1: \begin{definition} %Let %We will use the following definitions in our proof.

2: %We use the following definitions.

3: Define

4: \begin{enumerate}

5: 

6: %\item For simplicity we drop the subscript $j$ in the subspace and its estimates: $\P_{-1} := \P_{j-1}$, $\P := \P_j$, $\Phat_{-1} := \Phat_{j-1}$, $\Phat = \Phat_j$ and $\Phat_{k} := \Phat_{j, k}$ for $k = 1, 2, \cdots, K$.

7: %\item Subspace Estimation Error: $\zeta_k:=\SE(\Phat_{j, k}, \P_j)$ for $k=1,2,\dots, K$ and $j =0, 1, 2, \cdots, J$

8: 

9: %\item $\seinit:=\SE(\Phat_0,\P_0)$

10: %\item When stating Theorem \ref{thm1}, we just used numerical constants $c_1, c_2, C_1$ for simplicity. The result holds with $c_1=0.01$, $c_2 = 0.01$, and $C_1= 15\sqrt{\eta}$.

11: 

12: \item the constants used in Theorem \ref{thm1}: $c_1=0.01$, $c_2 = 0.01$, and $C_1= 15\sqrt{\eta}$

13: 

14: \item  $s:= \outfraccol \cdot n$

15: \item $\phi^+ = 1.2$

16: \item bound on $\outfracrow^\alpha$: $b_0:= 0.01/f^2$.%denote the bound on $\outfracrow$ given in the theorem

17: 

18: \item  $q_0 := 1.2(\zz + \SE(\P_{j-1}, \P_j))$, $q_{k} = (0.3)^{k}q_0$

19: %\item Subspace Error Bounds: $\zeta_{j, 0}^+ := q_0/4$, $\zeta_{j, k}^+ := (\phi^+/4)^{k-1} (q_0/4)$ for all $j$ and $k$. %$\zeta_k^+ := \phi^+(\zz + \Delta)/4^k$

20: 

21: \item $\et := \xhatt - \xt$. Since $\vt=0$,  $\et = \lt - \lhatt$

22: 

23: \item Events: $\Gamma_{0,0}:= \{\text{assumed bound on } \SE(\Phat_0,\P_0)\}$,

24: \\ $\Gamma_{0, k}:= \Gamma_{0,k-1} \cap \{\SE(\Phat_{0,k},\P_0) \le 0.3^k \SE(\Phat_0,\P_0) \}$, %$\Gamma_{1, 0}:= \Gamma_{0,K}$,

25: \\  $\Gamma_{j, 0} := \Gamma_{j-1, K}$, $\Gamma_{j, k} := \Gamma_{j, k-1} \cap \{\SE(\Phat_{j, k}, \P_j) \le q_{k-1}/4 \}$ for $j=1,2,\dots, J$ and $k=1,2,\dots, K$.

26: 

27: \item Using the expression for $K$ given in the theorem, and since $\Phat_j = \Phat_{j,k}$ (from the Algorithm), it follows that $\Gamma_{j,K}$ implies $\SE(\Phat_j, \P_j)=\SE(\Phat_{j, K}, \P_j) \leq \zz$.

28: \end{enumerate}

29: \label{defs}

30: \end{definition}

31: