The CASINO forum

Hi,

I've recently become interested in studying molecules adsorbing to metal surfaces. I have read some of the recent work by Tkatchenko and others discussing the issues with DFT for describing this interaction. It seems to me that DMC would be an excellent method for studying these types of systems. However, it doesn't appear that this method is commonly applied to these types of systems, even for typically challenging cases such as carbon monoxide on the platinum (111) surface. I am wondering why not? I know that finite size errors can be a complication, but would this make DMC results for these types of systems unreliable? Or, is there something more nefarious going on that I am missing?

Thanks,
Mike

Hi Mike,

I am actually working on exactly such systems. There is also some work by Dario Alfè on the dissociation of H2 on Mg(0001). However, for transition metal surfaces one has to meet some severe challenges in DMC. These challenges are mainly related to the use of PPs. To beat DFT you have to be reasonably sure that you actually get an accuracy of about 1kcal/mol. Hence, not only the statistical error bars have to be small, but one also has to be very careful about systematic errors. If you are interested in more details, I am happy to share my current results with you. Feel free to contact me!

All the best,
Katharina

Hi Mike,

Basically, I write in support of what Katharina wrote. Finite size effects can be dealt with but systematic errors related to PPs seem to be difficult to limt to 1 kcal/mol unless your slab geometries to be compared show a considerable cancellation of this systematic error. It seems difficult to compare 'isolated molecule' + clean surface with adsorbed molecule, for metals, particularly in the first transition series (Pt is a bit better) so, in addition to PP issues, there could well be FN problems that are amplified when the trial wave-functions are set up in this way.

I have results for copper surfaces that cover 2008-2014 some of which have been published but there is no completely satisfactory benchmark.


Best

Philip