A Counterexample to the Optimality Conjecture in Convex Quantum Channel Optimization

Introduction

The trace‑distance formulation of optimal state transformation is a cornerstone of convex quantum channel optimization. In a 2021 paper, Coutts, Girard, and Watrous conjectured that the sign function applied to the error matrix yields a unique dual certificate for every optimal solution. This elegant claim promised a universal, easily verifiable optimality condition. In a recent contribution, Jianting Yang demonstrates that the conjecture fails even in the smallest non‑trivial setting: a two‑dimensional Hilbert space. The counterexample not only overturns the conjecture but also highlights subtle pitfalls when extending classical convex‑optimization intuition to quantum settings.

The conjecture and its appeal

Coutts et al. considered the problem of minimizing the trace distance \(\|\rho - \Phi(\sigma)\|_1\) over all completely positive, trace‑preserving maps \(\Phi\) that act on a fixed input state \(\sigma\) to approximate a target state \(\rho\). By formulating the task as a semidefinite program (SDP), they identified a natural dual variable: the sign matrix \(\operatorname{sgn}(\Delta)\) of the error matrix \(\Delta = \rho - \Phi(\sigma)\). Their conjecture asserted that \(\operatorname{sgn}(\Delta)\) always satisfies the dual feasibility conditions and attains the optimal dual value, thereby certifying optimality of the primal solution.

The appeal of this conjecture lies in its simplicity. Computing the sign of a Hermitian matrix is straightforward, and if it indeed provided a universal certificate, practitioners could verify optimality without solving the full dual SDP. This would streamline analyses of quantum channel design, error mitigation, and resource‑theoretic transformations.

Constructing the counterexample

Yang’s counterexample exploits the geometry of the Bloch sphere. Consider two pure qubit states \(\ket{0}\) and \(\ket{+}= (\ket{0}+\ket{1})/\sqrt{2}\). Let the target state be \(\rho = \ket{0}\!\bra{0}\) and the input state be \(\sigma = \ket{+}\!\bra{+}\). The optimal channel for minimizing \(\|\rho-\Phi(\sigma)\|_1\) is a projective measurement onto the \(\{|0\rangle,|1\rangle\}\) basis followed by state preparation of \(|0\rangle\) conditioned on the outcome. This channel achieves a trace‑distance error of \(1/\sqrt{2}\).

When the sign function is applied to the resulting error matrix \(\Delta = \rho - \Phi(\sigma)\), the resulting matrix fails to satisfy the dual feasibility constraint \(\operatorname{sgn}(\Delta) \succeq 0\) required by the SDP dual. Explicit calculation shows a negative eigenvalue, meaning the sign matrix cannot serve as a valid dual certificate. Moreover, any perturbation of the channel that preserves optimality still yields a sign matrix with the same infeasibility, demonstrating that the issue is not an artefact of a particular optimal solution but inherent to the problem structure.

Implications and lessons

The counterexample forces a reassessment of how dual certificates are constructed in quantum SDP settings. While the sign function works for many classical convex problems (e.g., \(\ell_1\) minimization), quantum trace‑distance optimization introduces additional constraints—complete positivity and trace preservation—that interact non‑trivially with the sign operation. Yang’s result suggests that a more nuanced dual construction is needed, perhaps involving operator‑norm projections or additional symmetry‑exploiting techniques.

Beyond the specific conjecture, the work underscores the importance of low‑dimensional counterexamples in quantum information theory. Even the simplest qubit systems can reveal hidden complexities that may be invisible in higher‑dimensional numerical experiments. Researchers developing new optimality criteria should therefore test conjectures on the smallest possible Hilbert spaces before generalizing.

Practical take‑away

For practitioners working on convex quantum channel design, the safe approach remains to solve both the primal and dual SDPs explicitly, rather than relying on the sign‑function shortcut. When a candidate solution appears optimal, verify dual feasibility using the full dual constraints. If computational resources are limited, consider exploiting problem symmetries (as discussed in works by Ling, Bandeira, and others) to reduce the SDP size, but retain the rigorous dual verification step.

--- This article summarizes the key ideas of Jianting Yang’s counterexample without reproducing technical proofs. Readers interested in the detailed construction and its broader mathematical context are encouraged to consult the original paper in Quantum (2026, vol. 10, p. 2253).