4.3-exam2-2004¶
Quiz 2, 2004 · Section 4.3: Least Squares Approximations Source: exam2_1.pdf, p. 103
Problem¶
4 (28 pts.) For each of the following statements, determine if it is always true. If so, answer yes. Otherwise, answer no. Just circle yes or no. You do not need to justify your answer. However, we will be giving 4 points for a correct answer, 0 points for no answer at all, and -2 points for an incorrect answer.
(a) Yes -- No. Let \(F\) be the vector space of all \(3 \times 3\) matrices. Let \(T\) be the transformation that maps \(A \in F\) to \(\det(A) + \operatorname{trace}(A)\). Then \(T\) is a linear transformation.
(b) Yes -- No. Let \(A\) be the \(m \times n\) (edge-node) incidence matrix of a connected graph with \(n\) vertices and \(m\) edges. Then the left nullspace of \(A\) has dimension \(m - n + 1\).
(c) Yes -- No. An orthogonal matrix can have an eigenvalue equal to 0.
(d) Yes -- No. Let \(\hat{x}\) be the least-squares solution to \(Ax = b\). Then \(b - A\hat{x}\) is orthogonal to the column space of \(A\).
(e) Yes -- No. Let \(P\) be the projection matrix for projecting onto a subspace \(F\). Then \(I - P\) is the projection matrix for projecting onto \(F^\perp\).
(f) Yes -- No. There are values for \(a, b, c, d, e, f\) such that the following matrix \(A\) satisfies \(A^2 = 2A\):
(g) Yes -- No. Let \(0\) denote the \(5 \times 5\) matrix whose entries are all 0. If \(A^{10} = 0\) then \(A = 0\).
Solution
(a) No. \(T(2I) \neq 2T(I)\), for example. (b) Yes. The rank of the edge-node incidence matrix of a connected graph with \(n\) vertices is \(n - 1\), so the left nullspace has dimension \(m - (n - 1) = m - n + 1\). (c) No. An orthogonal matrix is non-singular: its columns are perpendicular and hence certainly independent, so none of its eigenvalues can be 0. (d) Yes. By definition of the least-squares solution, the residual \(b - A\hat{x}\) is orthogonal to the column space of \(A\). (e) Yes. The sum of the projections must be the identity: every \(b\) decomposes as \(b = Pb + (I - P)b\) with \(Pb\) in \(F\) and \((I - P)b\) in \(F^\perp\). (f) No. Multiplying by any eigenvector, and then considering a non-zero entry of it and dividing out, every eigenvalue must satisfy \(\lambda^2 = 2\lambda\), i.e., \(\lambda = 0\) or 2. But the sum of the eigenvalues of \(A\) is \(\operatorname{trace}(A) = 3\), which cannot be obtained by adding 0's and 2's. (g) No. Counterexample: \(A = 0\) except for \(A_{1,2} = 1\). Then \(A^2 = 0\), so \(A^{10} = 0\), but \(A \neq 0\).Keywords: true or false, linear transformation, least squares, projection, eigenvalues