Max-stable processes extend the multi-dimensional extreme value theory to the infinite-dimensional case. Similar to the one- and multi-dimensional case, such a process occurs as a limit value of the maxima of appropriately normalized independent copies of a stochastic process.
Let be any index set. A stochastic process is called max-stable if there are normalization constants such that it holds
for independent copies of the process
.
The one-dimensional marginal distributions of a max-stable process are given by one of the three univariate extreme value distributions. In the case of Fréchet distributed margins d. H. the normalization constants can be chosen as follows: .
General
Be independent copies of the stochastic process . There are now normalization constants , so apply
for and and the process is not degenerate, it is a max-stable process. A max-stable process with simple Fréchet-distributed margins can be constructed using its spectral representation.
^ Maximilian Zott: Extreme Value Theory in Higher Dimensions - Max-Stable Processes and Multivariate Records . ( uni-wuerzburg.de [accessed October 7, 2019]).
↑ Laurens de Haan: A Spectral Representation for Max-stable Processes. In: The Annals of Probability. 1984.