99 Herculis: Host to a Circumbinary Polar-ring Debris Disk

Abstract

We present resolved Herschel images of a circumbinary debris disk in the 99
Herculis system. The primary is a late F-type star. The binary orbit is well
characterised and we conclude that the disk is misaligned with the binary
plane. Two different models can explain the observed structure. The first model
is a ring of polar orbits that move in a plane perpendicular to the binary
pericenter direction. We favour this interpretation because it includes the
effect of secular perturbations and the disk can survive for Gyr timescales.
The second model is a misaligned ring. Because there is an ambiguity in the
orientation of the ring, which could be reflected in the sky plane, this ring
either has near-polar orbits similar to the first model, or has a 30 degree
misalignment. The misaligned ring, interpreted as the result of a recent
collision, is shown to be implausible from constraints on the collisional and
dynamical evolution. Because disk+star systems with separations similar to 99
Herculis should form coplanar, possible formation scenarios involve either a
close stellar encounter or binary exchange in the presence of circumstellar
and/or circumbinary disks. Discovery and characterisation of systems like 99
Herculis will help understand processes that result in planetary system
misalignment around both single and multiple stars.

Full text

Page 1

arXiv:1201.1911v1 [astro-ph.EP] 9 Jan 2012
Mon. Not. R. Astron. Soc. 000, 000–000 (0000) Printed 11 January 2012(MN LATEX style file v2.2)
99 Herculis: Host to a Circumbinary Polar-ring Debris Disk
G. M. Kennedy⋆1, M. C. Wyatt1, B. Sibthorpe2, G. Duchˆ ene3,4, P. Kalas4,
B. C. Matthews5,6, J. S. Greaves7, K. Y. L. Su8, M. P. Fitzgerald9,10
1Institute of Astronomy, University of Cambridge, Madingley Road, Cambridge CB3 0HA, UK
2UK Astronomy Technology Center, Royal Observatory, Blackford Hill, Edinburgh EH9 3HJ, UK
3Department of Astronomy, University of California, B-20 Hearst Field Annex, Berkeley, CA 94720-3411, USA
4Laboratoire d’Astrophysique, Observatoire de Grenoble, Universit´ e J. Fourier, CNRS, France
5Herzberg Institute of Astrophysics, National Research Council Canada, 5071 West Saanich Road., Victoria, BC, Canada, V9E 2E7, Canada
6University of Victoria, Finnerty Road, Victoria, BC, V8W 3P6, Canada
7School of Physics and Astronomy, University of St Andrews, North Haugh, St Andrews, Fife KY16 9SS, UK
8Steward Observatory, University of Arizona, 933 North Cherry Avenue, Tucson, AZ 85721, USA
9Institute of Geophysics and Planetary Physics, Lawrence Livermore National Laboratory, L-413, 7000 East Avenue, Livermore, CA 94550, USA
10Department of Physics and Astronomy, UCLA, Los Angeles, CA 90095-1547, USA
11 January 2012
ABSTRACT
We present resolved Herschel images of a circumbinary debris disk in the 99 Herculis
system. The primary is a late F-type star. The binary orbit is well characterised and we
concludethat the disk is misaligned with the binaryplane. Two differentmodels can ex-
plain the observedstructure.The first model is a ringof polar orbits that movein a plane
perpendicularto the binary pericenter direction. We favour this interpretationbecause it
includes the effect of secular perturbations and the disk can survive for Gyr timescales.
The second model is a misaligned ring. Because there is an ambiguity in the orientation
of the ring, which could be reflected in the sky plane, this ring either has near-polar
orbits similar to the first model, or has a 30 degree misalignment. The misaligned ring,
interpretedas theresultofa recentcollision,is showntobe implausiblefromconstraints
on the collisional and dynamical evolution. Because disk+star systems with separations
similar to 99 Herculis should form coplanar,possible formationscenarios involveeither
a close stellar encounter or binary exchange in the presence of circumstellar and/or cir-
cumbinary disks. Discovery and characterisation of systems like 99 Herculis will help
understand processes that result in planetary system misalignment around both single
and multiple stars.
Key words: circumstellar matter — stars: individual: 99 Herculis, HD 165908, HIP
88745, GJ704AB
1 INTRODUCTION
The Herschel Key Program DEBRIS (Dust Emission via a
BiasfreeReconnaissance intheInfrared/Submillimeter) hasob-
served a large sample of nearby stars to discover and charac-
terise extrasolar analogues to the Solar System’s asteroid and
Kuiper belts, collectively known as “debris disks.” The 3.5m
Herschel mirror diameter provides 6-7” resolution at 70-100µm
(Pilbratt et al. 2010), and as a consequence our survey has re-
solved many disks around stars in the Solar neighbourhood for
the first time (Matthews et al. 2010; Churcher et al. 2011).1
⋆Email: gkennedy@ast.cam.ac.uk
1Herschel is an ESA space observatory with science instruments pro-
vided by European-led Principal Investigator consortia and with impor-
tant participation from NASA.
Here we present resolved images of the 99 Herculis cir-
cumbinary disk. This system is particularly interesting because
unlike most debris disk+binary systems, the binary orbit is well
characterised. The combination of a known orbit and resolved
disk means we can compare their (different) inclinations and
consider circumbinary particle dynamics and formation scenar-
ios.
This system is a first step toward building on the binary
debris disk study of Trilling et al. (2007). Their Spitzer study
found that debris disks are as common in binary systems as
in single systems, but tend not to have separations in the 3-
30AU range. However, only some of their systems had detec-
tions at multiple wavelengths to constrain the disk location and
none were resolved, making the true dust location uncertain.
Resolved systems such as 99 Her remove this ambiguity, and
c ? 0000 RAS

Page 2

2
G. M. Kennedy et. al.
Table 1. 99Her orbital elements, system massand 1σ uncertainties. The
ascending node Ω is measured anti-clockwise from North. The longi-
tude of pericenter is measured anti-clockwise from the ascending node,
and projected onto the sky plane has a position angle of 163◦(i.e. is
slightly different to 41+116 because the orbit is inclined).
ParameterSymbol (unit) ValueUncertainty
Semi-major axis
Eccentricity
Inclination
Ascending node
Longitude of pericenter
Date of pericenter passage
Period
Total mass
a (”)
e
i (◦)
Ω (◦)
ω (◦)
T (yr)
P (yr)
Mtot(M⊙)
1.06
0.766
39
41
116
1997.62
56.3
1.4
0.02
0.004
2
2
2
0.05
0.1
0.1
provide crucial information on the disk location, stability and
dynamics.
This paper is laid out as follows. We first consider the stel-
lar and orbital properties of the 99 Her system, including the
possibility of a third component. Then we consider the Her-
schel image data and several different models that can explain
it. Finally, we discuss the implications of these models for the
formation of the system.
2 99 HERCULIS
The binary 99 Herculis (HD 165908, HIP 88745, GJ 704AB,
ADS 11077) contains the 37thclosest F star primary within the
volume limited Unbiased Nearby Stars sample (Phillips et al.
2010). The Catalogue of Components of Doubles and Multi-
ple systems (CCDM J18071+3034, Dommanget & Nys 2002)
lists three components, but using Hipparcos proper motions
Phillips et al. (2010) find that the 93” distant C component is
not bound to the system. The binary pair has been known since
1859, and consists of an F7V primary orbited by a K4V sec-
ondary. The primary is known to be metal poor with [Fe/H]
≈ −0.4 (e.g. Gratton et al. 1996; Adelman et al. 2000; Takeda
2007) and has an age consistent with the main-sequence (6 −
10Gyr from isochrone fitting, Nordstr¨ om et al. 2004; Takeda
2007).
2.1 Binary Configuration
To interpret the Herschel observations requires an understand-
ing of the binary configuration, which we address first. Typi-
cally, the important orbital elements in fitting binary orbits are
the semi-major axis, eccentricity, and period, which yieldphysi-
cal characteristics of the system (if the distance is known). Reg-
ular observations of 99 Her date back to 1859 and the binary
has completed nearly three revolutions since being discovered.
Additional observations since the previous orbital derivation
(S¨ oderhjelm 1999), have allowed us to derive an updated orbit.
Aside from a change of 180◦in the ascending node (based on
spectroscopic data Abt & Willmarth 2006), the orbital param-
eters have changed little; the main purpose of re-deriving the
orbit is to quantify the uncertainties.
The orbit was derived by fitting position angles (PAs) and
separations (ρ) from the Washington Double Star catalogue
Figure 1. 99 Her binary orbit as seen on the sky, with the line of nodes
and pericenter indicated. North is up and East is left. The stellar orbits
are shown over one (anti-clockwise) orbital period with black dots. Grey
dots (primary is the larger grey dot) show the positions at the PACS
observation epoch. Black dot sizes are scaled in an arbitrary way such
that larger dots are closer to Earth. The arrows indicate the direction of
motion and the scale bar indicates the binary semi-major axis of 1.06”
(16.5AU).
(Mason et al. 2011).2We included three additional observa-
tions; a Hubble Space Telescope (HST) Imaging Spectrograph
(STIS) acquisition image (epoch 2000.84, PA = 264 ± 2◦,
ρ = 0.54 ± 0.02”, Boesgaard et al. 2004), an adaptive op-
tics image taken using the Lick Observatory Shane 3m tele-
scope with the IRCAL near-IR camera as an ongoing search for
faint companions of stars in the UNS sample (epoch 2009.41,
PA = 309 ± 2◦, ρ = 1.12 ± 0.02”), and a Keck II NIRC2
L’ speckle image taken to look for a third companion (see §2.2,
epoch 2011.57, PA = 317±1◦, ρ = 1.20±0.014”). For visual
data we set uncertainties of 7◦to PAs and 0.5” to separations,
for Hipparcos data we used 5◦and 0.1”, and for speckle obser-
vations without quoted uncertainties we used 2◦and 0.04”. The
resulting orbital elements, shown in Table 1, vary only slightly
from those derived by S¨ oderhjelm (1999).3The best fit yields
χ2= 190 with 399 degrees of freedom. The fit is therefore rea-
sonable, and most likely the χ2per degrees of freedom is less
than unity because the uncertainties assigned to the visual data
are too conservative. If anything, the uncertainties quoted in Ta-
ble 1are therefore overestimated. However, visual data can have
unknown systematic uncertainties due to individual observers
and their equipment so we do not modify them (Hartkopf et al.
2001).
These data also allow us to derive a total system mass
of 1.4M⊙, where we have used a distance of 15.64pc
2http://ad.usno.navy.mil/wds/
3Afigure showing the S¨ oderhjelm (1999)orbit is available in the WDS
catalogue.
c ? 0000 RAS, MNRAS 000, 000–000

Page 3

A Circumbinary Polar Ring Debris Disk
3
Figure 2. Motion of the 99 Her system (filled dots) in sky coordinates
at three epochs. The epochs including the putative third component are
enclosed in boxes. The arrow shows the direction of the system center of
mass movement and the distance travelled in 5 years, and the grey lines
show the path traced out by each star. Star symbols show the position of
the third object observed in the STIS data in 2000 (dashed box) and by
Scardia et al. (2008) in 2005 (dotted box).
(van Leeuwen 2008). While the total mass is well constrained
by visual observations, the mass ratio must be derived from
either the differential luminosity of each star or radial ve-
locities. We use the spectroscopic mass function derived by
Abt & Willmarth (2006), yielding a mass ratio of 0.49, which
has an uncertainty of about 10%. The mass ratio from the dif-
ferential luminosity is 0.58, with a much larger (20%) uncer-
tainty (S¨ oderhjelm 1999). Using the spectroscopic result, the
primary (GJ 704A) has a mass of 0.94M⊙, while the secondary
(GJ 704B) is 0.46M⊙.
The system configuration is shown in Figure 1 and is in-
clined such that the primary is currently closer to Earth than
the secondary. The position of the B component relative to A
on the date it was observed by Herschel in late April 2010 was
PA=314◦at an observed separation of 1.15” (22.6AU when de-
projected), indicated by grey circles in the Figure.
2.2A Third Component?
While the STIS images clearly resolve the binary, there is a pos-
sible third component with PA ≈ 284◦and ρ ≈ 0.27” that
is about 2.4 times as faint as the B component. Scardia et al.
(2008) also report a third component (epoch 2005.8) at PA ≈
50◦and ρ ≈ 0.228” (no magnitude is given). However, while
they detected the secondary again in mid-2007, they do not re-
port any detection of the putative tertiary (Scardia et al. 2010).
The detected positions are shown as star symbols in sky co-
ordinates in Figure 2, which shows the motion of the 99 Her
system. The system proper motion is µαcosδ = −110.32mas
yr−1, µδ = 110.08mas yr−1(van Leeuwen 2007), and ac-
counts for the motion of the primary assuming the orbit derived
in S¨ oderhjelm (1999), which is very similar to ours. The small
proper motion uncertainty means STISand Scardia et al. (2008)
cannot have seen the same object if it is fixed on the sky. There
is no clear sign of a third component in the residuals from fitting
the orbit of the secondary.
Figure 3. Keck/NIRC2 adaptive optics image of 99 Her at 3.8µm,
cropped to about 1.5” around the primary. North is up and East is left.
The saturated A component is at the center of the frame.
To try and resolve this issue we obtained an adaptive op-
tics image of 99 Her at L’ (3.8 µm) using the NIRC2 camera at
Keck II on July 27, 2011, shown in Figure 3. We adopted the
narrow camera (10 mas/pixel) and used a five-point dither pat-
tern with three images obtained at each position consisting of
25 coadds of 0.181 seconds integration. The cumulative integra-
tiontimefor thefinal co-registered andcoadded image is67.875
seconds. The core of the A component point-spread-function is
highly saturated, which degrades the achievable astrometry. We
estimate the position of 99 Her A by determining the geomet-
ric center of the first diffraction ring. The position of 99 Her B
is taken from the centroid of the unsaturated core. The PA and
separation are quoted above.
There is no detection of the putative 99 Her C within 1.6”
of the primary in the combined image if it is only a factor 2.4
fainter than the B component, because it would appear 20 times
brighter thanthebrightest speckles. However, if itwere closer to
the primary than 0.2” it would currently be too close to detect. If
the object was fixed on the sky near either the 2000 or 2005 lo-
cations, it would have been detected in the individual pointings
of the five-point dither since each NIRC2 pointing has a field
of view of 10′′× 10′′. To be outside the field of view and still
bound to the primary, the tertiary must have an apocenter larger
than about 75AU (5”). An object in such an orbit would have
a period of at least 200 years, so could not have been detected
near the star in 2005 and be outside the NIRC2 field of view in
2011.
The non-detections by Scardia et al. (2010) and NIRC2
make the existence of the tertiary suspicious. It is implausible
that the object was too close to, or behind the star both in 2007
and 2011, because at a semi-major axis of 0.23” (3.5AU) from
the primary (similar to the projected separation) the orbital pe-
riod is 7 years. Therefore, the object would be on opposite sides
of the primary, and the two detections already rule out an edge-
on orbit. Even assuming a circular orbit, such an object is un-
likely to be dynamically stable, given the high eccentricity and
c ? 0000 RAS, MNRAS 000, 000–000

Page 4

4
G. M. Kennedy et. al.
small pericenter distance (4.1AU) of the known companion. A
tertiary at this separation would be subject to both short term
perturbations and possible close encounters. If the mutual incli-
nation were high enough, it would also be subject to Kozai cy-
cles (Lidov 1962; Kozai 1962) due to the secondary that could
result in a high eccentricity and further affect the orbital stabil-
ity.
While it may be worthy of further detection attempts, the
existence of this component appears doubtful and we do not
consider it further.
3IR AND SUB-MM DATA
3.1Observations
Herschel Photodetector and Array Camera & Spectrometer
(PACS, Poglitsch et al. 2010) data at 100 and 160µm were
taken in April 2010 during routine DEBRIS observations. Sub-
sequently, a Spectral & Photometric Imaging Receiver (SPIRE,
Griffin et al. 2010) observation was triggered by the large PACS
excess indicating the presence of a debris disk and a likely sub-
mm detection. The disk was detected, but not resolved with
SPIRE at 250 and 350µm. A 70µm PACS image was later ob-
tained to better resolve the disk. Because every PACS obser-
vation includes the 160µm band, we have two images at this
wavelength, which are combined to produce a single higher
S/N image. All observations were taken in the standard scan-
map modes for our survey; mini scan-maps for PACS data and
small maps for SPIRE.Datawere reduced using a near-standard
pipeline with the Herschel Interactive Processing Environment
(HIPE Version 7.0, Ott 2010). We decrease the noise slightly by
including some data taken as the telescope is accelerating and
decelerating at the start and end of each scan leg.
The high level of redundancy provided by PACS scan
maps means that the pixel size used to generate maps can be
smaller than the natural scale of 3.2”/pix at 70 and 100µm
and 6.4”/pix at 160µm via an implementation of the “drizzle”
method (Fruchter & Hook 2002). Our maps are generated at
1”/pix at 70 and 100µm and 2”/pix at 160µm. The benefit of
better image sampling comes at the cost of correlated noise
(Fruchter & Hook 2002), which we discuss below.
In addition to correlated noise, two characteristics of the
PACS instrument combine to make interpretation of the data
challenging. The PACS beam has significant power at large an-
gular scales; about 10% of the energy lies beyond 1 arcminute
and the beam extends to about 17 arcminutes (1000 arcsec).
While this extent is not a problem in itself, it becomes prob-
lematic because PACS data are subject to fairly strong 1/f (low
frequency) noise and must be high-pass filtered. The result is
that a source will have a flux that is 10-20% too low because the
“wings” of the source were filtered out. While this problem can
be circumvented with aperture photometry using the appropri-
ate aperture corrections derived from the full beam extent, the
uncorrected apertures typically used for extended sources will
result in underestimates of the source flux.4
Here, we correct the fluxes measured in apertures for 99
Her based on a comparison between PSF fitted and aperture
corrected measurement of bright point sources in the DEBRIS
4See http://herschel.esac.esa.int/twiki/bin/view/Public/WebHome for
details regarding the PACS beam extent and calibration.
survey with predictions from their stellar models (based on the
calibration of Rieke et al. 2008). These upward corrections are
16 ± 5%, 19 ± 5%, and 21 ± 5% at 70, 100, and 160µm re-
spectively. These factors depend somewhat on the specifics of
the data reduction, so are not universal. This method assumes
that the correction for 99 Her is the same as for a point source,
which is reasonable because the scale at which flux is lost due
to filtering the large beam is much larger than the source ex-
tent. The corrected PACS measurement is consistent with MIPS
70µm, so we do not investigate this issue further.
The beam extent and filtering is also important for re-
solved modelling because the stellar photospheric contribution
to the image is decreased. Therefore, in generating a synthetic
star+disk image to compare with a raw PACS observation, the
stellar photospheric flux should be decreased by the appropri-
ate factor noted above. Alternatively, the PACS image could be
scaled up by the appropriate factor and the true photospheric
flux used.
Table 2 shows the measured star+disk flux density in each
Herschel image. Uncertainties for PACSare derived empirically
by measuring the standard deviation of the same sized apertures
placed at random image locations with similar integration time
to the center (i.e. regions with a similar noise level).
The SPIRE observations of 99 Her are unresolved. The
disk is detected with reasonable S/N at 250µm, marginally de-
tected at 350µm, and not detected at 500µm. Fluxes are ex-
tracted with PSF fitting to minimise the contribution of back-
ground objects. Because all three bands are observed simultane-
ously (i.e. a single pointing), the PSF fitting implementation fits
all three bands at once. A detection in at least one band means
that all fluxes (or upper limits) are derived at the same sky posi-
tion.
Additional IR data exist for 99 Her, taken with the Multi-
band Imaging Photometer for Spitzer (MIPS, Rieke et al. 2004).
Only the star was detected at 24µm (270.3 ± 0.1mJy), but this
observation provides confirmation of the 99 Her stellar position
in the PACS images relative to a background object 1.8 arcmin
away to the SE (PA = 120◦) that is visible at 24, 70, and
100µm. The presence of an excess at 70µm (98 ± 5mJy com-
pared to the photospheric value of 30mJy) was in fact reported
by Koerner et al. (2010). They did not note either the circumbi-
narynature or thatthediskmay bemarginallyresolvedby MIPS
at 70µm. Because our study focuses on the spatial structure, we
use the higher resolution PACS data at 70µm, but include the
MIPS data to model the SED.
3.2 Basic image analysis
Figure 4 shows the Herschel PACS data. Compared to the beam
size, the disk is clearly resolved at all three wavelengths. At
160µm the peak is offset about 5” East relative to both the 70
and 100µm images. However, the disk is still visible at 160µm
asthe lower contours match the70 and 100µm imageswell. The
160µm peak is only 2-3σ more significant than these contours.
While such variations are possible due to noise, in this case the
offset is the same in both 160µm images, so could be real. The
fact that the peak extends over several pixels isnot evidence that
it isreal, because the pixels in these maps are correlated (see be-
low). If real, this component of the disk or background object
cannot be very bright at SPIRE wavelengths because the mea-
sured fluxes appear consistent with a blackbody fit to the disk
c ? 0000 RAS, MNRAS 000, 000–000

Page 5

A Circumbinary Polar Ring Debris Disk
5
20 100 -10-20
RA offset (")
-20
-10
0
10
20
DEC offset (")
-0.10
0.04
0.18
0.32
0.46
0.60
0.74
20100 -10-20
RA offset (")
-20
-10
0
10
20
DEC offset (")
-0.08
0.00
0.08
0.16
0.23
0.31
0.39
20 100 -10 -20
RA offset (")
-20
-10
0
10
20
DEC offset (")
-0.03
0.01
0.06
0.10
0.15
0.19
0.23
Figure 4. Herschel PACS images of the 99 Her system at 70µm (left) 100µm (center) and combined 160µm (right). North is up and East is left. Colour
bars show levels in units of mJy/arcsec2. Contours in the left two panels show the 160µm data from each observation in 5 levels from 3-9σ. The images
are centered on the primary position (plus symbol) derived from the 70µm image. Hatched circles show the FWHM beam sizes of 5.6”, 6.8”, and 11.5”
at 70, 100, and 160µm respectively.
Table 2. Herschel photometry of 99 Her. The disk is not detected at
500µm and can be considered a 3σ upper limit of 24mJy.
BandFlux (mJy)UncertaintyMethod
PACS70
PACS100
PACS160
SPIRE250
SPIRE350
SPIRE500
93
87
80
44
22
4
10
10
15
6
7
8
15” aperture
15” aperture
17” aperture
PSF fit
PSF fit
PSF fit
(see §4). Based on an analysis of all DEBRIS maps (that have
a constant depth), the chance of a 3σ or brighter background
source appearing within 10” of 99 Her at 160µm is about 5%
(Thureau et al in prep). Given that the 160µm offset is only
a 2-3σ effect (i.e. could be a 2-3σ background source super-
imposed on a smooth disk), the probability is actually higher
because the number of background sources increases strongly
with depth. These objects have typical temperatures of 20–40K
(e.g. Amblard et al. 2010), so could easily appear in only the
160µm image, particularly if the disk flux is decreasing at this
wavelength.
Wenow analysethePACSimagesusing2D Gaussianmod-
els to estimate the disk size, inclination, and position angle. A
2D Gaussian fits the star-subtracted PACS 100µm image fairly
well, with major and minor full-width half-maxima of 17.7 and
12.8” at a position angle of 78◦. Quadratically deconvolving
from the 6.7” FWHM beam assuming a circular ring implies
an inclination of 48◦from face-on and an estimated diameter
of 250AU. Gaussian fitting to star-subtracted images at both 70
and 160µm yields similar results.
As noted above, estimation of uncertainties in these pa-
rameters is non-trivial due to correlated noise, but made easier
by the constant depth of our survey. By inserting the best fit
Gaussian to the star-subtracted image of the 99 Her disk from
the 100µm image into 438 other 100µm maps at high coverage
positions offset from the intended target, we obtain a range of
fit parameters for hundreds of different realisations of the same
noise. This process yields an inclination of 45 ± 5◦and PA of
75 ± 8◦. Repeating the process, but using the best fit Gaussian
for the 70µm image yields an inclination of 44 ± 6◦and PA of
68 ± 9◦. Though the inclination of the disk is similar to the bi-
nary, the position angle is significantly different from the binary
line of nodes of 41±2◦. This difference means that the disk and
binary orbital planes are misaligned.
As a check on the above approach, we can correct for the
correlated noise directly. Fruchter & Hook (2002) show that for
a map that has sufficiently many dithers (corresponding in our
case to many timeline samples across each pixel), a noise “cor-
rection” factor of r/(1 − 1/3r) can be derived, where r is the
ratio of natural to actual pixel scales and is 3.2 for our PACS
maps. A correction factor of 3.6 for the measured pixel to pixel
noise is therefore required when estimating the uncertainty on
a fitted Gaussian. Including this factor at 70µm and calculating
the uncertainty by the standard ∆χ2method yields an inclina-
tion of 42 ± 7◦and a PA of 68 ± 9◦. At 100µm the result is
an inclination of 44 ± 6◦and a PA of 76 ± 8◦. These results
are therefore almost exactly the same as the empirical method
used above and therefore lead to the same conclusion of mis-
alignment.
As will become apparent in §5, there is reason to believe
that the disk plane could be perpendicular to the binary pericen-
ter direction. The projection of the binary pericenter direction
on the sky plane has a PA of 163 ± 2◦, and a line perpendicular
tothis hasa PA of 73±2◦. Therefore, the observed disk position
angle of about 72◦is consistent with being at 90◦to the binary
pericenter direction.
4 SED
The combination of all photometry for 99 Her allows mod-
elling of the spectral energy distribution (SED). The model
is separated into two components; a stellar atmosphere and
a disk. Due to being fairly bright (V
tem is saturated in the 2MASS catalogue. However, suffi-
cient optical photometry for each individual star and the pair
exists (Henry & McCarthy 1993; Hauck & Mermilliod 1997;
Perryman & ESA 1997; Mermilliod 2006), as well as infra-
red measurements of the AB pair from AKARI and IRAS
(Moshir & et al. 1990; Ishihara et al. 2010). These data were
used to find the best fitting stellar models via χ2minimisation.
This method uses synthetic photometry over known bandpasses
and has been validated against high S/N MIPS 24µm data for
∼ 5mag) the sys-
c ? 0000 RAS, MNRAS 000, 000–000

Page 6

6
G. M. Kennedy et. al.
Figure 5. SEDforthe 99 Her system (both stars) showing the stellar and
disk models (grey lines) and star+disk model (black line). The black-
body disk model is the solid grey line, and the physical grain model the
dashed line. Photometric measurements are shown as black filled cir-
cles, and synthetic photometry of the stellar atmosphere as open circles
(U − B, B − V , & b − y colours, and m1 and c1 Stromgren indices
were fitted but are not shown here). Black triangles mark upper limits
from IRAS at 60 and 100µm.
DEBRIS targets, showing that the photospheric fluxes are ac-
curate to a few percent for AFG-type stars. The stellar lumi-
nosities (L⋆,A = 1.96L⊙, L⋆,B = 0.14L⊙) and IR fluxes of
the individual components are consistent with the fit for the pair
(L⋆,AB = 2.08L⊙). The fit for the AB pair is shown in Figure
5.
The spatial structure of the disk can be modelled with dust
at a single radial distance of 120AU (i.e. thin compared to Her-
schel’s resolution, §5), so disk SED modelling can be decoupled
from the resolved modelling once this radial distance is known.
Because wehave measurements of thedisk emission at only five
wavelengths, we cannot strongly constrain the grain properties
and size distribution. We fit the data with a blackbody model,
and then compare the data with several “realistic” grain mod-
els (Li & Greenberg 1997; Augereau et al. 2001; Wyatt & Dent
2002).
Infittingablackbody weaccount for inefficient grainemis-
sion at long wavelengths by including parameters λ0 and β,
where the blackbody is modified by a factor (λ0/λ)βfor wave-
lengths longer than λ0. The best fitting model has a temperature
of 49K and fractional luminosity Ldisk/L⋆ = 1.4 × 10−5. The
SPIRE data are near the confusion limit of about 6mJy, so the
parameters β and λ0 are unconstrained within reasonable lim-
its by the data (based on previous sub-mm detections for other
disks we fix them to λ0 = 210µm and β = 1 in Figure 5
(Wyatt et al. 2007b)).
Assuming that grains absorb and emit like blackbodies, the
radial dust distanceimpliedby49K is45AU. Becausethediskis
observed at a radius of 120AU (i.e. is warmer than expected for
blackbodies at 120AU), the dust emission has at least some con-
tribution from grains small enough to emit inefficiently in the
70-350µm wavelength range. Because the SED alone is consis-
tent with a pure blackbody (i.e. with β = 0), we cannot make
such a statement without the resolved images. However, actu-
ally constraining the grain sizes is difficult because temperature
and emission are also affected by composition. We fit the data
by generating disk SEDs for grains at a range of semi-major
axes and choosing the one with the lowest χ2. If the dust semi-
major axis is different from the observed semi-major axis of
120AU the model parameters are changed and the model recal-
culated, thus iterating towards the best fit.
We model the dust with a standard diameter (D) distribu-
tion n(D) ∝ D2−3qwhere q = 1.9 (equivalently n(M) ∝
M−qwhere M is mass O’Brien & Greenberg 2003), with the
minimum size set by the blowout limit for the specific composi-
tion used (about 1.4µm) and a maximum size of 10cm. The size
distribution probably extends beyond 10cm, but objects larger
than 10cm contribute negligibly totheemission because the size
distribution slope means that smaller grains dominate. Prelimi-
nary tests found that icy grains provided a much better fit than
silicates. To refine the grain model so the SED and resolved ra-
dius agree, we introduced small amounts of amorphous silicates
to theinitiallyicymodel. The grains aretherefore essentially ice
mixed with a small fraction (fsil = 1.5%) of silicate. The icy
grainmodel isshown asadottedlineinFigure5.Thismodel has
a total dust surface area of 14AU2and a mass of order 10M⊕if
the size distribution is extrapolated up to 1000km size objects.
The parameters from of this model are degenerate for the
data in hand; for example the size distribution could be shal-
lower and the fraction of silicates higher (e.g. q = 1.83 and
fsil = 4%). If we allow the minimum grain size to be larger
than the blowout limit, the disk is well fit by amorphous sili-
cate grains with q = 1.9 and Dbl = 10µm. The disk spectrum
can even be fit with a single size population of 25µm icy grains.
However, the predictions for the flux at millimeter wavelengths
depend on the size distribution, with lower fluxes for steeper
size distributions. Therefore, grain properties and size distribu-
tion can be further constrained in the future with deep (sub)mm
continuum measurements.
In summary, it is hard to constrain the grain sizes or prop-
erties. There is a difference in the required minimum grain size
that depends on composition. Because icy grains are reflective
at optical wavelengths, a detection of the disk in scattered light
could constrain the albedo of particles, and therefore their com-
position.
5SPATIAL STRUCTURE
The PACS images of the 99 Her disk are resolved, which al-
lows modelling of the spatial distribution of grains that con-
tribute to the observed emission at each wavelength. We com-
pare synthetic images with the Herschel observations in several
steps: i) Generate a three dimensional distribution of surface
area σ(r,θ,φ), where the coordinates are centered on the pri-
mary star. ii) Generate a radial distribution of grain emission
properties. Because the SED can be modelled with blackbody
grains at 49K and the spatial structure modelled with a narrow
ring, there is no real need for a radial temperature dependence
and the grain properties are only a function of wavelength:
P(λ) = Bν(49K,λ). Practically, we use a radial temperature
dependence T ∝ r−1/2centered on the primary, normalised so
that thediskring temperatureis49K. Thisapproach ensures that
temperature differences due to non-axisymmetries (negligible
for 99 Her) are automatically taken into account. iii) Generate a
high resolution model as viewed from a specific direction. The
emission in a single pixel of angular size x from a single vol-
c ? 0000 RAS, MNRAS 000, 000–000

Page 7

A Circumbinary Polar Ring Debris Disk
7
Figure 6. Simple ring model where the disk is aligned with the binary orbital plane, compared to the PACS 70 (top row) and 100µm (bottom row)
observations. North is up and East is left. Columns show the data, model convolved with the PACS beam, a high resolution model, and the data-model
residuals from left to right. The model includes stellar photospheric fluxes, which lie in two pixels near the ring center but are not visible in the high
resolution plot. In the residual plots, contours show ±2 and 3σ in units of pixel to pixel RMS.
ume element in the three dimensional model dV viewed along
a vector R is dFν(λ,r,θ,φ) = P(λ)σ(r,θ,φ)dV , where
dV = x2d2dR, so dR is the length of the volume element,
and d is the distance to the particles from Earth (Wyatt et al.
1999). The emission is derived by integrating along the line
of sight R for each pixel in the synthetic image. The photo-
spheric fluxes for each star (decreased by the factors noted in
§3.1) are placed in the relevant pixels at this step. iv) Con-
volve the high resolution model with a high resolution tele-
scope+instrument beam, for which we use interpolated and ro-
tated PACS images of the star Arcturus.5v) Degrade the resolu-
tion to match the data. vi) Generate a map of residuals, defined
by (observed − model)/uncertainty, where the uncertainty
is the pixel to pixel RMS for that observation. We compute the
model χ2from pixels in a square crop around the disk.
A minor consideration is that in the general circumbinary
case the disk temperature is not axisymmetric because the disk
orbits the center of mass, not the primary. An axisymmetric disk
is therefore subject to a temperature asymmetry such that it will
be slightly hotter, and therefore brighter, where the distance to
the primary is smallest. This “binary offset” asymmetry will
rotate with the primary, and will be most pronounced in the
coplanar case. The result of this effect is similar to the offset
caused by perturbations from an eccentric object (“ pericenter
glow” Wyatt et al. 1999). However, the pericenter glow is off-
set towards the pericenter of the perturbing object, so does not
rotate unless the perturbing object’s pericenter precesses. The
offset from the primary and the pericenter glow are completely
5The observations are reduced in sky coordinates, so the spacecraft
orientation is different for each observation and the PSF must be rotated
accordingly. This rotation step could be avoided by reducing both in
spacecraft coordinates.
independent effects. Therefore, if the pericenter glow effect is
present, it will reinforce and cancel the binary offset effect, de-
pending on the relative magnitude of each each offset. Themag-
nitude of the binary offset effect is negligibly small (?1%) be-
cause the disk radius is much larger than the binary separation.
Because our model is centered on the system center of mass this
effect is taken into account anyway. We discuss the effect of the
binary on pericenter glow in §5.1.1.
To fit the data requires a handful of parameters, some are
required for all models and some are model specific. The disk
surface area, temperature, radius, width, and total opening angle
are the five physical parameters for a ring. The sky position an-
gle and inclination are two further parameters that set the orien-
tation, but can be fixed if the disk plane is assumed to be aligned
with the binary. In addition each observation has the stellar RA
and Dec as parameters to allow for the 2” 1σ pointing accu-
racy of Herschel. The position at 160µm is tied to the 100µm
position. There are therefore eleven possible parameters to fit
for the resolved observations at 70, 100, and 160µm. We fix the
disk temperature to 49K in all cases.
From the basic analysis (§3.2), a simple ring coplanar with
the binary does not appear a viable option. To emphasise this
point we show 70 and 100µm images of the best fitting copla-
nar model in Figure 6. This model was generated by the steps
outlined above, and the rightmost three panels are the results of
steps iv (convolved model), iii (high resolution model), and vi
(residuals). We fix the disk width to 20AU, the opening angle to
5◦, and the position angle and inclination to the binary plane, so
there are six free parameters (surface area, radius, and two pairs
of RA/Dec sky positions). While we include the 160µm data
in the fitting, it does not constrain the fit strongly due to low
S/N and always shows ∼2σ residual structure due to the off-
set peak. For comparison with the models below, the χ2value
c ? 0000 RAS, MNRAS 000, 000–000

Page 8

8
G. M. Kennedy et. al.
for all three PACS bands is 4278 with 3797 degrees of freedom.
The positive and negative residuals (rightmost panels) show that
the disk ansae in the model have the wrong orientation at both
wavelengths. It is clear that any structure symmetric about the
binary line of nodes will not be consistent with the observations
because the position angle is significantly different.
An alternative explanation for the misalignment between
the observed position angle and the binary line of nodes could
be that the dust does in fact lie in the binary plane, but that
the particles are on eccentric orbits with common pericenter di-
rections (i.e. the disk is elliptical and offset from the binary).
In principle, the observations can constrain the eccentricity and
pericenter direction. However, this model fails because the ec-
centricity needed to match the observed position angle istoo ex-
treme. In order to obtain an ellipse that lies in the binary orbital
plane and has a position angle and aspect ratio similar to the ob-
servations requires eccentricities?0.4. Theeccentricity of these
particles is so high that i) the ring is significantly offset from the
star and ii) the ring has an extreme pericenter glow asymmetry
at all wavelengths caused by particles residing at different stel-
locentric distances. Because the PACS 70µm image shows that
the star lies very near the disk center, such a strong offset is
ruled out.
Wenow consider tworelativelysimplemodelsthataccount
for themisalignment between the disk and binary orbital planes.
The first is based on the expected secular evolution of circumbi-
nary particles, and the second is a simple misaligned ring where
the disk position angle and inclination are free parameters.
5.1 Secularly perturbed polar ring
In this section we consider a ring inspired by the secular evo-
lution of circumbinary particles. This approach ensures that the
disk is stable over the stellar lifetime and encompasses the par-
ticle dynamics dictated by the binary. We first outline the dy-
namics of circumbinary particles, and then show the model for
the 99 Her disk.
5.1.1Dynamics
Particle dynamics are important for evolution and stability in
the 99 Her system. A circumbinary disk will have its inner edge
truncated, while circumstellar disks around either component
can be truncated at their outer edges. In addition, secular per-
turbations lead to precession of test particles’ nodes coupled
with inclination variations. We explore these dynamics using
the Swift HJS integrator (Beust 2003).
In general, disk truncation allows us to place limits on pos-
sible locations for disk particles. However, in the case of 99
Her there is no evidence for disk components orbiting only one
star, and the apparent circumbinary disk extent lies well beyond
∼30-60AU stabilitylimit at any inclination (Wiegert & Holman
1997; Doolin & Blundell 2011).
Circumbinary particles also undergo long-term dynami-
cal evolution due to secular perturbations. Because the binary
has a small mass ratio and high eccentricity, the dynamics are
not well described by the circular restricted three-body prob-
lem, commonly applied in the case of debris disks perturbed by
planets. Similar dynamics have previously been explored in the
context of the HD 98800 system (Verrier & Evans 2008, 2009)
Figure 7. Secular evolution of circumbinary particles in inclination (i)
and line of nodes (Ω) space. Particles begin at dots and move along the
curves due to perturbations from the binary. Particles that would appear
reflected in the x axis duplicate the spatial distribution so are not shown.
Crosses show the current location of particles in the two interpretations
of the transient ring model (§5.2). Over time these particles will sweep
out curves similar to particles 4 and 11. The long term structure of the
transient ring will therefore appear similar to either panel 4 or 11 in
Figure 8, depending on which inclination is correct.
and more generally (Farago & Laskar 2010; Doolin & Blundell
2011).
Thesestudiesshow that theinclination(i) andlineof nodes
(Ω) of circumbinary particles evolve due to perturbations from
the binary. Depending on the binary eccentricity and particle in-
clination, Ω can circulate (increase or decrease continuously) or
librate (oscillate about 90 or 270◦). Particles with low inclina-
tions stay on low inclination orbits, thus sweeping out a roughly
disk or torus-like volume over long timescales. Higher inclina-
tion particles are subject to nodal libration and large inclination
variations, thus sweeping out large parts of a sphere around the
binary. Most importantly for 99 Her, the orbits of particles with
Ω ≈ 90◦(or 270◦) and on near polar orbits will not change
much due to secular evolution, thus sweeping out a polar ring.
Figure 7 shows the secular evolution of 23 particles on ini-
tially circular orbits in complex inclination space. All particles
have initial nodes of 90◦relative to the binary pericenter and in-
clinations spread evenly between 0 and 180◦and are integrated
for 1Gyr (i.e. there are no other significant effects on such long
timescales). At 120AU, the time taken for a particle to complete
one cycle of secular evolution (make a loop in figure 7) varies in
the range 2-7×105years, with larger loops taking longer. These
times will also scale with particle semi-major axis. Particles 1-
12 are those with initial inclinations between 0-90◦that are suf-
ficient to describe the range of spatial structures as we cannot
distinguish between pro/retrograde orbits.
The particles can be split into two groups; those with low
inclinations whose nodes circulate (1–3) and those with high in-
clinations whose nodes librate about 90◦(4–12). The dividing
line (separatrix) between these families for the binary eccen-
tricity of 0.76 is 21◦when Ω = 90◦(or 270◦, Farago & Laskar
2010). While particles in the first group have i < 21◦when
Ω = 90◦, their inclinations when Ω = 0◦(or 180◦) can be
as high as 90◦. Thus, particles near the separatrix will sweep
out an entire spherical shell during their secular evolution. Sim-
ilarly, particles near the separatrix but in the second group also
c ? 0000 RAS, MNRAS 000, 000–000

Page 9

A Circumbinary Polar Ring Debris Disk
9
Figure 8. Debris structures derived from the secular evolution of particles 1-12 in Figure 7 as they would be seen in the 99 Her system. In panels 1 and
12 we have included the binary orbit from Figure 1. The orbit is not to scale in panel 1 for better visualisation, but is in panel 12. Panel 10 includes
a line along the binary pericenter direction that is obscured by the ring to show the orientation. In panels 1-3 particles nodes circulate and the binary
orbital plane is the plane of symmetry. In panel 1, the coplanar case, the position angle of the disk is aligned with the line of nodes. In panel 3 the
binary is nearly surrounded by a broad shell of particles. In panels 4-12 particles nodes librate and the plane of symmetry is perpendicular to the binary
pericenter direction. In panels 7-12 the observed position angle is perpendicular to the binary pericenter.
sweep out a spherical shell, though the orbital evolution is dif-
ferent.
To visualise the structures swept out by these families of
particles due to secular perturbations, Figure 8 shows the result-
ing debris structures for particles that follow each of the trajec-
tories1-12 from Figure7(left toright and down). Thestructures
are oriented as they would be seen on the sky in the 99 Her sys-
tem (i.e. have the same orientation with respect to the binary or-
bit shown in Figure 1). Each structure was generated by taking
the relevant particle at each time step and spawning 1000 addi-
tional particles spread randomly around the orbit. This process
was repeated for every time step, thus building up the spatial
density of a family of particles that follow a specific curve in
Figure 7. These structures are optically thin, which makes in-
terpreting them somewhat difficult. We have included a scaled
version of the binary orbit from Figure 1 in some panels in an
attempt to make the orientations clearer.
The first (top left) panel shows a circular orbit coplanar
with the binary. The PA is the binary line of nodes, and Figure 6
shows why a disk in the plane of the binary is not a satisfactory
match to the observations. The second and third panels are still
symmetric about thebinary orbital plane, but havea wider range
of inclinations and arean even poorer match tothe observations.
Panel 3 shows that while particle inclinations are restricted for
Ω = 90,270◦, they can be large for Ω = 0,180◦and result in
a “butterfly” structure when viewed down the binary pericenter
direction.
The remaining panels are for particles 4-12, whose nodes
librate and for which the plane of symmetry is perpendicular to
the binary pericenter direction. In panel 4 the range of nodes
and inclinations is so large that a particle sweeps out nearly an
entire spherical shell during a cycle of secular evolution (i.e.
the particle is near the separatrix). This range decreases as the
initial inclination nears a polar orbit, at which point the orbital
elements do not evolve and the resulting structure appears in
panel 12 as a simple ring. The key difference from the ring in
panel 1 is that this ring’s position angle is perpendicular to the
sky projection of the binary pericenter direction, and as noted in
§3.2 istherefore similar to the observed PA in the PACS images.
Secular perturbations from the binary also affect the long
term evolution of particle eccentricities and pericenter longi-
tudes. These effects are taken into account by our n-body ap-
proach. However, we noticed that the eccentricities imposed
(“forced”) on the particles are lower than would be expected for
c ? 0000 RAS, MNRAS 000, 000–000

Page 10

10
G. M. Kennedy et. al.
a lower mass companion. Further n-body simulations of copla-
nar particles show that for 99 Her with a mass ratio of 0.49 the
forced eccentricity at 120AU is about 0.03, but if the mass ratio
were 0.05 the forced e is 0.1.
This lack of significant eccentricity forcing is visible by
its absence in Figure 8, where the structures would be much
broader if there were a large range of particle eccentricities.
For example, if the mass of the secondary in the 99 Her system
were significantly smaller, the model in panel 1 would become
broader and offset from the binary center of mass, resulting in a
small pericenter glow effect.
This dependence suggests that a circumbinary disk’s struc-
ture may help constrain the binary mass ratio in cases where
it is uncertain. However, we cannot apply this idea to make a
better estimate of the 99 Her mass ratio because the PACS ob-
servations do not have enough resolution. In addition, at high
inclinations the particlebehaviour ismore complicated, because
polar particles switch between pro and retrograde orbits and do
not follow simple circles in complex eccentricity space.
5.1.2 Polar ring model
We now use the models from Figure 8 to fit the PACS obser-
vations. The model has only seven free parameters; the particle
semi-major axis and initial inclination, the surface area of dust,
and the same four RA/Dec positions. The dust temperature is
fixed to 49K. Using a semi-major axis of 120AU, each panel
was compared to the PACS images, setting the surface area in
grains for each model to obtain the least residual structure. Of
these we found that panel 9 was the best fit, shown in Figure
9. These particles follow near-polar orbits so we call this model
a “polar ring.” We find χ2= 3202. In terms of χ2the results
for panels 8 and 10 are similar, but slightly higher. The uncer-
tainty in the initial inclination is therefore about 10◦, and for the
semi-major axisabout 10AU.Thismodel ismuchbetter thanthe
coplanar model of Figure 6, with no overlapping residual struc-
ture at 70 and 100µm. The particles likely occupy a wider range
of orbits than a single semi-major axis with some non-zero ec-
centricity, which may account from some minor (2σ) structure
in the residuals at the disk ansae at 70µm. However, given that
this model stems directly from the secular evolution, has very
few free parameters, and accounts for the structure in all PACS
images, we consider it a plausible explanation.
5.2Transient ring model
A simple circular ring is a natural model to fit to the observa-
tions. This model has eight free parameters, with the width of
the ring fixed at 20AU and the opening angle fixed to 5◦. As ex-
pected fromthe simple analysis in§3.2 the position angle of this
ring is not aligned with the binary line of nodes, and is therefore
misaligned with the binary orbit.
The interpretationdepends on theorientation of thebest fit.
A misaligned ring with polar orbits and the correct line of nodes
would be considered further evidence in favour of the above
polar ring model. A ring with a non-polar orientation will be
spreadintoabroadstructurelikeoneof thepanelsinFigure8by
secular perturbations. The ring cannot be long-lived and could
therefore be the aftermath of a recent collision, seen after the
collision products have sheared into a ring, but before secular
perturbations destroy the ring structure. Thus we call this model
a “transient ring.”
This model is shown in Figure 10, and is a reasonable
match to the PACS observations. However, the residuals at
70µm show that the ring produces a structure that is slightly
too elliptical, compared to the more rectangular structure that is
observed and reproduced by the polar ring. This model also has
less emission at the stellar position than is observed. For this
model χ2= 3304. The disk is inclined 53◦from face-on and
the PA is 81◦. The uncertainties are similar to those derived for
the Gaussian fits in §3.2. The minimum relative inclination be-
tween the disk and binary orbital planes is therefore 32 degrees,
with a line of nodes with respect to the binary orbit of 139◦.
However, theinclination between the disk and binary plane
could also be 87◦if the disk were mirrored in the sky plane,
which means that the particles have near-polar orbits. These or-
bits are nearly the same as panel 12 of Figure 8 (the narrow
polar ring) because the line of nodes with respect to the binary
orbit is 276◦.
These two interpretations correspond to two points in Fig-
ure 7, shown as crosses. Over time the particles would spread
around to make two more lines similar to those drawn. The par-
ticles in the lower inclination case are close to the separatrix,
and would therefore sweep out a near-spherical shell like panel
4 of Figure 8. In this case, the long term evolution produces
structures that have the wrong position angle and are a poor
match to the observations. The higher inclination case is very
nearly a polar ring and would look very similar to panel 11.
Such a result is expected because we found above that the polar
ring model works well, and argues in favour of the polar-ring
interpretation.
We can in fact improve this simple ring model by increas-
ing the total disk opening angle (i.e. allowing a larger range
of inclinations), which emulates the range of inclinations that
result from the secular evolution. We find a best fit when the
particle inclinations are 25◦(total opening angle of 50◦), where
χ2= 3210. Thismodel looks very similar to thepreferred polar
ring model above, but is not generated in the same way, and will
therefore change somewhat due to secular perturbations over
time because the disk is not perfectly polar.
6DISCUSSION
We strongly favour the polar ring model as the best explanation
of the disk structure surrounding 99 Her. The polar ring is sta-
ble for the stellar lifetime, and takes the secular dynamics into
account. The transient ring model, where the disk orientation is
not fixed, also finds that the disk particles can have polar orbits.
However, because the ring could be mirrored in the sky plane
and appear the same, the ring could be misaligned with the bi-
nary orbital plane by about 30◦. Based on χ2and the residuals
the polar ring is marginally preferable over the transient ring.
However, given that a more realistic model with a range of par-
ticle radii and inclinations could improve the fit in each case, we
do not assign much importance to the relatively small χ2differ-
ences. Instead, we consider several constraints on the collisional
evolution that argue against the transient ring interpretation.
By considering the timescales for collisions and secular
evolution, we can estimate the likelihood of observing the prod-
ucts of a collision as a transient ring before it is spread into a
broader structure. Based on the observed radius and total cross-
c ? 0000 RAS, MNRAS 000, 000–000

Page 11

A Circumbinary Polar Ring Debris Disk
11
Figure 9. Polar ring model compared to the PACS 70 (top row) and 100µm (bottom row) observations. Layout is as for Figure 6.
Figure 10. Transient ring model compared to the PACS 70 (top row) and 100µm (bottom row) observations. Layout is as for Figure 6.
sectional area, the collisional lifetime of grains just above the
blowout size is about a million years (Wyatt et al. 2007a). The
emission could last much longer if larger objects exist in the
size distribution, and the lifetime scales with the maximum size
as?Dmax/Dbl, so depends on the size of the largest fragment
created in the collision. If the largest fragments are at least 1mm
in size the lifetime is at least 50Myr, and we would expect the
collisional cascade to be detectable for this length of time. The
secular precession timescale is about 0.5Myr, and it is reason-
able to assume that the ring structure would be erased by secular
perturbations within 10 secular timescales. Thus, the collisional
products would be observable as a ring for only 5Myr. Because
the collision time is longer than the secular time, the collision
products would spend at most a tenth of their lifetime appearing
as a misaligned ring, and the remainder as a broader structure.
That is, assuming such a collision did occur, we have less than
c ? 0000 RAS, MNRAS 000, 000–000

Page 12

12
G. M. Kennedy et. al.
a 1:10 chance of observing the collision products as a ring that
looks like the Herschel observations.6
While 1:10 is not unreasonable, this estimate does not con-
sider the object that must be destroyed to generate the observed
dust or the plausibility of a 1mm maximum size. To produce the
observed fractional luminosity, a parent body of at least 600km
in diameter must be broken into blowout sized grains. With the
more realistic assumption that the collision produced a range
of grain sizes, the parent body must be larger, about 2000km
if grains were as large as the 1mm assumed above (assuming
q = 11/6). Under the more realisticassumption of a wide range
of fragment sizes, up to 1km say, the parent body would need to
be roughly Earth-sized. However, for such large fragments the
dust lifetime would be 50Gyr and the chance of observing the
structure as a ring very unlikely (1:10,000).
We can estimate the ability of collisions to smash large ob-
jects into small pieces by considering their strength and possi-
ble collision velocities. The specific energy needed for catas-
trophic disruption, where the largest collision product is half
the original mass (i.e. very large), is roughly 1011erg/g for ob-
jects 2000km in size (Stewart & Leinhardt 2009). The energy
needed to disrupt an object so that the collision products are
all very small must be larger. The maximum collision energy
possible for circular orbits is for a collision between two equal
sized objects on pro and retrograde orbits. The collision energy
assuming such an impact at twice the orbital velocity of a few
km/s at 100AU is a few 1010erg/g. Therefore, only in the most
optimistic (highest velocity) case is the collision energy suffi-
cient to catastrophically disrupt 2000km objects. In the even of
a disruption, the lifetime of the collision products will be very
long because the largest remnant is about 1000km in size. In
the more realistic case where collision velocities are set by ob-
ject eccentricities and inclinations, disruption of large objects
at large semi-major axes is even more difficult. This difficulty,
combined with the smaller amount of starlight intercepted at
suchdistancesmeans thatsinglecollisionsonlyproduce aminor
increase over the background level of dust (Kenyon & Bromley
2005). These probability and collision arguments suggest that
a single collision is an extremely unlikely explanation for the
origin of the observed dust.
The polar ring model does not have these issues. The secu-
lar evolution of particles in the 99 Her system means that parti-
cles on polar orbits suffer only minor changes in inclination and
node (Fig 7). These orbits are therefore stable over the stellar
lifetime so the dust could be the steady state collision products
of the polar planetesimal belt. Initial misalignment is therefore
the only special requirement for the polar ring model. The ex-
cellent agreement between the PACS data and a simple model
generated by particles on these orbits argues strongly in favour
of this interpretation.
The question is then shifted to one of the origin of the mis-
alignment. Most binaries are thought to form through fragmen-
tation and subsequent accretion during collapse of a molecular
cloud (for a recent review see Goodwin et al. 2007). The result-
6Had we found that an eccentric ring could explain the data, the same
argument applied to ring spreading by pericenter precession would ap-
ply, with the same 1:10 result. The particles’ pericenter directions are
unlikely to be maintained through forcing by a third (circumbinary)
body as for a standard pericenter glow model, because the perturbing
body would be subject to the same pericenter precession.
ing binary systems should be aligned with their protoplanetary
disks when the separations are of order tens of AU (Bate et al.
2000). Given the 16AU separation of the 99 Her system, it
therefore seems that interactions during the subsequent phase
of dynamical cluster evolution are a more likely cause of a mis-
aligned disk.
Thereareseveral waysthatsuchaconfiguration couldarise
from interactions in a young stellar cluster. A close encounter in
which a binary captures some material from the outer regions of
a circumstellar disk hosted by another star seems possible. This
“disk exchange” scenario requires an encounter where the bi-
nary pericenter is perpendicular to the circumstellar disk plane,
and that the encounter distance and geometry captures material
into orbits similar to those observed for the debris disk (e.g.
most likely a prograde rather than retrograde encounter).
An alternative scenario is a stellar exchange reaction,
where a binary encounters a single star that harbours a cir-
cumstellar disk. During the exchange one of the binary com-
ponents is captured by the originally single star, and the other
leaves (e.g. Moeckel & Goddi 2011). The post-encounter con-
figuration is then a binary surrounded by a circumbinary disk.
If the binary pericenter direction were perpendicular to the disk
plane it could represent a young analogue of the 99 Her system.
Such an encounter would require that the disk is not irrepara-
bly damaged by large stellar excursions during the exchange
(Moeckel & Goddi 2011), but may also present a way to clear
inner disk regions, thus providing a possible reason that the 99
Her disk resides at 120AU and not closer, where it could still be
stable (see § 5.1.1).
Both scenarios require some degree of tuning; the encoun-
ters must happen with specific geometries to produce the ob-
served relative binary and disk orientations. However, differ-
ences in the surface brightness between the different models
in Figure 8 mean there could be some selection bias towards
more disk-like structures. The advantage of the disk exchange
scenario is that the cross section for interaction at a distance of
about 100AU is much higher than for stellar exchange, which
would need to have an encounter distance similar to the binary
semi-major axis. With a factor of about ten difference in the en-
counter impact parameter for each scenario, the close encounter
is therefore about 100 times more likely than the exchange (ig-
noring other constraints on geometry, configuration etc.).
In the absence of detailed simulations of encounter out-
comes, some data exist to help distinguish between these two
scenarios. Theminimum inclination of the stellar pole for the99
Her primary relative tothe binary orbital plane is20±10◦(Hale
1994). The inclination difference is therefore different from the
binary plane with 95% confidence, and is a hint that the system
may be the result of an exchange. However, the scatter in incli-
nation differences for binaries with separations similar to that
of 99 Her is about 20◦(Hale 1994), which may indicate that
systems with this separation are in fact aligned and the uncer-
tainties were underestimated, or that this scatter is the intrinsic
level of misalignment at these separations.
Though 99 Herculis is the first clear case of misalignment
between binary and disk planes, the GG Tauri system may show
asimilar signature. TheGG Tausystem consists of an Aa/Abbi-
nary surrounded by a circumbinary ring (Guilloteau et al. 1999;
Pi´ etu et al. 2011), and a more distant Ba/Bb pair that may be
bound (Beust & Dutrey 2006). It is not clear if the inner bi-
nary is misaligned with the circumbinary disk, but is suggested
because if they are aligned the ring’s inner edge is too distant
c ? 0000 RAS, MNRAS 000, 000–000

Page 13

A Circumbinary Polar Ring Debris Disk
13
to be set by the binary (Beust & Dutrey 2005, 2006; K¨ ohler
2011). However, there could also be problems if they are mis-
aligned, because the expected disk scale height due to pertur-
bations from the binary may be inconsistent with observations
(Beust & Dutrey 2006). Though uncertain, the possible mis-
alignment between the binary and ring planes shows that GG
Tau could be a young analogue of 99 Her-like systems.
7SUMMARY
We have modelled the resolved circumbinary debris disk in the
99 Her system. This disk is unusual because it appears mis-
aligned with the binary plane. It can be explained as either an
inclined transient ring due to a recent collision, or more likely a
ring of polar orbits. The transient ring is shown to be implausi-
ble from collisional arguments. While the inclined ring cannot
exist on long (secular) timescales, the polar ring can.
There appear tobetwopossible formation scenarios for the
polar ring model, which both invoke stellar encounters. The bi-
nary may have captured material from another stars’ circumstel-
lar disk, or a new binary may have formed in a stellar exchange
where oneof thesystems already contained acircumstellar disk.
While many binary and multiple systems are known to
have debris disks, none are resolved and have orbits charac-
terised as well as 99 Herculis. Future efforts should characterise
this system further to test our interpretation and attempt to find
more examples. A sample of resolved circumbinary disks would
test whether disk-binary misalignment is a common outcome of
star formation and cluster evolution, with implications for plan-
etary systems around both single and binary stars.
ACKNOWLEDGMENTS
We are grateful to the referee for a thorough reading of the
manuscript, especially for noting that previous 99 Her visual
orbits have the wrong ascending node. This research has made
use of the Washington Double Star Catalog maintained at the
U.S. Naval Observatory, and the SwiftVis n-body visualisation
software developed by Mark Lewis. We also thank Herve Beust
for use of the HJS code, and Paul Harvey for comments on a
draft of this article.
REFERENCES
Abt, H. A. & Willmarth, D. 2006, ApJS, 162, 207 [ADS]
Adelman, S. J., Caliskan, H., Kocer, D., Cay, I. H., & Gokmen
Tektunali, H. 2000, MNRAS, 316, 514 [ADS]
Amblard, A. et al. 2010, A&A, 518, L9 [ADS]
Augereau, J. C., Nelson, R. P., Lagrange, A. M., Papaloizou,
J. C. B., & Mouillet, D. 2001, A&A, 370, 447 [ADS]
Bate, M. R., Bonnell, I. A., Clarke, C. J., Lubow, S. H.,
Ogilvie, G. I., Pringle, J. E., & Tout, C. A. 2000, MNRAS,
317, 773 [ADS]
Beust, H. 2003, A&A, 400, 1129 [ADS]
Beust, H. & Dutrey, A. 2005, A&A, 439, 585 [ADS]
—. 2006, A&A, 446, 137 [ADS]
Boesgaard, A. M., McGrath, E. J., Lambert, D. L., & Cunha,
K. 2004, ApJ, 606, 306 [ADS]
Churcher, L. J., Wyatt, M. C., Duchˆ ene, G., Sibthorpe, B.,
Kennedy, G., Matthews, B. C., Kalas, P., Greaves, J., Su, K.,
& Rieke, G. 2011, MNRAS, 417, 1715 [ADS]
Dommanget, J. & Nys, O. 2002, VizieR Online Data Catalog,
1274, 0 [ADS]
Doolin, S. & Blundell, K. M. 2011, ArXiv e-prints,
(1108.4144) [ADS]
Farago, F. & Laskar, J. 2010, MNRAS, 401, 1189 [ADS]
Fruchter, A. S. & Hook, R. N. 2002, PASP, 114, 144 [ADS]
Goodwin, S. P., Kroupa, P., Goodman, A., & Burkert, A. 2007,
Protostars and Planets V, 133 [ADS]
Gratton, R. G., Carretta, E., & Castelli, F. 1996, VizieR Online
Data Catalog, 331, 40191 [ADS]
Griffin, M. J. et al. 2010, A&A, 518, L3 [ADS]
Guilloteau, S., Dutrey, A., & Simon, M. 1999, A&A, 348, 570
[ADS]
Hale, A. 1994, AJ, 107, 306 [ADS]
Hartkopf, W. I., Mason, B. D., & Worley, C. E. 2001, AJ, 122,
3472 [ADS]
Hauck, B. & Mermilliod, M. 1997, VizieR Online Data Cata-
log, 2215, 0 [ADS]
Henry, T. J. & McCarthy, Jr., D. W. 1993, AJ, 106, 773 [ADS]
Ishihara, D. et al. 2010, A&A, 514, A1 [ADS]
Kenyon, S. J. & Bromley, B. C. 2005, AJ, 130, 269 [ADS]
Koerner, D. W.,Kim, S.,Trilling, D.E.,Larson, H.,Cotera, A.,
Stapelfeldt, K. R., Wahhaj, Z., Fajardo-Acosta, S., Padgett,
D., & Backman, D. 2010, ApJ, 710, L26 [ADS]
K¨ ohler, R. 2011, A&A, 530, A126 [ADS]
Kozai, Y. 1962, AJ, 67, 591 [ADS]
Li, A. & Greenberg, J. M. 1997, A&A, 323, 566 [ADS]
Lidov, M. L. 1962, Planet. Space Sci., 9, 719 [ADS]
Mason, B. D., Wycoff, G. L., Hartkopf, W. I., Douglass, G. G.,
& Worley, C. E. 2011, VizieR Online Data Catalog, 1, 2026
[ADS]
Matthews, B. C. et al. 2010, A&A, 518, L135 [ADS]
Mermilliod, J. C. 2006, VizieR Online Data Catalog, 2168, 0
[ADS]
Moeckel, N. & Goddi, C. 2011, ArXiv e-prints, (1109.2007)
[ADS]
Moshir, M. & et al. 1990, in IRAS Faint Source Catalogue,
version 2.0 (1990), 0 [ADS]
Nordstr¨ om, B., Mayor, M., Andersen, J., Holmberg, J., Pont,
F., Jørgensen, B. R., Olsen, E. H., Udry, S., & Mowlavi, N.
2004, A&A, 418, 989 [ADS]
O’Brien, D. P. & Greenberg, R. 2003, Icarus, 164, 334 [ADS]
Ott, S. 2010, in Astronomical Society of the Pacific Con-
ference Series, Vol. 434, Astronomical Data Analysis Soft-
ware and Systems XIX, ed. Y. Mizumoto, K.-I. Morita, &
M. Ohishi, 139 [ADS]
Perryman, M. A. C. & ESA, eds. 1997, ESA Special Publica-
tion, Vol. 1200, The HIPPARCOS and TYCHO catalogues.
Astrometricand photometric star catalogues derived from the
ESA HIPPARCOS Space Astrometry Mission [ADS]
Phillips,N. M., Greaves, J. S., Dent, W. R. F.,Matthews, B. C.,
Holland, W. S., Wyatt, M. C., & Sibthorpe, B. 2010, MN-
RAS, 403, 1089 [ADS]
Pi´ etu, V., Gueth, F., Hily-Blant, P., Schuster, K.-F., & Pety, J.
2011, A&A, 528, A81 [ADS]
Pilbratt, G. L., Riedinger, J. R., Passvogel, T., Crone, G.,
Doyle, D., Gageur, U., Heras, A. M., Jewell, C., Metcalfe,
L., Ott, S., & Schmidt, M. 2010, A&A, 518, L1 [ADS]
Poglitsch, A. et al. 2010, A&A, 518, L2 [ADS]
c ? 0000 RAS, MNRAS 000, 000–000

Page 14

14
G. M. Kennedy et. al.
Rieke, G. H. et al. 2004, ApJS, 154, 25 [ADS]
—. 2008, AJ, 135, 2245 [ADS]
Scardia, M., Prieur, J.-L.,Pansecchi, L.,Argyle, R. W.,& Sala,
M. 2010, Astronomische Nachrichten, 331, 286 [ADS]
Scardia, M., Prieur, J.-L., Pansecchi, L., Argyle, R. W., Sala,
M., Basso, S., Ghigo, M., Koechlin, L., & Aristidi, E. 2008,
Astronomische Nachrichten, 329, 54 [ADS]
S¨ oderhjelm, S. 1999, A&A, 341, 121 [ADS]
Stewart, S. T. & Leinhardt, Z. M. 2009, ApJ, 691, L133 [ADS]
Takeda, Y. 2007, PASJ, 59, 335 [ADS]
Trilling, D. E., Stansberry, J. A., Stapelfeldt, K. R., Rieke,
G. H., Su, K. Y. L., Gray, R. O., Corbally, C. J., Bryden, G.,
Chen, C. H., Boden, A., & Beichman, C. A. 2007, ApJ, 658,
1289 [ADS]
van Leeuwen, F., ed. 2007, Astrophysics and Space Science
Library, Vol. 350, Hipparcos, the New Reduction of the Raw
Data [ADS]
van Leeuwen, F. 2008, VizieR Online Data Catalog, 1311, 0
[ADS]
Verrier, P.E.& Evans, N.W. 2008, MNRAS, 390, 1377 [ADS]
—. 2009, MNRAS, 394, 1721 [ADS]
Wiegert, P. A. & Holman, M. J. 1997, AJ, 113, 1445 [ADS]
Wyatt,M. C. &Dent, W. R. F.2002, MNRAS,334, 589 [ADS]
Wyatt, M. C., Dermott, S. F., Telesco, C. M., Fisher, R. S.,
Grogan, K., Holmes, E. K., & Pi˜ na, R. K. 1999, ApJ, 527,
918 [ADS]
Wyatt, M. C., Smith, R., Greaves, J. S., Beichman, C. A., Bry-
den, G., & Lisse, C. M. 2007a, ApJ, 658, 569 [ADS]
Wyatt, M. C., Smith, R., Su, K. Y. L., Rieke, G. H., Greaves,
J. S., Beichman, C. A., & Bryden, G. 2007b, ApJ, 663, 365
[ADS]
c ? 0000 RAS, MNRAS 000, 000–000