假设$x_t, t=0, 1,2,\cdots$为一个以0为反射壁的一维随机游走:
First, $x_0 = 0$, For $t \ge 0$, $x_{t+1} = x_t + 1$ with probablity 0.5, and $x_{t+1} = \max\{x_{t}-1, 0\}$ with probability 0.5.
Let $e_t$ denote the expectation of $x_t$:
$$e_t = E \left\{ x_t \right\}, t=1,2,3,\cdots$$显然,$e_t$单调非减。如果$e_t$收敛,令:
$$e_{\infty} = \lim_{t \to +\infty} e_t$$
那么,一定存在$N$,使得 $\forall n \ge N$,
$$e_{\infty} - 0.25 < e_n \le e_{\infty}$$
令$P(N,M)$表示从第$N$步到第$N+M-1$步,$x_t$至少回到原点一次的概率。
根据一维随机游走的常返性:
$$\lim_{M \to +\infty} P(N,M) = 1$$
显然,$P(N,M)$关于M单调非减。所以对于$N$,存在$M$,使得
$$P(N,M) > 0.5$$
易知:
$$e_{N+M} > e_N + 0.5 * P(N,M) > e_{\infty}$$
与假设矛盾。$\blacksquare$