Quantum channels with fixed finite-dimensional input and output form a compact convexbody. We ask how many continuous latent parameters are needed to encode an arbitrarychannel and reconstruct it with uniformly small diamond-norm error, when the encoder iscontinuous and the decoder is unrestricted. This is a continuous analogue of nonlinear width,distinct from tomography, which counts channel uses, and from programmable processors,which fix a decoding device. For a compact source the optimal error equals the minimalworst-case radius of the encoder’s fibers. Lower bounds follow from antipodal pairs in thesource and from latent collision topology via Borsuk–Ulam and the deleted-product coindex.Even input-independent output preparation forces error one below the output-state dimen-sion, with a Clifford analogue inside the unitary channels. We determine the largest diamondball contained in the channel body: it is centred at the completely depolarizing channel, isoptimal among all centres, and, together with a full-direction section, certifies the interme-diate lower envelope; the same channel is the optimal constant decoder, with exact coveringradius, so the two envelopes meet at the exact width at zero latent dimension. Constructiveupper bounds from convex projection and balanced pinching bracket the intermediate widths,and meet the lower envelope in the degenerate collapse case and from the affine dimensionon. Every finite-query adaptive network induces a continuous final code containing all sideinformation, so the same bottlenecks hold uniformly in the number of queries, for uniformlysuccessful postselection, and for sampled transcripts; a one-query informationally completemeasurement achieves the injective threshold, and repetition gives an O(N −1/2) expected-error bound. In the input-independent case the widths are governed by the companion’sflatness dichotomy: the flat value one fails for every output dimension at least three at thebottom two latent dimensions, the qutrit sitting exactly at 4/3 there, through a flag-quotientcohomology computation.
Quantum channels with fixed finite-dimensional input and output form a compact convexbody. We ask how many continuous latent parameters are needed to encode an arbitrarychannel and reconstruct it with uniformly small diamond-norm error, when the encoder iscontinuous and the decoder is unrestricted. This is a continuous analogue of nonlinear width,distinct from tomography, which counts channel uses, and from programmable processors,which fix a decoding device. For a compact source the optimal error equals the minimalworst-case radius of the encoder’s fibers. Lower bounds follow from antipodal pairs in thesource and from latent collision topology via Borsuk–Ulam and the deleted-product coindex.Even input-independent output preparation forces error one below the output-state dimen-sion, with a Clifford analogue inside the unitary channels. We determine the largest diamondball contained in the channel body: it is centred at the completely depolarizing channel, isoptimal among all centres, and, together with a full-direction section, certifies the interme-diate lower envelope; the same channel is the optimal constant decoder, with exact coveringradius, so the two envelopes meet at the exact width at zero latent dimension. Constructiveupper bounds from convex projection and balanced pinching bracket the intermediate widths,and meet the lower envelope in the degenerate collapse case and from the affine dimensionon. Every finite-query adaptive network induces a continuous final code containing all sideinformation, so the same bottlenecks hold uniformly in the number of queries, for uniformlysuccessful postselection, and for sampled transcripts; a one-query informationally completemeasurement achieves the injective threshold, and repetition gives an O(N −1/2) expected-error bound. In the input-independent case the widths are governed by the companion’sflatness dichotomy: the flat value one fails for every output dimension at least three at thebottom two latent dimensions, the qutrit sitting exactly at 4/3 there, through a flag-quotientcohomology computation.
We study tomography for quantum channels with input dimension d1, output dimension d2, and Kraus rank at most r, to within diamond norm error ε, using adaptive experiments that retain no quantum memory between channel queries.
- For quantum channels whose non-zero Choi eigenvalues are bounded below by Ω(d1/r), we establish optimal query upper and lower bounds Θ(d1d2r2/ε2). The upper bound is achieved by a nonadaptive algorithm that uses the estimator from [Surawy-Stepney et al., Quantum (2022)], together with a new diamond-norm analysis. The lower bound applies to arbitrary adaptive incoherent protocols and follows from a new local family of channels and a uniform one-query Fisher-information bound.
- For general channels, we establish an upper bound O(d1d2r2log(2d1)/ε2), nearly matching the above lower bound Ω(d1d2r2/ε2). To achieve this, we generalize the above nonadaptive algorithm by adapting the input state over O(log(2d1)) rounds with the Matrix Multiplicative Weight Update algorithm.
Quantum instruments describe the full statistics of a measurement together with theconditional post-measurement state, and they form the building blocks of adaptive protocolsand quantum networks. We study the geometry of the set of all n-outcome instrumentsacting between two finite-dimensional quantum systems, and we ask how many continuousreal parameters must be retained in order to encode an arbitrary instrument and later re-construct it with uniformly small error, under the convention of continuous encoding andunrestricted decoding. The geometry of the instrument body is determined exactly: itsaffine dimension is d2A(nd2B − 1); the largest diamond-norm ball contained in the body hasradius 2/(ndB min{dA, dB }) and is centred at the uniform depolarising instrument, whichis optimal among all centres; and the same instrument is optimal for the covering problem,with an exact covering radius linked to the in-radius by a complementarity identity. Everyradius and width carries a single-shot operational meaning through the optimal discrimi-nation error. On the compression side, antipodal constructions — output preparation, aninput-dependent observable sphere, and a full-direction section — produce a certified lowerenvelope on the optimal error; constructive upper bounds from constant codes, convex pro-jection, and balanced pinching meet the envelope exactly at zero latent dimension (whereboth equal the exact constant-code width), in the collapse case, and from the affine dimen-sion on, and bracket it in the intermediate regime; the zero-error threshold equals the affinedimension. For the input-independent case the flat-width classification is complete: the flatvalue one holds exactly on the classical n ≤ 4 and qubit n ≤ 2 bodies and fails at least4/3 on every other, including every state space of dimension ≥ 3 at the bottom and secondlatent dimensions alike, the qutrit exactly 4/3 at both, the second-index result resting on theintegral cohomology of the cyclic flag quotient. The single-outcome case recovers quantumchannels, and the one-dimensional-output case recovers finite-outcome measurements; thetrade-off between the number of outcomes and the robustness of the uniform instrument ismade explicit.
Consider quantum channels with input dimension d1, output dimension d2 and Kraus rank at most r. Any such channel must satisfy the constraint rd2≥d1, and the parameter regime rd2=d1 is called the boundary regime. In this paper, we show an optimal query lower bound Ω(rd1d2/ε2) for quantum channel tomography to within diamond norm error ε in the away-from-boundary regime rd2≥2d1, matching the existing upper bound O(rd1d2/ε2). In particular, this lower bound fully settles the query complexity for the commonly studied case of equal input and output dimensions d1=d2=d with r≥2, in sharp contrast to the unitary case r=1 where Heisenberg scaling Θ(d2/ε) is achievable.
How many black-box queries to a quantum channel are needed to learn its full classical description? This question lies at the heart of quantum channel tomography (also known as quantum process tomography), a fundamental task in the characterization and validation of quantum hardware. Despite extensive prior work, the optimal query complexity for quantum channel tomography is far from fully understood.
In this paper, we study tomography of an unknown quantum channel with input dimension d1, output dimension d2, and Kraus rank at most r, to within error ε. We identify the dilation rate τ=rd2/d1 (which always satisfies τ≥1 due to the trace preservation of quantum channels) as a key parameter, and establish that the optimal query complexity of channel tomography exhibits distinct scaling laws across three regimes of τ.
- In the boundary regime (τ=1): we show that the query complexity is Θ(rd1d2/ε) for Choi trace norm error ε, and is upper bounded by O(min{rd11.5d2/ε,rd1d2/ε2}) and lower bounded by Ω(rd1d2/ε) for diamond norm error ε.
- In the away-from-boundary regime (τ≥1+Ω(1)): we show that the query complexity is Θ(rd1d2/ε2) for both Choi trace norm and diamond norm errors ε.
Our results uncover a sharp Heisenberg-to-classical phase transition in the query complexity of quantum channel tomography: at τ=1, the optimal query complexity exhibits Heisenberg scaling 1/ε, whereas for τ≥1+Ω(1), it exhibits classical scaling 1/ε2. In addition, we show that in the near-boundary regime (1<τ<1+o(1)), the query complexity exhibits a mixture of Heisenberg and classical scaling behaviors.
Quantum channel tomography (also known as quantum process tomography) is a fundamental task in quantum information and quantum computing, aiming to learn the full description of an unknown quantum channel from black-box queries. For channels with input dimension d1, output dimension d2, and Kraus rank at most r, [CGO+26] identifies the dilation rate τ=rd2/d1≥1 as a key parameter governing a Heisenberg-to-classical transition in query complexity. While the optimal bounds in the boundary regime (τ=1) and the away-from-boundary regime (τ≥1+Ω(1)) are now known, the precise behavior in the near-boundary regime (1<τ<1+o(1)) has remained unresolved.
In this paper, we close this gap, thereby fully settling the query complexity of quantum channel tomography. Specifically, we show that the query complexity is
- Ω(rd1d2/ε2) for diamond norm error ε, matching the upper bound by [MB26, CGO+26],
- Θ(d12/ε+min{(τ−1)d12,(τ−1)2d12d22}/ε2) for Choi trace norm error ε.
Our results reveal two qualitatively different forms of the Heisenberg-to-classical transition. For diamond norm tomography, the Heisenberg scaling disappears suddenly as soon as τ leaves the boundary regime. In contrast, Choi trace norm tomography exhibits a gradual transition from Heisenberg to classical scaling as τ moves from the boundary regime to the away-from-boundary regime.
We determine the complete equal-reference Umegaki relative-entropy contraction coefficient of every finite-dimensional quantum channel. For its unnormalized Choi matrix J, let H be the largest output subspace with ℂⁿ ⊗ H ⊆ supp J, and let M(ρ) be the compression to H of Tr_A[(ρᵀ ⊗ I)J⁺]. The coefficient is 1 − min_ρ Tr_H M(ρ)⁻¹, with the minimum set to zero when H = {0}. A Kraus cancellation inequality gives the upper bound, and a weighted trace-zero operator basis gives matching faithful pairs with both marginals equal and reference dimension n². The complementary coefficient has a reciprocal radius formula, a finite semidefinite representation, and an exact product law. Strict contraction is equivalent to a nonzero completely positive replacer minorant. The formula also gives the exact coefficient of depolarizing channels throughout their physical parameter range. The general matching family has vanishing input divergence; its logarithmic rate is pointwise for a fixed faithful input density, with separate finite-accuracy bounds for optimization and conditioning. A different construction retains any prescribed positive limiting divergence when the coefficient is one. This cannot hold universally: a complete nonlinear bound for binary symmetric channels forces every maximizing sequence to lose input divergence, regardless of reference dimension. The positive construction still has vanishing trace distance and exponentially small regularization; no fixed-error or capacity conclusion follows. A visual guide records the mathematical routes, proof dependencies, and provenance of the result.