Expected number of semicircles

152 Views Asked by At

Choose a random semicircle on a circle with uniform probability. Then choose another, and another and another until the circle is fully covered. What is the expected number of steps in this process?

I've been trying to use: $E(x)=1\cdot P(x=1) + 2\cdot P(x=2) + 3 \cdot P(x=3) ... + n\cdot P(x=n)$ and have found that the first two terms are zero.

I'm not 100% sure about $3 \cdot P(x=3)$ but I think $P(x=3)=1/4$ with similar reasoning as 3blue1brown in his video about points on a circle. https://www.youtube.com/watch?v=OkmNXy7er84

From there, I'm pretty stuck.

2

There are 2 best solutions below

2
On

Firstly note that we can represent each semi-circle using its mid-point. Therefore, choosing a semi-circle uniformly at random is equivalent to choosing a point uniformly in the interval $[0, 2\pi)$. Consider a sample of $n$ such points given by $\{X_1, X_2, \dots, X_n\}$ and let $x_{\min}(n)$ and $x_{\max}(n)$ be the minimum and maximum values in this sample. Let us try to find the probability that this sample will not be able to cover the circle.

This sample will not be able to cover the circle if the semi-circles corresponding to $x_{\min}(n)$ and $x_{\max}(n)$ do not intersect. Since we are dealing with a circle, the interval of $[0, 2\pi)$ can be thought of as a periodic interval and the above condition can be written as $(2\pi + x_{\min}(n)) - x_{\max}(n) > \pi$ which is equivalent to $x_{\max}(n) - x_{\min}(n) < \pi$.

To compute the probability of the above event, first let us compute $\Pr(x_{\max}(n) < a + \pi | x_{\min}(n) = a )$ for some $a \in [0, 2\pi)$. Since the distribution is uniform, this evaulates to $\displaystyle \left(\frac{\pi}{(2\pi - a)}\right)^{n - 1}$. The exponent is $n - 1$ because only $n-1$ points remain after setting the minimum one equal to $a$. Using the distribution of the minimum of $n$ uniformly distributed random variables, one can write \begin{align} \Pr(x_{\max}(n) - x_{\min}(n) < \pi) & = n \int_{0}^{2\pi} \left(\frac{\pi}{(2\pi - a)}\right)^{n - 1} \left(1 - \frac{a}{2\pi}\right)^{n-1} \frac{da}{2 \pi} \\ & = \frac{n}{2^{n-1}} \end{align}

If $T$ denotes the random number of steps, then $P(T> n) = \Pr(x_{\max}(n) - x_{\min}(n) < \pi) = \frac{n}{2^{n-1}}$. Using the formula, $\displaystyle \mathbb{E}[T] = \sum_{n = 0}^{\infty} \Pr(T > n)$, you can compute the required value.

0
On

I'm disappointed now that I know I'm not the only one to have come up with this question (though I an surprised Grant of all people knows about this thing I thought was obscure), but I did spend a lot of time on it and believe I've solved it. Here goes. (Also I don't have much experience with typing mathematical notation so enjoy some badly formatted numbers!)

Step 1

Definitions first: I will define the probability that the circle is fully covered on the nth turn (and not on the (n-1)th turn) as "P(done by n)". I will also define that thee probability that the circle is fully covered by the nth turn (and maybe on the (n-1)th turn)as "P(finished by n)".

Now we try to prove a somewhat intuitive property that P(done by n) = P(finished by n) - P(finished by n-1). The way to do this is to imagine placing semicircles one at a time. If we stop at n-1, the probability the circle is covered is P(finished by n-1). If we instead stop at n, the probability the circle is covered is P(finished by n), but it is also P(finished by n-1) + P(done by n), which then implies our first statement. Thinking about what this means, it makes sense. Essentially, P(finished by 5) = P(finished by 4, implies finished by 5) + P(not finished by 4, done by 5).

Step 2

Now, the hard part. It'll be particularly hard for me to describe without pictures.

Let's try to compute P(finished by n) for an example, say n=5.

Instead of placing semicircles one at a time, let's randomly place their diameters all at once, and then randomly choose which half of the circle each semicircle will cover. This is essentially the same steps done in a different order. Now, WLOG, rotate the entire system (circle + semicircles) so one of the semicircles covers the bottom half of the circle. We now have 4 diameters (excluding the one that is currently horizontal), and a randomly chosen side of each diameter is covered.

We may now do complementary counting; instead of finding the probability that the whole circle is covered, let's find the probability that some of the table is uncovered.

Observation 1: The 4 lines divide the upper half of the circle into sectors. A sector can either be fully covered or fully exposed; it cannot be half covered.

Observation 2: When one sector is chosen to be uncovered, no other sectors can also be uncovered. This is because the two semicircles bordering said sector will together fill the entire circle outside of this sector. This means at most one sector can be uncovered.

Observation 3: Given its diameter, a semicircle only has one way to not cover a sector: it must cover the half of the circle not containing the sector, because by definition, the other half contains the sector which must remain uncovered.

Summary: We now know that only one of the sectors in the top half of the circle may be uncovered, and that there is only one way to arrange the semicircles so they do not cover said sector. In the general case, n-1 diameters divide the upper half of the circle into n sectors, so there are n ways to leave a sector uncovered.

Note that when we choose which side each semicircle should cover, we have 2 options per semicircle (excluding the one we already chose, that being the one covering the bottom half of the circle), which leads to 2^(n-1) ways to choose sides for the semicircles. This means the probability a sector remains uncovered is n/(2^(n-1)), so the probability the entire circle is covered is 1 - n/(2^(n-1)).

Step 3

The actual math begins.

Using our equation that P(done by n) = P(finished by n) - P(finished by n-1), we can say

P(done by n) = (1 - n/(2^(n-1))) - (1 - (n-1)/(2^(n-2)))

= 1 - n/(2^(n-1)) - 1 + (n-1)/(2^(n-2))

= - n/(2^(n-1)) + 2(n-1)/(2^(n-1))

= ((2n-2)-n)/(2^(n-1))

= (n-2)/(2^(n-1)).

Using the formula for expected value, the expected number of semicircles needed to cover the circles is (3)(3-2)/(2^(3-1)) + (4)(4-2)/(2^(4-1)) + (5)(5-2)/(2^(5-1)) + ... + (n)(n-2)/(2^(n-1)) + ...

This can be evaluated somewhat easily if you know what you're doing, but I don't, so here's my solution. This sum is the same as adding all of the following:

1/4 2/8 3/16 4/32 ...

1/4 2/8 3/16 4/32 ...

1/4 2/8 3/16 4/32 ...

2/8 3/16 4/32 ...

3/16 4/32 ...

4/32 ...

...

This can be rewritten as adding up the following:

1/4 2/8 3/16 4/32 ...

1/4 2/8 3/16 4/32 ...

1/4 2/8 3/16 4/32 ...

1/8 2/16 3/32 ...

1/16 2/32 ...

1/32 ...

...

and then also adding on these:

1/8 1/16 1/32 ...

2/16 2/32 ...

3/32 ...

...

Through some number manipulation, this can be simplified to

(1/4 + 2/8 + 3/16 + 4/32 + ...) * (3 + 1/2 + 1/4 + 1/8 + ... + 1/2 + 1/4 + 1/8 + ...)

And knowledge of some basic series will make it clear that this evaluates to

(1) * (3 + 1 + 1) = 5, which I believe to be the answer.