We pick randomly two points, $p_1(x_1,y_1)$ and $p_2(x_2,y_2)$ inside a circle of origin $S$ with radius $R$ and we draw two circle $C_{1-2} (p_1,\sqrt {|x_1-x_2|²+|y_1-y_2|²})$ and $C_{2-1} (p_2,\sqrt {|x_1-x_2|²+|y_1-y_2|²})$ centred at the two points $p_1$ and $p_2$ with radius equal to the distance between them. What is the expected area $E(AI)$ of the intersection of the two circles with condition that the intersection contains the origin $S$ ?
If we pik a third point $p_3$ and we draw two circles $C_{1-3}$ and $C_{3-1}$, what is the expected area of the intersection of the two circles $C_{1-3}$ and $C_{3-1}$ ,with condition that contains the origin, with $AI$?

This is part of a problem which consist in determining the region where a device is.

  • $\begingroup$ How do you "randomly" pick a point? By randomly picking their cartesian coordinates? Then the point may fall outside the circle. Or polar coordinates? Then the distribution is not density-uniform (higher concentration towards the circle center). $\endgroup$ – mbaitoff Apr 9 '13 at 11:01
  • $\begingroup$ @mbaitoff: You're right in pointing out that the question fails to specify a distribution, but you're not right in suggesting that there's a problem in specifying one. Presumably the points are to be chosen uniformly with respect to the standard area measure; my answered assumed that this was implied. $\endgroup$ – joriki Apr 9 '13 at 17:23

I'll calculate the result for two points randomly uniformly picked the unit disk; it scales with $R^2$.

Consider $p_1$ at distance $r_1$ from the origin. Where can $p_2$ lie such that the origin is closer to either point than the other point? For the origin to be closer to $p_1$ than $p_2$, we must have $p_2$ outside the circle around $p_1$ with radius $r_1$. For the origin to be closer to $p_2$ than $p_1$, we must have $p_2$ on the origin's side of the line equidistant from the origin and $p_2$.

Thus, given $p_1$ at $(0,r_1)$, the admissible region for $p_2$ is the difference of two circular segments, the segment of the circle around the origin with radius $1$ up to $y=r_1/2$ minus the segment of the circle around $p_1$ with radius $r_1$ up to $y=r_1/2$. The area whose expected value we seek is proportional to the squared distance $d^2$ between $p_1$ and $p_2$. Thus, to calculate the desired expected value, we need the area of a circular segment and the integral over a circular segment of the squared distance to a given point on its axis. In detail:

$$ \begin{align} A(r,a)&:=\int_{-r}^a\mathrm dy\int_{-\sqrt{r^2-y^2}}^{\sqrt{r^2-y^2}}\mathrm dx\;, \\ B(r,a,b)&:=\int_{-r}^a\mathrm dy\int_{-\sqrt{r^2-y^2}}^{\sqrt{r^2-y^2}}\mathrm dx\left(x^2+(y-b)^2\right)\;, \\ \langle d^2\rangle&=\frac{\displaystyle\int_0^1\mathrm dr_1r_1\left(B\left(1,\frac{r_1}2,r_1\right)-B\left(r_1,-\frac{r_1}2,0\right)\right)}{\displaystyle\int_0^1\mathrm dr_1r_1\left(A\left(1,\frac{r_1}2\right)-A\left(r_1,-\frac{r_1}2\right)\right)}\;. \end{align} $$

Some help from our electronic friends yields the integrals:

$$ \begin{align} A(r,a)=&a\sqrt{r^2-a^2}+r^2\arctan\frac a{\sqrt{r^2-a^2}}+\frac\pi2r^2\;, \\ B(r,a,b)=&\frac16\left( \left(6b^2a+8b\left(a^2-r^2\right)+r^2a+2a^3\right)\sqrt{r^2-a^2}\right. \\ &\left.+3r^2\left(r^2+2b^2\right)\left(\frac\pi2+\arctan\frac a{\sqrt{r^2-a^2}}\right)\right)\;. \end{align} $$

In the negative terms with $r=r_1$, the arctangent is


Then further integral crunching yields

$$ \begin{align} \langle d^2\rangle&=\frac{\displaystyle\frac1{288}(88\pi+81\sqrt3)-\frac1{288}\left(8\pi-3\sqrt3\right)}{\displaystyle\frac1{16}(4\pi+3\sqrt3)-\frac14\left(\frac\pi3-\frac{\sqrt3}4\right)} \\ &= \frac{\displaystyle\frac5{18}\pi+\frac7{24}\sqrt3}{\displaystyle\frac\pi6+\frac{\sqrt3}4} \\ &= \frac16\frac{20\pi+21\sqrt3}{2\pi+3\sqrt3} \\ &\approx1.44034\;, \end{align} $$

which is confirmed by numerical simulations. The average squared distance of two points uniformly randomly picked in the unit disk without conditions is $1$, so the condition that the lune determined by the points contain the origin increases the average squared distance considerably.

The area to be averaged is $2A(d,-d/2)=d^2(4\pi-3\sqrt3)/6$, so the expected area is

$$ \frac{4\pi-3\sqrt3}6\langle d^2\rangle=\frac1{36}\frac{(4\pi-3\sqrt3)(20\pi+21\sqrt3)}{2\pi+3\sqrt3}\approx1.76927\;. $$

The problem with three points seems considerably more involved.


Some more details on the calculation, to respond to the comments.

The basic idea is that the conditional expectation value of $d^2$ is the integral of $d^2$ over the volume satisfying the condition, divided by the volume satisfying the condition. Because we're dividing two integrals, we don't need to worry about constant factors. If we were instead calculating the probability for the condition to be fulfilled, we'd have to multiply the denominator by $2\pi$ for the missing integration over $\phi_1$ and divide it by $\pi^2$ for normalizing the density, for an overall factor of $2/\pi$, so the probability for the condition to be fulfilled is


But since we're dividing two integrals, this factor $2/\pi$ cancels. The integration is performed in polar coordinates for $p_1$ and in Cartesian coordinates for $p_2$; the origin for the coordinates is the centre of the unit circle in case of the polar coordinates for $p_1$ and the centre of the respective circle with radius $r$ in case of the Cartesian coordinates for $p_2$, and those are two different circles in the two terms of the difference.

The difference is the difference of the integrals over two different circular segments (as discussed in the comments). Both segments have the line equidistant between $p_1$ and the centre of the unit circle as their linear boundary. The positive term is the integral over the lower segment of the unit circle $C(O,1)$ defined by that line, and the negative term is the integral over the lower segment of the circle $C(p_1,r_1)$ defined by that line, where $O$ is the origin and $C(p,r)$ denotes the circle with origin $p$ and radius $r$.

Since the segments both have the same form and just differ in their centre and radius, the calculation is organized into calculating integrals $A$ and $B$ with general radius $r$ and boundary at $a$ and then substituting specific values in the two terms of the difference. Since Cartesian coordinates relative to the centre of the respective circle are used, the $y$ coordinate $b$ of $p_1$ relative to the respective centre also needs to be taken into account in calculating the squared distance $d^2$ from $p_1$.

So in the integrals $A$ and $B$ over a circular segment ($A$ being the integral of $1$, i.e. the volume, and $B$ being the integral of $d^2$), the parameter $r$ is the radius of the circle, the parameter $a$ is the upper $y$ coordinate (relative to the circle's centre) at which the circular segment ends, and the parameter $b$ is the $y$ coordinate of $p_1$ (relative to the circle's centre).

For the positive term, which is for the circular segment of the unit circle $C(O,1)$, the radius $r$ is $1$, the $y$ coordinate $a$ of the bisector relative to the centre at $O$ is $r_1/2$, and the $y$ coordinate $b$ of $p_1$ relative to the centre at $O$ is $r_1$ (since $p_1$ is at $(0,r_1)$). For the negative term, which is for the circular segment of the circle $C(p_1,r_1)$, the radius $r$ is $r_1$, the $y$ coordinate $a$ of the bisector relative to the centre at $p_1$ is $-r_1/2$, and the $y$ coordinate $b$ of $p_1$ relative to the centre at $p_1$ is $0$.

Since we're integrating over the lower segments of the respective circles, the $y$ coordinate runs from its minimal value $-r$ to its value $a$ at the linear boundary, and the $x$ coordinate takes its entire range of values from $-\sqrt{r^2-y^2}$ to $\sqrt{r^2-y^2}$.

This is all for the integration over $p_2$ given $p_1$ fixed at $(0,r_1)$. Then we also have to integrate over $p_1$. The integration over the polar angle $\phi_1$ of $p_1$ just yields a constant factor $2\pi$ that cancels (see above), since it's only the relative orientation of $p_1$ and $p_2$ that matters, not the absolute orientation of $p_1$. So we can just integrate the result we got for $p_1$ at $(0,r_1$) over $r_1$, taking into account the Jacobian $r_1$ of the polar coordinates.

| cite | improve this answer | |
  • $\begingroup$ Thanks, please it is not so obvious how you set conditions on $p_1$ and $p_2$ for the intersection to include the origin. For me the condition is, if we fix $p_1$ at distance $r_1$ from the origin, then $p_2$ must not be in the circular segment of $C(S,R)$ cut off by the line equidistant from $S$ and $p_1$. and also not in circular segment of $C(p_1,r_1)$ cut off by the line equidistant from $p_1$ and $S$ (the segment that contain the origin and not $P_1$). $\endgroup$ – sys Apr 9 '13 at 13:46
  • $\begingroup$ @sys: Yes, that's right. Another way of saying the same thing is that $p_2$ must be in the other segment of $C(S,R)$, but not in that segment of $C(p_1,r_1)$, and that's what I calculate: the integral over the other segment of $C(S,R)$ minus the integral over that segment of $C(p_1,r_1)$. $\endgroup$ – joriki Apr 9 '13 at 14:15
  • $\begingroup$ Thanks, that's great. What do you mean by $a$ and $b$ in the integral? And what about the case when adding a third point? $\endgroup$ – sys Apr 9 '13 at 15:12
  • $\begingroup$ @sys: $a$ and $b$ are just parameters to generalize the two segments. $a$ is the $y$ coordinate delimiting the segment, and $b$ is the distance of $p_1$ from the origin of the respective circle. These parameters are given concrete values for the two segments in the expression for $\langle d^2\rangle$. Regarding the third point, as I said, that seems rather involved; the intersection has a complicated shape and I don't see how to reformulate the problem to avoid dealing with that. It may well be tractable, but it looks like a big headache and I don't feel like trying it right now. $\endgroup$ – joriki Apr 9 '13 at 15:55
  • $\begingroup$ May be the function of the intersection of three circles can help? Can you please draw a figure to show the parameter $a$ $b$ and $y$, to make it more clear. $\endgroup$ – sys Apr 10 '13 at 0:11

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.