For the gaussian approximation on CESM matrix W is 5.7GB

[AlessandroLovo]

The matrix $W$ is what we use to compute the H2 norm of the projection pattern $p$ and thus is a $d\times d$ matrix that can get very very big.
However $W$ is also very sparse (99.98% of its entries are zero), so we should be able to save it in a much more memory efficient way

[AlessandroLovo]

We can use scipy.sparse.dia_matrix