\documentclass[11pt,letterpaper,english]{amsart}
\usepackage[T1]{fontenc}
\usepackage{lmodern,microtype}
\usepackage{amsmath,amsthm,amssymb,mathtools}
\usepackage[dvipsnames]{xcolor}
\usepackage[margin=1.5in]{geometry}
\linespread{1.1}
\usepackage{enumerate}
\usepackage[backref=page,linktocpage]{hyperref}
\usepackage{cleveref}
\IfFileExists{Uboondoxo.fd}{\usepackage[cal=boondoxo,scr=euler]{mathalfa}}{}
\hypersetup{colorlinks=true,linkbordercolor=white,linkcolor=NavyBlue,
anchorcolor=black,citecolor=NavyBlue,filecolor=cyan,menucolor=NavyBlue,
runcolor=cyan,urlcolor=NavyBlue,pdftitle={Ehrhart positivity for positroids and polypositroids}}
\makeatletter
\@namedef{subjclassname@2020}{\textup{2020} Mathematics Subject Classification}
\makeatother
\newcommand{\Q}{\mathbb{Q}}
\newcommand{\R}{\mathbb{R}}
\newcommand{\Z}{\mathbb{Z}}
\DeclareMathOperator{\vol}{vol}
\newtheoremstyle{teoremas}{12pt}{13pt}{\itshape}{}{\bfseries}{}{.5em}{}
\theoremstyle{teoremas}
\newtheorem{theorem}{Theorem}[section]
\newtheorem{corollary}[theorem]{Corollary}
\newtheorem{lemma}[theorem]{Lemma}
\newtheorem{proposition}[theorem]{Proposition}
\newcommand{\repeatedtheoremname}{}
\newtheorem*{repeatedtheorem}{\repeatedtheoremname}
\newenvironment{restate}[1]{%
\renewcommand{\repeatedtheoremname}{Theorem~\hyperref[#1]{\ref*{#1}}}%
\begin{repeatedtheorem}%
}{\end{repeatedtheorem}}
\newtheoremstyle{definition}{12pt}{12pt}{}{}{\bfseries}{}{.5em}{}
\theoremstyle{definition}
\newtheorem{definition}[theorem]{Definition}
\newtheorem{conjecture}[theorem]{Conjecture}
\newtheorem{openproblem}[theorem]{Open Problem}
\newtheorem{problem}[theorem]{Problem}
\newtheorem{question}[theorem]{Question}
\newtheorem{example}[theorem]{Example}
\newtheorem{remark}[theorem]{Remark}
\crefname{theorem}{theorem}{theorems}
\Crefname{theorem}{Theorem}{Theorems}
\crefname{lemma}{lemma}{lemmas}
\Crefname{lemma}{Lemma}{Lemmas}
\crefname{proposition}{proposition}{propositions}
\Crefname{proposition}{Proposition}{Propositions}
\crefname{corollary}{corollary}{corollaries}
\Crefname{corollary}{Corollary}{Corollaries}
\title[Ehrhart positivity for polypositroids]{Ehrhart positivity for polypositroids}
% The byline is left open for review.
\author[L.~Ferroni]{Luis Ferroni}
\address{Dipartimento di Matematica, Universit\`a di Pisa, Pisa, Italy}
\email{luis.ferroni@unipi.it}
\keywords{Polypositroid, positroid, Ehrhart polynomial, generalized permutohedron,
alcoved polytope, permutation}
\subjclass[2020]{Primary 52B20; Secondary 05B35, 05A15, 52B40}
\date{}
\allowdisplaybreaks
\begin{document}
\raggedbottom
\begin{abstract}
We prove that all coefficients of the Ehrhart polynomial of an integral
polypositroid are strictly positive. In particular, this proves the
positroid conjecture of Ferroni, Jochemko and Schr\"oter. More generally, inclusion of integral polypositroids in a common cyclic
order implies coefficientwise comparison of their Ehrhart polynomials.
Hence the hypersimplex dominates every positroid of the same rank and size.
We express each coefficient as a sum of relative face volumes multiplied
by explicit nonnegative rational weights. The weights count linear
extensions and specialize, for hypersimplices, to Ferroni's formula
involving weighted Lah and Eulerian numbers. In the appendix we exhibit
two representable matroids whose quartic Ehrhart coefficients exceed
those of the hypersimplices of the same rank and size.
\end{abstract}
\maketitle
\textbf{Disclaimer:} This text, including the mathematical proofs, was produced almost entirely by generative AI tools. The author's
contributions were limited to minor changes in the exposition, the order of presentation, and the choice of what to include or omit. The author claims no mathematical credit beyond his earlier work, written with several coauthors, on which these models were trained.
\section{Introduction}\label{sec:intro}
For a nonempty lattice polytope $P\subseteq\R^n$, its Ehrhart polynomial
$E_P(t)$ counts the points of $tP\cap\Z^n$ at nonnegative integers $t$.
Write $e_j(P)=[t^j]E_P(t)$ and $E_A\succeq E_B$ when
$E_A-E_B$ has nonnegative coefficients in the monomial basis. We call the line spanned by $e_i-e_j$ a root direction. We use
lattice-relative volume, normalized so that a fundamental
parallelepiped has volume one.
Give $[n]=\{1,\ldots,n\}$ a cyclic order and put
$x(I)=\sum_{i\in I}x_i$. A \emph{polypositroid} is a polytope in a
hyperplane $x([n])=r$ satisfying the two conditions below
\cite{LP}.
\begin{enumerate}[\normalfont (i)]
\item Each edge is parallel to a difference $e_i-e_j$ of coordinate
vectors.
\item Within its sum hyperplane the polytope is cut out by upper bounds
on $x(I)$ for cyclic intervals $I$.
\end{enumerate}
A lower bound is also permitted, by taking the complement interval.
A polypositroid is \emph{integral} if all its vertices are lattice points.
Its dimension throughout the paper is its affine dimension.
\begin{theorem}\label{thm:main}
Let $P$ be an integral polypositroid. Then $e_j(P)>0$ for every
$0\le j\le\dim P$. Moreover, if $P\subseteq Q$ are integral polypositroids in the same cyclic order, then $E_P\preceq E_Q$.
\end{theorem}
A positroid is a matroid representable by a real matrix whose maximal
minors in the given column order are nonnegative. Its base polytope
has root edge directions and a cyclic interval description
\cite[Propositions 5.5--5.7]{ARW}. This yields the following consequence.
\begin{corollary}[Ferroni--Jochemko--Schr\"oter conjecture]
For every positroid $M$, the Ehrhart polynomial of its base polytope
has strictly positive coefficients through its degree.
\end{corollary}
Ferroni, Jochemko and Schr\"oter obtained the rank-two case and proposed
the general statement in \cite[Conjecture 6.3]{FJS}. For lattice path
matroids, Ferroni, Morales and Panova used the order polynomials of fence
posets \cite{FMPskew,FMP}. Li and Shao proved positivity for
series--parallel matroids and constructed a transversal matroid with a
negative coefficient \cite{LS}. Liu and Zhang proved the paving
positroid case \cite{LZ}.
The proof works with sums along intervals in a fixed order. A small
perturbation makes the intervals tight at any vertex nested or disjoint.
We can then count integer vectors in each tangent cone by imposing one
bound at a time. The issue is how the boundary terms from different
cones fit together. Theorem~\ref{thm:halfbox} groups the terms by their
chosen equalities. The remaining signed sum is the indicator of a
single face. Opening a prefix facet gives coefficientwise domination over that
facet. We pass from facets to arbitrary inclusions by interpolating the
two systems of cyclic bounds and relaxing one bound at a time. Strict
positivity then follows by comparing $P$ with one of its faces in every
dimension.
Proposition~\ref{prop:beta} makes the face weights explicit. A weight
is a product of factors $a!b!/(a+b+1)!$, obtained by placing one label
between $a$ preceding labels and $b$ following labels. We use this
description in \S\ref{sec:comparison} to recover the weighted Lah numbers
for hypersimplices. Appendix~\ref{app:upper} disproves the upper bound
in Ferroni's two-sided conjecture for matroids \cite[Conjecture~1.5]{Fmin}.
The question for positroids, posed explicitly in \S\ref{sec:upper}, has
an affirmative answer by inclusion in the hypersimplex.
\section{Two elementary polyhedral identities}\label{sec:prelim}
\begin{lemma}[Closed tangent cones]\label{lem:bg}
Let $R$ be a nonempty polytope, and work in its affine hull. For a nonempty
face $F$ let $T_FR$ be the intersection of the closed facet halfspaces
whose bounding facets contain $F$. The empty intersection, corresponding
to $F=R$, is the whole affine hull. Then
\[
\sum_{\varnothing\ne F\le R}(-1)^{\dim F}\mathbf 1_{T_FR}
=\mathbf 1_R .
\]
\end{lemma}
\begin{proof}
Fix a point $z$ in the affine hull. Let $U$ be the union of the facets
whose halfspaces $z$ strictly violates, together with all their faces.
The terms missing from the complete face sum at $z$ are exactly the
nonempty faces of $U$. The complete face sum is $1$, by Euler's formula.
If $z\in R$, then $U$ is empty.
Suppose $z\notin R$. Strictly separate $z$ from $R$ by a hyperplane
$H$. Each ray from $z$ meeting $R$ has a unique first point $p$ of $R$.
Some facet through $p$ is strictly violated by $z$: otherwise a short
step from $p$ towards $z$ would satisfy every facet inequality.
Conversely, each point of a strictly violated facet is a first point.
Central projection from $z$ therefore gives a continuous bijection from
$U$ onto the convex section $H\cap\{z+\lambda(y-z):y\in R,\lambda\ge0\}$.
The domain is compact, so this is a homeomorphism. In particular $U$
is contractible, and its alternating face sum is $1$. Here we use the
elementary Euler formula for a finite polytopal complex: the alternating
sum of its nonempty cells is its Euler characteristic, as follows by
triangulating its cells and taking the alternating sum of the simplicial
chain ranks. Subtraction gives zero at $z$. The use of strictly violated
facets includes points $z$ lying on any of the other facet hyperplanes.
\end{proof}
\begin{lemma}[Two lattice summands]\label{lem:mixed}
If $P$ and $R$ are lattice polytopes in the same lattice, then
\[
B(a,b)=\#\bigl((aP+bR)\cap\mathbb Z^n\bigr)
\]
agrees with a single polynomial for all nonnegative integers $a,b$,
including the axes.
\end{lemma}
\begin{proof}
Take the cone spanned by $(v,1,0)$ for vertices $v$ of $P$ and
$(w,0,1)$ for vertices $w$ of $R$. Its lattice points of bidegree
$(a,b)$ are precisely the points being counted. Triangulate the cone
using its generating rays; for instance, triangulate the section
where the last two coordinates sum to one by successively pulling
vertices. Pulling a vertex means coning it to the triangulations of the
facets missing it, with the same vertex order used recursively on the
facets.
Choose a vector $h$ in the relative interior of the cone outside all spans
of facets of the triangulation. Assign $x$ to the unique maximal
simplicial cone containing $x+\epsilon h$ in its relative interior for
all sufficiently small $\epsilon>0$. This partitions the closed cone.
In the ray coordinates of a maximal cone, the coordinates of $x$
must be nonnegative, and strictly positive wherever the coordinate
of $h$ is negative. Suppose this cone has $p$ rays of the first
bidegree and $q$ of the second. Both are positive. In each group at
least one coordinate of $h$ is positive, since their group sums are
the two positive degree coordinates of $h$.
Every lattice point in the assigned part of this cone has a unique
expression as a nonnegative integer combination of its ray generators
plus a lattice point of their fundamental parallelepiped. Use $[0,1)$
for a closed ray coordinate and $(0,1]$ for an open one. In each group
at least one interval is $[0,1)$, so every such parallelepiped point
has integral bidegree $(i,j)$ with $0\le i
0}$,
then, for $j\ge 1$,
\begin{equation}\label{eq:asym}
e_j(R_L)=(DL)^j e_j(P)+O(L^{j-1}).
\end{equation}
Indeed, substitute $a=D(L-L_0)t$ and $b=t$ in $B(a,b)$.
Only terms $a^u b^{j-u}$ contribute to $[t^j]$, and the coefficient
of $a^j b^0$ is $e_j(P)$ by evaluation at $b=0$.
\section{A counting rule for a bounded sum}\label{sec:transfer}
Fix an integer $t\ge1$. We record the sum of integer coordinates by the
power of $z$ in their generating polynomial. Multiplying two such
polynomials concatenates the two lists of coordinates. We need a rule
for retaining totals in a closed interval that also records what happens
at either endpoint.
Use two symbols $S,D$. Their generating functions are $z^t$ and
$(1-z)^{-1}$, respectively. Thus $S$ adds $t$ to a total and $D$
allows any nonnegative integer contribution. We use finite linear
combinations of $S^j D^p$, $j,p\ge0$, with coefficients in $\Q[t]$.
These combinations also have a continuous interpretation, in which
each step either stays at its current position or moves to the right.
Choose a one-periodic function $m\colon\R\to[0,1]$ that is piecewise
affine on each period and takes uniformly distributed values there.
Point values at the finitely many breakpoints can be assigned arbitrarily,
subject to periodicity. Set
\[
(Sg)(u)=g(u+1),\qquad
(Dg)(u)=m(u)g(u)+t\int_{0}^{\infty} g(u+s)\,ds
\]
on piecewise continuous functions vanishing sufficiently far to the right.
