Search NASA⌕ Search

SEARCH · Search NASA

Results for “computer system benchmarking”

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 1,027 records · Page 57

HPC Analytics Support. Requirements for Uncertainty Quantification Benchmarks

This report outlines techniques for extending benchmark generation products so they support uncertainty quantification by benchmarked systems. We describe how uncertainty quantification requirements can be presented to candidate analytical tools supporting SPARQL. We describe benchmark data sets for evaluating uncertainty quantification, as well as an approach for using our benchmark generator to produce data sets for generating benchmark data sets.

benchmark↗

Amino Acid Encoding for Deep Learning Applications

Background: The number of applications of deep learning algorithms in bioinformatics is increasing as they usually achieve superior performance over classical approaches, especially, when bigger training datasets are available. In deep learning applications, discrete data, e.g. words or n-grams in language, or amino acids or nucleotides in bioinformatics, are generally represented as a continuous vector through an embedding matrix. Recently, learning this embedding matrix directly from the data as part of the continuous iteration of the model to optimize the target prediction – a process called ‘end-to-end learning’ – has led to state-of-the-art results in many fields. Although usage of embeddings is well described in the bioinformatics literature, the potential of end-to-end learning for single amino acids, as compared to more classical manually-curated encoding strategies, has not been systematically addressed. To this end, we compared classical encoding matrices, namely one-hot, VHSE8 and BLOSUM62, to end-to-end learning of amino acid embeddings for two different prediction tasks using three widely used architectures, namely recurrent neural networks (RNN), convolutional neural networks (CNN), and the hybrid CNN-RNN. Results: By using different deep learning architectures, we show that end-to-end learning is on par with classical encodings for embeddings of the same dimension even when limited training data is available, and might allow for a reduction in the embedding dimension without performance loss, which is critical when deploying the models to devices with limited computational capacities. We found that the embedding dimension is a major factor in controlling the model performance. Surprisingly, we observed that deep learning models are capable of learning from random vectors of appropriate dimension. Conclusion: Our study shows that end-to-end learning is a flexible and powerful method for amino acid encoding. Further, due to the flexibility of deep learning systems, amino acid encoding schemes should be benchmarked against random vectors of the same dimension to disentangle the information content provided by the encoding scheme from the distinguishability effect provided by the scheme.

Deep-learning↗

Amino Acid Encoding for Deep Learning Applications

Background: The number of applications of deep learning algorithms in bioinformatics is increasing as they usually achieve superior performance over classical approaches, especially, when bigger training datasets are available. In deep learning applications, discrete data, e.g. words or n-grams in language, or amino acids or nucleotides in bioinformatics, are generally represented as a continuous vector through an embedding matrix. Recently, learning this embedding matrix directly from the data as part of the continuous iteration of the model to optimize the target prediction – a process called ‘end-to-end learning’ – has led to state-ofthe-art results in many fields. Although usage of embeddings is well described in the bioinformatics literature, the potential of end-to-end learning for single amino acids, as compared to more classical manually-curated encoding strategies, has not been systematically addressed. To this end, we compared classical encoding matrices, namely one-hot, VHSE8 and BLOSUM62, to end-to-end learning of amino acid embeddings for two different prediction tasks using three widely used architectures, namely recurrent neural networks (RNN), convolutional neural networks (CNN), and the hybrid CNN-RNN. Results: By using different deep learning architectures, we show that end-to-end learning is on par with classical encodings for embeddings of the same dimension even when limited training data is available, and might allow for a reduction in the embedding dimension without performance loss, which is critical when deploying the models to devices with limited computational capacities. We found that the embedding dimension is a major factor in controlling the model performance. Surprisingly, we observed that deep learning models are capable of learning from random vectors of appropriate dimension. Conclusion: Our study shows that end-to-end learning is a flexible and powerful method for amino acid encoding. Further, due to the flexibility of deep learning systems, amino acid encoding schemes should be benchmarked against random vectors of the same dimension to disentangle the information content provided by the encoding scheme from the distinguishability effect provided by the scheme.

