Mechanics invariants and B-matrices

[endJunction]

This is part of the mechanics PR #1340.

Includes:

• deformation process specific invariants in the Kelvin mapping.
• B-matrices (like shape-matrices for mechanics) and Kelvin vector types.

For the tests the error bounds were eyeballed from a gnuplot graphics showing the error vs. input vector norm. Tried to make them as tight as possible. It might happen that they will need to be increased a little bit... we'll see.

Aside from the failing compiler (of the test code) this is ready for review.

Todo:

• Cite Nagel et al. Kelvin mapping.
• rename files/classes. Will do at the very end to reduce rereview:
  • MathLib/mechanics_invariants.h > MaterialLib/SolidModels/KelvinVector.h
  • Split off impl.h or even .cpp for specializations of det/inv.
• fix MSVC12 compilation.
• Possibly add test of LinearBMatrix using computeStrain function. Will not do.
• Test P_dev * P_sph == P_sph * P_dev == 0 6x6; P_dev + P_sph == identity 6x6;
• Move inverse out of Invariants.
• Maybe rename BMatrixPolicy::_columns to "number of DOF"? (And _rows to _kelvin_vector_size?)
• Merge/Unify ShapeMatrixPolicy and the BMatrixPolicy
  • Now you use the ::detail::EigenMatrixType, why not the respective type from ShapeMatricesType?
  • Wouldn't it be better to use the typedefs from ShapeMatrixPolicy here instead of always using fixed size matrices?
  • DynamicBMatrixPolicy Can be merged with the above into one type using only the typedefs from ShapeMatrixPolicyType.

[chleh]

maybe m(0,0) = v[0], or something like Eigen::Matrix<double, 1, 1> m = Eigen::Matrix<double, 1, 1>::Constant(v[0])?

[endJunction]

Jenkins, test this please.

The following ctests FAILED for some reason but they don't affect the review for now:
4 - MapGeometryToMeshSurface-MapGeometryToMeshSurface_Bode-time-diff (Failed)
6 - MapGeometryToMeshSurface-MapGeometryToMeshSurface_Naegelstedt-time-diff (Failed)

[chleh]

Can that be removed now? ✅

[chleh]

Only minor additional comments from my side. Basically:

• Eigen and auto
• == 0.0 vs. < eps
• some remaining displacementdim 1 code.

Afterwards 👍

[norihiro-w]

unsure: noalias() may influence optimization

[norihiro-w]

looks good, besides minor comments

[ogsbot]

Jenkins: OGS-6/Gui/Gui-Linux-PRs failed: https://svn.ufz.de:8443/job/OGS-6/job/Gui/job/Gui-Linux-PRs/2630/

[endJunction]

Can I rebase and merge, or is there more review to be expected?

The only test failing (again) is

[ RUN      ] NumLib.ComponentNormMultiComponent2
/home/travis/build/ufz/ogs/Tests/NumLib/TestComponentNorms.cpp:94: Failure
The difference between total_norm and compwise_total_norm is 8.8817841970012523e-15, which exceeds tolerance, where
total_norm evaluates to 14.747408246182694,
compwise_total_norm evaluates to 14.747408246182685, and
tolerance evaluates to 6.6613381477509392e-15.
[  FAILED  ] NumLib.ComponentNormMultiComponent2 (72 ms)

[chleh]

I'm fine with this PR. ⏩
I will raise the test tolerances since this test failed very often recently.

[endJunction]

Jenkins, test this please!

[endJunction]

It could happen that the error margins in the unit tests are too tight. In that case I'll increase the tolerance a little.

[ogsbot]

OpenGeoSys development has been moved to GitLab.

See this pull request on GitLab.