Skip to content

6.3-final-2011-strang

Final Exam, 2011 (strang) · Section 6.3: Symmetric Positive Definite Matrices Source: finalexam_2.pdf, p. 54

Problem

7. Suppose \(A\) is a positive definite symmetric matrix with \(n\) different eigenvalues: \(Ax_i = \lambda_i x_i\).

(a) What are the properties of those \(\lambda\)'s and \(x\)'s? How would you find an orthogonal matrix \(Q\) so that \(A = Q\Lambda Q^T\) with \(\Lambda = \mathrm{diag}(\lambda_1, \dots, \lambda_n)\)?

(b) I am looking for a symmetric positive definite matrix \(B\) with \(B^2 = A\) (a square root of \(A\)). What will be the eigenvectors and eigenvalues of \(B\)? Can you find a formula for \(B\) using \(Q\) and \(\Lambda\) from part (a)?

(c) What are the eigenvalues and eigenvectors of the matrix \(e^{-A}\)? Is this matrix also positive definite and why?

Solution (a) Since \(A\) is symmetric with distinct eigenvalues, the \(\lambda_i\) are real and positive (positive definiteness), and the \(x_i\) are mutually orthogonal; after normalization they are orthonormal. Take \(Q\) to have columns \(x_i/\|x_i\|\). Then \[ A = Q \Lambda Q^T, \qquad \Lambda = \mathrm{diag}(\lambda_1, \dots, \lambda_n). \] (b) The eigenvectors of \(B\) are the same \(x_i\), and the eigenvalues of \(B\) are \(\sqrt{\lambda_i} > 0\). The formula is \[ B = Q \Lambda^{1/2} Q^T, \qquad \Lambda^{1/2} = \mathrm{diag}(\sqrt{\lambda_1}, \dots, \sqrt{\lambda_n}). \] Then \(B^2 = Q\Lambda^{1/2}Q^T Q\Lambda^{1/2}Q^T = Q\Lambda Q^T = A\), and \(B\) is symmetric positive definite because \(\sqrt{\lambda_i} > 0\). (c) The eigenvalues of \(e^{-A}\) are \(e^{-\lambda_i}\) and its eigenvectors are the same \(x_i\): \[ e^{-A} = Q e^{-\Lambda} Q^T, \qquad e^{-\Lambda} = \mathrm{diag}(e^{-\lambda_1}, \dots, e^{-\lambda_n}). \] This matrix is positive definite: it is symmetric, and every eigenvalue \(e^{-\lambda_i}\) is positive.

Keywords: positive definite, orthogonal diagonalization, square root, matrix exponential