WEBVTT 00:00:00.000 --> 00:00:02.540 Welcome to today's deep dive. If you're joining 00:00:02.540 --> 00:00:05.580 us, you are likely the kind of person who doesn't 00:00:05.580 --> 00:00:08.820 just want to know what happened, but exactly 00:00:08.820 --> 00:00:11.039 how it happened, whether you're using this time 00:00:11.039 --> 00:00:13.480 to catch up on the cutting edge of physical cosmology 00:00:13.480 --> 00:00:16.949 or. prepping your brain for complex problem solving, 00:00:17.449 --> 00:00:20.129 or you just have this insatiable curiosity about 00:00:20.129 --> 00:00:22.570 our place in the cosmos, you are exactly where 00:00:22.570 --> 00:00:24.969 you need to be. Absolutely. Today, our mission 00:00:24.969 --> 00:00:28.109 is to understand how humanity maps the completely 00:00:28.109 --> 00:00:31.710 unmappable. We are doing a deep dive into simulating 00:00:31.710 --> 00:00:34.149 the universe, specifically looking inside the 00:00:34.149 --> 00:00:36.609 illustrious project and the supercomputer's mapping 00:00:36.609 --> 00:00:39.429 cosmic evolution. It's a massive topic. It really 00:00:39.429 --> 00:00:42.590 is. We're going to unpack dark matter, dark energy, 00:00:43.369 --> 00:00:46.439 galaxy formation and try to wrap our heads around 00:00:46.439 --> 00:00:48.880 how scientists have managed to simulate the entire 00:00:48.880 --> 00:00:52.259 history of the universe all 13 .8 billion years 00:00:52.259 --> 00:00:55.539 of it inside a machine. And we are using the 00:00:55.539 --> 00:00:57.679 detailed source material on the illustrious project 00:00:57.679 --> 00:01:01.020 as our roadmap today. It is truly a monumental 00:01:01.020 --> 00:01:03.640 achievement in astrophysics. Over the course 00:01:03.640 --> 00:01:05.280 of this deep dive, we're going to break down 00:01:05.280 --> 00:01:07.980 the timeline, the staggering computing power 00:01:07.980 --> 00:01:10.939 required to pull this off, the underlying physics 00:01:10.939 --> 00:01:14.040 encoded into the software, and the highly specialized 00:01:14.040 --> 00:01:16.739 spin -off projects that this original simulation 00:01:16.739 --> 00:01:19.579 spawned. Yeah, projects like Illustris TNG and 00:01:19.579 --> 00:01:22.000 Auriga, which are just fascinating. Exactly. 00:01:22.430 --> 00:01:24.790 What makes this so compelling is that it represents 00:01:24.790 --> 00:01:27.409 a massive paradigm shift in how we do science. 00:01:28.049 --> 00:01:30.750 I mean, for centuries, astronomy was purely observational. 00:01:30.870 --> 00:01:33.010 You just look up. Right, you build a bigger telescope, 00:01:33.170 --> 00:01:34.709 you look up, and you try to deduce the rules 00:01:34.709 --> 00:01:37.150 of the universe based on what you see. But the 00:01:37.150 --> 00:01:39.459 illustrious project... flips that entirely. It 00:01:39.459 --> 00:01:42.099 is experimental computational astronomy. OK, 00:01:42.120 --> 00:01:44.739 let's unpack this methodology. Imagine trying 00:01:44.739 --> 00:01:47.879 to bake a cake, right? But the recipe takes 13 00:01:47.879 --> 00:01:51.379 .8 billion years. It requires 25 terabytes of 00:01:51.379 --> 00:01:53.959 RAM. And the ingredients are things like dark 00:01:53.959 --> 00:01:56.700 matter and supermassive black holes. That is 00:01:56.700 --> 00:01:59.519 a very stressful baking show. Right. But instead 00:01:59.519 --> 00:02:02.000 of looking up at the finished product, the finished 00:02:02.000 --> 00:02:04.560 cake, you start with the raw ingredients and 00:02:04.560 --> 00:02:07.239 the laws of physics. put them in a digital box, 00:02:07.560 --> 00:02:09.539 and see if you can bake a universe that actually 00:02:09.539 --> 00:02:12.620 matches what we see through the telescopes. Exactly. 00:02:13.259 --> 00:02:15.360 And the underlying goal isn't just to make a 00:02:15.360 --> 00:02:18.120 visually stunning digital planetarium. This is 00:02:18.120 --> 00:02:21.319 about establishing a rigorous predictive theory 00:02:21.319 --> 00:02:25.400 of galaxy formation. By building a comprehensive 00:02:25.400 --> 00:02:28.680 physical model and running it forward, researchers 00:02:28.680 --> 00:02:30.780 are testing our fundamental understanding of 00:02:30.780 --> 00:02:32.659 reality. They want to see if the math actually 00:02:32.659 --> 00:02:35.060 works. Precisely. They want to know if the physics 00:02:35.060 --> 00:02:37.580 we think govern dark matter, dark energy, and 00:02:37.580 --> 00:02:40.080 standard matter are actually correct when applied 00:02:40.080 --> 00:02:42.479 on a cosmic time scale. Okay, but on the surface 00:02:42.479 --> 00:02:45.680 that almost sounds circular. Let me play Devil's 00:02:45.680 --> 00:02:47.939 Advocate for a second here. Go for it. If these 00:02:47.939 --> 00:02:50.479 scientists specifically key figures mentioned 00:02:50.479 --> 00:02:52.840 in our source, like Mark Vogelsberger at MIT 00:02:52.840 --> 00:02:55.500 and Volker Springel at the Max Planck Institute 00:02:55.500 --> 00:02:58.439 for Astrophysics, if they are writing the code 00:02:58.439 --> 00:03:00.740 and they are programming the simulation using 00:03:00.740 --> 00:03:03.400 the physics rules they already believe are true, 00:03:03.819 --> 00:03:05.840 aren't they just building a massive echo chamber? 00:03:05.979 --> 00:03:08.219 That's a really common critique. I mean, aren't 00:03:08.219 --> 00:03:11.159 they guaranteed to just get a universe that perfectly 00:03:11.159 --> 00:03:13.879 matches their preconceptions? That is the exact 00:03:13.879 --> 00:03:16.780 trap they had to avoid. And it's a brilliant 00:03:16.780 --> 00:03:20.000 question because it strikes at the heart of why 00:03:20.000 --> 00:03:23.180 this simulation is so incredibly difficult. The 00:03:23.180 --> 00:03:26.360 way they avoid the echo chamber is by strictly 00:03:26.360 --> 00:03:29.300 separating the micro rules from the macro results. 00:03:29.639 --> 00:03:31.840 Meaning they don't program a galaxy to look like 00:03:31.840 --> 00:03:34.139 a galaxy. Precisely. They do not write a single 00:03:34.139 --> 00:03:37.000 line of code that says make a spiral galaxy here 00:03:37.000 --> 00:03:39.580 or make an elliptical galaxy there. They only 00:03:39.580 --> 00:03:42.460 program the foundational microscopic rules of 00:03:42.460 --> 00:03:44.949 physics. Like gravity. Right. They code in the 00:03:44.949 --> 00:03:47.189 gravitational constant. They code the thermodynamic 00:03:47.189 --> 00:03:49.750 properties of gas. They code the rate at which 00:03:49.750 --> 00:03:53.189 gas cools and condenses to form stars. They drop 00:03:53.189 --> 00:03:56.310 in dark matter and dark energy based on the parameters 00:03:56.310 --> 00:03:58.389 measured by the cosmic microwave background. 00:03:58.490 --> 00:04:01.610 The echo of the Big Bang. Exactly. They set up 00:04:01.610 --> 00:04:04.199 these initial basic conditions as they existed 00:04:04.199 --> 00:04:07.180 right after the Big Bang, hit run, and let 13 00:04:07.180 --> 00:04:10.939 .8 billion years of virtual time unfold. So the 00:04:10.939 --> 00:04:13.860 macroscopic structures, the massive spiraling 00:04:13.860 --> 00:04:16.779 galaxies, the galaxy clusters, the cosmic web, 00:04:17.220 --> 00:04:19.819 all of that has to emerge entirely on its own, 00:04:20.079 --> 00:04:22.420 organically, just from the micro rules interacting 00:04:22.420 --> 00:04:25.180 over time. It's called emergent complexity. If 00:04:25.180 --> 00:04:27.459 they run the simulation and the resulting virtual 00:04:27.459 --> 00:04:29.459 universe is filled with galaxies that are way 00:04:29.459 --> 00:04:32.000 too massive or galaxies that don't cluster together 00:04:32.000 --> 00:04:34.360 the way real ones do, they don't tweak the galaxies. 00:04:34.540 --> 00:04:36.600 Because that would be cheating. Right. They realize 00:04:36.600 --> 00:04:39.040 that their understanding of the fundamental microphysics 00:04:39.040 --> 00:04:42.560 must be flawed. It serves as an unforgiving reality 00:04:42.560 --> 00:04:45.459 check. If the virtual universe doesn't mirror 00:04:45.459 --> 00:04:47.980 the observable universe, it means our textbook 00:04:47.980 --> 00:04:50.319 theories need adjusting. That makes perfect sense. 00:04:50.480 --> 00:04:52.959 The simulation is a stress test for our physics. 00:04:53.259 --> 00:04:55.660 Which brings us to the actual ingredients they 00:04:55.660 --> 00:04:59.100 put into this digital box. When you are simulating 00:04:59.100 --> 00:05:01.740 physical cosmology, you can't just throw in some 00:05:01.740 --> 00:05:04.220 gravity and a bunch of stars and call it a day. 00:05:04.319 --> 00:05:06.459 No, the universe is much messier than that. It 00:05:06.459 --> 00:05:09.379 really is. The simulation has to include the 00:05:09.379 --> 00:05:12.480 chaotic stuff. We're talking about star formation, 00:05:13.139 --> 00:05:15.740 supernova explosions, the formation of supermassive 00:05:15.740 --> 00:05:18.600 black holes, and of course, the invisible scaffolding 00:05:18.600 --> 00:05:21.319 of the universe, dark matter and dark energy. 00:05:21.639 --> 00:05:24.300 Out of all those ingredients, the one that often 00:05:24.300 --> 00:05:27.699 surprises people the most is the role of supermassive 00:05:27.699 --> 00:05:31.319 black holes in supernovae. If we want to understand 00:05:31.319 --> 00:05:34.339 how a galaxy forms, we have to understand a concept 00:05:34.339 --> 00:05:37.509 called feedback. Feedback. Because when we picture 00:05:37.509 --> 00:05:39.829 gravity, we usually just picture things falling 00:05:39.829 --> 00:05:42.490 inward. Gravity pulls matter together. So if 00:05:42.490 --> 00:05:44.209 a universe only had gravity, you'd eventually 00:05:44.209 --> 00:05:47.189 just get incredibly dense, dead clumps of matter. 00:05:47.389 --> 00:05:49.610 Everything would just collapse into itself. There 00:05:49.610 --> 00:05:51.970 has to be a push to counteract that pull. That 00:05:51.970 --> 00:05:55.209 push is the feedback. And it is arguably the 00:05:55.209 --> 00:05:57.509 most difficult thing to model correctly in a 00:05:57.509 --> 00:06:00.069 universe simulation. Let's look at supernovae 00:06:00.069 --> 00:06:03.250 first. When a massive star reaches the end of 00:06:03.250 --> 00:06:06.120 its life, It doesn't just quietly wink out. It 00:06:06.120 --> 00:06:08.420 explodes. It detonates in a supernova, one of 00:06:08.420 --> 00:06:11.040 those violent explosions in the universe. This 00:06:11.040 --> 00:06:14.000 explosion sends shockwaves out into the surrounding 00:06:14.000 --> 00:06:16.720 interstellar gas, heating it up and literally 00:06:16.720 --> 00:06:19.259 blowing it away. And since stars form from cold, 00:06:19.500 --> 00:06:22.399 dense gas, if a supernova blows all that gas 00:06:22.399 --> 00:06:24.459 away and heats it up, it's effectively shutting 00:06:24.459 --> 00:06:26.779 off the assembly line for new stars in that area. 00:06:27.000 --> 00:06:30.060 Exactly. It regulates star formation. Without 00:06:30.060 --> 00:06:32.620 supernova feedback, galaxies and simulations 00:06:32.620 --> 00:06:35.279 convert their gas into stars way too fast. They 00:06:35.279 --> 00:06:37.680 burn out. They do, resulting in virtual galaxies 00:06:37.680 --> 00:06:39.819 that are unnaturally bright and overly packed 00:06:39.819 --> 00:06:42.420 with stars compared to the real universe. So 00:06:42.420 --> 00:06:44.300 the simulation has to constantly track where 00:06:44.300 --> 00:06:46.459 massive stars are forming, calculate when they 00:06:46.459 --> 00:06:49.050 die. and model the subsequent thermal and kinetic 00:06:49.050 --> 00:06:51.170 energy injected back into the gap. And then you 00:06:51.170 --> 00:06:54.129 scale that up to the absolute extreme with supermassive 00:06:54.129 --> 00:06:56.509 black holes. We usually think of black holes 00:06:56.509 --> 00:06:59.209 as these ultimate cosmic vacuum cleaners that 00:06:59.209 --> 00:07:03.029 just eat everything. But the reality of a feeding 00:07:03.029 --> 00:07:05.589 black hole is profoundly explosive, isn't it? 00:07:05.829 --> 00:07:08.850 It is. When gas falls toward a supermassive black 00:07:08.850 --> 00:07:11.509 hole at the center of a galaxy, it forms an accretion 00:07:11.509 --> 00:07:15.089 disk. This material swirls inward at a significant 00:07:15.089 --> 00:07:17.120 fraction of the speed of light. The friction 00:07:17.120 --> 00:07:19.720 there must be insane. It's incomprehensible. 00:07:20.199 --> 00:07:22.319 The friction and magnetic forces within this 00:07:22.319 --> 00:07:25.079 disk generate massive amounts of heat and radiation. 00:07:25.680 --> 00:07:28.459 Furthermore, complex magnetic interactions can 00:07:28.459 --> 00:07:30.860 channel some of that infalling material into 00:07:30.860 --> 00:07:33.980 massive twin jets that blast outward into space. 00:07:34.220 --> 00:07:36.339 So while a black hole is consuming some matter, 00:07:36.600 --> 00:07:38.860 it's violently ejecting vast amounts of energy 00:07:38.860 --> 00:07:41.579 outward. The source calls this active galactic 00:07:41.579 --> 00:07:45.350 nucleus, or AGN, feedback. Correct. And AGN feedback 00:07:45.350 --> 00:07:47.689 is crucial for regulating the largest galaxies. 00:07:48.089 --> 00:07:50.250 Without supermassive black holes actively pushing 00:07:50.250 --> 00:07:52.069 back against the crushing weight of gravity, 00:07:52.449 --> 00:07:54.470 the largest galaxies in the universe would continue 00:07:54.470 --> 00:07:56.670 to grow out of control, becoming vastly larger 00:07:56.670 --> 00:07:58.610 than anything we actually observe. They would 00:07:58.610 --> 00:08:01.329 just eat everything around them. Right. The illustrious 00:08:01.329 --> 00:08:03.550 project had to encode the mathematics of this 00:08:03.550 --> 00:08:06.370 energetic feedback, the push and pull, the expanding 00:08:06.370 --> 00:08:09.449 and contracting to create the delicate, chaotic 00:08:09.449 --> 00:08:12.550 balance that allows beautiful, intricate structures 00:08:12.550 --> 00:08:15.550 to to form and stabilize over billions of years. 00:08:15.910 --> 00:08:18.750 It's basically a cosmic ecosystem. The predators, 00:08:18.990 --> 00:08:21.310 the black holes, and supernovae keep the population 00:08:21.310 --> 00:08:24.269 of stars in check so the galaxy as a whole can 00:08:24.269 --> 00:08:27.550 survive and evolve naturally. But simulating 00:08:27.550 --> 00:08:30.209 an ecosystem of that scale, tracking the gravity 00:08:30.209 --> 00:08:32.429 of dark matter, the expansion driven by dark 00:08:32.429 --> 00:08:36.230 energy, the fluid dynamics of gas, and the localized 00:08:36.230 --> 00:08:39.889 explosions of supernovae over 13 .8 billion years. 00:08:40.409 --> 00:08:42.509 That requires a machine that borders on science 00:08:42.509 --> 00:08:44.669 fiction. That requires computational brawn that 00:08:44.669 --> 00:08:46.370 most people have a hard time conceptualizing. 00:08:46.629 --> 00:08:48.970 You cannot do modern physical cosmology on a 00:08:48.970 --> 00:08:51.350 standard computer. Definitely not. The illustrious 00:08:51.350 --> 00:08:54.350 simulation relied on top -tier supercomputers. 00:08:54.750 --> 00:08:57.149 Specifically, the source mentions the Curie supercomputer 00:08:57.149 --> 00:09:00.629 located at the CEA in France and the SuperMUC 00:09:00.629 --> 00:09:02.649 supercomputer at the Leibniz Computing Center 00:09:02.649 --> 00:09:04.870 in Germany. Let's put some numbers to this because 00:09:04.870 --> 00:09:07.610 the scale is just phenomenal. To run the main 00:09:07.610 --> 00:09:11.750 Illustrious simulation, it took 19 million CPU 00:09:11.750 --> 00:09:16.490 hours. They used 8 ,192 CPU cores running in 00:09:16.490 --> 00:09:19.370 parallel. And the peak memory usage during this 00:09:19.370 --> 00:09:23.009 run hit 25 terabytes of RAM. That is a staggering 00:09:23.009 --> 00:09:24.809 amount of memory. Just to translate that for 00:09:24.809 --> 00:09:27.429 you listening. If you tried to run this exact 00:09:27.429 --> 00:09:30.750 simulation on a single high -end desktop processor, 00:09:31.190 --> 00:09:33.350 you would be waiting over 2 ,000 years for the 00:09:33.350 --> 00:09:36.470 progress bar to reach 100%. And that 2 ,000 -year 00:09:36.470 --> 00:09:39.470 figure assumes your machine doesn't crash, overheat, 00:09:39.549 --> 00:09:41.629 or experience a hardware failure, which is a 00:09:41.629 --> 00:09:43.889 major logistical hurdle when dealing with compute 00:09:43.889 --> 00:09:46.350 times of this magnitude. When you are running 00:09:46.350 --> 00:09:49.470 a simulation across 8 ,192 cores for months on 00:09:49.470 --> 00:09:51.730 end, the physical reality of the hardware becomes 00:09:51.730 --> 00:09:54.159 a factor. Right. Computers get hot. Exactly. 00:09:54.460 --> 00:09:56.440 Processors generate heat. They require massive 00:09:56.440 --> 00:09:58.539 cooling infrastructure. Individual nodes can 00:09:58.539 --> 00:10:00.639 fail. The software has to be robust enough to 00:10:00.639 --> 00:10:02.580 handle the sheer physical strain it puts on the 00:10:02.580 --> 00:10:04.559 data center. But the real magic trick is how 00:10:04.559 --> 00:10:08.879 you get 8 ,192 separate computer brains to perfectly 00:10:08.879 --> 00:10:11.200 collaborate on a single universe without stepping 00:10:11.200 --> 00:10:14.299 on each other's toes. I know in supercomputing, 00:10:14.500 --> 00:10:16.440 you can't just divide a task easily if everything 00:10:16.440 --> 00:10:19.259 is connected. If a supernova explodes in processor 00:10:19.259 --> 00:10:21.940 number 500, the shockwave is eventually going 00:10:21.940 --> 00:10:23.860 to hit the gas being calculated by processor 00:10:23.860 --> 00:10:26.039 number 600. They have to communicate instantly. 00:10:26.399 --> 00:10:28.740 This is where the MPI approach comes in. MPI 00:10:28.740 --> 00:10:32.440 stands for message passing interface. It is a 00:10:32.440 --> 00:10:34.879 standardized means of exchanging messages between 00:10:34.879 --> 00:10:37.139 multiple computers running a parallel program 00:10:37.139 --> 00:10:42.220 across distributed memory. Imagine 8 ,192 cartographers, 00:10:42.580 --> 00:10:44.879 all standing shoulder to shoulder, each tasked 00:10:44.879 --> 00:10:47.679 with drawing a small piece of a massive dynamic 00:10:47.679 --> 00:10:49.740 map. And the map is constantly changing. Right, 00:10:49.799 --> 00:10:52.259 so cartographer A needs to constantly tell cartographer 00:10:52.259 --> 00:10:54.940 B, hey, a massive hot gas just crossed the border 00:10:54.940 --> 00:10:57.820 from my section into yours. MPI is the highly 00:10:57.820 --> 00:10:59.919 optimized language and protocol they use to shout 00:10:59.919 --> 00:11:02.000 these updates to each other. millions of times 00:11:02.000 --> 00:11:04.379 a second. That sounds chaotic. It would be, without 00:11:04.379 --> 00:11:07.240 the interface. Minimizing the latency, the delay 00:11:07.240 --> 00:11:09.700 in that communication, is critical. If one core 00:11:09.700 --> 00:11:12.200 gets bogged down calculating a particularly complex 00:11:12.200 --> 00:11:16.299 galaxy collision, the other 8 ,191 cores might 00:11:16.299 --> 00:11:18.360 have to sit idle waiting for it to catch up, 00:11:18.580 --> 00:11:20.480 wasting precious compute time. So the software 00:11:20.480 --> 00:11:22.899 has to be incredibly smart about how it distributes 00:11:22.899 --> 00:11:26.120 the workload, ensuring no single processor becomes 00:11:26.120 --> 00:11:29.059 a bottleneck. Which brings us to the brains of 00:11:29.059 --> 00:11:31.500 the operation, the actual code running on those 00:11:31.500 --> 00:11:34.539 machines. The program used for Illustris was 00:11:34.539 --> 00:11:37.080 written by Volker Springel, and it's called a 00:11:37.080 --> 00:11:39.700 REPO. Springel had previously authored a widely 00:11:39.700 --> 00:11:43.340 used astrophysics code called Gay -Gidaro, but 00:11:43.340 --> 00:11:46.059 a REPO represented a monumental leap forward 00:11:46.059 --> 00:11:49.159 in how we simulate fluid dynamics in space. Before 00:11:49.159 --> 00:11:51.379 we get into the heavy math of Arepo, I just have 00:11:51.379 --> 00:11:53.580 to mention a fantastic piece of trivia from the 00:11:53.580 --> 00:11:56.519 source. The name Arepo isn't an acronym. It's 00:11:56.519 --> 00:11:58.320 actually derived from the Sator Square, which 00:11:58.320 --> 00:12:00.519 is an ancient Latin palindrome. Yes, the word 00:12:00.519 --> 00:12:03.279 square. It's a square containing five words, 00:12:03.860 --> 00:12:07.879 Sator, Arepo, Tenet, Opera, Rotus, that reads 00:12:07.879 --> 00:12:10.779 the same top to bottom, bottom to top, left to 00:12:10.779 --> 00:12:13.370 right, and right to left. It's been found in 00:12:13.370 --> 00:12:15.590 ruins all over the Roman Empire from Pompeii 00:12:15.590 --> 00:12:18.049 to Britain. I just love the juxtaposition there. 00:12:18.289 --> 00:12:21.269 You have the most advanced, bleeding -edge astrophysics 00:12:21.269 --> 00:12:24.250 code in human history, built to unravel the mysteries 00:12:24.250 --> 00:12:27.029 of dark matter, and it's named after a 2 ,000 00:12:27.029 --> 00:12:29.889 -year -old linguistic puzzle. It's a wonderful 00:12:29.889 --> 00:12:32.450 nod to the enduring human desire to find patterns 00:12:32.450 --> 00:12:34.690 and structure in the universe. And structure 00:12:34.690 --> 00:12:36.710 is exactly what the Arepo code is designed to 00:12:36.710 --> 00:12:39.639 handle. At its core, Arepo was built to solve 00:12:39.639 --> 00:12:42.580 the coupled equations of gravity and hydrodynamics. 00:12:43.000 --> 00:12:45.159 We talked briefly about gravity, but hydrodynamics 00:12:45.159 --> 00:12:47.700 is where this gets incredibly complex. When we 00:12:47.700 --> 00:12:49.620 look up at the night sky, we tend to think of 00:12:49.620 --> 00:12:52.000 space as a vacuum, entirely empty except for 00:12:52.000 --> 00:12:54.539 solid planets and burning stars. But that's not 00:12:54.539 --> 00:12:57.059 accurate, is it? Not at all. The space between 00:12:57.059 --> 00:12:59.100 stars, and especially the space between galaxies, 00:12:59.519 --> 00:13:02.100 is filled with the diffuse medium of gas, primarily 00:13:02.100 --> 00:13:05.039 hydrogen and helium. Over cosmic time scales 00:13:05.039 --> 00:13:07.720 and vast distances, this gas doesn't act like 00:13:07.720 --> 00:13:10.279 isolated particles. It acts like a fluid. Like 00:13:10.279 --> 00:13:13.120 an ocean. Very much like that. It flows in massive 00:13:13.120 --> 00:13:16.159 cosmic winds. It pools in gravitational wells. 00:13:16.460 --> 00:13:19.779 It creates eddies and shockwaves. Hydrodynamics 00:13:19.779 --> 00:13:22.759 is the study of fluids in motion. To simulate 00:13:22.759 --> 00:13:25.639 a universe, you must accurately simulate fluid 00:13:25.639 --> 00:13:28.919 dynamics on a galactic scale. And simulating 00:13:28.919 --> 00:13:31.580 fluids is notoriously difficult. Anyone who has 00:13:31.580 --> 00:13:33.500 ever watched an older video game struggle to 00:13:33.500 --> 00:13:35.840 make water look realistic knows that calculating 00:13:35.840 --> 00:13:38.639 how a liquid or gas moves and interacts with 00:13:38.639 --> 00:13:42.120 itself takes massive processing power. Now imagine 00:13:42.120 --> 00:13:44.519 simulating a fluid that spans billions of light 00:13:44.519 --> 00:13:46.720 years, where its density can vary by factors 00:13:46.720 --> 00:13:48.860 of millions depending on whether it's in the 00:13:48.889 --> 00:13:51.450 deep void of space or inside a collapsing star 00:13:51.450 --> 00:13:54.230 -forming cloud. So how does a repo do it? The 00:13:54.230 --> 00:13:56.230 text mentions a highly specific mathematical 00:13:56.230 --> 00:13:59.070 technique called a moving Voronor tessellation. 00:13:59.389 --> 00:14:01.350 This sounds intimidating, but it is actually 00:14:01.350 --> 00:14:03.649 a beautiful concept once you visualize it. Let's 00:14:03.649 --> 00:14:05.830 break it down, because this is really the secret 00:14:05.830 --> 00:14:08.809 sauce of the illustrious simulation. Historically, 00:14:09.029 --> 00:14:11.389 cosmologists use two different methods to simulate 00:14:11.389 --> 00:14:14.850 fluids. The first is an Eulerian approach, which 00:14:14.850 --> 00:14:18.409 involves throwing a fixed rigid 3D grid over 00:14:18.409 --> 00:14:22.049 the universe. Imagine a giant cosmic jungle gym. 00:14:22.649 --> 00:14:24.950 You calculate how much gas moves in and out of 00:14:24.950 --> 00:14:27.330 each square box in the grid. The problem with 00:14:27.330 --> 00:14:30.639 a fixed grid, I imagine, is resolution. The universe 00:14:30.639 --> 00:14:33.019 is mostly empty space. If your grid boxes are 00:14:33.019 --> 00:14:35.559 too big, you can't see the fine details of a 00:14:35.559 --> 00:14:37.960 galaxy forming. But if you make the boxes small 00:14:37.960 --> 00:14:40.539 enough to see a galaxy, you suddenly have trillions 00:14:40.539 --> 00:14:42.879 of empty boxes in the cosmic voids that your 00:14:42.879 --> 00:14:44.980 computer is wasting time calculating for no reason. 00:14:45.240 --> 00:14:47.379 Exactly. It's incredibly inefficient. You are 00:14:47.379 --> 00:14:50.659 burning CPU hours calculating nothing. The alternative 00:14:50.659 --> 00:14:54.039 method historically used was Lagrangian, specifically 00:14:54.039 --> 00:14:57.980 Smoothed Particle Hydrodynamics, or SPH. In this 00:14:57.980 --> 00:15:00.740 method, there is no grid. Instead, you represent 00:15:00.740 --> 00:15:03.120 the gas as millions of individual virtual particles, 00:15:03.500 --> 00:15:05.759 and you track where they move. This is great 00:15:05.759 --> 00:15:07.620 because your computing power naturally follows 00:15:07.620 --> 00:15:09.639 the matter. But I'm guessing SPH has its own 00:15:09.639 --> 00:15:13.019 drawbacks. It does. It struggles with fluid instabilities 00:15:13.019 --> 00:15:16.320 and sharp shock waves. When two gas clouds smash 00:15:16.320 --> 00:15:20.720 into each other, SPH sometimes artificially smooths 00:15:20.720 --> 00:15:23.659 out the collision, losing the chaotic turbulent 00:15:23.659 --> 00:15:26.259 details that are essential for mixing the gas. 00:15:26.460 --> 00:15:29.000 So, a fixed grid is bad at following the matter, 00:15:29.340 --> 00:15:31.539 and floating particles are bad at capturing sharp 00:15:31.539 --> 00:15:34.379 fluid dynamics. Volcker's spring oil needed a 00:15:34.379 --> 00:15:38.100 third way. And that is the moving Voronoi tessellation. 00:15:38.299 --> 00:15:40.720 Right. A Voronoi diagram is a way of dividing 00:15:40.720 --> 00:15:43.779 space into distinct regions based on distance 00:15:43.779 --> 00:15:46.620 to a specific set of points. The classic example 00:15:46.620 --> 00:15:49.059 is mapping cell phone towers. Okay, I can picture 00:15:49.059 --> 00:15:51.320 that. If you put 100 cell towers on a map, a 00:15:51.320 --> 00:15:53.360 Voronoi diagram draws boundaries around each 00:15:53.360 --> 00:15:56.080 tower, creating a polygon. Everywhere inside 00:15:56.080 --> 00:15:58.299 a specific polygon is closer to that tower than 00:15:58.299 --> 00:16:00.620 to any other tower. Okay, so take that flat map 00:16:00.620 --> 00:16:02.639 of cell towers, make it three -dimensional, and 00:16:02.639 --> 00:16:05.470 apply it to the universe. Exactly. A repo casts 00:16:05.470 --> 00:16:08.049 a brilliant mathematical net of these 3D Voronoi 00:16:08.049 --> 00:16:10.889 cells over the entire virtual universe. But here's 00:16:10.889 --> 00:16:13.559 the genius part. The net is not rigid. The points 00:16:13.559 --> 00:16:15.279 that define the cells are allowed to move with 00:16:15.279 --> 00:16:18.080 the flow of the cosmic gas. Oh, wow. So as gravity 00:16:18.080 --> 00:16:20.960 pulls gas together to form a galaxy, the cells 00:16:20.960 --> 00:16:23.559 of this mathematical net flow inward alongside 00:16:23.559 --> 00:16:26.419 the gas, shrinking and clustering tightly together 00:16:26.419 --> 00:16:28.620 where the density is highest. Exactly. It's an 00:16:28.620 --> 00:16:31.700 adaptive dynamic mesh. In the empty voids of 00:16:31.700 --> 00:16:34.279 space, the cells stretch out and become enormous, 00:16:34.639 --> 00:16:37.799 saving computing power. But in the dense, turbulent 00:16:37.799 --> 00:16:40.279 center of a forming galaxy, the cells become 00:16:40.279 --> 00:16:43.100 incredibly tiny and numerous, providing ultra 00:16:43.100 --> 00:16:44.879 -high resolution right where the physics are 00:16:44.879 --> 00:16:47.259 most complex. That is so smart. And because it's 00:16:47.259 --> 00:16:49.480 still technically a mesh, it handles fluid dynamics, 00:16:50.019 --> 00:16:52.240 turbulence, and shock waves beautifully. The 00:16:52.240 --> 00:16:54.720 moving Voronoi tessellation gave Arbo the best 00:16:54.720 --> 00:16:57.039 of both historical methods without the fatal 00:16:57.039 --> 00:16:59.840 drawbacks of either. It is an elegant, dynamic 00:16:59.840 --> 00:17:02.220 solution to mapping the unmappable. It allows 00:17:02.220 --> 00:17:04.539 the simulation to breathe, contracting its focus 00:17:04.539 --> 00:17:06.440 where the action is, and expanding where space 00:17:06.440 --> 00:17:09.680 is empty. They took this code, fired up the supercomputers, 00:17:09.779 --> 00:17:12.880 and let it run for 19 million CPU hours. They 00:17:12.880 --> 00:17:16.220 built a 13 .8 billion year universe. And then 00:17:16.220 --> 00:17:18.319 they did something that I think defines the best 00:17:18.319 --> 00:17:20.460 of modern science. They gave the universe away. 00:17:20.759 --> 00:17:23.000 The commitment to open science here is remarkable. 00:17:23.210 --> 00:17:27.109 In April 2015, just 11 months after the foundational 00:17:27.109 --> 00:17:29.309 papers describing the illustrious results were 00:17:29.309 --> 00:17:32.390 published, the project team publicly released 00:17:32.390 --> 00:17:34.230 the data products from all their simulations. 00:17:34.529 --> 00:17:36.670 They didn't lock it behind a paywall. No, and 00:17:36.670 --> 00:17:38.670 they didn't hoard it for their own institution's 00:17:38.670 --> 00:17:42.109 exclusive use. We need to quantify what giving 00:17:42.109 --> 00:17:44.950 away a universe looks like. We are talking about 00:17:44.950 --> 00:17:48.890 over 230 terabytes of cumulative data volume. 00:17:49.440 --> 00:17:51.319 During the simulation run, they couldn't just 00:17:51.319 --> 00:17:53.380 save a single video file of the whole thing. 00:17:54.000 --> 00:17:56.220 Instead, they took full snapshot particle data 00:17:56.220 --> 00:17:59.519 at 135 distinct time points throughout the 13 00:17:59.519 --> 00:18:02.819 .8 billion years, plus one final snapshot, making 00:18:02.819 --> 00:18:06.000 a total of 136 snapshots. Think of it like a 00:18:06.000 --> 00:18:08.500 time lapse of the cosmos. These snapshots contain 00:18:08.500 --> 00:18:11.619 the precise location, velocity, mass, and temperature 00:18:11.619 --> 00:18:14.099 of billions of resolution elements at that exact 00:18:14.099 --> 00:18:16.950 moment in cosmic history. But raw data of billions 00:18:16.950 --> 00:18:19.390 of particles isn't incredibly useful on its own. 00:18:19.410 --> 00:18:22.049 It's just a wall of numbers. The team went a 00:18:22.049 --> 00:18:24.609 step further and processed this data into usable 00:18:24.609 --> 00:18:27.690 tools for the wider astronomical community. They 00:18:27.690 --> 00:18:30.049 included things called group catalogs of individual 00:18:30.049 --> 00:18:32.869 halos and sub -halos. For you listening who might 00:18:32.869 --> 00:18:35.990 not know, what exactly is a halo in this context? 00:18:36.210 --> 00:18:38.990 In cosmology, a halo refers to a massive clump 00:18:38.990 --> 00:18:41.289 of dark matter. Dark matter doesn't interact 00:18:41.289 --> 00:18:44.049 with light, but it has mass, so it exerts gravity. 00:18:44.740 --> 00:18:47.319 In the early universe, dark matter clumped together 00:18:47.319 --> 00:18:50.359 into these vast, invisible halos. The gravity 00:18:50.359 --> 00:18:52.180 of these dark matter halos is what pulled in 00:18:52.180 --> 00:18:54.380 the normal gas and dust, eventually sparking 00:18:54.380 --> 00:18:56.480 star formation. So the dark matter halo is the 00:18:56.480 --> 00:18:58.579 invisible bowl and the galaxy is the soup that 00:18:58.579 --> 00:19:00.799 collects inside it. That's a great way to visualize 00:19:00.799 --> 00:19:03.579 it. So a halo is the main dark matter structure, 00:19:03.980 --> 00:19:06.240 and a sub -halo is a smaller clump of dark matter 00:19:06.240 --> 00:19:08.079 residing within it, which usually corresponds 00:19:08.079 --> 00:19:11.380 to a single galaxy or a satellite galaxy. By 00:19:11.380 --> 00:19:13.480 releasing these group catalogs, the illustrious 00:19:13.480 --> 00:19:16.599 team basically provided a full census of every 00:19:16.599 --> 00:19:19.160 single dark matter bowl and every single galaxy 00:19:19.160 --> 00:19:22.160 soup in their virtual universe, categorized by 00:19:22.160 --> 00:19:24.940 mass, size, and star formation rate. And they 00:19:24.940 --> 00:19:27.519 also included merger trees. I love this concept. 00:19:27.900 --> 00:19:29.500 A merger tree is exactly what it sounds like. 00:19:29.619 --> 00:19:31.980 It's a genealogical family tree for galaxies. 00:19:32.599 --> 00:19:35.140 Galaxies don't usually form in isolation. The 00:19:35.140 --> 00:19:37.660 hierarchical model of structure formation dictates 00:19:37.660 --> 00:19:39.700 that small things merge together to form bigger 00:19:39.700 --> 00:19:42.640 things. Small clumps of dark matter and early 00:19:42.640 --> 00:19:44.940 stars smash together to form dwarf galaxies. 00:19:45.420 --> 00:19:47.680 Those dwarf galaxies get pulled together by gravity 00:19:47.680 --> 00:19:50.460 and merge into massive spiral galaxies like our 00:19:50.460 --> 00:19:52.819 Milky Way. And eventually spiral galaxies can 00:19:52.819 --> 00:19:55.799 collide to form giant elliptical galaxies. Precisely. 00:19:56.009 --> 00:19:59.490 A merger tree allows a scientist to pick a massive 00:19:59.490 --> 00:20:03.230 galaxy in snapshot 135 the present day and trace 00:20:03.230 --> 00:20:05.950 its ancestry backward through time. They can 00:20:05.950 --> 00:20:08.289 see exactly which smaller galaxies collided, 00:20:08.529 --> 00:20:10.369 when they collided, and how those collisions 00:20:10.369 --> 00:20:12.950 triggered burst to star formation or fed the 00:20:12.950 --> 00:20:15.349 central black hole. It's tracing the lineage 00:20:15.349 --> 00:20:17.890 of a cosmic structure over billions of years. 00:20:18.410 --> 00:20:20.630 And the accessibility of this data is a game 00:20:20.630 --> 00:20:23.799 changer. Releasing 230 terabytes of data is one 00:20:23.799 --> 00:20:26.140 thing, but expecting researchers around the world 00:20:26.140 --> 00:20:28.400 to have the local storage and computing power 00:20:28.400 --> 00:20:31.319 to download and analyze it is unrealistic. So 00:20:31.319 --> 00:20:34.259 the illustrious team built a web -based API and 00:20:34.259 --> 00:20:36.599 application programming interface. Which effectively 00:20:36.599 --> 00:20:38.880 means you don't need a supercomputer in Germany 00:20:38.880 --> 00:20:41.460 to study this universe. A graduate student with 00:20:41.460 --> 00:20:43.779 a standard laptop in a developing country or 00:20:43.779 --> 00:20:45.460 a researcher at a small university without a 00:20:45.460 --> 00:20:48.180 massive funding grant can write a script, connect 00:20:48.180 --> 00:20:50.799 to the API via the internet, run a specific search 00:20:50.799 --> 00:20:53.519 query directly on the Illustra servers and download 00:20:53.519 --> 00:20:56.119 just the specific subset of data they need. It 00:20:56.119 --> 00:20:59.579 entirely democratizes the cosmos. It allows thousands 00:20:59.579 --> 00:21:02.279 of researchers to test their own unique hypotheses 00:21:02.279 --> 00:21:05.660 about galaxy rotation, black hole growth, or 00:21:05.660 --> 00:21:08.160 dark matter distribution against a state of the 00:21:08.160 --> 00:21:11.200 art simulation. It accelerates the pace of discovery 00:21:11.200 --> 00:21:13.859 across the entire field because the bottleneck 00:21:13.859 --> 00:21:16.119 of access has been removed. It's a beautiful 00:21:16.119 --> 00:21:18.099 model for how science should be shared. Now, 00:21:18.099 --> 00:21:20.759 we've been deep in the weeds of APIs, terabytes 00:21:20.759 --> 00:21:23.220 of RAM, and supercomputing architecture, but 00:21:23.220 --> 00:21:24.819 I want to pivot to something that highlights 00:21:24.819 --> 00:21:26.940 the broader cultural impact of this achievement. 00:21:27.640 --> 00:21:30.200 In December 2018, the illustrious simulation 00:21:30.200 --> 00:21:32.700 received a very unique honor from a surprisingly 00:21:32.700 --> 00:21:35.759 old -school institution. The German Postal Service 00:21:35.759 --> 00:21:37.839 recognized the project with a special series 00:21:37.839 --> 00:21:40.779 postage stamp. It is a wonderful, almost poetic 00:21:40.779 --> 00:21:43.910 contrast. On one hand, you have 230 terabytes 00:21:43.910 --> 00:21:46.349 of the most complex, dynamically generated digital 00:21:46.349 --> 00:21:49.269 information ever produced by humanity. And on 00:21:49.269 --> 00:21:51.549 the other hand, it is commemorated by being printed 00:21:51.549 --> 00:21:54.490 on a tiny physical square of paper that you lick 00:21:54.490 --> 00:21:56.769 and stick on a letter. I think it speaks volumes 00:21:56.769 --> 00:21:59.950 about how scientific milestones transcend academia. 00:22:00.869 --> 00:22:02.950 When a physics simulation gets put on a postage 00:22:02.950 --> 00:22:06.190 stamp, it means it has entered the cultural zeitgeist. 00:22:06.349 --> 00:22:09.450 It's a physical stamp of approval for a purely 00:22:09.450 --> 00:22:12.789 virtual universe. And I think a huge part of 00:22:12.789 --> 00:22:15.170 why it captured the public imagination is the 00:22:15.170 --> 00:22:18.289 visual aspect. The data visualizations generated 00:22:18.289 --> 00:22:20.609 by illustris are absolutely breathtaking. They 00:22:20.609 --> 00:22:23.130 really are. When you render the gas density, 00:22:23.210 --> 00:22:25.289 temperature, and metallicity of the illustris 00:22:25.289 --> 00:22:28.130 universe, it looks incredibly authentic. You 00:22:28.130 --> 00:22:30.869 see the glowing, delicate filaments of the cosmic 00:22:30.869 --> 00:22:33.750 web connecting massive glowing nodes of galaxy 00:22:33.750 --> 00:22:36.789 clusters. You see the distinct spiral arms of 00:22:36.789 --> 00:22:39.589 virtual galaxies rendered in vibrant colors that 00:22:39.589 --> 00:22:42.269 map to real physical properties. It bridges the 00:22:42.269 --> 00:22:44.869 gap between abstract math and human intuition. 00:22:45.369 --> 00:22:48.309 When people can literally see the output of Arepo's 00:22:48.309 --> 00:22:50.849 hydrodynamics, it makes the staggering complexity 00:22:50.849 --> 00:22:53.410 of physical cosmology feel tangible. But the 00:22:53.410 --> 00:22:55.549 scientific process doesn't stop at a successful 00:22:55.549 --> 00:22:57.910 proof of concept or a commemorative stamp. The 00:22:57.910 --> 00:22:59.829 original illustrious simulation was a triumph, 00:22:59.930 --> 00:23:01.869 but it wasn't perfect. Right, there were flaws. 00:23:02.160 --> 00:23:04.539 There were discrepancies between the simulated 00:23:04.539 --> 00:23:07.220 universe and the real one. Certain populations 00:23:07.220 --> 00:23:10.039 of galaxies didn't match perfectly, and the physics 00:23:10.039 --> 00:23:12.539 of black hole feedback needed refinement. Which 00:23:12.539 --> 00:23:15.460 is exactly how science is supposed to work. You 00:23:15.460 --> 00:23:17.880 build a tool, you push it to its absolute limits 00:23:17.880 --> 00:23:19.900 to see where it breaks, and then you build a 00:23:19.900 --> 00:23:22.480 better tool. Once they proved that the IREPO 00:23:22.480 --> 00:23:25.299 code could construct a universe, the team didn't 00:23:25.299 --> 00:23:27.359 just pat themselves on the back and move on. 00:23:27.759 --> 00:23:30.279 They initiated a series of highly specialized 00:23:30.480 --> 00:23:33.980 incredibly ambitious spin -off projects to ask 00:23:33.980 --> 00:23:37.039 deeper, more complex questions. If we consider 00:23:37.039 --> 00:23:39.480 the original Illustris simulation as a broad 00:23:39.480 --> 00:23:42.099 foundational map of the world, the spin -offs 00:23:42.099 --> 00:23:44.259 are like using specialized satellite imaging 00:23:44.259 --> 00:23:46.859 to zoom in on specific topographies, weather 00:23:46.859 --> 00:23:49.079 patterns, or demographics. The first and most 00:23:49.079 --> 00:23:51.660 prominent follow -up was Illustris TNG, which 00:23:51.660 --> 00:23:54.279 literally stands for The Next Generation. This 00:23:54.279 --> 00:23:56.859 project was presented in July 2017, again led 00:23:56.859 --> 00:23:58.819 by Volker Springel, alongside a broader team 00:23:58.819 --> 00:24:01.619 of scientists from in Germany and the U .S.? 00:24:01.619 --> 00:24:05.000 With TNG, they significantly upgraded the physics 00:24:05.000 --> 00:24:07.660 programmed into the simulation. The most crucial 00:24:07.660 --> 00:24:11.779 addition was a field of physics known as magnetohydrodynamics. 00:24:11.920 --> 00:24:13.920 And we talked about hydrodynamics, the flow of 00:24:13.920 --> 00:24:16.559 fluids. Adding the magneto prefix implies we 00:24:16.559 --> 00:24:19.339 are now tracking magnetic fields. Why is that 00:24:19.339 --> 00:24:22.259 necessary? Are magnetic fields really that important 00:24:22.259 --> 00:24:25.980 in the vacuum of space? They are profoundly important. 00:24:26.420 --> 00:24:29.000 While space is a vacuum by earthly standards, 00:24:29.599 --> 00:24:32.160 the diffuse gas that permeates the cosmos is 00:24:32.160 --> 00:24:34.740 often ionized, meaning the atoms have been stripped 00:24:34.740 --> 00:24:36.920 of electrons, turning the gas into a plasma. 00:24:37.579 --> 00:24:39.960 Plasmas are highly conductive and interact strongly 00:24:39.960 --> 00:24:42.319 with magnetic fields. So the magnetic fields 00:24:42.319 --> 00:24:44.980 can actually shape how the gas flows. Exactly. 00:24:45.150 --> 00:24:47.309 Magnetic fields can provide pressure that supports 00:24:47.309 --> 00:24:49.910 gas against gravitational collapse. They influence 00:24:49.910 --> 00:24:52.549 how cosmic rays travel through a galaxy. And 00:24:52.549 --> 00:24:55.309 crucially, they play a massive role in how supermassive 00:24:55.309 --> 00:24:58.049 black holes accrete matter and launch those massive 00:24:58.049 --> 00:25:00.829 powerful jets we discussed earlier. In the original 00:25:00.829 --> 00:25:02.890 illustrious simulation, the fluid dynamics were 00:25:02.890 --> 00:25:05.250 purely kinetic and thermal. By adding that ghetto 00:25:05.250 --> 00:25:07.809 hydrodynamics to illustrious TNG, they made the 00:25:07.809 --> 00:25:10.380 cosmic fluid vastly more realistic. But adding 00:25:10.380 --> 00:25:13.279 magnetic fields to the moving Voronoi tessellation 00:25:13.279 --> 00:25:15.680 means the mathematical complexity just went through 00:25:15.680 --> 00:25:18.960 the roof. It did. Tracking the strength and direction 00:25:18.960 --> 00:25:21.700 of magnetic fields across billions of moving 00:25:21.700 --> 00:25:24.579 computational cells requires an immense leap 00:25:24.579 --> 00:25:27.019 in processing power. Which explains the hardware 00:25:27.019 --> 00:25:29.960 upgrade. The flagship simulation of this new 00:25:29.960 --> 00:25:32.279 project was run on the Hazel Hen machine at the 00:25:32.279 --> 00:25:34.500 High Performance Computing Center in Stuttgart, 00:25:34.500 --> 00:25:38.740 Germany. They scaled up from the original 8 ,192 00:25:38.740 --> 00:25:42.000 computer cores to employing up to 25 ,000 computer 00:25:42.000 --> 00:25:45.000 cores working in parallel. And continuing their 00:25:45.000 --> 00:25:47.180 commitment to open science, they released the 00:25:47.180 --> 00:25:50.559 TNG data publicly in December 2018. But this 00:25:50.559 --> 00:25:53.099 time, they made it even more accessible by including 00:25:53.099 --> 00:25:55.319 a JupyterLab interface. For listeners who might 00:25:55.319 --> 00:25:57.660 not be in the computer science field, JupyterLab 00:25:57.660 --> 00:26:00.000 is a web -based interactive development environment. 00:26:00.440 --> 00:26:02.720 It allows researchers to write and execute code, 00:26:02.880 --> 00:26:05.779 like Python, directly in their browser. They 00:26:05.779 --> 00:26:07.940 can analyze the TNG data right on the project 00:26:07.940 --> 00:26:10.039 servers without having to download massive data 00:26:10.039 --> 00:26:12.440 sets or set up complex coding environments locally. 00:26:12.900 --> 00:26:14.859 It lowered the barrier to entry even further. 00:26:15.200 --> 00:26:18.160 Illustris TNG was a massive step forward, but 00:26:18.160 --> 00:26:20.500 it was still simulating a large representative 00:26:20.500 --> 00:26:23.319 volume of the universe. What if you want to understand 00:26:23.319 --> 00:26:26.119 something very specific, like the exact history 00:26:26.119 --> 00:26:29.460 of our own home galaxy? That brings us to the 00:26:29.460 --> 00:26:32.480 first of the highly specialized spin -offs, the 00:26:32.480 --> 00:26:35.420 Auriga Project. The Auriga Project is fascinating 00:26:35.420 --> 00:26:38.619 because it shifts the focus from the macro to 00:26:38.619 --> 00:26:41.140 the micro. The source notes that Auriga consists 00:26:41.140 --> 00:26:44.019 of high -resolution zoom simulations of Milky 00:26:44.019 --> 00:26:46.740 Way -like dark matter halos. Break down that 00:26:46.740 --> 00:26:49.089 term for us. Zoom simulation. Does that mean 00:26:49.089 --> 00:26:51.049 they just take the data from the main simulation 00:26:51.049 --> 00:26:53.569 and look really closely at one part? Not exactly. 00:26:53.789 --> 00:26:56.369 In a full box simulation like Illustris, the 00:26:56.369 --> 00:26:58.490 resolution, the smallest detail you can accurately 00:26:58.490 --> 00:27:01.130 model, is limited by the sheer size of the universe 00:27:01.130 --> 00:27:02.769 you're trying to fit into the computer's memory. 00:27:03.150 --> 00:27:05.009 If you want to study the fine details of how 00:27:05.009 --> 00:27:07.950 a galaxy's spiral arms form, or how a galaxy 00:27:07.950 --> 00:27:10.450 cannibalizes smaller dwarf galaxies over time, 00:27:10.869 --> 00:27:13.130 the full box resolution isn't sharp enough. It's 00:27:13.130 --> 00:27:15.329 like trying to read a license plate and a satellite 00:27:15.329 --> 00:27:18.140 photo of an entire continent. Right, so a zoom 00:27:18.140 --> 00:27:20.819 simulation uses a clever trick. They run a low 00:27:20.819 --> 00:27:23.619 resolution simulation of large volume of space 00:27:23.619 --> 00:27:26.660 and they identify a specific dark matter halo 00:27:26.660 --> 00:27:29.339 that has roughly the same mass and environment 00:27:29.339 --> 00:27:32.809 as our Milky Way. Then they rerun the simulation 00:27:32.809 --> 00:27:35.210 from the Big Bang, but they concentrate almost 00:27:35.210 --> 00:27:37.869 all of the computing power and millions of high 00:27:37.869 --> 00:27:40.190 -resolution particles specifically into that 00:27:40.190 --> 00:27:42.329 one small region of space. Well, the rest of 00:27:42.329 --> 00:27:44.470 the universe surrounding it is simulated at a 00:27:44.470 --> 00:27:46.829 very low resolution just to provide the correct 00:27:46.829 --> 00:27:48.950 gravitational background. Exactly. It allows 00:27:48.950 --> 00:27:51.210 them to study the incredibly intricate formation 00:27:51.210 --> 00:27:54.049 history of a Milky Way analog, its spiral structure, 00:27:54.269 --> 00:27:57.069 its central bulge, its surrounding halo of stars 00:27:57.069 --> 00:27:59.329 and breathtaking high definition, helping us 00:27:59.329 --> 00:28:01.529 understand how our own galactic backyard came 00:28:01.529 --> 00:28:03.910 to be. That is brilliant. OK, moving down the 00:28:03.910 --> 00:28:06.470 timeline. In 2021, we see the introduction of 00:28:06.470 --> 00:28:09.250 a spinoff called Thesen, and this one tackles 00:28:09.250 --> 00:28:12.109 a very specific and frankly mysterious era of 00:28:12.109 --> 00:28:15.029 cosmic history. The source describes Thesen as 00:28:15.029 --> 00:28:17.529 a radiative transfer version of illustris TNG 00:28:17.529 --> 00:28:20.470 designed to explore the epoch of reionization. 00:28:20.849 --> 00:28:23.150 I think we need to define some terms here, starting 00:28:23.150 --> 00:28:25.329 with the epoch of reionization. What is that? 00:28:25.529 --> 00:28:27.809 To understand reionization, we have to go back 00:28:27.809 --> 00:28:31.390 to the Cosmic Dark Ages, roughly 380 ,000 years 00:28:31.390 --> 00:28:34.269 after the Big Bang. At that time, the universe 00:28:34.269 --> 00:28:36.869 had expanded and cooled enough for protons and 00:28:36.869 --> 00:28:39.829 electrons to combine and form neutral hydrogen 00:28:39.829 --> 00:28:41.890 atoms. Which means the universe was no longer 00:28:41.890 --> 00:28:44.349 a hot, glowing plasma. Right. It became filled 00:28:44.349 --> 00:28:46.670 with a thick, opaque fog of neutral hydrogen 00:28:46.670 --> 00:28:49.750 gas. There were no stars, no galaxies, no sources 00:28:49.750 --> 00:28:52.839 of light. hence the Dark Ages. But eventually, 00:28:53.220 --> 00:28:54.859 gravity pulled enough gas together to ignite 00:28:54.859 --> 00:28:57.180 the very first stars and the first primitive 00:28:57.180 --> 00:28:59.359 galaxies. And when those stars turned on, they 00:28:59.359 --> 00:29:01.960 started pumping out ultraviolet radiation. Yes. 00:29:02.480 --> 00:29:04.920 And those high -energy UV photons slammed into 00:29:04.920 --> 00:29:07.259 the surrounding neutral hydrogen fog, tearing 00:29:07.259 --> 00:29:09.880 the electrons away from the protons. This process 00:29:09.880 --> 00:29:13.400 is called ionization. As more stars formed, these 00:29:13.400 --> 00:29:16.700 bubbles of ionized transparent gas grew and merged 00:29:16.700 --> 00:29:20.019 until eventually the entire universe was reionized. 00:29:20.359 --> 00:29:23.079 It was the dawn of light. The fog lifted and 00:29:23.079 --> 00:29:25.039 the universe became transparent to light, which 00:29:25.039 --> 00:29:27.910 is why we can see distant galaxies today. That 00:29:27.910 --> 00:29:29.970 is an incredibly poetic moment in the history 00:29:29.970 --> 00:29:32.769 of the cosmos, but simulating it requires a new 00:29:32.769 --> 00:29:35.630 type of physics, which the text calls radiative 00:29:35.630 --> 00:29:37.990 transfer. Tracking the gravity and fluid dynamics 00:29:37.990 --> 00:29:40.529 of the gas is hard enough. Radiative transfer 00:29:40.529 --> 00:29:43.130 requires calculating how individual packets of 00:29:43.130 --> 00:29:45.230 light photons travel through that moving fluid, 00:29:45.670 --> 00:29:47.730 how they are absorbed by the gas, and how that 00:29:47.730 --> 00:29:50.269 absorption heats and ionizes the gas, which in 00:29:50.269 --> 00:29:52.190 turn changes the pressure and the fluid dynamics. 00:29:52.670 --> 00:29:54.970 It's a deep, complex feedback loop between light 00:29:54.970 --> 00:29:57.309 and matter. Thiessen was built specifically to 00:29:57.309 --> 00:29:59.450 model this dawn of light and understand how the 00:29:59.450 --> 00:30:01.569 first stars transformed the entire universe. 00:30:01.789 --> 00:30:04.289 From the dawn of light, we move to the absolute 00:30:04.289 --> 00:30:07.809 giants of the cosmos. In 2022, the Millennium 00:30:07.809 --> 00:30:11.390 TNG project was launched. The text notes that 00:30:11.390 --> 00:30:14.309 it employs the galaxy formation model in a larger 00:30:14.309 --> 00:30:17.450 cosmological volume to explore the massive end 00:30:17.450 --> 00:30:20.630 of the halo mass function for detailed cosmological 00:30:20.630 --> 00:30:24.309 probe forecasts. There's a lot of dense terminology 00:30:24.309 --> 00:30:26.549 there. Let's start with the halo mass function. 00:30:27.009 --> 00:30:29.390 Simply put, the halo mass function is a demographic 00:30:29.390 --> 00:30:31.960 census of the universe. It is a mathematical 00:30:31.960 --> 00:30:34.740 curve that tells you how many dark matter halos 00:30:34.740 --> 00:30:37.500 exist at various different masses. So it tells 00:30:37.500 --> 00:30:39.859 you that there are trillions of small halos, 00:30:40.220 --> 00:30:42.680 billions of medium halos, and only a few massive 00:30:42.680 --> 00:30:45.859 ones. Correct. But to accurately plot the massive 00:30:45.859 --> 00:30:48.319 end of that curve to study the absolute largest 00:30:48.319 --> 00:30:50.619 dark matter structures in the universe, you run 00:30:50.619 --> 00:30:52.960 into a statistical problem. Because massive halos 00:30:52.960 --> 00:30:55.400 are extremely rare, you won't find many of them 00:30:55.400 --> 00:30:57.799 in a small or even medium sized simulation box. 00:30:58.019 --> 00:31:00.000 If you want to study the demographics of billionaires, 00:31:00.339 --> 00:31:02.380 surveying a small town won't work. You have to 00:31:02.380 --> 00:31:05.740 survey an entire country. Perfect analogy. Millennium 00:31:05.740 --> 00:31:08.619 TNG scaled up the volume of the simulation drastically. 00:31:08.799 --> 00:31:11.980 They needed an enormous virtual universe to ensure 00:31:11.980 --> 00:31:14.799 they captured a statistically significant sample 00:31:14.799 --> 00:31:17.279 of these extremely rare gigantic structures. 00:31:17.940 --> 00:31:19.720 They want to understand how these titans form 00:31:19.720 --> 00:31:22.740 and how they cluster, because the exact number 00:31:22.740 --> 00:31:25.579 of these massive halos is highly sensitive to 00:31:25.579 --> 00:31:27.740 the fundamental properties of dark energy and 00:31:27.740 --> 00:31:30.670 the total mass of neutrinos. By comparing the 00:31:30.670 --> 00:31:32.589 massive end of the halo mass function in the 00:31:32.589 --> 00:31:35.329 simulation to what telescopes like Euclid or 00:31:35.329 --> 00:31:37.450 the Nancy Grace Roman Space Telescope will observe, 00:31:38.109 --> 00:31:40.289 scientists can precisely calibrate our cosmological 00:31:40.289 --> 00:31:42.910 models. And that scales perfectly into the final 00:31:42.910 --> 00:31:45.369 spin -off mention, TNG cluster, introduced in 00:31:45.369 --> 00:31:48.109 2023. This is a suite of high -resolution zoom 00:31:48.109 --> 00:31:50.609 -in simulations specifically targeting galaxy 00:31:50.609 --> 00:31:53.670 clusters. If Millennium TNG was casting a wide 00:31:53.670 --> 00:31:57.420 net to find where the giants live, TNG clusters 00:31:57.420 --> 00:32:00.880 of putting those giants under a microscope. Galaxy 00:32:00.880 --> 00:32:02.799 clusters are the largest gravitationally bound 00:32:02.799 --> 00:32:05.480 objects in the universe. We are talking about 00:32:05.480 --> 00:32:07.559 structures containing hundreds or even thousands 00:32:07.559 --> 00:32:10.539 of individual galaxies all swarming around inside 00:32:10.539 --> 00:32:13.940 a colossal dark matter halo permeated by incredibly 00:32:13.940 --> 00:32:17.609 hot x -ray emitting gas. The physics in an environment 00:32:17.609 --> 00:32:20.029 like that must be incredibly extreme. You have 00:32:20.029 --> 00:32:22.190 galaxies moving at millions of miles an hour, 00:32:22.390 --> 00:32:24.109 smashing into each other, being stripped of their 00:32:24.109 --> 00:32:26.309 gas by the sheer friction of moving through the 00:32:26.309 --> 00:32:28.529 cluster environment. It is the ultimate cosmic 00:32:28.529 --> 00:32:31.490 mosh pit. TNG Cluster uses the zoom technique 00:32:31.490 --> 00:32:33.650 we discussed earlier to resolve the violent, 00:32:33.930 --> 00:32:36.650 complex astrophysics happening inside these massive 00:32:36.650 --> 00:32:39.230 clusters. How does a supermassive black hole 00:32:39.230 --> 00:32:41.730 in the central galaxy of a cluster manage to 00:32:41.730 --> 00:32:43.910 heat up gas that is millions of light years away? 00:32:44.039 --> 00:32:46.299 How do galaxies survive plunging through the 00:32:46.299 --> 00:32:48.380 center of the cluster? These are the frontier 00:32:48.380 --> 00:32:51.059 questions of astrophysics today. So let's pull 00:32:51.059 --> 00:32:53.680 all of this together. We started with the desire 00:32:53.680 --> 00:32:56.480 to map the unmappable. We looked at a single, 00:32:56.720 --> 00:32:59.119 mathematically elegant piece of code called a 00:32:59.119 --> 00:33:02.359 repo, named after a Roman palindrome. From that 00:33:02.359 --> 00:33:04.839 foundation, teams of brilliant scientists utilizing 00:33:04.839 --> 00:33:07.839 tens of thousands of supercomputer cores, millions 00:33:07.839 --> 00:33:11.200 of CPU hours, and an incredible grasp of gravity, 00:33:11.519 --> 00:33:13.720 hydrodynamics, and magnetic fields managed to 00:33:13.720 --> 00:33:17.559 simulate a 13 .8 billion year universe. They 00:33:17.559 --> 00:33:20.160 simulated the invisible scaffolding of dark matter, 00:33:20.579 --> 00:33:23.720 the chaotic explosion of supernovae, the terrifying 00:33:23.720 --> 00:33:26.460 feedback of supermassive black holes, the intricate 00:33:26.460 --> 00:33:29.119 dance of fluid dynamics, the dawn of light during 00:33:29.119 --> 00:33:31.339 reionization, and the assembly of the largest 00:33:31.339 --> 00:33:33.440 clusters in the cosmos. And then, instead of 00:33:33.440 --> 00:33:35.539 locking it away, they opened the doors. They 00:33:35.539 --> 00:33:38.740 provided APIs and web interfaces, handing a 230 00:33:38.740 --> 00:33:41.059 terabyte digital universe to the global scientific 00:33:41.059 --> 00:33:44.160 community. It sparked a golden age of computational 00:33:44.160 --> 00:33:46.819 cosmology, earning a place in cultural history 00:33:46.819 --> 00:33:48.980 right down to a commemorative postage stamp. 00:33:50.400 --> 00:33:53.059 and its TNG successors are more than just a map 00:33:53.059 --> 00:33:56.039 of stars. They are a testament to human ingenuity. 00:33:56.579 --> 00:33:58.640 By combining the complex math of microphysics, 00:33:59.119 --> 00:34:01.859 they created a sandbox for reality. They built 00:34:01.859 --> 00:34:03.880 an environment where we can test the very laws 00:34:03.880 --> 00:34:06.619 of nature and watch the consequences unfold and 00:34:06.619 --> 00:34:08.800 fast forward. And for you listening to this, 00:34:08.820 --> 00:34:11.260 whether you are trying to conceptualize the vast 00:34:11.260 --> 00:34:13.719 cosmic web or you are just looking for a bit 00:34:13.719 --> 00:34:15.860 of profound inspiration before you tackle your 00:34:15.860 --> 00:34:18.340 next big project, the ultimate takeaway here 00:34:18.340 --> 00:34:21.119 is the astonishing power of modeling complex 00:34:21.119 --> 00:34:23.820 systems. Think about the implications. If human 00:34:23.820 --> 00:34:26.059 beings can collaborate, write code, and build 00:34:26.059 --> 00:34:29.139 machines powerful enough to simulate 13 .8 billion 00:34:29.139 --> 00:34:31.300 years of galaxy formation down to the magnetic 00:34:31.300 --> 00:34:33.739 fields and the cooling rates of gas, what other 00:34:33.739 --> 00:34:36.360 massively complex, seemingly unsolvable problems 00:34:36.360 --> 00:34:38.980 can we model, understand, and ultimately solve 00:34:38.980 --> 00:34:41.440 here on Earth? It is a deeply empowering thought. 00:34:42.000 --> 00:34:43.980 But in the spirit of this deep dive, I want to 00:34:43.980 --> 00:34:46.119 leave you with an idea that is perhaps slightly 00:34:46.119 --> 00:34:48.699 more provocative to mull over. We've discussed 00:34:48.699 --> 00:34:51.199 the unbelievable detail of the illustrious simulation. 00:34:51.780 --> 00:34:54.679 It is so complex that researchers require merger 00:34:54.679 --> 00:34:57.699 trees just to track the billions of virtual cosmic 00:34:57.699 --> 00:35:00.400 events and genealogies over vast stretches of 00:35:00.400 --> 00:35:03.320 time. It is, for all intents and purposes, a 00:35:03.320 --> 00:35:05.719 functioning universe within a machine. The universe 00:35:05.719 --> 00:35:09.400 constrained by 25 terabytes of RAM. Indeed. But 00:35:09.400 --> 00:35:11.989 consider the trajectory of our technology. If 00:35:11.989 --> 00:35:14.289 a supercomputer in Germany in the 21st century 00:35:14.289 --> 00:35:18.050 can generate 136 distinct time snapshots of a 00:35:18.050 --> 00:35:20.670 virtual reality that remarkably mirrors our own 00:35:20.670 --> 00:35:22.909 observable universe, what happens in a thousand 00:35:22.909 --> 00:35:25.250 years, or a million years? If we can simulate 00:35:25.250 --> 00:35:27.969 reality this well now, how do we know for certain 00:35:27.969 --> 00:35:30.090 that our own universe, with all its galaxies, 00:35:30.389 --> 00:35:32.949 dark matter and cosmic webs, isn't just the ultimate, 00:35:33.329 --> 00:35:35.610 infinitely high -resolution simulation running 00:35:35.610 --> 00:35:38.710 on some unimaginable cosmic supercomputer? So 00:35:38.710 --> 00:35:41.550 that is the perfect mind -bending thought to 00:35:41.550 --> 00:35:45.489 end on. A reality check for reality itself. Thank 00:35:45.489 --> 00:35:47.210 you so much for joining us on this deep dive 00:35:47.210 --> 00:35:49.610 into the illustrious project and the incredible 00:35:49.610 --> 00:35:51.829 science of simulating the cosmos. Keep asking 00:35:51.829 --> 00:35:53.550 questions, keep wondering at the scale of it 00:35:53.550 --> 00:35:55.510 all, and we will catch you on the next one.