9316435 Gilbert This research is the application of new modeling strategies to problems in global seismology which have proven difficult to solve using conventional techniques. The problems of concern are highly nonlinear and/or have multiple solution minima, which makes gradient techniques unusable and general search techniques prohibitively expensive. Developments based on analogies with optimization in naturally occurring systems are leading to practical algorithms which can efficiently search model space. This class of algorithms includes simulated annealing, genetic algorithms, and evolutionary programming. The algorithms will be applied to two specific problems: spherically symmetric earth models from datasets of free-oscillation measurements, and waveform modeling of 1-D and 3-D structure. ***