Jansen–Rit source, runtime, and co-simulation fidelity

This page records the publication boundary, independent equation trace, five-runtime parity, failure atomicity, fixed-point envelope, and controlled benchmark used to promote JansenRitUnit.

Primary-source boundary

Jansen and Rit (1995) model one cortical column as interacting pyramidal, excitatory-interneuron, and inhibitory-interneuron populations. Equation (6) uses three second-order synaptic operators and the connectivity constants

[ C_1=C, \qquad C_2=0.8C, \qquad C_3=C_4=0.25C. ]

For the excitatory feedback term, the published wiring is (C_2S(C_1y_0)). The (C_2) factor therefore multiplies the sigmoid output; it does not scale the sigmoid input.

Primary source: Jansen and Rit 1995. Source-bound executable transcription: pinned Brian2 example.

Timestep and sigmoid convention

The publication defines continuous equations and does not require a particular numerical solver. The pinned Brian2 example uses explicit Euler at 0.1 ms; SC-NeuroCore adopts that step as the maintained default and records the solver choice as implementation scope rather than attributing it to the paper.

Brian2 names the complete sigmoid numerator e0=5. SC-NeuroCore names the half-maximum rate e0=2.5 and evaluates 2*e0/(1+exp(...)); the two conventions are algebraically identical at their defaults.

Independent reference trace

src/sc_neurocore/neurons/reference_trace_data/jansen_rit_eq6_euler_brian2.json pins a 256-step trace with:

• a deterministic, varied external pulse-density drive;
• every post-update state and the y1-y2 EEG proxy;
• the DOI, Brian2 URL, and exact source commit;
• first, final, minimum, maximum, and mean features for all seven traces.

tests/test_reference_jansen_rit.py independently re-derives equation (6) without importing production equation helpers, checks every recorded feature, and compares every production state exactly. The canonical interleaved little-endian float64 digest is 84e9273e381d543a5fb32a510a4c82ea977c32cdd267e88b790296e6e8364933. The deterministic drive is a parity protocol, not a paper-figure reproduction.

Numerical and atomic contract

Each scalar step:

1. validates parameters, the complete current state, and external drive;
2. evaluates all sigmoid responses from the pre-update state;
3. evaluates all six simultaneous Euler candidates;
4. validates the complete candidate;
5. commits all six states together.

The sigmoid uses a branch-stable exponential form. Public batch dispatchers also validate all seven trace shapes, finiteness, the eeg == y1 - y2 identity, and every final-state receipt before mutating the model instance. Empty batches preserve all six states.

Executable parity matrix

Runtime	Executed surface	Enrolled result
Python	scalar reference and atomic batch	reference
Rust engine	modular PyO3 scalar and batch	seven traces/six final states within 1e-11
Rust safety	standalone rustc --test module	equation wiring, validation, atomic mutation, reset
Julia	JuliaCall batch	seven traces/six final states within 1e-11
Go	generated C-shared ABI	seven traces/six final states within 1e-11
Mojo	exported shared-library C ABI	seven traces/six final states within 1e-8

The configured test begins from non-zero values for all six states, changes every public parameter, supplies a varied 128-step drive, and compares all seven trajectories plus all six final states. Explicit selection of an absent runtime fails rather than substituting the Python path.

Paired schema and Q32.32 co-simulation

The TOML and JSON schemas are structurally identical and their direct floating- point execution matches the hand model within 1e-13. One input edge carries one external-drive sample and advances all six states simultaneously.

Generated Q32.32 Verilog compiles with Icarus and preserves the varied 64-step trajectory within:

• maximum potential-state error below 0.02 mV;
• maximum derivative-state error below 3.2 mV/s;
• maximum EEG-proxy error below 0.02 mV.

These envelopes include fixed-point quantisation and the generic exponential lookup. They are an explicit bounded co-simulation result, not exact transcendental equivalence.

Controlled benchmark

benchmarks/bench_jansen_rit.py measures the same complete deterministic-drive batch through all five public dispatchers. It rejects an unpinned run, any unavailable runtime, a standalone Rust-safety failure, trace/final-state drift, or an invalid run size. The committed JSON binds raw timing samples, source hashes, exact loaded Rust/Go/Mojo binaries, tool versions, affinity, load, and every parity delta.

The record is one-logical-CPU, non-exclusive local regression evidence. It is not a production, cross-host, cross-framework, or hardware performance claim. The timing table is reproduced in Benchmark Comparison.

Boundaries

• The EEG proxy is a continuous mean-field potential, not a binary spike.
• The default Euler step follows the pinned executable transcription; it is not imposed by the continuous publication equations.
• The deterministic input protocol does not reproduce a published figure.
• Q32.32 co-simulation reaches H1 only.
• No synthesis, timing closure, formal equivalence, device resource, or PPA result is claimed.

Reproduction

The focused test invocation preloads the installed Rust extension before pytest adds repository support paths to the import search order.

Bash

rustc --edition 2021 --test \
  src/sc_neurocore/accel/rust/safety/jansen_rit.rs \
  -o /tmp/jansen_rit_tests
/tmp/jansen_rit_tests

PYTHONPATH=bridge:src:. .venv/bin/python - <<'PY'
import sc_neurocore_engine
import pytest

raise SystemExit(
    pytest.main(
        [
            "-q",
            "tests/test_model_jansen_rit.py",
            "tests/test_jansen_rit_dynamics.py",
            "tests/test_reference_jansen_rit.py",
            "tests/test_cosim_jansen_rit.py",
            "tests/test_jansen_rit_backends.py",
            "tests/test_bench_jansen_rit.py",
        ]
    )
)
PY

taskset -c 11 env PYTHONPATH=bridge:src:. .venv/bin/python \
  benchmarks/bench_jansen_rit.py \
  --json benchmarks/results/bench_jansen_rit.json