Skip to main content

Simple Random Walk

Let the state space be S=Z\mathcal{S} = \Z, a process XtX_t has probability pp moving forward and probability 1p1-p moving backward (i.e. pi,i+1=pp_{i, i+1} = p and pi,i1=1pp_{i, i-1} = 1-p). The process is called a simple random walk.

Gambler's Ruin

Gamblers's ruin is a very famous problem with a simple random walk.

Let zz be arbitrary state, then for each state xx we denote α(x)=P(Xn=z for some n0X0=x)\alpha(x) = P(X_n = z \text{ for some } n\ge 0|X_0 = x). Then we have a(x)=(1p)α(x1)+pα(x+1)a(x) = (1-p)\alpha(x-1) + p\alpha(x+1).

To solve this difference equation we have:

pa(x)+(1p)a(x)=(1p)α(x1)+pα(x+1)    p(a(x+1)a(x))=(1p)(a(x)a(x1))pa(x) + (1-p)a(x) = (1-p)\alpha(x-1) + p\alpha(x+1)\\ \implies p(a(x+1) - a(x)) = (1-p)(a(x) - a(x-1))

Then a(x+1)a(x)=1pp(a(x)a(x1))a(x+1) - a(x) = \frac{1-p}{p} (a(x) - a(x-1))

a(2)a(1)=1pp(a(1)a(0))    a(2)a(1)=1ppa(1)a(2) - a(1) = \frac{1-p}{p} (a(1) - a(0)) \implies a(2) - a(1) = \frac{1-p}{p} a(1)

and a(3)a(2)=1pp(a(2)a(1))    a(3)a(2)=(1pp)2a(1)a(3) - a(2) = \frac{1-p}{p} (a(2) - a(1)) \implies a(3) - a(2) = (\frac{1-p}{p})^2 a(1)

by induction, we have a(n+1)a(n)=(1pp)na(1)a(n+1) - a(n) = (\frac{1-p}{p})^n a(1)

And then we can have a(n)a(1)=a(n)a(n1)+a(n1)+a(2)a(1)=a(1)i=1n1(1pp)ia(n) - a(1) = a(n) - a(n-1) + a(n-1) - \cdots + a(2) - a(1) = a(1)\sum_{i=1}^{n-1}(\frac{1-p}{p})^i

That is, a(n)=i=0n1(1pp)i={na(1)p=0.5a(1)1(1pp)n1(1pp)p0.5a(n) = \sum_{i=0}^{n-1}(\frac{1-p}{p})^i=\begin{cases} n a(1) & p = 0.5 \\ a(1)\frac{1-(\frac{1-p}{p})^{n}}{1-(\frac{1-p}{p})} & p\ne 0.5 \end{cases}

By set n=zn = z we have 1=a(n)={na(1)p=0.5a(1)1(1pp)n1(1pp)p0.5    a(1)={1/np=0.51(1pp)1(1pp)np0.51 = a(n) = \begin{cases} n a(1) & p = 0.5 \\ a(1)\frac{1-(\frac{1-p}{p})^{n}}{1-(\frac{1-p}{p})} & p\ne 0.5 \end{cases}\implies a(1) = \begin{cases} 1/n & p = 0.5 \\ \frac{1-(\frac{1-p}{p})}{1-(\frac{1-p}{p})^{n}} & p\ne 0.5 \end{cases}

That is, 1in,α(i)={ia(1)=i/np=0.5a(1)1(1pp)i1(1pp)=1(1pp)1(1pp)n1(1pp)i1(1pp)=1(1pp)i1(1pp)np0.5\forall 1 \le i \le n, \alpha(i) = \begin{cases} i a(1) = i/n & p = 0.5 \\ a(1)\frac{1-(\frac{1-p}{p})^{i}}{1-(\frac{1-p}{p})} = \frac{1-(\frac{1-p}{p})}{1-(\frac{1-p}{p})^{n}}\frac{1-(\frac{1-p}{p})^{i}}{1-(\frac{1-p}{p})}=\frac{1-(\frac{1-p}{p})^{i}}{1-(\frac{1-p}{p})^{n}} & p\ne 0.5 \end{cases}

Recurrence

In case to know if such process is transient, positive recurrent or negative recurrent, we will need to use the theorem: An irreducible MC is transient iff satisfies followed:

  • 0α(x)10 \le \alpha(x) \le 1 for all xSx \in S.
  • α(z)=1inf{α(x):xS}=0\alpha(z) = 1\land \inf\{\alpha(x): x \in S\} = 0.
  • α(x)=ySpxyα(y)\alpha(x) = \sum_{y \in S} p_{xy} \alpha(y) for all xS,xzx \in S, x\ne z.

Since the 1st and 3rd conditions are automatically satisfied by the definition of α(x)\alpha(x), we only need to check the 2nd condition and it's conditional on pp.

If p<0.5p < 0.5, then all conditions are satisfied, then it's transient where limiα(i)=1(1pp)i1(1pp)n=0\lim_{i \to \infty} \alpha(i) = \frac{1-(\frac{1-p}{p})^{i}}{1-(\frac{1-p}{p})^{n}} = 0.

If p=0.5p = 0.5, then this random walk is null recurrent.