1: % aa.dem
2: % AA vers. 6, LaTeX class for Astronomy & Astrophysics
3: % demonstration file
4: % (c) Springer-Verlag HD
5: % revised by EDP Sciences
6: %-----------------------------------------------------------------------
7: %
8: %\documentclass[referee]{aa} % for a referee version
9: %\documentclass[onecolumn]{aa} % for a paper on 1 column
10: %\documentclass[longauth]{aa} % for the long lists of affiliations
11: %\documentclass[rnote]{aa} % for the research notes
12: %
13: \documentclass[letter]{aa}
14: %
15: \usepackage{graphicx}
16: \usepackage[T1]{fontenc}
17: \usepackage{txfonts}
18: \usepackage{natbib}
19: %\usepackage{amsmath, amsthm, amssymb}
20: \newcommand{\V}[1]{\textbf{#1}} % vector
21: \newcommand{\M}[1]{\mathbf{#1}} % matrix
22:
23: \bibpunct{(}{)}{;}{a}{}{,}
24: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
25: %
26: \begin{document}
27: %
28: \title{The limb-darkened Arcturus}
29: \subtitle{Imaging with the IOTA/IONIC interferometer}
30:
31: \author{S.~Lacour
32: \inst{1,2,3}
33: \and
34: S.~Meimon\inst{4}
35: \and
36: E.~Thi\'ebaut\inst{5}
37: \and
38: G.~Perrin\inst{1}
39: \and
40: T.~Verhoelst\inst{6,7}\fnmsep \thanks{Postdoctoral Fellow of the Fund
41: for Scientific Research,
42: Flanders}
43: \and
44: E.~Pedretti\inst{8}
45: \and
46: P.~A.~Schuller\inst{9,10}
47: \and
48: L.~Mugnier\inst{4}
49: \and
50: J.~Monnier\inst{11}
51: \and
52: J.P.~Berger\inst{3}
53: \and
54: X.~Haubois\inst{1}
55: \and
56: A.~Poncelet\inst{1}
57: \and
58: G.~Le~Besnerais\inst{4}
59: \and
60: K.~Eriksson\inst{12}
61: \and
62: R. Millan-Gabet\inst{13}
63: \and
64: M.~Lacasse\inst{9}
65: \and
66: W.~Traub\inst{9,14}
67: }
68:
69:
70: \offprints{S. Lacour}
71:
72: \institute{ Observatoire de Paris, LESIA, CNRS/UMR\,8109, 92190 Meudon,
73: France
74: \and Sydney University, School of Physics, N.S.W. 2006,
75: Australia
76: \and Laboratoire d'Astrophysique de Grenoble, CNRS/UMR\,5571,
77: 38041 Grenoble, France
78: \and Office National d'\'Etudes et de
79: Recherches A\'eronautiques, DOTA, 92322 Chatillon, France
80: \and
81: Centre de Recherche Astrophysique de Lyon, CNRS/UMR\,5574,
82: 69561 Saint Genis Laval, France
83: \and Instituut voor Sterrenkunde, K.U. Leuven, 3001
84: Leuven, Belgium
85: \and University of Manchester, Jodrell Bank Centre for Astrophysics,
86: Manchester, M13 9PL, U.K.
87: \and University of St Andrews, North Haugh,
88: St Andrews, KY16 9SS, Scotland, UK
89: \and Harvard-Smithsonian Center for
90: Astrophysics, Cambridge, MA, USA
91: \and Institut d'Astrophysique
92: Spatial, Universit\'e Paris-Sud, 91405 Orsay, France
93: \and Department of Astronomy, University of
94: Michigan, Ann Arbor, MI, USA
95: \and
96: Department of Astronomy and Space Physics, Uppsala University,
97: 75120 Uppsala, Sweden
98: \and Caltech/Michelson Science Center,
99: Pasadena, CA, USA
100: \and
101: Jet Propulsion Lab, M/S 301-451, 4800 Oak Grove Dr., Pasadena CA, 91109
102: }
103:
104: %\date{??,??}
105:
106:
107: \abstract % context heading (optional)
108: % {} leave it empty if necessary
109: {}
110: % aims heading (mandatory)
111: { This paper is an H band interferometric examination of
112: Arcturus, a star frequently used as a spatial and spectral
113: calibrator.}
114: % methods heading(mandatory)
115: { Using the IOTA 3 telescope interferometer, we performed
116: spectro-interferometric observation ($R\approx35$) of
117: Arcturus. Atmospheric models and prescriptions were fitted to the
118: data to derive the brightness distribution of the photosphere. Image
119: reconstruction was also obtained using two software algorithms:
120: \textsc{Wisard} and \textsc{Mira}.}
121: % results heading (mandatory)
122: { An achromatic power
123: law proved to be a good model of the brightness distribution, with a
124: limb darkening compatible with the one derived from atmospheric
125: model simulations using our \textsc{marcs} model. A Rosseland diameter of $21.05\pm0.21$ was derived,
126: corresponding to an effective temperature of $T_{\rm
127: eff}=4295\pm26$\,K. No companion was detected from the closure
128: phases, with an upper limit on the brightness ratio of
129: $8\times10^{-4}$ at 1AU. Dynamic range at such distance from the
130: photosphere was established at $1.5\times10^{-4}$
131: ($1\sigma\,$rms). An upper limit of $1.7\times10^{-3}$ was also
132: derived for the level of brightness asymmetries present on the
133: photosphere. }
134: % conclusions heading (optional), leave it empty if necessary
135: {}
136: \keywords{
137: techniques: interferometric --
138: stars: fundamental parameters --
139: infrared: stars -- stars: individual: Arcturus
140: }
141:
142: \maketitle
143: %
144: %________________________________________________________________
145:
146: \section{Introduction}
147:
148: Arcturus' curse is to be too popular. Many new generations of
149: instruments -- including interferometers -- observe it as a test
150: object. The reasons for that: this red giant is bright, large, and
151: spectrally well-defined. The curse is: since the instruments are new,
152: the star usually takes face on the bizarre systematic errors of
153: challenging observations. We can cite, among others, inexact diameter
154: and temperature measurements (prompting \citet{1999AJ....117.2998G} to
155: write an article entitled ``The Effective Temperature of Arcturus''),
156: or false duplicity observations \citep[``Arcturus as a Double Star''
157: by][]{1998Obs...118..299G}.
158:
159: This paper does not pretend to be an exception -- interferometry is
160: still a challenging technique. The main difference is in the
161: interferometer used: at the time of our observations, IOTA was more
162: wise than new \citep[we shall note that Arcturus has already been
163: observed several times by IOTA and led to three different
164: publications;][]{1996AJ....111.1705D,1998A&A...331..619P,2005A&A...435..289V}.
165: The initial goal of a new observation run was to leverage the more
166: extended capability of IOTA to show the ability of the interferometer
167: to do a reliable image of a commonly observed object.
168:
169: Indeed, image reconstruction is difficult. Even though it is routinely
170: performed by the current generation of radio interferometers, this
171: technique -- herein called ``regularized imaging'' -- remains marginal in optical interferometry. This is simply
172: due to a lack of spatial frequency coverage. Optical interferometers are
173: usually more difficult to build, and the complexity quickly increases
174: with the number of telescopes. Therefore, since the amount of
175: information accessible in the Fourier plane is sparse, our ability to
176: reconstruct a reliable image of a complex object is limited.
177:
178: A more common data analysis is to suppose the object to be conform to
179: a model -- or prescription. Originally, it consisted in fitting
180: visibility curves of uniform disks \citep[e.g.][]{1921ApJ....53..249M,
181: 1986A&A...166..204D}. With time, it included more complicated
182: models, e.g. limb-darkened disks
183: \citep[e.g.][MkIII]{1996A&A...312..160Q}, disks with spots
184: \citep[e.g.][COAST]{2000MNRAS.315..635Y}, disks with a molecular
185: envelope \citep[e.g.][IOTA]{2004A&A...426..279P}, etc. Both
186: methods, regularized imaging and model fitting, complement each
187: other. The role of regularized imaging is commonly to guide the
188: choice and complexity of a model. The role of model fitting is to
189: obtain the highest precision results on the parameters of the
190: model. The pitfall may be when the model does not best suit the
191: object, hence the need for quality regularization imaging.
192:
193:
194: Recently the maturation of interferometric facilities (e.g. IOTA,
195: CHARA, VLTI) reached the point where $u$-$v$ coverage (both in
196: amplitude and phase) allows regularized imaging
197: \citep{2007arXiv0706.0867M,lacour_these}. Here we present the data on
198: Arcturus, used as a test star for optical interferometry
199: reconstruction softwares. Several astrophysical issues also justify
200: this investigation: 1) what is the limb darkening? Is it compatible
201: with red giant atmosphere modeling
202: \citep{2000MNRAS.318..387D,2000A&A...363.1081C}? 2) are the disputed
203: previous detections of a companion compatible with our observations
204: \citep{1997A&A...323L..49P,2005A&A...435..289V,2007PASP..119..237B}?
205:
206:
207: The outline of this paper is as follows. Section~\ref{sc:obs} gives an
208: overview of the IOTA interferometer, describes the data reduction
209: process and shortly present the dataset. Section~\ref{sc:Com_model}
210: compares our data with atmosphere models, using either limb-darkening
211: prescriptions or a more evolved atmospheric simulation (the
212: \textsc{Marcs} model). Section~\ref{sc:comp} investigates on a
213: possible deviation from point symmetry. Results of image
214: reconstruction are presented in Section~\ref{sc:image}, and
215: Section~\ref{sc:conclusion} conclude.
216:
217: %__________________________________________________________________
218:
219: \section{Observation and data reduction} \label{sc:obs}
220:
221: \subsection{Description of IOTA observations}
222:
223: \begin{figure}
224: \centering
225: \includegraphics[scale=.6]{ABoo_UV.eps}
226: \caption{$u$-$v$ coverage. Maximum projected baseline length
227: is 37.7 meters. The lack of high frequency information
228: in the East-West direction is due to the geometry of
229: IOTA.}
230: \label{fig:UV_planes}
231: \end{figure}
232:
233: \begin{figure*}
234: \centering
235: \includegraphics[width=17cm]{Plot_wav.eps}
236: \caption{Single scan obtained on calibrator HD\,120477. It shows
237: the flux (arbitrary units) as a function of the optical path
238: difference (OPD) for the three baselines, and the seven
239: spectral channels. Eye inspection allows to observe the
240: decreasing frequency -- and therefore increasing wavelength --
241: from top to bottom. The bottom fringes correspond to the sum
242: of the spectral channels, showing a radical change in the
243: coherence length. Each datapoint is composed of 200 scans.}
244: \label{fig:fringes}
245: \end{figure*}
246:
247: The interferometric data presented herein were obtained using the IOTA
248: (Infrared-Optical Telescope Array) interferometer
249: \citep{2003SPIE.4838...45T}, a long baseline interferometer which
250: operates at near-infrared wavelengths. It consists of three 0.45 meter
251: telescopes movable among 17 stations along two orthogonal linear
252: arms. IOTA synthesizes a total aperture size of $35 \times 15\,$m,
253: corresponding to an angular resolution of $\approx 10 \times 23$
254: milliarcseconds at 1.65 $\mu$m. Visibility and closure phase
255: measurements were obtained using the integrated optics combiner IONIC
256: \citep{2003SPIE.4838.1099B}; light from the three telescopes is
257: focused into single-mode fibers and injected into the planar
258: integrated optics (IO) device. Six IO couplers allow recombinations
259: between each pair of telescopes. Fringe detection is done using a
260: Rockwell PICNIC detector \citep{2004PASP..116..377P}. The interference
261: fringes are temporally-modulated on the detector by scanning piezo
262: mirrors placed in two of the three beams of the interferometer.
263:
264: Observations were undertaken in the H band
265: ($1.5\,\mu$m$\,\leq\,\lambda\leq\,1.8\,\mu$m) divided into 7 spectral
266: channels. The science target observations are interleaved with
267: identical observations of unresolved or partially resolved stars, used
268: to calibrate the interferometer's instrumental response and effects of
269: atmospheric seeing on the visibility amplitudes. The calibrator
270: sources were chosen in two different catalogs:
271: \citet{2002A&A...393..183B} and \citet{2006A&A...447..783M}; using
272: criteria on the separation ($\lesssim 10$ degrees) and magnitude. The
273: calibrators are listed in Table~\ref{tb:calib}.
274:
275: \begin{table}
276: \caption{Calibrators}
277: \label{tb:calib}
278: \centering
279: \begin{tabular}{lccc}
280: \hline
281: \hline
282: &Calibrator & Spectral Type & UD diameter \\
283: \hline
284: 1&HD\,120477 & K5.5\,III & $4.460 \pm 0.050$ \\
285: 2&HD\,125560 & K3\,III & $1.910 \pm 0.021$ \\
286: 3&HD\,129972& G8.5\,III & $1.540 \pm 0.020$ \\
287: \hline
288: \end{tabular}
289: \end{table}
290:
291: Arcturus was observed in May 2006 during 5 nights and using 5 different
292: configurations of the interferometer. Full observation information can
293: be found in Table~\ref{tb:log}, including dates of observation,
294: interferometer configurations and calibrators.
295: Fig.~\ref{fig:UV_planes} shows the $u$-$v$ coverage achieved during this
296: observation run. The geometry of the IOTA interferometer and the
297: position of the star on the sky constrained the extent of frequency
298: coverage. We covered a frequency range equivalent to the one of an
299: elliptical telescope of aperture $38\times15$ meters, with a 20
300: degree inclination East of North.
301:
302: \begin{table}
303: \caption{Arcturus observing log}
304: \label{tb:log}
305: \centering
306: \begin{tabular}{ccl}
307: \hline \hline
308: Date & Interferometer & \multicolumn{1}{c}{Calibrator} \\
309: (UT) & Configuration$^{\mathrm{a}}$ & \multicolumn{1}{c}{(Table~\ref{tb:calib})} \\
310: \hline
311: 2006 May 11 & A15-B05-C10 & 1, 2, 3\\
312: 2006 May 12 & A15-B05-C00 & 1, 2, 3\\
313: 2006 May 13 & A15-B15-C00 & 1, 3\\
314: 2006 May 14 & A30-B15-C00 & 2\\
315: 2006 May 16 & A35-B15-C25 & 2, 3\\
316: \hline
317: \end{tabular}
318: \begin{list}{}{}
319: \item[$^{\mathrm{a}}$] Configuration refers to the location in meters
320: of telescopes A, B, C on the NE, SE and NE arms respectively.
321: \end{list}
322: \end{table}
323:
324:
325:
326: \subsection{Data reduction} \label{sc:data_red}
327:
328: Reduction of the IONIC visibility data was carried out using custom
329: software similar in its main principles to the one described by
330: \citet{1997A&AS..121..379C}. We measured the power spectrum of each
331: interferogram (proportional to the target squared visibility, $V^2$),
332: after correcting for intensity fluctuations and subtracting bias terms
333: from read noise, residual intensity fluctuations, and photon noise
334: \citep{2003A&A...398..385P}. Next, the data pipeline applies a
335: correction for the variable flux ratios for each baseline by using a
336: flux transfer matrix \citep{2001PASP..113..639M}. Finally, raw squared
337: visibilities are calibrated using the raw visibilities obtained by the
338: same means on the calibrator stars. Calibration accuracy had been
339: studied by extensive observation of the binary star $\lambda$ Vir. For
340: bright stars (H mag $\lesssim 5$), \citet{2007ApJ...659..626Z} have
341: validated 2 \% calibration error for $V^2$; corresponding to a 1\%
342: error in visibility. We therefore systematically added a 2\%
343: calibration error on all the squared visibilities present in this
344: paper.
345:
346: In order to measure the closure phase (CP), a fringe tracking
347: algorithm was applied in real-time while recording interferograms
348: \citep{2005ApOpt..44.5173P}, ensuring that interference occurs
349: simultaneously for all baselines. We required that interferograms are
350: detected for at least two of the three baselines in order to assure a
351: good closure phase measurement. This technique, called
352: ``baseline bootstrapping'' allowed precise visibility and closure
353: phase measurements for a third baseline with very small coherence
354: fringes. We followed the method of \citet{1996A&A...306L..13B} for
355: calculating the complex triple amplitude and deriving the closure
356: phase. Pair-wise combiners (such as IONIC) can have a large
357: instrumental offset for the closure phase which requires to be
358: calibrated by the closure phase of the calibrator stars. We noticed
359: very stable closure phase measurements during the nights with drifts
360: of less than a degree.
361:
362: \subsection{Wavelength calibration}
363: \label{sc:wave}
364:
365: \begin{figure}
366: \centering
367: \includegraphics[width=8.cm]{Transmission.eps}
368: \caption{Relative photometry between the different spectral
369: channels as a function of the wavelength (arbitrary vertical
370: units). The wavelength was determined by measuring the frequency
371: of the fringes as shown in Fig.~\ref{fig:fringes}.}
372: \label{fig:wav}
373: \end{figure}
374:
375:
376: \begin{figure}
377: \centering
378: \includegraphics[scale=.47]{ABoo_V22.eps}
379: \includegraphics[scale=.47]{ABoo_ResV22.eps}
380: \caption{Overview of the dataset: Visibility square measurements
381: as a function of the baseline length. The wavelength is
382: color coded with the same palette used in
383: Fig.~\ref{fig:UV_planes}. The solid curve correspond to the
384: visibility curve of a uniform stellar disk of angular diameter
385: 20.30 mas (not accounting for bandwidth smearing). The bottom
386: panel presents the residual of that fit, showing the clear
387: inconsistency of the second lobe.}
388: \label{fig:Dataset}
389: \end{figure}
390:
391: \begin{figure}
392: \centering
393: \includegraphics[scale=.47]{ABoo_CP.eps}
394: \includegraphics[scale=.47]{ABoo_ResCP.eps}
395: \caption{Overview of the dataset: closure phase measurement as a
396: function of observation datafile number. The bottom panel
397: shows the residual of fitting a simple limb darkening disk to
398: the data (power law with parameters as stated in
399: Table~\ref{tb:param}). Color legend is similar to the one used
400: in Fig.~\ref{fig:UV_planes} and~\ref{fig:Dataset}. Because
401: such a model corresponds to a symmetrical brightness
402: distribution, the closure phases are either 0 or 180
403: degrees. The reduced $\chi^2$ of the closure phase alone is
404: 1.06.}
405: \label{fig:Dataset2}
406: \end{figure}
407:
408: Spectral information was obtained by the means of a prism placed
409: between the integrated optics and the PICNIC camera
410: \citep{2003SPIE.4838.1225R}. The temporally-modulated fringes are
411: therefore spatially dispersed on the detector. To ensure well-defined
412: spectral edges, we also inserted a broad band H filter in the optical
413: path. Its bandpass is spatially equivalent to seven pixels on the
414: camera.
415:
416: Wavelength calibration of the spectral channels is however a difficult
417: and critical step. This is especially important since the prism was
418: removed and re-inserted (with a slightly different position) between the
419: night of the 13th and the 14th. Fortunately, the spectral wavelength
420: is coded in the data (see Fig.~\ref{fig:fringes}). The fringe
421: frequency (in pixels$^{-1}$) is directly proportional to the
422: wavenumber. The factor of proportionality is constant since the
423: modulation of the optical path is done by moving the piezo mirrors a
424: certain distance (step-like) between each pixels reads, even though
425: the steps are smoothed out by mirror/mount inertia. The relative
426: wavelength between each channel and each night was established this
427: way with a precision better than $0.1\%$. This level of precision was
428: achievable thanks to small differential piston variations due to good
429: seeing conditions and fast reading mode.
430: The relative wavelength between each baseline was also studied. To do
431: so, we compared the fringe frequency observed at a given spectral channel
432: between the three baselines. The frequency of the third baseline is
433: equal to the sum of the frequency of the two first, within $0.2\%$
434: error bars. This means that the different baselines are at equal
435: wavelength at a $0.2\%$ level.
436:
437: However, absolute calibration requires to know the exact angle of the
438: incoming beam on the piezo mirror. March 2007 narrow band observations
439: were used, and allowed to establish the speed of the optical path
440: modulation at $0.188\pm0.002\,\mu$m/sample for the first delay line,
441: and $0.195\pm0.002\,\mu$m/sample for the second delay line.
442: Optical path modulation was measured on the third baseline at
443: $0.383\pm0.004\,\mu$m/sample. A time sample correspond to the
444: integration time between two reads. The $\approx1\%$ error bars are
445: mainly due to uncertain changes in the angle of reflection which
446: occurred between March and May 2006.
447:
448: Figure~\ref{fig:wav} summarizes the wavelength calibration results by
449: plotting the average integrated flux on each spectral channel as a
450: function of wavelength. The fairly large error bars in wavelength are
451: mainly due to the uncertainty on the angle of reflection, and
452: correspond to a possible global shift in absolute calibration. In contrast, relative wavelength is precisely established, and shows that
453: a significant displacement of the prism occurred between the night of
454: the 13th (squares), and the night of the 14th (cross).
455:
456:
457:
458:
459:
460: \subsection{IOTA field of view}
461: \label{sc:FOV}
462:
463: The high resolution of IOTA has a counterpart: the field of view is
464: limited. The first limitation is due to the field of view of each
465: individual telescopes, delimited by the cone of acceptance of the
466: fibers on the sky. Such value is difficult to estimate, since it
467: depends on the interferometer as well as the atmospheric seeing. A
468: first order estimation is to neglect the atmospheric turbulence and to
469: consider the fiber core to be filling the diffraction pattern of the
470: telescopes. In this assumption, the field of view of the telescopes
471: reads:
472: \begin{equation}
473: FOV_{\rm telescopes}=\frac{\lambda}{D}\,,
474: \end{equation}
475: where D is the diameter of an individual telescope.
476:
477: A second limitation is the field of view of the interferometer. It is
478: delimited by the maximum distance between two objects which fringes
479: overlap on the detector. To be rigorous, one should take into
480: account parameters like the mode of recombinaison, the stroke of the
481: piezo (in the case of IONIC), and even the spectral energy
482: distribution of the target. However, to establish a simple relation,
483: we will only take into account the spectral bandwidth of a spectral
484: channel ($\Delta\lambda$), as well as the distance between two telescopes ($B$):
485: \begin{equation}
486: FOV_{\rm interferometer}=\frac{\lambda^2}{\Delta\lambda\,B}\,.
487: \end{equation}
488: Note that the interferometric field of view is baseline dependent. It
489: will be larger for shorter baselines, and smaller for longer
490: baselines. Moreover, this field limitation is valid only in the
491: direction along the baseline. Perpendicular to the baseline, the
492: bandpass does not cause any field limitation. All in all, it is
493: difficult to establish the field of view of an interferometer as a
494: whole. A conservative way to do so is to consider the maximum
495: baseline length for a given direction.
496:
497: In the North-East/South-West direction, using the spectral dispersion
498: mode of IOTA ($D=45$\,cm, $\Delta\lambda=40$\,nm and $B=35$\,m), the
499: field of view is not limited by the telescope ($FOV_{\rm
500: telescopes}=750$\,mas), but by the bandwidth. The field of view is
501: 350$\,$mas at 1.55\,$\mu$m, and 480$\,$mas at 1.80\,$\mu$m. In the
502: North-West/South-East direction, the shorter baselines ($B=15$\,m)
503: allow a larger interferometric field of view, hence a 750\,mas field
504: limitation due to the telescopes size.
505:
506: We will consider in the following of this paper a 400x750\,mas field
507: of view for IOTA\footnote{IOTA's field of view decrease when using large band
508: filters}.
509:
510: \subsection{The dataset}
511:
512: The dataset consist of 924 visibility measurements and 308 closure
513: phases. The $V^2$ are plotted as a function of the baseline length in
514: the upper panel of Fig.~\ref{fig:Dataset}. The CP are plotted in the
515: upper panel of Fig.~\ref{fig:Dataset2}. The frequency plane coverage
516: was previously presented in Fig.~\ref{fig:UV_planes}. The solid curve
517: corresponds to the best fit of a uniform disk. The residuals are
518: plotted on the lower panels. It is interesting to note that fringes
519: have been observed with a contrast below 1\%. Such a low
520: contrast exists thanks to the dispersive mode, which allows a deep
521: first null. If the full H band was observed, the effect of bandwidth
522: smearing would have limited the depth of the null to several
523: percents \citep{2005ApJ...626.1138P}. Probing into the null was
524: possible thanks to bootstrapping, two baselines of sufficient
525: contrast being enough to track the fringes on all the baselines.
526:
527: A few things are striking: first, the object is
528: relatively achromatic. This can be seen on the residuals of the
529: $V^2$. Secondly, the second lobe of the data is not well fitted by a
530: uniform disk. This is due to the presence of limb darkening. Thirdly,
531: the closure phases are close to zero or $\pi$. It means the object is
532: likely to be point symmetric.
533:
534:
535:
536: \section{Comparison with atmosphere models / prescriptions}
537: \label{sc:Com_model}
538:
539: \subsection{Fitting limb-darkening prescriptions} \label{sc:ima_param}
540:
541:
542: \begin{table}
543: \caption{Diameter and limb darkening measurements}
544: \label{tb:param}
545: \centering
546: \begin{tabular}{llc}
547: \hline \hline
548: Law & \multicolumn{1}{c}{parameters} & Reduced $\chi^2$ \\
549: \hline
550: Uniform & $\theta_{\rm UD} = 20.304 \pm 0.011$\,mas & 31\\
551: \hline
552: Power & $\theta_{\rm LD} = 20.900 \pm 0.007$\,mas & 1.962\\
553: & $\alpha = 0.258 \pm 0.003$ \\
554: \hline
555: Quadratic & $\theta_{\rm LD} = 20.922 \pm 0.036$\,mas & 1.959\\
556: & $a = 0.186 \pm 0.021$ \\
557: & $b = 0.298 \pm 0.053$ \\
558: \hline
559: Quadratic & $\theta_{\rm LD} = 20.931 \pm 0.004$\,mas & 2.956\\
560: Claret (2000)
561: & $a = 0.0291$ \\
562: & $b = 0.5107$ \\
563: \hline
564: Non-linear & $\theta_{\rm LD} = 20.863 \pm 0.004$\,mas & 2.013\\
565: Claret (2000)
566: & $a_1 = 0.8175$ \\
567: & $a_2 = 0.0827$ \\
568: & $a_3 = -0.4116$ \\
569: & $a_4 = 0.1864$ \\
570: \hline
571: {\sc marcs} model & $\theta_{\mathrm{Ross}}=21.05\pm0.01$\,mas & 2.080\\
572: \hline
573: \end{tabular}
574: \begin{list}{}{}
575: \item[Note --] { Errors bars are pure calculations based on the
576: second derivate of the $\chi^2$. They are not valid when assuming an
577: unrealistic model of the brightness distribution (for example a
578: uniform disk). Moreover, diameter errors do not include the 1\%
579: uncertainty due to an eventual wavelength miscalibration (see
580: section~\ref{sc:wave}).}
581: \end{list}
582: \end{table}
583:
584:
585: Since limb darkening is apparent, a logical first step is to fit a
586: model for the brightness distribution of the photosphere. Numerous
587: types of limb-darkening (LD) prescriptions exist in the literature. We
588: used two of them, which we supposed achromatic. A power law
589: \citep{1997A&A...327..199H}:
590: \begin{equation}
591: I(\mu)/I(1)=\mu^\alpha\,,
592: \label{eq:polaw}
593: \end{equation}
594: and a quadratic law \citep{1977A&A....61..809M}:
595: \begin{equation}
596: I(\mu)/I(1)=1-a(1-\mu)-b(1-\mu)^2 \,,
597: \label{eq:quadlaw}
598: \end{equation}
599: where $\mu=\sqrt{1-(2r/\theta_{\rm LD})^2)}$, $r$ being the angular distance from
600: the star center, and $\theta_{\rm LD}$ the angular diameter of the photosphere. In
601: terms of complex visibilities, the power law limb darkening prescription
602: yields:
603: \begin{equation}
604: V(v_r) = \sum_{k \geq 0}
605: \frac{\Gamma(\alpha/2+2)}{\Gamma(\alpha/2+k+2)
606: \Gamma(k+1)} \left( \frac{- (\pi v_r \theta_{\rm LD})^2}{4} \right)^k\,,
607: \label{eq:polaw_V}
608: \end{equation}
609: where $v_r$ is the radial spatial frequency and $\Gamma$ the Euler function (
610: $\Gamma(k+1)=k!$). On the other hand, the quadratic law yields:
611: \begin{equation}
612: V(v_r) =\frac{\displaystyle(1-a-b) \frac{J_1(\zeta)}{\zeta}+ \frac{a+2b}{\sqrt{2/\pi}}
613: \frac{J_{3/2}(\zeta)}{\zeta^{3/2}}-2b\frac{J_2(\zeta)}{\zeta^2}}{
614: \displaystyle {1/2}-{a/6}-{b/12}}
615: \end{equation}
616: where $\zeta=\pi v_r \theta_{\rm LD}$, $J_1$ and $J_2$ are the first and second-order Bessel function respectively, and:
617: \begin{equation}
618: J_{3/2}(\zeta)=\sqrt{\frac{2}{\pi \zeta}} \left(\frac{\sin(\zeta)}{\zeta}-\cos(\zeta)\right) \,.
619: \end{equation}
620:
621: Results for the fits are presented in Table~\ref{tb:param}. Using
622: Eq.~(\ref{eq:polaw}), we obtained a $\chi^2$ of 2413, for 1230 degrees
623: of freedom. The reduced $\chi^2$ ($\chi^2$ over the number of degrees
624: of freedom) does not improve significantly when using a two-parameter
625: prescription for limb darkening, prompting us to consider the power law
626: model as a sufficient approximation.
627:
628: There are two main explanations for the reduced $\chi^2$ different
629: from one: (i) an underestimation of the error bars, and (ii) an
630: inexact prescription of the brightness distribution by
631: Eqs.~(\ref{eq:polaw}) and~(\ref{eq:quadlaw}). Making the distinction
632: between these two is difficult. On the one hand, photometric
633: variations of the star are observed at the order of one percent
634: \citep{2003ApJ...591L.151R}, an indication that the brightness
635: distribution should not be as smooth and symmetric as our
636: prescriptions are. On the other hand, no deviation from point symmetry
637: is observed in the closure phases (section~\ref{sc:comp}) whose
638: reduced $\chi^2$, taken independently, is 1.015. The departure from
639: simple LD models could therefore only be explained by a missing
640: point-symmetric component. The residuals are discussed more throughly
641: in the last paragraph of Section~\ref{sc:model}.
642:
643: Whatever the cause, we decided to be as conservative as possible
644: by scaling the errors to a $\chi^2$ of one. The error bars stated
645: in Table~\ref{tb:param} are obtained by this mean. To decrease the
646: $\chi^2$, we explored -- and discarded -- two other alternatives. The
647: first one was to increase the error due to calibration (higher than
648: the 2\% justified in section~\ref{sc:data_red}). However, this
649: dramatically increased the errors bar on the lowest frequencies, which
650: is not desired since they are already well fitted by our
651: prescriptions. The second approach was to add an additive error due to
652: a potentially imperfect subtraction of the power spectrum bias. Such
653: an error was included at a 2\% level for faint objects by
654: \citet{2006ApJ...647..444M}. However, this dramatically increased the
655: errors bars on the lowest visibilities, which brings an unnecessary
656: bias on data whose first zero is already well fitted. In conclusion, a
657: global scaling of the error bars was seen as the best alternative.
658:
659:
660:
661: \subsection{Fitting the \textsc{Marcs} model}
662: \label{sc:model}
663:
664:
665: \begin{figure}
666: \centering
667: \resizebox{\hsize}{!}{\includegraphics{MARCSspectrum.ps}}
668: \caption{The synthetic H-band spectrum of the {\sc marcs} model
669: (solid line: in opacity sampling resolution, dotted line:
670: convolved to the instrumental spectral resolution -- shifted up by
671: five thousands Jensky) and the central wavelengths of the spectral
672: channels of IONIC. The peak in the spectrum corresponds to the
673: H$^-$ opacity minimum.}
674: \label{fig:MARCSspectrum}
675: \end{figure}
676: \begin{figure}
677: \centering
678: \resizebox{\hsize}{!}{\includegraphics{IOTAfit.ps}}
679: \caption{ Wavelength dependent visibility curves derived from the
680: \textsc{Marcs} atmospheric model. The color coding is similar to
681: the one used in Fig.~\ref{fig:MARCSspectrum}. IONIC data are
682: superimposed on the curves, and residuals are plotted in the lower
683: panel. The closure phase residuals are identical to
684: those of Fig.~\ref{fig:Dataset2}. }
685: \label{fig:MARCSfit}
686: \end{figure}
687:
688:
689: The \textsc{Marcs} atmospheric model was presented in
690: \citet{2005A&A...435..289V}. The models were originally constructed
691: and fine-tuned for the calibration of the ISO-SWS (Infrared Space
692: Observatory Short Wavelength Spectrometer) and checked against FTS
693: spectra \citep{Decin_these}. For the present study, we searched the
694: full Arcturus FTS spectral atlas \citep{1995PASP..107.1042H} in the H
695: band for peculiar spectral features. Lines are sparse and well
696: spaced. They belong mainly to CN, OH and some atomic transitions. The
697: IONIC data are therefore ideal to study the H$^-$ continuum, which has
698: its minimum (the transition between bound-free and free-free regimes)
699: within the bandpass sampled by our data. The only free parameter to
700: match model to observations is the angular diameter corresponding to
701: the outermost point in our model intensity profiles
702: ($\tau_{\mathrm{Ross}} = 10^{-7}$). Several models with stellar
703: parameters around those determined by \citet{2003A&A...400..709D} were
704: used, but they bring no significant improvement in $\chi^2$ compared
705: to the spectroscopically preferred model (T$_{\rm{eff}} = 4320\,$K,
706: $\log{g} = 1.5$, [Fe/H] = -0.5 and v$_{\rm{turb}} = 2$\,km\,s$^{-1}$).
707:
708: The synthetic H-band spectrum calculated from our model and the
709: comparison of our dataset with this model are shown in
710: Fig.\,\ref{fig:MARCSspectrum} and Fig.\,\ref{fig:MARCSfit}. We find
711: the best agreement for a diameter of $21.465\pm0.008$\,mas, which
712: corresponds to a $\tau_{\mathrm{Ross}} = 1$ diameter of
713: 21.05\,mas. This diameter is slightly larger than the one found in
714: Sect.\,\ref{sc:ima_param}: the star appears a little smaller at
715: wavelengths of minimal photospheric opacity than at the
716: Rosseland-averaged opacity.
717:
718: With a $\chi^2$ of 2, this fit is almost as good as it was when
719: fitting a free-parameter limb darkening prescription. This is an
720: overall confirmation of the validity of the \textsc{Marcs} modeling of
721: the limb darkening. Analysis of the visibility residuals plotted in
722: Fig.~\ref{fig:MARCSfit} indicates some possible shortcomings of the
723: model (supposing error bars are not underestimated, see
724: Section~\ref{sc:ima_param}). Indeed, the high $\chi^2$ can be
725: accounted for by two biases: a chromatic bias at low frequency
726: ($\approx9\,$M$\lambda$), another achromatic at high frequencies
727: (around the second nul). Accounting for these biases could be done by
728: (i) introducing a circumstellar emission of H$_2$O at a level of half
729: a percent \citep[water detection was reported by][although normal
730: hydrostatic model does not predict any in the
731: photosphere]{2002ApJ...580..447R}, and (ii) slightly modifying the limb
732: darkening distribution. However, such possibilities are at the limit
733: of what we think is reasonable to derive from our data, and no further
734: modeling was done to avoid over-interpretations.
735:
736:
737: \subsection{On the angular diameter of Arcturus}
738:
739: Numerous angular diameter measurements can be found in the
740: literature. Previous interferometric observations either use
741: uniform-disk fitting and apply limb-darkened corrections, or fit disks
742: whose limb-darkening is fixed by atmospheric models. At a wavelength
743: of $2.2\mu$m, \citet{1986A&A...166..204D} observed Arcturus with the
744: I2T interferometer and published a diameter of $\theta_{\rm UD}
745: =20.36\pm0.20$\,mas as well as a limb-darkened value $\theta_{\rm LD}
746: =20.95\pm0.20$\,mas. Previous measurements using the IOTA
747: interferometer exist too, and yielded in the K band $\theta_{\rm LD}=
748: 19.5\pm1.0$\,mas \citep[][using bulk optics]{1996AJ....111.1705D},
749: $\theta_{\rm LD}=20.91\pm0.08$\,mas \citep[][using
750: FLUOR]{1998A&A...331..619P} and $\theta_{\rm
751: Ross}=21.18\pm0.21$\,mas \citep[][also using
752: FLUOR]{2005A&A...435..289V}. In the visible,
753: \citet{2003AJ....126.2502M} observed Arcturus using the MarkIII
754: interferometer (450-800\,nm), and after correction for a substantial
755: limb darkening effect, published $\theta_{\rm LD}
756: =21.373\pm0.247$\,mas.
757:
758: From our dataset, and taking into account wavelength calibration
759: uncertainties, we derived $\theta_{\rm LD} =20.91\pm0.21$\,mas and
760: $\theta_{\rm Ross} =21.05\pm0.21$\,mas. These results are in agreement
761: with I2T, MarkIII and IOTA observations \citep[$1.5\sigma$ in][]{1996AJ....111.1705D}. It does not yield
762: an increase in terms of precision, but our measurements are indeed
763: interesting since, unlike the others, they did not require a
764: pre-defined value to account for limb darkening. Using the Rosseland
765: diameter and \citet{1999AJ....117.2998G} estimation of the integrated
766: flux ($F=(4.98\pm0.02)\times10^{-5}$\,erg\,cm$^{-1}$\,s$^{-1}$), we
767: can update their calculation of Arcturus' effective temperature to
768: $T_{\rm eff}=4295\pm26$\,K.
769:
770:
771: \subsection{On the limb darkening}
772:
773:
774: \begin{figure}
775: \centering
776: \resizebox{\hsize}{!}{\includegraphics{IP.ps}}
777: \caption{Intensity profiles of our {\sc marcs} model as colored
778: lines. For comparison, the best parametric fit of a power law is
779: represented with diamonds, the Kurucz model with triangle, and the
780: quadratic LD curve \citet{2000A&A...363.1081C} as a dotted
781: line. This last fit differs significantly form the others,
782: revealing a problem in the method used for limb darkening fitting
783: used by Claret.}
784: \label{fig:IP}
785: \end{figure}
786:
787: %_________________________________________ Fig. IP
788: %
789: \begin{figure}
790: \centering
791: \includegraphics[width=9cm]{alphas5.ps}
792: \caption{Limb darkening as a function of the wavelength. {\it
793: Upper panel}: the observed LD coefficients $\alpha$ (with error bars)
794: are compared with the LD coefficients derived by fitting a
795: $I(\mu) = \mu^{\alpha}$ profile to the {\sc Marcs} intensity
796: profiles of our dedicated Arcturus model (T$_{\rm{eff}} =
797: 4320\,$K, $\log{g} = 1.5$,
798: [Fe/H] = -0.5 and v$_{\rm{turb}} = 2$\,km\,s$^{-1}$). {\it
799: Lower panel}: influence of temperature and gravity on the limb
800: darkening.
801: % The offset between our dedicated Arcturus model and
802: % the grid of models is due to different metalicities, which is
803: % solar in the grid.
804: }
805: \label{fig:alphas}
806: \end{figure}
807: %
808: An important prospect of this work was to compare our limb darkening
809: measurements with existing atmospheric models. A first test was to
810: derive parameters of the limb darkening, and compare them with
811: published values in the literature. We were greatly surprised to see a
812: strong inadequacy between our measurements and the quadratic
813: parameters given by \citet{2000A&A...363.1081C} (see
814: Table~\ref{tb:param} and Fig.~\ref{fig:IP}). However, they also published
815: the values for a more complex 4-parameter non-linear law:
816: \begin{equation}
817: \frac{I(\mu)}{I(1)}=1-a_1(1-\sqrt{\mu})-a_2(1-\mu)-a_3(1-\mu^{3/2})-a_4(1-\mu^2) \,.
818: \label{eq:non_linear}
819: \end{equation}
820: \citet{2000A&A...363.1081C} claims that this four-parameter non-linear law
821: should give a more reliable estimation of the limb darkening. Using
822: his published parameters (assuming T$_{\rm{eff}} = 4250\,$K, $\log{g}
823: = 1.5$, [Fe/H] = -0.5 and v$_{\rm{turb}} = 2$\,km\,s$^{-1}$) we were
824: able to indeed confirm a correct fit.
825:
826:
827: \begin{figure*}
828: \centering
829: \includegraphics[width=7cm]{ABoo_Comp_Chi2.eps}\hspace{1cm}
830: \includegraphics[width=7cm]{ABoo_Comp_Flux.eps}
831: \caption{Non-detection of a companion to Arcturus. Left panel:
832: closure phase $\chi^2$ map. Right panel: $1\sigma$
833: upper limits of the brightness ratio of a companion. The
834: circles represent the limit of the photosphere. The field of
835: view of IOTA is 400x750\,mas (see discussion in
836: Section~\ref{sc:FOV}).}
837: \label{fig:Comp}
838: \end{figure*}
839:
840:
841: However, it is not the quadratic law which is intrinsically less able
842: to match the limb darkening: when leaving the parameters free to
843: adjust, the $\chi^2$ of a quadratic law is able to match the $\chi^2$
844: of the non-linear law. Therefore, the problem with the quadratic
845: values published by \citet{2000A&A...363.1081C} should lie in the
846: method to derive the parameters. To confirm this, we used the
847: \textsc{Atlas} model \citep{1979ApJS...40....1K} -- the one used by
848: Claret -- and we were able to obtain a good fit for the limb darkening
849: (see Fig.~\ref{fig:IP}). An explanation could be found in a paper
850: written by \citet{2007ApJ...656..483H}, in which the author states that
851: conventional stellar limb fitting methods (like the one used by
852: Claret) are biased.
853:
854: But the most striking results from Table~\ref{tb:param} is the
855: consistency in fitting quality achieved when using different limb
856: darkening laws (except when using Claret's quadratic value). We noted
857: that both \textsc{ Marcs} and \textsc{Atlas} models give similar fits,
858: showing an equivalent capacity to correctly model the atmosphere of
859: Arcturus. The reduced $\chi^2$ values are not exactly 1, but are close
860: to the ones obtained when fitting LD laws with freely variable
861: parameters. This is a good validation of both atmospheric modeling
862: softwares. Secondly, we do not note any difference in the
863: fitting quality between a power and a quadratic limb darkening
864: law. Furthermore, the likelihood does not increase when using a
865: 4-coefficient non-linear law (reduced $\chi^2$ of 1.97 -- we do not
866: present the results in Table~\ref{tb:param} since none of the
867: coefficients are properly constrained by our dataset). This is
868: because we do not have the necessary angular resolution to
869: distinguish the several limb darkening laws used here. To our
870: dataset, all of them are equally good. Therefore, for angular
871: resolution no greater than ours, we recommend using the power law
872: instead of the two other limb darkening laws tested in this work,
873: since it would use less free parameters while still being able to
874: correctly model the LD.
875:
876:
877: Finally, we investigated the spectral dependence of the limb
878: darkening. To do so, we fitted a limb-darkening power law with a
879: wavelength-dependent $\alpha$ value to the observations. The fit was
880: done using an achromatic diameter as, in principle, there is no such
881: thing as a different diameter at different wavelengths: a different
882: intensity profile, or in the case of a well-behaved star just a
883: different LD, mimics a different diameter at different wavelengths. In
884: the absence of extended molecular layers and other similar large
885: deviations from a normal photospheric IP, this effect is mostly
886: accounted for by the LD parameter. Similarly, we derived theoretical
887: $\alpha$ values from our preferred \textsc{marcs} model. The result
888: is summarized in Fig.~\ref{fig:alphas}. The general agreement is quite
889: good. The almost linear slope of the limb-darkening is in fact a
890: complex combination of opacity due to the H${^-}$ continuum and
891: molecular absorptions features. A minor discrepancy is an
892: overestimation of the LD at the blue end of the bandpass. We searched
893: a grid of models for possible improvement, but no significantly better
894: fit could be attained with reasonable stellar parameters.
895:
896:
897: \section{On the point symmetry of Arcturus}\label{sc:comp}
898:
899: \subsection{Fitting the closure phases}
900:
901:
902:
903: Closure phases (CP) are extremely sensitive to deviations from point
904: symmetric brightness distributions. For example, a binary of contrast
905: ratio 1:100 could induce closure phases of several tens of degrees at
906: low visibilities. The mean of our first 147 CP measurements (first two
907: days of observation) is 0.067 degree, with an average root mean square
908: of 0.34 degree. Such high quality data is therefore excellent for
909: probing a companion. When fitting a power law limb-darkened disk to
910: the data (both $V^2$ and CP; see section~\ref{sc:ima_param}), the
911: $\chi^2$ on the CP was 327 over 308 closure phases --- corresponding
912: to a reduced $\chi^2$ of 1.07. Fig.~\ref{fig:Dataset2} shows the CP as
913: well as the residual of the fit. The fit is, in our opinion,
914: satisfactory.
915:
916: An upper limit for the brightness ratio of a possible companion can be
917: obtained. We modified the visibility function presented in
918: Eq.~(\ref{eq:polaw_V}) to account for the presence of a point-like
919: off-centered source:
920: \begin{eqnarray}
921: V(u,v) &=& (1-K) \sum_{k \geq 0}
922: \frac{\Gamma(\alpha/2+2)}{\Gamma(\alpha/2+k+2)
923: \,k!} \left( \frac{- \pi^2 \theta_{\rm LD}^2(u^2+v^2)}{4} \right)^k \nonumber \\
924: &&+ K \exp\left(2i\pi(Xu-Yv)\right)\,.
925: \label{eq:V2_comp}
926: \end{eqnarray}
927: $K$ is the brightness ratio of the companion, $X$ and $Y$ its position
928: and $u$ and $v$ the spatial frequencies (arcsec$^{-1}$). The star
929: parameters ($\alpha$ and $\theta_{\rm LD}$) are fixed to the value
930: presented in Table~\ref{tb:param}. For each position of the companion
931: Eq.~(\ref{eq:V2_comp}) is computed, CP are derived and $K$ is adjusted
932: to minimize the $\chi^2$ on the closure phases. The minimum $\chi^2$
933: are plotted as a function of $X$ and $Y$ on the left panel of
934: Fig.~\ref{fig:Comp}. The general minimum $\chi^2$ for a companion
935: situated within the field of view of IOTA but further away than 1\,AU
936: of the star (400\,mas$>\sqrt{X^2+Y^2}>89$\,mas) is 299, with a
937: brightness ratio $K=(4\pm4)\times10^{-4}$. This is not significant
938: enough to be considered a detection. The values $(K+\sigma(K))$ can
939: nevertheless be used to derive upper limits for the brightness ratio
940: of a possible binary system. It is plotted in the right panel of
941: Fig.~\ref{fig:Comp}. The average dynamic range at 1\,AU of the star is
942: $1.5\times10^{-4}$.
943:
944: Closer to the photosphere, the $\chi^2$ can decrease
945: substantially. The minimum is 257, for $X=10$ and $Y=11$\,mas. It
946: still does not mean we have detected anything, since this value is
947: below the number of degrees of freedom. However, the fit can be used to
948: put upper limits on the brightness of a possible hotspot on the
949: photosphere. The maximum value for $K+\sigma(K)$ on the photosphere is
950: $1.7\times10^{-3}$. Note that the signature of an hotspot on the CP
951: gets smaller when it gets closer to the photocentre. Therefore, we
952: cannot exclude the presence of a bright hotspot coincidentally
953: situated in the middle of the photosphere.
954:
955:
956: \subsection{Presence of a companion?}
957:
958: Arcturus is often used for high-resolution spatial and spectral
959: calibration \citep{2000ApJ...534..907T,2003A&A...400..709D}. Such use
960: makes this star both very well known and very important to know. This
961: explains why, when {\it Hipparcos} flagged this star as a binary, it
962: stirred an important debate in the community. The absence of other
963: observational evidence \citep{1998Obs...118..299G}, uncertainties in
964: the {\it Hipparcos} detection \citep{1998Obs...118..365S} and finally
965: non-detection with adaptive optics observations
966: \citep{1999PASP..111..556T}, convinced the community they could keep
967: using Arcturus as a calibrator. Our results put an upper limit on the
968: brightness ratio of a possible companion to $8\times10^{-4}$ in the H
969: band.
970:
971: To make our results compatible with a binary system as proposed by
972: {\it Hipparcos} \citep{1997A&A...323L..49P} or
973: \citet{2005A&A...435..289V} ($\Delta m \approx 4$, $\rho \approx 230$\,mas,
974: $M \geq 0.7 M_\odot$), we would have to imagine either (i) a strong
975: dependence of the wavelength or (ii) an edge-on orbit with the
976: secondary occulted by the primary. Both possibilities can be ruled out
977: since (i) a differing spectral type would have aroused spectroscopists
978: and (ii) an edge-on orbit would have aroused people doing radial
979: velocity measurements. However, a lower mass planet of a few Jovian
980: masses, as proposed by \citet{1989PASP..101..147I,1993ApJ...413..339H}
981: and \citet{2007PASP..119..237B} is still a possibility. Our
982: measurement gives an upper limit on its relative magnitude in the H
983: band ($\Delta m>7.75$).
984:
985:
986:
987: \subsection{Asymmetric brightness distribution of the stellar surface}
988:
989:
990: Radial velocities \citep{1999ASPC..185..187M} as well as photometry
991: \citep{2003ApJ...591L.151R,2007arXiv0706.3346T} indicate variations of
992: a few days period. Photometric oscillations are especially notable,
993: with amplitude variation of up to a percent, well above what is
994: predicted by atmospheric models \citep{2001MNRAS.328..601D}. By
995: putting a $1.7\times10^{-3}$ 1\,$\sigma$ upper limit on the flux of an
996: eventual hotspot, our observations show that the temporal brightness
997: oscillations do not have a spatial counterpart. It means the source of
998: these variations is most likely not due to convection cells and/or
999: non-radial oscillations. It is interesting to note that interferometry
1000: could be a good tool to detect non-radial pulsation in variable stars
1001: ($\beta$ Cephei, ...).
1002:
1003: \section{Imaging Arcturus} \label{sc:image}
1004:
1005: \subsection{The \textsc{Wisard} and \textsc{Mira} reconstruction softwares}
1006:
1007: \begin{figure*}
1008: \centering \includegraphics[scale=.63,angle=-90]{Images_ABoo.eps}
1009: \caption{Three different images of Arcturus. On the left hand,
1010: image produced using a limb-darkening power law with parameter
1011: from Table~\ref{tb:param}. The two others images are obtained
1012: using two different softwares for regularized imaging. The
1013: right hand scale is derived using the {\it Hipparcos} parallax
1014: of $88.83\pm0.53$\,mas \citep{leeuwen07}. }
1015: \label{fig:Images}
1016: \end{figure*}
1017:
1018:
1019: Model fitting confines the image to be within the range of a
1020: pre-defined model. This is a perfect tool to derive parameters of
1021: astronomical objects whose morphology is already known. However, it
1022: would not reveal any unexpected phenomenon, hence the need for less
1023: constraining image reconstruction.
1024:
1025: The image is sought by minimizing a so-called \emph{cost function}
1026: which is the sum of a regularization term plus data related terms. The
1027: data terms enforce the agreement of the model image with the different
1028: kind of measured data (power spectrum, phase closures, complex
1029: visibilities, etc). The interpolation of missing data is allowed by
1030: the regularization and by strict constraints such as the positivity
1031: (which plays the role of a floating support constraint) and
1032: normalization.
1033:
1034: To validate this imaging process, we used two different reconstruction
1035: algorithms: \textsc{Wisard} and \textsc{Mira}. \textsc{Wisard}
1036: \citep{Meimon-t-05,Mugnier-l-08a} stands for ``Weak-phase
1037: Interferometric Sample Alternating Reconstruction Device''. Its
1038: approach consists in finding the image and the missing phase data
1039: jointly. This technique is called \textit{self-calibration} in
1040: radio-interferometry \citep{Cornwell-81} and has enabled reliable
1041: images to be reconstructed in situations of partial phase
1042: indetermination. The strength of \textsc{Wisard} is that it combines,
1043: within a Bayesian framework, a recently developed noise model
1044: approximation suited to optical interferometry data
1045: \citep{2005JOSAA..22.2348M}, and an edge-preserving regularization
1046: \citep{2004JOSAA..21.1841M} to deal with the sparsity of the data
1047: typical of optical interferometry.
1048:
1049: \textsc{Mira} \citep{Thiebaut_etal-2003-JENAM} stands for
1050: ``Multi-aperture Image Reconstruction Algorithm''. Compared to
1051: \textsc{Wisard}, \textsc{Mira} does not explicitly manage the missing
1052: Fourier phase information: all missing information is handled
1053: implicitly in the data related term. For instance, it is possible to
1054: reconstruct an image given only the Fourier modulus information
1055: \citep[power spectrum data;][]{Thiebaut-2007-GRETSI}. A second
1056: difference -- in these image reconstructions of Arcturus -- lies in
1057: the chosen prior. Instead of an edge preserving regularization, we
1058: used for this reconstruction a quadratic regularization criterion. To
1059: that end, we computed a prior which is a parametric model image of a
1060: stellar surface (a quadratic limb darkening law). This method, similar
1061: to that used by \citet{Monnier_etal-2007-Altair}, has the
1062: particularity of requiring a rough model of the observed object. It
1063: will inject more information into the reconstruction, which in turn
1064: can give wrong results if the prior model is not right.
1065:
1066: We shall note that both \textsc{Wisard} and \textsc{Mira} algorithms
1067: can use various types of priors (entropy, Tikhonov, etc). It is
1068: therefore possible to have both algorithms using the same prior. The
1069: phase management -- explicit or implicit -- will however be different.
1070:
1071: The images reconstructed by \textsc{Wisard} and \textsc{Mira} are
1072: shown in Fig.~\ref{fig:Images}. A third representation of Arcturus is
1073: also presented. It is an image reconstructed from the parameters
1074: derived by fitting a power law limb-darkening prescription on the data
1075: (values presented in Table~\ref{tb:param}). We tentatively call such
1076: type of image reconstruction ``Parametric imaging''. Cuts of the
1077: brightness distributions are presented Fig.~\ref{fig:Images_cut}. The
1078: similarity of the reconstructions is quite striking considering that
1079: the 2 reconstruction methods (i.e. data-fidelity terms), as well as the
1080: priors, are different.
1081:
1082:
1083: \subsection{Discussion of image reconstruction}
1084:
1085:
1086:
1087: The left hand image of Fig.~\ref{fig:Images} shows a featureless
1088: limb-darkened star. It is not a surprise since the image is strictly
1089: constrained by the prescription. However, the important result is the
1090: good fit of the prescription on our data. When doing parametric
1091: imaging, the $\chi^2$ is a strong information to judge the reliability
1092: of the image reconstruction. In this case, a reduced $\chi^2$ of 1.9
1093: for 2413 degrees of freedom is a good validation of the derived image.
1094:
1095: When dealing with regularized imaging, it is more difficult to judge
1096: the reliability of an image reconstruction. This is because the
1097: quantity which is minimized is a sum of a regularization term and a
1098: data term (generally, the $\chi^2$). The minimum of this {\em cost
1099: function} is therefore dependent on the regularization term, and no
1100: process is known which could use this minimum to judge on the quality
1101: of the reconstruction. The $\chi^2$ is still of interest, but does
1102: only give a partial view on the reliability of the reconstructed image:
1103: a reduced $\chi^2$ close to one is important, but it is not a
1104: quantity by itself which will ensure the quality of the image
1105: reconstruction.
1106:
1107: Ultimately, the quality of image reconstruction will be dependent on
1108: the choice of the regularization term. The closer to the object the
1109: regularization term brings us, the closer to the reality our image
1110: reconstruction will be. According to this philosophy, it is important
1111: to have a good estimation of the prior. We recommend for optical
1112: interferometrist to use an adjustable regularization term.
1113: Simultaneously or sequentially, a solution we propose consists in: (i)
1114: fit a parametric image which best describe the data, and (ii) find the
1115: image which best fit both the data and the parametric image. This
1116: technique was the one used with the \textsc{Mira} reconstruction
1117: software.
1118:
1119:
1120: \section{Conclusion}
1121: \label{sc:conclusion}
1122:
1123:
1124:
1125: \begin{figure}
1126: \centering \includegraphics[scale=.45]{Image_cut.eps}
1127: \caption{Intensity slices through the $x$ and $y$ axes of the
1128: three images of Arcturus presented in
1129: Fig.~\ref{fig:Images}. The solid curve correspond to the
1130: parametric image, the dashed curves to the \textsc{Wisard}
1131: reconstruction, and the dotted curves to the \textsc{Mira}
1132: reconstruction. If we suppose the limb darkening prescription
1133: to be the correct brightness distribution of the object, we
1134: can derive the residual of the reconstruction obtained by the
1135: two imaging softwares: the rms error is around 5\% for the
1136: \textsc{Wisard} prior (edge-preserving) and 2-3\% for
1137: \textsc{Mira} (limb-darkening prior).}
1138: \label{fig:Images_cut}
1139: \end{figure}
1140:
1141:
1142: In this paper, we presented an Arcturus dataset of interesting
1143: quality. With the IOTA/IONIC interferometer, we measured fringe
1144: contrasts of less than a percent, with errors bars in average below
1145: that level. Using this data, we fitted several models and
1146: prescriptions. The closure phases were well fitted by point-symmetric
1147: prescriptions. No companion at less than one AU was detected with an
1148: upper limit on its contrast ratio of $8\times 10^{-4}$. The same
1149: modeling of the closure phases allowed the derivation of an upper
1150: limit on the heterogeneity of the photosphere: no hotspot with a
1151: brightness above $1.7\times 10^{-3}$ the total flux of the photosphere
1152: was detected.
1153:
1154: We adjusted \textsc{Marcs} atmospheric models to the data. The derived
1155: Rosseland diameter equaled $21.05\pm0.21$ mas, most of the error bar
1156: being induced by non-trivial wavelength calibration. With a reduced
1157: $\chi^2$ of 2, it is interesting to note that atmosphere models of
1158: regular K giants can now be challenged by interferometry at a very
1159: fundamental level, even though spectroscopic agreement is
1160: near-perfect. Interestingly, we noted (i) a slight inconsistency in
1161: the magnitude of the limb-darkening at short wavelength ($\lambda
1162: \approx 1.55\,\mu$m; see Fig.~\ref{fig:alphas}), and (ii) a slight
1163: chromatic effect present in the residual (lower panel of
1164: Fig.~\ref{fig:MARCSfit}). This last results could hint the presence of
1165: a marginal ($\approx 0.5$\%) water vapor emission outside the
1166: photosphere.
1167:
1168: Finally, we imaged the photosphere using two different reconstruction
1169: algorithms (\textsc{Wisard} and \textsc{Mira}). Both produced
1170: realistic images, but highlight the difficulty to judge the
1171: reliability of regularized image reconstruction. Comparatively, we
1172: presented an image reconstructed from an ad-hoc prescription of a limb
1173: darkened stellar surface. The low number of free parameters, combined
1174: with a good fit to the data, hinted to us that the most realistic
1175: brightness distribution is in fact the one of a simple limb darkened
1176: disk.
1177:
1178:
1179: \begin{acknowledgements}
1180: SL acknowledges financial support through a {\it Lavoisier}
1181: fellowship. This work also received the support of PHASE, the high
1182: angular resolution partnership between ONERA, Observatoire de Paris,
1183: CNRS and University Denis Diderot Paris 7.
1184: \end{acknowledgements}
1185:
1186:
1187: \bibliographystyle{aa}
1188: \bibliography{ABbib,Cygbib}
1189:
1190: \end{document}
1191: