This preview shows page 1. Sign up to view the full content.
Unformatted text preview: absolute value j = 1 2 : : : J ; 1: Since this result is valid for j 2 1 J ; 1], it holds for the particular value of j where the
left side attains its maximum value. Thus, we can replace the left side absolute value by
its maximum norm to obtain ken+1k1 kenk1 + tk nk1:
Iterating the inequality over n ken+1k1 ken;1k1 + t(k nk1 + k
Thus,
where ;1 k n 1) ::: : kenk1 ke0 k1 + n t
= 0<k<n;1 k k k1:
max Neglecting round o errors, solutions of the partial di erential and di erence equations
both satisfy the same initial conditions, so ke0 k1 = 0. Additionally, if T is the total
time of interest, n t T for all combinations of n and t. Thus, kenk1 T : 10 Basic Theoretical Concepts Using (3.1.2c) we bound as tK + 2
where
Thus, K = 0 t max x 1 juttj
T0 x2 M 12 M = 0 t max x 1 juxxxxj:
T0
2 kenk1 T ( 2t K + 12x M ):
Letting x and t approach zero, we see that kenk1 ! 0 hence, the forward time centered space scheme (2.3.3) converges to the solution of the heatconduction problem
(2.3.1) in the maximum norm when r 1=2.
Example 3.1.7. We show that the forward timebackward space scheme (2.2.4) Ujn+1 = (1 ; jn)Ujn + jnUjn;1:
for the kinematic wave equation (2.2.1) is stable in the maximum norm when the Courant
n
number j a(Ujn ) t= x 2 (0 1]. (Recall that this condition must be imposed to
satisfy the Courant, Friedrichs, Lewy Theorem 2.2.1.)
We'll use the stability de nition (3.1.10) and begin by taking the absolute value of
the di erence scheme and using the triangular inequality to obtain jUjn+1 j j1 ; jnjjUjnj + j jnjjUjn;1j:
n
Since 0 < j 1 by the Courant, Friedrichs, Lewy Theorem, we can remove the absolute
n
value signs from the terms involving j to get jUjn+1j (1 ; jn)jUjnj + jnjUjn;1j:
Replacing the solution values on the right side by their maximum values jUjn j kUnk1 8j: As in Example 3.1.6, the inequality must hold for the value of j that maximizes its left
side thus, kUn+1 k1 kUnk1: 3.1. Consistency, Convergence, Stability 11 This inequality holds for all n and may be iterated to yield kUnk1 kU0k1
which establishes stability in the sense of (3.1.14b).
In the two previous examples, we analyzed the convergence and stability of nite
di erence schemes using very similar arguments. These arguments can be generalized
further to obtain a \maximum principle." Theorem 3.1.1. A su cient condition for stability of the onelevel nite di erence scheme Ujn+1 = X
jsj S csUjn+s in the maximum norm is that all coe cients cs , jsj S , be positive and add to unity.
Proof. The arguments follow those used in Examples 3.1.6 and 3.1.7.
Remark 12. A onelevel di erence scheme is one that only involves solutions at time
levels n and n + 1. A multilevel di erence scheme might involve time levels prior to n as
well.
Typically, we not only want to know that a given scheme converges but the rate at
which the numerical and exact solutions approach each other. This prompts the notion
of order of accuracy. De nition 3.1.11. A consistent nitedi erence approximation of a partial di erential
equation is of order p in time and q in space if
n
j = O( tp) + O( xq ): (3.1.15) Problems
1. Using (3.1.2b), we easily see that the forward timecentered space scheme (3.2.3a)
for the heat conduction equation is rst order in time and second order in space.
Show that this scheme has an O( x2) local discretization error when r = t= x2 =
1=6. 12 Basic Theoretical Concepts
2. Show that the Euclidean norm of a J J matrix A given by (3.1.12) follows from
from the de nition of a matrix norm (3.1.10) upon use of the Euclidean vector
norm (3.1.9b).
3. In Example 3.1.7, we used the maximum principle (Theorem 3.1.1) to establish
stability of the forward timebackward space scheme (2.2.4). It can likewise be
used to establish stability in the maximum norm of the forward timeforward space
n
scheme (2.2.2b) when ;1 j 0. Can it be applied to the forward timecentered
space scheme (2.2.5)? 3.2 Stability Analysis Using a Discrete Fourier Series
A discrete Fourier series can be used to analyze the stability of constant coe cient nitedi erence problems with periodic initial data in much the same way that an in nite
Fourier series can be used to solve constant coe cient partial di erential equations.
This method was introduced by John von Neumann and is called von Neumann stability
analysis.
Thus, suppose that the solution of a nite di erence problem is periodi...
View
Full
Document
This document was uploaded on 03/16/2014 for the course CSCI 6840 at Rensselaer Polytechnic Institute.
 Spring '14
 JosephE.Flaherty

Click to edit the document details