Blair Professor of Geology, Professor of Geosciences and Applied and Computational Mathematics, Princeton University
Director, Princeton Institute for Computational Science and Engineering
Former Director of Caltech's Seismo Lab
By David Zierler, Director of the Caltech Heritage Project
August 5, 2022
DAVID ZIERLER: Ok, this is David Zierler, director of the Caltech Heritage Project. It's Friday, August 5th, 2022. I'm very happy to be here with Professor Jeroen Tromp. Jeroen, it's great to be with you. Thank you so much for joining me today.
JEROEN TROMP: Oh, it's my absolute pleasure.
ZIERLER: Wonderful. Would you please tell me your title and institutional affiliation?
TROMP: I'm the Blair Professor of Geology and a Professor of Applied and Computational Mathematics at Princeton, where I am also the Director of the Princeton Institute for Computational Science and Engineering.
ZIERLER: You're on sabbatical now, on top of everything else?
TROMP: Yes, finally. This is the second sabbatical of my career. The last one was in 1998, when I went to Caltech and stayed.
ZIERLER: Jeroen, what are you working on, and is there anything specifically related to the fact that you are on sabbatical?
TROMP: Well, the sabbatical is a great time to focus on projects that you really want to bring to fruition. I started working on a monograph on theoretical and computational seismology that brings together notes from classes that I teach on these subjects. Last year, I made a real effort during the pandemic to write up a lot of my notes in a better format, and I'm hoping to take this year to whip it into proper shape for a monograph on that topic.
ZIERLER: Just an overall question; tell me about some of the big efforts in research that are happening in geosciences at Princeton. What's exciting these days?
TROMP: I assume we want to focus on my research efforts. For me the big change happened after I was at Princeton as a graduate student—I did my PhD here with Tony Dahlen and it was a totally theoretical affair. I studied the asymptotic behavior of surface waves. Since that time, I've really become interested in numerical simulations of seismic wave propagation, now culminating in actually using these numerical methods to do seismic inverse problems. If somebody had told me when I finished my PhD here that much later, we would be doing the things that we are doing now, I would have said, "No way, that is not going to happen; it's not feasible." But that's the reality of things.
I've become really interested in how you can use numerical methods and modern computers to do inverse problems using what is now known as Full Waveform Inversion, where you basically try to use every wiggle in a seismogram to try and constrain the structure. These days, that's not just in earthquake seismology. I have students and postdocs working on problems in regional seismology and global seismology that use these techniques, but I also have a couple of postdocs that are using these techniques for nondestructive testing and even for medical imaging. I have some grants now that look at ultrasound imaging, Full Waveform Inversion basically for mammography for example. I'm still extremely interested in further developing these techniques and improving the imaging algorithms. It's a very exciting, active area of research that I feel privileged to be a part of.
ZIERLER: For your research group, the name of it, Theoretical and Computational Seismology, where is the overlap between theory and computation, and where are these really distinct areas of research?
TROMP: It just reflects my research trajectory, I guess. I started out, as I mentioned, doing pure theory. My whole thesis was like that. Not an actual seismogram was in my thesis, an actual observed seismogram. When I was at Harvard and especially at Caltech, I really developed my interest in computational methods. There's this wonderful synergy between doing good, fundamental theoretical work and then taking advantage of what modern computers and algorithms have to offer to actually solve these equations computationally. That's really what I'm very excited about, that you can take the basic laws of physics, of seismology, use good numerical techniques and modern machinery, and just solve these equations. No more asymptotic methods or asymptotic approximations in that regard. I think it's an amazing time. Not just in geophysics, but even just in general in physics, this transformation has been phenomenal.
ZIERLER: Technologically, even over the past 20 years, what have been the computational advances that have made the field so vibrant? What can you do now that simply wasn't possible before?
TROMP: The first thing was the development of cheap, off-the-shelf computational hardware. That's really how I got started at Caltech. I was at Harvard at that time—I told you I went on sabbatical to Caltech—and I saw this change coming. I could see that computational science was going to be really important and that the hardware was getting there, to the point where you didn't have to go to a supercomputing center and you could build these supercomputers now, using off-the-shelf hardware. These were called Beowulf machines, where you basically cobbled together PCs to make a supercomputer. I went to talk to the administration at Harvard and I said, "We really need to have an effort focused on the computational side of things. We have theory covered. We have experiments covered. We need simulations. And that cannot be done at the departmental level. It really requires a university-wide approach to provide these resources to the community." And it fell on totally deaf ears there.
Caltech had been courting me for a while, and I had been saying no. Ed Stolper was the chairman at the time, and he's a very persistent individual. So, he sort of had this open invitation. When I started to get frustrated at Harvard, I took my sabbatical at Caltech just because I thought, "Well, I'd better just find out what would it be like to actually go there." Then I talked to Steve Koonin who was the provost at Caltech at the time. I said, "Here's what I'm thinking about computation and simulations." Koonin's reaction was the complete opposite of what I was hearing at Harvard. He said, "yeah, sounds great. Have at it. If you come here, that's what you should do." Finally, I just felt like I had no other choice. I had to make that move. I did and it was extremely important. What we did the moment we arrived at Caltech is we built this cluster in collaboration with Mike Gurnis, who is the current Seismo Lab director. It was pretty incredible what we did, in hindsight. We went and we bought I think it was 312 processors, and memory sticks, and PC boxes, and disks, and Gurnis and I, with two postdocs, Dimitri Komatitsch and Eh Tan, we literally built this thing with our hands. We put it on racks. I'm sure you've found pictures of this thing. It was in the Seismo Lab, on the floor there.
I remember we were putting that thing together and Helmberger wandered in. I don't know how well you knew him, but he sort of started to rummage around. He was looking at what I was doing, and he said, "So, in a way, this is sort of your lab," he said to me. I said, "Yeah." Then he looked up and he said—I was on the ladder putting some memory sticks in— "I'd be amazed if it actually works." And that was it, and then he walked out.
ZIERLER: [laughs] That's great.
TROMP: That was his blessing of the machine. But to his credit, once we got it up and running and it was working and we really were doing simulations, he totally embraced it. He really appreciated that kind of science and the fact that we could do these very careful accurate calculations of seismic wave propagation.
ZIERLER: Of course, this is all discussion about classical computing. A forward-looking question, given your interest in AI and deep learning and things like that, do you see a world where quantum computation might be relevant in seismology?
TROMP: I don't know. You asked about the trajectory, so that got me to Caltech, but the other main thing that happened since then is computing on graphics cards, GPU computing. We made that change too. That was a massive investment, really overhauling the code and making it portable for GPUs. That was transformative too. That gained us access to resources at Oak Ridge where we've now been since 2012, I think, so 10 years I've been using these Oak Ridge resources, calculating on these massive machines that are GPU-accelerated. Yes, I really believe in keeping up with the hardware and taking advantage of these trends, and quantum computing is one of these things that may or may not ultimately deliver; it's definitely something to keep in mind and to think about.
You mentioned AI and machine learning. That's a wonderful field and there are tremendous opportunities there. One of the things I'm actually thinking about while I'm here on sabbatical with a colleague of mine, Jeannot Trampert, who I've been visiting for 18 years in a row, we've been collaborating; he knows a lot about neural networks, and we're beginning to think about how we might do both forward simulations and inverse problems using what are called PINNs, Physics-Informed Neural Networks. People are beginning to show toy examples, where in 2D you can actually do inverse problems with these Physics-Informed Neural Networks, where the physics is one of the things that is part of the optimization in the neural network, and I see tremendous opportunities there. It shows you that this trajectory that we are on is by no means finished and who knows where it will lead in the future. The speedups that are being made, both in terms of what is happening with the hardware, and in terms of algorithms are absolutely tremendous. I just want to give one example of what is basically a transformative change in Classical Waveform Inversion. Classical, that means we're going back to 2005 when these ideas were invented. In those inverse problems, the problem scales linearly with the number of earthquakes. The cost goes linearly with the number of earthquakes. It's independent of the number of receivers, so that's great. You can have as much data as you possibly can at no extra cost, but it scales with the number of earthquakes.
In the last couple of years, we've invented a way of doing what is called source-encoded full waveform inversion, which makes the whole process also independent of the number of sources. Now, think about it; you go from a calculation that scales linearly with the number of earthquakes. Typically, our current global models use 2,000 earthquakes. You need to do 2,000 simulations at every iteration. We go to one. So, it's amazing. We do these super-simulations where we have all the earthquakes going off at the same time. All the data from all those events are assimilated at the same time. It sounds like it's not possible, right? Then, we use these encoding methods that enable us to extract using basically the Fourier components of the wave field, precisely each and every earthquake from that super-forward simulation, just using the orthogonality of the Fourier coefficients. And it works! We now go from taking one iteration per week on Oak Ridge machines to being able to do 50 iterations in a day. It's just another illustration of how things are changing very fast. It continues to be for me a very exciting time in this field.
ZIERLER: Drawing a line from the computational advances to the improvement of the research, the fundamental science, where can we take that to actual societal benefit, taking these technological improvements to improve seismic hazard or even the gold standard, perhaps dare I say one day, earthquake prediction? Where do you see trends in that regard?
TROMP: The ability to do these forward simulations faster and better, and more accurately, of course, is of great benefit to seismic hazard analysis, right? Because you can really run lots of scenario simulations. The quality of those simulations improves continually. There are definite benefits there. Personally, I've never been a believer in earthquake prediction. I have this feeling that the system is inherently chaotic and therefore unpredictable. When I was director of the Seismo Lab, I would always dismiss it and say, "Look that's not in the realm of possibilities, at least the way we currently see it." But what is in the realm of possibilities is to really do quantitative seismic hazard analysis so you can be prepared, so can really experience what it would be like to be in such an earthquake and act accordingly.
The other thing I would highlight is early warning. These are the themes that I was willing to engage in, but not earthquake prediction. I think honestly, as seismologists, we made a mistake to tout that, to float that possibility in the 1970s when we did, because now that's what people still want to hear and are waiting for, and are sort of expecting, and I'm not sure we'll ever be able to deliver. Having said that, my colleague here at Utrecht, Jeannot Trampert, he's beginning to think that again with machine learning and AI, maybe, if there is something that is a pattern that a network can latch onto and recognize, who knows? I'm willing to be agnostic, I guess.
ZIERLER: Just to go back, a few questions about your personal trajectory and interest. Even as an undergraduate in the Netherlands, were you interested in geophysics and earthquakes?
TROMP: I started in physics, and I really liked that. When I was an undergraduate, it was around the time when the U.S. decided to scrap its plans for a supercollider, and theoretical physics really started to look pretty dark. I took classes outside of theoretical physics just for breadth and to get an idea of what was happening. I took this one class by Guust Nolet, a seismologist at Utrecht, and it changed my life. It was this incredible course where he used all these tools from classical physics and did all these wonderful things that were really practical and connected to the real world. I just thought it was amazing. He was also a really approachable guy. And so it goes; I did my senior thesis with him and one of his graduate students and that's it. Everybody has a story like this, but that's how I became a seismologist.
They then told me about graduate school. I didn't know there was such a thing, but Roel had experienced this, Roel Snieder, my other advisor. He said, "There's this thing called graduate school. You apply, if you get in, you get an all-expenses paid education and I think you should go to Princeton." He had been there. He knew about Tony Dahlen, and he knew me pretty well. He said, "You would be a great match." It was the best advice. I sent in one application to go to graduate school to work with Tony Dahlen. [laughs] In hindsight, what was I thinking? I have no idea. But that's what I did, and I got in, and he was absolutely right. I was so pleased to be at Princeton as a grad student, and he was such a great, great adviser for me.
ZIERLER: What was he working on when you connected?
TROMP: Before I came to Princeton, I thought there were two Tony Dahlens. There was the Tony Dahlen who worked on wedge mechanics with John Suppe and these descriptions of critical tapers and how mountain belts form. Then there was the Tony Dahlen who worked on free oscillations and normal mode seismology. So I couldn't believe that this was a sole person —I started looking at these papers and—"It's both F.A. Dahlen! It cannot be! Are they brothers or what's going on?" It really was, of course, the same guy. But the topics were so different. He sort of divided his time half, half when I first got there, and then he sort of went back to being full-time seismology.
ZIERLER: Tell me about developing your thesis research. What were your interests? What were you exposed to that focused on what you wanted to do?
TROMP: I actually came and I had a couple of ideas that I ran by Tony and he liked one of them and so I worked on that for my first year. And then—this is how a wonderful thesis experience works—Tony made a suggestion. He said, "There's this book"—what was the guy's name? I want to say Willard or Wiggins, Wiggins maybe. Anyway, some really nice book on asymptotic methods. He said, "We should use these techniques to look at surface waves. I think it would work out very pretty." We did, and that turned out to be a really large portion of my thesis, looking at how surface waves travel in 3D models and using ray methods, caustics, and all these kinds of things. I got into Maslov theory and other asymptotic techniques that describe these sorts of problems. He was so great to interact with because we would be talking almost on a daily basis and standing in front of the blackboard, to work things out. In the end, I had an extremely productive time at Princeton, but it didn't feel like I was working my ass off; I just really loved it. I was one of the few graduate students that can say they were truly happy. I don't know what your experience was.
ZIERLER: I had a great time, too!
TROMP: But you probably also feel like most of your colleagues were miserable, or at least they acted like it, and that was certainly the case at Princeton; they were a miserable lot. They were not happy, and I was one of the few students who just—I didn't know what their problem was. I really had a great time.
ZIERLER: What do you see as your principal conclusions or contributions with your thesis?
TROMP: We did really nice fundamental work on the asymptotic behavior of surface waves in 3D models. Ultimately, we ended up putting a lot of that stuff in a monograph that Tony and I wrote together on theoretical global seismology that was sort of a treatise on every aspect of free oscillation seismology, which I really learned from him. That's something I did early on in my career, this collaboration with Tony to write this book, and it put together a lot of things, as I said, that he had worked on and that we did during my time at Princeton. I did that early when I was still an assistant professor at Harvard. I remember asking people about this and almost everybody said, "Don't do it. You're not going to get any credit for it, and it takes a lot of time and ultimately it won't look good for your package when you go up for tenure." In the end, I decided to ignore all of that advice and just do it because I really enjoyed doing it, and I knew Tony very well and I knew that we both had the personalities that once we decided we were going to do this, we were going to actually do it and bring it to fruition. And we did. In the end I have absolutely no regrets that that's what we worked on. That book is sort of a bit of a culmination of our work together.
ZIERLER: Even in graduate school, was the Department of Energy, the national lab effort in supercomputing, relevant? Was that necessary for what you needed to do?
TROMP: No. At that point I had a colleague—Tony was interested in that a little bit, and he had a student who worked on coupled mode methods and was using some DOE resources. I was not into that very much at all at that point. It's just that when I got to Harvard, after a while, maybe after a year or so, I became interested in these methods. I started to play with these numerical methods myself, initially with what are called pseudo-spectral methods, transform methods. I knew that finite-difference methods had been tried, but there were always problems with boundary conditions, and so I knew that that was not going to be the answer for global seismology. We tried these pseudospectral methods for global seismology. We worked for like a year and a half and we could do all these beautiful toy problems, but real problems like calculating synthetic seismograms for PREM or something like that was not possible. So, we never wrote anything up about it. Then, I became aware of the spectral element method and that was through work by Dimitri Komatitsch who was in Paris at the time. I really felt that that looked super promising, so I offered Dimitri a postdoc over e-mail.
I'd never met him, never talked to him. I tried to meet with him at a meeting in the Netherlands where he was presenting this work; he wasn't there. I wrote this e-mail to him afterward. I learned later that he decided to go to the beach instead of presenting his poster because he was finishing his thesis and he was only there for a day and a half, and he felt the beach was more important than his posters. [laughs] I offered him this postdoc. This must have been in May, and I didn't hear anything, so I thought, "Okay." Then in September, I get this e-mail from Dimitri—"Oh, sorry, I just found your e-mail. I defended my thesis in May." He's French, right, so he took the summer off. "I didn't look at any e-mail and I found your e-mail. Is it still good, your offer?" [laughs] I said, "Absolutely," so he came—again, sight unseen—in the middle of December, he came to Boston. That has been one of the great collaborations of my career. Despite that auspicious beginning, that really turned out to be one of the most important hires I ever made, and one of the most important collaborations I ever started. Dimitri came with me when I came to Caltech, and as I said, he was one of the two postdocs who worked on building that machine. That was a phenomenal collaboration.
ZIERLER: Did you leapfrog the postdoc entirely? Did you join Harvard as a faculty member from the beginning?
TROMP: Yeah, I did.
ZIERLER: Was that a rarity? Does that usually happen, or what was the circumstances?
TROMP: No, that's pretty unusual, even then. Now, I always kind of snicker when I hear discussions at faculty meetings about, "Well, this person is kind of young, and they're just into their postdoc" or this or that. I never say anything, but it's like, "Okay. Sometimes, you want to take a risk." Amazingly they decided to make me an offer. I was still finishing my thesis. Again, I asked for advice, and you sort of got two reactions. One is, "You've got to be nuts. If you go to Harvard, you're never going to get tenure. There are all these senior people there who will not be nice to you," et cetera, et cetera. Or, "It's a great opportunity. Go do your best and see what happens." I decided to do the latter, and again, I have no regrets. Harvard, I really enjoyed being there. It was a very supportive, interesting environment. Göran Ekström was there at the time, who was also hired as a junior person like me, and Adam Dziewoński was the senior person there. It was an exciting environment.
ZIERLER: And you got tenured. That worked.
TROMP: Yeah, I got tenure. I think I was very naive when I went there. I didn't really worry about it. I probably should have thought about these things much more. When I was almost up for promotion to associate professor, which is untenured at Harvard, Caltech offered me a tenured position. I said "No," and I stayed at Harvard for an untenured position. I don't know what I was thinking, or why I took that risk. Because it's a risk, right? At Harvard, it is really hard. When I was there, 17% of junior people were getting tenure. I remember going to see the dean when this was happening. He said, "Normally I don't meet with people at your stage of your career. You're an assistant professor and you don't have tenure. Harvard is an institution for the senior faculty." This is what he said to me. These were his words. And that's really what Harvard was like. He said, "I can't offer you anything, but I still hope you'll stay." That's sort of what I got. I stayed.
Yes, I did get tenure later, so it turned out okay. But it shows you one fundamental difference between Caltech as an institution and Harvard. Harvard really is an institution for senior people. Caltech is the example of an institution that just excels at bringing in outstanding young people, giving them everything that they need, bringing them through the ranks, and getting them to do their best work at the institution. That's my model for how an academic institution should be run. I still think Caltech is a brilliant place in this regard.
ZIERLER: Intellectually, technologically and even politically, tell me about the transition at Harvard where you realized you needed more computational powers, you weren't getting the support from Harvard, and how that factored into first Caltech's interest in recruiting you and ultimately what made that the best option for you.
TROMP: It was a very slow decision, as I mentioned. Stolper reached out to me early on in my career at Harvard. I really didn't feel like I wanted to go. I felt safe there and I was doing interesting work. I didn't want to take the risk of going to Caltech. Then they started to just be bold and made me this tenure offer; I still didn't take it. But I was sufficiently intrigued that when I finally did get tenure and I wasn't getting what I was looking for in terms of computational resources—Harvard is very much about, "What can we do for you? Can we buy you another piece of equipment?" Or this or that. I didn't want that. I decided to just go to Caltech and really, I wanted to know what it would be like culturally and what it was like to live there. Once I realized it was a great place and I knew the attitudes of Stolper and Koonin, I just felt I had to go. They made another offer. I remember Stolper came—it was at the end of my sabbatical. We were flying out of San Diego to go back to Boston. Stolper came to me at the airport. He said, "Before you go, I'm going to do this one last time, and here is what we'll do for you," and then he left. I remember turning to my wife at the time and I said, "This is it, right?" And she said "Yep." And that was it. We decided to go.
ZIERLER: That's amazing! [laughs]
ZIERLER: You were aware that you needed this computational power. You weren't getting it at Harvard. Can you just explain your access to the technology? How did you interface with these kinds of computers to know that this is what you needed to succeed in your research before you got to Caltech?
TROMP: Again, in hindsight, the risk was incredible because I knew I wanted to build a machine like that, but I had never built anything like it. We had software that could sort of run on these machines, and we had done experiments on other people's PC clusters, so we knew that we had software that could do this. But, we needed a machine. We needed a dedicated resource to be able to do this. What I asked for start-up from Caltech was money to build this damn thing. And we did! We spent, I can't remember now, almost half a million dollars on hardware—Mike Gurnis and I, together—and built this thing. I told you Helmberger's comment was. "I'm amazed it will ever work." I mean, this is 156 PCs with two processors each connected with Ethernet cables through one gigantic switch. That was the computer! And it worked.
ZIERLER: Do you have a sense of the budgetary environment at Caltech that made this such an easy offer? Do you think this was related to Gordon Moore's gift, or was this NSF-supported? What was supporting this?
TROMP: The Institute came up with this startup funding and I think—you spent that kind of money on a geochemist easy, right, if they need a mass spec or something like that. So, I think they were coming to grips with the fact that these days, even a computational scientist needs serious resources, and you need to make investments. They're willing to do that. That's where the money came from, and it paid off incredibly well. From there, we did manage to get incredible funding. We had this great collaboration with Dell for a long time. We buttered up this one guy who was in charge of scientific computing. At one point, I don't know how well you know this and whether you've ever seen a picture of it, but we had a machine in the basement; [laughs] it was incredible. It was ginormous! It was like a thousand nodes. This was for the Earth sciences, for geosciences. In the basement, that entire room was one giant computer. David Baltimore came to open up the facility! That was a combination of a donation from Dell, we got funding from I think the Moore foundation. It was incredible. The resources we managed to get were just really out of this world.
ZIERLER: Being recruited to Caltech, was becoming director of the Seismo Lab at some point in the future in the cards? Was that discussed? Was that part of what you were aiming for in coming here?
TROMP: Yeah, I was definitely interested in trying that and Stolper made it sort of a gentlemen's agreement. He said, "I want you to do that; I think you'd be great. I think we shouldn't do it right off the bat. You should get a feel for the place for a bit and then you can take over." That happened, I think a year and a half or two years after I was there, in consultation with everybody else. I felt comfortable with it. So, it was always sort of an unspoken or unwritten agreement.
ZIERLER: Even as an undergraduate when you first got interested in seismology to Princeton, to what extent did the Seismo Lab loom large in your imagination? Richter, Gutenberg, Benioff, Press, were these names that you appreciated? Did you understand what the Seismo Lab represented earlier in your career?
TROMP: Oh, absolutely, yeah. For me especially as an undergraduate looking at the world, it was Freeman Gilbert and George Backus at Scripps, the Seismo Lab, Kanamori, Anderson. Then John Woodhouse and Adam Dziewonski at Harvard. Well, and Tony Dahlen, Princeton. Those four places loomed large in my view of the world at that time.
ZIERLER: After Helmberger's blessing that this wouldn't work, what happened next? When did it start working?
TROMP: Oh, basically the moment we powered it up, it worked right. It worked as advertised. We had really done our homework right, and we knew how it needed to be built. There was an example of such a machine at Caltech. They were really the first pioneers, at CACR, to build one of these things. They had one of the prototypical Beowulfs. So, it's not like we're doing something that wasn't going to work. It was a very well-calculated investment and guess. It's really great that it did work the way it did, but it should. Nevertheless, it's a lot of moving parts on a machine like that, so we were always rebooting disks or doing practical things like replacing fans [laughs] and stuff like that. The machine only lasted 3 years, I think. At that point, you just have to let it go. We sold it to some school, I think; I can't remember. It has always been like that. The life expectancy of hardware like that is three, four years tops.
ZIERLER: It is a different stage in your career, a different perspective to what is happening around you, but comparing geophysics and seismology from Princeton to Harvard, what was different about Caltech, just the research culture, the interactions between faculty and students?
TROMP: Oh! Yeah, Caltech is so interactive and collaborative. Just the Seismo Lab coffee; twice a day, you have a chance to hang out and see each other. I don't know how much they still do that, but it was a very collaborative, interactive environment. I think for the right student, it is just amazing. The resources, whether it is finances, or intellectual, were sort of limitless, and the breadth of effort, too. The Seismo Lab at the time was like 11 or 12 faculty. I came from Harvard when there were maybe four geophysicists, five. So, it's a big lab with a lot of activity. The Division is even bigger, right? There too, I really felt it was very collaborative and supportive. The various groups really wanted the other groups to be successful. The emphasis really was on hiring quality young people, early-stage career. We'd hem and haw over those decisions, again, but then once they were made, done. That person was expected to succeed and given everything they needed to do so. Those meetings were also amazing. We would meet in the evening at Stolper's house. At 7:00 o'clock, all the faculty would be there in his living room talking about just this, making a hire.
ZIERLER: When it was time to become director of the Seismo Lab, to what extent were you aware of the history of past directors, the things that they had done, and to what extent did that influence the imprimatur that you wanted to have as director?
TROMP: Yeah, I was very aware of the lineage, in particular, Hiroo's influence. Anderson before that had done it for a very long time. I remember these stories that at some point there was one grant that supported the entire lab. One NSF grant. I mean, those were the days, right? What I lament most these days about being an academic is the amount of time and effort I need to spend writing proposals, trying to generate funding. In many ways, I feel it's a waste of time. I'd much rather focus on doing research or teaching or working with students. Those were very different days, clearly, right? They were just doing what they wanted to do. The money was there. They were of course not where they are now; at the different location where the lab used to be.
Hiroo was my next-door neighbor during my eight years at Caltech, so I knew him extremely well and we would always hang out together and talk. Even though we didn't really do that much research together, I always used him as a sounding board. I did that too if there were ever any issues related to being director. I tried to be a very available and approachable director who just tried to do what was best for the lab, listening to what people were looking for and providing the resources that were needed. Of course, the really important part has always been the network. I was very fortunate that Egill Hauksson at that time, was running it. He was doing it very efficiently. I think we had a very good working relationship as did Hiroo and Egill. That made that part of the job pretty straightforward and pretty easy. I was grateful for that.
ZIERLER: When did you start thinking about leaving Caltech, given what a great fit it was for you initially?
TROMP: It's almost entirely personal in nature, in that I traveled a lot while I was at Caltech. I flew typically 100,000 miles a year or something like this. Every time I flew into LA, I would come in and we'd come in over the San Bernadino Mountains most of the time, and 20 minutes later, you're still flying over Los Angeles. I would think, "What am I doing here?" I'm a Dutchman. Where I'm calling you from is my mother's house, which is where I spent most of my time growing up. It's five villages with 2,000 people, very rural. I'm not really a city guy. LA always sort of bothered me as a place to live, which is why I was very reticent to go to begin with. Nine months without a drop of rain for a Dutchman is almost intolerable.
TROMP: That's the only thing about my experience there that I did not take to. I did not agree well with Southern California and Pasadena. I tried to make the very best of it like everybody does. When I got a chance to come back to the East Coast and Princeton when Tony Dahlen was ill and then passed away and Guust Nolet was retiring, I sort of saw an opportunity to live a different life and still do top-notch science just in a different setting. That's why I left. There were some things I did not like about how Caltech was going about computing, and partly I'm to blame in that I embraced this model where we were putting the computational resources in the Lab and in the Division. These days, that's not very sensible anymore. You really want to have one central place for a university where you build and maintain these systems. These days it's not just about computing; it's also about the human resources. For example, at Princeton now as part of this Institute for Computational Science and Engineering that I direct, we not only provide computational hardware, but we also now have more than 18 research software engineers who help faculty on campus with software challenges. I really believe in this model where all the resources, human and otherwise, are provided centrally and distributed like that. Then, there's so much more bang for the buck.
Caltech in a way could have that through CACR, but this is where it became a little bit more like Harvard, where Chemistry, Physics, and Geoscience all had their own computational efforts, and we all felt like we needed to hold on to those. I think in the end it was a mistake. When I got a chance to be a part of starting that, sort of, from scratch at Princeton and doing it the way I was beginning to see it needed to be done, that was exciting. In the end and I can look back on now more than 12 years here, 14 years almost now, and really, we have made an incredible transition in this regard. That has been really good also for me as a scientist and as a researcher. We've been able to take advantage of those developments very effectively. This is all just to say that I left Caltech not easily and very reluctantly, not because I was unhappy there scientifically in any way, shape, or form, more because of other more personal reasons.
ZIERLER: As I'm sure you know, it has been a long-standing goal at Caltech to secure an endowment for the Seismo Lab. Were you involved in that effort at all as a director?
TROMP: Oh, very much. We tried so hard, Stolper and I. There were a number of individuals that we thought might be targets. I'd go visit these people, spend time with them, and try to butter them up. It never materialized. We always felt like, "How can this be? It's such an institution. We have such a well-defined need for long-term resources to support the network, support efforts in early warning, to support the computational things that we do." And just the postdocs and the students. We never managed to pull it off. It's one of the regrets of my tenure as director that I did not manage to make it happen.
ZIERLER: Caltech has had such success with endowments in so many other regards. What is it about seismology? It's such a proud history at Caltech. What makes the endowment so elusive, would you say?
TROMP: I don't know. I honestly don't know. I've always felt that if we had the right person who saw or who really could appreciate everything that goes on in that lab, it should be possible. But it's 14 years later, and it still hasn't happened, so it's not just me. [laughs]
ZIERLER: Right. [laughs] Don't take it personally. As director, what opportunities did you have for collaboration at the institutional level with the Survey, with the USGS?
TROMP: That was always a wonderful connection too. Lucy Jones was in charge at the time. Again, we had excellent connections. I remember the 2004 Sumatra earthquake happened during my tenure as director. Lucy and I said, "We have to do something, right? This is such an amazing earthquake, the first nine since 1964. We've got to do something." We decided to put on a show at the Beckman Hall on the campus, in the evening, sort of telling stories about this earthquake, what we had learned up to this point. It was open for the public. I couldn't believe it; that room was packed. People were so interested, and they really came out in droves to hear about this earthquake. The USGS and the Seismo Lab, we put on this show. It was really great. It's just one illustration of how effective we would work together. The Media Center within the Seismo Lab, we managed to get that completely renovated. Yeah, so that connection with her, and then Egill, and then the Lab; those three things that were an integral part of running the operation were super smooth and super easy.
ZIERLER: There's a long trajectory from the Seismo Lab many decades ago being of course totally a boy's club to where it is today—many women graduate students, several women tenured faculty members. Where do you see your time as director on that bigger transition to making it a more diverse place?
TROMP: While I was there, Joann was there already, but since then, during my time there, we brought in Jennifer Jackson and Nadia Lapusta. Already then, we had a really good balance, I think, of male and female graduate students, and it was a pretty amazing international affair. To be honest, I don't know what things look like these days in the Lab, but while I was there, I had probably an even number of male and female students and postdocs. Qinya Liu went on to become a very distinguished professor at the University of Toronto. Min Chen went on to have a great career at Rice. Unfortunately, she committed suicide while she was a professor at MSU. Vala Hjorliefsdottir was a female graduate student who is now back in Iceland. Christina Morency, she was a postdoc with me, now at Livermore Lab. Alessia Maggi was a postdoc; she's now is at the University of Strasbourg. Ying Zhou, she was a postdoc with me and went on to become a professor at Virginia Tech. I can go on. It was already pretty well balanced, I think, at that time and I hope that that has continued to be the case.
ZIERLER: What were some of the most significant graduate students you had when you were director?
TROMP: Carl Tape, of course, was one of my very successful graduate students. He really was the one who brought the inverse methods that we still continue to use to this date to early fruition. He did the first inversions, constructed the first model of Southern California. He used that PC cluster and ran it into the ground. I don't know if you talked to him. That was pretty incredible, what he managed to get out of that machine. Dimitri Komatitsch was my postdoc. He's the one who came with me and helped build the machine. He went on to a very illustrious career in France, extremely successful, but unfortunately he too ended up committing suicide. So, I've lost one graduate student and one postdoc that way.
ZIERLER: Have you kept up with the Seismo Lab over the years? Have you been in touch with the faculty? Are you aware of what's been happening?
TROMP: Not directly in the Lab itself, but of course I am in touch with people who work at the Lab all the time. I see them at meetings. Not so much of late, right? It has been a very strange time. But of course, I know most of these people really well and continue to be very impressed with everything that is happening in the Lab. But it's been a while since I've visited. I would like to come back sometime and see it again. I was there for some meeting in honor of Hiroo when he, I forget what he did, maybe he officially retired. [laughs]
ZIERLER: He still comes in every day, so retirement doesn't mean anything to him.
TROMP: I'm well aware of it. I used to joke that—I would come to my office around 8:00 and I would leave around 5:15, 5:30. When I came there, Hiroo was there; when I left, Hiroo was there. So I have no idea when he actually came to work or went home. [laughs]
ZIERLER: If not the endowment of course, what are you most proud of in your tenure as director of Seismo Lab?
TROMP: I think what worked well is how we developed the Seismo Lab into a real powerhouse in computational geophysics, and we provided the computational resources to do that. I think we also managed to maintain and continue to build out the network very well. While I was there, there was not a single really large earthquake that actually meant everybody would show up. I was director for five years; we didn't have any serious earthquakes during that time. To be [laughs] perfectly honest with you, I didn't mind. Maybe it's a professional hazard, but I used to think about earthquakes all the time while I was there. I would drive around, and I would sit under one of the overpasses. You know how they set up the lights, where you're under the overpass waiting for the light before you can go onto the freeway? Invariably, as I was sitting there, would think, "Not a good time for the big one." [laughs] I don't mind having not lived through the big one while I was there as director.
ZIERLER: Finally last question, looking to the future. It's as much a question about how you define your research agenda as it is about extrapolating technology trends in the field. Where do you see your research going five, ten, fifteen years out from here?
TROMP: We sort of touched on that at the very beginning when we talked about computational trends and these kinds of things. Where I see the real opportunities is in, again, harnessing what's happening in computational hardware, and combining that with what's happening scientifically. The example we talked about is PINNs, Physics-Informed Neural Networks. That looks super promising to me. You combine that with other advances like this source encoding that I mentioned that makes the process of doing inversions independent of both sources and receivers. That speeds things up beyond your imagination. I think being able to do things that fast, it just means that the quality of the images we will be able to produce and the resolution of the images will just be tremendous compared to where we are at this stage.
I think this will be a really exciting time also for graduate students and postdocs. They will be able to take advantage of all of these things and capitalize on all of this progress. That will bring opportunities for more complex physics. We are looking at anisotropy, so the directional dependence of seismic wave speed, but I think we can really do that much more accurately. The same thing with attenuation, so the dissipation of seismic energy. We're just beginning to scratch the surface of trying to infer what these variations look like three-dimensionally, say, inside the Earth's deep interior. I think all of those things are sort of almost there. It's going to happen in the next—hopefully before I retire, we're going to see tremendous progress in all of these areas.
ZIERLER: This has been a great conversation. It'll be a terrific addition to the Seismo Lab history project. So glad we connected. Thank you so much for doing this.
TROMP: It's wonderful that you're doing this.