Jacobian matrix and residual - test cases and discrepancies

General
1

Dear all,

I am doing some tests with NGSolve and nonlinear case studies. I am playing a bit around adding my own jacobian.

I am running two test case studies, one using a.Apply(gfu.vec, res) to get the residual (testbis.py) and the other expressing directly the residual (test.py). It seems that there is a key difference that I do not grasp during the assembly and if possible I would need some help to understand what happens behind the hood explaining the difference that I get in the number of iteration of the Newton and result.

Best regards,

Frederic

testbis.py (1.7 KB)

test.py (1.7 KB)

2

Hi Frederic,

you need to assemble the linear form, res_nl.Assemble(), otherwise res_nl.vec is zero.
Then it should work. The values might not coincide because of different integration rules.

Best,
Michael

3

Dear Michael,

Thanks, indeed it seems that was the missing item. Now, the convergence is very similar as well as the results.

best,

Frederic