Functional data analysis in the Banach space of continuous functions (Q2196214)

The paper develops data analysis methodology for functional time series in the space of all continuous functions because, according to the authors, all functions utilized for practical purpose are continuous. The paper consists of five sections. In the second section, authors of the paper provide some basic facts about \textit{Central Limit Theorem }and \textit{Invariance Principle} for \(C(T)\)-valued random variables, where \(C(T)\) is the set of continuous functions from \(T\) into real line \(\mathbb{R}\). The third section deals with the two sample problem on the space \(C([0,1])\). Let \(\{X_1,\ldots,X_m\}\) and \(\{Y_1,\ldots,Y_n\}\) be two independent samples of \(C([0,1])\)-valued random variables. Under suitable assumptions expectation functions \(\mu_1=\mathbb{E}X_1\) and \(\mu_2=\mathbb{E}Y_1\) exist together with the covariance kernels. The authors of the paper consider properties of the maximal deviation between two mean curves \[d_\infty=\|\mu_1-\mu_2\|=\sup_{t\in[0,1]}|\mu_1(t)-\mu_2(t)|\] and provide procedure for testing the hypotheses of relevant difference: \[ H_0: d_\infty\leqslant \Delta\ \ {\text{versus}}\ \ H_1: d_\infty>\Delta, \] where \(\Delta\geqslant 0\) is a constant determined by the user of test. In the fourth section of the paper, the change point problem is considered. The new results are presented for testing of a change-point for triangular arrays of \(C([0,1])\)-valued random variables satisfying suitable requirements with respect to metric \(\rho(s,t)=|s-t|^\theta\), \(\theta\in(0,1]\). The simulation study of the derived procedures is described in the last section of the paper. The detailed proofs of the new results and the detailed simulation study investigating the finite sample properties of the new methodology are given in the supplementary materials \url{doi:10.1214/19-AOS1842SUPPA} and \url{doi:10.1214/19-AOS1842SUPPB}.