HST Low-resolution Stellar Library

Stellar libraries are important tools used in far-flung corners of astronomy and astrophysics. They contain stellar spectra of a number of preselected stars in different wavelength regimes (ultraviolet (UV), visible, near-IR), a variety of spectral resolutions, and with varied attention to flux calibration. Examples include a library of stellar spectra by Jacoby et al. (1984), XSHOOTER (Verro et al. 2022a), MILES (Sánchez-Blázquez et al. 2006), Indo-US (Valdes et al. 2004), the Infrared Telescope Facility (Cesetti et al. 2013), ELODIE (Soubiran et al. 1998; Prugniel et al. 2007), Lick (Worthey et al. 1994, 2014), and UVES-POP (Bagnulo et al. 2003) libraries. Such libraries are often incorporated into stellar population synthesis models. For example, the MILES library (Sánchez-Blázquez et al. 2006) was used to compute simple stellar population (SSP) spectral energy distributions (SEDs) in the optical wavelength range with comprehensive metallicity coverage (Vazdekis et al. 2010; Falcón-Barroso et al. 2011). There are many other examples, such as Bruzual & Charlot (2003), Verro et al. (2022b), Le Borgne et al. (2004), Vazdekis et al. (2012), and Worthey et al. (2022b). On a star-by-star basis, libraries can be used to infer stellar parameters like T_eff, $\mathrm{log}\,g$, and [Fe/H] (e.g., Wu et al. 2011). Stellar libraries also find application in the study of stellar clusters (Alloin 1996; Deng & Xin 2010). One notable example is the BaSeL 3.1 stellar SED library (Lejeune et al. 1997, 1998; Westera & Buser 2003). This library is suitable for study of clusters at low metallicities, and has been exploited for the study of globular clusters (Bruzual et al. 1997; Kurth et al. 1999; Weiss & Salaris 1999), open clusters (Pols et al. 1998; Lastennet et al. 1999), and blue stragglers (Deng et al. 1999). When well flux-calibrated, stellar libraries are also very important for the characterization and performance evaluation of observational missions like Gaia (Lastennet et al. 2002; Sudzius & Vansevicius 2002). Several stellar libraries are built into the exposure time calculators for the Hubble Space Telescope (HST) and JWST. They even find use in educational products such as the University of Gettysburg's CLEA and VIREO or New Mexico State University's GEAS laboratory software packages to illustrate the trends among stellar spectra.

Spectral resolution and wavelength coverage vary among the various existing libraries (see Table 1 of Verro et al. 2022a), but none of them extend shortward of 300 nm into the UV regime except those of Wu et al. (1983b) and Fanelli et al. (1990), who present 172 and 218 stellar spectra, respectively, observed by the International Ultraviolet Explorer (IUE). An important motivation for the present HST-based library is to relieve the relative scarcity of spectral data in the UV.

Study of integrated spectra in the UV allows us access to the hottest stars, which are main-sequence turnoff stars with some blue-straggler contribution. For older stellar populations, UV-bright populations include blue horizontal branch and post–asymptotic giant branch (PAGB) stars (Koleva & Vazdekis 2012). An important goal is to isolate the various main sequences to chart the star formation history (SFH) of the Galaxy (Vazdekis et al. 2016). The d log age/d log Z = −3.2 age/metallicity degeneracy (Worthey 1994) becomes more like ≈ −1/1 in the UV. In UV, we have an abundance of strong absorption features that help constrain SFH, metallicity, and abundance ratios better (Ponder et al. 1998; Chavez et al. 2007; Toloba et al. 2009; Serven et al. 2010). Needless to say, if we want to extend the limit on redshift (z) for stellar population studies, the UV regime is of the utmost importance (Pettini et al. 2000; Daddi et al. 2005; van Dokkum & Brammer 2010).

Wu et al. (1983a) and Fanelli et al. (1992) gave the first large, systematic spectral library in UV using data from IUE. The library contained spectra of around 218 stars with a spectral resolution of 7 Å. HST's Space Telescope Imaging Spectrograph (STIS) improves upon IUE in both flux calibration and spectral resolution. Forty O, PAGB, and He-burning stars were observed with STIS to make a hot-star spectral library (Khan & Worthey 2018a). Made by stitching together spectra from three different gratings, these spectra have a wavelength coverage from ∼2000 Å to ∼10000 Å with a resolution of R ≈ λ/Δλ ∼ 1000. The hot-star library was modeled after an earlier effort called the Next Generation Spectral Library (NGSL; Gregg et al. 2006) which so far has not been completely described in the literature. The NGSL covers a wide range of stellar parameters, including metallicity (Heap & Lindler 2010; Koleva & Vazdekis 2012; Vazdekis et al. 2016). The original proposal was to obtain spectra of close to 600 stars via "snapshot" style programs (GO9088, GO9786, GO10222, and GO13776), in which single orbits left stranded between larger programs are exploited for short observations. Spectra of more than half of the stars that were observed (around 374 stars corresponding to proposals GO9088, GO9786, and GO10222) were reduced and made publicly available by Heap & Lindler (2009). The main intent of this paper is to provide a reduction of the full library to the community. The spectral quality is improved by applying additional corrections such as scattered-light, slit-off-center, and dust corrections.

We also investigate the Mg ii λ2800 feature, which is a pair of resonance lines designated by h and k. Boehm-Vitense (1981) used high-resolution IUE spectra on F stars to chart four origins for Mg2800 profile morphology: the main, broad stellar absorption feature, a narrower chromospheric emission core, a rare, even narrower self-absorption, and interstellar absorption. Fanelli et al. (1990) also noted that, when in emission, it probably indicates a chromospheric origin. Linsky & Ayres (1978) argue that most of the Ca ii emission (λ λ 3933, 3968) arises in the lower chromosphere, Mg ii in the middle chromosphere, and Lyα in the upper chromosphere, and that, together, these resonance features provide the bulk of the radiative cooling that occurs in the layers exterior to the photosphere.

The dynamo action brought about by differential stellar rotation is one of the most commonly accepted mechanisms for magnetic field generations in main-sequence stars (Hartmann & Noyes 1987; Fröhlich et al. 2012; Quentin & Tout 2018). Chromospheric activity is generally associated with strong magnetic fields (Musielak & Bielicz 1982; Brown et al. 2022). Since stellar rotation is expected to slow down over the lifetime of a star, activity can be presumed to decrease (Barry 1988). This leads to the possibility that chromospheric activity indicators (Ca ii, Mg ii, or Hα) may provide relatively precise chronometric information, at least in predefined spectral-type bands (Barry 1988). However, in general ages would be poorly constrained (Pace 2013). In addition, acoustic shocks without a magnetohydrodynamic component may also contribute to the chromospheric activity (Buchholz et al. 1998; Martínez et al. 2011; Pérez Martínez et al. 2014).

Connections between Mg2800 strengths and the astrophysics of stellar properties are still tenuous, but could eventually lead to realistic chromosphere models as a function of stellar type and magnetic field strength. In the meantime, we have various empirical clues. Houdebine & Stempels (1997) find that, at spectral type M1, metal-deficient stars are also activity deficient. Smith et al. (1991) compare Mg2800 with the Ca ii S index (Vaughan & Preston 1980), which measures the width of the emission rather than its strength. They also find that the available sample of 20 FGK stars can be separated into high-activity and low-activity groups at an approximately 4:16 ratio, but that Mg2800 displays a large range of values even among the low-activity group (Martínez et al. 2011; Pérez Martínez et al. 2014). Interstellar absorption usually dominates in OB stars (Khan & Worthey 2018b). Due to its wide coverage of parameter spaces, the present library can confirm or extend these trends.

This paper is organized as follows. We describe the observations and sample in Section 2. The data reduction process is detailed in Section 3, and additional corrections that affect the continuum shape in Section 4. The format of the data catalog is described in Section 5. In Section 6, we investigate the Mg ii 2800 feature and chart the systematics of chromospheric activity across the H-R diagram. We conclude in Section 7 with a summary of the results and a discussion of their implications.

The stars in the library were selected to cover T_eff–L–Z space insofar as the Galaxy could provide them. For example, given that the metal-poor components of the Milky Way are also ancient in age, no luminous, low-metallicity stars exist. The parameter coverage is shown in Figure 1 in $\mathrm{log}\,g$–$\mathrm{log}\,{T}_{\mathrm{eff}}$ space with metallicity indicated by symbol type. Figure 2 on the other hand shows the distribution of all the stars in different metallicity bins. The impact of including stars from GO13776 significantly improves coverage of T_eff–L–Z space, as shown in Figures 1 and 2. Several A-type field horizontal branch stars were observed to attempt to fill in the warm-and-metal-poor gap. Desirable faint stars, such as individual Small Magellanic Cloud stars, could not be observed due to the one-orbit limit on exposure time. The target list is hand selected, and should not be used for any statistical inferences. In addition, HST's SNAP mode selects from a larger input list according to schedulability, leading to further randomization.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

During the orbit in which they were targeted, the NGSL stars were observed by cycling through three different gratings. G230LB sees in UV (central wavelength of 2375 Å), G430L sees in blue (central wavelength of 4300 Å) and G750L sees in red (central wavelength of 7751 Å). The three gratings overlap at 2990–3060 Å and 5500–5650 Å (Gregg et al. 2006). The CCD detector was employed for these observations. UV exposure times were longer than exposures in the blue or red. A 02 slit, equivalent to ±2 pixels (Hernandez et al. 2012; Prichard et al. 2022) was used for all the observations and a fringe flat was taken for the G750L grating at the end of each sequence of exposures.

In addition to the usual observational defects (cosmic-ray hits, charge transfer efficiency effects, bad pixels, and photon noise) these data suffer from two additional sources of error that affect fluxing. First, the G230LB grating scatters red light into the UV, creating a spurious signal that must be corrected (Lindler & Heap 2010; Worthey et al. 2022a). Second, scatter in telescope pointing plus a narrow slit led to situations in which the jaws of the slit sliced off portions of the point-spread function (PSF). Because STIS is an off-axis instrument, the PSF is not symmetrical, so the resultant attenuation is wavelength dependent. Fortunately, both of these effects can be modeled, and we give details in Section 4.

Of minor note, STIS spectral flux calibrations have improved since the previous version of the NGSL library was placed at the Mikulski Archive for Space Telescopes (MAST).

All 556 targets from proposals GO9088, GO9786, GO10222, and GO13776 were reduced from raw observation files. Out of these 556 targets, 514 have been reduced completely and additional corrections have been applied. The remaining 42 targets have not been reduced either because of faulty fringe-flat files or because of the absence of one of the observations in UV, blue, or red. The raw files for all the observations (which include observations in UV, blue, and red as well as CCD flats) were downloaded from the Space Telescope Science Institute (STScI) archive. The reduction process is carried on using the stistools Python3 package developed by STScI. Out of these 514 targets, there is a duplication for one of the targets (GJ 614 and HD 145675) leaving us with 513 unique targets.

The reduction procedure consisted of several steps starting from cosmic-ray correction to combining disparate spectral windows into one continuous spectrum for each star.

3.1. Cosmic-Ray Correction

Cosmic-ray corrections are more crucial for observations using the G230LB grating that was used for longer-duration UV observations. This is illustrated in Figure 3 where cosmic rays are common in the G230LB exposure. Accordingly, all multiple UV observations were run through the ocrreject function of stistools. This function combined two sets of science observations in UV into a single file. In order to run ocrreject, we needed to have at least two observations at each pointing. Unfortunately, the UV raw files from proposals GO10222 and GO13776 did not have multiple UV exposures. For these, bad pixels were removed manually from the spectra.

Zoom In Zoom Out Reset image size

3.2. Defringing in the Red

Fringes are interference patterns caused by photons with wavelengths that are integral multiples of the width of the CCD layer. In STIS, fringe patterns are prominent redward of ∼7000 Å and reach a peak-to-peak amplitude of 25% at 9800 Å (Kimble et al. 1998; Malumuth et al. 2003). The G750L grating produces unwanted fringe patterns. Once per orbit, a fringe flat was obtained using the tungsten lamp on board HST.

The defringing process was carried out using the defringe tool of stistools (for details, see https://stistools.readthedocs.io/en/latest/). The following three methods were used in sequence for all the NGSL observations.

• 1.  
  normspflat: this method normalizes the fringe flat that is associated with each observation.
• 2.  
  mkfringeflat: this method cross correlates the normalized fringe flat with that of the observed spectrum to match the fringes between the two. It minimizes the rms within a given range of shift and scale values to find the best shift and scale.
• 3.  
  defringe: this method actually defringes the observed spectrum by removing the fringing pattern from the observed spectrum using the shifted and scaled fringe flat.

Figure 4 shows the red spectrum of HD102212 before and after defringing.

Zoom In Zoom Out Reset image size

While defringing the red spectra it was observed that no proper fringe flat is available for 27 targets. We dropped these stars from further analysis and thus reduced the total number of targets from 556 to 529. While trying to defringe red spectra from GO13776. although some of the targets from GO13776 have two or three red spectra, only one of them defringed properly. An investigation yielded an observing irregularity. For run GO13776, the G750L (red third of the spectrum) target exposures were preceded by a fringe flat through the 0.3 × 0.09 notch aperture, which is placed near row 512 of the chip (the UV and blue spectra were taken at the E1 pseudoaperture around row 900 of the CCD). The telescope was slewed to place the target star at row 512 of the chip rather than 900, and one exposure taken through the nominal 52 × 0.2 aperture. Due to an oversight, positional dithering occurred. The telescope was slewed 05, and an exposure was taken through the 52 × 0.2 aperture followed by an exposure through the 52 × 0.5 aperture. This last exposure eliminates edge effects and provides the best fluxing, but it cannot be fringe corrected using the data collected on orbit. Therefore, only one red spectrum (for each target) was used for run GO13776. For three targets from GO13776 (HD 65589, HD 84035, and HD 185264), none of the red observations could be satisfactorily defringed. These stars were also dropped from further analysis which reduced the total number of stars from 529 to 526.

Also unique to proposal GO13776 was that the last pair of red observations were often a pair obtained through the 02 and 05 apertures. Although these could not be defringed due to the shift along the aperture center line, they could be used to create a relative flux correction, should the star have been placed off the central line of the entrance aperture. A smoothed division of these two spectra was applied to the first, defringed observation in all cases where the complete set of observations exists.

3.3. 1D Extraction

3.4. Bad-pixel Handling

As mentioned in Section 3.1, a cosmic-ray rejection algorithm was not applied to blue and red observations. Even after applying cosmic-ray rejection to the UV observations, the UV spectra had leftover wild pixels of unusually high and nonastrophysical flux (or counts). In order to mitigate this problem, each observation for each target was checked manually for bad pixels and those pixel locations were flagged. This step generated a single text file for each target containing information on the number of bad pixels for each observation and values for those pixels. Figure 5 shows an example of the presence of bad pixels in the spectrum.

Zoom In Zoom Out Reset image size

We tried our best to remove as many bad pixels as possible from each spectrum, but there are some stars for which many bad pixels could not be removed cleanly.

3.5. Special Flux Scalings

3.6. Relative Velocities and Template Matching

The NGSL stars were chosen to encompass a broad interval of [Fe/H], $\mathrm{log}\,g$, and T_eff (Gregg et al. 2004). Galactic halo stars are mostly metal poor but can possess high relative velocity with respect to the local rest frame (Du et al. 2018). Thus, some of the stars in NGSL have relative velocities >250 km s⁻¹. This fact called for a relative velocity correction before bringing all the spectra to the rest frame. To be consistent, we applied the relative velocity correction to all 514 stars even when the effects would be negligible. The nonrelativistic formula was used to correct for the relative velocity:

$$\begin{eqnarray}&&d\lambda =\displaystyle \frac{v}{c}\times \lambda ,\end{eqnarray} \tag{ 1 }$$

where d λ is a correction to the wavelength λ, v is the relative velocity of the star in kilometers per second and c is the speed of light in kilometers per second. d λ was added or subtracted from corresponding λ values depending on the sign of v. The values of v were obtained from the SIMBAD astronomical database (Wenger et al. 2000).

After correcting for the relative velocities, residual shifts to the rest frame (vacuum wavelengths) were estimated by comparisons with template spectra. The choice of template spectrum was made based on the effective temperature of the particular star. The high-resolution templates were rebinned to match the observed wavelength points, then cross correlated. The following templates were adopted.

• 1.  
  Synthetic spectra were used for cool stars (T_eff < 5000 K) and warm stars (5000 K < T_eff < 8000 K). The synthetic spectra were generated using Worthey's (1994) model of stellar population.
• 2.  
  The observed spectrum of α Lyrae was used for hot stars (T_eff > 8000 K)

The cross correlation function (in angstroms) was fitted with a single peak Gaussian function. Figure 6 shows a part of the spectrum for HD 102212 and illustrates the amount of shift present in the observed spectrum with respect to the template. Correlation value as a function of shift is shown in Figure 7 (for the same star HD 102212). The same template was used for all the observations of a particular target. To speed convergence, we added initial shifts of 3, 9 and 14 Å to UV, blue, and red observations, respectively. This "pre-shift" evidently arises because wavelength calibrations were not performed on orbit for NGSL, and so a default wavelength solution was assigned.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

3.7. The Composite Spectrum

To assemble a single contiguous spectrum, we combined bad-pixel information and shift information from template matching to splice all the observations for a particular target into one final spectrum. The shift obtained for each observation was added algebraically to the wavelength values. While applying the bad-pixel information, we devised a method for suppressing the bad pixels. We first divided the range of each observation into 50 overlapping boxes of 40 pixels each. For each box, we found out the average flux weighted by the variance (f_box ) using the following formula:

$$\begin{eqnarray}&&{f}_{\mathrm{box}}=\displaystyle \frac{{f}_{1}{v}_{1}+{f}_{2}{v}_{2}+...+{f}_{40}{v}_{40}}{{v}_{1}+{v}_{1}+...+{v}_{40}},\end{eqnarray} \tag{ 2 }$$

where f_n is the flux at the nth wavelength value for a particular box and v_n is the corresponding variance (defined by ${v}_{n}=1/{e}_{n}^{2}$ where e_n is corresponding error in flux for that particular wavelength value). These flux values were then linearly fitted over the range of observation. Now, the flux at the previously identified bad pixels was set to a flux value according to this linearly extrapolated relation. It should be noted that the error values at the bad pixels were inflated by a factor of 1000 before calculating f_box. This was done to make sure that the erroneous pixels do not contribute much to the weighted average (as bad pixels generally have very high flux values).

Once the flux values at the bad pixels were set according to the abovementioned algorithm, we then calculated the weighted average flux value for all the observations of a particular type (for, e.g., UV, blue, or red) at a particular wavelength value. For example, if there are two UV observations for a particular target, then the average UV flux at the nth wavelength value (f^UV _n ) is given by

$$\begin{eqnarray}&&{f}_{n}^{{\rm{UV}}}=\displaystyle \frac{{f}_{n}^{1}{v}_{n}^{1}+{f}_{n}^{2}{v}_{n}^{2}}{{v}_{n}^{1}+{v}_{n}^{2}},\end{eqnarray} \tag{ 3 }$$

where f¹ _n and f² _n are UV fluxes at the nth wavelength value for the first and second observations, respectively, and v¹ _n and v² _n are corresponding variances as defined before. This formula can easily be generalized for more than or fewer than two observations. Once this operation was performed for all the observations of a target, we then combined all the observations to make a single spectrum for a target treating λ < 3057 Å as a UV observation, 3057 Å < λ < 5679 Å as a blue observation and λ > 5679 Å as a red observation. This algorithm does not apply without any caveats as sometimes the flux values at bad pixels were negative. Users are advised to be careful of such artifacts in the spectrum by considering the uncertainty we assign.

The G230LB grating scatters some red light onto the portions of the CCD where UV is expected (Worthey et al. 2022a). This is a problem mainly for cool stars (T_eff ≤ 5000 K) where we do not expect significant UV flux. This section summarizes the results from Worthey et al. (2022a) on scattered-light as well as slit-off-center corrections. We also applied these corrections to the 514 NGSL stars that we have reduced.

4.1. Scattered-light Correction

The scattered light (S(λ)) is approximated by the formula (Worthey et al. 2022a)

$$\begin{eqnarray}&&S(\lambda )={K}_{0}\times (1+0.00104\times (\lambda -2000)),\end{eqnarray} \tag{ 4 }$$

where K₀ is the scattered-light count rate at 2000 Å and λ is the wavelength. For targets with T_eff < 5000 K, K₀ is given by the median count rate around 2000 Å (median count rate for 1950 Å < λ < 2050 Å). Two stars in our list, HD 124547 and HD 200905, are spectroscopic binary stars with T_eff < 5000 K. For these two stars, K₀ calculated using the average count rate of around 2000 Å resulted in overcorrection of the spectra. After visually inspecting the spectrum for these two stars, the K₀ values were modified by hand to mitigate the problem of overcorrection. For targets with T_eff > 5000 K and for which V magnitudes (m_v ) are available, K₀ is given by

$$\begin{eqnarray}&&{K}_{0}=426\times {10}^{-0.4{m}_{v}}.\end{eqnarray} \tag{ 5 }$$

However, for some of the targets (with T_eff > 5000 K) m_v is not available. For such targets, K₀ is given by

$$\begin{eqnarray}&&{K}_{0}=1.78\times {10}^{-7}\times C,\end{eqnarray} \tag{ 6 }$$

where C is the integrated count rate between 2000 and 10000 Å. S(λ) was then subtracted from overall count at each λ. Figure 8 shows an example of scattered-light correction applied to the spectrum of HD102212.

Zoom In Zoom Out Reset image size

After applying the abovementioned formula of S(λ) for all the 514 stars, 96 stars (T_eff > 5000 K) were overcorrected and 8 stars (T_eff > 5000 K) were undercorrected as judged by inspection of the spectra. For these cases, the coefficient values (426 in Equation (5) and 1.78 × 10⁻⁷ in Equation (6)) were iteratively modified to calculate K₀ until the discrepant star fell among its peers in the UV. The updated K₀ values were then used to calculate S(λ) for those 104 targets.

4.2. Slit-off-center Correction

The NGSL targets were observed using the 02 slit. If the target is not placed at the center of the slit, light at the edges of the PSF gets attenuated by the slit edges. Because the STIS instrument is off axis, the PSF is asymmetric, and the attenuation is wavelength dependent. To correct for the attenuation effect, we use the attenuation factor (D_λ ) which is given by Worthey et al. (2022a):

$$\begin{eqnarray}&&{D}_{\lambda }=a+{bq}+{{cq}}^{2}+{{dq}}^{3}+{{eq}}^{4}+{{fq}}^{5}+{{gq}}^{6},\end{eqnarray} \tag{ 7 }$$

where $q=\sqrt{\lambda /4500}$. The coefficients for the above formula at different slit-off-center values are given in Table 3 of Worthey et al. (2022a). The slit-off-center value for each of the 514 NGSL spectra was calculated during the defringing process as outlined in Section 3.2. It is obvious that the slit-off-center values for our 514 targets were not matching the exact values given in Table 3 of Worthey et al. (2022a). The D_λ curve (as a function of λ) for each of our targets was calculated as a linearly interpolated curve between the two nearest D_λ curves (for which coefficients are available from Worthey et al. 2022a). Once the D_λ curve was calculated for each target, the flux of that target was divided by D_λ at each λ value.

4.3. Dust

We compiled interstellar dust extinction data for our 513-star library sample. Koleva & Vazdekis (2012) give nonnegative A_V values for around 341 stars. A_V for 44 stars are calculated by us (following Khan & Worthey 2018b) by matching an observed spectrum with a synthetic spectrum and then fitting a one-variable extinction law from Fitzpatrick (1999). The rest of the A_V values are taken from the GALExtin website version 1.2 (Amôres et al. 2021) using a three-dimensional Galactic extinction model by Drimmel et al. (2003). These A_V values are used to find the E(B − V) values using the following equation:

$$\begin{eqnarray}&&E(B-V)=\displaystyle \frac{{A}_{V}}{3.1}.\end{eqnarray} \tag{ 8 }$$

The extinction law of Fitzpatrick (1999) was used to correct the spectra to dust-free versions. Possible self-reddening for mass-losing stars was not considered. The E(B − V) values were also used to deredden the observed colors that we use for the analysis below.

5.1. Archived Spectra

All 513 spectra have been made available at http://astro.wsu.edu/hststarlib/ and at MAST as a High Level Science Product via doi:10.17909/kmdt-pw63 in separate FITS (Wells et al. 1981) files. Each FITS file contains five extensions, briefly described in Table 2.

Table 2. Brief Description of the FITS File Structure

Extension	Description
Primary	Contains no data. The header contains information about basic stellar parameters ([Fe/H], $\mathrm{log}\,g$, etc.) and averaged pointing information. Exposure-level pointing is available from the original MAST archive files.
Flux Table	Binary table extension with columns for wavelength (in angstroms), uncorrected flux, scattered-light-corrected flux, scattered-light- and slit-off-center-corrected flux, and scattered-light-, slit-off-center- and dust-corrected flux (fluxes are in erg s−1 cm−2 Å). Flux errors are also included as separate columns.
Count Rate Table	Binary table extension with columns for wavelength (in angstroms), uncorrected count rate, scattered-light-corrected count rate, scattered-light and slit-off-center-corrected count rate, and scattered-light, slit-off-center and dust-corrected count rate. Uncertainties are also included as separate columns.
Flux Table (Log Scale)	This binary table extension contains the same information as the Flux Table but the wavelengths are spaced on a logarithmic scale with $\mathrm{log}\,{\rm{\Delta }}\lambda$ = 0.0002.
Count Rate Table (Log Scale)	This binary table extension contains the same information as the Count Rate Table but the wavelengths are spaced on a logarithmic scale with $\mathrm{log}\,{\rm{\Delta }}\lambda$ = 0.0002.

Download table as:  ASCIITypeset image

Table 3 summarizes a mixture of astrophysical and reduction-specific metadata for each stellar target.

Table 3. Stellar Metadata

Simbad	Header	Teff	log g	[Fe/H]	B	V	π	${({M}_{V})}_{0}$	dSlit	vr	K0	AV	src	Mg2800	Hα	Hβ	Note
Name	Name	(K)	(dex)	(dex)	(mag)	mag	(mas)	(mag)	(pixel)	(km s−1)	(ADU)	(mag)		(mag)	(mag)	(mag)
HD 60319	HD060319	5907	4.03	−0.82	9.46	⋯	10.99	⋯	−0.20	−34.1	0.2	0.08	1	1.27	0.07	0.08	⋯
G 202-65	G202-65	6656	4.25	−1.37	⋯	⋯	3.88	⋯	1.00	−245.6	0.0	0.00	1	0.61	0.12	0.15	⋯
HD 185351	HD185351	4921	2.95	0.01	6.11	5.17	24.22	2.00	0.80	−6.6	5.2	0.09	1	0.45	0.05	0.04	⋯
HD 72184	HD072184	4643	2.84	0.23	7.01	⋯	14.55	⋯	−0.10	16.5	2.4	0.11	1	0.15	−0.01	0.03	⋯
HD 126614	HD126614	5453	3.87	0.53	9.66	8.79	13.65	4.41	−0.20	−32.9	0.2	0.05	1	0.67	0.03	0.08	⋯

Note. In this table, B and V are as observed (not dereddened), but ${({M}_{V})}_{0}$ is dereddened. The "src" column is for V-band extinction A_V : 1—Koleva & Vazdekis (2012); 2—our derivation based on comparison with synthetic templates; or 3—Drimmel et al. (2003). The "Note" column refers to objects noted in Section 5.2: 1—noisy; 2—possible extraction error; 3—chemically peculiar; 4—binary that does or may suffer from compositeness; 5—photometric variable.

Only a portion of this table is shown here to demonstrate its form and content. A machine-readable version of the full table is available.

Download table as:  DataTypeset image

5.2. Notable Objects

In this section, we explore the chromospheric activity of the 513 NGSL stars after full reduction, including extinction corrections. Wilson & Vainu Bappu (1957) showed that the absolute visual magnitudes (M_V ) of late-type stars correlate linearly with logarithm of H and K emission-line width of Ca ii (the Wilson–Bappu effect) and Mg2800 h and k share this behavior (Elgarøy et al. 1999; Cassatella et al. 2001). However, because our spectra are low resolution we could not reliably compute an analogous width for the twin Mg ii 2800 emission lines. We therefore measure overall strength only.

To summarize the strength of Mg ii 2800 emission, we adopt an equivalent-width-style index (Mg2800):

$$\begin{eqnarray}&&\mathrm{Mg}2800=-2.5\times {\mathrm{log}}_{10}\displaystyle \frac{\int {F}_{\lambda }^{i}\ d\lambda }{\int {F}_{\lambda }^{c}\ d\lambda },\end{eqnarray} \tag{ 9 }$$

where ${F}_{\lambda }^{i}$ is the observed flux within the spectral feature band and ${F}_{\lambda }^{c}$ is the expected flux without the spectral feature within the same band. We approximate ${F}_{\lambda }^{c}$ by defining a pseudocontinuum from side bands. A line is drawn between the central wavelengths and average flux values of the two side bands. The Mg2800 central feature band is defined as wavelengths between [2784, 2814 Å]. The blue side band is [2762, 2782 Å] and the red one is [2818, 2838 Å]. These definitions of feature and side bands are adopted from Fanelli et al. (1990). Figure 9 shows the adopted definitions for the Mg2800 central feature band and the two side bands (blue and red).

Zoom In Zoom Out Reset image size

We keep the units (magnitudes) adopted by Fanelli et al. (1990). A negative index value signifies net emission and a positive value signifies absorption. Figure 10 displays Mg2800 as a function of dereddened color for the library stars. Hot stars have negligible Mg2800 absorption. We also note that, although the sample contains some strongly active Be stars, these stars show no anomalous Mg2800 absorption or emission. Mg2800 absorption increases from A0 stars to Sunlike stars [(B − V)₀ = 0.65] and declines thereafter. In cool stars, both giants and dwarfs, chromospheric Mg2800 emission overtakes photospheric absorption at (B − V)₀ ≈ 1 and dominates for cooler stars. Figure 10 agrees well with Figure 5(c) of Fanelli et al. (1990).

Zoom In Zoom Out Reset image size

For the plots herein, the distinction between giants and dwarfs is approximated via the color–magnitude diagram (CMD) as shown in Figure 11. Stars warmer than (B − V)₀ = 0 or fainter than M_V = 3.0 were simply considered dwarfs regardless of their spectral type. For (B − V)₀ > 0, any star with M_V > 6.25 × (B − V)₀ − 2.5 is considered a dwarf whereas M_V < 6.25 × (B − V)₀ − 2.5 is considered a giant.

Zoom In Zoom Out Reset image size

The Figure 11 CMD is color coded by Mg2800 value. The verticality of the color bands shows again that both cool dwarfs and cool giants have similar Mg2800. Their chromospheres are similar by this measure despite vastly different size scales (∼0.1R_⊙ versus ∼100R_⊙). The emission gradually changes to absorption for warm stars and declines to near zero for hot stars. Note that some distant stars may have extra Mg2800 absorption due to warm interstellar material along the line of sight.

Even given the intentional diversity in sample selection, outliers are relatively few. One is a G9 giant HD 222093, at (B − V)₀ ≈ 1 and M_V ≈ 1 in Figure 11. It has a high value for Mg2800 absorption, signified by the red color in Figure 11. The star's spectrum shows emission peaks at the core of a broad absorption feature at 2800 Å, normal for a star whose absorption competes with emission at (B − V)₀ ≈ 1, but this star's emission is weak. HD 222093 also shows up in Figure 10 as the sole star with the highest Mg2800 absorption at (B − V)₀ ≈ 1.

Figure 12 plots Mg2800 versus metallicity, color coded by (B − V)₀. It is clear from this figure that no strong correlation exists between these two quantities in any color regime, particularly for cools. An anticorrelation among cool stars might have been expected from the Ca ii H and K results of Houdebine & Stempels (1997) who found that metal-poor stars are activity deficient, but we see no such trend. Peterson & Schrijver (1997) report that chromospheric characteristics do not have any metallicity dependence.

Zoom In Zoom Out Reset image size

A subtle declining trend among medium-temperature stars in Figure 12 deserves a note and an additional figure, namely Figure 13, which restricts the color range to be near solar (0.5 < (B − V)₀ < 0.8). Because these are positive values of Mg2800, indicating absorption, one might expect a monotonic increase of Mg2800 with [Fe/H]. Mg2800 absorption does increase for metal-poor stars (−2 < [Fe/H] < −1) but then the index value saturates and falls for metal-rich objects. Although Lovis et al. (2011) report not much of a correlation with the Ca ii lower envelope and metallicity, a decreasing trend can be seen in Meunier et al. (2022) for [Fe/H] > −1. With the help of synthetic spectra, two sequences of which are also plotted in Figure 13, the reason appears to be a simple curve of growth argument. Mg2800 is a resonance feature that scales approximately as the abundance of the Mg ii ion. It reaches full depth at [Fe/H] ∼ −1, but the flanking (in wavelength) absorption features from a plethora of atomic species are still weak. From [Fe/H] ∼ −1 and higher, these weak features will grow faster than the central Mg ii absorption pair. As the pseudocontinuum drops, the Mg2800 index drops. Parenthetically, the relatively poor agreement of synthetic spectra and observed spectra in Figure 13 should be no surprise. The UV spectrum is crowded, its lines have not received as much attention as optical ones, and for warm and cool stars the wavelength regime is on the blue side of the blackbody curve, exposing defects in the upper layers of the model atmosphere due to the absence of backwarming.

Zoom In Zoom Out Reset image size

Hα emission is a separate indicator of stellar chromospheric activity (Montes et al. 1995; Cincunegui et al. 2007; Gomes da Silva et al. 2014) and also magnetic flare activity. An index for the Hα feature is calculated using the passband definitions of Cohen et al. (1998) but here we convert it to magnitude units (Equation (9)). The spectral feature band is [6548, 6578 Å], the blue pseudocontinuum is [6420, 6455 Å], and the red pseudocontinuum is [6600, 6640 Å].

Mg2800 and Hα are plotted against each other in Figure 14. The strongest Hα emitters are Be stars, generally assumed to be young stars with disks (Gray & Corbally 2009). We might also expect to catch some flaring M dwarfs, but apparently none of the M dwarfs were observed during outbursts, as we see no cool dwarfs scattering to negative Hα values. The "triangle" in the positive–positive quadrant arises because peak Hα absorption occurs among hotter stars than peak Mg2800 absorption. Among cool stars with negative Mg2800, the mild correlation is due to expected Hα index absorption behavior from species unrelated to Hα itself, such as TiO (e.g., Valdes et al. 2004). That is, it is a consequence of the strong Mg2800–temperature anticorrelation in cool stars, and does not imply Hα emission at all. The chromospheric emission has been studied using Ca ii H and K and Hα as activity indicators. Some of these studies find a good correlation between these two indicators (please see the review by Linsky 2017, and references therein) whereas some studies find strong temporal anticorrelation between these two indices in a certain percentage of their selected stars (Cincunegui et al. 2007; Gomes da Silva et al. 2014). Meunier et al. (2022) suggests that this type of behavior needs to be explained by considering contributions from plages and filaments.

Zoom In Zoom Out Reset image size

Two stars lie at anomalously negative Hα values. They are HD 126327 (giant) and GL 109 (dwarf). Presumably, HST serendipitously observed these objects during flare events.

The correlation between Ca ii H and K core-emission strength (a third stellar activity indicator) and Hα emission is also well studied. Some authors report a positive correlation between the two (Pasquini & Pallavicini 1991; Montes et al. 1995), some a lack of correlation, and some a negative correlation (Cincunegui et al. 2007; Gomes da Silva et al. 2011). Our Mg2800 results shed little insight into this uncertain area.

This paper presents a new reduction of the Next Generation (HST/STIS low resolution) Spectral Library that includes updated flux calibration work, updated scattered-light corrections, and an increase in sample size (from 345 to 513) due to the inclusion of stars from run GO13776. This increases the parameter space coverage in $\mathrm{log}\,g$, T_eff, and [Fe/H] (Figures 1 and 2).

After correction for interstellar extinction, the spectra were used to explore the chromospheric activity of stars using the Mg ii 2800 h and k feature and Hα as likely indicators.

Against color, there is a gradual change of sign of Mg2800 from positive to negative (signifying absorption to emission transition) for both dwarfs and giants within 0.5 < (B − V)₀ < 1.5. From Figure 10 it is evident that the transition happens at (B − V)₀ = 1.0 or spectral class K3 for dwarfs, and (B − V)₀ = 1.12 or spectral class K4–K5 for giants. The color calibration of Worthey & Lee (2011) indicates that we expect dwarfs to have B − V bluer than giants by about 0.1 mag, so this crossover happens at about the same T_eff for both dwarfs and giants. Largely, this result is consistent with results from Gurzadian (1975) where it was shown that the Mg ii 2800 feature starts dominating in emission in K2 and later-type stars. The photospheric absorption gives way to strong chromospheric emission as the temperature drops. Temperature is the emphatic controlling parameter of Mg2800 emission; the cooler the star, the stronger the emission. [Fe/H] and log g have little influence on Mg2800, and we see no evidence of flare behavior.

We chart basic Hα and Hβ behavior in Figures 15 and 16, respectively. The peaks are the deep absorptions in A stars, and strongly negative values indicate that emission has overshadowed absorption. Figure 15 shows four stars with mild flares in progress: GJ 551, GJ 876, and GL 109 are dwarfs while HD 126327 is a giant. GJ 551 is Proxima Centauri and it shows up as a flaring dwarf in a 20 s cadence Transiting Exoplanet Survey Satellite monitoring campaign (Howard & MacGregor 2022). Evidence for flares in GJ 876 is reported in Froning et al. (2019). GL 109 is listed as an eruptive variable in SIMBAD and categorized as UC Cet–type flare star (Gershberg et al. 1999). HD 126327 is the only cool giant that seems to be flaring. Prominent TiO band absorption affects the coolest stars. Cool giants saturate at B − V ≈ 1.65 (Worthey & Lee 2011) but Hβ continues to increase in particular, not because of actual Hβ absorption, but because of the increasing influence of TiO features. Figure 17 shows the feature and side bands used in calculation of Hβ index strength. The hotter star (upper panel (a)) shows clear Hβ absorption. But for the cooler star (lower panel (b)), the values obtained for the Hβ strengths are mostly from TiO features. The hot dwarfs with Hα magnitudes less than −0.1 are Be stars.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The Mg ii 2800 line emission in UV is a major probe for chromospheric radiative loss (Linsky & Ayres 1978). From Figure 10 it is evident that there is scatter in the Mg ii 2800 line strength for a given temperature, but the character of that scatter might be astrophysical. Various studies have suggested the existence of a "basal" flux level for Mg ii 2800 that might indicate the level of an ongoing, persistent mechanism (acoustic waves are often cited) that can be supplemented by a more variable heating mechanism (such as magnetohydrodynamic shocks) that adds Mg emission to some stars but not others (Schrijver 1987; Strassmeier et al. 1994; Martínez et al. 2011).

Recast in terms of the Mg ii λ2800 flux emerging from the star's surface (F_λ ), the above authors found a "basal level" that increases with temperature. In order to confirm this, we select NGSL stars with T_eff < 5000 K and recast their emission-line strengths as emergent fluxes as in Martínez et al. (2011). The scheme follows Oranje et al. (1982), but is extended to account for interstellar extinction. Oranje et al. noted that

$$\begin{eqnarray}&&\displaystyle \frac{{F}_{\lambda }}{{f}_{\lambda }}=\displaystyle \frac{{F}_{\mathrm{bol}}}{{f}_{\mathrm{bol}}},\end{eqnarray} \tag{ 10 }$$

where F_λ is the star's outbound surface flux (erg cm⁻² s⁻¹) at some wavelength. For us, this wavelength is 2800 Å, and it is chromospheric in origin. The lower case f_λ is then the flux received at Earth. The right-hand side is the bolometric versions. This equation is only good in the limit of zero extinction. Extinction at wavelength λ (A_λ ) is defined by

$$\begin{eqnarray}&&{A}_{\lambda }=-2.5\mathrm{log}\ \displaystyle \frac{{f}_{\lambda }}{{f}_{0,\lambda }},\end{eqnarray} \tag{ 11 }$$

where f_0,λ is the extinction-corrected version of f_λ . This equation can be inverted to

$$\begin{eqnarray}&&{f}_{\lambda }={f}_{0,\lambda }{10}^{-0.4{A}_{\lambda }}={f}_{0,\lambda }\ g({A}_{\lambda }),\end{eqnarray} \tag{ 12 }$$

where the function g(A_λ ) is shorthand we introduce. By convention, A_λ is positive and thus f_λ is always less than f_0,λ and 0 < g(A_λ ) ≤ 1. Besides g(A_λ ), we also invent h(A_bol) to represent the extinction in bolometric quantities, which is more complicated to produce (it requires the integration of the dust-attenuated flux over all wavelengths and thus depends on the spectral type of the target star). Putting everything together,

$$\begin{eqnarray}&&{F}_{\lambda }={f}_{0,\lambda }\displaystyle \frac{{F}_{\mathrm{bol}}}{{f}_{0,\mathrm{bol}}}\displaystyle \frac{g({A}_{\lambda })}{h({A}_{\mathrm{bol}})}.\end{eqnarray} \tag{ 13 }$$

This can be rephrased in terms of T_eff by noting that:

$$\begin{eqnarray}&&{F}_{\mathrm{bol}}=\sigma {T}_{\mathrm{eff}}^{4},\end{eqnarray} \tag{ 14 }$$

where σ is the Stefan–Boltzmann constant and

$$\begin{eqnarray}&&{f}_{0,\mathrm{bol}}=B{10}^{-0.4(V+{\mathrm{BC}}_{V})},\end{eqnarray} \tag{ 15 }$$

where B is a zero-point adjustment between physical units and the astronomical magnitude scale, V is the apparent magnitude in the V band, and BC_V is the bolometric correction for the V band. The B value is obtained by noting that f_bol,⊙ = 1361 W m⁻², V_⊙ = −26.76 (Willmer 2018), and BC_V,⊙ = 0.09 (VandenBerg & Clem 2003).

Known T_eff, [Fe/H], and $\mathrm{log}\,g$ values for each star were used to interpolate a low-resolution synthetic flux from Worthey (1994). We applied a Fitzpatrick (1999) cubic spline extinction curve to this synthetic flux, then integrated (with and without extinction) to find h(A_bol). For the bolometric correction, we used the Worthey & Lee (2011) calibration, which also requires T, $\mathrm{log}\,g$, and [Fe/H]. We used these values and our Equation (8) to get A₂₈₀₀. The quantity f_0,bol was calculated by integrating the flux over the index band for Mg ii 2800. A linear pseudocontinuum calculated from the Mg ii 2800 passbands was subtracted before the integration.

Figure 18 shows the dependence of F_λ as a function of T_eff in a log–log scale. Thus transformed to surface-emergent flux, cool dwarfs are seen to emit an order of magnitude more Mg2800 flux per unit surface area, with two notable low-lying objects. As for giants, a number of cool giants have lower flux values than the basal line given by Martínez et al. (2011) (the solid green line in Figure 18). One giant (HD 222093) lies 2 orders of magnitude brighter than typical, and three stars lie off-scale on the low end. A likely explanation for the difference in the morphology of our figure versus Martínez et al.'s is our improved treatment of interstellar extinction. If we artificially set our extinctions to zero, the figure's morphology qualitatively matches that of Martinez et al. Despite our lower spectral resolution compared to IUE's, continuum subtraction is too minor to contribute significant error.

Zoom In Zoom Out Reset image size

Another giant, HD 126327, lies more than an order of magnitude lower than the line but also was caught flaring in Hα (Figure 14). This might indicate that stormy events in the photosphere and lower chromosphere might temporarily disrupt the middle chromosphere where Mg2800 arises. Figure 19 elucidates the fact that stars lying close to the green line in Figure 18 in fact have higher Mg ii λ2800 flux compared to stars lying way below the same green line in Figure 18.

Zoom In Zoom Out Reset image size

Figure 20 shows variation in the Mg ii 2800 spectral lines using observations from IUE, NGSL, and Worthey et al. (2022a) for the single star HD 102212. The observations were made in 1997, 2002, and 2021 for IUE, NGSL, and Worthey et al. (2022a), respectively. Mg2800 values for the three cases are −1.49 ± 0.05, −1.81 ± 0.003, and −2.26 ± 0.008 for IUE, NGSL, and Worthey et al. (2022a), respectively. The errors in Mg2800 values are calculated by taking into consideration the errors in flux at each pixel value and then propagating these errors while calculating Mg2800 values. Even admitting a few percent additional fluxing error, it is statistically certain that Mg2800 values show a temporal variation in HD 102212.

Zoom In Zoom Out Reset image size

Add this to HD 232078, a similar long-period variable listed in Section 5.2 that is probably also variable in Mg2800.

The Sun is known to have a ∼7% Mg2800 variation that correlates with the magnetic activity cycle (Deland & Cebula 1993). Buccino & Mauas (2008) report cyclic chromospheric activity in HD 22049 and HD 128621 using IUE spectral data. At visible wavelengths, some studies show an overall variation in chromospheric activity from Ca ii H and K lines. Baliunas et al. (1998) report that 85% of stars in the 40 yr HK Project at Mount Wilson Observatory showed either periodic (60%) or aperiodic (25%) variation in chromospheric activity. Temporal variation possibly separates magnetically driven chromospheric heating, which can be expected to be cyclic, from acoustic-wave-driven heating, which might be expected to be steadier. In this regard, HD 102212 is not an apt test case because it is a long-period variable star likely to experience considerable "activity variability" in its gaseous envelope.

We acknowledge with thanks the variable star observations from the AAVSO International Database contributed by observers worldwide and used in this research. This work is based on observations made with the NASA/ESA Hubble Space Telescope, program GO 16188, doi:10.17909/t9-d42d-z465. Support for this work was provided by NASA through grant No. HST-GO-16188.001-A from the Space Telescope Science Institute. STScI is operated by the Association of Universities for Research in Astronomy, Inc. under NASA contract NAS 5-26555. This research has made use of the SIMBAD database, operated at CDS, Strasbourg, France.