27.6 DIFFERENTIAL EQUATIONS
x h y(exact)
0. 01 0.1 0.5 1.0 1.5 2 3
0 (1) (1) (1) (1) (1) (1) (1) (1)
0.5 0.605 0.590 0.500 0 − 0. 500 − 1 − 2 0.607
1.0 0.366 0.349 0.250 0 0. 250 1 4 0.368
1.5 0.221 0.206 0.125 0 − 0. 125 − 1 − 8 0.223
2.0 0.134 0.122 0.063 0 0. 063 1 16 0.135
2.5 0.081 0.072 0.032 0 − 0. 032 − 1 − 32 0.082
3.0 0.049 0.042 0.016 0 0. 016 1 64 0.050
Table 27.10 The solutionyof differential equation (27.61) using the Euler
forward difference method for various values ofh. The exact solution is also
shown.
27.6.1 Difference equations
Consider the differential equation
dy
dx
=−y, y(0) = 1, (27.61)
and the possibility of solving it numerically by approximatingdy/dxby a finite
difference along the lines indicated in section 27.5. We start with the forward
difference
(
dy
dx
)
xi
≈
yi+1−yi
h
, (27.62)
where we use the notation of section 27.5 but withfreplaced byy.Inthis
particular case, it leads to the recurrence relation
yi+1=yi+h
(
dy
dx
)
i
=yi−hyi=(1−h)yi. (27.63)
Thus, sincey 0 =y(0) = 1 is given,y 1 =y(0 +h)=y(h) can be calculated, and
so on (this is theEulermethod). Table 27.10 shows the values ofy(x) obtained
if this is done using various values ofhand for selected values ofx. The exact
solution,y(x)=exp(−x), is also shown.
It is clear that to maintain anything like a reasonable accuracy only very small
steps,h, can be used. Indeed, ifhis taken to be too large, not only is the accuracy
bad but, as can be seen, forh>1 the calculated solution oscillates (when it
should be monotonic), and forh>2 it diverges. Equation (27.63) is of the form
yi+1=λyi, and a necessary condition for non-divergence is|λ|<1, i.e. 0<h<2,
though in no way does this ensure accuracy.
Part of this difficulty arises from the poor approximation (27.62); its right-
hand side is a closer approximation tody/dxevaluated atx=xi+h/2thanto
dy/dxatx=xi. This is the result of using a forward difference rather than the