Tensors and vectors

using MuesliMaterials

MUESLI carries its own small fixed-size types rather than using a general array library. This package wraps four of them, all in three dimensions:

TypeShapeNotes
Ivector3a vector
Itensor3 × 3a general second-order tensor
Istensor3 × 3symmetric; stores only six components
Itensor43 × 3 × 3 × 3a fourth-order tensor

Construction

Each type builds from the corresponding Julia array:

v = Ivector([1.0, 2.0, 3.0])
MuesliMaterials.IvectorAllocated with 3 entries
1.0
2.0
3.0
m = [1.0 2.0 3.0     4.0 5.0 6.0     7.0 8.0 9.0]t = Itensor(m)
MuesliMaterials.ItensorAllocated with 3x3 entries
1.0 2.0 3.0
4.0 5.0 6.0
7.0 8.0 9.0

Julia stores matrices column-major and MUESLI's constructors are row-major, so the conversion reorders the components. The round trip is faithful — in particular nothing is transposed:

t[1, 2], t[2, 1]
(2.0, 4.0)

Istensor is symmetric and keeps only six components, so it reads the symmetric part of what you give it:

s = Istensor([1.0 0.2 0.3              0.2 2.0 0.4              0.3 0.4 3.0])
MuesliMaterials.IstensorAllocated with 3x3 entries
1.0 0.2 0.3
0.2 2.0 0.4
0.3 0.4 3.0

A fourth-order tensor takes a 3×3×3×3 array:

c = Itensor4(reshape(collect(1.0:81.0), 3, 3, 3, 3))c[1, 2, 3, 1]
22.0

Sizes are checked, so a wrong shape is an error rather than a silent misread:

try    Itensor(zeros(2, 2))catch err    errend
ErrorException("Input has to be an array of size 3x3, got 4 element(s).")

They behave like Julia arrays

Indexing is 1-based, and slicing, axes, length and iteration all work:

t[1, :], t[:, 2]
([1.0, 2.0, 3.0], [2.0, 5.0, 8.0])
axes(t), length(t), ndims(t)
((Base.OneTo(3), Base.OneTo(3)), 9, 2)
sum(t)          # iteration
45.0

Assignment works too, including into slices:

t[1, 1] = 100.0t[2, 1:3] = [7.0, 8.0, 9.0]t
MuesliMaterials.ItensorAllocated with 3x3 entries
100.0 2.0 3.0
7.0 8.0 9.0
7.0 8.0 9.0

Converting back

convert produces an ordinary Julia array:

convert(Matrix{Float64}, s)
3×3 Matrix{Float64}:
 1.0  0.2  0.3
 0.2  2.0  0.4
 0.3  0.4  3.0
convert(Array{Float64}, c)[1, 2, 3, 1]
22.0

Interop with Tensors.jl

Loading Tensors.jl activates an extension that converts to its static tensor types, which is what you want inside an element routine:

using Tensorsconvert(Tensor{2, 3}, Itensor([1.0 2.0 3.0; 4.0 5.0 6.0; 7.0 8.0 9.0]))
3×3 Tensors.Tensor{2, 3, Float64, 9}:
 1.0  2.0  3.0
 4.0  5.0  6.0
 7.0  8.0  9.0
convert(SymmetricTensor{2, 3}, s)
3×3 Tensors.SymmetricTensor{2, 3, Float64, 6}:
 1.0  0.2  0.3
 0.2  2.0  0.4
 0.3  0.4  3.0