Equilibrium parameters: compute the stubbed flux bookkeeping fields, and collapse the duplicate psi-on-axis variables

[logan-nc]

Fallout from the gpec.h5 schema overhaul (#226, PR #364). While giving the Equilibrium/ scalars literature names I audited every field of Equilibrium.EquilibriumParameters against its assignment site, and two related problems turned up. Neither is a regression — both predate the schema work — but the rename made them visible, so they are worth fixing deliberately rather than carrying forward.

1. Several flux-bookkeeping fields are hardcoded constants, not computed

src/Equilibrium/Equilibrium.jl:370-381 is the sole assignment site for the whole flux-bookkeeping family, and most of it is literals:

pe.params.psi0                 = psio
pe.params.psi_axis             = pe.psio
pe.params.psi_boundary         = 1.0
pe.params.psi_boundary_norm    = 1.0
pe.params.psi_axis_norm        = 0.0
pe.params.psi_norm             = 0.0
pe.params.psi_axis_offset      = 0.0
pe.params.psi_boundary_offset  = 0.0
pe.params.psi_axis_sign        = 1
pe.params.psi_boundary_sign    = -1
pe.params.psi_boundary_zero    = false

Everything from psi_boundary down is a constant independent of the equilibrium being loaded. In normalized-flux terms psi_axis_norm = 0 and psi_boundary_norm = 1 are true by definition of psi_N, so those two are honest tautologies — but psi_boundary = 1.0 is written with units Wb/rad in the schema and is not the boundary poloidal flux of any real equilibrium, and the offsets/signs/psi_boundary_zero look like they were meant to record what the ingest actually did (Fortran GPEC derived the equivalent quantities from the g-file simag/sibry pair and the COCOS handling). The likely story is that the Julia port stubbed them and nothing downstream noticed.

Separately, four fields of the struct are never assigned anywhere in src/ — ro, zo, psio, b_norm. The HDF5 writer works around the first three by pulling them off PlasmaEquilibrium directly (equil.ro, equil.zo, equil.psio) and b_norm is silently skipped because it stays nothing.

Ask: compute these from the ingest rather than hardcoding them — real psi_axis/psi_boundary in Wb/rad, real offsets/signs reflecting the COCOS/normalization actually applied, a decision on b_norm (compute it or delete the field), and either populate params.ro/zo/psio or delete them in favour of the PlasmaEquilibrium fields so there is one source of truth.

2. Three variables are the same number for psi on axis

psi0, psi_axis, and psio all end up holding the total poloidal flux difference. psi_axis in particular is misnamed: it is assigned pe.psio, i.e. the total flux, not the flux at the axis. psi_axis_norm is likewise an exact duplicate of psi_norm (both 0.0).

For the schema overhaul I took the conservative route: Equilibrium/psi_total is written once (from equil.psio) and the duplicates psi0, psi_axis, psi_axis_norm are dropped from the HDF5 output via EQUIL_H5_SKIP in src/HDF5Schema.jl. That fixes the file, not the struct.

Ask: collapse the redundancy in EquilibriumParameters itself — either delete the duplicate fields, or, if each was intended to mean something distinct (axis flux vs. total flux vs. normalization constant), define those meanings and make the assignments match the names. Whatever survives should have a docstring entry stating which of the three it is; today a reader of the struct cannot tell them apart.

Notes

• Fixing (1) will move numbers only if a downstream consumer reads the stubbed values; a regression-harness run (regress --cases diiid_n1,solovev_n1 --refs develop,local) should confirm the blast radius before merging.
• Any change to which fields exist must be mirrored in EQUIL_H5_NAMES / EQUIL_H5_SKIP (src/HDF5Schema.jl) and the naming rules in docs/development/hdf5-conventions.md.

[logan-nc]

One trap to close when this is implemented: src/GeneralizedPerturbedEquilibrium.jl writes Equilibrium/psi_total, R_axis, and Z_axis with three explicit lines after the fieldnames(EquilibriumParameters) loop, pulling the values off PlasmaEquilibrium because params.psio/ro/zo are nothing. EQUIL_H5_NAMES also maps those same three fields, so the loop currently skips them only by virtue of the nothing check.

The moment params.ro/zo/psio are actually populated, the loop and the explicit writes will both try to create the same three datasets and the write will fail. Whichever change populates them must drop the explicit lines (or the map entries) in the same commit.

[logan-nc]

Assigning to @jhalpern30 — this is the same class of problem as the psihigh/psilim confusion you already untangled in the control-struct cleanup (#367, #383): fields whose names promise a distinction the assignments don't actually make, plus fields that are never populated at all.

The overlap is direct. EquilibriumParameters has psi0, psi_axis, and psio all carrying the total poloidal flux (with psi_axis misnamed — it is assigned pe.psio), psi_axis_norm duplicating psi_norm, most of the rest of the flux block hardcoded to literals rather than derived from the ingest, and ro/zo/psio/b_norm never assigned anywhere in src/. The HDF5 side is already handled — the duplicates are dropped from the output file — so what remains is the struct itself, which is squarely your cleanup.

See the comment above for the one trap: params.ro/zo/psio are mapped in EQUIL_H5_NAMES while the writer also emits those three datasets explicitly, so populating them without dropping the explicit lines will collide on dataset creation.