giant magnetized outflows from the centre of the milky way

LETTERdoi:10.1038/nature11734

Giant magnetized outflows from the centre of theMilky WayEttore Carretti1, Roland M. Crocker2,3, Lister Staveley-Smith4,5, Marijke Haverkorn6,7, Cormac Purcell8, B. M. Gaensler8,Gianni Bernardi9, Michael J. Kesteven10 & Sergio Poppi11

The nucleus of the Milky Way is known to harbour regions ofintense star formation activity as well as a supermassive black hole1.Recent observations have revealed regions of c-ray emission reach-ing far above and below the Galactic Centre (relative to the Galacticplane), the so-called ‘Fermi bubbles’2. It is uncertain whether thesewere generated by nuclear star formation or by quasar-like out-bursts of the central black hole3–6 and no information on the struc-tures’ magnetic field has been reported. Here we report observationsof two giant, linearly polarized radio lobes, containing three ridge-like substructures, emanating from the Galactic Centre. The lobeseach extend about 60 degrees in the Galactic bulge, closely corres-ponding to the Fermi bubbles, and are permeated by strong mag-netic fields of up to 15 microgauss. We conclude that the radio lobesoriginate in a biconical, star-formation-driven (rather than black-hole-driven) outflow from the Galaxy’s central 200 parsecs thattransports a huge amount of magnetic energy, about 1055 ergs, intothe Galactic halo. The ridges wind around this outflow and, wesuggest, constitute a ‘phonographic’ record of nuclear star forma-tion activity over at least ten million years.

We use the images of the recently concluded S-band Polarization AllSky Survey (S-PASS) that has mapped the polarized radio emission ofthe entire southern sky. The survey used the Parkes Radio Telescope ata frequency of 2,307 MHz, with 184 MHz bandwidth, and 99 angularresolution7.

The lobes we report here exhibit diffuse polarized emission (Fig. 1),an integrated total intensity flux of 21 kJy, and a high polarizationfraction of 25%. They trace the Fermi bubbles excepting the top western(that is, right) corners where they extend beyond the region covered bythe c-ray emission structure. Depolarization by H II regions establishesthat the lobes are almost certainly associated with the Galactic Centre(Fig. 2 and Supplementary Information), implying that their height is,8 kpc. Archival data of WMAP8 reveal the same structures at amicrowave frequency of 23 GHz (Fig. 3). The 2.3–23 GHz spectralindex a (with flux density S at frequency n modelled as Sn / na) oflinearly polarized emission interior to the lobes spans the range 21.0 to21.2, generally steepening with projected distance from the Galacticplane (see Supplementary Information). Along with the high polari-zation fraction, this phenomenology indicates that the lobes are dueto cosmic-ray electrons, transported from the plane, synchrotron-radiating in a partly ordered magnetic field.

Three distinct emission ridges that all curve towards Galactic westwith increasing Galactic latitude are visible within the lobes (Fig. 1);two other substructures proceeding roughly northwest and southwestfrom around the Galactic Centre hint at limb brightening in the bico-nical base of the lobes. These substructures all have counterparts inWMAP polarization maps (Fig. 3), and one of them9, already known

from radio continuum data as the Galactic Centre spur10, appears toconnect back to the Galactic Centre; we label the other substructuresthe northern and southern ridges. The ridges’ magnetic field directions(Fig. 3) curve, following their structures. The Galactic Centre spur andsouthern ridges also seem to have GeV c-ray counterparts (Fig. 2; alsocompare ref. 3). The two limb brightening spurs at the biconical lobebase are also visible in the WMAP map, where they appear to con-nect back to the Galactic Centre area. A possible third spur developsnortheast from the Galactic Centre. These limb brightening spurs arealso obvious in the Stokes U map as an X-shaped structure centred atthe Galactic Centre (Supplementary Fig. 3).

Such coincident, non-thermal radio, microwave and c-ray emissionindicates the presence of a non-thermal electron population coveringat least the energy range 1–100 GeV (Fig. 4) that is simultaneouslysynchrotron-radiating at radio and microwave frequencies andupscattering ambient radiation into c-rays by the inverse Comptonprocess. The widths of the ridges are remarkably constant at ,300 pcover their lengths. The ridges have polarization fractions of 25–31%(see Supplementary Information), similar to the average over the lobes.Given this emission and the stated polarization fractions, we infermagnetic field intensities of 6–12mG for the lobes and 13–15mG forthe ridges (see Figs 2 and 3, and Supplementary Information).

An important question about the Fermi bubbles is whether they areultimately powered by star formation or by activity of the Galaxy’scentral, supermassive black hole. Despite their very large extent, thec-ray bubbles and the X-shaped polarized microwave and X-ray struc-tures tracing their limb-brightened base11 have a narrow waist ofonly 100–200 pc diameter at the Galactic Centre. This matches theextent of the star-forming molecular gas ring (of ,3 3 107 solarmasses) recently demonstrated to occupy the region12. With 5–10%of the Galaxy’s molecular gas content1, star-formation activity in this‘central molecular zone’ is intense, accelerating a distinct cosmic raypopulation13,14 and driving an outflow11,15 of hot, thermal plasma,cosmic rays and ‘frozen-in’ magnetic field lines6,14,16.

One consequence of the region’s outflow is that the cosmic rayelectrons accelerated there (dominantly energized by supernovae) areadvected away before they lose much energy radiatively in situ14,16,17.This is revealed by the fact that the radio continuum flux on scales upto 800 pc around the Galactic Centre is in anomalous deficit withrespect to the expectation afforded by the empirical far-infrared/radiocontinuum correlation18. The total 2.3 GHz radio continuum fluxfrom the lobes of ,21 kJy, however, saturates this correlation as nor-malized to the 60 mm flux (2 MJy) of the inner ,160 pc diameterregion (ref. 19). Together with the morphological evidence, thisstrongly indicates that the lobes are illuminated by cosmic ray elec-trons accelerated in association with star formation within this region

1CSIRO Astronomy and Space Science, PO Box 276, Parkes, New South Wales 2870, Australia. 2Max-Planck-Institut fur Kernphysik, PO Box 103980, 69029 Heidelberg, Germany. 3Research School ofAstronomy and Astrophysics, Australian National University, Weston Creek, Australian Capital Territory 2611, Australia. 4International Centre for Radio Astronomy Research, M468, University of WesternAustralia, Crawley, Western Australia 6009, Australia. 5ARC Centre of Excellence for All-sky Astrophysics (CAASTRO), M468, University of Western Australia, 35 Stirling Highway, Crawley, Western Australia6009, Australia. 6Department of Astrophysics/IMAPP, Radboud University Nijmegen, PO Box 9010, 6500 GL Nijmegen, The Netherlands. 7Leiden Observatory, Leiden University, PO Box 9513, 2300 RALeiden, The Netherlands. 8Sydney Institute for Astronomy, School of Physics, The University of Sydney, New South Wales 2006, Australia. 9Harvard-Smithsonian Center for Astrophysics, 60 Garden Street,Cambridge, Massachusetts 02138, USA. 10CSIRO Astronomy and Space Science, PO Box 76, Epping, New South Wales 1710, Australia. 11INAF Osservatorio Astronomico di Cagliari, Strada 54 LocalitaPoggio dei Pini, I-09012 Capoterra (CA), Italy.

6 6 | N A T U R E | V O L 4 9 3 | 3 J A N U A R Y 2 0 1 3

Macmillan Publishers Limited. All rights reserved©2013

www.nature.com/doifinder/10.1038/nature11734

(see Supplementary Information), and that the lobes are not a result ofblack hole activity.

The ridges appear to be continuous windings of individual, colli-mated structures around a general biconical outflow out of the GalacticCentre. The sense of Galactic rotation (clockwise as seen from Galacticnorth) and angular-momentum conservation mean that the ridgesget ‘wound up’20 in the outflow with increasing distance from theplane, explaining the projected curvature of the visible, front-sideof the ridges towards Galactic west. Polarized, rear-side emission isattenuated, rendering it difficult to detect against the stronger emissionfrom the lobes’ front-side and the Galactic plane (Fig. 1 and Sup-plementary Information).

For cosmic ray electrons synchrotron-emitting at 2.3 GHz to be ableto ascend to the top of the northern ridge at ,7 kpc in the time it takesthem to cool (mostly via synchrotron emission itself) requires verticaltransport speeds of .500 km s21 (for a field of 15mG; see Fig. 4). Giventhe geometry of the Galactic Centre spur, the outflowing plasma ismoving at 1,000–1,100 km s21 (Fig. 4 and Supplementary Infor-mation), somewhat faster than the ,900 km s21 gravitational escapevelocity from the Galactic Centre region21, implying that 2.3-GHz-radiating electrons can, indeed, be advected to the top of the ridgesbefore they lose all their energy.

Given the calculated fields and the speed of the outflow, the totalmagnetic energy for each of the ridges, (4–9) 3 1052 erg (see Sup-plementary Information), is injected at a rate of ,1039 erg s21 over afew million years; this is very close to the rate at which independentmodelling6 suggests Galactic Centre star formation is injecting mag-netic energy into the region’s outflow. On the basis of the ridges’ indi-vidual energetics, geometry, outflow velocity, timescales and plasmacontent (see Supplementary Information), we suggest that their foot-points are energized by and rotate with the super-stellar clusters inha-biting1 the inner ,100 pc (in radius) of the Galaxy. In fact, we suggestthat the ridges constitute ‘phonographic’ recordings of the past,10 Myr of Galactic Centre star formation. Given its morphology,the Galactic Centre spur probably still has an active footprint. In con-trast, the northern and southern ridges seem not to connect to theplane at 2.3 GHz. This may indicate their footpoints are no longeractive, though the southern ridge may be connected to the plane bya c-ray counterpart (see Fig. 2). Unfortunately, present data do notallow us to trace the Galactic Centre spur all the way down to theplane: but a connection is plausible between this structure and one(or some combination) of the ,1u-scale radio continuum spurs15,22

emanating north of the star-forming giant molecular cloud complexesSagittarius B and C; a connection is also plausible with the bright,

0 0.017 0.034 0.051 0.068 0.085

P (Jy per beam)

0.1 0.12 0.14 0.15 0.17

Southern ridge

Galactic Centre spur

Northern ridge

Limb brightening spurs

Figure 1 | Linearly polarized intensity P at 2.3 GHz from S-PASS. The thickdashed lines delineate the radio lobes reported in this Letter, while the thindashed lines delimit the c-ray Fermi bubbles2. The map is in Galacticcoordinates, centred at the Galactic Centre with Galactic east to the left andGalactic north up; the Galactic plane runs horizontally across the centre of themap. The linearly polarized intensity flux density P (a function of the Stokesparameters Q and U,P:

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

Q2zU2p

) is indicated by the colour scale, and givenin units of Jy per beam with a beam size of 10.759 (1 Jy ; 10226 W m22 Hz21).The lobes’ edges follow the c-ray border up to Galactic latitude b < | 30 |u, fromwhich the radio emission extends. The three polarized radio ridges discussed inthe text are also indicated, along with the two limb brightening spurs. The

ridges appear to be the front side of a continuous winding of collimatedstructures around the general biconical outflow of the lobes (see text). TheGalactic Centre spur is nearly vertical at low latitude, possibly explained by aprojection effect if it is mostly at the front of the northern lobe. At its higherlatitudes, the Galactic Centre spur becomes roughly parallel with the northernridge (above), which itself exhibits little curvature; this is consistent with theoverall outflows becoming cylindrical above 4–5 kpc as previously suggested11.In such a geometry, synchrotron emission from the rear side of each cone isattenuated by a factor>2 with respect to the front side, rendering it difficult todetect the former against the foreground of the latter and of the Galactic plane(see Supplementary Information).

LETTER RESEARCH

3 J A N U A R Y 2 0 1 3 | V O L 4 9 3 | N A T U R E | 6 7


non-thermal ‘radio arc’1 (itself longitudinally coincident with the,4-Myr-old Quintuplet23 stellar cluster).

The magnetic energy content of both lobes is much larger than theridges, (1–3) 3 1055 erg. This suggests the magnetic fields of the lobesare the result of the accumulation of a number of star formationepisodes. Alternatively, if the lobes’ field structure were formed overthe same timescale as the ridges, it would have to be associated with

recent activity of the supermassive black hole, perhaps occurring inconcert with enhanced nuclear star-formation activity4.

Our data indicate that the process of gas accretion onto the Galacticnucleus inescapably involves star formation which, in turn, energizesan outflow. This carries away low-angular-momentum gas, cosmicrays and magnetic field lines, and has a number of important conse-quences. First, the dynamo activity in the Galactic Centre24, probablyrequired to generate its strong17 in situ field, requires the continualexpulsion of small-scale helical fields to prevent dynamo saturation25;the presence of the ridges high in the halo may attest to this process.Second, the lobes and ridges reveal how the very active star formationin the Galactic Centre generates and sustains a strong, large-scalemagnetic field structure in the Galactic halo. The effect of this on

0 0.017 0.034 0.051 0.068 0.085P (Jy per beam)

0.1 0.14 0.15 0.17

γ-ray spur

γ-ray spur

Depolarization/modulation area

0.12

0.000100.00012

Brightness temperature (K)0.0

Northern ridge

NW limb brightening

SW limb brighteningNE limb brightening

Galactic Centre spur

Southern ridge

Figure 3 | Polarized intensity and magnetic angles at 23 GHz fromWMAP8. The magnetic angle is orthogonal to the emission polarization angleand traces the magnetic field direction projected on to the plane of the sky(headless vector lines). The three ridges are obvious while traces of the radiolobes are visible (2.3 GHz edges shown by the black solid line). The magneticfield is aligned with the ridges and curves following their shape. Two spursmatch the lobe edges northwest and southwest of Galactic Centre and could belimb brightening of the lobes. A third limb brightening spur candidate is alsovisible northeast of the Galactic Centre. The map is in Galactic coordinates,

centred at the Galactic Centre. Grid lines are spaced by 15u. The emissionintensity is plotted as brightness temperature, in K. The vector line length isproportional to the polarized brightness temperature (the scale is shown by theline in the bottom-left corner, in K). Data have been binned in 1u3 1u pixels toimprove the signal-to-noise ratio. From a combined analysis of microwave andc-ray data (see also Supplementary Information) we can derive the followingmagnetic field limits (complementary to the equipartition limits reported in thetext and Fig. 2): for the overall lobes/bubbles, B . 9mG; and for the GalacticCentre spur, 11mG , B , 18mG.

Figure 2 | Lobes’ polarized intensity and c-ray spurs. Schematic rendering ofthe edges of two c-ray substructures evident in the 2–5 GeV Fermi data asdisplayed in figure 2 of ref. 2, which seem to be counterparts of the GalacticCentre spur and the southern ridge. The map is in Galactic coordinates, withGalactic east to the left and Galactic north up; the Galactic plane runshorizontally across the centre of the map approximately. The linearly polarizedintensity flux density P is indicated by the colour scale, and given in units of Jyper beam with a beam size of 10.759. The latter appears to be connected to theGalactic Centre by its c-ray counterpart. With the flux densities andpolarization fraction quoted in the text, we can infer equipartition26 magneticfield intensities of Beq < 6mG (1mG ; 10210 T) if the synchrotron-emittingelectrons occupy the entire volume of the lobes, or ,12mG if they occupy only a300-pc-thick skin (the width of the ridges). For the southern ridge, Beq < 13mG;for the Galactic Centre spur, Beq < 15mG; and, for the northern ridge,Beq < 14mG. Note the large area of depolarization and small-angular-scalesignal modulation visible across the Galactic plane extending up to | b | < 10uon either side of the Galactic Centre (thin dashed line). This depolarization isdue to Faraday rotation by a number of shells that match Ha emission regions27,most of them lying in the Sagittarius arm at distances from the Sun up to2.5 kpc, and some in the Scutum-Centaurus arm at ,3.5 kpc. The small-scalemodulation is associated with weaker Ha emission encompassing the sameH II regions and most probably associated with the same spiral arms. Thus2.5 kpc constitutes a lower limit to the lobes’ near-side distance and placesthe far side beyond 5.5 kpc from the Sun (compare ref. 9). Along with theirdirection in the sky, this suggests that the lobes are associated with the Galacticbulge and/or Centre.

RESEARCH LETTER

6 8 | N A T U R E | V O L 4 9 3 | 3 J A N U A R Y 2 0 1 3


the propagation of high-energy cosmic rays in the Galactic halo shouldbe considered. Third, the process of gas expulsion in the outflow mayexplain how the Milky Way’s supermassive black hole is kept relativelyquiescent1, despite sustained, inward movement of gas.

Received 8 August; accepted 26 October 2012.

1. Morris, M. & Serabyn, E. The Galactic Centre environment. Annu. Rev. Astron.Astrophys. 34, 645–701 (1996).

2. Su, M., Slatyer, T. R. & Finkbeiner, D. P. Giant gamma-ray bubbles from Fermi-LAT:active galactic nucleus activity or bipolar galactic wind? Astrophys. J. 724,1044–1082 (2010).

3. Su, M. & Finkbeiner, D. P. Evidence for gamma-ray jets in the Milky Way. Astrophys.J. 753, 61 (2012).

4. Zubovas, K., King, A. R. & Nayakshin, S. The Milky Way’s Fermi bubbles: echoes ofthe last quasar outburst? Mon. Not. R. Astron. Soc. 415, L21–L25 (2011).

5. Crocker, R. M. & Aharonian, F. Fermi bubbles: giant, multibillion-year-oldreservoirs of Galactic Center cosmic rays. Phys. Rev. Lett. 106, 101102 (2011).

6. Crocker, R. M. Non-thermal insights on mass and energy flows through theGalactic Centre and into the Fermi bubbles. Mon. Not. R. Astron. Soc. 423,3512–3539 (2012).

7. Carretti, E. in The Dynamic ISM: A Celebration of the Canadian Galactic Plane Survey(eds Kothes, R., Landecker, T. L. & Willis, A. G.) 276–287 (ASP Conf. Ser. CS-438,Astronomical Society of the Pacific, 2011).

8. Hinshaw, G. et al. Five-year Wilkinson Microwave Anisotropy Probe observations:data processing, sky maps, and basic results. Astrophys. J. 180 (suppl.), 225–245(2009).

9. Jones,D. I., Crocker, R.M., Reich,W., Ott, J.& Aharonian, F. A. Magnetic substructurein the northern Fermi bubble revealed by polarized microwave emission.Astrophys. J. 747, L12–L15 (2012).

10. Sofue, Y., Reich, W. & Reich, P. The Galactic center spur — A jet from the nucleus?Astrophys. J. 341, L47–L49 (1989).

11. Bland-Hawthorn, J. & Cohen, M. The large-scale bipolar wind in the GalacticCenter. Astrophys. J. 582, 246–256 (2003).

12. Molinari, S. et al. A 100 pc elliptical and twisted ring of cold and dense molecularclouds revealed by Herschel around the Galactic Center. Astrophys. J. 735,L33–L39 (2011).

13. Aharonian, F. A. et al.Discovery of very-high-energy c-rays from the Galactic Centreridge. Nature 439, 695–698 (2006).

14. Crocker, R. M. et al. Wild at heart: the particle astrophysics of the Galactic Centre.Mon. Not. R. Astron. Soc. 413, 763–788 (2011).

15. Law, C. J. A multiwavelength view of a mass outflow from the Galactic Center.Astrophys. J. 708, 474–484 (2010).

16. Crocker, R.M.et al. c-raysand the far-infrared-radio continuum correlation reveal apowerful Galactic Centre wind. Mon. Not. R. Astron. Soc. 411, L11–L15 (2011).

17. Crocker, R. M., Jones, D. I., Melia, F., Ott, J. & Protheroe, R. J. A lower limit of 50microgauss for the magnetic field near the Galactic Centre. Nature 463, 65–67(2010).

18. Yun, M. S., Reddy, N. A. & Condon, J. J. Radio properties of infrared-selectedgalaxies in the IRAS 2 Jy sample. Astrophys. J. 554, 803–822 (2001).

19. Launhardt, R., Zylka, R. & Mezger, P. G. The nuclear bulge of the Galaxy III. Largescale physical characteristics of stars and interstellar matter. Astron. Astrophys.384, 112–139 (2002).

20. Heesen, V., Beck, R., Krause, M. & Dettmar, R.-J. Cosmic rays and the magnetic fieldin the nearby starburst galaxy NGC 253 III. Helical magnetic fields in the nuclearoutflow. Astron. Astrophys. 535, A79 (2011).

21. Muno, M. P. et al. Diffuse X-ray emission in a deep Chandra image of the GalacticCenter. Astrophys. J. 613, 326–342 (2004).

22. Pohl, M., Reich, W. & Schlickeiser, R. Synchrotron modelling of the 400 pc spur atthe galactic center. Astron. Astrophys. 262, 441–454 (1992).

23. Hußmann, B., Stolte, A., Brandner, W. & Gennaro, M. The present-day massfunction of the Quintuplet cluster. Astron. Astrophys. 540, A57 (2012).

24. Ferriere, K. Interstellar magnetic fields in the Galactic center region. Astron.Astrophys. 505, 1183–1198 (2009).

25. Brandenburg, A. & Subramanian, K. Astrophysical magnetic fields and nonlineardynamo theory. Phys. Rep. 417, 1–209 (2005).

26. Beck, R. & Krause, M. Revised equipartition and minimum energy formula formagnetic field strength estimates from radio synchrotron observations. Astron.Nachr. 326, 414–427 (2005).

27. Gaustad, J. E., McCullough, P. R., Rosing, W. & Van Buren, D. A robotic wide-angleHa survey of the southern sky. Publ. Astron. Soc. Pacif. 113, 1326–1348 (2001).

28. Kulsrud, R. & Pearce, W. P. The effect of wave-particle interactions on thepropagation of cosmic rays. Astrophys. J. 156, 445–469 (1969).

29. Almy, R. C. et al. Distance limits on the bright X-ray emission toward the GalacticCenter: evidence for a very hot interstellar medium in the galactic X-ray bulge.Astrophys. J. 545, 290–300 (2000).

Supplementary Information is available in the online version of the paper.

Acknowledgements This work has been carried out in the framework of the S-bandPolarization All Sky Survey collaboration (S-PASS). We thank the Parkes Telescope stafffor support, both while setting up the non-standard observing mode and during theobserving runs. R.M.C. thanks F. Aharonian, R. Beck, G. Bicknell, D. Jones, C. Law,M. Morris, C. Pfrommer, W. Reich, A. Stolte, T. Porter and H. Volk for discussions, and theMax-Planck-Institut fur Kernphysik for supporting his research. R.M.C. alsoacknowledges the support of a Future Fellowship from the Australian Research Councilthroughgrant FT110100108. B.M.G. andC.P. acknowledge the support of an AustralianLaureate Fellowship from the Australian Research Council through grantFL100100114. M.H. acknowledges the support of research programme 639.042.915,which is partly financed by the Netherlands Organisation for Scientific Research (NWO).The Parkes Radio Telescope is part of the Australia Telescope National Facility, which isfunded by the Commonwealth of Australia for operation as a National Facility managedby CSIRO. We acknowledge the use of WMAP data and the HEALPix software package.

Author Contributions E.C. performed the S-PASS observations, was the leader of theproject, developed and performed the data reduction package, and did the mainanalysis and interpretation. R.M.C. provided theoretical analysis and interpretation.L.S.-S., M.H. and S.P. performed the S-PASS observations. M.J.K. performed thetelescope special set-up that allowed the survey execution. L.S.-S., M.H., B.M.G., G.B.,M.J.K. and S.P. were co-proposers and contributed to the definition of the project. C.P.performed the estimate of the Ha depolarizing region distance. E.C. and R.M.C. wrotethe paper together. All the authors discussed the results and commented on themanuscript.

Author Information Reprints and permissions information is available atwww.nature.com/reprints. The authors declare no competing financial interests.Readers are welcome to comment on the online version of the paper. Correspondenceand requests for materials should be addressed to E.C. ([email protected]).

8 9 10log[Ee (eV)]

6 μG

15 μG

2.3G

Hz,

15

μG

23G

Hz,

15

μG

1 G

eV (I

C)

50 G

eV (I

C)log[

Ran

ge (p

c)]

11 122

3

4

5

6

Figure 4 | The vertical range of cosmic ray electrons as a function of theirkinetic energy, Ee. Two cases are reported, for field amplitudes of 15 and 6mG(blue and red curves, respectively). Owing to geometrical uncertainties, adiabaticlosses cannot be determined so the plotted range (y axis) actually constitutes anupper limit. Electrons are taken to be transported with a speed given by the sumof the inferred vertical wind speed (1,100 km s21) and the vertical componentof the Alfven velocity in the magnetic field. The former is inferred from thegeometry of the northern ridge: if its footpoint has executed roughly half anorbit in the time the Galactic Centre spur has ascended to its total height of,4 kpc, its upward velocity must be close to9 1,000 km s21 3 (r/100 pc)21 3

vrot/(80 km s21), where we have normalized to a footpoint rotation speed of80 km s21 at a radius of 100 pc from the Galactic Centre12 (detailed analysisgives 1,100 km s21: see Supplementary Information). In a strong, regularmagnetic field, the electrons are expected to stream ahead of the gas at theAlfven velocity28 in either the ridges (B<15 mG, vvert

A <300 km s{1; this is alower limit given that nH=0:008 cm{3 on the basis of the ROSAT data29) or inthe large-scale field of the lobes (B<6 mG, vvert

A >100 km s{1 fornH=0:004 cm{3 in the lobes’ interior as again implied by the data). Alsoplotted as the vertical dashed lines are the characteristic energies of electronssynchrotron radiating at 2.3 and 23 GHz (for a 15 mG field) and into 1-GeV and50-GeV c-rays via inverse Compton (‘IC’) upscattering of a photonbackground with characteristic photon energy 1 eV; and the approximate 7 kpcdistance of the top of the northern ridge from the Galactic plane.

LETTER RESEARCH

3 J A N U A R Y 2 0 1 3 | V O L 4 9 3 | N A T U R E | 6 9



www.nature.com/reprints


mailto:[email protected]

W W W. N A T U R E . C O M / N A T U R E | 1

SUPPLEMENTARY INFORMATIONdoi:10.1038/nature11734

Supplementary Information to ‘Giant Magnetized Outflows from the Centre of theMilky Way’

Ettore Carretti,1 Roland M. Crocker,2,3 Lister Staveley-Smith4,5, Marijke Haverkorn6,7, Cormac Purcell8,B. M. Gaensler8, Gianni Bernardi9, Michael J. Kesteven10, and Sergio Poppi11

1CSIRO Astronomy and Space Science, PO Box 276, Parkes, NSW 2870, Australia2Max-Planck-Institut fur Kernphsik, P.O. Box 103980 Heidelberg, Germany3Research School of Astronomy & Astrophysics, Australian National University, Weston Creek, ACT 2611,Australia4International Centre for Radio Astronomy Research, M468, University of Western Australia, Crawley, WA6009, Australia5ARC Centre of Excellence for All-Sky Astrophysics6Department of Astrophysics/IMAPP, Radboud University Nijmegen, P.O. Box 9010, 6500 GL Nijmegen,The Netherlands7Leiden Observatory, Leiden University, P.O. Box 9513, 2300 RA Leiden, The Netherlands8Sydney Institute for Astronomy, School of Physics, The University of Sydney, NSW 2006, Australia9Harvard–Smithsonian Center for Astrophysics, 60 Garden Street, Cambridge, MA, 02138, USA10CSIRO Astronomy and Space Science, PO Box 76, Epping, NSW 1710, Australia11INAF Osservatorio Astronomico di Cagliari, St. 54 Loc. Poggio dei Pini, I-09012 Capoterra (CA), Italy

1 Data

We use the images of the recently concluded S-band Polarization All Sky Survey (S-PASS) that has mappedthe polarized radio emission of the entire southern sky with the Parkes Radio Telescope at a frequency of2307 MHz, with 184 MHz bandwidth, and 9’ angular resolution7 . See Table S1 for a description of theflux densities from the various objects we identify in the main text. For historical interest, Parkes telescopeinvestigations of the Galactic Centre (GC) – and even the discovery of a potential outflow from this region– date back some half a century30.

1.1 Inference of total intensity flux densities at 2.3 GHz

Confusion with Galactic foregrounds, especially free–free and HII regions, means that we cannot directlymeasure the total intensity flux density of the Lobes for |b|<∼ 15◦. We follow the following procedure tocircumvent this problem and estimate the total intensity flux density of the whole Lobes:

1. We measure the integrated polarized intensity of both (whole) Lobes (emission within the edges).

1structure solid angular 2.3 GHz pol. 2.3 GHzname angle width pol. flux frac. total flux

[deg2] [deg] density [Jy] density [Jy]S-PASS Lobesnorth 1751 2610 ± 100 0.25 ±0.02 10440 ± 450south 2009 2780 ± 110 0.26±0.02 10690 ± 450total 3760 5390 ± 150 0.26±0.02 21130 ± 720Northern Ridge 51.8 2.75 174 ± 14 0.31 ± 0.06 560 ± 110GC Spur 50.5 1.9 236 ± 8 0.25 ± 0.03 960 ± 132Southern Ridge 117.6 3.0 373 ± 35 0.31 ± 0.04 1215 ± 130

Table S1: Observed quantities

2. We measure the integrated total and polarized intensity from both Lobes at all latitudes |b| > 15◦

where the free-free emission is marginal compared to the synchrotron at this frequency.

3. On the basis of these measurements we infer the intrinsic polarization fraction of the synchrotronemission from the Lobes. This is done on the area mentioned above (|b| > 15◦), that is about 63% ofthe whole solid angle covered by the Lobes.

4. Assuming the same intrinsic polarization fraction, we infer the integrated, total intensity flux densityfrom the remaining 37% of the Lobes that we cannot measure directly.

1.2 Minimum distance to Lobes from (de)polarization phenomenology

Neither the γ-ray nor the microwave data1 allow us to infer the distance to the Lobes, but the 2.3 GHzS-PASS data show depolarization at low latitudes around the Galactic Plane that is very informative in thiscontext. S-PASS linear polarized emission and Stokes Q and U images reveal two strong depolarizationareas on either side of the Galactic Centre encompassed by a border of small scale modulated signal thatextends up to some |b| ≃ 10◦. Both are generated by Faraday Rotation effects generating depolarization(the former) and polarization angle modulation that generates small scale mixing of Stokes Q and U withoutsignificant depolarization (the latter). By pinning down the objects responsible for this depolarization wecan infer a lower limit on the distance to the Lobes.

Figure S1 shows a comparison between S-PASS Stokes Q (left panel) and the H-α emission as seen inthe SHASSA28 map (right panel). The S-PASS image reveals a number of circular, arc, and bow features inthe depolarization regions that match the H-alpha emission regions in SHASSA maps well. We investigatedthese associations to identify the individual H-α regions and found that most belong to the Sagittarius arm.Moreover, some regions of the more distant Scutum-Centaurus arm are evident emerging behind the Sagit-tarius arm, like the Scutum super–shell, generating depolarization as well. A few examples of these regions

1Evidence that a distinct, extended, thermal X-ray source – containing close to 1056 erg thermal energy – seen in the direction

of the Galactic Centre is actually located in its physical vicinity has previously been claimed29.

2

SUPPLEMENTARY INFORMATION

2 | W W W. N A T U R E . C O M / N A T U R E

RESEARCH

structure solid angular 2.3 GHz pol. 2.3 GHzname angle width pol. flux frac. total flux












2

structure solid angular 2.3 GHz pol. 2.3 GHzname angle width pol. flux frac. total flux












2

are: G4.28+0.55, G6.09-1.29, G17.4–4.55, G355.05+0.04, G347.70+1.90, G345.00+1.70. Thus, the H-αregions responsible for the depolarization are not objects in the local arm but are located at least 1.5–2.5 kpcfrom us. Depolarization against the Scutum-Centaurus arm occurs at 3.0-4.0 kpc from us.

The Faraday modulated region surrounding these two areas of depolarization corresponds to weakerH-α enclosing the same group of H-α regions and must be associated with the same spiral arms.

The large scale emission must come from the background of the depolarizing objects. The Lobes’front sides, then, must sit at least at 2.5 kpc from us if we conservatively only account for depolarizationby the Sagittarius arm objects. The transverse dimension of the lobes is some 50◦, so that, assuming acylindrical geometry, its centre has to be at least at 4.0 kpc from us and its far side at 5.5 kpc. This is alreadyin the bulge region.

In summary, the S-PASS data implies that the lobes must be located at least as distant as the rim of theGalactic Bulge, in the direction of the Galactic Centre, and huge (extending at least 4.0-5.0 kpc both northand south of the Galactic Plane, implying a minimum total vertical extent of 8.0-10.0 kpc).

2 Equipartition magnetic field calculation

2.1 Equipartition magnetic field calculation for entire structures

We assume that the average path length through the Lobes is 5 kpc and examine two limiting cases: i)synchrotron emission occurs through the entire volume of the Lobes in which case their 21 kJy flux densityimplies a 6 µG equipartition11 field as reported; ii) the synchrotron emission occurs in an ensheathingregion of 300 pc thickness (the approximate width of the Ridges) with a highly but not perfectly regularfield structure. Such a layer would have an equipartition field of 12 µG. Note that, given the rather highpolarization fraction at 2.3 GHz, limiting case i) is likely to be somewhat in tension with the inference31, 32

that – given a lack of observed polarized microwave emission coincident with the WMAP ‘haze’33 (see §3)– the interior, volume-filling field structure is highly turbulent (though see ref. 34). Calculated equipartitionmagnetic field amplitudes (and the lower and upper magnetic field amplitude limits determined below) forthe Lobes and for the Ridges are reported systematically in Table S2 (and consequent magnetic energydensities, total energies, inferred magnetic energy injection rates are presented in Table S3); the Ridges areremarkably similar in their gross characteristics despite their different ages. For the magnetic field energyof the entire structures (in the volume filling field scenario) we assume a total volume of 2 × 1067 cm3; themodelled, 300 pc-thick sheath has volume 1 × 1066 cm3.

3


SUPPLEMENTARY INFORMATION RESEARCH

are: G4.28+0.55, G6.09-1.29, G17.4–4.55, G355.05+0.04, G347.70+1.90, G345.00+1.70. Thus, the H-αregions responsible for the depolarization are not objects in the local arm but are located at least 1.5–2.5 kpcfrom us. Depolarization against the Scutum-Centaurus arm occurs at 3.0-4.0 kpc from us.

The Faraday modulated region surrounding these two areas of depolarization corresponds to weakerH-α enclosing the same group of H-α regions and must be associated with the same spiral arms.

The large scale emission must come from the background of the depolarizing objects. The Lobes’front sides, then, must sit at least at 2.5 kpc from us if we conservatively only account for depolarizationby the Sagittarius arm objects. The transverse dimension of the lobes is some 50◦, so that, assuming acylindrical geometry, its centre has to be at least at 4.0 kpc from us and its far side at 5.5 kpc. This is alreadyin the bulge region.

In summary, the S-PASS data implies that the lobes must be located at least as distant as the rim of theGalactic Bulge, in the direction of the Galactic Centre, and huge (extending at least 4.0-5.0 kpc both northand south of the Galactic Plane, implying a minimum total vertical extent of 8.0-10.0 kpc).

2 Equipartition magnetic field calculation

2.1 Equipartition magnetic field calculation for entire structures

We assume that the average path length through the Lobes is 5 kpc and examine two limiting cases: i)synchrotron emission occurs through the entire volume of the Lobes in which case their 21 kJy flux densityimplies a 6 µG equipartition11 field as reported; ii) the synchrotron emission occurs in an ensheathingregion of 300 pc thickness (the approximate width of the Ridges) with a highly but not perfectly regularfield structure. Such a layer would have an equipartition field of 12 µG. Note that, given the rather highpolarization fraction at 2.3 GHz, limiting case i) is likely to be somewhat in tension with the inference31, 32

that – given a lack of observed polarized microwave emission coincident with the WMAP ‘haze’33 (see §3)– the interior, volume-filling field structure is highly turbulent (though see ref. 34). Calculated equipartitionmagnetic field amplitudes (and the lower and upper magnetic field amplitude limits determined below) forthe Lobes and for the Ridges are reported systematically in Table S2 (and consequent magnetic energydensities, total energies, inferred magnetic energy injection rates are presented in Table S3); the Ridges areremarkably similar in their gross characteristics despite their different ages. For the magnetic field energyof the entire structures (in the volume filling field scenario) we assume a total volume of 2 × 1067 cm3; themodelled, 300 pc-thick sheath has volume 1 × 1066 cm3.

3

structure assumed physical height vertical assumed l.o.s. Beq[⋆] BBBnd

[†] Bmax[‡]

name distance width top extent pathlength [µG] [µG] [µG][pc] [pc] [pc] [pc] [pc]

S-PASS Lobesvolume 5 000 6 >9 Bvol

|| <25-filling [α23

2.3 = 1.15] [α3323 = −0.7] Bvol

tot < 43[§]

shell only 300 12 >9[α23

2.3 = −1.15] [α3323 = −0.7]

Northern Ridge 6 000 290 7000 2 100 290 14 [α232.3 = −1.15]

GC Spur 8 000 210 4 000 4 000 270 15 [α232.3 = −1.01] 11 − 18 [α33

23 = −1.25]Southern Ridge 6 000 320 7000 4 900 320 13 [α23

2.3 = −1.05]

Table S2: Derived quantities I: ⋆Equipartition11 magnetic field. Relative statistical error is 1% from uncertainty in total, 2.3 GHz flux and 6% fromuncertainty in 2.3 to 23 GHz spectral index. †Broadband limits on the allowed magnetic field amplitude determined from the consideration that electronssynchrotron radiating at microwave frequencies (and therefore contributing to the WMAP haze emission) will also inverse-Compton radiate into ∼GeVγ-rays (and therefore contribute to the Fermi Bubbles’ intensity). Note that the WMAP haze is significantly less extensive in b than the Bubbles or Lobesand that the limit only applies to the solid angle over which it is observed, roughly b <∼30◦. ‡Derived from tentative detection of polarized, 2.3 GHz emissionfrom rear surface of the outflow and consequent requirement that the change in the polarization angle due to differential Faraday rotation satisfy ∆θ < πover the 184 MHz bandwidth of the Parkes 2.3 GHz observations. §Derived from Bvol

|| <25 assuming a turbulent magnetic field.4



RESEARCH

2.1.1 Statistical uncertainties

Formally, the sources of statistical uncertainty in the determination of the equipartition field in the Lobesoriginate in errors in the measurement of the total flux density from these structures and the 2.3 to 23 GHzspectral index (see Table S1). The 3% relative error on the total flux density from the Lobes implies a partialcontribution of 1% to the relative statistical error on Beq. The 9% relative error on the spectral index impliesa partial contribution of 6% to the relative statistical error on Beq.

2.1.2 Systematic uncertainties

The systematic error on Beq is dominated by the uncertainty in K0, the proton-to-electron number den-sity ratio. We choose the theoretically-motivated35 and observationally-suggested (from local cosmic raymeasurements36) value of 100 for this parameter. It is to be admitted, however, that, in the unusual environ-ment of the Lobes and Ridges, we cannot be certain this value holds. Still, the dependence of Beq on K0 israther weak so that even variation of this parameter by fully an order of magnitude leads to only a ∼ 60%change in Beq (e.g., we have, for the volume-filling field Beq = {4, 6, 10} for K0 = {10, 100, 1000}).Moreover, given the timescale requisite to transport the cosmic rays from the plane and the much longercooling times5 of cosmic ray ions than electrons in the environment of the Lobes, we expect that – if any-thing – K0 = 100 is likely to be underestimate, implying that, conservatively, the equipartition magneticfield we estimate is likely to be lower than the real field. Finally, on the question of whether the physicalcircumstances in the Lobes and Ridges are such that equipartition actually holds or is, at least, a reason-able approximation, we explain immediately below how an analysis of the broadband data covering theLobes and the GC Spur implies lower limits to the real magnetic fields in these structures approaching theequipartition magnetic field values we obtain.

3 Broadband phenomenology

At lower Galactic latitudes the Fermi Bubbles – and the Lobes – are coincident with a non-thermal mi-crowave ‘haze’ found in total intensity WMAP 20-60 GHz data34, 37 of luminosity (1 − 5) × 1036 erg/s(cf. the 1-100 GeV luminosity of the Bubbles of 2 × 1037 erg/s[3]) and their edges are coincident withan hourglass-shaped X-ray structure seen at lower Galactic latitudes in ROSAT data29 (and attributed to anoutflow driven by Galactic centre star formation12 and also clearly evident in the Stokes U parameter map at23 GHz (see Figure S3). There are intriguing similarities and differences between emission seen in differentwavebands.

We find regions of emission coincident with the 2.3 GHz map not only in the microwaves but also inX-rays (Figure S2) and in γ-rays (Figure 2, main text). In the south west, a spur of X-ray emission appears towrap around the edge of the southern Fermi Bubble, paralleling but not exactly coincident with the SouthernRidge; this indicates this feature is not simply a limb-brightening in the cone of outflowing plasma (a γ-ray

5



feature coincident with the Southern Ridge and appearing to ‘wrap’ in the same fashion is also evident:see Figure 2 of the main text). Coincident, non-thermal emission in 2.3 GHz and 23 GHz polarization and∼GeV γ-rays is evident in the GC Spur, the Southern Ridge, and, indeed, over almost the entire extent ofthe Bubbles. This indicates a non-thermal electron population covering at least the energy range ∼(1-100)GeV (Figure 2, main text) that is simultaneously synchrotron radiating into radio/microwave frequenciesand up-scattering ambient light into γ-rays via the inverse Compton process2.

We find, however, that the broadband data cannot be explained with a single power-law electronpopulation: the spectrum between 2.3 and 23 GHz is considerably steeper (α <∼ − 1.0 for Fν ∝ να) thanthe very hard spectrum (−0.4 > α > −0.7) found34 over 23 to 41 GHz for the haze3. Moreover, polarized2.3 GHz emission is observed considerably outside the γ-ray-defined edges of the Bubbles at high Galacticlatitudes (and towards Galactic west). These considerations indicate that a second, high-energy and veryhard electron population is either locally accelerated (perhaps powered by magnetic field reconnection) orinjected as secondaries (from collisions between cosmic ray protons and the Bubbles’ low-density thermalplasma5 in situ. This is consistent with the fact that the cooling time of the high-energy electrons required togenerate the γ-rays is too short for these particles to be transported from the plane out to the full extensionof the Bubbles/Lobes given the speed of the outflow (Figure 4, main text).

3.1 Spectral index between 2.3 and 23 GHz polarized emission

The spectral index between polarized emission at 2.3 GHz measured by S-PASS and at 23 GHz measuredby WMAP is shown in Figure S4. S-PASS and WMAP polarization maps have been binned to 2◦ × 2◦

pixels to improve the signal–to–noise ratio of the 23 GHz data. Noise debiasing has been applied beforemeasurement of the spectral index. As stated in the main text, there is a clear tendency for the spectrum ofsynchrotron radiation to steepen with distance from the plane (the very flat spectrum in the the plane itselfis a spurious result of Faraday depolarization at 2.3 GHz near the Galactic plane). This is a clear indicationfor the ageing of the synchrotron-emitting cosmic electrons and consistent with their being transported outfrom the plane.

3.2 Broandband limits on magnetic field

Broadband considerations allow also us to derive a rough lower bound on the magnetic field intensitythroughout the volume of the Lobes/Bubbles: the magnetic field must be strong enough that the in situ

2We note that the γ-ray spectrum for the ‘jet’ feature identified by Su and Finkbeiner3 and claimed by us to be more-or-lesscoincident with the GC Spur, is distinct from that of the general Fermi Bubble emission surrounding it. In particular, the jetspectrum is both harder and does not exhibit the same low energy cut-off seen in the general Bubble spectral energy distributionbelow ∼1 GeV. This phenomenology is consistent with the jet γ-rays being largely supplied by inverse Compton emission dueto primary electrons advected from the plane while, in contrast, the general Bubble γ-ray emission might largely be supplied byproton-proton collisions5.

3Though note that the spectrum that we determine, on the basis of the WMAP data, for the GC Spur between 23 and 33 GHz is,at α

33

23 ≃ −1.25, considerably steeper than that determined for the haze.

6



RESEARCH

electron population (required to generate, via synchrotron emission, the observed non-thermal microwaveintensity) must not be so numerous that the inverse Compton emission4 from the same electrons surpassesthe observed γ-ray intensity in the Fermi band (E2

γdNγ/dEγ ≃ 4 × 10−7 GeV/cm2/s/sr at a few GeV fol-lowing ref. [2]). Such reasoning implies B >∼ 9 µG field for a hard-spectrum electron population with a 1eV cm−3 photon background to up-scatter and γe = 2.4 (for dNe/dEe ∝ E−γe

e , the particle spectral indexcorresponding to the steepest allowed spectral index from analysis of the haze emission, αhaze = −0.7[31],which generates the most conservative lower limit on the field amplitude), and B >∼ 16 µG for γe = 2.0(corresponding to αhaze = −0.5, the central value of the haze spectral index31).

Given the evidence that the coincident γ-ray and microwave emission originates from the same cosmicray electron population (indeed, from electrons in the same energy range), we may apply similar reasoningto the above to determine rough but robust limits to both the lower and upper allowed field strength in theGC Spur. Adopting the intensity reported3 for the jet-like γ-ray feature recently claimed in the Fermi data(also E2

γdNγ/dEγ ≃ 4×10−7 GeV/cm2/s/sr at a few GeV) which is coincident with the GC Spur (identifiedat radio continuum and microwave frequencies) at b ∼ 15 − 25◦ and using the spectral index measured byus between the polarized emission at 23 and 33 GHz, α33

23 ≃ 1.25, the polarized surface brightness at 23GHz (1520 Jy/sr) and assuming the polarization fraction of 0.25 measured at 2.3 GHz also applies at 23GHz, we derive a Stokes I surface brightness of 6100 Jy/sr and determine a lower limit to the total magneticfield amplitude in the GC Spur of 11 µG. We derive a conservative upper limit on the magnetic field fromdemanding that the cosmic ray electron population that supplies the inverse Compton γ-ray flux from theGC Spur saturate, via synchrotron emission, the whole total intensity at 23 GHz detected over the GC Spursolid angle, 16100 Jy/sr. This saturation point is attained for an ∼18 µG field, implying a rough upper limitto the field at this amplitude. In the case of the GC Spur, these lower and upper limits to the field imply that– if equipartition holds – the proton to electron number ratio, K0 is in the range 30 - 200.

4 Visibility of emission from rear windings at various wavelengths

Our explanation of the geometry of the Ridges – in particular their curvature to Galactic west – requiresthat, while they wrap around the entirety of the cones defined by the global outflow, rear-side emission fromthe Ridges is attenuated with respect to the front side emission (the emission from the putative rear part ofeach Ridge would curve to Galactic east contrary to observations). This relative attenuation must functionat 2.3, 23, and 33 GHz; we find that it cannot, then, be a result of simple Faraday depolarization which, forreasonable parameters of magnetic field intensity, plasma density, and path length through the volume of theLobes, could not appreciably Faraday rotate the polarization angle at microwave frequencies.

In fact, the relative attenuation of the rear, polarized emission can naturally be explained as a conse-quence of three simple effects which work equally well at radio continuum and microwave frequencies:

4In order to calculate the spectrum and luminosity of inverse Compton radiation we employ the RadiationField class39 fromthe GalProp code available at http://galprop.stanford.edu/code.php. RMC thanks Troy Porter for assistance with using GalProp’sGalactic interstellar radiation field data.

7



1. The synchrotron intensity scales approximately as B2perp, where Bperp is the component of the mag-

netic field perpendicular to the line-of-sight. From simple geometrical considerations (see caption tofigure S5 ) the vertical component of Bperp is appreciably less in the rear part of a conical outflowthan in the front part. The magnetic field direction, moreover, is largely vertical closer to the plane(figure 3 of the main text).

2. Furthermore, along any particular sightline the rear surface is intersected at a greater physical heightfrom the plane than the front surface. This has the consequence that the electron population on therear is ‘older’ (more cooled) than the front population.

3. A further (likely) consequence of the greater physical height of the intersection of any given line ofsight with the rear surface (relative to the front surface) is that the local magnetic field amplitudeat the rear on this line of sight is relatively attenuated (given that the magnetic field is also injectedas ‘frozen-in’ field lines in the plasma outflowing from the plane and will have had more time toreconnect/relax while ascending to a greater height above the plane).

Figure S5 shows the approximate ratio of front-side to rear-side synchrotron intensity taking these effectsinto account. It is also important to note that horizontal component of Bperp completely disappears at thetangent points of the projected outflow edges.

4.1 Upper limit on volume magnetic field from tentative detection of 2.3 GHz polarizedemission from rear surface

Note, however, that a blanket statement that polarized emission from the Ridges on the rear surface of theoutflow is invisible from our vantage point does not seem to be correct, though such emission is certainlyobscured as discussed in the previous section. A careful examination of the Southern lobe in the 2.3 GHzpolarization map (Figure S6 and Figure 1 in the main text) reveals features curving in the opposite senseto the Ridges. A clear ridge–like structure – with possible counterparts at other wavelengths – is a lineardepolarization feature running from (l, b) ∼ (350◦,−17◦) to (9,−32) (see Figure S6). A likely explanationof this feature is that it runs almost perpendicular to the Southern Ridge so that the polarization angles of thetwo structures are perpendicular (the magnetic angle is aligned with the Ridges). In turn, both Stokes Q andU have opposite signs for the two ridges and tend to cancel. An important implication of this explanation forthe phenomenology is that – even at the comparatively low frequency of 2.3 GHz – intrinsically polarizedemission from the rear surface is not Faraday depolarized by its passage through the magnetised plasmainhabiting the volume of the Southern lobe (rather the three geometrical factors outlined in the previoussection are responsible for the attenuation of the rear-side synchrotron emission relative to the front-side).We can use this inference to then place an upper limit on the magnetic field in this region of ∼ 43 µG: seeTable S2.

8



RESEARCH

5 More robust estimate on plasma outflow speed in ridges

The Galactic Centre Ridge diverges from the projected co-rotating point in a fashion that is consistent withangular momentum being conserved in the outflow and the Galactic Centre Spur being ‘wound-up’ in theoutflow21. This same analysis points to an initial ratio between the circulation and vertical velocities of theoutflow of vcirc/vvert ∼ 0.07. This generates a more accurate estimation of its ascension speed of ∼ 1100km/s vcirc/(80 km/s) where we normalise to an 80 km/s circulation speed in the inner ∼100 pc as suggestedby recent Herschel observations13 .

6 Global analysis

6.1 Considerations around the far infrared-radio continuum correlation

The 60 micron (infrared) flux density19 of the inner ∼200 pc × 80 pc region around the Galactic centre –essentially the Nuclear Bulge – is 2 MJy [20]. On the basis of the far infrared-radio continuum correlation19 ,this level of far infrared emission should be accompanied by a radio continuum flux density of 20.2 kJyat 1.4 GHz. In dramatic contrast, the detected radio continuum flux from the same region is ∼ 1.6 kJy at1.4 GHz, less than 10% of expectation or around 4σ shy of the correlation40 . Even integrating the radiocontinuum flux density out to scales of 800 pc in diameter (thereby encompassing the distinct ‘diffuse non-thermal source’ identified by LaRosa et al.41 surrounding the Galactic centre), the detected radio continuumflux reaches only 25% of expectation. As has been argued at length elsewhere40, the explanation for thisphenomenology is that the vast bulk of cosmic ray electrons – accelerated in concert with star formation(and consequent supernova activity) in the Galactic centre region – is advected out of the region before theelectrons can lose their energy, radiatively, in situ.

Similarly, the γ-ray luminosity of this same inner region is in significant deficit with respect to theexpectation42 were the hadronic cosmic rays accelerated in the region to lose their energy in situ (via col-lisions on ambient gas); i.e., the system is very far from a ‘calorimeter’. Again, the inference that can bemade is that the vast bulk of the hadronic cosmic rays also escape the region on an outflow40.

Where does this power represented by the escaping cosmic ray ions and electrons go? The γ-rayluminosity of the Fermi Bubbles matches the expectation if supplied by hadronic collisions of the cosmicray protons and ions leaving the Galactic centre5,6. Equally, the S-PASS data allow us to determine that thetotal radio continuum flux density from the Lobes is 21 kJy at 2.3 GHz or νFν = 4.9×10−10 erg/cm2/s; the20.2 kJy at 1.4 GHz predicted by the correlation corresponds42 to νFν = 2.3× 10−10 erg/cm2/s at 2.3 GHzor, the expected 2.3 GHz flux density is 11.2 kJy assuming a spectral index of -1.2 between 1.4 and 2.3 GHz.The observed and predicted total flux densities at 2.3 GHz are, therefore, within a factor 1.9 of each other,corresponding to a quite acceptable difference of ∼1.1 σ (adopting the 0.26 dex scatter in the empiricalcorrelation from ref. [19]). Given a number of uncertainties – particularly the effect that the spreading of theoutflow cones will mean that the r2-biased emission from the front of the outflow takes place significantly

9



closer than the ∼ 8 kpc distance to the GC (at which distance the Galactic centre star-formation-related farinfrared light is emitted) – we view this level of agreement, as stated in the main text, as a strong argumentthat the Lobes’ non-thermal radio emission is supplied by synchrotron emission from cosmic ray electronsaccelerated by star-formation activity in the Galactic Centre.

6.2 Gross Energetics

The current star-formation rate in the inner ∼ 100 pc (in radius) region around the GC is slightly below0.1 M⊙/year (see ref. [6] and references therein). Given the ∼ 109 M⊙ mass of the stellar populationinhabiting this region (the Nuclear Bulge20,43), the current value is close to the time-averaged value over thelast ∼ 10 Gyr. Using, conservatively, standard assumptions (i.e., the initial mass function, IMF, for the zero-age, main sequence stellar masses is given by a Kroupa44 IMF with a minimum stellar mass of 0.07 M⊙,the total mechanical energy release per core-collapse supernova is 1051 erg irrespective of the progenitor’szero-age, main sequence mass) this star formation rate translates to a mechanical power injection rate fromthe region’s core-collapse supernovae of ∼ 1040 erg/s[6,15]. By way of comparison, the power requisite toinflate the magnetic fields of the expanding Ridges and supply their cosmic ray content is ∼ 2 × 1039 erg/s(see Table S3) assuming that the equipartition approximation holds. These energetics can be satisfied by themechanical power available from the region’s supernovae under the most conservative assumptions.

structure volume uBeqUBeq

[†] age[⋆] texp[⋆] Emag

[‡]

name [cm3] [eV cm−3] [erg] [Myr] [Myr] [erg/s]S-PASS Lobesvol.-filling 2.0 × 1067 cm3 0.8 3 × 1055 300§ 300§shell only 1.2 × 1066 cm3 3 8 × 1054 90§ 90§northern ridge 5.0 × 1063 cm3 5 4 × 1052 4.7 1.4 8 × 1038

GC Spur 5.2 × 1063 cm3 4 7 × 1052 2.8 2.8 8 × 1038

southern ridge 1.4 × 1064 cm3 4 9 × 1052 4.7 3.4 9 × 1038

Table S3: Derived quantities II: ⋆Assumes expansion velocity vexp ≡ 1400 km/s. In principle, both thequoted ages and expansion times are lower limits because each Ridge structure disappears around the edgeof the general outflow. †Total magnetic energy assuming equipartition. ‡Emag ≡ UBeq

/texp. §Assumesmagnetic power injected at a rate 3 × 1039 erg/s.

7 Discussion of thermal X-ray fluxes

From ROSAT X-ray data covering the Lobes29,38 we find a background-subtracted count rate of 300×10−6

cnt/s/arcmin2 over the R6 band (0.91 - 1.31 keV) for the bright X-ray counterpart to the southern Ridge(see Figure S2). This corresponds to an intensity of ∼ 8 × 10−8 erg/cm2/s/sr. Obviously, this region isatypically bright in comparison to the whole solid angle of the Lobes but we use this intensity in the contextof generating various limits. (Also note, given the scale of other uncertainties, we are not correcting for

10


1 0 | W W W. N A T U R E . C O M / N A T U R E

RESEARCH

photoelectric absorption. This we expect to be a reasonably small correction given that we infer from thedata presented in ref. 38 an optical depth at ∼ keV of ∼ 0.4)

We first use this intensity to derive a (somewhat temperature-dependent) upper limit on the plasmadensity. Assuming a volume-filling plasma of ∼ 3 × 107 K, we derive from ref.45, an upper limit onthe density of this plasma of ∼ 3 × 10−3 cm−3. Assuming, more naturally, that the observed X-rays aregenerated by plasma entrained in the Southern Ridge outflow, assuming a 300 pc pathlength through thisstructure, we obtain a plasma density of ∼ 2 × 10−2 cm−3 for the same assumed temperature (we obtain∼ 9 × 10−3 cm−3 for T = 107 K) which defines an upper limit on the plasma density for the region. Fora 1100 km/s outflow, this density then implies a ∼ 2 − 4 × 10−2 M⊙/year mass flux along the SouthernRidge. Very roughly, this suggests that the mass flux along all of the ridges is ∼ 0.1M⊙/year, a comfortablefraction of the mass accretion rate on to the GC of ∼ 0.3M⊙/year (see ref. [6] and references therein).

7.1 Inferred Alfven velocities

Given the upper limit on the plasma density and assuming the equipartition magnetic field amplitude, wemay obtain the Alfven velocity, v2

A ≡ B2/(4πmpnp). The component of this resolved into the verticaldirection is vvert

A ≃ 300 km/s (for T ∼ 107 K and B = 15 µG).

8 Relating Ridges to GC super-stellar clusters

Assuming a reasonable fraction (∼20%) of the typical mechanical energy of a supernova (1051 erg) ends upin cosmic ray and magnetic energy, the Ridges each require 400 − 1000 supernovae or the formation of atotal stellar mass of (3 − 9) × 104 M⊙. This requires the accumulation of > (0.4 − 1) × 106 years’ starformation given the star-formation rate in the region. Such a timescale and the total stellar mass quoted arecomparable to those associated with the formation of the observed massive stellar clusters in the GalacticCentre (e.g., ref. 46). As we have already noted, the gross energetics of the Ridges can be supplied by core-collapse supernovae occurring with the frequency implied by the Galactic Centre’s current star-formationrate.

Note, however, a complicating factor: for any discrete star-formation event there is a delay of ∼ 3 Myr(e.g., ref. 47) between the onset of star-formation and the first core-collapse supernovae (originating in themost massive stars). In the strong tidal fields of the Galactic centre, moreover, stellar clusters are completelydisrupted over a timescale ∼ 10 Myr[48] or, at least, suffer sufficient dissolution that they become invisibleagainst the high stellar density background within a similar timeframe49. In general terms, this means that,whereas the most massive stars of the super-stellar clusters contribute to the outflows forming the Ridges(both in terms of their winds and their supernovae which occur soon enough after cluster formation that thecluster is still coherent), core-collapse supernovae arising in less massive stars are more broadly distributedthrough the region and would seem to be prime candidates for energising the general bi-conical outflow

11

W W W. N A T U R E . C O M / N A T U R E | 1 1


feeding into the extended Lobes/Bubbles. Consistent with this picture, the mechanical power injected by thecombined stellar winds of a ∼ 5 × 104 M⊙ super-stellar cluster (a reasonable estimate for, e.g. the initialmass of either the Arches or Quintuplet clusters) is at least a few ×1039 erg/s (see, e.g., Figure 1 of ref.50),enough to initially supply the Ridges’ magnetic fields and cosmic ray content. Likewise, the mass flux alongthe Southern Ridge represented by the outflowing plasma, ∼ 2−4×10−2 M⊙/year as derived above, is wellaccounted-for by mass loss due to massive stellar winds from a similarly-sized cluster (see, e.g., Figure 11of ref.50), allowing for an expected mass-loading of 3-10[51] and further mass injection by supernovae. Thismass flux also represents a comfortable fraction (∼ 10%) of the model-derived6 total plasma mass flux intothe entire outflow.

9 Collimation of the Ridges

One significant aspect of the winding’s phenomenology is that they remain coherent over many kpc withrather constant widths. Aside from the implication that the windings present a channel to deliver cosmic raysfrom the Galactic nucleus out into the halo with little adiabatic loss, their collimation likely implies a par-ticular magnetic field topology: a ‘force-free’ configuration where the toroidal and longitudinal componentsof the field satisfy Bφ >∼B|| and the magnetic structure is self-confined52 . Confirmation of this speculationand exactly how such a field configuration is produced is a subject of ongoing investigation.

References

[30] Kerr, F. J., & Sinclair, M. W. A Highly Symmetrical Pattern in the Continuum Emission from theGalactic Centre Region, Nature, 212, 166 (1966)

[31] Dobler, G., & Finkbeiner, D. P., Extended Anomalous Foreground Emission in the WMAP Three-YearData Astrophys. J., 680, 1222 (2008)

[32] McQuinn, M., & Zaldarriaga, M. Testing the Dark Matter Annihilation Model for the WMAP Haze,Mon. Not. Roy. Astron. Soc., 414, 3577 (2011)

[33] Gold, B., Odegard, N., Weiland, J. L., et al. Seven-year Wilkinson Microwave Anisotropy Probe(WMAP) Observations: Galactic Foreground Emission, Astrophys. J. Supp., 192, 15 (2011)

[34] Dobler, G. A Last Look at the Microwave Haze/Bubbles with WMAP, Astrophys. J., 750, 17 (2012)

[35] Bell, A. R. The acceleration of cosmic ray shock fronts – 1, Mon. Not. Roy. Astron. Soc., 182, 147(1978)

[36] Ginzburg, V. L., & Ptuskin, V. S. On the origin of cosmic rays: Some problems in high-energy astro-physics, Reviews of Modern Physics, 48, 161 (1976)

[37] Finkbeiner, D. P. Microwave Interstellar Medium Emission Observed by the Wilkinson MicrowaveAnisotropy Probe, Astrophys. J., 614, 186-193 (2004)

12



RESEARCH

[38] Snowden, S. L., et al. ROSAT Survey Diffuse X-Ray Background Maps. II., Astrophys. J., 485, 125(1997)

[39] Porter, T. A., Moskalenko, I. V., Strong, A. W., Orlando, E., & Bouchet, L. Inverse Compton Originof the Hard X-Ray and Soft Gamma-Ray Emission from the Galactic Ridge, Astrophys. J., 682, 400(2008)

[40] Crocker, R. M., Jones, D. I., Aharonian, F., et al. γ-rays and the far-infrared-radio continuum correla-tion reveal a powerful Galactic Centre wind, Mon. Not. Roy. Astron. Soc., 411, L11-L15 (2011a)

[41] LaRosa, T. N., Brogan, C. L., Shore, S. N., Lazio, T. J., Kassim, N. E., & Nord, M. E. Evidence ofa Weak Galactic Center Magnetic Field from Diffuse Low-Frequency Nonthermal Radio Emission,Astrophys. J., 626, L23 (2005)

[42] Thompson, T. A., Quataert, E., & Waxman, E. Starbursts and Extragalactic γ-ray Background, Astro-phys. J., 654, 219 (2007)

[43] Serabyn, E., & Morris, M., Nature, 382, 602 (1996)

[44] Kroupa, P. On the variation of the initial mass function, Mon. Not. Roy. Astron. Soc., 322, 231 (2001)

[45] Raymond, J. C., Cox, D. P., & Smith, B. W., Radiative cooling of a low-density plasma, Astrophys. J.,204, 290 (1976)

[46] Harfst, S., Portegies Zwart, S., & Stolte, A. Reconstructing the Arches cluster - I. Constraining theinitial conditions, Mon. Not. Roy. Astron. Soc., 409, 628-638 (2010)

[47] Mo, H., van den Bosch, F. C., & White, S. , Galaxy Formation and Evolution. Cambridge UniversityPress, 2010. ISBN: 9780521857932 (2010)

[48] Kim, S. S., Figer, D. F., Lee, H. M., & Morris, M. N-Body Simulations of Compact Young Clustersnear the Galactic Center, Astrophys. J., 545, 301 (2000)

[49] Portegies Zwart, S. F., Makino, J., McMillan, S. L. W., & Hut, P. The Lives and Deaths of Star Clustersnear the Galactic Center, Astrophys. J., 565, 265 (2002)

[50] Cote, B., Martel, H., Drissen, L., & Robert, C. Galactic outflows and evolution of the interstellarmedium, Mon. Not. Roy. Astron. Soc., 421, 847 (2012)

[51] Strickland, D. K., & Heckman, T. M. Supernova Feedback Efficiency and Mass Loading in the Star-burst and Galactic Superwind Exemplar M82, Astrophys. J., 697, 2030 (2009)

[52] Bicknell, G. V., & Li, J. The Snake: A Reconnecting Coil in a Twisted Magnetic Flux Tube, Astro-phys. J., 548, L69-L72 (2001)

13



-0.16 -0.13 -0.096 -0.064 -0.032 0.00016 0.032 0.064 0.096 0.13 0.16

-10 21 52 83 114 145 176 207 238 269 300

Figure S1: Top: Stokes Q image of the area around the Galactic Centre. The Galactic plane is horizontalacross the picture and the emission unit is Jy/beam with a beam of FWHM=10.75’. The green dashed lineindicates the two areas of depolarization on either side of the Galactic Centre and the belt encompassingthem of emission modulated to small angular scales by Faraday Rotation effects. Bottom: H-α emissionimage of the same area from the SHASSA survey. The emission unit is decirayleighs (dR); The resolutionis FHWM=6’. The area affected by Faraday Rotation effects is reported as well and corresponds to H-αemission regions from the Sagittarius and Scutum-Centaurus arms – see text.

14



RESEARCH

0 0.017 0.034 0.051 0.068 0.085 0.1 0.12 0.14 0.15 0.17

Figure S2: S-PASS and X-ray emission. Data are: i) polarized emission from S-PASS shown in colour– unit is Jy/beam); ii) X-ray emission as detected by ROSAT (white contour levels, ranging from 250 to550× 10−6cts/s/arcmin2 with steps of 75). ROSAT data are the average of the bands 5 and 6 and the band 7subtracted to remove the large scale emission and emphasise substructures. The thick dashed lines show theedges of the S-PASS Lobes and the thinner dashed lines the edges of the γ–ray Fermi Bubbles as defined bySu et al.2.

15



Figure S3: Linear polarization emission component Stokes U at 23 GHz from WMAP8. An X-shapestructure centred at the Galactic Centre matches the biconical Lobe base as traced by X-ray emission (cf.Figure 6c of ref. [12] and ref. [38]) and could be limb brightening of the Lobes (the 2.3 GHz Lobe edgesshown by the black solid line). Stokes U is less contaminated by spiral arm emission contamination thanStokes Q because the magnetic angle of the arm emission is largely parallel to the Galactic plane8. Themap is in Galactic coordinates, centred at the Galactic Centre. Grid lines are spaced by 15◦. The emissionintensity is in Brightness Temperature, the unit is K. Data have been binned in 1◦ × 1◦ pixels.

16



RESEARCH

Figure S4: Spectral index α between the 2.3 and 23 GHz polarized emission. The flux density S ismodelled as a power law of the frequency S ∝ να. The map is in Galactic coordinates, centred at theGalactic Centre. Grid lines are spaced by 15◦. S-PASS and WMAP linear polarized emission maps havebeen binned in 2◦ × 2◦ pixels to improve the Signal-to-Noise ratio of the latter.

17



Figure S5: The approximate ratio of front-side to rear-side synchrotron intensity as a function ofGalactic latitude. The relative attenuation of the rear side synchrotron emission with respect to the front sideis a consequence of three geometrical factors: i) (the vertical component of) the magnetic field perpendicularto the line of sight, Bperp, is relatively reduced on the rear surface by a factor ∼ cos(|b| − α)/ cos(|b| + α)where b is the latitude and α is the half opening angle of the outflow (the synchrotron intensity scales approx-imately as B2

perp); the front surface of the outflow intersects a given line of sight (los) at a smaller physicaldistance from the plane than the rear surface which has the dual effects that ii) the rear surface electrons arerelatively older (and therefore more cooled) along any given los and iii) the rear surface magnetic field alongthe given los might be expected to be relatively attenuated. We can approximately calculate the effects of ii)and iii) together by calculating the ratio of the total intensity along a given los (with given Galactic latitudeb) to the intensity along a line of sight at the (higher) latitude b′ which corresponds to the angle required suchthat this new los intersects the front surface at the same physical height above the plane that the originallos intersects the rear surface. Note that – consistent with the X-ray observations tracing the edge of theLobes/Bubbles relatively close to the plane12 – we set the outflow opening angle to be α = 60◦ in this plot.The plot also implicitly assumes that the magnetic field orientation is vertical; this is a good approximationover the latitude range of the plot.

18



RESEARCH

Figure S6: Black and white image of the polarised intensity at 2.3 GHz. A suspected rear windingrunning from (l, b) ∼ (350◦,−17◦) to (9,−32) is visible as a region of relatively low intensity (i.e., darker)within the box (white solid line) in the southern S-PASS Lobe. This region is likely dark because themagnetic field direction in a rear winding will be roughly perpendicular to the field in the front surfaceleading to cancellation of both Stokes U and Stokes Q parameters, hence cancellation of the total polarisedintensity along this line of sight.

19

giant magnetized outflows from the centre of the milky way

Education

galactic centre supplementary

galactic centre spur10

galactic centre relative

galactic centre area

galactic electrons

galactic bulge

galactic centremasses

galactic centre spur