Re: Number of MC steps in ABF integrate gives entirely diferent surfaces

From: Branko (bdrakuli_at_chem.bg.ac.yu)
Date: Thu Jan 14 2010 - 05:30:42 CST

Jerome,

Thank you very much for useful advices. First, excellent young physicist
from our institute of physics recompile source for win, but we have some
problems, i.e. if one give only name of grad file (with extension) and
put both .count and .grad file in same directory program works and
use default parameters and obtain results. But, if one try to specify
parameters program do not work. I believe that his solve this within few
days and send him your attached CVS to work on this version.
Linux version recompiles my colleague and friend skilled in programming
in Italy (author of VegaZZ) and repeat calculation for comparison. In
any way he also must recompile is from source code. Also it is confuse
me a bit when I see in console that data on PBC do not exist (0,0),
while simulation where performed with PBC. When fully functional version
will be finished, I strongly recommended to our physicist to put this
information on namd-l, and with your consent, i believe that his
contribution could be included in some of future NAMD releases. So far
2 colvars will by sufficient for my purposes, if you have a time
please check backward mails from my e-mail addresses, I send some longer
mail with asking for some advices, but do not post this on NAMD list,
because problem is not of general interest.

Thank you very much again

Branko

Jerome Henin wrote:
> Branko,
>
> In my experience, typical 2D grids require at least a few million
> steps for convergence (which should still run fairly quickly). Even
> simpler, if you don't specify the number of steps, the program will
> try to run until convergence, as explained in the manual. It can be
> useful to start off with a higher hill height parameter (maybe 0.05
> kcal/mol) to quickly obtain a slightly coarser PMF. Note that
> integrating PMFs in dimension higher than 2 is somewhat more involved.
>
> Also, I strongly recommend that you use the updated version of the
> code from NAMD CVS (I am attaching it for convenience). It uses a
> different choice of boundary conditions (regions not-sampled in the
> ABF run are now excluded), which gives more consistent results for
> poorly-sampled gradient fields.
>
> By the way, what did you have to do to make the program work?
>
> Best
> Jerome
>
> 2010/1/14 Branko <bdrakuli_at_chem.bg.ac.yu>:
>
>> Jerome,
>>
>> With the little help of my friends skilled in programming abf_integrate is
>> now fictional, but when one use different number of MC steps (10000 vs.
>> 2000000) resulted free energy surfaces are entirely different. All other
>> parameters are holding on default --t 500 --m 1 --h 0.01 --f 0.5. Please advise
>> me how rationally and reasonably to choose number of steps to obtain
>> realistic results. Thanks a lot in advance.
>>
>> Branko
>>
>> ------------------------------------------------------------------------
>>
>>
>> No virus found in this incoming message.
>> Checked by AVG - www.avg.com
>> Version: 9.0.725 / Virus Database: 270.14.139/2619 - Release Date: 01/13/10 20:35:00
>>
>>

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 05:22:39 CST