X-shooter Spectral Library (XSL)

• XSL is an empirical stellar spectral library offering simultaneous near-UV, optical, and near-IR data using ESO’s VLT/X-shooter.
• It provides moderate-to-high resolution spectra (R ~ 10,000) crucial for accurate stellar population synthesis and validating synthetic models.
• Releases deliver robust flux calibration, extensive parameter coverage, and targeted emphasis on cool evolved and variable stars.

The X-shooter Spectral Library (XSL) is a homogeneous empirical stellar spectral library built with ESO’s VLT/X-shooter spectrograph to provide simultaneous near-ultraviolet, optical, and near-infrared stellar spectra for stellar astrophysics and stellar population synthesis. Across its releases, XSL was designed to supply flux-consistent coverage over a wavelength range extending from the near-UV to the near-IR at a resolving power close to $R \sim 10{,}000$, with deliberate emphasis on cool evolved stars, long-period variables, carbon stars, and other phases that dominate the red and near-infrared light of galaxies (Chen et al., 2011, Verro et al., 2021).

1. Scientific motivation and conceptual scope

XSL was conceived to address a specific limitation of earlier empirical libraries: the absence of a single, simultaneous data set combining extended wavelength coverage, moderate-to-high spectral resolution, and a sample broad enough in effective temperature, surface gravity, and metallicity for population-synthesis work. Its scientific rationale is twofold. First, stellar population synthesis requires empirical spectra that sample the Hertzsprung–Russell diagram broadly enough to reproduce integrated-light spectra and infer ages, metallicities, and abundance patterns. Second, broad wavelength coverage is essential because different evolutionary phases contribute differently across bands; in particular, asymptotic giant branch stars dominate intermediate-age and old populations in the near-IR while contributing much less in the optical (Chen et al., 2011).

This motivation became more urgent as precision extragalactic spectroscopy increasingly depended on flux-consistent optical–near-IR spectra, and as the near-IR gained importance for facilities such as JWST and ELT. XSL was therefore designed to minimize the inter-library stitching problems that had historically complicated population synthesis, while also providing a stringent empirical benchmark for validating synthetic stellar spectra across the HR diagram (Lançon et al., 2018).

A central feature of the project is simultaneity. X-shooter records the UVB, VIS, and NIR arms in a single exposure, so the near-UV, optical, and near-IR segments correspond to the same instantaneous stellar state. For non-variable stars, this improves cross-arm spectrophotometric consistency. For variables such as Mira stars and other long-period variables, it is more consequential: it preserves phase coherence across the full spectral energy distribution rather than combining spectra obtained at different pulsation phases (Lançon et al., 2018).

2. Instrumentation, observing strategy, and spectral format

XSL is based on the VLT/X-shooter spectrograph, a three-arm cross-dispersed echelle instrument that operates simultaneously in the UVB, VIS, and NIR. In DR2 and DR3, the arm coverages are UVB 300–556 nm, VIS 533–1020 nm, and NIR 994–2480 nm, yielding a typical combined range from approximately 300 nm to approximately $2.45\,\mu$m. The resolving power is defined in the standard way as $R=\lambda/\Delta\lambda$ (Gonneau et al., 2020).

The narrow-slit settings used for XSL correspond to nominal resolving powers of approximately $R \approx 9{,}200$ in the UVB arm, $R \approx 11{,}000$ in the VIS arm, and $R \approx 7{,}770$ in the NIR arm. DR2 measured arm-dependent resolving powers of $R = 9{,}793$ for UVB, $R = 11{,}573$ for VIS, and $R = 7{,}956$ for NIR, with line-spread functions close to Gaussian in all three arms. Within a given arm, the resolving power is treated as constant with wavelength (Gonneau et al., 2020, Verro et al., 2021).

The observing strategy combines narrow-slit science exposures with wide-slit exposures used for flux calibration and slit-loss correction. Wide-slit spectra were obtained immediately after the narrow-slit observations. Narrow-slit observations were mostly taken in NODDING mode, while wide slits were taken in STARE mode. For sky subtraction, especially in the NIR where the background is severe, nodding along the slit was an integral part of the observing design (Chen et al., 2011, Gonneau et al., 2020).

The delivered products are rest-frame spectra, with wavelengths in air in DR2 and DR3. In DR2, the per-arm spectra are provided as FITS binary tables with wavelength, flux, and error columns; the calibrated flux unit is $\mathrm{erg\ s^{-1}\ cm^{-2}\ \AA^{-1}}$. DR3 extends this by distributing merged spectra over the full range when possible, together with metadata describing slit-loss correction, arm scaling, extinction estimates, and known quality flags (Gonneau et al., 2020, Verro et al., 2021).

3. Data releases, sample composition, and parameter coverage

XSL evolved through successive releases that expanded both spectral coverage and calibration fidelity.

Release	Content	Principal characteristics
DR1	246 spectra of 237 stars	UVB+VIS only, 3000–10200 Å, $2.45\,\mu$0
DR2	813 observations of 666 stars	Full three-arm coverage to $2.45\,\mu$1m, 2,388 arm spectra
DR3	830 spectra of 683 stars	Merged, de-reddened 350–2480 nm spectra, plus 20 added M dwarfs

DR1 established the library as a near-ultraviolet through optical data set spanning spectral types O–M and including LPV, C, and S stars (Chen et al., 2014). DR2 superseded that release by incorporating the NIR arm and providing 813 observations of 666 unique stars, with strong emphasis on red giants, red supergiants, AGB stars, carbon stars, and repeat observations of cool luminous variables (Gonneau et al., 2020). DR3 then provided 830 spectra of 683 unique stars as merged, de-reddened spectra over 350–2480 nm when possible, and filled a low-mass gap by adding 20 archival main-sequence M-dwarf spectra (Verro et al., 2021).

The parameter coverage derived for XSL is correspondingly broad. A homogeneous ULySS-based analysis provided atmospheric parameters for 754 spectra of 616 stars, spanning $2.45\,\mu$2, $2.45\,\mu$3, and $2.45\,\mu$4, with a couple of stars extending down to $2.45\,\mu$5 (Arentsen et al., 2019). DR3 describes this metallicity as an “equivalent metallicity” $2.45\,\mu$6 on the MILES scale for the majority of spectra (Verro et al., 2021).

The library’s sample design is not uniform in a purely statistical sense; it is intentionally enriched in evolved cool stars. That emphasis is methodological rather than incidental, because these objects dominate the near-IR light of intermediate-age and old stellar populations and are also the most difficult to represent with static model atmospheres (Chen et al., 2011, Lançon et al., 2018).

4. Reduction pipeline, flux calibration, and external validation

XSL reduction combines the ESO X-shooter pipeline with substantial custom post-processing. DR2 used the ESO X-shooter pipeline v2.6.8 in physical mode via EsoReflex, generated rectified 2D orders, and then performed custom 1D extraction based on Horne (1986), with robust bad-pixel rejection, inverse-variance weighting, and special handling of NIR detector artifacts such as the “chessboard effect” (Gonneau et al., 2020).

Telluric correction evolved significantly across releases. DR1 introduced a telluric-correction scheme based on a library of B/A-star telluric standards and principal-component reconstruction in the VIS arm, while using temporally close standards for cooler, more complex spectra (Chen et al., 2014). DR2 adopted molecfit, fitting molecular absorption by species appropriate to each arm and using an extended segmented approach in the VIS and NIR to determine precipitable water vapor globally and then fit local wavelength solutions and line-spread functions piecewise. Atmospheric extinction correction and response-curve calibration were then applied using spectrophotometric standards such as BD+17 4708, EG 274, Feige 110, LTT 3218, and LTT 7987 (Gonneau et al., 2020).

A crucial DR2 step was wavelength-dependent slit-loss correction. This was applied to approximately 85% of spectra by comparing wide-slit and narrow-slit observations, smoothing their ratio, fitting it with low-order polynomials, and multiplying the result into the narrow-slit spectrum. The remaining 15% were delivered without slit-loss correction and flagged accordingly (Gonneau et al., 2020). DR3 further introduced arm-scaling factors $2.45\,\mu$7 and $2.45\,\mu$8 relative to the VIS arm, and solved iteratively for those factors together with $2.45\,\mu$9 during full-spectrum merging and de-reddening (Verro et al., 2021).

Photometric validation showed that the DR2 synthetic broad-band colors agree on average with MILES and with the combined IRTF and Extended IRTF libraries to within approximately 1%, with typical errors on individual colors of 2–4%. Comparison with 2MASS showed systematics of up to 5% in some colors, attributed mostly to zero-point or transmission-curve errors, with larger dispersion for cool variables (Gonneau et al., 2020). DR3 validation against external libraries reported normalized rms deviations better than $R=\lambda/\Delta\lambda$0 for the majority of spectra in common with MILES, NGSL, and (E-)IRTF, and found a zero median offset with an rms scatter of 0.037 mag between Gaia published $R=\lambda/\Delta\lambda$1 colors and those synthesized from XSL DR3 spectra for 449 non-variable stars (Verro et al., 2021).

These improvements did not eliminate all continuum-shape issues. Independent optical IFU spectra from a MUSE subset of XSL stars showed that DR1 could differ from MUSE by 10–15% peak-to-peak across 4800–9300 Å in some cases, illustrating that order merging and slit losses remain structurally difficult problems even when line features are internally consistent (Ivanov et al., 2019). DR3’s merged spectra mitigate this, but dichroic regions, deep telluric bands, and residual arm-scale offsets remain documented caveats (Verro et al., 2021).

5. Atmospheric parameters, chemical characterization, and interpolation frameworks

The first homogeneous atmospheric-parameter catalogue for XSL was derived with ULySS using the empirical MILES interpolator as reference. The fitting used UVB and VIS spectra only, modeled as a multiplicative polynomial times a broadened template interpolator in $R=\lambda/\Delta\lambda$2. The reported precisions are 0.9% in $R=\lambda/\Delta\lambda$3, 0.14 dex in $R=\lambda/\Delta\lambda$4, and 0.06 dex in $R=\lambda/\Delta\lambda$5 for G- and K-type stars; 2.1%, 0.21 dex, and 0.22 dex for cool giants with $R=\lambda/\Delta\lambda$6; 1%, 0.14 dex, and 0.10 dex for other cool stars; and 2.6%, 0.20 dex, and 0.10 dex for hotter stars with $R=\lambda/\Delta\lambda$7K (Arentsen et al., 2019).

XSL later acquired an abundance dimension through the derivation of $R=\lambda/\Delta\lambda$8 and $R=\lambda/\Delta\lambda$9 for DR2 stars. Using the GAUGUIN automated abundance-estimation code, the analysis began from 611 stars and ultimately delivered precise abundances for 192 stars in magnesium, 217 stars in calcium, and 174 stars in both elements, within the domain $R \approx 9{,}200$0K, $R \approx 9{,}200$1, and $R \approx 9{,}200$2. The resulting abundance trends reproduce a plateau in the metal-poor regime followed by a decreasing trend even at supersolar metallicities, consistent with Galactic chemical-evolution expectations (Santos-Peral et al., 2023).

These parameter and abundance catalogues are not merely ancillary metadata. They support interpolation and synthesis frameworks built directly on XSL. In the XSL-based simple stellar population models, warm stars are represented with a global interpolator using polynomial expansions in $R \approx 9{,}200$3, $R \approx 9{,}200$4, and $R \approx 9{,}200$5, while cool dwarfs and hot stars use a local interpolator based on weighted averages in parameter cubes. Because TP-AGB stars are not well represented by pointwise interpolation alone, the models instead use average sequences for static O-rich giants, O-rich TP-AGB variables, and C-rich TP-AGB stars (Verro et al., 2021).

A plausible implication is that XSL’s long-term significance lies not only in its spectra themselves but in the combination of empirical spectra, homogeneous atmospheric parameters, and partial chemical characterization. That combination makes the library suitable for interpolation in both stellar-parameter space and, in a limited regime, abundance space (Arentsen et al., 2019, Santos-Peral et al., 2023).

6. Cool evolved stars, variability, and confrontation with model atmospheres

One of XSL’s distinctive scientific roles is its systematic treatment of cool evolved stars. The library contains dedicated subsets of carbon stars, oxygen-rich long-period variables, red supergiants, and related advanced evolutionary phases whose spectra are both astrophysically important and difficult to model (Chen et al., 2011, Verro et al., 2021).

For carbon stars, XSL assembled 35 simultaneous optical–near-IR spectra covering 0.3–2.4 $R \approx 9{,}200$6m at resolving power above approximately 8000. These data revealed a bimodality for stars redder than about $R \approx 9{,}200$7–1.6, divided by the presence or absence of the 1.53 $R \approx 9{,}200$8m absorption feature generally associated with HCN and C$R \approx 9{,}200$9H$R \approx 11{,}000$0. The stars showing that feature are essentially large-amplitude variables and also exhibit smoother near-IR spectra, weaker C$R \approx 11{,}000$1 bands, and signatures interpreted as circumstellar veiling by warm dust (Gonneau et al., 2016). Comparison with hydrostatic COMARCS models showed that broad-band colors and many molecular bands are reproduced well for $R \approx 11{,}000$2, whereas redder stars require additional reddening, dust emission, or both; for spectra with the 1.53 $R \approx 11{,}000$3m band, hydrostatic models are inadequate and dynamical models are required (Gonneau et al., 2017).

For oxygen-rich LPVs, XSL analyzed 160 spectra of 150 LPVs from the Milky Way and the Magellanic Clouds, focusing on $R \approx 11{,}000$4K. At fixed broad-band color, spectra with emission lines—often near maximum light—show more triangular spectral energy distributions, stronger TiO, VO, and H$R \approx 11{,}000$5O bands, a preferentially blue slope around 1 $R \approx 11{,}000$6m, and occasional probable TiO emission near 1.24 $R \approx 11{,}000$7m. Spectra without emission lines, at the same color, show stronger CO bandheads and more conspicuous Ca II triplet absorption (Lançon et al., 2018). Static PHOENIX models fit warm, small-amplitude LPVs near $R \approx 11{,}000$8K well, but cooler and larger-amplitude variables remain problematic; DARWIN dynamical models improve VO strengths, circumstellar dust effects, and the SED shape near 1 $R \approx 11{,}000$9m, yet still do not reproduce the most triangular observed spectra or the strongest phase-dependent molecular emission (Lançon et al., 2018).

A broader comparison between XSL DR2 and the PHOENIX-based Göttingen Spectral Library extended this conclusion across the HR diagram. With stellar parameters fixed to the DR2 catalogue, model spectral energy distributions are consistent with XSL for $R \approx 7{,}770$0K, but significant discrepancies appear below 5000 K. Allowing parameters to vary yields satisfactory SED fits down to about 4000 K, although UVB residuals and arm-to-arm parameter tensions persist, especially for cool giants and low-gravity stars (Lançon et al., 2020).

7. Stellar population synthesis, empirical leverage, and remaining limitations

XSL’s direct population-synthesis application is the construction of simple stellar population models from the near-ultraviolet to the near-infrared. The published XSL SSP models span 350–2480 nm at $R \approx 7{,}770$1, metallicities $R \approx 7{,}770$2, and ages above 50 Myr, though in practice they are safer above about 80 Myr at solar metallicity and above about 100 Myr at subsolar metallicity because of limitations in hot-star and supergiant coverage (Verro et al., 2021).

Within these models, the thermally pulsing AGB is explicitly important. Using PARSEC v1.2S + COLIBRI isochrones, TP-AGB stars contribute at least 40% of the $R \approx 7{,}770$3-band light between approximately 0.5 and 1.6 Gyr, peaking at approximately 0.8 Gyr and $R \approx 7{,}770$4 with approximately 55–60% of the $R \approx 7{,}770$5-band light. Above approximately 2 Gyr, the near-IR becomes dominated by the tip of the RGB (Verro et al., 2021). Because XSL includes separate average sequences for static O-rich giants, variable O-rich TP-AGB stars, and C-rich TP-AGB stars, its SSPs expand the range of predicted near-IR indices relative to other empirical-library models, including broader behavior in ZrO and Mg I $R \approx 7{,}770$6m (Verro et al., 2021).

The validation results are mixed in a way that is informative rather than contradictory. XSL SSPs reproduce the integrated optical colors of Coma cluster galaxies at the level of other semi-empirical models, and their optical diagnostic grids track standard trends closely. In the near-IR, however, the models differ noticeably from Coma cluster colors, with offsets of about 0.05–0.10 mag in $R \approx 7{,}770$7 or $R \approx 7{,}770$8 at fixed $R \approx 7{,}770$9 for solar and super-solar metallicities. The stated causes include TP-AGB treatment, separation of static and variable giants, merging across arms, abundance patterns, dust, and telluric residuals (Verro et al., 2021).

Several limitations therefore remain intrinsic to present-day use of XSL. Merged spectra still require caution around the dichroic intervals 545–590 nm and 994–1150 nm, and around the strongest telluric bands (Verro et al., 2021). Very cool giants remain incompletely represented at the extreme red end of some sequences, and variability is handled in practice by color-binned phase averaging rather than by cycle-resolved physical modeling (Verro et al., 2021, Lançon et al., 2018). XSL’s simultaneous coverage solves the problem of non-coherent optical–near-IR epochs, but it does not remove the underlying astrophysical diversity of pulsation phase, shock physics, dust formation, and molecular opacity errors. In that sense, the library functions both as an overview resource and as a stress test for stellar-atmosphere theory, especially below about 4000–5000 K (Lançon et al., 2018, Lançon et al., 2020).

XSL’s broader significance follows from that dual role. It is simultaneously an empirical library for optical–near-IR population synthesis and a benchmark against which synthetic spectra, interpolation schemes, and abundance-aware stellar-population models can be judged. The combination of simultaneous wavelength coverage, moderate-to-high resolution, dense evolved-star content, and progressively improved calibration has made it a reference data set for bridging optical and near-infrared spectroscopy in both stellar and extragalactic research (Gonneau et al., 2020, Verro et al., 2021, Verro et al., 2021).