6.3-exam3-2010-strang¶
Quiz 3, 2010 (strang) · Section 6.3: Symmetric Positive Definite Matrices Source: exam3_2.pdf, p. 15
Problem¶
2. (30 points) Suppose \(A\) is a positive definite symmetric \(n\) by \(n\) matrix.
(a) How do you know that \(A^{-1}\) is also positive definite? (We know \(A^{-1}\) is symmetric. I just had an e-mail from the International Monetary Fund with this question.)
(b) Suppose \(Q\) is any orthogonal \(n\) by \(n\) matrix. How do you know that \(QAQ^T = QAQ^{-1}\) is positive definite? Write down which test you are using.
(c) Show that the block matrix
\[
B = \begin{bmatrix} A & A \\ A & A \end{bmatrix}
\]
is positive semidefinite. How do you know \(B\) is not positive definite?
Solution
(a) Since a matrix is positive definite if and only if all its eigenvalues are positive, and since the eigenvalues of \(A^{-1}\) are simply the inverses of the eigenvalues of \(A\), the matrix \(A^{-1}\) is also positive definite (the inverse of a positive number is positive). (b) Using the energy test (\(x^TAx > 0\) for nonzero \(x\)), we find that \[ x^TQAQ^Tx = (Q^Tx)^TA(Q^Tx) > 0 \] for all nonzero \(x\) as well, since \(Q\) is invertible. Alternatively, using the positive eigenvalue test, \(A\) is similar to \(QAQ^{-1}\), and similar matrices have the same eigenvalues, so \(QAQ^{-1}\) also has all positive eigenvalues. (c) First, since \(B\) is singular, it cannot be positive definite (it has eigenvalue 0). However, the pivots of \(B\) are the pivots of \(A\) in the first \(n\) rows followed by 0s in the remaining rows, so by the pivot test \(B\) is still positive semidefinite. Similarly, the first \(n\) upper-left determinants of \(B\) are the same as those of \(A\), while the remaining ones are 0s, giving another proof. Finally, given a nonzero vector \[ u = \begin{bmatrix} x \\ y \end{bmatrix}, \] where \(x\) and \(y\) are vectors in \(\mathbb{R}^n\), one has \[ u^TBu = (x + y)^TA(x + y), \] which is nonnegative (and zero when \(x + y = 0\)).Keywords: positive definite, positive semidefinite, eigenvalues, similar matrices, block matrix