nested_dissection does not reduce the matrix pattern

[cesare-corrado]

Hi,
IU would like to use pymetis to improve the pattern of a sparse matrix obtained with finite element discretisation. I tested the algorithm using a Delaunay triangulation of a 2D rectangle. However, the pattern of the matrix after permutation is (in the best scenario) equal to that of the original matrix (see the attached figure). Could you help me with this, please?
I can share the code I used for this example, in case