Searcharxiv⌕ Search

arXiv subjects

Gábor Csányi

Publications and source records attributed to Gábor Csányi.

At least 19 recordsLinked to original sources
arXiv2026

A strategic roadmap for an atomistic machine-learning ecosystem

Data-driven machine learning (ML) techniques have become an essential tool in many domains of science. Their application to atomistic simulations of matter is particularly widespread and impactful. This success is due largely to the existence of a well-developed and established physics-based modeling framework, ranging from first-principles electronic-structure calculations to molecular dynamics and statistical sampling, into which ML was integrated naturally to reshape long-standing trade-offs between accuracy, efficiency, and scale. Nevertheless, this integration raises both conceptual and practical challenges, from choosing between data-centric and physics-based modeling approaches to adapting established software stacks to modern hardware accelerators and ML libraries. As the field evolves rapidly, fueled in part by widespread enthusiasm but also by tangible impact, it seems appropriate to take a moment to consider the current state of the art and open challenges, and reflect on what can be done to better coordinate efforts across the community. With this goal in mind, several members of this community met in Lausanne in January 2026 at CECAM to discuss algorithms, models, software and hardware infrastructure, and the most promising scientific applications that have become possible thanks to the use of artificial intelligence in atomic-scale simulations. This strategic roadmap paper summarizes the outcomes of these discussions, suggesting some long-term goals, and some concrete actions, to establish a healthy, sustainable and impactful atomistic ML ecosystem.

physics.chem-ph↗
arXiv2026

Tensor-reduced atomic density representations

Density based representations of atomic environments that are invariant under Euclidean symmetries have become a widely used tool in the machine learning of interatomic potentials, broader data-driven atomistic modelling and the visualisation and analysis of materials datasets.The standard mechanism used to incorporate chemical element information is to create separate densities for each element and form tensor products between them. This leads to a steep scaling in the size of the representation as the number of elements increases. Graph neural networks, which do not explicitly use density representations, escape this scaling by mapping the chemical element information into a fixed dimensional space in a learnable way. We recast this approach as tensor factorisation by exploiting the tensor structure of standard neighbour density based descriptors. In doing so, we form compact tensor-reduced representations whose size does not depend on the number of chemical elements, but remain systematically convergeable and are therefore applicable to a wide range of data analysis and regression tasks.

physics.chem-ph↗
arXiv2026

A fast summation method for the DFT-D3 dispersion correction

The DFT-D3 dispersion correction is routinely added to machine learning force fields (MLFFs) trained on dispersion-deficient functionals such as PBE. Its environment-dependent pair coefficients, however, break the atom-centered separability that fast summation methods require, forcing practitioners either to truncate D3 or to accept a substantial slowdown. We introduce FourierD3, a method that uses a functional low-rank decomposition to restore this separability and enable particle-mesh evaluation in $O(N\log N)$ time without a real-space cutoff on the dispersion sum.

physics.comp-ph↗
arXiv2026

Design Space of Self--Consistent Electrostatic Machine Learning Interatomic Potentials

Machine learning interatomic potentials (MLIPs) have become widely used tools in atomistic simulations. For much of the history of this field, the most commonly employed architectures were based on short-ranged atomic energy contributions, and the assumption of locality still persists in many modern foundation models. While this approach has enabled efficient and accurate modelling for many use cases, it poses intrinsic limitations for systems where long-range electrostatics, charge transfer, or induced polarization play a central role. A growing body of work has proposed extensions that incorporate electrostatic effects, ranging from locally predicted atomic charges to self-consistent models. While these models have demonstrated success for specific examples, their underlying assumptions, and fundamental limitations are not yet well understood. In this work, we present a framework for treating electrostatics in MLIPs by viewing existing models as coarse-grained approximations to density functional theory (DFT). This perspective makes explicit the approximations involved, clarifies the physical meaning of the learned quantities, and reveals connections and equivalences between several previously proposed models. Using this formalism, we identify key design choices that define a broader design space of self-consistent electrostatic MLIPs. We implement salient points in this space using the MACE architecture and a shared representation of the charge density, enabling controlled comparisons between different approaches. Finally, we evaluate these models on two instructive test cases: metal-water interfaces, which probe the contrasting electrostatic response of conducting and insulating systems, and charged vacancies in silicon dioxide. Our results highlight the limitations of existing approaches and demonstrate how more expressive self-consistent models are needed to resolve failures.

