LCO's Network of Robotic Echelle Spectrographs (NRES) will be six identical high-resolution (R~53,000), precise (≤ 3 m/s), optical (380-860 nm) echelle spectrographs, each fiber-fed (2.58" per fiber width) simultaneously by two 1 meter telescopes and a ThAr calibration source. Thus, NRES is to be a single, globally-distributed observing facility, composed of up to six units (each at a different site), using up to twelve 1-m telescopes.
We have received considerable support for NRES through NSF MRI (AST-1229720) and ATI (1407666 & 1508464) grants. The first spectrograph was installed at our Chilean site in March 2017. Additional spectrographs will be commissioned in Q4 of 2017.
Like all instruments at LCO, scheduling, observing, and data reduction are autonomous, and all data will be publicly available after a proprietary period.
The primary motivation for NRES is to study exoplanets. NRES will roughly double the radial velocity (RV) planet-vetting capacity nationwide and will achieve accuracy better than 3 m/s in reasonable exposure times for stars brighter than V = 12, enabling a large variety of planetary studies.
Spectroscopic Vetting of Transiting-Planet Candidates
Ground-based transiting planet searches produce hundreds of transiting planet candidates per year, but many of these are “astrophysical false positives” – binary or multiple stars masquerading as planets. In space, the CoRoT (Baglin et al., 2006) and Kepler (Borucki et al., 2003) missions have located thousands of candidates, but they also produced a signiﬁcant fraction of false positives. Looking forward, the Transiting Exoplanet Survey Satellite, TESS (Ricker et al., 2010), is now undergoing a NASA Phase-A study. If TESS is successful, then starting in 2017 it will produce an all-sky catalog containing an estimated 20,000 planet candidates circling bright (V ≲ 12.5) stars in the course of a 3-year mission.
Separating planets from false positives is efﬁciently done with (and often demands) RV measurements, to distinguish the reﬂex velocities of planets (typically m/s) from the velocities of stellar binaries (typically km/s). Moreover, knowing the mass, radius, and temperature of an exoplanet depends on knowing the same physical properties of its parent star. This requires spectroscopic classiﬁcation of the star to yield Teff, log g, metallicity, and v sin i. Accurate classiﬁcation is also necessary to exclude blended false positive scenarios (Torres et al., 2011). The problem of limited high-accuracy RV data was severe even before Kepler was launched. Kepler has made it a crisis, one that will become paralyzing if TESS ﬂies, unless there is signiﬁcant investment in these essential resources. Accordingly, NRES will provide an efﬁcient, uniform, automated source of spectra for stellar classiﬁcation and RV vetting, with a capacity matching the anticipated rate of planet candidate discoveries.
Precise RV Follow-up of Known Planets
Once planets (transiting or not) have been detected, continued precise spectroscopic follow-up is needed if the mass and orbital properties of the planets are to be known. Because multi-planet systems are common (e.g., Marcy et al., 2008
; Lissauer et al., 2011
), it is very desirable to maintain long-term RV monitoring of systems with known planets, to characterize any planetary siblings. NRES will support a program of long-term RV monitoring of a large selected sample of planet-bearing stars, with an accuracy adequate to study Neptune-mass planets in small orbits.
RV jitter: physical processes and possible amelioration [should we take this out?]
Many stars have higher-than-solar levels of magnetic activity (Basri et al., 2010
). Planets orbiting such stars are difﬁcult to follow up in the usual ways because their hosts suffer from radial velocity “jitter”. It may be possible to improve this situation by combining conventional RV data with diagnostics of stellar spots (e.g., Ca II K-line core intensity, Doppler imaging of starspots, or simultaneous multicolor photometry). To date, several workers have investigated this question, but with only modest success (e.g., Saar and Fischer, 2000
). Repeated observations with a relatively high time cadence (several observations per day) might however provide a point of entry. To enable such efforts, NRES will provide one-shot spectrum coverage from the Ca II H and K lines to the Ca II IR triplet (i.e., 380-860 nm). Such capability would be a direct response to the 2010 Decadal Survey
(Committee for a Decadal Survey of Astronomy and Astrophysics; National Research Council, 2010) mandate to pursue “...the aggressive development of ground-based high-precision radial velocity surveys of nearby stars at optical and near-infrared wavelengths (including efforts to determine the effects of stellar activity on these measurements).”
Each NRES spectrograph is fiber-fed simultaneously by two 1-meter telescopes and a thorium argon (ThAr) calibration source. The fiber feeds are mounted on their respective telescopes to provide on-axis guiding and to direct calibration light into "star" fibers. The fibers have octagonal cores for improved illumination stability (Brown et al. 1990, Bouchy, et al. 2013); they are 67 µm edge-to-edge, corresponding to 2.77 arcsec on the sky when fed with a reimaged f/5 beam from the telescope. The spectrograph’s optical design, produced by Dr. Stuart Barnes, gives almost full wavelength coverage between 380 nm and 860 nm, with typical spectral resolution of 48,000, and 4.47 pixels per fiber width. The optical/mechanical layout of the NRES spectrograph is shown in the photo below. The design uses an R4 echelle grating (41.6 l/mm, size 165 x 320 mm, from Newport) as the main dispersing element, with cross-dispersion from a 55-degree prism of PBM2Y glass. The 5-element all-refracting lens system serves as both collimator and camera. The science detector is a Fairchild 486 4Kx4K CCD, with 15um pixels. The CCD controller is a minor modification to a device designed and built at LCO for use in our ``Sinistro'' imaging instrument (Brown, et al. 2013). The spectrograph is bench-mounted in a temperature- and pressure-controlled chamber, itself located in an air-conditioned room, nicknamed the "Igloo", adjacent to the 1-m telescope domes. The average telescope-spectrograph fiber run is approximately 25 m.
Above: NRES in a clean room at LCO headquarters. The cylinder on the left is the exposure meter. The larger, ribbed cylinder in the center is the collimator for the CCD camera. The CCD controller is the red cube near the top of the photo. The echelle grating is inside the black box in the lower-right corner.
We adopted an optical design that is both simple and conventional in its general approach, being similar in concept to spectrographs designed for the Palomar East Arm Echelle (Libbrecht and Peri, 1995
), the Lick Automated Planet Finder (Radovan et al., 2010
), the Carnegie Planet Finder Spectrograph (Crane et al., 2006
), SOPHIE (Perruchot et al., 2008
), and the McDonald Observatory Sandiford Spectrograph (McCarthy et al., 1993
We applied the design philosophy of these instruments to our 1-m feeds, with the aim of achieving very high optical throughput, wide wavelength coverage, and simultaneous ﬁber input from two telescopes. The spectrograph will have only one moving part – the shutter – inside its environmental chamber. By eliminating mechanisms necessary to adjust optics, we simplify the design, and (more importantly) assure a system that has great intrinsic stability.
Above: a spectrum from the NRES prototype, installed on LCO's 0.8m telescope in the Sedgwick Reserve. The solid lines are the dispersed spectrum of HD 61421. The light blips beneath the solid lines are emission lines from the ThAr lamp. From left to right and top to bottom (like a book), the spectrum goes from blue to red.
The CCD detectors used for NRES will come from LCO’s stock of Fairchild 486 devices. These have 4K x 4K format with 15-micron square pixels, and are thinned and backside-illuminated with broadband antireﬂection coatings. Their quantum efﬁciency is 80% at 400 nm, peaks at 91% at 550 nm, and falls to 55% at 860 nm. The planned 62.5µm input ﬁbers will be over-resolved by a factor of 4.15. Oversampling that is very desirable for precise radial velocity measurement.
To operate these detectors, LCO has developed in-house an innovative and ﬂexible controller (Tufts et al., 2008
). This controller allows a wide range of formatting and readout speeds and other options, and adds almost no noise to the intrinsic CCD read noise. With it, we expect to achieve a read noise of about 7 e-/pixel at 1 MHz readout rate, requiring 16 s to read the full format. This is the same detector/controller that LCOGT uses for the standard imagers for our 1-m network.
Variations in spectrograph temperature and in barometric pressure cause spurious Doppler shifts, which must be calibrated out. Since calibration works better if the instrumental shifts are small to begin with, we keep the spectrograph wavelength scale constant within 30 m/s before calibration. We place the spectrograph in an environmental chamber that maintains temperature stability of 0.01 C and constant barometric pressure within 0.2 mbar. These requirements are echoed in many spectrographs that aim to achieve similar precision to NRES. We elected to maintain the spectrograph environmental chamber at near-atmospheric pressure, using active control. We chose this solution over a vacuum one because it allows a lighter, less expensive, structure and because the control mechanisms for the needed accuracy have already been tested to be relatively inexpensive and reliable. We describe the thermal control system below.
A multi-layer thermal control system is commonly used in high resolution spectrograph design. NRES’s environmental chamber consists of two nested aluminum shells, separated by an insulated gap. The outer shell insulated from the ambient air. The inner shell is relatively massive, sealed, and pressure-controlled. Its function is to hold the air inside at nearly-constant pressure (removing pressure variations from diurnal and weather-related processes) and to provide a long thermal time constant. This assures that the spectrograph doesn't undergo rapid temperature changes. The inner shell carries high-precision temperature sensors for monitoring performance, but it is not actively heated or cooled. The outer shell is a thin aluminium structure, also with multiple temperature sensors, but having heating elements bonded to it. The power to these elements is servo-controlled to keep the outer shell as nearly as possible at a constant temperature. We set the temperature 2 C above ambient, so that cooling is never required to maintain the desired outer box temperature.
Our acquisition software uses astrometry.net
(Lang et al., 2010
) to solve the image coordinates robustly in ≲ 3 seconds. Pointing is adjusted iteratively to place a target onto a known pixel for efﬁcient, robotic, spectroscopic acquisition. The analysis pipeline consists of modules for image calibration, spectrum extraction, ﬂux and wavelength calibration, radial velocity determination, stellar classiﬁcation, quality assurance, and interaction with the data archive. Extracted spectra and analysis results are stored in LCO science archive.
Given the scientific motivation for NRES, the prime ﬁgure-of-merit is how many distinct RV observations per hour of wall-clock time the facility can obtain at given precision and target star magnitude. The figure below compares this ﬁgure-of-merit for NRES (solid blue) with Keck/HIRES (dashed red). The figure refers to random errors (precision), not systematic ones (accuracy). The different lines represent a variety of Doppler precisions (1, 3, 10, and 30 m/s), labeled at the intersection between the red and blue lines, i.e. where NRES and Keck/HIRES are equally productive. To the left of these intersections (brighter stars), NRES is more efﬁcient than Keck/HIRES for the speciﬁed RV precision.
The startling conclusion from this figure is that even for RV precision as good as a few m/s, NRES outperforms Keck/HIRES for stars as faint as V = 12. And with a (30 m/s) precision sufﬁcient to detect Hot Jupiters, NRES can produce RV samples faster than Keck/HIRES to V = 15! Given that Keck has 8.33 times more collecting area than all twelve 1-meter telescopes combined, this behavior is surprising. The superior performance of NRES arises from three sources: 1. the number of telescopes, 2. optical efﬁciency, and 3. operational efﬁciency. To illustrate the latter two, consider stars with V = 9.5, with a desired precision of 3 m/s, indicated with dots on the above figure. In this parameter range, both Keck and NRES are limited by photon noise; detector read noise is negligible and noise from p-modes is unimportant. The ﬁgure-of-merit thus transforms simply to the number of detected photons per hour of wall-clock time. The optical efﬁciency of NRES will be higher than Keck/HIRES primarily because NRES will calibrate spectra using a ﬁber connected to a ThAr lamp, as opposed to the I2 absorption cell used on HIRES. This increases the usable wavelength range by a factor of about 3, avoids I2 pseudo-continuum absorption, and gives useful data at blue wavelengths where the density of absorption lines is large. As conﬁrmed by the HARPS spectrograph, the combined efﬁciency gain is about a factor of 6 (Pepe et al., 2003
). Also, the NRES optics have fewer surfaces, use prism (not grating) cross-dispersion, and have smaller “slit” losses, making them more transmissive than HIRES by a factor of about 2. Finally, at 3 m/s, the needed integration times are 8 min for NRES, but 1 min for Keck/HIRES. To observe many targets with a typical 1-min slew-settle time, the larger telescope loses a factor of 2 in duty cycle, whereas NRES loses only 6%. Combining these factors gives NRES the advantage by nearly a factor of 3. Note that a comparison between NRES and HARPS (which uses ﬁber/ThAr calibration) would be almost identical. Although the total aperture of NRES is about equal to that of HARPS, NRES also gains from smaller overheads and a simpliﬁed, prism-based design.
Above: Simulated performance of the NRES spectrograph as a function of stellar magnitude, for an integration time of 20 minutes. (a) S/N for three useful wavelength regimes, including for the Mt. Wilson S-index. (b) RV precision determined from photon plus read noise, for a variety of stellar types, assuming vsini = 2 km/s.
The figure on the left shows that in 20 min, NRES will provide classiﬁcations at S/N = 10 even below V = 14, corresponding to fairly faint Kepler targets. For V ≤ 12, which will cover all of the stars surveyed by TESS, S/N = 30 classiﬁcation is feasible. To classify the expected 20,000 TESS stars that will show transit-like signals requires about 6700 hours of 1-meter telescope time. Over a 3-year period, this is about 7% of the total observing time available to the network. For planet candidate vetting at vsini = 2 km/s for the same stars, RV measurements with accuracy of 30 m/s will need less than 8 min of observing time per star, or about 3% of the total network time. Vetting at V=14 will be possible, but slower. Higher accuracy of 3 m/s will be attainable in an hour for V = 12. For a plausibly-scaled program to measure orbital parameters, devoting 9% of the 1-m network observing time, NRES will provide 3 m/s Doppler accuracy for 500 stars, with 20 observing epochs per star over 3 years. (Of course, faster-rotating or lower-metallicity stars will take longer.) Performance at this level will be adequate for a large majority of the expected transiting planet follow-up; only the faintest and most difﬁcult targets, or those needing precision better than 3 m/s, will require observations from larger telescopes.
For favorable cases, the RM signal can exceed 50 m/s, while the smallest signal yet detected is about 1.5 m/s, corresponding to a Neptune-size body transiting the slow-rotating star HAT-P-11 (Winn et al., 2010
). To properly sample a RM radial velocity curve requires time resolution comparable to transit ingress/egress times – typically 5 minutes. Results from multiple transits or multiple telescopes may however be averaged together. We therefore target 3 m/s precision in 4 x 5-min observations on a typical star with V = 10. For large planets circling stars brighter than V ∼ 12 and vsini > 2 km/s, NRES will give RM data with usefully low noise within a single transit. For fainter stars or smaller planets, one may combine data from multiple telescopes and transits. For instance, the HAT-P-11b results by Winn et al. (2010
) based on 2 transits observed with Keck/HIRES could be reproduced by using pairs of telescopes (attached to the same spectrograph), and combining results from 4 transits. Because of LCO’s multiple sites, large number of telescopes, and ﬂexible scheduling, such coordinated observing strategies will be routine.
Ca II studies of stellar magnetic activity are promising. Observations of the S index (Vaughan et al., 1978
) that are precise to 1% or so should be possible in 20 minutes down to V = 7, and with 10% precision to V = 12. With NRES, one could for follow the entire target list of the Lowell stellar activity survey (147 stars, most brighter than V = 7.5; Hall et al., 2007
) with weekly activity measurements, using about 6% of the network capacity.
See the NRES status report (updated 1 June 2017).