Search NASA⌕ Search

SEARCH · Search NASA

Results for “CONVERGENCE”

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 37 records · Page 2
All sourcesNASA NTRSDOE OSTI
DOE OSTI2024

Newton-Raphson AC Power Flow Convergence Based on Deep Learning Initialization and Homotopy Continuation

Power flow forms the basis of many power system studies. With the increased penetration of renewable energy, grid planners tend to perform multiple power flow simulations under various operating conditions and not just selected snapshots at peak or light load conditions. Getting a converged AC power flow (ACPF) case remains a significant challenge for grid planners especially in large power grid networks. This paper proposes a two-stage approach to improve Newton-Raphson ACPF convergence and was applied to a 6102 bus Electric Reliability Council of Texas (ERCOT) system. The first stage utilizes a deep learning-based initializer with data re-training. Here a deep neural network (DNN) initializer is developed to provide better initial voltage magnitude and angle guesses to aid in power flow convergence. This is because Newton-Raphson ACPF is quite sensitive to the initial conditions and bad initialization could lead to divergence. The DNN initializer includes a data re-training framework that improves the initializer's performance when faced with limited training data. The DNN initializer successfully solved 3,285 cases out of 3,899 non-converging dispatch and performed better than random forest and DC power flow initialization methods. ACPF cases not solved in this first stage are then passed through a hot-starting algorithm based on homotopy continuation with switched shunt control. The hot-starting algorithm successfully converged 416 cases out of the remaining 614 non-converging ACPF dispatch. In conclusion, the combined two-stage approach achieved a 94.9% success rate, by converging a total of 3,701 cases out of the initial 3,899 unsolved cases.

Deep learning↗
NASA NTRS1995

Reliability enhancement of Navier-Stokes codes through convergence acceleration

Methods for enhancing the reliability of Navier-Stokes computer codes through improving convergence characteristics are presented. The improving of these characteristics decreases the likelihood of code unreliability and user interventions in a design environment. The problem referred to as a 'stiffness' in the governing equations for propulsion-related flowfields is investigated, particularly in regard to common sources of equation stiffness that lead to convergence degradation of CFD algorithms. Von Neumann stability theory is employed as a tool to study the convergence difficulties involved. Based on the stability results, improved algorithms are devised to ensure efficient convergence in different situations. A number of test cases are considered to confirm a correlation between stability theory and numerical convergence. The examples of turbulent and reacting flow are presented, and a generalized form of the preconditioning matrix is derived to handle these problems, i.e., the problems involving additional differential equations for describing the transport of turbulent kinetic energy, dissipation rate and chemical species. Algorithms for unsteady computations are considered. The extension of the preconditioning techniques and algorithms derived for Navier-Stokes computations to three-dimensional flow problems is discussed. New methods to accelerate the convergence of iterative schemes for the numerical integration of systems of partial differential equtions are developed, with a special emphasis on the acceleration of convergence on highly clustered grids.

Merkle, Charles L.↗
DOE OSTI2025

Convergence Analysis of the Alternating Anderson–Picard Method for Nonlinear Fixed-Point Problems

Anderson acceleration (AA) has been widely used to solve nonlinear fixed-point problems due to its rapid convergence. This work focuses on a variant of AA in which multiple Picard iterations are performed between each AA step, referred to as the Alternating Anderson–Picard (AAP) method. Furthermore, despite introducing more “slow” Picard iterations, this method has been shown to be efficient and even more robust in both linear and nonlinear cases. However, there is a lack of theoretical analysis for AAP in the nonlinear case. In this paper, we address this gap by establishing the equivalence between AAP and a multisecant-GMRES method that uses GMRES to solve a multisecant linear system at each iteration. From this perspective, we show that AAP “converges” to the Newton-GMRES method. Specifically, as the residual approaches zero, the multisecant matrix, the approximate Jacobian inverse, the search direction, and the optimization gain of AAP converge to their counterparts in the Newton-GMRES method. These connections provide insights for analyzing the asymptotic convergence properties of AAP. Consequently, we show that AAP is locally 𝑞-linear convergent and provide an upper bound for the convergence factor of AAP. To validate the theoretical results, numerical examples are provided.

Anderson acceleration↗
DOE OSTI2026

Convergence Criteria for Multiphysics Simulations

The behavior of engineered systems is often influenced by multiple physical phenomena, such as mechanical deformation, heat transfer, and chemical species transport and reactions. There are often strong interactions between these phenomena, and there is increasing interest in applying coupled-physics models to improve understanding of physical behavior under complex environmental conditions. Multiple simulation frameworks that facilitate coupled-physics simulations are in widespread use, and these employ a variety of techniques to account for interactions between those physics. Many frameworks solve the physics models independently and transfer results between them. Alternatively, a single monolithic system of equations for every physics model can be formed and solved. Each of these approaches has its benefits and drawbacks, and the optimal approach varies depending on the nature of the problem. The open-source MOOSE framework was developed targeting solution of large-scale multiphysics problems. Although it provides options for all these coupling approaches, its standard approach for multiphysics solutions is to form and solve a single monolithic system of equations containing the unknowns for all physics models. MOOSE provides a streamlined approach for users to define the solution variables, the terms in the partial differential equations pertaining to each variable, and interactions between solution variables. One aspect of the monolithic solution approach that can be problematic, however, is defining appropriate convergence criteria for the nonlinear system. A standard approach is to determine convergence is to simply take a norm of the residual vector corresponding to the full vector of unknowns. However, if the residual vector contains variables for multiple physics models, the magnitudes of those variables can differ significantly, and the variables can converge at significantly different rates from each other. It is important to ensure that the variables for each of the physics are converged, and also ensure that the convergence criteria are not excessively stringent in cases when there is little change in the solution. This talk presents representative multiphysics problems to highlight these issues, and shows strategies for convergence criteria in MOOSE that are robust for multiphysics models under a variety of conditions.

97 - MATHEMATICS AND COMPUTING↗
NASA NTRS1979

Use of global atmospheric data sets to test quasi-geostrophic eddy momentum flux convergence

The quasi-geostrophic relation between the fluxes of momentum, potential vorticity, and potential temperature is tested with global sets of atmospheric wind and temperature data by computing the convergence of momentum flux as a residual of the potential temperature and potential vorticity flux and comparing it to the momentum flux convergence computed directly. It is shown that in the troposphere between 18 N and 74 N the observed momentum flux convergence differs from the quasi-geostrophic convergence by 25%-60%, with the larger errors only occurring where the convergence is small. These results indicate that momentum flux convergence obtained from quasi-geostrophic theory is adequate for qualitative studies of the general circulation, and is comparable in accuracy to values obtained in general circulation models. For simple climate models and qualitative process studies, it can thus provide a useful approach.

Heck, W. J.↗
NASA NTRS1979

Acceleration of linear and logarithmic convergence

Eleven different methods for accelerating convergence of sequences and series have been tested and compared on a wide range of test problems, including both linearly and logarithmically convergent series, monotone and alternating series. All but one of these methods are already in the literature, and they include both linear and nonlinear methods. The only methods found to accelerate convergence across the board were the u and v transforms of Levin and the theta algorithm of Brezinski. The paper gives detailed comparisons of all the tested methods on the basis of number of correct digits in the answer as a function of number of terms of the series used. A theorem of Germain-Bonne states that methods of a certain form which are exact on geometric series will accelerate linear convergence. The theorem applies to theta sub 2, and we have extended it to apply to Levin's transforms. No corresponding theorem is known for logarithmic convergence, but u, v, and theta are exact on certain large classes of logarithmic series, and all tested methods lacking this property failed to accelerate some logarithmically convergent series.

Smith, D. A.↗
NASA NTRS1985

Convergence of Newton's method for a single real equation

Newton's method for finding the zeroes of a single real function is investigated in some detail. Convergence is generally checked using the Contraction Mapping Theorem which yields sufficient but not necessary conditions for convergence of the general single point iteration method. The resulting convergence intervals are frequently considerably smaller than actual convergence zones. For a specific single point iteration method, such as Newton's method, better estimates of regions of convergence should be possible. A technique is described which, under certain conditions (frequently satisfied by well behaved functions) gives much larger zones where convergence is guaranteed.

Campbell, C. W.↗
NASA NTRS1991

Surface wind convergence as a short-term predictor of cloud-to-ground lightning at Kennedy Space Center: A four-year summary and evaluation

Since 1986, USAF forecasters at NASA-Kennedy have had available a surface wind convergence technique for use during periods of convective development. In Florida during the summer, most of the thunderstorm development is forced by boundary layer processes. The basic premise is that the life cycle of convection is reflected in the surface wind field beneath these storms. Therefore the monitoring of the local surface divergence and/or convergence fields can be used to determine timing, location, longevity, and the lightning hazards which accompany these thunderstorms. This study evaluates four years of monitoring thunderstorm development using surface wind convergence, particularly the average over the area. Cloud-to-ground (CG) lightning is related in time and space with surface convergence for 346 days during the summers of 1987 through 1990 over the expanded wind network at KSC. The relationships are subdivided according to low level wind flow and midlevel moisture patterns. Results show a one in three chance of CG lightning when a convergence event is identified. However, when there is no convergence, the chance of CG lightning is negligible.

Watson, Andrew I.↗
NASA NTRS1992

Convergence characteristics of the multiple input, multiple output LMS algorithm

The convergence characteristics of the multiple input, multiple output LMS algorithm, as applied to active noise and vibration control systems, are examined. The mean square error during the convergence process, as well as the final converged value, are examined analytically and in computer simulation. It is shown that the ratio of number of error sensors to number of control sources has a significant influence upon both the converging and converged value of the mean square error. Other active control system variables, such as the inherent time delays and structural/acoustic transfer functions, are also shown to have a significant influence upon the convergence process.

Snyder, Scott D.↗
NASA NTRS1992

Non-LTE radiative transfer with lambda-acceleration - Convergence properties using exact full and diagonal lambda-operators

We investigate the convergence properties of Lambda-acceleration methods for non-LTE radiative transfer problems in planar and spherical geometry. Matrix elements of the 'exact' A-operator are used to accelerate convergence to a solution in which both the radiative transfer and atomic rate equations are simultaneously satisfied. Convergence properties of two-level and multilevel atomic systems are investigated for methods using: (1) the complete Lambda-operator, and (2) the diagonal of the Lambda-operator. We find that the convergence properties for the method utilizing the complete Lambda-operator are significantly better than those of the diagonal Lambda-operator method, often reducing the number of iterations needed for convergence by a factor of between two and seven. However, the overall computational time required for large scale calculations - that is, those with many atomic levels and spatial zones - is typically a factor of a few larger for the complete Lambda-operator method, suggesting that the approach should be best applied to problems in which convergence is especially difficult.

Macfarlane, J. J.↗
NASA NTRS1990

On domains of convergence in optimization problems

Numerical optimization algorithms require the knowledge of an initial set of design variables. Starting from an initial design x(sup 0), improved solutions are obtained by updating the design iteratively in a way prescribed by the particular algorithm used. If the algorithm is successful, convergence is achieved to a local optimal solution. Let A denote the iterative procedure that characterizes a typical optimization algorithm, applied to the problem: Find x belonging to R(sup n) that maximizes f(x) subject to x belonging to Omega contained in R(sup n). We are interested in problems with several local maxima (x(sub j))(sup *), j=1, ..., m, in the feasible design space Omega. In general, convergence of the algorithm A to a specific solution (x(sub j))(sup *) is determined by the choice of initial design x(sup 0). The domain of convergence D(sub j) of A associated with a local maximum (x(sub j))(sup *) is a subset of initial designs x(sup 0) in Omega such that the sequence (x(sup k)), k=0,1,2,... defined by x(sup k+1) = A(x(sup k)), k=0,1,... converges to (x(sub j))(sup *). The set D(sub j) is also called the basin of attraction of (x(sub j))(sup *). Cayley first proposed the problem of finding the basin of attraction for Newton's method in 1897. It has been shown that the basin of attraction for Newton's method exhibits chaotic behavior in problems with polynomial objective. This implies that there may be regions in the feasible design space where arbitrarily close starting points will converge to different local optimal solutions. Furthermore, the boundaries of the domains of convergence may have a very complex, even fractal structure. In this paper we show that even simple structural optimization problems solved using standard gradient based (first order) algorithms exhibit similar features.

Diaz, Alejandro R.↗
NASA NTRS2000

Convergent properties of vestibular-related brain stem neurons in the gerbil

Three classes of vestibular-related neurons were found in and near the prepositus and medial vestibular nuclei of alert or decerebrate gerbils, those responding to: horizontal translational motion, horizontal head rotation, or both. Their distribution ratios were 1:2:2, respectively. Many cells responsive to translational motion exhibited spatiotemporal characteristics with both response gain and phase varying as a function of the stimulus vector angle. Rotationally sensitive neurons were distributed as Type I, II, or III responses (sensitive to ipsilateral, contralateral, or both directions, respectively) in the ratios of 4:6:1. Four tested factors shaped the response dynamics of the sampled neurons: canal-otolith convergence, oculomotor-related activity, rotational Type (I or II), and the phase of the maximum response. Type I nonconvergent cells displayed increasing gains with increasing rotational stimulus frequency (0.1-2.0 Hz, 60 degrees /s), whereas Type II neurons with convergent inputs had response gains that markedly decreased with increasing translational stimulus frequency (0.25-2.0 Hz, +/-0.1 g). Type I convergent and Type II nonconvergent neurons exhibited essentially flat gains across the stimulus frequency range. Oculomotor-related activity was noted in 30% of the cells across all functional types, appearing as burst/pause discharge patterns related to the fast phase of nystagmus during head rotation. Oculomotor-related activity was correlated with enhanced dynamic range compared with the same category that had no oculomotor-related response. Finally, responses that were in-phase with head velocity during rotation exhibited greater gains with stimulus frequency increments than neurons with out-of-phase responses. In contrast, for translational motion, neurons out of phase with head acceleration exhibited low-pass characteristics, whereas in-phase neurons did not. Data from decerebrate preparations revealed that although similar response types could be detected, the sampled cells generally had lower background discharge rates, on average one-third lower response gains, and convergent properties that differed from those found in the alert animals. On the basis of the dynamic response of identified cell types, we propose a pair of models in which inhibitory input from vestibular-related neurons converges on oculomotor neurons with excitatory inputs from the vestibular nuclei. Simple signal convergence and combinations of different types of vestibular labyrinth information can enrich the dynamic characteristics of the rotational and translational vestibuloocular responses.

Non-NASA Center↗
NASA NTRSDate not supplied

Deep Neural Network Based Convergence Classification for Computational Fluid Dynamics

A supervised deep learning approach is coupled with heuristic convergence criteria to construct a classification model for detecting the completion (convergence) of computational fluid dynamics (CFD) simulations. Heuristic convergence criteria alone are not always sufficient and more complex decisions are often left to a human analyst. The proposed approach leverages heuristic convergence criteria as well as two deep neural network (DNN) models, one binary and one multi-class, to improve the efficiency and consistency of convergence classification across a wide range of flight regimes. The DNN models presented are each trained on a subset of ascent aerodynamic CFD simulations for NASA’s Space Launch System and were produced using NASA’s unstructured Navier-Stokes solver FUN3D. Individual solutions are analyzed intermittently and are classified as sufficiently converged, further iterations required, or switch from steady Reynolds Averaged Navier-Stokes (RANS) to unsteady RANS CFD based on the iterative histories of four aerodynamic coefficients. The implemented classification model is shown to produce solutions that closely correlate to solutions produced by a human analyst. This work lays groundwork for expanding the capabilities of DNNs for automating and improving more of the CFD process.

SLS↗
DOE OSTI2025

Revisiting Source Convergence Diagnostics in the KENO Monte Carlo Neutron Transport Codes [Abstract]

Monte Carlo criticality transport codes, which rely on the power iteration procedure, are a fundamental tool for nuclear criticality safety practitioners in assessing the neutron multiplication factor (k eff ) for problems involving fissile material. In these calculations, ensuring the convergence of both the fission source distributions and the k eff estimate for accurate results is crucial. However, a converged k eff estimate does not necessarily mean the fission source distribution is also converged because the fission source and flux distribution may continue to evolve even after k eff convergence. Therefore, most Monte Carlo transport criticality codes now offer various diagnostic tests to assess fission source convergence in addition to the k eff convergence by analyzing the trends of these quantities over multiple generations.

AZURE↗
DOE OSTI2026

Using Filter Methods to Guide Convergence for ADMM, with Applications to Nonnegative Matrix Factorization Problems

Nonconvex, nonlinear optimization problems arise naturally in parameter fitting and machine learning. While augmented Lagrangian methods have demonstrated robust convergence for classes of these problems, their convergence for block updates has been relatively unexplored outside of the context of the alternating direction method of multipliers (ADMM). ADMM has seen extensive use in these applications, but may exhibit uncertain convergence behavior in many practical nonconvex settings, and struggles with general nonlinear constraints. In contrast, filter methods have proved effective in enforcing convergence for sequential quadratic programming methods and interior point methods with feasibility criteria. We develop an ADMM-filter method for highly nonlinear and nonconvex problems. Here, we show convergence under mild assumptions for several types of coordinate descent schemes, and demonstrate our algorithm on nonnegative matrix factorization and completion problems in imaging and chemical spectrum analysis.

Nonconvex optimization↗
DOE OSTI2024

Comparison of Full-Field and Integrated CFD Convergence Based on Richardson Extrapolation

This work investigated the usefulness of Richardson extrapolation--based discretization error estimates across all points in a solution field to produce a spatial convergence field for a computational fluid dynamics (CFD) simulation. The presented work used previously developed methods for Richardson extrapolation to compute the convergence orders of a CFD simulation at all points of the base (coarsest) mesh solution. Three test cases of increasing complexity were considered: Poiseuille flow, incompressible flow around a sharp corner, and transonic flow over an RAE 2822 airfoil. These test cases highlighted the potential of the proposed method to identify error sources and their relation to the model system-response-quantity convergence orders. However, these test cases also revealed the immaturity of the proposed method stemming from the unreliability of computing observed convergence orders at single points. Nonetheless, the test cases highlighted that the observed convergence orders allow for a more accurate diagnosis of constructive and destructive error transport than mesh pair error estimates. In the long run, the proposed method can be a tool for developing efficient and advanced error management strategies like adaptive mesh refinement.

Weinmeister, Justin↗
NASA NTRS1991

The convergence rate of approximate solutions for nonlinear scalar conservation laws

The convergence rate is discussed of approximate solutions for the nonlinear scalar conservation law. The linear convergence theory is extended into a weak regime. The extension is based on the usual two ingredients of stability and consistency. On the one hand, the counterexamples show that one must strengthen the linearized L(sup 2)-stability requirement. It is assumed that the approximate solutions are Lip(sup +)-stable in the sense that they satisfy a one-sided Lipschitz condition, in agreement with Oleinik's E-condition for the entropy solution. On the other hand, the lack of smoothness requires to weaken the consistency requirement, which is measured in the Lip'-(semi)norm. It is proved for Lip(sup +)-stable approximate solutions, that their Lip'convergence rate to the entropy solution is of the same order as their Lip'-consistency. The Lip'-convergence rate is then converted into stronger L(sup p) convergence rate estimates.

Nessyahu, Haim↗
NASA NTRS1992

On the convergence of the coupled-wave approach for lamellar diffraction gratings

Among the many existing rigorous methods for analyzing diffraction of electromagnetic waves by diffraction gratings, the coupled-wave approach stands out because of its versatility and simplicity. It can be applied to volume gratings and surface relief gratings, and its numerical implementation is much simpler than others. In addition, its predictions were experimentally validated in several cases. These facts explain the popularity of the coupled-wave approach among many optical engineers in the field of diffractive optics. However, a comprehensive analysis of the convergence of the model predictions has never been presented, although several authors have recently reported convergence difficulties with the model when it is used for metallic gratings in TM polarization. Herein, three points are made: (1) in the TM case, the coupled-wave approach converges much slower than the modal approach of Botten et al; (2) the slow convergence is caused by the use of Fourier expansions for the permittivity and the fields in the grating region; and (3) is manifested by the slow convergence of the eigenvalues and the associated modal fields. The reader is assumed to be familiar with the mathematical formulations of the coupled-wave approach and the modal approach.

Li, Lifeng↗