Moving
hills - new parallelizable metadynamic method with
fixed amount of bias potential
Department
of Biochemistry and Microbiology, Faculty of Food and Biochemical Technology,
Institute of Chemical Technology, 166 28 Prague, Czech Republic
hosekp@vscht.cz
In
the last seven years, there has been significant slow-down in the increase of
computational power of single processor core. This limitation in molecular
simulations was overcome by parallelization. But parallelization of the
molecular dynamics and so the metadynamics, which we
are focus on, is also limited, because of an increased time spent for interprocess communication. One way to overcome the problem
is to reduce amount of data, which need to be transferred between processes.
Our contribution to this problematics is Moving hills method. This method use multiple parallel
simulation, same as Multiple walkers, but only one bias potential is assign to
each parallel and position of the potential is updated according immediate
state of the parallel. So the bias hill travels on the energy landscape as the
current conformation is evolving. Shape of bias potential summed from all parallels
converges as the system reaching an equilibration. The method was tested on
small molecule and results were compared with other metadynamic
methods. No significant differences in obtained free energy landscape were
found.
This project was supported by COST actions MultiGlycoNano (CM1102, LD13024) and GLISTEN (CM1207, LD14133). Access to computing and storage facilities MetaCentrum (LM2010005) and CERIT-SC (CZ. 1.05/3.2.00/08.0144) is greatly appreciated. Participation at the conference is supported by specific university research (MSMT No 21/2014).