Categories: Mathematics, Statistics.

Central limit theorem

In statistics, the central limit theorem states that the sum of many independent variables tends towards a normal distribution, even if the individual variables \(x_n\) follow different distributions.

For example, by taking \(M\) samples of size \(N\) from a population, and calculating \(M\) averages \(\mu_m\) (which involves summing over \(N\)), the resulting means \(\mu_m\) are normally distributed across the \(M\) samples if \(N\) is sufficiently large.

More formally, for \(N\) independent variables \(x_n\) with probability distributions \(p(x_n)\), the central limit theorem states the following, where we define the sum \(S\):

\[\begin{aligned} S = \sum_{n = 1}^N x_n \qquad \mu_S = \sum_{n = 1}^N \mu_n \qquad \sigma_S^2 = \sum_{n = 1}^N \sigma_n^2 \end{aligned}\]

And crucially, it states that the probability distribution \(p_N(S)\) of \(S\) for \(N\) variables will become a normal distribution when \(N\) goes to infinity:

\[\begin{aligned} \boxed{ \lim_{N \to \infty} \!\big(p_N(S)\big) = \frac{1}{\sigma_S \sqrt{2 \pi}} \: \exp\!\Big( -\frac{(\mu_S - S)^2}{2 \sigma_S^2} \Big) } \end{aligned}\]

We prove this below, but first we need to introduce some tools. Given a probability density \(p(x)\), its Fourier transform is called the characteristic function \(\phi(k)\):

\[\begin{aligned} \phi(k) = \int_{-\infty}^\infty p(x) \exp(i k x) \dd{x} \end{aligned}\]

Note that \(\phi(k)\) can be interpreted as the average of \(\exp(i k x)\). We take its Taylor expansion in two separate ways, where an overline denotes the mean:

\[\begin{aligned} \phi(k) = \sum_{n = 0}^\infty \frac{k^n}{n!} \: \phi^{(n)}(0) \qquad \phi(k) = \overline{\exp(i k x)} = \sum_{n = 0}^\infty \frac{(ik)^n}{n!} \overline{x^n} \end{aligned}\]

By comparing the coefficients of these two power series, we get a useful relation:

\[\begin{aligned} \phi^{(n)}(0) = i^n \: \overline{x^n} \end{aligned}\]

Next, the cumulants \(C^{(n)}\) are defined from the Taylor expansion of \(\ln\!\big(\phi(k)\big)\):

\[\begin{aligned} \ln\!\big( \phi(k) \big) = \sum_{n = 1}^\infty \frac{(ik)^n}{n!} C^{(n)} \quad \mathrm{where} \quad C^{(n)} = \frac{1}{i^n} \: \dv[n]{k} \Big(\ln\!\big(\phi(k)\big)\Big) \big|_{k = 0} \end{aligned}\]

The first two cumulants \(C^{(1)}\) and \(C^{(2)}\) are of particular interest, since they turn out to be the mean and the variance respectively, using our earlier relation:

\[\begin{aligned} C^{(1)} &= - i \dv{k} \Big(\ln\!\big(\phi(k)\big)\Big) \big|_{k = 0} = - i \frac{\phi'(0)}{\exp(0)} = \overline{x} \\ C^{(2)} &= - \dv[2]{k} \Big(\ln\!\big(\phi(k)\big)\Big) \big|_{k = 0} = \frac{\big(\phi'(0)\big)^2}{\exp(0)^2} - \frac{\phi''(0)}{\exp(0)} = - \overline{x}^2 + \overline{x^2} = \sigma^2 \end{aligned}\]

Let us now define \(S\) as the sum of \(N\) independent variables \(x_n\), in other words:

\[\begin{aligned} S = \sum_{n = 1}^N x_n = x_1 + x_2 + ... + x_N \end{aligned}\]

The probability density of \(S\) is then as follows, where \(p(x_n)\) are the densities of all the individual variables and \(\delta\) is the Dirac delta function:

\[\begin{aligned} p(S) &= \idotsint_{-\infty}^\infty \Big( \prod_{n = 1}^N p(x_n) \Big) \: \delta\Big( S - \sum_{n = 1}^N x_n \Big) \dd{x_1} \cdots \dd{x_N} \\ &= \Big( p_1 * \big( p_2 * ( ... * (p_N * \delta))\big)\Big)(S) \end{aligned}\]

