AMBER Archive (2005)

Subject: Re: AMBER: sander segmentation fault

From: Thomas E. Cheatham, III (cheatham_at_chpc.utah.edu)
Date: Wed Feb 02 2005 - 17:59:13 CST

> Out of curiosity along the lines of this thread--I seem to remember (memory
> backed up by some old notes...) that in an older version of the DNA tutorial (2
> years ago maybe--pre-amber8 release, but post amber7) it suggested you set ncyc
> larger than maxcyc for minimizations involving rigid water models like TIP3P and
> SPC/E, because conjugate gradient minimization on these rigid models often
> resluted in LINEMIN failures. Looking through the current version of the DNA
> tutorial today, it suggests more or less half steepest decent and half conjugate
> gradient, with no mention of a problem with rigid water. Has something changed
> in amber8 or did I misunderstand?

It's a matter of taste... Nothing really has changed with the minimizers
in sander for quite a while and no change has been made that will overcome
these issues (with SHAKE failures, linmin failures with conj. grad.,
etc). With (rigid) water present, it is useful to turn off SHAKE (to
avoid SHAKE failures). Also, as in the earlier DNA tutorial, I personally
always minimize with steepest descent (i.e. NCYC > MAXCYC); this is
"slower" in terms of its convergence but it is very stable. As the
tutorial evolved this may have changed, but the basic take home message is
that minimization (in my opinion and in the context of sander) is really
only important to locate a local minimum energy basin that avoids hot
spots (due to overlap or poor geometries) in the initial models.

In general, I only run between 500-1000 steps of minimization (steepest
descent) before starting MD. More is overkill (usually); if the system is
not ready for MD after 500-1000 steps of minimization, this usually points
to an underlying problem with my model (such as severely overlapping
atoms, ring fusion, etc) which may or may not be remedied with more
minimization steps.

-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu