Stellar population synthesis with more degrees of freedom than observables

In order to derive the stellar population of a galaxy or a star cluster, it is a common practice to fit its spectrum by a combination of spectra extracted from a data base (e.g. a library of stellar spectra). If the data to be fitted are equivalent widths, the combination is a non-linear one and the problem of finding the 'best' combination of stars that fits the data becomes complex. It is probably because of this complexity that the mathematical aspects of the problem did not receive a satisfying treatment; the question of the uniqueness of the solution, for example, was left in uncertainty. In this paper we complete the solution of the problem by considering the underdetermined case where there are fewer equivalent widths to fit than stars in the data base (the overdetermined case was treated previously). The underdetermined case is interesting to consider because it leaves space for the addition of supplementary astrophysical constraints. In fact, it is shown in this paper that when a solution exists it is generally not unique. There are infinitely many solutions, all of them contained within a convex polyhedron in the solutions vector space. The vertices of this polyhedron are extremal solutions of the stellar population synthesis. If no exact solution exists, an approximate solution can be found using the method described for the overdetermined case. Also provided is an algorithm able to solve the problem numerically; in particular all the vertices of the polyhedron are found.