using old KS orbitals/densities from previous calculations


Clicked A Few Times
Hello. I'm using nwchem to calculate binding energies of small molecules on an transition metal clusters. As some of these clusters are very large (particularly in comparison to the adsorbates) it would be useful if I could use the KS-orbitals/densities from the previously calculated bare cluster when starting the calculation of the cluster+adsorbate system. This should speed things up a bit, particularly on the first SCF cycle since whatever the electronic structure of the bare cluster was will be a better initial guess than the usual atomic density. I would have hoped that if the atoms were all in the same order in the input file, it would have sufficed to do the following:

start
echo
memory 4600 mb
geometry bridge1 noautoz noautosym
Pd 2.398765852 0 1.388678926
Pd 0 0 2.776431822
Pd 0.744233156 -2.228737442 1.472444413
Pd 0 0 0
H 0.760264707 0.749454007 1.44434155
H 1.38625867 -1.0868673 0.378189346
H -0.569927871 -0.01838549 -1.060303572
H -0.566033065 0.721597247 -1.057081848
end
basis set info etc....

dft
mult 1
xc PBE0
grid fine
vectors input bare_Pd4.movecs output nwchem.movecs
convergence energy 1.000000E-07
convergence density 1.000000E-05
convergence gradient 1E-05
iterations 1600
mulliken
end
task dft optimize


and that this would use the previously calculated density with just the usual atomic density guess for the
hydrogen atoms.

However, the output file indicates that the old .movecs were not used:
Either an incorrect movecs file was specified, or linear dependence has changed, or the basis name was changed.
Loading old vectors from job with title :
Load of old vectors failed. Forcing atomic density guess


It seems like there MUST be a way of doing this. Also, if there is a way, can someone please explain if anything special needs to be done if the total spin state of the system changes (say if it's O2 instead of H2 and total system goes from a triplet to a quintet).

Thanks for any help you can give.

Clicked A Few Times
As different atoms in geom mean different basic functions and different number of basis functions if you use basis sets from NWChem default libraries. That's the reason why you can't use the bare cluster as an initial guess for H2 one. There must be some way to keep the basis functions unchanged to do it.
But when you use a system with triplet O2 as an initial guess for a system with quintet O2 , I think it's OK.

Clicked A Few Times
Quote:Neil Nov 27th 1:33 pm
As different atoms in geom mean different basic functions and different number of basis functions if you use basis sets from NWChem default libraries. That's the reason why you can't use the bare cluster as an initial guess for H2 one. There must be some way to keep the basis functions unchanged to do it.
But when you use a system with triplet O2 as an initial guess for a system with quintet O2 , I think it's OK.



I'm not sure that I understand. The TM atoms are in the same order in both cases (say, the first 55 atoms) and the last two are the hydrogens. I don't see why there would be any basis set confusion. Also, you misunderstand what I was asking regarding the spin states. What I'd LIKE to be able to do is to calculate the bare cluster in, say, a triplet spin state and then add an O2 and see what happens if it retains two alpha electrons and hence the multiplicity goes up to five. Using the KS-orbitals/density (not just the geometry) from the cluster+adsorbate system converged on the triplet surface as an initial guess on the quintet surface leads to all sorts of problems in SCF convergence.

Forum Vet
Nope, that's not how it works. You can do what you want though, take a look at http://nwchemgit.github.io/index.php/Hartree_Fock_Theory_for_Molecules#Superposition_of_frag... .

Effectively, you also need to get the MOs for the adsorbate fragment separately, and then you can use "vectors fragment bare.movecs adsorbate.movecs output cluster.movecs".

Bert


Quote:Adampelzer Nov 25th 6:28 pm
Hello. I'm using nwchem to calculate binding energies of small molecules on an transition metal clusters. As some of these clusters are very large (particularly in comparison to the adsorbates) it would be useful if I could use the KS-orbitals/densities from the previously calculated bare cluster when starting the calculation of the cluster+adsorbate system. This should speed things up a bit, particularly on the first SCF cycle since whatever the electronic structure of the bare cluster was will be a better initial guess than the usual atomic density. I would have hoped that if the atoms were all in the same order in the input file, it would have sufficed to do the following:

start
echo
memory 4600 mb
geometry bridge1 noautoz noautosym
Pd 2.398765852 0 1.388678926
Pd 0 0 2.776431822
Pd 0.744233156 -2.228737442 1.472444413
Pd 0 0 0
H 0.760264707 0.749454007 1.44434155
H 1.38625867 -1.0868673 0.378189346
H -0.569927871 -0.01838549 -1.060303572
H -0.566033065 0.721597247 -1.057081848
end
basis set info etc....

dft
mult 1
xc PBE0
grid fine
vectors input bare_Pd4.movecs output nwchem.movecs
convergence energy 1.000000E-07
convergence density 1.000000E-05
convergence gradient 1E-05
iterations 1600
mulliken
end
task dft optimize


and that this would use the previously calculated density with just the usual atomic density guess for the
hydrogen atoms.

However, the output file indicates that the old .movecs were not used:
Either an incorrect movecs file was specified, or linear dependence has changed, or the basis name was changed.
Loading old vectors from job with title :
Load of old vectors failed. Forcing atomic density guess


It seems like there MUST be a way of doing this. Also, if there is a way, can someone please explain if anything special needs to be done if the total spin state of the system changes (say if it's O2 instead of H2 and total system goes from a triplet to a quintet).

Thanks for any help you can give.


Forum >> NWChem's corner >> Running NWChem