WEBVTT 1 00:00:00.000 --> 00:00:01.060 Mark Kushner: I'm hungry. 2 00:00:01.880 --> 00:00:16.940 Mark Kushner: We'll see you in the same room. Okay, welcome everybody. Today, it's, my great honor to introduce to you, Dr. Patrick Knapp from Pacific Fusion Corporation. Actually, so, 3 00:00:16.980 --> 00:00:32.719 Mark Kushner: Dr. Knapp got his, BS degree in electrical engineering at Syracuse University, and then where I first met Pat was we were both grads, grad school, lab mates, at Cornell, where we used to do experiments on the 1MAP COBRA facility at Cornell. 4 00:00:32.720 --> 00:00:37.230 Mark Kushner: And, Pat did a presentation on X-ray absorption spectroscopy. 5 00:00:37.230 --> 00:00:48.670 Mark Kushner: And then, we overlapped again when we worked together at Sandy and National Labs. Both worked on the Z machine, which is the world's largest, most powerful, pulse power device. For now. For now. For now. 6 00:00:48.780 --> 00:00:58.130 Mark Kushner: And Pat spearheaded a ton of, activity at Sandy. He spent 11 years there. He was the lead on over 100 experiments on the Z machine. 7 00:00:58.130 --> 00:01:19.519 Mark Kushner: He developed several new X-ray diagnostics for the facility, helped establish the Magnetized Liner Inertial Fusion Program, or MAGLIF, you've heard us talk about before. He was one of the key contributors to getting that going. And in doing that, he developed a method to measure fuel magnetization using secondary DT neutrons that get emitted from the fuel. 8 00:01:19.520 --> 00:01:28.710 Mark Kushner: He also developed several statistical analysis techniques, including using Bayesian inference, to analyze data and compare with simulations. 9 00:01:29.410 --> 00:01:54.089 Mark Kushner: Then after his 11 years at Sandia, he headed to Los Alamos for a short stint, and where he was leading a program… leading an effort to develop a pulse power program there to look at, to support not only the national ICF program, but also, stockpile stewardship applications in general. And then he, after Los Alamos, he went to Pacific Fusion, where he is today, and at Pacific, he is the, experimental physics lead 10 00:01:54.090 --> 00:02:01.279 Mark Kushner: Where he's developing platforms and analysis tools to support achieving facility gain. So, facility gain meaning 11 00:02:01.280 --> 00:02:13.819 Mark Kushner: more fusion energy coming out than you stored in all the capacitors to run the whole facility. So facility gain greater than 1, and fusion energy on the grid, using pulsar fusion, which he's going to be telling us about today. 12 00:02:13.920 --> 00:02:31.499 Mark Kushner: He's also… his responsibilities there, include de-risking novel target technologies for generating, validation data for the flash radiation magnetohydrodynamic MHD code, that was developed at University of Chicago. It's now at Rochester. A lot of people use it, specifically developing it further. 13 00:02:31.570 --> 00:02:47.750 Mark Kushner: He also leads the development of post-processing and synthetic data pipelines, which are going to help with design and optimization of the diagnostic suite for the forthcoming facility called DES, which stands for Demonstration System. That is the thing that's going to be bigger than Z. 14 00:02:47.800 --> 00:02:54.930 Mark Kushner: now being constructed. So, it's a very exciting time. So with that, let's, welcome our speaker. 15 00:02:56.150 --> 00:02:57.429 Mark Kushner: It's about this one. 16 00:02:57.430 --> 00:03:22.389 Mark Kushner: That's presented with the very famous Mimsy mug, and I know there's been a lot of confusion lately about whether it's still more rare than a Nobel Prize. I just looked it up, and I think there's been 230 Nobel victoriates, so assuming Mimsy's 17 years old, about 10 per year, we're probably getting close, but I think the joke still holds, that this is more rare. Alright, that's an honor. Thank you so much. 17 00:03:22.390 --> 00:03:26.740 Mark Kushner: There you go. Thank you. We should post the picture. 18 00:03:27.150 --> 00:03:29.959 Mark Kushner: Thank you. Thank you, I appreciate it. 19 00:03:29.970 --> 00:03:42.249 Mark Kushner: Thank you so much for having me here today. Hopefully folks online can hear me. Turn my microphone on, but if anyone wants to give a thumbs up, just so I can know for sure. But yeah, as Ryan said, I'm Pat Knapp. Thank you, guys. 20 00:03:42.250 --> 00:03:55.609 Mark Kushner: And I'm here presenting on behalf of the Pacific Fusion team, on pulsar-driven inertial fusion, which we believe is a practical and affordable approach to fusion energy. So I'll explain to you what that is and what we're doing. 21 00:03:55.610 --> 00:04:02.890 Mark Kushner: Here's an outline. I don't usually like outlines, but I have way too many slides today, so I think this is gonna help keep me on track and 22 00:04:02.890 --> 00:04:09.929 Mark Kushner: make sure we all know where we are. So, starting out, who are we? Who is Pacific Fusion? What are we doing? 23 00:04:10.120 --> 00:04:11.080 Mark Kushner: So… 24 00:04:11.080 --> 00:04:36.050 Mark Kushner: We're a private fusion company funded by venture capital. To date, we've raised more than $900 million, and we have almost 200 employees. I couldn't officially say 200 yet. We've got almost 200 employees. Additionally, approximately 40% of our workforce comes from the national labs. So, we're unique in that we're a fusion company where a significant amount of our workforce has experience in fusion at large-scale facilities. 25 00:04:36.050 --> 00:04:50.210 Mark Kushner: like you can only get at the national labs. So, the purpose of our company, we have two primary objectives in this phase of our company, is one, to achieve net facility gain. As Ryan said, that's more fusion energy out. 26 00:04:50.210 --> 00:05:14.950 Mark Kushner: then energy we put into the capacitors to charge our machine. So this would be a scientific first in the laboratory, and kind of the culmination of 70 years of fusion research. The other objective we have is to resolve hurdles to commercialization. So, our demonstration system, where we plan to achieve net facility gain, will be the kind of facility similar to the 27 00:05:14.950 --> 00:05:33.079 Mark Kushner: to the NIF or to Z, where you operate once a day, roughly. That is not suitable for energy generation. For energy generation, you need to operate at rep range, and so there's a lot of technological advancements, material science, engineering, that needs to be done to resolve those hurdles. 28 00:05:33.080 --> 00:05:38.130 Mark Kushner: So that we can actually do commercialization. So, these are our two objectives as a company. 29 00:05:39.650 --> 00:05:54.650 Mark Kushner: We're pursuing what we're calling pulsar-driven inertial fusion energy. This is basically inertial fusion energy using pulsed power as the source to drive our targets. I'll get into a lot more details about what that is, what that looks like, what the driver is. 30 00:05:54.650 --> 00:06:10.029 Mark Kushner: But this plot, I think, shows some really promising, results, and that gives us a lot of confidence that pulsar-driven IFE is a really valuable approach to take. So this plot shows, as a function of year. 31 00:06:10.030 --> 00:06:33.450 Mark Kushner: the, triple product achieved on various different facilities. So the triple product is density temperature times time, sometimes called P tau, where pressure is now density times temperature. And there's a threshold one needs to achieve that depends on temperature, above which you can get, a self… an igniting, self-propagating fusion reaction, and below that, you can't do that. 32 00:06:33.940 --> 00:06:45.350 Mark Kushner: It's different for different concepts, and in tokamak, magnetic confinement, we usually talk about Q, which is slightly different, but it's kind of the same general idea. 33 00:06:45.410 --> 00:07:04.729 Mark Kushner: What I want to point you to is this little cluster of purple points here, marked Z facility. So these are data points from the Magleif concept, and the really important point here is that the highest point measured on Magleif is the second highest P tau measured to date. 34 00:07:05.120 --> 00:07:11.029 Mark Kushner: on an inertial confinement fusion facility. The highest one was actually an igniting target on NIF. 35 00:07:11.800 --> 00:07:27.120 Mark Kushner: So, we've demonstrated, and I say we because I was part of this effort at the time, tremendous success in inertial fusion, using this pulsar-driven approach on Z over a very short period of time. 36 00:07:27.120 --> 00:07:45.090 Mark Kushner: laser-driven ICF stretches back, this says the 90s, it actually goes back further than that. Tokamaks have been studied for a very long time. So, one of the things that's really exciting about pulsar-driven fusion is that we've made so much progress in such a short amount of time. 37 00:07:45.090 --> 00:07:52.890 Mark Kushner: With very limited resources, that gives us a lot of confidence that there's a lot of headroom, there's a lot of unexplored territory, and it's really exciting. 38 00:07:53.290 --> 00:08:08.410 Mark Kushner: This is a cartoon of the magleif concept, so this is what's being studied on Z. Essentially, we have a cylinder, a can. In this case, on Z, it's usually beryllium. That can is filled with fusion fuel, and there's a window at the top. 39 00:08:08.410 --> 00:08:20.540 Mark Kushner: We pre-magnetize the target with a relatively low magnetic field, roughly 10 to 20 Tesla, and then we shine a bright laser into the target to preheat the gas. 40 00:08:20.540 --> 00:08:30.330 Mark Kushner: So the combination of this preheat and the magnetic field allows you to put some energy into the gas and then keep it there while you implode the target. 41 00:08:30.330 --> 00:08:40.939 Mark Kushner: So we implode the target over about… the current pulse that's used to drive the target is about 100 nanoseconds. It takes about 50 nanoseconds for the implosion to happen. 42 00:08:40.940 --> 00:09:05.579 Mark Kushner: Without that magnetic field, you'd leak all that energy away almost immediately through thermal conduction. So the magnetic field cuts off that process and allows us to keep that heat in while we compress it. And as we compress it, since we're not losing energy to thermal conduction, the energy in the internal energy in the gas, the density and temperature go up to eventually the point where they're sufficient to produce significant fusion reactions. 43 00:09:05.710 --> 00:09:16.649 Mark Kushner: So, at Sandia, this is primarily done with deuterium fuel, and it's really studying… the facility's not big enough to produce ignition, so we're studying the processes. 44 00:09:16.790 --> 00:09:25.479 Mark Kushner: the fundamental physics of what's going on, and how we think it might scale to higher yield. But this forms the basis of a lot of the work that Pacific Fusion is doing. 45 00:09:26.600 --> 00:09:51.589 Mark Kushner: Our recent publications, this QR code works, so if you want to check out our publications, it'll take you to one of our pages that has links to them. But this outlines our path. You know, we're developing Flash, as Ryan said. We're doing the science to validate our computational capabilities, to have confidence in our target designs. We're sharing with the community what our plans are. This is sort of an outline of what our facility 46 00:09:51.590 --> 00:09:54.360 Mark Kushner: is going to be, and how we're going to use it. 47 00:09:54.360 --> 00:10:04.890 Mark Kushner: And we are doing rigorous scientific work, theory, modeling, and experiments to underwrite what we're doing, and we're sharing that with the community. 48 00:10:06.660 --> 00:10:15.590 Mark Kushner: So Pacific Fusion right now has, 3 active, locations in the San Francisco Bay Area, so that's where the Pacific comes from. 49 00:10:15.590 --> 00:10:38.740 Mark Kushner: We started out in Fremont, which is down here, close to San Jose, at the south end of the San Francisco Bay. The Fremont Fusion Center is where our headquarters are, so this is where our executives' offices are, and it's also where our Pulsar engineering and pulse Power Design efforts are. This is where we're building our test module, demonstrating it, testing it, shaking it out. 50 00:10:38.740 --> 00:11:03.719 Mark Kushner: developing the core technology that's going to allow us to build this big machine. Our San Leandro build… oh, sorry, this is a picture of the Pulsar, partially obscured strategically, that, shows what we're building. Here is a picture of the high bay and our San Leandro Build Center. So this is, right here, just about 5 to 10 minutes south of the Oakland airport. 51 00:11:03.720 --> 00:11:28.720 Mark Kushner: to get to. I live in Albuquerque, and I travel there a lot, so it's a really great location for me. It works out nicely. But, so this is the high bay, the inside. We see an oil and a water tank. This is where we're developing all of our testing and prototyping capabilities. We're testing target fab and cryogenics there. We're testing our diagnostics, our positioner systems, alignment systems. 52 00:11:28.720 --> 00:11:41.990 Mark Kushner: We're building an x-ray lab right now so that we can characterize our x-ray diagnostics. So this is an enormous sort of 100,000 square foot warehouse where we're putting in all these different test capabilities. 53 00:11:41.990 --> 00:11:47.930 Mark Kushner: To develop and prototype the technologies that are going to go into our demonstration system. 54 00:11:47.960 --> 00:11:58.029 Mark Kushner: Target Fabrication is also there. Our experiments group, lives here, so this is where my team sits, along with the diagnostics group. 55 00:11:58.030 --> 00:12:11.019 Mark Kushner: And then our final facility in the Bay Area is the Livermore campus. It's a small office, near Livermore National Lab, where our target design, computation, and modeling and simulation groups sit. 56 00:12:11.040 --> 00:12:22.509 Mark Kushner: And they're… they're primarily responsible for developing Flash, validating it, developing material models, and then using Flash to design targets and design experiments. 57 00:12:24.640 --> 00:12:39.509 Mark Kushner: We are also expanding into New Mexico. So we've cited our demonstration system just south of Albuquerque in a little community called Mesa del Sol. This is a artist's rendering of our facility in Mesa del Sol. 58 00:12:39.510 --> 00:13:04.440 Mark Kushner: And here we have, the governor of New Mexico right here, with us when we announced our sighting. This is a handful of us. I'm back here, you can barely see me, at the site where we haven't broken ground yet. We also have this Los Lunas build center, so about 10 miles south of Albuquerque is this town, Los Lunas, where this enormous warehouse was empty. It's about a 250,000 square 59 00:13:04.440 --> 00:13:05.690 Mark Kushner: wonderful warehouse. 60 00:13:05.690 --> 00:13:10.399 Mark Kushner: So we leased that. It's right off of I-25, you can't miss it. 61 00:13:10.480 --> 00:13:21.200 Mark Kushner: And this is a picture of the inside. Sorry, we're losing… This is… we're some equipment being installed in there. And basically, while we're waiting for this building to be constructed. 62 00:13:21.360 --> 00:13:34.200 Mark Kushner: This facility is where we're going to be mass producing all of our components and all of our equipment that will then move into this facility when it's constructed. We're planning to break ground sometime this year, and it'll take about 18 months to build the building. 63 00:13:36.820 --> 00:13:43.510 Mark Kushner: Just so everybody knows, sometimes there's some questions about where New Mexico actually is. This is New Mexico. 64 00:13:43.800 --> 00:14:05.960 Mark Kushner: Albuquerque's right about in the center of New Mexico, and then our Mesa del Sol facility is just on the southern end, near the Isleta Pueblo, and the Isleta Amphitheater, if you want to go see an outdoor concert. It's right there. So we'll be, we'll be, hopefully not the loudest and most obnoxious neighbor in this, in this area. And so yeah, that's where we are. 65 00:14:08.290 --> 00:14:09.250 Mark Kushner: Okay. 66 00:14:09.250 --> 00:14:30.469 Mark Kushner: So, pulsar-driven fusion energy, pulsar IFE, leverages magnetic fields to relax the ignition requirements. So, I talked about magleif, what magleif was, how we needed the magnetic field to keep that energy in as we implode the target. It's also critical at stagnation when ignition is ideally happening, and that's because 67 00:14:30.710 --> 00:14:38.990 Mark Kushner: the targets don't have as high aerial density of fuel as something like a NIF target has. 68 00:14:38.990 --> 00:14:50.880 Mark Kushner: Because cylindrical convergence is not as favorable as spherical convergence to sort of accumulate aerial density the way a spherical implosion does. So we leverage the magnetic field to help relax that requirement. 69 00:14:50.880 --> 00:15:08.150 Mark Kushner: The way that works is when you have DT fusion happening, you produce a 14 MeV neutron and a 3.5 MeV alpha particle. That alpha particle, ideally what you want to do is trap that in your fuel so it deposits its energy in the fuel, heating it up more. 70 00:15:08.270 --> 00:15:25.970 Mark Kushner: So in regular inertial confinement, you need aerial density to do that. You need a lot of mass so that the alpha particles will slow down and deposit all their energy before they escape. So they, basically, if you think about them slowing down as they're running through, you want them to deposit all their energy and stop before they get out. 71 00:15:26.430 --> 00:15:30.389 Mark Kushner: In our case, with the magnetic field. 72 00:15:30.740 --> 00:15:41.120 Mark Kushner: If you have a strong enough magnetic field, which is parameterized by this parameter BR, it's the magnetic field times the radius, so it's analogous to aerial density, which is density times radius. 73 00:15:41.300 --> 00:15:55.000 Mark Kushner: You want this to be high enough so that the alpha particles' trajectories are curved by the magnetic field. You want them to be trapped, so you want their gyro orbit to be smaller than the size of your plasma, so that they don't escape. 74 00:15:55.110 --> 00:16:03.199 Mark Kushner: And that way, they can have more time before they leave your plasma to actually deposit the energy, because they just spend time gyrating energy fuel. 75 00:16:03.670 --> 00:16:04.750 Mark Kushner: So… 76 00:16:04.880 --> 00:16:16.000 Mark Kushner: the BR then takes the place of aerial density as your confinement parameter. That's a really high premium. And so this plot here is showing, as a function of BR, 77 00:16:16.110 --> 00:16:19.380 Mark Kushner: These and, aerial density of your fuel 78 00:16:19.540 --> 00:16:31.199 Mark Kushner: where you need to be up and to the right of these curves in order to have self-heating and have a hope at ignition. And so these curves are for different temperatures, so as you get to higher temperature. 79 00:16:31.200 --> 00:16:46.049 Mark Kushner: this curve comes down and down and down. And as you get to higher BR, you get to access this part of the curve where it drops steeply. And this is where you really benefit, right? So you relax your requirement for aerial density. 80 00:16:46.050 --> 00:16:52.459 Mark Kushner: And at high enough temperatures now, you can ignite without having to have nearly as much aerobensity. 81 00:16:52.900 --> 00:16:54.870 Mark Kushner: So this has implications. 82 00:16:54.870 --> 00:17:18.849 Mark Kushner: for how much driver energy you need, how much driver power you need in particular, and this allows us to use pulse power, which is comparatively lower power than a laser. The pulses are longer than typical laser fusion, and so therefore the powers are lower. So this allows us to use these slower implosions, lower power drivers, and still achieve ignition conditions. 83 00:17:21.119 --> 00:17:28.390 Mark Kushner: As I said, Magleif has also made steady progress over a very short period of time. This plot here shows work 84 00:17:28.390 --> 00:17:45.400 Mark Kushner: Using a statistical method that I helped develop to infer, the proximity to ignition from these implosions at these magleif experiments at Santia. The peak pressures that were inferred are above 2 gigabar, and temperatures above 3 kilovolts. 85 00:17:45.400 --> 00:18:04.579 Mark Kushner: And since I've left Sandia, they've done an experiment that, has gotten a value of this parameter chi, of 0.2. It was about .08 when I made this plot. So these two plot points here were the highest, and now there's another point, that's almost… that's more than double. 86 00:18:04.580 --> 00:18:17.790 Mark Kushner: And chi is essentially… it's analogous to p-tao, but basically it's normalized, so chi of 1 means you're at the self-heating threshold. Chi above 1 means you're self-heating, and below 1 means you're not. 87 00:18:18.100 --> 00:18:40.109 Mark Kushner: So CHI of 0.2 is actually quite remarkable for a machine that was never designed to get ignition-relevant conditions in the first place. Z is far too small to do that with not enough energy. Yet, the fact that over the course of less than 10 years, Sandia was able to make this kind of progress, improve the implosion quality, improve the initial conditions. 88 00:18:40.110 --> 00:18:47.680 Mark Kushner: Higher magnetic field, higher laser energy, lower mix. To get these conditions is really, really exciting. 89 00:18:48.060 --> 00:19:03.979 Mark Kushner: And we have a validated and verified method to actually understand exactly where we are in this space of CHI, basically how close you are to ignition. And so we're going to leverage this technique on our facility. 90 00:19:06.570 --> 00:19:07.600 Mark Kushner: So… 91 00:19:07.750 --> 00:19:21.820 Mark Kushner: I talked about, pulsar-driven fusion energy as an affordable and scalable approach. One of the reasons why it's affordable… so, affordable is relative. These facilities are quite expensive. But… 92 00:19:22.190 --> 00:19:40.689 Mark Kushner: Pulsar-driven fusion, if you look at the theory of scaling and compare it to laser ICF, has a much better, much more favorable scaling with driver energy than laser-driven ICF does. And the reason is because when you build a larger pulse power driver. 93 00:19:40.710 --> 00:19:44.859 Mark Kushner: You are not, generally speaking, lengthening the pulse. 94 00:19:44.860 --> 00:20:06.969 Mark Kushner: you're keeping the pulse length the same, so the implosion time is the same for your target, you're just increasing the energy and the peak current that you're supplying. Whereas when you scale a laser-driven target, you're generally doing what's called, Euler scaling, and so the target gets bigger, the laser energy increases, primarily by lengthening the pulse. So the implosions, 95 00:20:06.970 --> 00:20:18.859 Mark Kushner: The scaling is such that the implosion velocity stays constant, which limits your ability to increase the pressure, and therefore your fusion energy gain that you would get. 96 00:20:18.860 --> 00:20:35.319 Mark Kushner: So by applying this MHD scaling developed by Paul Schmidt and Daniel Ruiz, that conserves implosion time, your pressure goes up as your current increases, whereas with a laser, the pressure, stays constant. 97 00:20:35.320 --> 00:20:43.240 Mark Kushner: This allows the yield scaling to increase tremendously with your input energy, and so we get almost, 98 00:20:43.240 --> 00:20:57.429 Mark Kushner: your yield scales almost as the cube of the driver energy. The liner energy is essentially the same scaling as the driver energy, whereas for a laser, it's to the 1.5 power. So the scaling is extremely powerful. 99 00:20:57.430 --> 00:21:06.890 Mark Kushner: And this allows us now to build a driver at a reasonable and affordable scale with technology that we know how to build today. 100 00:21:07.100 --> 00:21:14.390 Mark Kushner: That leverages this scaling to get us over that hump and get us to ignition and, facility gain. 101 00:21:15.560 --> 00:21:32.760 Mark Kushner: So, as an illustration of that, I compare, sort of, the energy flow through a laser-driven system, laser indirect drive in this case, to a pulse power-driven system. And so, NIF stores about 400 megajoules of energy in the capacitors, and converts that to about 2 megajoules of laser energy. 102 00:21:32.760 --> 00:21:41.440 Mark Kushner: On our machine, we're going to store about 80 megajoules of energy, and that converts to 8 megajoules delivered to the target, and 103 00:21:41.530 --> 00:22:03.659 Mark Kushner: Sorry, no, more than that. But essentially, we end up, through the efficiency of our driver and the fact that we're able to store and couple 8 megajoules into the… into the liner, we can couple then a tremendous amount of energy into the fuel, which allows us to have larger fuel inventories and therefore larger yields. 104 00:22:03.660 --> 00:22:11.060 Mark Kushner: The laser, we have to convert that energy to laser light, which is inherently an inefficient process. 105 00:22:11.060 --> 00:22:26.190 Mark Kushner: absorb it into the capsule through either X-ray conversion or through direct drive, and then do work on the fuel, and then we're left with a sort of a very small amount of energy, and therefore, very little energy, very little fuel to work with. 106 00:22:26.190 --> 00:22:50.700 Mark Kushner: If you scale this up to proposed laser architectures that are much more efficient, and actually couple 10 megajoules of energy to the laser, we see that we're still, in our situation, able to couple 3 times the energy to our fuel, which allows us to have, again, larger fuel inventories, and then facility gain at what is actually comparable fuel gain to what you can get achieved with a laser-based system. 107 00:22:50.800 --> 00:22:58.639 Mark Kushner: Where fuel gain is the energy that you actually couple to the fuel rela… the energy infusion out relative to the energy coupled to the fuel. 108 00:23:02.000 --> 00:23:19.070 Mark Kushner: So, just to elaborate that on a little bit, I'll elaborate on that a little bit. We see some metrics here, 330 megajoules of energy stored in the capacitors, you have your 2 megajoules of laser light, this is on NIF, you have 0.2 milligrams of fusion fuel. 109 00:23:19.200 --> 00:23:36.329 Mark Kushner: And the yield on… this is a little out of date, not their best shot. I think they're up around 8 megajoules now. You get 5 megajoules of yield, so your target gain, is about 2.4. Your fuel gain is pretty large, 600, but your facility gain is very low. 110 00:23:36.610 --> 00:23:50.820 Mark Kushner: Now, what's interesting here, if you compare calculations with different liner materials on our facility, we store less energy, we couple more energy to the target, we have many more times fuel. 111 00:23:50.820 --> 00:23:55.919 Mark Kushner: So we have 10 milligrams of fuel in these targets, which is a lot, a lot of payload. 112 00:23:55.920 --> 00:24:14.309 Mark Kushner: But when you work the flow down, you see that the fuel gains are similar, the burn fractions are similar, yet we have facility gain. And so what this means is we can access facility gain, because we have these efficient drivers and a lot of energy to work with, at conditions that are comparable to what's already been achieved. 113 00:24:14.540 --> 00:24:26.560 Mark Kushner: So we don't have to sort of rely on things that we don't understand, conditions that haven't been accessed for our system to work. We know that it can be done. 114 00:24:28.320 --> 00:24:29.280 Mark Kushner: Okay. 115 00:24:29.380 --> 00:24:33.609 Mark Kushner: Alright, so now I'll talk a little bit about our simulation and target design efforts. 116 00:24:33.610 --> 00:24:55.879 Mark Kushner: So, as Ryan said, we're using the flash code, we're co-developing it. We have our own branch that we're able to develop on. Our next slide, I kind of details some of the things that we've developed. So, it's been widely used for over 20 years in the astrophysics and laboratory astrophysics community. More recently, it's been used for laser-driven HED and laser-driven infusion experiments. 117 00:24:55.880 --> 00:25:07.580 Mark Kushner: And then now, in part thanks to us, although others started using Flash before us for Z pinches, we've helped to really advance Flash's capabilities to model Z-pinches. 118 00:25:07.760 --> 00:25:09.260 Mark Kushner: perfusion efforts. 119 00:25:10.140 --> 00:25:12.950 Mark Kushner: So… Do you have… 120 00:25:16.260 --> 00:25:17.170 Mark Kushner: Sorry. 121 00:25:18.520 --> 00:25:20.430 Mark Kushner: Trying to get my video, so… 122 00:25:24.130 --> 00:25:24.990 Mark Kushner: There you go. 123 00:25:26.830 --> 00:25:42.310 Mark Kushner: Okay, sorry about that. So we've added a thermonuclear burn with alpha diffusion package, divergence-free implicit resistive MHC, we have a Python-based common modeling framework that really dramatically simplifies 124 00:25:42.310 --> 00:25:49.550 Mark Kushner: the use of Flash as a design tool, which is really, really helpful, especially for someone like me who's an experimentalist. 125 00:25:49.550 --> 00:26:14.419 Mark Kushner: This is a calculation of a magleth implosion. You see the current going up, eventually you'll see the laser heating happen right there, and then the implosion. So we've added also some interface capturing capabilities, magnetized anisotropic… anisotropic, plasma viscosity, radiation transport, really a whole host of, 126 00:26:14.420 --> 00:26:22.410 Mark Kushner: new capabilities, to Flash, which has really, improved its ability to agree with available data quite a bit. 127 00:26:22.730 --> 00:26:47.720 Mark Kushner: One of the ways we tested it is by performing this so-called Ruiz scaling, where, again, this is the similarity scaling theory developed by Daniel Ruiz and Paul Schmidt at Sandia National Labs. And essentially, it dictates as you… you start from an implosion, a target designed for Z, and then you turn up the current, and this theory dictates how the geometric 128 00:26:47.720 --> 00:27:10.489 Mark Kushner: and initial conditions like magnetic field, laser preheat, and fuel density need to change as you increase the current. And so, what happens is, as you increase the current, the target parameters change, but the current pulse, stays the same, except for a higher amplitude, and the implosion trajectory stays self-similar. 129 00:27:10.730 --> 00:27:21.769 Mark Kushner: And by doing this, you also preserve a lot of the physics of the implosion. You can preserve the radiation loss regime, the thermal conduction loss regime, and so you can say. 130 00:27:21.980 --> 00:27:33.509 Mark Kushner: this is what's called a conservative path in scaling, where you're saying, I'm not going to be more aggressive in any of the physics. I'm not going to bank on anything new happening, or 131 00:27:33.510 --> 00:27:44.679 Mark Kushner: getting a free lunch by some physics mechanism turning off, or get it scaling better than I expected, I'm gonna keep everything, to the extent possible, in the same regime. 132 00:27:44.720 --> 00:28:06.929 Mark Kushner: I'm just going to increase the current, and we'll get more fusion energy out. So this… this prescription tells you how the energy should… the fusion energy yield should scale, and these points show our calculations compared to Hydra and the theory. And so we agree really well, both with alpha on and alpha off, with how that scaling goes. And this gives us a lot of confidence 133 00:28:06.930 --> 00:28:16.530 Mark Kushner: that Flash is able to model these properties, not just at one point where we have experimental data, but across a wide range of conditions. 134 00:28:17.330 --> 00:28:35.219 Mark Kushner: Additionally, one of our interns this summer, your fellow student, Adam Badel, implemented the ability to use a screw-pinch boundary condition and study that in flash. So basically what this means is, typically we drive implosions 135 00:28:35.220 --> 00:28:48.649 Mark Kushner: with a return current can that just has straight posts, or maybe is a solid can, and that produces a purely azimuthal magnetic field around the target that drives the implosion. If you take those posts in your return current can and you twist them. 136 00:28:49.120 --> 00:29:08.659 Mark Kushner: So now you're adding a, azimuthyl component to the return current structure. You're adding a Z-directed magnetic field to the drive field. So by how aggressively you twist that, you can think in a limit, right, it turns into a solenoid, and you have a purely Z field being driven by your return current. 137 00:29:08.800 --> 00:29:15.510 Mark Kushner: But to the degree to which you twist that, you can change the ratio of the theta to the Z field. 138 00:29:15.510 --> 00:29:39.679 Mark Kushner: Which has a lot of interesting implications. It's been shown, with some theoretical and some experimental work that you can change the stability of the implosion by doing this. But it also has implications for how you can get that magnetic field you need to insulate your target into the liner. So that's what's being shown here. Adam did a calculation comparing to a 3D calculation. 139 00:29:39.680 --> 00:29:40.859 Mark Kushner: from a, 140 00:29:40.860 --> 00:29:53.260 Mark Kushner: code that Chris Jennings has at Sandia called Kraken. I believe Gabe Shipley did these, calculations. But basically showing that with the, drive magnetic field, the, the, 141 00:29:53.260 --> 00:30:09.900 Mark Kushner: sorry, screw pinch drive in 1D, Flash can get very close agreement to what was demonstrated in 3D. It's not perfect, but a lot of that is because we're comparing 1D and 3D calculations. But it shows here that you have 142 00:30:09.900 --> 00:30:12.210 Mark Kushner: The… 143 00:30:12.220 --> 00:30:25.230 Mark Kushner: Axial magnetic field in the case where you just pre-magnetize the target, and so you can see it's uniform everywhere initially, and then when the laser heating happens, there's all these interesting dynamics with how the magnetic field gets compressed up and redistributes. 144 00:30:25.230 --> 00:30:47.959 Mark Kushner: In this case, you have initially no axial magnetic field, and it's supplied by the boundary condition. Eventually, it leaks through. It doesn't get into, like, the center of the fuel, but there's a buffer layer that allows the preheat to be insulated, and maintain your heat through the implosion to get comparable performance. And you can do this now without the expensive magnetic field coils. 145 00:30:47.960 --> 00:30:55.290 Mark Kushner: that get destroyed on every experiment. So this is really attractive for fusion energy, because you're eliminating one of these complex subsystems. 146 00:30:56.100 --> 00:31:05.909 Mark Kushner: We're also doing kind of all the usual things you'd expect, for an ICF company to do with Flash, where we're looking at how different perturbations 147 00:31:05.910 --> 00:31:21.639 Mark Kushner: on the inside and the outside of the liner, on the fuel scale, and how they degrade your yield. So we have to do these studies. This will help us understand what our requirements are when we go to build these targets, how smooth different surfaces need to be. 148 00:31:21.640 --> 00:31:34.829 Mark Kushner: And then, we're also using Flash to design, high-yield fusion targets that are optimized for our driver. So we talked about that, scaling, study, and what happens when you take 149 00:31:34.830 --> 00:31:40.509 Mark Kushner: a Z target, and you scale it up to 60 megamps, the target has to get taller. 150 00:31:40.510 --> 00:31:55.229 Mark Kushner: This is just the way the scaling, math works out. The target gets taller. You know much about pulse power, longer targets are bad. They're higher inductance. That means you need more, for the same current getting to the target, you need more voltage. 151 00:31:55.730 --> 00:32:09.169 Mark Kushner: Higher voltages can be challenging for material physics aspects, where you have insulators in your system that might break down, so it gets harder and harder to couple higher voltage and higher current to a target. So we wanted to take that 152 00:32:09.380 --> 00:32:34.199 Mark Kushner: theoretical scaling as a starting point, where the target gets taller, and then work from there and say, how do we then redesign this so that it's more, a better match to our driver? So this is the Z current pulse compared to the demonstration system current pulse. It's also a little bit longer, so if you follow the scaling theory for adding time to your implosion, it would get even longer, so this hurts you even more. 153 00:32:34.970 --> 00:32:52.019 Mark Kushner: So, Paul Schmidt and company did the hard work of saying, you know, this is what the scaling theory says, but our driver doesn't like it, so let's go down to a shorter target. Let's optimize this target, we'll do the iterations, go through, and this is what you get. Essentially, we have 154 00:32:52.020 --> 00:33:04.290 Mark Kushner: An aluminum liner in this case, that's fairly fat, aspect ratio 4.2, 7 millimeter tall, and then we have a channel for the preheat laser to come in. 155 00:33:04.290 --> 00:33:29.250 Mark Kushner: And preheat the target. So I'm going to show you a movie of now an optimized, slightly different target. We'll let it repeat. So this one's slightly higher aspect ratio, but you can see the dynamics. There's some really interesting things that happen when you approach the design from this perspective. Laser preheat happens quite late in the implosion, but because the liner is a little bit thinner in this case. 156 00:33:29.250 --> 00:33:38.790 Mark Kushner: it moves faster. This target only uses about 20 kilojoules per centimeter of preheat energy, and only 5 Tesla of initial magnetic field. 157 00:33:38.790 --> 00:33:49.069 Mark Kushner: And it produces about, almost 600 megajoules of fusion energy in 2D calculations, for a facility gain factor of about 7. 158 00:33:50.840 --> 00:33:51.830 Mark Kushner: Okay. 159 00:33:51.990 --> 00:34:11.720 Mark Kushner: All right, so what is our facility? I've talked a lot about it, but I haven't shown you what it actually is. So this is a architectural design of what our actual facility looks like. We have our pulsar modules here, arrayed around the whole facility. There's 128 of them. These are the IMGs, impedance match marks generators. 160 00:34:11.719 --> 00:34:30.130 Mark Kushner: They couple through pulsar tubes into the water section here, into our tri-plate transmission lines, which then connect through a vacuum insulator stack to the vacuum section. The vacuum section is where the magic happens, right? Our target is going to be right at the center here. 161 00:34:30.130 --> 00:34:38.370 Mark Kushner: Connected, from the insulator stack through magnetically insulated transmission lines into, into the target. 162 00:34:38.960 --> 00:34:56.950 Mark Kushner: So, an important feature to note here is that our water section is required down here for electrical properties. It's the insulator in these transmission lines, deionized water. But we extended the water over top in this dome because it turns out it's a really good shield for neutrons. 163 00:34:56.949 --> 00:35:01.990 Mark Kushner: So, our primary radiological shield for this device is the water. 164 00:35:01.990 --> 00:35:15.890 Mark Kushner: We don't have 8-foot thick concrete walls designed in our building. We aren't doing it that way. We're using local shielding around our device by nature of this water as our radiological shield. 165 00:35:18.510 --> 00:35:41.909 Mark Kushner: So, a little bit about the impedance match marks generator. It's a modular pulse power system. It's made out of what are called bricks. This is an example of a brick right here. A brick is made of two capacitors and a switch connected together. This is sort of the same fundamental element that's used to build the maze generator in Professor McBride's lab, but it's arranged in a slightly different architecture. 166 00:35:41.910 --> 00:35:48.969 Mark Kushner: So we… we take the bricks, we put them together, and then we array them around in a circle on the anode. 167 00:35:48.970 --> 00:35:52.689 Mark Kushner: Of a… of this driver in what we call a stage. 168 00:35:52.690 --> 00:35:55.800 Mark Kushner: So, there's about 50,000 bricks total. 169 00:35:55.800 --> 00:36:19.899 Mark Kushner: There's 156 modules, 32 stages in a module, so we stack them up along there. They all discharge their energy through this anode-cathode gap that's tailored so that they go sort of along in a chain, and the impedance, as the voltage increases when a new stage discharges, the impedance changes so that it's matched the whole way across. 170 00:36:20.240 --> 00:36:39.789 Mark Kushner: So again, we arrange these all around. The pulsar tubes have different pulse lengths, so they have to be co-timed properly. And then all of that is sort of coupled into the target, which is a centimeter scale target system, which eventually is hopefully compatible with, Hertz rep rate. 171 00:36:41.860 --> 00:36:50.140 Mark Kushner: This is a little bit of a look at the inside of our machine. Basically, we're designing this facility, to operate 172 00:36:50.140 --> 00:36:58.940 Mark Kushner: based on a lot of important lessons learned from NIF and Z, we want it to be operated remotely, because when we're generating 100-plus megajoule yields. 173 00:36:58.940 --> 00:37:16.400 Mark Kushner: humans will not be able to enter the center section. It will be… all of the stainless steel components will be activated and highly radioactive, so it won't be a safe place to enter. So all of our diagnostics will be on manipulators that are inserted, can be removed remotely. 174 00:37:16.420 --> 00:37:21.829 Mark Kushner: Craned away to cool down if they need to be operated on for maintenance. 175 00:37:21.830 --> 00:37:46.650 Mark Kushner: But basically, the idea is we take our target, which is in what we call a cassette. It's a hermetically sealed, pre-built, vacuum chamber that gets inserted. That gets built offline and metrologized. There's some diagnostic components that'll actually be inside the cassette. We'll need to know exactly where they are relative to the target, so that our diagno… the rest of the 176 00:37:46.650 --> 00:37:56.540 Mark Kushner: diagnostic system can point accurately where it needs to point. That gets robotically inserted from the bottom up into the mitals. 177 00:37:56.540 --> 00:38:13.710 Mark Kushner: Everything gets precision aligned using a laser alignment system to the cassette. The cassette has windows for our diagnostics, so we do the shot, and then the cassette acts as actually a partial containment system for the blast. 178 00:38:13.710 --> 00:38:36.559 Mark Kushner: from the 8 megajoules of energy coupled to the target, and then the hopefully copious amounts of fusion energy that get released by the target. So this was designed to be done at sort of a once-a-day, shot rate at high yield. We can hopefully do more shots if we lower the energy coupled, because there'll be less maintenance, less things that need to be changed. 179 00:38:36.790 --> 00:38:39.319 Mark Kushner: But this is our concept of operations. 180 00:38:40.540 --> 00:38:48.090 Mark Kushner: Here is a cartoon of our diagnostic suite. We have a whole array of nuclear, X-ray, and optical diagnostics. 181 00:38:48.090 --> 00:39:12.580 Mark Kushner: That get, piped in in a variety of different ways. The next slide, shows actually what they are. So our nuclear diagnostics, we have a magnetic recoil spectrometer, activation samples, neutron imaging, we have an array of neutron time-of-flight detectors, several different gamma reaction history diagnostics, a proton recoil telescope, and then another activation array. 182 00:39:12.580 --> 00:39:31.749 Mark Kushner: I always forget which is which. One of them is geared towards looking at symmetry around the target, and the other is at quantifying the absolute yield. We have two, main flavors of optical diagnostic. Farad is a probe, basically an imaging laser that gets projected through the target to look at low-density plasmas. 183 00:39:31.750 --> 00:39:48.599 Mark Kushner: things like Faraday rotation and interferometry. And then VISOR and PDV primarily are going to be used to measure the current at the target. So basically, I talked about that return can. We have the current flowing through the target, and it has to go back out through the return current structure to get to ground. 184 00:39:48.780 --> 00:40:09.369 Mark Kushner: that return current structure experiences a J cross B force outward, the same way the liner in the inside experiences a J cross B force inward. So if we can measure that exploding velocity of the return current can, we can measure how much current was coupled to our target. And so our velocimetry diagnostics are primarily geared towards that, although there are some other use cases. 185 00:40:09.390 --> 00:40:31.790 Mark Kushner: And then X-rays, we have self-emission, time and space-resolved self-emission imaging, we have x-ray radiography, X-ray spectroscopy, burn history, and then an additional, Libra is our concept for doing X-ray… high-resolution time-resolved X-ray imaging at high yield. Galaxy is a pinhole camera. 186 00:40:31.790 --> 00:40:42.939 Mark Kushner: Pinhole cameras won't work at high yield, because it basically puts your detector in the direct line of sight of this onslaught of neutrons, and that won't work at 100 megajoule yields, so we need a different concept to do that. 187 00:40:44.260 --> 00:41:08.750 Mark Kushner: We're developing a comprehensive, comprehensive models of all of our diagnostics so that we can produce synthetic data. This is work that Patrick Adrian did on our X-ray, or sorry, neutron post-processing. So basically, this gives you a flavor of the workflow. Essentially, the same thing applies to X-rays. We do it the same way, but there's a few more steps for the neutronix modeling. So we start with either flash calculations. 188 00:41:08.750 --> 00:41:28.529 Mark Kushner: or this model called VERA, which I'll talk about in a minute. It's basically a reduced model of the implosion that runs really fast. And so then we trace the neutrons through the plasma, as well as charged particles of interest, so alpha particles, or knock-on, you know, deuterons and tritons that have been hit by a neutron and accelerated to high energies. 189 00:41:28.530 --> 00:41:42.940 Mark Kushner: We track them through the plasma, and this allows us to basically track the nuclear reactions going on in the plasma, downscatter of neutrons, a generation of tertiary, secondary neutrons, and then create, basically, a map 190 00:41:42.940 --> 00:41:47.580 Mark Kushner: Of all the particles, being emitted from the plasma. 191 00:41:47.820 --> 00:42:11.590 Mark Kushner: Then we can take that sort of two-dimensional flux surface of particles and post-process it in a variety of different ways to get, say, the neutron energy spectrum that the magnetic recoil spectrometer might see, or what the neutron image will look like. Or, in this case, this is the reconstruction of the sky map showing the uniformity of emission that our activation detectors see. 192 00:42:11.660 --> 00:42:31.700 Mark Kushner: So we have these full pipelines that we're developing for each of our diagnostics, so we can do… say, we can design a target, make a change to it in Flash, run the simulation, and make sure that our diagnostics are actually able to see what that change showed. It's one thing to look at a calculation and say the temperature changed, or the pressure changed. 193 00:42:31.700 --> 00:42:34.920 Mark Kushner: But actually being able to measure that is a very different thing. 194 00:42:34.920 --> 00:42:41.890 Mark Kushner: So we're trying to do that work up front, so that when we start operating our facility, we're ready to go with analyzing our data. 195 00:42:42.400 --> 00:43:07.200 Mark Kushner: We're also, speaking of high yield, designing many of our diagnostics to operate and thrive in the high yield environment. So this shows some OpenMC modeling we're doing to try to look at what the particle fluences are in different parts of our diagnostics. So this is our magnetic recoil spectrometer. Again, we have neutrons coming through, there's collimators. 196 00:43:07.210 --> 00:43:21.030 Mark Kushner: and openings that help clean things up. There's a beam dump at the back, and then here is where the magnet would be to bend the recoil protons or deuterons up to the detector. Our detector in this case is electronic. 197 00:43:21.790 --> 00:43:41.749 Mark Kushner: we… our electronic detector is going to be very sensitive to background x-rays and gammas, so we want to move it off to a place where it's very well shielded. All this gray area is water, so this is all in our water section, which makes it an ideal shield. We just need to get it far enough away from that primary beam so that the scattering environment, 198 00:43:41.750 --> 00:44:00.460 Mark Kushner: show, you know, is reduced sufficiently, and so this, both for, gammas as well as neutrons, shows that our, detector is in a safe place. So we're… we're looking at all this stuff, we're tweaking the designs, we're putting collimation and shielding in in the right places, so that our, diagnostics can operate effectively. 199 00:44:00.770 --> 00:44:01.710 Mark Kushner: Okay. 200 00:44:01.990 --> 00:44:25.029 Mark Kushner: Last section, hopefully I can get through this quickly. So, this is about some of the models we're developing to do our data analysis. So, Calvera is a simplified model of implosions. It's essentially an extension of Professor McBride's semi-analytic magleif model. If you haven't had a chance to go through this paper, I highly recommend it. 201 00:44:25.030 --> 00:44:48.219 Mark Kushner: It's long, but it's very complete. And so it gives you a really, really amazing overview of all the physics of Magleif, as well as, if you can follow the recipe, a way to build your own fast ODE model that solves most of the important physics. So we took that, we rebuilt it, instead of in MATLAB or straight Python, we built it in this language called JAX. 202 00:44:48.220 --> 00:44:53.630 Mark Kushner: Which is a Python-based language developed by Google that allows for compilation, so it's very fast. 203 00:44:53.630 --> 00:45:13.579 Mark Kushner: And it's also differentiable. So what that means is every single mathematical operator in your code has a derivative defined for it. So when you build a model and compile it, you're not only saying, if I give you these inputs, what's my output? You're also saying, what's the derivative of my output with respect to all these inputs? 204 00:45:13.680 --> 00:45:17.630 Mark Kushner: So, essentially, for free now, you get derivatives. 205 00:45:17.630 --> 00:45:35.069 Mark Kushner: Which lets you do a lot of really cool things, like sensitivity studies are essentially free. At every point, you know, you take a sampling over some grid of your outputs versus a variety of inputs, you get the sensitivity of your output to those inputs at each of those locations for free. 206 00:45:35.570 --> 00:45:46.010 Mark Kushner: It also means, you can use this as a model now in complex optimizations that require gradients to do the optimizations. 207 00:45:46.010 --> 00:46:10.340 Mark Kushner: or Markov chain Monte Carlo algorithms, like Hamiltonian Monte Carlo, that requires computing of the gradients of your probability function in order to find the next step. And it's a much more efficient way of doing Bayesian inference than standard Metropolis Hastings, and it also opens up a host of other sort of approximate inference methods that rely on optimization, which, again, is accelerated by having the gradients. 208 00:46:10.460 --> 00:46:26.449 Mark Kushner: So it's a really cool tool. I encourage you to look into JAX. There's some sharp edges and some difficulty in learning it, but it's really exciting, and I think the, sort of, sky's the limit for how to, what this can do for us as a scientist community. 209 00:46:26.990 --> 00:46:29.180 Mark Kushner: Okay. 210 00:46:29.290 --> 00:46:47.200 Mark Kushner: So, in addition to rewriting the code in JAX, we also made some changes to it. We put in some different models. I can't tell, is that moving? Yeah, it's moving. Okay. We put in some different models, to try to improve performance and improve fidelity. We also made it quasi-3D. 211 00:46:47.280 --> 00:47:06.940 Mark Kushner: So this is essentially a simulation of a stack of pizzas, you can imagine. So there's a bunch of pizzas stacked on top of each other, and then each slice of the pizza is an azimuthal wedge, right? So you've got a bunch of 1D problems that are now coupled together by the pressure in the fuel. 212 00:47:07.250 --> 00:47:32.229 Mark Kushner: And so this is an approximation, but, but it actually allows you to do a lot really quickly with understanding implosion symmetry and how you're sensitive to, asymmetries in the implosion. So this is a, example where we put a helical perturbation onto this, this implosion and simulated it in this quasi-3D. You get the helical 213 00:47:32.230 --> 00:47:33.720 Mark Kushner: Kink, and it bounces. 214 00:47:33.720 --> 00:47:42.110 Mark Kushner: This one took about 10 seconds to run, so we can do low-fidelity 3D calculations in seconds with this capability. 215 00:47:45.720 --> 00:47:56.500 Mark Kushner: We've also validated it against, Flash, so we did that same Ruiz scaling. We took the same… the exact same, input parameters and showed that we agree with Flash. 216 00:47:56.500 --> 00:48:21.489 Mark Kushner: And again, in this case, not just yield, but also pressure, temperature, bang time, sorry, these are small, a little hard to read, ROAR and, BR. So, we're looking at, at stagnation during burn, are we getting, sort of, the right properties of the fuel? And then you can also look at the implosions. The white lines are my calvera calculation, and the pseudocolor 217 00:48:21.490 --> 00:48:32.010 Mark Kushner: map is flashed behind it. So we can reproduce this scaling really well, which gives us confidence that even though we know this is a reduced physics model, it's still a very useful one. 218 00:48:32.910 --> 00:48:46.539 Mark Kushner: So what do we want to do with that? Well, this is an example now. In order to analyze our data that we get from stagnation, we can run this in a different way. Instead of running it from time equals zero, let's initialize the implosion in flight. 219 00:48:46.560 --> 00:49:05.800 Mark Kushner: Let's initialize the liner, velocity, and the fuel right here, basically where you're at peak velocity. So you're only modeling now a very small part of the implosion, but it's the part where all the interesting stuff happens, and all the things you care about measuring happen, right? So we can run this implosion. 220 00:49:05.800 --> 00:49:23.830 Mark Kushner: model it, and then post-process it, generate all our synthetic data. And now this is our engine for Bayesian inference. We now can use this to compare to the actual measurements we made, and do this in a statistically, quantifiable way to get uncertainties on 221 00:49:23.830 --> 00:49:36.389 Mark Kushner: quantities that occurred in our implosion, but we can't directly measure, like p-taw. P-Tao is sort of this aggregate quantity that you get from multiple different diagnostics. So this is our approach, this is how we're going to do it. 222 00:49:36.990 --> 00:49:52.329 Mark Kushner: Here is a validation case we just did recently, where we took a flash calculation, we initialized, at… we initialized calvera at peak velocity, and then we used Monte Carlo sampling to figure out what those initial conditions should be. 223 00:49:52.480 --> 00:50:12.349 Mark Kushner: because we didn't, you know, we pretend this is an experiment, we didn't know what the initial conditions are. What initial conditions were required in order to match a handful of observables from that flash simulation? So in this case, the observables are yield, bang time, the burn width, peak temperature. 224 00:50:12.350 --> 00:50:14.409 Mark Kushner: peak density and BR. 225 00:50:14.740 --> 00:50:37.089 Mark Kushner: So, we get really good agreement with some of them. The sort of blobby plots are the probability distribution from the posterior in the sampling, and then the red plot… the red points are what we extracted from the flash simulation. So we can do this. There's obviously a little room for improvement, but this is actually quite good. You can see the collection of light blue traces. 226 00:50:37.090 --> 00:50:55.129 Mark Kushner: is actually the ensemble of inner radius trajectories that we got from this inference, and they're very tightly clustered around the inner radius in dark blue that's extracted from the flash simulation. And then the temperature and density as a function of time of the fuel also agree quite well. 227 00:50:55.130 --> 00:51:03.209 Mark Kushner: So, this gives us a lot of confidence that we can use this as a tool to dig deeper into our experiments and understand what really happened. 228 00:51:04.100 --> 00:51:18.949 Mark Kushner: Finally, like, the last kind of cool model I wanted to talk about was one called Crispy. This is… Patrick Adrian wrote this. This is a 1D Lagrangian MHD code, fully functional Lagrangian MHD code, but it's all written in JAX. 229 00:51:19.280 --> 00:51:44.019 Mark Kushner: The reason to do that, again, is it's fast, and it gives us derivatives. So the main motivation is measuring the current. So this is a schematic of an experiment done on Z, where, the current, the velocity of this top plate was measured. So we have current flowing in this cavity, and it pushes this top plate up, as well as the return can out. Measure that velocity, and that allows us to back out what the current was. 230 00:51:44.020 --> 00:52:03.459 Mark Kushner: through a complicated inverse problem. So we wanted a fast code that simulates the forward problem, so we can do this same methodology of doing the forward modeling to fit to the experimental observable, in this case velocity, so that we can then back out what current was used to drive that with quantified uncertainties. 231 00:52:03.640 --> 00:52:22.799 Mark Kushner: Additionally, this plot… this shows a model of a flyer plate on Z, so a planar explosion, essentially, and it shows the different Lagrangian layers and how they evolve. This is the front surface of the flyer where you would measure its velocity. 232 00:52:22.800 --> 00:52:24.040 Mark Kushner: It turns out. 233 00:52:24.280 --> 00:52:40.330 Mark Kushner: when you do this, and this has been done on Z, there's a point where the magnetic field diffuses all the way through to that front surface, and now that front surface starts to accelerate because the magnetic field is sort of vaporizing and blowing off the outer surface. That time is called burn-through. 234 00:52:41.090 --> 00:52:51.429 Mark Kushner: The burn-through time is extremely sensitive to the electrical conductivity of your material at whatever point in its face space you're at, at that time when it burns through. 235 00:52:51.430 --> 00:53:05.770 Mark Kushner: And so what this plot here shows… so this… these black lines are trajectories of where, in a 60 megamp implosion, the liner will follow these trajectories. So at different points in the liner, they'll see different face space… regions of phase space. 236 00:53:05.770 --> 00:53:14.890 Mark Kushner: But this essentially shows the collection of points starting at, ambient aluminum conditions and getting compressed up where it flows through phase space. 237 00:53:15.160 --> 00:53:25.830 Mark Kushner: The color, color plot behind it shows the sensitivity computed through this model with JAX of 238 00:53:25.930 --> 00:53:40.490 Mark Kushner: that burn-through time to the conductivity. So what this tells us is, if we can measure the burn-through time with certain experiments, we can constrain the conductivity of aluminum in regions of the face space that are very important to us. 239 00:53:40.780 --> 00:53:59.919 Mark Kushner: And so this is really important for our modeling. This is important to do for materials maybe not like aluminum, that aren't as well understood, that we might want to use. But you can see most of the data for aluminum sits in the rarefied regime, so lower than ambient density, or at ambient density, but higher temperature. 240 00:54:00.180 --> 00:54:14.979 Mark Kushner: to the best of our knowledge, no one has made measurements of the aluminum conductivity over here. This plot shows that that can be done, and bonus, this plot was generated with one run of this CRISPI model. 241 00:54:15.280 --> 00:54:21.459 Mark Kushner: So that's what… that's kind of the magic that these… having auto-differentiable, physics models can give you. 242 00:54:22.510 --> 00:54:46.990 Mark Kushner: All right. So, last thing, we are excited to announce the Pacific Fusion Users Program. We just had a webinar, I think 2 weeks ago, kind of opening this up to the community. You can check out the QR code here, it'll take you to users.pacificfusion.com, I believe. And we're accepting what we're calling expressions of interest. We want to… we're soliciting feedback from the community 243 00:54:47.030 --> 00:54:49.919 Mark Kushner: What you might want to do on our facility. 244 00:54:49.960 --> 00:55:14.880 Mark Kushner: This is partly an information gathering exercise for us to say what capabilities might the community want that we're not planning for, so we can start to understand what it would take to actually enable different kinds of experiments than just the fusion ones that we're interested in. So, I definitely encourage you to check this out. This is intended to be very low barrier to entry. This is not a proposal, this is not a 245 00:55:14.880 --> 00:55:25.280 Mark Kushner: white paper, even. This is a couple paragraphs and a little bit of information about your use case, so that we can collect and then move on. We're going to have a formal call for proposals. 246 00:55:25.280 --> 00:55:27.439 Mark Kushner: In probably 2027. 247 00:55:27.440 --> 00:55:42.440 Mark Kushner: And so… so yeah, this is something we're really excited about. We're really excited about opening up our facility to the broader community for use, because there's lots of really cool things that we know people would like to do with this capability. 248 00:55:42.750 --> 00:55:44.720 Mark Kushner: And we'd like to help make that happen. 249 00:55:44.980 --> 00:55:51.340 Mark Kushner: So, this is my, conclusion, and I've sort of been going on for a long time, so I'll stop here. Thanks. 250 00:55:57.790 --> 00:56:00.429 Mark Kushner: All right, questions for Dr. Neck? 251 00:56:05.270 --> 00:56:10.189 Mark Kushner: Yes, absolutely sent it. It sounds like pretty much the physics is… 252 00:56:11.030 --> 00:56:26.069 Mark Kushner: Understood. Is there any outstanding questions? No, it's an MHD code. There are some extended MHD capabilities, but we're not currently using that. 253 00:56:26.070 --> 00:56:39.269 Mark Kushner: But, is there any tests? Yeah. Yeah, so, there's definitely… the physics is not, fully settled. There are… there's definite… once we're operating, there's research we need to do. 254 00:56:39.280 --> 00:56:54.390 Mark Kushner: Right? There are, there's definitely things that are, you know, there's risks that we need to mitigate. And so, the big ones, typically, you know, as all ICF, concepts share instabilities, right? 255 00:56:54.390 --> 00:57:08.279 Mark Kushner: no one fully understands how instabilities will scale, and when you're talking about MHD, the magneto-Rayleigh Taylor instability, you can actually get modifications to it by kinetic effects. The Hall term can have significant effects, it's believed. 256 00:57:08.280 --> 00:57:17.529 Mark Kushner: There's also the way instabilities are seeded, which often we think about the electromal instability as the seed for, 257 00:57:17.550 --> 00:57:21.269 Mark Kushner: that the MRT can then kind of take hold of and run away with. 258 00:57:21.590 --> 00:57:40.649 Mark Kushner: And the way these things scale to higher current, are not perfectly well understood. And those actually aren't quantities that were preserved in that scaling that I mentioned. And so one of the reasons why the liner got a lot fatter, going from Z scale to the high current scale, is to help 259 00:57:40.650 --> 00:57:49.099 Mark Kushner: give a little more robustness to instabilities, because there's concern that they could be more aggressive. So, one, instabilities are a big thing. 260 00:57:49.100 --> 00:58:09.049 Mark Kushner: fortunately, it's always much easier to study the thing you're worried about at the scale you care about than it is at a much smaller scale, and so that's what we're pushing for. So I think there's definitely… there's definitely work to do, things that aren't extremely well understood. Another thing that could happen is current losses. 261 00:58:09.050 --> 00:58:15.820 Mark Kushner: through the system, particularly as you get closer to the target, where the plasma environment becomes very, 262 00:58:16.340 --> 00:58:18.280 Mark Kushner: Very difficult to predict. 263 00:58:18.280 --> 00:58:39.669 Mark Kushner: We typically use PIC models to model this environment, but they… the densities start to get up to the regime where it becomes very difficult for PIC models to model them, and then the MHD codes are missing a lot of the physics that's important there. Extended MHD being just one of the things that can happen there. So we have… 264 00:58:39.670 --> 00:58:56.240 Mark Kushner: some understanding based on smaller-scale drivers and Z of how these things scale, but there's definitely questions about that. So, we're… we're planning a… we're planning a science campaign, essentially. We are not, expecting to… 265 00:58:56.270 --> 00:59:02.590 Mark Kushner: you know, hit it out of the park right out of the gate, and have to do just a little tweaking, right? 266 00:59:02.590 --> 00:59:19.700 Mark Kushner: that's one of the reasons we have all these diagnostics that we're building, is because we know we're going to need to investigate failure modes, and we're going to need to understand how to modify our targets to, overcome challenges that we weren't expecting. So we're… hopefully we're building that into the process, that's the goal. 267 00:59:23.080 --> 00:59:24.149 Mark Kushner: That person. 268 00:59:25.030 --> 00:59:34.359 Mark Kushner: So for these, cassettes, that you're, placing in, firing with ones a day, 269 00:59:34.510 --> 00:59:37.080 Mark Kushner: How… what is, 270 00:59:37.530 --> 00:59:47.749 Mark Kushner: like, the manufacturing lead time, per cassette, like, is it only one shot per cassette? Is there, like… Yeah. 271 00:59:47.880 --> 01:00:02.109 Mark Kushner: How is that pull scheme? Yeah, so the cassettes are disposable. They're, the… they're intended to sort of contain the blast, but they do that constructively… or, sorry, destructively. So they're not… they're not reusable. The… 272 01:00:02.110 --> 01:00:08.829 Mark Kushner: The goal is to get the sort of cycle for an experiment down to, at most, a couple of weeks. 273 01:00:08.850 --> 01:00:10.930 Mark Kushner: So, 274 01:00:10.930 --> 01:00:32.520 Mark Kushner: basically from specifying a design to building your cassette and having it ready to be fired to, at most, a couple of weeks. The reason for this is that we want to iterate as quickly as possible. We want to be able to change things, build a new target, and field it as quickly as possible. The typical design cycle on Z is more than 6 months. I think it's even longer than that on NIF. 275 01:00:32.520 --> 01:00:35.920 Mark Kushner: And so we really need to cut that down in order to make progress. 276 01:00:36.060 --> 01:00:42.739 Mark Kushner: But this is all work that is, you know, there's a prototype for our cassette system being built now. 277 01:00:43.040 --> 01:00:53.540 Mark Kushner: Right, so we're… we're actively studying and exploring manufacturing techniques. Do we do it on-site? Do we have other people build things, send them to us? 278 01:00:53.540 --> 01:01:03.469 Mark Kushner: Can we stockpile? Like, what components will be the same all the time, and we can stockpile, so this is just sort of the… we're wrestling with all this stuff at the moment as we speak. 279 01:01:06.440 --> 01:01:10.420 Mark Kushner: I think I… my… I was basically… 280 01:01:10.590 --> 01:01:18.650 Mark Kushner: My question was a follow-up to Louise's. So, you talked about validation in this context, but you're talking also… but you're also saying that you're missing physics, so… 281 01:01:19.260 --> 01:01:37.110 Mark Kushner: I'm… I'm curious, kind of, what you mean by validation, or else, what is, kind of, yeah. Yeah, so I… and maybe for what I showed, maybe verification was the better word. I always get the two mixed up. But, yeah, we were comparing to, code results and theory. 282 01:01:37.110 --> 01:01:40.740 Mark Kushner: So it's probably verification, 283 01:01:40.880 --> 01:01:54.330 Mark Kushner: But, the codes we're comparing to, codes like Hydra, have been validated pretty well against experimental data. Now, part of that validation process reveals that there are things that we know that are not there. 284 01:01:54.330 --> 01:02:01.570 Mark Kushner: Right? There's physics that we know that does not exist in the codes. The extent to which it matters is debatable. 285 01:02:01.570 --> 01:02:20.149 Mark Kushner: And the area of active research. So extended MHD is a great example. There's different camps of people that think it's really important, and there's some camps of people that think it's not really important at all, to perturbation, or… or maybe even not a player at all. So it's an active area of research that people are exploring. 286 01:02:20.150 --> 01:02:32.840 Mark Kushner: But I think we're… we're in no more of a… you know, this is a very common situation in high energy density physics and… and ICF, right? Lasers, they're LPI, right? 287 01:02:32.840 --> 01:02:42.319 Mark Kushner: the codes don't handle this. There might be some simplified models that are used to mock up various aspects of LPI, but it's something that you 288 01:02:42.600 --> 01:02:55.220 Mark Kushner: You model, you compare to experiments, and you go through this cycle to come to a modeling strategy that gives you, a useful predictive capability, I think is probably the best thing. 289 01:02:55.460 --> 01:02:58.929 Mark Kushner: It's useful when you can use it to design experiments. 290 01:02:59.150 --> 01:03:17.299 Mark Kushner: And the goal of designing those experiments is to either make progress on some objective, or shed light on the things that you don't understand, right? So, of course, the models aren't perfect. There's things we're missing, there's things that are left out for convenience because they're just too expensive to compute, there's things that we just downright don't know how to compute. 291 01:03:17.300 --> 01:03:27.089 Mark Kushner: And but these codes have a long history of being used to successfully design experiments on Z, on NIF, on Omega. 292 01:03:27.110 --> 01:03:35.990 Mark Kushner: And that's the body of work that we're comparing to, as well as the available experimental data, to show that flash 293 01:03:36.080 --> 01:03:39.609 Mark Kushner: Is in a point where it is a useful design tool. 294 01:03:39.640 --> 01:03:56.600 Mark Kushner: And in fact, we actually used it to design experiments on Zee that we conducted in October, and it was extremely successful, and compared extremely well to the, to the, experiments, once we got the data. So I think, 295 01:03:56.600 --> 01:04:00.499 Mark Kushner: We've done the work to show that Flash is a design tool. 296 01:04:00.500 --> 01:04:01.520 Mark Kushner: Essentially. 297 01:04:06.290 --> 01:04:09.170 Mark Kushner: Well, that's after you're both, but I'm… 298 01:04:09.390 --> 01:04:24.900 Mark Kushner: Yeah, so, yeah, great. It seems, if I understood it correctly, that your diagnostic inserters and sort of the diagnostics are built in water? Yes. 299 01:04:25.050 --> 01:04:39.090 Mark Kushner: And so I guess that adds a complication to building out your diagnostic insert and things like that. Does… does that introduce challenges in sort of using some of those things? 300 01:04:39.090 --> 01:04:45.160 Mark Kushner: facilities like Z and Omega and NIF have learned about diagnostic insertion and keeping your 301 01:04:45.160 --> 01:05:08.940 Mark Kushner: Yeah, so I think the water environment… deionized water can actually be problematic. It wants to pull ions out of the materials that are around it. But I think there's known strategies for how to deal with that. As an example, right, Z has this large deionized water tank with stainless steel everywhere, and the stainless steel is very stable. 302 01:05:08.940 --> 01:05:14.010 Mark Kushner: in deionized water. And so there's things we know. How do you construct a sealed 303 01:05:14.010 --> 01:05:30.669 Mark Kushner: container, so no water gets into it, with materials that won't react with the deionized water, won't degrade, or if they do, what's the schedule with which you need to, you know, look at them, replace bolts and gaskets and things like that? And so I think that's a really solvable problem. 304 01:05:30.670 --> 01:05:33.830 Mark Kushner: The problem that's a lot harder to solve 305 01:05:34.040 --> 01:05:40.989 Mark Kushner: is the problem of radiation, and water actually really helps us solve that problem. So our… our… 306 01:05:40.990 --> 01:06:03.770 Mark Kushner: insertion modules will actually take the diagnostic from the water section into the chamber, and so that, you know, we won't benefit a lot from distance and water in terms of shielding, but a lot of our diagnostics will actually live all their time, and they'll spend all their time in the water. The radiation power measurements, the neutron measurements. 307 01:06:03.770 --> 01:06:09.350 Mark Kushner: They will always be out there at large distance in the water, and I think the… 308 01:06:09.610 --> 01:06:26.589 Mark Kushner: the benefits you get from the radiation shielding far outweigh any challenges you… engineering challenges you come to by needing to make it hermetically sealed and stable. I think those are… those are very solvable challenges, and I think we'll run into issues with the radiation far before we run into issues with the water. 309 01:06:29.380 --> 01:06:51.910 Mark Kushner: I got one quick last question, maybe. You'll see these fats codes and decks, branching and searching and doing stuff makes me wonder, of course, about AI, and have you guys use any, looking into how to use that, or… Yeah, and we're always interested in how to use it and thinking about it. I would say the largest extent we're using it right now is as our co-pilot for writing these codes. 310 01:06:51.910 --> 01:06:58.090 Mark Kushner: A lot of this work, that I showed, you know, the sampling and this work. 311 01:06:58.090 --> 01:07:05.479 Mark Kushner: you know, I'd never written code in JAX before, and it's very different. Patrick Adrian has now become our, like, resident expert in JAX. 312 01:07:05.480 --> 01:07:10.949 Mark Kushner: But with a lot of help from ChatGPT and Copilot. 313 01:07:10.950 --> 01:07:32.300 Mark Kushner: And I have actually found it to be tremendously useful for prototyping codes and building things that I didn't know how to build. Actually, debugging in JAX is non-trivial. You can't… because it's compiled, you can't debug the way you normally do, and ChatGPT helped me solve that problem. So, I think, largely right now, we're using AI as an assistant in a lot of our workflows. 314 01:07:32.300 --> 01:07:37.000 Mark Kushner: But we're… we're interested in how we can use it to do more. 315 01:07:37.000 --> 01:07:47.859 Mark Kushner: But it hasn't been a, like, a priority at the company to… we don't have a person who's tasked with figuring that out. You know, how do we… how do we use AI in the future to, you know. 316 01:07:48.390 --> 01:07:53.319 Mark Kushner: give us a 10x speedup in everything we do. Like, we haven't done any… we don't have any initiatives like that. 317 01:07:54.510 --> 01:07:58.010 Mark Kushner: Alright, well, let's thank our speaker again. 318 01:07:59.090 --> 01:08:00.250 Mark Kushner: Thank you. 319 01:08:01.460 --> 01:08:03.759 Mark Kushner: Do you have any other questions? 320 01:08:19.390 --> 01:08:21.059 Mark Kushner: Yeah, we also didn't. 321 01:08:21.069 --> 01:08:22.689 Looja Tuladhar: Thank you. Thank you, thank you. 322 01:08:22.689 --> 01:08:33.079 Mark Kushner: Nice to meet you. How's it going?