Hesham ElAbd↗

Neutronics and Depletion Methods for Parametric Studies of Fluoride Salt Cooled High Temperature Reactors with Slab Fuel Geometry and Multi-Batch Fuel Management Schemes

The Advanced High-Temperature Reactor (AHTR) is a 3400 MWth fluoride-salt-cooled high-temperature reactor (FHR) that uses TRISO particle fuel compacted into slabs rather than spherical or cylindrical fuel compacts. Simplified methods are required for parametric design studies such that analyzing the entire feasible design space for an AHTR is tractable. These simplifications include fuel homogenization techniques to increase the speed of neutron transport calculations in depletion analysis and equilibrium depletion analysis methods to analyze systems with multi-batch fuel management schemes. This paper presents three elements of significant novelty. First, the Reactivity-Equivalent Physical Transformation (RPT) methodology usually applied in systems with coated-particle fuel in cylindrical and spherical geometries has been extended to slab geometries. Secondly, based on this newly developed RPT method for slab geometries, a methodology that uses Monte Carlo depletion approaches was further developed to search for the maximum discharge burnup in a multi-batch system by iteratively estimating the beginning of equilibrium cycle (BOEC) composition and sampling different discharge burnups. This Iterative Equilibrium Depletion Search (IEDS) method fully defines an equilibrium fuel cycle (keff, power, flux, and composition evolutions) but is computationally demanding, although feasible on single-processor workstations. Finally, an analytical method, the Non-Linear Reactivity Model, was developed by expanding the linear reactivity model to include an arbitrary number of higher order terms so that single-batch depletion results could be extrapolated to estimate the maximum discharge burnup and BOEC keff in systems with multi-batch fuel management schemes. Results from this method were benchmarked against equilibrium depletion analysis results using the IEDS.

fluoride-salt-cooled high-temperature reactor (FHR↗

Stereo Imaging Velocimetry of Mixing Driven by Buoyancy Induced Flow Fields

Mixing of two fluids generated by steady and particularly g-jitter acceleration is fundamental towards the understanding of transport phenomena in a microgravity environment. We propose to carry out flight and ground-based experiments to quantify flow fields due to g-jitter type of accelerations using Stereo Imaging Velocimetry (SIV), and measure the concentration field using laser fluorescence. The understanding of the effects of g-jitter on transport phenomena is of great practical interest to the microgravity community and impacts the design of experiments for the Space Shuttle as well as the International Space Station. The aim of our proposed research is to provide quantitative data to the community on the effects of g-jitter on flow fields due to mixing induced by buoyancy forces. The fundamental phenomenon of mixing occurs in a broad range of materials processing encompassing the growth of opto-electronic materials and semiconductors, (by directional freezing and physical vapor transport), to solution and protein crystal growth. In materials processing of these systems, crystal homogeneity, which is affected by the solutal field distribution, is one of the major issues. The understanding of fluid mixing driven by buoyancy forces, besides its importance as a topic in fundamental science, can contribute towards the understanding of how solutal fields behave under various body forces. The body forces of interest are steady acceleration and g-jitter acceleration as in a Space Shuttle environment or the International Space Station. Since control of the body force is important, the flight experiment will be carried out on a tunable microgravity vibration isolation mount, which will permit us to precisely input the desired forcing function to simulate a range of body forces. To that end, we propose to design a flight experiment that can only be carried out under microgravity conditions to fully exploit the effects of various body forces on fluid mixing. Recent flight experiments, by the P.I. through collaboration with the Canadian Space Agency (STS-85, August 1997), aimed at determining the stability of the interface between two miscible liquids inside an enclosure show that a long liquid column (5 cm) under microgravity isolation conditions can be stable, i.e. the interface remains sharp and vertical over a short time scale; thus transport occurs by molecular mass diffusion. On the other hand, when the two liquids were excited from a controlled vibration source (Microgravity Vibration Isolation Mount) two to four mode large amplitude quasi-stationary waves were observed. The data was limited to CCD recording of the dynamics of the interface between the two fluids. We propose to carry out flight experiments to quantify the dynamics of the flow field using Stereo Imaging Velocimetry and measure the concentration field using laser fluorescence. The results will serve as a basis to understand effects of g-jitter on transport phenomena, in this case mass diffusion. As the measurement of the kinematics of the flow field will shed light on the instability mechanism. The research will allow measurement of the flow field in microgravity environment to prove two hypotheses: (1) Maxwell's hypothesis: finite convection always exists in diffusing systems, and (2) Quasi-stationary waves inside a bounded enclosure in a microgravity environment is generated by Kelvin-Helmholtz instability; resonance of the interface which produces incipient mixing is due to Rayleigh-Taylor instability. The first hypothesis can be used as a benchmark experiment to illustrate diffusive mixing. The second hypothesis will lead to the understanding of g-jitter effects on buoyancy driven flow fields which occur in many situations involving materials processing, and other basic fluid physics phenomena. In addition, the second hypothesis will also provide insight in how Rayleigh-Taylor and Kelvin-Helmholtz instabilities propagate concentration fronts during mixing. Measurement of the flow field using SIV is important because it is the flow field which causes instability at the interface between the two fluids. Mixing driven by buoyancy induced flow fields will be addressed both experimentally and computationally. The experimental effort will address the kinematics of mixing: stretching, transport and chaos. Quantification of the mechanisms of mixing will consists of measuring the flow field using the SIV system at Glenn and capturing the dynamics of the interface, to measure mass transport, using a CCD camera. These experiments will be carried out within the framework of Earth's gravity and g-jitter microgravity acceleration as in a Space Shuttle environment or the International Space Station. The g-jitter will be induced and controlled using a tunable vibration isolation platform to isolate against vibration as well as input periodic and random vibration to the system. The parametric range of the microgravity experiment will be extended from the experiments on STS-85 to investigate higher mode quasi-stationary waves (8 to 12), as well as resonance regions which leads to chaos and turbulence. Ground-based experiments will focus on effects of vibration on stably stratified fluid layers in order to scale for possible scenarios in a microgravity environment. These vibrations will be subjected perpendicular to the concentration field on the ground since the parallel case can only be carried out in a microgravity environment. The concept of dynamical similarity will be applied to tune the experiments as closely as possible to a Space Shuttle environment or the International Space Station. The computational effort will take advantage of the Computational Laboratory at Glenn to corroborate the experimental findings with predictions of the dynamics of the flow field using the codes FLUENT (finite difference based) and FIDAP (finite element based). We will investigate two important cases, single-fluid model to address dilute systems with negligible jump in viscosity and the more general two-fluid model which accounts for finite jump in viscosity. Apart from its microgravity relevance, this experiment is well suited to study dynamics in nonlinear systems.

Duval, W. M. B.↗

Automatic Energy Schemes for High Performance Applications

Although high-performance computing traditionally focuses on the efficient execution of large-scale applications, both energy and power have become critical concerns when approaching exascale. Drastic increases in the power consumption of supercomputers affect significantly their operating costs and failure rates. In modern microprocessor architectures, equipped with dynamic voltage and frequency scaling (DVFS) and CPU clock modulation (throttling), the power consumption may be controlled in software. Additionally, network interconnect, such as Infiniband, may be exploited to maximize energy savings while the application performance loss and frequency switching overheads must be carefully balanced. This work first studies two important collective communication operations, all-to-all and allgather and proposes energy saving strategies on the per-call basis. Next, it targets point-to-point communications to group them into phases and apply frequency scaling to them to save energy by exploiting the architectural and communication stalls. Finally, it proposes an automatic runtime system which combines both collective and point-to-point communications into phases, and applies throttling to them apart from DVFS to maximize energy savings. The experimental results are presented for NAS parallel benchmark problems as well as for the realistic parallel electronic structure calculations performed by the widely used quantum chemistry package GAMESS. Close to the maximum energy savings were obtained with a substantially low performance loss on the given platform.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗

Improving IMRT-plan quality with MLC leaf position refinement post plan optimization

Purpose: In intensity-modulated radiation therapy (IMRT) planning, reducing the pencil-beam size may lead to a significant improvement in dose conformity, but also increase the time needed for the dose calculation and plan optimization. The authors develop and evaluate a postoptimization refinement (POpR) method, which makes fine adjustments to the multileaf collimator (MLC) leaf positions after plan optimization, enhancing the spatial precision and improving the plan quality without a significant impact on the computational burden. Methods: The authors' POpR method is implemented using a commercial treatment planning system based on direct aperture optimization. After an IMRT plan is optimized using pencil beams with regular pencil-beam step size, a greedy search is conducted by looping through all of the involved MLC leaves to see if moving the MLC leaf in or out by half of a pencil-beam step size will improve the objective function value. The half-sized pencil beams, which are used for updating dose distribution in the greedy search, are derived from the existing full-sized pencil beams without need for further pencil-beam dose calculations. A benchmark phantom case and a head-and-neck (HN) case are studied for testing the authors' POpR method. Results: Using a benchmark phantom and a HN case, the authors have verified that their POpR method can be an efficient technique in the IMRT planning process. Effectiveness of POpR is confirmed by noting significant improvements in objective function values. Dosimetric benefits of POpR are comparable to those of using a finer pencil-beam size from the optimization start, but with far less computation and time. Conclusions: The POpR is a feasible and practical method to significantly improve IMRT-plan quality without compromising the planning efficiency.

60 APPLIED LIFE SCIENCES↗

Quantum approximate optimization of the long-range Ising model with a trapped-ion quantum simulator

Quantum computers and simulators may offer significant advantages over their classical counterparts, providing insights into quantum many-body systems and possibly improving performance for solving exponentially hard problems, such as optimization and satisfiability. Here, we report the implementation of a low-depth Quantum Approximate Optimization Algorithm (QAOA) using an analog quantum simulator. We estimate the ground-state energy of the Transverse Field Ising Model with long-range interactions with tunable range, and we optimize the corresponding combinatorial classical problem by sampling the QAOA output with high-fidelity, single-shot, individual qubit measurements. We execute the algorithm with both an exhaustive search and closed-loop optimization of the variational parameters, approximating the ground-state energy with up to 40 trapped-ion qubits. We benchmark the experiment with bootstrapping heuristic methods scaling polynomially with the system size. We observe, in agreement with numerics, that the QAOA performance does not degrade significantly as we scale up the system size and that the runtime is approximately independent from the number of qubits. We finally give a comprehensive analysis of the errors occurring in our system, a crucial step in the path forward toward the application of the QAOA to more general problem instances.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Crystal Structure Prediction of Binary Alloys via Deep Potential

Predicting crystal structure has been a challenging problem in physics and materials science for a long time. A reliable energy calculation engine combined with an efficient global search algorithm, such as particle swarm optimization algorithm or genetic algorithm, is needed to conduct crystal structure prediction. In recent years, machine learning-based interatomic potential energy surface models have been proposed, potentially allowing us to perform crystal structure prediction for systems with the accuracy of density functional theory (DFT) and the speed of empirical force fields. In this paper, we employ a previously developed Deep Potential model to predict the intermetallic compound of the aluminum–magnesium system, and find six meta-stable phases with negative or nearly zero formation energy. In particular, Mg 12 Al 8 shows excellent ductility and Mg 5 Al 27 has a high Young's modulus. Based on our benchmark results, we propose a relatively robust structure screening criterion that selects potentially stable structures from the Deep Potential-based convex hull and performs DFT refinement. By using this criterion, the computational cost needed to construct the convex hull with ab initio accuracy can be dramatically reduced.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Bi-Level Integrated System Synthesis (BLISS)

BLISS is a method for optimization of engineering systems by decomposition. It separates the system level optimization, having a relatively small number of design variables, from the potentially numerous subsystem optimizations that may each have a large number of local design variables. The subsystem optimizations are autonomous and may be conducted concurrently. Subsystem and system optimizations alternate, linked by sensitivity data, producing a design improvement in each iteration. Starting from a best guess initial design, the method improves that design in iterative cycles, each cycle comprised of two steps. In step one, the system level variables are frozen and the improvement is achieved by separate, concurrent, and autonomous optimizations in the local variable subdomains. In step two, further improvement is sought in the space of the system level variables. Optimum sensitivity data link the second step to the first. The method prototype was implemented using MATLAB and iSIGHT programming software and tested on a simplified, conceptual level supersonic business jet design, and a detailed design of an electronic device. Satisfactory convergence and favorable agreement with the benchmark results were observed. Modularity of the method is intended to fit the human organization and map well on the computing technology of concurrent processing.

Sobieszczanski-Sobieski, Jaroslaw↗

Radiative Forcing by Well-Mixed Greenhouse Gases: Estimates from Climate Models in the Intergovernmental Panel on Climate Change (IPCC) Fourth Assessment Report (AR4)

The radiative effects from increased concentrations of well-mixed greenhouse gases (WMGHGs) represent the most significant and best understood anthropogenic forcing of the climate system. The most comprehensive tools for simulating past and future climates influenced by WMGHGs are fully coupled atmosphere-ocean general circulation models (AOGCMs). Because of the importance of WMGHGs as forcing agents it is essential that AOGCMs compute the radiative forcing by these gases as accurately as possible. We present the results of a radiative transfer model intercomparison between the forcings computed by the radiative parameterizations of AOGCMs and by benchmark line-by-line (LBL) codes. The comparison is focused on forcing by CO2, CH4, N2O, CFC-11, CFC-12, and the increased H2O expected in warmer climates. The models included in the intercomparison include several LBL codes and most of the global models submitted to the Intergovernmental Panel on Climate Change (IPCC) Fourth Assessment Report (AR4). In general, the LBL models are in excellent agreement with each other. However, in many cases, there are substantial discrepancies among the AOGCMs and between the AOGCMs and LBL codes. In some cases this is because the AOGCMs neglect particular absorbers, in particular the near-infrared effects of CH4 and N2O, while in others it is due to the methods for modeling the radiative processes. The biases in the AOGCM forcings are generally largest at the surface level. We quantify these differences and discuss the implications for interpreting variations in forcing and response across the multimodel ensemble of AOGCM simulations assembled for the IPCC AR4.

Collins, W. D.↗

Tuning collective communication for Partitioned Global Address Space programming models

Partitioned Global Address Space (PGAS) languages offer programmers the convenience of a shared memory programming style combined with locality control necessary to run on large-scale distributed memory systems. Even within a PGAS language programmers often need to perform global communication operations such as broadcasts or reductions, which are best performed as collective operations in which a group of threads work together to perform the operation. In this study we consider the problem of implementing collective communication within PGAS languages and explore some of the design trade-offs in both the interface and implementation. In particular, PGAS collectives have semantic issues that are different than in send–receive style message passing programs, and different implementation approaches that take advantage of the one-sided communication style in these languages. We present an implementation framework for PGAS collectives as part of the GASNet communication layer, which supports shared memory, distributed memory and hybrids. The framework supports a broad set of algorithms for each collective, over which the implementation may be automatically tuned. In conclusion, we demonstrate the benefit of optimized GASNet collectives using application benchmarks written in UPC, and demonstrate that the GASNet collectives can deliver scalable performance on a variety of state-of-the-art parallel machines including a Cray XT4, an IBM BlueGene/P, and a Sun Constellation system with InfiniBand interconnect.

97 MATHEMATICS AND COMPUTING↗

Development of a coupled experimental–computational approach for engineering optimization of spout-fluidized bed particle coating systems

The design of spout-fluidized bed (SFB) coating systems for nuclear particle fuels typically relies on trial-and-error processes, comprising iterative and time-consuming coating deposition experiments and post-deposition characterization. At an engineering scale, this approach to guided SFB system design is inefficient, highlighting the need for streamlined experimental methodologies which can correlate fluidization conditions to downstream coating outcomes. In this study, we combine time-resolved particle image velocimetry (PIV) with CFD–DEM simulations to benchmark hydrodynamic behavior in a 3D spout-fluidized bed. By exploiting easily accessible optical measurements of particle motion at the bed wall and within the spouting region, we obtain quantitative velocity fields that can be directly compared with model predictions of the occluded bed region, without resorting to complex imaging and characterization techniques such as X-ray or magnetic resonance tomography. Experimental benchmarking reveals strong agreement between CFD–DEM and PIV in the spout and annulus regions, while discrepancies near the wall highlight areas for future model development. Here, the proposed integrated experimental–numerical framework will enable a direct connection between measured variables and numerically predicted fluidization performance of dense, surrogate nuclear particle fuel feedstock such that experimental SFB component design can be rapidly evaluated, informing design decisions for nozzle geometry and operating conditions. Future work will extend this framework by correlating quantified fluidization metrics across nozzle geometries and operating conditions with the resulting coating morphology, microstructure, and uniformity. Establishing these correlations will enable predictive links between hydrodynamic performance and coating quality, providing a rational, scalable basis for optimizing SFB design prior to coating deposition.

CFD/DEM↗

Scattering matrix approach to the dissociative recombination of HCO{sup +} and N{sub 2}H{sup +}

We present a theoretical study of the indirect dissociative recombination of linear polyatomic ions at low collisional energies. The approach is based on the computation of the scattering matrix just above the ionization threshold and enables the explicit determination of all diabatic electronic couplings responsible for dissociative recombination. In addition, we use the multi-channel quantum-defect theory to demonstrate the precision of the scattering matrix by reproducing accurately ab initio Rydberg state energies of the neutral molecule. We consider the molecular ions N{sub 2}H{sup +} and HCO{sup +} as benchmark systems of astrophysical interest and improve former theoretical studies, which had repeatedly produced smaller cross sections than experimentally measured. Specifically, we demonstrate the crucial role of the previously overlooked stretching modes for linear polyatomic ions with large permanent dipole moment. The theoretical cross sections for both ions agree well with experimental data over a wide energy range. Finally, we consider the potential role of the HOC{sup +} isomer in the experimental cross sections of HCO{sup +} at energies below 10 meV.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Quantifying the structural uncertainty of the aerosol mixing state representation in a modal model

Abstract. Aerosol mixing state is an important emergent property that affects aerosol radiative forcing and aerosol–cloud interactions, but it has not been easy to constrain this property globally. This study aims to verify the global distribution of aerosol mixing state represented by modal models. To quantify the aerosol mixing state, we used the aerosol mixing state indices for submicron aerosol based on the mixing of optically absorbing and non-absorbing species (χo), the mixing of primary carbonaceous and non-primary carbonaceous species (χc), and the mixing of hygroscopic and non-hygroscopic species (χh). To achieve a spatiotemporal comparison, we calculated the mixing state indices using output from the Community Earth System Model with the four-mode version of the Modal Aerosol Module (MAM4) and compared the results with the mixing state indices from a benchmark machine-learned model trained on high-detail particle-resolved simulations from the particle-resolved stochastic aerosol model PartMC-MOSAIC. The two methods yielded very different spatial patterns of the mixing state indices. In some regions, the yearly averaged χ value computed by the MAM4 model differed by up to 70 percentage points from the benchmark values. These errors tended to be zonally structured, with the MAM4 model predicting a more internally mixed aerosol at low latitudes and a more externally mixed aerosol at high latitudes compared to the benchmark. Our study quantifies potential model bias in simulating mixing state in different regions and provides insights into potential improvements to model process representation for a more realistic simulation of aerosols towards better quantification of radiative forcing and aerosol–cloud interactions.

