1: \begin{abstract}

2: We develop a canonical framework for the study of the problem of

3: registration of multiple point processes subjected to warping, known as

4: the problem of separation of amplitude and phase variation. The

5: amplitude variation of a real random function $\{Y(x):x\in[0,1]\}$

6: corresponds to its random oscillations in the $y$-axis, typically

7: encapsulated by its (co)variation around a mean level. In contrast, its

8: phase variation refers to fluctuations in the $x$-axis, often caused by

9: random time changes. We formalise similar notions for a point process,

10: and nonparametrically separate them based on realisations of i.i.d.

11: copies $\{\Pi_i\}$ of the phase-varying point process. A key element in

12: our approach is to demonstrate that when the classical phase variation

13: assumptions of Functional Data Analysis (FDA) are applied to the point

14: process case, they become equivalent to conditions interpretable

15: through the prism of the theory of optimal transportation of measure.

16: We demonstrate that these induce a natural Wasserstein geometry

17: tailored to the warping problem, including a formal notion of bias

18: expressing over-registration. Within this framework, we construct

19: nonparametric estimators that tend to avoid over-registration in finite

20: samples. We show that they consistently estimate the warp maps,

21: consistently estimate the structural mean, and consistently register

22: the warped point processes, even in a sparse sampling regime. We also

23: establish convergence rates, and derive $\sqrt{n}$-consistency and a

24: central limit theorem in the Cox process case under dense sampling,

25: showing rate optimality of our structural mean estimator in that case.

26: \end{abstract}