Extremum statistics: A framework for data analysis

Abstract

Recent work has suggested that in highly correlated systems, such as sandpiles, turbulent fluids, ignited trees in forest fires and magnetization in a ferromagnet close to a critical point, the probability distribution of a global quantity (i.e. total energy dissipation, magnetization and so forth) that has been normalized to the first two moments follows a specific non Gaussian curve. This curve follows a form suggested by extremum statistics, which is specified by a single parameter a (a=1 corresponds to the Fisher-Tippett Type I ("Gumbel") distribution.) Here, we present a framework for testing for extremal statistics in a global observable. In any given system, we wish to obtain a in order to distinguish between the different Fisher-Tippett asymptotes, and to compare with the above work. The normalizations of the extremal curves are obtained as a function of a. We find that for realistic ranges of data, the various extremal distributions when normalized to the first two moments are difficult to distinguish. In addition, the convergence to the limiting extremal distributions for finite datasets is both slow and varies with the asymptote. However, when the third moment is expressed as a function of a this is found to be a more sensitive method.