Mystifying `sklearn.decomposition.SparsePCA` behavior

Question

I have been playing around with SparsePCA and I notice that on occasion some of the "principal components" are identically zero. This seems peculiar -