From: Michael Feig (feig_at_msu.edu)
Date: Tue Jan 20 2009 - 07:02:23 CST
Peter,
do you (or anybody else) know if the bug fixes that would allow TMD to work
in parallel are minor and can be easily applied to the 2.6 version?
If so, is it possible to post them to this list?
I haven't heard back about getting CVS access but I am not sure I really
want to deal with a 2.7 version that might have other issues (being a pre-
release version). If I can get TMD working with 2.6 that would allow me
to go forward with the simulations I plan to run.
Thanks,
Michael.
---
Michael Feig
Biochemistry & Molecular Biology
Chemistry
Computer Science & Engineering
Michigan State University, East Lansing, MI
> -----Original Message-----
> From: owner-namd-l_at_ks.uiuc.edu [mailto:owner-namd-l_at_ks.uiuc.edu] On Behalf
> Of Peter Freddolino
> Sent: Saturday, January 17, 2009 12:01 PM
> To: Michael Feig
> Cc: namd-l_at_ks.uiuc.edu
> Subject: Re: namd-l: TMD comments
>
> Dear Michael,
>
> Michael Feig wrote:
> > In learning about targeted MD (TMD) in NAMD I have
> > encountered a few issues that I would like to share
> > with the NAMD community.
> >
> > 1) I am interested in running TMD as a harmonic biasing
> > potential with a fixed reference RMSD for a certain time
> > (so that I can use appropriate equilibrium reweighting
> > techniques). This doesn't work with the current version
> > because the biasing term is only effective on one side
> > of the final RMSD target value. This may be a feature
> > for other purposes but in my case it makes TMD unusable.
> > This can be easily fixed by replacing
> >
> > if ((finalRMS < initialRMS && targetRMS <= curRMS) ||
> > (finalRMS >= initialRMS && targetRMS > curRMS)) {
> >
> > in lines 173-174 of GlobalMasterTMD.C with
> >
> > if (1) {
> >
> > A more elegant version could introduce an additional
> > keyword that allows a selection of a one-sided or two-sided
> > biasing potential.
>
> An even more elegant and general solution is present in the colvars
> module of namd (currently available in cvs, to be included with namd
> 2.7) contributed by Jerome Henin and Giacomo Fiorin. It allows the
> definition of several types of biasing potentials and their use with
> several biased dynamics methods. You should be able to find
> documentation in the ug subdirectory of the cvs tree.
>
> To answer your later question, the .tar.gz available through the webpage
> is *not* the cvs version; you need to instead follow the steps under the
> "Bleeding edge" heading at
> http://www.ks.uiuc.edu/Research/namd/development.html
>
> >
> > 2) The manual says that the biasing potential is
> >
> > U = k/2N [ RMS(t) - RMS*(t) ]^2
> >
> > After looking at the code and checking the derivatives I
> > think that this is incorrect. I think the potential is in
> > fact (as it should be):
> >
> > U = k/2 [ RMS(t) - RMS*(t) ]^2
> >
> > The 1/N factor comes in when calculating the derivatives/
> > forces, but the force constant in the potential does *not*
> > seem to be scaled by N in the current TMD implementation. Can
> > the developers confirm that?
>
> You're correct as far as I can tell; this will have to be fixed in the
> documentation. Thanks for pointing it out.
>
> >
> > 3) I have seen comments about problems with TMD in
> > parallel runs and that those problems have been fixed in the
> > 'CVS version'. Does that also apply to the source version
> > (tar.gz file) that is available through the standard NAMD
> > download mechanism or is the 'CVS version' a different version?
>
> You'll need to get the cvs version through the mechanism mentioned
> above. In any of the namd2.6 releases TMD will not work properly in
> parallel.
>
> >
> > 4) Finally, I have more general comments about the compilation
> > instructions (in notes.txt) and default Makefiles that ship
> > with the source distribution. They are not exactly up to
> > date and result in a number of minor problems on standard
> > modern UNIX distributions (in my case an EM64T system). In
> > particular, recent versions of gcc don't seem to work, current
> > versions of icc need different flags for generating static
> > executables that won't crash upon startup, the charm test
> > program is not where 'notes.txt' says it is etc. The problems
> > all seem to be known more or less and can be solved with
> > a little bit of help from Google but it would be nice if
> > NAMD could be compiled out of the box on Linux
> > workstations with recent compiler and OS versions.
>
> Please note that namd2.6 is at this point a bit over 2 years old, which
> is likely responsible for at least some of the difficulties you
> experienced. The binary releases are still fine, but if you're building
> from source I highly recommend using the development version. Please
> give the cvs version a shot (it also has updated release notes) and by
> all means let us know if you run into similar trouble so that things can
> be fixed before the 2.7 release.
>
> Best regards from an East Lansing native,
> Peter
>
> >
> > Thanks for any feedback.
> >
> > ---
> > Michael Feig
> > Biochemistry & Molecular Biology
> > Chemistry
> > Computer Science & Engineering
> > Michigan State University, East Lansing, MI
> >
> >
This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:52:16 CST