Hello, I’d like to run my optimisation problem based on time-distributed control demo in parallel. I modified the demo to have multiple PDEs (of the same type but with different coefficients) which are solved in the for-loop. for i in range(PDEs_numb): solve(a[i] == L[i], u[i]) I’d li…

The following does what it should, but might be done in more efficient way. Cuz now the same optimization problem is solved on each process. I tried to use single process, but didn’t manage to. I don’t know how to call functions on processes other the 0, inside process 0. I.e. as long the optimizati…

Thank you for your reply! I’ve made an extention of what was suggested in the post you attached by gathering solution from the processes in order to use it in forward model (solve_heat). There are 3 methods in MPI: Gather, GartherV and gather. GartherV is not provided by dolfin, Gather can’t handle…

Sry, I missed one line of code Running this with mpirun -n 4 python3 filename.py the error can be reproduced from dolfin import * import numpy as np comm = MPI.comm_world size = comm.Get_size() rank = comm.Get_rank() numDataPerRank = 1 def poisson(alpha): mesh = UnitSquareMesh(256, 256) …

It still seems like you haven’t read the previous post I linked to. You are in your example creating the mesh with the global MPI communicator, aka. a distributed mesh. What you could do is to use the comm_self communicator to solve one problem with one mesh per processor. Then you can gather the l…

I’ve read that post carefully and tried mesh = UnitSquareMesh(MPI.comm_self, as well. Now I understand that this is the main point of running my code in parallel, but it didn’t make any difference in terms of error I was getting, thats why I excludede it from MWE. But what I didn’t know, is that w…

I do not quite follow what you are trying to achieve with the time dependent problem. For each separate parameter input, you can solve the PDE on a separate processor. You cannot distribute time steps over multiple processors (as I think you also state in your setting). In the setting of the opti…

This is the initial problem: from fenics import * from fenics_adjoint import * from collections import OrderedDict data = Expression("16*x[0]*(x[0]-1)*x[1]*(x[1]-1)*sin(pi*t)", t=0, degree=4) nu = Constant(1e-5) # number of parapeters (i.e. PDEs) PDEs_numb = 30 # this is an array of 30 parameters…

I don’t see the immediate problem with the while-loop, as you can add an else statement for the processes that are not rank 0, and have them wait for data from the main process’ while loop. I could see how it is very inefficient with regards to a lot of data transfers between processes though, and …

First of all thank you for your detailed answer! I can’t get how else-statement could help construct forward model on single process and what should be under this else-statement? The first idea was to use the code I previously posted, just for finding PDE solutions in parallel and constructing for…

I haven’t worked with MPI directly, so I might be wrong or doing something that is bad practice, but in psuedo-code, this is what I imagined. Note that I do not advocate for this model, as I said previously I think it is better to split it up before the time while-loop. # Setup problem ... if ran…

Parallel for-loop with multiple solve-calls, MPI

dolfin-adjoint

dokken October 29, 2020, 5:45pm 2

I would suggest having a look at: How to solve in parallel for parametric study

Topic		Replies	Views
Parallelization with multiple solves dolfin-adjoint	0	644	October 31, 2019
Different results NewtonSolve parallel mpi.comm_world dolfinx mpi	13	277	January 30, 2025
Pyadjoint crash when using project function dolfin-adjoint	20	4526	January 20, 2021
IPOPT doesn't work in Parallel with MPI dolfin-adjoint	12	2189	August 20, 2025
Scalability for adaptive time-stepping General	3	407	August 28, 2023

Parallel for-loop with multiple solve-calls, MPI

Related topics