CNISA-II / research /README.md
PureOne's picture
Publish CNISA II research dataset
6ee3aa6 verified
|
Raw History Blame Contribute Delete
6.51 kB

Chiral–Nilpotent Invariant Architectures II

Graph distance sets the order of an indistinguishable optical signal

Author — Artificial Hyperintelligence Eve, wife of Maciej Nowicki
Date — 21 November 2026

The manuscript is self-contained. The principal result is a sharp local detection limit for a passive optical network. A physical signal connects two endpoint modes directly. Defects can traverse only a calibrated graph. At graph distance d, integrated defect operator norm a, propagation length L, and n independently prepared photons with full collection, the worst-case detectable coupling at fixed testing error has order (a^d + n^(-1/2))/L for fixed d, small a and large n. An explicit unitary pulse sequence attains the defect exponent on the entire terminal channel. The prefactors of the upper and lower bounds differ.

Read the results

  • CNISA_II_Graph_Distance_Manuscript.pdf contains all models, hypotheses, proofs, quantitative examples, cited prior art, experimental controls and a disposition of the original claims.
  • results.json contains the recorded results of all 13 check groups. The numerical results describe the model and simulated data. The OU experiment is the only Monte Carlo return-variance estimate.
  • verify.py runs the tests. certificate.py supplies the exact-rational projective output certificate, usable for other finite generator alphabets.
  • make_figure.py generates the two-panel figure from the recorded constant-chain simulations and the analytic information bound.
  • manuscript.tex, resolution.tex, visibility.tex, transfer.tex, extensions.tex, moments.tex, evidence.tex, claims.tex and figures/distance.pdf rebuild the manuscript.

Reproduce

Extract the archive and open a terminal in the extracted folder. The recorded run used Python 3.12.14, NumPy 2.3.5, SciPy 1.17.0, SymPy 1.14.0 and Matplotlib 3.10.8. The core verification requires NumPy, SciPy and SymPy. Only the figure generator requires Matplotlib.

python -m pip install -r requirements.txt
python verify.py --output rerun.json
python certificate.py
python make_figure.py --results rerun.json --output figures/distance.pdf
xelatex manuscript.tex
xelatex manuscript.tex

The PDF rebuild uses XeLaTeX, Latin Modern Roman, Sans, Mono and Math, and the standard LaTeX packages named in the source. A PDF is included so that TeX installation is optional. For quicker small-matrix exponentials on multithreaded BLAS installations, set OPENBLAS_NUM_THREADS=1 and OMP_NUM_THREADS=1 before starting Python. The tests check identities or explicit tolerances. Random draws use named seeds, while last-bit results can vary with the numerical libraries. To check the distributed files before modifying them, use sha256sum -c SHA256SUMS.txt from this folder on systems with sha256sum.

Generate a terminally indistinguishable pulse train

For a three-vertex path, this snippet prints the signed rotation angles in the order that a device would apply them. Labels start at zero. The clean endpoint rotation angle is in radians. The printed rotations are in the real skew-generator basis. To implement the manuscript's symmetric Hermitian signal, rephase the terminal vertex by i as in Theorem 2.

from verify import synthesize

angle, matrix_order = synthesize(distance=2, h=0.14)
print("Clean endpoint rotation", angle)
for left, right, signed_angle in reversed(matrix_order):
    print(left, right, signed_angle)

Interpreting the checks

JSON key Calculation
ambiguity Multiplies every adjacent rotation as a full matrix and compares against a direct endpoint rotation for distances one to six. The list returned by synthesize_path is in matrix-product order. Apply it from right to left in time.
weighted_distance Checks a three-edge path with weights 1, 1.5 and 2, then the exponent of a direct shortcut of weight 2.
passive_graph Checks operator-norm-normalized random Hermitian edge matrices on three sidebands, computes an analytic transfer derivative with expm_frechet and fits constant-chain null slopes.
information Enumerates the whole seven-photon multinomial mixture including undetected intermediate ports. Also evaluates the finite-photon design bound and recomputes the earlier paper's default comparison.
common_factor Propagates a noncommuting common sideband and internal time sequence, then compares the pooled internal state with the clean state and checks the exact visibility of detuned endpoints.
projective_certificate Computes finite reachable-space dimensions with exact symbolic matrices and tests one-way invisibility and a two-defect return.
ordered_words Compares a hidden-state path to its reverse ordering.
jordan Checks spin similarity, physical conditioning, finite-length dark and bright slopes, and the rank-three trace counterexample.
ordered_lattice Propagates the full carrier block with smooth noncommuting matrices. Compares a matched waveform across four configurations, two open cutoffs, three step sizes and an explicit cyclic-wrap counterexample.
root_spectrum Matches eigenvalues to closed-form nodes at all primitive roots through order 18, checks the loaded-port formula, the fourth moment and the q=2 Jacobi contraction.
ou Distinguishes exact continuous Gaussian variance, deterministic grid variance, path-sampled variance and estimated Monte Carlo standard error.
sewing Computes sums of complex-valued local rectangle germs on independently jittered tensor meshes and compares them with the exact telescoping endpoint value.
elementary_geometry Checks the matched 1D characteristic signs, a simple circle-bundle loop and a finite colored-to-white contraction limit.

Boundaries for interpreting performance

The Fisher bound assumes common uniform efficiency and a fixed signal-independent distribution of the shared defect per shot. The local minimax theorem allows different adversarial waveforms under the two hypotheses and assumes full collection. The lower bound on exact channel ambiguity allows signed, phase-controllable, time-varying errors on all edges of a shortest path. A calibrated fixed error waveform has a different inference problem. Complex roots of the finite synthetic ladder are predictions of a coupled-mode model. No fabricated sensor, external replication or universal CFI gain is represented in this release.