Generalized van Trees inequality: Local minimax bounds for non-smooth functionals and irregular statistical models

In a decision-theoretic framework, minimax lower bound provides the worst-case performance of estimators relative to a given class of statistical models. For parametric and semiparametric models, the H\'ajek--Le Cam local asymptotic minimax (LAM) theorem provides the optimal and sharp asymptotic lower bound. Despite its relative generality, this result comes with limitations as it only applies to the estimation of differentiable functionals under regular statistical models. On the other hand, non-asymptotic minimax lower bounds, such as those based on the reduction to hypothesis testing, do not often yield sharp asymptotic constants. Inspired by the recent improvement of the van Trees inequality and related methods in the literature, we provide new non-asymptotic minimax lower bounds under minimal regularity assumptions, which imply sharp asymptotic constants. The proposed lower bounds do not require the differentiability of functionals or regularity of statistical models, extending the efficiency theory to broader situations where standard approaches fail. Additionally, new lower bounds provide non-asymptotic constants, which can shed light on more refined fundamental limits of estimation in finite samples. We demonstrate that new lower bounds recover many classical results, including the LAM theorem and semiparametric efficiency bounds. We also illustrate the use of the new lower bound by deriving the local minimax lower bound for estimating the density at a point and directionally differentiable parameters.