In statistics and probability theory, the covariance matrix is a matrix of covariances between elements of a vector. It is the natural generalization to higher dimensions of the concept of the variance of a scalar-valued random variable.

If entries in the column vector

<math>X = \begin{bmatrix}X_1 \\ \vdots \\ X_n \end{bmatrix}</math>

are random variables, each with finite variance, then the covariance matrix Σ is the matrix whose (i, j) entry is the covariance

<math>

\Sigma_{ij} =\mathrm{E}\begin{bmatrix} (X_i – \mu_i)(X_j – \mu_j) \end{bmatrix} </math>

where

<math>\mu_i = \mathrm{E}(X_i)\,</math>

is the expected value of the ith entry in the vector X. In other words, we have

<math>

\Sigma = \begin{bmatrix}

\mathrm{E}[(X_1 - \mu_1)(X_1 - \mu_1)] & \mathrm{E}[(X_1 - \mu_1)(X_2 - \mu_2)] & \cdots & \mathrm{E}[(X_1 - \mu_1)(X_n - \mu_n)] \\ \\
\mathrm{E}[(X_2 - \mu_2)(X_1 - \mu_1)] & \mathrm{E}[(X_2 - \mu_2)(X_2 - \mu_2)] & \cdots & \mathrm{E}[(X_2 - \mu_2)(X_n - \mu_n)] \\ \\
\vdots & \vdots & \ddots & \vdots \\ \\
\mathrm{E}[(X_n - \mu_n)(X_1 - \mu_1)] & \mathrm{E}[(X_n - \mu_n)(X_2 - \mu_2)] & \cdots & \mathrm{E}[(X_n - \mu_n)(X_n - \mu_n)]

\end{bmatrix}. </math>

The definition above is equivalent to the matrix equality

<math>

\Sigma=\mathrm{E} \left[

\left(
\textbf{X} - \mathrm{E}[\textbf{X}]
\right)
\left(
\textbf{X} - \mathrm{E}[\textbf{X}]
\right)^\top

\right] </math>

This form can be seen as a generalization of the scalar-valued variance to higher dimensions. Recall that for a scalar-valued random variable X

<math>

\sigma^2 = \mathrm{var}(X) = \mathrm{E}[(X-\mu)^2], \, </math>

where

<math>\mu = \mathrm{E}(X).\,</math>

Nomenclatures differ. Some statisticians, following the probabilist William Feller, call this matrix the variance of the random vector <math>X</math>, because it is the natural generalization to higher dimensions of the 1-dimensional variance. Others call it the covariance matrix, because it is the matrix of covariances between the scalar components of the vector <math>X</math>. Thus

<math>

\operatorname{var}(\textbf{X}) = \operatorname{cov}(\textbf{X}) = \mathrm{E} \left[

(\textbf{X} - \mathrm{E} [\textbf{X}])
(\textbf{X} - \mathrm{E} [\textbf{X}])^\top

\right] </math>

However, the notation for the “cross-covariance” between two vectors is standard:

<math>

\operatorname{cov}(\textbf{X},\textbf{Y}) = \mathrm{E} \left[

(\textbf{X} - \mathrm{E}[\textbf{X}])
(\textbf{Y} - \mathrm{E}[\textbf{Y}])^\top

\right] </math>

The <math>var</math> notation is found in William Feller’s two-volume book An Introduction to Probability Theory and Its Applications, but both forms are quite standard and there is no ambiguity between them.

For <math>\Sigma=\mathrm{E} \left[ \left( \textbf{X} – \mathrm{E}[\textbf{X}] \right) \left( \textbf{X} – \mathrm{E}[\textbf{X}] \right)^\top \right]</math> and <math> \mu = \mathrm{E}(\textbf{X})</math> the following basic properties apply:

1. <math> \Sigma = \mathrm{E}(\mathbf{X X^\top}) – \mathbf{\mu}\mathbf{\mu^\top} </math>
   
   
2. <math> \mathbf{\Sigma}</math> is positive semi-definite
   
   
3. <math> \operatorname{var}(\mathbf{A X} + \mathbf{a}) = \mathbf{A}\, \operatorname{var}(\mathbf{X})\, \mathbf{A^\top} </math>
   
   
4. <math> \operatorname{cov}(\mathbf{X},\mathbf{Y}) = \operatorname{cov}(\mathbf{Y},\mathbf{X})</math>
   
   
5. <math> \operatorname{cov}(\mathbf{X_1} + \mathbf{X_2},\mathbf{Y}) = \operatorname{cov}(\mathbf{X_1},\mathbf{Y}) + \operatorname{cov}(\mathbf{X_2}, \mathbf{Y})</math>
   
   
6. If p = q, then <math>\operatorname{var}(\mathbf{X} + \mathbf{Y}) = \operatorname{var}(\mathbf{X}) + \operatorname{cov}(\mathbf{X},\mathbf{Y}) + \operatorname{cov}(\mathbf{Y}, \mathbf{X}) + \operatorname{var}(\mathbf{Y})</math>
   
   
7. <math>\operatorname{cov}(\mathbf{AX}, \mathbf{BY}) = \mathbf{A}\, \operatorname{cov}(\mathbf{X}, \mathbf{Y}) \,\mathbf{B}^\top</math>
   
   
8. If <math>\mathbf{X}</math> and <math>\mathbf{Y}</math> are independent, then <math>\operatorname{cov}(\mathbf{X}, \mathbf{Y}) = 0</math>
   
   

where <math>\mathbf{X}, \mathbf{X_1}</math> and <math>\mathbf{X_2}</math> are a random <math>\mathbf{(p \times 1)}</math> vectors, <math>\mathbf{Y}</math> is a random <math>\mathbf{(q \times 1)}</math> vector, <math>\mathbf{a}</math> is <math>\mathbf{(p \times 1)}</math> vector, <math>\mathbf{A}</math> and <math>\mathbf{B}</math> are <math>\mathbf{(p \times q)}</math> matrices.

This covariance matrix (though very simple) is a very useful tool in many very different areas. From it a transformation matrix can be derived that allows one to completely decorrelate the data or, from a different point of view, to find an optimal basis for representing the data in a compact way (see Rayleigh quotient for a formal proof and additional properties of covariance matrices). This is called principal components analysis (PCA) in statistics and Karhunen-Loève transform (KL-transform) in image processing.

From the identity

<math>\operatorname{var}(\mathbf{a^\top}\mathbf{X}) = \mathbf{a^\top} \operatorname{var}(\mathbf{X}) \mathbf{a}\,</math>

and the fact that the variance of any real-valued random variable is nonnegative, it follows immediately that only a nonnegative-definite matrix can be a covariance matrix. The converse question is whether every nonnegative-definite symmetric matrix is a covariance matrix. The answer is “yes”. To see this, suppose M is a p×p nonnegative-definite symmetric matrix. From the finite-dimensional case of the spectral theorem, it follows that M has a nonnegative symmetric square root, which let us call M^1/2. Let <math>\mathbf{X}</math> be any p×1 column vector-valued random variable whose covariance matrix is the p×p identity matrix. Then

<math>\operatorname{var}(M^{1/2}\mathbf{X}) = M^{1/2} (\operatorname{var}(\mathbf{X})) M^{1/2} = M.\,</math>

The variance of a complex scalar-valued random variable with expected value μ is conventionally defined using complex conjugation:

<math>

\operatorname{var}(z) = \operatorname{E} \left[

(z-\mu)(z-\mu)^{*}

\right] </math>

where the complex conjugate of a complex number <math>z</math> is denoted <math>z^{*}</math>.

If <math>Z</math> is a column-vector of complex-valued random variables, then we take the conjugate transpose by both transposing and conjugating, getting a square matrix:

<math>

\operatorname{E} \left[

(Z-\mu)(Z-\mu)^{*}

\right] </math>

where <math>Z^{*}</math> denotes the conjugate transpose, which is applicable to the scalar case since the transpose of a scalar is still a scalar.

LaTeX provides useful features for dealing with covariance matrices. These are available through the extendedmath package.

The derivation of the maximum-likelihood estimator of the covariance matrix of a multivariate normal distribution is perhaps surprisingly subtle. It involves the spectral theorem and the reason why it can be better to view a scalar as the trace of a 1 × 1 matrix than as a mere scalar. See estimation of covariance matrices.

• Covariance Matrix at MathWorld
• van Kampen, N. G. Stochastic processes in physics and chemistry. New York: North-Holland, 1981.

• Estimation of covariance matrices
• Multivariate statistics
• Sample covariance matrix

de:Kovarianzmatrix it:Matrice delle covarianze