1: \begin{abstract} 

2: We show that repulsive random variables can yield Monte Carlo

3:   methods with faster convergence rates than the typical  $N^{-1/2}$, where

4:   $N$ is the number of integrand evaluations. More precisely, we propose

5:   stochastic numerical quadratures involving determinantal point processes

6:   associated with multivariate orthogonal polynomials, and we obtain root mean square

7:   errors that decrease as $N^{-(1+1/d)/2}$, where $d$ is the dimension of the

8:   ambient space. First, we prove a central limit theorem (CLT) for the linear statistics of a class of determinantal point processes,

9:   when the reference measure is a product measure supported on a hypercube, which

10:   satisfies the Nevai-class regularity condition; a result which may be of

11:   independent interest. Next, we introduce a Monte Carlo method based on these

12:   determinantal point processes, and prove a CLT with explicit limiting variance

13:   for the quadrature error, when the reference measure satisfies a stronger

14:   regularity condition. As a corollary, by taking a specific reference measure

15:   and using a construction similar to importance sampling, we obtain a general

16:   Monte Carlo method, which applies to any measure with continuously derivable

17:   density. Loosely speaking, our method can be interpreted as a stochastic counterpart to

18:   Gaussian quadrature, which, at the price of some convergence rate, is easily

19:   generalizable to any dimension and has a more explicit error term.

20: \end{abstract}

21: