9th Indo-German Winter Academy

FUNDAMENTALS OF FINITE DIFFERENCE METHODS (FDM)

SHIPRA AGARWAL Chemical Engineering IIT Bombay

______________________________________________________________________________

Finite Difference Methods

Shipra Agarwal, IIT Bombay

FDM: Acknowledgement
It has been a great learning experience to prepare this lecture. I would like to take this opportunity to thank my supervisors: Professor G. Biswas Professor V. Buwa Professor Ruede
Major portion of this lecture has been prepared from the lecture notes of Professor G. Biswas.

______________________________________________________________________________

Finite Difference Methods

Shipra Agarwal, IIT Bombay

FDM: Outline
Partial Differential Equations
Discretization Elementary Finite Difference Quotients Taylor Series Expansion Truncation Error Feature of Finite Difference Equations Explicit and Implicit Finite Difference Method Error and Stability Analysis Fluid Flow Modeling Conservative Properties Transportive Properties
______________________________________________________________________________

Finite Difference Methods

Shipra Agarwal, IIT Bombay

FDM: Classification of Partial Differential Equations

Linear Equations: A,B,C,D,E and F are constants or f(x,y) Non-linear Equations: A,B,C,D,E and F contain or its

derivatives Quasilinear Equations: Important subclass of nonlinear equations. A,B,C,D,E and F may be function of or its first derivatives Homogeneous: G=0 Parabolic Equation: B2 - 4AC = 0 Elliptic Equation: B2 4AC < 0 Hyperbolic Equation: B2 4AC > 0

______________________________________________________________________________

Finite Difference Methods

Shipra Agarwal, IIT Bombay

FDM: PDE Contd.

Navier Stokes Equation is elliptic in space and parabolic

in time Laplace equation and Poisson Equation are elliptic Fluid flow problems have non-linear terms called advection and convection terms in momentum and energy equations respectively Unsteady 2-D problem represented as:
The equation is parabolic in time and elliptic in space
______________________________________________________________________________

Finite Difference Methods

Shipra Agarwal, IIT Bombay

FDM: Boundary and Initial Conditions:

Complete specification of geometry of interest, appropriate

boundary conditions Conservation equations to be applied within the domain No. of Boundary conditions required is order of highest derivative appearing in each independent variable Unsteady equations governed by a first derivative in time require initial condition to carry out the time integration Three types of spatial boundary conditions: Dirchlet Condition Neumann Condition Mixed Boundary Condition ______________________________________________________________________________
Finite Difference Methods Shipra Agarwal, IIT Bombay

FDM: Introduction
Basic philosophy: Replace derivatives of governing equations

with algebraic difference quotients

Results in a system of algebraic equations solvable for

dependent variables at discrete grid points

Analytical solutions provide closed-form expressions

variation of dependent variables in the domain points in the domain

Numerical solutions (finite difference) - values at discrete

______________________________________________________________________________

Finite Difference Methods

Shipra Agarwal, IIT Bombay

FDM: Discrete Grid Points

x and y spacing in positive x

and y direction x and y not necessarily uniform In some cases, numerical calculations performed on transformed computational plane having uniform spacing in transformed variables but non uniform spacing in physical plane Grid points identified by indices i and j in positive x and y direction respectively
______________________________________________________________________________

Finite Difference Methods

Shipra Agarwal, IIT Bombay

FDM: Elementary Finite Difference Quotients

Taylor Series Expansion:

For small x higher order terms can be neglected.

n-order accurate
Finite Difference Methods

Truncation error
Shipra Agarwal, IIT Bombay

______________________________________________________________________________

FDM: Elementary Finite Difference Quotients Contd.

Forward Difference Truncation Error (x)

Backward Difference

Truncation Error (x)

Central Difference

Truncation Error (x)2

Central Difference for Second Derivative ______________________________________________________________________________

Finite Difference Methods

Shipra Agarwal, IIT Bombay

FDM: Elementary Finite Difference Quotients Contd.

Central difference of second derivative

difference of first derivatives independent variables

forward backward differences of

Similar technique to generate mixed derivative (2u/xy):

______________________________________________________________________________

Finite Difference Methods

Shipra Agarwal, IIT Bombay

FDM: Finite Difference Equations

Basic concept: Replace each term of PDE with its finite

difference equivalent term Partial difference equation:

Using Forward time Central Space (FTCS)method of

discretization:

n: conditions at time t i: grid point in spatial dimension Truncation Error (TE) = (t, (x)2)
______________________________________________________________________________

Finite Difference Methods

Shipra Agarwal, IIT Bombay

FDM: Example 1D Poisson Equation

Boundary Value Problem: Central difference approximation

System of Linear Equations

______________________________________________________________________________

Finite Difference Methods

Shipra Agarwal, IIT Bombay

FDM: Example 1D Poisson Equation contd.

Can be represented in matrix form:

[A]u = F

______________________________________________________________________________

Finite Difference Methods

Shipra Agarwal, IIT Bombay

FDM: Consistency
A finite difference representation of a PDE is said to be

consistent if:

For equations where truncation error is (x) or (t) or

higher orders, TE vanishes as the mesh is refined However, for schemes where TE is (t/x), the scheme is not consistent unless mesh is refined in a manner such that t/x0 For the Dufort-Frankel differencing scheme (1953), if t/x does not tend to zero, a parabolic PDE may end up as a hyperbolic equation
______________________________________________________________________________

Finite Difference Methods

Shipra Agarwal, IIT Bombay

FDM: Convergence
A solution of the algebraic equations that approximate a PDE

is convergent if the approximate solution approaches the exact solution of the PDE for each value of the independent variable as the grid spacing tends to zero

RHS is the solution of algebraic equation

______________________________________________________________________________

Finite Difference Methods

Shipra Agarwal, IIT Bombay

FDM: Explicit Finite Difference Method

Explicit method uses the fact that we know the dependent

variable, u at all x at time t from initial conditions Since the equation contains only one unknown, (i.e. u at time t+t), it can be obtained directly from known values of u at t The solution takes the form of a marching procedure in steps of time
______________________________________________________________________________

Finite Difference Methods

Shipra Agarwal, IIT Bombay

FDM: Crank-Nicolson Implicit method

The unknown value u at time level (n+1) is expressed both in

terms of known quantities at n and unknown quantities at (n+1) The spatial differences on RHS are expressed in terms of averages between time level n and (n+1)

The above equation cannot result in a solution of

at grid point

i The eq. is written at all grid points resulting in a system of algebraic equations which can be solved simultaneously for u at all i at time level (n+1)
______________________________________________________________________________

Finite Difference Methods

Shipra Agarwal, IIT Bombay

FDM: Crank-Nicolson Implicit method contd.

The equation can be rearranged as

where r = t/(x)2 On application of eq. at all grid points from i=1 to i=k+1 , the system of eqs. with boundary conditions u=A at x=0 and u=D at x=L can be expressed in the form of Ax = C

A is the tridiagonal coefficient matrix and x is the solution vector. The

______________________________________________________________________________

eq. can be solved using Thomas Algorithm

Finite Difference Methods

Shipra Agarwal, IIT Bombay

FDM:2-D Heat Conduction Equation

u(temperature) = f(x,y,t)

Explicit Method Crank-Nicolson Method

where The resulting system of linear equations is a function of five unknowns have to be solved at (n+1) time level. The method is very time consuming. Instead we use Alternating Direction Implicit (ADI) methods ______________________________________________________________________________
Finite Difference Methods Shipra Agarwal, IIT Bombay

For a (6x6) computational grid, a system of 16 linear equations

FDM: Alternating Direction Implicit method (ADI)

ADI Two step Scheme
Step I

For each j, a tridiagonal matrix can be formed for varying i index and values from i=2 to (imax-1) at (n+1/2) can be obtained.
Step II

______________________________________________________________________________

Now, for each i, another tridiagonal matrix can be formed for varying j index and values from j=2 to (jmax-1) at nth level can be obtained.

Finite Difference Methods

Shipra Agarwal, IIT Bombay

FDM: Explicit and Implicit Methods

Explicit Method
Advantage: Simple solution

Implicit Method
Advantage: Stability maintained

algorithm
Disadvantage: For a given x,

over larger t, fewer time steps required

Disadvantage:
More involved procedure

stability constraints impose a maximum limit on t, many time steps required

More time consuming due to

matrix manipulations involved in the algorithm Due to larger t, truncation error is also larger
______________________________________________________________________________

Finite Difference Methods

Shipra Agarwal, IIT Bombay

FDM: Errors and Stability Analysis

A = Analytical solution of PDE

D = Exact solution of finite difference equation

N = Numerical solution from a real computer with finite

accuracy Discretization Error = A D = Truncation error + error introduced due to the treatment of boundary condition Round-off Error = = N D N=+D will be referred to as error henceforth
______________________________________________________________________________

Finite Difference Methods

Shipra Agarwal, IIT Bombay

FDM: Errors and Stability Analysis contd.

Consider the 1-D unsteady state heat conduction equation and its FDE

N must satisfy the finite difference equation.

Also, D being the exact solution also satisfies FDE Subtracting above 2 equations, we see that error also satisfies FDE.

If errors is shrink or remain same from step n to n+1, solution is

stable. Condition for stability is

Finite Difference Methods

______________________________________________________________________________

Shipra Agarwal, IIT Bombay

FDM: Von Neumann Stability Analysis

Assume distribution of errors along x-axis is given by Fourier series

and time wise distribution is exponential in t. Since the difference is linear, behavior of one term of series is same as the series
I: unit complex number k: wave number

Substituting the expression for in FDE and simplifying, we have for

stability

The stability requirement for which the solution of the difference

will be stable is
______________________________________________________________________________

Finite Difference Methods

Shipra Agarwal, IIT Bombay

FDM: First Order Wave Equation

Second order wave equation:

u: amplitude c: wave speed

Eigen values of [A] are 1 = +c and 2 = -c representing two travelling waves with speeds given by

Eulers equation may be treated as a system of first order

______________________________________________________________________________

wave equations

Finite Difference Methods

Shipra Agarwal, IIT Bombay

FDM: Stability of Hyperbolic and Elliptic Equations

Consider First order wave equation:

Representing spatial derivative by central difference and time

derivative by first order difference:

Time derivative employed is called Lax Method of Discretization (u(t)

is represented by average value between grid points i-1 and i+1

Using Von Neumann Stability Analysis,

we get

C: Courant number ______________________________________________________________________________

Amplification factor , G as For stability requirement |G| 1,

Finite Difference Methods

Shipra Agarwal, IIT Bombay

FDM: Fluid Flow Modeling

Fluid mechanics: More complex, governing PDEs form a

nonlinear system Burgers Equation: Includes time dependent, convective and diffusive term Here u: velocity : coefficient of viscosity : any property which can be transported or diffused Neglecting viscous term, remaining equation is a simple analog of Eulers equation
______________________________________________________________________________

Finite Difference Methods

Shipra Agarwal, IIT Bombay

FDM: Conservative Property

FDE possesses conservative property if it preserves integral

conservation relations of the continuum Consider Vorticity Transport Equation:

where is nabla, V is fluid velocity and is vorticity. Integrating over a fixed region we get, which can be written as i.e. rate of accumulation of in is equal to net advective flux rate plus net diffusive flux rate of across Ao into The concept of conservative property is to maintain this integral relation in finite difference representation. ______________________________________________________________________________
Finite Difference Methods Shipra Agarwal, IIT Bombay

FDM: Conservative Property contd.

Consider inviscid Burgers equation:
FDE analog

Evaluating the integral i=I2 ,

over a region running from i=I1 to

Thus, the FDE analogous to inviscid part of the integral has preserved the conservative property For the non-conservative form of inviscid Burgers equation:

i.e. FDE analog has failed to preserve the conservative property

______________________________________________________________________________

Finite Difference Methods

Shipra Agarwal, IIT Bombay

FDM: The Upwind Scheme

The inviscid Burgers equation in the following forms are

unconditionally unstable:

The equations can be made stable by using backward space

difference scheme if u > 0 and forward space difference scheme if u<0

Upwind method of discretization is necessary in convection

dominated flows

______________________________________________________________________________

Finite Difference Methods

Shipra Agarwal, IIT Bombay

FDM: Transportive Property

FDE formulation of a flow is said to possess the transportive property if

the effect of perturbation is convected only in the direction of velocity Consider model Burgers equation in conservative form and a perturbation m = in for u>0, all other =0 Using FTCS, we find the transportive property to be violated On the contrary when an upwind scheme is used,

Downstream Location (m+1) Point m of disturbance Upstream Location (m-1)

Upwind method maintains unidirectional flow of information. _____________________________________________________________________________ ______________________________________________________________________________

Finite Difference Methods

Shipra Agarwal, IIT Bombay

FDM: Upwind Differencing and Artificial Viscosity

Substituting Taylor series expansion in the upwind scheme of model Burgers equation

we get, where e is known as the numerical or artificial viscosity Considering u>0 and C<1 (due to CFL condition), e is always a nonzero positive quantity (so that the algorithm can work)
______________________________________________________________________________

Finite Difference Methods

Shipra Agarwal, IIT Bombay

FDM: Upwind Differencing and Artificial Viscosity contd.

For a steady state analysis:

Substituting Taylor series expansion in the upwind scheme of a two-

dimensional convective-diffusive equation

where

with
Some amount of upwind effect is necessary to maintain transportive
______________________________________________________________________________ _____________________________________________________________________________

property of flow equations.

Finite Difference Methods

Shipra Agarwal, IIT Bombay

FDM: Second Upwind Differencing or Hybrid Scheme

u: velocity in x direction

and For uR > 0 and uL > 0

______________________________________________________________________________

Finite Difference Methods

Shipra Agarwal, IIT Bombay

FDM: References
Prof. G. Biswass Lecture Notes Wikipedia http://www.mathematik.uni-dortmund.de

______________________________________________________________________________

Finite Difference Methods

Shipra Agarwal, IIT Bombay

Finite Difference Methods

THANK YOU

______________________________________________________________________________

Finite Difference Methods

Shipra Agarwal, IIT Bombay

Finite Difference Methods

APPENDIX

______________________________________________________________________________

Finite Difference Methods

Shipra Agarwal, IIT Bombay

FDM: Thomas Algorithm

Modified Gaussian matrix-solver applied to tridiagonal

system. Idea is to transform coefficient matrix into upper triangular form

______________________________________________________________________________

Finite Difference Methods

Shipra Agarwal, IIT Bombay

FDM: Thomas Algorithm contd.

Back Substitution:

______________________________________________________________________________

Finite Difference Methods

Shipra Agarwal, IIT Bombay