4
$\begingroup$

Let $A_1$ and $A_2$ be two $3\times 3$ hermitian matrices. Then their numerical range is defined as two-dimensional set

\begin{align} \mathbb{S}=\{\left[u^HA_1u,u^HA_2u\right]\in \mathbb{R}^2,u^Hu=1\} \end{align}

where $u$ is a unit-norm vector. Each of the $u^HA_iu,~i=1,2$ is a real number. Thus for a given $u$, I calculate the first co-ordinate as $u^HA_1u$ and second co-ordinate as $u^HA_2u$, then I have to vary $u$ through all possible unit-norm vectors. Thus $\mathbb{S}$ becomes a 2-D set.

EDIT If it is possible to find a solution for the real case, that is fine. But then the matrices $A_1$ and $A_2$ will be of size $6 \times 6$. The reason is you can convert the above set of equations in terms of real variables and real matrices in higher dimensions (twice the dimension to be exact).

Improve this question
$\endgroup$
2
  • $\begingroup$ What are the square brackets ? do you take u to be complex ? $\endgroup$
    – b.gates.you.know.what
    Apr 30, 2013 at 9:38
  • $\begingroup$ @b.gatessucks Please see the edit. Yes I take $u$ as complex, but if you know solution for real, that is fine as well $\endgroup$
    – dineshdileep
    Apr 30, 2013 at 9:41

1 Answer 1

Reset to default
7
$\begingroup$

If you consider only real unit vectors then one can parametrize them in spherical coordinates u = {Sin[t] Cos[f], Sin[t] Sin[f], Cos[t]}; this way you can then use ParametricPlot.

Example :

SeedRandom[6]
a1 = With[{tmp = RandomReal[{-1, 1}, {3, 3}] + I RandomReal[{-1, 1}, {3, 3}]}, 
         0.5 (tmp + ConjugateTranspose[tmp])];
a2 = With[{tmp = RandomReal[{-1, 1}, {3, 3}] + I RandomReal[{-1, 1}, {3, 3}]}, 
         0.5 (tmp + ConjugateTranspose[tmp])];

range[t_, f_] = {Dot[{Sin[t] Cos[f], Sin[t] Sin[f], Cos[t]}, a1, {Sin[t]Cos[f], Sin[t] Sin[f], Cos[t]}], 
                 Dot[{Sin[t] Cos[f], Sin[t] Sin[f], Cos[t]}, a2, {Sin[t] Cos[f], Sin[t] Sin[f], Cos[t]}]};

(* doubling the t-range to make the plot better *)
ParametricPlot[range[t, f], {t, 0, 2 Pi}, {f, 0, 2 Pi}, Mesh -> None]

enter image description here

Improve this answer
$\endgroup$
7
  • $\begingroup$ +1, nice!!. The reason I said real is fine is because, if one can do it for complex, then one can do it for real as well. If your complex matrix is $N\times N$, then you can convert it into a equation in real variables, but then matrix will be of size $2N \times 2N$. Thus in my case, $A_1$ and $A_2$ will be $6\times 6$. Is there any way to extend this solution to real case? $\endgroup$
    – dineshdileep
    Apr 30, 2013 at 15:14
  • $\begingroup$ @dineshdileep In the complex case you'd have 5 independent variables and I don't know of any way of handling that continuosly. You could generate a big list of points and then use ListPlot but the result will not be as nice. $\endgroup$
    – b.gates.you.know.what
    Apr 30, 2013 at 15:30
  • $\begingroup$ also, why is there some dark, more colored portion inside that shape. Is it because of complex part? $\endgroup$
    – dineshdileep
    Apr 30, 2013 at 16:53
  • $\begingroup$ I think it's just an artifact of the plotting routine; you can see a "skeleton" of the continuous plot with ParametricPlot[range[t, f], {t, 0, 2 Pi}, {f, 0, 2 Pi}, Mesh -> None]. However, hopefully some of the experts will have a better answer. $\endgroup$
    – b.gates.you.know.what
    Apr 30, 2013 at 17:43
  • $\begingroup$ When I try $EigenValues[a1]$ after executing your code, I don't get real numbers, why is that so? $\endgroup$
    – dineshdileep
    May 1, 2013 at 2:38

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

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