Episode 725 September 20269 min

Red Giants, Wave Dark Matter and a Palladium Anode

Six new papers, honestly explained: Hubble constant from red giants alone · Cosmic Daybreak at redshift fifteen · Quantum advantage on a cloud machine · Iridium-free anode catalysts · A health atlas of ageing · Interval training in women with obesity.

MP3

Chapters

The papers

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

Sponsor

Host 1Support for the Hastrup Review of Science comes from RefurbMiner.

Host 1One intelligent platform for your whole business. Invoicing, banking, and compliance in a single system, with every decision fully traceable. Privacy ready, and hosted in Europe.

Host 2Start your free fourteen day trial today. The link is in the show notes.

Cold open

Host 1Welcome to the Hastrup Review of Science — where artificial intelligence reads real science, so you don't have to pretend you understood the paper.

Host 1This week's best story: a distance ladder built without supernovae lands on both sides of the Hubble tension, and the authors explain exactly why.

Host 2It is a rare paper that tells you, with numbers, why it cannot settle the question it set out to answer.

Hubble constant from red giants alone

Host 1So what did they build?

Host 2Stiskalek, Desmond and Lavaux, in an arXiv preprint, measured the Hubble constant using only the tip of the red giant branch, a sharp brightness ceiling for old stars. They calibrate it on two geometric anchors, the Large Magellanic Cloud and a nearby galaxy whose distance is fixed geometrically by water masers orbiting its central black hole, and apply it to three hundred and ninety-four host galaxies. No supernova rung at all.

Host 1And the answer?

Host 2It depends on how you weight eighty reconstructions of the local velocity field. Weighted equally, sixty-eight point four, plus or minus five point nine kilometres per second per megaparsec; weighted by Bayesian evidence, seventy-two point three, plus or minus two point six.

Host 1Why the split?

Host 2Most hosts sit below a thousand kilometres per second of recession, where galaxies' own motions are comparable to cosmic expansion. The scatter between velocity reconstructions alone is five point two, and the higher value is driven by a single high-evidence realisation. The bottleneck is mapping local motions, not calibrating the stars.

Host 1What would fix it?

Host 2Red giant distances reaching beyond a thousand kilometres per second, plus sharper maps of the local universe. That would push the scatter well below five, and make this an independent check on the supernova ladders.

Host 1The full story is in this paper's deep dive, out this week in the feed.

Cosmic Daybreak at redshift fifteen

Host 1Staying with cosmology, but going much further back.

Host 2Zhang, Broadhurst and colleagues compiled all twenty-two galaxies above redshift ten with measured stellar ages from the James Webb Space Telescope. Subtracting each age from the cosmic age at its redshift gives a formation redshift, and those pile up against a hard edge at fifteen point one. No earlier star formation appears, and the absence beyond seventeen point three is significant at three sigma.

Host 1What does that edge mean?

Host 2Standard cold dark matter predicts a gradual dawn extending to earlier times. Wave dark matter, an ultralight boson, resists collapse until halos reach about four billion solar masses, and matching the edge gives a boson mass near two point two times ten to the minus twenty-two electronvolts.

Host 1How firm is it?

Host 2Twenty-two galaxies, model-dependent ages from different surveys, and a simple model with fixed star formation assumptions. It is a speculative preprint, but it makes a clean prediction: deeper spectroscopy should keep finding nothing beyond redshift fifteen. The full story is in this paper's deep dive, out this week in the feed.

Quantum advantage on a cloud machine

Host 1Next, quantum computing.

Host 2A team from the startup BlueQubit ran random-circuit sampling on sixty-one qubits of IBM's Nighthawk processor, through the standard cloud stack with no special calibration. At thirty-six cycles they collected one million samples in nineteen seconds.

Host 1And classically?

