CARMENES GTO: M-dwarf Exoplanet Survey

• CARMENES GTO is a high-precision, long-baseline radial-velocity survey of M dwarfs using a dual-channel spectrograph to detect low-mass exoplanets.
• It employs state-of-the-art optical and near-infrared spectroscopy with rigorous calibration and optimized scheduling for improved measurement accuracy.
• The survey provides comprehensive stellar activity diagnostics and public data releases, advancing our understanding of exoplanet occurrence and stellar properties.

Searching arXiv for CARMENES Guaranteed Time Observations and related survey papers. The CARMENES Guaranteed Time Observations (GTO) program is a long-baseline radial-velocity survey of nearby M dwarfs conducted with the dual-channel CARMENES spectrograph on the 3.5 m telescope at Calar Alto. Its primary goal is a high-precision search for low-mass planets, including super-Earths and Neptunes in or near the liquid-water habitable zone, through stabilized optical and near-infrared spectroscopy, homogeneous target monitoring, and activity diagnostics designed to distinguish Keplerian signals from stellar variability (Ribas et al., 2023).

1. Survey definition, scope, and target sample

CARMENES was conceived to deliver high-accuracy radial velocity measurements with long-term stability for temperate rocky-planet searches around nearby cool stars. In the GTO context, the survey ran from 1 Jan 2016 to 31 Dec 2020 with 750 nights reserved, and DR1 makes public all observations obtained during that interval: 19,633 spectra for 362 targets, of which 19,161 passed quality flags and 18,642 drift-corrected RVs were derived for 345 single-line targets (Ribas et al., 2023).

The target population is explicitly M-dwarf focused. Pre-survey catalog work described a two-step selection based on the Carmencita catalogue of approximately 2,200 M stars observable from Calar Alto and among the brightest of each spectral subtype, from which a high-priority GTO list of approximately 300 targets was drawn. The stated requirements included spectral types M0.0 V–M9.0 V, J-band brightness $J \le 11.5$ mag, singleness, and, whenever possible, low magnetic activity and slow rotation (Jeffers et al., 2018). DR1 characterizes the realized core sample somewhat differently: 362 M dwarfs with $\delta > -23^\circ$, selected to minimize biases in age, metallicity, and activity, with cuts imposed only on brightness and the exclusion of visual binaries with separations $<5''$; about 70% of all known M dwarfs within 10 pc and accessible from Calar Alto were included, corresponding to 48% of all known M dwarfs at 10 pc and 15% completeness at 20 pc (Ribas et al., 2023).

A common simplification is that the survey targeted only magnetically quiet stars. The literature is more nuanced. The input-catalog program emphasized low activity and slow rotation whenever possible (Jeffers et al., 2018), whereas DR1 stresses an approximately volume-limited, spectrally uniform sample that avoided preselection of “quiet” or “planet-hosting” stars (Ribas et al., 2023). This suggests a practical compromise between RV-precision optimization and broad census design.

Different CARMENES papers report different working samples because they address distinct phases of the program. Scheduling simulations used about 309 M dwarfs (Garcia-Piquer et al., 2017); the occurrence-rate study analyzed 329 nearby, bright M dwarfs and a high-cadence subsample of 71 stars (Sabotta et al., 2021); DR1 reports 362 targets (Ribas et al., 2023); and later wavelength-dependence work refers to more than 20,000 observations of some 350 nearby stars spanning more than eight years (Jeffers et al., 29 Oct 2025).

2. Instrument architecture and observing workflow

CARMENES is a fiber-fed, stabilized echelle spectrograph permanently mounted on the 3.5 m telescope at Calar Alto. It comprises two arms: a VIS channel covering $520\text{–}960$ nm at $R \approx 94\,600$, and a NIR channel covering $960\text{–}1710$ nm at $R \approx 80\,400$. Both channels are temperature- and pressure-stabilized to sub-m s$^{-1}$ long-term drift levels and use Fabry–Pérot etalons together with hollow-cathode lamps for wavelength calibration (Ribas et al., 2023).

The broad wavelength coverage is central to the program’s design. For bright early- and mid-M stars at $S/N=150$, the internal RV precision reaches approximately $1$ m s$\delta > -23^\circ$0, whereas for late-M targets it is $\delta > -23^\circ$1 m s$\delta > -23^\circ$2 (Ribas et al., 2023). In the first-planet discovery paper, the VIS channel achieved a median internal RV precision of $\delta > -23^\circ$3 m s$\delta > -23^\circ$4 for HD 147379, while the NIR channel yielded $\delta > -23^\circ$5 m s$\delta > -23^\circ$6, reflecting the lower NIR efficiency for early M dwarfs (Reiners et al., 2017). Later full-GTO analysis states that the VIS channel contains the bulk of the RV information, while the NIR arm initially suffered thermo-mechanical instabilities and delivered $\delta > -23^\circ$7 m s$\delta > -23^\circ$8 RVs, with upgrades in 2023 improving its performance (Jeffers et al., 29 Oct 2025).

The observing cadence is survey-like rather than target-of-opportunity by default. The GTO strategy typically used 1–3 visits per star per month, with higher cadence for promising targets, and integration times were set to achieve $\delta > -23^\circ$9 in the J band per visit (Reiners et al., 2017). In the 71-star occurrence-rate subsample, the effective cadence was roughly 10–20 visits per star per season, spanning periods from approximately 1 d up to about half the time baseline (Sabotta et al., 2021). DR1 reports a median of 30 visits per target, against a goal of at least 50 (Ribas et al., 2023).

Reduction and RV extraction are pipeline-based. The raw spectra are processed with caracal for extraction and wavelength calibration, while precise RVs are derived with serval template matching; raccoon provides cross-correlation-function indicators (Ribas et al., 2023). In the first discovery paper, the basic reduction steps are bias subtraction, flat-fielding, order extraction, and scattered-light removal; the nightly wavelength solution uses ThAr lamps and Fabry–Pérot etalons; and RV measurement proceeds in parallel through serval and a CCF method based on a weighted binary mask of approximately 3000 lines (Reiners et al., 2017).

3. Scheduling, cadence control, and operational optimization

Because the GTO survey comprises hundreds of targets over hundreds of nights, scheduling became a formal optimization problem rather than a purely manual process. The CAST scheduler models the campaign over a horizon of $<5''$0 nights and approximately 309 M dwarfs, with long-term variables $<5''$1 for target-night assignment, mid-term variables $<5''$2 for nightly start times, and a short-term repair heuristic for reactive execution (Garcia-Piquer et al., 2017).

Hard constraints include night-window bounds, minimum elevation, Moon separation, dome aperture and pointing limitations, continuous visibility duration, overlap and overhead handling, and environmental conditions such as humidity, wind, and temperature. Soft constraints are to maximize weighted observing time, equalize the number of observations among targets of equal priority, and optionally enforce cadence per target (Garcia-Piquer et al., 2017). The long-term and mid-term planners are both implemented with NSGA-II, using non-dominated sorting and crowding distance, with population size $<5''$3, initial population $<5''$4, 1000 generations, selection probability $<5''$5, crossover $<5''$6, and mutation $<5''$7 (Garcia-Piquer et al., 2017).

The mid-term objective function explicitly weights observations by priority and meridian proximity,

$<5''$8

while the short-term scheduler fills gaps by ranking feasible unscheduled targets according to five heuristics: fewest observations tonight, not already in the mid-term plan, highest priority, fewest total observations so far, and closest to the meridian (Garcia-Piquer et al., 2017).

Simulation results quantify the operational impact. From 50 independent trials of a 3-year survey, good-weather time was approximately 60% of total night hours, or about 6400 h. CAST used $<5''$9 of available time, corresponding to about 6342 h, of which $520\text{–}960$0 were science exposures and $520\text{–}960$1 were overheads. The simulations produced approximately $520\text{–}960$2 observations, or $520\text{–}960$3 visits per target, distributed uniformly across 309 targets (Garcia-Piquer et al., 2017).

The same framework was used for planet-yield simulations. Under pure photon noise, the optimized plan recovered $520\text{–}960$4 planets out of $520\text{–}960$5 generated planets, or $520\text{–}960$6 overall, and $520\text{–}960$7 out of $520\text{–}960$8 planets with $520\text{–}960$9 m s$R \approx 94\,600$0, or $R \approx 94\,600$1. With 3 m s$R \approx 94\,600$2 white stellar jitter, detection yield dropped to $R \approx 94\,600$3 planets overall and $R \approx 94\,600$4 for the $R \approx 94\,600$5 m s$R \approx 94\,600$6 subset (Garcia-Piquer et al., 2017). These figures clarify that CARMENES GTO performance depends not only on spectrograph stability but also on cadence design and stellar-noise control.

4. HD 147379 b as the first GTO detection

The first planet detected from CARMENES GTO RVs was HD 147379 b, orbiting the bright M0.0V star HD 147379 at 10.7 pc. The host star has $R \approx 94\,600$7 mag and $R \approx 94\,600$8. CARMENES observations taken between 2016 and 2017 revealed periodic RV variations with semi-amplitude $R \approx 94\,600$9 m s$960\text{–}1710$0 and period $960\text{–}1710$1 d; the signal is supported by HIRES/Keck observations obtained since 2000 (Reiners et al., 2017).

For this target, the CARMENES time series spans MJD 2457397–2457999, approximately 600 d, while the archival HIRES/Keck set contributes 30 points with $960\text{–}1710$2 m s$960\text{–}1710$3 (Reiners et al., 2017). Periodogram analysis of the CARMENES RVs shows a strong peak at $960\text{–}1710$4 d$960\text{–}1710$5, corresponding to $960\text{–}1710$6 d, and the combined data yield $960\text{–}1710$7, with false-alarm probability $960\text{–}1710$8 (Reiners et al., 2017). The Keplerian model was optimized with Nelder–Mead and uncertainty estimation with emcee; a single-planet plus jitter model was favored (Reiners et al., 2017).

The derived orbital parameters are $960\text{–}1710$9 m s$R \approx 80\,400$0, $R \approx 80\,400$1 d, and $R \approx 80\,400$2, with $R \approx 80\,400$3 at $R \approx 80\,400$4. Kepler’s third law gives

$R \approx 80\,400$5

leading to $R \approx 80\,400$6 au, and the RV semi-amplitude implies $R \approx 80\,400$7, approximately $R \approx 80\,400$8 (Reiners et al., 2017). The planet orbits inside the star’s temperate zone, where water could exist in liquid form (Reiners et al., 2017).

The detection is also a methodological demonstration of how CARMENES GTO treats stellar activity. The RVs and spectroscopic indicators show additional variability near 21.1 d and its first harmonic, interpreted as the stellar rotation period. A secondary peak at $R \approx 80\,400$9 d$^{-1}$0 appears in the RVs and in dLw, but no power at 86.5 d is found in any of the six activity indicators, which supports the planetary interpretation of the longer-period signal (Reiners et al., 2017).

5. Statistical yield, public release, and occurrence-rate inferences

DR1 places the first-planet result within a much broader survey context. By 2023, CARMENES GTO data products comprised 18,642 precise RVs for 345 targets, with time series of spectroscopic activity indicators accompanying the velocities. The public release contextualizes the exoplanet output as 33 new planets from the blind survey, 17 re-analysed known planets, and 26 confirmed planets from transiting-candidate follow-up (Ribas et al., 2023).

The occurrence-rate analyses proceed from injection-and-retrieval on pre-whitened RV time series. In the 71-star study, each star’s data were cleaned of known Keplerians and activity trends, then populated with synthetic circular orbits on a $^{-1}$1 log-uniform grid in $^{-1}$2, with detection defined by recovery of the injected period in a generalized Lomb–Scargle periodogram at $^{-1}$3 (Sabotta et al., 2021). The per-star detection probability map is

$^{-1}$4

For that 71-star subsample, 27 planets in 21 systems met the uniform detection-limit criteria. Summed over $^{-1}$5 d, the inferred occurrence rates are $^{-1}$6 giant planets $^{-1}$7 per star, $^{-1}$8 intermediate-mass planets $^{-1}$9 per star, and $S/N=150$0 low-mass planets $S/N=150$1 per star; the upper limit for hot Jupiters is $S/N=150$2 planets per star (Sabotta et al., 2021). The same study reports a stellar-mass dependence: for $S/N=150$3, planets more massive than $S/N=150$4 become rare, while low-mass planets with periods shorter than 10 d are overabundant (Sabotta et al., 2021).

DR1 extends occurrence-rate estimation to a much larger well-observed subsample of 238 targets. Over $S/N=150$5 and $S/N=150$6 d $S/N=150$7 d, the average occurrence is

$S/N=150$8

planets per star, and the fraction of stars with at least one planet in that domain is $S/N=150$9 (Ribas et al., 2023). The accompanying interpretation in DR1 is that nearly every M dwarf hosts at least one planet (Ribas et al., 2023).

The occurrence-rate literature also identifies an important bias issue. The 71-star sample was selected for having at least 50 observations and is explicitly skewed toward stars with deeper monitoring, higher mass, and lower activity; the giant-planet occurrence rate in that subsample could therefore be up to a factor of five higher than in the full 329-star GTO sample (Sabotta et al., 2021). This is not a contradiction with DR1’s broader occurrence estimate; rather, it is a reminder that completeness and target-selection functions must be matched to the question being asked.

6. Activity diagnostics, wavelength dependence, and legacy applications

Stellar magnetic activity is a central limitation for the GTO survey because spots, plages, and chromospheric inhomogeneities can induce apparent Doppler shifts of a few m s$1$0, comparable to or larger than the signals of low-mass planets (Perdelwitz et al., 2021). CARMENES addresses this both internally, through multiwavelength activity indicators, and externally, through complementary archives for diagnostics unavailable in the instrument bandpass.

Internally, the survey derives the chromatic index (CRX), differential line width (dLW), line-profile bisector slope (BIS), CCF FWHM and contrast, and chromospheric tracers including H$1$1, Ca II IRT, and Na I D (Ribas et al., 2023). In the later full-GTO wavelength-dependence analysis, CRX is defined as the slope of RV with logarithmic wavelength,

$1$2

implemented through a fit of

$1$3

across spectral orders (Jeffers et al., 29 Oct 2025). The physical premise is that activity-induced signals are chromatic, whereas Keplerian reflex motion is grey (Jeffers et al., 29 Oct 2025).

Using stars with at least $1$4 VIS measurements, the full-GTO study found that approximately 17% of GTO stars show a strong or moderate CRX–RV correlation. Thirty-nine stars define a significant “CRX-all” sample, with 13 forming a visually tight, approximately linear anti-correlation subset. Correcting velocities through

$1$5

reduces the RV rms by factors up to 3.9, and by $1$6 in nine stars (Jeffers et al., 29 Oct 2025). The best improvements occur for mid M dwarfs with moderate projected rotational velocities, moderate CRX gradients, and quasi-stable activity features (Jeffers et al., 29 Oct 2025).

Externally, Ca II H&K activity monitoring is supplied by archival spectroscopy because CARMENES does not cover the blue wavelengths of those lines. A dedicated catalog assembled 11,634 archival spectra of 186 GTO M dwarfs from ESPADONS, FEROS, HARPS, HIRES, NARVAL, TIGRE, and UVES, and derived homogeneous $1$7 time series through rectification against PHOENIX synthetic spectra (Perdelwitz et al., 2021). These series provide rotation-period and magnetic-cycle diagnostics on baselines up to approximately 15–20 yr, and the paper reports tentative discovery of three previously unknown activity cycles, including a $1$8 d cycle for Luyten’s Star and a $1$9 d cycle for HD 216899 (Perdelwitz et al., 2021).

Beyond exoplanet detection, DR1 emphasizes a broader scientific legacy. The high-resolution spectra and time series support stellar parameter determination, chemical abundances and Galactic kinematics, activity characterization, transmission spectroscopy of exoplanet atmospheres, and Rossiter–McLaughlin measurements (Ribas et al., 2023). This broader use is a direct consequence of the GTO design choice to collect a uniform, high-resolution, multi-year spectroscopic data set rather than only a narrowly optimized set of planet-search velocities.