ABSTRACT:
This chapter describes a variable metric minimizer, a program that finds the set of parameters to a function that will produce the smallest output value from that function.
The program uses methods now accepted as both robust and efficient. That is, these methods can handle a wide variety of functions and data without failing, and they can find a miniumum in a shorter time (fewer iterations) than other methods.
The methods are called quasi-Newton positive definite secant update methods. The variable metric part comes from looking at the global topology of the problem in addition to the local information given by the derivatives.
Complete C code is given. The user may choose an arbitrary function to minimize subject to his or her data.
A simplified explanation of the method is also included.
