Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “high performance space computing”

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 1,045 records · Page 58

Integration of Methane Steam Reforming and Water Gas Shift Reaction in a Pd/Au/Pd-Based Catalytic Membrane Reactor for Process Intensification

Palladium-based catalytic membrane reactors (CMRs) effectively remove H2 to induce higher conversions in methane steam reforming (MSR) and water-gas-shift reactions (WGS). Within such a context, this work evaluates the technical performance of a novel CMR, which utilizes two catalysts in series, rather than one. In the process system under consideration, the first catalyst, confined within the shell side of the reactor, reforms methane with water yielding H2, CO and CO2. After reforming is completed, a second catalyst, positioned in series, reacts with CO and water through the WGS reaction yielding pure H2O, CO2 and H2. A tubular composite asymmetric Pd/Au/Pd membrane is situated throughout the reactor to continuously remove the produced H2 and induce higher methane and CO conversions while yielding ultrapure H2 and compressed CO2 ready for dehydration. Experimental results involving (i) a conventional packed bed reactor packed (PBR) for MSR, (ii) a PBR with five layers of two catalysts in series and (iii) a CMR with two layers of two catalysts in series are comparatively assessed and thoroughly characterized. Furthermore, a comprehensive 2D computational fluid dynamics (CFD) model was developed to explore further the features of the proposed configuration. The reaction was studied at different process intensification-relevant conditions, such as space velocities, temperatures, pressures and initial feed gas composition. Finally, it is demonstrated that the above CMR module, which was operated for 600 h, displays quite high H2 permeance and purity, high CH4 conversion levels and reduced CO yields.

water-gas-shift reaction↗

The Pointing Self-calibration Algorithm for Aperture Synthesis Radio Telescopes

This paper is concerned with algorithms for calibration of direction-dependent effects (DDE) in aperture synthesis radio telescopes (ASRT). After correction of direction-independent effects (DIE) using self-calibration, imaging performance can be limited by the imprecise knowledge of the forward gain of the elements in the array. In general, the forward gain pattern is directionally dependent and varies with time due to a number of reasons. Some factors, such as rotation of the primary beam with Parallactic Angle for Azimuth–Elevation mount antennas are known a priori. Some, such as antenna pointing errors and structural deformation/projection effects for aperture-array elements cannot be measured a priori. Thus, in addition to algorithms to correct for DD effects known a priori, algorithms to solve for DD gains are required for high dynamic range imaging. Here, we discuss a mathematical framework for antenna-based DDE calibration algorithms and show that this framework leads to computationally efficient optimal algorithms that scale well in a parallel computing environment. As an example of an antenna-based DD calibration algorithm, we demonstrate the Pointing SelfCal (PSC) algorithm to solve for the antenna pointing errors. Our analysis show that the sensitivity of modern ASRT is sufficient to solve for antenna pointing errors and other DD effects. We also discuss the use of the PSC algorithm in real-time calibration systems and extensions for antenna Shape SelfCal algorithm for real-time tracking and corrections for pointing offsets and changes in antenna shape.

79 ASTRONOMY AND ASTROPHYSICS↗

Setting up a STAR Tier 2 Site at Golias/Prague Farm

High Energy Nuclear Physics (HENP) collaborations experience show that the computing resources available at a single site are often neither sufficient nor satisfy the need of remote collaborators eager to carry their analysis in the fastest and most convenient way. From latencies in the network connectivity to the lack of interactivity, work at distant computing centers is often inefficient. Having fully functional software stack on local resources is a strong enabler of science opportunities for any local group who can afford the time investment. The situation becomes more complex as vast amount of data are often needed to perform meaningful analysis. Prague's heavy-ions group participating in STAR experiment at RHIC has been a strong advocate of local computing as the most efficient means of data processing and physics analyses. To create an environment where science can freely thrive, a Tier 2 computing center was set up at a Regional Computing Center for Particle Physics called 'Golias'. It is the biggest farm in the Czech Republic fully dedicated for particle physics experiments. We report on our experience in setting up a fully functional Tier 2 center leveraging the minimal locally available human and financial resources. We discuss the solutions chosen to address storage space and analysis issues and the impact on the farms overall functionality. This includes a locally built STAR analysis framework, integration with a local DPM system (a cost effective storage solution), the influence of the availability and quality of the network connection to Tier 0 via a dedicated CESNET/ESnet link and the development of light-weight yet fully automated data transfer tools allowing the movement of entire datasets from BNL (Tier 0) to Golias (Tier 2). We will summarize the impact of the gained computing performance on the efficiency of the local physics group at offline physics analysis and show the feasibility of such a solution that can used by other groups as well.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Distributed-Memory Sparse Deep Neural Network Inference Using Global Arrays

Partitioned Global Address Space (PGAS) models exhibit tremendous promise in developing efficient and productive distributed-memory parallel applications. They have been used extensively in scientific computations due to conveniently offering a ``shared-memory''-like model and convenient interfaces that separate communication with synchronization. Traditionally, PGAS communication models have been applied to dense/contiguously distributed data, but most modern applications depict varied levels of sparsity. Existing PGAS models require certain adaptations to support distributed sparse computations, since associated computations often require matrix arithmetic, in addition to data movement. The Global Arrays toolkit from Pacific Northwest National Laboratory (PNNL) is one of the earliest PGAS models to combine one-sided data communication and distributed matrix operations and is still used in the popular NWChem quantum chemistry suite. Recently, we have expanded the Global Arrays toolkit to support common sparse operations, like sparse matrix-dense matrix multiplies (SpMM), sparse matrix-sparse matrix multiplication (SpGEMM) and Sampled Dense-Dense Matrix Multiplication (SDDMM). As it turns out, these operations are the bedrock of sparse Deep Learning (DL); sparse deep neural networks and Graph Neural Networks (GNNs) have gained increasing attention recently in achieving speedups on training and inference with reduced memory footprints. Unlike scientific applications in High Performance Computing (HPC), modern (distributed-memory capable) DL toolkits often rely on non-standardized and closed-source vendor software optimizations, creating challenges in software-hardware co-design at scale. Our goal is to support a variety of distributed-memory sparse matrix operations and helper functions in the newly created Sparse Global Arrays (SGA), such that it is possible to build portable and productive Machine Learning scenarios for algorithm/software and hardware codesign purposes. Contemporary data-parallel schemes for training/inference are undergoing a major overhaul since model replication limits scalability and causes resource inefficiencies. As such, we have adopted tensor parallelism in decomposing the model and inputs, to mitigate memory issues. Current implementation is built on top of MPI and uses CPUs to maximize the portability across the platforms.

Distributed computing, machine learning↗

Measuring Impact: Evaluating Thermal Zoning Simplification on Energy Efficiency Measures Analysis

Building Energy Modeling (BEM) is a versatile tool for designing, retrofitting, ensuring code compliance, obtaining certifications, qualifying for incentives, and enabling real-time building control. However, capturing all the details of building geometry for thermal zoning can be time-consuming, costly, and sometimes computationally challenging. As a result, modelers have been applying zoning simplification based on factors such as space functions and internal loads, as well as relying on their experience and judgment while adhering to zoning rules outlined in industry standards. Despite the prevalence of this common practice, a notable gap exists in the literature regarding studies quantifying the influence of simplified thermal zoning on the evaluation of Energy Efficiency Measures (EEMs). Recognizing this gap, this paper seeks to contribute to the field by enhancing the understanding of how the simplification of thermal zoning influences the evaluation of EEMs against a baseline design. The study utilized a medium office prototype model with a detailed floor plan featuring over 20 zones per floor covering diverse functional spaces with varying internal loads and occupancy schedules. A standard thermal zoning strategy outlined in ASHRAE Standard 90.1 Appendix G was employed as the simplified zoning method. This strategy condenses the zoning into a core zone and four perimeter zones per floor. It was compared with the detailed zoning approach, which involves one zone per space. Common Energy EEMs, such as enhanced envelope, high-efficiency appliances and equipment, and HVAC controls, were individually implemented and evaluated. The results indicate that the performance comparison between the two zoning methods varies depending on the type of measures considered. Basic measures, such as adding wall insulation, demonstrate similar energy impacts, while advanced HVAC control measures, such as static pressure reset, exhibit a more substantial difference that cannot be overlooked.

Xie, Jiarong↗

Inclined porous medium convection at large Rayleigh number

High-Rayleigh-number ( $Ra$ ) convection in an inclined two-dimensional porous layer is investigated using direct numerical simulations (DNS) and stability and variational upper-bound analyses. When the inclination angle $\unicode[STIX]{x1D719}$ of the layer satisfies $0^{\circ }<\unicode[STIX]{x1D719}\lesssim 25^{\circ }$ , DNS confirm that the flow exhibits a three-region wall-normal asymptotic structure in accord with the strictly horizontal ( $\unicode[STIX]{x1D719}=0^{\circ }$ ) case, except that as $\unicode[STIX]{x1D719}$ is increased the time-mean spacing between neighbouring interior plumes also increases substantially. Both DNS and upper-bound analysis indicate that the heat transport enhancement factor (i.e. the Nusselt number) $Nu\sim CRa$ with a $\unicode[STIX]{x1D719}$ -dependent prefactor $C$ . When $\unicode[STIX]{x1D719}>\unicode[STIX]{x1D719}_{t}$ , however, where $30^{\circ }<\unicode[STIX]{x1D719}_{t}<32^{\circ }$ independently of $Ra$ , the columnar flow structure is completely broken down: the flow transitions to a large-scale travelling-wave convective roll state, and the heat transport is significantly reduced. To better understand the physics of inclined porous medium convection at large $Ra$ and modest inclination angles, a spatial Floquet analysis is performed, yielding predictions of the linear stability of numerically computed, fully nonlinear steady convective states. The results show that there exist two types of instability when $\unicode[STIX]{x1D719}\neq 0^{\circ }$ : a bulk-mode instability and a wall-mode instability, consistent with previous findings for $\unicode[STIX]{x1D719}=0^{\circ }$ (Wen et al. , J. Fluid Mech. , vol. 772, 2015, pp. 197–224). The background flow induced by the inclination of the layer intensifies the bulk-mode instability during its subsequent nonlinear evolution, thereby favouring increased spacing between the interior plumes relative to that observed in convection in a horizontal porous layer.

Mechanics↗

Geometrical optics without singularities: using the ray time as the coordinate space

Geometrical optics (GO) is widely used for reduced modelling of waves in plasmas, but it fails near reflection points, where it predicts a spurious singularity of the wave amplitude. We show how to avoid this singularity by adopting a different representation of the wave equation. Instead of the physical coordinate 𝑥 and the wavevector 𝑘, we use the ray time 𝜏 as the new canonical coordinate and the ray energy ℎ as the associated canonical momentum. To derive the envelope equation in the 𝜏-representation, we construct the Weyl symbol calculus on the (𝜏,ℎ) space and show that the corresponding Weyl symbols are related to their (𝑥,𝑘) counterparts by the Airy transform. This allows us to express the coefficients in the envelope equation through the known properties of the original dispersion operator. When necessary, solutions of this equation can be mapped to the 𝑥-space using a generalised metaplectic transform. However, the field per se might not even be needed in practice. Instead, knowing the corresponding Wigner function usually suffices for linear and quasilinear calculations. As a Weyl symbol itself, the Wigner function can be mapped analytically, using the aforementioned Airy transform. We show that the standard Airy patterns that form in regions where conventional GO fails are successfully reproduced within metaplectic GO (MGO) simply by remapping the field from the 𝜏-space to the 𝑥-space. An extension to mode-converting waves is also presented. This formulation, which we call generalised MGO, can be particularly useful, for example, for reduced modelling of the O–X conversion in inhomogeneous plasma near the critical density, an effect that is important for fusion applications and also occurs in the ionosphere. Overall, MGO can replace GO for any practical purposes, because it better handles cutoffs and is similar otherwise.

plasma waves↗

Curvilinear finite-volume schemes using high-order compact interpolation

During the last years, the need of high fidelity simulations on complex geometries for aeroacoustics predictions has grown. Most of high fidelity numerical schemes, in terms of low dissipative and low dispersive effects, lie on finite-difference (FD) approach. But for industrial applications, FD schemes are less robust compared to finite-volume (FV) ones. Thus the present study focuses on the development of a new compact FV scheme for two- and three-dimensional applications. The proposed schemes are formulated in the physical space and not in the computational space as it is the case in most of the known works. Therefore, they are more appropriate for general grids. They are based on compact interpolation to approximate interface-averaged field values using known cell-averaged values. For each interface, the interpolation coefficients are determined by matching Taylor series expansions around the interface center. Two types of schemes can be distinguished. The first one uses only the curvilinear abscissa along a mesh line to derive a sixth-order compact interpolation formulae while the second, more general, uses coordinates in a spatial three-dimensional frame well chosen. This latter is formally sixth-order accurate in a preferred direction almost orthogonal to the interface and at most fourth-order accurate in transversal directions. For non-linear problems, different approaches can be used to keep the high-order scheme. However, in the present paper, a MUSCL-like formulation was sufficient to address the presented test cases. All schemes have been modified to treat multiblock and periodic interfaces in such a way that high-order accuracy, stability, good spectral resolution, conservativeness and low computational costs are guaranteed. This is a first step to insure good scalability of the schemes although parallel performances issues are not addressed. As high frequency waves, badly resolved, could be amplified and then destabilize the scheme, compact filtering operators have been used. Numerous test cases as the linear convection of a Gaussian wave, the convection of a Lamb-Oseen vortex and the diffraction of an acoustic wave on a plane have been realized to validate the schemes. The most efficient schemes are shown to be at least fifth-order accurate on linear and non-linear convection problems. They are also less dissipative and less dispersive on non-uniform curvilinear grids than schemes using implicit interpolation with constant coefficients of the same order on uniform cartesian grids.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Higher-order finite-difference formulation of periodic Orbital-free Density Functional Theory

We present a real-space formulation and higher-order finite-difference implementation of periodic Orbital-free Density Functional Theory (OF-DFT). Specifically, utilizing a local reformulation of the electrostatic and kernel terms, we develop a generalized framework for performing OF-DFT simulations with different variants of the electronic kinetic energy. In particular, we propose a self-consistent field (SCF) type fixed-point method for calculations involving linear-response kinetic energy functionals. In this framework, evaluation of both the electronic ground-state and forces on the nuclei are amenable to computations that scale linearly with the number of atoms. We develop a parallel implementation of this formulation using the finite-difference discretization. We demonstrate that higher-order finite-differences can achieve relatively large convergence rates with respect to mesh-size in both the energies and forces. Additionally, we establish that the fixed-point iteration converges rapidly, and that it can be further accelerated using extrapolation techniques like Anderson's mixing. We validate the accuracy of the results by comparing the energies and forces with plane-wave methods for selected examples, including the vacancy formation energy in Aluminum. Overall, the suitability of the proposed formulation for scalable high performance computing makes it an attractive choice for large-scale OF-DFT calculations consisting of thousands of atoms.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Sparse maps—A systematic infrastructure for reduced-scaling electronic structure methods. II. Linear scaling domain based pair natural orbital coupled cluster theory

