Convention regarding $\nabla u$

[srosenbu]

Currently, we use ufl.nabla_grad for $\nabla u$ which means $\partial u_j/\partial x_i$. Some other sources however define it as $\nabla u = \partial u_i/\partial_j$ which is represented in ufl as ufl.grad.

Which one do we want to use? Since this can be a really annoying source of error, we should make sure we use it consistently

[srosenbu]

@saifr68 we should probably deal with this before you merge your stress-rate branch

[saifr68]

I say let's just stick to what is already implemented and mention this in the documentation. And whenever an implementation method of some author is used we just check if the convention matches. If not, we make the necessary adjustments accordingly

[srosenbu]

I mean, my preferred way of implementing stuff is in Rust and there the memory-layout is column-wise instead of row-wise as it is in FEniCS. So I need to watch out for transposition anyways. So I am fine with either with a slight preference towards ufl.grad because the notation $\partial u_i/\partial x_j$ seems more intuitive to me.

Any more opinions on this @aradermacher , @pdiercks, @joergfunger ?

[aradermacher]

I'm fine with Saifs suggestion
Nevertheless we need to think about a documentation

[joergfunger]

I think the suggestion of @srosenbu is more intuitive, but just make sure it is correctly documented.