We have the matrix \begin{equation*}A=\begin{pmatrix}7 & \frac{1}{4} & \frac{1}{2} \\ \frac{1}{4} & -5 & 1 \\ \frac{1}{2} & 1 & \frac{3}{2}\end{pmatrix}\end{equation*}

with the eigenvalues $\lambda_1 < \lambda_2 < \lambda_3$.

Using the Gershgorin lemma I want to estimate the smallest eigenvalue $\lambda_1$ (using the smallest center of a Gershgorin circle).

The circles are \begin{align*}&K_1=\left \{|x-7|\leq \left |\frac{1}{4}\right |+\left |\frac{1}{2}\right |\right \}=\left \{|x-7|\leq 0.75\right \} \\ &K_2=\left \{|x-(-5)|\leq \left |\frac{1}{4}\right |+|1|\right \}=\left \{|x+5|\leq 1.25\right \} \\ &K_3=\left \{\left |x-\frac{3}{2}\right |\leq \left |\frac{1}{2}\right |+|1|\right \}=\left \{\left |x-\frac{3}{2}\right |\leq 1.5\right \} \end{align*}

Therefore, we get \begin{align*} -6.25\leq &\lambda_1 \leq -3.75 \\ 0\leq &\lambda_2\leq 3 \\ 6.25\leq &\lambda_3 \leq 7.75 \end{align*} right?

Next, I want to make a shift around this estimate and calculate approximately the eigenvalue using the Inverse Iteration (3 steps) with initial vector $x^{(0)} = (−1,0,0)^T$.

We have to consider the matrix $(A-sI)^{-1}$ and apply the power method, or not?But which $s$ do we have to take? Do we maybe consider as $s$ a number in the interval $[-6.25, -3.75]$, i.e. the interval of $\lambda_1$ ?


When you choose s to to be within domain $K_1$, the inverse powermethod will converge to the eigenvektor of the eigenpair $(\lambda_1, v_1)$. The closer s to the actual value of $\lambda_1$ is, the faster is the convergence.

  • $\begingroup$ The eigenvalues are: wolframalpha.com/input/… So, can we get for example $s=-5$ ? $\endgroup$ – Mary Star Jan 19 '18 at 16:31
  • $\begingroup$ if you select s=-5, you should converge to the eigenvector (0.0967097, -6.70119, 1) $\endgroup$ – Waterhouse Jan 19 '18 at 16:35
  • $\begingroup$ So, we have to find the inverse matrix $(A+5I)^{-1}$ and then apply the power method for the matrix $(A+5I)^{-1}$ and the initial vector $x^{(0}$, right? $\endgroup$ – Mary Star Jan 19 '18 at 18:23
  • $\begingroup$ You choose a $v_0$, an initial guess for the eigenvector ( not [0,0,0] ) and then you iterate as following: $v_{i+1}= \frac{(A+5I)^{-1}v_{i}}{||(A+5I)^{-1}v_{i}||} $ and you should compute the inverse once, not for every iteration. $\endgroup$ – Waterhouse Jan 19 '18 at 22:28
  • $\begingroup$ As initial vector it is given the $x^{(0)}=(-1,0,0)^T$. What do you mean by "you should compute the inverse once, not for every iteration" ? I have calculated (if I am not mistaken) $$\|(A+5I)^{-1}x^{(3)}\|_{\infty}=\frac{9255292892}{1414521867}\approx 6.54$$ Can we conclude from that something about the eigenvalues of $A$? Does it maybe hold that the eigenvalue of $A+5I$ will then be $\frac{1}{6.5}$ and the eigenvalue of $A$ is $\lambda$ such that $|\lambda+5|=\frac{1}{6.5}$ ? $\endgroup$ – Mary Star Jan 19 '18 at 23:13

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.