Editing Tokamak (section)

=== Other issues ===
Early tokamaks had an ion temperature limited by the empirical Artsimovich formula:<ref name="y331">{{cite journal |last=Strelkov |first=V.S. |date=1985-09-01 |title=Twenty-five years of tokamak research at the I.V. Kurchatov Institute |url=https://iopscience.iop.org/article/10.1088/0029-5515/25/9/033 |journal=Nuclear Fusion |volume=25 |issue=9 |pages=1189–1194 |doi=10.1088/0029-5515/25/9/033 |issn=0029-5515 |access-date=2025-04-15}}</ref>

<math>T_i = 5.9 \times 10^{-7} (n_e I B_t R^2 A^{-\frac{3}{2}})^{\frac{1}{3}}</math>

While the tokamak addresses the issue of plasma stability in a gross sense, plasmas are also subject to a number of dynamic instabilities. One of these, the [[kink instability]], is strongly suppressed by the tokamak layout, a side-effect of the high safety factors of tokamaks. The lack of kinks allowed the tokamak to operate at much higher temperatures than previous machines, and this allowed a host of new phenomena to appear.

One of these, the [[neoclassical transport|banana orbit]]s, is caused by the wide range of particle energies in a tokamak – much of the fuel is hot, but a certain percentage is much cooler. Due to the high twist of the fields in the tokamak, particles following their lines of force rapidly move towards the inner edge and then outer. As they move inward they are subject to increasing magnetic fields due to the smaller radius concentrating the field. The low-energy particles in the fuel will [[magnetic mirror|reflect]] off this increasing field and begin to travel backwards through the fuel, colliding with the higher energy nuclei and scattering them out of the plasma. This process causes fuel to be lost from the reactor, although this process is slow enough that a practical reactor is still well within reach.{{sfn|Wesson|1999|pp=15–18}}

Another instability is tearing instability. In 2024 researchers used [[reinforcement learning]] against a multimodal dynamic model to measure and forecast such instabilities based on signals from multiple diagnostics and actuators at 25 millisecond intervals. This forecast was used to reduce tearing instabilities in [[DIII-D (tokamak)|DIII-D6]], in the US. The reward function balanced the conflicting objectives of maximum plasma pressure and instability risks. In particular, the plasma actively tracked the stable path while maintaining H-mode performance.<ref>{{Cite journal |last1=Seo |first1=Jaemin |last2=Kim |first2=SangKyeun |last3=Jalalvand |first3=Azarakhsh |last4=Conlin |first4=Rory |last5=Rothstein |first5=Andrew |last6=Abbate |first6=Joseph |last7=Erickson |first7=Keith |last8=Wai |first8=Josiah |last9=Shousha |first9=Ricardo |last10=Kolemen |first10=Egemen |date=February 2024 |title=Avoiding fusion plasma tearing instability with deep reinforcement learning |journal=Nature |language=en |volume=626 |issue=8000 |pages=746–751 |doi=10.1038/s41586-024-07024-9 |pmid=38383624 |issn=1476-4687|pmc=10881383 |bibcode=2024Natur.626..746S }}</ref><ref>{{Cite web |last=Ate-a-Pi |date=February 26, 2024 |title=Deep learning fusion breakthrough |url=https://x.com/8teAPi/status/1762190250355658885?s=20 |website=X}}</ref>