incosistent DFT energies


Just Got Here
Dear NWChem users and developers,

I'm running some DFT calculations for a small cobalt oxide nanoparticle using CAM-B3LYP/6-31G, which ran fine in the Flux HPC cluster at the University of Michigan, using typically 48 cores. The NWChem which is installed there is version 6.1.1.

Recently I got access to the new Brazilian HPC facility named SDumont, with an much larger allocation (up to 1200 cores per job), but NWChem is producing erratic energies (the NWChem version is 6.6). I have repeated the same calculation using different number of nodes from 1 node (24 cores) to 25 nodes (600 cores) and SCF energies were the same during the first few steps of each calculation, then at some point the began to diverge from each other.

Final SCF energy differences are too large:

bench-01.out: Total DFT energy = -25618.475583943055
bench-02.out: Total DFT energy = -25618.497788784134
bench-03.out: Total DFT energy = -25618.516200558504
bench-04.out: Total DFT energy = -25618.497025334986
bench-05.out: Total DFT energy = -25618.495954391034
bench-06.out: Total DFT energy = -25618.515982887682
bench-07.out: Total DFT energy = -25618.408124515601
bench-08.out: Total DFT energy = -25618.472081296870
bench-09.out: Total DFT energy = -25618.515687469651

(the numbers in the output name stand for the number of nodes I have used)

I have repeated some of these calculations and energies were different from any previous values.

here's the input I'm running:



start bench-01

memory total 2500 stack 750 heap 750 global 1000 mb

charge 2

geometry
O  0.000000 2.123168 3.002613
Co 1.428727 2.020525 4.286181
O 1.283568 3.938407 4.286181
Co -0.000000 4.041050 2.857454
O -1.283568 3.938407 4.286181
Co -1.428727 2.020525 4.286181
O -0.000000 2.123168 5.569748
Co 0.000000 4.041050 -0.000000
O 0.000000 2.123168 -0.145159
Co 1.428727 2.020525 -1.428727
O 0.000001 2.123168 -2.712295
Co -1.428727 2.020525 -1.428727
O -1.573885 0.102642 -1.428728
Co -2.857454 -0.000000 -0.000000
O -2.857454 -1.917882 0.145159
Co -1.428727 -2.020525 1.428727
O -0.000001 -2.123168 2.712295
Co 1.428727 -2.020525 1.428728
O 1.573886 -0.102643 1.428728
Co 2.857454 0.000000 2.857455
O 4.141022 -0.102643 1.428728
Co 2.857454 -0.000000 0.000001
O 2.857454 1.917882 -0.145158
Co 2.857453 3.030787 1.428728
O 1.283569 4.143692 1.428726
O 2.857452 1.917884 3.002614
O 2.857454 -1.917882 0.145160
O 1.573886 0.102642 -1.428727
Co 0.000001 -1.010262 -1.428725
O 0.000002 -2.123167 0.145159
Co -2.857454 0.000000 2.857454
O -1.573886 -0.102643 1.428727
Co -0.000000 1.010262 1.428726
O -4.141022 -0.102643 1.428726
O -2.857454 1.917882 -0.145160
Co -2.857453 3.030787 1.428726
O -2.857454 1.917882 3.002611
O -1.283568 4.143691 1.428727
O -1.283568 3.938407 -1.428727
O 2.857453 -1.917882 2.712296
O -2.857455 -1.917882 2.712295
O 1.283568 3.938407 -1.428727
O 1.573884 0.102642 4.286181
Co -0.000001 -1.010262 4.286180
O -1.573885 0.102643 4.286180
H 4.641131 0.704307 1.463704
H 2.554397 -2.246781 -0.692982
H 2.519420 -2.296244 3.515462
H -2.554397 -2.246781 3.550437
H -2.519420 -2.296244 -0.658007
H -4.641131 0.704307 1.393751
H 0.034976 1.316218 6.069856
H 2.086734 4.316769 3.948148
H -2.121710 4.267306 3.983123
H 2.121710 4.267305 -1.125670
H -2.086734 4.316769 -1.090694
H -0.034975 1.316218 -3.212404
end

basis "ao basis" spherical
* library 6-31G
end

basis "cd basis" spherical
* library "Ahlrichs Coulomb Fitting"
end

dft
mult 16
iterations 1000
convergence energy 1e-6
xc xcamb88 1.00 lyp 0.81 vwn_5 0.19 hfexch 1.00
cam 0.33 cam_alpha 0.19 cam_beta 0.46
grid fine
tolerances accqrad 20
end

task dft energy




Any ideas where to begin?

Thanks!

Andre

Forum Regular
Do the 2nd Co and 22nd O actually have y and z coordinates or was there a mistake in copying your input file to the post?

Just Got Here
Quote:Sean Aug 10th 5:25 am
Do the 2nd Co and 22nd O actually have y and z coordinates or was there a mistake in copying your input file to the post?


Hi Sean,

it is a copy-and-paste error, they have coordinates (calculation runs fine in the older HPC allocation).

Andre


Forum >> NWChem's corner >> Running NWChem