Open cortner opened 2 years ago
I see! Is that the approach Ralf Drautz describes in his paper?
(1) is Ralf's approach. (2) is to stick with our approach but just a neater way to compute the coupling coefficients
could actually be a nice BSc or MSc project if you know somebody?
This is to discuss whether we want to (1) entirely move away from putting Euclidean tensors into the coupling matrix and instead just apply a transform at the end; or (2) keep the euclidean tensors but replace the tables for the coco initial conditions with an implementation.
This is based on the following observation: there is a matrix A (easy to make explicit of course) such that
x = r * A * Y1(x)
. This gives us in particular that a rotation matrix can be expanded as followsI.e. we can explicitly express Q in terms of D^1 matrices and from here on the computation of the coupling coefficients could be automated. This can of course be extended to tensors of arbitrary order.
I think we should FOR SURE to (2), but even consider (1). What would the performance downside be? Is there a benefit to having less generality in the possible coupling coefficients?
CC @MatthiasSachs