KineticForces - BUG FIX - Coulomb logarithm uses natural log, not log10

[logan-nc]

Summary

The Coulomb logarithm in the kinetic-profile collisionality (src/Equilibrium/KineticProfiles.jl) was computed with log10 where it must use natural log:

# before
ll = 17.3 - 0.5*log10(n_e/1.0e20) + 1.5*log10(T_e/1.602e-16)
# after
ll = 17.3 - 0.5*log(n_e/1.0e20)   + 1.5*log(T_e/1.602e-16)

The 17.3 / -0.5 / +1.5 coefficients are calibrated for natural log — they are the NRL Plasma Formulary electron-ion Coulomb logarithm

λ_ei = 23 − ln(nₑ^{1/2} · Tₑ^{−3/2}) (nₑ in cm⁻³, Tₑ in eV)

re-normalized to GPEC units (nₑ / 10²⁰ m⁻³, Tₑ / 1 keV = 1.602e-16 J): the constant becomes 17.24 ≈ 17.3, and the +1.5 coefficient on ln(Tₑ) is the signature of the Tₑ^{−3/2} regime (impossible with log₁₀). This also matches the Fortran reference pentrc/inputs.f90:238, which uses log() (natural log in Fortran).

Impact

lnΛ, and the collision frequencies nue/nui that scale linearly with it, ran ~13% low at a 20 keV core (19.2 vs the correct 21.7). The error vanishes near the reference point (nₑ=10²⁰, Tₑ=1 keV) and grows with temperature.

Verification

• Formulary + reference: derivation above; matches NRL and Fortran PENTRC.
• Regression (diiid_n1): NTV torque FGAR and NTV kinetic energy dW FGAR shift +0.06% / +0.22% — within tolerance, correct direction, small because that case sits near the log reference point. No other quantity is attributable to the change (the touched code is the collision frequency; the delta-prime/resonant-field scatter is pre-existing threaded-reduction non-determinism).

🤖 Generated with Claude Code

https://claude.ai/code/session_01SsLUydP2gJbE1tGoaHDiS1

[logan-nc]

@claude review this

[claude]

Reviewed the diff. Summary of findings:

Physics correctness — verified. Re-deriving the NRL formulary electron-ion Coulomb log λ_ei = 23 − ln(nₑ^½Tₑ^−³ᐟ²) (nₑ in cm⁻³, Tₑ in eV) and renormalizing to GPEC units (nₑ/10²⁰ m⁻³, Tₑ/1 keV) does reproduce the 17.3/−0.5/+1.5 coefficients under natural log (17.24, matching the PR's math), and the old log10 version only coincidentally agreed near the nₑ=10²⁰, Tₑ=1 keV reference point (where ln(1)=log₁₀(1)=0). At Tₑ=20 keV this reproduces the claimed ~19.2 vs ~21.7 discrepancy. The fix is correct.

Scope/isolation — good. Grepped the repo and this is the only place this Coulomb-log formula appears, so there's no other log10 instance of the same bug lurking elsewhere that needs the same fix.

Regression harness — used correctly per the repo policy; the reported shifts in NTV torque FGAR/NTV kinetic energy dW FGAR are consistent with the fix (small because diiid_n1 sits near the reference point) and no unrelated quantity moved.

Comment — three lines; could be tightened to one per the "concise, one line where possible" guidance in CLAUDE.md, but it's documenting a physics formula (encouraged) rather than session narration, so this is a minor nit, not a blocker.

Test coverage — this is the real gap. There's no unit test that pins down loglam/nue/nui for a known (n_e, T_e) input, which is exactly how a log10/log mixup like this survives silently for however long it did. The full-run regression harness only catches such issues via a knock-on effect a few floating-point-percent large, and only if the case happens to be far from the accidental-agreement point. Recommend adding a small unit test to test/runtests_kinetic.jl (or wherever load_kinetic_profiles is exercised) asserting loglam against a hand-computed value at Tₑ well above 1 keV (e.g. the 20 keV case cited in the PR description) — that would fail loudly under log10 and pass under the fix, and prevent regression of this specific bug independent of the full pipeline.

Minor/optional: the NRL formulary's 23−ln(...) form (used here) is the low/mid-temperature regime; for Tₑ ≳ 10·Z² eV the formulary switches to λ = 24 − ln(nₑ^½/Tₑ). At the 20 keV core cited in the PR description that threshold is far exceeded, so the regime choice itself may be worth a follow-up look — but that's a pre-existing modeling choice this PR doesn't touch, not something to hold up this fix on.