I would like to know how one finds the Kraus operators of a channel corresponding to a POVM.

Consider a POVM of the form $M_i$ such that $\sum_i M_i = \mathbb{I}$. I can represent this by a quantum channel with Kraus operators $\{ A_i\}$ such that $A_i = \vert i\rangle \otimes B_i$ (up to unitaries on $B$). In order for this to be consistent, it must be that $B^\dagger_i B_i = M_i$

I guess (correct me if this is wrong) that every positive operator $M_i$ can be written in the form $B^\dagger_i B_i$ but how can I explicitly show this and compute $B_i$ given the $M_i$?


1 Answer 1


For one: the $B_i$ will never be unique, because changing them to $B_i'=U_i B_i$, for $U_i$ an arbitrary unitary, does not affect the square: $$ B_i'^\dagger B_i' = B_i^\dagger U_i^\dagger U_i B_i = B_i^\dagger B_i. $$ Moreover, the set of possible choices is huge, given how widely that $U_i$ can roam. (In fact, there's very little that requires that the $B_i$ be operators within the same Hilbert space - they can go elsewhere too! This is reflected in the fact that the $U_i$ do not need to be rectangular, and they only need to satisfy the left-inverse relationship $U_i^\dagger U_i = \mathbb I$. The other product, $U_i U_i^\dagger$, must be an orthogonal projector, but it does not need to have full rank.)

This means that giving a single recipe is a hopeless task, but you can nevertheless still provide one way to get a suitable $B_i$, as a proof of principle. (This won't cover all the solutions, nor does it necessarily work well in practice or get the same operator that you'll get with easier-in-practice methods. But none of that matters.)

To get a solution, simply start by realizing that since the $M = M^\dagger \geq 0$ are hermitian and positive-semidefinite (dropping the subindex $i$ for simplicity), they always have an orthonormal eigenbasis, $$ M = \sum_m m |m⟩⟨m|, $$ with $m\geq 0$ for all $m$. Then $$ B = \sum_{m\neq 0} \sqrt{m} |m⟩⟨m|, $$ often denoted $B = \sqrt{M}$, is well-defined, and it is a suitable solution to the problem.


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.