Search NASA⌕ Search

SEARCH · Search NASA

Results for “Model Inference”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 1,729 records · Page 96
All sourcesNASA NTRSDOE OSTI
DOE OSTI2024

Identifying Heterogeneous Micromechanical Properties of Biological Tissues via Physics–Informed Neural Networks

The heterogeneous micromechanical properties of biological tissues have profound implications across diverse medical and engineering domains. However, identifying full-field heterogeneous elastic properties of soft materials using traditional engineering approaches is fundamentally challenging due to difficulties in estimating local stress fields. Recently, there has been a growing interest in data-driven models for learning full-field mechanical responses, such as displacement and strain, from experimental or synthetic data. However, research studies on inferring full-field elastic properties of materials, a more challenging problem, are scarce, particularly for large deformation, hyperelastic materials. Here, a physics-informed machine learning approach is proposed to identify the elasticity map in nonlinear, large deformation hyperelastic materials. This study reports the prediction accuracies and computational efficiency of physics-informed neural networks (PINNs) in inferring the heterogeneous elasticity maps across materials with structural complexity that closely resemble real tissue microstructure, such as brain, tricuspid valve, and breast cancer tissues. Further, the improved architecture is applied to three hyperelastic constitutive models: Neo-Hookean, Mooney Rivlin, and Gent. Furthermore, the improved network architecture consistently produces accurate estimations of heterogeneous elasticity maps, even when there is up to 10% noise present in the training data.

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2020

Modeling Stochastic Variability in Multiband Time-series Data

In preparation for the era of time-domain astronomy with upcoming large-scale surveys, we propose a state-space representation of a multivariate damped random walk process as a tool to analyze irregularly-spaced multifilter light curves with heteroscedastic measurement errors. We adopt a computationally efficient and scalable Kalman filtering approach to evaluate the likelihood function, leading to maximum O(k 3 n) complexity, where k is the number of available bands and n is the number of unique observation times across the k bands. This is a significant computational advantage over a commonly used univariate Gaussian process that can stack up all multiband light curves in one vector with maximum O(k 3 n 3 ) complexity. Using such efficient likelihood computation, we provide both maximum likelihood estimates and Bayesian posterior samples of the model parameters. Three numerical illustrations are presented: (i) analyzing simulated five-band light curves for a comparison with independent single-band fits; (ii) analyzing five-band light curves of a quasar obtained from the Sloan Digital Sky Survey Stripe 82 to estimate short-term variability and timescale; (iii) analyzing gravitationally lensed g- and r-band light curves of Q0957+561 to infer the time delay. Two R packages, Rdrw and timedelay, are publicly available to fit the proposed models.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2015

The dependence of the strength and thickness of field‐aligned currents on solar wind and ionospheric parameters

Abstract Sheared plasma flows at the low‐latitude boundary layer (LLBL) correlate well with early afternoon auroral arcs and upward field‐aligned currents. We present a simple analytic model that relates solar wind and ionospheric parameters to the strength and thickness of field‐aligned currents (Λ) in a region of sheared velocity, such as the LLBL. We compare the predictions of the model with DMSP observations and find remarkably good scaling of the upward region 1 currents with solar wind and ionospheric parameters in region located at the boundary layer or open field lines at 1100–1700 magnetic local time. We demonstrate that and Λ ∼ L when Λ/ L < 5 where L is the auroral electrostatic scale length. The sheared boundary layer thickness (Δ m ) is inferred to be around 3000 km, which appears to have weak dependence on V sw . J ∥ has dependencies on Δ m , Σ p , n sw , and V sw . The analytic model provides a simple way to organize data and to infer boundary layer structures from ionospheric data.

Johnson, Jay R.↗
DOE OSTI2021

Discrete-Direct Model Calibration and Uncertainty Propagation Method Confirmed on Multi-Parameter Plasticity Model Calibrated to Sparse Random Field Data

A discrete direct (DD) model calibration and uncertainty propagation approach is explained and demonstrated on a 4-parameter Johnson-Cook (J-C) strain-rate dependent material strength model for an aluminum alloy. The methodology's performance is characterized in many trials involving four random realizations of strain-rate dependent material-test data curves per trial, drawn from a large synthetic population. The J-C model is calibrated to particular combinations of the data curves to obtain calibration parameter sets which are then propagated to “Can Crush” structural model predictions to produce samples of predicted response variability. These are processed with appropriate sparse-sample uncertainty quantification (UQ) methods to estimate various statistics of response with an appropriate level of conservatism. This is tested on 16 output quantities (von Mises stresses and equivalent plastic strains) and it is shown that important statistics of the true variabilities of the 16 quantities are bounded with a high success rate that is reasonably predictable and controllable. The DD approach has several advantages over other calibration-UQ approaches like Bayesian inference for capturing and utilizing the information obtained from typically small numbers of replicate experiments in model calibration situations—especially when sparse replicate functional data are involved like force–displacement curves from material tests. The DD methodology is straightforward and efficient for calibration and propagation problems involving aleatory and epistemic uncertainties in calibration experiments, models, and procedures.

42 ENGINEERING↗
DOE OSTI2024

Constraining the Dense Matter Equation of State with New NICER Mass–Radius Measurements and New Chiral Effective Field Theory Inputs

Pulse profile modeling of X-ray data from the Neutron Star Interior Composition Explorer is now enabling precision inference of neutron star mass and radius. Combined with nuclear physics constraints from chiral effective field theory (χEFT), and masses and tidal deformabilities inferred from gravitational-wave detections of binary neutron star mergers, this has led to a steady improvement in our understanding of the dense matter equation of state (EOS). Here, we consider the impact of several new results: the radius measurement for the 1.42 M ⊙ pulsar PSR J0437−4715 presented by Choudhury et al., updates to the masses and radii of PSR J0740+6620 and PSR J0030+0451, and new χEFT results for neutron star matter up to 1.5 times nuclear saturation density. Using two different high-density EOS extensions—a piecewise-polytropic (PP) model and a model based on the speed of sound in a neutron star (CS)—we find the radius of a 1.4 M ⊙ (2.0 M ⊙ ) neutron star to be constrained to the 95% credible ranges ${12.28}_{-0.76}^{+0.50}$ km (${12.33}_{-1.34}^{+0.70}$ km) for the PP model and ${12.01}_{-0.75}^{+0.56}$ km (${11.55}_{-1.09}^{+0.94}$ km) for the CS model. The maximum neutron star mass is predicted to be ${2.15}_{-0.16}^{+0.14}$M ⊙ and ${2.08}_{-0.16}^{+0.28}$M ⊙ for the PP and CS models, respectively. We explore the sensitivity of our results to different orders and different densities up to which χEFT is used, and show how the astrophysical observations provide constraints for the pressure at intermediate densities. Moreover, we investigate the difference R 2.0 − R 1.4 of the radius of 2 M ⊙ and 1.4 M ⊙ neutron stars within our EOS inference.

gravitational wave sources↗
DOE OSTI2015

Constructing a folding model for protein S6 guided by native fluctuations deduced from NMR structures

The diversity in a set of protein nuclear magnetic resonance (NMR) structures provides an estimate of native state fluctuations that can be used to refine and enrich structure-based protein models (SBMs). Dynamics are an essential part of a protein’s functional native state. The dynamics in the native state are controlled by the same funneled energy landscape that guides the entire folding process. SBMs apply the principle of minimal frustration, drawn from energy landscape theory, to construct a funneled folding landscape for a given protein using only information from the native structure. On an energy landscape smoothed by evolution towards minimal frustration, geometrical constraints, imposed by the native structure, control the folding mechanism and shape the native dynamics revealed by the model. Native-state fluctuations can alternatively be estimated directly from the diversity in the set of NMR structures for a protein. Based on this information, we identify a highly flexible loop in the ribosomal protein S6 and modify the contact map in a SBM to accommodate the inferred dynamics. By taking into account the probable native state dynamics, the experimental transition state is recovered in the model, and the correct order of folding events is restored. Our study highlights how the shared energy landscape connects folding and function by showing that a better description of the native basin improves the prediction of the folding mechanism.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2017

A Semi-analytical Model for Wind-fed Black Hole High-mass X-Ray Binaries: State Transition Triggered by Magnetic Fields from the Companion Star

