An Introduction to Numerical Methods for the Solutions of Partial Differential Equations

Partial differential equations arise in formulations of problems involving functions of several variables such as the propagation of sound or heat, electrostatics, electrodynamics, fluid flow, and elasticity, etc. The present paper deals with a general introduction and classification of partial differential equations and the numerical methods available in the literature for the solution of partial differential equations.


Introduction
An equation involving derivatives or differentials of one or more dependent variables with respect to one or more independent variables is called a differential equation.The study of differential equations is a wide field in pure and applied mathematics, physics, meteorology, and engineering, etc.All of these disciplines are concerned with the properties of differential equations of various types.Pure mathematics focuses on the existence and uniqueness of solutions, while applied mathematics emphasizes the rigorous justification of the methods for approximating solutions.Differential equations play an important role in modeling virtually every physical, technical, or biological process, from celestial motion, to bridge design, and interactions between neurons.Differential equations which are used to solve real-life problems may not necessarily be directly solvable, that is, do not have closed form solutions.Instead, solutions can be approximated using numerical methods.Mathematicians also study weak solutions (relying on weak derivatives), which are types of solutions that do not have to be differentiable everywhere.This extension is often necessary for solutions to exist, and it also results in more physically reasonable properties of solutions, such as possible presence of shocks for equations of hyperbolic type.
The theory of differential equations is quite developed and the methods used to study them vary significantly with the type of the equation.
A differential equation involving derivatives with respect to single independent variable is called an ordinary differential equation.In the simplest form, the dependent variable is a real or complex valued function, but more generally, it may be vector-valued or matrix-valued: this corresponds to considering a system of ordinary differential equations for a single variable.Ordinary differential equations are classified according to the order of the highest derivative of the dependent variable with respect to the independent variable appearing in the equation.The most important cases for applications are first-order and second-order differential equations.In the classical literature, the distinction is also made between differential equations explicitly solved with respect to the highest derivative and differential equations in an implicit form.
A differential equation involving partial derivatives with respect to two or more independent variables is called partial differential equation.The partial differential equations can also be classified on basis of highest order derivative.
Some topics in differential geometry as minimal surfaces and imbedding problems, which give rise to the Monge-Ampere equations, have stimulated the analysis of partial differential equations, especially nonlinear equations.Moreover, the theory of systems of first order partial differential equations has a significant interaction with Lie theory and with the work of E. Cartan.
The development of partial differential equations in the 18th and 19th century is given in Kline's book [1].
Until the 1870, the study of partial differential equation was mainly concerned with heuristic methods for finding solutions of boundary value problems as well as explicit solutions for particular problems (for an example, the solution of Dirichlet boundary value problem = 0, u  introduced by Riemann).2 in , u    Poincaré [2] gave the first complete proof of the existence and uniqueness of a solution of the Laplace equation for any continuous Dirichlet boundary condition in 1890.In a fundamental paper of, Poincare [3] established the existence of an infinite sequence of eigenvalues and corresponding eigen-functions for the Laplace operator under the Dirichlet boundary condition.Picard applied the method of successive approximation to obtain solutions of nonlinear problems which were mild perturbations of uniquely solvable linear problems.The construction of elementary solutions and Green's functions for general higher order linear elliptic operators was carried through to the analytic case by E. E. Levi [4].Up to about 1920's solutions of partial differential equations were generally understood to be classical solutions, that is, for a differential operator of order k C .k Keeping in view the requirement of the new researchers, the present paper describes the basic fundamentals of partial differential equations which has been collected from a large number of research articles published in reputed journals and literature available in the books with the intension to provide all important relevant material in a condense form related to partial differential equations and numerical methods for their solutions.Also, since analytical and computational solution of partial differential equations is the major concern from the early years, this paper gives a small step towards the development of computational analysis of partial differential equations, which have lot of utilization in the field of science and engineering.

Classification of Partial Differential Equations
Both ordinary and partial differential equations are broadly classified as linear and nonlinear.A linear partial differential equation is one in which all of the partial derivatives appears in linear form and none of the coefficients depends on the dependent variables.The coefficient may be function of the independent variables.A nonlinear partial differential equation can be described as a partial differential equation involving nonlinear terms.

Types of Non-Linear Partial Differential Equations
The non-linear partial differential equations describe many different physical systems, ranging from gravitation to fluid dynamics and have been used in mathematics to solve problems such as Poincare conjecture and Calabi conjecture.
A non-linear partial differential equation is known as semi-linear if it is linear in highest order derivatives and the coefficients of highest order derivatives depend only on independent variables.

 
A nonlinear partial differential equation is known as quasi-linear if it is linear in highest order derivatives and the coefficients of highest order derivatives depend on independent variables as well on lesser order derivatives. (2) A nonlinear partial differential equation is known as fully non-linear if the coefficients depends on dependent variable or the derivatives appear in nonlinear form.
Example 2.1 = 0 is quasilinear = 0 is nonlinear where a, b are functions of x, y and c is function of x, y and f.But further classification into elliptic, hyperbolic, and parabolic equations, especially for second-order linear equations, is of utmost importance.For more study on linear and quasi linear elliptic equations, see [5,6].

Classification Based on Discriminant
The general quasi linear second-order non-homogeneous partial differential equation in two independent variable is = The classification of partial differential equations depends on the sign of discriminant as fallow:

Eigenvalue Based Classification of Partial Differential Equations
Since method for classification of partial differential equations fails if it is partial differential equation in three or more independent variable then we are not able to find discriminant.
Let us consider a general second order partial equation in n independent variables which also can be written in compact form as: Then coefficient matrix of highest order derivatives is where M  is Mac number.1) First we characterize by discriminant then all eigenvalues are nonzero and of same sign thus equation of flow is elliptic.
then all eigenvalues are zero thus equation of flow is parabolic.
then all eigenvalues are nonzero and of one eigenvalue of opposite sign thus equation of flow is hyperbolic.

Significance of Classification
The classification of a partial differential equations is intimately related to the characteristics of the partial differential equations.The characteristics are   1 n dimensional hyper-surfaces in n-dimensional hyperspace that have some very special features.In two-dimensional space, which is the case considered generally, characteristics are paths in the solution domain along which information propagates.In other words we can say information propagates throughout the solution domain along characteristics.

Classification by Physical Problems
Physical problems falls into one of the following general classification: 1) Equilibrium Problem; 2) Propagation Problem; 3) Eigenvalue Problem.Equilibrium Problem: Equilibrium problems are steady state problems in closed domain in which the solution ( , ) D x y ( , ) f x y is governed by an elliptic 0 , partial differential equations subject to boundary conditions specified at each point on the boundary of the domain.
Such type of problems have no real characteristic and thus the solution at every point in the solution domain is influenced by the solution at all other points and the solution at each points influence the solution at all other points.
Equilibrium problems are solved by method of relaxation numerically.
Propagation Problem: Propagation problems are initial value problems in open domains.Here by open domain means that open with respect to one of the independent variables.
Example 2.3 Eigenvalue Problem: Eigenvalue problem are special problems in which the solution exits only for special values (i.e.eigenvalues) of a parameter of the problem.The eigenvalues are to be determined in addition to the corresponding configuration of the system.

Types of Conditions
Initial Value Problem: An initial value problem is one in which the dependent variable and possibly its derivatives are specified initially (i.e. at time ) or at the same value of independent variable in the equation.Initial value problems are generally time-dependent problems.

Boundary Value Problems:
A boundary value problem is one in which the dependent variable and possibly its derivatives are specified at the extreme of the independent variable.For steady state equilibrium problems, the auxiliary conditions consists of boundary conditions on the entire boundary of the closed solution domain.There are three types of boundary condition.
Example 2.5 Let be a bounded domain in with a smooth boundary and let is called nonlinear elliptic boundary value problem.This type of BVP arises in several domains, for example in physical problems involving the steady-state temperature distribution see [7][8][9].1) Dirichlet boundary condition: The value of the function is specified on the boundary.The dependent variable of the partial differential equation are pre-scribed in domain at different points.For example if an iron rod had one end held at absolute zero then the value of the problem would be known at that point in space.A Dirichlet boundary condition imposed on an ordinary or a partial differential equation specifies the values of a solution is to take on the boundary of the domain.The question of finding solutions to such equations is known as the Dirichlet problem as in (10).

Various Methods for Solving Partial Differential Equation
In literature various method exits for solution of partial differential equations.Here we will discuss some of them briefly as following: 1) Finite Difference Method: The finite difference method is a numerical procedure which solves a partial differential equation by discretizing the continuous physical domain into a discrete finite difference grid, approximating the individual exact partial derivatives in the partial differential equations by algebraic finite difference approximations (i.e.FDA), substituting the FDA's into the partial differential equations to obtain an alge-braic finite difference equation(FDE), and solving the resulting algebraic finite difference equations for the dependent variable.
2) Finite Volume Method: The finite volume method is a method for representing and evaluating partial differential equations in the form of algebraic equations.Similar to the finite difference method, values are calculated at discrete places on a meshed geometry."Finite volume" refers to the small volume surrounding each node point on a mesh.In the finite volume method, volume integrals in a partial differential equation that contain a divergence term are converted to surface integrals, using the divergence theorem.These terms are then evaluated as fluxes at the surfaces of each finite volume.Because the flux entering a given volume is identical to that leaving the adjacent volume, these methods are conservative.The method is used in many computational fluid dynamics packages One advantage of the finite volume method over finite difference methods is that it does not require a structured mesh (although a structured mesh can also be used).Furthermore, the finite volume method is preferable to other methods as a result of the fact that boundary conditions can be applied non-invasively.This is true because the values of the conserved variables are located within the volume element, and not at nodes or surfaces.Finite volume methods are especially powerful on coarse nonuniform grids and in calculations where the mesh moves to track interfaces or shocks.
3) Finite Element Method: The finite element method, where functions are represented in terms of basis functions and the partial differential equations is solved in its integral (weak) form.In the finite element method (FEM) the domain is partitioned in a finite set of  elements , so that  and = .
i   Usually one takes for i triangles or quadrangles.Then the function is approximated by ), where i  are functions that are polynomials on each i  (i.e.piecewise polynomials).Usually the functions i  are continuous polynomials of a low degree.Further they are constructed so that their support extends only over a small number of elements.Now we will give detail discussion of Finite element method.

Finite Element Method
The key idea: The finite element is a numerical method like finite difference method but it is more general and powerful in its application to real-world problems that involve complicated physical geometry and boundary conditions.
In FEM, a given domain is viewed as a collection of sub-domains, and over each sub-domain the governing equation is approximated by any of the traditional variational methods.
The main reason behind taking approximate solution on a collection of sub-domains is the fact that it is easier to represent a complicated function as a collection of simple polynomials.
The method is characterized by three features: 1) The domain of the problem is represented by a collection of simple sub-domains called finite elements.The collection of finite elements is called the finite element mesh.
2) Over each finite element, the physical process is approximated by functions of the desired type and algebraic equations relating physical quantities at selective points, called nodes of the element are developed.
3) The element equation are assembled using continuity and/or "balance" of physical quantities.
In FEM, we seek an approximation u n of u in the form where j u are the values of n at the element nodes u j  are the interpolation function, j c are coefficients that are not associated with nodes, and π j are the associated approximation functions.Direct substitution of the such approximation into the governing differential equation does not always result, for an arbitrary choice of the data of the problem, in a necessary and sufficient no. of equations for the under-determined coefficients j u and j c Therefore a procedure whereby a necessary and sufficient number of equations can be obtained is needed.
There is only one method of finite element model of the same problem.There can be more than one finite element model of the same problem.The type of model depends on the differential equations, method used to derive the algebraic equations for the undetermined coefficients over an element, and nature of the approximations function used.

Variational Principles and Methods
The idea of using a variational formulation of a boundary value problem for its numerical solution goes back to Lord Rayleigh (1894,1896) and Ritz (1908), see, e.g., Kantorovich and Krylov [13].In Ritz's approach the approximation solution was sought as a finite linear com-bination of functions such as, for instance, polynomial or trigonometric polynomials.The use in this context of continuous piecewise linear approximating function based on triangulation adapted to the geometry of the domain proposed by Courant (1943) in the paper based on an address delivered to the American Mathematical Society in 1941.Even though this idea had appeared earlier, also in work by Courant himself (see Babuska [23]), this is often thought as the starting point of the finite element method, but the further development and analysis of the method would occur much later.
Classical sense of the variational principle is to find the extremum or the variables of the problem.The functional includes all the intrinsic features of the problem such as the governing equations, boundary conditions, and constraints condition, if any.
In solid and structural mechanics problems, the functional represents the total energy of the system and in other problems it is simply an integral representation of the governing equations.
First many problems of mechanics are posed in terms of finding the extremum and thus by their nature, can be formulated in terms of variational statement.Second, there are problems that can be formulated by other means, but these can also be formulated by means of variational principles.Third, variational formulations form a powerful basis for obtaining approximate solutions to practical problems, many of which are intractable otherwise.The principle of minimum total potential energy, for example, can be regarded as a substitute to the equations of equilibrium of an elastic body as well as a basis for the development of displacement finite element models that can be used to determine approximate displacement and stress fields in the body.

Variational Formulations
The classical sense of the phrase "Variational Formulation" refers to the construction of a functional or variational principle that is equivalent to the governing equations of the problem.The modern use of the phrase refers to the formulation in which the governing equations are translated into equivalent weighted integral statements that are not necessarily equivalent to a variational principle.
The importance of variational formulation of physical laws, in the modern or general sense of the phrase, goes for beyond its use as simply an alternate to other formulation.In fact, variational forms of the laws of continuum physics may be only natural and rigorously correct way to think of them.While all sufficiently smooth fields leads to meaningful variational forms the converse is not true.There exits physical phenomena which can be ade-quately modeled mathematically only in a variational setting they are nonsensical when viewed locally.
The starting point for the discussion of the finite element method is differential equations governing the physical phenomena under study.As such, we shall first discuss why integral statement of differential equations are needed.

Need for Weighted-Integral Statements
The weighted-integral statement are required in order to generate the necessary and sufficient number of algebraic equations to solve for the parameters j c in Equation ( 13) of approximate solution.=1 ( ) = .
The use of integral statement is equivalent to the governing differential equation is necessitated by the fact that substitution of Equation ( 13) into the governing differential does not always results in the required number of linearly independent algebraic equations for the unknown coefficients .j c One way to insure that there are exactly the same number of equations as there are unknowns is to require weighted integrals of the error in the equation to be zero.We can require the approximate solution to satisfy the given differential equation in the weighted integral sense, where is called residual.R

Linear and Bilinear Functional
Consider the integral expression For a given real function is a real number.Therefore, = ( ), ( ) u u x I u I can be viewed as an operator that transforms functions into real numbers, and such operators called "functionals".
is said to be linear in u if and only if it satisfies the relation for any real numbers  and  and dependent vari- A bilinear form is said to be symmetric in its ar (19) .

Variational Operator and First Variation
A f ( , ) B u v rgumen each of its a ts u and : Consider the function ( , , ).F x u u For any fixed value of the independent variable , x F depends on u and .u The change v  in u wh e er  is a constant uanand v is a fu tion of q tity nc x is called the variation of u and is fined by u de  is called variational operator.

The variation u
 of a function u represents an admissible change in t function ( ) u x at a fixed value of the independent variable .he x is specified value cannot be varied.Thus, the riatio of a function u is zero there because the specified on the boundary, the variation of u is zero there because the specified value cannot be va ed.Thus, the variation of a function u should satisfy the homogeneous form of the boundary conditions for u Fundamental lemma of variational c fu (22) holds for any arbitrary function alculus: The ndamental lemma of variations can be stated as follows for any integrable function ( ) These conditions are also known as Dirichle m on, which require specifica

Na and Essential Boun y Co
Essential boundary conditions which require v and ssible its derivatives to vanish at the boundary.Thus, we have Ŝpecify = 0, or = on the boundary.v u u t or geoetric boundary conditions.Natural boundary conditi tion of the coefficient of v (and possibly its derivatives).Thus we have Specify = , on the boundary.
The necessary condition for I to attain a minimum yields Now suppose that and v are selected such that = 0 for = , = 0 for = .
Then using the fundamental lemma of the calcu va y of th 2) lus of riations, we obtain the same Euler equation.Equations in (29) are satisfied identically for an e following combination:  nsider the problem of finding defined on a tw Co ( , ), u v o dimensional region  such that th ollowing functional is to be minimized , , , , )d d with condition Copyright © 2011 SciRes.
Equations ( 31) and ( 32) represent the essential boundary condition and Equations ( 33) and ( 34) represent the natural boundary conditions.The pair of elements (u,v) are called the primary variables and are secondary variables.
.2.5.Weak Form d to be a weighted integral statement .2.6.Ritz Method e coefficients Weak form is define of a differential equation in which the differentiation is transformed the dependent variable to the weight function such that all natural boundary conditions of the problem are also introduced.

4
In Ritz method, th j c w of the approximations are determined using the eak form of the problem, and hence choice of weight functions is restricted to the approximation function = .

j w 
Consider the variational problem resulting from the w (36) for all sufficiently differentiable fun v i eak form: find the solution u such that ( , ) = ( ) B w u l w ctions w that satisfy the homogeneous form of the specified essential boundary conditions on .
u In general, (, )  B   can be unsymmetric in w and and l is li the problem in (36) is equi alent to minim zation of the quadratic functional In Ritz method, we seek an approxima (3 (38) where the constants tion solution to 6) in the form of a finite series = ( , ), = ( ) The algebraic equations in Equation ( 39) can be expressed in matrix form as:

The Method of Weighted Residuals
The me igh thod of we ted residual can be described in its on generality by considering the operator equati where A is an operator (linear or nonlinear), often a differential operator, acting on the d f a k ependent variables and is nown function of the independent variables.
In the weighted residual method, the solution u is approximated, in much the same way as in Ritz method by the expression 0 0 =1 j uirements on except that the req 0  and j  for the weighted residual method are mo nge for the Ritz method As the name suggests, the parameter re stri nt than those . is referred as the Petrov-Galerkin method when .
the app required roximation functions are used in Galerkin method to be higher order than those in the Ritz method.For further readings we suggest to see [24][25][26][27][28].
3) The least-square method: In the least-square method, we determine the parameters j c by minimizing the integral of the square of the residual. 2( , )d = 0 4) The collocation method: In the collocation meate solution thod, we seek an approxim N U to Equation (44) in the form of N U N by quirin sidual to vanish identically at selected points re g the re = ( , , ) The selection of points The steps involved in finite element method of a problem: 1) Discretization of the given domain into a collection of prescribed finite element.3) Assembly of element equations to obtain the equatio e results.ite element method.

5.
In this section, we derive error estimates for the finite n of whole problem.4) Imposition of the boundary conditions of the problem.
5) Solution of the assembled equation.element method.The discrete v Lax-Milgram theorem solution to the discrete e o we use finite dimensional subspaces of our original Hilbert spaces.We have a more general case when we don't have such subspaces or when the operators in the variational equation are replaced by approximations (for instance by quadrature).We give also error estimates for this case.The error estimates depend on how good we can interpolate elements of Banach spaces subspaces of these Banach spaces, so we have to discuss the interpolation theory in Banach spaces, preceded by a necessary discussion of the formalism of the finite element method.This will give us estimates in the Sobolev norms : ; ;. k km q We will also give an estimate in the 2 Lnorm, but for this we need additional requirements on the problem we consider.
1) The error estimate should give an accurate measure of the discretization error for a wide range of mesh spacolynomial degrees.
2) The procedure should be inexpensive relative to the cost of obtaining the finit eans that error estimates should be calculated using only local computations, which typically require an effort comparable to the cost of ge x.
3) A technique that provides estimates of point-wise errors which can subsequently be used to calculate error measures in several norms is preferable to one that only works in a specific norm.Point-wise error estimates and error es de an indications as to where solution accuracy is insufficient and where refinement is needed.
A posteriori error estimates can roughly be divided into four categories: 1) Residual error estimates: Local finite element problems are created on either an element or a sub-domain and solved for the error estimate.The data s on e residual of the finite element solution.
2) Flux-projection lated by post processing the finite element solution.This flux is smoother than the original finite element flux and an error estimate is obtained from the difference of the two fluxes.
3) Extrapolation error estimates: Two finite element solutions having different orders or different meshes are compared and their differences used to provide an error estimate.
4) Interpolatio unds are used with estimates of the unknown constants.

Conclusions
e present paper gives a comprehensive overview of the fundamentals of partial differential equations and related to literature.Many fun n blance, and in some simple cases they coincide.Nevertheless, with its more systematic use of the variational approach, its greater flexibility, and the way it more easily lends itself to error analysis, the finite element method has become the dominating approach for tackling the partial differential equations together with their applications in science and engineering.

f 2 ) 3 ) 1 2
Neumann boundary condition: The value of derivative normal to the boundary is specified ( f n   is specified on the boundary).For example if one iron rod had heater at one end then energy would be added at a constant rate but the actual temperature would not be known.A Neumann boundary condition imposed on an ordinary or a partial differential equation specifies the derivative values of a solution is to take on the boundary of the domain.For example = ( , ),in u f x u   = 0,on .u x    Mixed boundary condition: The linear combination of Dirichlet and Neumann boundary conditions: specified on the boundary.Mixed boundary conditions are also known as Cauchy boundary condition.A Cauchy boundary condition imposed on an ordinary or a partial differential equation specifies both the values a solution of a differential equation is to take on the boundary of the domain and the normal derivative at the boundary.= ( , ),in u f x u  

Natural boundary conditions are also known
Ritz coefficients are determined such that (36) holds for each = ( =1: ).

N
are homogeneous, then  is taken to be zero and j  must still satisfy the same conditions, Note that the requirement that w be zero at the boundary conditions are specified is satisfied by the ice =

:
The weighted residual m

)
The Galerkin method: If the weight function i  is chosen to be equal to the approximation function i  -the we eth ighted residual m od is better known as Ga le in rkin method.The algebraic equations of the Galerk approximation ts.properties needed for probl ion of element equations for all typical elem ial equation over the typical element.(a) Construct the finite element mesh of the prescribed elemen (b) Number of nodes and elements.(c) Generate the geometry em. 2) Derivat ent in the mesh: (a) Construct the variational formulation of the given different (b) Assume that a typical dependent variable u is of the Select, if already available in the literature, or derive element interpolation functions i  and compute the element matrices.

6 )
Postprocessing of thAuthors can view the references[29][30][31][32][33][34][35][36][37][38][39][40][41][42][43][44][45][46][47] for more study on finError Estimateersion of the Generalized , gives the uniqueness of the quation and gives a first estimate f the error.This theorem is in fact only applicable when ings and p e element solution.This usually m nerating the stiffness matri timates in local (elemental) norms may also provi depend th error estimates: A new flux is calcu n error estimates: Interpolation error bo Th ols for their numerical solution available in the damental ideas and techniques in fiite difference and finite element methods have resemd by Department of cience and Technology, New Delhi, Government of e to thank the anonymous viewers for their valuable comments and suggestions to aux Derivees Partielles de athematique," American Journal of Mathe-, No. 3, 1890, pp.211-294.