I've got a reduction of the following partition problem to a certain scheduling problem:

Input: A list $a_1\leqslant\cdots\leqslant a_n$ of positive integers in non-decreasing order.

Question: Does there exist a vector $(x_1,\ldots,x_n)\in\{-1,1\}^n$ such that

\[\sum_{i=1}^na_ix_i=0\qquad\text{and}\] \[\sum_{i=1}^ka_ix_i\geqslant 0\quad\text{for all }k\in\{1,\ldots,n\}\]

Without the second condition it's just PARTITION, hence NP-hard. But the second condition seems to provide a lot of additional information. I'm wondering if there is an efficient way of deciding this variant. Or is it still hard?


1 Answer 1


Here is a reduction from PARTITION to this problem. Let $(a_1,\dots, a_n)$ be an instance of PARTITION. Assume that $a_1\leq a_2\leq \dots \leq a_n$.

Let $N$ be a “very large number”, e.g. $N = (\sum_{i=1}^n |a_i|) + 1$. Consider the instance $$\underbrace{N, \dots, N}_{5n \text{ times}}, N + a_1, \dots, N+a_n,\underbrace{4N, \dots, 4N}_{n \text{ times}}$$ of our problem.

  1. If there is a solution $x_1,\dots, x_n$ to PARTITION then $$\underbrace{1, \dots, 1}_{4n \text{ times}},-x_1,\dots,-x_n,x_1,\dots,x_n,\underbrace{-1,\dots,-1}_{n \text{ times}}$$ is a solution to our problem.

  2. If there is a solution $(x_1,\dots,x_{5n},y_1,\dots, y_n, z_1,\dots,z_n)$ to the instance of our problem (which we reduced an instance of PARTITION to), then $\sum_{i=1}^n a_i y_i \equiv 0 \pmod N$. Thus $$\sum_{i=1}^n a_i y_i = 0. $$ That is, $(y_1,\dots, y_n)$ is a solution to PARTITION.

  • $\begingroup$ Thanks Yury. In my application it is essential that the input list is ordered non-decreasingly, and the input $(N,a_1,\ldots,a_n,N)$ in your reduction is not. I'll modify the question to make the order requirement more explicit. $\endgroup$ Commented Apr 29, 2014 at 3:18
  • $\begingroup$ @thomas: I didn't notice that. Now I updated my solution. $\endgroup$
    – Yury
    Commented Apr 29, 2014 at 3:41

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.