For the shooting method (ref)
we consider a set of N
differential equations
for N variables.
We set n
initial conditions at t1
:
This means n
of the variables will have fixed initial values. This in turn means
that m = N-n
of the variables have unspecified inital values. We will call these
the free variables.
If we have m =
N-n final conditions at t2
:
then the goal is to vary the initial values of the m
free variables until we get the variables to match the boundary
conditions at t2.
For each set of these initial values, we integrate the variables
from t1
to t2.
(This is the shooting step.) The difference between the final
values and the desired boundary values is set up as a function to
be minimized by a root-finding
method, such as discrete
Newton method.
The process begins by setting the n
variables to their fixed values and guessing at values for the m
free variables. Lets label these initial values for the free variables
as
We then use an ODE method such as Euler or Runge-Kutta to integrate
the variables over the interval t1
to t2.
The boundary conditions demand that m
of the variables match specific values at t2
:
for
m of the variables
We create a function proportional to the total distance of the
final values from the boundary values as in
The root-finding method varys the
Z1(t1),..Zm(t1)
values until F equals zero.
In summary, the algorithm makes a guess at the initial values of
the free variables and integrates over
t1 to t2
interval, that is, it shoots towards the target. Based on
the distance of the values from the target values, the algorithm
modifies the initial values of the independent variables and tries
the integration again. This interative procedure eventually results
in a close hit on the final value. If the differential equations
are linear, then this method will require only one iteration.
Another technique is to shoot from both the initial and and final
boundaries and try to meet at an intermediate point (see ref).
Here we look only at the first technique as discussed in Demo
3.
|