Trying to run usermeeting 2019 example of moving meshes

General
1

using code provided and converted to jupyterlite, changed Draw and inverse solver from umfpack to sparsecholesky, runs but the Newton evolution is unstable, any ideas?

t = 0.004
Newton iteration 0
err = 0.47488232780718304
Newton iteration 1
err = 46.878802931960614
Newton iteration 2
err = 598089124.7894585
Newton iteration 3
err = 1.9174661607616916e+30
Newton iteration 4
err = 9.700328834485895e+94
Newton iteration 5
err = nan
Newton iteration 6
err = nan
Newton iteration 7
err = nan
Newton iteration 8
err = nan
Newton iteration 9
err = nan
Newton iteration 10
err = nan
Newton iteration 11
err = nan
Newton iteration 12
err = nan
Newton iteration 13

2

this simpler example shows the same issue; everything works up to the NS Newton part

since its restricted to NS, and even though it works as-is in non-jupyterlite, i have reduced the time step and increased nu, no effect.

3

ah - think the problem here is that sparsecholesky is not meaningful for non-symmetric problems, and may also not work for non-positive definite matrices.

4

thanks, i will look again at getting umfpack.

any others work?

although i have been using sparsecholesky with the NS examples.

5

I think the only available direct solver in jupyterlite that can solve nonsymmetric/ not pos definite is “superlu”