Frequency calculation for C70 fullerene crashing, do not understand error


Clicked A Few Times
I am doing DFT calculations with C70, and although my optimizations and single point calculations are going fine, my attempt to run a frequency calculation on the optimized geometry keeps crashing. I have been running it on large memory nodes. My first attempt, in which I ran on it on 8 nodes with only one process per node and 40 GB per process, made it this far and then crashed with this error:

HESSIAN: the one electron contributions are done in     101.1s

0:armci_malloc:malloc 2 failed: -1323215288
(rank:0 hostname:n162 pid:27425):ARMCI DASSERT fail. ../../ga-5-1/armci/src/memory/memory.c:PARMCI_Malloc():895 cond:0


When I doubled the number of nodes, it made it a little further and crashed again with the same error:

HESSIAN: the two electron contributions are done in    7412.0s

0:armci_malloc:malloc 2 failed: -1323215288
(rank:0 hostname:n181 pid:8082):ARMCI DASSERT fail. ../../ga-5-1/armci/src/memory/memory.c:PARMCI_Malloc():895 cond:0

The .e* file contains the error:

Last System Error Message from Task 7:: Numerical result out of range

I cut-and-pasted the input file below. I would be grateful for any suggestions on what could be causing this crash.

title "C70"
echo
memory total 40000 stack 10000 heap 10000 global 20000 mb
scratch_dir ./scratch
geometry autosym units angstroms noautoz
C 0 1.23334161 3.96232805
C 1.17297757 0.38112352 3.96232805
C 0.72494001 -0.99779432 3.96232805
C -0.72494001 -0.99779432 3.96232805
C -1.17297757 0.38112352 3.96232805
C 2.29219242 0.74477847 3.2162988
C 3.00148533 -0.24686632 2.43857472
C 2.57335703 -1.56450976 2.43857472
C 1.41665283 -1.94985534 3.2162988
C 0.69272615 -2.93086807 2.43857472
C -0.69272615 -2.93086807 2.43857472
C -1.41665283 -1.94985534 3.2162988
C -2.57335703 -1.56450976 2.43857472
C -3.00148533 -0.24686632 2.43857472
C -2.29219242 0.74477847 3.2162988
C -2.28314826 1.96394786 2.43857472
C -1.1622938 2.77829629 2.43857472
C 0 2.41015374 3.2162988
C 1.1622938 2.77829629 2.43857472
C 2.28314826 1.96394786 2.43857472
C -0.71596444 3.38763064 1.20439728
C 0.71596444 3.38763064 1.20439728
C 3.44307337 0.3659128 1.20439728
C 3.00058302 1.72775808 1.20439728
C -3.00058302 1.72775808 1.20439728
C -3.44307337 0.3659128 1.20439728
C -2.57042673 -2.31981742 1.20439728
C -1.41197193 -3.1614841 1.20439728
C 1.41197193 -3.1614841 1.20439728
C 2.57042673 -2.31981742 1.20439728
C 3.08266695 -1.77363472 0
C 3.53644628 -0.37704554 0
C 2.63942333 2.38370722 0
C 1.45141362 3.2468468 0
C -1.45141362 3.2468468 0
C -2.63942333 2.38370722 0
C -3.08266695 -1.77363472 0
C -3.53644628 -0.37704554 0
C 0.73423038 -3.47987376 0
C -0.73423038 -3.47987376 0
C 2.57042673 -2.31981742 -1.20439728
C 1.41197193 -3.1614841 -1.20439728
C 3.00058302 1.72775808 -1.20439728
C 3.44307337 0.3659128 -1.20439728
C 0.71596444 3.38763064 -1.20439728
C -0.71596444 3.38763064 -1.20439728
C -1.41197193 -3.1614841 -1.20439728
C -2.57042673 -2.31981742 -1.20439728
C -3.44307337 0.3659128 -1.20439728
C -3.00058302 1.72775808 -1.20439728
C -2.28314826 1.96394786 -2.43857472
C -1.1622938 2.77829629 -2.43857472
C 1.1622938 2.77829629 -2.43857472
C 2.28314826 1.96394786 -2.43857472
C 3.00148533 -0.24686632 -2.43857472
C 2.57335703 -1.56450976 -2.43857472
C 0.69272615 -2.93086807 -2.43857472
C -0.69272615 -2.93086807 -2.43857472
C -3.00148533 -0.24686632 -2.43857472
C -2.57335703 -1.56450976 -2.43857472
C 1.41665283 -1.94985534 -3.2162988
C -1.41665283 -1.94985534 -3.2162988
C -2.29219242 0.74477847 -3.2162988
C 0 2.41015374 -3.2162988
C 2.29219242 0.74477847 -3.2162988
C 0 1.23334161 -3.96232805
C 1.17297757 0.38112352 -3.96232805
C 0.72494001 -0.99779432 -3.96232805
C -1.17297757 0.38112352 -3.96232805
C -0.72494001 -0.99779432 -3.96232805
end
basis
carbon library 6-311G*
end
scf
 semidirect memsize 100000000 filesize 0
end
dft
xc b3lyp
print "information"
iterations 200
vectors input C70_optimization.movecs output C70.movecs
end
task dft freq

Clicked A Few Times
I had a same problem for C60. But I used "freq numerical" and worked fine for me.
If you have symmetry (I think at least can be D5h for your case) you have just 5 unique atoms. (I think)
I think you can get very accurate result (nearly same as analytical) if you have accurate integration.
I got accurate and correct results for C60 just by setting this criteria

 CONVERGENCE energy 1e-8 \
density 1e-6 \ ## maybe not necessary
gradient 1e-5 ##maybe not necessary
GRID fine
TOLERANCES tight

Also I optimized with the tight criteria.

Results were in excellent agreement with experiment.
and on 80 processor took less than 5 hours. (not too much)

I hope it help.
--
ººººººººººººººººººººººººººººººººººººººººººººººººººººººººº
S. Shahab Naghavi
(SISSA)
Scuola Internazionale Superiore di Studi Avanzati
Via Bonomea 265, Tireste, Italy
Phone: +39-403787448
Email(1): snaghavi@sissa.it
ººººººººººººººººººººººººººººººººººººººººººººººººººººººººº

Clicked A Few Times
Thank you so much, "freq numerical" worked!

Forum Vet
I was able to run this calculation with 16 processors using the following memory settings:

Input: memory 3500 mb

Job script: setenv ARMCI_DEFAULT_SHMMAX 2048

I think the huge memory allocations are creating the problems.

Bert


Forum >> NWChem's corner >> General Topics