1
$\begingroup$

For a class on Monte-Carlo simulation methods I want to do a comparison of Monte-Carlo integration and numerical integration on sparse grids. For this purpose I need a function $f$ defined over the $n$-dimensional hypercube $[-1,1]^n$ of which I know the value of the integral $\int_{[-1,1]^n} f$ in order to compare the calculation errors of the different methods. I thought about integrating the unit $n$-sphere as this is defined over the hypercube and the exact volume of the unit $n$-sphere is given by

$$V_n=\frac{\pi^{\frac{n}{2}}}{\Gamma (\frac{n}{2} +1)}$$

Now integrating the whole $n$-sphere will result in the integral being 0 as it is symmetric. That's why I thought about using only the upper unit $n$-sphere. Then the value of the integral would be given by $V_n/2$. However, I did not manage to come up with a function describing this upper half of the $n$-sphere. In the 2-dimensional case the upper sphere $U_2$ would be given by $U_2=\{ x_2= \sqrt{1-x_1^2},x_2\geq 0\}$ or in the more general case of the unit $n$-ball $U_n = \{ \sqrt{1- \sum_{i=1}^{n-1} x_i^2 }, x_n \geq 0 \}$.

Is there a way to express $U_n$ as a function that I could plug into my implemented integration algorithms to integrate it over the $n$-dimensional hypercube?

$\endgroup$
  • 1
    $\begingroup$ The assertion "[the unit $n$-sphere] is defined over the hypercube..." is not true if $n \geq 2$. (The "shadow" of an ordinary hemisphere is a disk, not a square.) Separately, it's not clear what you mean by "integrating the whole $n$-sphere will result in the integral being $0$ as it is symmetric". $\endgroup$ – Andrew D. Hwang Jan 19 '17 at 21:22
  • $\begingroup$ So on which set is the unit $n$-sphere defined? $\endgroup$ – YukiJ Jan 19 '17 at 21:26
  • 1
    $\begingroup$ I think you want: The unit sphere is the union of two graphs $x_{n+1} = \pm\sqrt{1 - \|x\|^{2}}$ over the unit ball, the set of $x = (x_{1}, \dots, x_{n})$ in $\mathbf{R}^{n}$ where $\|x\|^{2} \leq 1$. $\endgroup$ – Andrew D. Hwang Jan 19 '17 at 21:39
  • $\begingroup$ @YukiJ: Offering a bounty but not awarding it is like making a promise but not keeping it: a very ugly kind of behaviour. $\endgroup$ – Alex M. Jan 28 '17 at 9:18
  • $\begingroup$ I am sorry @AlexM. But I Received an e-mail that the bounty will be automatically given to the best answer? Why has this not happened?! $\endgroup$ – YukiJ Jan 28 '17 at 9:20
2
$\begingroup$

Why don't you try polynomials - the simplest of the non-trivial examples? Choose a degree $d \in \Bbb N$ and a set of coefficients $\{a_{i_1, \dots, i_n} \mid i_1 + \dots + i_n \le d\} \subset \Bbb C$ and consider the associated polynomial $p(x_1, \dots, x_n) = \sum _{i_1 + \dots + i_n \le d} a_{i_1, \dots, i_n} x_1 ^{i_1} \dots x_n ^{i_n}$. Then

$$\int \limits _{[-1,1]^n} p(x_1, \dots, x_n) \ \Bbb d x_1 \dots \Bbb d x_n = \int \limits _{-1} ^1 \dots \int \limits _{-1} ^1 \sum _{i_1 + \dots + i_n \le d} a_{i_1, \dots, i_n} x_1 ^{i_1} \dots x_n ^{i_n} \ \Bbb d x_1 \dots \Bbb d x_n = \\ \sum _{i_1 + \dots + i_n \le d} a_{i_1, \dots, i_n} \frac {1 + (-1)^{i_1}} {i_1 + 1} \dots \frac {1 + (-1)^{i_n}} {i_n + 1} ,$$

a formula simple enough to let you compare its value with the one that you obtain numerically.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.