Unformatted text preview: of Example 2.1.1
00 00 y 0 = y y (0) = 1 : Since we know the exact solution to this problem, we may use (2.1.12) to calculate
( )2
= 2 0max j ( )j = 2
tT
2 d h y 00 h t : With = ;100 and = 0 001, we have = 0 005. This value may be compared with
the actual local error 0.00484 that appears in the second line of Table 2.1.1. (Computed
local error values are not available for the other steps recorded Table 2.1.1. Why?) In
this case, the bound compares quite favorably with the exact local error. However, bear
in mind that we have very precise data for this example, including an exact solution.
With ( ) = , we have
h : d : d fty y j ( ) ; ( )j = j jj ; j
fty Hence, we may take
= 0 01 yields
T : L ftz y z: = j j = 100 as a Lipschitz constant. Then, using (2.1.13) with
001
j nj 0200 ( 100(0:01) ; 1)104 0 0859
: e e : : Comparing this bound and the actual error on the last line of Table 2.1.1, we see that the
bound is about 4.5 times the actual error. Comparing formula (2.1.13) with the results
of Table 2.1.2, reveals that both predict a linear convergence rate in .
Example 2.1.4. Let's solve the problem of Examples 2.1.1 and 2.1.3 with a di erent
choice of . Thus, we'll keep = ;100 and choose = 0 05 to obtain the di erence
equation (Example 2.1.1)
h h h yn = (1 + )
h yn;1 = ;4 yn;1 10 n> : 0 y0 =1 : Results, shown in Table 2.1.3, are very puzzling. As ( ) increases, the solution increases
exponentially in magnitude while oscillating from time steptotime step. The results are
clearly not approximating the decaying exponential solution of = ;100 .
n t y n 0
1
2
3
4
5
6
7
8 Table 2.1.3: The solution of
= 0 05.
h y 0 = y tn 0 y yn 0.00
1
0.05
4
0.10
16
0.15
64
0.20
256
0.25 1024
0.30 4096
0.35 16384
0.40 65536
, (0) = 1, by Euler's method for = ;100 and
y : The results of Example 2.1.4 are displaying a \numerical instability." Stability of
a numerical method is analogous to \well posedness" of a di erential equation. Thus,
we would like to know whether or not small changes in, e.g., the initial data or ( )
produce bounded changes in the solution of the di erence scheme.
fty De nition 2.1.5. Let n, 0, be the solution of a onestep numerical method with
initial condition 0 and let n be the solution of the same numerical method with a perturbed initial condition 0 + 0 . The onestep method is stable if there exists positive
constants ^ and such that
y y n> z y h k j n ; nj
y whenever j 0 j z k 8 nh T h 2 (0 ^ )
h (2.1.14) . Remark 2. A onestep method, like Euler's method, only requires information about
the solution at tn 1 to compute a solution at tn . We'll have to modify this de nition
when dealing with multistep methods.
Remark 3. The de nition is di cult to apply in practice. It is too dependent on
f (t y ).
; 11 Example 2.1.5. Although De nition 2.1.5 is di cult to apply, we can apply it to
Euler's method. Consider the original and perturbed problems
yn zn = zn;1 = + yn;1 ( hf tn;1 yn;1 ( hf tn;1 zn;1 Let
n + = zn ; ) )
= z0 yn y0 + 0: 0 n...
View
Full
Document
This document was uploaded on 03/16/2014 for the course CSCI 6820 at Rensselaer Polytechnic Institute.
 Spring '14
 JosephE.Flaherty

Click to edit the document details