Periodicity makes $S$ and $D$ commute. A product acts by taking its
steps in order, starting at $u$. Its expansion consists of atoms at
integer displacements and a density $K_O(u,w;m)$ at final positions
$w>u$. For example the box step
\begin{equation}\label{eq:box-step}
A_c=D+S^c(1-D),\qquad c\in\Z_{>0},
\end{equation}
has generating polynomial $1+z+\cdots+z^{ct}$. Its continuous rule
selects displacement $0$ with weight $m(u)$, displacement $c$ with
weight $1-m(u)$, or $s\in(0,c)$ with measure $t\,ds$.
\begin{lemma}[Closed truncation]\label{lem:cap}
For any such combination $O$ and integer $k\ge0$ there are combinations
$U_kO,V_kO$ which retain exactly the powers of $z$ of degrees at most
$kt$ and at least $kt$, respectively. Their continuous rules retain the
old atoms and density on the corresponding closed displacement range.
Each also adds one atom at displacement $k$. Putting $x=m(u)$, its
additional weight is respectively
\begin{equation}\label{eq:limits}
(1-x)\lim_{w\downarrow k} \frac{K_O(0,w;m_x^-)}{t},
\qquad x\lim_{w\uparrow k} \frac{K_O(0,w;m_x^-)}{t},
\end{equation}
where $m_x^-(h)=x$ at integers $h$ and
$m_x^-(s)=1-\{s\}$ elsewhere. The density at negative displacements is zero.
\end{lemma}
\begin{proof}
It suffices to treat $S^j D^p$ for all $p$ at once. Introduce a formal
variable $\lambda$ and write
\[
R=\sum_{p\ge0}\lambda^p D^p,\qquad
b=(1-\lambda)^{-1},\qquad a(u)=(1-\lambda m(u))^{-1}.
\]
An identity for these series gives a finite identity in $D$ upon taking
$[\lambda^p]$. On the discrete side
\[
R(z)=\frac{1-z}{1-\lambda-z},\qquad
[z^0]R(z)=b,\qquad [z^h]R(z)=\lambda b^{h+1}\quad(h\ge1).
\]
In particular the tail is geometric. For $jk$ they are $0$ and $S^jR$. The coefficients
$[\lambda^h]b^C$ are polynomials in $t$.
To read the same identities continuously, expand a word of $p$ steps
by its sequence of strictly increasing positions. At a visited
position $v$, an arbitrary string of stays has generating series $a(v)$.
If there are $q\ge1$ moves from $u$ to $w$, the $q-1$ intermediate
positions are ordered. Integrating the product of their $a$-weights
gives $(\int_u^w a(s)\,ds)^{q-1}/(q-1)!$. Summing over $q$ gives
the atom $a(u)$ and the density
\begin{equation}\label{eq:res}
K_R(u,w;m)=\lambda t\,a(u)a(w)
\exp\!\left(\lambda t\int_u^w a(s)\,ds\right),\qquad w>u .
\end{equation}
This computation is coefficientwise and retains the specified point
values at the two ends of the ordered sequence.
A complete period multiplies the exponential by $b^t$, since
$\lambda\int_0^1 a(s)\,ds=\lambda\int_0^1(1-\lambda y)^{-1}dy=\log b$.
Thus, above displacement $k$, the density of $S^jR$ agrees exactly with
that of $S^kb^CR$. Equation~\eqref{eq:cut-run} removes it on the upper
truncation and retains it on the lower truncation. Below $k$ the terms
shifted by $k$ contribute nothing. For $jk$ both limits vanish. Taking coefficients completes the proof.
\end{proof}
\begin{lemma}[A final slack coordinate]\label{lem:mass}
Fix $x\in[0,1]$ and evaluate the step rule from the origin with the
profile $m_x^-$. If the discrete specialization
$F_t(z)$ of $O$ is a polynomial and the step rule has bounded
displacement support, then $(O\mathbf1)(0)=F_t(1)$.
\end{lemma}
\begin{proof}
Append one unrestricted step, $H=OD$. Choose $k$ larger than every
displacement in the support of $O$, every power of $S$ occurring in $H$,
and $\deg(F_t)/t$. Then $[z^{kt}]H(z)=F_t(1)$, since the last coordinate
supplies the slack. The formulas in the proof of Lemma~\ref{lem:cap}
give the symbol identity
\[
(U_k+V_k-\mathrm{Id})H=S^k[z^{kt}]H(z)
\]
by coefficient extraction from $S^jR$. In the step rule there is no
old atom at $k$. Near $k$ the density of $H$ is $t(O\mathbf1)(0)$:
the last step must move, and it can reach the final position from
every endpoint of $O$. The two extra atom weights in \eqref{eq:limits}
therefore sum to $(1-x+x)(O\mathbf1)(0)$, since the reset at the
origin uses the same profile $m_x^-$. Thus the step rule of the
left-hand side is a single atom at $k$ of weight $(O\mathbf1)(0)$,
while that of the right-hand side is the atom $F_t(1)$ at $k$.
Comparing these atoms proves the identity.
\end{proof}
\section{A face formula for interval inequalities}\label{sec:local}
In this section intervals belong to one fixed linear order on $[d]$.
A family of intervals is \emph{laminar} if each pair is disjoint or nested.
A facet carries its interval $I$, the direction of its inequality
and its integer threshold $k_I$.
\begin{theorem}\label{thm:halfbox}
Let $Q\subseteq\R^d$ be a full-dimensional simple lattice polytope given
by interval bounds. Assume the facet intervals tight at each vertex
form a laminar family. There are weights $w_G\ge0$ on its nonempty faces
such that
\[
E_Q(t)=\sum_{\varnothing\ne G\le Q} w_G t^{\dim G}.
\]
Moreover $E_Q\succeq E_H$ whenever $H$ is a facet defined by a bound
on a prefix $[1,b]$.
\end{theorem}
\begin{proof}
Take an integral box strictly containing $Q$ and translate its
endpoints to $0,c_i$ in each coordinate. All facet thresholds of the
translated polytope are now positive. Fix $x\in[0,1]$ and use the
decreasing profile $m_x^-$ of Lemma~\ref{lem:cap}, starting at zero.
The box step in coordinate $i$ is $A_{c_i}$.
For a nonempty face $F$, let $B(F)$ be its containing facets. The
tangent polyhedron $T_F Q$ imposes just these bounds.
They are laminar, being tight at any vertex of $F$, and distinct
members of $B(F)$ have distinct intervals since $Q$ is
full-dimensional. So the coordinate enumerator in the clipped tangent
polyhedron has a symbol $O_F$. Read the coordinates from left to right,
multiplying the factors for each interval and applying its closed
truncation, proceeding from smaller intervals to larger ones.
This records each lattice vector once by the power of $z$ equal to
its coordinate sum. By Lemmas~\ref{lem:bg} and~\ref{lem:mass},
\begin{equation}\label{eq:bg}
E_Q(t)=\sum_{\varnothing\ne F\le Q}(-1)^{\dim F}(O_F\mathbf1)(0).
\end{equation}
\paragraph{Expanding a summand.}
At each bound Lemma~\ref{lem:cap} gives the restricted old measure plus
its extra boundary atom. Fix the set $S\subseteq B(F)$ where this atom
is chosen; we call its intervals \emph{pinned}. The construction below
also fixes whether each coordinate step uses an endpoint or its
continuous part. Put $\sigma_I=1$ for an upper pin and $-1$ for a lower
pin. In the density for $I$ evaluate at $k_I+\sigma_I\epsilon_I$,
with $0<\epsilon_I<1/2$. Its displacement in its parent is still
the exact integer $k_I$.
Order the pins by containment and put a formal root above them, even if
$[d]$ is pinned. The children of a node are its maximal proper pins.
The coordinates outside those children are its \emph{direct leaves}.
A pinned singleton therefore has one direct leaf. Writing $z_i$ for
the actual coordinate displacements, at each pin $I$ we have
\begin{equation}\label{eq:pivot}
\sum_{i\text{ direct in }I}z_i
=k_I+\sigma_I\epsilon_I-\sum_{J\text{ child of }I}k_J .
\end{equation}
The endpoint leaves have integer displacements. A nonzero term
therefore requires a continuous direct leaf at each pin. Use the
last such leaf to solve \eqref{eq:pivot}, with Jacobian one.
The pivot leaves at distinct pins are distinct. If $p$ leaves were
continuous, the resulting integral is over $p-|S|$ variables and
has the single factor $t^{p-|S|}$. Indeed every continuous step
supplies $t$ and every boundary density in \eqref{eq:limits} divides by $t$.
An unpinned bound just tests its interval sum. At a pin whose parent
profile has value $v$ at its starting position, the rule gives weight
$v$ for a lower pin and $1-v$ for an upper pin. Inside it, start
at zero with profile $m_v^-$; outside it resume after its integer
displacement $k_I$. Thus all profiles, starting positions and weights
are determined by $S$ and the leaf choices. All of these weights are nonnegative.
\paragraph{One parameter suffices.}
These integration rules follow by convolution of successive steps.
In a nested boundary density the inner limit is initially taken first.
To evaluate all limits, in each node replace its free continuous direct
leaf variables by their successive partial sums $q_1,q_2,\ldots$,
starting with $q_0=0$. This is triangular with determinant one. At a
pin the last partial sum has already been fixed at an integer plus
$\sigma_I\epsilon_I$ by \eqref{eq:pivot}; at the root all nonempty
partial sums are free. Child pins and endpoint leaves contribute
only integers to apparent displacements. Consequently a profile
argument is an integer plus one partial sum, and each unpinned bound
in its nearest enclosing node compares a difference $q_b-q_a$ with
an integer. The range indicators of pivot leaves have this same form.
Outside the null hyperplanes on which a free partial sum, or a
difference of distinct free partial sums, is integral, every
nonconstant comparison is eventually fixed. A constant comparison
either has no error or tests the sign of its own node's
$\sigma_I\epsilon_I$. At an exactly integral start the profile
inherits the value from its parent; at a pinned last partial sum it
approaches its integer from the indicated side. These rules retain
the contribution of a purely atomic subproduct. If an enclosing
density is evaluated at a noninteger total, a subproduct containing all
its direct continuous steps likewise has noninteger total; a proper
nonempty subset of these steps has integral total only on a null
hyperplane. Thus no choice of the point value of a continuous density
at an integer is required. The free variables can be confined to a
fixed bounded box and the integrands, after their power of $t$ is
removed, are bounded by one. Each weight and indicator depends on at
most one $\epsilon_I$ and converges, outside a null set, as that
$\epsilon_I\downarrow0$. Bounded convergence therefore shows that the
integrals converge along every joint approach of the errors to zero.
The joint limit thus exists and agrees with the iterated limit in the
prescribed nested order; in particular we may take all $\epsilon_I=\epsilon
\downarrow0$.
\paragraph{Summing over faces.}
For the fixed choices above the remaining faces in \eqref{eq:bg} are
precisely the nonempty faces of $G=Q\cap\{z(I)=k_I:I\in S\}$,
with $G=Q$ for $S=\varnothing$. Put
\[
\delta_A=\epsilon\sum_{I\in S,\ I\subseteq A}\sigma_I
\]
for every facet interval $A$. Summing \eqref{eq:pivot} gives
$z(I)=k_I+\delta_I$ for a pin $I$. An unpinned active interval $A$
sees each contained pin at its external displacement, so it tests
its bound at $k_A+\delta_A$ in the actual coordinates $z$.
These are the same shifted thresholds for every $F\le G$.
Shift all the facet thresholds in this way, taking equality at $S$,
and denote the resulting face by $G_\epsilon$. For sufficiently small
$\epsilon$ it has the labeled face lattice of $G$. Indeed each vertex
of the simple polytope $Q$ is the solution of a nonsingular active
system with the other inequalities strict. Its feasible systems
persist, and any other candidate independent system remains infeasible
for sufficiently small changes. This preserves the full labeled face
lattice and its face at $S$. In the affine space of the shifted pinned
equalities the remaining tests for $F$ are exactly the tangent
inequalities of $G_\epsilon$ at the corresponding face. The signed
sum of those tests over $F\le G$ is therefore $\mathbf1_{G_\epsilon}(z)$
by Lemma~\ref{lem:bg}, with the common weights unchanged.
Each grouped integral is now nonnegative, and so is its limit.
Moreover, every $G_\epsilon$ remains strictly inside the auxiliary box.
A leaf at a box endpoint then contributes zero after grouping.
Only the all-continuous choice remains and gives degree
$d-|S|=\dim G$. Every face of a simple polytope has exactly one such
set $S$, its containing facets. The surviving integral is $w_G$.
For the last assertion, make the bound defining $H$ strict precisely in
the tangent cones at $F\le H$. The signed indicator sum becomes
$\mathbf1_Q-\mathbf1_H$: the pieces removed on the supporting hyperplane
are exactly the tangent cones of $H$, with the same face signs.
For a strict upper cutoff use $O-V_kO$. By Lemma~\ref{lem:cap} this
restricts the old measure to displacements $0$, every point of $LP$ is in its relative
interior in the sum hyperplane.
Suppose two linear intervals overlap without either containing the other.
Then the corresponding cyclic bipartitions cross, regardless of which
side contains $n$. For the original intervals $A,B$, all four regions of
their Venn diagram are nonempty. Cutting the circle outside their union
shows that their intersection and union are both nonempty proper cyclic
intervals. Moreover,
\[
q_A+q_B-q_{A\cap B}-q_{A\cup B}
=2|A\setminus B|\,|B\setminus A|>0.
\]
Adding $La$ preserves this inequality by submodularity, and the four
small errors cannot exhaust its slack. Simultaneous equality in the
bounds on $A,B$ would now give
$x(A)+x(B)>x(A\cap B)+x(A\cup B)$, an impossibility. Thus the
intervals of coactive bounds in the chart form a laminar family.
Any $n-1$ independent bound normals determine a candidate vertex
affine in $L$. For any other bound, its slack at this candidate is
affine in $L$. The independent perturbations can avoid that slack
vanishing identically: the offset of the other bound has nonzero
coefficient and does not enter the candidate. There are finitely
many candidates and bounds. For all sufficiently large $L$ the
feasible bases thus stabilize and every vertex is simple.
Take a common denominator $D$ for the $c_I$ and put
$Q_L=D\widetilde Q_L$. In the cumulative coordinates
$y_i=x_1+\cdots+x_i$, $y_0=0$, a row is a signed difference $y_b-y_a$.
Independent difference rows form a forest on the index set; a cycle
gives a dependence, and a forest is independent by leaf removal.
In a full system their graph is a tree rooted at $0$, so integer
right-hand sides give integer coordinates by summing along paths.
Consequently $Q_L$ is integral.
For $L$ in the stable range, write each feasible vertex
as $DLv+w$. On dividing its equations and inequalities by $DL$ and
letting $L$ tend to infinity, one sees that $v\in P$ and the
same basis inequalities are tight at $v$. The cone of maximizing
functionals at $DLv+w$ consists of the nonnegative combinations of
the outward normals of these rows. This follows by duality from
the feasible infinitesimal directions at the vertex. Each such
functional is also maximized on $P$ at $v$. Take $L_0$ in the
stable range. Equality of support functions gives
\begin{equation}\label{eq:stable}
Q_L=D(L-L_0)P+Q_{L_0}\qquad(L\ge L_0)
\end{equation}
in the projected chart. In particular \eqref{eq:asym} and the face
formula give $e_j(P)\ge0$. Below we compare nested polytopes by using compatible perturbations of both.
\section{Monotonicity in a fixed cyclic order}\label{sec:inclusion}
The following is the only face comparison needed before taking limits.
Write $d=n-1$ for the dimension of the sum hyperplane.
\begin{lemma}\label{lem:truefacet}
If an integral polypositroid $R$ has dimension $d$ and $H$ is a facet
of $R$, then $E_R\succeq E_H$.
\end{lemma}
\begin{proof}
Choose a supporting cyclic bound $x(A)\le a_A$ of $H$. Rotate the
ground set so that $A$ starts at $1$, and omit its predecessor as the
last coordinate. The interval is now a prefix. For the deformation of
$R$ in \eqref{eq:stable}, let $H_L$ be the maximizing face in direction
$\mathbf1_A$. Faces exposed by a fixed functional add under Minkowski
addition, so $H_L=D(L-L_0)H+H_{L_0}$. For $L>L_0$ this face has
dimension at least $d-1$. The polytope $Q_L$ is full-dimensional, so
$H_L$ is a facet. Its supporting bound is the prefix bound on $A$:
in the full chart no other raw upper bound has the same outward normal.
By Theorem~\ref{thm:halfbox}, $E_{Q_L}\succeq E_{H_L}$.
Since $H$ and $H_{L_0}$ are lattice polytopes, \eqref{eq:asym} applies
to both sides; divide in degree $j\ge1$ by $(DL)^j$ and let $L\to\infty$.
Their constant coefficients agree.
\end{proof}
\begin{lemma}\label{lem:onerelax}
Suppose $K'\subsetneq K$ are full-dimensional integral polypositroids
in the same sum hyperplane and cyclic order, and that
$K'=K\cap\{x(A)\le a\}$ for one integral cyclic interval bound.
Then $E_K\succeq E_{K'}$.
\end{lemma}
\begin{proof}
The cutting plane meets the interior of $K$. Let $R$ be the other closed
half and $H$ their common facet. Every edge of $R$ in $H$ is an edge
of $H$, hence an edge of the polypositroid $K'$. At a relative interior
point of any other edge the cut is inactive, so its direction is that
of an edge of $K$. Thus $R$ has root edge directions. Its description
is still by integral cyclic bounds, since the new lower bound can
be written on the complementary interval. The cumulative-coordinate
tree argument of \S\ref{sec:deform} gives integrality.
Lemma~\ref{lem:truefacet} and inclusion--exclusion give
$E_K-E_{K'}=E_R-E_H\succeq0$.
\end{proof}
\begin{proof}[Proof of Theorem~\ref{thm:main}]
Suppose $P\subseteq Q$. Their coordinate sums agree and their tight
bounds satisfy $a_I(P)\le a_I(Q)$ for every cyclic interval $I$.
Use the same perturbation $c_I=q_I+\epsilon_I$ for both endpoints,
avoiding the finite exceptional systems in \S\ref{sec:deform} for
either polytope. Enlarge the common denominator $D$ to at least $10$.
For fixed $L$ the two systems have integral right-hand sides
\[
b^-_I=D(La_I(P)+c_I),\qquad b^+_I=D(La_I(Q)+c_I).
\]
Along the segment between $b^-$ and $b^+$, each crossing slack
$b_A+b_B-b_{A\cap B}-b_{A\cup B}$ is at least $(2-4/10)D$.
Indeed this holds at both ends by submodularity and the calculation
in \S\ref{sec:deform}.
Follow the componentwise floors of the segment from $b^-$ to $b^+$.
Resolve simultaneous jumps by increasing one entry at a time.
Every intermediate integer vector differs by at most one in each
entry from a point on the segment. Rounding can decrease any crossing
slack by at most four, so it remains strictly positive. The
intermediate polytopes contain the lower endpoint polytope and lie
in the upper one, so they are bounded and full-dimensional.
We check that they all have root edge directions.
At a relative interior point of an edge take $n-2$ independent chart
rows tight there. Positive crossing slack ensures that, after complementing the raw intervals containing
$n$, their supports are laminar. In the homogeneous equations for edge
directions, subtract from each selected interval its maximal proper
selected subintervals, and from $[n]$ all maximal selected intervals.
The resulting $n-1$ equations say that the coordinate displacement
sums to zero on each of $n-1$ disjoint blocks partitioning $[n]$.
None is empty, by independence; the last one contains $n$. Exactly
one block has size two, proving that the edge has a root direction.
Integrality follows from the difference rows as before. This proves
that the intermediate polytopes are integral polypositroids.
Successive distinct polytopes now satisfy Lemma~\ref{lem:onerelax}.
Chaining its inequalities compares the Ehrhart polynomials at the two
deformed endpoints. Dividing the degree-$j$ inequality by $(DL)^j$
and applying \eqref{eq:asym} proves $E_P\preceq E_Q$, also when
one or both original polytopes have smaller dimension. For $n=1$
both are the same point.
Finally, every nonempty face $F$ of an integral polypositroid is an
integral polypositroid for the induced cyclic interval description:
turn the defining tight bounds into equalities, and observe that its
edges are edges of the original polytope. Applying the inclusion
inequality to $F\subseteq P$ yields, in each degree,
\begin{equation}\label{eq:positive-face-bound}
e_j(P)\ \ge\ \max_{\substack{F\le P\\\dim F=j}}\vol_j(F)>0
\qquad(0\le j\le\dim P).
\end{equation}
Here a point has volume one.
\end{proof}
\section{Face volumes and linear extensions}\label{sec:beta}
For the closed polytope we take $m(u)=1-\{u\}$ off the integers and
$m(h)=1$ at integer $h$. For simple $Q$ in Theorem~\ref{thm:halfbox}, its own dilates $LQ$
already have a constant normal fan. The same computation covers the
possibly nonsimple $P$ using the deformations in
\eqref{eq:stable}. Put $d=n-1$ and let $G_L$ be a stable $j$-face of
$Q_L$. Its $d-j$ incident facets give a laminar set $S$ of signed
intervals. Choose a functional in the relative interior of their normal
cone and write $G$ for its maximizing face on $P$. From \eqref{eq:stable},
\[
G_L=D(L-L_0)G+G_{L_0},\qquad \dim G\le j .
\]
Make the containment forest of $S$, with an additional formal root. In
each node $H$ the \emph{direct leaves} are the individual coordinates
outside all its selected children. Each real node has at least one direct
leaf: otherwise its interval normal would be a sum of child normals.
For its $h_H$ direct leaves, in order, let $q_{H,a}$ be the sum of the
first $a$ coordinate values. The free prefixes are
$q_{H,1},\ldots,q_{H,h_H-1}$ at a real node and
$q_{H,1},\ldots,q_{H,h_H}$ at the formal root. There are $j$ such
variables. They give lattice coordinates in the pinned affine space:
differences recover the direct leaves, and each last leaf at a real node
then follows from the pinned sum minus its child sums. Thus arbitrarily
assigned integer free prefixes give integral leaf coordinates, and
conversely. This chart has the same volume form as in the face formula.
Write $u_{H,a}=\{q_{H,a}\}$ for each free prefix.
Here is a rule using only the marked intervals of $S$. Start with
$v_{\mathrm{root}}=1$. If $I$ is a selected child of $H$, let $a$ be
the number of direct leaves of $H$ preceding $I$. Its incoming value is
\[
v_I=\begin{cases}
v_H,& a=0,\\
1,& a=h_H>0,\ H\text{ a real upper pin},\\
0,& a=h_H>0,\ H\text{ a real lower pin},\\
1-u_{H,a},&\text{otherwise}.
\end{cases}
\]
The factor for $I$ is $v_I$ for a lower bound, $1-v_I$ for an upper
bound. If a factor is identically zero, put $\beta(S)=0$. Otherwise
the product of all the factors has the form
\[
W_S(u)=\prod_{k=1}^j u_k^{a_k}(1-u_k)^{b_k},\qquad
\beta(S)=\prod_{k=1}^j\frac{a_k!\,b_k!}{(a_k+b_k+1)!}.
\]
Empty products equal one. In the zero case set $W_S=0$.
\begin{proposition}\label{prop:beta}
For $0\le j\le \dim P$, one has
\[
e_j(P)=\sum_{S:\,\dim G=j}\beta(S)\,\vol_j(G).
\]
\end{proposition}
\begin{proof}
In the grouped integral for $G_L$ the enclosing box is strict, so its
only surviving leaf pattern integrates every leaf. The starting position
of $I$ inside its nearest selected parent $H$ is $q_{H,a}$ plus the
integer loads of preceding child pins. If $a=0$ its profile has the
value inherited from $H$. If the last direct leaf of a real parent has
already been passed, the total prefix is an integer plus
$\sigma_H\epsilon_H$ and its sawtooth limit is one on the right and
zero on the left. All other nonzero prefixes are free. Outside a null
set these give $1-\{q_{H,a}\}$. This proves the rule for $v_I$.
The indicator of the shifted face converges to the indicator of $G_L$
almost everywhere in its free coordinates, since for each fixed $L$
the labeled faces are stable and their vertices converge. For $j=0$
the unique point is retained. The contribution of $S$ therefore is
\[
\int_{\pi_S G_L} W_S(\{q\})\,dq ,
\]
where $\pi_S$ is the projection to the free prefixes. An integral translation of the enclosing box shifts these prefixes by integers and leaves the integrand unchanged.
On every integer unit cube the integrand has integral $\beta(S)$, by
$\int_0^1 u^a(1-u)^b\,du=a!b!/(a+b+1)!$.
If $\dim G=j>0$, then $\pi_S G_L$ has volume
$(DL)^j\vol_j(G)+O(L^{j-1})$. Its bounded number of facet hyperplanes
meet $O(L^{j-1})$ unit cubes: the entire face has diameter $O(L)$,
and cubes meeting a hyperplane lie in a slab of bounded thickness.
Thus its integral, divided by $(DL)^j$, tends to
$\beta(S)\vol_j(G)$. If $\dim G0$, the integer $d!\,\beta(S)$ counts linear extensions. Attach one label
to every free prefix and one label to every pin. For a factor $u_k$
require the pin label to precede prefix label $k$; for $1-u_k$ require
it to follow. A factor one gives an isolated label, and a zero factor
makes the count zero. There are $d$ labels in total. Conditional on
the prefix positions in $(0,1)$, independent uniform positions for
the pin labels satisfy the comparisons with probability $W_S(u)$.
All $d!$ relative orders are equally likely. This is also a direct
order count: each nontrivial connected component consists of prefix
label $k$, preceded by $a_k$ pin labels and followed by $b_k$ pin labels.
For a contributing face $G$ the number $j!\vol_j(G)$ counts unit
alcoves. Indeed in the original cumulative coordinates choose an
independent set of tight difference equalities spanning the annihilator
of the direction space of $G$, for instance those indexed by $S$.
Their graph on the cumulative coordinate indices is a forest. Choose
one representative for every connected component, fixing the component
of the origin at zero. All other coordinates differ from their
representative by integer constants. This also identifies the affine
lattice with $\Z^j$; every remaining interval bound is a difference
bound with an integral constant. Decomposing unit cubes according to
the strict order of the $j$ fractional parts partitions $G$ into unit
simplices, each of volume $1/j!$, up to their boundaries.
Consequently $d!j!e_j(P)$ counts the triples formed by a contributing
$S$, a unit alcove of $G$, and a linear extension of the indicated
comparisons. For $j=0$ the point is counted as one alcove.
\begin{example}
Take $0\le z_1,z_2\le2$ and $1\le z_1+z_2\le3$, the hexagonal chart
of the slice $x([3])=3$ of $[0,2]^3$. Every edge has lattice length
one. In the order lower and upper $z_1$, lower and upper $z_2$, lower
and upper total, their weights are $1,0,\frac12,\frac12,1,0$.
For example, at a bound on $z_2$ its label is compared with the free
prefix label from $z_1$, and one of their two orders contributes.
The vertex $(0,1)$ is the only one of weight one and the others have
weight zero. The area is three, giving $E(t)=1+3t+3t^2$.
\end{example}
\subsection{Lattice paths and the weighted Lah numbers}\label{sec:comparison}
In the lattice path proof, Ferroni, Morales and Panova group the cells of
a subdivision into snakes by the north-east path obtained by expanding
the diagonal steps of a Delannoy path. For a ribbon plane partition of
height at most $t$, summing over its marked high peaks cancels every
labeling that attains $t$ at one of those peaks. The cells still labeled
$t$ form an order filter. Removing it gives the fence order polynomials
in \cite[Theorems 3.7, 3.11 and 3.12]{FMP}. In our face formula the
analogous grouping fixes the intervals supplying boundary terms; the
remaining signed indicators sum to the indicator of one face. For nested
lattice path diagrams the path formula gives coefficientwise monotonicity
by inclusion of the admissible paths \cite[Theorem 4.1]{FMP}.
