A curvilinear method based on minimal-memory BFGS updates (Q711340)

For solving large scale smooth unconstrained optimization problems the authors propose an algorithm based on a modification of the minimal-memory BFGS method, that incorporates a curvilinear and linesearch method. To avoid the computationally expensive factorization and the storage of a matrix they utilize the L-BFGS method for computation of an approximate Hessian matrix, using information from the most previous iterations. This specific memoryless BFGS update is providing to determine analytically the eigenvalues of the approximate Hessian matrix B explicitly. Thus, the computation of the pair of search directions is obtained by performing a sequence of izmer products and vector summations. Finally, some results of numerical experiments for large scale problems are presented.