Conflicting nomenclatures and notations

Nomenclatures differ. Some statisticians, following the probabilist William Feller in his two-volume book An Introduction to Probability Theory and Its Applications,^[2] call the matrix ${\displaystyle \operatorname {K} _{\mathbf {X} \mathbf {X} }}$ the variance of the random vector ${\displaystyle \mathbf {X} }$, 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 ${\displaystyle \mathbf {X} }$.

${\displaystyle \operatorname {var} (\mathbf {X} )=\operatorname {cov} (\mathbf {X} ,\mathbf {X} )=\operatorname {E} \left[(\mathbf {X} -\operatorname {E} [\mathbf {X} ])(\mathbf {X} -\operatorname {E} [\mathbf {X} ])^{\rm {T}}\right].}$

Both forms are quite standard, and there is no ambiguity between them. The matrix ${\displaystyle \operatorname {K} _{\mathbf {X} \mathbf {X} }}$ is also often called the variance-covariance matrix, since the diagonal terms are in fact variances.

By comparison, the notation for the cross-covariance matrix between two vectors is

${\displaystyle \operatorname {cov} (\mathbf {X} ,\mathbf {Y} )=\operatorname {K} _{\mathbf {X} \mathbf {Y} }=\operatorname {E} \left[(\mathbf {X} -\operatorname {E} [\mathbf {X} ])(\mathbf {Y} -\operatorname {E} [\mathbf {Y} ])^{\rm {T}}\right].}$

Relation to the autocorrelation matrix

The auto-covariance matrix ${\displaystyle \operatorname {K} _{\mathbf {X} \mathbf {X} }}$ is related to the autocorrelation matrix ${\displaystyle \operatorname {R} _{\mathbf {X} \mathbf {X} }}$ by

${\displaystyle \operatorname {K} _{\mathbf {X} \mathbf {X} }=\operatorname {E} [(\mathbf {X} -\operatorname {E} [\mathbf {X} ])(\mathbf {X} -\operatorname {E} [\mathbf {X} ])^{\rm {T}}]=\operatorname {R} _{\mathbf {X} \mathbf {X} }-\operatorname {E} [\mathbf {X} ]\operatorname {E} [\mathbf {X} ]^{\rm {T}}}$

where the autocorrelation matrix is defined as ${\displaystyle \operatorname {R} _{\mathbf {X} \mathbf {X} }=\operatorname {E} [\mathbf {X} \mathbf {X} ^{\rm {T}}]}$.

Relation to the correlation matrix

An entity closely related to the covariance matrix is the matrix of Pearson product-moment correlation coefficients between each of the random variables in the random vector ${\displaystyle \mathbf {X} }$, which can be written as

${\displaystyle \operatorname {corr} (\mathbf {X} )={\big (}\operatorname {diag} (\operatorname {K} _{\mathbf {X} \mathbf {X} }){\big )}^{-{\frac {1}{2}}}\,\operatorname {K} _{\mathbf {X} \mathbf {X} }\,{\big (}\operatorname {diag} (\operatorname {K} _{\mathbf {X} \mathbf {X} }){\big )}^{-{\frac {1}{2}}},}$

where ${\displaystyle \operatorname {diag} (\operatorname {K} _{\mathbf {X} \mathbf {X} })}$ is the matrix of the diagonal elements of ${\displaystyle \operatorname {K} _{\mathbf {X} \mathbf {X} }}$ (i.e., a diagonal matrix of the variances of ${\displaystyle X_{i}}$ for ${\displaystyle i=1,\dots ,n}$).

Equivalently, the correlation matrix can be seen as the covariance matrix of the standardized random variables ${\displaystyle X_{i}/\sigma (X_{i})}$ for ${\displaystyle i=1,\dots ,n}$.

${\displaystyle \operatorname {corr} (\mathbf {X} )={\begin{bmatrix}1&{\frac {\operatorname {E} [(X_{1}-\mu _{1})(X_{2}-\mu _{2})]}{\sigma (X_{1})\sigma (X_{2})}}&\cdots &{\frac {\operatorname {E} [(X_{1}-\mu _{1})(X_{n}-\mu _{n})]}{\sigma (X_{1})\sigma (X_{n})}}\\\\{\frac {\operatorname {E} [(X_{2}-\mu _{2})(X_{1}-\mu _{1})]}{\sigma (X_{2})\sigma (X_{1})}}&1&\cdots &{\frac {\operatorname {E} [(X_{2}-\mu _{2})(X_{n}-\mu _{n})]}{\sigma (X_{2})\sigma (X_{n})}}\\\\\vdots &\vdots &\ddots &\vdots \\\\{\frac {\operatorname {E} [(X_{n}-\mu _{n})(X_{1}-\mu _{1})]}{\sigma (X_{n})\sigma (X_{1})}}&{\frac {\operatorname {E} [(X_{n}-\mu _{n})(X_{2}-\mu _{2})]}{\sigma (X_{n})\sigma (X_{2})}}&\cdots &1\end{bmatrix}}.}$

Each element on the principal diagonal of a correlation matrix is the correlation of a random variable with itself, which always equals 1. Each off-diagonal element is between −1 and +1 inclusive.

Inverse of the covariance matrix

The inverse of this matrix, ${\displaystyle \operatorname {K} _{\mathbf {X} \mathbf {X} }^{-1}}$, if it exists, is the inverse covariance matrix (or inverse concentration matrix), also known as the precision matrix (or concentration matrix).^[3]

Just as the covariance matrix can be written as the rescaling of a correlation matrix by the marginal variances:

${\displaystyle \operatorname {cov} (\mathbf {X} )={\begin{bmatrix}\sigma _{x_{1}}&&&0\\&\sigma _{x_{2}}\\&&\ddots \\0&&&\sigma _{x_{n}}\end{bmatrix}}{\begin{bmatrix}1&\rho _{x_{1},x_{2}}&\cdots &\rho _{x_{1},x_{n}}\\\rho _{x_{2},x_{1}}&1&\cdots &\rho _{x_{2},x_{n}}\\\vdots &\vdots &\ddots &\vdots \\\rho _{x_{n},x_{1}}&\rho _{x_{n},x_{2}}&\cdots &1\\\end{bmatrix}}{\begin{bmatrix}\sigma _{x_{1}}&&&0\\&\sigma _{x_{2}}\\&&\ddots \\0&&&\sigma _{x_{n}}\end{bmatrix}}}$

So, using the idea of partial correlation, and partial variance, the inverse covariance matrix can be expressed analogously:

${\displaystyle \operatorname {cov} (\mathbf {X} )^{-1}={\begin{bmatrix}{\frac {1}{\sigma _{x_{1}|x_{2}...}}}&&&0\\&{\frac {1}{\sigma _{x_{2}|x_{1},x_{3}...}}}\\&&\ddots \\0&&&{\frac {1}{\sigma _{x_{n}|x_{1}...x_{n-1}}}}\end{bmatrix}}{\begin{bmatrix}1&-\rho _{x_{1},x_{2}\mid x_{3}...}&\cdots &-\rho _{x_{1},x_{n}\mid x_{2}...x_{n-1}}\\-\rho _{x_{2},x_{1}\mid x_{3}...}&1&\cdots &-\rho _{x_{2},x_{n}\mid x_{1},x_{3}...x_{n-1}}\\\vdots &\vdots &\ddots &\vdots \\-\rho _{x_{n},x_{1}\mid x_{2}...x_{n-1}}&-\rho _{x_{n},x_{2}\mid x_{1},x_{3}...x_{n-1}}&\cdots &1\\\end{bmatrix}}{\begin{bmatrix}{\frac {1}{\sigma _{x_{1}|x_{2}...}}}&&&0\\&{\frac {1}{\sigma _{x_{2}|x_{1},x_{3}...}}}\\&&\ddots \\0&&&{\frac {1}{\sigma _{x_{n}|x_{1}...x_{n-1}}}}\end{bmatrix}}}$

This duality motivates a number of other dualities between marginalizing and conditioning for gaussian random variables.

Basic properties

For ${\displaystyle \operatorname {K} _{\mathbf {X} \mathbf {X} }=\operatorname {var} (\mathbf {X} )=\operatorname {E} \left[\left(\mathbf {X} -\operatorname {E} [\mathbf {X} ]\right)\left(\mathbf {X} -\operatorname {E} [\mathbf {X} ]\right)^{\rm {T}}\right]}$ and ${\displaystyle \mathbf {\mu _{X}} =\operatorname {E} [{\textbf {X}}]}$, where ${\displaystyle \mathbf {X} =(X_{1},\ldots ,X_{n})^{\rm {T}}}$ is a ${\displaystyle n}$-dimensional random variable, the following basic properties apply:^[4]

1. ${\displaystyle \operatorname {K} _{\mathbf {X} \mathbf {X} }=\operatorname {E} (\mathbf {XX^{\rm {T}}} )-\mathbf {\mu _{X}} \mathbf {\mu _{X}} ^{\rm {T}}}$
2. ${\displaystyle \operatorname {K} _{\mathbf {X} \mathbf {X} }\,}$ is positive-semidefinite, i.e. ${\displaystyle \mathbf {a} ^{T}\operatorname {K} _{\mathbf {X} \mathbf {X} }\mathbf {a} \geq 0\quad {\text{for all }}\mathbf {a} \in \mathbb {R} ^{n}}$
3. ${\displaystyle \operatorname {K} _{\mathbf {X} \mathbf {X} }\,}$ is symmetric, i.e. ${\displaystyle \operatorname {K} _{\mathbf {X} \mathbf {X} }^{\rm {T}}=\operatorname {K} _{\mathbf {X} \mathbf {X} }}$
4. For any constant (i.e. non-random) ${\displaystyle m\times n}$ matrix ${\displaystyle \mathbf {A} }$ and constant ${\displaystyle m\times 1}$ vector ${\displaystyle \mathbf {a} }$, one has ${\displaystyle \operatorname {var} (\mathbf {AX} +\mathbf {a} )=\mathbf {A} \,\operatorname {var} (\mathbf {X} )\,\mathbf {A} ^{\rm {T}}}$
5. If ${\displaystyle \mathbf {Y} }$ is another random vector with the same dimension as ${\displaystyle \mathbf {X} }$, then ${\displaystyle \operatorname {var} (\mathbf {X} +\mathbf {Y} )=\operatorname {var} (\mathbf {X} )+\operatorname {cov} (\mathbf {X} ,\mathbf {Y} )+\operatorname {cov} (\mathbf {Y} ,\mathbf {X} )+\operatorname {var} (\mathbf {Y} )}$ where ${\displaystyle \operatorname {cov} (\mathbf {X} ,\mathbf {Y} )}$ is the cross-covariance matrix of ${\displaystyle \mathbf {X} }$ and ${\displaystyle \mathbf {Y} }$.

Block matrices

The joint mean ${\displaystyle \mathbf {\mu } }$ and joint covariance matrix ${\displaystyle \mathbf {\Sigma } }$ of ${\displaystyle \mathbf {X} }$ and ${\displaystyle \mathbf {Y} }$ can be written in block form

${\displaystyle \mathbf {\mu } ={\begin{bmatrix}\mathbf {\mu _{X}} \\\mathbf {\mu _{Y}} \end{bmatrix}},\qquad \mathbf {\Sigma } ={\begin{bmatrix}\operatorname {K} _{\mathbf {XX} }&\operatorname {K} _{\mathbf {XY} }\\\operatorname {K} _{\mathbf {YX} }&\operatorname {K} _{\mathbf {YY} }\end{bmatrix}}}$

where ${\displaystyle \operatorname {K} _{\mathbf {XX} }=\operatorname {var} (\mathbf {X} )}$, ${\displaystyle \operatorname {K} _{\mathbf {YY} }=\operatorname {var} (\mathbf {Y} )}$ and ${\displaystyle \operatorname {K} _{\mathbf {XY} }=\operatorname {K} _{\mathbf {YX} }^{\rm {T}}=\operatorname {cov} (\mathbf {X} ,\mathbf {Y} )}$.

${\displaystyle \operatorname {K} _{\mathbf {XX} }}$ and ${\displaystyle \operatorname {K} _{\mathbf {YY} }}$ can be identified as the variance matrices of the marginal distributions for ${\displaystyle \mathbf {X} }$ and ${\displaystyle \mathbf {Y} }$ respectively.

If ${\displaystyle \mathbf {X} }$ and ${\displaystyle \mathbf {Y} }$ are jointly normally distributed,

${\displaystyle \mathbf {X} ,\mathbf {Y} \sim \ {\mathcal {N}}(\mathbf {\mu } ,\operatorname {\mathbf {\Sigma } } ),}$

then the conditional distribution for ${\displaystyle \mathbf {Y} }$ given ${\displaystyle \mathbf {X} }$ is given by

${\displaystyle \mathbf {Y} \mid \mathbf {X} \sim \ {\mathcal {N}}(\mathbf {\mu _{Y|X}} ,\operatorname {K} _{\mathbf {Y|X} }),}$^[5]

defined by conditional mean

${\displaystyle \mathbf {\mu _{Y|X}} =\mathbf {\mu _{Y}} +\operatorname {K} _{\mathbf {YX} }\operatorname {K} _{\mathbf {XX} }^{-1}\left(\mathbf {X} -\mathbf {\mu _{X}} \right)}$

and conditional variance

${\displaystyle \operatorname {K} _{\mathbf {Y|X} }=\operatorname {K} _{\mathbf {YY} }-\operatorname {K} _{\mathbf {YX} }\operatorname {K} _{\mathbf {XX} }^{-1}\operatorname {K} _{\mathbf {XY} }.}$

The matrix ${\displaystyle \operatorname {K} _{\mathbf {YX} }\operatorname {K} _{\mathbf {XX} }^{-1}}$ is known as the matrix of regression coefficients, while in linear algebra ${\displaystyle \operatorname {K} _{\mathbf {Y|X} }}$ is the Schur complement of ${\displaystyle \operatorname {K} _{\mathbf {XX} }}$ in ${\displaystyle \mathbf {\Sigma } }$.

The matrix of regression coefficients may often be given in transpose form, ${\displaystyle \operatorname {K} _{\mathbf {XX} }^{-1}\operatorname {K} _{\mathbf {XY} }}$, suitable for post-multiplying a row vector of explanatory variables ${\displaystyle \mathbf {X} ^{\rm {T}}}$ rather than pre-multiplying a column vector ${\displaystyle \mathbf {X} }$. In this form they correspond to the coefficients obtained by inverting the matrix of the normal equations of ordinary least squares (OLS).

Pseudo-covariance matrix

For complex random vectors, another kind of second central moment, the pseudo-covariance matrix (also called relation matrix) is defined as follows:

${\displaystyle \operatorname {J} _{\mathbf {Z} \mathbf {Z} }=\operatorname {cov} [\mathbf {Z} ,{\overline {\mathbf {Z} }}]=\operatorname {E} \left[(\mathbf {Z} -\mathbf {\mu _{Z}} )(\mathbf {Z} -\mathbf {\mu _{Z}} )^{\mathrm {T} }\right]}$

In contrast to the covariance matrix defined above, Hermitian transposition gets replaced by transposition in the definition. Its diagonal elements may be complex valued; it is a complex symmetric matrix.

Use in optimization

The evolution strategy, a particular family of Randomized Search Heuristics, fundamentally relies on a covariance matrix in its mechanism. The characteristic mutation operator draws the update step from a multivariate normal distribution using an evolving covariance matrix. There is a formal proof that the evolution strategy's covariance matrix adapts to the inverse of the Hessian matrix of the search landscape, up to a scalar factor and small random fluctuations (proven for a single-parent strategy and a static model, as the population size increases, relying on the quadratic approximation).^[9] Intuitively, this result is supported by the rationale that the optimal covariance distribution can offer mutation steps whose equidensity probability contours match the level sets of the landscape, and so they maximize the progress rate.

Covariance mapping

In covariance mapping the values of the ${\displaystyle \operatorname {cov} (\mathbf {X} ,\mathbf {Y} )}$ or ${\displaystyle \operatorname {pcov} (\mathbf {X} ,\mathbf {Y} \mid \mathbf {I} )}$ matrix are plotted as a 2-dimensional map. When vectors ${\displaystyle \mathbf {X} }$ and ${\displaystyle \mathbf {Y} }$ are discrete random functions, the map shows statistical relations between different regions of the random functions. Statistically independent regions of the functions show up on the map as zero-level flatland, while positive or negative correlations show up, respectively, as hills or valleys.

In practice the column vectors ${\displaystyle \mathbf {X} ,\mathbf {Y} }$, and ${\displaystyle \mathbf {I} }$ are acquired experimentally as rows of ${\displaystyle n}$ samples, e.g.

${\displaystyle [\mathbf {X} _{1},\mathbf {X} _{2},...\mathbf {X} _{n}]={\begin{bmatrix}X_{1}(t_{1})&X_{2}(t_{1})&\cdots &X_{n}(t_{1})\\\\X_{1}(t_{2})&X_{2}(t_{2})&\cdots &X_{n}(t_{2})\\\\\vdots &\vdots &\ddots &\vdots \\\\X_{1}(t_{m})&X_{2}(t_{m})&\cdots &X_{n}(t_{m})\end{bmatrix}},}$

where ${\displaystyle X_{j}(t_{i})}$ is the i-th discrete value in sample j of the random function ${\displaystyle X(t)}$. The expected values needed in the covariance formula are estimated using the sample mean, e.g.

${\displaystyle \langle \mathbf {X} \rangle ={\frac {1}{n}}\sum _{j=1}^{n}\mathbf {X} _{j}}$

and the covariance matrix is estimated by the sample covariance matrix

${\displaystyle \operatorname {cov} (\mathbf {X} ,\mathbf {Y} )\approx \langle \mathbf {XY^{\rm {T}}} \rangle -\langle \mathbf {X} \rangle \langle \mathbf {Y} ^{\rm {T}}\rangle ,}$

where the angular brackets denote sample averaging as before except that the Bessel's correction should be made to avoid bias. Using this estimation the partial covariance matrix can be calculated as

${\displaystyle \operatorname {pcov} (\mathbf {X} ,\mathbf {Y} \mid \mathbf {I} )=\operatorname {cov} (\mathbf {X} ,\mathbf {Y} )-\operatorname {cov} (\mathbf {X} ,\mathbf {I} )\left(\operatorname {cov} (\mathbf {I} ,\mathbf {I} )\backslash \operatorname {cov} (\mathbf {I} ,\mathbf {Y} )\right),}$

where the backslash denotes the left matrix division operator, which bypasses the requirement to invert a matrix and is available in some computational packages such as Matlab.^[10]

Figure 1: Construction of a partial covariance map of N₂ molecules undergoing Coulomb explosion induced by a free-electron laser.^[11] Panels a and b map the two terms of the covariance matrix, which is shown in panel c. Panel d maps common-mode correlations via intensity fluctuations of the laser. Panel e maps the partial covariance matrix that is corrected for the intensity fluctuations. Panel f shows that 10% overcorrection improves the map and makes ion-ion correlations clearly visible. Owing to momentum conservation these correlations appear as lines approximately perpendicular to the autocorrelation line (and to the periodic modulations which are caused by detector ringing).

Fig. 1 illustrates how a partial covariance map is constructed on an example of an experiment performed at the FLASH free-electron laser in Hamburg.^[11] The random function ${\displaystyle X(t)}$ is the time-of-flight spectrum of ions from a Coulomb explosion of nitrogen molecules multiply ionised by a laser pulse. Since only a few hundreds of molecules are ionised at each laser pulse, the single-shot spectra are highly fluctuating. However, collecting typically ${\displaystyle m=10^{4}}$ such spectra, ${\displaystyle \mathbf {X} _{j}(t)}$, and averaging them over ${\displaystyle j}$ produces a smooth spectrum ${\displaystyle \langle \mathbf {X} (t)\rangle }$, which is shown in red at the bottom of Fig. 1. The average spectrum ${\displaystyle \langle \mathbf {X} \rangle }$ reveals several nitrogen ions in a form of peaks broadened by their kinetic energy, but to find the correlations between the ionisation stages and the ion momenta requires calculating a covariance map.

In the example of Fig. 1 spectra ${\displaystyle \mathbf {X} _{j}(t)}$ and ${\displaystyle \mathbf {Y} _{j}(t)}$ are the same, except that the range of the time-of-flight ${\displaystyle t}$ differs. Panel a shows ${\displaystyle \langle \mathbf {XY^{\rm {T}}} \rangle }$, panel b shows ${\displaystyle \langle \mathbf {X} \rangle \langle \mathbf {Y^{\rm {T}}} \rangle }$ and panel c shows their difference, which is ${\displaystyle \operatorname {cov} (\mathbf {X} ,\mathbf {Y} )}$ (note a change in the colour scale). Unfortunately, this map is overwhelmed by uninteresting, common-mode correlations induced by laser intensity fluctuating from shot to shot. To suppress such correlations the laser intensity ${\displaystyle I_{j}}$ is recorded at every shot, put into ${\displaystyle \mathbf {I} }$ and ${\displaystyle \operatorname {pcov} (\mathbf {X} ,\mathbf {Y} \mid \mathbf {I} )}$ is calculated as panels d and e show. The suppression of the uninteresting correlations is, however, imperfect because there are other sources of common-mode fluctuations than the laser intensity and in principle all these sources should be monitored in vector ${\displaystyle \mathbf {I} }$. Yet in practice it is often sufficient to overcompensate the partial covariance correction as panel f shows, where interesting correlations of ion momenta are now clearly visible as straight lines centred on ionisation stages of atomic nitrogen.

Two-dimensional infrared spectroscopy

Two-dimensional infrared spectroscopy employs correlation analysis to obtain 2D spectra of the condensed phase. There are two versions of this analysis: synchronous and asynchronous. Mathematically, the former is expressed in terms of the sample covariance matrix and the technique is equivalent to covariance mapping.^[12]