Physicists finally observe strange isotope Oxygen 28 – raising fundamental questions

Nick Petrić Howe

Welcome back to the Nature Podcast. This week: the first observation of a long-sought isotope of oxygen...

Dan Fox

...and the AI that can beat humans at drone racing. I'm Dan Fox.

Nick Petrić Howe

And I'm Nick Petrić Howe.

<Music>

Nick Petrić Howe

After decades of experiments, physicists have observed an unusual isotope of oxygen, known as oxygen 28, for the very first time. This is a momentous discovery for nuclear physics and may challenge our theoretical models of how atomic nuclei work. But before we get to oxygen 28, and why it's got physicists so excited, we need to start with some background about the structure of atomic nuclei and explain a few bits of terminology. Atomic nuclei are made of protons or neutrons, with the number of protons determining the identity of the element. And both these subatomic particles are arranged in the atom in something akin to layers.

Ritu Kanungo

They are arranged like in an onion-like structure, so in other words in some discrete shells.

Nick Petrić Howe

This is Ritu Kanungo, a nuclear physicist who’s been writing a News and Views article about the new discovery. These layers, or shells as they are also known, are made up of varying numbers of protons and neutrons, which convey different properties to the nucleus.

Ritu Kanungo

And when one shell gets totally filled, that's what's called a magic number.

Nick Petrić Howe

Magic. That's the real term that physicists use. It refers to the specific number of protons or neutrons needed to complete one of these shells or layers, such as 2, 8, 20, or 28. In this magic state nuclei are what's known as bound. The neutrons and protons that make them up stick together better and the less liable to break apart. You can think of them as being more stable. This stability means that magical nuclei are more common in Nature. And some nuclei have this property for both their protons and neutrons, which makes them even more stable. This is what's called doubly magic.

Ritu Kanungo

Doubly magic means both neutrons close shelled, and protons close shell. Now if I have let's say, two neutrons and two protons, that will make it doubly magic. So a nucleus that has two neutrons and two protons is helium. So helium is the second most abundant element in the universe.

Nick Petrić Howe

Another example of a doubly magic nucleus is the oxygen all around us, oxygen 16. This is made up of 8 neutrons and 8 protons — both magic, so very stable. But Oxygen 16 is quite different from oxygen 28 a strange isotope that researchers have theorised would also be doubly magic. Oxygen 28 would still have 8 protons, a magic number, which is why it is still oxygen rather than another element. But it would have a magic 20 neutrons. Observing such an isotope with a huge imbalance between the number of protons and neutrons would allow scientists to put their theories about how nuclei, and how magic numbers work, to the test. So for decades researchers have been trying to find it, but without much success. Until now. This week, a paper in Nature shows the first observation of oxygen 28.

Takashi Nakamura

And the result was actually oxygen 28 was not doubly magic.

Nick Petrić Howe

This is Takashi Nakamura, one of the team behind this discovery. And if that sounds like a bit of a letdown, it isn't, but we'll come to that in a second. Now Takashi and the team have shown that rather than being stable, as would be predicted by a doubly magic Nature. Oxygen 28 only existed for a fleeting moment.

Takashi Nakamura

The lifetime is a well, we don't we cannot measure the lifetime, but we expect that lifetime is like 10 to the minus 21st seconds. The power of minus 21st. So that is I don't know. Very, very short right. Yeah, it just, yeah, it just made and just decays.

Nick Petrić Howe

This incredibly short existence was actually kind of expected. Oxygen 28 was known to be unbound, meaning that its neutrons would quickly fly off. So instead of looking for traces of a doubly magic nucleus Takashi and the team instead looked for the decay products of a short lived oxygen 28. They theorized that if oxygen 28 was not doubly magic, it would decay into four neutrons and one oxygen 24. So the team set about making an experiment that could detect these decay products, which was no mean feat, as neutrons are really hard to detect.

Takashi Nakamura

This detection of four neutrons was very, very tough. Nobody could do it. These neutrons are very like a ghost. Right? Neutrons are like neutral particles. So it can sometimes react with some material, then we know that neutrons are there, but sometimes it just scattered without leaving any signal. Right? So then it's like a ghost. So most of the experiments cannot measure three neutrons or four neutrons. So we needed to really set up a very good, sophisticated neutron detector system. So this is the first four neutron measurement ever done.

Nick Petrić Howe

To achieve this feat required an international collaboration and state of the art detection equipment to try and catch those ghostly neutrons. With that in place Takashi and the team started trying to make oxygen 28, however fleetingly. They started with a neutron rich isotope of calcium, known as calcium 48, which is relatively stable. Such an isotope can be broken down into different nuclei, including oxygen 28. And to trigger that process, they started by firing it at a beryllium target.

Takashi Nakamura

And then this hits the beryllium target, first, and then from calcium 48. Several neutrons protons are removed to produce fluorine 29. And fluorine 29 is short lived, but it's much more stable than oxygen 28 to survive more than milliseconds or something, so that it can be transported from the production point to the experimental point about, I think, 30 or 40 meters away. And during this 40 meters process, fluorine 29 was kind of purified. And then fluorine 29 hits a proton, to knock out this proton. So fluorine 29 has nine protons, okay. So nine protons minus one goes to eight protons, which is oxygen, right? So now we have oxygen 28.

Nick Petrić Howe

This experiment was actually conducted several years ago. But since then, a lot of maths and simulations have been required to really make sure that oxygen 28 was detected, or at least oxygen 24 and four neutrons. And that is what they present in this paper. The fact that oxygen 28 was not doubly magic was expected, but still raises questions, as it shows how our understanding of what makes something doubly magic is limited. The data the team gathered here, as well, will allow researchers to quantitatively probe how oxygen 28 challenges their expectations, something they could only theorize about before. And more broadly, this result implies there's a lot scientists still don't know about nuclear physics. In particular, the strong force that holds nuclei together, it may in fact challenge physicists concepts of how nuclei work. And so we'll help them understand some of the fundamental forces that helped make up all matter in the universe. He's Ritu, who you heard from earlier, to explain.

Ritu Kanungo

And the interesting thing is none of the theory predictions, which are really state of the art predictions, were able to explain or agrees really with the observed mass or the energy. So, now that brings a bigger question. If we know everything about Nature then there are no surprises we know it all. So, it should all fall in place. And therefore, this is pointing to the bigger question than whether our knowledge on Nature's strong interaction is complete, it clearly says it is not because if it is complete, then we would know everything and we do not know.

Nick Petrić Howe

That was Ritu Kanungo from St. Mary's University in Canada. You also heard from Takashi Nakamura from Tokyo Institute of Technology in Japan. For more on this story, check out the show notes for a link to Takashi's paper and to a News and Views article written by Ritu.

Dan Fox

Coming up, how an AI has been designed to take on the challenge of racing a drone in the real world. Right now though, it's time for the research highlights, with Noah Baker.

<Music>

Noah Baker

Venus flytraps can recognize when a wildfire is coming and snap shut to protect their flesh eating traps. The famously carnivorous plant lives in swamps where they're frequently exposed to fires. A team in Germany investigating this covered study plants with hay to mimic their natural habitat, and then set the area on fire. One of the plants survived and could still trap prey despite having burned. Intrigued the researchers brought lit matches close to the plants and found that they snapped shut as the flames approached. On examining the plants trigger hairs, which cause the traps to close around prey, the authors found a molecular heat sensor that activates the hairs if the temperature rises rapidly. The authors think that by triggering the traps to close, heat sensors protect the plants trigger hairs and so allow them to continue hunting prey throughout hot summers. You can read more on that in Current Biology.

<Music>

Noah Baker

An ordinary ballpoint pen loaded with conductive inks can write LEDs on to textiles, packaging and more. Perovskites are a class of semiconducting material that have excellent optical and electronic properties and hold promise for making efficient solar cells and LEDs. And unlike conventional semiconductors, such as silicon, perovskites can be handled in the form of solutions. And so a team in the USA formulated a series of viscous yet spreadable inks by dissolving various components in solvent mixtures. They loaded each ink into a commercial ballpoint pen and use the pens to write an LED. This consisted of a layer of the perovskite ink between an anode and a cathode made of a conductive polymer and nanometer wide silver threads. By switching between various perovskites the authors could make red, green and blue LEDs on various objects including rubber balloons and glass vials. According to the authors, this could be a handy low cost way to integrate simple displays and sensors into clothing and packaging. Read more in Nature Photonics.

