excitation: assign every current-source rhs entry at coincident elements

[KJ7LNW]

Description

An EX type-4 elementary current source placed exactly on the center of a segment or the centroid of a patch made nec2++ print garbage for that element and for the currents coupled to it.

etmns_current_source walks the n+m elements, measures the squared distance from the source point to each element center, and continues when that distance vanishes, src/nec_context.cpp:3419. The entries owned by the skipped element, e[i] for a segment or the two tangential slots for a patch, are never assigned, so the solver factors whatever the array held. The three sibling builders, etmns_voltage_source, etmns_linear_wave, and etmns_circular_wave, assign every row they own unconditionally.

current_vector is a member of the persisting nec_context, src/nec_context.h:849, sized at src/nec_context.cpp:140. safe_array::resize, src/safe_array.h:96-107, grows storage 1.5x and copies only head(_len), and the same vector carries solve results that c_geometry::cabc expands in place. The skipped slot therefore reads residue. Zero is the value the geometry calls for: an element centered on the source point receives no field from it.

Type of Change

• Bug fix (non-breaking change that fixes an issue)

Implementation Details

• nec_context::etmns_patch_base returns a patch's first tangential slot in the n+2m layout, replacing the inline n + patch_index*2 arithmetic in the builder.
• nec_context::etmns_zero_element assigns zero to one element's entries, one for a segment and two for a patch, off that same base, so the zeroing and the writing agree by construction on where a patch lives.
• The coincidence guard calls it before continue, so every index in [0, n_plus_2m) is assigned on return.

Reproduction

Three patches tile a plate in the xy-plane at x = +0.1, 0, -0.1, the source sits on the middle patch centroid at the origin, and a wire off to the side gives n > 0 segments so the patch slots are based at a nonzero n.

CM Elementary current source (EX type 4) coincident with a patch centroid.
CE
GW     1     3  5.00000E-01 -5.00000E-02  0.00000E+00  5.00000E-01  5.00000E-02  0.00000E+00  1.00000E-03
SP     0     0  1.00000E-01  0.00000E+00  0.00000E+00  9.00000E+01  0.00000E+00  1.00000E-02
SP     0     0  0.00000E+00  0.00000E+00  0.00000E+00  9.00000E+01  0.00000E+00  1.00000E-02
SP     0     0 -1.00000E-01  0.00000E+00  0.00000E+00  9.00000E+01  0.00000E+00  1.00000E-02
GE     0     0
FR     0     1     0     0  3.00000E+02  0.00000E+00  0.00000E+00  0.00000E+00  0.00000E+00  0.00000E+00
EX     4     0     0     0  0.00000E+00  0.00000E+00  0.00000E+00  9.00000E+01  0.00000E+00  1.00000E+00
RP     0    19    37  1000  0.00000E+00  0.00000E+00  1.00000E+01  1.00000E+01  0.00000E+00  0.00000E+00
EN     0     0     0     0  0.00000E+00  0.00000E+00  0.00000E+00  0.00000E+00  0.00000E+00  0.00000E+00

Before: patch 2 printed 9.88E-324 as its Y-real component, the double whose bit pattern is the integer 2, in a problem whose quantities span 1e-10 to 1e+1. The wire currents were asymmetric about the source plane.

After: patch 2 reads zero in every component and the wire currents are antisymmetric about that plane, segments 1 and 3 printing ±6.9547E-05 ∓1.5977E-05 with the center segment nulling at 2.2E-11, the symmetry a z-directed source imposes.

  PCH   --- PATCH CENTER ---     TANGENT VECTOR 1     TANGENT VECTOR 2    ------- X ------    ------- Y ------
    1   0.100   0.000   0.000  1.8800E+01   176.13  0.0000E+00     0.00 -1.88E+01  1.27E+00  0.00E+00  0.00E+00
    2   0.000   0.000   0.000  0.0000E+00     0.00  0.0000E+00     0.00 -0.00E+00 -0.00E+00  0.00E+00  0.00E+00
    3  -0.100   0.000   0.000  1.8800E+01    -3.87  0.0000E+00     0.00  1.88E+01 -1.27E+00  0.00E+00  0.00E+00

Cross-engine: before this change nec2++ disagreed with nec2c, xnec2c, and the NEC-2 Fortran on currents:current, rdpat:gtot, rdpat:sens, rdpat:tilt, and rdpat:axrt for this deck; after it, those agree.

Those three carry the same skip, at nec2c.c:4579, xnec2c/src/matrix.c:1066, and nec2dxs.f:4347 where the guard branches to the loop-terminating 24 CONTINUE. Their zero at that slot arrives implicitly, from static storage or a first-touch malloc, rather than from an explicit assignment; each reuses one right-hand-side buffer across frequency steps, so the skipped slot carries the previous step's solved current on a multi-frequency deck. Matching corrections are going to those engines as well, so every lineage assigns the entry explicitly.

Three rdpat:gtot rows remain at phi=120, theta=110, where the others print -18.53 dB and nec2++ prints -18.52 dB while the underlying E(theta) and E(phi) agree to five figures: one two-decimal print step. The companion deck that offsets the source and never reaches the guard carries the identical three rows.

The segment branch of the same guard, a type-4 source on a segment center, is covered by the same change.

Testing

• Tests pass locally