arXiv subjects

Hai Xiao

Publications and source records attributed to Hai Xiao.

12 recordsLinked to original sources

arXiv2026

Hessian-informed machine learning interatomic potential towards bridging theory and experiments

Local curvature of potential energy surfaces is critical for predicting certain experimental observables of molecules and materials from first principles, yet it remains far beyond reach for complex systems. In this work, we introduce a Hessian-informed Machine Learning Interatomic Potential (Hi-MLIP) that captures such curvature reliably, thereby enabling accurate analysis of associated thermodynamic and kinetic phenomena. To make Hessian supervision practically viable, we develop a highly efficient training protocol, termed Hessian INformed Training (HINT), achieving two to four orders of magnitude reduction for the requirement of expensive Hessian labels. HINT integrates critical techniques, including Hessian pre-training, configuration sampling, curriculum learning and stochastic projection Hessian loss. Enabled by HINT, Hi-MLIP significantly improves transition-state search and brings Gibbs free-energy predictions close to chemical accuracy especially in data-scarce regimes. Our framework also enables accurate treatment of strongly anharmonic hydrides, reproducing phonon renormalization and superconducting critical temperatures in close agreement with experiment while bypassing the computational bottleneck of anharmonic calculations. These results establish a practical route to enhancing curvature awareness of machine learning interatomic potentials, bridging simulation and experimental observables across a wide range of systems.

cs.LG↗

arXiv2025

In-Situ Growth of Halide Perovskite Single Crystals and Thin Films on Optical Fiber End Facets

Halide perovskites exhibit significant advantages for active optical components such as light emitting diodes, solar cells and photodetectors due to their excellent optoelectronic properties. Their nonlinear optical effects and other characteristics also make them suitable for integration into waveguide components, such as optical fibers, for applications like optical modulation. Although some efforts have been made to integrate perovskite nanomaterials with optical fibers, technological challenges have hindered reliable in-situ preparation methods. Herein, we propose an area-selective wetting strategy for optical fibers, which utilizes hydrophobic sidewalls and hydrophilic end facets to reliably hold small precursor droplets. By introducing a space confinement strategy to suppress the kinetics of solvent evaporation, Methylammonium lead bromide (MAPbBr3) perovskite single crystals were successfully grown in-situ on the fiber end facet. The versatility of this in-situ growth method for single crystals on fiber end facets of various sizes has also been verified. In a separate approach, the controllable in-situ preparation of CsPbBr3 polycrystalline thin films was achieved through vacuum-assisted rapid crystallization. Our strategy provides a controllable platform for the integration of perovskite materials and optical fibers, enabling further development in optical applications.

physics.optics↗

arXiv2025

Oxidation States in Solids from Data-Driven Paradigms

The oxidation state (OS) is an essential chemical concept that embodies chemical intuition but cannot be computed with well-defined physical laws. We establish a data-driven paradigm, with its implementation as Tsinghua Oxidation States in Solids (TOSS), to explicitly compute the OSs in crystal structures as the emergent properties from large-sized datasets based on Bayesian maximum a posteriori probability (MAP). TOSS employs two looping structures over the large-sized dataset of crystal structures to obtain an emergent library of distance distributions as the foundation for chemically intuitive understanding and then determine the OSs by minimizing a loss function for each structure based on MAP and distance distributions in the whole dataset. The application of TOSS to a dataset of $\mathrm{>}$1,000,000 crystal structures delivers a superior success rate, and using the resulting OSs as the dataset, we further train a data-driven alternative to TOSS based on graph convolutional networks. We expect TOSS and the ML-model-based alternative to find a wide spectrum of applications, and this work also demonstrates an encouraging example for the data-driven paradigms to explicitly compute the chemical intuition for tackling complex problems in chemistry.

physics.chem-ph↗

arXiv2025

AlphaNet: Scaling Up Local-frame-based Atomistic Interatomic Potential

Molecular dynamics simulations demand an unprecedented combination of accuracy and scalability to tackle grand challenges in catalysis and materials design. To bridge this gap, we present AlphaNet, a local-frame-based equivariant model that simultaneously improves computational efficiency and predictive precision for interatomic interactions. By constructing equivariant local frames with learnable geometric transitions, AlphaNet encodes atomic environments with enhanced representational capacity, achieving state-of-the-art accuracy in energy and force predictions. Extensive benchmarks on large-scale datasets spanning molecular reactions, crystal stability, and surface catalysis (Matbench Discovery and OC2M) demonstrate its superior performance over existing neural network interatomic potentials while ensuring scalability across diverse system sizes with varying types of interatomic interactions. The synergy of accuracy, efficiency, and transferability positions AlphaNet as a transformative tool for modeling multiscale phenomena, decoding dynamics in catalysis and functional interfaces, with direct implications for accelerating the discovery of complex molecular systems and functional materials.

cs.LG↗

arXiv2025

LOCAL: A Locality-based Active Learning Framework for Predicting the Stability of Dual-Atom Catalysts

Dual-atom catalysts supported on nitrogen-doped graphene (DAC/NG) are emerging as a family of promising catalysts that can overcome intrinsic limitations of single-atom catalysts. However, comprehensive assessment of their structural stability is prohibitively demanding due to a vast local configurational space. Here we introduce LOCAL, a locality-based framework that combines graph convolutional networks with active learning to efficiently predict DAC/NG stability by leveraging chemically intuitive locality quantified by crystal orbital Hamilton population analysis. We demonstrate the effectiveness of LOCAL over a comprehensive dataset of 611,648 DAC/NG structures, achieving a test mean absolute error of 0.15~eV while invoking density functional theory calculations for only 16,704 structures (2.7% of the dataset). Thus, LOCAL enables efficient and accurate construction of phase diagrams for DAC/NG across diverse compositions reciprocally validated with experimentally synthesized configurations for representative systems. Our framework composes an essential methodology for accelerating the discovery and optimization of high-performance complex catalysts.

physics.chem-ph↗

arXiv2024

Ensemble Learning of Machine Learning Force Fields

Machine learning force fields (MLFFs) are a promising approach to balance the accuracy of quantum mechanics with the efficiency of classical potentials, yet selecting an optimal model amid increasingly diverse architectures that delivers reliable force predictions and stable simulations remains a core pratical challenge. Here we introduce EL-MLFFs, an ensemble learning framework that uses a stacking methodology to integrate predictions from diverse base MLFFs. Our approach constructs a graph representation where a graph neural network (GNN) acts as a meta-model to refine the initial force predictions. We present two meta-model architectures: a computationally efficient direct fitting model and a physically-principled conservative model that ensures energy conservation. The framework is evaluated on a diverse range of systems, including single molecules (methane), surface chemistry (methanol/Cu(100)), molecular dynamics benchmarks (MD17), and the MatPES materials dataset. Results show that EL-MLFFs improves predictive accuracy across these domains. For molecular systems, it reduces force errors and improves the simulation stability compared to base models. For materials, the method yields lower formation energy errors on the WBM test set. The EL- MLFFs framework offers a systematic approach to address challenges of model selection and the accuracy-stability trade-off in molecular and materials simulations.

cs.LG↗

arXiv2021

Damage Detection in a laboratory-scale wellbore applying Time Reverse Nonlinear Elastic Wave Spectroscopy (TR NEWS)

Time Reverse Nonlinear Elastic Wave Spectroscopy (TR-NEWS) has been used to focus acoustic energy, and make measurements correlated with damage in a variety of industrial materials. Most studies using TR-NEWS in Earth Science have focused on relatively small objects and may have multiple acoustic sources. In Earth, within energy extraction settings, the structure and scale of wellbores makes acoustic focusing challenging. This paper explores the feasibility of applying TR NEWS for damage detection in wellbores by constructing a laboratory-scale wellbore, and using TR to focus and make dynamic linear and nonlinear elastic measurements. After successive cycles of induced, localized mechanical damage the sample, the hysteretic nonlinear elastic parameter α, increased with damage cycle indicating progressive mechanical damage. In addition to these strain-dependent changes, TR peak width and changes to peak amplitude near the damage sites was also observed. To deploy TR in a wellbore, it will be necessary to choose sensors that are suitable for the environment, and that can be distributed along the wellbore. Thus, this paper demonstrates that acoustic TR can be conducted using both an intrinsic Fabry Perot interferometer fiber optic strain sensor, and an intrinsic Michaelson interferometer fiber optic strain sensor, as a first step towards deployable sensing for TR in a wellbore.

physics.app-ph↗

arXiv2020

AI Online Filters to Real World Image Recognition

Deep artificial neural networks, trained with labeled data sets are widely used in numerous vision and robotics applications today. In terms of AI, these are called reflex models, referring to the fact that they do not self-evolve or actively adapt to environmental changes. As demand for intelligent robot control expands to many high level tasks, reinforcement learning and state based models play an increasingly important role. Herein, in computer vision and robotics domain, we study a novel approach to add reinforcement controls onto the image recognition reflex models to attain better overall performance, specifically to a wider environment range beyond what is expected of the task reflex models. Follow a common infrastructure with environment sensing and AI based modeling of self-adaptive agents, we implement multiple types of AI control agents. To the end, we provide comparative results of these agents with baseline, and an insightful analysis of their benefit to improve overall image recognition performance in real world.

cs.CV↗

arXiv2019

Online Learning to Estimate Warfarin Dose with Contextual Linear Bandits

Warfarin is one of the most commonly used oral blood anticoagulant agent in the world, the proper dose of Warfarin is difficult to establish not only because it is substantially variant among patients, but also adverse even severe consequences of taking an incorrect dose. Typical practice is to prescribe an initial dose, then doctor closely monitor patient response and adjust accordingly to the correct dosage. The three commonly used strategies for an initial dosage are the fixed-dose approach, the Warfarin Clinical algorithm, and the Pharmacogenetic algorithm developed by the IWPC (International Warfarin Pharmacogenetics Consortium). It is always best to prescribe correct initial dosage, motivated by this challenge, this work explores the performance of multi-armed bandit algorithms to best predict the correct dosage of Warfarin instead of trial-and-error procedure. Real data from the Pharmacogenetics and Pharmacogenomics Knowledge Base (PharmGKB) is used, with it a series of linear bandit algorithms and variants are developed and evaluated on Warfarin dataset. All proposed algorithms outperformed the fixed-dose baseline algorithm, and some even matched up the Warfarin Clinical Dosing Algorithm. In addition, a few promising future directions are given for further exploration and development.

cs.LG↗

arXiv2019

An Efficient Strategy to Generate Atom Connecting Positions

Atom connecting positions(ACPs) are positions where an atom is connecting to another one or a few atoms, which is needed when constructing final state used in chain-of-state(CoS) methods for transition state(TS) locating and minimum energy path(MEP) searching, especially with bond formation. However, ACPs are generated with chemical insight and experience, which is not only low efficient and time wasting, but the manually generated structure may be far from the optimized one. A efficient method is presented here for generating ACPs which is based on spherical optimization and VSEPR theory without manual interfering. Several examples are testified to prove the efficiency and robustness of the method.

physics.chem-ph↗

arXiv2019

Efficient Chain-of-States Approach for Locating Transition State via Spherical Optimization

The Chain-of-states(CoS) methods like nudge elastic band(NEB) method can be used to determine the minimum energy path (MEP) and transition state (TS) between two end local minima. However, the CoS methods are inefficient for difficult cases where the two ends are far apart with chemically insignificant part(s) in the MEP. We present here a method based on spherical optimization (SOPT), in which the SOPT method generates model end structures for CoS methods under the constraint of constant root-mean-square distance (RMSD) between two ends that is chosen to cover only the chemically significant part. We demonstrate the robustness and efficiency of our method with two examples, the CHOH dissociation channel and the first step of Aldol reaction. In both cases, the SOPT-based NEB calculations always reach the convergence to the correct MEPs with much less computational cost, whereas the regular NEB calculations fail under certain setups.

physics.chem-ph↗

arXiv2016

Two-dimensional Halide Perovskites: Tuning Electronic Activities of Defects

Two-dimensional (2D) halide perovskites are emerging as promising candidates for nano-electronics and optoelectronics. To realize their full potential, it is important to understand the role of those defects that can strongly impact material properties. In contrast to other popular 2D semiconductors (e.g. transition metal dichalcogenides MX2) for which defects typically induce harmful traps, we show that the electronic activities of defects in 2D perovskites are significantly tunable. For example, even with a fixed lattice orientation, one can change the synthesis conditions to convert a line defect (edge or grain boundary) from electron acceptor to inactive site without deep gap states. We show that this difference originates from the enhanced ionic bonding in these perovskites compared with MX2. The donors tend to have high formation energies, and the harmful defects are difficult to form at a low halide chemical potential. Thus we unveil unique properties of defects in 2D perovskites and suggest practical routes to improve them.

cond-mat.mtrl-sci↗