Matrix inversion formulae

Throughout the thesis we are using the matrix inversion lemma: having quadratic matrices A and B and a matrix X of corresponding number of columns and rows, we have the following equality (see for example [43, Appendix B] or [64])

$\displaystyle \left(\vphantom{ {\boldsymbol { A } }+{\boldsymbol { X } }{\boldsymbol { B } }{\boldsymbol { X } }^T}\right.$A + XBXT$\displaystyle \left.\vphantom{ {\boldsymbol { A } }+{\boldsymbol { X } }{\boldsymbol { B } }{\boldsymbol { X } }^T}\right)^{-1}_{}$ = A-1 - A-1X(B-1 + XTA-1X)-1XTA-1. (180)

The matrix inversion formula for a symmetric matrix with block sub-matrices is:
$\displaystyle \begin{bmatrix}
{\boldsymbol { A } }& {\boldsymbol { B } }\\
{\boldsymbol { B } }^T &{\boldsymbol { C } }
\end{bmatrix}^{-1}_{}$ = $\displaystyle \begin{bmatrix}
{\boldsymbol { D } }^{-1} & -{\boldsymbol { A } }...
...ol { B } }^T{\boldsymbol { A } }^{-1} & {\boldsymbol { E } }^{-1}
\end{bmatrix}$ (181)
       
  = $\displaystyle \begin{bmatrix}
{\boldsymbol { A } }^{-1} + {\boldsymbol { A } }^...
...dsymbol { D } }^{-1}{\boldsymbol { B } }{\boldsymbol { C } }^{-1}
\end{bmatrix}$ (182)
       
where    D   = A - BC-1BT  
E = C - BTA-1B   (183)

The formulae for the determinants of block matrices are also used in the thesis:

$\displaystyle \begin{vmatrix}{\boldsymbol { A } }& {\boldsymbol { B } }\\  {\boldsymbol { B } }^T &{\boldsymbol { C } } \end{vmatrix}$ = |A|  |E| = |C|  |D| (184)

A useful short guide to manipulating block matrices and computing determinants is provided by Sam Roweis, available at http://www.gatsby.ucl.ac.uk/~roweis/notes.html.