Nature, it seems, is the popular name
for milliards and milliards and milliards
of particles playing their infinite game
of billiards and billiards and billiards.
Solutions must be typeset in LaTeX and written in complete, grammatical sentences. You must list all your collaborators explicitly. Feel free to DM on Discord if you have any questions. Problems marked (Challenge) are optional but encouraged.
4.1Maps Between $G$-sets
We now have a good structural picture of $G$-sets: every $G$-set decomposes into orbits, which are themselves transitive $G$-sets. If $X$ is a transitive $G$-set, choosing a point $x \in X$ gives us an isomorphism $X \cong G/G_x$. Thus every $G$-set looks like a union of coset actions. After mastering a mathematical object, the next step is to study the maps between them.
Problem 4.1
Let $G$ be a finite group, let $X, Y$ be $G$-sets, and let $\vp: X \to Y$ be a $G$-map.
Prove that $\vp$ maps $G$-orbits in $X$ to $G$-orbits in $Y$.
Solution
Let $x \in X$. Since $\vp$ is a $G$-map, we have \[ \vp(Gx) = G\vp(x). \] Thus $\vp$ maps orbits to orbits.
Hence it follows that we can understand all $G$-maps by understanding the $G$-maps between transitive $G$-sets. Let $H, K \subseteq G$ be subgroups and let $g \in G$. Determine the necessary and sufficient conditions for there to exist a $G$-map $\vp : G/H \to G/K$ which sends $H \mapsto gK$.
Solution
If $\vp(H) = gK$ is a $G$-map, then for all $a \in G$ we have \[ \vp(aH) = a\vp(H) = agK. \] So there is at most one such $G$-map. If $h \in H$, then $hH = H$, hence applying $\vp$ we get \[ gK = \vp(H) = \vp(hH) = hgK. \] This is equivalent to $g^{-1}hg \in K$ for all $h \in H$. Hence a necessary condition is that $g^{-1}Hg \subseteq K$. Running this argument backwards, we can see this condition is also sufficient.
When are $G/H$ and $G/K$ isomorphic as $G$-sets?
Solution
The $G$-sets $G/H$ and $G/K$ are isomorphic if and only if $H$ and $K$ are conjugate subgroups. If $g^{-1}Hg = K$, then we also have $H = g K g^{-1}$. Hence the previous problem shows there is a $G$-map $\vp : G/H \to G/K$ determined by $\vp(H) = gK$ and a map $\psi : G/K \to G/H$ determined by $\psi(K) = g^{-1}H$. Thus $\psi\vp : G/H \to G/H$ is the $G$-map determined by \[ \psi\vp(H) = \psi(gK) = g\psi(K) = gg^{-1}H = H. \] The identity map $1 : G/H \to G/H$ also satisfies this, hence $\psi\vp = 1$. The identity $\vp\psi = 1$ follows similarly. Thus $\vp$ and $\psi$ are inverses, implying that $G/H \cong G/K$.
Recall that an automorphism of an object $X$ is an isomorphism $\vp : X \to X$. The collection $\aut(X)$ of all automorphisms of $X$ forms a group with respect to composition. Prove that the automorphism group of $G/H$ is isomorphic to $N_G(H)/H$ where $N_G(H)$ is the normalizer \[ N_G(H) = \{g \in G : g H g^{-1} = H\}. \]
Solution
A $G$-map $\vp : G/H \to G/H$ is determined by where it sends $H$. We know that $H \mapsto g^{-1}H$ is a $G$-map if and only if $gHg^{-1} \subseteq H$. Since $H$ is finite, this implies that $gH g^{-1} = H$. Thus we have a surjective homomorphism $\vp : N_G(H) \to \aut(G/H)$ given by $g \mapsto \vp_g : G/H \to G/H$ where $\vp_{g}$ is determined by $\vp_{g}(H) := g^{-1}H$. Note that the inverse here is important; without it we do not get a homomorphism:
\[\begin{aligned} \varphi_{g_1}\varphi_{g_2}(H) &= \varphi_{g_1}(g_2^{-1}H)\\ &= g_2^{-1}\varphi_{g_1}(H)\\ &= g_2^{-1}g_1^{-1}H\\ &= (g_1g_2)^{-1}H\\ &= \varphi_{g_1g_2}(H). \end{aligned}\]The first isomorphism theorem then implies that $\aut(G/H) \cong N_G(H)/\ker(\vp)$. Observe that $g \in \ker(\vp)$ if and only if $\vp_g$ is the identity automorphism of $G/H$, if and only if $gH = H$, if and only if $g \in H$. Hence $\ker(\vp) = H$. Therefore \[ N_G(H)/H \cong \aut(G/H). \]
4.2Products of $G$-Sets
There are two natural operations we can use to combine $G$-sets to get new $G$-sets: sums and products. We have already discussed how every $G$-set is a sum of transitive $G$-sets. But how do products fit into this? If $X, Y, Z$ are $G$-sets, then \[ X \times (Y + Z) \cong (X \times Y) + (X \times Z). \] (do you see why?) This implies that we can reduce the analysis of products to products of transitive $G$-sets (do you see how this follows?)
Problem 4.2
When we take the product of transitive $G$-sets, the result is typically not transitive anymore. The decomposition of $X \times Y$ tells us the ways in which elements of $X$ and $Y$ can relate to one another, up to symmetry.
For example, let $G = S_6$ and let $[6] = \{1,2,\ldots, 6\}$. (In general we write $[n] := \{1, 2, \ldots, n\}$.) Consider $X = \binom{[6]}{2}$ and $Y = \binom{[6]}{3}$ where $X$ is the set of 2 element subsets of $[6]$ and $Y$ is the set of 3 element subsets of $[6]$. Find representatives for each orbit of $S_6$ on $X \times Y$. Given a pair $(x,y) \in X \times Y$, can you succinctly describe the key feature of $x$ and $y$ which determines which orbit it belongs to?
Solution
The orbits are determined by the size of the overlap between $x$ and $y$. There are thus three orbits represented by
- $(\{1,2\}, \{3,4,5\})$
- $(\{1,2\}, \{2,3,4\})$
- $(\{1,2\}, \{1,2,3\})$.
Next let $G = D_6$ and let $V$ and $E$ be the sets of vertices and edges of a regular hexagon. These are both $D_6$-sets. Draw a picture to illustrate representatives for the orbits of $D_6$ on $V \times E$.
Solution
Fix a vertex $v_0$. Every orbit of $D_6$ on $V \times E$ contains a pair of the form $(v_0, e)$, and two such pairs $(v_0, e)$ and $(v_0, e')$ lie in the same orbit exactly when $e' = ge$ for some $g$ in the stabilizer of $v_0$. That stabilizer is $\{1, s\}$ where $s$ is the reflection across the axis through $v_0$ and $v_3$, and $s$ pairs up the six edges into three pairs. Hence there are three orbits, determined by how far the edge is from the vertex: the edge contains the vertex, the edge is one step away, or the edge is two steps away (adjacent to the opposite vertex).
Since $|V \times E| = 36$ and $D_6$ has order $12$, each of the three orbits has exactly $12$ elements, so the stabilizer of each pair $(v_0, e)$ is trivial.
Problem 4.3
We know that every transitive $G$-set is isomorphic to $G/H$ for some subgroup $H \subseteq G$. So if $H, K \subseteq G$, how do we decompose the $G$-set $G/H \times G/K$?
Prove that every orbit in $G/H \times G/K$ contains an element of the form $(H,gK)$ for some $g \in G$.
Solution
Every element of $G/H \times G/K$ looks like $(g_1H, g_2K)$ for some $g_1, g_2 \in G$. Multiplying by $g_1^{-1}$ gives us \[ g_1^{-1}(g_1 H, g_2 K) = (H, g_1^{-1}g_2K). \] Hence let $g := g_1^{-1}g_2$, then $(H, gK)$ belongs to the orbit of $(g_1H, g_2K)$.
What is the stabilizer of $(H,gK)$?
Solution
Note that $g'$ belongs to the stabilizer of $(H,gK)$ if and only if $(g'H, g'gK) = (H,gK)$, which is equivalent to $g' \in H$ and $g^{-1}g'g \in K$. And this is equivalent to $g' \in H \cap gKg^{-1}$. Hence the stabilizer of $(H, gK)$ is $H \cap gKg^{-1}$.
Given an element $g \in G$ and subgroups $H, K \subseteq G$, the set \[ HgK = \{hgk : h \in H, k\in K\} \] is called a double coset. Let $H\backslash G /K$ denote the set of all double cosets $H g K$. Prove there is a bijection between $H\backslash G /K$ and the orbits of $G$ on $G/H \times G/K$. Be sure to show that your function is well-defined!
Solution
Let $Y := (G/H \times G/K)/G$ denote the set of $G$-orbits on this product space. Let $\psi : H\backslash G /K \to Y$ be defined by $\psi(HgK) = G (H,gK)$. If $Hg_1K = Hg_2K$, then $g_2 = hg_1k$ for some $h \in H, k \in K$. Then
Hence $G (H, g_1K) = G(H,g_2K)$, which shows that $\psi$ is well-defined.
Part 1 shows that $\psi$ is surjective, hence it remains to show that $\psi$ is injective. Suppose that $G(H,g_1K) = G(H,g_2K)$. Then there exists some $g \in G$ such that \[ (H, g_2K) = g(H,g_1K) = (gH,gg_1K). \] Comparing first coordinates we see that $g \in H$ and that $g_2K = gg_1K$. Thus the coset equality criteria implies that $g_2 = gg_1k$ for some $k \in K$. Hence \[ Hg_2K = Hgg_1kK = Hg_1K. \] Thus $\psi$ is a bijection.
Thus double cosets $H\backslash G /K$ precisely capture the $G$-orbits of $G/H \times G/K$, hence determine the decomposition of this product. Hence if $X$ and $Y$ are transitive $G$-sets, then given $(x,y) \in X\times Y$ the double cosets $G_x\backslash G / G_y$ corresponds to the fundamental ways in which elements of the orbits of $x$ and $y$ relate to one another up to $G$-symmetry.
4.3Burnside Semiring
The operations of addition and multiplication of $G$-sets behave sort of like the usual operations of addition and multiplication in a ring. One difference is that identities like \[ X\times (Y + Z) \cong (X \times Y) + (X \times Z) \] only hold up to isomorphism, not equality. Hence we can't exactly make a ring out of $G$-sets. Another issue is that negatives don't make sense. But we can get around both these issues in the following way:
Definition 4.1
Let $G$ be a group. If $X$ is a $G$-set, let $[X]$ denote the isomorphism class of $X$. Let $B(G)$ denote the set of isomorphism classes of $G$-sets. We can define addition and multiplication operations on $B(G)$ by
Proposition 4.2
The set $B(G)$ forms a semiring1 with respect to the operations of addition and multiplication defined above. We call $B(G)$ the Burnside semiring of $G$.
I won't ask you to prove this since it's tedious to verify all the axioms. But let's think about this construction.
Problem 4.4
What are the 0 and 1 elements in $B(G)$? Construct $G$-sets that represent them and explain why they work as additive and multiplicative identities.
Solution
The empty set $\emptyset$ represents the 0 element of $B(G)$ and the singleton set $\{*\}$ with the trivial $G$-action represents the 1 element. The empty set is clearly an additive identity with respect to union, and the singleton is an identity with respect to products. Finally note that $X \times \emptyset = \emptyset$ for all sets $X$, which shows that $[X]0 = 0$.
The isomorphism classes $[G/H]$ corresponding to each conjugacy class of subgroups $H$ form an additive basis for $B(G)$. If $G = S_3$, there are 4 conjugacy classes of subgroups represented by $H = \langle \ve\rangle$, $\langle (12)\rangle$, $\langle (123)\rangle$, and $S_3$ itself. Write \[ 1 = [S_3/S_3], \quad T_2 = [S_3/\langle (123)\rangle], \quad T_3 = [S_3/\langle (12)\rangle], \quad T_6 = [S_3/\langle \ve\rangle], \] so that $T_n$ is the class of a transitive $S_3$-set with $n$ elements, and $1$ is the multiplicative identity from the previous part. Make a multiplication table which shows the product of each pair of elements in our additive basis $1, T_2, T_3, T_6$.
Solution
Concretely, $1$ is a point, $T_2$ is $S_3$ acting on $\{\pm 1\}$ by sign, $T_3$ is the natural action on $\{1,2,3\}$, and $T_6$ is $S_3$ acting on itself by left multiplication. Our previous problem tells us how to decompose a product: the orbits of $S_3/H \times S_3/K$ correspond to double cosets $HgK \in H\backslash S_3/K$, and the orbit of $(H, gK)$ is isomorphic to $S_3/(H \cap gKg^{-1})$. Since $1$ is the multiplicative identity, we only need to compute the products of $T_2$, $T_3$, and $T_6$.
- Products with $T_6$. If $H = \langle \ve \rangle$, then $H \cap gKg^{-1}$ is trivial for every $K$ and $g$, so every orbit of $S_3 \times S_3/K$ is isomorphic to $S_3$ itself. If $S_3/K$ has $n$ elements, then counting elements gives $6n = 6\cdot(\text{number of orbits})$, so there are $n$ orbits and $T_6T_n = nT_6$. Hence $T_6 T_2 = 2T_6$, $T_6T_3 = 3T_6$, and $T_6^2 = 6T_6$.
- $T_2^2$. The subgroup $A_3 = \langle (123) \rangle$ is normal, so $A_3 \cap gA_3g^{-1} = A_3$ for every $g$ and every orbit of $S_3/A_3 \times S_3/A_3$ is isomorphic to $S_3/A_3$. There are $4/2 = 2$ orbits, so $T_2^2 = 2T_2$. (Concretely, the orbits of $S_3$ on $\{\pm 1\}^2$ are $\{(1,1), (-1,-1)\}$ and $\{(1,-1),(-1,1)\}$.)
- $T_2T_3$. The intersection $\langle (123) \rangle \cap g\langle (12) \rangle g^{-1}$ is trivial by Lagrange's theorem, since the subgroups have orders $3$ and $2$. Thus every orbit of $\{\pm 1\} \times \{1,2,3\}$ is isomorphic to $S_3$, and since there are only $6$ elements, there is a single orbit: $T_2T_3 = T_6$.
- $T_3^2$. The action of $S_3$ on $\{1,2,3\}^2$ has two orbits: the diagonal $\{(i,i)\}$, which is isomorphic to $\{1,2,3\}$, and the six pairs $(i,j)$ with $i \neq j$, on which $S_3$ acts freely (a permutation fixing $i$ and $j$ also fixes the third point). Thus $T_3^2 = T_3 + T_6$. In terms of double cosets, if $H = \langle (12) \rangle$ then $H \backslash S_3 / H = \{H, H(13)H\}$ with $H \cap (13)H(13)^{-1} = \langle (12) \rangle \cap \langle (23) \rangle = \langle \ve \rangle$.
Putting this together, the multiplication table for $B(S_3)$ is: \[ \renewcommand{\arraystretch}{1.3} \begin{array}{c|cccc} \cdot & 1 & T_2 & T_3 & T_6 \\ \hline 1 & 1 & T_2 & T_3 & T_6 \\ T_2 & T_2 & 2T_2 & T_6 & 2T_6 \\ T_3 & T_3 & T_6 & T_3 + T_6 & 3T_6 \\ T_6 & T_6 & 2T_6 & 3T_6 & 6T_6 \end{array} \]
Let's put this table to work. If we have a set of combinatorial structures built on the labels $[3] = \{1,2,3\}$ (subsets, graphs, orderings, and so on), then $S_3$ acts on those structures by relabeling, and two structures are “the same up to relabeling” precisely when they lie in the same orbit. Products of such $S_3$-sets arise whenever we consider two structures on the same set of labels at once.
Problem 4.5
Let $\P$ be the set of all subsets of $[3]$ and let $\G$ be the set of all simple graphs2 with vertex set $[3]$ (so $\G$ has $2^3 = 8$ elements, one for each subset of the three possible edges). Both are $S_3$-sets via the relabeling action. Use the notation $T_2, T_3, T_6$ from the previous problem.
Express $[\P]$ as a sum of basis elements of $B(S_3)$.
Solution
The orbit of a subset $S \subseteq [3]$ is determined by $|S|$, since $S_3$ can move any $k$-element subset to any other. The subsets $\emptyset$ and $[3]$ are fixed by all of $S_3$, so each forms an orbit isomorphic to a point. The three $1$-element subsets form one orbit, and the stabilizer of $\{1\}$ is $\langle (23)\rangle$, which is conjugate to $\langle (12) \rangle$, so this orbit is isomorphic to $S_3/\langle (12) \rangle$. The same holds for the three $2$-element subsets (take complements). Therefore \[ [\P] = 1 + T_3 + T_3 + 1 = 2 + 2T_3. \]
Express $[\G]$ as a sum of basis elements of $B(S_3)$.
Solution
Up to relabeling there are four graphs on $[3]$, determined by the number of edges: the empty graph, a single edge, a path with two edges, and the triangle. The empty graph and the triangle are fixed by $S_3$, giving two orbits isomorphic to a point. The graph with the single edge $\{1,2\}$ has stabilizer $\langle (12) \rangle$, so the three single-edge graphs form an orbit isomorphic to $S_3/\langle (12) \rangle$. Likewise the path $1 - 2 - 3$ has stabilizer $\langle (13)\rangle$, so the three paths form an orbit isomorphic to $S_3/\langle (12) \rangle$. Therefore \[ [\G] = 2 + 2T_3. \] (Alternatively: a graph on $[3]$ is the same thing as a subset of the three edges, and $S_3$ permutes the edges the same way it permutes the vertices, so $\G \cong \P$ as $S_3$-sets.)
A graph with terminals is a graph on $[3]$ together with a distinguished subset of its vertices (the terminals), that is, an element of $\G \times \P$. Use the multiplication table for $B(S_3)$ to express $[\G \times \P]$ as a sum of basis elements. How many graphs with terminals are there up to relabeling?
Solution
Using the table for $B(S_3)$ and $T_3^2 = T_3 + T_6$,
\[\begin{aligned} [\G \times \P] &= (2 + 2T_3)(2 + 2T_3) = 4 + 8T_3 + 4T_3^2\\ &= 4 + 8T_3 + 4(T_3 + T_6) = 4 + 12T_3 + 4T_6. \end{aligned}\]Each basis element is a single orbit, so there are $4 + 12 + 4 = 20$ graphs with terminals up to relabeling: $4$ fixed by all of $S_3$, $12$ with orbits of size $3$, and $4$ with orbits of size $6$. As a check on the sizes, $4\cdot 1 + 12 \cdot 3 + 4 \cdot 6 = 64 = 8 \cdot 8 = |\G \times \P|$.
As a check, we can list the orbits directly. First choose a graph $\Gamma$ from each of the four orbits of $\G$, and then the orbits of $\G \times \P$ containing $(\Gamma, S)$ correspond to the orbits of the stabilizer $\aut(\Gamma)$ on subsets $S \subseteq [3]$. The size of the orbit of $(\Gamma, S)$ in $\G \times \P$ is $6/|\aut(\Gamma) \cap \mathrm{Stab}(S)|$.
- Empty graph, $\aut(\Gamma) = S_3$: the orbits of subsets are determined by size, giving $4$ orbits. For $S = \emptyset, [3]$ the pair $(\Gamma, S)$ is fixed by $S_3$ (orbit size $1$); for $|S| = 1, 2$ the orbit has size $3$.
- Triangle, $\aut(\Gamma) = S_3$: exactly the same, $2$ orbits of size $1$ and $2$ of size $3$.
- Single edge $\{1,2\}$, $\aut(\Gamma) = \langle (12)\rangle$: the subsets $\emptyset$, $\{3\}$, $\{1,2\}$, $\{1,2,3\}$ are each fixed by $(12)$ and give $4$ orbits of size $3$, while $\{1\} \sim \{2\}$ and $\{1,3\}\sim\{2,3\}$ give $2$ orbits of size $6$.
- Path $1 - 2 - 3$, $\aut(\Gamma) = \langle (13) \rangle$: the same count, $4$ orbits of size $3$ and $2$ orbits of size $6$.
Altogether we find $4$ orbits of size $1$, $4 + 4 + 4 = 12$ orbits of size $3$, and $2 + 2 = 4$ orbits of size $6$, for a total of $20$, exactly as predicted by $4 + 12T_3 + 4T_6$.