Re: Setting target minimization gradient?

From: Luis Rosales (
Date: Fri Apr 22 2005 - 19:49:18 CDT



 I posted that question... And I got a couple of nice answers, however I dont
know if those replys were posted also to the list...
I will paste them in this mail, credit due to the original author of those
posts ;)


I've been wondering about those questions too. I've been assuming the word
gradient refers to the rate of change of energy with respect to atomic
coordinates. There doesn't seem to be much interest in the topic on namd-l,
probably because minimization is usually computationally inexpensive
compared to other NAMD computations.

If you don't get an answer, you may want to take a look at the GROMACS
manual ( and other resources. I suppose you could take a
look at the source code as well.

Good luck,
Blake Charlebois

Hi Luis. There are at least three ways to accomplish this. You could write a
script that does convergence tests on the energy (which, as NAMD uses a shitty
local optimizer, won't tell you much) and sends the SIG_INT to NAMD2 upon
satisfaction of your criteria. Or you could hack the NAMD source and put such
a thing in (be careful, it was written in CHARM++ by a team of Hamsters), of
course. I did such a thing and put in my own global minimizer to compliment
the local CG minimizer. I think puting 'partial solutions' into the local min
from the global one is a good idea.

Hope this wasn't completely obvious.. Good luck on your studies.


Hope this helps, or at least makes you smile as I did when I was reading the
line about the team of hamsters...

Good luck,

Luis Rosales

---------- Original Message -----------
From: "Chang, Christopher" <>
To: <>
Sent: Fri, 22 Apr 2005 12:15:01 -0600
Subject: namd-l: Setting target minimization gradient?

> Hi,
> I found the message below posted on the namd-l archives in
> September of last year, unfortunately without a response. I have the
> same question--is there a means to set a target gradient for
> minimization in the NAMD config file, or is reaching a target
> gradient (as reported in many publications) a trial-and-error affair?
> Thanks,
> Chris
> _____________________
> Hi!!
> I am starting to play with namd, and from a small search on the list
> archives, I have read that one of the parameters to look during an
> energy minimization is the "Gradient Tolerance" as this is a good
> measure of the system convergence... However, I still have a couple
> of very basic questions: What does the gradient tolerance measure?
> Is the Gradient tolerance related to changes on energy, RMS, max
> derivative? I wonder if is possible to implement this on a
> configuration file, to mimimize the system for X steps or until
> gradient tolerance is <=Y ??? Thanks for your help, Luis Rosales
------- End of Original Message -------

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:40:41 CST