RE: amber-developers: State of NEB in AMBER10?

From: Ross Walker <ross.rosswalker.co.uk>
Date: Tue, 24 Apr 2007 09:36:02 -0700

Hi Wei,

> In amber 10, you need to use sander.MPI to do NEB simulation.

This is good but I have a number of concerns...

> There is a test case showing how to do this in
> amber10/test/full_pimd/Run.full_pimd.

These test cases appear nowhere in the standard test Makefile so I assume
haven't been run for ages and none of the other developers would be aware if
they are trampling on this code with their updates.

The Centroid test case fails with a namelist error:

Testing Centroid MD
forrtl: severe (19): invalid reference to variable in NAMELIST input, unit
5, file /usr/local/amber10/mpich2/test/full_pimd/cmd.in, line 21, position
11

All other test cases run but fail. The errors are reasonably innocuous in
that it appears that the saved files all just need updating. The save files
all have a zero temperature and in addition all new runs print the time to 5
decimal places instead of 3. E.g:

< NSTEP = 1 TIME(PS) = 0.001 TEMP(K) = 0. PRESS =
0. 
---
>  NSTEP =        1   TIME(PS) =     0.00100  TEMP(K) =   179.57  PRESS =
0.
13c13
<  NSTEP =        2   TIME(PS) =       0.002  TEMP(K) =     0.  PRESS =
0.
---
>  NSTEP =        2   TIME(PS) =     0.00200  TEMP(K) =   341.05  PRESS =
0.
So the problems with testing here concern me - have you validated that the
current tree indeed gets the correct results? You should really update the
test cases and have it run as part of the standard testing ASAP otherwise
there could be serious problems with other people updating the CVS tree.
So you might want to fix the printing of the time and update the test cases
with the correct temperatures. Meanwhile I'll run alanine dipeptide with the
new NEB and make sure it gets the correct pathway.
Also are there any qmmm NEB test cases?
On another note the following from numprocs.awk:
BEGIN{i=0}
{while ( i < NF ){if( $i == "-np" || $i == "-n"){i++;print $i; exit}; i++;}}
is not transferable. Not all mpi implementations have -n or -np followed by
the number of processors. For example on AIX the environment variables
MP_NODES and MP_TASKS_PER_NODE are used to define the number of processors
to run and then "poe" with no extra command line options is used to launch
the mpi job.
Hence we need to rethink how the test cases detect the number of processors.
While it isn't so much of a problem with Amber 10 while under development
once it is released I can just hear all the complaints on the amber mailing
list... :-(. So either we write a single centralized (probably in
$AMBERHOME/test/) numprocs.awk script that works for all mpi implementations
we can find or we find a more graceful way for the test case scripts to
handle sander quiting due to incompatible processor counts. Comments?
All the best
Ross
/\
\/
|\oss Walker
| HPC Consultant and Staff Scientist |
| San Diego Supercomputer Center |
| Tel: +1 858 822 0854 | EMail:- ross.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.
Received on Wed Apr 25 2007 - 06:07:34 PDT
Custom Search