Variationally enhanced sampling with permutationally invariant collective variables
Date issued
Authors
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
License
Abstract
Molecular dynamics (MD) simulations have become an indispensable
tool in understanding the physical world at a resolution of molecular
details. Currently, MD simulations are still limited to the microsecond
timescale in most circumstances. When there is a high free energy
barrier between metastable states, sampling is easily trapped in a local
free energy minimum. A variety of enhanced sampling methods and
algorithms have been developed to overcome such issues. In this thesis,
we present our work on both developments of algorithms and methods.
In terms of algorithm development, we have created an interface
between PLUMED 2, a software package that has implemented some
of the most prominent enhanced sampling methods, and the MD
engine ESPResSo++. We used the combination of the two packages to
study the first-order phase transition of the 128 monomer single-chain
smooth square-well polymer. In terms of method development, based
on the variationally enhanced sampling method, we have created
the variationally enhanced sampling with permutationally invariant
collective variables method so that such local collective variables can
be used in biased simulations. We have demonstrated the effectiveness
of the new method in phase transition studies of seven Lennard-
Jones particles in two-dimensional space and crystallization of bulk
sodium. We have also explored crystallization of ice and urea from
melt with the new method and discussed the limitations of the current
implementation encountered in these works.