Dihedral Groups

Take a square cut from cardboard, and place it on a table. Then ask a simple question. In how many ways can you pick it up and set it back down? It must occupy exactly the same outline. This innocent puzzle about physical symmetry turns out to encode something important. It represents one of the most important families of groups in all of algebra. The answer is eight ways for a square. In general, there are 2n2n ways for a regular nn-gon. This is not just a count. It is a rich algebraic structure called the dihedral group. It is generated by a single rotation rr and a single reflection ss. These satisfy the compact relation srs=r−1srs = r^{-1}. Dihedral groups matter because they are the smallest natural examples of non-abelian groups. Yet they are completely transparent. Every element, every product, every subgroup can be written down explicitly. Each can be checked by hand. They give the first real test of the abstract group axioms. This test is against genuine geometric intuition. They bridge Binary Operations and the Group Axioms and the concrete permutation viewpoint. That viewpoint comes from Symmetric Groups and Permutation Notation. Every symmetry of a polygon acts on its vertices. Hence it embeds as a permutation. The big picture is that dihedral groups sit at a crossroads. This crossroads joins geometry, combinatorics, and abstract algebra. They are the symmetry groups of the most symmetric planar figures. They appear as subgroups of every symmetric group SnS_n. They are the prototypical example of a group built by combination. This combination joins a cyclic group of rotations with an order-2 "flip." This construction will later be formalized as a semidirect product. Understanding them well also gives a first taste of something else. That is describing a group by generators and relations. This idea blossoms into the theory of Group Presentations and Free Groups. This article gives the precise definition of the dihedral group of order 2n2n. It proves the fundamental relation srs−1=r−1srs^{-1} = r^{-1}. It also proves the resulting normal form for elements. It works through the structure of small and asymmetric examples in complete detail. This includes subgroups, conjugacy, and the failure of commutativity.

Key takeaways
  • The dihedral group DnD_n (order 2n2n) is the symmetry group of a regular nn-gon: nn rotations and nn reflections.
  • It is generated by a rotation rr of order nn and a reflection ss of order 22, with the defining relation srs−1=r−1srs^{-1} = r^{-1}.
  • Every element has a unique normal form rir^i or risr^i s for 0≤i<n0 \le i < n, giving exactly 2n2n elements.
  • DnD_n is non-abelian for n≥3n \ge 3, making it the smallest standard family of non-abelian groups.
  • DnD_n embeds into SnS_n via its action on vertices, connecting geometry to permutation groups.
  • Key idea: a single non-commuting relation, srs=r−1srs = r^{-1}, completely determines the entire multiplication table of 2n2n symmetries.

Where this lives today

Dihedral groups are far more than a pedagogical toy. In chemistry and crystallography, molecules and crystal lattices with nn-fold rotational symmetry and mirror planes have symmetry described exactly by DnD_n (or its three-dimensional analogues), governing selection rules for spectroscopy. In coding theory and cryptography, dihedral groups appear in the study of dihedral codes and in group-based cryptographic schemes exploiting the non-abelian structure and the difficulty of certain algorithmic problems in DnD_n (e.g., the dihedral hidden subgroup problem, relevant to quantum algorithms). In computer graphics and computational geometry, detecting and exploiting polygonal symmetry (for compression, pattern recognition, or mesh simplification) amounts to identifying the dihedral symmetry group of a shape. In physics, dihedral groups classify vibrational modes of symmetric molecules and appear as point groups in the study of crystal symmetry and photonic/phononic band structure. Within pure mathematics, DnD_n is the first worked example when introducing Semidirect Products (since Dn≅Z/nZ⋊Z/2ZD_n \cong \mathbb{Z}/n\mathbb{Z} \rtimes \mathbb{Z}/2\mathbb{Z}), and it is the standard first nontrivial case when learning to build groups via Group Presentations and Free Groups.

Ask the tutor: How does the dihedral group DnD_n arise as a semidirect product of Z/nZ\mathbb{Z}/n\mathbb{Z} and Z/2Z\mathbb{Z}/2\mathbb{Z}?

DefinitionThe Dihedral Group DnD_n

Let n≥3n \ge 3 be an integer, and let PnP_n be a regular nn-gon in the plane, centered at the origin. The dihedral group of order 2n2n, denoted DnD_n, is the group of all rigid motions (isometries) of the plane that map PnP_n onto itself, under composition of functions. Concretely, DnD_n consists of:

  • nn rotations about the center, by angles 2πkn\frac{2\pi k}{n} for k=0,1,…,n−1k = 0, 1, \dots, n-1;
  • nn reflections across the nn axes of symmetry of the polygon (through opposite vertices, opposite edge midpoints, or a mix, depending on parity of nn).

Fixing rr to be the rotation by 2πn\frac{2\pi}{n} (counterclockwise) and ss to be any fixed reflection, one has the presentation

Dn=⟨r,s∣rn=e,  s2=e,  srs−1=r−1⟩,D_n = \langle r, s \mid r^n = e,\; s^2 = e,\; srs^{-1} = r^{-1} \rangle,

meaning: DnD_n is generated by rr and ss, these two relations (together with the group axioms) hold, and every relation among words in r,sr, s is a consequence of them. As a set,

Dn={ e,r,r2,…,rn−1,s,rs,r2s,…,rn−1s },D_n = \{\, e, r, r^2, \dots, r^{n-1}, s, rs, r^2 s, \dots, r^{n-1}s \,\},

a set of exactly 2n2n distinct elements, with composition determined by rn=er^n = e, s2=es^2 = e, and srs=r−1srs = r^{-1} (equivalently sr=r−1ssr = r^{-1}s). (Convention note: some texts write D2nD_{2n} for this same group, indexing by the order rather than by nn. This article uses DnD_n for the symmetries of the nn-gon, of order 2n2n.)

Intuition

Picture a regular pentagon lying on a table, its vertices labeled 1,2,3,4,51,2,3,4,5 in order. A symmetry is any way of picking the pentagon up, moving it in space, and setting it down so the outline lands exactly on itself — vertices may go to different labeled positions, but the shape is preserved. There are two fundamentally different kinds of motions available: you can spin it in the plane, or you can flip it over (equivalently, reflect it through an axis) before setting it down. The rotations alone form a copy of the cyclic group Z/nZ\mathbb{Z}/n\mathbb{Z}: rotating by 2πn\frac{2\pi}{n} a total of nn times returns you to the start. The single reflection ss is its own inverse, since reflecting twice through the same axis is the identity. The subtle and crucial fact — the heart of the whole theory — is what happens when you combine a rotation and a reflection in different orders. Rotating then reflecting is not the same as reflecting then rotating; instead, doing ss, then rr, then ss again has the net effect of undoing the rotation: srs−1=r−1srs^{-1} = r^{-1}. This single equation is the algebraic shadow of the fact that a reflection reverses orientation, turning "clockwise" rotations into "counterclockwise" ones. Because of this relation, every symmetry can be written uniquely as a rotation followed by at most one reflection: the elements rir^i (pure rotations) and risr^i s (a rotation composed with the fixed reflection ss) exhaust the group with no repeats, giving exactly 2n2n elements for the nn-gon.

Figure
The 8 symmetries of a square: rotations rir^i and reflections risr^i s, arranged as a Cayley-diagram hint
Square with vertices 1,2,3,4 — rotation r (90°) and reflection s 1 2 3 4 e (start) 1 2 3 4 r (rotate 90°) 2 1 4 3 s (reflect) D₄ = {e, r, r², r³, s, rs, r²s, r³s}, |D₄| = 8, non-abelian since sr ≠ rs
TheoremStructure and Order of DnD_n

Let n≥3n \ge 3, let DnD_n be the symmetry group of a regular nn-gon as above, let r∈Dnr \in D_n be rotation by 2πn\frac{2\pi}{n}, and let s∈Dns \in D_n be a fixed reflection. Then: 1. rr has order exactly nn, and ss has order exactly 22. 2. srs−1=r−1srs^{-1} = r^{-1} (equivalently sr=r−1ssr = r^{-1}s). 3. Every element of DnD_n can be written uniquely as rir^i or risr^i s with 0≤i≤n−10 \le i \le n-1; consequently ∣Dn∣=2n|D_n| = 2n. 4. DnD_n is non-abelian for every n≥3n \ge 3. 5. ⟨r⟩≅Z/nZ\langle r \rangle \cong \mathbb{Z}/n\mathbb{Z} is a normal subgroup of DnD_n of index 22.

Proof

We work with DnD_n realized geometrically as isometries of the plane fixing a regular nn-gon PnP_n centered at the origin, with vertices v0,…,vn−1v_0, \dots, v_{n-1} placed at angles 2πkn\frac{2\pi k}{n} on a circle. Let rr denote rotation by θ=2πn\theta = \frac{2\pi}{n} about the origin, and let ss denote reflection across the axis through v0v_0 and the origin (through the opposite vertex or edge midpoint, whichever the geometry dictates).

(1) Orders of rr and ss. As a rotation by angle θ\theta, the map rkr^k is rotation by kθ=2πknk\theta = \frac{2\pi k}{n}. This equals the identity rotation if and only if 2πkn\frac{2\pi k}{n} is an integer multiple of 2π2\pi, i.e. n∣kn \mid k. Hence the smallest positive kk with rk=er^k = e is k=nk = n, so rr has order exactly nn. For ss, reflecting across a fixed line twice returns every point to itself, so s2=es^2 = e. Since ss is a reflection (it reverses orientation) it is not the identity (which preserves orientation), so s≠es \ne e and ss has order exactly 22.

(2) The relation srs−1=r−1srs^{-1} = r^{-1}. We verify this by tracking the action on the plane using orientation. A rotation preserves orientation (it can be written as a linear map with determinant +1+1 in suitable coordinates), while a reflection reverses orientation (determinant −1-1). Fix coordinates so the axis of ss is the xx-axis; then

r=(cos⁡θ−sin⁡θsin⁡θcos⁡θ),s=(100−1),r = \begin{pmatrix}\cos\theta & -\sin\theta \\ \sin\theta & \cos\theta\end{pmatrix}, \qquad s = \begin{pmatrix}1 & 0 \\ 0 & -1\end{pmatrix},

as linear maps of R2\mathbb{R}^2 (both fix the origin, the center of the polygon). Direct matrix multiplication gives

srs−1=srs=(100−1)(cos⁡θ−sin⁡θsin⁡θcos⁡θ)(100−1)=(cos⁡θsin⁡θsin⁡θ−cos⁡θ)(100−1).srs^{-1} = srs = \begin{pmatrix}1&0\\0&-1\end{pmatrix}\begin{pmatrix}\cos\theta&-\sin\theta\\\sin\theta&\cos\theta\end{pmatrix}\begin{pmatrix}1&0\\0&-1\end{pmatrix} = \begin{pmatrix}\cos\theta & \sin\theta \\ \sin\theta & -\cos\theta\end{pmatrix}\begin{pmatrix}1&0\\0&-1\end{pmatrix}.

Carrying out this final multiplication:

(cos⁡θsin⁡θsin⁡θ−cos⁡θ)(100−1)=(cos⁡θ−sin⁡θsin⁡θcos⁡θ).\begin{pmatrix}\cos\theta & \sin\theta \\ \sin\theta & -\cos\theta\end{pmatrix}\begin{pmatrix}1&0\\0&-1\end{pmatrix} = \begin{pmatrix}\cos\theta & -\sin\theta \\ \sin\theta & \cos\theta\end{pmatrix}.

Wait — let us redo this multiplication carefully to avoid a sign slip, since precision matters here. We compute srs r first:

sr=(100−1)(cos⁡θ−sin⁡θsin⁡θcos⁡θ)=(cos⁡θ−sin⁡θ−sin⁡θ−cos⁡θ).sr = \begin{pmatrix}1&0\\0&-1\end{pmatrix}\begin{pmatrix}\cos\theta&-\sin\theta\\\sin\theta&\cos\theta\end{pmatrix} = \begin{pmatrix}\cos\theta & -\sin\theta \\ -\sin\theta & -\cos\theta\end{pmatrix}.

Then (sr)s(sr)s:

(sr)s=(cos⁡θ−sin⁡θ−sin⁡θ−cos⁡θ)(100−1)=(cos⁡θsin⁡θ−sin⁡θcos⁡θ).(sr)s = \begin{pmatrix}\cos\theta & -\sin\theta \\ -\sin\theta & -\cos\theta\end{pmatrix}\begin{pmatrix}1&0\\0&-1\end{pmatrix} = \begin{pmatrix}\cos\theta & \sin\theta \\ -\sin\theta & \cos\theta\end{pmatrix}.

This last matrix is exactly the rotation matrix for angle −θ-\theta, i.e. r−1r^{-1}. Since s−1=ss^{-1} = s (as s2=es^2=e), we conclude

srs−1=srs=r−1,srs^{-1} = srs = r^{-1},

which is relation (2). Equivalently, multiplying both sides on the left by ss and using s2=es^2=e: rs−1=sr−1rs^{-1} = sr^{-1}, i.e. sr=r−1ssr = r^{-1}s after taking inverses of both sides appropriately; directly from srs=r−1srs = r^{-1} we get sr=r−1ssr = r^{-1}s by right-multiplying by s−1=ss^{-1}=s.

(3) Normal form and order 2n2n. Existence. We show every element of DnD_n (as an abstract group generated by r,sr,s subject to the relations already verified) can be written as rir^i or risr^i s for some 0≤i≤n−10 \le i \le n-1. Any element of DnD_n is, by definition of "generated by," a finite product of the symbols r,r−1,s,s−1r, r^{-1}, s, s^{-1}. Since s−1=ss^{-1}=s and r−1=rn−1r^{-1} = r^{n-1} (as rn=er^n = e), every element is a word in rr and ss alone. We show by induction on word length that any such word reduces to the form rir^i or risr^i s. A word of length 00 is e=r0e = r^0. Suppose a word ww of length kk has been reduced to rir^i or risr^i s; we extend by one more letter, either rr or ss:

  • If w=riw = r^i and we append rr: get ri+1r^{i+1}, reduce the exponent mod nn using rn=er^n=e; still of the form ri′r^{i'}.
  • If w=riw = r^i and we append ss: get risr^i s, already in the desired form.
  • If w=risw = r^i s and we append rr: get risrr^i s r. Using the relation sr=r−1ssr = r^{-1}s, we have sr=r−1ssr = r^{-1}s, so risr=ri(sr)=ri(r−1s)=ri−1sr^i s r = r^i (sr) = r^i (r^{-1} s) = r^{i-1}s, of the desired form (with exponent reduced mod nn).
  • If w=risw = r^i s and we append ss: get riss=ris2=rir^i s s = r^i s^2 = r^i, of the desired form.

By induction, every word — hence every element of DnD_n — has the form rir^i or risr^i s with 0≤i≤n−10 \le i \le n-1 (reducing exponents mod nn). Uniqueness / distinctness of these 2n2n symbols as geometric transformations. We must check these 2n2n expressions denote 2n2n genuinely distinct isometries, so that no further collapsing occurs. First, r0,r1,…,rn−1r^0, r^1, \dots, r^{n-1} are nn distinct rotations, since ri=rjr^i = r^j with 0≤i<j≤n−10\le i<j\le n-1 would give rj−i=er^{j-i}=e with 0<j−i<n0 < j - i < n, contradicting that rr has order exactly nn (part 1). Second, each risr^i s is orientation-reversing (a composition of an orientation-preserving map rir^i with the orientation-reversing map ss), while each rjr^j is orientation-preserving; hence ris≠rjr^i s \ne r^j for any i,ji,j. Third, the risr^i s are pairwise distinct: if ris=rjsr^i s = r^j s for 0≤i,j≤n−10 \le i,j \le n-1, right-multiplying by s−1=ss^{-1}=s gives ri=rjr^i = r^j, hence i=ji = j as just shown. Therefore the 2n2n elements r0,…,rn−1,r0s,…,rn−1sr^0,\dots,r^{n-1}, r^0s,\dots,r^{n-1}s are pairwise distinct, and by existence they exhaust DnD_n. Hence ∣Dn∣=2n|D_n| = 2n.

(4) Non-commutativity. For n≥3n \ge 3, we have r≠r−1r \ne r^{-1}: indeed r=r−1r = r^{-1} would mean r2=er^2 = e, but rr has order n≥3n \ge 3 so r2≠er^2 \ne e (since 2<n2 < n when n≥3n \ge 3, and the order of rr is exactly nn). From relation (2), sr=r−1s≠rssr = r^{-1}s \ne rs (the last inequality because r−1≠rr^{-1} \ne r, and right-multiplication by ss is injective since ss is invertible). Hence sr≠rssr \ne rs, so DnD_n is non-abelian.

(5) Normality and index of ⟨r⟩\langle r \rangle. Let R=⟨r⟩={e,r,…,rn−1}R = \langle r \rangle = \{e, r, \dots, r^{n-1}\}, which is a subgroup of order nn by part (1) (it is cyclic, generated by an element of order nn, and closed under the group operations since it's a set of powers of rr). By part (3), Dn=R∪RsD_n = R \cup Rs (disjoint union, since Rs={ris:0≤i≤n−1}Rs = \{r^i s : 0\le i \le n-1\} consists exactly of the non-rotation elements), so [Dn:R]=2[D_n : R] = 2. Any subgroup of index 22 in a group GG is automatically normal: for g∈Gg \in G, either g∈Rg \in R, in which case gR=R=RggR = R = Rg trivially, or g∉Rg \notin R, in which case gRgR and RgRg are both equal to the unique non-identity coset (the complement of RR in GG, since there are only two cosets on each side), so gR=RggR = Rg. Hence R⊴DnR \trianglelefteq D_n. This completes the proof of all five parts. ■\blacksquare

∎

Worked examples

Example 1: The multiplication table entry for D5D_5 (the pentagon), computing a specific asymmetric product. Let n=5n=5, so D5D_5 has order 1010, generated by rr (rotation by 72°72°, order 55) and ss (a fixed reflection, order 22), with sr=r−1s=r4ssr = r^{-1}s = r^4 s. Compute the product (r3s)(r2)(r^3 s)(r^2) in normal form. We use the relation repeatedly, moving all rr's to the left of ss:

(r3s)(r2)=r3(sr2).(r^3 s)(r^2) = r^3 (s r^2).

First find sr2sr^2. From sr=r−1s=r4ssr = r^{-1}s = r^4 s, apply this twice:

sr2=(sr)r=(r4s)r=r4(sr)=r4(r4s)=r8s=r8  5s=r3s.sr^2 = (sr)r = (r^4 s) r = r^4 (sr) = r^4 (r^4 s) = r^8 s = r^{8 \bmod 5} s = r^3 s.

Therefore

(r3s)(r2)=r3(r3s)=r6s=r6  5s=rs.(r^3 s)(r^2) = r^3 (r^3 s) = r^6 s = r^{6 \bmod 5} s = r s.

So (r3s)(r2)=rs(r^3 s)(r^2) = rs, a single reflection in normal form, as required by the theorem's uniqueness statement (part 3). As a check, we verify with the general formula derived from the induction step in the proof: for any i,ki,k,

(ris)(rk)=ri(srk)=ri(r−ks)=ri−ks.(r^i s)(r^k) = r^i (s r^k) = r^i (r^{-k} s) = r^{i-k} s.

With i=3,k=2i=3, k=2: r3−2s=r1s=rsr^{3-2}s = r^1 s = rs. This matches, confirming correctness.

Example 2: Finding all subgroups of D4D_4 (the square, order 88) that contain a specific reflection, and checking normality. Let n=4n=4: D4={e,r,r2,r3,s,rs,r2s,r3s}D_4 = \{e, r, r^2, r^3, s, rs, r^2s, r^3s\}, with r4=er^4=e, s2=es^2=e, sr=r−1s=r3ssr=r^{-1}s = r^3 s. We find every subgroup H≤D4H \le D_4 containing the element rsrs, and determine which are normal. First, compute the order of rsrs: (rs)2=(rs)(rs)=r(sr)s=r(r3s)s=r4s2=e⋅e=e(rs)^2 = (rs)(rs) = r(sr)s = r(r^3 s)s = r^4 s^2 = e \cdot e = e. So rsrs has order 22 (it is not the identity, being a reflection).

Smallest subgroup: H1={e,rs}H_1 = \{e, rs\}, order 22. Check it's a subgroup: closed since (rs)2=e∈H1(rs)^2=e \in H_1; contains identity and inverses (rsrs is its own inverse). Valid. Is H1H_1 normal in D4D_4? Conjugate rsrs by rr:

r(rs)r−1=r2sr−1=r2sr3(since r−1=r3).r(rs)r^{-1} = r^2 s r^{-1} = r^2 s r^3 \quad (\text{since } r^{-1}=r^3).

Compute sr3sr^3: using sr=r−1s=r3ssr = r^{-1}s = r^3s repeatedly,

sr2=(sr)r=r3s⋅r=r3(sr)=r3(r3s)=r6s=r2s,sr^2 = (sr)r = r^3 s \cdot r = r^3(sr) = r^3(r^3 s) = r^6 s = r^2 s,
sr3=(sr2)r=r2s⋅r=r2(sr)=r2(r3s)=r5s=rs.sr^3 = (sr^2) r = r^2 s \cdot r = r^2 (sr) = r^2 (r^3 s) = r^5 s = r s.

So r2(sr3)=r2(rs)=r3sr^2 (sr^3) = r^2 (rs) = r^3 s. Thus r(rs)r−1=r3s∉H1r(rs)r^{-1} = r^3 s \notin H_1 (since H1={e,rs}H_1 = \{e, rs\} and r3s≠rs,er^3 s \ne rs, e). Hence H1H_1 is not normal.

A larger subgroup containing rsrs: consider H2={e,r2,s,r2s}H_2 = \{e, r^2, s, r^2 s\} — wait, we want subgroups containing rsrs specifically, and this one doesn't contain rsrs. Instead consider H3={e,r2,rs,r3s}H_3 = \{e, r^2, rs, r^3s\}. Check closure: (rs)(r2)=r(sr2)=r(r2s)=r3s∈H3(rs)(r^2) = r(sr^2)= r(r^2 s) = r^3 s \in H_3 (using sr2=r2ssr^2 = r^2 s computed above, since here sr2sr^2 conjugation-type simplification gives r−2s=r2sr^{-2}s = r^2 s as −2≡2(mod4)-2 \equiv 2 \pmod 4). Also (rs)(r3s)=r(sr3)s=r(rs)s=r2s2=r2∈H3(rs)(r^3 s) = r(sr^3)s = r(rs)s = r^2 s^2 = r^2 \in H_3. And (rs)(rs)=e(rs)(rs) = e as computed. One checks similarly all products of pairs from H3H_3 stay in H3H_3, and inverses: (r3s)−1=r3s(r^3s)^{-1} = r^3 s (order 2, by the same computation pattern as (rs)2=e(rs)^2=e generalized), (r2)−1=r2(r^2)^{-1}=r^2. So H3H_3 is a subgroup of order 44, hence index 22 in D4D_4, hence normal by the index-2 argument in part (5) of the theorem. This illustrates a genuine subtlety: the small subgroup generated by a single reflection is not normal, while a larger subgroup of index 22 containing that same reflection is normal — normality is not "inherited downward," and one must check it subgroup by subgroup.

CounterexampleDropping n≥3n \ge 3: why the pentagon-and-up restriction matters

The theorem requires n≥3n \ge 3 so that PnP_n is a genuine polygon with well-defined rotational and reflective symmetry generating a non-abelian group. Consider what happens if we try n=2n = 2. A "regular 22-gon" would be a degenerate figure — a line segment, or a "polygon" with two vertices. Its symmetry group consists of: the identity, the 180°180° rotation, and (attempting) two reflections. Let rr be rotation by π\pi (order 22, not order n=2n=2... but here order does equal 22, so part (1) still nominally holds), and let ss be a reflection across the segment's own axis. The trouble surfaces in part (4): the claim "DnD_n is non-abelian for n≥3n \ge 3" is precisely calibrated to exclude n=2n=2, because for a segment, rotation by 180°180° commutes with every reflection. Concretely, with r2=er^2 = e, the relation srs−1=r−1srs^{-1}=r^{-1} becomes srs=rsrs = r (since r−1=rr^{-1}=r when r2=er^2=e), i.e. sr=rssr=rs: the group is abelian. In fact the resulting group of order 44 is the Klein four-group Z/2×Z/2\mathbb{Z}/2 \times \mathbb{Z}/2, not a "new" non-abelian dihedral phenomenon — it coincides with its own rotation-reflection decomposition trivially because there's no room for a nontrivial inversion effect (r=r−1r=r^{-1} already). More importantly, the geometric definition breaks down: a "regular 22-gon" is not a polygon in the usual sense (it has no interior, two vertices, and its "edges" coincide), so the entire setup — vertices at angles 2πkn\frac{2\pi k}{n}, a well-defined notion of "the" polygon — becomes degenerate. This shows why the standing hypothesis n≥3n \geq 3 is not a cosmetic convenience but is load-bearing: it is exactly what guarantees r≠r−1r \ne r^{-1}, the key inequality used to prove non-commutativity in part (4) of the theorem.

Scope, edge cases & common traps

SituationWhat to watch for
Indexing conventionSome books write D2nD_{2n} for the order-2n2n group; always check which convention a source uses.
n=3n=3D3≅S3D_3 \cong S_3, the smallest non-abelian group, order 66.
nn even vs. oddFor even nn, reflection axes split into two conjugacy classes (through vertices, through edge midpoints); for odd nn, all reflections are conjugate.
Confusing DnD_n with Z/2nZ\mathbb{Z}/2n\mathbb{Z}Both have order... no — DnD_n has order 2n2n, Z/2nZ\mathbb{Z}/2n\mathbb{Z} also has order 2n2n, but DnD_n is non-abelian for n≥3n\ge3 while Z/2nZ\mathbb{Z}/2n\mathbb{Z} is always abelian: they are not isomorphic for n≥3n \ge 3.
The relation directionsrs−1=r−1srs^{-1}=r^{-1}, not srs−1=rsrs^{-1}=r; mixing these up silently turns DnD_n abelian.
Subgroup vs. normal subgroupAs Example 2 shows, a subgroup generated by a single reflection need not be normal even though the rotation subgroup always is.

The mathematician behind it

Origins · 1 minÉvariste Galois and how this was discovered
ÉG
Évariste Galois
1811–1832 · early 19th-century France, birth of group theory

Dihedral groups do not trace to a single named discoverer in the way a theorem might; they arise naturally from classical geometry going back to antiquity (the symmetries of regular polygons were understood, in effect, by Greek geometers). Yet their formal treatment as groups — abstract sets with an operation satisfying axioms, capable of encoding "the symmetries of a shape" as a purely algebraic object — belongs to the founding era of group theory in the early nineteenth century. Évariste Galois, working on the solvability of polynomial equations, was among the first to treat permutation groups as objects of intrinsic algebraic study, laying conceptual groundwork that later authors (Jordan, Klein, Cayley) used to formalize symmetry groups of geometric figures, including the dihedral groups, within the unified language of abstract group theory.

History
- ~300 BCE: Euclid's Elements catalogues the symmetries of regular polygons geometrically, without group language.
- 1830s: Galois develops permutation groups to analyze solvability of equations by radicals, seeding abstract group theory.
- 1854: Arthur Cayley gives one of the first abstract definitions of a finite group, enabling groups like DnD_n to be studied independent of geometric realization.
- 1870s–1880s: Felix Klein's Erlangen program frames geometry itself via symmetry groups, with dihedral groups as basic examples of finite symmetry groups of the plane.
- Early 20th century: Dihedral groups become standard textbook examples for illustrating non-abelian groups, semidirect products, and generator-relation presentations.
Why is DnD_n defined to have order 2n2n and not nn?

Because the full symmetry group of a regular nn-gon includes both the nn rotational symmetries (forming a cyclic subgroup of order nn) and the nn reflective symmetries (across the nn axes of symmetry). A polygon's shape can be preserved by turning it (rotation) or by flipping it over (reflection), and these are genuinely different operations — a reflection reverses orientation while a rotation preserves it, so no reflection equals any rotation. Counting both families gives n+n=2nn + n = 2n distinct symmetries total, matching the theorem's proof that the 2n2n symbols rir^i and risr^i s are pairwise distinct.

Is DnD_n the same group as Z/2nZ\mathbb{Z}/2n\mathbb{Z} since both have order 2n2n?

No. Having the same order does not imply isomorphism. Z/2nZ\mathbb{Z}/2n\mathbb{Z} is cyclic and therefore abelian, generated by a single element of order 2n2n. For n≥3n \ge 3, DnD_n is non-abelian (proven directly from sr=r−1ssr = r^{-1}s and r≠r−1r \ne r^{-1}), so it cannot be isomorphic to any abelian group, including Z/2nZ\mathbb{Z}/2n\mathbb{Z}. In fact DnD_n has no element of order 2n2n at all when n≥3n \ge 3: every reflection has order 22 and every rotation has order dividing nn.

Why does D3D_3 turn out to be isomorphic to S3S_3?

D3D_3 is the symmetry group of an equilateral triangle, and every symmetry of the triangle permutes its three vertices; conversely, every permutation of three labeled vertices of an equilateral triangle can be realized by some rigid motion (rotation or reflection) of the plane. This gives an injective homomorphism D3→S3D_3 \to S_3, and since both groups have order 66, the map is a bijective homomorphism, hence an isomorphism. This coincidence is special to n=3n=3: for n≥4n \ge 4, DnD_n embeds into SnS_n but is a proper subgroup, since ∣Sn∣=n!|S_n| = n! grows much faster than ∣Dn∣=2n|D_n| = 2n.

How do I quickly compute a product like ris⋅rjsr^i s \cdot r^j s without redoing the whole derivation each time?

Use the general identity derived from the defining relation: for any integers i,ji,j, srj=r−jss r^j = r^{-j} s, so

ris⋅rjs=ri(srj)s=ri(r−js)s=ri−js2=ri−j.r^i s \cdot r^j s = r^i (s r^j) s = r^i (r^{-j} s) s = r^{i-j} s^2 = r^{i-j}.

So the product of two "reflection-type" elements risr^i s and rjsr^j s is always a pure rotation ri−jr^{i-j} (reduce the exponent mod nn). This matches geometric intuition: composing two reflections yields a rotation by twice the angle between their axes.

Are all reflections in DnD_n conjugate to each other?

It depends on the parity of nn. For odd nn, all nn reflections form a single conjugacy class: each reflection axis passes through one vertex and the midpoint of the opposite side, and rotating the polygon maps any such axis to any other. For even nn, the reflections split into two conjugacy classes of size n/2n/2 each: those through pairs of opposite vertices, and those through pairs of opposite edge midpoints; a rotation by rr cannot map a vertex-axis to an edge-axis. This distinction is a frequent source of subtle errors when computing conjugacy classes or normal subgroups of DnD_n for even nn.

Does the abstract presentation ⟨r,s∣rn=e,s2=e,srs−1=r−1⟩\langle r,s \mid r^n=e, s^2=e, srs^{-1}=r^{-1}\rangle always give a group of order exactly 2n2n, or could there be "extra" hidden relations collapsing it further?

The geometric realization (as symmetries of an actual regular nn-gon) shows a group of order exactly 2n2n satisfying these relations exists, which proves the presentation cannot force any further collapse beyond what's already forced by rn=e,s2=e,srs−1=r−1r^n=e, s^2=e, srs^{-1}=r^{-1} — otherwise the geometric group would have fewer than 2n2n elements, contradicting the direct geometric count. This is a standard technique in the theory of Group Presentations and Free Groups: exhibiting a concrete model of the claimed size confirms the abstract presentation is not "over-determined."

Why is the rotation subgroup always normal but a single-reflection subgroup usually isn't?

The rotation subgroup ⟨r⟩\langle r \rangle has index 22 in DnD_n, and every index-22 subgroup is automatically normal (there are only two cosets, and left cosets equal right cosets by elimination). A subgroup generated by one reflection, like {e,s}\{e, s\}, has index nn in DnD_n; nothing forces it to be normal, and indeed conjugating ss by rr typically produces a different reflection rsr−1=r2s≠sr s r^{-1} = r^2 s \ne s (for n>2n>2), showing {e,s}\{e,s\} is not normal in general, exactly as illustrated in the worked subgroup example.

Before this

Where this leads