Phlegethon, a nearby 75° -long retrograde stellar stream
Rodrigo A. Ibata, Khyati Malhan, Nicolas F. Martin, Else Starkenburg
DDraft version September 12, 2018
Preprint typeset using L A TEX style emulateapj v. 12/16/11
PHLEGETHON, A NEARBY 75 ◦ -LONG RETROGRADE STELLAR STREAM Rodrigo A. Ibata , Khyati Malhan , Nicolas F. Martin , and Else Starkenburg Draft version September 12, 2018
ABSTRACTWe report the discovery of a 75 ◦ long stellar stream in Gaia DR2 catalog, found using the new STREAMFINDER algorithm. The structure is probably the remnant of a now fully disrupted globularcluster, lies ≈ . . . G ∼ . − ) with a mass of only 2580 ±
140 M (cid:12) , demonstrates the power of the
STREAMFINDER algorithm to detect even very nearby and ultra-faint structures. Due to its proximityand length we expect that Phlegethon will be a very useful probe of the Galactic acceleration field.
Keywords:
Galaxy: halo — Galaxy: stellar content — surveys — galaxies: formation — Galaxy:structure INTRODUCTION
The arrival of the second data release (DR2) of theGaia mission has opened up the field of Galactic Archeol-ogy to exciting new endeavors that were previously com-pletely out of reach. The excellent parallax and propermotion measurements (Gaia Collaboration et al. 2018d)of over a billion stars now allow the the dynamics ofthe various constituents of our Galaxy to be studied ingreat detail (Gaia Collaboration et al. 2018c,b), enablingprogress towards the ultimate goal of understanding theformation of our Galaxy, its stellar components and thedark matter.Among the Galactic components, stellar streams ac-count for only a very minor fraction of the total massbudget, yet their astrophysical interest far exceeds theirsmall contribution to our Galaxy. This importance stemspartly from the fact that they represent fossil remnantsof the accretion events that built up the Milky Way, giv-ing us a means to ascertain the number and provenanceof the building blocks of the Galactic halo (Johnstonet al. 2008). Streams also roughly delineate orbits inthe Galaxy, allowing one to probe the gradient of thegravitational potential (Ibata et al. 2001; Law & Ma-jewski 2010; Varghese et al. 2011; K¨upper et al. 2015;Bovy et al. 2016; Bonaca & Hogg 2018) in a way that isindependent of methods that make the often-unjustifiedassumption of dynamical equilibrium of a tracer popula-tion. Furthermore, the low velocity dispersion and finetransverse width of low-mass streams renders them ex-cellent probes of the small-scale substructure of the darkmatter halo (Ibata et al. 2002; Johnston et al. 2002; Carl-berg 2012; Erkal et al. 2016).In a previous contribution, we introduced a new algo-rithm (the
STREAMFINDER ; Malhan & Ibata 2018, here-after paper I) built specifically to search efficiently Observatoire Astronomique, Universit´e de Strasbourg,CNRS, 11, rue de l’Universit´e, F-67000 Strasbourg, France; [email protected] Max-Planck-Institut f¨ur Astronomie, K¨onigstuhl 17, D-69117 Heidelberg, Germany Leibniz Institute for Astrophyics Potsdam (AIP), An derSternwarte 16, D-14482 Potsdam, Germany through astrometric and photometric databases forstream-like structures. The first results of this algorithmapplied to Gaia DR2 were presented in Malhan et al.(2018, hereafter paper II), but were limited to distances > DATA AND STREAM ANALYSIS
All the data used in the present analysis were drawnfrom the Gaia DR2 catalog (Gaia Collaboration et al.2016, 2018a). As described in paper II, we extinction-corrected the survey using the Schlegel et al. (1998) dustmaps, and kept only those stars with G < .
5; this lim-iting magnitude ensures a homogeneous depth over thesky, given the additional selection criterion of choosingto analyze the sky at | b | > ◦ .A full exposition of the updated STREAMFINDER al-gorithm and contamination model will be provided inIbata et al. (2018, in prep), which includes a demonstra-tion that real structures are easily distinguishable fromfalse positives. Briefly, the algorithm works by examin-ing every star in the survey in turn, sampling the possi-ble orbits consistent with the observed photometry andkinematics, and finding the most likely stream fractiongiven a contamination model and a stream model. Theadopted contamination model is identical to that built forpaper II. It is an empirical model in sky position, color-magnitude and proper motion space (i.e. six-dimensions)constructed by spatially randomizing the Gaia countswith a 2 ◦ Gaussian. While the correlations between skyposition and color-magnitude are recorded as binned ar-rays, the color-magnitude and proper motion informationis condensed by the use of a 100-component Gaussianmixture model in each spatial bin. a r X i v : . [ a s t r o - ph . GA ] S e p Ibata, Malhan, Martin & Starkenburg b [ d e g ] [Fe/H] = 1.4, retrogradePhlegethon 300306090 [deg] b [ d e g ] Phlegethon[Fe/H] = 1.4, retrograde, v tan > 250 kms v tan [km s ] Figure 1.
Output of the
STREAMFINDER software in the distancerange [0 . , .
0] kpc over the region of sky − ◦ < (cid:96) < ◦ , − ◦ < b < − ◦ , assuming a stellar population of age 10 Gyr and[Fe / H] = − .
4. Only those stars are plotted that have retrogradeorbital solutions, and for which the stream detection significanceis greater than 8 σ . The upper panel shows the full range of tan-gential motions, while the bottom panel retains only those starswith v tan >
250 km s − . A clear stream, labelled “Phlegethon”, isdetected as a continuous linear structure in the panels. The colorof the points marks the tangential velocity of the stars, calculatedusing the observed Gaia proper motion in conjunction with thedistance solutions that the algorithm derives for every processedstar. The adopted stream model is very simple: we inte-grate along the sampled orbits for a half-length of 10 ◦ (i.e. total length 20 ◦ ), and over this length the streamcounts have uniform probability. Perpendicular to thestream model, the properties of all observables are takento be Gaussian, so the model has a Gaussian physicalwidth (selected here to be 100 pc), a Gaussian dispersionin both proper motion directions equivalent to 3 km s − ,and a dispersion in distance modulus of 0 .
05 mag. Thesewidth and velocity dispersion parameters are similar tothe properties of known globular cluster streams suchas the Palomar 5 stream (Dehnen et al. 2004; Ibataet al. 2016), while the distance modulus dispersion wasadopted to allow for a small mismatch between the color-magnitude behavior of the adopted stellar populationsmodel and that of the real stellar population.The
STREAMFINDER requires a model of the Galacticpotential in order to calculate orbits; for this we use therealistic Milky Way mass model of Dehnen & Binney(1998), their model ‘1’. The algorithm returns the best-fit orbit out of the sampled set for a given data point,the number of stars in the corresponding putative streamand the likelihood value relative to the model where thestream fraction is zero. All stars shown below have likeli-hoods equivalent to a stream detection exceeding the 8 σ level.
50 40 30 20 10 0 10 [mas yr ] b [ m a s y r ] [ d e g ] Figure 2.
Proper motion distribution of the sample of stars inthe lower panel of Figure 1. The color of the points marks Galacticlongitude, which can be seen to vary continuously along the arc-shaped feature in this distribution. This distinct structure was se-lected to lie within an irregular hand-drawn polygon in this propermotion plane; the resulting sample is shown with the larger (blackcircled) dots.
To convert the kinematics of the integrated orbitsinto the space of observables, we assume that the Sunlies 8 .
20 kpc from the Galactic center and 17 pc abovethe Galactic mid-plane (Karim & Mamajek 2017), thatthe local circular velocity is V circ = 240 km s − andthat the peculiar velocity of the Sun is ( u (cid:12) , v (cid:12) , w (cid:12) ) =(9 . , . , .
0) km s − (Reid et al. 2014; Sch¨onrich et al.2010).The main aim in developing the STREAMFINDER wasto try to find distant halo streams for which the Gaiaparallax measurements are poor. To circumvent this de-ficiency, we use stellar populations models to convert themeasured photometry into trial line of sight distance val-ues. To this end we adopted the PARSEC isochronemodels (Bressan et al. 2012), covering a range in ageand metallicity. Here, however, we report results us-ing a single model with age 10 Gyr and metallicity of[Fe / H] = − . STREAMFINDER au-tomatically include the best-fit orbit over the trial streamlength, from which one can naturally obtain an estimateof the sense of rotation of the survey stars with respect tothe Galaxy. This is possible because the algorithm findsa value for the missing line of sight velocity informationof a star in the Gaia DR2 catalog by requiring continuityto nearby stream candidates. The radial velocity is sam-pled between ± the escape velocity, and the value corre-sponding the the maximum-likelihood stream solution isretained. We showed in paper II that this works for thecase of well-measured streams such as GD-1 (Grillmair& Dionatos 2006). We use this orbital information toclassify stars as either prograde or retrograde. he Phlegethon stellar stream b [ d e g ] proper motion selection [ m a s y r ] best fit orbital model [deg] b [ d e g ] proper motion selection b [ m a s y r ] Figure 3.
Sky distribution of the selection of 321 stars drawnfrom Figure 2. A very clean stream-like structure is revealed bythe proper motion selection. The proper motion distribution alongthe stream in Galactic µ (cid:96) and µ b are color-coded on the top andbottom panels, respectively. The continuity of the proper motiondistribution is directly obvious from the individual proper motionmeasurements. The dashed line in the upper panel shows the pathof the best fit orbit described in Section 4. RESULTS
In Figure 1 (top panel) we show the 358 stars in theGaia DR2 catalog for which the
STREAMFINDER solutionsare retrograde in the area of sky − ◦ < (cid:96) < ◦ , − ◦ < b < − ◦ . The parameter choices discussedpreviously were used. The color of the points encodesthe tangential velocity v tan of the stars, derived from themeasured proper motions and the distances estimatedby the algorithm. A stream feature is clearly present inthis retrograde sample, which becomes even more evidentwhen selecting only those stars with v tan >
250 km s − ;this filtering yields 330 stars, shown on the bottom panel.We name this stream “Phlegethon”, after the river of theGreek Underworld. This structure forms a ∼ ◦ -longarc, skirting, in projection, the southern regions of theBulge.In Figure 2 we display the proper motion distributionof this sample of 330 candidate stream stars, which re-veals a continuous structure with a very large gradientin proper motion. The arc-like feature is highlightedwith large (circled) dots, which are colored accordingto Galactic longitude; the continuous variation in colordemonstrates the coherence in these three parameters.These stars were selected by simply drawing an irregularpolygon in the µ (cid:96) , µ b plane around this visually-obviousgrouping. The distribution on the sky of the 321 sourcesselected in that arc are shown in Figure 3, which revealsa very clean stream-like feature. The proper motions in µ (cid:96) (top panel) are strongly negative, meaning that thestars move towards the right in the Figure. Since thedistance solutions provided by the algorithm predict thatthe structure is located at a mean heliocentric distanceof 3 . ± .
28 kpc, the stars must be strongly lagging theSun. Indeed, towards (cid:96) = 15 ◦ , the proper motion in thelongitude direction reaches µ (cid:96) = −
38 mas yr − , implying [mas] [ m a s ] = 0.266 ± 0.008 mas Figure 4.
Parallax distribution of the stars selected in Figure 2.A maximum-likelihood fit to these 321 stars gives a mean parallaxof (cid:104) (cid:36) (cid:105) = 0 . ± .
008 mas, implying that the structure lies at amean distance of ∼ . a tangential velocity of v tan ∼
640 km s − with respect tothe observer. The proper motions in the Galactic latitudedirection µ b (bottom panel) can be seen to be consistentwith motion away from the plane moving rightwards from (cid:96) = 70 ◦ , reaching µ b = 0 at (cid:96) ∼ ◦ , and then the starsstart moving towards the Galactic plane at (cid:96) < ◦ .If the structure is really as close as the STREAMFINDER software predicts it to be, its distance should be easily re-solvable with Gaia’s parallax measurements. We displaythe parallaxes (cid:36) , along with their uncertainties δ(cid:36) inFigure 4. We implemented a simple maximum likelihoodalgorithm to calculate the mean parallax, assuming indi-vidual Gaussian parallax uncertainties on each star. Themean parallax was found to be (cid:104) (cid:36) (cid:105) = 0 . ± .
008 mas,i.e. ∼ . STREAMFINDER solutions.The color-magnitude distribution of the 321 stars inthe sample is displayed in Figure 5. These can beseen to conform well to the input template stellar pop-ulation model, with most stars lying on the main se-quence, and just a handful on the lower red giant branch.Given that the stream is approximately 75 ◦ long and 4 ◦ wide, we deduce a system surface brightness of Σ G =34 . − . Adopting the template stellar popu-lation model, and accounting for the fainter (unobserved)stars, implies a mass of 2580 ±
140 M (cid:12) over the observedregion, where we have assumed that the Gaia DR2 iscomplete to G = 19 . Ibata, Malhan, Martin & Starkenburg ( G BP G RP ) [mag] G [ m a g ] Padova 10Gyr, [Fe/H] = 1.4Phlegethon
Figure 5.
Color-magnitude distribution of the sample selected inFigure 2. The dots show the de-reddened Gaia G vs. G BP − G RP photometry of the stars in the stream. By construction, theseshould follow the chosen template stellar population model (greenline), but it is nevertheless a useful check to verify that the stars arenot clumped in an unphysical way on the CMD. The Padova modelshown here has been shifted to account for a distance modulus of12 .
88 mag, as measured in Figure 4.
20 15 10 5 0 5 10 15 20 x [kpc] y [ k p c ] R [kpc] z [ k p c ] Figure 6.
The orbital path of the Phlegethon stream. TheMCMC procedure described in the text was used to fit the fullsample simultaneously, yielding the best-fit orbit shown here. The x − y plane is shown on the left panel, and the R − z on the rightpanel, with the blue dots showing the path 1 Gyr into the pastand 1 Gyr into the future. The short red region corresponds to thearea where the observed stream stars are currently located. Theprogenitor clearly stayed close to the plane of the Galaxy on a ret-rograde tube orbit. In this Cartesian system, the Galactic centeris at the origin, and the Sun (marked with a large yellow dot) liesat ( x, y, z ) = ( − . , , . of − . ± . − and − . ± . − , respec-tively. They both lie very close to one another ( (cid:96) = 54 . ◦ and (cid:96) = 54 . ◦ , respectively). While the metallicity of thefirst star is unconstrained, the second star has a mea-sured spectroscopic metallicity of [Fe / H] = − . ± . REFINED ORBITAL FITS
Although
STREAMFINDER fits the orbits of the putativestreams, it is designed for stream detection rather thancareful fitting. We therefore updated the Lagrange-pointstripping method described in Varghese et al. (2011), butnow using the measured proper motions and parallax in-formation as constraints. The measured radial veloci-ties of the two SDSS stars were also used in the fittingprocedure. We dealt with the missing radial velocity in-formation for the remaining 319 stars by assigning themall a velocity of zero, but with a Gaussian uncertainty of10 km s − . We also updated the fitting procedure to usea custom-made MCMC driver package, discussed previ-ously in Ibata et al. (2013), that uses the affine-invariantensemble sampling method of Goodman & Weare (2010).Since we do not know the location of the progenitor rem-nant, we ignored the self-gravity of the stream in themodeling, i.e. we fitted orbits over the full length of thedetected structure.The fitting parameters are sky position α , δ , distance d , heliocentric velocity v h and proper motions µ α , µ δ . Wechose to anchor the solutions at δ = − ◦ , approximatelyhalf-way along the stream, leaving all other parametersfree to be varied by the algorithm (obviously, without ananchor line, the solution would have wandered over thefull length of the stream).After rejecting a burn-in phase of 10 steps, the MCMCprocedure was run for a further 10 iterations. The best-fit orbit is shown in Figure 6, integrated over a period of2 Gyr in the same Dehnen & Binney (1998) model em-ployed above for stream detection. The red region marksthe part of the orbit where we have currently detectedthe structure. The orbit is disk-like, but strongly retro-grade, possessing an apocenter at R = 19 . ± . . ± .
01 kpc and a maximum height fromthe Galactic plane of 7 . ± . ∼ . DISCUSSION AND CONCLUSIONS
In this paper, we present the discovery of a new streamstructure, that we named Phlegethon, found in the re-cently released Gaia DR2 dataset. Phlegethon, whichlies close to the Sun at a mean heliocentric distance of3 . . ◦ , corresponding to 88 pc at that helio-centric distance (for comparison, the Palomar 5 streamhas a dispersion of 58 pc, Ibata et al. 2016). This impliesthat the Phlegethon stream must be the remnant of adisrupted globular cluster rather than a dwarf galaxy.To ascertain whether Phlegethon could be related toany surviving globular cluster, in Figure 8 we comparethe orbital properties ( z -component of angular momen-tum L z and pericenter distance) of this stream with theglobular cluster sample recently analyzed by Gaia Col- he Phlegethon stellar stream [ m a s y r ] best fit orbital model [ m a s y r ] [ m a s ] [deg] v h e li o [ k m s ] Figure 7.
Comparison of the orbital solution (black dashed lines)to Gaia DR2 data (plotted in red along with their observationaluncertainties). The best orbit shown previously in Figure 6 is re-produced here for a visual comparison to the data. The modelcaptures the observed profiles in µ α , µ δ and (cid:36) , although someslight discrepancies towards the ends of the stream are apparent,possibly due to a mismatch between the real potential and themodel used here. The heliocentric radial velocity as a function ofGalactic longitude is shown on the bottom panel (this informationis missing in the Gaia DR2 catalog), but the positions of the twostars of our sample that were fortuitously measured by the SDSSare shown in blue. The orbital path in Galactic coordinates of thisbest-fit model was shown previously in Figure 3 (top panel). laboration et al. (2018b). We have labeled certain well-known retrograde globular clusters. Phlegethon clearlyhas extreme properties ( L z = 1769 ± − kpc inthe adopted model of the Galactic potential), with onlyNGC 3201 having a larger value of L z . Most impor-tantly, however, this analysis indicates that there are noknown globular clusters with similar dynamical proper-ties to this stream, so the progenitor has probably notsurvived (unless by chance it is hidden in the Galacticdisk).The very low mass of 2580 ±
140 M (cid:12) that we measurehere, and the old stellar population age, raise an interest-ing puzzle. Clearly, dynamical friction must have beencompletely unimportant in affecting the orbit of the glob-ular cluster progenitor, unless it arrived as part of a muchmore massive system, such as a dwarf galaxy. But the ab- L z [km s kpc] r p e r i [ k p c ] NGC 3201NGC 5466Omega Cen
Gaia globular clustersPhlegethon
Figure 8.
Orbital properties of the Phlegethon stream comparedto Milky Way globular clusters. The Galactic globular clusterswith orbits measured by Gaia (Gaia Collaboration et al. 2018b)are shown here in the L z r peri plane. Retrograde objects havepositive L z . The properties of Phlegethon are shown by a large reddot: it can be seen to lie far from any other globular cluster, so weconclude that the progenitor system is now completely dissolved.The uncertainties on L z and r peri derived here using the adoptedDehnen & Binney (1998) Galactic mass model are smaller than thesize of the dot; however, the real uncertainties on these values arelikely to be dominated by the reliability of our current knowledgeof the mass distribution in the Milky Way, which Phlegethon maysoon help to improve. sence of a large population of stars on similar retrogradeorbits suggests that this was not the case. Thus it wouldseem that the progenitor must have formed early in thelife of the Milky Way, and continued orbiting in this disk-like but retrograde manner. At some point over the last10 Gyr (the age of the stellar population as suggested bythe CMD properties of the stream) the progenitor be-came unbound, and began forming the stream we havedetected. It will be interesting to simulate this systemto determine the dynamical age of the stream, and givenits smooth appearance on the sky (Figure 3), examinethe expected effect of heating by dark matter sub-halosand the repeated crossings every ∼
200 Myr of the denseinner regions of the Galactic disk.It may prove very rewarding to hunt over the sky forstars with similar orbital properties. This would increasethe lever-arm for orbital-fit analyses, allowing one to bet-ter test Galactic mass models. It will also allow im-proved constraints to be placed on the initial mass ofPhlegethon’s progenitor.The detection of Phlegethon shows that the
STREAMFINDER algorithm can be successfully em-ployed to find even nearby structures of exceedinglylow surface brightness (Σ G ∼ . − ). Sofar, in paper II and in the present contribution wehave analyzed only a small part of the parameter spaceof possible stream structures; our next efforts willfocus on expanding the search criteria, and trying toexplore further these fossil remnants in the Milky Way, Ibata, Malhan, Martin & Starkenburg examining their implications for galaxy formation andthe dark matter problem.This work has made use of data from theEuropean Space Agency (ESA) mission
Gaia ( ), processedby the Gaia
Data Processing and Analysis Consortium(DPAC, ). Funding for the DPAC has beenprovided by national institutions, in particular theinstitutions participating in the
Gaia