Matrix F-distribution

Matrix ${\displaystyle F}$
Notation	${\displaystyle {\mathcal {F}}({\mathbf {\Psi } },\nu ,\delta )}$
Parameters	${\displaystyle \mathbf {\Psi } >0}$, ${\displaystyle p\times p}$ scale matrix (pos. def.) ${\displaystyle \nu >p-1}$ degrees of freedom (real) ${\displaystyle \delta >0}$ degrees of freedom (real)
Support	${\displaystyle \mathbf {X} }$ is p × p positive definite matrix
PDF	${\displaystyle {\frac {\Gamma _{p}\left({\frac {\nu +\delta +p-1}{2}}\right)}{\Gamma _{p}\left({\frac {\nu }{2}}\right)\Gamma _{p}\left({\frac {\delta +p-1}{2}}\right)|\mathbf {\Psi } |^{\frac {\nu }{2}}}}~|{\mathbf {X} }|^{\frac {\nu -p-1}{2}}|{\textbf {I}}_{p}+{\mathbf {X} }\mathbf {\Psi } ^{-1}|^{-{\frac {\nu +\delta +p-1}{2}}}}$ ${\displaystyle \Gamma _{p}}$ is the multivariate gamma function ${\displaystyle {\textbf {I}}_{p}}$ is the p × p identity matrix
Mean	${\displaystyle {\tfrac {\nu }{\delta -2}}\mathbf {\Psi } }$, for ${\displaystyle \delta >2.}$
Variance	see below

In statistics, the matrix F distribution (or matrix variate F distribution) is a matrix variate generalization of the F distribution which is defined on real-valued positive-definite matrices. In Bayesian statistics it can be used as the semi conjugate prior for the covariance matrix or precision matrix of multivariate normal distributions, and related distributions.^[1][2][3][4]

The probability density function of the matrix ${\displaystyle F}$ distribution is:

${\displaystyle f_{\mathbf {X} }({\mathbf {X} };{\mathbf {\Psi } },\nu ,\delta )={\frac {\Gamma _{p}\left({\frac {\nu +\delta +p-1}{2}}\right)}{\Gamma _{p}\left({\frac {\nu }{2}}\right)\Gamma _{p}\left({\frac {\delta +p-1}{2}}\right)|\mathbf {\Psi } |^{\frac {\nu }{2}}}}~|{\mathbf {X} }|^{\frac {\nu -p-1}{2}}|{\textbf {I}}_{p}+{\mathbf {X} }\mathbf {\Psi } ^{-1}|^{-{\frac {\nu +\delta +p-1}{2}}}}$

where ${\displaystyle \mathbf {X} }$ and ${\displaystyle {\mathbf {\Psi } }}$ are ${\displaystyle p\times p}$ positive definite matrices, ${\displaystyle |\cdot |}$ is the determinant, Γ_p(⋅) is the multivariate gamma function, and ${\displaystyle {\textbf {I}}_{p}}$ is the p × p identity matrix.

• The standard matrix F distribution, with an identity scale matrix ${\displaystyle \mathbf {I} _{p}}$, was originally derived by.^[1] When considering independent distributions,

${\displaystyle {\mathbf {\Phi } _{1}}\sim {\mathcal {W}}({\mathbf {I} _{p}},\nu )}$ and ${\displaystyle {\mathbf {\Phi } _{2}}\sim {\mathcal {W}}({\mathbf {I} _{p}},\delta +k-1)}$, and define ${\displaystyle \mathbf {X} ={\mathbf {\Phi } _{2}}^{-1/2}{\mathbf {\Phi } _{1}}{\mathbf {\Phi } _{2}}^{-1/2}}$, then ${\displaystyle \mathbf {X} \sim {\mathcal {F}}({\mathbf {I} _{p}},\nu ,\delta )}$.

• If ${\displaystyle {\mathbf {X} }|\mathbf {\Phi } \sim {\mathcal {W}}^{-1}({\mathbf {\Phi } },\delta +p-1)}$ and ${\displaystyle {\mathbf {\Phi } }\sim {\mathcal {W}}({\mathbf {\Psi } },\nu )}$, then, after integrating out ${\displaystyle \mathbf {\Phi } }$, ${\displaystyle \mathbf {X} }$ has a matrix F-distribution, i.e.,

${\displaystyle f_{\mathbf {X} |\mathbf {\Phi } ,\nu ,\delta }(\mathbf {X} )=\int f_{\mathbf {X} |\mathbf {\Phi } ,\delta +p-1}(\mathbf {X} )f_{\mathbf {\Phi } |\mathbf {\Psi } ,\nu }(\mathbf {\Phi } )d\mathbf {\Phi } .}$
This construction is useful to construct a semi-conjugate prior for a covariance matrix.^[3]

