Free Access
Issue
A&A
Volume 526, February 2011
Article Number A28
Number of page(s) 28
Section Catalogs and data
DOI https://doi.org/10.1051/0004-6361/201015704
Published online 16 December 2010

© ESO, 2011

1. Introduction

Type Ia supernovae (SNe Ia) as distance indicators provided the first direct evidence of the late-time acceleration of the Universe (Perlmutter et al. 1999; Riess et al. 1998). The observed faintness of high-z SNe Ia suggests there is an energy component with a negative pressure, which has been given the name dark energy. The origin and nature of the dark energy is still unknown, and precise measurements of its equation of state remain as one of the key goals for both cosmology and fundamental physics. To achieve the necessary precision using SNe Ia, large statistical samples and excellent control of potential systematic effects are imperative.

The Sloan Digital Sky Survey II (SDSS-II) Supernova Survey (York et al. 2000; Frieman et al. 2008) operated as a three-year (3 months per campaign) program (2005–2007), aimed at detecting a significant number of intermediate-redshift SNe Ia in a rolling survey. The survey provided multi-band (ugriz) photometric lightcurves for transient targets. The SN candidates detected in the galaxy subtraction pipeline were scheduled for spectroscopic observation, in order to provide spectral identification and an accurate redshift. SN Ia candidates were given highest priority for spectroscopic follow-up, however other SN types were also observed.

The first-year photometry and spectroscopy have been presented in Holtzman et al. (2008) and Zheng et al. (2008), respectively. First cosmological results, including a Hubble diagram consisting of 103 SNe Ia discovered during the first year of the survey can be found in Kessler et al. (2009) (see also Sollerman et al. 2009; Lampeitl et al. 2010).

In this article we present optical spectroscopy obtained with the ESO New Technology Telescope (NTT) and the Nordic Optical Telescope (NOT). In total, 290 spectra of SDSS-II SN candidates were obtained during 2006 and 2007. The data set contains 169 confirmed SN Ia spectra from 141 objects in the redshift range 0.03  < z  < 0.32. The redshift range complements previous and current SN Ia surveys and the spectra are, in general, of high signal-to-noise ratio (S/N).

The NTT and the NOT spectra constitute a natural subset of the complete SDSS sample since many of the observers were active at both telescopes guaranteeing a consistent observing strategy as well as similar choices of targets. Furthermore, all data were processed using the same pipeline, in a coherent manner including a detailed error analysis. The reduction procedure is described in this paper. The data set is presented as a documented library available for future studies of SNe Ia.

When our SN Ia spectra are combined with low-redshift spectroscopic samples, they cover a wide interval in cosmic time. They can then be used to search for signs of evolution of luminosity with distance, a potentially large systematic uncertainty in the use of SNe Ia as distance indicators (see e.g., Nordin et al. 2008). A statistical study of spectral features of SNe Ia could provide evidence for or against such evolution. In order to make such a study possible, correct estimates of the uncertainties are essential, especially if comparisons are to be made to a high-S/N local sample. In a companion paper, Nordin et al. (2011) present initial results of such a quantitative study. Similar studies, using data sets in other redshift ranges have previously been made (Hook et al. 2005; Blondin et al. 2006; Garavini et al. 2007; Foley et al. 2008; Bronder et al. 2008). Furthermore, spectroscopic features can potentially be used as brightness indicators (Bronder et al. 2008). Spectroscopic studies of SNe also play an important role to constrain the progenitor systems, and the explosion physics.

Other large spectroscopic samples which have been made accessible to the community include the low-redshift sample of Matheson et al. (2008) (432 spectra of 32 SNe Ia) and the high-redshift sample of Balland et al. (2009) (139 ESO/VLT spectra of 124 SNe Ia from SNLS with an average redshift of 0.63). The NTT/NOT data set complements these two well, having a mean redshift of 0.16 and a similar number of spectra as the SNLS spectra. The S/N of our data set is lower than the low-redshift sample and higher than the SNLS sample. Descriptions of other medium- and high-redshift samples have been presented in e.g. Howell et al. (2005); Foley et al. (2009); Zheng et al. (2008).

The NTT/NOT sample also contains spectra of core-collapse SNe. Some of the SNe IIP presented in this paper were used by D’Andrea et al. (2010) to investigate the method where luminosity is standardised using the ejecta velocity during plateau phase.

The paper is organised in the following manner. Section 2 describes the spectroscopic observations performed with the NTT and the NOT. In Sect. 3 the data set is presented and comparisons with other data sets are made. Section 4 describes the reduction method. The problems with differential atmospheric refraction and slit losses are addressed in Sect. 5. Section 6 deals with the host-galaxy subtraction and Sect. 7 with the typing. A couple of specific objects are briefly discussed in Sect. 8. This is followed by a summary in Sect. 9. The Appendix contains three tables describing the spectra: the observations (A.1), the typing and redshifts (A.2) and the quality of the data (A.3).

2. Observations

The NTT, located at the La Silla Observatory in Chile, was used for spectroscopic observations of SDSS-II SN candidates from September to December in 2006 and 20071. Thirty-four nights were awarded for the project, out of which 32 had sufficiently good conditions to obtain SN spectra. Through the course of these nights, 244 spectra of SN candidates were obtained. The primary mirror of the NTT has a diameter of 3.58-m. The observations were performed using the ESO Multi-Mode Instrument (EMMI; Dekker et al. 1986) in the Red Imaging and Low-Dispersion spectroscopy (RILD) mode using grism 2. Grism 2 provides a wavelength coverage from 3800 to 9200 Å. It has 300 grooves per mm, a wavelength dispersion of 1.74 Å per pixel, and a spatial resolution of 0166 per pixel, without CCD binning. During the observations a binning of 2  ×  2 was used, resulting in a resolving power R ≃ 570 at 6000 Å for a 10 slit. Slit widths of 10 or 15 were used depending on the seeing conditions.

The NOT is located at the Observatorio del Roque de los Muchachos on La Palma, Spain. Spectroscopic observations were conducted with this telescope during November 2006 and September and November 20072. Eleven nights were awarded for the project, out of which 9 nights had good enough observing conditions to obtain SN spectroscopy. During these nights 46 spectra of target SNe were obtained. The primary mirror of the NOT has a diameter of 2.56-m. Spectra were obtained with the Andalucia Faint Object Spectrograph and Camera (ALFOSC) using grism 4, which has 300 grooves per mm. This set-up provides a wavelength range from 3200 to 9100 Å with a wavelength dispersion of  ~3.0 Å per pixel and a spatial resolution of 019 per pixel. The resolving power is 710 for a 05 slit. Depending on the seeing, slit widths of 10 or 13 were used.

In general, to avoid observing erroneous objects such as asteroids and AGNs, each candidate had at least two epochs of photometry before they were placed in our spectroscopic queue. High priority was given to probable SNe Ia which appeared to have been discovered while still being on the rise. Furthermore, since several telescopes were taking spectra on the same nights, a division was made where faint targets were typically allocated to telescopes with a larger aperture. More details about the procedure for selecting the objects for spectroscopic observation can be found in Sako et al. (2008).

A typical exposure time for the spectroscopic observations was 1800 s, but depending on the magnitude of the SN and the observing conditions, exposure times were varied between 300 and 3600 s. In Table A.1 the exposure time for each individual spectrum is given.

As part of the observing programme, host-galaxy spectra of previously confirmed SNe Ia were obtained when SN candidates were lacking or the observational conditions were poor. These host galaxy spectra will not be presented in this paper.

3. Data set

The data set contains 290 spectra of 238 individual objects. Information about the spectra are given in the tables in the Appendix. Each object has been given a unique SDSS ID and the spectra have been given unique spectral numbers, SPIDs. Redshifts were obtained from Zheng et al. (in prep.). Out of the 238 targets, 84 had prior galaxy redshifts from SDSS DR7, 102 objects had the redshift determined from galaxy lines and 40 from SN features. Some of these redshifts were determined from NTT/NOT spectra, but also observations from other telescopes were used. Furthermore, there were two objects (ID 16391 and 16838) for which no reliable redshifts could be obtained from either the host galaxy or SN features, but where a redshift could be determined through a rough template fitting. For 10 objects, no redshift could be determined at all. Most of these objects are seemingly “hostless” events, several exhibiting lightcurves which differ significantly from that of SNe. The redshifts and the origin of the redshifts are listed in Table A.2. The redshift distribution of the objects is shown in Fig. 1, where the subset of objects which were classified as SNe Ia is indicated. The redshift range for all objects observed was z  =  0.016–0.487, and for the subset of SNe Ia, zSNIa  =  0.031–0.324.

thumbnail Fig. 1

Redshift distribution of the 228 observed objects at the NTT and the NOT with a known redshift. The subset of objects classified as SNe Ia is shown in the striped histogram. The bin size in the plot is 0.03.

Open with DEXTER

Our SN Ia sample contains 169 spectra of 141 individual objects, plus 3 likely SNe Ia. In addition, we also obtained 26 spectra of 23 SNe II and 12 spectra of 8 SNe Ib/c. Fourteen spectra were identified as galaxy spectra, which presumably were the result of poor seeing conditions or that the spectra were taken when the SN had faded significantly, becoming much fainter than the host galaxy. These spectra are useful as they provide spectroscopic redshift determination for the photometric sample of SNe. In Sect. 7 we describe how the spectroscopic typing was performed. The number of spectra and the number of unique objects are summarised in Table 1. For the first four years of ESSENCE, 55% of the spectroscopically observed objects were identified as SNe Ia, probable SNe Ia or core collapse supernovae (Foley et al. 2009). The corresponding number for the NTT/NOT data set is 72%. The SN fraction obtained depends on the design of the search, the redshift interval, the size of the telescope and the time spent on each supernova target. As two of several telescopes (with different sizes) involved in the follow-up of SDSS SNe, where different targets were assigned to different telescopes, the evaluation of the efficiency is further complicated.

Table 1

Number of spectra and unique objects.

The spectroscopic S/N distribution for the 290 NTT and NOT spectra is shown in Fig. 2. Values are computed for 10 Å bins and then averaged over the interval between 4000 and 6500 Å (in observed frame), which is the interval of most interest for SN features.

thumbnail Fig. 2

Distribution of the S/N for the spectra taken at the NTT and the NOT. The S/N is calculated in 10 Å bins and averaged over the interval between 4000 and 6500 Å (in observed frame).

Open with DEXTER

3.1. Lightcurve properties

SDSS multi-band photometry of the objects in our sample and their host galaxies were obtained from ugriz (Fukugita et al. 1996) observations using the method presented in Holtzman et al. (2008). Lightcurves of all confirmed SNe Ia were fitted using SALT (Guy et al. 2005). The same authors have since published a new fitter, SALT2 (Guy et al. 2007). Even though SALT2 in many respects is an improvement, we use the older code since it provides smooth lightcurves which are easy to use for interpolation, e.g. for calculating the photometry of the spectral epoch, to estimate the host galaxy contamination. The fitted parameters are: lightcurve width s (“stretch”), SALT colour c, rest-frame peak B-band magnitude, and time of B-band maximum light. However, 33 SN Ia lightcurves lack either pre- or post-maximum photometry. This leads to bigger uncertainties for the derived lightcurve parameters. In the remainder of this paper, we will refer to these as poor lightcurves. Removing all SNe Ia with poor lightcurves leaves 127 spectra of 108 different SNe Ia.

The epochs of the SN Ia spectra, calculated as the number of days in rest frame from B-band maximum light, are listed in Table A.2, and the sample distribution is shown in Fig. 3. In the table all epochs calculated from poor lightcurves are marked with a superscript “p”. The uncertainty in the epoch is obtained as either the uncertainty in peak date obtained from SALT or through error simulations, whichever is bigger. The error simulations were performed by randomising the photometry within the error bars and making new SALT fits, after which the spread in the peak date was studied. The epoch errors include the uncertainty in the photometry, but they do not include any potential systematic uncertainties in the SALT fitting procedure.

thumbnail Fig. 3

Epoch distribution of the SN Ia spectra observed at the NTT and the NOT, excluding the SNe with poor lightcurves. The un-filled histogram shows the full sample of SN Ia spectra, while the striped histogram shows the subset observed at the NOT. The epoch is defined as the number of days in rest frame since B-band maximum brightness, as obtained from the lightcurve. The 127 spectra are divided into epoch bins with a width of 3 days.

Open with DEXTER

thumbnail Fig. 4

Distribution of stretch (SALT s parameter) of the SNe Ia observed at the NTT and the NOT, excluding the SNe with poor lightcurves. The mean value is 0.95 and the standard deviation 0.11. The 108 objects are divided into bins of size 0.05. Converting the stretch to Δm15(B) following Jha et al. (2006), a mean value of 1.13 and a standard deviation of 0.22 is obtained.

Open with DEXTER

The stretch distribution, shown in Fig. 4, is similar to the distribution of the Constitution set (Hicken et al. 2009), both regarding the mean value and the shape. The Kolmogorov-Smirnov (K-S) test gives a probability of 35%, which is strong evidence against rejecting the assumption that the two samples of stretch belong to the same distribution. The distribution of colour (SALT c parameter), presented in Fig. 5, has a K-S probability of 4% when compared to the distribution of the Constitution sample of Hicken et al. (2009). We conclude that the lightcurve properties of the SNe Ia in our sample are statistically compatible with the SN lightcurves previously used in cosmological analyses.

thumbnail Fig. 5

Distribution of colour (SALT c parameter) for the SNe Ia observed at the NTT and the NOT, excluding the SNe with poor lightcurves. The mean value is 0.08 and the median 0.07. The standard deviation is 0.14. The 108 objects are divided into bins of size 0.05.

Open with DEXTER

3.2. Data archive

The spectra are publicly available in electronic format3. In addition to the calibrated spectra, a corresponding error spectrum is also provided. Versions of the spectra are available both with and without host-galaxy subtraction. Additional information about the spectra on the website; observing conditions, object type, days since maximum brightness, redshift, etc. are available in Appendix A of this paper.

4. Data reduction

The spectroscopic data were reduced using the Image Reduction and Analysis Facility (IRAF)4 and our own IDL routines. Below follows a step-by-step description of the reduction procedure.

Bias subtraction.

Bias frames were taken on every night. No spatial variations were detected so we used the CCD overscan region to subtract the bias.

Flat fielding.

The flat fields from the NTT observations had a peculiar feature at the blue end of the spectrum corresponding to wavelengths shorter than roughly 5200 Å (see the left panel of Fig. 6). This is an effect from the zeroth order image of the grism which, while being outside the CCD area, still produced a visible glow on the detector. The behaviour was not present in any of the other frames and was removed from the flat fields by fitting a surface over the region. In the right panel of Fig. 6, the corrected and normalised flat is shown. Some horizontal lines can be seen in the flat field image around 7600 Å. These are absorption lines due to water vapour and molecular oxygen, and are caused by the long light path inside the instrument. This leads to a variation of the order of 3.5% in the normalised flat5.

thumbnail Fig. 6

A normalised flat field from the NTT before (left) and after (right) the correction for the zeroth order light was applied. The figure shows only one of the two chips of the CCD. The vertical direction is the dispersion axis and the horizontal direction is the spatial axis. The scales are the same for the two images.

Open with DEXTER

For a given night of observation, one flat field was constructed for each grism/slit-width combination, which was then used to flat field the 2D spectral images of the standard stars and the SN candidates.

Spectral extraction.

The SN spectra were extracted using the optimal extraction algorithm of Horne (1986). An extraction window was chosen in the spatial direction for a narrow wavelength range and was then traced for all wavelengths. In the cases when the SN was not well separated from the host galaxy, a small spatial extraction window was used to minimise contamination from host-galaxy light. For the sky background subtraction, two small spatial ranges were defined on each side of the SN, and a fitted linear function was subtracted. For extended host galaxies, when the SN light was separated from the core of the host galaxy, this background fit also included host-galaxy light. Any residual host-galaxy light present in the spectra was removed at a later stage in the reduction process (see Sect. 6.2). When the combination of SN and host galaxy was complex, we experimented with different sizes of extraction apertures and higher orders of background fits, but without any significant improvement in the final result.

Wavelength calibration.

For wavelength calibration, spectra of a helium-argon lamp were taken at the NTT and of a helium lamp at the NOT. For each object spectrum, an arc spectrum was extracted with the same centring and trace as the object spectrum. A Chebyshev polynomial of fifth or sixth order was fitted to the identified wavelengths for the arc lamp emission lines. The solution was checked against locations of sky lines. The wavelength solution was then applied to the object spectrum.

Correction for second order contamination.

When obtaining spectra over a wide wavelength range in one single exposure, contamination from second order diffraction could lead to an erroneous flux where there is an overlap of the orders. The second order contamination can be circumvented by using a blocking filter for blue light or by ignoring the red end of the spectrum. However, this reduces the wavelength range of the spectrum. Another method is to double the exposure time and divide the observations into a blue part and a red part. However, for faint objects which require long exposure times, this is not desirable. A method has been developed by Stanishev (2007) to correct spectra, obtained at the NOT using grism 4, for second order contamination during the reduction phase. With this method the full spectrum can be used. Observations were made at the NTT using grism 2 to derive the necessary information to correct spectra from that telescope too. To be able to correct for the second order contamination we need to (1) find out the wavelength relation between the first-order spectrum and the second-order spectrum and (2) find the ratio of the efficiencies of the grism for the two orders. The wavelength overlap of the two orders were determined using observations of arc lamps, together with blocking filters. To determine the flux relation between the two orders, bright blue stars were observed with and without order-blocking filters. Szokoly et al. (2004) have developed a similar method.

In extreme cases the second order contamination can be as strong as the flux at long wavelengths. The size and shape of absorption lines at these wavelengths can also be affected. This is illustrated in Fig. 7, where two examples of second order corrected spectra are shown, one for each telescope. If the second order contamination is not removed, systematic effects could be introduced since blue SNe are more affected by the contamination than red SNe.

thumbnail Fig. 7

Second order corrections of two SNe, ID 17135 (SN 2006qm) observed at the NTT (top panel) and ID 17784 (SN 2007jg) observed at the NOT (bottom panel). The observed flux is displayed in black and the corrected flux in light grey. The spectra have been flux calibrated. No corrections due to telluric absorption, Milky Way extinction or effects from host-galaxy light have been applied to the spectra shown here. The inset in each panel shows the relative correction with respect to the observed flux.

Open with DEXTER

Flux calibration.

To correct the object spectra for the wavelength dependent sensitivity of the detector, the spectra were flux calibrated using spectrophotometric standard-stars. For the NTT, the standard stars were observed with a slit width of 50 and for the NOT with a slit of width 13. A wider slit width is used to make sure that the seeing does not affect the flux through the slit. All standard stars were observed at parallactic angle, to avoid effects from differential atmospheric diffraction. The standard-star spectra were processed in the same way as the object spectra.

Using IRAF, a sensitivity function was created for each instrument, slit width and observing session (of up to five nights). For the observations at the NOT during September and November 2007, a common sensitivity function was created since few standard star observations could be used. No significant variations with time were seen.

Corrections due to atmospheric extinction were implemented for La Silla and La Palma.

For a significant fraction of the SN observations, the parallactic angle was not used, causing differential slit losses due to atmospheric refraction. This problem will be discussed in detail in Sect. 5.

Telluric removal.

The spectra have been corrected for telluric absorption. Using standard-star observations the telluric absorptions were isolated and then combined into one telluric spectrum for each slit width and night of observations. The SN spectra were then corrected through division of a scaled telluric spectrum. The scale factor was determined as the optimal weight for removing any telluric absorption near 7600 Å. This is the dominant telluric region, while not containing any interesting SN features for objects in our redshift range. In some cases no weight could be found such that the telluric spectrum obtained during the same night could remove the telluric feature. If another telluric spectrum obtained in close connection to the SN spectrum could do the matching, this was used instead. Some spectra could not be corrected perfectly using any telluric spectrum. The increased uncertainty in the spectra due to possible un-corrected telluric absorption was estimated by calculating the deviation from a spline fit through the 7600 Å region, after correction. This error was then added to the SN error-spectrum.

Correction for dust extinction in the Milky Way.

The spectra were corrected for dust extinction in the Milky Way using RV = 3.1 and the colour excess E(B − V) from Schlegel et al. (1998). The extinction was assumed to follow the Cardelli et al. (1989) extinction law modified according to O’Donnell (1994). According to Schlegel et al. (1998), the uncertainty on E(B − V) is of the order of 16%. The errors due to the uncertainty in RV and the uncertainty in the extinction law was found to be of less importance. The wavelength-dependent errors from the Milky Way correction were added in quadrature to the existing error spectrum.

5. Differential atmospheric refraction and slit losses

Differential atmospheric refraction (DAR) occurs when an observed image is spread out with wavelength along the direction toward the horizon due to the wavelength dependence of the refraction index of air (see e.g. Filippenko 1982; Cuby et al. 1998). As a result, objects centred on the slit for one wavelength, may be partially or even fully outside the slit for shorter and longer wavelengths. The effect can be minimised by (1) observing at a low airmass and (2) rotating the slit to the parallactic angle, where the slit is parallel to the direction of the atmospheric refraction, i.e. normal to the horizon. Other factors affecting the differential slit loss are: slit width, seeing, morphology of the observed object, spectral coverage and the wavelength(s) for which the object is centred.

The spectroscopic observations of SNe presented in this paper were typically not made at parallactic angle6. Instead, the slit orientation was most often chosen to simultaneously give a spectrum of the host galaxy. The differential atmospheric refraction (in arcseconds) for a certain wavelength with respect to a reference wavelength was estimated using the formulae by Owens (1967). The reference wavelength is the wavelength with which the SN had been centred on the slit. To calculate the fraction of the flux within the slit for each wavelength we modelled the SN flux with a Gaussian distribution using the seeing at the time of observations. In Fig. 8, we show one of the observed spectra with an insignificant host-galaxy contamination and a severe differential slit loss. The spectrum is shown together with a SN Ia template from Hsiao et al. (2007) (which we from now on will refer to as Hsiao templates) in its original form as well as affected by the estimated differential slit loss. The slit loss affected template is a much better fit to the observed spectrum. The estimated slit loss at very short wavelengths is exaggerated, which seems to be true for a large fraction of the theoretically calculated slit loss functions. In most cases such a direct comparison cannot be made due to the complex interplay of host-galaxy contamination and slit loss effects. In Table A.3 we indicate the estimated effect from the differential atmospheric refraction on the spectra at an observed wavelength of 4000 Å. It should be noted that this is only a rough estimate of the differential slit loss since many simplifications are used in the calculations and the centring over the slit was not perfect. For a non-negligible fraction of the NTT/NOT spectra, the differential slit loss is severe at short wavelengths. This is due to large airmasses and large differences between the observing angle and the parallactic angle. The median value of the airmass of the observations is 1.4, with 39% of the observations having an airmass larger than 1.5. Large airmasses were unavoidable because of the position of the SDSS SN fields relative to the NTT and the NOT sites, during the observing season. Deviations from observations at the parallactic angle were also fairly large: nearly half of the spectra (46%) were observed with 45° or more from the parallactic angle. The centring of the objects on the slit was done at wavelengths above 4600 Å, typically at 6500 Å.

thumbnail Fig. 8

The observed spectrum of ID 16287 (SPID 1449) 2 days past maximum brightness is shown in black together with the Hsiao template at the same epoch in orange (upper panel). In the lower panel the Hsiao template has been multiplied with the theoretically calculated differential loss due to the atmospheric refraction.

Open with DEXTER

A polynomial multiplier is included in the host-galaxy subtraction to account for the differential slit loss effects. However, these empirically determined corrections differ from the theoretically estimated functions described above since the polynomial also accounts for reddening.

6. Host-galaxy contamination

A large fraction of the obtained SN spectra are contaminated by galaxy light. We present below an estimation of the galaxy contamination in the spectra from photometry and a description of the host-galaxy subtraction method applied to the data.

6.1. Estimating the degree of host-galaxy contamination

The host-galaxy contamination in the observed spectra was estimated using multi-band SDSS photometry. To obtain an estimate of the galaxy light in the slit, the measured photometric surface brightness of the galaxy at the position of the SN was scaled to the area defined by the slit width times the width of the spectral extraction window. The expected SN flux in the observed spectra was calculated using the SDSS lightcurve, interpolated to the night the spectrum was taken. For objects where good lightcurve fits were available, these were used for the interpolation. The amount of SN flux through the slit was estimated by modelling the SN flux as a Gaussian distribution with the width determined from the seeing. The seeing variations between the different filters were estimated following Schroeder (1987) (∝ λ-0.2). In Table A.3, the estimated contamination in the g-band (observed frame) for each observed spectrum is listed. The contamination values are rough estimates and they are more likely over estimated since we assume that all galaxies are extended over the whole extraction window. Out of the spectra, 31% have an estimated contamination less than 20%. The cases where extrapolation of the lightcurve was necessary are marked with a superscript of “e” in the table. These values should be considered less reliable.

6.2. Host-galaxy subtraction

In the past, several different techniques have been used to subtract the host-galaxy contribution from observed spectra, such as χ2 based template fitting using the spectrum of the actual host galaxy or templates, either varying the contamination level or estimating it from photometry (e.g. Howell et al. 2005; Ellis et al. 2008). For the vast majority of the NTT/NOT spectra, no good quality spectra of the host galaxy at the location of the SN were accessible at the time of the work and thus a χ2 based fitting using a real galaxy spectrum was rejected. To model the galaxy, principal component analysis (PCA) was chosen over template spectra. A similar method has also been used by Zheng et al. (2008). There are also host-galaxy subtraction methods which separate the two components, SN och galaxy, in the two dimensional spectrogram (Baumont et al. 2008). However, this requires higher resolution of the data than that of the NTT/NOT data set.

For the NTT/NOT data set a number of different methods to subtract host-galaxy light from spectra were explored. We found the most stable method to be a principal component analysis based subtraction of galaxy eigencomponent spectra. For a large fraction of the moderately contaminated spectra (10–70% contamination in the g-band) we found this method to yield good results.

It should be noted that some part of the host-galaxy light, at least for host galaxies with larger spatial extent on the CCD, were removed during a linear background fit in the reduction of the spectra. For these spectra the host-galaxy contamination that was fitted thus were the residuals from the reduction.

An additional difficulty in the host-galaxy subtraction was that a large fraction of the NTT/NOT spectra were not observed at parallactic angle (see Sect. 5), and were thus affected by a wavelength dependent flux loss.

In the PCA-based subtraction method described here, SN templates were used. A potential worry is that the usage of templates constructed from normal low-redshift SNe might affect the outcome of the subtraction. This concern and other concerns are addressed in Sect. 6.5.

PCA-based galaxy subtraction.

The spectral energy distribution (SED) of the host galaxy was estimated through minimising the difference between the observed spectrum and a combination of a SN template and a set of galaxy eigencomponent spectra. The minimisation can be described with the formula (1)where fSN is the SN template, gi the galaxy eigencomponent spectra, s is a second degree polynomial and, ai weights which are fitted in the subtraction.

thumbnail Fig. 9

The PCA-based host-galaxy subtraction for ID 16641 (SN 2006pr) at z  =  0.132. Left panel: the observed spectrum (black) is plotted together with the best fit (light grey). The galaxy template of the best fit is shown as a dotted line. The grey line shows the SN spectrum after subtraction of the estimated galaxy light. Right panel: the (host subtracted) SN spectrum is shown with and without correction for the wavelength dependent flux loss described by s (black/grey). The host galaxy contamination estimated from photometry (40%) agrees well with the fraction of host galaxy light indicated by the subtraction. The expected differential slit loss estimated following Sect. 5 (30% at 4000 Å) also seems close to what the best fit in the host-galaxy subtraction indicates (compare the curves at 4000 Å in the right panel).

Open with DEXTER

The galaxy eigencomponent spectra were created in a PCA analysis of 170 000 SDSS galaxy spectra (Yip et al. 2004). The three most dominant eigenspectra have been shown to describe 99% of all galaxies7. The fit can be made arbitrarily complex through the inclusion of more galaxy eigencomponent spectra (to the existing three), but this extension was not used in the final studies since the fits did not improve significantly while the computational demands increase drastically. In the fit we included the constraint that the total galaxy flux must be positive.

For the SN template in the fit we tried all Hsiao templates with epochs  ± 5 days from the spectral epoch as obtained from the photometric lightcurve. We also included templates of more peculiar SNe, SN 1991bg and SN 1991T (Nugent et al. 2002), to study how well the Hsiao templates worked and to, possibly, find peculiar SNe in our sample. None of our spectra were well fitted with the SN 1991bg template (there will be a bias against such objects due to their faintness). Some spectra had better fits with a SN 1991T template. The difference in the features sizes between the subtraction with the Hsiao template and the SN 1991T template was small in these cases.

To the SN template a second degree polynomial was multiplied which was introduced to account for reddening (e.g. due to host galaxy dust extinction) and differential slit loss effects. During observations, the object was centred on the slit in the red part of the spectrum, and thus differential slit losses due to atmospheric refraction (see Sect. 5), when present, predominantly affects the blue end (as would extinction). The second degree polynomial was fixed to have s ≡ 1 at the wavelength λ1 = 6600 Å and s < 1 for all other wavelengths. The value of λ1 was chosen to correspond to the wavelength where most spectra were centred on the slit. The function describing the differential slit loss is asymmetric around this wavelength, but since the fit during host-galaxy subtraction was only made between 4000 Å and 6000 Å, the behaviour of the slit loss function at longer wavelengths did not affect the fit. The polynomial was only multiplied with the SN SED, and not with that of the galaxy. The reason for this was that since galaxies are not point sources, they are less affected by differential slit losses. A separate polynomial could be modelled for the fitted galaxy, but this would be both hard to evaluate and too computationally demanding, considering the small effect. A test was performed where the same polynomial was added both to the galaxy and the SN, but this did not improve the quality of the fits.

The code for the PCA-based host-galaxy subtraction has only been applied to SNe Ia with known redshifts and a good lightcurve (photometry both before and after peak brightness). Two sample subtractions for one moderately and one highly host-galaxy contaminated spectrum can be seen in Figs. 9 and 10.

thumbnail Fig. 10

The PCA-based host-galaxy subtraction for the highly contaminated ID 12907 (SN 2006fv) at z  =  0.132. Left panel: the observed spectrum (black) is plotted together with the best fit (light grey). The galaxy template of the best fit is shown as a dotted line. The grey line shows the SN spectrum after subtraction of the estimated galaxy light. Right panel: the (host subtracted) SN spectrum is shown with and without correction for the wavelength dependent flux loss described by s (black/grey). The host galaxy contamination estimated from photometry (80%) agrees well with the fraction of host galaxy light indicated by the subtraction. The expected differential slit loss, see Sect. 5 (30% at 4000 Å) also seems compatible to what the best fit in the host-galaxy subtraction indicates (compare the curves at 4000 Å in the right panel).

Open with DEXTER

thumbnail Fig. 11

The difference between the g-band host-galaxy contamination that was introduced in the construction of fake spectra ( ) and the contamination that was obtained during host-galaxy subtraction ( ) for three different bins in  . The mean contamination for each bin is given in the respective panel as  ⟨ c ⟩ . The contamination was calculated as the percentage of the total flux in the observed SDSS g band which comes from the galaxy. The solid histogram corresponds to simulations without added noise while the dashed line corresponds to where noise is added. The dispersion is largest for small input contaminations. The number of outliers increase from 0.9% to 2% when noise is added.

Open with DEXTER

6.3. Host-galaxy simulations

Set-up.

To evaluate the subtraction methods, a large number of simulated “fake” spectra were constructed and run through the subtraction pipelines. Each fake spectrum was constructed through a combination of a SN and a galaxy spectrum, redshifted to some distance. Reddening and differential slit loss were added to the SN spectrum. Finally, noise was added. All parameters (redshift, spectral epoch, host-galaxy contamination, reddening, differential slit loss) were drawn from distributions that match the NTT/NOT data set. In constructing these fake spectra we avoided all templates/functions that were used in the host-galaxy subtraction pipeline (as this would make the fit trivial). More details about the different components of the fake spectra are given below:

  • Supernova spectra. All SN spectra that were used to con-struct fake spectra have high S/N andlow host-galaxy contamination. Their epochs are sim-ilar to the ones of the NTT/NOT spectra.Seven different spectra of normal SNe Iawere used: six of SN 2003du(epochs –6, –2, 4, 9, 10, 17)(Stanishev et al. 2007) andone of SN 1998aq at peak brightness (Branchet al. 2003). Furthermore,two spectra of the sub-luminous SN 1999by(epochs –5 and 3) (Garnavichet al. 2004) were used andtwo of the peculiar and luminous SN 1999aa(epochs –5 and 0) (Garaviniet al. 2004).

  • Reddening. Reddening was added to the SN spectrum using the Cardelli et al. (1989) extinction law with a total-to-selective extinction ratio RV of 2.1 and a colour excess E(B − V) drawn from the distribution of E(B − V) obtained from the NTT/NOT lightcurve fits. The lower value of RV compared to the Milky Way average was chosen to agree with the values often derived when looking at SNe Ia (see e.g. Altavilla et al. 2004; Nobili & Goobar 2008).

  • Galaxy spectra. In a first simulation series, four galaxy templates of varying type (elliptical, S0, Sa and Sb) from Kinney et al. (1996) were used together with three real galaxy spectra (host galaxy spectra for ID 7527, 13840 and 15381) observed at the NTT at the same time as the SN spectra presented here. The contamination level was randomly chosen between 0 and 70% for the g band. These simulations were later extended in a second series, where 50 randomly chosen SDSS galaxy spectra were used. Figures displayed here are based on the first run, but results are very similar when including the second set of galaxy spectra.

  • Redshifts. The object redshift was randomly drawn from the NTT/NOT redshift distribution.

  • Differential slit loss. To the SN spectra we multiplied functions modelling the expected differential slit loss due to atmospheric refraction for typical NTT/NOT observing conditions (see Sect. 5). These effects range from insignificant to severe. The slit loss functions are applied to the SN spectra but not to the galaxy spectra.

  • Noise. An S/N value was randomly chosen from the NTT/NOT spectral S/N distribution. Poisson noise was then added to the spectrum until the target S/N was achieved. The shape of the Poisson noise was determined as a linear combination of the input spectrum and a randomly chosen NTT/NOT sky spectrum. The linear combination was determined such that the highest S/N value in the NTT/NOT sample corresponded to no contribution from sky noise, the lowest S/N corresponded to complete dominance by sky noise and intermediate values to a combination of the two error sources. Simulations were done both with and without noise. As can be expected, added noise increased the dispersion around the true solution, but no significant bias effects were detected.

For the simulations without noise, 5000 fake spectra were constructed and about twice as many for the simulations with noise.

Evaluation of the host-galaxy simulations.

The simulations can be evaluated using a range of different tests. For all fake spectra we compared the input SN and galaxy spectra with the SN and galaxy as estimated by the host-galaxy subtraction pipeline. In Nordin et al. (2011) we study how spectral indicators are affected by the host-galaxy subtraction as a function of contamination level. A more direct way of evaluating subtractions is to compare the input contamination with what is obtained from the subtraction output. In Fig. 11 we show how the difference between input and fitted output contamination varies with input contamination (in the g band). The dispersion is largest for low contamination levels. When studying the contribution from different SN templates, it is found that for contamination levels roughly below 20%, some SN templates yield a small offset. This offset has its origin in a combination of two effects: first, some of the SN templates most likely contain some host galaxy light. This would be seen as a slight bias for low contamination levels, but should thus not be interpreted as a bias of the host subtraction method. Second, a bias can occur if the SN spectrum is imperfectly matched to the Hsiao template. The offset decreases with contamination level. For objects with very low contamination it can thus be advantageous not to subtract the estimated host galaxy light since the correction likely is smaller than the uncertainty in the correction. No trend with galaxy template or slit loss function could be seen. When noise is added to the fake spectra, the number of failed subtractions increase, from 0.9% to 2%. These failures are usually evident through a visual inspection of the subtraction. The percentage agrees with what is seen when evaluating subtractions of real SNe. These events likely correspond to random noise mimicking host galaxy or SN features.

6.4. Alternative methods

Several other host-galaxy subtraction methods were tried: varying the galaxy eigenspectra (their number and source), varying the limitations of the eigenspectra, using a large set of SN templates and using SDSS photometry to fix the fraction of galaxy light as well as the proportions of different galaxy eigenspectra. These alternative methods were found to be either less stable than the method above or equivalent (but involving more computational effort). It should be noted that for some individual SNe Ia, an alternative method might perform better. Alternative host-galaxy subtractions of spectra can be provided upon request.

6.5. Discussion on the host-galaxy subtraction

The host-subtraction method is not successful for all NTT/NOT SNe, especially for very faint or contaminated cases. It is, however, not obvious how to judge when a subtraction does fail. Both visual inspection and χ2 tests only indicate how well the subtracted spectrum match the local template, and can thus not be trusted to completely specify which subtractions succeed.

Some of the main worries with the subtraction technique used here are: (i) the observed SNe could differ from the local SN templates used e.g. if it is a peculiar type; (ii) the true galaxy contribution to the spectrum could be badly described by the use of galaxy eigencomponent spectra since the complete galaxy is not recorded in the slit; and (iii) the approximation of slit loss and reddening with a polynomial (applied to the SN but not to the galaxy SED) is a simplification. Regarding the first point, object typing was also performed prior to galaxy subtraction, which means that any truly odd SN would have been spotted at this stage unless it is heavily contaminated by galaxy light. Furthermore, the SN template is only used in the choice of what galaxy spectrum to remove and the number of bins used in the fit is large compared to the number of parameters which are fitted and thus the risk of affecting the spectrum severely is small. Regarding the second point, the three most dominant eigenspectra have been shown to be able to describe 99% of all galaxies (Yip et al. 2004), which means that they should be sufficient to describe our host galaxies as well. The third point, as well as the host-galaxy subtraction code in general, was tested through an extensive number of simulations as described in Sect. 6.3. The host-galaxy subtraction does fail in roughly one percent of all runs, but these cases usually fail completely, yielding unphysical results, and are thus easily detected.

In Nordin et al. (2011) we further test the host-galaxy subtraction code, by examining the effects on spectral features, in particular on pseudo-equivalent widths. We also compare the trends of pseudo-equivalent widths for the full sample with the subsample of spectra with low host-galaxy contamination and conclude that no bias is introduced.

7. Typing

The object spectra have been typed using SNID (SuperNova IDentification; Blondin & Tonry 2007), version 5.0. This algorithm cross-correlates the unknown spectrum against a set of template spectra. The template database includes SN spectra of different types and ages as well as non-SN spectra of galaxies, AGNs and stars. The redshift of the object can either be varied or fixed. In this version of the code, SNID distinguishes between the following types and subtypes (within parenthesis): Ia (Ia-norm, Ia-91T, Ia-91bg, Ia-csm, Ia-pec), Ib (Ib-norm, Ib-pec, IIb), Ic (Ic-norm, Ic-pec, Ic-broad), II (IIP, II-pec, IIn, IIL), Not-SN (AGN, Gal, LBV, M-star). The Ia-pec group consists of SNe of the type SN 2000cx and SN 2002cx. As a figure of merit for the classification we have used the rlap parameter (see Blondin & Tonry 2007, for a definition). A good correlation was considered to be obtained if rlap was greater or equal to five. Following Foley et al. (2009) we made four runs with SNID to determine the type, subtype, redshift and age, one at a time. This was done in the following manner:

  • First we attempted to determine the type. A type was identified if more than 50% of the templates with a good correlation belonged to this type and the best-match SN-template (highest rlap value) had the same type. When the redshift of the SN was known, we restricted the redshift range of SNID to this redshift (± 0.02).

  • If it was possible to determine the type, we continued with trying to determine the subtype. We fixed the type and the redshift (if it was known). A subtype was considered to be identified if more than 50% of the templates with a good correlation belonged to a specific subtype and the best-match spectrum was of the same subtype.

  • Regardless if a subtype was identified or not, we determined the best-fit redshift. We now locked SNID to the templates of the identified type or subtype and fitted for the redshift. The SNID redshift was defined as the median of the redshifts belonging to templates with a good correlation. The error was given as the standard deviation.

  • To determine the age, the type was fixed to the identified type or subtype. If we had a spectroscopic redshift, we fixed the redshift in SNID to this value. Otherwise we used the value obtained in the step before with a redshift error of 0.02. The SNID age was then defined as the median of all ages belonging to templates with a good correlation. The error of the age was given as the standard deviation of these values.

If the determined type and redshift were inconsistent in that the preferred type was an M-star or an LBV, while the redshift was too high for such an object to be detected, the type was set to unknown.

A problem with this method is that the composition of the template database affects the possibilities to detect different types. The SNID database has few spectra of core-collapse SNe and peculiar SNe Ia which makes these objects harder to type. It is also less effective for low S/N spectra than for those with high S/N.

The outcome of the SNID analysis is presented in Table A.2. The typing for each spectrum is presented in the column NTT/NOT type. When the type was determined using the host-galaxy subtracted spectrum, this is marked in the column for notes with the addition of an s. In some cases a spectrum only had a good match using SNID with one particular object in the database. In these cases we have marked this in the table. When no type could be determined this is marked with “–”.

The consistency in typing when several spectra of the same object existed was good. If a type could be determined, the same type was always obtained. In some cases, one of the spectra was left without a type due to, for example, noise making the classification difficult or it was classified as a galaxy, due to the faintness of the SN at that epoch.

After a visual inspection of the spectra, the typing from SNID was changed for 36 spectra. The majority of these, 20, could not be typed by SNID but could be clearly typed by visual inspection. There was also a group of spectra, 11, which had been typed by SNID, but with only few well matched template spectra, and, in the visual study, they were found to be of such low quality that it was not possible to say with certainty what the type was. Three spectra had a type which was changed, one from a Ib to a Ib/c and two from a Ia to a possible Ia. There were also two cases where the type was kept, but the subtype was removed. The spectra whose type was changed following the visual check are marked with a v in the column for notes in the table.

The NTT/NOT objects also have an SDSS typing based on the NTT and the NOT spectra in combination with spectra taken at other telescopes as a part of the SDSS-II SN Survey. This type is presented in Table A.2 under the column SDSS type. Due to the faintness of the SN in some spectra, not all NTT and NOT spectra could be typed and some were typed as galaxies, while the overall SDSS typing is a SN, based on spectra at other epochs from other telescopes.

In Fig. 12 the redshifts obtained from SNID are compared with the SDSS object redshifts from Zheng et al. (in prep.), which mainly come from the SDSS DR7 catalogue and measurements of galaxy lines in spectra obtained as a part of SDSS-II. Some are determined from SN features (see Sect. 3). There is a good agreement between the SNID redshifts and the SDSS redshifts, with a dispersion of 0.005 and a negligible bias. No dependence with redshift is detected.

thumbnail Fig. 12

A comparison of the SDSS object redshifts with the redshifts determined using SNID. The vertical error bars show the uncertainty in the SNID redshifts. The horizontal error bars show the uncertainty in the SDSS object redshifts. In most cases these are too small to be visible. All redshifts determined from SN features have an error of at least 0.005. The rms of the distribution is 0.005 (with a negligible bias). SNID redshifts which are calculated using less than five well matched spectra are marked with green triangles.

Open with DEXTER

In Fig. 13, the ages of the spectra as estimated with SNID are compared with the ages obtained from the lightcurves, (MJDspec − MJDmax)/(1 + z), for the spectra which we have classified as SNe Ia. A dispersion of 4 days was obtained. It should be noted that for young SNe, SNID generally estimates an older age compared to the SALT lightcurve fit and that the errors in the age from SNID are typically overestimated, especially at later epochs. This was also pointed out by Blondin & Tonry (2007). Since the estimated age error bars are symmetric (estimated as the standard deviation) and in many cases overestimed, there will be some cases where the minimum allowed age (from the error bar) will be unphyically low. We have not used any prior on the age from the lightcurve, which cause the large error bars and the bias in the epoch estimates at early epochs. It should be noted that the error bars, in a few cases, also could be underestimated. This occurs when the observed spectrum is only well fitted with few template SNe and these template epochs do not sample all different ages of a SN. One example is ID 16838 which is only well fit by spectra in the library of SN 1998S. For one spectrum (ID 20142), no error was estimated since only one template spectrum could fit the observed spectrum. We also test the hypothesis that the evolution of spectral features scale with the stretch factor, by multiplying the age obtained from SNID with s. The bias is slightly improved, while the spread increases.

thumbnail Fig. 13

A comparison of the age obtained using SNID and the age obtained from the lightcurve fit using SALT. The age is here defined as the number of days in rest frame since B-band maximum. SNID ages which are calculated using less than five well matched spectra are marked with green triangles.

Open with DEXTER

8. Some special objects

Possible non-normal SNe Ia.

Using SNID we find that the spectrum of ID 16333 (SN 2006on) is only well described by spectra of SN 1991T. The fit with SN 1991T at 5 days before maximum brightness is shown in Fig. 14. The peak brightness of the SN is well constrained by the lightcurve fit (the stretch and colour less so) and we find a peak B absolute magnitude of –19.06. The obtained stretch was 0.94 and the SALT c colour 0.19. Correcting the peak magnitude for stretch and colour, following MB + α(s − 1) − βc, gives a peak magnitude of –19.6, which is consistent with a SN 1991T-like SN.

thumbnail Fig. 14

SNID fit for ID 16333 (SN 2006on) with a SN 1991T template at 5 days before lightcurve maximum.

Open with DEXTER

The spectrum of ID 17176 (SN 2007ie) has strong indications of being a SN 2002cx-like object. Using SNID, the spectrum is best fit with a spectrum of SN 2002cx. However, also other classes of SNe Ia give sufficiently good fits to the spectrum, when allowing for a variation of 0.02 in redshift. In Fig. 15, some of the SNID fits with different template spectra are shown. The spectrum of SN 2002cx is a better fit to the double peaks at an observed wavelength of 5100 Å and 6100 Å than the other spectra (normal SN Ia, SN 1991T and SN 1991bg). Furthermore, the fit with the spectrum of SN 2002cx gives a better match with the redshift of the SN as obtained from SDSS DR7. If we reduce the allowed redshift range in SNID to  ± 0.01, SNID identifies the spectrum as a SN 2002cx-like object. By comparing the spectrum with the ones by Sahu et al. (2008) of SN 2005hk, which also is a SN 2002cx-like SN, we identify absorption by Fe ii λ5018, Fe ii λ5535, Fe ii λ6149 and Fe ii λ6247. The velocities in our spectrum for these lines are lower by 7–35% compared to a spectrum of SN 2005hk at 24 days past peak brightness. Compared to the spectrum at 38 days past peak, the velocities varies from being lower by 26% to being higher by 9%. The lightcurve, in this case, is not well constrained due to lack of photometry before maximum brightness. The best-fit absolute peak magnitude in the B-band from SALT is –18.2, which is comparable to what is expected for SN 2002cx-like objects.

thumbnail Fig. 15

The observed spectrum of ID 17176 (SN 2007ie) in black together with different template spectra as indicated in the legend. In the fit, the redshift has been fixed to 0.0935, the redshift of the object as obtained from SDSS DR7, with an allowed variation of 0.02. The fit with the spectrum of SN 2002cx gives the best match and the closest redshift compared to the one of the host galaxy.

Open with DEXTER

ID 19149 (SN 2007ni) is best fit with a SN 1991T-like spectrum according to SNID. It also has great similarities with SN 2002cx-like and SN 2000cx-like spectra, but not with spectra of normal SNe Ia. Comparisons from SNID with different templates are shown in Fig. 16. The lightcurve fit is well constrained and the obtained absolute peak magnitude in B is –19.47, which is somewhat brighter than a normal SN Ia. We could thus exclude an SN 2002cx interpretation. The stretch and SALT c colour are 1.04 and 0.09, respectively.

ID 20978 (SN 2007rl) is best fit in SNID with a spectrum of SN 2000cx. However, due to its high redshift, it is well fitted by many different templates. The peak brightness is well constrained by the photometry, while the stretch and colour are less so due to few observations past peak brightness. The absolute peak B band magnitude is –19.6.

Possible SNe IIn.

In the spectral fitting we find that the spectrum of ID 16838 is best fit with a SN IIn, similar to SN 1998S. Another potential SN IIn is ID 16668 (SN 2006pu).

Stellar tidal disruption event.

The object with SDSS identification 17237 has been suggested to be a stellar tidal disruption event (van Velzen et al. 2010), an event where a star passes close to a supermassive black hole in the centre of the galaxy without being completely destroyed.

9. Summary

We have presented observations and reductions of 290 spectra observed at the NTT and the NOT as a part of a programme to classify SNe discovered by the SDSS-II. The selection procedure, the observing strategy and the reductions have been performed in a coherent and well documented manner. The spectra have been corrected for telluric absorption. Objects classified as SNe Ia were processed through a host-galaxy subtraction pipeline. The host galaxy SEDs were estimated using a PCA-analysis where the difference between the observed spectrum and a combination of a SN template and galaxy eigencomponent spectra were minimised. A polynomial was multiplied to the SN templates to account for reddening effects as well as differential slit loss due to atmospheric refraction. The host-galaxy subtraction pipeline was evaluated using extensive simulations mimicking the distribution of parameters for the NTT/NOT data set. Furthermore, careful estimates of uncertainties have been made, thus making our data well-suited for systematic studies.

thumbnail Fig. 16

The SNID fit of ID 19149 (SN 2007ni) with the observed spectrum in black and different template spectra in colour as indicated in the legend.

Open with DEXTER

The spectra were classified using SNID, a template-matching algorithm. The age of the spectra estimated using SNID was compared to the age obtained from the lightcurve, in the case of SNe Ia, and was found to agree well, with an age dependent bias of a few days and an overall dispersion of 4 days. The typing was revised using visual inspection and changed in some cases; typically when the SNID classification was unreliable due to few matched spectra or when the spectrum could be typed by visual inspection but not with SNID.

In total, 207 spectra of 172 different confirmed SNe were obtained at the NTT and the NOT for the SDSS-II SN program. Out of these, 169 are SN Ia spectra, and some of these (the ones which will survive the strict cosmology lightcurve cuts) will be part of a larger sample to be included in the three year SDSS-II SN Hubble diagram. Four potential peculiar SNe Ia were identified from their spectra. ID 16333 (SN 2006on) and 19149 (SN 2007ni) are potential SN 1991T-like objects, while ID 17176 (SN 2007ie) is a potential SN 2002cx object. ID 20978 (SN 2007rl) is identified as a peculiar SN Ia using SNID, but due to its high redshift, it is difficult to determine a subtype. The sample also includes 26 spectra of 23 SNe II and 12 spectra of 8 SNe Ib/c.

The reduced spectra are made public together with the estimated uncertainty for each wavelength. For the subsample of SNe Ia, host-galaxy subtracted spectra are provided. All this data is accessible from the webpage: http://www.physto.se/~linda/spectra/nttnot.html.


1

The observations were acquired in the ESO programmes 077.A-0437, 078.A-0325, 079.A-0715 and 080.A-0024 under PI Robert Nichol.

2

The observations were acquired in the programmes with proposal numbers 34-004, 35-023 and 36-010 under PI Maximilian Stritzinger.

4

IRAF is distributed by the National Optical Astronomy Observatories, which are operated by the Association of Universities for Research in Astronomy, Inc., under cooperative agreement with the National Science Foundation.

5

See the documentation about EMMI www.eso.org/sci/facilities/lasilla/instruments/emmi

6

The observations of the standard stars were always made at parallactic angle.

7

Galaxy emission lines can have errors of up to 10% when using only three eigencomponent spectra (Yip et al. 2004). However, for our purposes it is enough with a well described continuum.

Acknowledgments

The authors would like to thank Johan Fynbo, Christa Gall and Christina Henriksson who all helped out at the NOT observations. L.Ö is partially supported by the Spanish Ministry of Science and Innovation (MICINN) through the Consolider Ingenio-2010 program, under project CSD2007-00060 “Physics of the Accelerating Universe (PAU)”. V.S is financially supported by FCT Portugal under program Ciência 2008. The Oskar Klein Centre is funded by the Swedish Research Council. The Dark Cosmology Centre is funded by the Danish National Research Foundation. Funding for the SDSS and SDSS-II has been provided by the Alfred P. Sloan Foundation, the Participating Institutions, the National Science Foundation, the US Department of Energy, the National Aeronautics and Space Administration, the Japanese Monbukagakusho, the Max Planck Society, and the Higher Education Funding Council for England. The SDSS Web Site is http://www.sdss.org/. The SDSS is managed by the Astrophysical Research Consortium for the Participating Institutions. The Participating Institutions are the American Museum of Natural History, Astrophysical Institute Potsdam, University of Basel, University of Cambridge, Case Western Reserve University, University of Chicago, Drexel University, Fermilab, the Institute for Advanced Study, the Japan Participation Group, Johns Hopkins University, the Joint Institute for Nuclear Astrophysics, the Kavli Institute for Particle Astrophysics and Cosmology, the Korean Scientist Group, the Chinese Academy of Sciences (LAMOST), Los Alamos National Laboratory, the Max-Planck-Institute for Astronomy (MPIA), the Max-Planck-Institute for Astrophysics (MPA), New Mexico State University, Ohio State University, University of Pittsburgh, University of Portsmouth, Princeton University, the United States Naval Observatory, and the University of Washington. The paper is partly based on observations made with the Nordic Optical Telescope, operated on the island of La Palma jointly by Denmark, Finland, Iceland, Norway, and Sweden, in the Spanish Observatorio del Roque de los Muchachos of the Instituto de Astrofisica de Canarias. The data have been taken using ALFOSC, which is owned by the Instituto de Astrofisica de Andalucia (IAA) and operated at the Nordic Optical Telescope under agreement between IAA and the NBI. The paper is partly based on observations collected at the New Technology Telescope, operated by the European Organisation for Astronomical Research in the Southern Hemisphere, Chile.

References

Appendix A: Tables

Table A.1

Information about the observations.

Table A.2

Spectroscopic typing and redshift determination.

Table A.3

Spectral quality.

All Tables

Table 1

Number of spectra and unique objects.

Table A.1

Information about the observations.

Table A.2

Spectroscopic typing and redshift determination.

Table A.3

Spectral quality.

All Figures

thumbnail Fig. 1

Redshift distribution of the 228 observed objects at the NTT and the NOT with a known redshift. The subset of objects classified as SNe Ia is shown in the striped histogram. The bin size in the plot is 0.03.

Open with DEXTER
In the text
thumbnail Fig. 2

Distribution of the S/N for the spectra taken at the NTT and the NOT. The S/N is calculated in 10 Å bins and averaged over the interval between 4000 and 6500 Å (in observed frame).

Open with DEXTER
In the text
thumbnail Fig. 3

Epoch distribution of the SN Ia spectra observed at the NTT and the NOT, excluding the SNe with poor lightcurves. The un-filled histogram shows the full sample of SN Ia spectra, while the striped histogram shows the subset observed at the NOT. The epoch is defined as the number of days in rest frame since B-band maximum brightness, as obtained from the lightcurve. The 127 spectra are divided into epoch bins with a width of 3 days.

Open with DEXTER
In the text
thumbnail Fig. 4

Distribution of stretch (SALT s parameter) of the SNe Ia observed at the NTT and the NOT, excluding the SNe with poor lightcurves. The mean value is 0.95 and the standard deviation 0.11. The 108 objects are divided into bins of size 0.05. Converting the stretch to Δm15(B) following Jha et al. (2006), a mean value of 1.13 and a standard deviation of 0.22 is obtained.

Open with DEXTER
In the text
thumbnail Fig. 5

Distribution of colour (SALT c parameter) for the SNe Ia observed at the NTT and the NOT, excluding the SNe with poor lightcurves. The mean value is 0.08 and the median 0.07. The standard deviation is 0.14. The 108 objects are divided into bins of size 0.05.

Open with DEXTER
In the text
thumbnail Fig. 6

A normalised flat field from the NTT before (left) and after (right) the correction for the zeroth order light was applied. The figure shows only one of the two chips of the CCD. The vertical direction is the dispersion axis and the horizontal direction is the spatial axis. The scales are the same for the two images.

Open with DEXTER
In the text
thumbnail Fig. 7

Second order corrections of two SNe, ID 17135 (SN 2006qm) observed at the NTT (top panel) and ID 17784 (SN 2007jg) observed at the NOT (bottom panel). The observed flux is displayed in black and the corrected flux in light grey. The spectra have been flux calibrated. No corrections due to telluric absorption, Milky Way extinction or effects from host-galaxy light have been applied to the spectra shown here. The inset in each panel shows the relative correction with respect to the observed flux.

Open with DEXTER
In the text
thumbnail Fig. 8

The observed spectrum of ID 16287 (SPID 1449) 2 days past maximum brightness is shown in black together with the Hsiao template at the same epoch in orange (upper panel). In the lower panel the Hsiao template has been multiplied with the theoretically calculated differential loss due to the atmospheric refraction.

Open with DEXTER
In the text
thumbnail Fig. 9

The PCA-based host-galaxy subtraction for ID 16641 (SN 2006pr) at z  =  0.132. Left panel: the observed spectrum (black) is plotted together with the best fit (light grey). The galaxy template of the best fit is shown as a dotted line. The grey line shows the SN spectrum after subtraction of the estimated galaxy light. Right panel: the (host subtracted) SN spectrum is shown with and without correction for the wavelength dependent flux loss described by s (black/grey). The host galaxy contamination estimated from photometry (40%) agrees well with the fraction of host galaxy light indicated by the subtraction. The expected differential slit loss estimated following Sect. 5 (30% at 4000 Å) also seems close to what the best fit in the host-galaxy subtraction indicates (compare the curves at 4000 Å in the right panel).

Open with DEXTER
In the text
thumbnail Fig. 10

The PCA-based host-galaxy subtraction for the highly contaminated ID 12907 (SN 2006fv) at z  =  0.132. Left panel: the observed spectrum (black) is plotted together with the best fit (light grey). The galaxy template of the best fit is shown as a dotted line. The grey line shows the SN spectrum after subtraction of the estimated galaxy light. Right panel: the (host subtracted) SN spectrum is shown with and without correction for the wavelength dependent flux loss described by s (black/grey). The host galaxy contamination estimated from photometry (80%) agrees well with the fraction of host galaxy light indicated by the subtraction. The expected differential slit loss, see Sect. 5 (30% at 4000 Å) also seems compatible to what the best fit in the host-galaxy subtraction indicates (compare the curves at 4000 Å in the right panel).

Open with DEXTER
In the text
thumbnail Fig. 11

The difference between the g-band host-galaxy contamination that was introduced in the construction of fake spectra ( ) and the contamination that was obtained during host-galaxy subtraction ( ) for three different bins in  . The mean contamination for each bin is given in the respective panel as  ⟨ c ⟩ . The contamination was calculated as the percentage of the total flux in the observed SDSS g band which comes from the galaxy. The solid histogram corresponds to simulations without added noise while the dashed line corresponds to where noise is added. The dispersion is largest for small input contaminations. The number of outliers increase from 0.9% to 2% when noise is added.

Open with DEXTER
In the text
thumbnail Fig. 12

A comparison of the SDSS object redshifts with the redshifts determined using SNID. The vertical error bars show the uncertainty in the SNID redshifts. The horizontal error bars show the uncertainty in the SDSS object redshifts. In most cases these are too small to be visible. All redshifts determined from SN features have an error of at least 0.005. The rms of the distribution is 0.005 (with a negligible bias). SNID redshifts which are calculated using less than five well matched spectra are marked with green triangles.

Open with DEXTER
In the text
thumbnail Fig. 13

A comparison of the age obtained using SNID and the age obtained from the lightcurve fit using SALT. The age is here defined as the number of days in rest frame since B-band maximum. SNID ages which are calculated using less than five well matched spectra are marked with green triangles.

Open with DEXTER
In the text
thumbnail Fig. 14

SNID fit for ID 16333 (SN 2006on) with a SN 1991T template at 5 days before lightcurve maximum.

Open with DEXTER
In the text
thumbnail Fig. 15

The observed spectrum of ID 17176 (SN 2007ie) in black together with different template spectra as indicated in the legend. In the fit, the redshift has been fixed to 0.0935, the redshift of the object as obtained from SDSS DR7, with an allowed variation of 0.02. The fit with the spectrum of SN 2002cx gives the best match and the closest redshift compared to the one of the host galaxy.

Open with DEXTER
In the text
thumbnail Fig. 16

The SNID fit of ID 19149 (SN 2007ni) with the observed spectrum in black and different template spectra in colour as indicated in the legend.

Open with DEXTER
In the text

Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.

Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.

Initial download of the metrics may take a while.