You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Is there a way to get fd.weight_matrix to produce matrices for, say, [1,0] and [0,1] at the same time? Not as a sum, but as two different operations. Under the hood, it would just be different right hand sides for the same core RBF matrix-vector system for each stencil. Solving against multiple right hand sides as a tall, skinny matrix should be much faster than re-inverting the system in multiple fd.weight_matrix calls.
There is not a way to do that, but I like the idea. I am trying to think of what the usage should look like. It seems like this should be implemented in a new function called weight_matrices, where specifying diffs=[[1, 0], [0, 1], [[2, 0], [0, 2]]] would be interpreted as requesting the matrices for d/dx, d/dy, and the Laplacian.
I am pretty immersed in another project right now, so I dont know when I will get around to this.
Is there a way to get fd.weight_matrix to produce matrices for, say, [1,0] and [0,1] at the same time? Not as a sum, but as two different operations. Under the hood, it would just be different right hand sides for the same core RBF matrix-vector system for each stencil. Solving against multiple right hand sides as a tall, skinny matrix should be much faster than re-inverting the system in multiple fd.weight_matrix calls.
See e.g. Appendix B of Flyer, Barnett & Wicker 2016 https://doi.org/10.1016/j.jcp.2016.02.078 where they do it this way in their MATLAB implementation.
The text was updated successfully, but these errors were encountered: