On the "energy" and "shift" keywords for the derivation of interatomic potentials

Hi everyone,

New user here! I have spent the last 2-3 days looking at the manual trying to assess whether GULP has all I need for my purposes, and it looks like it’s the case! Kudos for building and maintaining such a great software!

I have doubts about the use of “energy” as an observable, especially in conjunction with the “shift” keyword. I don’t really understand these two comments in example22 " […] get rid of the large core contribution by removing the number in front of the decimal place […] " and " […] uniform shift to the energies of all structures to remove the constant offset […] ".

I’ll make my case, so that maybe it will be easier for you to help me understand.
I’m planning to use CP2K for the ab initio data. Say that I get -32.7128736 eV and -31.8736453 as the total energy of my system for configuration 1 and 2, respectively.

  1. Which value would I use for the energy as an observable in the first and second structure, if I specify a “shift” value after the first structure?
  2. It is not clear to me what is the offset we are talking about, could you give me a clarification on that?

Thanks a lot,
Paolo Sebastiano Floris

Hi Paolo,

The point here is that the total energies you get from QM calculations are large numbers because they include core electrons (or pseudopotentials) etc & so it makes no sense to fit the energy that comes from such calculations with a force field. Instead what you need to fit are relative energies (i.e. energy differences between states). You can do this either by directly computing things that make sense in a force field context (e.g. binding energies) or just apply a shift that turns the absolute values into relative ones (which is what the shift does in GULP). You can also use the “reaction” option to make all energies into differences between configurations (this was added later on, whereas shift was available in the code from the start).
Hope that helps.


1 Like

It helps a lot, thank you very much!