Hubble finds dwarf galaxies have lost most of their oxygen
NGC 4449, one of the 37 dwarf galaxies in the GLOW sample. Credit: NASA, ESA, Hubble Heritage and Hubble Collaboration, Alessandra Aloisi (STScI, ESA), via NASA.
A galaxy's oxygen budget can be computed before anyone looks through a telescope. Stars make oxygen in their cores, so the quantity a galaxy ought to contain follows from how much stellar mass it has ever formed. Set that expectation against the gas, dust and starlight instruments actually turn up, and the accounts rarely balance. For Milky Way-like galaxies the shortfall is smaller, some of the missing material sitting in the halo of gas around the disk. For the smallest galaxies in the neighbourhood the figures have never come close. A survey of 37 dwarf galaxies within roughly 19.5 million light years of Earth now puts a number on the gap, and it is a large one: on average these systems keep only about a ninth of the oxygen their stars produced.
Oxygen is the standard stand-in for the whole family of heavy elements. It is abundant and comparatively straightforward to detect, and its production is well understood, so a measured abundance reads as a census of everything stars have forged. The choice of dwarf galaxies matters more than the choice of tracer. These systems occupy the mass range in which feedback from massive stars and supernovae shapes how a galaxy grows. They are also the ones thought to have done much of the reionization of the early universe, and there are enough of them that their outflows could seed the intergalactic medium with metals. A study from 2014, as the NASA release recounts, had already found that Milky Way-like star-forming galaxies shed much of their oxygen, some of it detectable in the diffuse haloes around their disks. The new work extends that accounting to low-mass galaxies across a wide range of environments, and finds retention fractions much smaller than the simulations predicted.
The project is called GLOW, for Galaxies Losing Oxygen via Winds, funded by NASA and the National Science Foundation. Kristen Wingfield McQuinn leads it. She is also mission head of the science operations centre for NASA's Nancy Grace Roman Space Telescope, based at the Telescope Science Institute in Baltimore, and an associate professor at Rutgers. Grace Telford of the University of Utah is co-investigator, and she set out the puzzle plainly in a NASA release: the oxygen had to have been inside these galaxies at some point, so the only way it can be absent now is if it was pushed out, which takes energy from massive stars and supernovae.
The measurement is bookkeeping across several archives. Hubble's optical images resolve individual stars in these galaxies, which is what makes the sample possible. Fitting the colour-magnitude diagrams against stellar evolution models then yields each galaxy's history of star formation and how its metallicity has climbed over time. From that the team estimates the total stellar mass ever formed, corrects for the share returned to the interstellar medium by winds and supernovae, and multiplies by a yield of newly synthesized oxygen per solar mass of gas turned into stars, 0.010 at low metallicity. Alongside that they add up the oxygen actually present: in the gas, from the Very Large Array and the Australian Telescope Compact Array; in the stars and stellar remnants, from the resolved populations Hubble photographed; and in the dust, while the stellar masses behind the calculation come from the retired Spitzer Space Telescope. The sample is restricted to systems within about 6 megaparsecs, close enough for the telescope to pick out individual stars, and to a stellar mass range of roughly 3 million to 3 billion Suns, where simulations expect the largest fraction of the oxygen to be blown out. The resulting census found retention running from 4 to 25 percent, mean near 11 percent, with most of the oxygen that survives sitting in the gas, ahead of the stars and the dust. NGC 4449, an irregular galaxy in the sample, retained about 17 percent, with 8.9 percent in its gas, 6.0 percent in its stars and 1.8 percent in dust.
The gap comes down to three possible explanations. The first is the circumgalactic medium, the diffuse halo of gas outside the disk. Surveys using the Cosmic Origins Spectrograph to hunt for highly ionized oxygen in absorption against background quasars find only 1 to 7 percent of the oxygen these galaxies made in their hot haloes, and correcting for cooler gas phases pushes the halo budget to perhaps 5 to 35 percent. Even the generous end of that range leaves most of the oxygen unaccounted for. The second possibility is that the models are wrong, and this is where the result gets uncomfortable: three separate cosmological simulations, all reproducing the observed mass-metallicity relation for these galaxies, predict that the dwarfs keep considerably more of their metals than the observations show. The third is that the models get the mass-metallicity relation right but the physics of the winds wrong.
That is the option the team argues for. They fitted a simple regulator model of galaxy evolution, tuning the wind's mass-loading factor, how enriched the expelled material is, and how much gas flows back onto a galaxy from the intergalactic medium. The best fit, to the data and to the stellar mass-halo mass relation seen in Milky Way satellites, gives a wind mass-loading factor that scales with stellar mass as a power of minus 0.45. The expelled material carries twice the metallicity of the gas it left, and accretion onto these low-mass systems runs 40 percent below the universal baryon fraction. The full paper sets out the uncertainties: the halo measurements that constrain where the oxygen went carry large corrections for ionisation state and for how much of the halo is filled with gas.
The result settles one thing: the models used for these systems have been too generous. Every simulation checked reproduces the mass-metallicity relation and still overpredicts retention, which puts the error in the feedback, not the mass-metallicity relation. Whether most of the missing oxygen has reached the intergalactic medium is still open, since the halo measurements carry large uncertainties, but the balance of the evidence points to a good deal of it having left the gravitational reach of these galaxies altogether. A contemporary summary put the headline plainly: gas-rich dwarfs hold about a tenth of the oxygen their stars have made.
The next steps are already funded. McQuinn has an approved programme to use the Roman Space Telescope to find fainter and smaller systems across a wide swath of sky, and Hubble and the James Webb Space Telescope would then be pointed at their stars to reconstruct how they formed. Telford is going the other way, using Hubble to zoom in on the massive stars that produce the oxygen and blow it out, to work out how those stars differ in small galaxies. The question the paper leaves open is quantitative and testable: how much of the oxygen a dwarf galaxy makes ends up in the halo, and how much is gone for good.
Featured blog posts
A Transformer model called EarlyDetect, trained on SDO/HMI helioseismic and magnetic observations by an NJIT-led team including Princeton and NASA Ames, forecasts the emergence of solar active regions on the Sun an average of 9.24 hours before they become visible, outperforming LSTM baselines and revealing that temporal smoothing filters hid the faintest precursor signals (Tirona et al., JGR Machine Learning and Computation, 14 August 2026) 
Sixty-three relays over your head: how the U.S. Space Development Agency quietly built a sky internet. Posted on May 24, 2026
New observations from the James Webb Space Telescope are providing the clearest and most detailed view yet of the environment where these carbon structures form and evolve. 
Engineers and scientists are assembling a spacecraft designed for a specific and increasingly important purpose: finding potentially hazardous objects before they find Earth. Posted on May 4, 2026
An international team of researchers announced that NASA's Curiosity rover had identified more than 20 distinct organic molecules preserved in ancient Martian rocks. 
To truly shorten the distances between planets, something more powerful is required—something that does not merely burn fuel, but transforms matter itself into energy. 
There are moments in engineering when progress is obvious. And then there are moments when progress looks like subtraction. 
For as long as humans have pushed aircraft beyond the speed of sound, there has been a cost to that achievement—an invisible but unmistakable shockwave that ripples across the sky and crashes into the ground as a sonic boom. 
There are moments in the history of technology when an idea appears so simple in form and so vast in implication that it changes how we think about the future. 
The American Institute of Aeronautics and Astronautics has released a groundbreaking report identifying ten technologies that will fundamentally reshape aerospace. 
Every propulsion revolution has been delayed not by imagination, but by power. 
The LASER taught us that energy does not need to explode outward to be useful. It can be channeled. It can be persuaded to exit matter in an orderly way. 
The story of the LASER does not begin with light, but with order. 
Today we are joined byYasunori Yamazaki, Chief Business Officer at Axelspace. Axelspace are pioneers of microsatellite technology advancing the frontiers of space business, reimagining traditional ways of using space, and creating a society where everyone on our planet can make space part of their life. 
Today we are joined by Logan Ryan Golema, Founder & Principal, and Vishal Singh, Chief Scientist at Lunargistics. Latest blog posts
McQuinn and colleagues used Hubble archival data, Spitzer, the Very Large Array and the Australian Telescope Compact Array to build an oxygen budget for 37 nearby dwarf galaxies and found that they retain only 4 to 25 percent of the oxygen their stars produced, a mean of about 11 percent, with most of the retained oxygen sitting in the gas, which implies supernova-driven winds strip low-mass galaxies far more efficiently than current simulations allow 
NASA's twin ESCAPADE spacecraft will execute the first powered Earth flyby ever attempted by a Mars-bound mission, burning their engines at closest approach to Earth on 7 and 9 November 2026 to trade Earth's gravity for speed toward Mars, a route reached by parking at the Earth-Sun L2 point for a year instead of waiting for the 26-month launch alignment 
Juno's Waves instrument caught an ion-acoustic wave at 5,610 Hz carrying clean second and third harmonics, and 18 to 20 electron cyclotron drift harmonics where Earth-bound instruments typically find three. 
SpaceX's Starship V4, the 142-metre 42-engine version of the rocket due to fly in 2027, is the configuration the Artemis IV architecture rests on, because the six-tanker-flight manifest Musk has been quoting only adds up if a single tanker can put 200 tonnes into orbit, and on November 7, 2026 Musk floated swapping the booster's outer ring to a 500-ton-thrust Raptor 4 
Vienna's prototype locked a 148-nanometre laser to a single 8.4-electron-volt transition inside a thorium nucleus and held it steady for a day without help from an atomic clock, and a Chinese group built a rival system in the same issue of Nature. The clock still trails the best optical atomic clocks on stability, but its sensitivity to drifting fundamental constants already beats theirs by orders of magnitude. 
A $10 billion delivery on-ramp, $1.3 billion in launch task orders and a $94 million rover award turn NASA's Moon Base from an architecture exercise into a procurement program that runs on commercial landers. 
"NASA's Voyager 1 crosses a distance of one light-day from Earth on 18 November 2026, 49 years after launch, becoming the first human-made object that far away, while running on two science instruments after a September power swap and an unexplained 48-hour signal blackout showed how thin the power margins have become"' 
Japan's Martian Moons eXploration spacecraft is on its way to the smallest of Mars's moons with a plan to land on it, scoop up ten grams of rubble, and drop the capsule into the Australian desert, giving laboratories their first physical material from the Mars system. 
NASA's Artemis III, launching in 2027 as a 14-day low Earth orbit mission, will dock Orion with a Blue Origin Blue Moon Mark 2 test article and a SpaceX Starship Version 3 test article in two back-to-back rendezvous demonstrations using two different docking geometries, with Orion controlling the docked stack for the Blue Moon leg and Starship controlling it for its own leg, per the Human Landing System program's July 2026 description and the two HLS engineers featured on episode 443 of the Houston We Have a Podcast 
Pulled out of survey images by software rather than by eye, the object 2026 ED138 sits 47 au from the Sun on an orbit tilted nearly 25 degrees out of the ecliptic, and its orbit says it has not been this close to the Sun since 1926. Whether it counts as a planet depends on an infrared measurement nobody has made yet. 
NASA's 2001 Mars Odyssey marks 25 years at Mars on October 24, 2026, still mapping the surface with THEMIS and still relaying rover data, but its share of the Mars Relay Network has shrunk to the smallest of any active orbiter while NASA runs out of hydrazine and Congress has to decide whether to keep funding it 
India's NavIC navigation system had shrunk to three working satellites after clock failures and a spacecraft that never reached its orbit; the 2.2-tonne NVS-03 launched on 30 October restores the four-satellite floor and adds an L1 signal that ordinary phone chipsets already understand. 
NASA's Ultraviolet Explorer cleared its make-or-break review on October 27, committing the agency to a launch by January 2031 at a lifecycle cost of $590 million. With a 75-centimetre mirror and newly developed ultraviolet CMOS detectors, the observatory will reach fifty to a hundred times deeper than the survey it inherits, and turn neutron-star mergers and supernovae into spectra within hours of an alert. 
Encoding one logical qubit into 73 trapped-ion qubits, Quantinuum found the failure rate falls by about twelve for every two-step increase in code distance, leaving a logical qubit roughly a hundred times more reliable than the physical qubits underneath it. The slope, published on 19 October 2026, is far steeper than the factor of two that superconducting surface codes had reached, which is what turns error correction from a physics result into an engineering plan. 
"NASA's Neil Gehrels Swift Observatory falls back into Earth's atmosphere on 4 November 2026, nearly 22 years after launch, after a $30 million commercial boost attempt with Katalyst's LINK servicing spacecraft failed when the tug tumbled out of control and NASA gave up on the capture"' 
Subscribe to blog posts using RSS
