Laurent Joubert

We present an efficient implementation of the Interacting Quantum Atoms (IQA) energy‑decomposition scheme in AMS, built within a Kohn–Sham DFT framework and using a QTAIM real‑space partitioning following the original IQA formalism and its DFT extensions. A distance‑adaptive algorithm treats short‑range atomic pairs through explicit 6D numerical integration, while medium‑ and long‑range interactions rely on multipolar and asymptotic approximations to reduce cost without losing chemical accuracy. Intra‑atomic Coulomb and exchange‑correlation self‑terms are reconstructed from total‑energy components, and hybrid functionals are handled through a combination of density‑based XC contributions and a Becke–Roussel‑type approximation of the HF‑like exchange fraction. The implementation supports open‑shell, scalar‑relativistic, and spin‑orbit two‑component calculations. Preliminary results indicate that most IQA energy components, and exchange in particular, can be efficiently learned through machine‑learning models. The second part of the talk introduces our recent work on exchange‑correlation effects in interatomic energies for pure density functionals. Standard LDA, GGA, and meta‑GGA functionals yield only intra‑atomic XC terms, meaning that “atoms will interact the one with the others only by classical electrostatics.” Using Green’s second identity, the XC energy is rewritten as a Coulomb‑type interaction between the density and an XC charge, restoring interatomic XC contributions and ensuring exact recovery of the self‑consistent DFT energy.