MIT 18.701 — Lecture 11

Point Groups, Lattices, Crystallographic Restriction, Group Actions

Last time, we classified all discrete subgroups of O2(R)O_2(\mathbb{R}). Our goal for today is to answer the following question:

Question. What do all the discrete symmetry groups GM2G \subseteq M_2 of a shape SR2S \subseteq \mathbb{R}^2 look like?

We'll set up today's discussion with the following theorem and definition.

Theorem. (Discrete Subgroups of R2\mathbb{R}^2) Every discrete subgroup of (R2,+)(\mathbb{R}^2, +) is either of the form Zv\mathbb{Z}v or Zv+Zw\mathbb{Z}v + \mathbb{Z}w, where vv and ww are linearly independent vectors in R2\mathbb{R}^2.

Proof: The proof is analysis-flavored, so we only provide a sketch here.

For any discrete subgroup H(R2,+)H \subseteq (\mathbb{R}^2, +), we may take vv to be the nonzero vector of least magnitude in HH, then take ww to be the vector of least magnitude in HZvH \setminus \mathbb{Z}v.

If there is still some xH(Zv+Zw)x \in H \setminus (\mathbb{Z}v + \mathbb{Z}w), then look for the “remainder” when xx is quotiented out by Zv+Zw\mathbb{Z}v + \mathbb{Z}w to contradict the minimality of v|v| and w|w|.   \blacksquare

Definition (Lattice). A subset of R2\mathbb{R}^2 is a lattice if it is of the form Zv\mathbb{Z}v or Zv+Zw\mathbb{Z}v + \mathbb{Z}w for linearly independent vv and ww.

§ Point Groups and Lattices

Definition (Point Group). Consider the map π:M2O2(R)\pi: M_2 \to O_2(\mathbb{R}) defined by π:(xAx+b)A\pi: (x \mapsto Ax + b) \mapsto A. Intuitively, π\pi filters out the translations from the isometries in M2M_2. Then for any discrete subgroup GG of M2M_2:

Why is L:=ker(π)GL := \ker(\pi) \cap G a lattice? Because an isometry f(x)=Ax+bf(x) = Ax + b is in ker(π)\ker(\pi) if and only if ff is a translation. So LL is a discrete group of translations of (R2,+)(\mathbb{R}^2, +), so it's a lattice by the “Discrete Subgroups of R2\mathbb{R}^2” theorem.

Example. The tilings of R2\mathbb{R}^2 shown below are the square tiling, the pythagorean tiling, and the weaving tiling.

The point groups of their symmetry groups are the following:

Remark. Importantly, the point group of a tiling is not the same as the group of isometries that fixes a single point. See the weaving tiling for an example of this.

§ Classifying Discrete Symmetry Groups: Mathematical Reasoning

Any discrete symmetry group GM2G \subseteq M_2 is built out of its point group G\overline{G} and its lattice LL. What can we say about the relationship between G\overline{G} and LL?

Theorem. (Point Groups are a Symmetry of the Lattice) Given a discrete subgroup GM2G \subseteq M_2, consider its corresponding point group G=π(G)\overline{G} = \pi(G) and lattice L=ker(π)GL = \ker(\pi) \cap G.

Then for any AGA \in \overline{G} and any x0Lx_0 \in L, we have Ax0LAx_0 \in L. In other words, G(L)=L\overline{G}(L) = L.

Proof: Since AGA \in \overline{G} and x0Lx_0 \in L, we know the following about GG:

The key idea is to consider the conjugation of f2f_2 by f1f_1, which is in GG.

f1f2f11:x  A(A1xA1b+x0)+b= x+Ax0.\begin{align*}f_1f_2f_1^{-1}: x \ \mapsto \ & A(A^{-1}x - A^{-1}b + x_0) + b \\ = \ & x + Ax_0.\end{align*}

Thus, xx+Ax0x \mapsto x + Ax_0 is in GG, so Ax0Ax_0 is in LL, as desired.   \blacksquare

What kind of point groups GO2(R)\overline{G} \subseteq O_2(\mathbb{R}) can preserve a lattice LL, then?

Theorem. (Crystallographic Restriction) Consider some point group GO2(R)\overline{G} \subseteq O_2(\mathbb{R}), which must look like CnC_n or DnD_n for some nn. Then if LL is a nontrivial lattice preserved by G\overline{G}, it must in fact be the case that n{1,2,3,4,6}n \in \{1, 2, 3, 4, 6\}.

Proof: The key idea is to consider the nonzero vector xx of minimal length in LL.

So the above implies n{1,2,3,4,6}n \in \{1, 2, 3, 4, 6\}, and we're done.   \blacksquare

Thus, our answer to the big question for today is the following:

Answer. If GM2G \subseteq M_2 is a discrete symmetry group with a nontrivial lattice, then G=π(G)\overline{G} = \pi(G)
must look like CnC_n or DnD_n for some n{1,2,3,4,6}n \in \{1, 2, 3, 4, 6\}.

The original question asked for all the discrete symmetry groups of some shape SR2S \subseteq \mathbb{R}^2. But our analysis so far has completely ignored the shape SS, so what remains is just shape-inspection for each candidate point group π(G)\pi(G).

This shape-inspection, unfortunately, is mostly tedious. So we'll just skip to the results.

§ Classifying Discrete Symmetry Groups: The Results

We categorize our results based on the rank of our lattice.

Example (Rank 0 Lattices). If the lattice LL has rank zero, then L={0}L = \{0\}, and every single point group CnC_n or DnD_n works (even if n∉{1,2,3,4,6}n \not \in \{1, 2, 3, 4, 6\}). This is uninteresting.

Example (Rank 1 Lattices: Frieze Groups). Consider the lattice L=ZvL = \mathbb{Z}v. Then it turns out G{C1,D1,C2,D2}\overline{G} \in \{C_1, D_1, C_2, D_2\}, achievable in 77 different ways. These are called Frieze Groups.

Example (Rank 2 Lattices: Wallpaper Groups). For L=Zv+ZwL= \mathbb{Z}v + \mathbb{Z}w, there are 1717 different Wallpaper Groups.

And these are all of the discrete symmetry groups of some shape SR2S \subseteq \mathbb{R}^2. The end.

Remark. These are not meant to be memorized, unless you're a crystallographer or something.

§ Group Actions and Cayley's Theorem

Now for something completely different.

Definition (Action). Let GG be a group and SS be a set. Then an action of GG on SS is a map G×SSG \times S \to S satisfying the following properties:

For ease of writing, we will often denote the image of (g,s)(g, s) by gsg \cdot s.

Example. Some common examples of group actions are:

Importantly, the third point in the example above provides the intuition behind the following theorem.

Theorem. (Cayley's Theorem) Every group GG is isomorphic to a subgroup of SnS_n, where n=Gn = |G|.

Proof: Think of GG as a group action on the set of its own elements. Expressed more formally, there exists a map f:GPerm(G)f: G \to \mathrm{Perm}(G), where f(g)f(g) is the map f(g):sgsf(g): s \mapsto gs for all sGs \in G.

Intuitively, ff is an injective homomorphism. Here's the formal mathematical jargon that justifies this intuition.

Thus, H=f(G)SnH = f(G) \subseteq S_n is a subgroup of SnS_n for which f:GHf: G \to H is an isomorphism, meaning GHG \cong H and win.   \blacksquare

Remark. This isomorphism from groups GG to subgroups of SnS_n is very “space-inefficient”. For example, the encoding of S8S_8 as a group action on its own elements would yield a representation of S8S_8 as a subgroup of S40320S_{40320}.