Helmholtz equation

In this example, we want to solve a (variant of) of the Helmholtz equation. The example is inspired by an dealii step_7 on the standard square.

\[ - \Delta u + u = f\]

With boundary conditions given by

\[u = g_1 \quad x \in \Gamma_1\]


\[n \cdot \nabla u = g_2 \quad x \in \Gamma_2\]

Here Γ₁ is the union of the top and the right boundary of the square, while Γ₂ is the union of the bottom and the left boundary.

We will use the following weak formulation:

\[\int \nabla δu \cdot \nabla u d\Omega + \int δu \cdot u d\Omega - \int δu \cdot f d\Omega + \int δu \cdot (n \cdot \nabla u - g_2) d\Gamma_2\]

where $δu$ is a suitable test function that satisfies:

\[δu = 0 \quad x \in \Gamma_1\]

and $u$ is a suitable function that satisfies:

\[u = g_1 \quad x \in \Gamma_1\]

The example highlights the following interesting features:

  • There are two kinds of boundary conditions, "Dirichlet" and "Von Neumann"
  • The example contains boundary integrals
  • The Dirichlet condition is imposed strongly and the Von Neumann condition is imposed weakly.
using JuAFEM
using Tensors
using SparseArrays
using LinearAlgebra

const ∇ = Tensors.gradient
const Δ = Tensors.hessian;

grid = generate_grid(Quadrilateral, (150, 150))

dim = 2
ip = Lagrange{dim, RefCube, 1}()
qr = QuadratureRule{dim, RefCube}(2)
qr_face = QuadratureRule{dim-1, RefCube}(2)
cellvalues = CellScalarValues(qr, ip);
facevalues = FaceScalarValues(qr_face, ip);

dh = DofHandler(grid)
push!(dh, :u, 1)
    :u, interpolation: Lagrange{2,RefCube,1}(), dim: 1
  Dofs per cell: 4
  Total dofs: 22801

We will set things up, so that a known analytic solution is approximately reproduced. This is a good testing strategy for PDE codes and known as the method of manufactured solutions.

function u_ana(x::Vec{2, T}) where {T}
    xs = (Vec{2}((-0.5,  0.5)),
          Vec{2}((-0.5, -0.5)),
          Vec{2}(( 0.5,  -0.5)))
    σ = 1/8
    s = zero(eltype(x))
    for i in 1:3
        s += exp(- norm(x - xs[i])^2 / σ^2)
    return max(1e-15 * one(T), s) # Denormals, be gone

dbcs = ConstraintHandler(dh)
  Not closed!

The (strong) Dirichlet boundary condition can be handled automatically by the JuAFEM library.

dbc = Dirichlet(:u, union(getfaceset(grid, "top"), getfaceset(grid, "right")), (x,t) -> u_ana(x))
add!(dbcs, dbc)
update!(dbcs, 0.0)

K = create_sparsity_pattern(dh);

function doassemble(cellvalues::CellScalarValues{dim}, facevalues::FaceScalarValues{dim},
                         K::SparseMatrixCSC, dh::DofHandler) where {dim}
    b = 1.0
    f = zeros(ndofs(dh))
    assembler = start_assemble(K, f)

    n_basefuncs = getnbasefunctions(cellvalues)
    global_dofs = zeros(Int, ndofs_per_cell(dh))

    fe = zeros(n_basefuncs) # Local force vector
    Ke = zeros(n_basefuncs, n_basefuncs) # Local stiffness mastrix

    @inbounds for (cellcount, cell) in enumerate(CellIterator(dh))
        fill!(Ke, 0)
        fill!(fe, 0)
        coords = getcoordinates(cell)

        reinit!(cellvalues, cell)

First we derive the non boundary part of the variation problem from the destined solution u_ana

\[\int \nabla δu \cdot \nabla u d\Omega + \int δu \cdot u d\Omega - \int δu \cdot f d\Omega\]
        for q_point in 1:getnquadpoints(cellvalues)
            dΩ = getdetJdV(cellvalues, q_point)
            coords_qp = spatial_coordinate(cellvalues, q_point, coords)
            f_true = -LinearAlgebra.tr(hessian(u_ana, coords_qp)) + u_ana(coords_qp)
            for i in 1:n_basefuncs
                δu = shape_value(cellvalues, q_point, i)
                ∇δu = shape_gradient(cellvalues, q_point, i)
                fe[i] += (δu * f_true) * dΩ
                for j in 1:n_basefuncs
                    u = shape_value(cellvalues, q_point, j)
                    ∇u = shape_gradient(cellvalues, q_point, j)
                    Ke[i, j] += (∇δu ⋅ ∇u + δu * u) * dΩ

Now we manually add the von Neumann boundary terms

\[\int δu \cdot (n \cdot \nabla u - g_2) d\Gamma_2\]
        for face in 1:nfaces(cell)
            if onboundary(cell, face) &&
                   ((cellcount, face) ∈ getfaceset(grid, "left") ||
                    (cellcount, face) ∈ getfaceset(grid, "bottom"))
                reinit!(facevalues, cell, face)
                for q_point in 1:getnquadpoints(facevalues)
                    coords_qp = spatial_coordinate(facevalues, q_point, coords)
                    n = getnormal(facevalues, q_point)
                    g = gradient(u_ana, coords_qp) ⋅ n
                    dΓ = getdetJdV(facevalues, q_point)
                    for i in 1:n_basefuncs
                        δu = shape_value(facevalues, q_point, i)
                        fe[i] += -(δu * g) * dΓ
                        for j in 1:n_basefuncs
                            ∇u = shape_gradient(cellvalues, q_point, j)
                            Ke[i, j] += (δu * ∇u ⋅ n) * dΓ

        celldofs!(global_dofs, cell)
        assemble!(assembler, global_dofs, fe, Ke)
    return K, f

K, f = doassemble(cellvalues, facevalues, K, dh);
apply!(K, f, dbcs)
u = Symmetric(K) \ f;

vtkfile = vtk_grid("helmholtz", dh)
vtk_point_data(vtkfile, dh, u)
println("Helmholtz successful")
Helmholtz successful

This page was generated using Literate.jl.