1: \begin{abstract}

2:  The subspace $\mathcal{S}$ is called the central subspace of $(y,\vx)$, if it is the minimal subspace such that $y\independent \vx | P_{\mathcal{S}}\vx$. Looking for the central subspace is commonly referred to as {\it `sufficient dimension reduction'} (SDR). 

3: Sliced inverse regression (SIR) is arguably one of the most commonly used sufficient dimension reduction algorithm with a wide range of applications. We consider the minimax estimation of the central space in high dimensional setting. Assuming the dimension $d$ of central space is fixed, the linearity condition and coverage condition, we establish the optimal rates of convergence for estimating 	the central space  under mild conditions. The lower bound is obtained by an application of Fano's lemma. . We also make a speculation when $d$ is diverging at moderate rate.

4: 

5: 

6: \end{abstract}

7: