AMBER Archive (2006)

Subject: RE: AMBER: MPI parallel problem in AMBER 8.0

From: Ross Walker (ross_at_rosswalker.co.uk)
Date: Tue May 02 2006 - 17:40:37 CDT

Dear MK,

> ELAN_EXCEPTION @ --: 6 (Initialisation error)
> elan_init: No capability, can't continue

This looks to me like you may still be using the MPI executable. How did you
run this job. Can you use a 'serial' compiled sander executable. That is one
that was not compiler with MPI support and run it interactively. I.e.
$AMBERHOME/exe/sander -O -i ... on the command line and not through any
form of batch scheduler.

> The bottom is the input I used for the test.
> Do you see any problem?

Did you get any output files at all? What do they show.

A couple of notes, some of which may or may not be problems:

> # Control section
> &cntrl

You should have only '1' space between the left margin and the &cntrl.

> ntwe = 200, ntwx = 200, ntwv = 200, ntpr = 200,

Just a note, do you really need velocity information (ntwv=200). If you
don't need it or don't know why you might need it you are probably best off
not writing it (ntwv=0).

> ntt = 1, temp0 = 300.0, tempi = 5.0, tautp = 1.0,

> scnb = 2.0, scee = 1.2, dielc = 1,

I wouldn't bother specifying these in your mdin file as they are the
defaults. Having less clutter in the input file can make debugging problems
easier. (Less chance for typos etc).

> ntb = 0, ntc = 2, ntf = 2,

You have ntb=0 which turns off periodic boundaries making this a gas phase
simulation (no igb) and yet you don't change cut from the default value of
8.0. This value of a cut off is much too small for a gas phase simulation
and in fact Amber will quit with an error message if you don't set it larger
than 8.0. This may be the cause of your error. In your mdout file it
probably says something about cut being unreasonably small for a gas phase
simulation.

My advice is that if you really really really really want to run a gas phase
simulation (i.e. not an implicit solvent simulation) then a more realistic
value of cut is 20.0 or greater. Or even better use NO CUT OFF at all (i.e.
set cut=9999.0 or some other very large value).

> nstlim = 250000, dt = 0.0010,

a 1fs time step is probably overkill when you are using shake. You can
probably get away with 1.5fs or 2.0fs if your system is fairly stable.

> ntt = 1, ntp = 0,

ntt=3 is generally the prefered option these days.

> ntx = 1, nmropt = 1
> &end

Some machines (compilers) don't like &end. Try using a '/' instead. This
should work with all compilers. I.e.

  ntx = 1, nmropt = 1
 /

Also make sure you have only '1' space before the /.

> &wt
> type = 'TEMP0', istep1 = 1, istep2 = 50000, value1 = 5.0,
> value2 =
> 300.0,
> &end

Avoid having a blank line before the &wt. This may or may not cause
problems. Also avoid having a trailing comma on the last entry in a
namelist. This can sometimes cause problems (although rarely). I.e.

 /
 &wt
  type = 'TEMP0', istep1 = 1, istep2 = 50000, value1 = 5.0, value2 = 300.0
 /
 &wt
 .....

> &wt
> type = 'TEMP0', istep1 = 30001, istep2 = 200000, value1 = 300.0,
> value2 = 300.0,
> &end

Your istep 1 here is smaller than istep2 above. This is illegal. Note, since
this line just maintains temp0 at 300K it is kind of redundant.

>
> &wt
> type='END'
> &end
>
> LISTOUT=POUT
> DISANG = restraints_2.in

Avoid the blank line before LISTOUT and it is probably best to not have
spaces around the '=' sign. Also make sure that the restraints_2.in file can
be seen and read by all nodes in the parallel calculation. I.e. make sure
the working directory is the same and that restraints_2.in is on a shared
drive.

I hope this helps,

All the best
Ross

/\
\/
|\oss Walker

| HPC Consultant and Staff Scientist |
| San Diego Supercomputer Center |
| Tel: +1 858 822 0854 | EMail:- ross_at_rosswalker.co.uk |
| http://www.rosswalker.co.uk | PGP Key available on request |

Note: Electronic Mail is not secure, has no guarantee of delivery, may not
be read every day, and should not be used for urgent or sensitive issues.

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