This thread is locked.Only browsing is available.
Top Page > Browsing
problem with cohesive energies
Date: 2008/12/04 23:27
Name: Vlad

Dear Prof. Ozaki and OpenMX users,

I have a problem calculating the cohesive energies with OpenMX-3.4 for atoms in molecules. The calculated values seem to be always strongly overestimated. To give the very simple example: the cohesive energy (or the bond dissociation energy if you want) in H2 molecule is calculated to be -6.2 eV using H5.5-s2 basis set with H_TM pseudopotential and 100 Ry energy cutoff in LDA. The experimental value is known to be -4.5 eV and is generally quite good reproduced with other DFT codes (+/-0.2 eV depending on basis type and size). But OpenMX gives a strong overestimation.

The runtest calculations showed that the OpenMX is correctly compiled and perfectly reproduces the test results. I have varied the size of the basis set and the energy cutoff, tried the BSSE correction, tried fixing the grid origin to keep the same atom position in molecule and single atom calculations relative to the grid, tried the orbital optimization: in all calculations the cohesive energy is overestimated by at least 15%.
Such the overestimation was similarly observed for the large variety of other smaller molecules and clusters. Could you probably tell me the reason for such a large error?

Thank you in advance!
Page: [1]

Re: problem with cohesive energies ( No.1 )
Date: 2008/12/09 06:16
Name: T.Ozaki


Using OpenMX3.4, I also calculated the cohesive energy of H2 molecule.
The following is my result.

VPS: H_CA.vps, scf.energycutoff=150Ryd, scf.XcType=LSDA-CA

********************* H6.5-s2 ****** H6.5-s3 ******* H6.5-s4 *******

NSP H (hartree)* -0.444608886408 -0.444608927193 -0.444608992979
SP H (hartree)** -0.477765305885 -0.477942135972 -0.477977695817
opt H2 (hartree)* -1.112057034790 -1.125275162130 -1.130369890431

Ecoh (hartree)** +0.156526423020 +0.169390890186 +0.174414498797
Ecoh (eV)******* +4.259240496797 +4.609295512851 +4.745992926765

The bond length of H2 molecule was optimized for each basis set.
Also, "NSP" and "SP" stand for the non spin polarized and spin polarized
calculations, respectively. Ecoh is defined to be 2*(SP H)-(opt H2).

My result looks better. Also, even if we use the shorter cutoff radius for
the basis set, I guess that the BSSE can be reduced by the
counterpoise scheme.


Re: problem with cohesive energies ( No.2 )
Date: 2008/12/05 16:48
Name: Vlad

Dear Prof. Ozaki,

thank you for your reply.

I could reproduce your results. And I have found the source of previous discrepancies: in the spin polarized calculations for single atoms I put the same initial charges for alpha and beta spin (e.g. 0.5 0.5 for hydrogen atom), thinking that the spin charges will be refined during SCF cycles. It's not the case. So with LSDA-CA and "Scf.SpinPolarization On" I got simply the same energies as for the non spin polarized calculations.

Re: problem with cohesive energies ( No.3 )
Date: 2008/12/11 21:58
Name: Vasilii Artyukhov

Dear Vlad,

since you're putting equal initial occupations for both spins, and there's no mechanism to break this symmetry during your calculation (like, e.g., spin-orbit coupling could do), there's simply no reason to expect the result to be different from that from a non-spin-polarized calculation. :)

Page: [1]