EMIS/ELibM Electronic Journals

Outdated Archival Version

These pages are not updated anymore. They reflect the state of 20 August 2005. For the current production of this journal, please refer to http://www.math.psu.edu/era/.


%_ **************************************************************************
%_ * The TeX source for AMS journal articles is the publishers TeX code     *
%_ * which may contain special commands defined for the AMS production      *
%_ * environment.  Therefore, it may not be possible to process these files *
%_ * through TeX without errors.  To display a typeset version of a journal *
%_ * article easily, we suggest that you retrieve the article in DVI,       *
%_ * PostScript, or PDF format.                                             *
%_ **************************************************************************
% Author Package file for use with AMS-LaTeX 1.2
\controldates{17-NOV-2004,17-NOV-2004,17-NOV-2004,17-NOV-2004}
 
\RequirePackage[warning,log]{snapshot}
\documentclass{era-l}
\issueinfo{10}{14}{}{2004}
\dateposted{December 1, 2004}
\pagespan{122}{134}
\PII{S 1079-6762(04)00137-4}


\copyrightinfo{2004}{}
\revertcopyright            

\usepackage{amsmath}
\usepackage{amssymb}
\usepackage{amsthm}
\usepackage{graphicx}

\theoremstyle{plain}
\newtheorem{lemma}{Lemma}[section]
\newtheorem{theorem}[lemma]{Theorem}
\newtheorem*{maintheorem}{Theorem}
\newtheorem{corollary}[lemma]{Corollary}
\newtheorem{proposition}[lemma]{Proposition}

\theoremstyle{definition}
\newtheorem{definition}[lemma]{Definition}
\newtheorem{example}[lemma]{Example}
\newtheorem{convention}[lemma]{Convention}  
\newtheorem{observation}[lemma]{Observation}  

\theoremstyle{remark}
\newtheorem*{remark}{Remark} 

\newcommand{\R}{\mathbb{R}}
\newcommand{\Q}{\mathbb{Q}}
\newcommand{\Z}{\mathbb{Z}}
\newcommand{\N}{\mathbb{N}}
\newcommand{\eps}{\varepsilon}

\DeclareMathOperator{\aff}{aff}
\DeclareMathOperator{\conv}{conv}
\DeclareMathOperator{\cone}{cone}
\DeclareMathOperator{\lin}{lin}
\DeclareMathOperator{\rank}{rank}
\DeclareMathOperator{\relint}{relint}
\DeclareMathOperator{\sign}{sign}
\DeclareMathOperator{\vol}{vol}
\newcommand{\myvec}[1]{{\boldsymbol {#1}}}
\newcommand{\vv}{\myvec{v}}
\newcommand{\uu}{\myvec{u}}
\newcommand{\ww}{\myvec{w}}
\newcommand{\xx}{\myvec{x}}
\newcommand{\yy}{\myvec{y}}
\newcommand{\bb}{\myvec{b}}
\newcommand{\cc}{\myvec{c}}
\newcommand{\ee}{\myvec{e}}
\newcommand{\nn}{\myvec{n}}
\newcommand{\zero}{\myvec{0}}
\newcommand{\F}{\mathcal{F}}

\graphicspath{{pictures/}}

\begin{document}
\title{Projected products of polygons}
\author{G\"unter M. Ziegler}
\address{Inst. Mathematics, MA 6-2, TU Berlin, D-10623 Berlin, Germany}
\email{ziegler@math.tu-berlin.de}
\thanks{Partially supported by Deutsche Forschungs-Gemeinschaft (DFG), via the
\emph{Matheon} Research Center ``Mathematics for Key Technologies'' (FZT86),
the Research Group ``Algorithms, Structure, Randomness'' (Project ZI 475/3),
and a Leibniz grant (ZI 475/4)}
\date{July 4, 2004}
\revdate{September 17, 2004}
\commby{Sergey Fomin}
\subjclass[2000]{Primary 52B05; Secondary 52B11, 52B12}
\keywords{Discrete geometry, convex polytopes, 
$f$-vectors, deformed products of polygons}
\begin{abstract}
It is an open problem to characterize the
 cone of $f$-vectors of $4$-dimensional convex polytopes.
The question whether the ``fatness'' of the $f$-vector of a 
$4$-polytope can be arbitrarily large is
a key problem in this context.

Here we construct a $2$-parameter family of $4$-dimensional polytopes
$\pi(P^{2r}_n)$ with extreme combinatorial structure. In this family, 
the ``fatness'' of the $f$-vector gets arbitrarily close to~$9$;
an analogous invariant of the flag vector, the ``complexity,'' 
gets arbitrarily close to~$16$.
\par
The polytopes are obtained from suitable deformed products of even
polygons by a projection to~$\mathbb{R}^4$.
\end{abstract}
\maketitle

\section{Introduction}
\subsection{{\itshape f}\/-vectors}

The combinatorial structure of a $d$-dimensional convex polytope
is given by the incidences between
its $k$-dimensional faces, for $0\le k\le d-1$.
In particular one looks at the faces of 
dimensions $0$, $1$, $d-2$, and $d-1$,
known as the \emph{vertices}, \emph{edges}, \emph{ridges}, and
\emph{facets}, respectively. 

To get an overview over enumerative and extremal properties
of the multitude of combinatorial types of $d$-polytopes, 
one tries to classify their \emph{$f$-vectors}, that is, the $d$-tuples
\[
f(P)\ :=\ (f_0,f_1,\dots,f_{d-2},f_{d-1})\ \in\Z^d,
\]
where $f_k$ denotes the number of $k$-dimensional faces of the
$d$-polytope~$P$.
The $f$-vector of any $d$-polytope is a point in $\R^d$,
but due to the Euler--Poincar\'e relation 
\[
f_0-f_1\pm\dots+(-1)^{d-1}f_{d-1}\ =\ 1-(-1)^d
\]
the set of all $f$-vectors of $d$-polytopes 
\[
\F_d\ :=\ 
\big\{f=(f_0,f_1,\dots,f_{d-1})\in\Z^d :
f=f(P)\textrm{ for a $d$-polytope }P\big\}
\]
has dimension $d-1$ \cite[Chapter 8]{Gr1-2}.


The $f$-vector
(and more so the flag vector of a $d$-polytope, as discussed below)
not only provides numerical data: it also
encodes various extremal properties.
So any attempt to characterize the $f$-vectors
of polytopes is closely linked to the
analysis and construction of extremal polytopes.

For example, a $d$-polytope is simplicial
if and only if its $f$-vector satisfies
$f_{d-2}=\frac d2 f_{d-1}$. Indeed,
each facet of a $d$-polytope is bounded by at least
$d$ ridges, while every ridge is in exactly $2$~facets.
Hence $2f_{d-2}\le d\, f_{d-1}$ is valid for all
polytopes, and the constraint is tight exactly 
if all facets are simplices.
Thus simplicial polytopes are \emph{extremal} in the sense that
they maximize a (linear) quantity among all $f$-vectors:
they maximize $2f_{d-2}-d\, f_{d-1}$.

In the study of $f$-vectors of $d$-polytopes, one tries
to find all such linear inequalities for the $f$-vectors, 
and to understand the extremal polytopes for which these inequalities
are tight.

My lecture notes \cite{Z99} contain
a more extensive discussion of the interplay between $f$-vector theory
and constructions of extremal polytopes that arises from this.

\subsection{Flag vectors}
Additional combinatorial information is contained in the 
\emph{flag vector} of a $d$-polytope, which in addition
to the components $f_k$ of the $f$-vector also records
the numbers $f_{k,\ell}$ of incidences of $k$-faces with $\ell$-faces
(that is, the numbers of pairs $(F,G)$ consisting of a
$k$-face $F$ contained in an $\ell$-face $G$, for $k<\ell$),
the numbers $f_{k,\ell,m}$ of chains of three 
faces $F\subset G\subset H$ of dimensions $k<\ell0$ there are polytopes of fatness larger
than $9-\eps$.
\end{theorem}

Thus the known polytopes now span a
hexagon, which is shaded in Figure~\ref{fig:phiplane}.

A flag vector parameter
that is similar to fatness, called \emph{complexity}~\cite{Z82},
is defined by
\[
C(P)\ :=\ \frac{f_{03}-20}{f_0+f_3-10}. 
\]
All $4$-polytopes satisfy $C(P)\ge3$.
Fatness and complexity are roughly within a factor of~$2$:
$C(P)\le 2F(P)-2$ and $F(P)\le 2C(P)-2$.
In particular, it is not known whether $C(P)$ can be arbitrarily
large.
Previously, the polytopes with the largest known 
complexity were the ``neighborly cubical polytopes''
of Joswig and Ziegler \cite{Z62}, of complexity $8-\eps$.
Our present construction yields ``neighborly cubical polytopes''
for $n=4$, but for $n,r\rightarrow\infty$ 
it yields complexity as large as $16-\eps$.

In the following two sections, we review the main ingredients
for the construction of $\pi(P_n^{2r})$.
The construction that yields Theorem~\ref{thm:ppp}
is described in Section~\ref{sec:construction}, with
a sketch of the proof for its correctness.
The flag vectors of the polytopes~$\pi(P_n^{2r})$ 
are computed in Section~\ref{sec:flag}, which yields
Theorem~\ref{thm:f-vectors}.
Detailed proofs, the combinatorial characterization of the
resulting polytopes, possible extensions, further remarkable aspects
(such as the polyhedral surfaces of high genus embedded in the
$2$-skeleta of the resulting $4$-polytopes; cf.\ 
\cite{schroeder04}) as well as necessity
of the restrictions (e.g., that $n$ must be even)
are topics of current research and will be presented later.

\subsection*{Acknowledgements}
The intuition for the construction given here 
grew from previous joint work and current discussions
with Nina Amenta, Michael Joswig, Raman Sanyal, \pagebreak
and Thilo Schr\"oder.

\section{Products and deformed products}\label{sec:products}

The combinatorial structure of the
products of polygons $(C_n)^r$ is easy to describe.
These are simple $2r$-polytopes, with $f_0=n^r$ vertices,
$f_1=rn^r$ edges, and $f_{2r-1}=rn$ facets.
In general, its non-empty faces are products of non-empty faces
of the polygons, so 
$\sum_{i=0}^{2r}f_i t^i=(n+nt+t^2)^r$.

The $2$-dimensional faces of~$(C_n)^r$, and thus
of any polytope combinatorially equivalent to~$(C_n)^r$,
may be split into two classes. There are 
$rn^{r-1}$ faces that are $n$-gons, to
which we refer as \emph{polygons}; they arise as products of one
of the $n$-gons with a vertex from each of the other factors.
There are also $\binom r2 n^r$ \emph{quadrilaterals} that 
(in $(C_n)^r$) arise
as products of edges from two of the factors with vertices from
the others. 
Thus, in total $(C_n)^r$ has $f_2=r n^{r-1} + \binom r2 n^r$ $2$-faces.

In the case $n=4$, the polygon $2$-faces of $(C_n)^r$
are $4$-gons, but we nevertheless treat the $r4^{r-1}$ polygons
and the $\binom{r}2 4^r$ quadrilaterals separately also in this case.

An inequality description for such a product polytope may be 
obtained as 
\begin{equation*}
\left(\raisebox{-39mm}{\includegraphics{era137el-pict-1}}\!\right)\xx 
\ \ \le\ \ 
\left(\raisebox{-39mm}{\includegraphics{era137el-pict-2}}\!\right),
\end{equation*}
assuming that $V\xx\le\bb$ is a correct description
for an $n$-gon. For this it is necessary and sufficient
that the row vectors $\vv_i$ of $V$ are non-zero and distinct
and that they positively span~$\R^2$,
that the components $b_i$ of~$\bb$ are positive,
and that the rescaled vectors $\tfrac1{b_i}\vv_i$ are in convex
position (the vertices of the polar of the polygon).

For this we say that a finite set of
vectors $\vv_1,\dots,\vv_k\in\R^d$ \emph{positively spans} if 
it satisfies the following equivalent conditions:
\begin{enumerate} 
\item[(i)] every vector $\xx\in\R^d$ is a linear combination of 
the vectors $\vv_i$, with non-negative \pagebreak coefficients;
\item[(ii)] every vector $\xx\in\R^d$ is a linear combination of 
the vectors $\vv_i$, with positive coefficients;
\item[(iii)] the vectors $\vv_i$ span $\R^d$, and
      $\zero\in\R^d$ is a linear combination of 
      the vectors $\vv_i$, with positive coefficients
      (that is, the vectors $\vv_i$ are \emph{positively dependent}).
\end{enumerate} 
In the following, we will need ``deformed products.''
(The deformations are more general than the ``rank~$1$'' deformations
as described in Amenta and Ziegler  \cite{Z51a}.)
For this, we look at systems of the form
\begin{equation*}
\left(\raisebox{-39mm}{\includegraphics{era137el-pict-3}}\!\right)\xx 
\ \ \le\ \ 
\left(\raisebox{-39mm}{\includegraphics{era137el-pict-4}}
\!\right).
\end{equation*} 
Given any such left-hand side matrix for such a system,
we can adapt the right-hand side
so that the resulting polytope is \emph{combinatorially equivalent}
to~$(C_n)^r$. For this all components of $\bb_k$ have to be sufficiently
large compared to $\bb_1,\dots,\bb_{k-1}$,
for $k=2,3,\dots,r$. (Compare \cite{Z51a} and \cite{Z62}.)

