I am totally at a loss with this question and don't even know where to begin.

Let $g:[0, 1]\rightarrow \mathbb{R}$ be a measurable and Lebesgue-integrable function. $U_1, U_2, \dots$ be a series of independent and on $[0, 1]$ equally distributed random variables.
Let $X_i := g(U_i) \qquad \forall \, i=1, 2, \dots$

Is it possible to approximate $\int_0^1g(x)dx$ with $\sum_{i=1}^{n}\frac{X_i}{n}$.
And if that is possible, in which way and what can I say about the approximation error.

Ok. I have an idea. Using Kolmogorov's strong law of large numbers which says

For equally distributed random variables $X_1, X_2, \dots \in L^1(\Omega, \mathcal{F}, \mathbb{P})$ is
$\frac{1}{n}\sum_{i=1}^{n}X_i \rightarrow \mathbb{E}X_1$ for $n\rightarrow \infty \, \mathbb{P}$ a.s.

I know that
$E[g(U_1)] = \int_{\mathbb{R}}g(x)f(x)d(x) = \int_{\mathbb{R}}g(x)1_{[0, 1]}d(x) = \int_{0}^1 g(x)d(x)$
and obviously
$\sum_{i=1}^{n}\frac{X_i}{n} = \frac{1}{n}\sum_{i=1}^{n}g(U_i)$

Now I only need to show that not only $U_i$ are random Variables but also $g(U_i)$. And how do I do that? Also, what about the approximation error?


1 Answer 1


Indeed, every function $g(U_i):\Omega\to\mathbb R$ is a random variable as the composition of the measurable functions $U_i:\Omega\to[0,1]$ and $g:[0,1]\to\mathbb R$. As you said, by the strong law of large numbers,
$$ V_n=\frac1n\sum\limits_{i=1}^ng(U_i) $$ is an approximation of $\mu=E(g(U_1))$ when $n\to\infty$ in the sense that $V_n\to \mu$ almost surely when $n\to\infty$, and $$\mu=\displaystyle\int_0^1g(x)\mathrm dx.$$ The approximation error between $V_n$ and $\mu$ can be quantified in at least two (related) ways. First, for every $n$, $$ E((V_n-\mu)^2)=\frac1n\sigma^2,\qquad\sigma^2=E((g(U_1)-\mu)^2). $$ Second, by the central limit theorem, when $n\to\infty$, $\sqrt{n}(V_n-\mu)$ converges in distribution to a centered normal distribution with variance $\sigma^2$. Both ways suggest a rate of convergence of order $1/\sqrt{n}$. Finally, $$ \sigma^2=\int_0^1g^2(x)\mathrm dx-\mu^2=\int_0^1(g(x)-\mu)^2\mathrm dx. $$ These remarks are the basis of the approximate computation of integrals by Monte Carlo methods.


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .