Markowitz-type heuristics for computing Jacobian matrices efficiently

摘要: We consider the problem of accumulating Jacobian matrix a nonlinear vector function by using minimal number arithmetic operations. Two new Markowitz-type heuristics are proposed for vertex elimination in linearized computational graphs, and their superiority over existing approaches is shown several tests. Similar ideas applied to derive edge techniques. The well known can be observed only partially discussed this paper. Nevertheless, significant improvements achieved both terms quality results robustness with respect different tiebreaking criteria.