Error reported for solving thermoelastic problem

[dokken]

@RokyC

  A_therm = dolfinx.fem.petsc.assemble_matrix(a_cpp_therm) 
    A_therm.assemble() 
    F_therm = dolfinx.fem.petsc.assemble_vector(f_cpp_therm)  
    F_therm.ghostUpdate(addv=petsc4py.PETSc.InsertMode.ADD, mode=petsc4py.PETSc.ScatterMode.REVERSE) 
    solve(A=A_therm,F=F_therm,X=thetanew,mesh=mesh)
    A_mecha = dolfinx.fem.petsc.assemble_matrix(a_cpp_mecha, bcs=[bc])
    A_mecha.assemble()
    F_mecha = dolfinx.fem.petsc.assemble_vector(f_cpp_mecha)
    dolfinx.fem.petsc.apply_lifting(F_mecha, [a_cpp_mecha], [bcs]) 
    F_mecha.ghostUpdate(addv=petsc4py.PETSc.InsertMode.ADD, mode=petsc4py.PETSc.ScatterMode.REVERSE)
    dolfinx.fem.petsc.set_bc(F_mecha, bcs)  
    solve(A=A_mecha,F=F_mecha,X=mechanew,mesh=mesh)

def solve(A,F,X,mesh):
    ksp = petsc4py.PETSc.KSP()
    ksp.create(mesh.comm)
    ksp.setOperators(A)        
    ksp.setType("preonly")  
    ksp.getPC().setType("lu")
    ksp.getPC().setFactorSolverType("mumps")
    ksp.setFromOptions()
    ksp.solve(F, X.vector)
    X.x.scatter_forward()

You should refactor this code such that you do not recreate matrices and solvers inside loops, as petsc has changed its garbage collection.

Manipulating vector data of fem.Function

This behavior is not safe, as dolfinx.fem.Function(..).vector does provide a view into data stored in the dolfinx.fem.Function(Vh). This data (and the underlying vector) is deleted when dl.fem.Functon(Vh) goes out of scope, which is right away. (This is due to PETSc’s modern garbage collection (dolfinx/function.py at main · FEniCS/dolfinx · GitHub, garbage collection different between petsc4py 3.17.4 and 3.18.3 (#1309) · Issues · PETSc / petsc · GitLab). Thus, if you want to use the petsc-vect…