Probe Operators in jaxfne v0.2.1+¶

Overview¶

The probe layer is a first-class component of the TFNE pipeline:

Emitter → Source → Field → Probe → Objective → Optimizer

Eight probe operators expose different aspects of neural/field state as named readouts. Each operator returns data plus a JSON-safe report declaring operator status, units, calibration metadata, scope declarations, and assumptions.

Operator Kinds¶

SPK¶

Purpose: Expose spike events or spike matrix.

Minimum output: - kind: spk - method: threshold_or_emitter_spike_array - units_or_status: binary_spike_indicator - operator_status: simulated_proxy

Shape examples: - Spike matrix: [T, N] (time × neurons) - Event table: rows with timestamp, neuron_id, area, layer, cell_type

v0.2.1 status: Simulated spike readout from emitter.

Vm¶

Purpose: Expose membrane voltage or emitter state variable.

Minimum output: - kind: vm - method: emitter_state_voltage_trace - units_or_status: mV_or_model_state_variable - operator_status: simulated_proxy

Scope note: The emitter voltage represents model state; the report declares its specific interpretation and scope.

v0.2.1 status: State voltage from emitter; proxy readout.

Source¶

Purpose: Expose current/source proxy or calibrated source trace used by the field layer.

Minimum output: - kind: source - method: declared_source_projection_or_proxy - source_decomposition: proxy_reduced_emitter (or other declared mode) - source_calibration_status: proxy_izhikevich_current (or calibrated status)

v0.2.1 status: Proxy source from emitter state; computational readout.

LFP-proxy¶

Purpose: Sample or average extracellular potential state at contacts.

Acceptable v0.2.1 implementation: - Point sample from phi_e proxy/solution - Finite-contact average - Laminar contact average over declared depths/layers

Minimum report: - kind: lfp_proxy - method: point_or_finite_contact_phi_proxy - units_or_status: proxy_voltage_units_or_V_if_calibrated - field_solver_status: linear_solver - amplitude_status: false

Important: v0.2.1 uses -proxy terminology to declare operator status explicitly. Proxy-scale operators are computational readouts suitable for tutorial and validation workflows.

v0.2.1 status: Laminar proxy-scale readout; calibration metadata available for calibrated workflows.

CSD-proxy¶

Purpose: Estimate current-source density profile or second spatial derivative/divergence proxy.

Acceptable v0.2.1 implementation: - div(J_e) when J_e exists - Second spatial derivative of laminar potential proxy - Source-profile proxy with explicit sign convention

Minimum report: - kind: csd_proxy - method: divergence_proxy_or_second_derivative_laminar - CSD_sign_convention: positive_equals_extracellular_source (or other declared convention) - units_or_status: proxy_A_m^-3_or_proxy_units

Required tests: - Constant potential gives near-zero CSD-proxy output - Linear potential gives near-zero second derivative - Finite output - Sign convention exported

v0.2.1 status: Laminar proxy-scale readout; calibration metadata available for calibrated workflows.

EEG-proxy¶

Purpose: Provide a simulated scalp-channel proxy readout using a declared toy or proxy lead field.

Acceptable v0.2.1 implementation:

y_eeg(t, c) = sum_k L_eeg[c, k] * s_k(t)

where s_k(t) is a declared source/current/potential feature and L_eeg is a toy or user-declared projection matrix.

Minimum report: - kind: eeg_proxy - method: linear_leadfield_proxy - leadfield_status: toy_or_declared_proxy - sensor_geometry_status: simulated_minimal - units_or_status: arbitrary_proxy_units - operator_status: simulated_proxy - amplitude_status: false

Scope: EEG readouts are computational proxies with declared toy leadfields. Calibration against empirical reference data is covered in Limitations and future plans.

v0.2.1 status: Simulated EEG-proxy readout; proxy-scale calibration for tutorial workflows.

MEG-proxy¶

Purpose: Provide a simulated magnetometer proxy readout using a declared current-orientation or lead-field proxy.

Acceptable v0.2.1 implementation:

y_meg(t, c) = sum_k L_meg[c, k] * j_oriented_k(t)

Minimum report: - kind: meg_proxy - method: linear_current_orientation_proxy - leadfield_status: toy_or_declared_proxy - sensor_geometry_status: simulated_minimal - orientation_convention: declared - units_or_status: arbitrary_proxy_units - operator_status: simulated_proxy - amplitude_status: false

Scope: MEG readouts are computational proxies with declared toy leadfields. Calibration against empirical reference data is covered in Limitations and future plans.

v0.2.1 status: Simulated MEG-proxy readout; proxy-scale calibration for tutorial workflows.

EMM-proxy¶

Name:

EMM = electromagnetic metabolism estimate proxy

Interpretation in v0.2.1:

normalized electrophysiological activity / electromagnetic energy cost proxy

Possible first operator:

EMM(t) = w_spk * normalized_spike_rate(t)
       + w_src * normalized(||source(t)||_1 or ||source(t)||_2^2)
       + w_field * normalized(||grad(phi_e)(t)||_2^2 or ||J_e(t)||_2^2)
       + w_syn * normalized_synaptic_activity_proxy(t)

Minimum report: - kind: emm_proxy - method: normalized_activity_field_source_cost_proxy - units_or_status: normalized_proxy_units - biophysical_calibration_status: uncalibrated_proxy - operator_status: simulated_proxy - amplitude_status: false

Important: EMM-proxy is valid for relative within-run comparisons. It represents a signaling-energy proxy suitable for optimization workflows in v0.2.x.

v0.2.1 status: Normalized activity cost proxy; exploratory metric for optimization.

Mathematical Forms¶

This section formalizes the eight operators in operator-form notation. These are computational forms declared as proxy-scale operators suitable for tutorial and model-development workflows.

SPK (Spike Detection)¶

\[\mathrm{SPK}_n(t)=\mathbb{1}[V_n(t)\geq \theta_n]\]

Spike indicator for neuron \(n\) at time \(t\): 1 if membrane voltage exceeds threshold \(\theta_n\), 0 otherwise.

Vm (Membrane Voltage)¶

\[\mathrm{Vm}_n(t)=V_n(t)\]

Direct readout of state voltage from emitter \(n\). Proxy readout.

Source¶

\[S_n(t)=f_{\mathrm{source}}(x_n(t),\theta_n)\]

Source/current proxy derived from emitter state \(x_n(t)\) and parameters \(\theta_n\). Status: proxy-scale units in v0.2.x; calibration path documented in the calibration guide.

LFP-proxy¶

\[\phi_{\mathrm{proxy}}(t,c)=\sum_{n=1}^{N}W_{cn}S_n(t), \quad \sum_{n=1}^{N}W_{cn}=1\]

Row-normalized kernel projection: potential at contact \(c\) is a weighted sum of sources, with row sums equal to one. Status: laminar proxy via solver-free kernel projection.

CSD-proxy¶

\[\mathrm{CSD}_{\mathrm{proxy}}(t,c) \approx \frac{\phi_{\mathrm{proxy}}(t,c+1)-2\phi_{\mathrm{proxy}}(t,c)+\phi_{\mathrm{proxy}}(t,c-1)}{(\Delta z)^2}\]

Second spatial derivative (Laplacian) of laminar potential proxy. Sign convention: positive_equals_extracellular_source.

EEG-proxy¶

\[Y_{\mathrm{EEG\text{-}proxy}}(t,s) = \sum_{c=1}^{C} L^{\mathrm{EEG}}_{sc} \phi_{\mathrm{proxy}}(t,c)\]

Linear leadfield projection: each scalp electrode \(s\) is a weighted combination of laminar potentials. \(L^{\mathrm{EEG}}\) is a declared toy/proxy leadfield suitable for tutorial workflows.

MEG-proxy¶

\[Y_{\mathrm{MEG\text{-}proxy}}(t,s) = \sum_{n=1}^{N} L^{\mathrm{MEG}}_{sn} o_n S_n(t)\]

Magnetometer readout: each sensor \(s\) sums orientation-weighted source projections. \(o_n\) is source orientation; \(L^{\mathrm{MEG}}\) is a proxy leadfield.

EMM-proxy (Activity/Cost Proxy)¶

\[\mathrm{EMM}_{\mathrm{proxy}}(t) = \alpha \|S(t,\cdot)\|_1 + \beta \|\phi_{\mathrm{proxy}}(t,\cdot)\|_1\]

Normalized electrophysiological activity cost proxy combining source and field energy norms. Represents signaling-energy proxy; valid for relative within-run comparisons and optimization workflows.

Probe Report Sidecar¶

\[R_k = \{\mathrm{kind}, \mathrm{method}, \mathrm{units\_or\_status}, \mathrm{operator\_status}, \mathrm{physical\_amplitude\_statement\_allowed}, \ldots\}\]

Each operator returns a JSON-safe report \(R_k\) declaring operator type, computation method, status (proxy/simulated), and statement constraints.

Current Status: v0.2.1 Simulated / Proxy¶

All eight operators in v0.2.1 are simulated or proxy readouts:

Operator	Status	Notes
SPK	Simulated spike readout	From emitter threshold or declared spike array
Vm	Simulated voltage trace	From emitter state variable
source	Simulated source proxy	From emitter state or declared mode
LFP-proxy	Laminar proxy readout	Point sample or contact average; solver-free proxy mode
CSD-proxy	Laminar proxy readout	Second derivative or divergence proxy; solver-free proxy mode
EEG-proxy	Simulated linear projection	Proxy implementation with toy leadfield; calibration metadata exposed
MEG-proxy	Simulated linear projection	Proxy implementation with toy leadfield; calibration metadata exposed
EMM-proxy	Normalized cost proxy	Relative metric for optimization; signaling-energy proxy

Future Path¶

The v0.2.x line preserves proxy operators as stable public readouts while adding clearer validation metadata, calibration specifications, and tutorial coverage.

Planned areas include: - v0.2.4–v0.2.6: field/proxy mathematics and admissibility diagnostics - v0.2.5 and v0.2.17: calibration specification and reporting workflows - v0.2.7–v0.2.15: Colab-ready tutorial stack and tutorial smoke tests - v0.2.13–v0.2.14: laminar profile templates using literature-derived technical references, including Lichtenfeld et al. (2024) and Mendoza-Halliday et al. (2024). These templates support declared profile construction and tutorial design; they reference literature without stating reproduction of the reference datasets. - v0.2.18–v0.2.21: operator status export, package audit, release candidate, and consolidated practical scaffold release

Calibration workflows and advanced tutorials are developed in the docs and examples.

Beyond v0.2.x (v0.3.x and later): - Receptor-level synaptic dynamics (synaptic current modeling) - Empirically calibrated source projection - Hodgkin-Huxley ion channels with biophysical parameters - Full-PDE field solver — see Limitations and future plans - Empirically calibrated LFP/CSD readouts with validated sign conventions - Physical leadfields from anatomically-validated head models - Whole-brain connectivity integrated with laminar columns - Plasticity and learning rules with evidence-backed parameters

API Example¶

import jaxfne
from jaxfne.fields import spk_probe, lfp_proxy_probe, emm_proxy_probe

# Simulate a small network
cfg = jaxfne.configuration().network(n=32).emitter(...).field(...).probe(...)
model = jaxfne.construct(cfg)
signals = model.simulate(jaxfne.simulation(duration_ms=1000.0))

# Apply probe operators
spk_readout = spk_probe(signals.spikes)
lfp_readout = lfp_proxy_probe(signals.field.lfp)
emm_readout = emm_proxy_probe(signals.field.lfp + signals.spikes.mean(axis=1))

# Each readout has JSON-safe report
print(spk_readout.report)
print(lfp_readout.report)
print(emm_readout.report)

Terminology Notes¶

Use *-proxy to denote declared computational operators: lfp_proxy, csd_proxy, eeg_proxy, meg_proxy are the canonical public labels in v0.2.1+.
This terminology explicitly declares computational intent and prevents informal analogy with empirically validated readouts.
All operators are "simulated" or "proxy" in v0.2.x. Statements of physical equivalence require separate calibration and validation evidence.
EMM-proxy is valid for relative within-run comparisons; it represents a signaling-energy proxy in v0.2.x.

Document version: v0.2.1
Last updated: 2026-05-20