Refactoring Manybody Calculator with distances squared

libAtoms / matscipy

Materials science with Python at the atomic-scale

http://libatoms.github.io/matscipy/

GNU Lesser General Public License v2.1

188 stars 55 forks source link

Refactoring Manybody Calculator with distances squared #100

Closed prs513rosewood closed 2 years ago

prs513rosewood commented 2 years ago

This PR is for the implementation of the manybody calculator in terms of squared distances, as well as cleanup of the interface.

For testing, the plan is to have three kinds of tests:

Potential tests: check inner potential derivatives with finite differences. These tests do not involve the calculator or atoms, they just check the correct implementation of the derivatives. Should be easy to set up, since now the functions take scalar variables as parameters.
Patch tests: check energy/force/stress/hessian with a restricted number of atoms at fixed positions. This tests the machinery of the manybody calculator
Application tests: check quantities of interest with a "real" system. This may involve energy minimization, elastic properties computation, etc., and check against values obtained with other methods / from literature

prs513rosewood commented 2 years ago

@griessej I've fixed some of the tests that were failing on my end, but some of the manybody tests are not passing, with tolerance errors. Do you have some time this week to look at it with me?

pastewka commented 2 years ago

Hi @griessej @prs513rosewood - what is the status of this MR? Can we merge it to master?

prs513rosewood commented 2 years ago

The PR should be good to go for monoatomic systems, but some tests are still failing. I have not yet made the right changes for heterogeneous systems, which should be confined to the hessian (all other terms should work for heterogeneous systems).

pastewka commented 2 years ago

Let's look at it Friday.

Lucas Frérot @.***> schrieb am Mi., 14. Sept. 2022, 09:38:

The PR should be good to go for monoatomic systems, but some tests are still failing. I have not yet made the right changes for heterogeneous systems, which should be confined to the hessian (all other terms should work for heterogeneous systems).

— Reply to this email directly, view it on GitHub https://github.com/libAtoms/matscipy/pull/100#issuecomment-1246367319, or unsubscribe https://github.com/notifications/unsubscribe-auth/AATJYVYUFNNA6JBVRGXQMTLV6F6GVANCNFSM5STG3LFA . You are receiving this because you commented.Message ID: @.***>

prs513rosewood commented 2 years ago

@pastewka I've fixed / marked as expected fails all the tests that were previously failing.