Consider the problem of finding a zero :9I of a smooth function f: :9I. Such problems are commonly solved by some variant of Newton’s method, combined with a continuation procedure if a sufficiently accurate initial approximation to :9I is not available. In a continuation procedure, the original root-finding problem is embedded in a family of similar problems through a smooth homotopy h: :9I. Under appropriate conditions, there is a smooth path :9I. The idea is to follow this path from tO to t1; there are a variety of ways of accomplishing this (see [1]). The homotopy parameter t may arise naturally in the context of the original problem so that the entire path is of interest, or may be introduced artificially; for example, by defining) h(x,t) = f(x) + (t-1)f(xO: 0E), for O :9I.
Among the aforementioned conditions is the assertion that :9I is nonsingular for tO:9I. If we reparameterize in terms of arclength :9I, the analogous condition for the existence of a smooth path is that
:9I have maximal rank, namely n. A point :9I is singular bu :9I has rank n is called a turning point; a point :9I at which :9I is rank deficient is called a bifurcation point. A root t¯ for which f’(x:A) is singular is called a singular root. In all three cases, if the rank in question is n-A-1 and certain conditions involving second derivatives are satisfied, the turning point, bifurcation point, or singular root is termed simple.
In the aforementioned root-finding context, a turning point encountered during the path-following process is merely a nuisance; it reflects the fact that t does not increase monotonically from tO to t1 along the path. In other contexts, the identification of turning points is an end in itself. Algorithms have been developed for this purpose which recast a simple turning-point problem as an enlarged nonsingular root-finding problem; for a review, see [2]. This paper is concerned with using these algorithms to solve simple singular root-finding problems, and thence simple bifurcation problems. By introducing a homotopy analogous to the artificial one noted above, the simple singular root-finding problem is recast as a simple turning-point problem. When recast as an enlarged root-finding problem, a simple bifurcation point becomes a simple singular root.
The presentation is uneven and somewhat obscure at times; it assumes familiarity with [3], and preferably also [2]. Theoretical justification for the application to these two problems, along the lines sketched above, of the method developed in [3] is presented. Alternate approaches are discussed briefly, and two examples are provided.