An invariance principle under the total variation distance (Q2342391): Difference between revisions

Let \({X_1},{X_2},\dots{}\) be a sequence of independent identically distributed random variables, \({\text{E}}{X_1} = 0, \) \({\text{D}}{X_1} = 1\), and let \({S_n} = ({X_1} + \dots{} + {X_n})/\sqrt n \). \textit{Yu. Prokhorov} [``On a local limit theorem for densities'' (Russian), Dokl. Akad. Nauk SSSR, n. Ser. 83, 797--800 (1952)] proved that \({S_n}\) converges in total variation to the standard normal distribution if and only if \({S_{{n_0}}}\) has an absolutely continuous component for some integer \({n_0}\). The authors extend this result to a sequence of homogeneous polynomials in \({X_i}\). These polynomials \[ {Q_n}(x) = \sum\limits_{{i_1},\dots{},{i_d} = 1}^{{N_n}} {{a_n}({i_1},\dots{},{i_d}){x_{{i_1}}}\dots{}{x_{{i_d}}}} , \;\;({x_1},\dots{},{x_{{N_n}}}) \in {\mathbb R^{{N_n}}} \] satisfy the following conditions:{\parindent=0.6cm\begin{itemize}\item[(i)] \({N_n} \to \infty \) as \(n \to \infty \); \item[(ii)] the coefficients \({a_n}({i_1},\dots{},{i_d})\) vanish on hyperdiagonals, that is, \({a_n}({i_1},\dots{},{i_d}) = 0\) if \({i_j} = {i_k}\) for some \(j \neq k\); \item[(iii)] the coefficients \({a_n}({i_1},\dots{},{i_d})\) are symmetric in the indices, that is, invariant under permutations; \item[(iv)] \(d!\sum\limits_{{i_1},\dots{},{i_d} = 1}^{{N_n}} {{a_n}{{({i_1},\dots{},{i_d})}^2}} = 1\) for all \(n \geqslant 1\); \item[(v)] \(\mathop {\lim }\limits_{n \to \infty } \mathop {\max }\limits_{1 \leqslant {i_1} \leqslant n} \sum\limits_{{i_2},\dots{},{i_d} = 1}^{{N_n}} {{a_n}{{({i_1},\dots{},{i_d})}^2}} = 0\). \end{itemize}}