Re: [AMBER] Regarding the MPI error in parallel running of 3D-RISM

From: George M. Giambasu <>
Date: Fri, 29 Jun 2018 10:02:53 -0400

The post you mentioned says that the error you get appears because you
might be using a different version of MPI to run your program than that
you used to compile Amber. This happens all the time when using
computing clusters. Here are few things to try:

  - make sure that the modules that were loaded at the compile time are
also loaded at the run time. You can list the available modules using
"module avail". In your batch script explicitly load those modules, eg:
module load intel/17.0.4 mvapich2/2.2.

  - make sure that the environmental variables such as PATH and
LD_LIBRARY_PATH are maintained in the batch script. First thing to do is
to source $AMBERHOME/ file in the batch script. Then add the
commands "echo $PATH" and "echo $LD_LIBRARY_PATH" in the batch script to
actually see what the compute nodes actually use and compare the results
to what you have on the login node where you compiled the code.

Let us know what you get from doing this. When you get back to us please
mention what AmberTools version you are using, compilers version and MPI
distribution and version.

(I do have doubts that rism3d will need a special MPI setup than what
one usually gets when using typical supercomputing facilities).



Dr. George M. Giambasu, Ph.D.
Rutgers, The State University of New Jersey
174 Frelinghuysen Road, Room 308E Piscataway,
NJ 08854, USA
Office: 1-848-445-5223
Cell: 1-612-644-8285
AMBER mailing list
Received on Fri Jun 29 2018 - 07:30:02 PDT
Custom Search