Accurate and Transferable DFT: Machine-Learned aPBE0 and Physics-Based XDM in FHI-Aims

Abstract:

I present the implementation and application of the accurate, machine-learned aPBE0 functional and the exchange-hole dipole moment (XDM) dispersion correction within the FHI-aims code. The aPBE0 functional, trained on benchmark molecular data, effectively mitigates delocalization error while retaining the computational cost of hybrid DFT. Its non-empirical design yields reliable energetics across a broad chemical space. Combined with XDM, which I have extended to solids and molecular crystals within FHI-aims, the approach delivers state-of-the-art accuracy in cohesive and lattice energy predictions. This combination offers a transferable, physically grounded framework for tackling noncovalent interactions in condensed phase systems.

Bio:

Alastair J. A. Price is a postdoctoral fellow at the University of Toronto and the Acceleration Consortium, working with Professors O. Anatole von Lilienfeld and Alán Aspuru-Guzik. He completed his PhD under Professor Erin R. Johnson at Dalhousie University, where he developed a new implementation of the XDM dispersion model. His research focuses on advancing accurate and transferable density functionals and dispersion corrections, including the machine-learned aPBE0 functional and XDM for solids and molecular crystals within FHI-aims.