Project Details
Description
Abstract
Molecular Dynamics simulations are powerful
tools to study problems of materials science,
nanoscience, and biology. It naturally provides
ample opportunities for interdisciplinary research
that requires knowledge in mathematics, statistics,
computer science, physics, materials and biology.
The focus of this project is on developing learning-based
computational and statistical methods for potential
energy landscape modeling to accelerate ab-initio
molecular dynamics simulations. The set of tools
developed will substantially expand the limits of time
and system size without compromising the precision and
quality of the ab-initio simulation results.
Hongyuan Zha, Qiang Du, Runze Li and Jorge Sofo will
investigate learning and
computational methods 1) to characterize both the local and global
structures of the low-dimensional manifold in which the simulation
really occurs through manifold learning from the trajectories of the
ab-initio simulation; 2) to identify and extract suitable clusters in
the reduced dimension spaces corresponding to regions in the
configuration space that naturally emerge from the ab-initio
simulation and are visited frequently by the particles throughout the
simulation; 3) to conduct efficient energy and force interpolation
using Gaussian Kriging models with penalized likelihood. In this
learning and computational
framework, the interpolated potential energy surface will be
evaluated and it will replace the costly ab-initio evaluation when its
precision is good enough. As the simulation evolves, the interpolated
potential energy surface will be retested to detect the
eventual need of a retraining in case the simulation is exploring
new regions of the configuration space.
Status | Finished |
---|---|
Effective start/end date | 10/1/04 → 9/30/08 |
Funding
- National Science Foundation: $215,000.00