[StuartBenjamin]

grid_type = "original" is meant to put the equilibrium on the input file's own ψ surfaces, but it took every one of them and ignored psilow and psihigh:

• Inverse files (inverse_run, inverse_chease4_run): the grid ran from the axis to the last file surface. On TokaMaker i-files RDCON and STRIDE then never return, because the top/bottom search in equil_out_sep_find does not converge on that surface and has no iteration limit. The sas_flag reform to psilim could not move the edge either.
• Direct files (direct_run): sq%xs=sq_in%xs(1:mpsi) assigns mpsi values to an array of mpsi+1, so sq%xs is reallocated with the wrong size and bounds. The shipped DIIID_ideal_example with grid_type="original" stops with "Took too many steps to find flux surf.".

This PR makes the original grid psilow, then the file's surfaces strictly between psilow and psihigh, then psihigh: the same end points as every other grid type. A small new grid_mod function, clipspace, does this for both readers. The other grid types are unchanged.

grid_type="original"	develop	this PR
DIIID_ideal_example (g-file 257×257), DCON and STRIDE	stops: "Took too many steps to find flux surf."	runs; psilim 0.995, as with ldp (DCON energy: see below)
DIII-D TokaMaker i-file 129×257, RDCON and STRIDE	does not return (killed after 180 s)	runs (STRIDE 4 s); psilim 0.9962, as with ldp
DIII-D TokaMaker g-file 257×257, STRIDE	does not return	runs

STRIDE Re Δ′ (n = 1, mtheta 128) with this PR, as the change against ldp with mpsi 4096 on the 513×1025 i-file of the same equilibrium (DIII-D 174956 at 1490 ms):

input and grid	q = 2	q = 3	q = 4	q = 5	q = 6
i-file 257×513, original	−1.6 %	−0.2 %	+0.8 %	+2.3 %	+8.2 %
i-file 513×1025, original	−0.5 %	+0.2 %	0.0 %	+0.9 %	−12 %
g-file 1025×1025, original	+1.1 %	+0.5 %	+0.4 %	0.0 %	+38 %
g-file 1025×1025, ldp mpsi 4096	+1.0 %	+6.9 %	−3.1 %	−4.1 %	+1.3 %

The original grid now converges with file size on the interior surfaces. q = 6 sits just inside psilim, where the file's own surfaces are sparser than a packed ldp grid.

Running is not the same as being accurate. The original grid is only as fine as the file's grid, and a g-file's grid is uniform in ψ, so it is coarse at the edge. On DIIID_ideal_example, DCON's least stable total energy with this PR is:

grid	mpsi	least stable energy
original	256 (file)	−1.516 (reported unstable)
ldp	128 / 256 / 512	0.772 / 0.798 / 0.803
pow1	256	0.801
rho	256 / 512 / 1024 / 2048	3.12 / 1.34 / 1.01 / 0.846

rho, also coarse at the edge, approaches the ldp value only slowly as mpsi grows, so this is edge resolution rather than a bug in the new grid. dcon/README and the input/equil.in comment now say this and recommend ldp unless the input grid is packed toward the edge. The README also corrects "psilow is used only for ldp": every grid type except rho uses it.

The scripts and inputs used are in the attached zip.

---

🤖 Generated with Claude Code

[StuartBenjamin]

Added a second commit: clipspace now also drops input nodes within round-off (1e-10 of the range) of psilow or psihigh. TokaMaker i-files put their first packed node at psi_N = 1e-5, and round-off can land it just above psilow = 1e-5. That left a ~1e-16-wide first interval on which RDCON and STRIDE never left the axis. Grids without such a node are unchanged.

[matt-pharr]

This is good. I appended a minor docs update on this for reference