Let $S\subseteq \{0,\ldots,n\}^d$ be a set of $d$-dimensional vectors of with bounded, natural, coordinates.
We are given that $$v'+v_1+\ldots+v_t=u'+u_1+\ldots+u_s$$ where $v_1,\ldots,v_t,u_1,\ldots,u_s,v',u'\in S$ (and the vectors are not necessarily distinct).
That is, two sets of vectors whose sums are equal.
I want to prove that, if $t$ and $s$ are large enough, then there exist subsets $I\subseteq \{1,\ldots,t\}$ and $J\subseteq \{1,\ldots > s\}$ such that $$\sum_{i\in I}v_i=\sum_{j\in J}u_j$$
Note that, without the assumption of the equal sum above, there may not be such sets (e.g., if all the $v_i$ are $(1,0)$, and all the $u_i$ are $(0,1)$). Also, for small $s,t$ there may not be such sets.
Some informal thoughts: My intuition is that for large enough $s,t$, we can force a lot of repetitions of vectors within the sets, and then we can ``tailor'' equal sums. This is somewhat akin to a vector version of Erdős-Ginzburg-Ziv (or the Van Emde Boas - Kruyswijk variation, which looks at vectors), but instead of looking at the finite abelian group, I have the sum above to bound the behaviour.
Also, I don't really care about tight bounds for $s,t$. They can be as large as needed (e.g., exponential, or even double exponential in $|S|,n$ is fine).