WEBVTT 1 00:00:03.399 --> 00:00:07.719 Welcome to Bedtime Astronomy. Explore the wonders of the cosmos 2 00:00:07.759 --> 00:00:12.279 with our soothing Bedtime Astronomy podcast. Each episode offers a 3 00:00:12.359 --> 00:00:16.320 gentle journey through the stars, planets, and beyond, perfect for 4 00:00:16.399 --> 00:00:20.239 unwinding after a long day. Let's travel through the mysteries 5 00:00:20.239 --> 00:00:22.440 of the universe as you drift off into a peaceful 6 00:00:22.480 --> 00:00:26.719 slumber under the night sky. 7 00:00:26.839 --> 00:00:30.199 Welcome everyone to what I think is a truly awe 8 00:00:30.239 --> 00:00:32.719 inspiring journey we're taking today right into the heart of 9 00:00:32.719 --> 00:00:35.640 the cosmos. Tonight, We're not just looking at the stars, 10 00:00:35.640 --> 00:00:38.920 you see. We're going to zoom out, way out, far 11 00:00:39.000 --> 00:00:42.399 beyond what our eyes or even our best telescopes can 12 00:00:42.439 --> 00:00:46.439 really show us. We want to see the universe's true skeleton. 13 00:00:47.119 --> 00:00:51.880 Imagine if you can uncovering this immense, sprawling architecture that 14 00:00:51.920 --> 00:00:53.840 basically dictates where everything is. 15 00:00:53.920 --> 00:00:55.000 And where it isn't. 16 00:00:54.920 --> 00:01:00.000 Exactly across distances that they defy comprehension, don't they. 17 00:01:00.039 --> 00:01:00.679 They absolutely do. 18 00:01:01.079 --> 00:01:04.760 That's precisely what we're here to explore, this mind bending 19 00:01:04.840 --> 00:01:08.359 reality of the cosmic web, that's the structure of our universe, 20 00:01:08.879 --> 00:01:12.719 and it's a challenge for scientists of well unimaginable scale 21 00:01:12.760 --> 00:01:15.079 and complexity. How do you even start to make sense 22 00:01:15.079 --> 00:01:19.239 of something so vast, so intricate, and so ancient. 23 00:01:19.439 --> 00:01:21.200 Right it sounds like science fiction almost, it. 24 00:01:21.200 --> 00:01:24.640 Really does, but it's very real, and thanks to some 25 00:01:24.680 --> 00:01:27.159 cutting edge tools, we're actually getting closer to mapping it 26 00:01:27.200 --> 00:01:28.079 than ever before. 27 00:01:28.719 --> 00:01:32.239 And that's exactly our mission. In this deep dive. We're 28 00:01:32.239 --> 00:01:36.319 going to reveal how an international team of scientists has developed, 29 00:01:37.000 --> 00:01:41.719 well a truly revolutionary shortcut to do just that, a shortcut. Okay, 30 00:01:41.799 --> 00:01:45.159 I'm intrigued. Yeah, they've built this ultra fast emulator. It's 31 00:01:45.200 --> 00:01:49.000 dubbed Effort dot JL, and it's capable of mapping the 32 00:01:49.079 --> 00:01:52.439 universe much faster and with incredible detail. 33 00:01:52.560 --> 00:01:52.959 Wow. 34 00:01:53.079 --> 00:01:55.599 And this isn't just about, you know, speeding up research. 35 00:01:55.640 --> 00:01:58.159 It's really about unlocking some of the deepest secrets of 36 00:01:58.200 --> 00:02:01.879 cosmic evolution itself. Okay, so we'll explore not just what 37 00:02:01.920 --> 00:02:05.760 this emulator does and how it works, but critically why 38 00:02:05.799 --> 00:02:09.199 it matters so profoundly for the future of cosmology and 39 00:02:09.240 --> 00:02:11.840 for our understanding of everything around us, really, from the 40 00:02:11.879 --> 00:02:14.159 smallest particles to the biggest structures out there. 41 00:02:14.199 --> 00:02:17.960 Okay, let's unpack the scale first, because it's mind bending. Right. 42 00:02:18.000 --> 00:02:20.360 When we talk about the universe, our brains, they kind 43 00:02:20.400 --> 00:02:23.120 of default to things we can grasp, like the size 44 00:02:23.120 --> 00:02:25.919 of Earth, maybe the Solar System. You're familiar scales, But 45 00:02:26.000 --> 00:02:29.879 what we're talking about here, it's just a whole different level, 46 00:02:29.879 --> 00:02:33.000 a different magnitude entirely. You look up at the night sky, 47 00:02:33.120 --> 00:02:38.719 you see galaxies, those magnificent swirls, billions of stars. They 48 00:02:38.759 --> 00:02:42.360 seem enormous to us, unfathomably. 49 00:02:41.439 --> 00:02:45.479 Large, just one a universe unto itself, almost exactly. 50 00:02:45.919 --> 00:02:48.800 But in the grand scheme, a galaxy, even our own 51 00:02:48.840 --> 00:02:51.840 Milky Way, it's basically nothing more than a tiny speck, 52 00:02:52.240 --> 00:02:54.080 an almost microscopic. 53 00:02:53.520 --> 00:02:55.960 Dot, right, hard to internalize that scale. 54 00:02:56.039 --> 00:02:59.840 It really is just one of countless tiny dots scattered 55 00:02:59.840 --> 00:03:04.280 across this unimaginable expanse. And crucially, these dots they don't 56 00:03:04.319 --> 00:03:05.599 just float randomly, do they. 57 00:03:05.719 --> 00:03:07.159 No, not at all. Gravity plays a. 58 00:03:07.159 --> 00:03:10.800 Huge rule, right, They have this profound tendency to aggregate, 59 00:03:11.199 --> 00:03:14.599 to clump together. Under that relentless pull of gravity. We 60 00:03:14.680 --> 00:03:18.479 see them forming these immense collections called clusters. 61 00:03:18.120 --> 00:03:20.479 One hundreds, sometimes thousands of galaxies together. 62 00:03:20.639 --> 00:03:22.719 Yeah, and then as if that wasn't big enough, these 63 00:03:22.759 --> 00:03:27.039 clusters aren't isolated either. They then aggregate into even larger structures. 64 00:03:27.120 --> 00:03:30.919 Scientists call them superclusters, our own Milky Way. For instance, 65 00:03:30.960 --> 00:03:34.280 it's part of the local group, which is itself sort 66 00:03:34.319 --> 00:03:37.400 of on the outskirts of the Virgo supercluster, which is 67 00:03:37.439 --> 00:03:41.479 part of lani Akiya exactly. He is supercluster. This scale 68 00:03:41.560 --> 00:03:43.560 just keeps ballooning outwards. It's incredible. 69 00:03:43.599 --> 00:03:44.159 It keeps going. 70 00:03:44.360 --> 00:03:47.000 But the aggregation doesn't stop there, does it. And this 71 00:03:47.199 --> 00:03:51.400 is where the picture really starts to come into focus. 72 00:03:51.479 --> 00:03:52.919 This is the cosmic web itself. 73 00:03:53.080 --> 00:03:56.639 Yeah, all these clusters and superclusters, they aren't just scattered 74 00:03:56.719 --> 00:04:02.360 randomly in space. They're woven together into the immense interconnected network, 75 00:04:02.840 --> 00:04:06.080 like a colossal cosmic spider web. 76 00:04:06.240 --> 00:04:07.240 It's a great analogy. 77 00:04:07.400 --> 00:04:10.879 This is the cosmic web, this incredible three dimensional skeleton 78 00:04:10.919 --> 00:04:14.520 of the universe. It's characterized by these long thread like 79 00:04:14.599 --> 00:04:16.839 structures filaments they call them filaments. 80 00:04:16.920 --> 00:04:20.079 Yes, that's where galaxies tend to cluster along these sort 81 00:04:20.079 --> 00:04:20.720 of cosmic. 82 00:04:20.519 --> 00:04:23.879 Highways, and these filaments can stretch for hundreds of millions 83 00:04:23.920 --> 00:04:28.199 of light years, forming the densest regions. And then between 84 00:04:28.319 --> 00:04:31.720 these filaments you find these immense almost completely empty regions 85 00:04:31.759 --> 00:04:33.680 called voids, vast voids. 86 00:04:33.800 --> 00:04:37.240 Yeah, hundreds of millions of light years across, often with 87 00:04:37.399 --> 00:04:38.959 very few galaxies, if any. 88 00:04:39.279 --> 00:04:41.720 So it's in these filaments that most of the well, 89 00:04:41.839 --> 00:04:46.040 the visible stuff, the observable matter in the universe is concentrated, 90 00:04:46.279 --> 00:04:49.600 giving the universe its fundamental, large scale structure. 91 00:04:49.800 --> 00:04:52.519 Yeah. What's truly fascinating here is just how difficult it 92 00:04:52.560 --> 00:04:56.319 is for our intuition, for us humans to really grasp 93 00:04:56.839 --> 00:05:01.439 structures this immense, and maybe more import how they even formed. 94 00:05:01.680 --> 00:05:02.920 Right, the formation process. 95 00:05:02.920 --> 00:05:06.319 We're talking about scales that just dwarf anything we experience 96 00:05:06.399 --> 00:05:08.639 day to day. And the cosmic web isn't just like 97 00:05:08.759 --> 00:05:12.279 a static picture, you know, a snapshot frozen in time. 98 00:05:12.439 --> 00:05:15.720 It evolved exactly. It's the culmination of billions of years 99 00:05:15.759 --> 00:05:19.680 of cosmic evolution, driven primarily by gravity. To really get this, 100 00:05:19.759 --> 00:05:22.160 you have to step back way back to the dawn 101 00:05:22.199 --> 00:05:22.720 of the universe. 102 00:05:22.800 --> 00:05:24.600 Okay, near the Big Bang, right after the Big Bang. 103 00:05:25.040 --> 00:05:28.639 Imagine the universe then this hot, dense, nearly uniform soup 104 00:05:28.680 --> 00:05:32.560 of particles. But nearly the keyword. 105 00:05:32.199 --> 00:05:33.920 There, Oh, not perfectly uniform. 106 00:05:34.199 --> 00:05:39.519 No, there were these infinitesimal quantum fluctuations in density, tiny 107 00:05:39.600 --> 00:05:42.839 tiny ripples in this cosmic ocean, so small they were 108 00:05:42.879 --> 00:05:47.720 imperceptible at first, but over billions of years, gravity began 109 00:05:47.759 --> 00:05:51.360 to relentlessly amplify these tiny ripples. And here's where the 110 00:05:51.399 --> 00:05:56.680 mysterious dark matter plays its huge unseen foundational role. 111 00:05:56.920 --> 00:05:59.720 Ah, dark matter the invisible stuff exactly. 112 00:06:00.079 --> 00:06:03.079 It doesn't interact with light, but it has immense gravitational 113 00:06:03.120 --> 00:06:06.360 pull and it started to clump together first. Because it 114 00:06:06.399 --> 00:06:08.920 doesn't interact with light or normal matter in the same way, 115 00:06:09.279 --> 00:06:13.000 it collapse more efficiently, forming this invisible scaffolding or skeleton 116 00:06:13.199 --> 00:06:14.319 throughout the early universe. 117 00:06:14.519 --> 00:06:16.680 So dark matter laid the groundwork. 118 00:06:16.399 --> 00:06:19.680 Pretty much normal matter, the stuff that forms stars, planets. 119 00:06:19.759 --> 00:06:23.600 You then gradually fell into these gravitational wells created by 120 00:06:23.639 --> 00:06:26.759 the dark matter. It essentially traced out this underlying structure. 121 00:06:26.439 --> 00:06:27.680 Like following a blueprint. 122 00:06:28.079 --> 00:06:30.759 Yeah, like an invisible blueprint forms first, and then the 123 00:06:30.839 --> 00:06:33.519 luminous flesh of the universe kind of coalesces around it. 124 00:06:33.759 --> 00:06:37.079 So it's gravity that dictated where matter is and crucially 125 00:06:37.079 --> 00:06:41.040 where it isn't. These structures are the direct consequence of 126 00:06:41.079 --> 00:06:45.839 the universe's initial conditions, it's expansion, its evolution. It basically 127 00:06:45.839 --> 00:06:48.360 tells the whole story from the Big Bang onwards. 128 00:06:48.560 --> 00:06:52.759 And that's exactly the central, just mind boggling question scientists face, 129 00:06:52.879 --> 00:06:55.839 isn't it. How do you even begin to see this, 130 00:06:56.480 --> 00:07:00.879 or study it, or truly understand something so vast, so intricate, 131 00:07:00.920 --> 00:07:01.920 so ancient. 132 00:07:01.839 --> 00:07:04.480 Especially when it's sculpted mainly by invisible dark. 133 00:07:04.240 --> 00:07:08.279 Matter right spanning nearly the entire observable universe. How do 134 00:07:08.360 --> 00:07:11.399 we map its fundamental properties, measure how it evolved over 135 00:07:11.480 --> 00:07:15.079 cosmic time, uncover the precise physics governing at all. It's 136 00:07:15.079 --> 00:07:17.160 not like we can just you send a probe out 137 00:07:17.160 --> 00:07:18.639 there to take pictures of the whole thing. 138 00:07:18.720 --> 00:07:19.759 Definitely not feasible. 139 00:07:19.839 --> 00:07:23.120 So this monumental challenge, it's driven scientists to develop some 140 00:07:23.199 --> 00:07:29.439 incredibly sophisticated methods combining meticulous observation with a deep theoretical insight, 141 00:07:30.199 --> 00:07:32.720 which I guess leads us to our next major point. 142 00:07:32.959 --> 00:07:37.759 Precisely to tackle this, the traditional and still incredibly powerful 143 00:07:37.759 --> 00:07:42.360 scientific approach involves theoretical models. Specifically, one of the most 144 00:07:42.360 --> 00:07:47.160 successful and robust frameworks for understanding the large scale structure 145 00:07:47.279 --> 00:07:50.399 is called ef TOW f LS. 146 00:07:50.160 --> 00:07:51.879 Ef toe FLSs. 147 00:07:52.000 --> 00:07:54.439 Okay. It stands for the effective field theory of large 148 00:07:54.439 --> 00:07:58.519 scale structure. Its fundamental purpose is to provide a comprehensive, 149 00:07:58.680 --> 00:08:00.839 rigorous description of the cosmic web. 150 00:08:01.040 --> 00:08:01.879 How does it do that? 151 00:08:02.000 --> 00:08:04.800 It does this by combining the known physics of the universe, 152 00:08:05.000 --> 00:08:09.120 everything from say, general relativities influence on gravity, to the 153 00:08:09.160 --> 00:08:12.199 properties of the fundamental particles making up matter and energy. 154 00:08:12.600 --> 00:08:15.079 It combines all that with the vast amounts of observational 155 00:08:15.120 --> 00:08:19.160 data we gather from astronomical instruments, telescopes, sky surveys, things 156 00:08:19.199 --> 00:08:19.480 like that. 157 00:08:19.560 --> 00:08:22.079 Okay, so it blends theory and observation exactly. 158 00:08:22.439 --> 00:08:24.959 Now, how do these models work well? Rather than trying 159 00:08:25.000 --> 00:08:27.920 to track every single particle or even every single galaxy individually, 160 00:08:27.959 --> 00:08:31.360 which would be utterly impossible just think of the numbers involved. 161 00:08:31.040 --> 00:08:32.720 Like mapping every grain of sand on. 162 00:08:32.720 --> 00:08:37.480 Earth right exactly that kind of impossible scale. Instead, ef 163 00:08:37.600 --> 00:08:42.279 to FLSs describes the cosmic web statistically. It focuses on 164 00:08:42.279 --> 00:08:45.840 the collective behavior the distribution of matter on very large scales. 165 00:08:46.279 --> 00:08:50.279 It averages out the messy complex details of individual galaxies 166 00:08:50.360 --> 00:08:50.879 or small. 167 00:08:50.759 --> 00:08:53.840 Clusters, so it looks at the big picture trends precisely. 168 00:08:54.559 --> 00:08:57.960 And by feeding these models with real observations, scientists can 169 00:08:58.000 --> 00:09:02.159 then estimate the key cosmological parameters that define the cosmic 170 00:09:02.159 --> 00:09:05.480 web structure. Parameters like things like the total. 171 00:09:05.200 --> 00:09:07.879 Density of matter and dark energy in the universe, the 172 00:09:07.879 --> 00:09:11.240 initial amplitude of those tiny fluctuations we talked about earlier, 173 00:09:11.480 --> 00:09:14.440 the rate the universe is expanding, things like that, gotcha, 174 00:09:14.600 --> 00:09:17.919 And by pinning down these parameters ef to FLSs lets 175 00:09:18.000 --> 00:09:20.879 us gain crucial insights into how the cosmic web formed, 176 00:09:21.159 --> 00:09:24.840 how it evolved, and the underlying cosmological principles that govern 177 00:09:24.919 --> 00:09:25.120 it all. 178 00:09:25.200 --> 00:09:27.919 Okay, here's where it gets really interesting but also maybe 179 00:09:27.919 --> 00:09:29.840 a little abstract. For those of us who aren't theoretical 180 00:09:29.879 --> 00:09:33.879 physicists right effective field theory of large scale structure, It 181 00:09:33.879 --> 00:09:35.120 sounds incredibly complex. 182 00:09:35.159 --> 00:09:35.720 It is complex. 183 00:09:35.799 --> 00:09:38.840 Yeah, for listeners who aren't steeped in this stuff, How 184 00:09:38.879 --> 00:09:42.159 can we really grasp what this theory does? How does 185 00:09:42.159 --> 00:09:46.200 it simplify such immense complexity without losing the essential truth? 186 00:09:46.879 --> 00:09:49.759 Is there like a way to picture it? An analogy? 187 00:09:49.879 --> 00:09:52.600 Maybe? Absolutely? And actually Marco Benici, one of the lead 188 00:09:52.639 --> 00:09:54.720 researches on the effort dot jail study we'll get to 189 00:09:55.399 --> 00:09:58.399 he offers a fantastic analogy that really helps make ethodo 190 00:09:58.480 --> 00:09:59.720 of LSS more. 191 00:09:59.519 --> 00:10:01.159 COMPREHENDI right, let's hear it. 192 00:10:01.320 --> 00:10:04.000 Okay, Imagine you're trying to study a glass of water, 193 00:10:04.799 --> 00:10:09.120 specifically how it moves when you stir it or pour it. Now, 194 00:10:09.159 --> 00:10:11.840 in theory, you could try to describe the water's movement 195 00:10:12.120 --> 00:10:15.799 by tracking every single individual atom or even the subatomic 196 00:10:15.840 --> 00:10:17.840 particles within them. That's the microscopic level. 197 00:10:17.879 --> 00:10:19.320 Okay, but that sounds impossible. 198 00:10:19.480 --> 00:10:22.360 Well, if you wanted to accurately describe what happens when 199 00:10:22.399 --> 00:10:25.679 the water moves, trying to calculate the interactions of countless 200 00:10:25.679 --> 00:10:30.039 trillions of atoms, it leads to this explosive growth of calculations. 201 00:10:30.720 --> 00:10:33.720 It would be practically, if not theoretically, impossible, to simulate 202 00:10:33.759 --> 00:10:36.799 even a few seconds of fluid movement at that fundamental 203 00:10:36.840 --> 00:10:41.159 atomic scale. Too much information exactly. So instead, what you 204 00:10:41.200 --> 00:10:44.840 do is you encode certain properties at the microscopic level. 205 00:10:45.120 --> 00:10:47.799 You understand atoms have mass, they attract each other, they collide, 206 00:10:47.840 --> 00:10:50.360 they have kinetic energy, et cetera, and then you observe 207 00:10:50.399 --> 00:10:53.759 their effect at the macroscopic level. The big picture, right, 208 00:10:54.039 --> 00:10:58.120 that macroscopic effect is the fluid's movement itself. You describe 209 00:10:58.120 --> 00:11:02.320 the water's flow using concepts like pressure, density, discosity. These 210 00:11:02.399 --> 00:11:05.559 things emerge from those underlying atomic interactions. But you don't 211 00:11:05.559 --> 00:11:08.799 need to know the exact position and velocity of every 212 00:11:08.840 --> 00:11:13.480 single molecule. You've effectively distilled the essential large scale behavior. 213 00:11:13.519 --> 00:11:16.559 Okay, that makes sense. You capture the overall flow without tracking. 214 00:11:16.240 --> 00:11:20.320 Every draw precisely. Now draw the direct parallel to the universe. 215 00:11:20.440 --> 00:11:22.960 The water in this example is the universe on very 216 00:11:23.039 --> 00:11:26.759 large scales, the cosmic web itself. The microscopic components are 217 00:11:26.799 --> 00:11:31.000 the small scale physical processes that influence that large scale structure. 218 00:11:31.240 --> 00:11:34.480 This includes things like the formation of individual galaxies, the 219 00:11:34.519 --> 00:11:39.120 incredibly complex dynamics within dense galaxy clusters, or the intricate 220 00:11:39.120 --> 00:11:43.279 behavior of gas and stars, phenomena that are just far 221 00:11:43.360 --> 00:11:47.000 too complex to simulate directly across the entire cosmos. If 222 00:11:47.039 --> 00:11:49.879 you want to understand the entire web much detail again, right, 223 00:11:49.960 --> 00:11:53.200 so Efedel files has steps in to effectively describe how 224 00:11:53.200 --> 00:11:57.360 these small scale processes collectively influence the large scale cosmic web, 225 00:11:57.399 --> 00:12:00.639 those filaments and voids, without needing to model every single 226 00:12:00.679 --> 00:12:04.159 tiny detail. It's about capturing the essential physics that shapes 227 00:12:04.200 --> 00:12:07.960 the universe on its grandest scales, providing a statistically accurate 228 00:12:08.000 --> 00:12:08.879 and predictive model. 229 00:12:09.240 --> 00:12:12.279 That analogy really helps distill the elegance and the power 230 00:12:12.360 --> 00:12:17.080 of ef FLIFLSS. It sounds incredibly effective, allowing scientists to 231 00:12:17.120 --> 00:12:20.360 make such headway in understanding the universe's grand design. It 232 00:12:20.440 --> 00:12:24.080 is very powerful theoretically, But despite its theoretical beauty, I 233 00:12:24.080 --> 00:12:27.279 imagine applying it to the universe's full scope, especially now 234 00:12:27.320 --> 00:12:30.000 with this explosion of new data we keep hearing about, 235 00:12:30.480 --> 00:12:33.840 presents some very tangible, very demanding challenges, doesn't it. 236 00:12:33.879 --> 00:12:33.960 Oh? 237 00:12:34.039 --> 00:12:35.080 Absolutely so? 238 00:12:35.600 --> 00:12:38.480 What does this all mean for actually using these models 239 00:12:38.519 --> 00:12:41.679 for processing the vast amounts of information we're collecting? Your 240 00:12:41.679 --> 00:12:45.759 analogy hinted at the computational challenge tracking every atom is impossible. 241 00:12:46.039 --> 00:12:49.080 Does that mean applying efto FLSs to the whole universe, 242 00:12:49.200 --> 00:12:53.639 even statistically, still demands a lot of time and computing resources. 243 00:12:53.720 --> 00:12:56.679 It absolutely does, And this is exactly where the theoretical 244 00:12:56.679 --> 00:13:01.960 elegance meets a very real, very practical bottleneck. While EFFLSS 245 00:13:02.080 --> 00:13:06.559 is a powerful theoretical framework, applying it directly exhaustively to 246 00:13:06.759 --> 00:13:11.399 real astronomical data is incredibly resource intensive. How intensive well, 247 00:13:11.440 --> 00:13:15.759 each calculation, each prediction from the model requires significant computational horsepower, 248 00:13:15.919 --> 00:13:19.440 often needing supercomputers running for days, sometimes even weeks, just 249 00:13:19.480 --> 00:13:20.639 to produce a single result. 250 00:13:20.679 --> 00:13:22.519 Wow, days or weeks for one run. 251 00:13:22.759 --> 00:13:26.840 Yeah, and here's the core problem. The astronomical data sets 252 00:13:26.840 --> 00:13:31.200 we have now aren't just growing, They're growing exponentially at 253 00:13:31.200 --> 00:13:35.000 a truly staggering rate. We are really entering a golden 254 00:13:35.080 --> 00:13:39.480 age of observational cosmology. We're collecting more information about the 255 00:13:39.559 --> 00:13:41.159 universe than ever before. 256 00:13:41.000 --> 00:13:43.000 Like the new surveys exactly. 257 00:13:43.360 --> 00:13:47.840 Think about monumental efforts like DSi, the Dark Energy Spectroscopic Instrument. 258 00:13:48.159 --> 00:13:51.559 It's already started releasing its first data badges mapping the 259 00:13:51.600 --> 00:13:54.480 positions of tens of millions of galaxies and quasars across 260 00:13:54.519 --> 00:13:57.000 a huge chunk of the universe ends of millions, and 261 00:13:57.039 --> 00:13:59.559 its main goal is to precisely measure the expansion history 262 00:13:59.559 --> 00:14:03.320 of the universe to understand dark energy. Then there's EUCLID, 263 00:14:03.519 --> 00:14:06.279 the European Space Agency's mission. It's coming online soon. 264 00:14:06.399 --> 00:14:07.240 What will EUCLID do. 265 00:14:07.519 --> 00:14:11.159 EUCLID promises to deliver an unprecedented volume of highly detailed 266 00:14:11.240 --> 00:14:14.919 data about the universe's large scale structure. It'll probe dark 267 00:14:15.039 --> 00:14:17.519 energy and dark matter by measuring the shapes and red 268 00:14:17.519 --> 00:14:21.480 shifts of over a billion galaxies across billions of light years. 269 00:14:21.600 --> 00:14:24.600 A billion galaxies, the data volumes must be immense. 270 00:14:24.960 --> 00:14:29.440 They are petabytes of data, so much information that are 271 00:14:29.519 --> 00:14:34.639 traditional analytical methods, even with sophisticated models like EFFOFLSS, they 272 00:14:34.679 --> 00:14:38.519 just struggle to keep up. It becomes practically impossible to 273 00:14:38.519 --> 00:14:43.519 do an exhaustive analysis, testing every possible parameter, combination, every scenario. 274 00:14:43.879 --> 00:14:46.559 Every time you want to check a hypothesis against this 275 00:14:46.759 --> 00:14:48.320 flood of new data, you. 276 00:14:48.200 --> 00:14:51.639 Need those supercomputers running constantly for weeks or months for 277 00:14:51.759 --> 00:14:55.639 each iteration, and that's just not practical for the iterative, 278 00:14:55.799 --> 00:14:59.279 explorative nature of science, especially with such huge data volumes. 279 00:14:59.480 --> 00:15:03.320 This compational barrier is a critical challenge. It necessitates a 280 00:15:03.399 --> 00:15:07.039 radically new approach, and that necessity, that computational bottle mick 281 00:15:07.120 --> 00:15:10.000 you just described, brings us directly to well a crucial 282 00:15:10.000 --> 00:15:13.320 answer that's transforming cosmology, the rise of emulators. 283 00:15:13.440 --> 00:15:13.960 Exactly. 284 00:15:14.000 --> 00:15:16.799 If our traditional models, powerful as they are, are just 285 00:15:16.960 --> 00:15:19.840 too slow to keep pace with this data deluge, then 286 00:15:19.840 --> 00:15:23.039 we need something faster. You use the word shortcut earlier, 287 00:15:23.120 --> 00:15:25.320 and that's essentially what an emulator is, isn't it in 288 00:15:25.399 --> 00:15:25.720 a sense? 289 00:15:25.799 --> 00:15:27.960 Yes, a very sophisticated shortcut. 290 00:15:28.159 --> 00:15:31.919 So a simple definition might be that emulators imitate how 291 00:15:31.960 --> 00:15:37.039 the full complex theoretical models respond to different inputs, but crucially, 292 00:15:37.840 --> 00:15:41.080 they operate much faster, much much faster. Instead of running 293 00:15:41.080 --> 00:15:43.840 that super intensive full model every single time you need 294 00:15:43.879 --> 00:15:46.960 a result, you have this high fidelity stand in, this 295 00:15:47.039 --> 00:15:51.080 digital doppelganger that gives you essentially the same answer, but 296 00:15:51.240 --> 00:15:52.200 almost instantly. 297 00:15:52.440 --> 00:15:55.759 That's the goal. Yes, high fidelity is key, and. 298 00:15:55.720 --> 00:15:59.600 This is absolutely critical given today's data volume and what 299 00:15:59.600 --> 00:16:02.440 we expel back from future surveys like DSi and EUCLID. 300 00:16:02.879 --> 00:16:05.919 As you just explained, it's simply not practical to do 301 00:16:05.960 --> 00:16:09.320 this exhaustively every time with the traditional methods. Emulators are 302 00:16:09.320 --> 00:16:13.440 becoming not just useful, but truly indispensable for modern cosmology. 303 00:16:13.519 --> 00:16:16.200 They really are. And to dive a bit deeper into 304 00:16:16.200 --> 00:16:18.919 the mechanism, what's fascinating here is that at its heart, 305 00:16:19.120 --> 00:16:22.559 an emulator is typically powered by a neural network ah 306 00:16:22.759 --> 00:16:26.279 AI machine learning exactly. This is where machine learning comes 307 00:16:26.320 --> 00:16:29.600 into play. Now, the neural network isn't explicitly programmed with 308 00:16:29.679 --> 00:16:32.200 the physics equations of the universe in the same way 309 00:16:32.240 --> 00:16:37.080 EFTO FLSs is. Instead, it learns from examples. Yeah, much 310 00:16:37.159 --> 00:16:38.960 like a student learns from practice problems. 311 00:16:39.000 --> 00:16:40.360 Okay, how does it learn well? 312 00:16:40.399 --> 00:16:44.240 The training process works like this. Scientists first run the 313 00:16:44.360 --> 00:16:49.480 full complex theoretical model like EFTO FLSs a carefully selected 314 00:16:49.519 --> 00:16:52.799 number of times, using various combinations of input parameters. 315 00:16:52.840 --> 00:16:55.679 The parameters we talked about before, like matter density. 316 00:16:55.879 --> 00:16:59.360 Right, things like matter density, dark energy amount, the initial 317 00:16:59.399 --> 00:17:02.919 fluctuation properties, expansion rate, and so on. For each set 318 00:17:02.960 --> 00:17:06.079 of inputs, the full model produces a prediction or an output, 319 00:17:06.119 --> 00:17:10.279 maybe a statistic about galaxy clustering or matter distribution. The 320 00:17:10.279 --> 00:17:14.200 neural network then learns to associate these input parameters with 321 00:17:14.279 --> 00:17:18.440 the model's already competed predictions. It builds an internal mathematical 322 00:17:18.480 --> 00:17:22.839 representation of this complex relationship between inputs and outputs. It 323 00:17:22.880 --> 00:17:23.359 finds the. 324 00:17:23.319 --> 00:17:25.519 Patterns, so it learns the input output. 325 00:17:25.200 --> 00:17:29.000 Map exactly After this intensive training phase, which might take 326 00:17:29.039 --> 00:17:32.920 a significant amount of computing time itself. Mind you, the 327 00:17:32.960 --> 00:17:37.559 neural network, now, functioning as an emulator, can generalize to 328 00:17:37.640 --> 00:17:39.640 combinations of parameters it hasn't. 329 00:17:39.319 --> 00:17:42.200 Seen, so it can predict new scenarios. 330 00:17:42.359 --> 00:17:45.119 Yes, if you give it a new set of cosmological 331 00:17:45.119 --> 00:17:48.680 parameters it wasn't explicitly trained on, it can quickly and 332 00:17:48.720 --> 00:17:52.200 accurately predict what the full theoretical model would have output it. 333 00:17:52.279 --> 00:17:55.839 That's powerful, it is, But there's a crucial distinction though. 334 00:17:56.400 --> 00:18:00.400 The emulator doesn't understand the physics itself, not in the 335 00:18:00.400 --> 00:18:02.839 way a human physicist does, or the way Eve of 336 00:18:02.880 --> 00:18:05.599 two LSS explicitly encodes it. 337 00:18:05.599 --> 00:18:07.119 It just knows the pattern right. 338 00:18:07.160 --> 00:18:10.240 It knows the theoretical model's responses very well and can 339 00:18:10.319 --> 00:18:13.440 anticipate what a would output for a new input based 340 00:18:13.440 --> 00:18:16.480 on the patterns and relationships it learned during training. It's 341 00:18:16.519 --> 00:18:20.720 like a highly skilled mimic reproducing the results without necessarily 342 00:18:20.759 --> 00:18:23.680 comprehending the underlying process. That makes a lot of sense. 343 00:18:23.720 --> 00:18:26.720 It learns the relationship the pattern rather than solving the 344 00:18:26.759 --> 00:18:31.559 equations directly, and that's incredibly powerful for speed. But okay, 345 00:18:31.599 --> 00:18:34.119 since this is a kind of shortcut in imitation, it 346 00:18:34.200 --> 00:18:38.880 raises a really critical intuitive question listeners might have. What's 347 00:18:38.920 --> 00:18:43.680 the risk of losing accuracy? Are we sacrificing precision for speed? 348 00:18:44.119 --> 00:18:45.880 That's the million dollar question, isn't it? 349 00:18:45.920 --> 00:18:48.880 Because when we're talking about mapping the universe. Good enough 350 00:18:49.400 --> 00:18:50.559 isn't really good enough? 351 00:18:50.640 --> 00:18:50.880 Is it? 352 00:18:51.000 --> 00:18:54.079 We need it to be fundamentally correct, scientifically robust. 353 00:18:54.200 --> 00:18:57.960 Absolutely, and that's a completely valid and crucial concern. It's 354 00:18:58.039 --> 00:19:00.920 one that the developers of these emulates spend a huge 355 00:19:00.920 --> 00:19:03.960 amount of effort addressing. The whole point of these advanced 356 00:19:04.000 --> 00:19:08.559 emulators is precisely to lighten the analysis without losing precision. Okay, 357 00:19:08.880 --> 00:19:11.599 it's not about cutting corners in the science. It's about 358 00:19:11.599 --> 00:19:14.960 finding a computationally efficient way to arrive at the same 359 00:19:15.480 --> 00:19:19.640 high precision answers that the full theoretical models would provide. 360 00:19:19.880 --> 00:19:24.200 Rigorous validation of this precision is paramount. Scientists simply cannot 361 00:19:24.240 --> 00:19:27.240 accept a tool that gives faster answers if those answers 362 00:19:27.240 --> 00:19:28.880 are wrong or less precise. 363 00:19:29.000 --> 00:19:30.240 So validation is key. 364 00:19:30.359 --> 00:19:33.359 Absolutely, And this commitment to accuracy is a key part 365 00:19:33.359 --> 00:19:36.160 of the groundbreaking work we're diving into today, as it's 366 00:19:36.200 --> 00:19:38.319 been robustly demonstrated. 367 00:19:37.839 --> 00:19:42.200 And that brings us seamlessly to the specific groundbreaking emulator 368 00:19:42.240 --> 00:19:44.799 that's really at the heart of this deep dive effort 369 00:19:44.839 --> 00:19:45.720 nott jail. 370 00:19:45.720 --> 00:19:47.039 Yes, effort, dot jail. 371 00:19:47.160 --> 00:19:50.200 This isn't just a theoretical concept or some lab experiment. 372 00:19:50.559 --> 00:19:54.319 It's a real published tool making significant waves in the 373 00:19:54.319 --> 00:19:55.880 cosmology community right now. 374 00:19:55.920 --> 00:19:56.440 It really is. 375 00:19:56.559 --> 00:19:59.200 It was developed through a truly collaborative effort, wasn't it 376 00:19:59.240 --> 00:20:03.720 an international team researchers from PINAF in Italy, the University 377 00:20:03.720 --> 00:20:07.519 of Parma, also Italy, and the University of Waterloo in Canada. 378 00:20:07.640 --> 00:20:09.480 That's right, a strong collaboration, and. 379 00:20:09.480 --> 00:20:13.599 Their work, titled Effort dot JL, a fast and differential 380 00:20:13.599 --> 00:20:16.319 emulator for the effective field theory of the large scale 381 00:20:16.359 --> 00:20:19.319 Structure of the Universe, was published in the Journal of 382 00:20:19.319 --> 00:20:22.799 Cosmology and Astroparticle Physics, a top journal indeed. 383 00:20:23.279 --> 00:20:26.559 So stepping back, the broader significance of effort dot JL 384 00:20:26.759 --> 00:20:31.200 lies in its primary game changing achievements, which really speak 385 00:20:31.240 --> 00:20:33.039 to that balance of speed and precision we were just 386 00:20:33.039 --> 00:20:33.519 talking about. 387 00:20:33.559 --> 00:20:34.759 Okay, what are the highlights? 388 00:20:35.079 --> 00:20:39.599 First and foremost, it delivers essentially the same correctness as 389 00:20:39.640 --> 00:20:44.240 the full complex ef to FLSs model it imitates. This 390 00:20:44.359 --> 00:20:47.440 is critical. It means the shortcut doesn't compromise the scientific 391 00:20:47.480 --> 00:20:49.559 integrity or the precision of the results. 392 00:20:49.640 --> 00:20:52.079 Same accuracy. That's huge, it is. 393 00:20:52.279 --> 00:20:55.359 What's even more impressive, perhaps is that in some scenarios 394 00:20:55.640 --> 00:20:59.240 it actually achieves finer detail than the original model could 395 00:20:59.279 --> 00:21:00.680 manage in a practice timeframe. 396 00:21:00.880 --> 00:21:03.000 Finer detail how so well. 397 00:21:03.039 --> 00:21:06.559 For instance, when trying to model the intricate gravitational dance 398 00:21:06.599 --> 00:21:10.960 within a really dense galaxy cluster or the precise distribution 399 00:21:11.119 --> 00:21:14.759 of matter within a cosmic suliment. Traditional, if to Fliss, 400 00:21:14.839 --> 00:21:17.440 might have had to average over certain small scale interactions 401 00:21:17.799 --> 00:21:21.119