1: \begin{abstract}

2: 	Assume that we are given a filtration $(\mathscr F_n)$ on a probability space

3: 	$(\Omega,\mathscr F,\mathbb P)$ of the form that each $\mathscr F_n$ is generated

4: 	by the partition of one atom of $\mathscr F_{n-1}$ into two atoms of $\mathscr F_n$

5: 	having positive measure. Additionally, assume  that we are given a finite-dimensional linear space $S$ of

6: 	$\mathscr F$-measurable, bounded functions on $\Omega$ so that on each 

7: 	atom $A$ of any $\sigma$-algebra $\mathscr F_n$, all $L^p$-norms of functions

8: 	in $S$ are comparable independently of $n$ or $A$.

9: 	Denote by

10: 	$S_n$ the space of functions that are given locally, on atoms of $\mathscr F_n$,

11: 	by functions in $S$ and by $P_n$ the orthoprojector (with respect to the inner 

12: 	product in $L^2(\Omega)$) onto $S_n$.

13: 

14: 	Since $S = \lin\{\charfun_\Omega\}$ satisfies the above assumption and $P_n$ is 

15: 	then the conditional expectation $\mathbb E_n$ with respect to $\mathscr F_n$, for such 

16: 	filtrations, martingales $(\mathbb E_n f)$ are special cases of our setting.

17: 	

18: 	We show in this article that certain convergence results

19: 	that are known for martingales (or rather martingale differences) are also 

20: 	true in the general framework described above. More precisely, we show that the  

21: 	differences $(P_n - P_{n-1})f$  converge unconditionally and are 

22: 	democratic in $L^p$ for $1<p<\infty$. This implies that those differences 

23: 	form a greedy basis in $L^p$-spaces for $1<p<\infty$.

24: \end{abstract}

25: