Multi-slot virtual channel conjugation

An exact signed simulation of the conjugate of an unknown channel, a closed formula for the state subcase, and certified adaptive advantage at three slots.

Full version (PDF, 7 pp)

Background

Some linear maps on quantum states are useful but not physical — complex conjugation and transposition among them. They can still be simulated by taking a signed combination of physical circuits and post-processing the outcomes, at a cost: the sum of absolute coefficients, which controls the sampling overhead. The framework for quantifying that cost is [5], and the circuit model for querying an unknown channel several times with memory is the quantum comb [6].

For an unknown channel given as a black box, the one-query optimum is known [2]. The natural next question — what repeated queries buy — is QIQCOP problem op_06e9f0c7b3b62f3b [1].

The problem

Let Λ be an unknown CPTP map, and let a slot be one call to it. Each branch of the protocol is a deterministic fixed-order n-slot comb with arbitrary finite quantum memory. The target is complex conjugation in fixed bases — not the adjoint. The cost is the infimum of the sum of absolute real coefficients over universal exact signed combinations of such branches.

Two things are being asked: how the cost behaves as n grows, and whether adaptive access — feeding one query’s output into the next — beats running the queries in parallel.

What was established

A general construction. An explicit physical parallel comb realises the target exactly at every finite n, built from unnormalised Choi matrices with swap pairs transposing the inserted channel. Its cost is controlled by a Jucys–Murphy content bound [3], and at any fixed nontrivial pair of dimensions the cost approaches one at the optimal order. The cost, not the channel error, is what has a limit — the simulation is exact throughout.

An exact formula for the state subcase. When the input dimension is one, the optimal cost is given in closed form by a finite sum over integer partitions with Weyl dimensions and rational arithmetic. It optimises over freely chosen CPTP branches rather than a fixed white-noise line.

Two queries, and where adaptivity starts. For all dimensions the optimal two-slot adaptive and parallel costs coincide — adaptivity buys nothing at two queries. At three and four qubit-channel slots, rational primal/dual witnesses put the optimum in certified intervals whose upper endpoints fall strictly below the corresponding parallel lower bounds. So three slots is the first point at which adaptive access can help at all in this problem.

Verification

The certificates are exact. Verification uses no optimizer, no numerical eigenvalue threshold and no saved floating-point seed; integer products carry explicit overflow bounds. The four-slot system is 3,330 adaptive (4,506 parallel) equations in 1,864 branch coefficients, reduced by exact highest-weight projection to nine positivity blocks and checked by rational elimination, with a dual witness supplying the lower bound. A separate cross-check certifies the primal’s universal action across all 1,820 channel-monomial groups without sampling channels.

Three AI reviewers worked independently — one rederiving the construction and lower bound before reading the proof, one checking the state formula and its prior-art context, one adversarial.

The adversarial pass found two real defects in the verifier, and this is the part worth dwelling on. The three-slot code converted saved rational dual equations to floating point when checking provenance, so a tiny rational perturbation of a right-hand side could pass that check and forge a lower bound. The four-slot verifier accepted misordered symmetric/skew metadata. Both were replaced by exact checks, with regressions that reject the malformed inputs. The original certificate rows passed independent exact provenance checks, so no endpoint moved — but a verifier that can be fooled is worth reporting whether or not it changed an answer.

What remains open

The broad question is unsolved, and should stay listed as such.

  1. No general formula for the optimum at arbitrary finite n.
  2. No proof that the leading coefficient converges for general channels; even in the qubit case only a range is proved.
  3. Fixed order only. Everything here concerns fixed-order combs, not indefinite causal order.
  4. No efficient synthesis. The constructions are not shown to compile to efficient circuits.
  5. Cost is not query count. Sampling variance scales with squared signed cost, but each run consumes n calls, so a lower coefficient alone does not prove fewer total channel queries.

Priority is not claimed: the literature search was limited, and absence of a matching source is not evidence of novelty. The pieces offered for scrutiny are the general-channel parallel lift, the sector-resolved state evaluation, the two-slot statements, and the finite adaptive certificates.

References

  1. QIQCOP, Multi-slot overhead of virtual channel conjugation, problem op_06e9f0c7b3b62f3b.
  2. C. Zhu, Z. Tang, G. Zhen, Y. Li, G. Bai and X. Wang, Simulation of adjoints and Petz recovery maps for unknown quantum channels, arXiv:2602.05828.
  3. V. Brzić, D. Grinko, M. Studziński and M. T. Quintino, Optimal pure state cloning and transposition are complementary channels, arXiv:2603.23628.
  4. Q. Dong, M. T. Quintino, A. Soeda and M. Murao, Implementing positive maps with multiple copies of an input state, Phys. Rev. A 99, 052352 (2019), arXiv:1808.05788.
  5. B. Regula, R. Takagi and M. Gu, Operational applications of the diamond norm and related measures in quantifying the non-physicality of quantum maps, Quantum 5, 522 (2021).
  6. G. Chiribella, G. M. D’Ariano and P. Perinotti, Theoretical framework for quantum networks, Phys. Rev. A 80, 022339 (2009).