Search NASA⌕ Search

SEARCH · Search NASA

Results for “predict”

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 55 records · Page 3

Prediction of performance and turbulence in ITER burning plasmas via nonlinear gyrokinetic profile prediction

Burning plasma performance, transport, and the effect of hydrogen isotope (H, D, D-T fuel mix) on confinement has been predicted for ITER baseline scenario (IBS) conditions using nonlinear gyrokinetic profile predictions. Accelerated by surrogate modeling (Rodriguez-Fernandez et al 2022 Nucl. Fusion 62 076036), high fidelity, nonlinear gyrokinetic simulations performed with the CGYRO code (Candy et al 2016 J. Comput. Phys. 324 73), were used to predict profiles of T i , T e , and n e while including the effects of alpha heating, auxiliary power (NBI + ECH), collisional energy exchange, and radiation losses inside of $r/a$ = 0.9. Predicted profiles and resulting energy confinement are found to produce fusion power and gain that are approximately consistent with mission goals ($P_\textrm{fusion} = 500$ MW at Q = 10) for the baseline scenario and exhibit energy confinement that is within 1σ of the H-mode energy confinement scaling. The power of the surrogate modeling technique is demonstrated through the prediction of alternative ITER scenarios with reduced computational cost. These scenarios include conditions with maximized fusion gain and an investigation of potential resonant magnetic perturbation (RMP) effects on performance with a minimal number of gyrokinetic profile iterations required (3–6). These predictions highlight the stiff ITG nature of the core turbulence predicted in the ITER baseline and demonstrate that $Q \gt$ 17 conditions may be accessible by reducing auxiliary input power while operating in IBS conditions. Prediction of full kinetic profiles allowed for the projection of hydrogen isotope effects around ITER baseline conditions. The gyrokinetic fuel ion species was varied from H, D, and 50/50 D-T and kinetic profiles were predicted. Results indicate that a weak or negligible isotope effect will be observed to arise from core turbulence in IBS conditions. The resulting energy confinement, turbulence, and density peaking, and the implications for ITER operations will be discussed.

gyrokinetics↗

Hybridizing Machine Learning and Physically-based Earth System Models to Improve Prediction of Multivariate Extreme Events (AI Exploration of Wildland Fire Prediction)

Focal Areas: This project responds to two focal areas identified in the DOE Call for AI4ESP White Papers: 1) Predictive modeling through the use of artificial intelligence (AI) techniques, and 2) insights gleaned from complex data using explainable AI and big data analytics. Science Challenge: Large wildland fires (hereafter wildfires) appearing as high-impact compound climate extreme events are closely related to hydroclimate and water cycle extremes that modulate surface fuel supply and combustibility. These compound events have multivariate climatic features (e.g., temperature, precipitation, relative humidity, wind, lightning) and societal drivers (e.g., forest management, land use change, human caused ignitions). Meanwhile, they induce strong feedbacks to the coupled atmosphere, biosphere, and hydrosphere by perturbing regional and global radiation budget as well as ecological, biogeochemical, and water cycles across multiple spatiotemporal scales. The nonlinear interactions between these natural and anthropogenic components of the Earth system are too complex to be completely and adequately represented in today’s Earth system models (ESMs). The inherent stochastic nature of fire activity at all scales further increases the difficulty of its prediction using ESMs that are usually developed from deterministic equations and parameterizations. Besides, concurrence of long-term (decadal to interdecadal) global climate change and fire regime shifts overlapping with short-term (intraseasonal to interannual) variations of regional fire weather and burning activity confound predictability of these compound extreme events. We propose to address the above scientific challenges by using machine learning (ML)-based data-driven modeling techniques to integrate observations and physically-based ESMs’ simulations in a computationally efficient hybrid prediction system. This prediction system is supposed to characterize the wildfire’s sensitivity to climate and exogenous drivers at high resolution (~ 0.25°) on subseasonal to seasonal (S2S) timescales providing improved predictability and explainability. We will use the system to help identify: (1) What are the computational elements of a hybrid system needed to predict compound climate extreme events such as global wildfires? (2) What are the key drivers (either natural or anthropogenic) that modulate short-term variations of multivariate fire weather and burning activity over different regions? How can one take advantage of those driver-response relationships to improve the predictability of large wildfires on S2S time scales? (3) What are the underlying physical mechanisms and sources of improved predictability? Which ML techniques are optimal in revealing and adapting these mechanisms?

54 ENVIRONMENTAL SCIENCES↗

Comparison of Optimal Energy Management Strategies Using Dynamic Programming, Model Predictive Control, and Constant Velocity Prediction

Due to the recent advancements in autonomous vehicle technology, future vehicle velocity predictions are becoming more robust which allows fuel economy (FE) improvements in hybrid electric vehicles through optimal energy management strategies (EMS). A real- world highway drive cycle (DC) and a controls-oriented 2017 Toyota Prius Prime model are used to study potential FE improvements. We proposed three important metrics for comparison: (1) perfect full drive cycle prediction using dynamic programming, (2) 10-second prediction horizon model predictive control (MPC), and (3) 10-second constant velocity prediction. These different velocity predictions are put into an optimal EMS derivation algorithm to derive optimal engine torque and engine speed. The results show that the constant velocity prediction algorithm outperformed the baseline control strategy but underperformed the MPC strategy with an average 1.58% and 2.45% of FE improvement with highway and city-highway DC. Also, using a 10-second prediction window MPC strategy provided FE improvement results close to the full drive cycle prediction case. MPC has the potential to achieve 60%-65% and 70% - 80% of global FE improvement over highway and city-highway DC respectively.

Patel, Amol Arvind↗

Comparison of Optimal Energy Management Strategies Using Dynamic Programming, Model Predictive Control, and Constant Velocity Prediction

Due to the recent advancements in autonomous vehicle technology, future vehicle velocity predictions are becoming more robust which allows fuel economy (FE) improvements in hybrid electric vehicles through optimal energy management strategies (EMS). A realworld highway drive cycle (DC) and a controls-oriented 2017 Toyota Prius Prime model are used to study potential FE improvements. We proposed three important metrics for comparison: (1) perfect full drive cycle prediction using dynamic programming, (2) 10-second prediction horizon model predictive control (MPC), and (3) 10-second constant velocity prediction. These different velocity predictions are put into an optimal EMS derivation algorithm to derive optimal engine torque and engine speed. The results show that the constant velocity prediction algorithm outperformed the baseline control strategy but underperformed the MPC strategy with an average 1.58% and 2.45% of FE improvement with highway and city-highway DC. Also, using a 10-second prediction window MPC strategy provided FE improvement results close to the full drive cycle prediction case. MPC has the potential to achieve 60%-65% and 70% - 80% of global FE improvement over highway and city-highway DC respectively.

Patel, Amol Arvind↗

The Solar Influencer Next Door: Predicting Low-Income Solar Referrals and Leads

Increasing the adoption of solar among low-to-moderate income (LMI) households remains an important policy goal because of its promise to simultaneously reduce energy burden and support the just distribution of benefits of renewable energy. However, scaling LMI solar remains challenging due to affordability and access issues. Most existing LMI adoption has occurred under public-funded programs, highlighting the importance of increasing the cost-effectiveness of these programs at scale. We develop a new household-level data set on LMI solar lead acquisition, referrals, and adoption to understand the processes through which LMI solar uptake has occurred in California. Then, we develop models to predict two sub-mechanisms in the solar adoption process: whether an otherwise qualified lead becomes "lost" i.e. non-responsive to outreach and, for existing clients, whether they refer solar to others. For the program analyzed, participants received their solar system at no cost, which deemphasizes economic drivers of solar adoption and could differ from other program experiences. Both models substantially improved the accuracy of prediction relative to a baseline. Overall, we find that peer effects and solar economics are important to predicting referrals, and household demographic factors in lead loss prediction. Finally, we find that referrals are both the highest quality and largest source of LMI solar leads, providing a promising mechanism to expand LMI programs further.

customer acquisition costs↗

A review of machine learning in building load prediction

The surge of machine learning in recent years has been empowering engineer modeling in various fields. The decreasing hardware cost, increasing data accessibility, and advances of building automation system (BAS) allow the collection and storage of a significant amount of building operation data. The two facts provide great opportunities of applying machine learning to building energy systems modeling and analysis. There are a great number of research papers on this topic but there lacks a comprehensive and general review to summarize the current development, limitations, gaps and future trend. In this review paper series, machine learning techniques in building energy system modeling and analysis are reviewed under the organization and logic of the machine learning definition by Tom M. Mitchell: a computer program is said to learn from experience E with respect to some class of tasks T and performance measure P if its performance at tasks in T, as measured by P, improves with experience E. This paper is the first part of the review paper series, which focuses on building load prediction. First, the applications of building load prediction model (task T) are reviewed. Then, the modeling algorithms improving machine learning performance and accuracy (performance P) are reviewed. At the same time, the literature on the data perspective for modeling (experience E), including data engineering from sensors level to data level, pre-processing, feature extraction and selection, is reviewed. Finally, what is well-studied and what is lacking but with great potential are concluded; the gaps between present and future utilization of machine learning techniques are identified; the future trend and development are also predicted. The target readers of this paper are not only researchers from the building side who can get exposed to cutting edge machine learning tools, but also those from machine learning side who can understand the potential and challenge to apply machine learning in buildings.

Liang, Zhang↗

Improving streamflow predictions across CONUS by integrating advanced machine learning models and diverse data

Accurate streamflow prediction is crucial to understand climate impacts on water resources and develop effective adaption strategies. A global long short-term memory (LSTM) model, using data from multiple basins, can enhance streamflow prediction, yet acquiring detailed basin attributes remains a challenge. To overcome this, we introduce the Geo-vision transformer (ViT)-LSTM model, a novel approach that enriches LSTM predictions by integrating basin attributes derived from remote sensing with a ViT architecture. Applied to 531 basins across the Contiguous United States, our method demonstrated superior prediction accuracy in both temporal and spatiotemporal extrapolation scenarios. Geo-ViT-LSTM marks a significant advancement in land surface modeling, providing a more comprehensive and effective tool for better understanding the environment responses to climate change.

Tayal, Kshitij↗

A machine learning approach for clinker quality prediction and nonlinear model predictive control design for a rotary cement kiln

Abstract Cement manufacturing is energy‐intensive (5Gj/t) and comprises a significant portion of the energy footprint of concrete systems. Incorporating modern monitoring, simulation and control systems will allow lower energy use, lower environmental impact, and lower costs of this widely used construction material. One of the goals of the CESMII roadmap project on the Smart Manufacturing of Cement included developing an analytical process model for clinker quality that includes the chemistry of the kiln feed and accounts for critical process variables. This predictive model will be used in nonlinear model predictive control system designed to significantly reduce process energy use while maintaining or improving product quality. In the cement manufacturing plant used in this study, the kiln feed (meal) is tested every 12 h and used to estimate the mineral composition of the cement kiln output (clinker) using the stoichiometry‐based Bogue's model and the expertise of the plant operators. During kiln operation, kiln output (clinker) is sampled and tested every 2 h to measure its chemical and mineral composition. The predicted and measured values of the clinker composition are used by the plant operators to adjust the kiln input stream and the production process characteristics to maintain stable operation and uniform product quality. However, the time delay between prediction and testing, along with inaccuracies inherent in the Bogue's model have made any process changes designed to minimize energy use problematic, especially in‐light of potential clinker quality issues that process changes often pose. A new analytical model that integrates quality information and process operation information has been developed from data collected from 2 years of production from an operating cement facility. To make the model fuel‐type‐independent, consumed heat energy was computed in the model instead of fuel type and amount. A Feedforward Network was trained and tailored from collected data. Many data‐based simulations were conducted to quantitatively evaluate the proposed model and the 5‐fold cross‐validation procedure was used to test the models. The resulting predictive model was shown to have a low root mean square error (MSE) with respect to the estimated clinker mineral composition compared to that using the industry standard “Bogue’ model”. The end goal of this work was to develop a single machine learning tool that allows the use of quality control data and process control variables to improve energy efficiency of the process in a continuous fashion. The proposed nonlinear model predictive control system (NMPC) can generate predicted kiln production characteristics based on manipulated variables in manner that accurately follows the target product quality values. Simulation results also show that the proposed model produced accurate predictions of kiln outputs that fell within the required constraints, while manipulating control variables within typical operational ranges.

Ali, Asem M.↗

Reduced model prediction of electron temperature profiles in microtearing-dominated National Spherical Torus eXperiment plasmas

A representative H-mode discharge from the National Spherical Torus eXperiment is studied in detail to utilize it as a basis for a time-evolving prediction of the electron temperature profile using an appropriate reduced transport model. The time evolution of characteristic plasma variables such as β{sub e}, ν{sub e}{sup ∗}, the MHD α parameter, and the gradient scale lengths of T{sub e}, T{sub i}, and n{sub e} were examined as a prelude to performing linear gyrokinetic calculations to determine the fastest growing micro instability at various times and locations throughout the discharge. The inferences from the parameter evolutions and the linear stability calculations were consistent. Early in the discharge, when β{sub e} and ν{sub e}{sup ∗} were relatively low, ballooning parity modes were dominant. As time progressed and both β{sub e} and ν{sub e}{sup ∗} increased, microtearing became the dominant low-k{sub θ} mode, especially in the outer half of the plasma. There are instances in time and radius, however, where other modes, at higher-k{sub θ}, may, in addition to microtearing, be important for driving electron transport. Given these results, the Rebut-Lallia-Watkins (RLW) electron thermal diffusivity model, which is based on microtearing-induced transport, was used to predict the time-evolving electron temperature across most of the profile. The results indicate that RLW does a good job of predicting T{sub e} for times and locations where microtearing was determined to be important, but not as well when microtearing was predicted to be stable or subdominant.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

Effects of relativity for atomization and isomerization energies of seaborgium carbonyl SgCO and seaborgium isocarbonyl SgOC: Relativity predicts SgOC to be more stable than SgCO

Our ab initio all-electron fully relativistic Dirac-Fock (DF) and nonrelativistic Hartree-Fock (NR) calculations for seaborgium isocarbonyl SgOC predict atomization energy (AE) of 13.04 and 11.05 eV, respectively. However, the corresponding DF and NR atomization energies for the seaborgium carbonyl SgCO are predicted as 12.75 and 12.45 eV, respectively. This is the first such result in Chemistry where an isocarbonyl (and especially for a system of superheavy element Sg) is predicted to be more stable at the DF level of theory than the corresponding carbonyl. The predicted energy for the formation of the carbonyl SgCO at the relativistic DF and NR levels of theory is -54.90 and -50.95 kJ /mol, whereas the corresponding energy of formation of the isocarbonyl SgOC is -64.44 and -18.64 kJ/mol, respectively. Ours are the first results of relativistic effects for isomerization and atomization energies of the superheavy seaborgium isocarbonyl SgOC and its isomer SgCO. Lastly, the formation of isocarbonyl SgOC, should be favored over the carbonyl isomer SgCO in the first step of the reaction Sg+CO →SgOC.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

A Data-driven Approach for Predicting Nepheline Crystallization in High-Level Waste Glasses

High-level waste (HLW) glasses with high aluminum content are prone to nepheline crystallization during the slow canister cooling that is experienced during large-scale production. Due to its detrimental effects on glass durability, nepheline precipitation must be avoided; however, developing robust, predictive models for nepheline crystallization behavior in HLW glasses is difficult due to their compositional complexity. Using overly conservative constraints to predict nepheline formation can limit the waste loading to lower than the achievable capacity. In this study, a robust data-driven model using five compositional features has been developed to predict nepheline formation. The analysis of the model and the data show that there is an overlap, instead of a distinct compositional boundary, between glasses that form and do not form nepheline. As a result, the model’s predictive accuracy is not the same throughout the feature space and instead is dependent on the location of the glass composition in the dimensionally reduced feature space.

Sargin, Irmak↗

PREDICTION OF CYCLIC VARIABILITY AND KNOCK-LIMITED SPARK ADVANCE (KLSA) IN SPARK-IGNITION (SI) ENGINE

Engine knock remains one of the major barriers to further improve the thermal efficiency of spark-ignition (SI) engines. SI engine is usually operated at knock-limited spark advance (KLSA) to achieve possibly maximum efficiency with given engine hardware and fuel properties. Co-optimization of fuels and engines is promising to improve engine efficiency, and predictive computational fluid dynamics (CFD) models can be used to facilitate this process. However, cyclic variability of SI engine demands that multi cycle results are required to capture the extreme conditions. In addition, Mach Courant Friedrichs-Lewy (CFL) number of 1 is desired to accurately predict the knock intensity (KI), resulting in unaffordable computational cost. In this study, a new approach to numerically predict KLSA using large Mach CFL of 50 with ten consecutive cycle simulation is proposed. This approach is validated against the experimental data for a boosted SI engine at multiple loads and spark timings with good agreements in terms of cylinder pressure, combustion phasing, and cyclic variation. Engine knock is predicted with early spark timing, indicated by significant pressure oscillation and end-gas heat release. Maximum amplitude of pressure oscillation analysis is performed to quantify the KI, and the slope change point in KI extrema is used to indicate the KLSA accurately. Using a smaller Mach CFL number of 5 also results in the same conclusions, thus demonstrating that this approach is insensitive to the Mach CFL number. The use of large Mach CFL number allows us to achieve fast turn-around time for multicycle engine CFD simulations.

CCV↗

TH-A-9A-01: Active Optical Flow Model: Predicting Voxel-Level Dose Prediction in Spine SBRT

Purpose: To predict voxel-level dose distribution and enable effective evaluation of cord dose sparing in spine SBRT. Methods: We present an active optical flow model (AOFM) to statistically describe cord dose variations and train a predictive model to represent correlations between AOFM and PTV contours. Thirty clinically accepted spine SBRT plans are evenly divided into training and testing datasets. The development of predictive model consists of 1) collecting a sequence of dose maps including PTV and OAR (spinal cord) as well as a set of associated PTV contours adjacent to OAR from the training dataset, 2) classifying data into five groups based on PTV's locations relative to OAR, two “Top”s, “Left”, “Right”, and “Bottom”, 3) randomly selecting a dose map as the reference in each group and applying rigid registration and optical flow deformation to match all other maps to the reference, 4) building AOFM by importing optical flow vectors and dose values into the principal component analysis (PCA), 5) applying another PCA to features of PTV and OAR contours to generate an active shape model (ASM), and 6) computing a linear regression model of correlations between AOFM and ASM.When predicting dose distribution of a new case in the testing dataset, the PTV is first assigned to a group based on its contour characteristics. Contour features are then transformed into ASM's principal coordinates of the selected group. Finally, voxel-level dose distribution is determined by mapping from the ASM space to the AOFM space using the predictive model. Results: The DVHs predicted by the AOFM-based model and those in clinical plans are comparable in training and testing datasets. At 2% volume the dose difference between predicted and clinical plans is 4.2±4.4% and 3.3±3.5% in the training and testing datasets, respectively. Conclusion: The AOFM is effective in predicting voxel-level dose distribution for spine SBRT. Partially supported by NIH/NCI under grant #R21CA161389 and a master research grant by Varian Medical System.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Data-driven predictive probability density function control of fiber length stochastic distribution shaping in refining process

Pulp is the most important raw material in paper in paper industries, whose Fiber Length Stochastic Distribution (FLSD) shaping directly determines the energy consumption and paper quality of the subsequent papermaking processes. However, the mean and variance are insufficient to describe the output FLSD shaping, which display non-Gaussian distribution properties. Therefore, the traditional control, method based on the mean and variance of fiber length is difficult to control the FLSD shaping effectively. In this paper, a novel data-driven predictive probability density function (PDF) control method is proposed for output FLSD shaping in refining process. Primarily, in order to improve the approximation accuracy of the PDF employing RBF neural network, the parameters (center value and width) of each RBF basis functions are tuned via utilizing iterative learning control (ILC) law, and the corresponding estimations of weights law can be obtained. Secondly, considering that the conventional linear model of weights vector has drawbacks of low accuracy and weak generalization ability, random vector functional link networks (RVFLNs) based data driven nonlinear modelling method is employed to characterize the prediction model between the input variables and weights vector. Finally, in order to reduce the randomness of the output FLSD, a minimum entropy control method under mean constraint is employed to design the predictive PDF controller for the output FLSD shaping with the help of the established nonlinear stochastic distribution model. Both simulations and experiments are given to illustrate the effectiveness and practicability of the proposed method.

Probability debsity function, Fiber length distrib↗

Comparative Study on the Machine Learning-Based Prediction of Adsorption Energies for Ring and Chain Species on Metal Catalyst Surfaces

Computation of adsorption and transition state energies for a large number of surface intermediates for numerous active site models pose significant computational overhead in computational screening of catalysts. Machine learning (ML) techniques can be used to predict part of these energies. To predict the energies, ML models need to be fed appropriate metal and species descriptors. For complex surface chemistries, the structures of the intermediate species can vary greatly. In this paper, working with the hydrodeoxygenation of succinic acid on six different metal surfaces, we have studied the effect of linear and non-linear ML models used along with pen-and-paper based species descriptors and two categories of metal descriptors on two different categories of intermediate species: chain and ring. More specifically, our computations include the prediction of chain species when trained on only chain species and also when trained on both chain and ring species. Similar computations were performed for predictions of ring species. In each case, results of linear ML models were compared with kernel based non-linear models. Our results indicate that ring species data does not improve the prediction of chain species. Similarly, chain species data does not improve the prediction of ring species. The use of non-linear ML models, however, did help to minimize the prediction errors compared to the linear models. Furthermore, the study also shows that electronic or adsorption energy based metal descriptors along with bond count based species fingerprints can achieve a mean absolute error (MAE) of less than 0.2 eV for complex chain molecules when used with an appropriate machine learning model.

Adsorption↗

Univariate Time Series Prediction of Solar Power Using a Hybrid Wavelet-ARMA-NARX Prediction Method

This paper proposes a new hybrid method for super short-term solar power prediction. Solar output power usually has a complex, nonstationary, and nonlinear characteristic due to intermittent and time varying behavior of solar radiance. In addition, solar power dynamics is fast and is inertia less. An accurate super short-time prediction is required to compensate for the fluctuations and reduce the impact of solar power penetration on the power system. The objective is to predict one step-ahead solar power generation based only on historical solar power time series data. The proposed method incorporates discrete wavelet transform (DWT), Auto-Regressive Moving Average (ARMA) models, and Recurrent Neural Networks (RNN), while the RNN architecture is based on Nonlinear Auto-Regressive models with eXogenous inputs (NARX). The wavelet transform is utilized to decompose the solar power time series into a set of richer-behaved forming series for prediction. ARMA model is employed as a linear predictor while NARX is used as a nonlinear pattern recognition tool to estimate and compensate the error of wavelet-ARMA prediction. The proposed method is applied to the data captured from UCLA solar PV panels and the results are compared with some of the common and most recent solar power prediction methods. The results validate the effectiveness of the proposed approach and show a considerable improvement in the prediction precision.

Nazaripouya, Hamidreza↗

HydraGNN_Predictive_GFM_2026 - Ensemble of predictive graph foundation models for atomistic materials modeling

This release contains data and parameters of HydraGNN-based graph foundation models trained as a result of the work published in the pre-print "Exascale Multi-Task Graph Foundation Models for Imbalanced, Multi-Fidelity Atomistic Data" by M. Lupo Pasini et al. (https://arxiv.org/abs/2604.15380). We jointly train on 16 open first-principles datasets (544+ million structures covering 85+ elements) using a multi-task architecture with per-dataset heads and a scalable ADIOS2/DDStore data pipeline. On Frontier, we execute six large-scale DeepHyper hyperparameter optimization campaigns in FP64 and promote the top-performing message-passing models to sustained 2,048-node training, yielding a PaiNN-based lead model. The version of HydraGNN used to generate the outputs provided in this release is HydraGNN v5.0 (https://github.com/ORNL/HydraGNN/releases/tag/v5.0) The list of datasets used for the training of the graph foundation model is the following: 1) Alexandria [1] 2) ANI1x [2] 3) MPTrj [3] 4) Open Catalyst 2020 (OC20) [4] 5) Open Catalyst 2022 (OC22) [5] 6) Open Catalyst 2025 (OC25) [6] 7) Open Direct ir Capture 2023 (ODAC23) [7] 8) Open Materials 2024 (OMat24) [8] 9) Open Molecules 2025 (OMol25) [9] 10) OMol25-neutral (subset of OMol25 that contains only molecules with zero total charge) 11) OMol25-non-neutral (subset of OMol25 that contains only molecules with non-zero total charge) 12) Open Polymers 2026 (OPoly2026) [10] 13) Nabla2DFT [11] 14) QCML [12] 15) QM7X [reference 13] 16) transition1x [14] Dataset references: [1] J. Schmidt et al., “A dataset of 175k stable and metastable materials calculated with the PBEsol and SCAN functionals,” Scientific Data, vol. 9, p. 64, 2022. [2] J. S. Smith et al., “The ANI-1ccx and ANI-1x data sets, coupled-cluster and density functional theory properties for molecules,” Scientific Data, vol. 7, p. 134, 2020. [Online]. Available: https: //www.nature.com/articles/s41597-020-0473-z [3] A. Jain et al., “Commentary: The Materials Project: A materials genome approach to accelerating materials innovation,” APL Materials, vol. 1, no. 1, p. 011002, 07 2013. [Online]. Available: https://doi.org/10.1063/1.4812323 [4] L. Chanussot et al., “Open catalyst 2020 (oc20) dataset and community challenges,” ACS Catalysis, vol. 11, no. 10, pp. 6059–6072, 2021. [Online]. Available: https://doi.org/10.1021/acscatal.0c04525 [5] K. Tran et al., “Open catalyst 2022 (oc22) dataset and challenges for oxidation electrocatalysts,” ACS Catalysis, vol. 13, no. 5, pp. 3066–3084, 2023. [Online]. Available: https://doi.org/10.1021/acscatal.2c05426 [6] S. J. Sahoo et al., “The open catalyst 2025 (oc25) dataset and models for solid-liquid interfaces,” arXiv preprint arXiv:2509.17862, 2025. [Online]. Available: https://arxiv.org/abs/2509.17862 [7] A. Sriram et al., “The open DAC 2023 dataset and challenges for sorbent discovery in direct air capture,” ACS Central Science, vol. 10, no. 5, pp. 923–941, 2024. [8] L. Barroso-Luque et al., “Open materials 2024 (omat24) inorganic materials dataset and models,” 2024. [Online]. Available: https://arxiv.org/abs/2410.12771 [9] D. S. Levine et al., “The open molecules 2025 (OMol25) dataset, evaluations, and models,” 2025. [Online]. Available: https://arxiv.org/abs/2505.08762 [10] D. S. Levine et al., The open polymers 2026 (OPoly26) dataset and evaluations,” arXiv preprint arXiv:2512.23117, 2025. [Online]. Available: https://arxiv.org/abs/2512.23117 [11] K. Khrabrov et al., “Nabla2dft: A universal quantum chemistry dataset of drug-like molecules and a benchmark for neural network potentials,” in NeurIPS 2024 Datasets and Benchmarks Track, 2024. [Online]. Available: https://openreview.net/forum?id=ElUrNM9U8c [12] S. Ganscha et al., “The QCML dataset, quantum chemistry reference data from 33.5M DFT and 14.7B semi-empirical calculations,” Scientific Data, vol. 12, p. 406, 2025. [13] J. Hoja et al., “QM7-X, a comprehensive dataset of quantum-mechanical properties spanning the chemical space of small organic molecules,” Scientific Data, vol. 8, p. 43, 2021. [Online]. Available: https://www.nature.com/articles/s41597-021-00812-2 [14] M. Schreiner et al., “Transition1x - a dataset for building generalizable reactive machine learning potentials,” Scientific Data, vol. 9, p. 779, 2022. The folder "datasets_ADIOS2_format" contains the set of pre-processed datasets in Adaptable I/O System (ADIOS) format (https://www.exascaleproject.org/research-project/adios/) that have been used for the development and training of GFMs in this work. The "datasets_ADIOS2_format" directory contains 2 sub-directories, one for the version "v1" of the datasets and one for the version "v2" of the datasets. The version "v1" of the datasets provides values of the total energy as they are extracted from the original data as it was released by the respective institutions. The version "v2" of the datasets provides values of the energy that have been realigned. The realignment was performed by training a linear regression model that predicts the total energy as a function of the chemical composition of the atomistic structure, and then subtract such prediction from the original value of the total energy. Both folders "v1" and "v2" contain 16 sub-directories, each corresponding to an ADIOS2-formatted dataset The folder "DeepHyper-results" contains the configurational files and model's parameters for all the 186 HPO trials that were successfully completed by the scalable hyperparameter optimization (HPO) runs on Frontier. The content of the folder "DeepHyper-results" I structured as follows: 1) task-list.txt: list of mpnn name, jobid, and deephyper task id 2) gfm_${MPNN}_${JOBID}_0.${TASKID}: run directory with checkpoint files 3) gfm_${MPNN}: deephyper summary directory (*.csv) for each specific MPNN type 4) deephyper-experiment-${JOBID}: output and error logs for each job The file "deephyper-sorted.csv" contains the details of each HydraGNN model built and tested by HPO, obtained by merging the (*.csv) filed from each HPO run executed. Out of all the HPO trials, we selected 10 to continue the training of the respective HydraGNN models. Due to limited computational budget available in the LRN070 allocation we could not complete the training till convergence for all these 10 selected models. The folder "models" contains multiple sub-folders, one per each HydraGNN model trained. Each model sub-folder contains the parameters of each HydraGNN model, with multiple checkpoint-restarts. The list of sub-folders are as follows: 1) multidataset_hpo-BEST1-fp64 2) multidataset_hpo-BEST2-fp64 3) multidataset_hpo-BEST3-fp64 4) multidataset_hpo-BEST4-fp64 5) multidataset_hpo-BEST5-fp64 6) multidataset_hpo-BEST6-fp64 7) multidataset_hpo-BEST7-fp64 8) multidataset_hpo-BEST8-fp64 9) multidataset_hpo-BEST9-fp64 10) multidataset_hpo-BEST10-fp64 Within each one of these folders, additional auxiliary log files are provided with descriptions about how the training proceeded. The lead PaiNN-model is contained inside "multidataset_hpo-BEST6-fp64". The file "mlp_branch_weights" contains the parameters of the multi-layer perceptron (MLP) used to reconcile the predictions of the 16 output decoding heads of the HydragNN architectures. The MLP takes in input the chemical composition of the atomistic structure and predicts averaging weights to linearly mix the predictions of each output decoding head toward consolidating them into a single one. The folder "1.1billion-structure-inference" contains 1.1 billion atomistic structures randomly generated. Each structures is associated with energy and forces predicted with the lead-PaiNN model combined with the MLP model for reconciliation of the multi-branch predictions generated by the 16 output decoding heads. The folder "1.1billion-structure-inference" contains 9,300 (*.tar.gz) subdirectories, one per Frontier compute node used to execute the inference at exascale. Once uncompressed, each (*.tar.gz) subdirectory contains an ADIOS2 (*.bp) file container, where each atomistic structure is stored as a PyTorch-Geometric Data object. The file "export_dataset_environment_variables.sh" contains the environment variables that need to be set before running the HydraGNN code to reproduce the results provided in this dataset release. The code that can be used to load the ADIOS2 files, load HydraGNN models, and run inference is available at: https://github.com/ORNL/HydraGNN/releases/tag/v5.0

36 MATERIALS SCIENCE↗