A View of Unconstrained Minimization Algorithms that Do Not Require Derivatives

Existing algorithms are examined, with particular attention given to their merits and defects, in order to identify the techniques which may give methods for minimizahon without derivatives that are better than the methods we have at present."Quasi-Newton" and "conjugate direction" algorithms are studmd, and it is noted that both these classes of methods have fundamental disadvantages To overcome these disadvantages another class of algorithms suggests itself, called "B-conjugate" methods.Some comments are made on the construction of good algorithms within this class.