Parallel low-memory quasi-Newton optimization algorithm for molecular structure

0398455 - ÚOCHB 2014 RIV NL eng J - Journal Article
Klemsa, Jakub - Řezáč, Jan
Parallel low-memory quasi-Newton optimization algorithm for molecular structure.
Chemical Physics Letters. Roč. 584, Oct 1 (2013), s. 10-13. ISSN 0009-2614. E-ISSN 1873-4448
R&D Projects: GA ČR GP13-01214P
Institutional support: RVO:61388963
Keywords : geometry optimization * parallelization * molecular graph
Subject RIV: CF - Physical ; Theoretical Chemistry
Impact factor: 1.991, year: 2013

We present a novel parallel gradient optimization algorithm designed for the optimization of molecular geometry - the parallel preconditioned LBFGS (PP-LBFGS) method. In each step, several additional gradient calculations (performed in parallel with the calculation of the potential) are used to improve the most important elements of the Hessian. The sparsity of the connectivity matrix and the graph theory are used to estimate multiple Hessian elements from each additional gradient calculation. The simplest variant of the algorithm, which requires 4 gradient evaluations per cycle, converges 2x-4x faster than the LBFGS algorithm, depending on the size of the system.
Permanent Link: http://hdl.handle.net/11104/0225938