Scientists have begun to leverage the power of AI to help us better understand the formation and dynamics of the universe. Paul Torrey is a computational astrophysicist with a research focus on the formation and evolution of our cosmos. Torrey’s research group builds, runs, and analyzes large-scale cosmological simulations that allow us to examine in a new way our assumption of the universe.
- Transcript
- 0:06
welcome to Who's in STEM i'm Ken Ono your host and the STEM adviser to the Provost and the Marvin Rosen Blum 0:13Professor of Mathematics at UVA this is the first episode of our third season 0:18the 34th episode and this is an episode that I'm really excited about because it 0:24reminds me of my youth our goal is to evoke flights of imagination and wonder by showcasing the cornucopia of all that 0:31is STEM at UVA the marvelous world of UVA science technology engineering and 0:37mathematics we're going to begin this episode with a special tribute to one of my first scientific heroes Carl Sean 0:46master astronomer planetary scientist and science communicator extraordinaire 0:53we inhabit a universe of galaxies there are unstructured blobs the irregular 1:00galaxies globular or elliptical galaxies and the graceful blue arms of spiral 1:08galaxies we've been investigating the galaxies their origins evolution and motions for less than a century these 1:16studies extend our understanding to the farthest reaches of the universe 1:23our ship of the imagination carries us to that ultimate frontier we view the 1:29cosmos on the grandest of scales the majesty of the galaxies is revealed by 1:36science sean was one of my heroes i loved his books and was a huge fan of 1:42his TV show called Cosmos because of his show I wanted to grow up to become an 1:47astronaut i wanted to be like Neil Armstrong as a kid my parents bought me a 1:53telescope from Toys R Us and I spent many summer nights in the 70s gazing at 1:58the moon and the nearby planets now make no mistake telescope sounds like fancy 2:04instrument but my telescope was little more than a white cardboard tube on a tripod fashioned with an eyepiece and 2:12two mirrors scientists you see have figured out that when light enters the tube and is reflected off the curved 2:19mirror and then onto a second mirror which then bounces the light into the eyepiece you could actually see the moon 2:26you could see the craters of the moon you could see the planets like Mars from your own very 2:33backyard now honestly it barely worked but it worked well enough to inspire a 2:39kid like me as I said I spent summer nights peering at the nearby planets and 2:45the craters of the moon now fast forward to today and it's now 2025 make no mistake astronomy today 2:53isn't the stuff of cardboard reflector telescopes and on a recent episode of 2:58Who's in STEM we talked about uh the work of UVA professor Brad Johnson and 3:04his work with the Simons Observatory in Chile you see there are many many kinds of telescopes now right you've probably 3:11heard the names Web Hubble and now this very special telescope that Brad is 3:17helping to construct in Chile in his work he's actually using his telescope to look back in time we talked about 3:25that so I invite you to go back and listen to that episode to learn more about it but just in a sentence he's 3:30trying to understand the origin of the universe by studying what's called the cosmic microwave background it's a great 3:37story but there is so much to astrophysics and astronomy here at the University of Virginia and actually in 3:44Charlottesville we'll get into that in a moment and I want to talk today about 3:50how the advances in computing and artificial intelligence is having an 3:56extraordinary impact on astronomy and astrophysics you see thanks to technology scientists 4:04have amassed extraordinary complex and massive data sets about the universe 4:09it's so massive that we need new science just to study it to process and analyze the data and where to 4:17turn well scientists have started to leverage AI physicists astronomers data 4:23scientists and statisticians are coming together to do a new kind of science we've actually already seen this before 4:30last fall two Nobel prizes were awarded based on science that made use of 4:36artificial intelligence a first this trend is continuing across all of the 4:41sciences and today we're talking about astronomy and astrophysics i'm talking about a new 4:48consortium called the NSF Simons AI Institute for Cosmic Origins or Cosmic 4:54AI and it's a pleasure to welcome one member of the team UVA's own Paul Tory a 5:01new assistant professor of astronomy and astrophysics and Paul it's a pleasure to have you here on who's in STEM thanks 5:07Ken pleasure to be here so we've been talking about AI for quite some time now 5:14and I think for most listeners when we talk about AI what pops to mind it's probably Siri or Chat GPT or perhaps uh 5:22the new school of data science here at the University of Virginia today I want to talk about astronomy i want to talk 5:28about astrophysics i want to talk about cosmic AI so Paul tell us about it yeah 5:34thanks Ken to start off we're super excited about it uh this new AI institute cosmic AI uh is funded by the 5:42National Science Foundation and the Simons Foundation and that's actually the same Simons as the Simons Observatory jim Simons he's done 5:49fantastic things and his foundation uh continues to do great work and so Cosmic AI is is an institute which is set up to 5:57try to bring AI into astronomy really that's the foundational goal is what we want to do is build an ecosystem where 6:04we can not only leverage AI for astronomy but then also use the use 6:09cases in astronomy to drive foundational changes or foundational advances in AI 6:14and as you're saying it's it's a remarkable time that we live in uh right now we're seeing uh huge advances happen 6:21in AI we see it happen basically every day but even with these advances there's still an open question which is to what 6:28extent will those uh techniques or approaches that are being developed 6:33actually be brought to bear to help us advance science and that's where this investment comes in that's where cosmic 6:39AI comes in we're a collaboration between a large number of scientists 6:44spread across many different institutions and so of course that's including UVA here it also includes UT 6:50Austin University of Utah UCLA but then we also have a series of uh federally funded facilities so specifically NRO 6:59which is the National Radio Astronomical Observatory that's actually headquartered here in Charlottesville right on grounds right so for listeners 7:07who often drive to Crosse on the way not far from Bors Head is a super important 7:13institute that's exactly right yeah this is a site where a significant fraction of the US's radio telescope resources 7:20are managed and in fact there's even another site uh which is even uh kind of embedded right within campus just a 7:26minute or two walk well five minute walk from the astronomy department we also have the Nor lab which is the optical 7:32and infrared counterpart of NOO and Slack and amongst these institutions there's a huge number of scientists that 7:39are spread between different fields so of course we've got a core of astronomers but then we also have data 7:45scientists computer scientists statisticians mathematicians and so on and so forth basically it's a it's a 7:50team that's assembled with the goal of being able to think about problems and how we can bring AI to bear on them 7:58while also driving forward the state of the field but cosmic AI efforts are much broader than that two other areas that 8:04are heavily represented here on grounds at UVA uh are surrogate model and 8:09explainable AI inference so Paul for most users of large language models chat 8:16GPT the experience is that uh the large language model is an amazing resource 8:22right it's it's something like being able to access the accumulation of human knowledge instantaneously but if I 8:28understand what you're saying correctly you're definitely going beyond that and this is an issue that is very much in the air we talk about AGI right uh 8:36generative intelligence is what you're describing something more along those lines so instead of just being a 8:42co-pilot that helps you do literature searches you're able to begin to simulate 8:48uh the phenomena that you want to study in in in the universe is that is would that be accurate so for right now the 8:55the large language models are restricted to the domain of really helping us to 9:00parse information in fact the way that I generally think about them is that where large language models really excel is 9:07that they're excellent at organizing information they help us parse through disperate different forms of information 9:14that can be collected in different places and allow us to query it in a much more natural way now we do also 9:20have tools that can help us with carrying out simulations and in fact that's where some of these surrogate 9:26models actually come in uh but at this stage we're not yet integrating them into the LLMs so for the surrogate 9:33models you know those do have the ability to enhance our simulations these surrogate models you know classically if 9:40I want to run a simulation you know let's say I'm thinking about gravity that's important for a lot of astrophysical topics if I want to run a 9:47simulation what do I do well I write a model i write down a model where I can update in time the positions and 9:55velocities and accelerations of a bunch of mass so if I care about gravity my positions get updated by my velocities 10:01my velocities by my accelerations and those accelerations are fundamentally set by the physics that's driving the 10:07system so in this case gravity but that involves the solution for a bunch of 10:14differential equations and we can do this it's no problem we've been doing it for for decades surrogate models in some sense 10:21offer us a new option what they do is they will observe a training set as 10:26we're training them they will observe a training set and provide a model that 10:31will update the state of the system without necessarily knowing the physical driver that's moving it forward and so 10:38it actually gives us a way to approximate the behavior of gravity without directly solving the 10:44differential equations that govern it now you might ask the question why would you do this the main advantage is that 10:50those surrogate models in principle can be factors of many faster than the classical approach and so this if 10:56harnessed correctly could actually lead to massive increases in the scale and scope of simulations we try to run and 11:01for someone like myself who thinks about simulations of the whole universe an increase in the size an increase in the 11:08resolution we can achieve is actually critical towards really pushing forward the state of our field so this this 11:14reminds me um so as a mathematician I like to tell the story that uh Newton's 11:20laws of gravity famously assisted astronomers to locate Neptune locate 11:27Pluto the ma the mathematics the equations together with actual astronomical observations worked 11:35together and this is of course the mathematics of gravity but on the other hand there's this famous story of the 11:41planet Vulcan right that astronomers and mathematicians believed existed well 11:47over maybe 150 years ago because if you apply the same mathematical equations of 11:52Newton to the easier more accessible observations of the planets that uh 11:57revolve around the sun in proximity to the the sun people believed there'd be a little planet called Vulcan that was 12:04very very close to the sun orbiting even inside the orbit of Mercury and of 12:11course there is no such planet unless you a Star Trek fan and as Einstein 12:16famously uh explained understanding gravity is like one of the hardest problems in science and this is what 12:22gave birth to the theory of relativity the the the massive size of the sun exerts relativistic effects that you 12:30have to take into account so as a mathematician I have to ask is one of 12:36the moonshot problems to use your simulations to to to seek further 12:42evidence or tests for Einstein's theory so there's a lot of moonshot type ideas 12:48we have within cosmic AI and what I'll say is that uh in order to get that sort of insight you actually hit upon the 12:55next idea which is explainable inference in fact these surrogate models if we think about applying them they function 13:02somewhat as black boxes but if we actually want to derive physical insight the sort of thing that would have happened when one probed what might be 13:09going on with a Vulcan-like planet uh that's where explainable inference can come in and so explainable inference is 13:15basically now asking the question if we train our models to very accurately emulate some sort of physical system it 13:21could actually include things like Mercury going around the sun though admittedly that's not necessarily what 13:26our team's expertise falls then what you can do is you can use techniques like symbolic regression to try to actually 13:33back out the sort of physical mechanisms that are at play and are driving the system forward and this gives us the 13:38real potential for discoveries so Paul this is a bold and exciting project and 13:43I'm glad that you are here at UVA and that UVA is participating in all of this i think it's pretty clear that this this 13:51project is not only worthwhile it could really be transformational to astrophysics and I don't even really 13:58really know where to start i have so many questions but maybe let's start with the most basic of questions in a 14:04funny way I mentioned Star Trek a moment ago and what pops to mind is another film maybe most of the listeners will 14:11recognize it it was a film that starred Keanu Reeves dressed in black wearing sunglasses i'm talking about the 14:16character Neo and the the Matrix that was a film that was based on a simulation how far off from that sci-fi 14:24Keanu Reeves Neo world are we in your work in Cosmic AI you know it's it's an 14:31amusing thought um but it at the same time it's also not the worst place to 14:36start you know we're not building the matrix but what we are doing uh quite deliberately is trying to build 14:43simulations that model large swaths of the universe we're going to evolve them in time we're going to take those we're 14:50going to take these swasts of the universe we take the initial conditions from very early times building on the 14:55sort of work that Brad Johnson does looking at those early universe initial conditions in the very early times the 15:01universe was very smooth nearly homogeneous but there were slight density fluctuations and through time 15:08gravity allows those to amplify what we want to do what my group does is we want 15:14to build simulations where we can put in what we think is all of the requisite physics that help us to produce a 15:21realistic galaxy population that would match the sort of things that we actually see in the night sky and we 15:27want to allow those models to evolve from soon after the big bang to the present day in order to actually test 15:33our understanding and our models of how galaxies form so that includes of course 15:39gravity i already mentioned that but it also includes hydrodnamics the sort of stuff that stops Earth's atmosphere from 15:46just collapsing down to be a single atom thick right hydronamics are important 15:51for getting gas into galaxies we also have to include a whole host of other physics that's things like 15:57radiative gas cooling the formation of stars when where and how do stars actually form what happens when they 16:03evolve when they move off the main sequence and you might have a supernova explosion how does that impact the gas 16:09in the galaxies and then we've got these beasts at the center of galaxies super massive black holes when they form and 16:15especially when they accrete material that can inject a huge amount of energy into galaxies we have to include all of that and so my group's focus is really 16:24trying to build those models with as much realism and completeness as we can and this is important because in 16:30astrophysics unlike say biology or chemistry we can't go to a laboratory and build a star or build a galaxy we 16:36can't do that and so if we have questions about how various mechanisms 16:42might actually impact uh our model how our assumptions about gravity might actually impact our model we have to 16:49have numerical simulations that can actually allow us to to test that paul this is this is really amazing your work 16:57at least in the lab sounds a whole lot like and and is analogous to the very 17:02famous experiments that were described in this book by Oparin many years ago called the origin of life where the idea 17:09was to come up with the axioms the the the the first constituents in in the 17:14early primordial soup on planet earth right what gave rise to the origin of 17:20life what chemicals do you put in this flask before you start you know sending electrical current through it and they 17:26discovered that you could make amino acids with just the right ingredients pretty easily and and so what you're 17:32describing is something like this it's it's the thought experiment that is getting to the bottom of what are the 17:39main ingredients right what are the axioms that gave birth but realistically 17:44practically to the universe so uh I I like to draw that parallel the origin of 17:50life and the origin of the universe are are two very similar questions but we live at a time where you can actually do 17:57the thought experiment for the universe yeah that's exciting absolutely and I mean to to build on that even one step 18:03more this is the way that we actually have evidence for something like dark 18:08matter so for the listeners what is dark matter dark matter is not something you don't want to talk about no happy to 18:15talk about it yeah if we our current model of the universe includes three key 18:20components we've got ordinary matter that's the stuff that you and I are made of the sun the earth everything we see 18:27all the stars all the galaxies in the universe that's ordinary matter that is a small fraction of the composition of 18:32the universe in our current model of the universe we have a much larger fraction of dark matter which is a substance that 18:40interacts gravitationally but in no significant other way with ordinary 18:45matter and so that's that's what I mean if we talk about how do we know that exists we see its influence on galaxies 18:52we see how it actually shapes the evolution of our universe and the need for it to get the cosmic web we observe 19:00but we don't or we're not able to probe it directly using telescopes or even at 19:06this point at this point using laboratory experiments or high energy 19:11colliders so Paul you said that a small fraction of the universe is actually 19:16visible to us tangible the dark matter is is the majority what percentage of the universe is made up of dark matter 19:23what do we theorize you're in the ballpark of you know 20% 4% of the 19:29universe is observable so a diminishingly small fraction an almost embarrassingly small fraction right 4% 19:35of the universe we think is observable something like 20% or so goes into the 19:41unobservable dark matter and then the last three quarters is an even more mysterious dark energy and that forms 19:48the three components of our cosmological model we can only see the tip of the iceberg but if we want to produce a 19:55model of our universe as we see it if we want to allow it to evolve as we see it evolving and have the properties that it 20:02that it's observed to have we need these other components or our model will be inadequate so this is quite analogous to 20:08people who think very deeply about large language models you wonder about the future of science what will be the role 20:15of mankind in the future of science when large language models are ubiquitous i 20:20think it will always be true that professional scientists experts will be required to fine-tune the large language 20:27models by getting the initial input data right the initial axioms because in your 20:33case if we were to run your simulations without the understanding of the existence of dark energy and dark matter 20:40your simulations would not be compatible with the actual observations of the world we live in so let me ask for 20:46example if we were to run your simulations say without the knowledge of 20:52dark energy would your simulation arrive at a conclusion that's just absurd yeah 20:58absolutely you know dark energy that really focuses on the expansion of the universe but if I go back for a second 21:03and think about dark matter there's so many really basic factors of the the the 21:10model prediction that are impacted by its presence you know the classic signpost is that the rotation curves of 21:17galaxies don't match what you'd actually expect in a model that lacks dark matter 21:24so what I mean by this is that we can step through a galaxy and look at how much mass we observe we can count up all 21:30the stars add up all the gas mass once we know that we can use a basic gravity calculation to figure out how fast it 21:37should be rotating but it turns out that if we do that on real galaxies we find there is a 21:43mismatch galaxies are rotating faster than you would anticipate that they 21:48should be rotating if you include ordinary matter alone if you ignore dark matter this is in fact the origin of 21:55dark matter it actually wasn't intended to be an exotic title just rather some matter which appeared to be present that 22:01could not be seen dark matter is is is what it was dubbed at the time and was what stuck around that dark matter is 22:08clearly having an influence on the scale of galaxies but we now know there's many other potential tensions that would 22:15arise if you failed to include dark matter another classic example is what we call the bullet cluster there's 22:20actually two clusters of galaxies that have collided with one another they've moved towards each other and in fact 22:26have passed through each other and by looking at that system we can trace out 22:32where the observable matter is and where most the mass is and we find those two are offset they're offset in a way that 22:39would really only be the case if you actually have some sort of dark matter there and so if you think about how this 22:46is all impacted in terms of needing a scientist in the loop yeah we need someone who's actually uh being creative 22:52about what those tensions look like about what sort of physical mechanisms we can implement to potentially address 22:57them and that's exactly uh where the the cosmic AI team comes in and where I anticipate scientists will continue to 23:03come in uh as we drive forward science so Paul this is really great cosmic AI I imagine when you run it and and you do 23:10your work the output is presented to you I presume in the form of graphs data 23:15sets so on and so forth but circling back to how we started our conversation my little tribute to Carl Sean I have to 23:23wonder can you visualize this videos can you offer a simulation on screen in film 23:30in the spirit of science communication absolutely and in fact that's uh one of the best parts of the job podcast might 23:36be the the most challenging environment where I could try to describe this for you because our simulations naturally 23:42lend themselves to visualizations and we can easily render them just to maybe 23:47think back to a previous time we did this we did this with a simulation known as Illustrous illustrous illustrous was 23:54a flagship simulation uh that we uh spent a lot of time making 24:00material that could be communicated to the public what we did is we effectively put a camera inside of the simulation 24:07and we allowed it to see various fields ordinary matter dark matter magnetic 24:13field strength whatever it might be and we allowed those cameras to move to actually see galaxies as they form and 24:20evolve what this does is allows you to basically transport yourself inside of 24:25the cosmic web and see it evolve for billions of years but just in a minute or two you can actually get direct 24:31intuition for how these galaxies are assembling that's useful for scientists but it's also crucially useful for uh 24:38communicating to the public the impact from Illustrous TNG was huge right these are 24:45actually hugely important uh simulations but uh if you look at the impact from 24:50the visualizations those themselves they've gotten very far out into the field as well those visualizations have 24:56showed up on book covers they shown up in planetariums they've certainly shown up in scientific talks as well uh and 25:02the reason why is because even if you don't have an expert understanding you can quickly appreciate the complexity of 25:08galaxy formation and get an understanding about how our universe is assembling on the largest scales in just 25:14a minute or two by watching those things come to life sitting here with my hat on as as the STEM advisor to the provost i 25:20I I can imagine courses taught using your product virtual reality so on and 25:26so forth this is hard work right so in your role uh as a communicator as a as a 25:35researcher all the hats that you have to wear to perform and produce these videos 25:42for example what's involved how hard is it what do you have to do well you know it starts by developing the actual 25:48simulation itself right once we have that on disk um then we can make a lot of choices about how we actually want to 25:54how we want to visualize it how we want to see it um we we make choices about 26:00what we actually want to represent to the person who's looking through are they seeing stars or gas or dark matter 26:06um and and we also make choices you know especially if we're talking about science communication about what we're 26:11hoping to communicate what we're hoping to teach so there's great ways that we can use these sorts of animations to uh 26:19to actually teach a physical concept perhaps it's gravity perhaps it's about the evolution of the cosmos the nature 26:26of dark matter um and and you know to be quite honest you know part of what goes into it also is understanding the extent 26:32to which we're trying to teach a lesson and the extent to which we're trying to just inspire folks you know these sorts of visualizations can be great tools for 26:39simply getting someone interested and so for for an animation that we're going to make for a VR interactive portal that we 26:47want to create if we're trying to get folks interested then we're just looking to have fun and so we're looking to 26:53animate something and visualize something in a way that's immediately going to capture someone's attention so Paul how do you build Illustrous yeah so 26:59Illustrous was a huge really huge flagship uh simulation effort and and in 27:05fact the history behind it is I think at least a little bit interesting really it started with the development of a new 27:11simulation code ARPO orepo was um unique and new in the 27:19sort of hydro solver that it employed the way it solved the hydro equations of motion it was this new tool built by 27:26Vulker Spring one of my senior mentors and it used something that we we call a 27:32moving mesh right what this does is it takes advantage of all of the strengths 27:38of grid- based hydrosolvers but it allows the mesh to move and 27:44distort itself and configure itself along with the fluid flow now this is really important for a cosmological 27:50simulation for a simulation of the universe because you start with a nearly smooth distribution of mass but mass 27:56flows into galaxies as it does that the moving mesh will enhance the resolution 28:04will put those resolution elements deep into the galaxies and allow us to really refine them well while maintaining the 28:09accuracy of using a grid-based hydrosolver so I was really lucky uh early on in grad school to get involved 28:15with some initial tests of cosmological simulations using a repo once we had 28:20those initial tests showing that in fact this hydrosoverver made a nice positive 28:26impact on uh the results that we got then we set out to put in a comprehensive galaxy information model 28:32that's where we really tried to work out how we could put in a model that would 28:37reproduce the galaxies as we see them in the night sky at least roughly a and and 28:43that work which at this point was done almost a little over a decade ago that 28:48formed the physics engine that ultimately underpinned Illustrous and with some modifications its successor 28:54illustrious TNG now Illustrous itself the simulation was a huge flagship 29:01simulation effort it was tens of millions of CPU hours ran for six months on supercomputing facilities uh and it 29:09it's characterized with really two key major uh advances the first is that it 29:15was a large volume what that means is that we had a large population of galaxies we could study that came from a 29:21large range of environments with a large range of formation histories and the second was that we used ambitiously high 29:28mass resolution what that meant is that we could not just have a lot of galaxies but also resolve them pretty well and 29:35that combination that's what was huge we've made the data from Illustrous public uh which has had a huge 29:40scientific impact at this point the data from Illustrous and TNG has been used in something like a thousand peer-reviewed 29:47paper which is really a huge impact and it's not necessarily because it's the absolute best model that's out there 29:53it's a really good model but it's because the model itself can generate 29:58interesting new physical insight and because we made it easy for people to access it and so uh Illustrous really 30:04kind of came about in that entire sequence first with the development of the code initial applications of cosmological simulations and ultimately 30:11through building of that galaxy formation model that can then be deployed on a large simulation at scale 30:17so I have to ask as you did all of this work you provide this tool to the 30:23community you as a researcher how have you been able to use your own work to 30:30update our understanding of the universe you know really in so many ways we've done a ton of work with with ustrous in 30:35fact it's quite varied the things that we've actually done uh if I kind of take a step back and think about the biggest 30:40things we've learned it's probably about the nature of cosmology itself about the nature of our universe itself perhaps 30:47the biggest thing we've learned is that the lambda cold dark matter our favored model of the universe the lambda cold 30:53dark matter model continues to work pretty darn well um lambda CDM basically 31:00has these principal components that we talked about previously the ordinary matter dark matter and dark energy and 31:07while we might have some uncertainty about what those components actually are we're able to integrate them into 31:15Illustratively straightforwardly and if you take dark matter as an 31:20example cold dark matter which is our currently favored model allows us to 31:26generally reproduce the clustering of galaxies and the structure of galaxies as we see them now you ask me what we 31:33learned right and that's where I think a really key point comes in we've learned 31:38that cold dark matter does really well on large scales but that as you push to 31:44smaller scales there are so-called small scale tensions right this is basically below the scale of the Milky Way in much 31:50lower mass galaxies we find that the abundances and the structure of those galaxies may actually have tension in 31:56fact likely have tension with observations well there's a transition phase where you need a different theory 32:02right same with gravity in general can you describe in friendly terms what you 32:08mean by cold dark matter yeah cold dark matter is really it's a paradigm not a 32:14model yeah what do I mean by that so it basically has two key components the 32:20first is that that dark matter does not strongly interact with other matter 32:25except through gravity okay that's in contrast to ordinary matter as I sit here in a chair I'm running into the 32:32chair right there's electromagnetic forces that stop me from falling through this chair but you know cold dark matter 32:38is is different than that right cold dark matter interacts gravitationally but does not interact in other strong 32:45ways and so what that means is we could have cold dark matter in this room with us right now we likely do have cold dark 32:50matter in this room with us right now passing through us passing through the Earth it does that no problem now the 32:56cold part implies something about how it move well could be the mass of the particle but really how it moved in the 33:03early universe hot dark matter would be something that's a relativistic species 33:09streams very fast would make it hard for structures to collapse on small scales 33:14cold dark matter is the opposite it's a big particle moves slowly and allows for 33:20an abundance of structure even at very small scales that's the model that's currently our favored one but it's not 33:26one without tension with observations this is kind of the question what's the composition of dark matter now hot dark 33:32matter is ruled out as being a large portion of the dark matter model but it could still be some fraction of it now 33:39for listeners I'd like to talk a little bit more about the dark matter because I think this is important and not because 33:46of the physics but because one of the most important figures in this story was 33:52a leading woman in science who happened to be the mother of one of my colleagues a very famous number theorist named call 33:59Rubin who works in my field by the way all the Reuben kids ended up becoming famous scientists vera Rubin I got to 34:07know before she passed away was a dominant figure she discovered dark 34:13matter and she went on to be a leading figure for women in science she was a 34:19strong supporter for women in science and I want to celebrate her today just 34:24like we opened with Carl Sean can you tell us about Vera and how important she 34:30is to all of your work yeah what a what a tremendous figure in our field so uh 34:36Vera Rubin really was responsible for seeing dark matter when no one else did what she did was try to understand 34:44uh the influence and of of gravity on the structure of galaxies and to map out 34:50in as much detail as they could how galaxies were rotating and compare that 34:56against the mass they could actually find you know at the time that she was doing this work she anticipated that she 35:04would be able to actually account for get a total census of the material in these galaxies but she found there was a 35:11discrepancy right and and think about what you might do in that moment right an almost embarrassing discrepancy she 35:18couldn't find a significant fraction of the material right but she wasn't wrong 35:23she in fact was exactly right she had the bravery to kind of confront that and say there's something else going on here 35:28there's something missing there is another type of matter in these galaxies there is a dark matter here right and 35:35that was a seminal contribution she was able to actually identify the fact that there is something else that no one else 35:41could see she was the first one to really identify it really she changed the landscape of our understanding of the universe thank you Paul i know 35:48that's a little bit out of the ordinary for our conversations that we have on who's in STEM but uh I knew her and and 35:55I'm glad that that you raised this i think she should have won the Nobel Prize and I think there's a second dark 36:01matter here not only did she discover dark matter but there's also the dark matter that for most of her career as a 36:08woman in science she was discriminated against so thank you for sharing that with me she was an amazing woman paul 36:16this has been a great conversation while we're still on topic because we do have to wrap up very soon here I have to ask 36:24what initially sparked your interest in computational astrophysics 36:31yeah well it certainly wasn't a linear path uh you know trying to not go back too far i I always was interested in uh 36:39math and eventually physics in in high school and that drove me towards wanting 36:44to be a physics major you know that the problem was I didn't know what a physics major would do right i had no clue how 36:51this would actually lead to possible career paths and so when I went to college I went to I went to undergrad at 36:57Cornell where I was an applied physics major and that's a crucial crucial little bit of information because the 37:02applied physics majors sit in the engineering college at Cornell what that meant was that all of my mentoring and 37:09all of my advising and all of my colleagues were coming from more of an engineering background and so kind of 37:14like a chameleon I took on that uh sort of environment and my first jobs were 37:19actually engineering jobs uh my first internship was uh at Lutron Electronics 37:25helping to redesign some light switches uh and I'm I'm really grateful for that job i actually learned a lot from it but 37:31I did want to seek out something else after that that maybe would be a different application of that sort of 37:37engineering principles so I I my my next internship was over at the Jet Propulsion Laboratory where I was 37:43studying hole thrusters and we were looking at what would happen if high energy ions came out of these electric 37:50propulsion systems and wrapped around the magnetic field lines ended up coming back towards the spacecraft right this 37:56is a bad idea if you design a spacecraft that is going to etch away at itself that's a very bad idea for the longevity 38:02of that spacecraft and so uh during that internship what I really learned was that I loved the numerical modeling and 38:09I loved being curious trying to understand the system trying to model the system and so from there I took one 38:15more step this time just kind of going over the edge into just pure science i worked with Joe Burns at Cornell trying 38:21to understand the dynamics in Saturn's rings specifically the wavy edges of the inky and keeler gaps you know there's 38:28these little moons that sit in each of those gaps and as material passes by them it pulls on those on that material 38:34the moons pull on them and so you can get these interesting patterns uh and I really as I was digging into that 38:39project just had a fantastic time and and that for me that combination of experiences allowed me to begin to drill 38:45down and say I love doing numerical work i love being curious i love trying to understand interesting systems and 38:52ultimately that that's what led me down this path to the computational astrophysics I do today well thank you 38:57for sharing Paul um the work that you do is certainly sounds a lot more exciting than studying the dynamics of light 39:03switches but I'm sure that was very important so last question an off-the-wall question tell us one fun 39:10fact about yourself for our listeners so that they can have an image of you beyond computational astrophysicist 39:17well as we sit here at UVA uh perhaps interesting know I I grew up here in Virginia i grew up in Fairfax County in 39:24a place called H hearnden i love bringing this up with my classes kind of asking them where folks are from if it's big enough class there's almost 39:30certainly person who grew up not that far away from meant I grew up really admiring this this university and so I 39:36went off to to college and grad school spent some time in New England uh spent five years at the University of Florida 39:42on the faculty there but always had a real interest in in cycling back and getting back to UVA and uh really 39:49excited to be here uh really consider myself lucky to be back at the University of Virginia so that's a 39:54little fact that I think hopefully resonates with students that my roots come not far from here came out of the 40:00the public school system in Virginia uh and I'm really excited to to now be working at this university that I always 40:06grew up admiring well you may feel lucky to be here at the University of Virginia but we feel lucky to have you this is 40:12great science you're a great professor i'm sure you're inspiring to your students how can students get involved 40:18in your work come on over to the astronomy department um come to our journal club discussions we have them 40:24twice a week come to our colloquia we have those once a week come listen to what the graduate students and the 40:30postocs and the faculty are doing get to know people embed yourself in that environment we are a very open community 40:38uh and there's a ton of opportunities for folks to to learn about what's happening in the department and 40:44ultimately step into positions where they're contributing to what's happening in the department so in the astronomy department we have a long history of 40:50involving undergraduates in research we have a long history of helping students 40:56really get up to the forefront of astronomy research uh those students need not be shy just come on over come 41:02to the journal clubs come to the colloquia learn about what's happening and use that as an opportunity to get to know some people as a launching point 41:08for getting involved well Paul great this has been really an inspiring and pleasurable 41:16conversation you are fulfilling President Ryan's mission to be great and good in everything that we do and I'm 41:22Kenono STEM adviser to the provost and the Marvin Rosenlum professor of mathematics and you've been listening to 41:28who's in STEM who's in STEM is a production of WTJU 91.1 FM and the office of the provost at 41:35the University of Virginia who's in STEM is produced by Kathern Kosaboom Cla Keren Benjamin Larson Mary Gardner McGee 41:41Katie Nichols and Ria Verma our music is composed and performed by Robert Schneider and John Ferguson of Apples 41:48and Stereo follow us on Facebook Instagram and Twitter listen and subscribe to Who's in STEM on Apple 41:53Podcast Spotify or wherever you get your podcasts we'll be back soon with another 41:58conversation about scientific and technological innovation at the university [Music]