Deep divefrom Episode 7Cosmology16 October 202623 min

The Hubble Constant From Red Giants Alone, No Supernovae

A deep dive into one paper: what was known before, what the researchers did, what the numbers say and what they don't.

The paper: Two-rung ladder: H0 from Tip of the Red Giant Branch and geometric anchors alone (Stiskalek, Desmond & Lavaux, arXiv preprint)

MP3

Chapters

Transcript

The script the two AI voices read from. The times are worked out from the script, not heard, so a line can start a few seconds early or late.

Read the whole episode

Cold open

Host 1You heard the trailer — this is the whole story. Welcome to the Hastrup Review of Science — where artificial intelligence reads real science, so you don't have to pretend you understood the paper. Today, an attempt to measure how fast the universe is expanding with only two steps, and no exploding stars at all.

Host 2The paper is called Two-rung ladder: the Hubble constant from Tip of the Red Giant Branch and geometric anchors alone. The authors are Richard Stiskalek at Oxford, Harry Desmond at Portsmouth, and Guilhem Lavaux at the Institut d'Astrophysique de Paris. It was posted to arXiv on September twenty-fifth, twenty twenty-six.

Host 1So a preprint.

Host 2Yes, formatted for the Monthly Notices of the Royal Astronomical Society but not yet peer reviewed. One more thing: the copy we worked from cuts off partway through the discussion, so the conclusions section and appendices are not covered here. Where I add textbook background, I will say so.

Host 1The claim, in two sentences.

Host 2Using about four hundred nearby galaxies with red giant distances, and no supernovae, they get a Hubble constant of sixty-eight point four plus or minus five point nine, or seventy-two point three plus or minus two point six, depending on how you combine their maps of the local universe. Their own conclusion is that this ladder cannot yet settle the Hubble tension, because the uncertainty comes from our map of how nearby galaxies move, not from the stars.

Why this field exists

Host 1Start with the basics. What is the Hubble constant?

Host 2Background, not from this paper: it is today's expansion rate of the universe, in kilometers per second per megaparsec. A megaparsec is about three point two six million light years. A value of seventy means a galaxy one megaparsec away recedes, on average, at seventy kilometers per second, and one ten megaparsecs away at seven hundred.

Host 1And to measure it you need two things for each galaxy.

Host 2A speed and a distance. The speed is easy: you measure the redshift of its light. The distance is the hard part, and astronomy's answer for a century has been the distance ladder.

Host 1Explain the ladder.

Host 2Background again. You start with objects whose distance you can measure by pure geometry, then use them to calibrate a type of star whose true brightness is predictable, then use those stars to calibrate something brighter that you can see farther away. Each step is a rung, and errors from the lower rungs carry upward.

Host 1What are the usual rungs?

Host 2In the two leading programmes, rung one is geometry, rung two is either pulsating Cepheid stars or the tip of the red giant branch, and rung three is Type Ia supernovae, which are bright enough to reach galaxies hundreds of megaparsecs away. This paper deliberately throws away rung three.

What was known before

Host 1Why would you throw away your brightest rung?

Host 2Because of the Hubble tension. Background figures: the Planck satellite's cosmic microwave background data, interpreted with the standard cosmological model, give about sixty-seven point four. The supernova team called Shoes, led by Adam Riess, measures about seventy-three.

Host 1And the paper's framing of that disagreement?

Host 2The paper says the Shoes result, using Cepheids and supernovae, sits about five sigma from Planck. The Carnegie-Chicago Hubble Program, led by Wendy Freedman, uses red giants and supernovae, and sits only about one point six sigma from Planck. Yet the two local results agree with each other at about one point two sigma.

Host 1So both local measurements share the supernova step.

Host 2Exactly. The paper lists the supernova rung's own problems: standardising their brightness, which host galaxies get selected, correlated errors, dust, and calibrating their absolute brightness. If something is wrong there, it would affect both teams, which motivates ladders that skip supernovae entirely.

Host 1Has anyone done that?

Host 2Several groups. The paper notes that routes replacing supernovae with surface brightness fluctuations, the Tully-Fisher relation for spiral galaxies, or Type Two supernovae typically land between seventy-three and seventy-six. The same authors previously did a Cepheid-only ladder, Cepheids plus geometry plus redshifts, and got seventy-one point one plus or minus one point four.

Host 1And a red-giant-only ladder?

Host 2The closest earlier work, by Kim and colleagues in twenty twenty, used thirty-three galaxies between our Local Group and the Virgo cluster, modelled as falling into Virgo. They got sixty-five point eight, with a statistical error of three point five and a systematic error of two point four.

The open question

Host 1What is the tip of the red giant branch, physically?

Host 2When a Sun-like star runs out of hydrogen in its core, it swells into a red giant, with a core of helium held up by electron degeneracy pressure. As the core grows, it eventually ignites helium in a sudden flash, and that happens at a nearly fixed core mass, so at a fixed composition the star reaches a nearly fixed peak luminosity.

Host 1So in any old galaxy the brightest red giants all top out at the same brightness.

Host 2Right. Plot a galaxy's red giants by brightness and there is a sharp cut-off at the bright end: that edge is the tip. Measure how bright the tip looks, compare with how bright it really is, and you have a distance.

Host 1How long has this been used?

Host 2The paper traces it to Mould and Kristian in nineteen eighty-six, and to Lee and colleagues in nineteen ninety-three, who automated it with an edge-detection filter borrowed from image processing. Madore and Freedman showed in nineteen ninety-five that you need at least about a hundred stars within one magnitude of the tip for an unbiased measurement.

Host 1And why is it attractive compared with Cepheids?

Host 2Red giants live in the old halo populations of galaxies, so they have different dust, star formation and metallicity problems from Cepheids, which sit in dusty young discs. An independent stellar calibrator is a check on the other.

Host 1So the open question is simple: can red giants plus geometry alone give a useful Hubble constant?

Host 2And the catch is distance. Red giants are faint, so the sample is local: every host is within a recession velocity of about two thousand one hundred fifty kilometers per second, and most are within one thousand.

Host 1Why does being close matter?

Host 2Because galaxies also have their own motions, called peculiar velocities, of a few hundred kilometers per second, from being pulled by nearby clusters. For a galaxy receding at five hundred kilometers per second, that is a large fraction of the signal, so the whole measurement depends on modelling those motions.

What the paper did

Host 1Walk me through the data.

Host 2The red giant distances come from the Extragalactic Distance Database catalogue compiled by Anand and colleagues in twenty twenty-one, from Hubble Space Telescope images of five hundred fifty-six nearby galaxies, four hundred eighty-nine with a tip measurement. That is roughly ten times more hosts than the Shoes Cepheid sample.

Host 1And after cleaning?

Host 2They drop the anchor galaxy and its satellite, six galaxies without a usable tip magnitude, one missing a colour correction, and thirty-three with no redshift on record, leaving four hundred forty-five. Then they cut forty-four very nearby objects within about two megaparsecs, mostly Local Group members, more than half in the Andromeda subgroup, whose motions swamp any expansion signal.

Host 1Where does that leave the final count?

Host 2A mask along the plane of the Milky Way removes seven more, so three hundred ninety-four galaxies enter the analysis. Eighty-seven percent are within one thousand kilometers per second, and half are within five hundred sixty.

Host 1Now the two geometric anchors.

Host 2The first is the Large Magellanic Cloud, with a distance from eclipsing binary stars, a distance modulus of eighteen point four seven seven, plus or minus zero point zero two six magnitudes. Converted, that is about fifty kiloparsecs, known to roughly one percent.

Host 1And the second?

Host 2The spiral galaxy Messier one hundred six, which has a disc of water masers orbiting its central black hole. Tracking those orbits gives a geometric distance of about seven and a half megaparsecs, to about one and a half percent, and both anchors have their own measured red giant tips.

Host 1How do they turn all that into a number?

Host 2They build a forward model: a simulation of how the catalogue would have come out for any given set of parameters, including the Hubble constant, and they ask which parameters reproduce what was actually observed. Each galaxy gets an unknown true distance, which predicts both its tip brightness and its redshift.

Host 1The redshift prediction is where the motions come in.

Host 2Yes. The predicted redshift is the expansion at that distance plus the galaxy's peculiar velocity, read off a map of the local universe called Manticore Local. That map also tells the model where galaxies are likely to sit, since galaxies cluster where matter is dense.

Host 1What is Manticore Local?

Host 2A reconstruction of the density and velocity of matter around us, built from about sixty-nine thousand galaxies in an all-sky redshift survey. An algorithm called Borg infers the initial conditions of the early universe and runs gravity forward to the present, so the map is physically consistent rather than just smoothed.

Host 1And it does not give one map.

Host 2It gives eighty, each an equally plausible version of the local universe consistent with the survey. They agree on large-scale flows but differ on small scales, which is exactly where these nearby galaxies live.

Host 1What about which galaxies made it into the catalogue? Nobody designed this sample.

Host 2Correct, it was pieced together from many Hubble programmes with no single selection rule. So the authors posit one: the chance of a galaxy being included falls off for faint tips, with the faint limit and how gradually it falls off inferred from the data.

Host 1Anything for where on the sky Hubble happened to look?

Host 2Yes, a sky exposure term. They split the sky into forty-eight equal patches and let each have its own weight for how heavily it was observed, with a prior that allows factors of a few between patches but stops everything piling into one.

Host 1And the redshift errors?

Host 2They compare a normal bell-curve error with a Student-t distribution, which has heavier tails and tolerates outliers. They also include a single bulk flow vector for coherent motion the map misses.

Host 1How do they handle eighty maps?

Host 2They run the full analysis separately on each map, using a standard Hamiltonian Monte Carlo sampler with five chains, then combine the eighty results in two ways. One weights every map equally; the other weights each map by its Bayesian evidence, which is how well that map explains the data.

A word from our sponsor

Host 1Before the results, a word from our sponsor.

Host 2Do you have things that keep slipping away from you just when you have them where you want them? The Hastrup Review of Science is brought to you by clothes pegs.

Host 1I prefer the wooden ones. A good strong spring, and they grip firmly without leaving a mark.

Host 2Clothes pegs hold on in any weather, and they come in generous packs. They are very reasonably priced and in stock right now.

Host 1I keep a few by the bed. You never know when something will need to be held tightly in place.

Host 2Clothes pegs. Our referral link is in the show notes.

Results

Host 1So, the number.

Host 2Weighting all eighty maps equally, sixty-eight point four plus or minus five point nine. Weighting by evidence, seventy-two point three plus or minus two point six, in kilometers per second per megaparsec.

Host 1Those are quite different. Where does the spread come from?

Host 2Each individual map gives a Hubble constant with an uncertainty of about two point six. But the best estimates from the eighty maps scatter around by five point two, and the equal-weight error of five point nine is essentially those two added in quadrature.

Host 1So the disagreement between maps is the dominant error.

Host 2Yes, and five point two is comparable to the whole gap between Planck and Shoes that this measurement was meant to judge.

Host 1And the evidence-weighted value. Why is it so much tighter?

Host 2Because one map wins outright. The effective number of maps contributing is one point zero zero out of eighty, and the next four best maps give central values anywhere from sixty-six point seven to eighty-two point two.

Host 1So seventy-two point three is essentially one map's answer.

Host 2That is the authors' own reading. With about four hundred galaxies, small differences in how well each map explains each galaxy multiply up, so one map pulls far ahead, and with a different set of eighty draws a different map would likely win.

Host 1Which number should we believe?

Host 2The paper does not pick one. The evidence-weighted value is the formally correct average but estimated from too few samples; the equal-weight value is not a proper average but shows the full spread, so together they bracket how much the answer depends on the map.

Host 1What about the star calibration itself?

Host 2The tip's absolute brightness comes out at minus four point zero three plus or minus zero point zero eight magnitudes, with an intrinsic scatter of zero point one zero magnitudes, stable across every variant. But the scatter was given an informative prior from a separate fit, so its stability is partly built in.

Host 1And the selection model. Did the data constrain it?

Host 2Yes: the faint limit comes out at twenty-four point zero seven magnitudes, inside the observed range, with a roll-off width of zero point nine three. So the data actually see where the catalogue thins out.

Robustness checks

Host 1Now the robustness checks. Start with the error distribution.

Host 2The heavy-tailed Student-t is decisively preferred over the bell curve. Its tail parameter is two point five nine, far from Gaussian, and the authors think the outliers are galaxies falling into Virgo, Fornax and Centaurus A, whose fast internal motions the map cannot resolve.

Host 1Does that choice change the answer?

Host 2Evidence-weighted, the bell curve gives sixty-seven point zero versus seventy-two point three. But map by map the two differ by only zero point eight plus or minus zero point nine, so the jump comes from a different map winning, not from the physics of the likelihood.

Host 1Smoothing the density map?

Host 2Map by map, going from four to eight megaparsecs of smoothing, in the map's units, moves the result by about a tenth of a kilometer per second per megaparsec, and removing smoothing by about a quarter. Yet removing it changes the evidence-weighted value to sixty-four point eight, again because a different map wins.

Host 1This is a pattern.

Host 2It is the paper's central point. At a fixed map, the error model, smoothing and sky weighting each move the answer by less than one, but across those same variants the evidence-weighted value ranges from sixty-three point five to seventy-two point four.

Host 1And the sky-exposure term?

Host 2Strongly preferred by the evidence, and needed for the simulated catalogues to look like the real one, yet it leaves the Hubble constant unchanged. Without it, a statistical test of the redshift distribution gives a p-value of zero point zero one two; with it, zero point two three.

Host 1Did everything fit that well?

Host 2Not quite. The magnitude distribution only matches marginally, with a p-value around zero point zero seven, because the real data have two peaks, near twenty-four and twenty-six magnitudes, that the smoother model does not reproduce.

The flow of the local universe

Host 1You mentioned the velocity model moves things more.

Host 2Yes. They tried adding a velocity monopole, a uniform inflow or outflow towards us, which looks a lot like a change in expansion rate. Map by map it raises the Hubble constant by five point four plus or minus three point four, and ninety-four percent of the maps move up.

Host 1So the headline is fragile to that term.

Host 2The equal-weight value goes to seventy-four point seven plus or minus five point two, while the evidence-weighted value moves only to seventy-three point six, because the winning map happens to shift less. The evidence neither favours nor disfavours the term, and the inferred inflow is consistent with zero.

Host 1What if you drop the map entirely?

Host 2With no reconstruction and only a single bulk flow, they get seventy point four plus or minus two point nine. But that model is overwhelmingly rejected by the evidence, and the bulk flow has to soak up a large five hundred sixty-two kilometers per second of local motion.

Host 1There is a result here I liked, about our own motion.

Host 2Yes, it is a good internal check. The map predicts our own velocity as three hundred sixty-eight kilometers per second, but the Local Group is known to move at six hundred twenty relative to the cosmic microwave background, so about two hundred fifty is missing.

Host 1And the bulk flow fills the gap.

Host 2Added to the map's velocity, the inferred bulk flow gives six hundred sixteen plus or minus twenty-nine kilometers per second, within two point seven degrees of the direction Planck measures. The red giants recover our motion through space independently, which suggests the model is doing something physically sensible.

Host 1But it also says the map is incomplete.

Host 2Exactly. The paper calls the reconstructed flow at our position colder than the dipole requires, likely because part of our motion is driven by structure beyond the survey volume.

Why it matters

Host 1Step back. Why does this matter?

Host 2The principle is that for very nearby galaxies, the bottleneck for measuring expansion is no longer the stars, it is our map of local motions. The authors argue that tip calibration and selection only start to matter once that limit is lifted.

Host 1And for the Hubble tension specifically?

Host 2It adds an independent route, with no supernovae and no Cepheids, but one not yet sharp enough to decide anything. The equal-weight result is consistent with both Planck and Shoes.

Host 1How does it compare with the other red giant results?

Host 2It agrees with the Carnegie-Chicago value of seventy point four, with a reanalysis called Cats at seventy-three point two, and with a ladder of supernovae in quiet elliptical galaxies at seventy-five point three. But it is about one point two to one point four times less precise than those when evidence-weighted, and about three times less precise when equal-weighted, and the authors say the agreement follows largely from that.

Host 1What does it not mean?

Host 2It does not mean the Hubble constant is sixty-eight, or seventy-two. It does not mean the red giant method is flawed, and it does not mean supernovae are vindicated or refuted; it is a careful statement about what this sample can and cannot do.

Caveats

Host 1The caveats.

Host 2First, the headline evidence-weighted number rests on a single map out of eighty. The authors say so plainly, and a different draw of eighty maps would likely give a different winner.

Host 1Second?

Host 2The selection function is assumed, not known. The archive does not record how the sample was assembled, so the magnitude window and the forty-eight sky weights are a posited stand-in for Hubble's real pointing history.

Host 1Third?

Host 2The thirty-three galaxies dropped for missing redshifts are not random: they are dwarf companions in the Messier eighty-one and Centaurus A groups at three to ten megaparsecs. The authors argue this should not bias the calibration, but say they did not test its effect on the sample's distribution.

Host 1And the galaxy bias model?

Host 2That is the rule mapping matter density to how many galaxies you expect. It has never been calibrated for nearby low-mass galaxies, and the authors cite work showing simpler versions already fail for dark matter halos, though they show the equal-weight result moves by less than one across the bias variants they tried.

Host 1Anything on the star side?

Host 2The colour slope is fixed at zero point two rather than fitted, the dust correction's uncertainty is ignored, and different Hubble programmes used different edge-detection methods. By the paper's estimate, a tenth of a magnitude error in the tip's true brightness would shift the Hubble constant by about four and a half percent. But it would shift every map equally, so none of these can create the scatter between maps.

Host 1And the obvious one.

Host 2It is an unrefereed preprint, and our copy of the text ends partway through the comparison with earlier work, so anything in the conclusions or the mock-data validation appendix is not reflected here.

What to watch next

Host 1What should we watch for next?

Host 2Deeper red giant samples. The James Webb Space Telescope has already measured a tip at thirty plus or minus four megaparsecs, and the catalogue authors project tip distances out to thirty or forty megaparsecs with modest exposures. At those distances peculiar velocities are a few percent of the signal, not tens of percent.

Host 1Is there a catch?

Host 2Two. That Webb measurement was a favourable case, a quiet dwarf galaxy with ninety-four usable stars, and crowded discs will need longer exposures. And current deep programmes pick galaxies because they host supernovae, so their selection comes from the third rung; the paper points to the proposed Roman Space Telescope wide-area programme as one with a well-specified selection, even though its targets are inherited the same way.

Host 1And on the map side?

Host 2Reconstructions constrained by galaxies with measured distances in exactly this local volume, not just by redshift surveys. That has been demonstrated on mock data, and together with deeper samples the authors expect it to bring the scatter between maps well below five point two.

Sign-off

Host 1That is the deep dive: Two-rung ladder, the Hubble constant from the tip of the red giant branch and geometric anchors alone, by Stiskalek, Desmond and Lavaux. The link is in the show notes.

Host 2Short version: three hundred ninety-four nearby galaxies and two geometric anchors give sixty-eight point four plus or minus five point nine, or seventy-two point three from one dominant map. The stars are not the limit; the map of local motions is.

Host 1Every Friday this same feed carries the roundup episode: five new papers as trailers, and you pick which become deep dives. This has been the Hastrup Review of Science.