• If ${\displaystyle {\mathbf {X} }|\mathbf {\Phi } \sim {\mathcal {W}}({\mathbf {\Phi } },\nu )}$ and ${\displaystyle {\mathbf {\Phi } }\sim {\mathcal {W}}^{-1}({\mathbf {\Psi } },\delta +p-1)}$, then, after integrating out ${\displaystyle \mathbf {\Phi } }$, ${\displaystyle \mathbf {X} }$ has a matrix F-distribution, i.e.,
  ${\displaystyle f_{\mathbf {X} |\mathbf {\Psi } ,\nu ,\delta }(\mathbf {X} )=\int f_{\mathbf {X} |\mathbf {\Phi } ,\nu }(\mathbf {X} )f_{\mathbf {\Phi } |\mathbf {\Psi } ,\delta +p-1}(\mathbf {\Phi } )d\mathbf {\Phi } .}$
  This construction is useful to construct a semi-conjugate prior for a precision matrix.^[4]

Suppose ${\displaystyle {\mathbf {A} }\sim F({\mathbf {\Psi } },\nu ,\delta )}$ has a matrix F distribution. Partition the matrices ${\displaystyle {\mathbf {A} }}$ and ${\displaystyle {\mathbf {\Psi } }}$ conformably with each other

${\displaystyle {\mathbf {A} }={\begin{bmatrix}\mathbf {A} _{11}&\mathbf {A} _{12}\\\mathbf {A} _{21}&\mathbf {A} _{22}\end{bmatrix}},\;{\mathbf {\Psi } }={\begin{bmatrix}\mathbf {\Psi } _{11}&\mathbf {\Psi } _{12}\\\mathbf {\Psi } _{21}&\mathbf {\Psi } _{22}\end{bmatrix}}}$

where ${\displaystyle {\mathbf {A} _{ij}}}$ and ${\displaystyle {\mathbf {\Psi } _{ij}}}$ are ${\displaystyle p_{i}\times p_{j}}$ matrices, then we have ${\displaystyle {\mathbf {A} _{11}}\sim F({\mathbf {\Psi } _{11}},\nu ,\delta )}$.

Let ${\displaystyle X\sim F({\mathbf {\Psi } },\nu ,\delta )}$.

The mean is given by: ${\displaystyle E(\mathbf {X} )={\frac {\nu }{\delta -2}}\mathbf {\Psi } .}$

The (co)variance of elements of ${\displaystyle \mathbf {X} }$ are given by:^[3]

${\displaystyle \operatorname {cov} (X_{ij},X_{ml})=\Psi _{ij}\Psi _{ml}{\tfrac {2\nu ^{2}+2\nu (\delta -2)}{(\delta -1)(\delta -2)^{2}(\delta -4)}}+(\Psi _{il}\Psi _{jm}+\Psi _{im}\Psi _{jl})\left({\tfrac {2\nu +\nu ^{2}(\delta -2)+\nu (\delta -2)}{(\delta -1)(\delta -2)^{2}(\delta -4)}}+{\tfrac {\nu }{(\delta -2)^{2}}}\right).}$

• The matrix F-distribution has also been termed the multivariate beta II distribution.^[5] See also,^[6] for a univariate version.
• A univariate version of the matrix F distribution is the F-distribution. With ${\displaystyle p=1}$ (i.e. univariate) and ${\displaystyle \mathbf {\Psi } =1}$, and ${\displaystyle x=\mathbf {X} }$, the probability density function of the matrix F distribution becomes the univariate (unscaled) F distribution:
  ${\displaystyle f_{x\mid \nu ,\delta }(x)=\operatorname {B} \left({\tfrac {\nu }{2}},{\tfrac {\delta }{2}}\right)^{-1}\left({\tfrac {\nu }{\delta }}\right)^{\nu /2}x^{\nu /2-1}\left(1+{\tfrac {\nu }{\delta }}\,x\right)^{-(\nu +\delta )/2},}$

• In the univariate case, with ${\displaystyle p=1}$ and ${\displaystyle x=\mathbf {X} }$, and when setting ${\displaystyle \nu =1}$, then ${\displaystyle {\sqrt {x}}}$ follows a half t distribution with scale parameter ${\displaystyle {\sqrt {\psi }}}$ and degrees of freedom ${\displaystyle \delta }$. The half t distribution is a common prior for standard deviations^[7]

• Inverse matrix gamma distribution
• Matrix normal distribution
• Wishart distribution
• Inverse Wishart distribution
• Complex inverse Wishart distribution