1: \begin{abstract}

2: We consider estimation in a sparse additive regression model with the design

3: points on a regular lattice. We establish

4: the minimax convergence rates over Sobolev classes and propose a

5: Fourier-based rate-optimal estimator which is adaptive to the unknown

6: sparsity and smoothness of the response function.

7: The estimator is derived within Bayesian formalism but

8: can be naturally viewed as a penalized

9: maximum likelihood estimator with the complexity penalties on the number of nonzero

10: univariate additive components of the response and on the numbers of the nonzero

11: coefficients of their Fourer expansions. We compare it with several

12: existing counterparts and perform a short simulation study to demonstrate its

13: performance.

14: \end{abstract}