arXiv++ Combinatorics

Voronoi defined two polyhedral partitions of the cone of se\mi\de\fi\nite forms into L-type domains and into perfect domains. Up to equivalence, there is only one domain that is simultaneously perfect and L-type. Voronoi called this domain {\em principal}. We show that closure of the principal domain may be identified with a cone of cut submodular set functions. Parallelotopes of the closed principal domain are zonotopes that are base polyhedra related to graphic unimodular sets of vectors.

A parallelotope is a polytope whose translation copies fill space without gaps and intersections by interior points. Voronoi conjectured that each parallelotope is an affine image of the Dirichlet domain of a lattice, which is a Voronoi polytope. We give several properties of a parallelotope and prove that each of them is equivalent to it is an affine image of a Voronoi polytope.

There are several works \cite{De} (and \cite{St}), \cite{En}, \cite{Co} and \cite{Va} enumerating four-dimensional parallelotopes. In this work we give a new enumeration showing that any four-dimensional parallelotope is either a zonotope or the Minkowski sum of a zonotope with the regular 24-cell $\{3,4,3\}$. Each zonotopal parallelotope is the Minkowski sum of segments whose generating vectors form a unimodular system. There are exactly 17 four-dimensional unimodular systems. Hence, there are 17 four-dimensional zonotopal parallelotopes. Other 35 four-dimensional parallelotopes are: the regular 24-cell $\{3,4,3\}$ and 34 sums of the regular parallelotope with non-zero zonotopal parallelotopes. For the nontrivial enumerating of the 34 sums we use a theorem discribing necessary and sufficient conditions when the Minkowski sum of a parallelotope with a segment is a parallelotope.

Voronoi conjectured that any parallelotope is affinely equivalent to a Voronoi polytope. A parallelotope is defined by a set of $m$ facet vectors $p_i$ and defines a set of $m$ lattice vectors $t_i$, $1\le i\le m$. We show that Voronoi's conjecture is true for an $n$-dimensional parallelotope $P$ if and only if there exist scalars $γ_i$ and a positive definite $n\times n$ matrix $Q$ such that $γ_i p_i=Qt_i$ for all $i$. In this case the quadratic form $f(x)=x^TQx$ is the metric form of $P$. As an example, we consider in detail the case of a zonotopal parallelotope. We show that $Q=(Z_βZ^T_β)^{-1}$ for a zonotopal parallelotope $P(Z)$ which is the Minkowski sum of column vectors $z_j$ of the $n\times r$ matrix $Z$. Columns of the matrix $Z_β$ are the vectors $\sqrt{2β_j}z_j$, where the scalars $β_j$, $1\le j\le r$, are such that the system of vectors $\{β_jz_j:1\le j\le r\}$ is unimodular. $P(Z)$ defines a dicing lattice which is the set of intersection points of the dicing family of hyperplanes $H(j,k)=\{x:x^T(β_jQz_j)=k\}$, where $k$ takes all integer values and $1\le j\le r$.

Non-rigidity degree of a lattice $L$, nrd$L$, is dimension of the L-type domain to which $L$ belongs. We complete here the table of nrd's of all root lattices and their duals; namely, the hardest remaining case of $D_n^*$, and the case of $E_7^*$ are decided. We describe explicitly the $L$-type domain ${\cal D}(D_n^*)$, $n \ge 4$. For $n$ odd, it is a non-simplicial polyhedral open cone of dimension $n$. For $n$ even, it is one-dimensional, i.e. for even $n$, $D_n^*$ is an edge form.