From: JťrŰme Hťnin (jhenin_at_ifr88.cnrs-mrs.fr)
Date: Fri Jan 27 2012 - 09:10:44 CST
2012/1/27 Ajasja Ljubetińć <ajasja.ljubetic_at_gmail.com>:
>> If you model the unbiased process as 2D diffusion on an effective
>> potential (i.e. the PMF), then ABF will (in time) erase the barriers, but
>> not change the intrinsic diffusion properties.
> Actually what I would like to do now is to model the (unbiased) dynamic
> behaviour of my colvars.
That is an intrinsically difficult question. A couple of colleagues
have been thinking hard about it for a while now, and they still don't
have a clear answer. For now, I would say that using ABF (and the
whole family of biased methods) essentially means that you give up on
getting dynamical information.
> I¬†would¬†like to¬†know¬†how far my colvars move in 10
> ns. I plan to preform a random walk on the obtained PMF, where each step
> will be accepted¬†according¬†to the Metropolis criterion (ie the probability
> of accepting the step is min(1,exp(dE/(kT)))). Since such simulations are
> computationaly¬†extremely¬†cheap I could get excellent sampling.
Sampling, certainly, but you already have that from ABF. And those
simulations would not give you access to any kind of accurate
dynamics. You might think of a kind of diffusive dynamics, using
diffusion coefficients to be determined (see below).
> The only
> thing I have to figure out is what is the (real) time of one step is.¬†Am I
> correct in guessing that it is¬†proportional¬†to
> the¬†autocorrelation¬†time¬†obtained¬†from the colvars¬†trajectory?
In a sense, yes, since that would be related to the diffusion
coefficient. There are several ways to access that, but I have to warn
you to be wary of the suitability of the dynamical model to begin
The only serious attempt I have made to go back to dynamics based on
ABF results was in the glycerol-in-GlpF work:
There is an attempt at predicting a mean first passage time, and
hence, the channel conductance - I am not really convinced about that
particular result. The method we used to get position-dependent
diffusion coefficients is from Gerhard Hummer, it is particularly
simple and effective.
If at all possible, I'd recommend getting dynamical data from unbiased
This archive was generated by hypermail 2.1.6 : Tue Dec 31 2013 - 23:21:36 CST