Searcharxiv⌕ Search

arXiv subjects

Bonan Zhu

Publications and source records attributed to Bonan Zhu.

14 recordsLinked to original sources

MatOS: An Agent Operating System for Materials Research

Materials research relies on scattered and growing data, knowledge, and simulation tools, and a single study often spans multi-day workflows across literature analysis, computation, and experiment. Large language model agents can now retrieve materials data, operate simulation codes, and close design-synthesis-characterization loops with autonomous laboratories. However, extending these agents with new capabilities, keeping long-running tasks on track, and letting them learn from past tasks and user feedback remain difficult. Here we present MatOS, an agent operating system for materials research with three core components: (1) a layered scientific asset graph that organizes data, knowledge, tools, skills, roles, and workflows as reusable assets; (2) a stateful adaptive workflow that runs tasks as explicit execution graphs with scientific checks, so that they can be resumed and traced; and (3) evidence-driven asset evolution, which revises these assets from execution records and user feedback. On ResearchClawBench, MatOS improves on the previously best-performing agent by 11.15% with the same base model and extends the cost-score Pareto frontier. With a common base model, it outperforms Codex CLI and Claude Code, exceeding Codex CLI by 2.13% on MatTools and 4.17% on SciAgentGym. In eight research cases, MatOS narrowed more than 30 million candidate structures to 97 solid-electrolyte candidates, identified 89 halide variants with higher computed conductivity than their parent structures, linked the force errors of 11 pretrained interatomic potentials to local atomic environments, and supported two new methods that reduce errors by 3.1% and 13.6% relative to their baselines. These results show that organizing research capabilities as reusable, evolving assets enables agents to carry out long, multistage materials studies. Code is available at https://github.com/DingyangLyu/MatOS.

cs.AI↗

MCSA: Multi-Cutoff Spectral Aggregation in SO(3) Equivariant GNNs for Interatomic Potentials

Machine Learning Interatomic Potentials (MLIPs) based on Equivariant Graph Neural Networks (EGNNs) have achieved remarkable progress in atomistic modeling. However, accurately representing atomic interactions across different spatial scales remains challenging. In this work, we investigate the radial spectral properties of SO(3)-equivariant message passing and establish that the cutoff radius controls the characteristic spatial-frequency scale of radial interactions. Based on this analysis, we propose Multi-Cutoff Spectral Aggregation (MCSA), a plug-and-play module that aggregates radial kernels with different cutoff radii to enrich multi-scale geometric representations. To achieve efficient implementation, we exploit the compact support of radial kernels to fuse cutoff-dependent radial weights before angular tensor-product operations, avoiding redundant angular computations while preserving rotational equivariance. Extensive experiments across representative EGNN backbones demonstrate consistent improvements in energy, force, and stress prediction, as well as enhanced accuracy in higher-order properties, including phonon dispersion and lattice thermal conductivity. Further evaluations on layered materials with distinct interaction scales demonstrate the effectiveness of physically motivated cutoff selection. Code is available at: https://github.com/JF-Tan/MCSA.

cs.LG↗

Understanding compression-extension error asymmetry in universal machine-learning interatomic potentials

Universal machine-learning interatomic potentials are increasingly used across diverse materials, yet their reliability under directional deformation remains unclear. We compare compression and extension along paired deformation paths from the same parent structures for eleven pretrained potentials. The benchmark comprises 1,963 Materials Project trajectory (MPtrj) parents and 1,286 experimental-source MinX-Eq parents at six strain magnitudes from 1% to 10%. Density-functional theory provides references for isotropic, uniaxial and biaxial loading in MPtrj and isotropic loading in MinX-Eq. At $\pm 10\%$, every model has a larger median force error under compression, with pooled isotropic compression-to-extension ratios of 2.12 and 3.30 for MPtrj and MinX-Eq. Decomposing error growth into DFT force-scale growth and relative-error evolution shows that relative errors decrease on both branches, faster under compression in MPtrj but slower under compression in MinX-Eq. Force-scale matching moves the error ratios toward unity, although MinX-Eq retains a compression-side excess across models. Radial and SOAP analyses place compressed structures at shorter separations and farther from public reference pools, while readout-associated model features show larger compression-side displacements. These results distinguish stronger physical responses from changes in relative predictive accuracy and provide a basis for assessing universal potentials in deformation simulations.

cond-mat.mtrl-sci↗

Origins of Universal Machine Learning Force-Field Errors in Multicomponent Materials

Universal machine learning force-field generalization to multicomponent environments generated by compositional design remains insufficiently assessed. We construct a benchmark of 7,599 multicomponent configurations inspired by high-entropy design, elemental substitution and anion mixing. Eleven pretrained models are evaluated against density functional theory for energies, forces and stresses, with assessment extended to elastic, vibrational and adsorption-related properties. Force errors are analysed through training-reference coverage, local geometric heterogeneity, distance directionality and elemental response. Distances to training-reference environments reveal a qualitative association between coverage differences and increasing errors, while substantial variation remains at similar distances. Higher-error groups show greater local geometric heterogeneity, although OMat24 provides broad coverage of these environments. Relative to training-reference pair medians, errors remain low near the median, rise steeply on the compression side and increase more weakly on the extension side. After matching element pairs and absolute distance deviations, compression-side force errors are 1.81-1.95 times extension-side errors. Model-predicted pairwise interaction curves show greater curvature under compression. Fitting difficulty in independent elemental systems correlates with electronic band-energy responses to atomic displacements and Fermi-level shifts, and a similar pattern is observed in multicomponent systems. In parameter-matched comparisons, spherical-harmonic representations with maximum degrees of 2 and 4 lower test force errors for 38 and 40 of 43 elements, respectively, while differences in elemental difficulty remain. These findings inform force-field selection for experimental compositional design and identify targets for training-data sampling and model representations.

cond-mat.mtrl-sci↗

Train for Accuracy, Execute at Scale: Architecture-Preserving Inference for Equivariant Atomistic Foundation Models

Equivariant atomistic foundation models provide broadly transferable interatomic potentials trained against quantum-mechanical reference data, but their repeated execution at simulation scale remains computationally and memory intensive. We present Symmetrix-XL, an inference engine that scales pretrained MACE checkpoints without retraining, distillation, or modification of their learned weights. It combines streamed-edge execution to avoid graph-wide materialization of expanded edge intermediates, model-specialized code generation to compile checkpoint-specific operators, and tiled execution to reuse a bounded device-memory workspace. On complete LAMMPS-step benchmarks, Symmetrix-XL reduces inference time by 3.1-5.0 times relative to ML-IAP + cuEquivariance across tested A100 and RTX 5090 workloads. On a single A100 80 GB GPU, the tested MACE-OMAT-0 capacity boundary increases from 24,565 atoms to 11.24 million atoms. This substantially lowers the hardware threshold for simulations that would otherwise require spatial decomposition across many GPUs and compute nodes. The same backend weak-scales to 703 million atoms on 64 A800 GPUs at 93.8 percent efficiency. Energy, force, stress, molecular-dynamics stability, and Matbench Discovery evaluations reproduce reference MACE behavior within measured tolerances. Case studies spanning solid-state, interfacial, and reactive systems demonstrate the increased capacity in realistic workflows. These results establish post-training execution as a scaling axis complementary to model redesign, compression, and distributed scale-out, and show that the practical accuracy-cost frontier depends on both model architecture and inference execution.

cond-mat.mtrl-sci↗

Identifying Highly Deformable van der Waals Layered Chalcogenides with Superior Thermoelectric Performance Using Deformability Factors and Interpretable Machine Learning

Van der Waals layered chalcogenide-based flexible thermoelectric devices show great potential for applications in wearable electronics. However, materials that are both highly deformable and exhibit superior thermoelectric performance are extremely limited. There is an urgent need for methods that can efficiently predict both deformability and thermoelectric performance to enable high-throughput screening of these materials. In this study, over 1000 van der Waals layered chalcogenides were high-throughput screened from material databases, the deformability of which were predicted with our previously developed deformability factor. An accurate and efficient model based on machine learning methods were developed to predict the thermoelectric properties. Several candidate materials with both deformability and thermoelectric potential were successfully discovered. Among them, NbSe2Br2 was verified by first principles calculations, achieving ZTmax value of 1.35 at 1000K, which is currently the highest value among flexible inorganic thermoelectric materials. And the power factor value of 8.1 μWcm-1K-2 at 300K also surpassed most organic and inorganic flexible thermoelectric materials. Its high deformability mainly attributed to the small slipping energy that allows interlayer slip and the small in-plane modulus that allows deformation before failure. The high ZTmax is mainly contributed by the extremely low thermal conductivity and the high Seebeck coefficient along the out-of-plane direction at high temperature. The high power factor at room temperature is mainly comes from the high conductivity in the in-plane direction. This study is expected to accelerate the development and application of flexible thermoelectric devices based on inorganic semiconductor materials.

cond-mat.mtrl-sci↗

doped: Python toolkit for robust and repeatable charged defect supercell calculations

Defects are a universal feature of crystalline solids, dictating the key properties and performance of many functional materials. Given their crucial importance yet inherent difficulty in measuring experimentally, computational methods (such as DFT and ML/classical force-fields) are widely used to predict defect behaviour at the atomic level and the resultant impact on macroscopic properties. Here we report doped, a Python package for the generation, pre-/post-processing, and analysis of defect supercell calculations. doped has been built to implement the defect simulation workflow in an efficient and user-friendly -- yet powerful and fully-flexible -- manner, with the goal of providing a robust general-purpose platform for conducting reproducible calculations of solid-state defect properties.

cond-mat.mtrl-sci↗

Developments and Further Applications of Ephemeral Data Derived Potentials

Machine-learned interatomic potentials are fast becoming an indispensable tool in computational materials science. One approach is the ephemeral data-derived potential (EDDP), which was designed to accelerate atomistic structure prediction. The EDDP is simple and cost-efficient. It relies on training data generated in small unit cells and is fit using a lightweight neural network, leading to smooth interactions which exhibit the robust transferability essential for structure prediction. Here, we present a variety of applications of EDDPs, enabled by recent developments of the open-source EDDP software. New features include interfaces to phonon and molecular dynamics codes, as well as deployment of the ensemble deviation for estimating the confidence in EDDP predictions. Through case studies ranging from elemental carbon and lead to the binary scandium hydride and the ternary zinc cyanide, we demonstrate that EDDPs can be trained to cover wide ranges of pressures and stoichiometries, and used to evaluate phonons, phase diagrams, superionicity, and thermal expansion. These developments complement continued success in accelerated structure prediction.

cond-mat.mtrl-sci↗

How to verify the precision of density-functional-theory implementations via reproducible and universal workflows

In the past decades many density-functional theory methods and codes adopting periodic boundary conditions have been developed and are now extensively used in condensed matter physics and materials science research. Only in 2016, however, their precision (i.e., to which extent properties computed with different codes agree among each other) was systematically assessed on elemental crystals: a first crucial step to evaluate the reliability of such computations. We discuss here general recommendations for verification studies aiming at further testing precision and transferability of density-functional-theory computational approaches and codes. We illustrate such recommendations using a greatly expanded protocol covering the whole periodic table from Z=1 to 96 and characterizing 10 prototypical cubic compounds for each element: 4 unaries and 6 oxides, spanning a wide range of coordination numbers and oxidation states. The primary outcome is a reference dataset of 960 equations of state cross-checked between two all-electron codes, then used to verify and improve nine pseudopotential-based approaches. Such effort is facilitated by deploying AiiDA common workflows that perform automatic input parameter selection, provide identical input/output interfaces across codes, and ensure full reproducibility. Finally, we discuss the extent to which the current results for total energies can be reused for different goals (e.g., obtaining formation energies).

cond-mat.mtrl-sci↗

Accessible Chemical Space for Metal Nitride Perovskites

Building on the extensive exploration of metal oxide and metal halide perovskites, metal nitride perovskites represent a largely unexplored class of materials. We report a multi-tier computational screening of this chemical space. From a pool of 3660 ABN$_3$ compositions covering I-VIII, II-VII, III-VI and IV-V oxidation state combinations, 279 are predicted to be chemically feasible. The ground-state structures of the 25 most promising candidate compositions were explored through enumeration over octahedral tilt systems and global optimisation. We predict 12 dynamically and thermodynamically stable nitride perovskite materials, including YMoN$_3$, YWN$_3$, ZrTaN$_3$, and LaMoN$_3$. These feature significant electric polarisation and low predicted switching electric field, showing similarities with metal oxide perovskites and making them attractive for ferroelectric memory devices

cond-mat.mtrl-sci↗

Accelerating Cathode Material Discovery through Ab Initio Random Structure Searching

