Large deviations for independent random walks (Q583708): Difference between revisions

In a system of independent random walks on \({\mathbb{Z}}\), let \(\xi_ n(x)\) denote the number of particles at x at time n, and let \(L_ n(x)=\xi_ 0(x)+...+\xi_ n(x)\) be the total occupation time of x by time n. The following large deviation asymptotics are derived for \(L_ n=L_ n(0)- L_ n(1):\) a) If \(\xi_ 0\) is a nonrandom initial configruation with \[ (*)\quad (2n)^{-1}\sum^{n}_{m=-n}\xi_ 0(m)\to 1\quad as\quad n\to \infty, \] then, for \(3/4<\alpha <2\) and \(a>0\), as \(n\to \infty\), \[ (1)\quad n^{(2-6\alpha)/5} \log P(L_ n>a n^{\alpha})\to - (5/4)(2a/3)^{6/5}. \] Moreover, under (*) and \(\alpha =3/4\), \[ (2)\quad n^{-1/2} \log P(L_ n>a n^{3/4})\to -I(a), \] where the large deviation functional I is determined by the asymptotic behavior of any single particle. b) If (*) above is replaced by (**) \(\xi_ 0(x)\), \(x\in {\mathbb{Z}}\), are i.i.d. Poisson with mean one, then assertion (2) retains, but with I(a) becoming strictly larger. Hence it follows that the Poisson process will more easily achieve large weighted occupation times. The difference in large deviations behavior is even more severe, when \(\alpha >3/4\). A lower bound \[ (3)\quad P(L_ n>n^{\alpha})\geq \exp (-2\gamma n^{\gamma} \log n), \] where \(\gamma =\alpha -1/4\), demonstrates this fact.