<Music>

Dan Fox

Artificial intelligences have been pretty successful at beating us humans at our own games. For a long time, AIs have excelled at chess. But more recently, even in fast paced video games like Starcraft, requiring split second reactions and decision making, AIs are coming out on top. But these all have something in common. The virtual world where the AI trains and the place where it competes are the same. This has meant that humans have remained dominant in the real world in things like sports, where it's a little different from a simulation. But perhaps not any more. Now, in a new paper in Nature, there's an AI that can beat champion drone races.

Thomas Bitmatta

Drone racing is a very fast pace sport, we have a lot of acceleration, where you're basically instantly hitting a top speed or within half a second.

Dan Fox

This is Thomas Bitmatta, a champion in what’s known as first person view or FPV drone racing, where contenders wear goggles to see what the drones see as it whips around the 3D course, ducking and weaving through a series of gates at high speeds in races that can be over in seconds.

Thomas Bitmatta

And that top speed can be a range of 100 kilometers an hour all the way through to above 200 kilometers an hour. In my case, that would be 217 is the best I've ever clocked.

Dan Fox

These dizzying speeds require competitors to have incredible reactions and spatial awareness. This makes it a tall challenge for an AI pilot. And here, AI is can't just be trained in a virtual world.

Elia Kaufmann

The interesting challenge of FPV Racing is actually that this challenge requires to push a real physical robots to its limits, which to my knowledge hasn't been done before.

Dan Fox

This is Elia Kaufmann, one of the team behind the new AI. Now, they’re not the first to try to make an AI to pilot racing drones, but the real world poses a lot of challenges. In something like chess or Starcraft the playing field can be exactly simulated and will function the same on the day of the competition. You don’t need to take into account subtle changes in wind for example. Not so for drone racing. Another problem is motion blur. When a camera moves really fast, like if it’s on a racing drone say, then the image blurs. This blurring makes it hard for AIs to interpret camera images and work out where the drone is. Kind of important knowledge when you’re in a race. So Elia and the team have developed Swift, a two-part AI system which determines where the drone is, and where to fly it, using on-board cameras and sensors ultimately allowing it to beat its human opponents. Swift starts, like most AIs start, with a whole bunch of simulating. The AI takes a representation of the course, and the route the drone has to take, and performs multiple virtual runs to determine the fastest way round it. In an hour, it can simulate an equivalent of around 23 days worth of flying around the course. But these simulated estimates are not enough, again, the real world comes to bite.

Elia Kaufmann

If you rely on simulation to do this, we need to very accurately match what we observe in the real world, so kind of how much uplift we see, how much drag we see with what we observe in the simulation. If this would not be the case, then you see a large drop in performance.

Dan Fox

If the real world is a bit different from the drones’ simulated expectations, then it can veer off course and crash. But this is where the Swift system can adapt. After the first bout of simulation, it takes a few test flights, and uses an external camera system to monitor the drone as it moves about the course. This, combined with the drones on-board camera and sensors, allows the AI to take into account things like air turbulence that differ from the simulations. And to deal with the motion blur, it uses a neural network to identify and track the gates as it flies around the course. Altogether this meant that the AI was able to work out the best racing-lines to take around the course, and estimate its position well enough to take them. The next step was to test it against some real-world opponents. In this case, a group of champion human drone racers, who’d had a week to practice on the course. And the AI beat them all. Here’s one of its unlucky opponents, Thomas Bitmatta, who you heard from earlier.

Thomas Bitmatta

Racing against the AI drone was a really fun challenge. It took near perfect lines and I honestly think in most club level races with the system as it is with a bigger drone that's quite a heavy machine, it would actually still outperform most human pilots, it flies that good.

Dan Fox

It wasn’t perfect though. While it won most races overall, out of seven races with Thomas it only won four. For example, sometimes it would crash into an opponent. These other drones were not there when the AI did its simulations and test runs, so it didn’t know to avoid them. And, while a human can recover from a crash, the AI hadn’t been trained to do this, leaving it floundering as the opponents raced on. The real world also had some more curveballs to throw at it, like lighting. If the lighting were to change, for example if the sun were to set, after the AI had done its fine-tuning, then it would struggle on the course.

Elia Kaufmann

One of the main limitations of the current system is that it's very sensitive to environmental changes, specifically illumination changes. So if we fine tune our system, based on, you know, environment conditions that are substantially different to the environment conditions that we then actually deploy the system we saw a large drop in robustness, which means that we frequently couldn't complete the lap or that we, yeah, that we crash into gates.

Dan Fox

Elia does think this can be overcome with more training data, but in general the real-world being different from what the drone expects is still a problem. Altogether though, Elia believes that this champion-beating AI is a milestone, and one that could have a range of applications, even in our fickle real world.

Elia Kaufmann

The insights that we gain by pushing these autonomous robots to the limits allow us to transfer these insights to other domains, for example, we could speed up drones that do search and rescue missions, we could speed up drones to do autonomous inspections of buildings, for example. Because at the end of the day, a drone is limited in flight time. So if we can transfer the insights that we gained with this research to make, in general, autonomous drones fly faster in other missions, we could improve the utility of drones.

Dan Fox

That was Elia Kaufmann, formally from the University of Zurich in Switzerland and currently at the drone manufacturer, Skydio, in the US. You also heard from Thomas Bitmatta, two-time multi GP International Open World Cup champion in first person view drone racing. For more on this story, check out the show notes where there'll be a link to a video I've made, showing the drones in action.

Nick Petrić Howe

Finally in the show, it's time for the Briefing Chat where we discuss a couple of articles that have been featured in the Nature Briefing. Dan, what have you been reading about this week?

Dan Fox

Well I've been reading in Nature about this week successful lunar landing Chandrayaan-3, India's lunar lander, touching down near the southern pole of the moon, and making India only the fourth country to ever have a successful controlled landing on the moon.

Nick Petrić Howe

Alright, Ben actually mentioned this last week when we're talking about Russia's sort of failed mission to the moon. And so it sounds like India was a lot more successful. I guess, I'm wondering what went right here?

Dan Fox

Well, I guess, you know, in some ways, it's it's good news for Russia because this is a success, built in part on the failures of Chandrayaan-2, which successfully launched an orbiter, with functioning instruments, but the lander carrying the moon rover crashed onto the lunar surface just at the final moments of its landing. And so the ISRO really, you know, took some lessons from this made several design changes to the lander rover portion of the mission, added a laser sensor to measure real time velocity of the spacecraft, improved algorithms for judging deviations in propulsion and trajectory, and then generally just made the lander sort of bigger and tougher. So more solar panels, more fuel, and sort of heavier, sturdier legs are able to handle a much faster landing velocity. So just you know, a bit of a tougher craft, they also gave themselves a much larger target area. So Chandrayaan-2 was aiming for a patch, half a kilometer by half a kilometer, Chandrayaan-3, on the other hand, was aiming for an area of four kilometers by 2.4 kilometers. So just that much bigger area for it to land on.

Nick Petrić Howe

Obviously, we've already talked about Russia's attempt, they were also aiming for the south pole. And there's been a few missions going for this sort of south pole of the moon. But it's all been quite difficult for everyone. So what are the challenges that Chandrayaan had to overcome in order to actually make a successful landing?

Dan Fox

Yeah, that's right. I mean, the poles are actually incredibly challenging places to land. And, by comparison, the Apollo missions sort of had it easy, they specifically aimed for somewhere that was easy to touch down on. The Moon's poles have very sort of rough rocky terrain, it's difficult to find sort of the flat spaces that are landers looking for. And obviously landing sort of on a slope or hitting a boulder is pretty much a failure for a lander. But also even getting into position to land at one of the poles is is more difficult. It requires entering a polar orbit, which requires additional energy to move the spacecraft into that position, and introduces other uncertainties around velocity and location of the spacecraft. So even just getting ready to land is a challenge. Then at the pole, as I mentioned, it's very rocky, it's difficult terrain to land on. But it also has some real extremes of light and dark, which make it difficult to see what's going on. There are areas that are completely in the dark, there're areas that are in very extreme angle of light. So that also obscures where they're going to touch down.

Nick Petrić Howe

And so now the lander is on the moon, what's the plan? What are they aiming to do now that they're there?

Dan Fox

Well, just touching down on the moon surface was one of the key objectives of this project. That's a massive success, as I mentioned at the start, that makes India only the fourth country, after the United States, Soviet Union and China to successfully land a craft on the surface of the moon. But now that landers down it's going to release a six wheeled robotic rover called Pragyan, we're just going to ramble around for the next 14 Earth days or one lunar day, doing experiments looking at the surface of the moon.

Nick Petrić Howe

Well, congrats to the Indian Space Agency for this successful mission. And I'm sure we'll be hearing more about it as the weeks go by, or at least the lunar days go by. But for my story this week, I've been looking at something that we've talked about a bit before on the Briefing Chat, which is about scientists whose first language isn't English. And the sort of difficulties and travails that come with that participating in science, which is predominantly conducted in English. And this article that was reading about in Nature is specifically looking at how journals accommodate scientists who are not native English speakers. And essentially, they're not doing a great job is the takeaway.

Dan Fox

So how are journals failing? You know, the portion of the scientific world who don't speak English?

Nick Petrić Howe

Well, this article is based on a study, which I must say is a preprint, so it hasn't been peer reviewed yet. But this is a study that has looked at biological sciences journals. And they've sort of looked at the guidelines that are available to authors when they submit their scientific manuscripts for publishing and peer review. And the vast, vast majority in fact, every single one of the ones they looked at, which was 736 journals, bar two don't actually have explicit policies of not rejecting manuscripts based on the quality of English, so they could, if they wish to, reject a manuscript based on English, at least according to their guidelines. And overall, the study finds there isn't a great amount done to help people whose first language isn't English, if they're trying to publish a paper,

Dan Fox

Wow, those are pretty stark numbers. Out of interest, do we know which are the two journals, which did have guidelines around this?

Nick Petrić Howe

We do. So, it was actually Nature Plants and Nature. And you know, for the basis of this, I should say that Nature's news team is editorially independent of its journal team. So those were the only two journals that had specific guidelines, that manuscripts would not be rejected solely on the grounds of perceived English quality. Now, other journals did have different things that were there to help researchers whose first language isn't English, but it was by far the minority. So for example, around 8% of journals made their guidelines available in a different language other than English, less than 7% allowed authors to publish articles in language other than English. And 10% explicitly allowed researchers to use references that were from papers published in a different language than English. So journals vary in how much they do. But I think it's fair to say the vast majority of the ones, at least looked at in this study, are not doing a great amount to help people whose first language isn't English.

Dan Fox

Do the authors of this study make any suggestions about what journals should be doing to improve this?

Nick Petrić Howe

Yeah, so in the article, there's a few different examples of different societies and journals that are doing things to try and help with this issue. So, the Society for the Study of Evolution has an English language mentoring program. So the authors there, they get to work with an editor who's specifically there to help with English and clarity of the text. Other societies have like buddy systems, or volunteers that can help with English language. And one other thing that the article touches on is just being mindful of the fact that someone's first language isn't English. There was a researcher who was interviewed for this article, whose first language was not English, and they said that they were deeply affected by what a reviewer wrote about their manuscript, the reviewer said that their sloppy language challenged the credibility of their work. And this researcher who was interviewed, they said that this review really destroyed their confidence in their science. So really trying to be mindful of the fact that not everyone is coming from the same level and understanding of English could be really helpful. And they do also suggest that maybe AI tools could help with this in the future. Perhaps there'll be AI proofreading services and things like that. Some journals do suggest that people use professional editing services. But, you know, that's not really affordable for everyone. So maybe the some of these AI tools can sort of fill in the gap. And finally, one thing that a researcher interviewed for this article suggests is maybe journals could offer an extended abstract for researchers that they could write in their native language to sort of explain the science clearly in simple words that then would be accessible for everyone.

Dan Fox

Well, hopefully, journals will be able to implement some of these suggestions in the paper and start taking these language considerations into account. Thanks, Nick. I think that's all we've got time for this week on the Briefing Chat. Listeners, for more on those stories, check out the show notes for some links and a link to where you can sign up to the Nature Briefing to get more stories like those direct to your inbox.

Nick Petrić Howe

That's all for this week. As always, you can keep in touch with us on X, we're @naturepodcast, or you can send an email to podcast@nature.com. I'm Nick Petrić Howe.

Dan Fox

And I'm Dan Fox. Thanks for listening.