As usual, we write the EPR pair as $$ \frac{1}{\sqrt2}(\left|00\right> + \left|11\right>). $$

A property of the EPR pair is that this definition is basis-independent, which means $$ \frac{1}{\sqrt2}(\left|00\right> + \left|11\right>) = \frac{1}{\sqrt2}(\left|uu\right> + \left|vv\right>) $$ as long as $\left|u\right>$ and $\left|v\right>$ are orthogonal in $\mathbb{R}^2$. (As pointed out by @ElioFabri, if we set $\left|u\right>=\left|0\right>$ and $\left|v\right>=i\left|1\right>$ then the equailty no longer holds. Hence we cannot put them in $\mathbb{C}^2$.)

Question: Is there any way to explain this other than an elementary direct computation?

  • $\begingroup$ A property of the EPR pair is that this definition is basis-independent I can't see it's true. If $|u\rangle=|0\rangle$ and $|v\rangle=i\,|1\rangle$ your identity doesn't hold. $\endgroup$
    – Elio Fabri
    Feb 21, 2019 at 14:00
  • $\begingroup$ @ElioFabri I think we need orthogonal basis. $\endgroup$
    – Lwins
    Feb 21, 2019 at 14:02
  • $\begingroup$ I think we need orthogonal basis. Sure. But $\{|0\rangle, i\,|1\rangle\}$ is orthogonal. $\endgroup$
    – Elio Fabri
    Feb 21, 2019 at 14:05
  • $\begingroup$ @ElioFabri You are right. I've modified the OP. $\endgroup$
    – Lwins
    Feb 21, 2019 at 14:43
  • $\begingroup$ as long as $|u\rangle$ and $|v\rangle$ are orthogonal in R2. I'm not sure how to interpret that. Do you mean that a real Hilbert space should be used? I would suggest the following. Starting from a base $\{|0\rangle, |1\rangle\}$, in order to keep the EPR pair invariant you may only use new bases obtained from the former through a real orthogonal matrix. $\endgroup$
    – Elio Fabri
    Feb 22, 2019 at 13:57

1 Answer 1


Here's a neat way to see this.

The main element is vectorization. Vectorization of a matrix $A=\sum A_{i,j}\vert i \rangle\langle j\vert$ is defined by $\vert\text{vec}({A})\rangle = \sum A_{i,j}\vert i \rangle\vert j\rangle$. Essentially, a bra turns into a ket. The cute thing about this is that you've now established a correspondence between matrices and bipartite pure states.

Next, consider what the identity matrix vectorizes to. You have $\sum_i \vert i \rangle\langle i\vert \rightarrow \sum_i \vert i \rangle\vert i\rangle $, which is an unnormalized Bell state. Indeed, if you chose to do it in two dimensions and the computational basis, you have your state $\vert 0\rangle\langle 0\vert + \vert 1\rangle\langle 1\vert \rightarrow \vert\Phi^+\rangle = \vert 00 \rangle + \vert 11\rangle$ upto normalization. You might be interested in the other Pauli matrices and what they do. I leave it to you to work it out but they simply give you the other Bell states.

But wait, you're free to write the identity as $\sum_i\vert i \rangle\langle i\vert$ in any orthonormal basis. So indeed you can have $i = \{0,1\}$ or $i = \{+,-\}$ or as you wanted to write it, $i = \{u,v\}$ for some $\vert u\rangle \perp \vert v\rangle$.

In some sense, the intuition is that because the identity matrix vectorizes to the $\vert\phi^+\rangle$ state, the freedom in representing the identity using any orthonormal basis carries over to the freedom in representing the vectorized state.

  • $\begingroup$ Elegant! I think there’s only one (maybe trivial) thing I cannot see through. That is, it is not obvious to me that the vectorization is base-independent. Would you please give me some hints? $\endgroup$
    – Lwins
    Feb 20, 2019 at 19:38
  • $\begingroup$ I'm not sure if that statement is true. What I can tell you is that there is a vectorization identity $A\otimes B^T\vert\text{vec}(X)\rangle = \vert\text{vec}(AXB)\rangle$. What you are doing when you go from $\vert 00\rangle + \vert 11\rangle$ to $\vert uu\rangle + \vert vv\rangle$ is to apply the unitary $U\otimes U^*$ (note the conjugate!) on $\vert\text{vec}(I)\rangle$. But when you apply the vectorization identity, you get $\vert\text{vec}(UIU^\dagger)\rangle = \vert\text{vec}(I)\rangle$. I'm not sure if there's a more general argument. $\endgroup$
    – rnva
    Feb 20, 2019 at 21:41
  • 1
    $\begingroup$ @nr2618 I'm not sure if that statement is true. It can't as vectorization isn't a linear map, turning a bra into a ket. $\endgroup$
    – Elio Fabri
    Feb 21, 2019 at 14:01

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.