Searcharxiv⌕ Search

arXiv subjects

Ying Wai Li

Publications and source records attributed to Ying Wai Li.

At least 19 recordsLinked to original sources
arXiv2026

Machine learning topological defect formation: When are the defects made?

Topological defects that form in a nonequilibrium second-order phase transition are presumably seeded by fluctuations of the order parameter in the vicinity of the critical point. Motivated by this conjecture that underlies the Kibble-Zurek mechanism (KZM), we investigate whether machine learning (ML) can anticipate their locations from the fluctuations in the ``impulse regime'', the time interval when the evolution of the order parameter cannot keep up with the conditions imposed by the quench. Going beyond the conventional KZM focus on defect density, we show that a recurrent neural network can predict locations of topological defects from short-time dynamical data deep within the impulse regime. We thus demonstrate that defects are sown in the immediate vicinity of the critical point. The seeds of defects, fluctuations imprinted on the evolving order parameter, are exponentially small near the critical point, but become amplified by the evolution and play a dominant role in breaking symmetry. This is before the freezeout time that concludes the impulse regime, and well before the order parameter assumes its final symmetry-broken configuration. Furthermore, we find that the predictive power of the ML also exhibits power-law scaling consistent with KZM.

cond-mat.stat-mech↗
arXiv2026

Discovery of novel magnetic Y-Mn-B compounds via advanced machine learning guided framework

Rare-earth transition-metal borides offer critical structural motifs for permanent-magnet design; however, the manganese-rich regions within these compositional phase spaces remain largely unexplored. In this work, we develop an advanced machine-learning-assisted discovery framework to explore Y-Mn-B ternary system. Starting from over one million hypothetical structures generated from known structures in databases, we filtered promising candidates by first applying graph neural networks to predict material stability, then using machine-learning-interatomic-potential to relax their structures, and finally validating the results with first-principles calculations. We identify 5 stable and near-stable Y-Mn-B phases along with 61 metastable compounds with the formation energy within 100 meV/atom with respect to the ternary convex hull. Among them, Y2Mn7B7 and YMn4B4 are structurally analogous to the previously synthesized $R_{1+ε}Fe_4B_4$ 1D incommensurate composite chain compounds. In striking contrast to the strongly suppressed Fe moments reported, our first-principles calculations reveal that the predicted Mn-chain phases preserve sizable local Mn moments (approximately 1.1 $μ_B$) and favored ferromagnetic ordering. Electronic structure analyses elucidate the microscopic origin of moment recovery via an enhanced exchange splitting driven by a Stoner-like instability. We also perform systematic Mn-Fe substitution to confirm the thermodynamic continuity and a monotonic enhancement of the macroscopic magnetization, from Fe to Mn. These findings indicate that targeted transition-metal substitution within a one-dimensional boride family can recover transition-metal magnetism, offering a physically interpretable route for designing new magnetic rare-earth transition-metal borides.

cond-mat.mtrl-sci↗
arXiv2026

exa-PD: A scalable high-performance workflow for multi-element phase diagram construction

Exa-PD is a highly parallelizable workflow designed for the construction of multi-element phase diagrams (PDs). It uses standard sampling techniques, molecular dynamics (MD) and Monte Carlo (MC) as implemented in the LAMMPS package, to simultaneously sample multiple phases over a fine temperature-composition mesh for free-energy calculations. Parsl serves as the global workflow engine, coordinating large ensembles of MD and MC tasks to achieve massive parallelization with strong scalability. The resulting free energies of liquid and solid phases are then fed to CALPHAD modeling via the PyCalphad package to construct multi-element PDs.

cond-mat.mtrl-sci↗
arXiv2026

Structural modulation, physical properties, and electronic band structure of the kagome metal UCr$_6$Ge$_6$

The chemical flexibility of the $RM_6X_6$ stoichiometry, where an $f$-block element is intercalated in the CoSn structure type, allows for the tuning of flatbands associated with kagome lattices to the Fermi level and for emergent phenomena due to interactions between the $f$- and $d$-electron lattices. Yet, 5$f$ members of the ``166" compounds are underrepresented compared with 4$f$ members. Here, we report single-crystal growth of UCr$_6$Ge$_6$, which crystallizes in a monoclinically distorted Y$_{0.5}$Co$_3$Ge$_3$-type structure. The real-space character of the modulation, which is unique within the $RM_6X_6$ family, is approximated by a 3$\times$1$\times$2 supercell of the average monoclinic cell. The compound has kagome-lattice flatbands near the Fermi level and a moderately enhanced electronic heat capacity, as evidenced by its low-temperature Sommerfeld coefficient ($γ=86.5$~mJ~mol$^{-1}$~K$^{-2}$) paired with band structure calculations. The small, isotropic magnetization and featureless resistivity of UCr$_6$Ge$_6$ suggest itinerant uranium 5$f$ electrons and Pauli paramagnetism. Angle-resolved photoemission spectroscopy results provide evidence for uranium 5$f$ weight at the Fermi level and for a flatband near the Fermi level associated with the chromium $3d$ kagome lattice. The isotropic magnetic behavior of the uranium 5$f$ electrons starkly contrasts with localized behavior in other uranium 166 compounds, highlighting the high tunability of the magnetic ground state across the material family.

cond-mat.str-el↗
arXiv2026

exaPD: A highly parallelizable workflow for multi-element phase diagram (PD) construction

Phase diagrams (PDs) illustrate the relative stability of competing phases under varying conditions, serving as critical tools for synthesizing complex materials. Reliable phase diagrams rely on precise free energy calculations, which are computationally intensive. We introduce exaPD, a user-friendly workflow that enables simultaneous sampling of multiple phases across a fine mesh of temperature and composition for free energy calculations. The package employs standard molecular dynamics (MD) and Monte Carlo (MC) sampling techniques, as implemented in the LAMMPS package. Various interatomic potentials are supported, including the neural network potentials with near {\it ab initio} accuracy. A global controller, built with Parsl, manages the MD/MC jobs to achieve massive parallelization with near ideal scalability. The resulting free energies of both liquid and solid phases, including solid solutions, are integrated into CALPHAD modeling using the PYCALPHAD package for constructing the phase diagram.

cond-mat.mtrl-sci↗
arXiv2026

Accelerating Multicanonical Sampling with Irreversibility

Flat-histogram Monte Carlo simulations are well-established, robust methods to perform random walks in a physical observable or parameter space, making them suitable for finding ground states or studying phase transitions in complex systems in statistical physics. However, their efficiency can be limited by the time to attain the desired flat distribution, which is generally unknown prior to the simulations. In particular, they might suffer from slowing down towards the end of a simulation due to the diffusive nature of random walks. In this work we apply irreversibility to the multicanonical Monte Carlo method via the lifting approach to alleviate this behavior. We achieve a 2-4 times speedup in ground-state search for a two-dimensional (2D) Ising model, and up to an order of magnitude of speedup for finding the ground-state energy in an Edwards-Anderson spin glass, compared to traditional multicanonical sampling. The round-trip times between ground states show a narrower distribution and are significantly shorter compared to the reversible counterpart, suggesting that a lower convergence time with a smaller time variance is feasible.

cond-mat.stat-mech↗
arXiv2025

exa-AMD: An Exascale-Ready Framework for Accelerating the Discovery and Design of Functional Materials

We present exa-AMD, an open-source, high-performance framework designed for accelerated materials discovery on modern supercomputers. exa-AMD overcomes key computational bottlenecks in large-scale structure prediction through task-based parallelization, adaptive load balancing, and optimized data management for CPU and GPU architectures. The framework automates the end-to-end workflow, from generating candidate structures to evaluating formation energies and updating phase diagrams. Its modular design allows users to easily replace or extend components with custom machine learning models, alternative initial structure templates, and future structure generators, enabling flexible integration with emerging AI approaches. We demonstrate strong scaling across high-performance computing platforms and highlight applications to Na-B-C, Ce-Co-B, and Fe-Co-Zr systems, establishing exa-AMD as a robust and exascale-ready tool for accelerating the discovery and design of functional materials. exa-AMD is publicly available on GitHub, with detailed documentation and reproducible test cases to support community engagement and collaborative research.

cond-mat.mtrl-sci↗
arXiv2025

Accelerated discovery and design of Fe-Co-Zr magnets with tunable magnetic anisotropy through machine learning and parallel computing

Rare earth (RE)-free permanent magnets, as alternative substitutes for RE-containing magnets for sustainable energy technologies and modern electronics, have attracted considerable interest. We performed a comprehensive search for new hard magnetic materials in the ternary Fe-Co-Zr space by leveraging a scalable, machine learning-assisted materials discovery framework running on GPU-enabled exascale computing resources. This framework integrates crystal graph convolutional neural network (CGCNN) machine learning (ML) method with first-principles calculations to efficiently navigate the vast composition-structure space. The efficiency and accuracy of the ML approach enable us to reveal 9 new thermodynamically stable ternary Fe-Co-Zr compounds and 81 promising low-energy metastable phases with their formation energies within 0.1 eV/atom above the convex hull. The predicted compounds span a wide range of crystal symmetries and magnetic behaviors, providing a rich platform for tuning functional properties. Based on the analysis of site-specific magnetic properties, we show that the Fe6Co17Zr6 compound obtained from our ML discovery can be further optimized by chemical doping. Chemical substitutions lead to a ternary Fe5Co18Zr6 phase with a strong anisotropy of K1 = 1.1 MJ/m3, and a stable quaternary magnetic Fe5Co16Zr6Mn4 compound.

cond-mat.mtrl-sci↗
arXiv2025

exa-AMD: A Scalable Workflow for Accelerating AI-Assisted Materials Discovery and Design

exa-AMD is a Python-based application designed to accelerate the discovery and design of functional materials by integrating AI/ML tools, materials databases, and quantum mechanical calculations into scalable, high-performance workflows. The execution model of exa-AMD relies on Parsl, a task-parallel programming library that enables a flexible execution of tasks on any computing resource from laptops to supercomputers. By using Parsl, exa-AMD is able to decouple the workflow logic from execution configuration, thereby empowering researchers to scale their workflows without having to reimplement them for each system.

cs.DC↗
arXiv2025

Teacher-student training improves accuracy and efficiency of machine learning interatomic potentials

Machine learning interatomic potentials (MLIPs) are revolutionizing the field of molecular dynamics (MD) simulations. Recent MLIPs have tended towards more complex architectures trained on larger datasets. The resulting increase in computational and memory costs may prohibit the application of these MLIPs to perform large-scale MD simulations. Here, we present a teacher-student training framework in which the latent knowledge from the teacher (atomic energies) is used to augment the students' training. We show that the light-weight student MLIPs have faster MD speeds at a fraction of the memory footprint compared to the teacher models. Remarkably, the student models can even surpass the accuracy of the teachers, even though both are trained on the same quantum chemistry dataset. Our work highlights a practical method for MLIPs to reduce the resources required for large-scale MD simulations.

physics.chem-ph↗
arXiv2025

Ensemble Knowledge Distillation for Machine Learning Interatomic Potentials

The quality of machine learning interatomic potentials (MLIPs) strongly depends on the quantity of training data as well as the quantum chemistry (QC) level of theory used. Datasets generated with high-fidelity QC methods are typically restricted to small molecules and may be missing energy gradients, which make it difficult to train accurate MLIPs. We present an ensemble knowledge distillation (EKD) method to improve MLIP accuracy when trained to energy-only datasets. First, multiple teacher models are trained to QC energies and then generate atomic forces for all configurations in the dataset. Next, the student MLIP is trained to both QC energies and to ensemble-averaged forces generated by the teacher models. We apply this workflow on the ANI-1ccx dataset where the configuration energies computed at the coupled cluster level of theory. The resulting student MLIPs achieve new state-of-the-art accuracy on the COMP6 benchmark and show improved stability for molecular dynamics simulations.

physics.chem-ph↗
arXiv2025

Sunny.jl: A Julia Package for Spin Dynamics

Sunny is a Julia package designed to serve the needs of the quantum magnetism community. It supports the specification of a very broad class of spin models and a diverse suite of numerical solvers. These include powerful methods for simulating spin dynamics both in and out of equilibrium. Uniquely, it features a broad generalization of classical and semiclassical approaches to SU(N) coherent states, which is useful for studying systems exhibiting strong spin-orbit coupling or local entanglement effects. Sunny also offers a well-developed framework for calculating the dynamical spin structure factor, enabling direct comparison with scattering experiments. Ease of use is a priority, with tools for symmetry-guided modeling and interactive visualization.

quant-ph↗
arXiv2024

Model-diff: A Tool for Comparative Study of Language Models in the Input Space

Comparing two (large) language models (LMs) side-by-side and pinpointing their prediction similarities and differences on the same set of inputs are crucial in many real-world scenarios, e.g., one can test if a licensed model was potentially plagiarized by another. Traditional analysis compares the LMs' outputs on some benchmark datasets, which only cover a limited number of inputs of designed perspectives for the intended applications. The benchmark datasets cannot prepare data to cover the test cases from unforeseen perspectives which can help us understand differences between models unbiasedly. In this paper, we propose a new model comparative analysis setting that considers a large input space where brute-force enumeration would be infeasible. The input space can be simply defined as all token sequences that a LM would produce low perplexity on -- we follow this definition in the paper as it would produce the most human-understandable inputs. We propose a novel framework \our that uses text generation by sampling and deweights the histogram of sampling statistics to estimate prediction differences between two LMs in this input space efficiently and unbiasedly. Our method achieves this by drawing and counting the inputs at each prediction difference value in negative log-likelihood. Experiments reveal for the first time the quantitative prediction differences between LMs in a large input space, potentially facilitating the model analysis for applications such as model plagiarism.

cs.CL↗
arXiv2024

High-Throughput Search and Prediction of Layered 4f-Materials

The development of multifunctional devices calls for the discovery of new layered materials with novel electronic properties. f-electron systems naturally host a rich set of competing and intertwining phases owning to the presence of strong spin-orbit coupling, electron-electron interactions, and hybridization between itinerant and local electrons. However, very little attention has been devoted to exploring the f-electron family of compounds for new promising layered material candidates. Here, we identify 295 rare earth compounds from across the lanthanide series of elements that exhibit a spectrum of lattice symmetries and electronic properties. In particular, we find metallic compounds and insulating systems with band gaps covering a 0.1 eV to 5.3 eV range which opens new possibilities in infrared quantum sensors, designer photocatalysts, and tunable transistors. The inclusion of 4f-states in a layered system also suggests the possibility of 2D confined heavy-fermion superconductivity and topological semimetals. Our study serves as a springboard to further systematic theoretical investigation of correlation-driven properties of the 4f and other 2D materials composed of heavy elements.

cond-mat.mtrl-sci↗
arXiv2024

Evaluation of human-model prediction difference on the Internet Scale of Data

Evaluating models on datasets often fails to capture their behavior when faced with unexpected and diverse types of inputs. It would be beneficial if we could evaluate the difference between human annotation and model prediction for an internet number of inputs, or more generally, for an input space that enumeration is computationally impractical. Traditional model evaluation methods rely on precision and recall (PR) as metrics, which are typically estimated by comparing human annotations with model predictions on a specific dataset. This is feasible because enumerating thousands of test inputs is manageable. However, estimating PR across a large input space is challenging because enumeration becomes computationally infeasible. We propose OmniInput, a novel approach to evaluate and compare NNs by the PR of an input space. OmniInput is distinctive from previous works as its estimated PR reflects the estimation of the differences between human annotation and model prediction in the input space which is usually too huge to be enumerated. We empirically validate our method within an enumerable input space, and our experiments demonstrate that OmniInput can effectively estimate and compare precision and recall for (large) language models within a broad input space that is not enumerable.

cs.LG↗
arXiv2024

Best of Both Worlds: Enforcing Detailed Balance in Machine Learning Models of Transition Rates

The slow microstructural evolution of materials often plays a key role in determining material properties. When the unit steps of the evolution process are slow, direct simulation approaches such as molecular dynamics become prohibitive and Kinetic Monte-Carlo (kMC) algorithms, where the state-to-state evolution of the system is represented in terms of a continuous-time Markov chain, are instead frequently relied upon to efficiently predict long-time evolution. The accuracy of kMC simulations however relies on the complete and accurate knowledge of reaction pathways and corresponding kinetics. This requirement becomes extremely stringent in complex systems such as concentrated alloys where the astronomical number of local atomic configurations makes the a priori tabulation of all possible transitions impractical. Machine learning models of transition kinetics have been used to mitigate this problem by enabling the efficient on-the-fly prediction of kinetic parameters. In this study, we show how physics-informed ML architectures can exactly enforce the detailed balance condition, by construction. Using the diffusion of a vacancy in a concentrated alloy as an example, we show that such ML architectures also exhibit superior performance in terms of prediction accuracy, demonstrating that the imposition of physical constraints can facilitate the accurate learning of barriers at no increase in computational cost.

cond-mat.mtrl-sci↗
arXiv2023

Machine learning potentials with Iterative Boltzmann Inversion: training to experiment

Methodologies for training machine learning potentials (MLPs) to quantum-mechanical simulation data have recently seen tremendous progress. Experimental data has a very different character than simulated data, and most MLP training procedures cannot be easily adapted to incorporate both types of data into the training process. We investigate a training procedure based on Iterative Boltzmann Inversion that produces a pair potential correction to an existing MLP, using equilibrium radial distribution function data. By applying these corrections to a MLP for pure aluminum based on Density Functional Theory, we observe that the resulting model largely addresses previous overstructuring in the melt phase. Interestingly, the corrected MLP also exhibits improved performance in predicting experimental diffusion constants, which are not included in the training procedure. The presented method does not require auto-differentiating through a molecular dynamics solver, and does not make assumptions about the MLP architecture. The results suggest a practical framework of incorporating experimental data into machine learning models to improve accuracy of molecular dynamics simulations.

physics.app-ph↗
arXiv2023

Reduction of Autocorrelation Times in Lattice Path Integral Quantum Monte Carlo via Direct Sampling of the Truncated Exponential Distribution

In Monte Carlo simulations, proposed configurations are accepted or rejected according to an acceptance ratio, which depends on an underlying probability distribution and an a priori sampling probability. By carefully selecting the probability distribution from which random variates are sampled, simulations can be made more efficient, by virtue of an autocorrelation time reduction. In this paper, we illustrate how to directly sample random variates from a two dimensional truncated exponential distribution. We show that our direct sampling approach converges faster to the target distribution compared to rejection sampling. The direct sampling of one and two dimensional truncated exponential distributions is then applied to a recent Path Integral Monte Carlo (PIMC) algorithm for the simulation of Bose-Hubbard lattice models at zero temperature. The new sampling method results in improved acceptance ratios and reduced autocorrelation times of estimators, providing an effective speed up of the simulation.

physics.comp-ph↗