Central limit theorem

Statement and intuition

The most commonly cited version applies to independent and identically distributed random variables with finite variance. Let X1, X2, ..., Xn denote such variables with mean μ and variance σ². The distribution of the sample mean (X1 + ⋯ + Xn)/n becomes increasingly close to a normal distribution with mean μ and standard deviation σ/√n as n grows large. Likewise, the sum X1 + ⋯ + Xn tends toward a normal distribution with mean nμ and standard deviation √nσ. This result provides a way to approximate probabilities for sums and averages even when the original variables are not normal.

Intuitively, each independent variable contributes a small random shift; when many such shifts are combined their idiosyncratic features cancel out and only a smooth, universal pattern remains. The theorem formalizes that idea: after appropriate centering and scaling, the combined distribution approaches a fixed shape (the normal curve), a phenomenon sometimes described as universality.

Conditions and generalizations

The classical theorem assumes identical distributions and finite variance, but several generalizations relax those assumptions. Conditions named after Lyapunov and Lindeberg allow variables to have different distributions while preventing any single term from dominating the total. More advanced forms cover dependent variables under mixing assumptions, triangular arrays, or situations with infinite variance that lead to stable but non-Gaussian limits. The various technical conditions control moments, tail behavior, or contribution balance so the convergence remains valid. Theorems about limiting distributions and aggregated behavior are part of a broader theory of asymptotic approximation.

History and significance

The central limit phenomenon was recognized in the 18th and 19th centuries through work by de Moivre and Laplace, who studied approximations to binomial probabilities. The result was gradually generalized and rigorously stated in the 19th and early 20th centuries; later contributions refined the required conditions and rates of convergence. The theorem is central to why the normal distribution is often called the normal or Gaussian law: many unrelated random mechanisms yield the same limit shape. Its historical development connects to the emergence of statistical inference, error analysis, and the mathematical theory of probability.

Applications and examples

Because of the central limit theorem, statisticians and scientists routinely use normal-based approximations when computing confidence intervals, hypothesis tests, and prediction bounds for means and totals. Examples include measurement errors averaging out in experiments, aggregated economic indicators, sampling distributions in surveys, and the justification for using least-squares methods. In practice, moderate sample sizes are often sufficient for good approximations, though skewed or heavy-tailed data may require larger samples or alternative techniques. Aggregated distributions arising in engineering, natural sciences, and finance frequently rely on CLT-based reasoning.

Important distinctions and notable facts

• The CLT concerns convergence in distribution; it does not guarantee almost sure or uniform convergence of densities.
• Finite variance is sufficient for the classical Gaussian limit; if variance is infinite, the limit may be a different stable distribution.
• Different versions (Lindeberg, Lyapunov, and Liapounov conditions) give criteria for non-identical variables; these control how much any single term can influence the total. Independent random variables under such conditions behave similarly in the limit.
• Rates of convergence can be quantified (Berry–Esseen bounds), which matter when assessing how large "large" must be in applications. Variance and other moments often appear in those bounds.

For readers seeking further detail, standard references explain rigorous proofs, refinements, and applications to dependent data, random processes, and high-dimensional problems. Topics closely related to the CLT include the law of large numbers, laws of iterated logarithm, and the study of stable laws. Finite moment conditions, expected value centering, and control of mean and standard deviation are recurring themes in those treatments.