# Exact spectral topology acceptance and construction research

Revised dossier, 9 October 2026 Australia/Brisbane. Decision: **Retain exact finite acceptance; defer the literal full constructor**. Buyer demand, commercial novelty and profitability remain unvalidated.

## Research finding

Family 178 claims deterministic polynomial-bit-time output of simple fixed-degree nonbipartite Ramanujan graphs on every sufficiently large even size, with degree-dependent exponent. Its algorithm section supplies a rational contrast-space test for the strict nonconstant spectral bound. Family 174 supplies a separate deterministic thin-tree construction for edge-connected multigraphs; these graph properties serve different objectives.

## Problem and buyer

Distributed-simulation and graph-benchmark teams need reproducible bounded-degree topologies at chosen sizes and an honest spectral-property record. A rounded eigenvalue near the threshold can be ambiguous. Operational network design also requires latency, bandwidth and failure-domain constraints beyond graph expansion.

## What the finding could enable

A library could combine a deterministic graph constructor, exact finite spectral audit and deployment-specific simulator adapter. The immediate prototype checks small supplied simple regular graphs using exact integer/Fraction arithmetic and the paper’s contrast quadratic form. A full new graph constructor remains to be translated and benchmarked.

## Technical and commercial limits

The source fixes `a=1/10000` and an auxiliary spectral-parameter group of `kh=ceil(10/(0.03a))=3,333,334` points below the largest primary, using both signs: 6,666,668 signed auxiliary points during early/cleanup phases before counting other groups. These are parameter points, not a claimed polynomial exponent or high-moment order. Processing/storage can be organized differently, so no memory lower bound is inferred. The construction’s sufficiently-large threshold, polynomial degrees and other constants remain numerically uninstantiated. The paper uses rational matrices, structural dynamic programs, candidate enumeration and repair; polynomial time alone does not establish practical crossover. Expansion is not a guarantee of real hardware throughput, congestion or resilience. The finite checker’s 64-vertex cap is a prototype budget, not a theorem restriction.

## Minimal architecture

Adjacency manifest -> loop/symmetry/regularity validation -> exact contrast-form LDL test -> saved rational pivots -> topology simulator -> construction and deployment cost report. Add the constructive paper algorithm only after its parameters are effective and measured. Thin-tree modules belong in a separate cut-sparsity layer.

## Existing alternatives and differentiation

NetworkX already includes explicit and random expander generators and expander checks. The proposed addition is the particular all-admissible-size deterministic construction, if usable, plus an exact rational spectral audit. Existing approximate or randomized generators are necessary practical baselines. [NetworkX graph generators](https://networkx.org/documentation/stable/reference/generators.html).

## Monetization hypothesis

Hypothesis: an open library with AUD 5,000–15,000 funded simulation or validation integrations. At an illustrative AUD 8,000 integration, 35 engineering hours costed at AUD 150/hour leave AUD 2,750 before maintenance, compute and overhead. A commercial topology product requires measured operational improvement beyond a theoretical spectral property.

## Validation experiment

The exact checker passes K4 and the Petersen graph, and rejects the bipartite cube, K3,3 and a disconnected pair of K4 graphs, matching their known nonconstant spectra. The source repair stage enumerates `P^2` affine seeds with prime `P` of order `n^Cprime`, where `Cprime` is a sufficiently large fixed constant; it can stop at the first passing candidate. All-degree or local-query claims are outside the fixed-degree adjacency-list theorem. No numerical crossover is established. Next compare exact acceptance on supplied existing-generator candidates before considering a translated full constructor or application latency benchmark.

## Conditions to reject or defer

Reject a paid constructor if degree-dependent costs are prohibitive, thresholds miss target sizes or existing random generators suffice. Defer claims of formal program verification or production network advantage. The finite exact check is evidence about its input graph under this implementation, not independent verification of the construction theorem.

## Next concrete action

Document all remaining `n0(d)` requirements and sufficiently-small/large constants for degree three, including auxiliary-group separation and repair precision. Keep the existing exact checker as an input-graph acceptance component. Do not substitute heuristic generation plus certification for the claimed deterministic polynomial constructor; a redesigned or translated constructor remains a separate R&D task.

## Pinned research sources

- Family 178: [Deterministic nonbipartite Ramanujan graphs in every fixed degree](https://github.com/openai/math/blob/fd4aeeb2ee4fc729c18d98444fed42fd0529eeeb/preprints/Deterministic-nonbipartite-Ramanujan-graphs-in-every-fixed-degree-September-23-2026/paper.pdf).
- Family 174: [The strong thin tree conjecture](https://github.com/openai/math/blob/fd4aeeb2ee4fc729c18d98444fed42fd0529eeeb/preprints/The-strong-thin-tree-conjecture-September-23-2026/paper.pdf).
- Family 174: [selected formal scope](https://github.com/openai/math/blob/fd4aeeb2ee4fc729c18d98444fed42fd0529eeeb/lean/docs/174.md); no independent check run here.

## Selected construction sections and saved parameters

The full introduction, setup/parameter hierarchy and algorithm section were read, together with the first 115 lines of accounts, the exact-repair theorem and the affine seed enumeration block. The full early, cleanup, structural and repair proofs remain pending. No family formal scope is listed. [Parameter/setup source](https://github.com/openai/math/blob/fd4aeeb2ee4fc729c18d98444fed42fd0529eeeb/preprints/Deterministic-nonbipartite-Ramanujan-graphs-in-every-fixed-degree-September-23-2026/build/sections/setup.tex), [repair source](https://github.com/openai/math/blob/fd4aeeb2ee4fc729c18d98444fed42fd0529eeeb/preprints/Deterministic-nonbipartite-Ramanujan-graphs-in-every-fixed-degree-September-23-2026/build/sections/repair.tex), [algorithm and exact contrast acceptance](https://github.com/openai/math/blob/fd4aeeb2ee4fc729c18d98444fed42fd0529eeeb/preprints/Deterministic-nonbipartite-Ramanujan-graphs-in-every-fixed-degree-September-23-2026/build/sections/algorithm.tex), [saved parameter record](../../snapshots/2026-10-08-baseline/spectral-constructor-parameters-2026-10-09-v1.json).