Welcome to Bedtime Astronomy. Explore the wonders of the cosmos with our soothing Bedtime Astronomie podcast. Each episode offers a gentle journey through the stars, planets, and beyond, perfect for unwinding after a long day. Let's travel through the mysteries of the universe as you drift off into a peaceful slumber under the night sky.
Imagine staring at the night sky without blinking, like, not for a minute, not for an hour, but for four straight.
Years, just an unyielding, relentless gaze locked onto.
The dark exactly, no glancing away, no resting your eyes. That is exactly what NASA's Transiting Exoplanet Survey satellite that are known as TESS has been doing. It has been staring completely unblinking at over two point two million stars. And it's not just looking at them in a passive, romantic sort of way.
No, it's watching for the absolutely faintest, most microscopic dimming of their starlight.
Yeah, we are talking about drops and lights so small that they push the absolute boundaries of our technological limits.
To really appreciate the scale of that microscopic dimming, we basically have to throw out our everyday understanding of brightness. How so, well, we are talking about a drop in stellar luminosity that is so minuscule it would be entirely imperceptible to the human eye, even to you are floating in a spacesuit right next to the telescope lenses. Yeah.
Tests is essentially this collection of ultra sensitive cameras designed to capture incredibly subtle dips in light, dips that occur when a planet passes directly in front of its host star from our specific vantage point here in the Solar System.
So it's kind of like a game of shadows played across trillions of miles of empty space.
That's exactly what it is.
And because TESS has been maintaining this unblinking vigil for its first four years of operation, it has accumulated a mountain of data that is frankly almost impossible for a human brain to comprehend.
Oh, the volume is staggering. Yeah, we are talking about continuous, unbroken light curves for two point two million distinct fusion reactors scattered across the galaxy.
But today we are jumping straight into a monumental breakthrough that just emerged from that exact mountain of.
Data, right the new research.
Yeah, astronomers at the University of Warwick have successfully waded through that digital ocean and validated over one hundred exoplanets. Specifically, they've handed us one hundred and eighteen validated planets, and the really exciting part is the thirty one newly detected ones.
Right, Exactly thirty one of those are completely fresh discoveries that literally nobody on Earth knew existed until the publication of this research, which is wild, it really is. Thirty one brand new worlds is a staggering yield for a single study, especially when you consider how thoroughly the low hanging fruit in exoplanet research has already been picked.
Yeah, we've been at this for a while now, we have.
But the Warwick team didn't stop there. Alongside those one hundred and eighteen validated planets, they also produce this massive, highly refined catalog of over two thousand high quality planet.
Candidates, and nearly one thousand of those candidates are entirely new to science too.
They've basically handed the global astronomical community a highly detailed, curated treasure map.
Okay, let's unpack this, because to really wrap your head around how difficult it is to find these things. Let's think about the method they're actually using, the transit method. Right in astrophysics, this is called the transit method. Imagine trying to spot a moth flying in front of a distant street light.
I love this analogy, thanks, But the street.
Light isn't down the block at the end of your street. It is hundreds of light years away. You can't see the moth itself. The moth is entirely invisible.
You're just looking for the shadow exactly.
You are solely trying to measure the exact fraction of a shadow that the moth casts on your eye as its tiny, fragile body blocks a microscopic sliver of the bulb's light for just a few hours.
It's a great way to picture it, but I'd actually take it a step further to highlight the sheer statistical improbability of what test is doing.
Okay, go for it.
So the distant street light is the host star and the moth is the exoplanet. Yes, but remember planets don't just fly around randomly. They orbit in a flat plane, right like a record player exactly, So for us to see that shadow, that temporary drop in the star's light curve. The geometry has to be absolutely perfect. The edge of that planetary system has to be perfectly aligned with our line of sight.
Oh, I say yeah.
If the system is tilted even a few degrees up or down relative to Earth, the planet just orbits above or below the star from our perspective, and we see absolutely nothing.
So not only are we looking for a moth's shadow from light years away, we are only able to see the tiny fraction of moths that happen to be flying on the exact right trajectory across our line of vision.
Precisely, which means for every planet test actually catches, there are likely dozens, if not hundreds of others orbiting just out of geometric alignment.
That's humbling. The universe is teeming with these things, but we are only catching the ones that cross our incredibly narrow.
Tripwire, and we are looking for a highly specific periodic dip in brightness. The moth has to circle the street light and block the light again and again on a predictable schedule.
Right. That periodicity is the first major clue it is.
But you know, the universe is a remarkably messy, chaotic place. A lot of things can cause a star's light to flicker or dim like what, well, stars have massive internal weather systems, They have sunspots larger than the Earth that rotate in and out of you. They have violent flares.
Oh right, so it's not a steady bolt exactly.
And the instruments on the satellite itself experienced thermal shifts that introduce noise into the data. Distinguishing the delicate, repeating shadow of our proverbial moth from all the other violent cosmic noise is the real challenge of modern astronomy.
Which brings us to why this matters to you listening right now. The focus of this breakthrough isn't just about finding a few new dots in the sky or adding a list of unpronounceable alphanumeric names to a stellar catalog.
Definitely not.
It's about a revolutionary new artificial intelligence tool that is completely changing how we map the universe. We are witnessing a fundamental shift in our astronomical capabilities.
We really are. The era of humans painstakingly plotting graphs to find planets is over. It had to end, didn't the Oh absolutely, when you were looking at continuous light readings from two point two million stars, taking measurements every two minutes. In some cases, the data volume becomes so vast that it is physically impossible for human eyes to process at all.
Not even massive international teams of human eyes.
Not a chance. You can't just throw a room full of tired graduate students at this problem and ask them to stare at squiggly lines on monitors all day.
Right, that sounds like torture.
It is. You need a way to separate the true planets from the cosmic illusions, and you need to do it at an industrial scale. And that absolute necessity is exactly what birth this new AI pipeline.
You know. I always find it fascinating how the biggest roadblock in a scientific field often isn't the technology used to gather the data, but our capacity to actually read it.
That's very common issue in modern science.
So what exactly is the bottleneck here? If tests is so incredibly sensitive, why is it so hard to just look at a dip in the light curve and confidently announced to the world, yep, we found another Earth. Pack your bags.
Well, the massive bottleneck in modern astronomy right now isn't detection at all. It's conformation. Okay, missions like tests or its legendary predecessor, the Kepler Space Telescope, are incredibly proficient at flagging anomalies. They routinely identify thousands of possible planet candidates.
They say, hey, the light dipped tier, you should look at this exactly.
But confirming which of those signals are real physical planets and which are illusions is painstakingly slow. The primary culprit holding everything up in the astronomical community is something we call false positives.
Because the universe loves to play tricks on us.
Oh, it really does. It generates phenomena that look exactly like what we are searching for but are entirely different beasts.
Right, and the most common, the most frustrating trick the universe plays on exoplanet hunters is the eclipsing binary star system.
Right, that is the big one. Yes. To understand this, we have to remember that our own sun, which is a solitary star floating alone in space, is actually somewhat of an anomaly. Really yeah, A huge percentage of the stars in our galaxy are binary systems, meaning two stars locked in a gravitational dance orbiting around a common center of.
Mass like a tattooine situation to use the mandatory Sci Fi reference two suns in the sky exactly that.
Now, imagine those two stars orbiting around each other from our specific line of sight. If the orbital plane is edge on to Earth, one star will periodically pass in front of the other. Okay, I'm picturing it. Sometimes the stars are of different sizes and temperatures. When a smaller, dimmer, cooler star passes in front of a larger, brighter, hotter star, it blocks a portion of the bright star's light.
Which creates a dimming effect in the overall light we receive in our telescopes.
Right and mechanically, mathematically, that specific dip in light can perfectly mimic the transit of a large Jupiter sized planet.
Let me make sure I'm visualizing this correctly. The telescope is basically just a bucket collecting photons.
That's a good way to put it.
It doesn't see a picture of a star. It just measures the amount of light hitting its sensor. So the telescope sees the light dropped by say one percent, flags it as a potential giant planet crossing the star, but it's actually just a smaller, fainter star getting in the way.
Yes, it's a stellar eclipse, not a planetary transit.
Man. That is deeply frustrating.
That's the core of the false positive problem. And it gets even more insidious. You have what are called grazing binaries. Grazing binaries, Yeah, this is where the two stars aren't perfectly aligned with us, but they just barely clip the edge of each other's disc as they.
Orbit, just a glancing glow.
Exactly, that tiny little clip that grazing pass creates a very shallow dip in the light curve, and a shallow dip is the exact signature we expect from a small Earth sized rocky planet. So a grazing binary star system can totally masquerade as a potentially habitable Earth like world.
So how did astronomers handle this before the AI? I mean, you have a list of thousands of these signals from tests. How do you figure out if it's a moth or just another slightly dimmer street light swinging in the wind?
Well? Distinguishing between a planetary transit and eclipsing binary historically required intense, incredibly time consuming follow up observation.
And pointing other telescopes at it.
Exactly you couldn't just use tests anymore. You had to requisition time on completely different, massive ground based telescopes on Earth. Oh wow, You'd use instruments called spectrographs to analyze the light spectrum of the star looking for minute gravitational wobbles. It's a technique called radial velocity.
How does that tell them? Apart?
A planet will cause au star to wobble very slightly, but a massive companion star will cause the primary star to yank back and forth violently.
But getting time on those giant ground based observatories is notoriously competitive. You can't just point the Keck telescope in Hawaii at two thousand different stars just to double check Tess's homework.
No, it would take decades and cost millions of dollars.
Which is precisely why we have a massive backlog of unconfirmed candidates just sitting on servers waiting to be vetted.
Exactly. The traditional scientific process simply could not keep pace with the sheer volume of data the satellite was beaming down to Earth every month. We were drowning in potential discoveries, completely paralyzed by the false.
Positive enter raven. Yeah, this is the newly developed AI pipeline from the team at the University of Warwick that is cutting through the backlog like a Scythe and raven stands for a ranking and validation of exoplanets. I always appreciate a backronym in astronomy.
They are very good at them.
They really are its core function. Its entire reason for existing is to solve this exact bottleneck. It is a bespoke software architecture designed to distinguish between actual planetary transits and these astrophysical impostures without needing years of expensive ground based telescope.
Time and the way to choose this is an absolute master class in modern computational astrophysics.
How does it actually work.
It doesn't just look at the overall depth of the shadow and make a guess. It uses deep machine learning, specifically advanced neural network architectures to analyze the entire shape.
Of the light curve, the shape of the curve itself.
Right. It looks at the duration of the dip, the speed at which the light drops the tiny, almost imperceptible variations at the edges of the shadow as the planet begins its crossing. It searches for the subtle mathematical signatures that separate a solid spherical planet from a glowing dynamic star.
Okay, let's unpack this methodology, because this is where I have to push back a little on how AI usually operate rates. When I was diving into the research papers detailing how they built Raven, I noticed something that sounded completely counterintuitive, almost paradoxical. The training data exactly. Usually, when you train an AI, you feed it real world examples. If you want an AI to recognize a cat, you
show it a million photographs of actual cats. Right, But the Warwick researchers didn't train this AI on the thousands of real exoplanets we've already discovered and confirmed over the last twenty years. Instead, they trained Raven on hundreds of thousands of realistically simulated planets and simulated stellar events.
Yes they did.
I have to admit I was scratching my head. We are training an AI on fake computer generated planets so it can find real ones. How does learning from a massive video game simulation of the universe give us an edge over just studying the real, messy data we already have from Kepler and Tess.
What's fascinating here is that your hesitation hits on the absolute most crucial debate in machine learning. Right now. Your confusion is exactly why this approach is so brilliant and why it represents such a leap forward.
Okay, explain that to me, it all comes down.
To how neural networks actually learn. Machine learning models excel at pattern recognition. They are vastly superior to humans at finding hidden correlations in high dimensional data. But to learn those patterns correctly, they need training data that is perfectly categorized. They need to know, without a shadow of a doubt, what the right.
Answer is, So they need an infallible answer key like if you are teaching a child math, you can't give them a textbook where half the answers in the back are wrong or you know, probably right, but we aren't totally sure exactly.
In computer science we call this the ground truth. If we were to train an AI exclusively on real data from tests or even our catalog of previously confirmed planets, we are fundamentally training it on data that is full of noise, uncorrected instrument artifacts, and most importantly, unknown.
Variables, even for the ones we think or confirmed, even.
Those even with planets we think we've confidently confirmed using ground based telescopes, there's always a margin of error. There's always a tiny, nagging percentage of doubt.
Kind of doubt.
Well, what if a faint background star is blending its light with our target, subtly altering the signal? What if our understanding of the star's mass is slightly off?
Ah, I see where this is going. If you train an AI on flawed, messy human data, it just inherits all of our own blind spots. Precisely, If the astronomical community has been systematically misclassifying a certain type of weird grazing binary star as a rocky planet for the last ten years, and we feed that into the AI, the AI will just learn to make that exact same mistake, only it will make that mistake a million times faster and with terrifying confidence.
And that is the absolute danger of black box AI in science. It amplifies human error. But by generating simulated events, the researchers hold the ultimate perfect answer.
Key Because as they built the simulation.
Right, they built incredibly complex mathematical models of stars, They generated artificial planetary transits across them, calculating the exact geometry of the shadows. They added synthetic telescope noise that perfectly mimics the thermal shifts and pixel bleed of the test cameras.
That's incredibly thorough.
And critically they generated artificial eclipsing binary stars and grazing binaries. Because the researchers created the simulation from the ground up, writing the laws of physics in code, they know the absolute ground truth of every single pixel of data.
They know with one hundred percent mathematical certainty, whether a specific simulated light curve was caused by a fake planet or a fake binary star exactly. And it makes total sense when you frame it like that, you aren't just showing it pictures of shadows. You are teaching the AI the exact, pure mathematical signature of a planet, completely divorced
from human error and observational bias. Yes, you're giving it the fundamental physics of an orbital transit isolated in a stere laboratory environment.
And by simulating hundreds of thousands of these events, they can expose the AI to scenarios that might be incredibly rare in our current real world catalogs, but theoretically possible in the wider universe, Like what well they can train the AI on what a planet looks like orbiting a highly active star covered in massive sunspots, or what a planet theory transit looks like if the planet has a massive ring system like Saturn.
Oh, that's so smart, right.
Once the neural network has completely mastered the physics of the simulation, once it can perfectly distinguish the fake planets from the fake stars, then and only then is it unleashed on the real, messy tests data.
And because it has internalized the underlying signature of a true planet so deeply, it can cut through the noise with incredible precision. It knows what a moth's shadow looks like, even if the street light is flickering, and even it fits raining precisely.
And it doesn't just give a yes or no answer. Raven provides a statistical probability.
Oh, it gives a confidence score exactly.
It looks at a candidate and says, based on my training, there is a ninety nine point nine percent chance this is a planetary transit, and only one point one percent chance it is a background eclipsing binary. When that probability crosses a certain rigorous threshold, the candidate is upgraded to a validated planet.
And the other thing that makes Raven such a paradigm shift is the workflow advantage. I was reading about how this used to be done and in the past, confirming a planet was this incredibly fragmented, piecemeal process oosen Nimare. You'd have one astronomer write a piece of software to clean the raw data. Then you'd pass that data to another team using a different algorithm to search for periodic dips. Then a human might literally eyeball the resulting.
Graphs yep, lots of staring at graphs.
Then if it looks promising, you'd feed it into a completely different statistical tool to calculate the probabilities of a false positive.
It was a very patchwork, fragmented ecosystem of tools. Every research group had their own preferred software packages written in different programming languages.
That was incredibly inefficient.
It inherently slows down the science, makes it difficult to reproduce results, and introduces opportunities for errors or data corruption at every single handoff point. It was well artisanal planet hunting, but.
Raven handles the entire process end to end. It's a unified ecosystem.
From start to finish.
It takes the raw, messy, noisy data straight from the satellite, processes the light curves, detects the potential periodic signals, vets those signals using its highly trained machine learning models to weed out the astrophysical imposters, and then statistically validates the surviving planets, all in one seamless automated pipeline.
That's the beauty of it.
It takes the grueling, repetitive heavy lifting out of the equation and just hands the astronomers a highly reliable, mathematically sound list of worlds.
It's the difference between a master mechanic trying to build a car from individual parts they ordered from a dozen different catalogs versus having a fully automated, state of the art robotic assembly lie perfect analogy. The efficiency and consistency are what allows us to look at two point two million stars and actually makes sense of him in the human lifetime.
So now that we understand how this AI works and why training it on a massive simulation was the secret to its success, we need to look at what it actually found when they turned it on. The fun part, because the AI didn't just find a bunch of generic, boring dead rocks floating in empty space. It uncovered some of the most extreme, chaotic and alien worlds imaginable. Hidden in that test data, the thirty one newly discovered planets
are not places you would ever want to visit. Let's talk about the exoplanet menagerie.
The diversity of planetary systems out there is truly staggering. Our own Solar system, with its neat division of inner rocky planets and outer gas giants, is just one tiny, specific flavor in a cosmic ice cream shop of bizarre configurations.
And it's important to note the specific focus of the Warwick team's study to understand what Raven was finding. The pipeline was specifically off demise to look for planets with very tight orbits. We are talking about planets that complete a full orbit around their star a full planetary year in less than sixteen days.
Sixteen days. Earth takes three hundred and sixty five days to make that trip. Mercury, the closest planet to our Sun, takes eighty.
Eight days, So a sixteen day year means these planets are hugging their host stars incredibly uncomfortably closely. They are practically skimming the surface.
They are deeply embedded in the intense gravitational and radiational environment of their stars. To give you a sense of scale, if Earth were on a sixteen day orbit, the Sun would take up a massive portion of the sky, the heat would be unimaginable.
And within that already extreme sixteen day parameter Raven managed to detect a population of what astronomers classify as ultra short period planets, or usps.
These are worlds that take the concept of a tight orbit to an absurd extreme.
These are planets that orbit their stars in less than twenty four hours.
Less than an Earth day to complete a full year.
I want to try and visualize that reality for a second. Jugene celebrating New Year's Eve. Every single day you wake up, it's a new year. You go to sleep, the year is over. Your entire planetary cycle season orbital dynamics is compressed into a single Earth day.
It's frantic.
A planet moving that incredibly fast, orbiting that perilously close to its host star is experiencing an environment that completely breaks the human imagination. At that proximity, the gravitational forces are immense. The gravity of the star almost certainly locks the planet in place. A phenomenon we call tidal locking.
It's the same physics that keeps one side of our Moon always facing Earth right, So for these ultra short period planets, one side of the planet always faces the nuclear furnace of the Sun and the other side is cast in eternal freezing darkness.
And the dayside of a tidally locked ultra short period planet isn't just hot in the way we can comprehend. It is a literal physical hellscape, like how hot. Because they are so close, the surface temperatures on the star facing side can easily reach several thousands of degrees. At those temperatures, rock doesn't just melt into magma. It vaporizes vaporized rock.
Wow.
Yeah, the extreme heat breaks down molecular bonds. You are likely looking at unimaginable landscapes dominated by global churning oceans of molten lava, and the atmosphere, if the stellar winds haven't entirely stripped away, would be composed of vaporized rock and heavy metals.
So we are talking about planets where the weather forecast might literally call for scattered showers of liquid iron or condensation of vaporized glass.
Pretty much.
Yeah, and thanks to this AI pipeline. These aren't just theoretical mathematical models on a whiteboard anymore. Raven is casually validating dozens of these nightmare worlds in a single sweep of the test data.
It's finding them so efficiently that we can start studying them as a broader population rather than just isolated freaks of nature.
And studying them tells us so much about the violent history of solar systems. Plants don't generally form that close to a star, do they know?
The environment is way too hot and volatile for dust and gas to clump together and form a planet.
So how did they get there?
Well, which means these ultra short period worlds likely formed much further out in their solar systems where it was cooler, and then, over millions of years, gravitational interactions with other planets dragged them inward. They spiraled closer and closer to the star until they parked in these extreme roasting orbits.
It's the kind of extreme physics that used to belong entirely to the realm of science fiction, and now it's just a Tuesday for an AI algorithm at Warwick. But the extreme environments Raven found aren't just limited to these solitary roosting rocks. The pipeline also proved incredibly adept at detecting close orbiting multiplanet systems. We are talking about previously unknown planetary pairs, or even entire groups of planets orbiting the same star, all packed into incredibly tight spaces.
This is where orbital mechanics gets incredibly complex, and where the AI truly flexes its computational muscle.
How tired are we talking?
Well? When we think of a Solar system picture, our own planets spread out over vast distances, neatly separated by millions of miles of empty space, Mercury than Venus than Earth, all keeping a respectful distance, right, But the universe has a way of packing planets together in configurations that completely challenge our traditional models of planetary formation.
A perfect specific example of this, to help visualize exactly the kind of chaos Raven is trained to untangle, is the Kepler eleven system. Now, Kepler eleven was discovered before Raven, but it perfectly illustrates the phenomenons.
A great example.
Kepler eleven is a sun like star relatively similar to our own, but it has six confirmed planets orbiting it, and those planets aren't spread out elegantly like ours. Five of the six planets and Kepler eleven orbit closer to their star than Mercury orbits our Sun. It's wild to think about imagine taking Venus, Earth, Mars, Jupiter, and Saturn and crushing all of their orbits down so they fit inside the orbit of Mercury. It is an unbelo believably crowded, claustrophobic neighborhood.
And from our observational perspective, watching the light curve of a crowded system like that is both chaotic and.
Beautiful because they're all blocking the light right.
Because the orbits are so tight and the planets are moving so fast, you don't just see one simple neat planetary transit at a time. You regularly have situations where two or even three planets are passing in front of the host stars simultaneously.
Which compounds the starlight dimming. You get a shadow within a shadow.
Precisely, the light drops when Planet A enters the disk. Then, while Planet A is still crossing, Planet B enters and the light drops further.
Like stair steps.
Exactly, Then Planet A leaves and the light goes back up a bit, but Planet B is still there. The shadow deepens and lightens in these complex overlapping asymmetrical steps, untangling a light curve like that. Trying to isolate the distinct mathematical shadow of three different planets overlapping each other while accounting for the gravitational tugs they exert on each other that slightly alter their transit times is a monumental mathematical puzzle.
It's like listening to a massive symphony orchestra playing a complex piece of music and being asked to perfectly isolate and transcribe the notes of three specific violins playing at the exact same time while the drummer is occasionally hitting the microphone stand.
That is a very accurate way to describe the noise.
A human brain struggles to separate those signals. But it's exactly the kind of complex overlapping pattern recognition that a well trained neural network like Raven is built to decode. It looks at the tangled mess of shadows and instantly identifies the individual players.
And discovering these compact multiplanet systems is vital because they are highly unstable on a cosmic timescale. The gravitational interactions in the system pack that tightly are immense.
So they're tugging on each other constantly.
Yeah. By cataloging them, we can study orbital resonances, which is how planets sync up their orbits to avoid colliding, and we can test our theories on planetary migration.
But amidst all this crowded, chaotic orbital traffic, amidst the lava worlds and the pack systems, the researchers also use raven to explore something entirely different. They use this incredible tool to look at a very specific and surprisingly empty region of space. I was fascinated by this part of the study. The AI helped clarify the boundaries of a rare class of planets found in an area known as the Neptunian Desert.
The Neptuny Desert is one of the most intriguing and hotly debated concepts in modern astrophysics. Essentially, the AI mapped out the demographics of planets and confirmed a region where theoretical physics strongly predicts planets of a certain size and composition should be almost entirely non existent.
Okay, this raises an important question, why is it a desert. If we are finding planets packed in everywhere else, if we are finding planets hugging their stars on sixteen hour orbits and magma oceans, why does physics say this specific zone, for this specific type of planet should be a barren wasteland.
It has to do with the dell violent balance between a planet's physical size, its internal composition, and its proximity to the intense radiation of its host star.
Okay, break that down for me to break down.
When we talk about Neptunian planets or sub Neptunes, we mean worlds roughly the size of our own ice giants Uranus and Neptune. These are planets that are significantly larger than Earth, but not quite as massive as Jupiter.
And what are they made of?
Crucially, they are composed of a relatively small rocky or icy core surrounded by a massive thick envelope of hydrogen and helium gas.
So they are basically a giant, dense ball of gas and ice with a little pit in the center.
Think of a peach, where the pit is the rocky core and the massive amount of flesh around it is the gas atmosphere. Now, if a planet like that forms further out in the Solar System. It's fine, it's coal, the gas is stable. But if that Neptunian planet migrates too close to its host star, say it gets dragged into that sixteen day or less orbital zone we were talking about earlier, it enters a profoundly hostile environment. It gets cooked, it is subjected to intense, searing stellar radiation.
We are talking about a relentless bombardment of X rays, extreme ultraviolet light, and powerful stellar winds constantly blasting the planet's upper atmosphere.
Okay, but a planet like Jupiter is also mostly gas, and we find hot jupiters orbiting incredibly close to their stars all the time. Why doesn' Jupiter get destroyed?
That is the crucial distinction. For a massive gas giant the size of Jupiter, its own internal gravity is immense. The gravitational pull of the planet is so strong that it can hold on to its massive gaseous atmosphere despite the intense radiation trying to blast it away. It's too heavy to be blown apart easily got it. On the other end of the spectrum. For a small, dense, rocky planet like Earth or Venus or one of those lava worlds we discussed, there isn't a massive thick envelope of
hydrogen gas to strip away in the first place. It's just a solid rock.
Ah, So a neptune sized planet is caught in the tragic metal ground.
It's the worst of worlds precisely. Its gravity isn't strong enough to tightly hold onto that massive, fluffy gaseous envelope against the relentless assault of the star's high energy photons. But it has too much gas to just ignore the radiation.
So what happens to the gas?
Theoretical models, specifically mechanisms like photoevaporation, where the light literally heats the gas until it escapes the planet's gravity, predict that the intense radiation should literally boil the atmosphere off into space.
Wow.
Over a span of millions or billions of years, the Neptune is violently dismantled. The gas is stripped away entirely, leaving behind nothing but its bare much smaller rocky core.
So the peach gets entirely eaten away by the star, leaving only the pit, and that pit just looks like a slightly large, rocky planet. Therefore, according to the physics of photo evaporation. Finding an intact, fluffy, neptune sized planet with its atmosphere still attached that close to a star should be statistically impossible.
It should be a desert, and yet we found some. Really Yeah, Raven's unparalleled precision allowed it to sift through the noise and validate a handful of worlds sitting right in the middle of this supposed wasteland. We are detecting intact neptunes surviving in a radiation environment that our current models say should have completely dismantled them eons.
Again, which means our understanding of the universe is broken, or at least incomplete. If these planets exist where they shouldn't, there are survival mechanisms at play that we don't fully grasp yet.
It forces the entire astrophysical community to reconsider the timeline and mechanics of planetary evolution. It poses massive questions like what, well are our models of photo evaporation too aggressive? Perhaps these neptunes only very recently arrived in their tight orbits. Maybe they migrated inward late in the system's life and the star just hasn't had enough time to strip them there. Yet we are catching them in the act of.
Dying, or maybe the planets themselves have defenses we didn't anticipate. What if they have internal planetary magnetic fields that are orders of magnitude stronger than anything we've observed, acting like a mass of deflector shield protecting their atmospheres from the stellar wind.
Exactly. These are the kinds of profound field altering discoveries that drive physics forward. When the data contradicts the theory, the theory must evolve finding these bizarre, extreme individual worlds, the lava planets, the crowded, chaotic systems, the impossible surviving neptunes is an incredible feat of engineering and software design.
But when I was looking at the scope of the Warwick team's work, I realized that the true power of this AI isn't just acting as a cosmic stamp collector building a gallery of space oddities and weird planets. It's about doing planetary census work on a galactic scale. We are moving from studying individual interesting trees to mapping the demographics of the entire forest. Let's shift gears and look at the broader demographics of the galaxy. Because the numbers Raven generated are mind bending.
This is where the Warwick Team's companion study, which was also publish recently, really changes the game for theorists.
Yeah, let's get into those numbers.
By having a data set that is so heavily vetted by Raven and so demonstrably free of those pesky false positives we discussed earlier, the researchers could confidently step back and perform massive demographic analysis. They mapped out the prevalence of these close in planets with unprecedented surgical detail. They weren't just asking what is out there? But how common is it?
And the big demographic statistic they unveiled is staggering. It completely reorients your perspective of the galaxy. They found that around nine to ten percent of all Sun like stars host a close end planet, meaning a planet orbiting in less than sixteen days.
Let's contextualize that number because it is massive. When we say some like stars, astronomers are specifically referring to what we call FG and K main sequence stars.
Okay, what does that mean for a lay person?
Stars are classified by their temperature and spectrum. F stars are a bit hotter and larger than our Sun. G stars are exactly like our Sun, and K stars are slightly cooler, smaller or dwarfs. These FGK stars make up a very significant, stable portion of the stellar population in the Milky Way. They are the prime real estate for finding planets and potentially life.
Here's where it gets really interesting. I want you to go outside tonight, assuming it's a clear night wherever you are listening from, and just look up at the night sky. Pick out the stars you can see with your naked eye. The vast majority of the individual stars you can see
are relatively close to us in the galactic neighborhood. Statistically speaking, roughly one in ten of the sunlike stars you were looking at right now has a planet whipping around it at breakneck speed, completing a year in less than sixteen days. It's hard to wrap your head around ten percent. It completely changes how you view the cosmos. The sky isn't just a collection of lonely isolated fusion reactors burning in the void. It is teeming with complex planetary systems operating
on hyper speed. Every tenth star you point your finger at has a world practically touching its surface.
It is a profoundly humbling to know that the night sky is that crowded. Now to be fair to the history of astronomy and to provide the proper context. This nine to ten percent figure isn't entirely new. It is largely consistent with the earlier estimates produced by NASA's legendary Kepler mission, which operated in the decade before tests. Kepler
was the great pioneer of exo planet demographics. It stared at a single fixed patch of sky in the Sickness Constellation for years, and it gave humanity our first, real, mathematically sound estimates of planetary occurrence rates.
Right. I'm glad you brought that up, because when I read the ten percent number, my first thought was, didn't we already know that from Kepler? If Kepler already gave us that ten percent estimate a decade ago, what makes the test data and the raven pipeline so groundbreaking today? Why is this a new breakthrough?
It fundamentally comes down to precision and confidence. Kepler gave us the initial estimate, but it was an estimate with relatively wide aerror bars.
So they weren't totally sure exactly.
Yea, there was still a margin of uncertainty because of the false positive rates and the limitations of the older software. Raven processing the incredibly rich test data mapp this exact same demographic, but with uncertainties that are up to ten times smaller.
Wow, ten times more precise. That isn't just a minor incremental update. That's a massive leap and resolution.
Think of it this way. Kepler proved the concept. It was like an early explorer sailing across the ocean and sketching a rough pencil outline of a newly discovered continent. They could tell you the continent was there and roughly how big it was. But Raven and tests are refining that map to a degree we've never seen.
They're mapping the actual terrain.
They are like a modern constellation of satellites providing a high definition, millimeter accurate topographical map of that same continent. When your statistical uncertainty shrink by a massive factor of ten, your theoretical models of how solar systems form suddenly have to become much much more rigorous to fit the new data. You can't hide sloppy physics inside way airbars anymore.
That makes perfect scent. The tighter the data, the harder it is for a bad theory to survive, And that exact precision is what allowed the Warwick researchers to finally quantify the Neptunian desert. We were just talking.
About, yes, the desert numbers.
We discuss the physics of why those planets are rare, how their atmospheres get boiled away. But the Raven pipeline allowed the researchers to move beyond just saying they are rare and actually provide the first direct, highly precise, empirical measurement of this phenomenon. They calculated that Neptunian planets surviving in tight orbits occur around a mere point zero eight percent of Sun like.
Stars, less than one tenth of one percent. If we connect this to the bigger picture of astrophysics, that point zero eight percent isn't just a neat piece of trivia for a textbook. It is a hard, unyielding mathematical boundary. It is an empirical constraint on the fundamental physics of solar system formation and atmospheric escape.
It's the universe drawing a literal line in the sand. It tells us, with absolute numerical precision, exactly where the tipping point is betwel in a planet surviving its stars, radiation and a planet being mercilessly boiled down to its bare, rocky core, and.
By establishing these hard boundaries the ten percent prevalence of short period planets the point zero eight percent survival rate in the extreme radiation of the Neptunian desert, the Raven team is providing the anchor points for every single theoretical astrophysicist trying to write the history of the galaxy.
So they have to build their models around these number.
Exactly when a scientist today sits down to build a complex computer simulation of how dust and gas coalesced to form planets, they have to ensure that their simulation naturally produces these exact percentages. If your computer model of a solar system forms Neptunes in tight orbits five percent of the time, you know your model is wrong, because the Raven data says it only happens point zero eight percent of the time. It keeps our theoretical physics strictly grounded in observational reality.
The extreme precision of these demographics fundamentally shifts how astronomers will plan their next decades of research, which brings us to the final and perhaps the most exciting part of this breakthrough, where we go from here because this paper isn't just the end of a study. It's not a closed book. It is the beginning of a whole new era of exploration. The Warwick team has embraced what I think is the absolute best part of modern science, the open source universe.
It is a vital transformative point. The development of a pipeline like Raven doesn't just generate astronomical discoveries in a vacuum. It simultaneously stress tests the absolute limits of artificial intelligence on incredibly difficult, noisy, real world research problems.
Oh that makes sense. A pushing AI forward to.
Absolutely The complex neural network architectures and mathematical frameworks they're building to find exoplanets push the boundaries of machine learning in ways it could eventually benefit completely different, unrelated fields of science. The way Raven analyzes time series data to find a tiny drop in light could theoretically be adapted to look for microscopic anomalies in medical data or seismic waves.
And crucially, the researchers haven't hoarded this AI or the massive catalogs of data produced. They didn't lock it behind a patent or a private university server. The team at WARG has released interactive tools, the underlying code and the full comprehensive catalogs of there one hundred and eighteen validated planets, and there are thousands of high quality candidates to the global scientific community.
It's a huge gift to the field.
They've essentially built a highly accurate AI verified roadmap of the local galaxy and they've just handed the keys to anyone on Earth with a telescope and an Internet connection.
And that roadmap is absolutely critical for the next generation of space observation because the telescopes that we are building now are marvels of engineering, but their time is precious. We have massive new missions on the immediate horizon, like the PLATO mission right yes, For example, the European Space Agency is preparing for the PLATO mission, which stands for Planetary Transits and Oscillations of Stars. It is set to launch in the near field future, and PLATO is a
massive step up. It is specifically designed to search for potentially habitable terrestrial Earth like planets orbiting in the habitable zones of Sun like stars.
But launching a space telescope costs billions of dollars and observing time is incredibly competitive. And expensive. You can't just point Plato blindly at the sky and hope for the best, and you certainly can't waste its valuable time looking at targets that turn out to be false positive grazing binaries.
Exactly, you need a rigorously vetted target list. You need to know exactly where to point the multi billion dollar instrument. The one hundred and eighteen planets dalinated by Raven and the thousands of pristine, high quality candidates that identified become that ultimate target.
List, so they know exactly where to look.
Ground based observatory is and upcoming space missions like Plato will use this exact catalog generated by the AI. They will know precisely which stars warrant the intense expensive follow up required to measure the exact mass of a planet or to use spectroscopy to look for the chemical signatures of water, oxygen, or methane in an alien atmosphere. Raven has cleared the brush so Plato can see the forest.
So what does this all mean When we take a step back from the convolutional neural networks, the sixteen day orbits, and the Neptunian deserts and look at the whole picture, we are seeing a profound evolution in how humanity interacts with the cosmos. Artificial intelligence has allowed us to take the raw, chaotic, impossibly noisy light from two point two million distant stars and to still it into a high fidelity, highly precise map of alien worlds.
It's beautiful, really.
We've automated the discovery of the universe. We have outsourced our sense of wonder to an algorithm, and the algorithm is showing us things we never could have found on our own.
It is the perfect synthesis of human ingenuity and machine capability. We built the magnificent telescope to gather the agent light, We built the intricate mathematical simulation to teach the machine the underlying physics of reality, and the machine tirelessly processed the sheer crushing volume with data that our biological brains never could have handled alone. Is a partnership that expands our reach into the dark.
It makes you wonder about the sheer volume of information we've already collected across all fields of science, which leaves you with a lingering question to ponder long after we finish today. If a newly designed AI pipeline train purely on simulated mathematical realities can uncover over one hundred hidden exotic worlds hiding in plain sight within a four year
old data set from tests. While what other fundamental earth shattering truths of the universe are currently sitting quietly in our existing data archives right here on Earth, just waiting for the right algorithm to finally open our eyes to them.
The data is already there, resting on our hard drives. The universe has already sent the message. We just need to learn how to build the right machine to read it.
It's exactly like looking for the impossibly faint shadow of a moth against a distant street light hundreds of light years away. The shadow is always there. Sometimes you just need a machine to tell you exactly how to look. Usai