TCBG Seminar

Computing Potentials of Mean Force Using the Adaptive Biasing Force Method

Dr. Jerome Henin
Equipe de chimie et biochimie theoriques
Universite Henri Poincare
Nancy, France

Wednesday, June 18, 2003
12:10 am (CT)
3269 Beckman Institute


Computing free energy variations along one particular degree of freedom requires an efficient sampling of phase space both along the coordinate of interest and other degrees of freedom. Several techniques were devised to achieve this, mainly by helping the system to cross free energy barriers, as in umbrella sampling. However, umbrella sampling relies on an initial guess about the free energy profile, which isn't always feasible. Other techniques constrain the selected coordinate, or decouple it form the other degrees of freedom. But then it is far from straightforward to assert that these other DOF are allowed to relax. In a nutshell, without a correct sampling, we do not know the free energy profile, and without a good estimate of this free energy profile, it is often very hard to ensure adequate sampling. The Adaptive Biasing Force (ABF) method was developed by Eric Darve and Andrew Pohorille as an attempt to solve this hen-and-egg problem. The derivative of the configurational free energy A with respect to a coordinate q can be expressed as the opposite of the average force exerted on this coordinate. The idea in ABF is to accumulate as a function on q, while applying a biasing force based on the current estimate of . Thus, the method requires no prior knowledge of the PMF. When converges towards dA/dq, the biasing force converges towards the "ideal" bias, corresponding to the opposite of the PMF. Applying this method to several test problems, the authors obtained almost homogenous sampling along q and reproduced accurately the well-known results.

Main TCBG Seminars page