MOLPRO at NERSC

MOLPRO is a complete system of ab initio programs for molecular electronic structure calculations, written and maintained by H.-J. Werner and P. J. Knowles, and containing contributions from a number of other authors. As distinct from other commonly used quantum chemistry packages, the emphasis is on highly accurate computations, with extensive treatment of the electron correlation problem through the multiconfiguration-reference CI, coupled cluster and associated methods. Using recently developed integral-direct local electron correlation methods, which significantly reduce the increase of the computational cost with molecular size, accurate ab initio calculations can be performed for much larger molecules than with most other programs.

The heart of the program consists of the multiconfiguration SCF, multireference CI, and coupled-cluster routines, and these are accompanied by a full set of supporting features. For a full description of the program's capabilities, see the MOLPRO quantum chemistry package web site.

• Access MOLPRO at NERSC
• Running MOLPRO at NERSC
• Running MOLPRO on Franklin
• Running MOLPRO on Carver
• Restart Capabilities
• Documentation and Help

Access MOLPRO at NERSC

MOLPRO is available at NERSC on Franklin (versions 2008.1 and 2009.1; the latter is the default) and Carver (version 2009.1 only).

You must use the module utility to access the MOLPRO application. To load the default version,

% module load molpro 

% module load molpro/2008.1

% module avail molpro

To see where the MOLPRO executables reside (the bin directory) and what environment variables it defines,

% module show molpro

% module show molpro           (or type module show molpro/2008.1)
-------------------------------------------------------------------
/usr/common/usg/Modules/modulefiles/molpro/2009.1:

prepend-path	 PATH /usr/common/usg/molpro/2009.1/bin 
prepend-path	 LD_LIBRARY_PATH /usr/common/usg/mkl/default/lib/em64t
setenv		 TMPDIR /global/scratch/sd/JOE_USER 
-------------------------------------------------------------------

Running MOLPRO at NERSC

You must use the batch system to run MOLPRO on the compute nodes. You can do this interactively or you can use a script. Examples of both are below.

Note that on all machines the file molpros is a script to run a serial (one-processor) version of the code and molprop is a script to run a parallel version.

Running on Franklin

To run a parallel job interactively on up to 512 cores, use the qsub -I -q interactive -V -l mppwidth=# command to request an interactive batch session. Here is an example, requesting 8 cores for the default time limit (30 minutes):

% qsub  -I -V -q interactive -l mppwidth=8 

%  cd $PBS_O_WORKDIR 
%  module load molpro 
%  molprop -n 8 your_molpro_inputfile_name 

Note that when the time limit (30 minutes is maximum) is reached the job will end and the session will exit.

To run a batch job on Franklin, use a job script similar to this one:

#PBS -S /usr/bin/tcsh
#PBS -N test_molpro
#PBS -q debug
#PBS -l mppwidth=8,walltime=30:00
#PBS -j oe
#PBS -m e
#PBS -V

cd $PBS_O_WORKDIR

module load molpro
molprop -n 8 h2o_opt_dflmp2.test

Put those commands or similar ones in a file, say, test1.pbs and then use the qsub command to submit the job:

% qsub test1.pbs 

#PBS -N test_molpro
#PBS -q debug
#PBS -l mppwidth=8,mppnppn=2,walltime=30:00
#PBS -j oe
#PBS -m e
#PBS -V

cd $PBS_O_WORKDIR

module load molpro
molprop -n 8 -N 2 pentane_dflccsd.test 

Note:

• We have seen cases where a job that runs across multiple nodes, when 4 cores per node are used, may hang randomly. Cray provided a fix to this on 4/17/2009 that is included in the molprop script. However, if you still run into this problem, please use a reduced number of cores per node, e.g., 2, or use only 1 node.
• If your job runs into a memory error, please try launching the MOLPRO binary directly through aprun, instead of a script. e.g.,

  module load molpro
  aprun -n 8 molprop_2008_1_Linux_XT_i8.exe pentane_dflccsd.test --no-xml-output
  

For general instructions on running jobs on franklin, see Running Jobs on Franklin.

Running on Carver

To run a job interactively on up to 64 cores, use the qsub -I -q interactive -V -l mppwidth=# command to request an interactive batch session. Here is an example, requesting 16 cores for the default time limit (30 minutes):

% qsub  -I -V -q interactive -l nodes=2:ppn=8 

%  cd $PBS_O_WORKDIR 
%  module load molpro 
%  molpros your_molpro_inputfile_name 

%  cd $PBS_O_WORKDIR 
%  module load molpro 
%  molprop -n 8 your_molpro_inputfile_name 

Note that when the time limit (30 minutes is maximum) is reached the job will end and the session will exit.

To run a parallel job on Carver use a job script similar to the example below, and submit it to the batch queue using qsub command.

#PBS -S /usr/bin/tcsh
#PBS -N test_mol
#PBS -l nodes=2:ppn=8
#PBS -q debug 
#PBS -l walltime=00:30:00
#PBS -j oe
#PBS -V

cd $PBS_O_WORKDIR
module load molpro
molprop -n 16 pentane_dflccsd.test

% qsub test1.pbs 

On Carver if you want to use more memory per process (and few cores per node) then use a script like this one:

#PBS -S /usr/bin/tcsh
#PBS -N test_mol
#PBS -l nodes=8:ppn=2
#PBS -q debug
#PBS -l walltime=00:30:00
#PBS -j oe
#PBS -V

cd $PBS_O_WORKDIR
module load molpro
molprop -n 16 pentane_dflccsd.test

This script will request that the MOLPRO job be run on eight nodes but using only two cores on each node. NOTE THAT YOUR REPO WILL BE CHARGED FOR THE USE OF ALL 64 CORES.

There are many other parameters to the PBS batch system. For more details see Running Jobs on Carver.

Restart Capabilities

By default, the job is run so that all MOLPRO files are generated in $TMPDIR. This is fine if the calculation finishes in one job, but does not provide for restarts. This section describes techniques which can be used to restart calculations.

MOLPRO has three main files which contain information which can be used for a restart: file 1 is the main file, holding basis set, geometry, and the one and two electron integrals; file 2 is the dump file and used to store the wavefunction information, i.e. orbitals, CI coefficients, and density matrices; file 3 is an auxiliary file which can be used in addition to file 2 for restart purposes. File 1 is usually too large to be saved in permanent storage

By putting the following lines in the input file, the wavefunction file (file number 2) can be saved as file "h2.wfu",and the auxiliary file (file number 3) saved as "h2.aux". By default, the files are saved to the subdirectory "wfu" of your home directory if the job runs out of time.

***,H2
file,2,h2.wfu,new
file,3,h2.aux,new
basis=vdz;
geometry={angstrom;h1;h2,h1,.74}
optscf;

The directory where the files are saved may be changed using the "-W" command line option.

These files enable some restarts to be performed, as they provide snapshots of the calculation as each module finishes. Unfortunately, restarting an incomplete SCF or CI calculation is not possible. To use the files in a restart, remove the "new" qualifier from the "file" command:

***,H2
file,2,h2.wfu
file,3,h2.aux
basis=vdz;
geometry={angstrom;h1;h2,h1,.74}
optscf;

Documentation and Help

MOLPRO User's manual.

On Carver you can find the full manual and a quickstart version in MOLPRO's doc subdirectory.