54 ENVIRONMENTAL SCIENCES↗

HOST combustion R and T overview

The overall objective of the Turbine Engine Hot Section Technology Combustion Project was to develop and verify improved and more accurate numerical analysis methods for increasing the ability to design with confidence combustion systems for advanced aircraft gas turbine engines. The objective was approached from two directions: computational and experimental. On the computational side, the approach was to first assess and evaluate existing combustor aerothermal analysis models. On the experimental side, three types of experiments are identified; first, fundamental experiments directed toward improved understanding of the flow physics and chemistry; second, experiments run to provide data for the empirical modeling of complex phenomena; and third, benchmark experiments for computer code validation.

Gaugler, Raymond E.↗

FAIR Surrogate Benchmarks Supporting AI and Simulation Research (Final Report)

Computational Science is being revolutionized by integrating AI and simulation and, in particular, by deep learning surrogate models that can replace all or part of traditional large‐scale HPC computations. Such surrogates can achieve remarkable performance improvements, as much as several orders of magnitude, and save both compute time and energy. The Surrogate Benchmark Initiative (SBI) project creates a community repository and FAIR (Findable, Accessible, Interoperable, and Reusable) data ecosystem for HPC application surrogate benchmarks. The SBI team comes from Argonne National Laboratory (ANL), Indiana University (IU), Rutgers University, the University of Tennessee, Knoxville (UTK), and the University of Virginia (UVA). SBI repositories include data, code, and all relevant collateral artifacts that the science and engineering community need to use and reuse these data sets and surrogates. SBI repositories generate active research from both the participants in SBI and the broad community of AI and domain scientists. This project develops surrogates that use several different neural nets to learn and quickly infer the results of simulations and data systems and captures them as surrogate benchmarks with a rich set of metadata covering: Data; Model; Metrics specification; Machine specification; and Science, Speed, and Power Results. We research FAIR metadata for these benchmarks. We develop application surrogate examples as benchmarks across many fields (ANL, UTK, IU, UVA). We also study non-Surrogate benchmarks that have many common features and similar issues as regards FAIRness. We work with MLCommons (UVA, UTK), which is a major machine learning benchmarking activity where we get metadata ontologies, software, and benchmarks, Benchmarks have datasets, models, and metadata and they need a technical framework developed by UTK and Rutgers and deployed by UVA. We study features of Surrogates including performance, training set size, and uncertainty quantification (Rutgers, UVA and IU).

97 MATHEMATICS AND COMPUTING↗

Second-Order Sensitivity Analysis of Uncollided Particle Contributions to Radiation Detector Responses

This work presents an application of Cacuci’s Second-Order Adjoint Sensitivity Analysis Methodology (2nd-ASAM) to the simplified Boltzmann equation that models the transport of uncollided particles through a medium to compute efficiently and exactly all of the first- and second-order derivatives (sensitivities) of a detector’s response with respect to the system’s isotopic number densities, microscopic cross sections, source emission rates, and detector response function. The off-the-shelf PARTISN multigroup discrete ordinates code is employed to solve the equations underlying the 2nd-ASAM. The accuracy of the results produced using PARTISN is verified by using the results of three test configurations: (1) a homogeneous sphere, for which the response is the exactly known total uncollided leakage, (2) a multiregion two-dimensional (r-z) cylinder, and (3) a two-region sphere for which the response is a reaction rate. For the homogeneous sphere, results for the total leakage as well as for the respective first- and second-order sensitivities are in excellent agreement with the exact benchmark values. For the nonanalytic problems, the results obtained by applying the 2nd-ASAM to compute sensitivities are in excellent agreement with central-difference estimates. The efficiency of the 2nd-ASAM is underscored by the fact that, for the cylinder, only 12 adjoint PARTISN computations were required by the 2nd-ASAM to compute all of the benchmark’s 18 first-order sensitivities and 224 second-order sensitivities, in contrast to the 877 PARTISN calculations needed to compute the respective sensitivities using central finite differences, and this number does not include the additional calculations that were required to find appropriate values of the perturbations to use for the central differences.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