A general Bell state:

$|\beta(a,b)\rangle = \frac{1}{\sqrt{2}}[|0,0 \oplus b\rangle + (-1)^{a}|1,1 \oplus b \rangle]$

$|\beta(0,0)\rangle = \frac{1}{2}[|00\rangle \langle 00| + |00\rangle \langle 11| + |11\rangle\langle00| + |11\rangle\langle11|]$

$|\beta(0,1)\rangle = \frac{1}{2}[|01\rangle\langle01| + |01\rangle\langle10| + |10\rangle\langle01| + |10\rangle\langle10|]$

$|\beta(1,0)\rangle = \frac{1}{2}[|00\rangle \langle00| - |00\rangle \langle11| - |11\rangle\langle00| + |11\rangle \langle11|]$

$|\beta(1,1)\rangle = \frac{1}{2}[|01\rangle \langle01| - |01\rangle\langle10| - |10\rangle\langle01| + |10\rangle\langle10|]$

Using the above one can mix the four bell state to arrive at $\rho = \sum_{a,b}p_{ab}|\beta(a,b)\rangle\langle\beta(a,b)|$ where $\sum_{ab}p_{ab} = 1$, $p_{ab} \geq 1$

Few questions:

  1. Each Bell state is a density matrix. If I take the expression for $\rho$: the trace of $\rho$ is no longer 1 - it fails the density operator test. What is going on?

  2. I want to determine if $\rho$ is entangled: Since $\rho$ is mixed by hypothesis, I can take the positive partial transposition of $\rho$ with respect to $b$. This is equivalent to taking the positive partial transpose of each of the Bell states above and sum them. Then compute the eigenvalues. If the eigenvalues are negative, $\rho$ is an entangled state. Is this correct?

extension to 2. I am informed that $|\beta((a,b)\rangle\ \langle\beta(a,b)| = \frac{1}{2}-|\beta(a,b)\rangle \langle\beta(a,b)||$ together with

$|\psi_{-}\rangle\langle\psi_{-}|^{T_{B}} = \frac{1}{2} - |\phi_{+}\rangle \langle \phi_{+}|$, $|\psi_{+}\rangle \langle \psi_{+}|^{T_{B}} = \frac{1}{2}-|\phi_{-}\rangle \langle \phi_{-}|$, $|\phi_{+}\rangle \langle \phi_{+}|^{T_{B}} = \frac{1}{2}-|\psi_{-}\rangle\langle\psi_{-}|$, $|\phi_{-}\rangle \langle \phi_{-}|^{T_{B}} = \frac{1}{2}-|\psi_{+}\rangle \langle \psi_{+}|$.

How does this relates to positive partial transposition and entanglement?

  • $\begingroup$ Why do you say the trace is not 1? Can you show us your calculation? So long as $\sum p_{ab}=1$, it should be 1. $\endgroup$
    – DaftWullie
    Commented Oct 25, 2023 at 9:42
  • $\begingroup$ @DaftWullie I am getting 2 probability coefficients along the diagonal for each Bell state matrix. $\endgroup$
    – librarian_
    Commented Oct 25, 2023 at 10:08
  • $\begingroup$ Yes, and each will be $p_{ab}/2$, so they add to $p_{ab}$, and when you sum over all $a,b$, you get $\sum p_{ab}=1$. $\endgroup$
    – DaftWullie
    Commented Oct 25, 2023 at 10:11
  • $\begingroup$ @DaftWullie You are right. I was an idiot for missing the $\frac{1}{2}$ $\endgroup$
    – librarian_
    Commented Oct 25, 2023 at 10:12

1 Answer 1


In answer to 2: Yes, this is correct. More carefully stated: it at least one eigenvalue is negative, the state is entangled. If all eigenvalues are non-negative, the state is separable.

In terms of doing the calculation, it looks like you've been told $$ |\beta(x,y)\rangle\langle\beta(x,y)|^{T_B}=\frac12-|\beta(\bar x,\bar y)\rangle\langle\beta(\bar x,\bar y)|. $$ Thus, $$ \rho^{T_B}=\frac12-\sum_{a,b}p_{\bar a,\bar b}|\beta(a,b)\rangle\langle\beta(a,b)|. $$ Your states $|\beta(a,b)\rangle$ are eigenvectors, so you can read off the eigenvalues. For example, $|\beta(0,0)\rangle$ is an eigenvector with eigenvalue $\frac12-p_{11}$. Thus, there is entanglement if $p_{11}>\frac12$. There's a lot of symmetry here, so the condition on entanglement becomes $\max_{a,b}p_{a,b}>\frac12$.

  • $\begingroup$ How does one arrive at this expression for the generalised Bell state matrix? Could you provide an example of reading off the eigenvalues here? $\endgroup$
    – librarian_
    Commented Oct 25, 2023 at 10:44
  • $\begingroup$ Do you mean eigenvalue $\frac{1}{2} - p_{00}$? I couldn't even verify the claim made by the OP regarding the partial transpose is correct. $\endgroup$
    – Physkid
    Commented Nov 3, 2023 at 15:18
  • $\begingroup$ No, I meant what I wrote ($x$ going to $\bar x$ is important). To verify the partial transpose claim, you can just write out the states one at a time. Personally, I like to use Pauli matrices: $|\beta(x,y)\rangle\langle\beta(x,y)|=\frac14(I+(-1)^yZZ+(-1)^xXX+(-1)^{x+y}YY)$. Since $X^T=X$, $Z^T=Z$ and $Y^T=-Y$, we can evaluate how these change very easily (without having to go through all $x$ and $y$ values separately). $\endgroup$
    – DaftWullie
    Commented Nov 3, 2023 at 15:35

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.