Expansions for the multivariate chi-square distribution

Three classes of expansions for the distribution function of the χ_k²(d, R)-distribution are given, where k denotes the dimension, d the degree of freedom, and R the “accompanying correlation matrix.” The first class generalizes the orthogonal series with generalized Laguerre polynomials, originally given by Krishnamoorthy and Parthasarathy [12]. The second class contains always absolutely convergent representations of the distribution function by univariate chi-square distributions and the third class provides also the probabilities for any unbounded rectangular regions. In particular, simple formulas are given for the three-variate case including singular correlation matrices R, which simplify the computation of third order Bonferroni inequalities, e.g., for the tail probabilities of max{χ_i²|1 ≤ i ≤ k} (k > 3).     

Asymptotic expansions in the singular value decomposition for cross covariance and correlation under nonnormality

Asymptotic cumulants of the distributions of the sample singular vectors and values of cross covariance and correlation matrices are obtained under nonnormality. The asymptotic cumulants are used to have the approximations of the distributions of the estimators by the Edgeworth expansions up to order O(1/n) and Hall’s method with variable transformation. The cases of Studentized estimators are also considered. As an application of the method, the distributions of the parameter estimators in the model of inter-battery factor analysis are expanded. Interpreting the singular vectors and values in the context of the factor model with distributional conditions, the asymptotic robustness of some lower-order normal-theory cumulants of the distributions of the sample singular vectors and values under nonnormality is shown.     

Markov-type Inequalities for Surface Gradients of Multivariate Polynomials

Let K ^d be a compact set with a smooth boundary and consider a polynomial p of total degree n such that p_C(K)1. Then we show that D_Tp(x)=o(n²) for any x Bd K and T a tangential direction at x. Moreover, the o(n²) term is given in terms of the modulus of smoothness of Bd K.     

Common Principal Components for Dependent Random Vectors

Let the kp-variate random vector X be partitioned into k subvectors X_i of dimension p each, and let the covariance matrix Ψ of X be partitioned analogously into submatrices Ψ_ij. The common principal component (CPC) model for dependent random vectors assumes the existence of an orthogonal p by p matrix β such that β^tΨ_ijβ is diagonal for all (i, j). After a formal definition of the model, normal theory maximum likelihood estimators are obtained. The asymptotic theory for the estimated orthogonal matrix is derived by a new technique of choosing proper subsets of functionally independent parameters.     

Covariance kernel and the central limit theorem in the total variation distance

We modify and generalize the idea of covariance kernels for Borel probability measures on R^d, and study the relation between the central limit theorem in the total variation distance and the convergence of covariance kernels.     

Derivatives of regularly varying functions in Rd and domains of attraction of stable distributions

In R² the integral of a regularly varying (RV) function f is regularly varying only if f is monotone. Generalization to R² of the one-dimensional result on regular variation of the derivative of an RV-function however is straightforward. Applications are given to limit theory for partial sums of i.i.d. positive random vectors in R²₊.     

Likelihood-Based Local Polynomial Fitting for Single-Index Models

The parametric generalized linear model assumes that the conditional distribution of a response Y given a d-dimensional covariate X belongs to an exponential family and that a known transformation of the regression function is linear in X. In this paper we relax the latter assumption by considering a nonparametric function of the linear combination β^TX, say η₀(β^TX). To estimate the coefficient vector β and the nonparametric component η₀ we consider local polynomial fits based on kernel weighted conditional likelihoods. We then obtain an estimator of the regression function by simply replacing β and η₀ in η₀(β^TX) by these estimators. We derive the asymptotic distributions of these estimators and give the results of some numerical experiments.     

Homogenization of the parabolic cauchy problem in the Sobolev class H<Superscript>1</Superscript>(R<Superscript>d</Superscript>)

Homogenization in the small period limit for the solution ue of the Cauchy problem for a parabolic equation in R^d is studied. The coefficients are assumed to be periodic in R^d with respect to the lattice ɛG. As ɛ → 0, the solution u ɛ converges in L2(R^d) to the solution u0 of the effective problem with constant coefficients. The solution u ɛis approximated in the norm of the Sobolev space H ¹(R^d) with error O( ɛ); this approximation is uniform with respect to the L2-norm of the initial data and contains a corrector term of order ɛ. The dependence of the constant in the error estimate on time t is given. Also, an approximation in H ¹(R^d) for the solution of the Cauchy problem for a nonhomogeneous parabolic equation is obtained.     

Groups of Units of Finite Commutative Group Rings*

Let R be a finite commutative ring with identity and ? _{p ^d} be the cyclic group of prime power order. Define R? _{p ^d} to mean the group ring of ? _{p ^d} over R. We determine the structure of the group of units of R? _{p ^d} in the case when R is generated by an element whose order is not divisible by p.     

Skew *-derivations with power-central values

Let R be a prime ring with involution * and d be a nonzero derivation on R such that d(x ^*) = -d(x)^* for all x ∈ R. Suppose that n ≥ 1 is a fixed integer. Then (I) if d(s) ⁿ = 0 for all s = s ^*, then R is either a commutative domain or an order in a 4-dimensional central simple algebra; (II) if d(s) ⁿ ∈Z, the center of R for all s = s ^*, then R is either a commutative domain or an order in a simple algebra of dimension 4 or 16 over its center.     

Weighted Approximation of Functions on the Unit Sphere

The direct and inverse theorems are established for the best approximation in the weighted L^p space on the unit sphere of R^d+1, in which the weight functions are invariant under finite reflection groups. The theorems are stated using a modulus of smoothness of higher order, which is proved to be equivalent to a K-functional defined using the power of the spherical h-Laplacian. Furthermore, similar results are also established for weighted approximation on the unit ball and on the simplex of R^d.     

Tame roots of wild quivers

We study the tame behaviour of the representations of wild quivers Q via tame roots. A positive root d of Q is called a tame root if d is sincere and for any positive sub-root d^′ of d we have q(d^′)0, where q(d^′) is the Tits form of Q. We prove that a sincere root is a tame root if and only if for any decomposition of d into a sum of positive sub-roots d=d¹++d^s, there is at most one dⁱ with q(dⁱ)=0 and q(d^j)=1. This is the essential property of a tame root and it is an alternative way to define tame roots. Then we give the canonical decomposition of a tame root. At the end we prove our main result that for any wild graph, there are only finitely many tame roots.     

Order Determination for Multivariate Autoregressive Processes Using Resampling Methods

LetX₁, …, X_nbe observations from a multivariate AR(p) model with unknown orderp. A resampling procedure is proposed for estimating the orderp. The classical criteria, such as AIC and BIC, estimate the orderpas the minimizer of the function[formula]wherenis the sample size,kis the order of the fitted model, Σ²_kis an estimate of the white noise covariance matrix, andC_nis a sequence of specified constants (for AIC,C_n=2m²/n, for Hannan and Quinn's modification of BIC,C_n=2m²(ln ln n)/n, wheremis the dimension of the data vector). A resampling scheme is proposed to estimate an improved penalty factorC_n. Conditional on the data, this procedure produces a consistent estimate ofp. Simulation results support the effectiveness of this procedure when compared with some of the traditional order selection criteria. Comments are also made on the use of Yule–Walker as opposed to conditional least squares estimations for order selection.     

Upper bounds for configurations and polytopes inR d

We give a new upper bound onn ^d(d+1)n on the number of realizable order types of simple configurations ofn points inR ^d , and ofn2d ² n on the number of realizable combinatorial types of simple configurations. It follows as a corollary of the first result that there are no more thann ^d(d+1)n combinatorially distinct labeled simplicial polytopes inR ^d withn vertices, which improves the best previous upper bound ofn ^{cn d/2}.Supported in part by NSF Grant DMS-8501492 and PSC-CUNY Grant 665258.Supported in part by NSF Grant DMS-8501947.     

The normal approximation rate for the drift estimator of multidimensional diffusions

We consider the problem of the density and drift estimation by the observation of a trajectory of an \mathbbR^d{\mathbb{R}^{d}}-dimensional homogeneous diffusion process with a unique invariant density. We construct estimators of the kernel type based on discretely sampled observations and study their asymptotic distribution. An estimate of the rate of normal approximation is given.     

Asymptotics of Green functions and Martin boundaries for elliptic operators with periodic coefficients

We give the asymptotics at infinity of a Green function for an elliptic equation with periodic coefficients on R^d. Basic ingredients in establishing the asymptotics are an integral representation of the Green function and the saddle point method. We also completely determine the Martin compactification of R^d with respect to an elliptic equation with periodic coefficients by using the exact asymptotics at infinity of the Green function.     

Inequalities for predictive ratios and posterior variances in natural exponential families

The predictive ratio is considered as a measure of spread for the predictive distribution. It is shown that, in the exponential families, ordering according to the predictive ratio is equivalent to ordering according to the posterior covariance matrix of the parameters. This result generalizes an inequality due to Chaloner and Duncan who consider the predictive ratio for a beta-binomial distribution and compare it with a predictive ratio for the binomial distribution with a degenerate prior. The predictive ratio at x₁ and x₂ is defined to be p_g(x₁)p_g(x₂)/[p_g( )]² = h_g(x₁, x₂), where p_g(x₁) = ∫ ƒ(x₁θ) g(θ) dθ is the predictive distribution of x₁ with respect to the prior g. We prove that h_g(x₁, x₂) ≥ h_g^*(x₁, x₂) for all x₁ and x₂ if ƒ(xθ) is in the natural exponential family and Cov_gx(θ) ≥ Cov_g^*x(θ) in the Loewner sense, for all x on a straight line from x₁ to x₂. We then restrict the class of prior distributions to the conjugate class and ask whether the posterior covariance inequality obtains if g and g^* differ in that the “sample size”     

Limit Processes for Age-Dependent Branching Particle Systems

We consider systems of spatially distributed branching particles in R ^d . The particle lifelengths are of general form, hence the time propagation of the system is typically not Markov. A natural time-space-mass scaling is applied to a sequence of particle systems and we derive limit results for the corresponding sequence of measure-valued processes. The limit is identified as the projection on R ^d of a superprocess in R ₊×R ^d . The additive functional characterizing the superprocess is the scaling limit of certain point processes, which count generations along a line of descent for the branching particles.     

Approximation of measures by Markov processes and homogeneous affine iterated function systems

It is shown that under certain conditions, attractive invariant measures for iterated function systems (indeed for Markov processes on locally compact spaces) depend continuously on parameters of the system. We discuss a special class of iterated function systems, the homogeneous affine ones, for which an inverse problem is easily solved in principle by Fourier transform methods. We show that a probability measureμ onR ⁿ can be approximated by invariant measures for finite iterated function systems of this class if \(\hat \mu (t)/\hat \mu (a^T t)\) is positive definite for some nonzero contractive linear mapa:R ⁿ →R ⁿ . Moments and cumulants are also discussed.     

Average width and optimal recovery of the anisotropic classesW r H α (R d ) in the metricC k (R d )

This paper concerns the problems of the average widths and the optimal recovery of the anisotropic Hölder classesW ^r H ^α (R ^d ) of smooth functions defined on the Euclidean spaceR ^d in the metricC ^k (R ^d ). The weak asymptotic behavior is established for the corresponding quantities.