The function K(t)=log E[e^{tX}] defined (when the moment‑generating function exists) whose derivatives at zero yield the cumulants of a random variable X; it encodes additive properties for independent sums.

Principle

Because the log of the moment‑generating function converts products of MGFs (for independent variables) into sums, derivatives of K at zero produce central measures (cumulants) that add under independence.

Demonstration

For X~N(μ,σ²), K(t)=μ t + (σ²/2) t²; the first two cumulants extracted by K'(0)=μ and K''(0)=σ² correspond to mean and variance, and higher cumulants vanish.

Misapplication

Using cumulant generating functions when the moment‑generating function does not exist (heavy‑tailed distributions) or misinterpreting cumulants as raw moments leads to errors; existence is local around t=0.

Consequence

Provides a compact tool for obtaining cumulants, studying convergence (via cumulant control), and simplifying analysis of sums of independent variables because cumulants add.

Reversal

Exponentiating K(t) returns the moment‑generating function M(t)=E[e^{tX}]; when M is known, cumulants follow from its logarithm rather than from raw moment derivatives.

Boundary

Defined only on the interval or neighborhood of t=0 where the moment‑generating function M(t) exists and is finite; characteristic functions (log of E[e^{itX}]) always exist but yield complex cumulant-like coefficients.

Semantic Tension

Often contrasted with moment generating and characteristic functions: cumulant generating function gives additive cumulants via logarithm of MGF, while moments come directly from MGF derivatives and characteristic functions avoid existence issues.

Synthesis

The cumulant generating function is the logarithm of the moment‑generating function whose derivatives at zero produce cumulants that summarize distribution shape and add for independent sums.