Peak finding through Scan Statistics

Abstract

We discuss the conditions under which Scan Statistics can be fruitfully implemented to signal a departure from the underlying probability model that describes the experimental data. It is shown that local perturbations (``bumps'' or ``excesses'' of events) are better dealt within this framework and, in general, tests based on these statistics provide a powerful and unbiased alternative to the traditional techniques related with the chi2 and Kolmogorov distributions. Approximate formulas for the computation of Scan Statistics in the range of interest for high energy and nuclear physics applications are also presented.