I am trying to formulate the calculation of conditional min-entropy as a semidefinite program. However, so far I have not been able to do so. Different sources formulate it differently. For example, in this highly influential paper, it has been formulated as:

$$H_{\text{min}}(A|B)_\rho = - \underset{\sigma_B} {\text{inf}} \ D_{\infty}(\rho_{AB} \| id_A \otimes \sigma_B) $$ Where $$\rho_{AB} \in \mathcal{H_A \otimes H_B}, \sigma \in \mathcal{H_B}$$ and $$D_{\infty}(\tau \| \tau') = \text{inf} \{\lambda \in \mathbb{R}: \tau \leq 2^{\lambda} \tau' \}$$

How do I formulate it into a semidefinite program? It is possible as is mentioned in this lecture.

A possible SDP program is given in Watrous's lecture:

$$\text{maximize}: <\rho, X>$$ $$\text{subject to}$$ $$Tr_X{X} == \mathcal{1}_Y$$ $$X \in \text{Pos}(X \otimes Y)$$

How do I write it in CVX or any other optimization system?

  • 2
    $\begingroup$ What do you want to know? How to write it as an SDP, or how to solve the SDP? $\endgroup$ Commented Nov 10, 2018 at 23:28
  • $\begingroup$ How could I write it in CVX properly? $\endgroup$ Commented Nov 11, 2018 at 3:51

1 Answer 1


I think I have an answer. The following should be the CVX code for one of the formulations found in this link.

cvx_begin sdp
variable X(2, 2) hermitian

minimize(trace(id' * X)) % id is eye(2)
subject to

kron(id, X) >= rho_ab % the tensor product of two density matrices a, b
X >= 0


The optimal value found in this program is $$\text{optval} = e^{-H_{\text{min}}(A|B)}.$$ So simple calculation would solve for ${H_{\text{min}}(A|B)}$. It turns out to be pretty simple at the end, given that the theoretical foundation leading up to this solution is not quite straightforward.

  • 1
    $\begingroup$ The implementation here is not fully correct. You should replace 'variable X(2,2) semidefinite' with 'variable X(2,2) hermitian'. Otherwise you are restricting the variable X to have real-valued entries and you will only get a lower bound on Hmin (and this lower bound can be quite a bit lower than the optimum). For example, you can find CQ states for which the above program would say Hmin(X|E) < 0, which is not possible for a cq-state. $\endgroup$
    – Rammus
    Commented Jan 21, 2020 at 18:46

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.