7

The symmetry group of a given $n$-ary truth function $f:\{0,1\}^n\to\{0,1\}$ is the permutation group consisting of all the $f$-preserving permutations of the $n$ input variable names: $\{\sigma\in S_n:f^\sigma=f\}$, where $f^\sigma(x_1,\dots,x_n):=f(x_{\sigma(1)},\dots,x_{\sigma(n)})$. For example, the truth function $f(x_1,x_2,x_3)=x_1\land(x_2\lor x_3)$ has symmetry group generated by the transposition $(23)$.

Does every subgroup of $S_n$ occur in this way?

Note that the permutation structure, not just the group structure, matters here. For example, $\langle(1234)\rangle$ and $\langle(1234)(56)\rangle$ are isomorphic as groups, but $\langle(1234)\rangle$ ought to be the symmetry group of something like

$$f(w,x,y,z)=g(w\to x,x\to y,y\to z,z\to w)$$

for some highly-symmetric $g$ (and possibly a higher-arity asymmetric operation in place of "$\to$"), whereas I can't think of a truth function with symmetry group generated by $(1234)(56)$.

Karl
  • 12,752
  • 2
    By rewriting "a->b" as "~a or b", I think your function f(u,v,w,x,y,z) just becomes the constant "true" function. So its symmetry group would be S_6. – pb1729 Sep 21 '24 at 07:05
  • @RickyTensor oops, thanks! I'm hoping the idea will still work if we use some other asymmetric 2-ary function in place of $\to$ and/or a different symmetric 6-ary function in place of the multi-$\lor$. (What about "exactly 3 of the 6 are true"?) – Karl Sep 21 '24 at 11:02

1 Answers1

3

Edit: The previous argument that was here was incorrect, but I don't think so. Boolean functions $f : \{ 0, 1 \}^n \to \{ 0, 1 \}$ are in bijection with subsets of $\{ 0, 1 \}^n$ (namely the subset $f^{-1}(1)$), or equivalently with sets of subsets of an $n$-element set $[n]$, and this bijection preserves symmetries; that is, the symmetries of $f$ are the symmetries of the corresponding set of subsets of $[n]$. Write $F_k \subset {[n] \choose k}$ for the set of subsets in $f^{-1}(1)$ of size $k$. Then the condition that $\sigma \in S_n$ is a symmetry of $f$ is equivalent to the condition that $\sigma F_k = F_k$ (setwise; my mistake previously was confusing this with pointwise) for all $k$.

The choice of collections of subsets $F_k$ amounts to a choice of (undirected) hypergraph structure on $[n]$, so the question is equivalent to asking whether every subgroup of $S_n$ appears as the automorphism group of a hypergraph structure on $[n]$.

I claim that $A_n$ is not the automorphism group of a hypergraph structure for $n \ge 3$. It suffices to observe that $A_n$ acts transitively on the set ${[n] \choose k}$ of subsets of size $k$, for all $1 \le k \le n-1$. (For $1 \le k \le n-2$ this follows from but is weaker than the statement that $A_n$ acts $k$-transitively, which involves ordered tuples. For $k = n-1$ it follows from $1$-transitivity.) It follows that if a hypergraph has automorphism group containing $A_n$ then for each $k$ it contains either all $k$-edges (subsets of size $k$) or none of them, so its automorphism group is $S_n$.

Qiaochu Yuan
  • 468,795
  • 1
    Possibly I'm being silly and misunderstanding something here, but what about the following boolean function of four variables corresponding to vertices of a square?: It returns true iff two neighbouring vertices are true and the rest are false. Intuitively, rotations and reflections should be symmetries, but eg. transposing two neighbouring vertices is not, since the resulting function can be true when two diagonally opposite vertices are true, whereas the original function could never be. – pb1729 Sep 21 '24 at 06:43
  • @Ricky: thanks for the correction. You're right, and in fact we can get all automorphism groups of graphs. I've edited the argument. – Qiaochu Yuan Sep 21 '24 at 18:51
  • Thanks! To build some intuition about this, I wonder if it makes sense to say that the fact that hypergraph edges are unordered sets is what makes hypergraphs "unable to see the difference" between $A_n$ and $S_n$. – Karl Sep 22 '24 at 23:26
  • Also, this post was a helpful reference for $A_n$ being $k$-transitive. – Karl Sep 22 '24 at 23:28