The OpenKIM project has been serving the atomistic simulation community since 2009. The project includes the largest repository of interatomic potentials on the web, an API and package manager for seamless download, installation, and use of these potentials in compatible simulation codes, and a distributed high-throughput computational pipeline that automatically checks each potential for coding integrity and evaluates its predictions of material properties. All content and calculation results are available on OpenKIM.

In this breakout session we’ll go through some hands-on Jupyter Notebooks outlining various features and capabilities of the NIST IPR tools atomman and iprPy. Use atomman to explore and fetch interatomic potentials and potential-specific relaxed crystals from the NIST IPR database. Use atomman to build, manipulate and analyze atomic configurations. Run LAMMPS simulations using atomman, allowing for methods/scripts that can be agnostic to potentials and crystals. Load an iprPy calculation, view its inputs, run the calculation and view results.

We will work with examples for treating different types of polymers, including star polymers and polymers composed out of pre-polymers. Users are advised to install EMC on their machine before coming to the breakout. EMC can be downloaded from https://montecarlo.sourceforge.net/. PC users should install Windows Subsystem for Linux with Ubuntu on their machine. WSL is installed by executing wsl.exe in a PowerShell. Ubuntu is installed after installing WSL by using the Software installation App.

Large language models (LLMs) are opening new possibilities for assisting with scientific simulations, but using them effectively requires careful prompting and validation, especially in domain-specific languages like LAMMPS. In this hands-on tutorial, participants will learn how to prompt large language models to generate LAMMPS input scripts and how to use a structure-aware static parser to catch common syntax and logic errors before running a simulation. We will explore typical failure modes, discuss prompt design strategies, and demonstrate a practical workflow for integrating AI tools into molecular dynamics projects.

Developing materials for fusion reactors is one of the leading challenges in developing fusion as a viable energy source. Plasma exposure and high temperatures at the plasma-material interface and neutron, helium, and tritium effects in structural materials will degrade the materials in the plasma-facing and structural components. Many of the processes that alter the microstructure of the material, leading to macroscopic damage and changes in material properties, occur at the atomistic scale, making molecular dynamics (MD) an essential tool in understanding the fundamental processes which drive radiation damage in these materials.

Machine-learning-based potentials (MLIPs) have taken the world of molecular dynamics by storm due to their dramatically improved accuracy compared to conventional empirical potentials. While near-quantum accuracy is (locally) in reach, MLIPs often show poor transferability to configurations that significantly diJer from their training data. This makes the curation of the training set critical but also challenging, often requiring multiple iterations, rigorous validation, and often human input. In this talk, I will discuss an automated approach to dataset generation based on an information-theoretic approach where the information entropy of the whole dataset, as measured in an ML feature space, is systematically maximized.

My talk will briefly review the early history of computational science, starting with the advances emanating from the US National Labs in the 1950’s. I will put special emphasis on the 1980’s and 1990’s; an important period for novel algorithms and methodologies that followed the democratization of high-performance computing.

One of the main challenges in understanding the cumulative effects of radiation damage in structural and functional materials lies in the ultrafast dynamics after an energetic collision with the lattice. The extremely fast timescale (ps) and large excess of energy (keV-MeV) imparted into the material during neutron and charge particle exposure are inadequately described by the assumptions of equilibrium dynamics. There is an urgent need for a modeling approach that can treat these conditions more realistically to predict these degradation properties.

Carbon nanoparticles (CNPs) are of tremendous interest for clean-energy technology due to the manifold of chemical, mechanical, electronic, and optoelectronic properties they can exhibit. However, practical application of these materials is hampered by current synthesis strategies. Low-pressure techniques such as chemical vapor deposition and flame pyrolysis are well understood but yield material with relatively low throughput. Conversely, high-pressure methods like ultrasound cavitation and detonation can greatly enhance throughput (e.g., enabling rates of up to kgs/μs), but the underlying phenomena are not well understood due to the highly dynamic nature of these processes, hampering tunability.

Machine-learning-based potentials (MLIPs) have taken the world of molecular dynamics by storm due to their dramatically improved accuracy compared to conventional empirical potentials. While near-quantum accuracy is (locally) in reach, MLIPs often show poor transferability to configurations that significantly diJer from their training data. This makes the curation of the training set critical but also challenging, often requiring multiple iterations, rigorous validation, and often human input. In this talk, I will discuss an automated approach to dataset generation based on an information-theoretic approach where the information entropy of the whole dataset, as measured in an ML feature space, is systematically maximized.

