Connectivity Metric Ranges#
Direction convention#
For xarray results returned by multitaper_connectivity,
result.sel(source="a", target="b") is the influence from a to b. The
lower-level Connectivity methods retain their historical array convention:
result[..., i, j] represents j -> i.
The following summarizes the theoretical value ranges for common connectivity metrics.
Metric |
Range |
Notes |
|---|---|---|
Power |
[0, ∞) |
Power spectral density; always non-negative. |
Coherency (complex) |
|C_{xy}(f)| ∈ [0, 1]; phase ∈ [−π, π] |
Values lie in the unit disk of the complex plane. |
Coherence phase |
[−π, π] |
Phase angles of complex coherency. |
Coherence magnitude |
[0, 1] |
0: none; 1: perfect linear dependence at frequency f. |
Imaginary coherence |
[0, 1] (magnitude) |
Raw imaginary part ranges [−1, 1]; magnitude version. |
Canonical coherence |
[0, 1] |
Maximal coherence between groups. |
Global coherence |
[0, 1] |
Fraction of total coherent power per component (eigenvalue / sum of eigenvalues, Cimenser et al. 2011); scale-invariant. |
PLV (Phase‑Locking Value) |
[0, 1] |
0: random phases; 1: constant phase diff. |
PLI (Phase‑Lag Index) |
[−1, 1] |
Signed version; unsigned in [0, 1]. |
wPLI (Weighted PLI) |
[−1, 1] |
Signed version (like PLI); weights by |Im(Sxy)|. Unsigned in [0, 1] via absolute value. |
dwPLI (Debiased PLI²) |
[−1/(N−1), 1] |
Unbiased squared PLI estimate; finite-sample values can be negative. |
dwPLI² (Debiased squared wPLI) |
[−1, 1] |
Unbiased weighted PLI² estimate; finite-sample values can be negative. |
PPC (Pairwise Phase Consistency) |
[−1/(N−1), 1] |
Unbiased phase consistency; finite-sample values can be negative. |
Spectral Granger causality |
[0, ∞) |
Non‑negative; no finite upper bound. |
DTF (Directed Transfer Function) |
[0, 1] (normalized) |
Proportion of inflow via transfer function. |
Directed coherence |
[0, 1] (normalized) |
DTF scaled by noise variance. |
PDC (Partial Directed Coherence) |
[0, 1] (normalized) |
Directional influence via AR coefficients. |
gPDC (Generalized PDC) |
[0, 1] (normalized) |
PDC scaled by noise variance. |
dDTF (Direct DTF) |
[0, 1] (normalized) |
DTF with partial coherence normalization. |
Group delay |
(−∞, ∞) |
Time delays; can be positive or negative. |
PSI (Phase Slope Index) |
(−∞, ∞) |
Signed; magnitude depends on phase slope. |
Notes#
For metrics theoretically bounded in [0, 1], numerical implementations should clamp to the interval after computation to avoid tiny overflows (e.g., 1 ± 1e−12). Unbiased estimators such as PPC and the debiased PLI variants must retain legitimate negative finite-sample values.
For phase‑based metrics, choose an expectation (sample) axis (trials/tapers/segments) and aggregate only across that axis.
Document the shapes/dtypes of inputs and outputs in each docstring per NumPy style.
Also document that for granger causality output [i,j] (row i, column j) corresponds to j –> i.
Implementation Details#
Bounded Metrics [0, 1]#
coherence_magnitude: Magnitude-squared coherence
imaginary_coherence: Magnitude of imaginary coherency
canonical_coherence: Maximal coherence between groups
phase_locking_value: Phase consistency measure
directed_transfer_function: Normalized directional influence
directed_coherence: DTF scaled by noise variance
partial_directed_coherence: Normalized causal influence
generalized_partial_directed_coherence: PDC scaled by noise variance
direct_directed_transfer_function: DTF with partial coherence normalization
global_coherence: Fraction of total coherent power per component (eigenvalue of the cross-spectral matrix over the sum of all eigenvalues, Cimenser et al. 2011); scale-invariant
Bounded Metrics [−1, 1]#
phase_lag_index: Signed phase consistency (unsigned version in [0, 1])
weighted_phase_lag_index: Signed weighted phase consistency
debiased_squared_weighted_phase_lag_index: Unbiased weighted PLI² estimate
Finite-sample unbiased metrics [−1/(N−1), 1]#
debiased_squared_phase_lag_index: Unbiased squared PLI estimate
pairwise_phase_consistency: Unbiased phase consistency
Phase Metrics [−π, π]#
coherence_phase: Phase angles of complex coherency
Unbounded Metrics [0, ∞)#
power: Power spectral density
pairwise_spectral_granger_prediction: Spectral Granger causality
Unbounded Metrics (−∞, ∞)#
group_delay: Time delays between signals
phase_slope_index: Directional phase coupling measure
Complex-valued Metrics#
coherency: Complex coherence with magnitude ∈ [0, 1] and phase ∈ [−π, π]