The Orbit-Stabilizer Theorem

The Orbit-Stabilizer Theorem is the computational backbone of finite group theory: it underlies the proof that conjugacy classes have size dividing $|G|$, which is the starting point of The Class Equation and hence of the existence proofs in Sylow theory. In combinatorics, it is the precise counting statement that Burnside Counting (the "lemma that is not Burnside's") builds upon, used to count distinct colorings of objects (necklaces, dice faces, chemical isomers) up to a symmetry group's action. In crystallography and physical chemistry, orbit-stabilizer reasoning determines the multiplicity of atomic positions within a unit cell under a crystallographic space group, directly informing site-symmetry tables. In computer science, it appears in algorithms for computing group orders and orbit sizes efficiently (e.g. the Schreier–Sims algorithm for permutation groups, central to computational group theory software), where stabilizer chains are built recursively using exactly this orbit/index relationship. In coding theory and cryptography, symmetry groups of codes and of algebraic curves are analyzed via orbit sizes to bound the number of equivalent representations of a structure, and in physics, orbit-stabilizer underlies the classification of particle states under symmetry groups (little groups / stabilizers of momentum vectors in the representation theory of the Poincaré group).

Ask the tutor: Can you show me how Orbit-Stabilizer is used inside the proof of Cauchy's theorem on elements of prime order?

Picture $G$ as a set of moves and $X$ as a stage of positions; acting sends a position to another position by applying a move. Fix one position $x$. As $g$ ranges over all of $G$, the point $g\cdot x$ sweeps out the orbit — everywhere $x$ can be sent. Some moves do nothing to $x$ at all; these form the stabilizer $G_x$, a subgroup because doing nothing then doing nothing is still doing nothing, and undoing a "do nothing" move is itself a "do nothing" move. The key insight is that two moves $g$ and $h$ send $x$ to the same place exactly when they differ by a stabilizing move: $g\cdot x = h\cdot x$ if and only if $h^{-1}g \in G_x$, i.e. $g$ and $h$ lie in the same left coset $hG_x$. So the orbit is not just related to the coset space $G/G_x$ — it is the coset space, relabeled by where each coset sends $x$. This is why counting an orbit reduces to counting cosets, and why $|G| = |\mathrm{Orb}(x)|\cdot|G_x|$ whenever $G$ is finite.

Example 1 (geometric, asymmetric orbit): the dihedral group acting on labeled vertices of a pentagon, with a non-vertex point thrown in for contrast. Let $G = D_5$, the dihedral group of symmetries of a regular pentagon with vertices labeled $1,2,3,4,5$ in cyclic order, so $|G| = 10$ (five rotations $e, r, r^2, r^3, r^4$ and five reflections $s, sr, sr^2, sr^3, sr^4$). Let $X = \{1,2,3,4,5\}$ be the vertex set, with $G$ acting by permuting labels according to the geometric symmetry. Take $x = 1$. Every vertex of a regular pentagon is symmetric to every other (the dihedral group acts transitively on vertices of a regular $n$-gon), so

The stabilizer $G_1$ consists of symmetries fixing vertex $1$. Rotations other than $e$ move every vertex, so no nontrivial rotation fixes $1$. Among reflections, exactly one axis of a regular pentagon passes through vertex $1$ and the midpoint of the opposite edge, and reflecting across it fixes vertex $1$ while swapping the other four vertices in pairs. Hence

where $s_1$ denotes that particular reflection. Orbit-Stabilizer predicts

which matches the direct count. Concretely, the bijection $\varphi$ sends the two-element coset $gG_1$ to the vertex $g\cdot 1$; for instance the coset $\{r, rs_1\}$ maps to $r\cdot 1 = 2$ (the vertex $1$ is rotated to), and one checks $rs_1 \cdot 1 = r\cdot(s_1\cdot 1) = r \cdot 1 = 2$ as well, confirming both coset representatives really do send $1$ to the same place, as the well-definedness argument in the proof guarantees. Now, for contrast, let $Y$ be the set of the five diagonals of the pentagon (segments joining non-adjacent vertices), and let $y_0$ be one fixed diagonal, say the one joining vertices $1$ and $3$. The group still acts on $Y$, again transitively (any diagonal can be rotated/reflected to any other), so $|\mathrm{Orb}(y_0)| = 5$ again, but now the stabilizer is different: the reflection fixing the diagonal $\{1,3\}$ setwise is the reflection through vertex $2$ (which swaps $1 \leftrightarrow 3$ and fixes $2,$ and swaps $4\leftrightarrow5$), together with identity — so again $|G_{y_0}| = 2$ and $10/2 = 5$. This illustrates that different actions of the same group can produce different stabilizers that nonetheless combine with orbit size to reproduce $|G|=10$ every time — the invariant is not the orbit or the stabilizer alone, but their product.

Example 2 (algebraic, asymmetric group and set): $GL_2(\mathbb{F}_3)$ acting on nonzero vectors of $\mathbb{F}_3^2$. Let $\mathbb{F}_3 = \{0,1,2\}$ be the field with three elements, and let $G = GL_2(\mathbb{F}_3)$, the group of invertible $2\times 2$ matrices over $\mathbb{F}_3$. Recall

by the standard formula $|GL_n(\mathbb{F}_q)| = \prod_{i=0}^{n-1}(q^n - q^i)$. Let $X = \mathbb{F}_3^2 \setminus \{0\}$, the eight nonzero vectors, with $G$ acting by matrix-vector multiplication: $A\cdot v = Av$. This is a genuine group action: $I\cdot v = v$, and $(AB)\cdot v = (AB)v = A(Bv) = A\cdot(B\cdot v)$ by associativity of matrix multiplication. Fix $x = \binom{1}{0}$. The orbit of $x$ is all of $X$: given any nonzero $v \in \mathbb{F}_3^2$, extend $\{v\}$ to a basis $\{v, w\}$ of $\mathbb{F}_3^2$ (possible since $v \ne 0$ and $\mathbb{F}_3^2$ is $2$-dimensional), and let $A$ be the matrix with columns $v, w$; then $A$ is invertible (its columns are a basis) and $A\cdot x = Av\cdot$-th column $= v$ — more precisely $A\binom{1}{0} = v$, the first column. So every nonzero vector is reachable, and

By Orbit-Stabilizer, $|G_x| = |G|/|\mathrm{Orb}(x)| = 48/8 = 6$. Let us verify this directly. $G_x$ consists of invertible matrices $A = \begin{pmatrix} a & b \\ c& d\end{pmatrix}$ with $A\binom{1}{0} = \binom{1}{0}$, i.e. $a = 1, c = 0$. So $A = \begin{pmatrix} 1 & b \\ 0 & d \end{pmatrix}$ with $d \ne 0$ (needed for invertibility, since $\det A = 1\cdot d - b\cdot 0 = d$) and $b \in \{0,1,2\}$ arbitrary, $d \in \{1,2\}$ (nonzero elements of $\mathbb{F}_3$). This gives $3 \times 2 = 6$ matrices, exactly matching the prediction $|G_x| = 6$. The bijection $\varphi: G/G_x \to X$ sends the coset of $A$ to the vector $Ax$, i.e. to the first column of $A$ — and indeed two matrices have the same first column exactly when they differ by right-multiplication by an upper-triangular matrix fixing $\binom{1}{0}$, which is precisely membership in the same left coset of $G_x$.