1 Definition

1.1 Formal definition

For a real-valued random variable \(X\), the moment generating function (MGF) is defined as \[ M_X(t)=\mathbb{E}\big[e^{tX}\big], \] for those values of \(t\) where the expectation exists and is finite.

1.2 Domain of existence

The set of \(t\in\mathbb{R}\) for which \(M_X(t)\) is finite forms the MGF’s domain. This domain is generally an open interval containing \(0\) (or, in some cases, a one-sided interval), because \(M_X(0)=\mathbb{E}[1]=1\) whenever the random variable is well-defined. If the tails of \(X\) are heavy, the expectation \(\mathbb{E}[e^{tX}]\) may diverge for some nonzero \(t\), limiting how far the MGF can be extended.

1.3 Relationship to the exponential function

The exponential function serves as a generating mechanism: expanding \(e^{tX}\) into its power series produces terms involving \(X^n\). When the relevant expectations exist, the MGF encodes the moments through derivatives at \(t=0\). Conceptually, the MGF acts like a transform that weights outcomes of \(X\) exponentially rather than linearly.

2 Basic properties

2.1 Normalization at zero

Because \(e^{0\cdot X}=1\), \[ M_X(0)=\mathbb{E}[1]=1, \] which provides a starting point for extracting information via derivatives.

2.2 Differentiation and moments

2.2.1 First moment

If \(M_X(t)\) is finite in a neighborhood of \(0\) and differentiable there, then \[ M_X'(0)=\mathbb{E}[X]. \] Thus the mean of \(X\) is obtained from the slope of the MGF at the origin.

2.2.2 Higher-order moments

Under analogous regularity conditions that justify differentiating under the expectation, \[ M_X^{(n)}(0)=\mathbb{E}[X^n] \] for integers \(n\ge 1\). Consequently, the MGF generates the entire moment sequence up to the highest order for which the corresponding derivatives exist at \(0\).

2.3 Linear transformations

For constants \(a\) and \(b\), let \(Y=aX+b\). Then, when the expectations exist, \[ M_Y(t)=\mathbb{E}\big[e^{t(aX+b)}\big]=e^{bt}\,\mathbb{E}\big[e^{(at)X}\big]=e^{bt}M_X(at). \] This property allows moment calculations for affine changes of variables.

2.4 Sums of independent random variables

If \(X\) and \(Y\) are independent, then for \(S=X+Y\), \[ M_S(t)=\mathbb{E}\big[e^{t(X+Y)}\big]=\mathbb{E}\big[e^{tX}e^{tY}\big] =\mathbb{E}\big[e^{tX}\big]\mathbb{E}\big[e^{tY}\big] = M_X(t)M_Y(t), \] again whenever both MGFs exist at the relevant \(t\). This multiplicative behavior is a key reason MGFs are convenient for handling sums.

3 Existence and uniqueness

3.1 Conditions for existence

MGFs require finite exponential moments. A sufficient condition is that there exists \(\epsilon>0\) such that \(\mathbb{E}[e^{tX}]<\infty\) for all \(t\in(-\epsilon,\epsilon)\). For many common light-tailed distributions (e.g., normal, many gamma-family cases), this neighborhood exists. For heavy-tailed distributions, the MGF may fail to exist except at \(t=0\), or only exist on one side of zero.

3.2 Distribution determination

Under standard conditions, the MGF uniquely determines the distribution: if two random variables have MGFs that agree in an interval around \(0\), then their distributions are the same. The intuition is that agreement of MGFs implies agreement of all moments within the interval, and under the usual analytic and integrability assumptions the transform pins down the underlying measure.

3.3 Comparison with characteristic functions

The characteristic function \(\phi_X(t)=\mathbb{E}[e^{itX}]\) always exists for real-valued \(X\) because \(e^{itX}=1\). In contrast, the MGF may diverge. Both transforms can be used to identify distributions; the characteristic function is therefore often preferred when the MGF does not exist globally, while the MGF offers simpler moment extraction through real derivatives at the origin.

4 Common examples

4.1 Discrete distributions

4.1.1 Bernoulli distribution

If \(X\sim\text{Bernoulli}(p)\) with \(\mathbb{P}(X=1)=p\) and \(\mathbb{P}(X=0)=1-p\), then \[ M_X(t)=(1-p)e^{t\cdot 0}+p e^{t\cdot 1}=(1-p)+pe^{t}. \] Differentiating at \(0\) yields \(\mathbb{E}[X]=p\).

4.1.2 Binomial distribution

For \(X\sim\text{Binomial}(n,p)\), \[ M_X(t)=\mathbb{E}[e^{tX}]=\sum_{k=0}^n e^{tk}\binom{n}{k}p^k(1-p)^{n-k}. \] Using the binomial theorem gives \[ M_X(t)=\big((1-p)+pe^{t}\big)^n. \]

4.1.3 Poisson distribution

If \(X\sim\text{Poisson}(\lambda)\), then \[ M_X(t)=\sum_{k=0}^\infty e^{tk} e^{-\lambda}\frac{\lambda^k}{k!} = e^{-\lambda}\sum_{k=0}^\infty \frac{(\lambda e^{t})^k}{k!} = e^{-\lambda} e^{\lambda e^{t}} = \exp\big(\lambda(e^{t}-1)\big). \]

4.2 Continuous distributions

4.2.1 Exponential distribution

Let \(X\sim\text{Exponential}(\lambda)\) with density \(\lambda e^{-\lambda x}\) for \(x\ge 0\). Then for \(t<\lambda\), \[ M_X(t)=\int_0^\infty e^{tx}\lambda e^{-\lambda x}\,dx =\lambda\int_0^\infty e^{-(\lambda-t)x}\,dx =\frac{\lambda}{\lambda-t}. \] The restriction \(t<\lambda\) reflects divergence when \(t\ge \lambda\).

4.2.2 Normal distribution

For \(X\sim \mathcal{N}(\mu,\sigma^2)\), \[ M_X(t)=\exp\left(\mu t+\frac{1}{2}\sigma^2 t^2\right). \] This closed form is central in many theoretical results, since normal sums and affine transforms behave neatly under this transform.

4.2.3 Gamma distribution

If \(X\sim \text{Gamma}(\alpha,\beta)\) with shape \(\alpha&gt;0\) and rate \(\beta&gt;0\) (density proportional to \(x^{\alpha-1}e^{-\beta x}\) on \(x\ge 0\)), then for \(t&lt;\beta\), \[ M_X(t)=\left(\frac{\beta}{\beta-t}\right)^{\alpha}. \] As with the exponential case, the finite exponential moment depends on staying below the rate parameter.

5 Applications

5.1 Calculating moments

The MGF provides a systematic method to compute moments: \[ \mathbb{E}[X^n]=M_X^{(n)}(0), \] whenever the derivatives exist. This approach is often faster than integrating \(x^n\) directly, especially for distributions with known transform forms.

5.2 Proving limit theorems

MGFs are widely used in proofs of limit theorems for sequences of random variables, such as the central limit theorem and variants. A common strategy is to compute the MGF of normalized sums and show convergence to the MGF of a limiting distribution. This requires conditions ensuring that the MGFs exist in a neighborhood of \(0\) and converge there.

5.3 Identifying distribution families

Because many distribution families have distinctive MGFs, one can sometimes identify a distribution by matching a proposed form of \(M_X(t)\) with a known template. This is particularly effective when the MGF is available in closed form and when uniqueness holds within the relevant interval around \(0\).

5.4 Analyzing sums and convolutions

For independent random variables, the product rule \[ M_{X+Y}(t)=M_X(t)M_Y(t) \] simplifies the analysis of sums, which correspond to convolutions at the level of densities or probability mass functions. By transforming the problem into multiplication in the MGF domain, one can obtain moments of sums without performing convolution integrals.

6.1 Cumulant generating function

The cumulant generating function (CGF) is defined as \[ K_X(t)=\log M_X(t), \] when \(M_X(t)\) is positive and finite. Derivatives of \(K_X\) at \(0\) yield cumulants, which include the mean and variance but also provide more detailed measures such as skewness- and kurtosis-related quantities.

6.2 Characteristic function

The characteristic function is \[ \phi_X(t)=\mathbb{E}[e^{itX}], \] and is analogous to the MGF but with \(t\) replaced by \(it\). While the MGF may fail to exist, the characteristic function always exists for real \(t\) and remains a standard tool for uniqueness and convergence in distribution.

6.3 Probability generating function

For nonnegative integer-valued random variables \(N\), the probability generating function (PGF) is \[ G_N(s)=\mathbb{E}[s^N]. \] Like the MGF, it encodes probabilities through an analytic transform. Under suitable conditions, derivatives at \(s=1\) produce factorial moments.

6.4 Laplace transform

For a nonnegative random variable \(X\), the Laplace transform is \[ \mathcal{L}_X(s)=\mathbb{E}[e^{-sX}], \quad s\ge 0. \] It is closely related to the MGF via a change of variable \(s=-t\). The Laplace transform is especially common in applied settings involving waiting times and differential equations.

7 Extensions and generalizations

7.1 Multivariate moment generating functions

For a random vector \(X=(X_1,\dots,X_d)\), the multivariate MGF is \[ M_X(t)=\mathbb{E}\big[\exp(t^\top X)\big], \quad t\in\mathbb{R}^d, \] where \(t^\top X=\sum_{i=1}^d t_i X_i\). Mixed partial derivatives at \(t=0\) recover mixed moments such as \(\mathbb{E}[X_i X_j]\).

7.2 Conditional moment generating functions

For events or sigma-fields \(\mathcal{G}\), the conditional MGF can be defined when it exists: \[ M_{X\mid \mathcal{G}}(t)=\mathbb{E}\big[e^{tX}\mid \mathcal{G}\big]. \] It is useful for studying randomness that evolves over time or for hierarchical models, enabling moment calculations under conditioning.

7.3 Matrix-valued and functional extensions

In more advanced settings, the role of \(X\) is played by objects such as random matrices or random functions. A matrix-valued transform may involve expectations of matrix exponentials (with appropriate definitions to preserve meaning), while functional analogues replace \(X\) by a random element in a function space. These extensions aim to preserve the generating-transform behavior, though existence and technical conditions can be substantially more intricate than in the scalar case.