How to solve an optimization problem over $n$, where $n$ is the sample size?

20 Views Asked by At

Consider $n$ i.i.d. (continues) random variables $X_1,...,X_n$ sampled a distribution with density function $f(x)$.

Let $g(x_1,...,x_n): \mathbb{R}^n\mapsto \mathbb{R}$ be a function defined over $x_1,...,x_n$. Define $G(n):=\int_{-\infty}^{\infty} g(x_1,...,x_n) \Pi_i^n f(x_i) dx_i$.

Consider the following optimization problem:

$$n^*:=\min \{n: G(n)>\xi\},$$

where $\xi$ is a real number between $(0,1)$.

This can be seen as to find the minimum sample size $n^*$ such that the statistic $T:=g(X_1,...,X_n)$ meets a certain probability constraint.

My question is how to solve the above optimization problem, when the function $g(\cdot)$ and the integral $G(n)$ are hard to be represented analytically?

As an example, consider $g(x_1,...,x_n):=\Phi_n(x_1,...,x_n)$ being the CDF of the $n$-variate Gaussian distribution (with diagonal covariance matrix) and $f(x)\sim N(0,1)$.

Any suggestion or reference is welcomed.