WEBVTT 1 00:00:00.170 --> 00:00:00.940 STFC-RAL-CR03-RAL-R61-2.01: Great. 2 00:00:04.960 --> 00:00:05.760 STFC-RAL-CR03-RAL-R61-2.01: Thanks. 3 00:00:08.010 --> 00:00:08.720 STFC-RAL-CR03-RAL-R61-2.01: Stop. 4 00:00:20.080 --> 00:00:22.189 STFC-RAL-CR03-RAL-R61-2.01: There are spots here. 5 00:00:22.300 --> 00:00:25.069 STFC-RAL-CR03-RAL-R61-2.01: Jennifer Smith. Yeah, as well. 6 00:00:26.610 --> 00:00:28.590 STFC-RAL-CR03-RAL-R61-2.01: More chess here. 7 00:00:48.250 --> 00:00:57.420 STFC-RAL-CR03-RAL-R61-2.01: I think it's gonna be the original one. 8 00:00:58.490 --> 00:01:04.019 STFC-RAL-CR03-RAL-R61-2.01: I just wanted to read you. 9 00:01:14.940 --> 00:01:19.430 STFC-RAL-CR03-RAL-R61-2.01: Does anybody know how to set the… Worth. 10 00:01:19.590 --> 00:01:28.189 STFC-RAL-CR03-RAL-R61-2.01: You've been just… Don't do another list, don't do another list. 11 00:01:31.470 --> 00:01:34.170 STFC-RAL-CR03-RAL-R61-2.01: Yeah, that should be fine. 12 00:01:35.710 --> 00:01:36.530 STFC-RAL-CR03-RAL-R61-2.01: Yes. 13 00:01:43.590 --> 00:01:46.129 STFC-RAL-CR03-RAL-R61-2.01: You can see it. 14 00:01:59.670 --> 00:02:02.560 STFC-RAL-CR03-RAL-R61-2.01: Okay, so, it's good to see you in the household. 15 00:02:02.740 --> 00:02:09.530 STFC-RAL-CR03-RAL-R61-2.01: Welcome, everyone. Good morning, and welcome to today's, PPDC Seminar. 16 00:02:10.330 --> 00:02:11.959 STFC-RAL-CR03-RAL-R61-2.01: I want you to click? 17 00:02:13.560 --> 00:02:18.010 STFC-RAL-CR03-RAL-R61-2.01: There are sports here. 18 00:02:20.170 --> 00:02:26.860 STFC-RAL-CR03-RAL-R61-2.01: Here is one. There's another one here. Another one is there. 19 00:02:34.780 --> 00:02:35.620 STFC-RAL-CR03-RAL-R61-2.01: Stick. 20 00:02:41.290 --> 00:02:48.319 STFC-RAL-CR03-RAL-R61-2.01: So, welcome, again, to this, today's, WPD and ZCC Seminar. Today's talk sits very 21 00:02:48.450 --> 00:02:53.180 STFC-RAL-CR03-RAL-R61-2.01: Nicely, at the intersection of quantum computing, algorithms, and scientific discovery. 22 00:02:53.310 --> 00:03:04.029 STFC-RAL-CR03-RAL-R61-2.01: The central question is a fascinating one. Where can quantum computers actually provide an advantage over classical computers, and how can that advantage translate into use for scientific discoveries? 23 00:03:04.320 --> 00:03:07.629 STFC-RAL-CR03-RAL-R61-2.01: A few words about, Harry himself. Harry Barnman? 24 00:03:07.800 --> 00:03:13.019 STFC-RAL-CR03-RAL-R61-2.01: Is Chief Scientist for Quantum Algorithms and Innovation at Quantino. 25 00:03:13.260 --> 00:03:19.580 STFC-RAL-CR03-RAL-R61-2.01: and Professor of Algorithms, Complexity Theory, and Quantum Computing at the University of Amsterdam. 26 00:03:20.150 --> 00:03:29.340 STFC-RAL-CR03-RAL-R61-2.01: He's also the founding director of QSoft, the Dutch research center for quantum software, which he co-founded in 2015. 27 00:03:30.210 --> 00:03:34.799 STFC-RAL-CR03-RAL-R61-2.01: For those less familiar with Continuum, it is one of the leading companies 28 00:03:34.910 --> 00:03:43.169 STFC-RAL-CR03-RAL-R61-2.01: working on trapped IM, quantum computer… computing, and quantum softwares, bringing together quantum hardware, algorithms, and applications. 29 00:03:43.500 --> 00:03:54.890 STFC-RAL-CR03-RAL-R61-2.01: Harry has been one of the pioneers of quantum computing research in Netherlands and internationally. He established one of the earliest quantum computing research groups in the Netherlands in the late 1990s. 30 00:03:55.030 --> 00:04:03.829 STFC-RAL-CR03-RAL-R61-2.01: And his work has made foundational contributions to the quantum algorithms, communication, complexity, and computational complexity theory. 31 00:04:04.750 --> 00:04:10.639 STFC-RAL-CR03-RAL-R61-2.01: His research has helped establish both the power of quantum computation, and importantly, its limitations. 32 00:04:11.190 --> 00:04:19.610 STFC-RAL-CR03-RAL-R61-2.01: He was elected to the Royal Netherlands Academy of Arts and Sciences in 2020, and has received a Wiki grant, among numerous other distinctions. 33 00:04:20.019 --> 00:04:23.369 STFC-RAL-CR03-RAL-R61-2.01: And serves on several international scientific advisory boards. 34 00:04:23.890 --> 00:04:33.749 STFC-RAL-CR03-RAL-R61-2.01: And I have to admit, I have a little bit of personal interest in today's seminar. I'm using a continuum quantum computing for some of our research. 35 00:04:33.750 --> 00:04:51.019 STFC-RAL-CR03-RAL-R61-2.01: with NQCC colleagues, and also, I happen to own 3 shares of Monty Income. So, Harry, there is at least one shareholder in the Seminar who is particularly interested in Monty Income. Jokes aside, it is a real pleasure to have you here. 36 00:04:51.290 --> 00:04:59.870 STFC-RAL-CR03-RAL-R61-2.01: And I'm very interested to hear your perspective on complete quantum advantages to actual scientific discovery. So with that, please join me in welcoming, Harry Perman. 37 00:05:05.080 --> 00:05:09.369 STFC-RAL-CR03-RAL-R61-2.01: Thank you very much. Is there a chair for you? Yeah, yeah. Yeah. 38 00:05:09.980 --> 00:05:25.839 STFC-RAL-CR03-RAL-R61-2.01: you're almost majority shareholder. Thank you very much for inviting me and for the very nice introduction. Today, I want to talk a bit about 39 00:05:26.310 --> 00:05:28.969 STFC-RAL-CR03-RAL-R61-2.01: My vision of what you… 40 00:05:29.130 --> 00:05:40.399 STFC-RAL-CR03-RAL-R61-2.01: can do with a quantum computer, and how to approach that. And I chose this title, What Should Quantum Supercomputers Compute? 41 00:05:40.590 --> 00:05:45.249 STFC-RAL-CR03-RAL-R61-2.01: Because I think there is, there is a great, 42 00:05:45.460 --> 00:05:55.639 STFC-RAL-CR03-RAL-R61-2.01: promise in using quantum computers, AI, and HPC. But of course, my focus will be mostly on quantum computing and quantum algorithms. 43 00:05:57.020 --> 00:06:07.039 STFC-RAL-CR03-RAL-R61-2.01: But before I continue with this, I want to tell you a little bit a story about technology, scientific discovery, and 44 00:06:07.040 --> 00:06:17.080 STFC-RAL-CR03-RAL-R61-2.01: commercialization, and I want to go back, to 1674, to be precise, a couple of centuries back, to the Netherlands. 45 00:06:17.250 --> 00:06:35.700 STFC-RAL-CR03-RAL-R61-2.01: in Delft, where there was a gentleman called Antoni von Leuvenuch, and you see him sitting here, and he was a merchant in Kos. He was not at all a scientist. But, for his own amusement, he made a microscope, and you can see here a picture of that microscope. 46 00:06:35.810 --> 00:06:43.840 STFC-RAL-CR03-RAL-R61-2.01: Where the lens is really where I'm pointing at here. There's a tiny little lens, which he made himself. 47 00:06:43.940 --> 00:06:57.409 STFC-RAL-CR03-RAL-R61-2.01: And this was really a big feat of craftsmanship, because this microscope was magnifying 300 times better than any microscope around it at the time. 48 00:06:57.990 --> 00:07:01.909 STFC-RAL-CR03-RAL-R61-2.01: And so, what did Antoni do with this microscope? 49 00:07:02.070 --> 00:07:19.339 STFC-RAL-CR03-RAL-R61-2.01: Well, he started to play with it. He started to put all kinds of things under it, under the lens, and looked at it. I won't tell you exactly what kinds he tried, but one of the things he tried was a drop of water, a clear drop of water from the pond behind his house. 50 00:07:19.550 --> 00:07:27.539 STFC-RAL-CR03-RAL-R61-2.01: And he wasn't expecting to see anything, but to his bewilderment, he could really see what nobody had seen before. 51 00:07:27.660 --> 00:07:30.370 STFC-RAL-CR03-RAL-R61-2.01: And he saw something that looked like this. 52 00:07:30.570 --> 00:07:34.410 STFC-RAL-CR03-RAL-R61-2.01: We saw little animals rolling around in this… 53 00:07:34.870 --> 00:07:39.240 STFC-RAL-CR03-RAL-R61-2.01: Clear drop of water. And he actually makes, pictures of it. 54 00:07:39.840 --> 00:07:43.290 STFC-RAL-CR03-RAL-R61-2.01: And this is what it looked like, according to him. 55 00:07:43.430 --> 00:07:49.200 STFC-RAL-CR03-RAL-R61-2.01: And nobody believed him at the time that these things were actually independent. 56 00:07:49.490 --> 00:07:51.480 STFC-RAL-CR03-RAL-R61-2.01: But they were. 57 00:07:51.600 --> 00:08:08.930 STFC-RAL-CR03-RAL-R61-2.01: And, he called them, little animals in Latin, he called them Animoculus. But now we know they are microorganisms, they are bacteria, protozoa, algae, and this was a really very important discovery, because this was the birth of microbiology. 58 00:08:10.400 --> 00:08:25.110 STFC-RAL-CR03-RAL-R61-2.01: And then, of course, this opened up a whole lot of possibilities, mainly it led into industrial advantage, and it unlocked advances in medicine, agriculture, food production, biotechnology, etc. 59 00:08:25.640 --> 00:08:28.990 STFC-RAL-CR03-RAL-R61-2.01: And so why am I telling this story? Because there are… 60 00:08:29.330 --> 00:08:32.799 STFC-RAL-CR03-RAL-R61-2.01: Kind of three phases that you have to go through. 61 00:08:32.960 --> 00:08:37.699 STFC-RAL-CR03-RAL-R61-2.01: And the first phase was that you had to get better equipment, better hardware. 62 00:08:38.140 --> 00:08:39.429 STFC-RAL-CR03-RAL-R61-2.01: And then… 63 00:08:39.730 --> 00:08:49.999 STFC-RAL-CR03-RAL-R61-2.01: Antoni went into this scientific discovery phase, this plane with the machine, with the microscope in this case, and then a bit later came industrial advanced. 64 00:08:50.520 --> 00:08:57.569 STFC-RAL-CR03-RAL-R61-2.01: And with quantum computers, it's exactly the same thing. So, we at Continuum, and not only we, but 65 00:08:57.670 --> 00:09:15.110 STFC-RAL-CR03-RAL-R61-2.01: Similarly, we also are working on better hardware, and here you can see the roadmap that we have laid out for the coming years. We are now in 2025, where we have available online Helios, which is about 98 qubits. 66 00:09:15.270 --> 00:09:27.969 STFC-RAL-CR03-RAL-R61-2.01: And what is important is that the physical 2-qubit error is, small, is, 5, less than 5 times 10 to the minus 4, so that means that you can do 67 00:09:28.350 --> 00:09:35.979 STFC-RAL-CR03-RAL-R61-2.01: Roughly… 10,000 to 5,000 gates with… and still get a meaningful answer out. 68 00:09:36.860 --> 00:09:43.269 STFC-RAL-CR03-RAL-R61-2.01: And then, of course, you can add error correction and fault tolerance to it, or in this case, error correction to get 69 00:09:43.490 --> 00:09:58.900 STFC-RAL-CR03-RAL-R61-2.01: even more stable qubits, but then you get fewer of them. It's called logical qubits. Then in 2027, so next year, and this is already up and running in our labs, we double the number of qubits, we even make the error smaller. 70 00:09:59.250 --> 00:10:08.340 STFC-RAL-CR03-RAL-R61-2.01: And then in 2029, we will unveil Apollo, which will have thousands of qubits, and then in 2020, 30 71 00:10:08.340 --> 00:10:20.230 STFC-RAL-CR03-RAL-R61-2.01: we will have, Lumos, which has about a million kilos. And so, this is sort of the roadmap that we have laid out, and by the way, we have H1 and H2 available, I will say a little bit about it. 72 00:10:20.630 --> 00:10:21.410 STFC-RAL-CR03-RAL-R61-2.01: But… 73 00:10:22.250 --> 00:10:37.379 STFC-RAL-CR03-RAL-R61-2.01: going back to the story of von Leeuwenhoek, we are making better hardware, so now we are in the phase where we have to play with it, where we are in the scientific discovery phase, and then much later, industrial advantage comes. 74 00:10:38.090 --> 00:10:55.690 STFC-RAL-CR03-RAL-R61-2.01: And my bosses want to have this industrial advantage to be as forward as possible, and I sort of… I'm a scientist, and I want to have as much fun as possible in this space, and I guess we are all here scientists, so you probably agree with me on that. 75 00:10:55.850 --> 00:11:03.210 STFC-RAL-CR03-RAL-R61-2.01: So we are in this phase, in the scientific discovery phase, and I hope, and I expect also, that with this 76 00:11:03.290 --> 00:11:07.610 STFC-RAL-CR03-RAL-R61-2.01: new device, we can see things that we couldn't see before, or 77 00:11:07.640 --> 00:11:20.539 STFC-RAL-CR03-RAL-R61-2.01: rather compute things that we could not compute before. And we're really here in a realm with 98 qubits where you cannot simulate this anymore on a classical computer. It's just simply too 78 00:11:20.540 --> 00:11:30.650 STFC-RAL-CR03-RAL-R61-2.01: too big of a… of a state factor if you wanted to simulate that. And so the goal is to find algorithms and applications that can do something meaningful and something interesting. 79 00:11:32.310 --> 00:11:36.310 STFC-RAL-CR03-RAL-R61-2.01: Okay, so I have to say a little bit about Continium, because that's where I work. 80 00:11:36.440 --> 00:11:45.339 STFC-RAL-CR03-RAL-R61-2.01: We are a company that makes qubits, quantum computer that's based on trapped ions. 81 00:11:45.340 --> 00:11:56.510 STFC-RAL-CR03-RAL-R61-2.01: And each ion in itself is perfect and identical, and these ions, they carry the cupids, the suitable protein. 82 00:11:56.540 --> 00:12:02.279 STFC-RAL-CR03-RAL-R61-2.01: Ground state and an excited state form the 0 and the 1 of these ions. 83 00:12:02.840 --> 00:12:16.530 STFC-RAL-CR03-RAL-R61-2.01: And here you can see our device, H2. It's based on a quantum charge coupled device, and so we're pushing around these ions, which have charge. And you can see here. 84 00:12:16.530 --> 00:12:29.469 STFC-RAL-CR03-RAL-R61-2.01: a real… a real-life movie of the ions moving around in our trap. So our ions are not fixed at a certain location, but they can move around, and they can also overtake each other. They can switch places. 85 00:12:29.470 --> 00:12:42.760 STFC-RAL-CR03-RAL-R61-2.01: And then, the goal is that if you want to do an operation on two qubits, you first bring them together by letting one overtake the other, so that they're next to each other, and then you move them into these gate zones, these gray areas that you see here. 86 00:12:42.760 --> 00:12:52.449 STFC-RAL-CR03-RAL-R61-2.01: These are gate zones, and once you move these ions in there, you can hit them with the laser pulse, and then that… that effectively does an operation on them. 87 00:12:52.700 --> 00:13:04.469 STFC-RAL-CR03-RAL-R61-2.01: a C0 gate, or actually in our case, it's a ZZ gate, or you can have one ion in there and do a single rotation on your unit. So this is… this is how it operates. 88 00:13:04.680 --> 00:13:10.680 STFC-RAL-CR03-RAL-R61-2.01: I'm a computer scientist, so for me this looks like complete nonsense, but it actually does work. 89 00:13:10.900 --> 00:13:13.570 STFC-RAL-CR03-RAL-R61-2.01: Go forwards. 90 00:13:13.760 --> 00:13:32.919 STFC-RAL-CR03-RAL-R61-2.01: Yeah, so it has very low error. That's the lowest error in the industry. Another thing that we can do is we can, in the middle of the computation, we can do a measurement. By the way, you also do a measurement by putting qubits in this zone, and hitting it with another laser and see what light comes back that then measures these qubits. 91 00:13:33.070 --> 00:13:46.020 STFC-RAL-CR03-RAL-R61-2.01: And we can do mid-circuit measurements, so while the algorithm is still running, we can measure a few of these qubits, see what the answers are, and depending on the answer, change what the next gates are going to be. 92 00:13:46.020 --> 00:13:56.380 STFC-RAL-CR03-RAL-R61-2.01: operated on the remaining qubits. And even the qubits that are measured can be put back into the computation, and have to be initialized to zero. 93 00:13:57.900 --> 00:14:14.929 STFC-RAL-CR03-RAL-R61-2.01: So then, about a year ago, we unveiled Helios, which is the current best model, which has 98 qubits, and still has even a higher fidelity, so it has lower error than the previous one you saw, and the layout 94 00:14:15.070 --> 00:14:21.929 STFC-RAL-CR03-RAL-R61-2.01: of this one is slightly different than this racetrack that we had. It is kind of, 95 00:14:22.000 --> 00:14:32.630 STFC-RAL-CR03-RAL-R61-2.01: It has one big ring where the ions can move around, as before, and then we have these two legs that you see here, and then the ions can sort of be shuttled out. 96 00:14:32.650 --> 00:14:47.459 STFC-RAL-CR03-RAL-R61-2.01: And the eyes that need to be together can be shuttled out in the right amount, in the right time, so that they exit one after the other. And then here are these gate zones again. We have 8 now, and then they enter these gate zones. 97 00:14:47.660 --> 00:14:57.370 STFC-RAL-CR03-RAL-R61-2.01: Again, the same principle, lasers hit them, and then they fed back into this ring, and they circle around again, and they exit again in the right order. 98 00:14:57.820 --> 00:15:03.330 STFC-RAL-CR03-RAL-R61-2.01: So this is how we, operate. By the way, our, ions are… 99 00:15:03.760 --> 00:15:15.949 STFC-RAL-CR03-RAL-R61-2.01: Barium at the moment, and we have another ion that sits always next to it, so it's really a little… a little crystal. And the other one, the ecturium 100 00:15:15.950 --> 00:15:28.099 STFC-RAL-CR03-RAL-R61-2.01: ion is there to cool down the system. So, these ions, they heat up when you do these operations on them, and when they heat up too much, they lose their quantum information, so you don't want that, so you want to cool them down. 101 00:15:28.100 --> 00:15:40.669 STFC-RAL-CR03-RAL-R61-2.01: But if you cool them down, you might also change the state of the ion, and thereby lose the information. And so, what happens is that you cool down its neighboring ion. 102 00:15:40.670 --> 00:15:54.269 STFC-RAL-CR03-RAL-R61-2.01: And now we have, no worries. We have, a pinpoint, okay? So you pull down the neighboring ion, and then that, by sympathetic pulling, pulls the one next to it. 103 00:15:54.330 --> 00:15:59.799 STFC-RAL-CR03-RAL-R61-2.01: Without disturbing its, quantum information, its internal state. 104 00:16:00.530 --> 00:16:03.149 STFC-RAL-CR03-RAL-R61-2.01: And here you see an actual picture from the trap. 105 00:16:03.380 --> 00:16:11.119 STFC-RAL-CR03-RAL-R61-2.01: That we have. Unfortunately, I cannot show you the moving ions anymore, but these are the ions in our trap. 106 00:16:11.380 --> 00:16:13.650 STFC-RAL-CR03-RAL-R61-2.01: At some… at some snapshot. 107 00:16:15.520 --> 00:16:31.209 STFC-RAL-CR03-RAL-R61-2.01: So these are the three models that you've seen, actually. I didn't talk about H1, which was just a line, then we had this racetrack, and now we have this funny-shaped thing. The next one over, which is Sol, is now a square lattice. So you can sort of see that 108 00:16:31.210 --> 00:16:49.340 STFC-RAL-CR03-RAL-R61-2.01: heli is somehow in between the racetrack and the square lattice. We sort of practice a little bit with this crossover point, where ions can cross over from one track to the other. The next one is really a grid track, where we have gate zones at the end. 109 00:16:49.400 --> 00:17:00.799 STFC-RAL-CR03-RAL-R61-2.01: of the trap, and these ions can actually move around and make these sharp images that you can see, these sharp turns that you can see here. And this, again, is a footage of a real experiment. 110 00:17:03.000 --> 00:17:18.089 STFC-RAL-CR03-RAL-R61-2.01: Let me show you a little bit how a circuit runs on our machine. Imagine that this is the circuit that you want to run, so you have 3 qubits at play, I color-coded them red, green, and blue, and that you want to do Hadamar and a Hadamar on the 111 00:17:18.109 --> 00:17:28.420 STFC-RAL-CR03-RAL-R61-2.01: on the green and the red, and then you want to do a CNOT from the red to the green, a CNOT from the red to the blue, and then you do a measurement. 112 00:17:31.030 --> 00:17:47.260 STFC-RAL-CR03-RAL-R61-2.01: So, it starts out with these three ions over here, and by the way, these gray areas are the gate zones, and as you can see, we want to do a Hadenmar on the green one and on the red one, so that means that you hit them with that laser, and they instantiate the 113 00:17:47.380 --> 00:17:49.559 STFC-RAL-CR03-RAL-R61-2.01: The blue one, to zero. 114 00:17:50.570 --> 00:17:53.300 STFC-RAL-CR03-RAL-R61-2.01: So now the next thing, 115 00:17:53.660 --> 00:18:09.910 STFC-RAL-CR03-RAL-R61-2.01: is… well, these are… these are the Hadimars, so the first one was industrialization, the second time step is now doing the Hadimar on the red one and the green one. And now we want to do a CNOT, right from the red one to the green one. 116 00:18:10.090 --> 00:18:26.309 STFC-RAL-CR03-RAL-R61-2.01: over here, which means that we have to bring the green one and the red one next to each other, so we can actually move it over into the… so that they're now together in the same gate zone. And then we apply another laser pulse to it, which actually does this C-0. 117 00:18:26.540 --> 00:18:32.249 STFC-RAL-CR03-RAL-R61-2.01: And now we want to do a CNOT between the red one and the blue one, so we move over 118 00:18:32.360 --> 00:18:45.339 STFC-RAL-CR03-RAL-R61-2.01: the red one over the green one, and we moved to the blue one, and we hit it with a laser again. So now we have done the whole circuit, and then we do a laser pose to measure this particular QH. 119 00:18:45.410 --> 00:18:52.829 STFC-RAL-CR03-RAL-R61-2.01: And this is how it acts in our machine, of course, much faster than I just did, and also with all these gate zones in parallel. 120 00:18:53.350 --> 00:19:01.729 STFC-RAL-CR03-RAL-R61-2.01: But I find it really amazing that this actually works, that this does something. And I will show you some results that we have. 121 00:19:05.850 --> 00:19:14.100 STFC-RAL-CR03-RAL-R61-2.01: So, of course, the question is, what on earth can you compute with this machine that you couldn't already compute on a classical computer? And this has been… 122 00:19:14.290 --> 00:19:29.539 STFC-RAL-CR03-RAL-R61-2.01: haunting me since the mid-90s, and I still don't… I mean, I have a little bit of an answer, but I… I don't… I don't have the full answer here. But first, I want to tell you the story of, complexity theory a little bit, and how 123 00:19:29.570 --> 00:19:40.689 STFC-RAL-CR03-RAL-R61-2.01: we complexity theorists looked at the world, and the way we looked at the world was we looked at the worst-case analysis for a particular problem. So if you have a computational problem that you want to solve. 124 00:19:40.710 --> 00:19:56.150 STFC-RAL-CR03-RAL-R61-2.01: Then one measure that you can attach to that, sort of the time it takes to solve this problem, is by looking at the hardest instance among all the instances, and say, I have an algorithm that minimizes the time it takes to solve the hardest instance. 125 00:19:56.150 --> 00:20:00.870 STFC-RAL-CR03-RAL-R61-2.01: among all the instances. And that gives you a guarantee on the running time 126 00:20:00.960 --> 00:20:06.299 STFC-RAL-CR03-RAL-R61-2.01: Always, right? No matter what instance you put into the algorithm, it will never run more than 127 00:20:06.400 --> 00:20:09.460 STFC-RAL-CR03-RAL-R61-2.01: The time it took to run on the harvest, just by definition. 128 00:20:09.700 --> 00:20:13.559 STFC-RAL-CR03-RAL-R61-2.01: And that way, we have developed algorithms for a long time. 129 00:20:13.760 --> 00:20:21.539 STFC-RAL-CR03-RAL-R61-2.01: But that may be not a good measure, because it might be that the hardest instance is not the instance that you're interested in at all. 130 00:20:21.620 --> 00:20:25.409 STFC-RAL-CR03-RAL-R61-2.01: That there is an instance that can be solved much faster. 131 00:20:25.450 --> 00:20:40.029 STFC-RAL-CR03-RAL-R61-2.01: but isn't the hardest one, and you only care about solving this particular instance, and not about this exotic instance that's sort of out there, that you never encounter. So, what we need to do is to look at what 132 00:20:40.030 --> 00:20:51.409 STFC-RAL-CR03-RAL-R61-2.01: people call heuristics, and we need to have a quantum version of these heuristics. We need to argue about special, typical instances, and what are the best algorithms for these instances. 133 00:20:52.840 --> 00:21:08.780 STFC-RAL-CR03-RAL-R61-2.01: Now, this is, of course, difficult, and in practice, what we have been doing, and by the way, AI is an example of this. Like, if you look at what AI can do, if you model that as a problem, and you look at the worst-case instance for that problem. 134 00:21:08.780 --> 00:21:17.909 STFC-RAL-CR03-RAL-R61-2.01: AI should never be able to do what it is doing now. And it's able to do what it's doing now, because it only works for specific instances. It doesn't work for the worst case. 135 00:21:18.230 --> 00:21:24.540 STFC-RAL-CR03-RAL-R61-2.01: And so, how did AI work? Well, we had a big computer, and we could run it, and we could play with it. 136 00:21:24.650 --> 00:21:36.790 STFC-RAL-CR03-RAL-R61-2.01: So, this is what we need to do with the quantum computer as well, and the machines that are coming out are almost and are suitable to do that, and we need to have a little bit better ones. But we also developed a theory 137 00:21:36.910 --> 00:21:44.050 STFC-RAL-CR03-RAL-R61-2.01: to reason about these quantum heuristics, and there's a definition here that I want to throw out. 138 00:21:44.190 --> 00:22:03.629 STFC-RAL-CR03-RAL-R61-2.01: Mainly because I think it's… we have a very… we have a paper that makes this all precise, but I think it's very important to have a name for these instances, and these instances, I call them Queasy, or we call them queasy, where queasy stands for, on one hand, quantum easy. 139 00:22:03.690 --> 00:22:16.499 STFC-RAL-CR03-RAL-R61-2.01: And on the other hand, classical algorithms feel queasy. They cannot solve this problem very well. And so the queasy instances are the instances for a particular problem, where we have a fast content algorithm 140 00:22:16.500 --> 00:22:31.049 STFC-RAL-CR03-RAL-R61-2.01: For which there does not exist a classical opening. And we made this all precise using ideas from comorb complexity and instance complexity. I won't go into details, but at least this gives you a little bit of a handle to reason about these things. 141 00:22:33.280 --> 00:22:36.269 STFC-RAL-CR03-RAL-R61-2.01: And so the intuition is the following. 142 00:22:36.670 --> 00:22:45.479 STFC-RAL-CR03-RAL-R61-2.01: Here, I have a problem called satisfy. Who of you have heard of satisfy as a computational problem? 143 00:22:46.060 --> 00:22:47.270 STFC-RAL-CR03-RAL-R61-2.01: Nobody. 144 00:22:47.500 --> 00:22:51.259 STFC-RAL-CR03-RAL-R61-2.01: It's only bump. Bump, too. Two people. Very good. 145 00:22:51.390 --> 00:23:00.930 STFC-RAL-CR03-RAL-R61-2.01: Well, not very good, actually. Because I wanted to skip over this. The satisfiability is a… an optimization problem. 146 00:23:01.150 --> 00:23:13.220 STFC-RAL-CR03-RAL-R61-2.01: It's a famous optimization problem in computer science, and the question is, I give you a formula with variables and AND, and OR clauses and NOTs. 147 00:23:13.520 --> 00:23:23.940 STFC-RAL-CR03-RAL-R61-2.01: And, so a big formula, and then the question is, can you find an assignment to the variables, which are Boolean, which can be 0 or 1, such that this formula evaluates to true? 148 00:23:24.710 --> 00:23:35.429 STFC-RAL-CR03-RAL-R61-2.01: And this is sort of a… the… sort of a very famous problem, which is called NP-complete. And, what does that mean? It means that we… 149 00:23:35.930 --> 00:23:53.679 STFC-RAL-CR03-RAL-R61-2.01: don't have a fast algorithm for this, but we would like… we'd love to have a fast algorithm for this. And there are many, many other problems that are NP-complete that have the same property as satisfy. We would love to have a fast algorithm for them, but we don't. We only have exponentially slow algorithms for this. 150 00:23:53.700 --> 00:24:02.300 STFC-RAL-CR03-RAL-R61-2.01: And, they're NP-complete, because if you can solve one of these problems efficiently, then all the others are also solvable efficiently. 151 00:24:02.380 --> 00:24:16.269 STFC-RAL-CR03-RAL-R61-2.01: Maybe you've heard of the troubling salesman problem, or hacking problems, and there's hundreds of thousands of problems out there that, like, for example, computing the train schedule for the train, an optimal train schedule, is also an example of this. 152 00:24:16.330 --> 00:24:27.200 STFC-RAL-CR03-RAL-R61-2.01: So, to think of satisfiability as whatever your favorite hard problem is… By the way, solving ground states for classical Hamilton use is also one that you can do. 153 00:24:27.520 --> 00:24:37.449 STFC-RAL-CR03-RAL-R61-2.01: reduce, or is equivalent to satisfyability. And so satisfiability has instances in them that… that are hard, and instances in them that are easy. 154 00:24:37.870 --> 00:24:50.240 STFC-RAL-CR03-RAL-R61-2.01: that you can solve classically, easily, and these are what we solve every day on our computers. And then there is this new region now, the queasy Instances, which, which, which I have… 155 00:24:50.240 --> 00:25:05.010 STFC-RAL-CR03-RAL-R61-2.01: shown here. And typically, the analysis of satisfy would go to the worst case, to the hardest instances, so that would be focusing on this red area, whereas I think we shouldn't focus at the red area, we should focus at this purple area. 156 00:25:05.930 --> 00:25:08.140 STFC-RAL-CR03-RAL-R61-2.01: And so, 157 00:25:08.310 --> 00:25:16.809 STFC-RAL-CR03-RAL-R61-2.01: Here's another example, factoring, which I guess you know, given the number factored in its prime factors, to which we do have a 158 00:25:16.930 --> 00:25:29.720 STFC-RAL-CR03-RAL-R61-2.01: fast quantum algorithm due to Peter Shore, but we don't have a fast classical algorithm, so all the instances that are not easily solvable on a classical computer are crazy instances. 159 00:25:31.480 --> 00:25:33.949 STFC-RAL-CR03-RAL-R61-2.01: Okay, so this is a little bit tough. 160 00:25:34.380 --> 00:25:43.830 STFC-RAL-CR03-RAL-R61-2.01: computational complexity theory, but of course, I want to tell you what can you do with this new hardware? What are the problems that we are trying to solve? 161 00:25:45.760 --> 00:25:53.650 STFC-RAL-CR03-RAL-R61-2.01: And… There are two routes, really, to this quantum advantage and useful scientific applications. 162 00:25:54.670 --> 00:26:02.449 STFC-RAL-CR03-RAL-R61-2.01: And, one route is bottom-up. Start with a funky quantum algorithm, or a quantum trick. 163 00:26:02.660 --> 00:26:16.849 STFC-RAL-CR03-RAL-R61-2.01: And then, and then see where that leads you to, and the other is top-down. Start with a problem that you already wanted to solve, or maybe are already solving classically, and then see if there's somehow a subroutine in there 164 00:26:16.850 --> 00:26:31.640 STFC-RAL-CR03-RAL-R61-2.01: Or maybe the whole problem itself, that could be… could be replaced by a content algorithm, and then try to massage this content algorithm so that overall the whole complexity of this algorithm is better than what you had before. 165 00:26:34.610 --> 00:26:38.559 STFC-RAL-CR03-RAL-R61-2.01: So the bottom up, and actually, I want to argue that 166 00:26:39.510 --> 00:26:54.809 STFC-RAL-CR03-RAL-R61-2.01: we focus… we focus a lot on the top-down, and we should maintain doing that, but we should also focus on the bottom-up, which we're not doing so much. And so, the bottom-up, really starts with discover quantum native primitive, then 167 00:26:54.960 --> 00:26:59.180 STFC-RAL-CR03-RAL-R61-2.01: Try to do some resource estimate and validate it on hardware. 168 00:26:59.340 --> 00:27:10.859 STFC-RAL-CR03-RAL-R61-2.01: And then see if you can find a use case that it sort of can help, or that it can solve. Or maybe it gives you a completely new use case that nobody ever thought of, that you can… that you can solve. 169 00:27:11.720 --> 00:27:19.390 STFC-RAL-CR03-RAL-R61-2.01: And I want to give you three examples of this. The first one is complement sampling, which I've been working on myself. 170 00:27:19.610 --> 00:27:34.270 STFC-RAL-CR03-RAL-R61-2.01: And I'm quite excited about it. And then the other two are quantum topological data analysis and interacting electron dynamics. I'm going to show each of these, how we work with them on continuum. 171 00:27:34.660 --> 00:27:39.009 STFC-RAL-CR03-RAL-R61-2.01: Then the top-down, really, you start with an existing use case. 172 00:27:39.230 --> 00:27:55.050 STFC-RAL-CR03-RAL-R61-2.01: You do resource estimates and look at an end-to-end workflow that is classical and see what part of it you can make quantum, and then you validate small examples on hardware and try to… as the hardware grows, you try to make this, 173 00:27:55.090 --> 00:28:02.539 STFC-RAL-CR03-RAL-R61-2.01: These algorithms bigger and bigger until, hopefully, at some point, they will actually outperform the best classical algorithms. 174 00:28:03.040 --> 00:28:17.420 STFC-RAL-CR03-RAL-R61-2.01: And examples here are many, for example, material design, quantum chemistry, optimization, drug design, and there's really a big list of potential use cases that people look at. But I want to focus 175 00:28:17.420 --> 00:28:24.630 STFC-RAL-CR03-RAL-R61-2.01: on this part here, on the bottom-up. And by the way, the whole field started with this bottom-up approach. 176 00:28:24.710 --> 00:28:30.649 STFC-RAL-CR03-RAL-R61-2.01: Because the field started with an algorithm of David Deutsch, actually not far away here, from here, in Oxford. 177 00:28:30.770 --> 00:28:35.210 STFC-RAL-CR03-RAL-R61-2.01: Which just did a very silly thing. It could compute the parity of two 178 00:28:35.410 --> 00:28:44.900 STFC-RAL-CR03-RAL-R61-2.01: two Boolean variables with just one query, whereas classically you need two, and then that lets Adye Deutsch and Richard Josa 179 00:28:44.900 --> 00:28:57.199 STFC-RAL-CR03-RAL-R61-2.01: to come up with a problem that is called the Deutsch-Joseph problem, and then that led, eventually, to Peter Shore's factoring algorithm. But Peter Shore didn't start with factoring and worked his way back. 180 00:28:57.300 --> 00:29:03.820 STFC-RAL-CR03-RAL-R61-2.01: to this… the Deutschild and Deutsche's problem, it went the other way. It started with some 181 00:29:03.990 --> 00:29:10.109 STFC-RAL-CR03-RAL-R61-2.01: freaking funny content algorithm that was better than classical, and then that led to an application. 182 00:29:10.790 --> 00:29:13.860 STFC-RAL-CR03-RAL-R61-2.01: And I want to sort of focus more on that. 183 00:29:14.760 --> 00:29:18.249 STFC-RAL-CR03-RAL-R61-2.01: And of course, there's going to be interplay between these two. 184 00:29:18.390 --> 00:29:30.500 STFC-RAL-CR03-RAL-R61-2.01: And, all of this is sort of within the back of my mind. I want to hunt for these squeezy instances, because those are the ones that are actually interesting for a quantum computer. 185 00:29:30.740 --> 00:29:44.069 STFC-RAL-CR03-RAL-R61-2.01: And then, because the talk is also a little bit about how AI works, we really use AI whenever and wherever we can. And I guess you guys are doing that probably too, because it's getting so, so good and so smart, this. 186 00:29:44.210 --> 00:29:46.869 STFC-RAL-CR03-RAL-R61-2.01: You can use it for almost anything. 187 00:29:47.290 --> 00:29:51.290 STFC-RAL-CR03-RAL-R61-2.01: Maybe next time. I won't be standing here anymore. 188 00:29:51.440 --> 00:29:55.270 STFC-RAL-CR03-RAL-R61-2.01: But hopefully, for a little while, I can still enjoy it. 189 00:29:55.920 --> 00:29:59.359 STFC-RAL-CR03-RAL-R61-2.01: And here you see that in a, in a, in a picture. 