[krystophny]

read_kin replaces every zero $\omega_E$ knot by $10^{-9}$ rad/s before it rescales the rotation (pentrc/inputs.f90#L290-L306). The replacement is there because the rescaling divides by $\omega_E$. Here $\omega_E$ is the $E\times B$ frequency after the input factor wefac, and wpfac scales the total rotation $\omega_E + \omega_{*n} + \omega_{*T}$, where $\omega_{*n}$ and $\omega_{*T}$ are the density- and temperature-gradient frequencies (wdian, wdiat). As a result, a run with wefac = 0 and wpfac = 1 uses $10^{-9}$ rad/s instead of zero, and every knot where the profile itself is zero is shifted in the same way.

The rescaling now uses the equivalent form without the division, and the replacement and its warning are removed:

$$\omega_E \leftarrow \omega_E + (w_p - 1),(\omega_E + \omega_{*n} + \omega_{*T}), \qquad w_p = \texttt{wpfac}.$$

This keeps exact zeros, and with wpfac = 1 it leaves $\omega_E$ bit-for-bit unchanged; the old expression changed it by rounding.

regression/test/pentrc_zero_omega_e.f90 reads docs/examples/regression_solovev_kinetic_example/kinetic.txt. It is run by hand against the built PENTRC objects and is not part of CI. Without the fix it stops with max |omega_E| for wefac = 0: 1.0E-09; with the fix it passes. The Julia GPEC reforms $\omega_E$ without dividing, so it needs no counterpart.

Upstream: PrincetonUniversity#275.

Chris&AI