Exact SOAP cutoff treatment about dscribe

[lauri-codes]

Hi Marcel,

Nice to hear that you are not seeing any major differences in performance!

The difference between quippy SOAP and dscribe SOAP has been discussed in the old project repository. Please check that discussion, although it does not provide a direct answer. Fully answering this question would require someone who knows both the quippy code and the dscribe code in and out.

Regarding the "padding" that is added in extending the system for periodic systems: If you use the polynomial radial basis set (rbf="polynomial"), anything beyond rcut is never taken into account, as the radial basis goes exactly to zero after the cutoff. So the padding is actually entirely unnecessary (maybe I could take it out with an if-statement...) However, the default radial basis set uses Gaussians (rbf="gto") which don't exactly go to zero at the cutoff, just to a very low value. Because of this, a padding is needed to avoid boundary effects from atoms that are close to the cutoff value and would be visible in this "tail" of the Gaussians.

Hope this helps

[sirmarcel]

Okay, thanks for the clarification! I think that's enough for now. :)