Associate Professor of Physics, University of Colorado and Associate Fellow, JILA
By David Zierler, Director of the Caltech Heritage Project
January 26, 2022
DAVID ZIERLER: This is David Zierler, Director of the Caltech Heritage Project. It is Wednesday, January 26th, 2022. I am so happy to be here with Professor Graeme Smith. Graeme, it's good to be with you. Thank you for joining me today.
GRAEME SMITH: Hi.
ZIERLER: Graeme, I'd like to start if you can tell me please your title and institutional affiliation?
SMITH: Oh, sure. I'm an Associate Professor of Physics, and a Fellow of JILA, at the University of Colorado in Boulder.
ZIERLER: Graeme, how does that work in terms of your responsibilities? Is it essentially a joint appointment between Physics and JILA, or do you have a primary appointment with interests on the other?
SMITH: It's a little complicated. JILA's a joint institute, joint between the university and NIST. The fellows there are all either employed by the University of Colorado or NIST. I'm employed by the university. I'm a faculty member in the Physics Department, but I sit in a building which is different from the physics building. It's right next to it. It's the JILA building. That's where my students sit. I have responsibilities in both institutions.
ZIERLER: In terms of your funding, does it all come from JILA, or do you get grants from NSF, other kinds of places that are funding quantum information?
SMITH: Very little money comes—well, it almost all comes from grants.
ZIERLER: What is the larger role that you see from JILA in this effort to build a quantum computer? What are they contributing to this broader effort?
SMITH: Building a quantum computer is not the primary goal of JILA. It's a side effect of much of the research that's going on there. But, I think, basically you will find nobody whose primary goal is to build a quantum computer at JILA. It's largely an atomic and molecular—an AMO physics institute. Much of the interests, to the extent that there is a vision, it comes from the overlap of academic interests and the interests of NIST. NIST, of course, is very interested in precision measurement, and that's, I would say, is kind of the primary driving thing going on. Then, the reason people are involved in quantum information and quantum computing at JILA, is that, when you make an attempt to build really precise instruments, you end up doing a lot of things that are very similar to manipulating quantum information, or even building quantum computers. But it's not that we're like somebody's trying to make the next quantum computer with the largest number of qubits, or something.
ZIERLER: It's an interesting way that you phrased that. That quantum computing is almost like a side effect of what's going on at JILA. I wonder if you can explain, what are some of the research projects at JILA that, even in a happenstance manner, might actually contribute to some future scalable quantum computer?
SMITH: Sure, let's see. First I'll tell you about the experimental work. I'm a theorist, but there are really great experiments here. Jun Ye, for instance, builds just the best atomic clocks. This involves managing the interactions between a bunch of atoms and trying to make them as isolated as possible, and high-fidelity as possible. That effort is kind of inevitably—well, I would say, is moving towards making a small quantum computer. A highly-controlled quantum system out of neutral atoms. We don't have anyone working on ion traps here at JILA, but at NIST they have basically the best ion trap group in the world. They, I guess, got into the business for building clocks, but it turns out they're also very, very good quantum computers.
ZIERLER: Just as a snapshot in time, what are you doing, and what are some of the exciting developments in your group these days?
SMITH: I'm doing a lot of things right now. Maybe we'll talk about this later, but when I get into this stuff, I was really properly an information theorist, doing communications problems, stuff like that. I still do that sort of thing. But since coming to JILA, I have gotten more interested in metrology problems. One metrology problem we've been looking at lately is, tradeoffs in sensing AC magnetic fields. So, somehow strategies you use for measuring the strength of a magnetic field at one frequency, are, in some fundamental way, required to be insensitive to the magnetic field at some other frequencies. That's what I've been thinking a lot about lately, how to quantify and use fundamental constraints on sensing to better understand strategies for doing stuff like, measuring very weak signals. That's one thing.
Another direction—well, I can tell you about the information theory stuff. It's still most of what I do. That involves really trying to understand, given a noisy communication link, how can you use it, through error correction, to mimic a noiseless communication link. How many qubits can you send across a noisy channel? With what fidelity? What kind of codes should you use to achieve that rate? How can you combine different resources in a way that is helpful for enhancing the capacities of the individual channels?
ZIERLER: Your students, the graduate students and postdocs in your group, is their primary affiliation at Colorado or JILA, or how does that work?
SMITH: The postdocs are affiliated with JILA, and the students are all affiliated with Physics and JILA.
ZIERLER: In terms of instrumentation, of course you're a theorist, but the world-class instrumentation at JILA, in what ways is that relevant for the questions you're after?
SMITH: This magnetic field sensing problem just came up literally out of discussions with James Thompson, who is very interested in understanding how quantum effects can be used to do better sensing. More generally, we have collaborations in the form of large grants with a large number of PIs where maybe what I would offer is some theoretical support for the more experimental parts, like maybe trying to figure out, are there ways we can do error correction—experiments using the systems that people already have, or are there ways that we could perhaps run small algorithms in a meaningful way.
ZIERLER: A broad historical question—given that you were at Caltech, really at the origins of IQI, is your sense that over the past 20 years, interactions between theorists and experimentalists have reached a point where breakthroughs in quantum computing are much more feasible now than they might have been 20 years ago?
SMITH: It's kind of a leading question. I'm not sure I would put it that way. I think there's been slow and steady progress on experimental efforts, and now they're getting to the point where they're genuinely interesting in terms of what they can do. Twenty years ago, people were trying to get good two qubit gates, and they didn't have them. There's not a whole lot of, I guess—I mean, there's theory you can do in terms of modeling how they're trying to do their gates, but there's not a whole lot of theory that's very deep, that you can do, on the computational capabilities of two qubits, or something like that. Now people are starting to have bigger systems. They're still nosier than they need to be, but they're getting to be interesting. I think there's a lot of enthusiasm and hope for what that will mean for the future.
ZIERLER: Just some nomenclature questions, you mentioned metrology. Just what is metrology?
SMITH: It's the science of measuring things. Usually, one would think of measuring frequencies, time, magnetic field strengths, charges.
ZIERLER: What are the most important measurements, as they relate to quantum information and quantum computing? What are the units, what are the benchmarks?
