pmodb NetCDF field-function reshape corrupts theta=0 complex value and omits endpoint

[krystophny]

Summary

The four complex scalar field-function variables written by gpout_pmodb have a one-point-short RESHAPE extent. Their NetCDF poloidal dimension contains mthsurf+1 points (theta=0...1), but lines 2991--3002 reshape the source arrays to (/mstep,mthsurf,2/). The result corrupts the complex value at theta=0 and leaves the periodic endpoint as fill.

Affected variables:

• b_eul_fun
• b_lag_fun
• Bdivxi_perp_fun
• Bkappaxi_perp_fun

The adjacent normal-field writer correctly uses (/mstep,mthsurf+1,2/) at lines 3635--3638.

Reproduction

Revision: 19b43523684d3336856462608d1ac56f6ed5f308.

I built GPEC with GNU Fortran, OpenBLAS, and system netCDF, ran the repository's regression_solovev_kinetic_example, and replaced its coil drive by one explicitly complex boundary harmonic:

coil_flag = .false.
harmonic_flag = .true.
cosmn(2) = 1e-4
sinmn(2) = 2e-4

I ran the same DCON solution twice with (ip_direction,bt_direction)=(positive,positive) and (positive,negative), giving NetCDF helicity=+1 and -1. The internal b_lag spectra are bitwise identical. For theta_dcon indices 1--63, every field-function value obeys the source's expected -helicity imaginary-part map exactly. At index 0, however, both runs contain the same complex value rather than conjugates; its imaginary part is identically copied from its real part. At index 64 (theta_dcon=1) both components are fill.

Representative b_lag_fun(theta=0,psi index 100) values:

helicity=+1:  1.6715577267257276e-4 + 1.6715577267257276e-4 i
helicity=-1:  1.6715577267257276e-4 + 1.6715577267257276e-4 i

At an interior point the expected map works:

helicity=+1:  1.5994490543356555e-4 + 1.4473644169856975e-4 i
helicity=-1:  1.5994490543356555e-4 - 1.4473644169856975e-4 i

Cause

The arrays eulbparfout, lagbparfout, divxprpfout, and curvfout have the periodic 0:mthsurf poloidal extent. The NetCDF theta coordinate also has mthsurf+1 entries. The four writes nevertheless use:

(/mstep,mthsurf,2/)

so the concatenated real/imaginary source is split at the wrong offset and supplied with one missing poloidal plane.

Expected

Use (/mstep,mthsurf+1,2/), add a complex nonzero-sine regression, and assert:

• both components at theta=0 are correct;
• the periodic endpoint equals theta=0;
• reversing helicity reverses only the packed imaginary component for the same native solution.

Scope

This is an output serialization defect in the four pmodb field-function variables. The Fourier-space b_eul, b_lag, Bdivxi_perp, and Bkappaxi_perp variables are not affected. No production source was changed during this audit.