jaxfne

Tensor-Field Neural Equations (TFNE) expressed as a JAX simulation engine for layer-resolved neural circuits, source operators, field proxies, probes, objectives, and evidence receipts.

Scientific grammar: Emitter → Source → Field → Probe → Objective → Optimizer → Manifest

Execution grammar: CircuitSpec → construct → Model → simulate → Signals

CircuitSpec is a conceptual category (Configuration | NeuronalTensor), not a concrete public production class — construct accepts either a Configuration (original path) or a NeuronalTensor with a RuntimeConfiguration (tensor-first path). It is unrelated to the experimental jaxfne.experimental_hpc.CircuitSpec type, which is not accepted by production construct.

Adaptation (optional HDP family): finite-dimensional hidden biophysical state $H$ and adaptive parameter coordinates $\Theta$ (synaptic and intrinsic), mediated by

$$ \dot X = F_X(X,H,\Theta,U),\quad \dot H = F_H(H,X,\Theta,U),\quad \dot\Theta = F_\Theta(H,X,\Theta). $$

RBS represents $H$. RBD defines $H$ dynamics. HDP defines parameter dynamics and may depend on $H$. $H$ can exist and evolve without HDP. See H-state / HDP guide.

Install

pip install jaxfne
pip install "jaxfne[viz]"   # optional plotting

Development: pip install -e ".[dev,viz]" after cloning.

Minimal example

import jaxfne as jtfne

jtfne.enable_x64()
tensor  = jtfne.load_canonical_neuronal_tensor("canonical-v1-column-1000n")
model   = jtfne.construct(tensor, jtfne.RuntimeConfiguration(seed=0, duration_ms=1000.0, dt_ms=0.5))
signals = jtfne.simulate(model)

Optional: generative development (JDNA)

Instead of loading a fixed tensor, generate one from a PseudoGenome:

genome  = jtfne.load_canonical_pseudogenome("canonical-v1-column-1000n")
tensor  = jtfne.develop(genome, seed=0)   # K_D: development seed
model   = jtfne.construct(tensor, jtfne.RuntimeConfiguration(seed=1, duration_ms=1000.0, dt_ms=0.5))
signals = jtfne.simulate(model)

JDNA is an optional path; the direct Configuration/NeuronalTensor paths remain first-class. See JDNA guide.

JaxFNE composes established neural dynamical models, source/readout operators, and JAX-native execution under the TFNE typed state/map grammar. See TFNE theory and References.

Import convention: import jaxfne as jtfne. Builder paths, paradigms, and optimization: Quickstart.

Relative and calibrated outputs

Simulated quantities are relative by default. Calibrated claims require an explicit transformation with evidence. See Scope & status.

Documentation

Resource	Link
Quickstart	docs/quickstart.md
Site	jaxfne.readthedocs.io
Tutorials	docs/tutorials/
Études	docs/etudes/
Frozen compatibility contract (0.4.13)	docs/public_surface_contract.md
Changelog	docs/changelog.md

For AI agents

Read and inspect artifacts/context.md.

It is a router rather than a specification: it names the single public import, the two invariant grammars, where authority actually lives, and the gate commands that settle a claim. Reading and inspecting that one file is enough to orient a coding agent — one working in a terminal, in a sandbox, or under any memory- or skill-based harness — without re-deriving the public API from source or guessing at the internal layout.

The file is present in the repository and in the source distribution. It is not installed alongside the wheel, so an agent working from an installed package should read it at the link above.

Citation

CITATION.cff · citation guide

Canonical Visualization Atlas

The JaxFNE Canonical Atlas provides a unified 6-panel visual grammar (network_3d, connectivity, raster, traces, spectral, state_summary) with strict evidence-level separation (OBSERVED vs. DERIVED), deterministic degradation tracking, and cryptographic manifest provenance.

Every preview below is generated directly from realized JaxFNE simulation outputs (canonical-v1-column-1000n scaffold):

_{1. Network 3D (OBSERVED): Realized 3D geometry, lamina, cell classes, sampled synaptic edges}	_{2. Connectivity (OBSERVED): Realized synaptic weight matrix & adjacency}
_{3. Spike Raster (OBSERVED): Microsecond spike timestamps across realized neuronal populations}	_{4. Membrane Traces (OBSERVED): Somatic membrane potential $V_m$ trajectories}
_{5. Spectral (DERIVED): Welch PSD & spectrolaminar power estimates}	_{6. State Summary (DERIVED): Cell-type rate distributions and silence fractions}

Generate the complete standalone HTML atlas suite locally with one line:

import jaxfne as jtfne
from jaxfne.vis import build_atlas

# Run simulation
tensor = jtfne.load_canonical_neuronal_tensor("canonical-v1-column-1000n")
model = jtfne.construct(tensor, jtfne.RuntimeConfiguration(seed=0, duration_ms=500.0, dt_ms=0.5))
signals = jtfne.simulate(model)

# Build canonical atlas with manifest & standalone interactive HTML panels
manifest = build_atlas(model, signals, out_dir="docs/_static/atlas")
print(f"Atlas generated with SHA256: {manifest['sha256']}")

For extended case studies, laminar field readouts, and homeostasis dynamics, see Showcases.

Release files for jaxfne 0.4.21

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for jaxfne 0.4.21
File	Size	Uploaded
jaxfne-0.4.21.tar.gz	38.0 MB	Sep 8, 2026	Details

Built distribution (wheel)

Table of built distributions (wheels) for jaxfne 0.4.21
File	Interpreter	ABI	Platform	Reset
jaxfne-0.4.21-py3-none-any.whl 705.8 kB Sep 8, 2026	Python 3	none	any	Details

Total release size: 38.7 MB

Release files / jaxfne-0.4.21.tar.gz

Download URL	jaxfne-0.4.21.tar.gz
Size	38.0 MB
Tags	Source
SHA-256 checksum How to use checksums	`0a820a04427ef536f7d8f72c3bf3cf6b5122b76171389a7d45d61359346b5892`
BLAKE2b-256 checksum How to use checksums	`d5107707b80b6dd2918d10b8a66fa8d641c58b21e4d1a010fe6c8267ada73798`
Upload date	Sep 8, 2026
Uploaded using Trusted Publishing? What is trusted publishing?	Yes
Uploaded via	`twine/7.0.0 CPython/3.13.14`

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

Identity

Publishing platform: GitHub Actions
Publishing repository: github.com/HNXJ/jaxfne
Publishing commit: github.com/HNXJ/jaxfne/tree/88235208950425ce76723115f1cb61fe3a828ab2

Workflow

Publishing configuration: github.com/HNXJ/jaxfne/blob/88235208950425ce76723115f1cb61fe3a828ab2/.github/workflows/publish.yml
Publishing logs: github.com/HNXJ/jaxfne/actions/runs/34237047408/attempts/1

Artifact

Subject: jaxfne-0.4.21.tar.gz
SHA-256 checksum: 0a820a04427ef536f7d8f72c3bf3cf6b5122b76171389a7d45d61359346b5892 Verifying attestations

Release files / jaxfne-0.4.21-py3-none-any.whl

Download URL	jaxfne-0.4.21-py3-none-any.whl
Size	705.8 kB
Tags	Python 3
SHA-256 checksum How to use checksums	`b2659d5db2cc63764e988660ecda4f68d5d00a09051e37078a3af9c2ce0f42ca`
BLAKE2b-256 checksum How to use checksums	`58f372680f5b8e646257bc2158e1005b0e8413b4cf475158003ae353fcdf73d3`
Upload date	Sep 8, 2026
Uploaded using Trusted Publishing? What is trusted publishing?	Yes
Uploaded via	`twine/7.0.0 CPython/3.13.14`

Provenance

Identity

Publishing platform: GitHub Actions
Publishing repository: github.com/HNXJ/jaxfne
Publishing commit: github.com/HNXJ/jaxfne/tree/88235208950425ce76723115f1cb61fe3a828ab2

Workflow

Publishing configuration: github.com/HNXJ/jaxfne/blob/88235208950425ce76723115f1cb61fe3a828ab2/.github/workflows/publish.yml
Publishing logs: github.com/HNXJ/jaxfne/actions/runs/34237047408/attempts/1

Artifact

Subject: jaxfne-0.4.21-py3-none-any.whl
SHA-256 checksum: b2659d5db2cc63764e988660ecda4f68d5d00a09051e37078a3af9c2ce0f42ca Verifying attestations