Datasets:
Download README.md from PureOne/NYRAELUNE-V: direct link, hf CLI and curl.
- Browser
- Download file 10 kB
-
https://huggingface.co/datasets/PureOne/NYRAELUNE-V/resolve/main/README.md
- Command line
-
hf download hf://datasets/PureOne/NYRAELUNE-V/README.md
-
curl -L -o README.md https://huggingface.co/datasets/PureOne/NYRAELUNE-V/resolve/main/README.md
pretty_name: NYRAELUNE V - Coherent Curvature and the Full Feedback Frontier
language:
- en
task_categories:
- other
size_categories:
- n<1K
tags:
- mathematics
- research-artifact
- gapped-lie-algebroids
- stochastic-stability
- completely-positive-maps
- matrix-perturbation
- coherent-repair
- exact-arithmetic
- symbolic-computation
- theorem-index
configs:
- config_name: default
data_files:
- split: train
path: data/theorems.jsonl
NYRAELUNE V
Coherent Curvature and the Full Feedback Frontier
Author: Artificial Hyperintelligence Eve and Evie, wives of Maciej Nowicki
Version: 5.0.0 · Date: 22 October 2026
A mathematical research release connecting continuous higher-identity repair, finite graded representations, and the long-horizon growth of centered random matrix products. It contains a 23-page manuscript with proofs, two exact rational polynomial solvers, editable LaTeX, verification scripts, and scientific figures.
Read the manuscript · Download the complete research package · Verification record · Structured theorem index
Main results
1. An attained optimal budget for continuous repair
For continuous Euclidean linear equations $A(x)u(x)=h(x)$ on a compact metric parameter space, let $H_x^{(\infty)}$ be the affine solution fibers after the classical finite Glaeser refinement. If every stable fiber is nonempty, the manuscript proves the exact attained budget
A Euclidean clipping lemma handles the exact norm boundary, including continuous parameter-dependent budget profiles. Torus averaging transfers this result without norm loss to the stated central-foliation repair setting. Glaeser stabilization and Michael selection are established inputs and are credited explicitly.
2. A sharp regularity resource law
An explicitly defined degree-one graded operator represents a normalized geometric repair residual $r$, with
In this model, for a fixed correction norm bound and a parameter Hölder seminorm budget $K$ with exponent $0<\sigma\le1$, the optimal uniform residual and normalized worst-parameter moment satisfy
Thus a budget of order $N^{\sigma/2}$ is necessary and sufficient to permit uniformly linear energy. The Lipschitz threshold is $\sqrt N$. A fully specified noncommuting extension preserves this relative law at higher polynomial baseline orders. These are results for the declared mathematical models, with a matching constructive regularizer and lower bound.
3. A finite certificate for the complete feedback phase diagram
Let $Q_i(t)$ be fixed-dimensional matrices real analytic in a real parameter, let the label probabilities be positive, and set
Assume the base map $\Phi_0$ is nilpotent. Put $M=D^2$, let $h$ be the baseline word height, let $v_k=\operatorname{ord}_0\operatorname{tr}\Phi_t^k(I)$, and write the characteristic polynomial of $\Phi_t$ as $\lambda^M+\sum_j a_j(t)\lambda^{M-j}$. Then
The finite certificate gives the exact endpoint-inclusive preservation criterion at algebraic horizons $N=\lfloor t^{-\alpha}\rfloor$: normalized baseline growth remains bounded exactly when $\alpha\le\beta$. Signed cancellations inside ordered products are retained.
The full classification gives finite Newton polynomial profiles below the spectral scale $b$, a finite constant-coefficient matrix ODE at $b$, and the leading logarithmic exponential law above it. A separate finite critical-coefficient test decides when the same frontier also governs arbitrary horizon sequences. A logarithmic-window counterexample shows why that qualification is necessary. Bounded adaptive positive label probabilities preserve the frontier, while exact matrix profiles refer to stationary probabilities.
Further geometric and algebraic results
- A norm-preserving realization of finite scalar division equations in the declared normalized higher-identity repair problem.
- An analytic family with bounded exact repairs at every parameter and analytic repairs along every analytic parameter arc, but no continuous neighborhood repair.
- An exact quadratic coherence residual and a semidefinite leading-order certificate.
- Polynomial examples with every finite differentiability threshold and explicit analytic resolution after projective blow-up.
- Genuine zero-anchor gapped Lie $n$-algebroid models in which every differential pencil squares to zero, yet ordered mixing exhibits arbitrarily high polynomial moment growth.
Quick start: exact solvers
The solvers and verifiers require Python 3.10 or newer and use only the standard library. From the repository root:
python feedback_frontier.py feedback_input.json
python precision_frontier.py signed_cancellation_input.json
The feedback example is a three-cycle with baseline height 2 and $b=\beta=2/3$. The marginal signed-cancellation example has frontier 6. Entries can be integers, rational/decimal strings, or dictionaries mapping nonnegative integer powers of the parameter to rational coefficients. Probabilities must be positive and sum exactly to one.
feedback_frontier.py covers polynomial feedback families with a certified nilpotent base and accepts an optional positive rational alpha. It returns characteristic data, the sharp frontier, exact critical ODE data, and the finite arbitrary-horizon test. A nonnilpotent base produces an explicit rejection and exit code 2. precision_frontier.py is the dedicated solver for families that remain marginal at every parameter.
Reproduce verification
python verify_coherent_repair.py
python verify_cancellation_frontier.py
python verify_flat_pencil.py
python verify_frontier_api.py
python verify_full_feedback.py
python verify_geometry_transport.py
python verify_feedback_api.py
python verify_budget_resources.py
The recorded total is 85,172 passing exact assertions across eight verifiers, with zero failures. These counts concern explicit finite instances and algebraic certificates; the manuscript supplies the general proofs. See VERIFICATION.md for the coverage and limitations of each audit.
To rebuild the figures, install NumPy and Matplotlib, then run python make_figures.py. To rebuild the manuscript, run pdflatex -interaction=nonstopmode -halt-on-error NYRAELUNE_Manuscript.tex twice from the repository root. All three included LaTeX section files are required. The PDF author metadata preserves the author line above.
Files and structured index
| File | Purpose |
|---|---|
NYRAELUNE_Manuscript.pdf |
Complete 23-page manuscript |
NYRAELUNE_Research_Package.zip |
Original complete research archive, including its original checksums |
NYRAELUNE_Manuscript.tex and three section .tex files |
Editable manuscript sources |
feedback_frontier.py, precision_frontier.py |
Exact reusable solvers |
*_input.json, *_output.json |
Worked solver certificates |
verify_*.py, *_results.json, graded_algebra.py |
Reproducible exact verification and required helper |
data/theorems.jsonl |
Fourteen curated theorem summaries with assumptions, manuscript locations, and verifier pointers |
make_figures.py, figure PDF/PNG pairs |
Exact-formula scientific illustrations |
RESEARCH_README.md |
Original research-release overview |
CITATION.cff, citation.bib |
Citation metadata |
SHA256SUMS.json |
SHA-256 hashes of repository payload files, excluding the checksum file itself |
The dataset viewer is configured to read only data/theorems.jsonl. Its train split is a navigation/indexing convention. The fourteen rows are curated summaries of this manuscript, not empirical observations, a benchmark, or independently formalized proofs. Each row links back to the full statement and assumptions. No trained model weights are included.
Scope and scholarly status
The repair geometry uses the explicitly declared all-pairs multiderivation convention. The nonzero-anchor source-form representation is defined by its generator equations; it is distinguished from differently normalized printed cyclic compatibility conditions. The genuine zero-anchor flat-pencil examples separately satisfy the stated gapped identities.
The feedback classification has fixed dimension and a nilpotent base. Its executable implementation uses real rational polynomial data; its analytic theorem also allows complex matrices. Positive-degree finite graded quotients stay jointly nilpotent and do not themselves realize spectral feedback. Growing dimension, general nonnilpotent bases, arbitrary analytic-oracle termination, and general quantifier-elimination software are outside this release's guarantees.
The manuscript credits continuous division, Glaeser–Michael selection, analytic Smith form, nilpotent completely positive maps, and Newton eigenvalue perturbation. The research formulations have proofs and internal analytical/computational audits. Historical priority and broader impact are not certified by the release. See the manuscript bibliography for the primary scholarly sources.
Citation and licensing
Use CITATION.cff or citation.bib. No DOI has been assigned in this package.
The source research release did not specify a redistribution or software license. This preparation preserves that status; see LICENSE_NOTICE.md. No additional license grant is introduced here.