Introduce arbitrary orientation for lumped ports #75

…ce currents. The original specification using keywords still works, but now also provide an optional interface to specify the direction using a 3 vector.

…aNode system

* Both make use of a reduce/gather paradigm to compute bounding boxes/balls for a collection of points. Given this, they do not scale well with growing ranks. They are however capable of handling arbitrary orientation of two dimensional ports. * The algorithms rely on sorting tricks applied to containers of the vertices * The uniform lumped port was tested by applying a 30 degree rotation about all three axes to all the vertices of the rings mesh. * The uniform lumped port also performs an axis snapping for user specified direction vectors, this means the resulting calculations are forgiving on precision issues. * Calculations were tested by applying rotations to the global mesh and checking that the resulting areas agree with the integrated values

…tes from a discovered axis, rather than snap to axis

…0,1). - Most of the values are perturbed only on the 3rd sf or so. - For the probe data, it is necessary to apply the axis angle transformation on the original data - In julia: ``` julia> AngleAxis(pi/3, 0,0,1) * [-6.975740032e-01, -3.331547360e+00, +1.022439079e+02] 3-element StaticArraysCore.SVector{3, Float64} with indices SOneTo(3): 2.5364176460709804 -2.269890487790808 102.2439079 julia> AngleAxis(pi/3, 0,0,1) * [-2.108257694e-03, -1.094675073e-02, +1.007018922e+01] 3-element StaticArraysCore.SVector{3, Float64} with indices SOneTo(3): 0.008426035374075847 -0.007299180085728002 10.07018922 ``` which agree fairly closely with the newly generated values ``` i, B_x[1], B_y[1], B_z[1] 1.000000e+00, -2.744244415e+00, +2.362252272e+00, +1.021127464e+02 2.000000e+00, -8.553328102e-03, +7.180956729e-03, +1.007140281e+01 ```

… with theta = 0.0

- All the non-probe data is very close to the original values, modulo a few sf - The probe data requires a transformation of the rebaselined data: ``` julia> Ref(AngleAxis(pi/6, 0,0,1)) .* [[-1.934600007e+00, +1.332005966e+00, +1.007689041e+02], [-2.056769746e-03, +3.073509047e-03, +1.007227123e+01]] 2-element Vector{StaticArraysCore.SVector{3, Float64}}: [-2.3414157352235527, 0.18625100104843154, 100.7689041] [-0.0033179693732712672, 0.0016333520404633007, 10.07227123] ``` which compares to ``` i, B_x[1], B_y[1], B_z[1] 1.000000e+00, -2.471968830e+00, +3.021551217e-01, +1.012769086e+02 2.000000e+00, -7.789873394e-03, +1.048294735e-03, +1.006816729e+01 ``` the additional discrepancy can be partially attributed to breaking of rotational symmetry, as the axis does not pass through a vertex of the bounding box of the domain. Better agreement can be found with pi/3 around (1,1,1), but (0,0,1) is chosen for ease of understanding in the examples. Note: the data was generated on a mac M1 with a tighter tolerance of 1e-9, as this was required to reduce discrepancy in the values of the second probe and to agree with CI.

- Update schema to account for CoordinateSystem - Fix bugs with config parsing and update some more config files - Refactor point cloud calculations to reduce duplication and hide local functions

…ammar

…s' into hughcars/lumped-port-orientation-dev

…nd comments and allowing GlobalMinLoc/MaxLoc to use arbitrary type for the second argument as supported by the available MPI types

…for maximally separated points to define initial diagonal/diameter

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Introduce arbitrary orientation for lumped ports #75

Introduce arbitrary orientation for lumped ports #75

Commits on Aug 22, 2023