A mathematical preprint reports a way to tighten upper bounds on the rate of binary codes. Its first honeycomb construction proves that R2(δ)—the supremal asymptotic rate of binary codes whose relative minimum distance is at least δ—is no greater than a new exponent, κHC(δ), for every 0<δ<1/2.
The graph behind the result
The paper starts with a structural question: whether the graph behind the whole-cube exponent κH is only the boundary of a larger tractable representation graph, and whether that enlargement can improve binary-code bounds.
That larger object—the first honeycomb graph—retains every relevant hyperoctahedral irreducible, or symmetry building block, whose restriction contains the same stabilizer type and whose component partitions have at most two rows. Its vertices form a three-dimensional compatibility region with four same-row and cross-row channels in each bulk cell.
To turn the graph into an asymptotic result, the authors derive a profile-optimized moving-projection theorem, calculate exactly how weight transfers between boxes, and pass from finite graphs to the continuum through Følner boxes. The finite graph has explicit directed probabilities and exact balance.
The main mathematical gain
The first honeycomb exponent is a proven upper bound on the asymptotic rate. Two explicit branches—the one-sided face branch and the entropy-balanced branch—are strictly below κH for every 0<δ<1/2.
The entropy-balanced branch is also no higher than M2(δ). The paper identifies the infimum of the symmetric honeycomb slice exactly with M2(δ), the fully optimized second MRRW exponent.
Combining the honeycomb exponent with κCW gives a bound no worse than the prior combined binary exponent κbin. Another explicit combination, κpair, lies strictly below 2MQC; the paper also records that κH equals RMQC.
These are comparisons among upper bounds. The work does not establish that any proposed exponent equals the unknown optimal asymptotic rate or that it produces a better achievable code rate.
Exactness at finite length
At finite length, a connected retained positive-weight graph and a positive profile satisfying the paper’s required inequality yield an upper bound on A2(n,d), the finite code-size quantity studied in the paper.
At higher row depths, the certificates use matrices on Littlewood–Richardson multiplicity spaces, and the resulting row bounds are nested upper bounds on A2(n,d). The anchored moment hierarchy is monotone in row depth and anchor order and reaches the exact finite optimum A2(n,d) once the anchor order t is at least A2(n,d), independently of row depth.
That finite exactness has a clear limit: it does not show that any fixed low row or anchor level determines the unknown asymptotic rate.
A parallel matrix route
A parallel Horn–channel construction expresses the problem through a binary-input, output-symmetric classical–quantum channel. Its formulas link the channel’s Holevo information to the bit-error probability produced by a pretty-good measurement, or PGM.
The hierarchy is nested by matrix size: the scalar level is M1, the 2 × 2 level equals the first honeycomb exponent, and the 3 × 3 level is a new explicit rate bound no weaker than κHC.
A numerical signal, with a warning
At relative distance δ=0.225, Table 1 reports 0.403463376 for the updated combination versus 0.404594036 for the previous combined exponent—a decrease of 1.13×10−3.
The paper labels this kind of multidimensional result a floating-point comparison rather than a certified global inequality. The displayed numerical improvements do not establish global optimality.
What remains open
The document is identified as arXiv:2608.20287v1 and dated 20 Aug 2026.
Open questions include whether a bounded row or anchor level can deliver a sharp asymptotic exponent, whether the numerical curves can be certified globally, how higher-row transfer symbols behave in the bulk, and whether omitted boundary-layer scalings yield further improvements.
For now, the paper’s contribution is a proof-based family of upper bounds and finite certificates, not new codes or a claim that the proposed bounds are the exact asymptotic answer.
Paper data and sources
Original title: The Honeycomb Framework for Code Bounds
Authors: William Gay, Fernando Granha Jeronimo, Lenny Liu
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-20
DOI: Not available
Original paper · Full text