A new cluster algorithm for orientation data

An algorithm to classify data points on the sphere in distinct cluster groups is defined. The characteristics of the cluster groups and the rule for assigning data to the groups are related to a continuous differentiable density estimation. The modes of the estimated density are assumed to be representative of the groups; data points are then assigned to the mode reached by the steepest ascent. The major advantage of this procedure is its sensitivity in detecting cluster groups independently of their geometry and configuration. As a consequence, the procedure is capable of handling orientation data that may be arranged in girdles.