Signatures of algebraic curves via numerical algebraic geometry (Q2674020)

The topic of this paper is to decide whether two plane algebraic curves are equal up to action of the group of rigid body motions or other subgroups of the projective group. Problems of this type can be studied via differential invariants. For example in the case of rigid body motions the curves' arclength and curvature must match. It is, however, conceptually simpler, also for implementation, to merge suitable differential invariants into a ``signature curve''. An alternative approach uses ``joint signatures'' which rely only on invariants of order zero and are considered to be more robust against noise. The thus obtained theory is not only capable of detecting symmetries between two curves, it can also report the size of the symmetry group. Details of definition and precise statements on the information value to be extracted are too technical to be reported here. Besides improving theoretical underpinnings, this article introduces for the first time numerical algebraic geometry to signature based symmetry detection between curves. The curves are represented by witness sets or sampling oracles, respectively and membership tests are performed via homotopy continuation. This leads to a probabilistic algorithm that is both feasible for curves of degree \(d \le 6\) (the paper also reports successful runs for \(d \le 10\)) and robust against noise. The theory comes with an implementation in \texttt{Macaulay2} that is publicly available. It is described in Section~4 together with particular examples and experimental results.