I worked on a previous project that was sort of an ancestor to this. We used grid computing (before "cloud" was even a thing) to rapidly scale up compute- right after an experiment, you want to recalculate the optimal parameters based on what you just learned from the experiment and then start over as fast as possible. This was well before anybody could say "reinforcement learning" or "neural network" around physicists without being pilloried.
Very cool, I came several years following and we too had a grid for similar tuning work on LHC. My fondest memories were the bizarre hangups that could occur in custom stuff like this, and the way you'd find out that you done fucked up. My favorite time was when I broke a database by sending an accidental "query all the stuff", but it was really stupid, like from the console all you had to do was type a partial "ls" and it would assume a wildcard. So when the authority in my group attached me in the email and said I broke the main monitoring database I told them they had stupid defaults and never heard anything again. My second favorite memory was when I found a way to gain production access to grid resources just by naming your output datasets in a particular way. I was sat down in a hush ceremony with the Pakistani post-doc from my group and the Russian? programmer of the system and asked to keep it quiet. I kept it quiet, and am pretty sure I failed some sort of spy test that day. It wouldn't be the last time I felt like I failed a spy test, I guess that is my third fondest memory from my time in grad school.
Well, my nickname in my first job out of grad school was Murphy (as in murphys law). I was essentially a dowsing rod for corner cases and it was very much not intentional and extremely distressing. You could say I have some experience.
I've worked in a number of factories and it seems like everyone of them has a Murphy. The people who get confronted with problems that begin with everyone saying 'thats not actually possible'. I will always remember the poor guy who had to figure out the local fire department's radio repeater was rebooting ONE SINGLE piece of equipment in a factory from about a third of a mile away.
I have always wondered if there is a way to detect this particular trait either using MRI or possibly smell. Once you are in a space you can usually identify them because everyone seems to slightly avoid their workspace out of fear of being infected.
> I will always remember the poor guy who had to figure out the local fire department's radio repeater was rebooting ONE SINGLE piece of equipment in a factory from about a third of a mile away.
Old relays (especially when switching large DC currents) are fantastic for this purpose. The arc will essentially give off a mini EMP across the whole spectrum and woe to you if your gear isn't shielded and grounded properly.
I have a family member like that. She walks into the room: all computing equipment mysteriously crashes. I haven't been able to pin it on a real world cause yet.
I know this is maybe weird, but I kind of wish programmers and other "technical" people dressed up at least a little bit more for the office. It makes the work environment feel a bit tidier and more professional to me.
It's not weird, it's controlling. Needing everyone to wear mono-colored shirts with buttons on it to feel tidy is a you problem.
A company I worked for (that I did generally like working for) implemented a suit and tie policy for management and above. It didn't affect me but I started looking for the door immediately and left a few months after.
Don't want to be in a company that is still clinging to 19th century values.
I always wear a collared shirt to work because I like them and they do look neat, but I cannot stand people that demand it from others.
Hygiene is pretty important in the business I work since some of our devices come in contact with drinking water. But I can assure you that hygiene often does not correlate with outside appearances.
Still, we have rules for people that have customer contact because people are still judgemental, even if they are often demonstrably wrong.
To underscore, I would bet that washing hands frequently, wearing gloves when touching food and masks/hood when working near food is much more important than the difference between wearing a fresh white-collar shirt vs a sweat shirt you change every other day.
That is vastly more important. It is mandatory to wash your hands before visiting production as well as changing clothes. Regardless of how clean they look, they will contain foreign matter. The cafeteria is pretty much declared a hazard zone, not only because of that one colleague who has a weird cheese experiment going on in the fridge.
Downside for management is that we are only allowed to drink water in our offices and are not allowed to eat outside the cafeteria. With time I have grown to like it that way, even if I am a coffee junkie. But it never smells bad in the office in exchange.
There is still some strange association with cleanliness and neat shirts.
The problem is, that alot of humanity still relies on cliches to communicate "base-truth". Suite and tie looks professional in movies, so you get more credit and respect for superficial looks on the job.
One must not support such a ridiculous look based culture by heart, its instead more a - "I want to reap the benefits & rewards from idiots in awed reference to work uniforms"
Ironically the same happened to the hoodie, after various hackermovies. Wear a hoodie and sit with a laptop, obviously improves your coding ability.
I've often wondered about dress code for programmers. From a quick poll conducted at lunchtime, we concluded that there definitely is a way to dress for acceptance among devs, it's just hoodies and witty T-shirts instead of suits and ties.
Several people confirmed that they would probably be biased against applicants for dev positions coming in for an interview in a full on business suit with tie and slicked back hair (think "slick salesman" look). Perhaps not justified or fair, but interesting as a data point nonetheless. N=4 though, so don't take it too seriously.
The distinction, and its relevance, are obvious. If you're going to feign ignorance I have better things to do with my day than argue with you about it.
Yes, there is a distinction, no it is not relevant. Just like to a blind person it would not matter what you wore so therefore 'looks' and 'what you are wearing' are closely related, related enough to both be used in this context.
One of the funniest things I've ever seen was behind the scenes at TGIF at Google. Sundar was wearing some nice bay area gear (ie, ready to climb a mountain or meeting a corporate board room) and there were a couple PMs nearby wondering what brand he was wearing and how they could buy it.
I'm just an old man, pointing out that it's all been done before, and it will all be done again. The big difference is it looks like they're in the realtime controller for the plasma, rather than determining static settings for a subsequent run.
Yours was a tools and infrastructure project, and this is about learned, autonomous control. They are completely different goals.
> determining static settings for a subsequent run.
AFAICT, your software didn't do that. Rather it provided infrastructure for the researchers to do that themselves more efficiently.
From your paper:
> The ultimate goal of FusionGrid is to allow scientists at geographically distributed sites to participate fully in experimental and computational activities...
Could you be a bit clearer why you're being critical? Much of what DeepMind does is rooted in prior projects, they just sprinkle a little more RL and have a great press team. Nobody would use their system for real in production, just like when they published how they could run Google datacenters 5% cheaper in real time- turns out, nobody actually deployed that because the settings it came up with don't make sense to humans.
I'm not trying to be critical of your project, sorry if it came off that way. I'm just trying to clarify when you say "it's all been done before", exactly what in the current work had been done in your project. Although they are both in the fusion domain, I don't see how the current work builds on anything done in your work, even in a very distant way.
The root enabling factor was not only the reinforcement learning, it was the simulation. Put differently, any environment that you can simulate within sufficient accuracy, you can derive robust and efficient control by reinforcement learning.
I'm probably being excessively pedantic but I would suggest sufficient fidelity is a more appropriate phrase than sufficient accuracy.
The exact precision doesn't seem to be the underlying problem being solve, what matters is being able to craft a stronger understanding of relationships of underlying phenomena that can then be leveraged to product new control theories.
I think they are just talking about how accurately the model reflects the real world.
If someone draws a map and it has some streets with the wrong names, you say, "This map is not accurate." It doesn't mean you are talking about something numerical.
I would argue it isn't. I love that graphic (I use it in classes frequently) but it's limiting here.
Accuracy is about how close to the underlying behavior that is being modeled is. In this case, the goal is modeling the controllers they are using.
Precision is about how stable I can make my controller parameters and how stable they can make the behavior or the underlying fusion.
What they did was changed the fidelity of the model. Rather than parameterizing it using a known controller model (e.g., a PID controller) they used the neural network to fundamentally discover the important parameters that they can understand and how they interact.
The simulation can be accurate or not accurate entirely independently of how much fidelity of the actual phenomenon you model. You need to have a high fidelity model, and that model needs to be accurate, and they behavior needs to be precise and stable.
An example would be how wing profiles in aircraft are simulated/modeled. I can do it in 2D with high accuracy and high precision - it's a GOOD 2D simulation. But its low fidelity. I might be better off with a lower accuracy model of a high fidelity model, like a 3D model that lets me look at how the ends of the wingtip perform as well. Increasing the accuracy of that model might make the results less stable (because numerical techniques...great). A lot of improvements in computing power and simulation techniques have resulted in the ability to simulate systems more completely and learn about the interactions - not just making the simulation more accurate to what it contains.
If the reward function is sufficiently dense, then likely yes. RL uses function approximators to solve problems, so in theory the perturbations of the chaotic system shouldn't matter as much.
How to simulate human humor responses? Then we can evolve AIs that are funnier and more resourceful than anyone on the planet.
How to simulate human connection, attraction and arousal responses? Then we could have bots that are smoother talking and far more preferable to human lovers.
How to simulate human responses to nutrition? Then we could have individualized recommendations for diets, achieving goals like holistic health or muscle definition and addressing issues like diabetes.
These are all complex systems where figuring out how to simulate them is a much difficult task than even reinventing reinforcement learning from scratch.
AFAIU this is entirely simulated - if it turns out the simulation doesn't perfectly match reality, what are the challenges in re-adjusting? Does the RL network need to be re-trained with massive amounts of data on a new improved simulation or can it be fine tuned on the real thing?
Hi, author here. We validated the neural network controller on the real plant in a couple of experiments. All experiments in the paper are based on data from running the controller on an actual tokamak.
There is a video of one of these experiments in the linked blogpost (figure 2), where we show imagery taken from a camera looking into the vessel. Next to it is a reconstruction of the shot, visualised on a cross-section of the torus.
The simulation not perfectly matching reality is indeed a major challenge, and the nature paper is mainly about showing how you can overcome that challenge. As others have said, plasma is a pain to model, and you cannot collect a ton of data on these machines. But it seems technology is coming at a point where we know how to work within these limits and still get it to work.
Very cool, and thanks for replying to laymen comments such as mine.
Do you think this line of research will eventually feed back into improved designs of electromagnets, chambers and so on? My limited understanding of the field is that the optimal reactor design is a moving target that's changed a lot over the years due to engineering improvements in manufacturing, material sciences and so on.
Yeah, I think the main contribution of this paper in the long run, might be that the time is right for optimizing the controller together with the hardware?
I'm imagining optimizing the plasma configuration, reactor shape, coil/sensor placement and controller all in one giant computational optimization. All the algorithmic components and models humanity needs to attempt this seem to be there.
Well as long as there have been electronic computers they have been applied they have been applied to the problems of magnetic confinement. For tokamaks static configurations and dynamic controls have been simulated. For stellarators geometry optimizations have been done. The time has always been right for this application.
Stellarators are particularly fun. First you optimize a last closed magnetic flux surface (a 3D sheet) with relatively simple plasma models, then you optimize coils (you need to weight things like magnetic field error, gaps big enough for heating and diagnostics, and coil complexity a la minimum radius), then you perform more accurate simulations to start the overall optimization loop again.
There are 4 episodes on fusion. You're probably referencing 312 which talks about simulating stellarator plasmas a bit.
Episode 22 is an intro to fusion power research and tokamaks, but could be skipped if you are aware of the big concepts.
Episode 157 is an interview of a director at ITER: the largest fusion power project in terms of scale, ambition, and funding. ITER is expected to demonstrate net energy gain fusion within the next ten years. Episode 157 gives a good sense of where the cutting edge of fusion power research exists currently and its potential place in society.
Episode 304 interviews the authors of a recent popular science book on tokamaks. This episode covers the history of fusion power research, and the economics involved in future power plants.
Episode 312 is a set of interviews at W7-X: the current largest stellarator. This episode take a deeper dive into what fusion research is like and how threads are being tugged on.
This same podcast also has interesting episodes on plasma physics and superconductors. The superconductor episode is especially worth listening to more than once.
Hi!
this is fantastic article! Years ago I was able to visit the "Wendelstein 7X" stelarator-type fusion reactor in greifswald (Germany) when it was still assembled. It was so cool.
Do you think this approach of controlling plasmas also translates to these types of reactors?
Reading your article I didn't understand a lot of things and wondered if there was some kind of fusion-reactor-physics word-list where the most important keywords are explained a little. (without reading a whole book on plasma physics :-))
Also the timescales are fascinating. How long is 200ms in a tokamak? How long does this reactor need to ignite or shut down the plasma?
I think one of the challenges with stellarators is that they are harder to simulate due to the twisted geometry of the chamber (with a tokamak you can often simulate a single 2D slice of the torus).
All optimized stellarator designs are periodic, so only one period (with correct boundary mapping) needs to be simulated. But yeah, it's a much larger search space.
This is why fundamental plasma physics is so valuable to fund. ML can come up with good models that fit the data in the domains that there are empirical datasets to train against, but it can't reliably extrapolate outside of those domains.
So this is a good tool for the computationalists but we still need the experimentalists. This is science.
To clarify, this result did not have the benefit of a large labeled dataset. They used a simulator and got working results without _any_ finetuning on data measured from the real world.
> So this is a good tool for the computationalists but we still need the experimentalists. This is science.
I would say that this is a good example of collaboration between the groups.
Probably massive amounts of data but it depends on how far off the simulation is from reality and the expected cost of collecting that data vs running the simulation a few billion times.
Not far off and you want perfect results? Go with actual data. Very far off where it's not even usable? You'd need to go back to the training environment and redo the reward function.
Generally speaking (and I'm not sure if this is the case from reading the link but I may have missed it) reinforcement learning for optimal control is done to fine tune while traditional control methods are used for coarse adjustments. Since this is deepmind they probably want to use RL for everything
Too little, too late. The traditional fission industry is being killed by high CAPEX and nothing about the current tokamak based designs suggests they will be any cheaper to build for a given power rating:
- They need large, highly advanced cryocooled superconducting magnets in very close proximity to a hundred million degree plasma. This only makes economic sense in massive, and expensive plants.
- They are a very strong source of fast neutrons useful to transmute cheap depleted uranium into plutonium, so carry massive proliferation risks, need close regulatory scrutiny and will require mounts of paperwork to operate, thus exceptionally inflexible to improvements and rapid iteration. Just like the current fission crop.
- Aneutronic fusion is a currently a purely theoretical concept, in the last 70 years nobody has been able to contain even the much cooler D-T plasma for economically viable durations and temperatures.
- The structure of the reactor becomes radiologically active and cleanup operations must be considered. Highly penetrating neutron radiation means some radiation will escape regardless of containment, requiring a radiological exclusion zone. No Mr. Fusion in your car, sorry.
- They operate and must breed sensitive nuclear materials - Tritium, a well known component of boosted thermonuclear weapons. The limited efficiency of tritium production from lithium-6 might require obtaining some from fission reactors to top up the fuel cycle and keep fusion reactors operating.
So when you draw the line, a life time of magnetic containment research has produced a speculative design that even if it were to work, which it doesn't, would be, in the best case scenario, comparable to existing fission designs that are being phased out for cost and risk issues.
A PhD money pit with zero chance of ever building anything useful. I know it, you know it, I'm ready for the proverbial downvote protecting one's paycheck from their own knowledge.
>The limited efficiency of tritium production from lithium-6 might require obtaining some from fission reactors to top up the fuel cycle and keep fusion reactors operating.
Since nobody has yet demonstrated tritium breeding, a citation is actually not needed.
There exist purely theoretical models where neutron multiplication is used so that the single fast neutron generated in a D+T reaction can breed more than a single tritium atom when it hits li-6 (which would be clearly insufficient to have a self fueling rector).
It's all a complex mess of absorption cross sections and neutron spectra, dependent on lithium blanket geometries, coolant and structural material parasitic interactions. The tritium breeding efficiency today sits at exactly 0% and until that changes, concerns that it may never reach the required 110% are at least somewhat warranted.
From the talks I've been to the targeted performance of blankets is 120% and simulations predict this should be possible without prohibitively thick blankets.
ITER's TBM will be doing important work to test theory.
It isn't clear to me that external sources of tritium will be necessary. Between Li-6 fission and Be neutron multiplication there is theoretically plenty of neutrons to go around even though they aren't easy to capture.
Viability of blanket breeding hasn't been demonstrated to work yet, but my original point is that it hasn't been demonstrated to not work. I see now the "might" in your statement.
Sure, by the "it hasn't been demonstrated to not work" standard the whole field should generate cheap electricity tomorrow. Just that it's so unimaginably, fiendishly hard to make it work, perhaps the most challenging technical problem in history. It might never work, for interesting values of never.
My point was that fueling is in itself a very difficult and complex problem that has yet to be solved, far from the "it runs on seawater" popular tropes.
When SPARC ignites, 5 years from now, and when it achieves long burns at Q=10, after a good few more years, then I will redact the "nobody has been able to contain D-T plasma for economically viable durations and temperatures" part. The rest of the points will probably remain true decades into the future, though.
This is a bit off-topic, but can anyone here explain the expected economic advantages of fusion power over fission? My understanding is that fusion plants would also use heat-driven turbines, and be pretty capital intensive. Would they fill a different niche, or are they expected to be fundamentally more cost-effective?
- The end product of Deuterium + Tritium is regular, stable Helium, making waste disposal both safe, and cheap
- The input of the process, heavy water, while not safe, is way less dangerous than thorium or uranium
- The whole process is no way involved with nuclear weapons, making security concerns much less relevant
- Since the process produces magnetically charged plasma, steam turbines are not necessary, a solution of directly harvesting energy with electromagnets was proposed.
Direct harvesting is not possible with the fusion reactions we are currently working on. The products are electrically neutral, so the only way to harvest the energy is as heat.
Helium-3 fusion produces charged particles, which it might be possible to directly harvest as electricity. The He3 fusion reaction has a high activation energy, which is currently unachievable, and there is no Helium-3 on Earth. We'd have to get it from space, somehow. (The movie "Moon" was set on a Helium-3 mining outpost on the Moon.)
Conveniently, pure deuterium fusion (D-D) is easier than D-He3 fusion, and the output of D-D fusion is half He3, and half tritium which decays to He3 with a 12-year half-life. So if we can get net power from He3, we can make He3 from deuterium and generate energy in the process.
D-D fusion does produce neutrons but they're much lower energy than D-T neutrons. Fusion startup Helion is working on a hybrid D-D/D-He3 reactor, saying the combination will produce only 6% of its energy as neutron radiation, low enough so they can do direct conversion.
They've built half a dozen reactors, and now they're working on a seventh that they'll use for a net power attempt around 2025. They recently had a fundraising round led by Sam Altman, and raised $500M with another $1.7 billion of commitments based on milestones.
There is no usefully available Helium-3 on the moon, either. The only practical source known is decay of tritium, which must itself be synthesized by nuclear reaction, and is, for military use.
Another choice is fusion of ordinary hydrogen, usually labeled "p" for the proton, with boron, B, thus "pB". But that is even harder to achieve. Still, it is being worked on.
There is no route to commercially viable fusion extracting heat from Tokamak reactors, as any such reactor would need to be enormously bigger and much more expensive to operate than the same-rated fission reactor, which is not today competitive, and gets less so all the time.
Some people hope that something can be learned from Tokamak work that might be applicable to potentially practical designs, but the money is all going to Tokamak, while the others mostly go begging.
(Note that you can drink a lot of pure heavy water without worrying about any health impact. I still wouldn't recommend it, but bad things would be unlikely to happen by accident.)
> You could consume a single glass of heavy water without suffering any major ill effects, however, should you drink any appreciable volume of it, you might begin to feel dizzy. That's because the density difference between regular water and heavy water would alter the density of the fluid in your inner ear
> magic heavy chemical water with different nuclear properties...
That seems a bit dismissive. I would naively assume that there have been exactly zero studies to see if there are any problems with babies drinking heavy water, beyond some LD50 extrapolation.
Another big one is the default-to-safety nature of fusion vs. fission. With fission, if things go wrong, the nuclear process often speeds up and can run out of control. With fusion, generally when things go wrong temperatures dissipate and the fusion process halts naturally.
The amount of helium will be much lesser than what you think it is. If we produce all the energy in the world using fusion, it will create less helium per day than a tank used by kids helium balloon shop.
> If one ton of deuterium were to be consumed through the fusion reaction with tritium, the energy released would be 8.4 × 10^20 joules[1]
That's 0.84 exajoule(233 TWh) per kilo of deuterium or 0.42 exajoule(117 TWh) per kg of helium produced. World energy consumption is 1.6 exajoule per day[2] so less than 4 kg helium will be removed per day if energy extraction is perfect or few 10s of kg assuming imperfection
Your energy estimate is off, but also ~10kg of helium is ~56 cubic meters or ~2,000 cubic feet that’s several large helium tanks which run around 291 cubic feet as that’s a serious pain to move around without a hand truck.
Actual fuel fuel estimate: “a 1000 MW coal-fired power plant requires 2.7 million tonnes of coal per year, a fusion plant of the kind envisioned for the second half of this century will only require 250 kilos of fuel per year, half of it deuterium, half of it tritium.” https://www.iter.org/sci/FusionFuels
“Global electricity consumption in 2019 was 22,848 terawatt-hour”
22,848 * 1000 / 365 / 24 = 2608 different 1GW reactors each producing 250kg of helium per year. So 652,000 kg/year if all the worlds electricity was made from fusion or ~3,650,000 cubic meters or ~130,000,000 cubic feet of helium.
PS: Efficiency numbers could wildly change those estimates, but that’s the rough ballpark for electricity let alone stuff like transportation or home heating etc.
You are talking if helium is stored in atmospheric pressure. It is generally sold compressed like all other gas. Search google for 10kg helium tank. In my area it is available for $30.
That’s the weight of the tank not the helium inside it. They should advertise it as filling ~30 helium balloons. Which obviously weigh far less than 10kg or they wouldn’t float.
The ~300 cubic foot tanks are 9 inches in diameter, 55 inches tall, with about 130 pounds and contain about 1.5kg of helium.
Missing, arguably the most important factor: It is not a runaway reaction. You don't need a giant pool of water with functioning pumps to mitigate a disaster. There are some newer designs that are self-regulating but surprised to see this is not at the top of the list.
It's not possible to have a runaway reaction in a light water reactor either. The major incidents (TMI and Fukushima) happened as the reactor cooled down. It's just that the power output of the decay heat is enough to cause problems.
Heavy water is not a radiological hazard, but it's toxic [1]. I doubt there have been definitive experiments on humans (for obvious reasons). I wouldn't try substituting it for ordinary water.
It is toxic when you ingest >20% of your body weight. Let's just say that that sort of thing won't happen by accident.
(And to point out the obvious, every other liquid apart from drinking water is more toxic when you ingest literal bucketfuls, including harmless household liquids like vinegar, shampoo, ethanol, olive oil.)
I think I heard on a podcast the other day that a 10ft x 10ft tank of heavy water would produce the same amount of energy the entire world consumes in a year (or some crazy stat like that).
> - Since the process produces magnetically charged plasma, steam turbines are not necessary, a solution of directly harvesting energy with electromagnets was proposed.
But wouldn't that require aneutronic fusion to be viable? I had thought that the prevalence of neutron radiation otherwise would have made directly tapping the plasma for electricity impractical.
Yes. But mostly aneutronic fusion is possible, in principle.
"Side reactions" produce neutrons and gamma rays, and fusion products can get involved in side reactions too, also producing neutrons and gamma rays. If you can keep recirculating the desired reactants, filtering out the products and side products, those reactions can be kept to a low level.
As the other commenter mentioned, fuel is limitless and the waste product doesn’t need special disposal. This isn’t just a technical advantage over fission, but also a political one.
In its current form though, the CapEx for fusion projects is huge. Fusion won’t play a big part as an energy source if reactors take 10+ years to come online and cost tens of billions of dollars. Fusion proponents will argue that costs will come down, but if the political and entrepreneurial pressure to reduce cost isn’t there, fusion will end up the same as fission.
As you say, fusion technology inherently has less need for strict regulation. Less harmful waste. Less runaway reaction. Less arms proliferation.
This potentially creates room for fusion that never existed for fission:
- innovation and entrepreneurial pressure. Best case, we get SpaceX efficiency fusion innovators versus SLS pork barrel subsidy patients.
- insurability. Fusion energy production might become privately insurable.
- political support. If the political buyin required is mainly limited to capex, that's a very big advantage versus fission.
As for short to medium term political motivation, one might hope fusion energy research is to benefit from the tension between Russian and western leadership.
On the political motivations: one can hope. My view of things at this point is that those with power in the global playing field want to maintain it. Both Russia and the US supply the world with a lot of oil. This grants a lot of political power. If you make oil obsolete then you remove your lever.
This sounds like a great thing for humanity, but humanity's interests are not properly represented by the existing political power structures.
Thanks. So it sounds like you're saying that, unless there is some radically different plant design in the future, that there won't be an economic advantage to fusion over fission, at least in the near term.
Well fission energy would be much cheaper if most of our energy was produced that way. Economies of scale provide huge opportunities that are currently limited by political will. I don't see a reason fusion can't quickly become a large scale source of power, because unlike fission, there would be much less political backlash.
I am sad to report that in South Korea, which produces fission power very well and does excellent fusion research, the leading anti-nuclear activist (also a member of congress) is vocally against fusion and already proposed to cut 100% of fusion research budget in 2018. (People thought that's too extreme so she haven't repeated it so far, but I don't think her position changed in any way.)
Being against wasting tax money is a principled stance. There is no possibility of current work leading to practical power on a time scale of less than multiple decades, and even then only if a mostly aneutronic form is achieved, which very few are working on.
Does fooling around with Tokamaks and stellarators count as scientific research?
Certainly, containing fusing plasma magnetically requires developing a much better understanding than we now have of plasma fluid dynamics under conditions of high pressure and complicated, variable magnetic field geometry, but studying plasma fluid dynamics for itself does not need multi-billion dollar equipment. Nobody wants to give plasma physicists a dime just to learn about plasma, even though it would be fantastically cheaper than what is being done instead.
Right, the current crop of Tokamak designs getting the lion's share of funding would cost much more than 10x, per watt out, than fission, and the reactors would cost that much more to build, besides. Since fission is uncompetitive, that kind of fusion is completely impractical.
Such reactors would also quickly destroy themselves, so would have no opportunity to pay back the investment.
The only hope for useful fusion is if work on existing designs turns out to be applicable to actually practical, aneutronic forms. Current spending on those is negligible.
Thanks for actually answering my question! I guess no one wants to be a downer, but I'm pretty shocked that no one else in this thread, or that I've talked to about this issue, or anything I've read about fusion, has mentioned that current designs aren't on track to be remotely cost effective, which is what you seem to be saying.
Most people think the fusion idea is cool enough to be enthused over, and assume all the deep technical difficulties, beyond actual fusing itself, will evaporate, because that would be cool too.
Consider that the majority of startup companies with apparently good ideas never get to market. People like to think the hardest problem they know of right now is all that matters, but often it is a boring problem that sinks the company. Maybe it is technically solvable, but costs enough to destroy the value proposition.
So long as solar and wind costs are still falling fast, any prediction about the viability of competing tech is at best provisional.
Solar and wind benefit from the opposite effect: there are known problems, but they have lots of known viable solutions that are just competing for which ones (plural) will end up cheapest, or have the most side benefits, or are easiest to deploy.
It used to be that poor round-trip efficiency would sink a storage technology, but generation has become so cheap that losses matter less than other considerations. 50% loss? Build out more panels!
Hydrogen still has awful efficiency, electrolysers and fuel cells need platinum-group metals, and liquid hydrogen needs really rigorous handling, but H2 is so useful that those don't matter. Efficiency and cost will only improve. Ammonia synthesis is similar: super-useful, but maybe easier to make starting with water. (There will be a lot of waste oxygen soon.)
Iron-air batteries likewise have poor efficiency, and low discharge rate, but the material basis is very, very cheap. You can gang up thousands in parallel to get the rate you need, and stick lithium or lead cells on the front to handle load spikes. Useless for cars, fine for utilities.
Liquifying air is very mature tech, so as the basis for a storage medium it's a safe bet. Storage capacity grows with cheap tankage. And, excess LN2 is valuable, so when your tanks are full you still have revenue.
It turns out there are myriad elevated basins that would be perfect for pumped hydro storage, another very mature technology. Unlike hydro generation, you don't need a whole watershed and river valley, just hills with a dip.
Even if fusion fizzles for utilities, the mostly-aneutronic sort might be perfect for outer solar system propulsion, where a completely different set of constraints apply. And, military deployments will often not be able to lay out much solar where they land. So, even though fission is, relative to solar, super-expensive, places can be found where nothing else will do.
You make an interesting point and I am myself absolutely not knowledgeable in that field but there are some obvious points that you elude in your response.
Fission (will?) have the advantage of being able to produce energy whenever we need it. Solar or wind need storage. And as far as I know, there are no viable storage solution as of today.
> It turns out there are myriad elevated basins that would be perfect for pumped hydro storage, another very mature technology. Unlike hydro generation, you don't need a whole watershed and river valley, just hills with a dip.
In some countries, all possible places of hydro generation have already been used. What would be the potential for new hydro solutions in Europe for example, where hydro has been exploited for decades? Again, not an expert, but I don't think that there is an obvious path in countries where population density is pretty high and without large swathe of lands, to build hydro storage to be able to produce enough energy on a sufficient long period of time. Curious to know what you think.
Using any sort of battery, based on a chemical process, will also probably have high impact on the environment. Current battery relies on rare earth material or industrial processes that are very impactful. Creating enough batteries to ensure safe power distribution for billions of people will probably be terrible for the environment.
My point is that there is no silver bullet as of now so putting all you eggs in the same basket does not seem to be a sane strategy. Investigating fusion is worth a shot I think.
And if we worry about money, there's plenty of money to go around. We are talking about the survival of civilization here. 16 billions were poured into the a company providing ways to share pictures of your baby to your high school friends ten years ago (yes Facebook). I am sure we can find the money to finance Fusion AND research on energy storage. It's a question of political will. In the end, we'll get what we deserve...
Thank you. This is already four years old. Their system seems to have a constructed tank, rather than relying on a natural basin. That is fine when storage need is limited; and tankage is relatively cheap, as construction goes.
Where an elevated natural basin can be found, that can radically increase the storage capacity from hours to, potentially, weeks, and for even less expense: just the penstock needs to be built. Elevated basins are much more common than the elevated river valleys needed for pure hydro generation. A hybrid approach is to wall up one end of an elevated box canyon: a dam, technically, but inflow is pumped from below rather than drainage from above. There is some construction cost, but radically less per unit volume of storage than a complete tank. Dams with penstocks are extremely mature tech.
Is that to run or to build currently? And is this not due to the fact that fission is an exisitng technology and fusion new? The new technology would be much more expensive until it becomes know and then it the price comes down as they understand how to build it more effiently.
There is no reasonable expectation that Tokamak fusion operating or capital cost will fall much.
While fusion doesn't need containment for radioactive fuel, a reactor must be much larger than a fission plant because the volumetric energy flux density of fusing plasma is enormously lower than of uranium. And, the heat has to be collected by blasting neutrons right through the magnetic coils and into a "blanket" of thousands of tons of molten, radioactive lithium, in pipes all around, the which does need to be contained. People get testy when that much molten, radioactive lithium runs downhill.
The lithium needs to be confined in plumbing which will be weakened by the neutron blast and need to be replaced frequently, but will be deadly radioactive so need to be replaced using robots. That plumbing is really most of the reactor. Probably it should all be underground, so that when it is seen to cost more to refurbish every couple of years than it is worth, it is already buried.
Meanwhile, the lithium needs to be processed continuously to extract transmuted tritium to use for fuel. It is hard to imagine molten lithium processing being as cheap as managing the water and steam in a fission reactor.
Then, you need to move the heat from the lithium into liquid that will not pick up its radioactivity, thence to water for steam for the turbines. And, you need to maintain the steam turbines frequently, same as in a fission reactor.
Contrast this to negligible upkeep cost for solar and wind, which mostly amounts to unbolting and replacing them as they pass two decades of service. Your storage method might need some upkeep, but you chose it for its low cost.
Fusion definitely has long term economic advantages over fission. The fuel is the most plentiful in the universe and the energy released from fusion is much higher than fission.
Okay, but the other commenters are saying that fuel is around 6% of the cost of running a fission plant. Do you really think fuel availability will be a decisive factor in say, the next 60 years?
Fission is already not competitive, so a technology much more expensive to build and operate is even less so. In ten years, solar and wind will be even cheaper than today, and will be supplying most of our energy needs.
Fission is non-competitive only because governments don't want it to be. France's nuclear fleet built in the 60s/70s with more primitive tech has capital/operating costs of 7c/kWh, which is fairly similar to commercial solar or wind in France - and is the cheapest option once you double the cost of the latter to account for extra storage/smart grids/other intermittency handling.
(Not every place is Southern California where there are no clouds and the sun shines all year. Solar is cheap in some places, but not at high latitudes.)
Fission is not competitive because of high operational expenses, very long construction, and very high capital commitment, none of which will follow plummeting solar/wind/storage costs. So it gets less competitive with each passing day.
All of these points are up for debate of course, but:
* compared to fission, the nuclear waste management for fusion looks to be done within a human generation, rather than outlast human civilisation. So how you discount the future has a large impact on the trade-off between fusion and fission.
* the fuel for fission is uranium, a hard to get and limited resource. The fuel for fusion is hydrogen. Initially only the rarer hydrogen isotopes, but we might eventually get fusion to work for the more common isotopes as well.
The fuel for DT fusion is heavy water and lithium. Everyone leaves out the lithium part for some reason. They are considerably more rare than hydrogen. But more common than uranium for sure.
I'm a proponent of fission so obviously I don't see the lithium thing as a huge problem. It's just that I don't like the "it runs on just water" takes. Honesty works better in the long run.
Okay, but is the fuel a major cost for fission plants? And it's not clear to me how nuclear waste can both be extremely long-lived and also very dangerous that entire time.
Fuel is not a major cost right now, but it inevitably will be at some point. Current estimates are that we have 80-250 years of uranium left [0], so we could use a different energy source for the long run.
The difference in radioactive danger, is because not all nuclear reactions are treated equally. Tritium decays once, emitting a 0.019MeV beta particle [1]. Uranium creates an avalanche of particles [2] for a total of 51.7 MeV. So it has the leeway to be both more radioactive and to be it for longer. But both the type and energy of the radiation have intricate effects on how they interact with biology and so how dangerous they really are.
That said, exactly how radioactive fusion waste will be is a bit a philosophical problem, as nobody knows what the minimal requirements are for a functional plant.
> As of 2017, identified uranium reserves recoverable at US$130/kg were 6.14 million tons (compared to 5.72 million tons in 2015). At the rate of consumption in 2017, these reserves are sufficient for slightly over 130 years of supply. The identified reserves as of 2017 recoverable at US$260/kg are 7.99 million tons (compared to 7.64 million tons in 2015).
I wouldn't be too worried right now though, because Table 1.1 in uranium-2018 shows that the known recoverable uranium sources @ $40/kgU grew 50% between 2015 and 2017, and known recoverable uranium sources @ $80/kgU grew 4.6% over the same time period.
No one's going to go around prospecting unless they think they can make a profit doing it.
Finally some numbers! Thank you. So again, this sounds like more evidence against an economic advantage of fusion over fission, is that correct in your opinion?
> And it's not clear to me how nuclear waste can both be extremely long-lived and also very dangerous that entire time.
This is a really good question, and as someone who knows nothing about this, here's what I've found:
There's 4 kinds of nuclear waste:
- Very low-level waste
- Low-level waste
- Intermediate-level waste
- High-level waste
You just dump very low-level waste into a landfill[1]. This stuff is basically random concrete, etc. that comes from demolishing a nuclear power plant.
Low-level waste is still pretty boring stuff like clothing and rags that got irradiated somehow, and is 90% of the volume and 1% the radioactivity of radioactive waste[1]. The radioactivity in this mostly comes from atoms with a half-life of less than 5 years, although it seems like trace amounts of slightly long-lasting radioactive isotopes are allowed[2].
Intermediate-level waste looks like it's pretty varied things: sludges, fuel cladding, reactor parts from decomissioning. It's 7% of the volume and has 4% of the radioactivity[1]. Looks like this stuff is generally pretty long-lived: takes about 1000 years to become 10x as radioactive as low-level waste, and 100k years to become as radioactive as low-level waste.
High-level waste is basically spent fuel. It's special in that it requires some kind of cooling, at least for a few decades. It's 3% of the volume and 95% of radioactivity[1]. It can often be re-processed into more fuel. After about 200 years, it becomes about as radioactive as intermediate-level waste, and 100k years to become as radioactive as low-level waste.
Nuclear waste is not actually that dangerous that long. Everyone (including the planners) like to hype it up for their own reasons.
Fusion has many fusion pathways, some of which are pretty light on the dangerous radiation, others that are pretty heavy on it. Most advocates don’t know which one they’re advocating for.
The radiation ‘activates’ and damages the interior of the fusion reactor, making it radioactive. This is not a solved problem yet. It may never be.
> Nuclear waste is not actually that dangerous that long. Everyone (including the planners) like to hype it up for their own reasons.
Can you elaborate? E.g. Plutonium-239 has a half life of 24000 years. That hardly sounds like "not that long"? Leftover uranium-238 even stays radioactive for billions of years (half life of 4.5 billion years).
The more radioactive it is, by it’s very nature, the shorter it’s half life. That is literally what is going on. Decay means release of radiation. Decay means less of the original radioactive material.
Some of these elements have pretty high overall energy levels released in their decay chains (so it’s not just one decay) some less - but sources that are more radioactive are decaying and releasing energy faster, have shorter half lives, and are very dangerous for shorter periods of time.
You can literally buy Uranium 238 ore through the mail and handle it with no more special precautions than washing your hands afterwards and not eating it. It’s seriously fine.
Except in a few spots where it was heavily concentrated, most of the legitimately dangerous stuff has decayed to ‘meh, not that bad’ levels already even at Chernobyl.
It’s still not a good idea to lick it, or spend all your time in the main reactor hall, but give it another 50 years and you’ll probably be able pet the elephants foot on a tour.
Hiroshima and Nagasaki has been fine for awhile. People do tours at ground zero of the Trinity test site.
Some of these elements are chemically active in weird ways, and even without the radioactivity, eating plutonium, cesium, or uranium will be a bad time. Same with mercury, lead, cadmium, etc. so I’m not advocating for being careless with them.
But the idea of a big chunk of radioactive waste being a glowing orb in 10k years is fantasy.
Plutonium, anyway, is not waste, but fuel. And weapons material.
But there are other long-lived isotopes, as well. Safe disposal is not a difficult technical problem, but is a political football causing limitless distraction.
Yes, fusion is considerably more capital intensive than fission power, but the fuel is just water. You don’t need to worry about digging the fuel out of the earth, maintaining a costly disposal scheme for hazardous waste, etc.
Nuclear byproducts storage (let's not call it waste since MOX fuel is the future for humanity) do not cost much, it's not a point that is salient enough to male fusion relevant in any way.
Thanks, but is the fuel cost a significant factor in fission plants? And the waste disposal? There's already lots of radioactive material in the ground, after all, for example in uranium mines.
Fuel cost is not significant, waste disposal even less so. (In fact, usual accounting of nuclear fuel cost includes waste disposal cost. People argue cost is not accounted in full, but as accounted, waste disposal typically costs 1/6 of fuel cost.)
In terms of cost, 80% of nuclear power is CAPEX. Among OPEX, about 1/3 is fuel, rest is operation and maintenance. So from total cost, about 6% is fuel, and about 1% is waste disposal.
Still, fission operational expense is very, very large when compared to solar and wind, which also have radically lower capital cost. Operational expense of neutron-emission fusion would be radically more than for fission: not for the fuel, but for everything else.
Building one power plant is incredibly expensive. Building it again costs less. Building it 1000 times costs much less per plant. We're not building thousands of fission plants because of politics. Fusion might be an easier sell.
This isn't really true. Yes first of a kind is expensive and there is learning curve, but learning is serial and iteration cycle is long. You can't build 1000 plants in parallel and learn 1000x. Solar power cost is falling fast because iteration cycle is short.
South Korea built 28 reactors over 36 years and cost fell 1.5% per year, totalling 40% reduction. Even with this South Korean nuclear power is only about as half cheap as solar power locally (as of 2020), and South Korea is a very poor country for solar power considering its latitude and weather.
ITER plans to test breeding tritium from lithium. It should work in theory, but there is no practical experience yet. Current tritium used in fusion research is entirely produced from fission reactor moderated by heavy water, and there is only one production facility in the world at Canada.
Which will never work because of conduction losses in the coils. You can't just put solid materials inside your confinement volume and expect to make power.
But, even less is being spent on that than on other aneutronic designs, which themselves are on shoestring budgets. Theoretical reasons make it seem like an unpromising avenue.
Confinement approaches aren't inherently neutronic or aneutronic. What matters is the temperature, desnity, and confinement time (lawson criterion). Get it sufficiently high then DT becomes a breeze. Get it a factor of 500 higher and pB11 is on the table.
No one serious talks about aneutronic fusion because we need to walk 1 mph before we sprint at 500 mph. We'll seriously discuss aneutronic fusion in 200 years when it's relevant.
There are no proliferation concerns with fusion. You can't make a nuclear bomb with a fusion reactor. They'll be largely free of the nuclear materials and technologies handling and export regulations. Quite a few international deals for fission plants over the years have been scuttled over concerns about misapplication of the reactors or fuel.
You need to start with fissile material to make weapons grade fissile material. If there is no peaceful application for fissile material then a global ban treaty complete with trust-but-verify policies is within the realm of reason.
Regardless of this hypothetical: the fact remains that no part of a fusion reactor increases weapons proliferation risks, unlike a fission reactor. You could plop one down anywhere on the planet and locally source fuel.
If you want a neutron source then you can make one in your basement with a fusor or linear magnetic mirror.
One is not obliged to turn ordinary, mined uranium into weapons-grade stuff, in a neutron-emitting fusion system, but you could if you cared to.
I.e., a nation that had control of such a fusion plant would have little difficulty attaching an enrichment process without interfering with power output.
Fortunately, no economically practical power generation system can be built using hot-neutron fusion, so it is an idle concern, but almost all the money being spent on fusion pursues that impractical goal.
You can make a centrifugal magnetic mirror the size of a truck for a few million bucks that will weapons grade as much fissile material as you want. It doesn't take new science or technology to do. If you want to keep weapons grade fissile material out of someone's hands then your best hope is preventing them from having fissile material (or at least controlling every pound of the stuff).
Why is Iran running hundreds of centrifuges if they could cheaply use one of these things to make it into plutonium and separate that out by trivial chemical processes?
Maybe because it's the US cold war blueprint and doesn't require any innovation. To be clear: no public program has used magnetic mirrors for enrichment but they are so cheap/small and so hot that they really are a sure thing. Maybe MRI magnets are difficult to source. I feel like the true answer is not rooted in any technical explanation but rather some kind of political explanation.
If I was a warlord trying to make industrial quantities of weapons grade fissile material in 2022 I personally wouldn't go the centrifuge route, but everyone's a critic.
The most intense non-fission, non-fusion neutron source in the world in 2017, Oak Ridge Spallation Neutron Source, produced ~1e10 neutrons / second. If every last neutron emitted could be cooled and made to be absorbed by a ready U-238 nucleus, decaying after a few days to Pu-239, if I figure right we ought to get a microgram in 28 days, or a Kg in 78My. Pu-239's half-life is only 24Ky, decaying to (equally useful) U-235 with half-life a solid 0.7Gy, so only a little bit turns to lead before we finish.
Keeping our copy of the Oak Ridge source operating, and the current world order destabilized, for those 78My seems difficult; and we anyway have made only a Kg. So, this seems like not a practical way to generate a geopolitically effective amount of Pu-239 or U-235.
Magnetic mirrors are fusion devices, but they are pulsed and have conduction losses due to the central electrode. This electrode greatly increases confinement performance but is obviously an impossibility for power generation. Where magnetic mirrors really thrive is in having very hot plasma in relatively small devices [0].
So the idea is not to use magnetic mirrors for power generation, but as a DD-fusion-powered cheap fast neutron source. I have a hunch that radioisotope companies and government organizations are doing this right now.
I'm reminded of Alastair Reynolds' sci-fi short story "Weather"
<spoilers follow>
The twist of the story is that the alien/incomprehensible starship drives are controlled by an embedded, augmented, disembodied human brain. I've long felt the gimmick was a bit silly because what system would require an actual human brain over a normal control algorithm.
Saw the headline, immediately thought of the same short story. Excerpts where they encounter said brain:
"It's a monstrosity."
"Not to us," she said sharply. "We see a thing of wonder and beauty."
"No," I said firmly. "Let's be clear about this. What you're showing me here is a human brain, a living mind, turned into some kind of slave."
"No slavery is involved," Weather said. "The mind chose this vocation willingly."
"It *chose* this?"
"It's considered a great honour. Even in [our] society, even given all that we have learned about the maximization of our mental resources, on a few are ever born who have the skills necessary to tame and manage the reactions in the heart of a [star]-drive."
…
"The degree of concentration is quite intense. He can barely spare any resources for what we might call normal thought. He's in a state of permanent unconscious flow, like someone engaged in an enormously challenging game. But now the game has begun to get the better of him. It isn't fun anymore. And yet he knows the cost of failure."
…
"It'll be more like one very long dream. Someone else's dream, certainly, but I don't doubt that there'll be a certain rapturous quality to it. I remember how good it felt to find an elegant solution, when the parameters looked so unpromising. Like making the most beautiful music imaginable. I don't think anyone can really know how that feels unless they've also held some of that fire in their minds. It's ecstasy, Inigo, when it goes right."
"And when it goes wrong?"
Chris Bishop and others were working on this with neural nets for the Tokamak at Harwell back in the 1980s which illustrates just how difficult this problem is.
The best part was the name of our collaborator- "General Atomics" https://en.wikipedia.org/wiki/General_Atomics
Here's a writeup, https://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.58...