Re: abf convergence problems

From: Jorgen Simonsen (jorgen589_at_gmail.com)
Date: Thu Jan 22 2009 - 14:37:05 CST

ok - I have just expanded the calculation with 3 ns more - I know that the
answer to the question is that it depends on ..... as I don't have much
experience on the ABF - on average how long time would you expect for it to
converge?

On Thu, Jan 22, 2009 at 9:17 PM, Jerome Henin <jhenin_at_cmm.chem.upenn.edu>wrote:

> One last thing... with an 8-residue peptide, you probably shouldn't
> expect things to converge in 12 ns. ABF is taking care of one phi
> angle - convergence depends on what all the other degrees of freedom
> are doing.
>
> Jerome
>
> On Thu, Jan 22, 2009 at 1:56 PM, Jorgen Simonsen <jorgen589_at_gmail.com>
> wrote:
> > Hi,
> >
> >
> > I am using
> >
> > abf xiMin -180.0
> > abf xiMax 0.0
> >
> > with a dxi of 5 I have seen this size of sample in other simulations
> where
> > they got convergence of the angle.
> >
> > On Thu, Jan 22, 2009 at 6:52 PM, Jerome Henin <jhenin_at_cmm.chem.upenn.edu
> >
> > wrote:
> >>
> >> Hi Jorgen,
> >>
> >> I recommend running the calculation in two separate windows (say,
> >> [-180:0] and [0:180]). There are several reasons to do that:
> >> 1) it's easier to get even sampling in narrower windows
> >> 2) you can run them in parallel (linear scaling!)
> >> 3) it avoids the behavior where the angle always rotates in the same
> >> direction, which gives an asymmetric (and wrong) PMF.
> >>
> >> Jerome
> >>
> >> On Thu, Jan 22, 2009 at 12:16 PM, Jorgen Simonsen <jorgen589_at_gmail.com>
> >> wrote:
> >> > Hi all,
> >> >
> >> > I have been using the ABF methods in order to estimate some energy
> >> > around a
> >> > phi-angle in a peptide with 8 residues. I have applied the following
> >> > specifications
> >> >
> >> > fullSamples 5000
> >> > dSmooth 0.1
> >> > forceConst 1.0
> >> >
> >> > I have been running the simulation for 12 ns and the distribution is
> not
> >> > uniform and some of the gradients are around -0.15. Is there a way to
> >> > speed
> >> > up convergence - any comments or suggestions appreciated.
> >> >
> >> > Thanks in advance
> >> >
> >> > Jorgen
> >> >
> >
> >
>

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:50:24 CST