Optimal precision dependence of diamond-norm channel tomography
- Field
- Topics
Problem
Does tomography of an arbitrary \(d\)-dimensional quantum channel in diamond norm require and suffice with \(\Theta(d^4/\varepsilon^2)\) channel uses?
For \(d\geq2\), let \(\Lambda:\mathcal L(\mathbb C^d)\to\mathcal L(\mathbb C^d)\) be an unknown completely positive, trace-preserving channel with no Kraus-rank promise. Let \(Q_\diamond(d,\varepsilon)\) be the minimum worst-case number of ordinary channel uses needed to output \(\widehat\Lambda\) such that
The supremum in Eq. (1) is over states on \(\mathbb C^d\otimes\mathbb C^d\). Adaptive inputs, ancillas, quantum memory, and collective measurements are allowed, but no purification of the channel environment is supplied. Is \(Q_\diamond(d,\varepsilon)=\Theta(d^4/\varepsilon^2)\) uniformly in \(d\) and sufficiently small \(\varepsilon\)?
Source
This is an editor-formulated unrestricted-rank specialization of the joint dimension–accuracy question discussed by Mele and Bittel [Mele25]. Subsequent lower bounds settle this specialization; the original question is retained here as a solved problem.
Progress
Reports do not certify correctness or automatically change the problem's status. Progress policy.
General channels of input dimension \(d_1\), output dimension \(d_2\), and Kraus rank at most \(r\) admit diamond-norm tomography with \(O(rd_1d_2/\varepsilon^2)\) ordinary channel queries [Mele25], [Chen25]. Setting \(d_1=d_2=d\) and \(r=d^2\) gives the required upper bound.
Chen, Zhang, and Yu prove the matching lower bound \(\Omega(rd_1d_2/\varepsilon^2)\) when \(rd_2\geq2d_1\) (Theorem 1.1 and Corollary 1.2 of the January 2026 preprint) [CZY26]. The proof uses sequential channel testers, so it permits adaptive coherent queries and quantum memory; it is not restricted to tomography of independently prepared Choi states.
The subsequent joint preprint of Chen, Girardi, Oufkir, Yu, and Zhang subsumes that result. Corollary 1.3 gives \(\Theta(rd^2/\varepsilon^{\min\{r,2\}})\) for equal input and output dimensions; Theorems 3.4 and 6.2 supply the upper and lower bounds, with the query model defined in Section 2.3. Substituting \(r=d^2\) yields
\begin{equation}Q_\diamond(d,\varepsilon)=\Theta(d^4/\varepsilon^2).\tag{2} \end{equation}The rate is uniform in the dimension and sufficiently small error [CGOYZ26]. These are preprint results.
- Reported progress: PR #80.
Comment
The product dependence is established, not merely the two one-parameter scalings separately. The lower bound covers the ordinary black-box model in the statement, including adaptive queries, and the upper bound needs no access to a purification of the channel environment. The dimension is taken to be \(d\geq2\); at \(d=1\) there is a unique channel and zero queries suffice. The explicit large-dimension cutoff in Theorem 6.2 does not change the asymptotic claim: for the finitely many smaller nontrivial dimensions, preparation channels reduce the task to state tomography and give a uniform lower bound after decreasing its constant. The cited resolution predates this weekly audit and corrects an omission from the previous progress summary.
References
- [Mele25]
- A. A. Mele and L. Bittel, "Optimal learning of quantum channels in diamond distance," arXiv preprint (2025), version 3, 15 June 2026.arXiv
- [Chen25]
- K. Chen, N. Yu, and Z. Zhang, "Quantum channel tomography and estimation by local test," arXiv preprint (2025), version 2, 2 February 2026.arXiv