Free Access
Issue
A&A
Volume 526, February 2011
Article Number L4
Number of page(s) 4
Section Letters
DOI https://doi.org/10.1051/0004-6361/201015679
Published online 13 December 2010

© ESO, 2010

1. Introduction

Solar twins, defined as spectroscopically identical to the Sun (Cayrel de Strobel et al. 1981), are important because they allow precise differential analysis relative to the Sun (Ramírez et al. 2009; Meléndez et al. 2009). The brightest solar twin is 18 Sco (HD 146233, HIP 79672; V = 5.5), whose mean atmospheric parameters are Teff = 5813 ± 21 K, log g = 4.45 ± 0.02 and [Fe/H] = 0.04 ± 0.01 (Takeda & Tajitsu 2009; Ramírez et al. 2009; Sousa et al. 2008; Meléndez & Ramírez 2007; Takeda et al. 2007; Meléndez et al. 2006; Valenti & Fischer 2005). Its rotation rate and magnetic field are also similar to solar ones (Petit et al. 2008). Its position in the H-R diagram indicates that the star should be slightly younger and more massive than the Sun (do Nascimento et al. 2009, and references therein).

During the past decade, asteroseismology and interferometry have arisen as powerful techniques for constraining stellar parameters (e.g., Cunha et al. 2007; Creevey et al. 2007). Asteroseismology involves measuring the global oscillation modes of a star (which for Sun-like stars are pressure-driven p modes). It is the only observational technique that is directly sensitive to the deeper layers of the stellar interior, since the characteristics of the modes depend on the regions through which the waves travel. Interferometry requires long-baseline interferometers capable of resolving distant stars, hence allowing measurement of their radii.

These techniques have already been combined to study the bright sub-giant β Hyi (North et al. 2007), for which a mass was derived through homology relations. Here, we apply a similar method to 18 Sco. In Sect. 2 we present the asteroseismic data and describe the method used to derive the average large frequency separation. In Sect. 3 we describe the interferometric measurements that, combined with the parallax, allow us to estimate the radius. In Sect. 4 we use these quantities to derive the mass.

2. Asteroseismology

Detecting solar-like oscillations in a fifth-magnitude star from the ground is challenging and only a few instruments offer the required efficiency and high precision. We observed 18 Sco using the HARPS spectrograph on the 3.6-m telescope at La Silla Observatory, Chile (Mayor et al. 2003). The data were collected over 12 nights from 10 to 21 May 20091. We used the high-efficiency mode with an average exposure time of 99.6 s. This resulted in a typical signal-to-noise ratio at 550 nm of 158, with some exposures reaching as high as 240. The measured radial-velocity time series (filtered for low-frequency variations) is shown in Fig. 1. We obtained 2833 points with uncertainties in general below 2 m s-1. The average dispersion per night is  ~1.11 m s-1 and can be attributed mostly to p modes.

The power spectrum, calculated using the measurement uncertainties as weights, is shown in Fig. 2. The spectral window W, which is the Fourier transform of the observing window, w(t) =  ∑ δ(t − ti) (with ti the mid-exposure time of the ith exposure), is shown in the inset of Fig. 2. Strong aliases caused by the daily gaps appear on both sides of the central peak at multiples of  ±11.57 μHz. The power spectrum shows a clear excess around 3 mHz that is characteristic of solar-like oscillations, reaching  ~0.04 m2 s-2 (corresponding to amplitudes  ~20 cm s-1).

The median sampling time was 135.0 s, including the read-out (~22.6 s). This leads to an equivalent Nyquist frequency of 3.7 mHz. In Fig. 2, we clearly see a steep rise in power at 7 mHz, corresponding to the folded low-frequency increase. This also causes the bump that appears between 3.7 mHz and  ~5.5 mHz, which is an alias of the oscillation spectrum.

thumbnail Fig. 1

Time series of radial velocities (upper panel) and their uncertainties (lower panel) from HARPS observations of 18 Sco.

Open with DEXTER

The asymptotic relation for high-order p modes is νn,l = (n + l/2 + ϵsν, where νn,l is the frequency of the mode with radial order n and angular degree l, the average large separation Δν and a surface phase offset ϵs (Tassoul 1980). There is a periodicity of Δν/2 in the frequency distribution, implying that there will be a local maximum at this value in the autocorrelation function (ACF) of the signal y(t), defined by Ryy(τ) = E[y ∗ (t)y(t + τ)] (where E is the expectation value and y ∗  the complex conjugate of y). The ACF for 18 Sco is shown in Fig. 3. Since the signal is irregularly sampled with daily gaps, it was computed by applying the Wiener-Khinchine theorem.

The large separation estimator is thus simply (1)where T is the domain in which we search for this maximum, which we set at T = [13   000,25   000] s (i.e. in the frequency range 40–80 μHz), and H the Fourier transform of a filter h(ν) that truncates the power spectrum. Indeed, only the range 1500–3700 μHz is considered when computing (using an FFT algorithm) the Fourier transform of the spectrum. We used zero-padding to ensure that the ACF was evaluated at points separated by an “equivalent frequency resolution”  ~0.01 μHz. As noted by Roxburgh (2009), the width of h(ν) affects the localization of , which is a limitation of the method.

thumbnail Fig. 2

Power spectrum of 18 Sco, evaluated using a weighted Lomb-Scargle “periodogram”. The vertical grey dashed line marks the location of the equivalent Nyquist frequency. The inset shows the spectral window W, normalized to its maximum.

Open with DEXTER

The next step is to obtain information about the statistical properties of the estimator that accounts for the noise in the data. We can write (2)where y = [y0,...,yN] are the measured values of the radial velocity at times t0,...,tN, x = [x0...,xN] are the true values of the radial velocity and ε = [ε0,...,εN] is a vector gathering the noise contributions from observational errors. Our goal is to estimate the probability density of conditional on y, . To do so, we used a Monte Carlo approach to error propagation. We assumed that the noise in the data is a series of realizations of independent random variables distributed according to the Gaussian distributions , with the σi given by the uncertainties on the data. We simulated time series by generating new realizations of the noise distributed according to the at each ti and adding them to yi. For each artificial set of data ya, we estimated .

Our process for generating the artificial data means that it satisfies (3)with ε the artificially generated noise, and ε and ε both realizations of the same distribution at time ti. Ideally, one wishes to estimate the large separation from x. One possibility would be to estimate it from several measurements, on different telescopes at the same times ti. A second way would be to generate the ya from a model reproducing the data y, then perturbing the output of this model, rather than the real observations, which is the classical procedure of Monte Carlo estimation of parameters. Unfortunately, the knowledge of Δν alone does not permit such a model to be set up.

It thus has to be assumed that this bias will not be too severe, which can be crudely checked graphically with an échelle diagram (see Fig. 4). Further estimations of the large separations, using individual frequencies, may give us some information on its magnitude. However, the error bars are representative of the error propagation: were we able to correct for the bias induced by random observational noise, we would expect our confidence interval on the large separation to be the same.

Figure 4 shows the results for our Monte Carlo suite of 10 000 time series. It closely follows a Gaussian distribution with parameters μ = 134.4 μHz and σ = 0.3 μHz. The right panel represents the échelle diagram of the observations using this value for the mean large separation.

thumbnail Fig. 3

Filtered autocorrelation function for the observed data. The shaded area marks the interval T, in which we searched for the local maximum corresponding to Δν/2. F is the Fourier transform.

Open with DEXTER

3. Interferometry

To measure the angular diameter of 18 Sco, which is expected to only be about 0.7 mas, we used long-baseline interferometry at visible wavelengths. We used the PAVO beam combiner (Precision Astronomical Visible Observations; Ireland et al. 2008) at the CHARA array (Center for High Angular Resolution Astronomy; ten Brummelaar et al. 2005). We obtained four calibrated sets of observations on 18 July 2009 using the S1-W2 (211 m) baseline.

PAVO records fringes in 38 wavelength channels centred on the R band (λc ≃ 700 nm). The raw data were reduced using the PAVO data analysis pipeline (Ireland et al. in prep.). To enhance the signal-to-noise ratio, the analysis pipeline can average over several wavelength channels, and for 18 Sco we found an optimal smoothing width of five channels. Excluding four channels on each end because of edge effects, this resulted in six independent data points per scan and hence a total of 24 independent visibility measurements for 18 Sco.

Table 1 lists the three stars we used to calibrate the visibilities. We estimated their angular diameters, θ, using the VK calibration of Kervella et al. (2004). Although the internal precision of this calibration, as well as the uncertainties in the photometry for all three stars, is better than 1%, we assume here conservative uncertainties of 5% for each calibrator (van Belle & van Belle 2005). These incorporate the unknown orientation and expected oblateness in fast rotators (Royer et al. 2002). These were chosen to be single stars with predicted diameters at least a factor of two smaller than 18 Sco and to be nearby on the sky (at a separation d < 10°). Each of the four scans of 18 Sco was calibrated, using a weighted mean of the calibrators bracketing the scan. All scans contributing to a bracket were made within a time interval of 15 min. The final calibrated squared-visibility measurements are shown in Fig. 5 as a function of spatial frequency.

thumbnail Fig. 4

Left panel: échelle diagram corresponding to the mean value of . Right panel: results from the Monte Carlo experiment. The histogram shows the distribution of the actual realizations, and the continuous line the Gaussian with the corresponding mean and variance.

Open with DEXTER

To determine the angular diameter, corrected for limb darkening, we fitted the following model to the data (Hanbury Brown et al. 1974): (4)with x = πBθLDλ-1. Here, V is the visibility, μλ the linear limb-darkening coefficient, Jn(x) the nth order Bessel function, B the projected baseline, θLD the limb-darkened angular diameter, and λ the wavelength at which the observations were done. In our analysis, we used μλ = 0.607 ± 0.012, which is interpolated at the Teff, log g, and metallicity of 18 Sco in the R-filter given in the catalog of Claret (2000). For all wavelength channels, we assumed an absolute error of 5 nm (~0.5%).

Interferometric measurements are often dominated by systematic errors and therefore require a careful analysis of all error sources. To arrive at realistic uncertainties for the angular diameter, we performed a series of 104 Monte Carlo simulations as follow. For each simulation, we drew realizations from the observed values (assuming they correspond to the parameters of Gaussian distributions) for the calibrator angular diameters, limb darkening coefficient, and wavelength channels. With these parameters we then calibrated the raw visibility measurements and fit the angular diameter θLD to the calibrated data using a least-squares minimization algorithm. Finally, we generated for each simulation a random sample of 200 normally distributed points with a mean corresponding to the fitted diameter and a standard deviation corresponding to the formal uncertainty (scaled so that χ2 = 1) of the fit. For each Monte-Carlo simulation these 200 points were stored to make up the final distribution containing 2 × 106 points. This procedure was carried out for all independent measurements in our data.

The resulting distribution for the diameter of 18 Sco is shown in Fig. 5, along with the best-fitting model. The mean and standard deviation of this distribution yield θLD = 0.6759 ± 0.0062 mas. Combined with the Hipparcos parallax of 71.94 ± 0.37 mas (van Leeuwen 2007), we find the radius of 18 Sco to be R/R = 1.010 ± 0.009. We conclude that the radius of 18 Sco is the same as the Sun, within an uncertainty of 0.9%.

thumbnail Fig. 5

Calibrated squared visibilities for 18 Sco. The red line represents the best model. Each symbol type corresponds to one scan, the visibility averaged over five wavelength channels. Upper-right inset: same model represented on a larger scale. Lower-left inset: distribution for angular diameter (mean and standard deviation represented by vertical lines).

Open with DEXTER

Table 1

Properties of the calibrators used for 18 Sco.

4. Mass of 18 Sco

Gough (1990) pointed out that the homology relation: (5)holds for main-sequence stars even outside the zero-age main sequence. Other studies have confirmed this picture (e.g., Stello et al. 2009).

Applying the ACF method described above to a one-year BiSON time series of the Sun (Broomhall et al. 2009, and references therein), we found the solar large separation to be 135.229 ± 0.003 μHz. With our radius measurement this gives a mass M = 1.02 ± 0.03 M for 18 Sco. The agreement is good with the published estimates derived from indirect methods, such as comparison between spectro- or photometric observations and stellar evolutionary tracks (Valenti & Fischer 2005; Meléndez & Ramírez 2007; Takeda et al. 2007; Sousa et al. 2008; do Nascimento et al. 2009).

The assumption of homology is in general well-supported by the comparison to models. Considering a small departure from it in the form Δν ∝ (1 + c)M1/2R−3/2 (with c ≪ 1, being a function of the structure of the star), we then have for a scaling relative to the Sun, c = (Δνν)(M/M)1/2(R/R)−3/2 − 1. For stellar ages characteristics of those quoted for 18 Sco (the dependence on the mass and the metallicity being weak), this quantity may contribute to an additional  ~0.2%–0.4% on the total error on the mass.

The impact of filtering the ACF is not completely negligible, and if we vary the lower limit of T, between 1500 μHz and 2000 μHz, the final estimate may vary by  ~0.01 M, emphasizing the need for individual frequency measurements.

5. Conclusion

We presented the first asteroseismic and interferometric measurements for the solar twin 18 Sco. These allowed us to estimate a mass for this star independent of the previous spectro-photometric studies, which are still being confirmed. This work shows the possibilities offered by asteroseismology, even from a ground-based single site, and iby nterferometry. Our results confirm that 18 Sco is remarkably similar to the Sun in both radius and mass.

The next step will involve measuring the individual oscillation frequencies and performing full modelling using all the available observations. It will hopefully reduce the uncertainty on the estimated age, improving our knowledge of the physical state of 18 Sco (do Nascimento et al. 2009). This will provide a more precise picture of its interior and give information on the depth of its external convective zone (Monteiro et al. 2000), which is necessary if one wishes to study its magnetic activity cycle (Petit et al. 2008).


1

An attempt to reduce the daily aliases, with simultaneous observations on SOPHIE (Observatoire de Haute-Provence, run ID: 09A.PNPS.THEA), was unfortunately plagued by bad weather.

Acknowledgments

This work was supported by grants SFRH/BPD/47994/2008, PTDC/CTE-AST/098754/2008, and PTDC/CTE-AST/66181/2006, from FCT/MCTES and FEDER, Portugal. This research was supported by the Australian Research Council (project number DP0878674). Access to CHARA was funded by the AMRFP (grant 09/10-O-02), supported by the Commonwealth of Australia under the International Science Linkages programme. The CHARA Array is owned by Georgia State University. Additional funding for the CHARA Array is provided by the National Science Foundation under grant AST09-08253, by the W. M. Keck Foundation, and the NASA Exoplanet Science Center.

References

All Tables

Table 1

Properties of the calibrators used for 18 Sco.

All Figures

thumbnail Fig. 1

Time series of radial velocities (upper panel) and their uncertainties (lower panel) from HARPS observations of 18 Sco.

Open with DEXTER
In the text
thumbnail Fig. 2

Power spectrum of 18 Sco, evaluated using a weighted Lomb-Scargle “periodogram”. The vertical grey dashed line marks the location of the equivalent Nyquist frequency. The inset shows the spectral window W, normalized to its maximum.

Open with DEXTER
In the text
thumbnail Fig. 3

Filtered autocorrelation function for the observed data. The shaded area marks the interval T, in which we searched for the local maximum corresponding to Δν/2. F is the Fourier transform.

Open with DEXTER
In the text
thumbnail Fig. 4

Left panel: échelle diagram corresponding to the mean value of . Right panel: results from the Monte Carlo experiment. The histogram shows the distribution of the actual realizations, and the continuous line the Gaussian with the corresponding mean and variance.

Open with DEXTER
In the text
thumbnail Fig. 5

Calibrated squared visibilities for 18 Sco. The red line represents the best model. Each symbol type corresponds to one scan, the visibility averaged over five wavelength channels. Upper-right inset: same model represented on a larger scale. Lower-left inset: distribution for angular diameter (mean and standard deviation represented by vertical lines).

Open with DEXTER
In the text

Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.

Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.

Initial download of the metrics may take a while.