Could we use Jacobian Solver Instead In parallel solving

[fishboy82]

Hi all:
I am just reading the topic "Solving a single island across multiple cores" ,and thus read the paper "High-Performance Physical Simulations on Next-Generation Architecture with Many Cores", And I think the need to reorder constraints is because of the use of PGS solver ,the PGS solver 's iterator is dependent internal the Xk's value dependent on the x1...xk-1 's value calculated in this iterator . So I am think whether we could use Jacobian method that only use the value in the previous iterator step,In this situation we don't need to reorder the constraint and can have max parallel, Of course the Jacobian solver will converg not as quick as PGS Solver .But due to it's capability of maxmize parallel,Is it a valuable choice when do Parallel SOlver .It's my crazy idea may be ...... Do not laugh me because I am a beginner of this field..

[Dirk Gregorius]

Theoretically this makes sense. The question is if the speed gain due to parallesim is larger than the lost due to worse convrgence. So let us know when you find out.

[fishboy82]

So I will try to see whether this make sense in general