arX

iv:1

504.

0512

5v1

[ast

ro-p

h.H

E]

20 A

pr 2

015

SUBMITTED TO APJPreprint typeset using LATEX style emulateapj v. 5/2/11

ECHELLE SPECTROSCOPY OFγ-RAY BINARY 1FGL J1018.6−5856

IDEL R. WAISBERG1 AND ROGERW. ROMANI 1

Submitted to ApJ

ABSTRACTWe observed theFermi-discoveredγ-ray binary 1FGL J1018.6−5856 at 20 epochs over 50 days using the

CHIRON spectrograph, obtaining spectra atR ∼25,000 coveringλλ4090 − 8908A. The average spectrumconfirms an O6 V((f)) spectral type and extinctionE(B − V ) = 1.35± 0.04. Variable absorption line equiv-alent widths suggest substantial contamination by wind line features. The limited S/N ratio hindered accuratecontinuum definition and prevented measurement of a high quality radial velocity curve. Nevertheless, thebest data indicate a radial velocity amplitude≤ 40km s−1 for the HeII lines and substantially lower for H I.We argue that this indicates a most likely compact object mass< 2.2M⊙. While black hole solutions are notexcluded, a neutron star source of theγ-ray emission seems preferred.

Subject headings: gamma rays: stars — binaries: spectroscopic

1. INTRODUCTION

To date only five binaries with high mass secondaries havebeen detected in the GeV band; 1FGL 1018.6−5856 is theonly one whose binary period was discovered at these ener-gies (Corbet et al. 2011; Ackerman et al. 2012). The natureof the γ-producing compact object in these binaries is un-clear: it might be a jet-producing black hole, in analogy toγ-loud blazars (the ‘µQSO model’), or theγ-rays may begenerated by the magnetosphere or wind termination shockof an energetic pulsar (the ‘PSR/PWN model’). This lattermodel certainly applies to the PSR B1259−63 system, whoseGeV emission is produced via Inverse Compton scatteringoff wind shock particles when the PWN is compressed nearperiastron. For 1FGL 1018.6−5856 and the other three sys-tems (LS I+61o303, LS 5039 and Cygnus X-3), the identityof the compact object is presently unknown. Discovery ofγ-ray pulsations would certainly indicate a neutron star pri-mary. However, the poorly known acceleration in the binariesmakesγ-ray pulse searches prohibitively expensive and thedense plasma produced by the massive secondary winds frus-trates radio pulse searches. An alternative approach is to con-strain the compact object mass via measurement of the sec-ondary’s radial velocity. Although the high mass and strongwinds of the secondary also make this very challenging, evencrude constraints can be useful; for example a minimum com-panion mass> 3M⊙ would certainly indicate a black hole.

After discovery of a 16.5 dγ-ray modulation (Corbet et al.2011), the associatedSwift X-ray source was identifiedwith a bright optical counterpart. A low resolution spec-trum with the SAAO 1.9 m shows an O6V((f)) companionwith an estimated mass≈ 20 − 30M⊙ and distance≈ 5 kpc(Ackerman et al. 2012). This system is also a radio sourceandSwift XRT (An et al 2013) and ATCA monitoring detectthe orbital modulation in the X-ray and radio band, respec-tively. Optical photometric monitoring detects no variation,although present constraints are weak. These properties aresimilar to those of the more compact (Pb=3.9 d)γ-ray binaryLS 5039. There spectroscopic and photometric studies (e.g.Sarty et al. 2011) indicate an O6.5V((f)) secondary, a sub-stantial eccentricitye = 0.24 ± 0.08, and a modest mass

1 Department of Physics, Stanford University, Stanford, CA 94305-4060, USA; rwr@astro.stanford.edu, idelw@stanford.edu

function f(m) = 0.0049 ± 0.0006M⊙. The limit on LS5039’s photometric variability, 2mmag, implies a high sys-tem inclination and/or a small compact object mass. Sincethe inclination is poorly determined, they can only concludethatM1 > 1.8M⊙, allowing both black hole and neutron starsolutions.

In this paper we describe a project to study the radial ve-locity of 1FGL 1018.6−5856 using echelle spectroscopy. Weadopt the orbital period estimate of Coley et al. (2014)Pb =16.549± 0.007d with epoch MJD55403.3± 0.4 correspond-ing to theγ-ray peak described in Ackerman et al. (2012). AnindependentSwift/Fermi analysis (An, priv. comm.) con-firms this ephemeris. While our results appear to show sys-tematic variation with orbital phase, it is unclear if we haveisolated the true radial velocity curve. Nevertheless, ourmea-surements improve the characterization of the companion star,limit the amplitude of its motion and point to the sort of ob-servations needed for a definitive study.

2. OBSERVATIONS AND CALIBRATIONS

The 16.5 d binary period is particularly inconvenient, beingtoo long to cover in a dedicated run, but too short for effec-tive coverage with ad hoc∼ monthly observations. The bestsolution was queue observing and we were awarded 25h ofqueue time (NOAO-11B-0092) on the CTIO 1.5 m using theCHIRON fiber-coupled echelle spectrograph (Schwab et al.2010). 1FGL 1018.6−5856 is relatively faint for CHIRONat V = 12.5, so CHIRON was run in the slitless (2.5′′ fiberaperture) mode for an effective resolutionR ∼ 25, 000. TheCCDs were binned4×4 to maximize S/N and minimize read-out time. The spectra covered 4090−8908A, with substantialorder overlap except in the reddest orders.

Observations were attempted roughly every other nightfrom March 02, 2012 (JD 2455989) to April 20, 2012 (JD2456022). Four nights were lost to system failures andweather; spectra were obtained on 20 nights (Table 1). Werequested 3×1000s exposure on 1FGL 1018.6−5856, onenight had only a single exposure. For each epoch we obtainedThAr arc exposures immediately before the target observa-tions for wavelength calibration. For the first epoch only wealso observed the flux standard HD60753. CHIRON was un-dergoing final commissioning during these observations, andthe obtained S/N was substantially lower than predicted by

2 Waisberg et al.

TABLE 1CHIRON OBSERVATIONS

Date Nexp S/N6000A

φB

03/02 3 36 0.4103/06 1 24 0.6503/10 3 12 0.8903/14 3 33 0.1303/16 3 32 0.2503/18 3 44 0.3703/20 3 37 0.4903/22 3 33 0.6103/24 3 3 0.7303/26 3 31 0.8503/28 3 18 0.9803/30 3 18 0.0904/02 3 28 0.2804/04 3 31 0.4004/06 3 21 0.5104/08 3 8 0.6304/10 3 22 0.7604/12 3 20 0.8704/18 3 30 0.2404/20 3 29 0.38

the design plots (predicted peak S/N/extracted pixel∼ 40 ineach 1000s exposure). In practice, we obtained S/N/extractedpixel ∼ 25 at the order peaks in our best exposures (in linewith recent revised 6.5% instrument efficiency). More typi-cal S/N/extracted pixel was≈ 17, with a few epochs muchworse. We were not able to ascertain from the night logs whythe S/N on source varied so dramatically, although from thelow apparent flux at some epochs, we suspect that poor trans-parency and imperfect fiber acquisition may play a role. In theend we analyzed the 17 epochs with S/N> 15 in the summed,combined spectra. These spectra sample four binary orbits.

Although the CHIRON data distribution included a stan-dard pipeline extraction of the echelle spectra, this was op-timized for the relatively red target stars of the instrument’sprincipal (radial velocity planet search) program. This ex-traction coveredλλ4504 − 8908A (62 orders); for our blue06V target 14 additional orders were measurable, extendingthe spectral coverage to 4090A.

Standard CCD processing was performed with the IRAFpackage. A master bias was assembled and subtracted fromthe target, flat and lamp images. The spectra for each epochcover 3200s. As this is only 0.2% of the orbital period,intra-epoch velocity smearing was negligible and the threetarget exposures for each night were median combined withaverageσ-clipping to increase S/N and reject cosmic rayevents. Combined flat field images were also prepared foreach night. These were used to define the apertures for eachorder. Sufficient inter-order pixels were available to defineflanking ‘background’ windows for each order. A linearfit to these backgrounds was used as a local scattered lightsubtraction. All extractions were performed using varianceweighting, modeled by the CCD read noise (4.5e−) and gain(1.3e−/ADU).

The extracted science spectra for each order were dividedby the extracted flat spectrum. This removed the blaze func-tion and summed pixel-to-pixel variations, while maintaininga relatively stable response from the smooth lamp spectrum.In principle, the true response could be determined from thestandard star spectrum. In practice, for the queue-selectedstandard the tabulated fluxes were derived from a spectralmodel and were too sparse (10A grid) to fit an accurate sensi-

FIG. 1.— Fluxed CHIRON spectrum from March 02, 2012, corrected foratmospheric and interstellar (EB−V = 1.35) extinction. Telluric featureshave not been removed. The strongest night sky and absorption lines arevisible, along with inter-order gaps for the reddest five orders.

tivity function for each order. Accordingly we used the lamp-flattened spectra directly. The wavelength calibration wasde-termined from the nightly ThAr spectra. Non-linear solutionswere fit, with a typical rms residual 0.01A. Note that althoughCHIRON is designed to be quite stable, night-to-night shiftsin the spectrum position on the chip were significant. Aper-ture migrations (1-2 pixels) were treated by re-centering or-ders on the flat before extraction (and importing these tracesto the arc and science spectra). Shifts along the dispersiondi-rection (∼ 0.1A∼ 5 − 10km s−1) were removed by runningIRAF taskre-identify on each night’s ThAr arcs to derive thewavelength calibration.

After trimming very low S/N data at the edges of eachechelle order, we merged all orders to a single spectrum,average-combining the overlap regions (figure 1). Given ourrelatively modest S/N for most of the spectra, this proce-dure generally produced an adequate match between orders.For some epochs, small flux gradients across orders and poorinter-order match were however present. Such imperfectionscan be attributed to differences in illumination paths of and/orlight path shifts between flat and science images. These maybe removed by computation of a smoothly varying residualblaze function (Skoda et al. 2008). Although we attemptedsuch corrections, the S/N of our targets and standard expo-sures did not allow significant improvement. Finally, we di-vided the merged spectra by a low order function to removethe global lamp response and create normalized spectra ap-propriate to our kinematic/line study. The S/N weighted com-bined normalized spectrum from all epochs is shown in figure2. An absolute fluxed spectrum was computed from the firstnight’s data where the adjacent standard was used to normal-ize the fluxes. For each epoch we corrected the normalizedspectrum to the heliocenter using the IRAF routinesrvcorrectanddopcor.

3. SPECTRUM ANALYSIS

1FGL J1018.6−5856 3

FIG. 2.— Normalized average spectrum. Each sub-panel covers 1000A.The strongest lines used for the radial velocity analysis are labeled, addi-tional lines from these species are marked. The synthesizedDIB transmissionspectrum is generated from tables in Jenniskens et al. (1996) (normalized toEB−V = 1.35) and is plotted offset by -0.1.

The overall spectrum (figure 1) is quite consistent with theSAAO spectrum of Ackerman et al. (2012), with HI, He Iand HeII lines competing with strong interstellar absorption.The HeIIλ4541/HeIλ4471 ratio is a good diagnostic of sub-class for O-type stars (Gray et al. 2009). From our combinedspectrum we measure2.3 ± 0.2 for this ratio, indicating O6(Jascheck & Jascheck 1995). The NIII emission line triplet4634/4640/4642 is not present, or even weakly in absorption.Even higher S/N is needed to measure the state of this line.Its weakness argues for a dwarf classification; overall the des-ignation O6 V((f)) is supported by this spectrum. The photo-spheric lines are broad, reaching FWHM 350km s−1. Therewere no sky fibers for sky subtraction and so a few of thebrightest night sky lines (e.g. 5577A ) are visible. A broademission feature at 4293A appears in most of the spectra, withhighly variable flux and high equivalent width even when thetarget was weak. As this corresponds to no strong astronom-ical line, we suspect a spectrograph artifact from local light.In figure 2 we label the strongest photospheric features.

