Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

AI for real-time fusion plasma behavior prediction and manipulation

control.princeton.edu

Post reply on HN
Print view View original

151–160 of 167 posts

Re: AI for real-time fusion plasma behavior prediction and manipulation

#151

Post by DoctorOetker »

fragmede wrote: post #112
ben_w wrote: post #98

Earlier quoted context omitted.

> Deuterium is also not renewable, even if it is more abundant than uranium Technology correct, in that after around a hundred trillion years even the red dwarf stars will have stopped burning hydrogen. But last I checked as yet there is no known way to harness the only (and even then merely suspected) infinitely renewable energy source: the expansion of the universe.

Just tie the end of an infinitely long string to the edge of the universe and have it pull on a generator to spin it.

I would certainly like to see serious critical analysis and calculations of such hypothetical setups by physicists.

Would such a setup slow down the local expansion (action and reaction)?

Since iron is essentially a nuclear ground state, a steel cable being lengthened seems like the least worse mass loss imaginable.

Re: AI for real-time fusion plasma behavior prediction and manipulation

#152

Post by fragmede »

koe123 wrote: post #144
fragmede wrote: post #113

Earlier quoted context omitted.

> Many on Hacker News fantasize about fusion (not fission) reactors. These fusion reactors will be an intense source of fast neutrons. All the hardware in a fusion reactor will become radioactive. Not to mention the gamma rays. My personal ideology about fusion aside, it should be mentioned there is an easy fix for these radiation problems. What you do is put the fusion reactor in space, and collect the energy with s…

For such an approach I’ve always thought it seemed a risk to launch something like that into the air. E.g. what happens if the rocket explodes while taking off? Or something bad happens when in space? Will it rain nuclear material?

Obviously you just put the nuclear material inside of the in-flight data recorder so it will survive a rocket failure.

If you're being serious, Cassini had those kinds of questions with its launch about its RTG but that didn't have enough nuclear material for it to be a problem.

If we were to try and use a fusion reaction in space, we'd probably use the existing one.

Re: AI for real-time fusion plasma behavior prediction and manipulation

#153

Post by nightowl_games »

rnhmjoj wrote: post #139
nightowl_games wrote: post #119

Earlier quoted context omitted.

Isn't there a lot of daylight between old Fortran code and AI? What if we rewrote the old algorithms in C with modern techniques? Multitthreading? Or GPU compute? If there's value there, I could do these things. Probably wouldn't take that long

Rewriting the old Fortran code in C will probably make it slower with new bugs. A smarter thing to do when picking up terrible code written by physicists is to document everything you can, write tests and then start refactoring bit by bit using modern Fortran features (yes, the latest standard is 2023). Fortran compilers had more than 40 years to become pretty good at generating efficient code; they can make assumpti…

> Rewriting the old Fortran code in C will probably make it slower with new bugs.

Then it's not done yet.

Re: AI for real-time fusion plasma behavior prediction and manipulation

#154

Post by varjag »

lazide wrote: post #148
varjag wrote: post #147

Earlier quoted context omitted.

Perhaps in 2024. Not in 2004.

since the 90’s [ https://en.m.wikipedia.org/wiki/Convolutional_neural_network ]

I know the history and the fist USPS applications. My point is in any random industrial vision setup on the verge of millennium you'd unlikely found any sort of NN.

Re: AI for real-time fusion plasma behavior prediction and manipulation

#155

Post by ben_w »

DoctorOetker wrote: post #149
ben_w wrote: post #133

Earlier quoted context omitted.

I am already aware of those things. On earth, there is an estimated 4.85×10e13 tonnes of deuterium; the energy density is 3.4x10e14 J/kg, giving a total yield of 1.649e31 joules. If you deleted the sun, this would be sufficient to maintain the current temperature of the Earth for ~9.5 million years: https://www.wolframalpha.com/input?i=%281.649×10%5E31+joules... At "merely" the level of current human power consumptio…

Do you have an opinion on using N2O (laughing gas) as an energy carrier? 2 molecules of N2O exothermically react to form 2 x N2 and 1 x O2 molecules, approximately the same composition as our atmosphere. It is a very potent greenhouse gas, so quite disturbing on that front. I've been making calculation for designing earth suits, where the suit replaces the home, internal showering, ventilation, heat recovery etc. Usi…

I have no opinions about most chemistry. I do know that the heaviest recreational users develop issues due to it being neurotoxic, so that's worth considering.

I don't know if your designs are technical or world-building for a story? If the latter, I'd suggest https://worldbuilding.stackexchange.com as I've had good conversations there, if the former perhaps (but not as a recommendation because I'm not a chemist) https://chemistry.stackexchange.com would help?

Re: AI for real-time fusion plasma behavior prediction and manipulation

#156

Post by lazide »

varjag wrote: post #154
lazide wrote: post #148

Earlier quoted context omitted.

since the 90’s [ https://en.m.wikipedia.org/wiki/Convolutional_neural_network ]

I know the history and the fist USPS applications. My point is in any random industrial vision setup on the verge of millennium you'd unlikely found any sort of NN.

What would you find instead?

Re: AI for real-time fusion plasma behavior prediction and manipulation

#157

Post by varjag »

lazide wrote: post #156
varjag wrote: post #154

Earlier quoted context omitted.

I know the history and the fist USPS applications. My point is in any random industrial vision setup on the verge of millennium you'd unlikely found any sort of NN.

What would you find instead?

https://en.wikipedia.org/wiki/Feature_(computer_vision)#Dete...

Re: AI for real-time fusion plasma behavior prediction and manipulation

#158

Post by lazide »

varjag wrote: post #157
lazide wrote: post #156

Earlier quoted context omitted.

What would you find instead?

https://en.wikipedia.org/wiki/Feature_(computer_vision)#Dete...

None of those work for many common cases - which is why CNN’s revolutionized actual use of computer vision.

Re: AI for real-time fusion plasma behavior prediction and manipulation

#159

Post by chriskanan »

noobermin wrote: post #89
chriskanan wrote: post #74

There is a lot of AI research in the nuclear fusion space. For inertial confinement fusion (a competing technology to magnetic confinement fusion, e.g., tokamaks) the National Ignition Facility (NIF) used it for their experiment that resulted in "ignition." My lab is collaborating with researchers at the Laboratory for Laser Energetics to use AI to improve inertial confinement fusion (ICF). We recently put out this p…

Here we go with the CS people saying "the old fortran codes are terrible." Yeah, the "high error" between code predictions and laser shots is because LPI is inherently noisy and it's essentially impossible to fully control conditions. I would work on expensive sims for days to weeks and the experiments would see differences that would be off from that from an order or mag because their focal point is off by a micron.…

Based on what the physicists have told me, the main issue with the Fortran code is that it is SLOW and hard to accelerate. Because the codebase is 30+ years old, much it sounds like it isn't well understood. They think it would benefit a lot from things like GPU acceleration, but that's hard to do with it in its current state.

Otherwise, the major issue is just that it makes a lot of assumptions and has a lot of inaccuracies, but that issue isn't due to Fortran but just because it doesn't capture all of the physics and there is a sim2real gap. So it DOES NOT have the right physics, but re-writing it probably wouldn't result in the "right" physics either.

The main version of the code they run uses 1-D physics (so one spatial dimension), which takes about 8 minutes to run on a modern CPU. The code can't be parallelized without being re-written. Running the 3D version of the code takes 1+ weeks for a single run, and that still has a big sim2real gap.

Re: AI for real-time fusion plasma behavior prediction and manipulation

#160

Post by chriskanan »

nightowl_games wrote: post #119
chriskanan wrote: post #74

There is a lot of AI research in the nuclear fusion space. For inertial confinement fusion (a competing technology to magnetic confinement fusion, e.g., tokamaks) the National Ignition Facility (NIF) used it for their experiment that resulted in "ignition." My lab is collaborating with researchers at the Laboratory for Laser Energetics to use AI to improve inertial confinement fusion (ICF). We recently put out this p…

Isn't there a lot of daylight between old Fortran code and AI? What if we rewrote the old algorithms in C with modern techniques? Multitthreading? Or GPU compute? If there's value there, I could do these things. Probably wouldn't take that long

This is what they would like to do, but it is hard to get funding for such an effort or to prioritize such work. It sounds like they think it would take a long time, and it wouldn't yield high impact papers from just doing that exercise.

Plus, even after doing that, there would still be a sim2real gap. The goal of our research is to use physics informed deep learning and methods with strong inductive biases, combined with transfer learning and low-shot learning to overcome the sim2real gap.

Post reply on HN