The Multivariate Distribution
The multivariate distribution of a random variable \(x \in \mathbb{R}^n\) is written as \(x \sim \mathcal{N}(\mu, \Sigma)\) where \(\mu \in \mathbb{R}^n\) is the mean vector and \(\Sigma \in \mathbb{R}^{n \times n}\) is the covariance matrix. \(\Sigma\geqslant 0\) is symmetric and positive semi-definite. Note that the multivariate distribution generalizes the one-dimensional normal distribution to n-dimensions. The probability density function is given by:
For this multivariate distribution the expectation is given by:
Also, the covariance matrix, which generalizes the notion of variance, is given by:
A multivariate distribution with a zero mean vector and an identity matrix as the covariance matrix is known as the standard normal distribution.
Marginals and Conditionals of Gaussians
Suppose that we have a vector-valued random variable \(x\sim \mathcal{N}(\mu,\Sigma)\) where:
Here \(x_1, \mu_1 \in \mathcal{R}^{n_1}\), \(x_2, \mu_2 \in \mathcal{R}^{n_2}\), \(x, \mu \in \mathcal{R}^{n_1+n_2}\), \(\Sigma_{11} \in \mathcal{R}^{n_1 \times n_1}\), \(\Sigma_{12} \in \mathcal{R}^{n_1 \times n_2}\), \(\Sigma_{21} \in \mathcal{R}^{n_2 \times n_1}\), \(\Sigma_{22} \in \mathcal{R}^{n_2 \times n_2}\) and \(\Sigma \in \mathcal{R}^{(n_1+n_2) \times (n_1+n_2)}\).
Note that \(\mathbb{E}(x_1)=\mu_1\)and \(\mathbb{E}(x_2)=\mu_2\). Also, note that:
Note that \(Cov(x_1) = (x_1 - \mu_1)(x_1 - \mu_1)^T = \Sigma_{11}\) and similarly \(Cov(x_2) = \Sigma_{22}\).
Note that \(x\) represents the joint multivariate density of \(x_1\) and \(x_2\). We have also found the marginal distributions of \(x_1\) and \(x_2\) to be \(\mathcal{N}(\mu_1,\Sigma_{11})\) and \(\mathcal{N}(\mu_2,\Sigma_{22})\) respectively.
It may also be shown that the conditional distribution \(x_1 \vert x_2\) is given by \(\mathcal{N}(\mu_{1 \vert 2},\Sigma_{1 \vert 2})\), where: