1: \begin{abstract}

2: This paper studies a \textit{partial functional partially

3: linear single-index model} that consists of a functional linear component as

4: well as a linear single-index component. This model generalizes many

5: well-known existing models and is suitable for more complicated data

6: structures. However, its estimation inherits the difficulties and complexities

7: from both components and makes it a challenging problem, which calls for new

8: methodology. We propose a novel profile B-spline method to estimate the

9: parameters by approximating the unknown nonparametric link function in the

10: single-index component part with B-spline, while the linear slope function in

11: the functional component part is estimated by the functional principal

12: component basis. The consistency and asymptotic normality of the parametric

13: estimators are derived, and the global convergence of the proposed estimator

14: of the linear slope function is also established. More excitingly, the latter

15: convergence is optimal in the minimax sense. A two-stage procedure is

16: implemented to estimate the nonparametric link function, and the resulting

17: estimator possesses the optimal global rate of convergence. Furthermore, the

18: convergence rate of the mean squared prediction error for a predictor is also

19: obtained. Empirical properties of the proposed procedures are studied through

20: Monte Carlo simulations. A real data example is also analyzed to illustrate

21: the power and flexibility of the proposed methodology.

22: \end{abstract}

23: