Table of Contents
Fetching ...

Quasi-Monte Carlo for unbounded integrands with importance sampling

Du Ouyang, Xiaoqun Wang, Zhijian He

TL;DR

Superisingly, it is found that using importance sampling with t distribution as the proposal can improve the root mean squared error of RQMC from $O(n^{-1+2M+\epsilon})$ to $O ( n^{-3/2+\EPsilon)$ for any $M\in(0,1/2)$.

Abstract

We consider the problem of estimating an expectation $ \mathbb{E}\left[ h(W)\right]$ by quasi-Monte Carlo (QMC) methods, where $ h $ is an unbounded smooth function on $ \mathbb{R}^d $ and $ W$ is a standard normal distributed random variable. To study rates of convergence for QMC on unbounded integrands, we use a smoothed projection operator to project the output of $W$ to a bounded region, which differs from the strategy of avoiding the singularities along the boundary of the unit cube $ [0,1]^d $ in 10.1137/S0036144504441573. The error is then bounded by the quadrature error of the transformed integrand and the projection error. If the function $h(\boldsymbol{x})$ and its mixed partial derivatives do not grow too fast as the Euclidean norm $|\boldsymbol{x}|$ goes to infinity, we obtain an error rate of $O(n^{-1+ε})$ for QMC and randomized QMC (RQMC) with a sample size $n$ and an arbitrarily small $ε>0$. However, the rate turns out to be $O(n^{-1+2M+ε})$ if the functions grow exponentially with a rate of $O(\exp\{M|\boldsymbol{x}|^2\})$ for a constant $M\in(0,1/2)$. Superisingly, we find that using importance sampling with t distribution as the proposal can improve the root mean squared error of RQMC from $O(n^{-1+2M+ε})$ to $O( n^{-3/2+ε})$ for any $M\in(0,1/2)$.

Quasi-Monte Carlo for unbounded integrands with importance sampling

TL;DR

Superisingly, it is found that using importance sampling with t distribution as the proposal can improve the root mean squared error of RQMC from to for any .

Abstract

We consider the problem of estimating an expectation by quasi-Monte Carlo (QMC) methods, where is an unbounded smooth function on and is a standard normal distributed random variable. To study rates of convergence for QMC on unbounded integrands, we use a smoothed projection operator to project the output of to a bounded region, which differs from the strategy of avoiding the singularities along the boundary of the unit cube in 10.1137/S0036144504441573. The error is then bounded by the quadrature error of the transformed integrand and the projection error. If the function and its mixed partial derivatives do not grow too fast as the Euclidean norm goes to infinity, we obtain an error rate of for QMC and randomized QMC (RQMC) with a sample size and an arbitrarily small . However, the rate turns out to be if the functions grow exponentially with a rate of for a constant . Superisingly, we find that using importance sampling with t distribution as the proposal can improve the root mean squared error of RQMC from to for any .
Paper Structure (13 sections, 20 theorems, 111 equations, 3 figures)

This paper contains 13 sections, 20 theorems, 111 equations, 3 figures.

Key Result

Proposition 2.6

\newlabelprop:scr10 If $\left\{ \boldsymbol{a}_j\right\}$ is a $(\lambda,t,m,d)$-net in base b and $\left\{ \boldsymbol{y}_j\right\}$ is the scrambled version of $\left\{ \boldsymbol{a}_j\right\}$, then $\left\{ \boldsymbol{y}_j\right\}$ is a $(\lambda,t,m,d)$-net in base b with probability $1$.

Figures (3)

  • Figure 1: One-dimensional smoothed projection operator with $\varepsilon = 1$
  • Figure 1: RMSEs for the test function $h$ with $d = 5$ and $\nu = 3$. The RMSEs are computed based on $100$ repetitions. The slopes of the gray dashed lines are $-1$ and $-3/2$.
  • Figure 2: RMSEs for the test function $h$ with $d = 30$ and $\nu = 3$. The RMSEs are computed based on $100$ repetitions. The slopes of the gray dashed lines are $-1$ and $-3/2$.

Theorems & Definitions (53)

  • Definition 2.1
  • Definition 2.2
  • Definition 2.3
  • Definition 2.4
  • Definition 2.5
  • Proposition 2.6
  • Proposition 2.7
  • Definition 3.1
  • Definition 3.2
  • Definition 3.3
  • ...and 43 more