TURBOMOLE Users Forum
TURBOMOLE Modules => Jobex: Structure Optimization and Molecular Dynamics => Topic started by: marand on July 07, 2014, 02:08:10 pm

Dear Users and Developers
I have encountered a strange problem while trying to optimize geometry of a 27atom organic molecule at the CC2/augccpVTZ level of theory. The symetry group is Cs.
Optimization crashes during the first step, whe after the initial dscf calculations the ricc2 module is envoked. The calculations are initialise and fail at this poit (the excerpt from job.last follows):
total memory allocated for calculation of (QP)**(1/2) : 33 MiB
calculation of (PQ) ...
time in lpzwei cpu: 0.17 sec wall: 0.17 sec ratio: 1.0
calculation of the Cholesky decomposition of (PQ)**(1) ...
time in invmetri cpu: 23.74 sec wall: 1.12 sec ratio: 21.3
threshold for RMS(d[D]) in SCF was : 0.10E06
integral neglect threshold : 0.28E11
derivative integral neglect threshold : 0.10E07
setting up bound for integral derivative estimation
increment for numerical differentiation : 0.00050000
=========================================================================
Energy of reference wave function is 1009.4912800250000
Maximum orbital residual is 0.4573730786200E04
Number of symmetrynonredundant auxiliary basis functions: 2303
Block lengths for integral files:
frozen occupied (BOI): 1 MiB
active occupied (BJI): 1 MiB
active virtual (BAI): 17 MiB
frozen virtual (BGI): 0 MiB
general (BTI): 18 MiB
=========================================================================
error in gradient step
I receive no further comment or error message.
I am using redundant internal coordinates and maximum of 3500 MB of core memory. The calculations run without correlating the core electrons.
I am guessing that there must be some default limit exceeded, because the same computations uwith smaller basis sets (augccpVDZ, ccpVTZ) run smoothly. Please, point me to the possible reason for the crash of my calculations.
My best regards
Marcin Andrzejak

Hi,
the 'error in gradient step' is a bit confusing since it does not stem from ricc2 itself, but from jobex which noticed an incomplete ricc2 run.
So the question is why ricc2 failed to run. In almost all cases the problem is caused by memory problems or limits:
 Is your stack size limit sufficient?
 Did you run the parallel version and asked for too much memory for each process?
 How large is the maximum virtual memory your system can handle? This can be a limit if the size of swap is too low.
Some hints are given here:
http://www.turboforum.com/index.php/topic,23.0.html (http://www.turboforum.com/index.php/topic,23.0.html)
If all that does not help, contact the support!
Regards,
Uwe

If ricc2 crashed without printing at the end an error message it could be a memory (stack size) limit that
is exceeded or it was an MPI parallel calculation and the error occured in one of the slave processes which then aborted all the other processes.
Check if there are any error messages from the operating system of the Fortran runtime environment in the error output of your calculation (or batch job).
For MPI parallel calculations check if any of the slave*.output files contains at the end an error message.
Christof

Hi,
If I'm using MPI and see that slave4.output ends in "**** dscf : all done ****" but the others (1 to 3) seems to be interrupted, i.e they end like that:
calculating CC ground state density
a semicanonical algorithm will be used
density nr. cpu/min wall/min L R

1 2.97 2.97 L0 R0

time in cc_1den cpu: 2 min 58 s wall: 2 min 58 s ratio: 1.0
What can I learn? where is the problem? why is my excited states optimization fails?
Thanks!

dscf and ricc2 use a slightly different numbering for the output files. ricc2 write the output for the first slave to the terminal.
So this is normal.
Also the output you get in slave*ouput for 13 from ricc2 is normal. It doesn't contain any error message.
Is there an error message from the first slave (terminal output)? Any error messages from the system or MPI?
Christof