[molpro-user] different DFT energies from Molpro and other softwares (Molcas, Gaussian09) for open-shell systems

[gosia olejniczak]

Dear Molpro users.

I'm comparing DFT energies between Molpro and two other softwares: Molcas
and Gaussian09.

I use:
1. two Hamiltonians: nonrelativistic and DKH2
2. HF method, DFT method with LDA (VWN5) and B3LYP (with VWN3) functionals
3. two types of basis sets: ANO contracted basis sets and uncontracted
basis - Dunning for lighter and Dyall for heavier elements
4. restricted calculations for closed-shell and unrestricted for open-shell
test systems

In order to compare the correct things, i take care of the following:
1. i check basis sets are really the same (i use spherical harmonics)
2. i set the same thresholds for integral screening
3. i set the same thresholds for SCF convergence
4. if DKH2 Hamiltonian is used, all calculations are done with point
nucleus model
5. if DFT calculations are done, i set the grid of the analogous quality
6. all calculations for open-shell systems are done with unrestricted HF/DFT
7. all calculations are run without symmetry

As a result, i can reach a very good agreement between scf energies from
all software for closed-shell systems. For example for Ne atom with ANO
basis set i get:

                    RHF (nonrel)       SVWN5 (nonrel)       B3LYP (nonrel)

 -----------------------------------------------------------------------------------------------------------
      g09        -128.523833639      -128.208032842        -128.958538025
     molpro     -128.52383364       -128.20803284          -128.95853804
     molcas    -128.5238336388    -128.2080328424       -128.9585380272

                    RHF (DKH2)         SVWN5 (DKH2)         B3LYP (DKH2)

 -----------------------------------------------------------------------------------------------------------
      g09         -128.683953423      -128.367112213        -129.118503285
     molpro      -128.68395342       -128.36711221          -129.11850330
     molcas     -128.6839534241    -128.3671122136       -129.1185032879



When i do calculations for open-shell systems i have very good agreement
for HF method, but when i do DFT, Molpro always gives noticeably higher
energy than Molcas/Gaussian09.

It is not a question of converging to different states. I looked at O atom
and O2 molecule.

For example for O2 molecule with ANO contracted basis set, HF energies
agree very well up to 1d-09, but for DFT agrrement is only at 1d-6 (i take
care of tight convergence criteria and dense grids, so that it should not
be due to numerical noise). these are the results:


                    UHF (nonrel)         UVWN5 (nonrel)      UB3LYP (nonrel)

 -------------------------------------------------------------------------------------------------------
       g09        -149.676204517     -149.321224439       -150.383816961
     molpro      -149.676204516     -149.3212235135     -150.383815130439
     molcas     -149.676204517     -149.3212244405      -150.3838170099

                    UHF (DKH2)        UVWN5 (DKH2)        UB3LYP (DKH2)

 -------------------------------------------------------------------------------------------------------
       g09         -149.793906497     -149.437758947          -150.501389228
     molpro       -149.793906497     -149.437758012997    -150.501387385591
     molcas      -149.793906499     -149.4377589499       -150.5013892789


Conclusions for uncontracted basis set are analogous, so that it is not due
to contraction scheme.
For molecules with heavier open-shell elements, the discrepancy between
Molpro DFT and DFT from other software is even bigger (but there there is
additional question of converging to the same ground states).


Do you know what can be the source of these differencies? The
implementation of functionals should be the same (at least as far as i can
trace it on software's websites), but maybe there is some specificity
related to unrestricted DFT implementation?

Below i paste my Molpro input for O2, i can send you corresponding inputs
for Gaussian09 and Molcas if helpful,

best regards,
gosia olejniczak


--------------------------------------------------------------------------
 memory, 1000,m;
 file,1, O2.int, new;
 file,2, O2.wfu, new;

gthresh,energy=1.d-9,twoint=1.d-18,grid=1.d-9
gprint,basis;
gprint,orbitals;

symmetry nosym

include, O.tzp

geometry={
O  0.0  0.0  0.0;
O  0.0  0.0  1.2;
}

{uhf,accu=9.0;
WF,16,1,2}
{pop; individual;}
{pop; density,type=spin}

{UKS, LDA, accu=9.0;
WF,16,1,2}
{pop; individual;}
{pop; density,type=spin}

{UKS, B3LYP3, accu=9.0;
WF,16,1,2}
{pop; individual;}
{pop; density,type=spin}

INT,DKROLL=1

{uhf, accu=9.0;
WF,16,1,2}
{pop; individual;}
{pop; density,type=spin}

{UKS, LDA, accu=9.0;
WF,16,1,2}
{pop; individual;}
{pop; density,type=spin}

{UKS, B3LYP3, accu=9.0;
WF,16,1,2}
{pop; individual;}
{pop; density,type=spin}
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.molpro.net/pipermail/molpro-user/attachments/20140521/d8fda32d/attachment-0001.html>