arXiv subjects

Justin S. Smith

Publications and source records attributed to Justin S. Smith.

At least 19 recordsLinked to original sources

arXiv2026

Projected Hessian Learning: Fast Curvature Supervision for Accurate Machine-Learning Interatomic Potentials

The Hessian matrix (second derivatives) encodes far richer local curvature of the potential energy surface than energies and forces alone. However, training machine-learning interatomic potentials (MLIPs) with full Hessians is often impractical because explicitly forming and storing Hessian matrices scales quadratically in cost and memory. We introduce Projected Hessian Learning (PHL), a scalable second-order training framework that injects curvature information using only Hessian-vector products (HVPs). Rather than constructing the Hessian, PHL projects curvature along stochastic probe directions and uses an unbiased stochastic trace-based loss with favorable system-size scaling, enabling curvature-informed training without quadratic memory growth. We benchmark PHL on a chemically diverse dataset of reactants, products, transition states, intrinsic reaction coordinates, and normal-mode sampled geometries computed at omegaB97XD/6-31G(d). We compare energy-force training (E-F), two HVP-based schemes (E-F-HVP with one-hot or randomized probes), and full energy-force-Hessian training (E-F-H). With randomized probes per minibatch, both HVP schemes match full-Hessian training in energy, force, and Hessian accuracy while delivering >24x epoch speedups for the small molecular systems studied. In a fixed-probe regime with one HVP per molecule, randomized projections consistently outperform one-column probing, especially for far-from-equilibrium geometries. Overall, PHL replaces explicit Hessian supervision with force-complexity curvature training, retaining most second-order accuracy gains while scaling to larger, more complex molecular systems.

physics.chem-ph↗

arXiv2025

Multi-fidelity learning for interatomic potentials: Low-level forces and high-level energies are all you need

The promise of machine learning interatomic potentials (MLIPs) has led to an abundance of public quantum mechanical (QM) training datasets. The quality of an MLIP is directly limited by the accuracy of the energies and atomic forces in the training dataset. Unfortunately, most of these datasets are computed with relatively low-accuracy QM methods, e.g., density functional theory with a moderate basis set. Due to the increased computational cost of more accurate QM methods, e.g., coupled-cluster theory with a complete basis set extrapolation, most high-accuracy datasets are much smaller and often do not contain atomic forces. The lack of high-accuracy atomic forces is quite troubling, as training with force data greatly improves the stability and quality of the MLIP compared to training to energy alone. Because most datasets are computed with a unique level of theory, traditional single-fidelity learning is not capable of leveraging the vast amounts of published QM data. In this study, we apply multi-fidelity learning to train an MLIP to multiple QM datasets of different levels of accuracy, i.e., levels of fidelity. Specifically, we perform three test cases to demonstrate that multi-fidelity learning with both low-level forces and high-level energies yields an extremely accurate MLIP -- far more accurate than a single-fidelity MLIP trained solely to high-level energies and almost as accurate as a single-fidelity MLIP trained directly to high-level energies and forces. Therefore, multi-fidelity learning greatly alleviates the need for generating large and expensive datasets containing high-accuracy atomic forces and allows for more effective training to existing high-accuracy energy-only datasets. Indeed, low-accuracy atomic forces and high-accuracy energies are all that are needed to achieve a high-accuracy MLIP with multi-fidelity learning.

physics.comp-ph↗

arXiv2025

Reactive Chemistry at Unrestricted Coupled Cluster Level: High-throughput Calculations for Training Machine Learning Potentials

Accurately modeling chemical reactions at the atomistic level requires high-level electronic structure theory due to the presence of unpaired electrons and the need to properly describe bond breaking and making energetics. Commonly used approaches such as Density Functional Theory (DFT) frequently fail for this task due to deficiencies that are well recognized. However, for high-fidelity approaches, creating large datasets of energies and forces for reactive processes to train machine learning interatomic potentials or force fields is daunting. For example, the use of the unrestricted coupled cluster level of theory has previously been seen as unfeasible due to high computational costs, the lack of analytical gradients in many computational codes, and additional challenges such as constructing suitable basis set corrections for forces. In this work, we develop new methods and workflows to overcome the challenges inherent to automating unrestricted coupled cluster calculations. Using these advancements, we create a dataset of gas-phase reactions containing energies and forces for 3119 different organic molecules configurations calculated at the gold-standard level of unrestricted CCSD(T) (coupled cluster singles doubles and perturbative triples). With this dataset, we provide an analysis of the differences between the density functional and unrestricted CCSD(T) descriptions. We develop a transferable machine learning interatomic potential for gas-phase reactions, trained on unrestricted CCSD(T) data, and demonstrate the advantages of transitioning away from DFT data. Transitioning from training to DFT to training to UCCSD(T) datasets yields an improvement of more than 0.1 eV/Å in force accuracy and over 0.1 eV in activation energy reproduction.

physics.chem-ph↗

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

Optimizing Data Distribution and Kernel Performance for Efficient Training of Chemistry Foundation Models: A Case Study with MACE

Chemistry Foundation Models (CFMs) that leverage Graph Neural Networks (GNNs) operating on 3D molecular graph structures are becoming indispensable tools for computational chemists and materials scientists. These models facilitate the understanding of matter and the discovery of new molecules and materials. In contrast to GNNs operating on a large homogeneous graphs, GNNs used by CFMs process a large number of geometric graphs of varying sizes, requiring different optimization strategies than those developed for large homogeneous GNNs. This paper presents optimizations for two critical phases of CFM training: data distribution and model training, targeting MACE - a state-of-the-art CFM. We address the challenge of load balancing in data distribution by formulating it as a multi-objective bin packing problem. We propose an iterative algorithm that provides a highly effective, fast, and practical solution, ensuring efficient data distribution. For the training phase, we identify symmetric tensor contraction as the key computational kernel in MACE and optimize this kernel to improve the overall performance. Our combined approach of balanced data distribution and kernel optimization significantly enhances the training process of MACE. Experimental results demonstrate a substantial speedup, reducing per-epoch execution time for training from 12 to 2 minutes on 740 GPUs with a 2.6M sample dataset.

cs.DC↗

arXiv2025

Does Hessian Data Improve the Performance of Machine Learning Potentials?

Integrating machine learning into reactive chemistry, materials discovery, and drug design is revolutionizing the development of novel molecules and materials. Machine Learning Interatomic Potentials (MLIPs) accurately predict energies and forces at quantum chemistry levels, surpassing traditional methods. Incorporating force fitting into MLIP training significantly improves the representation of potential-energy surfaces (PES), enhancing model transferability and reliability. This study introduces and evaluates incorporating Hessian matrix training into MLIPs, capturing second-order curvature information of PES. Our analysis specifically examines MLIPs trained solely on stable molecular geometries, assessing their extrapolation capabilities to non-equilibrium configurations. We show that integrating Hessian information substantially improves MLIP performance in predicting energies, forces, and Hessians for non-equilibrium structures. Hessian-trained MLIPs notably enhance reaction pathway modeling, transition state identification, and vibrational spectra accuracy, benefiting molecular dynamics simulations and Nudged Elastic Band (NEB) calculations. By comparing models trained with various combinations of energy, force, and Hessian data on a small-molecule reactive dataset, we demonstrate Hessian inclusion leads to improved accuracy in reaction modeling and vibrational analyses while simultaneously reducing the total data needed for effective training. The primary trade-off is increased computational expense, as Hessian training demands more resources than conventional methods. Our results offer comprehensive insights into the strengths and limitations of Hessian integration in MLIP training, enabling practitioners in computational chemistry to make informed decisions aligned with their research goals and available computational resources.

physics.chem-ph↗

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

Lightweight and Effective Tensor Sensitivity for Atomistic Neural Networks

Atomistic machine learning focuses on the creation of models which obey fundamental symmetries of atomistic configurations, such as permutation, translation, and rotation invariances. In many of these schemes, translation and rotation invariance are achieved by building on scalar invariants, e.g., distances between atom pairs. There is growing interest in molecular representations that work internally with higher rank rotational tensors, e.g., vector displacements between atoms, and tensor products thereof. Here we present a framework for extending the Hierarchically Interacting Particle Neural Network (HIP-NN) with Tensor Sensitivity information (HIP-NN-TS) from each local atomic environment. Crucially, the method employs a weight tying strategy that allows direct incorporation of many-body information while adding very few model parameters. We show that HIP-NN-TS is more accurate than HIP-NN, with negligible increase in parameter count, for several datasets and network sizes. As the dataset becomes more complex, tensor sensitivities provide greater improvements to model accuracy. In particular, HIP-NN-TS achieves a record mean absolute error of 0.927 kcal/mol for conformational energy variation on the challenging COMP6 benchmark, which includes a broad set of organic molecules. We also compare the computational performance of HIP-NN-TS to HIP-NN and other models in the literature.

physics.chem-ph↗

arXiv2021

Quantum-based Molecular Dynamics Simulations Using Tensor Cores

Tensor cores, along with tensor processing units, represent a new form of hardware acceleration specifically designed for deep neural network calculations in artificial intelligence applications. Tensor cores provide extraordinary computational speed and energy efficiency, but with the caveat that they were designed for tensor contractions (matrix-matrix multiplications) using only low-precision floating point operations. In spite of this, we demonstrate how tensor cores can be applied with high efficiency to the challenging and numerically sensitive problem of quantum-based Born-Oppenheimer molecular dynamics, which requires highly accurate electronic structure optimizations and conservative force evaluations. The interatomic forces are calculated on-the-fly from an electronic structure that is obtained from a generalized deep neural network, where the computational structure naturally takes advantage of the exceptional processing power of the tensor cores and allows for high performance in excess of 100 Tflops on the tensor cores of a single Nvidia A100 GPU. Stable molecular dynamics trajectories are generated using the framework of extended Lagrangian Born-Oppenheimer molecular dynamics, which combines computational efficiency with long-term stability, even when using approximate charge relaxations and force evaluations that are limited in accuracy by the numerically noisy conditions caused by the low precision tensor core floating-point operations. A canonical ensemble simulation scheme is also presented, where the additional numerical noise in the calculated forces is absorbed into a Langevin-like dynamics.

physics.comp-ph↗

arXiv2021

NavTuner: Learning a Scene-Sensitive Family of Navigation Policies

The advent of deep learning has inspired research into end-to-end learning for a variety of problem domains in robotics. For navigation, the resulting methods may not have the generalization properties desired let alone match the performance of traditional methods. Instead of learning a navigation policy, we explore learning an adaptive policy in the parameter space of an existing navigation module. Having adaptive parameters provides the navigation module with a family of policies that can be dynamically reconfigured based on the local scene structure, and addresses the common assertion in machine learning that engineered solutions are inflexible. Of the methods tested, reinforcement learning (RL) is shown to provide a significant performance boost to a modern navigation method through reduced sensitivity of its success rate to environmental clutter. The outcomes indicate that RL as a meta-policy learner, or dynamic parameter tuner, effectively robustifies algorithms sensitive to external, measurable nuisance factors.

cs.RO↗

arXiv2020

Automated discovery of a robust interatomic potential for aluminum

Accuracy of molecular dynamics simulations depends crucially on the interatomic potential used to generate forces. The gold standard would be first-principles quantum mechanics (QM) calculations, but these become prohibitively expensive at large simulation scales. Machine learning (ML) based potentials aim for faithful emulation of QM at drastically reduced computational cost. The accuracy and robustness of an ML potential is primarily limited by the quality and diversity of the training dataset. Using the principles of active learning (AL), we present a highly automated approach to dataset construction. The strategy is to use the ML potential under development to sample new atomic configurations and, whenever a configuration is reached for which the ML uncertainty is sufficiently large, collect new QM data. Here, we seek to push the limits of automation, removing as much expert knowledge from the AL process as possible. All sampling is performed using MD simulations starting from an initially disordered configuration, and undergoing non-equilibrium dynamics as driven by time-varying applied temperatures. We demonstrate this approach by building an ML potential for aluminum (ANI-Al). After many AL iterations, ANI-Al teaches itself to predict properties like the radial distribution function in melt, liquid-solid coexistence curve, and crystal properties such as defect energies and barriers. To demonstrate transferability, we perform a 1.3M atom shock simulation, and show that ANI-Al predictions agree very well with DFT calculations on local atomic environments sampled from the nonequilibrium dynamics. Interestingly, the configurations appearing in shock appear to have been well sampled in the AL training dataset, in a way that we illustrate visually.

cond-mat.mtrl-sci↗

arXiv2020

Good Graph to Optimize: Cost-Effective, Budget-Aware Bundle Adjustment in Visual SLAM

The cost-efficiency of visual(-inertial) SLAM (VSLAM) is a critical characteristic of resource-limited applications. While hardware and algorithm advances have been significantly improved the cost-efficiency of VSLAM front-ends, the cost-efficiency of VSLAM back-ends remains a bottleneck. This paper describes a novel, rigorous method to improve the cost-efficiency of local BA in a BA-based VSLAM back-end. An efficient algorithm, called Good Graph, is developed to select size-reduced graphs optimized in local BA with condition preservation. To better suit BA-based VSLAM back-ends, the Good Graph predicts future estimation needs, dynamically assigns an appropriate size budget, and selects a condition-maximized subgraph for BA estimation. Evaluations are conducted on two scenarios: 1) VSLAM as standalone process, and 2) VSLAM as part of closed-loop navigation system. Results from the first scenario show Good Graph improves accuracy and robustness of VSLAM estimation, when computational limits exist. Results from the second scenario, indicate that Good Graph benefits the trajectory tracking performance of VSLAM-based closed-loop navigation systems, which is a primary application of VSLAM.

cs.CV↗

arXiv2020

Simple and efficient algorithms for training machine learning potentials to force data

Abstract Machine learning models, trained on data from ab initio quantum simulations, are yielding molecular dynamics potentials with unprecedented accuracy. One limiting factor is the quantity of available training data, which can be expensive to obtain. A quantum simulation often provides all atomic forces, in addition to the total energy of the system. These forces provide much more information than the energy alone. It may appear that training a model to this large quantity of force data would introduce significant computational costs. Actually, training to all available force data should only be a few times more expensive than training to energies alone. Here, we present a new algorithm for efficient force training, and benchmark its accuracy by training to forces from real-world datasets for organic chemistry and bulk aluminum.

physics.comp-ph↗

arXiv2020

Closed-Loop Benchmarking of Stereo Visual-Inertial SLAM Systems: Understanding the Impact of Drift and Latency on Tracking Accuracy

Visual-inertial SLAM is essential for robot navigation in GPS-denied environments, e.g. indoor, underground. Conventionally, the performance of visual-inertial SLAM is evaluated with open-loop analysis, with a focus on the drift level of SLAM systems. In this paper, we raise the question on the importance of visual estimation latency in closed-loop navigation tasks, such as accurate trajectory tracking. To understand the impact of both drift and latency on visual-inertial SLAM systems, a closed-loop benchmarking simulation is conducted, where a robot is commanded to follow a desired trajectory using the feedback from visual-inertial estimation. By extensively evaluating the trajectory tracking performance of representative state-of-the-art visual-inertial SLAM systems, we reveal the importance of latency reduction in visual estimation module of these systems. The findings suggest directions of future improvements for visual-inertial SLAM.

cs.RO↗

arXiv2019

Machine Learned Hückel Theory: Interfacing Physics and Deep Neural Networks

The Hückel Hamiltonian is an incredibly simple tight-binding model famed for its ability to capture qualitative physics phenomena arising from electron interactions in molecules and materials. Part of its simplicity arises from using only two types of empirically fit physics-motivated parameters: the first describes the orbital energies on each atom and the second describes electronic interactions and bonding between atoms. By replacing these traditionally static parameters with dynamically predicted values, we vastly increase the accuracy of the extended Hückel model. The dynamic values are generated with a deep neural network, which is trained to reproduce orbital energies and densities derived from density functional theory. The resulting model retains interpretability while the deep neural network parameterization is smooth, accurate, and reproduces insightful features of the original static parameterization. Finally, we demonstrate that the Hückel model, and not the deep neural network, is responsible for capturing intricate orbital interactions in two molecular case studies. Overall, this work shows the promise of utilizing machine learning to formulate simple, accurate, and dynamically parameterized physics models.

cond-mat.dis-nn↗

arXiv2019

Autonomous, Monocular, Vision-Based Snake Robot Navigation and Traversal of Cluttered Environments using Rectilinear Gait Motion

Rectilinear forms of snake-like robotic locomotion are anticipated to be an advantage in obstacle-strewn scenarios characterizing urban disaster zones, subterranean collapses, and other natural environments. The elongated, laterally-narrow footprint associated with these motion strategies is well-suited to traversal of confined spaces and narrow pathways. Navigation and path planning in the absence of global sensing, however, remains a pivotal challenge to be addressed prior to practical deployment of these robotic mechanisms. Several challenges related to visual processing and localization need to be resolved to to enable navigation. As a first pass in this direction, we equip a wireless, monocular color camera to the head of a robotic snake. Visiual odometry and mapping from ORB-SLAM permits self-localization in planar, obstacle-strewn environments. Ground plane traversability segmentation in conjunction with perception-space collision detection permits path planning for navigation. A previously presented dynamical reduction of rectilinear snake locomotion to a non-holonomic kinematic vehicle informs both SLAM and planning. The simplified motion model is then applied to track planned trajectories through an obstacle configuration. This navigational framework enables a snake-like robotic platform to autonomously navigate and traverse unknown scenarios with only monocular vision.

cs.RO↗

arXiv2019

Machine learning for molecular dynamics with strongly correlated electrons

We use machine learning to enable large-scale molecular dynamics (MD) of a correlated electron model under the Gutzwiller approximation scheme. This model exhibits a Mott transition as a function of on-site Coulomb repulsion $U$. The repeated solution of the Gutzwiller self-consistency equations would be prohibitively expensive for large-scale MD simulations. We show that machine learning models of the Gutzwiller potential energy can be remarkably accurate. The models, which are trained with $N=33$ atoms, enable highly accurate MD simulations at much larger scales ($N\gtrsim10^{3}$). We investigate the physics of the smooth Mott crossover in the fluid phase.

cond-mat.str-el↗

arXiv2018

Less is more: sampling chemical space with active learning

The development of accurate and transferable machine learning (ML) potentials for predicting molecular energetics is a challenging task. The process of data generation to train such ML potentials is a task neither well understood nor researched in detail. In this work, we present a fully automated approach for the generation of datasets with the intent of training universal ML potentials. It is based on the concept of active learning (AL) via Query by Committee (QBC), which uses the disagreement between an ensemble of ML potentials to infer the reliability of the ensemble's prediction. QBC allows the presented AL algorithm to automatically sample regions of chemical space where the ML potential fails to accurately predict the potential energy. AL improves the overall fitness of ANAKIN-ME (ANI) deep learning potentials in rigorous test cases by mitigating human biases in deciding what new training data to use. AL also reduces the training set size to a fraction of the data required when using naive random sampling techniques. To provide validation of our AL approach we develop the COMP6 benchmark (publicly available on GitHub), which contains a diverse set of organic molecules. Through the AL process, it is shown that the AL-based potentials perform as well as the ANI-1 potential on COMP6 with only 10% of the data, and vastly outperforms ANI-1 with 25% the amount of data. Finally, we show that our proposed AL technique develops a universal ANI potential (ANI-1x) that provides accurate energy and force predictions on the entire COMP6 benchmark. This universal ML potential achieves a level of accuracy on par with the best ML potentials for single molecule or materials, while remaining applicable to the general class of organic molecules comprised of the elements CHNO.

physics.comp-ph↗