Function assert_matrix_derivativefd_rel 

pub fn assert_matrix_derivativefd_rel(
    fd: &Array2<f64>,
    analytic: &Array2<f64>,
    rel_tol: f64,
    label: &str,
)

Asserts that a finite difference dense matrix matches an analytically computed directional derivative matrix to a relative tolerance rel_tol·(1 + |analytic|), plus component-wise sign agreement.

Use this (rather than the absolute-tolerance assert_matrix_derivativefd) when the comparison’s dominant components are O(0.1–1) and the finite difference is contaminated by a small, non-smooth solver channel — e.g. an adaptive PIRLS stabilization ridge whose magnitude shifts discontinuously across the ± FD re-solves. There the exact analytic IFT derivative (which correctly excludes that solver-only ridge) and the FD disagree by a fixed fraction of the component magnitude, not a fixed absolute amount, so an absolute bound tuned for the small components is spuriously tight on the large ones. The two underlying derivative channels are validated separately against their own FDs, so this asserts the composite to the achievable relative precision rather than weakening the per-channel checks (gam#855).