Canonical: https://knowlpedia.com/large-deviations/cramers-theorem/

# Cramér's theorem

Large deviations for empirical means of independent identically distributed real random variables.

**Cramér's theorem:** Let \((X_i)_{i\ge 1}\) be an [i.i.d. sequence](https://knowlpedia.com/probability/iid-sequence/) of real-valued [random variables](https://knowlpedia.com/probability/random-variable/). Assume the [moment generating function](https://knowlpedia.com/probability/moment-generating-function/) \(M(\theta)=\mathbb{E}[e^{\theta X_1}]\) is finite for all \(\theta\) in some open interval containing \(0\), and let \(\Lambda(\theta)=\log M(\theta)\) be the [log moment generating function](https://knowlpedia.com/large-deviations/log-moment-generating-function/). Define the empirical mean \(\overline X_n=\frac{1}{n}\sum_{i=1}^n X_i\). Then \((\overline X_n)\) satisfies a [large deviation principle](https://knowlpedia.com/large-deviations/large-deviation-principle/) on \(\mathbb{R}\) with speed \(n\) and [good rate function](https://knowlpedia.com/large-deviations/good-rate-function/)
\[
I(x)=\sup_{\theta\in\mathbb{R}}\bigl\{\theta x-\Lambda(\theta)\bigr\}.
\]

The rate function \(I\) is the [Cramér transform](https://knowlpedia.com/large-deviations/cramer-transform/), i.e. the [Legendre–Fenchel transform](https://knowlpedia.com/convex-analysis/legendre-fenchel-transform/) ([Fenchel conjugate](https://knowlpedia.com/convex-analysis/convex-conjugate-fenchel/)) of \(\Lambda\).
