Re: restarting ABF simulation

From: Jerome Henin (jhenin_at_cmm.chem.upenn.edu)
Date: Tue Apr 15 2008 - 13:16:28 CDT

Vaithee,
In the cases where you have duplicated data in the average force F,
you can subtract in the final PMF it by calculating something like:
F = ( N2F2 - N1F1 ) / (N2 - N1)
where F1 and F2 are the average force at point 1 and 2 in the
simulation, and N1 and N2 the corresponding number of samples.
Jerome

On Tue, Apr 15, 2008 at 12:16 PM, Subramanian Vaitheeswaran
<vaithee_at_umd.edu> wrote:
> Jerome,
>
> I use "abf inFiles" to combine data from multiple trajectories as well. For example, in SET1:
> trajectory1: abf inFiles{}
> trajectory2: abf inFiles{}
> trajectory3: abf inFiles{./trajectory1/abf.dat ./trajectory2/abf.dat}
>
> And in SET2:
> trajectory1: abf inFiles{../SET1/trajectory3/abf.dat}
> trajectory2: abf inFiles{../SET1/trajectory3/abf.dat}
> trajectory3: abf inFiles{./trajectory1/abf.dat ./trajectory2/abf.dat ../SET1/trajectory3/abf.dat}
>
> The multiple trajectories improve sampling and the input for the pmf calculation in each trajectory is the best current estimate of the pmf. I realize that the data from SET1 (and from SET2 if I run a third set) is being duplicated, but the sampling is still from the same ensemble and the duplication should have a negligible effect. Do you see anything wrong with this?
>
> thanks,
> Vaithee
>
>
> ---- Original message ----
> >Date: Mon, 14 Apr 2008 14:58:57 -0400
> >From: "Jerome Henin" <jhenin_at_cmm.chem.upenn.edu>
> >Subject: Re: namd-l: restarting ABF simulation
>
>
> >To: "Subramanian Vaitheeswaran" <vaithee_at_umd.edu>
> >Cc: namd-l_at_ks.uiuc.edu
> >
> >Vaithee,
> >It would actually be somewhat wrong, because in effect the data from
> >early runs would be duplicated and get an excessive statistical
> >weight. I am saying "somewhat" wrong because in theory, if you have
> >independent chunks of data sampled according to the canonical (or NPT)
> >ensemble, the reweighted average still obeys the right distribution.
> >In practice, though, it amounts to discarding some information from
> >the later runs.
> >Jerome
> >
> >On Mon, Apr 14, 2008 at 1:21 PM, Subramanian Vaitheeswaran
> ><vaithee_at_umd.edu> wrote:
> >> Hi Jerome,
> >>
> >> This brings up a question about "abf inFiles" I have had for some time. I thought including all the previous output files as Luca proposed for a sequence of restarted runs, is not wrong, merely unnecessary. It would be a big help if you could clarify this point.
> >>
> >> thanks,
> >> Vaithee
> >>
> >>
> >>
> >> ---- Original message ----
> >> >Date: Mon, 14 Apr 2008 09:54:07 -0400
> >> >From: "Jerome Henin" <jhenin_at_cmm.chem.upenn.edu>
> >> >Subject: Re: namd-l: restarting ABF simulation
> >> >To: "Luca Bellucci" <bellucci14_at_unisi.it>
> >> >Cc: namd-l_at_ks.uiuc.edu
> >> >
> >> >Luca:
> >> >No, you haven't clarified it! But it's not your fault. The manual is
> >> >not clear enough on this question...
> >> >
> >> >When we wrote "This command may also be used to combine data obtained
> >> >from separate runs.", we meant *independent* runs covering *different*
> >> >ranges of the reaction coordinate ("windows"). This does NOT apply to
> >> >a sequence of runs restarted from one another, because all previous
> >> >data provided as input is already carried over in the output.
> >> >
> >> >This means that under normal circumstances, inFiles should really have
> >> >only ONE argument.
> >> >
> >> >Jerome
> >> >
> >> >
> >> >On Mon, Apr 14, 2008 at 9:04 AM, Luca Bellucci <bellucci14_at_unisi.it> wrote:
> >> >> Hi Jerome,
> >> >> I have just read the manual....
> >> >> then to combine multiple abf runs I should to use
> >> >> abf inFiles md01-file.dat md02-file.dat md03-file.dat
> >> >> I hope to have clarified the use of "abf inFiles" command for other user.
> >> >> Anyway thanks for replay.
> >> >> Luca
> >> >>
> >> >>
> >> >>
> >> >> > Luca,
> >> >> > Just restart the NAMD simulation as you would normally do, providing
> >> >> > the same ABF parameters as in the first run, with the "abf inFiles"
> >> >> > parameter pointing to the previous ABF data file containing the free
> >> >> > energy, its gradient and the sampling histogram.
> >> >> > Jerome
> >> >> >
> >> >> > On Mon, Apr 14, 2008 at 5:14 AM, Luca Bellucci <bellucci14_at_unisi.it> wrote:
> >> >> > > Dear all,
> >> >> > > My simulation is prematurely finished (6ns of 10ns),
> >> >> > > How restart ABF simulation?
> >> >> > >
> >> >> > > any suggestions are well accepted
> >> >> > > Luca
> >> >> > >
> >> >> > >
> >> >> >
> >> >>
> >> >>
> >> >
> >>
>

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:49:23 CST