Tuesday, November 17, 2026

Non-Local Plasmon Dispersion: When the Local Response Breaks Down

Non-Local Plasmon Dispersion: When the Local Response Breaks Down

In standard macroscopic electrodynamics, the Local Response Approximation (LRA) assumes that the current density J(r) at a specific point depends only on the electric field E(r) at that exact same point. While powerful for bulk metals, LRA breaks down when structural dimensions shrink below 10 nm or when fields vary rapidly across atomic scales (such as in nanogaps, sharp tips, or toroidal vortex cores).

Non-local plasmon dispersion introduces spatial dispersion, meaning the electron gas's response at point r depends on fields across a finite neighborhood around r. This non-locality is governed by fundamental microscopic length scales, primarily the Fermi velocity (vF) and quantum/thermal pressure.

The Physical Mechanisms

Internal Electron Pressure (Fermi/Quantum Pressure): Unlike a classical cold plasma, a degenerate electron gas possesses kinetic energy tied to its density gradient. Compressing electron clouds creates a restoring quantum pressure gradient (∇P or Bohm-like quantum potentials). This acts like an elastic spring constant for the electron fluid.

Wavevector Cutoff & Finite-Size Limits: In local models, the field at a sharp tip or in a vanishing gap diverges. Non-local models introduce a natural wavevector cutoff (approaching ω / vF), beyond which Landau damping and spatial dispersion suppress high-frequency singularity collapse.

Mathematical Formulation in Spectral Space

In hydrodynamic frameworks, non-local response alters Maxwell's equations or the associated velocity-vorticity transport equations by introducing a wavevector-dependent screening kernel. The standard hydrodynamic Drude result is the longitudinal dielectric function

εL(k, ω) = 1 − ωp2 / (ω2 − β2k2),

which gives the longitudinal plasmon dispersion

ω2 = ωp2 + β2k2,   with β2 ~ (3/5) vF2 (Thomas–Fermi).

That is the dispersion relation the hydrodynamic model actually predicts: a plasma frequency plus a pressure term that grows with wavevector. It is the reference against which any simulation kernel has to be judged.

For instance, the relation between the streamfunction ψ̂ and vorticity ω̂ transitions from a pure Poisson equation to a dispersed operator:

ψ̂(k) = ω̂(k) / [ k2 (1 + λp2 k2) ]

The k2 term represents the standard local geometric diffusion and long-range Coulomb coupling. The λp2k2 term is a low-pass momentum filter — it suppresses the streamfunction at high k, because the denominator grows with k. (The original draft called it high-pass; that was backwards, and it matters: the kernel damps short-wavelength velocity, it does not amplify it.)

Consequences for Nanoscale and Toroidal Systems

Blueshifting of Resonance Spectra: Non-local pressure terms provide an extra restoring force, systematically shifting localized surface plasmon resonances to higher frequencies relative to the local model. Qualified, because electron spill-out — which the hydrodynamic model omits — shifts resonances the other way, toward the red. The net shift in a real particle is the balance of the two, not a one-way blueshift.

Suppression of Field Divergences: In extreme sub-nanometer geometries (e.g., dimer gaps), local models predict infinite field enhancement. Non-local dispersion smooths out these artificial infinities by accounting for electron spill-out and pressure dispersion. The hydrodynamic model gives a large-wavevector cutoff inversely proportional to the Fermi velocity, and a large but finite maximal enhancement.

Phase Mixing & Wave Dispersion: Adding λp to a vorticity solver suppresses grid-scale velocity and smooths quantum-hydrodynamic phase mixing. What it does not do, on its own, is introduce a finite group velocity or an ω(k) dispersion relation — the kernel contains no wave equation and no plasma frequency, and a temporal FFT of the simulated field peaks at ω = 0 for most wavenumbers. A genuine group velocity needs a wave mode with ω(k) ≠ 0, e.g. the ω2 = ωp2 + β2k2 above, simulated as a wave, not as advected vorticity.

Appendix: The Simulation Kernel, Tested

The kernel above was implemented in a 2D pseudospectral vorticity-transport solver (N = 128, 200 steps, dt = 0.01, ν = 0.001) and run. Two fatal bugs in the original code had to be fixed first, and they are recorded here because the result depends on them: np.ifft2 does not exist in NumPy (the correct call is np.fft.ifft2), and the wavenumber formula np.fft.fftfreq(N, d=2π/N) * N is wrong by a factor N/(2π) ≈ 20.37 — the correct integer modes are np.fft.fftfreq(N) * N. With the wrong wavenumbers the run overflowed to NaN by step ~20 (ν·Kmax2·dt = 17.0, against a stability limit of ~0.5). With the corrected wavenumbers (ν·Kmax2·dt = 0.041) the run is stable: final max|ω| ≈ 2.34, rms ≈ 1.03.

One more number, because it sets the scale honestly: in the code λp = 0.08 is dimensionless, and the grid spacing is dx = 2π/128 ≈ 0.049, so λp is about 1.6 grid cells. A real Thomas–Fermi length in a metal is ~0.1 nm. The code value is a grid-scale regulariser, not a measured screening length, because no physical units are defined in the material. Treat all quantities as dimensionless unless a material (n, m*, κ, vF) is specified.

Sources consulted in review: the hydrodynamic Drude model and non-local response literature, arXiv:1208.0146 and 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). Draft only — not published.

 

No comments:

Post a Comment