SMITH: One thing would be how precise—well, this is where sort of the two interests slightly diverge. The most important measurement for an atomic clock, is how precisely can you measure an amount of time. It turns out that when you build a system that's really good at measuring time, it's a really good clock, you kind of have to have, at least, a pretty good quantum computer. The thing that's not really as fundamental—or not as essential for clock measurements, is having two qubit gates. You can just think of the clocks are literally little atoms that spin around like this. Maybe it's not as essential to have really high-fidelity two qubit gates for an atomic clock, but it is essential for a quantum computer.
The other work here at JILA that I hadn't mentioned, is really in super-conducting circuits, which also are sort of a platform for building quantum computers that we have people here working on, but again, with more metrological focus. Like Konrad Lehnert here works on super-conducting circuits, and builds really great amplifiers that are good for sensing very small amount of noise—or, sorry, very small signals. They also find use in readouts for quantum computers.
ZIERLER: If we can imagine a Venn diagram, two really broad definitional questions, there's quantum information and there's quantum computing. What's the shaded area and what are the unshaded areas?
SMITH: I'm tempted to say that the notion of quantum information science is a bigger set, with quantum computing as a subset of it. That's probably controversial.
ZIERLER: You would say then that all of quantum computing belongs in the larger discipline of quantum information?
SMITH: Quantum information science, I would put it.
ZIERLER: Other nomenclature questions, additivity, what does additivity mean?
SMITH: It means when I have two separate resources, and I have a way to quantify how good they are, I can put them together, and the joint resource is better than the sum of the individual—oh, wait, sorry. It means that if it's additive, it means that the joint resource is equal to the sum of the individual resources. So, if I have 10 dollars, and I add 5 dollars to it, the value is 15 dollars. It means things interact in a very simple way. That's additivity.
ZIERLER: Where might that slot in, or how do you use that as a tool, for quantum information?
SMITH: That, so far, has been a necessary a property of communication channels, in order to allow us to quantify their value for communication problems.
ZIERLER: What about quantum anealers, what is that?
SMITH: There's a historical answer and a technical—the technical answer is, they are devices that are meant to implement a version of the adiabatic quantum algorithm. They are built without much regard for whether they can maintain quantum effects in them. A really great quantum annealer, which I don't believe people have actually made, would be something that coherently—has high-fidelity qubits, and you can use as a way to find the solution to some combinatorial optimization problem, by gradually transforming the Hamiltonian of the system from a very simple one to a complicated one. It's a way of mimicking, well, the simulated annealing algorithm from the 80s, where you start with a random choice for basically the ground state of some Hamiltonian you're trying to find, and then you slowly—you mimic some temperatures, you try to flip the spins if that gives you a better configuration than the original configuration you had. You start at a high temperature and gradually ramp the temperature down. The hope is, that finds the ground state of the Hamiltonian. A true quantum annealer is a quantum version of that algorithm.
ZIERLER: You mentioned that there's a historical side to this as well. How far back does this go?
SMITH: Oh, not that long. Just historically, people started thinking of quantum computers when there was this company that wanted to have a quantum computer, but didn't have a quantum computer. They just threw together a bunch of stuff and said it was a quantum computer, and the algorithm they ran was the quantum annealing algorithm.
ZIERLER: Graeme, the phrase quantum internet, something that you've worked on, is this something that's developing in parallel with quantum computers, or is it sequential? In other words, we can't really think about a quantum internet, until we have quantum computers to create that network.
SMITH: No, you can think about a quantum internet, without quantum computers, or without big quantum computers. You can imagine a network with a few qubits here, and a few qubits over there, and very limited operations that you can do here, in each node, and they're all connected by optical fiber. You can imagine having a big quantum network with very primitive nodes. Given that, you wouldn't be able to do very much computationally, I guess, but you maybe be able to do stuff like key distribution or generating entanglement for other reasons.
ZIERLER: Your work on the mathematical properties of entropy, how does that fit in with your overall research agenda?
SMITH: When you're trying to quantify the capacity of a communication channel, it's almost—well, it's always in terms of entropies. The goal really of my investigation of properties of entropies is essentially always to understand what's the—how the mathematics can be used to better understand the capacities of communication links.
ZIERLER: Sort of an overall question to your research motivations, when, for you and your lab, are the most important questions simply fundamental research, and when are you thinking about practical applications that serve as a framework for the kinds of things you want to think about?
SMIITH: I think the lesson of information theory, and certainly of quantum information theory, is that if you pursue fundamental questions, and really try to push the limits of what an information processing system could do, in principle, you inevitably shed these extremely useful ideas, that can impact practice. Possibly in the future. At this point, always in the future, but that can kind of guide our expectations and our hopes for what quantum technologies might become.
ZIERLER: A question about thinking about practical applications. Of course, there are major corporations that are involved in building quantum computers. To the extent that it's a horserace, if that's the right metaphor, that companies like Amazon and Google, IBM, Honeywell, the list goes on, are all racing towards a quantum computer—to the extent you're paying attention to these developments, and the different ways that each of these companies are going about that, does any one approach or technology or theoretical proposition seem most exciting to you, or most realistic?
SMITH: Yes. I would say, if it's a horserace, the horses are all in the barn chewing oats. The companies that recognize this are the ones that are most promising. So, I would say Amazon, is really doing things right, in terms of basically, if I can say this, a bullshit-free way of doing things. Essentially every other company is—over the past years, people became more and more comfortable using the word lie to talk about when people lie. I'm not sure I want to say that they're peddling lies, but they're peddling untruths, let's say.
ZIERLER: There's a lot of hype in this world, you're saying.
SMITH: Well, it's not hype, that I would say. I would say, it's deliberate misrepresentation of the situation.
ZIERLER: Huh. That the field is a lot more—advances are not nearly as far along as these companies would have us believe.
SMITH: That's correct, yes. We all know that a quantum computer, that would have any kind of economic impact or any kind of use, is ages away, decades away. We all know, also, that the—well, let me give you an example. I can't remember the company, but they're partnering with some car company to develop better batteries, or something. They made a huge announcement about this great new partnership, and finally you dig deep enough into the article, they're running something on—no, they hope to be running something on ten qubits or twelve qubits, with hundreds of gates, in order to solve this problem. That's a very hard experiment to do, and I hope that they can manage to do that experiment they hope to do. It can offer nothing in the way of developing batteries for cars. Computations that small, with that few gates, are easily—I mean, I'm a terrible programmer, I could code it up, and I could simulate it on my laptop. The path to an actually useful quantum computer is much longer than what they're representing.
ZIERLER: In this context, usefulness obviously has an economic component. What about some of the exciting prospects that a quantum computer might have for physics research itself? Do you see what you're doing, what's happening at Colorado, at JILA, as contributing to some of those questions?
SMITH: Yes. What I'm doing, I would say, probably our work looking a little bit at error correction. Both in the context of communication, but also in the context of computation, may well be helpful in that way. The work that Ana Maria Rey is doing with analog simulations, using Jun's clock basically, I think will have much more impact on physics in, say, the next ten years, than quantum computing efforts will. There's sort of this division between analog simulations and digital simulations. Analog simulation, well, you have this really great device. You don't have full control over it. You can't do an arbitrary computation. But you can give it the Hamiltonian that you're interested in, and you can learn a lot about that Hamiltonian in that system. I would say, that path is going to be much more promising than trying to do things on explicitly built quantum computers.
ZIERLER: Let's take it back to an historical context. I'm very interested. When you were at the University of Toronto, this is the very beginning, the late 90s, the early 2000s, this is the very beginning of people starting to talk about quantum information. Were you aware of those developments, were you aware even before you got to Caltech, of some of the things that John Preskill was thinking about at that time, for example?
SMITH: No, not really, no.
ZIERLER: Was quantum information a thing, even, at the University of Toronto? Were you aware of it as a field at all?
SMITH: I think Daniel Lidar was there, in the Chemistry Department. I might have gone to his talk, but it didn't interest me in any way.
ZIERLER: Going to Caltech, for graduate school, quantum information was not on your radar, was not the motivating factor?
ZIERLER: What did you go to Caltech for, what was the game plan, from the beginning?
SMITH: From the beginning. It depends when the plan was. I hoped to do high-energy physics, high-energy theory, when I went there. But, in my final year of undergraduate, I kind of soured on that idea, and began looking around for other things to do. I poked around to all the graduate schools I had been admitted to, to see who would let me defer for a year, if possible. I was going to figure things out. Many of them were kind of fussy about it, but at Caltech they said, "Sure, just let us know whether or not you're coming before September, and that's fine." So I thought, OK, I don't have anything else to do. I didn't have a good plan, so I said, I'll go to Caltech. They'll pay me to be a TA, and I'll take some classes or something, and I'll figure things out there. Then, there was a process. But, the flexibility of Caltech was basically it. It was like, one other place said, "Here's this formal thing you have to sign, and you have to let us know by this date, and this is how it's going to work." At Caltech, they didn't care. If I was going to show up, I could show up. If I didn't feel like showing up, just let them know.
ZIERLER: Was the IQI operational by the time you got there, or how did you become aware of the IQI?
SMITH: It must have been operational, but I was really not paying attention to anything in the Physics Department. I went and took a bunch of classes—actually, my first semester there were no physics classes. I took some CNS, computational neuroscience classes, and I took a political science class, and maybe one other, I forget what. I was just really not planning to stay. But, one of my friends, Ben Toner, he went deliberately to work with John. I guess he must have been already working at the IQI, in our first year there.
ZIERLER: What was the initial point of connection, what was compelling to you when you first got to know IQI and what was happening?
SMITH: The thing I really liked was, it was sort of an enormous group, and it seemed like the people in it were a lot of fun and very collaborative and nice. That's really—and actually, the science that they were doing, as far as I could tell, seemed mathy and interesting. I sort of poked around to see if there would be something there, and eventually there was.
ZIERLER: What did you want to work on, when you got the sense that this would be your speciality, this would be your focus?
SMITH: I just wanted to find something I could do. But I talked to a lot of the people there. I mean, at the time, I don't know what it's like right now, but at the time, basically, there was lots of postdocs, maybe five postdocs or more, and you would just sort of go around and talk to them, and see if they had any ideas you could work on. Then, if something sounded interesting, you would do it. I did that with a few people, and eventually I found some people to work with. Debbie Leung and Patrick Hayden. They had some problem that seemed really cool to me, involving random matrices and stuff, and a lot of math I didn't know. So, I started working on it.
ZIERLER: What was interesting, what were some of the big ideas at that point, or even the frontiers of knowledge, if you will?
SMITH: What was happening, right then, was people were—let me say something about information theory. Information theory, do you know about information theory?
SMITH: Good. The coolest thing about information theory to me is this random coding strategy. If you take a classical channel, and you want to know what it's capacity is, you just kind of optimize over the input distribution. You can choose a code randomly according to that distribution, and that's just optimal, if you choose it of the right size, you achieve capacity and you can't do any better. What people were learning right then was, how to do this in a quantum context. It's not as simple—well, it is as simple when I use words. The math of how that works, people kind of had been struggling with. There was, basically, Holevo, Schumacher, and Westmoreland, knew how to do this classical signals—or, sorry, for quantum signals, but classical messages. People had feelings about how coding could work for choosing random quantum error correcting codes, but hadn't really mastered, until right at this point, how to show achievability of coherent information. Which is one of the—coherent information is what you get if you choose random quantum codes, kind of in an analogy with classical constructions.
I think, it would have been in 2002, Peter Shor gave a talk at QIP about the achievability of coherent information, and people believed that was a correct argument. It hasn't ever been written up, because shortly after that, Igor Devetak came up with a really crisp, rigorous proof. Seth Lloyd had sort of conjectured this value for—or this achievability of coherent information in late 90s, but again, they didn't really have the tools to deal with random coding at that point. People weren't able to establish it until, I guess it was—well, 2002 Shor, and then 2003 Devetak had done the argument.
The other cool thing that was happening right then, that is very connected, is Debbie and Patrick and—well, a variety of people—Debbie and Patrick and maybe some of the IBM crowd and Peter Shor, had really begun to show that when you look at a random quantum state, where you have fixed dimensions of—random bipartite state where you have fixed dimensions on A and B, and some environment E, they really were learning new things about the entanglement in a typical state. This is the aspects of generic entanglement paper. I don't know if you know it.
ZIERLER: What was the research culture like at the IQI? Were graduate students working with other graduate students, with postdocs? Where did people like John Preskill and Alexei Kitaev interface with students? How did all of that work when you were there?
SMITH: It was mostly graduate students working with postdocs or other graduate students. Both Kitaev and John, they had their work, that they were working on. Some people were collaborating with them on that. But, if you were working with some other postdoc, generally he wouldn't be literally collaborating, but he would be aware of what was going on. I guess the point of contact for everybody would be the weekly group meeting on, I forget what day of the week it was.
ZIERLER: I think it was Wednesday, I keep on hearing Wednesdays.
SMITH: Right, you're right. Then you had to present what you'd done that week. People would ask questions about it. You had to try to make it sound like it had been productive and was interesting.
ZIERLER: Graeme, what was the process for weaving all of these ideas into what would be become your thesis research?
SMITH: For me, one of the biggest—I was working with Debbie and Patrick, and I got very interested in this additivity question for coherent information. Which led to me doing an internship at IBM Research, with Charlie Bennett and John Preskill—no, Charlie Bennett and John Smolin, sorry. That was kind of the thing where my research started to actually work, and things started to actually go well. It was a group where, I guess, their style was very informal, and very, maybe, qualitative, I could say. That worked well for me. We wrote several papers. John Smolin became one of my best collaborators. I don't know, I just worked with them, I worked with Debbie and Patrick, and I wrote a few papers. In terms of thesis, it's literally you staple the papers together, and that's it.
ZIERLER: What were some of the unifying themes from the different papers, if there was a big idea that motivated all of the different things that you were working on?
SMITH: The big idea that motivated all the different things I was working on was that, well, Igor and the others had shown the achievability of coherent information for quantum communication, and the non-additivity of coherent information really added a richness to the quantum capacity problem that wasn't present in the classical case, and, in fact, that we still don't understand. The theme was kind of coming at this non-additivity of coherent information from different directions. I guess there was a little bit of additivity stuff too, showing if we had this simplification, if we had this, I guess, simpler form than the coherent information, we would be able to just have a clean additive theory instead of non-additivity.
ZIERLER: In what ways did you see what you were working on as slotting in or responsive to some of the broader questions that your fellow graduate students, postdocs, what was happening at IQI more generally in those years?
SMITH: There were several exciting things. I think it fit right in the center of what Patrick and Debbie were doing on understanding properties of entanglement in high-dimensional spaces, and more generally, in the project of turning Shannon theory into a fully quantum theory. That's where it sat, I guess, to me. There were a lot of cool things happening, that I wasn't really doing. Like, Guifré Vidal and Frank Verstraete were discovering, or rediscovering DMRG and then generalizing it to matrix product states and PEPS, and stuff like that. I watched that and was interested. I didn't do anything in that area.
ZIERLER: Graeme, what was John Preskill's style like as a graduate mentor? How often would you contact him, how hands-off or hands-on would he be, how did that work, that relationship?
SMITH: Extremely hands-off, except for the fact that you'd go to group meeting, and he's going to ask you what you've been working on, and we felt very much like we had to have something substantial to say. One week, if you've done nothing, that's fine, but a few weeks, and you better have at least done a little bit of work. He seemed to be paying attention, and knew about that kind of stuff. When I was just starting, I kind of got stuck. I talked to a lot of different people, and didn't really have any good—any projects I could work on, or I could get to work. I later learned Debbie and Patrick, as a joke, gave me an extremely hard problem, that's still not solved. That didn't work either.
ZIERLER: Which problem is that?
SMITH: It's kind of a Hamiltonian simulation problem. There's a really nice clean answer for two qubits. They said, generalize this to higher dimensional systems, and there's not a clean answer, and it's not very nice. You basically cannot solve it.
Anyway, I was very, very frustrated, so I sent John an email. Of course, I sent one of multiple paragraphs, or something. "Can we meet to discuss some like maybe new directions I could work on?" We met, and he said, "What have you been doing so far, first of all?" So I told him. Then he said, "What seems interesting?" I told him some of the things that I thought sounded interesting. He said, "That sounds good, you should work on those." I had been hoping, right, I show up, he says to solve this problem, I take the problem, I solve it, I give it back. But that's absolutely not how John worked. At the time, he was at a very extreme end of hands-off, but, also at the time, he was very open to—he would take you as a student. That was another reason I joined his group, because there was no—there didn't seem to be at least—any hurdles. If you go and say, "Hey, I'm interested, can I work in your group?" he'd say, "Sure, come to the group meeting." That's the end of it. Many other people you have to prove that you have some qualifications or capabilities, but that wasn't required. But, the flipside of that was that, nothing was provided.
ZIERLER: Looking back, I'm sure that has yielded benefits, just in terms of you figuring out how to do stuff on your own.
SMITH: Sure. It's not quite how I advise, but there's a touch of that. Because struggling through problems is—you're going to have to do it at some point. Might as well do it, I guess, early. The reason it did work, though, anyway, was that the postdocs, they were very happy to give you a problem. They wanted help with their work. It was good for them to get experience mentoring, and stuff like that. I think it wouldn't have worked, and probably wouldn't have been this way, if there were no postdocs around.
ZIERLER: To go back to this idea that it's worse than hype in terms of how realistic, how feasible, a scalable quantum computer is, circa 2022, looking back to your Caltech years, what was the level of optimism, or even exuberance, about the creation of a quantum computer, relative to now?
SMITH: It was believed to be hard. I would say that most of the people I knew in the—I guess, what I want to say is, the quantum computer isn't the point. The ideas are the point, and the ideas have already transformed physics in ways that are more substantial than anything that's going to come out of a quantum computer, as far as I would guess. Like, if there's no quantum computer ever, we still have a great new way of understanding physics, that's basically providing a new language to describe stuff that we couldn't even have imagined 20 years ago.
ZIERLER: That statement, what would be the banner headline that people need to appreciate, in terms of the things that quantum information has pushed forward in physics?
SMITH: The banner headline, as in like, what are some—well, OK, how's this, entanglement is everywhere. It's literally in everything. In condensed matter systems—people in condensed matter, some of the most exciting stuff on the theory side, is just understanding entanglement structure in material systems. Of course, there's the high-energy connection, that I'm sure you're talking with other people about. To me, I guess, the thing that I've learned that I didn't know at the time, is just the extent to which it is really all entanglement everywhere, even now. Even in your everyday life, it's just a matter of, some of it sometimes it gets hidden because of the way we interact with the world. I think that is far more interesting than estimating reaction rates of different kinds of chemistry problems.
ZIERLER: Graeme, the idea that now we appreciate that entanglement is everywhere, so before, 20 years ago, in the 20th Century, what would we have said as an alternative? Entanglement is somewhere? It's nowhere? What's the distinction?
SMITH: We wouldn't have said entanglement. We would have said the world is classical, when it's macroscopic. When you're dealing with atoms and stuff, you better use some quantum mechanics, but observers always collapse any kind of quantum states before they can get too big. Literally, I don't think I—I maybe heard the word entanglement once or twice, when I was an undergraduate. It just was not that relevant. It was something that these annoying foundations people who want to talk about what's the nature of reality, always bring up. But, I would have told you, it's stupid, use the wavefunction. If you have a mixed state, which is really part of an entangled state, you're doing it wrong, you should use the Schrödinger equation, and use the whole—just include as many systems as are required, before you have a proper wavefunction.
I don't know. Maybe also that's me. I know there were people in the 80s and so on, that had more sophisticated understandings of entanglement, but there weren't that many, and there wasn't much beyond qualitative discussions of how spooky it is, and how weird it is. It was considered this thing that if you really forced yourself, you could get into this regime, and I guess that had some spiritual meaning, or something. I don't know. Even people who don't work on quantum information would think it's crazy to try to talk about quantum mechanics without entanglement now.
ZIERLER: Interesting. Besides John, who else was on your committee?
SMITH: My manager at IBM—I went to work at IBM afterwards, out of Caltech. He described it as an inside job, because John Smolin from IBM was there. Igor Devetak was at USC at the time, but he has just left IBM. I had two out of five from IBM, and really more my buddies than impartial observers. Then, there was Oskar Painter. He was my roommate's advisor, so I thought, he's a good person to ask. And John. Then, Bob McCleese was supposed to be on my committee, but had a medical emergency on that day, so he didn't show up. Actually, one thing that John—I was very impressed, basically John called the dean and got permission immediately that we could proceed with the committee, even though I didn't have the right number of Caltech faculty on it. Because with these outside guys, we only had two Caltech faculty, and you're supposed to have three. I was worried, oh, shit, I'm going to have do this some other day. I'm so tense already. But, it went fine. Then, usually they're closed—well, they are closed, in the Physics Department, defenses, but Debbie Leung was in town from Waterloo, where she had moved, and Aram Harrow just happened to be visiting, so he came too. They were all there. They were both there. Also, two IBM people.
ZIERLER: Now, was the IBM postdoc wrapped up before you accepted to be at Bristol for a year? Meaning, you spent, it seems like it was almost like a gap year, between Caltech and IBM.
SMITH: Yeah. Basically I narrowed it down to, I want to go to Bristol or IBM, for postdoc, and I went to Bristol, and almost immediately decided, I should have gone to IBM, and got to work on going to IBM instead.
ZIERLER: Was Bristol a center of quantum information, what was happening at that point?
SMITH: It was amazing. Aram Harrow was a professor there, at the time. Andreas Winter was there. Those are the two I went to work with. There was also Richard Jozsa, and the other people who are still there, like—who else?—Noah Linden and Sandu Popescu. So, in the UK, it was the place.
ZIERLER: But for you, what was the deciding factor, where you realized you should have gone to IBM initially?
SMITH: There was a bit of a culture shock, for me, in terms of just living in England, and how things were expected to be done there. At Caltech, postdocs were extremely independent, and more or less, worked on what they want, and were more or less senior people. At Bristol, that was very much not the case. I knew that it would be like that—it would be much better—well, I knew I would be basically an independent researcher at IBM, so I arranged to go there. In fact, when I was employed by Bristol, I spent a good part of the time visiting IBM. I don't know, just somehow it didn't work out very well in Bristol, for me.
ZIERLER: Did you defer your acceptance at IBM, or you had to reapply?
SMITH: I think they hired somebody else. I mean, I had to reapply. It was more like they had to get another line that they could use, and that took some time.
ZIERLER: What was happening, when you get to IBM in 2007, what's happening in quantum at IBM? Is there a center, is it a division? How big is it?
SMITH: It's a tiny little group of really fantastic people, working on what they believed to be, or the managers thought of as, very long term, distant future stuff. So, the people there were Charlie Bennett, David DiVincenzo, Barbara Terhal, John Smolin, and Sergey Bravyi, who, you probably talked to him. They were the staff members, and I was a postdoc, and I think Panos Aliferis was also a postdoc there, at the same time, or—I can't remember exactly how that worked. They had one postdoc, and all these people—oh, that's the theory group. There was a very small experimental effort, maybe two research staff members and a postdoc, or something like that. The theory group was amazing. People were just constantly—they would have ideas, and you could just, you'd sit there at lunch, and suddenly you'd have three things to work on, and you'd go work on it. There was just a lot of creative ideas being generated in really abstract—well, in information theory side, and also Barbara was doing some really cool computational complexity stuff, too.
ZIERLER: Now as you said, this is far-reaching stuff. I assume that you're operating really in a fundamental research environment. There's no discussion about how any of this work might be applicable, in an economic sense, to IBM's mission?
SMITH: It was always required that you be able to tell a story about how there's a path from what I'm doing now, to possible impact on the business much later. But it was understood that much later meant decades. It was a small enough group that it was tolerated that we do basically fundamental stuff, even curiosity-driven stuff, as long as it's in the general area of quantum information.
ZIERLER: Graeme, I'm curious, at IBM, given that IBM institutionally has been thinking about these things arguably longer than any other institution, I wonder if that was something that you felt, when you got there? That there was a history to these idea, unique to IBM?
SMITH: Absolutely. There was a straight line from Rolf Landauer, who, he had Charlie Bennett as a postdoc, who was still there, and was actively involved in the research. There was a flavor of work that was done, and also sort of a way of working that was very different from what was going on elsewhere. It was pretty non-technical, and very conceptual, but ultimately there was math and physics underlying it. But it was really very much the sort of stuff that you could do a lot of it in your head, and thinking about things in a somewhat different way. That was the thing that really made things work for me, when I got there. I kind of have tried to adopt. I would say, Charlie and—yeah, that's very much a reflection of Charlie, and also of David, who had been there ages too.
ZIERLER: Were there new ideas or approaches to quantum information that you were exposed to, at IBM, that you had not seen at either Caltech or Bristol?
SMITH: That's hard to answer, because, basically from my fourth year at Caltech, I was in constant contact with a group at IBM. The way I would think of it is, it's sort of playing games with quantum theory really, to see what the structure looks like, and just saying, "Hey, look, what if there are these three people that are trying to do this thing, and there's this wrinkle that this person's trying to get access to the thing, but the other one doesn't—" It really was a lot more about stories, about people just trying to play funny games. That was, I guess to me, the difference that was there. I think I'd put it that way.
Basically, when I was thinking, sure I could go to IBM or Bristol. I was asking Debbie, because she knows very well both places, or knew very well both places. Somehow, basically, I asked something like—she was telling me, "IBM, it would be a very good place to be." I said, "Oh, good, do you they have different techniques that I could learn from them?" Meaning, ways of analyzing random matrices or something. She just kind of stopped and said, "No, they don't use techniques there so much as thinking about things in a different way." I decided what I really needed was—well, I went to Bristol because I thought, what I'm going to do is learn properly some rigorous new powerful techniques, and then I'm going to go solving problems. But that's actually, I don't like that model of working, in turns out. Techniques are fine, but I think clean ideas are better.
ZIERLER: Is this the first time you worked on cryptography, when you were at IBM?
SMITH: No, when I was a grad student I worked on looking at non-additivity in BB84 protocol. That's actually something that arose because Michael Nielson had run up a visiting fellows program for grad students. You just apply, and you can go to Australia for six weeks, or something. So, this was amazing to me. I'd never been to Australia. I went to Queensland, and lots of other people were there. I met Joe Renes, and he told me about cryptography, quantum cryptography, and how it's sort of connected to coherent information. We wrote a little paper together, with John, with John Smolin. That's the first time I thought about cryptography.
ZIERLER: When the postdoc was winding down, were you applying more widely, or that was a very easy transition into a staff position?
SMITH: I was applying more widely. That was the—I don't know what we call it. It was the global financial crisis, so you remember there was this meltdown due to housing, and stuff like that?
SMITH: I was applying, and Charlie and David were working to get me a job also as a research staff, and probably my first—I think, yes, my first choice was to get a job as a research staff member there. Ultimately it worked out, although it was much slower because of this financial crisis. IBM went into sort of a phase where they didn't feel like hiring people. But I was looking for jobs. I wasn't very successful looking for jobs. Nobody was really hiring in this area, at that time.
ZIERLER: When you joined the staff, did that effect the kind of research you were doing?
SMITH: Not really. I mean, a little bit. But, there was still a lot of freedom to be working on things basically from a curiosity-driven perspective, with the understanding that it should be generally in a direction that would be valuable to IBM. It was still, at the time, what you would generate, the value that you would generate, was, visibility for the lab and potentially useful things for later. That did begin to change around the time I became staff, maybe a few years later, and really transformed into, at this point it's a matter of, are you helping the superconducting circuits experiments improve? That's how I ended up over here.
ZIERLER: Now at this point, as I always try to do, I'm establishing the historical narrative. Where was your sense of progress, at this point, 10 years in, 15 years of the things that you had been working on? Where was the progress, and where was it simply, new projects, new research endeavors, that may not have necessarily been connected with what you had done earlier, and therefore, the notion of progress might not have been relevant?
SMITH: The progress that happened, I guess, somewhere towards the end of my IBM postdoc, was, we began to understand non-additivity. We began to realize—in my thesis, I had this formula that I thought, this captures the quantum capacity, I just have to show it's an achievable rate. What we discovered, a few years later, while I was at IBM, was, in fact—well, basically, if that had been the correct formula, then we would have had a nice additive formula for quantum capacities, and the world would have been simple and beautiful. The progress we made was to really discover that, essentially, all quantum capacities, all the different capacities for different kinds of communication, are not additive. Instead, better coding strategies than the random coding strategies seem to exist, or at least we don't have a good way to formalize the structure that needs to exist in quantum error correcting codes to make them really good. So, non-additivity of quantum capacity was something that was really exciting, at that time, and closely connected to my thesis. Something that happened shortly after that, was Matt Hastings showed non-additivity of Holevo information. Which was even more surprising to us.
ZIERLER: Why was that surprising?
SMITH: There was some forms of non-additivity of coherent information that had already existed from the 90s, that kind of told us, yeah, you're going to need to do better than random coding, and hinted at possible non-additivities of quantum capacities, as well. Until then, there had not been any examples of non-additivity of Holevo information. It was hoped that the Holevo information was, in fact, the classical capacity of a classical channel. There had been lots of numerical investigations that seemed to suggest—well, that showed additivity for many cases. There was a definite direction to the conjecture, and the conjecture was that it was additive.
ZIERLER: As Microsoft, and Amazon, Google, and others, started to jump into the quantum computer game, I'm curious how this registered for you, or for IBM, as these developments were happening?
SMITH: As these developments were happening, IBM was also growing substantially its experimental effort in super-conducting circuits. The key thing I think I should mention is D-Wave. We talked about them before. But D-Wave was out selling devices that were super classical, and saying they were quantum computers, and they seemed to be making money, and this was causing executives and managers to contact us and be like, "Hey, why aren't we also doing this?" Basically, the scientists had been telling them, "Nah, this is going to take a long time, it's going to be a hard slog." The super-conducting experiments were very primitive at the time. Then, there became a lot of pressure from the top to really find some way to make this commercially viable. That led to a really uncomfortable dynamic where, the more you would exaggerate about what quantum computers could do, or when quantum computers might arrive, the more success you would have, and the more resources you would get. This led to rapid growth in the experimental effort at IBM, as well. Also, I guess, the deterioration and perhaps elimination of actual science—or fundamental science happening at the lab.
ZIERLER: I guess then this would not be unconnected from your decision to join JILA and Colorado, in 2016?
SMITH: No, it's not. I saw I had to leave, and was very happy to find a position.
ZIERLER: What has come of IBM's efforts, since 2016?
SMITH: They made a website where you can go and interface with lots of their experiments. They've made really a large—they have a large effort to convince people that they should be interfacing with that website.
ZIERLER: Tell me about the opportunity at JILA and Colorado. How widely were you applying and was the market better than first time around when you were a postdoc at IBM?
SMITH: The market was better. I got to tell you about this fun thing we did at IBM—so John and I, there were these people who kept telling us, "Hey, D-Wave is really eating your lunch here." We would tell them, "Don't worry about D-Wave, it's not a quantum computer, forget it." They sort of didn't trust us in some sense. We had to do more—like, if I told you my socks are a quantum computer, how much proof do you need that I'm lying? Not much, but they kind of demanded more and more. Ultimately, John and I started writing these almost—well, sarcastic papers, let's say. Polemical papers. Where, every time D-Wave had a claim that they had some quantum effects in their device, we thought it'd be fun to say, "We also have quantum effects in our classical simulation of balls and springs." We wrote a series of papers there, that had some impact, because they sort of influenced the way people began to think about testing claims about quantum computers. What it convinced me, was that it's not a matter of people misunderstanding what the situation is in the quantum computing world, it's a matter of people really wishing the situation was different, and simply pretending. Evidence is not going to matter in this discussion of quantum computers.
I bring that up because those papers that we wrote, while we wrote them basically as jokes, did contribute to some visibility for me, and I think did help me find a position. The market was better when I went looking for a job. But also, I was in a better position, having done that research.
ZIERLER: To clarify, the joint appointment between JILA and Physics at Colorado, that was baked in from the beginning?
SMITH: Yes, basically, when we do a search here, the way it works is, if you do AMO in the Physics Department, you're probably going to sit in JILA. JILA will run the search. There are extra people you will talk to, to make sure JILA will be happy to hire you, as well as Physics. The joint appointment was baked in, that's right.
ZIERLER: Did you think about other industrial labs, or after that experience at IBM you specifically knew you wanted to be in an academic environment, or a university environment, I should say?
SMITH: I did two things. I thought about other industrial positions, sort of not in quantum computing. I started looking—we were near New York, I thought maybe I could work at a hedge fund. I started looking into that. Turns out I'm not a very good programmer, so they're not interested. Then, I was looking exclusively for academic positions.
ZIERLER: What was the game plan as you joined the faculty? You wanted to build your group—what were the big questions motivating you, circa 2016-2017?
SMITH: The same one that—in the information theory section, just capacities, understand quantum coding, and how to build better correcting codes for noisy quantum channels. That is, I would say, a long-term, slow-moving thing, that at the moment my approach is finding better and better, and cleaner and cleaner, examples of the effects that we're trying to understand. I hope that they will reveal ideas—no, they have revealed some ideas. But I hope that these examples which we've been amassing are giving us a better idea of how to think about quantum coding, that will let us really solve the quantum capacity problem. Now, that's not something I can't count on happening. So, another goal I had was really understanding the information theory of networks. Multiple users interacting through some communication channel. Then, I guess my third goal was, I wanted to develop some things to work with—I don't want to sit here and be in a vacuum. There's not a—well, there wasn't, especially then, a significant straight quantum information effort. I wanted to develop connections and collaborations with people at JILA. So, more metrology stuff.
ZIERLER: Did you take on graduate students right away?
SMITH: I hired a postdoc right away. I had some graduate students by the second semester there.
ZIERLER: As you were putting together your own group, thinking about your postdoc at Bristol and IBM, being a graduate student at Caltech, what were your goals in terms of being a mentor, providing a framework for the next generation to start thinking about these things?
SMITH: I wanted to, I guess, make sure that many of the ideas that we had developed continued—basically, I wanted to make sure Shannon theory continued to be pursued in a sort of vigorous way. Because it's hard and slow, and also not directly connected to corporate—
SMITH: —interests. Profits, yes. I wanted to make space for people to work on, I think, more important and more fundamental questions. I wanted to I guess make sure that they were—basically, put my students and postdocs first, rather than try to sort of grind them up and turn them into papers. I thought quite a bit about it, and there's definitely a hint of my advisor's hands-off approach, but I didn't fully take on that idea, and I don't think it would have been a good idea if I did. So, we collaborated, but I tried to make sure that people were able to follow their interests. Perhaps that's because I was leaving a place where I was certainly not able to follow my interests.
ZIERLER: Graeme, I'm curious what interactions or opportunities you have to work with undergraduates at Colorado?
SMITH: In terms of research?
ZIERLER: Or teaching.
SMITH: Obviously I teach with them a lot. I've developed some courses on—we have a third-year undergraduate course on quantum computing, that I put together with Alex Kolla and Murray Holland, here. That's been a lot of fun. I gradually am, very gradually—my approach to teaching, to see if I could teach very well, was, I'm going to do advanced classes first, and keep getting more and more basic, to see if I can still manage first-year students, or something. I'm doing a similar thing with undergraduate research. I'm just beginning to do undergraduate research with the students, because I feel comfortable now that—basically, I was sure I could handle a postdoc, I learned how to handle graduate students, at this point, I think I can manage undergraduate research as well.
ZIERLER: I'm curious, teaching an undergraduate course on quantum computing, just as a sense of how mature the field is, or your expectations. What grounding do students need to have in order to succeed in a third-year quantum computing course? What are the things that you expect in terms of knowledge or prerequisites?
SMITH: We had a pretty firm conviction that the prerequisites would be linear algebra, and that's it. So, we have students from the Physics Department and students from the Computer Science Department and from the Math Department. We totally screwed up at the beginning, because they all had classes in linear algebra, but they didn't know any of the linear algebra they would actually need to do the class. We ended up introducing into it a two-week module, just basically introducing, reminding everyone about eigenvalues and eigenvectors, and diagonalizing things, and unitaries, and that's sort of it. I guess the prerequisite that we have is still just some class in linear algebra, and a willingness to do this boot camp at the beginning.
ZIERLER: Graeme, bringing the conversation right up to the present, just in terms of something we're all dealing with, the pandemic. Has this been a productive time, just being more at home, having more bandwidth perhaps to think about these problems, or not necessarily?
SMITH: It's been deeply unproductive. I have a 6-year old and a 3-year old, and my wife has a job that is, I guess, far more stressful and rigid in terms of the time commitments. I have been surviving this pandemic only because of my students and my postdocs and collaborators. They are being productive. I am almost an observer. That's all I can manage, I really have discarded almost—well, I'm getting back to it actually. This semester I don't have to teach. My kids, one of them is in school now, and next week, the other one will be back to school. This is my recovery semester, but it's been—nobody I know got sick, nobody's died, the coup failed. In terms of the whole world, I'm doing pretty good, but it's the worst time I've ever experienced.
ZIERLER: Yeah. That's a great point as we close out our conversation, a few questions looking to the future. Hopefully, for everyone's sake, for you in particular, this will be a recovery semester. What are you looking forward to jumping back into, what's next on your plate?
SMITH: I've got an idea for an actual way to develop a coding strategy for quantum channels. It's based on an old classical result of Marton, on how to code for broadcast channels in a novel way. Broadcast channels is two receivers, one sender. That's relevant, because a quantum channel just does have two receivers and one sender. The two receivers are the output of the channel, and the environment that's screwing up all of the inputs. So, my hope is—well, there is a qualitative analogy between broadcast channels and quantum channels. I'm specifically going to try to develop a coding strategy based on Marton's coding strategy, to turn it into a quantum coding strategy for quantum channels. That's my excitement about doing something technical.
ZIERLER: Who's funding that?
SMITH: Who's funding that? First of all, I don't need somebody to fund what I think about. Which is amazing to me. Anyway. Let's see, it sort of overlaps with two of my projects, so I guess there's an ARO MURI on quantum networks that I have a small part dedicated to novel coding strategies. Then, there's a CAREER Award I have from NSF, that, non-additivity is kind of at the heart of it. So, NSF and ARO.
Let me think about what I am excited about. I am also excited that this paper on sensing magnetic fields has been in the works—actually, I started it before I came to JILA. I met James during my interviews, and then he contacted me, and we started talking, and we had this idea, seven years ago. Gradually, it came to fruition, and we managed to actually solve the problem. We were kind of stuck for years, because I had the wrong conjecture, and luckily, one of my students went and did an internship at Honeywell, and happened to be talking to them about some noise spectroscopy stuff. They came up with the strategy that just sort of broke the conjecture, and also made just everything fall into place, and the paper will finally come out, within a month or so.
ZIERLER: Oh, wow.
SMITH: What I want to do with that, is look more into whether it can be applied—so the point of that is that there are strategies that give you broad sensitivity across many frequencies, that seem to outperform standard strategies people are using to do magnetic field sensing, and we're wondering whether that'll have applicability to axion detection. Basically, there's a lot of physics I have to understand, because I'm less detail-oriented, as I guess I emphasized with this idea of how IBM worked. They do very simple things. But to know whether this is actually going to be a useful thing, we're going to have to do some details. So, I'll learn some physics and see if this can be helpful with axion detection.
ZIERLER: Is that to say that you might be jumping into the dark matter game, to some degree?
SMITH: Yes, basically in a very distant way. I mean actually, we do have dark—Konrad, here at JILA, works on, I forget what the name of the thing is, but he builds the amplifiers for the dark matter detectors, because qubits it turns out are pretty good—or, at least super–conducting resonators are very good sensors, because what an axion would do is basically generate a little bit of extra magnetic field in there, and sensing that magnetic field is exactly what Konrad's amplifiers are good for. That's really dabbling, but I think it could be—look, if we find a way to speed up the searches, that would be amazing. I think there's a non-zero probability, not fifty-fifty probability, but a non-zero one, that we can actually offer some insight into how to do those searches.
ZIERLER: Finally, Graeme, last question that will have both retrospective and a future-looking perspective to it. In thinking about Caltech, IQI/IQIM, at 25, I wonder if you can reflect a little bit about, at the institutional level, reflecting on your own experiences, those of your colleagues, those who have passed through the program, those are there right now, what do you see broadly as IQI/IQIM's contributions in all of these areas—quantum information, quantum computing, and all the things that people are doing there?
SMITH: Early on, it played an essential role. It was the place to go for people if they were doing a postdoc. You look at any of—well, not any—the postdoc market was, can you get a job at Caltech or are you going somewhere else. Many of the best people went through there, and it was very important that it existed at the time. This work that got done in the early years was really quite amazing. I don't know, you just look around at the people who are leaders now, and chances are you find they did a postdoc at Caltech at some point.
There are other places that now do quantum information. One of the things it did, and I think John has done, is help make it a little bit more—well, it is more respectable. It's OK to do quantum information now, it's not crackpot stuff. That wasn't the case everywhere, but it was the case at Caltech when I was there. In terms of now, I think they're doing really—and I'm not sure the extent to which—from the outside, I don't see quite how the division between, like, what Amazon is doing and what IQIM is doing. It's a little bit blurry. I think that's an amazing thing too. Because, every other company is essentially going it on their own. What Amazon is doing, I guess, is they're doing their program and they're doing it in tight collaboration with really the best scientists. I think it's demonstrating a different model for how you can approach doing industrial research. Not just in quantum computing, but just overall. I think it's ensuring a high-level of integrity to the work that gets done. It's providing a model for other places that might like to do a similar thing. I think that's a really important influence they're having. You might say, hey, there's all this sort of hype and stuff, what I'm going to do is, I'll do science, maybe I'll work on something else. Or, you can kind of drag the people doing the hype-y things into being honest, by working with them, and showing them how to be honest, and still make impact. I think that's an important role that the IQIM is playing now.
I think also, I guess, the students that go through there are also—at the time, when I was there, students who went and worked with John at IQI, it was kind of like freaks and losers. Like, they tried something else and it didn't work. I would put myself in that category, but it was like, very few people went to work there with this in mind. It was kind of a fallback. We maybe did interesting work, or maybe we didn't. But right now, the best students in the world are clamoring to work right there. I think, making sure they're educated in a way that allows them to critically evaluate progress in quantum computing, claims about quantum computers. They have role models who are optimistic and excited and honest, as well. That is something that, again—seeing that you can be gung-ho, really excited about a direction, and still not fall in love with the idea so much that you can't see straight, and you can't tell the truth, is really important for people who are going to be leading in this area, in the future. So, I think educating the students in a way, well, in the right kind of science, is another thing that IQIM can do. Or is doing.
ZIERLER: Graeme, on that note, this has been a terrific conversation. I'm so glad we were able to do this. I'd like to thank you so much.
SMITH: Great, thanks.