Failing tests for off-diagonal elasticity tensor components for hyperelastic models

[th-skam]

Some context

There are a number of classes implementing material models in Sofa/Component/SolidMechanics/FEM/HyperElastic/src/sofa/component/solidmechanics/fem/hyperelastic/material. The Sofa/Component/SolidMechanics/FEM/HyperElastic/tests/Material_test.cpp file tests 3 functions: deriveSPKTensor, applyElasticityTensor and ElasticityTensor, implemented by these classes.
Here we are looking at models that already pass these tests.

The issue

Currently, the test for the ElasticityTensor is incomplete as it checks only the diagonal components of the tensor.

sofa/Sofa/Component/SolidMechanics/FEM/HyperElastic/tests/Material_test.cpp

Lines 189 to 190 in 98ae907

	// for (sofa::Size i = 0; i < Matrix6::size(); ++i)
	for (sofa::Size i : {0, 2, 5}) //only the diagonal terms

If you activate the off-diagonal test, all components fail. But... it's just that the off-diagonal entries are multiplied by 2. e.g. from the VerondaWestman

This is the case for all models and I would assume that this multiplication is put in place on purpose. Maybe it's convenient to store the tensor like this (due to symmetry)?

Does anyone know whether this is by design or a mistake?

[alxbilger]

Not sure if this is the reason, but in Voigt notation, the off-diagonal terms have a factor 2.

[th-skam]

That's true but in the end it's all a matter of what the code that uses the tensor would do with it, hence the question.
I'm not sure where to look at to confirm this.

[th-skam]

In fact, maybe that's our answer:

sofa/Sofa/Component/SolidMechanics/FEM/HyperElastic/src/sofa/component/solidmechanics/fem/hyperelastic/StandardTetrahedralFEMForceField.inl

Lines 419 to 423 in 98ae907

	myMaterial->ElasticityTensor(tetInfo,globalParameters,outputTensor);
	Matrix63 mBl=matB[l];
	mBl(1,0)/=2;mBl(1,1)/=2;mBl(1,2)/=2;mBl(3,0)/=2;mBl(3,1)/=2;mBl(3,2)/=2;mBl(4,0)/=2;mBl(4,1)/=2;mBl(4,2)/=2;
	N=(matB[k].transposed()*outputTensor*mBl);