Multivariate Laplace's approximation with estimated error and application to limit theorems

DOI10.1016/J.JAT.2019.105305MaRDI QIDQ2334385zbMATH OpenOpenAlexFDO

Publication date 7 November 2019

Published in Journal of Approximation Theory (Search for Journal in Brave)

Full work available at URL https://arxiv.org/abs/1502.03266

asymptotic approximation of integrals central limit theorem error estimates law of large numbers limit theorems multivariate Laplace's method

Mathematics Subject Classification ID

Asymptotic approximations, asymptotic expansions (steepest descent, etc.) (41A60) Central limit and other weak theorems (60F05)

Abstract: In this paper we obtain an approximation for the multivariate Laplace's integral with a large parameter and estimate error term for two cases, when the maximum of the exponent is in the interior of the domain and on the boundary. We are specifically interested in the situation when the function in the exponent depends on the large parameter. As an application we prove weak law of large numbers and central limit theorem. The second result gives different limiting distributions for two cases mentioned above. When the maximum of the exponent is in the interior of the domain it is Normal distribution and if it is on the boundary, it is Exponential in one direction of integration and Normal in other directions.

Review text

The authors study the asymptotics of the multi-dimensional Laplace integrals \[ I_1(N):=\int_\Omega g(x)e^{Nf(x,N)}dx \] and \[ I_2(N):=\int_{\Omega\cap\{x;x_1\ge 0\}} g(x)e^{Nf(x,N)}dx \] where \(\Omega\subset R^m\) is an open set, the functions \(f\) and \(g\) are sufficiently regular, and \(N\in Z^+\) is a large parameter. In the first integral, \(f(\cdot,N)\) has a unique maximum \(x^*(N)\) in the interior of \(\Omega\), whereas in the second one, \(f(\cdot,N)\) has a unique non-critical maximum on the boundary \(\{x;x_1= 0\}\). The asymptotic theory of these integrals is well-known, and we can find in the literature different asymptotic expansions for them. The form of these expansions depends on the location and on the order of the critical points of the exponent. The authors go a step further and derive the following error bounds for the first order Laplace approximation: \noindent (1) For the second integral with \(\Omega=[0,\infty)\) (univariate integral), the authors derive the following formula \[ \int_\Omega g(x)e^{Nf(x,N)}dx=E^{\frac{Nf(0,N)}{N}}\left(\frac{g(0)}{\vert f'(0,N)\vert}+\frac{w(N)}{N}\right), \] with \(w(N)=\mathcal{O}(1)\) as \(N\to\infty\). Moreover, an explicit bound for \(w(N)\) is given in terms of bounds for the first partial derivatives of the phase function \(f\). \noindent (2) For the first integral, when the maximum \(x^*\) is an interior point, the authors derive the following formula \[ \int_\Omega g(x)e^{Nf(x,N)}dx=E^{Nf(x^*,N)}\left(\frac{2\pi}{N}\right)^{m/2}\left(\frac{g(x^*)}{\sqrt{\vert D^2 f(x^*,N)\vert}}+\frac{w(N)}{\sqrt{N}}\right), \] with \(w(N)=\mathcal{O}(1)\) as \(N\to\infty\) an a similar explicit bound for \(w(N)\). \noindent (3) For the second integral, when the maximum \(x^*\) is on the boundary, the authors derive the following formula \[ \int_{\Omega\cap\{x;x_1\ge 0\}} g(x)e^{Nf(x,N)}dx=E^{\frac{Nf(x^*,N)}{N}}\left(\frac{2\pi}{N}\right)^{(m-1)/2}\left(\frac{g(x^*)}{\sqrt{\vert D_y^2 f(x^*,N)\vert}}+\frac{w(N)}{\sqrt{N}}\right), \] with \(w(N)=\mathcal{O}(1)\) as \(N\to\infty\) an a similar explicit bound for \(w(N)\). The symbol \(D_y^2f\) represents the Hessian matrix of \(f\) as a function of all its variables except \(x_1\). As an application, the authors prove the weak law of large numbers and the central limit theorem, with different limit distributions depending on the location of the maximum \(x^*\) of the phase function: when the maximum is in the interior of the integration domain, the limit distribution is a Normal distribution. When the maximum is on the boundary, the limit distribution is exponential in one direction and Normal in the other directions.

Recommendations

Cites work

Cited in

(14)

This page was built for publication: Multivariate Laplace's approximation with estimated error and application to limit theorems

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2334385)