How many points in the square

Determine the largest number μ \mu (to 3 decimal places) such that for any set C C of 20 points in the interior of the unit square U U , there exists a rectangle T T contained in U U that has the following 3 properties:

  1. The sides of T T are parallel to the sides of U U .
  2. The interior of T T contains exactly one point of C C .
  3. The area of T T is at least μ \mu .

Bonus: Generalize this for 4 n 4n points.


The answer is 0.083.

This section requires Javascript.
You are seeing this because something didn't load right. We suggest you, (a) try refreshing the page, (b) enabling javascript if it is disabled on your browser and, finally, (c) loading the non-javascript version of this page . We're sorry about the hassle.

2 solutions

Sharky Kesa
Dec 13, 2016

If you were trying small cases for n n in the generalised case, you might notice that μ = 1 2 n + 2 \mu = \frac {1}{2n+2} seems to work. Let's try and prove this.

Let's first try to establish an upper bound on μ \mu . We will prove μ 1 2 n + 2 \mu \leq \frac {1}{2n+2} . Suppose μ > 1 2 n + 2 \mu>\frac {1}{2n+2} . Consider the n n points ( 1 n + 1 , 1 2 ) , ( 2 n + 1 , 1 2 ) , , ( n n + 1 , 1 2 ) (\frac {1}{n+1}, \frac {1}{2}), (\frac {2}{n+1}, \frac {1}{2}), \ldots, (\frac {n}{n+1}, \frac {1}{2}) . Now split each of these points into four points by adding ( ± ϵ , ± ϵ ) (\pm \epsilon, \pm \epsilon) to each point. A rectangle that contains exactly one of these has height less than 1 2 + ϵ \frac {1}{2} + \epsilon and base less than 1 n + 1 + ϵ \frac {1}{n+1} + \epsilon . Thus, such a rectangle has area less than ( 1 n + 1 + ϵ ) ( 1 2 + ϵ ) < μ (\frac {1}{n+1} + \epsilon)(\frac {1}{2} + \epsilon) < \mu if ϵ \epsilon is small enough. Thus, we have a contradiction. Hence, μ 1 2 n + 2 \mu \leq \frac {1}{2n+2} . We shall now prove that μ = 1 2 n + 2 \mu = \frac {1}{2n+2} . First, though, let us solve what amounts to the one-dimensional version of the problem as in the following lemma.

Lemma. Let t 1 < t 2 < . . . < t k t_1 < t_2 < ... < t_k be points in the open interval ( 0 , 1 ) (0, 1) . Then there exists an open sub-interval I ( 0 , 1 ) I \subseteq (0, 1) of length 1 k 2 + 1 \frac {1}{\left \lfloor \frac{k}{2} \right \rfloor+1} that contains exactly one of the t i t_i .

Proof. If k = 2 h + 1 k=2h+1 is odd, consider the following h + 1 h+1 intervals.

( 0 , t 2 ) , ( t 2 , t 4 ) , ( t 4 , t 6 ) , , ( t 2 h 2 , t 2 h ) , ( t 2 h , 1 ) (0, t_2), (t_2, t_4), (t_4, t_6), \ldots, (t_{2h-2}, t_{2h}), (t_{2h}, 1)

Each such interval contains exactly one t i t_i . Moreover, the sum of the lengths of those intervals is 1. Hence, one of them has length at leat 1 h + 1 = 1 k 2 + 1 \frac {1}{h+1} = \frac {1}{\left \lfloor \frac{k}{2} \right \rfloor+1} , as required.

If k = 2 h k=2h is even, consider the following h + 1 h+1 intervals.

( 0 , t 2 ) , ( t 2 , t 4 ) , ( t 4 , t 6 ) , , ( t 2 h 2 , t 2 h ) , ( t 2 h 1 , 1 ) (0, t_2), (t_2, t_4), (t_4, t_6), \ldots, (t_{2h-2}, t_{2h}), (t_{2h-1}, 1)

Each such interval contains exactly one t i t_i . Moreover, the sum of their lengths is 1 + t 2 h t 2 h 1 1+t_{2h}-t_{2h-1} , which is more than 1. Hence, one of them has length greater than 1 h + 1 = 1 k 2 + 1 \frac {1}{h+1} = \frac {1}{\left \lfloor \frac{k}{2} \right \rfloor+1} , as required. _{\square}

Second, we now solve the 2D version of this problem. Let x 1 < x 2 < . . . < x k x_1 < x_2 < ... < x_k denote the set of x x -coordinates of the points in C C . Note that if all x x -coordinates of points in C C are different, then k = 4 n k=4n , otherwise k < 4 n k<4n . We also define x 0 = 0 x_0 = 0 and x k + 1 = 1 x_{k+1}=1 , so that x 0 < x 1 < < x k < x k + 1 x_0 < x_1 < \ldots < x_k < x_{k+1} . For i = 0 , 1 , 2 , , k , k + 1 i = 0, 1, 2, \ldots, k, k+1 , let l i l_i be the vertical line through x i x_i , and for i = 1 , 2 , . . . , k i=1, 2, ..., k , let m i = C l i m_i = |C \cap l_i| . Note that i = 1 k m i = 4 n \sum_{i=1}^k m_i = 4n . Applying the lemma along line l i l_i , there is a vertical interval I I of length 1 m i 2 + 1 \frac {1}{\left \lfloor \frac {m_i}{2} \right \rfloor + 1} that contains exactly one point of C l i C \cap l_i . Thus the rectangle bounded by the horizontal projections of I I onto l i 1 l_{i-1} and l i + 1 l_{i+1} contains exactly one point of C C . The area of this rectangle is x i + 1 x i 1 m i 2 + 1 \frac {x_{i+1} - x_{i-1}}{\left \lfloor \frac {m_i}{2} \right \rfloor + 1} , and we have one such rectangle for each i i .

Suppose, for the sake of contradiction, that all such areas are less than 1 2 n + 2 \frac {1}{2n+2} . It follows that

x i + 1 x i 1 < m i 2 + 1 2 n + 2 , for i = 1 , 2 , , k . ( 1 ) x_{i+1} - x_{i-1} < \dfrac {\left \lfloor \frac {m_i}{2} \right \rfloor +1}{2n+2}, \quad \text{for } i=1, 2, \ldots, k. \quad (1)

However,

( x k + 1 x k ) + ( x 1 x 0 ) + i = 1 k ( x i + 1 x i 1 ) = 2 ( x k + 1 x 0 ) = 2. ( 2 ) (x_{k+1} - x_k) + (x_1 - x_0) + \displaystyle \sum_{i=1}^k (x_{i+1} - x_{i-1}) = 2(x_{k+1} - x_0) = 2. \quad (2)

We know x k + 1 x k < x k + 1 x k 1 x_{k+1} - x_k < x_{k+1} - x_{k-1} , and x 1 x 0 < x 2 x 0 x_1 - x_0 < x_2 - x_0 . Using these and ( 1 ) (1) and ( 2 ) (2) , we have

4 n + 4 < ( m k 2 + 1 ) + ( m 1 2 + 1 ) + i = 1 k ( m i 2 + 1 ) = 2 m 1 2 + 2 m k 2 + 4 + i = 2 k 1 ( m i 2 + 1 ) m 1 + m 2 + 4 + i = 2 k 1 m i ( since 2 x 2 x , and m i 2 m i 1 for m i N + ) = 4 n + 4 , \begin{aligned} 4n+4 &< \left (\left \lfloor \dfrac {m_k}{2} \right \rfloor + 1 \right ) + \left (\left \lfloor \dfrac {m_1}{2} \right \rfloor + 1 \right ) + \displaystyle \sum_{i=1}^k \left (\left \lfloor \dfrac {m_i}{2} \right \rfloor + 1 \right )\\ &= 2 \left \lfloor \dfrac {m_1}{2} \right \rfloor + 2 \left \lfloor \dfrac {m_k}{2} \right \rfloor + 4 + \displaystyle \sum_{i=2}^{k-1} \left (\left \lfloor \dfrac {m_i}{2} \right \rfloor + 1 \right )\\ &\leq m_1 + m_2 + 4 + \displaystyle \sum_{i=2}^{k-1} m_i \quad (\text{since } 2 \left \lfloor \frac {x}{2} \right \rfloor \leq x, \text{ and } \left \lfloor \frac {m_i}{2} \right \rfloor \leq m_i - 1 \text{ for } m_i \in \mathbb{N}^{+})\\ &= 4n+4, \end{aligned}

which is a contradiction. Therefore, μ = 1 2 n + 2 \mu = \frac {1}{2n+2} , so we substitute n = 5 n=5 to get the value of μ \mu as 1 12 0.083 \frac{1}{12} \approx 0.083 .

Nice proof, Sharky!

I wonder what μ \mu would be for, say, 5 5 points...

Geoff Pilling - 4 years, 6 months ago

Hey @Sharky Kesa here's one for you, inspired by this problem! :0)

Enjoy!

Geoff Pilling - 4 years, 6 months ago
Geoff Pilling
Dec 12, 2016

Whoa... I love this problem! (+1)

Here is a justification for 1 12 \frac{1}{12} . See Sharky's solution for a more complete explanation about why/how the construction is optimal...

Although I don't yet have a proof that 1 12 \frac{1}{12} (which leads to the answer of 13 13 above) is the largest value for μ \mu , I do have an example for which you can achieve it, where you distribute the points as follows:

  • Four points forming a "tiny" box with the same orientation as U U around ( n 6 , 1 2 ) (\frac{n}{6},\frac{1}{2}) where n n ranges from 1 1 to 5 5 .

Or more explicity:

  • ( n 6 + δ , 1 2 + δ ) (\frac{n}{6}+\delta,\frac{1}{2}+\delta) where n n ranges from 1 1 to 5 5
  • ( n 6 + δ , 1 2 δ ) (\frac{n}{6}+\delta,\frac{1}{2}-\delta) where n n ranges from 1 1 to 5 5
  • ( n 6 δ , 1 2 + δ ) (\frac{n}{6}-\delta,\frac{1}{2}+\delta) where n n ranges from 1 1 to 5 5
  • ( n 6 δ , 1 2 δ ) (\frac{n}{6}-\delta,\frac{1}{2}-\delta) where n n ranges from 1 1 to 5 5

Let δ 0 \delta \rightarrow 0

Making this comment to seek clarity in your statements. You might be saying the same thing, so I'm just confirming

Doesn't this construction show that an upper bound is 1/12? IE take any E > 1 12 E > \frac{1}{12} , find the suitable δ \delta and show that any rectangle of size E E must contain at least 2 points. I believe that the missing step is instead why "Given any square, we can find such a rectangle", which shows that this can be achieved hence it's the least upper bound, IE maximum.

Calvin Lin Staff - 4 years, 6 months ago

Log in to reply

Yes. This is more or less what I was trying to say. I also forgot to mention that we want to let δ 0 \delta \rightarrow 0

I've just added this above.

Essentially, for the construction above, it is impossible to pick a rectangle with area μ > 1 12 \mu \gt \frac{1}{12} that contains only one point. Note, you can find a rectangle with zero points that is larger than 1 12 \frac{1}{12} , i.e. the bottom half of the square, just below the lowest point, but that isn't allowed by criteria (ii).

What I haven't yet managed to show is that it isn't possible to find a construction for which there is a upper bound for μ \mu which is lower than 1 12 \frac{1}{12} .

By the way, did I mention that I love this problem? Lots of potential for follow up questions! (i.e. What if you have 5 or more dots and the number of dots isn't divisible by n n ? - Surprisingly non trivial to solve)

Geoff Pilling - 4 years, 6 months ago

I'll post the full solution. :)

Sharky Kesa - 4 years, 6 months ago

Log in to reply

Cool... Looking forward to it, @Sharky Kesa !

Geoff Pilling - 4 years, 6 months ago

Log in to reply

LOL, it's been posted already.

Sharky Kesa - 4 years, 6 months ago

Log in to reply

@Sharky Kesa Ooops, not sure how I missed it... Nice solution! I wonder what μ \mu would be for, say, 5 5 points...

Geoff Pilling - 4 years, 6 months ago

Geoff, were you still going to edit this one? You could just change it into a "here's a justication for 1/12, see Sharky's solution for the reason the construction is optimal."

Jason Dyer Staff - 4 years, 5 months ago

Log in to reply

Done..................

Geoff Pilling - 4 years, 5 months ago

0 pending reports

×

Problem Loading...

Note Loading...

Set Loading...