Enable to perform this operation

[yolhan83]

Hello, I'm sure the doc may mention it but from a quick reading I couldn't find a way out or just what's the issue in the following mwe

using TensorOperations
T = Float64
Z = zeros(T,512,8,24192);    # 512 X  8  X 24192
X = randn(T,512,16,24192); # 512 X 16 X 24192
Y = randn(T,16,8,24192);     #  16  X  8  X 24192
@tensor Z[a, d, n] = X[a, b,n] * Y[b, d,n]

I get TensorOperations.IndexError{String}("non-matching indices between left and right hand side: Z[a, d, n] = X[a, b, n] * Y[b, d, n]").

Thank you.

[lkdvos]

TensorOperations is a bit strict with the operations it supports, we require very strict Einsum notation: every index has to appear twice.
Either once before and once after the equals sign, denoting open/uncontracted/... indices, or twice after the equals sign, denoting traced/contracted/... indices.
In your example, the index n appears 3 times, and the specific operation you have is not supported by TensorOperations. In principle you could hijack the system and implement this by adding a delta array (d[n1, n2, n] which is 1 iff n1 == n2 == n). And using the following contraction:

d = zeros(size(Z, 3), size(X, 3), size(Y, 3))
for i in 1:size(Z, 3)
    d[i, i, i] = 1
end
@tensor Z[a, d, n] = X[a, b,n1] * Y[b, d,n2] * d[n1, n2, n]

[yolhan83]

Thank you !