Research Director · CNRS
I work on the gas in and around galaxies — how it cools, how it is set in motion, and what that tells us about why galaxies grow the way they do.
Centre de Recherche Astrophysique de Lyon (UMR 5574)
CNRS · Université Claude Bernard Lyon 1 · École normale supérieure de Lyon
I have been a research director at the CNRS since 2006, at GEPI, then the Institut d’astrophysique de Paris, and since 2021 at CRAL in Lyon, which I directed from January 2021 through the end of July 2024. Before coming to France I held positions at the Max-Planck-Institut für extraterrestrische Physik in Munich, Leiden Observatory, and Lawrence Livermore National Laboratory.
I am the author or co-author of more than 300 refereed papers, with some 31,400 citations and a Hirsch index of 92. I have supervised fifteen doctoral students and about ten postdoctoral researchers; nearly all have gone on to academic or teaching careers.
I have sat on many of the major telescope and instrument review panels of the past three decades, including the NSF extragalactic review board, the Hubble Space Telescope time allocation committee for cycles 16, 31 and 32, the ESO Observing Programmes Committee, and the JWST galactic-ecosystems panel as one of the ESA representatives. I was project scientist for KMOS, for LUCI on the Large Binocular Telescope, and for EAGLE, a direct precursor of MOSAIC for the ELT.
What ties my work together is the gas: where it is, what phase it is in, and how it loses the energy it has. Underneath all of it is one question — galaxies turn only a small fraction of the available baryons into stars, and after three decades we still cannot say confidently why. Answering that means finding the material that did not collapse and accounting for the energy that kept it from collapsing, which is what most of the work below is trying to do.
Most of a galaxy’s baryons are not in the galaxy. What sits in the halo, in which phase, and whether it is arriving or leaving are all still badly constrained at high redshift — and the cold, molecular component, the part that can actually form stars, is the hardest of all to detect. With ALMA, the VLA and MUSE we have found it to be far more abundant than anyone expected: in some protoclusters there is as much molecular gas outside the galaxies as within them, and in one case a 100-kpc stream of it running into a radio galaxy. Whether these reservoirs are extended disks or something closer to spherical, and how much of the material is genuinely inflowing rather than debris, are still open.
Every simulation of galaxy formation needs winds, and almost none can tell you how much mass they carry. I measure outflows in the phases that hold the mass rather than the phases that are easy to see, which means working across ionized, atomic and molecular tracers at once; the recurring surprise is how often the obvious driver fails, with radiation pressure falling an order of magnitude short and the radio jet doing the work. The other half of the same problem is where that energy goes once it is in the gas. In diffuse molecular gas it is radiated by H₂, which turns the rotational lines into a measurement of a dissipation rate rather than just a mass — the reasoning behind our JWST programs on the Spiderweb and on cluster filaments.
Wherever the hot atmosphere of a cluster or group has low enough entropy, filaments of cold gas appear. Why they are filamentary, what excites them, and why they persist are all unsettled. Our MUSE surveys established how common they are and tied them to short cooling times and low central entropy, which is what would be expected if they condense out of the hot halo; the JWST MIRI follow-up on seven brightest cluster galaxies is meant to settle what heats them, and whether a central molecular disk forms that could feed the active nucleus and close the feedback loop.
The gas that matters most for star formation is cold, dense and faint, and at high redshift it is usually beyond reach in emission. Absorption against a bright radio or millimeter continuum gets around this, because the signal scales with the background source instead of falling away with distance. We used it to detect CO, HCO⁺ and HCN in a radio galaxy at z = 3.4, and as the resolution improves the absorbing complex keeps breaking into narrower components — the internal structure of individual molecular clouds, twelve billion years ago. We are now looking for complex, possibly prebiotic molecules the same way. It is a risky approach and I present it as one.
Galaxies at z > 2 sustain specific star-formation rates that are all but unheard of nearby, and almost everything one would want to measure about them — how much gas they hold, the shape of a line profile, whether material is still arriving — sits at the edge of what any high-redshift observation can do. A small number of local galaxies do reach those rates, and those can be observed properly. With W. van Driel and R. Minchin I have put together H I 21 cm observations of 373 such galaxies, from the Nançay Radio Telescope and Arecibo, giving 328 clear detections. They hold reservoirs ample enough to sustain their growth, and the asymmetries in their line profiles point to gas still falling in — which would tie the high rates to accretion rather than to mergers. The molecular half, from IRAM 30 m and APEX CO observations already in hand, comes next.
Ours is the only galaxy whose chemical evolution can be read star by star, which makes it the one place a formation history can be tested in detail rather than inferred. With colleagues at Paris Observatory I worked for several years on the two-phase growth of the disk, on the α-element bimodality that separates thin from thick, and on the old G-dwarf problem — arguing that the thick disk pre-enriched the inner Galaxy to solar metallicity, and that the bimodality requires neither mergers nor radial migration. Some of that is now widely accepted; several parts remain contested, which is fair.
I serve on the steering committees and executive boards of four next-generation facilities, which takes roughly twenty days a year.
The first-light visible and near-infrared integral-field spectrograph for ESO’s Extremely Large Telescope. CRAL is responsible for the slicer unit, the relay optics and the science software. The project is negotiating design rescopes with ESO and revising its consortium agreement following the decision to use MORPHEO as the adaptive-optics feed.
elt.eso.org/instrument/HARMONI ↗The 4-metre Multi-Object Spectroscopic Telescope, a wide-field fibre-fed spectrograph on ESO’s VISTA telescope at Paranal. The low-resolution spectrograph was built entirely at CRAL and has been delivered and through assembly, integration and verification; the high-resolution spectrograph is finishing commissioning.
4most.eu ↗A blue-optimized wide-field integral-field spectrograph for the VLT, extending MUSE into the ultraviolet. ESO has committed about 9 M€. Phase B is underway, with a preliminary design review about a year away.
bluemuse.univ-lyon1.fr ↗The Wide-field Spectroscopic Telescope, a proposed 10–12 m survey facility to be submitted to ESO in June 2027. I helped develop the EU research-infrastructure grant of about 3 M€ that funds the definition phase, for which CRAL is the lead institute.
wstelescope.com ↗Three programs occupy most of my time. All three are about the same thing from different directions: how gas dissipates its energy, cools, and forms stars in places where that should be difficult.
We hold a substantial JWST MIRI imaging and IFU program on the Spiderweb radio galaxy at z = 2.16, one of the most spectacular known protoclusters, where stars are forming across nearly 100 kpc of circumgalactic gas.
The question is whether molecular hydrogen is the dominant cooling channel in this turbulent, multi-phase medium. As shocks and turbulent mixing cascade kinetic energy to small scales, it should be radiated in the pure rotational H₂ lines, which carry the bulk of the emitting mass and set the cooling rate. The Spiderweb is the most distant galaxy for which Spitzer detected H₂ 0–0 S(3) and S(5), and so is likely the best system in which JWST can capture enough of them before they redshift out of MIRI’s reach.
The data are in hand and give the first spatially resolved energy budget for warm H₂ cooling across a protocluster.
We have MIRI imaging and IFU spectroscopy of seven brightest cluster galaxies, continuing the MUSE survey that revealed ionized filaments reaching 30 kpc in low-entropy systems.
The spectra are essentially a complete mid-infrared census: [Ne II] and [Ne III], PAH features at 7.7, 11.3 and 17 µm, the warm H₂ 0–0 S(1)–S(9) ladder, high-ionization metal lines and hydrogen recombination lines. We aim to measure the total molecular mass, map excitation and kinematic gradients, look for compact molecular disks near the central AGN, close the energy budget, and establish why star formation does and does not occur in the filaments.
These trace two stages of the feedback cycle: the condensation of cold filaments out of the hot intracluster medium, and the formation of a central disk that may then feed the AGN and close the loop.
What limits our understanding of star formation in the early Universe is that we cannot follow gas cooling from diffuse warm phases down to the cold, dense scales where stars and planets actually form. This is a deliberately high-risk program to detect absorption from complex — possibly prebiotic — molecules against the bright radio and millimeter continuum of compact high-redshift radio galaxies.
Our target is B2 0902+34 at z = 3.4, where we have already detected CO(0–1), HCO⁺(0–1) and HCN(0–1) absorption. New sub-arcsecond JVLA data break the absorbing complex into distinct velocity components as resolution increases, consistent with an unresolved hierarchy of dense clouds no more than a few hundred parsecs across.
Roughly 50 hours of JVLA time is now being taken, with ALMA observations of methanimine (CH₂NH) and methylamine (CH₃NH₂) still to come. A detection would be among the highest-redshift complex organic molecules found.
Nine papers from recent years, chosen because each answered a question I could not have answered before, not because of where they appeared. Each opens to a short summary and the figure that carries the result.
ALMA observations of the [C I](1–0) line around the radio galaxy 4C 41.17 at z = 3.8, which sits inside a large Lyα halo. [C I] traces neutral atomic and molecular hydrogen, so it reaches gas that Lyα cannot: Lyα only shows material above about 10⁴ K, whereas the gas that actually fuels star formation is at 10–100 K.
A filamentary stream extending 100 kpc and connecting to the galaxy, containing (6.7 ± 2.2) × 10¹⁰ M☉ of molecular hydrogen — comparable to the stellar mass of the galaxy itself. Velocities decline steadily toward the galaxy, as expected for infall along a cosmic filament rather than for an outflow.
Cold streams penetrating dark-matter halos have been predicted by simulations for two decades, but the observational evidence had come from Lyα, which traces gas far too warm to form stars. Here the raw material itself is detected. Even at a continuous 250 M☉ yr⁻¹, 4C 41.17 would exhaust its own molecular gas within about 560 Myr; the stream is what can keep it going.
MUSE and ALMA observations of 4C 19.71, combining the rest-ultraviolet emission lines (Lyα, C IV, He II, [C III]) with [C I] and the dust continuum, so that the ionized and molecular phases of the same halo are characterized together rather than separately.
A narrow [C I] component — FWHM about 100 km s⁻¹, offset only a few km s⁻¹ from systemic — lies roughly 75 kpc from the galaxy: dynamically quiescent gas far out in a halo whose ultraviolet lines are otherwise broad and disturbed. Photoionization modeling of the northern region gives a low metallicity, 0.03 < Z/Z☉ < 0.1.
Near-pristine gas at 75 kpc is more easily explained as recently accreted than as expelled. Quiet accretion and jet-driven feedback are therefore going on in the same halo at the same time, which is awkward for any picture in which a powerful radio source simply sweeps its environment clean.
SED fitting with BEAGLE of the ten gravitationally lensed z ~ 9–16 candidates found behind the cluster SMACS J0723.3−7327 — among the very first JWST deep-field results. We used all three published strong-lensing models of the cluster, so that magnification systematics are carried through the fit, and dynamical arguments to bound the ages.
Most of the sample is low mass, 10⁷–10⁸ M☉, with ages of 10–100 Myr; a few reach 10⁹–10¹⁰ M☉ where a Balmer break is detected at z ~ 9–10. Ultraviolet slopes reach β ~ −3, implying almost no dust (AV ≲ 0.02). The mass–luminosity relation is shallow and the specific star-formation rates sit above the main sequence, with no significant evolution in either.
How much could legitimately be claimed about these objects depended entirely on how carefully the magnification and its uncertainty were handled. Propagating three independent lens models is what makes the resulting masses defensible, and the exercise set a standard for the lensed high-redshift candidates that followed.
ALMA observations of hydrogen fluoride HF J = 1–0, water H₂O (2₂₀–2₁₁) and the 1.2 THz rest-frame continuum in BR 1202−0725, a group at z = 4.7 containing a quasar, a submillimeter galaxy and two Lyα emitters. Fluorine is almost entirely locked into HF, which makes the line an unusually clean probe of irradiated molecular gas.
HF is seen in emission in the quasar and possibly in absorption in the submillimeter galaxy; water appears in emission in both. The quasar is the most luminous HF J = 1–0 emitter yet found, and its ratio of HF line to infrared luminosity matches that of local active nuclei and of the Orion Bar — a consistency holding over roughly ten orders of magnitude in infrared luminosity. Radiative transfer modeling gives excitation either by electrons and H₂ in molecular plasma irradiated by the AGN and its starburst, or predominantly by H₂ in gas near 120 K and 10⁵ cm⁻³.
The constant luminosity ratio from the Orion Bar to a quasar at z = 4.7 is the striking part. It suggests HF emission tracks irradiated molecular gas nearly independently of scale, which would make it a usable diagnostic in systems where almost nothing else can be measured.
A deep ATCA CO(1–0) mosaic of the Spiderweb protocluster field. CO(1–0) is the most direct tracer of total molecular mass, and reaching it at z = 2 over a whole overdensity is expensive enough that it had not been done.
Forty-six CO(1–0) detections spanning z = 2.09–2.22 — the largest set of molecular gas measurements in a protocluster — and from them the first constraint on the CO luminosity function in an overdense environment. Its amplitude is 1.6 ± 0.5 orders of magnitude above the field at z ~ 2, and an order of magnitude above what the SHARK semi-analytic model predicts for protoclusters.
Protoclusters are not merely crowded; they are far richer in molecular gas than either field galaxies or the models anticipate. The order-of-magnitude gap with SHARK is the interesting part, since it is the models rather than the observations that have to give way.
Chemo-dynamical simulations of the formation of Milky Way-type galaxies, asking where the two α-sequences that distinguish the thin from the thick disk actually come from — a bimodality usually attributed to mergers or to radial migration.
The two sequences form in quite different environments. The high-α sequence comes from an early burst in a turbulent, compact gaseous disk that becomes the thick disk. The low-α population is quiescent star formation fed by slow accretion onto a radially extended thin disk. Feedback-driven outflows during thick-disk formation enrich the surrounding halo, which then feeds the disk on a longer timescale. In the thin-disk phase the chemical evolution reaches an equilibrium metallicity where stars pile up; that equilibrium falls toward the outer disk, and the ridge line it traces is the low-α sequence. A second route appears in one simulation, where a rapid feedback-driven shutdown suppresses halo enrichment and the gas accreted afterward dilutes the interstellar medium.
The bimodality needs neither mergers nor radial migration — it falls out of dissipative collapse and disk growth. If that is right, it should be common among Milky Way-mass galaxies, and the chemistry of our own disk is a record of its star-formation history rather than of its accretion history.
A 154-pointing IRAM NOEMA mosaic of CO(1–0) in and around M82, complemented by zero-spacing data, reaching about 30 pc and 5 km s⁻¹ — enough to resolve the starburst disk, the outflow and the tidal streamers in a single dataset.
About two thousand clouds were decomposed. To first order their properties, and their size–line width relations, are similar regardless of environment. Their distribution in σ²/R against column density indicates that external pressure does not set their parameters in the outflow or the streamers. Clouds in the streamers keep roughly constant size (R ~ 50 pc) and mass (~10⁵ M☉) with projected distance, whereas clouds in the outflow shrink in both toward the south.
The streamers and the outflow behave differently, and both are routinely lumped together as molecular gas leaving a starburst. M82 is the nearest system where the distinction can be drawn cloud by cloud, which makes it the reference for interpreting far coarser ALMA observations of distant starbursts.
MUSE observations of eighteen local (z ≤ 0.017) brightest group galaxies, asking where the cold gas in group centrals comes from. The question is settled neither in groups nor in clusters, and groups are much the less studied of the two.
Ten of the eighteen show complex filament networks reaching about 10 kpc; two show compact disks (< 3 kpc) and five extended ones (> 5 kpc). The stellar kinematics are rotation-dominated and quite unlike the disturbed filamentary gas, and in most systems the ionized gas is kinematically decoupled from the stars — which points to an external origin. Hα luminosity correlates with molecular gas mass, and the filaments and compact disks occur in systems with low central entropy and low tcool/teddy.
The condensation of hot halo gas through thermal instability that is invoked for cluster centrals appears to operate in groups as well, consistent with chaotic cold accretion. But rotating disks are more common here than in cluster centrals, so wet mergers likely contribute too: the cold gas does not have a single origin.
ALMA [C II] 158 µm and continuum observations of four radio AGN at z ≈ 3.5, concentrating on eight companion cloud systems found within a few tens of kiloparsecs, with archival [C I] — a tracer of H₂ — reanalyzed to check the hosts themselves.
Most of the cold gas is not at the position of the AGN. The [C II] data at 0.2″ together with the reanalyzed [C I] confirm that the hosts are gas-poor while their companions are gas-rich. [C II] luminosities span 2.8 × 10⁸ to 4.2 × 10⁹ L☉, and the ratio to infrared luminosity, about 9.4 × 10⁻⁴, is unremarkable compared with the literature.
Two readings remain open and the data do not yet separate them. The companions may be clouds stripped during mergers, in which case they are a plausible trigger for the radio-loud phase; or they may record negative feedback, with the AGN having heated and cleared its own host. Either way it is the asymmetry between host and companion that has to be explained.
Complete lists: arXiv · NASA ADS · Google Scholar
I am glad to hear from prospective doctoral students and postdoctoral researchers interested in the physics of the circumgalactic medium, high-redshift galaxy evolution, molecular gas, or multi-wavelength observational work. A short description of your background and what you want to work on is the most useful thing to send.