3.1. Interstellar Absorption

Our measurement of the photospheric features is com-plicated by particularly strong interstellar absorption.Atomic features (e.g. Mg I 5174, Na I 5892) providenarrow absorptions, while Diffuse Interstellar Bands (DIBs)are particularly prominent in the spectrum. To identifythese, we use the catalog of Jenniskens et al. (1996, see alsohttp://leonid.arc.nasa.gov/DIBcatalog.html)to synthesize a DIB absorption spectrum, combining Gaus-sian features at the catalogedλeff , kinematic and relativeequivalent widths (figure 2). Comparison with the companionspectrum shows that many subtle broad features have aDIB origin. Note that the relative feature strength shownhere is only typical; substantial variation exists betweenvarious Galactic sight-lines. The many DIB features make

TABLE 2REDDENING FROMDIB M EASUREMENTS

DIB a† b† EW EB−V

A A

5780 0.506 ± 0.050 −0.014 ± 0.032 0.709± 0.009 1.43± 0.165797 0.124 ± 0.014 −0.003 ± 0.009 0.134± 0.006 1.10± 0.146196 0.048 ± 0.004 0.001 ± 0.003 0.063± 0.005 1.29± 0.126379 0.057 ± 0.012 0.004 ± 0.008 0.119± 0.005 2.00± 0.446613 0.192 ± 0.026 0.003 ± 0.017 0.225± 0.006 1.16± 0.186660 0.014 ± 0.004 0.009 ± 0.003 0.043± 0.004 2.54± 0.82

†Coefficients from Kos et al. (2013).

it particularly challenging to define the true photosphericcontinuum, in turn this can subtly affect our line centroidestimates.

The DIBs are believed to be generated by PAHs associatedwith interstellar dust. Thus one expects a correlation betweenDIB strength and interstellar reddening from dust scattering.Kos et al. (2013) lists fit linear relations for several strongDIBs of the formEW = a×EB−V + b. These relations de-pend on whether the sight-line is ‘UV shielded’ (ζ sight-line)or not (σ sight-line), and theλ5797/λ5780 EW ratio is usedto discriminate these cases. Our value is0.19 ± 0.01, belowthe 0.3 threshold, so we treat this as aσ sight-line. In table 2we apply the tabulated coefficients to several well-measuredDIBs to estimate the reddening. The inferredEB−V have asurprisingly large scatter, with a weighted averageEB−V =1.35±0.04 (bootstrap error). This value agrees very well withthe estimateEB−V = 1.34±0.04 determined by Napoli et al.(2011) from Stromgren photometry. It is similarly in reason-able accord with theNH estimated from the X-ray spectrum.In turn, this agreement supports the photometric distance es-timated = 5+4.6

−2.1 kpc.

3.2. Radial Velocity Measurements

To confirm the stability of our wavelength solutions, wemeasured the night sky linesλ5577.3387 A andλ6300.304 Ain each processed spectrum, before heliocentric correctionswere applied. The rms wavelength variations were0.0106 Aand0.0124 A respectively, i.e.σRV = 0.6km s−1.

A second stability check was made measuring the verystrong Na(D) interstellar doubletλ5889.951/5895.924A inspectra corrected to the heliocenter. The line centroids dis-play mean redshifts of 2.88 and 2.85km s−1, with spectrum-to-spectrum rms variation of 0.88 and 0.74km s−1, respec-tively. We conclude that our wavelength solutions are good tobetter than 1km s−1.

To attempt to measure the photospheric velocity of thecompanion, we must follow the centroids of the broad (∼

300km s−1) H I, He I and HeII lines, to a fewkm s−1. Thisis challenging since the imperfect inter-order fluxing causessignificant gradients and steps in the continuum, with a char-acteristic scale comparable to the typical 50A (3000km s−1)order width. Further, for several of the photospheric linesthere are strong DIBs in the line wings, which can perturbthe continuum. To mitigate this latter effect, we divide thespectrum by the model DIB spectrum. While imperfectly re-moving all DIBs, this does significantly flatten the continuumin the region of the strongest features.

