4
$\begingroup$

Given an inner product, one can define a projection and a norm. Can we do the opposite?

That is, suppose we have:

  • a complex vector space V
  • a norm $|V|^2 : V \rightarrow \mathbb{R}$ such that:
    • is posite definite
    • $|\alpha V|^2 = |\alpha|^2 |V|$ with $\alpha \in \mathbb{C}$
  • a family of projections $\forall v \in V \; \exists \mathbf{P}_v : V \rightarrow V$ such that:
    • $\mathbf{P}_v ^2 = \mathbf{P}_v$
    • $\mathbf{P}_v (\alpha w_1 + \beta w_2) = \alpha \mathbf{P}_v (w_1) + \beta \mathbf{P}_v (w_2)$
    • $\forall w \in V$ $\mathbf{P}_v(w)=\alpha v$ with $\alpha \in \mathbb{C}$
    • $\mathbf{P}_v(v)=v$

Is the map $\langle \cdot , \cdot \rangle : V \times V \rightarrow \mathbb{C}$ defined such that $\mathbf{P}_v(w)=\frac{\langle v, w \rangle}{|v|^2} v$ an inner product?

$\langle \cdot , \cdot \rangle$ is positive definite. $\mathbf{P}_v(v) = v$. $\langle v, v \rangle = |v|^2$. The norm induced by the map is the original norm which is positive definite.

$\langle \cdot , \cdot \rangle$ is linear in the second argument because the projection is linear.

Conjugate symmetry is the only thing that I am missing. Is it an extra requirement, or can it be derived by the previous?

$\endgroup$
3
  • $\begingroup$ You also need some axiom on the family of projections relating the indexes, i.e., the indexes $v$ in $P_v$ should satisfy some property. Otherwise $P_v=0$ satisfies all of this but is clearly useless. Something like $|v_1|^2P_{v_1}+|v_2|^2P_{v_2}=|v_1+v_2|^2P_{v_1+v_2}$, maybe? (But not this, because again $P_v=0$ is also a counter-example). Ok, I think you also need $P_v(v)=v/|v|$, so these counter-examples don't work anymore $\endgroup$ – Luiz Cordeiro Sep 15 '16 at 20:01
  • $\begingroup$ The third condition was supposed to do that... Maybe I should just require $\mathbf{P}_v (v) = v$ more explicitly? $\endgroup$ – Carcassi Sep 15 '16 at 20:09
  • $\begingroup$ Yes, this is necessary (I made a mistake: it should've been $P_v(v)=v$, as you noted), otherwise, we have some counter-examples as above. This should also be neccessary for the inner-product to be non-degenerate. $\endgroup$ – Luiz Cordeiro Sep 15 '16 at 20:11
1
$\begingroup$

The family of projections with one-dimensional range is a red herring: It is equivalent to require a family of linear functionals (if you assume the axiom of choice). In the case of a normed space, it is equivalent to require a family of bounded projections/bounded linear functionals by Hahn-Banach (again, if you assume the axiom of choice).

Since you require nothing on the family of functionals/projections with respect to $v$, there are plenty of functionals/projections violating the conjugate symmetry.

A prominent family of examples are semi-inner products, see https://en.wikipedia.org/wiki/L-semi-inner_product (The examples given there are linear in the first argument instead of the second, but this is of course just a notational difference.)

$\endgroup$
1
  • $\begingroup$ Thanks! Counter-examples makes it a clear no. $\endgroup$ – Carcassi Feb 18 at 16:44

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.