Fisher-Tippett-Gnedenko Theorem

This article needs to be linked to other articles. In particular: throughout, including category You can help $\mathsf{Pr} \infty \mathsf{fWiki}$ by adding these links. To discuss this page in more detail, feel free to use the talk page. When this work has been completed, you may remove this instance of {{MissingLinks}} from the code.

This page has been identified as a candidate for refactoring of advanced complexity. In particular: into separate pages with well-defined theorem and definitions as relevant Until this has been finished, please leave {{Refactor}} in the code. New contributors: Refactoring is a task which is expected to be undertaken by experienced editors only. Because of the underlying complexity of the work needed, it is recommended that you do not embark on a refactoring task until you have become familiar with the structural nature of pages of $\mathsf{Pr} \infty \mathsf{fWiki}$. To discuss this page in more detail, feel free to use the talk page. When this work has been completed, you may remove this instance of {{Refactor}} from the code.

Theorem

In general, for iid $\set {X_i} _{i\mathop = 1}^n$ with cdf $\map F x$ we have $\mathbb P \sqbrk {\max_{i \mathop = 1}^n \set {X_i} > x} = \map {F^n} x$.

Let $x_* := \sup \set {x: \map F x < 1}$.

Then as $n \to \infty$:

$\ds \max_{i \mathop = 1, \mathop \ldots, n} \set {X_i} \stackrel {prob} {\to} x_*$

We normalize it by constants $a_n > 0, b_n$ such that:

$\dfrac {\ds \max_{i \mathop = 1, \mathop \ldots, n} \set {X_i} - b_n} {a_n} \stackrel {dis} {\to} X_*$

for some non-degenerate $X_*$ with distribution $\map G x$.

We have the following three possible limits

• Frechet: If $x > 0$, $\map G x = \map \exp {-x^{-a} }$ for some $a > 0$. If $x \le 0$, $\map G x = 0$.

• Gumbel: If $x \in \R$, $\map G x = \map \exp {-e^{-\frac {x - \mu} {\sigma} } }$ for $\mu\in \R, \sigma>0$.

• Weibull: If $x \ge 0$, $\map G x = 1 - \map \exp {-\paren {\dfrac x \lambda}^k}$ for $\lambda, k > 0$. If $x < 0$ , $\map G x = 0$.

Fisher-Tippet-Gnedenko theorem

$(1): \quad \map G x$ will be Frechet if and only if $\map F x < 1$ and $\ds \lim_{t \mathop \to +\infty} \dfrac {1 - \map F {t x} } {1 - \map F t} = x^{-\theta}$ for $x > 0$ and $\theta > 0$. Here we can set $b_n := 0$ and $a_n := F^{-1} \paren {1 - \dfrac 1 n}$.

$(2): \quad \map G x$ will be Weibull if and only if $x_* : = \sup \set {x: \map F x < 1} < \infty$ and $\ds \lim_{t \mathop \to +\infty} \dfrac {1 - \map F {x_* + t x} } {1 - \map F {x_* - t} } = \paren {-x}^{-\theta}$ for $x < 0$ and $\theta > 0$. Here we can set $b_n := x_*$ and $a_n := x_* - \map {F^{-1} } {1 - \dfrac 1 n}$.

$(3): \quad G$ will be Gumbel if the density $\map f x := \dfrac \d {\d x} \map F x > 0$ and it is differentiable in $\openint {x_1} {x_*}$ for some $x_1$, and $\ds \lim_{x \mathop \to x_*} \map {\dfrac \d {\d x} } {\dfrac {1 - \map F x} {\map f x} } = 0$. Here we can set $b_n := \map {F^{-1} } {1 - \dfrac 1 n}$ and $a_n := \dfrac 1 {n \map f {b_n} }$.

Proof

Proof under construction

Source of Name

This entry was named for Ronald Aylmer Fisher, Leonard Henry Caleb Tippett and Boris Vladimirovich Gnedenko.