Almost simplicial polytopes I. The lower and upper bound theorems
Extracted source text ยท arXiv 1510.08258
\documentclass[11pt,a4paper]{amsart}
%---AMS PACKAGES------
\usepackage{amsfonts,amsmath, amssymb}
\usepackage{color}
%---MATH FONTS--------
%\usepackage{iwona}
%\usepackage{arev}
%\usepackage{cmbright}
%\usepackage{fourier}
%\usepackage{pxfonts}
\usepackage{mathpazo}
%\usepackage{lmodern}
%\usepackage[T1]{fontenc}
%\usepackage[Garamond]{mathdesign}
%---PAGE SETUP-------
\usepackage{geometry} % See geometry.pdf to learn the layout options. There are lots.
\geometry{} % ... or a4paper or a5paper or ...
%\geometry{landscape} % Activate for for rotated page geometry
%\usepackage[parfill]{parskip} % Activate to begin paragraphs with an empty line rather than an indent
\usepackage{anysize}
\marginsize{3cm}{3cm}{3cm}{3cm}%\marginsize{left}{right}{top}{bottom}
\renewcommand{\baselinestretch}{1.25}
\setlength{\footnotesep}{\baselinestretch\footnotesep}
%\renewcommand{\thefootnote}{\fnsymbol{footnote}}
\setlength{\parindent}{0cm}
\setlength{\parskip}{2ex}
%---REFERENCES------
\usepackage[sort&compress,square,comma,numbers]{natbib}
%---WATERMARKS------
%\usepackage{draftwatermark}
%\SetWatermarkFontSize{5cm}
%\SetWatermarkScale{2}
%\SetWatermarkLightness{0.82}
%\SetWatermarkText{Draft-Not for public distribution}
%---MISCELLANEOUS---
\usepackage{graphicx,float,enumerate}
\usepackage[ruled,boxed,commentsnumbered,norelsize]{algorithm2e}
\usepackage{verbatim}
\usepackage{url}
\usepackage{epstopdf}
%\DeclareGraphicsRule{.tif}{png}{.png}{`convert #1 `dirname #1`/`basename #1 .tif`.png}
\usepackage[capitalise]{cleveref}
%---MATH ENVIRONMENTS---
\newtheorem{theorem}{Theorem}[section]
\newtheorem{corollary}[theorem]{Corollary}
\newtheorem{lemma}[theorem]{Lemma}
\newtheorem{proposition}[theorem]{Proposition}
\newtheorem{problem}[theorem]{Problem}
\newtheorem{claim}{Claim}
\newtheorem{conjecture}[theorem]{Conjecture}
\crefname{conjecture}{Conjecture}{Conjectures}
\newtheorem{question}[theorem]{Question}
\crefname{question}{Question}{Questions}
%\theoremstyle{definition}
\newtheorem{definition}[theorem]{Definition}
\theoremstyle{remark}
\newtheorem{remark}[theorem]{Remark}
%---MATH OPERATORS & SYMBOLS----
\usepackage{mathtools}
\DeclarePairedDelimiter\ceil{\lceil}{\rceil}
\DeclarePairedDelimiter\floor{\lfloor}{\rfloor}
\DeclareMathOperator{\dist}{dist}
%\DeclareMathOperator{\deg}{deg}
\DeclareMathOperator{\conv}{conv}
\DeclareMathOperator{\ver}{vert}
\DeclareMathOperator{\st}{star}
\DeclareMathOperator{\lk}{link}
\DeclareMathOperator{\aff}{aff}
\DeclareMathOperator{\ri}{ri}
\DeclareMathOperator{\tr}{tr}
\DeclareMathOperator{\pyr}{pyr}
\DeclareMathOperator{\intP}{int}
\newcommand{\D}{\Delta}
\newcommand{\R}{\mathcal{R}}
\newcommand{\Sp}{\mathcal{S}}
\newcommand{\Pl}{\mathcal{P}}
\newcommand{\E}{\mathcal{E}}
\newcommand{\C}{\mathcal{C}}
%\newcommand{\P}{\mathcal{P}}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\newcommand{\marrow}{\marginpar{$\longleftarrow$}}
\newcommand{\eran}[1]{\textsc{\textcolor{red}{Eran says:}} \marrow \textsf{#1}}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\newcommand{\guille}[1]{\textsc{\textcolor{blue}{Guillermo says:}} \marrow \textsf{#1}}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\date{\today}
\title[Almost simplicial polytopes]{Almost simplicial polytopes: the lower and upper bound theorems}
\author{Eran Nevo}
\address{Institute of Mathematics, Hebrew University of Jerusalem}
\email{\texttt{nevo@math.huji.ac.il}}
\author{Guillermo Pineda-Villavicencio}
\author{Julien Ugon}
\author{David Yost}
\address{Centre for Informatics and Applied Optimisation, Federation University Australia}
\email{\texttt{\{g.pinedavillavicencio,j.ugon,d.yost\}@federation.edu.au}}
%\thanks{Thanks}
\keywords{polytope; simplicial polytope; almost simplicial polytope; Lower Bound theorem; Upper Bound theorem; graph rigidity; $h$-vector; $f$-vector}
\subjclass[2010]{Primary 52B05; Secondary 52B12, 52B22}
\thanks{Research of E.~Nevo was partially supported by Israel Science Foundation grants ISF-805/11 and ISF-1695/15.
}
\begin{document}
\maketitle
\begin{abstract}
We study $n$-vertex $d$-dimensional polytopes with at most one nonsimplex facet with, say, $d+s$ vertices, called {\it almost simplicial polytopes}. We provide tight lower and upper bound theorems for these polytopes as functions of $d,n$ and $s$,
thus generalizing the classical Lower Bound Theorem by Barnette and Upper Bound Theorem by McMullen, which treat the case of $s=0$.
We characterize the minimizers and provide examples of maximizers, for any $d$. Our construction of maximizers is a generalization of cyclic polytopes, based on a suitable variation of the moment curve, and is of independent interest.
\end{abstract}
%\tableofcontents
\section{Introduction}
In 1970 McMullen \cite{McM70a} proved the Upper Bound Theorem (UBT) for {\it simplicial polytopes}, polytopes with each facet being a simplex, while between 1971 and 1973 Barnette \cite{Bar71,Bar73} proved the Lower Bound Theorem (LBT) for the same polytopes. Both results are major achievements in the combinatorial theory of polytopes; see, e.g., the books \cite{Gru03, Zie95} for further details and discussion.
These results can be phrased as follows: let $C(d,n)$ (resp. $S(d,n)$) denote a cyclic (resp.~stacked) $d$-polytope on $n$ vertices, and for a polytope $P$ let $f_i(P)$ denote the number of its $i$-dimensional faces. Then the classical LBT and UBT read as follows.
\begin{theorem}[(Classical LBT and UBT)]
For any simplicial $d$-polytope $P$ on $n$ vertices, and any $0\le i\le d-1$,
\[f_i(S(d,n))\le f_i(P) \le f_i(C(d,n))
.\]
\end{theorem}
The numbers $f_i(S(d,n))$ and $f_i(C(d,n))$ are explicit known functions of $(d,n,i)$, to be discussed later.
We generalize the UBT and LBT to the following context:
consider a pair $(P,F)$ where $P$ is a polytope, $F$ is a facet of $P$, and all facets of $P$ different from $F$ are simplices. We call such a polytope
$P$ an \emph{almost simplicial polytope} (ASP) and a pair $(P,F)$ an ASP-pair. In this case, the facet $F$ is necessarily simplicial.
We will be interested only in the combinatorics of $P$, thus the ASP-pair $(P,F)$ is equivalent to specifying a regular triangulation of $F$ admitting a lifting of its vertices that leaves the vertices of $F$ fixed; we are interested in the simplicial ball $P':=\partial P - \{F\}$.
Let $\mathcal{P}(d,n,s)$ denote the family of $d$-polytopes $P$ on $n$-vertices such that $(P,F)$ is an ASP-pair, where $F$ has $d+s$ vertices ($s\ge 0$). Note that $\mathcal{P}(d,n,0)$ consists of the simplicial $d$-polytopes on $n$ vertices.
In this paper, we define certain polytopes $C(d,n,s),S(d,n,s)\in \Pl(d,n,s)$, explicitly compute their face numbers, and show the following.
\begin{theorem}[(LBT and UBT for ASP)]\label{thm:Main}
For any $d,n,s$, any polytope $P\in \Pl(d,n,s)$, and any $0\le i\le d-1$,
\[f_i(S(d,n,s))\le f_i(P) \le f_i(C(d,n,s))
.\]
Further, for $d\geq 4$, the polytopes $P\in \Pl(d,n,s)$ with $f_i(P)=f_i(S(d,n,s))$ for \emph{some} $1\le i\le d-1$ are characterized combinatorially, and satisfy the above equality for \emph{all} $0\le i\le d-1$.
\end{theorem}
The characterization of the equality case above generalizes Kalai's result \cite{Kal87} that for $d\geq 4$ equality in the classical LBT holds for some $1\le i\le d-1$ if and only if $P$ is stacked.
The polytopes $C(d,n,s)$ form an ASP analog of cyclic polytopes and satisfy a combinatorial Gale-evenness type description of their facets.
Billera and Lee \cite{BilLee81a} considered the notion of polytope pairs. In particular, their results give tight upper and lower bound theorems for the face numbers of simplicial $(d-1)$-dimensional balls of the ``\emph{polytope-antistar}" form; that is, balls of the form $\partial Q-v$, where $Q$ is a simplicial $d$-polytope and $v$ is a vertex of $Q$ that is deleted.
These bounds are given as functions of $d,f_0(\partial Q-v),f_0(Q/v)$, where $Q/v$ denotes the vertex figure of $v$ in $Q$. For an ASP-pair $(P,F)$, let $Q$ be obtained from $P$ by stacking a pyramid over $F$ with a new vertex $v$. Then
$F\cong Q/v$ and $P'=\partial P - \{F\} = \partial Q-v$. Thus, our balls $P'$ form a subfamily of the balls $\partial Q-v$ considered in \cite{BilLee81a}. The bounds we obtain in Theorem~\ref{thm:Main} are strictly stronger than those of \cite{BilLee81a} which apply to all polytope-antistar balls.
Let $f(P)=(1,f_0(P),f_1(P),\cdots,f_{d-1}(P))$ denote the {\it $f$-vector} of $P$, a vector recording the face numbers of $P$.
The following problem naturally arises.
\begin{problem}
Characterize the pairs of $f$-vectors $(f(P),f(F))$ for ASP-pairs $(P,F)$.
\end{problem}
A solution to the problem above would generalize the well known $g$-theorem characterizing the face numbers of simplicial polytopes, conjectured by McMullen \cite{McMullen-g-conj} and proved by Billera-Lee \cite{Billera-Lee} (sufficiency) and Stanley \cite{Sta80} (necessity); the $g$-theorem solves the case $s=0$ and provides some restrictions when $s>0$.
We leave this general problem to a future study.
We remark that for the corresponding problem for the larger family of polytope pairs \cite{BilLee81a}, currently there is no conjectured characterization, after Kolins' \cite{Kolins11} counterexamples to the characterization conjectured by Billera and Lee \cite{BilLee81a}.
%However, we give an ASP analogue to
%the Generalized Lower Bound Theorem \cite{McMWal71,MurNev13};
%see Theorem~\ref{thm:GLBT_ASP}.
The proof of the LBT for ASP and the characterization of the equality cases are based on framework-rigidity arguments (cf.~Kalai~\cite{Kal87}) and on an adaptation of the well known McMullen-Perles-Walkup reduction (MPW) \cite[Sec.~5]{Kal87} to ASP; see Section~\ref{sec:LBT}.
The numerical bounds obtained in the UBT for ASP are a special case of a recent result of Adiprasito and Sanyal~\cite[Thm.~3.9]{AdiSan14}, who proved the bounds for homology balls whose boundary is an induced subcomplex. While their proof relies on machinery from commutative algebra, our proof is elementary and is based on a suitable shelling of $P$. Further, our construction of maximizers $C(d,n,s)$ is a generalization of cyclic polytopes, based on a suitable variation of the moment curve, and is of independent interest;
see Section~\ref{sec:UBT}.
%The rest of the paper is structured as follows.
\section{Preliminaries}\label{sec:prelim}
For undefined terminology and notation, see \cite{Zie95} for polytopes and complexes, or \cite[Sec.~2]{Kal87} for framework rigidity.
\subsection{Polytopes and simplicial complexes}
The $k$-dimensional faces of a polyhedral complex $\Delta$ are called
$k$-faces, where the empty face has dimension $-1$.
For a simplicial complex $\Delta$ of dimension $d-1$, the numbers $f_k(\Delta)$ are then related to the $h$-numbers
$h_k(\Delta):=\sum_{i=0}^k (-1)^{k-i}{d-i \choose k-i}f_{i-1}(\Delta)$ by
\begin{align}\label{eq:fk_hk}
f_{k-1}(\Delta)=\sum_{i=0}^k {d-i\choose k-i}h_i(\Delta).
\end{align}
The $h$-vector of $\Delta$, $(\ldots,h_k,h_{k+1},\ldots)$, can be considered as an infinite sequence if we let $h_k(\Delta)=0$ for $k>d$ and $k<0$.
The $g$-numbers
%of $\Delta$
are defined by $g_k(\Delta)=h_k(\Delta)-h_{k-1}(\Delta)$.
For an ASP pair $(P,F)$, where $P$ is $d$-dimensional,
the following version of the Dehn-Somerville equations
%for simplicial polytopes \cite[Thm.~8.21]{Zie95}
applies to the complex $P'=\partial P-\{F\}$.
\begin{proposition}[{\cite[Thm.~18.3.6]{GooORo04}}, Dehn-Somerville Equations for $P'$]\label{prop:Dehm-Somerville}
The $h$-vector of the simplicial $(d-1)$-ball
$P'$ with boundary $\partial F$
satisfies
%\begin{align*}
%h_k(P')&=h_{d-k}(P')+\sum_{j=0}^k{d-j\choose k-j}(-1)^{k-j}f_{j-1}(F)\quad \text{for $k=0,\ldots,d$.}
%\end{align*}
%Equivalently,
for $k=0,\ldots,d$
%we have
\begin{align}\label{eq:DSEq-hP}
%h_k(P')=h_{d-k}(P')+h_{k}(F)-h_{k-1}(F)\; \mathrm{i.e.}\;
h_k(P')&=h_{d-k}(P')+g_{k}(\partial F).
\end{align}
\end{proposition}
Note that $h_{k}(P')=0$ and $h_k(\partial F)=0$ for $k\ge d$ and $h_{d-1}(\partial F)=1$.
%Following \cite[Sec.~8.4]{Zie95}, we define the following notation.
%\begin{align*}
%\displaystyle\sideset{}{^*}\sum_{i=0}^{d/2} T_i&=\begin{cases}
%\displaystyle \sum_{i=0}^{\floor{d/2}}T_i,&\text{if $d$ is odd};\\
%\displaystyle\sum_{i=0}^{\floor{d/2}-1}T_i+\frac{1}{2}T_{\floor{d/2}},&\text{if $d$ is even}.
%\end{cases}
%\end{align*}
%\eran{}
%and obtain from (\ref{eq:DSEq-hP})
%applied to $f_{k-1}(P')=\sum_{i=0}^k {d-i\choose k-i}h_i(P')$ that
%\begin{align}\label{eq:fk_hk}
%f_{k-1}(P')&=\displaystyle \sideset{}{^*}\sum_{i=0}^{d/2}{d-i\choose k-i} h_{i}(P')+{i\choose k-d+i}h_{d-i}(P'),\; \text{for $0\le k\le d$}.
%\end{align}
Let $2^A$ denote the simplicial complex generated by the set $A$; it is a simplex. Sometimes we abbreviate this complex by $A$, when the context is clear.
Say $\Delta$ is \emph{pure} if all its maximal faces, called \emph{facets}, have the same dimension, and a pure simplicial complex $\Delta$ is \emph{shellable} if its facets can be ordered $F_1,F_2,\ldots$ such that for each $j>1$, $2^{F_j}$ intersects the complex $\cup_{i<j}2^{F_i}$ in a pure codimension $1$ subcomplex of $2^{F_j}$. Such an order is called a \emph{shelling order} or \emph{shelling process} of $\Delta$.
For a shelling order, the set of faces $2^{F_j} - \cup_{i<j}2^{F_i}$ has a unique minimal element, called the \emph{restriction face} of $F_j$, denoted $R_j$.
For \emph{any} shelling of $\Delta$, $h_i(\Delta)$ equals the number of facets in the shelling whose restriction face has size $i$; cf. \cite[Thm.~8.19]{Zie95}.
%independent of the particular shelling chosen , and depends only on the face numbers of $\mathcal{C}$.
%We next recall some necessary definitions.
Note that $P'$ is shellable, by a Bruggesser-Mani line shelling \cite[Prop.~2]{BruMan71}.
The \emph{link} of a face $F$ in the simplicial complex $\Delta$ is $\lk_{\Delta}(F):=\{T\in \Delta:\ T\cap F =\emptyset,\ F\cup T\in \Delta\}$, and its \emph{star}, $\st_{\Delta}(F)$ is the complex $\cup_{F\subseteq T}2^T$. Thus, using the \emph{join} operator on simplicial complexes, we obtain $2^F * \lk_{\Delta}(F)=\st_{\Delta}(F)$. For a general polyhedral complex, the star of a face $F$ is the polyhedral subcomplex formed by all faces containing $F$, and their faces.
For a vertex $v$ in a polytope $Q$, its \emph{vertex figure}
$Q/v$ is a codimension $1$ polytope obtained by intersecting $Q$ with a hyperplane $H$ \emph{below} $v$, that is, $v$ is on one side of $H$ and the other vertices of $Q$ are on the other side. If $\st_Q(v)$ is simplicial then the boundary complex of $Q/v$ coincides with $\lk_{Q}(v)$.
A subcomplex $K$ of $\Delta$ is {\it induced} if it contains all the faces in $\Delta$ which only involve vertices in $K$. Note that, for an ASP-pair $(P,F)$, $\partial F$ is an induced subcomplex of $P'$, by convexity.
A polytope is {\it $k$-neighborly} if each subset of at most $k$ vertices forms the vertex set of a face. A $\floor{d/2}$-neighborly $d$-polytope is simply called {\it neighborly}. A polytope is {\it $k$-simplicial} if each $k$-face is a simplex.
The underlying set $|\mathcal{C}|$ of a polyhedral complex $\mathcal{C}$ is the point set $\cup_{Q\in \mathcal{C}} Q$ of its geometric realization.
A \emph{refinement} (or subdivision) of $\mathcal{C}$ is another polyhedral complex $\mathcal{D}$ such that $|\mathcal{D}|=|\mathcal{C}|$ and for any face $F\in \mathcal{D}$ there exists a face $T\in \mathcal{C}$ such that $|F|\subseteq |T|$.
Let $G$ be a proper face of a polytope $Q$. A point $w$ is {\it beyond} $G$ (with respect to $Q$) if
%\textcolor{blue}{
(i) $w$ is not on any hyperplane supporting a facet of $Q$, (ii)
$w$ and the interior of $Q$ lie on different sides of any hyperplane supporting a facet containing $G$, but (iii) on the same side of every other facet-defining hyperplane which does not contain $G$.
For an ASP-pair $(P,F)$ we will consider the simplicial polytope $Q$ obtained as the convex hull of $P$ and a vertex $y$ beyond $F$.
A simplicial complex $\Delta$ is a \emph{homology sphere} (over a fixed field $\mathbf{k}$) if for any face $F\in \Delta$, the reduced homology groups $\tilde{H}_i(\lk_{\Delta}(F); \mathbf{k})\cong \tilde{H}_i(S^{\dim \Delta -\dim F -1}; \mathbf{k})$ for all $i$, where $S^j$ is the $j$-dimensional sphere.
Say $\Delta$ is a \emph{homology ball} if $\tilde{H}_i(\lk_{\Delta}(F); \mathbf{k})$ vanishes for $i<\dim \Delta -\dim F -1$ and is isomorphic to either $0$ or $\mathbf{k}$ for $i=\dim \Delta -\dim F -1$. Furthermore, the boundary complex $\partial \Delta$ of $\Delta$, consisting of all faces $F$ for which $\tilde{H}_{\dim \Delta -\dim F -1}(\lk_{\Delta}(F); \mathbf{k})=0$, is a homology sphere (of codimension $1$). In particular, simplicial spheres (resp. balls) are homology spheres (resp. balls).
A polytope is {\it stacked} if it can be obtained from a simplex by repeatedly taking the convex hull with a vertex beyond some facet.
A homology sphere is {\it stacked} if it is combinatorially isomorphic to the boundary complex of a stacked polytope.
\subsection{Rigidity}
We mostly follow the presentation in Kalai's \cite{Kal87}.
Let $G=(V,E)$ be a graph, and $\rm{dist}(a,b)$ denote Euclidean distance
between points $a$ and $b$ in a Euclidean space.
A $d$-embedding $f:V\rightarrow \mathbf{R}^{d}$ is called \emph{rigid}
if there exists an $\epsilon>0$ such that if $g:V\rightarrow
\mathbf{R}^{d}$ satisfies $\textrm{dist}(f(v),g(v))<\epsilon $ for every
$v\in V$ and $\rm{dist}(g(u),g(w))=\rm{dist}(f(u),f(w))$ for every $\{u,w\}\in E$,
then $\rm{dist}(g(u),g(w))=\rm{dist}(f(u),f(w))$ for every $u,w\in V$.
%Loosely speaking, $f$ is rigid if any perturbation of it which
%preserves the lengths of the edges is induced by an isometry of $\mathbf{R}^d$.
$G$ is called \emph{generically d-rigid} if the set of its rigid
$d$-embeddings is open and dense in the topological vector space
of all of its $d$-embeddings.
Given a $d$-embedding $f:V\rightarrow \mathbf{R}^{d}$, a \emph{stress}
on $f$ is a function $w:E\rightarrow \mathbf{R}$ such that for
every vertex $v\in V$
\[\sum_{u:\{v,u\}\in E}w(\{v,u\})(f(v)-f(u)) =0.\]
The stresses on $f$ form a vector space, called the \emph{stress space}.
Its dimension is the same for all generic $d$-embeddings (namely, for an open and dense set in the space of all $d$-embeddings of $G$). A graph
$G$ is called \emph{generically d-stress free} if this dimension is zero.
If a generic $f:V\rightarrow \mathbf{R}^{d}$ is rigid, then
$f_1(G)\ge df_0(G)-{d+1 \choose 2}$.
Thus, if $\Delta$ is a simplicial complex of dimension $d-1$ whose $1$-skeleton is generically $d$-rigid, then
$f_1(\Delta)\ge df_0(\Delta)-{d+1 \choose 2}$, and $g_2(\Delta)$ is the dimension of the stress space of any generic embedding.
Based on these observations for $\Delta$ the boundary of a simplicial $d$-polytope with $d\ge 3$, and more general complexes, Kalai~\cite{Kal87} extended the LBT and characterized the minimizers.
For a $d$-polytope $P$ with a simplicial $2$-skeleton, the toric $g_2(P)$ equals $g_2(\partial P):=f_1(P)-df_0(P)+\binom{d+1}{2}$, and by
%\textcolor{blue}{
a result of Alexandrov (cf.~Whiteley~\cite{Whi84}), it
equals the dimension of the stress space of the $1$-skeleton of $P$.
For our LBT for ASP, we will need the following very special case of Kalai's monotonicity\footnote{Kalai's monotonicity conjecture on the toric $g$-polynomials, asserting that $g(P)\ge g(F)g(P/F)$ coefficientwise for any face $F$ of $P$, was first proved for rational polytopes by Braden and MacPherson \cite{BraMac99}. Later, using the theory of combinatorial intersection homology, Braden \cite{Bra06} proved Kalai's conjecture in full generality.},
which Kalai proved using rigidity arguments.
\begin{theorem}[(Kalai's Monotonicity {\cite[Thm.~4.1]{Kal94}}, weak form)]\label{thm:kalaiMonotonicity}
Let $d\ge 4$, $P$ a $d$-polytope with a simplicial $2$-skeleton, and $F$ a facet of $P$. Then
\[g_2(P)\ge g_2(F).\]
Equivalently,
$f_1(P)-f_1(F)\ge (df_0(P)-\binom{d+1}{2})-((d-1)f_0(F)-\binom{d}{2})$.
\end{theorem}
\section{A lower bound theorem for almost simplicial polytopes}
\label{sec:LBT}
Recall that a simplicial $d$-polytope is called \emph{stacked} if it can be obtained from a $d$-simplex by repeated \emph{stacking}, namely, adding a vertex beyond a facet and taking the convex hull. While stacked $d$-polytopes on $n$ vertices, denoted $S(d,n)$, may have different combinatorial structures, they all have the same $f$-vector, given by \[f_k(S(d,n))=\phi_{k}(d,n):=\begin{cases}
{d\choose k}n-{d+1\choose k+1}k&\text{for $k=1,\ldots,d-2$}\\
(d-1)n-(d+1)(d-2) &\text{for $k=d-1$}.\end{cases}\]
For any integers $d\ge 3$, $s\ge 0$ and $n\ge d+s+1$, let $F$ be a stacked $(d-1)$-polytope with $d+s$ vertices.
Construct a pyramid over $F$ and then stack $n-d-s-1$ times over facets of the resulting polytope which are
%\eran{See how I defined "beyond" and "stacking".}
different from $F$
%not contained in $\aff F$, the affine space spanned by $F$,
to obtain a polytope $S(d,n,s)$ in $\Pl(d,n,s)$.
One easily computes the $f$-vector of $S(d,n,s)$, since refining $F$ by its (unique) stacked triangulation refines the boundary complex of $S(d,n,s)$ to a stacked simplicial sphere with $f$-vector $f(S(d,n))$. We obtain
\[f(S(d,n,s)) = f(S(d,n)) - (0,0,\cdots,0,s,s)
.\]
Note that for $n\ge s+4$, any $P\in \Pl(3,n,s)$ has $f$-vector
$f(P)=(1,n,3n-6-s,2n-4-s)=f(S(3,n,s))$.
%=f(S(3,n))-(0,0,s,s)$.
We are ready to state the LBT for ASP; its minimizers will be characterized later (see \cref{thm:LBT>4,thm:LBT=4}).
\begin{theorem}[(LBT for ASP)]\label{thm:LBT}
Let $d\ge 3$, $s\ge 0$, $n\ge d+s+1$. Then for any $P\in \Pl(d,n,s)$ and $1\le i \le d-1$ we have
\[f_i(S(d,n,s))\le f_i(P)
.\]
\end{theorem}
\begin{proof}
We proceed by induction on $d$, the case $d=3$ was verified above.
Let $d\ge 4$.
As $P$ is $2$-simplicial,
by a result of Whiteley~\cite[Thm. 8.6]{Whi84}, the $1$-skeleton of $P$ is generically $d$-rigid, hence $f_1(P)\ge \phi_{1}(d,n)$, and by the MPW reduction (as usual, counting pairs $(v,A)$ such that $v$ is a vertex in an $i$-face $A$), $f_i(P)\ge \phi_{i}(d,n)$ for all $2\le i\le d-3$ as well; see \cite[Thm.~12.2]{Kal87}
\footnote{Kalai's theorem contains a typo. It includes the case $i=k$, while it holds only for $i<k$, where $P$ is $k$-simplicial.
%, namely its $k$-skeleton is a simplicial complex.
Our ASP $P$ is $(d-2)$-simplicial.}.
Denote by $(P,F)$ the ASP-pair, and by $\deg_P(v)$ the degree of a vertex $v$ in the $1$-skeleton of $P$. We now prove the inequality for the facets, by a variation of the MPW reduction.
Note that the vertex figure $P/v$ in $P$ of any vertex $v\in \ver F$ is an ASP (with $\deg_P(v)$ vertices), while for any vertex $v\in \ver P\setminus \ver F$ $P/v$ is a simplicial polytope; cf. \cite[Thm.~11.5]{Bro83}. Furthermore, for a vertex $v\in \ver F$, letting $s_v:=\deg_F(v)-(d-1)\ge 0$ gives
$P/v \in \Pl(d-1,\deg_P(v),s_v)$.
Double counting the number of pairs $(v,A)$ for a vertex $v$ in a facet $A$ of $P$, we obtain the following inequalities:
\begin{multline*}
d(f_{d-1}(P)-1) + (d+s) = \sum_{v\in \ver P}f_{d-2}(\lk_P(v))\\
\ge
\sum_{v\in \ver P \setminus \ver F}((d-2)\deg_P(v) - d(d-3))
+ \sum_{v\in \ver F}((d-2)\deg_P(v) - d(d-3)-s_v)\\
= 2(d-2)f_1(P) - d(d-3)f_0(P) - 2f_1(F) + (d-1)(d+s)\\
\ge
2(d-2)\left[d f_0(P)-\binom{d+1}{2}\right] - d(d-3)f_0(P) - 2\left[(d-1)f_0(F)-\binom{d}{2}\right] +(d-1)(d+s)\\
= d(d-1)f_0(P) - d(d+1)(d-2) - s(d-1),
\end{multline*}
where the first inequality is by the induction hypothesis and the second inequality is by
Kalai's monotonicity Theorem~\ref{thm:kalaiMonotonicity}
and the LBT inequality for $f_1(P)$.
Comparing the LHS with the RHS gives
\[f_{d-1}(P)\ge \phi_{d-1}(d,n)-s.\]
The inequality for $f_{d-2}(P)$ follows from the inequality for $f_{d-1}(P)$ by double counting.
Since any ridge in $P$ is contained in exactly two facets, counting the number of pairs $(R,A)$ for a ridge $R$ in a facet $A$ of $P$, we obtain that
\[2f_{d-2}(P)=d(f_{d-1}(P)-1)+f_{d-2}(F).\]
Applying the classical LBT to the simplicial $(d-1)$-polytope $F$ with $f_0(F)=d+s$, we get
\[2f_{d-2}(P)\ge d(f_{d-1}(P)-1)+ (d-2)(d+s) - d(d-3),\]
and applying the lower bound for $f_{d-1}(P)$ yields, after dividing both sides by $2$,
the desired lower bound $f_{d-2}(P)\ge \phi_{d-2}(d,n)-s$.
\end{proof}
We now turn our attention to characterizing the minimizers of Theorem~\ref{thm:LBT}. We start with some terminology and background.
A proper subset $A$ of the vertices of a $d$-polytope $P$ is called a \emph{missing $k$-face} of $P$ if the cardinality of $A$ is $k+1$, the simplex on $A$ is not a face of $P$, but for any proper subset $B$ of $A$ the simplex on $B$ is a face of $P$. If $A$ is a missing $(d-1)$-face of $P$ then adding the simplex $A$ cuts $P$ into two $d$-polytopes $P_1, P_2$, glued along the simplex $A$. We denote this operation by $P=P_1\# P_2$. Repeating this procedure on each $P_i$ until no piece $P_i$ contains a missing $(d-1)$-face results in a decomposition
$P=P_1\# P_2\#\cdots\# P_t$,
where intersections along missing $(d-1)$-faces of $P$ define the edges of a tree whose vertices are the $P_i$'s.
Note that for $d\ge 3$ a decomposition of $P$ as above is uniquely defined; just insert all the missing $(d-1)$-faces. Call such a decomposition the \emph{prime decomposition} of $P$, and call each $P_i$ a \emph{prime factor} of $P$. Denote by $\Delta_P$ the polyhedral complex defined by the prime decomposition of $P$.
Then a simplicial $d$-polytope $P$ is stacked if and only if all its prime factors are $d$-simplices. This definition immediately extends to polyhedral spheres where the operation $\#$ corresponds to the topological connected sum.
%$g_2(P)=g_2(P_1)+g_2(P_2)$
We start with the characterization of the minimizers for the easier case $d>4$.
\begin{theorem}[(Characterization of minimizers for $d>4$)]\label{thm:LBT>4}
Let $d>4$ and $P\in \Pl(d,n,s)$. Let $\Delta_F$ be the polyhedral complex corresponding to the prime decomposition of $F$, and let $\Delta$ be the refinement of the boundary complex $\partial P$ of $P$ obtained by refining $F$ by $\Delta_F$. Assume there is some $1\le i\le d-1$ for which $f_i(P)=f_i(S(d,n,s))$.
Then, all prime factors in the prime decomposition of $\Delta$ are $d$-simplices.
In particular, $f(P)=f(S(d,n,s))$.
\end{theorem}
\begin{remark}
Let $Q$ be a polytope, $G$ a facet of $Q$ and $H$ the hyperplane containing $G$. An \emph{$H$-stacking} on $Q$ is the operation of (i) adding a new vertex $w$ in $H$, beyond a facet of $G$ (with respect to $G$)
such that perturbing $w$ from $H$ to the side of the interior of $Q$ makes $w$ beyond a facet of $Q$, and (ii) taking the convex hull of $w$ and $Q$.
The minimizers $P$ considered in Theorem \ref{thm:LBT>4} are precisely the polytopes that can be obtained by the following recursive procedure: start with a $d$-simplex having a facet in a hyperplane $H$, and repeatedly either $H$-stack or (usual) stack over a facet not in $H$.
\end{remark}
\begin{proof}[Proof of Theorem~\ref{thm:LBT>4}]
By the classical MPW reduction for $1\leq i\leq d-3$ and the variation of it we used in the proof of Theorem~\ref{thm:LBT} for $d-2\leq i\leq d-1$, equality for some $1\leq i\leq d-1$ implies equality for $i=1$, so
it is enough to consider the case $i=1$.
From Kalai's monotonicity (Theorem~\ref{thm:kalaiMonotonicity}) and our assumption $g_2(P)=0$, it follows that $g_2(F)=0$.
As $F$ is simplicial of dimension $\ge 4$, Kalai's \cite[Thm.~1.1(ii)]{Kal87}
says that $F$ is stacked, thus $\Delta$ is a simplicial $(d-1)$-sphere. Since $g_2(\Delta)=0$, by \cite[Thm.~1.1(ii)]{Kal87} again, $\Delta$ is stacked, as desired.
In particular, as $f(P)=f(\Delta)-(0,\ldots,0,s,s)=f(S(d,n))-(0,\ldots,0,s,s)$ we conclude that
$f(P)=f(S(d,n,s))$.
\end{proof}
For $d=4$, $F$ need not be stacked. For example, the pyramid over any simplicial $3$-polytope is a minimizer. We obtain the following characterization of minimizers.
\begin{theorem}[(Characterization of minimizers for $d=4$)]\label{thm:LBT=4}
Let $P\in \Pl(4,n,s)$,
and keep the notation of Theorem~\ref{thm:LBT>4}.
%Let $\Delta_F$ be the polyhedral complex corresponding to the prime factorization of $F$, and $\Delta$ be the refinement of the boundary complex $\partial P$ of $P$ obtained by refining $F$ by $\Delta_F$.
Assume there is some $1\le i\le d-1$ for which $f_i(P)=f_i(S(d,n,s))$.
Then, the prime factors in the prime decomposition of $\Delta$ are either $d$-simplices with no facet contained in $|F|$, or pyramids over prime factors of $F$.
\end{theorem}
In order to prove this theorem we first need to show generic $d$-rigidity for the $1$-skeleton of a much larger class of complexes.
Let $\C_{k}$ be the family of homology $k$-balls $\Delta$ such that:
\begin{itemize}
\item the induced subcomplex $\Delta[I]$ on the set $I$ of internal vertices has a connected $1$-skeleton, and
\item for any edge $e$ in the boundary complex $\partial\Delta$, there exists a $2$-simplex $T$, $e\subset T$, such that $T$ has a vertex in $I$.
\end{itemize}
%%%%%%%%%%%%%
Note that any homology $k$-ball $\Delta$ whose boundary $\partial\Delta$ is an induced subcomplex is in $\C_k$\footnote{This follows from Alexander duality, however we preferred to give an easier argument.}.
Indeed, as $\partial\Delta$ is induced, any facet of $\Delta$ intersects $I$ nontrivially. Assume by contradiction that the graph $\Delta[I]_{\leq 1}$ is disconnected, say equals the disjoint union of nontrivial graphs $G_1$ and $G_2$. As $\Delta$ is facet-connected it has facets $F_1$ and $F_2$ whose intersection $S$ has codimension $1$ and $F_i$ has a vertex in $G_i$ for $i=1,2$. Then $S$ is disjoint from $I$, so $S\in \partial\Delta$. This is a contradiction as $S$ is contained in two facets of $\Delta$, not in one.
In particular,
for $P\in \Pl(d,n,s)$,
the simplicial complex $P'=\partial P-\{F\}$ is in $\C_{d-1}$.
\begin{lemma}\label{lem:rigidity_C_d}
Let $d\ge 4$. The $1$-skeleton of any $\Delta\in \C_{d-1}$ is generically $d$-rigid. Thus, $f_1(\Delta)\ge df_0(\Delta) - \binom{d+1}{2}$.
\end{lemma}
\begin{proof}
The proof follows from Kalai's proof of the classical LBT. Apply \cite[Prop. 6.4]{Kal87} with the tree $T$ there be a spanning tree of the $1$-skeleton of $\Delta[I]$.
%; for completeness we give a brief description.
%For any $v\in I$, $\lk_{\Delta}(v)$ is a homology sphere of dimension $\ge 2$, hence its $1$-skeleton is generically $(d-1)$-rigid. Then the Cone Lemma (cf. \cite[Sec.~3]{Kal87}) says the $1$-skeleton of $\st_{\Delta}(v)$ is generically $d$-rigid. From the Gluing Lemma
%\cite{Asi-Roth2}
%\cite[Lem.~11.1.9]{Whi96}
%and the first item in the definition of $\C_{d-1}$, it then ensues that the $1$-skeleton $G$ of $\cup_{v\in I}\st_{\Delta}(v)$ is also generically $d$-rigid. The definition of $\C_{d-1}$ finally gives that $G$ is the $1$-skeleton of $\Delta$.
\end{proof}
\begin{proof}[Proof of Theorem~\ref{thm:LBT=4}]
Consider a prime factor $L$ of $\Delta$.
Then $L$ is a $4$-polytope
which is $2$-simplicial so it has a generically $4$-rigid $1$-skeleton by \cite[Thm. 8.6]{Whi84}.
As $g_2(P)=0$, the $1$-skeleton of $L$, denoted by $G$, must be generically $4$-stress free. Thus, $g_2(L)=0$.
If $L$ does not contain a facet in $\Delta_F$, then $L$ is simplicial, with $g_2(L)=0$, hence is stacked by \cite[Thm.~1.1]{Kal87}. Being also prime, $L$ is a $4$-simplex.
Assume then that $L$ contains a facet $F''$ contained in $|F|$, so $(L,F'')$ is an ASP-pair. If $L$ has a unique vertex outside $F''$, then $L$ is a pyramid over a prime factor of $F$ and we are done. Assume the contrary, so there is an edge $vu\in G$ with $v,u\notin F''$ (for concreteness, taking $v,u$ to be the highest two vertices of $L$ above the hyperplane of $F$ works).
First we show that $vu$ satisfies the link condition $\lk_L(v)\cap \lk_L(u) = \lk_L(vu)$, which guarantees that contracting the edge $vu$ in the simplicial complex $\partial L-\{F''\}$ results in $\tilde{\Delta}\in \C_3$; see e.g.\cite[Prop.2.4]{Nevo-Novinsky}\footnote{To apply \cite[Prop.~2.4]{Nevo-Novinsky}, phrased for homology spheres, simply cone the boundary of the homology ball $\Delta$ to obtain a homology sphere.}.
Indeed, if $vu$ fails the link condition it means that $vu$ is contained in a missing face $M$, with $3$ or $4$ vertices. Now, $M$ cannot have $4$ vertices as $L$ is prime. If $M=vuz$ then $uz$ is an edge of $L$ \emph{not} in $\lk_L(v)$. Since $\lk_L(v)$ is a homology $2$-sphere (thus, a simplicial $2$-sphere), its $1$-skeleton is generically $3$-rigid. Consequently, the $1$-skeleton of $\st_L(v)$ is
generically $4$-rigid, and adding $uz$ to it yields a $4$-stress in $G$, a contradiction.
Let $m$ be the number of vertices in the cycle $\lk_L(vu)$, then
$f_1(\tilde{\Delta})=f_1(L)-m-1$ and $f_0(\tilde{\Delta})=f_0(L)-1$, which implies that $g_2(L)=g_2(\tilde{\Delta})+(m-3)$.
If $m>3$, then applying Lemma~\ref{lem:rigidity_C_d} to $\tilde{\Delta}$ yields $g_2(L)>0$, a contradiction. So assume $m=3$.
Denote by $x,y,z$ the vertices of $\lk_L(vu)$.
If the triangle $xyz\in L$, then, as $L$ is prime, both tetrahedra $xyzv, xyzu$ are faces of $L$, so $L$ is the $4$-simplex $xyzuv$, a contradiction (as it has a facet $F''$ in $F$).
We are left to consider the case $xyz\notin L$.
The argument here is inspired by Barnette~\cite[Thm.~2]{Bar71}.
In this case, the $3$-ball formed by the join $vu * \partial(xyz)$ is an induced subcomplex of $\partial L-\{F''\}$. Now replace it by $\partial(vu) * xyz$ (this is a bistellar move) to obtain from $\partial L-\{F''\}$ the complex $\Delta"$.
Clearly $\Delta"$ is a homology $3$-ball, and any edge on its boundary is part of a $2$-simplex with an internal vertex (just take the same one as in $\partial L-\{F''\}$). To show $\Delta"\in\C_3$ we are left to show that the graph on the internal vertices $I$ of $\Delta"$ is connected.
Assume not, namely removing the edge $uv$ disconnects the induced graph on $I$ in $\partial L-\{F''\}$. In particular, $x,y,z\in F''$. But $xyz\notin L$, so $xyz$ is a missing face of $F''$, contradicting that $F''$ is a prime factor of $F$.
We conclude that $\Delta"\in C_3$, thus, by Lemma~\ref{lem:rigidity_C_d}, $\Delta"\cup\{vu\}$ has a nonzero $4$-stress. However, the $1$-skeletons of $\Delta"\cup\{vu\}$ and of $L$ are equal graphs
so $g_2(L)>0$, a contradiction.
The proof is then complete.
\end{proof}
%\begin{remark}
%The above shows that, for any $k\ge 3$, the lower bounds of Theorem~\ref{thm:LBT} are valid for any complex in $\C_k$, and the minimizers in $\C_k$ are exactly the complexes $\partial P-\{F\}$ described in Theorems~\ref{thm:LBT>4} and \ref{thm:LBT=4}.
%\end{remark}
\section{An upper bound theorem for almost simplicial polytopes}
\label{sec:UBT}
%\subsection*{Preliminaries}
Throughout this section, we let $P\in \Pl(d,n,s)$ denote an almost simplicial polytope, $(P,F)$ the ASP-pair, and $P'=\partial P \setminus \{F\}$ the corresponding shellable simplicial $(d-1)$-ball. Recall that $\partial P'=\partial F$ is an induced subcomplex of $P'$.
\begin{comment}
Then $f_{k-1}(P)$ \cite[Eq.~8.25]{Zie95} can expressed as
\begin{align}
f_{k-1}(P)&=\begin{cases}
\displaystyle \sum_{i=0}^{\floor{d/2}}h_i(P){d-i\choose k-i}+h_{d-i}(P){i\choose k-d+i},&\text{if $d$ is odd};\\
\displaystyle \sum_{i=0}^{d/2-1}h_i(P){d-i\choose k-i}+ h_{d-i}(P){i\choose k-d+i}+h_{d/2}(P){d-d/2\choose k-d/2},&\text{if $d$ is even}.
\end{cases}\label{eq:fk_hk}
\end{align}
\end{comment}
\subsection{ASP generalization of cyclic polytopes}
\label{subsec:almostCycPol}
The {\it moment curve} in $\mathbf{R}^d$ is defined by $t\mapsto(t,t^2,\ldots,t^d)$ for $t\in \mathbf{R}^d$, and the convex hull of any $n$ points on it gives, combinatorially, the cyclic polytope $C(d,n)$; see, for instance, \cite[Example 0.6]{Zie95}. We extend this construction by considering curves $x(t)$ of the form $(t,t^2,\ldots,t^{d-r},p_1(t),\ldots,p_r(t))$, where $p_i(t)$ are continuous functions in $t$ for $i=1,\ldots,r$.
Later, a special choice of the curve $x(t)$ and points on it will give, by taking the convex hull, our maximizer polytope $C(d,n,s)$.
We let $V(t_1,\ldots,t_{l})$ denote the {\it Vandermonde determinant} on variables $t_1,\ldots,t_l$.
\[V(t_1,\ldots,t_{l}):=\begin{vmatrix}
1 &1&\cdots &1\\
t_1&t_2&\cdots &t_{l}\\
t_1^2&t_2^2&\cdots &t_{d}^l\\
\vdots&\vdots&\cdots&\vdots\\
t_1^{l-1}&t_2^{l-1}&\cdots &t_{l}^{l-1}
\end{vmatrix}
= \prod_{1\le i<j\le l}(t_j-t_i).
\]
\begin{lemma}\label{lem:neighborly}
Consider the curve $x(t)$. Then the following holds:
\begin{enumerate}
\item Any $d-r+1$ points on the curve $x(t)$ are affinely independent.
\item For any $n$ distinct numbers $t_1,\ldots,t_n$, the polytope $Q=\conv (\{x(t_1),\ldots,x(t_{n})\})$ is $(d-r-1)$-simplicial.
%\item Any $d-r+1$ points define a $(d-r)$-flat in $\mathbf R^d$.
\item The polytope $Q$ is $\floor{(d-r)/2}$-neighbourly.
\end{enumerate}
\end{lemma}
\begin{proof}Consider $n$ real numbers $t_1<\ldots<t_{n}$ and the corresponding points $x(t_i)$.
% and the matrix
\begin{comment}
\[\begin{pmatrix}
1 &1&\cdots &1&1\\
t_1&t_2&\cdots &t_{d}&t\\
t_1^2&t_2^2&\cdots &t_{d}^2&t^2\\
\vdots&\vdots&\cdots&\vdots&\vdots\\
t_1^{d-r}&t_2^{d-r}&\cdots &t_{d}^{d-r}&t^{d-r}\\
p_1(t_1)&p_1(t_2)&\cdots &p_1(t_{d})&p_1(t)\\
\vdots&\vdots&\cdots&\vdots&\vdots\\
p_r(t_1)&p_r(t_2)&\cdots &p_r(t_{d})&p_r(t)
\end{pmatrix}.\]
\end{comment}
From any $d-r+1$ points $x(t_{i_1}),\ldots, x(t_{i_{d-r+1}})$,
we form a matrix by considering them as its columns, in this same order, and adding a row of ones as the first row. The top
$(d-r+1)\times (d-r+1)$
square submatrix of this martix has
determinant $V(t_{i_1},\ldots,t_{i_{d-r+1}})$ which is nonzero, which in turn implies the first assertion.
The second assertions follows immediately from the first.
To prove the third assertion proceed as in \cite[Sec.~4.7]{Gru03}. Consider a set $S_k=\{x(t_{i_j}):j=1,\ldots,k\}$, $1\le i_j\le n$,
%of $\ver Q$
with $k\le \floor{(d-r)/2}$ and the polynomial
\[\beta(t)=\prod_{i=1}^k(t-t_{i_j})^2=\beta_0+\beta_1t+\cdots+\beta_{2k}t^{2k}.\]
Let $b=(\beta_1,\ldots,\beta_{2k},0,\ldots,0)$ be a vector in $\mathbf{R}^d$ and $H=\{x\in \mathbf{R}^d: x\cdot b=-\beta_0\}$ a hyperplane in $\mathbf{R}^d$. Here $\cdot$ denotes the dot product of vectors.
All the points in $S_k$ are clearly contained in $H$, and for any other $x(t_{l})\in \{x(t_1),\ldots,x(t_n)\}\setminus S_k$ we have $x(t_l)\cdot b=-\beta_0+\beta(t_l)>-\beta_0$.
Thus, $S_k$ is the vertex set of a simplex face of $Q$.
\end{proof}
Let $n$ and $s$ be fixed integers with $n>d+s$ and $s\ge 0$ and
consider the curve $y(t)=(t,t^2,\ldots,t^{d-1},p(t))$, where \[p(t):=(n-1)^{(t-1)(d-1)}t(t+1)\cdots(t+d+s-1).\]
Let $t_i=-s-d+i$ for $i=1,\ldots,n$
and let $C(d,n,s):=\conv (\{y(t_1),\ldots,y(t_{n})\})$. Also, let $T=\{t_{i}:i=1,\ldots,n\}$, $I=\{t_{i}:i=1,\ldots,d+s\}$ and $y(S):=\{y(t_{i}):t_i\in S\}$ for $S\subset T$.
The following proposition collects a number of properties of the $d$-polytope $C(d,n,s)$.
\begin{proposition}\label{prop:C(d,n,s)} The $d$-polytope $C(d,n,s)$ ($n>d+s$) satisfies the following properties.
\begin{enumerate}
\item $C(d,n,s)\in \Pl(d,n,s)$.
\item {\bf Gale's evenness condition}: A $d$-subset $S_d$ of $\ver C(d,n,s)$ such that $S_d\not \subset I$ forms a simplex facet if and only if, for any two elements $u,v\in T\setminus S_d$, the number of elements of $S_d$ between $u$ and $v$ on the curve $y(t)$ is even.
\end{enumerate}
\end{proposition}
\begin{proof}
(1)
We first show that the first $d+s$ vertices span a facet $F$. Let $z=(z_1,\ldots,z_d)\in \mathbf{R}^d$ and let
\[D((t_1,t_2,\ldots,t_d);z):=\begin{vmatrix}
1 &1&\cdots &1&1\\
t_1&t_2&\cdots &t_{d}&z_1\\
t_1^2&t_2^2&\cdots &t_{d}^2&z_2\\
\vdots&\vdots&\cdots&\vdots&\vdots\\
t_1^{d-1}&t_2^{d-1}&\cdots &t_{d}^{d-1}&z_{d-1}\\
p(t_1)&p(t_2)&\cdots &p(t_{d})&z_d\\
\end{vmatrix}.\]
Let $D(z):=D((t_1,t_2,\ldots,t_d);z)$ and consider the hyperplane $H_D:=\{z\in \mathbf{R}^d: D(z)=0\}$. The points $y(t_i)$ ($i=1,\ldots,d+s$) are all contained in $H_D$,
since the last row of $D(z)$ vanishes at all these points $y(t_i)$; recall that $p(t_i)=0$ for $i=1,\ldots,d+s$. Also, by \cref{lem:neighborly}, any $d$ of these points $y(t_i)$ ($i=1,\ldots,d+s$) are affinely independent. So in fact $H_D$ equals the affine span of the points $y(t_i)$ ($i=1,\ldots,d+s$).
Let $y(t^*)\in \ver C(d,n,s)\setminus y(I)$, then $D(y(t^*))=p(t^*)V(t_1,\ldots,t_d)>0$ since $p(t^*)>0$ and $V(t_1,\ldots,t_d)>0$.
Thus, $F$ is a facet of $C(d,n,s)$.
We now show that every other facet is a simplex.
Consider any $(d+1)$-set $\{t_{i_1}<\ldots <t_{i_d}<t_{i_{d+1}}=t^*\}\subset T$ not contained in $I$. Thus, $t^*\in T\setminus I$.
Consider the determinant $E(z):=D((t_{i_1},t_{i_2},\ldots,t_{i_d});z)$.
\begin{comment}
\[E(y):=\begin{vmatrix}
1 &1&\cdots &1&1\\
t_{i_1}&t_{i_2}&\cdots &t_{i_d}&y_1\\
t_{i_1}^2&t_{i_2}^2&\cdots &t_{i_d}^2&y_2\\
\vdots&\vdots&\cdots&\vdots&\vdots\\
t_{i_1}^{d-1}&t_{i_2}^{d-1}&\cdots &t_{i_d}^{d-1}&y_{d-1}\\
p(t_{i_1})&p(t_{i_2})&\cdots &p(t_{i_d})&y_d\\
\end{vmatrix}.\]
\end{comment}
The hyperplane $H_E:=\{z\in \mathbf{R}^d: E(z)=0\}$ contains all the points $y(t_{i_j})$ ($j=1,\ldots,d$).
We need to show that $E(y(t^*))\ne 0$.
Note that $p(t)=0$ for $t\in I$ and $p(t)>0$ for $t\in T\setminus I$. Also, note that $|t_a-t_b|\le n-1$ for $t_a,t_b\in [-s-d+1,-s-d+n]$. For the sake of clarity assume $d$ is odd; the case of even $d$ is analogous. Computing $E(y(t^{*}))$ by expanding w.r.t. the last row gives
\begin{align*}\left(p(t^{*})V(t_{i_1},\ldots,t_{i_d})-p(t_{i_d})V(t_{i_1},\ldots,t_{i_{d-1}},t^*)\right)+\cdots\\
\quad +\left(p(t_{i_2})V(t_{i_1},t_{i_3}\ldots,t^{*})-p(t_{i_1})V(t_{i_2},\ldots,t^*)\right).
\end{align*}
The definition of $p(t)$ implies that each pair-summand is
nonnegative and the first pair-summand is
positive, and so the determinant is positive.
Indeed, for $j>1$, if $p(t_{i_j})=0$ then also $p(t_{i_{j-1}})=0$ and the corresponding pair-summand vanishes. Otherwise, let
$V(j):=V(t_{i_1},\ldots t_{i_{j-1}},t_{i_{j+1}},\ldots,t_{i_{d+1}})$ for short.
Then,
\[p(t_{i_j})V(j)\ge (n-1)^{(d-1)(t_{i_j}-1)}\prod_{l=0}^{d+s-1}(t_{i_{j-1}}+l) \frac{V(j-1)}{(n-1)^{d-1}}\ge p(t_{i_{j-1}})V(j-1).\]
This completes the proof of the first assertion.
(2)
Consider a set $S_d=\{t_{i_1}<\ldots <t_{i_d}\}\not \subset I$.
%Fix a vertex in $S_d$, say $y(t_{i_j})$, and consider
%%\textcolor{blue}{
%real numbers $t^*$ and $t^{**}$ such that $t^*,t^{**}\in T$ and $t_{i_{j-1}}<t^*<t_{i_j}<t^{**}<t_{i_{j+1}}$. If $t_{i_j}=t_{i_1}=-s-d-1$ we only look at $t^{**}$, while if $t_{i_j}=t_{i_d}=-s-d+n$ we only look at $t^{*}$. We show that $E(y(t^*))$ and $E(y(t^{**}))$ have different signs, which in turn implies the second assertion of the proposition.
%%%%%%%%%
Let $t^*\in T$, $t_{i_{j-1}}<t^*<t_{i_j}$ (include also the cases $t^*<t_{i_1}$ with $j=1$ and $t_{i_d}<t^*$ where we put $j=d+1$).
%%%%%%%%
%First consider $E(y(t^*))$.
From the above reasoning we see that if the column $y(t^*)$ in the determinant $E(y(t^*))$ is placed between the columns $y(t_{i_{j-1}})$ and $y(t_{i_j}) $ then the resulting determinant is positive. To achieve this, we swap $d-j+1$ times the column $y(t^*)$, which gives that the sign of $E(y(t^*))$ is $(-1)^{d-j+1}$.
%Analogously, we get that the sign of $E(y(t^{**}))$ is $(-1)^{d-j}$.
Consequently, on the curve $y(t)$, between $[-s-d+1,-s-d+n]$, the determinant $E(y(t^*))$ changes sign whenever the variable passes through one of the values $t_{i_{j}}$ ($i=1,\ldots,d$), and we are done.
\end{proof}
A polytope $C(d,n,s)$ will be called {\it almost cyclic}. Having established in \cref{lem:neighborly} that $C(d,n,s)$ is $\floor{(d-1)/2}$-neighbourly, we can compute its $h$-vector, in steps.
Recall that $P'=\partial P \setminus \{F\}$.
\begin{proposition}\label{prop:neighbourlyASP} Let $P\in \Pl(d,n,s)$ be $\floor{(d-1)/2}$-neighbourly, and $(P,F)$ the ASP-pair. Then,
%If $d$ is odd, then
\begin{align*}
h_{k}(P')&= {n-d-1+k\choose k},&\text{if $0\le k\le \floor{(d-1)/2}$};\\
h_{d-k}(P')&= {n-d-1+k\choose k}-{s+k-1\choose k},&\text{if $1\le k\le \floor{(d-1)/2}$}.
\end{align*}
%If $d$ is even, then
%\begin{align*}
%h_{k}(P)&= {n-d-1+k\choose k},&\text{if $0\le k\le \floor{(d-1)/2}$};\\
%h_{d-k}(P)&= {n-d-1+k\choose k}-{s+k-1\choose k},&\text{if $1\le k\le \floor{d/2}-1$}.
%\end{align*}
\end{proposition}
\begin{proof} First note that $f_{k-1}(P')= {n\choose k}$ for $k\le \floor{(d-1)/2}$. Thus, it follows that
%From \cref{eq:fk_hk}
\[h_k(P')=\sum_{i=0}^k(-1)^{k-i}{d-i\choose k-i}{n\choose i}={n-d-1+k\choose k}.\]
%To see that the last equality holds, consider the known identity (cf.~\cite[p.~149]{Gru03})
%\[\sum_{i=0}^k(-1)^i{d-i\choose k-i}{n\choose i}=\sum_{i=0}^k(-1)^i{d-i\choose d-k}{n\choose i}={d-n\choose k} \quad \text{(for $0\le k\le d$)}\]
%together with the basic fact that ${d-n\choose k}=(-1)^k{n-d-1+k\choose k}$.
We now consider the remaining values of $k$.
Using that $F$ is a neighborly simplicial $(d-1)$-polytope
%and applying the $g$-theorem for shellable simplicial polytopes \cite[Cor.~8.38]{Zie95} to $F$,
we obtain that
$g_{k}(F)= {d+s-(d-1)+k-2\choose k}$, for $0\le k\le \floor{(d-1)/2}$. Thus, from \cref{eq:DSEq-hP} of Proposition~\ref{prop:Dehm-Somerville} it follows, for $1\le k\le \floor{(d-1)/2}$, that
\[h_{d-k}(P')= {n-d-1+k\choose k}-{s+k-1\choose k}.\]
\end{proof}
Observe that, for even $d$, being $\floor{(d-1)/2}$-neighbourly does not determine the value of $h_{d/2}(P')$.
\begin{comment}
\begin{question}
\eran{It will be surprising if the answer is NO. Let's check Billera-Lee to get an idea for a construction.}
For $s>0$ let $P$ be a $\floor{(d-1)/2}$-neighbourly, almost simplicial $d$-polytope of even $d$. Does there exist a $P$ with $h_{d/2}(P)<{n-d-1+d/2\choose d/2}-{s+d/2-1\choose d/2}$? This amounts to proving that there exists a $P$ with $f_{d/2-1}(P)<{n\choose d/2}-g_k(F)$, with $g_{k}(F)={s+d/2-1\choose d/2}$.
\end{question}
\end{comment}
With the help of Gale's evenness condition we can compute the number of facets of $C(d,n,s)$, and together with \cref{prop:neighbourlyASP} and \cref{eq:fk_hk}, we can compute $h_{d/2}(C(d,n,s))$ for any even $d$ as well.
\begin{proposition}\label{prop:facetsC(d,n,s)}
For the ASP-pair $(C(d,n,s),F)$ with $d$ even, consider
the simplicial ball $C':=C(d,n,s)-\{F\}$. Then
%The number of facets of a $d$-polytope $C(d,n,s)$ with $n>d+s$ is
%\textcolor{blue}{
\begin{align*}
f_{d-1}(C')=\left(\binom{n-d/2 -1}{d/2} + \sum_{i=0}^{d/2 -1} 2 \binom{n-d-1+i}{i}\right) - \binom{s+ d/2}{d/2}.
%f_{d-1}(C(d,n,s))=\begin{cases}
%\displaystyle{n-d/2\choose d/2}+ {n-d/2\choose d/2}-{s+d/2\choose d/2}+1&\text{if $d$ is even,}\\
%\displaystyle 2{n-\floor{d/2}-1\choose \floor{d/2}}-{s+\floor{d/2}\choose \floor{d/2}}+1&\text{if $d$ is odd}.
%\end{cases}
\end{align*}
\end{proposition}
\begin{proof}
%There is exactly one nonsimplex facet in $C(d,n,s)$, counted by the one in the formula.
The counting argument for the facets of $C'$, based on Gale evenness, goes as in the proof of the number of facets of cyclic polytopes (cf.~\cite[Cor.~8.28]{Zie95}),
%\cite[Thm.~5.4.5]{Mat02a}),
with the difference that we discard the Gale $d$-tuples formed solely by the first $d+s$ vertices, thus we discard exactly $\binom{s+ d/2}{d/2}$ of them.
\end{proof}
\begin{corollary}\label{thm:(k-1)C(d,n,s)}
The $h$-numbers of $C'$ are given by:
%The $k$-th entry of the $h$-vector of an almost cyclic $d$-polytope $C(d,n,s)$ with $n:=f_0(P)$ and $n>d+s$ is given by the expression:
\begin{align*}
h_{k}(C')&={n-d-1+k\choose k},&\text{if $0\le k\le \floor{(d-1)/2}$};\\
h_{d-k}(C')&= {n-d-1+k\choose k}-{s+k-1\choose k},&\text{if $1\le k\le \floor{d/2}$}.
\end{align*}
\end{corollary}
\begin{proof} The case of odd $d$ was already established by \cref{prop:neighbourlyASP} since $C(d,n,s)$ is $\floor{(d-1)/2}$-neighbourly.
For the case of even $d$ it remains to compute $h_{d/2}(P)$.
Equating the corresponding expression in Proposition \ref{prop:facetsC(d,n,s)} with the expression of $f_{d-1}$ in \cref{eq:fk_hk}, after substituting the known values of $h_k$ for $k\neq d/2$, gives
\begin{align*}h_{d/2}(C')&={n-d/2-1\choose d/2} + \sum_{i=0}^{d/2 -1}{s+i-1\choose i} -{s+d/2\choose d/2}\\
\quad &= {n-d/2-1\choose d/2} -{s+d/2-1\choose d/2}
,
\end{align*}
as desired.
\end{proof}
\begin{comment}
\section{Theorem 3.10 from AS's paper}
Let $h_{k}(\Delta,\partial \Delta):=h_{k}(\Delta)+h_{k-1}(\partial \Delta)-h_{k}(\partial \Delta)$, where $\Delta$ is a simplicial $d$-ball and $\partial \Delta$ is the boundary. For an ASP $d$-polytope $P$, we have $P=\Delta$, $f_0(P)=n$, $F=\partial \Delta$, $f_0(F)=d+s$ and $h_{d-k}(\Delta)=h_{k}(\Delta,\partial \Delta)$. Thus, the inequalities of Theorem 3.10 can be written as:
\begin{align*}
h_{k}(\Delta)&\le {n+d-k-2\choose d+1-d+k}={n+d-k-2\choose 1+k},&\text{if $0\le k\le \floor{(d-1)/2}$}\\
h_{d-k}(\Delta)&\le {n-d+k-2\choose k} -{s+k-2\choose k},&\text{if $0\le k\le \floor{(d-1)/2}$}.
\end{align*}
\end{comment}
\subsection{An upper bound theorem for almost simplicial polytopes}
We are now in a position to state an upper bound theorem for almost simplicial polytopes $P\in \Pl(d,n,s)$.
% in terms of $d,n,s$.
\begin{theorem}[(UBT for ASP)]\label{thm:UBT_ASP}
Any almost simplicial polytope $P\in \Pl(d,n,s)$ satisfies
\begin{align}
h_{k}(P')&\le {n-d-1+k\choose k},&\text{if $0\le k\le \floor{(d-1)/2}$};
\label{eq:hk}\\
h_{d-k}(P')&\le {n-d-1+k\choose k}-{s+k-1\choose k},&\text{if $1\le k\le \floor{d/2}$}.
\label{eq:hd-k}
\end{align}
Thus,
\[f_{i-1}(P)\le f_{i-1}(C(d,n,s)\quad \text{for $i=1,2,\ldots,d$},\]
for the almost cyclic $d$-polytope $C(d,n,s)$. Equality for some $f_{i-1}$ with $\floor{(d-1)/2}\le i\le d$ implies that $P$ is $\floor{(d-1)/2}$-neighbourly.
\end{theorem}
\begin{proof}[Proof of \cref{thm:UBT_ASP} via {\cite[Thm.~3.9]{AdiSan14}}]
The inequalities on $h_k(P')$ hold for $0\le k\le d-1$ by \cite[Thm.~3.9]{AdiSan14}, as $P'$ is a special case of a homology ball whose boundary is an induced subcomplex.
From Corollary~\ref{thm:(k-1)C(d,n,s)} and \cref{eq:fk_hk} the inequality $f_{i-1}(P)\le f_{i-1}(C(d,n,s)$ follows.
%\textcolor{blue}{(for $i=d$, add one to both sides to take care of the nonsimplex facets)}.
Equality for some $f_{i-1}$ with $d\ge i\ge \floor{(d-1)/2}$ implies, by \cref{eq:fk_hk}, the equality $h_{k}(P')={n-d-1+k\choose k}$ for all $0\le k\le \floor{(d-1)/2}$, and thus, again by \cref{eq:fk_hk},
that $P$ is $\floor{(d-1)/2}$-neighbourly.
\end{proof}
\begin{remark}[More maximizers.]\label{rem:MoreMaximizers}
As is the case with neighborly polytopes, we expect that there are many combinatorially distinct ASPs achieving the upper bounds in the UBT for ASP. Here we sketch another such construction, based on a certain perturbation of the Cayley polytope constructed by Karavelas and Tzanaki \cite[Sec.5]{KarTza12}: there, two neighborly $(d-1)$-polytopes $P_1$ and $P_2$ are placed in parallel hyperplanes in $\mathbf{R}^d$ so that the Cayley polytope $P=\conv(P_1\cup P_2)$ is $\floor{\frac{d-1}{2}}$-neighborly and all $\floor{d/2}$-subsets with a vertex in $P_1$ and a vertex in $P_2$ form faces of $P$. Let $n=f_0(P)$ and $d+s=f_0(P_2)$.
Thus, for $d$ odd, any small enough perturbation of the vertices of $P_1$ into general position will change $P$, by considering the new convex hull, into $Q\in \Pl(d,n,s)$ ($(Q,P_2)$ is the ASP-pair) with $f(Q)=f(C(d,n,s))$; so $Q$ is a maximizer.
For $d$ even, in order for $Q$ to be a maximizer, we need the small perturbation be such that all $d/2$-subsets of $\ver(P_1)$ become faces of $Q$. To achieve this, we recall more from the construction of \cite{KarTza12}, and make a variation:
consider the images of the functions $\gamma_1(t,z_1,z_2)=(t,z_1 t^{d-1},t^2,t^3,\ldots,t^{d-2},z_2 t^d)\subset \mathbf{R}^d$ and
$\gamma_2(t,z_1)=(z_1 t^{d-1},t,t^2,\ldots,t^{d-2},-1)\subset \mathbf{R}^{d-1}\times\{-1\} \subset \mathbf{R}^d$.
The vertices of $P_1$ (resp. $P_2$) are on appropriate locations on the curve $\gamma_1(t,z_1^*,0)\subset \mathbf{R}^{d-1}\times\{0\}$ (resp. $\gamma_2(t,z_1^*)$, for small enough fixed $z_1^*>0$. It is possible to show, by appropriate determinant computation, that for a small enough fixed $z_2^*>0$, perturbing the vertices $\gamma_1(t_i,z_1^*,0)$ of $P_1$ to $\gamma_1(t_i,z_1^*,z_2^*)$ makes all $d/2$-subsets of $P_1$ faces of $Q$; so $Q$ is a maximizer.
\end{remark}
We proceed by producing an alternative and elementary proof of the UBT for ASP, via shelling. This will take the rest of this section.
Our proof follows ideas from the proof of the classical UBT by McMullen, cf.~\cite[Sec.8.4]{Zie95}, and from a recent work of Karavelas and Tzanaki \cite{KarTza12}. The key new ingredient is Lemma~\ref{lem:key} below, for which we need some preparation.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%The proof of \cref{thm:UBT_ASP} is decomposed into a couple of lemmas.
Let $P\in \Pl(d,n,s)$ and $(P,F)$ an ASP-pair. Let $Q$ be a polytope obtained from $P$ by stacking a new vertex $y$ beyond $F$. The $d$-polytope $Q$ is simplicial. The set of proper faces of $Q$ is the disjoint union of the faces of the complex $P':=\partial P -\{F\}$ and the faces of $Q$ which contain $y$. Consequently, for all $k\ge0$,
\begin{equation}\label{eq:hQ=P+F}
h_k(Q)=h_k(P')+h_{k-1}(F).
\end{equation}
Recall the star $\st_{\mathcal{C}}(F)$ of a face $F$ in a polytopal complex $\mathcal C$ is the polytopal subcomplex generated by all the faces of $\mathcal{C}$ containing $F$.
\begin{comment}Following \cite[p.~241]{Zie95}, we say that a facet $K$ of $P$ is {\it visible} from the point $w$ if $w$ belongs to the open halfspace determined by $\aff K$ which does not meet $P$. Also, the point $w$ is {\it beyond} a face $G$ of $P$ if the facets of $P$ containing $G$ are precisely those that are visible from $w$.
\end{comment}
\begin{comment}
Next we establish a particular shelling order of the polytope $Q$, but first we need an observation from \cite{Cou06}.
\begin{lemma}{\cite[Lem.~1]{Cou06}}\label{lem:shellingOrderCou}
Let $\mathcal{C}$ be a shellable polytopal complex and let $F_1,\ldots,F_q,H_{1},\ldots,H_p$ be a shelling order of $\mathcal{C}$. If $F'_1,\ldots,F_q'$ is a shelling order of the complex $F_1\cup\cdots\cup F_q$ then the ordering $F'_1,\ldots,F'_q,H_{1},\ldots,H_p$ is also a shelling of $\mathcal{C}$.
\end{lemma}
\end{comment}
We will use a line shelling of $Q$ with some special properties:
\begin{lemma}\label{lem:shellingOrder}
Let $(P,F)$ be an ASP-pair, and $v\in F$ a vertex.
Then we can choose the aforementioned vertex $y$ beyond $F$ such that, for $(P,F,y,Q)$ as above, there is a line shelling of $Q$ which shells the star of $y$ first and then proceeds to shell the rest of the star of $v$. \end{lemma}
%We proceed as in the proofs of \cite[Thm.~8.12, Cor.~8.13]{Zie95}.
\begin{proof}
%A shelling order as guaranteed in Lemma~\ref{lem:shellingOrder} can be achieved by a line shelling:
For an oriented line $\ell$ that shells $P$, with $F$ being first followed by the rest of the facets in the star of $v$ (cf.~\cite[Thm.~8.12, Cor.~8.13]{Zie95}\footnote{Here we use the extension of the notion of shellability to polyhedral complexes.}), place $y$ on $\ell$ beyond $F$ to make $Q$. Now perturb $\ell$, intersecting $Q$ near $y$, to obtain the desired line shelling.
%There is a shelling order $S(P)$ of $P$ which ends with the facets of $\st(v)$ and has the facet $F$ as the last facet. Indeed, choose a point $w_v$ beyond the vertex $v$ and a point $w_F$ beyond the facet $F$. Let $\ell$ be the line determined by $w_v$ and $w_F$ such that $\ell$ hits the interior of $P$. Perturb $\ell$ to general position if necessary (cf.~\cite[Lem.~3.2]{Zie95}). Orienting the line $\ell$ from $w_F$ to $w_v$ produces a shelling which starts with the facets of $\st(v)$ and has $F$ as the first facet. The reverse order of this shelling is the desired shelling $S(P)$ of $P$.
%Then the shelling $S(P)$ without $F$, denoted $S'(P')$, is a shelling of the complex $P'$ (recall $P':=\partial P-\{F\}$).
%The polytope $Q$ admits a shelling order which ends with the facets $H_{1},\ldots,H_p$ of the open star of $y$; just consider a point $w_y$ beyond $y$ and a line $\ell$ in general position which passes through $y$. The complex obtained from $Q$ by removing the facets containing $y$ is $P'$. From \cref{lem:shellingOrderCou} it follows that the concatenation of $S'(P)$ and
%the facets $H_{1},\ldots,H_p$ is a shelling order of $Q$ whose reverse order satisfies the conditions of the lemma.
\end{proof}
%\begin{remark}
%\end{remark}
\begin{comment}
We conclude that for $(P,F,y,Q)$ as above, and any vertex $v\in F$, there is a shelling of $Q$ (not necessarily a line shelling) that shells the star of $y$ first and then proceeds to shell the rest of the star of $v$.
\end{comment}
Consider any vertex $v\in \ver Q$ and let $S(Q)$ be a shelling of the facets of $Q$. Then, clearly,
\begin{itemize}
\item The restriction of $S(Q)$ to $\st_Q(v)$ yields a shelling of $\st_Q(v)$ (cf. \cite[Lem.~8.7]{Zie95}); denote it by $S_v(Q)$.
\item A shelling of $\st_Q(v)$ induces a shelling of $\lk_Q(v)$ by deleting $v$ from the facets.
\item Since $\partial F=\lk_Q(y)$, it follows that $S(Q)$ induces a shelling $S(F)$ of $F$.
\item Recursively, $S(F)$ induces a shelling of $\lk_F(v)$ if $v\in \ver F$.
\end{itemize}
%In the following we see the polytope $P$ as a simplicial $(d-1)$-complex with induced boundary $F$. Following \cite[Lem.~8.26]{Zie95}, for convenience, we let $\mathcal C/v:=\lk(v,\mathcal C)$.
%In order to compare successive $h$-numbers of $P'=\partial P - \{F\}$,
Now consider any vertex $v\in F$, a shelling $S(Q)$ as guaranteed in Lemma~\ref{lem:shellingOrder}, and the induced shellings $S(Q/v)$ of $Q/v$, $S(F)$ of $F$ and $S(F/v)$ of $F/v$. Recall that if $\st_K(v)$ is simplicial then the boundary complex of $K/v$ coincides with $\lk_{K}(v)$.
Following \cite[Sec.~4]{KarTza12}, call the facet
$F_j$ of $Q$ \emph{active} if it is the new facet to be added to the shelling process $S(Q)$. Let $F_j|F$ be the active facet of $S(F)$ which is the restriction of $F_j$ to $F$ (if $y\in F_j$). Let $F_j/v$ be the active facet of $S(Q/v)$ induced by $F_j$ (if $v\in F_j$), $F_j|F/v$ be the active facet of $S(F/v)$ induced by $F_j|F$ (if $vy\subset F_j$), and $F_j|v$ be the active facet of $S_v(Q)$ induced by $F_j$ (if $v\in F_j$); so $F_j|v=\{v\}\cup F_j/v$ in this case.
Let $R_j\subseteq F_j$, $R_j/v\subseteq F_j/v$, $R_j|F\subseteq F_j|F$, $R_j|F/v\subseteq F_j|F/v$, and $R_j|v$ be the corresponding new minimal faces in the shellings $S(Q)$, $S(Q/v)$, $S(F)$, $S(F/v)$, and $S_v(Q)$ respectively.
Finally, let $h_k^j(Q)$ denote the value of $h_k$ up to step $j$, namely $h_k(\cup_{i\le j}F_i)$, and similarly for the other complexes.
The following key lemma allows us to relate the difference in $h$-numbers along a shelling of $Q$ and $F$ to that of $Q/v$ and $F/v$.
%, and ultimately will allow us to prove the UBT for ASP along the lines of McMullen's proof for the simplicial case \cite{McM70a}.
%Our proof is inspired by the proof of \cite[Lem.~8.26]{Zie95} and the analysis in \cite[Sec.~4]{KarTza12}.
%Pick a vertex $v\in \ver F$. Consider a shelling $S(Q)=F_1,\ldots, F_s$ of the facets of $Q$, or of $Q$ for short, which satisfies the conditions of \cref{lem:shellingOrder}, that is, a shelling that shells the star of $y$ first and then proceeds to shell the rest of the star of $v$. Also, consider the induced shellings $S(Q/v)$ of $Q/v$, $S(F)$ of $F$ and $S(F/v)$ of $F/v$.
%We begin with the following claim.
\begin{lemma}\label{lem:key}
For any vertex $v\in \ver F$ and at any step $j$ of the shelling $S(Q)$ which is guaranteed by \cref{lem:shellingOrder}, we have, for all $k\ge 0$, that
\[h^j_{k}(Q)-h^j_k(Q/v) \ge h^j_k(F)-h^j_k(F/v).\]
\end{lemma}
\begin{proof}
While shelling $\st_Q(y)$, the minimal face $R_j$ of $F_j$ in $S(Q)$ and the minimal face $R_j|F$ of $F_j|F$ in $S(F)$ coincide at every step, since $F=Q/y$ and $S(Q)$ shells the star of $y$ first.
Therefore, while shelling $\st_Q(y)$, for all $k\ge0$, it follows that \[h^j_k(Q)=h^j_k(F).\]
For the same reason, if $F_j\in \st_Q(v)\cap \st_Q(y)$, then, regardless of whether $v\in R_j$ or $v\not\in R_j$, we have, for all $k\ge 0$, that
\[h^j_k(Q/v)=h^j_k(F/v).\]
Thus, while shelling $\st_Q(y)$, it follows for all $k\ge0$ that
\[h^j_{k}(Q)-h^j_k(Q/v)= h^j_k(F)-h^j_k(F/v).\]
After the shelling has left $\st_Q(y)$, we get no new contributions to $h_k(F)$ or $h_k(F/v)$ for all $k\ge0$, so the RHS does not change.
After shelling $\st_Q(y)$ and while still shelling $\st_Q(v)$, we have that the minimal faces $R_j$ and $R_j/v$ of $S(Q)$ and $S(Q/v)$, respectively, coincide, so the LHS does not change either.
To see that $R_j=R_j/v$, first note that
%\eran{It seems you miss something here, please read this part carefully.}
%$v\not\in R_j$ (as $v\in \st_Q(y)$, which was already shelled) and that
$R_j/v \subseteq R_j$ (as the complex at the $j$-th step of $S(Q)$ contains the complex at the $j$-th step of $S_v(Q)$). We show the reverse containment.
Assume by contradiction that there is a facet $F"$ of $F_j$
which is in the subcomplex $\cup_{i< j}F_i$ of $Q$ but not in
the subcomplex $\cup_{i< j,\ v\in F_i}F_i$ of $\st_Q(v)$.
As we have already left $\st_Q(y)$, $y\notin F_j$ so $F"$ is a facet of $F$.
However, also $v\in F$, so
we must have $v\in F"$, as otherwise,
by convexity, $|F_j| \subset |F|$, a contradiction.
But then the (unique) facet $F_i$ in $\st_Q(y)$ containing $F"$ is also in $\st_Q(v)$, a contradiction.
%\textcolor{blue}{ Indeed, suppose, by way of contradiction, that $v\in R_j$. Then, $y\not \in F_j$, and $F_j\setminus v\subset F_i$ for some $1\le i<j$. It then follows that $F_i\in \st(y,Q)\setminus \st(v,Q)$. Thus, $F_j\setminus v$ is a facet of $F$, which would imply that the facet $F_j$ is contained in $F$ (since $v\in F$), {\Large implying that $F_j$ is not convex}. Recall that $F_j$ is not the last facet in $S(Q)$.}
Thus, for all $k\ge 0$,
\[ h^j_{k}(Q)-h^j_k(Q/v)= h^j_k(F)-h^j_k(F/v). \]
After the shelling has left $\st(v,Q)$ we may get new contributions to $h_k(Q)$ but not any more to $h_k(Q/v)$. This concludes the proof of the lemma.
\end{proof}
\begin{proposition}\label{lem:UBTh-Aux}
Let $P\in \Pl(d,n,s)$ and $(P,F)$ be an ASP-pair.
Then, for all $k\ge 0$, we have
\begin{align*}
h_{d-(k+1)}( P')&\le\frac{n-d+k}{k+1}h_{d-k}(P')+\frac{n-(d+s)}{k+1}g_k(F).
\end{align*}
Equivalently, for all $k\ge 0$, we have
\begin{align*}
h_{k+1}( P')&\le\frac{n-d+k}{k+1}h_{k}( P')-\frac{s+k}{k+1}g_k( F)+g_{k+1}( F).
\end{align*}
\end{proposition}
\begin{proof}
The second inequality follows from the first by the Dehn-Sommerville relations (\ref{eq:DSEq-hP}).
For the first inequality, first
%From \cite[Eq~8.27a]{Zie95} in the proof of \cite[Lem.~8.26]{Zie95},
we have the following sequence of equalities.
\begin{align}
\sum_{v\in \ver Q} h_k(Q/v)&=(k+1)h_{k+1}(Q)+(d-k)h_{k}(Q)\nonumber\\
\ &=(k+1)h_{d-(k+1)}( Q)+(d-k)h_{d-k}( Q)
%; \text{since $h_{d-k}( Q)=h_k( Q)$};
\nonumber\\
\ &=(k+1)\left(h_{d-(k+1)}( P')+h_{d-(k+1)-1}( F)\right) +(d-k)h_{d-k}( P')\nonumber\\
&\quad +(d-k)h_{d-k-1}( F)
%,\, (\text{by \cref{eq:hQ=P+F}});
\nonumber\\
\ &=\left((k+1)h_{d-(k+1)}( P')+(d-k)h_{d-k}( P')\right)+(k+1)h_{k+1}( F)\nonumber\\
&\quad+(d-1-k)h_{k}( F)+h_{k}( F)
%,\, \text{since $h_{d-1-k}( F)=h_k( F)$};
\nonumber\\
\ &=(k+1)h_{d-(k+1)}( P')+(d-k)h_{d-k}( P')+h_{k}(F)+\sum_{v\in \ver F} h_k(F/v).
\label{eq:SumlkQ}
\end{align}
For the first equality, see, e.g., \cite[Eq.~8.27a]{Zie95}, while for the second, use Dehn-Sommerville \cref{eq:DSEq-hP} for $Q$. The third equality follows from \cref{eq:hQ=P+F}, the forth from \cref{eq:DSEq-hP} again, this time for $F$, and the last equality follows from \cite[Eq.~8.27a]{Zie95} again.
%Using that $ F=Q/y$ and that $ P\setminus F= Q\setminus (\{y\}\cup F)$, with the help of \begin{align*}\sum_{v\in \ver Q} h_k(Q/v)&=\sum_{v\in \ver Q\setminus (\{y\}\cup F)} h_k(Q/v)+\sum_{v\in \ver F} h_k(Q/v)+h_k(Q/y),\end{align*}
As $ F\cong Q/y$, \cref{eq:SumlkQ} then becomes
\begin{align}
%\sum_{v\in \ver P\setminus F} h_k(Q/v)+\sum_{v\in \ver F} \left(h_k(Q/v)- h_k(F/v)\right)&=(k+1)h_{d-(k+1)}( P) +(d-k)h_{d-k}( P);\nonumber\\
\sum_{v\in \ver P'\setminus \ver F} h_k(Q/v)+\sum_{v\in \ver F} \left(h_k(Q/v)- h_k(F/v)\right)&=(k+1)h_{d-(k+1)}( P')+(d-k)h_{d-k}( P')
\label{eq:SumlkQ1}.
\end{align}
From Lemma~\ref{lem:key} and the fact that any vertex $v\in \ver P'\setminus \ver F$ has the same link in both $Q$ and $P'$, it then follows that
\begin{align}
\sum_{v\in \ver P'\setminus \ver F} h_k(Q/v)+\sum_{v\in \ver F} \left(h_k(Q/v)- h_k(F/v)\right)&\le \sum_{v\in \ver P'\setminus \ver F} h_k( P'/v)\nonumber\\
&\quad+\sum_{v\in \ver F} \left(h_k( Q)- h_k( F)\right)
\label{eq:SumlkQ2}.
\end{align}
Let $v\in \ver P'\setminus \ver F$. There is
a shelling of $P'$ that shells $\st_{P'}(v)$ first
-- just perturb a line through an interior point of $F$ and $v$, so it still intersects $P$ near those two points, to obtain such line shelling.
Such shelling
shows that $h_k(P'/v) \le h_k( P')$ for all $k\ge0$. Consequently, from \cref{eq:SumlkQ1,eq:SumlkQ2} it follows that
\begin{align*}
(k+1)h_{d-(k+1)}( P')+(d-k)h_{d-k}( P')&\le \sum_{v\in \ver P'\setminus \ver F} h_k( P')+\sum_{v\in \ver F} \left(h_k( Q)- h_k( F)\right)\\
&= (n-(d+s))h_k( P')+(d+s)(h_k( Q)- h_k( F))\\
%&\le (n-(d+s))h_k( P)+(d+s)(h_k( P)+h_{k-1}( F))\\
%&\quad- (d+s)h_k( F),\; \text{by \cref{eq:hQ=P+F}};\\
&= nh_k( P')+(d+s)(h_{k-1}( F)- h_k( F))\\
&= n(h_{d-k}( P')+g_k( F))-(d+s)g_k( F).
\end{align*}
The last two equations ensue from \cref{eq:hQ=P+F} and \cref{eq:DSEq-hP}, respectively. Hence, the desired inequality follows.
%\begin{align*}
%(k+1)h_{d-(k+1)}( P)&\le(n-d+k)h_{d-k}( P)+(n-(d+s))(h_{k}( F)-h_{k-1}( F)).
%\end{align*}
%The other inequality follows from applying \cref{eq:DSEq-hP} to the previous inequality. This completes the proof of the lemma.
\end{proof}
\begin{comment}
\begin{lemma}Let $P$ be an almost simplicial polytope with at most one nonsimplex facet of $d+s$ vertices ($s\ge 0$) and let $n:=f_0(P)$ such that $n>d+s$. Then for $0\le k\le d-2$,
\begin{align*}
h_{k}(P)&\le {n-d-1+k\choose k},
\end{align*}
where $h_k(P)$ is the $k$th entry of the $h$-vector associated with a shelling in $\Sp^l$ (i.e.~ the set of shellings in which the nonsimplex facet is last).
\end{lemma}
\begin{proof} The proof follows from reasoning as in \cite[Lem.~8.26]{Zie95}. In following the proof of the second part of \cite[Lem.~8.26]{Zie95}, note that, for any facet $F$ and any vertex $v\not \in \ver F$ of a polytope, the polytope admits a shelling which shells the star of $v$ first and has $F$ as the last facet. Furthermore, for any facet $F$ and any vertex $v\in \ver F$ of a polytope, the polytope admits a shelling which ends with the star of $v$ and has $F$ as the last facet.
\end{proof}
\begin{lemma}\label{lem:h_D/2}Let $P$ be an almost simplicial polytope with at most one nonsimplex facet of $d+s$ vertices ($s\ge 0$) and let $n:=f_0(P)$ such that $n>d+s$. Then
\begin{align*}
h_{\ceil{d/2}}(P)\le\begin{cases}
\displaystyle{n-\floor{d/2}-2\choose \floor{d/2}}-{s+\floor{d/2}-1\choose\floor{d/2}}&\text{if $d$ is odd},\\
\displaystyle{n-\floor{d/2}-1\choose \floor{d/2}}-{s+\floor{d/2}-1\choose\floor{d/2}}&\text{if $d$ is even}.
\end{cases}
\end{align*}
\end{lemma}
\begin{proof}Something should go here.
\end{proof}
\end{comment}
%%%%%%%%%%%%%%%%%%%%%%%%%%
We are now in a position to prove the inequalities of Theorem~\ref{thm:UBT_ASP} using the shelling approach.
%\begin{lemma}\label{lem:UBTh}Let $P$ be an almost simplicial polytope with at most one nonsimplex facet $F$ of $d+s$ vertices ($s\ge 0$) and let $n:=f_0(P)$ be such that $n>d+s$. Then, for all $k\ge0$, we have
%\begin{align}
%h_{k}( P)&\le {n-d-1+k\choose k},&\text{if $0\le k\le %\floor{(d-1)/2}$};\label{eq:hk}\\
%h_{d-k}( P)&\le {n-d-1+k\choose k}-{s+k-1\choose k},&\text{if $1\le k\le \floor{d/2}$}\label{eq:hd-k}.
%\end{align}
%\end{lemma}
\begin{proof}[Proof of Theorem \ref{thm:UBT_ASP} via Shellings]
%Our proof proceeds by induction on $k$ for each inequality, with the help of \cref{lem:UBTh-Aux}.
The inequalities \cref{eq:hk} for $P'$ in Theorem \ref{thm:UBT_ASP} hold for any Cohen-Macaulay complex, and can also be proved exactly as in \cite[Lem.~8.26]{Zie95}. We prove \cref{eq:hd-k} by induction on $k$. The case $k=1$ holds with equality by \cref{eq:DSEq-hP}:
$h_{d-1}( P')= n-d-s$.
Suppose now that the inequality \cref{eq:hd-k} holds for $k-1\le \floor{d/2}-1$. By \cref{lem:UBTh-Aux},
\begin{align*}
h_{d-k}( P')&\le \frac{n-d+k-1}{k}\left( {n-d-1+k-1\choose k-1}-{s+k-2\choose k-1}\right)+\frac{n-(d+s)}{k}g_{k-1}( F).
\end{align*}
From the application of the $g$-theorem\footnote{In fact, this consequence follows easily just from the fact that $\partial F$ is a Cohen-Macaulay complex.}, cf. \cite[Cor.~8.38]{Zie95}, to $F$, it follows that
$g_{k-1}( F)\le {d+s-(d-1)+k-3\choose k-1}$ for $0\le k-1\le \floor{d/2}-1$. Thus, the previous inequality becomes
\begin{align*}
h_{d-k}( P')&\le \frac{n-d+k-1}{k}\left( {n-d-1+k-1\choose k-1}-{s+k-2\choose k-1}\right)+\frac{n-(d+s)}{k} {s+k-2\choose k-1}\\
&= {n-d-1+k\choose k} -{s+k-1\choose k},\;
\end{align*}
%The last inequality follows from the equalities ${n-d-1+k\choose k}=\frac{n-d+k-1}{k} {n-d-1+k-1\choose k-1}$ and ${s+k-1\choose k}=\frac{s+k-1}{k}{s+k-2\choose k-1}$.
%We now prove \cref{eq:hk} by induction $k$. The base $k=0$ is trivial, as it says that $h_0( P)\le 1$, which is true. Assume that the inequality is true for $0\le k-1\le\floor{(d-1)/2}-1$. From \cref{lem:UBTh-Aux} it ensues that
%\begin{align*}
%h_{k}( P)-g_k( F)&\le \frac{n-d+k-1}{k}h_{k-1}( P)-\frac{s+k-1}{k}g_{k-1}( F);\\
%&\le {n-d-1+k\choose k}-\frac{s+k-1}{k}g_{k-1}( F),\; \text{by the induction hypothesis}.\end{align*}
%From applying the $g$-theorem for simplicial polytopes \cite[Cor.~8.38]{Zie95} to $F$, it ensues that
%$\frac{s+k-1}{k}g_{k-1}( F)\ge g_{k}( F)$ for $0\le k-1\le \floor{(d-1)/2}-1$. Thus, for $0\le k\le \floor{(d-1)/2}$ we have
%\begin{align*}
%h_{k}( P)&\le {n-d-1+k\choose k},
%\end{align*}
%as derired.
as desired.\end{proof}
%\begin{proof}[Proof of \cref{thm:UBT_ASP}] The theorem immediately follows from \cref{lem:UBTh}.
%\end{proof}
\begin{comment}
\section{A generalized lower bound theorem for almost simplicial polytopes}
Recall that $P'=\partial P -\{F\}$. We find it convenient to state the results here using the \emph{toric $g$-vector} of $P$ (rather than the simplicial $h$-vectors of $P'$ and $\partial F$; both are possible).
The toric $h$- and $g$-vectors were introduced by Stanley~\cite{Stanley:GeneralizedH-vectors} and the reader is referred there for further background.
They are defined by a polynomial recurrence
\[h(P,t)=\sum_{0\le i\le \dim P}h_i(P)t^i :=\sum_{F<P}g(F,t)(t-1)^{\dim(P)-\dim(F)-1},\]
where the sum runs over all proper faces $F$ of the polytope $P$, setting \[g(P,t)=\sum_{0\le i\le \floor{d/2}}g_i(P)t^i\; \text{and}\;g_i(P)=h_i(P)-h_{i-1}(P),\] as in the simplicial case, and the initial conditions $h(\emptyset,t)=1=g(\emptyset,t)$.
Stanley showed that $h(P,t)$ is a palindrome.
Karu \cite{Kar06} showed that for any polytope $P$, $g(P,t)$ is nonnegative coefficientwise, by proving a hard Lefschetz theorem for the combinatorial intersection homology module associated with $P$.
The structure of polytopes with $g_2=0$ is not known, cf.~\cite[Conj.~11,12]{Kal94}, however, the results of Section~\ref{sec:LBT} do provide a characterization for almost simplicial polytopes, similar to the simplicial case. In this section, we
extend these results to higher $g$-numbers, similar to the classical generalized lower bound theorem (GLBT); see \cite{McMWal71,MurNev13}.
Recall that a simplicial $d$-polytope $P$ is \emph{$(r-1)$-stacked} if the simplicial complex $P(d-r)$, formed by all subsets of $\ver P$ whose $(d-r+1)$-subsets are faces of $\partial P$, triangulates $P$.
Similarly, extend this notion to any geometric triangulation of a polytope boundary.
\begin{theorem}\label{thm:GLBT_ASP}
Let $P\in \Pl(d,n,s)$ with $g_k(P)=0$ for some $1\le k<d/2$, and let $(P,F)$ be an ASP-pair. Then $F$ is $(k-1)$-stacked.
Furthermore, let $P"$ be the simplicial complex refining $\partial P$ by replacing $F$ by its $(k-1)$-stacked triangulation $F(d-1-k)$.
Then $P"(d-k)$ triangulates $P$.
\end{theorem}
\begin{proof}
By Kalai's monotonicity, cf.\cite{Bra06}, $g_k(F)=0$, so from the GLBT \cite{MurNev13} it follows that $F(d-1-k)$ triangulates $F$.
By a very recent result of Adiprasito \cite[Theorem 7.2]{AdipIIver1}, $F(d-1-k)$ is regular, so lifting it by a small enough amount into convex position gives that $P"$ is isomorphic to the boundary complex of a simplicial $d$-polytope $Q$. Further, as $g_k(P")=0$ holds\footnote{This can be shown directly from the definition, or by noting that the intersection homology modules of the complete fans of $P$ and of $P"$ coincide up to degree $k$.},
the GLBT applied to $Q$
%(indeed $g_k(Q)=g_k(P")=0$)
gives that $Q$ is $(k-1)$-stacked, thus
$P"(d-k)$
%triangulates $P"$, hence
triangulates $P$.
\end{proof}
\begin{problem}
What can be said in the case $2<k=d/2$? (Compare with Theorem~\ref{thm:LBT=4} for $2=k=d/2$.)
\end{problem}
\end{comment}
\section*{Acknowledgments}
We thank Karim Adiprasito, Gil Kalai and Isabella Novik for helpful feedback on an earlier version of this paper.
%We end with the following problem.
%\begin{problem}
%Let $2\le k=d/2$ and $P\in \Pl(d,n,s)$ with $g_k(P)=0$. Must $\partial P$ have a $(k-1)$-stacked triangulation? NO, as our d=4, k=2 minimizers show.
%\end{problem}
%---REFERENCES---
%\bibliographystyle{amsplain}
%\bibliography{GPinedaVillavicencioBibDesk}
%\begin{comment}
\providecommand{\bysame}{\leavevmode\hbox to3em{\hrulefill}\thinspace}
\providecommand{\MR}{\relax\ifhmode\unskip\space\fi MR }
% \MRhref is called by the amsart/book/proc definition of \MR.
\providecommand{\MRhref}[2]{%
\href{http://www.ams.org/mathscinet-getitem?mr=#1}{#2}
}
\providecommand{\href}[2]{#2}
\begin{thebibliography}{10}
\bibitem{AdiSan14}
K.~A. Adiprasito and R.~Sanyal, \emph{Relative stanley-reisner theory and upper bound theorems for {M}inkowski sums}, Publications Math\'ematiques. Institut de Hautes \'Etudes Scientifiques \textbf{124} (2016), 99--163.
\bibitem{Asi-Roth2}
L. Asimow and B. Roth,
\emph{The rigidity of graphs. {II}},
J. Math. Anal. Appl.,
\textbf{68} (1979), 171--190.
\bibitem{Bar71}
D.~W. Barnette, \emph{The minimum number of vertices of a simple polytope},
Israel J. Math. \textbf{10} (1971), 121--125. \MR{0298553 (45 \#7605)}
\bibitem{Bar73}
\bysame, \emph{A proof of the lower bound conjecture for convex polytopes},
Pacific J. Math. \textbf{46} (1973), 349--354. \MR{0328773 (48 \#7115)}
\bibitem{Billera-Lee}
L.~J. Billera and C.~W. Lee,
\emph{A proof of the sufficiency of {M}c{M}ullen's conditions for
{$f$}-vectors of simplicial convex polytopes},
J. Combin. Theory Ser. A,
\textbf{31},
(1981),
237--255.
\bibitem{BilLee81a}
\bysame, \emph{The numbers of faces of polytope pairs and
unbounded polyhedra}, European J. Combin. \textbf{2} (1981), no.~4, 307--322.
\MR{638405 (82m:52005)}
\bibitem{Bra06}
T.~Braden, \emph{Remarks on the combinatorial intersection cohomology of fans},
Pure Appl. Math. Q. \textbf{2} (2006), no.~4, part 2, 1149--1186. \MR{2282417
(2008c:14031)}
\bibitem{BraMac99}
T.~Braden and R.~D. MacPherson, \emph{Intersection homology of toric varieties
and a conjecture of {K}alai}, Comment. Math. Helv. \textbf{74} (1999), no.~3,
442--455. \MR{1710686 (2000h:14018)}
\bibitem{Bro83}
A.~Br{\o}ndsted, \emph{An introduction to convex polytopes}, Graduate Texts in Mathematics, vol.~90, Springer-Verlag, New York, 1983. \MR{683612 (84d:52009)}
\bibitem{BruMan71}
H.~Bruggesser and P.~Mani, \emph{Shellable decompositions of cells and
spheres}, Math. Scand. \textbf{29} (1971), 197--205 (1972). \MR{0328944 (48
\#7286)}
\bibitem{GooORo04}
J.~E. Goodman and J.~O'Rourke (eds.), \emph{Handbook of discrete and
computational geometry}, 2nd ed., Discrete Mathematics and its Applications
(Boca Raton), Chapman \& Hall/CRC, Boca Raton, FL, 2004. \MR{2082993
(2005j:52001)}
\bibitem{Gru03}
B.~Gr{{\"u}}nbaum, \emph{Convex polytopes}, 2nd ed., Graduate Texts in
Mathematics, vol. 221, Springer-Verlag, New York, 2003, Prepared and with a
preface by V. Kaibel, V. Klee and G. M. Ziegler. \MR{1976856 (2004b:52001)}
\bibitem{Kal87}
G.~Kalai, \emph{Rigidity and the lower bound theorem. {I}}, Invent. Math.
\textbf{88} (1987), no.~1, 125--151. \MR{877009 (88b:52014)}
\bibitem{Kal94}
\bysame, \emph{Some aspects of the combinatorial theory of convex polytopes},
Polytopes: abstract, convex and computational ({S}carborough, {ON}, 1993),
NATO Adv. Sci. Inst. Ser. C Math. Phys. Sci., vol. 440, Kluwer Acad. Publ.,
Dordrecht, 1994, pp.~205--229. \MR{1322063 (96b:52018)}
\bibitem{KarTza12}
M.~I. Karavelas and E.~Tzanaki, \emph{The maximum number of faces of the
{M}inkowski sum of two convex polytopes}, Proceedings of the {T}wenty-{T}hird
{A}nnual {ACM}-{SIAM} {S}ymposium on {D}iscrete {A}lgorithms, ACM, New York,
2012, pp.~11--22. \MR{3205193}
\bibitem{Kar06}
K.~Karu, \emph{Hard {L}efschetz theorem for nonrational polytopes},
Invent. Math., \textbf{157} (2004), 419--447.
\bibitem{Kolins11}
S.~Kolins, \emph{$f$-vectors of triangulated balls},
Disc. Comp. Geom., \textbf{46} (2011), 427--446.
%\bibitem{Mat02a}
%J.~Matou{\v{s}}ek, \emph{Lectures on discrete geometry}, Graduate Texts in
% Mathematics, vol. 212, Springer-Verlag, New York, 2002. \MR{1899299
% (2003f:52011)}
\bibitem{McM70a}
P.~McMullen, \emph{The maximum numbers of faces of a convex polytope},
Mathematika \textbf{17} (1970), 179--184. \MR{0283691 (44 \#921)}
\bibitem{McMullen-g-conj}
\bysame,
\emph{The numbers of faces of simplicial polytopes},
Israel J. Math.,
\textbf{9},
1971,
559--570.
\bibitem{McMWal71}
P.~McMullen and D.~W. Walkup, \emph{A generalized lower-bound conjecture for
simplicial polytopes}, Mathematika \textbf{18} (1971), 264--273. \MR{0298557
(45 \#7609)}
\bibitem{MurNev13}
S.~Murai and E.~Nevo, \emph{On the generalized lower bound conjecture for
polytopes and spheres}, Acta Math. \textbf{210} (2013), no.~1, 185--202.
\MR{3037614}
\bibitem{Nevo-Novinsky}
E.~Nevo and E.~Novinsky,
\emph{A characterization of simplicial polytopes with {$g_2=1$}},
J. Combin. Theory Ser. A,
\textbf{118},
(2011) 387--395.
\bibitem{Sta80}
R.~P. Stanley, \emph{The number of faces of a simplicial convex polytope},
Adv. Math. \textbf{35} (1980), 236--238.
\bibitem{Stanley:GeneralizedH-vectors}
\bysame,
\emph{Generalized {$H$}-vectors, intersection cohomology of toric
varieties, and related results},
Commutative algebra and combinatorics (Kyoto, 1985),
{Adv. Stud. Pure Math.},
\textbf{11}, 187--213,
North-Holland, Amsterdam, 1987.
\bibitem{Whi84}
W.~Whiteley, \emph{Infinitesimally rigid polyhedra. {I}. {S}tatics of
frameworks}, Trans. Amer. Math. Soc. \textbf{285} (1984), no.~2, 431--465.
\MR{752486 (86c:52010)}
%\bibitem{Whi96}
%\bysame, \emph{Some matroids from discrete applied geometry}, Matroid theory
% ({S}eattle, {WA}, 1995), Contemp. Math., vol. 197, Amer. Math. Soc.,
% Providence, RI, 1996, pp.~171--311. \MR{1411692 (97h:05040)}
\bibitem{Zie95}
G.~M. Ziegler, \emph{Lectures on polytopes}, Graduate Texts in Mathematics,
vol. 152, Springer-Verlag, New York, 1995. \MR{1311028 (96a:52011)}
\end{thebibliography}
%\end{comment}
\end{document}