EMLP Discussion Update about dmol-book HOT 4 CLOSED

from dmol-book.

Yes this can be true for non permutation representations and is also why for visualization the recommendation for non permutation representations is to use vis(repin,repout,cluster=False) otherwise the visualization can be misleading. It may be possible to add functionality to vis to check whether the rep is a permutation and set cluster accordingly.

A good example is T(3)(SO(3)) which has the equivariant solution εᵢⱼₖ (the levi civita tensor) .
It forms a one dimensional equivariant subspace but has both positive and negative entries.

Although actually now that I'm looking at your example, maybe this is not what you were referring to?

from dmol-book.

@mfinzi I was trying to examine in my plots what is the effect of the equivariant linear basis on the weights. You're the expert, can you review these few paragraphs and tell me if I'm understanding emlp correctly? Thanks!

https://whitead.github.io/dmol-book/dl/Equivariant.html#how-the-constraints-work

from dmol-book.

Fixed by 700c749

from dmol-book.