We propose a mechanism of state transition in wind-fed black hole (BH) binaries (high-mass X-ray binaries) such as Cyg X-1 and LMC X-1. Modeling a line-driven stellar wind from the companion by two-dimensional hydrodynamical calculations, we investigate the processes of wind capture by, and accretion onto, the BH. We assume that the wind acceleration is terminated at the He ii ionization front because ions responsible for line-driven acceleration are ionized within the front, i.e., the He iii region. It is found that the mass accretion rate inferred from the luminosity is remarkably smaller than the capture rate. Considering the difference, we construct a model for the state transition based on the accretion flow being controlled by magnetorotational instability. The outer flow is torus-like, and plays an important role to trigger the transition. The model can explain why state transition does occur in Cyg X-1, while not in LMC X-1. Cyg X-1 exhibits a relatively low luminosity, and then the He ii ionization front is located and can move between the companion and BH, depending on its ionizing photon flux. On the other hand, LMC X-1 exhibits too high luminosity for the front to move considerably; the front is too close to the companion atmosphere. The model also predicts that each state of high-soft or low-hard would last fairly long because the luminosity depends weakly on the wind velocity. In the context of the model, the state transition is triggered by a fluctuation of the magnetic field when its amplitude becomes comparable to the field strength in the torus-like outer flow.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2025

A Bayesian approach to time-domain photonic Doppler velocimetry analysis

Photonic Doppler velocimetry (PDV) is an established technique for measuring the velocities of fast-moving surfaces in high-energy-density experiments. In the standard approach to PDV analysis, the short-time Fourier transform (STFT) is used to generate a spectrogram from which the velocity history of the target is inferred. The user chooses the form, duration, and separation of the window function. Here, in this study, we present a Bayesian approach to infer the velocity directly from the PDV oscilloscope trace, without using the spectrogram for analysis. This is clearly a difficult inference problem due to the highly periodic nature of the data, but we find that with carefully chosen prior distributions for the model parameters, we can accurately recover the injected velocity from synthetic data. We validate this method using PDV data collected at the STAR two-stage light gas gun at Sandia National Laboratories, recovering shock-front velocities in quartz that are consistent with those inferred using the STFT-based approach and are interpolated across regions of low signal-to-noise data. Although this method does not rely on the same user choices as the STFT, we caution that it can be prone to misspecification if the chosen model is not sufficient to capture the velocity behavior. Analysis using posterior predictive checks can be used to establish whether a better model is required, although more complex models come with additional computational cost, often taking more than several hours to converge when sampling the Bayesian posterior. We, therefore, recommend it be viewed as a complementary method to that of the STFT-based approach.

Allison, James R. [First Light Fusion Ltd., Yarnto↗
DOE OSTI2016

Systematic network assessment of the carcinogenic activities of cadmium

Cadmium has been defined as type I carcinogen for humans, but the underlying mechanisms of its carcinogenic activity and its influence on protein-protein interactions in cells are not fully elucidated. The aim of the current study was to evaluate, systematically, the carcinogenic activity of cadmium with systems biology approaches. From a literature search of 209 studies that performed with cellular models, 208 proteins influenced by cadmium exposure were identified. All of these were assessed by Western blotting and were recognized as key nodes in network analyses. The protein-protein functional interaction networks were constructed with NetBox software and visualized with Cytoscape software. These cadmium-rewired genes were used to construct a scale-free, highly connected biological protein interaction network with 850 nodes and 8770 edges. Of the network, nine key modules were identified and 60 key signaling pathways, including the estrogen, RAS, PI3K-Akt, NF-κB, HIF-1α, Jak-STAT, and TGF-β signaling pathways, were significantly enriched. With breast cancer, colorectal and prostate cancer cellular models, we validated the key node genes in the network that had been previously reported or inferred form the network by Western blotting methods, including STAT3, JNK, p38, SMAD2/3, P65, AKT1, and HIF-1α. These results suggested the established network was robust and provided a systematic view of the carcinogenic activities of cadmium in human. - Highlights: • A cadmium-influenced network with 850 nodes and 8770 edges was established. • The cadmium-rewired gene network was scale-free and highly connected. • Nine modules were identified, and 60 key signaling pathways related to cadmium-induced carcinogenesis were found. • Key mediators in the network were validated in multiple cellular models.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2015

Design of experiments and data analysis challenges in calibration for forensics applications

Forensic science aims to infer characteristics of source terms using measured observables. Our focus is on statistical design of experiments and data analysis challenges arising in nuclear forensics. More specifically, we focus on inferring aspects of experimental conditions (of a process to produce product Pu oxide powder), such as temperature, nitric acid concentration, and Pu concentration, using measured features of the product Pu oxide powder. The measured features, Y, include trace chemical concentrations and particle morphology such as particle size and shape of the produced Pu oxide power particles. Making inferences about the nature of inputs X that were used to create nuclear materials having particular characteristics, Y, is an inverse problem. Therefore, statistical analysis can be used to identify the best set (or sets) of Xs for a new set of observed responses Y. One can fit a model (or models) such as Υ = f(Χ) + error , for each of the responses, based on a calibration experiment and then “invert” to solve for the best set of Xs for a new set of Ys. This perspectives paper uses archived experimental data to consider aspects of data collection and experiment design for the calibration data to maximize the quality of the predicted Ys in the forward models; that is, we assume that well-estimated forward models are effective in the inverse problem. In addition, we consider how to identify a best solution for the inferred X, and evaluate the quality of the result and its robustness to a variety of initial assumptions, and different correlation structures between the responses. In addition, we also briefly review recent advances in metrology issues related to characterizing particle morphology measurements used in the response vector, Y.

36 MATERIALS SCIENCE↗
DOE OSTI2025

Using Machine Learning to Generate a GISS ModelE Calibrated Physics Ensemble (CPE)

A neural network (NN) surrogate of the NASA GISS ModelE atmosphere (version E3) is trained on a perturbed parameter ensemble (PPE) spanning 45 physics parameters and 36 outputs. The NN is leveraged in a Markov Chain Monte Carlo (MCMC) Bayesian parameter inference framework to generate a second posterior constrained ensemble coined a “calibrated physics ensemble,” or CPE. The CPE members are characterized by diverse parameter combinations and are, by definition, close to top-of-atmosphere radiative balance, and must broadly agree with numerous hydrologic, energy cycle and radiative forcing metrics simultaneously. Global observations of numerous cloud, environment, and radiation properties (provided by global satellite products) are crucial for CPE generation. The inference framework explicitly accounts for discrepancies (or biases) in satellite products during CPE generation. We demonstrate that product discrepancies strongly impact calibration of important model parameter settings (e.g., convective plume entrainment rates; fall speed for cloud ice). Structural improvements new to E3 are retained across CPE members (e.g., stratocumulus simulation). Notably, the framework improved the simulation of shallow cumulus and Amazon rainfall while not degrading radiation fields, an upgrade that neither default parameters nor Latin Hypercube parameter searching achieved. Analyses of the initial PPE suggested several parameters were unimportant for output variation. However, many “unimportant” parameters were needed for CPE generation, a result that brings to the forefront how parameter importance should be determined in PPEs. From the CPE, two diverse 45-dimensional parameter configurations are retained to generate radiatively-balanced, auto-tuned atmospheres that were used in two E3 submissions to CMIP6.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2021

Wolf-Rayet Galaxies in SDSS-IV MaNGA. II. Metallicity Dependence of the High-mass Slope of the Stellar Initial Mass Function

As hosts of living high-mass stars, Wolf-Rayet (WR) regions or WR galaxies are ideal objects for constraining the high-mass end of the stellar initial mass function (IMF). We construct a large sample of 910 WR galaxies/regions that cover a wide range of stellar metallicity (from Z ~ 0.001 to 0.03) by combining three catalogs of WR galaxies/regions previously selected from the SDSS and SDSS-IV/MaNGA surveys. We measure the equivalent widths of the WR blue bump at ~4650 Å for each spectrum. They are compared with predictions from stellar evolutionary models Starburst99 and BPASS, with different IMF assumptions (high-mass slope α of the IMF ranging from 1.0 to 3.3). Both singular evolution and binary evolution are considered. We also use a Bayesian inference code to perform full spectral fitting to WR spectra with stellar population spectra from BPASS as fitting templates. We then make a model selection among different α assumptions based on Bayesian evidence. These analyses have consistently led to a positive correlation of the IMF high-mass slope α with stellar metallicity Z, i.e., with a steeper IMF (more bottom-heavy) at higher metallicities. Specifically, an IMF with α = 1.00 is preferred at the lowest metallicity (Z ~ 0.001), and an Salpeter or even steeper IMF is preferred at the highest metallicity (Z ~ 0.03). These conclusions hold even when binary population models are adopted.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2020

A Bayesian Approach to Regional Decadal Predictability: Sparse Parameter Estimation in High-Dimensional Linear Inverse Models of High-Latitude Sea Surface Temperature Variability

Stochastic reduced models are an important tool in climate systems whose many spatial and temporal scales cannot be fully discretized or underlying physics may not be fully accounted for. One form of reduced model, the linear inverse model (LIM), has been widely used for regional climate predictability studies—typically focusing more on tropical or midlatitude studies. However, most LIM fitting techniques rely on point estimation techniques deriving from fluctuation–dissipation theory. In this methodological study we explore the use of Bayesian inference techniques for LIM parameter estimation of sea surface temperature (SST), to quantify the skillful decadal predictability of Bayesian LIM models at high latitudes. We show that Bayesian methods, when compared to traditional point estimation methods for LIM-type models, provide better calibrated probabilistic skill, while simultaneously providing better point estimates due to the regularization effect of the prior distribution in high-dimensional problems. We compare the effect of several priors, as well as maximum likelihood estimates, on 1) estimating parameter values on a perfect model experiment and 2) producing calibrated 1-yr SST anomaly forecast distributions using a preindustrial control run of the Community Earth System Model (CESM). Finally, we employ a host of probabilistic skill metrics to determine the extent to which an LIM can forecast SST anomalies at high latitudes. We find that the choice of prior distribution has an appreciable impact on estimation outcomes, and priors that emphasize physically relevant properties enhance the model’s ability to capture variability of SST anomalies.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2022

Combining biomarker and virus phylogenetic models improves HIV-1 epidemiological source identification

To identify and stop active HIV transmission chains new epidemiological techniques are needed. Here, we describe the development of a multi-biomarker augmentation to phylogenetic inference of the underlying transmission history in a local population. HIV biomarkers are measurable biological quantities that have some relationship to the amount of time someone has been infected with HIV. To train our model, we used five biomarkers based on real data from serological assays, HIV sequence data, and target cell counts in longitudinally followed, untreated patients with known infection times. The biomarkers were modeled with a mixed effects framework to allow for patient specific variation and general trends, and fit to patient data using Markov Chain Monte Carlo (MCMC) methods. Subsequently, the density of the unobserved infection time conditional on observed biomarkers were obtained by integrating out the random effects from the model fit. This probabilistic information about infection times was incorporated into the likelihood function for the transmission history and phylogenetic tree reconstruction, informed by the HIV sequence data. To critically test our methodology, we developed a coalescent-based simulation framework that generates phylogenies and biomarkers given a specific or general transmission history. Testing on many epidemiological scenarios showed that biomarker augmented phylogenetics can reach 90% accuracy under idealized situations. Under realistic within-host HIV-1 evolution, involving substantial within-host diversification and frequent transmission of multiple lineages, the average accuracy was at about 50% in transmission clusters involving 5–50 hosts. Realistic biomarker data added on average 16 percentage points over using the phylogeny alone. Using more biomarkers improved the performance. Shorter temporal spacing between transmission events and increased transmission heterogeneity reduced reconstruction accuracy, but larger clusters were not harder to get right. More sequence data per infected host also improved accuracy. We show that the method is robust to incomplete sampling and that adding biomarkers improves reconstructions of real HIV-1 transmission histories. The technology presented here could allow for better prevention programs by providing data for locally informed and tailored strategies.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2019

Parameter Estimation and Uncertainty Quantification for Systems Biology Models

Mathematical models can provide quantitative insights into immunoreceptor signaling, and other biological processes, but require parameterization and uncertainty quantification before reliable predictions become possible. We review currently available methods and software tools to address these problems. We consider gradient-based and gradient-free methods for point estimation of parameter values, and methods of profile likelihood, bootstrapping, and Bayesian inference for uncertainty quantification. We consider recent and potential future applications of these methods to systems-level modeling of immune-related phenomena.

