Skip to content

Central Limit Theorem

As large \(n\) experiments with \(\mathbf{X}_1, \mathbf{X}_2, \mathbf{X}_3, \cdots, \mathbf{X}_n\) as i.i.d. random variables are conducted, the sample mean will converge to the expected value. This leads to the Central Limit Theorem, regardless of the distribution of \(\mathbf{X}_i\)'s.

With the sample mean:

\[ \bar{\mathbf{X}}_n = \frac{1}{n}\sum_{i=1}^n \mathbf{X}_i \sim \mathcal{N}(\mu, \frac{\sigma^2}{n}) \]

Proof

The distribution of the normalized sum is:

\[ \begin{align*} S_n &= \sum_{i=1}^n \mathbf{X}_i \\ NS_n &= \frac{S_n}{\sigma \sqrt{n}} \end{align*} \]

Proof Goal: The distribution of \(NS_n\) approaches the standard normal distribution \(\mathcal{N}(0,1)\) as \(n \to \infty\).

\[ \lim_{n \to \infty} P(NS_n \leq z) = \Phi(z) \]

where \(\Phi(z)\) is the cumulative distribution function (CDF) of the standard normal distribution. The moment-generating function \(m(t)\) for \(\mathcal{N}(0,1)\) is:

\[ m(t) = e^{\frac{t^2}{2}} \]

Taylor Expansion of \(m(t)\):

\[ m(t) = \underbrace{m(0)}_{=1} + \underbrace{tm'(0)}_{=0} + \underbrace{\frac{t^2}{2}m''(0)}_{\frac{t^2}{2}} + \underbrace{\cdots}_{\mathcal{E}(t^3)} \]

If \(\mathbf{Y} = b\mathbf{X}\), then \(m_{\mathbf{Y}} = m_{\mathbf{X}}(bt)\), so:

\[ \begin{align*} m_{\mathbf{Z}}(t) &= m_{S_n}\left(\frac{t}{\sigma \sqrt{n}}\right) \\ &= \left[m\left(\frac{t}{\sigma \sqrt{n}}\right)\right]^n \end{align*} \]

where:

\[ \begin{align*} m\left(\frac{t}{\sigma \sqrt{n}}\right) &= 1 + \frac{1}{2}\sigma^2\left(\frac{t^2}{\sigma^2 n}\right) + \mathcal{E}\left(\frac{t^3}{\sigma^3 n^{3/2}}\right) \\ &= 1 + \frac{1}{2}\left(\frac{t^2}{n}\right) + \mathcal{E}_n(\cdots) \end{align*} \]

Now, take the limit as \(n \to \infty\) of \(m_{\mathbf{Z}}(t)\):

\[ \begin{align*} \lim_{n \to \infty} m_{\mathbf{Z}}(t) &= \lim_{n \to \infty} \left(1 + \frac{1}{2}\left(\frac{t^2}{n}\right) + \mathcal{E}_n\right) \\ &= \lim_{n \to \infty} \left(1 + \frac{t^2}{2n}\right)^n \\ &= e^{\frac{t^2}{2}}, \quad \text{since} \quad \lim_{n \to \infty} \left(1 + \frac{a}{n}\right)^n = e^a \end{align*} \]

This proves that the moment-generating function of \(\mathcal{N}(0,1)\) is \(e^{\frac{t^2}{2}}\).