Reverse osmosis polyamide membranes are of significant importance due to their widespread applications in water purification and desalination processes. The critical properties of the membrane, such as thickness, chemical composition, crosslinking structure, and surface roughness, are sensitive to the conditions of the interfacial polymerization process, posing a remaining challenge for their precise prediction. In this work, extensive molecular dynamics simulations and experiments were performed to study the polymerization and crosslinking of trimesoyl chloride (TMC) and metaphenylene diamine (MPD) at the interface of water and organic solvent.

Machine learning interatomic potentials (MLIPs) have become essential for extending the reach of atomistic simulations for larger system and longer time scales, enabling the study of complex chemical processes with near ab initio accuracy. However, constructing reactive and transferable MLIPs remains challenging, due to the need for high quality training datasets that include rare-events, high energy intermediates and transition states. In this work we present SPARC (Smart Potential with Rare Event and Continuous Learning), a modular Python workflow package design to automate the construction of reactive MLIPs and to generate near accurate potential energy surface (PES) model.

High-speed communication uses the GHz band and requires low energy loss during transmission. To reduce energy loss at GHz frequencies, materials with a low dielectric constant (Dk) and low dielectric loss (Df) should be used in circuits. The Df value is influenced by dipolar orientation polarization arising from relaxational motion in polymers. Understanding the relationship between molecular structure and Df will aid in designing low-Df materials, thereby accelerating material discovery.

We report classical molecular dynamics predictions of the shock Hugoniot locus and aspects of post-shock stress and structure relaxation in amorphous, inert polydimethylsiloxane (PDMS), as functions of degree of polymerization (\(𝑛_{\text{DOP}}\)), cross- linking density (\(\rho_{xl}\)), and initial temperature (\(T_0\)). The results were obtained using a nonreactive united-atom force field comprising harmonic covalent bonds, angles, and dihedrals, and 12-6-1 interactions for non-bonded pairs. Starting from initially monodisperse systems with \(n_{\text{DOP}}\) = 32, 64, 128 or 256, spatial proximity-based cross linking was performed at a prescribed \(𝑇_0\) until the desired \(\rho_{xl}\) was achieved (\(0.

The solid-melt interfacial energy and its anisotropy are fundamental parameters that govern the final microstructure of cast metals. The magnitude of the interfacial energy dictates the nucleation frequency, while its crystalline anisotropy determines the growth direction of dendrites. While the effect of solute type and concentration on these properties has been well studied, the influence of steep thermal gradients—typical of rapid solidification processes like metal additive manufacturing and welding—remains largely unexplored.

We simulate a series of model polymer composites, composed of linear polymer strands and spherical, monodisperse filler particles (FP). These molecular dynamics simulations implement a coarse-grained, bead-spring force field and we vary several formulation parameters to study their respective influences on material properties. Uniaxial extension of the simulation cells allows direct comparison of mechanical reinforcement (or weakening) provided by the FP. We focus on the formation of microscopic voids during simulated tensile testing of glassy polymer composites and quantify how the characteristic spatial and morphological arrangement of these voids is a function of interaction potentials used in the simulation.

Universal machine learning interatomic potentials (MLIPs) are transforming materials science by providing the quantum-level accuracy essential for novel scientific discovery and the computational efficiency required for large-scale industrial applications. Here, we introduce Matlantis’s latest results in developing a universal MLIP, PreFerred Potential (PFP) version 8. PFP v8 has been trained on large-scale DFT datasets, including a PBE-GGA dataset covering arbitrary combinations of 96 elements, and an r2SCAN dataset covering 70 elements.

Heat-Assisted Magnetic Recording (HAMR) stands at the forefront of next-generation data storage technologies, offering a promising path to significantly increase the areal density of hard disk drives. Recent advancements have brought HAMR from an idea to a reality, particularly through innovations in component miniaturization. As these components operate at the nanoscale, understanding their behavior requires tools capable of capturing atomic-level interactions. Molecular dynamics (MD) simulations provide a powerful framework for exploring the complex physics governing the optical elements of the recording head and the thermal properties of the media.

We seek to understand and predict the response of biological membranes to mechanical shock. Semi-flexible and rigid models are commonly used to model water in such systems under equilibrium conditions, with relatively minor effects on many equilibrium thermodynamic properties. Two important exceptions are the heat capacity and the Grüneisen parameter; both of which manifest strong quantum mechanical behaviors. The nascent thermodynamic state behind a shock is sensitive to both the isochoric heat capacity and the Grüneisen parameter; together, they largely determine the temperature increase across a shock discontinuity.

Advances in metallurgy have revealed new classes of complex concentrated alloys with impressive thermomechanical, chemical, and other properties. Unraveling their underlying metallurgical mechanisms for processes like segregation and formation of precipitates that control alloy properties remains an elusive but necessary prerequisite for predicting and enhancing their performance. A powerful technique for simulating these processes is Hybrid Monte Carlo (MC)/Molecular Dynamics (MD), which combines MC steps, to rapidly alter alloy chemistry, with MD relaxations, to explore atomistic trajectories.

Aspherix is an advanced Discrete Element Method (DEM) simulation platform for modeling granular materials, developed as the commercial successor to the open-source LIGGGHTS code, itself based on the LAMMPS molecular dynamics engine. Building on these robust foundations, Aspherix delivers significant improvements in computational performance, physical modeling capabilities, and simulation flexibility. Key features include a modern graphical user interface (GUI) and an intuitive, natural-language-inspired command syntax, both designed to streamline simulation setup and lower the barrier to entry for new users.

The Rankine-Hugoniot Equation is a tenet of shock physics, predicting the existence of a relationship between initial and final states within a shock wave based only on simple conservation laws and a steady-state assumption. Experiments in highly porous materials appear inconsistent with Rankine-Hugoniot predictions derived from first-principle calculations. Resolving these inconsistencies is challenging due to the inherent variability in porous target fabrication, that often leads to large experimental uncertainties. Here, we use molecular dynamics to explore the conditions in which highly porous materials (silica glass, copper, and a Lennard-Jones model system) stray from the Rankine-Hugoniot predictions.

Polyethylene fiber films with highly oriented lamellae exhibit superior thermal and mechanical properties. Stretch-induced crystallization (SIC) using molecular dynamics simulations in LAMMPS accurately mimics fiber film formation, providing valuable insights into crystallization kinetics, inner structure, and mechanical properties. In this investigation, the morphological and mechanical (Uniaxial elongation) analyses complement the SIC method’s effectiveness in designing high-performance fiber films. The one-dimensional density (1D) along the Z-axis shows that lamellar orientation and thickness change with temperature and strain rates.

In fusion reactors, the integrity of plasma-facing components is critically challenged by high-velocity tungsten (W) dust impacts, particularly during events like runaway electron terminations [1]. These incidents can propel dust particles at velocities potentially exceeding several km/s, leading to severe material erosion and damage. Current models predominantly address only low-velocity impacts, revealing a significant gap in our understanding necessary for reactors such as ITER and DEMO [2]. To address this, our study employs large-scale molecular dynamics (MD) simulations to investigate the effects of high-velocity dust impacts on W walls.

Training and deploying machine-learned interatomic potentials (MLIPs) can require significant effort and non-trivial software infrastructure. In this talk, I will discuss the impact of the recent overhaul of the NequIP code [1] for training and deploying deep equivariant MLIPs, in particular the message-passing NequIP and the strictly local Allegro. This software development effort aims to produce a modular and extensible framework to keep up with the rapid advancements in machine learning for materials and molecules.

LAMMPS currently supports GPU acceleration through the KOKKOS and GPU packages. While the KOKOS package has been under active development, the GPU package is mostly under maintenance mode with occasional updates. To many LAMMPS users, it is useful to know which package would give optimal performance for a given problem. In this talk, I will present my recent contribution to both packages by adding new features and improving the GPU speedup.

High-resolution transmission electron microscopy (HR-TEM) is a powerful tool for visualizing atomic structures and characterizing material behavior. Its natural companion in the modeling and simulation space is molecular dynamics (MD), which matches HR-TEM length scales and can provide a wealth of complementary information. A common approach in HR-TEM studies using MD is to identify a set of defects of interest in the physical sample, then reconstruct said defects virtually in an MD-friendly file format by hand or via atomistic simulation tools.

The LAMMPS homepage stated for many years that LAMMPS does not have a GUI. When preparing a LAMMPS tutorial and looking for a set of tools that is consistent across major platforms for editing input files, visualizing the simulated system, and plotting thermodynamic data that the results were somewhat lacking. Thus, the idea was born to write a cross-platform graphical user interface that would include basic versions of such tools and LAMMPS in a single executable.

Since its inception in 1995, LAMMPS has grown to be a world class molecular dynamics code, with thousands of users, over one million lines of code, and multi-scale simulation capabilities. In this presentation, I will discuss how LAMMPS has adapted to the modern heterogeneous computing landscape by integrating the Kokkos performance portability library with a modular and user-friendly approach. I will also discuss the seamless ways in which LAMMPS supports advances in GPU-accelerated machine-learned models for more physically realistic simulations.

Stretched polymer materials exhibit interesting two-dimensional scattering patterns (2DSPs) that originated from network heterogeneity and bond breakages. Abnormal butterfly patterns (ABPs) were widely observed in scattering experiments, where ABPs in the small angle region give rise to a peak spot on the q-axis in the stretching direction and a dark streak on the q-axis perpendicular. We performed our large-scale CGMD simulation with large-scale system size to describe small-angle 2DSPs and confirmed the reproduction of the ABPs in various polymer network systems.

I will first briefly summarize the so-called Modified Johnson-Cook (M-JC) continuum material model for mesoscale simulations of HMX under extreme conditions and then attempt to confront the three questions posed in the title. The M-JC model builds from a standard Johnson-Cook form for material strength that, for explosives such as HMX, is typically calibrated empirically—e.g., by fitting to wave-profile data from macroscale flyer-plate shock experiments—and often with limited treatment of the underlying ‘sub-grid’ physics.

Machine-learned interatomic potentials (MLIPs) can offer near first-principles accuracy but are computationally expensive, limiting their application to large-scale molecular dynamics simulations. Inspired by quantum mechanics/molecular mechanics (QM/MM) methods this talk will present ML-MIX, a CPU and GPU compatible LAMMPS package to accelerate simulations by spatially force-mixing interatomic potentials of different complexities. By restricting the use of costly MLIPs to only the necessary regions of the simulation domain, ML-MIX enables researchers with constrained computational resources to overcome traditional cost–accuracy trade-offs.

Friction of aluminum borates was investigated using molecular dynamics simulations. Contrary to conventional expectations, it is found that specific combinations of temperature and velocity lead to unexpected increases in the coefficient of friction (COF), influenced by the elemental distribution in the lubricant’s surface layer. While Alumina borate generally maintains its structure across various conditions, certain thermal and mechanical environments cause deviations that negatively affect COF and hardness. It is also noted that water molecules on the surface reduces COF and enhance hardness, due to water-lubricant interactions.

Large language models (LLMs) offer new capabilities for bridging natural language and domain-specific languages in scientific computing. Here, we investigate the use of LLMs to translate from English task descriptions to LAMMPS input scripts. Our workflow pairs the LLM-generated input scripts with our newly developed parser to check structural correctness, validate syntax, and assist in debugging. We systematically evaluate model performance across tasks of increasing complexity, from basic thermalization to multi-fix, multi-region configurations.

The NIST Interatomic Potentials Repository has long served as host for LAMMPS-compatible classical interatomic potentials, and has provided tools for finding, accessing, and evaluating the hosted potentials. This talk will focus on how our framework for measuring predicted materials properties across the potentials has been extended to explore complex defect structures (dislocations and grain boundaries), and dynamic liquid and crystal structure predictions. Calculation results are made available both on the main repository website as well as from within our queryable public database.

Glassy carbon (g-C), also known as vitreous carbon, is a metastable, pure sp² carbon phase composed of a random assembly of twisted graphene fragments, resulting in a material with a lower density than graphite. The amorphous nature of g-C enables the exploration of novel bonding configurations and phase transitions that are challenging to observe in crystalline materials. We investigate the metastability of g-C and its transformation into diamond, as well as metastable and stable carbon liquids, across a wide range of pressures and temperatures using large-scale molecular dynamics (MD) simulations.