Domain based local pair natural orbital coupled cluster theory with single-, double-, and perturbative triple excitations (DLPNO-CCSD(T)) is a highly efficient local correlation method. It is known to be accurate and robust and can be used in a black box fashion in order to obtain coupled cluster quality total energies for large molecules with several hundred atoms. While previous implementations showed near linear scaling up to a few hundred atoms, several nonlinear scaling steps limited the applicability of the method for very large systems. In this work, these limitations are overcome and a linear scaling DLPNO-CCSD(T) method for closed shell systems is reported. The new implementation is based on the concept of sparse maps that was introduced in Part I of this series [P. Pinski, C. Riplinger, E. F. Valeev, and F. Neese, J. Chem. Phys. 143, 034108 (2015)]. Using the sparse map infrastructure, all essential computational steps (integral transformation and storage, initial guess, pair natural orbital construction, amplitude iterations, triples correction) are achieved in a linear scaling fashion. In addition, a number of additional algorithmic improvements are reported that lead to significant speedups of the method. The new, linear-scaling DLPNO-CCSD(T) implementation typically is 7 times faster than the previous implementation and consumes 4 times less disk space for large three-dimensional systems. For linear systems, the performance gains and memory savings are substantially larger. Calculations with more than 20 000 basis functions and 1000 atoms are reported in this work. In all cases, the time required for the coupled cluster step is comparable to or lower than for the preceding Hartree-Fock calculation, even if this is carried out with the efficient resolution-of-the-identity and chain-of-spheres approximations. The new implementation even reduces the error in absolute correlation energies by about a factor of two, compared to the already accurate previous implementation.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Revisiting lattice thermal transport in PbTe: The crucial role of quartic anharmonicity

Here, we perform a first-principles study of lattice thermal transport in PbTe by explicitly considering anharmonicity up to 4th order. To determine the temperature-dependent lattice constant of PbTe beyond quasiharmonic approximation, we introduce a simple yet effective scheme to account for anharmonic phonon renormalization at finite temperature. Moreover, we explicitly compute mode-resolved phonon lifetimes by including both three- and four-phonon scatterings. We find that (1) anharmonic phonon renormalization leads to strong vibrational frequency shifts which improve the agreement between simulated and experimental lattice constants; (2) these frequency shifts lead to a significant increase in lattice thermal conductivity (κ l ) because of reduced phonon scattering phase space; and (3) four-phonon scatterings are responsible for severe reduction in κ l on top of three-phonon scatterings, making κ l consistent with experiments. Our results suggest that the predicted κ l and its temperature dependence without considering thermal expansion, anharmonic phonon renormalization and four-phonon scatterings could accidentally agree with experiments due to error cancellation. Our study not only deepens the understanding of lattice thermal transport in PbTe but also exemplifies a widely applicable approach to investigate lattice dynamics and thermal transport properties from first-principles calculations including high-order anharmonicity.

36 MATERIALS SCIENCE↗

SU-E-T-795: Validations of Dose Calculation Accuracy of Acuros BV in High-Dose-Rate (HDR) Brachytherapy with a Shielded Cylinder Applicator Using Monte Carlo Simulation

Purpose: Acuros BV has become available to perform accurate dose calculations in high-dose-rate (HDR) brachytherapy with phantom heterogeneity considered by solving the Boltzmann transport equation. In this work, we performed validation studies regarding the dose calculation accuracy of Acuros BV in cases with a shielded cylinder applicator using Monte Carlo (MC) simulations. Methods: Fifteen cases were considered in our studies, covering five different diameters of the applicator and three different shielding degrees. For each case, a digital phantom was created in Varian BrachyVision with the cylinder applicator inserted in the middle of a large water phantom. A treatment plan with eight dwell positions was generated for these fifteen cases. Dose calculations were performed with Acuros BV. We then generated a voxelized phantom of the same geometry, and the materials were modeled according to the vendor’s specifications. MC dose calculations were then performed using our in-house developed fast MC dose engine for HDR brachytherapy (gBMC) on a GPU platform, which is able to simulate both photon transport and electron transport in a voxelized geometry. A phase-space file for the Ir-192 HDR source was used as a source model for MC simulations. Results: Satisfactory agreements between the dose distributions calculated by Acuros BV and those calculated by gBMC were observed in all cases. Quantitatively, we computed point-wise dose difference within the region that receives a dose higher than 10% of the reference dose, defined to be the dose at 5mm outward away from the applicator surface. The mean dose difference was ∼0.45%–0.51% and the 95-percentile maximum difference was ∼1.24%–1.47%. Conclusion: Acuros BV is able to accurately perform dose calculations in HDR brachytherapy with a shielded cylinder applicator.

60 APPLIED LIFE SCIENCES↗

Learning nonlinear level sets for dimensionality reduction in function approximation

We developed a Nonlinear Level-set Learning (NLL) method for dimensionality reduction in high-dimensional function approximation with small data. This work is motivated by a variety of design tasks in real-world engineering applications, where practitioners would replace their computationally intensive physical models (e.g., high-resolution fluid simulators) with fast-to-evaluate predictive machine learning models, so as to accelerate the engineering design processes. There are two major challenges in constructing such predictive models: (a) high-dimensional inputs (e.g., many independent design parameters) and (b) small training data, generated by running extremely time-consuming simulations. Thus, reducing the input dimension is critical to alleviate the over-fitting issue caused by data insufficiency. Existing methods, including sliced inverse regression and active subspace approaches, reduce the input dimension by learning a linear coordinate transformation; our main contribution is to extend the transformation approach to a nonlinear regime. Specifically, we exploit reversible networks (RevNets) to learn nonlinear level sets of a high-dimensional function and parameterize its level sets in low-dimensional spaces. A new loss function was designed to utilize samples of the target functions' gradient to encourage the transformed function to be sensitive to only a few transformed coordinates. The NLL approach is demonstrated by applying it to three 2D functions and two 20D functions for showing the improved approximation accuracy with the use of nonlinear transformation, as well as to an 8D composite material design problem for optimizing the buckling-resistance performance of composite shells of rocket inter-stages.

Zhang, Guannan↗

Reweighted ℓ{sub 1} minimization method for stochastic elliptic differential equations

We consider elliptic stochastic partial differential equations (SPDEs) with random coefficients and solve them by expanding the solution using generalized polynomial chaos (gPC). Under some mild conditions on the coefficients, the solution is “sparse” in the random space, i.e., only a small number of gPC basis makes considerable contribution to the solution. To exploit this sparsity, we employ reweighted l{sub 1} minimization to recover the coefficients of the gPC expansion. We also combine this method with random sampling points based on the Chebyshev probability measure to further increase the accuracy of the recovery of the gPC coefficients. We first present a one-dimensional test to demonstrate the main idea, and then we consider 14 and 40 dimensional elliptic SPDEs to demonstrate the significant improvement of this method over the standard l{sub 1} minimization method. For moderately high dimensional (∼10) problems, the combination of Chebyshev measure with reweighted l{sub 1} minimization performs well while for higher dimensional problems, reweighted l{sub 1} only is sufficient. The proposed approach is especially suitable for problems for which the deterministic solver is very expensive since it reuses the sampling results and exploits all the information available from limited sources.

97 MATHEMATICS AND COMPUTING↗

Determining Interactions in PSA models: Application to a Space PSA

This paper addresses use of an importance measure interaction study of a probabilistic risk analysis (PSA) performed for a hypothetical aerospace lunar mission. The PSA methods used in this study follow the general guidance provided in the NASA Probabilistic Risk Assessment Procedures Guide for NASA Managers and Practitioners. For the PSA portion, we used phased-based event tree and fault tree logic structures are used to model a lunar mission, including multiple phases (from launch to return to the Earth surface) and multiple critical systems. Details of the analysis results are not provided in this paper – instead specific basic events are denoted by number (e.g., the first event is 1, the second is 2, and so on). However, in the model, we used approximately 150 fault trees and over 800 basic events. Following analysis and truncation of cut sets, we were left with about 400 basic events to evaluate. We used this model to explore interactions between different basic events and systems. These sensitivity studies provide high-level insights into features of the PSA for the hypothetical lunar mission.

97 MATHEMATICS AND COMPUTING↗

Scientific and Computational Challenges of the Fusion Simulation Program (FSP)

This paper highlights the scientific and computational challenges facing the Fusion Simulation Program (FSP) a major national initiative in the United States with the primary objective being to enable scientific discovery of important new plasma phenomena with associated understanding that emerges only upon integration. This requires developing a predictive integrated simulation capability for magnetically-confined fusion plasmas that are properly validated against experiments in regimes relevant for producing practical fusion energy. It is expected to provide a suite of advanced modeling tools for reliably predicting fusion device behavior with comprehensive and targeted science-based simulations of nonlinearly-coupled phenomena in the core plasma, edge plasma, and wall region on time and space scales required for fusion energy production. As such, it will strive to embody the most current theoretical and experimental understanding of magnetic fusion plasmas and to provide a living framework for the simulation of such plasmas as the associated physics understanding continues to advance over the next several decades. Substantive progress on answering the outstanding scientific questions in the field will drive the FSP toward its ultimate goal of developing the ability to predict the behavior of plasma discharges in toroidal magnetic fusion devices with high physics fidelity on all relevant time and space scales. From a computational perspective, this will demand computing resources in the petascale range and beyond together with the associated multi-core algorithmic formulation needed to address burning plasma issues relevant to ITER - a multibillion dollar collaborative experiment involving seven international partners representing over half the world's population. Even more powerful exascale platforms will be needed to meet the future challenges of designing a demonstration fusion reactor (DEMO). Analogous to other major applied physics modeling projects (e.g., Climate Modeling), the FSP will need to develop software in close collaboration with computers scientists and applied mathematicians and validated against experimental data from tokamaks around the world. Specific examples of expected advances needed to enable such a comprehensive integrated modeling capability and possible "co-design" approaches will be discussed. __________________________________________________

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

Residual-based error correction for neural operator accelerated infinite-dimensional Bayesian inverse problems

We explore using neural operators, or neural network representations of nonlinear maps between function spaces, to accelerate infinite-dimensional Bayesian inverse problems (BIPs) with models governed by nonlinear parametric partial differential equations (PDEs). Neural operators have gained significant attention in recent years for their ability to approximate the parameter-to-solution maps defined by PDEs using as training data solutions of PDEs at a limited number of parameter samples. The computational cost of BIPs can be drastically reduced if the large number of PDE solves required for posterior characterization are replaced with evaluations of trained neural operators. However, reducing error in the resulting BIP solutions via reducing the approximation error of the neural operators in training can be challenging and unreliable. We provide an a priori error bound result that implies certain BIPs can be ill-conditioned to the approximation error of neural operators, thus leading to inaccessible accuracy requirements in training. To reliably deploy neural operators in BIPs, we consider a strategy for enhancing the performance of neural operators: correcting the prediction of a trained neural operator by solving a linear variational problem based on the PDE residual. We show that a trained neural operator with error correction can achieve a quadratic reduction of its approximation error, all while retaining substantial computational speedups of posterior sampling when models are governed by highly nonlinear PDEs. The strategy is applied to two numerical examples of BIPs based on a nonlinear reaction–diffusion problem and deformation of hyperelastic materials. We demonstrate that posterior representations of the two BIPs produced using trained neural operators are greatly and consistently enhanced by error correction.

97 MATHEMATICS AND COMPUTING↗

Accurate Binding Energies for Lithium Polysulfides and Assessment of Density Functionals for Lithium–Sulfur Battery Research

Lithium–sulfur batteries have high theoretical energy density, but a better knowledge of their intimate structural details will be helpful in improving their conductivity and long-term cycling behavior. In order to identify the stationary configurations of lithium polysulfides (Li 2 S n , 2 ≤ n ≤ 8) formed in the charging and discharging processes of the lithium–sulfur batteries, ab initio molecular dynamics was employed to sample the configuration space of Li 2 S n , followed by optimization of structures by CCSD(T)-F12b/aug-cc-pVDZ. Here, using the optimized stationary points, we have created the LiSAE38 benchmark database of atomization energies (AEs) of 38 lithium polysulfide isomers by using the higher-level WMS and W3X-L methods. In addition, the performances of 39 density functionals have been assessed against the benchmark AEs and the relative stabilities of Li 2 S n isomers. On the basis of the assessments with the def2-QZVP basis set, the PW6B95, B97-1, B3LYP-D3, TPSS, and DSD-PBEP86 density functionals are the most accurate for the AEs, whereas the mPW2-PLYP-D, DSD-PBEP86, PW6B95, HSE06, and PBEQIDH functionals are most accurate for the relative energies. Local functionals are of special interest because of their lower cost (faster timings in terms of computer resources) for large systems; among the tested local functionals, MN15-L and revM06-L are the most accurate for the calculations of relative stabilities.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