RE: Minimization convergence issue

From: Brittany Morgan (
Date: Wed Mar 04 2009 - 13:36:38 CST

Hi Chris,


Your suggestion of using a dcdfreq of 1 worked; I found the problem but I am
not sure why it is happening. It seems to be a problem with amide protons.
In the initial structure, they are fine, but after minimization, almost all
of the amide protons are rotated so that they are inside of the carbonyl
carbon! The minimization is actually causing the problem. It doesn't matter
if I use rigidBonds on or off during minimization. Once I start the
dynamics, some of them begin to oscillate wildly, causing the Rattle


I think that something must be going wrong when I generate the psf (using
psfgen), but nothing appears wrong visually and I get no errors or warnings
of any kind. I use 'pdbalias atom XXX H HN' for each residue (where XXX is
the 3 letter residue code), but it reads in all of the coordinates correctly
without an issue.


Any idea what might be causing this?






From: [] On Behalf Of Chris
Sent: Wednesday, March 04, 2009 12:20 PM
To: Brittany Morgan
Subject: Re: namd-l: Minimization convergence issue



What is the timestep for the beginning of the MD (not minimization) and the
first Rattle failure? If first Rattle failure occurs several steps after
the beginning of the MD, try dropping your dcdfreq to 1, collecting all the
frames and then visualizing them (specifically looking at atoms near and
interacting with those atoms suffering from the Rattle failure) to see if
you notice anything obviously problematic.


Chris Harrison, Ph.D.
Theoretical and Computational Biophysics Group
NIH Resource for Macromolecular Modeling and Bioinformatics
Beckman Institute for Advanced Science and Technology
University of Illinois, 405 N. Mathews Ave., Urbana, IL 61801                            Voice: 217-244-1733               Fax: 217-244-6078
On Mon, Mar 2, 2009 at 10:53 AM, Brittany Morgan <>
Hi Peter,
There are no additional constraints or restraints on the system. I haven't
tried generating the psf using my old procedure, because it is very tedious.
However, comparing the original PDB structure with my new structure in VMD
shows them to be identical (same positions, no extra bonds or atoms). The
only difference from the original PDB (NMR structure with all hydrogen atoms
present) is that mine is solvated and ionized.
Could it be an issue with atom names? I did a lot of aliasing with psfgen
and in the end I had no errors, warnings or spurious bonds. But, it might be
possible to have swapped the names of two atoms without generating extra
bonds (say, two hydrogens connected to the same carbon), but if it doesn't
generate extra bonds, I fail to see how it would be an issue in the
- Show quoted text -
-----Original Message-----
From: Peter Freddolino []
Sent: Monday, March 02, 2009 11:30 AM
To: Brittany Morgan
Subject: Re: namd-l: Minimization convergence issue
Hi Brittany,
are you placing any additional constraints or restraints on the system?
If you visually inspect your structure (maybe comparing the same
structure generated using your new and old procedures), does everything
look ok? Did you get any warnings from psfgen? It's odd that you'd have
difficulties with structures made by psfgen if the initial coordinates
are ok...
Brittany Morgan wrote:
> Hello all,
> I am having a problem with convergence of the minimization routine. The
> gradient tolerance convergences rapidly at first, but then fluctuates
> around 40-100 and will not continue to decrease. If I try to go ahead
> and run dynamics, even with a very conservative time step (0.5fs with a
> 1e-6 SHAKE tolerance), I get a RATTLE constraint error, but it is not
> the same atom every time.
> I have tried SHAKE on or off during minimization, varying the length of
> minimization from 1,000 to 20,000 steps and I also tried using the
> velocity quenching algorithm for 1,000 steps initially and then doing
> additional conjugate gradient minimization. Turning SHAKE off during
> minimization helped some (fluctuates between 10-40) for one structure at
> least, but it didn't solve the problem.
> I have never had this problem in the past, and now it is happening for 4
> structures. The difference is that I am using psfgen (using a script,
> not the GUI) to generate the input files for NAMD instead of using
> CHARMM. I can't find any difference when comparing the output of psfgen
> with the original PDB files, and the structure after minimization is not
> very different from the starting state. I'd prefer not to go back to
> using CHARMM, since it is a pain comparatively for me, but I don't know
> what else to try.
> Any suggestions?
> Thanks,
> Brittany Morgan

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:52:26 CST