In other words, the integrals pick out all combinations of \(x_n\) which add up to the desired \(S\)-value, and multiply the probabilities \(p(x_1) p(x_2) \cdots p(x_N)\) of each such case. This is a convolution, so the convolution theorem states that it is a product in the Fourier domain:

\[\begin{aligned} \phi_S(k) = \prod_{n = 1}^N \phi_n(k) \end{aligned}\]

By taking the logarithm of both sides, the product becomes a sum, which we further expand:

\[\begin{aligned} \ln\!\big(\phi_S(k)\big) = \sum_{n = 1}^N \ln\!\big(\phi_n(k)\big) = \sum_{n = 1}^N \sum_{m = 1}^{\infty} \frac{(ik)^m}{m!} C_n^{(m)} \end{aligned}\]

Consequently, the cumulants \(C^{(m)}\) stack additively for the sum \(S\) of independent variables \(x_m\), and therefore the means \(C^{(1)}\) and variances \(C^{(2)}\) do too:

\[\begin{aligned} C_S^{(m)} = \sum_{n = 1}^N C_n^{(m)} = C_1^{(m)} + C_2^{(m)} + ... + C_N^{(m)} \end{aligned}\]

We now introduce the scaled sum \(z\) as the new combined variable:

\[\begin{aligned} z = \frac{S}{\sqrt{N}} = \frac{1}{\sqrt{N}} (x_1 + x_2 + ... + x_N) \end{aligned}\]

Its characteristic function \(\phi_z(k)\) is then as follows, with \(\sqrt{N}\) appearing in the arguments of \(\phi_n\):

\[\begin{aligned} \phi_z(k) &= \idotsint \Big( \prod_{n = 1}^N p(x_n) \Big) \: \delta\Big( z - \frac{1}{\sqrt{N}} \sum_{n = 1}^N x_n \Big) \exp(i k z) \dd{x_1} \cdots \dd{x_N} \\ &= \idotsint \Big( \prod_{n = 1}^N p(x_n) \Big) \exp\!\Big( i \frac{k}{\sqrt{N}} \sum_{n = 1}^N x_n \Big) \dd{x_1} \cdots \dd{x_N} \\ &= \prod_{n = 1}^N \phi_n\Big(\frac{k}{\sqrt{N}}\Big) \end{aligned}\]

By expanding \(\ln\!\big(\phi_z(k)\big)\) in terms of its cumulants \(C^{(m)}\) and introducing \(\kappa = k / \sqrt{N}\), we see that the higher-order terms become smaller for larger \(N\):

\[\begin{gathered} \ln\!\big( \phi_z(k) \big) = \sum_{m = 1}^\infty \frac{(ik)^m}{m!} C^{(m)} \\ C^{(m)} = \frac{1}{i^m} \dv[m]{k} \sum_{n = 1}^N \ln\!\bigg( \phi_n\Big(\frac{k}{\sqrt{N}}\Big) \bigg) = \frac{1}{i^m N^{m/2}} \dv[m]{\kappa} \sum_{n = 1}^N \ln\!\big( \phi_n(\kappa) \big) \end{gathered}\]

For sufficiently large \(N\), we can therefore approximate it using just the first two terms:

\[\begin{aligned} \ln\!\big( \phi_z(k) \big) &\approx i k C^{(1)} - \frac{k^2}{2} C^{(2)} = i k \overline{z} - \frac{k^2}{2} \sigma_z^2 \\ \phi_z(k) &\approx \exp(i k \overline{z}) \exp\!(- k^2 \sigma_z^2 / 2) \end{aligned}\]

We take its inverse Fourier transform to get the density \(p(z)\), which turns out to be a Gaussian normal distribution, which is even already normalized:

\[\begin{aligned} p(z) = \hat{\mathcal{F}}^{-1} \{\phi_z(k)\} &= \frac{1}{2 \pi} \int_{-\infty}^\infty \exp\!\big(\!-\! i k (z - \overline{z})\big) \exp(- k^2 \sigma_z^2 / 2) \dd{k} \\ &= \frac{1}{\sqrt{2 \pi \sigma_z^2}} \exp\!\Big(\!-\! \frac{(z - \overline{z})^2}{2 \sigma_z^2} \Big) \end{aligned}\]

Therefore, the sum of many independent variables tends to a normal distribution, regardless of the densities of the individual variables.


  1. H. Gould, J. Tobochnik, Statistical and thermal physics, 2nd edition, Princeton.

© Marcus R.A. Newman, a.k.a. "Prefetch". Available under CC BY-SA 4.0.