Skip to content

Forward Gaussian rounding: parity-flux obstruction below half slack - #7841

Open
CubeJerry wants to merge 1 commit into
CrocSwap:mainfrom
CubeJerry:research/forward-gaussian-rounding-20261120
Open

CubeJerry wants to merge 1 commit into
CrocSwap:mainfrom
CubeJerry:research/forward-gaussian-rounding-20261120

Conversation

@CubeJerry

Copy link
Copy Markdown
Contributor

Result

Any total injective error-one Hadamard/Gaussian rounder with forward
radial slack below one half must change infinitely many inputs whose
ideal images are already integral. The proof gives a quantitative bound:
at slack at most 3/10, at least (8m) even-sum inputs of norm at most
(4m+2) must change, for every (m\geq0).

In exact half-grid coordinates, an odd-width cube has equal input/output
color counts but eight inaccessible output corners. Each parity-changing
exit forces a nearby exact-valued input to change. If (K_N) counts
these changes by transformed output cube, the result is

[
K_{N+2}-K_{N-2}\geq8
\quad\text{when}\quad
\delta<N-\sqrt{N^2-N+1}.
]

The parity-preserving subclass consequently has forward optimum one half,
using #7569's credited upper construction. Surjectivity, exact sums,
phase covariance and period identities are not premises of the new
lower theorem.

Why this helps

#7651 already attains the unrestricted finite threshold through radius
24. This contribution identifies what any all-lattice extension must
change: a finite repair of exact-valued inputs cannot suffice. It does
not claim the unrestricted optimum or repeat larger finite-radius work.
#7696 and #7714/#7743 concern additional exact-sum and iterate contracts.

Checks and limitations

Run:

python3 -B scripts/research_package.py --package research/gaussian-forward-parity-flux

The exact checks cover 2,401 coordinate inputs, four complete cube graphs,
49,664 two-edge choices, equality/escape controls, 9,600 rational cutoffs,
and a separately verified 3,121-input unrestricted finite matching.
Holding its even inputs exact gives only 3,113 matches, with an explicit
321-input/313-neighbor Hall witness certifying the upper bound separately
from the search. Collision and nonlocal-edge mutations reject. The
universal statement rests on the written proof, not finite exhaustion.

No selected caller, physical tariff, positive-density claim or
unrestricted all-lattice construction is supplied. All five foundational
assumptions remain; there is no new kappa.

Sources and credit

Base: e545235. Exact predecessor heads,
scope, proof, receipt and remaining questions are in the package.
Prepared for Jerry Chen / @CubeJerry with substantial OpenAI assistance.
Credit #7569 for the half-grid coordinates and upper construction,
#7571 and Nick Petalas's #7651 for the finite Hall problem, Ali Romman's
#6509 for the Gaussian pair, and #6619/#2589 for the radial-obstruction
lineage. Model-assisted review is not human or formal-kernel acceptance.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant