EXPLANATION FILE OF PROGRAM ZMUELLER
====================================
Mueller's Method in the Complex Plane
-------------------------------------
Mueller's one-dimensional parabolic method (see file Mueller.txt) can be
applied directly to complex plane calculations by replacing the real variables
and functions contained in the subroutine by their complex plane counterparts.
The difficulty in doing this resides in the large number of complex plane al-
gebraic manipulations which must be performed.
This problem can be approached in two ways. The first way would be to per-
form aIl the complex plane calculations by calling complex numbers subrouti-
nes. However, the resulting program would not be efficient in the total length
of code, and it would also execute slowly. The second approach is to algebra-
ically rewrite the basic procedure so that the complex plane calculations are
broken down into a series of real-number operations. The result is some com-
plicated algebra, but a reasonably short and fast subroutine. The latter ap-
proach is preferable.
Program Zmueller is a real-number, algebraic implementation of Mueller's me-
thod in the complex plane. The subroutine requires as input an initial guess,
(xO, yO); two initial bounds (BI for xO; B2 for yO); a convergence criterion,
E; and a limit on the number of iterations, N. The subroutine returns the new
estimate, (x, y), and the number of iterations actually performed, K.
We can use this program to examine the test function f(z) = z^2 + 1. The
convergence rate is very impressive. It appears that only two iterations are
required no matter what the initial guess isl The reason for this high conver-
gence rate is simple: ZMUELLER, in effect, fits a complex-plane parabola to
the test funcion. However, f(z) is only of the second degree, and in princi-
ple, the fitted parabola coincident with f(z) throughout the complex plane.
Therefore, the roots are located on the first pass (again, in principle).
Thus, we would expect only one iteration to be necessary, if it were not for
round-off error. The second iteration is, in essence, a cleanup operation.
f(z) = z^2 + 1 was an ideal function for treatment with ZMUELLER. Functions
that do not have more than two roots close together are usually very amenable
to solution by this algorithm. The high curvature associated with functions
like f(z) = (z + 1)^5 can slow down convergence, as we saw in other sections,
but the root is eventually found. ZMUELLER is a fairly reliable subroutine.
The utility of ZMUELLER is not strictly limited to analytic functions of z.
This routine can also be employed to search for the roots of other functions
of x and y simply by using mu(x,y) to represent the function (in the "Func-
tions" subroutine), and setting nu(x,y) = 0. A difficult example of this (dif-
ficult because of the high curvature in the functional form) is demonstrated.
Again the fifth-degree function, (x + 1)^5, is probed and a root found.
This particular example was chosen in part to show that the convergence pro-
perties of ZMUELLER for one- and two-dimensional problems are similar to those
of the previous two Mueller method subroutines (see file Mueller.txt). Also,
this example is actually one-dimensional: f(x,y) = f(x). However, the routine
followed a path through the (x,y) plane. The calculated values of y have no
significance in this case. ~
In conclusion, ZMUELLER is a very versatile algorithm that can be used for
finding the roots of one- and two-dimensional functions in the complex plane.
The subroutine can also be extended to the solution of simulianeous two-dimen-
sional functions. Simply represent one function with mu(x,y), and the other
with nu(x,y).
[Note that it is not required that mu(x,y) + i nu(x,y) be an analytic function
of z: the Cauchy-Riemann equations need not be satisfied.) This is left as an
exercise for you.
From [BIBLI 01].
----------------------------------------------
End of file Zmueller.txt