AMBER Archive (2004)

Subject: RE: AMBER: MD problem (very long parhabs endless runtime)

From: Ross Walker (ross_at_rosswalker.co.uk)
Date: Sat Oct 23 2004 - 13:53:22 CDT

Hi Daniel,

> No I didn't get any output.

The fact that you get no output at all suggests that something is wrong with
your copy of sander rather than with your input file. Try running something
very simple like the gas phase MD provided with the DNA tutorial:
http://amber.scripps.edu/tutorial/polyA-polyT_New/index.html

> My system is much smaller. It owns only about 10000 atoms including
> about 2000 Water molecules.

This should run in seconds so something is obviously wrong.
 
> the test cases
> >all pass then this is much more unlikely.
> Uh. I didn't check that. Where do I find these test cases ?

See $AMBERHOME/test/

Ideally you should follow the installation instructions on page 7 of the
users manual. Essentially once you have built amber and the executables have
been copied to the $AMBERHOME/exe directory you should then do:

cd $AMBERHOME/test/
make test

This will run a whole series of tests. They should all pass. Each test
should take no more than a few seconds to run. Any failures should be noted
and the diff files checked. If the differences are simply rounding errors
(last decimal place) then they can be safely ignored. Anything bigger should
be more carefully investigated.
 
> I wanted to set it to one just to check wheteher MD works
> onmy system.
> If it runs I want to
> perform a run of much more steps.

Ok, that's fine. I suggest, however, that you look into doing a more complex
equilibration of your system rather than just the MD scheme you have
proposed. See the Amber DNA tutorial for an example of how to minimise a
system and then slowly heat it and then equilibrate it. I.e. simply starting
a system off at 300K with constant pressure can lead to instabilities. You
would probably be better to start at 0K and ramp the temperature up to 300K
with constant volume over about 20ps before switching to constant pressure.
This is secondary to your problems of not obtaining any output however.
 
> >> ntwv = 1,
> >Do you 'really' want to write out the velocities every step?
> If at all?
> Only for my test run later I would set this value to 100.

My question is though, do you 'really' want the velocities as a function of
time. Typically people never use the velocities, they just want the
positions and so ntwv is usually set to 0. Unless of course you have a
specific reason for wanting the velocities.

>> stable simulation. I would avoid belly for anything other than
>> minimisation...!!! And even then you are probably better
>> using restraints instead.

> Oh, that's a good hint. But why is belly so bad ?

The problem is that in doing a constant pressure simulation you use
isotropic position scaling of the atoms to correct the pressure towards your
target pressure. Now, with the belly option any atoms subject to belly
constraints will not be allowed to move. As a result you will have a section
of your system not subject to the isotropic scaling and so essentially you
are trying to do constant pressure with only a small part of your system.
Hence the instabilities. If, however, you do restraints using a harmonic
potential (ntr=1) then the restrained atoms will be allowed to move, and so
can be adjusted by the constant pressure algorithm, but will feel a force
that restores them to the reference structure. This is much less severe than
the belly option and so should, while providing you with essentially the
same effect as you desire from belly, give a much more stable simulation
that has much better pressure regulation.

Let me know how you get on with the test cases.

All the best
Ross

/\
\/
|\oss Walker

| Department of Molecular Biology TPC15 |
| The Scripps Research Institute |
| Tel:- +1 858 784 8889 | EMail:- ross_at_rosswalker.co.uk |
| http://www.rosswalker.co.uk/ | PGP Key available on request |

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