LiYang

I have confirmed that the performance degradation is caused by PR #2623. It might be due to the inaccuracy of the stiffness matrix.

> See also #2826 Appreciate. This issue will be closed once #2826 is addressed.

> Hello, @karimifard and myself have been testing the `pknViscosityDominated_smoke.xml` case, and we could see a big difference before and after this PR #2481, which quite changes the `xml`. We...

> > Hello, @karimifard and myself have been testing the `pknViscosityDominated_smoke.xml` case, and we could see a big difference before and after this PR #2481, which quite changes the `xml`....

> Hi, Have you checked if `kgdViscosityDominated_poroelastic_benchmark.xml` is exactly the same for "`develop`" and "`develop 94e338c`" cases? Thanks! Yes, I actually compiled `develop` and `develop 94e338c` as two singularity image...

> > @karimifard @TotoGaz any update? > > So far I haven't find anything "wrong". At this point the performance difference seems to be due to nonlinearity treatment. Essentially the...

Hi @paveltomin @TotoGaz @jhuang2601 @karimifard , I have resolved the performance issue as described in #2922 . I kindly requests you to check the PR and provide your feedback when...

> The issue is that this will effectively change the solution. I can see how it would help with convergence because it removes a nonlinearity but it is essentially treating...

Updating the stencil weights at the end of each Newton iteration is a standard Newton iteration, not lagging. $$J(y_k^{n+1})(y_{k+1}^{n+1} - y_k^{n+1}) = -R(y_k^{n+1})$$ $k$ is the Newton iteration count and...

> > Updating the stencil weights at the end of each Newton iteration is a standard Newton iteration, not lagging. J(ykn+1)(yk+1n+1−ykn+1)=−R(ykn+1) > > k is the Newton iteration count and...