SMARTY: The mileS Moderate resolution neAr-infRared sTellar librarY

Most of the observed galaxies cannot be resolved into individual stars and are studied through their integrated spectrum using simple stellar populations (SSPs) models, with stellar libraries being a key ingredient in building them. Spectroscopic observations are increasingly being directed towards the near-infrared (NIR), where much is yet to be explored. SSPs in the NIR are still limited, and there are inconsistencies between different sets of models. One of the ways to minimize this problem is to have reliable NIR stellar libraries. The main goal of this work is to present SMARTY (mileS Moderate resolution neAr-infRared sTellar librarY) a ~0.9-2.4$\mu$m stellar spectral library composed of 31 stars observed with the Gemini Near-IR Spectrograph (GNIRS) at the 8.1m Gemini North telescope and make it available to the community. The stars were chosen from the SMARTY library, for which the atmospheric parameters are reliable (and well tested), to populate different regions of the Hertzsprung-Russell (HR) diagram. Furthermore, five of these stars have NIR spectra available that we use to assess the quality of SMARTY. The remaining 26 stars are presented for the first time in the NIR. We compared the observed SMARTY spectra with synthetic and interpolated spectra, finding a mean difference of ~20% in the equivalent widths and ~1% in the overall continuum shape in both sets of comparisons. We computed the spectrophotometric broadband magnitudes and colours and compared them with the 2MASS ones, resulting in mean differences up to 0.07 and 0.10mag in magnitudes and colours, respectively. In general, a small difference was noted between the SMARTY spectra corrected using the continuum from the interpolated and the theoretical stars.


INTRODUCTION
Galaxies in the local universe are the final product of a very long process, which depends on a combination of internal (e.g.star formation, stellar and chemical evolution) and external processes (e.g.environment).They display a wide range of properties, such as luminosities, stellar masses, gas, and dust content (e.g.Conselice 2014;Sánchez et al. 2018;Sánchez 2020;Sánchez et al. 2021;Riffel et al. 2021Riffel et al. , 2022Riffel et al. , 2023)).The determination of many of these properties relies on the correct characterisation of their stellar content, which, in turn, depends on reliable stellar population models.One of the most common ways of modelling integrated stellar populations is through ★ E-mail: michele.bertoldo@ufrgs.brspectral fitting, which can, for example, combine simple stellar populations (SSPs) in different proportions to build the composite stellar population that best describes the observed galaxy (Tinsley 1968;Fernandes et al. 2005;Walcher et al. 2011;Conroy 2013;Gomes & Papaderos 2017;Cappellari 2023).Thus, the SSPs are the most important ingredient in this type of characterisation.
To build up reliable SSP models, one needs several ingredients.Stellar libraries are one of the most fundamental ones (e.g.Worthey et al. 1992;McWilliam 1997;Sánchez-Blázquez et al. 2006;Gustafsson et al. 2008;Husser, T.-O. et al. 2013;Coelho 2014;Chen et al. 2014;Villaume et al. 2017;Knowles et al. 2021, and references therein), and can be either theoretical or empirical.Empirical libraries depend on existing observed stellar spectra, for which high signal-to-noise data can be obtained only for nearby stars.Thus, em-pirical libraries are restricted to nearby and bright objects, leading to libraries biased to metallicity and abundance ratios of stars in the solar neighbourhood.Besides, these libraries have limited coverage of atmospheric parameters and spectral resolution.Theoretical libraries, on the other hand, can encompass a wide range of parameters, including the possibility of high-resolution spectra.However, these libraries depend on our knowledge of the physics of stellar atmospheres and data of atomic and molecular opacities.Thus, both are important since they complement each other, not only in their use but in their assembly.For instance, theoretical libraries are tested and calibrated using empirical libraries (Coelho 2009;Arentsen et al. 2019;Coelho et al. 2020;Lançon et al. 2021) while empirical libraries rely on theoretical libraries for estimating the atmospheric parameters of stars.Both types of libraries have been intensely evolving in the last years and, at least in the optical, can confidently be used to reproduce the integrated spectra of stellar systems (Martins et al. 2019; see also Moura et al. 2019;Rennó et al. 2020).
Ideally, for a given set of atmospheric parameters, theoretical and empirical spectra should be equivalent; however, a complete understanding of stellar physics (as well as atmospheric, atomic, and molecular parameters) and greater observational power would be necessary for this scenario to be achieved.A way to overcome this is to use combined empirical and theoretical stellar libraries.For instance, Westera et al. (2002) used empirical data to correct the spectral energy distributions of the BaseL Stellar Library (BaSeL).Coelho (2014) used observed stellar spectra to test her theoretical stellar library.Empirical and theoretical stellar libraries can also be used to build semi-empirical stellar population synthesis models in methods called differential stellar population (Walcher et al. 2009) and flexible SPS (FSPS, Conroy et al. 2009), as well as in abundance ratio variations (e.g.Knowles et al. 2021Knowles et al. , 2023)).
Additionally, understanding the stellar populations using the NIR spectral region is now of utmost importance since the JWST is producing amazing data in this spectral region (e.g.Luhman et al. 2024;Marino et al. 2024;Boyett et al. 2024).This spectral region is interesting since it is less affected by dust extinction than the optical, allowing the study of the light content inside optically obscured regions (e.g.Riffel et al. 2006Riffel et al. , 2015;;Riffel et al. 2019, and references therein).Besides that, models have predicted that cold evolved stars dominate the NIR emission in galaxies.In particular, the thermally pulsing asymptotic giant branch (TP-AGB), phase of cold, intermediate-mass giants stars of difficult modelling, is believed to contribute up to 80% in  band luminosity for intermediate-age populations (0.2 − 2 Gyr) (Maraston et al. 2006;Salaris et al. 2014), but with a limited impact on the spectral features (Riffel et al. 2015;Röck et al. 2017;Eftekhari et al. 2021Eftekhari et al. , 2022).See, for example, Verro et al. (2022b) for a discussion on the effect of these stars in the building of SSP models.
The path towards developing robust SSP models involves comparing empirical and synthetic stellar spectral libraries across the Parameter coverage of existing empirical NIR stellar libraries in the log  vs.  eff plane.The blue markers and the violet dots are the stars from XSL (683 stars with  ∼ 10 000; Chen et al. 2014;Gonneau et al. 2020;Verro et al. 2022a) and IRTF+EIRTF (210+287 stars with  ∼ 2 000; Cushing et al. 2005;Rayner et al. 2009;Villaume et al. 2017), respectively.The stars presented in this work are indicated by the black crosses (31 stars with  ∼ 1 300).
wavelength ranges of photospheric emission.For instance, the theory of stellar physics enters all SSP models, even when this is only implicit in the association of fundamental stellar parameters with empirical spectral library stars (Lançon et al. 2021).To shed some light on our understanding of the stellar populations in the NIR spectral regions, two aspects are thus fundamental: ) expand the existing stellar libraries on the NIR since they do not completely populate the HR diagram (e.g.Cushing et al. 2005;Rayner et al. 2009;Meneses-Goytia et al. 2015;Röck et al. 2016;Villaume et al. 2017;Lançon et al. 2021) and ) to fine-tune the NIR theoretical libraries, comparing them with the empirical stellar libraries (e.g.Coelho 2014).
Aimed at helping to overcome these problems, here we present a new set of NIR stellar spectra of a sub-sample of miles stars (Sánchez-Blázquez et al. 2006;Falcón-Barroso et al. 2011), whose atmospheric parameters have been previously determined (Cenarro et al. 2007;García Pérez et al. 2021).For this purpose, we selected a sub-sample of stars from the miles stellar library, which have been used to test theoretical stellar spectra in the optical region by Coelho (2014, see § 2 for more details).For the selected miles stars, we obtained the NIR data using the Gemini Near-IR Spectrograph (GNIRS) on the Gemini North telescope, from ∼ 0.9 to ∼ 2.4 µm at a moderate spectral resolution ( ∼ 1 300).
This paper is organised as follows: in Sect.2, we describe the sample selection, our Gemini observations and data reduction.In Sect.3, we describe the processes applied to calibrate the flux in order to fine-tune the spectra quality.Finally, our last remarks are given in Sect. 4.

Sample Selection
Our sample selection was based on a systematic comparison between the miles optical observations and a grid of synthetic spectra available in Coelho (2014).This author grouped the miles stars in bins of  eff and log  with widths given by the uncertainties in these parameters.Our intention was to have from 6 to 8 stars for each log  and [Fe/H] with effective temperature varying around 100 K to help correct absorption lines of synthetic spectra.However, due to limitations of observing time, we were able to observe only a sub-sample of these stars.We selected the stars giving priority to hotter stars, which are lacking in the IRTF (Infrared Telescope Facility Spectral Library; contains 210 stars within 0.8 − 5.4 µm with medium-resolution  ∼ 2 000 observed with the cross-dispersed spectrograph SpeX from the NASA Infrared Telescope Facility on Mauna Kea, Hawaii; Cushing et al. 2005;Rayner et al. 2009) and EIRTF libraries (Extended IRTF Spectral Library; contains 287 stars observed with the SpeX within 0.7 − 2.5 µm with  ∼ 2 000; Villaume et al. 2017), and also to have a diverse distribution on the HR diagram.
The final sample comprises 31 stars listed in Tab. 1.The sample is well distributed in the atmospheric parameter space (see Fig. 1) and with data available in the optical.Of this sample, 5 stars are common with other NIR libraries (IRTF, EIRTF, and XSL, the Xshooter Spectral Library; contains 683 stars within 0.35 − 2.48 µm with moderate-to-high resolution  ∼ 10 000 observed with the Xshooter three-arm spectrograph of the Very Large Telescope on Cerro Paranal, Chile; Chen et al. 2014;Gonneau et al. 2020;Verro et al. 2022a) to be used as a control sample, and 26 are observed for the first time in the NIR.The spectra in the optical region (3 525−7 500 Å) are available in the miles library (Sánchez-Blázquez et al. 2006;Falcón-Barroso et al. 2011) with a resolution of 2.5 Å (in FWHM), while the NIR spectral range was observed with GNIRS at  ∼ 1 300.A comprehensive description of the observation process and data reduction is provided in § 2.2.

Observations and data reduction
The NIR spectra were obtained using the cross-dispersed (XD) mode of GNIRS on the 8.1 m Gemini North telescope in Mauna Kea, Hawaii.With the "long blue" camera with the LXD prism, 10 l/mm grating and 0.10 ′′ wide slit, this mode gives simultaneous spectral coverage from ∼ 0.83 − 2.5 µm at  ∼ 1 300 with a pixel scale of 0.05 ′′ /pix.To remove the sky emission, the targets have been observed in the ABBA-type pattern, with the source always on the slit.One telluric star per object was observed (either before or after the observations) to remove the telluric bands that plague the NIR spectral range.Individual and total exposure times varied depending on the object's brightness and the likely observing conditions (see below) and are given in Tab. 2.
The slit was orientated close to the mean parallactic angle during the observations of both the science target and standard star.This procedure was adopted to minimise the effects of differential atmospheric refraction, which can be important over this wide wavelength range, especially at low elevations.
Data reduction was carried out by a slightly modified version of XDGNIRS (Mason et al. 2015) pipeline, V1.9, which is available at https://xdgnirs.readthedocs.io/en/latest/.Standard CCD procedures, such as bias subtraction, flat-fielding, and wavelength calibration, were followed and implemented via customary iraf (Tody 1986) tasks.Uncertainties were estimated from electron counts due to the science targets and atmospheric emission, as well as characteristic read noise and dark current values of the detector.
The two most critical aspects of the data reduction were removing  2016) and (5)-( 6), from Sánchez-Blázquez et al. (2006).The superscript letter after the control star name identifies the library in which it is also present: a in XSL; b in IRTF; and c in EIRTF.telluric features and matching the sensitivity function between different orders.Regarding the former, the reference spectrum of telluric standards was first treated with an algorithm to remove hydrogen lines from the star's atmosphere based on a direct comparison with a high-resolution and high signal-to-noise spectrum of Vega.As for the latter, scale factors for small multiplicative corrections between the sensitivity functions of different diffraction orders were estimated by minimising the quadratic difference between overlapping regions of the spectrum.The same telluric standard star was used for flux calibration.

DATA QUALITY
Many of the smarty observations at Gemini were carried out under poor weather conditions (see Tab. 2).Despite that, we achieved a good telluric correction, but we needed to apply an independent flux calibration to our data in order to correct the 'steps' in the continuum.These irregularities likely stem from challenges to match the sensitivity function of different orders.The independent flux calibration was performed according to the following procedures.First, we normalised the smarty spectra, leading to a spectrum of pure absorption features,  norm .We then multiplied the normalised spectra by the continuum flux,  C , from a reference spectra, using:    HD 145976 and HD 205512 ( eff = 7 186, 6 927 and 4 703 K, respectively; in EIRTF).The shape of the continuum of these stars has been used as a reference to fine-tune the flux calibration of the smarty counterpart.Note that these stars have a good  eff coverage.
(ii) interpolated stars: For all smarty stars, we have done the independent flux calibration using the e-miles interpolator (see Vazdekis et al. 2003Vazdekis et al. , 2016, for more details) to interpolate among 180 IRTF plus 200 EIRTF stellar spectra to compute a spectrum which best matches the smarty stars stellar parameters adopting a local interpolation scheme 1 .Therefore, as IRTF+EIRTF do not have a significant number of stars hotter than ∼ 7 000 K, the interpolated corrected flux is not recommended for stars above this temperature since the box can be bigger than typical uncertainties in the determination of the parameters.
(iii) synthetic stars: Following Coelho et al. ( 2020), we computed synthetic spectra for the 31 smarty stars using as input the values computed by Prugniel et al. (2011) and Sharma et al. (2016) for effective temperature and surface gravity, and approximated values of metallicity ([Fe/H] = −0.1,0.0 or 0.2).The continuum of these stars has been used as a reference for the independent flux calibration. 1 The interpolator selects stars whose parameters are within a box around the requested parametric point ( eff , log , [Fe/H]), which is divided into eight boxes.If no stars are found in any of these boxes, it can be expanded up to a limit.Thus, the larger the density of stars around the point, the smaller the box is.
It is worth noting that the spectra of colder ( eff < 6 000 K) are not well predicted by the models2 .
To fit the continuum, we smoothed the spectra using LOESS3 , discarding the low flux values (i.e., the absorption features) from the spectra in each iteration by adopting different values for the lower and upper -clipping factors.This process was repeated until only the data points of the continuum were left to be fitted.To properly fit the continuum of different spectral regions, the fit was done independently within wavelength ranges, as illustrated in Fig. 2. The whole procedure was performed interactively by visually inspecting the fits and changing the fitting parameters for each spectrum to achieve a good continuum fit.The parameters that can be adjusted in our approach are the lower and upper -clipping factors, the number of -clipping iterations, and the LOESS smoothing parameter, , which corresponds to the fraction of total number of data points that are used in each local fit.
In Fig. 2, we illustrate the continuum fitting process for a hot (HD 27295,  eff ∼ 11 000 K) and a cold star (HD 205512,  eff ∼ 4 700 K) as examples; similar figures for the other smarty stars are shown in Appendix A. The same approach was adopted to fit the continuum of the reference stars to obtain  C .Before fitting the continuum, all spectra were degraded to the same resolution of  = 1 300.The initial resolution of smarty spectra, shown in Fig. 3, was obtained from the arc lamp spectra.
The continuum-corrected smarty spectra were multiplied by a factor so that the total flux within the , , and  bands ( J ,  H , and  H ) of our final spectra is consistent with that of the 2MASS photometry; i.e., the factor was chosen so that it leads to ( The line-of-sight velocities used to correct the smarty spectra were determined through cross-correlation with the theoretical spectra using the task xcsao from the iraf rvsao package (for details, see Tonry & Davis 1979;Kurtz et al. 1992;Mink & Kurtz 1998).To avoid the regions with telluric contamination, the cross-correlation was performed separately within the J, H, and K wavelength ranges, and we adopted the result with the lowest uncertainty.
The flux-corrected smarty spectra are available at www.if.ufrgs.br/~riffel/smarty/.We recommend using smarty with the continuum from the common stars when available, from interpolated stars when  eff ≲ 7500 K, and from synthetic stars when  eff ≳ 7500 K.

Comparison with literature data
To assess the uncertainties of our approach, in Fig. 4, we show the smarty spectra calibrated using  C from the common, interpolated and synthetic stars (as listed above).In this figure, we show the five stars that also have spectra available in other NIR libraries, and to evaluate possible differences in a more quantitative way, we computed the pixel deviation following the equation: where   is the flux of the smarty star (for each of the independent calibration procedures), and  is the flux of the common star (taken from the different libraries).We show the mean value of the pixel deviation ( Δ) and its standard deviation ( Δ ) for the three independent flux calibration procedures in the different windows.As can be seen, there is a very good agreement for all 5 stars, with differences up to Δ = 0.03 (for -band of HD 087822), but for almost all the cases, Δ ∼ 0.01.

Index comparisons
The equivalent widths (EW) of absorption lines are one of the most adequate ways to compare the underlying spectrum of stars.To better compare the smarty spectra with the predictions of their synthetic and interpolated versions, we have measured the EW of the ions of H i, Mg i, Fe i, Mn i, Al i, Si i, Na i, Ca i and molecular absorptions of CN and CO.For this, we have used a Python modified version of the code pacce (Riffel & Borges Vale 2011) with the definitions for the indices presented in (Riffel et al. 2019), except for the Pa  and Br  indices, where the definitions of Eftekhari et al. (2021) and Kleinmann & Hall (1986) have respectively been used.Before measuring the EW, all the spectra have been homogenised to a uniform spectral resolution of  = 1 300 (see Fig. 3).In Fig. 5, we compare the indices measured in the smarty with the predicted values of the interpolated and synthetic spectra.We show the measured value, for each index, in the smarty stars in the x-axis and the difference between the smarty stars with those measured (Δ) in the synthetic spectra (crosses) and the interpolated ones (circles) in the y-axis.The smarty measurements are colour-coded according to the star's temperature.The one-to-one relation is represented by the full line, while the standard deviation of the difference between the measurements of the smarty and synthetic stars is represented as a dashed line.
In general, there is a good agreement between the smarty and other (synthetic and interpolated) values.However, a large spread is observed among different indices.Although for some indices, the standard deviation of the differences (indicated by the dashed lines in Fig. 5) is small, measurements for individual elements in individual stars may show very large differences (see Fig. 6).For instance, the indices Mn i, H i (especially Pa ), and CN are similar for both the smarty spectra corrected by interpolated and by synthetic stars, with relative differences within 5%.The worst indices were the CO bands, Si i and Ca i, reaching up to a relative difference of up to 2 times.Also, in some cases, the difference between the synthetic and interpolated measurements is too large (e.g.Si i  15 800 Å), where the interpolated values are smaller than the synthetic ones, which can reach up to 2 Å larger than the interpolated ones for the cooler stars.
Taking into account all the indices at the same time, we found that (EW S −EW synth )/EW S = −0.2±0.84 and (EW S −EW interp )/EW S = −0.13± 0.81, indicating that, within the errors, both interpolated and synthetic spectra are in good agreement, with the interpolated EW predictions being somehow in better agreement than those obtained from the synthetic ones.
Finally, the source of these discrepancies is unclear, and addressing them is out of the scope of this study.However, we speculate that regarding the synthetic spectra, the spread may come from differences in the chemical abundance pattern (the synthetic spectra adopt solar-scaled abundances) or inaccuracies in the model opacities (see Coelho 2014;Coelho et al. 2020).In the case of the interpolated stars, the spectrum is computed by mixing observed spectra of different stars (see Vazdekis et al. 2003), thus in terms of individual elements, it may produce some deviations (e.g.differences in elemental abundances may produce different indices).

Comparison with 2MASS photometry
To check the overall accuracy of the final corrected smarty spectra4 , we compared the spectrophotometric magnitudes derived from the smarty spectra with the values available at the 2MASS point source catalogue (PSC, Skrutskie et al. 2006).
In Fig. 7, we compare smarty and 2MASS magnitudes in ,  and  bands.We can see that there is a very good agreement between smarty and 2MASS magnitudes, with maximum differences that do not exceed ∼ 0.2 mag.The mean differences in magnitude for the stars corrected by interpolated and synthetic spectra are, respectively, −0.01 and 0.0 mag for the  band; −0.02 and −0.03 mag for the  band; and 0.06 and 0.07 mag for the  band.
The colour indices obtained directly from the smarty spectra are compared with the 2MASS ones in Fig. 8, where a good agreement can be observed with the maximum differences being smaller than ∼ 0.3 mag for all colour indices.However, the smarty spectra lead to slightly lower (bluer) colour indices compared to the values from the 2MASS photometry (with mean differences in the range of −0.10 to −0.07 mag in both the  −  and  −  indices).Part of this small systematic difference might be due to Galactic extinction corrections, which are not applied to the 2MASS magnitudes, but, since we use the continuum from the reference stars, the smarty spectra are "implicitly" extinction corrected.

FINAL REMARKS
We presented a stellar spectral library with 31 stars covering the wavelength range from 0.9 to 2.4 µm observed with GNIRS at Gemini North Telescope.The smarty is publicly available at www.if.ufrgs.br/~riffel/smarty/.To ensure the spectra quality, we corrected the flux using the continuum of reference stellar spectra from three different sources: ) stars in common with other NIR empirical libraries; ) spectra obtained through interpolation of the empirical IRTF+EIRTF library; and ) theoretical spectra based on Coelho et al. (2020), extended to cover our wavelength range.
The average flux difference between smarty and the reference spectra is ≲ 2%, as can be seen in Fig. 4, where we compare the smarty spectra corrected with the continuum from the three sources mentioned above with the spectra of stars in common with other libraries.In each panel, for each set of measurements, we show the mean and standard deviation of the differences between smarty and interpolated and synthetic measurements (Δ ±  Δ ).The solid and the dashed lines represent  = 0 and  = ± Δ , where  Δ is the standard deviation of the differences between the smarty spectral indices and those from interpolated stellar spectra.
T eff (K)  We also investigated our data reliability by comparing the equivalent widths measured in the smarty spectra with those obtained from synthetic spectra computed with the atmospheric parameters of the smarty stars and interpolated from the IRTF+EIRTF stars.We find good agreement between the EW values; however, large differences can be observed for a few individual stars.
We have also compared the magnitudes and colour indices from smarty spectra with those from 2MASS photometry.The comparison reveals a very good agreement between smarty and 2MASS magnitudes, with mean differences from −0.01 to 0.07 mag and standard deviation from 0.04 to 0.07 mag in the , , and  bands.A good agreement is also observed for the colour indices, with mean differences from −0.10 to 0.03 mag for the ( − ), ( − ), and ( − ) indices.A small difference was noted between the smarty spectra corrected using the continuum from the interpolated and the theoretical stars.Comparison between the 2MASS magnitudes with those from the smarty spectra corrected using as the continuum from the interpolated (circles) and theoretical stars (crosses), colour-coded by effective temperature.Upper panels: J, H, and K magnitudes from the smarty spectra vs. 2MASS magnitudes.
The black solid and dashed lines indicate  =  and  =  ± 0.2, respectively.Bottom panels: distributions of the differences between the 2MASS and smarty magnitudes obtained with smarty spectra corrected using the interpolated (filled grey histograms) and theoretical stars (black hashed histograms).The distribution mean values and standard deviations are indicated in each panel.The solid and the dashed lines are obtained by smoothing the positions of the data points using a Gaussian kernel with a standard deviation equal to half of the standard deviation of the data points.
grant PID2021-123313NA-I00 and PID2022-140869NB-I00 from the Spanish Ministry of Science and Innovation.This publication makes use of data products from the Two Micron All Sky Survey, which is a joint project of the University of Massachusetts and the Infrared Processing and Analysis Center/California Institute of Technology, funded by the National Aeronautics and Space Administration and the National Science Foundation.

Figure 2 .Figure 3 .
Figure2.Illustration of the continuum-fit process for a hot (HD 27295,  eff ∼ 11 000 K) and a cold star (HD 205512,  eff ∼ 4 700 K).The upper plot in each panel shows the original flux (black line) and the fitted continuum (red line), and the lower plot shows the resulting normalised flux.The normalised spectra of the synthetic and interpolated stars in the lower plots are also shown as the orange and green lines, respectively.To properly fit the continuum of different spectral regions and to correct the 'steps' in the flux that could not be removed during the data reduction, the fit was done independently within wavelength ranges indicated by the vertical grey dashed lines (see text for details).The figures showing the fitted continuum for all the stars are in the Appendix A (Fig.A1).

ΔΔΔΔFigure 4 .
Figure 4. Comparison between smarty and the reference spectra from XSL, IRTF or EIRTF for the five stars in common with these libraries.The upper plot in each panel is the flux, and the lower plot shows the relative difference between the smarty and reference spectra.The dashed and dotted lines indicate 0 ± 0.05, respectively.The shaded areas are omitted due to telluric contamination.

Figure 5 .
Figure5.Differences between spectral indices measured in smarty and the interpolated (circles) and the synthetic (crosses) stellar spectra vs. values obtained for smarty.The symbols are colour-coded by effective temperature, and the error bars are shown for the measurements made in smarty spectra with the continuum from interpolated stars.In each panel, for each set of measurements, we show the mean and standard deviation of the differences between smarty and interpolated and synthetic measurements (Δ ±  Δ ).The solid and the dashed lines represent  = 0 and  = ± Δ , where  Δ is the standard deviation of the differences between the smarty spectral indices and those from interpolated stellar spectra.

Figure 6 .
Figure 6.Relative differences between smarty spectral indices and those measured in the interpolated (upper panel) and the synthetic (lower panel) stellar spectra, grouped by chemical element or molecule.In both panels, the symbols are colour-coded by effective temperature.The figure shows only stars for which EW/ EW ⩾ 2, where EW and  EW are the smarty index value and its uncertainty, respectively.The median values and interquartile ranges (IQR) of the relative differences are shown for each index.
Figure 7.Comparison between the 2MASS magnitudes with those from the smarty spectra corrected using as the continuum from the interpolated (circles) and theoretical stars (crosses), colour-coded by effective temperature.Upper panels: J, H, and K magnitudes from the smarty spectra vs. 2MASS magnitudes.The black solid and dashed lines indicate  =  and  =  ± 0.2, respectively.Bottom panels: distributions of the differences between the 2MASS and smarty magnitudes obtained with smarty spectra corrected using the interpolated (filled grey histograms) and theoretical stars (black hashed histograms).The distribution mean values and standard deviations are indicated in each panel.The solid and the dashed lines are obtained by smoothing the positions of the data points using a Gaussian kernel with a standard deviation equal to half of the standard deviation of the data points.

Figure 8 .Figure A1 .
Figure 8.Comparison between the 2MASS colour indices with those from the smarty spectra corrected using as the continuum from the interpolated (circles) and theoretical stars (crosses), colour-coded by effective temperature.Upper panels: (  −  ), (  −  ), and ( −  ) from the smarty spectra vs. 2MASS colour indices.The black solid and dashed lines indicate  =  and  =  ± 0.2, respectively.Bottom panels: distributions of the differences between the 2MASS and smarty magnitudes obtained with smarty spectra corrected using the interpolated (filled grey histograms) and theoretical stars (black hashed histograms).The distribution mean values and standard deviations are indicated in each panel.The solid and the dashed lines are obtained by smoothing the positions of the data points using a Gaussian kernel with a standard deviation equal to half of the standard deviation of the data points.

Table 1 .
Stellar atmospheric parameters and other information for the smarty stars.

Table 2 .
Observation log for the smarty stars.