Corey O’Meara (E.ON): Finding Real Quantum Advantage in Grid Optimization

Yuval Boger interviews Corey O’Meara, Chief Quantum Scientist at E.ON Digital Technology. They discuss how E.ON’s five-person quantum team works with business units to find high-value use cases, especially where classical methods may hit bottlenecks in optimization and machine learning. Corey describes the company’s work on peer-to-peer energy trading and other grid-related applications, and shares his perspective on cloud access, hybrid deployment, and the accelerating progress toward fault-tolerant quantum computing.

Transcript

Corey: Hello, Corey, and thank you for joining me today. Thanks for having me.

Yuval: So who are you and what do you do?

Corey: Yeah, so my name is Corey and I’m chief quantum scientist at E.on Digital Technology. And there I’m technically leading the applied quantum computing program at one of Europe’s largest utility companies.

Yuval: E.on, as you mentioned, is a very large company. How large is the Quantum team and when did it get started?

Corey: Yeah, great question. So we’re a pretty modest team. We’re five full-time employees on the project and the team. And we started in around 2020. It’s kind of an interesting story. So in the early days, we were approached– so I should preface it that the department where it lands is in– it used to be called the Data and AI Department, or DataOn. And we basically were looking at how to apply machine learning and optimization for different business use cases across the company. And at the time, 2019, 2020, quantum started to get a little bit of a buzz in the industry sector. And so it kind of came actually from the board level. So it came from top down. They asked, hey, you guys in the department down here, you know, working on internal tooling, what is this quantum computing thing? And I had originally done my PhD in quantum computing actually. So we started discussing and came up with a short little kind of proof of concept about a business use case. And that sort of started the momentum and the internal traction. And over the years, yeah, we’ve grown the team. We’ve gotten more external funding, more internal funding, more stakeholder buy-in and yeah, have been working on finding where can quantum advantage be achieved for real world business use case scenarios.

Yuval: – I would imagine that if the funding, the initial start came five years ago, then at some point the board says, okay, how’s this going? And when are we gonna get results beyond academic papers from it? What are you telling them?

Corey: – Yes, so we have a very good relationship with the board. So I’m very privileged. the quantum activities are completely supported by them at the highest level. And we basically, everyone’s aware. So one of the main goals that I’ve personally had is to bring the hype down internally and seeing what is possible on today’s or yesterday’s quantum computers, but also today’s quantum computers and try to manage expectations essentially right and talk about where we are in the quantum advantage journey and the board really believes in innovation for the right targeted business use cases so they’re aware that fault tolerant is some years away I know QA is you know actively working on this on this mission and that’s fantastic so they know there’s no ROI in the immediate future. We will wait some more years till fault tolerant is here and start to see these applications roll out for different business use cases.

Yuval: If a colleague from another energy provider, say a US energy provider, approaches you and says, “Cory, we’ve been doing this for five years. When should we get started? What should we expect? And how soon do you think we can get something that’s truly valuable to the

Corey: business out of it, what would you say? I would say there may be even a little bit too late honestly. It’s been really challenging but rewarding building our internal capabilities, onboarding our organization. I mean you know it was hard enough a couple years ago to talk about machine learning or AI. AI is now the for that right and this sort of blew up everything and now every organization around the world knows the value of AI or machine learning right and quantum is such a foreign concept to business units to stakeholders it really is its own special thing that is like literally hard to grasp for some of the world’s most famous physicists for a hundred years right so to communicate what are the kinds of problems in the different business units where you can potentially use a couple algorithms that may have a provable advantage you really have to educate both like the management right so the ones that are allowing to work on certain projects but you have to get to the low level to the nitty-gritty of the problems and often work with the engineers who really know the domain right because where you find out where you can maybe apply quantum where the bottlenecks are in those different applications so this is not something that happens overnight and even if there is a fault tolerant quantum computer you know in a couple years from now and we can we can run real algorithms on this that give provable quantum advantages you know it’s not going to be an And I think to figure out where to even use it in your organization. And yeah, so I would say that it’s good to, I mean, I’m proud that we’re some of the pioneers in this space to sort of look at these early stage quantum applications. But you know, I think the first move advantage is going to be, you know, very important due to the complexity of applying this kind of technology.

Yuval: Is the team primarily quantum physicists and software engineers or is your five-person team have a lot of domain expertise that comes from other disciplines?

Corey: We did the first way, so the first order. So we’re all PhDs in physics or quantum computing and yeah, so ex-professors, postdocs, etc. And then when they join E.ON, we learn about the energy sector.

Yuval: When you started identifying use cases, obviously having directives or support from the board is super important to get other people engaged. But how often do you review these use cases? Or is it the case that department heads just come to you and say, “Oh, I know you’re doing quantum and we’ve got this interesting problem that maybe you could help us with?”

Corey: Yeah, that’s a great question so It’s a little bit of both. I would say so at first we were on our own we didn’t have really a broad enough internal network at Eon so I mean Yeah, you know operates in 15 countries across Europe right and It’s it’s difficult to sort of I mean at the beginning of the journey five years ago I mean quantum algorithm development is changing all the time You know and especially things how they were five years ago is different how they are now besides a handful of like classic quantum algorithms right Grover Shore HHL whatever So it started off with us trying to look at what problems were people in our department building internal tooling for that was a bit that would have a bottleneck let’s say and So that came from internal knowledge of the tools. We were building classically because we could see like oh Classically this works fine, but actually at scale we’re going to hit runtime limits So that was kind of the first few projects and over time we’ve educated the company enough that we have training workshops We have internal Yeah, and internal trainings and lecture series and things like this where you know the quantum hype and people get interested and Then they come to us sometimes, so we’ll get random emails. Yeah. Yeah, I’m working on this problem, and you know Do you think this would be good use for quantum computing and then you know we? Discuss with the business unit and we try to look to see okay when you go into the nitty-gritty details of it Is this something that is for either near-term? potential or something that’s more requiring a fault tolerant quantum computer.

Yuval: So you basically try to understand whether this is a problem worth solving in terms of are you reaching classical limits? And then can it be solved in a reasonable amount of time calendar wise as opposed to waiting for a large scale fault tolerant? Is that about right?

Corey: Yeah, exactly. we I mean for any use case especially let’s talk about optimization problems so especially quantum optimization we now look at okay I mean you know just because a problem is can be phrased in a common rhetorical way such that it’s you know an NP hard problem it doesn’t mean that that’s actually hard to run in production for a business, you know, it could be that, I mean, essentially it boils down to the actual instance of the problem. And so we only really tackle combinatorical optimization problems if we can show or there’s known literature which shows that these problems are hard enough and they scale with exponential runtime using let’s say Gurobi or you know maybe some heuristic or something and then I believe that that’s a prerequisite for needing quantum optimization because otherwise yeah you’re running on a quantum computer but for what reason right so even though there’s no of course you know guaranteed asymptotic speedup using something like QAOA or some other quantum optimization algorithm or quantum annealing, but you can, you know, at least work on something that’s already hard. And so that that’s usually the beginning of our projects when we get approached or we start digging into a new problem. We look at, okay, what’s your data look like? How big is your optimization problem instances? Like something if it requires 10,000 variables and then it gets hard, or a million variables and then it gets hard, well that’s not really a fit for any quantum computer in the near future at all. So we particularly try to look at challenging problems that are hard enough at a small scale where we try to push the limits of today’s machines.

Yuval: So in a sense it’s not about big data but about complex data.

Corey: Yes, exactly. Complex data in terms of optimization and also quantum machine learning. So looking at, again, there’s some kinds of difficult, not necessarily from a complexity theoretic perspective per se, but if your current machine learning algorithms struggle, like if current state of the art algorithms struggle, let’s say, then that could be a good use case where you can maybe try to augment it with a quantum computer. And there’s some new research out there that suggests that combining quantum and classical in this way for hybrid or however you want to call it, machine learning, that there’s some provable types of quantum advantages.

Yuval: If you found a problem or if you found a solution to a problem that is sufficiently interesting for the business, would you want to own a quantum computer or would you use one on the cloud?

Corey: I mean, I would want to own a quantum computer, yes. [laughs] I don’t know, I’m sure QAIR or IBM also would like EON to own a quantum computer. No. It’s a good question. I mean, the thing is, with the cloud access, the way you access it these days, I think probably fine for the foreseeable future in the sense of decades. This is just my guess, right, but these are large very specialized machines and you know if there’s a whole bunch of them accessible via the cloud I mean it’ll be interesting once quantum advantage is really proven even from a scientific non real-world application that will be interesting. Then proving it for practical quantum advantage, right? I often think like what will happen to industries where they haven’t been looking into this yet And I hope that it’s going to be a mad rush to get access to these machines once It’s really you know once there’s no shadow of a doubt You know that these things are achievable on devices that are actually built today or in the near future It’s a good question how much queuing can you have and how many physical computers can you build? but one thing that’s interesting about our use cases is Oftentimes things have to be computed in Essentially near real time in the energy sector so we’re always trying to balance the Supply and demand of electricity in the grid and so there’s energy markets right that you buy and you trade and you sell electricity which is a traded commodity and and we’re also live generating electricity and transmitting it and operating the electrical grid. And so what I find really cool about that is it’s a guaranteed time window. You know, some things, okay, you can talk about logistics and maybe truck routing or I don’t know, yeah, you can have something run, but maybe do that schedule once a week, I don’t know, once a day maybe. But in the energy sector because it’s such a constantly changing dynamic thing it We have our domain itself has a dedicated runtime window so any kind of potential speed up we’re fixed in inside this window and so going back to the question about Having our own quantum computer probably one day that would be great where we just plug in and we run so it kind of depends on how things are deployed in operations in the future and how what kinds of critical applications you know are running on these computers and when there’s SLA agreements right for getting access to to cloud-based quantum computers I mean we’re not there yet of course but if real critical things run on there then there’s definitely a possibility in the future. In the near

Yuval: term I could probably help you get a quantum computer from Lego. It’s much easier to maintain. It doesn’t take a lot of space, very energy efficient, even though for you guys energy is not the problem. I’m wondering if you think about quantum computing as a business service under these SLAs, will you always have a classical backup? You know, if the computer is not available or something happened, do you have a sort of a backup algorithm that you can always sort of run classically?

Corey: I mean we have some papers out about hybrid software architecture, so integrating enterprise, you know, Docker containers and sort of cloud deployments into the Azure cloud and things like this. So we have some work in there. Some of our government grants that we are working on with our collaborators from universities and research centers, there’s usually a piece to our work where we already are trying to think a few steps ahead about how would we integrate that into our existing cloud infrastructure. So I think this is a definite topic, but I also think that, going back to the hard problem instance example I was talking about, I mean, I would imagine if there were some kind of test or how you say, like a try-catch statement almost, right? You’re writing code and then you try it and okay, there’s a timeout limit because maybe that particular instance of the problem is extremely hard and your normal heuristic won’t run, right? So you run and you solve it on the quantum computer. You could go that direction or you can just run everything on the quantum computer. you try it running there that little subroutine right before you post process it classically but having a fail-safe you know that definitely would be important I would imagine in the future at least for the sector I’m working in right so there’s I’m sure other sectors maybe where you can’t do this like molecular simulation I’m not sure if you want to take a big hit to your accuracy let’s say if you go back to kind of classical models but in our scenario I think the big question will be if you go to a classical heuristic and a fallback if something is wrong with your quantum computer connection how much are you losing overall in whatever your use cases right because you I would expect you’ll be losing accuracy in your in your answer as it’s just an approximation, let’s say.

Yuval: I saw that you published several works on energy pricing and coalition structure generation and vehicle to grid scheduling. Could you talk about one or two that you’re particularly proud of?

Corey: Sure, yeah. So in terms of the coalition formation, I think this is a nice one. So one of the things we’re looking at is how can we enable a new business model for peer-to-peer energy trading. So if you have solar panels on your roof or you have maybe an electric car that’s plugged in at home, you might have a little power wall or an extra battery. know we’re moving more towards this concept of decentralized energy generation and sharing so the idea is if you have solar panels in your roof and maybe a couple people down down the road do in your neighborhood and then your other neighbor needs electricity and he has a smart meter it doesn’t really make a lot of sense to keep things the traditional way where you use you know a steam turbine or a gas-powered turbine spinning hundreds of kilometers away to transmit that electricity to power your neighbor, when you have that, you have excess. And so this whole concept of how do you optimally share locally stored and locally generated electricity such that you get a perfect energy balance locally. wins here in this scenario and it turns out this is a really, this is an extremely hard problem so finding out the best way to group prosumers as we call it, so, or peers, the best way to group them so that they share and locally balance their electricity with to excess that’s produced. This scales– this is exponential scaling if you try to run this on classical computers. There’s heuristics for it, which also scale poorly. And what we did was we were working with a startup from Munich called Aquarius GmbH and the University of Munich, LMU, and the University of Oxford, where we looked at, okay, how can we formulate this problem such that it can run on a quantum computer in a sort of hybrid algorithmic approach? So we knew some collaborators from, also in Germany, from DFKI, and they developed a nice algorithm for essentially sequentially applying MaxCut algorithm. So it’s a hybrid workflow. So we took it, tuned it a little bit and really formulated real world graphs and data. And in the end, to our surprise, honestly, we actually proved a kind of quantum or let’s prove this is a strong word. Let’s say we demonstrated evidence of quantum scaling advantage for running this on a D-Wave quantum computer, so using quantum annealing. Now we weren’t the first to demonstrate quantum scaling advantage, right? So the folks at J.P. Morgan had a very nice paper in science advances on this for those so-called labs problem and also USC had a PRL on this paper on this topic but it wasn’t for our use case. It was essentially for mathematically interesting optimization problems, but not really founded in the real world. And yeah, our paper was just recently published maybe two or three months ago in IOP QST journal. So finished the peer review process and amended the draft over some time. And so we’re quite proud of that result it’s you know if I were to sell it a little bit here I would say it’s you know the first real-world business relevant application and proof or evidence of quantum scaling advantage. How do you go about selecting the

Yuval: quantum computer that you use to run the algorithms? Yeah so we are I mean there’s

Corey: so many options out there right so I mean this is a technological race that you’re you’re a part of and you know when we started five years ago at the time I think IBM had 27 qubit chips and D-Wave had their D-Wave advantage 5000 qubit chip I believe and we essentially signed up with both providers because okay quantum annealing at the time and still is it’s it’s excellent for running quantum optimization so we were using that mainly for quantum optimization projects honestly for the first few years and then we’re also interested in other things that aren’t optimization right so you had mentioned earlier energy contract pricing and sort of portfolio related topics. We’re also working on quantum machine learning and not that you can’t run that on quantum annealing, but generally that’s better suited for gate-based quantum computers. Yeah, we also have partnered with IBM and they’ve helped us upskill our team and learn how to use these things in the real world over the last couple years and we’re very happy with with both of our

Yuval: hardware providers. How difficult do you anticipate it would be to move this code to another quantum computer if something came out that was exciting enough performance-wise or a number of qubits or error rate that that would be

Corey: tempting to do so? I think it’s definitely possible I mean it depends on the exact architectures and the basis gate sets of course, right? But I mean the algorithms we develop for universal gate based computers, I mean it’s essentially down to sort of compilation level details I would imagine, right? So it’s definitely something that we think about, that we want to keep things relatively agnostic for the future but yeah it’s definitely an interesting race to be part of somehow I suppose and we’re really looking forward to seeing how it evolves in the next five years will be very interesting and then of course the next decade but it’s a great time to see what’s what’s happening every month there’s some new new new some new news breakthrough by you quantum computing

Yuval: hardware company guys. We’re getting close to the end of our conversations today unfortunately but I wanted to ask you a couple of questions about time and price. When you’re thinking about these algorithms, you know the vehicle-to-grid or the other one, the coalition structure generation that you mentioned, how long does it take to run? Is it a couple seconds? Is it a

Corey: few hours? Let me think off the top of my head it was I mean it’s definitely not hours so yeah some like a few minutes I think so particularly for optimization problems right so quantum machine learning is a different thing especially if you’re trying to train on a quantum computer and you’re and we were rerunning things many times. But now for optimization, it was a matter of minutes.

Yuval: And thus, given the potential benefits of optimizing this, I mean, I’m sure you, as a company, there’s a lot of money that changes hands and there’s grids and so on. So in a sense, it doesn’t really matter how much a quantum computer costs per hour. if it’s $5,000 or $20,000, if the problem is solved and the payoff is much larger, I would assume. Is that correct?

Corey: – Yeah, that’s something that we look into when we’re doing use case development. So we look at, is there a classical bottleneck here? So is there a bottleneck either today or tomorrow once the grid is more digitalized, we have more pipelines in the cloud, we have more information to process. So we look at that. And then the other topic we look at is, yeah, if you can speed it up, what would the financial gain be? Because, you know, just because you can, I mean, that’s the difference between proving a scientific result and talking about business, right? So if we had some use case and we proved undeniably quantum advantage, as a scientist, I would be thrilled and I think the whole community would be thrilled that a real business proved quantum advantage and we get it into, you know, cover page of nature, right? But yeah, at the end of the day, if the business says to me, “Yeah, well, okay, that means nothing for us. I mean, so, you know, that’s not going to have any monetary value at the end of the day, or that doesn’t help the grid stability, that doesn’t make anything more safe for us, that doesn’t make the grid more resilient, that doesn’t, you know.” Then at the end of the day, that’s just a cool science result and doesn’t really have an effect for the business at the end of the day. So, the use cases we work on, we try to aim for things that having maybe a little bit of a speed up might have a large cost function sort of value at the end of the day.

Yuval: Next to last question. You’ve been doing this for a while. What’s new in the quantum world in the last 12 months? What have you learned or that has surprised you in the way quantum is developing?

Corey: Yeah, that’s a loaded question. I mean, there’s news every week these days. I think that, I mean, besides the incredible progress in hardware, so it’s clear I’m not a quantum hardware guy. My PhD originally was in math of quantum computing and now it’s in work in quantum applications. So I won’t say too much about the hardware besides the fact that I find the roadmaps default tolerance amazing. You know, 20 years ago when I started learning about quantum computing, they were just a dream. There was a couple qubit machines, you know, five qubit, maybe seven qubit NMR machines. And to see where we’re at today, not only am I using daily, you know, 150, 156 qubits, there are roadmaps out there to show that you can really have error corrected logical qubits at scale, or you can run, you know, hundreds of millions of gates. I mean, this is really science fiction stuff. And it was science fiction even five years ago. When you say it, that’s the big joke, right? It was always, ah, it’s five to 10 years away. But now companies really put their name on the line that it really is within five years, you know? So from a hardware perspective, that’s that. And from a algorithmic perspective, I really like these new efforts where people are looking at, you know, so for example, the new decoded quantum interferometry result from the Google team. So these novel algorithm developments are absolutely critical for the whole field. Because if you think about the handful of quantum algorithms that have provable asymptotic advantages, right, that people have been playing for years with trying to find it, “Ah, can you apply HHL to this?” Or, “Ah, can you apply Grover or some variant of it to this?” I mean, there’s only a handful of them. And, you know, once these machines are built, they should be doing real things that actually have quantum advantage, right? And the problem is, is that they’re so darn hard to come up with. (laughs) the algorithms and their inventors or discoverers, whatever you want to say, that’s why they’re all quite famous within the community, because it just requires such a deep knowledge of, yeah, computer science, mathematics, complexity theory, etc, etc. And so I’m really thrilled with the new algorithm developments such as that. And last, a hypothetical, if you could have

Yuval: with one of the quantum greats, dead or alive, who would that be?

Corey: – The quantum greats, oh. Yeah, well, people probably listen, I’m sure they listen to your podcast, so I have to be careful what I say, right? So, (laughs) I would love to have dinner with a previous guest of yours. I think, I don’t know when this will air, but a recent guest you’ve had, but Scott Aronson. I think that would be, I think he’s brilliant. And the interview I heard was also fantastic. And I think the skepticism of people like him in the field, I think is very important, especially to balance the hype with the academic and the scientific. And I try to sit in the middle there of those two sides of the spectrum. I recognize the skepticism, but I also recognize the need, of course, from business perspective, that things have to kind of continue rolling there. So I think that would be my answer.

Yuval: And Scott also always mentions Gil Kalai as one of the quantum skeptics. So wait for the Gil Kalai episode. I recorded it just yesterday.

Corey: Oh, then I’m looking forward to it. Corey, thank you so much for joining me today. Yeah, you’re welcome. Thank you very much for having me. It’s been a pleasure.

—-

Yuval Boger is the Chief Commercial Officer of QuEra Computing.