Preface
The idea of the book
By an alphabet we mean the set A N = { 0 , 1 , … , N − 1 } \mathcal{A}_N = \{0,1,\ldots,N-1\} A N = { 0 , 1 , … , N − 1 } of the first N N N non-negative integers; it is the model of a finite set of size N N N . Its elements are labels : a finite set X X X with N N N elements is identified with A N \mathcal{A}_N A N through a bijection λ : X → A N \lambda:X\to\mathcal{A}_N λ : X → A N , its labelling . A label is an integer in its own right and should not be confused with a digit. Each label m m m has its own digits, in a base or in a mixed radix adapted to the structure at hand, and it is in these digits that operations on the alphabet are written.
The idea of the book is to describe every finite group G G G as an alphabet endowed with the group operation, that is, as a pair ( A N , ∗ ) (\mathcal{A}_N,*) ( A N , ∗ ) , where N = ∣ G ∣ N=|G| N = ∣ G ∣ is the order of the group and ∗ * ∗ is the product of G G G transported to A N \mathcal{A}_N A N . This is achieved for all finite groups, and it is explained in detail for all the infinite families of finite simple groups—the cyclic groups of prime order, the alternating groups and the groups of Lie type—and for the 26 sporadic groups.
A finite operation on an alphabet is read off the digits of a single integer, its code . The basic tool is the digit function
d N ( k , x ) = ⌊ x N k ⌋ − N ⌊ x N k + 1 ⌋ d_N(k,x)=\left\lfloor\frac{x}{N^k}\right\rfloor-N\left\lfloor\frac{x}{N^{k+1}}\right\rfloor d N ( k , x ) = ⌊ N k x ⌋ − N ⌊ N k + 1 x ⌋
which returns the k k k -th digit of x x x in base N N N . An r r r -ary operation on A N \mathcal{A}_N A N is the integer whose N r N^r N r base-N N N digits are its values, so a magma, a permutation or a whole Cayley table is one number. A group of order N N N becomes a digital law : a labelling λ : G → A N \lambda:G\to\mathcal{A}_N λ : G → A N , its inverse, and the transported product m ∗ n = λ ( λ − 1 ( m ) λ − 1 ( n ) ) m*n=\lambda(\lambda^{-1}(m)\lambda^{-1}(n)) m ∗ n = λ ( λ − 1 ( m ) λ − 1 ( n )) written as a closed formula in the digits of m m m and n n n . For a cyclic group this formula is addition modulo N N N ; for a metacyclic group it is the two-digit rule m ∗ n = d q ( 0 , m + r m 1 n 1 ) + q d s ( 0 , m 1 + n 1 ) m*n=d_q(0,m+r^{m_1}n_1)+q\,d_s(0,m_1+n_1) m ∗ n = d q ( 0 , m + r m 1 n 1 ) + q d s ( 0 , m 1 + n 1 ) ; for a finite simple group it packs the coordinates of a normal form or of a stabiliser chain into mixed-radix digits. Every finite group has such a law, since any bijection transports the product. The point of the book is that the bijection can be chosen so that the law is explicit, short and computable, and that the structural properties of the group—associativity, inverses, conjugacy, normal subgroups, characters—become identities between digits of codes.
The book has grown in three directions, and it is worth saying plainly what each contributes. The first is foundational: the digit function and the B \mathcal{B} B -function, codes of permutations and of r r r -ary operators, and the observation that every finite structure is a single integer. The second is a re-reading of the general theory of finite groups in digital terms: the construction of laws by diagrams, the classical theorems, conjugacy classes, stabiliser chains and characters. Here the abstract statements are kept, and the digital form supplies explicit and checkable instances of them. The third is a catalogue: laws for all groups of order at most 32, for the finite quotients of the plane and space crystallographic groups, for the symmetric and alternating groups, for the families of finite simple groups of Lie type and for the sporadic groups, all in one format, so that they can be compared, transmitted and verified as integer arithmetic. The catalogue is accompanied by worked products in the largest groups and by a software archive that recomputes every law, table and example of the book. Nothing in the book changes what is known about the structure of these groups; the classical background is that of the standard texts [1, 2, 3, 4, 5]. What changes is that the groups become concrete data.
Outline of the chapters
Foundations (Chapters 1 and 2). Chapter 1 introduces the digit function and the digital expansion of integers (Section 1.1), and the B \mathcal{B} B -function, which tests digits and expresses comparisons and carries of positional arithmetic in closed form (Section 1.3). Chapter 2 begins with permutations as codes—composition, inversion, cycles, conjugacy and sign (Section 2.1)—and then encodes arbitrary finite magmas and r r r -ary operators, their homomorphisms and relabellings, and direct products. Everything that follows uses these two chapters.
General theory in digital form (Chapters 3–6). Chapter 3 is a method for producing group laws: the digits of a label are arranged in layers, and a diagram of arrows prescribes how carries couple them. Associativity is built in by construction, and a gallery of recipes recovers many familiar groups, including simple ones. Chapter 4 restates the main theorems—Cayley, Lagrange, group actions, conjugacy classes and the counting of colourings (Pólya–Redfield and de Bruijn), the isomorphism theorems, the Sylow theorems and the series theorems of Zassenhaus and Schreier—for digital laws, with proofs that run in the digits. Chapter 5 is algorithmic. It presents the Schreier–Sims algorithm [6, 7, 8, 9], whose base and strong generating set is precisely a mixed-radix labelling, and then computes conjugacy classes (their basic theory is in Section 4.4.2), power maps, rationality and fusion, ending with a complete example for M 12 M_{12} M 12 . Chapter 6 develops ordinary character theory—Maschke, Schur and Wedderburn, the orthogonality relations, integrality, induction and restriction—and turns it into a roadmap for computing character tables from class data [3, 4, 10, 11].
Families of groups (Chapters 7–9). Chapter 7 gives the laws of cyclic groups, additive and multiplicative, and of finite Abelian groups as direct sums, with explicit isomorphisms for coprime factors. Chapter 8 treats semidirect products, dicyclic, dihedral and metacyclic groups, wreath products and transitive actions; its Section 8.4 computes the characters of metacyclic, dihedral and dicyclic groups directly from the two digits of their laws. Chapter 9 classifies the groups of order at most 32 and gives a law for each of them [12, 13].
Crystallographic groups (Chapter 10). Chapter 10 gives digital laws for the finite quotients G / m L G/mL G / m L of the 17 plane groups and of the 230 space groups. An element { R i ∣ t i + 24 s } \{R_i \mid t_i+24s\} { R i ∣ t i + 24 s } receives the label i m 3 + s 1 m 2 + s 2 m + s 3 i m^3+s_1m^2+s_2m+s_3 i m 3 + s 1 m 2 + s 2 m + s 3 ; the product adds the rotated translation digits, and the carry c ( i , j ) c(i,j) c ( i , j ) is an integer cocycle determined by the conventional operations. The quotients modulo 2 T 2T 2 T , 6 T 6T 6 T and 2 L 2L 2 L are identified in the small groups library and separated by their invariants.
Finite simple groups (Chapters 11–13). Chapter 11 treats Sym n \operatorname{Sym}_n Sym n and Alt n \operatorname{Alt}_n Alt n : generators and presentations, conjugacy classes, simplicity, maximal subgroups, automorphisms including the outer automorphism of Sym 6 \operatorname{Sym}_6 Sym 6 , the characters of Sym n \operatorname{Sym}_n Sym n and Alt n \operatorname{Alt}_n Alt n [14, 15, 16, 17, 18], and the Lehmer and rotation laws. Chapter 12 builds finite fields in digital form and then the groups of Lie type, from PSL ( 2 , N ) \operatorname{PSL}(2,N) PSL ( 2 , N ) through the classical families, the Suzuki and Ree groups, G 2 ( q ) G_2(q) G 2 ( q ) and 3 D 4 ( q ) {}^{3}D_4(q) 3 D 4 ( q ) to F 4 ( q ) F_4(q) F 4 ( q ) , 2 F 4 ( q ) {}^{2}F_4(q) 2 F 4 ( q ) and the groups of type E E E . Their product rules are transcriptions of the Bruhat normal form and the Chevalley commutator relations [19, 20, 21, 22, 23, 24, 25]. Chapter 13 gives laws for the sporadic groups, from the Mathieu groups through the Conway, Fischer and Janko groups to the Monster and the Baby Monster, relative to standard generators [26, 27, 28, 29]. It ends with canonical labels and with the recomputation of a character table from a printed law.
Worked products and software (Chapters 14 and 15). Chapter 14 multiplies explicit labels in PSL 2 ( 7 ) \operatorname{PSL}_2(7) PSL 2 ( 7 ) , PSL 3 ( 2 ) \operatorname{PSL}_3(2) PSL 3 ( 2 ) , Sz ( 8 ) \operatorname{Sz}(8) Sz ( 8 ) , 2 G 2 ( 3 ) {}^{2}\operatorname{G}_2(3) 2 G 2 ( 3 ) , M 24 \operatorname{M}_{24} M 24 , Co 0 \operatorname{Co}_0 Co 0 , J 4 \operatorname{J}_4 J 4 , Th \operatorname{Th} Th and the Monster, and computes the inverses and orders of the factors and products within the digital laws themselves, with an independent check by decoding. Chapter 15 describes the published archives, their installation and layout, the reproduction workflow, and each program and data file, including those for the crystallographic groups.
Chapter 16 contains solutions to the exercises of all chapters.
Dependencies and reading paths
Figure 1 shows the main dependencies. Chapter 1 is used everywhere, and the permutation codes of § 7.1 reappear wherever groups act. Chapter 4 builds on the diagrammatic construction of Chapter 3. Chapter 5 combines the Schreier–Sims algorithm with conjugacy classes, and Chapter 6 uses both parts. The families of Chapters 7–9 need essentially only Chapters 1 and 2; the exceptions are § 14.4 , which uses Chapter 6, and § 14.13 , which uses Chapters 4 and 6. These chapters are placed after the general theory so that Chapters 3–6 form one block, but a reader who prefers examples first can read them directly after Chapter 2. Chapters 11–13 draw on all of this: Chapter 11 on § 7.1 and on Chapters 5 and 6, Chapter 12 on Chapter 11, and Chapter 13 on the stabiliser chains of Chapter 5, the transitive laws of Theorem 8.32 and the groups of Lie type. Chapter 10 uses the semidirect products of Chapter 8 and the identification of small groups in Chapter 9. Chapter 14 draws on Chapters 12 and 13, and Chapter 15 can be consulted at any point alongside the chapter whose computations it reproduces.
Figure 1. Main dependencies between the chapters. An arrow A → B A \to B A → B means that Chapter B B B uses results of Chapter A A A ; Chapter 1 and § 7.1 are used throughout, and only their first arrows are drawn. The dashed arrows into Chapter 8 mark § 14.4 and § 14.13 . Chapter 15, which documents the programs of all chapters, is not drawn.
Three routes through the book are natural. A first course in finite groups in digital form reads Chapters 1 and 2, then Chapters 7–9, and then Chapter 4. A reader interested in computation reads Chapters 1 and 2, then Chapters 5 and 6. A reader interested in the simple groups can go from Chapters 1 and 2 directly to Chapters 11–13, consulting Chapters 5–6 and Chapter 8 when they are cited.
Verification and reproducibility
Every law in the book is a finite object, and most of them are small enough to be checked exhaustively. Generated tables—for instance the character data of § 14.4 and the tables of Chapter 11—are produced by scripts kept with the source of the book. They rebuild each group from its printed law, verify the stated identities in exact arithmetic and compare the results with GAP [30]; the single command python3 scripts/sn_verify.py regenerates and checks all of them. For the large groups the laws are tied to standard generators [26, 27] and come with fingerprints, so that they can be reproduced independently of this book. Chapter 15 lists the published archives and describes every program and data file; the same scripts reproduce the crystallographic laws and identifications of Chapter 10 and the worked products and inverses of Chapter 14.
Conventions
Throughout, A N = { 0 , 1 , … , N − 1 } \mathcal{A}_N = \{0,1,\ldots,N-1\} A N = { 0 , 1 , … , N − 1 } , d N ( k , x ) d_N(k,x) d N ( k , x ) is the k k k -th base-N N N digit of x x x , and D h ( m ) D_h(m) D h ( m ) is the h h h -th digit of m m m in a mixed radix fixed by the context. Permutations are composed from right to left, as in (12) . Group laws are written m ∗ n m * n m ∗ n on labels; when a group is given by generators,
x a y α x^{a}y^{\alpha} x a y α denotes the element whose label has digits a a a and α \alpha α . The preface closes with a survey of related work and a list of what we believe is new.
A digital law is a bijection λ : G → A ∣ G ∣ \lambda\colon G \to A_{\lvert G\rvert} λ : G → A ∣ G ∣ together with the transported product m ∗ n = λ ( λ − 1 ( m ) λ − 1 ( n ) ) m \ast n = \lambda\bigl(\lambda^{-1}(m)\lambda^{-1}(n)\bigr) m ∗ n = λ ( λ − 1 ( m ) λ − 1 ( n ) ) . Bijections of this kind are not new; what this book adds is a uniform, explicit and reproducible catalogue of them. We list the known ingredients first.
Stabiliser chains. Sims introduced bases and strong generating sets [6]: for a base ( b 1 , … , b k ) (b_1,\ldots,b_k) ( b 1 , … , b k ) every element of a permutation group factors uniquely as s 1 , t 1 ⋯ s k , t k s_{1,t_1}\cdots s_{k,t_k} s 1 , t 1 ⋯ s k , t k , with s l , t s_{l,t} s l , t running over a transversal of the l l l -th basic orbit. Reading the indices t l t_l t l in mixed radix is exactly the chain law of Chapter 13 and the iterated transitive law of Theorem 8.32 . This factorisation is the basis of permutation group algorithms [7, 8, 9]; GAP enumerates the elements of a permutation group through it [30], Knuth uses “Sims tables” to generate all elements of a group [31], and Jerrum studied compact representations built on it [32]. The bijection underlying a chain law is therefore classical; the contribution here is to treat the resulting integer law as the object of study and to publish it.
Ranking and unranking. The permutation codes of Section 2.1 are relatives of the Lehmer code and the factorial number system [33]. Ranking and unranking combinatorial objects, including permutations in lexicographic order and in linear time, is a developed subject [34, 35, 36]. Its emphasis is on efficient bijections for whole classes of objects, not on transporting a group product.
Normal forms in groups of Lie type. The laws of Chapter 12 pack the coordinates of the Bruhat normal form G = ⨆ w U T w U ˙ w G = \bigsqcup_{w} U T w \dot{U}_{w} G = ⨆ w U T w U ˙ w into one integer. The normal form and the Chevalley commutator relations that drive the product are classical [20, 21], and algorithms that multiply in these coordinates are implemented in Magma [37]. Our product rules for 3 D 4 ( q ) {}^{3}D_{4}(q) 3 D 4 ( q ) , F 4 ( q ) F_{4}(q) F 4 ( q ) , 2 F 4 ( q ) {}^{2}F_{4}(q) 2 F 4 ( q ) and the E E E series are explicit transcriptions of this theory, checked against matrix models.
Canonical representatives. The lexicographically least element of a coset, and more generally the smallest image under a group, is a standard tool in backtrack search [7, 8, 38]. The canonical lexicographic law uses this greedy step; what is new is to use it to fix a labelling that depends only on the permutation group.
The Monster and the Baby Monster. The smallest permutation degrees of B \mathbb{B} B and M \mathbb{M} M , about 1.4 ⋅ 10 10 1.4\cdot10^{10} 1.4 ⋅ 1 0 10 and 10 20 10^{20} 1 0 20 , exclude the orbit enumeration that works for the other sporadic groups, so the coset digits must be computed from the structure. Our starting point is the Python package mmgroup of Seysen, which multiplies elements of M \mathbb{M} M and reduces them to a normal form, together with the 196884 196884 196884 -dimensional representation, in which the 2 A 2A 2 A involutions correspond to axes. The package is a model of M \mathbb{M} M , not a labelling of it: it does not give a canonical coset representative, and we construct one from the orbits of K = C G x 0 ( β + ) K = C_{G_{x_0}}(\beta^+) K = C G x 0 ( β + ) on the axes.
Crystallographic groups. The 17 plane and 230 space groups and their symmetry operations are classical and tabulated in the International Tables for Crystallography [39]. A space group is an extension of its lattice by its point group, described by a cohomology class with values in the translations, and its finite quotients by sublattices are a standard tool for studying subgroups of finite index. The laws of Chapter 10 make this extension explicit as integer arithmetic: the label is the point-group index followed by the translation digits modulo m m m , and the carry is the integer cocycle c ( i , j ) c(i,j) c ( i , j ) of the conventional operations. The quotients modulo 2 T 2T 2 T , 6 T 6T 6 T and 2 L 2L 2 L are identified with GAP [30].
Groups as data. The Atlas [26] and the online Atlas of group representations [27] fix standard generators, which is what makes the generator fingerprints of this book reproducible. Integer-labelled group operations have recently become benchmarks in machine learning [40, 41], where they are mostly limited to cyclic and small symmetric groups.
What is new. To our knowledge the following are not available elsewhere in one place:
a digital law, with explicit λ \lambda λ , λ − 1 \lambda^{-1} λ − 1 and ∗ \ast ∗ on A ∣ G ∣ \mathcal{A}_{|G|} A ∣ G ∣ , for every family of finite simple groups treated here, in a single format;
explicit coordinate product rules for the exceptional families, verified over every field tested;
canonical laws that depend only on the group and its standard generators, with fingerprints that allow independent reproduction;
complete laws for all groups of order at most 32;
digital laws for the finite quotients G / m L G/mL G / m L of all 17 plane and 230 space crystallographic groups, with integer cocycles as carries, and identifications of the quotients modulo 2 T 2T 2 T , 6 T 6T 6 T and 2 L 2L 2 L in the small groups library (Chapter 10);
the observation, carried out for M 11 M_{11} M 11 , that the printed laws suffice to recompute classical invariants such as the character table;
a method that produces associative laws layer by layer from diagrams of carries (Chapter 3), and digital versions of the main theorems of finite group theory (Chapter 4);
character theory carried out on codes, with closed formulas in the digits of the labels for the characters of metacyclic, dihedral and dicyclic groups (§ 14.4 ) and with generated character data for Sym n \operatorname{Sym}_n Sym n and Alt n \operatorname{Alt}_n Alt n (Chapter 11);
a digital law for M \mathsf{M} M whose coset digit indexes the 2 A 2A 2 A involution (equivalently its axis) and whose fibre digit is a law of 2. B 2.B 2. B (Proposition 13.74), and a digital law for B \mathsf{B} B obtained from it (Theorem 13.72), in both cases with canonical coset representatives built from orbits of axes and checked against mmgroup on random elements;
worked products, inverses and orders of explicit labels in groups from PSL 2 ( 7 ) \operatorname{PSL}_2(7) PSL 2 ( 7 ) to the Monster, computed within the digital laws and checked independently by decoding (Chapter 14);
generated tables whose scripts rebuild each group from its printed law, verify the stated identities exactly and compare them with GAP [30], published with all other programs and data in the versioned archives described in Chapter 15.
Nothing in this book changes what is known about the structure of these groups. The aim is to make them explicit, computable and comparable as integer arithmetic.
The digit function
Digits of integers and the digital expansion
Definition 1.1. Let N ≥ 2 N \ge2 N ≥ 2 be an integer. For x ∈ R x \in\mathbb{R} x ∈ R and k ∈ Z k \in\mathbb{Z} k ∈ Z the digit function is
d N ( k , x ) = ⌊ x N k ⌋ − N ⌊ x N k + 1 ⌋ (1) d_N(k,x)=\left\lfloor\frac{x}{N^k}\right\rfloor-N\left\lfloor\frac{x}{N^{k+1}}\right\rfloor \tag*{(1)} d N ( k , x ) = ⌊ N k x ⌋ − N ⌊ N k + 1 x ⌋ (1) The set A N = { 0 , 1 , … , N − 1 } \mathcal{A}_N=\{0,1,\ldots,N-1\} A N = { 0 , 1 , … , N − 1 } is the digital alphabet of size N N N .
Lemma 1.2 (Basic properties). Let x ∈ R x \in\mathbb{R} x ∈ R and k ∈ Z k \in\mathbb{Z} k ∈ Z .
d N ( k , x ) ∈ A N d_N(k,x) \in\mathcal{A}_N d N ( k , x ) ∈ A N and d N ( k , x ) = d N ( 0 , x / N k ) d_N(k,x)=d_N(0,x/N^k) d N ( k , x ) = d N ( 0 , x / N k ) .
If x ∈ Z x \in\mathbb{Z} x ∈ Z , then d N ( 0 , x ) = x m o d N d_N(0,x)=x \bmod N d N ( 0 , x ) = x mod N .
If x ∈ Z x \in\mathbb{Z} x ∈ Z and 0 ≤ x < N M 0 \le x < N^M 0 ≤ x < N M , then d N ( k , x ) = 0 d_N(k,x)=0 d N ( k , x ) = 0 for k ≥ M k \ge M k ≥ M and x = ∑ k = 0 M − 1 d N ( k , x ) N k x=\sum_{k=0}^{M-1}d_N(k,x)N^k x = ∑ k = 0 M − 1 d N ( k , x ) N k .
Proof. Put y = x / N k y=x/N^k y = x / N k . Since ⌊ ⌊ y ⌋ / N ⌋ = ⌊ y / N ⌋ \left\lfloor\lfloor y\rfloor/N\right\rfloor=\lfloor y/N\rfloor ⌊ ⌊ y ⌋ / N ⌋ = ⌊ y / N ⌋ , Euclidean division gives ⌊ y ⌋ = N ⌊ y / N ⌋ + r \lfloor y\rfloor=N\lfloor y/N\rfloor+r ⌊ y ⌋ = N ⌊ y / N ⌋ + r with r ∈ A N r \in\mathcal{A}_N r ∈ A N , and r = d N ( k , x ) r=d_N(k,x) r = d N ( k , x ) ; this proves (1) and (2). For (3) the sum telescopes:
∑ k = 0 M − 1 ( ⌊ x N k ⌋ − N ⌊ x N k + 1 ⌋ ) N k = x − N M ⌊ x N M ⌋ = x . \sum_{k=0}^{M-1}\left(\left\lfloor\frac{x}{N^k}\right\rfloor-N\left\lfloor\frac{x}{N^{k+1}}\right\rfloor\right)N^k=x-N^M\left\lfloor\frac{x}{N^M}\right\rfloor=x. k = 0 ∑ M − 1 ( ⌊ N k x ⌋ − N ⌊ N k + 1 x ⌋ ) N k = x − N M ⌊ N M x ⌋ = x .
Lemma 1.3 (Modular arithmetic). For x , y ∈ Z x,y \in\mathbb{Z} x , y ∈ Z , d N ( 0 , x ± y ) = d N ( 0 , d N ( 0 , x ) ± d N ( 0 , y ) ) d_N(0,x\pm y)=d_N(0,d_N(0,x)\pm d_N(0,y)) d N ( 0 , x ± y ) = d N ( 0 , d N ( 0 , x ) ± d N ( 0 , y )) and d N ( 0 , x y ) = d N ( 0 , d N ( 0 , x ) d N ( 0 , y ) ) d_N(0,xy)=d_N(0,d_N(0,x)d_N(0,y)) d N ( 0 , x y ) = d N ( 0 , d N ( 0 , x ) d N ( 0 , y )) .
Proof. Immediate from Lemma 1.2(2).
Lemma 1.4 (Block digits). Let n ≥ 1 n \ge1 n ≥ 1 , x ∈ Z ≥ 0 x \in\mathbb{Z}_{\ge0} x ∈ Z ≥ 0 , a ∈ Z ≥ 0 a \in\mathbb{Z}_{\ge0} a ∈ Z ≥ 0 and b ∈ A n b \in\mathcal{A}_n b ∈ A n . Then
d N ( a n + b , x ) = d N ( b , d N n ( a , x ) ) . (2) d_N(an+b,x)=d_N(b,d_{N^n}(a,x)). \tag*{(2)} d N ( an + b , x ) = d N ( b , d N n ( a , x )) . (2)
Proof.
d N n ( a , x ) = ⌊ x N a n ⌋ m o d N n , d N ( a n + b , x ) = ⌊ x N a n + b ⌋ m o d N . d_{N^n}(a,x)=\left\lfloor\frac{x}{N^{an}}\right\rfloor\bmod N^n,\qquad d_N(an+b,x)=\left\lfloor\frac{x}{N^{an+b}}\right\rfloor\bmod N. d N n ( a , x ) = ⌊ N an x ⌋ mod N n , d N ( an + b , x ) = ⌊ N an + b x ⌋ mod N .
Because b < n b<n b < n , the second expression is the b b b -th base-N N N digit of the first.
Definition 1.5 (Mixed radix). Let p 0 , … , p ℓ − 1 ≥ 2 p_0,\ldots,p_{\ell-1} \ge2 p 0 , … , p ℓ − 1 ≥ 2 and Π h = ∏ j < h p j \Pi_h=\prod_{j<h}p_j Π h = ∏ j < h p j (so Π 0 = 1 \Pi_0=1 Π 0 = 1 ). The h h h -th mixed-radix digit of x x x is
D h ( x ) = d p h ( 0 , x Π h ) . (3) D_h(x)=d_{p_h}\left(0,\frac{x}{\Pi_h}\right). \tag*{(3)} D h ( x ) = d p h ( 0 , Π h x ) . (3)
Lemma 1.6. The map x ↦ ( D 0 ( x ) , … , D ℓ − 1 ( x ) ) x \mapsto(D_0(x),\ldots,D_{\ell-1}(x)) x ↦ ( D 0 ( x ) , … , D ℓ − 1 ( x )) is a bijection A Π ℓ → A p 0 × ⋯ × A p ℓ − 1 A_{\Pi_\ell} \to A_{p_0} \times\cdots\times A_{p_{\ell-1}} A Π ℓ → A p 0 × ⋯ × A p ℓ − 1 , with inverse ( x h ) ↦ ∑ h x h P h (x_h) \mapsto\sum_h x_hP_h ( x h ) ↦ ∑ h x h P h .
Proof. Since ⌊ ⌊ x / Π h ⌋ / p h ⌋ = ⌊ x / Π h + 1 ⌋ \left\lfloor\left\lfloor x/\Pi_h\right\rfloor/p_h\right\rfloor=\left\lfloor x/\Pi_{h+1}\right\rfloor ⌊ ⌊ x / Π h ⌋ / p h ⌋ = ⌊ x / Π h + 1 ⌋ , we have
D h ( x ) = ⌊ x / Π h ⌋ − p h ⌊ x / Π h + 1 ⌋ . D_h(x)=\left\lfloor x/\Pi_h\right\rfloor-p_h\left\lfloor x/\Pi_{h+1}\right\rfloor. D h ( x ) = ⌊ x / Π h ⌋ − p h ⌊ x / Π h + 1 ⌋ . Moreover, ∑ h D h ( x ) Π h \sum_h D_h(x)\Pi_h ∑ h D h ( x ) Π h telescopes to x − Π ℓ ⌊ x / Π ℓ ⌋ = x x-\Pi_\ell\lfloor x/\Pi_\ell\rfloor=x x − Π ℓ ⌊ x / Π ℓ ⌋ = x . So the map is injective; both sets have Π ℓ \Pi_\ell Π ℓ elements. □
Theorem 1.7 (Coding principle). Let r ≥ 1 r\ge1 r ≥ 1 . Every map f : A N r → A N f:A_N^r\to A_N f : A N r → A N is of the form
f ( x 0 , … , x r − 1 ) = d N ( ∑ k = 0 r − 1 x k N k , R ) (4) f(x_0,\ldots,x_{r-1})=d_N\left(\sum_{k=0}^{r-1}x_kN^k,R\right) \tag*{(4)} f ( x 0 , … , x r − 1 ) = d N ( k = 0 ∑ r − 1 x k N k , R ) (4) for a unique integer R ∈ [ 0 , N N r ) R\in[0,N^{N^r}) R ∈ [ 0 , N N r ) , the code of f f f , namely R = ∑ x ∈ A N r f ( x ) N ∑ k x k N k R=\sum_{x\in A_N^r}f(x)N^{\sum_k x_kN^k} R = ∑ x ∈ A N r f ( x ) N ∑ k x k N k . Hence there are exactly N N r N^{N^r} N N r r r r -ary operators on A N A_N A N .
Proof. By Lemma 1.2(3), x ↦ ∑ k x k N k x\mapsto\sum_kx_kN^k x ↦ ∑ k x k N k is a bijection A N r → A N r A_N^r\to A_{N^r} A N r → A N r and R ↦ ( d N ( n , R ) ) n < N r R\mapsto(d_N(n,R))_{n<N^r} R ↦ ( d N ( n , R ) ) n < N r is a bijection A N N r → A N N r A_{N^{N^r}}\to A_N^{N^r} A N N r → A N N r . □
Computing digits in practice
The digit function (1) is a closed formula: it gives the digit of x x x in position k k k directly, for real x x x and every integer k k k . This section compares it with the way digits are extracted in computation, and with Lehmer’s code of a permutation [33], which is the mixed-radix expansion used for S y m n \mathrm{Sym}_n Sym n in Definition 11.1.
Sequential extraction
For a radix p h = 1 p_h=1 p h = 1 the right-hand side of (1) still makes sense and vanishes identically; we then read (3) as D h ( x ) = 0 D_h(x)=0 D h ( x ) = 0 . This is the case of the last radix in Definition 11.1.
Proposition 1.8 (Sequential digits). Let p 0 , … , p ℓ − 1 ≥ 1 p_0,\ldots,p_{\ell-1}\ge1 p 0 , … , p ℓ − 1 ≥ 1 be integers, Π h = ∏ j < h p j \Pi_h=\prod_{j<h}p_j Π h = ∏ j < h p j and x ∈ Z ≥ 0 x\in\mathbb{Z}_{\ge0} x ∈ Z ≥ 0 . Put x 0 = x x_0=x x 0 = x and x h + 1 = ⌊ x h / p h ⌋ x_{h+1}=\lfloor x_h/p_h\rfloor x h + 1 = ⌊ x h / p h ⌋ . Then x h = ⌊ x / Π h ⌋ x_h=\lfloor x/\Pi_h\rfloor x h = ⌊ x / Π h ⌋ and
D h ( x ) = x h − p h x h + 1 = x h m o d p h ( 0 ≤ h < ℓ ) . (5) D_h(x)=x_h-p_hx_{h+1}=x_h\bmod p_h \qquad(0\le h<\ell). \tag*{(5)} D h ( x ) = x h − p h x h + 1 = x h mod p h ( 0 ≤ h < ℓ ) . (5) In particular d N ( k , x ) = x k − N x k + 1 d_N(k,x)=x_k-Nx_{k+1} d N ( k , x ) = x k − N x k + 1 for the constant radix p h = N p_h=N p h = N .
Proof. For an integer p ≥ 1 p\ge1 p ≥ 1 and real y y y one has ⌊ ⌊ y ⌋ / p ⌋ = ⌊ y / p ⌋ \lfloor\lfloor y\rfloor/p\rfloor=\lfloor y/p\rfloor ⌊⌊ y ⌋ / p ⌋ = ⌊ y / p ⌋ . By induction, x h + 1 = ⌊ ⌊ x / Π h ⌋ / p h ⌋ = ⌊ x / Π h + 1 ⌋ x_{h+1}=\lfloor\lfloor x/\Pi_h\rfloor/p_h\rfloor=\lfloor x/\Pi_{h+1}\rfloor x h + 1 = ⌊⌊ x / Π h ⌋ / p h ⌋ = ⌊ x / Π h + 1 ⌋ . By (3) and (1) , D h ( x ) = ⌊ x / Π h ⌋ − p h ⌊ x / Π h + 1 ⌋ = x h − p h x h + 1 D_h(x)=\lfloor x/\Pi_h\rfloor-p_h\lfloor x/\Pi_{h+1}\rfloor=x_h-p_hx_{h+1} D h ( x ) = ⌊ x / Π h ⌋ − p h ⌊ x / Π h + 1 ⌋ = x h − p h x h + 1 , which is the remainder of x h x_h x h modulo p h p_h p h . □
Proposition 1.8 is the familiar rule “divide and keep the remainder”. It is formula (1) evaluated with the quotient carried from one position to the next, instead of being recomputed from x x x each time.
Remark 1.9 (Cost). Exact integers are stored in base 2 w 2^w 2 w for a word size w w w , typically w = 30 w=30 w = 30 or w = 64 w=64 w = 64 . Let x x x occupy L L L words and let every radix fit in one word. A division by a one-word integer costs O ( L ) O(L) O ( L ) operations, so (5) gives all ℓ \ell ℓ digits in O ( ℓ L ) O(\ell L) O ( ℓ L ) operations. Evaluating (1) or (3) separately in every position instead divides x x x by the multiword integers Π h \Pi_h Π h and Π h + 1 \Pi_{h+1} Π h + 1 . With schoolbook division, dividing L L L words by s s s words costs about s ( L − s ) s(L-s) s ( L − s ) operations; summed over the positions this is of order L 3 L^3 L 3 when ℓ \ell ℓ is proportional to L L L . The closed formula is therefore the right tool for one digit in a prescribed position and for proofs, and the recurrence is the right tool for a full expansion.
Remark 1.10 (Divide and conquer). Lemma 1.4 gives a faster method for very long expansions. If 0 ≤ x < N 2 n 0 \le x < N^{2n} 0 ≤ x < N 2 n , write x = r + N n q x = r + N^n q x = r + N n q with r = d N n ( 0 , x ) r = d_{N^n}(0,x) r = d N n ( 0 , x ) and q = d N n ( 1 , x ) q = d_{N^n}(1,x) q = d N n ( 1 , x ) . Then d N ( b , x ) = d N ( b , r ) d_N(b,x) = d_N(b,r) d N ( b , x ) = d N ( b , r ) and d N ( n + b , x ) = d N ( b , q ) d_N(n+b,x) = d_N(b,q) d N ( n + b , x ) = d N ( b , q ) for b ∈ A n b \in A_n b ∈ A n . Applied recursively, this halves the problem at each step; with fast multiplication and division the whole expansion costs only a logarithmic factor more than one product of L L L -word integers. The same splitting works in mixed radix, with x = ( x m o d Π h ) + Π h ⌊ x / Π h ⌋ x = (x \bmod\Pi_h) + \Pi_h \lfloor x/\Pi_h \rfloor x = ( x mod Π h ) + Π h ⌊ x / Π h ⌋ . Arbitrary-precision libraries convert long binary integers to decimal in this way.
Remark 1.11 (Exact arithmetic). Formula (1) is stated for real x x x , but it must be evaluated with exact floors. In IEEE double precision every integer is exact only up to 2 53 2^{53} 2 53 , and 18 ! < 2 53 < 19 ! 18! < 2^{53} < 19! 18 ! < 2 53 < 19 ! . Hence the Lehmer labels of Sym n \operatorname{Sym}_n Sym n (Definition 11.1 ) can be handled in double precision only for n ≤ 18 n \le18 n ≤ 18 , and with unsigned 64-bit integers only for n ≤ 20 n \le20 n ≤ 20 , since 20 ! < 2 64 < 21 ! 20! < 2^{64} < 21! 20 ! < 2 64 < 21 ! . Beyond this one needs arbitrary-precision integers, for which (5) is the native operation. Converting such an integer to decimal for printing is itself the expansion (1) with N = 10 N = 10 N = 10 ; for integers with many thousands of decimal digits it is often the most expensive step of a computation.
Lehmer's code and the Lehmer label
For R ∈ Sym n R \in\operatorname{Sym}_n R ∈ Sym n let a i ( R ) = # { j > i : R ( j ) < R ( i ) } ∈ [ 0 , n − i ) a_i(R) = \#\{j > i : R(j) < R(i)\} \in[0,n-i) a i ( R ) = # { j > i : R ( j ) < R ( i )} ∈ [ 0 , n − i ) , as in Definition 11.1 . The sequence ( a 0 ( R ) , … , a n − 1 ( R ) ) (a_0(R), \ldots, a_{n-1}(R)) ( a 0 ( R ) , … , a n − 1 ( R )) is Lehmer's code of R R R . Classically it is read in the factorial number system,
L ( R ) = ∑ i = 0 n − 1 a i ( R ) ( n − 1 − i ) ! , L(R) = \sum_{i=0}^{n-1} a_i(R)(n-1-i)!, L ( R ) = i = 0 ∑ n − 1 a i ( R ) ( n − 1 − i )! ,
whereas this book uses the Lehmer label
λ ( R ) = ∑ i = 0 n − 1 a i ( R ) Π i , Π i = ∏ j < i ( n − j ) = n ! ( n − i ) ! . \lambda(R) = \sum_{i=0}^{n-1} a_i(R)\Pi_i,\qquad\Pi_i = \prod_{j<i}(n-j) = \frac{n!}{(n-i)!}. λ ( R ) = i = 0 ∑ n − 1 a i ( R ) Π i , Π i = j < i ∏ ( n − j ) = ( n − i )! n ! .
Proposition 1.12 (Two readings of one code). Both λ \lambda λ and L L L are bijections Sym n → [ 0 , n ! ) \operatorname{Sym}_n \to[0,n!) Sym n → [ 0 , n !) , and λ ( id ) = L ( id ) = 0 \lambda(\operatorname{id}) = L(\operatorname{id}) = 0 λ ( id ) = L ( id ) = 0 .
The mixed-radix digits of λ ( R ) \lambda(R) λ ( R ) for the radices ( n , n − 1 , … , 1 ) (n,n-1,\ldots,1) ( n , n − 1 , … , 1 ) are a 0 ( R ) , … , a n − 1 ( R ) a_0(R),\ldots,a_{n-1}(R) a 0 ( R ) , … , a n − 1 ( R ) ; thus a 0 ( R ) a_0(R) a 0 ( R ) is the least significant digit.
The mixed-radix digits of L ( R ) L(R) L ( R ) for the radices ( 1 , 2 , … , n ) (1,2,\ldots,n) ( 1 , 2 , … , n ) are a n − 1 ( R ) , … , a 0 ( R ) a_{n-1}(R),\ldots,a_0(R) a n − 1 ( R ) , … , a 0 ( R ) ; thus a 0 ( R ) a_0(R) a 0 ( R ) is the most significant digit. Moreover L ( R ) L(R) L ( R ) is the rank of R R R when Sym n \operatorname{Sym}_n Sym n is listed in lexicographic order of one-line words.
Hence λ \lambda λ and L L L read the same digit string in opposite directions.
Proof. (1) is Theorem 11.2 . For (2), the radices q k = k + 1 q_k = k+1 q k = k + 1 have place values ∏ j < k ( j + 1 ) = k ! \prod_{j<k}(j+1) = k! ∏ j < k ( j + 1 ) = k ! , and the digit in position k k k is a n − 1 − k ( R ) ∈ [ 0 , k + 1 ) a_{n-1-k}(R) \in[0,k+1) a n − 1 − k ( R ) ∈ [ 0 , k + 1 ) , so Lemma 1.6 gives the bijection. Suppose that R R R precedes S S S lexicographically, and let i i i be the first position with R ( i ) ≠ S ( i ) R(i) \ne S(i) R ( i ) = S ( i ) , so that R ( i ) < S ( i ) R(i) < S(i) R ( i ) < S ( i ) . The values in the positions before i i i agree, hence a j ( R ) = a j ( S ) a_j(R) = a_j(S) a j ( R ) = a j ( S ) for j < i j < i j < i ; and a i ( R ) < a i ( S ) a_i(R) < a_i(S) a i ( R ) < a i ( S ) , because a i a_i a i is the rank of the value in position i i i among the values not yet used. Since ∑ j > i a j ( R ) ( n − 1 − j ) ! ≤ ∑ j > i ( n − 1 − j ) ( n − 1 − j ) ! = ( n − 1 − i ) ! − 1 \sum_{j>i} a_j(R)(n-1-j)! \le\sum_{j>i}(n-1-j)(n-1-j)! = (n-1-i)!-1 ∑ j > i a j ( R ) ( n − 1 − j )! ≤ ∑ j > i ( n − 1 − j ) ( n − 1 − j )! = ( n − 1 − i )! − 1 , it follows that L ( R ) < L ( S ) L(R) < L(S) L ( R ) < L ( S ) . □
The difference matters in computation; Table 1.1 lists the main points. The low digits of λ ( R ) \lambda(R) λ ( R ) describe the beginning of the word of R R R : a 0 ( R ) = λ ( R ) m o d n a_0(R) = \lambda(R) \bmod n a 0 ( R ) = λ ( R ) mod n costs one division by a one-word integer, whereas a 0 ( R ) = ⌊ L ( R ) / ( n − 1 ) ! ⌋ a_0(R) = \lfloor L(R)/(n-1)!\rfloor a 0 ( R ) = ⌊ L ( R ) / ( n − 1 )!⌋ needs a division by a multiword integer. More generally λ ( R ) m o d Π h \lambda(R) \bmod\Pi_h λ ( R ) mod Π h is determined by ( R ( 0 ) , … , R ( h − 1 ) ) (R(0),\ldots,R(h-1)) ( R ( 0 ) , … , R ( h − 1 )) , so it is constant on each left coset R H RH R H of the pointwise stabiliser H H H of { 0 , … , h − 1 } \{0,\ldots,h-1\} { 0 , … , h − 1 } : if S ∈ H S \in H S ∈ H , then ( R ∘ S ) ( i ) = R ( i ) (R \circ S)(i) = R(i) ( R ∘ S ) ( i ) = R ( i ) for i < h i < h i < h . Truncating the label modulo Π h \Pi_h Π h is therefore a step down the point-stabiliser chain.
Table 1. Two readings of Lehmer’s code of R ∈ Sym n R\in\operatorname{Sym}_n R ∈ Sym n .
Example 1.13. Let n = 4 n=4 n = 4 and R = ( 2 , 0 , 3 , 1 ) R=(2,0,3,1) R = ( 2 , 0 , 3 , 1 ) in one-line notation. Then ( a 0 , a 1 , a 2 , a 3 ) = ( 2 , 0 , 1 , 0 ) (a_0,a_1,a_2,a_3)=(2,0,1,0) ( a 0 , a 1 , a 2 , a 3 ) = ( 2 , 0 , 1 , 0 ) , so λ ( R ) = 2 ⋅ 1 + 0 ⋅ 4 + 1 ⋅ 12 + 0 ⋅ 24 = 14 \lambda(R)=2\cdot1+0\cdot4+1\cdot12+0\cdot24=14 λ ( R ) = 2 ⋅ 1 + 0 ⋅ 4 + 1 ⋅ 12 + 0 ⋅ 24 = 14 and L ( R ) = 2 ⋅ 3 ! + 0 ⋅ 2 ! + 1 ⋅ 1 ! + 0 ⋅ 0 ! = 13 L(R)=2\cdot3!+0\cdot2!+1\cdot1!+0\cdot0!=13 L ( R ) = 2 ⋅ 3 ! + 0 ⋅ 2 ! + 1 ⋅ 1 ! + 0 ⋅ 0 ! = 13 . Indeed twelve words begin with 0 or 1, and ( 2 , 0 , 1 , 3 ) (2,0,1,3) ( 2 , 0 , 1 , 3 ) is the only word beginning with 2 that precedes R R R . Proposition 1.8 recovers the code from 14 with one small division per digit,
14 = 3 ⋅ 4 + 2 , 3 = 1 ⋅ 3 + 0 , 1 = 0 ⋅ 2 + 1 , 0 = 0 ⋅ 1 + 0 , 14=3\cdot4+2,\qquad3=1\cdot3+0,\qquad1=0\cdot2+1,\qquad0=0\cdot1+0, 14 = 3 ⋅ 4 + 2 , 3 = 1 ⋅ 3 + 0 , 1 = 0 ⋅ 2 + 1 , 0 = 0 ⋅ 1 + 0 ,
while the closed formula gives a single digit directly, for instance D 2 ( 14 ) = ⌊ 14 / 12 ⌋ − 2 ⌊ 14 / 24 ⌋ = 1 D_2(14)=\lfloor14/12\rfloor-2\lfloor14/24\rfloor=1 D 2 ( 14 ) = ⌊ 14/12 ⌋ − 2 ⌊ 14/24 ⌋ = 1 .
Remark 1.14 (Which definition is practical). Lehmer’s code is not a different definition of digits: it is a choice of radices and of the order of the digits, and once these are fixed its digits are the D h ( x ) D_h(x) D h ( x ) of (3) . As a definition, the digit function (1) is the more useful one. It is exact and closed, it applies to real x x x and to negative k k k (for instance d N ( − 1 , x ) = ⌊ N x ⌋ − N ⌊ x ⌋ d_N(-1,x)=\lfloor Nx\rfloor-N\lfloor x\rfloor d N ( − 1 , x ) = ⌊ N x ⌋ − N ⌊ x ⌋ is the first digit after the point), and the identities of this chapter are proved from it. As an algorithm it is practical provided that the floors are exact integer divisions and that a full expansion is computed by (5) , or by Remark 1.10 for very long integers. The accompanying scripts follow exactly this pattern.
The B \mathcal{B} B -function
The digit function extracts the digits of a code; the B \mathcal{B} B -function tests them. Together they are the two basic building blocks of digital mathematics: every finite law, every comparison of labels and every carry of positional arithmetic is a closed formula in these two functions. This section collects the properties of the B \mathcal{B} B -function needed later, following the treatment in the digital mathematics notes (Chapter 3, Section 4.8 and Section 11.4 there). We write it as B \mathcal{B} B , in calligraphic type, which also keeps it apart from the digit labels A , B , C A,B,C A , B , C used in Chapter 3.
Definition and values
Definition 1.15. For real x , y x,y x , y the B \mathcal{B} B -function is
B ( x , y ) = 1 2 ( sgn ( x + y ) − sgn ( x − y ) ) , \mathcal{B}(x,y)=\frac{1}{2}\bigl(\operatorname{sgn}(x+y)-\operatorname{sgn}(x-y)\bigr), B ( x , y ) = 2 1 ( sgn ( x + y ) − sgn ( x − y ) ) ,
and we abbreviate B ( x ) = B ( x , 1 2 ) \mathcal{B}(x)=\mathcal{B}(x,\frac{1}{2}) B ( x ) = B ( x , 2 1 ) .
Proposition 1.16.
For y > 0 y>0 y > 0 , B ( x , y ) = 1 \mathcal{B}(x,y)=1 B ( x , y ) = 1 if ∣ x ∣ < y \lvert x\rvert<y ∣ x ∣ < y , B ( x , y ) = 1 2 \mathcal{B}(x,y)=\frac{1}{2} B ( x , y ) = 2 1 if ∣ x ∣ = y \lvert x\rvert=y ∣ x ∣ = y , and B ( x , y ) = 0 \mathcal{B}(x,y)=0 B ( x , y ) = 0 if ∣ x ∣ > y \lvert x\rvert>y ∣ x ∣ > y . Moreover B ( x , 0 ) = 0 \mathcal{B}(x,0)=0 B ( x , 0 ) = 0 and B ( − x , y ) = B ( x , y ) \mathcal{B}(-x,y)=\mathcal{B}(x,y) B ( − x , y ) = B ( x , y ) .
For an integer u u u , B ( u ) = 1 \mathcal{B}(u)=1 B ( u ) = 1 if u = 0 u=0 u = 0 and B ( u ) = 0 \mathcal{B}(u)=0 B ( u ) = 0 otherwise. In particular B ( u ) = d n ( u , 1 ) \mathcal{B}(u)=d_n(u,1) B ( u ) = d n ( u , 1 ) for u ∈ A n u\in\mathcal{A}_n u ∈ A n , the digital Kronecker delta used earlier.
For integers a ≤ b a\leq b a ≤ b and x x x , B ( x − a + b 2 , b − a + 1 2 ) = 1 \mathcal{B}\left(x-\frac{a+b}{2},\frac{b-a+1}{2}\right)=1 B ( x − 2 a + b , 2 b − a + 1 ) = 1 if a ≤ x ≤ b a\leq x\leq b a ≤ x ≤ b and 0 otherwise.
Proof. (1) If ∣ x ∣ < y |x|<y ∣ x ∣ < y then x + y > 0 > x − y x+y>0>x-y x + y > 0 > x − y and the bracket is 2; if x = ± y x=\pm y x = ± y exactly one sign vanishes and the other is ± 1 \pm1 ± 1 with the right sign; if ∣ x ∣ > y |x|>y ∣ x ∣ > y both signs agree. For y = 0 y=0 y = 0 the two signs coincide, and sgn \operatorname{sgn} sgn is odd, which gives the symmetry. (2) is (1) with y = 1 2 y=\frac{1}{2} y = 2 1 , since a nonzero integer has ∣ u ∣ ≥ 1 |u|\ge1 ∣ u ∣ ≥ 1 . (3) The condition ∣ x − a + b 2 ∣ < b − a 2 \left|x-\frac{a+b}{2}\right|<\frac{b-a}{2} x − 2 a + b < 2 b − a means a − 1 2 < x < b + 1 2 a-\frac{1}{2}<x<b+\frac{1}{2} a − 2 1 < x < b + 2 1 , and the boundary values are half-integers, never attained by x x x .
Thus B \mathcal{B} B is the indicator of an open interval (with the value 1 2 \frac{1}{2} 2 1 on its ends), and on integers it is an exact if-then-else : the map sending a a a to b b b and every other integer to c c c is f ( n ) = c + ( b − c ) B ( n − a ) f(n)=c+(b-c)\mathcal{B}(n-a) f ( n ) = c + ( b − c ) B ( n − a ) .
Identities
Proposition 1.17 (Splitting and block coalescence). For real x , y , z , m , ε x,y,z,m,\varepsilon x , y , z , m , ε and n ≥ 1 n\ge1 n ≥ 1 ,
B ( x , y + z ) = B ( x + z , y ) + B ( x − y , z ) , (6) \mathcal{B}(x,y+z)=\mathcal{B}(x+z,y)+\mathcal{B}(x-y,z), \tag*{(6)} B ( x , y + z ) = B ( x + z , y ) + B ( x − y , z ) , (6) ∑ k = 0 n − 1 B ( m − 2 k ε , ε ) = B ( m − ( n − 1 ) ε , n ε ) . (7) \sum_{k=0}^{n-1}\mathcal{B}(m-2k\varepsilon,\varepsilon)=\mathcal{B}(m-(n-1)\varepsilon,n\varepsilon). \tag*{(7)} k = 0 ∑ n − 1 B ( m − 2 k ε , ε ) = B ( m − ( n − 1 ) ε , n ε ) . (7)
Proof. The right side of (6) is 1 2 ( sgn ( x + y + z ) − sgn ( x + z − y ) + sgn ( x − y + z ) − sgn ( x − y − z ) ) \frac{1}{2}\bigl(\operatorname{sgn}(x+y+z)-\operatorname{sgn}(x+z-y)+\operatorname{sgn}(x-y+z)-\operatorname{sgn}(x-y-z)\bigr) 2 1 ( sgn ( x + y + z ) − sgn ( x + z − y ) + sgn ( x − y + z ) − sgn ( x − y − z ) ) ; the middle terms cancel and what remains is B ( x , y + z ) \mathcal{B}(x,y+z) B ( x , y + z ) . For (7) use induction on n n n : the case n = 1 n=1 n = 1 is trivial, and (6) with x = m − n ε x=m-n\varepsilon x = m − n ε , y = n ε y=n\varepsilon y = n ε , z = ε z=\varepsilon z = ε gives B ( m − n ε , ( n + 1 ) ε ) = B ( m − ( n − 1 ) ε , n ε ) + B ( m − 2 n ε , ε ) \mathcal{B}(m-n\varepsilon,(n+1)\varepsilon)=\mathcal{B}(m-(n-1)\varepsilon,n\varepsilon)+\mathcal{B}(m-2n\varepsilon,\varepsilon) B ( m − n ε , ( n + 1 ) ε ) = B ( m − ( n − 1 ) ε , n ε ) + B ( m − 2 n ε , ε ) .
Splitting cuts an interval into two adjacent ones; block coalescence glues n n n adjacent intervals of width 2 ε 2\varepsilon 2 ε into one. With ε = 1 2 \varepsilon=\frac{1}{2} ε = 2 1 it says that a sum of consecutive Kronecker deltas is a single B \mathcal{B} B -function, which is how carries will be written below.
Proposition 1.18 (Integer deltas). For integers x , a , b x,a,b x , a , b and any finite set S S S of integers containing a a a ,
B ( x − a ) 2 = B ( x − a ) , B ( x − a ) B ( x − b ) = B ( x − a ) B ( a − b ) , \mathcal{B}(x-a)^2=\mathcal{B}(x-a),\qquad\mathcal{B}(x-a)\mathcal{B}(x-b)=\mathcal{B}(x-a)\mathcal{B}(a-b), B ( x − a ) 2 = B ( x − a ) , B ( x − a ) B ( x − b ) = B ( x − a ) B ( a − b ) , B ( x − a ) = ∑ k ∈ S B ( x − k ) B ( k − a ) . \mathcal{B}(x-a)=\sum_{k\in S}\mathcal{B}(x-k)\mathcal{B}(k-a). B ( x − a ) = k ∈ S ∑ B ( x − k ) B ( k − a ) .
Proof. All three follow from Proposition 1.16 (2): the deltas take values 0, 1; both sides of the second identity are 1 exactly when x = a = b x=a=b x = a = b ; in the sum only k = a k=a k = a can contribute, and it contributes B ( x − a ) \mathcal{B}(x-a) B ( x − a ) .
The last identity (propagation ) may be iterated: inserting any number of summed deltas between x x x and a a a changes nothing. It is the digital form of composing Kronecker deltas.
The deltas are also the evaluation kernel of every finite operation: a code is turned into a function by selecting one of its digits.
Proposition 1.19 (Evaluation kernel). Let R ∈ [ 0 , N N 2 ) R\in[0,N^{N^2}) R ∈ [ 0 , N N 2 ) be the code of a binary law ∗ \ast ∗ on A N \mathcal{A}_N A N . Then for a , b ∈ A N a,b\in\mathcal{A}_N a , b ∈ A N
a ∗ b = d N ( a N + b , R ) = ∑ n = 0 N 2 − 1 d N ( n , R ) B ( n − a N − b ) . a\ast b=d_N(aN+b,R)=\sum_{n=0}^{N^2-1}d_N(n,R)\mathcal{B}(n-aN-b). a ∗ b = d N ( a N + b , R ) = n = 0 ∑ N 2 − 1 d N ( n , R ) B ( n − a N − b ) . More generally, if R ∈ [ 0 , N N r ) R\in[0,N^{N^r}) R ∈ [ 0 , N N r ) is the code of an r r r -ary operator on A N \mathcal{A}_N A N , then
R ( x 0 , … , x r − 1 ) = ∑ n = 0 N r − 1 d N ( n , R ) B ( n − ∑ j = 0 r − 1 x j N j ) . R(x_0,\ldots,x_{r-1})=\sum_{n=0}^{N^r-1}d_N(n,R)\mathcal{B}\left(n-\sum_{j=0}^{r-1}x_jN^j\right). R ( x 0 , … , x r − 1 ) = n = 0 ∑ N r − 1 d N ( n , R ) B ( n − j = 0 ∑ r − 1 x j N j ) .
Proof. By Proposition 1.16 (2) only the term with n = a N + b n=aN+b n = a N + b (respectively n = ∑ j x j N j n=\sum_j x_jN^j n = ∑ j x j N j ) survives, and it equals the digit selected by the coding of Theorem 1.7 .
The first form reads one digit of R R R ; the second writes the operator as a linear combination of deltas with the digits of R R R as coefficients, which is the form used to manipulate tables algebraically (for instance in Theorem 1.20 below).
The B \mathcal{B} B -function and the digits
Theorem 1.20 (Factorization). For integers N ≥ 2 N\ge2 N ≥ 2 and a , b ∈ [ 0 , N M ) a,b\in[0,N^M) a , b ∈ [ 0 , N M ) ,
B ( a − b ) = ∏ k = 0 M − 1 B ( d N ( k , a ) − d N ( k , b ) ) . \mathcal{B}(a-b)=\prod_{k=0}^{M-1}\mathcal{B}\bigl(d_N(k,a)-d_N(k,b)\bigr). B ( a − b ) = k = 0 ∏ M − 1 B ( d N ( k , a ) − d N ( k , b ) ) .
The same holds for mixed-radix labels: two labels are equal if and only if all their digits are.
Proof. By uniqueness of the radix-N N N (or mixed-radix) representation, a = b a=b a = b if and only if all digits agree, and the product is 1 exactly in that case.
Corollary 1.21 (Boolean form). For N = 2 N=2 N = 2 , writing a k = d 2 ( k , a ) a_k=d_2(k,a) a k = d 2 ( k , a ) and b k = d 2 ( k , b ) b_k=d_2(k,b) b k = d 2 ( k , b ) ,
B ( a − b ) = ∏ k = 0 M − 1 ( 1 − a k − b k + 2 a k b k ) . \mathcal{B}(a-b)=\prod_{k=0}^{M-1}(1-a_k-b_k+2a_kb_k). B ( a − b ) = k = 0 ∏ M − 1 ( 1 − a k − b k + 2 a k b k ) .
Proof. By propagation over S = { 0 , 1 } S=\{0,1\} S = { 0 , 1 } , B ( a k − b k ) = B ( a k ) B ( b k ) + B ( a k − 1 ) B ( 1 − b k ) = ( 1 − a k ) ( 1 − b k ) + a k b k \mathcal{B}(a_k-b_k)=\mathcal{B}(a_k)\mathcal{B}(b_k)+\mathcal{B}(a_k-1)\mathcal{B}(1-b_k)=(1-a_k)(1-b_k)+a_kb_k B ( a k − b k ) = B ( a k ) B ( b k ) + B ( a k − 1 ) B ( 1 − b k ) = ( 1 − a k ) ( 1 − b k ) + a k b k .
Applied to codes, Theorem 1.20 turns the equality of two magmas into an entrywise comparison of their tables: for codes R , R ′ R,R' R , R ′ of binary laws on A N \mathcal{A}_N A N ,
B ( R − R ′ ) = ∏ m = 0 N 2 − 1 B ( d N ( m , R ) − d N ( m , R ′ ) ) , ∑ m = 0 N 2 − 1 ( 1 − B ( d N ( m , R ) − d N ( m , R ′ ) ) ) \mathcal{B}(R-R')=\prod_{m=0}^{N^2-1}\mathcal{B}\bigl(d_N(m,R)-d_N(m,R')\bigr),\qquad\sum_{m=0}^{N^2-1}\left(1-\mathcal{B}\bigl(d_N(m,R)-d_N(m,R')\bigr)\right) B ( R − R ′ ) = m = 0 ∏ N 2 − 1 B ( d N ( m , R ) − d N ( m , R ′ ) ) , m = 0 ∑ N 2 − 1 ( 1 − B ( d N ( m , R ) − d N ( m , R ′ ) ) )
being, respectively, the test R = R ′ R=R' R = R ′ and the number of entries in which the two tables differ.
Carries
Proposition 1.22 (The carry as a B \mathcal{B} B -function). For N ≥ 2 N\ge2 N ≥ 2 and u , v ∈ A N u,v\in\mathcal{A}_N u , v ∈ A N let c N ( u , v ) = ⌊ ( u + v ) / N ⌋ c_N(u,v)=\left\lfloor(u+v)/N\right\rfloor c N ( u , v ) = ⌊ ( u + v ) / N ⌋ . Then
c N ( u , v ) = B ( u + v − 3 N − 2 2 , N − 1 2 ) , u + v = d N ( 0 , u + v ) + N c N ( u , v ) . c_N(u,v)=\mathcal{B}\left(u+v-\frac{3N-2}{2},\frac{N-1}{2}\right),\qquad u+v=d_N(0,u+v)+Nc_N(u,v). c N ( u , v ) = B ( u + v − 2 3 N − 2 , 2 N − 1 ) , u + v = d N ( 0 , u + v ) + N c N ( u , v ) .
Proof. Since 0 ≤ u + v ≤ 2 N − 2 0\le u+v\le2N-2 0 ≤ u + v ≤ 2 N − 2 , the carry is 1 exactly when N ≤ u + v ≤ 2 N − 2 N\le u+v\le2N-2 N ≤ u + v ≤ 2 N − 2 ; apply Proposition 1.16 (3) with a = N a=N a = N , b = 2 N − 2 b=2N-2 b = 2 N − 2 . Equivalently, the carry is the sum of the deltas B ( u + v − s ) \mathcal{B}(u+v-s) B ( u + v − s ) for s = N , … , 2 N − 2 s=N,\ldots,2N-2 s = N , … , 2 N − 2 , coalesced by (1.7). The second identity is the definition of the lowest digit.
Proposition 1.23 (The carry is a cocycle). For u , v , w ∈ A N u,v,w\in\mathcal{A}_N u , v , w ∈ A N ,
c N ( u , v ) + c N ( d N ( 0 , u + v ) , w ) = c N ( v , w ) + c N ( u , d N ( 0 , v + w ) ) = ⌊ u + v + w N ⌋ . c_N(u,v)+c_N\bigl(d_N(0,u+v),w\bigr)=c_N(v,w)+c_N\bigl(u,d_N(0,v+w)\bigr)=\left\lfloor\frac{u+v+w}{N}\right\rfloor. c N ( u , v ) + c N ( d N ( 0 , u + v ) , w ) = c N ( v , w ) + c N ( u , d N ( 0 , v + w ) ) = ⌊ N u + v + w ⌋ .
Proof. By Proposition 1.22 applied twice,
u + v + w = d N ( 0 , u + v ) + w + N c N ( u , v ) = d N ( 0 , u + v + w ) + N [ c N ( u , v ) + c N ( d N ( 0 , u + v ) , w ) ] , u+v+w=d_N(0,u+v)+w+N c_N(u,v)=d_N(0,u+v+w)+N\left[c_N(u,v)+c_N\left(d_N(0,u+v),w\right)\right], u + v + w = d N ( 0 , u + v ) + w + N c N ( u , v ) = d N ( 0 , u + v + w ) + N [ c N ( u , v ) + c N ( d N ( 0 , u + v ) , w ) ] ,
and the bracket is ⌊ ( u + v + w ) / N ⌋ \lfloor(u+v+w)/N\rfloor ⌊( u + v + w ) / N ⌋ because the lowest digit is less than N N N . Grouping v + w v+w v + w first gives the other expression.
The ordinary addition of integers thus splits into two parts. For a , b ∈ [ 0 , N M ) a,b\in[0,N^M) a , b ∈ [ 0 , N M ) define the carry-free sum and the carry part
a ⊕ N b = ∑ k = 0 M − 1 N k d N ( 0 , d N ( k , a ) + d N ( k , b ) ) , a + ^ N b = a + b − a ⊕ N b . a\oplus_N b=\sum_{k=0}^{M-1}N^k d_N\left(0,d_N(k,a)+d_N(k,b)\right),\qquad a\mathbin{\hat{+}_N}b=a+b-a\oplus_N b. a ⊕ N b = k = 0 ∑ M − 1 N k d N ( 0 , d N ( k , a ) + d N ( k , b ) ) , a + ^ N b = a + b − a ⊕ N b .
The carry-free sum adds each digit modulo N N N independently of the others: it is the law of C N ⊕ ⋯ ⊕ C N C_N\oplus\cdots\oplus C_N C N ⊕ ⋯ ⊕ C N , an operator acting position by position (a graded operator in the sense of Definition 3.2 ). All coupling between the positions sits in the carry part. For instance, in base 10, 56782 + 36754 = 93536 56782+36754=93536 56782 + 36754 = 93536 splits as 82436 + 11100 82436+11100 82436 + 11100 . Chapter 3 generalises exactly this picture: a diagram without arrows is a graded operator, a direct sum of its layers, and the carry arrows are arbitrary admissible replacements of the carry c N c_N c N (§ 9.1.3 ).
Remark 1.24. Carries also count. By Kummer’s theorem the exponent of p p p in ( a + b a ) \binom{a+b}{a} ( a a + b ) is the number of carries in the base-p p p addition a + b a+b a + b , that is, the number of positions at which the propagated carry is 1. This is the form in which carries enter the proof of the Sylow theorems in Chapter 4 (Lemma 4.28 and the discussion after it).
Finite sets and characteristic functions
In the rest of the book every finite object is a set of integer codes: subgroups, cosets and classes are subsets of an alphabet A n A_n A n . This section records how such subsets are handled with digits and with the B \mathcal{B} B -function. For integers x x x we only use that B ( x ) = 1 \mathcal{B}(x)=1 B ( x ) = 1 if x = 0 x=0 x = 0 and B ( x ) = 0 \mathcal{B}(x)=0 B ( x ) = 0 otherwise (Section 1.3 and the exercises below).
Definition 1.25. Let S ⊆ A n S\subseteq A_n S ⊆ A n . Its characteristic function and its code are
1 S ( x ) = ∑ s ∈ S B ( x − s ) ( x ∈ A n ) , R S = ∑ s ∈ S 2 s . (8) \mathbf{1}_S(x)=\sum_{s\in S}\mathcal{B}(x-s)\quad(x\in A_n),\qquad R_S=\sum_{s\in S}2^s. \tag*{(8)} 1 S ( x ) = s ∈ S ∑ B ( x − s ) ( x ∈ A n ) , R S = s ∈ S ∑ 2 s . (8)
Proposition 1.26. For S ⊆ A n S\subseteq A_n S ⊆ A n and x ∈ A n x\in A_n x ∈ A n one has 1 S ( x ) = d 2 ( x , R S ) \mathbf{1}_S(x)=d_2(x,R_S) 1 S ( x ) = d 2 ( x , R S ) and ∣ S ∣ = ∑ x ∈ A n 1 S ( x ) |S|=\sum_{x\in A_n}\mathbf{1}_S(x) ∣ S ∣ = ∑ x ∈ A n 1 S ( x ) . The map S ↦ R S S\mapsto R_S S ↦ R S is a bijection from the subsets of A n A_n A n onto A 2 n A_{2^n} A 2 n ; in particular A n A_n A n has 2 n 2^n 2 n subsets.
Proof. At most one term of the sum in (1.8) is nonzero, and it equals 1 exactly when x ∈ S x\in S x ∈ S ; summing over x x x counts each element of S S S once. The binary digits of R S R_S R S are 1 S ( 0 ) , … , 1 S ( n − 1 ) \mathbf{1}_S(0),\ldots,\mathbf{1}_S(n-1) 1 S ( 0 ) , … , 1 S ( n − 1 ) , so the last claim is Lemma 1.2 (3) with p = 2 p=2 p = 2 .
Proposition 1.27 (Boolean operations). For S , T ⊆ A n S,T\subseteq A_n S , T ⊆ A n ,
1 S ∩ T = 1 S 1 T , 1 S ∪ T = 1 S + 1 T − 1 S 1 T , 1 S △ T = 1 S + 1 T − 21 S 1 T , 1 A n ∖ S = 1 − 1 S , \mathbf{1}_{S\cap T}=\mathbf{1}_S\mathbf{1}_T,\qquad\mathbf{1}_{S\cup T}=\mathbf{1}_S+\mathbf{1}_T-\mathbf{1}_S\mathbf{1}_T,\qquad\mathbf{1}_{S\triangle T}=\mathbf{1}_S+\mathbf{1}_T-2\mathbf{1}_S\mathbf{1}_T,\qquad\mathbf{1}_{A_n\setminus S}=1-\mathbf{1}_S, 1 S ∩ T = 1 S 1 T , 1 S ∪ T = 1 S + 1 T − 1 S 1 T , 1 S △ T = 1 S + 1 T − 2 1 S 1 T , 1 A n ∖ S = 1 − 1 S , and
R S ∩ T = ∑ k < n d 2 ( k , R S ) d 2 ( k , R T ) 2 k . R_{S\cap T}=\sum_{k<n}d_2(k,R_S)d_2(k,R_T)2^k. R S ∩ T = k < n ∑ d 2 ( k , R S ) d 2 ( k , R T ) 2 k . If P ⊆ A n P\subseteq A_n P ⊆ A n is the set of elements with a given property and A ⊆ A n A\subseteq A_n A ⊆ A n , the statements “some x ∈ A x\in A x ∈ A has the property” and “every x ∈ A x\in A x ∈ A has the property” have the truth values
1 − B ( ∑ x ∈ A 1 P ( x ) ) and ∏ x ∈ A 1 P ( x ) . 1-\mathcal{B}\left(\sum_{x\in A}\mathbf{1}_P(x)\right)\qquad\text{and}\qquad\prod_{x\in A}\mathbf{1}_P(x). 1 − B ( x ∈ A ∑ 1 P ( x ) ) and x ∈ A ∏ 1 P ( x ) .
Proof. Each identity is checked on the four possible values of ( 1 S ( x ) , 1 T ( x ) ) (\mathbf{1}_S(x),\mathbf{1}_T(x)) ( 1 S ( x ) , 1 T ( x )) , and the formula for R S ∩ T R_{S\cap T} R S ∩ T then follows from Proposition 1.26 . The sum in the first truth value is a non-negative integer that vanishes exactly when no x ∈ A x \in A x ∈ A lies in P P P .
