2
$\begingroup$

Let $Z$ be a random variable with cumulative distribution function: \begin{equation} F_Z(x) = \frac{1}{4 \delta^2}\left( \arccos\left( \frac{2-x^2}{2} \right) - (\pi-2\delta)\right)^2, \end{equation} and $V$ be uniformly distributed on $[\mu -\nu, \mu+\nu]$, such that: \begin{equation} f_V(x)=\frac{1}{2\nu}. \end{equation} $Z$ represents distance (of a line in the unit circle with a starting point on the circle boundary uniformly on $[-\delta, \delta]$ and end point uniformly on $[\pi-\delta, \pi+\delta]$, derived by hand and confirmed by simulation) and $V$ represents velocity. Assume $\delta< \frac{1}{2} \pi$ and $\nu<\mu$.

I am looking for the distribution of time $Q:=\frac{Z}{V}$ (time an object spends inside the circle when traveling along a line with distance $Z$ and with velocity $V$). For possible values of $Q$ we have $Q \in [\frac{\sqrt{2-2\cos(\pi-2\delta)}}{\mu+\nu}, \frac{2}{\mu-\nu}]$.

On the wikipedia page https://en.wikipedia.org/wiki/Ratio_distribution, the ratio distribution is derived using (in my notation) $f_Z$ and $f_V$. Finding antiderivatives when $|y|$ is introduced turns out to be quite challenging, so I wanted to try a more intuitive approach.

My reasoning to find the distribution of $Q$ is like this (using only one integral instead of the double integral the wikipedia article uses): \begin{equation} \mathbb{P}(Q \leq x) = \mathbb{P}(\frac{Z}{V} \leq x) \\ = \mathbb{P}(Z \leq xV) \\ = F_Z(xV)\\ = \int_{y=\mu-\nu}^{\mu+\nu}f_Y(y)F_Z(xV|V=y)dy \\ =\int_{y=\mu-\nu}^{\mu+\nu}\frac{1}{2\nu}F_Z(xy)dy. \\ \end{equation} When computing this by hand, I find a long expression that by no means represents a probability distribution. Also, asking Mathematica to compute the integral gives no usable answer.

What am I doing wrong here? Thanks in advance!

  • 0
    I have not check all the details in your expression, but it looks ok. The approach you adopted is exactly the same way to derive the wikipedia result which in terms of the pdf. It is normal that the integral has no closed-form anti-derivative, and even mathematica may not know how to deal with this. By the way what is your result and how do you determine that "by no means represents a probability distribution"?2017-01-23
  • 0
    Thanks for your comment! What I mean with "by no means represents a probability distribution" is: when I pick some specific values for $\delta, \mu$ and $\nu$ (without violating the assumptions $\delta < 1/2 \pi$, $\nu<\mu$), the probability that $Q \leq$ its minimum value, becomes -13.1359.. As this probability should be equal to 0, I am sure something went wrong.2017-01-23
  • 0
    It is not clear how the (correct) formula $$P(Q\leqslant x)=\frac1{2\nu}\int_{\mu-\nu}^{\mu+\nu}F_Z(xy)dy$$ could give a negative result since $F_Z\geqslant0$ everywhere. A point to mention though, is that $F_Z(x)$ is equal to the expression in the RHS of the first identity in your post, **only for some restricted values of $x$**, namely, for $$2\cos\delta\leqslant x\leqslant2$$ Outside of this interval, $F_Z(x)$ is *not* equal to this expression but to $0$ or $1$.2017-01-24
  • 0
    I guess it is good news for me that you think the $\mathbb{P}(Q \leq x)$ expression is correct :). Back to my calculations for now.. Could you explain how you find the interval of $x$ for $F_Z(x)$? I did take into account that the expression $F_Z(x)$ is only true for certain values of $x$, but I found: \begin{equation} \sqrt{2-2\cos(\pi-2\delta)} \leq x \leq 2. \end{equation}2017-01-24
  • 0
    $$\cos(\pi-2\delta)=-\cos(2\delta)=1-2\cos^2(\delta)$$2017-01-24
  • 0
    Thanks for the comment! Should have been able to see that one myself.2017-01-25

0 Answers 0