Search NASA⌕ Search

SEARCH · Search NASA

Results for “adjoint method”

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 235 records · Page 13

SU-G-TeP1-15: Toward a Novel GPU Accelerated Deterministic Solution to the Linear Boltzmann Transport Equation

Purpose: To develop a Graphic Processor Unit (GPU) accelerated deterministic solution to the Linear Boltzmann Transport Equation (LBTE) for accurate dose calculations in radiotherapy (RT). A deterministic solution yields the potential for major speed improvements due to the sparse matrix-vector and vector-vector multiplications and would thus be of benefit to RT. Methods: In order to leverage the massively parallel architecture of GPUs, the first order LBTE was reformulated as a second order self-adjoint equation using the Least Squares Finite Element Method (LSFEM). This produces a symmetric positive-definite matrix which is efficiently solved using a parallelized conjugate gradient (CG) solver. The LSFEM formalism is applied in space, discrete ordinates is applied in angle, and the Multigroup method is applied in energy. The final linear system of equations produced is tightly coupled in space and angle. Our code written in CUDA-C was benchmarked on an Nvidia GeForce TITAN-X GPU against an Intel i7-6700K CPU. A spatial mesh of 30,950 tetrahedral elements was used with an S4 angular approximation. Results: To avoid repeating a full computationally intensive finite element matrix assembly at each Multigroup energy, a novel mapping algorithm was developed which minimized the operations required at each energy. Additionally, a parallelized memory mapping for the kronecker product between the sparse spatial and angular matrices, including Dirichlet boundary conditions, was created. Atomicity is preserved by graph-coloring overlapping nodes into separate kernel launches. The one-time mapping calculations for matrix assembly, kronecker product, and boundary condition application took 452±1ms on GPU. Matrix assembly for 16 energy groups took 556±3s on CPU, and 358±2ms on GPU using the mappings developed. The CG solver took 93±1s on CPU, and 468±2ms on GPU. Conclusion: Three computationally intensive subroutines in deterministically solving the LBTE have been formulated on GPU, resulting in two orders of magnitude speedup. Funding support from Natural Sciences and Engineering Research Council and Alberta Innovates Health Solutions. Dr. Fallone is a co-founder and CEO of MagnetTx Oncology Solutions (under discussions to license Alberta bi-planar linac MR for commercialization).

60 APPLIED LIFE SCIENCES↗

A non-intrusive parallel-in-time approach for simultaneous optimization with unsteady PDEs

This paper presents a non-intrusive framework for integrating existing unsteady partial differential equation (PDE) solvers into a parallel-in-time simultaneous optimization algorithm. The time-parallelization is provided by the non-intrusive software library XBraid [Parallel multigrid in time, software available at http://llnl.gov/casc/xbraid], which applies an iterative multigrid reduction technique to the time domain of existing time-marching schemes for solving unsteady PDEs. Its general user-interface has been extended in [S. Günther, N.R. Gauger, and J.B. Schroder, A non-intrusive parallel-in-time adjoint solver with the XBraid library, Comput. Vis. Sci. 19 (2018), pp. 85–95. Available at arXiv, math.OC/1705.00663] for computing adjoint sensitivities such that gradients of output quantities with respect to design changes can be computed parallel-in-time alongside with the primal PDE solution. In this paper, the primal and adjoint XBraid iterations are embedded into a simultaneous optimization framework, namely the One-shot method. In this method, design updates towards optimality are employed after each state and adjoint update such that optimality and feasibility of the design and the PDE solution are reached simultaneously. Finally, the time-parallel optimization method is validated on an advection-dominated flow control problem which shows significant speedup over a classical time-serial optimization algorithm.

97 MATHEMATICS AND COMPUTING↗

Deep learning closure models for large-eddy simulation of flows around bluff bodies

Near-wall flow simulation remains a central challenge in aerodynamics modelling: Reynolds-averaged Navier–Stokes predictions of separated flows are often inaccurate, and large-eddy simulation (LES) can require prohibitively small near-wall mesh sizes. A deep learning (DL) closure model for LES is developed by introducing untrained neural networks into the governing equations and training in situ for incompressible flows around rectangular prisms at moderate Reynolds numbers. The DL-LES models are trained using adjoint partial differential equation (PDE) optimization methods to match, as closely as possible, direct numerical simulation (DNS) data. They are then evaluated out-of-sample – for aspect ratios, Reynolds numbers and bluff-body geometries not included in the training data – and compared with standard LES models. The DL-LES models outperform these models and are able to achieve accurate LES predictions on a relatively coarse mesh (downsampled from the DNS mesh by factors of four or eight in each Cartesian direction). We study the accuracy of the DL-LES model for predicting the drag coefficient, near-wall and far-field mean flow, and resolved Reynolds stress. A crucial challenge is that the LES quantities of interest are the steady-state flow statistics; for example, a time-averaged velocity component $\langle {u}_i\rangle (x) = \lim _{t \rightarrow \infty } ({1}/{t}) \int _0^t u_i(s,x)\, {\rm d}s$ . Calculating the steady-state flow statistics therefore requires simulating the DL-LES equations over a large number of flow times through the domain. It is a non-trivial question whether an unsteady PDE model with a functional form defined by a deep neural network can remain stable and accurate on $t \in [0, \infty )$ , especially when trained over comparatively short time intervals. Our results demonstrate that the DL-LES models are accurate and stable over long time horizons, which enables the estimation of the steady-state mean velocity, fluctuations and drag coefficient of turbulent flows around bluff bodies relevant to aerodynamics applications.

Mechanics↗

Constructing entanglement witnesses for infinite-dimensional systems

It is shown that every entangled state in an infinite-dimensional composite system has a simple entanglement witness of the form {alpha}I+T with {alpha} a non-negative number and T a finite rank self-adjoint operator. We also provide two methods of constructing an entanglement witness and apply them to obtain some entangled states that cannot be detected by the positive partial transpose criterion and the realignment criterion.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Advanced Mesh-Enabled Monte Carlo Capability for Multi-Physics Reactor Analysis

This project will accumulate high-precision fluxes throughout reactor geometry on a non-orthogonal grid of cells to support multi-physics coupling, in order to more accurately calculate parameters such as reactivity coefficients and to generate multi-group cross sections. This work will be based upon recent developments to incorporate advanced geometry and mesh capability in a modular Monte Carlo toolkit with computational science technology that is in use in related reactor simulation software development. Coupling this capability with production-scale Monte Carlo radiation transport codes can provide advanced and extensible test-beds for these developments. Continuous energy Monte Carlo methods are generally considered to be the most accurate computational tool for simulating radiation transport in complex geometries, particularly neutron transport in reactors. Nevertheless, there are several limitations for their use in reactor analysis. Most significantly, there is a trade-off between the fidelity of results in phase space, statistical accuracy, and the amount of computer time required for simulation. Consequently, to achieve an acceptable level of statistical convergence in high-fidelity results required for modern coupled multi-physics analysis, the required computer time makes Monte Carlo methods prohibitive for design iterations and detailed whole-core analysis. More subtly, the statistical uncertainty is typically not uniform throughout the domain, and the simulation quality is limited by the regions with the largest statistical uncertainty. In addition, the formulation of neutron scattering laws in continuous energy Monte Carlo methods makes it difficult to calculate adjoint neutron fluxes required to properly determine important reactivity parameters. Finally, most Monte Carlo codes available for reactor analysis have relied on orthogonal hexahedral grids for tallies that do not conform to the geometric boundaries and are thus generally not well-suited to coupling with the unstructured meshes that are used in other physics simulations.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

Unstructured discontinuous Galerkin for seismic inversion.

This abstract explores the potential advantages of discontinuous Galerkin (DG) methods for the time-domain inversion of media parameters within the earth's interior. In particular, DG methods enable local polynomial refinement to better capture localized geological features within an area of interest while also allowing the use of unstructured meshes that can accurately capture discontinuous material interfaces. This abstract describes our initial findings when using DG methods combined with Runge-Kutta time integration and adjoint-based optimization algorithms for full-waveform inversion. Our initial results suggest that DG methods allow great flexibility in matching the media characteristics (faults, ocean bottom and salt structures) while also providing higher fidelity representations in target regions. Time-domain inversion using discontinuous Galerkin on unstructured meshes and with local polynomial refinement is shown to better capture localized geological features and accurately capture discontinuous-material interfaces. These approaches provide the ability to surgically refine representations in order to improve predicted models for specific geological features. Our future work will entail automated extensions to directly incorporate local refinement and adaptive unstructured meshes within the inversion process.

58 GEOSCIENCES↗

SU-D-210-03: Limited-View Multi-Source Quantitative Photoacoustic Tomography

Purpose: This work is to investigate a novel limited-view multi-source acquisition scheme for the direct and simultaneous reconstruction of optical coefficients in quantitative photoacoustic tomography (QPAT), which has potentially improved signal-to-noise ratio and reduced data acquisition time. Methods: Conventional QPAT is often considered in two steps: first to reconstruct the initial acoustic pressure from the full-view ultrasonic data after each optical illumination, and then to quantitatively reconstruct optical coefficients (e.g., absorption and scattering coefficients) from the initial acoustic pressure, using multi-source or multi-wavelength scheme.Based on a novel limited-view multi-source scheme here, We have to consider the direct reconstruction of optical coefficients from the ultrasonic data, since the initial acoustic pressure can no longer be reconstructed as an intermediate variable due to the incomplete acoustic data in the proposed limited-view scheme. In this work, based on a coupled photo-acoustic forward model combining diffusion approximation and wave equation, we develop a limited-memory Quasi-Newton method (LBFGS) for image reconstruction that utilizes the adjoint forward problem for fast computation of gradients. Furthermore, the tensor framelet sparsity is utilized to improve the image reconstruction which is solved by Alternative Direction Method of Multipliers (ADMM). Results: The simulation was performed on a modified Shepp-Logan phantom to validate the feasibility of the proposed limited-view scheme and its corresponding image reconstruction algorithms. Conclusion: A limited-view multi-source QPAT scheme is proposed, i.e., the partial-view acoustic data acquisition accompanying each optical illumination, and then the simultaneous rotations of both optical sources and ultrasonic detectors for next optical illumination. Moreover, LBFGS and ADMM algorithms are developed for the direct reconstruction of optical coefficients from the acoustic data. Jing Feng and Hao Gao were partially supported by the NSFC (#11405105), the 973 Program (#2015CB856000) and the Shanghai Pujiang Talent Program (#14PJ1404500)

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

A flexible linear diffusion acceleration to k-eigenvalue neutron transport with SN discontinuous finite element method

In this paper, we derive a flexible linear diffusion acceleration (LDA) for k-eigenvalue neutron transport discretized with discontinuous finite element method (DFEM) and discrete ordinates(SN). This LDA is based on our two pieces of previous works: the flexible non linear diffusion acceleration (NDA) for DFEM-SN and LDA for k-eigenvalue neutron transport using pre-conditioned Jacobian-free Newton-Krylov with self-adjoint angular flux (SAAF), continuous finite element method(CFEM), and SN. We point out the differences between LDA and NDA for DFEM-SN and the difference between DFEM-SN and SAAF-CFEM-SN for LDA. Numerical tests are presented to compare the convergence behaviour of NDA and LDA. (authors)

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

Singular value decomposition of adjoint flux distributions for Monte Carlo variance reduction

Monte Carlo (MC) shielding calculations often use weight windows (WWs) and biased sources formed from a deterministic estimate of the adjoint flux to improve the convergence rate of tallies. This requires a significant amount of computer memory, which can limit the memory available for high-resolution tally output. A new method is proposed for reducing these memory requirements by using singular value decomposition (SVD) in linear or logarithmic space to approximate the adjoint flux. This method’s performance is evaluated using the Shift and Denovo codes for streaming and diffusion base case problems, followed by problems using the Westinghouse AP1000 and the Joint European Torus. The log SVD reduced WW memory requirements by an order of magnitude in all cases without a significant performance penalty. Additionally, the linear SVD reduced biased source memory requirements by an order of magnitude, but further investigation is needed to account for observed limitations.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

Evolution of the coupling constant in SU(2) lattice gauge theory with two adjoint fermions

We measure the evolution of the coupling constant using the Schroedinger functional method in the lattice formulation of SU(2) gauge theory with two massless Dirac fermions in the adjoint representation. We observe strong evidence for an infrared fixed point, where the theory becomes conformal. We measure the {beta} function and the coupling constant as a function of the energy sca0008.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

Parallelizable adjoint stencil computations using transposed forward-mode algorithmic differentiation

Algorithmic differentiation (AD) is a tool for generating discrete adjoint solvers, which efficiently compute gradients of functions with many inputs, for example for use in gradient-based optimization. AD is often applied to large computations such as stencil operators, which are an important part of most structured-mesh PDE solvers. Stencil computations are often parallelized, for example by using OpenMP, and optimized by using techniques such as cache-blocking and tiling to fully utilize multicore CPUs and many-core accelerators and GPUs. Differentiating these codes with conventional reverse-mode AD results in adjoint codes that cannot be expressed as stencil operations and may not be easily parallelizable. They thus leave most of the compute power of modern architectures unused. We present a method that combines forward-mode AD and loop transformation to generate adjoint solvers that use the same memory access pattern as the original computation that they are derived from and can benefit from the same optimization techniques. The effectiveness of this method is demonstrated by generating a scalable adjoint CFD solver for multicore CPUs and Xeon Phi accelerators.

algorithmic differentiation↗

Generalized Perturbation Theory-Free Sensitivity Analysis for Eigenvalue Problems

Generalized perturbation theory (GPT) has been recognized as the most computationally efficient approach for performing sensitivity analysis for models with many input parameters, which renders forward sensitivity analysis computationally overwhelming. In critical systems, GPT involves the solution of the adjoint form of the eigenvalue problem with a response-dependent fixed source. Although conceptually simple to implement, most neutronics codes that can solve the adjoint eigenvalue problem do not have a GPT capability unless envisioned during code development. We introduce in this manuscript a reduced-order modeling approach based on subspace methods that requires the solution of the fundamental adjoint equations but allows the generation of response sensitivities without the need to set up GPT equations, and that provides an estimate of the error resulting from the reduction. Moreover, the new approach solves the eigenvalue problem independently of the number or type of responses. This allows for an efficient computation of sensitivities when many responses are required. This paper introduces the theory and implementation details of the GPT-free approach and describes how the errors could be estimated as part of the analysis. The applicability is demonstrated by estimating the variations in the flux distribution everywhere in the phase space of a fast critical sphere and a high-temperature gas-cooled reactor prismatic lattice. The variations generated by the GPT-free approach are benchmarked to the exact variations generated by direct forward perturbations.

99 GENERAL AND MISCELLANEOUS↗

Hybrid PN-SN Calculations with SAAF for the Multiscale Transport Capability in Rattlesnake

Two interface conditions, the Lagrange multiplier method and the upwinding method, for hybrid \pn-\sn calculations is proposed for the self-adjoint angular flux (SAAF) formulation of the transport equation using the continuous finite element method (FEM) for spatial discretization. These interface conditions are implemented in Rattlesnake, the radiation transport application built on MOOSE, for the on-going multiscale transport simulation effort at INL. For smoothing the solution at the interface for the Lagrange multiplier method, a method based on \sn Lagrange interpolation on the sphere is proposed. Numerical results indicate that the interface conditions give the expected convergence.

97 MATHEMATICS AND COMPUTING↗

A Feynman-Kac based numerical method for the exit time probability of a class of transport problems

The exit time probability, which gives the likelihood that an initial condition leaves a prescribed region of the phase space of a dynamical system at, or before, a given time, is arguably one of the most natural and important transport problems. In this work, we present an accurate and efficient numerical method for computing this probability for systems described by non-autonomous (time-dependent) stochastic differential equations (SDEs) or their equivalent Fokker-Planck partial differential equations. The method is based on the direct approximation of the Feynman-Kac formula that establishes a link between the adjoint Fokker-Planck equation and the forward SDE. The Feynman-Kac formula is approximated using the Gauss-Hermite quadrature rules and piecewise cubic Hermite interpolating polynomials, and a GPU accelerated matrix representation is used to compute the entire time evolution of the exit time probability using a single pass of the algorithm. The method is unconditionally stable, exhibits second order convergence in space, first order convergence in time, and it is straightforward to parallelize. Applications are presented to the advection diffusion of a passive tracer in a fluid flow exhibiting chaotic advection, and to the runaway acceleration of electrons in a plasma in the presence of an electric field, collisions, and radiation damping. Benchmarks against analytical solutions as well as comparisons with explicit and implicit finite difference standard methods for the adjoint Fokker-Planck equation are presented.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Stabilized FE simulation of prototype thermal-hydraulics problems with integrated adjoint-based capabilities

A critical aspect of applying modern computational solution methods to complex multiphysics systems of relevance to nuclear reactor modeling, is the assessment of the predictive capability of specific proposed mathematical models. In this respect the understanding of numerical error, the sensitivity of the solution to parameters associated with input data, boundary condition uncertainty, and mathematical models is critical. Additionally, the ability to evaluate and or approximate the model efficiently, to allow development of a reasonable level of statistical diagnostics of the mathematical model and the physical system, is of central importance. In this study we report on initial efforts to apply integrated adjoint-based computational analysis and automatic differentiation tools to begin to address these issues. The study is carried out in the context of a Reynolds averaged Navier–Stokes approximation to turbulent fluid flow and heat transfer using a particular spatial discretization based on implicit fully-coupled stabilized FE methods. Initial results are presented that show the promise of these computational techniques in the context of nuclear reactor relevant prototype thermal-hydraulics problems.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Automated Hybrid Variance Reduction on Advanced Architectures in the Shift Monte Carlo Code

Monte Carlo transport methods are the most accurate schemes for solving problems with complex energy and spatial features, but they come with a high computational cost. Although hybrid methods have enabled the use of Monte Carlo transport for a large class of problems, they still require significant computing resources. Modern multicore CPUs with large numbers of compute cores and graphical processing units (GPUs) provide opportunities to optimize the memory and run-time costs of hybrid Monte Carlo methods. This paper documents the development and analysis of three Monte Carlo transport algorithms that support hybrid transport using the consistent adjoint-driven importance sampling (CADIS) and forward-weighted CADIS methods in the Shift Monte Carlo code: history-based transport using static and dynamic threading on multicore CPUs and event-based transport enabling weight window tracking on GPUs. The results are shown for two challenging hybrid problems on the Frontier supercomputer at the Oak Ridge Leadership Computing Facility. The results show that all three methods yield good performance and enable solutions of difficult fixed-source transport problems in less than 2 min on 20 nodes of Frontier. Dynamic threading was observed to give up to 20% better scaling behavior than static threading. Moreover, the AMD Instinct 250X GPU was found to give 9 to 11 times greater throughput per graphics compute die than the best CPU performance. In conclusion, additional opportunities for optimization of hybrid transport on GPUs are discussed.

Denovo↗

Multi-fidelity thermal modeling of laser powder bed additive manufacturing

Laser powder bed fusion (LPBF) Additive manufacturing (AM) has attracted interest as an agile method of building production metal parts to reduce design-build-test cycle times for systems. However, predicting part performance is difficult due to inherent process variabilities. This makes qualification challenging. Computational process models have attempted to address some of these challenges, including mesoscale, full physics models and reduced fidelity conduction models. The goal of this work is credible multi-fidelity modeling of the LPBF process by investigating methods for estimating the error between models of two different fidelities. Two methods of error estimation are investigated, adjoint-based error estimation and Bayesian calibration. Adjoint-based error estimation is found to effectively bounding the error between the two models, but with very conservative bounds, making predictions highly uncertain. Bayesian parameter calibration applied to conduction model heat source parameters is found to effectively bound the observed error between the models for melt pool morphology quantities of interest. However, the calibrations do not effectively bound the error in heat distribution.

36 MATERIALS SCIENCE↗