For the higher order extended wave action equation we need to computer higher order derivatives on unstructured grids, any hints suggestions are welcome.
This is difficult and depends on the kind of grid. For a totally arbitrary irregular grid Baumgardners method is in principle possible. However, it is computationally for higher order so expensive that Baumgardner did use it for second order only. If you can exploit some regularity of the grid, such as some of the points lining up, quite efficient methods are possible. See mon sea rev 136 no 7 2683 for this and the citation of Baumgardners work
we are seeking something for arbitrary irregular grids, since we are using a priori error estimates and in future even adaptive grids, there is no chance for exploring grid regularities, however, many thanks for the hint!
For arbitrary grids, even for 1st order elements, you can compute the L^2 projection of the derivative over the mesh, integrate by parts and solve the finite element problem. Then go on until you get to the derivative you want.
My inclination, at least for a start, would be to do a least-squares reconstruction. With a sufficiently large stencil, this can give you as many derivatives as you need. For instance, to get third derivatives in 2D to second order accuracy, you'd need a reconstruction that goes as high as fourth derivatives. This is a total of 15 terms; I'd use about 50% more data than this so the L2 reconstruction can filter out noise. (And yes, this requires smooth data, but that's always true for derivatives...)
See the paper below, or any of Tim Barth's stuff on reconstruction, for mathematical details of the process.
Article A High-Order-Accurate Unstructured Mesh Finite-Volume Scheme...