Approximation of functions of large matrices with Kronecker structure (Q504162): Difference between revisions

This paper derives a computational strategy -- based on Krylov methods -- to evaluate matrix functions \(f({\mathcal A}) b\) effectively and efficiently where \({\mathcal A} = M_1 \otimes I + I \otimes M_2\) is a two term sum of Kronecker products, \(f\) is a regular function and \(b\) represents a low rank vectorized matrix. Specific examples involve derivations or numerical tests and comparisons with earlier methods for square root functions, exponential functions, the matrix inverse function, matrix sine and cosine functions, graph and network analyses, and for completely monotonic functions. Some convergence results are included, as well as a guide on how to deal similarly with multiterm Kronecker sums. The numerical experiments show the savings in memory and computer time when exploiting a Kronecker sum structure.