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2022

TDCOSMO. VII. Boxyness/discyness in lensing galaxies: Detectability and impact on H 0

In the context of gravitational lensing, the density profile of lensing galaxies is often considered to be perfectly elliptical. Potential angular structures are generally ignored, except to explain flux ratios of point-like sources (i.e. flux ratio anomalies). Surprisingly, the impact of azimuthal structures on extended images of the source has not been characterised, nor has its impact on the H 0 inference. We address this task by creating mock images of a point source embedded in an extended source and lensed by an elliptical galaxy on which multipolar components are added to emulate boxy or discy isodensity contours. Modelling such images with a density profile free of angular structure allows us to explore the detectability of image deformation induced by the multipoles in the residual frame. Multipole deformations are almost always detectable for our highest signal-to-noise ratio (S/N) mock data. However, the detectability depends on the lens ellipticity and Einstein radius, on the S/N of the data, and on the specific lens modelling strategy. Multipoles also introduce small changes to the time-delays. We therefore quantify how undetected multipoles would impact H 0 inference. When no multipoles are detected in the residuals, the impact on H 0 for a given lens is in general less than a few k ms -1 Mpc -1 , but in the worst-case scenario, combining low S/N in the ring and large intrinsic boxyness or discyness, the bias on H0 can reach 10-12 km s -1 Mpc -1 . If we now look at the inference on H 0 from a population of lensing galaxies with a distribution of multipoles representative of what is found in the light profile of elliptical galaxies, we find a systematic bias on H 0 of less than 1%. A comparison of our mock systems to the state-of-the-art time-delay lens sample studied by the H0LiCOW and TDCOSMO collaborations indicates that multipoles are currently unlikely to be a source of substantial systematic bias on the inferred value of H 0 from time-delay lenses.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2018

Errors in nanoparticle growth rates inferred from measurements in chemically reacting aerosol systems

In systems in which aerosols are being formed by chemical transformations,individual particles grow due to the addition of molecular species. Efforts to improve our understanding of particle growth often focus on attempts to reconcile observed growth rates with values calculated from models. However,because it is typically not possible to measure the growth rates of individual particles in chemically reacting systems, they must be inferred from measurements of aerosol properties such as size distributions, particle number concentrations, etc. Furthermore this work discusses errors in growth rates obtained using methods that are commonly employed for analyzing atmospheric data. We analyze “data” obtained by simulating the formation of aerosols in a system in which a single chemical species is formed at a constant rate, R . We show that the maximum overestimation error in measured growth rates occurs for collision-controlled nucleation in a single-component system in the absence of a preexisting aerosol, wall losses, evaporation or dilution,as this leads to the highest concentrations of nucleated particles. Those high concentrations lead to high coagulation rates that cause the nucleation mode to grow faster than would be caused by vapor condensation alone. We also show that preexisting particles, when coupled with evaporation, can significantly decrease the concentration of nucleated particles. This can lead to decreased discrepancies between measured growth rate and true growth rate by reducing coagulation among nucleated particles. However, as particle sink processes become stronger, measured growth rates can potentially be lower than true particle growth rates. We briefly discuss nucleation scenarios in which the observed growth rate approaches zero while the true growth rate does not.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2020

On-the-fly active learning of interpretable Bayesian force fields for atomistic rare events

Abstract Machine learned force fields typically require manual construction of training sets consisting of thousands of first principles calculations, which can result in low training efficiency and unpredictable errors when applied to structures not represented in the training set of the model. This severely limits the practical application of these models in systems with dynamics governed by important rare events, such as chemical reactions and diffusion. We present an adaptive Bayesian inference method for automating the training of interpretable, low-dimensional, and multi-element interatomic force fields using structures drawn on the fly from molecular dynamics simulations. Within an active learning framework, the internal uncertainty of a Gaussian process regression model is used to decide whether to accept the model prediction or to perform a first principles calculation to augment the training set of the model. The method is applied to a range of single- and multi-element systems and shown to achieve a favorable balance of accuracy and computational efficiency, while requiring a minimal amount of ab initio training data. We provide a fully open-source implementation of our method, as well as a procedure to map trained models to computationally efficient tabulated force fields.

36 MATERIALS SCIENCE↗