convention for strain/stress tensor rotation

[jbjacob94]

I noticed an inconsistency in how second-order tensors (stress/strain) are rotated between coordinate frames in different parts of ImageD11.

ImageD11.stress rotates stress from the reference (crystal) frame to the lab frame using an active rotation convention.

sLab = F.U.dot(sRef).dot(F.U.T)

The strain used there, derived from ImageD11.finite_strain, appears to follow the same convention.

In contrast, ImageD11.sinograms.tensormap applies the passive rotation when transforming tensors.
res[...] = U.T.dot(tensor_sample).dot(U)

Which one is correct?

[jonwright]

Do you have an example showing the difference? I thought both codes are tested against real data for samples under uniaxial load.

[jbjacob94]

Ok I found where the problem is. Both use active convention but there is an error in tensormap.eps_sample.

in:
@Property
def eps_sample(self):
[...]
elif 'eps_crystal' in self.keys():
# we have strain in crystal reference frame
# rotate it into sample frame and return
print('Rotating eps_crystal into sample frame')
eps_sample_map = tensor_sample_to_crystal(self.eps_crystal, self.U)
self.add_map('eps_sample', eps_sample_map)
return eps_sample_map
[...]

If eps_crystal has already been defined, the function to compute eps_sample is
eps_sample_map = tensor_sample_to_crystal(self.eps_crystal, self.U)

which actually does the reverse operation: eps_sample -> eps_crystal instead of eps_crystal -> eps_sample

The problem only appears when eps_crystal has been previously computed and you're then trying to get eps_sample, so it may have passed under screening during testing.

edit: created pull request to fix the problem