Bad rates of convergence for the central limit theorem in Hilbert space (Q5896401)

The authors show that one can smoothly renorm the Hilbert space \(\ell^ 2\) such that the rate of convergence in the central limit theorem becomes very poor. Their result reads: Let \(\epsilon>0\) and a real sequence \(\xi_ n\to 0\), \(n\to \infty\), be given. There is a norm \(N(\cdot)\) on \(\ell^ 2\) and a sequence \((X_ i)_{i\in {\mathbb{N}}}\) of i.i.d. \(\ell^ 2\)-valued r.v. such that the following holds: (a) \((1-\epsilon)\| x\| \leq N(x)\leq \| x\|\); \(x\in \ell^ 2;\) (b) \(N(\cdot)\) is infinitely many times differentiable, and each of its differentials is bounded on the unit sphere; (c) For infinitely many values of n, \(\sup_{t\geq 0}| P\{N(n^{- {1\over2}}\sum^{n}_{i=1}X_ i)\leq t\}-\gamma \{x;N(x)<t\}| \geq \xi_ n\), where \(\gamma\) is the Gaussian measure with the same covariance as \(X_ i.\) This result is based on the fact that for \(N(\cdot)\) fails the following condition that the usual norm \(\| \cdot \|\) fulfills, namely, \(\gamma \{s\leq \| x\| \leq t\}\leq C(t-s)\) for a constant C and all \(0\leq s<t\).