Natural density distribution of Hermite normal forms of integer matrices

From MaRDI portal
Publication:640872



Abstract: The Hermite Normal Form (HNF) is a canonical representation of matrices over any principal ideal domain. Over the integers, the distribution of the HNFs of randomly looking matrices is far from uniform. The aim of this article is to present an explicit computation of this distribution together with some applications. More precisely, for integer matrices whose entries are upper bounded in absolute value by a large bound, we compute the asymptotic number of such matrices whose HNF has a prescribed diagonal structure. We apply these results to the analysis of some procedures and algorithms whose dynamics depend on the HNF of randomly looking integer matrices.


The author analyses the probability that the Hermite Normal Form (HNF) of a random \(n\times m\) integer matrix has a given diagonal. The starting point is Hermite's result which states that for any \(n\times m\) integer matrix \(A\) there exists an \(n\times n\) unimodular matrix \(U\) and a unique \(n\times n\) integer matrix \(H\) in HNF such that \(A=UH\); \(H\) is called the HNF of \(A\) and it has \(r\leq n\) non-zero rows, for each row \(i\leq r\) the first non-zero entry \(h_{ij_i}\) is strictly positive, \(j_1<j_2<\cdots<j_r\) and for each \(1\leq k\leq i\leq r\) the entries \(h_{kj_i}\) of \(H\) satisfy \(0\leq h_{kj_i}<h_{ij_i}\). An algorithm is proposed which compute the HNF of a matrix \(A\). The paper presents the concept of natural density in \({\mathbb Z}^k\) which is used to obtain the main result of the paper (Proposition 6) which provides the frequency of appearance of a given non-zero diagonal in the HNF of a random looking matrix and shows that the density of these HNFs is 0. A corollary proves that given \(d_1,d_2,\dots,d_{n-1}\in{\mathbb N}^*\), the natural density of \(n\times n\) matrices whose HNF has the diagonal \((d_1,d_2,\dots,d_{n-1},\det A/\prod_{i=1}^{n-1}d_i)\) is \[ (\zeta(n)\zeta(n-1)\cdots\zeta(2)d_1^nd_2^{n-1}\cdots d_{n-1}^2)^{-1}, \] where \(\zeta\) is the usual zeta function. Two applications are presented, concerning the selection of random lattices that arise in cryptology, respectively the distribution of gcd\((\det[A|x],\det[A|y])\), where \(A\) is a random looking \(n\times(n-1)\) integer matrix and \(x,y\) are two \(n\)-vectors.











This page was built for publication: Natural density distribution of Hermite normal forms of integer matrices

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q640872)