A Reanalysis of the VLT2/UVES M Dwarf Planet Search, 2000-2007

The ESO VLT telescope complex, Paranal, Chile
The ESO VLT telescope complex, Paranal, ChileCredit: ESO/H.H.Heyer


When the UVES M Dwarf Planet Search program began in March 2000, fewer than 50 giant exoplanets were known, only one of which was hosted by an M dwarf. There were no known terrestrial mass planets. Super-Earths were unknown.

Compared to solar-type stars, terrestrial mass and potentially habitable planets around M dwarfs are much easier to detect. The lower mass of the host star yields a larger Doppler velocity amplitude. Due to the lower luminosity of M dwarfs, the potentially habitable zone is much closer to the host star. This further increases the Doppler velocity amplitude, and allows many orbits to be observed in a relatively short period of time. The primary problem with observing M dwarfs is their intrinsic faintness.

The UVES M Dwarf team (Kürster et al. 2003) presciently combined the advantages of the large aperture VLT-UT2 telescope (ESO Paranal Observatory, Chile) with the spectacular new UVES spectrometer (Dekker, et al. 2000), a cross dispersed echelle with a resolution of 130K, to survey 40 nearby M dwarfs.

The decision by the UVES M Dwarf Planet team to chase potentially habitable planets with an untested instrument was courageous. They would need to push velocity precision to the 1 m/s level in an era when state-of-the-art precision ranged from 3 to 10 m/s.

While no planets were found in this initial study (Zechmeister, Kürster, & Endl 2009; hereafter ZKE2009), it represents a ground breaking effort, combining a bold vision with state-of-the-art instrumentation.

A Reanalysis of the UVES Data

As part of the "Pale Red Dot" program we reanalyzed the UVES data for proxima Centauri in 2016, starting with the reduced 1-D spectra from the UVES M Dwarf team. We were able to reduced the velocity scatter (RMS) from 3.6 to 2.3 m/s. This improved UVES data showed a dominant periodicity of 11.2 days, and confirmed the signal found in the HARPS data. The resulting planet is ~1.3 earth-masses, has a velocity semi-amplitude of 1.4 m/s (Anglada-Escude et al. 2016, Nature Cover).

Motivated by this result, we harvested all of the raw data from the UVES M Dwarf Planet Search from the ESO archives and have written custom packages to generate 1-D spectra from the raw data, and velocities from the 1-D spectra. The median improvement in the velocity RMS from the new analysis is 1.8 m/s. Six of the 38 M dwarfs from the original study had a velocity RMS < 4 m/s. In the reanalysis presented here, 22 of these stars have velocity RMS < 4 m/s.

The 35 most stable stars from this study are shown in Figure 1 from ZKE2009. We have reproduced this figure with the newly reanalyzed data (Figure 3 from Butler et al. 2019).

The 35 most stable stars from ZKE2009

Figure 1 from ZKE2009.
Figure 3 from Butler et al. 2019.

The most obvious difference between the ZKE2009 velocities and our reanalysis are the stars with significant proper motion, notably Barnard's star (GL 699). The barycentric corrections computed in ZKE2009 do not include proper motion. Each star is assumed to have fixed coordinates. The linear fit to the high proper motion stars in ZKE2009 are from a model of the secular acceleration due to the stellar proper motion. Our barycentric correction code includes the stellar proper motion information. The stellar coordinates are advanced by the proper motion. Our final barycentric correction includes the effect of proper motion, so the resulting velocities do not show the effect of secular acceleration.

The table to the right shows a comparison of the velocities from ZKE2009 and our reanalysis. The columns are:

1) Star Name

2) Star Name from ZKE2009

3) stellar spectral type

4) star V magnitude

5) number of observations

6) velocity RMS from ZKE2009

7) velocity RMS from our reanalysis

8) the quadrature difference of columns 6 and 7

Reanalysis of the UVES M Dwarf planet search program

The first 33 stars have a velocity RMS < 6 m/s, and are deemed "stable". The median improvement in the velocity RMS from the reanalysis is 1.8 m/s.

If we assume that sources for velocity scatter add in quadrature, then the final velocity RMS for any given star is given by:

Where J is the contribution from stellar jitter, P is the contribution from orbiting planets, N is the contribution from photon noise, and R is the contribution from the velocity reduction package.

For each star in this program, the same data has now been analyzed by independent raw reduction and velocity reduction packages. For each star the velocity RMS contributions from stellar jitter (J), orbiting planets (P), and photon noise (N) is identical, only the contribution from the reduction packages differ. The reduction in the velocity scatter due to the reduction package (R) can be estimated as the quadrature difference of the velocity RMS of the original UVES data set with respect to the new reduction. This is shown in the final column of Table 1. The median reduction in ``R'', the velocity scatter due to the Doppler velocity reduction code, is 4 m/s.

The original goal of the UVES M Dwarf team was to find terrestrial mass and potentially habitable planets around the nearest M dwarfs. This goal has now been realized. In addition to the planet around proxima Cen, this data has contributed to the discoveries of a cold super-Earth around Barnard's Star (Ribas et al. 2018), and potentially habitable super-Earths around GJ180 and GJ229A (Feng et al. 2019).

There are two data sets below. The first is the complete set of velocities (uves_vels.tar). The second is the binned set of data (uves_vels_binned.tar). Often two or more consecutive observations are taken of a star. Faint stars require more than 20 minutes to achieve the necessary signal-to-noise. For these cases multiple shorter observations are taken to prevent a single exposures from going longer than 15 or 20 minutes. Longer exposures suffer from more cosmic ray hits, and they have a larger uncertainty in the photon-weighted mid-point of the observation. Multiple exposures are taken of bright stars because they would saturate the CCD detector in less than 5 minutes. We prefer that the total exposure time on a star be at least 5 to 10 minutes to average over the stellar seismology timescale. The "uves_vels_binned" file averages all the observations of a given star taken within 2 hours.

These are both "tar" files. Upon downloading them you will need to "un-tar" them. In a standard UNIX X-windows environment you can do this as follows:

UNIX> tar xvf uves_vels.tar

The data is written as standard ascii (text) files designed to be plugged directly into the Systemic Planet Fitting package. A tutorial on running Systemic can be found at https://www.stefanom.io/projects.

As an example, we show the first 5 observations for GJ551. The columns are:

If you use any of this data, please cite our paper:

Butler, Jones, Feng, Tuomi, Anglada-Escude & Keiser 2019, AJ, accepted


A Reanalysis of the UVES M Dwarf Planet Search Program - The Astronomical Journal, 158:251 (9pp), 2019 December

R. P. Butler , H. R. A. Jones, F. Feng , M. Tuomi, G. Anglada-Escudé, and Sandy Keiser

Search for Nearby Earth Analogs. II. detection of five new planets, eight planet candidates, and confirmation of three planets around nine nearby M dwarfs

Fabo Feng, R. Paul Butler, Stephen A. Shectman, Jeffrey D. Crane, Steve Vogt, John Chambers, Hugh R. A. Jones,4 Sharon Xuesong Wang, Johanna K. Teske, Jenn Burt, Matías R. Díaz, and Ian B. Thompson

Download files: