Struct MultiplicativeBaselineModel 

pub struct MultiplicativeBaselineModel<M: FitModel> { /* private fields */ }

Bounded multiplicative polynomial baseline (issue #635):

y(E) = B(E) · T_inner(E),   B(E) = b0 + b1·z + b2·z²,   z = ln(E / E_ref)

where E_ref = √(E_min·E_max) (see baseline_reference_energy) and T_inner is any inner FitModel — typically the bare transmission model, or NormalizedTransmissionModel when the SAMMY additive background is also configured (the baseline is the OUTERMOST factor).

INTENTIONAL DEPARTURE from SAMMY

SAMMY’s modern data-reduction path applies a SCALAR normalization plus additive backgrounds only: T_obs = Anorm·T + BackA + BackB/√E + BackC·√E + BackD·exp(−BackF/√E) (cro/mnrm1.f90, subroutine Norm, applied to every data type via the/ZeroKCrossCorrections_M.f90). SAMMY’s nearest analogue to an energy-dependent multiplicative normalization is the DORMANT legacy power-law Anorm = Anrm(1) + Anrm(2)·E^Anrm(3) (acs/macs4.f90:440–450, Find_Www_Yyy), which is not reachable from the modern reconstruction path. This low-order ln-E polynomial baseline is a NEREIDS extension motivated by the IPTS-37432 campaign (findings A3/A5): real VENUS sample/open-beam ratios sit a few % from unity with smooth energy dependence, and freeing the SAMMY Anorm together with temperature and density is degenerate on such data (observed: T → 4471 K, n +76 %, χ²/ν 932, with no warning). The bounded multiplicative form fitted jointly with temperature at fixed density produced χ²/ν ≈ 2–8 across the 20-run campaign.

Because b0 is exactly degenerate with Anorm, the pipeline rejects a free Anorm alongside ANY configured baseline — including a fully frozen one (see nereids-pipeline::validate_multiplicative_baseline). A frozen-b0 + free-Anorm combination would be well-posed, but supporting it buys nothing (Anorm would just play b0’s role at a rescaled value) and splits the normalization story across two knobs; the sanctioned combination is Anorm held fixed.

Index invariant

The baseline indices (b0_index, b1_index, b2_index) must NOT designate a parameter the inner model reads: the analytic Jacobian filters the baseline indices out of the inner free set, so such a collision cannot be detected and the column would silently omit B(E) × ∂T_inner/∂p. Aliasing AMONG the baseline indices themselves IS supported — the Jacobian columns accumulate.

Implementations

impl<M: FitModel> MultiplicativeBaselineModel<M>

pub fn new( inner: M, energies: &[f64], e_ref: f64, b0_index: usize, b1_index: usize, b2_index: usize, ) -> Self

Create the wrapper. e_ref is normally baseline_reference_energy(energies); it is passed explicitly so result consumers can reconstruct B(E) with the exact same reference.

pub fn with_active_mask(self, mask: Option<&[bool]>) -> Self

Scope the runtime positivity guard to the given active mask (None = all bins active, the default). See the active_mask field doc for why masked bins must be exempt.

Trait Implementations

impl<M: FitModel> FitModel for MultiplicativeBaselineModel<M>

fn evaluate(&self, params: &[f64]) -> Result<Vec<f64>, FittingError>

Evaluate the model for the given parameters.

fn analytical_jacobian( &self, params: &[f64], free_param_indices: &[usize], y_current: &[f64], ) -> Option<FlatMatrix>

Optionally provide an analytical Jacobian.

impl<M: FitModel> ForwardModel for MultiplicativeBaselineModel<M>

fn predict(&self, params: &[f64]) -> Result<Vec<f64>, FittingError>

Predict model output for the given parameter vector.

fn jacobian( &self, params: &[f64], free_param_indices: &[usize], y_current: &[f64], ) -> Option<Vec<Vec<f64>>>

Analytical Jacobian (column-major layout).

fn n_data(&self) -> usize

Number of data points in the model output.

fn n_params(&self) -> usize

Number of parameters (total, including fixed).