1: \begin{abstract}

2: Functional data that are nonnegative and have a constrained integral

3: can be considered as samples of one-dimensional density functions. Such

4: data are ubiquitous. Due to the inherent constraints, densities do not

5: live in a vector space and, therefore, commonly used Hilbert space

6: based methods of functional data analysis are not applicable. To

7: address this problem, we introduce a transformation approach, mapping

8: probability densities to a Hilbert space of functions through a

9: continuous and invertible map. Basic methods of functional data

10: analysis, such as the construction of functional modes of variation,

11: functional regression or classification, are then implemented by using

12: representations of the densities in this linear space. Representations

13: of the densities themselves are obtained by applying the inverse map

14: from the linear functional space to the density space. Transformations

15: of interest include log quantile density and log hazard

16: transformations, among others. Rates of convergence are derived for the

17: representations that are obtained for a general class of

18: transformations under certain structural properties. If the

19: subject-specific densities need to be estimated from data, these rates

20: correspond to the optimal rates of convergence for density estimation.

21: The proposed methods are illustrated through simulations and

22: applications in brain imaging.

23: \end{abstract}