All photospheric lines were fit with simple Gaussian pro-files (test Voigt fitting showed negligible Lorentzian wings).The fitting used the IRAF deblending routine, where the ef-

4 Waisberg et al.

FIG. 3.— Radial Velocity measurements for three species. Points withcircles are from high S/N spectra; those with crosses from lower S/N data.Black squares show the mean velocity of the two strongest lines for H I andHe II for the high S/N spectra; large systematic scatter is seen. The sampleK = 40 km s

−1 radial velocity curve plotted provides a reasonable matchto the HeII average points; the H I amplitude is∼ 2.5× smaller (see text).

fective noise was determined for each line by a measurementof the spectrum rms in the nearby, line-free continuum. Thestatistical errors on the fit parameters (line centroid, strengthand, optionally kinematic width) are provided by the routine,based on Monte Carlo simulations. For the lower S/N lines thekinematic width was held fixed at the value measured from theaverage spectrum (5-10A FWHM).

We also attempted to determine the line centroids usingthe so-called “mirroring” method (Parimucha & Skoda 2008).This allows one to separately probe the line wings and core.Although we can see that our lines are significantly asym-metric, suggesting absorption in wind/outflow components,we lacked the S/N to measure these structures accurately andcompare them across lines and species. This prevented usfrom using the line shapes to improve our estimate of the sys-tematic (photospheric) radial velocity.

4. RADIAL VELOCITY VARIATIONS

To test for velocity variation, we performed a weightedleast-squares fit for the mean velocity of each species. Table3 shows the velocities, errors, goodness of fit and the proba-bility of constancy. It is unclear how to compare mean veloci-ties between lines, as the effective rest wavelengths depend onthe excitation. Also, the measured line centers are somewhatuncertain due to continuum irregularities; for example HeII6527 appears substantially redshifted, due to a continuum gra-dient and superposed DIB. Finally, the lower excitation linescan be affected by wind absorption and may be blue shiftedwith respect to HeII (Casares et al. 2005; Sarty et al. 2011).

Overall, we infer a net systemic radial velocity of∼40km s−1, but a precise value requires cross-correlation witha simultaneously observed RV standard of similar spectralclass. The reducedχ2 values are, however, interesting. Fromthe low probabilities, we infer that the the strong lines do in-deed show statistically significant variation.

However, there are certainly large systematic effects, andit

TABLE 3TEST FORRADIAL VELOCITY VARIATIONS

Species λrestΓ χ2/DoF Log(Prob)

A in air km s−1

Hα 6562.808 43.0± 1.1 14.7 −3.9Hβ 4861.34 22.4± 3.0 3.6 −1.2Hγ 4340.48 25.5± 36.7 3.3 −1.2He I 5875.65 72.9± 4.3 20.8 −5.3He II 4685.682 44.1± 20.9 10.7 −3.0He II 5411.52 63.3± 3.3 15.5 −4.1He II 6527.099 125.8 ± 15.4 5.6 −1.7

FIG. 4.— Equivalent width and its variation for strong absorption lines. Hβand HeII 4686 absorption appears to weaken near phaseφB = 0.2.is unclear how much of this variation is associated with thephotosphere radial velocity. To illustrate, we average thera-dial velocity from the two strongest H I lines (Hα and Hβ) andtwo strongest HeII lines (λ4686 andλ5412) and plot the val-ues for the highest quality (S/N6000 > 20) spectra (Figure 3).Even disegarding the mean radial velocityΓ offset betweenlines, we see that the fluctuations are several times the statis-tical error. Nevertheless, both sets of velocities show similartrends, with relatively low values in the rangeφB = 0.0−0.2.SinceφB = 0 is the γ-/X-ray light curve maximum, it istempting to identify this excursion with periastron and inter-pret the radial velocity measurements as showing significanteccentricity. The two species HI and HeII show mean veloc-itiesΓH I = 31.9km s−1 andΓHe II = 46.2km s−1 and RMSdispersionsσH I = 11.1km s−1 andσHe II = 29.2km s−1.

4.1. Equivalent Width Variations

Following Sarty et al. (2011), we checked the equivalentwidths of our strongest absorption lines. Hα, the strongestatomic absorption has an average equivalent width of−2.2A.Like these authors we also see evidence for equivalent widthvariation through the orbit, especially for Hβ and HeII 4686(Figure 4). However, in our case the minimum absorptionstrengths appear to lie atφB ≈ 0.2, close to compact objectsuperior conjunction for the orbit suggested in figure 3 (i.e.when viewing the side of the companion away from the com-

1FGL J1018.6−5856 5

pact object). This might indicate denser wind flow throughthe L3 point. Inspection of the strongest lines show variablewings, suggesting time or phase-dependent views of such out-flow.

As for LS 5039, we find that our observed Hα equiv-alent width, −1.87 to −2.29A , is smaller than the ex-pected≈ −3.27A photospheric width for this spectral class(McSwain et al. 2004). Scaling from those authors’ estimates(i.e. assuming similar companion mass and asymptotic windvelocity), we find that our lower observed residual absorptionequivalent width implies a somewhat larger mass flux, namelyM ≈ 0.7− 1.0× 10−7M⊙ y−1.

4.2. Radial Velocity Model and Implied Mass

Labeling the primary (compact object) 1 and the visible sec-ondary 2, we have a single line spectroscopic binary, with ra-dial velocity measurements constraining the mass function

f(M1) =M3

1

(M1 +M2)2sin3i =

P

2πGK3

2(1 − e2)3

2

with K2 = ∆v2/2 the radial velocity amplitude of the sec-ondary andi the unknown system inclination. Sincesini < 1andM2 > 0, f(M1) provides a lower bound to the massof the primary. Casares et al. (2005) give the mass of an O6V((f)) star as20.0− 26.4M⊙.

In view of the large and poorly understood systematics, oneshould not directly fit our radial velocities. However, a ba-sic model (figure 3) gives an idea of the allowed amplitude.Here a curve is plotted forK = 40km s−1, Γ = 40km s−1,e = 0.2, and longitude of ascending nodeω = 0.6π. Thisprovides a reasonable match to the range of the HeII line; theH I range isK ≈ 15km s−1. Larger amplitudes or eccentrici-ties provide a poor match to the data, so we take this as an up-per limit on the allowed radial velocity amplitude. Sarty etal.(2011) adopt the HeII radial velocity curve as minimizingwind contamination and best following the photospheric mo-tion. Their curve for HI has a somewhat smaller amplitudethan that for HeII ; for our data the range of HI velocities is∼ 2.5× smaller. Thus we follow Sarty et al. (2011) in adopt-ing the HeII velocities, but consider the possibility that theseare systematics dominated and that the stronger H I lines re-flect the true radial velocity amplitude.

Figure 5 shows the corresponding allowed regions in themass-mass plane for circular orbits. For the allowed rangeof eccentricityM1 will be up to 7% smaller. The solid linesshow theK = 40km s−1 constraints suggested by HeII , thedashed lines show the range from the Balmer lines.

5. DISCUSSION AND CONCLUSIONS

Our attempt to measure the radial velocity curve of 1FGL1018.6−5856 has been frustrated by a large scatter, appar-ently due to systematic errors. In part these are certainly dueto the imprecision of the continuum calibration in our echellespectra. In part they are also likely due to variable absorptionline components of a stellar wind unpredictably shifting theline centroids.

However, we do see that the maximum radial velocity am-plitude allowed by our data is∼ 40km s−1. If this is the trueradial velocity amplitude, then we would require a compactobject mass> 1.7M⊙ and 50% of the allowed parameterspace hasM1 > 2.2M⊙. Since the likely maximum neutronstar mass is≈ 2.5M⊙, this would seem to suggest that theγ-ray source is a black hole. However, one should compare

with the masses of other stellar mass black holes. Farr et al.(2011) find that at 90% confidence the minimumMBH instellar mass binaries is> 4.5M⊙. Although they do not sep-arately evaluate this for the high mass systems, like 1FGL1018.6−5856, their figure 2 indicates most probable masses10− 20M⊙ and 90% lower mass limits> 10M⊙. From fig-ure 5 we see that we requirei < 30◦ for M1 > 4.5M⊙ (apriori probability 13%), while forM1 > 10M⊙ we requirei < 14◦ (probability 3%). Thus the likelihood of having atypical stellar black hole mass seems modest. When we con-sider that systematic errors may be contributing to the HeIIradial velocity width and that the H I radial velocity amplitudemay be closer to the intrinsic amplitude, this only allowsM1

to exceed4.5M⊙ for i < 11◦ (probability 2%). Thus we con-clude that the modest radial velocity amplitude observed for1FGL 1018.6−5856 prefers a neutron star. However, thesestatistical statements do not exclude either option and a cleanmeasurement of the companion radial velocity would help thesituation greatly.

The way forward on determining the mass of 1FGLJ1018.6−5856 doubtless starts with improved spectroscopy.Our data show that the photospheric lines are quite broad,and likely variable. Accordingly a future campaign on small-medium class telescopes should obtain long-slit, moderateresolution (R ∼ 3000− 6000) spectra covering 4000-5500A,with a collection of several template O6 V radial velocitycomparisons. Queue/service observing is almost certainlyre-quired in view of the largePb. Higher resolution echelle datafrom a large telescope can be of use, but will require higherS/N to allow deconvolution of the line shape and isolation ofthe systematic radial velocity. If the minimum mass exceeds3M⊙, we can be confident of a black hole primary. But inthe more likely event that lower masses are allowed, the incli-nation becomes critical and one will require very high preci-sion photometry to constrain the small ellipsoidal variations,

FIG. 5.— The binary mass-mass plane. For the HeII-estimated radial veloc-ity amplitude, we see that a neutron star (NS) solution requires small compan-ion masses and/or large inclinations (i > 50◦); roughly half of the allowedparameter space is consistent with a black hole (BH), although typical BHmasses require very small inclinations. For the H I amplitude, a neutron staris highly preferred, requiring small inclinations to reacheven1.4M⊙.

6 Waisberg et al.

or constraints from light curve fitting at other wavebands.

We thank Robin Corbet, Malcolm Coe, Richard Dubois andMatthew Kerr for assistance with original proposal. HongJunAn helped with useful discussions about the orbit ephemerisand Matt Giguere assisted with the CHIRON data acquisition.

This project made use of data obtained at the CTIO 1.5 mtelescope run by the SMARTS consortium. CTIO is operatedby the Association of Universities for Research in Astronomy,under contract with the National Science Foundation. Thiswork was supported in part by NASA grants NNX11AO44Gand NAS5-00147.

REFERENCES

Ackermann, M. et al. 2012, Science, 335, 198An, H.J., et al. 2013, ApJ, 775, 135Casares, J., et al. 2005, MNRAS, 364, 899Coley, J.B. et al. 2014, HEAD 1412210Corbet, R. H. D. et al. 2011, ATEL #3221Farr, W. M. et al. 2011, ApJ, 741, 103Gray, R. O. 2009 “A Digital Spectral Classification Atlas v.1.07”

(ned.ipac.caltech.edu/level5/Gray/Gray1.html)

Jascheck, P. & Jascheck, M. 1995, “The Behavior of Chemical Elements inStars”, (Cambridge University Press: Cambridge)

Jenniskens P. et al. 1996, AA, 313, 649Kos, J. & Zwitter, T. 2013 ApJ 774 72McSwain, M. V. et al. 2004, ApJ, 600, 927Napoli, V.J. et al. 2011, PASP, 123, 1293Parimucha, S. & Skoda, P. 2007, IAUS, 240, 486Sarty, G.E. et al. 2011, MNRAS 411, 1293Schwab, C. et al. 2010, SPIE conf. proc., 7735, 77354Skoda, P. et al. 2008, SPIE conf. proc. 7014, 70145