Research
Past our observational horizons: measuring the half of cosmic star formation hidden by dust.
Roughly half of all the stars ever formed were born behind veils of cosmic dust: their light absorbed and re-radiated in the infrared, invisible to the optical surveys that shaped our textbook picture of galaxy evolution. In the first billion years after the Big Bang, that hidden half is essentially unmeasured. My research revolves around one question: how much of the early Universe's star formation is hidden by dust — and when did the hidden Universe come to dominate? I answer it with the world's largest infrared and radio telescope arrays, above all ALMA: measuring the dust directly at the highest redshifts (§1), securing the cosmic distances that make those measurements possible (§2), calibrating the physics on the brightest dusty galaxies in the sky (§3), mapping the crowded environments that bias our census (§4), and helping build the instruments that will finish the job (§5).
This programme runs on competitive telescope time — including fifteen accepted ALMA programmes as PI and the 250-hour z-GAL NOEMA Large Programme as co-PI — and its datasets have supported bachelor, master and PhD thesis projects across four countries. Each thread below links to the papers behind it; the full list lives on the publications page.
A quick primer: galaxy formation in a nutshell
Setting the stage
Boom! The Big Bang started both space and time, together with an incomprehensible amount of energy. Through several fascinating processes this energy expanded with the Universe, cooled, and condensed into matter. The distribution of matter mere seconds after the Big Bang formed the 'nucleation sites' for galaxies, and its imprint can still be seen today in the Cosmic Microwave Background — light emitted when the Universe suddenly became transparent. Since then the Universe has expanded about 1300-fold, stretching that light from optical colours into the mm-domain (a redshift, z, of 1300). After this came the cosmic dark ages: nothing but slowly-cooling neutral hydrogen and helium gently collapsing.
Forming the first stars
Around a redshift of 30 to 15, collapsing clumps of gas awaken the first stars in the Universe (some theories predict giant stars a thousand times the mass of our Sun). Stars burn hydrogen, but need carbon, nitrogen and oxygen to do so efficiently. Single stars are far too faint to see at these distances; we can, however, find whole galaxies out to redshifts ~8 to 10 with the Hubble Space Telescope. On the left is one such galaxy, where we also detected the fingerprints of carbon and oxygen (Bakx et al. 2020).
Messy, dusty business
While we can barely detect dust at redshifts of 8 to 10, lower-redshift sources appear to have dust in excess. Early observations from a Hawaiian telescope (the JCMT) found 'monster' galaxies where dust obscured so much star formation that even super-computer models still cannot reproduce them. Worse, these sources were often completely invisible in optical light. It became clear that star formation occurs almost everywhere across these galaxies — and that there is a suspicious lack of them in our local Universe.
1. Dust in the first billion years
Even 600 million years after the Big Bang, much of galaxy growth is hidden behind cosmic dust. Characterising that dust — how much, how warm, how quickly it forms — is the core of my current research, and the measurement on which everything else on this page converges.
Warm dust at the highest redshifts
![Warm dust and [C II] emission in a z = 8.3 galaxy](images/research/SED_ALMA_B9.png)
I made one of the most distant detections of ionized carbon and uncovered unusually warm dust in a galaxy at z=8.3 (Bakx et al. 2020). New ALMA Band 9 observations later measured that dust at ~90 K and confirmed the galaxy as an ultra-luminous infrared galaxy — the most distant direct dust detection to date, with 93% of its star formation hidden from UV view (Bakx et al. 2025).
The temperature of cosmic dust
Using challenging short-wavelength ALMA observations, I made the first direct measurement of dust temperature in the Epoch of Reionization, at z=7.13 (Bakx et al. 2021). Dust temperature is the decisive lever: because infrared luminosity climbs steeply with temperature, a few warm galaxies can hide most of the light. It sets how much hidden star formation we infer, yet had only ever been guessed indirectly at these distances — where the field assumes a temperature, I measure it.
How quickly did dust form?
In PIXIEDust, I combined ~200 hours of ALMA and NOEMA data on the ten most distant galaxies into the deepest dust-stacking experiment in the early Universe to date (Bakx et al. 2025). The remarkably low dust limits point to inefficient dust build-up in the first 600 million years — a direct challenge to current chemical-evolution models, which I am now testing with dust-evolution modelling. Together with the warm ultra-luminous galaxy above, these two results bracket the question: somewhere between them, the Universe's star formation went dark.
2. Cosmic distances & the JWST + ALMA frontier
Before we can study a distant galaxy, we need to know how far away it is — its redshift. Distances are the prerequisite for every dust measurement above, and at the very edge of the observable Universe they take the world's best telescopes working together.
Pinpointing distances at scale
Measuring redshifts is expensive in telescope time, so I lead international redshift surveys on the world's largest (sub)mm arrays — together the largest sample of monster galaxies with known distances — and build the methods that make them efficient. A graphical redshift technique from IRAM 30m data (Bakx et al. 2020) and the "High-z Sudoku" diagnostic (Bakx & Dannerbauer 2022) roughly double the efficiency of (sub)mm redshift searches and have been adopted by other groups. This underpins the large NOEMA (z-GAL) and ALMA (BEARS) survey collaborations I help lead.
Reaching the first galaxies with JWST + ALMA
Combining the James Webb Space Telescope with ALMA reaches the very edge of the observable Universe. In the first such synergy, I worked hard to spectroscopically confirm a z > 12 galaxy — among the most distant ever spectroscopically confirmed (Bakx et al. 2023). I also asked why photometric and spectroscopic redshifts of the earliest galaxies sometimes disagree, tracing it to an Eddington bias first described back in 1913 (Serjeant & Bakx 2023, Nature Astronomy). These results contribute to landmark confirmations of the most distant galaxies known (Castellano et al. 2024) out to extreme redshifts of even z=14 (Carniani et al. 2024).
3. Monster galaxies: the Universe's most extreme star factories
In 1997, far-infrared surveys revealed a new kind of galaxy: home to the most violent bursts of star formation in the cosmos, yet nearly invisible in optical light. These galaxies are both a subject in their own right and the calibration laboratory for my early-Universe census: bright enough to measure dust physics in detail, and — through gravitational lensing — resolvable down to the scales where stars actually form. For my PhD I catalogued the 209 brightest, most distant such galaxies from the 660 deg² Herschel-ATLAS survey — the Herschel Bright Sources (HerBS) sample (Bakx et al. 2018). Many are so bright that current models struggle to explain them at all.
Most turn out to be gravitationally lensed by a foreground galaxy. I developed a statistical method to identify those foreground lenses (Bakx et al. 2020), later verified it observationally, and extrapolated it to an estimated ~3000 lenses across the Herschel fields (FLASH, Bakx et al. 2024). Because sub-mm-selected lenses are picked purely on the background source, they offer a clean, complementary probe of cosmology.
Characterising hundreds of these galaxies one by one is impractical, so I built an efficient way to survey them: the ANGELS strategy detected 66 emission and absorption lines across 19 galaxies in six ALMA bands in just six and a half hours (Bakx et al. 2024) — a template for the kind of large, systematic studies the upcoming ALMA Wideband Sensitivity Upgrade will enable.
4. What galaxies are made of — and where they live
Gas conditions & resolved studies
Spectral lines reveal the conditions inside these galaxies — their gas density, radiation field and dynamics. Selecting galaxies by their [C II] emission, rather than by UV or infrared light, gives a far less biased census of the first billion years; doing so, I find more moderate gas conditions than UV-selected studies had suggested (Bakx et al. 2024). With resolved observations and gravitational-lensing reconstructions, I trace gas and star formation down to the scale of individual giant molecular clouds (~200 pc) — revealing super-linear star-formation relations at the highest gas densities — and large [O III] surveys at z=3–4 (Breathless BEARS, Bakx et al. 2026) connect these detailed views to the wider galaxy population (Harikane et al. 2020). These resolved studies set the clock of my census: how long galaxies spend in their brief, dust-hidden phase.
Galaxies in crowds: overdensities & protoclusters
Galaxies are not built in isolation. The most massive systems often sit in dense cosmic neighbourhoods — overdensities and protoclusters — that mark where the largest structures of today's Universe first assembled. I discovered a dusty protocluster surrounding the binary galaxy HerBS-70 at z=2.3 (Bakx et al. 2024; Press release), and its follow-up grew into my ongoing 40+ hour ALMA programme ProtoCluster 3D, which maps 44 candidate structures in three dimensions with [C II] — building the homogeneous census of early overdensities the field currently lacks. At far higher redshift, I help confirm and characterise the cores of the earliest known overdensities, including the z=7.88 structure revealed with NIRSpec/JWST (RIOJA, Hashimoto et al. 2023). Because the most distant dust detections all sit in crowded regions, this environment map is what turns individual discoveries into a fair census of the whole Universe.
5. Building the instruments
The science above is only possible with the right hardware, so I also help build the next generation of (sub)mm instruments. I contributed to DESHIMA, an on-chip superconducting spectrometer demonstrated on the ASTE telescope (Endo et al. 2019, Nature Astronomy), and I produced the first scientific paper using ALMA's new Band 2 receiver (Bakx et al. 2026).
As co-leader of the ALMA2040 high-redshift working group — 250 participants across five continents — I help guide ALMA's development over the coming decades, and I compiled the technical papers behind every ALMA receiver to give that instrumental work proper visibility (Bakx & Conway 2024). I am also a Co-I on the DESHIMA, TYFOON, FINER and AtLAST instrument and telescope projects — ensuring the questions raised above can be answered with tomorrow's facilities.
6. Where this is heading
The threads above converge on one measurement: the fraction of star formation hidden by dust across the first billion years. My current results bracket it — a galaxy at z=8.3 forming 93% of its stars behind warm dust on one side, the remarkably dust-poor PIXIEDust stack on the other — and my ongoing programmes are designed to close that bracket: short-wavelength ALMA campaigns that measure dust temperatures rather than assume them, deeper stacking experiments positioned on JWST catalogues, and the ProtoCluster 3D environment census. Together they will answer when the Universe's star formation went dark — and whether that happened before or after the Universe reionized. Every stellar-mass, dust and metal budget of the first billion years will need to be weighed against that measurement, and the instruments I help design will carry the census to the fainter galaxies that make up most of the Universe.