Le séminaire a lieu le mardi 14 mars 2006 de
11h00
à
12h00
dans la salle de réunion
de la Nef (Observatoire de Nice).
Often in nature the temporal distribution of inhomogeneous stochastic
point processes can be modelled as a realization of renewal Poisson processes
with a variable rate. Here one of the classical examples is investigated,
namely the temporal distribution of polarity reversals of the geomagnetic
field. In spite of the commonly used underlying hypothesis, it is
shown that this process strongly depart from a Poisson statistics, including
variable rate Poisson statistics. The origin of this failure can be attributed
to the presence of temporal clustering. A Lévy statistics is able
to re-produce paleomagnetic data, thus suggesting the presence of long-range
correlations in the underlying dynamo process. Our results introduce a
constraint for geodynamo modelling, and the statistical tools used here
can be viewed as a benchmark for further applications to various stochastic point processes.
Finally, the features of some dynamo models in this perspective are discussed.