# Computing $P(\limsup S_n=\infty)$ and $P(\liminf S_n=-\infty)$ for a random walk

Let $X_i$ be i.i.d. non-degenerate random variables with $EX_i=0$, and consider the random walk $S_n = \sum_{i=1}^n X_i$.

If $X_i$ was a simple random walk (i.e., $P(X_i = -1) = 1/2 = P(X_i = 1)$), I am able to show that $$P(\limsup S_n = \infty) = 1 = P(\liminf S_n = -\infty).$$

In my proof of the above statement, I do so by first showing that infinitely often the random walk must take $M$ consecutive positive steps infinitely often, for all $M\in\mathbb{N}$, with probability $1$. Similarly, one can show that the random walk must take $M$ consecutive negative steps infinitely often with probability $1$. (I am also aware that there are other questions on this site that address the above statement.)

However, my question is can this be generalized beyond this simple case of a simple random walk for $X_i$ following an arbitrary, non degenerate distribution with mean $0$? And if so, how does one go about a proof? In the case of the simple random walk we had a nice clean characterization of the distribution of the $X_i$, but in a generalized version we do not. Thanks!

• The probabilities are both equal to 1. When variance is finite this follows immediately from the Law of Iterated Logarithm: en.wikipedia.org/wiki/Law_of_the_iterated_logarithm. The result is true in general but I don't recall the proof at this moment. – Kavi Rama Murthy Feb 22 '18 at 7:41
• FYI, the fact "that infinitely often the random walk must take M consecutive positive steps infinitely often, for all M∈ℕ, with probability 1" does not imply that the limsup is + infinity almost surely. – Did Feb 22 '18 at 13:47
• @Did - I know. I did not mean to say that this statement implies the result. It was an intermediate step I showed in the simple Bernoulli case, and put it there in case similar sorts of arguments could be used more generally. But, perhaps, the comment was unnecessary. – Satana Feb 22 '18 at 13:56

The probabilities are both equal to $1$ for any non-trivial random walk $S_n = \sum_{i=1}^n X_i$ satisfying $\mathbb{E}(X_i)=0$; this is a direct consequence of a statement by Chung-Fuchs:
Let $(X_i)_{i \in \mathbb{N}}$ be a sequence of non-trivial iid $\mathbb{R}$-valued random variables, and denote by $S_n := \sum_{i=1}^n X_i$ the associated random walk. Then the following statements hold true.
• $\mathbb{E}X_1>0 \iff \lim_{n \to \infty} S_n = \infty$ almost surely
• $\mathbb{E}X_1<0 \iff \lim_{n \to \infty} S_n = -\infty$ almost surely
• $\mathbb{E}X_1 = 0 \iff - \infty = \liminf_{n \to \infty} S_n < \limsup_{n \to \infty} S_n = \infty$ almost surely.
The proof is relatively easy if the steps $X_i$ are bounded (i.e. $\mathbb{P}(X_1 \leq K)=1$ for some $K>0$); for general random variables $X_i$ the proof is more involved.