Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “surrogate systems modeling”

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 343 records · Page 19
All sourcesNASA NTRSDOE OSTI
DOE OSTI2024

Probabilistic projections of the Amery Ice Shelf catchment, Antarctica, under conditions of high ice-shelf basal melt

Abstract. Antarctica's Lambert Glacier drains about one-sixth of the ice from the East Antarctic Ice Sheet and is considered stable due to the strong buttressing provided by the Amery Ice Shelf. While previous projections of the sea-level contribution from this sector of the ice sheet have predicted significant mass loss only with near-complete removal of the ice shelf, the ocean warming necessary for this was deemed unlikely. Recent climate projections through 2300 indicate that sufficient ocean warming is a distinct possibility after 2100. This work explores the impact of parametric uncertainty on projections of the response of the Lambert–Amery system (hereafter “the Amery sector”) to abrupt ocean warming through Bayesian calibration of a perturbed-parameter ice-sheet model ensemble. We address the computational cost of uncertainty quantification for ice-sheet model projections via statistical emulation, which employs surrogate models for fast and inexpensive parameter space exploration while retaining critical features of the high-fidelity simulations. To this end, we build Gaussian process (GP) emulators from simulations of the Amery sector at a medium resolution (4–20 km mesh) using the Model for Prediction Across Scales (MPAS)-Albany Land Ice (MALI) model. We consider six input parameters that control basal friction, ice stiffness, calving, and ice-shelf basal melting. From these, we generate 200 perturbed input parameter initializations using space filling Sobol sampling. For our end-to-end probabilistic modeling workflow, we first train emulators on the simulation ensemble and then calibrate the input parameters using observations of the mass balance, grounding line movement, and calving front movement with priors assigned via expert knowledge. Next, we use MALI to project a subset of simulations to 2300 using ocean and atmosphere forcings from a climate model for both low- and high-greenhouse-gas-emission scenarios. From these simulation outputs, we build multivariate emulators by combining GP regression with principal component dimension reduction to emulate multivariate sea-level contribution time series data from the MALI simulations. We then use these emulators to propagate uncertainty from model input parameters to predictions of glacier mass loss through 2300, demonstrating that the calibrated posterior distributions have both greater mass loss and reduced variance compared to the uncalibrated prior distributions. Parametric uncertainty is large enough through about 2130 that the two projections under different emission scenarios are indistinguishable from one another. However, after rapid ocean warming in the first half of the 22nd century, the projections become statistically distinct within decades. Overall, this study demonstrates an efficient Bayesian calibration and uncertainty propagation workflow for ice-sheet model projections and identifies the potential for large sea-level rise contributions from the Amery sector of the Antarctic Ice Sheet after 2100 under high-greenhouse-gas-emission scenarios.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2022

Uncertainty Aware Deep Learning for Particle Accelerators

Standard deep learning models for classification and regression applications are ideal for capturing complex system dynamics. However, their predictions can be arbitrarily inaccurate when the input samples are not similar to the training data. Implementation of distance aware uncertainty estimation can be used to detect these scenarios and provide a level of confidence associated with their predictions. In this paper, we present results from using Deep Gaussian Process Approximation (DGPA) methods for errant beam prediction at Spallation Neutron Source (SNS) accelerator (classification) and we provide an uncertainty aware surrogate model for the Fermi National Accelerator Lab (FNAL) Booster Accelerator Complex (regression).

Rajput, Kishansingh↗
DOE OSTI2025

Application of machine learning interatomic potentials in heterogeneous catalysis

