A question about the computation of noise variance in modified k-means algorithm

[kekehia123]

In Pascual-Marqui et al. (1995)'s paper, the noise variance is calculated as follows:

in which "The orthogonal squared distance between each measurement vector and each microstate is computed" as

I'm wondering why the distance is defined like this. It seems the second term with a square does not match the first term. It seems that an absolute operation is proper? Does anyone have some explanations? Thanks a lot!