Search NASA⌕ Search

SEARCH · Search NASA

Results for “algorithmic differentiation”

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 217 records · Page 12
All sourcesNASA NTRSDOE OSTI
DOE OSTI2018

Neural Algorithms for Low Power Implementation of Partial Differential Equations

The rise of low-power neuromorphic hardware has the potential to change high-performance computing; however much of the focus on brain-inspired hardware has been on machine learning applications. A low-power solution for solving partial differential equations could radically change how we approach large-scale computing in the future. The random walk is a fundamental stochastic process that underlies many numerical tasks in scientific computing applications. We consider here two neural algorithms that can be used to efficiently implement random walks on spiking neuromorphic hardware. The first method tracks the positions of individual walkers independently by using a modular code inspired by grid cells in the brain. The second method tracks the densities of random walkers at each spatial location directly. We present the scaling complexity of each of these methods and illustrate their ability to model random walkers under different probabilistic conditions. Finally, we present implementations of these algorithms on neuromorphic hardware.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2022

Learning infinite-horizon average-reward restless multi-action bandits via index awareness

We consider the online restless bandits with average-reward and multiple actions, where the state of each arm evolves according to a Markov decision process (MDP), and the reward of pulling an arm depends on both the current state of the corresponding MDP and the action taken. Since finding the optimal control is typically intractable for restless bandits, existing learning algorithms are often computationally expensive or with a regret bound that is exponential in the number of arms and states. In this paper, we advocate \textit{index-aware reinforcement learning} (RL) solutions to design RL algorithms operating on a much smaller dimensional subspace by exploiting the inherent structure in restless bandits. Specifically, we first propose novel index policies to address dimensionality concerns, which are provably optimal. We then leverage the indices to develop two low-complexity index-aware RL algorithms, namely, (i) GM-R2MAB, which has access to a generative model; and (ii) UC-R2MAB, which learns the model using an upper confidence style online exploitation method. We prove that both algorithms achieve a sub-linear regret that is only polynomial in the number of arms and states. A key differentiator between our algorithms and existing ones stems from the fact that our RL algorithms contain a novel exploitation that leverages our proposed provably optimal index policies for decision-makings.

Xiong, Guojun↗
NASA NTRS2003

GOME Total Ozone and Calibration Error Derived Usign Version 8 TOMS Algorithm

The Global Ozone Monitoring Experiment (GOME) is a hyper-spectral satellite instrument measuring the ultraviolet backscatter at relatively high spectral resolution. GOME radiances have been slit averaged to emulate measurements of the Total Ozone Mapping Spectrometer (TOMS) made at discrete wavelengths and processed using the new TOMS Version 8 Ozone Algorithm. Compared to Differential Optical Absorption Spectroscopy (DOAS) techniques based on local structure in the Huggins Bands, the TOMS uses differential absorption between a pair of wavelengths including the local stiucture as well as the background continuum. This makes the TOMS Algorithm more sensitive to ozone, but it also makes the algorithm more sensitive to instrument calibration errors. While calibration adjustments are not needed for the fitting techniques like the DOAS employed in GOME algorithms, some adjustment is necessary when applying the TOMS Algorithm to GOME. Using spectral discrimination at near ultraviolet wavelength channels unabsorbed by ozone, the GOME wavelength dependent calibration drift is estimated and then checked using pair justification. In addition, the day one calibration offset is estimated based on the residuals of the Version 8 TOMS Algorithm. The estimated drift in the 2b detector of GOME is small through the first four years and then increases rapidly to +5% in normalized radiance at 331 nm relative to 385 nm by mid 2000. The lb detector appears to be quite well behaved throughout this time period.

Gleason, J.↗
NASA NTRS1990

A novel multistage estimation of signal parameters

A multistage estimation scheme is presented for estimating the parameters of a received carrier signal possibly phase-modulated by unknown data and experiencing very high Doppler, Doppler rate, etc. Such a situation arises, for example, in the case of the Global Positioning Systems (GPS). In the proposed scheme, the first-stage estimator operates as a coarse estimator of the frequency and its derivatives, resulting in higher rms estimation errors but with a relatively small probability of the frequency estimation error exceeding one-half of the sampling frequency (an event termed cycle slip). The second stage of the estimator operates on the error signal available from the first stage, refining the overall estimates, and in the process also reduces the number of cycle slips. The first-stage algorithm is a modified least-squares algorithm operating on the differential signal model and referred to as differential least squares (DLS). The second-stage algorithm is an extended Kalman filter, which yields the estimate of the phase as well as refining the frequency estimate. A major advantage of the is a reduction in the threshold for the received carrier power-to-noise power spectral density ratio (CNR) as compared with the threshold achievable by either of the algorithms alone.

Kumar, Rajendra↗
DOE OSTI2017

Bayesian calibration of terrestrial ecosystem models: A study of advanced Markov chain Monte Carlo methods

Calibration of terrestrial ecosystem models is important but challenging. Bayesian inference implemented by Markov chain Monte Carlo (MCMC) sampling provides a comprehensive framework to estimate model parameters and associated uncertainties using their posterior distributions. The effectiveness and efficiency of the method strongly depend on the MCMC algorithm used. In this study, a Differential Evolution Adaptive Metropolis (DREAM) algorithm was used to estimate posterior distributions of 21 parameters for the data assimilation linked ecosystem carbon (DALEC) model using 14 years of daily net ecosystem exchange data collected at the Harvard Forest Environmental Measurement Site eddy-flux tower. The DREAM is a multi-chain method and uses differential evolution technique for chain movement, allowing it to be efficiently applied to high-dimensional problems, and can reliably estimate heavy-tailed and multimodal distributions that are difficult for single-chain schemes using a Gaussian proposal distribution. The results were evaluated against the popular Adaptive Metropolis (AM) scheme. DREAM indicated that two parameters controlling autumn phenology have multiple modes in their posterior distributions while AM only identified one mode. The calibration of DREAM resulted in a better model fit and predictive performance compared to the AM. DREAM provides means for a good exploration of the posterior distributions of model parameters. Lastly, it reduces the risk of false convergence to a local optimum and potentially improves the predictive performance of the calibrated model.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2022

Multiobjective optimization of nuclear microreactor reactivity control system operation with swarm and evolutionary algorithms

To improve the marketability of novel microreactor designs, there is a need for automated and optimal control of these reactors. This paper presents a methodology for performing multiobjective optimization of control drum operation for a microreactor under normal and off-nominal conditions. Here, two different case studies are used where the control drum configuration is optimized for the reactor to be critical with some desired power distribution that would satisfy peaking limits. A surrogate model for power distribution is developed based on a feedforward neural network. The process for determining weights for scalarization of the multiobjective optimization problem is also detailed. Six optimization algorithms: evolutionary strategies, differential evolution, grey wolf optimization, Harris hawks optimization, moth flame optimization and particle swarm optimization, are all applied to these cases and the results analyzed. Although all these algorithms will demonstrate optima-seeking behavior, for real-time control it is necessary to identify the best algorithm to efficiently provide reasonable optima without operator interference. The moth flame optimization algorithm was found to perform particularly well on both cases. Overall, it was found that the algorithms capable of supplying the best optima were also the most consistent. Finally, the found optima were verified with the original model used to train surrogates.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗
NASA NTRS1991

Asymptotic integration algorithms for nonhomogeneous, nonlinear, first order, ordinary differential equations

New methods for integrating systems of stiff, nonlinear, first order, ordinary differential equations are developed by casting the differential equations into integral form. Nonlinear recursive relations are obtained that allow the solution to a system of equations at time t plus delta t to be obtained in terms of the solution at time t in explicit and implicit forms. Examples of accuracy obtained with the new technique are given by considering systems of nonlinear, first order equations which arise in the study of unified models of viscoplastic behaviors, the spread of the AIDS virus, and predator-prey populations. In general, the new implicit algorithm is unconditionally stable, and has a Jacobian of smaller dimension than that which is acquired by current implicit methods, such as the Euler backward difference algorithm; yet, it gives superior accuracy. The asymptotic explicit and implicit algorithms are suitable for solutions that are of the growing and decaying exponential kinds, respectively, whilst the implicit Euler-Maclaurin algorithm is superior when the solution oscillates, i.e., when there are regions in which both growing and decaying exponential solutions exist.

