1: \begin{proof}

2: (i) Consider a path as in (3.7). The point is that, as we run through the 

3: vertices $\gamma_i$, each time the addition of a $\gamma_i$ increases the 

4: dimension of the space $H_P$ by one, the edge corresponding to the commutator $[\gamma_i,\gamma_{i-1}]$ is independent of all the previous edges along the path. This is an immediate consequence of our definition of $G_{m,k}$, along with an application of LIP if one wants to be completely rigorous. The inequality for 

5: $\mathbb{P}(B_P)$ follows immediately in turn. When the dimension of $H_P$ is maximal then all the edges along the path are independent, which gives equality in that case.

6: \\

7: \\

8: (ii) Consider paths as in (3.7) again. We must estimate the number of ways 

9: we can choose the ordered sequence $(\gamma_1,...,\gamma_{l-1})$ of vertices, assuming that dim$(H_P) = t$. We can just as well start with $a$ and $b$, which span a $2$-dimensional space, and choose the remaining $\gamma_i$ in order, so that the dimension increases by $t-2$ in all. Each time we choose a new $\gamma_i$ we must decide whether to increase the dimension by one or not. Each time we do the former, there are $(1-o_{n}(1))n$ choices for $\gamma_i$. Each time we do the latter, there are certainly no more than $2^{l} = O_l(1)$ choices for $\gamma_i$. We will get another $O_l(1)$ factor from the freedom to choose on which $t-2$ occasions the dimension is to be increased. But clearly the result is that

10: we have $\Theta_l(1) \cdot n^{t-2}$ choices for the path, as claimed. 

11: \\

12: \\

13: The argument for (iii) and (iv) follows the same lines as that for (i), while the argument for (v) follows the same lines as that for (ii). We omit further details.    

14: \end{proof}

15: