The Astrophysical Journal, 757:133 (14pp), 2012 October 1 doi:10.1088/0004-637X/757/2/133
C
2012. The American Astronomical Society. All rights reserved. Printed in the U.S.A.
THREE NEW ECLIPSING WHITE-DWARF–M-DWARF BINARIES DISCOVERED IN A SEARCH FOR
TRANSITING PLANETS AROUND M-DWARFS
Nicholas M. Law
1,11
, Adam L. Kraus
2,12
, Rachel Street
3
, Benjamin J. Fulton
3
, Lynne A. Hillenbrand
4
,
Avi Shporer
3,5
, Tim Lister
3
, Christoph Baranec
4
, Joshua S. Bloom
6
, Khanh Bui
4
, Mahesh P. Burse
7
,
S. Bradley Cenko
6
,H.K.Das
7
, Jack. T. C. Davis
4
, Richard G. Dekany
4
, Alexei V. Filippenko
6
, Mansi M. Kasliwal
8
,
S. R. Kulkarni
4
, Peter Nugent
9
, Eran O. Ofek
4
, Dovi Poznanski
10
, Robert M. Quimby
4
, A. N. Ramaprakash
7
,
Reed Riddle
4
, Jeffrey M. Silverman
6
, Suresh Sivanandam
1,11
, and Shriharsh P. Tendulkar
4
1
Dunlap Institute for Astronomy and Astrophysics, University of Toronto, 50 St. George St., Toronto, ON M5S 3H4, Canada
2
Institute for Astronomy, University of Hawaii, 2680 Woodlawn Drive, Honolulu, HI 96822, USA
3
Las Cumbres Observatory Global Telescope Network, Inc., 6740 Cortona Dr. Suite 102, Santa Barbara, CA 93117, USA
4
Cahill Center for Astrophysics, California Institute of Technology, Pasadena, CA 91125, USA
5
Department of Physics, Broida Hall, University of California, Santa Barbara, CA 93106, USA
6
Department of Astronomy, University of California, Berkeley, CA 94720-3411, USA
7
Inter-University Centre for Astronomy and Astrophysics, Ganeshkhind, Pune-411007, India
8
Carnegie Institution for Science, 813 Santa Barbara St., Pasadena, CA 91101, USA
9
Computational Cosmology Center, Lawrence Berkeley National Laboratory, 1 Cyclotron Road, Berkeley, CA 94720, USA
10
School of Physics and Astronomy, Tel-Aviv University, Tel Aviv 69978, Israel
Received 2011 December 2; accepted 2012 July 30; published 2012 September 11
ABSTRACT
We present three new eclipsing white-dwarf/M-dwarf binary systems discovered during a search for transiting
planets around M-dwarfs. Unlike most known eclipsing systems of this type, the optical and infrared emission is
dominated by the M-dwarf components, and the systems have optical colors and discovery light curves consistent
with being Jupiter-radius transiting planets around early M-dwarfs. We detail the PTF/M-dwarf transiting planet
survey, part of the Palomar Transient Factory (PTF). We present a graphics processing unit (GPU)-based box-
least-squares search for transits that runs approximately 8 × faster than similar algorithms implemented on general
purpose systems. For the discovered systems, we decompose low-resolution spectra of the systems into white-dwarf
and M-dwarf components, and use radial velocity measurements and cooling models to estimate masses and radii
for the white dwarfs. The systems are compact, with periods between 0.35 and 0.45 days and semimajor axes of
approximately 2 R
(0.01 AU). The M-dwarfs have masses of approximately 0.35 M
, and the white dwarfs have
hydrogen-rich atmospheres with temperatures of around 8000 K and have masses of approximately 0.5 M
.Weuse
the Robo-AO laser guide star adaptive optics system to tentatively identify one of the objects as a triple system. We
also use high-cadence photometry to put an upper limit on the white-dwarf radius of 0.025 R
(95% confidence)
in one of the systems. Accounting for our detection efficiency and geometric factors, we estimate that 0.08%
+0.10%
−0.05%
(90% confidence) of M-dwarfs are in these short-period, post-common-envelope white-dwarf/M-dwarf binaries
where the optical light is dominated by the M-dwarf. The lack of detections at shorter periods, despite near-100%
detection efficiency for such systems, suggests that binaries including these relatively low-temperature white dwarfs
are preferentially found at relatively large orbital radii. Similar eclipsing binary systems can have arbitrarily small
eclipse depths in red bands and generate plausible small-planet-transit light curves. As such, these systems are a
source of false positives for M-dwarf transiting planet searches. We present several ways to rapidly distinguish
these binaries from transiting planet systems.
Key words: binaries: eclipsing – methods: data analysis – planets and satellites: detection – stars: low-mass –
techniques: photometric – white dwarfs
Online-only material: color figures
1. INTRODUCTION
Large numbers of non-eclipsing white-dwarf/main-sequence
binaries have been discovered in the Sloan Digital Sky Survey
(SDSS) and other surveys (e.g., Rebassa-Mansergas et al. 2012;
Bianchi et al. 2007, and references therein). For low-mass stars
in particular there is a bridge in color between white dwarfs
and M-dwarfs. The bridge is interpreted as being due to rare
white-dwarf/M-dwarf binaries, at a ratio with respect to single
stars of ∼1:2300 (Smol
ˇ
ci
´
cetal.2004).
White-dwarf/M-dwarf eclipsing systems are much rarer, and
almost all have been discovered by searching for white dwarfs
11
Dunlap Fellow.
12
Hubble Fellow.
displaying very deep eclipses of up to several magnitudes (e.g.,
Nelson & Young 1970; Green et al. 1978; Fulbright et al. 1993;
Bruch & Diaz 1998; Drake et al. 2010). These searches find
systems containing relatively hot (>12,000 K) white dwarfs
and mid-to-late M-dwarfs. The discovery rate of these systems
(e.g., Drake et al. 2010; Parsons et al. 2011, 2012) is increasing
with the advent of large sky surveys. These binaries survived
the common-envelope phase of their evolution and many will
become cataclysmic variables (e.g., Nebot G
´
omez-Mor
´
an et al.
2011), and so the properties and number statistics of these
systems can provide windows into two important areas of stellar
evolution. Precision measurements of the systems allow the
determination of the masses and radii of two types of stars for
which there are relatively few measurements (Nebot G
´
omez-
Mor
´
an et al. 2009; Kraus et al. 2011; Pyrzas et al. 2012,
1
The Astrophysical Journal, 757:133 (14pp), 2012 October 1 Law et al.
2009), along with a verification of close binary evolution models
(Pyrzas et al. 2009).
In this paper, we present three eclipsing white-dwarf/
M-dwarf systems discovered during the PTF/M-dwarfs search
for transiting planets around M-dwarfs. In contrast to most
known eclipsing systems of this type, the systems detected in
this survey have optical and infrared emission dominated by
the M-dwarf component and contain relatively low-temperature
(8000 K) white dwarfs and relatively early M-dwarfs. The shape
of the light curves of the detected systems is similar to that ex-
pected for transiting giant planets around M-dwarfs, in particular
in having a flat-bottomed eclipse with a depth of 1%–20% in
red optical bands.
The PTF/M-dwarfs survey (Law et al. 2011) is a search
for transiting planets around 100,000 M-dwarfs. The survey is
performed with the Palomar Transient Factory (PTF) camera
(Rahmer et al. 2008; Law et al. 2009a, 2010a) on the 48 inch
Samuel Oschin telescope at Palomar Observatory and is a
Key Project of the PTF (Law et al. 2009a; Rau et al. 2009).
The PTF/M-dwarfs survey is designed to complement other
M-dwarf transiting planet surveys such as MEarth (e.g.,
Charbonneau et al. 2009; Irwin et al. 2010), the WFCam transit
survey (Sip
˝
ocz et al. 2011), and the M-dwarfs in the Kepler
mission target list (Borucki et al. 2011), by covering a much
larger number of M-dwarfs at somewhat lower sensitivity. The
survey achieves photometric precisions of a few percent for
∼100,000 targets, and few-millimag precision around a subset
of ∼10,000 M-dwarfs. These systems offer much larger transit
depths compared to solar-type stars, while their very red col-
ors compared to most other stars in the field greatly reduce the
probability of a blended eclipse producing a difficult-to-detect
transit false positive.
The three eclipsing systems presented here were originally
detected as Jupiter-sized exoplanet candidates during the first
year of operations of the survey. Followup of the candidates
showed large color changes during eclipse and very large radial
velocity (RV) signals, suggesting a hidden hot companion. In
this paper, we detail the properties of these eclipsing white-
dwarf/M-dwarf systems and explore ways to distinguish them
from true planetary transits.
The paper is organized as follows. In Section 2, we describe
the PTF/M-dwarfs survey, its precision photometry methods,
and its target detection strategies, including a new method of
performing a box-least-squares (BLS) transit search in parallel
on Graphics Processing Unit (GPU) hardware. Section 3 details
the three new eclipsing white-dwarf/M-dwarf systems and
describes follow-up photometric, low-resolution spectroscopic,
and RV observations, which are used to determine masses
and radii for the system components in Section 4. Section 5
determines the frequency of eclipsing binaries such as these and
discusses ways to distinguish them from transiting exoplanets.
2. THE PTF/M-DWARFS SURVEY
The PTF/M-dwarfs project consists of a transiting planet sur-
vey on the robotic 48 inch Samuel Oschin telescope (hereafter
P48), photometric follow-up using the Palomar 60 inch tele-
scope, the Byrne Observatory at Sedgwick Reserve (hereafter
BOS), the LCOGT Faulkes-North and Faulkes-South telescopes
(FTN and FTS), and RV follow-up with the High-Resolution
Echelle Spectrometer (HIRES; Vogt et al. 1994) instrument on
Keck I.
The8deg
2
camera (with 7.26 useful square degrees) on the
P48 telescope allows the survey to cover ≈3000 M-dwarfs in
Tab le 1
The Specifications of the PTF Camera and the PTF/M-dwarfs Survey
P48 PTF camera specifications
Telescope Palomar 48 inch (1.2 m) Samuel Oschin
Camera field dimensions 3.
◦
50 ×2.
◦
31
Camera field of view 8.07 deg
2
Light-sensitive area 7.26 deg
2
Image quality 2.0 arcsec FWHM in median seeing
Filters g
and Mould-R; other bands available
CCD specs 2 K × 4KMIT/LL 3 edge-butted CCDs
Plate scale 1.
01 pixel
−1
Readout noise <12 e
−
Readout speed 35 s, entire 100 MPix array
PTF/M-dwarfs survey characteristics
Targets Late-K, M, and L dwarfs with m
R
< 18
Survey sky area 29 deg
2
every 2 months
Target locations 20
◦
< Galactic latitude < 35
◦
Targets covered ≈12,000 every 2 months
Observations per night 5 hr
Exposure time 60 s
Cadence 15–25 minutes
Observation length 1–3 months
Efficiency 66% open shutter (slew during readout)
Saturation level m
R
≈ 14 (seeing dependant, in 60 s)
Sensitivity (median) m
R
≈ 21 in 60 s, 5σ
m
g
≈ 21.3in60s,5σ
Photometric stability 3 mmag (brightest targets)
10% (faintest targets)
Followup Photometric: Palomar 60 inch, FTN/S, BOS
Low-res spectroscopic: Lick Shane-3 m
Radial velocity: Keck I/HIRES
each pointing (Figure 1), for a total of around 100,000 targets
per year at Galactic latitudes of 20
◦
–35
◦
. The PTF/M-dwarfs
survey typically observes several fields with an approximately
20 minute cadence for 4–5 hr per night. Individual fields are
typically observed for several months, and observations
are performed throughout the year. All PTF/M-dwarfs data are
acquired with a 60 s exposure time in the PTF camera’s Mould-R
filter (similar to the SDSS r filter). Data taking is interleaved with
the PTF supernova survey which generally operates on 1–3 days
cadences, and so scheduling constraints lead to a variety of
final cadences for PTF/M-dwarfs fields. Table 1 summarizes
the specifications of the PTF camera and the PTF/M-dwarfs
survey.
The PTF/M-dwarfs survey is sensitive to equal-mass binaries
around all M-dwarfs in the survey fields brighter than m
R
≈ 20.
In practice, the mass and radius determination and follow-up
observation of the faintest systems is extremely challenging,
so we impose a discovery magnitude limit of m
R
≈ 18. At
that magnitude, the typical P48 data photometric precision is a
few percent per data point, allowing immediate high-precision
constraints on the system properties in discovery data. Given
the saturation and faint cutoff limits of the survey, its effective
distance ranges are 200–1300 pc for M0 dwarfs, 50–290 pc for
M5 dwarfs, and 10–70 pc for M9 dwarfs.
2.1. P48 Data Reduction
Reduction of the PTF/M-dwarfs data is performed in two
steps: the standard PTF real-time data reduction software first
calibrates the images, and then a custom pipeline performs
source extraction, association, and precision photometry.
2
The Astrophysical Journal, 757:133 (14pp), 2012 October 1 Law et al.
Figure 1. PTF camera image of a survey field centered at α = 17:28, δ = +57:22, and covering 3.
◦
50 × 2.
◦
31. The highlighted points show the 2851 stars with
photometrically estimated spectral types later than K4 and photometric stability better than 5%. The colors of the points correspond to the stellar temperatures, with
K4 as yellow and late-M-dwarfs as red. The pipeline has removed stars with possible photometric precision problems such as proximity to a bright star or a bright
ghost image. North is up and east is to the right.
(A color version of this figure is available in the online journal.)
Immediately after observations, PTF data are transferred
to Lawrence Berkeley National Laboratory where cross-talk
corrections are applied to each chip, standard bias/overscan
subtraction is performed, and a superflat based on recently
acquired data is applied.
After the calibrated data are transferred to the Dunlap Institute
for Astronomy and Astrophysics at University of Toronto, the
PTF/M-dwarfs pipeline extracts sources from the calibrated
images, produces an optimal photometric solution, associates
sources in images taken at different times, and applies a range
of eclipse-detection algorithms to the resulting light curves. We
summarize the system here; the pipeline is described in more
detail in Law et al. (2011).
Initial source extraction is performed on individual CCD
chip images by SExtractor (Bertin & Arnouts 1996)using
radius-optimized aperture photometry with a locally optimized
background. The extracted sources are filtered to remove those
close to bad pixels, diffraction spikes from bright stars, and those
that may be affected by nearby sources. Heliocentric Julian dates
(HJD) are used for time measurements throughout the pipeline.
The photometric zero points for each epoch are initially esti-
mated based on either SDSS or USNO-B1 photometry for bright
stars in the field. The pipeline then optimizes the zero point of
each epoch to minimize the median photometric variability of
all the remaining sources. This first optimization typically im-
proves the long-term photometric stability to below the percent
level. The pipeline then filters the generated light curves, search-
ing for epochs that produce anomalous photometry for a large
fraction of the sources; those epochs are usually those affected
by clouds, moonlight, or some other effect that varies across the
images. Typically 0%–2% of epochs are flagged by this process
and are removed from further consideration. A second iteration
of variable-source removal and zero-point optimization is then
performed. The final zero points are applied to each light curve,
along with flags for poor conditions, nearby sources that could
cause confusion, bad pixels, and other problems that could af-
fect the photometry. Running on a 2.5 GHz quad-core desktop
computer the pipeline processes a 300 epoch set of 11 chips
(54 GB of image data) in less than 24 hr.
The pipeline typically achieves a photometric stability of
3–5 mmag over periods of months (Figure 2). The photometric
precision is photon limited for all sources fainter than m
R
≈ 16,
except in regions of crowding or nebulosity. The pipeline has
been used for several PTF programs such as the open cluster
rotation project (Ag
¨
ueros et al. 2011).
2.2. Eclipsing Binary Detection
The pipeline produces light curves for 25,000–100,000 stars
per field. The large number of light curves makes the pro-
cessing time involved in searching for eclipses an important
consideration. An initial cut is made on the basis of the esti-
mated spectral type of the source, based on photometry from the
USNO-B1, Two Micron All Sky Survey (2MASS), and (where
3
The Astrophysical Journal, 757:133 (14pp), 2012 October 1 Law et al.
0.001
0.01
0.1
1
14 15 16 17 18 19 20
RMS stability / magnitudes
R-magnitude
Figure 2. Photometric stability achieved in a typical PTF/M-dwarfs field in
140 epochs spanning 92 days of observations. Each point corresponds to the
root-mean-square (rms) variability of one of the 23,713 stars in this field with
magnitude 14 <m
R
< 20. For clarity, 50% of the fainter stars have been
removed from this plot. The best stability achieved in this field is 4 mmag, with
some CCD chips and sky regions having somewhat lower stability at 5–7 mmag.
The few stars with very high variability compared to others at similar magnitudes
are astrophysically varying sources such as eclipsing binaries, RR Lyrae stars,
and other variables.
(A color version of this figure is available in the online journal.)
available) SDSS data (Monet et al. 2003; Skrutskie et al. 2006;
York et al. 2000). The photometric data are fit to updated ver-
sions of the spectral energy distributions (SEDs) given in Kraus
& Hillenbrand (2007), yielding an accuracy of approximately 1
subclass. Sources with late-K, M, or L estimated spectral types
are passed to our eclipse-search algorithms.
2.2.1. High-variability Source Searches
Many eclipsing binaries have eclipse signals in the tens-of-
percent range as well as large eclipse duty cycles. These systems
have significantly increased photometric variability compared to
nearby stars of similar magnitude, and so a simple variability
search can rapidly find them (Figure 2). We estimate the value
and scatter of the locally expected photometric stability as a
function of stellar magnitude in 0.1 mag bins, using a sigma-
clipped average of the stars detected on each chip. Objects
that show a more than 2σ increased variability compared to
the ensemble expectation are flagged for further review. Of the
systems presented here, only PTFEB11.441 was detected in this
manner; the photometric variations from the smaller eclipse
depths of the other systems required more computationally
intensive algorithms.
2.2.2. A Parallel Eclipse Search Using Graphics Processing Units
Systems with smaller eclipse depths and/or longer periods
and reduced duty cycles may have only slightly increased
photometric variability, necessitating a more sensitive search.
We use a standard BLS (Kov
´
acs et al. 2002; Tingley 2003)
algorithm to phase the light curves at all possible periods and
search for a transit-like signal.
This algorithm requires the testing of thousands of periods and
transit phases and is thus computationally expensive. However,
the problem is easily parallelizable as an arbitrary number of
light curves and periods can be tested simultaneously. We take
advantage of this by implementing the BLS algorithm on a GPU
that can perform hundreds of computations in parallel.
Our BLS search is run on an NVIDIA Tesla C2050 GPU
that contains 448 cores operating at 1.15 GHz, for a total of
1.03 Tflops in single-precision floating point arithmetic. The
BLS search is coded in CUDA and is called from the PyCUDA
python module (Kl
¨
ockner et al. 2009).
Perhaps the simplest parallel-processing technique for the
BLS algorithm is to have each GPU thread perform all the
calculations for one source; in this way several hundred sources
could be tested simultaneously. However, the BLS algorithm
requires at least one access to the full light-curve data at each
test period. The per-thread memory in typical GPUs is too
small to contain a full light curve, necessitating frequent calls
to the GPU global memory. These calls are slow, even when
synchronized across threads, greatly limiting the speed of the
GPU implementation.
Instead, we consider sources sequentially, and simultaneously
test hundreds of different periods on a single source. The
processing of each light curve proceeds as follows. First, the
light-curve data are read into fast shared memory by each thread
block (with memory accesses coalesced for speed). Next, each
thread picks an untested period, phases the light curve at that
period and then tests a range of durations and phases.
In testing with typical PTF/M-dwarfs data, our GPU BLS
algorithm operates approximately eight times faster than a multi-
threaded program executing on all cores of an Intel Core-2 Quad
Core CPU running at 2.50 GHz. With the GPU approximately
10,000 light curves can be fully searched for transit or eclipse
events each hour.
3. DISCOVERIES AND FOLLOW-UP OBSERVATIONS
We discovered three eclipsing M-dwarf/white dwarf binary
systems in a search of ≈45,000 M-dwarfs from the first year
of PTF/M-dwarf operations. Each system was detected as a
high-confidence planet candidate, with a 5%–20% depth flat-
bottomed eclipse in the R band with duration consistent with a
transiting Jupiter-radius planet. Images of the field around each
target are shown in Figure 3, and the P48 detection light curves
of the systems are shown in Figure 4.
The PTF names of the three detected sources are PTF1
J004546.0+415030.0, PTF1 J015256.6+384413.4, and PTF1
J015524.7+373153.8. For brevity, we hereafter refer to the
sources by their PTF/M-dwarfs survey internal names, which
are based on their decimal right-ascension coordinates:
PTFEB11.441, PTFEB28.235, and PTFEB28.852, respectively.
3.1. Follow-up Measurements
After the systems were detected in PTF data as planet
candidates we followed the standard PTF/M-dwarfs follow-
up strategy: high-cadence multi-color photometry with the
Palomar 60 inch, BOS, and FTN, along with RV observations
with the HIRES spectrograph on Keck I. The multi-color
photometry (Section 3.1.1) rapidly revealed that these objects
had a strongly varying eclipse depth with color (Figures 5 and 6),
suggesting we were seeing eclipses between two self-luminous
objects with different temperatures. Low-resolution spectra
(Section 3.1.4) were sufficient to immediately confirm a white-
dwarf component in one system (PTFEB11.441), but required
detailed modeling to recover the white-dwarf components in
the other two systems (Section 4.1). RV observations were
scheduled to allow determinations of the white-dwarf mass.
4
The Astrophysical Journal, 757:133 (14pp), 2012 October 1 Law et al.
Figure 3. Images of the newly discovered systems taken with the PTF camera (left-to-right: PTFEB11.441, PTFEB28.235, and PTFEB28.852). North is up and east
is to the left; each image shows a 400×400 arcsec cutout from the 3.
◦
50 ×2.
◦
31 PTF camera field.
(A color version of this figure is available in the online journal.)
16.20
16.30
16.40
16.50
16.60
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35
R-magnitude
Time / days
16.50
16.60
16.70
16.80
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35
R-magnitude
Time / days
17.00
17.10
17.20
17.30
17.40
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45
R-magnitude
Time / days
Figure 4. Discovery P48 light curves for the three eclipsing systems (top-to-
bottom: PTFEB11.441, PTFEB28.235, and PTFEB28.852). The colors of the
points correspond to the time the data point was measured: black are the oldest
points and bright yellow are the newest. PTFEB11.441 was observed during a
high cadence run targeted at M31, hence the large density of points all taken on
a single night.
(A color version of this figure is available in the online journal.)
3.1.1. High-cadence, Multi-color Photometry with BOS
Multi-color photometry data for each of the targets were gath-
ered with the RC Optics 0.8 m telescope at Byrne Observatory
at Sedgwick reserve near Santa Ynez, CA. The BOS telescope
is equipped with a Santa Barbara Instrument Group STL-6303E
0.60
0.65
0.70
0.75
0.80
0.85
0.90
0.95
1.00
1.05
0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08
Fractional flux
Time / days
z filter
g filter
Figure 5. Faulkes-North (z filter) and BOS (g filter) follow-up photometry of
PTFEB28.852. The eclipse is detected at high significance in both bands, but is
approximately 10× deeper in the g filter, suggesting the secondary eclipse of a
hot body.
(A color version of this figure is available in the online journal.)
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
1.10
1.20
1.30
0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16 0.18
Fractional flux
Time / days
g (480nm)
r (620nm)
i (760nm)
Figure 6. Dunlap Institute Arctic Telescope photometry of PTFEB11.441,
showing the varying eclipse depth with wavelength. The g, r,andi filters (with
midpoints shown in the figure) were alternated in sequence to produce a nearly
simultaneous multi-color light curve. To guide the eye, each color has a fit to
the light curve using the system’s eclipse parameters derived in Section 4.5.
(A color version of this figure is available in the online journal.)
5
