Super_lu dist memory error

Question

Hello,

I am trying to solve a Navier-Stokes system over a very large mesh with the direct solver super_lu_dist.

For some reason when I run the simulation I get an error that states the computer runned out of memory, which is completely false (I am runing the code in only one thread in a computer with 100GB of ram memory).

Note: if I coarse the mesh then the solver works perfectly, with good results. Nevertheless, for theoretical reasons, I need to solve the problem in a very fine mesh.

Thank you in advice guys

Comments

How many degrees of freedom are in your problem, and which element type and discretisation scheme are you using?

---

Total # of dofs: 1 646 725
Elements: P1 bubble for velocity and P1 for pressure
The discretization in time is implicit except for the non-linear terms, where a semi-implicit approach is used (which avoid to solve a non-linear system of equations).

Thx!

---

I don't know the details of the element type you're using. But Taylor-Hood elements should fit into memory and superlu_dist (or mumps) should work fine.

---

for some reason I cannot use mumps. Back in the day I use it in an old installation I had with fenics 1.6.

Now, if I do list_linear_solver_methods() I get

LU method | Description

default | default LU solver
petsc | PETSc built in LU solver
superlu_dist | Parallel SuperLU

Do you know why this happen?

Thank you again!