Supplementary material for the paper The emergence of clusters in self attention dynamics

Neural Information Processing Systems 

This appendix is organized as follows: Appendix A: Well-posedness results. Throughout the remainder of the paper, we use the terminology "tokens" Definition 3 (Equi-compactly supported curves) . To prove Proposition A.1, we show a more general result concerning global existence and uniqueness We will make use of the following lemma regarding ( 6). R. We now show ( 8). R), which finally leads us to ( 9).