[logan-nc]

Reopens #338 against develop. #338 was approved but merged into its stacked base pentrc-edge-collisionality after #336 had already merged, so its changes never reached develop. This branch is the same content, plus 8b6f0d6, which nests the kinetic_format check inside if(present(format)) as suggested in the #338 review.

This is a slim version of the supplied-rate feature from #322, rebuilt on #336's way of handling ν.

What it does

Setting kinetic_format = "rates-v1" in PENT_INPUT reads a 9-column kinetic_file:

psi_n  n_i  n_e  T_i  T_e  omega_E  logLambda  nu_i(s^-1)  nu_e(s^-1)

The last three columns replace PENTRC's internal Coulomb/impurity collision model. The format and keyword are the same as #322, so existing adapter files work unchanged.

Default: "legacy" is the default and leaves behavior unchanged.

Validation: the run stops with a clear error stop message if:

• rates-v1 gets a table without exactly 9 columns;
• logLambda, nu_i or nu_e is non-positive;
• kinetic_format is not one of the two values.

Interpolation:

• The supplied columns are interpolated in log on the input table with the same pchip fit as n and T, then exponentiated onto read_kin's 101 nodes. This keeps the nodes positive, even when extrapolating past the end of the table.
• From there the columns go through PENTRC - BUG FIX - Stop the edge collision frequency spline from ringing (#321) #336's existing pchip ν slopes, so ν stays positive between nodes as well.
• No changes to torque.F90 or the integrand.

nfac/tfac: these scale n and T after ν is set, so they don't rescale supplied ν. This matches the behavior #322 documents.

DCON: kin_flag passes the same kinetic_format (dcon/dcon.F), so DCON and PENTRC read the table the same way.

Diff: +37/−10 in 4 files. There is no new build target, test harness or generated code, because the testing setup is frozen for this release.

How it differs from #322

	#322	here
Where ν is evaluated	exactly from a log-spline at each query, through new kin_eval wrappers	the 101-node kin grid with pchip slopes
Rotation of exactly 0	kept	goes through the existing replacement of 0 with 1e-9
wpfac	required to be 1	unrestricted (with 1, the rotation manipulation already does nothing)
Input validation	at every query, with bit-level finite checks	once, when the table is read
Generated code, build targets, Python oracles	added	none

Verification

These were throwaway runs on feynman of the regression_solovev_kinetic_example chain (DCON with kin_flag → GPEC → PENTRC), built with gfortran. They compare this branch against #336.

run	PENTRC T_total (fgar)	δW_k total	DCON W_t eigenvalues
#336, legacy	5.703490e-3	2.122400e-5	16.65462609, 216.66187207
this PR, legacy	bit-identical	bit-identical	bit-identical
this PR, 9-column file without kinetic_format	bit-identical (extra columns ignored, as before)	bit-identical	bit-identical
this PR, rates-v1 round trip*	5.703994e-3 (+8.8e-5 rel.)	2.122443e-5 (+2.1e-5 rel.)	16.65462844, 216.66187123

*In the round trip, the legacy formula's logΛ, ν_i and ν_e were written into the example's own table as columns 7–9 and fed back in with rates-v1.

• Inside the table: supplied ν_i matches the legacy ν_i to ≤ 5e-7 relative for ψ < 0.99.
• Last 0.01 in ψ: the two differ there because the paths extrapolate differently past the table's end (ψ = 0.998).
  • Legacy floors T at 1 meV, which puts ν ≈ 1e12 at ψ = 1.
  • rates-v1 extrapolates log ν smoothly, which puts ν ≈ 1e9 there.
  • Both values are far into the collisional regime, so torque moves by only 9e-5.

All PENTRC output files from the two legacy runs are bit-identical. The DCON outputs differ only in their CPU and wall-time stamps.

The three bad inputs (8 columns, a negative ν, kinetic_format="rates_v1") each stop with their message.

https://claude.ai/code/session_01VMDGyfQKRZQMdk3NJXsKQE

🤖 Generated with Claude Code