Preprint: Quantum decoding expands a window for polynomial states

A new quantum-information preprint reports a modest but consistent gap in a nonlinear optimization benchmark: an ideal quantum calculation averaged about 0.643134, while the median score from the tested classical searches was 0.60666. The paired quantum-minus-classical gap was positive for all 10 independent coefficient vectors, with a median of 0.036479, or about 3.65 percentage points. The authors stress that this is not a classical lower bound. It compares an ideal quantum expectation with searches run under finite classical budgets.

A transfer circuit built around relative decoding

The paper's broader proposal is polynomial state transduction. An efficiently preparable polynomial state in a source quantum system is transferred to a corresponding target state through a unital *-homomorphism, a structure-preserving map between the two operator algebras.

The proposed circuit has four coherent stages. It first applies a source operator-Fourier transform, computes the target labels and phases, erases the source label by relative decoding, and then applies the inverse target operator-Fourier transform.

The efficiency claim is conditional. It assumes compatible source and target bases, efficient preparation of the source state, efficient operator-Fourier transforms, efficient computation of labels and phases, and an efficient coherent relative decoder. The paper does not present those ingredients as automatic for arbitrary source-target pairs.

The mathematical window

The central mathematical test concerns the degree of the polynomial. The paper says all operator inner products on the degree-D source subspace are preserved when $2D < drel$. Here D is the source polynomial degree bound, and drel is the relative distance, the separation measure used by the relative decoder. This is a conditional algebraic guarantee under the circuit assumptions.

Relative distance is at least as large as ordinary distance in the paper's construction. More importantly, the authors give families in which ordinary distance remains bounded as the system grows, while relative distance grows linearly with system size. That creates the possibility of certifying higher-degree polynomial filters, although the result is a family construction rather than a guarantee for every quantum system.

For Hamiltonian energy expectations, the source-computable window is narrower: it requires $2D + 1 < drel$, whereas guide transduction itself requires $2D < drel$. Within the trace-matching window, however, transduction adds no advantage for estimating expectations; any potential gain concerns target samples or quantities outside that window.

The benchmark's narrow lead

The numerical test used a nonlinear benchmark. Methods and search budgets were fixed on separate tuning instances before the evaluation on 10 new random coefficient vectors. Each vector was paired with two predicate-set choices and two solver streams per choice; deep runs examined approximately 5 million candidates. The reported setup used p = 2053, n = 205 and guide degree D = 50.

That design makes the comparison concrete but narrow. The quantum figure is an ideal mean, while the classical result is a median from finite-budget searches. The authors therefore do not treat the gap as evidence about the best score any classical method could achieve.

What the result still leaves open

The framework also gives a transfer statement for Gibbs states, the thermal states associated with a Hamiltonian. Its trace-distance error bound is $\delta_{\mathrm{src}} + 2\eta$: source preparation error plus twice the polynomial approximation error. The same bound applies to the Gibbs marginal after the reference and workspace registers are discarded. This conclusion remains conditional on the source-preparation and transduction assumptions.

The authors extend the framework to fermionic Majorana systems and to qudit and occupation-truncated bosonic systems using Weyl operators. In a fermionic example, relative distance is twice the number of source blocks, while ordinary distance is 3. The relative-decoding circuit supports polynomial degree up to one less than the number of blocks, compared with degree at most 1 under ordinary unique decoding.

The benchmark does not settle the larger question of practical advantage. It compares an ideal quantum expectation with finite-budget classical searches on 10 coefficient vectors, and its positive gap does not establish a classical lower bound. Nor does a mean-score comparison by itself show superiority for preparing the full target output distribution; within the trace-matching window, the method adds no advantage for expectation estimation.

The next test for the proposal is therefore end-to-end: whether the assumed source preparation, Fourier transforms, label and phase computation, and coherent relative decoding can be built efficiently for useful source-target pairs. The paper reports the nonlinear implementation, reference instances, polynomial witnesses and reproduction scripts as publicly available in RDQINumerics release v1.0.0. The manuscript is an arXiv version 1 preprint dated 28 September 2026.

Paper data and sources

Original title: From Simple Sources to Quantum Advantage: Homomorphic Polynomial Transduction via Relative Decoding
Authors: Zhong-Xia Shang, Daniel Stilck França
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-09-28
DOI: Not available
Original paper · Full text