For $1\le k0$, the bound on $q$ itself must be its upper
bound: a lower bound on $x_q$ would be an upper bound on the chart total and has
boundary factor zero at the origin. Its upper bound is a lower bound on
the chart total, which is sampled from the left. All fixed steps after
the last free step of the chart therefore start just below integers
and must also be upper. This accounts for precisely $r$ fixed ones.
The sum over all free sets and choices of bounds is invariant under
rotation of $q$. Average over $q$ before summing over free sets.
For fixed block sizes $b_1,\ldots,b_m$, the average is
$\prod_i H_{b_i}(z)$, since summing $z^r$ over the positions of block
$i$ gives $b_i H_{b_i}(z)$ and $\sum_i b_i=n$. Marking one of the $m$
free positions converts the cyclic blocks into a composition of $n$
with a chosen first position. Hence
\[
\sum_S \beta(S)z^{\ell(S)}
=\frac nm\sum_{\substack{b_1+\cdots+b_m=n\\b_i\ge1}}
\prod_{i=1}^m H_{b_i}(z).
\]
For a prescribed ordered list of block sizes, distributing $[n]$ among
the blocks gives $n!/\prod b_i!$ choices. In a block of size $b$ the
generating polynomial for its linear orders by the rank minus one of
the first entry is $(b-1)!(1+\cdots+z^{b-1})=b!H_b(z)$.
Dividing by $m!$ to forget the order of the blocks proves
\eqref{eq:weighted-lah-markers}.
To see the Eulerian factor as well, use the simple slices of $[0,L]^n$
at total $kL+1$ for integral $L\ge2$. Maximizing a strictly ordered objective fills the first $k$ positions
with $L$ and the next with $1$. Hence these slices equal
$(L-1)\Delta_{k,n}+\Delta_{k+1,n}$ by their support functions.
Their projected facets are singletons and the total interval, so
Proposition~\ref{prop:beta} applies to this simple deformation (its
proof uses only simplicity, laminar tight intervals and a relation of
the form \eqref{eq:stable}, here with $D=L_0=1$). The limiting
$j$-face for $S$ is a hypersimplex on its $j+1$ free coordinates
of rank $k-\ell(S)$ whenever $1\le k-\ell(S)\le j$.
Its relative volume is $A(j,k-\ell(S)-1)/j!$: across consecutive
partial sums the integer part increases exactly at descents of their
fractional parts. Here $A(j,h)$ counts permutations of $[j]$ with
$h$ descents and is zero when $h$ is outside $0,\ldots,j-1$. This recovers the coefficient formula
\[
(n-1)!\,e_j(\Delta_{k,n})=
\sum_\ell W(\ell,n,j+1)A(j,k-1-\ell)
\]
of \cite[Theorem 4.3]{FerroniHyp}.
\subsection{The hypersimplex upper bound for positroids}\label{sec:upper}
For a matroid $M$, put $E_M=E_{P(M)}$. The bases of a rank-$k$ matroid
on $[n]$ give vertices of $\Delta_{k,n}=P(U_{k,n})$, so
$E_M(t)\le E_{U_{k,n}}(t)$ for every positive integer $t$.
Appendix~\ref{app:upper} shows how the coefficientwise inequality can fail.
The counterexamples do not admit positroid orders.
\begin{question}\label{q:posupper}
Does $E_M\preceq E_{U_{k,n}}$ hold for every rank-$k$ positroid $M$ on
$[n]$?
\end{question}
\begin{corollary}\label{cor:uniform}
The answer is yes. Every positroid is coefficientwise dominated by
the hypersimplex of the same rank and ground-set size.
\end{corollary}
\begin{proof}
Take $P=P(M)$ and $Q=\Delta_{k,n}$ in Theorem~\ref{thm:main}. Both
are integral polypositroids in the positroid order, and $P\subseteq Q$.
\end{proof}
The bound was previously known for lattice path matroids and for elementary split
matroids \cite[Theorem 4.1 and Corollary 4.2]{FMP}.
An exact finite check also verified the comparison for all 125,672
positroids on $[n]$ with $n\le 8$ in the standard cyclic order,
including disconnected positroids. The counts for $n=1,\ldots,8$ are
\[
2,\quad5,\quad16,\quad65,\quad326,\quad1957,\quad13700,\quad109601 .
\]
We enumerated decorated permutations and their Grassmann necklaces
$(I_1,\ldots,I_n)$ as in \cite[Sections 4--5]{ARW}. For an interval $J$
starting at $a$, the greedy basis $I_a$ gives $r(J)=|I_a\cap J|$.
For $t=1,\ldots,n-1$ we counted the points of $t\Delta_{k,n}$ satisfying
$x(J)\le t\,r(J)$ for every cyclic interval $J$. Both polynomials have
degree at most $n-1$ and constant term one, so these counts suffice for
exact interpolation in the monomial basis. Cyclic rotation and duality
were used to reduce the enumeration.
\appendix
\section{The hypersimplex upper bound fails for matroids}
\label{app:upper}
Ferroni conjectured that a
connected rank-$k$ matroid on $n$ elements satisfies
$E_M\preceq E_{U_{k,n}}$ as part of a two-sided coefficient bound
\cite[Conjecture~1.5]{Fmin}. The following examples disprove the upper bound.
\begin{theorem}\label{thm:upper-counterexample}
There is a connected, self-dual, real-representable matroid $M$ of rank
$23$ on $46$ elements such that
\[
[t^4](E_M-E_{U_{23,46}})
=\frac{82498066396874632518608872598768884067}
{2449057743451980875173812652800000}>0 .
\]
There is also a connected, self-dual, $\mathbb{Q}$-realizable example
of rank $25$ on $50$ elements. In both cases the coefficients exceeding
those of the hypersimplex occur precisely in degrees $4,5,6,7$.
\end{theorem}
\begin{proof}
Let $A$ have size $m$. On $A\cup\{p\}$ take the rank-$r$ matroid $F$
whose bases are all $r$-subsets except a specified family
$\mathcal H\subseteq\binom Ar$ of size $h$. We use the following
two families:
\begin{enumerate}[\normalfont (i)]
\item $A=\Z/23\Z$, $r=11$, and $\mathcal H$ consists of the
zero-sum $11$-subsets in this cyclic group. Translation acts transitively
on the possible sums, giving $h=\binom{23}{11}/23=58786$.
\item $A=\{-12,-11,\ldots,12\}$, $r=12$, and $\mathcal H$ consists
of the $12$-subsets with integer sum zero. Here
$h=[u^{12}z^0]\prod_{a=-12}^{12}(1+uz^a)=110780$.
\end{enumerate}
In either case two forbidden sets have at most $r-2$ common elements,
since $r-1$ elements determine the last one from the sum. Thus the
allowed sets satisfy basis exchange. Among two or more candidates for one
exchange at most one is forbidden, and with only one candidate it is
the other basis. Every $(r-1)$-subset extends to a basis and every
$(r+1)$-subset contains one.
