Search NASA⌕ Search

SEARCH · Search NASA

Results for “ML”

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 199 records · Page 11
All sourcesNASA NTRSDOE OSTI
DOE OSTI2024

Temperature-dependent mechanical properties and crystal plasticity parameters for additively manufactured Haynes-214 alloy: Experiments and numerical modeling

Our experimental mechanical testing data demonstrated that the additively manufactured (AM) laser powder bed fusion (L-PBF) Haynes-214 alloy exhibits non-linear mechanical properties as the temperature rises from ambient to 870 °C. Crystal plasticity (CP) simulations provide an effective approach to gaining deeper insights into microstructure-property linkages under thermomechanical loading. This method can reduce the need for costly high-temperature mechanical testing while accounting for the effects of crystallographic texture and grain morphology on the mechanical behavior of AM materials. However, calibrating a CP model is time-consuming because individual simulations are computationally expensive and hundreds (or more) of iterations over parameter sets may be required. To address this issue, we have designed a machine learning-differential evolution (ML-DE) CP framework that can accurately interpolate the tensile properties of AM L-PBF Haynes-214 alloy across a wide temperature range from ambient to 870 °C, with minimal reliance on experimental data. The framework uses electron backscatter diffraction (EBSD) measurements to generate statistically equivalent microstructural volume elements to serve as inputs to the CP modeling framework. Stress–strain curves were generated from 1000 CP simulations, which serve as the training data set for the three ML regression algorithms explored: linear, extra-trees, and multi-layer perceptron. These three regression models were independently evaluated to compare their efficiency and identify the most suitable algorithm for the given problem. Results revealed that the extra-trees ML regressor outperforms the other models in both qualitative and quantitative aspects with an R 2 of 0.98. Subsequently, the differential evolution optimization approach is employed to calibrate the ML-based CP material parameters with experimental results obtained at various temperatures. Finally, temperature-dependent CP material parameters are formulated. The effectiveness and efficiency of the designed framework are validated through comparison with experimental results, demonstrating a high degree of agreement. These calibrated parametric constitutive equations enable further use of the CP model to study the deformation behavior of this alloy under a wide range of thermo-mechanical loading conditions.

36 MATERIALS SCIENCE↗
DOE OSTI2025

Generalizable machine learning potentials for quantum-accurate predictions of non-equilibrium behavior in 2D materials

Machine learning interatomic potentials (ML-IAPs) are emerging as transformative tools in materials modeling, promising quantum-level accuracy at a fraction of the computational cost. However, their ability to generalize beyond equilibrium configurations and to reliably capture defect- and temperature-driven behavior remains underexplored. Here, we develop and benchmark two state-of-the-art ML-IAPs, Spectral Neighbor Analysis Potential (SNAP) and Allegro, on a comprehensive dataset for monolayer MoSe₂. Using density functional theory (DFT) as the reference, we evaluate their performance in capturing stress–strain behavior, phase transition energetics, defect evolution, edge stability, and fracture toughness. Allegro, a deep equivariant neural network potential, surpasses both SNAP and the classical Tersoff potential in accuracy, efficiency, and transferability. Importantly, both ML potentials accurately reproduce experimental fracture measurements and ab initio predictions of inversion domain formation—phenomena well beyond their training sets. Our findings establish ML-IAPs as viable replacements for traditional force fields in the study of non-equilibrium mechanical phenomena, enabling large-scale, high-fidelity simulations in 2D materials and beyond. In conclusion, this work provides a broadly applicable framework for data-driven modeling of structural and functional transformations under extreme conditions.

2D materials↗
DOE OSTI2025

HydroEcoLSTM: A Python package with graphical user interface for hydro-ecological modeling with long short-term memory neural network

Machine learning (ML) is emerging as a promising tool for modeling hydro-ecological processes due to the increasing availability of large environmental data. However, the use of ML requires sufficient programming knowledge due to a lack of a graphical user interface (GUI). In this study, we introduced a GUI package, named HydroEcoLSTM, with the long short-term memory network (LSTM) as the core model, that allows non-ML experts to utilize their domain knowledge to construct complex ML models. We demonstrated the functionalities of HydroEcoLSTM with two practical examples, including (1) predictions of streamflow in both gauged and ungauged catchments and (2) predictions of multiple outputs (i.e., streamflow and isotope transport from two catchments). The simulation results obtained in both case experiments are satisfactory. In the first example, the average Nash–Sutcliffe Efficiency (NSE) for streamflow simulation during the testing period is 0.79 while the application of the trained model in two assumed ungauged catchments also achieves the average NSE of 0.68. In the second example, the average NSE for streamflow and instream isotope simulation during the testing period is 0.71. Ultimately, applications of HydroEcoLSTM with real-world examples demonstrate its potential use for practical applications and research without requiring extensive coding skills.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2024

A machine-learning-aided data recovery approach for predicting multi-material thermal behaviors in advanced test reactor capsules

Instrumented experiments conducted at test reactors are essential to the deployment of new advanced reactor systems. Designing new experiments and generating data on specific reactor conditions require significant investments in terms of both time and cost. Finite element analysis software can be used to create high-fidelity models of experiment environments in order to support the actual experiments, but computation time remains a concern in terms of applying outcomes to real-time usage of data (e.g., a digital twin [DT]). Here, the present research proposes a machine-learning (ML) aided approach to making temperature and displacement predictions based on the thickness of the outer gas gap on the experimental capsule used for in-pile demonstration of a novel new thermal conductivity probe in the Advanced Test Reactor (ATR). This capsule consisted of U10Zr fuel, a rodlet, sodium, and inner and outer capsules. Gas gaps existed between the fuel and the rodlet, and between the inner and the outer capsule. The learning data pertained to an experimental capsule's radial distributions of temperature and displacement, as obtained based on Abaqus and the physical features. For the first step of ML sequence, the temperature was predicted using three positional parameters. Next, the displacement was predicted using seven additional parameters. Each physical feature was normalized in order to be both nondimensional and standardized. The temperature and displacement predictions showed good agreement with the simulation results in all cases involving interpolation and extrapolation. Furthermore, data similarity enhancement increased the similarity between the training and the target data, thereby increasing the predictive accuracy of the ML models. In certain extrapolation cases involving limited original ML model accuracy, data similarity enhancement and data recovery was able to somewhat improve this accuracy.

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗
DOE OSTI2026

Machine-learning modeling of magnetization dynamics in quasi-equilibrium and driven metallic spin systems

Here, we present a perspective on recent progress in machine-learning (ML) force-field approaches for large-scale Landau–Lifshitz–Gilbert (LLG) simulations of metallic spin systems. Building on a generalization of the Behler–Parrinello (BP) architecture originally developed for quantum molecular dynamics, we develop scalable and transferable ML models that faithfully capture the complex, environment-dependent electron-mediated exchange fields characteristic of itinerant magnets. A central ingredient of this framework is the implementation of symmetry-aware magnetic descriptors based on group-theoretical bispectrum formalisms. Leveraging these ML force fields, LLG simulations faithfully reproduce hallmark non-collinear magnetic orders—such as the 120° and tetrahedral states—on the triangular lattice, and successfully capture the complex spin textures emerging in the mixed-phase states of a square-lattice double-exchange model under thermal quench. We further discuss a generalized potential theory that extends the BP formalism to incorporate both conservative and nonconservative electronic torques, thereby enabling ML models to learn nonequilibrium exchange fields from computationally demanding microscopic approaches such as nonequilibrium Green’s-function techniques. This extension yields quantitatively accurate predictions of voltage-driven domain-wall motion and establishes a foundation for quantum-accurate, multiscale modeling of nonequilibrium spin dynamics and spintronic functionalities.

Descriptors↗
DOE OSTI2024

Glass Design Using Machine Learning Property Models with Prediction Uncertainties: Nuclear Waste Glass Formulation

The United States Department of Energy is responsible for managing the legacy nuclear waste stored in underground tanks at the Hanford Site. The waste will be separately vitrified as low-activity waste and high-level waste fractions. Waste glass formulation algorithms have been traditionally developed using partial quadratic mixture property-composition models. Recently, machine learning (ML) techniques have been used to predict glass properties and discover new glass materials for nuclear waste vitrification, and these advancements can be utilized to improve waste glass composition design. In this proof-of-principle study, ML algorithms such as Gaussian process regression (GPR) were used to interpolate glass properties (e.g., viscosity, electrical conductivity, chemical durability). After selecting appropriate sets of GPR hyper-parameters for each property, an optimization program was developed to formulate glass compositions to maximize waste loading while simultaneously satisfying property within constraints. The results of the ML-based waste loadings and glass compositions were compared to those obtained using the traditional methods. Comparing to the previous glass design framework, the ML-based optimization methods offer improved glass designs and a streamlined approach to generation of optimally designed data and near real-time updates.

glass formulation, machine learning, constraints, ↗
DOE OSTI2024

Modeling injection-induced fault slip using long short-term memory networks

Stress changes due to changes in fluid pressure and temperature in a faulted formation may lead to the opening/shearing of the fault. This can be due to subsurface (geo)engineering activities such as fluid injections and geologic disposal of nuclear waste. Such activities are expected to rise in the future making it necessary to assess their short- and long-term safety. Here, a new machine learning (ML) approach to model pore pressure and fault displacements in response to high-pressure fluid injection cycles is developed. The focus is on fault behavior near the injection borehole. To capture the temporal dependencies in the data, long short-term memory (LSTM) networks are utilized. To prevent error accumulation within the forecast window, four critical measures to train a robust LSTM model for predicting fault response are highlighted: (i) setting an appropriate value of LSTM lag, (ii) calibrating the LSTM cell dimension, (iii) learning rate reduction during weight optimization, and (iv) not adopting an independent injection cycle as a validation set. Several numerical experiments were conducted, which demonstrated that the ML model can capture peaks in pressure and associated fault displacement that accompany an increase in fluid injection. The model also captured the decay in pressure and displacement during the injection shut-in period. Further, the ability of an ML model to highlight key changes in fault hydromechanical activation processes was investigated, which shows that ML can be used to monitor risk of fault activation and leakage during high pressure fluid injections.

58 GEOSCIENCES↗
DOE OSTI2024

Machine learning-guided discovery of polymer membranes for CO 2 separation with genetic algorithm

Designing polymer membranes with high gas permeability and selectivity is a difficult multi-task constrained problem due to the trade-off between these two properties. In this work, we present a machine learning (ML) driven genetic algorithm to tackle the design problem of polymer membranes for CO 2 separation from N 2 and O 2 . Using literature data of permeability for three gases, we constructed multiple ML models with different fingerprinting featurization schemes to predict gas permeabilities. Then, we employed a genetic algorithm to design new polymers and evaluated their performance using our ML models. We were able to identify new polymer membranes that are promising for both CO 2 /N 2 and CO 2 /O 2 separations. Further, the top discovered polymers are predicted to have high glass transition temperatures. Similarly, the pyridine functionality was found in ≈20% of the predicted polymers. This framework can be used to design polymers for any application involving constrained optimization. Finally, we outlined the challenges and opportunities with using ML guided data-driven inverse design of polymers.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2026

Machine learning-based bias-corrected future projections of ozone concentrations from a chemistry-climate model

Reliable projection of future near-surface ozone is crucial for air quality management and health risk assessment. However, potential biases in spatial distribution, magnitude and trends in ozone concentrations simulated by global chemistry-climate models limit their applicability in regional-scale evaluations. In this study, LightGBM, a machine learning (ML) algorithm is applied to correct biases in CESM2-simulated ozone concentrations over China, the United States and Europe and calibrate future ozone projections under two diverse Shared Socioeconomic Pathways (SSP1-2.6 and SSP5-8.5) scenarios from 2020 to 2060. The ML-based correction significantly improves the spatial distribution and reduces the model bias by 40%–60%. It also reverses the potentially incorrect trend of ozone change under SSP1-2.6 in eastern China. When applying ML-based bias correction to CESM2 future projections, warm season mean ozone concentrations decrease across China, the United States, and Europe by –13.5, –17.9, and –13.7 µg/m³, respectively, between 2020 and 2060 in SSP1-2.6, while they increase by 9.4, 2.0, and 5.2 µg/m³ in SSP5-8.5. Decomposition analysis show that changes in anthropogenic emissions dominate future ozone changes in both scenarios, while strong climate penalty from ozone changes occurs in polluted eastern China and climate benefit is found in western China, the United States and Europe under SSP5-8.5. These findings demonstrate the value of combining ML with chemistry-climate models to produce more accurate air quality projections, thereby informing more effective and region-specific environmental protection strategies.

Chemistry Model↗
DOE OSTI2024

Machine Learning a Simple Interpretable Short-Range Potential for Silica

A wide array of models, spanning from computationally expensive ab initio methods to a spectrum of force-field approaches, have been developed and employed to probe silica polymorphs and understand growth processes and atomic-level dynamical transitions in silica. However, the quest for a model capable of making accurate predictions with high computational efficiency for various silica polymorphs is still ongoing. Recent developments in short-range machine-learned models, such as GAP and NNPScan, have shown promise in providing reasonable descriptions of silica, but their computational cost remains high compared to force fields such as BKS which are based on simple interpretable functional forms. Here, in this study, we build on the recent success of our reinforcement learning (RL) workflow to derive a new set of optimal parameters for a promising short-range BKS-based model proposed by Soules. We use RL to navigate the eight-dimensional parameter space of the Soules potential using an experimental training data set that includes both local and global structural features from approximately 21 experimentally realized silica polymorphs, including high density phases and porous zeolites. We compare the performance of our machine-learned ML-Soules model with other high quality models including our recent machine-learned parametrization of BKS (ML-BKS), a machine-learned potential (GAP), as well as predictions of ab initio calculations with the highly fidelity SCAN functional. The ML-Soules accurately captures the relative energetic ordering of various polymorphs as well as their structural features at a significantly reduced computational expense. The ML-Soules model also reasonably captures the structure, density, and elastic constants of quartz, as well as metastable silica polymorphs. We further discuss the limitations of the Soules functional form and propose potential enhancements, including the incorporation of additional three-body terms and/or the utilization of different short-ranged functional forms to achieve greater accuracy for both global and local features in the modeling of silica while retaining low computational cost.

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

Similarity Metric for Data Optimization and Efficient Training of Reactive Machine Learning Force Fields for Hydrocarbon Radiolysis

Radiolysis is a common approach to sterilize polymers, chemically modify them for upcycling, and accelerate their decomposition for recycling purposes. Reactive molecular dynamics (MD) simulations provide a powerful tool to generate atomic-level trajectories of the reactive processes and quantify radiolytic chemical degradation pathways. For this, machine learning (ML) surrogate models for reactive force fields with quantum mechanical accuracy are now widely used, which require ML training data sets that can provide information on atomic environments for target chemical systems. However, radiolysis chemistry can be highly complex and diverse, which poses significant challenges for generating training data to parametrize ML models. In this regard, we developed a method for optimizing the training data set using a cosine similarity metric to help guide training set selection for radiolysis of polyethylene, a model hydrocarbon polymer, as well as to enhance the transferability of our reactive ML force field (MLFF) to a variety of molecular and polymeric systems. Our approach performs atom-by-atom comparisons between local atomic environments to pinpoint important data points associated with rare and localized events, such as radiolysis damage within structures. We apply this approach to train the Chebyshev Interaction Model for Efficient Simulation (ChIMES) MLFF model, which expresses the atomic interaction potentials in terms of linear combinations of many-body Chebyshev polynomials. We first show that our method can reduce our training set size by ∼70% while improving overall accuracy compared to more standard MD model fitting approaches. We then validate our optimum model against diverse hydrocarbon simulation data, including simple alkanes and systems with unsaturated carbon bonds, over a wide range of thermodynamic conditions. Finally, we use our ChIMES model to perform MD simulations of radiolytic damage with large-scale systems that help avoid system size effects. Overall, our approach yields an MD force field that retains most of the accuracy of the underlying quantum method while yielding many orders of improvement in computational efficiency. In conclusion, our efforts will have impact on future hydrocarbon polymer radiolysis studies, where the chemical details of the polymer–radiation interactions can have a strong effect on the resulting products observed in experiments.

Hydrocarbons↗
DOE OSTI2024

Machine Learning and IAST-Aided High-Throughput Screening of Cationic and Silica Zeolites for Alkane Capture, Storage, and Separations

We present an approach for quantitatively predicting the temperature-dependent single-component adsorption behavior of linear alkanes in silica and Na-exchanged cationic zeolites using machine learning (ML) models trained from extensive molecular simulations based on force fields with coupled cluster accuracy. A high-performing classification model was developed to distinguish between instances with negligible and non-negligible adsorption. Subsequently, two ML models were trained to predict the single-component adsorption loading and the heat of adsorption at any pressure at 300 K for any zeolite topology and silicon-to-aluminum ratio. The ML models were trained on International Zeolite Association (IZA) zeolites, and their transferability to hypothetical zeolites was successfully validated. We then expand the power of these predictions to adsorbed mixtures at arbitrary temperatures by integrating them with the Clausius–Clapeyron equation and ideal adsorbed solution theory (IAST). This approach was validated and then applied to a temperature swing adsorption separation process to demonstrate its practical utility. We demonstrate how predictions from this ML-enabled approach can allow the selection of high-performing materials that are then validated using detailed molecular simulations based on quantitatively accurate force fields.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2024

Machine Learning-Guided Exploration of Ternary Metal Borohydrides

We employ deep machine learning (ML) combined with first-principles calculations to explore energetically favorable ternary metal borohydrides. Using La–B–H as a prototype system, we demonstrate that iteratively trained ML models can efficiently screen hundreds of thousands of hypothetical structures and accurately select a small fraction of promising structures and compositions for further studies by first-principles calculations. Such an ML-guided approach dramatically accelerates the pace of materials discovery. A number of new La–B–H ternary compounds with formation energies within 100 meV/atom above the known ternary convex hull are discovered, including a known stable La(BH 4 ) 3 phase. Moreover, by replacing La with Group 1, 2, 3, 13, and 14 elements in the four lowest-energy La–B–H structures from our ML-guided predictions, several low-energy X–B-H (X = Mg, Ca, Sr, Ba, Sc, Y, Ac, Al, Ga, In, Si, Ge, Sn, Pb) compounds are predicted.

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

Encapsulation of Monolayer 2D Materials Using Kinetic Energy-Controlled Pulsed Laser Deposition

The integration of monolayer (ML) two-dimensional (2D) materials into next-generation microelectronics, optoelectronics, and sensors is hindered by their sensitivity to environmental exposure. Deposition of additional layers for encapsulation or growth on ML 2D materials by versatile but energetic plasma techniques such as pulsed laser deposition (PLD) has not been considered at the monolayer level because of potential damage caused by hyperthermal species with kinetic energies (KEs) exceeding the threshold displacement energy (TDE) of the ML. Here, we describe a general strategy to understand and mitigate damage during PLD by reducing the incident KE of ablated species below the TDE of the 2D monolayer using background gas collisions. Ion flux diagnostics, combined with in situ Raman spectroscopy of monolayer graphene during PLD of amorphous boron nitride (a-BN) as a dielectric encapsulation layer, show that damage is primarily correlated with fast ions that penetrate the background gas in accordance with Beer’s Law and are often overlooked in ICCD imaging due to the dominance of the bright, delayed plasma luminescence. Significantly, if fast ions are eliminated and a ∼2 nm-thick a-BN layer is “soft landed”, the monolayer graphene is effectively protected from damage by high KE species in the boron nitride plasma plume. Deposited a-BN films display a characteristic dielectric constant of 3.6 at 100 kHz and tunable charge injection properties. Our results enable PLD as a viable option for encapsulation and thin film growth onto ML 2D materials, with implications for both fundamental research and device integration.

2D materials↗
DOE OSTI2024

High-Throughput Screening and Accurate Prediction of Ionic Liquid Viscosities Using Interpretable Machine Learning

Ionic liquids (ILs) are a novel group of green solvents with great promise for various industrial applications, including carbon capture and lignocellulosic biomass deconstruction. However, the use of ILs at the industrial scale remains challenging due to their high viscosities at ambient temperatures. To develop ILs with lower viscosities, a systematic study of their quantitative structure–property relationship (QSPR) is desirable. Here, we developed four machine learning (ML) models to predict viscosity at various temperature and pressure ranges, trained over a wide range of ILs consisting of various cationic and anionic families. ML methods including two-factor polynomial regression (two-factor PR), support vector regression (SVR), feed-forward neural networks (FFNN), and categorical boosting (CATBoost) were developed based on features that have proven useful in previous ML studies: COSMO-RS (conductor-like screening model for real solvents)-derived surface screening charge densities (sigma profiles). FFNN and CATBoost were the most accurate in predicting IL viscosities with lower average absolute relative deviation and higher R2 values on the test set. Tanimoto similarity scores were calculated to characterize the chemical space and structural similarity of the investigated ions. Furthermore, SHapley Additive exPlanation (SHAP) analysis was employed to interpret the ML results. Temperature, the polar area of ILs, and the nonpolar regions of ions are key features that influence the viscosity predictions. Importantly, the IL viscosity prediction here is the most accurate reported to date.

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

Advancing AI-Driven Analysis in X-ray Absorption Spectroscopy: Spectral Domain Mapping and Universal Models

In recent years, rapid progress has been made in developing artificial intelligence (AI) and machine learning (ML) methods for X-ray absorption spectroscopy (XAS) analysis. Compared to traditional XAS analysis methods, AI/ML approaches offer dramatic improvements in efficiency and help eliminate human bias. To advance this field, we advocate an AI-driven XAS analysis pipeline that features several interconnected key building blocks: benchmarks, workflows, databases, and AI/ML models. Specifically, we present two case studies for XAS ML. In the first study, we demonstrate the importance of reconciling the discrepancies between simulation and experiment using spectral domain mapping (SDM). Our ML model, which is trained solely on simulated spectra, predicts an incorrect oxidation state trend for Ti atoms in a combinatorial zinc titanate film. After transforming the experimental spectra into a simulation-like representation using SDM, the same model successfully recovers the correct oxidation state trend. In the second study, we explore the development of universal XAS ML models that are trained on the entire periodic table, which enables them to leverage common trends across elements. Looking ahead, we envision that an AI-driven pipeline can unlock the potential of real-time XAS analysis to accelerate scientific discovery.

36 MATERIALS SCIENCE↗
DOE OSTI2024

Snow Distribution Patterns Revisited: A Physics-Based and Machine Learning Hybrid Approach to Snow Distribution Mapping in the Sub-Arctic

Snowpack distribution in Arctic and alpine landscapes often occurs in repeating, year-to-year patterns due to local topographic, weather, and vegetation characteristics. Previous studies have suggested that with years of observational data, these snow distribution patterns can be statistically integrated into a snow process modeling workflow. Recent advances in snow hydrology and machine learning (ML) have increased our ability to predict snowpack distribution using in-situ observations, remote sensing data sets, and simple landscape characteristics that can be easily obtained for most environments. Here, we propose a hybrid approach to couple a ML snow distribution pattern (MLSDP) map with a physics-based, snow process model. We trained a random forest ML algorithm on tens of thousands of snow survey observations from a subarctic study area on the Seward Peninsula, Alaska, collected during peak snow water equivalent (SWE). We validated hybrid model outputs using in-situ snow depth and SWE observations, as well as a light detection and ranging data set and a distributed temperature profiling sensor data set. When the hybrid results were compared with the physics-based method, the hybrid method more accurately depicted the spatial patterns of the snowpack, areas of drifting snow, and years when no in-situ observations were used in the random forest ML training data set. The hybrid method also showed improvements in root mean squared error at 61% of locations where time-series estimations of snow depth were observed. These results can be applied to any physics-based model to improve the snow distribution patterning to reflect observed conditions in high latitude and high elevation cold region environments.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2026

Hierarchical Testing of a Hybrid Machine Learning‐Physics Global Atmosphere Model

Machine learning (ML)-based models have demonstrated high skill and computational efficiency, often outperforming conventional physics-based models in weather and subseasonal predictions. While prior studies have assessed their fidelity in capturing synoptic-scale atmospheric dynamics, their performance across timescales and under out-of-distribution forcing, such as +3K or +4K uniform-warming forcings, and the sources of biases remain elusive, to establish the model's reliability for Earth science. Here, we design three sets of experiments targeting synoptic-scale phenomena, interannual variability, and out-of-distribution uniform-warming forcings. We evaluate the Neural General Circulation Model (NeuralGCM), a hybrid model integrating a dynamical core with ML-based component, against observations and physics-based Earth system models (ESMs). At the synoptic scale, NeuralGCM captures the evolution and propagation of extratropical cyclones with performance comparable to ESMs. At the interannual scale, when forced by El Niño-Southern Oscillation sea surface temperature (SST) anomalies, NeuralGCM successfully reproduces associated teleconnection patterns but exhibits deficiencies in capturing nonlinear response. Under out-of-distribution uniform-warming forcings, NeuralGCM simulates similar responses in global-average temperature and precipitation and reproduces large-scale tropospheric circulation features similar to those in ESMs. Notable weaknesses include overestimating the tracks and spatial extent of extratropical cyclones, biases in the teleconnected wave train triggered by tropical SST anomalies, and differences in upper-level warming and stratospheric circulation responses to SST warming compared to physics-based ESMs. The causes of these weaknesses were explored. Despite the noted weaknesses, NeuralGCM reproduces responses across experiments reasonably and performs comparably to ESMs. By integrating a dynamical core with ML, NeuralGCM shows potential for developing ML-based ESMs.

global warming↗