Andrew Childs (BS '00, Postdoc '04-'07), Computer Scientist and Quantum Physicist

Arriving as an undergraduate at Caltech in the mid-1990s, Andrew Childs found himself in a unique position. At a time when quantum information was just emerging as a new area of physics research, and for most students who don't have well-formed ideas about what they want to pursue as they begin college, Childs had already learned about the possibilities of quantum computers and he knew this was an exciting path worth pursuing.

In the discussion below, Childs reflects on the special opportunities he had as an undergraduate to experience the early formative years of quantum information research, and how John Preskill was generous with his time and included Childs in all the exciting work in this new field. Following his graduate work in quantum algorithms at MIT, Childs returned to Caltech, where the quantum effort had been formalized into the Institute for Quantum Information (before the M, "Matter" was later added to create IQIM).

Now at the University of Maryland, Childs has taken a leading role in making College Park a "hub" of quantum information work, where several interdisciplinary centers on campus are partnering across a large landscape of government and private enterprise initiatives to transform the fundamental research aspects of quantum information into scalable and societally useful quantum computers. As Childs indicates, exactly what this will look like and when it will be achieved remains a matter of prediction, and in the meantime, there remains much important work - best done in the spirit of collaboration - to be done at the leading edge of both physics and computer science.

DAVID ZIERLER: This is David Zierler, Director of the Caltech Heritage Project. It is Wednesday, January 19th, 2022. I'm delighted to be here with Professor Andrew Childs. Andrew, great to be with you. Thank you so much for joining me today.

ANDREW CHILDS: Thank you. It's great to talk with you, David.

ZIERLER: Andrew, to start, I know this is going to be a complicated answer, but can you give me your titles and institutional affiliations?

CHILDS: Yeah. It's probably more complicated than it needs to be. So I'm a professor in the Computer Science Department at the University of Maryland. And there's also a kind of interdisciplinary computing institute at Maryland called the Institute for Advanced Computer Studies that I'm also a part of. And I co-direct something called QuICS, the Joint Center for Quantum Information and Computer Science, which is a center that sits within UMIACS at the University of Maryland but is also a partnership with NIST. And I also recently started directing a new institute that we just launched. It's the NSF Quantum Leap Challenge Institute for Robust Quantum Simulation. It's an institute that spans five universities, headquartered at Maryland, about all things related to quantum simulation.

ZIERLER: Now your home department is computer science?

CHILDS: Computer science. Yeah, that's my main department.

ZIERLER: Do you have a joint appointment or a courtesy appointment with physics or how does that work?

CHILDS: Yeah. At Maryland, we call them affiliate faculty appointments. It's like a courtesy appointment in the Physics Department at Maryland. I also have a similar kind of connection with the applied math graduate program here. So yeah, I have lots of connections to lots of different things.

College Park as a Quantum Hub

ZIERLER: Andrew at a broad level, the tweet you sent out on January 13th announcing all of these new positions, this has created quite a stir. I wonder if you can explain what is so exciting, what is all of the growth that's happening at UMD, and what are the funding sources that are making this growth possible?

CHILDS: Yeah. Well, Maryland has a long history in quantum information and has had strength in this area long before I was here. And I think that's part of the excitement, that there's already a lot going on. But there's a commitment to continue to grow and to pursue new research directions and these kinds of things. Our most recent hiring in quantum information has been funded by the state government.

ZIERLER: Because you have a background in physics this is really a generational question, where 20 years ago, 25 years ago if you were interested in quantum information, the department of computer science would say maybe that's really more physics, and you would go to physics, and they would say I think that's more computer science.

CHILDS: Yes.

ZIERLER: Has that sort of entirely gone away in terms of the approach to quantum information and the appreciation among both departments in the interdisciplinary nature of the field?

CHILDS: Yeah. My feeling is that it's pretty much gone away. I don't know that it's completely gone away everywhere, but it seems to have certainly gone away at Maryland and quantum information is very much a part of the landscape, and that includes people doing quantum information in the computer science department. I have a background in physics, but I came to the computer science department here several years ago feeling maybe a little bit like an outsider. Although before I was here, I was in a math department at the University of Waterloo, so I was used to being outside the physics community that I had grown up in. I felt maybe in some ways not fully a computer scientist, but the department was very welcoming.

Computer science is an area that has evolved a lot and that involves connections to many things. There are people who do computational biology in my department, and people who do computational science related to classical physics. So the department has been very welcoming, and I feel like it really has changed a lot over the years. Now, many places are recognizing the importance of quantum information and hiring people to study it. Clearly, Caltech got it a long time ago and had people doing quantum computing within physics and computer science and other departments, but it's become, I would say, much more mainstream.

ZIERLER: Andrew, tell me the origin story of the NSF Quantum Leap Challenge Institute for Robust Quantum Simulation. How did that get started and was the creation of IQIM sort of a model in the creation of the Center at University of Maryland?

CHILDS: The program got started in response to the call for Quantum Leap Challenge Institutes that NSF put out. It was a component of the National Quantum Initiative that was signed into law several years ago. As part of that legislation there was this idea that NSF and also DOE would spin up these new centers. So we were responding to that call for new institutes. The selection process was done in two rounds. There were three institutes that were funded in the first year and then two more that were funded in the second year, and it was in that second year that we had applied. We wanted to do something that would leverage the strengths we have at Maryland and at the other collaborating universities.

There are many people here doing stuff related to quantum simulation. We thought it was a good way to pull together research going on in theory and experiment and across different departments, which seemed well aligned with what NSF was looking for. I don't know if the IQIM was specifically a model, although definitely the kind of activity that's gone on at IQIM and other NSF institutes definitely informed how we thought about things. Certainly, the time that I spent at Caltech informed how I thought about research centers—although there was no M when I was there, it was IQI.

ZIERLER: Right.

CHILDS: But one of the great experiences that I had there was working with the other IQI postdocs. There was a great group of people doing really quite different things, but who talked to each other a lot about research and collaborated. And it was just a really great environment that John had created. So that was really fantastic and that's definitely something that we've tried to reproduce through the postdoc program that we have at QuICS and that we're trying to incorporate as part of this Institute for Robust Quantum Simulation also.

ZIERLER: Now, it would seem on the surface at least that there would be a lot of overlap between the new NSF Center and what QuICS already is doing. So what is that overlap, and in what ways does the new NSF Center allow you to achieve things far beyond what would be possible only with QuICS?

CHILDS: There's definitely a lot of overlap in the sense that many of the Maryland faculty who are involved with the Robust Quantum Simulation Institute are part of QuICS or JQI, the other big joint institute that we have at Maryland. But there are several ways that the Institute for Robust Quantum Simulation adds value. There's a new emphasis on quantum simulation in particular, which helps provide focus. Also, the RQS Institute is not just a Maryland thing. We also have collaborators at Duke, Princeton, Yale, and North Carolina State, so it's creating a community that's definitely broader than just what's going on at Maryland. There's a big educational focus also as part of the program. We have a broad range of educational activities from outreach at the high school level through undergraduate and graduate education, postdoctoral training, training for post-degree professionals. So there's a lot of new stuff that's being enabled by this institute.

ZIERLER: Andrew, a question about the name. Of course, the idea of robust quantum information or robust quantum computation, that's throughout the literature going back twenty years.

CHILDS: Correct.

NSF and Robust Quantum Computation

ZIERLER: I wonder if you can explain a little what it means in the context of the creation of this NSF Center?

CHILDS: Yeah. It means we want to have quantum simulators that behave well, and there are several aspects that tie into the research aims of the institute. One is to be able to do verification, to have some confidence that your simulation results are correct. This is a challenging thing because the whole reason we want to do quantum simulation is because quantum systems are hard to simulate. So we can't just do some classical computation on the side that we can check things against, because that's exactly why we built the quantum device in the first place. So verifying that quantum simulations were done in the right way is a challenging thing. So that's one of the things that we're focusing on.

And in general, building quantum computers and quantum information processing devices where errors can be controlled and where you can have confidence that the system is behaving in a clean way, this is a major challenge for quantum computing in general, so understanding ways of mitigating and maybe even effectively eliminating noise, yeah, that's really an essential challenge.

ZIERLER: Just a nomenclature question, quantum simulators, in what way is that term interchangeable with quantum computers and where might there be separation between the terms?

CHILDS: It's not exactly the same thing, although there's definitely overlap. So I would say broadly speaking there's two kinds of quantum simulators that people think about, analog quantum simulators and digital quantum simulators. A digital quantum simulator is basically a quantum computer. If you can build a quantum computer, you can use it to do quantum simulations using networks of gates that you can design. And that's the kind of quantum simulation that I think about the most.

But there's a lot of work on quantum simulation where the goal is basically to build some physical system that allows you to model the physics of a system that you want to understand. You can try to engineer a Hamiltonian that describes the system whose properties you're interested in understanding. And there's been more progress toward realizing these kinds of simulators. People have been able to push further in terms of the scale of simulators that they can build and the kinds of phenomena they can explore, with the tradeoff being that it's maybe harder to get some guarantees of robustness, verifying that outputs are correct and understanding that noise is controlled and things like that. There's no theory of fault tolerance for analog simulators the way there is a theory of fault tolerance for digital quantum computers.

ZIERLER: Andrew, are there corporate partners with the University of Maryland that are contributing to this quest for quantum computation in the way that you see, for example, Amazon and Caltech, to take one of many examples? Do you have that at the University of Maryland?

CHILDS: Yeah, there's some of that. I'm not deeply involved with those kinds of collaborations, but we certainly interact with researchers at places like Microsoft, Amazon, Google, and IBM. We're very plugged into quantum computing research in general.

ZIERLER: To the extent that this is a horse race, if that's the right way to think about it, at least insofar as the major companies are involved, Microsoft, Amazon, Honeywell, the list goes, right?

CHILDS: Right.

ZIERLER: Do you see that they are essentially aiming for the same outcome but using different means to get there or how do you understand this race, at least from a corporate perspective, to achieve quantum computing?

CHILDS: This is not something I think about a lot. My general sense is that it's very exploratory. I think nobody really knows what exactly quantum computers are going to be useful for in a commercial sense. I think many companies want to explore what the potential of quantum computers might be, and so a way to do that is to have a research group that's going to explore quantum computing and see where it goes. I can't really speak to the motivations of people at companies and exactly what they're hoping to get out of building quantum computers, but I think there's just a general sense that there's a lot of potential. We know there are things we can do with quantum computers that are beyond the capabilities of classical computers. And it seems very likely that the more we explore them, the more we'll discover unexpected things about what they're able to do. And so I guess my take is that probably we just need to explore this more, and I think—or maybe I hope—that this is the perspective that companies are taking.

ZIERLER: I guess if I could refine the question, as you well know the process, even the engineering by which Microsoft is looking to create a quantum computer, is very different from Google.

CHILDS: Right.

ZIERLER: So the question is, are those different processes leading to the same outcome or are different processes leading to the idea that there isn't a single quantum computer, there are many types of quantum computers?

CHILDS: Oh. Yeah, I see what you mean. There's this notion of a universal quantum computer that's a device that can run any efficient quantum computation. And that's one mathematical concept that we can imagine instantiating in many different ways. So in that sense, I think there's one thing that all these companies are trying to realize, potentially in very different platforms. There are also ideas for devices that are maybe not universal quantum computers but can run limited kinds of quantum processes, such as quantum annealing, or other things that maybe we can do without the full control that you would need to do universal quantum computing, but would still maybe let us do something interesting. An analog quantum simulator actually is an example of something like this, but there's also all this stuff that goes under the umbrella of what John [Preskill] called NISQ. I'm sure you've come across this acronym, right?

ZIERLER: Yeah.

CHILDS: Noisy Intermediate-Scale Quantum devices. So I think definitely companies are interested in exploring that, and people in the community in general are interested in exploring what you can do in that regime. And that is maybe something of a qualitatively different kind than a universal quantum computer. But if you think about the long game and what we hope we can do eventually when we have digital quantum computers, conceptually to me that's one thing that everybody is trying to eventually realize.

ZIERLER: Now, of course, for you, operating in an academic environment, you mentioned question marks about what potential commercial applications are.

CHILDS: Right.

ZIERLER: What about just from a basic science perspective?

CHILDS: Yeah.

The Benefits to Fundamental Physics

ZIERLER: What are some of the promises that quantum computation might offer for, for example, say physics research?

CHILDS: Yeah. Well, I think that's actually probably the first target, and I think it's more plausible that in the shorter term we can do things where quantum simulators or other kinds of quantum computations will let us learn interesting things about physics. I think there are examples of condensed matter models, for example, that people would like to understand better. For example, what is the nature of some phase transition as you vary a given parameter? People sometimes try to answer those kinds of questions with classical computation, but they just can't push things very far because quantum mechanics is hard to simulate. So that's the kind of thing that I think once we have even fairly rudimentary quantum simulators, we can actually answer questions about physics that people care about, but that are beyond the kind of computation that we can do with classical computing devices.

ZIERLER: Andrew, I'm curious for you and for the various quantum information endeavors that are happening at Maryland, just as a matter of geography, being in the Washington hub close to NIST, DOE, the list goes on, in what ways is that proximity really advantageous for you and your colleagues?

CHILDS: I think it definitely is. I think that the engagement that we have with NIST is absolutely transformational. That's the reason that we have QuICS, this quantum computing center that I'm a part of, and it's the reason that JQI is here. Those are both joint centers with NIST. And yeah, NIST's involvement is so essential to the success of these activities. There are many scientists from NIST who are really outstanding researchers who are contributing to the work that's going on and enriching the research community that we have as part of these centers. And beyond the specific engagement with NIST, it's definitely much easier to talk to funding agencies and folks at federal research labs that happen to be in the area. There's just a lot of that kind of stuff in the DC area and yeah, it makes a difference to have those connections nearby.

ZIERLER: Just a snapshot in time, what are you currently working on these days?

CHILDS: Oh, many things. Let's see. For example, I have a paper now that we're just trying to finish up which is about simulating quantum mechanics in real space, so if you have particles interacting through given potentials, how efficiently can you simulate their dynamics, taking into account the error that results when you discretize space to put it on a computer. This is actually a very basic question that was one of the things that people first considered when writing down quantum simulation algorithms but somehow, there are things that still haven't been worked out, so that's been fun to come back to. I also have some projects about verifying quantum computations using interactive protocols. There are some really interesting questions there.

ZIERLER: Institutionally, what are some unique opportunities of collaboration at the University of Maryland between experimenters and theorists?

CHILDS: I've found Maryland to be a really collaborative place. One of the things that I really enjoyed in coming here was forming new kinds of collaborations. I actually have not collaborated so much with experimentalists here, more with other theory colleagues. But definitely, yeah, there's a lot of great quantum physics experiment that's going on. I haven't personally collaborated with experimental colleagues, but in general, there is a lot of collaboration between theorists and experimentalists here.

ZIERLER: Andrew, some fun questions because I get as many answers as people I ask them to, just the state of the field. So at a very basic level, do we yet have a functional quantum computer, however you might define that?

CHILDS: "Functional." Maybe I won't exactly answer the question. I think we have great proof-of-concept devices. We've had them for a while, but they're much better now than they were, I don't know, 10, 15 years ago. At the same time, they're not the dream of a scalable, fault-tolerant quantum computer. We haven't seen really fault-tolerant implementations of quantum computation yet, and I feel like that's something we at some level need to have a real quantum computer, to really be able to have arbitrary control and be able to dial in any quantum circuit. The theorist's dream that we think about when we design algorithms, it seems like we're very far from having that, but at the same time we've come a long way and it seems like there's a path to getting there.

ZIERLER: To get there, what do we need to do both from a theoretical and experimental and even perhaps an engineering perspective to create a real quantum computer?

CHILDS: Well, I think there's just a lot of work that has to go into building devices that we can scale up, where we can keep the noise low enough, and where we can have control of the device that allows us to get into this fault-tolerant regime. And so on the theory side I think there are really interesting challenges about—this is not what I work on, but I think it's really important—finding more efficient fault-tolerant protocols so we can make the overhead from trying to do quantum computation fault tolerantly as low as possible. There's a lot of overhead there and as a result there's a lot of scope, I think, for making things more efficient and having to pay less in encoding your qubits in a fault-tolerant way.

On the algorithm side, which is where I work, there's definitely lots of scope for finding new applications but also for taking problems that we know how to solve and making the solutions even more efficient, making the quantum circuits smaller, optimizing the performance of algorithms. That's something that I've worked on a lot for quantum simulation algorithms specifically. But I think a lot of the progress has to come on the experimental side. It's just really hard to build qubits that we have good control over.

ZIERLER: Because you've been involved in these questions for so long, where do you see measuring progress over the decades as a source of being able to extrapolate to some degree into the future as in, okay, we've gotten this far, we've solved this, here's what we have to do next? How might you translate that very roughly into a timetable of achieving the dream, as you say?

CHILDS: You mean in terms of experimental progress, progress just toward building a quantum computer?

ZIERLER: All of the above. All of the things that it will take to get there, however you define that benchmark.

CHILDS: I don't know. It's hard to make predications and I generally prefer not to try to extrapolate too much. I just don't think I'm very qualified to do it. It's really hard to say. I think it's definitely wrong to just focus on how many qubits can we build because I think, as many people recognize, it's important to consider how well can you control them and how deep are the quantum circuits that you can reliably do, which correlates pretty directly with the error rates. So I think that's an important thing to track, not just how many qubits can you make but how many qubits can you make of a quality that's good enough that you can, in principle, hope you can scale them up. But mostly I find it just very hard to plausibly extrapolate.

ZIERLER: Andrew, last question before we go back to undergraduate days. Are there limitations in classical computing that serve as a guidepost to help conceptualize what quantum computing can do or at least do better than classical computers?

CHILDS: Yeah. Well, definitely from the standpoint of computational complexity theory there are. A big one is the difference between the difficulty of solving problems and verifying solutions of problems—this is the P versus NP question. There are many things that we believe but don't know how to prove, like the difficulty of NP-complete problems. We have a lot of belief that's been built up over the years that these problems are hard for classical computers, but also, we believe they should be hard for quantum computers. This puts a lot of constraints, assuming that that's true, on the kinds of problems that we can hope to efficiently solve with quantum computers. And so we have to try to carve out a space that somehow is beyond what we can do classically but we're not looking at problems that are so hard that we don't expect to be able to solve them efficiently with quantum computers.

Caltech at the Dawn of Quantum Information

ZIERLER: All right. So let's now go back to undergraduate days. First, root me in the chronology. What year did you arrive at Caltech as a freshman?

CHILDS: I guess it was 1996.

ZIERLER: Okay. So 1996. Obviously, this wouldn't have been on your radar at the beginning, but do you remember, was John Preskill talking about quantum information? Did you interact with him at all?

CHILDS: Oh, yeah. Yeah, I did. And it was on my radar, actually. I was interested in physics as a high school student and read popular science books and stuff like that, and I think I learned about quantum computing from articles by Tom Siegfried in the Dallas Morning News. Caltech was very much on the map even at that point as a place that was doing quantum information, so I think there was stuff in those pieces about work going on at Caltech and maybe at JPL even. So I knew about it when I went there.

And I was interested in getting involved in quantum computing research, so I started going to John's group meetings. He ran these fantastic group meetings. Michael Nielsen was there at the time, along with Chris Fuchs and Steven Van Enk. There was a great group of people there in the early days, pushing things forward. So I went to his group meetings, and I did a SURF with John the summer after my sophomore year, along with Joe Renes. Joe was a Caltech undergrad at the time, and he's also still working in quantum information. We wrote a paper called Quantum Information and Precision Measurement, which was about using computational ideas to think about protocols and limitations for measuring parameters of physical systems. John's vision was that we could use quantum information ideas to think about limits on measurement and how you could maybe improve measurement. Actually, I just saw recently that Caltech is launching a new precision measurement institute.

ZIERLER: Oh yes.

CHILDS: And there's going to be a new building. And somehow, I don't know, it took me back.

ZIERLER: [laughs]

CHILDS: John wrote this article around that same time, in the mid '90s, about ways he thought that quantum information was going to revolutionize other areas of physics. And there were two things he identified, one of which was precision measurement. He saw that there should be interesting connections between quantum computing and measurement theory.

ZIERLER: Now, was it precocious of you, looking back, as an undergraduate to be hanging out in the group meetings or were there other undergraduates there who shared your interests?

CHILDS: I don't know if there were other undergraduates that went to the group meetings. I don't know. But I was excited about it, and I went to these meetings and people would present stuff and I sat in the back and took notes. And I don't know how much of it I understood. I wonder if I have those notebooks somewhere. But it definitely immersed me in the culture for sure. And I learned a lot through working with John and Joe on that project. It was fantastic.

ZIERLER: So John was accessible to you as an undergraduate, you could spend some time with him?

CHILDS: Yeah. Yeah, definitely. Well, I did that project with him, and we would meet every week for 10 weeks or something like that over a summer. Yeah, I learned a lot.

ZIERLER: What was that project? What did you work on?

CHILDS: It was this precision measurement thing. We thought about how you could distinguish parameters of Hamiltonians more efficiently using quantum information ideas, using the quantum Fourier transform to resolve a frequency more efficiently, or some other setting where you have some collection of Hamiltonians and you know that the system is governed by one of them and you want to figure out which one it is. How can you use the fewest resources to identify those dynamics?

ZIERLER: Were there other physics professors that were involved in this, or it was essentially only John at this point?

CHILDS: I guess it was mainly John. Maybe also Jeff Kimble. But yeah, I think at the group meetings that I went to, John was the only faculty member.

ZIERLER: Now, you were a physics major, that was your major at Caltech?

CHILDS: Yeah. Right, right.

ZIERLER: Did you see the connections even as an undergraduate in the way that John conveyed his excitement about how at least initially quantum information would be great for physics research?

CHILDS: Well, I saw that he was making those connections, yeah. I think that made sense. I think mostly I was interested in understanding more physics. And I really like finding interesting math problems through physics research, so that was one of the things that appealed to me. So I was happy to identify some mathematical challenge and work on that. That was one of the things that definitely excited me about that project and about quantum information in general.

ZIERLER: By the time you were ready to start thinking about graduate school were you specifically thinking about quantum information and people to work with in that field?

CHILDS: Yeah. It was what I wanted to do. I wanted to work on quantum computing. I would go and talk with John about the continuation of that research project that we had done. It took longer than just the summer to write the paper, so I would go talk with him periodically about that, but also about advice on where to go for graduate school and stuff like that. And so that was definitely helpful in figuring out where to go and what to do.

ZIERLER: I asked you about your sense of the cusp of breakthroughs now in 2022, right?

CHILDS: Right.

ZIERLER: Looking back 20 or 25 years, what was exciting? What seemed to be on the horizon when you first started to think about these things?

CHILDS: People were definitely talking about building quantum computers, and I guess I had no concept of what the timescale for that might be like or what it might eventually look like, what it might take to get there. Another thing that I did when I was an undergrad at Caltech, I don't know if John suggested this or how exactly it happened, but I spent some summers at IBM Research with Ike Chuang, who was doing NMR quantum computing at that time. And I think it was clear at that time that that was not really a scalable path to quantum computers, but it was a way to sort of get your hands on something that was like a quantum computer in some ways and learn something about how it operates. So I think that was important for me in seeing more about what the path to building a quantum computer might look like. But I think I also recognized—also probably in large part through learning about quantum information from John and from his lecture notes for his course, which are fantastic and which I learned a lot of quantum information from—was the importance of developing quantum algorithms and coming up with new applications of quantum computers. That was something he emphasized as an important direction, and that's what I decided to focus on.

ZIERLER: Andrew, the field was so young at that point. When you started to think about other programs, presumably the culture at Caltech, was don't stay for graduate school, go somewhere else.

CHILDS: Right.

ZIERLER: Were there places that had, even if you can call it, established quantum information programs or people who were working on these things was like John was at Caltech? What did the landscape look like circa 2000?

CHILDS: I don't think there were really places that had established programs in quantum information as such, but there were definitely strong places where people had been working on quantum computing in a serious way for a while, so those were the places that most appealed to me. I ended up going to MIT to work with Eddie Farhi and I think that was a great choice. It worked out really well. He had been working with several collaborators on quantum algorithms and quantum computing for a while. I remember that he was kind of hesitant to take PhD students in the area. Maybe I was his first PhD student whose main topic was quantum computing.

ZIERLER: Just because the field is so iffy at that point?

CHILDS: Right. Exactly.

ZIERLER: It's like a moral issue, you can't do this?

CHILDS: Exactly. I think he wasn't sure. Does it make sense to work with someone on this and they'll have no future because it could just be a flash in the pan.

ZIERLER: Yeah. [Laugh]

CHILDS: I really think he was concerned about that, and he hesitantly agreed to work with me on it, but I think it turned out okay.

ZIERLER: Now, had John and Eddie, had they collaborated? Did John give the advice that Eddie might be a good person to work with?

CHILDS: Yeah. I think he probably did. I don't remember a specific moment where he said that, but I suspect that he did. They definitely knew each other. They both came from a particle physics background.

Quantum Algorithms at MIT

ZIERLER: When you got to MIT was it similar in the sense—was Eddie a one-man show like John was at Caltech? Was the group any larger than Eddie, so to speak?

CHILDS: Yeah. Yeah, it was bigger. Eddie had several collaborators that he worked with on quantum computing for a long time. Sam Gutmann, who actually was a childhood friend of his, has been a long-term collaborator. Sam was at Northeastern University in the math department. And Jeffrey Goldstone, who's an eminent particle physicist at the Center for Theoretical Physics there at MIT, worked with them on quantum computing. And also, they had been working with Mike Sipser who is a well-known complexity theorist. The four of them had been working together for several years. They wrote five or six papers about algorithms and lower bounds for quantum computers. And then I think Mike eventually decided around the time of their proposal of the adiabatic algorithm for optimization problems that he didn't want to think so much anymore about quantum computing. That was right about at the time that I showed up at MIT, in 2000, so he had sort of left the group. But yeah, there were several people that Eddie had been thinking about quantum computing regularly with by the time that I got there.

ZIERLER: Now, it's unique for you because most people, particularly of this generation, did not think so deeply about quantum information as an undergraduate at least. So for you when you got to MIT, did you have some kind of an idea of what you wanted to work on or were you really an open book when you got to MIT, you were open to all kinds of things that might be possible?

CHILDS: I wanted to work on quantum algorithms. I think I had really decided that's what I wanted to do. And it was natural because that was what Eddie was doing. So the kinds of problems that he had in mind to work on were in that direction, and it's kind of what I wanted to do so, yeah, it worked out pretty smoothly.

ZIERLER: Tell me about the paper Robustness in Adiabatic Quantum Computation that you were a coauthor along with Eddie and John in 2001, how did that come together?

CHILDS: Eddie and his collaborators had proposed this idea of adiabatic quantum computing and were thinking about it as sort of like an algorithmic tool. But I guess he had this idea that maybe there should be some inherent robustness to this method because of properties of adiabatic evolution: it doesn't matter exactly what path you go along as long as you get to the right place. It's not exactly topological protection, but it's something of a similar kind of flavor. And my recollection is that he had started working on this with John and they had some discussions about what they maybe could say about this. And at some point, Eddie asked if I would be interested in joining and working with them, maybe because they decided that they wanted to have some sort of numerical calculations as part of the paper and they needed a graduate student to help them with that.

ZIERLER: [Laugh]

CHILDS: Although I have to say Eddie is very proficient with MATLAB and coding stuff up. He actually doesn't mind getting his hands dirty. But anyway, they involved me, and I did some simulations and also worked on formulating an open quantum system model we could use to explore what happens if the system is cold. If you couple your system to a bath at some low temperature, you might hope that somehow this could even help to keep the system in its ground state, and certainly shouldn't hurt. So yeah, I sort of helped with developing a model for that and also doing some numerics, which were very limited, actually, but we did what we could.

ZIERLER: Andrew, for the non-specialist audience that will be reading this transcript at some point, I wonder if you can explain how adiabatic modifies quantum computation? What does adiabatic mean in this context?

CHILDS: Adiabatic basically means changing parameters of the system slowly enough that you can keep it in its ground state. You can imagine that if you change systems suddenly, they behave differently than if you change them really gradually. If you have a cup of water and it's very full and you try to carry it, if you do that really gradually it's okay, the water will just stay flat. But if you really move it fast, the water will slosh around. You're exciting oscillations that maybe you don't want to if you don't want to get wet. So the idea of adiabatic quantum computing is that you change parameters of your system slowly enough that it will track its ground state. And so what you try to do is you try to start from some system in a ground state that you could easily make and then you have some system you can build whose ground state encodes the solution of some hard computational problem, but maybe it's not obvious how to prepare that, because the problem is hard.

So what you do is you start with this easy thing, and you slowly morph it into the thing whose ground state is this hard-to-find thing. And there's this theorem called the adiabatic theorem that says that if you go slowly enough, you will end up in the ground state of the hard thing, and that will allow you to solve the problem. So that's the idea of the adiabatic algorithm. Of course, the whole challenge with this is how slowly do you have to go? How long is it going to take you to walk across the room with the cup of water? And it turns out that that's a really hard thing to understand. For the cup of water it's not so hard to understand, but to reach the solution of a hard computational problem, it's a hard thing to understand.

ZIERLER: Now, as you mentioned earlier, Eddie's idea that there needed to be, as you said, some inherent robustness here, you already explained what robustness means in this context, but how do you make it inherent, what does that mean?

CHILDS: Oh, yeah. What I mean is just that somehow there could be robustness that's already in the algorithm, so it's not an additional feature that you have to introduce. There's this theory of fault-tolerant quantum computing that tells you how to encode quantum information using error correcting codes and perform gates in a fault-tolerant way on the encoded information, but that's not inherent robustness because you add it by a careful encoding. So when I say that maybe the adiabatic algorithm could be inherently robust, what I mean is that maybe you could just implement it the way it was proposed, and actually things would be robust, in the sense that if you didn't do it exactly right or errors happened in the course of trying to run the algorithm, that actually it would be okay.

ZIERLER: Now, looking back, obviously this is a landmark paper in the history of quantum information, but that's retrospective. How was it received at the time?

CHILDS: Well, I don't know if I would say it's a landmark paper. I don't know. I think that's overly generous. I think at the end of the day there is some robustness related to what we were able to show but I think it's somewhat limited. And I think it's actually still to this day sort of an open question how robust you can actually make adiabatic computation.

ZIERLER: Is this one of those areas that needs to be resolved?

CHILDS: Yeah, I think so. I think so, actually. I think there are still interesting questions about how robust adiabatic computation really is, or maybe it would be worthwhile to try to add robustness through something like error correcting codes. Eddie has another paper with some other collaborators about proposals for that, and I think there are interesting questions about how to push that further that really haven't been explored much. But I think there are also really important unresolved questions about the basic computational power of adiabatic optimization. As I was saying, it's really hard to understand how slowly you have to go, and that really relates to how effective the method is. If you have to go exponentially slowly then it's not such a useful method. And it's hard to understand to what extent that's the case. I think that's still very much open and there's a lot that's not understood about the power of adiabatic optimization.

ZIERLER: What was Eddie's style like as a graduate mentor? In other words, how hands-on was he as you were starting to think about what ultimately would become your thesis research?

CHILDS: He was pretty hands-on in the sense that he would help to find a problem and we would have regular meetings to have discussions and go over things on the board. And we really met every week and talked a lot about what we were doing. At the same time, I had other projects that I developed on my own and he was perfectly happy for me to do something else through some other collaboration. I had some papers that I wrote just with Jeffrey, and I had papers that I wrote just with external collaborators. Eddie was very flexible, but I would say he was on the more hands-on side, and I think it was helpful.

ZIERLER: Tell me about the ideas that became eventually your thesis. How did that come together?

CHILDS: Well, that paper on the robustness of adiabatic optimization was a part of it, but there was also other stuff in there—it was kind of an assembly of various different things that I had done during the time that I spent as a PhD student. For example, the work that I did on spatial search with Jeffrey Goldstone was in there. There's kind of an interesting story there, as well. There was a talk that we had as part of the seminar series that Eddie ran on topics related to quantum computing where Scott Aaronson had visited to tell us about this paper he had written on spatial search with Andris Ambainis. And it was this interesting question about how efficiently you can search a region of space. Suppose you have a quantum robot that somehow can go into a superposition of different locations—this is very science fiction—but how rapidly could you use that to find a treasure chest or something that's located somewhere. It's similar in spirit to the Grover's search algorithm, which gets a quadratic speedup for finding a marked item in the setting where you have, a black box that tells you whether a given item is the thing you're looking for.

But the spatial search question has this additional complication that the geometry matters. Scott told us about his work on this, and we realized—or maybe Scott had suggested—that it would be interesting to consider how well quantum walk algorithms would be able to solve this problem. Jeffrey and I were able to understand something about that, and that became a piece of my thesis. Actually, yeah, I guess I should mention also there's other stuff about quantum walks in my thesis. This really grew out of work that Eddie had done with Sam Gutmann before I joined as a PhD student. They were looking for ways that quantum mechanics could speed up classical processes. They didn't call it a quantum walk at the time, but they defined a process that was a quantum analog of a random walk and showed you could do certain things faster using it. And then I got involved in work that developed this model of a quantum walk and explored its computational power.

ZIERLER: And if we can unpack the title a little bit, Quantum Information Processing in Continuous Time, what is continuous time?

CHILDS: Right. Often when we think about quantum circuits we think about discrete sequences of gates. You have some elementary gates, CNOT gates, and Hadamard gates, and Toffoli gates, and certain kinds of gates that act discretely at certain times. That's a very powerful model for thinking about what quantum computers could do, but you can also think about quantum computations that operate through what we call Hamiltonian dynamics, where there's this thing called the Schrödinger equation that governs how quantum systems evolve continuously in time in quantum mechanics. You know, time is a continuous variable and over some amount of time some evolution happens as determined by this operator called the Hamiltonian. So these quantum walks that I was just describing, they're an example of quantum processes that are defined by specifying some Hamiltonian and then letting the system evolve continuously in time according to the Schrödinger equation. And adiabatic algorithms are another example of that—the algorithm is specified by this thing, the Hamiltonian, that tells you instantaneously in time how is the system going to be pushed forward, and then you solve the Schrödinger equation to figure out what happens to the system when you let things evolve for a long time.

ZIERLER: Besides Eddie who else was on your thesis committee?

CHILDS: Jeffrey and I guess—were there only three people? There was also a condensed matter theorist named Leonid Levitov.

ZIERLER: Anything memorable from the oral defense?

CHILDS: No, I don't really remember a lot. It was kind of a blur.

ZIERLER: Was it the kind of thing where you really knew more about the topic than the professors? Was it more of a conversation?

CHILDS: I think it was friendly and it was a conversation. I wouldn't say that I knew more about it than the professors, though. [Laugh] They're pretty smart cookies.

Return to IQI

ZIERLER: [Laugh] When it was time to start thinking about postdocs were you keyed in at Caltech still? Were you following what was going on there, the growth of the IQI?

CHILDS: Oh, for sure. Yeah. Yeah, definitely. I guess the IQI started right after I left Caltech.

ZIERLER: Right.

CHILDS: I guess it was 2000 or something, something like that.

ZIERLER: Right.

CHILDS: Yeah. I think I sort of just missed the launch of the IQI, but I definitely knew about it. I was in touch with people at Caltech and I went to quantum computing conferences and stuff, and I knew people from there so, yeah, I was very aware of that happening.

ZIERLER: Was that the place to go to? Did you apply more widely?

CHILDS: I don't remember where else I might've applied. I think it was probably the main place I wanted to go, and I don't even remember. I remember other places that I applied for graduate school. I was also thinking about going to Berkeley to work with Umesh Vazirani but decided to go to MIT instead. But yeah, I don't remember where else I applied for postdocs.

ZIERLER: Now, to the extent that you thought about these things or attached meaning to it, being a DuBridge postdoctoral scholar, with Lee DuBridge being, of course, the past president of Caltech, did that sort of convey how important this endeavor was to Caltech, to be named in the honor of a past president?

CHILDS: I don't think I appreciated it at the time. I think I just wanted to go to Caltech and study quantum computing some more.

ZIERLER: What were the big, exciting things at that point? What did you want to work on next?

CHILDS: At the time that I was finishing my PhD and moving to Caltech I was really interested in this thing called the hidden subgroup problem, which is a kind of a problem that generalizes the core of Shor's algorithm, the factoring algorithm. So the idea is to somehow find hidden symmetries in systems. You can view the factoring algorithm as somehow finding some hidden periodicity that reveals the factors of numbers and there was this idea that you could somehow generalize that to the setting of nonabelian groups. And we sort of knew that that could potentially have interesting applications. Actually, a bunch of progress was made on this nonabelian hidden subgroup problem, although unfortunately not that leads to solutions of the problems that motivated it in the first place. But yeah, I was very interested in that and so that was something that I spent a lot of time working on. But I also found new things to work on and made new connections by talking to the other postdocs at the IQI. That was one of the really great things about being there.

ZIERLER: So coming back to Caltech having left right before the creation of the IQI, since you were there when the IQI existed at least informally, John ran these meetings, people were talking about quantum information, in those four years that you were gone in what ways had quantum information grown both scientifically and administratively at Caltech?

CHILDS: Definitely the field had advanced in terms of what was known. Also I had learned a lot in those four years, so I certainly had a very different perspective. Administratively there was an office that had been set up where the postdocs sat which was not in Lauritsen, where John sat with the high-energy physics group. There was this separate outpost where the postdocs sat. But often we would meet John for lunch. It was great.

ZIERLER: I'm getting a sense that food was really an important part of things for the IQI. [Laugh]

CHILDS: Yes. Yes, definitely.

ZIERLER: How many postdocs were in your cohort?

CHILDS: There were five or six postdocs at a time. Now here at QuICS, we have many more postdocs than that, but for the time it was a big group, and we had a lot of great discussions. I was there at the same time as Frank Verstraete. Who else? Sergey Bravyi was there at that time, and Robert Raussendorf. And Pawel Wocjan, who I wrote a bunch of papers with. Several other people, too. There were quite a few people there, and it was fantastic.

ZIERLER: And what was the culture of collaboration like? Was everybody sort of working together, bouncing ideas? Was it single author papers that you would present? How did those things work out?

CHILDS: It was not a lot of single author papers. It was mostly, yeah, people kind of getting together and working in small groups. I wouldn't say that everybody worked with everybody else on everything because people had their own specializations. I wrote a bunch of papers with Pawel because somehow we had some common interests, and I never wrote a paper with Frank or with Robert or with Sergey because they were interested in different things. But we definitely talked with each other about what we were working on and asked questions about technical stuff. It was very interactive. And there were also these group meetings that I guess looked very different during the time that I was there as a postdoc because somehow the activity had grown a lot. There were all of these postdocs that would go but also all of John's graduate students and other students in adjacent groups. And these meetings got very large and were sort of epic in scope. They would go for, like, three hours. Everybody would give short updates on what they had been thinking about recently. And then we would have a recess and then somebody would give a presentation for an hour and a half on something that they were working on. Yeah, it was fantastic.

ZIERLER: What aspects of your postdoc were a refinement or an expansion on the ideas that you pursued in graduate school and what were totally new concepts, new ideas for you?

CHILDS: I was still fundamentally thinking about quantum algorithms, so it was in that way connected a lot to stuff I had been thinking about. But yeah, as I mentioned, I sort of branched out and thought about this new direction of this hidden subgroup problem and I learned more about new mathematical things, group representation theory and this thing called Schur-Weyl duality and some mathematical ideas you could use to exploit certain kinds of structures that arise in those problems, so it was a lot of fun to learn that stuff. But yeah, it was definitely at some level really very much a continuation of this idea of understanding what is the power of quantum computers and looking for specific problems where you could say something new about that.

ZIERLER: Now, in these early years, I'm curious about your sense looking back of how parochial IQI was in the sense that it was the center of the universe, and you didn't really need to follow what was going on so much elsewhere in the field, or was it not like that and you were more broadly connected?

CHILDS: Yeah, I don't think it was ever like that. I remember at these group meetings that I would go to, the group meetings that John would run when I was an undergrad when I was sitting in the back of the room taking notes, they would discuss papers on the arXiv that came from all over the world. It was not just that people were locally working on what everyone at Caltech thought was interesting. They were really following research that was happening all over the place. And so I think, yeah, it was the case then and it was definitely the case when I was a postdoc. Quantum information was not as big then as it is now but there was definitely a worldwide community and people were coming from all over the place to Caltech and talking with people all over the place, going to conferences all over the world. It was pretty far reaching.

ZIERLER: Andrew, just a few more questions to wrap up. When you joined the faculty at Waterloo what were some of the things that you brought with you from Caltech, just from a research perspective, from the perspective of collaboration, and just from the big ideas that you wanted to pursue in your first faculty job?

CHILDS: Let's see. I'm trying to think back. What were the things that I was working on when I first went there? I have to kind of remind myself. I finished at Caltech in 2007.

ZIERLER: So your papers around that time were Quantum Algorithms for Hidden Nonlinear Structures, for example.

CHILDS: Yeah. That was a Caltech paper really. That was a paper that I wrote with Leonard Schulman who was at Caltech—I guess still is at Caltech.

ZIERLER: Yeah.

CHILDS: But that was something that came up through interacting with him at Caltech. I wrote this paper about universal computation by quantum walk which is definitely something that I had been working on at Caltech and presented at one of these long group meetings that I mentioned. And that was something that probably got finished during the time that I was first an assistant professor, but that was definitely something that led to interesting future work. I had a project that I did with David Gosset, who was a postdoc there at IQC, and Zak Webb, who was my PhD student, which was about generalizing that to a setting where you have multiple particles. And this somehow allows you to do interesting universal quantum computations on much smaller graphs and explore a large Hilbert space with a small graph. So that was something that I did at Waterloo that was definitely an outgrowth of stuff that I had done at Caltech. The stuff that we did on algorithms for formula evaluation, for evaluating AND-OR trees, was also something that I guess I worked on when I was first at Waterloo, and it was sort of a generalization of ideas coming from some of these quantum walk algorithms.

Emphasis on Collaboration

ZIERLER: Finally, Andrew, last question. Looking to the future again, in order for you to extrapolate a little, to the extent that the creation of true quantum computing—obviously, it's a global worldwide effort that's going to involve all kinds of different researchers coming from all different kinds of specialties. From your perspective, your unique perspective having been there at Caltech really from the beginning as a postdoc and what you brought with you all the way to the present, what do you see as some of the hallmarks of—if you can imagine, like, a genetic imprint from Caltech on quantum computing, both from a research perspective, the theory, and also just the general approach, the big ideas that will ultimately make this possible?

CHILDS: I don't know. That's a big question, and a lot of the time I have my head down and I'm sort of looking at the particular problems in front of me. I'm not sure I have a great answer. Definitely in general terms the experiences that I had at Caltech and the kinds of things that I learned when I was there have very much informed the way that I think about quantum computing and quantum information. I think the spirit of collaboration that we had during the times that I was there, especially when I was there as a postdoc, is really something that I've tried to reproduce in other places where I've been—to have people very freely discuss their scientific ideas and talk about what problems they think are important and what ideas they have for approaching them. So I think that's definitely something that I've taken from that experience.

ZIERLER: No, that's fine. That's fine. This has been a great conversation. I'm so glad we were able to do this. I'd like to thank you so much.

CHILDS: Yeah. Thank you.

[END]