SEARCH · Engineering Papers

Results for “inverse problems”

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.

SourceResearch connectionFrom yearThrough year

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 1,261 records · Page 70

All sources NASA NTRS DOE OSTI

DOE OSTI2020

Quantifying the generalization error in deep learning in terms of data distribution and neural network smoothness

We report the accuracy of deep learning, i.e., deep neural networks, can be characterized by dividing the total error into three main types: approximation error, optimization error, and generalization error. Whereas there are some satisfactory answers to the problems of approximation and optimization, much less is known about the theory of generalization. Most existing theoretical works for generalization fail to explain the performance of neural networks in practice. To derive a meaningful bound, we study the generalization error of neural networks for classification problems in terms of data distribution and neural network smoothness. We introduce the cover complexity (CC) to measure the difficulty of learning a data set and the inverse of the modulus of continuity to quantify neural network smoothness. A quantitative bound for expected accuracy/error is derived by considering both the CC and neural network smoothness. Although most of the analysis is general and not specific to neural networks, we validate our theoretical assumptions and results numerically for neural networks by several data sets of images. The numerical results confirm that the expected error of trained networks scaled with the square root of the number of classes has a linear relationship with respect to the CC. We also observe a clear consistency between test loss and neural network smoothness during the training process. In addition, we demonstrate empirically that the neural network smoothness decreases when the network size increases whereas the smoothness is insensitive to training dataset size.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2024

Efficient Mixed-Precision Matrix Factorization of the Inverse Overlap Matrix in Electronic Structure Calculations with AI-Hardware and GPUs

In recent years, a new kind of accelerated hardware has gained popularity in the artificial intelligence (AI) community which enables extremely high-performance tensor contractions in reduced precision for deep neural network calculations. In this article, we exploit Nvidia Tensor cores, a prototypical example of such AI-hardware, to develop a mixed precision approach for computing a dense matrix factorization of the inverse overlap matrix in electronic structure theory, S –1 . This factorization of S –1 , written as ZZT = S –1 , is used to transform the general matrix eigenvalue problem into a standard matrix eigenvalue problem. Here we present a mixed precision iterative refinement algorithm where Z is given recursively using matrix–matrix multiplications and can be computed with high performance on Tensor cores. To understand the performance and accuracy of Tensor cores, comparisons are made to GPU-only implementations in single and double precision. Additionally, we propose a nonparametric stopping criteria which is robust in the face of lower precision floating point operations. The algorithm is particularly useful when we have a good initial guess to Z, for example, from previous time steps in quantum-mechanical molecular dynamics simulations or from a previous iteration in a geometry optimization.

36 MATERIALS SCIENCE↗

DOE OSTI2024

Quantum Complexity of the Kronecker Coefficients

Whether or not the Kronecker coefficients of the symmetric group count some set of combinatorial objects is a longstanding open question. In this work we show that a given Kronecker coefficient is proportional to the rank of a projector that can be measured efficiently using a quantum computer. In other words a Kronecker coefficient counts the dimension of the vector space spanned by the accepting witnesses of a QMA verifier, where QMA is the quantum analogue of NP . This implies that approximating the Kronecker coefficients to within a given relative error is not harder than a certain natural class of problems that captures the complexity of estimating thermal properties of quantum many-body systems. A second consequence is that deciding positivity of Kronecker coefficients is contained in QMA , complementing a recent NP -hardness result of Ikenmeyer, Mulmuley, and Walter. We obtain similar results for the related problem of approximating row sums of the character table of the symmetric group. Finally, we discuss an efficient quantum algorithm that approximates normalized Kronecker coefficients to inverse-polynomial additive error. Published by the American Physical Society 2024

Physics↗

NASA NTRS1979

Off-axis impact of unidirectional composites with cracks: Dynamic stress intensification

The dynamic response of unidirectional composites under off axis (angle loading) impact is analyzed by assuming that the composite contains an initial flaw in the matrix material. The analytical method utilizes Fourier transform for the space variable and Laplace transform for the time variable. The off axis impact is separated into two parts, one being symmetric and the other skew-symmetric with reference to the crack plane. Transient boundary conditions of normal and shear tractions are applied to a crack embedded in the matrix of the unidirectional composite. The two boundary conditions are solved independently and the results superimposed. Mathematically, these conditions reduce the problem to a system of dual integral equations which are solved in the Laplace transform plane for the transformation of the dynamic stress intensity factor. The time inversion is carried out numerically for various combinations of the material properties of the composite and the results are displayed graphically.

Sih, G. C.↗

NASA NTRS1984

A pseudospectral Legendre method for hyperbolic equations with an improved stability condition

A new pseudospectral method is introduced for solving hyperbolic partial differential equations. This method uses different grid points than previously used pseudospectral methods: in fact the grid are related to the zeroes of the Legendre polynomials. The main advantage of this method is that the allowable time step is proportional to the inverse of the number of grid points 1/N rather than to 1/n(2) (as in the case of other pseudospectral methods applied to mixed initial boundary value problems). A highly accurate time discretization suitable for these spectral methods is discussed.

Tal-Ezer, H.↗

NASA NTRS1986

A VLSI architecture for performing finite field arithmetic with reduced table look-up

A new table look-up method for finding the log and antilog of finite field elements has been developed by N. Glover. In his method, the log and antilog of a field element is found by the use of several smaller tables. The method is based on a use of the Chinese Remainder Theorem. The technique often results in a significant reduction in the memory requirements of the problem. A VLSI architecture is developed for a special case of this new algorithm to perform finite field arithmetic including multiplication, division, and the finding of an inverse element in the finite field.

Hsu, I. S.↗

NASA NTRS1986

A pseudospectral Legendre method for hyperbolic equations with an improved stability condition

A new pseudospectral method is introduced for solving hyperbolic partial differential equations. This method uses different grid points than previously used pseudospectral methods: in fact the grid points are related to the zeroes of the Legendre polynomials. The main advantage of this method is that the allowable time step is proportional to the inverse of the number of grid points 1/N rather than to 1/n(2) (as in the case of other pseudospectral methods applied to mixed initial boundary value problems). A highly accurate time discretization suitable for these spectral methods is discussed.

Tal-Ezer, Hillel↗

NASA NTRS1989

Analysis of modified SMI method for adaptive array weight control

An adaptive array is applied to the problem of receiving a desired signal in the presence of weak interference signals which need to be suppressed. A modification, suggested by Gupta, of the sample matrix inversion (SMI) algorithm controls the array weights. In the modified SMI algorithm, interference suppression is increased by subtracting a fraction F of the noise power from the diagonal elements of the estimated covariance matrix. Given the true covariance matrix and the desired signal direction, the modified algorithm is shown to maximize a well-defined, intuitive output power ratio criterion. Expressions are derived for the expected value and variance of the array weights and output powers as a function of the fraction F and the number of snapshots used in the covariance matrix estimate. These expressions are compared with computer simulation and good agreement is found. A trade-off is found to exist between the desired level of interference suppression and the number of snapshots required in order to achieve that level with some certainty. The removal of noise eigenvectors from the covariance matrix inverse is also discussed with respect to this application. Finally, the type and severity of errors which occur in the covariance matrix estimate are characterized through simulation.

Dilsavor, R. L.↗

NASA NTRS1989

Transition regime droplet growth and evaporation - An integrodifferential variational approach

A variational principle based on the integrodifferential form of the Boltzmann equation, which allows very general forms of the collision integral and of the boundary operator, has been used to compute the mass flux to a spherical particle. For simplicity the problem has been restricted to the one-speed, constant cross-section approximation, the black sphere problem of neutron transport theory. Using the Hilbert expansion of the solution far from the sphere as a trial function leads to a rational expression in the inverse Knudsen number for the extrapolation distance, which is simply related to the number flux to the sphere. A five-term expansion gives results that are no more than 6 percent in error when compared to an accurate numerical solution. The technique can be applied to other forms of the collision model and other boundary conditions.

Cipolla, J. W., Jr.↗

NASA NTRS1991

Polynomial Transformations For Discrete-Time Linear Systems

Transformations based on polynomial matrices of finite degree developed for use in computing functions for compensation, inversion, and approximation of discrete-time, multivariable, linear systems. Method derived from z-transform transfer-function form of matrices. Applicable to cascade-compensation problems in design of control systems.

Baram, Yoram↗

NASA NTRS1991

A unified viscous theory of lift and drag of 2-D thin airfoils and 3-D thin wings

A unified viscous theory of 2-D thin airfoils and 3-D thin wings is developed with numerical examples. The viscous theory of the load distribution is unique and tends to the classical inviscid result with Kutta condition in the high Reynolds number limit. A new theory of 2-D section induced drag is introduced with specific applications to three cases of interest: (1) constant angle of attack; (2) parabolic camber; and (3) a flapped airfoil. The first case is also extended to a profiled leading edge foil. The well-known drag due to absence of leading edge suction is derived from the viscous theory. It is independent of Reynolds number for zero thickness and varies inversely with the square root of the Reynolds number based on the leading edge radius for profiled sections. The role of turbulence in the section induced drag problem is discussed. A theory of minimum section induced drag is derived and applied. For low Reynolds number the minimum drag load tends to the constant angle of attack solution and for high Reynolds number to an approximation of the parabolic camber solution. The parabolic camber section induced drag is about 4 percent greater than the ideal minimum at high Reynolds number. Two new concepts, the viscous induced drag angle and the viscous induced separation potential are introduced. The separation potential is calculated for three 2-D cases and for a 3-D rectangular wing. The potential is calculated with input from a standard doublet lattice wing code without recourse to any boundary layer calculations. Separation is indicated in regions where it is observed experimentally. The classical induced drag is recovered in the 3-D high Reynolds number limit with an additional contribution that is Reynold number dependent. The 3-D viscous theory of minimum induced drag yields an equation for the optimal spanwise and chordwise load distribution. The design of optimal wing tip planforms and camber distributions is possible with the viscous 3-D wing theory.

Yates, John E.↗

NASA NTRS1995

Tests of two convection theories for red giant and red supergiant envelopes

Two theories of stellar envelope convection are considered here in the context of red giants and red supergiants of intermediate to high mass: Boehm-Vitense's standard mixing-length theory (MLT) and Canuto & Mazzitelli's new theory incorporating the full spectrum of turbulence (FST). Both theories assume incompressible convection. Two formulations of the convective mixing length are also evaluated: l proportional to the local pressure scale height (H(sub P)) and l proportional to the distance from the upper boundary of the convection zone (z). Applications to test both theories are made by calculating stellar evolutionary sequences into the red zone (z). Applications to test both theories are made by calculating stellar evolutionary sequences into the red phase of core helium burning. Since the theoretically predicted effective temperatures for cool stars are known to be sensitive to the assigned value of the mixing length, this quantity has been individually calibrated for each evolutionary sequence. The calibration is done in a composite Hertzsprung-Russell diagram for the red giant and red supergiant members of well-observed Galactic open clusters. The MLT model requires the constant of proportionality for the convective mixing length to vary by a small but statistically significant amount with stellar mass, whereas the FST model succeeds in all cases with the mixing lenghth simply set equal to z. The structure of the deep stellar interior, however, remains very nearly unaffected by the choices of convection theory and mixing lenghth. Inside the convective envelope itself, a density inversion always occurs, but is somewhat smaller for the convectively more efficient MLT model. On physical grounds the FST model is preferable, and seems to alleviate the problem of finding the proper mixing length.

Stothers, Richard B.↗

NASA NTRS2019

Computational Bayesian Methods Applied to Complex Problems in Bio and Astro Statistics

In this dissertation we apply computational Bayesian methods to three distinct problems. In the first chapter, we address the issue of unrealistic covariance matrices used to estimate collision probabilities. We model covariance matrices with a Bayesian Normal-Inverse-Wishart model, which we fit with Gibbs sampling. In the second chapter, we are interested in determining the sample sizes necessary to achieve a particular interval width and establish non-inferiority in the analysis of prevalences using two fallible tests. To this end, we use a third order asymptotic approximation. In the third chapter, we wish to synthesize evidence across multiple domains in measurements taken longitudinally across time, featuring a substantial amount of structurally missing data, and fit the model with Hamiltonian Monte Carlo in a simulation to analyze how estimates of a parameter of interest change across sample sizes.

Elrod, Chris↗

DOE OSTI2023

Modular chip-integrated photonic control of artificial atoms in diamond waveguides

A central goal in creating long-distance quantum networks and distributed quantum computing is the development of interconnected and individually controlled qubit nodes. Atom-like emitters in diamond have emerged as a leading system for optically networked quantum memories, motivating the development of visible-spectrum, multi-channel photonic integrated circuit (PIC) systems for scalable atom control. However, it has remained an open challenge to realize optical programmability with a qubit layer that can achieve high optical detection probability over many optical channels. Here, we address this problem by introducing a modular architecture of piezoelectrically actuated atom-control PICs (APICs) and artificial atoms embedded in diamond nanostructures designed for high-efficiency free-space collection. The high-speed four-channel APIC is based on a splitting tree mesh with triple-phase shifter Mach–Zehnder interferometers. This design simultaneously achieves optically broadband operation at visible wavelengths, high-fidelity switching (>40dB) at low voltages, submicrosecond modulation timescales (>30MHz), and minimal channel-to-channel crosstalk for repeatable optical pulse carving. Via a reconfigurable free-space interconnect, we use the APIC to address single silicon vacancy color centers in individual diamond waveguides with inverse tapered couplers, achieving efficient single photon detection probabilities (∼15%) and second-order autocorrelation measurements g (2) (0)<0.14 for all channels. The modularity of this distributed APIC–quantum memory system simplifies the quantum control problem, potentially enabling further scaling to thousands of channels.

47 OTHER INSTRUMENTATION↗

NASA NTRS1992

Human-like agents with posture planning ability

Human body models are geometric structures which may be ultimately controlled by kinematically manipulating their joints, but for animation, it is desirable to control them in terms of task-level goals. We address a fundamental problem in achieving task-level postural goals: controlling massively redundant degrees of freedom. We reduce the degrees of freedom by introducing significant control points and vectors, e.g., pelvis forward vector, palm up vector, and torso up vector, etc. This reduced set of parameters are used to enumerate primitive motions and motion dependencies among them, and thus to select from a small set of alternative postures (e.g., bend vs. squat to lower shoulder height). A plan for a given goal is found by incrementally constructing a goal/constraint set based on the given goal, motion dependencies, collision avoidance requirements, and discovered failures. Global postures satisfying a given goal/constraint set are determined with the help of incremental mental simulation which uses a robust inverse kinematics algorithm. The contributions of the present work are: (1) There is no need to specify beforehand the final goal configuration, which is unrealistic for the human body, and (2) the degrees of freedom problem becomes easier by representing body configurations in terms of 'lumped' control parameters, that is, control points and vectors.

Jung, Moon R.↗

NASA NTRS2017

Iterative Discrete Ordinates Solution of the Equation for the Surface-Reflected Radiance

This paper presents a new method of numerical solution of the integral equation for the radiance reflected from an anisotropic surface. The equation relates the radiance at the surface level with BRDF and solutions of the standard radiative transfer problems for a slab with no reflection on its surfaces. It is also shown that the kernel of the equation satisfies the condition of the existence of a unique solution and the convergence of the successive approximations to that solution. The developed method features two basic steps: discretization on a 2D quadrature, and solving the resulting system of algebraic equations with successive over-relaxation method based on the Gauss-Seidel iterative process. Presented numerical examples show good coincidence between the surface-reflected radiance obtained with DISORT and the proposed method. Analysis of contributions of the direct and diffuse (but not yet reflected) parts of the downward radiance to the total solution is performed. Together, they represent a very good initial guess for the iterative process. This fact ensures fast convergence. The numerical evidence is given that the fastest convergence occurs with the relaxation parameter of 1 (no relaxation). An integral equation for BRDF is derived as inversion of the original equation. The potential of this new equation for BRDF retrievals is analyzed. The approach is found not viable as the BRDF equation appears to be an ill-posed problem, and it requires knowledge the surface-reflected radiance on the entire domain of both Sun and viewing zenith angles.

Alexander Radkevich↗

NASA NTRS1987

Progress in the prediction of unsteady heat transfer on turbines blades

Progress toward developing a general method for predicting unsteady heat transfer on turbine blades subject to blade-passing frequencies and Reynolds numbers relevant to the Space Shuttle Main Engine (SSME) is discussed. The method employs an invisid/viscous interactive procedure which has been tested extensively for steady subsonic and transonic external airfoil problems. One such example is shown. The agreement with experimental data and with Navier-Stokes calculations yields confidence in the method. The technique is extended to account for wake generated unsteadiness. The flow reversals around the stagnation point caused by the nonuniform onset velocity are accounted for by using the Characteristic Box scheme developed by Cebeci and Stewartson. The coupling between the inviscid and viscous methods is achieved by using a special procedure, which, with a novel inverse finite-difference boundary-layer method, allows the calculations to be performed for a wide range of flow conditions, including separation. Preliminary results are presented for the stagnation region of turbine blades for both laminar and turbulent flows. A laminar model problem corresponding to a flow on a circular cylinder which experiences the periodic passing of wakes from turbine blades is presented to demonstrate the ability of the method to calculate flow reversals around the stagnation region.

Cebeci, T.↗

DOE OSTI2021

Efficient ptychographic phase retrieval via a matrix-free Levenberg-Marquardt algorithm

The phase retrieval problem, where one aims to recover a complex-valued image from far-field intensity measurements, is a classic problem encountered in a range of imaging applications. Modern phase retrieval approaches usually rely on gradient descent methods in a nonlinear minimization framework. Calculating closed-form gradients for use in these methods is tedious work, and formulating second order derivatives is even more laborious. Additionally, second order techniques often require the storage and inversion of large matrices of partial derivatives, with memory requirements that can be prohibitive for data-rich imaging modalities. We use a reverse-mode automatic differentiation (AD) framework to implement an efficient matrix-free version of the Levenberg-Marquardt (LM) algorithm, a longstanding method that finds popular use in nonlinear least-square minimization problems but which has seen little use in phase retrieval. Furthermore, we extend the basic LM algorithm so that it can be applied for more general constrained optimization problems (including phase retrieval problems) beyond just the least-square applications. Since we use AD, we only need to specify the physics-based forward model for a specific imaging application; the first and second-order derivative terms are calculated automatically through matrix-vector products, without explicitly forming the large Jacobian or Gauss-Newton matrices typically required for the LM method. We demonstrate that this algorithm can be used to solve both the unconstrained ptychographic object retrieval problem and the constrained “blind” ptychographic object and probe retrieval problems, under the popular Gaussian noise model as well as the Poisson noise model. We compare this algorithm to state-of-the-art first order ptychographic reconstruction methods to demonstrate empirically that this method outperforms best-in-class first-order methods: it provides excellent convergence guarantees with (in many cases) a superlinear rate of convergence, all with a computational cost comparable to, or lower than, the tested first-order algorithms.

36 MATERIALS SCIENCE↗