6.3-final-2009¶
Final Exam, 2009 · Section 6.3: Symmetric Positive Definite Matrices Source: finalexam_1.pdf, p. 232
Problem¶
Problem 6: Consider the matrix
(i) \(A\) has one eigenvalue \(\lambda = -1\), and the other eigenvalue is a double root of \(\det(A - \lambda I)\). What is the other eigenvalue? (Very little calculation required.)
(ii) Is \(A\) defective? Why or why not?
(iii) Using the above \(A\), suppose we want to solve the equation
where \(c\) is some real number, for some initial condition \(u(0)\).
(a) For what values of \(c\) will the solutions \(u(t)\) always go to zero as \(t \to \infty\)?
(b) For what values of \(c\) will the solutions \(u(t)\) typically diverge (\(\|u(t)\| \to \infty\)) as \(t \to \infty\)?
(c) For what values of \(c\) will the solutions \(u(t)\) approach a constant vector (possibly zero) as \(t \to \infty\)?
Solution
(18 points = 6 + 6 + 6 (2 + 2 + 2)) (i) Let \(\lambda_1 = -1\) and let \(\lambda_2 = \lambda_3\) denote the double root. Then from the trace of \(A\), we have \(\lambda_1 + 2\lambda_2 = \text{trace}(A) = 3\). Hence, \(\lambda_2 = 2\). (ii) \(A\) is not defective. There are two ways to see it. For one way, since \(A\) is symmetric, it is always non-defective; for another way, we compute \(A - \lambda_2 I = \begin{bmatrix} -1 & -1 & -1 \\ -1 & -1 & -1 \\ -1 & -1 & -1 \end{bmatrix}\), which has rank 1 and hence its nullspace is 2-dimensional. (iii) The key point here is that \(A + cI\) would have eigenvalues \(\lambda_1 + c\) and \(\lambda_2 + c\) (with multiplicity 2). An alternative point of view is as follows. If we write the initial condition \(u(t) = c_1(t)v_1 + c_2(t)v_2 + c_3(t)v_3\), then the differential equation becomes \[ \frac{dc_1(t)}{dt} v_1 + \frac{dc_2(t)}{dt} v_2 + \frac{dc_3(t)}{dt} v_3 = c_1\lambda_1 v_1 + c_2\lambda_2 v_2 + c_3\lambda_3 v_3 + cc_1 v_1 + cc_2 v_2 + cc_3 v_3. \] We have \[ \left\{ \begin{aligned} \frac{dc_1(t)}{dt} &= c_1\lambda_1 + cc_1, \Rightarrow c_1 = e^{(\lambda_1+c)t}; \\ \frac{dc_2(t)}{dt} &= c_2\lambda_2 + cc_2, \Rightarrow c_2 = e^{(\lambda_2+c)t}; \\ \frac{dc_3(t)}{dt} &= c_3\lambda_3 + cc_3, \Rightarrow c_3 = e^{(\lambda_3+c)t}. \end{aligned} \right. \] (a) If we require \(u(t)\) to always go to zero as \(t \to \infty\), \(\lambda_1 + c < 0\) and \(\lambda_2 + c = \lambda_3 + c < 0\). Hence, we require \(c < -2\). (b) If the solution \(u(t)\) typically diverges, we need either \(\lambda_1 + c > 0\) or \(\lambda_2 + c = \lambda_3 + c > 0\). Hence, we require \(c > -2\). (c) If we allow the solution to approach some constant vector, we allow the extreme case of (a), that is to say \(c \le -2\).Keywords: symmetric matrix, eigenvalues, stability, differential equations