On the exponential rate of convergence of the least squares estimator in the nonlinear regression model with Gaussian errors (Q796937)

Let \({\hat \theta}{}_ n\) be a least squares estimator in a nonlinear regression model \(Y_ n=g_ n(\theta)+\epsilon_ n\), \(n\geq 1\), \(\theta \in \Theta \subset R\), \(\epsilon_ n\) i.i.d. (0,1)-normal r.v. Let K be compact in \(\Theta\). Suppose there exist constants \(k_ i>0\), \(i=1,2\), such that \[ nk_ 1(\theta_ 1-\theta_ 2)^ 2\leq \psi_ n(\theta_ 1,\theta_ 2)\leq nk_ 2(\theta_ 1-\theta_ 2)^ 2 \] for all \(\theta_ 1,\theta_ 2\in \Theta\), where \(\psi_ n(\theta_ 1,\theta_ 2)=\sum^{n}_{i=1}[g_ i(\theta_ 1)-g_ i(\theta_ 2)]^ 2.\) Then there exist constants B, b, both positive, depending on K such that \(\sup_{\theta \in K}P_{\theta}\{n^{{1\over2}}| {\hat \theta}_ n-\theta |>\rho \}\leq Be^{b\rho^ 2}\) for any \(\rho>0\) and \(n\geq 1\).