Finding positive matrices subject to linear restrictions (Q996327): Difference between revisions

The problem of finding a positive definite \(l \times l\) matrix \(X=(x_{jk})_{j,k=1}^l\) of numbers \(x_{jk}\) satisfying a given set of \(n\) nonhomogeneous linear conditions: \[ \sum_{j,k=1}^l \alpha_{ijk}x_{jk}=\beta_i,\quad i=1,2,\dots,n, \] is studied. The author characterizes the existence of a such positive definite matrix by the existence of a minimum for an associated function \(V\), smooth and strictly convex on \(\mathbb R^n\). If there exist solutions \(X>0\), then \(\lim_{\| x\| \rightarrow \infty} V(x)= +\infty\) and the critical point \(x^0\) of \(V\) can be approximated by the conjugate gradient method. Knowing \(x^0\) provides, by a simple analytic formula, the unique solution \(X\) maximizing the entropy \[ -tr(X \ln{X})=-\sum_{j=1}^l \lambda_j \ln{\lambda_j}, \] where \(\lambda_1, \dots, \lambda_l\) are the eigenvalues of \(X\) and \(tr\) denotes the trace of a matrix, subject to the given restrictions. The author also presents related results in the semipositive definite case.