Thermal Stability, Negative Thermal Expansion, and Pore Accessibility of HKUST‐1 From M3GNet Machine Learning Molecular Dynamics

The temperature‐dependent structure of metal–organic frameworks such as HKUST‐1 is difficult to model with classical force fields. Here, isothermal–isobaric molecular dynamics driven by the M3GNet universal machine learning potential, applied without system‐specific parameterization, characterizes HKUST‐1 from 77 to 573 K. The relaxed framework reproduces the experimental lattice parameter and cell volume to within 0.24% and 0.71%, and exhibits negative thermal expansion, contracting by 0.47% with a linear coefficient of −9.44 × 10 −6 K −1 . Three measures confirm structural integrity: the root mean square deviation stays below 0.8 Å and scales harmonically with temperature; radial distribution function peaks remain sharp, with Cu‐centered distances invariant and linker distances contracting less than the lattice; and velocity autocorrelation relaxation times increase with atomic mass and vary weakly with temperature. The vibrational density of states reveals Cu paddle‐wheel modes invariant within resolution, while linker deformation modes harden by 2.6–3.3 cm −1 per 100 K and the C─H stretch softens through anharmonicity. Averaged over 31 snapshots, the pore‐limiting diameter narrows by 8.4%, yet surface area, pore volume, and gas accessibility are retained. HKUST‐1 thus preserves crystallinity and porosity, establishing universal potentials as a transferable route for thermal characterization of frameworks.