Theorem 1.28 (Multiplicities). Let S 1 , … , S m ⊂ A n S_1,\ldots,S_m \subset A_n S 1 , … , S m ⊂ A n and let c ( x ) = ∑ j = 1 m 1 S j ( x ) c(x)=\sum_{j=1}^{m}\mathbf{1}_{S_j}(x) c ( x ) = ∑ j = 1 m 1 S j ( x ) be the number of these sets containing x x x . Then x x x lies in exactly k k k of the sets if and only if B ( k − c ( x ) ) = 1 \mathcal{B}(k-c(x))=1 B ( k − c ( x )) = 1 , and, for k ≥ 1 k\geq1 k ≥ 1 , in at least one and at most k k k of them if and only if B ( k + 1 − 2 c ( x ) , k ) = 1 \mathcal{B}(k+1-2c(x),k)=1 B ( k + 1 − 2 c ( x ) , k ) = 1 . In particular
1 S 1 ∪ ⋯ ∪ S m ( x ) = 1 − B ( c ( x ) ) = B ( m + 1 − 2 c ( x ) , m ) . (9) \mathbf{1}_{S_1\cup\cdots\cup S_m}(x)=1-\mathcal{B}(c(x))=\mathcal{B}(m+1-2c(x),m). \tag*{(9)} 1 S 1 ∪ ⋯ ∪ S m ( x ) = 1 − B ( c ( x )) = B ( m + 1 − 2 c ( x ) , m ) . (9)
Proof. The first claim holds because k − c ( x ) k-c(x) k − c ( x ) is an integer. For the second, ∣ k + 1 − 2 c ∣ < k |k+1-2c|<k ∣ k + 1 − 2 c ∣ < k holds exactly when 1 ≤ c ≤ k 1\leq c\leq k 1 ≤ c ≤ k , and ∣ k + 1 − 2 c ∣ = k |k+1-2c|=k ∣ k + 1 − 2 c ∣ = k is impossible for an integer c c c , so the boundary convention of the two-argument B \mathcal{B} B -function plays no role. For (9) take k = m k=m k = m and note that c ( x ) ≤ m c(x)\leq m c ( x ) ≤ m .
Remark 1.29. When a further set S S S joins the union, the right-hand side of (9) changes linearly: the first argument gains 1 − 21 S ( x ) 1-2\mathbf{1}_S(x) 1 − 2 1 S ( x ) and the second gains 1 1 1 . The classical expansion of Corollary 1.30 needs 2 m − 1 2^m-1 2 m − 1 terms instead of m m m .
Corollary 1.30 (Inclusion–exclusion). With the notation of Theorem 1.28 , and with ⋂ j ∈ ∅ S j = A n \bigcap_{j\in\varnothing}S_j=A_n ⋂ j ∈ ∅ S j = A n ,
B ( c ( x ) ) = ∏ j = 1 m ( 1 − 1 S j ( x ) ) = ∑ J ⊆ { 1 , … , m } ( − 1 ) ∣ J ∣ 1 ⋂ j ∈ J S j ( x ) , \mathcal{B}(c(x))=\prod_{j=1}^{m}(1-\mathbf{1}_{S_j}(x))=\sum_{J\subseteq\{1,\ldots,m\}}(-1)^{|J|}\mathbf{1}_{\bigcap_{j\in J}S_j}(x), B ( c ( x )) = j = 1 ∏ m ( 1 − 1 S j ( x )) = J ⊆ { 1 , … , m } ∑ ( − 1 ) ∣ J ∣ 1 ⋂ j ∈ J S j ( x ) , ∣ A n ∖ ⋃ j = 1 m S j ∣ = ∑ J ⊆ { 1 , … , m } ( − 1 ) ∣ J ∣ ∣ ⋂ j ∈ J S j ∣ . \left|A_n\setminus\bigcup_{j=1}^{m}S_j\right|=\sum_{J\subseteq\{1,\ldots,m\}}(-1)^{|J|}\left|\bigcap_{j\in J}S_j\right|. A n ∖ j = 1 ⋃ m S j = J ⊆ { 1 , … , m } ∑ ( − 1 ) ∣ J ∣ j ∈ J ⋂ S j .
Proof. Both B ( c ( x ) ) \mathcal{B}(c(x)) B ( c ( x )) and the product equal 1 exactly when x x x lies in no S j S_j S j , and 0 otherwise. Expand the product and use ∏ j ∈ J 1 S j = 1 ⋂ j ∈ J S j \prod_{j\in J}\mathbf{1}_{S_j}=\mathbf{1}_{\bigcap_{j\in J}S_j} ∏ j ∈ J 1 S j = 1 ⋂ j ∈ J S j (Proposition 1.27 ); then sum over x x x (Proposition 1.26 ).
Example 1.31 (Generators of C 12 C_{12} C 12 ). An element of C 12 C_{12} C 12 , coded by A 12 A_{12} A 12 , generates the group exactly when it lies in no maximal subgroup. The maximal subgroups are M 1 = { 0 , 2 , … , 10 } M_1=\{0,2,\ldots,10\} M 1 = { 0 , 2 , … , 10 } and M 2 = { 0 , 3 , 6 , 9 } M_2=\{0,3,6,9\} M 2 = { 0 , 3 , 6 , 9 } , with codes R M 1 = 1365 R_{M_1}=1365 R M 1 = 1365 and R M 2 = 585 R_{M_2}=585 R M 2 = 585 . Their common binary digits are in positions 0 and 6, so R M 1 ∩ M 2 = 65 R_{M_1\cap M_2}=65 R M 1 ∩ M 2 = 65 and M 1 ∩ M 2 = { 0 , 6 } M_1\cap M_2=\{0,6\} M 1 ∩ M 2 = { 0 , 6 } . Corollary 1.30 gives 12 − 6 − 4 + 2 = 4 12-6-4+2=4 12 − 6 − 4 + 2 = 4 generators, namely 1, 5, 7, 11, in agreement with φ ( 12 ) = 4 \varphi(12)=4 φ ( 12 ) = 4 .
Relations and quotient sets
A binary relation on A n A_n A n is a function r : A n 2 → { 0 , 1 } r:A_n^2\to\{0,1\} r : A n 2 → { 0 , 1 } ; via ( x , y ) ↦ x + n y (x,y)\mapsto x+ny ( x , y ) ↦ x + n y it is a subset of A n 2 A_{n^2} A n 2 and so, by Proposition 1.26 , it is determined by the code ∑ x , y r ( x , y ) 2 x + n y \sum_{x,y}r(x,y)2^{x+ny} ∑ x , y r ( x , y ) 2 x + n y . It is an equivalence relation if r ( x , x ) = 1 r(x,x)=1 r ( x , x ) = 1 , r ( x , y ) = r ( y , x ) r(x,y)=r(y,x) r ( x , y ) = r ( y , x ) and r ( x , y ) r ( y , z ) ≤ r ( x , z ) r(x,y)r(y,z)\leq r(x,z) r ( x , y ) r ( y , z ) ≤ r ( x , z ) for all x , y , z x,y,z x , y , z . The class of x x x then has characteristic function 1 [ x ] ( y ) = r ( x , y ) \mathbf{1}_{[x]}(y)=r(x,y) 1 [ x ] ( y ) = r ( x , y ) , and the classes form a partition of A n A_n A n .
Proposition 1.32. Let r r r be an equivalence relation on A n A_n A n . The number of its classes is
∑ x ∈ A n 1 ∑ y ∈ A n r ( x , y ) = ∑ x ∈ A n ∏ y < x ( 1 − r ( x , y ) ) . \sum_{x\in A_n}\frac{1}{\sum_{y\in A_n}r(x,y)}=\sum_{x\in A_n}\prod_{y<x}(1-r(x,y)). x ∈ A n ∑ ∑ y ∈ A n r ( x , y ) 1 = x ∈ A n ∑ y < x ∏ ( 1 − r ( x , y )) .
Proof. In the first sum each class C C C contributes ∣ C ∣ |C| ∣ C ∣ terms equal to 1 / ∣ C ∣ 1/|C| 1/∣ C ∣ . In the second, the product equals 1 exactly when x x x is the least element of its class, so each class is counted once.
Example 1.33. For d ∣ n d \mid n d ∣ n , congruence modulo d d d on A n \mathcal{A}_n A n is r ( x , y ) = B ( d d ( 0 , x − y ) ) r(x,y)=\mathcal{B}(d_d(0,x-y)) r ( x , y ) = B ( d d ( 0 , x − y )) by Lemma 1.2 (2). Every class has n / d n/d n / d elements, so Proposition 1.32 gives n ⋅ d / n = d n\cdot d/n=d n ⋅ d / n = d classes, with least elements 0 , 1 , … , d − 1 0,1,\ldots,d-1 0 , 1 , … , d − 1 . Once a group law is given on A n \mathcal{A}_n A n , the cosets of a subgroup and the conjugacy classes arise in the same way, and Proposition 1.32 counts them.
Exercises
Prove directly from (1) that d N ( k , x ) ∈ A N d_N(k,x)\in\mathcal{A}_N d N ( k , x ) ∈ A N for every integer x ≥ 0 x\geq0 x ≥ 0 and that x = ∑ k ≥ 0 d N ( k , x ) N k x=\sum_{k\geq0}d_N(k,x)N^k x = ∑ k ≥ 0 d N ( k , x ) N k .
Show that d N ( k , x + N k + 1 y ) = d N ( k , x ) d_N(k,x+N^{k+1}y)=d_N(k,x) d N ( k , x + N k + 1 y ) = d N ( k , x ) and d N ( k , N x ) = d N ( k − 1 , x ) d_N(k,Nx)=d_N(k-1,x) d N ( k , N x ) = d N ( k − 1 , x ) for all integers x , y x,y x , y and k ≥ 1 k\geq1 k ≥ 1 .
Show that d N 2 ( k , x ) = d N ( 2 k , x ) + N d N ( 2 k + 1 , x ) d_{N^2}(k,x)=d_N(2k,x)+Nd_N(2k+1,x) d N 2 ( k , x ) = d N ( 2 k , x ) + N d N ( 2 k + 1 , x ) for every integer x ≥ 0 x\geq0 x ≥ 0 , and state the analogous rule for base N n N^n N n .
Using Theorem 1.7 , show that exactly N N ( N + 1 ) / 2 N^{N(N+1)/2} N N ( N + 1 ) /2 of the codes in A N N 2 \mathcal{A}_N^{N^2} A N N 2 define commutative binary operations on A N \mathcal{A}_N A N .
Show that for integers x x x one has B ( x ) = 1 \mathcal{B}(x)=1 B ( x ) = 1 if x = 0 x=0 x = 0 and B ( x ) = 0 \mathcal{B}(x)=0 B ( x ) = 0 otherwise, and that for integers x x x and n ≥ 1 n\geq1 n ≥ 1 , B ( x − n − 1 2 , n 2 ) = 1 \mathcal{B}\left(x-\frac{n-1}{2},\frac{n}{2}\right)=1 B ( x − 2 n − 1 , 2 n ) = 1 if 0 ≤ x < n 0\leq x<n 0 ≤ x < n and 0 0 0 otherwise. Write the indicator of A n \mathcal{A}_n A n and the Kronecker delta δ m , n \delta_{m,n} δ m , n in this form.
Let a , b ≥ 0 a,b\geq0 a , b ≥ 0 and let c k = ⌊ ( a m o d N k ) + ( b m o d N k ) N k ⌋ c_k=\left\lfloor\frac{(a\bmod N^k)+(b\bmod N^k)}{N^k}\right\rfloor c k = ⌊ N k ( a mod N k ) + ( b mod N k ) ⌋ . Show that c k ∈ { 0 , 1 } c_k\in\{0,1\} c k ∈ { 0 , 1 } and d N ( k , a + b ) ≡ d N ( k , a ) + d N ( k , b ) + c k ( m o d N ) d_N(k,a+b)\equiv d_N(k,a)+d_N(k,b)+c_k\pmod N d N ( k , a + b ) ≡ d N ( k , a ) + d N ( k , b ) + c k ( mod N ) , and compare with Proposition 1.22.
Using Kummer’s theorem, show that ( 2 n n ) \binom{2n}{n} ( n 2 n ) is even for every n ≥ 1 n\geq1 n ≥ 1 , and that an odd prime p p p does not divide ( 2 n n ) \binom{2n}{n} ( n 2 n ) if and only if every base-p p p digit of n n n is less than p / 2 p/2 p /2 .
Show that R S ∪ T = R S + R T − R S ∩ T R_{S\cup T}=R_S+R_T-R_{S\cap T} R S ∪ T = R S + R T − R S ∩ T and R S △ T = R S + R T − 2 R S ∩ T R_{S\triangle T}=R_S+R_T-2R_{S\cap T} R S △ T = R S + R T − 2 R S ∩ T for S , T ⊆ A n S,T\subseteq\mathcal{A}_n S , T ⊆ A n .
In the notation of Theorem 1.28 , show that exactly ∑ x ∈ A n B ( k − c ( x ) ) \sum_{x\in\mathcal{A}_n}\mathcal{B}(k-c(x)) ∑ x ∈ A n B ( k − c ( x )) elements lie in exactly k k k of the sets, and that ∑ k = 1 m B ( k − c ( x ) ) = 1 − B ( c ( x ) ) \sum_{k=1}^{m}\mathcal{B}(k-c(x))=1-\mathcal{B}(c(x)) ∑ k = 1 m B ( k − c ( x )) = 1 − B ( c ( x )) .
Use Corollary 1.30 to show that A n \mathcal{A}_n A n has n ! ∑ k = 0 n ( − 1 ) k / k ! n!\sum_{k=0}^{n}(-1)^k/k! n ! ∑ k = 0 n ( − 1 ) k / k ! permutations without fixed points.
Apply Corollary 1.30 to the maximal subgroups of C n C_n C n to prove φ ( n ) = n ∏ p ∣ n ( 1 − 1 / p ) \varphi(n)=n\prod_{p\mid n}(1-1/p) φ ( n ) = n ∏ p ∣ n ( 1 − 1/ p ) .
Let f : A n → A m f:\mathcal{A}_n\to\mathcal{A}_m f : A n → A m . Show that r ( x , y ) = B ( f ( x ) − f ( y ) ) r(x,y)=\mathcal{B}(f(x)-f(y)) r ( x , y ) = B ( f ( x ) − f ( y )) is an equivalence relation whose classes are the fibres of f f f , and that it has as many classes as f f f has values.
Show that a relation r r r on A n \mathcal{A}_n A n is transitive if and only if ∏ x , y , z ( 1 − r ( x , y ) r ( y , z ) + r ( x , y ) r ( y , z ) r ( x , z ) ) = 1 \prod_{x,y,z}(1-r(x,y)r(y,z)+r(x,y)r(y,z)r(x,z))=1 ∏ x , y , z ( 1 − r ( x , y ) r ( y , z ) + r ( x , y ) r ( y , z ) r ( x , z )) = 1 .
Finite magmas and r r r -ary operators
Permutations
Permutation codes
Definition 2.1. A permutation code of degree N N N is an integer R ∈ [ 0 , N N ) R \in[0,N^N) R ∈ [ 0 , N N ) whose N N N digits a n = d N ( n , R ) a_n=d_N(n,R) a n = d N ( n , R ) , n ∈ A N n \in A_N n ∈ A N , are pairwise distinct. It acts by R ( x ) = d N ( x , R ) R(x)=d_N(x,R) R ( x ) = d N ( x , R ) . The set of such codes is Sym N \operatorname{Sym}_N Sym N ; by Theorem 1.7 (with r = 1 r=1 r = 1 ) it is in bijection with the bijections of A N A_N A N , so ∣ Sym N ∣ = N ! |\operatorname{Sym}_N|=N! ∣ Sym N ∣ = N ! . In Cauchy notation, R = ( 0 1 ⋯ N − 1 a 0 a 1 ⋯ a N − 1 ) R=\left(\begin{smallmatrix}0&1&\cdots&N-1\\a_0&a_1&\cdots&a_{N-1}\end{smallmatrix}\right) R = ( 0 a 0 1 a 1 ⋯ ⋯ N − 1 a N − 1 ) .
Proposition 2.2. For R , S ∈ Sym N R,S \in\operatorname{Sym}_N R , S ∈ Sym N :
ι N = ∑ n = 0 N − 1 n N n = N ( N N + 1 − 2 N N + 1 ) ( N − 1 ) 2 (identity), (10) \iota_N=\sum_{n=0}^{N-1}nN^n=\frac{N(N^{N+1}-2N^N+1)}{(N-1)^2}\qquad\text{(identity),} \tag*{(10)} ι N = n = 0 ∑ N − 1 n N n = ( N − 1 ) 2 N ( N N + 1 − 2 N N + 1 ) (identity), (10) R − 1 = ∑ n = 0 N − 1 n N d N ( n , R ) (inverse), (11) R^{-1}=\sum_{n=0}^{N-1}nN^{d_N(n,R)}\qquad\text{(inverse),} \tag*{(11)} R − 1 = n = 0 ∑ N − 1 n N d N ( n , R ) (inverse), (11) R ∘ S = ∑ n = 0 N − 1 d N ( d N ( n , S ) , R ) N n (composition, right to left). (12) R\circ S=\sum_{n=0}^{N-1}d_N(d_N(n,S),R)N^n\qquad\text{(composition, right to left).} \tag*{(12)} R ∘ S = n = 0 ∑ N − 1 d N ( d N ( n , S ) , R ) N n (composition, right to left). (12) With ∘ \circ ∘ , Sym N \operatorname{Sym}_N Sym N is a group, the symmetric group.
Proof. Each formula is (4) for the map in question: ι N \iota_N ι N has digit n n n at position n n n ; R − 1 R^{-1} R − 1 has digit n n n at position R ( n ) R(n) R ( n ) ; R ∘ S R\circ S R ∘ S has digit R ( S ( n ) ) R(S(n)) R ( S ( n )) at position $n. The closed form of ι N \iota_N ι N is the sum ∑ n < N n x n \sum_{n<N}nx^n ∑ n < N n x n at x = N x=N x = N . Composition of maps is associative. □ \square □
Example 2.3. For N = 6 N=6 N = 6 , ( 0 1 2 3 4 5 3 5 2 1 4 0 ) \left(\begin{smallmatrix}0&1&2&3&4&5\\3&5&2&1&4&0\end{smallmatrix}\right) ( 0 3 1 5 2 2 3 1 4 4 5 0 ) has code 3 + 5 ⋅ 6 + 2 ⋅ 6 2 + 1 ⋅ 6 3 + 4 ⋅ 6 4 = 5505 3+5\cdot6+2\cdot6^2+1\cdot6^3+4\cdot6^4=5505 3 + 5 ⋅ 6 + 2 ⋅ 6 2 + 1 ⋅ 6 3 + 4 ⋅ 6 4 = 5505 , cycle form ( 0315 ) (0315) ( 0315 ) , and its inverse has code 13055 13055 13055 .
Cycles and conjugacy
Proposition 2.4 (Cycle code). Suppose R R R decomposes into disjoint cycles
( j 0 ( h ) j 1 ( h ) ⋯ j s h − 1 ( h ) ) , h = 1 , … , c , \left(j_0^{(h)}\ j_1^{(h)}\ \cdots\ j_{s_h-1}^{(h)}\right),\qquad h=1,\ldots,c, ( j 0 ( h ) j 1 ( h ) ⋯ j s h − 1 ( h ) ) , h = 1 , … , c , with fixed points included as 1-cycles. Then
R = ∑ h = 1 c ∑ k = 0 s h − 1 j d s h ( 0 , k + 1 ) ( h ) N j k ( h ) . (13) R=\sum_{h=1}^{c}\sum_{k=0}^{s_h-1}j_{d_{s_h}(0,k+1)}^{(h)}N^{j_k^{(h)}}. \tag*{(13)} R = h = 1 ∑ c k = 0 ∑ s h − 1 j d s h ( 0 , k + 1 ) ( h ) N j k ( h ) . (13)
Proof. The digit of R R R at position j k ( h ) j_k^{(h)} j k ( h ) must be its image j k + 1 ( h ) j_{k+1}^{(h)} j k + 1 ( h ) , indices mod s h s_h s h ; every position occurs exactly once.
Example 2.5. ( 236 ) ∘ ( 01246 ) = ( 0136 ) ( 24 ) (236) \circ(01246) = (0136)(24) ( 236 ) ∘ ( 01246 ) = ( 0136 ) ( 24 ) in Sym 7 \operatorname{Sym}_7 Sym 7 : by (12) , 0 ↦ 1 ↦ 1 0 \mapsto1 \mapsto1 0 ↦ 1 ↦ 1 , 1 ↦ 2 ↦ 3 1 \mapsto2 \mapsto3 1 ↦ 2 ↦ 3 , 2 ↦ 4 ↦ 4 2 \mapsto4 \mapsto4 2 ↦ 4 ↦ 4 , 3 ↦ 3 ↦ 6 3 \mapsto3 \mapsto6 3 ↦ 3 ↦ 6 , 4 ↦ 6 ↦ 2 4 \mapsto6 \mapsto2 4 ↦ 6 ↦ 2 , 6 ↦ 0 ↦ 0 6 \mapsto0 \mapsto0 6 ↦ 0 ↦ 0 .
Proposition 2.6 (Conjugation). For R , S ∈ Sym N R,S \in\operatorname{Sym}_N R , S ∈ Sym N ,
S ∘ R ∘ S − 1 = ∑ n = 0 N − 1 d N ( d N ( n , R ) , S ) N d N ( n , S ) . S \circ R \circ S^{-1} = \sum_{n=0}^{N-1} d_N(d_N(n,R),S)N^{d_N(n,S)}. S ∘ R ∘ S − 1 = n = 0 ∑ N − 1 d N ( d N ( n , R ) , S ) N d N ( n , S ) . Conjugation replaces each symbol of the cycle form of R R R by its image under S S S . Hence two permutations are conjugate if and only if they have the same cycle type, and the conjugacy classes of Sym N \operatorname{Sym}_N Sym N correspond to the partitions of N N N .
Proof. S R S − 1 SRS^{-1} S R S − 1 sends S ( n ) S(n) S ( n ) to S ( R ( n ) ) S(R(n)) S ( R ( n )) , which is the stated digit placement. Conversely, given two permutations with the same cycle type, the bijection matching their cycles symbol by symbol is a conjugating S S S .
Sign and the alternating group
Definition 2.7. The sign of R ∈ Sym N R \in\operatorname{Sym}_N R ∈ Sym N is sgn R = \operatorname{sgn} R = sgn R =
∏ 0 ≤ i < j ≤ N − 1 d N ( j , R ) − d N ( i , R ) j − i ∈ { ± 1 } . \prod_{0 \le i < j \le N-1} \frac{d_N(j,R)-d_N(i,R)}{j-i} \in\{\pm1\}. 0 ≤ i < j ≤ N − 1 ∏ j − i d N ( j , R ) − d N ( i , R ) ∈ { ± 1 } .
Theorem 2.8. sgn : Sym N → { ± 1 } \operatorname{sgn}:\operatorname{Sym}_N \to\{\pm1\} sgn : Sym N → { ± 1 } is a homomorphism; an s s s -cycle has sign ( − 1 ) s − 1 (-1)^{s-1} ( − 1 ) s − 1 ; and sgn R = ( − 1 ) N − c ( R ) \operatorname{sgn} R = (-1)^{N-c(R)} sgn R = ( − 1 ) N − c ( R ) , where c ( R ) c(R) c ( R ) is the number of cycles of R R R (fixed points included).
Proof. sgn ( R ∘ S ) = ∏ i < j R ( S ( j ) ) − R ( S ( i ) ) S ( j ) − S ( i ) ∏ i < j S ( j ) − S ( i ) j − i \operatorname{sgn}(R \circ S) = \prod_{i<j}\frac{R(S(j))-R(S(i))}{S(j)-S(i)}\prod_{i<j}\frac{S(j)-S(i)}{j-i} sgn ( R ∘ S ) = ∏ i < j S ( j ) − S ( i ) R ( S ( j )) − R ( S ( i )) ∏ i < j j − i S ( j ) − S ( i ) , and the first factor is sgn R \operatorname{sgn} R sgn R after reindexing over unordered pairs { S ( i ) , S ( j ) } \{S(i),S(j)\} { S ( i ) , S ( j )} . A transposition ( j k ) (j\ k) ( j k ) , j < k j<k j < k , reverses exactly the 2 ( k − j ) − 1 2(k-j)-1 2 ( k − j ) − 1 pairs it meets, so has sign − 1 -1 − 1 . Moreover ( j 0 j 1 ⋯ j s − 1 ) = ( j 0 j s − 1 ) ∘ ⋯ ∘ ( j 0 j 2 ) ∘ ( j 0 j 1 ) (j_0\,j_1\,\cdots\,j_{s-1})=(j_0\,j_{s-1})\circ\cdots\circ(j_0\,j_2)\circ(j_0\,j_1) ( j 0 j 1 ⋯ j s − 1 ) = ( j 0 j s − 1 ) ∘ ⋯ ∘ ( j 0 j 2 ) ∘ ( j 0 j 1 ) , which gives ( − 1 ) s − 1 (-1)^{s-1} ( − 1 ) s − 1 , and then ∏ h ( − 1 ) s h − 1 = ( − 1 ) N − c \prod_h(-1)^{s_h-1}=(-1)^{N-c} ∏ h ( − 1 ) s h − 1 = ( − 1 ) N − c .
Definition 2.9. The alternating group is Alt N = ker sgn \operatorname{Alt}_N=\ker\operatorname{sgn} Alt N = ker sgn . For N ≥ 2 N \ge2 N ≥ 2 it has index 2 and order N ! / 2 N!/2 N ! /2 .
Example 2.10 (Sym3 _3 3 ). The six permutation codes of degree 3 are
Table 2.10.
so Alt 3 = { 7 , 11 , 21 } \operatorname{Alt}_3=\{7,11,21\} Alt 3 = { 7 , 11 , 21 } and ι 3 = 21 \iota_3=21 ι 3 = 21 .
Magmas and their codes
Definition 2.11. A magma of order N N N is a map ∗ : A N × A N → A N \ast:\mathcal{A}_N \times\mathcal{A}_N \to\mathcal{A}_N ∗ : A N × A N → A N . By Theorem 1.7 (with x 0 = b x_0=b x 0 = b , x 1 = a x_1=a x 1 = a ) there is a unique R ∈ [ 0 , N N 2 ) R \in[0,N^{N^2}) R ∈ [ 0 , N N 2 ) such that
a ∗ b = d N ( a N + b , R ) . (14) a \ast b=d_N(aN+b,R). \tag*{(14)} a ∗ b = d N ( a N + b , R ) . (14) We write ( A N , R ) (\mathcal{A}_N,R) ( A N , R ) for the magma. There are N N 2 N^{N^2} N N 2 magmas of order N N N .
Proposition 2.12 (Translations are block digits). Let λ a ( b ) = a ∗ b \lambda_a(b)=a*b λ a ( b ) = a ∗ b and ρ b ( a ) = a ∗ b \rho_b(a)=a*b ρ b ( a ) = a ∗ b . Then λ a \lambda_a λ a has the (not necessarily bijective) code L a = d N N ( a , R ) L_a=d_{N^N}(a,R) L a = d N N ( a , R ) , ρ b \rho_b ρ b has code C b = ∑ a d N ( a N + b , R ) N a C_b=\sum_a d_N(aN+b,R)N^a C b = ∑ a d N ( a N + b , R ) N a , and
R = ∑ a = 0 N − 1 L a ( N N ) a (15) R=\sum_{a=0}^{N-1}L_a\left(N^N\right)^a \tag*{(15)} R = a = 0 ∑ N − 1 L a ( N N ) a (15)
Proof. Lemma 1.4 with n = N n=N n = N gives d N ( a N + b , R ) = d N ( b , L a ) d_N(aN+b,R)=d_N(b,L_a) d N ( a N + b , R ) = d N ( b , L a ) ; (15) is Lemma 1.2 (3) in base N N N^N N N .
Definition 2.13 (Axioms). A magma ( A N , R ) (\mathcal{A}_N,R) ( A N , R ) is
a semigroup if it is associative, a ∗ ( b ∗ c ) = ( a ∗ b ) ∗ c a*(b*c)=(a*b)*c a ∗ ( b ∗ c ) = ( a ∗ b ) ∗ c ;
commutative if a ∗ b = b ∗ a a*b=b*a a ∗ b = b ∗ a ;
a monoid if it is a semigroup in which 0 0 0 is an identity, a ∗ 0 = 0 ∗ a = a a*0=0*a=a a ∗ 0 = 0 ∗ a = a ;
a quasigroup if every λ a \lambda_a λ a and every ρ b \rho_b ρ b is bijective (the Cayley table is a Latin square);
a loop if it is a quasigroup in which 0 0 0 is an identity;
a group if it is an associative loop. A commutative group is Abelian .
Placing the identity at 0 0 0 loses no generality, by relabelling (Proposition 2.20).
Proposition 2.14 (Digital form of the axioms). For a magma ( A N , R ) (\mathcal{A}_N,R) ( A N , R ) :
it is commutative if and only if R = R o p R=R^{\mathrm{op}} R = R op , where R o p = ∑ a , b d N ( b N + a , R ) N a N + b R^{\mathrm{op}}=\sum_{a,b}d_N(bN+a,R)N^{aN+b} R op = ∑ a , b d N ( b N + a , R ) N a N + b ;
0 0 0 is an identity if and only if L 0 = C 0 = ι N L_0=C_0=\iota_N L 0 = C 0 = ι N ;
it is a quasigroup if and only if L a , C b ∈ Sym N L_a,C_b\in\operatorname{Sym}_N L a , C b ∈ Sym N for all a , b a,b a , b ;
if it is a quasigroup, then
∑ a d N ( a N + b , R ) = ∑ b d N ( a N + b , R ) = ( N 2 ) , ∑ a , b d N ( a N + b , R ) = N ( N 2 ) . \sum_a d_N(aN+b,R)=\sum_b d_N(aN+b,R)=\binom{N}{2},\qquad\sum_{a,b}d_N(aN+b,R)=N\binom{N}{2}. a ∑ d N ( a N + b , R ) = b ∑ d N ( a N + b , R ) = ( 2 N ) , a , b ∑ d N ( a N + b , R ) = N ( 2 N ) .
Proof. (1)–(3) restate the definitions through Proposition 2.12 . For (4), each row and each column contains every element of A N \mathcal{A}_N A N once.
Theorem 2.15 (Digital Cayley theorem). A loop ( A N , R ) (\mathcal{A}_N,R) ( A N , R ) is a group if and only if
L a ∗ b = L a ∘ L b for all a , b ∈ A N . (16) L_{a*b}=L_a\circ L_b\qquad\text{for all }a,b\in\mathcal{A}_N. \tag*{(16)} L a ∗ b = L a ∘ L b for all a , b ∈ A N . (16) In that case a ↦ L a = d N N ( a , R ) a\mapsto L_a=d_{N^N}(a,R) a ↦ L a = d N N ( a , R ) is an injective homomorphism into Sym N \operatorname{Sym}_N Sym N ; every group of order N N N is a subgroup of Sym N \operatorname{Sym}_N Sym N .
Proof. Associativity for all c c c says exactly λ a ∘ λ b = λ a ∗ b \lambda_a\circ\lambda_b=\lambda_{a*b} λ a ∘ λ b = λ a ∗ b , i.e. (16) by (12) . Injectivity: L a ( 0 ) = a ∗ 0 = a L_a(0)=a*0=a L a ( 0 ) = a ∗ 0 = a .
Proposition 2.16 (Inverses). In a loop, each a a a has a unique right inverse and a unique left inverse, namely a ρ = L a − 1 ( 0 ) = d N ( 0 , ⋅ ) a^\rho=L_a^{-1}(0)=d_N(0,\cdot) a ρ = L a − 1 ( 0 ) = d N ( 0 , ⋅ ) -position of 0 0 0 in L a L_a L a , i.e. a ρ = d N ( 0 , ∑ n N d N ( n , L a ) ) a^\rho=d_N\left(0,\sum_n N^{d_N(n,L_a)}\right) a ρ = d N ( 0 , ∑ n N d N ( n , L a ) ) , and symmetrically with C a C_a C a . In a group they coincide.
Proof. a ∗ x = 0 a*x=0 a ∗ x = 0 has the unique solution x = λ a − 1 ( 0 ) x=\lambda_a^{-1}(0) x = λ a − 1 ( 0 ) , whose code is given by (11) . In a group, a λ = a λ ∗ ( a ∗ a ρ ) = ( a λ ∗ a ) ∗ a ρ = a ρ a^\lambda=a^\lambda*(a*a^\rho)=(a^\lambda*a)*a^\rho=a^\rho a λ = a λ ∗ ( a ∗ a ρ ) = ( a λ ∗ a ) ∗ a ρ = a ρ .
Definition 2.17. A Moufang loop is a loop satisfying the equivalent identities z ∗ ( x ∗ ( z ∗ y ) ) = ( ( z ∗ x ) ∗ z ) ∗ y z \ast(x \ast(z \ast y)) = ((z \ast x) \ast z) \ast y z ∗ ( x ∗ ( z ∗ y )) = (( z ∗ x ) ∗ z ) ∗ y , x ∗ ( z ∗ ( y ∗ z ) ) = ( ( x ∗ z ) ∗ y ) ∗ z x \ast(z \ast(y \ast z)) = ((x \ast z) \ast y) \ast z x ∗ ( z ∗ ( y ∗ z )) = (( x ∗ z ) ∗ y ) ∗ z , ( z ∗ x ) ∗ ( y ∗ z ) = ( z ∗ ( x ∗ y ) ) ∗ z (z \ast x) \ast(y \ast z) = (z \ast(x \ast y)) \ast z ( z ∗ x ) ∗ ( y ∗ z ) = ( z ∗ ( x ∗ y )) ∗ z , ( z ∗ x ) ∗ ( y ∗ z ) = z ∗ ( ( x ∗ y ) ∗ z ) (z \ast x) \ast(y \ast z) = z \ast((x \ast y) \ast z) ( z ∗ x ) ∗ ( y ∗ z ) = z ∗ (( x ∗ y ) ∗ z ) .
Example 2.18. The smallest nonassociative Moufang loop, M ( S 3 , 2 ) M(S_3,2) M ( S 3 , 2 ) , has order 12:
Table 2.18.
Homomorphisms and relabelling
Definition 2.19. A homomorphism f : ( A N , R ) → ( A q , S ) f \colon(A_N,R) \to(A_q,S) f : ( A N , R ) → ( A q , S ) is a map f : A N → A q f \colon A_N \to A_q f : A N → A q with
f ( d N ( a N + b , R ) ) = d q ( f ( a ) q + f ( b ) , S ) for all a , b . (17) f\bigl(d_N(aN+b,R)\bigr)=d_q\bigl(f(a)q+f(b),S\bigr)\qquad\text{for all }a,b. \tag*{(17)} f ( d N ( a N + b , R ) ) = d q ( f ( a ) q + f ( b ) , S ) for all a , b . (17)
A bijective homomorphism is an isomorphism (then q = N q=N q = N ); an isomorphism of a magma onto itself is an automorphism . For groups, ker f = f − 1 ( 0 ) \ker f=f^{-1}(0) ker f = f − 1 ( 0 ) .
Proposition 2.20 (Relabelling). For F ∈ Sym N F\in\operatorname{Sym}_N F ∈ Sym N , the unique code R F R^F R F making F F F an isomorphism ( A N , R ) → ( A N , R F ) (A_N,R)\to(A_N,R^F) ( A N , R ) → ( A N , R F ) is
R F = ∑ a , b = 0 N − 1 d N ( d N ( a N + b , R ) , F ) N d N ( a , F ) N + d N ( b , F ) . (18) R^F=\sum_{a,b=0}^{N-1}d_N\bigl(d_N(aN+b,R),F\bigr)N^{d_N(a,F)N+d_N(b,F)}. \tag*{(18)} R F = a , b = 0 ∑ N − 1 d N ( d N ( a N + b , R ) , F ) N d N ( a , F ) N + d N ( b , F ) . (18)
This is an action , ( R F ) G = R G ∘ F (R^F)^G=R^{G\circ F} ( R F ) G = R G ∘ F , so Aut ( A N , R ) = { F ∈ Sym N : R F = R } \operatorname{Aut}(A_N,R)=\{F\in\operatorname{Sym}_N:R^F=R\} Aut ( A N , R ) = { F ∈ Sym N : R F = R } is a subgroup of Sym N \operatorname{Sym}_N Sym N .
Proof. (17) prescribes the digit of R F R^F R F at position F ( a ) N + F ( b ) F(a)N+F(b) F ( a ) N + F ( b ) ; apply (4) . G ∘ F G\circ F G ∘ F is an isomorphism R → ( R F ) G R\to(R^F)^G R → ( R F ) G , and the code making it so is unique. A stabiliser of an action is a subgroup.
Counting magmas up to symmetry
The relabelling action of Proposition 2.20 moves each digit of a code twice: position a N + b aN+b a N + b is relocated to F ( a ) N + F ( b ) F(a)N+F(b) F ( a ) N + F ( b ) , and the digit value is renamed by F F F . Counting its orbits is therefore a problem of the type treated in § 10.4.4 . For F ∈ Sym N F\in\operatorname{Sym}_N F ∈ Sym N and k ≥ 1 k\geq1 k ≥ 1 let fix k ( F ) \operatorname{fix}_k(F) fix k ( F ) be the number of i ∈ A N i\in A_N i ∈ A N with F k ( i ) = i F^k(i)=i F k ( i ) = i , and for α , β ∈ Sym N \alpha,\beta\in\operatorname{Sym}_N α , β ∈ Sym N let e k ( α , β ) e_k(\alpha,\beta) e k ( α , β ) be the number of cycles of length k k k of the permutation ( a , b ) ↦ ( α ( a ) , β ( b ) ) (a,b)\mapsto(\alpha(a),\beta(b)) ( a , b ) ↦ ( α ( a ) , β ( b )) of A N 2 A_N^2 A N 2 . A cycle of length i i i of α \alpha α and a cycle of length j j j of β \beta β give gcd ( i , j ) \gcd(i,j) g cd( i , j ) cycles of length lcm ( i , j ) \operatorname{lcm}(i,j) lcm ( i , j ) on their product, and this determines e k e_k e k from the cycle types of α \alpha α and β \beta β .
Theorem 2.21 (Magmas up to isomorphism). The number of isomorphism classes of magmas on A N A_N A N is
1 N ! ∑ F ∈ Sym N ∏ k ≥ 1 fix k ( F ) e k ( F , F ) . \frac{1}{N!}\sum_{F\in\operatorname{Sym}_N}\prod_{k\geq1}\operatorname{fix}_k(F)^{e_k(F,F)}. N ! 1 F ∈ Sym N ∑ k ≥ 1 ∏ fix k ( F ) e k ( F , F ) .
Proof. By Proposition 2.20 and (18) , F F F fixes the code R R R if and only if d N ( F ( a ) N + F ( b ) , R ) = F ( d N ( a N + b , R ) ) d_N(F(a)N+F(b),R)=F(d_N(aN+b,R)) d N ( F ( a ) N + F ( b ) , R ) = F ( d N ( a N + b , R )) for all a , b a,b a , b . Along a cycle of length k k k of ( a , b ) ↦ ( F ( a ) , F ( b ) ) (a,b)\mapsto(F(a),F(b)) ( a , b ) ↦ ( F ( a ) , F ( b )) the digit at one position determines all digits on the cycle, and the condition closes up if and only if the starting digit y y y satisfies F k ( y ) = y F^{k}(y)=y F k ( y ) = y . So F F F fixes ∏ k fix k ( F ) e k ( F , F ) \prod_k\operatorname{fix}_k(F)^{e_k(F,F)} ∏ k fix k ( F ) e k ( F , F ) codes, and the claim is Burnside’s lemma (Theorem 4.6 ). □
Two magmas R , S R,S R , S on A N \mathcal{A}_N A N are isotopic if S ( α ( a ) , β ( b ) ) = γ ( R ( a , b ) ) S(\alpha(a),\beta(b))=\gamma(R(a,b)) S ( α ( a ) , β ( b )) = γ ( R ( a , b )) for some α , β , γ ∈ Sym N \alpha,\beta,\gamma\in\operatorname{Sym}_N α , β , γ ∈ Sym N ; isomorphism is the case α = β = γ \alpha=\beta=\gamma α = β = γ . The group Sym N 3 \operatorname{Sym}_N^3 Sym N 3 acts on the codes, and the argument of the proof, with ( α , β ) (\alpha,\beta) ( α , β ) acting on positions and γ \gamma γ on values, is Theorem 4.22 .
Theorem 2.22 (Magmas up to isotopy). The number of isotopy classes of magmas on A N \mathcal{A}_N A N is
1 ( N ! ) 3 ∑ α , β , γ ∈ Sym N ∏ k ≥ 1 fix k ( γ ) e k ( α , β ) . \frac{1}{(N!)^3}\sum_{\alpha,\beta,\gamma\in\operatorname{Sym}_N}\prod_{k\ge1}\operatorname{fix}_k(\gamma)^{e_k(\alpha,\beta)}. ( N ! ) 3 1 α , β , γ ∈ Sym N ∑ k ≥ 1 ∏ fix k ( γ ) e k ( α , β ) .
Proposition 2.23 (Magmas up to opposite). The number of operators on A N \mathcal{A}_N A N up to the exchange R ( a , b ) ↔ R ( b , a ) R(a,b)\leftrightarrow R(b,a) R ( a , b ) ↔ R ( b , a ) , that is, of magmas up to passing to the opposite magma, is 1 2 ( N N 2 + N N ( N + 1 ) / 2 ) \frac{1}{2}(N^{N^2}+N^{N(N+1)/2}) 2 1 ( N N 2 + N N ( N + 1 ) /2 ) .
Proof. The exchange of the two coordinates is an involution of A N 2 \mathcal{A}_N^2 A N 2 with N N N fixed points and N ( N − 1 ) / 2 N(N-1)/2 N ( N − 1 ) /2 two-cycles, hence N ( N + 1 ) / 2 N(N+1)/2 N ( N + 1 ) /2 cycles. Apply Theorem 4.21 to Z 2 \mathbb{Z}_2 Z 2 acting on the N 2 N^2 N 2 positions with N N N colours. □
Table 2.1. Magmas of small order up to symmetry. Verified by scripts/polya_verify.py by formula and, for N ≤ 3 N\le3 N ≤ 3 , by direct orbit enumeration.
Example 2.24. For N = 3 N=3 N = 3 the identity fixes all 3 9 3^9 3 9 codes. A transposition has one fixed point and acts on A 3 2 \mathcal{A}_3^2 A 3 2 with one fixed pair and four 2-cycles, so it fixes fix 1 1 fix 2 4 = 1 ⋅ 3 4 = 81 \operatorname{fix}_1^1\operatorname{fix}_2^4=1\cdot3^4=81 fix 1 1 fix 2 4 = 1 ⋅ 3 4 = 81 codes; a 3-cycle has three 3-cycles on pairs and fixes fix 3 3 = 27 \operatorname{fix}_3^3=27 fix 3 3 = 27 codes. Hence 1 6 ( 19683 + 3 ⋅ 81 + 2 ⋅ 27 ) = 3330 \frac{1}{6}(19683+3\cdot81+2\cdot27)=3330 6 1 ( 19683 + 3 ⋅ 81 + 2 ⋅ 27 ) = 3330 .
Direct products
Definition 2.25. Given magmas ∗ h \ast_h ∗ h on A p h \mathcal{A}_{p_h} A p h , h < ℓ h<\ell h < ℓ , their direct product is the magma on A ∏ h p h \mathcal{A}_{\prod_h p_h} A ∏ h p h
m ∗ n = ∑ h = 0 ℓ − 1 ( D h ( m ) ∗ h D h ( n ) ) ∏ i < h p i h . (19) m\ast n=\sum_{h=0}^{\ell-1}(D_h(m)\ast_hD_h(n))\prod_{i<h}p_i^h. \tag*{(19)} m ∗ n = h = 0 ∑ ℓ − 1 ( D h ( m ) ∗ h D h ( n )) i < h ∏ p i h . (19)
If all factors equal ∗ \ast ∗ on A N \mathcal{A}_N A N , this is m ∗ n = ∑ h ( d N ( h , m ) ∗ d N ( h , n ) ) N h m\ast n=\sum_h(d_N(h,m)\ast d_N(h,n))N^h m ∗ n = ∑ h ( d N ( h , m ) ∗ d N ( h , n )) N h .
Proposition 2.26. Each projection D h D_h D h is a surjective homomorphism onto ∗ h \ast_h ∗ h . The product is associative, commutative, a quasigroup, has identity 0 0 0 , or is a group if and only if every factor has the same property.
Proof. By Lemma 1.6 , D h ( m ∗ n ) = D h ( m ) ∗ h D h ( n ) D_h(m\ast n)=D_h(m)\ast_hD_h(n) D h ( m ∗ n ) = D h ( m ) ∗ h D h ( n ) , so every identity holds in the product exactly when it holds digit by digit. □
r-ary operators
Definition 2.27. An r r r -ary operator on A N A_N A N is a map A N r → A N A_N^r \to A_N A N r → A N , written by Theorem 1.7 as y = d N ( ∑ k < r x k N k , R ) y=d_N\left(\sum_{k<r}x_kN^k,R\right) y = d N ( ∑ k < r x k N k , R ) with code R ∈ [ 0 , N N r ) R\in[0,N^{N^r}) R ∈ [ 0 , N N r ) . The argument x k x_k x k is the k k k -th digit of the index. Magmas are the case r = 2 r=2 r = 2 with x 1 = a x_1=a x 1 = a and x 0 = b x_0=b x 0 = b , as in (14) , so that the left operand is the most significant digit; permutations and other unary maps are the case r = 1 r=1 r = 1 .
Proposition 2.28. The j j j -th projection ( x 0 , … , x r − 1 ) ↦ x j (x_0,\ldots,x_{r-1})\mapsto x_j ( x 0 , … , x r − 1 ) ↦ x j has code ∑ n < N r d N ( j , n ) N n \sum_{n<N^r}d_N(j,n)N^n ∑ n < N r d N ( j , n ) N n . If R R R is r r r -ary and S 0 , … , S r − 1 S_0,\ldots,S_{r-1} S 0 , … , S r − 1 are s s s -ary, then the composite x ↦ R ( S 0 ( x ) , … , S r − 1 ( x ) ) x\mapsto R(S_0(x),\ldots,S_{r-1}(x)) x ↦ R ( S 0 ( x ) , … , S r − 1 ( x )) has code ∑ n < N s d N ( ∑ k < r d N ( n , S k ) N k , R ) N n \sum_{n<N^s}d_N\left(\sum_{k<r}d_N(n,S_k)N^k,R\right)N^n ∑ n < N s d N ( ∑ k < r d N ( n , S k ) N k , R ) N n .
Proof. Both are (4) applied to the map in question, with n = ∑ k x k N k n=\sum_kx_kN^k n = ∑ k x k N k and x j = d N ( j , n ) x_j=d_N(j,n) x j = d N ( j , n ) . □
Clones and identities
Definition 2.29. A clone on A N A_N A N is a family of finitary operators on A N A_N A N that contains all projections and is closed under the composition of Proposition 2.28 . The clone Clo ( R ) \operatorname{Clo}(R) Clo ( R ) of a magma ( A N , R ) (A_N,R) ( A N , R ) is the smallest clone containing ∗ \ast ∗ ; its members are the term operations of the magma, and Clo s ( R ) \operatorname{Clo}_s(R) Clo s ( R ) denotes its s s s -ary part.
Since a ∗ b a\ast b a ∗ b has x 1 = a x_1=a x 1 = a and x 0 = b x_0=b x 0 = b , the term operation S ∗ T S\ast T S ∗ T of two s s s -ary term operations has code R ∘ ( T , S ) R\circ(T,S) R ∘ ( T , S ) in the notation of Proposition 2.28 . Thus Clo s ( R ) \operatorname{Clo}_s(R) Clo s ( R ) is the closure of the s s s projection codes under one explicit integer map.
Proposition 2.30. Let ( A N , R ) (A_N,R) ( A N , R ) be a magma. An identity u = v u=v u = v between two terms in s s s variables holds if and only if the codes of the term operations of u u u and v v v coincide. The set Clo s ( R ) ⊆ [ 0 , N N s ) \operatorname{Clo}_s(R)\subseteq[0,N^{N^s}) Clo s ( R ) ⊆ [ 0 , N N s ) is finite and is computed by finitely many compositions.
Proof. By Theorem 1.7 two s s s -ary operators are equal if and only if their codes are equal. There are at most N N s N^{N^s} N N s codes, so the closure stabilises after finitely many steps. □
Associativity is the case s = 3 s=3 s = 3 , u = ( x ∗ y ) ∗ z u=(x\ast y)\ast z u = ( x ∗ y ) ∗ z , v = x ∗ ( y ∗ z ) v=x\ast(y\ast z) v = x ∗ ( y ∗ z ) ; the two codes are the ternary codes L ( R ) L(R) L ( R ) and R ′ ( R ) R'(R) R ′ ( R ) of (36) , and (38) counts the triples where they differ.
The ternary law of a group
In a group, put
m ( x , y , z ) = x ∗ y − 1 ∗ z . (20) m(x,y,z)=x\ast y^{-1}\ast z. \tag*{(20)} m ( x , y , z ) = x ∗ y − 1 ∗ z . (20)
Then m ( x , y , y ) = x m(x,y,y)=x m ( x , y , y ) = x and m ( x , x , y ) = y m(x,x,y)=y m ( x , x , y ) = y , so m m m is a Mal’cev term; by Mal’cev’s theorem a variety has such a term if and only if any two congruences of each of its algebras permute [46]. The term m m m also satisfies
m ( x , y , y ) = x = m ( y , y , x ) , m ( m ( x , y , z ) , u , v ) = m ( x , y , m ( z , u , v ) ) . (21) m(x,y,y)=x=m(y,y,x),\qquad m(m(x,y,z),u,v)=m(x,y,m(z,u,v)). \tag*{(21)} m ( x , y , y ) = x = m ( y , y , x ) , m ( m ( x , y , z ) , u , v ) = m ( x , y , m ( z , u , v )) . (21)
and a nonempty set with a ternary law satisfying (21) is a heap [47, 48].
Proposition 2.32. Let m m m be a heap law on A N \mathcal{A}_N A N and e ∈ A N e \in\mathcal{A}_N e ∈ A N . Then x ∗ y = m ( x , e , y ) x * y = m(x,e,y) x ∗ y = m ( x , e , y ) is a group law with identity e e e and inverse y − 1 = m ( e , y , e ) y^{-1} = m(e,y,e) y − 1 = m ( e , y , e ) , and m ( x , y , z ) = x ∗ y − 1 ∗ z m(x,y,z) = x * y^{-1} * z m ( x , y , z ) = x ∗ y − 1 ∗ z . Conversely, every group law gives the heap law (20) .
Proof. Associativity of ∗ * ∗ is the second identity of (21) with y = u = e y = u = e y = u = e . The first identities give m ( e , e , y ) = y m(e,e,y) = y m ( e , e , y ) = y and m ( x , e , e ) = x m(x,e,e) = x m ( x , e , e ) = x . For the inverse, y ∗ m ( e , y , e ) = m ( y , e , m ( e , y , e ) ) = m ( m ( y , e , e ) , y , e ) = m ( y , y , e ) = e y * m(e,y,e) = m(y,e,m(e,y,e)) = m(m(y,e,e),y,e) = m(y,y,e) = e y ∗ m ( e , y , e ) = m ( y , e , m ( e , y , e )) = m ( m ( y , e , e ) , y , e ) = m ( y , y , e ) = e , and similarly on the left. Finally x ∗ y − 1 ∗ z = m ( x , e , m ( m ( e , y , e ) , e , z ) ) = m ( x , e , m ( e , y , z ) ) = m ( m ( x , e , e ) , y , z ) = m ( x , y , z ) x * y^{-1} * z = m(x,e,m(m(e,y,e),e,z)) = m(x,e,m(e,y,z)) = m(m(x,e,e),y,z) = m(x,y,z) x ∗ y − 1 ∗ z = m ( x , e , m ( m ( e , y , e ) , e , z )) = m ( x , e , m ( e , y , z )) = m ( m ( x , e , e ) , y , z ) = m ( x , y , z ) . The converse is a direct check.
The heap forgets only the choice of identity: the N N N choices of e e e give isomorphic groups, by the translation x ↦ m ( x , e , e ′ ) x \mapsto m(x,e,e') x ↦ m ( x , e , e ′ ) . In digital form the code of m m m is obtained from the code R R R and the inversion code by two compositions in Proposition 2.28.
Cochains
Let Q Q Q be a finite group acting on a finite abelian group A A A , with ∣ Q ∣ = q |Q| = q ∣ Q ∣ = q and ∣ A ∣ = n |A| = n ∣ A ∣ = n , labelled by A q \mathcal{A}_q A q and A n \mathcal{A}_n A n with identities at 0 0 0 . An r r r -cochain is a map f : Q r → A f : Q^r \to A f : Q r → A ; the set of them is C r ( Q , A ) = A Q r C^r(Q,A) = A^{Q^r} C r ( Q , A ) = A Q r . The proof of Theorem 1.7, with input base q q q and output base n n n , gives
f ( x 0 , … , x r − 1 ) = d n ( ∑ k < r x k q k , F ) , 0 ≤ F < n q r , (22) f(x_0,\ldots,x_{r-1}) = d_n\left(\sum_{k<r} x_kq^k,F\right), \qquad0 \le F < n^{q^r}, \tag*{(22)} f ( x 0 , … , x r − 1 ) = d n ( k < r ∑ x k q k , F ) , 0 ≤ F < n q r , (22)
so ∣ C r ( Q , A ) ∣ = n q r |C^r(Q,A)| = n^{q^r} ∣ C r ( Q , A ) ∣ = n q r , and the normalised cochains, which vanish when some argument is 0 0 0 , number n ( q − 1 ) r n^{(q-1)^r} n ( q − 1 ) r . The carry code F F F of Chapter 3 is (22) for r = 2 r = 2 r = 2 . The low degrees have the following meaning [49]: 1-cocycles are crossed homomorphisms f ( x y ) = f ( x ) + x ⋅ f ( y ) f(xy) = f(x) + x \cdot f(y) f ( x y ) = f ( x ) + x ⋅ f ( y ) , and H 1 ( Q , A ) H^1(Q,A) H 1 ( Q , A ) classifies the complements of A A A in A ⋊ Q A \rtimes Q A ⋊ Q up to conjugacy; 2-cocycles are the factor sets of extension theory, and H 2 ( Q , A ) H^2(Q,A) H 2 ( Q , A ) classifies the extensions with the given action. The action itself is not a cochain: it is the separate datum carried by the action code Φ \Phi Φ . A first example is Proposition 1.23: the carry c N : A N × A N → Z c_N : \mathcal{A}_N \times\mathcal{A}_N \to\mathbb{Z} c N : A N × A N → Z is a normalised 2-cocycle of Z N \mathbb{Z}_N Z N with values in Z \mathbb{Z} Z (trivial action), and the extension it defines is 0 → Z → N Z → Z N → 0 0 \to\mathbb{Z} \xrightarrow{N} \mathbb{Z} \to\mathbb{Z}_N \to0 0 → Z N Z → Z N → 0 , that is, positional notation itself.
Polyadic groups
For r ≥ 3 r \ge3 r ≥ 3 , an r r r -ary group is a set G G G with an r r r -ary law f f f such that the value
f ( x 1 , … , x i , f ( x i + 1 , … , x i + r ) , x i + r + 1 , … , x 2 r − 1 ) f(x_1,\ldots,x_i,f(x_{i+1},\ldots,x_{i+r}),x_{i+r+1},\ldots,x_{2r-1}) f ( x 1 , … , x i , f ( x i + 1 , … , x i + r ) , x i + r + 1 , … , x 2 r − 1 )
does not depend on i ∈ { 0 , … , r − 1 } i \in\{0,\ldots,r-1\} i ∈ { 0 , … , r − 1 } , and every equation f ( a 1 , … , a i − 1 , x , a i + 1 , … , a r ) = b f(a_1,\ldots,a_{i-1},x,a_{i+1},\ldots,a_r) = b f ( a 1 , … , a i − 1 , x , a i + 1 , … , a r ) = b has a unique solution x x x [50]. Post’s coset theorem embeds G G G in a group G ∗ G^* G ∗ with a normal subgroup G 0 G_0 G 0 such that G ∗ / G 0 G^*/G_0 G ∗ / G 0 is cyclic of order r − 1 r-1 r − 1 , G G G is a generating coset of G 0 G_0 G 0 and f f f is the product of r r r factors in G ∗ G^* G ∗ [51].
Theorem 2.33 (Hosszú–Gluskin [52, 53]). Let ( G , f ) (G,f) ( G , f ) be an r r r -ary group, r ≥ 3 r \ge3 r ≥ 3 . There are a group law on G G G , an automorphism θ \theta θ and an element b b b with θ ( b ) = b \theta(b) = b θ ( b ) = b and θ r − 1 ( x ) = b x b − 1 \theta^{r-1}(x) = bxb^{-1} θ r − 1 ( x ) = b x b − 1 for all x x x , such that
f ( x 1 , … , x r ) = x 1 θ ( x 2 ) θ 2 ( x 3 ) ⋯ θ r − 1 ( x r ) b . f(x_1,\ldots,x_r) = x_1\theta(x_2)\theta^2(x_3)\cdots\theta^{r-1}(x_r)b. f ( x 1 , … , x r ) = x 1 θ ( x 2 ) θ 2 ( x 3 ) ⋯ θ r − 1 ( x r ) b . Conversely, any such data define an r r r -ary group.
For example, in an abelian group with r = 3 r = 3 r = 3 , θ ( x ) = − x \theta(x) = -x θ ( x ) = − x and b = 0 b = 0 b = 0 one gets f ( x , y , z ) = x − y + z f(x,y,z) = x - y + z f ( x , y , z ) = x − y + z , which is the heap law (20) . Digitally, Theorem 2.33 replaces a search through all N N r N^{N^r} N N r codes by a search through group codes on A N \mathcal{A}_N A N , automorphisms and elements.
Exercises
Show that the identity permutation code of degree N N N is ι N = ∑ i < N i N i = N ( 1 + ( N − 2 ) N N ) ( N − 1 ) 2 \iota_N = \sum_{i<N} iN^i = \frac{N\left(1+(N-2)N^N\right)}{(N-1)^2} ι N = ∑ i < N i N i = ( N − 1 ) 2 N ( 1 + ( N − 2 ) N N ) , and check ι 3 = 21 \iota_3 = 21 ι 3 = 21 .
Show that Sym 3 = { 5 , 7 , 11 , 15 , 19 , 21 } \operatorname{Sym}_3 = \{5,7,11,15,19,21\} Sym 3 = { 5 , 7 , 11 , 15 , 19 , 21 } , identify the three transpositions, and write the composition table of these codes using (12) .
Show that the sign of a permutation code R R R is ( − 1 ) ι ( R ) (-1)^{\iota(R)} ( − 1 ) ι ( R ) , where ι ( R ) \iota(R) ι ( R ) is the number of pairs i < j i<j i < j with d N ( i , R ) > d N ( j , R ) d_N(i,R)>d_N(j,R) d N ( i , R ) > d N ( j , R ) , and compute the sign of each element of Sym 3 \operatorname{Sym}_3 Sym 3 (compare Theorem 2.8 ).
Show that there are N N 2 N^{N^2} N N 2 magmas on A N A_N A N , of which N N ( N + 1 ) / 2 N^{N(N+1)/2} N N ( N + 1 ) /2 are commutative and N ( N − 1 ) 2 N^{(N-1)^2} N ( N − 1 ) 2 have 0 0 0 as a two-sided identity.
Using the relabelling action of Proposition 2.20, show that there are exactly 10 magmas of order 2 up to isomorphism.
Show that addition modulo 3 is the only group law on A 3 A_3 A 3 with identity 0.
Write the code of the direct product C 2 ⊕ C 2 C_2\oplus C_2 C 2 ⊕ C 2 given by (19) and show that it differs from the code of C 4 C_4 C 4 although both laws live on A 4 A_4 A 4 .
Show that there are exactly 5 magmas on A 2 A_2 A 2 up to isotopy, and give the code of one representative of each class. Explain why the projections a ∗ b = a a*b=a a ∗ b = a and a ∗ b = b a*b=b a ∗ b = b are not isotopic, although they are exchanged by passing to the opposite magma.
Use Theorem 2.21 to show that there are 3330 magmas of order 3 up to isomorphism.
Show that every group law gives a heap by (20) , that is, verify (21) . For a heap law m m m on A N A_N A N and e , e ′ ∈ A N e,e'\in A_N e , e ′ ∈ A N , show that x ↦ m ( x , e , e ′ ) x\mapsto m(x,e,e') x ↦ m ( x , e , e ′ ) is an isomorphism from the group x ∗ e y = m ( x , e , y ) x*_{e}y=m(x,e,y) x ∗ e y = m ( x , e , y ) of Proposition 2.32 onto x ∗ e ′ y = m ( x , e ′ , y ) x*_{e'}y=m(x,e',y) x ∗ e ′ y = m ( x , e ′ , y ) .
Let ( A N , + ) (A_N,+) ( A N , + ) be an abelian group. Show that f ( x , y , z ) = x − y + z f(x,y,z)=x-y+z f ( x , y , z ) = x − y + z and g ( x , y , z ) = x + y + z + 1 g(x,y,z)=x+y+z+1 g ( x , y , z ) = x + y + z + 1 are ternary groups, and find the automorphism θ \theta θ and the element b b b of Theorem 2.33 for each.
Let Q = A = Z 2 Q=A=Z_2 Q = A = Z 2 with trivial action. Show that there are exactly two normalised 2-cochains, give their codes F F F by (22) , show that both are cocycles and that the only normalised coboundary is 0, so that H 2 ( Z 2 , Z 2 ) ≅ Z 2 H^2(Z_2,Z_2)\cong Z_2 H 2 ( Z 2 , Z 2 ) ≅ Z 2 . Identify the two extensions.
Diagrammatic construction of finite groups
The diagrammatic method for discrete operators [54] exposes the layers into which an operator r R N rR_N r R N on N = p 0 p 1 ⋯ p M − 1 N=p_0p_1\cdots p_{M-1} N = p 0 p 1 ⋯ p M − 1 symbols decomposes, and represents the couplings between layers by arrows. In this chapter we specialise the method to groups. The aim is constructive: we want a single, uniform procedure that produces finite groups of large order from a small amount of data, together with an explicit translation of every diagram into digital mathematics, i.e. into a formula built only from the digit function and integer codes. Questions of classification and uniqueness are deliberately left aside.
The main message is the following. For a group, every node of a diagram is a code of a smaller group, and every arrow carries exactly two codes: an action code Φ \Phi Φ and a carry code F F F . Once these codes are given, the code R R R of the whole group follows mechanically, even when R R R itself is far too large to be written down.
Digital labelling and layers
Throughout, a law of composition on A N = { 0 , 1 , … , N − 1 } A_N=\{0,1,\ldots,N-1\} A N = { 0 , 1 , … , N − 1 } is written, as in the chapter on magmas,
a ∗ b = d N ( a N + b , R ) , d N ( k , x ) = ⌊ x N k ⌋ − N ⌊ x N k + 1 ⌋ , (23) a*b=d_N(aN+b,R),\qquad d_N(k,x)=\left\lfloor\frac{x}{N^k}\right\rfloor-N\left\lfloor\frac{x}{N^{k+1}}\right\rfloor, \tag*{(23)} a ∗ b = d N ( a N + b , R ) , d N ( k , x ) = ⌊ N k x ⌋ − N ⌊ N k + 1 x ⌋ , (23)
so that the left operand a a a is the most significant digit of the index a N + b aN+b a N + b and R = ∑ m = 0 N 2 − 1 N m a m R=\sum_{m=0}^{N^2-1}N^m a_m R = ∑ m = 0 N 2 − 1 N m a m with a m = d N ( 1 , m ) ∗ d N ( 0 , m ) a_m=d_N(1,m)*d_N(0,m) a m = d N ( 1 , m ) ∗ d N ( 0 , m ) .
Let N = p 0 p 1 ⋯ p M − 1 N=p_0p_1\cdots p_{M-1} N = p 0 p 1 ⋯ p M − 1 . Every x ∈ A N x\in A_N x ∈ A N has the mixed-radix (digital) expansion
x = ∑ h = 0 M − 1 x ( h ) Π h , x ( h ) = d p h ( 0 , x Π h ) ∈ A p h , Π h = ∏ m = 0 h − 1 p m , (24) x=\sum_{h=0}^{M-1}x^{(h)}\Pi_h,\qquad x^{(h)}=d_{p_h}\left(0,\frac{x}{\Pi_h}\right)\in A_{p_h},\qquad\Pi_h=\prod_{m=0}^{h-1}p_m, \tag*{(24)} x = h = 0 ∑ M − 1 x ( h ) Π h , x ( h ) = d p h ( 0 , Π h x ) ∈ A p h , Π h = m = 0 ∏ h − 1 p m , (24)
where x ( h ) x^{(h)} x ( h ) is the layer projector of x x x on layer h h h . Layer h = 0 h=0 h = 0 is the ground layer, layers h > 0 h>0 h > 0 are lifted layers. We shall also use the upper block of x x x above layer h h h ,
U h ( x ) = ⌊ x Π h + 1 ⌋ ∈ A Q h , Q h = ∏ m = h + 1 M − 1 p m , (25) U_h(x)=\left\lfloor\frac{x}{\Pi_{h+1}}\right\rfloor\in A_{Q_h},\qquad Q_h=\prod_{m=h+1}^{M-1}p_m, \tag*{(25)} U h ( x ) = ⌊ Π h + 1 x ⌋ ∈ A Q h , Q h = m = h + 1 ∏ M − 1 p m , (25)
which collects all digits of x x x on the layers h + 1 , … , M − 1 h+1,\ldots,M-1 h + 1 , … , M − 1 .
Graded operators
The simplest diagrams have no arrows at all. They are characterised by the following notion, which applies to operators of any arity. Recall Π h = ∏ m < h p m \Pi_h = \prod_{m<h} p_m Π h = ∏ m < h p m from Eq. (24) .
Definition 3.2. An r r r -ary operator y = R ( x 0 , … , x r − 1 ) y = R(x_0,\ldots,x_{r-1}) y = R ( x 0 , … , x r − 1 ) on A N A_N A N , N = p 0 ⋯ p M − 1 N = p_0\cdots p_{M-1} N = p 0 ⋯ p M − 1 , is graded (for this ordering of the layers) if for every h h h the output digit y ( h ) y^{(h)} y ( h ) depends only on the input digits x 0 ( h ) , … , x r − 1 ( h ) x_0^{(h)},\ldots,x_{r-1}^{(h)} x 0 ( h ) , … , x r − 1 ( h ) of the same layer, that is, if there are r r r -ary operators A ( h ) A^{(h)} A ( h ) on A p h A_{p_h} A p h , the layer operators , with
y ( h ) = A ( h ) ( x 0 ( h ) , … , x r − 1 ( h ) ) , h = 0 , … , M − 1. y^{(h)} = A^{(h)}\left(x_0^{(h)},\ldots,x_{r-1}^{(h)}\right), \qquad h = 0,\ldots,M-1. y ( h ) = A ( h ) ( x 0 ( h ) , … , x r − 1 ( h ) ) , h = 0 , … , M − 1. Otherwise the layers are coupled .
Proposition 3.3. An r r r -ary operator R R R on A N A_N A N is graded if and only if it is the direct product of r r r -ary operators A ( h ) A^{(h)} A ( h ) on A p h A_{p_h} A p h ,
R ( x 0 , … , x r − 1 ) = ∑ h = 0 M − 1 A ( h ) ( x 0 ( h ) , … , x r − 1 ( h ) ) Π h . R(x_0,\ldots,x_{r-1}) = \sum_{h=0}^{M-1} A^{(h)}\left(x_0^{(h)},\ldots,x_{r-1}^{(h)}\right)\Pi_h. R ( x 0 , … , x r − 1 ) = h = 0 ∑ M − 1 A ( h ) ( x 0 ( h ) , … , x r − 1 ( h ) ) Π h . The layer operators are then determined by R R R : their codes are
A ( h ) = ∑ n = 0 p h r − 1 p h n d p h ( 0 , R ( d p h ( 0 , n ) Π h , … , d p h ( r − 1 , n ) Π h ) Π h ) . A^{(h)} = \sum_{n=0}^{p_h^r-1} p_h^n d_{p_h}\left(0,\frac{R\left(d_{p_h}(0,n)\Pi_h,\ldots,d_{p_h}(r-1,n)\Pi_h\right)}{\Pi_h}\right). A ( h ) = n = 0 ∑ p h r − 1 p h n d p h ( 0 , Π h R ( d p h ( 0 , n ) Π h , … , d p h ( r − 1 , n ) Π h ) ) . For r = 2 r=2 r = 2 and groups this is the direct product of Proposition 2.26.
Proof. If R R R is graded, expanding the output as y = ∑ h y ( h ) Π h y = \sum_h y^{(h)}\Pi_h y = ∑ h y ( h ) Π h gives the displayed formula. Conversely, the formula says precisely that the h h h -th output digit is A ( h ) A^{(h)} A ( h ) applied to the h h h -th input digits. To read off A ( h ) A^{(h)} A ( h ) , evaluate R R R on inputs whose only nonzero digits lie on layer h h h , namely x k = d p h ( k , n ) Π h x_k = d_{p_h}(k,n)\Pi_h x k = d p h ( k , n ) Π h , and take the layer projector of the output; by the coding theorem this gives the n n n -th digit of the code of A ( h ) A^{(h)} A ( h ) .
In the language of diagrams, a graded law is a diagram without arrows: carries and actions are exactly what couple the layers. The carry-free sum ⊕ p \oplus_p ⊕ p of Section 1.3 is graded with M M M layers equal to p p p , while ordinary addition modulo p M p^M p M is coupled through its carries (Proposition 1.22).
The building block
The whole construction rests on one elementary diagram (Figure 3.1 ): a model of order N = n q N = nq N = n q with a hollow ground layer K K K on n n n symbols, a solid lifted layer Q Q Q on q q q symbols, and one arrow from Q Q Q to K K K .
Figure 3.1. The building block. The lifted layer Q Q Q is a layer operator (solid node); the ground layer K K K (hollow node) receives one arrow carrying the action code Φ \Phi Φ and the carry code F F F .
Definition 3.5 (Data of a diagram). The building block is specified by four integers:
The building block
Table 3.1.
Thus Φ \Phi Φ is the code of a map A q × A n → A n A_q \times A_n \to A_n A q × A n → A n (for each A ∈ A q A \in A_q A ∈ A q a unary operator 1 R n 1R_n 1 R n acting on the ground layer), and F F F is the code of a map A q × A q → A n A_q \times A_q \to A_n A q × A q → A n .
For a , b ∈ A N a,b \in A_N a , b ∈ A N we write a ( 0 ) = d n ( 0 , a ) a^{(0)}=d_n(0,a) a ( 0 ) = d n ( 0 , a ) , a ( 1 ) = d q ( 0 , a / n ) a^{(1)}=d_q(0,a/n) a ( 1 ) = d q ( 0 , a / n ) and similarly for b b b .
What the lifted layer does. Being a layer operator, it multiplies its own digits with its own table:
y ( 1 ) = d q ( b ( 1 ) + q a ( 1 ) , Q ) . (26) y^{(1)}=d_q\left(b^{(1)}+qa^{(1)},Q\right). \tag*{(26)} y ( 1 ) = d q ( b ( 1 ) + q a ( 1 ) , Q ) . (26)
What the ground layer does. It multiplies its own digits, twisted by the information arriving through the arrow:
y ( 0 ) = d n ( d n ( b ( 1 ) + q a ( 1 ) , F ) + n d n ( d n ( b ( 0 ) + n a ( 1 ) , Φ ) + n a ( 0 ) , K ) , K ) (27) y^{(0)}=d_n\left(d_n\left(b^{(1)}+qa^{(1)},F\right)+n\,d_n\left(d_n\left(b^{(0)}+na^{(1)},\Phi\right)+na^{(0)},K\right),K\right) \tag*{(27)} y ( 0 ) = d n ( d n ( b ( 1 ) + q a ( 1 ) , F ) + n d n ( d n ( b ( 0 ) + n a ( 1 ) , Φ ) + n a ( 0 ) , K ) , K ) (27)
Read from the inside out: (i) the action arrow relabels the right operand, b ( 0 ) ↦ φ a ( 1 ) ( b ( 0 ) ) b^{(0)}\mapsto\varphi_{a^{(1)}}(b^{(0)}) b ( 0 ) ↦ φ a ( 1 ) ( b ( 0 ) ) ; (ii) the ground layer multiplies, a ( 0 ) ∗ K φ a ( 1 ) ( b ( 0 ) ) a^{(0)}*_{K}\varphi_{a^{(1)}}(b^{(0)}) a ( 0 ) ∗ K φ a ( 1 ) ( b ( 0 ) ) ; (iii) the carry arrow inserts f ( a ( 1 ) , b ( 1 ) ) f(a^{(1)},b^{(1)}) f ( a ( 1 ) , b ( 1 ) ) .
Reassembly. By the decomposition property of the digit function,
a ∗ b = y ( 0 ) + n y ( 1 ) . (28) a*b=y^{(0)}+ny^{(1)}. \tag*{(28)} a ∗ b = y ( 0 ) + n y ( 1 ) . (28)
In ordinary notation, writing a = ( a ( 0 ) , a ( 1 ) ) a=(a^{(0)},a^{(1)}) a = ( a ( 0 ) , a ( 1 ) ) , Eqs. (26) –(28) state
( a ( 0 ) , a ( 1 ) ) ∗ ( b ( 0 ) , b ( 1 ) ) = ( a ( 0 ) ∗ K φ a ( 1 ) ( b ( 0 ) ) ∗ K f ( a ( 1 ) , b ( 1 ) ) , a ( 1 ) ∗ Q b ( 1 ) ) , (29) \left(a^{(0)},a^{(1)}\right)*\left(b^{(0)},b^{(1)}\right)=\left(a^{(0)}*_{K}\varphi_{a^{(1)}}\left(b^{(0)}\right)*_{K}f\left(a^{(1)},b^{(1)}\right),a^{(1)}*_{Q}b^{(1)}\right), \tag*{(29)} ( a ( 0 ) , a ( 1 ) ) ∗ ( b ( 0 ) , b ( 1 ) ) = ( a ( 0 ) ∗ K φ a ( 1 ) ( b ( 0 ) ) ∗ K f ( a ( 1 ) , b ( 1 ) ) , a ( 1 ) ∗ Q b ( 1 ) ) , (29)
which is the general extension of K K K by Q Q Q .
Link with the general arrow formula. In the diagrammatic method a hollow node is described by one code A n ′ A_{n'} A n ′ for each configuration n ′ n' n ′ of the source layer. Here n ′ = b ( 1 ) + q a ( 1 ) ∈ A q 2 n'=b^{(1)}+qa^{(1)}\in A_{q^2} n ′ = b ( 1 ) + q a ( 1 ) ∈ A q 2 and
y ( 0 ) = ∑ n ′ = 0 q 2 − 1 d n ( b ( 0 ) + n a ( 0 ) , A n ′ ) d q ( n ′ − b ( 1 ) − q a ( 1 ) , 1 ) , (30) y^{(0)}=\sum_{n'=0}^{q^2-1}d_n\left(b^{(0)}+na^{(0)},A_{n'}\right)d_q\left(n'-b^{(1)}-qa^{(1)},1\right), \tag*{(30)} y ( 0 ) = n ′ = 0 ∑ q 2 − 1 d n ( b ( 0 ) + n a ( 0 ) , A n ′ ) d q ( n ′ − b ( 1 ) − q a ( 1 ) , 1 ) , (30)
where d q ( k , 1 ) = δ k , 0 d_q(k,1)=\delta_{k,0} d q ( k , 1 ) = δ k , 0 for integer k k k , and
A n ′ = ∑ m = 0 n 2 − 1 n m d n ( d n ( n ′ , F ) + n d n ( d n ( d n ( 0 , m ) + n d q ( 1 , n ′ ) , Φ ) + n d n ( 1 , m ) , K ) , K ) . (31) A_{n'}=\sum_{m=0}^{n^2-1}n^m d_n\left(d_n(n',F)+n d_n\left(d_n\left(d_n(0,m)+n d_q(1,n'),\Phi\right)+n d_n(1,m),K\right),K\right). \tag*{(31)} A n ′ = m = 0 ∑ n 2 − 1 n m d n ( d n ( n ′ , F ) + n d n ( d n ( d n ( 0 , m ) + n d q ( 1 , n ′ ) , Φ ) + n d n ( 1 , m ) , K ) , K ) . (31)
An arbitrary arrow may carry any family of codes A n ′ A_{n'} A n ′ ; the arrows that produce groups are those for which the whole family is generated by the three codes K K K , Φ \Phi Φ , F F F as in Eq. (31) .
Code of the group. Finally, the code of the model is
R = ∑ m = 0 N 2 − 1 N m [ y ( 0 ) ( a , b ) + n y ( 1 ) ( a , b ) ] a = d N ( 1 , m ) , b = d N ( 0 , m ) . (32) R=\sum_{m=0}^{N^2-1}N^m\left[y^{(0)}(a,b)+n y^{(1)}(a,b)\right]_{a=d_N(1,m),\,b=d_N(0,m)}. \tag*{(32)} R = m = 0 ∑ N 2 − 1 N m [ y ( 0 ) ( a , b ) + n y ( 1 ) ( a , b ) ] a = d N ( 1 , m ) , b = d N ( 0 , m ) . (32)
Admissible codes
The four codes define a group when the following conditions hold.
Normalisation of the action: d n ( v , Φ ) = v d_n(v,\Phi)=v d n ( v , Φ ) = v for 0 ≤ v < n 0\leq v<n 0 ≤ v < n (i.e. φ 0 = i d \varphi_0=\mathrm{id} φ 0 = id ).
Normalisation of the carry: d n ( m , F ) = 0 d_n(m,F)=0 d n ( m , F ) = 0 whenever m < q m<q m < q or d q ( 0 , m ) = 0 d_q(0,m)=0 d q ( 0 , m ) = 0 (i.e. f ( 0 , B ) = f ( A , 0 ) = 0 f(0,B)=f(A,0)=0 f ( 0 , B ) = f ( A , 0 ) = 0 ). Then 0 is the neutral element.
Each φ A \varphi_A φ A is an automorphism of K K K :
d n ( d n ( v + n u , K ) + n A , Φ ) = d n ( d n ( v + n A , Φ ) + n d n ( u + n A , Φ ) , K ) . d_n\left(d_n(v+nu,K)+nA,\Phi\right)=d_n\left(d_n(v+nA,\Phi)+n d_n(u+nA,\Phi),K\right). d n ( d n ( v + n u , K ) + n A , Φ ) = d n ( d n ( v + n A , Φ ) + n d n ( u + n A , Φ ) , K ) .
Compatibility of action and carry: φ A ( φ B ( v ) ) ∗ K f ( A , B ) = f ( A , B ) ∗ K φ A ∗ Q B ( v ) \varphi_A(\varphi_B(v))*_K f(A,B)=f(A,B)*_K\varphi_{A*_Q B}(v) φ A ( φ B ( v )) ∗ K f ( A , B ) = f ( A , B ) ∗ K φ A ∗ Q B ( v ) .
Cocycle condition: f ( A , B ) ∗ K f ( A ∗ Q B , C ) = φ A ( f ( B , C ) ) ∗ K f ( A , B ∗ Q C ) f(A,B)*_K f(A*_Q B,C)=\varphi_A(f(B,C))*_K f(A,B*_Q C) f ( A , B ) ∗ K f ( A ∗ Q B , C ) = φ A ( f ( B , C )) ∗ K f ( A , B ∗ Q C ) .
Conditions 4 and 5 become digital identities on substituting φ A ( v ) = d n ( v + n A , Φ ) \varphi_A(v)=d_n(v+nA,\Phi) φ A ( v ) = d n ( v + n A , Φ ) , f ( A , B ) = d n ( B + q A , F ) f(A,B)=d_n(B+qA,F) f ( A , B ) = d n ( B + q A , F ) , u ∗ K v = d n ( v + n u , K ) u*_K v=d_n(v+nu,K) u ∗ K v = d n ( v + n u , K ) and A ∗ Q B = d q ( B + q A , Q ) A*_Q B=d_q(B+qA,Q) A ∗ Q B = d q ( B + q A , Q ) . They are exactly what is obtained by expanding ( a ∗ b ) ∗ c = a ∗ ( b ∗ c ) (a*b)*c=a*(b*c) ( a ∗ b ) ∗ c = a ∗ ( b ∗ c ) with Eq. (29) , first with b ( 0 ) = c ( 0 ) = 0 b^{(0)}=c^{(0)}=0 b ( 0 ) = c ( 0 ) = 0 and then with c ( 1 ) = 0 c^{(1)}=0 c ( 1 ) = 0 .
Remark 3.6. For construction purposes it is often simplest to impose 1 and 2 and to test associativity of the resulting table directly. Different admissible codes may produce the same group: changing the labels of K K K by a map depending on the Q Q Q -digit changes F F F without changing the group.
Kinds of arrows
Let Φ i d = ∑ A = 0 q − 1 ∑ v = 0 n − 1 v n v + n A \Phi_{\mathrm{id}}=\sum_{A=0}^{q-1}\sum_{v=0}^{n-1}v n^{v+nA} Φ id = ∑ A = 0 q − 1 ∑ v = 0 n − 1 v n v + n A denote the trivial action code.
Table 2.
When K = Z n K=\mathbb{Z}_n K = Z n and φ A ( v ) = λ A v \varphi_A(v)=\lambda_Av φ A ( v ) = λ A v is multiplication by a unit, Eq. (27) simplifies to
y ( 0 ) = d n ( 0 , a ( 0 ) + λ a ( 1 ) b ( 0 ) + d n ( b ( 1 ) + q a ( 1 ) , F ) ) , (33) y^{(0)}=d_n\left(0,a^{(0)}+\lambda_{a^{(1)}}b^{(0)}+d_n\left(b^{(1)}+qa^{(1)},F\right)\right), \tag*{(33)} y ( 0 ) = d n ( 0 , a ( 0 ) + λ a ( 1 ) b ( 0 ) + d n ( b ( 1 ) + q a ( 1 ) , F ) ) , (33)
since d n ( 0 , x ) d_n(0,x) d n ( 0 , x ) is the residue of x x x modulo n n n .
Stacking layers: rank trees
Larger groups are obtained by using an already built block as the solid node Q Q Q of a new block. For N = p 0 p 1 ⋯ p M − 1 N = p_0p_1\cdots p_{M-1} N = p 0 p 1 ⋯ p M − 1 the top layer h = M − 1 h = M - 1 h = M − 1 is solid, and each layer h < M − 1 h < M - 1 h < M − 1 is a hollow node with code K h K_h K h receiving arrows from the upper block U h U_h U h , with an action code Φ h ∈ [ 0 , p h Q h ) \Phi_h \in[0,p_h^{Q_h}) Φ h ∈ [ 0 , p h Q h ) and a carry code F h ∈ [ 0 , p h Q h ) F_h \in[0,p_h^{Q_h}) F h ∈ [ 0 , p h Q h ) . The layer outputs are
y ( h ) = d p h ( d p h ( U h ( b ) + Q h U h ( a ) , F h ) + p h d p h ( d p h ( b ( h ) + p h U h ( a ) , Φ h ) + p h a ( h ) , K h ) , K h ) (34) y^{(h)} = d_{p_h}\left(d_{p_h}(U_h(b) + Q_hU_h(a),F_h) + p_hd_{p_h}\left(d_{p_h}(b^{(h)} + p_hU_h(a),\Phi_h) + p_ha^{(h)},K_h\right),K_h\right) \tag*{(34)} y ( h ) = d p h ( d p h ( U h ( b ) + Q h U h ( a ) , F h ) + p h d p h ( d p h ( b ( h ) + p h U h ( a ) , Φ h ) + p h a ( h ) , K h ) , K h ) (34)
and
a ∗ b = ∑ h = 0 M − 1 y ( h ) Π h . (35) a * b = \sum_{h=0}^{M-1} y^{(h)}\Pi_h. \tag*{(35)} a ∗ b = h = 0 ∑ M − 1 y ( h ) Π h . (35)
For the top layer Q M − 1 = 1 Q_{M-1} = 1 Q M − 1 = 1 , F M − 1 = 0 F_{M-1} = 0 F M − 1 = 0 and Φ M − 1 \Phi_{M-1} Φ M − 1 is trivial, so that Eq. (34) reduces to the layer operator y ( M − 1 ) = d p M − 1 ( b ( M − 1 ) + p M − 1 a ( M − 1 ) , K M − 1 ) y^{(M-1)} = d_{p_{M-1}}(b^{(M-1)} + p_{M-1}a^{(M-1)},K_{M-1}) y ( M − 1 ) = d p M − 1 ( b ( M − 1 ) + p M − 1 a ( M − 1 ) , K M − 1 ) . The evaluation runs from the top down: y ( h ) y^{(h)} y ( h ) only needs the digits of the layers h ′ ≥ h h' \ge h h ′ ≥ h . An arrow is drawn from layer h ′ > h h' > h h ′ > h into layer h h h if and only if Φ h \Phi_h Φ h or F h F_h F h actually depends on the digit d p h ′ ( 0 , U h / ∏ m = h + 1 h ′ − 1 p m ) d_{p_{h'}}(0,U_h/\prod_{m=h+1}^{h'-1}p_m) d p h ′ ( 0 , U h / ∏ m = h + 1 h ′ − 1 p m ) of the upper block. Any node may itself be a composite block (for instance a non-abelian simple group), which is then treated as a single node with its own code.
Worked examples
Example 3.7 (Cyclic groups Z p 2 \mathbb{Z}_{p^2} Z p 2 : carry arrows). Take n = q = p n = q = p n = q = p , K = Q K = Q K = Q = the code of Z p \mathbb{Z}_p Z p , Φ = Φ i d \Phi= \Phi_{\mathrm{id}} Φ = Φ id and d p ( B + p A , F ) = ⌊ ( A + B ) / p ⌋ d_p(B + pA,F) = \lfloor(A+B)/p\rfloor d p ( B + p A , F ) = ⌊( A + B ) / p ⌋ . Then
y ( 1 ) = d p ( 0 , a ( 1 ) + b ( 1 ) ) , y ( 0 ) = d p ( 0 , a ( 0 ) + b ( 0 ) + ⌊ a ( 1 ) + b ( 1 ) p ⌋ ) y^{(1)} = d_p(0,a^{(1)} + b^{(1)}), \qquad y^{(0)} = d_p\left(0,a^{(0)} + b^{(0)} + \left\lfloor\frac{a^{(1)} + b^{(1)}}{p}\right\rfloor\right) y ( 1 ) = d p ( 0 , a ( 1 ) + b ( 1 ) ) , y ( 0 ) = d p ( 0 , a ( 0 ) + b ( 0 ) + ⌊ p a ( 1 ) + b ( 1 ) ⌋ )
and x ↦ x ( 1 ) + p x ( 0 ) x \mapsto x^{(1)} + px^{(0)} x ↦ x ( 1 ) + p x ( 0 ) is an isomorphism onto Z p 2 \mathbb{Z}_{p^2} Z p 2 : the carry arrow is nothing but the carry of base-p p p addition. Iterating with three layers gives Z 27 \mathbb{Z}_{27} Z 27 , with a code R R R of 1044 decimal digits.
Example 3.8 (Dihedral and metacyclic groups: action arrows). For D n D_n D n take K = Z n K = \mathbb{Z}_n K = Z n , Q = Z 2 Q = \mathbb{Z}_2 Q = Z 2 , F = 0 F = 0 F = 0 and Φ = ∑ v = 0 n − 1 v n v + ∑ v = 0 n − 1 d n ( 0 , n − v ) n n + v \Phi= \sum_{v=0}^{n-1}v\,n^v + \sum_{v=0}^{n-1}d_n(0,n-v)n^{n+v} Φ = ∑ v = 0 n − 1 v n v + ∑ v = 0 n − 1 d n ( 0 , n − v ) n n + v , i.e. φ 1 ( v ) = − v \varphi_1(v) = -v φ 1 ( v ) = − v . By Eq. (33) , y ( 0 ) = d n ( 0 , a ( 0 ) + ( − 1 ) a ( 1 ) b ( 0 ) ) y^{(0)} = d_n(0,a^{(0)} + (-1)^{a^{(1)}}b^{(0)}) y ( 0 ) = d n ( 0 , a ( 0 ) + ( − 1 ) a ( 1 ) b ( 0 ) ) . With K = Z 7 K = \mathbb{Z}_7 K = Z 7 , Q = Z 3 Q = \mathbb{Z}_3 Q = Z 3 and d 7 ( v + 7 A , Φ ) = d 7 ( 0 , 2 A v ) d_7(v + 7A,\Phi) = d_7(0,2^Av) d 7 ( v + 7 A , Φ ) = d 7 ( 0 , 2 A v ) one obtains the Frobenius group Z 7 ⋊ Z 3 \mathbb{Z}_7 \rtimes\mathbb{Z}_3 Z 7 ⋊ Z 3 of order 21.
Example 3.9 (The quaternion group: action and carry). Take n = 4 n = 4 n = 4 , q = 2 q = 2 q = 2 , K K K = the code of Z 4 \mathbb{Z}_4 Z 4 , Q = 6 Q = 6 Q = 6 (the code of Z 2 \mathbb{Z}_2 Z 2 ), and
Φ = 27876 , F = 128. \Phi= 27876, \qquad F = 128. Φ = 27876 , F = 128.
The base-4 digits of Φ \Phi Φ are ( 0 , 1 , 2 , 3 , 0 , 3 , 2 , 1 ) (0,1,2,3,0,3,2,1) ( 0 , 1 , 2 , 3 , 0 , 3 , 2 , 1 ) (identity, then v ↦ − v v \mapsto-v v ↦ − v ) and the only non-zero digit of F F F is d 4 ( 3 , F ) = f ( 1 , 1 ) = 2 d_4(3,F)=f(1,1)=2 d 4 ( 3 , F ) = f ( 1 , 1 ) = 2 . Equations (26) and (33) give
y ( 1 ) = d 2 ( 0 , a ( 1 ) + b ( 1 ) ) , y ( 0 ) = d 4 ( 0 , a ( 0 ) + ( − 1 ) a ( 1 ) b ( 0 ) + 2 a ( 1 ) b ( 1 ) ) y^{(1)}=d_2(0,a^{(1)}+b^{(1)}), \qquad y^{(0)}=d_4(0,a^{(0)}+(-1)^{a^{(1)}}b^{(0)}+2a^{(1)}b^{(1)}) y ( 1 ) = d 2 ( 0 , a ( 1 ) + b ( 1 ) ) , y ( 0 ) = d 4 ( 0 , a ( 0 ) + ( − 1 ) a ( 1 ) b ( 0 ) + 2 a ( 1 ) b ( 1 ) )
and a ∗ b = y ( 0 ) + 4 y ( 1 ) a*b=y^{(0)}+4y^{(1)} a ∗ b = y ( 0 ) + 4 y ( 1 ) is Q 8 Q_8 Q 8 (one element of order 2, six of order 4).
Example 3.10 (A 4 A_4 A 4 : a composite node). Label the Klein group Z 2 2 \mathbb{Z}_2^2 Z 2 2 by v = v 0 + 2 v 1 v=v_0+2v_1 v = v 0 + 2 v 1 , so that its code K K K is that of v ∗ w = d 2 ( 0 , v 0 + w 0 ) + 2 d 2 ( 0 , v 1 + w 1 ) v*w=d_2(0,v_0+w_0)+2d_2(0,v_1+w_1) v ∗ w = d 2 ( 0 , v 0 + w 0 ) + 2 d 2 ( 0 , v 1 + w 1 ) . The matrix M ( v 0 , v 1 ) = ( v 1 , v 0 + v 1 ) M(v_0,v_1)=(v_1,v_0+v_1) M ( v 0 , v 1 ) = ( v 1 , v 0 + v 1 ) acts as the permutation ( 0 , 1 , 2 , 3 ) ↦ ( 0 , 2 , 3 , 1 ) (0,1,2,3)\mapsto(0,2,3,1) ( 0 , 1 , 2 , 3 ) ↦ ( 0 , 2 , 3 , 1 ) . With Q = Z 3 Q=\mathbb{Z}_3 Q = Z 3 , F = 0 F=0 F = 0 and Φ \Phi Φ the code with base-4 digits ( 0 , 1 , 2 , 3 , 0 , 2 , 3 , 1 , 0 , 3 , 1 , 2 ) (0,1,2,3,0,2,3,1,0,3,1,2) ( 0 , 1 , 2 , 3 , 0 , 2 , 3 , 1 , 0 , 3 , 1 , 2 ) (i.e. φ A = M A \varphi_A=M^A φ A = M A ) the block is A 4 A_4 A 4 .
Example 3.11 (S 4 S_4 S 4 : three layers). First build S 3 S_3 S 3 from K = Z 3 K=\mathbb{Z}_3 K = Z 3 , Q = Z 2 Q=\mathbb{Z}_2 Q = Z 2 with φ 1 ( v ) = − v \varphi_1(v)=-v φ 1 ( v ) = − v ; an element u = u 0 + 3 u 1 u=u_0+3u_1 u = u 0 + 3 u 1 of this block is r u 0 s u 1 r^{u_0}s^{u_1} r u 0 s u 1 . Then use it as the upper block of the Klein layer, with φ u = M u 0 S u 1 \varphi_u=M^{u_0}S^{u_1} φ u = M u 0 S u 1 , where S ( v 0 , v 1 ) = ( v 1 , v 0 ) S(v_0,v_1)=(v_1,v_0) S ( v 0 , v 1 ) = ( v 1 , v 0 ) acts as ( 0 , 1 , 2 , 3 ) ↦ ( 0 , 2 , 1 , 3 ) (0,1,2,3)\mapsto(0,2,1,3) ( 0 , 1 , 2 , 3 ) ↦ ( 0 , 2 , 1 , 3 ) , and all carries zero. Eq. (34) with p 0 = 4 p_0=4 p 0 = 4 , p 1 = 3 p_1=3 p 1 = 3 , p 2 = 2 p_2=2 p 2 = 2 gives S 4 S_4 S 4 (Figure 3.2 ); its code R R R has 794 decimal digits, while the diagram is specified by a handful of small codes.
