Controlling plasma-wall interactions is critical to achieve high performance in present day tokamaks, and will continue to be the case in the approach to practical fusion reactors. Tungsten (W) is the main candidate as plasma facing material for a fusion reactors and will be exclusively used in the ITER divertor [1]. Outstanding technical issues are still to be overcome, for instance erosion/redeposition from plasma sputtering and disruptions, dust and flake generation.

We applied LAMMPS to study several important biophysical problems at different scales. First, by using a one-particle-thick lipid bilayer model we developed for LAMMPS, we simulated the endocytosis of Covid-19 virus at the coarse-grained molecular dynamics scale. Second, we developed a multi scale model of red blood cells (RBCs) based on LAMMPS and simulated the splenic filtration of RBCs at the cellular level. Third, we studied the blood rheology from healthy and sickle cell patients using dissipative particle dynamics (DPD), and simulated how the viscosity changes with varying percentage of rigid sickle cells and compared with experiments at the blood/tissue level.

Machine learning (ML) enables the development of interatomic potentials that promise the accuracy of first principles methods while retaining the low cost and parallel efficiency of empirical potentials. While ML potentials traditionally use atom-centered descriptors as inputs, different models such as linear regression and neural networks can map these descriptors to atomic energies and forces. This begs the question: what is the improvement in accuracy due to model complexity irrespective of choice of descriptors?

Allegro and FLARE are two very different packages for constructing machine learning potentials that are fast, accurate, and suitable for extreme-scale molecular dynamics simulations. Allegro uses PyTorch for efficient equivariant potentials with state-of-the-art accuracy, while FLARE is a sparse Gaussian process potential with an optimized C++ training backend leveraging Kokkos, OpenMP, and MPI for state-of-the-art performance, and a user-friendly Python frontend. We will compare and contrast the two methods, discuss lessons learned, and show scientific applications ranging from catalysis to biology.

Authors: Tesia Janicki, Carlos Chacon, Edwin Chiu, Jason Gibson, Scott Grutzik, Khalid Hattar, Richard Hennig, Calvin Parkin, Jennie Podlevsky, Aashique Rezwan, Chris Bishop, J. Matthew D. Lane Amorphous silicon nitride is a common layer stack component in microelectronics devices. Crystal defects formed during fabrication can produce materials with undesirable properties. To enact measures which avoid spurious crystallization, we must first understand the crystal growth mechanism. We explore this mechanism in a multiscale, interdisciplinary approach spanning experiment and meso- and atomistic-scale models.

The amount of energy stored in gas hydrate reservoirs is more than twice that in conventional hydrocarbon resources, rendering their economical exploration crucial for the world’s future energy security. This study introduces the first large-scale simulation of the isobaric thermal dissociation of methane hydrates, the most abundant source of gas hydrate reservoirs, under a symmetric thermal gradient. Leveraging the monoatomic water (mW) model and Stillinger-Weber (SW) potential, we simulated a system encompassing a hundred-fold more sI unit cells than in previous research studies, thus enabling the observation of unique phenomena unattainable within smaller simulation domains.

Machine learned potentials inherently perform worse on structures outside of the training set than structures directly trained to. This presents a challenge in developing stable, transferable potentials for high temperature applications as there is plenty of energy for the atoms to move away from ground state and other expected configurations. We will summarize lessons learned while developing the W-ZrC SNAP potential to study material behavior under fusion reactor conditions (≥1000K). We will discuss progress and challenges in developing a W-ZrC-H ACE potential, which will enable the study of plasma-material interactions.

Classical molecular dynamics simulations with the constant potential method (CPM) are an increasingly important tool for models with metallic electrodes at an electrostatic potential. In the CPM charges of individual electrode atoms are set to meet the applied potential. Especially charging mechanisms in next-generation energy storage devices have been studied with the CPM [1, 2]. As an implementation of the CPM, the ELECTRODE package for LAMMPS is presented [3]. This package features a particle-mesh solver to greatly reduce computation times of the long-range Coulomb interactions [4].

The bulk dynamics of granular systems are significantly affected by the shape of the constituent particles. [1][2][3] As an added, but necessary, complication, particle shape can change over time through the process of abrasion. [4] However, the current LAMMPS implementation does not allow for abrasion to be captured. This represents a major shortcoming, and area of interest, as in reality all granular particles abrade and change shape to some extent. [5] Hence, this current research proposes to implement abradable non-spherical particles into LAMMPS through a new fix abrasion.

Authors: Joseph Monti, Joel Clemmer, Ishan Srivastava, Leo Silbert, Gary Grest, Jeremy Lechman Power-law size distributions abound in natural and processed granular materials. To study the structure and properties of granular packings with such underlying size distributions, we present results of discrete element method jamming simulations with varying distribution exponent and span, with the latter ranging up to a 200:1 maximum particle size ratio in three dimensions. Our work underscores the importance of balancing the relative abundance of large-large and small-small particle contacts to optimize packing density and reveals the role and abundance of mechanically unstable particles present in the packing.

Biochar is a highly porous carbonaceous material commonly produced by the thermal decomposition of lignocellulosic material. It is an excellent, low cost adsorbent and is of great interest in the remediation of heavy metals from water. The surface properties of Biochar can be modified to enhance the efficiency of adsorption of specific compounds. However, experimental studies into the ad- sorbate - adsorbent interactions remain difficult and tedious. Although several in-silico studies have been carried out into these interactions using Density Func- tional Theory (DFT), these studies use graphitic models and single molecules in -lieu of biochar, which are not accurate representations of amorphous biochar.

Machine-learned interatomic potentials (ML-IAPs) are used to study many important physical and chemical phenomena, but often fall short for atomic systems with long-range interactions. For example, using an ML-IAP to study reactions in oxides is limited because of the large degree of charge transfer and long-range electrostatics. Methods that account for charge transfer and incorporate electrostatics into ML-IAPs are usually limited by cost and accuracy. Oxides and analogous systems represent a weak point in the modeling capabilities of ML-IAPs for these reasons.

Reactive molecular dynamics (MD) is a powerful tool for atomistic-scale modeling of a diverse range of chemical processes. However, scaling these simulations to large systems and long times scales remains a challenge because of the complexity of the potential energy function required. The authors previously developed a heuristic approach, called REACTER, that incorporates reactivity in MD simulations in a less general but much more computationally efficient manner. REACTER uses standard, fixed valence force fields as the underlying potential energy surface for describing all interatomic interactions but adds a procedure for enforcing user-defined reactions that occur when certain geometric constraints on relative atomic positions are satisfied.

The SLLOD equations of motion have long been used to simulate homogeneous flow. However, they were initially designed only for atomic fluids, and this was the algorithm implemented in LAMMPS (fix nvt/sllod and fix nvt/uef). Despite this, most contemporary systems of interest deal with molecular flow, in which case a modified version of the SLLOD algorithm, operating on molecular centres of mass rather than individual atoms, is required. This talk describes a new package, developed as part of a Pawsey Centre for Extreme Scale Readiness (PaCER) project, which implements the molecular SLLOD algorithm and includes support for GPU acceleration via Kokkos.

In this talk we will present a tool to simulate long-term diffusive mass transport in systems with atomic scale resolution [1]. The implemented framework is based on a non-equilibrium statistical thermo-chemo-mechanical formulation of atomic systems where the effective transport rates are computed using a kinematic diffusion law [2,3]. Our implementation is built as an add-on to the Large-scale Atomic/Molecular Massively Parallel Simulator (LAMMPS) code [4]. In applications involving diffusive mass transport, this framework is able to simulate problems of technological interest for exceedingly large time scales using an atomistic description, which are not reachable with the state-of-the-art molecular dynamics techniques.

The OpenKIM project (https://openkim.org) archives, tests and deploys interatomic potentials for use in production molecular dynamics codes like LAMMPS. Over the last couple of years OpenKIM has introduced several new capabilities that will be described in this talk, including: (1) the “Crystal Genome” framework with properties for all known crystals; (2) support for bonded force fields; (3) support for arbitrary PyTorch machine learning (ML) potentials; (4) the “Deep Citation” framework for identifying potential past usage; and a comparison tool and ML-based recommender system for potential selection.

The response of polymers to uniaxial elongational flow affects the structure of polymers which is critical for their processing. The more structured the polymer, the more complex its response becomes. Here we probe the effects of uniaxial elongational flow on entangled, linear polymer melts. The polymers are depicted by a bead spring model with 5% randomly incorporated interacting associating beads, as the interaction strengths vary from 1kBT to 10kBT, using molecular dynamics simulations.

The pyiron framework is an abstraction layer to orchestrate parameter studies for atomistic simulation, covering both empirical interatomic simulations as well as ab-initio methods like density functional theory (DFT). Based on a generic interface atomistic structures can be seamlessly transferred between simulation codes to enable the coupling of different levels of theory all within the same simulation protocol developed in the python programming language. For two selected examples we demonstrate how this data-driven approach provides new scientific insights: First for the fitting of an interatomic machine learning potential, the dependence of the hyperparameters cut-off radius and number of descriptors is systematically analyzed by coupling LAMMPS and FitSNAP inside pyiron.

Morphogenesis is a complex, multiscale process which is extremely important to understand and equally difficult to model. While continuum mechanics deals poorly with non-linear, large deformations caused by individual cells migrating or proliferating, some discrete models fail to capture the underlying multiscale mechanics. We present an enhanced version of the subcellular element modeling (SEM2) framework to tackle this problem. SEM models tissue mechanics by describing cells as ensembles of particles whose interactions are governed by empirically defined potentials.

C. Miguel Barriuso Gutiérrez1, José Martín-Roca1,2, Valentino Bianco2, Ignacio Pagonabarraga3,4,5 and Chantal Valeriani1,6* 1 Departamento de Estructura de la Materia, Física Térmica y Electrónica, Universidad Complutense de Madrid, Madrid, Spain 2 Departamento de Química Física, Facultad de Química, Universidad Complutense de Madrid, Madrid, Spain 3 Departament de Física de la Matèria Condesada, Facultat de Física, Universitat de Barcelona, Barcelona, Spain 4 Universitat de Barcelona Institute of Complex Systems, Barcelona, Spain 5 Centre Européen de Calcul Atomique et Moléculaire (CECAM), Ecole Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland 6 GISC—Grupo Interdisciplinar de Sistemas Complejos, Madrid, Spain

High entropy alloys (HEA) are being intensely studied due to their extraordinary mechanical properties, radiation damage, corrosion resistance, etcetera. Most simulation studies have focused on random HEA, without any chemical ordering, and we have studied collision cascades [1] and mechanical properties of fcc [2,3] and bcc alloys [4,5], including porosity and grain boundaries in some cases. Due to large lattice distortion, Machine Learning can help to avoid significant noise in standard structure detection methods [6].

The USER-SELM package provides methods for incorporating into models continuum hydrodynamic coupling with thermal fluctuations into LAMMPS simulations. Motivations include stochastic and deterministic simulations of dynamic implicit-solvent coarse-grained models (for example colloids / polymers / membranes), general fluid-structure interactions subject to thermal fluctuations (selms / immersed boundary models), and shear boundary conditions for micro-rheology studies (for example rheology of soft materials / complex fluids). We discuss the latest updates to the package and example use cases for how to set up models.