Hi,
just some general remarks about the comparison of total energies or properties from different quantum chemistry programs:
1. Basis set
it is obvious that you have to use the very same basis set, that's what you were asking for.
2. DFT functional
many people try to compare DFT results, so make sure that you are really using the very same functional. With Gaussian, you could either use VWN 5 instead of the Gaussian default VWN 3 for all functionals that include VWN (like B3-LYP), or use the VWN 3 functional within Turbomole instead of the default VWN 5 (only possible for B3-LYP).
3. DFT grid size
Gaussian and Turbomole are using completely different numerical integration grids for the DFT quadrature. So comparing just the number of grid points is not sufficient.
4. RI
Do not switch on RI in Turbomole if you want to compare the total energy to Gaussian. And I am not sure if Gaussian's version of RI (they call it density fitting) gives the very same results as Turbomole if you are using the Turbomole auxiliary basis set for both programs. That would be nice to know, so if you are going to try it, please tell us
.
5. SCF convergence criteria
That is a bit more tricky, because the number of SCF iterations and the total energy both depend on the thresholds for neglecting integrals. In Turbomole, you can change $scfconv for the energy convergence and $denconv for the density convergence. The integral neglect threshold is determined by those two keywords (more or less), but the actual threshold when evaluating the integrals does also depend on several other things like the sum of the l-quantum numbers of the integral...
I have no idea how you can change the threshold in Gaussian, or how to compare it at all, but both programs will print some number in the output...
6. Geometry convergence criteria
Turbomole has two optimizer: relax and statpt. The default, when calling jobex, is to use relax. relax just checks for the difference in energy and the gradient norm. If you call jobex with -statpt, then the (newer) optimizer statpt will be used. statpt checks for the energy change, RMS of displacement, RMS of gradient, the maximum displacement and the maximum gradient. The number of geometry cycles will, of course, depend on all those thresholds. Please make sure that all thresholds that are used in both program packages are the same.
7. Start geometry
Gaussian automatically moves a molecule in its center of mass - Turbomole does not (unless you do not have C1 symmetry, or unless you are using internal coordinates during an optimization - then, the information about the absolute position is lost after the first geometry cycle due to the transformation from internal to cartesian coordinates). That has no effect on energy or gradients. But if you do something like NMR chemical shieldings, the GIAO integrals are defined by a vector that points from the origin to the atoms. If your input structure is far away from the origin, the numerical errors of the GIAO integrals might get quite large, and the different contributions to the shielding tensor do not cancel each other completely. Move your structure to CMS before starting such a calculation (within define, in the first menu, say 'm' for 'manipulate geometry', and then 'inert').
Hence, I would just compare the Hartree-Fock single point energy on the same input structure, using the same basis set, without RI/density fitting...
Regards,
Uwe