Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “model interpretability”

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 739 records · Page 41
All sourcesNASA NTRSDOE OSTI
DOE OSTI2024

Tokamak divertor plasma emulation with machine learning

Abstract Future tokamak devices that aim to create conditions relevant to power plant operations must consider strategies for mitigating damage to plasma facing components in the divertor. One of the goals of MAST-U tokamak operations is to inform these considerations by researching advanced divertor configurations that aid stable plasma detachment. Machine design, scenario planning and detachment control would all greatly benefit from tools that enable rapid calculation of scenario-relevant quantities given some input parameters. This paper presents a method for generating large, simulated scrape-off layer data sets, which was applied to generate a data set of steady-state Hermes-3 simulations of the MAST-U tokamak. A machine learning model was constructed using a Bayesian approach to hyperparameter optimisation to predict diagnosable output quantities given control-relevant input features. The resulting best-performing model, which is based on a feedforward neural network, achieves high accuracy when predicting electron temperature at the divertor target and carbon impurity radiation front position and runs in around 1 ms in inference mode. Techniques for interpreting the predictions made by the model were applied, and a high-resolution parameter scan of upstream conditions was performed to demonstrate the utility of rapidly generating accurate predictions using the emulator. This work represents a step forward in the design of machine learning-driven emulators of tokamak exhaust simulation codes in operational modes relevant to divertor detachment control and plasma scenario design.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2021

Identifying Key Drivers of Wildfires in the Contiguous US Using Machine Learning and Game Theory Interpretation

Abstract Understanding the complex interrelationships between wildfire and its environmental and anthropogenic controls is crucial for wildfire modeling and management. Although machine learning (ML) models have yielded significant improvements in wildfire predictions, their limited interpretability has been an obstacle for their use in advancing understanding of wildfires. This study builds an ML model incorporating predictors of local meteorology, land‐surface characteristics, and socioeconomic variables to predict monthly burned area at grid cells of 0.25° × 0.25° resolution over the contiguous United States. Besides these predictors, we construct and include predictors representing the large‐scale circulation patterns conducive to wildfires, which largely improves the temporal correlations in several regions by 14%–44%. The Shapley additive explanation is introduced to quantify the contributions of the predictors to burned area. Results show a key role of longitude and latitude in delineating fire regimes with different temporal patterns of burned area. The model captures the physical relationship between burned area and vapor pressure deficit, relative humidity (RH), and energy release component (ERC), in agreement with the prior findings. Aggregating the contribution of predictor variables of all the grids by region, analyses show that ERC is the major contributor accounting for 14%–27% to large burned areas in the western US. In contrast, there is no leading factor contributing to large burned areas in the eastern US, although large‐scale circulation patterns featuring less active upper‐level ridge‐trough and low RH two months earlier in winter contribute relatively more to large burned areas in spring in the southeastern US.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2022

Non‐Linear Dimensionality Reduction With a Variational Encoder Decoder to Understand Convective Processes in Climate Models

Abstract Deep learning can accurately represent sub‐grid‐scale convective processes in climate models, learning from high resolution simulations. However, deep learning methods usually lack interpretability due to large internal dimensionality, resulting in reduced trustworthiness in these methods. Here, we use Variational Encoder Decoder structures (VED), a non‐linear dimensionality reduction technique, to learn and understand convective processes in an aquaplanet superparameterized climate model simulation, where deep convective processes are simulated explicitly. We show that similar to previous deep learning studies based on feed‐forward neural nets, the VED is capable of learning and accurately reproducing convective processes. In contrast to past work, we show this can be achieved by compressing the original information into only five latent nodes. As a result, the VED can be used to understand convective processes and delineate modes of convection through the exploration of its latent dimensions. A close investigation of the latent space enables the identification of different convective regimes: (a) stable conditions are clearly distinguished from deep convection with low outgoing longwave radiation and strong precipitation; (b) high optically thin cirrus‐like clouds are separated from low optically thick cumulus clouds; and (c) shallow convective processes are associated with large‐scale moisture content and surface diabatic heating. Our results demonstrate that VEDs can accurately represent convective processes in climate models, while enabling interpretability and better understanding of sub‐grid‐scale physical processes, paving the way to increasingly interpretable machine learning parameterizations with promising generative properties.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2020

Context for interpreting equilibrium climate sensitivity and transient climate response from the CMIP6 Earth system models

For the current generation of earth system models participating in the Coupled Model Intercomparison Project Phase 6 (CMIP6), the range of equilibrium climate sensitivity (ECS, a hypothetical value of global warming at equilibrium for a doubling of CO 2 ) is 1.8°C to 5.6°C, the largest of any generation of models dating to the 1990s. Meanwhile, the range of transient climate response (TCR, the surface temperature warming around the time of CO 2 doubling in a 1% per year CO 2 increase simulation) for the CMIP6 models of 1.7°C (1.3°C to 3.0°C) is only slightly larger than for the CMIP3 and CMIP5 models. Here we review and synthesize the latest developments in ECS and TCR values in CMIP, compile possible reasons for the current values as supplied by the modeling groups, and highlight future directions. Cloud feedbacks and cloud-aerosol interactions are the most likely contributors to the high values and increased range of ECS in CMIP6.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2021

Uncovering electronic and geometric descriptors of chemical activity for metal alloys and oxides using unsupervised machine learning

Here, we show that unsupervised machine learning (ML) using principal component analysis (PCA) provides a straightforward pathway for developing accurate and interpretable electronic-structure descriptors of the chemical and catalytic properties of materials. We demonstrate the approach by finding chemisorption descriptors for metal alloys and surface oxygens on metals and metal oxides. In both cases, the principal component (PC) descriptors yield ML models that predict the material’s chemical properties with competitive accuracy compared to ML models built using established descriptors. Importantly, interpreting the electronic-structure patterns captured by each PC descriptor via signal reconstruction suggests potential design motifs for future electronic-structure descriptor design and allows us to identify links between a material’s geometric and catalytic properties. Ultimately, we show that the unsupervised ML approach provides a route to find electronic-structure descriptors of the catalytic properties of materials that readily connect to geometric structure and composition.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2025

A Representation Fusion Framework for Decoupling Diagnostic Information in Multimodal Learning

Modern medicine increasingly relies on multimodal data, ranging from clinical notes to imaging and genomics, to guide diagnosis and treatment. However, integrating these heterogeneous data sources in a principled and interpretable manner remains a major challenge. We present MODES (Multi-mOdal Disentangled Embedding Space), a representation fusion framework that explicitly separates shared and modality-specific factors of variation, offering a structured latent space for multimodal information that improves both prediction and interpretability. By leveraging pre-trained unimodal foundation models, MODES mitigates the dependency on extensive paired datasets, crucial in data-scarce clinical settings. We introduce a masking strategy that optimizes representation dimensionality by eliminating low-information dimensions, to achieve compact, information-rich representations. Our framework demonstrates superior performance in predicting diagnoses and phenotypes compared to unimodal and conventional fusion models. MODES also enables robust diagnostic inference in missing data scenarios, offering an opportunity toward interpretable and efficient multimodal diagnostics in personalized healthcare.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2021

Chemical insights into the multi-regime low-temperature oxidation of di-n-propyl ether: Jet-stirred reactor experiments and kinetic modeling

To further understand the combustion characteristics and the reaction pathways of acyclic ethers, in this work the oxidation of di-n-propyl ether (DPE) was investigated in a jet-stirred reactor (JSR) combined with a photoionization molecular-beam mass spectrometer. The experiments were carried out at near-atmospheric pressure (700 Torr) and over a temperature range of 425–850 K. Based on the experimental data and previous studies on ether oxidation, a new kinetic model was constructed and used to interpret the oxidation chemistry of DPE. In DPE oxidation, a high reactivity at low temperatures and two negative temperature coefficient (NTC) zones were observed. These behaviors are explained in this work by taking advantage of the obtained species information and the modeling analyses: the two NTC zones are caused by the competition of chain branching and termination reactions of the fuel itself and specific oxidation intermediates, respectively. Furthermore, the general requirements to have double-NTC behavior are discussed. A variety of crucial fuel-specific C 6 species, such as ketohydroperoxides and diones, were detected in the species pool of DPE oxidation. Their formation pathways are illuminated based on rate-of-production (ROP) analyses. Propanal was identified as the most abundant small molecule intermediate, and its related reactions have an important impact on the oxidation process of DPE. Both acetic acid and propionic acid were detected in high concentrations. A new formation pathway of propionic acid is proposed and incorporated into the kinetic model to achieve a more accurate prediction for propionic acid mole fractions.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2023

Influence of aluminum incorporation and aqueous conditions on metal ion release of high-Ni transition metal oxide nanomaterials

Developing a materials perspective of how to control the degradation and negative impact of complex metal oxides requires an integrated understanding of how these nanomaterials transform in the environment and interact with biological systems. Doping with aluminum is known to stabilize oxide materials, but has not been assessed cohesively from synthesis to environmental fate and biological impact. In the present study, the influence of aluminum doping on metal ion release from transition metal oxides was investigated by comparing aqueous transformations of lithium nickel cobalt aluminum oxide (LiNi 0.82 Co 0.15 Al 0.03 O 2 ; NCA) and lithium nickel cobalt oxide (LiNi 0.80 Co 0.20 O 2 ; NC) nanoparticles and by calculating the energetics of metal release using a density functional theory (DFT) and thermodynamics method. Two model environmental organisms were used to assess biological impact, and metal ion release was compared for NCA and NC nanoparticles incubated in their respective growth media: moderately hard reconstituted water (MHRW) for the freshwater invertebrate Daphnia magna (D. magna) and minimal growth medium for the Gram-negative bacterium Shewanella oneidensis MR-1 (S. oneidensis). Further, the amount of metal ions released was reduced for NCA compared to NC in MHRW, which correlated to changes in the modeled energetics of release upon Al substitution in the lattice. In minimal medium, metal ion release was approximately an order of magnitude higher compared to MHRW and was similar to the stoichiometry of the bulk nanoparticles for both NCA and NC. Interpretation of the release profiles and modeling indicated that the increase in total metal ion release and the reduced influence of Al doping arises from lactate complexation of metal ions in solution. The relative biological impacts of NC and NCA exposure for both S. oneidensis and D. magna were consistent with the metal release trends observed for minimal medium and MHRW, respectively. Together, these results demonstrate how a combined experimental and computational approach provides valuable insight into the aqueous transformations and biological impacts of complex metal oxide nanoparticles.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2022

Training calibration-based counterfactual explainers for deep learning models in medical image analysis

The rapid adoption of artificial intelligence methods in healthcare is coupled with the critical need for techniques to rigorously introspect models and thereby ensure that they behave reliably. This has led to the design of explainable AI techniques that uncover the relationships between discernible data signatures and model predictions. In this context, counterfactual explanations that synthesize small, interpretable changes to a given query while producing desired changes in model predictions have become popular. This under-constrained, inverse problem is vulnerable to introducing irrelevant feature manipulations, particularly when the model’s predictions are not well-calibrated. Hence, in this paper, we propose the TraCE (training calibration-based explainers) technique, which utilizes a novel uncertainty-based interval calibration strategy for reliably synthesizing counterfactuals. Given the wide-spread adoption of machine-learned solutions in radiology, our study focuses on deep models used for identifying anomalies in chest X-ray images. Using rigorous empirical studies, we demonstrate the superiority of TraCE explanations over several state-of-the-art baseline approaches, in terms of several widely adopted evaluation metrics. Our findings show that TraCE can be used to obtain a holistic understanding of deep models by enabling progressive exploration of decision boundaries, to detect shortcuts, and to infer relationships between patient attributes and disease severity.

62 RADIOLOGY AND NUCLEAR MEDICINE↗
DOE OSTI2024

Mechanisms for Microseismicity Occurrence Due to CO 2 Injection at Decatur, Illinois: A Coupled Multiphase Flow and Geomechanics Perspective

Here, we numerically investigate the mechanisms that resulted in induced seismicity occurrence associated with CO 2 injection at the Illinois Basin–Decatur Project (IBDP). We build a geologically consistent model that honors key stratigraphic horizons and 3D fault surfaces interpreted using surface seismic data and microseismicity locations. We populate our model with reservoir and geomechanical properties estimated using well-log and core data. We then performed coupled multiphase flow and geomechanics modeling to investigate the impact of CO 2 injection on fault stability using the Coulomb failure criteria. We calibrate our flow model using measured reservoir pressure during the CO 2 injection phase. Our model results show that pore-pressure diffusion along faults connecting the injection interval to the basement is essential to explain the destabilization of the regions where microseismicity occurred, and that poroelastic stresses alone would result in stabilization of those regions. Slip tendency analysis indicates that, due to their orientations with respect to the maximum horizontal stress direction, the faults where the microseismicity occurred were very close to failure prior to injection. These model results highlight the importance of accurate subsurface fault characterization for CO 2 sequestration operations.

58 GEOSCIENCES↗
DOE OSTI2026

Fantômas unconfined: global QCD fits with Bézier parameterizations

Fantômas is a C++ toolkit for exploring the parametrization dependence of parton distribution functions (PDFs) and other correlator functions in quantum chromodynamics (QCD). Fantômas facilitates the generation of adaptable polynomial parametrizations for PDFs, called metamorphs, to find best-fit PDF solutions and quantify the epistemic uncertainty associated with the parametrizations during their fitting. The method employs Bézier curves as universal approximators for a variety of PDF shapes. Integrated into the xFitter framework for the global QCD analysis, Fantômas provides a foundation for general models of PDFs, while reducing the computational time compared to the approaches utilizing traditional polynomial parametrizations as well as providing an interpretable alternative to neural-network-based models. This paper outlines the structure and practical usage of the Fantômas toolkit, including its inputs, outputs, and implementation within xFitter. It also provides a practical example of using Fantômas for uncertainty quantification as well as the combination of PDF fits into a single ensemble.

Bézier curves↗
DOE OSTI2022

Nuclear's Role in the U.S. Electricity System: A Multi-Model Inter-Comparison Analysis

Multiple capacity expansion models (CEMs) for the U.S. power system represent the balance of options among generation, transmission, and storage assets that can satisfy electric loads, operating and planning reserves, and policy requirements. These models are typically set up to find the least-cost portfolio of assets that meet specified requirements, and model decisions can include both investments in new, and retirement of existing, resources. The scenarios explored by CEMs can help inform strategies for meeting future electricity and energy needs under a range of future conditions. However, projections can differ between models, sometimes dramatically, for a seemingly similar scenario. Differences in model coverage, structure, and input assumptions contribute to the range of model outcomes. Understanding what drives the biggest differences in model outputs improves model insights and provides context for interpreting results. This summary presents analysis that was performed through a forum of analysts who own, update, and apply CEMs, as well as nuclear experts from national laboratories, industry, and the research community. The following sections describe methods, results, and findings from an original, innovative inter-model comparison that provides insights into what drives the greatest differences in nuclear retirement and deployment projections across models and a range of technology, market, and policy conditions.

capacity expansion model↗
DOE OSTI2023

Initial overview, description, and assessment of the DELFIC atmospheric transport model

We present an overview and description of the atmospheric transport model (ATM) component of the Defense Land Fallout Interpretative Code (DELFIC) and its revision history. In particular, we discuss the hybrid Eulerian-Lagrangian model of the DELFIC diffusive transport module (DTM), its assumptions and constraints, and assess its capability for modeling the transport, dispersion, and deposition of nuclear fallout debris. This hybrid model is an efficient transport model for the wide range of particle sizes found in nuclear fallout debris. Furthermore, we validate the 1979 version of DELFIC using the 1979 test case input and output presented in its documentation. Except for some rounding differences and exposure rate differences from the OPM, the DTM output is in very good agreement with the 1979 results. Overall, the DELFIC DTM provides a reasonable and rapid approximation to the transport and deposition of local nuclear fallout debris.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2022

Model-based compositional predictions for a differential scanning calorimetry/thermogravimetric analysis-mass spectrometry system used for heat of vaporization measurements

A Differential Scanning Calorimetry/Thermogravimetric Analysis (DSC/TGA) procedure for determination of Heat of Vaporization (HOV) of fuel and surrogate fuel mixtures has been previously utilized for evaluation of oxygenate impacts on spark ignition fuels. This analysis has been further leveraged with mass spectrometry (MS) to elucidate vapor phase compositional changes during evaporation. In this study we focus on a simple system of three compounds (ethanol, n-hexane, and n-octane) to provide a model-based understanding of the evaporation process under these experimental conditions. An Aspen Plus model was set up to provide liquid and vapor phase concentration predictions during mixture evaporation using the initial mixture composition, and measured temperature and mass loss profiles from DSC/TGA experiments. Agreements between the trends of the MS ion count profiles and model composition predictions provide a degree of validation of this modeling approach that can allow the interpretation and extrapolation of DSC/TGA measurements for more complex fuel mixtures.

33 ADVANCED PROPULSION SYSTEMS↗
DOE OSTI2022

Revisiting stellar properties of star-forming galaxies with stellar and nebular spectral modelling

Spectral synthesis is a powerful tool for interpreting the physical properties of galaxies by decomposing their spectral energy distributions (SEDs) into the main luminosity contributors (e.g. stellar populations of distinct age and metallicity or ionised gas). However, the impact nebular emission has on the inferred properties of star-forming (SF) galaxies has been largely overlooked over the years, with unknown ramifications to the current understanding of galaxy evolution. The objective of this work is to estimate the relations between stellar properties (e.g. total mass, mean age, and mean metallicity) of SF galaxies by simultaneously fitting the stellar and nebular continua and comparing them to the results derived through the more common purely stellar spectral synthesis approach. The main galaxy sample from SDSS DR7 was analysed with two distinct population synthesis codes: FADO, which estimates self-consistently both the stellar and nebular contributions to the SED, and the original version of STARLIGHT, as representative of purely stellar population synthesis codes. Differences between codes regarding average mass, mean age and mean metallicity values can go as high as ~0.06 dex for the overall population of galaxies and ~0.12 dex for SF galaxies (galaxies with EW(Hα) > 3 Å), with the most prominent difference between both codes in the two populations being in the light-weighted mean stellar age. FADO presents a broader range of mean stellar ages and metallicities for SF galaxies than STARLIGHT, with the latter code preferring metallicity solutions around the solar value (Z ⊙ = 0.02). A closer look into the average light- and mass-weighted star formation histories of intensively SF galaxies (EW(Hα) > 75 Å) reveals that the light contributions of simple stellar populations (SSPs) younger than ≤10 7 (10 9 ) years in STARLIGHT are higher by ~5.41% (9.11%) compared to FADO. Moreover, FADO presents higher light contributions from SSPs with metallicity ≤ Z ⊙ /200 (Z ⊙ /50) of around 8.05% (13.51%) when compared with STARLIGHT. This suggests that STARLIGHT is underestimating the average light-weighted age of intensively SF galaxies by up to ~0.17 dex and overestimating the light-weighted metallicity by up to ~0.13 dex compared to FADO (or vice versa). The comparison between the average stellar properties of passive, SF and intensively SF galaxy samples also reveals that differences between codes increase with increasing EW(Hα) and decreasing total stellar mass. Moreover, comparing SF results from FADO in a purely stellar mode with the previous results qualitatively suggests that differences between codes are primarily due to mathematical and statistical differences and secondarily due to the impact of the nebular continuum modelling approach (or lack thereof). However, it is challenging to adequately quantify the relative role of each factor since they are likely interconnected. This work finds indirect evidence that a purely stellar population synthesis approach negatively impacts the inferred stellar properties (e.g. mean age and mean metallicity) of galaxies with relatively high star formation rates (e.g. dwarf spirals, ‘green peas’, and starburst galaxies). In turn, this can bias interpretations of fundamental relations such as the mass-age or mass-metallicity, which are factors worth bearing in mind in light of future high-resolution spectroscopic surveys at higher redshifts (e.g. MOONS and 4MOST-4HS).

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2023

Search for a high-mass dimuon resonance produced in association with b quark jets at $ \sqrt{s} $ = 13 TeV

A search for high-mass dimuon resonance production in association with one or more b quark jets is presented. The study uses proton-proton collision data collected with the CMS detector at the LHC corresponding to an integrated luminosity of 138 fb$^{−1}$ at a center-of-mass energy of 13 TeV. Model-independent limits are derived on the number of signal events with exactly one or more than one b quark jet. Results are also interpreted in a lepton-flavor-universal model with Z′ boson couplings to a bb quark pair (g$_{b}$), an sb quark pair (g$_{b}$δ$_{bs}$), and any same-flavor charged lepton (g$_{ℓ}$) or neutrino pair (g$_{ν}$), with |g$_{ν}$| = |g$_{ℓ}$|. For a Z′ boson with a mass $ {m}_{{\textrm{Z}}^{\prime }} $ = 350 GeV (2 TeV) and |δ$_{bs}$| < 0.25, the majority of the parameter space with 0.0057 < |g$_{ℓ}$| < 0.35 (0.25 < |g$_{ℓ}$| < 0.43) and 0.0079 < |g$_{b}$| < 0.46 (0.34 < |g$_{b}$| < 0.57) is excluded at 95% confidence level. Finally, constraints are set on a Z′ model with parameters consistent with low-energy b → sℓℓ measurements. In this scenario, most of the allowed parameter space is excluded for a Z′ boson with 350 < $ {m}_{{\textrm{Z}}^{\prime }} $ < 500 GeV, while the constraints are less stringent for higher $ {m}_{{\textrm{Z}}^{\prime }} $ hypotheses. This is the first dedicated search at the LHC for a high-mass dimuon resonance produced in association with multiple b quark jets, and the constraints obtained on models with this signature are the most stringent to date.[graphic not available: see fulltext]

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
DOE OSTI2022

Comprehensive compartmental model and calibration algorithm for the study of clinical implications of the population-level spread of COVID-19: a study protocol

The complex dynamics of the coronavirus disease 2019 (COVID-19) pandemic has made obtaining reliable long-term forecasts of the disease progression difficult. Simple mechanistic models with deterministic parameters are useful for short-term predictions but have ultimately been unsuccessful in extrapolating the trajectory of the pandemic because of unmodelled dynamics and the unrealistic level of certainty that is assumed in the predictions. We propose a 22-compartment epidemiological model that includes compartments not previously considered concurrently, to account for the effects of vaccination, asymptomatic individuals, inadequate access to hospital care, post-acute COVID-19 and recovery with long-term health complications. Additionally, new connections between compartments introduce new dynamics to the system and provide a framework to study the sensitivity of model outputs to several concurrent effects, including temporary immunity, vaccination rate and vaccine effectiveness. Subject to data availability for a given region, we discuss a means by which population demographics (age, comorbidity, socioeconomic status, sex and geographical location) and clinically relevant information (different variants, different vaccines) can be incorporated within the 22-compartment framework. Considering a probabilistic interpretation of the parameters allows the model's predictions to reflect the current state of uncertainty about the model parameters and model states. We propose the use of a sparse Bayesian learning algorithm for parameter calibration and model selection. This methodology considers a combination of prescribed parameter prior distributions for parameters that are known to be essential to the modelled dynamics and automatic relevance determination priors for parameters whose relevance is questionable. This is useful as it helps prevent overfitting the available epidemiological data when calibrating the parameters of the proposed model. Population-level administrative health data will serve as partial observations of the model states.

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2023

Graph-EAM: An Interpretable and Efficient Graph Neural Network Potential Framework

The development of deep learning interatomic potentials has enabled efficient and accurate computations in quantum chemistry and materials science, circumventing computationally expensive ab initio calculations. However, the huge number of learnable parameters in deep learning models and their complex architectures hinder physical interpretability and affect the robustness of the derived potential. In this work, we propose graph-EAM, a lightweight graph neural network (GNN) inspired by the empirical embedded atom method to model the interatomic potential of single-element structures. Four material systems: platinum, niobium, silicon, and amorphous-carbon, for which quantum simulation data sets are publicly available, are examined to demonstrate that graph-EAM can achieve high energy and force prediction accuracy-comparable or better than existing state-of-the-art machine learning models-with much fewer parameters. It is also shown that the explicit inclusion of the angular information via three-body atomic density increases the prediction accuracy. In conclusion, the accuracy and efficiency of potentials obtained from graph-EAM can help accelerate the molecular dynamics simulation.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