arXiv subjects

Brennon L. Shanks

Publications and source records attributed to Brennon L. Shanks.

7 recordsLinked to original sources

arXiv2026

Uncertainty quantification design principles for machine learning interatomic potentials: lessons learned from hierarchical Bayesian inference

Reliable uncertainty quantification (UQ) remains a major challenge for machine learning interatomic potentials (MLIPs). Here, we benchmark UQ strategies for message passing neural networks such as MACE and Gaussian process (GP)-based interatomic potentials using coupled-cluster reference data for argon as a test case. We develop a hierarchical GP potential combining physics-informed priors with Bayesian propagation of hyperparameter uncertainty that retains useful discrimination of prediction errors while remaining conservative rather than overconfident, with a probabilistic structure that allows the sources of predictive uncertainty to be explicitly dissected. These results expose important distinctions between accuracy, calibration, sharpness, and error discrimination and provide practical design principles for MLIP UQ in uncertainty-critical applications.

cond-mat.mtrl-sci↗

arXiv2026

Uncertainty Quantification for Free Energy Calculations by Generalized Hierarchical Bayesian Inference

Free energy calculations are routinely used to study molecular processes inaccessible to unbiased molecular dynamics, but their utility ultimately depends on knowing when and how much their predictions can be trusted. Uncertainty estimation is therefore essential for distinguishing genuine physical features of a free energy profile from artifacts arising from limited simulation data or inadequate sampling. Gaussian processes have emerged as a powerful framework for reconstructing free energy profiles together with predictive uncertainties. However, existing implementations typically condition on fixed hyperparameters and observation noise, preventing predictive uncertainties from adapting to the information content of the simulation data. Here, we develop a generalized hierarchical Gaussian process framework that accounts for these neglected sources of uncertainty. Applications to umbrella sampling and extended Lagrangian metadynamics of peptide-lipid membrane interactions demonstrate that the resulting uncertainty estimates track reconstruction errors across a wide range of sampling and data conditions.

physics.chem-ph↗

arXiv2026

Physics-Informed Gaussian Process Inference of Liquid Structure from Scattering Data

We present a nonparametric Bayesian framework to infer radial distribution functions from experimental scattering measurements with uncertainty quantification using non-stationary Gaussian processes. The Gaussian process prior mean and kernel functions are designed to mitigate well-known numerical challenges with the Fourier transform, including discrete measurement binning and detector windowing, while encoding fundamental yet minimal physical knowledge of liquid structure. We demonstrate uncertainty propagation of the Gaussian process posterior to unmeasured quantities of interest. Experimental radial distribution functions of liquid argon and water with uncertainty quantification are provided as both a proof of principle for the method and a benchmark for molecular models. The full implementation is available on GitHub at: https://github.com/hoepfnergroup/LiquidStructureGP-Sullivan.

physics.chem-ph↗

arXiv2025

Bayesian learning for accurate and robust biomolecular force fields

Molecular dynamics is a valuable tool to probe biological processes at the atomistic level - a resolution often elusive to experiments. However, the credibility of molecular models is limited by the accuracy of the underlying force field, which is often parametrized relying on ad hoc assumptions. To address this gap, we present a Bayesian framework for learning physically grounded parameters directly from ab initio molecular dynamics data. By representing both model parameters and data probabilistically, the framework yields interpretable, statistically rigorous models in which uncertainty and transferability emerge naturally from the learning process. This approach provides a transparent, data-driven foundation for developing predictive molecular models and enhances confidence in computational descriptions of biophysical systems. We demonstrate the method using 18 biologically relevant molecular fragments that capture key motifs in proteins, nucleic acids, and lipids, and, as a proof of concept, apply it to calcium binding to troponin - a central event in cardiac regulation.

physics.chem-ph↗

arXiv2025

Uncertainty-Aware Liquid State Modeling from Experimental Scattering Measurements

This dissertation is founded on the central notion that structural correlations in dense fluids, such as dense gases, liquids, and glasses, are directly related to fundamental interatomic forces. This relationship was identified early in the development of statistical theories of fluids through the mathematical formulations of Gibbs in the 1910s. However, it took nearly 80 years before practical implementations of structure-based theories became widely used for interpreting and understanding the atomic structures of fluids from experimental X-ray and neutron scattering data. The breakthrough in successfully applying structure-potential relations is largely attributed to the advancements in molecular mechanics simulations and the enhancement of computational resources. Despite advancements in understanding the relationship between structure and interatomic forces, a significant gap remains. Current techniques for interpreting experimental scattering measurements are widely used, yet there is little evidence that they yield physically accurate predictions for interatomic forces. In fact, it is generally assumed that these methods produce interatomic forces that poorly model the atomistic and thermodynamic behavior of fluids, rendering them unreliable and non-transferable. This thesis aims to address these limitations by refining the statistical theory, computational methods, and philosophical approach to structure-based analyses, thereby developing more robust and accurate techniques for characterizing structure-potential relationships.

physics.chem-ph↗

arXiv2025

Experimental Evidence of Quantum Drude Oscillator Behavior in Liquids Revealed with Probabilistic Iterative Boltzmann Inversion

The first experimental evidence of quantum Drude oscillator behavior in liquids is determined using probabilistic machine learning-augmented iterative Boltzmann inversion applied to noble gas radial distribution functions. Furthermore, classical force fields for noble gases are shown to be reduced to a single parameter through simple empirical relations linked to atomic dipole polarizability. These findings highlight how neutron scattering data can inspire innovative force field design and offer insight into interatomic forces to advance molecular simulations.

physics.chem-ph↗

arXiv2024

Bayesian Analysis Reveals the Key to Extracting Pair Potentials from Neutron Scattering Data

The inverse problem of statistical mechanics is an unsolved, century-old challenge to learn classical pair potentials directly from experimental scattering data. This problem was extensively investigated in the 20th century but was eventually eclipsed by standard methods of benchmarking pair potentials to macroscopic thermodynamic data. However, it is becoming increasingly clear that existing force field models fail to reliably reproduce fluid structures even in simple liquids, which can result in reduced transferability and substantial misrepresentations of thermophysical behavior and self-assembly. In this study, we revisited the structure inverse problem for a classical Mie fluid to determine to what extent experimental uncertainty in neutron scattering data influences the ability to recover classical pair potentials. Bayesian uncertainty quantification was used to show that structure factors with random noise smaller than 0.005 to $\sim30$ A$^{-1}$ are required to accurately recover pair potentials from neutron scattering. Notably, modern neutron instruments can achieve this precision to extract classical force models to within approximately $\pm$ 1.3 for the repulsive exponent, $\pm$ 0.068 A$^{-1}$ for atomic size, and 0.024 kcal/mol in the potential well-depth with 95\% confidence. Our results suggest the exciting possibility of improving molecular simulation accuracy through the incorporation of neutron scattering data, advancement in structural modeling, and extraction of model-independent measurements of local atomic forces in real fluids.

cond-mat.stat-mech↗