Deep divefrom Episode 9Cosmology6 November 202619 min

Watching Galaxies Fall Into Virgo to Weigh the Expansion

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: Debiasing Local Cluster Infall Reveals H0: Application to the Virgo Cluster (Carreres et al., 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, how fast the universe is expanding, measured by watching galaxies fall into the Virgo Cluster, and why the old way of doing that came out too low.

Host 2The paper is Debiasing Local Cluster Infall Reveals H nought: Application to the Virgo Cluster, by Bastien Carreres, David Benisty, Jenny Wagner, Avery Erlenbach, Dan Scolnic and Adam Riess, from Duke, Bar-Ilan, Helsinki, the Space Telescope Science Institute and Johns Hopkins. It is a preprint on arXiv, so it has not been through peer review yet. Their claim is that the standard recipes for this measurement are biased low, and that once you test them in a simulated universe and correct for it, the Virgo answer moves up.

Why measure the expansion

Host 1Let me start at the very beginning: what is the Hubble constant, and why does anyone care about one number so much?

Host 2It is the rate at which the universe is expanding today. In the wider picture, it goes back to Edwin Hubble and others in the nineteen twenties, who noticed that more distant galaxies recede faster, roughly in proportion to their distance. The constant is that proportion, and it sets the age and scale of everything in cosmology.

Host 1The unit always sounds strange to me, so what does kilometres per second per megaparsec actually mean?

Host 2A megaparsec is about three point three million light years. A value of seventy means a galaxy one megaparsec away recedes at about seventy kilometres per second, one twice as far recedes at about one hundred and forty, and so on. I will just say the number from here on, and it always means those units.

The tension

Host 1So what was already known before this paper, and why is there a fight about it?

Host 2There are two camps that disagree, and the paper quotes both. Measurements in the nearby universe, using Cepheid stars and supernovae, give seventy three point zero, plus or minus one point zero, and a newer network of many distance methods gives seventy three point five, plus or minus zero point eight. Measurements from the cosmic microwave background, the afterglow of the early universe, read through the standard cosmological model, give sixty seven point one nine, plus or minus zero point three eight.

Host 1That gap between sixty seven and seventy three sounds small, so is it actually a big deal?

Host 2It is about six units with error bars of around one, so it is far bigger than the uncertainties. The paper describes it as having reached the five sigma level, which is the conventional threshold where physicists stop calling something a fluke. It has survived a decade of better data, which is why people want completely independent ways to measure the same number.

Falling into Virgo

Host 1And how does a cluster of galaxies give you an independent way to measure it?

Host 2A big cluster has so much mass that it pulls nearby galaxies towards it, fighting the cosmic expansion. Very far away, galaxies just ride the expansion, and their speed grows in step with distance; close in, they are slowed down or even falling in. The shape of that transition depends on two things, the cluster's mass and the expansion rate, so fitting it tells you both at once.

Host 1Is this a new idea, or have people been doing it for a while?

Host 2It goes back to Allan Sandage in nineteen eighty six, with Jim Peebles including the cosmological constant in the equations a couple of years earlier. It has since been applied to many galaxy groups and clusters. The paper notes that those results mostly land between sixty and eighty, with uncertainties around ten percent.

Host 1You mentioned galaxies falling in close to the cluster; is there a specific line where they stop moving away and start falling?

Host 2Yes, it is called the zero velocity radius. It is the distance from the centre at which a galaxy's fall towards the cluster exactly cancels the expansion, so it is standing still relative to the cluster. Inside it galaxies are falling in, outside it they are still receding, and for Virgo it is expected to be somewhere between five and seven megaparsecs.

The question

Host 1So what exactly is the open question that this team set out to answer?

Host 2Whether the formulas people use to fit that infall curve actually give the right answer. All of them assume an isolated, perfectly spherical lump collapsing in a smooth expanding universe. Real clusters are lumpy, still growing and surrounded by other structure, and the authors say nobody had tested these formulas against a realistic simulated universe before.

Host 1Which formulas are we talking about, in plain words?

Host 2There are three in common use. One by Peirani and de Freitas Pacheco, and one by Peñarrubia and colleagues, are quick approximate fitting curves that turn out to be the same equation with different tuning numbers. The third, by Baushev in twenty twenty, is the full solution of the collapse equations, slower to compute but without the shortcut.

Host 1Why would a quick approximation get the expansion rate wrong, rather than just be a bit noisy?

Host 2Because of what happens far from the cluster, where the curve should settle down to pure expansion. The approximate formulas head towards a line that is thirty eight percent or forty nine percent steeper than plain expansion, depending on the version. The fit anchors the expansion rate to that far-off slope, so an overly steep slope drags the inferred value down.

Host 1So if I have this right, the shortcut formula has a built-in exaggeration far out, and to compensate the fit chooses a smaller expansion rate?

Host 2That is right. The full Baushev solution approaches plain expansion by construction, so it does not have that particular problem. The question then becomes whether it has other problems.

The simulated universe

Host 1How do you actually test a formula when nobody knows the true answer for a real cluster?

Host 2You use a simulated universe where the answer is known. They used IllustrisTNG three hundred, a large cosmological simulation of a cube about three hundred megaparsecs on a side, with dark matter, gas and galaxies, and haloes ranging from small groups to massive clusters. Its expansion rate is set in advance, so any fit can be compared with the truth.

Host 1What does halo mean here, and are they using every one in the simulation?

Host 2A halo is the clump of mostly dark matter that a group or cluster of galaxies sits in. They kept only isolated ones, with no neighbour of at least a tenth of the mass inside the zero velocity radius. Each had to have at least ten smaller galaxies between zero point eight and two point five times that radius, each galaxy above a billion solar masses.

Host 1And in the simulation, do they know exactly how each little galaxy is moving?

Host 2Yes, the full three dimensional positions and velocities, which no telescope gives you. They still added realistic errors, five percent on distances and three kilometres per second on velocities. They also fitted an extra scatter term, so the leftover jitter of galaxies around the smooth curve doubles as a quality score for each fit.

What the simulation showed

Host 1So with all that perfect information, how wrong were the formulas?

Host 2Both were biased low. Averaged over the haloes, the quick Peñarrubia formula came out about fifteen and a half units too low, and the full Baushev solution about twelve and a half too low. For massive haloes with very little scatter around the curve, the Baushev bias shrank towards zero, but the quick formula stayed clearly off.

Host 1Twelve units too low sounds enormous next to a tension of six, so is that the right comparison?

Host 2It is the right worry. A bias that size is twice the whole disagreement the method is meant to help settle. The paper also finds that the spread from halo to halo is far larger than the error bar each single fit reports, which tells you real haloes vary in ways the spherical model does not capture.

Host 1Does the size of the structure matter, say a small group versus a big cluster?

Host 2Yes. Above about ten thousand billion solar masses the bias levels off, and clusters give tighter answers than groups, because their deeper gravity drowns out neighbours and they have more galaxies to fit. The authors say results from smaller groups should be treated with caution.

Turning to Virgo

Host 1I'm guessing the real sky is harder than a simulation, so what changes when you point this at Virgo?

Host 2Two big things. For a real galaxy you only measure the speed along our line of sight, not its full motion, so you need a model to convert that into a fall towards Virgo's centre. And the sample is incomplete, since we mostly see galaxies on our near side of the cluster.

Host 1How do you turn a speed towards us into a speed towards Virgo, when they point in different directions?

Host 2There are two standard conversions, called the minor and major infall models. Both work best for galaxies roughly in line with us and Virgo, and both struggle for galaxies off to the side; the major one actually blows up at a right angle. That is why earlier studies threw out galaxies at awkward angles.

Host 1Which galaxies did they use, and how do they know how far away each one is?

Host 2They use a method called the tip of the red giant branch, which works because the brightest red giant stars in a galaxy all peak at nearly the same brightness, so their apparent faintness gives the distance. They assembled two samples, one hundred and seven galaxies from Anand and colleagues, and seventy one from Kashibadze and colleagues. Combined, that is one hundred and twenty one galaxies, against thirty three in the most recent earlier analysis by Kim and colleagues.

Host 1Were the cluster's own distance and speed taken as known, or measured as well?

Host 2They took sixteen point five megaparsecs for the distance and nine hundred and eighty eight kilometres per second for its speed relative to our Local Group, with uncertainties. In the main fit, five parameters are fitted together: the expansion rate, the mass, the scatter, and the cluster's distance and speed, with those last two held near their earlier values. A simpler three parameter version keeps distance and speed fixed.

Mock Virgos

Host 1Then how do they work out the bias for Virgo specifically, rather than for clusters in general?

Host 2They built mock observations. They took simulated haloes heavier than a hundred thousand billion solar masses, a lower bound for Virgo from X-ray data, and placed pretend observers about sixteen and a half megaparsecs away, with the same kind of measurement limits. Then they ran the whole pipeline, including the line of sight conversion, on those fakes.

Host 1Did throwing out the galaxies at awkward angles help, like the earlier studies assumed?

Host 2No, and that is one of the clearer findings. In the mocks, those angle cuts made the bias worse, and keeping everything gave the smallest bias for the minor model. The major model needed a moderate cut just to keep enough fits working, and was worse overall.

Host 1What were the actual bias numbers for the mock Virgos?

Host 2For the minor model with no cuts, the answer came out seven point two units too low, from one hundred and seventy two good fits. For the major model with its cut, it was thirteen point nine too low, from eighty seven fits. The spread between haloes was about seven to eight units in both cases.

A word from our sponsor

Host 2This episode is brought to you by Slowdown, the welcome mat that brings visiting relatives to a gentle stop at exactly the right distance from your door. Slowdown is announcing a voluntary recall.

Host 1I own two. My uncle now halts at the garden gate and orbits it until supper.

Host 2Slowdown mats were calibrated on simulated relatives. Real relatives arrived seven point two minutes early, so every mat has been corrected by adding seven point two minutes.

Host 1Mine has also pulled in the neighbours, the postman and, since Tuesday, a small dwarf galaxy, which it will not give back.

Host 2Replacement mats cannot be returned, as they are now receding from us at seventy four kilometres per second per megaparsec. Slowdown. Everyone arrives eventually.

Results

Host 1So what number did they actually get for Virgo from the real galaxies?

Host 2With the five parameter fit on all one hundred and twenty one galaxies, the minor model gives seventy four point four, plus or minus about three, and the major model gives seventy three point four, plus or minus about four. Those are statistical errors only, before any correction. One galaxy, I C three three four one, was dropped as a four sigma outlier that destabilised the fits.

Host 1And then they add back the bias they found in the mock clusters?

Host 2Exactly, they assume the bias simply adds on. The minor model goes from seventy four point four to eighty one point six, plus or minus seven point eight, and the major from seventy three point four to eighty seven point three, plus or minus eight point eight, now including systematic errors. The authors favour the minor model, because it had the smaller bias in the mocks.

Host 1Eighty one is higher than either side of the tension, so does this mean the universe is expanding even faster than the high camp says?

Host 2No, and the paper does not claim that. With an error of nearly eight, eighty one point six is about one sigma from seventy three, and much further from sixty seven, but this is one cluster and a first attempt at correction. The fairest reading is that the method, once corrected, does not point towards the low value, and is not yet precise enough to referee.

Host 1Did the two galaxy samples agree with each other, given they were measured by different teams?

Host 2Yes, within their error bars for each model, which is why the authors felt entitled to combine them. Interestingly, combining them barely shrank the uncertainty. The authors read that as a sign that simply adding more galaxies around Virgo will not tighten the answer much.

Host 1If more galaxies don't help, where is the uncertainty actually coming from?

Host 2They broke the error down by switching off each source in turn. In the main fit, roughly half comes from the leftover scatter of galaxies around the curve, and roughly half from Virgo's own speed. The measurement errors on individual galaxy distances and velocities account for less than one percent.

Host 1Why would the cluster's own speed matter so much for the expansion rate?

Host 2Because the two trade off against each other in the fit. Shifting Virgo's speed by one standard error, about sixty kilometres per second, moves the result by about three units. When the speed was held fixed in the three parameter fit, the minor model dropped to sixty seven point one, a shift of about seven, while the major model barely moved.

Host 1So the minor model's answer swings between sixty seven and seventy four depending on one assumption about Virgo's speed?

Host 2Yes, that is a fair summary. In the five parameter fit, the minor model pulled Virgo's speed about two standard errors away from its prior value, which the authors flag as a possible bias in that model that their mocks did not reveal. That is why they conclude neither model can be favoured outright for the real data.

Why it matters

Host 1Stepping back, why does this matter for anyone outside this small corner of astronomy?

Host 2The principle is simple: before you trust a measuring formula, run it on a fake universe where you know the answer. Here that test shows the widely used shortcut formulas tend to land low, and that the tidy habit of cutting awkward galaxies makes it worse. That helps explain why earlier Virgo results, like Kim and colleagues at sixty five point eight, sat on the low side.

Host 1Does this change anything for, say, the people working on the Hubble tension?

Host 2It gives them a reason to re-examine cluster infall results that seemed to lean towards the low camp. It also gives a recipe for others: the same reasoning applies to groups like the one around M eighty one and Centaurus A. It does not resolve the tension, and it does not add a new precise measurement to the pile.

Caveats

Host 1What are the honest weak points the authors themselves admit?

Host 2The biggest is whether the mock clusters really represent Virgo, a single structure that the authors call most likely not typical. Their halo selection could be too broad or too narrow, and the bias changes with how strictly the scatter quality cut is set. They call the correction a first attempt.

Host 1Is there anything baked into the method that could tilt the answer in a way we can't see?

Host 2Yes, a few things. The correction assumes the bias just adds on, which they have not tested with simulations run at different expansion rates. And the simulation itself is built on the standard cosmological model with its own fixed expansion rate, so the correction inherits that model's assumptions.

Host 1And the real data, is there anything uneven about which galaxies we can see?

Host 2We see mostly the near side of Virgo, and every distance comes from a single method. Virgo itself has at least three major parts, so treating it as one sphere is an approximation. The paper also notes that comparing directly with earlier Virgo results is not straightforward, because each study made different choices.

What comes next

Host 1So what should we watch for next, and what would actually make this measurement sharper?

Host 2A better handle on Virgo's own speed is the clearest lever, because it drives half the error. The authors also want simulations tailored to clusters like the one being measured, and a collapse model more general than a perfect sphere. Spectroscopic surveys and telescopes such as the James Webb Space Telescope, DESI and Four MOST will supply far more velocities and precise distances, with Fornax and Virgo already being studied and Coma planned.

Sign-off

Host 1So the takeaway is that cluster infall is a promising independent check, but not one you can lean on yet?

Host 2That is fair. Fitted with the full collapse model and corrected using simulated clusters, Virgo gives about eighty one point six, plus or minus seven point eight, with the favoured model. The bigger lesson is that the shortcut formulas and the angle cuts used before tend to pull the answer low.

Host 1That is the deep dive: Debiasing Local Cluster Infall Reveals H nought, by Carreres, Benisty, Wagner and colleagues, a preprint on arXiv. Every weekend this same feed carries the roundup, the week's new papers in one short episode. This has been the Hastrup Review of Science.