On the rank of a random binary matrix

From MaRDI portal





Summary: We study the rank of a random \(n \times m\) matrix \(\mathbf{A}_{n,m;k}\) with entries from \(GF(2)\), and exactly \(k\) unit entries in each column, the other entries being zero. The columns are chosen independently and uniformly at random from the set of all \(\binom{n}{k}\) such columns. We obtain an asymptotically correct estimate for the rank as a function of the number of columns \(m\) in terms of \(c,n,k\), and where \(m=cn/k\). The matrix \(\mathbf{A}_{n,m;k}\) forms the vertex-edge incidence matrix of a \(k\)-uniform random hypergraph \(H\). The rank of \(\mathbf{A}_{n,m;k}\) can be expressed as follows. Let \(|C_2|\) be the number of vertices of the 2-core of \(H\), and \(|E(C_2)|\) the number of edges. Let \(m^*\) be the value of \(m\) for which \(|C_2|= |E(C_2)|\). Then w.h.p. for \(m<m^*\) the rank of \(\mathbf{A}_{n,m;k}\) is asymptotic to \(m\), and for \(m \ge m^*\) the rank is asymptotic to \(m-|E(C_2)|+|C_2|\). In addition, assign i.i.d. \(U[0,1]\) weights \(X_i, i \in{1,2,\ldots m}\) to the columns, and define the weight of a set of columns \(S\) as \(X(S)=\sum_{j \in S} X_j\). Define a basis as a set of \(n-\mathbb{1} (k\text{ even})\) linearly independent columns. We obtain an asymptotically correct estimate for the minimum weight basis. This generalises the well-known result of \textit{A. M. Frieze} [Discrete Appl. Math. 10, 47--56 (1985; Zbl 0578.05015)] that, for \(k=2\), the expected length of a minimum weight spanning tree tends to \(\zeta(3)\sim 1.202\).











This page was built for publication: On the rank of a random binary matrix

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2327226)