Exploratory analysis of particle data using random cells (bins) with fixed contents of observed events

We present a simple algorithm to group events observed (rather than predicted) in a bounded region of the multidimensional phase space into bins of fixed and nearly equal content. Thereby the randomness is completely transferred from the observed to the predicted bin contents: with a simple null hypothesis the predicted content of any single bin is random and Erlang distributed. We advocate a particular chi-squared quantity Gs2 for testing the goodness of fits to data binned in this way. The size of the bins obtained with the algorithm is inversely proportional to the smoothed local event density. This fact facilitates the pattern recognition in multidimensional distributions via contrast enhancement of their one- or two-dimensional projections.