Supplementary Materials for Learning Physical Dynamics with Subequivariant Graph Neural Networks

In the main paper, we have sketched the definitions and conclusions related to E(3) equivariance and subequivariance. Here, we introduce more details to facilitate the understanding for these conceptions. Basically, E(3) is a group of orthogonal transformations (rotations and reflections) and translations. Moreover, by joining the analyses in GMN along with the conclusion by [11], we immediately have the universality of the formulation in Eq. (13): Proposition 1 ([5, 11]). Let f be defined by Eq. (13).