# What's the best way to think about the covariance matrix?

Let $X$ be a random vector with covariance matrix $\Sigma$.

People often describe $\Sigma$ in terms of its components: $\Sigma_{ij}$ is the covariance of the $i$th and $j$th components of $X$.

But in linear algebra, thinking about a matrix in terms of its components is often discouraged. It is often more enlightening to avoid thinking in terms of components.

So what is the "best" way to think about $\Sigma$, particularly for someone who likes linear algebra?

I know that $\Sigma = \mathbb E((X - \mu)(X - \mu)^T)$, where $\mu = \mathbb E(X)$. But I think I am still missing something, because I'm not sure what to make of that formula. Does this formula shed light on what $\Sigma$ really is and why we care about it?

## 1 Answer

The best way to think about the covariance matrix: this is the matrix of the quadratic form

$$\beta\in\Bbb R^d \to var \langle \beta, X\rangle$$

Indeed, \begin{align} var \langle \beta, X\rangle &= var \sum_{i=1}^d \beta_i X_i \\ &= \sum_{i=1}^d \sum_{j=1}^d \beta_i \beta_j cov( X_i ,X_j) \\ &= \sum_{i=1}^d \sum_{j=1}^d \beta_i \beta_j \Sigma_{i,j} \\ &= \beta^T\Sigma\beta \\ \end{align}

• Thank you, great answer. – eternalGoldenBraid Apr 30 '15 at 20:07