No. 008 · Stars & statistics

The stars whose winds fell off a cliff

Hubble measured the winds of 29 massive stars in metal-poor galaxies and found them about half as fast as their metal-rich twins, when the textbook trend predicted a tenth. Spotting a drop that big takes about 10 stars. Spotting the drop everyone expected would have taken 180.

The hottest, heaviest stars live fast and die young, and while they live they blow. Starlight pushes gas off their surfaces at thousands of kilometres per second, a stellar wind strong enough to strip away a star's outer layers and to stir the galaxy around it. The push works through metal atoms (to an astronomer, anything heavier than helium), which catch the light like sails.

Fewer metals should mean weaker winds. That matters because the first galaxies had almost no metals, and the James Webb Space Telescope keeps finding them stranger than the models built from our own metal-rich galaxy predict. Nobody can see single stars that far away. So a team led by Grace Telford at the University of Utah went looking for the next best thing: hot blue O stars in small, nearby dwarf galaxies that never got around to making many metals.

The survey is called TEMPOS. It gathered Hubble ultraviolet spectra of 29 O stars in six dwarf galaxies, all with less than a fifth of the Sun's metal content, and it was published in September in the Astrophysical Journal Supplement Series.

“It’s a sample of 29 stars, which doesn’t sound like a lot.”

That is Telford, speaking to the University of Utah's news office. Each star cost up to 35 hours of telescope time. The interesting question is whether 29 was enough to find anything at all.

Reading a speed out of light

A wind announces itself in the spectrum. Gas between us and the star is rushing toward us, so it absorbs light at slightly shorter, bluer wavelengths than it would at rest. The fastest gas absorbs the bluest light, so the blue edge of that absorption marks the wind's top speed. For one star in the galaxy IC 1613 the edge of a carbon line sits nearly ten angstroms blue of where it belongs, which works out to about 1,900 kilometres per second.

Now compare twins. That star and one in the galaxy Sextans A are both classed as O7.5 giants: the same temperature, the same stage of life. IC 1613 has about 14 percent of the Sun's oxygen; Sextans A has 8. A smooth trend measured on 149 stars in the Magellanic Clouds says wind speed scales with metal content to the power 0.22, so the Sextans A twin should be about 12 percent slower. It is closer to 30 percent slower.

Across six matched pairs in the survey, five fall further than the trend allows, and a typical twin is about 43 percent slower. Across all 25 stars with model-fitted speeds, the extremely metal-poor group averages 789 kilometres per second against 1,467 for the rest. The paper calls the steeper drop tentative, which is the right word. But it is a big drop.

Why the size of the drop decides the size of the sample

Here is the part that turns a sample of 29 from thin to plenty. How many stars you need to see a difference depends on how big the difference is compared with the scatter between individual stars. A long-standing statistician's rule of thumb puts it as sixteen times the square of the scatter divided by the square of the difference, per group.

The scatter in these wind speeds is about 340 kilometres per second. The trend predicted a difference of roughly 140. Plug that in and you need about 90 stars per group, 180 in all: at TEMPOS's rate of about nine Hubble orbits per new star, some 1,650 orbits, more than three times what Hubble's flagship massive-star program spent. The difference that actually turned up was about 680 kilometres per second. That needs 5 stars per group.

The difference is 4.7 times bigger, so the sample can be 22 times smaller: the square law. Twenty-nine stars was enough because the effect was large. Had the winds behaved the way everyone expected, this survey could never have told.

The part that isn't settled

“Metal content” turns out to be more than one number. It is usually measured with oxygen, because glowing gas makes oxygen easy to see. But iron does much of the pushing. In the galaxy NGC 3109 the two disagree badly: about a tenth of the Sun's oxygen and a fifth of its iron. And NGC 3109's stars have the fastest winds in the survey.

It is tempting to say iron explains it. The trouble is that all three NGC 3109 stars observed are supergiants or bright giants, the most luminous kinds of O star, which drive fast winds anywhere. Iron and luminosity are tangled together, and three stars cannot untangle them. The honest verdict is that the evidence does not yet decide. The team plans to measure the stars' own iron directly with new optical spectra from Keck.

Sources

O. G. Telford et al., “The Treasury of Extremely Metal-Poor O Stars,” Astrophysical Journal Supplement Series (2026), doi:10.3847/1538-4365/ae95f6; arXiv:2609.20928. Source of every measured figure: wind speeds from its Table 5, oxygen and iron from Table 1, star types from Table 2, 110 new orbits for 12 stars from Section 4. L. Potter, “Galaxies in the early universe are weird. These stars may explain why,” @theU, University of Utah, 21 September 2026 (the quotation). C. Hawcroft et al. 2024, Astronomy & Astrophysics 688, A105 (the 0.22 trend). R. Lehr 1992, Statistics in Medicine 11, 1099 (the sample-size rule). Modelled, not measured: the trend-predicted speeds (an extrapolation below where the trend was measured), the sample sizes (a rule of thumb assuming bell-shaped scatter) and the orbit costs (TEMPOS’s average rate). The calculator uses the same rule.