Dear NWChem developers/users,
I am currently doing SO-DFT calculations to get an approximate evaluation of the effect of spin-orbit interaction on the energy of metal atoms and molecules. I am using the SO-ECPs from the Stuttgart group with the correlation consistent cc-pVnZ basis sets and the B3LYP functional.
A major problem that I encountered regards the expectation value of S^2. For example, in the case of the Bismuth atom (spin multiplicity 4 in the ground state), I obtain a value of about 3.36, far from the exact value of 3.75. The same result is obtained regardless details in the input file such as using a preliminary SCF and/or non-SO DFT calculation or performing, as in the example given in the NWChem manual, the only SO-DFT calculation (note that simple DFT calculations give correct S^2 values in all cases). I attach two slightly different input files for Bi at the end of this message. Is anything wrong in my inputs?
Note also that in treating other species it happens that, depending on the input details and also on the NWChem version used (I have used the 5.1.1., 6.0, 6.1, 6.3 releases), sometimes the S^2 values of the SO-DFT calculation improve (typically when using, as detailed above, preliminary SCF and DFT calculations), getting close to the correct values. However, better values of S^2 always correspond to higher total energies and viceversa. All this seems to me to make questionable the results.
I would be very grateful if someone could suggest to me how to improve the SO-DFT calculations in order to obtain the correct values of S^2 (and energies) especially for the case of the Bismuth atom.
Thank you very much
Guido
INPUT 1
start b3lyp Bi
title "b3lyo Bi"
memory 4000 mb
set int:txs:limxmem 50000000
geometry units angstrom
Bi 0 0 0
end
BASIS "ao basis" PRINT spherical
[Here Bi cc-pVTZ basis set]
END
ECP
[Here scalar ECP]
END
SO
[Here SO ECP]
END
ecce_print ecce.out
dft
xc b3lyp
iterations 200
mult 4
convergence energy 1.000E-08
convergence density 1.000E-05
convergence gradient 1.000E-05
iterations 200
grid fine
end
task sodft
INPUT 2
start b3lyp Bi
title "b3lyo Bi"
memory 4000 mb
set int:txs:limxmem 50000000
geometry units angstrom
Bi 0 0 0
end
[Here basis set and ECPs as before]
ecce_print ecce.out
set tolguess 1e-8
scf
quartet
tol2e 1e-10
adapt off
sym off
end
dft
xc b3lyp
iterations 200
odft
mult 4
adapt off
sym off
grid xfine
end
task scf energy
task dft energy
task sodft Energy
|