Conference Tools for 2025 TMS Annual Meeting & Exhibition

Login

Register as a New User

Help

Submit An Abstract

Propose A Symposium

Presenter/Author Tools

Organizer/Editor Tools

About this Abstract
Meeting	2025 TMS Annual Meeting & Exhibition
Symposium	Bridging Scale Gaps in Multiscale Materials Modeling in the Age of Artificial Intelligence
Presentation Title	UF3: Fast and Interpretable MLIP for High-Performance Molecular Dynamics
Author(s)	Ajinkya C. Hire, Hendrik Krass, Stephen Xie, Jason Gibson, Michael MacIsaac, Sung Hoon Jung, Matthias Rupp, Richard Hennig
On-Site Speaker (Planned)	Ajinkya C. Hire
Abstract Scope	Ultra-fast forcefield (UF3) is a fast and interpretable machine-learned interatomic potential (MLIP) that expresses many-body energy as a function of effective two- and three-body potential. In this talk, I will present the LAMMPS implementation of UF3 MLIP for both CPUs and GPUs. Our benchmark simulations show that UF3 outperforms state-of-the-art MLIPs in terms of molecular dynamics speed for potentials of similar accuracy. UF3 also scales efficiently with respect to the number of atoms in the simulation on GPUs. I will also outline our UltraFast workflow of fitting UF3 potentials to genetic algorithm-generated ab-initio data using the Nb-Sn system as an example, highlighting the data efficiency of UF3. The UltraFast workflow allows the development of an MLIP in less than two minutes on a single core machine from a dataset consisting of 4300 structures and 80000 forces. This short featurization and training times enable the rapid optimization of model hyperparameters.
Proceedings Inclusion?	Planned:
Keywords	Computational Materials Science & Engineering,

OTHER PAPERS PLANNED FOR THIS SYMPOSIUM

A Machine Learning-Assisted Dislocation Density-Based Crystal Plasticity Model for FCC Aluminum

AI-Enabled Upscaling of Ab Initio Thermodynamics for 3C-SiC(100) Surface Reconstructions

An Ultra-Fast Machine-Learning Potentials to Investigate the Phonon-Dislocation Interaction of Lead Selenide

AtomAgents: Alloy Design and Discovery Through Physics-Aware Multi-Modal Multi-Agent Artificial Intelligence

Atomistically-Informed Discrete Dislocation Dynamics Simulations of Shock in Aluminum

Atomistically Informed Mesoscale Modeling of Deformation Behavior of Bulk Metallic Glasses

Bridging Scales in Metal Plasticity: The Roles of Theory, Data Science, and Computing

Coarse-Graining Atomistic Simulation Data with Physics-Guided Gaussian Process Regression

Complex Structure of Liquid and Machine-Learning

Computational Studies on Statistical Features of Dislocation Glide Energetics in Refractory Complex Concentrated Alloys

Developing Data-Driven Dislocation Mobility Laws for BCC Metals

Developing On-Demand, Highly Efficient Digital Twins with DFT Accuracy for Iterative Alloy Discovery Frameworks

Discovering New Mechanisms of Grain Growth with a Machine Learning Model Trained on Experimental and Simulation Data

Efficient High-Throughput Ab Initio Prediction of Liquidus Curves

Engineering the Crack-Tip Material Composition to Enhance the Microplasticity in Refractory Complex Concentrated Alloys

First-Principles Models of Solute-Defect Interactions in Alloys

Influence of Surface Structure on Graphene Formation via Thermal Decomposition of Silicon Carbide

Integrating AI for High-Dimensional Saddle Point Sampling

Interplay between Hydrogen and Screw Dislocation in bcc-Fe: a Neural-network Potential Study

Machine Learning-Enhanced Multiscale Modeling of Solidification

Machine Learning - Kinetic Monte Carlo Investigation on Sluggish Interstitial Diffusion in Fe-Ni-Cr-Cu-Co High Entropy Alloys

Machine Learning for the Efficient Identification of High-Performance Metal-Doped Transition Metal Compounds for Hydrogen Evolution Catalysis

Machine Learning Potentials for Chemically Complex Alloys

Material-Agnostic Training Data Generation for Machine-Learning Interatomic Potentials

Mechanism-Based Data-Driven Exploration of Complex Concentrated Alloys with Enhanced Mechanical Performance

Mesoscale Investigation of Dislocation-Grain Boundary Interactions in Metals and Alloys

Modelling Helium Bubble Evolution and Grain Decohesion in Nanostructured Tungsten Using ML-Based Interatomic Potential

Molecular Dynamic Studies of Strain Rate Effects on Screw Dislocation Mobility In BCC Metals

Multiscale Computation-Experiment Study of Advanced Materials with Characteristic Microstructure

Multiscale Computational Tools and AI Integration Using Chocolate as a Frugal Model System in Self-Driving Lab

Multiscale Modeling for Studying Corrosion-Induced Hydrogen Embrittlement in Zirconium

Neural Network Kinetics: Exploring Diffusion Multiplicity and Chemical Ordering in Compositionally Complex Materials

Pathways to the 7 � 7 Surface Reconstruction of Si(111) Revealed by Machine-Learning Molecular Dynamics Simulations

Peierls-Nabarro Modeling of Dislocations in High Entropy Alloys

Quantifying Chemical Short-Range Order in Metallic Alloys

Realizing High-Throughput Multi-Scale Simulations of Materials Through Machine Learning

Rethinking Materials Simulations; Blending Direct Numerical Simulations with Machine-Learning Strategies

Revealing the Impact of Hydrogen on Iron: Large-Scale Quantitative Atomistic Analysis with Highly Accurate and Transferrable Machine Learning Interatomic Potentials

Simulation-Informed Models for Amorphous Metal Mechanical Property Prediction

Study of Xe Binding in Ag-Exchange Chabazite for Radio-Nuclide Absorption

Surrogate Models in First-Principles Statistical Mechanics Methods

The Connection Between Atomistic Defect Clusters and Geometrically Necessary Dislocations in Irradiated Nanocrystals

UF3: Fast and Interpretable MLIP for High-Performance Molecular Dynamics

Understanding Microstructural Evolution Using Graph Attention Networks

Questions about ProgramMaster? Contact programming@programmaster.org