How quantum computers model molecules

Welcome to the Coherence Times, where we bring coherence to the entangled world of quantum computing. I'm your host, Ryan Mandelbound. Every other week, I will bring you stories about how scientists, developers, and businesses are making quantum computing a reality. We'll explore the latest research and development, highlight the latest advances in the field, and see how users are trying to extend quantum to realworld use cases. >> [music] >> Last year, we told you that simulating chemistry would be one of the first valuable use cases for quantum. There were already a few promising simulations back then. For example, Cleveland Clinic and IBM had used quantum to study molecules with up to 36 atoms, and we're about to publish another paper simulating a 303 atom mini protein. Now, it's mindboggling to think about how far things have come since then. Just a few months ago, IBM and Cleveland Clinic teamed up with Reken in Japan to publish a simulation of a protein with over 12,000 atoms. So, [music] I wanted to bring on some folks who could talk a bit about the progress on this work and where things are going to go next. First, returning from our very first episode is Dr. Jamie Garcia, director of strategic growth and quantum partnerships at IBM, who's also an expert when it comes to doing chemistry with quantum. And joining us for the first time is [music] Dr. Kenneth Murs, staff scientist from the Cleveland Clinic Center for Computational Life Sciences, who leads the team responsible for these exciting new demonstrations. [music] So Jamie, just as a quick recap, can you give me a sense of what makes quantum computing good at simulating chemistry? Yeah. So quantum computing holds a lot of promise for simulating chemistry problems. Um mainly like the premise for this is based off of the idea that quantum computers operate fundamentally under you know the principles of quantum mechanics which is basically the OS system for the universe if you will. And so because of that uh chemistry is really well modeled by uh quantum computers and it holds a lot of promise for the future and doing things um beyond what we can do classically today with approximate methods. >> What makes chemistry problems so hard for classical computers? >> Yeah. So classical computers uh are basically fundamentally use one and zeros to represent systems. This doesn't work particularly well when you're trying to model molecular systems. Uh, molecular systems are incredibly complicated and you can kind of think about it of like they're built up of these subatomic particles including electrons and electrons and how electrons interact with each other dictates a lot of what you know materials and chemistry properties um and reactivity and everything else that we that we know about how chemistry actually works is governed. So what we really need is to be able to have some sort of computational platform that can manage those kinds of interactions as they scale. Um so essentially if you had if you can imagine having to model every single electron with electron B and electron C and electron D and then doing that over and over again that just gets really computationally expensive on a classical computer. So the idea and the promise is when you're using a quantum computer that it now it's fundamentally behaving under the principles of quantum mechanics which is also how you know molecules behave. It's the same rules that govern molecular behavior as well. And so the promise is that you can actually be able to use algorithms that then really much more accurately and comprehensively describe the molecular systems that we're interested in and then we can understand the behavior better. So maybe I can take us even further step back and ask you Kenny, why would we want to simulate chemistry or simulate molecules in the first place? >> Yeah, I mean you know molecules this particular uh drug discovery type things. The constitution of a molecule really governs its behavior um quite intimately actually. So the classic example people use is aolyomide right. So one form of theolyomide is a mutagen traden right and causes lots of you know birth effects and things of that sort. The other one is works very effectively for morning sickness. So it's just it's just one change in the chyal center in the molecule. So the the different interactions that a molecule presents say to a receptor are critically important for its function. And so that's why you really want to understand these at the fundamental level. You know, I guess what I'm hearing is right like you can just twist a molecule around, change its shape and then form kind of function follows form in a way. >> Yeah. If you think about like a receptor site and you have say a lead molecule, right? You can look at that receptor site and hypothesize about where you may want to add to that molecule. The addition could be something as simple as a methyl group or a benzene group or a carbonal. Um and the the goal there would be to optimize its interaction with the receptor to make it more efficacious as a potential drug candidate. Right? You would get a better binding affinity um for the receptor for a given drug molecule. So this is the classic thing that's done you know week after week around the world in pharmaceutical and biotech companies where they're trying to optimize the characteristics of the molecule given a particular receptor. How are they simulating these molecules today and kind of you know doing this workflow? >> Yeah. Uh so the computational chemistry has evolved over the years. Uh you know initially was microraphics or just simple energy minimizations using a force field or very simple um representation of a molecule as balls and springs. Um then it evolved into more sophisticated docking studies where you would take a receptor and you would just place um potentially millions of molecules into the pocket to find the one that had the best complement uh with the you know the receptor. And this is done again with kind of force field based methods. You know uh this is still a widely used approach but it's kind of a 50-50 approach. um it's usually like for hypothesis or idea generation rather than quantitation. Uh more recently people started using molecamics uh so-called free energy calculations where you can actually simulate the behavior of these molecules and get the relative free energy differences between you know adding methyl group or adding benzel group. Um some of these now are done with quantum mechanics and we actually have reported on archive actually doing these kinds of calculations using quantum computing uh you know where we calculate the relative free energy of binding of of different uh molecules to the same receptor. So these are sort of um this kind of where things are now I would say u what we're doing I believe is more pioneering work. uh it may be something that's going to be you know more routinely used in the coming years as you know the technology gets better. >> Let's talk about that. Yeah, let's start talking about that work. So maybe Jamie, can you talk to me about how this collaboration started like uh you know who reached out to who? How the first conversations go? I think early days uh it had to do with leadership uh having a vision for a collaboration and and really I give uh Cleveland Clinic a ton of credit on this for understanding and realizing the potential that you know quantum computing could potentially have on their own research. Um so I think it started very early. I mean Cleveland Clinic was one of our very first. So, uh, you know, partnerships, large partnerships and, uh, by the way, thank you for, uh, being one of the firsts on that and we we went through a lot of, uh, different trials and tribulations as, as we went on, but um, certainly I think it started at a high level and then, you know, uh, folks started getting really interested in using the quantum computer and that was where I would say a lot of the magic happened in the partnership because once people started actually using it and really working together and working through problems and identifying like what they thought could potentially bring an advantage for using a quantum computer for relevant problems for the Cleveland Clinic. Um that was really what needed to happen to like really start catalyzing and then working with Kenny and and your group has been amazing. Um the progress is just incredible. So, you know, you talk about doing some of these smaller simulations and then in 18 months, you know, you go from like 10 atoms all the way up to like almost 13,000. It's uh incredibly rapid growth and uh evolving research. So, I think uh that's that's sort of the vision. I think we're continuing down that path and and trying to look at problems that are really interesting and relevant to Cleveland Clinic and also uh uses the quantum computer in the in the right way. I mean, and then Kenny, how did it feel to kind of go to your leadership and your team and say, "All right, we're going to get into quantum computing." >> You know, I think I realized early on the importance of having access to cutting edge hardware, right? If you really want to kind of explore new things, and this goes back to like say vector computation to parallel computing, getting a hold of some of the earliest parallel devices, um, GPUs, right? And um you know Cleveland Clinic you know again like Jamie said you got to give them a lot of credit for you know jumping into this and being one of the the first you know installs of the on-prem device you know in the United States. Um but that was very very attractive to me right and I I had always kind of been following the quantum computing area. I have been interested in fragmentbased methods which is what this uh calculation is based on uh for some years and I I just thought this is a perfect marriage potentially right and so I joined uh Cleveland Clinic and then you know IBM's been amazing in terms of training my team right because we started pretty much at ground zero you know how to use the device um so the training is awesome that you know um IBM's offered and then just the scientists right and also the sharing culture. uh we were using something called a variational quantum uh solver uh initially to try to do some of these calculations and you know we weren't making much progress let's let's be honest right and then uh IBM introduced us to SQD and it became very clear that the subspace quantum diagonalization approach was going to be the way to go right and they didn't have to share that with us right um they shared it a few weeks early before it came out on archive you know giving us a boost right and So immediately this whole journey started from you know we had the idea we had the molecules then we finally had the technology to go after it and then uh were able to very quickly you know study some you know smaller systems methane dimer water dimer u but it became pretty clear that we could go much larger and then um we discovered a method called embedding wave function method it's uh developed by George Booth out of the UK and this allowed us to go at the atom level and this allowed us to be right in the sweet spot for SQD, right? In terms of the number of cubits like 40 to 80 cubits and uh we're able to put it together to go to this large scale uh calculation. So that's kind of really the journey I would say and it's you know just everyone being super supportive right and the IBM quantum network connecting us with say with rican and um you know I think the rest is just everything fell into place and we're off to the races. Maybe we can go and tell a bit more of that story. Like Jamie, what made it like so VQE wasn't working? And then what made this SQD algorithm that Kenny's talking about kind of a more valuable approach to tackling these kinds of problems? >> Yeah, I mean, I think they're they're different algorithms, right? And I think um you know, we we've been using VQE for quite some time um since it was it was really popularized in around 2014. Um and and so people started seeing it as being one of the first uh ways that you could actually start studying chemistry using a quantum computer because it was efficient. Um which is great. Um but at the same time the method is is set up in such a way that you know the impacts of noise is actually pretty substantial. And so while you're still, you know, um, in a in a world where we don't have error correction yet, um, you're always going to have an impact by that. So the problem was really like how do you scale and how do you get to places that, you know, to larger molecules that are more relevant um, you know, it was really great that we were able to simulate lithium hydide and hydrogen and burillium hydide, you know, the 2018 nature paper. In fact, it was rather inspiring, I would have to say, because at that point in time, we weren't sure uh what quantum computers were going to be able to do. And so, to actually see a tangible example of like here's a ground state energy calculation of small molecules, but still relevant molecules and real molecules, I think, was an inspiring moment. So, it's just about, you know, it's about algorithm development. It's about understanding the systems that you're looking at and putting the pieces together in the right way. So, you know, Kenny and the team, you know, very cleverly put together, you know, EWF alongside SQD. And really, it was understanding how these things fit together that I think was what enabled uh them to be able to produce the results they did. The other thing I want to point out too is that it's it takes a village in a lot of ways. So you really need like people who are coming together with those different backgrounds whether it be from the HPC world um whether it's from the you know computational chemistry world quantum information science and getting again like coming together and really thinking through a problem um in detail is really what can advance us to that next phase. So I think it's a combination of having people that have an intuition for the systems they're studying and combined with those who have the experience with the algorithm development and understanding how to map things on a quantum hardware and then working together with the HPC and the classical resources as well to stitch together a complete solution to a cobble. >> You know BQE really was a pioneering method as Jamie is showing and or had had described. Um for us we were trying to go to like the bigger molecules and it was the errors and things like that were really consuming us right. Um so I think we just ran into the kind of wall if you will in terms of the range of applicability for for VQE and SQD opened up the the next branch. So I agree with Jamie like some of these early papers were were very inspiring right and actually did inspire me in the sense that you know I looked at these and I said okay you know you can do small molecules so if we could do something at an atom level you know we could go probably quite bigger right and much larger and so I think that indeed bore itself out from this this early pioneering work. Yes. >> So maybe Kenny can you walk us through then the sort of workflow of how this simulation actually works? Yeah, sure. So the, you know, it was known since probably like the early 90s that you could take a larger quantum chemistry problem and break it up into fragments and then there was different ways in which you could stitch it back together to get answers that were very comparable to doing the full calculation. You know, prior to this time period, you know, the thought of breaking a molecule apart in a quantum calculation was kind of anthma, right? It wasn't something you would think about doing. Um but then pioneering work by say way tao yang and others in the early 90s you know kind of broke that barrier and allowed you to start thinking about it. So the way you look at and it's actually proteins are interesting because they're insulators right so they're not you know particularly conducting a lot of the chemistry is very local and so you could take a protein molecule and you could go well we early work went residue by residue amino acid residue uh in this particular work we went atom by atom uh and using the CWF method we could take like a glycine amino acid which consists of an NH so we could treat the nitrogen by itself with an environment. The hydrogen by itself with an environment, the C alpha with um just a carbon and its environment, the hydrogen's attached to the C alpha and then the carbonal. We could treat the the carbon of the carbonal and the O all independently, right? And then we can do the calculation and then there's a procedure by which you can stitch the individual energies together to estimate the the energy of the global system which matches the uh full calculation uh quite closely. So this is really the power of the method like uh instead of trying to solve a global problem right which would be you know I don't know tens of thousands of orbitals electrons you know we can break it down into smaller components and then stitch it together to get an estimate uh of the total energy of the system. So that's the power of the approach I think. >> Got it. So I mean this is kind of the beauty of it right is and this this is what you've been doing sort of since the beginning right is taking these large molecules splitting them up into individual fragments and then basically by characterizing each of these fragments you can kind of stitch it together to create a good approximation of how the entire thing should act. I mean that's in a nutshell. I mean we can dive into like some of the real technical things but I don't think you know I think you know if you understand that basic concept you you're 90% of the way there and then you know we started with a 300 atom this TRP cage molecule this mini protein and then we sort of the next step was coming out with this paper we simulated two proteins that are 11,000 12,000 atoms um which sounds kind of mindboggling to me that just in 6 months we can kind of scale up like that um so maybe Kenny you can walk me through how that's even possible. Well, I mean, so we had done some simple molecules like test cases like peptides and things like that and you know went really fast. We were really surprised, you know, uh is quite efficient using just sort of um IBM Cleveland, you know, one quantum device and our local compute resources. Uh and so we were able to quickly scale up to the 300 atoms. Uh and that worked quite it was in the gas phase of course. Um, and that went really well. And you know, we had been in contact with Rean and IBM and IBM was interested in in sponsoring a Gordon Bell submission and so we we put it all together and I thought, well, gee, you know, we'll go for 10 or 15. Oh, that's going to be so challenging, right? Um but we're able to pull it off in this in this particular case it was in water and we were looking at a drug-like molecule binding you know that looking at computing the binding energy and uh I think the realization as I said IBM thing you know I I undersshot I mean we could have probably double or tripled the size and and actually been able to do that as well. So it really uh you know just everything like Jamie said you know it's like takes a village and just like everything just fell in place right and we were able to line our ducks up and away we go >> on seeing the scaling of this you know when you first saw this 12,000 atom paper um what was it like how did you react >> I mean for me it was remarkable right like we we had started publishing some work with uh some partners in 2018 where we were looking at you know how to how to do energy calculations for lithium hydide and you know seeing this scale up has just been uh really remarkable. At the time when we were when we were doing this a lot of you know when I was talking to folks in the in the chemistry community about it they were probably a little underwhelmed with the size of the molecules we were looking at. And so it often times came back where the comment would be like, "Well, that's great, but you know, I needed to be able to do something much larger." And so I feel like now like I my reaction to this is that we can say that scaling is no longer the issue with quantum uh and using quantum computers for uh studying chemistry. So I think it's a huge moment. I think it's really important for our field and a very good demonstration of how you can combine methods and algorithms to get to something much much bigger than people would expect. >> I mean, Kenny, these, you know, what Jamie's just said is really interesting to me, right? Because now we've caught the attention of these chemists and of these biologists. Um, and what's cool, I think, is that these 12,000 these these atoms, these proteins that you guys are simulating are are legit. Like, these are real molecules that people have heard of. Can you kind of walk us through what these molecules do and why they're so exciting? >> Yeah, I mean one's tripson uh benzamadine. So benzamadine is a small inhibitor. It's electrostatic driven. It's it forms a salt bridge with the protein. And uh so that was one side of the equation, right? We wanted something that was more like polar hydrogen bonding electrostatic interactions to see how you know could we do those kind of calculations. The other one um T4 lysosyme with nbutylenzene bound uh bound to it and that was on the hydrophobic side. So a typical drug molecule will have hydrogen bonding you know polar interactions but it also have non-polar and a lot of times the non-polar interactions are the major driver in a drug molecule binding to a receptor. So both of we sort of covered both ranges of these uh kinds of interactions. And just to respond to uh Jaime's comment, when I jumped over to Cleveland Clinic, I had several colleagues say to me, "Oh, MS, what the hell are you doing? You know, you're, you know, you'll be lucky if you do, you know, I don't know, methane or, you know, propane in the next five years, right?" And I sort of took that personally, right? And so it became my goal to is go as big as possible to exactly what Jamie said, to shut up the excuse my frame. I mean you know quiet the um naysayers and and I agree you know the once we have this kind of thing mechanism or structure in place workflow we can go after things like bigger basis sets you know more sophisticated treatments um it's you know this workflows um hardware agnostic solver agnostic so you know we can plug in anything you know newer methods QPE you know quantum phase his estimation could you know be plugged in tomorrow if the machine was available to to do these kinds of calculations. So really like as Jamie highlights I think the problem now isn't so much can we scale we check we can do that right now it's really like how do we get even better answers you know better basis sets higher quality calculations so that's the exciting part for me I think Jamie this scaling is um not just a quantum story and I think you got into it a little bit before but part of what makes this method so interesting to me is the fact that we're pulling all of this um sort of supercomput high performance compute even GPUs in way um can you talk about the contribution there and how REK got involved and what we're doing on this quantum ccentric supercomputing front? >> Yeah, so I think that you know one of the most exciting things about recent developments has been the integration of quantum computers with HPC and AI. And one of the the doors that that opens is that you can start really developing these customized workflows that leverage capabilities of CPUs and GPUs and QUUs in the correct fashion. So you just you assign the QPU to the part of the problem where it makes sense to use a QPU and the same thing for the GPU and the CPU. Um so I think all of it has to to come together. um the and it has a lot to do uh with the algorithm development and basically how you're actually setting your problem up and structuring it to leverage a heterogeneous uh compute structure. Um it will be even more interesting as time goes on and we start seeing more and more examples of exactly how workflows can leverage each piece of that architecture for different parts of the problem over time. um because I I think we're just going to start seeing cumulative examples of these um over time build up so that you can start getting to a place where you can generalize a little bit more. Right now I think it's still very bespoke. We need to focus in on the problem that we're looking at. We need to really kind of leverage the um infrastructure that we have access to and and utilize the quantum uh compute in the right way as a you know as a part of this overall structure. So I think that's what we're starting to see examples of and I think that's really what enabled the you know getting to the scales that we needed to for some of the things that you know Cleveland Clinic is interested in is other as well as other biologists. Um and I think we're we're just going to see more and more of that and creative science and and uh experimentation to figure out like where where each part of the compute is best suited in a problem solving. >> Got it. I mean, and then Kenny sort of taking that back to you. Can you walk me through how that actually worked, you know, in practice as you guys worked on this 12,000 atom study? >> Yeah, I mean, in terms of the collaboration, you know, Recan was just just fabulous. You know, I've known people at Recan for some years, you know, Sosan uh and then I met through the quantum network, IBM quantum network, you know, who are like the big drivers there. And, uh, you know, Sosan made sure we had the compute we needed. So we got two full days of Fugaku to do these calculations. Also they have a quantum device as does Cleveland Clinic. So we were given uh about 50 hours both sides allocation on these quantum devices um for a cumulative like uh 1.3 billion shots uh to to get these samples. U so that was the the quantum side. the there's a couple steps right to initiate it you need to do just a simple heartchie f calculation on the whole protein right and so this is done just using classical methods uh GPUs um and then there's some circuit parameterization steps and this can be done on CPUs u and then once you have the the circuits parameterized you do the shots on the quantum device and then reassembly occurs on classical hardware so that's sort of the the overall step The rate of limiting step is actually the diagonalization of the hilbert spaces that you get for the individual fragments. Uh that was the most expensive part and that's what predominantly used uh two full days for the two proteins of fugako to uh do those calculations. Got it. Yeah. And just to you know folks might not have heard some of these before. I want to make sure they understand that the way that this sort of algorithms actually working is you can imagine taking all of the you know most of quantum most chemistry can be represented in like these large what we call tensors but really it's just big old lists of numbers. Um so what we're trying to do is find essentially a because we need to approximate this. It's way too many numbers in order to actually explain this. We have to find a good subset of these numbers of this table that could well approximate our system. And then that is what the quantum computer does. And then we send that subset over to a classical computer that does the diagonalization which really is just saying turns that list of numbers into something that we can easily calculate the you know energies we're interested in from. Right. >> Yeah. Exactly. The I mean what's the dream? The dream is to have a quantum device that will return the u bitst strings and only those bitst strings that are needed to reach chemical accuracy for the calculation. That's the dream, right? And as you point out, I mean, currently the methods SQD and extended SQD and trim SQD, which are different variants and they improve as you go along that series, um, are attempts to, you know, reach that level of sophistication and, you know, my expectation over time and as Jamie is highlighting, there'll be a lot of, you know, intense interest in trying to actually achieve the dream, right? And um so stay tuned right I think there will be new methods and new approaches coming out and then once we get to the point where we can you know say we need 1.3 trillion Slater determinants you know to solve a particular problem but we only need 10,000 to get chemical accuracy right so eventually we'll have methods that start you know getting closer to that and then that's going to speed up the the whole calculation and then the accuracy will just continue to improve. Jamie, can you tell me about what the methods are that are going to take us there? I mean, I know that there's SQD is almost like uh it's what we're doing today, but in the future, we're going to get into things like quantum phase estimation, you know, things like that. >> Yes. So, I think you're uh you're aware that error correction and the possibility to, you know, run algorithms on fall tolerant devices are coming sooner uh than anyone even realizes. And so I think like the the magic in in sort of the research and the art of the science, if you will, is to start seeing how we can test out small examples of that and then ultimately take lessons that are learned from exploring that for chemistry problems alongside what we already know and what we've built up as a body of knowledge and then bringing together new methods. So that's what I'm predicting is going to happen is we start having you know the hardware made a available and you know start actually having you know logical cubits that you can you can try running some small circuits on. Um, and then I think at some point, you know, we're going to be building up, you know, our capabilities and then there's going to be an intersection point uh there with like all of a sudden you can say like this is hands down the best method, you know, that we could possibly use to study this, you know, molecular or biological system. And I think that that's the part the moment in time that I'm personally very excited about because I think at that point you're going to say like okay quantum is really bringing something new from a computational standpoint for the field of chemistry that we needed. you know, we really needed this to be able to to simulate um the systems that we want to simulate and to be able to guide experiments and find new therapeutics and find new catalysts and all the things that that we think that this has the potential for. So I think that that's if I had to predict, you know, it it kind of follows the error correction roadmap that we have set out in front of us um as to being able to try out different algorithms and then starting to piece things together in the right way to really really get to those uh you know substantial and valuable solutions. >> Yeah. I mean the way to think about it is like you had VQE, right? And so BQE had its moment in the sun sort of maybe fading a little bit for for computational quantum chemistry right let's just say uh because I know other fields are still using it sqd is now come to the four next week there may be something totally different right you know so things are evolving really quickly um on the way to say a quantum phase estimation a fault tolerant quantum device >> I'm sorry and get away from that sort of variable um kind approach where you have a variational approach to to the algorithm. Um, so those these this is where we're going to start I think seeing little glimpses of that and then I think like I said you know the the beauty of it will be when people start pulling it together to solve like these real problems. Kenny, once that's available, what does your workflow start looking like as a chemist? >> You know, I mean, basically the workflow is there. We we need to sort of get it optimized. Um, and this, you know, getting it scripted up and so we're working with Rekin and IBM to do that. Uh, you know, make it available in Kizkit. And then, you know, for us, we're starting to already explore ways we can do larger basis sets. Uh, we can do layering, right? So we can have you know a larger base set where it matters and and we can use a smaller one you know further out you know so there's just to me you know once I had this in place it's like there's hundred different directions we can go uh we're looking at circuit optimization as well with the collaboration with Sergey Stchuk at at Oxford. Um so we're we're going after this in many many different directions uh and continue to work you know closely with IBM on uh you know trying to do innovative things. The one thing I'll say that we haven't really touched on Jamie is you know we've really talked about bi biology right but this method can be applied to materials and an example is we've looked at something called florine lithium burillium molten assults in collaboration with DOE >> the five yeah [laughter] exactly I know it sounds like flipper or something right >> so it's not you know exclusive to proteins right Jamie I mean we could apply this all sorts of other different problems right so that's the another exciting aspect of it that we can explore in the coming weeks and months >> and I think we're going to take lessons learned from like people studying this for optimization and partial differential equations and like you know it's it can come from other parts of study and and the the really you know the brilliance is when you apply that to the problem and that has real application. Um so I think that we're going to just be taking you know lessons learned from all this but yeah I mean certainly materials uh and then you know the future you know thinking of systems that are just dynamic in nature and starting to look at other properties you know beyond ground state energies like what else can we explore um I think those are those are the really interesting open scientific questions. >> Can you guys like imagine with me what a like what it might look like when this stuff goes into production? I mean is there going to be a time when you know there's I we want to cure a disease and so you just ask your new found quantum centric supercomputer what the uh best molecule to solve it looks like is that what we're thinking so you know I think this is an interesting question right so the way I way I really describe it and you know Jamie knows this well is quantum you know quantum mechanics is the foundational method of computational chemistry if you take quantum mechanics out we have we don't have a method that can guarantee uh exact or near matching with experiment right so I think it's a the fundamental method right so there's a number of different ways you can go with this you could use it in simulations like I described to get free energies um certainly you could go that way another way to think about it is we and this is an area where there's not much data right uh we could evaluate potentially millions of protein liant complexes or poses, excuse me, and get ultra high accuracy energies and this then can be complemented say with an alpha fold type strategy uh to actually create like say machine learning uh type approaches to solve this problem. So I think it's as a data generation tool it could be really really powerful as well as a tool to understand you know very complex interactions and as you know Jamie mentioned uh excited states uh photos problems all these kinds of things where you know you really need this you know high level quantum rigor to to address these problems so that's kind of how I view it evolving um you know being a molecular guy I'd really love to be able to do it on the molecular side but I think there's also the realization of the power of machine learning and and how that you know this approach being a fundamental you know the cornerstone or foundation of our field could actually help propel you know methods like machine learning is just a stocastic you know approach you know you can't it depends on the data right but now if you're providing it with highle you know structural data from x-ray or you know uh different approaches combined with ultra you know ultra accur but very accurate energies. Um, you know, you have another direction, another avenue to go to solve drug discovery problems. >> So, right now it's like we're starting to see the very beginning examples of like where experiment is used to validate uh solutions that we're getting you know using quantum computing as a part of the workflow for the solution. Um and I think that that is going to continue right where you use experiment to validate. My hope and dream as an experimentalist uh myself is that someday we can use simulation to actually give us you know enough of like a reliable insight as to how the problem actually would evolve and experimental parameters that we would need that we could perform you know these uh calculations in silicico first and then take that and go run the experiment. So it's sort of like uh taking what is already starting to happen and then just like extending that out even further and that ultimately I think would save a lot of time um and a lot of money and a lot of resources uh if we could do that. >> Yeah. sort of like an iterative loop, right? Computation, make suggestions, if it works, you know, great. If it not, you know, feedback and Yeah, for sure. >> Right. Right. >> I mean, it's kind of beautiful when you're all putting it this way, which is you're really thinking beyond just, you know, what will quantum computing do as a tool for kind of day-to-day drug discovery, but really, how will quantum computing revolutionize the science of drug discovery overall? I mean, you know, from this like I imagine a molecule, I can use a quantum computer to just create it to even what Kenny is saying, which is like I can just create a suite of molecules that don't exist and then use machine learning methods to help uncover things about biology and chemistry overall. I mean, these are like these, you know, we're think we're not just shooting for like uh for tomorrow. We're shooting for the stars over here is what I'm hearing. >> Yeah. I mean, it's a dream, right? you know, can you, you know, actually get to the point where you can get these chemically accurate calculations done routinely to generate data sets for machine learning and also use them for exactly what Jamie is saying use it an iterative process to optimize catalyst or you know protein league interactions or photophysical problem properties you know whatever you might be interested in um I think that's been the dream all along right and I think you know uh there's this famous diary a quote which basically says, you know, all of chemistry and much of biology, you know, is defined by quantum mechanics, but the equations are too hard to solve. But we're getting to the point now where that's no longer the case, right? And then this dream could be potentially um realized. Um you know, I've been doing computational chemistry for a while and you know, it's exciting to see that we're we're tending in that direction. So, how big a quantum computer are we going to need to sort of achieve the dream of this uh of Cleveland clinics and recends and IBMs and the worlds? >> I don't know. Maybe maybe you should take that one. You know, as a computational chemist, you know, the answer is always as big as possible. >> Yeah. A bajillion cubits. Uh I think you know when we start seeing like uh actual like insights like and gaining insights into nature that we were not able to glean before cuz right like a lot of what we're talking about drug discovery, new materials discovery, catalyst discovery is actually really hard. It's really hard. Um there's not it's you know we've been chemistry is an old field. So there's, you know, most of the reactions are known. If I add A and B, I'm going to get C, you know. Um, and so how to change that or make something new, um, can be a real challenge. So I think once we start getting those insights and start being able to run circuits, I don't know how big quantum computer you're going to need for that. I do think that it's probably more going to be about uh sort of the art of putting together all of the different types of computation um so that you do gain these insights right um I think it's going to be more about that from a algorithms perspective yes running larger circuits and more complex circuits is what we need and going in the direction of error correction is what we need to do and so once we start getting you know larger and larger quantum systems that can run, you know, some of these like QPE and some of these other uh algorithms for chemistry. I think, you know, that's going to unlock a lot of that. So, it's going to have to happen at the same time. Um, so I think that's I I I you we used to see resource estimations that would be in the billions of of cubits. I don't know that that's true. Um, if you you all these different dimensions of compute together. >> No, I think that's right. Right. It's, you know, of course more is better, but you know, if you gave me a 100, 200 logical cubits today, I think we could do pretty amazing stuff already with EW with kind of this this workflow. So, that's why I'm really optimistic. Um, that we're not going to need to get to, you know, ridiculously large systems. Um, and it also depends on what you want to do. Like if you wanted to solve a whole protein on device, you know, that's a different challenge than using kind of these fragment based methods. And you know, this method is very tunable, right? So you can go bigger basis sets, but instead of doing atom by atom, we can go by functional group to functional group or we can go amino acid by amino acid um depending on the resource availability, right? and and again maybe we only focus the highle quantum calculations on the regions uh that really matter while the rest are handled you know using more simplified methods so there's like Jamie said there's many layers to this uh problem right it's not that we need a huge number of cubits to solve 20,000 atoms on device right um we don't need to go that way >> and I think that's why the focus has to be on algorithms and applications is because of that it's that's where the value capture is but also you know thinking about from the algorithm standpoint even with shores there's been some really clever you know advances in how to parallelize it and so you get this to much reduced like orders of magnitude reduc uh circuit depths needed for it so it's like all of these things combined I think is going to uh sort of dictate when that moment in time happens >> and I mean I agree with Jamie on like the circuit optimization right I like we're using just LUCJS you know which is kind of a standard but I think there's a lot more we can do to by improving that and the only thing I I'll say you know we're sort of what 100 200 cubit devices right [snorts] what happens when we have 10,000 or 100,000 you know with these are not unreasonable you know targets for cubits and what is the coding model for that you know I mean the circuit optim so I think there's just some really fascinating things that are going to be coming down the pike in terms of just you know how do you deal with these kinds of systems right >> yeah I mean what I'm sort of hearing which is I mean kind of amazing to think about is these 12,000 atom studies sure we're modeling proteins that exist but we're already straining these classical methods so we're already starting to push the envelope and so you can imagine that you know of course we're not solving all of quantum chemistry tomorrow but you know really even in the next couple years as you start to get access to these larger systems. As sort of error correction begins to hit, you know, even in the next couple two to three years, we're going to start seeing science being pushed forward and things uncovered that simply couldn't be covered uncovered just with a classical computer. And then it's that matter of, you know, you're talking about like it's not even about building better quantum computers, but in fact, it's about figuring out better ways to use the existing quantum computers. You're talking about new algorithms. You're talking about new mappings. the existing quantum computers. We're talking about, you know, new ways to run circuits so that we can do so in fewer gates, higher fidelities, error mitigation tricks. There's all sorts of stuff that like, you know, there's a lot to be excited about. I mean, that's what I'm hearing. >> Yeah, exactly. The, you know, I take it from the classical world, right? So, the, you know, the classical chips have gone faster. Yeah. I mean, that that's for sure. But the there's an equal contribution for algorithms, right? and new approaches which was equal to the performance in the in the hardware right and that cannot be forgotten. So not only are we going to get this you know interesting hardware stacks in the quantum computing world but there's also you know great room for algorithmic and novel improvements in circuit design things like that. So that's another certainly exciting and it should deliver just amount as much punch as the the hardware itself will deliver. I think that's a good place to ask our final question. Uh and so one at a time I will ask you let's start uh maybe Jamie first. What about the future of quantum computing gets you the most excited? >> Um the future of quantum computing is bright uh as you know the hardware continues to progress as we start figuring out how to leverage heterogeneous comput architectures to solve problems for chemistry. um you know this this topic's near and dear to my heart uh I think there's a lot that quantum will bring to the field of chemistry and so I think that and sort of the vision of the future uh saving resources saving time finding new catalysts you know and that's the dream and I think that's what gets me the most excited about the future of quanting computer >> you know I don't know if I can really say much more yeah you know I'm just we we have a construct now where we can start exploring you know pushing the boundaries and on on bigger systems on real molecules and I think that's very exciting both on new algorithms and hardware you know new hardware generations well thank you so much for a great conversation um Kenny and Jamie this was really fun and I will uh definitely be speaking to both of you soon I'm sure that's it for this episode of the coherence times if you enjoyed the conversation please be sure to subscribe wherever you get your podcasts. Comment in the comments section and share it with somebody who's curious about quantum. You can find us on Spotify, Apple Podcasts, and YouTube via the research channel. And for more episodes, resources, and deep dives, please visit us at ibm.com/think/mpodcasts. I'm Ryan Mandlebomb. Thanks for tuning in. And remember, the quantum future isn't just coming, we're building it right now.