Heterogeneous catalysts are crucial in modern societies as they promote sustainability by enabling lower-energy pathways for various chemical reactions. While Density Functional Theory (DFT) computations can provide critical insights into how heterogeneous catalysts operate at the atomic level, they are limited by computational costs and unfavorable scaling with system size. Recently, machine learning interatomic potentials (MLIPs) have emerged as a promising alternative to DFT, offering near-DFT accuracy at significantly reduced cost. Here, in this perspective, we discuss the application of MLIPs in heterogeneous catalyst modeling as a surrogate for DFT. We detail how MLIPs have been applied in thermal catalysis to probe active sites, enable studying complex metallic and nanoporous catalysts, and investigate the reconstruction of catalytic surfaces. We review the use of MLIPs in electrocatalysis and photocatalysis, emphasizing their capabilities in studying transition metal oxide surfaces and solid–liquid interfaces. We also discuss the current limitations of MLIPs, particularly their challenges with transferability and description of non-local interactions. Finally, we conclude by identifying promising and underexplored domains in which MLIPs can further advance our understanding of heterogeneous catalysts.

Catalytic surfaces↗
DOE OSTI2020

Fast Computational Algorithms for Partial Differential Equations and Uncertainty Quantifications

This project concerned the construction, testing and analysis of computational algorithms for solving parameterized and stochastic partial differential equations. The study and understanding of equations of this type is of fundamental importance in numerous engineering and scientific applications. Examples include simulation of plasma dynamics in models of electric propulsion and nuclear fusion, simulation of multiphase flows, such as the flow of water, gas and oil in reservoirs, and structural analysis of the dependence of structures on materials. Parametrization is used in such settings when properties of the models such as viscosity of fluids or electric resistivity of materials are not precisely understood and instead are treated as random variables. The resulting solutions are themselves random, and having such solutions will enable engineers to use probabilistic methods to assess the likelihood of events, for example, whether a pollutant in a liquid will exceed a limit, and to use such analyses to develop ways to ensure positive outcomes. Construction of accurate (high resolution) computational solutions is expensive, requiring significant computer time and computational resources, and there is need to reduce computational cost to make simulation useful and effective. The aim of the project was to construct algorithms to efficiently compute surrogate solutions to parameterized problems to allow for efficient and accurate simulation. The technical approach used focused on two related strategies, based on rank-reduction methods and reduced-order models. These methods construct surrogate solutions of parameter-dependent models by projection or interpolation into low-dimensional approximation spaces. Cost savings are achieved if the low-dimensional spaces can be identified and constructed efficiently and if the resulting low-dimensional algebraic systems can be solved cheaply. Accomplishments include: Theoretical and empirical demonstration of the effectiveness of fast multigrid solution strategies for computing low-rank representations of parameter-dependent solutions to discrete partial differential equations, including the first proof establishing so-called textbook convergence properties for low-rank methods. Development of efficient solution algorithms for solving nonlinear parameter-dependent partial differential equations used in models of fluid dynamics. Developent of efficient algorithms for low-rank representation of solutions of time-dependent simulations of fluid dynamics using multi-dimensional tensor representations of solutions.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2020

Assessment of the 3420 Building Filtered Exhaust Stack Sampling Probe Location: Stack Verification Following Fan and Air Blender Additions

The Pacific Northwest National Laboratory (PNNL) 3420 Building, which is part of the Physical Sciences Facility (PSF), houses radiological capabilities that results in the requirement that emissions monitoring must be conducted for potential radionuclides in the exhaust air discharge of this building. The air monitoring system is required to conform to Title 40 of the Code of Federal Regulations part 61 (40 CFR 61) Subpart H, which in turns requires a sampling probe in the exhaust stream to conform to the criteria of American National Standards Institute / Health Physics Society (ANSI/HPS) N13.1-2011, Sampling and Monitoring Releases of Airborne Radioactive Substances from the Stack and Ducts of Nuclear Facilities. To support the air emissions permit for the 3420 Building on the PNNL campus, stack testing that used computational fluid dynamics (CFD) modeling as a surrogate stack and verification tests of velocity uniformity and flow angle on the retrofitted facility stack was performed. The ANSI/HPS N13.1-2011 criteria for the air monitoring probe location are that velocity uniformity, gaseous tracer uniformity, and particulate tracer uniformity must be less than or equal to 20%COV. Furthermore, no point in the sampling location may have a gaseous tracer concentration that varies from the mean concentration by more than 30%. Additionally, the flow angle at the sampling location must not be more than 20°. The CFD modeling of the stack, as reported by Recknagle et al. (2018) demonstrated that the stack meets the criteria at the probe location. The velocity uniformity and flow angle results from the 3420 stack verification tests, performed in October 2020, demonstrated that the CFD model results may be used to support the qualification of the stack sampling location. The measured velocity uniformity verification test result was 1.4%COV. This value is well within the uniformity criterion, which is that the velocity uniformity be =20%COV. Additionally, this value is well within the criterion that the actual stack measurement must be within 5% of the surrogate stack result of 2.1%COV when all four fans were operating. Additionally, the measured average flow angle at the 3420 stack monitor location was 15.5 degrees. Although this is higher than expected based on the CFD model, the result is =20 degrees, so the criterion is met. Based on these stack verification test results, the reconfigured 3420 Building filtered exhaust stack meets the qualification criteria given in the ANSI/HPS N13.1-2011 standard. Further changes to the system configuration or operating conditions that are outside the bounds described in this and the CFD report (Recknagle et al., 2018) may require additional tests and additional analysis to determine compliance with the standard.

3420 Building↗
DOE OSTI2024

Towards accelerating particle-resolved direct numerical simulation with neural operators

In this paper, we present our ongoing work aimed at accelerating a particle-resolved direct numerical simulation model designed to study aerosol–cloud–turbulence interactions. The dynamical model consists of two main components—a set of fluid dynamics equations for air velocity, temperature, and humidity, coupled with a set of equations for particle (i.e., cloud droplet) tracing. Rather than attempting to replace the original numerical solution method in its entirety with a machine learning (ML) method, we consider developing a hybrid approach. We exploit the potential of neural operator learning to yield fast and accurate surrogate models and, in this study, develop such surrogates for the velocity and vorticity fields. We discuss results from numerical experiments designed to assess the performance of ML architectures under consideration as well as their suitability for capturing the behavior of relevant dynamical systems.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2022

Milestone 1.2.11: H 2 Production from Surrogate Non-Native Corrosion Plumes on Aluminum 6061-T6 Fuel Cladding Surrogates

Thick, localized, “non-native” corrosion plumes have been identified on Advanced Test Reactor fuel elements, raising concern on their impact on the radiolytic formation of molecular hydrogen gas (H 2 ) from aluminum-clad spent nuclear fuel (ASNF) under proposed extended (> 50 years) dry storage conditions. Here, we report our findings on H 2 generation from the gamma irradiation (up to 52 MGy) of surrogate “non-native” corrosion plume coupons: ambient-temperature-corroded (~350 days in water) aluminum alloy 6061 (AA6061) coupons in helium gas environments with ~0% added relative humidity. Additionally, we provide a comparison of proposed ASNF drying techniques— vacuum drying only, vacuum drying + 100 °C for 4 hr, and vacuum drying + 220 °C for 4 hr—on the yield of H 2 from these surrogate systems. The presented data indicates that similar amounts of H 2 (~2 × 10–3 µmol J–1) are formed from gamma irradiated AA6061 coupons corroded under different temperature regimes, i.e., ambient/350 days vs. 90 C/30 days. These findings validate current, complimentary modelling predictions based on high-temperature-corrosion irradiation data only. Further, the application of a heat treatment procedure (100 and 220 °C), in conjunction with vacuum drying, accelerated the rate at which a steady-state H 2 yield was attained, in comparison to vacuum only, due to the removal of H 2 precursors in the form of adsorbed waters. Interestingly, within the confidence limits of our measurements, negligible difference in total H 2 yield was found between the two investigated heat treatment procedures.

38 RADIATION CHEMISTRY, RADIOCHEMISTRY, AND NUCLEA↗
DOE OSTI2023

Integrated Process Testing of MSR Salt Spill Accidents

Part of the licensing process for new nuclear reactors requires vendors to assess the potential consequences of identified accident scenarios using accident progression modeling. The accident scenario that will likely be evaluated by all molten salt reactor (MSR) developers is a spill of radionuclide-bearing fuel salt onto the reactor containment floor (i.e., a salt spill accident). The development of accident progression models requires experimental data to inform which processes to incorporate, to enable the calculation of parameters to model these processes, and to validate the model predictions. The data should quantify the sensitivities of key processes (e.g., molten salt spreading, heat transfer, containment structure corrosion, radionuclide vaporization, and aerosol generation) towards the initial conditions of the spill, the ambient environment, and the features of the containment. In addition, results from integrated process tests that quantify coupled processes are required to validate systems-level models. This report documents results from integrated process tests conducted on simulated molten salt spill accidents. The generated experimental data simultaneously quantify the heat transfer behavior of the spilled salt, compositional changes to the bulk salt, and the release of surrogate fission products from the spilled salt as aerosol particles. All tests that were conducted used FLiNaK doped with surrogate fission products, and the variables that were evaluated included the initial salt temperature and the concentration of surrogate fission products present in the salt. The major accomplishments of this work include identifying surrogate fuel salt compositions that provide insight into the dispersal behavior of radionuclides of potential significance to the source term, employing previously developed methods and measurement techniques to simultaneously measure key processes, generating data on the coupled processes of molten salt heat transfer and surrogate fission product release as aerosol particles, demonstrating new test methods for real-time monitoring of the flow rate of the spill and aerosol size quantification in an argon atmosphere, and developing a mass transfer model for cesium and iodine release from molten FLiNaK to provide insight into aerosol formation by vapor condensation. The same methodology applied herein can be employed to study different salt compositions of interest to MSR developers, different environmental conditions, and other variables that are relevant to postulated accident scenarios. The insights gained from these integrated process tests conducted at a laboratory scale will be incorporated into future integral effects tests conducted at an engineering scale.

20 FOSSIL-FUELED POWER PLANTS↗
DOE OSTI2026

Heliostat Sizing Methodology for Solar Heat for Industrial Processes

This study presents a method to obtain a heliostat size that minimizes the levelized cost of a heliostat-based concentrating solar thermal system for industrial process heat (IPH) applications at operating temperatures from 565 to 1550 degrees Celsius. The method extends prior work by embedding a routine for system design that obtains near-optimal subsystem sizes, increasing the fidelity of drive cost functions, and adding an optical performance model. An illustrative business case is developed for Daggett, California, targeting specified annual thermal energy outputs of 50 to 400 GWhth. Optical performance is modeled using verified estimates from the literature. A surrogate heliostat cost model, derived from commercial heliostat designs and scaled for production volume, installation, and operations and maintenance costs, is used to develop cost functions. Results show that heliostat size strongly affects the levelized cost of heat (LCOH), producing a characteristic U-shaped trend with a robust near-optimal window of 8 - 12 m2; the heliostat size producing the lowest project cost in our study grows slightly as the project size increases, and is reduced as the operating temperature increases. The findings in this study are consistent with the general trend of smaller heliostats under deployment at existing projects for high-temperature industrial process heat and reflect the significant reduction in power electronics and other per-heliostat costs. The methodology we propose is general and can be tailored to revised cost curves as the technology continues to evolve.

