Abstract
We present the largest Cepheid sample in M31 based on the complete Pan-STARRS1 survey of Andromeda (PAndromeda) in the rP1, iP1, and gP1 bands. We find 2686 Cepheids with 1662 fundamental-mode Cepheids, 307 first-overtone Cepheids, 278 type II Cepheids, and 439 Cepheids with undetermined Cepheid type. Using the method developed by Kodric et al., we identify Cepheids by using a three-dimensional parameter space of Fourier parameters of the Cepheid light curves combined with a color cut and other selection criteria. This is an unbiased approach to identify Cepheids and results in a homogeneous Cepheid sample. The period–luminosity relations obtained for our sample have smaller dispersions than in our previous work. We find a broken slope that we previously observed with HST data in Kodric et al., albeit with a lower significance.
Export citation and abstract BibTeX RIS
1. Introduction
Cepheids are standard candles, and their period–luminosity relation (PLR), sometimes referred to as the Leavitt law (Leavitt 1908; Leavitt & Pickering 1912), can be used to determine extragalactic distances. In order to be used for distance determination, the PLR needs to be calibrated by using Cepheids with already known distances. There are multiple galaxies (so-called anchor galaxies) with well-known distances that host the Cepheids used for this calibration. Typically, one or a combination of these anchor galaxies are used for the calibration: the Milky Way, the Magellanic Clouds, M106, and the Andromeda galaxy (M31). Cepheids in the Milky Way need very precise parallaxes. Until recently, the number of Cepheids with accurate enough parallaxes was very limited. The Gaia mission (Gaia Collaboration et al. 2016) remedied this problem and will keep improving the parallaxes and therefore the precision of the calibration (see Riess et al. 2018 for an application of the Gaia DR2 parallaxes to Milky Way Cepheids). The Magellanic Clouds have been studied extensively (see, e.g., Soszyński et al. 2015 and Udalski et al. 2015 for the OGLE survey and Gieren et al. 2015 for the Araucaria project), and there are a lot of known Cepheids. Pietrzyński et al. (2013) used eclipsing binaries to determine the distance to the Large Magellanic Cloud (LMC) with an uncertainty of only 2%. For M106, a very accurate distance has been determined geometrically by a maser (Humphreys et al. 2013), and it is therefore also used as an anchor galaxy (Hoffmann & Macri 2015). Galaxy M31 is relatively nearby, the disk has a solar metallicity, and the full range of the PLR can be observed from ground- and space-based observatories in the optical and near-infrared. These advantages are partly compromised by the large extent on the sky, and significant crowding caused by the large inclination are major issues for an M31 Cepheid survey. There have, however, been multiple surveys of M31 Cepheids, and hence there are multiple Cepheid samples: the PAndromeda Cepheid sample (Kodric et al. 2013; hereafter K13) with 2009 Cepheids, the Vilardell et al. (2006) sample with 416 Cepheids, the 332 Cepheid sample from the DIRECT survey (Kaluzny et al. 1998, 1999; Stanek et al. 1998, 1999; Mochejska et al. 1999; Bonanos et al. 2003), and the WECAPP sample (Fliri et al. 2006) with 126 Cepheids. There are also Cepheid samples making use of the PHAT HST survey of M31 (Dalcanton et al. 2012): the Riess et al. (2012) sample with 68 Cepheids, the Kodric et al. (2015; hereafter K15) sample with 371 Cepheids, and the Wagner-Kaiser et al. (2015) sample with 175 Cepheids. See also Lee (2017) for a summary of time-domain studies in M31.
The PLR calibration is not the only goal of the PAndromeda project; it is also intended to study systematic effects on the PLR. The effect of metallicity on the PLR needs further investigation (see, e.g., Freedman & Madore 2011; Majaess et al. 2011). A metallicity effect on the PLR has to be accounted for when using the low-metallicity Magellanic Cloud PLR to determine the distance to a galaxy with a Milky Way–like metallicity like the Andromeda galaxy (M31). Another issue with the PLR is the possible existence of a broken slope (Sandage et al. 2009). The existence of a broken slope is heavily discussed in the literature. Ngeow et al. (2008) found a broken slope but linear relations in the Ks band and the Wesenheit PLR. Inno et al. (2013), however, found linear Wesenheit PLRs, while García-Varela et al. (2013) found no linear relations but rather exponential ones. In K15, we found a significant broken slope in all relations, including the Wesenheit PLR. The Cepheid sample we obtain here is used in Kodric et al. (2018; hereafter K18b), where we combine it with the PHAT data to obtain the largest HST Cepheid sample for M31 and investigate the broken slope. Since Cepheids are part of the distance ladder used to determine the Hubble constant (H0), the study of these systematic effects is not only important for the value and error budget of H0 but also for ascertaining how universal the PLR is in different galaxies. Riess et al. (2016) used Cepheids as part of their distance ladder in order to determine H0 with an uncertainty level of only 2.4%. Freedman & Madore (2010) summarized previous projects focused on determining H0 (for a more recent summary, see Riess et al. 2016 and references therein). Another uncertainty in the determination of H0 is the handling of outliers in the PLR, which is still discussed in the literature (e.g., Efstathiou 2014; Becker et al. 2015).
This paper is structured as follows. In Section 2, we discuss the PAndromeda survey and the data reduction procedure used for this paper, which is different from that of K13. In Section 3, we describe the Cepheid detection and classification. Section 4 describes the PAndromeda Cepheid catalog. The results are discussed in Section 5, followed by the conclusion in Section 6.
2. PAndromeda Data
PAndromeda is a Pan-STARRS1 (PS1; Kaiser et al. 2002; Hodapp et al. 2004; Tonry & Onaka 2009; Tonry et al. 2012; Chambers et al. 2016) survey dedicated to monitoring the Andromeda galaxy (M31). The original goal of the survey was microlensing (Lee et al. 2012), but the focus has shifted toward Cepheids (K13; Lee et al. 2013; K15), eclipsing binaries (Lee et al. 2014a), and luminous blue variables (Lee et al. 2014b). The 1.8 m PS1 telescope has a very large field of view (FOV) with ∼7 deg2, which makes it ideal for observing M31, since the complete disk can be observed in one exposure.
2.1. PAndromeda Survey
The galaxy M31 was observed from 2009 August 31 to 2010 January 21, 2010 July 23 to 2010 December 27, 2011 July 25 to 2011 November 22, and 2012 July 28 to 2012 October 31, with usually two visits per night in the gP1, rP1, and iP1 bands (see Tonry et al. 2012 for information on the PS1 filter system). The general exposure time is 60 s per frame, and each epoch (visit) is a stack of typically seven frames in the rP1 band and five frames in the gP1 and iP1 bands. The creation of this visit stack is described in the next section. Figure 1 shows the distribution of the number of frames making up one epoch. In total, there are 420 epochs in the rP1 band, 262 epochs in the iP1 band, and 56 epochs in the gP1 band. Due to masking (which will be discussed in the following section), the light curves usually have less epochs, so this is the maximum number of epochs. In addition, we disregard epochs that are based on less than three frames. In total, PAndromeda has taken 2511 frames in the rP1 band with a combined exposure time of 157,904 s (≈43.86 hr), 1260 frames in the iP1 band with a total of 75,840 s (≈21.07 hr), and 246 frames in the gP1 band with a combined exposure time of 15,048 s (≈4.18 hr).
The gigapixel camera (GPC) consists of 60 orthogonal transfer arrays (OTAs), where each OTA is made up of an array of 8 × 8 cells. One OTA has an FOV of 21' × 21'. The pixel scale is 025 pixel–1. The PS1 image-processing pipeline (IPP; Magnier 2006) provides us with astrometrically aligned frames (the frames are also bias- and flat-field-corrected). During the alignment, the exposures are remapped to a common grid, the so-called skycells. Only the first data reduction steps are done by the IPP; the other steps, done by ourselves, are discussed in the next section. Compared to K13, we changed the layout and size of the skycells so that they now cover 6647 pixels × 6647 pixels (i.e., 277 × 277) and use the intrinsic pixel scale of 025. Also, we defined the skycells such that they overlap (by 6'), in order to have a simple way to check our data reduction for consistency. The covered area is ∼7 deg2 in the rP1 and iP1 bands and ∼6.8 deg2 in the gP1 band. The central six skycells were split during our data reduction into four smaller skycells with 3524 pixels × 3524 pixels (i.e., 147 × 147), for reasons discussed in the next section. The area covered in the gP1 band is smaller, since skycells 003, 004, 005, 006, 012, and 028 are not used due to the lack of data in their skycells. The area in K13 (with ∼2.6 deg2) is much smaller than the area analyzed here. The skycell layout for this work (hereafter K18a), as well as for K13, is shown in the Appendix.
2.2. PAndromeda Data Reduction
The data reduction is based on our previous pipeline used in K13. The overall approach of using difference imaging (Alard & Lupton 1998) is the same. Apart from changing certain details, the biggest change is to mask a large fraction of the images in order to minimize systematic errors that are due to the poor PS1 image quality. The trade-off for less systematics is a larger statistical error caused by the rigorous masking. This was mainly done to help with the microlensing search (Lee et al. 2012), which was originally the main scientific goal of PAndromeda. As far as Cepheid light curves are concerned, the fact of having more statistical noise is usually negated by the large increase in epochs. But some area is lost or has substantially less epochs due to the masking. Therefore, we do not recover all Cepheids found in K13 and all eclipsing binaries found in Lee et al. (2014a).
As mentioned in the previous section, our data reduction starts with astrometrically aligned frames provided by the IPP (Magnier 2006) that are already debiased and flat-field-corrected. Our pipeline (see also Lee et al. 2012; Koppenhoefer et al. 2013; K13) uses the AstroWISE data-management environment (Valentijn et al. 2007). In the first step, where we ingest the frames provided by IPP into the AstroWISE database, we apply multiple masks. We use most of the masking flags already provided by IPP. The quality of these original masks was not sufficient, so we improved those masks and provided them to the IPP in order to reprocess all of our raw data using these new mask files. This step was necessary since the masks are based on the defects on the OTA cell level, and applying the masks before the skycell regridding is computationally more efficient. We also use the ghost masks provided by IPP but significantly increase the masking, since the position of the ghost model is sometimes off. We also mask satellite tracks using the linear Hough transform algorithm that is implemented in AstroWISE, and we use manually created masks that we have for a few hundred frames. Saturated pixels are also masked. Since there are offsets between the OTAs, we only use the dominant OTA in each skycell and mask the other OTAs. This implies that we mask up to 75% for the rare case where four OTAs are in one frame with the same area. This OTA masking is quite extreme, but it ensures that there are no systematic jumps inside a frame and that we can photometrically align the frames to each other. After applying these masks, we only use those frames that have more than 20% unmasked pixels in the central region of the frame (pixels 1661–4985 in x and y). The mean masking over all frames in all skycells is 64% in the rP1 and iP1 bands and 65% in the gP1 band. The mean masking per skycell ranges from 48% to 83% in the rP1 band, 48% to 82% in the iP1 band, and 43% to 85% in the gP1 band.
As can be seen in Figures 27–29, the central six skycells are each divided into four smaller skycells. The point-spread function (PSF) varies slightly over the FOV of the GPC. The PSF is broadest on the edges of the GPC but also blurred in the center of the GPC. Over the PAndromeda observing campaign, the center of the GPC was pointed on different skycells. Due to this, the central six skycells showed the largest PSF variation. By splitting them into smaller frames, the variation inside these new skycells is significantly reduced. This ensures that our PSF photometry, which uses a constant PSF model for a skycell, performs well.
In order to perform difference imaging, we need to construct so-called reference frames. A reference frame is a stack of the images with the best PSF. The frames are photometrically aligned so that the sky background and zero point are the same. Masked areas in one frame (if the masked region is not too large) are replaced from another frame with an almost similar PSF. The frames are assigned weights according to the inverse of the squared product of the seeing and the mean error. The reference frame is the weighted stack of these frames. In the rP1 and iP1 bands, we use 100 frames for the reference frame. For the gP1 band, we use 40 frames. Additionally, we require the original position angle to be zero or a multiple of 90° in the rP1 and iP1 bands. The reason for this is that we want to limit the increased noise caused by, e.g., chip gaps or bleeding to the four cardinal directions. The observing strategy required the second visit in one night to have a position angle of 90°. Unfortunately, there were also a significant number of frames taken with a random position angle. We disregard those for the reference frame in the rP1 and iP1 bands but have to use them in the gP1 band, because in this band we do not have enough data. In K13, we used the 70 best-seeing frames, but here we also selected according to masking. Since we increased the masking drastically, we have to make sure that the reference frames cover as much area as possible. Everything that is masked in the reference frame will be lost, even if there are frames that have data in these masked regions. Due to the fact that the masking fraction and the seeing are not correlated, we first select seven times more frames for the reference frame where the masking is the lowest and then select the best-seeing frames from those frames. The frame that is used to photometrically align the other frames is one of the 10 best-seeing frames that overlaps the most with the other frames. The reason for this is that the alignment works better the more area the frames have in common. Since we wanted to limit the area lost due to the new strict masking, we also require that a frame does not increase the masking in the reference frame too much. Pixels that are masked in all but one frame will still be dominant in the reference frame. During the creation of the reference frame, each frame contributing to it is checked if including it in the stack would decrease the unmasked area in the reference frame by more than 1%. If that is the case, the frame will be dropped from the reference frame. Therefore, the final reference frame can include less than 100 (rP1 and iP1 bands) or 40 (gP1 band) frames, respectively. Each reference frame is calculated twice. The first time, the sources are detected using SExtractor (Bertin & Arnouts 1996). The reference frame is inspected, and if there are, e.g., satellite tracks missed by the algorithm or ghosts, etc., we create manual masks for the input frames. This first reference frame is then used to photometrically align the input frames for the creation of the final second reference frame. In the second run, we use DAOPHOT (Stetson 1987) to detect the sources. Due to the crowding in M31, SExtractor sometimes determines a wrong position of a star, while DAOPHOT is optimized for high crowding. We perform PSF photometry on the positions identified with DAOPHOT on a background-subtracted reference frame. The neighboring stars are iteratively subtracted in order to account for the crowding. As in K13, we fit a bicubic spline model to determine the background (sky and M31). For each skycell, we determine a PSF model but do not vary it over the reference frame. The flux calibration is done as in K13; however, we do not use the Sloan Digital Sky Survey (SDSS; Abazajian et al. 2009) for the calibration but rather the PS1 PV2 catalog (Chambers et al. 2016). The reason for this is that the SDSS catalog is very sparsely populated in the central region of M31. With the PV2 catalog, we have a lot of comparison stars, so that we are able to also correct for PSF variation over the field by modeling the zero point with a spline.
For the difference imaging, we also need to create so-called visit stacks, which are stacks of all frames from one visit. Ideally, there should have been two visits of M31 per night, one in the first half of the night and one in the second half. The data taken have significant deviations from that. Sometimes the second visit was done right after the first, or there were three visits or visits with less frames. Also, the total integration time per visit changed several times in order to balance the total exposure time between the rP1 and iP1 bands. So, instead of automatically splitting the data of one night into two visit stacks, we grouped all the frames by hand into visit stacks with a stack depth as homogeneous as possible. The frames of a visit stack are photometrically aligned such that the sky background and zero point are the same. The weighted stack of these frames is photometrically aligned to the reference frame such that the visit stack and reference frame have the same zero point and sky background. Note that the visit stack has a smaller signal-to-noise ratio, since the reference frame is much deeper and the PSF is different, i.e., usually worse. As described in K13, the PSF of the reference frame is aligned to the visit stack by determining a convolution kernel when calculating the difference frame. Compared to K13, we decreased the size of the area one convolution kernel uses in order to better account for differences not only in PSF size but also PSF shape. We also masked out all sources brighter than 18 mag in the rP1 band, since those bright sources are not relevant for our science cases and are producing residuals in the difference frame. The PSF photometry on the difference frame uses the PSF constructed from the convolved reference frame.
In order to obtain the flux in one epoch of the light curve of a resolved source, the flux in the difference frame is added to the flux in the reference frame. Since the skycells overlap, the light curve can be constructed as a combination of measurements from different skycells. For the case where a resolved source has measurements of one epoch in multiple skycells, we use the measurement with the smallest error. For our Cepheid detection, we only use epochs in the light curve that have a signal-to-noise ratio larger than two and where the visit stack has more than two frames (the published light curves include those epochs for consistency). In total, we have detected 7,226,125 unique sources, where 3,348,137 unique sources have data in at least the rP1 band. For all sources that have rP1-band data, we determine the periodicity of the light curve with SigSpec (Reegen 2007). In K13, we explained in detail how we determined the periods and period errors. We have not changed any parameter of the period determination, since the procedure described in K13 works very well. For the period error determination, we increased the number of bootstrapping samples from 1000 in K13 to 10,000. The error rescaling factor described in K13 that was necessary due to the change of the pixel scale is as expected, 1.0, and therefore not necessary anymore. We use the AB magnitude system throughout this paper and in the published data.
3. Cepheid Detection and Classification
The goal is to find as many Cepheids as possible in an unbiased way. We could look through all periodic light curves by eye and select those that have a typical shape for a Cepheid. The problem with this approach is that different pulsators are sometimes hard to distinguish, especially in a noisy light curve. Also, the selection would be unreproducible and not objective. Therefore, we use essentially the same procedure we developed in K13 in order to be as unbiased as possible. We apply some selection criteria and manually determine a Cepheid sample from high signal-to-noise light curves and use the Fourier parameters of this sample to construct a three-dimensional parameter space. The parameters of the Cepheid light curves have to be within this parameter space and are further constrained by a color cut.
In order to find the Cepheids in our data, we apply several selection criteria. From the 3,348,137 resolved sources in the rP1 band, we start by selecting those that are also resolved in the iP1 band. We also require the source to be periodic in both bands and that the periods are similar to 1% . Additionally, we only select sources where the period in the rP1 band is . The light curves of the resulting 21,313 sources are fitted in order to obtain the Fourier coefficients A21 and φ21. Here A21 is the amplitude ratio of the first two Fourier components and φ21 is the phase difference between those two coefficients. The Fourier decomposition is done in the same way as described in K13; however, we use less degrees of freedom for the gP1 band, i.e., 3 instead of 5. This is because we have less epochs in the gP1 band. From the fit, we calculate the mean magnitude by determining the mean flux from the fitted curve and convert the mean flux back to a magnitude. The determination of the error of the mean magnitude is more difficult, since the error is not just the fit error of the mean magnitude. In K13, we used the photometric error determined in the reference frame as the mean magnitude error, since this was the dominating error. While it is still true that the error in the reference frame is the dominating error, we cannot just use this error because the light curve might consist of epochs from multiple reference frames. We determine the mean magnitude error by constructing a sample of 10,000 light-curve realizations. For each realization, we determine the mean magnitude by fitting the corresponding light curve. We construct a new light curve for each realization where each epoch's magnitude is the sum of the flux from the difference frame and a new flux from the corresponding reference frame. This new reference frame flux is drawn from a Gaussian distribution around the measured reference frame magnitude with a width of the measured photometric error. From the distribution of these 10,000 mean magnitudes, we calculate the mean magnitude error. For example, a light curve that consists of data from four different skycells is fitted 10,000 times. Each time, the four reference frame magnitudes are modified by drawing four new magnitudes from four Gaussian distributions with the width of the corresponding reference frame error. In this way, we obtain a new light curve where each epoch's magnitude is modified by changing the corresponding reference frame flux. If the light curve consists almost exclusively of epochs from one skycell, the error of the corresponding reference frame will dominate the mean magnitude error. This is not always the case, however; hence, we have to use this sophisticated procedure to determine the mean magnitude error. We also calculate the errors of A21 and φ21 simultaneously with calculating the mean magnitude error using the same technique. The errors of A21 and φ21 determined with this approach are much smaller than the fit errors of the Fourier decomposition. Therefore, we use the errors obtained from the Fourier decomposition, but the other errors are included in the published data (see the Appendix).
The Wesenheit, a reddening-free magnitude (Madore 1976, 1982; Opolski 1983), is defined as in K13:
with R = 3.86.
The selection criteria described in the next subsections could be performed in any order. The published data also include those light curves that do not make the selection criteria. We performed the cuts in the order described in the next subsections. Therefore, any fit parameters that were calculated after applying a selection criterion or cut are not available for such light curves excluded before performing the corresponding fit.
3.1. Manual Classification
As described previously, we need to identify a certain number of Cepheids manually so that we are able to construct a three-dimensional parameter space that enables us to find Cepheids in as unbiased a way as possible. Of course, the manual selection of Cepheids by light-curve shape is biased to some degree. It is especially difficult to determine the variable type for noisy light curves. As described previously, we start with 21,313 sources but only select those where the period significance determined by SigSpec is larger than 25. This leaves us with 10,562 light curves. From these, we preselected 2807 by using only those that are not noisy, and the light curves that even remotely resemble Cepheids, i.e., Mira-like light curves, were also selected. The selection is only based on the shape of the light curves in all available bands. So, we do not look at the position in the PLR. In order to be as unbiased as possible, four of the authors independently assigned flags to those 2807 light curves. We then matched the four classifications and reviewed cases where we disagreed. We disregarded sources where we could not reach a consensus of opinion. We end up with 1966 light curves that are manually classified as Cepheids based only on their light-curve shape. We apply the color cut described in the next subsection and clip 295 sources, leaving us with 1671 Cepheids. The color cut can be seen in Figure 2. The number of clipped sources is quite high considering that they were manually identified to have a Cepheid-like light curve. The color cut used in K13, shown with cyan lines, would cut less objects, but the new color cut (magenta lines) works better in selecting Cepheids in the complete sample.
Download figure:
Standard image High-resolution image3.2. Instability Strip Cuts
As already mentioned, we use a color cut for a better identification of Cepheids. We use the reddening-free Wesenheit and the rP1 – iP1 color for the selection, since we do not have gP1 data for all light curves. In order to define the color cut, we use the same approach as in K13. Theoretical pulsation models provide us with the information on how the instability strip edges depend on the effective temperature (Teff) and luminosity (L). We then use isochrones to translate the instability strip defined by Teff and L to the observable color and Wesenheit. In K13, we used the Fiorentino et al. (2002) instability strip but had to extend the edges by 0.2 mag in order to also include fainter but obvious Cepheids that would otherwise not have been covered, which makes the reliability of the instability strip parameters somewhat doubtful. Now we use the Anderson et al. (2016) instability strip edge models. As can be seen in Figure 3, those also include fainter and cooler Cepheids. We include all their models from Tables A.1 to A.6, i.e., all three metallicities (Z = 0.014, 0.006, and 0.002) and all three rotations (ωini = 0.0, 0.5, and 0.9). The Anderson et al. (2016) instability strip edges are used to define an area in which we require our Cepheids to reside. The area is defined by lines connecting the points (3.755, 4.8), (3.890, 1.5), (3.550, 4.8), (3.740, 2.934), and (3.812, 1.5), as shown in Figure 3. In the next step, we use the PARSEC isochrones (Bressan et al. 2012; Chen et al. 2014, 2015; Tang et al. 2014) and require the Teff and L of the isochrones to be within the area defined by the instability strip edges. This defines an area in the Wesenheit–color diagram, as seen in Figure 4. This region, which is defined by lines connecting the points (0.111, 16.450), (−0.010, 20.350), (−0.126, 25.900), (0.033, 25.134), (0.092, 23.222), (0.194, 20.430), (0.294, 19.003), (0.536, 16.680), and (1.067, 14.223), is used to perform the color cut. The area shown in Figure 4 is also valid for all metallicities used in the Anderson et al. (2016) models. In order not to clutter up the plot, we only show the solar-metallicity isochrones. The isochrones can also be used to determine the instability strip in other filter bands. We use this in K18b, where we identify our Cepheid sample in HST data. Since the color cut uses the intrinsic color, we have to correct for the extinction. As in K13, we use the Montalto et al. (2009) color-excess map and the foreground extinction toward M31 determined by Schlegel et al. (1998) (E(B − V)fg = 0.062) combined with the correction factor 0.86 from Schlafly & Finkbeiner (2011). The Montalto et al. (2009) map gives the color excess for a line of sight going completely through M31. Therefore, we only correct by half the color excess, assuming the Cepheid is only affected by half the dust. This corresponds to a probability argument, where sources are in front of an infinitely thin dust layer or behind, e.g., the extinction correction in the rP1 band, as
where Rr,P1 = 2.5535 (Ri,P1 = 1.8928 and Rg,P1 = 3.5258) for a reddening law of RV = 3.1. The assumption that the Cepheid is obscured by only half the dust can be refined in some cases. An object satisfies the color selection if it is within the area defined above, also considering its errors. So each object is not a point in the Wesenheit–color space but rather a box, where the height of the box is the error of the Wesenheit. The width of this box is the error of the color combined with the full color excess of that object. So the right side of the box is where only the foreground extinction has been corrected, and the left side is where the foreground plus the full color excess is used. In an edge-on view this corresponds to a location on the top of the disk, where only the foreground extinction has to be considered, or the bottom, where additionally the full color excess has to be taken into account. In case this box intersects the color-cut area defined by the instability strip edges, half the width of the box that is inside the color-cut area and is due to the color excess is used to correct the extinction. So we use the color-cut area to narrow down the range of color excess the object can have.
Download figure:
Standard image High-resolution imageDownload figure:
Standard image High-resolution imageIn the following, we use the extinction-corrected magnitudes, i.e., rP1 denotes the extinction-corrected magnitude. The uncertainty caused from the color excess is not propagated into the magnitude errors.
3.3. Type Classification
Fundamental-mode (FM) and first-overtone (FO) Cepheids, as well as type II (T2) Cepheids, occupy different locations in the A21–φ21–P space, which can be used to identify the Cepheid type (see, e.g., Udalski et al. 1999; Vilardell et al. 2007; K13). As in K13, we use two projections of this space, i.e., A21–P (see Figure 5) and φ21–P (see Figure 6), for the type classification. In K13, we used the A21–P projection to distinguish between FM and FO Cepheids and the φ21–P space to separate T2 and FM Cepheids. Here we additionally use an area in the φ21–P projection to find T2 Cepheids. The different Cepheid types could also be identified in the PLR, but by using this three-dimensional space, i.e., using the Fourier parameters, we only use the light-curve shape to determine the Cepheid type. The FO and FM PLRs have a small separation compared to their dispersions. Therefore, defining a line in the PLR that would distinguish both Cepheid types would bias the resulting PLRs. In the three-dimensional Fourier space, we also see overlaps between the different Cepheid type sequences. To address this issue, we require the Cepheids to be completely inside the defined areas, including their errors. Consequently, some Cepheids are in a transition region and therefore cannot be assigned a definite Cepheid type. We label those as the unclassified (UN) Cepheid type. In K13, we called the Cepheids that were clipped in the Wesenheit relation UN Cepheids. As we will discuss later, we still perform outlier clipping in the Wesenheit PLR and call the clipped Cepheids UN Cepheids. Therefore, in this paper, UN Cepheids can arise from two reasons. We do not distinguish between the two and call both UN Cepheids. We separate the areas by defining the following linear parameterization:
and
In order for a Cepheid to be assigned the FO type, the following two equations have to be true (see Figure 5):
and
For it to be a T2 Cepheid, the equations (see Figure 6)
and
or the equations (see Figure 5)
and
have to be true, i.e., (Equation (11) AND Equation (12) AND Equation (13)) OR (Equations (14) AND (15)). For it to be considered an FM Cepheid, at least one of the equations (see Figure 5)
and
has to be true, and additionally, at least one of the equations (see Figure 6)
and
has to be true. Also, at least one of the equations (see Figure 6)
and
has to be true; i.e., for an FM Cepheid, the logical connection is (Equation (16) OR Equation (17)) AND (Equations (18) OR Equation (19) OR Equation (20)) AND (Equation (21) OR Equation (22)). In case the Cepheid cannot be classified as FM, FO, or T2, it is classified as UN. Figure 7 shows that the type classification of the manually selected Cepheid sample performs reasonably well considering the dispersion in the PLRs.
Download figure:
Standard image High-resolution imageDownload figure:
Standard image High-resolution imageDownload figure:
Standard image High-resolution image3.4. Selection Criteria
In order to facilitate the automatic classification of Cepheids, additional selection criteria are needed. The combination of color cut and three-dimensional parameter space selection alone does not suffice to obtain a clean Cepheid sample. Table 1 summarizes the selection criteria and provides information on how many objects are rejected in each step. The cuts could be applied in any order without changing the resulting sample. The published Cepheid catalog also includes the tables and light curves of the clipped objects, as well as information on the clipping reason in the form of a bit flag. Some fits or, e.g., the final type determination are performed after certain cuts have been applied in order to save computation time. Therefore, the catalog of the clipped objects might not have all the information. The selection criteria are applied to the manually classified Cepheid sample, as well as the sample that is selected using the three-dimensional parameter space (3D sample). Table 1 provides the number of clipped objects for each sample. The selection criteria are applied first to the manual sample so that the three-dimensional parameter space can be defined (as explained in the next section). The selection criteria are then applied again to the resulting 3D sample. The selection criteria in Table 1 are applied to the rP1 and iP1 bands, while the criteria in Table 2 are applied to the gP1 band. Objects clipped due to the criteria in Table 1 are removed from the final Cepheid sample, while objects clipped only in the gP1 band (Table 2) are flagged and therefore not used for the gP1 PLR or the gP1 sample in general.
Table 1. Selection Criteria Used in the rP1 and iP1 Bands
Band | Selection Criterion | Manual | 3D | Flag | |
---|---|---|---|---|---|
I | All | Color cut | 295/1966 | 1922/3390 | 1 |
II | rP1 and iP1 | 15 mag ≤ m(t) ≤ 25 mag | 0/1671 | 0/1468 | 2 |
III | rP1 and iP1 | 0.1 mag ≤ max(m(t)) − min(m(t)) ≤ 1.75 mag | 1/1671 | 72/1468 | 4 |
IV | rP1 and iP1 | 20/1671 | 343/1468 | 8 | |
V | rP1 and iP1 | max(Θi − Θi+1) ≤ 0.25 | 0/1671 | 1/1468 | 16 |
VI | All | Same type from light-curve bootstrap ≥90% | 3/1671 | 73/1468 | 32 |
rP1 and iP1 | Final sample | 1640 | 1046 |
Note. The number of remaining Cepheids for the manual and 3D sample does not have to add up to the final number, since some cuts are performed at the same time and some objects fulfill multiple criteria. The columns Manual and 3D provide the number of clipped objects and the total number of objects still present at this step. The Flag column provides a bit flag so that the reason for the clipping can be found in the published data. The magnitude of the fitted light curve is denoted as m(t), while the observed magnitudes are called mi. Here Θi is the phase of the ith observed epoch, and Θi+1 is the next largest phase (which is not necessarily the next epoch).
Download table as: ASCIITypeset image
Table 2. Selection Criteria Used in the gP1 Band, Otherwise See Table 1
Band | Selection Criterion | Manual | 3D | Flag | |
---|---|---|---|---|---|
I | All | Color cut | 295/1966 | 1922/3390 | 1 |
II | gP1 | 15 mag ≤ m(t) ≤ 25 mag | 162/1671 | 391/1468 | 2 |
III | gP1 | 0.1 mag ≤ max(m(t)) − min(m(t)) ≤ 2.00 mag | 198/1671 | 504/1468 | 4 |
IV | gP1 | 24/1671 | 362/1468 | 8 | |
V | gP1 | max(Θi − Θi+1) ≤ 0.25 | 175/1671 | 246/1468 | 16 |
VI | All | Same type from light-curve bootstrap ≥90% | 3/1671 | 73/1468 | 32 |
gP1 | Final sample | 1302 | 595 |
Download table as: ASCIITypeset image
Criterion I, the color cut discussed previously, is an important selection criterion for the 3D sample. Selection criterion II is a magnitude cut that ensures that the fitted light curve does not overshoot. As can be seen in Table 1, this cut is not used for the rP1 and iP1 band, but the cut is relevant for the gP1 band (Table 2). This is not surprising, since the gP1 band has relatively few epochs, and the light curves therefore have gaps that facilitate overshoots in the light-curve fitting. Criterion III is an amplitude cut, where we reject small amplitudes because those are difficult to distinguish from noise. Amplitudes are smaller for longer wavelengths (Madore & Freedman 1991), and therefore we have a different upper amplitude cut in the gP1 band. In criterion IV, we make a noise cut by comparing the median absolute deviation of the observed magnitudes with the fitted magnitudes in relation to the amplitude of the light curve. Large gaps in the light curve cause problems in fitting the light curve. As mentioned previously, the gaps might cause overshoots, but even without overshoots, the shape of the light curve might not be determined well enough for the type classification or the determination of the Fourier parameters A21 and φ21. Therefore, we introduce criterion V, where we require that the largest gap is smaller than a quarter phase of the light curve. As we mentioned previously, we perform 10,000 bootstrapping realizations for determining the magnitude error and the errors of the Fourier parameters A21 and φ21. For each of those realizations, we determine the Cepheid type and require in criterion VI that the Cepheid type is the same in more than 9000 realizations. There are also two additional flags (64 and 128) not mentioned in Tables 1 and 2. Flag 64 is assigned to seven manually classified Cepheids, and they are removed from the Cepheid sample because of their location in the 3D parameter space. The reason for this is explained in the next subsection. Flag 128 is assigned 19 times in the final Cepheid catalog (so those Cepheids are still included in the catalog) in order to mark Cepheids where the light curve from visual inspection does not look like a typical Cepheid light curve. We keep those Cepheids in the sample so that the selection of Cepheids is as objective as possible. The results do not change if we disregard those Cepheids.
3.5. 3D Classification
Similar to K13, we use the 3D parameter space A21–φ21–P to select light curves that have a shape typical for a Cepheid. The advantages of this approach are that it is less biased than visual selection and the selection is reproducible. Nevertheless, the approach requires the manual sample in order to define this 3D space. In K13, we spanned a grid in this 3D space and selected those grid points where one manually classified Cepheid resides. By doing this, we had to choose six parameters, i.e., the grid size and starting point for each dimension. Here we improve the method by using only three parameters. We define a sphere around each point in the manual sample. In order for an object to be selected, it has to be inside at least one of the spheres that are made up by the manual sample in the A21–φ21–P space. One of the free parameters is the radius of the sphere, and the other two are the normalizations of the other two dimensions, i.e., how each dimension is scaled compared to the other dimensions. We choose A21–(2π)−1φ21–log(P) as the new 3D space. We choose it such that the values of the manual sample have a similar range. The reason for this is that we can use a sphere and do not have to use an ellipsoid. For example, if we use a radius of 0.2 in A21–φ21–P space, that would be a large difference in A21, while it is only a small change in φ21 and almost no change in P. In order to determine the radius of the sphere, we use the manual sample. We calculate for each Cepheid in the manual sample the distance in A21–(2π)−1φ21–log(P) space to the next closest Cepheid. The cumulative distribution function of this distance is then used to choose the radius of the sphere. We select a radius of 0.088 for the sphere (99.5% of Cepheids are closer than this distance to their next neighbor). Seven Cepheids have a larger distance to the next neighbor. We choose to cut those seven Cepheids because they are in a region of the 3D space that is sparsely sampled, and if they were to be included, the radius of the sphere would have to be increased significantly, which makes the selection worse (a larger radius means a more fuzzy selection). If the seven Cepheids were left in the sample but the 0.088 radius were used, then the seven Cepheids would be rejected by the 3D selection process if the manual sample were subjected to the 3D selection (which it is not because it is used to define it). So, for consistency reasons, the seven Cepheids are excluded. As described in the previous subsection, the seven Cepheids are flagged with the 64 bit flag. In the Appendix, the location of those seven Cepheids in the 3D space can be seen in Figures 32–34. The 3D classification finds an additional 3390 Cepheid candidates. As can be seen in Figure 8, 1922 of those are clipped due to the color cut.
Download figure:
Standard image High-resolution image3.6. Outlier Clipping
Figure 9 shows the Wesenheit PLR of the combined manual and 3D sample after all selection criteria have been applied. As can be seen, there are outliers. There are different reasons for these outliers: crowding, misclassification, misidentification, blending, and extinction. In K15, we discussed those reasons in more detail, but we want to point out that the crowding problem is significantly more severe in these ground-based optical data than in the HST data. In order to remove the outliers from the sample, we perform an iterative κ–σ clipping with the median absolute deviation of the residuals as the magnitude error. This method that we developed in K15 is very robust and is also used in K18b, as well as in Riess et al. (2016) and Hoffmann et al. (2016). As in K15 and K18b, we use κ = 4. The clipped Wesenheit PLR shown in Figure 10 has no outliers anymore. In Figure 11, we show all Cepheids that were clipped, as well as the UN Cepheids. The UN Cepheids in this figure were assigned the UN type because the Cepheids are occupying a transitional region where the type classification is unsecure (see Section 3.3). In K13, the UN type was assigned only to the clipped Cepheids. Here the UN type is assigned to all Cepheids shown in Figure 11. After this point, we do not distinguish between the two reasons a Cepheid could have been assigned the UN type. The 3D parameter space of the final clipped sample is shown in Figures 12 and 13. In these figures, we already do not distinguish between the two UN type origins. The 3D parameter space is more densely sampled, and of course, it is similar to the manual sample shown in Figures 5 and 6, since the additional Cepheids have been selected by this 3D space. It can also be seen that in the final sample, the transitional region between Cepheid types is more prominently populated, and therefore there are more UN Cepheids in these regions than in the manual sample.
Download figure:
Standard image High-resolution imageDownload figure:
Standard image High-resolution imageDownload figure:
Standard image High-resolution imageDownload figure:
Standard image High-resolution imageDownload figure:
Standard image High-resolution image4. PAndromeda Cepheid Catalog
The PAndromeda Cepheid catalog consists of 1662 FM Cepheids, 307 FO Cepheids, 278 T2 Cepheids, and 439 UN Cepheids and is the largest and most homogeneous catalog of Cepheids in M31. For all Cepheids, we provide the light curves, as well as the fitted parameters in electronic form. In the Appendix, we show an excerpt of the published tables and briefly discuss the fit parameters. We also provide the light curves and fit parameters of 2670 sources that have been cut for the various reasons discussed in the previous section. The Appendix also includes a comparison to the K13 data. As can be seen in Figure 14, many Cepheids in the rP1 band have a lot of epochs, and therefore the periods can be determined very accurately, as seen in Figure 15. An example light curve is shown in Figure 16. Typically, the K13 data have less epochs, as is also the case in the example light curve, and the phase zero point is arbitrary and therefore will be different. Note that because of the strict masking, some epochs present in K13 might be missing here. In some extreme cases, there are no epochs from a certain year of the observing campaign due to the masking. This is in part the reason that some Cepheids identified in K13 are missing in this sample. Some areas are completely masked, and those Cepheids can no longer be recovered. Here 140 Cepheids from K13 have no data due to masking, and 148 Cepheids do not make the initial cut, where we require the periods in the rP1 and iP1 bands to be similar to 1% and in a period range between 1.5 and 150 days. Those Cepheids are not included in the published light curves of the selected sources, because we do not publish the sources that do not make the first cut due to their large quantity. These 148 Cepheids are a rather large number of missing Cepheids, since all K13 Cepheids should make this simple cut. The strict masking can remove a significant number of epochs in one band such that the period cannot be determined precisely anymore. This is the reason, then, that the Cepheids do not make the first cut. Of the remaining 1721 Cepheids from K13, 1453 are part of the PAndromeda Cepheid catalog. The other 268 Cepheids are cut for various reasons. As with the other sources that have been cut, the light curves are also published, and the reason for the cut is encoded in the bit flag. A cross-identification table is provided with the published data. In the Appendix, we compare the Cepheid type of the two samples. The spatial distribution of FM and FO Cepheids shown in Figure 17 follows the ring structure in M31, while the T2 Cepheids shown in Figure 18 trace the M31 halo.
Download figure:
Standard image High-resolution imageDownload figure:
Standard image High-resolution imageDownload figure:
Standard image High-resolution imageDownload figure:
Standard image High-resolution imageDownload figure:
Standard image High-resolution image5. Results
The PLR fit parameters are summarized in Table 3, while the corresponding fits are shown in Figures 19–22. The T2 Cepheids are not the subject of this work, so they are shown without a fit. Fit parameters for long-period Cepheids () are also provided. Due to the fact that the intrinsic dispersion of the PLR is much larger than the typical photometric error, we do not use the photometric errors shown in Figures 19–22 as a weight in the fit. Rather, we assign the same weight to each Cepheid. An alternative to this approach is the method used in Efstathiou (2014), namely, to quadratically add an internal dispersion factor σint to all photometric errors such that the χ2 of the fit is 1. The resulting PLR fit parameters of this method are within the errors shown in Table 3. This is not surprising, since the resulting σint is much larger than the typical photometric error, and thus the weight is almost the same for each Cepheid. The error of each fit parameter is in the ±σ range around the fit parameter determined from a distribution of 10,000 bootstrapping samples. As mentioned previously, not all Cepheids have color-excess information from the Montalto et al. (2009) map, and therefore some Cepheid magnitudes are only corrected for foreground extinction. If we remove these from the PLR fit, the result stays the same within the given errors. Of course, some Cepheids might not be included in our sample because they are rejected in the color cut due to the missing color-excess information. The PLR dispersion is significantly reduced (by ≈25%) in the Wesenheit compared to K13. Also, the dispersion for the long-period sample is greatly reduced. The FM and FO PLR dispersion is also reduced or stays the same, as in the case of the rP1-band PLR. Figure 23 shows a comparison of the slope values to other work. Compared to K13, our new slopes are shallower. The theoretical predictions of Bono et al. (2010) are quite close to our slopes. As in Bono et al. (2010) and Di Criscienzo et al. (2013), we also observe that the long-period sample slope is steeper than the slope of the complete sample. Due to the large scatter in the gP1 PLR, the slopes of the long-period sample and the complete sample are almost identical. Our slopes are steeper than the slopes for the LMC and SMC in Soszyński et al. (2015).
Download figure:
Standard image High-resolution imageDownload figure:
Standard image High-resolution imageDownload figure:
Standard image High-resolution imageDownload figure:
Standard image High-resolution imageDownload figure:
Standard image High-resolution imageTable 3. PLR Fit Parameters
# | Band | Type | Range | Nfit | a (log P = 1) | Slope b | σ |
---|---|---|---|---|---|---|---|
1 | Wesenheit | FM | All | 1662 | 19.752 (0.008) | −3.323 (0.026) | 0.327 |
2 | Wesenheit | FM | log(P) > 1 | 422 | 19.760 (0.020) | −3.247 (0.058) | 0.265 |
3 | Wesenheit | FO | All | 307 | 19.256 (0.071) | −3.065 (0.131) | 0.317 |
4 | rP1 | FM | All | 1662 | 20.371 (0.011) | −2.548 (0.035) | 0.354 |
5 | rP1 | FM | log(P) > 1 | 422 | 20.365 (0.028) | −2.443 (0.105) | 0.398 |
6 | rP1 | FO | All | 307 | 19.811 (0.065) | −2.456 (0.115) | 0.259 |
7 | iP1 | FM | All | 1662 | 20.212 (0.009) | −2.749 (0.029) | 0.304 |
8 | iP1 | FM | log(P) > 1 | 422 | 20.210 (0.023) | −2.651 (0.083) | 0.322 |
9 | iP1 | FO | All | 307 | 19.667 (0.060) | −2.615 (0.105) | 0.243 |
10 | gP1 | FM | All | 1298 | 20.822 (0.018) | −2.202 (0.058) | 0.433 |
11 | gP1 | FM | log(P) > 1 | 289 | 20.843 (0.051) | −2.186 (0.262) | 0.533 |
12 | gP1 | FO | All | 246 | 20.319 (0.087) | −1.997 (0.152) | 0.298 |
Note. The magnitude errors were set to the same value. The errors of the fitted parameters were determined with the bootstrapping method.
Download table as: ASCIITypeset image
In K15, we found a broken slope in the HST F110W and F160W data. The large dispersion in the ground-based data could make the identification of the broken slope more difficult. Table 4 shows the broken-slope fit with a common suspension point at . These fits are shown in Figures 19–22. As in K15, we test how significant the broken slope is by using the bootstrapping method. Figure 24 shows that the broken slope is not as significant as in K15, since the contour lines overlap. But it also shows that there are clearly two distinct distributions, except in the gP1 band. This is consistent with the significantly disjunct slopes shown in Table 4. In the gP1 band, the error of the slope for the long-period Cepheids is large, which is consistent with the fact that the long-period Cepheids in the gP1 band have a large scatter, and there are relatively few of them. So the fact that we do not observe the broken slope in the gP1 band might be due to the few epochs in the gP1 band, since a large error of the long-period slope can mask a possible broken slope in the gP1 band.
Download figure:
Standard image High-resolution imageTable 4. Broken-slope PLR Fit Parameters
# | Band | Nfit | σ | |||
---|---|---|---|---|---|---|
1 | Wesenheit | 1662 | −3.461 (0.051) | −3.131 (0.050) | 19.715 (0.014) | 0.326 |
2 | rP1 | 1662 | −2.679 (0.054) | −2.368 (0.086) | 20.336 (0.017) | 0.353 |
3 | iP1 | 1662 | −2.883 (0.046) | −2.565 (0.068) | 20.176 (0.014) | 0.303 |
4 | gP1 | 1298 | −2.286 (0.077) | −2.021 (0.189) | 20.799 (0.025) | 0.433 |
Note. Two lines with a common suspension point at are fitted. The magnitude errors were set to the same value. The errors of the fitted parameters were determined with the bootstrapping method.
Download table as: ASCIITypeset image
While the dispersion of the FO PLR decreased compared to K13, it is still not as nicely separated from the FM Cepheid PLR as in the HST data in K15 and K18b. This can best be seen in Figure 11, where some of the FO Cepheids overlay the FM Cepheids and vice versa (in Figure 19, we offset the FO Cepheids so that the two relations can be seen without the overlap). The FO-to-FM ratio is 0.18 and therefore twice the ratio in K13. This is closer to the ratio of 0.27 reported by Vilardell et al. (2007) for M31 and the ratio of 0.13 from the General Catalog of Variable Stars (Samus et al. 2017) for the Milky Way. As already described in K13, lower-metallicity objects like the SMC and LMC have higher FO-to-FM ratios on the order of 0.4–0.7 (Udalski et al. 1999). However, there are significant uncertainties associated with our FO-to-FM ratio, mostly stemming from the type classification and the fact that we do not perform a completeness analysis. Especially, the lower-period cut is influencing our result.
We also determine amplitude ratios for our sample. Table 5 summarizes the fits shown in Figure 25. The amplitude ratio between the rP1 and iP1 bands decreases slightly compared to K13. As expected, the amplitude ratio between the rP1 and gP1 bands is smaller than 1, since the Cepheid amplitudes decrease with increasing wavelength (Madore & Freedman 1991). The obtained fits show no or only slight dependence on period. We also determine the phase lag between the three different bands. As can be seen in Figure 26, the phase lag is zero and independent of the period. This fact enabled us in K18b to correct the random phased HST data.
Download figure:
Standard image High-resolution imageDownload figure:
Standard image High-resolution imageTable 5. Amplitude Ratio Fit Parameters
# | Amplitude Ratio | Type | Range | Nfit | a (log P = 1) | Slope b | σ |
---|---|---|---|---|---|---|---|
1 | FM | All | 1662 | 1.224 (0.003) | 0.009 (0.012) | 0.137 | |
2 | FM | All | 1298 | 0.694 (0.005) | 0.059 (0.018) | 0.137 | |
3 | FO | All | 307 | 1.286 (0.032) | 0.113 (0.059) | 0.151 | |
4 | FO | All | 246 | 0.656 (0.032) | 0.005 (0.060) | 0.133 |
Note. The errors of the fitted parameters were determined with the bootstrapping method.
Download table as: ASCIITypeset image
6. Conclusion
We use the PAndromeda data to obtain the largest and most homogeneous Cepheid sample in M31. We find 1662 FM Cepheids, 307 FO Cepheids, 278 T2 Cepheids, and 439 Cepheids with unclear Cepheid type. Similar to K13, we use a combination of a 3D Fourier parameter space, a color cut, and other selection criteria to find the Cepheids in as unbiased and reproducible a way as possible. We improved this approach by implementing a new type classification and using an improved color cut using the Anderson et al. (2016) instability strip edge models. The Anderson et al. (2016) instability strip edge models describe our data much better than the Fiorentino et al. (2002) instability strip edge models. However, the Anderson et al. (2016) instability strip does not cover a substantial fraction of the Cepheids that we classified manually (see Figure 2).
The PLRs we obtain in the rP1, iP1, and gP1 bands (see Figures 19–22) have smaller dispersions than in K13. This is achieved by a rigorous masking of the PS1 data and improved data reduction. As in K15, with HST data, we find a broken slope (see Figure 24), except in the gP1 band. The broken slope is not as significant as in K15.
We used the Cepheid sample obtained here in Kodric et al. (2018), where we obtained space-based HST PLRs in two near-infrared and two optical bands. The Kodric et al. (2018) Cepheid sample is the largest Cepheid sample in M31 with HST data. The PLR dispersions in the HST optical bands are similar to the dispersions obtained here with ground-based data. This means that the crowding effect on our ground-based but mean phase-corrected photometry is comparable to the effect of using random phased data in the HST sample. In K18b, we showed that the color selection can cause the broken slope. Therefore, the new improved color cut could contribute to the observed broken slope.
The complete Cepheid catalog, including the light curves and objects clipped during the selection process, can be found electronically in the CDS database.
This research was supported by the DFG cluster of excellence Origin and Structure of the Universe (www.universe-cluster.de).
Appendix:
A.1. Skycell Layout
The skycell layout discussed in Section 2.1 is shown in Figures 27–29 for K18a. Figures 30 and 31 show the skycell layout in K13.
Download figure:
Standard image High-resolution imageDownload figure:
Standard image High-resolution imageDownload figure:
Standard image High-resolution imageDownload figure:
Standard image High-resolution imageDownload figure:
Standard image High-resolution imageA.2. Flag 64
The bit flag 64 is assigned seven times, as explained in Section 3.5. The seven manually classified Cepheids have a larger distance to their respective next Cepheid neighbor in 3D space. Figures 32 and 33 show the location of these seven Cepheids in the amplitude ratio (A21) and phase difference (φ21) diagrams. Figure 34 shows where these seven Cepheids reside in the period Wesenheit diagram.
Download figure:
Standard image High-resolution imageDownload figure:
Standard image High-resolution imageDownload figure:
Standard image High-resolution imageA.3. Comparison with K13
Since our data have more epochs and cover a larger time span than K13, the periods are also more precise. The long periods show a larger relative change in Figure 35, since for these periods, we now cover multiple pulsation cycles. We compare the rP1- and iP1-band magnitudes in Figures 36 and 37 for the 1445 Cepheids that are in both samples. There is no offset, but some Cepheids show a large magnitude change. The large magnitude differences are explained with a bad SExtractor position (on which the forced photometry was carried out) in K13 due to crowding. Most of these had a source position between two stars, and the brighter magnitude from this sample is now correct, since the position is correctly identified. But there are also a few cases where the new magnitude is fainter. This happens when there are two very close stars where the PSFs overlap and the pulsation was previously wrongly attributed to the brighter source. In Figure 38, we show how the Cepheid type changes. Of the 1445 Cepheids that are in both samples, 1224 do not change the Cepheid type. As can be seen, most change from UN to FM and from FM to UN, and almost all changes except for nine cases involve the UN type. A cross-match table is provided on the CDS.
Download figure:
Standard image High-resolution imageDownload figure:
Standard image High-resolution imageDownload figure:
Standard image High-resolution imageDownload figure:
Standard image High-resolution imageA.4. Excerpt of the Published Tables
Tables 6–11 give a description of the published data. The tables include the clipped sources, but not all fits have necessarily been performed for those objects, depending at which point the object was cut. In those cases, the columns show −1 or −99 as a default value. Some values can also be so small that they are shown as 0.0 in the tables. The errors of some values have been determined by different methods. We use the errors shown in Table 6. The method used to determine the magnitude errors is described in Section 3. A distribution of magnitudes is obtained, and the error is calculated from this distribution in two slightly different approaches. The first method is to determine the error from the ±σ/2 range around the mean magnitude from the light-curve fit. The other method is to determine the error from the 1σ range of the distribution regardless of the fitted mean magnitude from the light curve. The two errors are the same if the mean of the distribution is the same as the mean magnitude from the light-curve fit. This also implies that the difference of these two error estimates is a tracer of how symmetric the distribution is around the mean magnitude from the light curve. A good example for this is the period error, where the distribution is obtained by using the bootstrapping method on the light curve and the error is calculated with the two methods. The period calculated from the light curve can be so much on the edge of the distribution of periods obtained from the bootstrapping that it is outside the 1σ range of the distribution, and therefore the error from the ±σ/2 range around the mean period is not defined anymore. The period of a light curve that is changed drastically when bootstrapping the light curve would show this behavior. Note that these Cepheids do not have poor period estimates but rather all epochs are important for the period determination. The error from the 1σ range can always be calculated, and therefore it is used as the period error. The ±σ/2 range period error would not be defined for all Cepheids. The different error estimates are not used in this work to perform additional cuts, but we include them in the published data for the interested reader.
Table 6. Description of main.dat
Column # | Label | Explanation |
---|---|---|
1 | PSO id | PS1 identifier |
2 | id | Identifier used through the K18a data |
3 | R.A. | Right ascension (J2000.0) |
4 | Decl. | Declination (J2000.0) |
5 | Period in the rP1 band (determined with SigSpec) | |
6 | Period error in the rP1 band determined with half the width of | |
the 1σ interval of the period distribution obtained by the bootstrapping method | ||
7 | rP1,0 | Extinction-corrected rP1-band magnitude (mean flux of the |
light curve converted to magnitudes) | ||
8 | re,P1 | Magnitude error in the rP1 band |
determined by the method described in Section 3 | ||
9 | iP1,0 | Extinction-corrected iP1-band magnitude |
10 | ie,P1 | Magnitude error in the iP1 band (see Section 3) |
11 | gP1,0 | Extinction-corrected gP1-band magnitude |
12 | ge,P1 | Magnitude error in the gP1 band (see Section 3) |
13 | W | Wesenheit magnitude as defined in Equation (1) |
14 | We | Wesenheit magnitude error |
determined with the same method as re,P1; see Section 3 | ||
15 | A21 | Amplitude ratio of the first two Fourier components |
in the rP1 band (see K13 Equation (5)) | ||
16 | Ae,21 | Error of the amplitude ratio A21 |
obtained from the error of the Fourier decomposition | ||
17 | Phase difference of the first two Fourier components | |
in the rP1 band (see K13 Equation (6)) | ||
18 | φe,21 | Error of φ21 (same method used as for Ae,21) |
19 | sample | Manual in case the Cepheid was classified manually and 3D in case the Cepheid |
was classified with the 3D parameter space | ||
20 | Bit flag for the rP1 and/or iP1 bands as described in Section 3.4 | |
21 | Bit flag for the gP1 band (see Section 3.4) | |
22 | type | Cepheid type (see Section 3.3) |
23 | catalog | Flag indicating whether the object is part of the catalog (C) |
or has been clipped/rejected (R) |
Only a portion of this table is shown here to demonstrate its form and content. A machine-readable version of the full table is available.
Table 7. Description of lc-info.dat
Column # | Label | Explanation |
---|---|---|
1 | id | Identifier used through the K18a data |
2 | Number of epochs in the rP1-band light curve | |
3 | Number of epochs in the iP1-band light curve | |
4 | Number of epochs in the gP1-band light curve | |
5 | Period in the iP1 band | |
6 | Period in the gP1 band | |
7 | SigSpec significance of the rP1-band period (see Section 3 in K13) | |
8 | Significance of the iP1-band period | |
9 | Significance of the gP1-band period | |
10 | Number of epochs used for the Cepheid light curve in the rP1 band | |
(all epochs are required to have a signal-to-noise ratio larger than two, and | ||
the corresponding visit stack has to have more than two frames) | ||
11 | Number of epochs used for the Cepheid light curve in the iP1 band | |
12 | Number of epochs used for the Cepheid light curve in the gP1 band | |
13 | Minimum magnitude (i.e., brightest magnitude) | |
of the Fourier fit to the light curve in the rP1 band | ||
14 | mini | Minimum magnitude of the Fourier fit to the light curve in the iP1 band |
15 | ming | Minimum magnitude of the Fourier fit to the light curve in the gP1 band |
16 | maxr | Maximum magnitude (i.e., faintest magnitude) |
of the Fourier fit to the light curve in the rP1 band | ||
17 | maxi | Maximum magnitude of the Fourier fit to the light curve in the iP1 band |
18 | maxg | Maximum magnitude of the Fourier fit to the light curve in the gP1 band |
19 | rP1 | rP1-band magnitude (i.e., not extinction-corrected as the rP1,0 magnitude but |
also determined with the mean flux of the light curve converted to magnitudes) | ||
20 | iP1 | iP1-band magnitude |
21 | gP1 | gP1-band magnitude |
22 | Amplitude ratio in the iP1 band (analog to A21 but for the iP1 band) | |
23 | Amplitude ratio in the gP1 band | |
24 | Error of the amplitude ratio in the iP1 band | |
(analog to , i.e., error from the Fourier decomposition) | ||
25 | Error of the amplitude ratio in the gP1 band | |
26 | Phase difference of the first two Fourier coefficients in the iP1 band | |
(analog to ) | ||
27 | Phase difference in the gP1-band | |
28 | Error of the phase difference in the iP1 band (analog to ; | |
the method described in Section 3 is used to determine the error) | ||
29 | Error of the phase difference in the gP1 band | |
30–40 | Fourier coefficients in the rP1 band as defined in Equation (3) in K13 | |
(the order of the fit is N = 5, so there are 11 Fourier parameters) | ||
41–51 | Fourier coefficients in the iP1 band | |
(the order of the fit is N = 5, therefore there are 11 Fourier parameters) | ||
52–62 | Fourier coefficients in the gP1 band | |
(the order of the fit is N = 3, but there are still 11 columns where the columns | ||
for the higher orders are set to a default value of −1) | ||
63–73 | Fit error of the Fourier coefficients in the rP1 band | |
74–84 | Fit error of the Fourier coefficients in the iP1 band | |
85–95 | Fit error of the Fourier coefficients in the gP1 band | |
96 | Largest phase (Θ) gap in the folded rP1-band light curve () | |
97 | Largest phase gap in the folded iP1-band light curve | |
98 | Largest phase gap in the folded gP1-band light curve | |
99 | catalog | Flag indicating whether the object is part of the catalog (C) |
or has been clipped/rejected (R) |
Only a portion of this table is shown here to demonstrate its form and content. A machine-readable version of the full table is available.
Table 8. Description of other.dat
Column # | Label | Explanation |
---|---|---|
1 | id | Identifier used through the K18a data |
2 | pretype | Cepheid type before the outlier clipping (see Section 3.6) |
3 | Number of different skycells in which the Cepheid is in the rP1 band | |
4 | Number of different skycells in which the Cepheid is in the iP1 band | |
5 | Number of different skycells in which the Cepheid is in the gP1 band | |
6 | Decline/rise factor in the rP1 band as defined in Section 5 in K13 | |
7 | Decline/rise factor in the iP1 band | |
8 | Decline/rise factor in the gP1 band | |
9 | Noise factor in the rP1 band as defined by selection criterion IV in Table 1 | |
10 | Noise factor in the iP1 band as defined by selection criterion IV in Table 1 | |
11 | Noise factor in the gP1 band as defined by selection criterion IV in Table 2 | |
12 | Color excess from the Montalto et al. (2009) map | |
but improved where possible by the instability strip as discussed in Section 3.2 | ||
13 | Magnitude error in the rP1 band determined from the 1σ range method | |
14 | Magnitude error in the iP1 band determined from the 1σ range method | |
15 | Magnitude error in the gP1 band determined from the 1σ range method | |
16 | Percentage () of rP1-band epochs used in all 10,000 light-curve realizations | |
(see Section 3; epochs can be cut if the sum of the drawn reference frame flux | ||
and difference frame flux is negative or the signal-to-noise ratio is smaller than two) | ||
17 | Percentage of iP1-band epochs (analog to ) | |
18 | Percentage of gP1-band epochs (analog to ) | |
19 | Mean rP1-band magnitude of the distribution used to calculate the error | |
20 | Mean iP1-band magnitude of the distribution used to calculate the error | |
21 | Mean gP1-band magnitude of the distribution used to calculate the error | |
22 | Standard deviation of the rP1-band distribution used to calculate the error | |
23 | Standard deviation of the iP1-band distribution used to calculate the error | |
24 | Standard deviation of the gP1-band distribution used to calculate the error | |
25 | Amplitude ratio error in the rP1 band determined from the 1σ range method | |
26 | Amplitude ratio error in the iP1 band determined from the 1σ range method | |
27 | Amplitude ratio error in the gP1 band determined from the 1σ range method | |
28 | Amplitude ratio error in the rP1 band determined from the ±σ/2 range method | |
29 | Amplitude ratio error in the iP1 band determined from the ±σ/2 range method | |
30 | Amplitude ratio error in the gP1 band determined from the ±σ/2 range method | |
31 | Phase difference error in the rP1 band determined from the 1σ range method | |
32 | Phase difference error in the iP1 band determined from the 1σ range method | |
33 | Phase difference error in the gP1 band determined from the 1σ range method | |
34 | Phase difference error in the rP1 band determined from the ±σ/2 range method | |
35 | Phase difference error in the iP1 band determined from the ±σ/2 range method | |
36 | Phase difference error in the gP1 band determined from the ±σ/2 range method | |
37 | Period error in the iP1 band determined with the 1σ range method | |
38 | Period error in the gP1 band determined with the 1σ range method | |
39 | Period error in the rP1 band determined with the range method | |
40 | Period error in the iP1 band determined with the range method | |
41 | Period error in the gP1 band determined with the range method | |
42 | Percentage () of the 10,000 bootstrapped light curves in the rP1 band | |
for which SigSpec was able to determine a period | ||
43 | Percentage () of the 10,000 bootstrapped light curves in the iP1 band | |
for which SigSpec was able to determine a period | ||
44 | Percentage () of the 10,000 bootstrapped light curves in the gP1 band | |
for which SigSpec was able to determine a period | ||
45 | Amplitude ratio error in the rP1 band determined from the 1σ range method | |
where the distribution is obtained from 10,000 bootstrapped light curves | ||
from the measured light curve (i.e., not the method described in Section 3, | ||
where the reference frame flux is drawn from a distribution) | ||
46 | Amplitude ratio error in the rP1 band determined from the range method | |
with bootstrapping | ||
47 | Phase difference error in the rP1 band determined from the 1σ range method | |
with bootstrapping | ||
48 | Phase difference error in the rP1 band determined from the range method | |
with bootstrapping | ||
49 | Percentage () of how often the FM type is assigned in the 10,000 realizations | |
(i.e., criterion VI in Section 3.4; the Cepheid type is the pretype, | ||
i.e., the type before the outlier clipping) | ||
50 | Percentage of how often the FO type is assigned | |
51 | Percentage of how often the T2 type is assigned | |
52 | Percentage of how often the UN type is assigned | |
53 | Smallest possible color excess determined from the Montalto et al. (2009) map | |
and the instability strip edges, as described in Section 3.2 | ||
54 | Largest possible color excess | |
55 | catalog | Flag indicating whether the object is part of the catalog (C) |
or has been clipped/rejected (R) |
Only a portion of this table is shown here to demonstrate its form and content. A machine-readable version of the full table is available.
Table 9. Description of match.dat
Column # | Label | Explanation |
---|---|---|
1 | name | Identifier from the literature sample |
2 | id | Identifier used through the K18a data |
3 | sR.A. | Literature sample right ascension (J2000.0) |
4 | sDecl. | Literature sample declination (J2000.0) |
5 | R.A. | Literature sample right ascension (J2000.0) |
6 | Decl. | Literature sample declination (J2000.0) |
7 | sP | Literature sample period |
8 | PrP1 | Period in the rP1 band (determined with SigSpec) |
9 | sample | Literature sample identifier; DIR for the DIRECT sample (Stanek et al. 1998), |
R12 for the Riess et al. (2012) sample, | ||
V06 for the Vilardell et al. (2006) sample, | ||
and WEC for the WECAPP sample (Fliri et al. 2006) | ||
10 | catalog | Flag indicating whether the object is part of the catalog (C) |
or has been clipped/rejected (R) |
Only a portion of this table is shown here to demonstrate its form and content. A machine-readable version of the full table is available.
Download table as: DataTypeset image
Table 10. Description of match-K13.dat
Column # | Label | Explanation |
---|---|---|
1 | name | Identifier from the K13 sample |
2 | id | id > 0: identifier used through the K18a data |
id = 0: rejected due to the first cut criterion (see Section 3) | ||
id = −1: masked area in K18a | ||
3 | flag | Flag indicating whether the object is part of the catalog (C) |
or has been clipped/rejected (R) | ||
4 | rK13,P1,0 | K13 extinction-corrected rP1-band magnitude |
5 | iK13,P1,0 | K13 extinction-corrected iP1-band magnitude |
6 | PrP1 | K13 period in the rP1 band (determined with SigSpec) |
7 | type | K13 Cepheid type |
8 | rP1,0 | K18a extinction-corrected rP1-band magnitude |
9 | iP1,0 | K18a extinction-corrected iP1-band magnitude |
10 | PrP1 | K18a period in the rP1 band (determined with SigSpec) |
11 | type | K18a Cepheid type (see Section 3.3) |
Only a portion of this table is shown here to demonstrate its form and content. A machine-readable version of the full table is available.
Download table as: DataTypeset image
Table 11. Description of Light-curve Tables in the .tar.gz Package (FOLDER/cep_cut_FILTER_ID.dat)
Column # | Label | Explanation |
---|---|---|
1 | date | Modified Heliocentric Julian Date |
2 | mag | Magnitude |
3 | magerr | Magnitude error |
4 | nexp | Number of exposures contributing to the visit stack (see Section 2.2) |
5 | t | Total exposure time |
6 | sc | Skycell identifier (see Figures 27–29) |
Note. FILTER: r or i or g; FOLDER: can be lc_FILTER or lc_cut_FILTER; ID: K18a identifier. All of the light-curve tables are available in the .tar.gz package in the Astronomical Journal. The metadata header for each file is shown here for guidance regarding its form and content.
Download table as: ASCIITypeset image
In addition to the K13 cross-match table mentioned in the previous subsection, we also provide a cross-matching table for the Vilardell et al. (2006) sample with 416 Cepheids, the DIRECT sample with 332 Cepheids (Kaluzny et al. 1998, 1999; Stanek et al. 1998, 1999; Mochejska et al. 1999; Bonanos et al. 2003), the WECAPP sample with 126 Cepheids (Fliri et al. 2006), and the Riess et al. (2012) sample with 68 Cepheids. Due to the masking discussed in Section 2.1, we cannot match all literature Cepheids. With a matching radius of 2'', we match 349 Cepheids of the Vilardell et al. (2006) sample of 416 Cepheids. Eight of the Vilardell et al. (2006) sample have two matches within the 2'' radius. There are 277 Cepheids in the K18a sample, and 72 are clipped. Of the 332 DIRECT Cepheids, we match 316. There are 14 Cepheids with two matches inside the 2'' matching radius, 267 are in the K18a sample, and 49 are clipped. Of the 126 WECAPP Cepheids, we match 64. Thirty are in the K18a sample while 34 are clipped, and two Cepheids have two matches inside the 2'' matching radius. We match 61 of the 68 Riess et al. (2012) Cepheids. Three have two matches within the 2'' radius, and 55 are in K18a, while nine are clipped. In Figure 39, we show the comparison of the literature periods with those that are in our K18a sample (i.e., the clipped Cepheids are not included). The Vilardell et al. (2006) sample and the WECAPP sample have the most epochs of the literature samples, and therefore the periods match very well to our K18a periods. Since our K18a sample covers a larger baseline, the long periods are determined more precisely than in the literature samples.
Download figure:
Standard image High-resolution image