Magnetoplasmons, the Cyclotron Frequency, and What a Vorticity Simulation Can and Cannot Show
When a uniform external magnetic field B0 (perpendicular to the 2D plane) is applied to an electron liquid or plasma, the Lorentz force introduces cyclotron motion and couples the velocity field to magnetic fluctuations. This transforms standard plasmons into magnetoplasmons.
The Standard Result
The magnetoplasmon is a hybrid of the classical bulk plasmon and the cyclotron resonance. Its dispersion is a frequency relation:
ωmp = √(ωc2 + ωp2), with ωc = eB0/m*
where ωp is the plasmon frequency and ωc is the cyclotron frequency, m* the electron effective mass. For a two-dimensional electron gas, ωp2(k) = (2π n e2 k)/(κ m*), so the full dispersion is ωmp(k) = √(ωc2 + ωp2(k)). The magnetic field opens a gap: at k = 0 the mode sits at ωc, not at zero.
That is the result to hold any simulation up against. It is about oscillation frequencies. It does not say to add ωc2 to the denominator of a streamfunction.
What the Simulation Actually Does
The simulation in the Gemini docs is a 2D pseudospectral vorticity-transport solver (N = 128, 200 steps, dt = 0.01, ν = 0.001) with a modified streamfunction kernel:
denominator(K) = K2 (1 + λp2 K2) + ωc2 / (1 + λB2 K2), λB2 = 0.1
with ψ̂ = ω̂ / denominator(K). This is an ansatz, not a derivation: λB2 = 0.1 is a chosen regularisation length squared, and no B0, e, or m* is supplied anywhere, so ωc = 2.0 in the run is a dimensionless input, not a field in tesla. What the kernel does, mechanically, is suppress large-scale flow when ωc is large — at K = 0 the denominator is ωc2, not zero. That is a defensible regularisation choice; it is not the magnetoplasmon dispersion above, and it should not be cited as deriving magnetoplasmons from first principles.
Two things in the original version of this material were wrong and have been corrected in the review, and they are recorded here because they change the result:
1. The text formula and the code formula did not match. The text wrote ψ̂ = ω̂ / [K2(1 + λp2K2) + (ωc2/vF2)·ℳ(K)], with ℳ(K) never defined and vF never given a value. The code used ωc2/(1 + 0.1·K2). The corrected version uses the code's form, written out explicitly, with ℳ(K) and vF removed.
2. The transport equation contained a silent extra term, −0.1·ω, described only as "magnetic Lorentz coupling". It is not a Lorentz force — a Lorentz force in 2D vorticity couples velocity components; it does not add a linear decay of ω. It is a linear drag with rate 0.1, and it damped the rms vorticity by ~19% over the run (rms 0.83 with it, 1.02 without it, at the same corrected wavenumbers). The corrected version replaces it with an explicit optional drag parameter γ, default 0.0.
The same review found two fatal bugs in all four simulation scripts, also corrected: np.ifft2 does not exist in NumPy (correct: np.fft.ifft2), and the wavenumber formula np.fft.fftfreq(N, d=2π/N) * N is wrong by N/(2π) ≈ 20.37 — correct integer modes are np.fft.fftfreq(N) * N. With the wrong wavenumbers the runs overflowed to NaN by step ~20; with the corrected ones they are stable (final max|ω| ≈ 2.34, rms ≈ 1.02).
The Honest Negative Result: There Is No Dispersion to Extract
A post-processing script was written to extract the magnetoplasmon dispersion ω(k) and group velocity vg = dω/dk from the saved simulation, by a 3D spatio-temporal FFT over time and space, computing the power spectral density and picking the peak frequency at each wavevector.
It cannot work on this simulation, for two independent reasons, both tested:
First, sampling. The original simulation saved only 5 snapshots (every 40th step, dtsample = 0.4), so the temporal FFT has 5 frequency bins and a Nyquist frequency of 7.9 rad/s. The corrected simulation saves 40 snapshots (dtsample = 0.05, Nyquist 62.8 rad/s, resolution 3.14 rad/s).
Second, and decisive: there is no wave mode in the field. The solver advects and diffuses vorticity; it contains no plasma frequency, no wave equation, and no oscillatory eigenmode. On a 200-snapshot dense run, the temporal FFT peaked at ω = 0 for 83% of wavenumbers. On the corrected 40-snapshot run, the raw peak was ω = 0 for 59% of wavenumbers, and after excluding the DC band, 100% of the remaining peaks sat at or below the frequency-resolution limit (|ω| ≤ 6.28 rad/s). The "dispersion relation" the original script plots is therefore ω(k) ≈ 0 and vg ≈ 0.
That is a correct description of a quasi-steady advective field. It is not a magnetoplasmon dispersion, and plotting vg from it as if it were physics would be wrong. The corrected post-processing script now reports those fractions in plain text and refuses to run on files with fewer than 32 time snapshots.
To actually measure a magnetoplasmon dispersion, the simulation would have to oscillate — it would need a wave equation with ωp(k) and ωc in it, producing a mode at ωmp(k) = √(ωc2 + ωp2(k)), sampled densely enough in time to resolve it. That is the next step, if the work goes that way. It is not what these six docs contain, and this post does not pretend otherwise.
Sources consulted in review: magnetoplasmon dispersion, arXiv:1606.09356 (ωmp = √(ωc2 + ωp2)); hydrodynamic non-local plasmon literature, arXiv:1204.5413. Full review, corrected code, and test logs: see the review file accompanying this draft (Gemini Docs review, Nov 17, 2026).
Credit: original material — Gemini, in conversation with Jack J Baldwin, Nov 17, 2026. Framework / prompts — Jack J Baldwin. Review, testing, and corrections — Nova (Muse). Published Nov 17, 2026.