Systematic errors due to pseudopotentials

[Philip_Hoggan]

Having used CASINO for while to evaluate reaction barriers for diatomics on metal surfaces, now is the crunch! Our benchmark for H2 dissociation on Cu(111) agrees well with the best mesurements.

Should it?

We use a Troullier Martins PP that shows a systemtic error around 1kcal/mol in PBE DFT compared to more complete PPs (with 17 instead of 11 active electrons through including the 3p shell). We are worried about this since the error we'd like to claim is 0.5kcal/mol, based on the DMC with twist-averging and extrapolation to cater for finite size effects (these are also large in the DFT calculation).

Now, the PBE DFT barrier is way off the measured one (too low). Could the error due to the PP also be very different in the DMC result and how could it be estimated? A new QMC run with a better PP is likely to take ages and this PP has produced good results for DMC reaction brriers, although the measured values are not within a kcal/mol in the cases we previously studied.

Any help on this would be most welcome.

[Dario_Alfe]

Dear Philip,

In my view there are a few issues here:

- you say there are large finite size effects in DFT too, but it is not clear if you are referring to k-point sampling of genuine H2-H2 interactions on the surface. Of course, the latter have nothing
to do with QMC finite size effects, and so if you use extrapolation to get rid of those it is not clear what functional form you should use. In general, however, I find it difficult to extrapolate to
infinite size reliably, even when the finite size are just QMC effects. If your extrapolated values are very different from those obtained without extrapolation then I would be wary.

- The PP with 17 electrons is not necessarily better that that with 11 electrons, have you tested them against a reliable all-electron calculation ?

- Yes, the PP error could be different in QMC than in DFT. One way to test it would be to do a high quality (e.g. CCSD(T) ) calculation on a small cluster, perhaps carved from your real system, and
compare with QMC.

Dario

[Philip_Hoggan]

Dear Dario,

Thanks for the reply!

I think the finite-size issue is under control.

You have experience of the TM PP in QMC for metal surfaces. Have you gathered evidence to support the conjecture that the PP error in QMC is (much) less than in DFT (PBE) for the Troullier-Martins scheme of PPs?

Best,

Philip

[Dario_Alfe]

Dear Phil,

No, I don't have any evidence of the PP error being necessarily smaller in QMC.

Best wishes,

Dario