How to relay %nproc=x to Gaussian09?

[Michal Kochman]

Dear Newton-X Developers and Users,

I am learning to use Newton-X (version NX-1.4.0-2), and I seem to have stumbled onto a problem when running nonadiabatic dynamics at the TD-DFT level with Gaussian09 (Revision A.02). I am running a slightly modified version of the test simulation MD-SICH4-G09-TDDFT-NAD:

==> control.dyn <==
              &input
              Nat   = 6
              nstat = 3 
              nstatdyn = 2
              dt    = 0.50
              t     = 0.0
              tmax  = 25.0
              prog  = 6.5
              thres = 100
              /&end

==> JOB_AD/basis <==
3-21g

==> JOB_AD/gaussian.com <==
%chk=gaussian
%nproc=4
%rwf=gaussian
%mem=4GB
#td(Nstates=3,Root=2) 3-21g BHandHLYP nosymm

comment

0   1
     Si        -0.010544         0.010835          0.808752
      C         0.015819        -0.050721         -0.931069
      H         0.009875         1.149573          1.543340
      H         0.172225        -1.176299          1.509994
      H         0.071356         1.042598         -1.452426
      H        -0.149187        -0.712162         -1.664254




==> veloc <==
    0.000010936   -0.000066200   -0.000002385
    0.000017729   -0.000268794    0.000167248
   -0.000729330    0.000915395    0.000595681
    0.000465234    0.001110822   -0.000214325
    0.000691764    0.002843392   -0.001622418
   -0.000935033    0.000121951   -0.000690283

==> geom <==
 Si   14.0   -0.01992571    0.02047582    1.52831928   27.97692840
 C     6.0    0.02989350   -0.09584871   -1.75946482   12.00000000
 H     1.0    0.01866037    2.17237820    2.91648925    1.00782504
 H     1.0    0.32545742   -2.22288223    2.85347524    1.00782504
 H     1.0    0.13484275    1.97022545   -2.74468749    1.00782504
 H     1.0   -0.28192308   -1.34579170   -3.14498462    1.00782504


The job submission script for my cluster is as follows:

==> job.bash <==
#!/bin/bash
#$ -cwd -V
#$ -N nx_test
#$ -q ecdf
#$ -l h_rt=48:00:00
#$ -pe memory-2G 4
#$ -R y
#$ -P chemistry_pccp_morrison

jobname=job
nproc=4

export NX=/exports/work/chemistry_pccp_morrison/michal/newton_x/NX-1.4.0-2/bin

export g09root=/exports/work/chemistry_rcf/software
export GAUSS_SCRDIR=$TMPDIR
source $g09root/g09/bsd/g09.profile

. /etc/profile.d/modules.sh
date
$NX/moldyn.pl > logfile
date


It contains an nproc=4 line as required by the cluster to run Gaussian09 in parallel.

The only modifications to the original test simulation MD-SICH4-G09-TDDFT-NAD are the value of tmax in control.dyn, the amount of memory in JOB_AD/gaussian.com (%mem=4GB) and, crucially, the addition of the %nproc=4 flag in JOB_AD/gaussian.com . When the simulation is running, the %nproc=4 flag does not appear in TEMP/gaussian.com , and Gaussian09 is not being executed in parallel mode. The key line "Will use up to    4 processors via shared memory." does not appear in TEMP/gaussian.log.

Regarding the nproc flag, the newton-x manual states that "To run a calculation in parallel, the number of processors has to be specified with the keyword %nproc in the GAUSSIAN input. It has also to be specified in the submission script." As far as I can tell, I am following these instructions to the letter. I would therefore like to ask you, is the reason why %nproc=x does not appear in the Gaussian09 input file, and how should I correct it?

(By the way, I am using a compatibility fix for Gaussian09 Revision A.02, which was kindly issued to my colleague Arend Dijkstra by Rachel Crespo and comes in the form of a modified run_cioverlap_g09.pl, but the nproc issue happens both with the old and the new versions of run_cioverlap_g09.pl, so I believe that this compatibility fix is not the source of the problem.)

Thank you very much for any help and advice.

Best regards,

Michal Kochman
Postdoctoral researcher
CFEL, DESY, Hamburg

[Michal Kochman]

Dear All,

I just wanted to report that I've found a walkaround, maybe this will be useful to someone:

Instead of setting the number of processors using %nproc, it is possible to create a "Default.Route" file which will specify the default number of processors for a job. More information on how to do this can be found at the Gaussian website:
http://www.gaussian.com/g_tech/g_ur/m_defroute.htm

[Vernon Crowell]

Hello,
Would it be possible for you to send me the version of run_cioverlap_g09.pl which fixes compatibility issues
with Gaussian09 Revision A.02, or at least let me know what's been changed from the original?
Thanks
Vernon

[Michal Kochman]

[Michal Kochman]

I just attached the modified run_cioverlap_g09.pl file to the previous post. I hope Rachel Crespo and other developers approve of me posting it here! It's meant to simply replace the old file (it lives in the 'bin' directory). It's not necessary to recompile Newton-X.

Best wishes,

Michal

[Rachel Crespo]

Dear Michal, 

I haven't checked the discussion list for such a long time, the changes in run_cioverlap_g09.pl are available in the last 

distributed version. 

Did you solve the problems with "%nproc"? 

Best wishes 

Rachel