0
$\begingroup$

Let $\alpha=(m_{0}(p\xi+pu(0)),m_{0}(p\xi+pu(1)),...,m_{0}(p\xi+pu(q-1)) )^T$ and

$\mathbb{M}(\xi)=I_{q}-\alpha\overline{\alpha }^{T}=(\beta_{1},\beta_{2},...,\beta_{q})$

enter image description here

How these eigenvalues of the matrix are obtained?

If we further assume that:

$$\sum_{k=0}^{q-1}\left | m_{0}(p\xi+pu(k)) \right |^2\leq 1$$

Then the unit eigen-vectors of the matrix $\mathbb{M}(\xi)$ can be represented by enter image description here enter image description here

How these unit eigen-vectors are obtained?

$\endgroup$
  • 1
    $\begingroup$ What are your Full attempts ? What I can say is the following: if we where with real components, expression $I-\alpha\alpha^T$ is a projection on the hyperplane defined normal to vector $\alpha$, thus an eigenvalue 1 (associated with eignevector $\alpha$) and (q-1) eigenvalues 0 (asociated with a basis of vectors orthogonal to $\alpha$). This can be slightly different with complex coordinates. $\endgroup$ – Jean Marie Apr 10 '16 at 19:54
  • 1
    $\begingroup$ @JeanMarie: $\alpha$ is not necessarily a unit vector ... $\endgroup$ – Catalin Zara Apr 12 '16 at 14:46
  • 1
    $\begingroup$ @JeanMarie, why is it slightly different with complex coordinates? We're just doing orthogonal projections in a Hilbert space. $\endgroup$ – Jon Warneke Apr 13 '16 at 13:41
  • 1
    $\begingroup$ @Catalin Zara and Jon Warneke : I agree $\endgroup$ – Jean Marie Apr 13 '16 at 15:27
2
$\begingroup$

We know $\alpha$ is an eigenvector of $\mathbb{M}$ by $$ \mathbb{M}\alpha=\alpha - \lvert\alpha\rvert^2 \alpha=(1-\lvert\alpha \rvert^2) \alpha, $$ as long as $\alpha \neq 0$. We note that $\delta_q=\frac{\alpha}{\lvert \alpha\rvert}$.

Further we know, if we have a vector $\eta$ with $\bar{\alpha}^\top\eta=0$ we get $$ \mathbb{M}\eta=\eta - \alpha\cdot 0=1\cdot\eta$$ thus, all vectors $\eta$ that are orthogonal to $\alpha$ are eigenvectors of $\mathbb{M}$ to the eigenvalue $1$.

Additionally, $\mathbb{M}$ is hermitian, i.e. $\bar{\mathbb{M}}^\top=I-\alpha\bar{\alpha}^\top=\mathbb{M}$. Hence, we know that $\mathbb{M}$ is diagonalizable with orthonormal eigenvectors.

Now we investigate, how to define an orthogonal basis for the eigenspace of the eigenvalue $1$, that is also orthogonal to the eigenvector $\alpha$ of the eigenvalue $1-\lvert\alpha \rvert^2$.

For simplicity, we say $$\alpha=\begin{pmatrix} a \\ b \\ c \\ d\end{pmatrix}.$$ Then the vector $$ \beta = \begin{pmatrix} -\bar{b} \\ \bar{a} \\ 0 \\ 0\end{pmatrix} $$ is orthogonal to $\alpha$. Now we need a third vector that is orthogonal to $\alpha$ and $\beta$. For $\beta$ it is easy, we just take the first two entries of $\alpha$ again $$\hat{\gamma}=\begin{pmatrix} a \\ b \\ \gamma_3 \\ 0\end{pmatrix} $$ but of course $\hat{\gamma}$ is not longer orthogonal to $\alpha$ and we get $$ \bar{\alpha}^\top\hat{\gamma}=\lvert a\rvert^2 +\lvert b\rvert^2 +\bar{c}\gamma_3.$$ Now we have an idea for $\gamma$ $$ \gamma=\begin{pmatrix} a\bar{c} \\ b\bar{c} \\ -(\lvert a\rvert^2 +\lvert b\rvert^2) \\ 0\end{pmatrix} $$ The next (and last vector in this simplified case) has to be orthogonal to the first three, thus we use $$ \delta=\begin{pmatrix} a\bar{d} \\ b\bar{d} \\ c\bar{d}\\ -(\lvert a\rvert^2 +\lvert b\rvert^2+\lvert c\rvert^2)\end{pmatrix}.$$

Now we can divide them by their norm to get an orthonormal basis (see $\Omega_k$).

I hope this clarified one idea, to get an orthogonal basis from a vector.

On the other hand, the vectors $\delta_1$ and $\delta_q$ in your question are not orthogonal, maybe just a typo.

$\endgroup$
  • 1
    $\begingroup$ Edit: added something like 'why are vectors orthogonal to $\alpha$ eigenvectors of $\mathbb{M}$' $\endgroup$ – Hajo Apr 13 '16 at 14:24
  • $\begingroup$ what do you mean by "we know that $\mathbb{M}$ is diagonalizable with orthonormal eigenvectors." ? $\endgroup$ – Ehsan zarei Apr 14 '16 at 5:59
  • 1
    $\begingroup$ @ehsan a hermitian matrix is a normal matrix and normal matrices have this property $\endgroup$ – Hajo Apr 14 '16 at 7:26

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.