Challenging Atomistic Foundation Models on Disordered Carbons

Joseph Hart; Samuel G. H. Brookes; Fabian L. Thiemann; Christoph Schran

Home/Archive/arXiv

arXiv · 2610.26701

Challenging Atomistic Foundation Models on Disordered Carbons

Joseph Hart·Samuel G. H. Brookes·Fabian L. Thiemann·Christoph Schran

View original source ↗Read source document ↗Export metadata ↓

Abstract

Foundation machine-learned interatomic potentials (MLIPs) promise broad transferability, yet their out-of-domain reliability is hard to establish. Here, we use pure and O/H-containing disordered carbon spanning multiple phases, competing hybridisations and porous frameworks to stress-test six popular foundation MLIPs. We first screen all six on density functional theory (DFT) energy and force errors and binding curves, then subject four to melt-quench-anneal molecular dynamics and evaluate the resulting structural observables. Binding curves and O/H-containing trajectories reveal unphysical extrapolation and dynamical failures that low DFT energy and force errors fail to reveal. Of the models tested, MACE-MH-1 is the most consistently robust, with low DFT errors and stable MD. Despite differing architectures, OMAT-based models share structural biases, highlighting the importance of training coverage. All four models generate structures of nanoporous carbon that reproduce the principal experimental and specialised carbon potential signatures. Incorporating O/H leaves short-range order largely unchanged but shifts pore-size distributions towards larger pores. These results emphasise the need for application-specific validation and extend molecular dynamics modelling of nanoporous carbon towards experimentally realistic structures.

Explore related subjects

physics.chem-ph cond-mat.mtrl-sci physics.comp-ph

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

Joseph Hart, Samuel G. H. Brookes, Fabian L. Thiemann, Christoph Schran. 2026-10-19. Challenging Atomistic Foundation Models on Disordered Carbons. https://arxiv.org/abs/2610.26701

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Quantum-Accurate Vibrational and Conformational Dynamics from Small Molecules to Proteins with Machine-Learned Force Fields

Biomolecular thermodynamics and spectroscopy depend on relative conformer energies, local curvatures, and collective dipole fluctuations on the potential-energy surface. Conventional molecular mechanics force fields enable large-scale simulations, but their fixed functional forms can misrepresent infrared intensities, mode character, and environment-dependent vibrational response. Here we assess general-purpose machine-learned force fields across small molecules, finite-temperature infrared spectra, gas-phase peptides, and monomeric, oligomeric, and solvated protein assemblies. To enable this analysis, we introduce QVib, a dataset of 293 molecules and 1365 conformers, together with peptide amide-band benchmarks and p53 oligomerization-domain models, to evaluate vibrational transferability from DFT references to experimental spectra. Across these systems, machine-learned force fields substantially improve over molecular mechanics in reproducing DFT-level forces, vibrational frequencies, densities of states, mode eigenvectors, conformational energetics, and experimental infrared spectra. Among models with explicit long-range electrostatics, SO3LR provides the most favourable accuracy-cost balance for the biomolecular systems considered. These results show that machine-learned force-field dynamics can recover collective, environment-dependent vibrational landscapes at near-DFT fidelity, enabling spectroscopically validated biomolecular simulations at force-field-like cost.

physics.chem-ph↗

arXiv2026

Smoothed truncated Coulomb potential for periodic Gaussian-basis Hartree--Fock exchange

Truncated Coulomb (TC) potentials reduce finite-size errors and accelerate thermodynamic-limit convergence in periodic Hartree--Fock (HF) calculations, but their use with Gaussian basis sets is complicated by the evaluation of electron repulsion integrals (ERIs), particularly for nonspherical truncation domains and all-electron calculations. We introduce the smoothed truncated Coulomb (sTC) potential as a systematically improvable approximation to a parent TC potential. A real-space Gaussian convolution smooths the sharp truncation boundary, and a single dimensionless parameter, $η$, controls the width of the smoothing window, which can be tightened to systematically approach the TC reference. The smoothing by sTC enables a dual-space algorithm for evaluating periodic Gaussian-basis ERIs that requires neither a large plane-wave basis nor new molecular integral kernels and is applicable to both pseudopotential and all-electron calculations, including the important Wigner--Seitz-cell truncation boundaries. Benchmarks spanning insulating, semiconducting, layered, metallic, and molecular-crystal systems show that sTC-based HF closely reproduces TC results in pseudopotential calculations and extends TC-quality calculations to all-electron settings. Across these systems, sTC substantially improves thermodynamic-limit convergence relative to the probe-charge Ewald method while remaining practical for all-electron calculations in which direct TC-based calculations are computationally challenging.

physics.chem-ph↗

arXiv2026

Intermolecular Packing and Cooperative Stabilization in Nucleotide Monophosphate Crystals

We examine the known polymorphs of cytidine 5$^\prime$-monophosphate (CMP) using dispersion-corrected density functional theory. The monoclinic phase is predicted to be the most stable, lying 1.76 and 3.00 kcal/mol per molecular unit below the orthorhombic and triclinic forms, respectively. Monoclinic CMP contains a ribose-based H-bonding motif along the crystallographic $\vec{b}$ direction, comprising alternating intra- and intermolecular interactions, motivating an investigation of cooperative stabilization along this pathway. We employ chain-of-molecules (COM) and chain-of-slabs (COS) models to quantify the direction- and size-dependent stabilization and extend the analysis to the corresponding 5$^\prime$-monophosphates of thymidine, uridine, adenosine, and guanosine. Successive dimer-addition energies reveal the strongest size-dependent cooperative stabilization along the $\vec{b}$ direction, whereas growth along $\vec{a}$ and $\vec{c}$ exhibits substantially weaker size dependence. Among the nucleotides examined, adenosine 5$^\prime$-monophosphate shows the strongest chain-growth stabilization in the COM model, whereas the uridine variant shows the strongest stabilization in the COS model. Guanosine 5$^\prime$-monophosphate exhibits the weakest response in both models. Minimal-basis iterative stockholder (MBIS) analysis further supports molecular polarization along the growing assemblies. These results demonstrate how nucleobase identity and the surrounding crystal environment modulate direction-dependent cooperative stabilization associated with extended hydrogen-bond networks.

physics.chem-ph↗