Scaling of condition number in 1D, 2D, and 3D

[nate]

It would likely benefit you to use assemble_system which preserves symmetry of the underlying operator when applying boundary conditions. Particularly for the efficient computation of eigenvalue approximations.

This ambiguity (unless you read the documentation in detail) is a “feature” of legacy DOLFIN, and one of many reasons I highly recommend using DOLFINx instead of legacy DOLFIN.

See, e.g.,

How to implement the equivalence of `assemble_system` in DOLFINx

Hello, I met an issue when trying to convert my old dolfin code into DOLFINx. Here I was trying to assemble the Jacobian matrix with Dirichlet boundary conditions, but got different results with the two libraries. In the old dolfin code, I used the assemble_system() which would preserve the symmetry of the Jacobian matrix (the A matrix in the code example). While in DOLFINx, I used assemble_matrix() from dolfinx.fem.petsc, where the symmetry is also preserved, but the numerical values are differ…