The choice of cathode material in Li-ion batteries (LIBs) underpins their overall performance. Discovering new cathode materials is a slow process, and all major commercial cathode materials are still based on those identified in the 1990s. Materials discovery using high-throughput calculations has attracted great research interest, however, reliance on databases of existing materials begs the question of whether these approaches are applicable for finding truly novel materials. In this work, we demonstrate that ab-initio random structure searching (AIRSS), a first-principles structure prediction methods that does not rely on any pre-existing data, can locate low energy structures of complex cathode materials efficiently based only on chemical composition. We use AIRSS to explore three Fe-containing polyanion compounds as low-cost cathodes. Using known quaternary LiFePO4 and quinary LiFeSO4F cathodes as examples, we easily reproduce the known polymorphs, in addition to predicting other, hitherto unknown, low energy polymorphs, and even finding a new polymorph of LiFeSO4F which is more stable than the known ones. We then explore the phase space for Fe-containing fluoroxalates, predicting a range of redox-active phases that have yet to be experimentally synthesized, demonstrating the suitability of AIRSS as a tool for accelerating the discovery of novel cathode materials.

cond-mat.mtrl-sci↗

Materials Informatics Reveals Unexplored Structure Space in Cuprate Superconductors

High-temperature superconducting cuprates have the potential to be transformative in a wide range of energy applications. In this work we analyse the corpus of historical data about cuprates using materials informatics and re-examine how their structures are related to their critical temperatures (Tc). The available data is highly clustered and no single database contains all the features of interest to properly examine trends. To work around these issues we employ a linear calibration approach that allows us to utilise multiple data sources -- combining fine resolution data for which the Tc is unknown with coarse resolution data where it is known. The hybrid data set constructed enables us to explore the trends in Tc with the apical and in-plane copper-oxygen distances. We show that large regions of the materials space have yet to be explored and highlight how novel experiments relying on nano-engineering of the crystal structure may enable us to explore such new regions. Based on the trends identified we propose that single layer Bi-based cuprates are good candidate systems for such experiments.

cond-mat.supr-con↗

Common workflows for computing material properties using different quantum engines

The prediction of material properties through electronic-structure simulations based on density-functional theory has become routinely common, thanks, in part, to the steady increase in the number and robustness of available simulation packages. This plurality of codes and methods aiming to solve similar problems is both a boon and a burden. While providing great opportunities for cross-verification, these packages adopt different methods, algorithms, and paradigms, making it challenging to choose, master, and efficiently use any one for a given task. Leveraging recent advances in managing reproducible scientific workflows, we demonstrate how developing common interfaces for workflows that automatically compute material properties can tackle the challenge mentioned above, greatly simplifying interoperability and cross-verification. We introduce design rules for reproducible and reusable code-agnostic workflow interfaces to compute well-defined material properties, which we implement for eleven different quantum engines and use to compute three different material properties. Each implementation encodes carefully selected simulation parameters and workflow logic, making the implementer's expertise of the quantum engine directly available to non-experts. Full provenance and reproducibility of the workflows is guaranteed through the use of the AiiDA infrastructure. All workflows are made available as open-source and come pre-installed with the Quantum Mobile virtual machine, making their use straightforward.

cond-mat.mtrl-sci↗

Ab initio random structure searching for battery cathode materials

Cathodes are critical components of rechargeable batteries. Conventionally, the search for cathode materials relies on experimental trial-and-error and a traversing of existing computational/experimental databases. While these methods have led to the discovery of several commercially-viable cathode materials, the chemical space explored so far is limited and many phases will have been overlooked, in particular those that are metastable. We describe a computational framework for battery cathode exploration, based on ab initio random structure searching (AIRSS), an approach that samples local minima on the potential energy surface to identify new crystal structures. We show that, by delimiting the search space using a number of constraints, including chemically aware minimum interatomic separations, cell volumes, and space group symmetries, AIRSS can efficiently predict both thermodynamically stable and metastable cathode materials. Specifically, we investigate LiCoO2, LiFePO4, and LixCuyFz to demonstrate the efficiency of the method by rediscovering the known crystal structures of these cathode materials. The effect of parameters, such as minimum separations and symmetries, on the efficiency of the sampling is discussed in detail. The adaptation of the minimum interatomic distances, on a species-pair basis, from low-energy optimized structures to efficiently capture the local coordination environment of atoms, is explored. A family of novel cathode materials based, on the transition-metal oxalates,is proposed. They demonstrate superb energy density, oxygen-redox stability, and lithium diffusion properties. This article serves both as an introduction to the computational framework, and as a guide to battery cathode material discovery using AIRSS.

physics.chem-ph↗