Form the $2$-sum $M=F\oplus_2F^*$ at $p$, on two disjoint copies of $A$.
Its bases are obtained by taking a basis from each factor with exactly
one containing $p$ and removing $p$. Thus its rank is $m$, its ground
set has size $2m$, and exchanging the copies identifies $M$ with its dual.
The element $p$ is neither a loop nor a coloop, and $F/p=U_{r-1,m}$
is connected. It follows that $F$ is connected: any component other
than the one containing $p$ would survive contraction as a separate
component, as would the other elements of the component containing $p$.
The dual is likewise connected. In the $2$-sum, two
elements on opposite sides belong to a common circuit: take a circuit
through each of them and $p$ in its factor, and remove $p$ from their
union. Thus the two sides lie in a single connected component.
Put $B_l(t,b)=[z^b](1+z+\cdots+z^t)^l$ and write $f_t(s)$ for the
number of lattice points of $tP(F)$ with $p$-coordinate $s$. We claim
\begin{equation}\label{eq:appendix-squares}
E_M(t)=\sum_{s=0}^t f_t(s)^2,\qquad
E_{U_{m,2m}}(t)=B_{2m}(t,mt).
\end{equation}
The fiber count is
\begin{equation}
f_t(s)=B_m(t,rt-s)-
h\sum_{q=0}^{t-s-1}
\binom{s+q+r-1}{r-1}\binom{q+m-r-1}{m-r-1}.\label{eq:appendix-slice}
\end{equation}
For the first identity, complement the basis from the second factor.
The resulting pairs of bases of $F$ have equal membership of $p$.
Their convex hull consists of all pairs in $P(F)^2$ with equal
$p$-coordinate. Indeed, in convex decompositions of two points at the
same height, the total weight in each $p$-membership class agrees, so
the two distributions can be coupled independently within each nonempty
class. After dilation, projecting away $p$ and complementing the second
copy preserves the lattice, since $s=rt-x(A)$ recovers the omitted
coordinate and the complement is $y_a\mapsto t-y_a$.
For the fiber count, the matroid rank inequalities on this slice are
the box bounds, the equation $x(A)=rt-s$, and $x(H)\le(r-1)t$ for
$H\in\mathcal H$. The rank assertions above give this description
directly from the base-polytope inequalities. Two distinct forbidden
inequalities cannot both be violated, since
\[
x(H)+x(J)\le x(A)+t|H\cap J|\le 2(r-1)t .
\]
For a violation at $H$ let $q=x(A\setminus H)$. The $r$ deficits
$t-x_i$ on $H$ sum to $s+q0]
\end{verbatim}
\endgroup
At $(m,r,h)=(23,11,58786)$ this gives the displayed fraction and
$[4,5,6,7]$. At $(25,12,110780)$ it gives the same degree list and
quartic excess
\[
\frac{12728700083893901323277370650378549292808943}
{50380056652099388979375534843279360000}.
\]
To verify representability in (i), take $z_a=\exp(2\pi ia/23)$ and use
the columns $(z_a,z_a^2,\ldots,z_a^{10},z_a^{11}-1)^{\mathsf T}$ and
$v_p=e_{11}$. On eleven finite columns the determinant is the Vandermonde
times $(\prod_a z_a)-1$, by expansion in the last row. Minors containing
$p$ are nonzero by Vandermonde. Rescaling each finite column by
$z_a^{-17}$ changes the first ten rows into the powers $7,\ldots,16$
modulo $23$, which occur in conjugate pairs. Replace each pair by its
real and imaginary parts. The last row is $z_a^{-6}-z_a^6$ on the
finite columns, so multiplying it by $i$ and rescaling $v_p$ gives a
real matrix. In (ii) use the integral columns
$(1,a,\ldots,a^{10},a^{12})^{\mathsf T}$ and $v_p=e_{12}$.
On twelve finite columns the determinant is the Vandermonde times
their sum, by reducing $X^{12}$ modulo $\prod_a(X-a)$.
Again the minors containing $p$ are nonzero.
Representability over either field is preserved by duality and by
$2$-sums: for the latter, identify the two lines representing $p$
in the direct sum of the representing spaces and delete $p$.
\end{proof}
Neither example is a positroid in any cyclic order. Indeed,
$M|A=F\setminus p$, as follows from the basis description of the
$2$-sum. Each $H\in\mathcal H$ supports a distinct facet of this
restriction. The bases $H-i+j$, for $i\in H$ and $j\in A\setminus H$,
span the product of two simplices of total dimension $m-2$ on the
bounding hyperplane. There is also a basis disjoint from $H$, since
$m-r\ge r+1$. A positroid on $m$ elements can have at most $m$ such
facets with a prescribed support cardinality $r$, since the support
must be a cyclic interval \cite[Proposition~5.7]{ARW}. In both cases
$h>m$, and restrictions of positroids preserve positroidality
\cite[Proposition~3.5]{ARW}.
\begin{thebibliography}{FMP26}
\bibitem[ARW16]{ARW}
F. Ardila, F. Rinc\'on and L. Williams,
\emph{Positroids and non-crossing partitions},
Trans. Amer. Math. Soc. \textbf{368} (2016), 337--363.
\bibitem[Fer21]{FerroniHyp}
L. Ferroni, \emph{Hypersimplices are Ehrhart positive},
J. Combin. Theory Ser. A \textbf{178} (2021), 105365.
\bibitem[Fer22]{Fmin}
L. Ferroni, \emph{On the Ehrhart polynomial of minimal matroids},
Discrete Comput. Geom. \textbf{68} (2022), 255--273.
\bibitem[FJS22]{FJS}
L. Ferroni, K. Jochemko and B. Schr\"oter,
\emph{Ehrhart polynomials of rank two matroids},
Adv. Appl. Math. \textbf{141} (2022), 102410.
\bibitem[FMP25]{FMPskew}
L. Ferroni, A. H. Morales and G. Panova,
\emph{Skew shapes, Ehrhart positivity and beyond},
Proc. Lond. Math. Soc. (3) \textbf{133} (2026), e70204.
\bibitem[FMP26]{FMP}
L. Ferroni, A. H. Morales and G. Panova,
\emph{Ehrhart positivity for lattice path matroids},
\href{https://arxiv.org/abs/2605.22673}{arXiv:2605.22673}.
\bibitem[LP24]{LP}
T. Lam and A. Postnikov, \emph{Polypositroids},
Forum Math. Sigma \textbf{12} (2024), e42.
\bibitem[LS26]{LS}
Y. Li and S. Shao, \emph{Series-parallel extensions and Ehrhart positivity},
\href{https://arxiv.org/abs/2610.03122}{arXiv:2610.03122}.
\bibitem[LZ26]{LZ}
F. Liu and Z. Zhang, \emph{Ehrhart positivity for paving positroids},
\href{https://arxiv.org/abs/2610.21501}{arXiv:2610.21501}.
\end{thebibliography}
\end{document}