Host 2Their tensor-network estimate is about one hundred and ten years on the Frontier supercomputer. Two independent fidelity estimators, mirror benchmarking and patched cross-entropy benchmarking, agree at every measured depth, and put fidelity at about two in a thousand. The newer chip raised per-cycle fidelity from zero point eight three six to zero point eight seven two, which compounds to more than tenfold at these depths.

Host 1Caveats?

Host 2The full circuit cannot be verified directly, the classical estimate uses only the contraction methods they tried, and better classical algorithms have erased such claims before. Random-circuit sampling also solves no useful problem; what is new is that anyone with a cloud account can reproduce it. The full story is in this paper's deep dive, out this week in the feed.

Iridium-free anode catalysts

Host 1Now, green hydrogen.

Host 2Proton-exchange-membrane electrolysers need iridium or ruthenium at the anode, both scarce. A group at Lila Sciences used a closed-loop platform, more than ninety percent automated, with machine-learning models picking each next experiment, to screen two thousand nine hundred and forty-two oxide catalysts across twenty-six elements.

Host 1What came out?

Host 2A palladium oxide with dilute indium and manganese stayed below half a volt of overpotential for one thousand hours in one-molar sulfuric acid. Plain palladium oxide crossed that line at around two hundred hours, and the ruthenium oxide benchmark film delaminated at about six hundred. In a retrospective test, the learning agent advanced the activity and stability frontier faster than fixed Bayesian optimisation or a language model choosing experiments alone.

Host 1What should we hold back on?

Host 2These are thin films at ten milliamps per square centimetre, far below industrial current densities, and palladium is itself a platinum-group metal. The authors are company employees, patents may be filed, and it is a preprint. The full story is in this paper's deep dive, out this week in the feed.

Sponsor break

Host 2Do you have problems with things flapping around in the wind? Then you are in luck.

Host 2This week's sponsor is clothes pegs. Strong, reliable, and designed to hold firmly exactly where you need it, for as long as you need it.

Host 1I use them every morning. There is nothing like a firm, dependable grip before breakfast.

Host 2They come in wood or plastic, in packs large enough for the whole household, and they are very reasonably priced. They are in stock right now.

Host 1And when the job is done, one gentle squeeze and they help you get off the line.

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

A health atlas of ageing

Host 1Back to science, and to ageing.

Host 2Zhang and colleagues, in a Nature Portfolio ageing journal, built a health atlas from routine check-ups in Zhejiang: one hundred and eighty-nine thousand people, ninety-four million observations of ninety-nine biomarkers, grouped into ten organ-system domains. Different systems drift from normal at different, sex-specific ages.

Host 1Does the drift predict anything?

Host 2A higher average burden across domains was linked to mortality with a hazard ratio of one point zero five eight per standard deviation, a modest effect with a tight confidence interval. In exploratory analyses, instability over repeated visits mattered, while a single snapshot's spread did not after adjustment. A model trained on the raw measurements predicted seven future clinical abnormalities with an area under the curve of zero point eight zero two.

Host 1Limits?

Host 2It is observational, from one Chinese province, and the instability finding is exploratory. We could not obtain the full text, so this trailer rests on the abstract. The full story is in this paper's deep dive, out this week in the feed.

Interval training in women with obesity

Host 1Finally, training.

Host 2A preregistered meta-analysis pooled fifteen randomised trials of high-intensity interval training in adult women with overweight or obesity, against controls doing no structured exercise. Effects are in Hedges' g, a standardised effect size.

Host 1The numbers?

Host 2Body weight fell by a small g of minus zero point three three, but fat mass fell by minus zero point six six and body-fat percentage by minus zero point seven six. Peak oxygen uptake improved by a large g of one point zero nine, while total cholesterol and triglycerides did not change significantly.

Host 1So the scale undersells it.

Host 2Yes, but the training groups had eight to thirty women each, heterogeneity was high for several outcomes, and certainty was rated low to very low. The comparison is against doing nothing, not against steady endurance work. For anyone tracking fat loss, the lesson holds: measure body composition, not just the scale.