Candidate H{\alpha} emission and absorption line sources in the Galactic Bulge Survey

4.5 MB·

by  T. Wevers

MNRAS000,1–12(2016) Preprint12January2017 CompiledusingMNRASLATEXstylefilev3.0
Candidate Hα emission and absorption line sources in the
Galactic Bulge Survey
T. Wevers1(cid:63), P. G. Jonker2,1, G. Nelemans1,3, M. A. P. Torres2,1, P. J. Groot1,
D. Steeghs4, T. J. Maccarone5, R. I. Hynes6, C. Heinke7, C. Britt6,8
1Department of Astrophysics/IMAPP, Radboud University, P.O. Box 9010, 6500 GL Nijmegen, The Netherlands
2SRON, Netherlands Institute for Space Research, Sorbonnelaan 2, 3584 CA Utrecht, The Netherlands
7 3Institute for Astronomy, KU Leuven, Celestijnenlaan 200D, 3001 Leuven, Belgium
1 4Department of Physics, University of Warwick, Coventry CV4 7AL, UK
0 5Department of Physics and Astronomy, Texas Tech University, Box 41051, Lubbock, TX 79409-1051, USA
2 6Department of Physics and Astronomy, Louisiana State University, Baton Rouge, LA 70803-4001, USA
7Department of Physics, University of Alberta, CCLS 4-183, Edmonton, AB T6G 2E1, Canada
n
8Department of Physics and Astronomy, Michigan State University, 5678 Wilson Road, Lansing, MI 48824, USA
a
J

0
Accepted2016November28.Received2016November25;inoriginalform2016October14
1

]
E ABSTRACT
H WepresentacatalogueofcandidateHαemissionandabsorptionlinesourcesandblue
objects in the Galactic Bulge Survey (GBS) region. We use a point source catalogue
.
h oftheGBSfields(twostripsof(l×b)=(6◦×1◦)centredatb=1.5◦ aboveandbelow
p the Galactic centre), covering the magnitude range 16 (cid:54) r(cid:48) (cid:54) 22.5. We utilize (r(cid:48)–i(cid:48),
-
o r(cid:48)–Hα) colour-colour diagrams to select Hα emission and absorption line candidates,
r and also identify blue objects (compared to field stars) using the r(cid:48)–i(cid:48) colour index.
t
s We identify 1337 Hα emission line candidates and 336 Hα absorption line candidates.
a These catalogues likely contain a plethora of sources, ranging from active (binary)
[ stars,early-typeemissionlineobjects,cataclysmicvariables(CVs)andlow-massX-ray

1 binaries (LMXBs) to background active galactic nuclei (AGN). The 389 blue objects
v we identify are likely systems containing a compact object, such as CVs, planetary
9 nebulae and LMXBs. Hot subluminous dwarfs (sdO/B stars) are also expected to
6 be found as blue outliers. Crossmatching our outliers with the GBS X-ray catalogue
7 yields sixteen sources, including seven (magnetic) CVs and one qLMXB candidate
2 amongtheemissionlinecandidates,andonebackgroundAGNfortheabsorptionline
0 candidates. One of the blue outliers is a high state AM CVn system. Spectroscopic
.
1 observations combined with the multi-wavelength coverage of this area, including X-
0 ray, ultraviolet and (time-resolved) optical and infrared observations, can be used to
7 further constrain the nature of individual sources.
1
Key words: Galaxy: bulge – stars: emission line – cataclysmic variables – binaries:
:
v symbiotic – white dwarfs – galaxies: active
i
X
r
a
1 INTRODUCTION (Schwopeetal.2000;Rattietal.2012),orthenatureofthe
compactobjectand/ordonorstar(Steeghs&Casares2002;
Thepresenceofanionisingradiationfieldcanleadtohydro-
van Spaandonk et al. 2010; Casares 2015, 2016).
gen emission lines, while the presence of neutral hydrogen
can result in absorption features in the optical spectrum of Historically, large scale photometric Hα surveys with
astronomical objects. From the properties of the H Balmer modest spatial resolution focussed on extended sources of
linesonecaninfercharacteristicsofthesystemunderstudy. emission to study star-forming regions, galaxy groups and
For example, the properties of single and/or double-peaked supernova remnants (Davies et al. 1976, and references
lines can allow us to infer geometrical properties (Horne & there-in). More recently, higher resolution surveys such as
Marsh 1986), the presence or absence of an accretion disc theINTPhotometricHαsurvey(IPHAS;Drewetal.2005,
Barentsen et al. 2014) have focussed on the Galactic plane
touncoverandstudycompactsourcesofemission,typically
(cid:63) Email:[email protected] associatedwithvariousstagesofstellarevolution.Themost
(cid:13)c 2016TheAuthors
2 Wevers et al.
noteworthy Hα survey covering the Galactic bulge is the 2016a). This catalogue consists of optical photometry ob-
photographic Anglo-Australian Observatory UK Schmidt tained using the Mosaic-2 camera on the Victor M. Blanco
Telescope Supercosmos Hα Survey (Parker et al. 2005), go- telescope, located at the Cerro Tololo Inter-American Ob-
ing down to R∼19.5 mag in the latitude range |b| (cid:54) 10◦. servatory (CTIO), in three filters: r(cid:48), i(cid:48) and Hα. The areas
Currently ongoing is the VST Photometric Hα Survey of coveredarecentredonb=1.5◦aboveandbelowtheGalactic
the Southern Galactic Plane and Bulge (VPHAS+, Drew centre,andconsistoftwostripsspanning(l×b)=(6◦×1◦).
etal.2014)whichwillcovertheGalacticbulgeandplanein Intotal,64fields,eachconsistingof8frames,wereobserved
5 filters down to at least 20th magnitude. twice.Oneofthetwoexposureswasoffsetby∼1.2arcminin
The analysis of colour-colour diagrams (CCDs) to right ascension and declination to fill the gaps between the
search for Hα emission line objects has been introduced by detectors.Thereforewehave1024observedframesintotal.
the IPHAS collaboration. Witham et al. (2008) present a Themean5σ limitingdepthoftheobservationsisr(cid:48)=22.5,
method and first results of this effort. A variety of source i(cid:48)=21.1 mag. The point source catalogue includes objects
classes, including CVs (see also Witham et al. 2006, 2007), thathavebeendetectedwithasignal-to-noiseratioofmore
early-type emission line stars (Corradi et al. 2008, 2010; than 5 in all bands. For more details about the optical cat-
Drew et al. 2008), active late-type stars, young stellar ob- alogue we refer the reader to Wevers et al. (2016a).
jects (Vink et al. 2008) and planetary nebulae (Viironen
et al. 2009; Sabin et al. 2010) have been identified. For an
exampleoftheexpectedsourceclassesandtheirlocationin
2.1.1 Global photometric calibration
the CCD we refer to fig. 1 in Corradi et al. (2008).
ThepresenceofanultravioletorX-rayphotonfieldcan The observations consist of two strips of overlapping fields
ionise hydrogen atoms in its direct environment, and hence aboveandbelowtheGalacticCentre,respectively(seefig.1
leadtoanHαemissionlineintheopticalspectrum.Binary inWeversetal.2016a).Weusetheoverlapbetweentheseob-
systemscontainingacompactobject,suchasawhitedwarf servations to apply a photometric calibration with the goal
(WD), neutron star (NS) or black hole (BH), are examples of getting all the photometry on the same absolute scale.
of (transient) Hα emitters, with the strength and width of There is no overlap between northern and southern fields,
the emission line depending on the primary mass, mass ac- sowecalibratethemindependently.Tothisend,weusethe
cretion rate and inclination angle with respect to the line methoddevelopedbyGlazebrooketal.(1994)(seealsoBar-
of sight (Casares 2015). Other excitation mechanisms (e.g. entsen et al. 2014 for an application of this method to the
collisionalexcitationinthestellarcorona)canalsoexciteH IPHAS dataset).
atoms and induce spectral line emission. In short, the goal is to minimize the magnitude offsets
InadditiontoHαemissionlineobjects,somesystemsshow between stars that are present on overlapping fields using
Hα in absorption. If the strength of this absorption line is a set of anchor fields for which the photometry is thought
stronger than that of a normal main sequence (MS) star, to be well determined. The reference fields are chosen on 2
it will appear as an outlier below the locus of stars in a photometric nights, namely MJD 53912 and 59315 (see ta-
CCD. For example, single H-rich (DA) WDs are known to ble 1 in Wevers et al. 2016a and Section 3.4). We denote
exhibit a very broad Hα absorption line. C-rich and S-type the magnitude offset between stars on overlapping fields as
asymptotic giant branch (AGB) stars have molecular ZrO ∆ =(cid:104)m −m (cid:105).Inordertokeepthesolutionfromdrifting
ij i j
absorption bands in their spectra that coincide with the lo- arbitrarilyfarfromthevaluesoftheanchorfields,thediffer-
cation of the Hα line (e.g. ZrOλ6456). These objects will ence in zeropoint values across the fields is also minimized.
hence also appear to have a deficit of flux in the Hα filter This problem can be solved as a linear least-squares prob-
relative to the r(cid:48)-band and appear as outliers to the locus lembecausethemagnitudesandthezeropointsarelinearly
ofobjectswhichdonotexhibitthesefeaturesintheirspec- related. Following Glazebrook et al. (1994) and setting the
trum. Late-type variable stars can cover a large range in weights w =1, we minimize the sum:
ij
colour space depending on the relative strength of molecu-
lar absorption bands (e.g. TiO, VO, ZrO) that are located
N N
in the r(cid:48), i(cid:48) and Hα filters. S =(cid:88)(cid:88)θ (∆ +a −a )2 (1)
ij ij i j
Inthiswork,weusetheopticalobservationsfromWev-
i=1j=1
ersetal.(2016a),takenaspartoftheGalacticBulgeSurvey
(GBS; Jonker et al. 2011, 2014), to search for sources with whereθij isanoverlapfunctionthatis1whenthereisover-
excess Hα emission and absorption signatures compared to lap and 0 otherwise, ai are the zeropoints to solve for and
normalMSstars.Wealsoidentifyblueoutlierswithrespect aj the zeropoints of overlapping fields to ai. N is the total
to field stars in the CCDs. The structure of this article is number of fields included in the least-squares problem.
as follows: Section 3 outlines the method used to identify Minimisingthesuminequation1byvaryingaiisequiv-
outliers,andinSection4wepresenttheresults.Wediscuss alent to solving δS =0, which yields the matrix equation
δai
our findings in Section 5 and summarise in Section 6.
N
(cid:88)
A a =b (2)
ij j i
j=1
2 DATA
where
2.1 Photometry
N
As the starting point of our work, we use the optical point (cid:88)
A =θ −δ θ (3)
ij ij ij jk
source catalogue covering the GBS fields (Wevers et al.
k=1
MNRAS000,1–12(2016)
Candidate Hα outliers towards the Galactic bulge 3
and
Table 1.Estimateofthedistancerangeandobservedcoloursof
N starswithdifferentspectraltypesinfieldS01(detector2).
(cid:88)
b = θ ∆ (4)
i ij ij
j=1 Sp.type Mr(cid:48) (r(cid:48)–i(cid:48))syn d(kpc) Ar(cid:48) (r(cid:48)–i(cid:48))obs
Wenowsolvetheleast-squaresproblembykeepingthe A0V 0.77 0 2.6–5 4.4–8.4 1.1–2.2
solutions of the anchor fields fixed, while the zeropoints of G0V 4.26 0.39 1.3–3.4 2.2–5.7 1.0–1.9
the other fields are allowed to vary to optimize the solution K0V 5.67 0.48 0.9–2.8 1.5–4.7 0.9–1.7
as a global photometric calibration. M0V 11.72 0.96 0.5–1.8 0.8–3.0 1.2–1.8
2.1.2 Colour-colour diagrams
(locatedslightlytotherightandbelowtheorangesynthetic
Before we move to the details of the selection criteria for
track) and the locus of reddened stars. We note that in the
outliers,wefirstintroducethenecessarytoolswewilluseto
CCD there is an offset between the synthetic track and the
find them: colour-colour diagrams and synthetic photome-
observedunreddenedMS.Thisoffsetshowsthattheunred-
try. We merge the nominal and offset observations in each
dened MS in this case is nevertheless slightly reddened to
filter (taken within minutes of each other), and create (r(cid:48)– about E(B−V)(cid:54)0.5.
i(cid:48),r(cid:48)–Hα) colour-colour diagrams. This gives us a total 512
ComparingtheobservedCCDandtheunreddenedsyn-
frameswhichformthebasisofourwork.Weexcludesources
thetic track reveals that our catalogue apparently contains
withsaturatedphotometryinouranalysis.Weusethesame
no unreddened stars of early spectral types. The observed
basic techniques as presented by the IPHAS collaboration
unreddened MS population typically contains only stars of
(Drew et al. 2005; Witham et al. 2006).
spectral type K0V or later. Taking the absolute magnitude
Weuseasetofsyntheticspectra(Pickles1998)tocreate
from Schmidt-Kaler (1982) and colour from Pecaut & Ma-
synthetic photometry for spectral types ranging from O5V
majek(2013),andthecolourtransformationgivenbyJester
to M5V using the CTIO filters1. We redden these spectra etal.(2005),wefindthataK0Vstarobservedatr(cid:48)=17(the
with increasing E(B−V) to estimate the colours of stars
typicalsaturationlimitoftheopticalcatalogue)islocatedat
at a range of reddening values (hence distances). We con-
adistanceof∼1kpc.Unreddenedstarswithaspectraltype
sider solar-metallicity MS and giant stars. The binning of
earlier than K0V (hence intrinsically brighter) and located
the spectra is sufficiently small (5˚A) that we can use them
within1kpcaresaturated.Thephotometricobservationsof
to compute synthetic photometry for our r(cid:48) and i(cid:48) filters
saturatedsourcesareunreliableandwediscardtheminour
as well as for the narrow-band Hα filter. We recompute the
analysis.
gridsofthesefilterprofilestomatchthebinningofthespec-
We use the absolute magnitudes from Schmidt-Kaler
tra,meaningthatforeachspectralbinwecomputethefilter
(1982)togetherwiththe3DreddeningmapfromSchultheis
transmissionvalueatthemidpointofthebin.Wedefinethe
et al. (2014) at (l, b) = (-2.8, -1.8) to estimate the distance
synthetic colours in the Vega system as
ranges for stars of different spectral types in our catalogue.
The3Dreddeningmapisconvertedtother(cid:48)-bandfollowing
Schlegel et al. (1998). In Table 1 we show the results for
(cid:82) (cid:82)
T1,λFλdλ T2,λFλdλ different spectral types. We use the distance modulus to
m1−m2 =−2.5log (cid:82) T F dλ+2.5log(cid:82) T F dλ estimate the observable distance range as:
1,λ λ,V 2,λ λ,V
(5)
log d (pc)=0.2×(r(cid:48)−Mr(cid:48) −Ar(cid:48)(d)+5) (6)
where the filter transmission profiles are labeled T , F is
the synthetic spectrum per spectral type and F x isλthe where Mr(cid:48) is the absolute magnitude, r(cid:48) the apparent (ob-
spectrum of Vega (Bohlin 2007). It is possible toλ,cVompute served) magnitude and Ar(cid:48) the extinction in the r(cid:48)-band
an upper limit to the r(cid:48)–Hα colour of any physical object obtained from the Schultheis et al. (2014) reddening map.
TheresultsofthiscalculationarevisualisedinFigure2.We
based on synthetic photometry of a pure Hα emission line
spectrum(Drewetal.2005).Theupperlimitisr(cid:48)–Hα=3.3 assumethatthesaturationlimitofourcatalogueisr(cid:48)=17,
and the limiting magnitude is r(cid:48)=22.5 (marked by dashed
fortheusedfiltercombinations,hencewediscardallobjects
thathaveanobservedr(cid:48)–Hαcolourabovethisvalue.Such horizontallines),givingrisetotherangesshowninTable1.
Usingoursyntheticphotometry,wecanalsoinfertherange
valuesareindicativeofabadcrossmatchbetweenthethree
ofr(cid:48)–i(cid:48) colourswhichdifferentspectraltypesoccupyinthe
optical bands, or detector artifacts.
In Figure 1 we show an example of a CCD of field S01 CCD, assuming that E(r(cid:48)–i(cid:48))=0.26×Ar(cid:48) (Schlegel et al.
1998).
(detector 2), centred at Galactic coordinates (l,b)=(–2.81,
We conclude that the locus of reddened stars at r(cid:48)–
–1.75). Observed colours are plotted in black, and overlaid
i(cid:48)∼1.4 consists of reddened MS stars with spectral type
are synthetic tracks for MS stars (orange squares and blue
earlierthanM0V.Theobjectslocatedbeyondr(cid:48)–i(cid:48)∼2are
triangleswithE(B−V)=0and1,respectively)andgiants
giants, as MS stars are too faint to be observable at those
(red diamonds, E(B−V)=2). The two main populations
reddening values.
of stars that can be identified are the unreddened MS stars
Webrieflynotethattherearetwooutliersaround(r(cid:48)–
i(cid:48), r(cid:48)–Hα)=(0.1,0). As we explained above, these can not
1 http://svo2.cab.inta-csic.es/svo/theory/fps3/index.php?mode beordinaryearly-typeMSstarsbecausetheywouldappear
=browse&gname=CTIO saturated in our observations.
MNRAS000,1–12(2016)
4 Wevers et al.
1.2
Early A0V reddening line
MS, E(B-V) = 0
MS, E(B-V) = 1
1.0
Giants, E(B-V) = 2 M5
0.8
)
g
a
0.6
m
M0
(

α
H
0.4
-

r0 K0
G0
0.2
0.0
A0
0.2
1 0 1 2 3 4 5
r - i (mag)
0 0
Figure1.TheblackpointsshowtheobservedcoloursofsourcesonfieldS01(detector2).Thedashedlinesarethespectralsequencesof
unreddenedMSstarsfromO5VtoM5V,withE(B−V)=0and1fororangesquaresandbluetriangles,respectively.Thereddiamonds
show colours of giants with spectral types ranging from O8III to M5III, reddened to E(B−V)=2. The green squares show the early
A-typeMSline,whichindicatesthelowerlimitforA0VMSstarsintheCCD.Thearrowindicatestheeffectofreddening∆E(B−V)=1.
Thetwooutliersmarkedbyblackstarsaround(0.1,0)areverybluecomparedtotheotherfieldstars,makingthempotentiallyinteresting
sources.
2.2 Spectroscopy consists of 1024 observations, observed on 8 nights, span-
ning a large range of stellar densities, dust extinctions and
A 500s spectrum of the optical counterpart to the X-ray
photometricuncertainties.Henceitisunavoidablethatany
source CX2 (Jonker et al. 2011) was taken on 2010 July
selectionmethodwillfailinsomecases.Wetakeaconserva-
8 with the ESO Faint Object Spectrograph and Camera
tive approach and prioritize minimising false positives over
(EFOSC2,Buzzonietal.1984)attheESONewTechnology
completeness. This implies that our catalogue will not be
Telescope (NTT). We used grism #13 combined with a 1
complete. Below we introduce the automatic identification
arcsec slit, resulting in a spectral resolution of R∼300 and
algorithm together with a list of criteria that must be ful-
awavelengthcoveragerangingfrom3700–9300˚A.Wedebi-
filled for the results to be deemed trustworthy. CCDs that
asedandcorrectedfortheCCDflatfieldresponse,andwave-
failtomeetthesecriteriahavebeenrejectedfromautomatic
lengthcalibrationwasperformedusingaHeArarclamp.We
processing and are instead inspected manually.
normalisedthespectrumbyfittingcubicsplinestothecon-
tinuum in molly. We set out to identify the outliers from the main fea-
turesthatarepresentintheCCD:theunreddenedMSand
reddened sequence. We make a distinction between these
two features and fit them independently. We identify Hα
3 OUTLIER IDENTIFICATION
emission line sources from both the unreddened and red-
Wedevisedaselectionmethodthatresultsintheautomatic dened loci of objects. Additionally we identify absorption
identificationofHαemissionandabsorptionlinecandidates line sources from the reddened locus, which should be lo-
and blue sources (all referred to as outliers). Our dataset cated below the main population in the CCD. Unreddened
MNRAS000,1–12(2016)
Candidate Hα outliers towards the Galactic bulge 5
denedMS,wecalculatetheslopeofourfitineachiteration.
16 If the slope of the fit is more shallow than the slope of the
A0V
G0V synthetic track between spectral types K5V and M5V, we
17 K0V deemourfitunsatisfactoryandapplyanadditionaliteration
M0V
18 (i.e. we force the fit upward). We iterate for a maximum of
5 times; CCDs for which the fit has not converged at that
g)19 point will be inspected manually for outliers. We also place
a
(m20 aconstraintonther(cid:48)−i(cid:48) colourofanunreddenedMSstar,
r0 to distinguish those sources from reddened stars with Hα
21
inemission.Thesesourcesmayoccupythesameparameter
22 space in the CCD. However, r(cid:48)–i(cid:48) increases for later spec-
traltypes.Weconservativelyestimatefromtheextentofthe
23 unreddened MS in our CCDs that the highest r(cid:48)–i(cid:48) colour
0 1 2 3 4 5 an unreddened late-type MS star can have is r(cid:48)–i(cid:48)=3.5.
d (kpc)
Once we have identified the unreddened MS, we deter-
Figure 2. Estimated distance ranges for MS stars of different mine the iteratively 4σ-clipped scatter around the fit. We
spectral types present in the optical GBS catalogue, assuming a define outliers as sources with an excess Hα emission con-
saturation limit of r(cid:48)=17 (upper horizontal dashed line) and a tribution, quantified as follows:
limitingmagnitudeofr(cid:48)=22.5(lowerhorizontaldashedline).See
(cid:113)
Section2.1.2fordetails. (r(cid:48)−Hα) −(r(cid:48)−Hα) (cid:62)C× σ2+σ2 (7)
obs fit s phot
Here σ represents the scatter of datapoints around the fit,
stars with strong absorption line features may overlap with s
andσ isthephotometricmeasurementerrorforther(cid:48)–
morereddenedobjects(andconverselyHαemittersfromthe phot
Hα colour index. C is a constant which we set to 4. An
reddened locus may overlap with the unreddened MS) and
exampleisshowninFigure3.Theresultingfitisoverplotted
cannotbedistinguishedbasedonaCCDalone.Includinga
as the upper solid line, while the dashed lines indicate the
colour-magnitudediagramintheanalysismayhelptobreak
4σ scatter. Note that this is only a mean representation of
this degeneracy, but this is beyond the scope of this work.
thescatter,asitdiffersforeachindividualmeasurement.As
Weperformtheanalysis(describedindetailbelow)for
two magnitude bins, one including sources with r(cid:48)(cid:54)19.5 an illustration we include the photometric uncertainties in
and one containing sources with r(cid:48)(cid:62)19.5. The motivation r(cid:48)–Hα.
tousetwomagnitudebinsisthefactthatatfaintermagni-
tudes,thephotometricuncertaintiesincreaseandhencethe
3.2 Outliers from the locus of reddened stars
scatter of stars in the CCD also increases. If we combine
sourceswithsmallandlargephotometricerrorsinthesame For the next step, we remove all sources belonging to the
CCD,ourselectioncriteriawillbedominatedbytheintrin- unreddened MS (defined as all datapoints that are within
sic scatter of the faint sources. This may preclude us from 4σ of the final fit) and continue our analysis with the re-
identifying sources with small photometric uncertainties as maining objects. If the slope of our fit after 5 iterations is
significant outliers. We use the value r(cid:48)=19.5 because the stillmoreshallowthanthatofthesyntheticphotometry,we
peak of the distribution of magnitudes in the r(cid:48)-band typi- cannotidentifytheunreddenedMS.Inthatcaseweremove
cally occurs around this magnitude. It is approximately in allpointsabovethesynthetictrackwithE(B−V)=1,with
the middle between the saturation limit and the 5σ detec- theexceptionofsourcesthathaver(cid:48)–i(cid:48)(cid:62)3.5(asthesecan-
tionlimit,androughlythecompletenesslimitoftheoptical not be part of theunreddened MS). We arenow leftwith a
catalogue (Wevers et al. 2016a). sample of reddened stars, and continue to identify outliers
byfittingastraightlinetotheremainingobjects.Asforthe
unreddened case, we determine the iteratively 4σ-clipped
3.1 Outliers from the unreddened MS
scatter to obtain the best fit and the scatter around it. For
WedefineunreddenedobjectsassourcesthathaveE(B−V) sourcesthatarelocatedbelowthelocusofreddenedobjects
(cid:54) 1 (i.e. all objects that lie above the synthetic track with (the absorption line sources) the left hand side of Equation
E(B−V)=1,seeFig.1).AsnotedalreadyinWithametal. 7changestoanabsolutevalue.Moreover,sourcesthathave
(2006),fittingastraightlinetothisselectionofobjectsmay Hαinabsorptionwillhaveacomparativelylowersignal-to-
not converge onto the observed unreddened MS. The solu- noise ratio in the Hα filter (see Section 4). It is expected
tion proposed by these authors is to iteratively force the fit that the scatter of these sources in the CCD will be larger
upwards, and we do the same here. After an initial fit to than the typical scatter of the locus of stars. We therefore
all points with E(B −V)(cid:54)1, we select the objects above useamoreconservativevalueofC=5fortheidentification
the fitted line and iterate, forcing the fit up towards the of absorption line candidates.
unreddenedMS.Inpractice,theshapeoftheCCDisdeter- Visual inspection of the Hα absorption line candidates
minedbythedetectionlimitofourobservations,thestellar shows that we identify many sources that fall partially off
density and reddening along the line of sight. CCDs along the detector when the Hα filter is mounted but not when
different lines of sight have a different shape depending on the r(cid:48) filter is present. This slight shift in the focal plane is
these parameters, hence they require a different number of likely introduced by the different optical path the incoming
iterations for the fit to converge onto the unreddened MS. lightfollowswhentheHαfilterismounted.Toremovethese
To establish whether or not our fit represents the unred- spurioussources,werequirethatthesourcepositionismore
MNRAS000,1–12(2016)
6 Wevers et al.
than20pixelsfromthedetectoredges.Figure3showsanex- First of all, there are 75 frames for which the slope of
ampleoftwoCCDsforthebright(panela)andfaint(panel our fit after five iterations is still more shallow than the
b) magnitude bins. In particular panel b illustrates the di- slope of the synthetic track. As was mentioned earlier, this
versityofphotometricmeasurementerrors(evenforstarsin can arise due to a combination of the stellar density and
thesamemagnitudebin),whichcangreatlyinfluenceoutlier reddeningalongthelineofsighttogetherwithourselection
detection.Theemissionlinecandidatefromtheunreddened criteria for (un)reddened objects. In 9 cases the iterative
MS is inside the mean 4σ boundaries, but still a significant procedurereducedthenumberofstarsavailableforthefitto
outlier because of the very small photometric uncertainty. onlyahandful.Visualinspectionshowsthatnooutlierswere
Theemissionlinecandidateatr(cid:48)–i(cid:48)=3.8isalsoconsistent missedintheseframes.Intheremaining66framestheupper
with the expected locus of unreddened stars. However, the part of the population in the CCD is well identified, even
highr(cid:48)–i(cid:48)colourimpliesitcannotbeanunreddenedobject, though the slope is more shallow than our threshold value.
sowedetectitasanoutliertothelocusofreddenedsources. The outliers we find in these frames are robustly identified
and included in the final sample.
Secondly, the presence of Hii regions can affect the
3.3 Blue outliers CCDbyincreasingthenumberofapparentHαemissionline
sources due to increased non-homogeneous extended emis-
In Figure 3, the fit to the unreddened MS (the upper solid sion. This is the case in fields S04 and N15. In simbad we
line) is satisfactory in panel a, with a slope consistent with find Hii regions LBN1120 and LBN9 for S04 and N15, re-
thatoftheunreddenedMSsyntheticphotometry.Thereare
spectively.TheCCDsoffieldS07containaverylargenum-
two objects with very blue colours compared to the fields
ber of Hα emission and absorption line outliers. We find
stars in the diagram (marked by blue squares). However,
an open star cluster in this field, containing many bright
wedonotidentifythemasoutliersbecausetheyfallwithin
(V∼6)starsthataresaturatedinourobservations.Weat-
thelimitsofthe4σ regionsaroundourfittedlines.Because
tributethelargenumberofoutlierstothepresenceofthese
these are potentially interesting sources, we use a different
bright stars, which cause blooming of charge in the detec-
selection algorithm based on their r(cid:48)–i(cid:48) colour.
torsthatleadtofalsesourcedetectionsandcolours,and/or
We fit the distribution in r(cid:48)–i(cid:48) with a Gaussian mix-
erroneous matches (see Wevers et al. 2016a for a discussion
ture model using the python Machine Learning package
about the effect of saturated sources on e.g. source detec-
sci-kit learn (Pedregosa et al. 2011). Based on the shape of
tion).
the histogram, which is typically single- or double-peaked,
weconsidertwomodels,oneconsistingofoneGaussiandis-
tribution and one consisting of the sum of two Gaussians.
We note that there is no reason to believe that the blue
4 RESULTS
edgeofthisdistributionshouldfollowaone-sidedGaussian
distribution function. The shape of this colour distribution 4.1 Outlier populations
is determined by many factors, including the stellar den-
Wefindatotalof389blueoutliers,336absorptionlinecan-
sity and magnitude distribution of stars along the line of
didatesand1337emissionlinecandidatesinourphotometric
sight, our survey detection limits and the effects of redden-
catalogue.Weshowthemagnitudeandcolourdistributions
ing. Typically the blue edge of the distribution is sharper
oftheemissionandabsorptionlinecandidatesintheleftand
than Gaussian (Carmona-Ruiz et al. in prep.), hence our
rightpanelsofFigure5,respectively.Weidentify62sources
approximation by a Gaussian distribution is a conservative
thatarebothblueandhavesignsofexcessHαemission.We
approach. We show a result of fitting two Gaussian distri-
also find 3 sources that are blue and candidate Hα absorp-
butions to the number of stars as a function of their r(cid:48)–i(cid:48)
tion sources. The sample properties of the blue outliers are
colours in Figure 4 for the same field as in Figure 3.
shown in Figure 6. The distribution of emission line candi-
Wedeterminethewidth,heightandpeakpositionusing
dates, in particular the decrease of the number of outliers
an iterative maximum likelihood estimate approach. Given
in the bin 19.5(cid:54) r(cid:48)(cid:54)20, is introduced by the division into
the best fit parameters, we flag all sources more than 5σ
two magnitude bins at that r(cid:48)=19.5. The increase of Hα
away from the peak of the distribution as outliers (marked
emissionlinecandidatesuptor(cid:48)=19.5canbeexplainedby
inthefigurebythedashedverticalline).Inthecasethatthe
two effects: an increasing number of stars at fainter magni-
sum of two Gaussians works best, we select the blue (low-
tudes, and an increased occurrence of emission line sources
est r(cid:48)–i(cid:48) peak position) Gaussian component to determine
at fainter magnitudes. For the fainter magnitude bin, the
which sources are blue outliers.
intrinsic scatter of stars is much larger (due to the larger
measurement errors), leading to a reduction in the number
of observed outliers compared to the bright bin.
3.4 Quality control
Similarly, there is a large decrease in the number of
Themethodsthatweemploytofindoutliersdonotrelyon absorption line candidates for r(cid:48)(cid:62)19.5. This can be under-
any underlying physical models that accurately predict the stood as a combination of two effects. As mentioned above,
shapesorpositionsofthepopulationsofsourceswearetry- the increased scatter for faint stars decreases the number
ing to describe. We have optimised our methods such that ofoutliersweidentify.Moreover,absorptionlinecandidates
theyworkforthemajorityoftheframes,butvisualinspec- have a lower flux in the Hα filter, hence they are detected
tion shows that in some cases our selection procedures do withalowersignal-to-noiseratiocomparedtoothersources
not yield satisfactory results. We therefore visually inspect withsimilarr(cid:48)-bandmagnitudes.Thismeansthatitismuch
every CCD to reject all anomalous frames. harder to detect absorption line candidates at fainter mag-
MNRAS000,1–12(2016)
Candidate Hα outliers towards the Galactic bulge 7
2.0
a b
1.0
1.5
0.8
) )
g g
a a
m0.6 m1.0
( (
α α
H H
- -

r00.4 r0
0.5
0.2
0.0
0.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 0 1 2 3 4 5
r0 - i0 (mag) r0 - i0 (mag)
Figure 3. Panel a: CCD of all stars brighter than r(cid:48)(cid:54)19.5 on field S20, detector 6. The red solid lines indicate the fits to the loci of
stars;dashedlinesindicatethe4σ scatteraroundthesefits.BlacksquaresareidentifiedaspartoftheunreddenedMS,greendiamonds
belongtothereddenedlocusofstars.Blueoutliersareareplottedinblue,candidateHαoutliersasgoldstars.Thereddeningvectorfor
∆E(B−V)=1isshownasanarrow.Panelb:same,butforallstarsfainterthanr(cid:48)(cid:62)19.5.
3.0
a b
1.0
2.5
0.8
)2.0 )
d d
e e
s s
ali ali0.6
m1.5 m
r r
o o
n n
( (
N 1.0 N 0.4
0.5 0.2
0.0 0.0
0.0 0.5 1.0 1.5 2.0 2.5 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
r0 - i0 (mag) r0 - i0 (mag)
Figure 4. Panel a: normalised r(cid:48)–i(cid:48) colour distribution of all stars brighter than r(cid:48)(cid:54)19.5 on field S20, detector 6. The blue (dashed)
andred(dash-dotted)linesshowtheGaussiandistributions;theblacksolidlineshowsthesumofbothcomponents.Theverticaldashed
line shows the 5σ limit below which sources are flagged as outliers. There are two outliers around r(cid:48)–i(cid:48)=0.2, marked with an arrow.
Panelb:sameaspanela,butforallstarsfainterthanr(cid:48)(cid:62)19.5.Noblueoutliersareidentified.
MNRAS000,1–12(2016)
8 Wevers et al.
Table 2.Exampleofthetablescontainingtheinformationofthesourcesidentifiedasoutliers,includingtheposition,magnitudesand
colours.ThephotometricmeasurementsarequotedinVegamagnitudes.
RA(◦) Dec(◦) r(cid:48) σr(cid:48) i(cid:48) σi(cid:48) Hα σHα r(cid:48)–i(cid:48) r(cid:48)–Hα
266.344818 –32.464782 19.15 0.02 16.81 0.01 18.08 0.02 2.34 1.07
266.251831 –32.328826 17.32 0.02 15.50 0.01 16.19 0.02 1.82 1.13
266.208557 –32.298542 19.01 0.02 17.00 0.01 18.05 0.02 2.01 0.96
266.442108 –32.151733 19.31 0.02 17.37 0.01 18.37 0.02 1.94 0.94
266.280395 –32.102737 17.96 0.02 16.71 0.01 16.80 0.02 1.25 1.16
350 400 450 300 140 250
300 350 400 120
250
200
350
300
s250 s 100
ce 300 ce200
sour200 250 250 sour 80 150
of 200 of 150
er 150 200 er 60
b b 100
m 150 m
u 150 u100
N100 N 40
100
100
50
50
50 50 50 20
0 0 0 0 0 0
16 18 20 22 2 1 0 1 2 3 4 0 0.5 1 1.52 2.5 3 16 18 20 22 1 0 1 2 3 4 5 6 3 2 1 0 1
r0 r0 - i0 r0 - Hα r0 r0 - i0 r0 - Hα
Figure 5. Magnitude and colour distribution of the sample of candidate Hα emission (left three panels) and absorption (right three
panels)lineoutliers.Anexplanationforthedecreasednumberofoutliersinthemagnitudebin19.5(cid:54)r(cid:48)(cid:54)20isgiveninthetext.
outliers and a catalogue of blue outliers with a 4σ selection
80 160 200
limit.
70 140
60 120 150
s 4.2 Optical counterparts to X-ray sources
e
c
ur50 100 Now that we have identified the outlier candidates in the
o
of s40 80 100 optical photometry, we cross-correlate the sample with the
er GBS X-ray sources from Jonker et al. (2014). We use the
b
m30 60 4σ X-ray error circle, defined in Wevers et al. (2016a), as
u
N the crossmatching radius. We find thirteen matches among
20 40 50
the emission line candidates, while for the absorption line
10 20 candidates we find one potential counterparts and among
theblueoutlierswefindtwomatches.InTable3wepresent
0 16 18 20 22 -01.5 -1-0.5 0 0.5 1 1.5 03 2 1 0 1 the optical counterparts.
r0 r0 - i0 r0 - Hα
Figure 6.SameasFigure5,butfortheblueoutliers. 5 DISCUSSION
5.1 Optical counterparts to X-ray sources
A significant fraction of the optical counterparts to X-ray
nitudes, and results in a strong decrease in the number we sources found in this work have already been classified. In
can identify. this regard, of the thirteen emission line candidates, seven
Table 2 shows an example of the information avail- have been previously studied and classified as CVs, three
able for the outliers. The full tables can be found in the of which are magnetic systems (Table 3). This is not sur-
online material, and will also be made available in elec- prising, since magnetic systems are known to have a higher
tronic form through the Vizier database (http://vizier.u- X-raytoopticalfluxratio,sotheycanbedetectedatlarger
strasbg.fr). Here we will include the outlier catalogues de- distances(inX-rays)comparedtonon-magneticCVs.Inad-
scribed above, and in addition we will add outlier samples dition,twoHαemissionlineobjectswereidentifiedasdwarf
with slightly less restrictive selection critera. In particular, novae, while the remaining two are a dwarf nova candidate
wewilladdanemissionlinecataloguewhichincludesall3σ andeitheraCVoraqLMXB,respectively.Onebackground
MNRAS000,1–12(2016)
Candidate Hα outliers towards the Galactic bulge 9
Table 3. Properties of the optical counterparts (identified as outliers) to GBS X-ray sources. The position of the optical counterpart
is given in degrees. The uncertainties of these positions can be found in Wevers et al. (2016a). All magnitudes are given in the Vega
system.Thenumbersinbracketscorrespondtotheuncertaintyonthelastdigit.EWistheequivalentwidthoftheHαlinein˚A,where
a negative value indicates emission. In the comments we give the classification (if available) and the most relevant reference work. DN
standsfordwarfnova,IPforintermediatepolar,qLMXBforquiescentlowmassX-raybinary,andSy1forSeyfert1galaxy.
CXID RA(◦) Dec(◦) r(cid:48) i(cid:48) Hα r(cid:48)–i(cid:48) r(cid:48)–Hα EW FX Comments
Fopt
Emissionlinesources
CX5 265.038086 –28.790512 19.03(2) 18.00(1) 18.03(2) 1.03 1.00 –50 19.8 IPa
CX21 265.390747 –28.676245 19.14(2) 18.46(1) 17.37(2) 0.68 1.77 11.5
CX37 264.371490 –29.467827 19.01(2) 18.35(2) 18.21(2) 0.66 0.80 –45 6.4 IPa
CX81 266.109680 –27.323917 20.81(3) 19.95(3) 19.46(3) 0.86 1.35 16.7 DNb
CX93 266.186615 –26.058407 17.81(1) 16.66(1) 17.01(1) 1.15 0.80 –18.4 0.73 CVc
CX118 264.709259 –28.802433 17.90(2) 16.94(1) 17.04(2) 0.96 0.86 0.81
CX142 266.015655 –31.384815 21.23(3) 20.21(2) 20.28(5) 1.02 0.95 –59 13.5 DNd
CX207 266.606201 –26.526419 20.02(2) 19.25(3) 19.16(3) 0.77 0.86 –83 4.4 IPd
CX585 265.953796 –31.416586 19.95(2) 19.23(1) 18.65(2) 0.72 1.30 1.9
CX645 266.639374 –26.387234 19.43(2) 18.90(2) 18.77(2) 0.53 0.66 1.4 CV/qLMXBb
CX982 267.191925 -30.660831 21.9(1) 20.02(8) 19.96(5) 1.9 2.0 0.24 DN?b
CX1061 265.886413 -26.750890 18.77(2) 16.77(1) 17.72(2) 2.00 1.05 0.01
CXB279 266.340759 –32.160236 18.60(2) 16.80(1) 17.76(2) 1.80 0.84 0.17
Absorptionlinesources
CX2 264.368256 –29.133940 18.38(2) 16.78(1) 17.96(2) 1.60 0.42 –490(5) 86 Sy1e
Bluesources
CX361 267.781891 –29.677025 17.45(1) 17.64(2) 17.29(1) -0.19 0.16 0.63 AMCVnf
CXB34 266.870453 –32.244946 21.39(4) 20.97(7) 21.34(1) 0.42 0.05 3.3
aBrittetal.(2013),bBrittetal.(2014),cRattietal.(2013),dTorresetal.(2014),eMaccaroneetal.(2012)fWeversetal.(2016b)
AGNwasidentifiedasanHαabsorptionlinecandidate,and withmid-IRcataloguessuchastheSpitzerBulgecatalogue
one blue outlier was classified as a high state AM CVn sys- (Uttenthaler et al. 2010), the GLIMPSE catalogue (Spitzer
tem. Science2009)andAllWISEcatalogues(Cutri&etal.2014).
5.3 Colour-colour diagram of outliers
5.2 Multi-wavelength counterparts
WediscusstheCCDofalltheoutliersandtheimplications
Anumberofobservingprogramswitharcsecondspatialres-
for the identification of different source classes in the dia-
olution have imaged the Galactic bulge, producing source
gram. We show the position of all outliers in the CCD in
catalogues which are potentially useful to further constrain
Figure 7. The emission line candidates are plotted as grey
the nature of objects in our sample. Below, we give a brief
triangles, while blue outliers are marked as blue circles and
overviewofthewavelengthcoverageandpropertiesofsome
absorption line candidates are shown as green diamonds.
of these catalogues. A detailed study of multi-wavelength
CounterpartstoGBSX-raysourcesaremarkedasredstars.
crossmatcheswithourcataloguesisbeyondthescopeofthis
Thegapbetweenemissionandabsorptionlinesourcesmarks
paper.
the global position of the locus of objects in all GBS fields.
Atopticalwavelengths,theOGLEsurvey(Udalskietal.
2015)hasobservedalargepartoftheGalacticbulgeduring
multiple observing seasons with a cadence on occasion as
5.3.1 Emission line sources
short as 20 min, adding temporal information to our opti-
cal colours. In the near future VPHAS+ (Drew et al. 2014) We interpret the spread of the emission line candidates (in
will complement our obervations by adding catalogues in r(cid:48)–Hα)astwodifferentpopulations.Thelargestpopulation
the u(cid:48), g(cid:48) and z(cid:48) bands, as well as temporal colour and as- consistsoftheoutliersfromtheunreddenedMS,andconsti-
trometric information on a ∼10 year baseline. For some tutestheupperpopulationofemitters.Inadditiontothese,
sources, proper motion measurements are available (e.g. another track of outliers at lower r(cid:48)–Hα with a shallower
Sumi et al. 2004, Fedorov et al. 2009). Narrow-band Hei slope can be identified. These systems are likely reddened
5875˚A information is also available from the UV EXcess stars with Hα in emission.
survey (Groot et al. 2009). Other catalogues that overlap We expect the sources that have r(cid:48)–i(cid:48) (cid:54) 1 and Hα
include the XMM-Newton serendipitous UV survey cata- in emission (the region of overlap between the blue outliers
logue (Page et al. 2012), while the Vista Variables in the andemitters)tobegood(non-magnetic)CVcandidates.At
Via Lactea survey (Saito et al. 2012) and UKIDDS (Lucas somewhatreddercoloursweexpectotherHαexcesssources
etal.2008)surveyscanconstraintheNIRpartofthespec- such as intermediate polars (IPs) and (quiescent) LMXBs.
tralenergydistribution.PartoftheGBSfootprintcoincides For example, NS LMXBs and long period BH LMXBs are
MNRAS000,1–12(2016)
10 Wevers et al.
3
2
1
α
H

-

r0
0
1
2
1 0 1 2 3 4 5 6
r - i
0 0
Figure 7. Colour-colour diagram of the three samples of outliers. Blue outliers are shown as blue circles, while emitter candidates are
shownasgreytrianglesandabsorptionlinecandidatesasgreendiamonds.RedstarsindicatecounterpartstoX-raysourcesinTable3.
more X-ray bright, hence they can be detected further out we expect the blue sources to be relatively close to Earth.
in X-rays and their optical colours will be reddened by in- The differential reddening in the r(cid:48)-band with respect to i(cid:48)
terstellardustextinction.Intrinsicallymoreredobjectssuch is∆(r(cid:48)–i(cid:48))=0.26×Ar(cid:48) (Schlegeletal.1998),whereAr(cid:48) is
as flare stars, and other sources such as chromospherically theextinctioninther(cid:48)-band(inmag).Assuminganintrinsic
active (binary) stars and early-type emission line stars are colourofr(cid:48)–i(cid:48)=0(correspondingtoanA0Vspectraltype,
also expected to be found as Hα emitters. Planetary nebu- and typical for a blackbody spectrum with T∼10000 K),
lae, whose spectrum is dominated by emission lines with a this implies a typical reddening of Ar(cid:48)∼1 mag which in
low continuum contribution, should show up as very large turn means that the source should be nearby (cid:46) 1 kpc.
Hα excess sources up to the r(cid:48)–Hα=3.3 limit.
Blue outliers are identified up to r(cid:48)–i(cid:48)=1, indicating
Crossmatching these sources with the SIMBAD
thatwearenotjustfindingforegroundobjectsbutalsosys-
database yields a variety of sources (Table 4), indicating
temsatdistanceswherethedustextinctionbecomesappre-
thatindeedoursamplecouldspanthewholerangeofstellar
evolution. We find, among others, two young stellar object ciable(Ar(cid:48)∼4mag).Forexample,insomelinesofsightwith
lowextinctionbackgroundAGNmaybepartoftheblueout-
candidates,fivevariablestars,twoasymptoticgiantbranch
lierpopulation.Someofthe(dust)extinctionmayalsobein-
candidates and four systems containing a WD.
trinsictothesystem.Hotsubluminousdwarfs(sdO/Bstars)
are also expected to be found as blue outliers with respect
tofieldstars.Typicalhotsubdwarfstarsfainterthang=17
5.3.2 Blue sources
maghavedistancesexceeding4kpc(Geieretal.2011),im-
The population of blue outliers likely consists of systems plying that these blue objects could appear as some of the
containingacompactobjectsuchasawhitedwarf,neutron reddest of the blue outliers. The sample of 62 blue outliers
starorblackhole.Nearbyearly-typeMSstarsaresaturated thatarealsoidentifiedasemissionlinecandidateslikelycon-
in our observations, hence they cannot populate the blue sistsofnearbyCVswhoseopticalspectrumisdominatedby
partoftheCCD.Althoughsomeoutliershavebluercolours the accretion disk, and in addition PN are known to pop-
thanfieldstarsintheCCD,theircoloursarenotextremely ulate this part of colour space (e.g. Corradi et al. 2008).
blue. The median (r(cid:48)–i(cid:48)) colour is 0.26, and because r(cid:48)–i(cid:48) BluesourcesshowingindicationsofstrongHαabsorption(3
increases with distance (due to interstellar dust extinction) sources in our sample) are likely H-dominated (DA) WDs,
MNRAS000,1–12(2016)