While simple to understand, Euler's method
for integrating ordinary differential equations can have a great deal
of numerical error.
Euler's method can be expressed simply as
New = Old + Change
where the change is determined based on known information at the
old time. However, if the rate at which things are changing is also
changing over time, this can result in an error. Consider a dropped
ball at constant acceleration. The initial velocity of the ball is zero.
Clearly if the "change" in the balls position, or the velocity, is assumed
to be zero for the entire time step, then the ball will not move
at all during the time step. Suppose instead we could use the average
of the speed of the ball at the beginning and end of the timestep. This
would reduce our error.
The improved Euler method (also called the midpoint Euler method
or second order Runge Kutta method) attempts to do just this.
Consider the equation dx/dt = f(x).
What we want is xnew = xold + dt*0.5*
(dx/dt|old + dx/dt|new). The problem is that
until we know what xnew is, we cannot do this.
So what we do is the following, we shoot forward using dx/dt|old,
and use that to create an interim value for xnew.