DFT Calculations converge ok in direct mode but not when integrals are stored incore
Just Got Here
1:27:30 PM PDT - Thu, Apr 16th 2015
Dear All,

I am benchmarking different options for running some DFT claculations and having access to large memory machines my intention is t run some fully incore calculations. For the example benchmark I am using (a C60) molecule, no matter the basis set, converges nicely in a few iterations in direct mode but when run incore fails miserably to reach convergence. I attach the two input files below. Is it possible that the starting guess is somehow different for the two? Any insight will be grately appreciated.

I am running branch 6.5 revision 26243

Cheers,

Cristián

INCORE

title "C60"
memory 2000 mw
  1. print low
echo

start C60

geometry "system" units angstroms nocenter noautoz noautosym noprint
load format xyz "C60.xyz"
end

set geometry "system"

basis "ao basis"
 * library 3-21G
end
basis "cd basis"
* library "DGauss A1 DFT Coulomb Fitting"
end


dft
 xc xpbe96 cpbe96

 semidirect memsize 1500000000 filesize 0
end

task dft energy


DIRECT

title "rubish"
memory 2000 mw
  1. print low
echo

start C60

geometry "system" units angstroms nocenter noautoz noautosym noprint
load format xyz "C60.xyz"
end

set geometry "system"

basis "ao basis"
* library 3-21G
end
basis "cd basis"
* library "DGauss A1 DFT Coulomb Fitting"
end


dft
 print low "convergence"

 xc xpbe96 cpbe96

 direct
end

task dft energy

Forum Vet
2:12:16 PM PDT - Thu, Apr 16th 2015
Cristián
Please remove the line
semidirect memsize 1500000000 filesize 0
from your first input file, since
1) it has no effect on the SCF when CD basis set is used
2) it's used only when the starting guess is generated and, as you can see,
it has the adverse effect of generating a poorer guess than the default one.

My suggestion is not to play with the semidirect options, since -- as you can see -- can have unpredictable effects on the calculation accuracy


Forum >> NWChem's corner >> Running NWChem