Appropriate Density Fitting Basis Sets for RPAX2

[Connor Fawcett]

Hello all,

I'm currently trying to determine the appropriate density fitting basis set for use with RPA programs such as RPAX2 or ACFDT2. I've noticed an example of a "typical" RPAX2 correlation energy calculation uses mp2fit basis set. 

Is the choice of density fitting basis set solely based on the type of calculation being performed (mp2fit for RPA v.s. jkfit for HF as an example)? My instinct is to use the corresponding df mp2fit basis set, though I'd like to make sure that this is correct. I didn't see any indication of what the appropriate df basis set to use is in either the Density fitting or Basis input sections for this particular application.

For reference, I am planning on studying adsorption of various molecules on transition metals. I intend to use the aug-ccPVDZ basis set (with Stuttgart ECPs for the metals). Rather than using a slab model, I will be using DFET to optimize an embedding potential which can recreate environmental effects on a "cluster" of metal atoms.

I noted that there are also RPA algorithms available without density fitting, though I worry that it will be difficult to converge and highly memory intensive even for the smaller systems I'd like to study. If these computations are reasonable, I'll plan on comparing the results from either implementation.

Please let me know what you think!

Best,

Connor Fawcett

[andreas...@gmail.com]

Dear Connor,

typically the density fitting basis set needs to be chosen such that it can optimally fit the respective orbital products that are being used with the method. In case of RPAX2 and related methods that is orbital products of the type phi_occ*phi_vir (occupied*virtual orbital products), so the ideal fitting basis sets will be the corresponding MP2Fit basis sets that were developed for the respective orbital basis sets. This is not to say that MP2Fit is the most optimal choice here generally, yet special fitting basis sets for DF-RPA to my knowledge haven't been developed so far, so it is the best choice you have in Molpro.

In case of HF and DFT methods the fitting basis sets also need to cover the case of orbital products with two occupieds (phi1_occ*phi2_occ) and therefore it is recommended to use the fitting basis stes which were designed for this, namely JKFit ones.

In either case, if you want to make sure that the DF error in the calculation is minimised, you can also choose either one of the available fitting basis sets, but use one where the cardinal number is increased compared to the orbital basis set (use triple-zeta or even quadruple-zeta instead of double-zeta, for instance). Both with JKfit and MP2Fit basis sets you should converge to the same result if only the fitting basis set is chosen large enough to describe all possible orbital products accurately.

Best wishes,

Andreas

[Connor Fawcett]

Dear Andreas,

Thank you so much for your quick response!

This has certainly cleared up my confusion. I didn't know that the JKFit basis sets cover occupied*virtual orbital products in addition to occupied*occupied. To ensure my study is rigorous I will plan on trying both types of fitting basis sets alongside  at least the double and triple zeta orbital basis sets.

Thanks for the help!

Best,

Connor