For $U(\theta,\theta^2)$ the likelihood falls in $\theta$, so take the least feasible value $\sqrt{X_{(3)}}$

Let $X_1,X_2,X_3$ be iid r.v.s with $U(\theta,\theta^2)$; $\theta>1$. The maximum likelihood estimator (mle) of $\theta$ is: