The enigma of GCIRS 3 - Constraining the properties of the mid-infrared reference star of the central parsec of the Milky Way with optical long baseline interferometry

Abstract

GCIRS3 is the most prominent MIR source in the central pc of the Galaxy. NIR spectroscopy failed to solve the enigma of its nature. The properties of extreme individual objects of the central stellar cluster contribute to our knowledge of star and dust formation close to a supermassive black hole. We initiated an interferometric experiment to understand IRS3 and investigate its properties as spectroscopic and interferometric reference star at 10um. VISIR imaging separates a compact source from diffuse, surrounding emission. The VLTI/MIDI instrument was used to measure visibilities at 10mas resolution of that compact 10um source, still unresolved by a single VLT. Photometry data were added to enable simple SED- and full radiative transfer-models of the data. The luminosity and size estimates show that IRS3 is probably a cool carbon star enshrouded by a complex dust distribution. Dust temperatures were derived. The coinciding interpretation of multiple datasets confirm dust emission at several spatial scales. The IF data resolve the innermost area of dust formation. Despite observed deep silicate absorption towards IRS3 we favor a carbon rich chemistry of the circumstellar dust shell. The silicate absorption most probably takes place in the outer diffuse dust, which is mostly ignored by MIDI measurements. This indicates physically and chemically distinct conditions of the local dust, changing with the distance to IRS3. We have demonstrated that optical long baseline interferometry at infrared wavelengths is an indispensable tool to investigate sources at the Galactic Center. Our findings suggest further studies of the composition of interstellar dust and the shape of the 10um silicate feature at this outstanding region. Comment: accepted by A&A, now in press; 19 pages, 22 figures, 5 tables

Full text

Page 1

arXiv:0711.0249v3 [astro-ph] 15 Feb 2008
Astronomy & Astrophysics manuscript no. 6733
February 15, 2008
c ? ESO 2008
The enigma of GCIRS 3
Constraining the properties of the mid-infrared reference star of the central
parsec of the Milky Way with optical long-baseline interferometry.⋆
J.-U. Pott1,2, A. Eckart1, A. Glindemann2, R. Sch¨ odel1, T. Viehmann1, and M. Robberto3
1I. Physikalisches Institut, University of Cologne, Z¨ ulpicher Str. 77, D-50937 K¨ oln, Germany
e-mail: pott@ph1.uni-koeln.de
2European Southern Observatory (ESO), Karl-Schwarzschildstr. 2, D-85748 Garching bei M¨ unchen, Germany
3Space Telescope Science Institute, 3700 San Martin Drive, Baltimore, MD 21218, USA
Received ¡date¿ / Accepted ¡date¿
ABSTRACT
Context. GCIRS 3 is the most prominent MIR-source in the central parsec of the Galaxy. NIR spectroscopy has failed to solve the
enigma of its nature. The properties and peculiarities of extreme individual objects in the central stellar cluster contribute to our
knowledge of star and dust formation close to a supermassive black hole.
Aims. We initiated an unprecedented interferometric experiment to understand the nature of GCIRS 3, where we investigate its
properties as a spectroscopic and interferometric reference star at 10 µm.
Methods. VLT/VISIR imaging separates a compact source from diffuse, surrounding emission. The VLTI/MIDI instrument was
used to measure spectroscopically resolved visibility moduli at an angular resolution of ∼ 10 mas of that compact 10 µm source, still
unresolved by a single VLT. Recent NIR/MIR photometry data were added to enable simple SED- and full radiative transfer-modeling
of the data.
Results. The luminosity and size estimates show that IRS 3 is probably a cool carbon star enshrouded by a complex dust distribution.
Blackbody temperatures werederived. Thecoinciding interpretationof singletelescope andinterferometricdataconfirmdust emission
from several different spatial scales. The interferometric data resolve the inner rim of dust formation. Despite observed deep silicate
absorption towards GCIRS 3, we favor a carbon-rich circumstellar dust shell. The silicate absorption most probably takes place in
the outer diffuse dust, which is mostly ignored by MIDI measurements, but very observable in complementary VLT/VISIR data. This
indicates physically and chemically distinct conditions of the local dust, changing with the distance to GCIRS 3.
Conclusions. We have demonstrated that optical long baseline interferometry at infrared wavelengths is an indispensable tool for
investigating sources at the Galactic center. Our findings suggest further studies of the composition of interstellar dust and the shape
of the 10 µm silicate feature in this extraordinary region.
Key words. Galaxy: center – AGB: dust shells – Techniques: interferometric
1. Introduction
At a distance of ∼7.6 kpc (Eisenhauer et al. 2005), the center
of the Milky Way is by far the closest center of a large spiral
galaxy. Its astrophysical properties can be studied on a unique
angular scale of ∼ 40 mpc/arcsec, which is two orders of mag-
nitude smaller than the angular scale at the nucleus of M31,
the next comparablegalaxy(McConnachie et al. 2005). Star for-
mation and the kinematics of the central stellar cluster can be
studied in the region of direct influence of the supermassive
black hole (SMBH) at the dynamic center of the Milky Way
(Eckart & Genzel 1996; Ghez et al. 2000; Sch¨ odel et al. 2002).
The active history of star formation, despite the tidal forces
of the SMBH, is manifested in the existence of numerous mas-
sive, young stars in the central cluster (Krabbe et al. 1995;
Genzel et al. 2000; Eckart et al. 2004; Moultaka et al. 2004).
The 7 most luminous (L > 105.75L⊙), moderately hot blue su-
Send offprint requests to: J.-U. Pott, now at W.M. Keck Observatory:
jpott@keck.hawaii.edu
⋆Based on observations obtained at the European Southern
Observatory, Paranal, Chile (programs 073.B-0249, 075.B-0113,
077.B-0028).
pergiants (T < 104.5K) provide about 50% of the flux ionizing
the region (Blum et al. 1995; Krabbe et al. 1995; Najarro et al.
1997). Schr¨ oder et al. (2003) have shown that a few tens of
carbon-rich supergiants can produce about 50% of the mass loss
of a large stellar sample with solar neighborhood characteris-
tics. Thus, similar to the ionizing flux, the mass loss and dust-
formation properties of a stellar cluster can be dominated by
a few prominent stars. These facts underline the importance of
studying individual extreme objects like GCIRS 31to under-
stand properties of the entire surrounding stellar environment.
Therecentadventofmidinfrared2(MIR)instrumentson8m
class telescopes enables us to study in detail the thermal dust at
the Galactic center (GC) at unprecedented angular resolution.
The investigationof the circum- and interstellar dust distribution
at the GC uncovers stellar mass loss, zones of wind interaction,
formation history, evolution, and kinematics.
Photometric and spectral properties of dusty stars at the
GC have recently been published by Moultaka et al. (2004) and
Viehmann et al. (2005, 2006). Despite an average optical extinc-
1in the following IRS 3
23-20 µm covering the atmospheric L, M, N, and Q−windows

Page 2

2J.-U. Pott et al.: The enigma of GCIRS 3
Fig.1. Flux-calibrated Lucy-Richardson deconvolved image at
8.59 µm after restoration with a 250 mas Gaussian beam. The
pixel scale is 75 mas per pixel. The logarithmic contours levels
are 1.6n·7 mJy. The flux scale is given on top. IRS 3 and Sgr
A* are highlighted. Details of the data reduction are given in
Sch¨ odel et al. (2007b).
tion of AV ∼25 (Scoville et al. 2003; Viehmann et al. 2005),
near-infrared spectroscopy and imaging reveal the nature of the
underlyingdust-embeddedstars in most cases, since the dust ex-
tinction is wavelength-dependentand decreases from the optical
towards longer wavelengths.
This article focuses on the most prominent of the MIR
bright dusty sources, IRS 3, the embedded stellar source, which
still eludes any spectral classification (Paumard et al. 2006;
Tanner et al. 2006). It is located within the central 20”. A re-
cent state-of-the-art narrow-band MIR image at 8.6 µm with an
angular resolution of only 250 mas is shown in Fig. 1. The ex-
tended and diffuse dust emission surrounding IRS 3 is visible at
unprecedented angular resolution in this image.
The NIR extinction studies reveal a spatial variation of only
∼10% (AK
gion (Scoville et al. 2003; Moultaka et al. 2004; Sch¨ odel et al.
2007a).
In contrast, narrow-band, N-band photometry and spec-
troscopic observations are interpreted to indicate a signifi-
cant amount of additional 10 µm silicate absorption along
the line of sight towards IRS 3 with respect to other parts
of the central 20” (Becklin et al. 1978; Roche & Aitken 1985;
Viehmann et al. 2006). The published intrinsic optical depths of
about τ9.8(IRS 3) ∼ 1, in addition to the average τ9.8(GC) ≈
3.5 (e.g. Roche & Aitken 1985), still underestimate the true
value due to source confusion problems. While our new high-
resolution VISIR data clearly indicate that more than 50%
of the N-flux is diffuse, extended emission at 0.3” resolu-
tion, one byproduct of the MIDI observations is the calibrated,
low-resolution spectra of the compact emission, which show
τ9.8(IRS 3) to be much larger than the value given above.
Rieke et al.(1978) foundthespectro-photometricMIR prop-
ertiesofIRS 3toresembleeitheryoungstarsorOH/IRstars.The
≈ 3) of the interstellar extinction over this re-
Fig.2.Part of the sciencedata takenon 2005-05-25.We overplot
the different reduction methods (coherent and incoherent fringe
averaging and static and dynamic detector apertures) to demon-
strate the different results. The labelling indicates the artificial
wavelength shifts successively applied to increase the readabil-
ity of the plot. The relative errors of each method is given. The
rightmost error bars present the merged, maximum error bars.
We based our analysis on these merged data.
latter interpretation is opposed by the lack of OH-maser emis-
sion. With an MIR color temperature of ∼ 400 K, the central
emission of IRS 3 was found to be (together with the nearby
GCIRS 7) the hottest and most compact of the sources dominat-
ing the thermal dust irradiation from the GC (Rieke et al. 1978;
Gezari et al. 1985; Smith et al. 1990). Extended dust emission
around IRS 3 interacting with external stellar winds has been
found in recent L− and M−band observations (Viehmann et al.
2005). While hot star hypotheses are given by some authors
(Krabbe et al. 1995; Tanner et al. 2003), the lack of ionizing gas
leads Roche & Aitken (1985) to the assumption of IRS 3 being
a cool dust-enshrouded star.
Within the past two years we collected a unique dataset of
optical long-baseline interferometric (OLBI) data of IRS 3 at
10 µm. These constitute the first successful OLBI observation
of an object within the central parsec of our galaxy, opening
the window to NIR/MIR GC observations at highest angular
resolution (Pott et al. 2005a). In this article we show that the
OLBI data strongly support the former hypothesis of a cool
dust-enshrouded star. Furthermore our results shed light on the
amount and spectral shape of the interstellar 10 µm extinction
towards IRS 3 at unprecedented angular resolution. Since MIDI
is a relatively new instrument, the achievable precision of vis-
ibility measurements of such distant, challenging targets is not
commonknowledge yet. In Sect. 2 we therefore describe the ob-
servations, the extensive data-reduction and calibration efforts,
and the evaluationof different data reductiontechniquesin some
detail to ease and stimulate similar experiments and the compar-
ison of theiroutcomes.Thenthe immediateobservationalresults
are given (Sect. 3), followed by a detailed discussion of the re-
sults in the astrophysical context (Sect. 4) and a summary of our
conclusions in Sect.5.
2. Observations and data reduction
In 2004 we started an observing campaign to study the bright-
est MIR-excess sources in the central parsec with MIDI at the
highest angular resolution available today. The MID-Infrared
interferometric instrument (MIDI) combines the light of two

Page 3

J.-U. Pott et al.: The enigma of GCIRS 33
Table 1. Observinglog of IRS 3. The wavelength dependent TF
has been estimated for each night on the basis of regularly con-
ducted calibrator measurements, typically about once per hour
(Sect. 2). The number of calibrators used per night is given, and
the applied parameters are given in Table 2.
BaselinePB
[m]
PA Airmass
[1]
Accuracy
[%] [deg E of N]
-
46
52
55
53
-
11
50
-
89
-
89
90
-
139
-
101
124
-
112
136
#1 night: 2004-07-07
U2-U3
U2-U3
U2-U3
U2-U3
#2 night: 2004-07-08
U2-U3
U2-U3
#3 night: 2005-05-25
U3-U4
#4 night: 2005-06-23
U3-U4
U3-U4
#5 night: 2005-06-27
U3-U4
#6 night: 2005-07-20
U3-U4
U3-U4
#7 night: 2005-08-23
U3-U4
U3-U4
5 calibrators
45.7
42.3
37.0
26.4
1.03
1.16
1.45
2.63
12
12
12
14
8 calibrators
44.9
43.5
1.38
1.11
19
14
7 calibrators
45.81.36 14
4 calibrators
45
46.8
1.39
1.33
23
16
4 calibrators
541.2911
7 calibrators
58.6
59.6
1.07
1.08
22
22
2 calibrators
62.5
54.9
1.00
1.25
20
21
Table 2. Parameters of the calibrators used for estimating the
transfer function.
Name Spectral
type
K3.5III
G5II
K0IIIb
G8III
M1III
K5III
K5III
K0III
G9III
K1IIIb
M4III
K1IIIb
K0II
G8IV
G6/G8III
G7III
K1III
K4III
Diameter
[mas]
4.42±0.03
3.25±0.02
5.34±0.03
2.50 ±0.01
2.79±0.04
2.62±0.01
2.28±0.01
3.48±0.02
2.16±0.01
4.00±0.03
3.43±0.04
3.76±0.03
2.49±0.01
2.07±0.01
2.33±0.02
2.08±0.02
3.18±0.02
3.45±0.02
Flux dens.
[Jy @12µm]
22.0
15.1
36.4
8.4
8.0
7.5
5.7
15.5
5.9
20.1
11.9
17.1
7.6
5.9
7.5
5.9
12.4
13.2
Nighta
HD107446
HD109379
HD123139
HD134505
HD142804
HD152820
HD160668
HD165135
HD169767
HD169916
HD173484
HD177716
HD178345
HD188512
HD192947
HD213009
HD218594
HD220704
4
1
3
4
7
3,4
3,5
1-7
3
4
5,6
2
5-7
1-3
1
5
3
7
aThe night in which the calibrator was used
8.2m unit telescopes of the ESO Paranal Observatory in Chile
(Leinert et al. 2003). We used the standard 0.5”×2” slit and dis-
persedthelightovertheentire N−band(8-13µm)with theprism
providing a spectral resolution of R ∼ 30. The first MIR fringes
of IRS 3 were recorded successfully on the night of 7 July 2004.
The whole dataset comprises 14 independentvisibility measure-
ments of IRS 3 with interlaced calibration measurements. The
projected baseline length (PB) and position angle (PA) of each
measurement are given in Table 1 for the beginningof the fringe
measurement. The last column gives the median relative uncer-
Fig.3. The influence of faulty background estimation on the fi-
nalaccuracyfordatafrom2005-05-25.Thestandardreductionis
plotted (solid line) with respective uncertainties (small errorbars
to the right of larger ones). It suggests a significant deviation
from a smooth, flat visibility spectrum (dotted line: weighted
quadratic fit over the silicate absorption feature). In particular
the visibility increase (solid line) at λ ≤ 11 µm towards the ab-
sorption center cannot be confirmed if we considerthe impact of
backgroundsubtraction errors (large error bars). To demonstrate
the flux dependence of the importance of background accuracy,
we overplotted the total flux spectrum (in Jy).
tainty obtained for the calibrated visibility outside the low-flux
region of the silicate absorption.
2.1. Interferometric calibrator stars
In Table 2 we list the used calibrators with their main fea-
tures. The angular diameter ascertainments result from fitting
A9 and M model atmosphere spectral energy distribu-
tions (SEDs) to optical and NIR photometry. The chosen mod-
els (Kurucz 1992, 1994; Plez et al. 1992) adopt solar metallic-
ity. Details of the model fitting are given in van Boekel (2004,
chapter 5). The modeled 12µm flux densities listed in Table 2
are consistent within ?5% with the work of Cohen et al. (1999),
who present a list of absolutely calibrated infrared spectra.
Furthermore, none of those calibrators shows an MIR-excess,
defined as having a measured 12 µm flux density more than 3 σ
above the synthetic spectra fitted to the optical-NIR data. Such
an MIR-excess would indicate the existence of (extended) dust
shells. Dust shells, which can be expected to exist around K-M
giants,radiate thestellar fluxat longerwavelengthsand decrease
the visibility amplitudes of the entire object; i.e., they deteri-
orate the calibrator properties. The absence of an MIR-excess
also makes all visibility calibrators usable as potential photo-
metric calibrators, down to the 5% uncertainty, which is often
not reached due to atmosphericand instrumental variability.The
photometric variability and the fitting uncertainty of the angular
diameter of all the calibrationstars in Table 2 affect the accuracy
of the derived transfer functions by less then 1% at all VLTI
baselines (≤ 200 m).
2.2. Calibration and absolute accuracy
The immediate measurand, the raw visibility Vraw, is the ratio of
measured correlated (FC) and total flux (FT). The final quantity

Page 4

4J.-U. Pott et al.: The enigma of GCIRS 3
Fig.4. Transfer functions in gray, derived from the individual
calibrator measurements of 2005-05-25. The error bars around
5% indicate the statistical error of the single data frames of one
measurement, which is slightly smaller than the absolute vari-
ation in the TF over the night. The mean TF, which we used
for this night, is overplotted in bold solid black style with the
standard deviation. The two remote TF (gray), highlighted with
a diamond symbol, were rejected from calculating the mean be-
cause of abnormal behavior relative to the average of the good
TF.
Fig.5. Different analyses of the TF, derived from calibrator
measurementson 2005-05-25.To increase the clarity of the plot,
the wavelengths of two overplots have been shifted, as indi-
cated by the respectively shifted labeling. Left error bars: av-
erage (mean & standard deviation) of the individual TF. The
two other data sets are averagedin normalizedspace (divisionof
the individualTF by their integral).Central: maximumdifferen-
tial fluctuation between adjacent spectral channels, presenting a
conservative maximum of the spectral shape error of TF. Right:
standard deviation of the normalization factor, representing the
orderofmagnitudeofthe variabilityoftransmissionbetweenthe
FCand FTmeasurement.
ofinterestis thecalibratedvisibility(Vcal),whichis computedby
dividing the raw visibility (Vraw) by the interferometric transfer
function (TF) of the observation. Propagation of errors relates
the uncertainties ∆ to each other:
Vcalib=Vraw
TF
→
?∆Vcalib
Vcalib
?2
?2
=
?∆Vraw
Vraw
?2
?2???????calibrator
+
?∆TF
TF
?2
(1)
TF =Vraw
Videal
?????calibrator
→
?∆TF
TF
=
?∆Vraw
Vraw
.
(2)
TheTF is derivedfromthe Vrawmeasurementofa calibratorstar
andits intrinsicvisibility(Eq.2).Theuncertaintyofthediameter
of the calibrator is usually too small to affect ∆TF (Table 2).
Thus ∆Vcalibsuffers twice from the accuracy of the estimation of
Vraw(Eq. 1). This central accuracy can be estimated by means
of the following considerations and investigations of calibrator
data sets.
All constant defects of Vraware corrected for by multiplica-
tion with TF and thus have only a minor influence on the final
accuracy; e.g., the time delay between the measurement of FC
and FTis only used a few minutes in the high-sensitivity mode
of MIDI. This reduces the variation of atmospheric transmission
in the thermal infrared, due to airmass difference and temporal
fluctuations, to such a level that a small and time-constant im-
pact on the quotient FC/FTcan be assumed. We provethis by the
followingtest, visualizedin Fig. 5. If theatmosphericandinstru-
mental flux transmission change randomly at a significant level
between the FCand FTmeasurements, the orders of magnitude
of Vrawand the resulting TF are affected and the absolute accu-
racy of the calibrated visibility is worse than the relative spectral
accuracy. To derive this spectral accuracy we linearly normalize
each individual TF of a night by dividing through the integral
over a fixed part of the N-band and estimate the maximum scat-
ter in each spectral channel.If this scatter is significantly smaller
than the scatter of the normalization factors, a strong transmis-
sion variability has been observed (assuming constant intrinsic
visibility). Figure 5 shows that this is not the case. Note in the
same figure the increase in the shape error around the atmo-
spheric ozone absorption band at 9.7 µm and at the borders of
the N-band, where the applicability of simple linear normaliza-
tion decreases.
Faulty and unequal background suppression during reduc-
tion of both data sets also does not vary on a significant level for
the bright calibratorstars. The influence of backgroundnoise for
faint science targets is discussed at the end of the section.
As a matter of fact, Leinert et al. (2004) state that the ac-
curacy of Vrawis dominated by the accuracy of the overlap be-
tween the interfering beams, which critically relies on the wave-
frontcorrectionsduringobservations.This is difficultto quantify
and thus cannot be easily incorporated into the TF. Usually the
quality of the beam overlap does not vary statistically during
one observation. This means that there is no mean overlap accu-
racy to correctly describe a single observation and that the TF,
if derived from one calibrator measurement, may not be appli-
cable to a subsequent science observation. We confirm this on
the basis of our dataset. A single observation of both FCand FT
consists of several frames that are averaged during data reduc-
tion. We estimate the standard deviation of the subsequent scans
and analyze the background in chopped photometry data to de-
rive a statistical uncertainty for each measurement of Vraw. The
resulting total statistical uncertaintyσ(Vraw) is similar for the in-
dividual calibrator observations, but it cannot fully explain the
larger scatter of individually estimated TF over the respective
observing night. Figure 4 demonstrates that the standard devia-
tionoverseveralcalibratorsis onlyas small as expectedfromthe
estimated uncertainty of the individual measurements (∼ 5%), if
outliers are rejected from the average. That means that the abso-
lute accuracy of the visibility measurement can change signifi-
cantly with each new pointing.
Furthermore, the overnight scatter of TF is significantly re-
duced in the 2005 data due to the more stable VLTI feeding by
the higher-orderAO-system M instead of the earlier tip-tilt-
only correction of the S units (Arsenault et al. 2003). This
also confirms the dominating influence of varying beam overlap

Page 5

J.-U. Pott et al.: The enigma of GCIRS 35
and flux concentration between different pointings on the visi-
bility accuracy.
Theabovegiventests andconsiderationsunderlinetheorigin
of the accuracyof the Vrawmeasurementandjustify the accuracy
of a single measurementonly being given by the TF statistics of
several calibrator measurements. We calculated the mean and
standard deviation of the different TF to quantify TF and ∆TF.
Note that the presented estimation of errors even holds for an
extendedcalibrator with visibilities significantly lower than one,
providedthat the diameter of the calibratoris knownat sufficient
precision. AlthoughVraw,calibratorwill systematically changeover
the night with changing projected baseline length, this effect is
annihilated by calculating TF (Eq. 2).
Since the number of calibrators per night is typically around
8, one TF strongly deviating from the others can influence the
mean and standard deviation significantly. To avoid this situa-
tion we did not use the median, but rejected the anomalous TF
manually (on average one per night, Fig. 4). This has the advan-
tage that the meanandstandarddeviationofthe resultingsample
better represents the spectral shape of the TF reducing the prob-
ability of artificial spectral features in the calibrated visibility
spectrum.
To complete Eq. 1, we have to know the relative uncer-
tainty of Vraw of the measured target. Qualitatively the origin
is similar, as discussed above for the calibrators, but the flux of
the science targets can be significantly smaller than the calibra-
tor fluxes. Since our calibrators cover a range of 5-35 Jy, we
searched for any flux dependence of ∆Vrawon the basis of our
data. We checked for every night whether on average the scatter
of the TF of the fainter calibrators is greater than the scatter of
the TF of brighter ones, but did not find any such flux depen-
dence. Thus we assume a flux-independent relative uncertainty
of Vrawresulting in the uncertainty of the calibrated visibility:
∆Vcalib
Vcalib
=
√2 ·∆TF
TF
.
(3)
This estimation might still underestimate the relative accuracy
of the final calibrated visibility of the science target, since IRS 3
was observed with an off-axis AO-guide star, which decreases
the AO performance with respect to the on-axis guiding on
the calibrators. We compared the PSF of the photometric data
of IRS 3 with the calibrator measurements and specified three
datasets with calibrator-likePSF. These are highlightedthrough-
out the analysis and provide the tightest constraints on our inter-
pretation of the data (cf. Sect. 2.4 & 3).
If the correlated flux drops significantly below 5 Jy and ap-
proaches the sensitivity limits, an increase in the relative un-
certainties will probably occur due to the increased influence
of noise; but outside the deep silicate absorption, the correlated
spectrum of IRS 3 is above this limit. Furthermore, imperfect
background subtraction has an increasing influence on the final
accuracy with decreasing source flux. To account for this, we
estimated the background level by reducing sky-frames without
the source. Typically the background randomly varies between
zero and a value close to Fbg, where Fbgis the sum of the mean
andstandarddeviationofall spectralchannelsofthebackground
frames. This leads to a maximum background induced error in-
terval of
Vraw,bg∈
?FC− FC,bg
FT
,
FC
FT− FT,bg
?
,
(4)
which dominates the flanks of the strong silicate absorption to-
wardsIRS 3.A putativeincreaseinvisibilitytowardstheabsorp-
tion center, which appears to be present after applying the stan-
dard reduction, cannot be verified after incorporating the back-
ground uncertainty (Fig. 3). Indeed, not all datasets show such
an increase after the standard reduction.
2.3. Data reduction
We used the data reduction package M+E provided by the
MIDI consortium3. This package offers two different methods
for reducing the data: incoherent averaging of the fringe power
in each spectral bin over several scans, and coherent averaging
of the single dispersed scans. In the N−band, the latter method
can provide the differential phase information in addition to the
visibility amplitude, if the atmospheric and instrumental delay
and dispersion have been removed properly. A more detailed
description of both methods and their realization in M+E
is given by Leinert et al. (2004) and Jaffe (2004), respectively.
Since noise always contributespositively to the powerspectrum,
it is not automatically reduced by averaging incoherently over
several scans. In contrast, the coherent integration reduces the
statistical noise of the fringe data by averaging, which makes
this method favorable at very low correlated flux (below about
1 Jy) and low SNR. Correlated fluxes down to 0.1 Jy could be
estimated by this method (Jaffe, private communication). Since
our MIDI data is usually well above this limit, we found consis-
tentresultsofbothreductionalgorithmsthroughoutthecomplete
dataset (Fig. 2).
The constantly changing baseline projection due to earth ro-
tation limits the integration times of the TC measurement. To
reduce the noise level, which is intrinsically high in the thermal
infrared, only the pixels with the highest SNR should be con-
sidered by the reduction algorithm. This is achieved by detec-
tor aperture masks, which can lead to reduced SNR if the cho-
sen aperture is too large, too small, or misplaced with respect
to the incident source photons. This effect is strongly enhanced
by beam distortions and motions that are only partially corrected
by the AO system, which holds especially for our data since we
had to lock the AO on an off-axis guide star 35” away from the
GC. Therefore we reduced the full dataset four times: with both
coherent and incoherent fringe averaging and by applying two
different apertures (or detector masks), a wide standard one with
fixed location and width and a narrower one that is chosen dy-
namically for the PSF of every observation to trace the beam
maximum and to have an optimized width (this is the narrow
mask in Fig. 2).
After calibrating the data and estimating its accuracy as
described in Sect. 2.2, the four spectra (Vcalib,1..4± ∆Vcalib,1..4)
were averaged. To estimate the accuracy conservatively, we se-
lecteda maximum∆Vcalibincludingall fouruncertaintyintervals
(Fig. 2).
Note that the apparent correlation between systematically
higher visibilities and the use of the narrower mask in this fig-
ure is not a general systematic effect, but depends on the data
set. Other data show an apparent correlation between visibil-
ity and averaging method, independent of the chosen mask.
Consequentlywe do notpreferonesingle reductionmethodover
anotherbutusethemergeddata.Furthermore,thedifferencesare
significant with respectto the intrinsicuncertainties,resultingin
clearly increased uncertainties of the merged data. The mean fi-
3See
and
Currently
http://www.strw.leidenuniv.nl/∼nevec/MIDI/.
the
links
ESO webpages forgeneralinformation
(http://www.eso.org/instruments/midi/).
datareductionpackage theisprovidedon:

Page 6

6J.-U. Pott et al.: The enigma of GCIRS 3
nal relative accuracyof each mergedscientific dataset is given in
Table 1.
2.4. Photometric Calibration
MIDI observations provide the astronomer with spectra of the
total flux of the target in addition to the visibility modulus.
We flux-calibrated the spectra on the basis of the regularly ob-
served interferometric calibrator stars, typically late type gi-
ants (Sect. 2.1). We fitted an airmass-dependent system re-
sponse to the calibrator measurements of each night to flux-
calibrate the science data (for a more detailed description see
van Boekel et al. 2005). Spectra that were obviously faulty were
not taken into account. Such a time-independent system re-
sponse model makes it possible to use all the good calibra-
tor measurements of the night, which is especially favorable if
the broad band spectrum of the science target is not known.
Furthermore, the airmass dependence can minimize the impact
of larger distances between the calibrators and the target. In
some nights only one star was closer than 10◦to the target be-
cause of scheduling requirements. On normal nights, incorpo-
rating airmass-dependence reduced the calibration uncertainties
by up to 5%. But then the model includes uncertainties due to
varying atmospheric transmission and the instrumental through-
put over the night, resulting in final photometric accuracies of
about 5-10%, which dominate the intrinsic uncertainties of the
used calibrator spectra.
The Gaussian detector masks (Sect.2.3) that are typicallyap-
plied do not affect the visibility calculation but only the photom-
etry. If the science target is not completely unresolved by the
single telescope PSF (Fig. 14) or if such a weighting mask is
not well-centered on the brightest pixels, the measured flux is
decreased.We reducedthephotometryseparatelywithoutapply-
inganymasktotakecareofanysuchbias.Furthermore,itturned
out that a lot of datasets on the target have at least one beam of
significantly lower quality than the other one. Since the calibra-
tor measurementsdo not show such a strong beam variation,this
effect is assumed to result from the use of an off-axis AO guide
star duringthe observations of IRS 3, decreasingthe accuracyof
the wave-frontcorrection(Sect.2.3). A manualselection of good
datasets facilitates a final photo-spectrometric accuracy of less
than 10%, which is decreased towards low fluxes due to the re-
maining background. The result is shown in the lower spectrum
in Fig. 10).
3. Results
In this section we present the direct results of our interferomet-
ric measurements. First the measured and calibrated data are
shown. Then in the following sections we discuss how the ob-
served spectra constrain the underlying brightness distribution.
This discussion outlines the average properties of IRS 3 in the
MIR at MIDI resolution.
3.1. Visibility moduli
The measured visibility moduli are given in Table 3 . Following
Sect. 2.2 we did not find within the uncertainties any deviation
from a smooth visibility slope over the full N−band. Since no
correlated fluxes have been measured at the center of the silicate
absorption, we give the mean visibility and its accuracy for two
adjacent wavelength intervals in Table 3. The fully calibrated
visibility spectra are shown in Fig. 6. Note the two outliers at the
Table 3. Measured mean visibility moduli below and above the
silicate absorption, centered at 9.8 µm.
Julian datePBa
[m]
PAa
Visibility
[deg E of N] 8-8.7 µm11.8-13 µm
#1 night: 2004-07-07
2453194.7
2453194.7
2453194.8
2453194.9
#2 night: 2004-07-08
2453195.5
2453195.7
#3 night: 2005-05-25
2453516.6
#4 night: 2005-06-23
2453545.5
2453545.5
#5 night: 2005-06-27
2453549.8
#6 night: 2005-07-20
2453572.5
2453572.7
#7 night: 2005-08-23
2453606.5
2453606.6
45.7
42.3
37.0
26.4
46
52
55
53
0.17 ± 0.05
0.20 ± 0.06
0.24 ± 0.07
0.47 ± 0.19
0.22 ± 0.06
0.22 ± 0.05
0.17 ± 0.03
0.10 ± 0.05
0.11 ± 0.04
0.13 ± 0.02
0.10 ± 0.04
0.08 ± 0.03
0.05 ± 0.02
0.05 ± 0.02
0.27 ± 0.04
0.28 ± 0.04
0.34 ± 0.04
0.53 ± 0.10
0.32 ± 0.06
0.26 ± 0.04
0.25 ± 0.04
0.18 ± 0.05
0.22 ± 0.04
0.22 ± 0.03
0.21 ± 0.05
0.18 ± 0.04
0.15 ± 0.04
0.12 ± 0.04
44.9
43.5
11
50
45.8 89
45
46.8
89
90
54 139
58.6
59.6
101
124
62.5
54.9
112
136
aPB and PA stand for projected baseline length and position angle
and characterize the interferometric resolution at the time of the obser-
vation.
bottom of the upper and the middle panels. With respect to their
projected baseline length, they show visibilities too low to be
consistent with the other data. This is most probably an artefact
due to bad beam overlap at those observing times.
3.2. Probing the circular symmetry in uv-space
Circular symmetry in uv-space implies circular symmetry of the
brightness distribution. To probe the variation in the measured
visibility with changing position angle (PA), we compared our
data at a fixed uv-radius, where most data were obtained. In
Fig. 7 we show the uv-coverageof the entire dataset, overplotted
with a ring of constant uv-radius(5 Mλ). Note that the definition
of uv-radius (= projected baselinelength/observing wavelength)
leads to radial lines in the figure and to the situation where along
the circle datapoints at different PA might be observed at differ-
ent wavelengths. The uv-radius gives the angular resolution and
should show similar visibilities at different PA in the case of cir-
cular symmetry. In Fig. 8 the mean visibilities at Ruv =5 Mλ
of each dataset are plotted. The overplotted horizontal indicates
that the full dataset still conforms to total circular symmetry.
Although some data points do not perfectly coincide with circu-
lar symmetry, it has to be remembered that the shown error bars
of single measurements may be underestimated, since their es-
timation relies on averaging several calibrator measurements of
the observing night, leading to an average uncertainty that may
be exceeded in individual cases.
On the other hand, the drawn horizontal shows that a slight
deviation from circular symmetry is possible. The values around
PA=(120±30◦) appear to lie on average below the values at
smaller PA. Possible reasons for such circular asymmetry are
discussed in Sect. 4.4.
But it has to be mentioned that the data points with the best
photometric quality (used to fit the horizontal in Fig. 8) have

Page 7

J.-U. Pott et al.: The enigma of GCIRS 37
Fig.6. Calibrated visibility spectra ordered following the inter-
ferometric resolution to show the dependence on the projected
baseline length (given to the right). Each curve is plotted in a
different style. For the sake of clarity we plotted only errorbars
for the solid curves, and the other curves have uncertainties on a
similar relative scale. The gray dotted lines show the used linear
interpolation over the deep silicate absorption, where we do not
have reliable data. Note the different scaling of the panels.
nearly identical visibilities at different PA. And a comparison
with the uv-coverage (Fig. 7) shows that the longer wavelength
spectral channels have been considered at those PA with a ten-
dency towards lower visibilities (Fig. 8), i.e. the possibly indi-
cated deviation from circular symmetry in Fig. 8 might in fact
derive from the slightly wavelength-dependent size of IRS 3.
We do not expect to observe wavelength-dependent sizes due
to line emission or absorption of a certain layer, since the vis-
ibility spectrum over the N−band is apparently free of spectral
line features within the uncertainties.
But a slight change in the shape and the size scale of the
brightness distribution of the dust can be expected due to typi-
cally lower temperatures of larger, outer dust layers dominating
at longer wavelengths. This interpretation can easily explain the
deviationsfromperfectcircularsymmetryinFig.8 andis further
backed up by a slight increase of size with wavelength indicated
by the wavelength dependent analysis of the data (Sect. 3.4.2).
Circular symmetry implies a vanishing differential phase,
which remains in the data after coherent averaging and calibra-
Fig.7. uv-coverageof all our observations. The overplotted ring
indicates a uv-radius of 5 Mλ. We have investigated the degree
of circular symmetry along this annulus (Fig. 8). North is up,
East to the left.
Fig.8. Measured visibilities at a uv-radiusof 5 Mλ. To guide the
reader’s eye we have overplotted a horizontal representing total
circular symmetry, which was fitted to the data of best photo-
metric quality in both beams (three thick error bars). If a linear
correlation with arbitrary slope is fitted to the full dataset, and a
negative slope shows a ∼ 10% smaller reduced χ2, which could
indicate asymmetry or a wavelength-dependentsize (Sect. 3.2).
tion. We did not find differential phases along the N−band spec-
trum larger than the remaining scatter of less than ± 5◦around
zero. Based on the circular symmetry found, we show the PA-
averaged data plotted over the uv-radius in the lower panel of
Fig. 9. Also the black solid line in both panels is the wavelength
average of all data and lies smoothly between data sets at 8.5 µ
and 12.5 µm (upper panel). This instead indicates a slight size
increase in the brightness distribution with wavelength than a
qualitativechange because the overalltrend of the visibility with
baseline length coding the brightness distribution is similar over
the N-band.
Naturally the merged data represent an average bright-
ness distribution at the cost of losing the slight wavelength-
dependencebutgainingadditionalandmoreaccuratedatapoints,
since we can use all spectral channels together.Note in the lower
panel in Fig. 9 that statistical data averaging at one uv-radius
normally reduces the errorbars with respect to the original data.
The wavelength-averaged data is used only in Sect. 3.4.1 to de-
rive an overall description of the underlying brightness distri-

Page 8

8J.-U. Pott et al.: The enigma of GCIRS 3
Fig.9. Azimuthal data average in both panels. Upper panel: er-
ror bars represent the data taken at 8.5 (gray)and 12.5 µm (black
& diamonds), respectively. In the lower panel, the same spec-
tral average (black solid line and error bars) is shown with all
measuredvisibilities (errorbars in light gray).The three datasets
with the best beam quality are highlighted in dark gray. To cal-
culate the average, we interpolated linearly over the silicate ab-
sorption where necessary.
bution, which itself is used as a basis for describing the wave-
length dependence of IRS 3 by fitting the average model to the
wavelength-dependent data (Sect. 3.4.2). The three datasets of
highest photometric quality are highlighted. This suggests that,
beyond ∼ 6 Mλ, the averaged uv-data (black solid line) indicate
visibilities, which are too low since they lie below the best data.
3.3. Photo-spectrometry
In Fig. 10, the flux-calibrated spectrum of IRS 3 is shown based
onthereductiondescribedinSect.2.4.Onlythedataofbestpho-
tometricqualityandAO correctionhavebeentakenintoaccount.
The three resulting data sets, observed in July 2004, May 2005,
and June 2005, do not show significant photometric variability
beyond the general uncertainties. This agrees with the study of
flux-variable sources in the GC by Ott et al. (1999), who find no
variability for IRS 3.
3.4. Morphological interpretation of the visibility data
In this section we explain the MIDI data by a model for the
brightness distribution, which is as simple as possible, but as
complex as needed. Although the results cannot compete with
Fig.10. Flux-calibrated and dereddened photometry. The up-
per spectrum is dereddened with τ9.8 = 7.2, and its error
bars indicate the wavelength intervals used to fit the temper-
ature. The middle spectrum is dereddened with τ9.8 = 3.3,
which corresponds to the standard average optical extinction of
AV=25 towards the central parsec, assuming the extinction law
by Moneti et al. (2001). The lower spectrum is the extinguished,
measured spectrum. The black solid lines show the extinguished
and dereddened χ2-minimized temperature fit of T=410 K.
Fig.11. Best-fit models of the brightness distribution with a sin-
gle component model; left panel shows a uniform disc, right
panel shows a Gaussian. The angular diameter is indicated.
detailed radiative transfer models, they summarize the average
order-of-magnitudeproperties in terms of morphological shape,
size, and flux. Such a heuristic model is therefore an impor-
tant check and starting point for furtheranalysis, discussion, and
interpretation of the data (Sect. 4). Thanks to the large multi-
wavelength data set, we get a non-trivial model of the dust dis-
tribution around IRS 3.
3.4.1. General structure
Following the results of Sect. 3.2, we assume a circularly
symmetric brightness distribution, at first of a wavelength-
independent size, to derive the general shape of the bright-
ness distribution. The full dataset is shown in a radial uv-plot
in Fig. 9. The two simplest, but often applicable, circularly
symmetric brightness distributions are a uniform disc and a
Gaussian. Figure 11 demonstrates that the visibility moduli can-
not be described by such a simple single component.Despite the
conservative error estimates, neither model can reproduce the
data even approximately.

Page 9

J.-U. Pott et al.: The enigma of GCIRS 39
Fig.12.Best-fit models of the brightnessdistributionwith a two-
component model. Up: two uniform disc components; Bottom:
two Gaussian components. The error bars are the azimuthally
averageddata (Fig. 9, Sect. 3.4.1). In addition,the probedspatial
scales are indicatedby verticallines.To the rightofeachof these
lines, the visibility of a Gaussian component of the indicated
FWHM would contribute less than 10% of its flux. The dashed
line stems froma modeladdinga thirdGaussian of arbitrarysize
larger than 80 mas and illustrates that up to 40% of the total flux
could have been resolved out by the interferometer.
In contrast, two superposed components of different sizes
andfluxratiosaresufficienttomodelthedata(Fig.12).Bothtwo
uniform discs and two Gaussians fit the error bars satisfactorily,
but the Gaussians are closer to the measured data: the reduced
best-fitχ2is abouttentimessmallerforthetwo-Gaussianmodel.
Furthermore, the data reduction indicates that the visibility at
lowuv-radii(∼ 2.5Mλ)is probablyoverestimatedandshouldbe
expected to lie in the lower half of the indicated error bars. This
would favor a Gaussian shape for the larger component. But as
shown by the dashed line in the lower panel of Fig. 12, the inter-
mediate 50 mas spatial scales are relatively loosely constrained
to contribute between 20% and 70% of the total flux detected by
MIDI. Up to 40% of the total flux could have been resolved out
by the interferometerdueto the lack of shorterbaseline informa-
tion. Since with the current visibility dataset there are no means
to further constrain the outer flux contributions, we stay with the
two component model representing the data and outline where
further analysis suggest that indeed some source flux could have
been resolved out (Sect. 4.3). However, the measured correlated
flux at the higher spatial frequencies (the smaller component of
the model), which is most important for the further analysis and
contributes about one third of the total flux, are not significantly
influenced by this uncertainty.
At longer uv-radii the smaller component dominates. A
closer coincidence with the data suggests a Gaussian shape for
the smaller component, too. But the final distinction between
Gaussian and disc shape of the smaller component requires ad-
ditional data at longer baselines. In the case of a uniform disc,
the visibility would increase again around ∼ 10 Mλ as indi-
cated by the overplotted model in Fig 12 (upper panel). We in-
vestigated the uv-space ≥ 12 Mλ with the longest UT-baseline
(UT1-4: 130m) without fringe detection. This might support a
Gaussian shape for the smaller component. But it is also possi-
ble that the visibility increase, indicative of a uniformdisc shape
of the inner component, was too small to be detected with these
long baselines. However, this non-detection at the long base-
line supports the primary finding of our source modeling that
the spatial scales of the smaller component are also resolved by
the VLTI leading to significant constraints on the physical inter-
pretation of the data (Sect. 4).
At this point there is no indication that the two Gaussian
components used to represent the data must stand for two phys-
ically distinct entities, such as two dust layers of different radii.
At the moment they simply appear convenientfor describing the
data, and the larger Gaussian can be seen as representing the
wings of the observed brightness distribution. In the later sec-
tions,physicalparameterssuchas flux-calibratedspectraandde-
rived temperatures support the idea of several physical compo-
nents, but a more complicatedly shaped single structure cannot
be completely excluded. In general the situation is similar to an-
alyzing the uv-dataof radio-interferometricobservations(e.g. of
a quasar radio jet in Pott et al. 2005b), where Gaussian com-
ponents are used to represent mean properties of the observed
structure, such as size, flux, and location. To avoid confusion we
simply keep referringto the two Gaussians as being two compo-
nents. This understanding is further supported by the later anal-
ysis.
3.4.2. Wavelength dependence
The spectroscopically resolved MIDI data allow the investiga-
tion of wavelengthdependenceof the observedbrightnessdistri-
bution. We base this analysis on the two-Gaussian model of the
previous section, since it fits the wavelength-averaged data set
perfectly. A model-independentanalysis of the observed bright-
ness distribution by Fourier-transforming the data (based on
the fundamental van Cittert-Zernike theorem of interferometry,
e.g. Labeyrie et al. 2006) confirms of our model approach, but
the limited spatial frequency coverage hampers further model-
independent conclusions from such an analysis. We describe the
existing wavelength-dependence of the data in terms of chang-
ing sizes and relative flux contributions of the two Gaussians,
keeping the comment at the end of the previous section in
mind. Furthermore, the change of the fit parameters with wave-
length turns out to be reasonably smooth (Fig. 13). This finding
backs the application of the wavelength-independent model of
Sect. 3.4.1 as a basis for fitting the wavelength dependence.
We fit intensity and size of two superposed Gaussians to the
data (Fig. 13). We bin the data to a 0.5 µm sampling,using error-
weightedvisibility averages.The unreliabledata aroundthe cen-
ter of the silicate absorption are interpolated using a χ2-fit of a
quadratic curve to the spectral channels of good SNR. In Fig. 6

Page 10

10J.-U. Pott et al.: The enigma of GCIRS 3
Fig.13. Wavelength-dependent model consisting of two
Gaussians. The trends are discussed in the text. Around 10 µm,
the Gaussians are fitted to the interpolated data points. Is/lde-
note the intensities (in arbitrary units) and θs/lthe FWHM of the
smaller and the larger components, respectively. The flux ratio
calculates as (Isθ2
of the inner component are analyzed by RT calculations. The
Gaussian parameters have been fitted to the directly measured
visibilities in the spatial frequency domain.
s)/(Ilθ2
l). In Sect. 4.5 the physical properties
the individualregionsof reliable data are indicatedby errorbars.
The interpolated data are used in the central wavelength interval
only, which is devoid of error bars. The respective dependence
of the visibilities on the position angle is more visible in Fig. 8.
In Fig. 13 the resulting best-fit parameters are shown. Using
χ2-minimization, the uncertainties of the fit are nearly the same
as the scatter of the data around the overplotted linear correla-
tions. A slightly larger systematic uncertainty might be intro-
duced by the estimated errors of the individual data points. To
address this, we re-fit the data with increased weighting of the
three datasets of best photometric quality. The overall trends
are similar, but the size of the larger component in the two-
componentmodelapproachmaybeunderestimatedinFig.13by
5-10 mas, which indicates more flux on low spatial frequencies.
The FWHM of the smaller component, θs, may show a slight
size-increase with wavelength by about 3 mas, and the flux ratio
increases towards the smaller component(Fs/Fl∼ 0.7). This in-
creased flux ratio can be understood by lower photometric qual-
ity, typically decreasing the average visibility due to an imper-
fect beam overlap. But a decreased visibility means increased
relative brightness of the larger component. Thus, probably a
few of the data sets of lower photometric quality systematically
show visibilities that are too small, thus artificially implying
structures that are less concentrated than the real ones (see the
bottom curves of the upper and central panel of Fig. 6, which
Fig.15. Zoom of the lower spectrum in Fig. 10 into the wave-
length interval of the 11.3 µm feature of SiC. The measured data
is plotted in (dashed)gray,and the black solid line represents the
best-fit extinguished blackbody SED of T = 410 K.
apparently do not follow the visibility trend of the other curves
with respect to the baseline lengths).
The most intriguing result is that the smaller component (θs)
shows a roughly constant FWHM of about 18 mas, while the
larger shows a significant linear size increase with wavelength.
This might be an indication that we directly resolve the inner
zoneof dust formationat all wavelengths,whichhas a fixed size,
and that the dust shell might be carbon-rich (Ivezic & Elitzur
1996b).
Furthermore,the relative flux contributionof the larger com-
ponent increases with wavelength, suggesting that the larger
component represents the outer, cooler dust around the central
object. Again this is a more qualitative statement. The second
component stands for the more extended flux of IRS 3, its prop-
erties averageoverthe conditionsat the outerparts, but a smooth
transition between the inner and outer regions around the star
cannot be excluded. Temperatures and luminosities of dust on
the size scales of both componentsare derivedin Sect. 4.3 based
purely on the interferometric data.
4. Discussion
4.1. Interstellar absorption and the composition of the
circumstellar absorbing dust
Following the most recent published results, we assume a spec-
tral profile of the interstellar absorption towards the GC, as
published by Moneti et al. (2001), and an average visual ex-
tinction of AV = 25 towards the GC (Scoville et al. 2003;
Viehmann et al. 2005). Moneti et al. (2001) incorporated into
their model that the mean interstellar dust towards sources in
the central 2 pc of the Galaxy shows a relatively stronger sili-
cate absorption than in the solar neighborhood(Roche & Aitken
1985).
In the NIR no strong local increase in the interstellar extinc-
tion towards the region of IRS 3 is found (e.g. most recently
confirmed by Sch¨ odel et al. 2007a). However, several authors
claim additional silicate absorption in the N-band only along
the line of sight to IRS 3, possibly intrinsic to the source (e.g.
Roche & Aitken 1985; Viehmann et al. 2006).
A first glimpse of the probable location of this additional
extinction is given by our high-resolution single-telescope N-
band imaging with the new VLT/VISIR instrument. In Fig. 14
the complete emission of IRS 3 (left panel), unresolved by ear-
lier imaging, is clearly resolved into a diffuse and a compact
component(central and right panels, respectively).Althoughthe
total diffuse flux is even larger than the compact flux, its surface

Page 11

J.-U. Pott et al.: The enigma of GCIRS 311
Fig.14. Our VISIR 8.6 µm imaging data of a 4” field of view centered on IRS 3. Left: A zoom of Fig. 1. Middle: The diffuse
emission with subtracted PSF. Right: The flux-calibrated PSF, i.e. the unresolved emission seen by MIDI. The data reduction is
detailed in (Sch¨ odel et al. 2007b). All images have the same color coding as indicated at the top and the same contour lines for
ease of comparison. The logarithmic contours levels are 1.6n·7 mJy. The integrated flux of the different components is given in the
figures. Note that the diffuse flux peaks below 40 mJy/pix.
brightness is very low and hidden in the noise of all MIDI data
due to the shorterintegrationtimes. Thus most of the diffuse flux
is not included in the single-telescope MIDI photometry (FT,λ),
although it is observed at similar spatial resolution to the VISIR
data. Therefore the flux-calibrated MIDI FTspectrum (Fig. 10)
fits the unresolved flux shown in the right panel. The two com-
ponents, fitted to the interferometricdata (including the possible
fraction of fully resolved flux; Sect. 3.4) and discussed in the
following sections, together make up the unresolved emission
in the right panel of Fig. 14, since the single-telescope observa-
tions provide an angular resolution of about 250 mas. To clarify
the situation, we speak of local, interstellar silicate absorption
in addition to the GC average, if the absorbing silicate is located
in the dust, which radiates the diffuse emission that is directly
visible only in the VISIR data. In contrast, absorption in the in-
ner dust components, resolved by the MIDI flux and visibility
estimates, is labeled as intrinsic, circumstellar absorption.
Our MIDI spectro-photometry confirms the existence of a
broad 9.8 µm silicate absorption feature remaining in the data
after correction for standard GC extinction. We show the mea-
sured spectrum in Fig. 10 (lower curve). For the dereddening of
the spectrum, we used the µCep emission profile of the silicate
feature, as realized in the extinction law by Moneti et al. (2001)
for lines of sight to the GC. Several authors state that this pro-
file matches both the local ISM absorption and the GC interstel-
lar silicate absorption profile best (although at different relative
optical depths; Roche & Aitken 1984, 1985; Chiar & Tielens
2006).
Our IRS 3 spectrum of a spectral resolution of R = 30 shows
this coincidence perfectly. We reddened a single blackbody of
variable temperature with an absorption spectrum of the nor-
malized shape of the silicate feature seen in emission towards
µCep (see Moneti et al. 2001) and a variable optical depth τ9.8.
The shapes of the observed and dereddened spectra coincide
with the spectra shown by Roche & Aitken (1985) (Fig. 10).
The best χ2-fitted parameters are a blackbody temperature of
T = (410 ± 30) and τ9.8 = (7 ± 0.5). This temperature re-
sembles other N−band measurements (e.g. Gezari et al. 1985,
found a color-temperature of about 400 K), but is below the
800K derivedfrom K− and L−banddata (Moultaka et al. 2004).
This indicates that a single blackbody may not be appropriate
for describing the complete NIR-MIR SED. Although the hotter
component,dominating the K− and L−band, cannot be resolved
against the cooler outer dust shell by the single telescope spec-
trum or image at 10 µm with additional SED information, our
interferometric 10 µm data alone can resolve it (Sect. 4.3).
Since the center of the absorption feature is hidden in the
noise of the background subtraction, a certain level of uncer-
tainty remains in the estimation of τ9.8, but the spectrophotomet-
ric quality of the wings is good enough to exclude τ9.8 ≤ 6.5.
Assuming AV = 25, this means AV/τ9.8 ≤ 4. This is a re-
markable result, since it doubles the silicate MIR optical depth
towards IRS 3 with respect to the average of the GC region
((AV/τ9.8)GC ≈ 8-10; Roche & Aitken 1985), which itself is
twice as deep as AV/τ9.8 in the solar neighborhood. This is
shown by the middle spectrum in Fig. 10, which is the mea-
sured spectrum corrected for standard GC values of extinction
(AV= 25 and (AV/τ9.8)GC). The remaining silicate absorption is
obvious. The aforementioned authors quantify for the first time
an extra τ9.8≈ 0.8 for IRS 3 in addition to already enhanced the
GC-average. Although the spectral resolution of both datasets is
comparable,the spatial resolution of the MIDI photometrydata4
is increased by at least an order of magnitude.
In addition, at 11.3 µm a significant drop in the data below
thefitis suggested(Fig.15),evenonthelogarithmicscaleshown
in Fig. 10. This further absorption feature, in addition to the
dominating broad silicate absorption, can be attributed to SiC,
which peaks around 11.3 µm and has a much narrower spectral
width than the interstellar silicate feature. If we exclude the data
fromthewavelengthregionaround11.3µmfortheχ2minimiza-
tion, the discrepancy between data and fit around the center of
the SiC feature becomes even stronger, although the fitted tem-
4which is the VLT 10 µm resolution of ∼ 250 mas, not the interfero-
metric resolution

Page 12

12J.-U. Pott et al.: The enigma of GCIRS 3
perature and τ9.8remain constant in the given interval of uncer-
tainties. This further supports the existence of a SiC absorption
feature towards IRS 3, but the sampling of the applied extinction
law, which does not contain the SiC feature, is with 0.5 µm at
MIR wavelengths not high enough to properly sample the SiC
feature. The kink of the blackbody fit at 11.5 µm is probably
an artificial feature. Thus a definitive answer regarding the exis-
tence of absorbing SiC cannot be given.
Arguments against an intrinsic silicate absorption in the im-
mediate circumstellar dusty environment of IRS 3, which could
be evoked by a deep O-rich dust shell, are:
– Most silicate-rich dust shells show the silicate feature in
emission. Similarly the SiC feature is typically found in
emission in the dust shells of evolved stars. Since our esti-
mated optical depth τ9.8is already very deep, an even larger
amount of absorbing dust would be necessary to overcome
the circumstellar emission and result in such strong features
as observed.
– The spectral shape of the observed silicate absorption coin-
cides perfectly with the interstellar absorption features. No
indications of circumstellar crystalline silicates are found,
although spectroscopic data with higher SNR and spectral
resolution covering the full N−band are needed to investi-
gate the spectral shape in more detail.
Thus a circumstellar dust shell free of a significant amount
of silicates appears to be a reasonable assumption for the imme-
diate environment of IRS 3. This is confirmed by our visibility
data, which do not show any spectral feature coinciding with the
broad shape of the 9.8-silicate feature at any baseline length.
Such a lack of an intrinsic, circumstellar silicate-rich dust
shell and the deep interstellar silicate absorption would favor the
brightIRS 3 to be the primarytarget forestimatingthe true spec-
tral shape of the interstellar absorption in the N−band towards
central GC sources at the high spatial and spectral resolution
now available at ground-based 8 m class telescopes.
4.2. Dust temperatures from the spectral energy distribution
The MIR regime is dominated by thermal dust emission. We
have shown in the previous section that the N−band spectrum
of IRS 3 can be fitted convincingly by a single reddened black-
body. But published studies of stellar dust shells show that the
full infrared photometric information is required to describe the
optical and physical properties of the shells. Accordingly, we
investigated the complete infrared wavelength range from 1.6-
20 µm, available at a spatial resolution sufficient for distinguish-
ing IRS 3 from other sources. The corresponding dereddened
SED is shown in Fig. 16. The bolometric flux ratio indicated in
the figure is better constrained by the MIDI visibility estimates
(Sect. 4.3).
A second star close to IRS 3 has recently been classified
as a Wolf-Rayet star (Paumard et al. 2006). It is unresolved in
the published medium-resolution NACO data (Viehmann et al.
2005), but we confirm this secondary star on high-resolution
NACO images, showing about 30% and less than 10% of the
IRS 3 H- and K-band fluxes, respectively. For our SED fit in
Fig.16weusedtheaccordinglyreducedpublishedH magnitude.
Since the companion is located about 120 mas east of IRS 3, its
MIR flux could also contribute to the MIDI data. But a signifi-
cant contribution should show up as a binary pattern in the visi-
bility data (Sect. 4.4), which was not observed. Thus we assume
negligible contamination of the flux of IRS 3 longward of 2 µm
by the WR-star, which is further supported by the bluer NIR
Fig.16. SED temperature fit (solid line) as the superposition
of two spatially unresolved blackbody SED (dashed lines). The
dotted line close to the cooler and presumably larger Tlcom-
ponent shows a blackbody SED of T
fitted to the N-band spectrum alone (Sect. 4.1). The possi-
ble bolometric flux ratio range is given in the plot. Data are
taken from Viehmann et al. (2005, H−, K−, L−, M−bands) and
Viehmann et al. (2006, Q−band) and corrected (see text). The
N−band data was taken from our MIDI observations at 8 and
13 micron, outside the silicate absorption feature. The data was
extinction-correctedfor AV=25.
= 410 K, which was
SED of the secondaryas measuredwith NACO and expected for
a hot WR.
To account for the uncertainties in the amount of interstel-
lar silicate absorption towards IRS 3 (Sect. 4.1), we used only
the MIDI fluxes at 8 and 13 µm outside the 9.8 µm absorp-
tion feature for the MIR-SED. All data were dereddened with
the Moneti extinction law scaled to AV=25. As for the N-band,
a broad interstellar silicate absorption feature is located in the
Q−band. At 20 µm, only narrow-bandphotometryinside the sil-
icate absorption was available. We dereddened this Q-band data
by an optical depth of τQ ∼ 3.5, which is derived from scaling
the interstellar extinction law to fit our measured enhanced τ9.8
towards IRS 3, exceeding the average GC values. At least the
plotted SED data for λ ≤ 13µm should be free of any significant
flux contribution of the diffuse VISIR component, because of its
low surface brightness (Sect. 4.1) and presumably cool temper-
ature. Only the VISIR Q-band data may contain a fraction of
the diffuse flux; although similar to the MIDI spectra, these Q-
band data do not reach the low noise level of the 8.6 µm VISIR
imaging data.
We successfully fit the extinction-corrected SED with two
blackbodyspectra. The lower temperature,which we attribute to
the outer component(Tl= (440±50)K), coincides with the sin-
gle blackbody temperature fitted to the N−band spectrum alone
(Sect.4.1). This single blackbody SED is overplotted in Fig. 16,
which shows immediately that for λ ≤ 5µm additional flux by
hotter dust is required to explain the measured infrared SED.
This is consistent with the findings of Moultaka et al. (2004),
who fitted a blackbody temperature of 800 K to their 2-4 µm
data. That hotter dust emission appears at shorter wavelengths
limitstheopticaldepthoftheoutercoolerdust,assumingthatthe
hotter dust is located inside the cooler dust (see also Sect. 4.3).
The calibration of the Q−band emission, apparently too faint to
fit the model, was very difficult and the deviations can be fully

Page 13

J.-U. Pott et al.: The enigma of GCIRS 313
Table 4. Temperatures from the N−band MIDI data based on
size and relative flux estimates of the two-Gaussian model and
τ = 0.5 of the outer dust.
Properties Inner (20 mas)
8 µm
35
18
(610±180)
(830±120)
Outer (50 mas)
8 µm
21..55
40
(460±100)
(480..630±70)
(400..580±90)
13 µm
32
18
13 µm
25..66
55
Fa[Jy]
θa[mas]
TFratio[K]
Tsphere, 8[K]
Tsphere, 13[K](1020±210)
aThe individual component fluxes and sizes have uncertainties of
about 15% and 10%, respectively, within the model. The additional un-
certainty range of the flux contribution of the larger component due to
the lack of data at short spatial frequencies (Sect. 3.4.1) is indicated
with dots.
attributed to calibration errors (Viehmann,priv. communication)
including the uncertain amount of interstellar silicate extinction
in that band.
We can state, summarizing these considerations, that no fur-
ther blackbody component is needed to model the full near-
and mid-infrared SED. In particular this excludes any signifi-
cant stellar contribution to the NIR fluxes of the SED, emerging
from IRS 3 itself or a second star inside the PSF. This leads
to two possible interpretations: either the inner and hotter dust
component is optically thick at NIR wavelengths avoiding any
direct detection of stellar light or the enshrouded star has a very
hotcontinuumemission.Theabsenceofanystellarphotospheric
lines in NIR spectra of IRS 3 (Tanner et al. 2006) supports the
optical thickness of the circumstellar dust at these wavelengths.
4.3. Dust temperatures from the interferometric data
In contrast to the previous section, here we present a derivation
of temperatures from the spatially resolved MIDI observations
in MIR as a further step in interpreting this (and similar) inter-
ferometric data.
The simplest morphological interpretation of the data con-
sists of two circular symmetric Gaussian components enshroud-
ing the same central object (Sect. 3.4). One-dimensional radia-
tive transfer calculations confirm bell-shaped brightness distri-
butions of circumstellar dust shells (Ivezic & Elitzur 1996b).
That we can distinguish at least two dust components suggests
theouteroneis opticallythinintherespectivewavelengthsrange
and physically separated. That is, the observed total flux consti-
tutes of the sum of the flux of both components, adjusted by the
optical depth τ. From the observed total fluxes and flux ratios,
we calculate the component fluxes as
Fd
tot,λ= Fs,λ+ Fl,λ= e−τλF′
= (Rλ+ 1)Fl,λ
s,λ+ (1 − e−τλ)F′
l,λ
(5)
where Fd
between the inner and the outer dust shells as observed, and F′
aretheintrinsicfluxdensitiescorrectedfortheopticaldepth. All
measurable quantities in Eq. 5 are supposed to be wavelength-
dependent. In other sections we present strong indications that
the absorbing silicate is not located in the inner circumstellar
dust (Sect. 4.2 & 4.5). Nevertheless, here we confine the calcu-
lation to the edges of the N−band outside the silicate feature to
minimize the possible corruption of the results by faulty correc-
tion for the interstellar extinction.
totis the total dereddenedflux(Sect. 4.1), R the fluxratio
s,l
In Table 4 the component fluxes and FWHM-sizes for the
two-Gaussian model of Sect. 3.4 and the temperatures derived
here are given: Tsphereis the brightness temperature of a spher-
ical blackbody of radius θ/2 at GC distance emitting the ob-
served flux, and TFratiothe blackbody color temperature derived
solely from the 8-13 µm color of each component. We corrected
both component fluxes for an optical depth of the outer shell of
τ = 0.5. Since typically the optical depths at 8 and 13 µm are
comparable, the color temperature does not change with τ.
Furthermore,RT calculations of spherical dust shells as used
in Sect. 4.5 indicate that a black body spectrum of TFratiooften
convincinglyfitstheMIRtotalspectrumofadustembeddedstar,
particularly well for carbon-rich dust shells. But the respective
Tsphereof such models, calculated again with a Gaussian bright-
ness distribution approximation, is often significantly higher
than TFratioof the same RT model. This is in line with the find-
ings presented in Table 4 showing that the temperatures derived
here only give general trends and temperature ranges without
imposing tight constraints on the physical temperature.
We find:
– A reasonable increase in TFratioof the inner component with
respect to the outer one. This coincides perfectly with the
analysis of the infrared SED in the previous section. The
MIR-interferometric data alone show the inner, hotter dust
component in contrast to spatially unresolved N−band pho-
tometry.But the physical properties of the inner dust are bet-
ter confined by the complete IR-SED and by more detailed
radiative transfer calculations (Sect.4.5)
– Physically reasonable temperatures can only be obtained
with an optical depth of τ8/13∼ 0.5 due to the outer dust in
addition to the known interstellar extinction AV= 25. This
backs the large τ9.8discussed earlier found towards IRS 3
but suggests at the same time that this additional silicate ab-
sorption does not occur in the innermost circumstellar dust.
– Constraining the optical depth is also important for deriv-
ing the intrinsic luminosity of ∼ 5 · 104L⊙only from the
inner component fluxes. This luminosity is slightly higher
than earlier estimates (Becklin et al. 1978) that were based
solely on spatially unresolved N-band photometry. Thus it
can now be excluded that any additional source outside the
central 20 mas significantly contributes to the flux and heat-
ing of the dust around IRS 3, a finding confirmed in the next
section.
4.4. Circular symmetry
In
Paumard et al. (2006) classify the star IRS 3E, only 120 mas
east of IRS 3, as a Wolf-Rayet star of type WC 5/6. Since
Wolf-Rayet stars of this spectral type are known to be (often
strong) dust formers, it is possible that IRS 3E is still visible
in the N−band. Additionally its spectral classification as a
carbon-rich WC-star does not conflict with our interpretation of
the spectral data in terms of a lack of silicate emission.
The visibility pattern of such a binary system with 120 mas
separation and east-west orientation is shown in Fig. 17 for sev-
eral Ruv. The shown pattern has been calculated for two stars of
equal brightness, and both are individually unresolvedby the in-
terferometer. IRS 3 is resolved by MIDI, which would decrease
the amplitude of the variation shown in Fig. 17, but the variation
per PA is defined by the binary separation only; i.e., the imprint
of a 120 mas binary system should show several ripples over
180◦rotation, which cannot be confirmed with our data (Fig. 8).
their spectroscopicsurveyofthecentral cluster,

Page 14

14J.-U. Pott et al.: The enigma of GCIRS 3
Fig.17. Visibility patterns for a 120 mas east-west binary at sev-
eral uv-radii. The individual components are not resolved.
Furthermore, we did not observe neither photometric variability
beyond the calibration uncertainties (Sect. 3.3) or any non-zero
differential phases (Sect. 3.2).
The simplest deviation from circular symmetry is an ellip-
tical morphology or, more generally, a brightness distribution
of different apparent extension in orthogonal directions. The
analysis of Fig. 8 allows such an interpretation: Towards PA =
120◦the measured visibilities appear to be slightly lower than
in orthogonal direction, necessitating a larger extension of the
brightness distribution in this direction. Such a lateral contrac-
tion could be evoked by the strong stellar wind of a nearby star,
but more probably this deviation from circular symmetry can be
attributed to a wavelength dependent size (cf. Sect. 3.2).
4.5. Radiative transfer models
The Gaussian model analysis of the data demonstrated that the
interferometer resolves the inner dust around IRS 3. Thus we
apply a physically self-consistent radiative transfer (RT) model
here to further investigate the physical and chemical properties
of the inner dust around IRS 3 addressed before by the smaller
of the two Gaussians.
Because of the circular symmetry of the source, we use the
one-dimensional code D (Ivezic et al. 1999). Since IRS 3
appears to be very luminous, isolated, and surrounded by a lot
of dust, it is the most reasonable to assume that IRS 3 is a post
main-sequence star with strong stellar winds and massive dust
formation.We followed a heuristic approach and calculated four
distinct scenarios spanning the space of possible stellar param-
eters: hot and cold, realized by stellar effective temperatures of
T∗,hot = 2.5 · 104K and T∗,cold = 3 · 103K; C-rich with C/O
abundance ratios beyond 1, realized by a domination of amor-
phous carbon grains in the circumstellar dust and O-rich with a
dust composition dominated by warm amorphous silicates.
We applied radial density profiles dominated by radiation
pressure of the star (Ivezic & Elitzur 1995). While the chemical
compositionand temperaturesof the dust stronglyinfluences the
infrared spectrum, the stellar effective temperature T∗, and lumi-
nosity L∗scale the physical size of the system in dependence on
Tdust, the temperatureof the sublimation zone, where dust distri-
bution starts. For the grain size distribution we used a power-law
distribution as published by Mathis et al. (1977) and an upper
Fig.18. Data equal to the one in Fig. 16, but further corrected
to match the intrinsic emission of IRS 3 and its surrounding cir-
cumstellar dust shell. The corrections are discussed in the text,
and lead to the following intrinsic fluxes: 0.05, 0.5, 13.5, 30, 50,
46, 7.0 Jy at 1.6, 2.1, 3.8, 4.7, 8, 13, 19 µm. The best RT model
SED were overplotted, the respective optical depths τ13are 0.4
for both C-rich models and 1.6, and 1.3 for the cold and hot star
oxygen-dominatedmodels, respectively.
limit on the grain size of a = 0.25 µm, which was successfully
applied in similar experiments.
To demonstrate the different constraints imposed by the data
we present our results in two parts. First the measured SED can
berelatedtotheopticaldepthτofthedustandtothetemperature
Tdustat the inner boundary of the dust shell. In the next step the
remainingambiguitiesregardingthe stellar temperatureand dust
composition can be scrutinized further by comparing calculated
visibilities of the RT models to the MIDI data.
4.5.1. Modeling the spectral energy distribution
This part is aggravated by the fact that IRS 3 is observed both
through a large amount of interstellar extinction towards the GC
and through the local (partly diffuse) dust of non-negligible op-
tical depth at MIR wavelengths as shown in earlier sections.
Some uncertainty derives from the fact that the shape of the in-
terstellar extinction only at wavelengths shorter than 8 µm is
well observed and shown to be relatively constant throughout
the Galaxy(Indebetouw et al. 2005). Inadditionwe do notknow
the exact dust composition and resulting shape of the interstel-
lar extinction at MIR wavelengths within the central parsec and
around IRS 3 in particular. Thus the dereddened SED as the ba-
sis of this section is expected to create uncertainties on the order
of magnitude of tens of percent in flux.
This naturally affects the accuracy of the estimation of Tdust
and τ, which both determine the shape of the intrinsic SED.
Similarly we cannot meaningfully investigate the gradual im-
portance of secondary ingoing parameters like the density pro-
file, chemistry (e.g. existence and amount of SiC, crystalline
olivines), etc. These parameters cannot be fine-tuned unambigu-
ously by comparison with the dereddened stellar spectrum, so
fixed standard values are used.
Since the relative depth of the interstellar silicate absorption
is unclear, we excluded the data around 9.8 µm for the SED fits
and used only data at the edges of the N-band. In addition we
corrected the SED for using only 40, 30, and 20 % at 8, 13,

Page 15

J.-U. Pott et al.: The enigma of GCIRS 315
and 19 µm of the single telescope flux making allowance for the
found interferometric flux ratios, thereby eliminating the emis-
sive contributions from the outer cooler dust. Furthermore, we
increasedtheopticaldepthat λ ≥ 8 µm beyondthe AV= 25seen
at NIR wavelengths to account for the extinction caused by the
outer flux seen by the interferometer (Sect. 4.3). This is realized
by upscaling the assumed shape of the interstellar extinction to-
wards the GC (Moneti et al. 2001) to match at λ ≥ 8 µm an addi-
tional τ13≈ 0.5 (Table 4). This opacity would already explain at
least half the τ9.8excess (Fig. 10) if the same interstellar GC ex-
tinction shape would again be used for the outer dust addressed
by the larger Gaussian in our simple two-component model. It
is conceivable that, by confining this correction to λ ≥ 8 µm,
we slightly underestimatethe dereddenedfluxes at shorterwave-
lengths, which should also be affected by the outer dust.
The remainig excess extinction at 9.8 µm could stem from
either fractionally enhanced silicate dust in the central parsec in
general or in the outer and (partially diffuse) dust around IRS 3
only. In the next section we aim at investigating whether this re-
mainingτ9.8excess towards IRS 3 couldoriginatefromthe inner
dustat 20 masscales andis thus producedinan oxygen-richdust
shell. Finding the latter would suggest IRS 3 to contribute to its
possibly silicate-enriched environs.
We decided to fix Tdustfor all models at a typical value for
subliming dust of 1200 K. Tests show that by varying this Tdust
by up to 300 K, we still reproduce the earlier dust temperature
findings and generate model SEDs with deviations from the data
that can easily be explained by slightly varying the optical depth
intheouterdustofthestellarsurrounding.However,sublimation
temperatures well below 900 K can be excluded. Furthermore,
the optical depth needed to create model SED that comply with
the data is not negligible.This in turn enablesus to scrutinize the
dust composition in the next section since the different optical
properties of carbon and oxygen-rich dust show stronger impact
in the case of higher optical depths.
The probed optical depth range covered two orders of mag-
nitude (τ13= 0.1..10). The best-fit SED models under the given
assumptions are shown in Fig. 18. The main results can be sum-
marized as follows:
– the O-rich SEDs show much poorer coincidence with the
dereddened data at longer wavelengths;
– the MIR optical depth of the O-rich models is with τO,13∼
1.5 even outside the silicate feature that significantly larger
than for C-rich dust (τC,13 ∼ 0.4); this is the only finding
that would comply with a scenario in which the τ9.8excess
originates in the innermost dust;
– the photometry data alone can be fitted convincingly by
single-shell dust models, and for each dust chemistry both
hot and cold star models show fits of similar goodness
4.5.2. Modeling the visibilities
Similar to the flux correction in Sect. 4.5.1 we have to alter the
theoretical visibilities of our RT models here to account for the
outer flux probed predominantly by the short baselines prior to
comparing the model to the data. To conform this correction to
the findingsof the previoussections, we add to the respectiveRT
model a Gaussian component of twice the RT model flux and of
FWHM of 35, and 45 mas at 8, and 13 µm, respectively. Again
these parameters reflect average properties of the dust emission
at lower spatial frequencies and are variable within the ranges
implied by the findings of Sect. 3.4.
Fig.19.Measured visibilities on the edges of the N-band outside
the silicate feature. The data at 13 µm are highlighted by stars.
The respectivevisibilities of RT models plus outerGaussian (see
text) are overplotted as indicated. Quantitative details of the RT
models are given in Table 5
Fig.20. Measured visibilities at 11.5 µm in the wing of the sil-
icate feature. The best cool star RT models plus Gaussian are
overplotted,andthedottedlinesrepresentthemodelsat11.5µm.
While the C-rich model at 11.5 µm lies between the model visi-
bilities at 8 and 13 µm, as does the data, the O-rich visibilities at
11.5µm are lower than ones on the edgeof the N-bandat similar
baseline lengths.