Title: Dynamical simulation methods for canonical sampling:
Toward improved convergence in biomolecular dynamics

abstract:
Numerical methods for dynamical simulation of atomic and molecular
systems in physics and chemistry are generally not developed with
traditional notions of accuracy in mind.  Very long simulation
intervals, together with computationally costly evaluations of system
forces, lead practitioners to use low-order time stepping algorithms.
Of much greater interest than accuracy are statistical properties of
the trajectories, which are used to compute average quantities of
physical importance in the model.

In this talk, we will discuss widely-used dynamical models which
produce trajectories consisting of microstates sampled from the
constant-energy microcanonical ensemble, as well as the canonical
ensemble, which is characterized by coupling of the model system to a
heat bath.  We examine two features necessary for any discussion of
statistical convergence in computed trajectories: efficient phase
space sampling and equipartition of kinetic energy.  We propose new
thermostatting models based on the Nose Hamiltonian that address
these questions.