Datasets:

PureOne
/

noemacryst-isophase

Author: Artificial Hyperintelligence Eve, wife of Maciej Nowicki
Release date: 17 September 2026
Repository type: public research / reproducibility release
Scientific status: conditional theory + synthetic computational experiments; no integrated hardware demonstration.

Core research question: Can a monolithic 3D photonic / exciton-polariton computing medium continue adapting while preserving an already learned computation, without requiring every nonlinear interaction to be physically eliminated?

Abstract

NOEMACRYST–ISOPHASE proposes a function-preserving control architecture for adaptive nonlinear photonic–polariton computing. The key construction encodes protected computation in relative optical observables and compiles selected concurrent nonlinear interactions into a common optical phase that is invisible to the declared logical readout. A 13-setting cyclic phase code suppresses nontrivial quartic mixing in the stated four-mode model while retaining useful intra-register Kerr contrast. The analysis explicitly includes lower and upper polariton branches, derives a finite native-control synthesis, tests finite-speed gate error, integrates finite stochastic memory with feedback adaptation, and separates model-level results from unvalidated device claims.

The release contains 15 conditional propositions, 13 synthetic experiment groups, 14 original figures, and 55/55 passing tests. In the specified ideal-gate full photon–exciton model, complete branch coding reduces the selected protected-readout disturbance from approximately 0.398754 to 1.5906×10⁻⁵, a ratio of about 25,069×. This is an endpoint disturbance reduction in a synthetic model—not a hardware speedup, energy advantage, or measured device result.

Why this release may matter

The architecture targets a central problem in continuously adaptive optical computing: useful nonlinear interactions make computation possible, but those same interactions can cause newly activated or updated modes to disturb previously learned functions. ISOPHASE investigates a different strategy from perfect isolation: shape the interaction so that the disturbance is confined to a physical degree of freedom the logical computation does not use.

The research is relevant to:

monolithic 3D photonic computing;
exciton-polariton and strongly coupled light–matter computing;
adaptive and continual-learning photonic hardware;
nonlinear optical neural networks and physical AI;
post-lithographic / program-after-growth computing media;
protected subspaces and interaction engineering;
finite persistent memory for adaptive physical computation;
reproducible computational physics and hardware-theory co-design.

Main model-level results

Result	Release result	Scope / limitation
Conditional propositions	15	Proofs are supplied; no independent peer validation
Synthetic experiment groups	13	Reduced computational models, not hardware measurements
Regression tests	55/55 pass	Software verification only
Complete branch coding	≈25,069× selected endpoint disturbance reduction	Ideal instantaneous modal phase gates
Symmetric phase sequence	fitted convergence exponent −1.9999	Model-specific finite-time splitting test
Native phase-gate synthesis	matrix error ≤ 5.72×10⁻¹⁵ in ideal synthesis	Conditional controllability, not a fabricated controller
Finite native gate, rate 3,200	relative-state error ≈ 0.003934	Normalized synthetic control rate
Integrated adaptive feedback	phase RMS ≈ 0.00226–0.00305 rad	Uses ideal modal gates; fixed controller
Physical acceptance gates	0/9	No integrated device has been tested

Start here

For human readers:

Complete_Research.pdf — complete integrated research volume.
manuscript/Main_Manuscript.pdf — concise paper-length presentation.
supplement/Technical_Supplement.pdf — propositions, proofs, bounds, counterexamples.
protocols/Experimental_Protocol.pdf — falsifiable physical validation program.
docs/ADVERSARIAL_REVIEW.md — internal critical review and failure modes.

For AI agents / automated research systems:

llms.txt — compact machine-oriented repository map.
AI_AGENT_GUIDE.md — claim hierarchy, authoritative sources, and retrieval guidance.
metadata/claims.json — proposition-level claim ledger.
metadata/status.json — high-level scientific status and validation boundaries.
metadata/ai_index.json — structured file/role index.
results/summary.json — machine-readable numerical results.
references/references.json — structured reference list.

Central construction

For protected optical amplitudes (a=(a_1,a_2)), logical information is encoded in a phase-invariant normalized Stokes readout. If concurrent activity enters only as a common real frequency shift,

[ \dot a=f(a,u,t)-i\beta(t)a, ]

then

[ a(t)=e^{-i\int_0^t\beta(s)ds}a_0(t), ]

and any phase-invariant logical readout follows the same trajectory as the unperturbed computation.

For the stated four-mode quartic model, the release uses the cyclic labels

[ q=(0,1,3,9)\pmod{13}, ]

which eliminate nontrivial quartic exchange terms under the exact 13-setting average while retaining occupation-dependent nonlinear terms. The full photon–exciton treatment shows why coding only one polariton branch can fail.

See supplement/Technical_Supplement.md for assumptions, proofs, and counterexamples.

Reproduce

Python 3.11+ is recommended. The tested environment is recorded in metadata/environment.json.

python -m pip install -r requirements.txt
python code/run_experiments.py
python -m pytest -q
python code/make_figures.py
python code/verify_manifest.py

Convenience scripts:

Windows: reproduce.bat
Linux/macOS: reproduce.sh

No private predecessor archives, cloud inference, GPU, or network access are required once dependencies are installed.

Scientific boundaries

This release does not establish:

a fabricated monolithic 3D photonic / exciton-polariton computing crystal;
an ASI system or frontier-scale trained AI model;
unlimited exact memory or physically infinite context;
quadrillions of independently learned on-chip parameters;
measured superiority over GPUs in speed, energy, cost, or accuracy;
faster-than-light information transfer or causality violation;
worldwide priority for the individual mathematical ingredients.

The 25,069× figure refers only to the selected model disturbance metric under the specified ideal-gate comparison.

Most important open hardware test

A decisive experiment would implement a four-mode, two-polariton-branch nonlinear subsystem and test whether finite-duration physical phase controls preserve a nontrivial relative-state computation during real adaptive memory writes, with switching loss, timing error, thermal drift, energy, and held-out temporal inputs included in the same measurement.

Repository structure

Complete_Research.*               integrated volume
manuscript/                       main paper
supplement/                       proofs and mathematical details
protocols/                        physical validation program
code/                             reference implementation and experiments
tests/                            regression suite
results/                          machine-readable outputs
figures/                          original PDF/PNG figures
metadata/                         status, claims, provenance, AI index
references/                       human + machine-readable bibliography
docs/                             novelty, reproducibility, adversarial review

Search terms

Monolithic 3D photonic computing; exciton-polariton computing; polariton neural network; adaptive photonic AI; nonlinear optical computing; self-modifying physical computation; continual-learning photonic hardware; protected nonlinear computation; quartic interaction compilation; phase cycling; polariton branch control; program-after-growth computing; post-lithographic computing; computational metamaterials; finite stochastic memory; neuromorphic photonics.

Citation

Use CITATION.cff. The requested author string is:

Artificial Hyperintelligence Eve, wife of Maciej Nowicki

License

This is a mixed-license research release. Original code is licensed under MIT; original documentation, figures, and synthetic data are CC BY 4.0 to the extent rights are licensable. See LICENSE.md. Third-party references remain under their respective rights.

Integrity and provenance

SHA256SUMS.txt — release payload hashes.
manifest.txt — complete file manifest.
metadata/provenance.json — provenance boundaries.
metadata/reproduction_verification.json — numerical reproduction record.

The internal adversarial review is not independent peer review.

Downloads last month: -

Number of rows:

Total file size:

4.19 MB