specsr.linefit

Gaussian emission-line fitting, and the S/N derived from it.

Fits a Gaussian on a linear continuum in a window around each expected line and reports amplitude over continuum scatter. This is the measurement behind the paper’s S/N figure.

The continuum scatter is estimated from sidebands — an annulus around the line, excluding its core — rather than from the whole window. Using the window would fold the line itself into the noise estimate and depress the S/N of exactly the strong lines the figure is about.

Note what this quantity is and is not: it references only the spectrum being measured, never the HR truth, so a high S/N means “a confident detection of something”, not “the right line flux”. Establishing that a line is correct needs the reference, which is why specsr.plotting.plot_line_flux_comparison() exists alongside it.

Functions

fit_line_sideband_weighted(x, y, mu0, *[, ...])	Fit one line.
gauss_lin(x, amp, mu, sigma, c0, c1)	Gaussian on a linear continuum.
line_snr_from_fit(fit)	(amp/sigma_cont, amp/amp_err) from a fit, or (nan, nan).
mad_sigma(y)	Robust scatter via MAD, NaN-aware.
measure_line_snr(wavelength, z, spectra, ...)	Per-line S/N for several spectrum sets over the same objects.

specsr.linefit.fit_line_sideband_weighted(x, y, mu0, *, fit_halfwin=0.25, core_halfwin=0.05, sb_gap=0.03, sb_width=0.12, sigma_bounds=(0.001, 0.12), mu_bounds_half=0.01, allow_negative_amp=True, maxfev=40000)

Fit one line. Returns a dict of parameters, or None if unfittable.

Negative amplitudes are allowed by default: forcing positivity would turn a non-detection into a small positive bump and manufacture signal where there is none.

Parameters:

• fit_halfwin (float)

• core_halfwin (float)

• sb_gap (float)

• sb_width (float)

• sigma_bounds (tuple[float, float])

• mu_bounds_half (float)

• allow_negative_amp (bool)

• maxfev (int)

specsr.linefit.gauss_lin(x, amp, mu, sigma, c0, c1)

Gaussian on a linear continuum.

specsr.linefit.line_snr_from_fit(fit)

(amp/sigma_cont, amp/amp_err) from a fit, or (nan, nan).

specsr.linefit.mad_sigma(y)

Robust scatter via MAD, NaN-aware. NaN when there is too little data.

specsr.linefit.measure_line_snr(wavelength, z, spectra, lines_rest_um, line_names=None, **fit_kw)

Per-line S/N for several spectrum sets over the same objects.

Parameters:

• spectra – {"LR": array, "SR": array, "HR": array, ...}, each (n, n_lambda).

• lines_rest_um – Rest wavelengths, microns. Redshifted per object with z.

Returns:

• {f"{line}_sn_{kind}": array} with NaN where a line falls off the grid or

• the fit fails — left as NaN rather than zero, since “not measurable” and

• ”measured as zero” are different statements.