As I understand, a set of effects $\mathcal{E}\subset\mathcal{E}(\mathcal{H})$ are called informational complete if given $\sigma \in \mathcal{T}\mathcal(H)$, $\operatorname{tr}[\sigma E]=0$ for all $E\in\mathcal{E}$, then $\sigma=0$.

This is also equivalent to: if $\operatorname{tr}[\sigma_1 E]=\operatorname{tr}[\sigma_2 E]$ for all $E\in\mathcal{E}$ then $\sigma_1=\sigma_2$.

However, if we are performing state tomography, then our state is not a general operator in the trace class. Rather, it has trace one.

In other words, now $\rho\in\mathcal{S}(\mathcal{H})\subset \mathcal{T}(\mathcal{H})$, where $\mathcal{S}(\mathcal{H}):=\{\rho\in \mathcal{T}(\mathcal{H})\,|\,\rho\ge 0,\,\operatorname{tr}[\rho]=1\}$.

Therefore, for the definition of tomographical completeness, can we replace $\mathcal{T}(\mathcal{H})$ with $\mathcal{S}(\mathcal{H})$ in the definition of informational completeness?

In other words, can we now define tomographical completeness as: given $\rho_1, \rho_2 \in \mathcal{S}(\mathcal{H})$ if $\operatorname{tr}[\rho_1 E]=\operatorname{tr}[\rho_2 E]$ for all $E\in\mathcal{E}$ then $\rho_1=\rho_2$?


1 Answer 1


Yes it's the same thing. If a set of effects $\{E_i\}_i$ linearly spans the set of all density matrices (i.e. the unit-trace positive semidefinite operators) then it also spans all operators. Trivially because any operator can be written as a linear combination of density matrices.

In practice, the terms "tomographically complete" and "informationally complete" are typically used as synonyms in the literature.

  • $\begingroup$ Do you mean that whether it is $\sigma \in \mathcal{T}(H)$ or $\sigma \in \mathcal{S}(\mathcal{H})$ does not make any difference? My gut feeling is that it does make a difference as $\sigma \in \mathcal{S}(\mathcal{H})$ has $d^2 - 1$ degrees of freedom while $\sigma \in \mathcal{T}(\mathcal{H})$ has $d^2$ degrees of freedom, where $d$ is the dimension of the Hilbert space. Hence, tomographical completeness differs from informational completeness in that it requires one less effect. $\endgroup$
    – Godfly666
    Commented Apr 22 at 11:20
  • $\begingroup$ I mean that (at least in finite dimensions) a set of effects spans all density matrices iff it spans all operators. Think of a qubit: you can't span all qubit states without spanning $I,X,Y,Z$, and thus all Hermitian operators. Formally, any operator decomposes uniquely in terms of 2 Hermitians, any Hermitian decomposes as $H=P-Q$ with $P,Q\ge0$, and any PSD is a scalar multiple of a density matrix. Thus any linear operator is a linear combination of density matrices $\endgroup$
    – glS
    Commented Apr 22 at 12:15
  • $\begingroup$ A qubit has the density operator $\rho = \frac{I + xX + yY + zZ}{2}$ and hence has only three degrees of freedom. So, aren't the effects $\{\Pi_1, \Pi_2, \Pi_3\} = \{\frac{I+X}{2}, \frac{I+Y}{2}, \frac{I+Z}{2}\}$ tomographically complete? Solving the equations $\operatorname{tr}[\rho \Pi_k] = p_k, \quad k=1,2,3$ can uniquely determine the qubit $\rho$. However, informational completeness requires four effects. $\endgroup$
    – Godfly666
    Commented Apr 23 at 9:45
  • $\begingroup$ @Godfly666 that's true, though that set of effects wouldn't represent a physical measurement due to the lack of normalisation: you don't have $\sum_i \Pi_i=I$. I guess the proper way to state this is then that there is no difference between informational and tomographic completeness as you defined them, for any set of effects that forms a POVM. But if you're just going for the mathematical abstraction, sure, you can have a set of effects which span a basis of traceless Hermitians for the space, but not the identity $\endgroup$
    – glS
    Commented Apr 24 at 7:25
  • $\begingroup$ In a typical physical experiment, we don't perform a single-setting measurement where there is a single POVM. Instead, we use multiple measurement settings, each of which is a POVM. For example, to measure a qubit, we usually don't use the SIC-POVM with four elements. Instead, we use three measurement settings $\{\frac{1\pm X}{2}\}$, $\{\frac{1\pm Y}{2}\}$, $\{\frac{1\pm Z}{2}\}$. That's where my doubt comes from. $\endgroup$
    – Godfly666
    Commented Apr 24 at 8:38

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.