Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

This is particularly directed at those who have Grimmett & Stirzaker, Probability and random processes (2005), at hand. It pertains to the proof step prior to equation (10), p. 166. For others:

  • $X_i$ are i.i.d. integer-valued random variables with $\mathbb{P}(X_i \leq 1) = 1$ and $\mathbb{P}(X_i = 1) > 0$.
  • $T_b=\min\{n:\sum_{i=1}^n X_i = b\}>0$ is the first hitting time of the point $b$.
  • $G(z) = \mathbb{E}\left(z^{-X_1}\right) = \sum_{n=-\infty}^1 z^{-n} \mathbb{P}(X_1=n)$
  • $F_b(z) = \mathbb{E}\left(z^{T_b}\right) = \sum_{n=0}^\infty z^n \mathbb{P}(T_b=n)$
  • equation (9): $F_b(z) = F_1(z)^b$ for $b \geq 1$

I'm not understanding how to prove $$ \mathbb{E}(\mathbb{E}(z^{T_1}|X_1)) = \mathbb{E}(z^{1+T_{1-X_1}}) = z \mathbb{E}\left(F_{1-X_1}(z)\right) = z \mathbb{E}\left(F_1(z)^{1-X_1}\right) = z F_1(z)G(F_1(z)) $$

In J. G. Wendel, "Left-continuous random walk and the Lagrange expansion" (1975), he argues in essence that $$ \begin{align*} \mathbb{E}(\mathbb{E}(z^{T_1}|X_1)) & = \sum_{n=-1}^\infty \mathbb{E}(z^{T_1}|X_1=-n)\ \mathbb{P}(X_1=-n) \\ & = z \mathbb{P}(X_1=1) + \sum_{n=0}^\infty \mathbb{E}(z^{T_1}|X_1=-n)\ \mathbb{P}(X_1=-n) & T_1=1 \Leftrightarrow X_1=1 \\ & = z \mathbb{P}(X_1=1) + \sum_{n=0}^\infty \mathbb{E}(z^{1+T_{1+n}})\ \mathbb{P}(X_1=-n) & \textrm{homogeneity} \\ & = z \mathbb{P}(X_1=1) + z \sum_{n=0}^\infty F_{1+n}(z)\ \mathbb{P}(X_1=-n) & \textrm{definition of }F \\ & = z \mathbb{P}(X_1=1) + z \sum_{n=0}^\infty F_1(z)^{1+n}\ \mathbb{P}(X_1=-n) & \textrm{equation (9)} \\ & = z F_1(z) \sum_{n=-1}^\infty F_1(z)^n\ \mathbb{P}(X_1=-n) \\ & = z F_1(z) G(F_1(z)) & \textrm{definition of }G \end{align*} $$

In the second line, $n=-1$ is treated specially, because the temporal and spatial homogeneity assumptions don't apply to it. Specifically, the further time required to hit 1 is 0; applying the homogeneity assumptions would imply $T_0=0$, which isn't allowed, hitting times are positive.

In the fifth line, applying equation (9), you can see that $n=-1$ needs to be treated specially there as well.

So although $\mathbb{E}(\mathbb{E}(z^{T_1}|X_1)) = z F_1(z)G(F_1(z))$, it seems that Grimmett's intermediate steps are nonsensical. Or can you see his logic?

Incidentally, $$ \mathbb{E}\left(F_1(z)^{1-X_1}\right) = \sum_{n=0}^\infty F_1(z)^n\ \mathbb{P}(1-X_1=n) = \sum_{n=-1}^\infty F_1(z)^{1+n}\ \mathbb{P}(X_1=-n) = F_1(z)G(F_1(z)) $$ so the subsequent application of Lagrange's inversion formula on p. 166 is still done correctly.

share|cite|improve this question
Thanks for pointing this out, Mike. There is indeed potential for confusion over the definition of $T_0$. We need $T_0=0$ at this stage (but not earlier nor later), and we are adding a note in the next reprint. – user34990 Jul 3 '12 at 14:25
@Geoffrey: Hello, and welcome to math.SE! I have converted your answer to a comment. Because you do not have 50 reputation points yet, you are only able to comment on your own questions and answers, so you were forced to post an answer; however, the "add comment" button will appear for you once you gain 50 points. Here is an explanation of reputation points. – Zev Chonoles Jul 3 '12 at 21:16

The key is that one defines $T_0=0$. Thus, $\mathrm E(z^{T_1}\mid X_1=1)=z$ is also $\mathrm E(z^{1+T_0})$ and, for every $k\leqslant0$, $\mathrm E(z^{T_1}\mid X_1=k)=\mathrm E(z^{1+T_{1-k}})$. Likewise, $F_0(z)=\mathrm E(z^{T_0})=1$, hence equation (9) holds for $b=0$ as well, and the computation of $\mathrm E(z^{T_1})$ which you recall becomes direct.

share|cite|improve this answer
You can't define $T_0=0$. It has a meaning already: the time that the walk returns to the origin. I'll add Grimmett's definition above. – Mike Feb 29 '12 at 7:04
Nevertheless this is the (only) way to make true G&S computations. In other words, if one sets $T_0=\inf\{n\geqslant1\mid S_n=0\}$, the line you try to understand is false but if one sets $T_0=\inf\{n\geqslant0\mid S_n=0\}$, that is $T_0=0$, this line holds. (Here, $S_0=0$ and $S_n=X_1+\cdots+X_n$.) – Did Feb 29 '12 at 7:58
By the way, you might wish to indicate where in the book G&S define $T_0$ in relation to equation (10) the way you say they do. – Did Mar 2 '12 at 8:43
You raise a good point. Middle of p. 164, $T_r$ is defined. But $S_n$ is defined on p. 162, and there $n=0$ would be an element of the sequence $S$. But that confuses the definition of $S_n=\sum_{i=1}^n X_i$. But in support of my interpretation of $T_0$, bottom of p. 162, "the random time $T_0$ until the particle makes its first return to the origin". – Mike Mar 2 '12 at 19:32
Then again, the definition of $T_0$ on p. 162 may be overridden by the one on p. 164. Hard to tell. It's all just a convention that needs clarification, and I've brought this post to Grimmett's attention. – Mike Mar 2 '12 at 19:46

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.