A parallel searching and sorting method to compute entropies from time series

A parallel method is presented for computing the generalized entropies K_q from a time series with a direct approach, i.e. by counting the occurrences of the various box sequences in the phase-space. The parallelization is accomplished by distributing the sequence table among processors, and efficient searching and sorting procedures are employed. A model of the algorithm is used to evaluate computational performance under different operative settings. The algorithm is tested on a distributed memory multiprocessor, the Transtech PARAstation/i860 with eigth nodes; it shows good performance, allowing fast analysis of very long time series.