Awesome

FEMBasis.jl contains interpolation routines for standard finite element function spaces. Given ansatz and coordinates of domain, interpolation functions are calculated symbolically in a very general way to get efficient code. As a concrete example, to generate basis functions for a standard 10-node tetrahedron one can write

code = FEMBasis.create_basis(
    :Tet10,
    "10 node quadratic tetrahedral element",
    [
     (0.0, 0.0, 0.0), # N1
     (1.0, 0.0, 0.0), # N2
     (0.0, 1.0, 0.0), # N3
     (0.0, 0.0, 1.0), # N4
     (0.5, 0.0, 0.0), # N5
     (0.5, 0.5, 0.0), # N6
     (0.0, 0.5, 0.0), # N7
     (0.0, 0.0, 0.5), # N8
     (0.5, 0.0, 0.5), # N9
     (0.0, 0.5, 0.5), # N10
    ],
    :(1 + u + v + w + u*v + v*w + w*u + u^2 + v^2 + w^2),
   )

The resulting code is

    struct Tet10 <: FEMBasis.AbstractBasis{3}
    end
    Base.@pure function Base.size(::Type{Tet10})
            return (3, 10)
        end
    function Base.size(::Type{Tet10}, j::Int)
        j == 1 && return 3
        j == 2 && return 10
    end
    Base.@pure function Base.length(::Type{Tet10})
            return 10
        end
    function FEMBasis.get_reference_element_coordinates(::Type{Tet10})
        return Tensors.Tensor{1,3,Float64,3}[[0.0, 0.0, 0.0], [1.0, 0.0, 0.0], [0.0, 1.0, 0.0], [0.0, 0.0, 1.0], [0.5, 0.0, 0.0], [0.5, 0.5, 0.0], [0.0, 0.5, 0.0], [0.0, 0.0, 0.5], [0.5, 0.0, 0.5], [0.0, 0.5, 0.5]]
    end
    function FEMBasis.eval_basis!(::Type{Tet10}, N::Vector{<:Number}, xi::Vec)
        assert length(N) == 10
        (u, v, w) = xi
        begin
            N[1] = 1.0 + -3.0u + -3.0v + -3.0w + 4.0 * (u * v) + 4.0 * (v * w) + 4.0 * (w * u) + 2.0 * u ^ 2 + 2.0 * v ^ 2 + 2.0 * w ^ 2
            N[2] = -u + 2.0 * u ^ 2
            N[3] = -v + 2.0 * v ^ 2
            N[4] = -w + 2.0 * w ^ 2
            N[5] = 4.0u + -4.0 * (u * v) + -4.0 * (w * u) + -4.0 * u ^ 2
            N[6] = +(4.0 * (u * v))
            N[7] = 4.0v + -4.0 * (u * v) + -4.0 * (v * w) + -4.0 * v ^ 2
            N[8] = 4.0w + -4.0 * (v * w) + -4.0 * (w * u) + -4.0 * w ^ 2
            N[9] = +(4.0 * (w * u))
            N[10] = +(4.0 * (v * w))
        end
        return N
    end
    function FEMBasis.eval_dbasis!(::Type{Tet10}, dN::Vector{<:Vec{3}}, xi::Vec)
        @assert length(dN) == 10
        (u, v, w) = xi
        begin
            dN[1] = Vec(-3.0 + 4.0v + 4.0w + 2.0 * (2u), -3.0 + 4.0u + 4.0w + 2.0 * (2v), -3.0 + 4.0v + 4.0u + 2.0 * (2w))
            dN[2] = Vec(-1 + 2.0 * (2u), 0, 0)
            dN[3] = Vec(0, -1 + 2.0 * (2v), 0)
            dN[4] = Vec(0, 0, -1 + 2.0 * (2w))
            dN[5] = Vec(4.0 + -4.0v + -4.0w + -4.0 * (2u), -4.0u, -4.0u)
            dN[6] = Vec(4.0v, 4.0u, 0)
            dN[7] = Vec(-4.0v, 4.0 + -4.0u + -4.0w + -4.0 * (2v), -4.0v)
            dN[8] = Vec(-4.0w, -4.0w, 4.0 + -4.0v + -4.0u + -4.0 * (2w))
            dN[9] = Vec(4.0w, 0, 4.0u)
            dN[10] = Vec(0, 4.0w, 4.0v)
        end
        return dN
    end
end

Also more unusual elements can be defined. For example, pyramid element cannot be descibed with ansatz, but it's still possible to implement by defining shape functions, Calculus.jl is taking care of defining partial derivatives of function:

code = FEMBasis.create_basis(
    :Pyr5,
    "5 node linear pyramid element",
    [
     (-1.0, -1.0, -1.0), # N1
     ( 1.0, -1.0, -1.0), # N2
     ( 1.0,  1.0, -1.0), # N3
     (-1.0,  1.0, -1.0), # N4
     ( 0.0,  0.0,  1.0), # N5
    ],
    [
     :(1/8 * (1-u) * (1-v) * (1-w)),
     :(1/8 * (1+u) * (1-v) * (1-w)),
     :(1/8 * (1+u) * (1+v) * (1-w)),
     :(1/8 * (1-u) * (1+v) * (1-w)),
     :(1/2 * (1+w)),
    ],
   )
eval(code)

Basis function can have internal variables if needed, e.g. variable dof basis like hierarchical basis functions or NURBS.

It's also possible to do some very common FEM calculations, like calculate Jacobian or gradient of some variable with respect to some coordinates. For example, to calculate displacement gradient du/dX in unit square [0,1]^2, one could write:

using Tensors
B = Quad4()
X = Vec.([(0.0, 0.0), (1.0, 0.0), (1.0, 1.0), (0.0, 1.0)])
u = Vec.([(0.0, 0.0), (1.0, -1.0), (2.0, 3.0), (0.0, 0.0)])
grad(B, u, X, Vec(0.0, 0.0))

Result is

2×2 Tensors.Tensor{2,2,Float64,4}:
 1.5  0.5
 1.0  2.0