Linear 2D Elasticity - pure Neumann problem - Krylov null space - convergence problem

Question

Dear,

I have convergence problems with Krylov solver for my problem. I would like to find a solution of 2D linear elastic plate with pure Neumann problem (only traction on the boundary). My geometry is Rectangle with Rectangular hole, where stress in x-direction is specified on a left and right part of outer rectangle. Solution has problems with the Krylov solver. Even if hole is removed, problem remains.

Thanks for any help, hints.

Regards,
AG

Below is my working example:


from mshr import *
from dolfin import *

#set_log_level(1)

# Test for PETSc
if not has_linear_algebra_backend("PETSc"):
    print("DOLFIN has not been configured with PETSc. Exiting.")
    exit()
    
# Do it parallel
parameters["num_threads"] = 4;

# Set backend to PETSC
parameters["linear_algebra_backend"] = "PETSc"

class FreeBoundary(SubDomain):
    def inside(self, x, on_boundary):
        return on_boundary

class LeftBoundary(SubDomain):
    def inside(self, x, on_boundary):
        return on_boundary and abs(x[0] + 2) < DOLFIN_EPS

class RightBoundary(SubDomain):
    def inside(self, x, on_boundary):
        return on_boundary and abs(x[0] - 2) < DOLFIN_EPS


def build_nullspace(V, x):
    #Function to build null space for 2D elasticity
    #
    # Create list of vectors for null space
    nullspace_basis = [x.copy() for i in range(2)]

    # Build translational null space basis - 2D problem
    #
    V.sub(0).dofmap().set(nullspace_basis[0], 1.0);
    V.sub(1).dofmap().set(nullspace_basis[1], 1.0);

    for x in nullspace_basis:
        x.apply("insert")

    # Create vector space basis and orthogonalize
    basis = VectorSpaceBasis(nullspace_basis)
    basis.orthonormalize()

    return basis

# Initialize subdomain instances
free_b = FreeBoundary()
left_b = LeftBoundary()
right_b = RightBoundary()
    
# Create geo and mesh
res = 100
r1 = Rectangle(Point(-2., -2.), Point(2., 2.))
r2 = Rectangle(Point(-0.5, -0.5), Point(0.5, 0.5))
geo = r1-r2
mesh = generate_mesh(geo,res)

# Initialize mesh function for boundary domains
boundaries = FacetFunction("size_t", mesh)
boundaries.set_all(0)
free_b.mark(boundaries, 1)
left_b.mark(boundaries, 2)
right_b.mark(boundaries, 3)

# Create function space
V = VectorFunctionSpace(mesh, 'Lagrange', 1)

# Create test and trial functions, and source term
u = TrialFunction(V)
v = TestFunction(V)

# Define new measures associated with the exterior boundaries
ds = Measure('ds', domain=mesh, subdomain_data=boundaries)

# Elasticity parameters
E, nu = 10, 0.3
mu, lmbda = E/(2.0*(1.0 + nu)), E*nu/((1.0 + nu)*(1.0 -2.0*nu))

# Stress relation
I = Identity(2)
epsilon = sym(nabla_grad(u))
sigma = 2*mu*epsilon + lmbda*tr(epsilon)*I

# Boundary conditions - Neumann (stress on L&R side)
gL = Constant((-1.0, 0.0))
gR = Constant((1.0, 0.0))

# Governing balance equation
a = inner(sigma, sym(nabla_grad(v)))*dx
L = dot(gL, v)*ds(2) + dot(gR, v)*ds(3)

A, b = assemble_system(a, L)
print "Norm(b)=",b.norm("l2") # Check 

# Set up PDE and solve
u = Function(V)

# AMG). The solution vector is passed so that it can be copied to
# generate compatible vectors for the nullspace.
null_space = build_nullspace(V, u.vector())

# Attach near nullspace to matrix and orthogonalize b
as_backend_type(A).set_near_nullspace(null_space)
null_space.orthogonalize(b) 

# Create PETSC smoothed aggregation AMG preconditioner and attach near
# null space
pc = PETScPreconditioner("petsc_amg")

# Use Chebyshev smoothing for multigrid
PETScOptions.set("mg_levels_ksp_type", "chebyshev")
PETScOptions.set("mg_levels_pc_type", "jacobi")

# Improve estimate of eigenvalues for Chebyshev smoothing
PETScOptions.set("mg_levels_esteig_ksp_type", "cg")
PETScOptions.set("mg_levels_ksp_chebyshev_esteig_steps", 50)

# Create CG Krylov solver and turn convergence monitoring on
solver = PETScKrylovSolver("cg", pc)
solver.parameters["monitor_convergence"] = True

# Set matrix operator
solver.set_operator(A);

# Compute solution
solver.solve(u.vector(), b);

plot(u, mode = "displacement", interactive=True, scalarbar=True, title="u")

Hi, the first problem I see is with your nullspace. The space of rigid motions in 2d is three dimensional - 2 translations and 1 rotation. Your nullspace basis has no rotations. — MiroK, Apr 13, 2016

Gusar · Answer 1 · Apr 13, 2016

Now it works. Changed to "bicgstab" KrylovSolver.

u = Function(V)

null_space = build_nullspace(V, u.vector())

as_backend_type(A).set_near_nullspace(null_space)
null_space.orthogonalize(b)

solver = KrylovSolver("bicgstab")
solver.parameters["relative_tolerance"] = 1e-10
solver.parameters["monitor_convergence"] = True

solver.set_operator(A);

solver.solve(u.vector(), b);

Linear 2D Elasticity - pure Neumann problem - Krylov null space - convergence problem

1 Answer