Generalization of uMLIPs to diverse MOFs and derived properties

Ascertain whether the agreement of universal machine learning interatomic potentials (uMLIPs) with ab initio energies and forces on selected metal–organic frameworks extends to a broader variety of MOFs and to derived properties, including bulk modulus, constant-volume heat capacity, and host–guest interaction energies.

Background

Prior uMLIP studies have presented proof-of-principle results on a few prototypical MOFs, suggesting reasonable agreement with ab initio references. Whether such agreement holds across a chemically and structurally diverse set of frameworks and for downstream properties that probe different aspects of the potential energy surface remained explicitly unclear.

This benchmark targets exactly those derived properties—global PES curvature (bulk modulus), local curvature around equilibrium (heat capacity), and guest–host interactions—providing a stringent test of uMLIP transferability beyond isolated examples.