Canonical: https://knowlpedia.com/probability/cumulant-generating-function/

# Cumulant generating function

The logarithm of the moment generating function, when the latter is finite near zero.

A **cumulant generating function** is the real-valued [function](https://knowlpedia.com/shared-foundations/function/) \(K_X\) associated to a [random variable](https://knowlpedia.com/probability/random-variable/) \(X\) whose [moment generating function](https://knowlpedia.com/probability/moment-generating-function/) \(M_X(t)=\mathbb{E}[e^{tX}]\) is finite on an open interval containing \(0\), defined by
\[
K_X(t) \;=\; \log M_X(t),
\]

for all \(t\) where \(M_X(t)\) is finite.

## Remarks

Derivatives of \(K_X\) at \(0\) (when they exist) produce the [cumulants](https://knowlpedia.com/probability/cumulant/) of \(X\); in particular, this links \(K_X\) to [expectation](https://knowlpedia.com/probability/expectation/) and [variance](https://knowlpedia.com/probability/variance/). If \(X\) and \(Y\) are [independent random variables](https://knowlpedia.com/probability/independence-random-variables/) and both cumulant generating functions exist near \(0\), then \(K_{X+Y}(t)=K_X(t)+K_Y(t)\).

## Examples

- If \(X\sim \mathcal{N}(\mu,\sigma^2)\), then \(K_X(t)=\mu t+\tfrac12\sigma^2 t^2\) (finite for all \(t\in\mathbb{R}\)).
- If \(X\sim \mathrm{Bernoulli}(p)\), then \(K_X(t)=\log\!\bigl((1-p)+p\,e^{t}\bigr)\).
