Mixing Double Precision Single Precision in QM/MM simulations
Title: Mixing Double Precision Single Precision in QM/MM simulations
SNIC Project: SNIC 2018/7-65
Project Type: SNIC Small Compute
Principal Investigator: Pedro Ojeda <pedroojeda2011@gmail.com>
Affiliation: Umeå universitet
Duration: 2018-11-06 – 2020-12-01
Classification: 10407
Keywords:

Abstract

We aim at using mixed double precision and single precision in expensive routines of matrix diagonalization, and forces accumulation within the QM/MM framework. These routines are the bottleneck in current QM/MM simulations where a self-consistent field procedure is required. A similar scheme for forces calculations is commonly used in classical MD simulations but this scheme has not been extensively explore in QM/MM simulations due to the tight criterion for the density matrix convergence.