Skip to content

6.3-exam3-2011-strang

Quiz 3, 2011 (strang) · Section 6.3: Symmetric Positive Definite Matrices Source: exam3_2.pdf, p. 31

Problem

3. Suppose \(A\) is a real \(m\) by \(n\) matrix.

(a) Prove that the symmetric matrix \(A^TA\) has the property \(x^T(A^TA)x \ge 0\) for every vector \(x\) in \(\mathbb{R}^n\). Explain each step in your reason.

(b) According to part (a), the matrix \(A^TA\) is positive semidefinite at least -- and possibly positive definite. Under what condition on \(A\) is \(A^TA\) positive definite?

(c) If \(m < n\) prove that \(A^TA\) is not positive definite.

Solution (a) \[ x^T(A^TA)x = (x^TA^T)Ax = (Ax)^T(Ax) = (Ax) \cdot (Ax) = \|Ax\|^2 \ge 0. \] (b) We want to see under what condition \(x^T(A^TA)x = 0\) implies \(x = 0\). So let \(x^TA^TAx = 0\). By part (a) we get \((Ax) \cdot (Ax) = 0\), so \(Ax = 0\). Hence, to get \(x = 0\) from \(Ax = 0\), we need \(N(A) = \{0\}\), i.e. \(A\) must have independent columns. (c) We use part (b) and show that if \(m < n\) then \(N(A) \neq \{0\}\). We know that \(\dim N(A) = n - r\), where \(r = \operatorname{rank}(A)\). But \(r \le m\), so \[ \dim N(A) = n - r \ge n - m. \] Since \(m < n\), we have \(n - m > 0\), and therefore \(N(A) \neq \{0\}\), so \(A^TA\) is not positive definite.

Keywords: positive semidefinite, positive definite, Gram matrix, nullspace, rank