60

Here's a riddle that I've been struggling with for a while:

Let $A$ be a list of $n$ integers between 1 and $k$. Let $B$ be a list of $k$ integers between 1 and $n$. Prove that there's a non-empty subset of $A$ and a (non-empty) subset of $B$ having the same sum.

Example: Say $n=3,\ k=5$, and $A=\{3,4,5\},\ B=\{1,1,2,3,3\}$. Then we can find $\{1,3,3\}\subset B$ and $\{3,4\}\subset A$ with the same sum (I know there're are simpler solutions in this example, it's just for demonstration).

I tried to attack it from different directions: induction, pigeon-hole, combinatorics, but I couldn't make it work. Suggestions?

Addition

As this was so highly voted, I thought I should tell how I came about this riddle - it's an interesting story: I heard it from a friend of mine about 10 years ago when I just finished high school and he just graduated in math. I remeber him telling me the riddle and his solution, and I thought "math is so cool, some day I'll also have a degree in math and will be able to solve riddles like this". I don't know why I suddenly remembered this conversation, and why I remember only the problem and not the solution. but it turns out that now I have a degree, but I still can't.

yohBS
  • 2,918
  • 3
    Are you familiar with Ferrers diagrams? The Ferrers diagram for $n$ integers between $1$ and $k$ is the transpose of a Ferrers diagram for $k$ integers between $1$ and $n$. Perhpas one gets some mileage out of this observation. – Gerry Myerson Feb 18 '13 at 23:28
  • I am curious if stars and bars is relevant to this problem. – Average Feb 18 '13 at 23:20
  • 5
    You should probably make the distinction between set and multiset. As sets, ${1,1,2,3,3}$ and ${1,2,3}$ are equal. Repetitions are not counted for sets, whereas multisets do mind them (but not the order, of course). – Asaf Karagila Feb 19 '13 at 00:41
  • How can there be a list of n integers between 1 and k & a list of k integers between 1 and n unless n=k?and if n=k,both the lists will be just the same...and so it is obvious that there's a non-empty subset of A and a (non-empty) subset of B having the same sum. – Shagun25 Feb 19 '13 at 14:15
  • 6
    @user62965: You misunderstand. Repetitions are allowed, as the OP's example makes clear. – Peter Smith Feb 19 '13 at 14:22
  • 2
    After 18hrs without an answer, I'm beginning to feel not so bad about failing to crack this one .... – Peter Smith Feb 19 '13 at 17:18
  • 5
    out of curiosity, how did you come across this ? – Albert Feb 19 '13 at 17:28
  • 1
    @Glougloubarbaki I added that to the question... – yohBS Feb 19 '13 at 20:34

2 Answers2

29

Nice problem! I almost hate to post a solution. If you like puzzles and haven't put in any time on this one yet, I encourage you not to read further.

Imagine writing the numbers in $A$ on a stack of cards, one number per card. We write the numbers of $B$ on a separate stack, again one per card. We then recursively define a sequence $s_j$ as follows:

Initial value: $s_0=0$.

If $s_j \leq 0$, look at the top card of the $A$ stack; let the number written on it be $a$. Set $s_{j+1} = s_j+a$, and remove that card from the stack.

If $s_j > 0$, look at the top card of the $B$ stack; let the number written on it be $b$. Set $s_{j+1} = s_j-b$, and remove that card from the stack.

Continue until you attempt to draw a card from an empty stack.

Example: Taking $A = ( 3,4,5 )$ and $B = (1,1,2,3,3)$ as in the OP, we have $$\begin{array}{|rrrrr|} \hline s & A \ \mbox{deck} & B \ \mbox{deck} & A \ \mbox{cards used} & B \ \mbox{cards used} \\ \hline 0 & 345 & 11233 & & \\ \hline 3 & 45 & 11233 & 3 & \\ \hline 2 & 45 & 1233 & & 1 \\ \hline 1 & 45 & 233 & & 1 \\ \hline -1 & 45 & 33 & & 2 \\ \hline 3 & 5 & 33 & 4 & \\ \hline 0 & 5 & 3 & & 3 \\ \hline 5 & & 3 & 5 & \\ \hline 2 & & & & 3 \\ \hline \end{array}$$

At this point we stop, because the next step would be to draw from the $B$ deck, but the $B$ deck is empty. (In this example, the $A$ deck is also empty, but that is a coincidence; they don't have to both run out.)

Lemma 1: The numbers $s_j$ are always between $-n+1$ and $k$.

Proof by induction on $j$. The base case is true. If $s_j$ is between $-n+1$ and $0$, then $s_{j+1}$ is between $-n+k+1$ and $k$; if $s_j$ is between $1$ and $k$ then $s_{j+1}$ is between $-n+1$ and $-n+k$. Since the intervals $[-n+1, 0]$ and $[1, k]$ cover every integer in $[-n+1, k]$, this shows that, if $s_j \in [-n+1, k]$ then $s_{j+1} \in [-n+1,k]$. $\square$.

Lemma 2: The sequence $s_j$ repeats a value. (In the example, the values $0$, $2$ and $3$ are repeated.)

Let's suppose that the game ends when we try to draw from the $B$ stack; the other case is similar. So we must make $k+1$ attempts to draw from the $B$-stack (including the attempt that fails). At the time that we make each attempt, $s_j$ is positive. So we have $k+1$ positive values of $s_j$. By Lemma 1, each of these values lies in $[1,k]$. So some value must appear more than once. $\square$

Proof of the result Let $s_i=s_j$. Then the set of $A$ cards which are dealt between $s_i$ and $s_j$ must have the same value as the set of $B$ cards. For example, if we use the repeated $3$'s in the example sequence, then we see that $-1-1-2+4=0$ or, in other words, $4=1+1+2$. $\square$

user1001001
  • 5,236
David E Speyer
  • 64,196
4

I've done this problem in my exam today. It's a nice problem and here is my solutuion: First, I denote $A=\{x_1,x_2,...,x_n\}$, $1\leq x_1\leq x_2\leq ...\leq x_n\leq k$ and $B=\{y_1,y_2,...,y_k\}$, $1\leq y_1\leq y_2\leq...\leq y_k\leq n$. Set $a_p=\sum_{i=1}^{p}x_i$ ($1\leq p\leq n$) and $b_q=\sum_{j=1}^{q} y_j$ ($1\leq q\leq k$).Without lost of general I can assume that $a_n\leq b_k$. From that, $\forall 1\leq p\leq n$, there exists $1\leq f(p)\leq k$ is the smallest index that $a_p\leq b_{f(p)}$. $$b_{f(1)}-a_1;...;b_{f(n)}-a_n$$ I comment that all $b_{f(i)}-a_i<n$ (if $n\leq b_{f(p)}-a_p$ with some $p$ $\rightarrow n<b_{f(p)}\rightarrow f(p)>1$, so we have $n\leq b_{f(p)-1}+y_{f(p)}-a_p\rightarrow 0\leq n-y_{f(p)}\leq b_{f(p)-1}-a_p\rightarrow a_p\leq b_{f(p)-1}$ (contradiction because I choose $f(p)$ is the smallest index))

If $b_{f(p)}-a_p=0$ with some $p$, I have the proof, if not, by the pingeonhole principle, we have $b_{f(r)}-a_r=b_{f(s)}-a_s$ with some $1\leq r< s\leq n$ $$\rightarrow b_{f(s)}-b_{f(r)}=a_s-a_r$$ $$\rightarrow \sum_{i=r+1}^{s}x_i=\sum_{j=f(r)+1}^{f(s)}y_j$$ So i have the proof. (do i have mistake in somewwhere?)

Zootopia
  • 753