Large deviations for sums of random variables related to a chain (Q1088281): Difference between revisions

A homogeneous Markov chain \(\{\xi_ i\), \(i=0,1,...\}\) with the state space \((X,{\mathcal F})\), initial distribution \(Q(\cdot)\) and transition function \(P(\cdot,\cdot)\) is considered. For a real \({\mathcal F}\)-measurable bounded function \(f: X\to {\mathbb{R}}\) denote \(S_ n=f(\xi_ 1)+...+f(\xi_ n)\). It is stated that if \(Q(\cdot)\) and \(P(x,\cdot)\) are absolutely continuous w.r.t. some probability measure \(\nu(\cdot)\), \(q(x)\) and \(p(x,y)\) are their densities which are a.s. positive, \(q\in L^ s(X,\nu)\), \(1<s<\infty\), and \[ \int\left(\int | p(x,y)|^ s \nu(dx)\right)^{s'/s}\nu(dy)< \infty, \] where \(1/s+1/s'=1\), then the finite limits \[ \lim_{n\to \infty} n^{-1}ES_ n=M\quad and\quad \lim_{n\to \infty} n^{-1}E(S_ n-ES_ n)^ 2=d^ 2 \] exist. Let \(r(h)\) be the spectral radius of the operator T(h) (h\(\in {\mathbb{R}})\) defined on \(L^ s(X,\nu)\) by \[ (T(h)g)=\int g(y)\exp (hf(y))p(x,y)\nu (dy) \] and \(\mu_ 1=\sup_{h>0}\{d r(h)/d h\}\). If \(d^ 2>0\), then \[ \lim_{n\to \infty} n^{-1}\ln P(S_ n>na) = \begin{cases} \inf_{h>0} (- ah+r(h)), & \text{for \(M<a<\mu_ 1,\)} \\ -\infty,& \text{for \(a>\mu_ 1\).} \end{cases} \] A more general result is proved without absolute continuity of the measures \(Q(\cdot)\) and \(P(x,\cdot)\) and boundedness of f.