Behavior Of Numerical Methods Research Articles

On Some Works of Boris Teodorovich Polyak on the Convergence of Gradient Methods and Their Development

The paper presents a review of the current state of subgradient and accelerated convex optimization methods, including the cases with the presence of noise and access to various information about the objective function (function value, gradient, stochastic gradient, higher derivatives). For nonconvex problems, the Polyak–Lojasiewicz condition is considered and a review of the main results is given. The behavior of numerical methods in the presence of a sharp minimum is considered. The aim of this review is to show the influence of the works of B.T. Polyak (1935–2023) on gradient optimization methods and their surroundings on the modern development of numerical optimization methods.

Discretization limits of lattice‐Boltzmann methods for studying immiscible two‐phase flow in porous media

SummaryDigital images of porous media often include features approaching the image resolution length scale. The behavior of numerical methods at low resolution is therefore important even for well‐resolved systems. We study the behavior of the Shan‐Chen (SC) and Rothman‐Keller (RK) multicomponent lattice‐Boltzmann models in situations where the fluid‐fluid interfacial radius of curvature and/or the feature size of the medium approaches the discrete unit size of the computational grid. Various simple, small‐scale test geometries are considered, and a drainage test is also performed in a Bentheimer sandstone sample. We find that both RK and SC models show very high ultimate limits: in ideal conditions the models can simulate static fluid configuration with acceptable accuracy in tubes as small as three lattice units across for RK model (six lattice units for SC model) and with an interfacial radius of curvature of two lattice units for RK and SC models. However, the stability of the models is affected when operating in these extreme discrete limits: in certain circumstances the models exhibit behaviors ranging from loss of accuracy to numerical instability. We discuss the circumstances where these behaviors occur and the ramifications for larger‐scale fluid displacement simulations in porous media, along with strategies to mitigate the most severe effects. Overall we find that the RK model, with modern enhancements, exhibits fewer instabilities and is more suitable for systems of low fluid‐fluid miscibility. The shortcomings of the SC model seem to arise predominantly from the high, strongly pressure‐dependent miscibility of the two fluid components.

Features of behavior of numerical methods for solving Volterra integral equations of the second kind

Systems of second-kind Volterra integral equations with stiff and oscillating components are considered. An implicit noniterative method of the second order is proposed for the numerical solution of such problems. The efficiency of the method is demonstrated using several examples.

On the spectrum of the electric field integral equation and the convergence of the moment method

Existing convergence estimates for numerical scattering methods based on boundary integral equations are asymptotic in the limit of vanishing discretization length, and break down as the electrical size of the problem grows. In order to analyse the efficiency and accuracy of numerical methods for the large scattering problems of interest in computational electromagnetics, we study the spectrum of the electric field integral equation (EFIE) for an infinite, conducting strip for both the TM (weakly singular kernel) and TE polarizations (hypersingular kernel). Due to the self-coupling of surface wave modes, the condition number of the discretized integral equation increases as the square root of the electrical size of the strip for both polarizations. From the spectrum of the EFIE, the solution error introduced by discretization of the integral equation can also be estimated. Away from the edge singularities of the solution, the error is second order in the discretization length for low-order bases with exact integration of matrix elements, and is first order if an approximate quadrature rule is employed. Comparison with numerical results demonstrates the validity of these condition number and solution error estimates. The spectral theory offers insights into the behaviour of numerical methods commonly observed in computational electromagnetics. Copyright © 2001 John Wiley & Sons, Ltd.

Accuracy of the method of moments for scattering by a cylinder

We study the accuracy and convergence of the method of moments for numerical scattering computations for an important benchmark geometry: the finite circular cylinder. From the spectral decomposition of the electric-field integral equation for this scatterer, we determine the condition number of the moment matrix and the dependence of solution error on the choice of basis functions, discretization density, polarization of the incident field, and the numerical quadrature rule used to evaluate moment-matrix elements. The analysis is carried out for both the TM polarization (weakly singular kernel) and TE polarization (hypersingular kernel). These results provide insights into empirical observations of the convergence behavior of numerical methods in computational electromagnetics.

Weak Nonlinear Stability of Implicit Runge-Kutta Methods

Two slightly different test problems have been used to examine nonlinear stability behaviour of numerical methods for solving systems of ordinary differential equations. These test problems, for BN-stability and for monotonicity, both lead to the concept of algebraic stability. In this paper, it is shown that these two problems may be combined to yield weaker stability conditions. Nonautonomous systems give a more restrictive stability condition than autonomous systems. Irreducible methods with some zero quadrature weights can be stable in this weak sense