Notes on the roots of Ehrhart polynomials (Q2385152): Difference between revisions

Given a lattice polytope \(P \subset {\mathbb R}^d\) (i.e., the vertices of \(P\) have integer coordinates), let \(L_P(t) := \# (tP \cap {\mathbb Z}^d)\) be its lattice-point enumerating function, where we consider \(t\) as a positive integer variable. Ehrhart's famous theorem asserts that \(L_P(t) = c_n t^n + c_{ n-1 } t^{ n-1 } + \cdots + c_0\) is a polynomial in \(t\) of degree \(n = \text{ dim} P\), where \(c_n\) is the volume of \(P\), \(c_{ n-1 }\) equals \(1 \over 2\) the sum of the volume of the facets of \(P\), and \(c_0 = 1\) (here we measure volume of a polytope \(Q\) relative to the lattice in the affine span of \(Q\)). The authors prove that in the above setting, \(\text{vol}\, P \geq {{ n I + 1 } \over { n! }}\), where \(I\) denotes the number of interior lattice points in \(P\). They show that this bound is best possible, for any \(I\). If we denote the roots of \(L_P\) by \(-\gamma_1, -\gamma_2, \dots, -\gamma_n\), then Ehrhart's theorem implies that \(\gamma_1 \gamma_2 \cdots \gamma_n = {1 \over {\text{vol\,} P }}\) and thus the authors' theorem implies the following bound for the geometric mean of the roots of \(L_P\): \[ \left( \gamma_1 \gamma_2 \cdots \gamma_n \right)^{ 1/n } \leq (n!)^{ 1/n } (nI+1)^{ -1/n } . \] If in addition \(P\) is centrally symmetric, the authors conjecture that \(\text{vol} \,P \geq {{ 2^{ n-1 } (I+1) } \over { n! }}\) and prove this conjecture for the case that \(P\) has \(2n\) vertices. The simplices \[ S_n(l) := \text{ conv} \left\{ e_1, e_2, \dots, e_n, -l \sum_{ j=1 }^n e_j \right\} \] (where \(e_j\) denotes the \(j\)th unit vector) play a central role in the paper; note that \(S_n(l)\) contains exactly \(l\) interior lattice points. The authors prove that all roots of \(L_{ S_n(1) }\) have real part \(- {1 \over 2}\) and that the root \(\alpha_n\) of \(L_{ S_n(1) }\) with maximal norm satisfies \(\left| \alpha + {1 \over 2} \right| = {{ n(n+2) } \over { 2 \pi }} + O(1)\), which proves that a recent bound established by Braun on the roots of Ehrhart polynomials is essentially tight. The authors also prove that if all roots of \(L_P\) have real part \(- {1 \over 2}\) then \(P\) is a reflexive polytope (up to unimodular transformation) of volume \(\leq 2^n\). The paper finishes by proving that the roots of \(L_P\) in the case \(\text{ dim} P = 3\) lie in the set \([-3, -1] \cup \left\{ a+bi : \, -1 \leq a < 1, \, a^2 + b^2 \leq 3 \right\}\) and that the bounds on \(a\) and \(a^2 + b^2\) are tight.