I want to prove that if $ \langle \psi | A | \psi \rangle = 1$ for all $ \psi ,$ then $A=I .$

Let's write $A$ and $\psi$ in the same basis.

$$ \begin{alignat}{7} \left\langle \psi \middle| A \middle| \psi \right\rangle & ~=~ && \left( \sum_w \alpha_w^* \langle w | \right) \left(\sum_{pq} \gamma_{pq} |p\rangle \langle q| \right) \left(\sum_v \alpha_v | v \rangle \right) \\[5px] & ~=~ && \sum_{wvpq} \alpha_w^* \gamma_{pq} \alpha_v \langle w | p \rangle \langle q | v \rangle \\[5px] & ~=~ && \sum_{wv} \alpha_w^* \gamma_{wv} \alpha_v \end{alignat} $$

which is equal to $1 .$

We know that the $\alpha$'s can be anything, and we need to prove that the $\gamma$'s where $w = v$ are $1$ and when $w \neq v$ is $0 .$

How do I proceed?

  • 1
    $\begingroup$ Can we prove it in the opposite direction (i.e. contra-positive)? if $A \neq I$, then $\langle \psi | A | \psi \rangle \neq 1$ for some $\psi$ $\endgroup$ – K_inverse Jan 21 at 8:37
  • 1
    $\begingroup$ You say the $\alpha$'s can be anything but surely they need to be normalised? $\endgroup$ – jacob1729 Jan 21 at 10:17
  • $\begingroup$ @jacob1729, true, they need to be normalised $\endgroup$ – Mahathi Vempati Jan 21 at 10:18
  • $\begingroup$ @MahathiVempati if $\alpha$'s are normalised then one (long winded) way of writing the number 1 is $1=\sum \alpha^*_w \delta_{wv}\alpha_v$. $\endgroup$ – jacob1729 Jan 21 at 11:31

We have the freedom to expand $A$ in its eigenbasis, then \begin{equation} A = \sum_{\alpha} \vert\alpha\rangle \langle \alpha\vert A\vert\alpha\rangle \langle \alpha\vert = \sum_{\alpha} \vert\alpha\rangle \langle \alpha\vert = 1, \end{equation} where in the second equality we used that by assumption $\langle \alpha\vert A\vert\alpha\rangle=1$.

  • $\begingroup$ Thank you! Is it possible to continue my line of proof, though? $\endgroup$ – Mahathi Vempati Jan 21 at 9:12
  • 3
    $\begingroup$ This is slick but assumes that $A$ can be diagonalized.. $\endgroup$ – lcv Jan 21 at 19:19
  • $\begingroup$ @MahathiVempati I sort of addressed your comment. $\endgroup$ – lcv Jan 22 at 9:03

Loewe's proof is very slick but assumes that $A$ can be diagonalized. Here is a proof with no such restriction.

First, setting $B=A-I$ the statements becomes

$$ \langle \psi, B \psi \rangle=0 ,\ \ \forall \parallel \psi \parallel=1 \Rightarrow B = 0. $$

By going to a basis, reasoning as the OP, one obtains

$$ \sum_{i,j} \psi_i^\ast \psi_j B_{i,j} =0 $$

from which one obtains that the hermitian part of $B$ must be zero. I.e., the statements holds if $B$ (and hence $A$) is hermitian. Loewe's proves it for the more general case for which $A$ can be diagonalized.

For full generality the trick is to realize that we also have information on the off-diagonal elements of $B$ via the polarization identity.


$$ q(\psi) = \langle \psi, B\psi \rangle $$


$$ \langle \phi, B \psi \rangle = \frac{1}{4} \sum_{n=0}^3 i^n q(\phi + i^n \psi) $$

but by assumption $q(\xi)=0$ for all $\xi$, hence $\langle \phi, B \psi \rangle =0$ for all $\phi,\psi$, hence $B=0$.


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.