In this talk, I will highlight an application for which polymeric properties are derived. These properties need the production of many data points, for which I used our internal workflow, which includes LAMMPS. The data points are combined and used in parameterizing a single equation, which is subsequently used to predict properties as a function of composition and temperature.

Spin-Lattice Dynamics (SLD) simulations, implemented via LAMMPS, provide a powerful framework to study the coupled evolution of magnetization and lattice structure in complex magnetic nanosystems. Unlike micromagnetic approaches, SLD resolves atomic-scale features—such as defects, strain, and spin-lattice coupling—enabling unique insights into emergent phenomena. This talk will highlight three applications. Magnetic domain wall (DW) motion in Fe nanowires can provide a simple system to understand a problem of technological relevance. In perfect single crystals motion proceeds in agreement with a simple analytic model, depending on anisotropy, exchange strength, damping and external field.

Skyrmions are topologically protected spin textures with promising applications in spintronics due to their nanoscale size, mobility, and stability. Accurately modeling their behavior under external perturbations, such as mechanical strain, requires approaches that go beyond conventional micromagnetic or spin dynamics simulations. In this work, we use Spin-Lattice Dynamics (SLD) simulations implemented in LAMMPS to investigate the influence of uniaxial strain on skyrmion properties in two-dimensional magnetic systems. Unlike micromagnetic or standard spin dynamics approaches, SLD allows for the fully atomistic treatment of coupled spin and lattice degrees of freedom, enabling us to capture strain-induced effects with higher fidelity, including local lattice distortions and thermal fluctuations.

Machine Learning Potentials (MLPs) have emerged as powerful tools for modeling interatomic interactions in materials science. The MLIP package provides an efficient and versatile framework for realizing MLPs within the LAMMPS molecular dynamics simulation package. In this talk, we present an improved MLIP package that includes enhanced functionality and performance. The package is tested for various elements, including aluminum, silicon, and magnesiumBy leveraging the capabilities of tensor networks, we demonstrate the potential of MLPs for modeling complex interatomic interactions.

The OpenKIM project has been serving the atomistic simulation community since 2009. The project includes the largest repository of interatomic potentials on the web, an API and package manager for seamless download, installation, and use of these potentials in compatible simulation codes, and a distributed high-throughput computational pipeline that automatically checks each potential for coding integrity and evaluates its predictions of material properties. This last component – the evaluation of material properties – has been undergoing a major rework in the past several years.

Despite the rapid rise of machine-learned potentials, the use of more classical force fields, such as fixed-bond or reactive force fields like ReaxFF, still plays a central role in materials engineering. The main advantages of classical potentials are the large variety of functional forms, parameters, and existing tooling to allow for many materials to be simulated at speeds orders of magnitude quicker. Additionally, the transferability and relative ease of adding onto existing classical potentials for new materials exploration make fixed-bond force fields still relevant today in the world of machine-learning-based potentials.

I will highlight some of the new features in LAMMPS, added since the last workshop two years ago.

Skyrmions are topologically protected spin textures with promising applications in spintronics due to their nanoscale size, mobility, and stability. Accurately modeling their behavior under external perturbations, such as mechanical strain, requires approaches that go beyond conventional micromagnetic or spin dynamics simulations. In this work, we use Spin-Lattice Dynamics (SLD) simulations implemented in LAMMPS to investigate the influence of uniaxial strain on skyrmion properties in two-dimensional magnetic systems. Unlike micromagnetic or standard spin dynamics approaches, SLD allows for the fully atomistic treatment of coupled spin and lattice degrees of freedom, enabling us to capture strain-induced effects with higher fidelity, including local lattice distortions and thermal fluctuations.

Machine Learning Potentials (MLPs) have emerged as powerful tools for modeling interatomic interactions in materials science. The MLIP package provides an efficient and versatile framework for realizing MLPs within the LAMMPS molecular dynamics simulation package. In this talk, we present an improved MLIP package that includes enhanced functionality and performance. The package is tested for various elements, including aluminum, silicon, and magnesiumBy leveraging the capabilities of tensor networks, we demonstrate the potential of MLPs for modeling complex interatomic interactions.