ESPRESSO@VLT:
On-Sky Performance and First Results

What ESPRESSO is designed to solve

ESPRESSO is the Echelle SPectrograph for Rocky Exoplanets and Stable Spectroscopic Observations. In 1-UT High Resolution mode, the paper reports resolving power around R ≈ 140,000 over 378.2–788.7 nm; the Ultra High Resolution mode reaches approximately R ≈ 190,000; and the 4-UT mode provides an equivalent 16 m aperture at approximately R ≈ 70,000.[1]

The performance numbers define the engineering problem. The paper reports throughput above 10% near 550 nm under nominal 0.65 arcsec seeing, radial-velocity precision better than 25 cm/s over one night, and better than 50 cm/s over several months. These values are described as limited by photon noise and stellar jitter, consistent with a 10 cm/s instrumental precision floor.[1]

Coudé train: the instrument begins before the bench

ESPRESSO receives light from the VLT Unit Telescopes through the Coudé train before the beam reaches the Combined Coudé Laboratory. The front-end module includes atmospheric-dispersion correction, field and pupil stabilisation, and calibration-light injection. The telescope-to-lab optical path is therefore part of the instrument, not merely delivery hardware.

This matters because detector motion or wavelength drift cannot automatically be assigned to the bench. It may originate in input illumination, pupil motion, fibre injection, calibration geometry, pressure and temperature changes, or a coupled combination of these effects.

Fibre feed, pseudo-slit, and scrambling

The fibre feed uses octagonal fibres and a double scrambler. The paper describes the scrambler as exchanging near-field and far-field illumination so guiding residuals and pupil wobble are suppressed before they become spectral-format changes. A stable wavelength solution is not sufficient if the light enters the spectrograph in a different spatial state from exposure to exposure.

The pseudo-slit and spectral-format figures make this visible. HR, UHR and MR modes map fibre outputs differently onto the detector. Residual near-field or far-field variation can alter line shapes, extraction weights or order-dependent centroids.

Vacuum and multi-shell thermal stability

ESPRESSO is placed inside a large vacuum vessel in the Combined Coudé Laboratory. The vacuum removes air-column refractive-index variations from the optical path, while multiple actively controlled thermal shells reduce the disturbance that reaches the optical bench. This is a cascade, not a single-loop control strategy.

The missing piece, from a control-analysis point of view, is the frequency-domain response. An RMS temperature number does not show whether remaining thermal power lives on exposure timescales, nightly timescales or seasonal timescales.

Resolving power is not a single number

The resolving-power maps are more useful than a headline R value because they show dependence on observing mode, detector position, order, slice and wavelength. For precision work, resolution should be treated as a mapped quantity across the extracted format, not as one scalar label.

Throughput and efficiency: high resolution is not free

The efficiency plots show that throughput cannot be separated from resolution and seeing. HR mode reaches more than 10% photon-detection efficiency near 550 nm, while the blue end drops substantially. UHR mode increases resolving power but makes aperture, seeing, slit loss and fibre size more restrictive. The design trade-off is therefore not “higher resolving power is always better”.

If a design chases resolution without tracking throughput, radial-velocity precision can worsen because photon noise grows faster than information content improves. Resolution, throughput, fibre diameter, seeing, detector efficiency and target type belong in the same error budget.

RV precision: short-term and long-term validation

The internal precision versus S/N plot links photon noise, spectral type and information content. The seven-hour HD 85512 time series reports approximately 28 cm/s RMS for single 300 s exposures, close to the 25 cm/s photon-noise estimate, and binning five points reduces the scatter to roughly 8 cm/s.[1]

The long-baseline τ Ceti plots show the operational reality. The paper distinguishes data before and after the June 2019 fibre upgrade, with raw scatter changing from around 0.94 m/s before the intervention to around 0.72 m/s after it, and nightly-binned post-upgrade scatter around 0.58 m/s.[1] Short-term precision and long-baseline repeatability are not the same performance claim.

Calibration strategy: ThAr, Fabry-Pérot, and the LFC boundary

ThAr lamps provide absolute wavelength anchors, while a white-light Fabry-Pérot etalon provides dense drift monitoring. The laser-frequency comb was intended to provide a more complete calibration reference, but at the time of the performance paper it was not yet fully reliable or routine.[1]

Later calibration papers sharpen this section. Schmidt et al. tested a 34 GHz EOM comb at ESPRESSO and found that the source itself was stable, while calibration-unit injection geometry could introduce local wavelength-solution systematics.[2] Kreider et al. showed that Fabry-Pérot resonators can drift chromatically, which means an FP grid cannot always be treated as a rigid ruler.[3]

Open technical questions after this reading

• Thermal transfer function: What is the frequency-domain transfer function from room to shell, vessel, bench and wavelength solution?
• Fibre-link root cause: The paper reports a fibre-link throughput deficiency that required repair. Was the dominant issue alignment, coating degradation, stress, modal behaviour, or another mechanism?
• Calibration injection repeatability: If later comb tests reveal calibration-unit injection systematics, what tolerance should be assigned to the fibre-head or switching mechanism?
• Scrambling residuals: What residual near-field or pupil variation remains at the pseudo-slit under different seeing and guiding conditions?
• 4-UT stability: Has the 4-UT mode been demonstrated over the same long-baseline RV stability window as the 1-UT mode?

Connection to EXOhSPEC

This paper gives a checklist rather than a recipe. For EXOhSPEC, the writing should state where the disturbance enters, what each control or shielding layer attenuates, and which measured optical quantity proves that the correction is working. Temperature stability alone is not the final metric; the final metric is whether the optical path, spectral format, centroid, or wavelength solution becomes more stable.

The same point changes how TEC feedback should be presented. A TEC loop should appear as one element of a stability architecture: environmental sensing, thermal actuation, optical-path monitoring, image or echellogram tracking, calibration diagnostics, and a quantified residual error budget.

Following the full technical argument

Extraction, observing modes and the information preserved in the data

The instrument description is not complete at the vacuum vessel. Acquisition, guiding, the sliced fibre images, detector sampling and the reduction chain determine which information survives into a spectrum. The raw-format figure makes that connection tangible: calibration and science traces have a specific geometry that a pipeline must model consistently.

Observing modes trade collecting area, resolving power, sampling and sensitivity. These are not cosmetic settings. A binning choice changes read-noise and sampling behaviour, while a different fibre aperture changes the seeing-dependent throughput. Comparing modes fairly requires a common science objective and an explicit exposure-time or precision metric.

A reduced spectrum should also be read as a measurement carrying atmospheric structure and processing choices. The blue and red examples contain different information and different potential contaminants. Wide wavelength coverage increases scientific leverage, but also expands the burden of throughput characterisation, wavelength calibration and telluric treatment.

Performance is a family of experiments, not one record number

The source supplies several complementary tests: resolution maps, efficiency versus wavelength and seeing, S/N versus magnitude, internal RV uncertainty, time-domain repeatability and telescope-path equivalence. Each controls a different part of the error budget. No single plot replaces the others.

The upper envelope of an S/N–magnitude relation describes favourable observing conditions rather than a guarantee for every exposure. Likewise, an internal photon-noise estimate concerns information in a measured spectrum, not the full uncertainty after stellar variability, interventions and calibration systematics are included. Distinguishing these definitions prevents an impressive number from acquiring a broader meaning than the experiment supports.

Interventions need explicit treatment. An improvement in throughput after a fibre change is scientifically valuable, but combining the time series without considering offsets or changed instrument response could confuse an RV analysis. A commissioning paper is most useful when it records those operational details rather than presenting a single unbroken idealised instrument.

The science cases explain why the engineering trade-offs matter

The guaranteed-time programme connects the design to more than one kind of measurement. A high-S/N bright-star spectrum and a faint quasar spectrum place different demands on resolution, collecting area and calibration fidelity. The science examples should therefore be read as demonstrations of distinct capabilities rather than as one universal observing recipe.

For a planet-search example, phase-folded RV agreement checks a repeatable signal, while binning and high-cadence observations probe the residual noise and stellar variability. A clean folded curve does not imply that the unmodelled time series is white noise. That is why the adjacent binning and p-mode figures deserve their own reading.

For absorption-line science, stable spectral structure and accurate local wavelength information can matter differently from a star’s global Doppler shift. The quasar examples broaden the calibration lesson: a wavelength distortion that partly averages out in one estimator may remain important in another. The instrument has to be characterised for the observable actually being inferred.

Reading based on the full manuscript, including methods and discussion—not only its abstract. These explanations are original commentary; they are not a reproduced analysis or new experiment. Original manuscript · expanded 21 October 2026.

Figure-by-figure reading

The figure panels below anchor the discussion to specific instrument-design and performance figures from Pepe et al. They are included where they clarify architecture, spectral format, resolving power, throughput, radial-velocity precision, and telescope-path equivalence.

Further source figures: reduction, sensitivity and science cases

Read the paper plots alongside the explanations below. Each discussion keeps its source figure number and links to the original caption and context. Openly licensed panels are shown here; other figures open at the original paper. The explanations distinguish the plotted evidence from what it cannot establish.

References

1. F. Pepe et al. (2021), ESPRESSO@VLT: On-sky performance and first results, Astronomy & Astrophysics 645, A96; arXiv:2010.00316.
2. T. M. Schmidt et al. (2026), Test of a 34 GHz EOM laser frequency comb at ESPRESSO, arXiv:2603.17908.
3. M. K. Kreider et al. (2025), Quantifying broadband chromatic drifts in Fabry-Pérot resonators for exoplanet science, Nature Astronomy / arXiv:2210.10988.