3
$\begingroup$

There are seven cups, each with a water tap that adds water into it at the same rate. There are also three buckets. At any point, if any of these conditions is met, you pour water from the cups into the buckets according to the following rules:

  • If there is one unit of water in cups $1,2,3,4$ combined, you pour all that water into the first bucket.
  • If there is one unit of water in cups $4,5,6$ combined, you pour all that water into the second bucket.
  • If there is one unit of water in cups $6,7$ combined, you pour all that water into the third bucket.

As time goes on, what does the ratio between water in the three buckets converge to?

My computer program shows that it the limit of this ratio is $9:4:4$, but it is not clear to me how to prove it formally. Can we set up some kind of "steady-state" equations? More interestingly, can we determine the answer $9:4:4$ without using a program?

$\endgroup$
7
  • $\begingroup$ Indeed combined, the cups get $7$ units of water per unit of time and so the buckets must too. So if your computer is correct then in $17$ units of time the first bucket gets $63$ units, and each of the other buckets $28$ units. Is there a cycle? Is there the possibility cups $4$ or $6$ do not know which bucket to go into? $\endgroup$
    – Henry
    Commented Oct 21, 2021 at 0:17
  • $\begingroup$ @Henry After $17\cdot 7 = 119$ times pouring into the buckets, the split is indeed $63,28,28$. But it doesn't cycle here - there is still water left in cups $1,2,3,4,5$. $\endgroup$
    – pi66
    Commented Oct 21, 2021 at 0:36
  • $\begingroup$ Experimentally, it seems like there's no cycle, but there's a sort of limiting cycle: if you look at states that are $17$ bucket-pours apart, they get closer and closer as time goes on. We may be able to prove that as long as the current state is sufficiently close to some theoretical limit, the next $17$ steps are predictable, and then the result is even closer to that limit. $\endgroup$ Commented Oct 21, 2021 at 0:40
  • $\begingroup$ So after lots of $17$s, the amount left in the first four cups is about $0.090250244859752$ each and in the fifth cup about $0.410066898214241$ with the last two empty and then the drops are into buckets $A,B,A,C,A,B,A,A,C,B,A,A,C,B,A,A,C$ and then you are back there again. I am surprised that the first batch of $17$ with its different starting position (especially with cup $5$) matches this pattern $\endgroup$
    – Henry
    Commented Oct 21, 2021 at 1:03
  • $\begingroup$ depending on which subsets of cups combine their amount into one unit result, by which rules, into pouring into which bucket(s), one gets (or not) certain convergence. But, is the order of rules important? I guess so, since, combined amount of water in different subsets of cups reaching one unit can simultaneously meet more than one condition to decide to poor somewhere. $\endgroup$ Commented Oct 21, 2021 at 1:07

2 Answers 2

3
$\begingroup$

I only have a terrible proof that the pattern continues.

If the cups are in a state $(a,a,a,a,b,0,0)$ with $\frac1{12} < a \le \frac18$ and $\frac{788a-61}{148} \le b \le \frac{8a+1}{4}$, then $17$ steps later they end up in the state $(a',a',a',a',b',0,0)$ with $a' = \frac{3644 a-640 b+1805}{20736}$ and $b' = \frac{320 a-49 b+2117}{5184}$, adding $9$ units of water to the first bucket and $4$ units of water to each of the other two. The new $a'$ and $b'$ also satisfy the inequalities we assumed about $a$ and $b$.

Moreover, after $17$ steps from the state $(0,0,0,0,0,0,0)$, we arrive at a state $(a,a,a,a,b,0,0)$ with $a = \frac{1805}{20736}$ and $b = \frac{2117}{5184}$, adding $9$ units of water to the first bucket and $4$ units of water to each of the other two. These values of $a$ and $b$ satisfy the inequalities in the previous paragraph.

All this could in theory be checked by hand, very painfully; it requires verifying a finite number of inequalities at each of the $17$ steps. Of course, I did it in Mathematica.

As a point of curiosity, if the cups are at some point in the state $$ \left(\frac{99885}{1106756}, \frac{99885}{1106756}, \frac{99885}{1106756}, \frac{99885}{1106756}, \frac{113461}{276689}, 0,0\right) $$ then $17$ steps later they return to that exact state, adding $9$ units of water to the first bucket and $4$ units of water to each of the other two. This state is the limit of the $17$-step cycle we're looking at above.

$\endgroup$
0
$\begingroup$

I think you can show this using Birkhoff's ergodic theorem. Represent the amount of water in the first three glasses as $x_1$, and the amount of water in glass $i$ for $i \ge 4$ as $x_{i-2}$. Then you can treat the amount of water in all the glasses as a point $x \in \mathbb{R}^5$. You can define a function $f:\mathbb{R}^5 \to \mathbb{R}^5$ by saying $f(x)$ is the amount of water in each glass just before you empty a bucket (let's assume the starting amounts of water in each glass aren't rationally related, so that no two rules ever activate at the same time). I think (though I haven't checked) that $f$ is ergodic, i.e. if you choose a random initial condition and repeat this process infinitely many times, then eventually you get arbitrarily close to every point in the space. Also, $f$ is measure-preserving, since $f$ maps any subset of the set of possible values of $x$ to an equal-volume set. Therefore, Birkhoff's ergodic theorem says that the proportion of the $\{x,f(x),f^2(x),\ldots\}$ where bucket 1 is the next bucket to be filled, equals the fraction of the possible volume where bucket 1 is the next bucket to be filled.

$\endgroup$
1
  • 1
    $\begingroup$ Experimentally, it doesn't seem to be ergodic; rather, the sequence $x, f^{17}(x), f^{34}(x), \dots$ appears to converge to a limit I can't quite identify. $\endgroup$ Commented Oct 21, 2021 at 0:34

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .