Objectives and success metrics:

4. Introduce new descriptors and regression models into Tadah!MLIP, validate and verify these on test case materials and implement them efficiently in Tadah!MD. Success metrics: Implementation of GAP and ACE models, and nonlinear kernel regression.

During Dr Kirsz’ PhD, we have made potentials using pairwise forces (Krypton), pair functionals (Nitrogen) and many-body terms (Tantalum) which cover the range of descriptors currently implemented. Some features, notably directional bonding, have not been tested and are expected to require more advanced descriptors. Here, we will add popular descriptors such as three-body, dihedral, SOAP, and cluster expansions (ACE, PACE and MACE). We will also add neural network capability. During Dr Kirsz’ PhD, we have made potentials using pairwise forces (Krypton), pair functionals (Nitrogen) and many body (Tantalum) which cover the range of descriptors currently implemented. Some features, notably directional bonding, have not been tested and are expected to require these more advanced descriptors. Unlike WP1-3, this is a more open-ended workpackage and once those descriptors are implemented, we will add others per external demand and as time permits.