Why does conditioning on this probability give me this other probability?

95 Views Asked by At

From Stochastic Processes by Ross

At the top of page 75 (in the proof of lemma 2.3.4 on the bottom of page 74), the author writes (Summarized here):

Let $\tau_1,...,\tau_n$ denote the ordered values from a set of $n$ independent uniform $(0,t)$ random variables. Let $Y_1, \dots, Y_n$ be independent and iid nonnegative random variables that are also independent of $\{\tau_1, \dots, \tau_n\}$. Then $P(Y_1 + \dots + Y_k < \tau_k, k = 1 \dots n | Y_1 + \dots + Y_n = y) = \{1 - y/t \text{ if } 0 < y < t, 0 \text{ otherwise }\}$. We prove by induction. Suppose it is true for $n-1$. We condition on the values of $Y_1 + \dots + Y_{n-1}$ and $\tau_n$ and then using the fact that conditional on $\tau_n = u, \tau_1 \dots \tau_{n-1}$ are distributed as the order statistics from a set of $n-1$ uniform $(0,u)$ random variables. Doing so for $s < y$ implies $$P(Y_1 + \dots + Y_k < \tau_k, k = 1 \dots n | Y_1 + \dots + Y_{n-1} = s, \tau_n = u, Y_1 + \dots + Y_n = y)$$ $$= P(Y_1 + \dots + Y_k < \tau_k^{*}, k = 1 \dots n-1 | Y_1 + ... + Y_{n-1} = s)$$ for $y < u$, where $\tau_k^{*}$ are the ordered values from a set of $n-1$ independent uniform $(0,u)$ random variables.

I don't understand how he gets $$P(Y_1 + \dots + Y_k < \tau_k, k = 1 \dots n | Y_1 + \dots + Y_{n-1} = s, \tau_n = u, Y_1 + \dots + Y_n = y)$$ $$= P(Y_1 + \dots + Y_k < \tau_k^{*}, k = 1 \dots n-1 | Y_1 + ... + Y_{n-1} = s)$$ from "conditional on $\tau_n = u, \tau_1 \dots \tau_{n-1}$ are distributed as the order statistics from a set of $n-1$ uniform $(0,u)$ random variables."

It seems as if the part before the conditional in $P(Y_1 + \dots + Y_k < \tau_k, k = 1 \dots n | Y_1 + \dots + Y_{n-1} = s, \tau_n = u, Y_1 + \dots + Y_n = y)$ is not concerning the $\tau_n$ individually, but considering two random variables ($Y's$ and $\tau$'s) so I don't understand from which axioms of probability/theorems he is deriving this.

Anyone have any ideas?