)a+j.\sum_{k\geq0}\frac{(a)_{k}}{k!}k^{j}x^{k} =\sum_{k\geq j}\frac{(a)_{k}}{(k-j)!}x^{k} =(a)_{j}x^{j}\sum_{\ell\geq0}\frac{(a+j)_{\ell}}{\ell!}x^{\ell} =\frac{(a)_{j}x^{j}}{(1-x)^{a+j}}.

Multiplying by cjc_{j} and summing over jj yields (7). When m=deg⁡Fm=\deg F, the numerator in (7) is exactly (1−x)mLa(F)(x1−x)(1-x)^{m}\mathcal{L}_{a}(F)\left(\frac{x}{1-x}\right), which gives (8). Substituting x=y/(1+y)x=y/(1+y) into (8) gives (9). Finally, evaluating (8) at x=0x=0 and x=1x=1 yields c0=F(0)c_{0}=F(0) and cm(a)m=(a)mlc⁡(F)c_{m}(a)_{m}=(a)_{m}\operatorname{lc}(F). This proves (10). □

Two auxiliary identities

Lemma 2.2. For i,j≥0i,j\geq0,

titj=∑ℓ=0min⁡(i,j)(iℓ)(jℓ)ℓ!ti+j−ℓ.(11)t^{i}t^{j}=\sum_{\ell=0}^{\min(i,j)}\binom{i}{\ell}\binom{j}{\ell}\ell!t^{i+j-\ell}. \tag*{(11)}

Proof. Since both sides are polynomials in tt, it suffices to assume that tt is a sufficiently large integer. For such tt, we give a combinatorial proof. The left-hand side counts pairs of ordered lists of lengths ii and jj chosen from tt labels, with no repetition within each list. Classify such a pair by the number ℓ\ell of common labels. There are (iℓ)(jℓ)ℓ!\binom{i}{\ell}\binom{j}{\ell}\ell! ways to choose and match these common labels. Once these are fixed, the remaining i+j−ℓi+j-\ell positions must receive distinct labels, yielding ti+j−ℓt^{i+j-\ell} choices. This completes the proof. □

Lemma 2.3. If I,JI,J are nonnegative integers and b>0b>0, then

∑r=0min⁡(I,J)I!J!r!(b)r=(b+J)I(b)I.(12)\sum_{r=0}^{\min(I,J)}\frac{I!J!}{r!(b)_r}=\frac{(b+J)_I}{(b)_I}. \tag*{(12)}

Proof. We use the generalized binomial coefficients (uv)=u(u−1)⋯(u−v+1)/v!\binom{u}{v}=u(u-1)\cdots(u-v+1)/v! for real uu and integers v≥0v\ge0. Multiplying the left-hand side by (b)I/I!(b)_I/I! gives

∑r=0min⁡(I,J)(b+I−1I−r)(Jr).\sum_{r=0}^{\min(I,J)}\binom{b+I-1}{I-r}\binom{J}{r}.

This sum is exactly the coefficient of zIz^I in the product (1+z)b+I−1(1+z)J=(1+z)b+I+J−1(1+z)^{b+I-1}(1+z)^J=(1+z)^{b+I+J-1}, which evaluates to (b+I+J−1I)=(b+J)I/I!\binom{b+I+J-1}{I}=(b+J)_I/I!. This is an instance of the Vandermonde convolution. Dividing by (b)I/I!(b)_I/I! completes the proof. □

The multiplication formula

For f,g∈R[y]f,g\in\mathbb{R}[y], we define the induced operation

f⋆ag=La(La−1(f)La−1(g)).(13)f\star_a g=\mathcal{L}_a(\mathcal{L}_a^{-1}(f)\mathcal{L}_a^{-1}(g)). \tag*{(13)}

This operation is bilinear, commutative, and associative. The next proposition gives an explicit formula for it.

Proposition 2.4. For any a>0a>0 and f,g∈R[y]f,g\in\mathbb{R}[y], we have

(f⋆ag)(y)=∑k≥0[y(y+1)]kk!(a)kf(k)(y)g(k)(y).(14)(f\star_a g)(y)=\sum_{k\ge0}\frac{[y(y+1)]^k}{k!(a)_k}f^{(k)}(y)g^{(k)}(y). \tag*{(14)}

If ff and gg are nonzero, the sum is finite with k≤min⁡(deg⁡f,deg⁡g)k\le\min(\deg f,\deg g).

Proof. Since both sides are bilinear, it suffices to verify the identity for monomials f(y)=yif(y)=y^i and g(y)=yjg(y)=y^j. Using La−1(yi)=ti/(a)i\mathcal{L}_a^{-1}(y^i)=t^i/(a)_i and Lemma 2.2, we compute

yi⋆ayj=∑ℓ=0min⁡(i,j)Aℓyi+j−ℓ,Aℓ=(iℓ)(jℓ)ℓ!(a)i+j−ℓ(a)i(a)j.y^i\star_a y^j=\sum_{\ell=0}^{\min(i,j)}A_\ell y^{i+j-\ell},\qquad A_\ell=\binom{i}{\ell}\binom{j}{\ell}\ell!\frac{(a)_{i+j-\ell}}{(a)_i(a)_j}.

On the other hand, the right-hand side of (14) becomes

∑k=0min⁡(i,j)ik‾jk‾k!(a)kyi+j−k(1+y)k.\sum_{k=0}^{\min(i,j)}\frac{i^{\underline{k}}j^{\underline{k}}}{k!(a)_k}y^{i+j-k}(1+y)^k.

Its coefficient of yi+j−ℓy^{i+j-\ell} is

Bℓ=∑k=ℓmin⁡(i,j)ik‾jk‾k!(a)k(kℓ).B_\ell=\sum_{k=\ell}^{\min(i,j)}\frac{i^{\underline{k}}j^{\underline{k}}}{k!(a)_k}\binom{k}{\ell}.

Put I=i−ℓI=i-\ell, J=j−ℓJ=j-\ell, and k=ℓ+rk=\ell+r. Splitting the rising and falling factorials at ℓ\ell yields

Bℓ=iℓ‾jℓ‾ℓ!(a)ℓ∑r=0min⁡(I,J)I!J!r!(a+ℓ)r=iℓ‾jℓ‾ℓ!(a)ℓ(a+j)i−ℓ(a+ℓ)i−ℓ,B_\ell=\frac{i^{\underline{\ell}}j^{\underline{\ell}}}{\ell!(a)_\ell}\sum_{r=0}^{\min(I,J)}\frac{I!J!}{r!(a+\ell)_r} =\frac{i^{\underline{\ell}}j^{\underline{\ell}}}{\ell!(a)_\ell}\frac{(a+j)_{i-\ell}}{(a+\ell)_{i-\ell}},

where the last equality uses Lemma 2.3 with b=a+ℓb=a+\ell. Now

iℓ‾jℓ‾ℓ!=(iℓ)(jℓ)ℓ!,(a)ℓ(a+ℓ)i−ℓ=(a)i,(a+j)i−ℓ=(a)i+j−ℓ(a)j.\frac{i^{\underline{\ell}}j^{\underline{\ell}}}{\ell!}=\binom{i}{\ell}\binom{j}{\ell}\ell!,\qquad (a)_\ell(a+\ell)_{i-\ell}=(a)_i,\qquad (a+j)_{i-\ell}=\frac{(a)_{i+j-\ell}}{(a)_j}.

Thus Bℓ=AℓB_\ell=A_\ell for every ℓ\ell. This completes the proof. □

Preservation of real-rootedness

We now prove Theorem 1.1. Brändén and Ottergren [6] classified multiplier sequences for generalized Laguerre bases, establishing a powerful criterion for real-rootedness preservation in that setting. Remarkably, our weighted Euler numerators map precisely onto their characterizing polynomial, which allows us to invoke their classification directly and bypass the standard technical estimates. (For completeness, a direct proof using the classical Pólya–Schur criterion and Brändén’s bilinear theorem is given in Appendix A.)

The change of variables and the location of zeros

For a polynomial PP of degree at most mm, define

f(y)=(1+y)mP(y1+y).f(y) = (1+y)^m P\left(\frac{y}{1+y}\right).

It is straightforward to check that ff is a polynomial: writing P(x)=∑j=0mpjxjP(x)=\sum_{j=0}^{m}p_jx^j yields f(y)=∑j=0mpjyj(1+y)m−jf(y)=\sum_{j=0}^{m}p_jy^j(1+y)^{m-j}. The inverse substitution is

P(x)=(1−x)mf(x1−x).P(x)=(1-x)^m f\left(\frac{x}{1-x}\right).

In particular, applying (9) gives

La(F)(y)=(1+y)mHa,F(y1+y),m=deg⁡F.(15)\mathcal{L}_a(F)(y)=(1+y)^m H_{a,F}\left(\frac{y}{1+y}\right), \qquad m=\deg F. \tag*{(15)}

Lemma 3.1. Let P∈R[x]P\in\mathbb{R}[x] be a nonzero polynomial with deg⁡P≤m\deg P\le m for some m≥0m\ge0. If every zero of PP is real and nonpositive, then f(y)=(1+y)mP(y1+y)f(y)=(1+y)^mP\left(\frac{y}{1+y}\right) has degree mm and all its zeros in [−1,0][-1,0]. Conversely, if a nonzero polynomial ff has degree mm and all its zeros in [−1,0][-1,0], then P(x)=(1−x)mf(x1−x)P(x)=(1-x)^mf\left(\frac{x}{1-x}\right) is a nonzero polynomial of degree at most mm with only real, nonpositive zeros. Under this correspondence, we have P(0)=f(0)P(0)=f(0); hence P(0)≠0P(0)\ne0 if and only if f(0)≠0f(0)\ne0. Moreover, if ff has positive leading coefficient, then PP has nonnegative coefficients.

Proof. Write P(x)=c∏ν=1d(x+rν)P(x)=c\prod_{\nu=1}^{d}(x+r_\nu) with c≠0c\ne0 and rν≥0r_\nu\ge0, where d=deg⁡Pd=\deg P. Then

f(y)=c(1+y)m−d∏ν=1d((1+rν)y+rν).f(y)=c(1+y)^{m-d}\prod_{\nu=1}^{d}\bigl((1+r_\nu)y+r_\nu\bigr).

This has degree mm, with roots at −rν/(1+rν)∈(−1,0]-r_\nu/(1+r_\nu)\in(-1,0] for each ν\nu, and at −1-1 with multiplicity m−dm-d.

Conversely, suppose f(y)=c∏ν=1m(y+ρν)f(y)=c\prod_{\nu=1}^{m}(y+\rho_\nu) with c≠0c\ne0 and 0≤ρν≤10\le\rho_\nu\le1. The inverse substitution yields

P(x)=c∏ν=1m(ρν+(1−ρν)x).P(x)=c\prod_{\nu=1}^{m}\bigl(\rho_\nu+(1-\rho_\nu)x\bigr).

When ρν<1\rho_\nu<1, this contributes a nonpositive root −ρν/(1−ρν)-\rho_\nu/(1-\rho_\nu); when ρν=1\rho_\nu=1, the factor is the constant 11. Therefore PP is nonzero, and its degree is mm minus the multiplicity of −1-1 as a root of ff. The equality P(0)=f(0)P(0)=f(0) gives the claim about the origin. Finally, all factors in the displayed expression for PP have nonnegative coefficients, so the same holds for PP when c>0c>0. □

Multiplier sequences for the Laguerre basis

Fix a>0a>0 and put α=a−1>−1\alpha=a-1>-1. Let

Ln(α)(u)=∑j=0n(−1)jj!(n+αn−j)uj,n≥0,L_n^{(\alpha)}(u)=\sum_{j=0}^{n}\frac{(-1)^j}{j!}\binom{n+\alpha}{n-j}u^j,\qquad n\ge0,

be the generalized Laguerre polynomials. Since their degree is exactly nn, they form a basis of $\mathbb{R}[u].

For a nonzero polynomial F∈R[t]F \in\mathbb{R}[t], define the linear operator TFT_F on R[u]\mathbb{R}[u] by

TF(Ln(α)(u))=F(n)Ln(α)(u),n≥0.T_F\left(L_n^{(\alpha)}(u)\right)=F(n)L_n^{(\alpha)}(u),\qquad n\geq0.

(Here and below, preservation of real-rootedness includes mapping to the zero polynomial.)

We use the following characterization due to Brändén and Ottergren [6 Theorem 1.1], stated here in a form adapted to our notation. Given real sequences {aj}j≥0\{a_j\}_{j\geq0} and {λn}n≥0\{\lambda_n\}_{n\geq0} related by

λn=∑j=0naj(nj),\lambda_n=\sum_{j=0}^{n}a_j\binom{n}{j},

assume that {λn}\{\lambda_n\} is not a trivial sequence (i.e., it does not consist of at most two consecutive nonzero terms). Define the power series

p(y)=∑j=0∞(j+αj)ajyj.p(y)=\sum_{j=0}^{\infty}\binom{j+\alpha}{j}a_jy^j.

Then the operator T(Ln(α)(u))=λnLn(α)(u)T(L_n^{(\alpha)}(u))=\lambda_nL_n^{(\alpha)}(u) preserves real-rootedness if and only if p(y)p(y) is a real-rooted polynomial with all its zeros contained in [−1,0][-1,0].

We now apply this to the sequence λn=F(n)\lambda_n=F(n). Writing F(t)=∑j=0mcjtjF(t)=\sum_{j=0}^{m}c_jt^j and using tj=j!(tj)t^j=j!\binom{t}{j}, we obtain

F(n)=∑j=0m(j!cj)(nj).F(n)=\sum_{j=0}^{m}(j!c_j)\binom{n}{j}.

Thus, the sequence corresponds to aj=j!cja_j=j!c_j for 0≤j≤m0\leq j\leq m, and aj=0a_j=0 for j>mj>m. Since FF is a nonzero polynomial, it has at most mm roots, so F(n)≠0F(n)\neq0 for infinitely many nn. Hence the sequence (F(n))n≥0(F(n))_{n\geq0} is not trivial. For this sequence, the power series p(y)p(y) truncates to the polynomial

p(y)=∑j=0m(j+αj)(j!cj)yj=∑j=0m(j+α−1j)(j!cj)yj=∑j=0mcj(α)jyj.(16)p(y)=\sum_{j=0}^{m}\binom{j+\alpha}{j}(j!c_j)y^j =\sum_{j=0}^{m}\binom{j+\alpha-1}{j}(j!c_j)y^j =\sum_{j=0}^{m}c_j(\alpha)_jy^j. \tag*{(16)}

Comparing this with (5), we see that p(y)p(y) is precisely Lα(F)(y)\mathcal{L}_\alpha(F)(y). Therefore, the theorem of [6 Theorem 1.1] gives

TF preserves real-rootedness⟺Lα(F) has all its zeros in [−1,0].(17)T_F\text{ preserves real-rootedness}\Longleftrightarrow\mathcal{L}_\alpha(F)\text{ has all its zeros in }[-1,0]. \tag*{(17)}

By Lemma 3.1 and (15), the right-hand side is equivalent to saying that all zeros of Hα,FH_{\alpha,F} are real and nonpositive. We conclude that

TF preserves real-rootedness⟺Hα,F has only real nonpositive zeros.(18)T_F\text{ preserves real-rootedness}\Longleftrightarrow H_{\alpha,F}\text{ has only real nonpositive zeros}. \tag*{(18)}

Now we are ready to prove the real-rootedness of the Euler numerators.

Proof of Theorem 1.1. Assume that Hα,FH_{\alpha,F} and Hα,GH_{\alpha,G} have only real nonpositive zeros. By (18), both TFT_F and TGT_G preserve real-rootedness; consequently, their composition TF∘TGT_F\circ T_G also preserves real-rootedness. For every n≥0n\geq0, we compute

(TF∘TG)(Ln(α))=TF(G(n)Ln(α))=F(n)G(n)Ln(α)=TFG(Ln(α)).(T_F\circ T_G)\left(L_n^{(\alpha)}\right) =T_F\left(G(n)L_n^{(\alpha)}\right) =F(n)G(n)L_n^{(\alpha)} =T_{FG}\left(L_n^{(\alpha)}\right).

Since these two linear operators agree on the basis {Ln(α)}\{L_n^{(\alpha)}\}, they must be identical. Thus, TFGT_{FG} also preserves real-rootedness. By applying the equivalence (18) once more, we deduce that all zeros of Hα,FGH_{\alpha,FG} are real and nonpositive. ∎

Preservation of γ\gamma-positivity

We now focus on Theorem 1.2. We begin by fixing terminology. A nonzero polynomial P∈R[x]P \in\mathbb{R}[x] is symmetric of degree m∈Z≥0m \in\mathbb{Z}_{\ge0} if

deg⁡P≤mandP(x)=xmP(1/x).\deg P \le m \qquad\text{and}\qquad P(x)=x^{m}P(1/x).

Equivalently, this means pj=pm−jp_j=p_{m-j} for every jj when writing P(x)=∑j=0mpjxjP(x)=\sum_{j=0}^{m}p_jx^j. If J=supp⁡(P)J=\operatorname{supp}(P) denotes the set of indices with nonzero coefficients, then such an mm is unique and equals min⁡J+max⁡J\min J+\max J. We call this the symmetric degree of PP and denote it by

sdeg⁡(P)=min⁡{j:pj≠0}+max⁡{j:pj≠0}.\operatorname{sdeg}(P)=\min\{j:p_j\ne0\}+\max\{j:p_j\ne0\}.

The following basic properties follow directly from the definition:

(S1) sdeg⁡(1)=0\operatorname{sdeg}(1)=0, sdeg⁡(1+x)=1\operatorname{sdeg}(1+x)=1, sdeg⁡(x)=2\operatorname{sdeg}(x)=2, and sdeg⁡(xn)=2n\operatorname{sdeg}(x^n)=2n for n≥0n\ge0.

(S2) sdeg⁡(PQ)=sdeg⁡(P)+sdeg⁡(Q)\operatorname{sdeg}(PQ)=\operatorname{sdeg}(P)+\operatorname{sdeg}(Q) for nonzero P,Q∈R[x]P,Q\in\mathbb{R}[x].

(S3) sdeg⁡(P)=deg⁡P\operatorname{sdeg}(P)=\deg P if and only if p0≠0p_0\ne0.

By (S3), the symmetric degree can exceed the ordinary degree exactly when the constant term vanishes. For instance, deg⁡(ax)=1\deg(ax)=1 but sdeg⁡(ax)=2\operatorname{sdeg}(ax)=2 for any a≠0a\ne0.

Recall from (2) that Γm\Gamma_m is the cone generated by the polynomials xi(1+x)m−2ix^i(1+x)^{m-2i}.

The polynomials

Bm,i(x)=xi(1+x)m−2i,0≤i≤⌊m/2⌋,(19)B_{m,i}(x)=x^i(1+x)^{m-2i},\qquad0\le i\le\lfloor m/2\rfloor, \tag*{(19)}

form a basis of the space of polynomials symmetric of degree mm. Indeed, the space has dimension ⌊m/2⌋+1\lfloor m/2\rfloor+1, and the degrees deg⁡Bm,i=m−i\deg B_{m,i}=m-i are all distinct for 0≤i≤⌊m/2⌋0\le i\le\lfloor m/2\rfloor. Therefore, every such PP has a unique expansion

P(x)=∑i=0⌊m/2⌋γixi(1+x)m−2i.(20)P(x)=\sum_{i=0}^{\lfloor m/2\rfloor}\gamma_i x^i(1+x)^{m-2i}. \tag*{(20)}

Consequently, P∈ΓmP\in\Gamma_m if and only if sdeg⁡(P)=m\operatorname{sdeg}(P)=m and all coefficients γi\gamma_i in this expansion are nonnegative. Equivalently, Γm\Gamma_m is precisely the cone generated by the Bm,iB_{m,i}.

We will use the following elementary properties:

(G1) xk(1+x)ℓ∈Γ2k+ℓx^k(1+x)^\ell\in\Gamma_{2k+\ell} for all k,ℓ≥0k,\ell\ge0. In particular, x∈Γ2x\in\Gamma_2 and 1+x∈Γ11+x\in\Gamma_1.

(G2) Γm⋅Γn⊆Γm+n\Gamma_m\cdot\Gamma_n\subseteq\Gamma_{m+n}: if P∈ΓmP\in\Gamma_m and Q∈ΓnQ\in\Gamma_n, then PQ∈Γm+nPQ\in\Gamma_{m+n}. Indeed, Bm,iBn,j=Bm+n,i+jB_{m,i}B_{n,j}=B_{m+n,i+j}, so multiplying the γ\gamma-expansions (20) gives a nonnegative expansion in the basis of Γm+n\Gamma_{m+n}.

(G3) For α>0\alpha>0, (x+α)(x+α−1)∈Γ2(x+\alpha)(x+\alpha^{-1})\in\Gamma_2. Indeed,

(x+α)(x+α−1)=x2+(α+α−1)x+1=(1+x)2+(α+α−1−2)x,(x+\alpha)(x+\alpha^{-1})=x^2+(\alpha+\alpha^{-1})x+1=(1+x)^2+(\alpha+\alpha^{-1}-2)x,

and the coefficient of xx is nonnegative.

Note that (G3) is the only step where the reciprocal pairing of roots appears. Combining (G1)–(G3) with the factorization of real-rooted polynomials yields the following classical result, which is due to Brändén [4 Lemma 4.1] and Gal [9 Remark 3.1.1]; see also [1 Section 1].

Theorem 4.1 (Brändén–Gal). Let P∈R≥0[x]P\in\mathbb{R}_{\ge0}[x] be symmetric of degree mm. If all zeros of PP are real and nonpositive, then P∈ΓmP\in\Gamma_m.

Next, we verify that symmetry is preserved under our operation.

Proposition 4.2. Let a>0a>0, and let F,G∈R[t]F,G\in\mathbb{R}[t] be nonzero polynomials of degrees m,nm,n, respectively. If Ha,FH_{a,F} and Ha,GH_{a,G} are symmetric of degrees mm and nn, then Ha,FGH_{a,F}G is symmetric of degree m+nm+n.

Proof. Put P=Ha,FP=H_{a,F} and f=La(F)f=\mathcal{L}_a(F). By (8), we have

f(y)=(1+y)mP(y1+y).(21)f(y)=(1+y)^mP\left(\frac{y}{1+y}\right). \tag*{(21)}

We claim that PP is symmetric of degree mm if and only if

f(−1−y)=(−1)mf(y).(22)f(-1-y)=(-1)^m f(y). \tag*{(22)}

For the forward implication, suppose P(x)=xmP(1/x)P(x)=x^mP(1/x). Substituting y=−1−xy=-1-x into (21) yields

f(−1−x)=(−x)mP(−1−x−x)=(−x)mP(1+xx)=(−x)m(1+xx)mP(x1+x)=(−1)mf(x).\begin{aligned} f(-1-x)&=(-x)^mP\left(\frac{-1-x}{-x}\right)\\ &=(-x)^mP\left(\frac{1+x}{x}\right)\\ &=(-x)^m\left(\frac{1+x}{x}\right)^mP\left(\frac{x}{1+x}\right)=(-1)^m f(x). \end{aligned}

The backward implication is obtained similarly using the substitution y=x1−xy=\frac{x}{1-x}.

Analogously, for g=La(G)g=\mathcal{L}_a(G), we have g(−1−y)=(−1)ng(y)g(-1-y)=(-1)^n g(y). Differentiating kk times gives

f(k)(−1−y)=(−1)m+kf(k)(y),g(k)(−1−y)=(−1)n+kg(k)(y).f^{(k)}(-1-y)=(-1)^{m+k}f^{(k)}(y),\qquad g^{(k)}(-1-y)=(-1)^{n+k}g^{(k)}(y).

Since (−1−y)(−y)=y(y+1)(-1-y)(-y)=y(y+1), each summand in (14) acquires the same factor (−1)m+n(-1)^{m+n} under y↦−1−yy\mapsto-1-y. Hence

(f⋆ag)(−1−y)=(−1)m+n(f⋆ag)(y).(f\star_a g)(-1-y)=(-1)^{m+n}(f\star_a g)(y).

Because f⋆ag=La(FG)f\star_a g=\mathcal{L}_a(FG), we apply the equivalence (22) once again to conclude that Ha,FGH_{a,FG} is symmetric of degree m+nm+n. □

We now prove Theorem 1.2. Note that the two gamma-positive inputs are not required to be real-rooted.

Proof of Theorem 1.2. Write

Ha,F(x)=∑iγiBm,i(x),Ha,G(x)=∑jηjBn,j(x),H_{a,F}(x)=\sum_i\gamma_iB_{m,i}(x),\qquad H_{a,G}(x)=\sum_j\eta_jB_{n,j}(x),

with all γi,ηj≥0\gamma_i,\eta_j\geq0.

For each 0≤i≤⌊m/2⌋0\leq i\leq\lfloor m/2\rfloor, define

fi(y)=(1+y)mBm,i(y1+y)=[y(1+y)]i(1+2y)m−2i,Fi=La−1(fi).(23)f_i(y)=(1+y)^mB_{m,i}\left(\frac{y}{1+y}\right)=[y(1+y)]^i(1+2y)^{m-2i},\qquad F_i=\mathcal{L}_a^{-1}(f_i). \tag*{(23)}

Although Bm,iB_{m,i} has degree m−im-i, the polynomial fif_i has degree exactly mm, with leading coefficient 2m−2i2^{m-2i}. Since La\mathcal{L}_a preserves degrees,

deg⁡Fi=m,lc⁡(Fi)=2m−2i(a)m>0,Ha,Fi=Bm,i.\deg F_i=m,\qquad\operatorname{lc}(F_i)=\frac{2^{m-2i}}{(a)_m}>0,\qquad H_{a,F_i}=B_{m,i}.

The polynomials GjG_j are defined analogously, and we have

deg⁡Gj=n,lc⁡(Gj)=2n−2j(a)n>0,Ha,Gj=Bn,j.\deg G_j=n,\qquad\operatorname{lc}(G_j)=\frac{2^{n-2j}}{(a)_n}>0,\qquad H_{a,G_j}=B_{n,j}.

Fix i,ji,j and set Uij=Ha,FiGjU_{ij}=H_{a,F_iG_j}. We show that Uij∈Γm+nU_{ij}\in\Gamma_{m+n} by verifying the conditions of the Brändén–Gal theorem.

First, since Bm,iB_{m,i} and Bn,jB_{n,j} have no zeros outside {0,−1}\{0,-1\}, Theorem 1.1 implies that all zeros of UijU_{ij} are real and nonpositive.

Second, by (10) of Proposition 2.1, we have

Uij(1)=Ha,FiGj(1)=(a)m+nlc⁡(Fi)lc⁡(Gj)>0.U_{ij}(1)=H_{a,F_iG_j}(1)=(a)_{m+n}\operatorname{lc}(F_i)\operatorname{lc}(G_j)>0.

Since UijU_{ij} has only real nonpositive zeros, being positive at 11 ensures that all its coefficients are nonnegative.

Third, by Proposition 4.2, UijU_{ij} is symmetric of degree m+nm+n. Theorem 4.1 therefore applies to yield Uij∈Γm+nU_{ij} \in\Gamma_{m+n}. Finally, by the linearity of La−1\mathcal{L}_a^{-1} and (23), we have

F=∑iγiFi,G=∑jηjGj.F = \sum_i \gamma_i F_i,\qquad G = \sum_j \eta_j G_j.

All products FiGjF_iG_j have degree m+nm+n, so their weighted Euler series share the same denominator (1−x)a+m+n(1-x)^{a+m+n}. Consequently,

Ha,FG(x)=∑i,jγiηjUij(x).H_{a,FG}(x) = \sum_{i,j} \gamma_i\eta_j U_{ij}(x).

The theorem then follows, since the cone Γm+n\Gamma_{m+n} is closed under nonnegative linear combinations. □

Remark 4.3. The requirement that deg⁡F=m\deg F=m and deg⁡G=n\deg G=n in Theorem 1.2 is essential. The map F↦Ha,FF \mapsto H_{a,F} is not globally linear: the denominator of its generating function depends on deg⁡F\deg F. For example, with a=1a=1, H1,t=xH_{1,t}=x and H1,1=1H_{1,1}=1, but

H1,t+1=1≠x+1=H1,t+H1,1.H_{1,t+1}=1\ne x+1=H_{1,t}+H_{1,1}.

By imposing deg⁡F=m\deg F=m and Ha,F∈ΓmH_{a,F}\in\Gamma_m, the ordinary degree and the symmetric degree are forced to coincide. This unifies the denominators, ensuring that the linear combinations used in the proof (such as F=∑iγiFiF=\sum_i\gamma_iF_i) are well-defined.

Without this degree condition, the conclusion can fail. For instance, take a=1a=1 and F(t)=G(t)=(t+1)2F(t)=G(t)=(t+1)^2. Then H1,F=H1,G=1+x∈Γ1H_{1,F}=H_{1,G}=1+x\in\Gamma_1. Direct computation yields H1,FG=1+11x+11x2+x3∈Γ3H_{1,FG}=1+11x+11x^2+x^3\in\Gamma_3, which does not belong to the predicted Γ1+1=Γ2\Gamma_{1+1}=\Gamma_2. This shows that the degree condition is indeed necessary.

A common simplex factor

Ehrhart series and the Pochhammer weight

Let PP be a dd-dimensional lattice polytope. We work in its affine span and, after translating by a lattice vertex, regard PP as a full-dimensional polytope in Rd\mathbb{R}^d with the induced lattice; the lattice on a Cartesian product is the product lattice (see [2] for standard background). The Ehrhart function of PP is

LP(k)=∣kP∩Zd∣,k∈Z≥0.L_P(k)=\lvert kP\cap\mathbb{Z}^d\rvert,\qquad k\in\mathbb{Z}_{\geq0}.

Ehrhart’s theorem states that LP(k)L_P(k) is a polynomial in kk of degree dd, with LP(0)=1L_P(0)=1 [8], and its generating function takes the form

∑k≥0LP(k)xk=hP∗(x)(1−x)d+1.(24)\sum_{k\geq0}L_P(k)x^k=\frac{h_P^*(x)}{(1-x)^{d+1}}. \tag*{(24)}

The numerator hP∗(x)h_P^*(x) is called the h∗h^*-polynomial of PP. It has degree at most dd, nonnegative integer coefficients, and constant term 11; see [16, 2]. In particular, if hP∗h_P^* is real-rooted, its zeros are strictly negative. (As usual, nonzero constants are considered real-rooted.)

For two lattice polytopes PP and QQ, the Cartesian product satisfies

LP×Q(k)=LP(k)LQ(k).(25)L_{P\times Q}(k)=L_P(k)L_Q(k). \tag*{(25)}

The preservation of real-rootedness under this product is a classical result of Wagner [18 Theorem 0.2]; see Liu [11 Section 2] for a broader survey of operations on lattice polytopes.

Fix an integer r≥1r\geq1 and let

Δr−1=conv⁡(0,e1,…,er−1)⊆Rr−1\Delta_{r-1}=\operatorname{conv}(0,e_1,\ldots,e_{r-1})\subseteq\mathbb{R}^{r-1}

be the standard simplex, with Δ0\Delta_0 being a point. The lattice points of kΔr−1k\Delta_{r-1} correspond to nonnegative integer tuples (u1,…,ur−1)(u_1,\ldots,u_{r-1}) satisfying u1+⋯+ur−1≤ku_1+\cdots+u_{r-1}\leq k. By setting ur=k−u1−⋯−ur−1u_r=k-u_1-\cdots-u_{r-1}, these are in bijection with the nonnegative integer solutions of u1+⋯+ur=ku_1+\cdots+u_r=k. Consequently,

LΔr−1(k)=(k+r−1r−1)=(r)kk!.(26)L_{\Delta_{r-1}}(k)=\binom{k+r-1}{r-1}=\frac{(r)_k}{k!}. \tag*{(26)}

Equivalently, the polynomial Wr(t)=(t+r−1r−1)W_r(t)=\binom{t+r-1}{r-1} is the Ehrhart polynomial of Δr−1\Delta_{r-1} (with the empty product being 11 for r=1r=1). The Pochhammer weight (a)k/k!(a)_k/k! therefore has a direct geometric interpretation at every positive integer parameter.

Proposition 5.1. Let PP be a lattice polytope of dimension dd, and let r≥1r\geq1 be an integer. Then

Hr,LP(x)=hP×Δr−1∗(x).(27)H_{r,L_P}(x)=h^*_{P\times\Delta_{r-1}}(x). \tag*{(27)}

In particular, deg⁡hP×Δr−1∗≤d\deg h^*_{P\times\Delta_{r-1}}\leq d.

Proof. By (25) and (26), we have

∑k≥0(r)kk!LP(k)xk=∑k≥0LP×Δr−1(k)xk.\sum_{k\geq0}\frac{(r)_k}{k!}L_P(k)x^k=\sum_{k\geq0}L_{P\times\Delta_{r-1}}(k)x^k.

The left-hand side is the weighted Euler series associated to F=LPF=L_P. Since deg⁡LP=d\deg L_P=d, its denominator is (1−x)r+d(1-x)^{r+d}. On the other hand, the polytope P×Δr−1P\times\Delta_{r-1} has dimension d+r−1d+r-1, so its Ehrhart series shares the same denominator (1−x)r+d(1-x)^{r+d}. Comparing the numerators yields (27). The degree bound then follows from the degree bound for Hr,LPH_{r,L_P}. □

Products with one common simplex

We now prove Theorems 1.3 and 1.4. Using Proposition 5.1 to translate into the algebraic setting, both proofs are immediate applications of our main results (Theorems 1.1 and 1.2). Afterward, we explain a geometric feature of the output.

Proof of Theorems 1.3 and 1.4. Put F=LPF=L_P and G=LQG=L_Q. Then deg⁡F=d\deg F=d, deg⁡G=e\deg G=e, and LP×Q=FGL_{P\times Q}=FG. By Proposition 5.1, we have

Hr,F=hP×Δr−1∗,Hr,G=hQ×Δr−1∗,Hr,FG=hP×Q×Δr−1∗.H_{r,F}=h^*_{P\times\Delta_{r-1}},\qquad H_{r,G}=h^*_{Q\times\Delta_{r-1}},\qquad H_{r,FG}=h^*_{P\times Q\times\Delta_{r-1}}.

Under the hypotheses of Theorem 1.3, the first two numerators are real-rooted. Since Ehrhart numerators have nonnegative coefficients and constant term 11, all their zeros are strictly negative. Applying Theorem 1.1 with a=ra=r shows that Hr,FGH_{r,FG} has only real nonpositive zeros, proving Theorem 1.3.

Under the hypotheses of Theorem 1.4, we instead have Hr,F∈ΓdH_{r,F}\in\Gamma_d and Hr,G∈ΓeH_{r,G}\in\Gamma_e. Together with deg⁡F=d\deg F=d and deg⁡G=e\deg G=e, these are precisely the hypotheses of Theorem 1.2 with a=ra=r. Hence

Hr,FG∈Γd+e,H_{r,FG}\in\Gamma_{d+e},

which proves Theorem 1.4. □

Corollary 5.2. For each positive integer rr, the class

Cr={P:hP×Δr−1∗ is real-rooted}\mathcal{C}_r=\{P:h^*_{P\times\Delta_{r-1}}\text{ is real-rooted}\}

is closed under Cartesian products. More generally, if P1,…,Ps∈CrP_1,\ldots,P_s\in\mathcal{C}_r, then hP1×⋯×Ps×Δr−1∗h^*_{P_1\times\cdots\times P_s\times\Delta_{r-1}} is real-rooted. Likewise, if P1,…,PsP_1,\ldots,P_s are lattice polytopes with hPi×Δr−1∗∈Γdih^*_{P_i\times\Delta_{r-1}}\in\Gamma_{d_i} for every ii, where di=dim⁡Pid_i=\dim P_i, then the same output belongs to Γd1+⋯+ds\Gamma_{d_1+\cdots+d_s}.

Proof. The assertion for two factors is the combination of Theorem 1.3 and Theorem 1.4. Apply them successively to P1×⋯×PjP_1\times\cdots\times P_j and Pj+1P_{j+1}, keeping the same parameter rr. At the jjth step the symmetry degree in the second assertion is d1+⋯+djd_1+\cdots+d_j. □

A common Gorenstein index

We next discuss a geometric condition that supplies the symmetry degrees required in Theorem 1.4. A full-dimensional lattice polytope A⊆RDA \subseteq\mathbb{R}^{D} is reflexive if 00 lies in its interior and its polar

A∨={y∈RD:⟨y,x⟩≤1 for every x∈A}A^{\vee}=\{y\in\mathbb{R}^{D}:\langle y,x\rangle\le1\text{ for every }x\in A\}

is a lattice polytope. Equivalently, every facet inequality can be written as ⟨n,x⟩≤1\langle n,x\rangle\le1 for some primitive integer vector nn.

The codegree of a lattice polytope PP is

codeg⁡P=min⁡{k≥1:relint⁡(kP)∩Zd≠∅}.\operatorname{codeg} P=\min\{k\ge1:\operatorname{relint}(kP)\cap\mathbb{Z}^{d}\ne\varnothing\}.

We say PP is Gorenstein of index rr if r=codeg⁡Pr=\operatorname{codeg} P, the polytope rPrP has a unique interior lattice point uu, and rP−urP-u is reflexive. These definitions are taken in the affine lattice of PP. A point is Gorenstein of index 1. We refer to [7, 2] for the standard connections with Ehrhart series.

We will need the following well-known properties of Gorenstein polytopes; see [2] and [12 Theorem 4.3].

Lemma 5.3. Let AA be a DD-dimensional Gorenstein lattice polytope of index rr. Then

hA∗(x)=xD+1−rhA∗(1/x),deg⁡hA∗=D+1−r.(28)h_A^*(x)=x^{D+1-r}h_A^*(1/x),\qquad\deg h_A^*=D+1-r. \tag*{(28)}

Moreover, the Cartesian product of two Gorenstein lattice polytopes of the same index rr is Gorenstein of index $r.

For r≥2r\ge2, the polytope rΔr−1r\Delta_{r-1} has a unique interior lattice point, namely the vector 1=(1,…,1)∈Zr−1\mathbf{1}=(1,\ldots,1)\in\mathbb{Z}^{r-1}. Indeed, the interior of kΔr−1k\Delta_{r-1} consists of points with strictly positive coordinates and coordinate sum less than kk. Since 1\mathbf{1} has coordinate sum r−1<rr-1<r, it lies in the interior of rΔr−1r\Delta_{r-1}; moreover, no such point exists for k<rk<r. Translating by −1-\mathbf{1}, the facet inequalities become

−xi≤1(1≤i≤r−1),x1+⋯+xr−1≤1.-x_i\le1\quad(1\le i\le r-1),\qquad x_1+\cdots+x_{r-1}\le1.

Since the right-hand sides are all 1, the polytope is reflexive. Hence Δr−1\Delta_{r-1} is Gorenstein of index rr. (For r=1r=1, this holds trivially by our convention for a point.)

Corollary 5.4. Let PP and QQ be Gorenstein lattice polytopes of the same index rr. If hP×Δr−1∗h^*_{P\times\Delta_{r-1}} and hQ×Δr−1∗h^*_{Q\times\Delta_{r-1}} are gamma-positive, then hP×Q×Δr−1∗h^*_{P\times Q\times\Delta_{r-1}} is gamma-positive.

Proof. Put d=dim⁡Pd=\dim P and e=dim⁡Qe=\dim Q. Lemma 5.3 shows that P×Δr−1P\times\Delta_{r-1} is Gorenstein of index rr. Its dimension is d+r−1d+r-1, so its h∗h^*-polynomial has degree and symmetry degree (d+r−1)+1−r=d(d+r-1)+1-r=d. Similarly, the symmetry degree for Q×Δr−1Q\times\Delta_{r-1} is ee. The two gamma-positive inputs therefore belong to Γd\Gamma_d and Γe\Gamma_e, respectively. Theorem 1.4 gives the conclusion, with symmetry degree d+ed+e. □

The Reeve tetrahedron

Consider the Reeve tetrahedron [15]

R=conv⁡(0,(1,0,0),(0,1,0),(1,1,2)).R=\operatorname{conv}\left(0,(1,0,0),(0,1,0),(1,1,2)\right).

A direct computation of its Ehrhart series gives

LR(k)=(k+1)((k+1)2+2)3,hR∗(x)=1+x2.(29)L_R(k)=\frac{(k+1)\left((k+1)^2+2\right)}{3},\qquad h_R^*(x)=1+x^2. \tag*{(29)}

This numerator is not real-rooted. We give two applications of our theorems to this non-real-rooted example.

First application: adjoining a segment. Let Δ1\Delta_1 be the unit segment. Since its Ehrhart polynomial is k+1k+1, multiplying the kkth series coefficient by k+1k+1 acts as 1+xddx1+x\frac{d}{dx}. We obtain

∑k≥0LR×Δ1(k)xk=(1+xddx)1+x2(1−x)4=(1−x)(1+3x2)+4x(1+x2)(1−x)5=(1+x)3(1−x)5.\begin{aligned} \sum_{k\geq0} L_{R\times\Delta_1}(k)x^k=\left(1+x\frac{d}{dx}\right)\frac{1+x^2}{(1-x)^4} \\ &=\frac{(1-x)(1+3x^2)+4x(1+x^2)}{(1-x)^5}=\frac{(1+x)^3}{(1-x)^5}. \end{aligned}

Thus hR×Δ1∗(x)=(1+x)3h^{*}_{R\times\Delta_1}(x)=(1+x)^3. In particular, the non-real-rooted numerator 1+x21+x^2 is repaired by adjoining a single segment: R∉C1R\notin C_1 but R∈C2R\in C_2. Moreover, since the output belongs to Γ3\Gamma_3, Corollary 5.2 implies that for every s≥1s\geq1, the numerator hR×s×Δ1∗(x)h^{*}_{R^{\times s}\times\Delta_1}(x) is real-rooted and belongs to Γ3s\Gamma_{3s}.

Second application: Cartesian powers. Next we consider the powers R×sR^{\times s} without adjoining any simplex factor. For s≥1s\geq1, define

Bs(x)=hR×s∗(x),∑k≥0LR(k)sxk=Bs(x)(1−x)3s+1.(30)B_s(x)=h^{*}_{R^{\times s}}(x),\qquad\sum_{k\geq0}L_R(k)^s x^k=\frac{B_s(x)}{(1-x)^{3s+1}}. \tag*{(30)}

We will show that BsB_s has only real negative roots if and only if s≥3s\geq3.

Since Bs(0)=1B_s(0)=1, zero is never a root. For s=1s=1, B1(x)=1+x2B_1(x)=1+x^2 has no real roots. For s=2s=2, a direct computation gives

B2(x)=(1+x)(x4+8x3+22x2+8x+1),B_2(x)=(1+x)(x^4+8x^3+22x^2+8x+1),

and the quartic factor equals (x2+4x+1)2+4x2>0(x^2+4x+1)^2+4x^2>0 for all real xx. Hence B2B_2 has only one real root. Thus the threshold s≥3s\geq3 is sharp.

We now prove the assertion for s≥3s\geq3. By Theorem 1.1 with a=1a=1, the product of two real-rooted polynomials with real nonpositive zeros has real nonpositive zeros. Since LR×p(k)=LR(k)pL_{R^{\times p}}(k)=L_R(k)^p, this implies

Bp, Bq real-rooted ⟹ Bp+q real-rooted.(31)B_p,\ B_q\ \text{real-rooted}\ \Longrightarrow\ B_{p+q}\ \text{real-rooted}. \tag*{(31)}

It therefore suffices to verify the base cases s=3,4,5s=3,4,5, as the identities 6=3+36=3+3, 7=3+47=3+4, 8=3+58=3+5, and the induction step s=3+(s−3)s=3+(s-3) for s≥6s\geq6 complete the proof.

To check these base cases, we make the substitution u=−x−1/xu=-x-1/x. For u>2u>2, the roots of x2+ux+1=0x^2+ux+1=0 are real and negative. We express BsB_s in terms of a lower-degree monic polynomial qsq_s. Put d=3s−1d=3s-1, ℓ=⌊d/2⌋\ell=\lfloor d/2\rfloor, and ε=d−2ℓ\varepsilon=d-2\ell. Then

Bs(x)=(−1)ℓxℓ(1+x)εqs(−x−1/x).B_s(x)=(-1)^\ell x^\ell(1+x)^\varepsilon q_s(-x-1/x).

The extra factor 1+x1+x occurs when ss is even. The coefficients of BsB_s are given by LRL_R via the generating function.

For s=3s=3, B3B_3 has degree 8:

B3(x)=1+54x+736x2+3274x3+5310x4+3274x5+736x6+54x7+x8.B_3(x)=1+54x+736x^2+3274x^3+5310x^4+3274x^5+736x^6+54x^7+x^8.

It can be written as B3(x)=x4q3(−x−1/x)B_3(x)=x^4q_3(-x-1/x), with

q3(u)=u4−54u3+732u2−3112u+3840.q_3(u)=u^4-54u^3+732u^2-3112u+3840.

Evaluating q3q_3 at u=2,3,5,12,36u=2,3,5,12,36 gives 128,−285,455,−672,672128,-285,455,-672,672. The signs alternate, so the intermediate value theorem ensures a root in each interval (2,3),(3,5),(5,12),(12,36)(2,3),(3,5),(5,12),(12,36). All evaluations use exact rational arithmetic. Since deg⁡q3=4\deg q_3=4, all its roots are real and lie in (2,∞)(2,\infty). Each root uiu_i yields two negative roots of B3B_3 via x2+uix+1=0x^2+u_i x+1=0. Thus all 8 roots of B3B_3 are real and negative.

For s=4,5s=4,5, the verification is analogous. The corresponding polynomials and their alternating evaluations are:

B4(x)=−(1+x)x5q4(−x−1/x),B_4(x)=-(1+x)x^5q_4(-x-1/x),
q4(u)=u5−242u4+11144u3−149008u2+673616u−920480,q_4(u)=u^5-242u^4+11144u^3-149008u^2+673616u-920480,

evaluated at u=2,3,5,12,38,187u=2,3,5,12,38,187 yields −,+,−,+,−,+-,+,-,+,-,+.

B5(x)=−x7q5(−x−1/x),B_5(x)=-x^7q_5(-x-1/x),
q5(u)=u7−1008u6+144780u5−5502080u4+75156720u3−428237568u2+1041166144u−892108800,\begin{aligned} q_5(u)={}&u^7-1008u^6+144780u^5-5502080u^4+75156720u^3\\ &-428237568u^2+1041166144u-892108800, \end{aligned}

evaluated at u=2,52,3,5,11,30,115,845u=2,\frac{5}{2},3,5,11,30,115,845 yields −,+,−,+,−,+,−,+-,+,-,+,-,+,-,+.

Since deg⁡q4=5\deg q_4=5 and deg⁡q5=7\deg q_5=7, the alternating signs certify that all their roots are real and lie in (2,∞)(2,\infty). Translating back via u=−x−1/xu=-x-1/x (and accounting for the extra factor 1+x1+x when s=4s=4) shows that all roots of B4B_4 and B5B_5 are real and negative. This completes the proof of the base cases.

By (31), BsB_s has only real negative roots for all s≥3s\geq3.

Acknowledgements

Acknowledgements. Guoce Xin was partially supported by the National Natural Science Foundation of China (No. 12571355).

Appendix A. An alternative proof via the Pólya–Schur criterion

This appendix provides an alternative proof of Theorem 1.1 that avoids the Brändén–Ottergren characterization. Instead, it relies solely on the classical Pólya–Schur criterion and Brändén’s bilinear preservation theorem. The argument follows the structure of § 3 but substitutes the algebraic core.

A Pochhammer multiplier sequence

Definition A.1. A sequence (λk)k≥0(\lambda_k)_{k\geq0} of real numbers is a multiplier sequence if the coefficient operator

Tλ(∑k=0Nckzk)=∑k=0NλkckzkT_\lambda\left(\sum_{k=0}^{N}c_kz^k\right)=\sum_{k=0}^{N}\lambda_kc_kz^k

maps every real-rooted polynomial to a real-rooted polynomial or to the zero polynomial.

We use the following finite form of the Pólya–Schur criterion [14]; see also [3 Theorem 3.14].

Theorem A.2 (Pólya–Schur criterion). Let (λk)k≥0(\lambda_k)_{k\geq0} be a sequence of nonnegative real numbers. It is a multiplier sequence if and only if, for every integer N≥0N\geq0, the polynomial

JN(z)=∑k=0N(Nk)λkzkJ_N(z)=\sum_{k=0}^{N}\binom{N}{k}\lambda_kz^k

has only real, nonpositive zeros or is identically zero.

Lemma A.3. For every a>0a>0, the sequence (1/(a)k)k≥0(1/(a)_k)_{k\geq0} is a positive multiplier sequence.

Proof. We verify Theorem A.2 using generalized Laguerre polynomials. For α>−1\alpha>-1 and N≥0N\geq0, define

LN(α)(u)=∑k=0N(−1)kk!(N+αN−k)uk.(32)L_N^{(\alpha)}(u)=\sum_{k=0}^{N}\frac{(-1)^k}{k!}\binom{N+\alpha}{N-k}u^k. \tag*{(32)}

Here (Nr)=Nr/r!\binom{N}{r}=N^{r}/r! for r≥0r\ge0. Formula (32) is the standard finite expansion of the generalized Laguerre polynomial [13 Eq. 18.5.12]. Its leading coefficient is (−1)N/N!(-1)^{N}/N!, so its degree is exactly NN.

To locate the zeros, we use the Rodrigues identity

LN(α)(u)=euu−αN!dNduN(e−uuN+α),u>0.(33)L_{N}^{(\alpha)}(u)=\frac{e^{u}u^{-\alpha}}{N!}\frac{d^{N}}{du^{N}}\left(e^{-u}u^{N+\alpha}\right),\qquad u>0. \tag*{(33)}

Indeed, differentiating the exponential factor kk times gives

dNduN(e−uuN+α)=e−u∑k=0N(Nk)(−1)k(N+α)N−kuα+k,\frac{d^{N}}{du^{N}}\left(e^{-u}u^{N+\alpha}\right)=e^{-u}\sum_{k=0}^{N}\binom{N}{k}(-1)^{k}(N+\alpha)_{N-k}u^{\alpha+k},

and multiplying by euu−α/N!e^{u}u^{-\alpha}/N! recovers (32).

Let p∈R[u]p\in\mathbb{R}[u] have degree less than NN. Integrating by parts NN times in (33) gives

∫0∞LN(α)(u)p(u)uαe−u du=(−1)NN!∫0∞p(N)(u)e−uuN+α du=0.(34)\int_{0}^{\infty}L_{N}^{(\alpha)}(u)p(u)u^{\alpha}e^{-u}\,du=\frac{(-1)^{N}}{N!}\int_{0}^{\infty}p^{(N)}(u)e^{-u}u^{N+\alpha}\,du=0. \tag*{(34)}

To justify this, put q(u)=e−uuN+αq(u)=e^{-u}u^{N+\alpha}. For 0≤r≤N−10\le r\le N-1, q(r)(u)=O(uN+α−r)q^{(r)}(u)=O(u^{N+\alpha-r}) as u→0+u\to0^{+}, and N+α−r≥α+1>0N+\alpha-r\ge\alpha+1>0. Thus all boundary terms at 00 vanish. At infinity, qq and its derivatives decay exponentially, so the other boundary terms also vanish. All integrals converge since the least power at 00 is uαu^{\alpha}, with α>−1\alpha>-1.

Assume N≥1N\ge1, and let u1,…,usu_{1},\ldots,u_{s} be the distinct zeros in (0,∞)(0,\infty) at which LN(α)L_{N}^{(\alpha)} changes sign. If s<Ns<N, take p(u)=∏ν=1s(u−uν)p(u)=\prod_{\nu=1}^{s}(u-u_{\nu}). The polynomial LN(α)pL_{N}^{(\alpha)}p then has constant sign on (0,∞)(0,\infty) apart from its zeros, and it is not identically zero. Since uαe−u>0u^{\alpha}e^{-u}>0 on this interval, its weighted integral cannot vanish. This contradicts (34), because deg⁡p=s<N\deg p=s<N. Consequently s≥Ns\ge N. As LN(α)L_{N}^{(\alpha)} has degree NN, it has exactly NN distinct positive zeros, each of multiplicity one.

Now set α=a−1\alpha=a-1. For 0≤k≤N0\le k\le N,

(N+a−1N−k)=(a)N(a)k(N−k)!.\binom{N+a-1}{N-k}=\frac{(a)_{N}}{(a)_{k}(N-k)!}.

It follows from (32) that

∑k=0N(Nk)zk(a)k=N!(a)NLN(a−1)(−z).(35)\sum_{k=0}^{N}\binom{N}{k}\frac{z^{k}}{(a)_{k}}=\frac{N!}{(a)_{N}}L_{N}^{(a-1)}(-z). \tag*{(35)}

For N≥1N\ge1 all zeros of this polynomial are strictly negative; for N=0N=0 it equals 1. Theorem A.2 proves the multiplier property, and positivity follows from a>0a>0. □

Interval preservation for the induced product

We use the interval form of Brändén’s bilinear preservation theorem [3 Theorem 3.11].

Theorem A.4 (Brändén). Let (λk)k≥0(\lambda_k)_{k\ge0} be a nonnegative multiplier sequence and let α<β\alpha<\beta be real numbers. If f∈R[y]f\in\mathbb{R}[y] is real-rooted and all zeros of g∈R[y]g\in\mathbb{R}[y] lie in [α,β][\alpha,\beta], then

∑k≥0λkk!(y−α)k(y−β)kf(k)(y)g(k)(y)\sum_{k\ge0}\frac{\lambda_k}{k!}(y-\alpha)^k(y-\beta)^k f^{(k)}(y)g^{(k)}(y)

is real-rooted or identically zero. (The sum is finite, and the closed interval permits zeros at its endpoints.)

Proposition A.5. Fix a>0a>0. Let f,g∈R[y]f,g\in\mathbb{R}[y] be nonzero polynomials with positive leading coefficients, of degrees m,nm,n, respectively. If all their zeros belong to [−1,0][-1,0], then f⋆agf\star_{a}g has degree m+nm+n, positive leading coefficient, and all its zeros in [−1,0][-1,0]. Moreover,

(f⋆ag)(0)=f(0)g(0),(f⋆ag)(−1)=f(−1)g(−1).(36)(f\star_{a}g)(0)=f(0)g(0),\qquad(f\star_{a}g)(-1)=f(-1)g(-1). \tag*{(36)}

Proof. By (14),

h(y):=(f⋆ag)(y)=∑k=0min⁡(m,n)[y(y+1)]kk!(a)kf(k)(y)g(k)(y).(37)h(y) := (f \star_a g)(y) = \sum_{k=0}^{\min(m,n)} \frac{[y(y+1)]^k}{k!(a)_k} f^{(k)}(y)g^{(k)}(y). \tag*{(37)}

Lemma A.3 allows us to apply Theorem A.4 with

λk=1(a)k,α=−1,β=0.\lambda_k = \frac{1}{(a)_k}, \qquad\alpha=-1, \qquad\beta=0.

Here ff is real-rooted, gg has all its zeros in [−1,0][-1,0], and (y−α)k(y−β)k=[y(y+1)]k(y-\alpha)^k(y-\beta)^k=[y(y+1)]^k. The theorem therefore shows that hh is real-rooted or identically zero.

Every summand in (37) has degree at most m+nm+n. Its coefficient of ym+ny^{m+n} is nonnegative, and their sum is

lc⁡(f)lc⁡(g)∑k=0min⁡(m,n)mk‾nk‾k!(a)k>0.(38)\operatorname{lc}(f)\operatorname{lc}(g)\sum_{k=0}^{\min(m,n)}\frac{m^{\underline{k}}n^{\underline{k}}}{k!(a)_k}>0. \tag*{(38)}

Thus hh is nonzero, has degree m+nm+n, and has positive leading coefficient.

It remains to locate its real zeros. Write f(y)=c∏ν=1m(y+ρν)f(y)=c\prod_{\nu=1}^{m}(y+\rho_\nu), where c>0c>0 and 0≤ρν≤10\leq\rho_\nu\leq1. For 0≤k≤m0\leq k\leq m and y>0y>0, each term in the product-rule expansion of f(k)(y)f^{(k)}(y) is positive. For y<−1y<-1, each such term has sign (−1)m−k(-1)^{m-k}. The same statements hold for gg.

For y>0y>0, all summands in (37) are positive. For y<−1y<-1, their signs are all (−1)m−k+n−k=(−1)m+n(-1)^{m-k+n-k}=(-1)^{m+n}, since y(y+1)>0y(y+1)>0. In either interval the summand with k=0k=0 is nonzero, so hh has no zeros there. Together with real-rootedness, this proves that all its zeros lie in [−1,0][-1,0]. The same argument applies when m=0m=0 or n=0n=0. Finally, evaluating (37) at 00 and −1-1 leaves only its k=0k=0 summand and proves (36). □

Real-rootedness of the Euler numerators

Alternative proof of Theorem 1.1. Set m=deg⁡Fm=\deg F and n=deg⁡Gn=\deg G. Multiplying FF and GG by nonzero real constants rescales their numerators and the numerator of FGFG. We may therefore assume that P=Ha,FP=H_{a,F} and Q=Ha,GQ=H_{a,G} have positive leading coefficients. Since all their zeros are nonpositive, their coefficients are nonnegative, and P(1),Q(1)>0P(1),Q(1)>0.

Set f=La(F)f=\mathcal{L}_a(F) and g=La(G)g=\mathcal{L}_a(G). By (15) and Lemma 3.1, these polynomials have degrees m,nm,n and all their zeros in [−1,0][-1,0]. Their leading coefficients are P(1)P(1) and Q(1)Q(1), hence positive. The definition of the induced product gives

La(FG)=f⋆ag.\mathcal{L}_a(FG)=f\star_a g.

Proposition A.5 shows that this polynomial has degree m+nm+n, positive leading coefficient, and all its zeros in [−1,0][-1,0]. Since deg⁡(FG)=m+n\deg(FG)=m+n, applying (8) to FGFG and then Lemma 3.1 proves that Ha,FGH_{a,FG} has only real, nonpositive zeros.

If both input numerators have only strictly negative zeros, then f(0)=P(0)≠0f(0)=P(0)\ne0 and g(0)=Q(0)≠0g(0)=Q(0)\ne0. By (36), (f⋆ag)(0)≠0(f\star_a g)(0)\ne0, so the output numerator has no zero at the origin. Its zeros are consequently strictly negative.

Finally, if the original input numerators already have nonnegative coefficients, the initial rescaling is unnecessary. The positive leading coefficient of f⋆agf\star_a g and Lemma 3.1 then show that Ha,FGH_{a,FG} has nonnegative coefficients.

References

School of Mathematical Sciences, Capital Normal University, Beijing 100048, P.R. China Email address: [email protected]

School of Mathematical Sciences, Capital Normal University, Beijing 100048, P.R. China Email address: [email protected]

References

  1. [1]C. A. Athanasiadis, Gamma-positivity in combinatorics and geometry, Sém. Lothar. Combin. 77 (2016–2018), Article B77i, 64 pp.arxiv.org/abs/1711.05983
  2. [2]M. Beck and S. Robins, Computing the Continuous Discretely: Integer-Point Enumeration in Polyhedra, second ed., Undergraduate Texts in Mathematics, Springer, New York, 2015. doi:10.1007/978-1-4939-2969-6.DOI
  3. [3]P. Brändén, On linear transformations preserving the Pólya frequency property, Trans. Amer. Math. Soc. 358 (2006), no. 8, 3697–3716. doi:10.1090/S0002-9947-06-03856-6.DOI
  4. [4]P. Brändén, Actions on permutations and unimodality of descent polynomials, European J. Combin. 29 (2008), no. 2, 514–531.DOI
  5. [5]P. Brändén, L. Ferroni, and K. Jochemko, Preservation of inequalities under Hadamard products, Trans. Amer. Math. Soc. (2026), published electronically September 22, 2026. doi:10.1090/tran/9683. Preprint: arXiv:2408.12386v2.DOI
  6. [6]P. Brändén and E. Ottergren, A characterization of multiplier sequences for generalized Laguerre bases, Constr. Approx. 39 (2014), no. 3, 585–596. doi:10.1007/s00365-013-9204-4.DOI
  7. [7]W. Bruns and T. Römer, h-Vectors of Gorenstein polytopes, J. Combin. Theory Ser. A 114 (2007), no. 1, 65–76. doi:10.1016/j.jcta.2006.03.003.DOI
  8. [8]E. Ehrhart, Sur les polyèdres rationnels homothétiques à n dimensions, C. R. Acad. Sci. Paris 254 (1962), 616–618.
  9. [9]S. R. Gal, Real root conjecture fails for five- and higher-dimensional spheres, Discrete Comput. Geom. 34 (2005), no. 2, 269–284.DOI
  10. [11]F. Liu, Ehrhart properties under operations on lattice polytopes, preprint (2026), arXiv:2610.06338v2.arxiv.org/abs/2610.06338
  11. [12]F. Liu, S. Tao, and G. Xin, Ehrhart Theory of the Join of Two Lattice Polytopes, preprint (2026), arXiv:2606.18794v1.arxiv.org/abs/2606.18794
  12. [13]National Institute of Standards and Technology, NIST Digital Library of Mathematical Functions, Chapter 18, Orthogonal polynomials, especially Equation 18.5.12. https://dlmf.nist.gov/18.5.E12, accessed October 27, 2026.
  13. [14]G. Pólya and I. Schur, Über zwei Arten von Faktorenfolgen in der Theorie der algebraischen Gleichungen, J. Reine Angew. Math. 144 (1914), 89–113. doi:10.1515/crll.1914.144.89.
  14. [15]J. E. Reeve, On the volume of lattice polyhedra, Proc. London Math. Soc. (3) 7 (1957), 378–395. doi:10.1112/plms/s3-7.1.378.DOI
  15. [16]R. P. Stanley, Decompositions of rational convex polytopes, Ann. Discrete Math. 6 (1980), 333–342. doi:10.1016/S0167-5060(08)70717-9.DOI
  16. [18]D. G. Wagner, Total positivity of Hadamard products, J. Math. Anal. Appl. 163 (1992), no. 2, 459–483. doi:10.1016/0022-247X(92)90261-B.DOI

Paper details

vlnc:W_wf8sgk2c
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents