Efficient Implementation of Adaptively Biased Molecular Dynamics

Thursday, July 26, 2007 - 11:40am - 12:00pm
EE/CS 3-180
Celeste Sagui (North Carolina State University)
Joint work with Volodymyr Babin and Christopher Roland
(Department of Physics, North Carolina State University,
Raleigh, NC 27695-8202).

The Adaptively Biased Molecular Dynamics (ABMD) method
presented here corresponds to the general class of
nonequilibrium dynamics methods that use the history of the
sampling process to bias the dynamics, thereby effectively
flattening the free energy surface of the chosen order
parameter(s). These adaptive methods, are a generalization of
umbrella sampling with a time-dependent potential, and include
the Wang-Landau approach, the adaptive biasing force method,
and non-equilibrium metadynamics (MTD). Recently, we
implemented a variation of MTD in the AMBER package [V. Babin,
C. Roland, T.A. Darden and C. Sagui, J. Chem. Phys.
125, 204909
(2006)]. In spite of the efficient implementation, MTD still
has a square dependence on the sampling time t, i.e., scales as
O(t2) , which can be a significant problem in large systems
with non-negligible entropy. In addition, the large number of
parameters in MTD influences the output in an entangled way.
In the present work, we seek to cure these problems by carrying
out an efficient implementation of ABMD. The method has only
two parameters and scales linearly with time, i.e., it is O(t).
Comparative results for the folding of small peptides are