adler-j / bwgan Goto Github PK

We are doing project on your paper. We've read your code for sobolev_filter and faced problems with understanding the trick you use for \xi grid. As far as we understood, the straight-forward way to do scaling is to make a meshgrid using range(sx) / sx and range(sy) / sy. But we found out that you do it in a symmetric-way, i.e. minimum(x, sx - x) / (sx // 2). We wonder why you do it like that? Are there any reasons like numerical stability, etc.? It is also interesting why one needs to limit |\xi| <= 5, does constant "c" in your code have any correlation with this constraint? And if yes, why it is not in the Sobolev-norm equation?