Rank-deficient matrices as a computational tool (Q4383440)

Parameter-dependent matrices \(A(\alpha)\), \(\alpha \in \mathbb R^k\), are considered. A method based on a bordered extension of \(A(\alpha)\) is proposed to detect rank changes and to compute \(\alpha\) for which \(A(\alpha)\) has a prescribed (low) rank deficiency. In detail, a square non-singular matrix \(M(\alpha)\) is created by adding \(m_1\) rows and \(m_2\) columns to \(A(\alpha)\). Then, auxiliary systems \(M(\alpha)b_i(\alpha)=c_i\), \(i=1,\dots,m\), are introduced and solved. By means of \(b_i(\alpha)\), a matrix \(G(\alpha)\in \mathbb R^{m_2\times m_1}\) having the same rank defect as \(A(\alpha)\) is set up. If \(m_1\) and \(m_2\) are chosen minimal then \(G(\alpha)\) is the zero matrix. Thus \(G(\alpha)\) can be examined instead of \(A(\alpha)\) as illustrated by Example 1. Using this approach and adding some more analysis, the authors solve a minimum-norm least squares problem (Example 2). In Example 3, singularities characterized by the rank defect of the Jacobian matrix corresponding to a discretized reaction-diffusion equation (the Brusselator) are detected. Example 4 shows detection of the Hopf points in a eutrophication model (nonlinear ordinary differential equations).