kinextract.compute_losvd_roughness

kinextract.compute_losvd_roughness(b: ndarray) → float[source]

Peak-normalised maximum absolute second difference of a LOSVD array.

Measures local curvature: smooth LOSVDs (e.g. a Gaussian with sigma ~ 3 LOSVD bins) have roughness ~ 0.10; jagged noise spikes produce values > 0.20. Used by _auto_select_xlam() (and logged as a diagnostic at every point of the xlam grid search) to judge whether the current regularisation strength is sufficient to suppress noise-driven wiggles in the recovered LOSVD.

Parameters:: b (ndarray) – LOSVD histogram to assess (negative values are clipped to zero before computing curvature).
Returns:: Roughness statistic: the maximum absolute second difference of the (clipped) LOSVD, divided by its peak value. Returns 0.0 if b has fewer than 3 elements, and inf if the peak is essentially zero (a degenerate LOSVD is treated as maximally rough so that xlam selection pushes toward stronger regularisation).
Return type:: float