Deep divefrom Episode 10Cosmology6 November 202623 min

Two Masses for One Neutrino, Both Too Light

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: Disentangling geometry & growth in cosmic neutrino mass constraints: a more physical split and diagnostic (Ge 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, why the universe seems to weigh its neutrinos lighter than physics allows, and a new way to ask which part of the evidence is doing the pulling.

Host 2The paper is Disentangling geometry and growth in cosmic neutrino mass constraints: a more physical split and diagnostic, by Fei Ge, Frank Qu, Hanyue Wang, Emmanuel Schaan, Kimmy Wu and Alexander Friedland, from Caltech, Stanford and the SLAC National Accelerator Laboratory. It is a preprint on arXiv from October, so it has not been through peer review yet. Their idea is to split the neutrino mass into two copies, one for how the universe stretches distances and one for how it grows lumps, and see whether both tell the same story.

Why weigh a ghost particle

Host 1Before we get to copies of a mass, can you tell me what a neutrino actually is, and why anyone weighs it with the whole universe?

Host 2A neutrino is a tiny, electrically neutral particle that barely interacts with anything; trillions pass through you every second. In the wider picture, physicists long assumed they had no mass at all, until experiments around the turn of the century saw them change type in flight, called oscillation, which is only possible if they have mass. Those experiments measure the differences between the masses, not the masses themselves.

Host 1So oscillation experiments tell us neutrinos are not weightless, but not how heavy they are?

Host 2Exactly, but the differences do set a floor. If the three masses are ordered the usual way, the sum must be at least fifty-eight thousandths of an electronvolt, the number the paper uses; in the other possible ordering, called inverted, the floor is about a tenth of an electronvolt. An electronvolt is a unit of energy and mass used for particles, and an electron weighs about half a million of them, so we are talking about something extraordinarily light.

Host 1And where does the universe come into it, if the particle is that light?

Host 2Because there are so many of them, left over from the Big Bang, that together they add up to a small but measurable share of the cosmos. Their combined mass changes how fast the universe expands and how quickly matter clumps into galaxies. Cosmology measures the total of the three masses, the sum, and the paper writes it as sum m nu.

What was known before

Host 1What are the telescopes actually looking at when they do this?

Host 2Two main kinds of data. The cosmic microwave background is the oldest light we can see, released when the universe was about three hundred and eighty thousand years old, which is the wider picture; its pattern of hot and cold spots, and the way that light gets bent by matter on its way to us, called CMB lensing. The other is baryon acoustic oscillations, or BAO, a fixed ruler left by sound waves in the early universe, which galaxy surveys use to measure distances.

Host 1Which specific experiments feed into this paper?

Host 2For the microwave background they use the Planck satellite's final maps, plus a lensing measurement that combines Planck with the Atacama Cosmology Telescope and the South Pole Telescope, with a joint signal-to-noise of sixty-one, the most precise lensing measurement to date. For distances they use the second data release of the Dark Energy Spectroscopic Instrument, DESI, covering redshifts from about zero point three to two point three. Redshift is a measure of how far back in time you are looking.

Host 1And what did all of that say about the neutrino mass before this paper?

Host 2That it is suspiciously small. The latest joint analysis the paper cites puts the sum below fifty-five thousandths of an electronvolt at ninety-five percent confidence, which is below the floor of fifty-eight from oscillations. Their own combination gives an upper limit of about sixty thousandths, which rules out the inverted ordering at more than three sigma and sits right on the edge of the normal one.

Host 1When you say three sigma, how sure is that in everyday terms?

Host 2Sigma counts how many standard errors a result is from a value. Roughly, one sigma happens by chance about a third of the time, two sigma about one time in twenty, and three sigma a bit more than one time in four hundred. Physicists usually treat anything under three as a hint, not a discovery.

The puzzle

Host 1So cosmology seems to be saying the neutrinos weigh less than the particle physics says they possibly can?

Host 2Yes, and that is the tension. When the data are allowed to go past zero, which the field does as a mathematical extrapolation, they lean towards what people call negative neutrino mass. Nobody thinks neutrinos literally weigh less than nothing; it means the data prefer the opposite of the effect that massive neutrinos would have.

Host 1What have people suggested might be causing it?

Host 2The paper lists several. Some groups argue it is a mismatch between the total amount of matter that DESI prefers and the dark and ordinary matter the microwave background prefers; others blame the lensing signal being slightly too strong. Others again point to the reionization optical depth, or to dark energy that changes over time, or to a slightly curved universe.

Host 1That optical depth sounds like jargon; what is it in plain words?

Host 2It measures how much the microwave light was scattered by gas when the first stars switched on and ionised the universe. It is the least well pinned-down of the standard parameters, resting almost entirely on Planck's largest-scale polarisation data, which give about zero point zero five four. The paper notes that a value around zero point zero nine would remove the neutrino tension.

Host 1So with all those candidates on the table, what exactly were these authors trying to find out?

Host 2Which of two physical channels drives the low mass: geometry or growth. Geometry is how neutrino mass changes the expansion rate and therefore the distances we infer; growth is how it changes the way lumps of matter build up. They also want to know whether the two channels agree with each other, because the standard model of cosmology says they must.

Host 1Why not just ask which telescope is to blame, and leave the physics out of it?

Host 2They tried that first. They swapped real data for fake, perfect data from a chosen model, one dataset at a time, to see whose removal fixes the problem. When the fake universe was the one Planck prefers, swapping out DESI mostly fixed it, and swapping out lensing as well brought the mass back to fifty-eight thousandths.

Host 1So that would point the finger at DESI, wouldn't it?

Host 2Only if you assume Planck is the truth. When they built the fake universe from the best fit to Planck and DESI together, no single swap made much difference; only replacing two or all three moved the answer back. Their conclusion is that the problem is a mild disagreement between Planck and DESI, just under three sigma, rather than one bad dataset.

Host 1And why does that push them towards splitting by physics instead of by dataset?

Host 2Because the datasets do not line up with the physics. DESI only measures geometry, but the microwave background carries both geometry and growth in the same pattern. Splitting by telescope cannot tell you which physical effect is behind the pull.

Host 1There was one more subtlety in the paper about why the limit looks so tight; what was that?

Host 2They point out that a tight upper limit can come from two things: a genuinely better measurement, or a distribution that has slid towards negative values and been chopped off at zero. If the mass is only allowed to be positive, the visible part of a slid distribution looks narrow even when the measurement has not improved. Untangling those two is part of why they test with fake data and allow negative values.

Two copies of one mass

Host 1So how do you give one particle two different masses inside a computer?

Host 2By running the standard calculation twice and stitching the results together. They modified a widely used program called CAMB, which predicts what the microwave sky and the BAO distances should look like for a given universe. One run uses the geometry mass to set the expansion rate for every distance; a second run uses the growth mass to set the expansion rate while the lumps evolve.

Host 1What exactly falls on the geometry side and what on the growth side?

Host 2The geometry mass controls the BAO distances, the angular distance to the microwave background, and the geometric factors inside the polarisation and lensing calculations. The growth mass controls the evolution of every kind of fluctuation, including the dark matter, the gas, the light and the gravitational potentials themselves. In the final step, the geometry from the first run is combined with the fluctuations from the second to produce the predicted sky.

Host 1If the two runs describe universes with different expansion rates, don't their clocks drift out of step?

Host 2They do, and that was the fiddly part. The two universes can have different ages, so matching them by time would leave gaps at the end of the calculation. Instead they line up the two runs by redshift, so each step in one corresponds to the same moment of stretching in the other, and they use the same grid of wavelengths for both.

Host 1How do they know the stitched version isn't just broken?

Host 2They checked that when the two masses are set equal, the stitched code gives the same microwave predictions as the unmodified program, within numerical precision. That is the minimum any such split must do: reduce to the ordinary model with one neutrino mass. They also say plainly that their matching choice is not the only possible one.

Host 1And how do they handle negative masses, if a negative mass isn't a real thing?

Host 2They extend each prediction past zero in a straight line, mirroring the effect around zero mass, a method used in earlier work. So a negative value means the data want the observables shifted the opposite way from what adding neutrino mass would do. Then they explore the possibilities with a standard sampling method called Markov chain Monte Carlo, and they have published their modified code.

Host 1What does changing each mass actually do to the sky, so I can picture it?

Host 2Raising the geometry mass from fifty-eight thousandths to a tenth of an electronvolt mostly slides the peaks of the microwave pattern sideways, because the distance to that light changes. Raising the growth mass by the same amount barely touches the raw pattern, under zero point two percent, but lowers the lensing signal by about two percent at the finest scales. The lensed microwave pattern then drops about one and a half percent at small scales.

Host 1Two percent sounds tiny; is that something a telescope can really see?

Host 2For lensing, yes, which is why the growth channel leans on it. The lensing measurement they use has a signal-to-noise of sixty-one, so a shift of a couple of percent at the scales it covers is within reach. The raw pattern change of a fifth of a percent is much harder.

How sharp each channel is

Host 1Before seeing the real answer, did they check how sharply each channel can measure in principle?

Host 2Yes, with a forecast at today's precision. From the microwave background alone the geometry mass is essentially unmeasured, with an uncertainty of about one electronvolt, while the growth mass gets about a quarter of an electronvolt. The reason is that the extra lensing smearing already built into the observed spectra measures growth; with unlensed spectra both channels are equally poor.

Host 1Why is the microwave background so blind to the geometry mass on its own?

Host 2Because the distance to that light can be kept the same by trading neutrino mass against the expansion rate today and a little dark matter. The microwave background pins the angle very precisely but cannot tell those apart. The paper says an uncertainty around one electronvolt should be read as unconstrained, not as a measurement.

Host 1And what happens to those numbers when you add the lensing and the BAO?

Host 2Adding lensing improves the growth channel by forty-five percent, to about a seventh of an electronvolt, and barely moves geometry. Adding BAO improves the geometry channel fourteen-fold, to about seventy-six thousandths, because BAO pins the total matter fraction independently. With everything combined, geometry is about one point seven times sharper than growth: seventy-four thousandths against one hundred and twenty-five thousandths.

Host 1Were those forecasts trustworthy, or just optimistic arithmetic?

Host 2Somewhat optimistic. Fake-data runs with the full sampling gave uncertainties of about ninety-five thousandths for geometry and one hundred and ten thousandths for growth, so the quick forecast was twenty-eight and twelve percent too rosy. That is why the authors quote the full runs for the real results.

A word from our sponsor

Host 1This episode is brought to you by Two Tape, the measuring tape with one side for distance and a separate side for growth.

Host 2I measured my hallway with both sides. They agreed to within zero point four sigma.

Host 1Terms and conditions. Readings below zero are an extrapolation, valid at linear order only. Two Tape is not to be used on anything that is far away and growing at the same time.

Host 2Mine reads minus seven centimetres for the garden shed. The shed is now slightly inside the house.

Host 1Never lock the two sides together, or everything will measure shorter than the smallest thing allowed. Two Tape. Some lengths may already be behind you.

What the real data say

Host 1Back to the real data, then: when you plug in the actual measurements, what do the two channels say?

Host 2Both lean low. With everything combined and negative values allowed, the geometry mass comes out at minus seven hundredths of an electronvolt, plus or minus ten hundredths, and the growth mass at minus thirteen hundredths, give or take about eleven. That puts geometry about one point three sigma and growth about one point eight sigma below the oscillation floor.

Host 1So neither channel, taken alone, is in serious trouble with particle physics?

Host 2Right, neither one excludes the floor on its own. If you insist both masses are positive, geometry stays under about eighty-one thousandths and growth under about one hundred and thirteen thousandths, at ninety-five percent confidence; both limits sit above fifty-eight. The very tight single-parameter limit of about sixty thousandths only appears when you force the two to be the same number.

Host 1Why would forcing them equal make the limit so much tighter than either one alone?

Host 2Because each channel trades off against the expansion rate today in a different direction. Geometry slides steeply, growth much more gently, and when you demand one shared mass you keep only the small region where both agree. So the strong standard bound comes from combining two channels with different trade-offs, and it depends on the assumption that one mass governs both.

Host 1And do the two channels agree with each other, which was the test the standard model has to pass?

Host 2They do. The difference, geometry minus growth, is seven hundredths plus or minus seventeen, which is only zero point four sigma from zero, so the standard model passes this internal check. But both are pulled low together: the point where both equal fifty-eight thousandths still lies outside the ninety-five percent region of their joint result.

Host 1So if I have this right, geometry and growth tell the same story, and that shared story is the one in tension with the oscillation floor?

Host 2That is how the authors put it, almost word for word. Current data are consistent with both channels agreeing, and that agreed answer is in mild tension with the minimum. Mild is the right word: the individual channels are at one to two sigma.

Host 1Did they find which channel carries more of the pull away from the expected value?

Host 2Comparing the fake-data runs with the real ones, the geometry mass moved down by about one point nine sigma and the growth mass by about zero point nine. So more of the shift sits in geometry, which fits their earlier finding that the matter-density disagreement between Planck and DESI is what the neutrino mass is absorbing. They caution that this comparison assumes Planck's preferred universe is the truth, and the two shifts are correlated, so the significances are only indicative.

Host 1How do these numbers compare with earlier attempts to split the mass?

Host 2They are broadly consistent. One earlier study split by observable and found minus fifteen hundredths for clustering and minus three hundredths for BAO, which matches their growth and geometry numbers. Another, using background information only, found more negative values; this paper's geometry number is less negative, which the authors expect because the deficit is now shared with the growth channel.

The next telescopes as a diagnostic

Host 1Looking ahead, what do they expect the next generation of telescopes to achieve?

Host 2They forecast for the South Pole Telescope's upgraded survey, the Advanced Simons Observatory, and a future survey like the proposed CMB Stage Four, each combined with DESI's third data release. With DESI added, the geometry uncertainty drops to fifty, forty-three and thirty-eight thousandths, and growth to ninety-seven, eighty-two and seventy-two thousandths. Without BAO, geometry stays roughly ten times weaker than growth.

Host 1So the two channels will be improved by different machines?

Host 2Yes. Once DESI's third release is in, geometry is limited mainly by BAO, and the three microwave surveys differ by only about a third; growth keeps improving with better microwave and lensing data. If the geometry value of minus seven hundredths persists, the Advanced Simons Observatory plus DESI would put the oscillation floor three sigma away using geometry alone, and CMB Stage Four three point four.

Host 1You mentioned a diagnostic in the title; how does this split help tell the possible explanations apart?

Host 2They ran forecasts where the analysis deliberately uses the wrong model. If the real universe has dark energy that evolves, using DESI's own best fit for that, but you analyse it assuming a constant one, the geometry mass is dragged down by three point one to four point two sigma, while growth moves only one point one to one point three. If instead the true optical depth is zero point zero nine but you assume about zero point zero five four, it reverses: growth is dragged down by two point seven to three point six sigma, geometry by at most a quarter of a sigma.

Host 1So it works like a fingerprint: a big miss in geometry points to dark energy, a big miss in growth points to the optical depth?

Host 2That is the idea, with one correction. You have to look at how far each channel sits from the oscillation floor, not only at the difference between them. In the dark energy case the difference stays within about zero point seven sigma of zero, so the agreement test alone would miss it; a single combined mass would miss both, because it responds to each.

What it means and what it doesn't

Host 1What does any of this mean for someone who isn't a cosmologist?

Host 2Directly, nothing changes in your life. Indirectly, it is a cross-check on the standard picture of the universe: if neutrino masses from the sky and from the laboratory end up genuinely incompatible, something in that picture, or in a measurement, is wrong. This paper gives a way to point at which part.

Host 1What can we not conclude from this paper, so nobody overreads it?

Host 2We cannot conclude that neutrinos have negative mass, or that the standard model of cosmology has failed; the check between channels passes. We also cannot conclude which explanation is right, since the diagnostic cases are forecasts for future data, not findings. And the tension itself remains mild, at one to two sigma per channel.

Caveats

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

Host 2The split is phenomenological, meaning a bookkeeping device rather than a physical theory; no real particle has two masses. The way the two runs are matched is a choice, and negative values are a straight-line extrapolation. The quick forecasts proved twenty-eight percent optimistic for geometry, and the future ones are forecasts around an assumed true mass of fifty-eight thousandths.

Host 1Are there choices in the data that could change the answer?

Host 2Some. They use only one Planck likelihood for the main microwave data, fix the number of neutrino species at the standard three point zero four four, and use data from the whole of cosmic history. They note that restricting to data free of the late universe could give a different result and leave that for future work.

Host 1Is there anything about the comparison between fake and real data that I should be careful with?

Host 2Yes: the fake data assume Planck's best-fit universe is the truth, so the finding that geometry carries more of the shift is conditional on that. And because fake and real runs share the same Planck data, the two shifts are correlated. The authors themselves call those significances indicative.

What to watch

Host 1So what should listeners keep an eye on over the next few years?

Host 2New DESI data releases and the Simons Observatory, which between them would sharpen the geometry channel to about four hundredths of an electronvolt. A better measurement of the optical depth, which would test the growth side directly. And whether other groups adopt this split, which is easy to try now that the code is public.

Host 1If the geometry number stays negative while growth comes back up, what would that tell us?

Host 2By their forecasts, that pattern would point towards the expansion history, such as dark energy that changes over time, rather than the optical depth. The reverse would point at the optical depth or the large-scale polarisation data. If both stay low together and agree, the puzzle sits in something both channels share.

Sign-off

Host 1So the short version is: two copies of the neutrino mass, both a little too light, still agreeing with each other, and a tool that should tell us later why?

Host 2That is a fair summary. Geometry gives minus seven hundredths and growth minus thirteen, each one to two sigma below the floor, and their difference passes the consistency test. The main contribution is a cleaner method and a diagnostic that future data can actually use.

Host 1That is the deep dive: Disentangling geometry and growth in cosmic neutrino mass constraints, by Fei Ge 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.