Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

Looking into the distribution of a Fabius random variable: $$ X := \sum_{k=1}^\infty 2^{-k} u_k $$ where $u_k$ are i.i.d. uniform variables on a unit interval, I encountered the following expression for its probability density: $$ f_X(x) = \frac{1}{\pi} \int_0^\infty \left( \prod_{k=1}^\infty \operatorname{\rm sinc}\left( \frac{t}{2^{k+1}} \right) \right) \cos \left( t \left( x- \frac{1}{2} \right) \right) \mathrm{d} t $$ It seems, numerically, that $f\left(\frac{1}{2} \right) = 2$, but my several attempts to prove this were not successful.

Any ideas how to approach this are much appreciated.

share|improve this question

2 Answers 2

up vote 9 down vote accepted

From Theorem 1 (equation (19) on page 5) of Surprising Sinc Sums and Integrals, we have $$\frac{1}{\pi} \int_0^\infty \left( \prod_{k=1}^N \operatorname{\rm sinc}\left( \frac{t}{2^{k+1}} \right) \right) \mathrm{d} t=2$$ for all $N<\infty$. I suppose you can justify letting $N\to \infty$ to get your result.


One of the surprises in that paper concerns a similar integral $$ \int_0^\infty \left( \prod_{k=0}^N \operatorname{\rm sinc}\left( \frac{t}{2{k+1}} \right) \right) \mathrm{d} t.$$ This turns out to be equal to $\pi/2$ when $0\leq N\leq 6$, but is slightly less than $\pi/2$ when $N=7$.

share|improve this answer

Since $u_k$ and $1-u_k$ are equal in distribution, it follows that $$ X \stackrel{d}{=} \sum_{k=1}^\infty 2^{-k} \left(1-u_k\right) = \sum_{k=1}^\infty 2^{-k} - X = 1 - X $$ Since $u_k \geq 0$, it follows that $\mathbb{P}\left(0 \leqslant X \leqslant 1\right) = 1$.

Therefore $f_X(x) = 0$ for $x < 0$ or $x>1$. It also shows that $f_X(x) = f_X(1-x)$ for $0<x<1$.

Notice that $$ \begin{eqnarray} f_X^\prime(x) &=& \frac{1}{\pi} \int_0^\infty \left(- t \sin\left( t \left(x-\frac{1}{2} \right) \right) \right) \operatorname{\rm sinc}\left( \frac{t}{4} \right) \prod_{k=2}^\infty \operatorname{\rm sinc}\left( \frac{t}{2^{k+1}} \right) \mathrm{d} t \end{eqnarray} $$ Now using $t \cdot \operatorname{\rm sinc}\left( \frac{t}{4} \right) = 4 \sin\left( \frac{t}{4} \right)$, and $$ \sin\left( t \left(x-\frac{1}{2} \right) \right) \sin\left(\frac{t}{4} \right) = \frac{1}{2} \left[ \cos\left( \frac{t}{2} \left( (2 x-1) -\frac{1}{2} \right) \right) - \cos\left( \frac{t}{2} \left( 2 x -\frac{1}{2} \right) \right) \right] $$ which gives $$ \begin{eqnarray} f_X^\prime(x) &=& 4 f_X(2 x) - 4 f_X(2x-1) \end{eqnarray} $$

Now let $0 < z \leqslant \frac{1}{2}$. Then $f_X^\prime(z) = 4 f_X(2z)$ and thus $$ f_X(z) - f_X(0) = \int_0^{z} 4 f_X(2x) \mathrm{d} x = 2 \left(F_X(2z) - F_X(0)\right) $$

Clearly $F(0) = 0$, $F(1) = 1$ and $F_X\left(\frac{1}{2} \right) = \frac{1}{2}$, since the distribution is symmetric about $x=\frac{1}{2}$, thus $$ f_X\left(\frac{1}{2}\right) = f_X\left(0\right) + 2 \qquad f_X\left(\frac{1}{4}\right) = f_X\left(0\right) + 1 $$

If I now assume that $f_X(0) = 0$, the result follows.

Added A missing proof of $f_X(0)=0$, as well as alternative proof of $f_X\left(\frac{1}{2}\right) = 2$ results from writing: $$ X \stackrel{d}{=} \frac{u_1}{2} + \frac{1}{2} \sum_{k=2}^\infty 2^{-k+1} u_k \stackrel{d}{=} \frac{u_1+X}{2} $$ This equality in distribution implies the following integral equation for $f_X$: $$ f_X(x) = 2 \int_0^1 f_X(2x -u) \mathrm{d} u $$ Substituting $x=0$ we get $f_X(0) = 2 \int_0^1 f_X(-u) \mathrm{d} u =0$, since $f_X(x)=0$ for $x<0$. Incidentally, using $x=\frac{1}{2}$ gives the desired identity as well: $$ f_X\left(\frac{1}{2}\right) = 2 \int_0^1 f_X(1-u) \mathrm{d} u = 2 \int_0^1 f_X(u) \mathrm{d} u = 2 $$

share|improve this answer
2  
The probabilist in me likes this kind of argument! –  Byron Schmuland Jan 6 '12 at 21:01

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

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