190 00:29:59.470 --> 00:30:15.549 STFC-RAL-CR03-RAL-R61-2.01: So the top-down is you start with no use cases, and you sort of map it onto existing content algorithms, and the bottom up is to start with sort of funny quantum algorithms and see where they lead to. 191 00:30:15.550 --> 00:30:23.160 STFC-RAL-CR03-RAL-R61-2.01: for these two Hamiltonian simulations and QTTA, we already found some use cases, and for complement something, we are still looking. 192 00:30:23.160 --> 00:30:34.350 STFC-RAL-CR03-RAL-R61-2.01: And I sort of see it as sort of a… sort of an approach where you can now, when you do the top-down, you can reach other use cases that… that you would actually miss if you were starting from the top-down. 193 00:30:36.590 --> 00:30:37.490 STFC-RAL-CR03-RAL-R61-2.01: Fair enough. 194 00:30:37.940 --> 00:30:42.270 STFC-RAL-CR03-RAL-R61-2.01: By the way, if you have any questions, then please stop me. 195 00:30:43.700 --> 00:30:51.539 STFC-RAL-CR03-RAL-R61-2.01: So I want to talk a little bit about complement sampling, which appeared earlier this year in PRL. 196 00:30:52.120 --> 00:30:57.370 STFC-RAL-CR03-RAL-R61-2.01: Which was, done with, with these people, Benedetti and Beckermans. 197 00:30:58.380 --> 00:31:00.799 STFC-RAL-CR03-RAL-R61-2.01: And here's the idea. So this is… 198 00:31:01.960 --> 00:31:09.609 STFC-RAL-CR03-RAL-R61-2.01: top-down, right? So I'm just telling you something that I found funny, and we can hopefully find some good applications of this. So the idea is the following. 199 00:31:09.860 --> 00:31:15.220 STFC-RAL-CR03-RAL-R61-2.01: I start with a probability distribution D1, 200 00:31:15.780 --> 00:31:19.070 STFC-RAL-CR03-RAL-R61-2.01: And I want to produce a distribution detail. 201 00:31:20.160 --> 00:31:35.020 STFC-RAL-CR03-RAL-R61-2.01: And the rules of the game are as follows. I get one sample from distribution B1, and then I can compute, and then from that, I have to produce a sample from distribution D2. 202 00:31:35.710 --> 00:31:44.269 STFC-RAL-CR03-RAL-R61-2.01: For example, and this is very, very well studied in computer science, I start… I start out with the uniform distribution. 203 00:31:44.400 --> 00:31:59.649 STFC-RAL-CR03-RAL-R61-2.01: and then I feed that into a circuit, classical polynomial time circuit, efficient circuit, and then what comes out is a sample from distribution D2. And this is… all these distributions that you can make this way are called 204 00:31:59.660 --> 00:32:04.479 STFC-RAL-CR03-RAL-R61-2.01: efficiently computable distribution. So it's very easy to get the uniform distribution. 205 00:32:04.530 --> 00:32:12.539 STFC-RAL-CR03-RAL-R61-2.01: And with a little bit of effort, you can get all these other distributions out of it. So this is a paradigm that's very well, very well studied. 206 00:32:18.590 --> 00:32:19.980 STFC-RAL-CR03-RAL-R61-2.01: That's where it stops. 207 00:32:20.510 --> 00:32:21.630 STFC-RAL-CR03-RAL-R61-2.01: Let's see… 208 00:32:27.040 --> 00:32:31.679 STFC-RAL-CR03-RAL-R61-2.01: Seems to be frozen. Yeah, well, that's… I don't know. 209 00:32:50.030 --> 00:32:54.249 STFC-RAL-CR03-RAL-R61-2.01: In my point of view, one goes fine, but I think direct growth. 210 00:33:02.140 --> 00:33:03.660 STFC-RAL-CR03-RAL-R61-2.01: It's difficult, Jonathan. 211 00:33:25.920 --> 00:33:27.340 STFC-RAL-CR03-RAL-R61-2.01: If you don't you? 212 00:33:36.200 --> 00:33:39.740 STFC-RAL-CR03-RAL-R61-2.01: Yeah, sorry about that. Don't know why that happened. 213 00:33:39.940 --> 00:33:41.699 STFC-RAL-CR03-RAL-R61-2.01: Okay, so… 214 00:33:42.600 --> 00:33:56.799 STFC-RAL-CR03-RAL-R61-2.01: And I want to compare this classical setting, where you have the sample from this solution B1, produce a sample from this solution to do, with the classical, with a quantum one, where quantum sample is a coherent superposition over the 215 00:33:56.920 --> 00:34:06.789 STFC-RAL-CR03-RAL-R61-2.01: the strings in D1. So, a quantum sample looks like this state, where you have a superposition over all the strings in S, 216 00:34:07.070 --> 00:34:23.999 STFC-RAL-CR03-RAL-R61-2.01: And the DX squared is the probability of observing X. So it's a simple way of representing the distribution in a coherent way. And by the way, if you give this to a classical person, the only thing, or classical thing it can do is measure it. 217 00:34:24.000 --> 00:34:28.309 STFC-RAL-CR03-RAL-R61-2.01: And it will get… get a sample with, according to distribution, D. 218 00:34:29.230 --> 00:34:36.410 STFC-RAL-CR03-RAL-R61-2.01: By the way, are you familiar with this notation, and sort of these tricks. 219 00:34:37.250 --> 00:34:43.939 STFC-RAL-CR03-RAL-R61-2.01: Okay, so… The goal is the following. So, suppose that I give you a distribution 220 00:34:44.320 --> 00:34:46.200 STFC-RAL-CR03-RAL-R61-2.01: I give… I have a set. 221 00:34:46.360 --> 00:34:59.970 STFC-RAL-CR03-RAL-R61-2.01: that contains half of the elements of all the elements, of all the 2 to the n elements that I have of length… length n. So I have a universe, universe, and I have a set that sort of splits the universe perfectly in half. 222 00:35:00.070 --> 00:35:12.000 STFC-RAL-CR03-RAL-R61-2.01: But I don't know how it's split. I only know that it splits it in half. And I have a unit, and my distribution, D1, is uniform over all the strings in this first half. 223 00:35:12.350 --> 00:35:21.070 STFC-RAL-CR03-RAL-R61-2.01: And the goal is, given a sample from the first half, produce a sample from the complement, from the half that doesn't have any support. 224 00:35:22.750 --> 00:35:26.210 STFC-RAL-CR03-RAL-R61-2.01: So, in a game, it looks like this. We have a referee. 225 00:35:26.790 --> 00:35:31.959 STFC-RAL-CR03-RAL-R61-2.01: That chooses a random set from all the sets that have had the universe. 226 00:35:32.110 --> 00:35:37.439 STFC-RAL-CR03-RAL-R61-2.01: And then, it gives a classical player a sample Y from S, 227 00:35:37.550 --> 00:35:43.679 STFC-RAL-CR03-RAL-R61-2.01: And then the classical player, having seen Y, has to produce a string Y that's not in S. 228 00:35:44.520 --> 00:35:58.960 STFC-RAL-CR03-RAL-R61-2.01: Now, this is extremely difficult. For example, if the set… if I have the strings 0, sorry, 1, 2, 3, 4, 5, 6, 7, 8, so I have 8 strings in total, and I have a… and my set S is the first half, so it's 1, 2, 3, and 4. 229 00:35:59.070 --> 00:36:01.739 STFC-RAL-CR03-RAL-R61-2.01: 5, 6, 7, 8 are the complements. 230 00:36:01.960 --> 00:36:10.790 STFC-RAL-CR03-RAL-R61-2.01: But you don't know that it is. 1, 2, 3, 4, and now what you get is you get the string from S, say, 3, and now the goal for you is to produce 231 00:36:11.320 --> 00:36:14.460 STFC-RAL-CR03-RAL-R61-2.01: Either 5, 6, 7, or 8. 232 00:36:15.650 --> 00:36:18.609 STFC-RAL-CR03-RAL-R61-2.01: Now, since you don't know what the set is. 233 00:36:18.690 --> 00:36:37.200 STFC-RAL-CR03-RAL-R61-2.01: the best thing you can do is produce a random string that's not the one that you saw. The one that you saw, for sure, is not correct. And all the other ones are equally likely, as far as you know, to be in S or outside of S. So you just pick one at random, and then that, with probabilities slightly better than a half, gives you the right answer. So… 234 00:36:37.860 --> 00:36:55.479 STFC-RAL-CR03-RAL-R61-2.01: It's hard, classically, and you can prove very easily that the best thing you can do is really pick… pick a string, Y prime, that is not the one you got, and give that back. And that is correct, it's probably a half, plus 1 over 2 to the n, so basically a half. Slightly better than a half. 235 00:36:56.110 --> 00:37:09.979 STFC-RAL-CR03-RAL-R61-2.01: Now, quantumly, and oh, by the way, it's very efficiently verifiable for the referee whether the string he receives back is correctly not in S, or… because he knows what S and what SR is. 236 00:37:11.050 --> 00:37:17.849 STFC-RAL-CR03-RAL-R61-2.01: Now, quantumly, to my surprise, actually, it turned out that there is a very efficient algorithm, and I'll show you in a minute. 237 00:37:18.000 --> 00:37:29.320 STFC-RAL-CR03-RAL-R61-2.01: that, this is the algorithm that actually can do this perfectly. So there is… if you get coherent superposition over the strings in S, so if I give you superposition over. 238 00:37:29.330 --> 00:37:47.180 STFC-RAL-CR03-RAL-R61-2.01: Now, what's it, 1, 2, 3, 4? So, the sum of 1 half, 1 plus 2 plus 3 plus 4, then this algorithm that is on the next slide will produce a superposition over the other strings. And it will do this for any set S, without knowing what S is. 239 00:37:47.640 --> 00:37:53.570 STFC-RAL-CR03-RAL-R61-2.01: So here's a circuit. It's a very simple circuit. It contains of, Hammer gates. 240 00:37:53.700 --> 00:37:58.429 STFC-RAL-CR03-RAL-R61-2.01: and a Z gate, and it has, in the middle here, a very big 241 00:37:58.650 --> 00:38:06.189 STFC-RAL-CR03-RAL-R61-2.01: Control, control, control, not gate. And this is really a big, a big end gate, or it's also called a totally gate. 242 00:38:06.340 --> 00:38:13.490 STFC-RAL-CR03-RAL-R61-2.01: This, this, this sort of troubles this, this bit, even only if all these bits are, zero. 243 00:38:16.000 --> 00:38:30.920 STFC-RAL-CR03-RAL-R61-2.01: And now it turns out that if you put S in there, then S bar comes out on the other side. So if I put in the superposition over, for example, 1, 2, 3, 4, then what comes out here is a superposition 5, 6, 7, 8. 244 00:38:31.040 --> 00:38:39.010 STFC-RAL-CR03-RAL-R61-2.01: And again, no matter what this is, the superposition of the complement will come out. It's kind of a little bit strange that it does that so well. 245 00:38:39.320 --> 00:38:54.169 STFC-RAL-CR03-RAL-R61-2.01: Until my colleague observed that it's actually something that was studied before by Groffer, and maybe… have you heard of Groffer's algorithm? So this is, like, one of the famous algorithms that came after Peter Shore. 246 00:38:54.170 --> 00:38:59.489 STFC-RAL-CR03-RAL-R61-2.01: That, shows that you can search dramatically faster in the database. 247 00:38:59.490 --> 00:39:14.139 STFC-RAL-CR03-RAL-R61-2.01: And if you look at one building block of that algorithm, that's a Grover diffusion, then that… what it does, it sort of reflects the amplitudes around the mean of the amplitudes, and it sort of 248 00:39:14.360 --> 00:39:27.929 STFC-RAL-CR03-RAL-R61-2.01: flops them around. So, here we have a superposition over all the strings in S, and all the strings not in S don't have any amplitudes. And then one Grover iterate, or one Grover diffusion, actually 249 00:39:28.000 --> 00:39:41.709 STFC-RAL-CR03-RAL-R61-2.01: makes these guys all get diffused to zero, whereas the zero ones get diffused to what the amplitude of the original ones were. And so one step really 250 00:39:41.910 --> 00:39:46.589 STFC-RAL-CR03-RAL-R61-2.01: Gives all the amplitudes, now to the… to the other strings. 251 00:39:47.240 --> 00:39:56.160 STFC-RAL-CR03-RAL-R61-2.01: And this is basically why it works, but that's not how I discovered it. I just did the calculations, and to my surprise, it did what it did. 252 00:39:58.560 --> 00:40:01.050 STFC-RAL-CR03-RAL-R61-2.01: So here we have this very good 253 00:40:01.460 --> 00:40:09.440 STFC-RAL-CR03-RAL-R61-2.01: experiment for quantum supremacy, where we have a game where we can actually find a good S, 254 00:40:10.020 --> 00:40:22.569 STFC-RAL-CR03-RAL-R61-2.01: show that classically giving you a string in the complement is very hard, but quantumly, we have this easy circuit that should, in principle, do it perfectly. Now, there is one remaining problem, and that is 255 00:40:22.950 --> 00:40:30.809 STFC-RAL-CR03-RAL-R61-2.01: this works, and this hardness result really is only hard, because we argued about the random set S. 256 00:40:31.160 --> 00:40:41.010 STFC-RAL-CR03-RAL-R61-2.01: But it may be very difficult to produce superposition over a random set test. Actually, provably, this requires an exponentially large quantum computer to do. 257 00:40:41.050 --> 00:41:00.140 STFC-RAL-CR03-RAL-R61-2.01: So we cannot do that. We have to only consider random or sets S, that are easily, gender-ageable, and we found a way to do that. But first, I have to tell you that we can compute how well an experiment works by computing how much, how well, 258 00:41:00.490 --> 00:41:10.199 STFC-RAL-CR03-RAL-R61-2.01: The quantum computer works divided by how well the classical computer works, and how well it works means how well it can do better than a half. 259 00:41:10.330 --> 00:41:22.579 STFC-RAL-CR03-RAL-R61-2.01: And quantumly, we saw that… classically, we saw that it can only do 1 over 2 to the n better than a half, and quantumly, we saw that it can do perfect, in theory, so that's 1 half better than a half. 260 00:41:22.820 --> 00:41:32.480 STFC-RAL-CR03-RAL-R61-2.01: And so this ratio becomes, in the perfect case, 2 to the n minus 1. It's an exponential violation of, sort of, classicality. 261 00:41:33.580 --> 00:41:45.460 STFC-RAL-CR03-RAL-R61-2.01: And we also came with a good idea, and by the way, we used the help of AI here, we came with a family of easily computable sets S that can be 262 00:41:45.460 --> 00:41:53.629 STFC-RAL-CR03-RAL-R61-2.01: That can be implemented. And then still have this classically hard, property. 263 00:41:53.650 --> 00:41:56.659 STFC-RAL-CR03-RAL-R61-2.01: So you don't need to compute a very difficult X. 264 00:41:58.240 --> 00:42:03.810 STFC-RAL-CR03-RAL-R61-2.01: And then, of course, now that Prof is in the pudding, we implemented this whole thing on our 265 00:42:04.050 --> 00:42:05.950 STFC-RAL-CR03-RAL-R61-2.01: Quantum computer. 266 00:42:06.560 --> 00:42:23.900 STFC-RAL-CR03-RAL-R61-2.01: And this is what I find really amazing. So now imagine that we implement this on this islands that are moving around, and that go into these gait zones, and we implemented this algorithm, and here's what happens, and so focus on this appeared a couple of weeks ago in Nature Communications, and this is… this line here is the 267 00:42:24.010 --> 00:42:28.029 STFC-RAL-CR03-RAL-R61-2.01: It's the optimal Sort of ratio that you can have. 268 00:42:28.270 --> 00:42:44.419 STFC-RAL-CR03-RAL-R61-2.01: When there's no errors in the machine, and we sort of were able to go up to strings of length 35, or even a little bit bigger, and we get almost perfect fit with this, with this optimal exponential, violation. 269 00:42:45.260 --> 00:42:50.310 STFC-RAL-CR03-RAL-R61-2.01: And you see that it's starting to filter off here a little bit because of the errors in the machine. 270 00:42:50.560 --> 00:43:07.610 STFC-RAL-CR03-RAL-R61-2.01: So we need… as the inputs become bigger, you need more and more gates, and hence the errors start to accumulate. But still, up to 35, we could get a violation, which is of order 2 to the 35, right? It's exponential in this number, so you get an exponential advantage here. 271 00:43:08.200 --> 00:43:09.560 STFC-RAL-CR03-RAL-R61-2.01: And it's verified. 272 00:43:11.550 --> 00:43:27.490 STFC-RAL-CR03-RAL-R61-2.01: what we're doing now is trying to find applications of this. So this is like a funky, nice quantum algorithm. What it can do, we don't know yet. We found some applications in cryptography, and we're searching at the moment for other applications. 273 00:43:28.250 --> 00:43:33.599 STFC-RAL-CR03-RAL-R61-2.01: Okay, this is what I wanted to tell you about, complements, and we have much time do you already? 274 00:43:33.730 --> 00:43:37.649 STFC-RAL-CR03-RAL-R61-2.01: Still have, I can talk a little bit more, right? 275 00:43:38.960 --> 00:43:45.170 STFC-RAL-CR03-RAL-R61-2.01: The other is, about quantum topological data analysis. 276 00:43:45.470 --> 00:43:49.739 STFC-RAL-CR03-RAL-R61-2.01: And this is a team of Adam Connolly at Continuum, who's doing that. 277 00:43:50.280 --> 00:43:56.960 STFC-RAL-CR03-RAL-R61-2.01: And, the goal here is the following. If you have data that can be represented as a graph. 278 00:43:57.300 --> 00:44:13.479 STFC-RAL-CR03-RAL-R61-2.01: for example, interaction of people in an interaction graph, or in biology, which molecules interact with another molecule in the cell, then you can represent that as a graph, and a graph has notes. 279 00:44:13.480 --> 00:44:19.180 STFC-RAL-CR03-RAL-R61-2.01: Of the objects, and which ones are connected, or which ones interact with each other, and they have an edge between them. 280 00:44:19.410 --> 00:44:24.979 STFC-RAL-CR03-RAL-R61-2.01: But this is a very useful object to study and to analyze. 281 00:44:24.980 --> 00:44:42.490 STFC-RAL-CR03-RAL-R61-2.01: But there's actually more information in this data than just this interaction between individuals, individual nodes. And this is what topological data analysis tries to extract from the data. And the idea here is that instead of looking at this 282 00:44:42.490 --> 00:44:59.630 STFC-RAL-CR03-RAL-R61-2.01: point and just these node and edge relationships, look at subsets of nodes and how they are connected to other subsets of nodes. And note now that this object becomes much bigger. Like, a graph has only n nodes and most n squared edges, so you can 283 00:44:59.830 --> 00:45:03.869 STFC-RAL-CR03-RAL-R61-2.01: Right? Any graph has a big matrix, where you put the notes 284 00:45:04.380 --> 00:45:11.229 STFC-RAL-CR03-RAL-R61-2.01: on one side, and on the other side, and at position rj, you put a 1 if i interacts with J. 285 00:45:11.540 --> 00:45:26.340 STFC-RAL-CR03-RAL-R61-2.01: Now, if I want to know something about these higher dimensional structures, what happens in… with K nodes, and how they interact with K other nodes, my object becomes much larger, it becomes of size n to the K by n to the K. 286 00:45:26.420 --> 00:45:33.349 STFC-RAL-CR03-RAL-R61-2.01: And if K grows, like, order N over 2, then this is an exponential by an exponential large matrix. 287 00:45:33.690 --> 00:45:48.470 STFC-RAL-CR03-RAL-R61-2.01: But it does have this information in there. And by the way, if you have these matrices, then often all the information is in the eigenvalues of this matrix, and that's also the case in topological data analysis. 288 00:45:49.140 --> 00:46:00.139 STFC-RAL-CR03-RAL-R61-2.01: As you can imagine that either you miss this higher dimensional structure because you don't look at these subsets of K and other subsets of K, 289 00:46:00.320 --> 00:46:05.969 STFC-RAL-CR03-RAL-R61-2.01: Or, it takes a tremendous amount of time to compute this, because then you have to analyze this huge matrix. 290 00:46:07.620 --> 00:46:10.170 STFC-RAL-CR03-RAL-R61-2.01: So here is what I… 291 00:46:10.310 --> 00:46:28.979 STFC-RAL-CR03-RAL-R61-2.01: I already described, so this Laplacian matrix is this N to the k by n to the K size matrix, so this huge object that… whose eigenvalues you actually want to know, and they tell you what the topological structure is of this, of this data that you have at hand. And 292 00:46:29.330 --> 00:46:31.949 STFC-RAL-CR03-RAL-R61-2.01: So it grows exponential, but… 293 00:46:32.380 --> 00:46:48.840 STFC-RAL-CR03-RAL-R61-2.01: We have developed a quantum algorithm that doesn't have to compute this whole big structure. It can somehow superposition with some very nice tricks, access this data, and efficiently get some version of this higher dimensional 294 00:46:48.950 --> 00:46:51.260 STFC-RAL-CR03-RAL-R61-2.01: Features out of the data. 295 00:46:51.470 --> 00:46:55.930 STFC-RAL-CR03-RAL-R61-2.01: And this is what I call quantum topological data analysis, and here you see 296 00:46:56.170 --> 00:47:08.939 STFC-RAL-CR03-RAL-R61-2.01: sort of a schematic of… on the… on the top, you see the… a classical algorithm that sort of… it doesn't really matter how it works, but this is important here if you want to figure out 297 00:47:09.160 --> 00:47:23.069 STFC-RAL-CR03-RAL-R61-2.01: what this eigenvalue is, so this is lambda, that's the thing that you're trying to assess, then the shot count grows exponentially in 1 over lambda. So the classical algorithms run 298 00:47:23.250 --> 00:47:40.009 STFC-RAL-CR03-RAL-R61-2.01: exponential in 1 over lambda, whereas the quantum circuit that we developed runs linear in 1 over lambda. So there is an exponential difference in runtime in order to figure out what this lambda is, right? Lambda is the thing that tells you this higher dimensional structure. 299 00:47:41.880 --> 00:48:01.099 STFC-RAL-CR03-RAL-R61-2.01: And actually, it's kind of not as nice as I say here, because we don't get this lambda output, we get a normalized version, and we also don't get a normalized version, we get an approximation to that, because there's errors, and this algorithm isn't perfect. But it gives you an sort of a piece of this lambda 300 00:48:01.160 --> 00:48:04.890 STFC-RAL-CR03-RAL-R61-2.01: A piece of the cake that you couldn't get before. 301 00:48:05.610 --> 00:48:15.979 STFC-RAL-CR03-RAL-R61-2.01: And, actually, it turned out also that we're not really computing these lumnas or these Betty numbers themselves, but we… 302 00:48:16.200 --> 00:48:30.479 STFC-RAL-CR03-RAL-R61-2.01: calculate moments of this Laplacian. So, doesn't really matter, but there are sort of properties that you can get out that resemble topological data analysis that, that we can actually get from our machine. 303 00:48:32.090 --> 00:48:38.330 STFC-RAL-CR03-RAL-R61-2.01: And so, we applied this to a use case with, together with SoftBank in Japan. 304 00:48:38.440 --> 00:48:48.560 STFC-RAL-CR03-RAL-R61-2.01: And the idea was that maybe we can get from their telephone call representation data, we can maybe use this higher dimensional 305 00:48:48.750 --> 00:48:52.240 STFC-RAL-CR03-RAL-R61-2.01: Teachers to see if there's fraud or not fraud. 306 00:48:52.440 --> 00:49:00.630 STFC-RAL-CR03-RAL-R61-2.01: And so here… and it actually… it kind of worked, because you see that the… the blue line is what you sort of get with, sort of. 307 00:49:00.630 --> 00:49:12.769 STFC-RAL-CR03-RAL-R61-2.01: are the normal… the normal data points, whereas these orange ones are the fraudulent ones, and you can see that the topological information is different in these two. 308 00:49:12.810 --> 00:49:15.469 STFC-RAL-CR03-RAL-R61-2.01: So this higher dimension of features 309 00:49:15.900 --> 00:49:27.749 STFC-RAL-CR03-RAL-R61-2.01: are able… enable you to tell fraudulent data points from non-fraudulent parts. Now, there may be other methods, by the way, that do the same thing. The point here is that this particular 310 00:49:27.860 --> 00:49:29.909 STFC-RAL-CR03-RAL-R61-2.01: Method also works. 311 00:49:30.190 --> 00:49:45.480 STFC-RAL-CR03-RAL-R61-2.01: And, so then you can imagine that these features, these topological features, you feed them into some other machine learning or AI technique, which then has extra information about the data, which then can then hopefully better classify. 312 00:49:45.540 --> 00:49:58.359 STFC-RAL-CR03-RAL-R61-2.01: what it can do. Again, this is an example of top… of bottom-up, because we started with this funny quantum algorithm that didn't really compute exactly what we wanted, but some approximation of an approximation of something 313 00:49:58.360 --> 00:50:05.880 STFC-RAL-CR03-RAL-R61-2.01: that resembled it, but it's cool. We don't know how to calculate the classical, and it turns out that it actually is UX score for something. 314 00:50:06.100 --> 00:50:13.560 STFC-RAL-CR03-RAL-R61-2.01: And we're also trying this on other sets of data, for example, on biological data sets, to see whether it can help us. 315 00:50:13.930 --> 00:50:30.030 STFC-RAL-CR03-RAL-R61-2.01: Of course, it will become only very interesting when we can compute it on our… when we can use this on our bigger machines, where you cannot simulate anymore. Currently, these data points can still be simulated on a classical computer, but as the quantum computer grows. 316 00:50:30.030 --> 00:50:34.259 STFC-RAL-CR03-RAL-R61-2.01: We will be able to see or compute things that we couldn't compute before. 317 00:50:34.780 --> 00:50:46.619 STFC-RAL-CR03-RAL-R61-2.01: And so this is kind of our pipeline, where we have the classical algorithm, which is called COL, which up till now, this is all the data that we got was computed with COLE, can compute 318 00:50:46.790 --> 00:50:59.649 STFC-RAL-CR03-RAL-R61-2.01: classically, up to a certain point, how well it works, and what the data does. And then, whenever Cole finds something interesting, then we can sort of run it on bigger instances, and hopefully 319 00:51:00.060 --> 00:51:02.079 STFC-RAL-CR03-RAL-R61-2.01: If the benefits that are terrible. 320 00:51:03.860 --> 00:51:09.190 STFC-RAL-CR03-RAL-R61-2.01: Then, finally, I want to tell you a little bit of simulation of physics, which I guess is actually closest. 321 00:51:09.410 --> 00:51:13.659 STFC-RAL-CR03-RAL-R61-2.01: to what you guys are doing, so maybe I should have focused more on that. 322 00:51:14.710 --> 00:51:20.069 STFC-RAL-CR03-RAL-R61-2.01: And the example here is high-temperature superconductors. 323 00:51:20.230 --> 00:51:29.229 STFC-RAL-CR03-RAL-R61-2.01: And these are really of interest to many, many people, maybe even to ask you some, some would be… it would be beneficial to have these. I don't know. 324 00:51:29.230 --> 00:51:40.510 STFC-RAL-CR03-RAL-R61-2.01: your particle physicist. Yeah, the accelerators, yeah. But, but, I mean, as many companies who want that, for example, nuclear fusion, nuclear magnetic resonance, etc. 325 00:51:40.730 --> 00:51:52.950 STFC-RAL-CR03-RAL-R61-2.01: And, sort of the way it works is that you have these, these, objects which are, almost yttrium-barium copper oxides, and you sort of sprinkle them in certain. 326 00:51:52.980 --> 00:52:06.000 STFC-RAL-CR03-RAL-R61-2.01: way, and then if you do that in the right amounts, it's kind of almost like magic, then they become superconducting. But nobody understands exactly how and why, and with this one, it was recently found. 327 00:52:06.240 --> 00:52:18.940 STFC-RAL-CR03-RAL-R61-2.01: that if you shine some light on this object, then for a very short amount of time, you get superconductivity, but it's not stable. It's only for a very short amount of time. 328 00:52:18.940 --> 00:52:28.999 STFC-RAL-CR03-RAL-R61-2.01: And actually, nobody really understands why, so that… but there are some competing hypotheses why this… why this is the case. I just want to show you an example. 329 00:52:29.240 --> 00:52:43.890 STFC-RAL-CR03-RAL-R61-2.01: of how we can use a quantum computer to distinguish these hypotheses. So if we can figure out which one is correct, that can then maybe help us understand why this superconductivity happened, and maybe we can make it stable for longer periods of time. 330 00:52:44.080 --> 00:52:49.549 STFC-RAL-CR03-RAL-R61-2.01: And one such hypothesis postulates that the superconductor is described by an extended 331 00:52:49.550 --> 00:53:10.710 STFC-RAL-CR03-RAL-R61-2.01: Fermi-Hubber model. That's actually the normal way of describing these systems, but it's extended in this case, the Fermi-Hubert model, and that the laser destroys stripe ordering that competes with superconductivity. And so, this is a hypothesis that we can test on a quantum computer that's difficult to test on a classical computer. 332 00:53:11.210 --> 00:53:16.419 STFC-RAL-CR03-RAL-R61-2.01: And here is how the algorithm works. So here you have this extended 333 00:53:16.510 --> 00:53:31.010 STFC-RAL-CR03-RAL-R61-2.01: Franklin Hubbard, Hamiltonian, where this T prime, term here, T prime S, indicates the light that was turned on and off for a certain amount of time. 334 00:53:31.090 --> 00:53:37.859 STFC-RAL-CR03-RAL-R61-2.01: And, you want to now test whether, the… you want to do the following. 335 00:53:37.930 --> 00:53:51.999 STFC-RAL-CR03-RAL-R61-2.01: So, you prepare a room temperature state of the extended Herbert model. So, on our digital computer, we produce a state that is this, this we can do, and then we simulate this Hamiltonian. 336 00:53:52.070 --> 00:54:07.380 STFC-RAL-CR03-RAL-R61-2.01: And we change this parameter T over time, and sort of see what the difference is when… by turning on T or not turning on T prime. And then, we measure both superconductivity and stripe order, which is something you can also do. 337 00:54:09.430 --> 00:54:19.890 STFC-RAL-CR03-RAL-R61-2.01: By the way, this is how you measure that. How do we measure superconductivity? We do this by the Meissner effect. You can somehow put a current 338 00:54:19.890 --> 00:54:43.209 STFC-RAL-CR03-RAL-R61-2.01: through this… through this object, all, of course, in digital form, and measure whether this thing becomes conducting or superconducting. Oh, and the current will develop in response to… oh, sorry, you put the magnetic field, and then a current should come through. And if this is there, then it's a conductor, and if it's… sorry, if it's not there, it's a conductor, and otherwise it's a superconductor. 339 00:54:43.250 --> 00:54:46.659 STFC-RAL-CR03-RAL-R61-2.01: And this you can then do on our own computer. 340 00:54:48.100 --> 00:54:56.389 STFC-RAL-CR03-RAL-R61-2.01: I also want to say that about a year ago on our Helios, somehow the first steps towards such algorithms were made. 341 00:54:56.390 --> 00:55:14.649 STFC-RAL-CR03-RAL-R61-2.01: by this team, where they were able to do this on a reasonably large grid using all our 100 qubits, and they did some of these Hamiltonian simulations in the art. And I guess for all you guys, maybe this Hamiltonian simulation is something that's very interesting. 342 00:55:14.680 --> 00:55:23.289 STFC-RAL-CR03-RAL-R61-2.01: And that's, you know, Bri and I were talking a little bit, this is something very difficult to compute classically, and also this dynamics is very hard to compute classically. 343 00:55:23.440 --> 00:55:28.800 STFC-RAL-CR03-RAL-R61-2.01: But on the quantum computer, in some sense, this is almost what they're made to do. 344 00:55:30.790 --> 00:55:32.179 STFC-RAL-CR03-RAL-R61-2.01: Okay, I'm… 345 00:55:32.880 --> 00:55:47.810 STFC-RAL-CR03-RAL-R61-2.01: Summarizing, so I really want to advocate that we are in the scientific discovery phase at the moment, and that these industrial applications will come a little bit later, although, of course, we're trying to bring them forward as much as we can. 346 00:55:47.810 --> 00:55:53.590 STFC-RAL-CR03-RAL-R61-2.01: And by the way, there is merit in working with customers to see if we can find these applications, because 347 00:55:53.590 --> 00:55:55.420 STFC-RAL-CR03-RAL-R61-2.01: This way, you can already 348 00:55:55.460 --> 00:56:02.000 STFC-RAL-CR03-RAL-R61-2.01: Find out which algorithms you have, and which algorithms you can still develop, and still 349 00:56:02.070 --> 00:56:06.350 STFC-RAL-CR03-RAL-R61-2.01: Tweak so that it actually works for these… for these applications. 350 00:56:08.410 --> 00:56:28.050 STFC-RAL-CR03-RAL-R61-2.01: I also told you a little bit about quantum heuristics, how we should focus on actual instances and not on worst-case instances, and I told you about queasy instances, don't forget them, queasy instances. It's a cool, cool name. And then, I discussed two principles, top-down and bottom-up. 351 00:56:28.250 --> 00:56:32.050 STFC-RAL-CR03-RAL-R61-2.01: Approach to getting new quantum algorithms and applications. 352 00:56:32.760 --> 00:56:42.899 STFC-RAL-CR03-RAL-R61-2.01: And of course, use AI whenever you can, and I described these three examples, complement sampling, quantum topological data masks, and this interacting electron dynamics, or… 353 00:56:43.130 --> 00:56:45.970 STFC-RAL-CR03-RAL-R61-2.01: simulation of, of Hamiltonians. 354 00:56:46.490 --> 00:56:48.609 STFC-RAL-CR03-RAL-R61-2.01: And, that's what I wanted to tell you. 355 00:56:56.940 --> 00:56:57.890 STFC-RAL-CR03-RAL-R61-2.01: horses. 356 00:57:00.180 --> 00:57:03.030 STFC-RAL-CR03-RAL-R61-2.01: Unbelievable. So… 357 00:57:03.050 --> 00:57:20.819 STFC-RAL-CR03-RAL-R61-2.01: One thing, and I missed the first couple of moments of your talk, so you may have answered it then, but I wanted to understand better how you're using, you know, sometimes hardware, sometimes simulation, what your engagement is with the company that… I understand you're an academic, right, but you're working with Continuum, so how… 358 00:57:20.950 --> 00:57:39.520 STFC-RAL-CR03-RAL-R61-2.01: How vital is it to have that link, rather than you simulating, this? Classically unique. Yes, so how far… because you mentioned at some point that the classical simulation could prove that some of this works, and then at some point you have to go to the hardware. What triggers you to do that at some point? 359 00:57:39.580 --> 00:57:43.400 STFC-RAL-CR03-RAL-R61-2.01: So, that's an excellent question. So, the… 360 00:57:44.060 --> 00:57:48.280 STFC-RAL-CR03-RAL-R61-2.01: All of these problems that I discussed, they have quantum advantage. 361 00:57:48.770 --> 00:58:02.910 STFC-RAL-CR03-RAL-R61-2.01: So I believe. Which means that at some point, when this has become a little bit larger than what we studied, or already some of them have this property, you can no longer use a classical computer to compute these properties. 362 00:58:03.190 --> 00:58:12.559 STFC-RAL-CR03-RAL-R61-2.01: But the classical methods are still very useful, because it allows you to validate on smaller instances that your computer and your algorithms are actually doing what they're supposed to do. 363 00:58:12.900 --> 00:58:24.190 STFC-RAL-CR03-RAL-R61-2.01: But at some point, you cannot use your classical algorithms anymore, because they would just simply take too long, and you have to… you have to use the quantum hardware to do so. And we're currently 364 00:58:24.380 --> 00:58:28.840 STFC-RAL-CR03-RAL-R61-2.01: Kind of at the, at the, at sort of the borderline where the classical 365 00:58:28.980 --> 00:58:35.480 STFC-RAL-CR03-RAL-R61-2.01: I can still keep up with the quantum, but the quantum is sort of soon overtaking the barcode for these problems. 366 00:58:36.210 --> 00:58:41.079 STFC-RAL-CR03-RAL-R61-2.01: And the interaction with the actual hardware, at what point the… 367 00:58:41.500 --> 00:58:50.850 STFC-RAL-CR03-RAL-R61-2.01: at what point do you… do you need to use the very best and not… can you… can you get insights on how the very best would work from the less good, quantum computers? 368 00:58:51.120 --> 00:58:56.320 STFC-RAL-CR03-RAL-R61-2.01: You see what I mean? Can you use them to simulate the next version of themselves? 369 00:58:56.850 --> 00:59:02.650 STFC-RAL-CR03-RAL-R61-2.01: Well, I mean, if you want to have more qubits… Yeah, exactly, yeah. And so you need more qubits. 370 00:59:02.660 --> 00:59:20.859 STFC-RAL-CR03-RAL-R61-2.01: I suppose what I mean is, in going from classical computing to a fewer qubit machine, that presumably does get you closer, though, to understanding the larger qubit machine. Can you use the fewer qubit machine to, in effect, simulate a larger qubit machine in a better way than the classical could? Do you see what I'm… 371 00:59:20.860 --> 00:59:30.600 STFC-RAL-CR03-RAL-R61-2.01: Yeah, it's a good question. There are some results where you say… where they say, suppose that you want to have 100 qubits, but you only have 99. 372 00:59:30.770 --> 00:59:40.579 STFC-RAL-CR03-RAL-R61-2.01: can you simulate with the 99 and the 100 one? And, you can, but there's an exponential blow-up in each qubit that you add, so… 373 00:59:40.780 --> 00:59:45.819 STFC-RAL-CR03-RAL-R61-2.01: So the answer is we went from 100 to 200. 374 00:59:46.110 --> 00:59:50.310 STFC-RAL-CR03-RAL-R61-2.01: You cannot simulate a 200 machine with a 100 cubic machine. 375 00:59:50.480 --> 00:59:55.490 STFC-RAL-CR03-RAL-R61-2.01: That will… that will cost you 200-300 Loba, which is… 376 00:59:55.710 --> 01:00:01.730 STFC-RAL-CR03-RAL-R61-2.01: too much to handle. So you can maybe get by with one or two qubits, if you're lucky. 377 01:00:01.930 --> 01:00:08.200 STFC-RAL-CR03-RAL-R61-2.01: But really, you want to have more qubits. That's awesome. We're selling more qubits. Makes sense, of course. 378 01:00:08.200 --> 01:00:23.720 STFC-RAL-CR03-RAL-R61-2.01: It's good for us. I have a second one, if I may, which is about, you said use AI whenever you can. Yes. So, you talked about, you know, top-down or bottom-up, and I suppose, are you using AI to identify what those use cases are as well? Is it good at spotting what a good use case is? 379 01:00:23.720 --> 01:00:28.419 STFC-RAL-CR03-RAL-R61-2.01: So far, I don't think it was that good at doing that, so you see a lot of 380 01:00:28.870 --> 01:00:33.749 STFC-RAL-CR03-RAL-R61-2.01: Of lists out there, which are, to my mind, a little bit fantasy use cases. 381 01:00:34.020 --> 01:00:39.319 STFC-RAL-CR03-RAL-R61-2.01: What I meant more was, like, if you have a particular 382 01:00:39.940 --> 01:00:48.250 STFC-RAL-CR03-RAL-R61-2.01: use case in mind, or if you have a particular algorithm in mind, if you have a particular specific question, then we use AI a lot to figure out 383 01:00:48.400 --> 01:00:52.890 STFC-RAL-CR03-RAL-R61-2.01: To help us, identifying how to go further. 384 01:00:53.680 --> 01:01:03.000 STFC-RAL-CR03-RAL-R61-2.01: But maybe eventually it will be also very good. But we're using… we have a whole… a whole team, maybe let me say a little bit about that. That's the AI team that does 385 01:01:03.010 --> 01:01:21.659 STFC-RAL-CR03-RAL-R61-2.01: three things. One is it uses AI for quantum. So, for example, these pulses that… maybe, I'm not sure if you saw, but there are these pulses that we need to tune to get our qubits to actually get the right gates applied to them. Optimizing these pulses, we can use AI to do that. 386 01:01:21.660 --> 01:01:27.120 STFC-RAL-CR03-RAL-R61-2.01: Another one is using better air quality codes, quantum error collecting codes, who use AI 387 01:01:27.120 --> 01:01:44.259 STFC-RAL-CR03-RAL-R61-2.01: to help us optimize that. Then there… so this is the AI for quantum, then it's also the other way around, which this topological data analysis is kind of an example of, where you can use the quantum computer to compute some classical data that you can then feed into your AI so that that becomes better. 388 01:01:44.620 --> 01:01:51.860 STFC-RAL-CR03-RAL-R61-2.01: So we call that quantum data. It's really classical data that comes from a quantum computer. That was hard to compute classically. 389 01:01:52.340 --> 01:01:56.490 STFC-RAL-CR03-RAL-R61-2.01: And then there's a loop, which we call the Gen QAI loop. 390 01:01:56.640 --> 01:02:03.970 STFC-RAL-CR03-RAL-R61-2.01: where this sort of perpetually feeds into each other. So you produce a circuit that produces some quantum states. 391 01:02:03.970 --> 01:02:20.410 STFC-RAL-CR03-RAL-R61-2.01: and you compute the properties of that quantum state, and then, by the way, the quantum state, on your quantum computer, and then there's maybe some approximation of the ground state to some Hamiltonian, then that feeds back into your classical part, your classical AI, which then starts to 392 01:02:20.430 --> 01:02:26.899 STFC-RAL-CR03-RAL-R61-2.01: Compute a bit more, and then produces a new quantum circuit that is run on the computer, and this whole loop 393 01:02:27.070 --> 01:02:31.379 STFC-RAL-CR03-RAL-R61-2.01: is how you can envision quantum and AI all work together. 394 01:02:31.710 --> 01:02:36.880 STFC-RAL-CR03-RAL-R61-2.01: And I also want to stress that, in some sense, quantum is very good, because 395 01:02:36.910 --> 01:02:54.139 STFC-RAL-CR03-RAL-R61-2.01: it probably cannot be simulated completely by AI. AI can do a lot of things, but I don't think it can simulate a quantum computer, at least we don't believe so. So, AI and quantum are in some way orthogonal, and they kind of help each other, rather than competing. 396 01:02:56.490 --> 01:02:57.270 STFC-RAL-CR03-RAL-R61-2.01: Sweet. 397 01:02:57.960 --> 01:03:01.569 STFC-RAL-CR03-RAL-R61-2.01: Maybe it's a little bit of a theoretical question at the moment, but, 398 01:03:02.070 --> 01:03:07.129 STFC-RAL-CR03-RAL-R61-2.01: Doesn't this quantum enhanced AI have problems with GDPR and sustainability rules? 399 01:03:08.550 --> 01:03:20.799 STFC-RAL-CR03-RAL-R61-2.01: Yeah, probably. Well, GPR, not so sure. It depends a little bit what you put, what kind of data you put, who didn't put. Yeah. 400 01:03:23.550 --> 01:03:29.329 STFC-RAL-CR03-RAL-R61-2.01: I don't think the problems are… so, for example, when it's biological data, I don't think the problems are… 401 01:03:30.570 --> 01:03:41.950 STFC-RAL-CR03-RAL-R61-2.01: More complicated than they already are with classical computers, because data's there as well, so you just have to be sure that it doesn't leak, or that it's protected well enough. 402 01:03:42.410 --> 01:03:46.880 STFC-RAL-CR03-RAL-R61-2.01: And that… Yeah. 403 01:03:47.160 --> 01:03:50.970 STFC-RAL-CR03-RAL-R61-2.01: And get… 404 01:03:51.220 --> 01:03:59.880 STFC-RAL-CR03-RAL-R61-2.01: I don't know. I mean, I think we have already struggled with classical AI to make it explainable, and maybe we won't even succeed in doing that. 405 01:04:00.070 --> 01:04:09.169 STFC-RAL-CR03-RAL-R61-2.01: In my mind, we may… maybe we don't want to be able to explain what it does, but we want to be able to trust what it does. 406 01:04:09.710 --> 01:04:28.919 STFC-RAL-CR03-RAL-R61-2.01: And so… So we need to have verification methods that allow us to be certain, or almost certain, that what we got from the quantum computer, or from the AI, or the combination thereof, actually is what we wanted, that it completed. 407 01:04:28.990 --> 01:04:41.890 STFC-RAL-CR03-RAL-R61-2.01: And there are techniques, we're working on that, and there are nice techniques that you can use. And for the TDPR, there is actually something that's called blind quantum computing that allows you to compute 408 01:04:42.770 --> 01:04:47.009 STFC-RAL-CR03-RAL-R61-2.01: An algorithm, or data, and an algorithm on the computer. 409 01:04:47.630 --> 01:04:50.979 STFC-RAL-CR03-RAL-R61-2.01: Without a quantum computer knowing what it's computing. 410 01:04:51.410 --> 01:05:04.450 STFC-RAL-CR03-RAL-R61-2.01: And this you can have classically only other assumptions, like cryptographic assumptions, but in a quantum case, you can actually show that you don't need these assumptions, so that this assumption is perfectly secure. 411 01:05:04.560 --> 01:05:10.090 STFC-RAL-CR03-RAL-R61-2.01: So, in some sense, quantum can even be safer to use than classical. 412 01:05:10.730 --> 01:05:12.630 STFC-RAL-CR03-RAL-R61-2.01: But I'm a little bit out on it. 413 01:05:13.560 --> 01:05:14.250 STFC-RAL-CR03-RAL-R61-2.01: Nope. 414 01:05:14.400 --> 01:05:27.410 STFC-RAL-CR03-RAL-R61-2.01: About the error correction that you were talking about, how do you actually implement it in your existing machines, and what do you anticipate the evolution being in the million qubit, machine? 415 01:05:27.900 --> 01:05:47.669 STFC-RAL-CR03-RAL-R61-2.01: So, the way we implemented this, I talked about this mid-circuit measurement, and, before it, at the very beginning. So, in the middle of your circuit, you can measure a few qubits, and depending on what the values are, these measurements, change the gates of future qubits. This is exactly what error correction and problem computing 416 01:05:47.860 --> 01:05:48.890 STFC-RAL-CR03-RAL-R61-2.01: Thus. 417 01:05:49.150 --> 01:05:57.380 STFC-RAL-CR03-RAL-R61-2.01: And so we actually have already papers out where we show, in real time, how this can reduce the error. 418 01:05:57.990 --> 01:06:08.009 STFC-RAL-CR03-RAL-R61-2.01: In the future, this is the way to go, because you need to have the logical error so low that you can run a long, long algorithm. 419 01:06:08.160 --> 01:06:26.869 STFC-RAL-CR03-RAL-R61-2.01: So, for those who don't know, there is a threshold theorem that says if your physical error is below a certain value, then you can make the error arbitrarily small by using multiple qubits as one qubit, and then look at how the error is of this particular 420 01:06:27.010 --> 01:06:28.610 STFC-RAL-CR03-RAL-R61-2.01: ensemble digits. 421 01:06:29.640 --> 01:06:39.509 STFC-RAL-CR03-RAL-R61-2.01: And yeah, we have several, several proposals and ways of doing it. And in practical hardware, do you do computation and FPGAs, or… 422 01:06:39.870 --> 01:06:41.980 STFC-RAL-CR03-RAL-R61-2.01: Yeah, take it. 423 01:06:42.420 --> 01:06:46.940 STFC-RAL-CR03-RAL-R61-2.01: I think we do, actually, yeah. But now, again, that's not really mine. 424 01:06:47.400 --> 01:06:57.029 STFC-RAL-CR03-RAL-R61-2.01: I don't know exactly well enough how they do it, but then you need… in order to do this error correction, you need to have very fast computers close to the metal for it to work. 425 01:06:59.500 --> 01:07:19.429 STFC-RAL-CR03-RAL-R61-2.01: Yeah, because you need, from the… from the measurement outcome, which is called a similar measurement, that tells you what kind of error happened, you actually need to do quite some compute to figure out what that error was. So you need some classic, some heavy-duty classical compute, as close and as fast as possible to your quant environment. 426 01:07:23.760 --> 01:07:27.659 STFC-RAL-CR03-RAL-R61-2.01: Thank you for the wonderful talk. I actually want to ask more than one question. 427 01:07:27.820 --> 01:07:29.200 STFC-RAL-CR03-RAL-R61-2.01: Okay, thank you. 428 01:07:29.350 --> 01:07:34.499 STFC-RAL-CR03-RAL-R61-2.01: So, first, like, I… my, field is neutrinos. 429 01:07:34.620 --> 01:07:41.589 STFC-RAL-CR03-RAL-R61-2.01: And, one thing that's very important in our field is to evaluate, like, nuclear effects after interaction. 430 01:07:41.910 --> 01:07:44.140 STFC-RAL-CR03-RAL-R61-2.01: So I wanted to ask, like. 431 01:07:44.330 --> 01:07:53.980 STFC-RAL-CR03-RAL-R61-2.01: If quantum computers can, be fast and efficient at, simulating nuclear deterrence. 432 01:07:55.740 --> 01:08:04.709 STFC-RAL-CR03-RAL-R61-2.01: I think we kind of talked, well, a bit of… it depends. If your interactions are quantum mechanical in nature, which I believe they are. 433 01:08:04.960 --> 01:08:09.589 STFC-RAL-CR03-RAL-R61-2.01: Then, on a computer is a good, a good look to use. 434 01:08:09.740 --> 01:08:19.939 STFC-RAL-CR03-RAL-R61-2.01: However, if your interactions are just classical, or described by classical mechanics, you don't… you shouldn't use a quantum computer, because it's slow and noisy and whatnot. 435 01:08:20.060 --> 01:08:31.870 STFC-RAL-CR03-RAL-R61-2.01: If it is really a quantum mechanical description, and if it has… like, for example, in quantum chemistry, if it has this strongly correlated system, so there's a lot of entanglement. 436 01:08:32.050 --> 01:08:43.729 STFC-RAL-CR03-RAL-R61-2.01: that is also being formatted or at play when you analyze your system, then a quantum computer is a good one to use, because it can do that. And your classical methods 437 01:08:43.830 --> 01:08:51.800 STFC-RAL-CR03-RAL-R61-2.01: Or, in my terms, if your neutrino question, it's a queasy question, you should use a quantum computer, otherwise not. 438 01:08:52.229 --> 01:09:06.309 STFC-RAL-CR03-RAL-R61-2.01: So, it's just a… probably from the error correction, I mean, Gerard and I, we've been using competitive analysis for Ivan and ZoomMartin, and we do realize that, you know, the error rates give us error rates have been continuous are much, much lower comparatively. 439 01:09:06.430 --> 01:09:11.950 STFC-RAL-CR03-RAL-R61-2.01: industry has sort of acknowledged it by multiple companies announcing that they're going to also have 440 01:09:12.689 --> 01:09:22.459 STFC-RAL-CR03-RAL-R61-2.01: Do you think Continu would be able to keep its advantage on that? How is the competition trend looking? 441 01:09:23.130 --> 01:09:41.140 STFC-RAL-CR03-RAL-R61-2.01: I certainly hope so. I work for Martin, but at the moment… well, it's hard to predict, right? But at the moment, our machines have the highest fidelities, or the lower, lowest error, and we have the highest number of qubit counts. 442 01:09:41.660 --> 01:09:52.530 STFC-RAL-CR03-RAL-R61-2.01: with those fidelities. For example, IBM has more qubits, but the fidelities are lower. By the way, fidelity is not the only thing that is important, also is important 443 01:09:52.689 --> 01:09:59.039 STFC-RAL-CR03-RAL-R61-2.01: the time it takes to do your computation, or the throughput. And actually, continue is not that good, because it's a slow 444 01:09:59.280 --> 01:10:02.539 STFC-RAL-CR03-RAL-R61-2.01: sheet. So although the fidelities are… are… 445 01:10:03.070 --> 01:10:06.790 STFC-RAL-CR03-RAL-R61-2.01: Hi, and we have this other thing, this old-to-all connectivity. 446 01:10:07.110 --> 01:10:24.210 STFC-RAL-CR03-RAL-R61-2.01: our machine is slow, but this all-to-all connectivity and lower error rates allow for, sort of, more efficient error correcting codes, and so overall, that is an advantage we have, and the disadvantage is speed. And for example, IBM has a very fast machine. 447 01:10:24.360 --> 01:10:35.510 STFC-RAL-CR03-RAL-R61-2.01: But its fidelities are not very good, and so now the question is, which of these two is going to win? Because they need to do more error correction. They don't have all-to-all connectivity, so they have to do swaps, so they'll launch also 448 01:10:36.270 --> 01:10:37.610 STFC-RAL-CR03-RAL-R61-2.01: Without the air. 449 01:10:38.560 --> 01:10:45.670 STFC-RAL-CR03-RAL-R61-2.01: it's… it's unclear which one is, which one's better, but I think the predictions are that they're about… about the same speed. 450 01:10:46.090 --> 01:10:49.399 STFC-RAL-CR03-RAL-R61-2.01: If you take everything into account. 451 01:10:49.830 --> 01:10:54.080 STFC-RAL-CR03-RAL-R61-2.01: But in some sense, I mean, I don't know. I mean, I surely hope that continues. 452 01:10:54.240 --> 01:11:10.619 STFC-RAL-CR03-RAL-R61-2.01: will be the one that is the winner, and if I were CEO, I would am not, I would say, yes, of course, continuing to maintain its competitive advantage. If there are any questions online, please raise your hands. Can I ask the other questions as well? 453 01:11:10.680 --> 01:11:21.990 STFC-RAL-CR03-RAL-R61-2.01: Thank you. So it might be, like, in a completely different direction, but are there, like, any cases where classical computers are actually faster than quantum computers? 454 01:11:22.150 --> 01:11:37.849 STFC-RAL-CR03-RAL-R61-2.01: Oh, yes. Oh, yeah, they are, yeah. Actually, almost everything you can think of. For example, if you… if you want to do, if you use your text processor, or your work program. 455 01:11:37.950 --> 01:11:49.250 STFC-RAL-CR03-RAL-R61-2.01: Well, you don't want to do that on a quantum computer. First of all, you don't have a lot of qubits or bits available. Second of all, it would be full of errors, and it would be extremely slow, so it would better use your browser. 456 01:11:49.880 --> 01:11:58.560 STFC-RAL-CR03-RAL-R61-2.01: And this is true for almost all of the tasks that you use your computer for. Actually, if you are using your computer and it works well. 457 01:11:58.930 --> 01:12:07.650 STFC-RAL-CR03-RAL-R61-2.01: then you use it, right? It ain't broke. But for those problems where it doesn't work. 458 01:12:07.750 --> 01:12:12.349 STFC-RAL-CR03-RAL-R61-2.01: It might be the case that the comment would be, like, enough. 459 01:12:12.790 --> 01:12:14.230 STFC-RAL-CR03-RAL-R61-2.01: Nice. Yeah. 460 01:12:15.180 --> 01:12:25.010 STFC-RAL-CR03-RAL-R61-2.01: Well, it's something I ask for, like, other, like, quantum computing seminars, so is there, like, any chance of, a quantum memory system? 461 01:12:25.350 --> 01:12:34.619 STFC-RAL-CR03-RAL-R61-2.01: Because, like, if you have a computer, you… it's not only the CPU that does everything, you… you need RAM registers, so is there any… 462 01:12:35.310 --> 01:12:39.350 STFC-RAL-CR03-RAL-R61-2.01: Yes. Yes. I mean, in theory, we know exactly how to do it. 463 01:12:41.310 --> 01:12:51.340 STFC-RAL-CR03-RAL-R61-2.01: We will need the quantum memory. I think there is this thing called quantum RAM, where you want a classical… you want your classical data accessible 464 01:12:51.590 --> 01:12:56.989 STFC-RAL-CR03-RAL-R61-2.01: in superposition in a coherent way. It's called QRAM, and also there are proposals. 465 01:12:57.130 --> 01:13:03.750 STFC-RAL-CR03-RAL-R61-2.01: We're doing taxes, so… Only proposals, it hasn't been demonstrated to work in real life. 466 01:13:05.490 --> 01:13:13.980 STFC-RAL-CR03-RAL-R61-2.01: one reason I was thinking about quantum memory is that, I realized that if you look at these supercomputers, like, like, 467 01:13:14.550 --> 01:13:28.040 STFC-RAL-CR03-RAL-R61-2.01: is it called Frontier, and IBM, and Fugargo. They actually don't operate for too long. I mean, they work for a day, maybe, or maybe less, and then they break down some component phase. 468 01:13:28.190 --> 01:13:36.940 STFC-RAL-CR03-RAL-R61-2.01: And what they do is they do this checkpointing, so they come… every now and then, they just store the whole contents of the memory. 469 01:13:37.110 --> 01:13:52.340 STFC-RAL-CR03-RAL-R61-2.01: And then they compute, and if they compute it again for a certain amount of time, they update the checkpoint. But if in between, somehow the machine breaks, they can go back to the previously stored checkpoint and start from there, so not always lost. 470 01:13:52.420 --> 01:14:03.469 STFC-RAL-CR03-RAL-R61-2.01: I was thinking quantum memory would be… oh, and by the way, quantum computers, I don't expect them to be better than these supercomputers. They're probably also gonna, sort of, break it down if you run them for too long. 471 01:14:04.280 --> 01:14:06.570 STFC-RAL-CR03-RAL-R61-2.01: So we need a quantum version of this. 472 01:14:06.720 --> 01:14:20.180 STFC-RAL-CR03-RAL-R61-2.01: And I haven't been able to figure out how to do that, because it's… it's complicated, because you can't just store quantum information and continue on still. Quantum nature allows you… doesn't allow you to copy information. 473 01:14:20.550 --> 01:14:27.169 STFC-RAL-CR03-RAL-R61-2.01: So you need to come up with the smart trick. But I certainly think that there will be, in the future, modalities to do 474 01:14:27.290 --> 01:14:30.589 STFC-RAL-CR03-RAL-R61-2.01: With quantum memory and other modalities that do fast computing. 475 01:14:30.830 --> 01:14:35.769 STFC-RAL-CR03-RAL-R61-2.01: I don't think there will be this distinctive, but we're very far apart from that at the moment. 476 01:14:36.360 --> 01:14:37.620 STFC-RAL-CR03-RAL-R61-2.01: Number of questions? 477 01:14:39.000 --> 01:14:52.399 STFC-RAL-CR03-RAL-R61-2.01: So, you're both a professor at university and an officer at Continuum. Yes. Do you think there's a reason why quantum computing seems to be so dominated by companies and not universities, or, like, places like, you know, Row or CERN or whatever? 478 01:14:52.590 --> 01:14:55.040 STFC-RAL-CR03-RAL-R61-2.01: Why are private corporations so much better? 479 01:14:55.250 --> 01:14:58.220 STFC-RAL-CR03-RAL-R61-2.01: Scenes and making and pulling from the cures. 480 01:14:58.370 --> 01:15:15.829 STFC-RAL-CR03-RAL-R61-2.01: Well, let me say that, no, I don't know what the answer here. It should be friendly to the university, too. The thing is, the whole thing started in university, right? University, in academia, 30 years ago. 481 01:15:15.850 --> 01:15:25.869 STFC-RAL-CR03-RAL-R61-2.01: And, then the first systems, I mean, Jericho was one of them, were built in the labs, and QCubits worked, and sort of systems grew. 482 01:15:25.870 --> 01:15:37.149 STFC-RAL-CR03-RAL-R61-2.01: And now, we're at a stage that it's really a lot of engineering to build these bigger systems, and you can no longer expect academia to do that, first of all, because it's too expensive. 483 01:15:37.150 --> 01:15:39.980 STFC-RAL-CR03-RAL-R61-2.01: Costs a shitload of money to do this. 484 01:15:39.980 --> 01:15:46.499 STFC-RAL-CR03-RAL-R61-2.01: the system. Second, it also requires more than 4 years of work, so you cannot have a PhD 485 01:15:47.150 --> 01:16:02.580 STFC-RAL-CR03-RAL-R61-2.01: work on it, right? You want someone to work on it for an extensive period of time. So, the reason why you see companies now take the forefront in building color communities is because of these two reasons. Mostly because of these two instances. 486 01:16:02.680 --> 01:16:03.439 STFC-RAL-CR03-RAL-R61-2.01: Thank you. 487 01:16:04.430 --> 01:16:09.960 STFC-RAL-CR03-RAL-R61-2.01: But it doesn't mean, by the way, that there's no place for academia. I think there's a big place for academia. 488 01:16:10.660 --> 01:16:18.260 STFC-RAL-CR03-RAL-R61-2.01: For example, in developing new algorithms, but also maybe developing new techniques, or at least questioning our memory, or applying computers. 489 01:16:18.960 --> 01:16:21.140 STFC-RAL-CR03-RAL-R61-2.01: We have one more to ask questions, comments? 490 01:16:23.160 --> 01:16:40.500 STFC-RAL-CR03-RAL-R61-2.01: You mentioned earlier coming up with algorithms to solve, like, for example, I think you said the complement song thing you came in? Yeah, yeah. How… what is your, like, process for actually solving that? I wish that I had a process, I mean, I don't know. Sort of… 491 01:16:40.550 --> 01:16:48.669 STFC-RAL-CR03-RAL-R61-2.01: Yeah. It's like doing science, like, what's the process for doing science and coming up with a good idea? You start out by reading 492 01:16:48.940 --> 01:16:50.400 STFC-RAL-CR03-RAL-R61-2.01: Papers, and… 493 01:16:51.090 --> 01:16:58.750 STFC-RAL-CR03-RAL-R61-2.01: focusing on a particular problem, and at some point, I don't know how it works, but maybe you guys know, but for me, then sometimes something comes, you wake up. 494 01:16:58.850 --> 01:17:09.250 STFC-RAL-CR03-RAL-R61-2.01: One night, or you've just… you found something, or you're just playing around, and you see, oh, this is funny, let's explore a little bit further. It's really a bit of luck, I would say, but… 495 01:17:10.650 --> 01:17:13.459 STFC-RAL-CR03-RAL-R61-2.01: I don't know how that works. 496 01:17:14.340 --> 01:17:22.359 STFC-RAL-CR03-RAL-R61-2.01: Let's… let's maybe call it intuition. Let's end there and thank our speakers a little… 497 01:17:28.020 --> 01:17:39.580 STFC-RAL-CR03-RAL-R61-2.01: will be joining us for lunch, at the team, so please, join us and continue the conversation there. Thank you. Thank you, everybody. 498 01:17:40.390 --> 01:17:45.680 STFC-RAL-CR03-RAL-R61-2.01: Well, so,