The results show that incorrect source-level phase energetics can be reversed through target-level fine-tuning, and suggest a practical multi-fidelity strategy in which pretraining prioritizes broad, consistent, and affordable data, while compact target-level datasets impose energetics through application-specific fine-tuning.
Abstract
Foundation machine-learned force fields (MLFFs) are often pretrained on broad materials datasets whose electronic-structure conventions may not reproduce the phase energetics required for a specific correlated material. Using NiO as a case study, we examine whether incorrect source-level phase energetics can be corrected efficiently through target-level fine-tuning. Along a common structural interpolation, non-spin-polarized PBE and ferromagnetic PBE+U predict opposite energetic orderings of the octahedral Oct and square-planar Sqr phases. Pretrained DPA-4 models adapt rapidly to the NiO PBE+U surface, reaching energy and force root-mean-square errors (RMSEs) of approximately 0.5 meV/atom and 30 meV/{\AA}, respectively, with approximately 170 PBE+U labels. Crucially, models previously fine-tuned to the opposing no-U surface recover the qualitative PBE+U phase ordering with nearly the same target-data efficiency as models fine-tuned directly from their respective pretrained initializations. Our results show that incorrect source-level phase energetics can be reversed through target-level fine-tuning, and suggest a practical multi-fidelity strategy in which pretraining prioritizes broad, consistent, and affordable data, while compact target-level datasets impose energetics through application-specific fine-tuning.
Foundation machine-learning interatomic potentials (MLIPs) enable atomistic simulations at substantially lower computational cost than first-principles methods, but their reliability across structural geometries remains insufficiently understood. Here, we construct a density-functional-theory dataset of ZrO2 configurations spanning bulk, slab, particle, neck, and atomically thin wire environments motivated by an experimentally observed ZrO2 desintering process involving neck thinning and atomic wire formation. We first benchmark 26 pretrained MLIPs and observe pronounced geometry-dependent degradation in zero-shot predictions. Without any training, after only reference-energy alignment, the best zero-shot model (ORB-V3) reaches energy and force root-mean-square errors of 6 meV/atom and 197.3 meV/{\AA}, respectively, with the largest force errors in neck and wire configurations. We then compare zero-shot inference, fine-tuning, and training from scratch strategies. Fine-tuning yields lower energy and force errors than training from scratch, while both require comparable wall-clock time. Geometry-specific fine-tuning improves in-domain accuracy but frequently produces negative transfer to other structural classes, whereas mixed-geometry fine-tuning reduces cross-geometry errors. Evaluations of elastic and vibrational properties, surface energies, and neck dynamics further show that rankings based on average energy and force errors do not universally predict property-level behavior. These results demonstrate that geometry-diverse target data and independent physical validations are necessary when adapting foundation MLIPs to low-coordination (ionic) nanostructures.
P. Zanineli, B. Focassio, G. R. Schleder· 0 citations
Pretrained machine-learning interatomic potentials, so-called universal or foundation models offer an appealing starting point for atomistic simulations, but their accuracy for material-specific observables often remains limited without additional reference data (fine-tuning). Here, we systematically quantify how much first-principles data are required to convert universal models into ab initio-accurate material-specific potentials, and ask whether fine-tuning is necessarily preferable to training from scratch. We compare five universal MLIP frameworks, MACE-MP-0, SevenNet-0, GRACE-1L-OAM, MatterSim-v1-5M and ORB-v2, across seven chemically diverse systems incorporating rare and reactive events. Fine-tuning on only 10 AIMD-derived configurations is insufficient for the investigated systems; 200 configurations succeed in favorable cases, but the outcome remains strongly system-dependent. By contrast, 2000 AIMD configurations constitute a robust default, yielding low force and energy errors and reproducing the target material-specific observables. Moderately dense sub-sampling of the AIMD trajectory reduces the required trajectory length tenfold with little loss in model quality. Training from scratch on the same datasets is competitive with, and often slightly more accurate than, naive fine-tuning for MACE and SevenNet, whereas GRACE requires more data. The energy profile for a sulfur-vacancy jump in MoS$_2$ reveals that low trajectory-level errors do not guarantee a correct reaction profile, highlighting the need for observable-level validation. Finally, we show that averaging independently trained models improves predictions in scarce-data regimes at no additional first-principles cost. Together, these results provide practical guidelines for converting limited AIMD reference data into reliable material-specific MLIPs for nanosecond-timescale simulations at near-DFT accuracy.
Accurate prediction of electronic Hamiltonians would enable broad property inference while avoiding the high computational cost of Density Functional Theory (DFT). However, progress toward general-purpose materials foundation models is limited by a data bottleneck: existing Hamiltonian datasets are typically small, lack structural diversity, and often omit essential relativistic physics such as spin--orbit coupling (SOC). We therefore construct UniHam, a large-scale Hamiltonian dataset and benchmark suite comprising 100,000+ DFT-computed complex-valued Hermitian Hamiltonians with full SOC, covering 72 elements and a wide range of crystal geometries and symmetries (spanning diverse lattice types and space-group families). Building on UniHam, we benchmark two representative state-of-the-art models under a standardized protocol and introduce complementary evaluation metrics that jointly assess three dimensions: (i) Hamiltonian reconstruction accuracy, (ii) out-of-distribution (OOD) generalization across composition/symmetry shifts, and (iii) the ability to support downstream property prediction from the predicted Hamiltonians. Experiments on UniHam demonstrate that the proposed benchmark and metrics effectively differentiate model capabilities, revealing intrinsic SOC- and element-dependent failure modes, large variations in compositional OOD robustness, and the necessity of spectral-level evaluation to assess whether Hamiltonian predictions reliably support downstream electronic-structure properties. Overall, UniHam provides a reproducible, SOC-complete benchmark that can sharpen model comparisons and accelerate the development of next-generation foundation models for quantum materials.
Yuewen Huang, Pin Chen, Yutong Lu· Proceedings of the 32nd ACM...· 0 citations
A structure-aware graph neural network is trained to predict cross-functional energy residuals and align inconsistent DFT energy scales, which enables reliable predictions of phase stability, battery voltage profiles, and reaction thermodynamics, while allowing the integration of multi-source DFT data to advance the development of high-performance materials foundation models.
Yi-Dong Huang, Teng-Long Lu, Hanwen Kang et al.· arXiv.org· 0 citations
Machine Learning interatomic potentials (MLIPs) have emerged as powerful tools for molecular dynamics (MD) simulations with their competitive accuracy and computational efficiency. However, MLIPs often exhibit unphysical behavior when encountering configurations that deviate significantly from their training data distribution, leading to simulation instabilities and unreliable dynamics. This limits their reliability for materials simulations. We therefore present a physics-informed pretraining strategy that leverages simple empirical potentials to improve the robustness and stability of MLIPs for MD simulations. We demonstrate this approach through a pretraining-finetuning pipeline where MLIPs are initially pretrained on data labeled with embedded atom model (EAM) potentials and subsequently finetuned on the quantum mechanical ground truth data. Evaluation across three material systems (phosphorus, silica, and a subset of Materials Project) and three representative MLIP architectures (CGCNN, M3GNet, and TorchMD-NET) demonstrates that this physics-informed pretraining consistently improves both prediction accuracy as well as stability in MD compared to the baseline models.
Qian-Yu Zheng, Victor Fung· Journal of Chemical Informat...· 0 citations
Density-functional theory (DFT) has been the workhorse of first-principles calculations for decades, and DFT-derived energies and forces are now widely used to train machine learning models of inter-atomic potentials. However, DFT’s single-particle treatment of exchange-correlation functionals severely limits accuracy for materials with open d- and f-shell elements, and ML models trained on such data inherit this limitation. Dynamical mean-field theory (DMFT) addresses this limitation by explicitly incorporating local electronic correlations, albeit at a significantly higher computational cost. In this work, we develop deep-learning models trained on ab-initio DFT+DMFT calculations to predict electronic self-energies from non-interacting Green’s functions. Using the correlated metal SrVO
3
as a prototype, we show that accurate self-energy predictions can be achieved from small datasets. Through transfer-learning, models pre-trained on SrVO
3
successfully predict the self-energies of CaVO
3
, BaVO
3
and SrNbO
3
, despite differences in composition and electronic structure. Moreover, models pretrained on SrVO
3
and SrNbO
3
can predict self-energy of BaNbO
3
without training on its self-energy. This approach captures temperature variation, extends beyond d
1
perovskites and drastically reduces computational time. These results establish deep-learning as an efficient surrogate for computationally demanding DMFT calculations, enabling rapid prediction of correlation-driven properties, paving the way for a transformative shift in materials theory.
