Corentin
Ravoux

I measure how cosmic structure grows and how the Universe expands, from nearby supernovae to the Lyman-α forest.

Chargé de recherche, CNRS
LPCA, Université Clermont Auvergne
DESI, ZTF, Rubin-LSST

I am a CNRS researcher at the Laboratoire de Physique de Clermont Auvergne (LPCA), working on the large-scale structure of the Universe with spectroscopic and photometric surveys.

My PhD (Université Paris-Saclay, 2022) measured the matter distribution with the Lyman-α forest of eBOSS and of the first DESI data. I started working on peculiar velocities as a postdoc at CPPM in Marseille (2022–2023), and joined the CNRS in November 2023. My work sits at two ends of the same axis: the local Universe, where supernova velocities measure how structure grows today, and the high-redshift intergalactic medium, where the Lyman-α forest measures the matter power spectrum on some of the smallest scales accessible to cosmology.

DESI: Builder; co-lead of the Lyman-α software topical group
Rubin-LSST: Co-lead of the DESC Peculiar Velocities topical team; member of the DESC collaboration council; AuxTel atmospheric-calibration team
ZTF: Type Ia supernova cosmology: growth rate and H₀-dipole analyses
Supervision: One PhD co-supervised (C. Barjou-Delayre, 2023–2026); four master's internships co-supervised, one mentored
Teaching: Tutorials designed and taught at a French engineering school, a PhD summer school in Mexico and three French summer schools (notebooks)

z ≈ 0.05, the local Universe

Growth of structure from peculiar velocities

Galaxies fall towards mass. Measure that infall and you measure the growth rate of structure, through the combination fσ₈: a test of gravity on cosmological scales.

Type Ia supernovae are the most precise distance indicators at low redshift (about 6% in distance per object, against about 20% for Tully–Fisher and fundamental-plane galaxies), so the residuals of their Hubble diagram carry the velocity field. I developed a generalised framework for field-level, likelihood-based inference of fσ₈ from velocity and density fields, released as flip and now used in DESI, ZTF and Rubin-LSST DESC. On realistic simulations of six years of ZTF supernovae and the DESI Bright Galaxy Survey (DR3), adding the galaxy density field reduces the uncertainty on fσ₈ by up to 50%, to about 10%, without bias.

For Rubin-LSST, simulations of the full survey show that fσ₈ can be measured to 10% over 0.02 < z < 0.14 in the most realistic case (photometric classification, non-Ia contamination, missing host redshifts), and to better than 18% in each of three redshift bins. I estimated how many host redshifts DESI and 4MOST will provide; intrinsic-scatter modelling turns out to be one of the main systematics.

At that precision, photometric calibration takes over. With the AuxTel team I added Gaia spectra as standard stars for measuring the atmospheric transmission above the Rubin Observatory, alongside the CALSPEC standards, and a master's internship I co-supervised derived corrections that remove most of the residual difference between the two. A paper is in preparation.

Forecast fσ₈ uncertainties against redshift for ZTF, ZTF with DESI, LSST, DESI and Euclid, over ΛCDM, f(R) and nDGP growth curves — Forecast fσ₈ uncertainties from realistic simulations of supernova peculiar velocities (ZTF, ZTF combined with DESI galaxies, LSST), with Fisher forecasts for DESI and Euclid redshift-space distortions, against f(R) and nDGP predictions computed with EFTCAMB. These models change the expansion history by less than 0.1%, so growth is where they can be caught.

z ≈ 0.1, isotropy

Anisotropy and cosmic dipoles

The cosmological principle is an assumption, not a measurement. Supernovae can test it directly.

error on the H₀ dipole amplitude: 0.33km s⁻¹ Mpc⁻¹
error on its right ascension: 3.4°
error on its declination: 6.1°

Recovery of a 3 km s⁻¹ Mpc⁻¹ dipole injected into realistic simulations of the ZTF DR2 supernova sample. arXiv:2601.11139

The same Hubble-diagram residuals that give the velocity field also let you fit the expansion rate as a function of direction. With Chloé Barjou-Delayre, whose PhD I co-supervised, we built a likelihood fit of an H₀ dipole and tested it on realistic simulations of the ZTF sample: the injected dipole is recovered without bias, whatever H₀ is assumed. Peculiar velocities are a major contaminant, because a coherent bulk flow can mimic an expansion dipole, and selection effects in the survey add a systematic error that the error model includes.

z ≈ 0.3–2.3, a standard ruler

Baryon acoustic oscillations with DESI

The sound horizon left a fixed length scale imprinted in the matter distribution. Measuring it at many redshifts, with galaxies and with the Lyman-α forest, turns it into a ruler for the expansion history.

Shift of the BAO scale along and across the line of sight in the Lyman-alpha and quasar cross-correlation, against redshift, for several quasar biases — Shift of the BAO scale in the Lyman-α × quasar cross-correlation, along (top) and across (bottom) the line of sight, from z = 2 to 4 and for a range of quasar bias parameters. It never exceeds 0.3%. arXiv:2412.06892

Inside DESI I work on the Lyman-α side of that measurement, the catalogues and the analysis pipeline the forest correlations are built from, and I co-lead the Lyman-α software topical group. That work runs through the DR1 and DR2 Lyman-α BAO analyses. Combined with the CMB and supernovae, the DR2 BAO measurement prefers an evolving dark energy over ΛCDM at 2.8σ to 4.2σ, depending on the supernova sample.

How far does non-linear physics move the peak? On the ACCEL² simulations, which I co-initiated and co-lead, an effective-field-theory analysis led by R. de Belsunce found shifts of the BAO scale of at most 0.3%, for the forest auto-correlation and for its cross-correlation with quasars. That keeps the forest a trustworthy ruler at z ≈ 2.3, and the estimate entered the DESI DR2 Lyman-α BAO and full-shape analyses as their theoretical systematic.

z ≈ 2.5, the intergalactic medium in 3D

Tomographic maps, and the voids inside them

Every distant quasar is a backlight: the neutral hydrogen between us and it absorbs at the Lyman-α wavelength, redshifted, leaving a forest of absorption lines. With enough sightlines close together, the forest becomes a three-dimensional map.

Three-dimensional rendering of the tomographic map of Stripe 82, showing overdense and underdense iso-surfaces — The Stripe 82 tomographic map in three dimensions: iso-surfaces of the reconstructed flux contrast, overdense in red, underdense in blue. arXiv:2004.01448

From the eBOSS quasars of the SDSS Stripe 82 field (220 deg², 37 sightlines per square degree), I reconstructed the flux-contrast field between the sightlines with a Wiener filter. The map covers 0.94 h⁻³ Gpc³ at 13 h⁻¹ Mpc resolution, the largest-volume three-dimensional map of matter fluctuations at high redshift. The pipeline is open, as lelantos.

A map lets you ask questions a power spectrum cannot: where are the underdense regions? We identified voids in the map and measured their cross-correlation with the forest. Its quadrupole carries the flow of matter out of the voids: redshift-space distortions detected at 8σ at a median redshift of 2.49, the first detection of flows around voids that far back, with a model that accounts for the sightlines being parallel.

Two-dimensional void and Lyman-alpha forest cross-correlation against transverse and line-of-sight separation — Cross-correlation between cosmic voids and the Lyman-α forest, ξ × r, against separation across (r⊥) and along (r∥) the line of sight, in h⁻¹ Mpc. arXiv:2203.11045

z ≈ 2.2–4.2, the intergalactic medium

The Lyman-α forest power spectrum

Along each sightline, the forest traces the matter distribution down to about 1 h⁻¹ Mpc, smaller scales than any galaxy survey can reach.

DESI DR1 one-dimensional Lyman-alpha power spectrum in eleven redshift bins — The DESI DR1 one-dimensional Lyman-α power spectrum, Δ²_1D = k P_1D / π, in eleven redshift bins from z = 2.2 to 4.2 and up to k = 2 Å⁻¹, measured with the FFT estimator. arXiv:2505.09493

I lead the measurement of the one-dimensional Lyman-α power spectrum, P1D, with the Fast Fourier Transform estimator in DESI: first on the early data (26,330 quasar spectra from the Early Data Release and the first two months of the survey), then on Data Release 1. Together with an independent quadratic estimator, the DR1 measurement is the most precise P1D to date. It constrains the small-scale matter power spectrum and, through it, the running of the primordial spectral index, neutrino masses, warm dark matter and the thermal history of the intergalactic medium.

The DR1 cosmological analysis measures the amplitude and slope of the linear power spectrum at z = 3, Δ²_★ = 0.379 ± 0.032 and n_★ = −2.309 ± 0.019, and, combined with the CMB and DESI BAO, tightens the constraints on N_eff and on the running of the spectral index.

Code I build and maintain

Each measurement above comes with its software. My main packages are open, documented and versioned, with a licence and a Zenodo DOI; the collaboration code I contribute to, such as picca for the DESI Lyman-α analyses, is open too.

Supernovae and peculiar velocities

Lyman-α forest

Teaching

lymanalpha-tutorials Notebooks from the tutorials I teach in schools and summer schools, built around applications of the Lyman-α forest.

About