kinextract.build_initial_guess_nonparam

kinextract.build_initial_guess_nonparam(st: FitState, coff_init: float, coff2_init: float, b_bounds: tuple = (1e-06, 1.0), w_bounds: tuple = (1e-05, 1.0), coff_bounds: tuple | None = None, coff2_bounds: tuple = (-0.7, 0.7), amp_bounds: tuple = (1e-08, 1000000000000.0)) → tuple[ndarray, ndarray, ndarray][source]

Build the initial non-parametric-LOSVD parameter vector and its bounds.

Assembles the flat parameter vector used by the optimizer, concatenating (in order): nl non-parametric LOSVD histogram bin weights (each initialized to a flat 1/nl, i.e. a uniform velocity distribution), nt stellar template weights (each initialized to a flat 1/nt), then optionally a continuum/wavelength-offset block (depending on st.icoff), a global amplitude parameter (if st.fit_global_amp), and a continuum-polynomial coefficient (if st.continuum_poly_mode is not "none"). The corresponding lower- and upper-bound vectors are built in the same order.

Parameters:

  • st (FitState) – Fit state; supplies nl (number of LOSVD velocity bins), nt (number of stellar templates), icoff (which continuum-offset parameterization is active), fit_global_amp, continuum_poly_mode, continuum_poly_bound, t (template matrix, for the global-amplitude initial guess), and gerr/g.

  • coff_init (float) – Initial value for the primary continuum/wavelength offset coefficient (used when st.icoff == 1).

  • coff2_init (float) – Initial value for the secondary continuum/wavelength offset coefficient (used when st.icoff is 1 or 2).

  • b_bounds (tuple of (float, float), optional) – (lower, upper) bounds applied to every LOSVD bin weight.

  • w_bounds (tuple of (float, float), optional) – (lower, upper) bounds applied to every template weight; the effective upper bound is widened to max(w_bounds[1], 1 + 1e-6) (see Notes).

  • coff_bounds (tuple of (float, float), optional) – (lower, upper) bounds for the primary offset coefficient (only used when st.icoff == 1). Defaults to (coff_init - 0.8, coff_init + 4.0) if not given. The lower bound must be greater than -1 (see Raises).

  • coff2_bounds (tuple of (float, float), optional) – (lower, upper) bounds for the secondary offset coefficient (used when st.icoff is 1 or 2).

  • amp_bounds (tuple of (float, float), optional) – (lower, upper) bounds for the global amplitude parameter (only used when st.fit_global_amp is True).

Returns:

  • x0 (ndarray) – Initial flat parameter vector for the optimizer.

  • lb (ndarray) – Lower bounds, same length and ordering as x0.

  • ub (ndarray) – Upper bounds, same length and ordering as x0.

Raises:

ValueError – If the resolved lower bound on the primary continuum offset is <= -1, since the model evaluation divides by (coff + 1) and such a bound would allow a zero or negative denominator.

Notes

The template-weight upper bound is raised to at least 1.0 + 1e-6 so that the initial value (1/nt) coincides with the lower bound only in the degenerate nt == 1 case, never with the upper bound; starting exactly at an upper bound can cause some optimizers to treat the parameter as already converged.