25
$\begingroup$

Define the functor $x^3 + 1 \colon \mathbf{Set} \rightarrow \mathbf{Set}$ by sending maps $f \colon X \rightarrow Y$ to maps $$f^3 + 1 \colon X^3 \sqcup \{*\} \rightarrow Y^3 \sqcup \{*\}$$ via $$(f^3 + 1)(x_1, x_2, x_3) = (f(x_1), f(x_2), f(x_3))$$ and $$(f^3 + 1)(*) = *$$ and I'm aware there's a more abstract way to define it, but this has the benefit of writing out the functor explicitly.

I am curious if $x^3 + 1$ is naturally isomorphic to the product of two functors, neither of which are naturally isomorphic to the constant functor to a singleton. I chose $x^3 + 1$ in particular because it factors into $x + 1$ and $x^2 - x + 1$ as an integer polynomial, but it is irreducible as a non-negative integer polynomial. I'm pessimistic that such a $x^2 - x + 1$ functor exists, but I'm interested if one could prove it doesn't exist. I tried to construct one via "map $X$ to the set of injective maps from $2$, plus an additional point", but I was able to prove that its product with the $x + 1$ functor isn't naturally isomorphic to $x^3 + 1$. It was pretty close to working though, I got a bijection that was natural only with injections.

$\endgroup$
3
  • 4
    $\begingroup$ I can show using the representation theory of $S_n$ that the only possibility for this $x^2-x+1$ functor is the one you mentioned; how did you show that the product isn't naturally isomorphic to $x^3+1$? $\endgroup$ Commented Apr 14, 2025 at 22:02
  • $\begingroup$ @dhy the issue occurs on the n -> {0} constant maps. One path goes (x, y, z) -> (0, 0, 0) -> (*, 0), and the other path goes (x, y, z) -> nattrans(x, y, z) -> ..., so every tuple (x, y, z) must be sent to a non-star element of $n + 1$. (One can show that nattrans(0, 0, 0) must go to (*, 0).) $\endgroup$ Commented Apr 14, 2025 at 22:52
  • $\begingroup$ That being said, it would be interesting if a result could be found for a different category. The bijection I did find is natural in injections specifically because I just need x = y iff f(x) = f(y) for the casework. $\endgroup$ Commented Apr 14, 2025 at 23:00

1 Answer 1

17
$\begingroup$

Assume that $x^3+1$ factored as a product of two nontrivial functors $f(x)$ and $g(x)$. Then if we took $X=[n]:=\{1,2,\cdots,n\},$ we would need the $S_n$-set $[n]^3 \sqcup \{*\}$ to be a product of two $S_n$-sets $f([n])$ and $g([n])$. Looking at $S_n$-fixed points, we see that for $n\geq 3$ $f([n])$ (resp. $g([n])$) must have a unique $S_n$-fixed point, which we will denote $*_f$ (resp. $*_g$).

If $f([n])\neq\{*_f\},$ then $f([n])$ must have size at least $n$ (by the classification of small $S_n$-sets used below), and so $g([n])$ must have size at most $n^2$. Take $n$ sufficiently large. Then, there is a classification of $S_n$-sets with at most $n^2$ elements (see https://math.stackexchange.com/questions/3536888/large-subgroups-of-symmetric-group). In particular, each $S_n$-orbit with at most $n^2$ elements has an $A_{n-2}$ fixed point, and $[n]^3+1$ has $9$ $A_{n-2}$ fixed points, so $f([n])$ and $g([n])$ have at most $9$ $S_n$-orbits each. It follows that, for a fixed $n$, the size of $f([n])$ must be the value at $n$ of one of a finite number of polynomials. Therefore, for a fixed (still sufficiently large $n$), $f([n])$ and $g([n])$ either have sizes $1$ and $n^3+1$ or $n+1$ and $n^2-n+1$.

Assume that for all sufficiently large $n$, $f([n])$ and $g([n])$ are nontrivial. Fix a value of $n$ and assume WLOG that $f([n])$ has $n+1$ elements. By the classification above, it must be isomorphic as a $S_n$-set to the union of $*_f$ and $S_n/S_{n-1}.$ Similarly, using that $g([n])$ has at most $9$ $S_n$-orbits, the classification tells us that $g([n])$ must be a union of $*_g$ and orbits of the form $S_n/G$, where $G$ is $S_{n-2}$, $S_{n-2}\times S_2$,$A_{n-2}\times S_2$, or the index $2$ subgroup $H:=\{\sigma,\operatorname{sign}(\sigma)\}\subseteq S_{n-2}\times S_2.$ Taking invariants with respect to $A_{n-2}\times S_2$ rules out the middle two cases and taking invariants with respect to $H$ rules out the last case, so we must have $g([n])\cong{*_g}\cup S_n/S_{n-2}$ as a $S_n$-set.

Now consider the map $[n]\rightarrow[n-1]$ which sends $i$ to $i$ for $i\leq n-1$ and sends $n$ to $n-1.$ If $f([n-1])$ had size $(n-1)^2-(n-1)+1$, then the map $f([n])\rightarrow f([n-1])$ could not be surjective, which contradicts the observation that $[n]^3+1\rightarrow[n-1]^3+1$ is surjective. So we can assume $f([n-1])$ has size $n$ (and so the previous paragraph applies to $f([n-1])$ and $g([n-1])$). There are three $S_{n-2}$-invariant points of $g([n])$, and the map $g([n])\rightarrow g([n-1])$ must send them all to $*_g$ (the only $S_{n-2}$-invariant point of $g([n-1])$). But that would imply that the fiber of $*$ along $[n]^3+1\rightarrow[n-1]^3+1$ has size at least $3$, which is not true (it consists only of $*$).

We have shown that we can find arbitrary large $n$ such that $f([n])$ or $g([n])$ is trivial. WLOG, assume that there exists arbitrary large $n$ with $f([n])$ trivial. Then for any $[m]$, if we choose an injection $[m]\rightarrow[n]$ with $f([n])$ trivial, the map $f([m])\rightarrow f([n])$ must be an injection, so $f([m])$ must be trivial, as desired.

$\endgroup$
1
  • 5
    $\begingroup$ I suppose the original question was about all sets, not just finite ones. But if I am not mistaken, then $f$ and $g$ have to preserve filtered colimits since $f \times g$ does, giving the result for arbitrary sets as well. $\endgroup$ Commented Apr 15, 2025 at 9:53

You must log in to answer this question.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.