ESPRESSO Radial Velocities: Techniques in Exoplanet Detection

Updated 22 January 2026

ESPRESSO Radial Velocities is an ultra-stable, high-resolution spectrograph achieving sub-10 cm s⁻¹ precision, essential for detecting rocky exoplanets.
It integrates advanced calibration (via laser frequency combs), fiber scrambling, and innovative data reduction to minimize instrumental drifts and stellar noise.
Its dual RV extraction methods—cross-correlation and template matching—coupled with robust error budgeting enable reliable planet detection across diverse observing modes.

The ESPRESSO (Echelle SPectrograph for Rocky Exoplanets and Stable Spectroscopic Observations) instrument is an ultra-stable, fiber-fed, high-resolution spectrograph installed on the Very Large Telescope (VLT) at ESO/Paranal, explicitly designed to achieve extreme radial-velocity (RV) precision (approaching and surpassing 10 cm s⁻¹) for exoplanet detection and precision astrophysics. ESPRESSO operates in both single-telescope high-resolution and multi-telescope modes, leveraging advanced hardware and sophisticated data reduction methodologies to deliver RVs with unmatched precision and stability for bright, quiet G, K, and M dwarfs (Hernández et al., 2017, Pepe et al., 2020, Nielsen et al., 2022, Figueira et al., 10 Jul 2025).

1. Instrumental Design and Calibration Framework

ESPRESSO is engineered for sub-10 cm s⁻¹ RV stability by combining multiple hardware innovations:

Optomechanical Environment: The spectrograph optics and bench are housed in a vacuum vessel (≤10⁻⁵ mbar) inside a nested multi-shell thermally stabilized room (ΔT≤0.001 K over days), eliminating environmental drifts (Pepe et al., 2014, Hernández et al., 2017, Pepe et al., 2020). All optomechanical components rest on vibration-dampened mounts and are constructed with matched coefficients of thermal expansion (CTEs) for maximal stability.
Fiber-Feed and Scrambling: Light is injected via octagonal or square fibers and a double-scrambler system, achieving near-field and far-field illumination stability at the 10⁻⁴ level. This suppresses guiding and seeing-induced RV drifts (Nielsen et al., 2022, Pepe et al., 2014).
Spectral Resolution and Observing Modes:
- Single-UT HR: R≃134 000, 1″ aperture, total efficiency ≃11%.
- Single-UT UHR: R≃200 000, 0.5″ aperture.
- Multi-UT MR: R≃59 000, effective 16 m collecting area (Pepe et al., 2014, Hernández et al., 2017).
Wavelength Calibration:
- Laser Frequency Comb (LFC): Provides an absolute, dense, and stable frequency grid with stability <2 cm s⁻¹ over one night, traceable to atomic clocks (Nielsen et al., 2022, Pepe et al., 2014).
- Fabry–Pérot Etalon and ThAr Lamps: Serve as backup and for long-term drift monitoring.
- The simultaneous calibration fiber allows real-time correction to sub-10 cm s⁻¹ instrumental drifts.
CCD System: Homogeneous, thermally stabilized 92 × 92 mm CCDs, ensuring pixel-position fidelity at the nanometer scale (Hernández et al., 2017).

2. Radial-Velocity Extraction Methods

ESPRESSO implements two primary approaches for RV measurement:

Cross-Correlation Function (CCF) with Binary/Weighted Mask:
- The extracted, normalized spectrum $f(\lambda)$ is cross-correlated with a spectral-type-specific numerical mask $M_i(\lambda)$ :
$\mathrm{CCF}(v) = \sum_{i} w_i \int f(\lambda) M_i[\lambda(1+v/c)] d\lambda,$

where $w_i$ are line weights reflecting depth, S/N, and blending (Nielsen et al., 2022). - The CCF profile is fit with a Gaussian (or advanced models), and the centroid gives the RV.
Template-Matching (S-BART, TERRA, SERVAL):
- Constructs a high-SNR master template $T(\lambda)$ by co-adding multiple spectra. The observed spectrum is fit via least-squares or a semi-Bayesian framework:
$\chi^2(v, \alpha) = \sum_k \left[ f_k - \alpha T_k(1+v/c) \right]^2 / \sigma_k^2$

This approach improves photon-noise performance (by 10-20% in M/K dwarfs) and is robust to line-blending (Silva et al., 2022, Nielsen et al., 2022).
Data Reduction Software (DRS): Implements these algorithms and enforces "pixel conservation"—tracking both original and merged flux/variance to allow true $\chi^2$ -based fitting and error propagation (Cupani et al., 2018).

3. Error Budget, Precision Limits, and Noise Mitigation

The RV error budget is expressible as:

$\sigma_{\rm tot}^2 = \sigma_{\rm ph}^2 + \sigma_{\rm cal}^2 + \sigma_{\rm drift}^2 + \sigma_{\rm star}^2 + \sigma_{\rm jit}^2,$

where:

$\sigma_{\rm ph}$ : photon-noise error;
$\sigma_{\rm cal}$ : calibration source (LFC/etalon) uncertainties, ∼5–10 cm s⁻¹;
$\sigma_{\rm drift}$ : residual instrument drifts after simultaneous calibration, <$10$ cm s⁻¹;
$\sigma_{\rm star}$ : stellar jitter (granulation, oscillations, activity), ranging 10–100 cm s⁻¹;
$\sigma_{\rm jit}$ : additional white or correlated "jitter," often fitted per time series (Lillo-Box et al., 2021, Figueira et al., 10 Jul 2025).

Stellar noise is further divided:

P-modes: Averaged out by 10–20 min exposures; forward modeling or Gaussian Process (GP) modeling used to characterize remaining jitter at the ~10 cm s⁻¹ level (Figueira et al., 10 Jul 2025).
Granulation/Supergranulation: Modeled as Harvey functions in the RV power spectrum; sets a floor of 20–70 cm s⁻¹ for Sun-like stars (Figueira et al., 10 Jul 2025).
Rotationally-Modulated Activity: Quasi-periodic GP kernels, coupled with activity diagnostics (e.g., CCF FWHM, S-index, line-by-line analysis), are routinely deployed for decorrelation (Faria et al., 2019, Netto et al., 2021, Lillo-Box et al., 2021).

Telluric contamination ("micro-tellurics") and binary contamination are addressed by:

Line-by-line masking of affected orders/pixels, deep atmospheric modeling (TAPAS/LBLRTM), and variance propagation through the reduction chain (Cunha et al., 2014, Cunha et al., 2012).
CCF bisector span and spectral mask tests for the identification of blends.

Nightly zero-point scatter (NZP) after all corrections is empirically measured at 0.6–0.7 m s⁻¹, including stellar, photon, and instrumental effects; S-BART/Template-matching achieves night-to-night precision of ∼10 cm s⁻¹ for M dwarfs (Silva et al., 2022).

4. Science Reach and Notable Applications

Detection Limits: ESPRESSO's photon-noise floor is ∼10 cm s⁻¹ (V < 8.6, S/N > 650, R ≃ 134 000, 20 min exposure); real single-exposure precision on-sky is 10–25 cm s⁻¹ for G/K/M dwarfs, with long-term (years) stability of 20–40 cm s⁻¹ (Figueira et al., 10 Jul 2025, Pepe et al., 2020).
Survey Strategy: The ESPRESSO GTO campaign targets a curated sample of 45 GKM dwarfs (V<12), optimized for intrinsic stability (log R′_HK < –4.80, v sin i < 5 km s⁻¹, single stars) and habitable zone (HZ) sensitivity at the Earth-mass level (Hojjatpanah et al., 2019).

Recent Planet Discoveries (examples):

Proxima Centauri b/d: Confirmed and refined with ESPRESSO. Proxima d recovery at $K=0.44\pm0.13$ m s⁻¹ (planet mass $\sim0.26\,M_{\oplus}$ ), RV rms 0.27 m s⁻¹ after GP decorrelation (Mascareño et al., 2020, Nielsen et al., 2022).
Barnard's Star: Four sub-Earth-mass planets ( $K<50$ cm s⁻¹, $m\sin i$ 0.19–0.34 $M_{\oplus}$ ), ESPRESSO residuals 23 cm s⁻¹, demonstrating sensitivity to terrestrial-mass planets in short orbits (Basant et al., 11 Mar 2025).
HD 22496b and HIP 29442 system: Characterized planets with $K=2$ –5 m s⁻¹ and demonstrated robust activity filtering via GP+Keplerian modeling (Lillo-Box et al., 2021, Damasso et al., 2023).
Empirical Null Results: On stars like HD10700, RV sensitivity to $K$ as low as 10 cm s⁻¹ allows exclusion of $1.7\,M_{\oplus}$ planets at $P<100$ d and $2$– $5\,M_{\oplus}$ in the HZ (Figueira et al., 10 Jul 2025).

See the table below for a sample of ESPRESSO precision metrics from recent surveys:

Star/Survey	Timescale	Achieved Precision	Science Outcome
τ Ceti (HD10700)	1 hr	10 cm s⁻¹	Intraday p-mode floor
τ Ceti (HD10700)	1 yr	20–30 cm s⁻¹	Bona fide planetary sensitivity
Barnard's Star	2.5 yr	23 cm s⁻¹ (rms)	4 sub-Earth planets
Proxima Centauri	~7 mo	26 cm s⁻¹ (photon noise)	Sub-Earth discovery, GP disentangling
HD22496b	2.5 yr	18 cm s⁻¹ (photon noise)	Rocky/gaseous boundary

5. Advanced Data Modeling: Gaussian Processes and Planet Detection

Detection and characterization utilize Bayesian hierarchical models that jointly fit Keplerian orbits and correlated-noise GPs:

Quasi-periodic GP kernels, incorporating rotation periods and active-region evolution timescales, are shown to be critical for activity-dominated stars (Faria et al., 2019, Netto et al., 2021, Lillo-Box et al., 2021).
Model selection is based on Bayesian evidence, with $\Delta \ln \mathcal{Z} > 5$ as a detection threshold (Basant et al., 11 Mar 2025).

Multi-instrument modeling (e.g., ESPRESSO + MAROON-X) involves fitting independent offsets and jitter terms per dataset, enabling robust planet confirmation and activity discrimination even with signals $K<30$ cm s⁻¹ (Basant et al., 11 Mar 2025).

6. Limitations, Systematics, and Path Forward

Instrumental Stability: Demonstrated intra-night drift tracking at <10 cm s⁻¹, but years-long stability is set by residual calibration errors, hardware interventions, and occasional fiber changes (requiring offset terms) (Pepe et al., 2020, Figueira et al., 10 Jul 2025).
Stellar Noise Floor: Granulation/supergranulation and magnetic activity place an intrinsic floor of 20–70 cm s⁻¹ for bright, quiet solar-type stars—even with perfect instrument performance (Figueira et al., 10 Jul 2025).
Contaminants: Unresolved companions (Δmag<10 for G/K, Δmag<8 for M at 10 cm s⁻¹ level) remain a systematic limitation; careful CCF and AO imaging mitigation is mandated (Cunha et al., 2012, Hojjatpanah et al., 2019).
Tellurics and Atmosphere: Micro-telluric lines can induce RV errors at the 10–100 cm s⁻¹ scale; division by high-resolution models (TAPAS/LBLRTM), real-time water column adjustment, and pixel-level masking are required to reach cm s⁻¹ accuracy (Cunha et al., 2014).

Development Directions:

Further LFC improvements for absolute calibration (<1 cm s⁻¹ long-term).
Next-gen fiber scrambling and modal noise suppression.
Advanced line-by-line RV computation, including activity-informed weighting and chromatic diagnostics (Nielsen et al., 2022).
Integration with ELT-class spectrographs (e.g., ANDES) targeting the cm s⁻¹ domain.

7. Impact on Exoplanet Science and Fundamental Physics

ESPRESSO has established a new regime in RV precision, enabling:

Routine detection/confirmation of sub-Earth-mass exoplanets ( $K<50$ cm s⁻¹) in the nearest stars.
Mass/radius bulk composition studies for small and temperate exoplanets when combined with transit photometry (TESS, PLATO).
Constraints on variability of fundamental constants (α, μ) at the ppm level via high-precision QSO absorption-line measurements (Hernández et al., 2017).
Refinement of atmospheric dynamics, rotation, and convective processes in exoplanet and stellar characterization (Cristo et al., 2023).

ESPRESSO's legacy is codified as the technical and operational benchmark for high-precision RV surveys as the community advances towards detecting true Earth analogs in the habitable zones of solar-type and M dwarf stars (Figueira et al., 10 Jul 2025, Nielsen et al., 2022).