Walker, K. P.↗
DOE OSTI2017

Bayesian calibration of terrestrial ecosystem models: a study of advanced Markov chain Monte Carlo methods

Calibration of terrestrial ecosystem models is important but challenging. Bayesian inference implemented by Markov chain Monte Carlo (MCMC) sampling provides a comprehensive framework to estimate model parameters and associated uncertainties using their posterior distributions. The effectiveness and efficiency of the method strongly depend on the MCMC algorithm used. In this work, a differential evolution adaptive Metropolis (DREAM) algorithm is used to estimate posterior distributions of 21 parameters for the data assimilation linked ecosystem carbon (DALEC) model using 14 years of daily net ecosystem exchange data collected at the Harvard Forest Environmental Measurement Site eddy-flux tower. The calibration of DREAM results in a better model fit and predictive performance compared to the popular adaptive Metropolis (AM) scheme. Moreover, DREAM indicates that two parameters controlling autumn phenology have multiple modes in their posterior distributions while AM only identifies one mode. The application suggests that DREAM is very suitable to calibrate complex terrestrial ecosystem models, where the uncertain parameter size is usually large and existence of local optima is always a concern. In addition, this effort justifies the assumptions of the error model used in Bayesian calibration according to the residual analysis. Here, the result indicates that a heteroscedastic, correlated, Gaussian error model is appropriate for the problem, and the consequent constructed likelihood function can alleviate the underestimation of parameter uncertainty that is usually caused by using uncorrelated error models.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2019

Algorithms and algebraic solutions of decay chain differential equations for stable and unstable nuclide fractionation

Radioactive decay processes, such as alpha decay, produce decay chains where the mass numbers of nuclides decrease as larger nuclides expel energetic particles to form smaller nuclides. Under these conditions, the coefficient matrix that describes the differential rate expressions for radioactive decay can be made lower triangular. With this special structure, formulating an algebraic solution to the decay chains can be done by first formulating the eigenvectors that make up the coefficient matrix, which can then be solved using forward substitution for a lower triangular matrix. This work details the derivation of algebraic solutions for decay chains of any number of stable and unstable nuclides with any number of branching based on this eigenvector analysis. A prototype computational code was developed to validate and compare this methodology against a number of other methods for solving similar systems. A two-phase sorting algorithm yielding the lower triangular matrix structure was established to apply the developed algebraic solutions for decay chains involving beta-emitting radionuclides transformed into daughter nuclides without change in their mass number. In conclusion, the methodologies produced in this work provide an efficient way to estimate nuclide fractions from natural decay processes.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗
DOE OSTI2011

A compact digital time differential perturbed angular correlation-spectrometer using field programmable gate arrays and various timestamp algorithms

A user-friendly fully digital time differential perturbed angular correlation (TDPAC)-spectrometer with six detectors and fast digitizers using field programmable gate arrays (FPGA) is described and performance data are given. The new spectrometer has an online data analysis feature, a compact size, and a time resolution such as conventional analog spectrometers. Its calculation intensive part was implemented inside the digitizer. This gives the possibility to change parameters (energy windows, constant fraction trigger delay) and see their influence immediately in the {gamma}-{gamma} correlation diagrams. Tests were performed which showed that the time resolution using a {sup 60}Co source with energy window set at 1.17 MeV and 1.33 MeV is 265 ps with LaBr{sub 3}(Ce) scintillators and 254 ps with BaF{sub 2} scintillators. A true constant fraction algorithm turned out to be slightly better than the constant fraction of amplitude method. The spectrometer performance was tested with a TDPAC measurement using a {sup 44}Ti in rutile source and a positron lifetime measurement using {sup 22}Na. The maximum possible data rate of the spectrometer is 1.1 x 10{sup 6} {gamma} quanta per detector and second.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗
DOE OSTI2016

Exploring the Interplay of Resilience and Energy Consumption for a Task-Based Partial Differential Equations Preconditioner

We discuss algorithm-based resilience to silent data corruption (SDC) in a task- based domain-decomposition preconditioner for partial differential equations (PDEs). The algorithm exploits a reformulation of the PDE as a sampling problem, followed by a solution update through data manipulation that is resilient to SDC. The implementation is based on a server-client model where all state information is held by the servers, while clients are designed solely as computational units. Scalability tests run up to ~ 51 K cores show a parallel efficiency greater than 90%. We use a 2D elliptic PDE and a fault model based on random single bit-flip to demonstrate the resilience of the application to synthetically injected SDC. We discuss two fault scenarios: one based on the corruption of all data of a target task, and the other involving the corruption of a single data point. We show that for our application, given the test problem considered, a four-fold increase in the number of faults only yields a 2% change in the overhead to overcome their presence, from 7% to 9%. We then discuss potential savings in energy consumption via dynamics voltage/frequency scaling, and its interplay with fault-rates, and application overhead.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2022

Performance of New Near-Real-Time PERSIANN Product (PDIR-Now) for Atmospheric River Events over the Russian River Basin, California

Most heavy precipitation events and extreme flooding over the U.S. Pacific coast can be linked to prevalent atmospheric river (AR) conditions. Thus, reliable quantitative precipitation estimation with a rich spatiotemporal resolution is vital for water management and early warning systems of flooding and landslides over these regions. At the same time, high-quality near-real-time measurements of AR precipitation remain challenging due to the complex topographic features of land surface and meteorological conditions of the region: specifically, orographic features occlude radar measurements while infrared-based algorithms face challenges, differentiating between both cold brightband (BB) precipitation and the warmer nonbrightband (NBB) precipitation. It should be noted that the latter precipitation is characterized by greater orographic enhancement. In this study, we evaluate the performance of a recently developed near-real-time satellite precipitation algorithm: Precipitation Estimation from Remotely Sensed Information Using Artificial Neural Networks (PERSIANN) Dynamic Infrared–Rain Rate-Now (PDIR-Now). This model is primarily dependent on infrared information from geostationary satellites as input; consequently, PDIR-Now has the advantage of short data latency, 15–60-min delay between observation to precipitation product delivery. The performance of PDIR-Now is analyzed with a focus on AR-related events for cases dominated by NBB and BB precipitation over the Russian River basin. In our investigations, we utilize S-band (3-GHz) precipitation profilers with Joss/Parsivel disdrometer measurements at the Middletown and Santa Rosa stations to classify BB and NBB precipitation events. In general, our analysis shows that PDIR-Now is more skillful in retrieving precipitation rates over both BB and NBB events across the topologically complex study area as compared to PERSIANN-Cloud Classification System (CCS). Also, we discuss the performance of well-known operational near-real-time precipitation products from 2017 to 2019. Conventional categorical and volumetric categorical indices, as well as continuous statistical metrics, are used to show the differences between various high-resolution precipitation products such as Multi-Radar Multi-Sensor (MRMS).

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2022

Field Work Proposal ERKJ358: Black-box training for scientific machine learning models (Final Report)

The overarching goal of this project is to develop a scalable black-box training capability for scientific machine learning (SciML) problems that are non-trainable with existing automatic differentiation (AD)-based algorithms. AD assumes that a loss function can be decomposed into a sequence of elementary operations whose derivatives are known. This assumption is violated when the loss function includes a black-box physical model (e.g., a legacy simulator). The current strategy, converting a black-box simulator to an AD-enabled code via differential programming, is inflexible and time-, labor-consuming. Thus, black-box optimization is a main workhorse for training SciML models, e.g., in scientific reinforcement learning, hyper-parameter fine tuning, designing SciML models with adversarial robustness, etc.

97 MATHEMATICS AND COMPUTING↗
NASA NTRS1994

Aspects of Unstructured Grids and Finite-Volume Solvers for the Euler and Navier-Stokes Equations

One of the major achievements in engineering science has been the development of computer algorithms for solving nonlinear differential equations such as the Navier-Stokes equations. These algorithms are now used in the practical engineering design of devices such as cars and airplanes as well as theoretical studies of complex phenomena such as fluid turbulence. In past years, limited computer resources have motivated the development of efficient numerical methods in computational fluid dynamics (CFD) utilizing structured meshes. These meshes are comprised of systematic arrays of quadrilateral or hexahedral cells. The use of structured meshes greatly simplifies the implementation of CFD algorithms on conventional computers. Structured meshes also permit the use of highly efficient solution techniques such as alternating direction implicit (ADI) iteration schemes or multigrid. Following the dramatic improvement in computing speed in recent years, emphasis has shifted towards the design of algorithms capable of treating complex geometries. The automatic generation of structured grids about complex geometries is problematic. Unstructured grids offer one promising alternative technique for treating these general geometries. Unstructured meshes have irregular connectivity and usually contain triangles and/or quadrilaterals in two dimensions and tetrahedra and/or hexahedra in three dimensions. The generation and use of unstructured grids poses new challenges in computational fluid dynamics. This is true for both grid generation as well as for the design of algorithms for flow solution. The purpose of these notes is to present recent developments in the unstructured grid generation and flow solution technology.

Barth, T. J.↗
DOE OSTI2018

On the numerical algorithms of parametrization method for solving a two-point boundary-value problem for impulsive systems of loaded differential equations

A linear two-point boundary value problem for a system of loaded differential equations with impulse effect is investigated. Values in the previous impulse points are taken into consideration in the conditions of impulse effect. The considered problem is reduced to an equivalent multi-point boundary value problem for the system of ordinary differential equations with parameters. A numerical implementation of parametrization method is offered using the Runge–Kutta method of 4th-order accuracy for solving the Cauchy problems for ordinary differential equations. The constructed numerical algorithms are illustrated by examples.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2022

Physics constrained learning for data-driven inverse modeling from sparse observations

Deep neural networks (DNN) have been used to model nonlinear relations between physical quantities. Those DNNs are embedded in physical systems described by partial differential equations (PDE) and trained by minimizing a loss function that measures the discrepancy between predictions and observations in some chosen norm. This loss function often includes the PDE constraints as a penalty term when only sparse observations are available. As a result, the PDE is only satisfied approximately by the solution. However, the penalty term typically slows down the convergence of the optimizer for stiff problems. We present a new approach that trains the embedded DNNs while numerically satisfying the PDE constraints. We develop an algorithm that enables differentiating both explicit and implicit numerical solvers in reverse-mode automatic differentiation. This allows the gradients of the DNNs and the PDE solvers to be computed in a unified framework. We demonstrate that our approach enjoys faster convergence and better stability in relatively stiff problems compared to the penalty method. Furthermore, our approach allows for the potential to solve and accelerate a wide range of data-driven inverse modeling, where the physical constraints are described by PDEs and need to be satisfied accurately.

97 MATHEMATICS AND COMPUTING↗
NASA NTRS1987

The hypercluster: A parallel processing test-bed architecture for computational mechanics applications

The development of numerical methods and software tools for parallel processors can be aided through the use of a hardware test-bed. The test-bed architecture must be flexible enough to support investigations into architecture-algorithm interactions. One way to implement a test-bed is to use a commercial parallel processor. Unfortunately, most commercial parallel processors are fixed in their interconnection and/or processor architecture. In this paper, we describe a modified n cube architecture, called the hypercluster, which is a superset of many other processor and interconnection architectures. The hypercluster is intended to support research into parallel processing of computational fluid and structural mechanics problems which may require a number of different architectural configurations. An example of how a typical partial differential equation solution algorithm maps on to the hypercluster is given.

Blech, Richard A.↗
DOE OSTI2020

Drive-pressure optimization in ramp-wave compression experiments through differential evolution

Ramp-wave dynamic-compression experiments are used to examine quasi-isentropic loading paths in materials. The gradual and continuous increase in pressure created by ramp waves make these types of experiments ideal for studying nonequilibrium material behavior, such as solidification kinetics. In ramp-wave compression experiments, the input drive pressure to the experimental setup may be exerted through one of a number of different mechanisms (e.g., magnetic fields, gas-gun-driven impactors, or high-energy lasers) and is generally required for simulating such experiments. Yet, regardless of the specific mechanism, this drive pressure cannot be measured directly (measurements are generally taken at a location near the back of the experimental setup through a transparent window), leading to an inverse problem where one must determine the drive pressure at the front of the experimental setup (i.e., the input) that corresponds to the particle velocity (the output) measured near the back of the experimental setup. Furthermore, we solve this inverse problem using a heuristic optimization algorithm, known as differential evolution, coupled with a multiphysics, hydrodynamics code that simulates the compression of the experimental setup. By running many rounds of forward simulations of the experimental setup, our optimization process iteratively searches for a drive pressure that is optimized to closely reproduce the experimentally measured particle velocity near the back of the experimental setup. While our optimization methodology requires a significant number of hydrodynamics simulations to be conducted, many of these can be performed in parallel, which greatly reduces the time cost of our methodology. One novel aspect of our method for determining the drive pressure is that it does not require physical modeling of the drive mechanism and can thus be broadly applied to many types of ramp-compression experiments, regardless of the drive mechanism.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