Numerical continuum mechanics (Q441295)

The book under review is based on lecture courses given by the author over a period of 15 years at the Moscow Institute of Physics and Technology and the Ziolkovsky Russian State Technology University. It is arranged for engineering physics students rather than for computational mathematicians. It is assumed that the reader is familiar with basic concepts of continuum mechanics and computational mathematics, but the material is presented to such an extent that no additional literature has to be used. The aim of the book is the presentation of the main ideas, the applications, and limitations of the methods rather than a rigorous statement of grounds and proofs of theorems. The vii + 429 pages long monograph consists of a preface, three parts with a total of 8 chapters, a bibliography, and an index. The chapters are divided into sections and subsections. Exercises to improve the skill of the reader are formulated at the end of all chapters of Parts 2 and 3, i.e., for the numerical methods. Solutions of the problems are not given. The self-consistent Part 1 is devoted to basic equations of continuum mechanics (thermodynamics, elastic, plastic, elastoplastic, elastoviscoplastic, and composite media). They are presented in differential and integral forms including variational principles in order to prepare the discretization of the equations in the subsequent sections. The formulations are in comparison with traditional continuum mechanics monographs more adapted to the aim of their numerical solution including nonlinear problems and special practical problems, such as damages and large elastoplastic deformations. The theory of finite difference equations is treated in Part 2 which are used in applications with a focus on the algorithms, their stability, and accuracy of the results, mostly omitting proofs. The three chapters of this part contain the basic theory of finite difference schemes, algorithms for the solution of systems of linear and nonlinear algebraic equations including minimization methods, and methods for solving boundary value problems for systems of ordinary and partial differential equations. Especially, the solutions of stiff, singularly perturbed, and nonlinear problems are included. Other methods, such as finite elements, finite volume, spectral and collocation methods, are not a subject of this monograph. The methods of Part 2 are finally used in the last part for the development of special numerical algorithms for problems of continuum mechanics, mostly described in form of the equations given in Part 1. Section 1 of Part 1 starts with the description of the motion of continuous medium in the simplest reference frame, the rectangular Cartesian coordinate system, considering the motion relative to this fixed frame. The Eulerian and Lagrangian description of motion including the relationship between them are treated. The deformation of a continuous medium is described by the change of the square distance between two infinitely near points in Lagrangian variables introducing the Green-Lagrange finite strain tensor. Alternatively, the Euler-Almansi finite strain tensor is deduced. Additionally, the rates of deformation of a continuous medium are derived using the velocity gradients. The laws of conservation of mass, energy, linear and angular momentum are formulated in Section 2. They are presented in integral and differential form. The different forms are introduced because they are equivalent in continuum but not in a discretization form. Contingent on the application, one or the other discretization form is to be selected. The first and second law of thermodynamics are discussed in Section 3 in order to prepare the derivation of constitutive equations for a specific medium, such as viscous compressible heat-conducting gases, thermoplastic isotropic, composite, rigid-plastic, and elastoplastic media, in Section 4. For these purposes, the quantities of the conservation laws are linked to kinematic and thermal properties of the media, determining deformations and temperature. In order to also describe complex rheological media, the dynamic and kinematic variables are completed by additional so-called internal variables. Introducing an additional thermodynamic principle, the principle of maximum rate of dissipation, after all the general form of constitutive equations, composed of the stress equation, entropy equation, structural parameter equation, and heat equation, is formulated. Based on the theory of plastic flow, equations of elastoplastic and elastoviscoplastic media are derived in Section 5. The subject of Section 6 is the phenomenological description and the explanatory statement of elastoplastic deformations under dynamic loading on the basis of the dislocation theory. Alternatively to the differential forms, used in the preceding sections, energy forms of the equations of motion are considered in Section 7. The two integral forms are related to the principle of virtual displacements and the principle of virtual velocities and are equivalent to a differential equation of the mass conservation law treated in Section 1. The energy representation is associated with the weak formulation of the original continuum mechanics problem in terms of kinematic variables which often has advantages for variational methods. For certain classes, the integral forms related to the principles of virtual displacements and velocities can be used to develop variational principles that allow to reduce the problem to the minimization of a functional. Section 8 is devoted to Lagrange's, Hamilton's, Castigliano's, and a general variational principle for the solution of the problems. The results of the preceding sections are valid for small deformations. The consideration of large strains, needed for instance in applications of metal forming, requires special consideration. The motion of solids at large deformations, the so-called finite deformations in the kinematics of continuous media, is the subject of Section 9. Topics are the corresponding Lagrangian and Eulerian description, including the derivation of the Green-Lagrange and Euler-Almansi tensor that are used to measure the deformation. Differentiation of these tensors gives the rate of change of the deformation measures. Section 10 is devoted to the stress state. The reader is made familiar with the Cauchy stress tensor, the Piola-Kirchhoff stress tensors, the Kirchhoff stress tensor, and the measures of the rate of change of the stress tensors. The variational principles of virtual displacements and velocities introduced in Section 8 are generalized to the problem of finite deformations in Section 11. Based on the fact that the work of all external forces over virtual displacements is accompanied by the work done by the internal forces, the reader is introduced into the topics of virtual work and the constitutive equations under finite deformations. Part 2, ``The basics of the theory of finite difference schemes'', consists of Chapters 2 to 4. Chapter 2 starts with some historical remarks: Until the advent of computers about 1940, the differential and integral calculus was the main tool to study the problems of solid mechanics. But, after this time, the subject of this book, the mathematical formulation in terms of functions of continuous arguments and the subsequent discretization, has been booming. In order to introduce the reader to the theory of finite difference equations, the author starts universally with the approximation of a problem in the form on an operator equation for a function of a continuous argument. Three steps are emphasized, the grid approximation of the domain of definition and its boundary, the choice of the grid functions defined on the discrete domain, and the substitution of the differential operator by a difference analogon defined on the discrete domain, which reduce the continuous problem to an algebraic system of equations for the function values at the discrete points. In order to prepare the estimation of approximation errors, stability, and convergence problems for the finite difference equations, the functions of the continuous argument are interpreted as elements of a functional space and the set of the function values at the discrete points as a vector space including the norms generated by the scalar products of the function spaces (i.e., the space of square integrable functions \(L^2\) and the Sobolev space \(W^2\)) and of the corresponding vector spaces. The different norms of the functional spaces and the corresponding vector spaces are used to estimate the difference between the two solutions. The author points out that the approximation errors can be estimated in the space of the functional space, used in the finite element method, or by projection of the functions of the functional space into the corresponding vector space with a projection operator, as done in the theory of finite difference equations. The determination of the global approximation error is demonstrated for a family of difference operators and the corresponding norms. Especially, the local approximation error is evaluated using a Taylor series as general example for any difference operator. Other topics are the different orders of magnitude of the approximation error that depends on the norm, the chosen stencil, and the set of nodes involved. The Richardson extrapolation formula is presented to increase the approximation order using embedded grids with decreasing step sizes rather than only more nodes. It is outlined that the approximation condition is a necessary condition for the convergence of the finite difference solution to the solution of the corresponding differential equation, but it is not sufficient. Thus, the notion of the stability of the finite difference system is introduced, and Lax's convergence theorem is formulated. To illustrate the problem, an example of an unstable finite difference scheme is given, here originating from the different orders of the differential equation and the difference approximation. Further topics are the numerical integration of the Cauchy problem of first- and higher-order ordinary differential equations, including stiff and singularly perturbed systems, stability analysis, and examples. The Euler scheme, the Adams-Bashforth scheme, Runge-Kutta schemes, and Gear' implicit method (for stiff systems) are considered. The author goes on with finite difference schemes for one-dimensional partial differential equations. He starts with the hyperbolic wave equation in displacements, the acoustic wave equation, consisting of the equation of motion, the compatibility equation and Hooke's law, and the parabolic unsteady heat equation, in order to introduce different finite difference approximations, such as the cross and leapfrog schemes, the Lax-Friedrichs and the Lax-Wendroff schemes, explicit (e.g., forward Euler, Allen-Cheng, Du Fort-Frankel) and implicit methods (e.g., backward Euler) including the problem of numerical viscosity, the consideration of boundary points, initial and boundary value problems, stability analysis (von Neumann stability condition, Courant-Friedrichs-Lewy (CFL) condition), and step size selection. Systems of algebraic equations, arising from the discretization of differential equations, often have to be solved in the inner cycle of complex programs. Thus, efficient and accurate algorithms are needed for these problems. Direct methods for low-dimensional systems and iterative algorithms for large-scale problems are treated in Chapter 3. The author starts with the definition of the matrix norm and the condition number of a matrix in order to treat accuracy problems in the solution of inhomogeneous linear equations. For instance, using a tridiagonal \(n \times n\) matrix, which results from the discretization of a two-point boundary value problem of a second-order differential equation, it is shown that the condition number increases essentially with an excessively fine partitioning, i.e., the improvement of the approximation accuracy by finer partitioning can result in ill-conditioned systems. The Gaussian elimination (in the kind of matrix factorization into a lower and an upper triangular matrix, backward and forward substitution, as well with multiple right-hand sides) with partial pivoting, but without an estimation of the condition number, for general matrices and the Cholesky decomposition for positive definite matrices are presented. Special direct algorithms for tridiagonal, band and sparse matrices avoiding the storage of the zero elements and saving computing times are mentioned but not presented. There are no hints to modern implementations which employ modern cache-based architectures and more general memory hierarchies. A regularization technique for ill-conditioned systems is outlined. Iterative methods for linear systems of equations are needed for large scale, especially sparse systems. Iterative methods require fewer memory in comparison with the direct algorithms. The reader is introduced into two relaxation methods, the classical Jacobi and Seidel method, dividing the original matrix into a diagonal and two triangular matrices. The convergence is accelerated by a Tschebychev iterative process. A Krylov subspace method for nonsymmetric systems and preconditioning techniques would be an enrichment in this field. Linear systems can be considered as special case of systems of nonlinear equations. Thus, the iterative methods for the nonlinear systems can be treated as a generalization of the linear methods. Nonlinear equations may not have a unique solution but, considering the iterative methods, it is assumed that the user has found separately a domain in which the problem has a unique solution. The author starts with the statement that a solution of a nonlinear equation can be treated as solution to the problem to find a stationary point. Applying the method of successive approximation, sufficient convergence conditions are formulated with help of the Lipschitz condition and the norm of the Jacobian matrix. In contrast to this linear convergent method, the Newton-Raphson method has quadratic convergence. The Newton-Raphson method has the disadvantage that one has to calculate the Jacobian matrix and to solve a linear system at each iteration step. This time-consuming calculation can be reduced by quasi-Newton methods applying an updating of the Jacobian matrix during the iterations, but this is only indicated. Mentioned is the secant method that replaces the derivatives in the Jacobian matrix with a finite difference, resulting in a two-step method. In general, one has to distinguish two kinds of the Newton-Raphson methods. On the one hand, the nonlinear system is linearized, that means the, method consists of two stages, the external cycle of the Newton-Raphson method and the internal cycle with the solution of linear systems. Otherwise, a generalized linear iterative method (for instance, a nonlinear Seidel algorithm) is applied to the nonlinear system and at each stage \(m\) nonlinear equations (\(m\) is the number of unknowns) are solved by the Newton-Raphson method for a nonlinear equation in one variable. In addition, the nonstationary Newton method is demonstrated including an optimal step selection. Two decent methods, the coordinate and the steepest decent method, are presented as methods for the minimization of functions. For the minimization of a quadratic function in \(m\) unknowns a conjugate gradient method, known for the solution of linear systems with positive symmetric matrices, is described. Chapter 4 is focused on the numerical solution of boundary value problems for differential equations. At first, the user is advised against problems with the property that small deviations in the boundary conditions can result in arbitrarily large errors, demonstrated by a stiff two-point boundary value problem, difficulties that need special methods. Then the shooting, the quasi-linearization, and the sweep method for the solution of boundary value problems for ordinary differential equations are treated. The sweep method is also formulated for an initial value problem of the one-dimensional heat equation. Other topics are Fourier's method of separation of variables and the numerical solution of the Poisson equation using a five-point stencil and proving the stability of this finite difference scheme with the help of the maximum principle. The numerical ideas treated in Part 2 are refined and extended to special numerical methods of continuum mechanics problems given in the first part. Chapters 5 to 8 of Part 3 contain extensions to two and three dimensions, to unsteady problems with discontinuous solutions, and new problems that have appeared so far only in journal publications. Chapter 5 is devoted to wave propagation equations which describe different physical problems but are mathematically equivalent. The author starts, based on a numerical analysis of explicit and implicit schemes for longitudinal, transverse, and coupled longitudinal and transverse equations, with instructions which corresponding algorithms should be applied for a special problem. The next topic consists in solution methods (direct and inverse methods, Courant-Isaacson-Rees grid-characteristic scheme) for nonlinear hyperbolic wave propagation systems by the method of characteristics, demonstrated for a nonlinear elastic and an elastoviscoplastic beam. Treated are also associated discontinuous solutions, stability problems, singularly perturbed systems, schemes for stiff problems, characteristic schemes of higher-order accuracy, two- and three-dimensional characteristic schemes, and axisymmetric problems. Another topic is a problem composed of a hyperbolic and a parabolic equation, given by the motion of a heat conducting compressible gas in the form of the equation of motion and the heat conduction equation, respectively. Finite difference splitting methods are the subject of Chapter 6. Multidimensional problems can be reduced to a successive solution of simpler problems applying splitting methods. Explicit and implicit schemes are formulated. The stability of the methods is investigated. Especially, explicit and implicit splitting schemes are presented for 2D/3D heat and wave equation, calculating a set of one-dimensional difference equations rather than the multidimensional equation. Furthermore, Navier-Stokes equations for a thermally conductive viscous gas are treated using splitting schemes. Taking into account that solids possess viscoplastic and elastic properties, these methods are extended for the treatment of elastoviscoplastic dynamic problems, by splitting the corresponding constitutive equations. Another topic is the integration of elastoviscoplastic flow equations, formulating the flow as a single variation inequality for the elastic as well as for plastic domains. The approximation of the conservation laws on curvilinear polygonal meshes with the necessity to approximate the derivatives of a function \(f(x, y)\) inside the mesh cell is considered in Chapter 7. The `formulas for natural approximation of derivatives' are deduced. These formulas possess inter alia the property that a conservative law is exactly approximated on any irregular mesh and results in the so-called conservative finite difference schemes. Another topic is the treatment of the particle-in-cell method for continuum mechanics with large deformations problems, including limitations and modifications as well as applications in the field of elastoviscoplastic problems. The last sections of Chapter 7 are about the optimal choice of a mesh, mesh adaptions, moving meshes, grid refinement strategies, and hybrid schemes of variable order of approximation at internal nodes, including demonstrations on elastoviscoplastic problems. The last chapter is devoted to the concepts of damage and continuum fracture of elastoplastic and elastoviscoplastic materials under quasistatic and dynamic thermomechanical loading. Topics are the presentation, analysis, and numerical realization of the Gurson-Tvergaard-Needlman model and of micromechanical multiscale models including examples.