06.06.2013 Views

Theory of Statistics - George Mason University

Theory of Statistics - George Mason University

Theory of Statistics - George Mason University

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

0.3 Some Basics <strong>of</strong> Linear Algebra 791<br />

where 0 represents a submatrix with all zeros, and X represents a general<br />

submatrix, with at least some nonzeros. The diag(·) function previously introduced<br />

for a vector is also defined for a list <strong>of</strong> matrices:<br />

diag(A1, A2, . . ., Ak)<br />

denotes the block diagonal matrix with submatrices A1, A2, . . ., Ak along the<br />

diagonal and zeros elsewhere.<br />

0.3.2.13 Inverses <strong>of</strong> Partitioned Matrices<br />

If A is nonsingular, and can be partitioned as<br />

<br />

A11 A12<br />

A = ,<br />

A21 A22<br />

where both A11 and A22 are nonsingular, it is easy to see that the inverse <strong>of</strong><br />

A is given by<br />

A −1 =<br />

⎡<br />

⎣ A−1 11<br />

+ A−1 11 A12Z −1A21A −1<br />

11 −A−1 11<br />

−Z −1 A21A −1<br />

11<br />

A12Z −1<br />

Z −1<br />

where Z = A22 − A21A −1<br />

11 A12. In this partitioning Z is called the Schur complement<br />

<strong>of</strong> A11 in A.<br />

If<br />

A = [Xy] T [Xy]<br />

and is partitioned as into X T X and y T y on the diagonal, and X is <strong>of</strong> full<br />

column rank, then the Schur complement <strong>of</strong> X T X in [Xy] T [Xy] is<br />

y T y − y T X(X T X) −1 X T y.<br />

This particular partitioning is useful in linear regression analysis, where this<br />

Schur complement is the residual sum <strong>of</strong> squares.<br />

0.3.2.14 Gramian Matrices and Generalized Inverses<br />

A matrix <strong>of</strong> the form Z T Z is called a Gramian matrix. Such matrices arise<br />

<strong>of</strong>ten in statistical applications.<br />

Some interesting properties <strong>of</strong> a Gramian matrix Z T Z are<br />

• Z T Z is symmetric;<br />

• Z T Z is <strong>of</strong> full rank if and only if Z is <strong>of</strong> full column rank, or, more generally,<br />

rank(Z T Z) = rank(Z);<br />

<strong>Theory</strong> <strong>of</strong> <strong>Statistics</strong> c○2000–2013 James E. Gentle<br />

⎤<br />

⎦,

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!