Force-field parametrization, multiple birds with multiple stones – with REST2
This blog post is a idea for a force-field parametrization strategy, which takes advantage of replica exchanges between different parametrizations of the same system.
One of the problems with force-field parametrization is that a large number of parameter space has to be searched. A series of parameter values is often evaluated to try and match an experimental observable. This evaluation is usually done sequentially, slowing down the process. Additionally, if the parameter choice introduces an energy barrier, it may not be possible to sample efficiently and obtain convergence of a numerical quantity, to be compared with experiment.
To enhance sampling, temperature replica exchange can be employed. Replicas of the same hamiltonian are run at different temperatures end exchanges are attempted according to the boltzman criterion. Importantly, the energies of the neighboring replicas are compared.
It should be advantageous, in the force-field parametrization to take advantage of this idea. Instead of running at multiple temperatures, one could run multiple hamiltonians (corresponding to different parameter sets of interest). Each replica is reduced to its potential energy, and potential energies between neighboring replicas can be compared.
The simplest example of this idea could involve taking an existing molecule model, such as the TIP3P water and introduce some perturbation λ generating replicas with similar VdW and charge parameters to the old model. For each replica, one can then compute an observable – density, heat capacity, whatever – compare to experiment and see which parameter set performs best. The ability to exchange with neighbor replicas is going to speedup convergence of the observable.
One of the problems with force-field parametrization is that a large number of parameter space has to be searched. A series of parameter values is often evaluated to try and match an experimental observable. This evaluation is usually done sequentially, slowing down the process. Additionally, if the parameter choice introduces an energy barrier, it may not be possible to sample efficiently and obtain convergence of a numerical quantity, to be compared with experiment.
To enhance sampling, temperature replica exchange can be employed. Replicas of the same hamiltonian are run at different temperatures end exchanges are attempted according to the boltzman criterion. Importantly, the energies of the neighboring replicas are compared.
It should be advantageous, in the force-field parametrization to take advantage of this idea. Instead of running at multiple temperatures, one could run multiple hamiltonians (corresponding to different parameter sets of interest). Each replica is reduced to its potential energy, and potential energies between neighboring replicas can be compared.
The simplest example of this idea could involve taking an existing molecule model, such as the TIP3P water and introduce some perturbation λ generating replicas with similar VdW and charge parameters to the old model. For each replica, one can then compute an observable – density, heat capacity, whatever – compare to experiment and see which parameter set performs best. The ability to exchange with neighbor replicas is going to speedup convergence of the observable.
Comments
Post a Comment