14 SOLAR ENERGY↗
DOE OSTI2025

Divertor Plasma Detachment Control Neural Network

DivControlNN is a state-of-the-art software tool that leverages advanced machine learning techniques to predict and control divertor plasma behavior in fusion reactors. Plasma, a highly energetic and electrically charged gas, requires meticulous management to protect reactor components and maintain optimal energy production. Conventional simulation methods, although extremely detailed, typically demand extensive computational time-making them unsuitable for real-time control scenarios. DivControlNN addresses this challenge by learning from tens of thousands of high-fidelity simulations, thereby creating a rapid surrogate model that can deliver near-instantaneous predictions. At the core of its functionality is a sophisticated technique known as latent space mapping, which condenses complex, high-dimensional plasma data into a compact, lower-dimensional representation. This streamlined representation enables the system to quickly forecast essential plasma properties and determine the precise conditions required for effective detachment. Detachment is a crucial process in which the plasma is cooled before reaching the divertor plates, thereby reducing heat loads and mitigating material erosion. In recent experiments conducted on the KSTAR tokamak in South Korea, DivControlNN successfully guided the detachment process without any fine-tuning-even when applied to a new tungsten divertor configuration. By achieving a computational speed-up of over one hundred million times compared to traditional simulation methods while maintaining low prediction errors, DivControlNN stands to significantly enhance real-time control and diagnostic capabilities in future fusion reactors. This breakthrough paves the way for safer, more reliable reactor operation and represents a major advancement toward realizing fusion energy as a practical, sustainable, and clean power source.

Xu, Xueqiao [Lawrence Livermore National Laborator↗
DOE OSTI2021

An adaptive approach to machine learning for compact particle accelerators

Abstract Machine learning (ML) tools are able to learn relationships between the inputs and outputs of large complex systems directly from data. However, for time-varying systems, the predictive capabilities of ML tools degrade if the systems are no longer accurately represented by the data with which the ML models were trained. For complex systems, re-training is only possible if the changes are slow relative to the rate at which large numbers of new input-output training data can be non-invasively recorded. In this work, we present an approach to deep learning for time-varying systems that does not require re-training, but uses instead an adaptive feedback in the architecture of deep convolutional neural networks (CNN). The feedback is based only on available system output measurements and is applied in the encoded low-dimensional dense layers of the encoder-decoder CNNs. First, we develop an inverse model of a complex accelerator system to map output beam measurements to input beam distributions, while both the accelerator components and the unknown input beam distribution vary rapidly with time. We then demonstrate our method on experimental measurements of the input and output beam distributions of the HiRES ultra-fast electron diffraction (UED) beam line at Lawrence Berkeley National Laboratory, and showcase its ability for automatic tracking of the time varying photocathode quantum efficiency map. Our method can be successfully used to aid both physics and ML-based surrogate online models to provide non-invasive beam diagnostics.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2024

Numerical Simulation and Experimental Comparison of System Analysis Module 1D Mixing Model for Cold Shock Transients in the Gallium Thermal-Hydraulic Mixing Facility

Abstract Liquid metals are being investigated as coolants in many advanced reactor designs because of their high thermal conductivity and effectiveness at high temperatures. However, they often pose challenges to reactor operation and safety because of the complex thermal mixing and stratification in the plenum of pool-type reactor designs. The advanced system analysis code System Analysis Module (SAM) currently under development at Argonne National Laboratory aims to develop and implement thermal mixing models to accurately capture these complex thermal fluid behaviors. In this study, the SAM thermal mixing model was compared against experimental data from the Gallium Thermal-Hydraulic Experiment facility, a scaled liquid metal test facility that uses gallium as a surrogate fluid to investigate the stratification and thermal mixing of low-Prandtl-number fluids in the upper plenum of a liquid metal-cooled reactor. Two cold shock transient cases were used: one with stable stratified flow (Ri = 32) and one with stronger thermal mixing (Ri = 0.5). The resultant temperatures were then compared with the experimental temperatures over the entire plenum to assess the ability of the mixing models to capture the thermal behavior and to better correspond mixing parameters to various flow scenarios. Generally, the zero-dimensional mixing model was more capable of capturing the bulk temperature of the component modeled assuming that an accurate mass flow rate was provided, but it was inherently unable to capture thermal gradients in space. The one-dimensional mixing model was capable of capturing that the thermal gradients provided accurate selection of the mixing coefficients. Further, the temperature at the outlet junction was compared over time for each of the mixing models with the recorded experimental temperature. The implemented mixing models demonstrated the ability to effectively capture the overall thermal behavior for stronger mixing scenarios but struggled with more stably stratified flows. It was found that a system analysis code's covering of the entire range of different operating conditions still remains a challenging task, and it is suggested that further model and closure improvements are necessary to accurately capture complex thermal mixing and stratification phenomena.

stratification↗
DOE OSTI2025

Artificial Intelligence/Deep Learning FRNN Software for Prediction & Real-Time Control of DIII-D Plasma Control System (PCS)

This collaborative project integrated an improved version of the Artificial Intelligence/Deep Learning FRNN prediction and control software into the real-time DIII-D PCS (plasma control system). A key AI/DL software challenge is to build a modern high-performance computing (HPC) enabled “synthetic plasma simulator” capable of carrying out HPC-driven real-time plasma control applications. This involves development of a deep learning framework to train the surrogate model for a first-principles-based instability analysis simulator (“SGTC”) derived from the global gyrokinetic code GTC. The role of SGTC is to provide accurate and detailed plasma instability information from a real-time AI-based simulator capability to complement the deep learning prediction and control from experimentally-measured signals, such as ECE Imaging, supplemented by synthetic SGTC-ECEI.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2026

Deriving effective electrode–ion interactions from free-energy profiles at electrochemical interfaces

Understanding ion adsorption at electrified metal–electrolyte interfaces is essential for accurate modeling of electrochemical systems. Here, in this study, we systematically investigate the free energy profiles of Na + , Cl − , and F − ions at the Au(111)–water interface using enhanced sampling molecular dynamics with both classical force fields and machine-learned interatomic potentials (MLIPs). Our classical metadynamics results reveal a strong dependence of predicted ion adsorption on the Lennard-Jones parameters, highlighting that—without due care—standard mixing rules can lead to qualitatively incorrect descriptions of ion–metal interactions. We present a systematic methodology for tuning the cross term LJ parameters to control adsorption energetics in agreement with more accurate models. As a surrogate for an ab initio model, we employed the recently released Universal Models for Atoms MLIP, which validates classical trends and displays strong specific adsorption for chloride, weak adsorption for fluoride, and no specific adsorption for sodium, in agreement with experimental and theoretical expectations. By integrating molecular-level adsorption free energies into continuum models of the electric double layer, we show that specific ion adsorption substantially alters the interfacial ion population, the potential of zero charge, and the differential capacitance of the system. Our results underscore the critical importance of force field parameterization and advanced interatomic potentials for the predictive modeling of ion-specific effects at electrified interfaces and provide a robust framework for bridging molecular simulations and continuum electrochemical models.

Roncoroni, Fabrice [Lawrence Berkeley National Lab↗
DOE OSTI2024

Colloidal Stability of PFSA-Ionomer Dispersions Part II: Determination of Suspension pH Using Single-Ion Potential Energies

Perfluorosulfonic acid (PFSA) ionomers serve a vital role in the performance and stability of fuel-cell catalyst layers. These properties, in turn, depend on the colloidal processing of precursor inks. Here, to understand the colloidal structure of fuel-cell catalyst layers, we explore the aggregation of PFSA ionomers dissolved in water/alcohol solutions and relate the predicted aggregation to experimental measurements of solution pH. Not all side chains contribute to measured pH because of burying inside particle aggregates. To account for the measured degree of dissociation, a new description is developed for how PFSA aggregates interact with each other. The developed single-counterion electrostatic repulsive pair potential from Part I is incorporated into the Smoluchowski collision-based kinetics of interacting aggregates with buried side chains. We demonstrate that the surrounding solvent mixture affects the degree of aggregation as well as the pH of the system primarily through the solution dielectric permittivity, which drives the strength of the interparticle repulsive energies. Successful pH prediction of Nafion ionomer dispersions in water/n-propanol solutions validates the numerical calculations. Nafion-dispersion pH measurements serve as a surrogate for Nafion particle-size distributions. The model and framework can be leveraged to explore different ink formulations.

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

Neural network model of neutral beam injection in the EAST tokamak to enable fast transport simulations

The neutral beam injection (NBI) system in EAST produces energetic neutral particles, which collide with electrons and ions in tokamak plasmas and heat the plasmas through Coulomb collisions. Moreover, it drives a non-inductive source of current, due to the charge-exchange collision between neutral particles and ions, and injects toroidal torque, which generates a toroidal rotation of the plasma. The effect caused by the NBI system, such as plasma heating, current drive, total neutron rate, momentum transfer, and shine-through, are modeled by a comprehensive module called NUBEAM. However, NUBEAM is computationally intensive since it relies on Monte Carlo methods. In this work, a neural network model has been developed as a surrogate model for NUBEAM in EAST. The database for neural-network model training, validation and testing is generated by running TRANSP for experimental discharges from recent EAST campaigns (after the latest NBI upgrade) while using the NUBEAM module. Simulation results illustrate that the trained neural network has the capability of replicating the predictions made by NUBEAM while demanding a significantly shorter execution time. Finally, these results indicate that surrogate models like the one proposed in this work could enable fast transport simulations for EAST after integrating them into a control-oriented predictive code such as COTSIM.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2021

Statistical and systematic uncertainties in extracting the source properties of neutron star-black hole binaries with gravitational waves

Gravitational waves emitted by neutron star black hole mergers encode key properties of neutron stars - such as their size, maximum mass and spins - and black holes. However, the presence of matter and the high mass ratio makes generating long and accurate waveforms from these systems hard to do with numerical relativity, and not much is known about systematic uncertainties due to waveform modeling. In this work, we simulate gravitational waves from neutron star black hole mergers by hybridizing numerical relativity waveforms produced with the SpEC code with a recent numerical relativity surrogate NRHybSur3dq8Tidal. These signals are analyzed using a range of available waveform families, and statistical and systematic errors are reported. We find that at a network signal-to-noise ratio (SNR) of 30, statistical uncertainties are usually larger than systematic offsets, while at an SNR of 70 the two become comparable. The individual black hole and neutron star masses, as well as the mass ratios, are typically measured very precisely, though not always accurately at high SNR. At a SNR of 30 the neutron star tidal deformability can only be bound from above, while for louder sources it can be measured and constrained away from zero. All neutron stars in our simulations are non-spinning, but in no case we can constrain the neutron star spin to be smaller than ~0.4 (90% credible interval). Waveform families whose late inspiral has been tuned specifically for neutron star black hole signals typically yield the most accurate characterization of the source parameters. Their measurements are in tension with those obtained using waveform families tuned against binary neutron stars, even for mass ratios that could be relevant for both binary neutron stars and neutron star black holes mergers.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2022

Scalable training of graph convolutional neural networks for fast and accurate predictions of HOMO-LUMO gap in molecules

Abstract Graph Convolutional Neural Network (GCNN) is a popular class of deep learning (DL) models in material science to predict material properties from the graph representation of molecular structures. Training an accurate and comprehensive GCNN surrogate for molecular design requires large-scale graph datasets and is usually a time-consuming process. Recent advances in GPUs and distributed computing open a path to reduce the computational cost for GCNN training effectively. However, efficient utilization of high performance computing (HPC) resources for training requires simultaneously optimizing large-scale data management and scalable stochastic batched optimization techniques. In this work, we focus on building GCNN models on HPC systems to predict material properties of millions of molecules. We use HydraGNN, our in-house library for large-scale GCNN training, leveraging distributed data parallelism in PyTorch. We use ADIOS, a high-performance data management framework for efficient storage and reading of large molecular graph data. We perform parallel training on two open-source large-scale graph datasets to build a GCNN predictor for an important quantum property known as the HOMO-LUMO gap. We measure the scalability, accuracy, and convergence of our approach on two DOE supercomputers: the Summit supercomputer at the Oak Ridge Leadership Computing Facility (OLCF) and the Perlmutter system at the National Energy Research Scientific Computing Center (NERSC). We present our experimental results with HydraGNN showing (i) reduction of data loading time up to 4.2 times compared with a conventional method and (ii) linear scaling performance for training up to 1024 GPUs on both Summit and Perlmutter.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