arXiv subjects

Paulette Clancy

Publications and source records attributed to Paulette Clancy.

11 recordsLinked to original sources

arXiv2026

ARIA: A Causal-Aware Framework for Rescuing LLM Reasoning in Trustworthy Materials Discovery

Generative models have revolutionized the process of materials discovery, yet they often fail to satisfy underlying physical causality. Through an analysis of Large Language Models (LLMs) augmented with knowledge graphs derived from current literature, we uncover a phenomenon termed contextual tunneling, where models "over-anchor" on narrow, retrieved evidence while suppressing global physical reasoning. To address this problem, we introduce ARIA, a causal-aware framework that conditions knowledge use on mechanistic completeness. ARIA routes each query through a three-tier cascade: (i) direct causal reasoning when complete evidence chains of Process-Structure-Property (PSP) are available, (ii) physics-informed analogical transfer for sparse or novel material systems, and (iii) explicit parametric fallback when external evidence is incomplete. As a proof of concept, we construct a Knowledge Graph (KG) containing 2,839 extracted PSP relations from peer-reviewed articles in the materials literature and evaluate ARIA on forward prediction and inverse design tasks for two-dimensional (2D) materials. ARIA mitigates contextual tunneling, improves over unaugmented and naive KG-augmented baselines, and provides further gains when an online literature search is used for evidence enrichment. Crucially, ARIA produces auditable causal traces, enabling physically grounded and trustworthy AI-assisted materials discovery.

cs.AI↗

arXiv2026

Can Coding Agents Reproduce Findings in Computational Materials Science?

Large language models are increasingly deployed as autonomous coding agents and have achieved remarkably strong performance on software engineering benchmarks. However, it is unclear whether such success transfers to computational scientific workflows, where tasks require not only strong coding ability, but also the ability to navigate complex, domain-specific procedures and to interpret results in the context of scientific claims. To address this question, we present AutoMat, a benchmark for evaluating LLM-based agents' ability to reproduce claims from computational materials science. AutoMat poses three interrelated challenges: recovering underspecified computational procedures, navigating specialized toolchains, and determining whether the resulting evidence supports a claim. By working closely with subject matter experts, we curate a set of claims from real materials science papers to test whether coding agents can recover and execute the end-to-end workflow needed to support (or undermine) such claims. We then evaluate multiple representative coding agent settings across several foundation models. Our results show that current LLM-based agents obtain low overall success rates on AutoMat, with the best-performing setting achieving a success rate of only 53%. Error analysis further reveals that agents perform worst when workflows must be reconstructed from paper text alone and that they fail primarily due to incomplete procedures, methodological deviations, and execution fragility. Taken together, these findings position AutoMat as both a benchmark for computational scientific reproducibility and a tool for diagnosing the current limitations of agentic systems in AI-for-science settings.

cs.CL↗

arXiv2025

Synthesis and Characterization of Ultrasonically Atomized Al-Based Alloy Powders for Tunable Thermal Reactivity

Reactive aluminum (Al) alloy powders are promising for advanced manufacturing, joining, and energetic applications, yet scalable routes that couple controlled reactivity with safe handling remain limited. While nanoscale Al powders ignite readily, their agglomeration, handling, and safety limit broad deployment. Here, we manufacture micron-sized Al-based powders produced by ultrasonic atomization (UA), targeting a balance of enhanced reactivity and process robustness. Binary systems (AlCu, AlSi, AlMg) and pure Al were synthesized, and their morphology, phases present, thermal stability, and oxidation behavior were characterized using XRD, SEM, and DTA/TGA in an Ar/O2 environment. We show that alloy selection and UA-controlled microstructure can modify the native Al2O3 passivation, alter oxidation pathways, and shift thermal onsets/exotherms. The results establish a manufacturing-forward framework for designing micron-sized powders with tunable ignition/oxidation behavior.

cond-mat.mtrl-sci↗

arXiv2025

Migration as a Probe: A Generalizable Benchmark Framework for Specialist vs. Generalist Machine-Learned Force Fields

Machine-learned force fields (MLFFs), especially pre-trained foundation models, are transforming computational materials science by enabling ab initio-level accuracy at molecular dynamics scales. Yet their rapid rise raises a key question: should researchers train specialist models from scratch, fine-tune generalist foundation models, or use hybrid approaches? The trade-offs in data efficiency, accuracy, cost, and robustness to out-of-distribution failure remain unclear. We introduce a benchmarking framework using defect migration pathways, evaluated through nudged elastic band trajectories, as diagnostic probes that test both interpolation and extrapolation. Using Cr-doped Sb2Te3 as a representative two-dimensional material, we benchmark multiple training paradigms within the MACE architecture across equilibrium, kinetic (atomic migration), and mechanical (interlayer sliding) tasks. Fine-tuned models substantially outperform from-scratch and zero-shot approaches for kinetic properties but show partial loss of long-range physics. Representational analysis reveals distinct, non-overlapping latent encodings, indicating that different training strategies learn different aspects of system physics. This framework provides practical guidelines for MLFF development and establishes migration-based probes as efficient diagnostics linking performance to learned representations, guiding future uncertainty-aware active learning.

physics.chem-ph↗

arXiv2024

Low-energy pathways lead to self-healing defects in CsPbBr$_3$

Self-regulation of free charge carriers in perovskites via Schottky defect formation has been posited as the origin of the well-known defect tolerance of metal halide perovskite materials that are promising candidates for photovoltaic applications, like solar cells. Understanding the mechanisms of self-regulation, here for a representative of more commercially viable all-inorganic perovskites, promises to lead to the fabrication of better-performing solar cell materials with higher efficiencies. We investigated different mechanisms and pathways of the diffusion and recombination of interstitials and vacancies (Schottky pairs) in CsPbBr$_3$. We use Nudged Elastic Band calculations and ab initio-derived pseudopotentials within Quantum ESPRESSO to determine energies of formation, migration, and activation for these defects. Our calculations uncover defect pathways capable of producing an activation energy at or below the value of 0.53~eV observed for the slow, temperature-dependent recovery of light-induced conductivity in CsPbBr$_3$. Our work reveals the existence of a low-energy diffusion pathway involving a concerted "domino effect" interstitial mechanism, with the net result that interstitials can diffuse more readily over long distances than expected. Importantly, this observation suggests that defect self-healing can be promoted if the "domino effect" strategy can be engaged.

cond-mat.mtrl-sci↗

arXiv2024

Machine Learning-guided accelerated discovery of structure-property correlations in lean magnesium alloys for biomedical applications

Magnesium alloys are emerging as promising alternatives to traditional orthopedic implant materials thanks to their biodegradability, biocompatibility, and impressive mechanical characteristics. However, their rapid in-vivo degradation presents challenges, notably in upholding mechanical integrity over time. This study investigates the impact of high-temperature thermal processing on the mechanical and degradation attributes of a lean Mg-Zn-Ca-Mn alloy, ZX10. Utilizing rapid, cost-efficient characterization methods like X-ray diffraction and optical, we swiftly examine microstructural changes post-thermal treatment. Employing Pearson correlation coefficient analysis, we unveil the relationship between microstructural properties and critical targets (properties): hardness and corrosion resistance. Additionally, leveraging the least absolute shrinkage and selection operator (LASSO), we pinpoint the dominant microstructural factors among closely correlated variables. Our findings underscore the significant role of grain size refinement in strengthening and the predominance of the ternary Ca2Mg6Zn3 phase in corrosion behavior. This suggests that achieving an optimal blend of strength and corrosion resistance is attainable through fine grains and reduced concentration of ternary phases. This thorough investigation furnishes valuable insights into the intricate interplay of processing, structure, and properties in magnesium alloys, thereby advancing the development of superior biodegradable implant materials.

cond-mat.mtrl-sci↗

arXiv2023

Embedding theory in ML toward real-time tracking of structural dynamics through hyperspectral datasets

In-situ Electron Energy Loss Spectroscopy (EELS) is an instrumental technique that has traditionally been used to understand how the choice of materials processing has the ability to change local structure and composition. However, more recent advances to observe and react to transient changes occurring at the ultrafast timescales that are now possible with EELS and Transmission Electron Microscopy (TEM) will require new frameworks for characterization and analysis. We describe a machine learning (ML) framework for the rapid assessment and characterization of in operando EELS Spectrum Images (EELS-SI) without the need for many labeled training datapoints as typically required for deep learning classification methods. By embedding computationally generated structures and experimental datasets into an equivalent latent space through Variational Autoencoders (VAE), we effectively predict the structural changes at latency scales relevant to closed-loop processing within the TEM. The framework described in this study is a critical step in enabling automated, on-the-fly synthesis and characterization which will greatly advance capabilities for materials discovery and precision engineering of functional materials at the atomic scale.

cond-mat.mtrl-sci↗

arXiv2023

Crystal-GFN: sampling crystals with desirable properties and constraints

The discovery of novel solid-state materials, such as electrocatalysts, super-ionic conductors, or photovoltaic materials, plays a critical role in addressing various global challenges. It has, for instance, the potential to significantly improve the efficiency of renewable energy production and storage, thereby making substantial contributions to climate crisis mitigation strategies. In this paper, we introduce Crystal-GFN, a generative model of crystal structures possessing desirable properties and constraints. Operating as a multi-environment, continuous-discrete GFlowNet, it sequentially samples structural attributes of crystalline materials, namely space group, composition and lattice parameters. This domain-inspired approach enables the flexible incorporation of physicochemical and geometric hard constraints. We demonstrate the capabilities of Crystal-GFN to efficiently discover diverse and valid crystals with various properties: low predicted formation energy (median -3.2 eV/atom), band gap close to a target value and high density. Overall, Crystal-GFN is a crystal generation method that addresses several existing challenges in the literature and opens promising paths for accelerating materials discovery with machine learning.

cs.LG↗

arXiv2010

Accurate implementation of leaping in space: The spatial partitioned-leaping algorithm

There is a great need for accurate and efficient computational approaches that can account for both the discrete and stochastic nature of chemical interactions as well as spatial inhomogeneities and diffusion. This is particularly true in biology and nanoscale materials science, where the common assumptions of deterministic dynamics and well-mixed reaction volumes often break down. In this article, we present a spatial version of the partitioned-leaping algorithm (PLA), a multiscale accelerated-stochastic simulation approach built upon the tau-leaping framework of Gillespie. We pay special attention to the details of the implementation, particularly as it pertains to the time step calculation procedure. We point out conceptual errors that have been made in this regard in prior implementations of spatial tau-leaping and illustrate the manifestation of these errors through practical examples. Finally, we discuss the fundamental difficulties associated with incorporating efficient exact-stochastic techniques, such as the next-subvolume method, into a spatial-leaping framework and suggest possible solutions.

physics.chem-ph↗

arXiv2007

Quantifying Stochastic Effects in Biochemical Reaction Networks using Partitioned Leaping

"Leaping" methods show great promise for significantly accelerating stochastic simulations of complex biochemical reaction networks. However, few practical applications of leaping have appeared in the literature to date. Here, we address this issue using the "partitioned leaping algorithm" (PLA) [L.A. Harris and P. Clancy, J. Chem. Phys. 125, 144107 (2006)], a recently-introduced multiscale leaping approach. We use the PLA to investigate stochastic effects in two model biochemical reaction networks. The networks that we consider are simple enough so as to be accessible to our intuition but sufficiently complex so as to be generally representative of real biological systems. We demonstrate how the PLA allows us to quantify subtle effects of stochasticity in these systems that would be difficult to ascertain otherwise as well as not-so-subtle behaviors that would strain commonly-used "exact" stochastic methods. We also illustrate bottlenecks that can hinder the approach and exemplify and discuss possible strategies for overcoming them. Overall, our aim is to aid and motivate future applications of leaping by providing stark illustrations of the benefits of the method while at the same time elucidating obstacles that are often encountered in practice.

q-bio.SC↗

arXiv2006

A "partitioned leaping" approach for multiscale modeling of chemical reaction dynamics

We present a novel multiscale simulation approach for modeling stochasticity in chemical reaction networks. The approach seamlessly integrates exact-stochastic and "leaping" methodologies into a single "partitioned leaping" algorithmic framework. The technique correctly accounts for stochastic noise at significantly reduced computational cost, requires the definition of only three model-independent parameters and is particularly well-suited for simulating systems containing widely disparate species populations. We present the theoretical foundations of partitioned leaping, discuss various options for its practical implementation and demonstrate the utility of the method via illustrative examples.

physics.chem-ph↗