A Programming Language for Quantum Simulations with Xiaodi Wu
Welcome to the New Quantum Era. I'm your host, Sebastian Hassinger, and this is the podcast where I bring you conversations with leading researchers in quantum information science and technology. Today's episode comes to you from Albuquerque, New Mexico, where I'm attending the IEEE Quantum Week for 2025. This annual event, officially the IEEE International Conference on Quantum Computing and Engineering or QCE, has actually become one of my favorite conferences in the quantum field. The organizers have done a really fantastic job bringing together engineers, computer scientists, and the broader physics community.
Sebastian Hassinger:That intersection, that interdisciplinary interaction is so vital because the cutting edge sort of science advances in quantum information need to be paired with robust engineering in order to achieve sort of larger scale, more reliable devices and also to find applications in various fields for those devices. So next year's conference will be in Toronto. I'm already looking forward to it, actually. This is the last day of this year, and I'm I'm thinking about next year's already, which I think is a testament to the quality. So special thanks to Carmen and Hausi who helped arrange space for me to record the interview that I'm gonna bring you today.
Sebastian Hassinger:My guest is doctor Xiaodi Wu, associate professor at the University of Maryland in the Department of Computer Science. He is also a fellow at the Joint Center for Quantum Information and Computer Science or QUICS and an Amazon scholar. He's collaborating closely with the Amazon Braket team, part time. And that's actually how I got to know him personally through our interactions at AWS. Although I did know of him and I was following his research when I was still at IBM.
Sebastian Hassinger:Doctor Wu's work and his group span a broad range of projects at the intersection of computer science and quantum computing. With a background in both fields, Jiaudi has been focused on uniting the strengths of each domain to solve meaningful problems. In this episode, we're gonna dive into his group's recent release from 2024 of Simu Q, which is an open source domain specific programming language, which is designed to make quantum simulations more accessible and productive and easier from an end user perspective. Given that the most likely near term applications of quantum hardware is going to be simulations of systems from nature quantum systems. The timing and the focus of this work is especially relevant.
Sebastian Hassinger:So we're gonna talk about quantum simulations, SimuQ, about programming languages, about his own unique career path, and the challenges and opportunities of quantum simulation and the evolution of programming languages themselves. So I hope you enjoy the conversation. I sure did. Shadi, thank you very much for joining me. We're at the IEEE Quantum Week in Albuquerque.
Sebastian Hassinger:Thanks to the organizers for, for getting us this room to record. You've been in the field for a long time and you've been involved in a lot of different really interesting projects. I wonder, can you start just by sort of, telling us like a brief introduction and how you got to where you are right now?
Xiaodi Wu:Okay. Yeah. Yeah. Thanks for having me here, Sebastian. Actually, we also work together at Amazon.
Xiaodi Wu:That's right. For information. Yeah. So, thanks for everyone, lesson to the episode. So for, I'm Xiaodi Wu, currently our associate professor at the University of Maryland.
Xiaodi Wu:I'm part of the QUICS, which stands for quantum information and computer science, which is joint center between NIST and UMD.
Sebastian Hassinger:So
Xiaodi Wu:I think I got you into the field at the very early, I mean, me, in my undergrad, when that's like around like 2004 or 02/05, at that time I got into the program back in Tsinghua where I studied mathematics and physics, which, you know, very science oriented, right? But I did had experience, you know, do a lot of programming, computer science stuff in my high school. Right after got admitted to the program, Andy Yao, who was a Turing awardee, and he returned back to Tsinghua. And he actually, I think he started all the field about computer science and complexity in China, because before him, nobody really knew that field in China. So he has done a lot of introductory courses into the field.
Xiaodi Wu:I'm really lucky to get into that, his course and learn about all the stuff. By the way, also lucky enough to get to the lecture by Chen Lin Yang. He only did one lecturing in China on and fundamental I'm really lucky to sit in his class as well.
Sebastian Hassinger:That's cool.
Xiaodi Wu:And so, but then I quickly realized I want to do something that is computer science plus physics. And what's cool about that? So, you know, at the time there is a concept of quantum computing, which is exactly computer science plus physics.
Sebastian Hassinger:Yep. So
Xiaodi Wu:I don't think I really spend much time thinking about why I want to get in the field or what are the other alternatives. It's just very natural for me.
Sebastian Hassinger:Right, I
Xiaodi Wu:want do something out of this interdisciplinary field. And maybe as I, you you're gonna see my, you know, the research trajectory, I really enjoy being interdisciplinary and maybe too much interdisciplinary for now.
Sebastian Hassinger:So that's always a trade off, isn't it? Because you bring, different perspectives to sort of cross those boundaries. Yeah. But also it can be a challenge to keep focus on one thing to get deep enough to actually, like really make a fundamental contribution to that specialty.
Xiaodi Wu:Yeah, exactly. Yeah. But I think in my experience, actually really lucky to really get you in the field and also just witness the very huge growth of the field in the last decade. And that also is the reason I can work in a very interdisciplinary way. So when I finished my undergrad in Tsinghua and got into The United States.
Xiaodi Wu:I went to University of Michigan, I worked with Yao Yunxi. By the way, he actually now left Michigan, he's working in the industry now.
Sebastian Hassinger:Where is he at in industry?
Xiaodi Wu:He was the lead of the Alibaba, Quantum Lab, but the whole lab has, you know, is gone. So, but I think he's just continue working some other, in a way, which I don't, not entirely clear about. Yeah. But at that time, I very motivated to study the intersection of a theoretical computer Science and the Quantum. I think that's still super niche field at the time.
Xiaodi Wu:And Quantum Computing wasn't a real thing. Nobody knows, there isn't any machine. Right. And I think even at that time, a lot of techniques we're seeing today, superconducting, transparent, they're just reinvented. Yeah.
Xiaodi Wu:So nothing really
Sebastian Hassinger:When did you do your PhD? When did you get,
Xiaodi Wu:do you finish it? I got into the program in 02/2008. Okay. And then five years in Michigan. Yeah, yeah, yeah.
Xiaodi Wu:Okay. So, at that time I was really into Stereoctica Embrace Science and actually I did a lot of complexity theory, start with my work, basically I did like a summer intern with John Watrous. And that time, the biggest thing about starting the complexity for QIP, the And Quantum Interactive Proof the summer I got there, they actually had a big breakthrough. That's the QIP equals PSPACE result. But before I know about that, I look into the problem and I have my separate thinking about this.
Xiaodi Wu:I actually didn't tell John that I'm working on a little bit on that. I'm working on something actually much simpler, less ambitious. It was drawn and actually we didn't publish anything on that either. But then I tried to just think of this for myself. And actually right after the paper up here online, I just talked to John, so I have something different from what they have.
Xiaodi Wu:I think it turns out to be, I think, conceptually simpler proof. Interesting. So there is a paper I read about that, but it's something people already show. Right, right, Yeah, but just, it's concurrently. I didn't really ask him about what he's thinking about.
Xiaodi Wu:So I think this means independently, but just he had to pay for Interesting. But that's an amazing experience. I think that the time, capacity theory for content is really exciting. Or maybe because there isn't too much you can do. Yeah.
Sebastian Hassinger:Can sit in a room and think about it really hard.
Xiaodi Wu:Yeah, yeah. So I spent a lot of time in my PhD thinking about all kinds of complexity theory. So, there are things about a quantum interactive proof system, and then there are problems about how the entanglement between system can give different power over to capacity system. At that time, I was working on QMA2, quantum alerting also problem with the two perverse. I had some results later with both Yao Yun and also in my PhD, I had the result with Aaron and Anand.
Xiaodi Wu:That's my postdoc at MIT. So that's all like a topic that really excites me until I get to the point I think people see new possibility for other types since. So I worked a little bit on crypto, especially device independent crypto at the time, they called it like certified randomness by Nature Lords. Right. Yeah.
Xiaodi Wu:So that's actually a topic picked up by Yao Yun and, know, in the Carl Miller, who is also a failure at the QuicksNow, was working with Yao Yun on that topic. And I was also kind of interested in that. So we have a few projects going looking into the crypto stuff, especially about the randomness generation.
Sebastian Hassinger:And did that include sort of quantum sources of randomness? Yeah.
Xiaodi Wu:I think this is something super interesting from, you know, physics or, you know, some philosophical perspective. It's really talking about whether the nature is deterministic or you have inherent like in a randomness inside of And if it's the case, can you really certify that there is such Right. Right. And then the device
Sebastian Hassinger:It's funny, isn't it? Because I mean, on the one hand, it's
Xiaodi Wu:just a
Sebastian Hassinger:mechanical there's a need for random numbers to generate, cryptographic keys, etcetera. But on the other hand, it's it's like a like an ontological question about the universe. Yes. Whether there is such a thing as randomness or not.
Xiaodi Wu:So, of course, you know, we come you know, I think a lot of physicists also started this topic at the time that that was a hot topic. It has fundamental, I'd say, physics and philosophical interest, but also can lead you to something practical, right? So random number generators, and that's basically the key for building a really secure crypto system. I remember, you will see the certified readiness this year, our nature paper with JP Morgan running since the continuum, but using the theory from Scott Aronson and his team. Right.
Xiaodi Wu:But that time we were really looking for something with much higher security guarantee because it's a serial paper. Right. And we tried to see without trust anything about your devices, just looking at input output, what it can say. It turns out, after a few papers, if you don't have that, know you just need a little bit of randomness to start with, and then you can certify a tons of randomness out of the devices. I don't know whether that's something practical, but from the foundation's perspective, this is really amazing.
Xiaodi Wu:So you could say either the word is like a super deterministic, so nothing really is random, or you can just spend a little bit of randomness, you can certify in a basically infinite amount of randomness out of the word.
Sebastian Hassinger:That's interesting. Yeah. That's also, that's an interesting result. So that was during your postdoc at MIT?
Xiaodi Wu:Yeah, that's really near my graduation as PhD, and also I spent most of my time in my postdoc working on this. I mean, in addition to the capacity series,
Sebastian Hassinger:just Right,
Xiaodi Wu:right, right. Yeah. So that basically summarise what I have done in my PhD and in the postdoc.
Sebastian Hassinger:Right, And then you landed at UMD?
Xiaodi Wu:Oh, no, I spent like two years in Oregon.
Sebastian Hassinger:Oh, okay.
Xiaodi Wu:Yeah, before I joined UMD. At that time, you know, there also, the job market for quantum is pretty bad, you know?
Sebastian Hassinger:Dave Weineland is at U, Oregon analyst.
Xiaodi Wu:Yeah. I think Oregon has a very good experimental physicist. Now, Dave Weineland, actually he moved to Oregon after I left. That's too bad. I know he actually has a lot of connections.
Xiaodi Wu:He had lot of students at Oregon. So, I know that. But just like at that time, quantum computing wasn't really really hot topic.
Sebastian Hassinger:Right.
Xiaodi Wu:And still very niche. People don't know why they want to invest
Sebastian Hassinger:in it. It's only slightly less niche now, I would say. Yeah. But It's a lot more, but it's still slightly niche. Yeah.
Xiaodi Wu:I mean, I won't argue with that. But just like, there's national interest in that there. And definitely there are also tons of, you know, progress post-
Sebastian Hassinger:Yeah, absolutely. That's the exciting thing is how much has changed in the last ten years. It's incredible. Okay. So Oregon and then UMD.
Xiaodi Wu:Yeah. But that's the time I, you know, my research interests start to just a shift.
Sebastian Hassinger:Yeah. In one way.
Xiaodi Wu:I think this is a very From my inside, this is really something I had from my education in undergraduate, especially as a physicist student. Not just pure as a theoretical physicist, I always remember the professors that have been teaching us courses, they'll say, you should need to have a tooth fit for studying physics. So you need to understand theory, but also you want to see experiments. You want to see something real, but otherwise you're doing mathematics. And of course, mathematicians are great, but if you want to understand the nature, you need to have both.
Xiaodi Wu:So I think I have done a lot of theory. I really want to see how quantum computer can work in practice. And part of the reason I can probably see a little bit more than other computer scientists, because I have a physics background. So I didn't do a physics PhD, but I can try to understand what's really going on. And I understand those physics still fascinating to me.
Xiaodi Wu:So, and of course at the time, people have, the technique has matured to the stage where you do see some early prototype of quantum computer appearing. And then actually spend a lot of time looking into that and also need to think about what it can do. So that's like one motivation for just stepping out of just pure theoretic study, try to see what's possible. But still, it's different fields. And another motivation is that I actually had my undergrad advisor visit me in Oregon once.
Xiaodi Wu:He's working on the, he used to work on the logic, then formal method, program language, this kind of stuff, and he moved to work on this part of the sense for quantum. And before his visit, I wasn't really sure what one people can outside the theory for But computer I think he just showed me there are a lot of interests since that one can do, but not just only one can do, but one need to do, right? So in order to make the computer really working for the user to use it. There are many other sense in theory, because theory, you started very abstract, clean mathematical models, you know, something in principle, but if want to make sense work, are lot many other things Well, you need to
Sebastian Hassinger:is there a sense also that, I mean, you know, experimental physics makes devices that embody a particular experiment. Computing computing science makes a general purpose machine. Mhmm. And and there's there's there's sort of this phase of interplay between sort of experimental physics right now and computer engineering in a sense. Like, to help the physicists build these larger scale systems, we we sort of have to blend physics and and computer science, just like you were saying Yeah.
Xiaodi Wu:Yeah. But it's also interesting to just learn the history. Yeah, yeah, Because I think, I definitely, too young to know anything about history if I don't really spend time in looking to that, because nobody's really teaching a lot of things that we have done in the past. But if you look into the history, you're going to see, we probably start with like a special purpose machine. Right, you mean classical?
Xiaodi Wu:Even for classical, because people need to kill it up. People need like the resource to do that. They know when we do that, this can be put in some use case
Sebastian Hassinger:and then
Xiaodi Wu:you get the financial support.
Sebastian Hassinger:Right. Well, and it was, it was physics experiments in fact, right? I mean, ENIAC and Von Neumann's machine in Princeton were both being, they were both accelerators for physics computation, physics based computation. Yeah.
Xiaodi Wu:Basically for scientific computation at the very beginning. But, you know, people will buy that, right? So they have utility. Right. And I think, you know, this is something I learned, especially after I joined Amazon, I didn't really bring the economic into the picture, if I'm not really stepping into the industry.
Xiaodi Wu:Thought everything is academic, we solve the problem, nice. But then you realize if you really want to solve the problem, shouldn't forget the economies side of everything, because you need those resources. And you need the users and you need the people to invest into this. And then nothing will happen without any Right, of But anyway, I said, that's the fascinating thing I didn't know in my PhD. I got into maybe other part of computer science and then going from there, see not just computer science, maybe like electrical engineering.
Xiaodi Wu:And I basically look into everything we probably need in order to build our classical computers. That's not something people are teaching. You need to spend a lot of time really think about how this happened in the history. Of course, I really want to have someone really teaching me how this happened, or how, in a sense, technology just pivot from one to another. So, what's the motivation behind that?
Xiaodi Wu:It could be technology reason, or maybe because there's some other economic reason. But you do see
Sebastian Hassinger:Like market driven or technology. Yeah,
Xiaodi Wu:they're definitely not like one straight line. So there has been many pivots in both the way you build the classic computers and the application in the targets. So many things you can see. Of course, I still don't have a very good answer or understanding for everything, but you see, this is very complicated.
Sebastian Hassinger:Yeah. So you draw on the lessons of classical computing to sort of identify areas where you might want to do research? Is that sort of how you're
Xiaodi Wu:using Yes. So, I mean, of course, the field is huge. It's like just embodied in the entire computing industry. That's huge. I'm not saying I know everything.
Xiaodi Wu:No. I just say, okay, I just realized there's so many things that I don't know. Well, that's always a good thing for a researcher. You have your own curiosity, right? So you want to know.
Xiaodi Wu:There are so many things you don't know, right? So you want to just pick something to learn. But of course, I can learn everything, especially in my undergrad background, the knowledge I learned in my PhD, somehow limit what since I can explore. I only can explore since one by one, maybe gradually enlarge the scope of those things I'm doing. But for me, I think there could be some random walk into the field, but also I feel if you're looking into what the quantum computer is one day, that can be a reality.
Xiaodi Wu:What are the necessary parts are there, but what are the missing parts for today?
Sebastian Hassinger:Right, right, right.
Xiaodi Wu:So, indeed, there are actually many. They're not just one, There are many. But I just started to pick something that I think is more tangible for me to get started. And the first thing I get into is outside of theory is doing program language. I I think I learned a lot by just the understanding of field.
Xiaodi Wu:I still don't treat myself as expert in that field. I mean, I think there are tons of other classical knowledge about programming language that I don't know. But I gradually to see how that community thinks
Sebastian Hassinger:about
Xiaodi Wu:systems, what they think about computer science, for example, and what kind of research they think they'll be important.
Sebastian Hassinger:Did you start to design your own quantum computing language or were you using existing tools or other?
Xiaodi Wu:Yeah, mean, I'm not starting with designing my own language. You probably just mentioned CMQI, I can get to that point later. Right, right. Yeah. So I didn't start with doing my own programming language because I wouldn't know why want to start and design a new one.
Xiaodi Wu:So I start with some technical question. And I think that technical question is I shouldn't say that's like a first principle kind of question because there's a lot of technical questions we want you to address. The one I get started is to really helping people see how they can verify the quantum program they write is And before I really get to the point I didn't know, actually debugging, testing is a big market in practice. Companies spend tons of resources just to help people write the correct program. So that's the first time I realized, okay, human factors is a big thing.
Xiaodi Wu:Before that, I said, okay, mathematically this works in principle. How can we make things run? But then I realized, okay, if you really want to build this in practice, you need to deal with human factors. Not just human factors, sometimes it's just so hard to make things correct, even when you know you can do this in principle. And then we know there's a bigger failure,
Sebastian Hassinger:so like That's larger scary experiment
Xiaodi Wu:coming I in the wasn't really appreciating that. Until the moment I get to the field, I see, okay, people are making that mistake. Actually, you should assume that's what will be happening. That's something will happen. Not assume this won't happen.
Xiaodi Wu:Right. I think that's a big change of how
Sebastian Hassinger:That's interesting.
Xiaodi Wu:The view about the different assumptions from Yeah, yeah. You know, that's still a challenging problem, right? So we have something, think about, basically try to see how classical techniques for really helping people writing correct software to migrate to QUANTON. I think that's maybe still a bit too far from what we currently have. We don't write a super complicated QUANTON software right But you know, research It's coming though.
Xiaodi Wu:Research is a little bit into the future. Yeah, exactly. So we have some proposal for how you do this, using a Tony horse where there's a hard logic for verifying the correctness, but then there's also a very formal method way, writing the software, doing something people call correct by construction. So that's actually related to the automatic theorem prover. I think this just gets hotter in the general audience now because Tarantao says you want to do this for also verifying the proof for mathematical the papers.
Xiaodi Wu:But at the time, people show you can do this to verify your software is
Sebastian Hassinger:correct. Interesting.
Xiaodi Wu:And that's people call it corrected by construction. And we basically find out there are many classical techniques for testing debugging, they don't really extend to quantum. But correct by construction approach, that can extend, although that's expensive. So nothing really practical implication for now, but we do know if you want one day there's certain quantum software you want to have a 100% confident that is And
Sebastian Hassinger:that might be really critical once we get to the size of system that you can't simulate classically at all. Exactly. You have to sort of fire it off into the quantum computer and like, you don't know how to actually track its progress.
Xiaodi Wu:So you know what people do for this in classical domain? They write such a same for GCC compiler because compiler gonna be used a lot. You don't want a compiler that has error. Actually, GCC compiler has a lot of errors. And then people write this for all operating systems.
Xiaodi Wu:Because Loads are the very core software package, you have to have 100% correctness, especially if you're running this, say, on Airbus, or maybe either Euros That's or super impressive, those research they really have in application in real world, in a very critical way. And I think quantum computers should be added to the list. We have the large system, we want to have 100% correct cohort. I think that time, my research can help. That's interesting.
Xiaodi Wu:Course, it's still far from there. Yeah, so
Sebastian Hassinger:So what about So that's sort of the long term. There's stuff that you've been doing that's more about, you know, how to get value out of the the devices that are being built today. Right? Yeah.
Xiaodi Wu:I mean Yeah.
Sebastian Hassinger:So so what are some examples of that that your group is you're getting
Xiaodi Wu:So I think this is just to connect to the how, you know, you what kind of other topic I pick up along So the I talk about the first topic in a programming language. But then if you're doing that, nothing is cooler than if you have your own language. And actually, we see there are so many different languages. What really inspires me, at the very beginning, say, okay, am I really in the position to create something new as a programming language? You're say, okay, maybe you're just a single person, you're not a company, you don't have enough workforce and you're not, say, done tons of we don't have tons of application either, right?
Xiaodi Wu:So what makes you think there some way you can make a new programming language and maybe that programming language will be useful? And then what really changed my mind is that I'm really happy that when I started working on this, and there was a conference in the program language community, they called the history of program language. I think that conference only held every decade. But when I started, there won. And so the purpose of doing that, they just have people writing about how the famous programming language in the history, either popular using industry or not so popular in the industry, but people think that provides you a lot of conceptual They try to have people try to say, what's the motivation?
Xiaodi Wu:Why they view that? Why this And is why does it change some field? And actually that's the time about the COVID, that's really pandemic time. You don't have a lot of other things to do. I don't need to teach for example.
Xiaodi Wu:So I have the luxury to really read through basically almost all the survey paper about a lot of like existing famous programming language for some reason. I think I learned a lot by just reading that because that's really giving me a very different idea about that domain.
Sebastian Hassinger:So
Xiaodi Wu:why people think certain programming language is important in history. And I think that message was later amplified by the Turing Award, I forget which year, to the compiler. What's the second one? Maybe you should help me.
Sebastian Hassinger:Yeah, yeah, The second one. Yes.
Xiaodi Wu:So to give a Turing lecture, they really emphasize the abstraction. It's important and a really powerful tool by computer scientists to help people basically solve all kinds of tasks. Some of the tasks, they don't really like computing tasks, but we can solve them by computers. So really building the right abstraction. Think about how you can not just have abstraction to think, but also all those abstraction, you can do automation.
Xiaodi Wu:You can compute on top of that.
Sebastian Hassinger:So
Xiaodi Wu:that's like, I think very powerful, you know.
Sebastian Hassinger:That's one of the things I find so fascinating about this stage of quantum computing is that classical computing, you know, you look at your phone or your laptop, that's seventy years of abstractions. Right?
Xiaodi Wu:I mean,
Sebastian Hassinger:there's so many layers, as you said, so many pivots. Some of them are technology driven. Some of them are marketing driven or market driven. And with Quantum, we have almost no abstraction at this point.
Xiaodi Wu:Mean I think we do have abstraction. It's it's hard to argue whether that's the correct one. But probably we don't know what's the correct one. Right. But but I think that's fascinating to think about.
Sebastian Hassinger:Yeah.
Xiaodi Wu:So, I think the more I think about the problem, the more I talk to other communities about that, I realise, okay, people do want abstraction, but they have different expectations for the abstractions. So I think from the program language perspective, you basically want to have a way easy for the users to really program your applications.
Sebastian Hassinger:And
Xiaodi Wu:you don't just program in Python or C, actually there are many other programming languages, especially for certain domain. You want to do modelling, simulation, or other types of things for machine learning, you have tons of different abstractions that help you to solve your problems. But of course, for quantum, we don't know what's the QAP yet. So it's hard. But on the other side, you see, at that time, I think it's still possible, lot of cases still the same today, even though I should say there are a lot of progress from other people.
Xiaodi Wu:But if you look at what we have, say, Qiskit or even Bracket or Circ, other sense, you see, this is really like some Python way for writing down our circuit.
Sebastian Hassinger:Yeah, exactly.
Xiaodi Wu:And I call this Python based or maybe twenty first century punch card.
Sebastian Hassinger:Right.
Xiaodi Wu:Right. So definitely it's more automatic. You have very nice UI. So you do this on your laptop, but you're just doing punch cards, essentially. And of course, I'm not saying you shouldn't do that.
Xiaodi Wu:Because we start with punch cards for IBM machine, like in all the time. But what really drives IBM to a success at a time is that people, they realize they cannot just stay at the punch cards.
Sebastian Hassinger:Right.
Xiaodi Wu:So they identify the application, they try to see how to connect, make things easy for the domain users. And that's the time they invented Fortran.
Sebastian Hassinger:Right.
Xiaodi Wu:And that was a huge success in the industry, even though not a lot of people using Fortran today. Yeah. Right. But that gives you some things to think about, right? So, can we invent something like Fortran for Quanta?
Xiaodi Wu:Again, I think that the real challenge for thinking that way for quantum computing is that we don't know the Qt app.
Sebastian Hassinger:Right. But we have some guesses of where the first use cases are gonna emerge. Yeah, mean Right, Kleinman had a really good guess and we're pretty sure he was right. That's some stimulating nature is probably gonna be what quantum computers are good at.
Xiaodi Wu:That's actually exactly the point I was gonna mention about. Right. I I you know, when I say, you know, we don't know the killer app is that well, I'm not saying we don't have proposals. Yes. We just say we don't know those things will really matureize, right?
Xiaodi Wu:So you know for the classical domain, when you build a computer, you know you want to do something. And you know somebody will pay for that. So that's not a big issue for them. But now, like you said, the only thing we know for sure is that you can simulate quantum physics. That wasn't really like computing for many computer scientists, but still, I think that's a valid point.
Xiaodi Wu:So quantum computer can be used to simulate quantum physics. And there are tons of interest from, you know, the science community. Yeah. Let's see how we can make use of that.
Sebastian Hassinger:Yeah. And advanced manufacturing. I mean, you can see Airbus and BMW and LG and Hyundai, like they're, they're actively exploring what they will be able to do as, you know, for improving inputs into their manufacturing process
Xiaodi Wu:or Yeah, these exactly. Other So I don't think there's any like, you know, doubts in the large community that people say quantum simulation wouldn't be an application case. Actually, this will be, I think, the least controversial cases. People will say quantum computing can be good at that. Then that comes to me as to say, this actually is a safe choice.
Xiaodi Wu:So, know something you want to do. Can we build abstraction around that?
Sebastian Hassinger:Right.
Xiaodi Wu:So, that's to me is that I know I want to build abstractions. Right. And there's something I can definitely try. And it's a safe target and it also exercise for me. So, and then that started many things, not just in programming language, basically tried to see, can I help people to think about quantum computing, both the way you use the machine or the way you design applications based on the Hamiltonian abstraction?
Xiaodi Wu:So this is slightly different from the gate abstraction. The gate abstraction, we have that for a long time. Think there are good reasons to think about gates, but there are also I think scientific reasons are super clean mathematic model.
Sebastian Hassinger:There are a
Xiaodi Wu:lot things we can say, especially in complexity theory. You have a way to count the resource gate by gate. So if you don't have a way to measure the resource, there is no way you can build a complexity theory. But the Hamiltonian is something hard to build something major that is agnostic of the hardware. But on the other side, you'll say, Hamiltonian is so natural.
Xiaodi Wu:So can we directly think about Hamiltonian? Because it's close to our target applications. So this is higher level abstraction than gates and potentially also closer to your implementation because when you try to implement your quantum computers, you still run Hamiltonian. You use the dynamic to do something. Of course, a lot of people know how to use that to do gate.
Xiaodi Wu:But also, we know there are certain Hamiltonians who can do them directly. Maybe with better control, you can do them faithfully. So I think that that time, essentially, the Hamiltonian simulation or Hamiltonian itself is a good target.
Sebastian Hassinger:In some ways, it's a Hamiltonian I mean, like the CUERA system, for example, they don't implement gates yet on the Aquila system, and they talk about it as an analog Hamiltonian simulation. So is a Hamiltonian almost more analog in nature than and gates are more digital? Is a useful distinction or
Xiaodi Wu:Yeah, I think that's something I don't think I have the final answer. Like I said, I read into the history. Because for me, I didn't know all these ideas, so I know people have their own opinion about analogue digital, what's the difference. But for me, the first thing this also works when I try to develop abstraction. I want you to look into the history, like I said.
Xiaodi Wu:So feel there are good reasons and bad reasons for using either of them. But then it's really not a debate for which one to use. It's more about which one is more suitable Right? For the technology you So, something I learned from, I still say, limited reading of the history, is that the analogs in, they're cheap, so they came early and they
Sebastian Hassinger:have
Xiaodi Wu:been used for some of the tasks in the history.
Sebastian Hassinger:I think
Xiaodi Wu:some of the Apollo project, you make use of that. And they can be very useful for doing a lot of signal processing as well. But of course, they're not precise. They're more noisy sensitive and it's super hard to control an analog machine for a very long time. So people know what's good and bad about it.
Xiaodi Wu:But also, of course, for digital ones, you can do something super precise. And you basically lower the control task. So because now you just need to implement a single gate. You do that gate well and you can have tons of gates to give you something you want. So basically, I think that's a reduction of the difficulty of controlling the entire thing into the controlling of a single thing, plus you can scale them.
Xiaodi Wu:But once you can scale them, then that's going to be something you want to do. And also, digital sense gives you more generality. So that means I just need to build this one thing. I can do many other things.
Sebastian Hassinger:Like LEGO blocks. You can build anything out
Xiaodi Wu:of Exactly. And of course, because of the moral law, this becomes super cheap now.
Sebastian Hassinger:Right.
Xiaodi Wu:And people just use that. Right. And I'll say, for most of us, people in my age, maybe younger than me, we probably haven't seen anything out of your life. It's really hard to try to convince you why there's still value over there. But actually, if you look into more about the reason, about why something is good or something still has a limitation, like I said, the assumption for digital is that you need to be able to do them a lot of gates, At a very cheap cost.
Xiaodi Wu:And if you cannot do that, then you run into other troubles. I think this is something you really like. Digital is a model in principle it's going to And you know you can build a gate today. Wonderful, right? So you have a clear logical path leading But to the ultimate the difficulty is actually how you scale them.
Sebastian Hassinger:So
Xiaodi Wu:it's not something you see right away. You just push the conceptual difficulty into the future, so we can handle that. Maybe we can. Maybe. Marsalor says there's a lot of things you can do, integrate circuits, and then we are as a design, you're pushing to the size of your transistor in So your that happened, but there are many other reasons to make that happen.
Xiaodi Wu:Are we ready to get that? I don't know. Or how many years we need to get that, I don't know. But on the other side, analog is saying something cheap, maybe we can find some application for that. And maybe it's more suitable before we have our digital ones.
Xiaodi Wu:And as I said, maybe that's something we can find the first useful applications. I don't know. And I would just say, someone should have started that. And regarding for in my case, I try to make some abstraction and basically change how people think about that, then whether you implement that in an analog way or digital way doesn't really to me.
Sebastian Hassinger:The abstraction is still Hamiltonian, the implementation could
Xiaodi Wu:be either. Do either way. I don't care. I think that's the power of abstraction. So what I want to do is that, can I think about the whole quantum computer using abstraction that based on Hamiltonian?
Xiaodi Wu:And of course, there are problems that you don't want to use the Hamiltonian as
Sebastian Hassinger:an abstraction.
Xiaodi Wu:So I'm not claiming this is something you want to do always. There are certain, in a lot of existing, algorithm about the discord problems you don't want to use Hamiltonian to So worry for me, it's more like, what are the good use cases for Hamiltonians? When we think about Hamiltonians, can we think about something new? That is really hard to perceive if you just think about circuits. I think that's the question I try to I'm not saying Hamiltonian is No.
Xiaodi Wu:Can totally see it can solve all the problems. But I really want to see what this kind of abstraction is good for. And can this kind of abstraction lead to something new that is hard to think about in And a certain
Sebastian Hassinger:so you have released this early version of this programming language,
Xiaodi Wu:right?
Sebastian Hassinger:Would you call it a programming language or it's a framework or like, how do you sort of think of the code that you're It's CMUQ, is the Yeah, the
Xiaodi Wu:CMUQ. So, like I said, when you think about abstraction, you have many things you can do. One thing, of course, is building the program language tool. And I also have projects that using this kind of abstraction, we can think about new applications. And we just prove sense there as well with this kind of new angle, which I think it's really hard to think about if you were not using Hamiltonian abstraction.
Xiaodi Wu:Right. But in the circuit, there's no way I can think about that. But just to
Sebastian Hassinger:What are some examples that are use cases that yield themselves to the Hamiltonian abstraction better than the gate abstraction.
Xiaodi Wu:So that's actually has something to do with the problem with the continuous nature. So as you can see, what the gate is good about, this is both true for quantum and classical. We know definitely a lot since about classical, they're about digital design and everything boils down to boolean functions. Actually, that's the beauty of the digital thing. Super precise, digital and very resilient.
Xiaodi Wu:But you know, you just, you have a lot of tools to deal with like a digital sense. Right. But there are a lot of applications that are continuous. Right. For example, machine learning today, right?
Xiaodi Wu:So all the training, you know, Yeah. And they're continuous optimization problem, weights are continuous. Actually, you want things to be differentiable. Yeah. So that's one example of continuous.
Xiaodi Wu:And there are tons of examples, say, in scientific computing.
Sebastian Hassinger:Interesting.
Xiaodi Wu:Like PDE solvers, right? People want to model real problem, like an engineering problem, mechanical problem, or even probably in nature. So a lot of the models, they're continuous. So they're not digital. Of course, if you have really
Sebastian Hassinger:good You can create a digital You
Xiaodi Wu:can create a digital model and mimic the continuous nature. I think it's still super fascinating. I think a lot of people don't really realise that. We have fine appersitions in your computer. So we have limited digits in whatever, like floating number or whatever.
Xiaodi Wu:Yeah. I mean,
Sebastian Hassinger:it's I always think of it in terms of media, like music, for example. We just sort of go to 44 kilohertz and past that, your ear can't really hear. Yeah. So who cares? Right?
Sebastian Hassinger:So but it's still I mean, you're dropping all kinds of information that's there. It's just not relevant to the use case of listing. Exactly.
Xiaodi Wu:So I think this is fascinating. We use floating numbers with digital machine. We can do a lot of things continuous in So the customer I think this is really, really great. But of course, we don't have this kind of synchroquantum. Don't have enough gates.
Xiaodi Wu:So, the question is that if we think about problems that will be more natural, their continuous nature, You and the
Sebastian Hassinger:mentioned signal processing and analog classical Yeah,
Xiaodi Wu:I think there's even something very fundamental, like a gradient descent. So gradient descent, people know that's one magic box behind all the ML stuff. Nobody know why it works. It works. So the gradient descent help you, SGD, or Adam, this is the basic training tool for deep neural networks.
Xiaodi Wu:Know deep neural network definitely not convex function, so there's no theoretic guarantee. You can get somewhere, but it works pretty well. Gradients then help you to identify reasonable models with maybe tons of computing Right. So, I think that's still a magic. Nobody has a very good theoretical explanation.
Xiaodi Wu:But if you look it back, gradient descent is a very fundamental algorithm. It's a fundamental algorithm, I think, both in theory and in practice. And if you think about the circuit model, it's pretty hard to think about what would be the quantum version of that. You think GradedSens is so simple algorithm, classically. So, how can you make a quantum version of that?
Xiaodi Wu:Presumably, you should be having some new quantum feature and hopefully it should be better than the classical gradient descent or maybe also the many other classical algorithms for some tasks. So I think that's just one example why I think switching to think about Hamiltonian really helps. Because if you don't think about a circuit, you don't want to say, I want to design the algorithm by very clever combination of the circuits. Then you say, okay, how can maybe there's a way I can rethink about the aggression in continuous nature. Say, I like a gradient descent.
Xiaodi Wu:What this really is? This is really just a dynamical system. It's a classical, very spatial dynamical It can help you, drive you to say the point where like a gradient zero, right? So people call, not necessarily always even to the local minimum, but going But to it works so well. You can think this is just a classical dynamical system.
Xiaodi Wu:And that's basically the reason we can think about the quantum version when we move into thinking that way. So if we have something that is a classical dynamical system, then you can think, maybe I can design something that is a quantum dynamical system that has some kind of connection with the gradient descent.
Sebastian Hassinger:I see.
Xiaodi Wu:And I think that's the big conceptual step we had.
Sebastian Hassinger:And have you, in SimiQ, have you actually sort of done a quantum gradient descent, the equivalent of it?
Xiaodi Wu:We actually have a software package implement that aggression on top of CMUQ. So CMUQ is more like, I say, it's abstraction for the users. So you want to say, I just want to do some kind of Hamiltonian simulation. Can I write a simple program to do that? So what would be the simplest thing you can do?
Xiaodi Wu:Basically, if you say your domain users, physicists or material scientists or whatever, you write down your Hamiltonian. You say, okay, quantum computer helped me figure out how I can run this Hamiltonian. All I want to know is some property about this Hamiltonian. And I basically give you a way to model what you want
Sebastian Hassinger:to do.
Xiaodi Wu:And I think that's really inspired by the There was a programming language called Simula. And Simula, the Galler Turing Award, by the way, around February. And it's really the first type of modeling language to help people to model probabilistic systems, especially the Keynes system at that time. Help people to write down the probabilistic model and simulate that. Of course, people will compile that to, say, I think Argo 60 at the time or later on maybe to Fortune.
Xiaodi Wu:But it gives the users a super clean abstraction.
Sebastian Hassinger:I see.
Xiaodi Wu:All you need to focus on is your problem. And your problem at that time is say, just tell me the configuration of your probabilistic model. And the program language or the compiler will do the rest for you. So I divide something very similar to that. Basically, from the user perspective, this is the Hamiltonian I want to address.
Sebastian Hassinger:And it's open source, SimiQ?
Xiaodi Wu:It's open sourced. I think we have done the first version compiler. We built the interface. So basically, we put Hamiltonian as the first class object. So, if your task has something to do with Hamiltonian, it's super easy for you to write the program.
Sebastian Hassinger:That's cool.
Xiaodi Wu:And then the compiler will handle how we can actually run that particular And you can actually I think we built a prototype of the compiler. I shouldn't say this is something really great, but we show this is possible. We can compile to a few of the existing machines through their public API. We've done this through Bracket. We can connect to Qiskit if you have the IBM We can also have this through D Wave.
Xiaodi Wu:And we also have connected to Qtip, if you want to simulate that. So we met this connection. Think there has been some follow-up work to improve our compiler. I think definitely there are a lot of things that can do. But I really see the values that we build a different abstraction of its own kind.
Xiaodi Wu:Something really good for the task if you want to simulate a Hamiltonian. That's your major task. Maybe in the future, can build a more user friendly language on top of that. But we showed this direction is possible. And like I said, we actually also added to some new application here.
Xiaodi Wu:Not for the nature science purpose, for something computational.
Sebastian Hassinger:Right. Right. Which is bringing you back to where you started, which is you wanted to find a way to combine physics and computer science. And you have. Yeah.
Sebastian Hassinger:Mean,
Xiaodi Wu:hopefully, I think it's still some our proposal. I don't think it has been widely accepted as something. We think it's I have strong belief in that, but I think I still need to convince the community.
Sebastian Hassinger:The broader community.
Xiaodi Wu:And also, of course, by doing that, also just to try to see whether I can convince myself or I see something wrong with that, and of course I can But so far I see no competing evidence. But just going back to that, so we have a program language that help people to run their Hamiltonian based applications. I think the major contribution for that project is we show this kind of abstraction can work. In that project, we basically show you can reproduce nature papers, all the figures, with just like maybe 200 lines of code. Actually, the language is not designed for the Nature paper.
Xiaodi Wu:So we just say, this is a good abstraction that can save you a lot of effort. Also, you basically really we try to sit in the shoes of those domain users for this. So if you really think about how much an installation is your task, you should find this language pretty easy to use.
Sebastian Hassinger:That's great.
Xiaodi Wu:And of course, how we can make the compiler work, that's taking a good challenge. Or maybe there's going to be some balance between how you can decorate more information in the language, in the path to the compiler because compiler is not super intelligent. But I think this shows that direction is possible.
Sebastian Hassinger:And
Xiaodi Wu:I can totally see if we identify our killer app for quantum computing in the future where simultaneous simulation will be a major part of it, you can definitely borrow ideas I'm on this not saying you should use it. Actually, a lot of cases in the development of programming languages that you prove something, the concept is
Sebastian Hassinger:And those ideas get incorporated
Xiaodi Wu:into Yeah, yeah, yeah. The language itself, they might die, but the idea will exist for a long time.
Sebastian Hassinger:That's fascinating, Shadia. That's great. I've been very curious about the work you're doing in semi q. So I'm glad we did sort of a deep dive into And hopefully, you know, if some people hear the podcast, then you'll get some inbound interest. You'll get some more adopt these and kicking the tires, so to speak.
Xiaodi Wu:I just like, this gives you more conceptual ideas. Think as an academic group, usually it's not the case that we can maintain a large really user community in the program language. We're just shorthanded to do that. Yeah, of course. But I hope the future developers from either industry or maybe academia, they find this concept useful.
Xiaodi Wu:They can grow their own program language, but they can use our, you know, some of our concepts. Think that that'll be super heavy. Yeah. Well,
Sebastian Hassinger:thank you very much for your time and for telling us about CMQ and your journey. So I really enjoyed it. Thank you.
Xiaodi Wu:Okay. Thank you.
Sebastian Hassinger:Thank you for listening to this episode of the New Quantum Era hosted by me, Sebastian Hassinger, with theme music by OCH. You can find all past episodes at www.newquantumera.com or connect with the show on blue sky at newquantumera.com. Special thanks to the organizers of IEEE Quantum Week for their support of this episode and to Xiaodi Wu for sharing his insights and experiences. If you're enjoying the podcast, please subscribe and let your quantum curious friends know about us. Thanks again for joining me.
