# They Said It Was Impossible… This Simulation Solved It

## Метаданные

- **Канал:** Two Minute Papers
- **YouTube:** https://www.youtube.com/watch?v=9Mcv9vpGW5Q
- **Дата:** 25.01.2026
- **Длительность:** 14:13
- **Просмотры:** 85,689

## Описание

❤️ Check out Lambda here and sign up for their GPU Cloud: https://lambda.ai/papers

📝 The paper is available here:
https://visualcomputing.ist.ac.at/publications/2025/HomogenizedSand/

Previous Disney grains paper:
https://la.disneyresearch.com/publication/multi-scale-modeling-and-rendering-of-granular-materials/

Our Patreon if you wish to support us: https://www.patreon.com/TwoMinutePapers

🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
Adam Bridges, Benji Rabhan, B Shang, Cameron Navor, Christian Ahlin, Eric T, Fred R, Gordon Child, Juan Benet, Michael Tedder, Owen Skarpness, Richard Sundvall, Ryan Stankye, Steef, Taras Bobrovytsky, Tazaur Sagenclaw, Tybie Fitzhugh, Ueli Gallizzi
 
My research: https://cg.tuwien.ac.at/~zsolnai/

## Содержание

### [0:00](https://www.youtube.com/watch?v=9Mcv9vpGW5Q) Segment 1 (00:00 - 05:00)

I got a real treat for you fellow scholars. You know that this video is longer than usual and it is because this is one of the best research papers I've seen in a while and most definitely the hardest. No question about that. You have to see this because it makes the impossible possible. And what the heck is this? Is this a nasty bug in the Matrix? Nope, not at all. This is perfect. What? How can that be? Okay, let me try to explain. I try my best. Especially because this paper is kicking my rear end. I got to tell you that. I'm pretty sure I have smoke coming out of my ears by now. But I think I got it. I hope. Sort of. I think around the time when I worked at the Disney Research Lab in Switzerland, an insane work came out. A castle made of sand. So, what's so crazy about this? Well, check this out. Whoa. Oh, yes. This is made out of tiny grains. And the computer runs a light simulation program that computes what this would really look like in reality up [clears throat] to billions of little grains. The algorithm behind it is kind of insane, and no AI was used whatsoever here. Now, this was 10 years ago, and ever since then, I was thinking, what if we could even simulate how they move? That would be truly insane. But it won't happen. Why? Well, here is a bunch of stuff. Let's run a rigid body simulation. Love it. Realistic, yes, but it only works for smaller object counts. You can't do this with billions of grains. That's crazy. But in this paper they say they can do the impossible. They don't promise billions of particles but they do something that really stunned me. And this work is from Professor Chris Witton's lab. A true master. If you see his name, you know it's going to be good. Love him. Okay, I am ready. Let's go. What does it say? Our plastic flow handles Draer Prager and more cool. Okay, so what the heck does that mean? Well, the Draker Prager model means smooth and simple sand. We simplify the math by imagining perfect smooth objects. Essentially assuming the sand behaves like a pile of slippery marbles. Now, more cool means a model that acknowledges that sand is jagged. It knows how real grains lock together and have specific weak spots where they prefer to slip. Now let's look side by side. They look very similar. Of course, these are good, commonly used models for this kind of thing. But here comes the problem. Check this. Uh-oh. They used to be the same, but this time differentto. Why are they differentto? This is physics. Shouldn't they be the same? Well, in the first example, we had low internal friction. But when we start increasing parameters like friction, well, different models are good at different things and are bad at others. So this is why I say we'll never get quick and accurate granular simulations with traditional techniques. But now here is the new technique simulating a collapsing cylinder from many circular grains. Okay, but now check this out. Oh my. Now we have a bunch of grains built from these hexopods. Oo. Now this one collapses differently. Why? Well, because these star-shaped grains hook into each other and they clamp together. So if you're in a car race and you want to throw a bunch of catrops to sabotage the next guy, this is what you should be doing. Very important lesson. And you can even create a kind of object that interlocks with itself even better. So much so that it barely falls apart. We'll build a crazy scene with these in a moment. Before that, if you put them in an hourglass, you get this. Oh, a big clump refusing to move. But if you use the regular circular grains as a reference, let's have a look. Oh my, look at how beautifully and obediently everyone goes through. Now, this other kind of object can still go through, but with a bit more resistance. And it clumps like caramel on top of your ice cream. Wow. So good. Now, time for something crazier. A sand bridge. Let's start with the spherical grains. I love the appearance of this support. And when we

### [5:00](https://www.youtube.com/watch?v=9Mcv9vpGW5Q&t=300s) Segment 2 (05:00 - 10:00)

remove it, we get classic dry sand behavior. The grains are round, so they can't get tangled up with each other. So the bridge collapses immediately. Then below we get a flat pile because the only thing holding the grains is simple friction which isn't enough to fight gravity. Next up these funny door handles that they call the loss. These are like those huge concrete weights that are used to protect seaw walls. But for me door handle it is. So they interlock a bit but not that much. Look, the ridge collapses, but notice how it forms a much steeper heap at the bottom compared to the spheres. Now, let's break out the hexopods, catrops. And this is where things start to get sticky. Love it. They have these six little arms. And even without any glue or water to hold them together, what happens? Well, it crumbles into more chunky cohesive lumps rather than flowing like a fluid. It even has these two cat ears. Love it. So cool. Now check this out. The Deca fangs. Let's see what these 12 fang-like hooks can do. What pile will they form? And oh my what? No pile. What happened? Why? Well, fellow scholars, hold on to your papers because this is where the magic happens. This technique can simulate that these little hooks lock together so tightly that the material suddenly stops acting like sand. It now acts more like a solid elastic body. It even jiggles. The individual grains are effectively trapped turning a pile of particles into a sturdy structure. How sturdy? H. You are wise fellow scholars. You know already that. We'll see about that in a moment. Now onward to the ultimate test, siege warfare. We fire a virtual sandball at these castles to see who survives. And then I'll try to explain how it works. I hope because this is really advanced stuff. I'll try my best. Starting with the spheres. Utter destruction. Since there is zero cohesion to hold the grains together, the castle collapses instantly. It creates a nice little splash, but as a fortress, zero out of 10. Get out of here. Wait, want to see a close-up? Look at that. I get tingly just watching this. Love it. Next, the door handles. They put up a slightly better fight because of that higher friction we saw earlier, but ultimately the impact is just too much. But I like the lump a bit better. One out of 10. Now, close-up time. Oh my, look. You see each and every single one of them. Goodness. Okay, Calrops. So, what happens to the castle? Well, exactly what happens to some YouTube channels after being sold to private equity companies. And I can't wait for the close-up. Now, hold on to your papers, fellow scholars, for these guys. Wow. So good. Okay. And now we have the high-speed projectile slammed into the castle and it bounces. What? Huh? It didn't just survive. It absorbed the energy like a block of rubber. This confirms that our pile of loose grains has effectively transformed into a solid elastic object. It just jiggles for a while and goes back to sleep. That is why it jiggles. Absolutely incredible. And now the closeup. This is god tier material. Mother of papers. Look at that. So, how on earth did they simulate so many of these complex shapes without melting a data center into a puddle? Well, they didn't. They used a clever cheat code called numerical homogenization. Let me fire up this nice little experiment in the meantime. So instead of simulating every single grain in that castle, which would be impossible, they took a tiny box. They packed it with just a few thousand grains and squashed it thousands of times to measure exactly how it resists force, beat it up a little until it gives up its mathematical secrets. Then they treat this box like a 3D wallpaper that repeats endlessly in all directions. This is the key. So once they figured out the math of how these hooks tangle, they taught those rules to a larger, faster simulation. This allows them to render these massive scenes in a fraction of the time. This is incredible. And this is why you need all

### [10:00](https://www.youtube.com/watch?v=9Mcv9vpGW5Q&t=600s) Segment 3 (10:00 - 14:00)

this crazy mathematics. Okay, let me explain and you'll see that it will be easy to understand. Now, promise that you won't close the video. Come on, promise. Okay, so why the heck do you need all of these crazy integrals with tensor products within them? This is a math house. Dear fellow scholars, this is two minute papers with Dr. Koa Eher. Well, these integrals compute the homogenized Koshi stress tensor. In simple words, they calculate the average pressure for the entire box of sand. But they don't measure the force inside every hard grain, which is impossible. No, no. Here they calculate forces pushing on just the outer walls of the box. And that is genius. Why? Well, imagine a bunch of rockers inside a mosh pit. To figure out the pressure of the crowd, you don't ask every single person how squished they feel. Well, you can, but I ain't going there, brother. Instead, you just measure how hard the crowd is pushing against the walls of the room. And the tensor product here is just the math tool that translates these complex wall pushes into a single room pressure score. Yeah, a big bunch of stuff collapses down into just one number. Yes, these huge formulas actually make things simpler. Wow. And just imagine that we can do all of this with a computer and nothing but human ingenuity. No AI needed here. But it's not perfect. I'll tell you about a weak point in a moment. Now, this is absolutely brilliant, genius work that almost no one talks about in today's media. Works like this don't get the spotlight. So, who is going to stand up for these amazing works? These are the works that push humanity forward. I mean, come on. Solving the physics of millions of grains by crushing a tiny virtual box. Incredible. We try to save these hidden gems from obscurity. Come help us. Make sure to subscribe, hit the bell, and leave a kind comment. I love to read your comments. Now, a weak point. To measure the physics of just those pink hexopods, the computer had to crunch numbers for 75 hours. That is almost a month of calculation just to learn the rules for one grain shape. Also, this method assumes that grains are hard as rock, so no squishy jelly beans are allowed. However, it just made the impossible possible. This is the most important step. Optimization can come after. Remember the first law of papers which says that research is a process. Do not look at where we are. will be two more papers down the line. So, what did we learn today? Well, simple. If you see a paper about rendering or simulating grains, for the love of God, run or watch two minute papers. Okay, kidding. So, what we learned is that you don't need to simulate a billion collisions to build a sand castle. You just need to squash a tiny box of grains in a virtual hydraulic press until they give up their mathematical secrets. What a time to be alive. And please note that folks at Lambda are now our [clears throat] main supporters. So please give them some love. Click and sign up in the video description through our link so we can do more work like this. I'd love to. Here you see me running the full Deepseek AI model through Lambda GPU Cloud. 671 billion parameters running super fast and super reliably. This is insane. I love it. and I use it on a regular basis. Lambda provides you with powerful NVIDIA GPUs to run your own chatbots and experiments. Seriously, try it out now at lambda. ai/papers AI/papers or click the link in the description.

---
*Источник: https://ekstraktznaniy.ru/video/11404*