Search NASA⌕ Search

SEARCH · Search NASA

Results for “model complexity”

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 469 records · Page 26
All sourcesNASA NTRSDOE OSTI
DOE OSTI2020

Smart Proxy Modeling Application of Artificial Intelligence & Machine Learning in Computational Fluid Dynamics

Smart proxy technology leverages the art of artificial intelligence and machine learning in order to build accurate and very fast proxy models for highly complex numerical simulation models. In this project, smart proxy technology is used to replicate the results of a highly complex single phase CFD simulation with reasonable degree of accuracy while reducing the computational cost associated with such CFD simulations. The CFD model under study simulates the combustion of natural gas under various conditions such as varying natural gas composition and flow rate, inlet air flow rate and temperature, and outlet pressure in a High-Pressure Combustion Facility (B6 Combustor) with more than 4 million simulation cells. Only eight CFD simulation runs were used to create a smart proxy model that replicates the detail distribution of Pressure, Temperature, Nitrogen, Oxygen and Carbon-dioxide concentration in the CFD simulation model in seconds with less than 10% error.

03 NATURAL GAS↗
DOE OSTI2024

A Bayesian Learning Approach to Wireless Outdoor Heatmap Construction using Deep Gaussian Process

We present a novel Bayesian learning approach to outdoor radio heatmap construction utilizing deep Gaussian process (GP). The proposed approach employs a two-layer hierarchy which consists of two cascaded Gaussian processes that are capable of modeling more complex input-output relations than standard single-layer Gaussian processes. Since deriving the exact model likelihood is challenging, a lower bound is optimized instead so that gradient descent-based methods can be performed to find out the optimal model parameters. Typically, inducing points are used in GPs to facilitate low-rank approximation of covariance (kernel) matrices for computation speedup. However, the inaccuracy induced by inducing points can accumulate when stacking multiple layers of GP which may hinder the performance of deep GP. Moreover, since inducing points need to be learned, having them at all layers of deep GP also incurs computational burden. To overcome the above challenges, in contrast to the canonical deep GP model, we use a modified architecture where a full standard GP resides in the first layer and inducing points are only introduced for the second layer. This modified architecture strikes a balance between model accuracy and training complexity. In the proposed model, the noise parameter of the first GP layer is also eliminated to improve the training efficiency as the noise parameter at the output of the second layer suffices to model the uncertainty in the output. The proposed approach is evaluated on real-world datasets, in the form of location-Received Signal Strength (RSS) pairs, collected from the Platform for Open Wireless Data-driven Experimental Research (POWDER) located at the campus of the University of Utah. Experiment results show that the proposed approach can achieve smaller prediction errors on various training and testing data configurations than DNN-based and GP-based methods.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗
DOE OSTI2020

Land Model Testbed: Accelerating Development, Benchmarking and Analysis of Land Surface Models

A Land Model Testbed (LMT), designed to provide a computational framework for systematically assessing model fidelity and supporting rapid development of complex multiscale models, offers a general-purpose workflow for conducting large ensemble simulations of multiple land surface models, post-processing large volumes of model output, and evaluating model results. It leverages existing tools for launching model simulations and the International Land Model Benchmarking (ILAMB) package for assessing model fidelity through comparison with best-available observational datasets. Increased complexity and proliferation of uncertain parameters in process representations in land surface models has driven the need for frequent and intensive testing and evaluating of models to quantify uncertainties and optimize parameters such that results are consistent with observations. The LMT described here meets these needs by providing tools to run thousands of ensemble simulations simultaneously and post-process their output files, by automating execution of an enhanced version of ILAMB with site-specific benchmarks and multivariate functional relationships, and by offering ensemble diagnostics and a customizable dashboard for displaying model performance metrics and associated graphics. We envision the LMT capabilities will serve as a foundational computational resource for a proposed user facility focused on terrestrial multiscale model--data integration.

Sreepathi, Sarat↗
DOE OSTI2022

Adaptive Generation of Training Data for ML Reduced Model Creation

Machine learning proxy models are often used to speed up or completely replace complex computational models. The greatly reduced and deterministic computational costs enable new use cases such as digital twin control systems and global optimization. The challenge of building these proxy models is generating the training data. A naive uniform sampling of the input space can result in a non-uniform sampling of the output space of a model. This can cause gaps in the training data coverage that can miss finer scale details resulting in poor accuracy. While larger and larger data sets could eventually fill in these gaps, the computational burden of full-scale simulation codes can make this prohibitive. In this paper, we present an adaptive data generation method that utilizes uncertainty estimation to identify regions where training data should be augmented. By targeting data generation to areas of need, representative data sets can be generated efficiently. The effectiveness of this method will be demonstrated on a simple one-dimensional function and a complex multidimensional physics model.

Cianciosa, Mark↗
DOE OSTI2023

Generating MCNP Input Files for Unstructured Mesh Geometries

The Los Alamos National Laboratory’s (LANL) Monte Carlo N-Particle (MCNP)1 transport code version 6.3 (also known as MCNP6.3) has the capability for tracking particles on unstructured mesh (UM) geometry models embedded into constructive solid geometry (CSG) cells. This feature has been developed for performing calculations of complex geometry models because manually creating CSG models is time-consuming and error-prone as the complexities of geometries increase. A UM geometry model is a collection of finite elements representing a solid geometry. The first step of the MCNP UM calculation is using other software packages to create a finite element mesh representation of a solid 3D geometry because the MCNP code cannot be used to generate a UM model. Computer-aided design (CAD) software is typically used to create a solid geometry model, which is later imported into mesh generation software to create a UM model. Some mesh generation software packages may also be used to create solid geometries and thus CAD files are not needed. The MCNP UM feature was originally designed for models generated by the Abaqus/CAE software suite. The MCNP code version 6.0 and later can process UM models formatted as Abaqus input files. Starting with a 6.3 version, the MCNP code can process HDF5 mesh input files. We only focus on the UM models formatted as Abaqus input files in this report since currently no external software can be used to generate HDF5 mesh input files for MCNP UM calculations. The MCNP code version 6.3 can be used to convert the Abaqus mesh input files into the HDF5 mesh input files, but this option is typically used by the MCNP code development team to test the HDF5 mesh input file feature. Several software packages (such as Abaqus, Attila4MC, or Cubit) can be used to create the Abaqus input files for MCNP UM calculations. An MCNP UM calculation using an Abaqus model requires two input file types: MCNP and Abaqus input files. The Abaqus input files needed for MCNP UM calcu lations must have the correct Abaqus syntax and meet the additional requirements by the MCNP code. The MCNP code can process only Abaqus input files that make use of part and assembly definitions, where elements in each part must be grouped into one or more element sets (i.e., elset) using *Elset keyword lines with specified naming formats. The MCNP and Abaqus input files required for MCNP UM simulations must be related; pseudo-cells in an MCNP input file must be constructed from mesh model data from an Abaqus input file. For large complex UM models, it is tedious to manually create MCNP UM input files. The um pre op (unstructured mesh pre operations) program with the -m option can be used to create a skeleton MCNP input file from an Abaqus input file [6]. Since the um pre op program was written in Fortran and was not written for optimized performance, this program is a deprecated feature in the MCNP code version 6.3 and may be removed in the next release of the code. To improve calculation flow of multiphysics calculations, a Python3 code called write mcnp um input has been developed to generate an MCNP input file instead of using the um_pre_op -m option. This Python code was initially released to the public in 2020. We have updated this Python code for MCNP6.3 and it was used to generate the MCNP input files used to verify the MCNP6.3 code. The write_mcnp_um_input code is included with the MCNP6.3 code package which will be released to the public through the Radiation Safety Information Computational Center (RSICC) at Oak Ridge National Laboratory. This report is a revision of LA-UR-20-27139 report.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗
DOE OSTI2023

Evaluation of an open forecasting challenge to assess skill of West Nile virus neuroinvasive disease prediction

Abstract Background West Nile virus (WNV) is the leading cause of mosquito-borne illness in the continental USA. WNV occurrence has high spatiotemporal variation, and current approaches to targeted control of the virus are limited, making forecasting a public health priority. However, little research has been done to compare strengths and weaknesses of WNV disease forecasting approaches on the national scale. We used forecasts submitted to the 2020 WNV Forecasting Challenge, an open challenge organized by the Centers for Disease Control and Prevention, to assess the status of WNV neuroinvasive disease (WNND) prediction and identify avenues for improvement. Methods We performed a multi-model comparative assessment of probabilistic forecasts submitted by 15 teams for annual WNND cases in US counties for 2020 and assessed forecast accuracy, calibration, and discriminatory power. In the evaluation, we included forecasts produced by comparison models of varying complexity as benchmarks of forecast performance. We also used regression analysis to identify modeling approaches and contextual factors that were associated with forecast skill. Results Simple models based on historical WNND cases generally scored better than more complex models and combined higher discriminatory power with better calibration of uncertainty. Forecast skill improved across updated forecast submissions submitted during the 2020 season. Among models using additional data, inclusion of climate or human demographic data was associated with higher skill, while inclusion of mosquito or land use data was associated with lower skill. We also identified population size, extreme minimum winter temperature, and interannual variation in WNND cases as county-level characteristics associated with variation in forecast skill. Conclusions Historical WNND cases were strong predictors of future cases with minimal increase in skill achieved by models that included other factors. Although opportunities might exist to specifically improve predictions for areas with large populations and low or high winter temperatures, areas with high case-count variability are intrinsically more difficult to predict. Also, the prediction of outbreaks, which are outliers relative to typical case numbers, remains difficult. Further improvements to prediction could be obtained with improved calibration of forecast uncertainty and access to real-time data streams (e.g. current weather and preliminary human cases). Graphical Abstract

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2020

Extended Finite Element Based Approach in Additive Manufacturing Modeling for Optimizing Highly Complex Manifold in Protonic Ceramic Electrochemical Cells

The objective of this project is to develop a novel numerical approach based on the extended finite element method (XFEM) for modeling moving boundaries of material deposited in additive manufacturing (AM) process with improved accuracy and reduced computational cost. One of the major challenges in simulating AM processes is that the boundaries of the computational domain need to change as material is deposited. Previous approaches have achieved this by activating new finite elements, but this requires a high level of mesh refinement to capture small movement of the boundary. XFEM allows the solution boundary to move independently of the mesh, permitting smooth representation of the evolution of the boundary as material is deposited. As a result, this new method will significantly improve the simulation accuracy while reducing the computational cost compared with existing approaches. Furthermore, this project will bring significant improvements for the design and AM process optimization in an iterative fashion among numerical simuation, parameter optimization, and experimental validation. Specifically, this project will support the protonic ceramic electrochemical cells (PCEC) stack development at INL by providing insights into the density, residual stress, thermal and mechanical properties of the PCEC manifold and interconnect, which is critical in enhancing the system lifetime and reducing the overall cost. Upon the success of proposed development and validation of the proposed approach, simulation will be applied to determine an optimal set of AM process parameters for the PCEC manifold production. This project will consolidate Idaho National Laboratory (INL)’s simulation capabilities provided by the MOOSE framework and the Valhalla AM simulation application, with encouraging expansion to emerging PCEC applications and AM technologies at INL and beyond.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2023

Meshing, Language Server Protocol, and other User-Oriented MOOSE Framework Improvements to Enhance Reactor Analysis Capabilities and Workflows

The open-source Multiphysics Object-Oriented Simulation Environment (MOOSE) framework underpins most of the Nuclear Energy Advanced Modeling and Simulation (NEAMS) physics applications and coupling methods. Users interact with MOOSE in multiple ways to enable the construction of complex multiphysics models, including but not limited to compilation, input creation and syntax validation, meshing, solving the physics problem via MOOSE-based solvers and coupling, and output inspection. In FY23, numerous enhancements have been made to MOOSE to enhance the user experience. Reactor-oriented meshing capabilities in MOOSE have been expanded, an online tutorial was created and hosted on the MOOSE site, and a hands-on workshop was held for more than 100 users. The Language Server Protocol (LSP) capability has been implemented in MOOSE to better communicate correct syntax to the NEAMS Workbench user interface which is commonly used to validate input and submit jobs by users. Finally, several other user-facing improvements were made including the implementation of improved screen output and logging control, enhancements to the initial condition system, and enhancements of the MultiApp system commonly used for coupling. Collectively, these enhancements were driven by user needs and directly improve user ability to construct complex multiphysics models for nuclear reactors.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗
DOE OSTI2024

SysCaps (Language Interfaces for Simulation Surrogates of Complex Systems) [SWR-24-97]

You've found the official code repository for the paper "SysCaps: Language Interfaces for Simulation Surrogates of Complex Systems," presented at the Foundation Models for Science: Progress, Opportunities, and Challenges workshop at NeurIPS 2024. Our paper conjectures that interfaces (both text templates as well as conversational) makes interacting with simulation surrogate models for complex systems more intuitive and accessible for both non-experts and experts. "System captions", or SysCaps, are text-based descriptions of systems based on information contained in simulation metadata. Our paper's goal is to train multimodal regression models that take text inputs (SysCaps) and timeseries inputs (exogenous system conditions such as hourly weather) and regress timeseries simulation outputs (e.g. hourly building energy consumption). The experiments in our paper with building and wind farm simulators, which can be reproduced using this codebase, aim to help us understand whether a) accurate regression in this setting is possible and b) if so, how well can we do it. Paper: https://arxiv.org/abs/2405.19653

Emami, Patrick↗
DOE OSTI2020

Numerical Modeling Experiments of Combined Effects of Ice and Permafrost on Delta Morphodynamics

This dataset contains output from numerical model experiments to test the combined effects of ice thickness and permafrost erodibility on delta dynamics. All experiments were conducted using DeltaRCM, a reduced complexity cellular model. Details of the model setup and experiments are described in full in Piliouras et al., (in revision). This dataset contains results from a total of 48 experiments, representing 16 different experimental conditions, each of which was run in triplicate. There is a control run (3 total experiments) with no ice and no permafrost. We used permafrost erodibilities (here described as erodibility relative to that of non-permafrost, and therefore < 1) of 0.65, 0.8, and 0.95, where 0.95 represents permafrost being only 5% less erodible than non-permafrost. We used ice thicknesses of 0.5 m, 3 m, and 4.5 m in a 5 m deep basin. We conducted experiments for each possible combination of permafrost erodibility and ice thickness (9 conditions, 27 experiments). We also provide results for experiments with only ice or permafrost (6 conditions, 18 experiments). The excel table (csv file) provides the ice and permafrost conditions for each run ID. Zip files contain model outputs and other variables generated during analysis (see ReadMe file) as numpy arrays (.npy files), which can be accessed using python.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2021

A software environment for effective reliability management for pulsed power design

The reliable design of magnetically insulated transmission lines (MITLs) for very high current pulsed power machines must be accomplished in the future by utilizing a variety of sophisticated modeling tools. The complexity of the models required is high and the number of sub-models and approximations large. The potential for significant analyst error using a single tool is large, with possible reliability issues associated with the plasma modeling tools themselves or the chosen approach by the analyst to solve a given problem. Here, we report on a software infrastructure design that provides a workable framework for building self-consistent models and constraining feedback to limit analyst error. The framework and associated tools aid the development of physical intuition, the development of increasingly sophisticated models, and the comparison of performance results. The work lays the computational foundation for designing state-of-the-art pulsed-power experiments. The design and useful features of this environment are described. We discuss the utility of the Git source code management system and a GitLab interface for use in project management that extends beyond software development tasks.

42 ENGINEERING↗
DOE OSTI2020

Multi-model Hydroclimate Projections for the Alabama-Coosa-Tallapoosa River Basin in the Southeastern United States

Abstract This study uses a high-resolution, process-based modeling framework to assess the impacts of changing climate on water resources for the Alabama-Coosa-Tallapoosa River Basin in the southeastern United States. A 33-member ensemble of hydrologic projections was generated using 3 distributed hydrologic models (Precipitation-Runoff Modeling System, Variable Infiltration Capacity, and Distributed Hydrology Soil Vegetation Model) of different complexity. These hydrologic models were driven by dynamically downscaled and bias-corrected future climate simulations from 11 Coupled Model Intercomparison Project Phase 5 global climate models under Representative Concentration Pathway 8.5 emission scenario, with 40 years each in baseline (1966–2005) and future (2011–2050) periods. The hydroclimate response, in general, projects an increase in mean seasonal precipitation, runoff, and streamflow. The high and low flows are projected to increase and decrease, respectively, in general, suggesting increased likelihood of extreme rainfall events and intensification of the hydrologic cycle. The uncertainty associated with the ensemble hydroclimate response, analyzed through an analysis of variance technique, suggests that the choice of climate model is more critical than the choice of hydrologic model for the studied region. This study provides in-depth insights of hydroclimate response and associated uncertainties to support informed decisions by water resource managers.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2023

Topological structure of complex predictions

Abstract Current complex prediction models are the result of fitting deep neural networks, graph convolutional networks or transducers to a set of training data. A key challenge with these models is that they are highly parameterized, which makes describing and interpreting the prediction strategies difficult. We use topological data analysis to transform these complex prediction models into a simplified topological view of the prediction landscape. The result is a map of the predictions that enables inspection of the model results with more specificity than dimensionality-reduction methods such as tSNE and UMAP. The methods scale up to large datasets across different domains. We present a case study of a transformer-based model previously designed to predict expression levels of a piece of DNA in thousands of genomic tracks. When the model is used to study mutations in the BRCA1 gene, our topological analysis shows that it is sensitive to the location of a mutation and the exon structure of BRCA1 in ways that cannot be found with tools based on dimensionality reduction. Moreover, the topological framework offers multiple ways to inspect results, including an error estimate that is more accurate than model uncertainty. Further studies show how these ideas produce useful results in graph-based learning and image classification.

Computer Science↗
DOE OSTI2020

High-Fidelity Large-Signal Order Reduction Approach for Composite Load Model

With the increasing penetration of electronic loads and distributed energy resources, conventional load models cannot capture their dynamics. Therefore, a new comprehensive composite load model is developed by Western Electricity Coordinating Council (WECC). However, this model is a complex high-order non-linear system with multi-time-scale property, which poses challenges on power system studies with heavy computational burden. In order to reduce the model complexity, the authors firstly develop a large-signal order reduction (LSOR) method using singular perturbation theory. In this method, the fast dynamics are integrated into the slow ones to preserve transient characteristics of the former. Then, accuracy assessment conditions are proposed and embedded into the LSOR to improve and guarantee the accuracy of reduced-order model. Finally, the reduced-order WECC composite load model is derived by using the proposed algorithm. Overall, simulation results show that the reduced-order large-signal model significantly alleviates the computational burden while maintaining similar dynamic responses as the original composite load model.

24 POWER TRANSMISSION AND DISTRIBUTION↗
DOE OSTI2026

Model Parameter Development for Complex Materials: Species-Specific Diffusion Barriers in 316 Stainless Steel from Systematic DFT Calculations

Vacancy-mediated diffusion barriers in 316 stainless steel have been systematically calculated using density functional theory to provide essential parameters for mesoscale microstructure evolution models. A statistical sampling approach employing 210 nudged elastic band calculations across multiple special quasi-random structures captures the effects of local chemical environments in this concentrated alloy. The computational methodology addresses challenges specific to chemically disordered systems, including proper magnetic treatment throughout multi-step calculations and validation against experimental structural properties. The calculated activation barriers reveal clear species-dependent diffusion behavior with the hierarchy Ni >> Fe ˜ Cr >> Mo. Nickel exhibits the highest barriers (0.74–1.31 eV, mean 1.045 eV), confirming its role as the slowest-diffusing major component. Iron and chromium show similar moderate barriers averaging 0.587 eV and 0.522 eV, respectively. Remarkably, molybdenum demonstrates exceptionally low barriers (0.12–0.28 eV, mean 0.194 eV), suggesting much higher mobility than previously recognized and potentially significant implications for precipitation kinetics and microstructure evolution. The barrier ranges remain consistent across different 316 SS compositions, supporting parameter transferability for modeling applications. The overall mean barrier of 0.64 eV provides a practical approximation for phase field simulations, while species-specific values enable detailed treatments of diffusion-controlled processes. This systematic approach establishes a validated framework for generating diffusion parameters in other concentrated alloys where experimental data are limited, while providing the first systematic set of species-specific barriers for predictive modeling of 316 stainless steel microstructure evolution.

36 MATERIALS SCIENCE↗
DOE OSTI2023

Generating Mixed Patterns of Residential Segregation: An Evolutionary Approach

The Schelling model of residential segregation has demonstrated that even the slightest preference for neighbors of the same race can be amplified into community-wide segregation. However, these models are unable to simulate mixed, coexisting patterns of segregation and integration, which have been seen to exist in cities. Using evolutionary model discovery we demonstrate how including social factors beyond racial bias when modeling relocation behavior enables the emergence of strongly mixed patterns. Our results indicate that the emergence of mixed patterns is better explained by multiple factors influencing the decision to relocate; the most important being the interaction of nonlinear, rapidly diminishing racial bias with a recent, historical tendency to move. Additionally, preference for less isolated neighborhoods or preference for neighborhoods with longer residing neighbors may produce weaker mixed patterns. Finally, this work highlights the importance of exploring the influence of multiple hypothesized factors of decision making, and their interactions, within agent rules, when studying emergent outcomes generated by agent-based models of complex social systems.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2023

Athena-I CUBIT Journal Files

The Monte Carlo N-Particle (MCNP)1 transport code version 6 (also known as MCNP6) has the capability for tracking particles on unstructured mesh (UM) geometry models embedded into con structive solid geometry (CSG) cells. This feature has been developed for performing calculations of complex geometry models because manually creating CSG models is a time-consuming and error prone process as the complexities of geometries increase. The UM capability was originally designed to work with UM models created with the Abaqus software and ASCII input files that it generates. The MCNP code version 6.0 and later can process UM models formatted as Abaqus input files. Starting with version 6.3, MCNP can also process HDF5 mesh input files. External codes must be used to generate Abaqus input files for MCNP UM calculations. CUBIT, the Sandia National Laboratory automated mesh generation toolkit, can generate a UM model formatted as an Abaqus input file. However, the Abaqus input files exported from CUBIT cannot be used for MCNP simulations because it lacks the proper syntax. A Python script was developed to convert an Abaqus file created by CUBIT to an Abaqus file format that MCNP can process. Creating UM models for complex geometries is not an easy task. The process of creating UM models in CUBIT for MCNP simulations is detailed in. CUBIT provides several user interface options including a graphical user interface (GUI) and a command line interface. A GUI provides an easy way to use CUBIT without learning the CUBIT command syntax. When using CUBIT with either interface option, command lines are written into an ASCII file known as a journal file; this journal file can be edited and archived so that it can be played back in CUBIT to automatically generate a UM model. This report describes the CUBIT journal files of the UM models developed for Athena-I. The Athena platform, an energy-tuning assembly, was developed to spectrally shape the National Ignition Facility (NIF) deuterium-tritium fusion neutron source to a thermonuclear (fusion) plus prompt fission neutron spectrum with capability to act as a short-pulse neutron source. MCNP6 was used for the Athena experiment design analysis.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗
DOE OSTI2026

When does global attention help: a unified empirical study on atomistic graph learning

Graph neural networks (GNNs) are widely used as surrogates for costly experiments and first-principles simulations to study the behavior of compounds at atomistic scale, and their architectural complexity is constantly increasing to enable the modeling of complex physics. While most recent GNNs combine more traditional message passing neural networks (MPNNs) layers to model short-range interactions with more advanced graph transformers (GTs) with global attention mechanisms to model long-range interactions, it is still unclear when global attention mechanisms provide real benefits over well-tuned MPNN layers due to inconsistent implementations, features, or hyperparameter tuning. We introduce the first unified, reproducible benchmarking framework–built on HydraGNN–that enables seamless switching among four controlled model classes: MPNN, MPNN with chemistry/topology encoders, GPS-style hybrids of MPNN with global attention, and fully fused localglobal models with encoders. Using seven diverse open-source datasets for benchmarking across regression and classification tasks, we systematically isolate the contributions of message passing, global attention, and encoder-based feature augmentation. Our study shows that encoder-augmented MPNNs form a robust baseline, while fused localglobal models yield the clearest benefits for properties governed by long-range interaction effects. We further quantify the accuracycompute trade-offs of attention, reporting its overhead in memory. Together, these results establish the first controlled evaluation of global attention in atomistic graph learning and provide a reproducible testbed for future model development.

Equivariant graph neural networks↗