Explicit rates of approximation in the CLT for quadratic forms

central limit theorem concentration functions convergence rates ellipsoids hyperboloids lattice point problem multidimensional spaces quadratic forms theta-series

Mathematics Subject Classification ID

Central limit and other weak theorems (60F05) Asymptotic distribution theory in statistics (62E20)

Abstract: Let

X, X_{1}, X_{2}, l d o t s

be i.i.d.

{m a t h b b R}^{d}

-valued real random vectors. Assume that

m a t h b f E X = 0

,

o p e r a t o r n a m e c o v X = m a t h b b C

,

m a t h b f E V e r t X V e r t^{2} = s i g m a^{2}

and that

X

is not concentrated in a proper subspace of

m a t h b b R^{d}

. Let

G

be a mean zero Gaussian random vector with the same covariance operator as that of

X

. We study the distributions of nondegenerate quadratic forms

m a t h b b Q [S_{N}]

of the normalized sums

S_{N} = N^{- 1 / 2} (X_{1} + c d o t s + X_{N})

and show that, without any additional conditions, [Delta_Nstackrel{mathrm{def}}{=}sup_x�igl |mathbf{P}�igl{mathbb{Q}[S_N]leq x�igr}-mathbf{P}�igl{mathbb{Q}[G]leq x�igr}�igr|={mathcal{O}}�igl(N^{-1}�igr),] provided that

d g e q 5

and the fourth moment of

X

exists. Furthermore, we provide explicit bounds of order

m a t h c a l O (N^{- 1})

for

D e l t a_{N}

for the rate of approximation by short asymptotic expansions and for the concentration functions of the random variables

m a t h b b Q [S_{N} + a]

,

a i n {m a t h b b R}^{d}

. The order of the bound is optimal. It extends previous results of Bentkus and G"{o}tze [Probab. Theory Related Fields 109 (1997a) 367-416] (for

d g e 9

) to the case

d g e 5

, which is the smallest possible dimension for such a bound. Moreover, we show that, in the finite dimensional case and for isometric

m a t h b b Q

, the implied constant in

m a t h c a l O (N^{- 1})

has the form

c_{d} s i g m a^{d} (d e t m a t h b b C)^{- 1 / 2} m a t h b f E | m a t h b b C^{- 1 / 2} X |^{4}

with some

c_{d}

depending on

d

only. This answers a long standing question about optimal rates in the central limit theorem for quadratic forms starting with a seminal paper by Ess'{e}en [Acta Math. 77 (1945) 1-125].

Review text

Let \(X,X_1,X_2,\dots\) be i.i.d. \(\mathbb{R}^d\)-valued random vectors with mean zero. Here, \(\mathbb{R}^d\) is equipped with the scalar product \(\langle x,y\rangle=x_1y_1+\cdots+x_dy_d\). The authors consider the central limit theorem for the quadratic form \[ \mathbb{Q}[S_N]=\langle \mathbb{Q}S_N,S_N \rangle\,, \] where \(\mathbb{Q}:\mathbb{R}^d\mapsto\mathbb{R}^d\) is a linear symmetric bounded operator and \(S_N=N^{-1/2}(X_1+\cdots+X_N)\). Define the distribution functions \(F_a(x)=\operatorname{P}(\mathbb{Q}[S_N-a]\leq x)\) and \(H_a(x)=\operatorname{P}(\mathbb{Q}[G-a]\leq x)\), where \(G\) is a mean zero Gaussian random vector with the same covariance operator as that of \(X\). The main result of the paper is an explicit bound on \[ \sup_{x\in\mathbb{R}}\left| F_a(x)-H_a(x)-E_a(x) \right|\,, \] where \(E_a(x)\) is the Edgeworth correction. This bound applies to dimensions \(5\leq d<\infty\) and gives, in particular, that \[ \Delta_N=\sup_{x\in\mathbb{R}}\left| \operatorname{P}(\mathbb{Q}[S_N]\leq x)-\operatorname{P}(\mathbb{Q}[G]\leq x) \right|=\mathcal{O}(N^{-1})\,, \] provided that the fourth moment of \(X\) exists. These results generalise and improve several earlier results, by extending the range of \(d\) for which they may be applied and by improving the bounds obtained. The dimension condition \(d\geq5\) is the best possible for which we have \(\Delta_N=\mathcal{O}(N^{-1})\). The authors also give bounds on the concentration function \[ \sup_{a\in\mathbb{R}^d,x\in\mathbb{R}}\operatorname{P}(x\leq\mathbb{Q}[N^{1/2}S_N-a]\leq x+\lambda)\,, \] for \(\lambda\geq0\). The proofs of the main results use number theoretic arguments.

Recommendations

Cites work

Cited in

(17)

This page was built for publication: Explicit rates of approximation in the CLT for quadratic forms

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2438752)