arXiv subjects

Linda Hung

Publications and source records attributed to Linda Hung.

9 recordsLinked to original sources

arXiv2026

Bridging electrode preparation and electrocatalyst performance with physics-based causal AI

State-of-the-art artificial intelligence (AI) and Machine-Learning (ML) tools have not yet enabled rapid design of next-generation materials. Detailed physical understanding of how material properties affect device performance is required to advance materials development. For example, optimization of ink parameters for electrocatalysts has no known physical mathematical model and thus insights are difficult to translate from material studies to device studies. Herein, we demonstrate how to use the emerging AI tool, physics-based structural causal models (SCMs), to extract quantitative causative insights from complex heterogeneous electrochemical systems with small (n < 10), but multi-modal datasets (modes > 10). Our SCM quantitatively separates the role that varying the support-to-catalyst ratios and total material loadings plays on catalytic performance. The proof of concept model developed in this work enables root-cause-analysis on the cyclic voltammograms of manganese-antimony oxide oxygen reduction electrocatalysts on Vulcan carbon supports tested in alkaline media using a rotating disc electrode device configuration. Our preliminary causal analyses quantitatively disentangle how the catalyst performance is affected by the number of active sites versus the thickness of the electrode. To the best of our knowledge, this is the first demonstration of physics-based SCMs applied to electrochemical materials and their performance.

cond-mat.mtrl-sci↗

arXiv2025

XxaCT-NN: Structure Agnostic Multimodal Learning for Materials Science

Recent advances in materials discovery have been driven by structure-based models, particularly those using crystal graphs. While effective for computational datasets, these models are impractical for real-world applications where atomic structures are often unknown or difficult to obtain. We propose a scalable multimodal framework that learns directly from elemental composition and X-ray diffraction (XRD) -- two of the more available modalities in experimental workflows without requiring crystal structure input. Our architecture integrates modality-specific encoders with a cross-attention fusion module and is trained on the 5-million-sample Alexandria dataset. We present masked XRD modeling (MXM), and apply MXM and contrastive alignment as self-supervised pretraining strategies. Pretraining yields faster convergence (up to 4.2x speedup) and improves both accuracy and representation quality. We further demonstrate that multimodal performance scales more favorably with dataset size than unimodal baselines, with gains compounding at larger data regimes. Our results establish a path toward structure-free, experimentally grounded foundation models for materials science.

cs.LG↗

arXiv2024

UniMat: Unifying Materials Embeddings through Multi-modal Learning

Materials science datasets are inherently heterogeneous and are available in different modalities such as characterization spectra, atomic structures, microscopic images, and text-based synthesis conditions. The advancements in multi-modal learning, particularly in vision and language models, have opened new avenues for integrating data in different forms. In this work, we evaluate common techniques in multi-modal learning (alignment and fusion) in unifying some of the most important modalities in materials science: atomic structure, X-ray diffraction patterns (XRD), and composition. We show that structure graph modality can be enhanced by aligning with XRD patterns. Additionally, we show that aligning and fusing more experimentally accessible data formats, such as XRD patterns and compositions, can create more robust joint embeddings than individual modalities across various tasks. This lays the groundwork for future studies aiming to exploit the full potential of multi-modal data in materials science, facilitating more informed decision-making in materials design and discovery.

cs.LG↗

arXiv2023

The Role of Reference Points in Machine-Learned Atomistic Simulation Models

This paper introduces the Chemical Environment Modeling Theory (CEMT), a novel, generalized framework designed to overcome the limitations inherent in traditional atom-centered Machine Learning Force Field (MLFF) models, widely used in atomistic simulations of chemical systems. CEMT demonstrated enhanced flexibility and adaptability by allowing reference points to exist anywhere within the modeled domain and thus, enabling the study of various model architectures. Utilizing Gaussian Multipole (GMP) featurization functions, several models with different reference point sets, including finite difference grid-centered and bond-centered models, were tested to analyze the variance in capabilities intrinsic to models built on distinct reference points. The results underscore the potential of non-atom-centered reference points in force training, revealing variations in prediction accuracy, inference speed and learning efficiency. Finally, a unique connection between CEMT and real-space orbital-free finite element Density Functional Theory (FE-DFT) is established, and the implications include the enhancement of data efficiency and robustness. It allows the leveraging of spatially-resolved energy densities and charge densities from FE-DFT calculations, as well as serving as a pivotal step towards integrating known quantum-mechanical laws into the architecture of ML models.

physics.chem-ph↗

arXiv2019

Network analysis of synthesizable materials discovery

Assessing the synthesizability of inorganic materials is a grand challenge for accelerating their discovery using computations. Synthesis of a material is a complex process that depends not only on its thermodynamic stability with respect to others, but also on factors from kinetics, to advances in synthesis techniques, to the availability of precursors. This complexity makes the development of a general theory or first-principles approach to synthesizability currently impractical. Here we show how an alternative pathway to predicting synthesizability emerges from the dynamics of the materials stability network: a scale-free network constructed by combining the convex free-energy surface of inorganic materials computed by high-throughput density functional theory and their experimental discovery timelines extracted from citations. The time-evolution of the underlying network properties allows us to use machine-learning to predict the likelihood that hypothetical, computer-generated materials will be amenable to successful experimental synthesis.

cond-mat.mtrl-sci↗

arXiv2018

Mechanical control of crystal symmetry and superconductivity in Weyl semimetal MoTe$_2$

The non-centrosymmetric Weyl semimetal candidate, MoTe$_2$ was investigated through neutron diffraction and transport measurements at pressures up to 1.5 GPa and at temperatures down to 40 mK. Centrosymmetric and non-centrosymmetric structural phases were found to coexist in the superconducting state. Density Functional Theory (DFT) calculations reveal that the strength of the electron-phonon coupling is similar for both crystal structures. Furthermore, it was found that by controlling non-hydrostatic components of stress, it is possible to mechanically control the ground state crystal structure. This allows for the tuning of crystal symmetry in the superconducting phase from centrosymmetric to non-centrosymmetric. DFT calculations support this strain control of crystal structure. This mechanical control of crystal symmetry gives a route to tuning the band topology of MoTe$_2$ and possibly the topology of the superconducting state.

cond-mat.str-el↗

arXiv2017

First-principles study of magnetism, lattice dynamics, and superconductivity in LaFeSiH$_x$

The structural, electronic, magnetic, and vibrational properties of LaFeSiH$_x$ for $x$ between 0 and 1 are investigated using density functional calculations. We find that the electronic and magnetic properties are strongly controlled by the hydrogen concentration $x$ in LaFeSiH$_x$. While fully hydrogenated LaFeSiH has a striped antiferromagnetic ground state, the underdoped LaFeSiH$_x$ for $x\leq0.75$ is not magnetic within the virtual crystal approximation or with explicit doping of supercells. The antiferromagnetic configuration breaks the symmetry of Fe $d$ orbitals and increases electron-phonon coupling up to $50\%$, especially for modes in the 20-50 meV range that are associated with Fe atomic movement. We find competing nearest and next-nearest neighbor exchange interactions and significant spin-phonon coupling, qualitatively similar but smaller in magnitude compared those found in LaOFeAs superconductors. The superconducting $T_c$ for antiferromagnetic LaFeSiH$_x$, assuming conventional superconductivity via McMillan's equation, therefore is computed to be 2-10 K, in contrast to $T_c\approx0$ for the nonmagnetic material. We also predict that the LaFeSiH$_x$ could be a good proton conductor due to phase stability with a wide range of hydrogen concentration $x < 1$.

cond-mat.supr-con↗

arXiv2016

Interpretation of monoclinic hafnia valence electron energy loss spectra by TDDFT

We present the valence electron energy-loss spectrum and the dielectric function of monoclinic hafnia (m-HfO$_2$) obtained from time-dependent density-functional theory (TDDFT) predictions and compared to energy-filtered spectroscopic imaging measurements in a high-resolution transmission-electron microscope. Fermi's Golden Rule density-functional theory (DFT) calculations can capture the qualitative features of the energy-loss spectrum, but we find that TDDFT, which accounts for local-field effects, provides nearly quantitative agreement with experiment. Using the DFT density of states and TDDFT dielectric functions, we characterize the excitations that result in the m-HfO$_2$ energy loss spectrum. The sole plasmon occurs between 13-16 eV, although the peaks $\sim$28 and above 40 eV are also due to collective excitations. We furthermore elaborate on the first-principles techniques used, their accuracy, and remaining discrepancies among spectra. More specifically, we assess the influence of Hf semicore electrons (5$p$ and 4$f$) on the energy-loss spectrum, and find that the inclusion of transitions from the 4$f$ band damps the energy-loss intensity in the region above 13 eV. We study the impact of many-body effects in a DFT framework using the adiabatic local-density approximation (ALDA) exchange-correlation kernel, as well as from a many-body perspective using a $GW$-derived electronic structure to account for self-energy corrections. These results demonstrate some cancellation of errors between self-energy and excitonic effects, even for excitations from the Hf $4f$ shell. We also simulate the dispersion with increasing momentum transfer for plasmon and collective excitation peaks.

cond-mat.mtrl-sci↗

arXiv2014

Evidence for anisotropic dielectric properties of monoclinic hafnia using high-resolution TEM valence electron energy-loss spectroscopy and ab initio time-dependent density-functional theory

The effect of nanocrystal orientation on the energy loss spectra of monoclinic hafnia (m-HfO$_2$) is measured by high resolution transmission electron microscopy (HRTEM) and valence energy loss spectroscopy (VEELS) on high quality samples. For the same momentum-transfer directions, the dielectric properties are also calculated ab initio by time-dependent density-functional theory (TDDFT). Experiments and simulations evidence anisotropy in the dielectric properties of m-HfO$_2$, most notably with the direction-dependent oscillator strength of the main bulk plasmon. The anisotropic nature of m-HfO$_2$ may contribute to the differences among VEELS spectra reported in literature. The good agreement between the complex dielectric permittivity extracted from VEELS with nanometer spatial resolution, TDDFT modeling, and past literature demonstrates that the present HRTEM-VEELS device-oriented methodology is a possible solution to the difficult nanocharacterization challenges given in the International Technology Roadmap for Semiconductors.

cond-mat.mtrl-sci↗