Loading [MathJax]/extensions/TeX/boldsymbol.js

 

 

 

Discussion of Jacobi's method for eigenvalues

Defining the quantities \tan\theta = t= s/c and \cot 2\theta=\tau = \frac{a_{ll}-a_{kk}}{2a_{kl}}, we obtain the quadratic equation (using \cot 2\theta=1/2(\cot \theta-\tan\theta) t^2+2\tau t-1= 0, resulting in t = -\tau \pm \sqrt{1+\tau^2}, and c and s are easily obtained via c = \frac{1}{\sqrt{1+t^2}}, and s=tc . Convince yourself that we have |\theta| \le \pi/4 . This has the effect of minimizing the difference between the matrices \mathbf{B} and \mathbf{A} since \mathrm{norm}(\mathbf{B}-\mathbf{A})_F^2=4(1-c)\sum_{i=1,i\ne k,l}^n(a_{ik}^2+a_{il}^2) +\frac{2a_{kl}^2}{c^2}.