New Sum-of-Squares Hierarchy Solves Quantum Channel Coding Optimization

A sharper route through a hard quantum optimization problem

A research preprint posted on September 9, 2026, presents a new Sum-of-Squares, or SOS, hierarchy for computing the optimal success probability of sending classical messages through a single use of a quantum channel [1]. The work, titled “A Sum-of-Squares Hierarchy with Quadratic Convergence for Quantum Channel Coding,” is by Hoang Ta of Hanoi University of Science and Technology and Hoang Anh Tran of the National University of Singapore [1]. Its core claim is technical but important: the authors construct a Hermitian SOS hierarchy whose relaxation error falls quadratically with the hierarchy level, while remaining tied to the channel’s advantage over random guessing [2].

The optimization target is the one-shot success probability, denoted in the paper as Psucc(Φ, k), where Φ is a quantum channel and k is the number of equiprobable classical messages [2]. In operational terms, a sender chooses quantum states to encode the messages, a noisy quantum channel transforms those states, and a receiver applies a measurement to guess which message was sent [2]. The mathematical problem is to optimize both the encoding and decoding so that the average probability of a correct guess is as high as possible [2].

That problem is not merely inconvenient. The paper states that computing the optimal success probability for a single use of a quantum channel is NP-hard, and that the difficulty persists even when there are only two messages [1]. This binary case is especially significant because it connects channel coding to the trace-norm contraction coefficient, a quantity measuring how much a quantum channel can reduce distinguishability between input states [2]. For two messages, the paper uses the identity Psucc(Φ, 2) = (1 + ηtr(Φ))/2, making the coding problem a direct route to estimating ηtr(Φ) [2].

What the new hierarchy changes

The immediate advance is not a claim that NP-hardness disappears. Rather, the paper offers a structured sequence of semidefinite programs that produce upper bounds converging to the true optimum [2]. At each level L, the hierarchy returns a value UL(Φ, k), and the authors prove that these values form a nonincreasing sequence of upper bounds on Psucc(Φ, k) [2]. This matters because practical approaches to hard optimization problems often depend on relaxations: one solves a tractable problem that bounds the hard one, then improves the bound by increasing a level or degree parameter.

The novelty lies in the convergence rate. Existing semidefinite programming approaches based on symmetric extensions are described in the preprint as having an a priori error estimate that decays like the inverse square root of the extension level [1]. By contrast, the new SOS hierarchy has a quadratic error guarantee in the level L [2]. In the authors’ main convergence theorem, the additive gap UL(Φ, k) − Psucc(Φ, k) is bounded by a constant times dA²/L² multiplied by Psucc(Φ, k) − 1/k, where dA is the input dimension [2].

That last factor is more than a detail. The term Psucc(Φ, k) − 1/k is the advantage over random guessing, because 1/k is the success probability of a purely random guess among k equally likely messages [2]. If a noisy channel gives only a small improvement over guessing, the guarantee tightens proportionally [2]. The authors also point out that the stated convergence bound has no dependence on the output dimension dB, even though the semidefinite program’s actual size still depends on the dimensions of the systems involved [2].

How the proof reframes channel coding

The paper’s strategy combines several ideas from quantum information and polynomial optimization. First, for a fixed decoding measurement, the input states can be taken to be pure states [2]. The authors parametrize those pure states by real unit vectors, turning the input optimization into a problem over a product of spheres [2]. Then they use state-discrimination duality to remove the decoding measurement from the formulation [2].

This dual step is important because it converts the original operational coding problem into a certificate problem. Instead of directly optimizing a measurement, an upper bound can be certified by a Hermitian matrix-valued field satisfying positivity constraints across the product of spheres [2]. At hierarchy level L, the method restricts that matrix field to polynomial form and enforces positivity through finite-degree SOS certificates [2]. Those certificates can be represented by positive semidefinite Gram matrices, which is what turns the relaxation into a finite-dimensional semidefinite program [2].

A central technical obstacle is that an optimal dual field need not be polynomial or smooth [2]. In the binary case, for example, a natural optimal object involves a matrix absolute value, which can be nonsmooth [2]. The paper avoids requiring a polynomial approximation of that object. Instead, it applies positive squared-polynomial kernels on products of spheres directly to the positive slack fields that arise from the dual formulation [2]. The authors then correct the effect of the kernel on the common harmonic structure of the channel output fields, yielding feasible polynomial dual certificates [2].

This kernel-and-correction mechanism is the engine behind the convergence rate. The channel-output fields share a constant component and a degree-two spherical harmonic component [2]. A normalized squared kernel leaves constants fixed and scales the degree-two component by an eigenvalue [2]. By choosing a kernel polynomial that maximizes the relevant eigenvalue, the authors obtain a distortion bound that falls like dA²/L² [2]. This is the mathematical source of the quadratic convergence result [2].

Why the binary-message case is a useful benchmark

For k = 2, the paper translates the success-probability hierarchy into a multiplicative upper approximation for the trace-norm contraction coefficient [2]. Defining EL(Φ) = 2UL(Φ, 2) − 1, the authors prove ηtr(Φ) ≤ EL(Φ) ≤ min{1, (1 + ρ2dA,L)ηtr(Φ)} [2]. They further bound EL(Φ) − ηtr(Φ) by a constant times dA²/L² times ηtr(Φ) [2].

This is meaningful in the strong-contraction regime, where ηtr(Φ) is small [2]. A purely additive approximation can look numerically small while still being large relative to the true contraction coefficient. The multiplicative statement instead controls the relative error over positive ηtr(Φ), and the authors show exactness at every hierarchy level when ηtr(Φ) is either 0 or 1 [2]. In physical language, the hierarchy behaves especially cleanly at the endpoints: channels that erase distinguishability completely and channels that preserve distinguishability perfectly are captured exactly [2].

The connection to contraction is not only mathematical. Trace distance is a standard distinguishability measure in quantum theory, and contraction coefficients quantify how quickly distinguishability can be lost under repeated noise. The paper notes that when input and output spaces match, the trace distance after n successive applications of a channel is bounded by ηtr(Φ)n times the initial trace distance [2]. That makes better approximation of ηtr(Φ) relevant to assessing memory loss, mixing, and information degradation in quantum processes [2].

Numerical evidence: promising, but not a theorem of universal first-level exactness

The preprint also reports numerical experiments comparing the SOS hierarchy with the extension hierarchy at the first level for binary channel coding [2]. The authors study an exactly solvable benchmark and 40 random qubit-to-qutrit channels, including extension-hierarchy variants with and without positive partial transpose, or PPT, constraints [2]. For the random channels, the samples include 20 channels of Choi rank 6 and 20 of Choi rank 3, generated through a QR factorization of complex Ginibre matrices [2].

The reported numerical results are striking. Across all 40 random qubit-to-qutrit channels, the repaired first-level SOS gaps relative to the reference value are at most 4.10 × 10^-7 [2]. The mean first-level extension gaps are much larger: 0.101499 without PPT and 0.101377 with PPT [2]. The repaired SOS value is smaller than both computed extension values in every instance, with a minimum observed separation greater than 0.0164 [2].

The authors are careful about what these computations do and do not prove. The reference values are obtained by evaluating over 8,192 spherical grid points and refining selected directions by local optimization, which gives an achievable lower estimate but not a certified global optimum [2]. The numerical solves return “optimal_inaccurate,” and the feasibility checks are carried out in floating-point arithmetic rather than interval arithmetic [2]. The experiments support numerical tightness of the first SOS level on the sampled channels, but the paper explicitly says they do not establish exactness for all qubit-to-qutrit channels [2].

The current state of the story

As of the September 9, 2026 posting, the development is a newly available preprint rather than a peer-reviewed journal article [1]. The arXiv record lists the subject areas as quantum physics, information theory, and optimization and control, which reflects the paper’s hybrid role: it is simultaneously about quantum communication, convex relaxation, polynomial certificates, and computational complexity [1]. A search-indexed arXiv mirror also surfaced the paper under the same title and abstract, reinforcing that the current public record is centered on the preprint itself rather than on broader outside coverage [3].

The practical impact will depend on several follow-up questions. One is implementation: semidefinite programs grow quickly with hierarchy level, and the authors themselves distinguish convergence guarantees from runtime conclusions [2]. Another is empirical scope: the first-level results for sampled qubit-to-qutrit channels are encouraging, but broader numerical testing would be needed across larger dimensions, different channel families, and higher numbers of messages [2]. A third is theoretical sharpness: future work may examine whether the quadratic rate is optimal for this coding formulation or whether special channel classes admit still stronger guarantees.

Still, the conceptual advance is clear. The paper brings Hermitian sum-of-squares certificates to the unassisted one-shot quantum channel coding problem and proves a convergence rate that improves the known hierarchy landscape described by the authors [1]. For a field where optimization problems quickly become NP-hard, the ability to trade SDP size for a quantified and rapidly shrinking error bound is valuable in its own right [2]. The result does not make quantum channel coding easy, but it gives researchers a sharper mathematical instrument for bounding one of its central one-shot quantities [2].

Sources from the last 72 hours

1. [1][2609.09629] A Sum-of-Squares Hierarchy with Quadratic Convergence for Quantum Channel CodingSep 9, 2026, 2:40 AM UTC
2. [2]A Sum-of-Squares Hierarchy with Quadratic Convergence for Quantum Channel CodingSep 9, 2026, 2:40 AM UTC
3. [3]arXiv Troller record: A Sum-of-Squares Hierarchy with Quadratic Convergence for Quantum Channel CodingSep 9, 2026, 2:40 AM UTC

AI-generated article based on recent web research, then preserved as a dated editorial snapshot.