\section{Projections}\label{sec:projections}

We will work with a  rather restrictive concept of
faces ``being preserved under projection.''

\begin{definition}[Strictly preserving faces under projection]%
\label{def:strictly}%
Let $\pi:P\rightarrow Q=\pi(P)$ be a projection of polytopes.
Then a face $G\subseteq P$ is 
\emph{strictly preserved} if
\begin{enumerate} 
\item[(i)]\label{ci} its image $\pi(G)$ is a face of~$Q$,
\item[(ii)]\label{cii} the map $G\rightarrow\pi(G)$ is a bijection, and
\item[(iii)]\label{ciii} the preimage of the image is $G$, that is, $\pi^{-1}(\pi(G))=G$.
\end{enumerate} 
\end{definition}

In the definition, conditions (ii) and (iii) are both needed.
Indeed, in the projection ``to the second coordinate'' displayed in our figure, 
the vertex \pagebreak $v$ is strictly preserved, 
but the vertex~$w$ and the edge $e$ are not: 
for $w$ condition (iii) fails, while for $e$ condition (ii) is
violated. 
\begin{equation*} 
\includegraphics{era137el-pict-5}
\end{equation*}

For simplicity, the following characterization result
is given only in a coordinatized version, for the
projection ``to the last $d$ coordinates.''

We say that a vector $\cc$ \emph{defines} the face 
$G\subseteq P$ given by all the points of $P$ that have
maximal scalar product with $\cc$. This describes exactly
all the vectors in the relative interior of the
\emph{normal cone} of~$G$. If $P$ is full-dimensional,
this interior of the normal cone consists of all the
positive combinations of outer facet normals $\nn_F^{}$ to the facets
$F\subset P$ that contain~$G$.
(Compare \cite[Sections 2.1, 3.2, 7.1]{Z35}.)

\begin{proposition}\label{prop:preserve}%
Let $\pi:\R^{e+d}\rightarrow\R^d$, $(\xx',\xx'')\mapsto\xx''$ be the
projection to the last~$d$ coordinates, and
let $P\subset\R^{e+d}$ be an $(e+d)$-dimensional polytope,
and let $G$ be a face of~$P$.
Then the following three conditions are equivalent:
\begin{enumerate} 
\item[(1)]
$G$ is strictly preserved by the projection $\pi:P\rightarrow\pi(P)=Q$.
\item[(2)]
Any $\cc'\in\R^e$ arises as the first $e$ components 
of a vector $(\cc',\cc'')$ that defines~$G$.
\item[(3)]
The vectors $\nn_F'$, given by the first $e$ components 
of the normal vectors $(\nn_F',\nn_F'')=\nn^{}_F$ to facets $F$ of~$P$ that
contain $G$, positively span $\R^e$.
\end{enumerate}
\end{proposition}

\begin{proof}
Here we only establish ``$(3)\Rightarrow(1)$,'' which is used
in the following.

If the vectors $\nn_F'$ are positively dependent,
then some positive combination of the vectors $(\nn_F',\nn_F'')=\nn^{}_F$ 
yields $(\zero,\cc'')=:\cc$.
A point $\xx\in P$ lies in the face $G\subseteq P$ if and only
if its scalar product with each facet normal $\nn^{}_F$
is maximal. This happens if and only if
$\cc^t\xx$ is maximal, that is, iff
$(\cc'')^t\xx''$ is maximal under the restriction $\xx''\in\pi(P)$.
Thus we have established that under the assumption (3),
$\pi(G)=:\bar G$ is a face of~$\pi(P)$, and 
$\pi^{-1}(\pi(G))=\pi^{-1}(\bar G)=G$; that is,
conditions (i) and (iii) of Definition~\ref{def:strictly}
are satisfied.

For part~(ii) of Definition~\ref{def:strictly}, 
we have to show that $G\rightarrow\pi(G)$ is injective.
Assume that $\xx=(\xx',\xx'')$ and $\yy=(\yy',\yy'')$ are
points in~$G$ with $\pi(\xx)=\pi(\yy)$, that is, $\xx''=\yy''$.
For each normal vector $\nn^{}_F=(\nn_F',\nn_F'')$ we have
$\nn^{}_F{}^t\xx=\nn^{}_F{}^t\yy$ and 
$(\nn_F'')^t\xx''=(\nn_F'')^t\yy''$,
which implies $(\nn_F')^t\xx''=(\nn_F')^t\yy''$.
But the vectors $\nn_F'$ that correspond to the various facets $F$ that contain~$G$ are
positively spanning in~$\R^e$, which implies $\xx'=\yy'$.
\end{proof}

\section{Construction}\label{sec:construction}

\begin{proposition}[Construction for the proof of Theorem~\ref{thm:ppp}]
For $n\ge4$ even and $r\ge2$, let $P_n^{2r}$ be defined by
the linear inequality system
\begin{equation*}
\left(\raisebox{-55mm}{\includegraphics{era137el-pict-6}}
\right)\xx 
\ \ \le\ \ 
\left(\raisebox{-55mm}{\includegraphics{era137el-pict-7}}
\right).
\end{equation*}
Here the left-hand side coefficient matrix 
$A_{n,r}^\eps\in\R^{rn\times 2r}$
contains blocks of size $n\times 2$, where
\begin{equation*}
V=\raisebox{-8mm}{\includegraphics{era137el-pict-8}}
\longrightarrow
V^\eps=\raisebox{-8mm}{\includegraphics{era137el-pict-9}},
\qquad
W=\raisebox{-8mm}{\includegraphics{era137el-pict-10}},
\qquad
U=\raisebox{-8mm}{\includegraphics{era137el-pict-11}}
\ \in\ \R^{n\times2},
\end{equation*} 
with
\begin{gather*}
\vv_0=(1,0),\  
\vv_1=(0,0)=\zero,\  
\ww_0=(0,1),\  
\ww_1=(-3,-\tfrac23),\\
\uu_0=(-\tfrac{31}4,\tfrac12),\  
\uu_1=(9,-\tfrac23).
\end{gather*}
The block $V^\eps$ arises from $V$ by an $\eps$-perturbation:
\[
\vv_i^\eps\ =\ \begin{cases}
\hphantom{\eps}\big(1-\eps (n-2-2i)^2,\eps (n-2-2i)\big)
&\textrm{for }i=0,2,4,\ldots,n-2,\\
         {\eps}\big(1-\eps (n-2-2i)^2,\eps (n-2-2i)\big)
&\textrm{for }i=1,3,5,\ldots,n-3,\\
\eps(-1,0)\ =\ (-\eps,0)                     
&\textrm{for }i=n-1
\end{cases}
\]
for a sufficiently small $\eps>0$.
All entries of $A^\eps_{n,r}$ outside the $r+(r-1)+(r-2)=3r-3$ blocks
of types $V^\eps$, $W$ and $U$ are zero.

Let the right-hand side vector be such that $\bb_1$ is given by 
$b_{1,i}=1$ for even $i$, and $b_{1,i}=\eps$ for odd~$i$,
and by $\bb_k=M^{k-1}\bb_1$ for sufficiently large $M$.

Then $P_n^{2r}$ has the properties claimed by Theorem~\ref{thm:ppp}.
In particular, it is a deformed product of $r$ $n$-gons, and
all its polygon $2$-faces survive the projection to 
the last $4$ coordinates.
\end{proposition}

\begin{proof}
The rows $\vv_i^\eps$ of $V^\eps$ are
indeed in cyclic order:
\begin{equation*} 
\includegraphics[scale=.95]{era137el-pict-12}
\end{equation*}
Moreover, rescaled as 
$\frac1{b_{k,i}}\vv_i^\eps=\frac1{M^{k-1}\eps}\vv_i^\eps$ for odd~$i$ and as
$\frac1{b_{k,i}}\vv_i^\eps=\frac1{M^{k-1}    }\vv_i^\eps$ for even $i$ 
they are in convex position, if $\eps$ is small; so
 $V^\eps\xx\le\bb_i$ defines a convex $n$-gon.
Thus for sufficiently small $\eps$ and sufficiently large~$M$,
the polytope $P_{2r}$ is indeed a deformed product of polygons,
as discussed in Section~\ref{sec:products}.

Now we show that for sufficiently small $\eps$,
all the polygon $2$-faces of $P_n^{2r}$ survive the projection
to the last $4$ coordinates. For this, we verify that the 
left-hand side matrix with $V$-blocks instead of $V^\eps$-blocks,
which we denote by $A_{n,r}^0=A_{n,r}$,
satisfies the linear algebra condition
dictated by Proposition~\ref{prop:preserve}(3).
This is sufficient, since the ``positively spanning'' condition is
stable under perturbation by a small~$\eps$.

Any polygon $2$-face $G$ of the simple $2r$-polytope $P^{2r}_n$ is defined
by  the facet normals to the $2r-2$ facets that contain~$G$.
The facet normals correspond to the rows of the inequality system,
and thus for the facet normals of a polygon $2$-face 
one has to choose two cyclically adjacent rows from each block
(corresponding to a vertex from each factor polygon), 
except from one of the blocks no row is taken.
Moreover, due to the structure of the matrices $U$, $V$, and $W$,
in which rows alternate, any choice of two cyclically-adjacent
rows from a block yields the same pair of rows (only 
the order is not clear, but it also does not matter).

Thus, to apply Proposition~\ref{prop:preserve}(3) we have to show:
\begin{quote}\emph{%
If one of the $r$ pairs of rows is deleted from the reduced matrix
\begin{equation*}
A'_{n,r}\ \ =\ \ 
\left(
\raisebox{-25mm}{\includegraphics{era137el-pict-13}}
\right)
\ \ \in\ \R^{2r\times (2r-4)},
\end{equation*} 
then the remaining $2r-2$ rows\\
{\rm(a)} \ span $\R^{2r-4}$, and\\
{\rm(b)} \ have a linear dependence with strictly positive coefficients.}
\end{quote}

Let us establish (b) first. 
For this, let 
\[
\alpha_k\ :=\ 2^k+2^{-k}-2\qquad\textrm{and}\qquad
\beta_k \ :=\ 2^k+\tfrac54 2^{-k}-\tfrac94.
\]
These sequences are designed to be non-negative,
$\alpha_k,\beta_k\ge0$ for all $k\in\Z$, with equality only for $k=0$.
Thus for (b) it suffices to verify that
\begin{quote}\emph{%
  For any $1\le t\le r$, the rows of $A'_{n,r}$ 
  are positively dependent with coefficients $\alpha_{k-t}$ for the
  even-index row from the $k$-th block, and $ \beta_{k-t}$ for the
  odd-index row from the $k$-th block,}
\end{quote}
since the (two) vectors in the $t$-th block thus get
zero coefficients, so they may be
deleted from any linear dependence (with otherwise positive
coefficients).
Thus we are led to the condition
\[
\alpha_{k-1}\vv_0 + 
\alpha_k    \ww_0 + \beta_k    \ww_1 + 
\alpha_{k+1}\uu_0 + \beta_{k+1}\uu_1 \ =\ \zero,
\]
which is needed to hold for $k\le |r-2|$, but which we impose for all
$k\in\Z$.
The choice of vectors $\vv_0,\ww_0,\ww_1,\uu_0,\uu_1$ 
is designed to satisfy this condition. Indeed,
except for the choice of a basis, which we took to be  
$\vv_0=(1,0)$ and $\ww_0=(0,1)$, the 
configuration of five vectors $\vv_0,\ww_0,\ww_1,\uu_0,\uu_1$
is uniquely determined by the condition. 

For property (a), we have to show that if one of the $r$ pairs of rows
is deleted from the matrix $A'_{n,r}$, 
then the resulting matrix still has full rank.
If the first or the second pair of rows is deleted,
then we still have the last $2r-4$ rows, and they
form a block upper triangular matrix, which has full
rank since its diagonal block
\begin{equation*} 
\Big(
\raisebox{-2.5mm}{\includegraphics{era137el-pict-14}}
\Big)
\end{equation*} 
is non-singular.
If a later pair of rows is deleted, then we are faced
with the task to show that the $2k\times 2k$ matrices 
$M_k$ of the form
\begin{equation*}
M_k\ \ :=\ \ 
\left(
\raisebox{-15.5mm}{\includegraphics{era137el-pict-15}}
\right)
\ \ \in\ \R^{2k\times 2k}
\end{equation*} 
are non-singular. To verify this
(without proving explicitly that
$\det M_k=\frac{(2^k-1)^2}{3^k}$, which
might need combinatorial ingenuity)
we use our knowledge about row combinations
of $M_k$. Indeed, if we sum the rows of $M_k$ with
coefficients 
$(\alpha_0,\beta_0,\alpha_1,\ldots,\break\alpha_{k-1},\beta_{k-1})$,
then this will result in the linear combination
of the three rows of the matrix
\begin{equation*}
H_3\ \ =\ \ 
\left(
\raisebox{-4mm}{\includegraphics{era137el-pict-16}}
\right)\ \in\ \R^{3\times2r}
\end{equation*} 
with the coefficients $(-\alpha_{-1},-\alpha_k,-\beta_k)$,
since $\vv_1=\zero$. Similarly, 
if we sum the rows of $M_k$ with coefficients 
$(\alpha_1,\beta_1,\alpha_2,\ldots,\alpha_k,\beta_k)$,
then we get a linear combination of the same three rows,
with coefficients $(-\alpha_0,-\alpha_{k+1},-\beta_{k+1})$.
And if we use coefficients
$(\alpha_2,\beta_2,\alpha_3,\ldots,\alpha_{k+1},\beta_{k+1})$
to sum the rows of $M_k$, then the result will be a sum 
with coefficients $(-\alpha_1,-\alpha_{k+2},-\beta_{k+2})$.
The coefficient matrix
\[
\begin{pmatrix}
-\alpha_{-1} & -\alpha_k & -\beta_k\\
-\alpha_0 & -\alpha_{k+1} & -\beta_{k+1}\\
-\alpha_1 & -\alpha_{k+2} & -\beta_{k+2}
\end{pmatrix}
\]
is non-singular for $k\ge0$: its determinant is
$\tfrac38(2^k-1+2^{-k-2})$.
Thus the full row space
of $H_3$ is contained in the row space of $M_k$.
In particular, we find the unit vectors
$\ee_{2k-1},\ee_{2k}\in\R^{2k}$
in the row space of $H_3$, and thus of $M_k$, and
this allows us to complete the argument by induction.
\end{proof}


\section{Flag vectors}\label{sec:flag}

The following result includes Theorem~\ref{thm:f-vectors}.  Note that
its proof relies only on the properties of~$\pi(P_n^{2r})$ that are
guaranteed by the statement of Theorem~\ref{thm:ppp}; it does not
refer to the specific combinatorial type of the polytopes constructed
in Section~\ref{sec:construction}, in our proof of
Theorem~\ref{thm:ppp}.

\begin{theorem}\label{cor:flag}
The $4$-polytope $\pi(P_n^{2r})$
has the flag vector 
\begin{align*}
(f_0,f_1,f_2,f_3;f_{03}) & =  
(n^r,rn^r, \tfrac54rn^r\!-\tfrac32n^r\!+rn^{r-1}, 
           \tfrac14rn^r\!-\tfrac12n^r\!+rn^{r-1}; 4rn^r\!-4n^r)\\
&= (4n, 4rn,   5rn-6n+4r, rn-2n+4r; 16rn-16n)\cdot \tfrac14 n^{r-1}.
\end{align*}
\end{theorem}

\begin{proof}
We obtain $f_0=n^r$ and $f_1=rn^r$ from the
products $(C_n)^r$, which are simple $2r$-polytopes
with $n^r$ vertices. With the abbreviation $N:=\frac14n^{r-1}$
this yields $f_0=4nN$ vertices and $f_1=4rnN$ edges for $\pi(P_n^{2r})$.

The products $(C_n)^r$ have $P:=rn^{r-1}=4rN$ polygon $2$-faces.
In the projection, all these are preserved,
in addition to some of the quadrilateral $2$-faces.

The projected polytope has two types of facets.
There are ``prism'' facets, which involve two of the
polygons, as well as ``cube'' facets,
which in $(C_n)^r$ arise as products of three edges and
$r-3$ vertices, but contain no polygon $2$-faces.
Thus each prism facet is bounded by two polygons,
and each polygon lies in two prism facets.
Hence there are $P=4rN$ prism facets, as well as
some number $C\ge0$ of cube facets.

Now double counting of ridges yields
$6C+(n+2)P=2f_2$. Thus with the Euler equation we get
$C=\frac14(r-2)n^r=(rn-2n)N$.
Finally, counting the vertex-facet incidences according to facets
yields $f_{03}=8C+2nP=(8rn-16n+8rn)N$.
\end{proof}

\begin{thebibliography}{10}

\bibitem{Z51a}
{\sc N.~Amenta and G.~M. Ziegler}, {\em Deformed products and maximal shadows},
Advances in Discrete and Computational Geometry (South Hadley, MA, 1996),
B.~Chazelle, J.~E. Goodman, and R.~Pollack, eds., vol.~223 of Contemporary
Mathematics, Amer. Math. Soc., Providence, RI, 1998, pp.~57--90.
\MR{1661377 (2000a:52019)}

\bibitem{Bar4}
{\sc D.~W. Barnette}, {\em Projections of $3$-polytopes}, Israel J. Math. 8
(1970), pp.~304--308.
\MR{0262923 (41:7528)}

\bibitem{Bay}
{\sc M.~M. Bayer}, {\em The extended $f$-vectors of $4$-polytopes}, J.
Combinatorial Theory, Ser. A, 44 (1987), pp.~141--151.
\MR{0871395 (88b:52009)}

\bibitem{BaBi}
{\sc M.~M. Bayer and L.~J. Billera}, {\em Generalized {D}ehn-{S}ommerville
relations for polytopes, spheres and {E}ulerian partially ordered sets},
Inventiones Math. 79 (1985), pp.~143--157.
\MR{0774533 (86f:52010b)}

\bibitem{Z80}
{\sc D.~Eppstein, G.~Kuperberg, and G.~M. Ziegler}, {\em Fat $4$-polytopes and
fatter $3$-spheres}, Discrete Geometry: {I}n honor of {W. Kuperberg}'s
60th birthday, A.~Bezdek, ed., vol.~253 of Pure and Applied Mathematics,
Marcel Dekker Inc., New York, 2003, pp.~239--265.
\newblock \texttt{arXiv:math.CO/0204007}.
\MR{2034720 (2004j:52009)}

\bibitem{Gevay}
{\sc G.~G\'evay}, {\em Kepler hypersolids}, Intuitive Geometry (Szeged,
1991), vol.~63 of Colloq. Math. Soc. J\'anos Bolyai, Amsterdam, 1994,
North-Holland, pp.~119--129.
\MR{1383617 (97a:52013)}

\bibitem{Gr1-2}
{\sc B.~Gr{\"u}nbaum}, {\em Convex {P}olytopes}, vol.~221 of Graduate Texts in
Math., Springer-Verlag, New York, 2003.
Second edition edited by V. Kaibel, V. Klee and G. M. Ziegler
(original edition: Interscience, London 1967).
\MR{1976856 (2004b:52001)}

\bibitem{Z59}
{\sc A.~H\"oppner and G.~M. Ziegler}, {\em A census of flag-vectors of
$4$-polytopes}, in Polytopes --- Combinatorics and Computation, G.~Kalai and
G.~M. Ziegler, eds., vol.~29 of DMV Seminars, Birkh\"auser-Verlag, Basel,
2000, pp.~105--110.
\MR{1785294 (2001e:52026)}

\bibitem{Z62}
{\sc M.~Joswig and G.~M. Ziegler}, {\em Neighborly cubical polytopes}, Discrete
Comput. Geometry (Gr\"unbaum Festschrift (G. Kalai, V. Klee, eds.))
24 (2000), pp.~325--344.
\newblock \texttt{arXiv:math.CO/9812033}.
\MR{1758054 (2001f:52019)}

\bibitem{kalai87:_rigid_i}
{\sc G.~Kalai}, {\em Rigidity and the lower bound theorem, {I}}, Inventiones
Math. 88 (1987), pp.~125--151.
\MR{0877009 (88b:52014)}

\bibitem{paffenholz-pc}
{\sc A.~Paffenholz}, {\em New polytopes from products}.
Preprint, TU Berlin, November 2004, 22~pages. 
\texttt{arXiv:math.MG/0411092}.

\bibitem{Z89}
{\sc A.~Paffenholz and G.~M. Ziegler}, {\em The {$E_t$}-construction for
lattices, spheres and polytopes}.
{Discrete Comput. Geometry (Billera Festschrift 
(M. Bayer, C. Lee, B. Sturmfels, eds.))}, in
print; published online August 23, 2004; \texttt{arXiv:math.MG/0304492}.

\bibitem{Schla}
{\sc L.~Schl\"afli}, {\em Theorie der vielfachen Kontinuit\"at}, Denkschriften
der Schwei\-ze\-ri\-schen naturforschenden Gesellschaft, Vol.~38, pp.~1--237,
Z\"urcher und Furrer, Z\"urich, 1901.
Written in 1850--1852. Reprinted in: Ludwig Schl\"afli, 1814--1895,
Gesammelte Mathemati\-sche Abhandlungen, Vol. I, Birkh\"auser, Basel, 1950, pp.
167--387.
\MR{0034587 (11:611b)}

\bibitem{schroeder04}
{\sc T.~Schr\"oder}, {\em On neighborly cubical spheres and polytopes}.
Work in progress, TU Berlin, 2004.

\bibitem{Sta7}
{\sc R.~P. Stanley}, {\em Generalized $h$-vectors, intersection cohomology of
toric varieties, and related results}, Commutative Algebra and
Combinatorics, M.~Nagata and H.~Matsumura, eds., vol.~11 of Advanced Studies
in Pure Mathematics, Kinokuniya, Tokyo, 1987, pp.~187--213.
\MR{0951205 (89f:52016)}

\bibitem{Stei3}
{\sc E.~Steinitz}, {\em {\"U}ber die {E}ulerschen {P}olyederrelationen}, Archiv
f\"ur Mathematik und Physik 11 (1906), pp.~86--88.

\bibitem{Z35}
{\sc G.~M. Ziegler}, {\em Lectures on {P}olytopes}, vol.~152 of Graduate Texts
in Mathematics, Springer-Verlag, New York, 1995.
Revised edition, 1998; ``Updates, corrections, and more'' at
\texttt{www.math.tu-berlin.de/~ziegler}.
\MR{1311028 (96a:52011)}

\bibitem{Z82}
\bysame, {\em Face numbers of
$4$-polytopes and $3$-spheres}, Proceedings of the International Congress
of Mathematicians (ICM 2002, Beijing), L.~Tatsien, ed., vol.~III, 
Higher Education Press, Beijing,
2002, pp.~625--634.
\newblock \texttt{arXiv:math.MG/0208073}.
\MR{1957513 (2003i:00010b)}

\bibitem{Z99}
\bysame, {\em Convex polytopes:
{E}xtremal constructions and $f$-vector shapes}.
Park City Mathematical Institute (PCMI 2004) Lecture Notes. With an
Appendix by Th. Schr\"oder and N. Witte, 2004.
Preprint, TU Berlin, November 2004, 73 pages.

\end{thebibliography}

\end{document}