checkFESByAutoDiff Test fails for the vanishing stress case when displacements are zero

[tarun-mitruka]

When the displacements are zero, the strains are zero. This doesn't perform any iteration in the Newton-Raphson method when a VanishingStress material is used along with automatic differentiation. AutoDiff doesn't see the Newton-Raphson method when no iterations are performed and thereby doesn't provide the correct stiffness matrix. This, for instance, makes the plane stress case behave like a plane strain case, resulting in wrong results when the Poisson's ratio is non-zero.

Possible fixes:

• Perturb the strain vector by 1e-10 or similar for the vanishing stress material case
• Force one iteration in the Newton-Raphson method by default
• Add a boolean flag in the Newton-Raphson method to force one iteration
• ...

[tarun-mitruka]

@rath3t Which fix option would you recommend? On one hand I think the Newton-Raphson method should not be modified for such a fix as then Trust Region and NewtonRaphsonWithSubsidiaryFunction should also be modified to have a symmetry on how the number of iterations are counted. On the other hand, perturbing the strain can also may lead to some inaccuracies depending on the perturbation. Is there a better way? The branch bug/nlelastic now reproduces this error.

[rath3t]

You can add the minIter parameter to the Newton-Rhapson settings and make it default to =0. This is the cleanest way I suppose