AMBER Archive (2002)

Subject: Re: minimization in amber7 for nonperiodic systems

From: Giulio Rastelli (rastelli.giulio_at_unimo.it)
Date: Wed Aug 14 2002 - 10:03:11 CDT

Dear David and others,
thanks for pointing out a possible solution.
In the tests I run so far, the use of an infinite cutoff (1) solves the
LINMIN failure.
Also by using a short cutoff and by setting nbflag=0 and skinnb=0 (2),
which restores the use of the non-bonded
list update, I didn't have any failure in minimization;
I also tested the difference between (1) and (2) with molecular dynamics
at 300K:
(1) was pretty stable, I got structures in agreement with experiment and
low
energy fluctuations because the non-bonded list is never updated.
(2) I was able to get reasonable results only with cutoff>15angstroms
and a
more frequent non-bonded list update (nsnb=10 instead of 25, default).
Obviously, this behaviour is dependent on the system I am investigating,
but
it is just worth to give an idea.
Regards,

Giulio Rastelli

   
David Case wrote:
> nbflag=0, skinnb=0
> Several people have noted recently that some types of minimization in
> Amber 7 perform very poorly. Here are things that should help; I
> am primarily thinking of non-PME (ntb=0) systems:
>
> (1) If you can at all afford it, use a very large cutoff (e.g. 999.).
> Conjugate gradient minimization is very sensitve to cutoffs, and in Amber
> (all versions) it is often a poor economy to try to skimp on the cutoff
> during minimization. Also, turn off SHAKE (set ntf=ntc=1); you *can* leave
> SHAKE on, but only for short minimizations whose purpose is to remove bad
> contacts before MD can begin.
>
> (2) If you really cannot use an infinite cutoff, use as large a one as you
> can (especially when igb=0), and set nbflag=0, skinnb=0. in the &ewald
> namelist. [Notes: yes, this should have been in the Users' Manual; no,
> the fact that these variables are in the &ewald namelist does *not* mean
> that they are only important for PME simulations, although it is certainly
> understandable that a user would think that.]
>
> Setting nbflag=0 makes nsnb (the frequency of nonbonded updates) an
> important variable. I haven't done lots of tests, but the default value
> of 25 seems reasonable good: you get 25 steps of conjugate gradient before
> re-making the non-bonded list, which triggers a "restart" of the conjugate
> gradient procedure. But especially with non-periodic systems with water,
> there is a real limit to what one can do with minimization using the
> current Amber cutoff scheme. But nbflag=0, skinnb=0. *should* get you back
> the performance one used to get in earlier versions. Future releases will
> make these the default for minimization.
>
> (3) Mike and Tom: the above is another reason why we need to implement
> a switching scheme more general than erf/erfc (and/or implement non-periodic
> PME....)
>
> Thanks to those of you that have brought the problem to our attention;
> please let us know if items (1) or (2) above don't help.
>
> ...regards....dac
> --
>
> ==================================================================
> David A. Case | e-mail: case_at_scripps.edu
> Dept. of Molecular Biology, TPC15 | fax: +1-858-784-8896
> The Scripps Research Institute | phone: +1-858-784-9768
> 10550 N. Torrey Pines Rd. | home page:
> La Jolla CA 92037 USA | http://www.scripps.edu/case
> ================================================================== 

-- 

Giulio Rastelli
Dipartimento di Scienze Farmaceutiche
Universita di Modena e Reggio Emilia
Via Campi 183
41100 Modena - ITALY
-------------------------------------
tel 0039-059-2055145
fax 0039-059-2055131
-------------------------------------