You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This PR changes all links to the new location of the
repository. Additionally pybind11 is now written consistently
and according to their naming convention.
"Given that the number of features grows exponentially with body order, some kind of truncation must be applied. NICE features incorporate a Contraction step to compress information at each body order iteration. See the [NICE library](https://github.com/cosmo-epfl/nice/) for a full-fledged implementation. \n",
665
+
"Given that the number of features grows exponentially with body order, some kind of truncation must be applied. NICE features incorporate a Contraction step to compress information at each body order iteration. See the [NICE library](https://github.com/lab-cosmo/nice/) for a full-fledged implementation. \n",
666
666
"\n",
667
667
"One way to check the amount of information that is lost due to truncation of the NICE iteration (or of the basis set!) is to apply the sum rule from [Goscinski et al. (2021)](https://arxiv.org/pdf/2105.08717), which is a consequence of the orthogonality of CG coefficients. Basically, _for each environment_ the norm of the full $\\nu$-neighbors feature vector is equal to the norm of the $\\nu=1$ equivariants (the density expansion coefficients) raised to the power $\\nu$. Note the norm is to be intended over all equivariant terms, i.e.\n",
0 commit comments