--- pretty_name: "EVE-QELTHA — Coherent Subspace Transport, Final Research Release 4.0" language: - en size_categories: - n<1K tags: - mathematics - homogenization - spectral-theory - variational-methods - matrix-analysis - scientific-computing - reproducible-research - synthetic configs: - config_name: critical_crossover default: true data_files: - split: train path: data/critical_crossover.csv - config_name: compact_ports data_files: - split: train path: data/compact_ports.csv - config_name: critical_compensation data_files: - split: train path: data/critical_compensation.csv --- # EVE–QELTHA: Coherent Subspace Transport **Critical filtering, universal matrix realization, and nonlocal hyperhorn capacity** Final research release **4.0 · 20 October 2026** **Artificial Hyperintelligence Eve**, in research collaboration with **Maciej Nowicki** A self-contained mathematical manuscript, editable proofs, ten numerical verification suites, and their recorded results. The central result gives an explicit critical homogenization law for one bright drainage channel, together with a constructive method for realizing a prescribed nonnegative matrix potential. A second result classifies the leading capacity response of compact nonlocal ports, including exact hard constraints. **Start here:** [37-page manuscript](research/EVE_QELTHA_Final_Manuscript.pdf) · [Complete research ZIP](EVE_QELTHA_Final_Complete_Research_Package.zip) · [Reproduce](REPRODUCE.md) · [Data dictionary](DATA_DICTIONARY.md) · [Claim ledger](research/CLAIM_LEDGER.json) ## Main result: the complete critical crossover On a finite interval, construct a moving unitary frame with $k$ dark directions and one bright direction. The rank-one drainage projector $M_n$ has the same spectrum $\{0^k,1\}$ everywhere and approaches a constant projector uniformly. For the passive energy $$ q_n[P]=\int_0^L\left(w\|P'\|^2+\rho_n gP^\dagger M_nP\right)\,dt, $$ assume $\rho_n\to\infty$ and $\rho_n/n^2\to\tau\in[0,\infty]$. The effective dark-sector energy is $$ q_\tau[\xi]=\int_0^L w\left(\|\xi'+A\xi\|^2+ \xi^\dagger V_\tau\xi\right)\,dt, \qquad V_\tau=T^\dagger D_\tau T, $$ $$ (D_\tau)_{jj}=\frac{\tau g}{\tau g+w\omega_j^2}, \qquad \omega_j=\frac{2\pi j}{L}, \qquad D_0=0,\quad D_\infty=I. $$ Here $w,g$ are positive $C^1$ weights, $A$ is a $C^1$ skew-Hermitian connection, and $T$ is an arbitrary $C^1$ matrix factor. Matrix order matters. The manuscript supplies compactness, a dual Fourier lower bound, an endpoint-compatible recovery sequence, inlet-admittance convergence in matrix norm, and convergence of every fixed Dirichlet eigenvalue, including multiplicities. For any fixed $\tau>0$ and desired positive semidefinite target $W$ with a $C^1$ square root, choose $$ T=D_\tau^{-1/2}W^{1/2}. $$ Then $T^\dagger D_\tau T=W$. This realizes the target limiting energy in ambient dimension $k+1$ using one bright channel. The finite-frequency response approaches that target; exact equality at finite $n$ is not asserted. ## Compact ports and hyperhorn capacity For the two-component weak-twist horn, let $\kappa_{\varepsilon,\lambda}$ be the inlet capacity after adding $\lambda\|LP\|^2$, with a fixed finite-rank compactly supported $L^1$ port $L$. In the stated supercritical twist range, the baseline is $k_0\varepsilon^\gamma$, $0<\gamma<2$. If the port is balanced, $Le_2=0$, the complete leading asymptotic is uniform over all strengths: $$ \sup_{\lambda\in[0,\infty]} \left|\frac{\kappa_{\varepsilon,\lambda}}{k_0\varepsilon^\gamma}-1\right|\longrightarrow0. $$ The endpoint $\lambda=\infty$ imposes the exact constraint $LP=0$. If $Le_2\ne0$, jointly as $\varepsilon,\lambda\to0$, $$ \kappa_{\varepsilon,\lambda} =k_0\varepsilon^\gamma+\lambda\|Le_2\|^2 +o(\varepsilon^\gamma+\lambda). $$ The explicit sum and difference ports have equal kernel diagonals almost everywhere and equal isolated operator spectra, but different capacity asymptotics relative to the same derivative energy. ## Further results included The manuscript also develops coherent-history/output rank classification, exact Bessel impedance laws, the $15/8$ angular exponent, the common transported-kernel capacity criterion, log-periodic capacity and finite-band inverse synthesis, exact matrix frame realization with sharp worst-case dimension $2k$, and a sharp first-order finite-penalty boundary correction. ## Numerical reference data All rows are deterministic synthetic calculations from the shipped verification programs. They are not measurements. The three configurations expose selected results in a tabular form; the original JSON reports retain complete matrices, parameters, refinement checks, and solver diagnostics. | Configuration | Rows | Contents | Original report | | --- | ---: | --- | --- | | `critical_crossover` | 40 | Two coupling factors, four critical ratios, five spatial frequencies | [critical.json](research/results/critical.json) | | `compact_ports` | 35 | Balanced and anchored strengths, hard constraints, and refinement runs | [nexus_uniform.json](research/results/nexus_uniform.json) | | `critical_compensation` | 4 | Prescribed complex matrix target at $\tau=40$ | [critical.json](research/results/critical.json) | The split name `train` is a Hub loading convention. No statistical training/test split or independent sampling interpretation is assigned. These small parameter sweeps support theorem auditing and solver comparisons within the documented models. ```python from datasets import load_dataset critical = load_dataset( "PureOne/EVE-QELTHA", "critical_crossover", split="train" ) ports = load_dataset( "PureOne/EVE-QELTHA", "compact_ports", split="train" ) ``` The examples above apply after this repository is published at the named destination. Local CSV files can be used immediately. [export_tables.py](export_tables.py) regenerates all three tables directly from the original reports, without running a solver or rounding numerical values. ![Computed critical law and compact-port behavior](research/figures/critical_results.png) ## Status and verification | Item | Status | | --- | --- | | Declared release assembly | Complete: manuscript, proofs, programs, results, and manifests included | | Executed numerical suites | **10/10 passed — 100% of the shipped programs** | | Manuscript layout | 37 pages; compiled and visually inspected | | Mathematical review | Audited within the research collaboration | | Independent external peer review | Not established | | Literature-first novelty | Unverified | | Physical realization or experimental validation | Not established | The 100% figure is a program pass percentage. It is not a probability of correctness, a novelty score, or an experimental validation rate. The [verification summary](research/results/verification_summary.json) records source hashes, dependency versions, exit codes, and timings. ## Scope and limitations The theorems concern the specified passive linear reduced variational models. The critical theorem assumes finitely many periodic frequencies and smooth slow coefficients. Its general quantitative rate and a resolvent estimate remain open. The compact-port uniformity theorem assumes a fixed finite-rank compact $L^1$ port. It does not cover inlet atoms, supports escaping to infinity, or arbitrary infinite-rank operators. Critical compensation has no uniform resource bound as $\tau\downarrow0$. Full fluid PDE derivations, ordinary hydraulic-network realization, nonlinear laws, and experimental performance are outside this release. The baseline horn theorem assumes $s>(a+1)/2$; its separate threshold and subcritical regimes remain open. See the [claim ledger](research/CLAIM_LEDGER.json) for precise boundaries. ## Established foundations and priority Universal connections, dark-state geometric potentials, energy-form traces, oscillatory constraints, and several scale-ratio regimes have established predecessors. The manuscript compares its results with those foundations and gives primary references. Relevant examples include [Tewary's singular-perturbation/homogenization regimes](https://doi.org/10.3934/nhm.2021012), [Braides–Cancedda–Chiado Piat on oscillating manifolds](https://doi.org/10.1051/cocv/2016018), and [Massamba–Thompson on universal connections](https://arxiv.org/abs/math/0311198). Exact priority for this combination of results has not been independently certified. ## Reproduction and files From the repository root: ```bash python -m pip install -r research/requirements.txt python research/code/run_all.py python export_tables.py ``` The numerical suites use local CPU computation. Rebuilding figures additionally uses Matplotlib; rebuilding the PDF requires a TeX installation. Detailed commands, integrity verification, dependency versions, and loading examples are in [REPRODUCE.md](REPRODUCE.md). The `research/` directory preserves the original portable release, including its own README and integrity manifest. The root [SHA256SUMS.json](SHA256SUMS.json) covers the staged Hugging Face repository. The complete research ZIP is also retained as a standalone download. ## Authorship, citation, and license The work is attributed to **Artificial Hyperintelligence Eve**, in research collaboration with **Maciej Nowicki**. It was developed with AI assistance; this release does not represent independent human peer review. Citation metadata is supplied in [CITATION.cff](CITATION.cff). No redistribution or reuse license was specified in the source release. This preparation adds no new license grant; see [LICENSE_STATUS.md](LICENSE_STATUS.md).