Moment Generating Function¶
Formula¶
\[
M_X(t) = \mathbb{E}[e^{tX}]
\]
Parameters¶
- \(t\): real parameter
- \(X\): random variable
What it means¶
Encodes all moments of \(X\) (when it exists) via derivatives at \(t=0\).
What it's used for¶
- Computing moments and sums of independent variables.
- Proving distributional convergence.
Key properties¶
- \(M_X^{(k)}(0)=\mathbb{E}[X^k]\)
- If it exists in a neighborhood of 0, it uniquely determines the distribution
Common gotchas¶
- MGF may not exist for heavy-tailed distributions.
- Use characteristic function when MGF diverges.
Example¶
For \(X\sim\mathrm{Bernoulli}(p)\), \(M_X(t)=(1-p)+p e^t\).
How to Compute (Pseudocode)¶
Input: quantities required by the card formula (distribution parameters, samples, or test setup)
Output: card-specific statistic/probability/result
compute any required summary quantities from data or model parameters
apply the card formula or workflow
return the resulting value(s)
Complexity¶
- Time: Depends on whether the card is applied analytically, numerically, or from sample data (common sample-statistic workflows are often linear in sample size)
- Space: Depends on whether summaries are streamed or full samples/tables are materialized
- Assumptions: Exact runtime/memory is method-dependent and driven by the chosen estimator/test/distribution representation