Backward Method
:
Instead of using Taylor's formula at the point x0+h
one can also use x0-h and develop Euler's backward formula
in the following manner. Consider the Taylor series at x = x0
-
h
y(x0 - h) = y(x0) - hy'(x0)
+ (h2/2) y''(x0) + . . .
If we truncate the infinite series on the right hand side after second
term, we can write
|
y(x0 - h) = y(x0) - hy'(x0) |
|
|
-y(x0) = -y(x0 - h) - h f(x0, y0) |
|
or |
yi = yi-1 + h fi |
for any i |
since fi = f(xi, yi) has
also the unknown yi, the formula has more than one term
which has the unknown and hence the scheme is an implicit one.
Local truncation error is (h2/2) y''(x)
where
xi
< x < xi+1.
That is, the order of the scheme is same as Euler forward scheme, which
is very convenient to use because of its explicit nature whereas the backward
scheme is an implicit one in which an approximation to yi
has to be computed with help of some other scheme. Hence using Euler
Forward scheme is time saving process than Euler backward scheme with the
same order of accuracy.
Mid-point
Method :
Euler's forward and backward schemes
yi+1 = yi + h f(xi, yi)
yi = yi-1 + h f(xi, yi)
can be clubbed together to get a second order scheme of the form
yi+1 = yi-1 + 2h f(xi, yi)
The only drawback of this scheme is at i = 0, the scheme uses
the value of the dependent variable at i = -1 which is not available.
Hence one has to use some other scheme to compute the value of the dependent
variable at first node (next to the initial point). However from second
node onwards the solution can be obtained with second order accuracy using
the mid point Nystrom method.
Local truncation error :
Ti+1 = y(xi+1) - yi+1
= y(xi+1)
- yi-1 + 2h f(xi, yi)
= h3/3 y'''(x)
where xi < x <
xi+1.
or |Ti+1| < h3/3 M
where M = |
max
|
|y'''(x)|
|
x Î (x0,
xn)
|