physics.chem-ph↗
arXiv2026

Fine-tuning MLIP foundation models: strategies for accuracy and transferability

Adapting machine-learned interatomic potential (MLIP) foundation models to specialised tasks through fine-tuning is an increasingly important practice, yet systematic guidance on when and how to fine-tune is currently limited. We evaluate seven fine-tuning strategies -- naive full-parameter updates, two layer-freezing variants, Low-Rank Adaptation (LoRA), multihead replay, pseudolabelled replay, and replay combined with LoRA -- across five chemically diverse benchmarks (aqueous NaCl, ice polymorphs, S$_\mathrm{N}$2 reactions, SPICE biomolecules, and lithium electrolytes), three generations of foundation models, and training sets spanning five orders of magnitude. To support this evaluation we implement three capabilities in the MACE codebase: LoRA adapted for equivariant message-passing architectures, including both scalar and equivariant linear layers; pseudolabelled replay, which decouples the replay data source from the original pretraining corpus; and model-aware atomic reference energy (E0) reestimation for fine-tuning workflows. We find that foundation model quality, correct E0 initialisation, and well-chosen hyperparameters are prerequisites whose impact routinely exceeds that of the fine-tuning strategy itself. Once these prerequisites are met, most strategies achieve strong target-task accuracy, consistently surpassing models trained from scratch. The practical distinction depends on deployment scope: naive fine-tuning offers the best convergence for single-system applications, while multihead replay -- with either original or pseudolabelled data -- is the only approach tested that consistently preserves out-of-distribution robustness, maintaining both pretraining-distribution accuracy for broader deployment and many-body short-range repulsion.

physics.chem-ph↗
arXiv2026

DFT Accuracy on Crystal Structure Prediction with Machine Learning Interatomic Potentials

We present an evaluation of CSP-MACE-Å, a machine learning interatomic potential intended to replace DFT in crystal structure prediction (CSP). We decompose the total energy into separate intramolecular and intermolecular components. For the intramolecular component, we adopt the MACE-POLAR architecture and train it on the OMol25 dataset. The intermolecular component combines three terms: an intermolecular contribution from the MACE-POLAR model, a long-range dispersion term with the functional form of the XDM correction, and a learned delta model trained to reproduce B86bPBE-XDM intermolecular energies. The learned delta model is trained on residual intermolecular targets derived from 50,000 B86bPBE-XDM calculations on molecular crystal structures. On an evaluation set composed of 19 compounds, including a salt, selected from AstraZeneca's previous CSP publications, CSP-MACE-Å achieves performance comparable to PBE DFT with the Neumann-Perrin dispersion correction. On a second evaluation set composed of 28 compounds, including cocrystals and salts, collated from the seven CSP blind tests, CSP-MACE-Å achieves performance close to B86bPBE-XDM DFT. In both evaluation sets, reranking with harmonic free energies substantially improves performance relative to ranking by energy alone. Across our evaluation suite, CSP-MACE-Å is shown to outperform the MACE-POLAR-1 and UMA-OMC foundation models. Lastly, on a set of five compounds, CSP-MACE-Å is shown to capture temperature-dependent trends in the relative stability of polymorphs through estimation of the free energy under the harmonic approximation. By running multiple orders of magnitude faster than DFT, CSP-MACE-Å enables energy and free energy evaluation of far more candidate structures, providing greater confidence when derisking solid forms.

physics.chem-ph↗
arXiv2026

Systematic Fine-Tuning of MACE Interatomic Potentials for Catalysis

Once trained, machine-learned interatomic potentials (MLIPs) provide a fast and accurate way to study catalytic reaction pathways, but their performance strongly depends on the training set. Here, we compare nine MLIPs trained with different data sets and strategies, including from-scratch (FS) training and fine-tuning (FT) of large foundation models. The models are evaluated on reaction energies, $E_{r}$, and reaction energy barriers, $E_{a}$, for 141 reactions, including CO$_2$ reduction to C$_2$ and C$_3$ products, propane dehydrogenation, hydrogen intercalation on Pd, and out-of-distribution oxygen evolution reaction (OER) on metal oxides. FS models trained with 5%--10% perturbed high-energy configurations from molecular dynamics or contour exploration reduce the error by more than twofold compared with models trained only on relaxation trajectories. In contrast, FT MLIPs are less sensitive to sampling and transfer well to out-of-distribution reactions. An MLIP fine-tuned on metallic catalysts achieves a 0.30 eV MAE for OER on iridium oxide polymorphs, outperforming out-of-the-box MACE-MH-1 by 0.08 eV and the best FS model by 0.14 eV. A model fine-tuned to O and OH adsorption on metal oxides gives a 0.19 eV reaction-barrier MAE for out-of-distribution CO$_2$RR on Cu, comparable to an FS model trained on in-distribution C--C bond-breaking reactions. Finally, a large MLIP fine-tuned on 49,860 configurations gives the best overall performance across metallic and metal-oxide catalysts and was used to screen a large left-out set of bimetallic alloys, achieving a 0.15 eV MAE for $E_{r}$, even for adsorbates on unseen Miller-index surfaces such as (532). This work identifies the training configurations needed for accurate FS and FT MLIPs for catalytic reaction modeling.

physics.chem-ph↗
arXiv2026

Roadmap on Advancements of the FHI-aims Software Package

Electronic-structure theory is the foundation of the description of materials including multiscale modeling of their properties and functions. Obviously, without sufficient accuracy at the base, reliable predictions are unlikely at any level that follows. The software package FHI-aims has proven to be a game changer for accurate free-energy calculations because of its scalability, numerical precision, and its efficient handling of density functional theory (DFT) with hybrid functionals and van der Waals interactions. It treats molecules, clusters, and extended systems (solids and liquids) on an equal footing. Besides DFT, FHI-aims also includes quantum-chemistry methods, descriptions for excited states and vibrations, and calculations of various types of transport. Recent advancements address the integration of FHI-aims into an increasing number of workflows and various artificial intelligence (AI) methods. This Roadmap describes the state-of-the-art of FHI-aims and advancements that are currently ongoing or planned.

cond-mat.mtrl-sci↗
arXiv2026

Atomic-scale order enables high thermal boundary conductance at $β$-Ga$_2$O$_3$/4H-SiC interfaces

Thermal boundary conductance (TBC) at dissimilar interfaces imposes a fundamental limit on electronic device performance, yet predicting and understanding heat transport across realistic, disordered boundaries remains elusive. Here, we develop a computational framework that combines machine-learned interatomic potentials with lattice dynamics to address the long-standing problem of how interfacial structure, from disordered to atomically sharp, affects thermal transport in the technologically important $β$-Ga$_2$O$_3$/4H-SiC heterostructure. By explicitly accounting for phonon wave-particle duality, we show that interfacial disorder introduces additional interfacial phonon modes that facilitate vibrational impedance matching between the two highly dissimilar crystals, yet it simultaneously disrupts interfacial phonon coherence and limits the potential heat-transport benefit. Our atomistic simulations further indicate that restoring atomic-scale order preserves coherence and yields markedly higher conductance. These insights motivate the controlled epitaxial growth of $β$-Ga$_2$O$_3$/4H-SiC heterostructures with systematically tuned interfacial order. Experimental measurements validate our predictions, achieving a record-high TBC of 231 MW m$^{-2}$ K$^{-1}$ at atomically sharp junctions. Beyond the immediate implications for $β$-Ga$_2$O$_3$-based power electronics, our results establish the preservation of interfacial phonon coherence as an effective strategy for mitigating thermal bottlenecks in mismatched systems.

cond-mat.mtrl-sci↗
arXiv2026

Equivariant Many-body Message Passing Interatomic Potentials for Magnetic Materials

Magnetism governs key properties of materials used in energy, data storage, and spintronic technologies, yet its complex coupling to lattice and electronic degrees of freedom challenges conventional first-principles approaches. We introduce an equivariant message-passing graph neural network that embeds atomic magnetic moments as explicit degrees of freedom, enabling the learning of magnetic interactions beyond collinear approximations. The model learns physically consistent and transferable representations of magnetic behaviour and can incorporate spin-orbit coupling, achieving near density-functional-theory accuracy with strong data efficiency across diverse magnetic systems by fine-tuning from a pre-trained model. Applications to structural transformations, finite-temperature magnetic phenomena, and materials screening for strongly spin-orbit coupled materials demonstrate transferable magnetic behaviour, establishing a practical foundation for data-driven, high-throughput discovery of complex magnetic materials.

cond-mat.mtrl-sci↗
arXiv2026

The Open Molecules 2025 (OMol25) Dataset, Evaluations, and Models

Machine learning (ML) models hold the promise of transforming atomic simulations by delivering quantum chemical accuracy at a fraction of the computational cost. Realization of this potential would enable high-throughout, high-accuracy molecular screening campaigns to explore vast regions of chemical space and facilitate ab initio simulations at sizes and time scales that were previously inaccessible. However, a fundamental challenge to creating ML models that perform well across molecular chemistry is the lack of comprehensive data for training. Despite substantial efforts in data generation, no large-scale molecular dataset exists that combines broad chemical diversity with a high level of accuracy. To address this gap, Meta FAIR introduces Open Molecules 2025 (OMol25), a large-scale dataset composed of more than 100 million density functional theory (DFT) calculations at the $ω$B97M-V/def2-TZVPD level of theory, representing billions of CPU core-hours of compute. OMol25 uniquely blends elemental, chemical, and structural diversity including: 83 elements, a wide-range of intra- and intermolecular interactions, explicit solvation, variable charge/spin, conformers, and reactive structures. There are ~83M unique molecular systems in OMol25 covering small molecules, biomolecules, metal complexes, and electrolytes, including structures obtained from existing datasets. OMol25 also greatly expands on the size of systems typically included in DFT datasets, with systems of up to 350 atoms. In addition to the public release of the data, we provide baseline models and a comprehensive set of model evaluations to encourage community engagement in developing the next-generation ML models for molecular chemistry.

physics.chem-ph↗
arXiv2026

False Metallization in Short-Ranged Machine Learned Interatomic Potentials

Machine learned interatomic potentials (MLIPs) have enabled atomistic simulations with ab initio accuracy for a fraction of the computational cost. However, many widely used MLIPs are short-ranged and do not accurately capture long-ranged electrostatic interactions. At interfaces with polar solvents, such as water, this deficiency can drive unphysical long-distance dipolar alignment far away from the interface. Here we reveal that neglecting long-ranged physics leads to spurious metallization of the water layer due to artificially large fluctuations of the total solvent dipole, similar to the electron rearrangement observed to prevent polar catastrophes at polar interfaces. This metallization is eliminated in MLIPs that explicitly include long-ranged electrostatics. Our results showcase a fundamental flaw of short-ranged MLIPs, highlighting that long-ranged electrostatics are essential for studying systems with a polar-liquid component, especially if one is interested in electronic properties.

physics.chem-ph↗
arXiv2026

Regularity Priors for the Linear Atomic Cluster Expansion

Machine-learned interatomic potentials enable large systems to be simulated for long time scales at near ab-initio accuracy. This accuracy is achieved by fitting extremely flexible model architectures to high quality reference data. In practice, this flexibility can cause unwanted behavior such as jagged predicted potential energy surfaces and generally poor out-of-distribution behavior. We investigate a general strategy for incorporating prior beliefs on the regularity of the target energy into linear Atomic Cluster Expansion (ACE) models and explore to what extent this approach improves the quality of the fitted models. Our main focus is an over-regularisation that replicates the Gaussian broadening used in Smooth Overlap of Atomic Positions (SOAP) descriptors within the ACE framework. Numerical tests indicate that the exact form of the prior is non-critical but that including such a prior leads to significant improvement in test errors, consistent repulsion at close-approach, eliminates spurious false minima in the potential energy and enhances stability during molecular dynamics simulations.

physics.chem-ph↗
arXiv2026

MACE-POLAR-1: A Polarisable Electrostatic Foundation Model for Molecular Chemistry

Accurate modelling of electrostatic interactions and charge transfer is fundamental to computational chemistry, yet most machine learning interatomic potentials (MLIPs) rely on local atomic descriptors that cannot capture long-range electrostatic effects. We present a new electrostatic foundation model for molecular chemistry that extends the MACE architecture with explicit treatment of long-range interactions and electrostatic induction. Our approach combines local many-body geometric features with a non-self-consistent field formalism that updates learnable charge and spin densities through polarisable iterations to model induction, followed by global charge equilibration via learnable Fukui functions to control total charge and total spin. This design enables an accurate and physical description of systems with varying charge and spin states while maintaining computational efficiency. Trained on the OMol25 dataset of 100 million hybrid DFT calculations, our models achieve chemical accuracy across diverse benchmarks, with accuracy competitive with hybrid DFT on thermochemistry, reaction barriers, conformational energies, and transition metal complexes. Notably, we demonstrate that the inclusion of long-range electrostatics leads to a large improvement in the description of non-covalent interactions and supramolecular complexes over non-electrostatic models, including sub-kcal/mol prediction of molecular crystal formation energy in the X23-DMC dataset and a fourfold improvement over short-ranged models on protein-ligand interactions. The model's ability to handle variable charge and spin states, respond to external fields, provide interpretable spin-resolved charge densities, and maintain accuracy from small molecules to protein-ligand complexes positions it as a versatile tool for computational molecular chemistry and drug discovery.

physics.chem-ph↗
arXiv2026

Better without U: Impact of Selective Hubbard U Correction on Foundational MLIPs

The training of foundational machine learning interatomic potentials (fMLIPs) relies on diverse databases with energies and forces calculated using ab initio methods. We show that fMLIPs trained on large datasets such as MPtrj, Alexandria, and OMat24 encode inconsistencies from the Materials Project's selective use of the Hubbard U correction, which is applied to certain transition metals only if O or F atoms are present in the simulation cell. This inconsistent use of +U creates two incompatible potential-energy surfaces (PES): a lower-energy GGA surface and a higher-energy GGA+U one. When trained on both, MLIPs interpolate between them, leading to systematic underbinding, or even spurious repulsion, between U-corrected metals and oxygen- or fluorine-containing species. Models such as MACE-OMAT and -MPA exhibit repulsion between U-corrected metals and their oxides, limiting their value for studying catalysis and oxidation. We link the severity of this pathology to the oxygen number density in U-corrected training configurations. This explains why OMAT-trained models are most affected and suggests the issue might worsen as expanding future datasets increasingly include configurations with low oxygen content, such as those generated through combinatorial exploration of multi-element or defect-containing systems. Our simple per-U-corrected-atom shift aligns PBE+U and PBE energies for identical structures, yielding a smoother PES compared to existing correction schemes, which target phase diagram accuracy. As a result, models trained on datasets with our shift applied exhibit smaller mean absolute errors for the adsorption energies of oxygen on U-corrected elemental slabs. Since datasets omitting +U entirely (e.g. MatPES, MP-ALOE) avoid these pathologies, we recommend excluding +U in future fMLIP datasets. For existing datasets, our post-hoc correction provides a low-cost improvement.

physics.chem-ph↗
arXiv2025

Cross Learning between Electronic Structure Theories for Unifying Molecular, Surface, and Inorganic Crystal Foundation Force Fields

Creating a single unified interatomic potential capable of attaining ab initio accuracy across all chemistry remains a long-standing challenge in computational chemistry and materials science. This work introduces a training protocol for foundation machine-learning interatomic potentials (MLIPs) that bridge molecular, surface, and materials chemistry through cross-domain learning. First, we introduce enhancements to the MACE architecture that improve its performance on chemically diverse databases by increasing weight sharing across chemical elements and introducing non-linear factors into the tensor decomposition of the product basis. Second, we develop a multi-head replay post-training methodology that enables efficient knowledge transfer across diverse chemical domains. By fine-tuning on datasets at different levels of electronic structure theory, including inorganic crystals, molecular systems, surface chemistry, and reactive organic chemistry, we demonstrate that a single unified model achieves state-of-the-art performance across several chemical domains. Comprehensive benchmarking reveals superior cross-domain transferability compared with existing specialised and multi-task models, with notable improvements in molecular and surface properties while maintaining state-of-the-art performance in materials-property prediction.

physics.chem-ph↗
arXiv2025

How Accurate Are DFT Forces? Unexpectedly Large Uncertainties in Molecular Datasets

Training of general-purpose machine learning interatomic potentials (MLIPs) relies on large datasets with properties usually computed with density functional theory (DFT). A pre-requisite for accurate MLIPs is that the DFT data are well converged to minimize numerical errors. A possible symptom of errors in DFT force components is nonzero net force. Here, we consider net forces in datasets including SPICE, Transition1x, ANI-1x, ANI-1xbb, AIMNet2, QCML, and OMol25. Several of these datasets suffer from significant nonzero DFT net forces. We also quantify individual force component errors by comparison to recomputed forces using more reliable DFT settings at the same level of theory, and we find significant discrepancies in force components averaging from 1.7 meV/Å in the SPICE dataset to 33.2 meV/Å in the ANI-1x dataset. These findings underscore the importance of well converged DFT data as increasingly accurate MLIP architectures become available.

physics.chem-ph↗
arXiv2025

Bond-Network Entropy Governs Heat Transport in Coordination-Disordered Solids

Understanding how the vibrational and thermal properties of solids are influenced by atomistic structural disorder is of fundamental scientific interest, and paramount to designing materials for next-generation energy technologies. While several studies indicate that structural disorder strongly influences the thermal conductivity, the fundamental physics governing the disorder-conductivity relation remains elusive. Here we show that order-of-magnitude, disorder-induced variations of conductivity in network solids can be predicted from a bond-network entropy, an atomistic structural descriptor that quantifies heterogeneity in the topology of the atomic-bond network. We employ the Wigner formulation of thermal transport to demonstrate the existence of a relation between the bond-network entropy, and observables such as smoothness of the vibrational density of states (VDOS) and macroscopic conductivity. We also show that the smoothing of the VDOS encodes information about the thermal resistance induced by disorder, and can be directly related to phenomenological models for phonon-disorder scattering based on the semiclassical Peierls-Boltzmann equation. Our findings rationalize the conductivity variations of disordered carbon polymorphs ranging from nanoporous electrodes to defective graphite used as a moderator in nuclear reactors.

cond-mat.mtrl-sci↗