Special numerical techniques are already needed to deal with n à n matrices for large n. Tensor data are of size n à n Ã...à n=nd, where nd exceeds the computer memory by far. Since standard methods fail, a particular tensor calculus is needed to treat such problems.