Expressing $C(x) = \tilde{x} = (\langle x,a_i \rangle )$ as a product of matrices in the form $Cx = \tilde{x}$

74 Views Asked by At

Le that $(a_i)^{n}_{i=1}$ be an orthonormal basis and $C(x)$ be a transformation defined as follows:

$$C(x) = \tilde{x} = \left( \begin{array}{c} \langle x, a_1 \rangle\\ \vdots \\ \langle x, a_k \rangle\\ \vdots \\ \langle x, a_n \rangle\\ \end{array} \right)$$

is there a way to express that (linear) function with matrix multiplication?

Note that $a_i$ makes an orthonormal basis and each $a_i \in \mathbb{C}^d$ and $\langle a, b \rangle = a^*b = a^H b$.

i.e. what I am looking for looks something like this:

$$C(x) = Cx = \tilde{x}$$

where my intuition says C should be a unitary matrix in terms of the basis $a_i$.


This is what I have tried:

It seems to me that the only way to do this is to define C as follows. Let the columns of C be $a_i$. Then take the matrix product of x and the hermitian C i.e. let C(x) = :

$$C^*x = \tilde{x}$$

which doesn't seem entirely correct to me.

Let me tell you why I think its not right. First in this context, $\langle a, b \rangle = a^Hb = a^*b$ and hence $\langle a, b \rangle \neq \langle b, a \rangle $ in general.

$$C^*x = = \left( \begin{array}{c} a_1\\ \vdots \\ a_k\\ \vdots \\ a_n\\ \end{array} \right) x = \left( \begin{array}{c} \langle a_1, x \rangle\\ \vdots \\ \langle a_k, x \rangle\\ \vdots \\ \langle a_n, x \rangle\\ \end{array} \right) $$

and as I pointed out $ \langle a_k, x \rangle \neq \langle x , a_k \rangle $. So my answer is wrong ... but close to what I wanted, it just has the inner products the wrong way round.

However, it seems that in the real case $C^* = C$, so that would be right answer...no? Since in the real case the inner product $\langle a , b \rangle$ is a symmetric function.. is that right in this case?

Thought it doesn't answer my original question though.


I was expecting C(x) to have the following properties:

  1. C is a linear function
  2. it can be identified with d by d matrix with columns given by the orthonormal basis
  3. it is a unitary transformation ( $C^*C = I$)

1 seems obvious to me because we need $C(x + y) = C(x) + C(y)$ which seems clear it obeys it because of the properties of inner products.

2 and 3 are the ones I am having a hard time obtaining.


As an example, I was told the Fourier transform works with the following basis:

$a_k = \frac{1}{\sqrt{d} }(1, e^{-2 \pi i j \frac{1}{d}} , \dots, e^{-2 \pi i j \frac{g}{d}} , \dots, e^{-2 \pi i j \frac{d-1}{d}} )$

with $a_k$ as the columns, but I am skeptical because I can't make it work for a general orthonormal basis.

Any help is appreciated.

1

There are 1 best solutions below

2
On

As $\langle x, a \rangle$ is conjugate-linear in $x$, rather than linear, there's no way you can get that by a matrix multiplication, which is a linear operator.