$\texttt{HostSub_GP}$: Precise Galaxy Background Subtraction in Transient Long-slit Spectroscopy with Gaussian Processes

TL;DR

The paper tackles the challenge of removing host galaxy contamination from long-slit spectra of extragalactic transients. It introduces HostSub_GP, a GP-based framework that leverages archival multi-band galaxy images to construct a prior on the 2D galaxy background and couples it with a second GP to model spectrum deviations, enabling precise background subtraction. The method is validated on synthetic MUSE data and applied to two Keck LRIS transients (SN 2019eix and AT 2019qiz), demonstrating improved recovery of faint spectral features and yielding refined insights into progenitor properties and nuclear activity. The work provides a scalable, open-source toolkit built on JAX/tinygp that can enhance the quality of time-domain spectroscopy and support more reliable transient classifications and physical inferences.

Abstract

We present a novel host galaxy subtraction technique in long-slit spectroscopy for extragalactic transients. Unlike classic methods which generally estimate the background using a simple linear interpolation of local galaxy flux in the 2D spectrum, our approach leverages multi-band archival images of the host galaxies to model the background emission from the galaxy in the 2D spectrum. Such imaging encodes the wavelength-dependent galaxy profile along the slit, and is readily accessible through wide-field imaging surveys. We construct a smooth prior for the 2D galaxy profile with a Gaussian process (GP) based on these reference images, and use another GP to model the correlated deviations from the prior in the observed spectrum. This enables accurate inference of the galaxy flux blended with the transient. On synthetic long-slit data of a spiral galaxy extracted from a Multi Unit Spectroscopic Explorer hyper-spectral cube, the GP method remains robust as long as the host galaxy is spatially resolved and consistently outperforms classic methods. We apply the method to archival Keck spectra of two real transients, SN 2019eix and AT 2019qiz, to further demonstrate how the method uniquely recovers weak spectral features amid strong galaxy contamination, enabling refined constraints on the properties of both transients. We have released the software implementation, $\texttt{HostSub_GP}$, a scalable toolkit that leverages $\texttt{JAX}$, with an MIT license.

Figures (9)

• Figure 1: The data reduction framework in HostSub_GP. The 2D co-addition will only be performed if multiple exposures exist.
• Figure 2: An example of the derived $F_\mathrm{sub}$ and $\xi_\mathrm{sub}$ from the preprocessed data with adaptive binning around galaxy emission lines. Upper panel: the example preprocessed 2D spectrum after co-addition, rectification, and global background subtraction. The edges of the sky regions (green), host regions (salmon), and transient mask (red) are displayed. The bad pixels are indicated with red colors. Middle panel: the batched 2D galaxy profile $\xi_\mathrm{sub}$. Dashed gray lines indicate batch edges. Bottom panel: the 1D galaxy spectrum $F_\mathrm{sub}$ used to normalize the galaxy profile.
• Figure 3: An example of constructing $\xi_\mathrm{sub,img}$, which we use as the mean function in modeling $\xi_\mathrm{sub}$ with a GP (Eq. \ref{['eq:2D_GP']}). Top panels: example archival PS images in $grizy$ filters. The position of a 1-long slit is displayed as the blue patch. A 24-wide host region and a transient mask with a width of twice the seeing are indicated as the salmon and red patches, respectively. Middle panel: the throughout function of each filter and the corresponding effective wavelength. Bottom panel: the smooth $\xi_\mathrm{sub,img}$ built with the galaxy light profiles (within the host region) at five discrete wavelengths with a 2D GP.
• Figure 4: The GP method systematically improves the galaxy background estimates on the MUSE synthetic dataset of the spiral galaxy LCRS B110709.2-121854. The first two columns illustrate the distribution of fractional flux residuals (residuals normalized by the original galaxy flux in the PS $i$ filters at randomly drawn slit locations) as a function of wavelengths in each of the 25 Å bins, resulting from the classic linear and B-spline interpolation methods, and the GP method. The first column covers the full wavelength range, whereas the second column zooms in on the region containing the strong H$\alpha$ and [N2] emission lines. Solid lines denote the median residuals across 100 random draws; the darker and lighter shaded regions represent the 68% and 95% percentiles, respectively. As a comparison, the black dashed line denotes the zero residual level. The $\sigma$ values in the legends quantify the standard deviation of the 100 median fractional residuals, each computed across the full wavelength range for a single random draw. The rightmost panel shows the synthetic image in the $i$ band convolved with a Gaussian kernel to downgrade the spatial resolution from 0.68 to 1.2. Overlaid are the locations of the randomly drawn slits, color coded by the corresponding absolute values of fractional residuals (median across all wavelengths) from the GP method. The typical aperture mask width (twice the seeing, 2.4) is indicated, though the mask width is determined case by case in the seeing matching.
• Figure 5: Same as Figure \ref{['fig:muse_spiral']}, but showing that the GP method still outperforms the classic methods in both the precision and accuracy on the synthetic dataset where the MUSE hyper-spectral data cube has been spatially binned by a factor of 4, simulating observations of a less resolved galaxy scaled to 25% of the angular size under the same seeing conditions. Again, seeings in the synthetic photometry on the rightmost panels have been downgraded to $\sim$1.2.