Fix $N$ and suppose $\{X_n\}_{k=1}^{N}$ are i.i.d steps that are $\pm 1$ with equal probability. Then $S_n = \sum_{k\leq n} X_k $ is a simple random walk, and (with the right scaling) we know that the path $S_n$ converges to a Brownian motion as $N \to \infty$(This is an application Donsker's Theorem).

I want to consider a related, but slightly different set up. Fix $N$ and suppose $\{X_n\}_{k=1}^{N}$ are i.i.d steps that are $+1$ with probability $\frac{1}{2} + \frac{\lambda / 2}{\sqrt{N}} $ and $-1$ with probability $\frac{1}{2} - \frac{\lambda / 2}{\sqrt{N}}$. Then and $S_n = \sum_{k\leq n} X_k $ is a biased random walk. I want to show that (in the same scaling as above) $S_n$ converges to a Brownian motion with drift $\lambda$ as $N \to \infty$.

My attempt:
I think we can do this by some manipulation and then applying Donsker's Theorem. Let $Y_k = \sqrt{1 - \frac{\lambda^2}{n}}\left( X_k - \frac{\lambda}{\sqrt{n}} \right)$ so that the $Y_k$'s are mean $0$ and variance $1$. Then applying Donsker's Theorem to the $Y_k$'s gets us very close to the result I want. However, there is a pesky factor of $\sqrt{1-\frac{\lambda^2}{n}}$ that is in front.

I think the following is true and can fix the problem: If $f_n \Rightarrow f$ are distributions in $C[0,1]$ and if $c_n \to 1$ are constants, then $c_n f_n \Rightarrow f$ in $C[0,1]$ too. My "proof" of this involves tightness and Prohorov's Theorem, so I'm hoping for an easier alternative.

  • 1
    $\begingroup$ Do you know proofs of Donsker's theorem without Prokhorov theorem? $\endgroup$ – Davide Giraudo Jan 4 '13 at 22:58

Yes, $f_n \to f$ in distribution and $c_n \to 1$ implies $c_n \cdot f_n \to f$ in distribution. A short proof of this assertion is found here.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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