HARPS-N observes the Sun as a star

The radial velocity (RV) technique is successful at finding super-Earth planets orbiting bright stars (Mayor et al. 2011; Howard et al. 2010), as well as extremely short period Earth-mass planets (Dressing et al. 2015; Dumusque et al. 2014b; Pepe et al. 2013; Howard et al. 2013; Dumusque et al. 2012). However, RV detection of Earth-twins, i.e., long-period terrestrial worlds around solar-type stars, requires an order of magnitude improvement in precision to $\sim$10 cm s${}^{-1}$ . Among the multitude challenges to reaching such RV precision, the most important ones are: (i) stable, long-term wavelength calibration of the astrophysical spectrograph; and (ii) accounting for the effects of stellar noise, i.e., RV perturbations induced by stellar surface inhomogeneities, which are typically a few  m s${}^{-1}$ for solar-type stars. The first challenge has been successfully addressed with astro-comb calibrators, based on laser frequency combs locked to atomic clocks (Glenday et al. 2015; Phillips et al. 2012; Wilken et al. 2012, 2010). To meet the second challenge, it will be crucial to better understand RV noise induced by stars and develop mitigation techniques, particularly as current and future high-resolution spectrographs such as HARPS (with its newly commissionned astro-comb, Wilken et al. 2012; Mayor et al. 2003), HARPS-N (Cosentino et al. 2012), ESPRESSO (Pepe et al. 2014), and G-CLEF (Szentgyorgyi et al. 2014) should reach long-term instrumental RV precision allowing Earth-twin detection.

Stellar RV noise can be differentiated into four categories (given the current state of knowledge) :

• stellar oscillations, on a timescale of a few minutes for solar-like stars, produced by pressure waves propagating in stellar interiors (Dumusque et al. 2011b; Arentoft et al. 2008; Kjeldsen et al. 2005);

• stellar granulation and supergranulation, on a timescale from a few minutes to 48 hours, resulting from convectively driven outflowing material reaching the stellar surface (Dumusque et al. 2011b; Del Moro et al. 2004; Del Moro 2004);

• short-term stellar activity, on the timescale of the stellar rotational period, induced by rotation in the presence of evolving and decaying surface inhomogeneities including spots and plages (e.g., Borgniet et al. 2015; Dumusque et al. 2014a; Boisse et al. 2012; Saar 2009; Meunier et al. 2010; Saar & Donahue 1997); and

• long-term stellar activity, on a timescale of several years, induced by stellar magnetic cycles and/or surface flows (Meunier & Lagrange 2013; Lovis et al. 2011; Dumusque et al. 2011c; Makarov 2010).

Due to their semi-periodic nature, stellar oscillations can be averaged out efficiently by taking measurements with exposure times longer than the timescale of these perturbations (Dumusque et al. 2011b). This observational strategy is now commonly employed in high precision RV surveys, e.g., by imposing a minimum exposure time of fifteen minutes. Stellar granulation can be mitigated somewhat by observing a star several times per night, with the maximum possible spread between measurements (Dumusque et al. 2011b). However, this approach is only partially effective, and better mitigation methods need to be developed. For short-term activity, some correction techniques have been investigated (Meunier & Lagrange 2013; Aigrain et al. 2012; Boisse et al. 2011, 2009), but none are fully satisfactory. Finally, long-term activity seems to correlate well with the calcium chromospheric activity index, which provides a promising approach to mitigation of this source of stellar RV noise (Meunier & Lagrange 2013; Dumusque et al. 2012).

Characterizing the effects of stellar noise sources on stellar spectra with sufficient precision to allow the detection of true Earth twins from RV measurements is a challenging task. An orbiting planet will produce on the stellar spectrum a pure shift of all the spectral lines, while stellar noise will modify the shape of spectral lines. The two effects are different in nature and should be distinguishable. However, when considering an Earth-like planet orbiting a quiet star like the Sun, both the planetary and stellar signals will induce shifts or shape variations of spectral line smaller than a thousand of a detector pixel. Therefore, we cannot distinguish one effect from the other on single spectral lines. To get around the problem, we need to average thousands of spectral lines to construct observables that are sensitive to stellar noise. However this is extremely challenging as this noise affects each spectral line differently.

So far, it was only possible to obtain precise, frequent, full-disk RV measurements derived from the full optical spectrum for stars other than the Sun. However, the surfaces of other stars are not resolvable; and therefore all the information used to study surface inhomogeneities is indirect, including: photometric flux (e.g., Aigrain et al. 2012; Boisse et al. 2009), bisector of spectral lines or of the cross correlation function depending on the activity level of the star (e.g., Dumusque et al. 2014a; Figueira et al. 2013; Queloz et al. 2001; Vogt et al. 1987), or the calcium chromospheric activity index (Noyes et al. 1984). Because it is extremely difficult to infer from such indirect observables the size, location, and contrast of surface inhomogeneities, understanding in detail the RV perturbations produced by these features is a major challenge.

As a way forward, we note that we might better understand the effect of surface inhomogeneities on RVs by obtaining precise full-disk RV measurements of the Sun. This approach of observing the Sun as a star allows us to directly correlate any change in surface inhomogeneities observed by solar satellites like the Solar Dynamics Observatory (SDO, Pesnell et al. 2012) with variations in the full-disk RV. Thus, we built a small solar telescope to feed a full-disk unresolved image of the Sun into the HARPS-N spectrograph located at the Telescopio Nazionale Galileo at the Observatorio del Roque de los Muchachos, Spain. We plan to use this instrument to observe the Sun several hours every clear day over the next two to three years, employing the exquisite long-term sub - m s${}^{-1}$ precision of HARPS-N calibrated by an astro-comb. Note that this project is not the only one to feed full-disk sunlight into a high-precision spectrograph (Strassmeier et al. 2015; Probst et al. 2015; Kjeldsen et al. 2008). However, our setup observes the Sun using exactly the same instrumental configuration as for other HARPS-N targets, which allows for a direct comparison between solar and stellar spectra.

The solar telescope is a simple instrument built from off-the-shelf components. Details of this instrument and its application to solar observations will be provided in a future publication (Glenday et al., in prep.). Here we provide a brief summary. The solar telescope consists of a $3^{\prime \prime }$ $f=200$ mm achromatic lens, which feeds an integrating sphere to scramble the solar disk. A guide camera mounted on the telescope frame provides sufficient guiding to keep the 2 mm solar image centered on a 6 mm input aperture. These components are mounted on an amateur telescope tracking mount following the Sun. The integrating sphere couples to a 300 micron core optical fiber which is fed to the HARPS-N spectrograph via its calibration unit. Light from the Sun is coupled out of the integrating sphere to HARPS-N with a uniformity of better than ${10}^{-4}$ as measured both in the laboratory and on-sky. This good uniformity is crucial to average out efficiently the 4  km s${}^{-1}$ RV shift that can be seen between the blueshifted approaching limb and the redshifted receding limb of the Sun. Given these 4  km s${}^{-1}$ , a uniformity better than ${10}^{-4}$ enables a quantitative comparison of disk-integrated RV below the 10 cm s${}^{-1}$ level.

The throughput of the solar telescope allows for exposures as short as 20 seconds. As described above, to reduce the effects of the solar 5-minute oscillations, we typically expose for five minutes to average over one such oscillation period. However, this instrument can also be used in its 20 second exposure time mode ($\approx 45$ seconds including readout) where the five minute oscillation signal is easily observable (Glenday et al., in prep.). Such a mode may prove useful for asteroseismology studies, as well as testing the effectiveness of different exposure times and observing strategies for averaging out oscillation and granulation jitter.

First tests of the HARPS-N solar telescope show sub - m s${}^{-1}$ RV precision over a week timescale. More extensive RV measurements with the same quality, combined with the wealth of information available from solar satellites, should allow us to develop a better understanding of the RV variation induced by solar surface inhomogeneities. We expect this will lead to better correction techniques for stellar noise applicable to bright stellar RV targets, and eventually will enable the RV detection of Venus, thus demonstrating the possibility of detecting Earth twins around other stars using Doppler spectroscopy.