1
$\begingroup$

I am investigating a particular map from a product of three-spheres to the moduli space of (non-negative, real edge weight) networks. The map in question is

$$f: \smash{\left( \mathbb{S}^3 \right)}^N \rightarrow \mathbb{R}_{\ge 0}^N$$

where $N=\binom{n}{2}$ is the number of edges of a complete graph. We are placing a qubit on each edge of the underlying graph, given by $g: \mathbb{R}_{\ge 0}^N \rightarrow \{0,1\}^N$, a semi-ring morphism.

A qubit lives in the Bloch sphere, $\mathbb{S}^3$. Define $e_{ij}:=e_{ij}^0 \lvert0\rangle + e_{ij}^1 \rvert1\rangle$, with $\lvert e_{ij}^0\rvert^2 + \lvert e_{ij}^1\rvert^2 = 1$, and $e_{ij}^k \in \mathbb{C}$. $f$ is given by applying $e_{ij} \rightarrow \lvert e_{ij}^1\rvert^2$ for $1 \le i < j \le n$.

To get unlabeled networks, quotient by the permutation group $\Sigma_n$ where $\sigma \cdot e_{ij} = e_{\sigma(i)\sigma(j)}$.

Let $Y_n=(\mathbb{S}^3)^N / \Sigma_n$ and $X_N=(\mathbb{S}^3)^N/\Sigma_N$. What is $H^*(Y_n;\mathbb{Z})$?

$\endgroup$
15
  • 2
    $\begingroup$ The fiber over a generic point is a product of tori, while over a point with some of the coordinates $1$ (or $0$) the corresponding tori are replaced with circles, so it is not a fibration. I think. $\endgroup$ Commented Sep 2, 2023 at 16:19
  • 5
    $\begingroup$ Let’s be clear. ChatGPT does not, and cannot, “compute” anything at all. That’s not what it does. $\endgroup$
    – HJRW
    Commented Sep 5, 2023 at 14:29
  • 3
    $\begingroup$ @JacksonWalters The process by which I would write a paragraph like your ChatGPT examples is in fact substantially different from the process that I use to write a normal sentence (such as this one). If I followed the process of simply writing down the most likely next token, then with high probability my statements would be false. $\endgroup$ Commented Sep 5, 2023 at 15:03
  • 5
    $\begingroup$ @JacksonWalters I asked ChatGPT to compute the integral homology groups of the free group $F_2$. For $n=0, 1, 2, 3$ it gave me $\mathbf{Z}, \mathbf{Z}^2, \mathbf{Z}/2\mathbf{Z}, \mathbf{Z}$, and for $n=4$ it gave me $S_3$ (!). It's total nonsense. Saying that it is "with high likelihood wrong" is a massive understatement. The fact that ChatGPT is "writing clearly" is not a redeeming factor by any stretch of the imagination. $\endgroup$ Commented Sep 6, 2023 at 2:09
  • 3
    $\begingroup$ Today I also learnt from ChatGPT that "$H_8(F_2; \mathbf{Z}) \cong \zeta(3)$ (the Riemann zeta function at 3). This is the exceptional group, and it's a nontrivial result in homotopy theory related to the so-called 'Kummer congruences.' It's a manifestation of a more complex and less straightforward structure in the homology." How's that for clear writing? $\endgroup$ Commented Sep 6, 2023 at 2:11

1 Answer 1

3
+50
$\begingroup$

$H^*(Y_n;\mathbb{Q}) \cong (H^*((S^3)^N;\mathbb{Q}))^{\Sigma_n}$.

Hence with rational coefficients this reduces to analyzing the invariant subspace of the action of $\Sigma_n$ on $H^*((S^3)^N;\mathbb{Q})\cong (\mathbb{Q}^2)^{\otimes N}$.

Let $H^*(S^3;\mathbb{Q})$ be generated by $s^0 \in H^0(S^3)$ and $s^3\in H^3(S^3)$. We have a projection $p_i: (S^3)^N \to S^3$ to the $i$th factor. Then $H^*((S^3)^N)$ has a basis $p_1^*(s^{i_1})\cup \cdots \cup p_N^*(s^{i_N})$, where $i_j\in \{0,3\}$. We may identify $H^*((S^3)^N)$ with $H^*(S^3)^{\otimes N}$ via the Kunneth formula with basis $s^{i_1}\otimes\cdots\otimes s^{i_N}$ (note that this is a graded commutative tensor product of graded commutative algebras).

For a vector space $V$ over $\mathbb{Q}$ and a finite group $G$ acting on $V$, we have a retract $av: V\to V^G$ (the fixed points of the $G$-action on $V$) given by $av(v)=\frac{1}{|G|}\sum_{g\in G} g(v)$. Hence for any basis of $V$, the image under $av$ will span $V^G$.

Let’s apply this to the $\Sigma_N$ action on $H^*((S^3)^N)\cong H^3(S^3)^{\otimes N}$. For $\sigma \in \Sigma_N$, suppose that $(i_1,\ldots, i_N)=(i_{\sigma(1)},\ldots,i_{\sigma(N)})$. Consider the subset $J=\{ j | i_j=3\}$, then $\sigma$ restricts to a permutation of this set. Let $\epsilon_J(\sigma)$ be the sign of $\sigma_{|J}$. Then $\sigma (s^{i_1}\otimes \cdots\otimes s^{i_N})= s^{i_{\sigma(1)}}\otimes\cdots\otimes s^{i_{\sigma(N)}} = \epsilon_J(\sigma) s^{i_1}\otimes \cdots\otimes s^{i_N}$ due to the rule for changing signs of permutations of graded commutative tensor products. Thus we see that $av(s^{i_1}\otimes \cdots \otimes s^{i_N})$ (with associated $J=\{j|i_j=3\}$) will $= 0$ if there is $\sigma\in \Sigma_N$ such that $\sigma(J)=J$ and $\epsilon_J(\sigma)=-1$, and otherwise will $= \frac{1}{N!} \sum_{\sigma\in\Sigma_N} s^{i_{\sigma(1)}}\otimes \cdots \otimes s^{i_{\sigma(N)}} \neq 0$.

Now lets apply this to $H^*(Y_n;\mathbb{Q})\cong H^*((S^3)^N;\mathbb{Q})^{\Sigma_n} \cong (H^*(S^3)^{\otimes N})^{\Sigma_n}$, $N=\binom{n}{2}$.

We get a generating set for $H^*((S^3)^N)$ corresponding to subgraphs of the complete graph $K_n$ on $n$ vertices where if an edge does not connect $i,j$ we choose the generator $s^0$ of $H^0(S^3)$ corresponding to the $ij$ factor and if an edge connects $i,j$, we choose the generator $s^3$ of $H^3(S^3)$. Then $\Sigma_n$ acts on the vertices and hence on the edges. For a generator corresponding to the graph $\Gamma$, and a permuation $\sigma\in \Sigma_n$ preserving $\Gamma$, we may consider the sign of the action of $\Sigma_n$ on $E(\Gamma)$. If this sign is trivial for all $\sigma\in Aut(\Gamma) < \Sigma_n$, then the averaging operator on the corresponding generator will be non-zero, and we get an element of $H^*((S^3)^N)^{\Sigma_n}-\{0\}$. If the sign is non-trivial, then the averaging operator will be zero. One may also see that the elements will be linearly independent for each isomorphism type of graph, since the corresponding tensor factors will be distinct.

Thus the dimension of the fixed subspace, and hence $H^*(Y_n;\mathbb{Q})$, will be the number of isomorphism classes of simple graphs $\Gamma$ on $n$ vertices such that the sign of the automorphism group $Aut(\Gamma)$ acting on $E(\Gamma)$ is trivial.

$\endgroup$
3
  • $\begingroup$ This is a beautiful answer, thank you. $\endgroup$ Commented Sep 11, 2023 at 14:04
  • 1
    $\begingroup$ @JacksonWalters: a couple of comments: 1. It might be helpful to edit the question to clarify what is being asked. It is confusing since it was edited several times. It took me a couple of reads and looking at the comments to understand what you were asking. 2. I assumed coefficients Q which works essentially because the transfer map is inverse to pullback on the invariant sub space. With positive characteristic coefficients the answer may be more complicated. $\endgroup$
    – Ian Agol
    Commented Sep 11, 2023 at 17:33
  • $\begingroup$ 1) I'll edit the question to clarify what is being asked, which was "what is $H^*((S^3)^N/\Sigma_n;\mathbb{Z})$?". It was confusing because the question evolved from studying "properties" of the fibration to simply asking for the cohomology. 2) $\mathbb{Q}$ coefficients are fine if it simplifies the computation significantly, which it does here. Perhaps the torsion is interesting, I don't know. $\endgroup$ Commented Sep 12, 2023 at 15:29

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .