Sorry, it’s me agin. It is a continuous question based on my last post , I have now fix the issuse with assembling the vector, now there is not error reported when I run the program. But I meet another problem: the solver gives me different solution when work in parallel. Since this doesn’t happend…

As I just posted in the other post, you need to use a function with the MPC function space to store your solution, if you want back-substitution to work properly, for the same reason of using the MPC function space for the RHS vector.

[image] WjjSteve: b.scatter_rev(std::plus<T>()); fem::set_bc(b.mutable_array(), {}); There should be a scatter forward after set_bc. Also there should be one after solve, Pre-backsubstitute. Please note that with only snippets, its hard to diagnose issues/bugs in your code

Thanks for replying! May I ask when should the function scatter_fwd() be applied? I read the Poisson demo here, and it seems to me that, this scatter_fwd function is not applied after set_bc, I guess it is because we only need to scatter forward when work in parallel? And this doesn’t really solv…

[image] WjjSteve: May I ask when should the function scatter_fwd() be applied? I read the Poisson demo here, and it seems to me that, this scatter_fwd function is not applied after set_bc, I guess it is because we only need to scatter forward when work in parallel? Scatter forward should be a…

Thanks for your patience and suggestions! Now I include all of the codes, if we consider the simplest case: f = g. g = cos(2pi x) in [0, 1] with periodic BC. Then the solution of f should be a cosine function as well, i.e., we should get f(0) = f(1) = 1 But when I solve this equation in paralle…

Hi, thanks for sharing! I will read through it. So have you solved it now?Or do you know how could fix it? It seems like the slaves dof is not in the same process with the master dof, so it is not properly updated.

Hi, Yes I just found that! By using u(mpc->function_space()), now it works! Thanks very much!

DOLFINX_MPC produce different results when work in parallel

dolfinx

tuderic January 9, 2024, 7:58am 6

Is it possible that your problem is also related to this post? In my case, the slave DOFs are also not updated by the backsubstitution process.

Topic		Replies	Views
MPC backsubstitution in parallel dolfinx mpi , dolfinx_mpc	28	916	January 9, 2024
Problem with dolfinx_mpc when assemble vetors in paraller dolfinx assembly , mpi	5	260	January 7, 2024
Discontinuities at MPI partition boundaries with manual periodic MPC on large high-order curved mesh dolfinx mpi , dolfinx_mpc	4	70	May 29, 2026
Different results in serial and parallel run dolfinx dolfinx	23	2778	October 26, 2020
Dolfinx_mpc parallel setup dolfinx boundary-conditions	1	49	February 10, 2026

DOLFINX_MPC produce different results when work in parallel

Related topics