5.2-final-2012-b¶
Final Exam, 2012 · Section 5.2: Computing and Using Determinants Source: finalexam_2.pdf, pp. 66-67
Problem¶
6 (20 pts.) This problem concerns matrices whose entries are taken from the values +1 and -1. In other words, the general form of this matrix is
We will call these matrices \(\pm 1\) matrices.
One \(3 \times 3\) example of such a matrix is
(a) (5 pts.) Find a two by two example of a \(\pm 1\) matrix with eigenvalues \(\lambda_1 = 0\) and \(\lambda_2 = 0\) or prove it is impossible.
(b) (5 pts.) Suppose \(A\) is a \(10 \times 10\) example of a \(\pm 1\) matrix. Compute \(\sigma_1^2 + \sigma_2^2 + \dots + \sigma_{10}^2\).
(c) (5 pts.) The big determinant formula for a \(5 \times 5\) \(A\) has exactly __ terms. A computer package for matrices computes that the determinant of a \(\pm 1\) matrix that is \(5 \times 5\) is an odd integer. If this is possible exhibit such a \(\pm 1\) matrix, if not argue clearly why the package must not be giving the right answer for this \(5 \times 5\) matrix.
(d) (5 pts.) For every \(n\), construct a \(\pm 1\) matrix \(A_n\) with \((n - 1)\) eigenvalues exactly equal to 2. (Hint: Think about \(A_n - 2I\).)
Solution
(a) The matrix \[ \begin{bmatrix} 1 & 1 \\ -1 & -1 \end{bmatrix} \] has trace 0 and determinant 0, so its eigenvalues are \[ \lambda_1 = \lambda_2 = 0. \] (b) The squared singular values of a matrix are the eigenvalues of \(A^T A\), so \[ \sigma_1^2 + \sigma_2^2 + \dots + \sigma_{10}^2 = \mathrm{trace}(A^T A) = \text{sum of the squares of all 100 entries}. \] Every entry of a \(\pm 1\) matrix squares to 1, so the answer is \[ \sigma_1^2 + \sigma_2^2 + \dots + \sigma_{10}^2 = 100. \] (c) The big determinant formula has \[ 5! = 120 \] terms. It is not possible for the determinant of a \(5 \times 5\) \(\pm 1\) matrix to be odd: every term in the big formula is a product of 5 entries, hence equals \(\pm 1\) and is odd, but the sum of an even number (120) of odd integers is always even. The package must be wrong. (d) Let \(E_n\) be the \(n \times n\) all-ones matrix and take \[ A_n = 2I_n - E_n. \] The diagonal entries of \(A_n\) are \(2 - 1 = 1\) and the off-diagonal entries are \(0 - 1 = -1\), so \(A_n\) is indeed a \(\pm 1\) matrix. The matrix \(E_n\) has eigenvalues \(n\) (for the all-ones eigenvector) and 0 (multiplicity \(n - 1\)), so \(A_n = 2I_n - E_n\) has eigenvalues \[ 2 - n, \underbrace{2, 2, \dots, 2}_{n-1}. \] Thus exactly \(n - 1\) eigenvalues of \(A_n\) are equal to 2. (This matches the hint: \(A_n - 2I = -E_n\) has rank 1, so \(\lambda = 2\) has an \((n - 1)\)-dimensional eigenspace.)Keywords: sign matrix, big determinant formula, singular values, eigenvalues