# Simulating ALL 100 billion stars in the Milky Way for the first time (with the help of AI?!)

## Метаданные

- **Канал:** Dr. Becky
- **YouTube:** https://www.youtube.com/watch?v=fFpW5W06kV4
- **Дата:** 04.12.2025
- **Длительность:** 11:08
- **Просмотры:** 85,771
- **Источник:** https://ekstraktznaniy.ru/video/15130

## Описание

AD | For 48 hours, enjoy 15% OFF on all Hoverpens with code DrBecky, or click on the link https://noviumdesign.shop/DrBecky with free shipping to most countries. Also on Amazon: https://noviumdesign.shop/YQjn0y | One way that us astrophysicists try to make sense of what we see out in the Universe, is to simulate what’s going on in a computer. These simulations are an astrophysicists lab experiment - we can’t poke and prod things, but we can change inputs and tweak equations to test ideas about how things like galaxies form, evolve, and interact. Simulations reveal how the complex processes of physics, like gravity, gas particle interactions, magnetic fields, and blasts from supernova—combine over millions or billions of years. But there is a big barrier to us understanding the universe through simulations and that is computing power. We can’t simulate EVERYTHING, it’s too much of a drain on resources. So we have to drop the resolution. For example, when we simulate our galaxy the Milky

## Транскрипт

### Introduction []

When we look out into the vastness of the universe, we see many things that we still don't understand. But one way that us astrophysicists try to make sense of what we're seeing is to simulate what's going on in a computer. From the relatively simple simulation of a single supernova when a star dies and implodes to the less simple formation of an entire solar system to the even more complex simulating an entire galaxy or even the entire universe. These simulations are an astrophysicist's lab experiment. We can't poke and prod things like other scientists, but we can change inputs and tweak equations to test our ideas about how things like galaxies form, evolve, and interact. Simulations reveal how all the complex processes of physics like gravity, gas particle interactions, magnetic fields, and blasts from supernova combine over millions or billions of years. But there is a big barrier to us understanding the universe through simulations and that is computing power. We can't simulate everything. It's just too much of a drain on resources. So we have to drop the resolution. For example, when we simulate our galaxy, the Milky Way, instead of having a 100 billion particles in a simulation, each representing a single star, a simulation might only have a billion particles with each particle representing a cluster of roughly a 100 stars. But then that means that the fine details of each individual stars life and death get lost. And the problem is that those smallcale events can then have ripple effects across the whole galaxy. So, if we want to be able to run a simulation of the Milky Way with a particle representing every single star and see how it changed over at least a billion years, that would take us 36 years to run that simulation with the current best supercomputers. But what if we could help the computers take a shortcut by using machine learning or AI? That's what this research paper from Hiroshima and collaborators that was published this past month has claimed to have done. Make the first starbystar simulation of the Milky Way, which doesn't take 36 years to run, but instead just 115 days. So, in this video, I'm going to explain how they did that. But before I

### Novium AD [2:24]

do, Christmas is coming up very soon. And maybe you're struggling to think of the perfect gift for someone and might have spotted this beauty on the desk or shelves behind me during my videos. This is a Novium hover pen inspired by spacecrafts and engineered to defy gravity. They float on their docks using permanent magnets. There's no electricity needed. And they're refillable. Just look how cool the HoverPen Interstellar looks. It floats in its dock at a 23. 5° tilt, a nice nod to the Earth's axial tilt, which I love. Plus, it comes in either space black, starite silver, Mars magma, or Neptune blue. Or the premium edition also has a real meteorite shard embedded in it, which is older than the Earth itself, which just makes me want it even more. The Interstellar pen even received the time award for one of the best inventions of the year. But maybe like me, you occasionally want to be fancy and write with a fountain pen or a roller ball, in which case you should get the hover pen future with its interchangeable tip. They feel so premium and high-end. And honestly, I think it's the perfect Christmas gift for the space nerd in your life. Who knows, maybe that is you. If you use the link in the description or scan the QR code on screen or use the code Dr. Becky at checkout, you'll get 15% off all their hover pens. An offer which is only available for the next 48 hours. And don't worry, there's free international shipping available to most countries. So, thanks again to Novian for sponsoring this video. And now, let's get back to this breakthrough from this

### Why do simulations take so long to run? [3:54]

paper by Hiroshima and collaborators and chat first about why our current simulations of the Milky Way with 100 billion star particles take so long to run. Yes. Okay, there's the problem of the sheer number of stars involved there, but don't forget, you've also got to add in some particles for the gas in your galaxy, plus the dark matter as well. And the main force affecting all of those will be gravity. With every particle, whether star, dark matter, or a gas particle, pulling on every other particle. So the more particles you add to your simulation, the number of interactions rises incredibly fast. So much so that if you double the number of stars, the number of gravitational calculations doesn't just double, it squares. But it's not just gravity you're simulating. You also have to keep track of things like magnetic fields and the temperature and energy exchange between all of those particles with the cooling and heating going on. Plus, you have to keep track of the elements making up all the stars and the gas. So whether it's hydrogen or oxygen or nitrogen, then you've got huge differences in scales involved. Yes, distance scales is the obvious one. So a star and a planet system are tiny compared to the galaxy as a whole, right? a fraction of a lightyear compared to 100,000 light years across for the galaxy, but also 600,000 lighty dark matter halo that the galaxy is embedded in. But you've also got temperature scales. So a supernova can reach 10 million degrees, whereas interstellar gas clouds are down at 10° above absolute zero. And the biggest kicker of all is the time scales involved. Because a supernova going off and then dissipating all of that material and energy outwards happens on time scales of a few minutes to a few years. Whereas one full rotation of the entire galaxy takes a few hundred million years. So, previous simulations of galaxies have just not had the spatial resolution to cover the very small distance scales involved or the time resolution to cover the drastic changes when a supernova happens. And it's this issue with supernovi that keeps coming to the forefront of this problem. Because even though it is just one star, right, relatively tiny on a galactic scale, a supernova releases an enormous amount of energy in a very small region in just a few seconds, which then has ripple effects outwards. That blast can heat gas, trigger new star formation, or even blow material out of a galaxy. So if a simulation smooths over this event either in space or in time, it risks missing important ripple effects on the whole galaxy. Now one way of solving this is to use what's known as sub grid models where the resolution gets finer closer and closer to the supernova and stays coarse across the whole galaxy. But that still leaves the time resolution issue, i. e. the number of years between each step in your simulation. Because processing smaller time steps means recording what all the particles in your simulation are doing and what their properties are more often, which takes more time and more

### How does AI help? [7:00]

computational resources. And so you might think, well, let's just throw everything at this problem, then get all the supercomputers on it working together. But that won't speed things up because your efficiency decreases because all the computers have to talk to each other. This communication actually becomes slower than the calculations needed for each time step in your simulation. This is why the current state-of-the-art simulations would take 36 years to run an accurate simulation of the Milky Way with 100 billion star particles. And of course, as much as we can dream that one day quantum computers might speed up some of these calculations needed for a simulation like this, that is a long way off. So, a lot of astrophysicists and computer scientists have been working on what can we do in the meantime, which is where this paper from Hiroshima and collaborators comes in. And their idea was not to model individual supernova directly at all, but instead use the huge amount of detailed simulations of individual supernova that have already been run in the past and train a deep learning algorithm called a neural network on those individual supernova simulations in order to then predict the properties of any given supernova in their much larger galaxy simulation. Essentially, they kept a section of the computer to one side outside of the main simulation, and its job was to identify when there's a star going supernova in the galaxy, feed in its current properties of that star, and then use the AI algorithm to predict how much energy and heavy elements that supernova will throw out and in which directions and what impact that will have on the surrounding particles in the galaxy simulation near the supernova. And critically, it did this over many much shorter time steps before feeding that information back to the main simulation only after one much longer time step of the main simulation had run. That way you keep track of all the details on the small scale but only record the impact on the large scale. And it's this use of AI in this way that allows you to do this so quickly and efficiently because you can run those calculations in parallel to the main simulation. Parallel means you can throw more supercomputers at this and you don't need them to communicate. It's all of

### The future [9:24]

this combined that gives you that factor of 113 times speed up compared to the current state-of-the-art simulations. That is an incredible number. And it's not just astrophysics that will benefit from this breakthrough. Any simulation that has smallcale events that have ripple effects on large scales, especially when large time scales are involved, will benefit from this technique. Think climate modeling, small scale changes yearon-year to the weather versus long-term changes to the climate as a whole. So, I'm sure like me, you're excited about this technique and you're now thinking, okay, show me the simulation. Was it any good? Does it work using this predictive neural network instead of modeling the supernova directly? Well, we don't have those answers just yet. This paper from Hiroshima and collaborators focused on the computer science and not the astrophysics. This paper's focus was on the time that it would take this simulation to run just as a proof of concept and not the actual output of the simulation just yet. So that will be the next step here to verify the results that you get from this simulation and ensure that the use of this predictive AI model doesn't bias you or give you a different answer compared to actually modeling the supernova directly. But given the fact that the simulation like that would only take 100 days or so to run, I hope we won't be waiting around too long for those results. Make sure you subscribe to my channel so you don't miss out when those get published. And don't forget that link in the description to get yourself a Novium hover pen. An amazing Christmas gift to yourself, but also the science enthusiast in your life.