COSMO treatment yields large differences in relative energies

[vsharma]

Dear everyone,

I have some molecular systems that I optimized (at BP86/def2-SVP) and then performed a single-point calculation (at B3-LYP/def2-TZVP) on the optimized geometries. When I compared the relative energies between different systems, I got reasonable numbers, lets say 10 - 100 kcal/mol. Now when I treat the same systems with COSMO (at B3-LYP/def2-TZVP and eps=4, 20, 50....etc) and then compare the 'same' relative energies, the results are a bit too large, such as the relative energy value which was 20 kcal/mol (at B3-LYP/def2-TZVP) becomes almost 2400 kcal/mol (at B3-LYP/def2-TZVP and eps=4). I have used Turbomole before, but always before relative energies varied only a little when COSMO treatment was performed.

I can see in output (in COSMO run) that during each SCF iteration energy remains indeed close to what it was in a 'non' COSMO treatment, but a great drop in total energy occurs after ENERGY CONVERGED statement in the output file, but that could be because of the switch of the grid size to a more finer one.

Could you please suggest possible reasons for such large changes in relative energies upon COSMO treatment ? Is it possible that the two above points are linked ? I think I may 'not' have made a mistake somewhere.

many thanks,

best regards,

Vivek

[uwe]

Hello,

could you please send a bug report to turbomole at cosmologic.de and include your control and output files.

There was a problem in version 6.4 with MPI parallel COSMO calculations on some systems and it would be very important to check if you ran into that.

Regards,

Uwe

[vsharma]

Hello again,

Yes, all my calculations were performed on Turbomole 6.4 and I can confirm by running Turbomole 6.31 that this problem occurred with Turbomole 6.4/COSMO/MPI implementation (in complete agreement to your reply). I got rather okay results (relative energies) with Turbomole 6.31. As you suggested, I will send the relevant files to turbomole at cosmologic.de.

Many thanks for your reply.

best regards,

Vivek