AI Framework Controls Fusion Plasma in Milliseconds
Princeton Plasma Physics Laboratory and Princeton University researchers have demonstrated PACMAN, an AI software framework that monitors and controls tokamak fusion plasma on millisecond timescales. In tests on the DIII-D facility, it predicted a tearing-mode instability roughly 200 milliseconds ahead and adjusted parameters to prevent it, while coordinating multiple heating systems and meeting density and rotation targets. The modular system runs a continuous loop of data collection, prediction, and command issuance far faster than human operators, who respond on the order of seconds.
Fusion plasmas can destabilize in thousandths of a second, outpacing conventional control and rendering full physics simulations too slow for real-time use. PACMAN integrates multiple machine-learning models—including reinforcement learning—into a shared architecture with built-in hardware safety limits. Humans still set high-level objectives and review results after each shot. The design allows new models to be swapped in days rather than months.
Key uncertainties remain around generalization to larger or differently configured devices and long-term reliability under continuous operation. Performance gains must still be balanced against any residual risk of unforeseen plasma behaviors that models have not yet encountered.
Sources: ScienceDaily (Princeton University / Nuclear Fusion), PPPL.
Cellular Enzyme Reads Eight-Letter Genetic Alphabet
UC San Diego researchers have shown that bacterial RNA polymerase accurately transcribes an expanded “hachimoji” genetic alphabet containing eight letters instead of the four used by all known life. Cryo-electron microscopy revealed that the enzyme recognizes two synthetic base pairs through structural and biochemical cues largely shared with natural bases. A companion study found the polymerase can also handle a hydrophobic unnatural pair that lacks conventional hydrogen bonds.
The work demonstrates that existing cellular machinery can process synthetic genetic information without wholesale redesign. Expanded alphabets have already been used to create DNA molecules that recognize specific cancer cells; the new structural data provide a foundation for more complex engineered systems that could produce novel compounds or perform non-natural functions.
Open questions include long-term stability of the synthetic bases inside living cells, fidelity over many replication cycles, and whether the full suite of cellular enzymes (polymerases, ligases, repair systems) will tolerate the expanded code equally well. Scaling from in-vitro transcription to functional organisms remains a substantial engineering challenge.
Sources: ScienceDaily (UC San Diego / Nature Communications, PNAS).
Earth’s Magnetic Field Used as Dark-Matter Detector
A Kyoto University-led team treated the Earth-ionosphere cavity as a planet-scale resonator to search for ultralight axions and dark photons. Analyzing a decade of geomagnetic data, they placed new limits on axion-photon coupling roughly 100 times tighter than prior laboratory bounds in the relevant mass range and identified several narrow-band signal candidates consistent with dark-photon signatures. The theoretical framework extends reliable predictions up to about 30 Hz by incorporating atmospheric conductivity.
Traditional searches rely on strong laboratory magnets over limited volumes. Using Earth’s natural magnetic environment and resonant cavity dramatically increases effective detector size for the lightest proposed dark-matter particles. Axion signals are predicted to vary geographically while dark-photon signals should be more uniform, offering a potential discriminator.
The candidate signals have not been confirmed as dark matter; terrestrial or instrumental origins remain possible. Further multi-site measurements and refined noise rejection will be required to determine whether any excess is astrophysical.
Sources: ScienceDaily (Kyoto University / PTEP, Physical Review D).