For which integers $n>1$ is there a set of positive integers $S\subseteq \mathbb{N}$ with $n$ elements, and for every $s\in S$ the set $S\setminus\{s\}$ can be partitioned into two subsets with equal sum?
-
${1,3,5,7,9}$ almost works; $s=3$ is the problem. – Brian Hopkins Jul 17 '21 at 21:19
-
1$n=5$ is impossible, but $S={1,3,5,7,9,11,13}$ works fine. Is the answer all odd $n\ge7$? – bof Jul 18 '21 at 01:33
-
1There was a Putnam problem like this (maybe exactly this) some decades back. – Gerry Myerson Jul 18 '21 at 03:52
-
1@Gerry I know a problem on partitioning a multiset of positive reals without any term by two parts with equal sums and also equal size (then all elements must be equal). – Fedor Petrov Jul 18 '21 at 04:37
-
1@FedorPetrov That was Macalester Problem of the Week 1252, and Stan Wagon gave two references: The USSR Problem Book by Shklarsky, Chentov, and Yaglom, Freeman, 1962, and Peter Winkler, Mathematical Mindbenders, AK Peters, 2007, p. 23. – RobPratt Jul 18 '21 at 04:45
-
@FedorPetrov That problem (equal sums and also equal size) has appeared on math.se a couple of times, e.g. here: https://math.stackexchange.com/questions/1002267/after-removing-any-part-the-rest-can-be-split-evenly-consequences – bof Jul 18 '21 at 04:59
-
@Fedor, yes, the Putnam problem had that extra requirement. – Gerry Myerson Jul 18 '21 at 06:30
-
1See also https://mathoverflow.net/questions/105400/extending-an-assignment-property-from-q-to-r-or-c for the equal size version, including the question of what happens in abelian groups. – Gerry Myerson Jul 18 '21 at 08:56
1 Answers
As conjectured by bof, the answer is all odd $n \geq 7$.
Proof. Let $S$ be a set of positive integers such that $S \setminus \{s\}$ can be partitioned into two sets of equal sum for all $s \in S$. By parity considerations, note that all elements of $S$ are either all odd or all even. If all elements of $S$ are even, then the set obtained from $S$ by dividing every element by $2$ is also a valid set. Repeating this argument, we may assume that all elements of $S$ are odd. This implies that $n$ must be odd. Clearly, $n \notin \{1,3\}$ and bof has shown $n=5$ is also impossible (see comment below). To complete the proof we now show that every odd $n \geq 7$ is possible.
As noted by bof, $S=\{1,3,5,7,9,11,13\}$ shows that $n=7$ is possible, and it is easy to check $S=\{1,3,5,7,9,11,13,15,17\}$ is also valid. We will show by strong induction on $k$ that $S=\{1,3, \dots, 4k+1\}$ is a valid set. If $s \in \{1, 3, \dots, 4k-7\}$, then $\{1, 3, \dots, 4k-7\} \setminus \{s\}$, can be partitioned into two equal sum sets by induction. Since $\{4k-5, 4k-3, 4k-1, 4k+1\}$ can also be partitioned into two equal sum sets, we are done. If $s \in \{4k-5, 4k-3\}$, then by induction $\{1, 3, \dots, 4k-3\} \setminus \{s\}$ can be partitioned into two equal sum sets $S_1$ and $S_2$. Suppose that $1 \in S_1$. Then $(S_1 \setminus \{1\}) \cup \{4k+1\}$ and $S_2 \cup \{1, 4k-1\}$ are equal sum subsets of $S \setminus \{s\}$. Suppose $s=4k-1$. By induction, there is a partition of $\{1, 3, \dots, 4k-5\}$ into two equal sum sets $S_1$ and $S_2$. Suppose $4k-5 \in S_1$, Then there must be some $\ell \in S_1$ such that $\ell-2 \in S_2$. We obtain the required partition of $S \setminus \{s\}$ by swapping $\ell$ and $\ell-2$, adding $4k+1$ to $S_1$, and adding $4k-3$ to $S_2$. Finally, suppose $s=4k+1$. By induction, there is a partition of $\{1, 3, \dots, 4k-7\} \cup \{4k-3\}$ into two equal sum sets $S_1$ and $S_2$. Suppose $4k-7 \in S_1$. Then there must be some $\ell \in S_1$ such that $\ell-2 \in S_2$. The required partition of $S \setminus \{4k+1\}$ is obtained by swapping $\ell$ and $\ell-2$, adding $4k-1$ to $S_1$, and adding $4k-5$ to $S_2$.
- 31,500
-
How do you conclude that 4 divides all elements? (I would divide them all by 2 instead.) – Fedor Petrov Jul 18 '21 at 04:23
-
-
3To see that $n=5$ is impossible: Suppose $S={a,b,c,d,e}$, $a\lt b\lt c\lt d\lt e$. Assume for a contradiction that both $S\setminus{a}$ and $S\setminus{b}$ can be split into two sets with equal sums. Then we have: $$\text{either }b+e=c+d\text{ or }e=b+c+d$$ and $$\text{either }a+e=c+d\text{ or }e=a+c+d$$ and we very quickly reach a contradiction. – bof Jul 18 '21 at 04:48