🚀 Space

Henrietta Leavitt's 1912 Data Was Re-Analyzed With Modern Tools. Her Scatter Dropped by Half.

Breuval, Huang, and Riess took the same 25 Cepheid variable stars Henrietta Leavitt measured from photographic plates 113 years ago and re-observed them with Gaia parallaxes and CCD photometry. The period-luminosity scatter fell by a factor of two. The foundation beneath every modern Hubble constant measurement just got a physical.

By Lena Okafor · Space Economy · June 26, 2026 · ☕ 8 min read

Twenty-five stars. That's it. That's the entire dataset underpinning one of the most important relationships in astrophysics, specifically the cosmic distance ladder that carries every measurement of the universe's expansion rate on its back. In 1912, Henrietta Swan Leavitt published Harvard College Observatory Circular No. 173, documenting a tight correlation between the pulsation period of Cepheid variable stars in the Small Magellanic Cloud and their apparent brightness: brighter Cepheids pulsed more slowly, and if you measured the period you could calculate the luminosity, and from the luminosity you could derive the distance, which meant that for the first time in history a human being had built a ruler that worked across intergalactic space. She built it from two dozen data points extracted by eye from glass photographic plates coated in silver halide emulsion.

Now Louise Breuval (ESA/STScI), Caroline Huang (Harvard CfA), and Adam Riess (STScI/Johns Hopkins) have gone back to those same 25 stars and re-measured them with 21st-century instruments. Gaia parallaxes, CCD photometry, automated light-curve fitting algorithms that would have seemed like science fiction to a woman sitting at a desk in Cambridge with a magnifying loupe and a notebook. Their paper in the Publications of the Astronomical Society of the Pacific (arXiv: 2502.17438) reports what Leavitt might have expected but couldn't prove: her results are in "excellent agreement" with modern observations, and when modern photometry replaces silver halide, the scatter around the period-luminosity relation drops by a factor of two.

From Glass Plates to Gaia

Leavitt never received proper credit. She was a "computer" at Harvard College Observatory, one of the women hired by director Edward Pickering to catalog stellar photographs at a fraction of what male astronomers earned, and her 1912 paper was published under Pickering's name because that was how things worked at Harvard's observatory when you were a woman doing the actual science while a man ran the department. She died of cancer in 1921 at age 53. When Gösta Mittag-Leffler attempted to nominate her for the Nobel Prize in 1926, he discovered she had been dead for five years.

What she left behind was a relationship so robust that Edwin Hubble used it seven years later to prove that the Andromeda nebula was a separate galaxy, and then again in 1929 to demonstrate that the universe was expanding, a discovery whose downstream implications included the Big Bang theory, the cosmic microwave background prediction, and eventually the accelerating expansion that won Riess the 2011 Nobel Prize. His SH0ES team's latest measurement (H₀ = 73.17 ± 0.86 km/s/Mpc) still runs through the Cepheid pipeline Leavitt invented.

So when critics of the Hubble constant suggest that systematic errors in Cepheid measurements might explain the tension with the cosmic microwave background value (67.4 ± 0.5 km/s/Mpc from Planck), they're arguing, implicitly, that something about the way we measure these pulsating stars has been consistently wrong since 1912.

Breuval's team decided to test that directly.

What They Did

They identified Leavitt's original 25 Cepheids in the SMC using modern catalogs (cross-matching through multiple catalog generations: Harvard designation to HD number to Gaia DR3 source ID), then re-observed them using CCD photometry and fitted contemporary period-luminosity relations to both datasets: Leavitt's original photographic magnitudes measured from glass plates under a desk lamp, and the modern measurements of the identical stars captured by instruments that can resolve individual photons across a dynamic range Leavitt's emulsions couldn't touch.

Two findings dominate.

First: scatter. Leavitt's photographic data produces a period-luminosity relation with roughly 0.4 magnitudes of residual between each star's measured brightness and the best-fit line. Modern data for the same 25 stars cuts that to approximately 0.2 magnitudes. Halved.

Second: bias. Leavitt's data shows a systematic brightward bias at the short-period end, where stars with fast pulsation and lower intrinsic luminosity appear brighter in her measurements than modern instruments record. Breuval attributes this to two factors that compound each other: the non-linear response of photographic plates (silver halide emulsions compress the brightness range at the faint end, clipping the dimmest stars upward toward the detection threshold) and crowding in the SMC (at Leavitt's photographic resolution, faint Cepheids blend with neighboring stars, inflating their apparent brightness by an amount that depends on the local stellar density). Neither error was avoidable in 1912.

What Halving the Scatter Actually Means

Here is an original calculation that puts the scatter reduction in human-scale terms. Astronomical magnitudes convert to distances via the relation d = 10^((m−M+5)/5), where m is apparent magnitude and M is absolute magnitude, and scatter in the period-luminosity relation propagates directly into distance uncertainty because any error in the inferred absolute magnitude translates, through that exponential, into a proportional error in the distance you calculate.

For a galaxy at the distance of the Small Magellanic Cloud (roughly 60 kiloparsecs), 0.4 magnitudes of PL scatter translates to a distance uncertainty of about ±12%. Cut that scatter to 0.2 mag: ±6%. Concretely, that's the difference between placing the SMC somewhere between 53 and 67 kpc versus pinning it between 56 and 64 kpc.

At cosmological distances where the Hubble constant is measured, these fractional errors compound through the distance ladder, with each rung's uncertainty adding in quadrature to the next. Tightening the first rung does not eliminate the tension between SH0ES and Planck. But it constrains the space where systematic errors could hide, and if the very first dataset, the 25 hand-measured stars from glass plates in 1912, agrees with modern CCD photometry to within the scatter reduction Breuval reports, the argument that Cepheid calibration harbors large hidden systematics loses much of its footing.

Strongest Counterargument

Critics of Cepheid-based distance measurements have a real point, and it has nothing to do with photographic plates. MF25, a 2025 proposal by Mortsell and Fransson, argues that Cepheid metallicity dependence (how a star's chemical composition affects its brightness) introduces a systematic bias that mimics a higher Hubble constant. If metal-rich Cepheids in host galaxies are intrinsically brighter than the calibrators in the Milky Way and Magellanic Clouds, the distance ladder underestimates distances and overestimates H₀.

Breuval herself has addressed this argument in a separate July 2025 paper defending the SH0ES metallicity corrections, but the debate is unresolved. And there is a more fundamental issue: re-analyzing Leavitt's 25 SMC stars doesn't test the metallicity question at all, because those 25 stars all sit in the same low-metallicity environment. Validating the PL relation in one galaxy doesn't validate the corrections applied when that relation is transported to galaxies with different chemical compositions.

In other words, Breuval's paper proves Leavitt was right about these 25 stars. It does not prove that the cosmic distance ladder built on top of them is free of systematic errors introduced at later rungs.

Limitations

This analysis rests on 25 stars in a single galaxy. It is, by design, a historical validation exercise, not a calibration improvement. Modern Cepheid analyses use thousands of stars across dozens of galaxies; nothing about Breuval's re-analysis changes those calibrations directly. The scatter reduction quantified here reflects the improvement from CCD photometry over photographic plates, a gain the field already assumed but had not formally measured against Leavitt's original sample.

Additionally, the brightward bias Breuval identifies at short periods could cut in either direction for the Hubble tension debate. If similar biases persist in modern photometric pipelines (from crowding in more distant, less-resolved galaxies) they would push H₀ higher. Breuval's team flags this but does not quantify the residual effect in contemporary surveys.

Finally, Leavitt's notebooks are the primary historical source, but their provenance is imperfect. Some of the star identifications required cross-matching through multiple catalog generations (Harvard designation → HD number → Gaia DR3 source ID). Breuval reports all 25 matches are secure, but any misidentification would corrupt the comparison.

The Hubble Tension, Briefly

For context: the Hubble tension is the disagreement between two methods of measuring how fast the universe is expanding, and the numbers are stubborn. SH0ES, using Cepheids and Type Ia supernovae as distance markers, measures H₀ at 73.17 ± 0.86 km/s/Mpc. Planck, using the cosmic microwave background (the afterglow of the Big Bang, observed from a satellite measuring temperature fluctuations across the entire sky with microkelvin precision), measures 67.4 ± 0.5. That discrepancy exceeds 5σ — a level where statisticians stop calling it a fluctuation and start calling it a crisis.

In January 2026, Högås proposed that using physically-motivated Bayesian priors could reduce the SH0ES value from 73.0 to 70.6, cutting the tension from 5σ to 2σ. Meanwhile, Casertano et al. (2026) demonstrated that no single measurement method, if excluded, shifts H₀ significantly, ruling out the "one bad measurement" explanation and leaving the field with a tension that is distributed across multiple independent rungs of the ladder, none of which breaks when you remove it.

Breuval's re-analysis enters this landscape as a meta-argument. It says: the pipeline works. The woman who started it was right. The instrument has improved but the signal was always there.

What You Can Do With This

If you teach astronomy, this paper is a gift. Leavitt's original 25 data points are reproduced in Breuval's supplementary material alongside their modern counterparts. You can assign students to plot both datasets, fit their own period-luminosity relations, and calculate the scatter reduction themselves. Few exercises better illustrate how scientific measurement improves without the underlying science changing.

If you follow the Hubble tension debate, watch for the next round of SH0ES papers using Gaia DR4 parallaxes, expected in late 2026 or 2027. Breuval's re-analysis sets a floor on how good Cepheid calibration can get in the SMC. If the tension persists when SMC Cepheids are measured at ±6% precision instead of ±12%, the explanation increasingly points toward new physics rather than bad measurements.

If you work in data science or metrology, consider the methodological precedent. Re-analyzing a 113-year-old dataset with modern tools is not common practice outside astronomy, but the logic applies anywhere foundational measurements anchor long chains of inference. Breuval's approach (same stars, same relationship, different instruments) is a template for validating whether legacy calibrations still hold.

The Bottom Line

Henrietta Leavitt measured 25 stars by squinting at glass plates under a desk lamp in Cambridge, Massachusetts, and got a result that held up for 113 years, across every revolution in observational astronomy from CCDs to space telescopes to billion-star parallax catalogs. Louise Breuval pointed those modern instruments at the same stars and found that Leavitt was right — that her scatter was twice what contemporary tools produce, and that the biases she couldn't avoid are now quantified, bounded, and small. None of this resolves the Hubble tension. What it does is eliminate a specific class of objections: the ones suggesting that the woman who started the distance ladder got the first rung wrong.