Finite Elements and Approximation
By O. C. Zienkiewicz and K. Morgan
4.5/5
()
About this ebook
Starting with continuum boundary value problems and the need for numerical discretization, the text examines finite difference methods, weighted residual methods in the context of continuous trial functions, and piecewise defined trial functions and the finite element method. Additional topics include higher order finite element approximation, mapping and numerical integration, variational methods, and partial discretization and time-dependent problems. A survey of generalized finite elements and error estimates concludes the text.
O. C. Zienkiewicz
O. C. Zienkiewicz was one of the early pioneers of the finite element method and is internationally recognized as a leading figure in its development and wide-ranging application. He was awarded numerous honorary degrees, medals and awards over his career, including the Royal Medal of the Royal Society and Commander of the British Empire (CBE). He was a founding author of The Finite Element Method books and developed them through six editions over 40 years up to his death in 2009. Previous positions held by O.C. Zienkiewicz include UNESCO Professor of Numerical Methods in Engineering at the International Centre for Numerical Methods in Engineering, Barcelona, Director of the Institute for Numerical Methods in Engineering at the University of Wales, Swansea, U.K.
Related to Finite Elements and Approximation
Related ebooks
An Introduction to Computational Fluid Mechanics by Example Rating: 5 out of 5 stars5/5Dynamic Programming: Models and Applications Rating: 2 out of 5 stars2/5Analysis of Numerical Methods Rating: 3 out of 5 stars3/5Differential Equations for Engineers and Scientists Rating: 0 out of 5 stars0 ratingsComputational Fluid Dynamics with Moving Boundaries Rating: 0 out of 5 stars0 ratingsAdvanced Mathematics for Engineers and Scientists Rating: 4 out of 5 stars4/5Complex Variable Methods in Elasticity Rating: 0 out of 5 stars0 ratingsApplied Partial Differential Equations Rating: 5 out of 5 stars5/5Numerical Methods for Scientists and Engineers Rating: 4 out of 5 stars4/5Nonlinear Differential Equations Rating: 0 out of 5 stars0 ratingsIntroduction to Finite Element Analysis: Formulation, Verification and Validation Rating: 0 out of 5 stars0 ratingsMechanics of Composite Materials Rating: 3 out of 5 stars3/5Applied Complex Variables Rating: 5 out of 5 stars5/5Introduction to Differential Geometry for Engineers Rating: 0 out of 5 stars0 ratingsAn Introduction to the Theory of Elasticity Rating: 4 out of 5 stars4/5Splines and Variational Methods Rating: 5 out of 5 stars5/5Methods of Applied Mathematics Rating: 3 out of 5 stars3/5Boundary Value Problems and Fourier Expansions Rating: 0 out of 5 stars0 ratingsBasic Abstract Algebra: For Graduate Students and Advanced Undergraduates Rating: 4 out of 5 stars4/5Equilibrium Statistical Mechanics Rating: 4 out of 5 stars4/5First-Order Partial Differential Equations, Vol. 1 Rating: 5 out of 5 stars5/5Elementary Matrix Algebra Rating: 3 out of 5 stars3/5Numerical Methods: Design, Analysis, and Computer Implementation of Algorithms Rating: 4 out of 5 stars4/5Introduction to the Explicit Finite Element Method for Nonlinear Transient Dynamics Rating: 0 out of 5 stars0 ratingsDifferential Forms with Applications to the Physical Sciences Rating: 5 out of 5 stars5/5Finite Elements of Nonlinear Continua Rating: 4 out of 5 stars4/5Finite Element Methods Rating: 0 out of 5 stars0 ratingsProgramming the Finite Element Method Rating: 0 out of 5 stars0 ratingsNonlinear Finite Element Analysis of Solids and Structures Rating: 0 out of 5 stars0 ratingsTheory of Elasticity and Stress Concentration Rating: 0 out of 5 stars0 ratings
Technology & Engineering For You
The Big Book of Hacks: 264 Amazing DIY Tech Projects Rating: 4 out of 5 stars4/5A Night to Remember: The Sinking of the Titanic Rating: 4 out of 5 stars4/5The Art of War Rating: 4 out of 5 stars4/5The Invisible Rainbow: A History of Electricity and Life Rating: 5 out of 5 stars5/5Vanderbilt: The Rise and Fall of an American Dynasty Rating: 4 out of 5 stars4/5Navy SEALs Bug-In 2025 Guide: Transform Your Home into an Impenetrable Fortress of Security and Resilience Rating: 0 out of 5 stars0 ratingsThe Art of War Rating: 4 out of 5 stars4/5Death in Mud Lick: A Coal Country Fight against the Drug Companies That Delivered the Opioid Epidemic Rating: 4 out of 5 stars4/5Ultralearning: Master Hard Skills, Outsmart the Competition, and Accelerate Your Career Rating: 4 out of 5 stars4/5Longitude: The True Story of a Lone Genius Who Solved the Greatest Scientific Problem of His Time Rating: 4 out of 5 stars4/5Nuclear War: A Scenario Rating: 4 out of 5 stars4/580/20 Principle: The Secret to Working Less and Making More Rating: 5 out of 5 stars5/5The Homeowner's DIY Guide to Electrical Wiring Rating: 5 out of 5 stars5/5The Big Book of Maker Skills: Tools & Techniques for Building Great Tech Projects Rating: 4 out of 5 stars4/5The CIA Lockpicking Manual Rating: 5 out of 5 stars5/5The ChatGPT Millionaire Handbook: Make Money Online With the Power of AI Technology Rating: 4 out of 5 stars4/5Selfie: How We Became So Self-Obsessed and What It's Doing to Us Rating: 4 out of 5 stars4/5The Systems Thinker: Essential Thinking Skills For Solving Problems, Managing Chaos, Rating: 4 out of 5 stars4/5On War: With linked Table of Contents Rating: 4 out of 5 stars4/5Rethinking Narcissism: The Bad---and Surprising Good---About Feeling Special Rating: 4 out of 5 stars4/5My Inventions: The Autobiography of Nikola Tesla Rating: 4 out of 5 stars4/5The Wuhan Cover-Up: And the Terrifying Bioweapons Arms Race Rating: 4 out of 5 stars4/5Summary of Nicolas Cole's The Art and Business of Online Writing Rating: 4 out of 5 stars4/5Seeing Further: The Story of Science and the Royal Society Rating: 4 out of 5 stars4/5Some Future Day: How AI Is Going to Change Everything Rating: 0 out of 5 stars0 ratingsArtificial Intelligence: A Guide for Thinking Humans Rating: 4 out of 5 stars4/5
Related categories
Reviews for Finite Elements and Approximation
4 ratings0 reviews
Book preview
Finite Elements and Approximation - O. C. Zienkiewicz
CHAPTER ONE
Continuum Boundary Value Problems and the Need for Numerical Discretization. Finite Difference Methods
1.1.INTRODUCTION
While searching for a quantitative description of physical phenomena, the engineer or the physicist establishes generally a system of ordinary or partial differential equations valid in a certain region (or domain) and imposes on this system suitable boundary and initial conditions. At this stage the mathematical model is complete, and for practical applications merely
a solution for a particular set of numerical data is needed. Here, however, come the major difficulties, as only the very simplest forms of equations, within geometrically trivial boundaries, are capable of being solved exactly with available mathematical methods. Ordinary differential equations with constant coefficients are one of the few examples for which standard solution procedures are available—and even here, with a large number of dependent variables, considerable difficulties are encountered.
To overcome such difficulties and to enlist the aid of the most powerful tool developed in this century—the digital computer—it is necessary to recast the problem in a purely algebraic form, involving only the basic arithmetic operations. To achieve this, various forms of discretization of the continuum problem defined by the differential equations can be used. In such a discretization the infinite set of numbers representing the unknown function or functions is replaced by a finite number of unknown parameters, and this process, in general, requires some form of approximation.
Of the various forms of discretization which are possible, one of the simplest is the finite difference process. In this chapter we describe some of the essentials of this process to set the stage, but the remainder of this book is concerned with various trial function approximations falling under the general classification of finite element methods. The reader will find later that even the finite difference process can be included as a subclass of this more general category.
Before proceeding further we shall focus our attention on some particular problems which will serve as a basis for later examples. It is clearly impossible to deal in detail in a book of this length with a wide range of physical problems, each requiring an introduction to its background. It is our hope, however, that the few examples chosen will serve to introduce the general principles of approximation, which the readers can then apply to their own particular special cases.
1.2.SOME EXAMPLES OF CONTINUUM PROBLEMS
Consider the example of Fig. 1.1a in which a problem of heat flow in a two-dimensional domain Ω is presented. If the heat flowing in the direction of the x and y axes per unit length and in unit time is denoted by qx and qy respectively, the difference D between outflow and inflow for an element of size dx dy is given as
For conservation of heat, this quantity must be equal to the sum of the heat generated in the element in unit time, say, Q dx dy, where Q may vary with position and time, and the heat released in unit time due to the temperature change, namely, − where c is the specific heat, ρ is the density and (x, y, t) is the temperature distribution. Clearly, this requirement of equality leads to the differential relationship
which has to be satisfied throughout the problem domain Ω.
FIGURE 1.1. Examples of continuum problems. (a) Two-dimensional heat conduction. (b) One-dimensional heat conduction.
Introducing now a physical law governing the heat flow in an isotropic medium,¹ we can write, for the flow component in any direction n,
where k is a property of the medium known as the conductivity. Specifically, in the x and y directions we can then write for an isotropic material
Relationships (1.2) and (1.4) define a system of differential equations governing the problem at hand, and which now requires solution for the three dependent variables ax, qy and .
Such a solution needs the specification of initial conditions at time, say, t = t0 (e.g., the distribution of temperature may be given everywhere in Ω at this time) and of boundary conditions on the surface or boundary Γ of the problem. Typically two different kinds of boundary condition may be involved.
In the first condition, say applicable on a portion Γ of the boundary, the values of the temperature are specified as so we nave
A boundary condition of this form is frequently referred to as being a Dirichlet boundary condition.
In the second condition, applicable on the remainder Γq of the boundary, the values of the heat outflow in the direction n normal to the boundary are prescribed as . Then we can write
or, alternatively,
This type of boundary condition is often called a Neumann boundary condition.
The problem now is completely defined by Eq. (1.2), (1.4), (1.5), and (1.6), and numbers representing the distribution of , qx, and qy at all times can, in principle, be obtained by the solution of this set of equations.
This problem may be expressed in an alternative form by using Eq. (1.4) to eliminate the quantities qx and qy from Eq. (1.2), and now a higher order differential equation in a single independent variable results. Performing this elimination produces the equation
which once again requires the specification of initial and boundary conditions.
In the above we have been concerned with a problem defined in time and space domains, with the former requiring the specification of initial conditions. The independent variables here were x, y, and t. If steady-state conditions are assumed (i.e., the problem is invariant with time and so 6666 6666t = 0), the governing equation (1.2) or (1.7) simplifies. In the latter case we have
which for solution requires only the imposition of boundary conditions of the form (1.5) and (1.6). Such boundary value problems will be the subject of discussion of the major part of this book, but in Chapter 7 we shall return to time-dependent equations and consider possible methods for their solution.
While we have written here the governing equations for a two-dimensional situation, this could have easily been extended to three dimensions to deal with more general problems. On the other hand, in some problems only a one-dimensional variation occurs; in Fig. 1.1b, for instance, we consider the heat flow through a slab in which conditions do not vary with y. Then, from Eq. (1.8), we have for steady state an ordinary differential equation
and the problem domain
is now simply the range 0 x Lx.
Such an ordinary differential equation can be solved analytically, but we shall use it and similar equations extensively to illustrate the application of discretization procedures. This will enable us to demonstrate the accuracy of approximate methods by comparing their results with the exact solutions.
The problem of heat flow just described is typical of many other physical situations and indeed can be identified with problems such as the following.
1. Irrotational ideal fluid flow. If we put k = 1, Q = 0, then Eq. (1.8) reduces to a simple Laplacian form;
which is the equation governing the distribution of the potential in irrotational ideal fluid flow.
2. Flow of fluid through porous media. Here we take Q = 0 and identify k as the medium permeability. The hydraulic head then satisfies Eq. (1.8).
3. Small deformation of membranes under a lateral load. With k = 1 and Q defined to be the ratio of the lateral load intensity to the in-plane tension of the membrane, Eq. (1.8) is the equation governing the transverse membrane deflection .
Other applications will occur to the reader familiar with different physical and engineering problems, and from time to time we shall introduce in this book different applications of the above differential equation and indeed other systems of differential equations.
Although at such times the full exploration of the origin and derivation of such equations may not always be apparent to all readers, we hope that the procedures of mathematical discretization adopted to produce a solution will be clear in each case.
1.3.FINITE DIFFERENCES IN ONE DIMENSION
Suppose we are faced with a simple one-dimensional boundary value problem, that is, we wish to determine a function (x) which satisfies a given differential equation in the region 0 x Lx, together with appropriate boundary conditions at x = 0 and x = Lx. As we have just seen, a typical example of this type of problem would be that of calculating the temperature distribution (x) through a slab of thickness Lx, of thermal conductivity k, with the faces x = 0 and x = Lx maintained at given temperatures and , respectively, and with heat generation at a rate Q(x) per unit length in the slab. The governing differential equation for this problem is given by Eq. (1.9), which reduces to the equation
if we make the assumption that the material thermal conductivity is constant. The associated boundary conditions are of the type given in Eq. (1.5) and can be written as
To solve this problem by the finite difference method we begin by differencing the independent variable x, that is, we construct a set (or grid or mesh) of L +1 discrete, equally spaced grid points xl (l = 0, 1, 2,..., L) on the range 0 x Lx (see Fig. 1.2) with x0 = 0, xL = Lx, and xl+1 − xl = Δx.
The next step is to replace those terms in the differential equation that involve differentiation by terms involving algebraic operations only. This process, of necessity, involves an approximation and can be accomplished by making use of the finite difference approximations to function derivatives. The manner in which such approximations can be made are now discussed.
1.3.1. The Finite Difference Approximation of Derivatives
Using Taylor’s theorem with remainder we can write, exactly,
where θ1, is some number in the range 0 θ1 1. Using the subscript l to denote an evaluation at x = xl this can be written
FIGURE 1.2. Construction of a finite difference mesh over the interval 0 x Lx.
and therefore
This leads to the so-called forward difference approximation of the first derivative of a function in which
The error E in this approximation can be seen to be given by
and as E is equal to a constant multiplied by Δx, we say that this error is O(Δx). This is known as the order of the error.
The exact magnitude of the error cannot be obtained from this expression, as the actual value of θ1 is not given by Taylor’s theorem, but it follows that
Figure 1.3 shows a graphical interpretation of the approximation that we have derived mathematically. The first derivative of (x) at x = xl is the slope of the tangent to the curve y = (x) at this point, that is, the slope of the line AB. The forward difference approximation is the slope of the Une AC, and it can be seen that the slope of this line approaches that of the line AB as the mesh spacing Δx gets smaller.
FIGURE 1.3. A graphical interpretation of some finite difference approximations to d /dx. Forward difference—slope of AC; backward difference—slope of DA; central difference—slope of DC.
In a similar manner we can use Taylor’s theorem to obtain
where 0 θ2 1. Rewriting this expression in the form
we can produce the backward difference approximation
The error E in this approximation is again O(Δx), and now
The graphical representation of the backward difference approximation can be seen in Figure 1.3; the slope of the line AB is now approximated by the slope of the line AD.
In both the forward and the backward difference approximations the error is of the same order, that is, O(Δx). However, if we replace the expansions of Eqs. (1.14) and (1.19) by
then a more accurate representation for the first derivative can be obtained by subtracting Eq. (1.23b) from Eq. (1.23a). The resulting equation
can be used to derive the central difference approximation
and the error E in this approximation satisfies
As the error here is O(Δx²), this should now be a better representation than either the forward or the backward difference approximation. This can again be seen in Fig. 1.3, where the graphical interpretation is that we are now approximating to the slope of the line AB by the slope of the line DC. Again, adding the Taylor expansions
we find that the terms involving the first and third derivatives disappear. The result is that
and so we can approximate the second derivative by
The error E in this approximation is O(Δx²) and satisfies
These approximations to first and second derivatives are sufficient for our present purposes, but approximations (of increasing complexity) to higher order derivatives can be obtained in a similar manner if so required. This point is briefly considered further in Section 1.10.
1.3.2. Solution of a Differentia! Equation by the Finite Difference Method
If we evaluate Eq. (1.11) at a typical grid point xl, we obtain, exactly,
and using the approximation of Eq. (1.29) for the second derivative produces the equation
An equation of this form arises at each of the interior grid points xl (l = 1, 2,..., L − 1) on the finite difference mesh. Writing down these equations separately gives [changing the sign and inserting the boundary conditions of Eq. (1.12)]
If is a column vector whose transpose is ( ), tnen this set of equations may be written as a single matrix equation,
where
Thus the original problem of determining an unknown continuous function (x) has been replaced by the problem of solving a matrix equation for the discrete set of values .
The finite difference method will therefore give information about the function values at the mesh points, but it gives us no information about the function values between these points. Indeed we have only approximated to the governing equation at a discrete number of points and not throughout the region.
The solution to Eq. (1.34) can be efficiently computed by noting that the matrix K is symmetric, positive definite and tridiagonal, and then using an inversion algorithm specifically designed for such an equation system.² It must be remembered that the resulting solution only approximates to the exact solution of the problem as originally posed because of the approximation involved in replacing Eq. (1.31) by Eq. (1.32). However, the fact that the error in the approximation is O(Δx²) indicates that reducing the mesh spacing should reduce the error involved and produce a more accurate solution.
The practical use of the finite difference method will now be illustrated by applying the general theory outlined above to a particular simple example.
Example 1.1
It is required to obtain the solution of the equation d² /dx² − = 0 which satisfies the boundary conditions = 0 at x = 0 and = 1 at x = 1. A mesh spacing is chosen, as shown in Fig. 1.4, and the solution may then be found by the finite difference method.
The only unknowns are 1 and 2, the values of the solution at the points and , respectively, while the given boundary conditions imply that 0 = 0 and 3 = 1.
The equation evaluated at a general grid point xl is
which can be expressed in finite difference form, using the approximation of Eq. (1.29), as
Using this equation for l = 1 and l = 2, that is, the two interior points, and inserting the known conditions, gives
FIGURE 1.4. Finite difference mesh adopted for the solution of Examples 1.1 and 1.2.
with solution
The reader can readily obtain the analytical solution for this simple example and the above results can then be compared with the exact grid point values of 0.2889 and 0.6102, respectively.
If the calculation is repeated with , the solution of the resulting equation system gives values of 0.2890 and 0.6104 for the finite difference approximations to the value of at and , respectively. The improvement in the accuracy of the finite difference approximations with a decrease in mesh spacing is then apparent.
EXERCISES
1.1. Solve the equation subject to the conditions = 1 at x = 0 and = 0 at x = 1, using a mesh point spacing Δx = 0.25. Compare the resulting finite difference solution with the exact solution.
1.2. The distribution of bending moment M in a beam subjected to loading by a distributed load w(x) per unit length satisfies the equation d²M/dx² = w(x). A beam of unit length is simply supported (i.e., M = 0) at both ends and carries a load w(x) = sin πx per unit length. Calculate the distribution of the bending moment by the finite difference method using a mesh point spacing Δx = 0.25.
1.3. The equation governing the variation in temperature T in a viscous fluid flowing between two parallel plates (y = 0 and y = 2 H) is given to be
where µ, k, and U are the viscosity, thermal conductivity, and maximum velocity of the fluid respectively. If µ = 0.1, k = 0.08, H = 3.0, and U = 3.0, calculate the temperature distribution when one plate is held at T = 0 and the other at T = 5, using the finite difference method and a mesh point spacing Δy = 0.5 H.
1.4. A cable at tension T is held fixed at its ends x = 0 and x = 1 and rests on an elastic foundation of stiffness k. When the cable is loaded transversely with load w per unit length, the deflection of the cable osatisfies the equation Solve this equation by the finite difference method for the case k/T = 1, w/T = 1, using a mesh point spacing Δx = 0.1 and a suitable computer program. Compare your solutions with the exact answer to this problem.
1.5. On a finite difference mesh with point spacing Δx, the first derivative is to be replaced by the approximation where a and b are constants. Show that if it is required that this approximation be exact whenever is a linear function of x, then , that is, the forward difference method. If the same requirement is applied to the approximation show that the result is the backward difference method. (Hint: First move the origin to x = xl using X = x − xl and then require equality for both = 1 and = X.)
1.6. Prove that if the approximation is exact whenever is a quadratic function of x, then a, b, and c are such that the approximation is that of Eq. (1.29). Show that this approximation is also exact if is a cubic.
1.7. Construct an approximation
by requiring that the approximation be exact if is a quartic function of x. What is the order of the error in this approximation?
1.4.DERIVATIVE BOUNDARY CONDITIONS
Frequently in real problems one (or more) of the associated boundary conditions may be expressed in terms of a derivative; for example, returning to our heat flow problem of Section 1.3, suppose that we now assume that the surface x = Lx of the slab is subjected to a condition of prescribed heat flux q across the surface (i.e., a condition of the form of Eq. (1.6) is to be applied). Using Eq. (1.6), the appropriate condition at x = Lx is now not that the temperature itself is specified at this point, but that the gradient of the temperature is specified, namely,
Then if we repeat the work of the previous section-and write down the finite difference equation at each interior point, we obtain
which, since L is now unknown, is a set of L − 1 equations in the L unknowns . The missing equation has to be provided by the boundary condition of Eq. (1.37), which can be written as
If the derivative is replaced by the backward difference approximation of Eq. (1.21), then this condition becomes
which together with Eq. (1.38) produces a complete set of L equations in the unknowns .
Example 1.2
Return again to the equation considered in Example 1.1, but subjected now to the boundary conditions = 0 at x = 0 and d /dx = 1 at x = 1. If the finite difference mesh shown in Fig. 1.4 is used, the unknowns are 1, 2, and 3, while the boundary conditions give .
The finite difference approximation of the governing equation at x1 and at x2 becomes
and using a backward difference representation of the derivative boundary condition at x3 produces
The solution of this set of equations can be found to be
and the corresponding exact solutions in this case are 0.2200, 0.4648, and 0.7616.
There is an inconsistency in the above analysis in that we have represented the differential equation to within an error which is O(Δx²), whereas our backward difference approximation to the derivative means that