This is a good question, Hissashi Rocha. In the article, I wrote that multiplying Z^T by Z, we get the covariance matrix up to a constant. The phrase “up to a constant” simply means that we’ll be off by a constant factor — in this case, we’re off by a factor of n (perhaps n-1 if we’re dealing with sample covariances, but I can’t say that with certainty at the moment). We could multiply everything by n, but it won’t impact our final results. We’re really only interested in the relationships between variables, so scaling everything effectively does nothing.