anim_to_vtk: 3DELEM/SPHELEM TENSORS blocks transpose the YZ and XZ shear components

[pmquang87]

Summary

anim_to_vtk writes the symmetric 3×3 matrix of every TENSORS 3DELEM_* and TENSORS SPHELEM_* block with the YZ and XZ shear components transposed: SYZ lands at positions (0,2)/(2,0) and SXZ at (1,2)/(2,1), which is the reverse of the standard symmetric-tensor layout. The 2D (shell) tensor path is correct; only the 6-component 3D and SPH paths are affected.

Evidence

1. Component order of the A-file tensor record is XX, YY, ZZ, XY, YZ, XZ.
The engine documents it in engine/source/output/anim/generate/srota6.F (lines 193–205):

C    1     SX
C    2     SY
C    3     SZ
C    4     SXY
C    5     SYZ
C    6     SXZ
        L12    =TENS(4)
        L23    =TENS(5)
        L13    =TENS(6)

tensor6.F then writes the six values straight through (R4(1)=TENS(1,N) … R4(6)=TENS(6,N), CALL WRITE_R_C(R4,6)), and anim_to_vtk reads them flat with no permutation (Ufread(tensVal3DA, sizeof(float), nbElts3D*6*nbTens3D, inf)). So in the tool's 0-based indexing: +3 = SXY, +4 = SYZ, +5 = SXZ.

2. The 3D emit loop places index 4 (SYZ) in row 0 and index 5 (SXZ) in row 1 — src/anim_to_vtk.cpp lines 1473–1475:

cout << tensVal3DA[(iel * 6) + ...] << " " << tensVal3DA[(iel * 6) + 3 + ...] << " " << tensVal3DA[(iel * 6) + 4 + ...] << "\n";
cout << tensVal3DA[(iel * 6) + 3 + ...] << " " << tensVal3DA[(iel * 6) + 1 + ...] << " " << tensVal3DA[(iel * 6) + 5 + ...] << "\n";
cout << tensVal3DA[(iel * 6) + 4 + ...] << " " << tensVal3DA[(iel * 6) + 5 + ...] << " " << tensVal3DA[(iel * 6) + 2 + ...] << "\n";

The emitted matrix is therefore

XX  XY  YZ
XY  YY  XZ
YZ  XZ  ZZ

whereas the standard symmetric layout (what VTK consumers such as ParaView assume when computing eigenvalues/eigenvectors) is

XX  XY  XZ
XY  YY  YZ
XZ  YZ  ZZ

3. The SPH loop has the identical pattern — lines 1575–1577 (tensValSPH, same 0,3,4 / 3,1,5 / 4,5,2 offsets).

4. The 2D shell path is correct for contrast — plane stress [s0 s2 0 / s2 s1 0 / 0 0 0], textbook.

Impact

• Von Mises computed from the emitted tensor is unaffected (the swap permutes the off-diagonal squares).
• Principal stresses and principal directions are wrong whenever SYZ ≠ SXZ, since the swap changes the third invariant (det). Any downstream eigen-analysis — ParaView's Principal Stress filters, tensor glyphs, user scripts — silently produces incorrect results for solid and SPH tensors.
• Downstream converters that have hard-coded the transposed layout to compensate will need to track this fix if applied.

Suggested fix

Emit row 0 as indices 0, 3, 5, row 1 as 3, 1, 4, row 2 as 5, 4, 2 in both loops (3D at lines 1473–1475 and SPH at 1575–1577).

Observed in main @ 4e52942 and in the shipped anim_to_vtk_win64.exe from the 2026-05-20 binary bundle.