Emergence of Correlations Between Records and Maximum in Random Time Series
In the authors' words
We compute exactly the covariance between the maximum and the number of records in a discrete time series of entries as a quantitative probe of the history leading to an extreme event. For continuous IID series, with a common probability distribution function (PDF) , we find the striking result that and are uncorrelated for any . We show that this independence is rather fragile and the covariance between and becomes nonzero when becomes discontinuous due to a delta-component at with an amplitude . This time series models the occurrence of dry days with probability in a precipitation time series when there is no rainfall. We show that there is an interesting shape transition at a critical value where the covariance , as a function of , changes from monotonic () to non monotonic behavior (for ). We compute the critical strength exactly. Furthermore, for large , we show that decays as a power law, with an exponent that depends on the tail of . We then compare our theoretical predictions with real precipitation data from several weather stations across Germany, finding a very good agreement.
Appeared: Monday, October 19. arXiv. Preprint, not yet peer-reviewed.
Authors' comment: Letter: 5+2 pages and 6 figures; Supp. Mat.: 15 pages and 5 figures