Figure 3.2. A three-layer diagram for S 4 S_4 S 4 (Example 3.11). Only action arrows occur.
Example 3.12 (S 5 S_5 S 5 : a simple group as a node). Let K K K be the code of A 5 A_5 A 5 on 60 symbols (for instance, from a labelling of the even permutations of five letters), Q = Z 2 Q=\mathbb{Z}_2 Q = Z 2 , F = 0 F=0 F = 0 , and let φ 1 \varphi_1 φ 1 be conjugation by a transposition. The block is S 5 S_5 S 5 , of order 120; its code R R R has 29940 decimal digits.
All the codes above were evaluated with Eqs. (26) –(32) and (34) –(35) ; in each case the resulting R R R reproduces, through Eq. (23) , an associative table with the expected element orders:
Table 3.
Example 3.13 (All diagrams of three Z 2 \mathbb{Z}_2 Z 2 layers). Take p 0 = p 1 = p 2 = 2 p_0=p_1=p_2=2 p 0 = p 1 = p 2 = 2 with only carry arrows (the only automorphism of Z 2 \mathbb{Z}_2 Z 2 is the identity). After normalisation, the carry into layer 1 has one free digit and the carry into layer 0 has nine, giving 1024 diagrams. Exactly 24 of them are associative, and they produce all groups of order 8: Z 2 3 \mathbb{Z}_2^3 Z 2 3 (2 diagrams), Z 4 × Z 2 \mathbb{Z}_4\times\mathbb{Z}_2 Z 4 × Z 2 (10), Z 8 \mathbb{Z}_8 Z 8 (4), D 4 D_4 D 4 (6) and Q 8 Q_8 Q 8 (2).
Mixed layers and simple groups
In the diagrammatic method a layer is nothing but a projector on a digit of the mixed-radix representation, with a radix that divides the order of the group, and an arrow means that layers are mixed: an incoming arrow indicates that a layer contains material of another layer. Every group of order N = p 0 p 1 ⋯ p M − 1 N=p_0p_1\cdots p_{M-1} N = p 0 p 1 ⋯ p M − 1 can be labelled in this way for every factorisation of N N N and for every bijection between its elements and A N A_N A N . The arrows depend on the labelling, and the task of the method is to find labellings with as few arrows as possible. This point of view yields a characterisation of simple groups by their diagrams.
Arrows as mixing
Definition 3.14 (Arrow). Let a ∗ b = d N ( a N + b , R ) a*b=d_N(aN+b,R) a ∗ b = d N ( a N + b , R ) , let N = p 0 p 1 ⋯ p M − 1 N=p_0p_1\cdots p_{M-1} N = p 0 p 1 ⋯ p M − 1 , and let x ( h ) x^{(h)} x ( h ) denote the layer projectors (24) . For j ≠ h j\ne h j = h there is an arrow j → h j\to h j → h if the output layer y ( h ) = ( a ∗ b ) ( h ) y^{(h)}=(a*b)^{(h)} y ( h ) = ( a ∗ b ) ( h ) , regarded as a function of the 2 M 2M 2 M digits a ( m ) , b ( m ) a^{(m)},b^{(m)} a ( m ) , b ( m ) , is not constant in a ( j ) a^{(j)} a ( j ) or not constant in b ( j ) b^{(j)} b ( j ) . A layer receiving no arrow is a solid node, a layer receiving arrows is a hollow node. A set S S S of layers is closed if there is no arrow j → h j\to h j → h with j ∉ S j\notin S j ∈ / S and h ∈ S h\in S h ∈ S .
For S ⊆ { 0 , … , M − 1 } S \subseteq\{0,\ldots,M-1\} S ⊆ { 0 , … , M − 1 } let π S \pi_S π S be the projector that keeps the digits of the layers in S S S , and let p S = ∏ h ∈ S p h p_S=\prod_{h\in S}p_h p S = ∏ h ∈ S p h ; π S \pi_S π S maps A N A_N A N onto a set of p S p_S p S digit strings, which we label in mixed radix by A p S A_{p_S} A p S .
Proposition 3.15 (Closed sets of layers are quotients). If S S S is closed, then π S ( a ∗ b ) \pi_S(a*b) π S ( a ∗ b ) depends only on π S ( a ) \pi_S(a) π S ( a ) and π S ( b ) \pi_S(b) π S ( b ) , and
R S = ∑ m = 0 p S 2 − 1 p S m π S ( a ∗ b ) , π S ( a ) = d p S ( 1 , m ) , π S ( b ) = d p S ( 0 , m ) , R_S=\sum_{m=0}^{p_S^2-1}p_S^m\pi_S(a*b),\qquad\pi_S(a)=d_{p_S}(1,m),\quad\pi_S(b)=d_{p_S}(0,m), R S = m = 0 ∑ p S 2 − 1 p S m π S ( a ∗ b ) , π S ( a ) = d p S ( 1 , m ) , π S ( b ) = d p S ( 0 , m ) , is the code of a group on p S p_S p S symbols. The projector π S \pi_S π S is a surjective homomorphism, and its kernel K S K_S K S is a normal subgroup of order N / p S N/p_S N / p S .
Proof. Changing the digits of a a a and b b b outside S S S one at a time does not change any layer of S S S , because no arrow enters S S S from outside; hence π S ( a ∗ b ) \pi_S(a*b) π S ( a ∗ b ) depends only on π S ( a ) , π S ( b ) \pi_S(a),\pi_S(b) π S ( a ) , π S ( b ) and R S R_S R S is well defined. By construction π S ( a ∗ b ) = π S ( a ) ∗ S π S ( b ) \pi_S(a*b)=\pi_S(a)*_S\pi_S(b) π S ( a ∗ b ) = π S ( a ) ∗ S π S ( b ) , so ∗ S *_S ∗ S inherits associativity, neutral element and inverses, and π S \pi_S π S is a surjective homomorphism. Its kernel is normal and all its fibres have N / p S N/p_S N / p S elements. □
Conversely, let K K K be a normal subgroup and label x = x ( 0 ) + ∣ K ∣ x ( 1 ) x=x^{(0)}+|K|x^{(1)} x = x ( 0 ) + ∣ K ∣ x ( 1 ) , where x ( 1 ) x^{(1)} x ( 1 ) is the coset x K xK x K (in any numbering) and x ( 0 ) x^{(0)} x ( 0 ) the position of x x x inside its coset. Since ( a b ) K = ( a K ) ( b K ) (ab)K=(aK)(bK) ( ab ) K = ( a K ) ( b K ) , the top layer receives no arrow: it is a solid node with the code of G / K G/K G / K .
Lemma 3.16. A directed graph on M ≥ 2 M\ge2 M ≥ 2 vertices has a closed set of vertices other than ∅ \varnothing ∅ and the whole set if and only if it is not strongly connected.
Proof. If the graph is not strongly connected there are vertices u , v u,v u , v such that v v v cannot be reached from u u u . The set of vertices from which v v v can be reached contains v v v , not u u u , and no arrow enters it from outside, since otherwise its source could reach v v v . Conversely, no arrow leads from the complement of a closed proper set into it, so its vertices cannot be reached from the complement. □
Theorem 3.17 (Simple groups). A finite group G G G of composite order is simple if and only if, for every factorisation ∣ G ∣ = p 0 ⋯ p M − 1 |G|=p_0\cdots p_{M-1} ∣ G ∣ = p 0 ⋯ p M − 1 with M ≥ 2 M\ge2 M ≥ 2 and every labelling of its elements, the arrow graph is strongly connected.
Proof. If some arrow graph is not strongly connected, Lemma 3.16 gives a closed set S S S with ∅ ≠ S ≠ { 0 , … , M − 1 } \varnothing\ne S\ne\{0,\ldots,M-1\} ∅ = S = { 0 , … , M − 1 } . Since all p h ≥ 2 p_h\ge2 p h ≥ 2 , we have 1 < p S < ∣ G ∣ 1<p_S<|G| 1 < p S < ∣ G ∣ , and Proposition 3.15 gives a normal subgroup K S K_S K S of order ∣ G ∣ / p S |G|/p_S ∣ G ∣/ p S , neither trivial nor equal to G G G . Conversely, a normal subgroup K ≠ 1 , G K\ne1,G K = 1 , G gives, with the labelling described above and p 0 = ∣ K ∣ p_0=|K| p 0 = ∣ K ∣ , p 1 = ∣ G / K ∣ p_1=|G/K| p 1 = ∣ G / K ∣ , a top layer without incoming arrows. □
Thus the layers of a simple group can never be separated: whatever the factorisation and the labelling, the mixing between layers can be reduced but not removed. For groups of prime order there is only one layer and the statement is empty.
Subgroup nodes and tower labellings
Subgroups are a weaker notion than closed sets of layers. If the elements whose digits vanish on all layers outside a set L L L are closed under the product, they form a subgroup, but this only constrains the product when both inputs have those digits equal to zero, and mixing is still allowed. Simple groups therefore have plenty of subgroup nodes, while they never have closed sets of layers.
A chain of subgroups 1 = H 0 < H 1 < ⋯ < H M = G 1=H_0<H_1<\cdots<H_M=G 1 = H 0 < H 1 < ⋯ < H M = G gives a tower labelling. Choose for each k k k a transversal of the left cosets of H k H_k H k in H k + 1 H_{k+1} H k + 1 whose first element is the identity, and write every element uniquely as g = t M − 1 t M − 2 ⋯ t 0 g=t_{M-1}t_{M-2}\cdots t_0 g = t M − 1 t M − 2 ⋯ t 0 with t k t_k t k in the k k k -th transversal. The digit x ( k ) x^{(k)} x ( k ) is the index of t k t_k t k , with radix p k = [ H k + 1 : H k ] p_k=[H_{k+1}:H_k] p k = [ H k + 1 : H k ] . In this labelling H h + 1 H_{h+1} H h + 1 is exactly the set of elements with U h ( x ) = 0 U_h(x)=0 U h ( x ) = 0 , and U h ( x ) U_h(x) U h ( x ) labels the left coset x H h + 1 xH_{h+1} x H h + 1 .
Proposition 3.18. In a tower labelling, U h ( a ∗ b ) U_h(a*b) U h ( a ∗ b ) depends on b b b only through U h ( b ) U_h(b) U h ( b ) . Hence the right operand sends arrows only downwards, and every upward arrow comes from the left operand. The upper block above layer h h h is closed if and only if H h + 1 H_{h+1} H h + 1 is normal in G G G .
Proof. U h ( a ∗ b ) U_h(a*b) U h ( a ∗ b ) labels a b H h + 1 = a ( b H h + 1 ) abH_{h+1}=a\left(bH_{h+1}\right) ab H h + 1 = a ( b H h + 1 ) , which depends on a a a and on the coset b H h + 1 bH_{h+1} b H h + 1 , i.e. on U h ( b ) U_h(b) U h ( b ) . If H h + 1 H_{h+1} H h + 1 is normal, a b H h + 1 = ( a H h + 1 ) ( b H h + 1 ) abH_{h+1}=(aH_{h+1})(bH_{h+1}) ab H h + 1 = ( a H h + 1 ) ( b H h + 1 ) and no arrow enters the upper block. If the upper block is closed, Proposition 3.15 shows that H h + 1 H_{h+1} H h + 1 , the kernel of x ↦ U h ( x ) x\mapsto U_h(x) x ↦ U h ( x ) , is normal.
In a tower labelling, the upward arrows are given by the action of G G G on the cosets of the subgroups of the chain; changing the transversals only relabels the cosets and cannot remove them unless the subgroups are normal.
Examples
The arrows of the following labellings were computed exactly from the full multiplication tables, for S 4 S_4 S 4 and A 5 A_5 A 5 (radices listed from the ground layer up):
Table 1 (PDF p. 40).
In the tower labelling of S 4 S_4 S 4 , layer 3 is closed (the sign homomorphism onto Z 2 \mathbb{Z}_2 Z 2 ) and layers { 2 , 3 } \{2,3\} { 2 , 3 } are closed (the quotient S 4 / V ≅ S 3 S_4/V\cong S_3 S 4 / V ≅ S 3 ), in agreement with Proposition 3.15. In A 5 A_5 A 5 every layer contains material of every other layer. The random labellings show that strong connectivity is the generic situation, also for groups that are not simple: Theorem 3.17 states that for simple groups no labelling escapes it.
Example 3.19 (The tower of M 11 M_{11} M 11 ). The chain 1 < Z 2 < Z 4 < Q 8 < M 9 < M 10 < M 11 1<\mathbb{Z}_2<\mathbb{Z}_4<Q_8<M_9<M_{10}<M_{11} 1 < Z 2 < Z 4 < Q 8 < M 9 < M 10 < M 11 of point stabilisers (with Q 8 Q_8 Q 8 refined by its cyclic subgroups) gives the radices 2 , 2 , 2 , 9 , 10 , 11 2,2,2,9,10,11 2 , 2 , 2 , 9 , 10 , 11 . Inside the Q 8 Q_8 Q 8 node the two upper layers receive no arrows and the ground layer receives a carry, which is the diagram of the quaternion group obtained in the worked examples of this chapter. In a sample of 3000 random pairs, every layer of M 9 M_9 M 9 , M 10 M_{10} M 10 and M 11 M_{11} M 11 above the Q 8 Q_8 Q 8 node receives arrows from all lower layers, through the left operand, as required by Proposition 3.18 and Theorem 3.17 .
A gallery of recipes
Generalised quaternion groups Q 2 k Q_{2^k} Q 2 k : K = Z 2 k − 1 K=\mathbb{Z}_{2^{k-1}} K = Z 2 k − 1 , Q = Z 2 Q=\mathbb{Z}_2 Q = Z 2 , φ 1 ( v ) = − v \varphi_1(v)=-v φ 1 ( v ) = − v , f ( 1 , 1 ) = 2 k − 2 f(1,1)=2^{k-2} f ( 1 , 1 ) = 2 k − 2 .
Heisenberg groups of order p 3 p^3 p 3 : Q = Z p × Z p Q=\mathbb{Z}_p\times\mathbb{Z}_p Q = Z p × Z p (a graded block, no internal arrows), K = Z p K=\mathbb{Z}_p K = Z p , trivial action, f ( ( x 1 , y 1 ) , ( x 2 , y 2 ) ) = d p ( 0 , x 1 y 2 ) f((x_1,y_1),(x_2,y_2))=d_p(0,x_1y_2) f (( x 1 , y 1 ) , ( x 2 , y 2 )) = d p ( 0 , x 1 y 2 ) .
Central extensions of a simple node : K = Z 2 K=\mathbb{Z}_2 K = Z 2 , Q = A 5 Q=A_5 Q = A 5 , trivial action and a non-trivial normalised cocycle F F F give the double cover S L ( 2 , 5 ) SL(2,5) S L ( 2 , 5 ) of order 120.
Wreath products: K = A 5 × A 5 K = A_5 \times A_5 K = A 5 × A 5 (graded block on 3600 symbols, v = v 0 + 60 v 1 v = v_0 + 60v_1 v = v 0 + 60 v 1 ), Q = Z 2 Q = \mathbb{Z}_2 Q = Z 2 , F = 0 F = 0 F = 0 and φ 1 ( v 0 + 60 v 1 ) = v 1 + 60 v 0 \varphi_1(v_0 + 60v_1) = v_1 + 60v_0 φ 1 ( v 0 + 60 v 1 ) = v 1 + 60 v 0 give A 5 ≀ Z 2 A_5 \wr\mathbb{Z}_2 A 5 ≀ Z 2 , of order 7200.
Simple groups: by Theorem 3.17 their arrow graph is strongly connected for every labelling, so no set of layers is ever closed. They are generated from permutation codes (Proposition 3.21) or from matrix groups acting on layered vectors, e.g. G L ( 3 , 2 ) ≅ P S L ( 2 , 7 ) GL(3,2) \cong PSL(2,7) G L ( 3 , 2 ) ≅ P S L ( 2 , 7 ) acting on A 8 ≅ F 2 3 A_8 \cong\mathbb{F}_2^3 A 8 ≅ F 2 3 , and their subgroup chains give tower labellings with subgroup nodes and mixed layers (§ 8.8 ).
Associativity by construction
Associativity is the only axiom that involves three elements. Since ternary operators can be built from binary operators acting twice, associativity can be formulated as an identity between two ternary codes. This formulation suggests two ways of constructing associative codes instead of searching for them: through the rows of the table, which are unary codes, and through the carry arrows of a diagram, which are solutions of linear equations.
Associativity as an identity between ternary codes
Let a ∗ b = d N ( a N + b , R ) a*b=d_N(aN+b,R) a ∗ b = d N ( a N + b , R ) . Composing the binary operator with itself in the two possible ways gives the ternary operators
T L ( a , b , c ) = d N ( c + N d N ( b + N a , R ) , R ) , T R ( a , b , c ) = d N ( d N ( c + N b , R ) + N a , R ) , (36) T_L(a,b,c)=d_N(c+Nd_N(b+Na,R),R), \qquad T_R(a,b,c)=d_N(d_N(c+Nb,R)+Na,R), \tag*{(36)} T L ( a , b , c ) = d N ( c + N d N ( b + N a , R ) , R ) , T R ( a , b , c ) = d N ( d N ( c + N b , R ) + N a , R ) , (36)
i.e. T L ( a , b , c ) = ( a ∗ b ) ∗ c T_L(a,b,c)=(a*b)*c T L ( a , b , c ) = ( a ∗ b ) ∗ c and T R ( a , b , c ) = a ∗ ( b ∗ c ) T_R(a,b,c)=a*(b*c) T R ( a , b , c ) = a ∗ ( b ∗ c ) . Their codes, in the universal normal form of ternary operators, are
L ( R ) = ∑ m = 0 N 3 − 1 N m T L ( d N ( 2 , m ) , d N ( 1 , m ) , d N ( 0 , m ) ) , L(R)=\sum_{m=0}^{N^3-1}N^mT_L(d_N(2,m),d_N(1,m),d_N(0,m)), L ( R ) = m = 0 ∑ N 3 − 1 N m T L ( d N ( 2 , m ) , d N ( 1 , m ) , d N ( 0 , m )) ,
R ′ ( R ) = ∑ m = 0 N 3 − 1 N m T R ( d N ( 2 , m ) , d N ( 1 , m ) , d N ( 0 , m ) ) . (37) R'(R)=\sum_{m=0}^{N^3-1}N^mT_R(d_N(2,m),d_N(1,m),d_N(0,m)). \tag*{(37)} R ′ ( R ) = m = 0 ∑ N 3 − 1 N m T R ( d N ( 2 , m ) , d N ( 1 , m ) , d N ( 0 , m )) . (37)
The operator with code R R R is associative if and only if L ( R ) = R ′ ( R ) L(R)=R'(R) L ( R ) = R ′ ( R ) . The number of positions in which the two codes differ,
D ( R ) = ∑ m = 0 N 3 − 1 ( 1 − d N ( d N ( m , L ( R ) ) − d N ( m , R ′ ( R ) ) , 1 ) ) , (38) D(R)=\sum_{m=0}^{N^3-1}\left(1-d_N\left(d_N(m,L(R))-d_N(m,R'(R)),1\right)\right), \tag*{(38)} D ( R ) = m = 0 ∑ N 3 − 1 ( 1 − d N ( d N ( m , L ( R )) − d N ( m , R ′ ( R )) , 1 ) ) , (38)
is the associativity defect : D ( R ) = 0 D(R)=0 D ( R ) = 0 characterises associative operators, and D ( R ) D(R) D ( R ) can be used as an objective function when one explores magmas, quasigroups or loops. As a search over all N N 2 N^{N^2} N N 2 codes, however, this criterion is hopeless except for very small N N N . The rest of the section shows how to avoid the search.
Rows as unary codes
The base-N N N^N N N digits of R R R are unary codes: since the index b + N a b+Na b + N a has a a a as its most significant digit,
R = ∑ a = 0 N − 1 Λ a ( N N ) a , Λ a = d N N ( a , R ) , a ∗ b = d N ( b , Λ a ) . (39) R=\sum_{a=0}^{N-1}\Lambda_a(N^N)^a,\qquad\Lambda_a=d_{N^N}(a,R),\qquad a*b=d_N(b,\Lambda_a). \tag*{(39)} R = a = 0 ∑ N − 1 Λ a ( N N ) a , Λ a = d N N ( a , R ) , a ∗ b = d N ( b , Λ a ) . (39)
Λ a \Lambda_a Λ a is the code of left multiplication by a a a , i.e. of the row a a a of the table. The composition of two unary codes Λ \Lambda Λ , Λ ′ \Lambda' Λ ′ is the unary code
Λ ∘ Λ ′ = ∑ v = 0 N − 1 N v d N ( d N ( v , Λ ′ ) , Λ ) . (40) \Lambda\circ\Lambda' = \sum_{v=0}^{N-1} N^v d_N(d_N(v,\Lambda'),\Lambda). \tag*{(40)} Λ ∘ Λ ′ = v = 0 ∑ N − 1 N v d N ( d N ( v , Λ ′ ) , Λ ) . (40)
Proposition 3.20. The operator with code R R R is associative if and only if
Λ a ∗ b = Λ a ∘ Λ b for all a , b ∈ A N . \Lambda_{a*b} = \Lambda_a \circ\Lambda_b \qquad\text{for all } a,b \in A_N. Λ a ∗ b = Λ a ∘ Λ b for all a , b ∈ A N . Proof. By Eqs. (36) and (39) ,
T L ( a , b , c ) = d N ( c , Λ a ∗ b ) , T R ( a , b , c ) = d N ( d N ( c , Λ b ) , Λ a ) = d N ( c , Λ a ∘ Λ b ) T_L(a,b,c) = d_N(c,\Lambda_{a*b}), \qquad T_R(a,b,c) = d_N(d_N(c,\Lambda_b),\Lambda_a) = d_N(c,\Lambda_a \circ\Lambda_b) T L ( a , b , c ) = d N ( c , Λ a ∗ b ) , T R ( a , b , c ) = d N ( d N ( c , Λ b ) , Λ a ) = d N ( c , Λ a ∘ Λ b ) by Eq. (40) . The two ternary codes coincide for all c c c exactly when the two unary codes coincide.
Read backwards, Proposition 3.20 is a construction, and associativity is inherited from the composition of maps, which is always associative.
Proposition 3.21 (Construction from unary codes). Let H H H be a set of N N N unary codes on some alphabet A s A_s A s , containing the identity and closed under composition (40) , and let ϕ : H → A N \phi:H \to A_N ϕ : H → A N be a bijection with ϕ ( id ) = 0 \phi(\operatorname{id})=0 ϕ ( id ) = 0 . Then
a ∗ b = ϕ ( ϕ − 1 ( a ) ∘ ϕ − 1 ( b ) ) a * b = \phi\left(\phi^{-1}(a) \circ\phi^{-1}(b)\right) a ∗ b = ϕ ( ϕ − 1 ( a ) ∘ ϕ − 1 ( b ) ) is associative with neutral element 0 0 0 , and its code is obtained from Eq. (39) . If the elements of H H H are permutations, ∗ * ∗ is a group law.
In practice H H H is generated by a few permutation codes (for instance the generators of a simple group), all products are formed by Eq. (40) , and the elements are labelled in the order in which they appear. This is how simple groups enter the diagrams as single nodes.
Proposition 3.22 (Generator test). Let X ⊂ A N X \subset A_N X ⊂ A N generate A N A_N A N under ∗ * ∗ . If ( x ∗ g ) ∗ y = x ∗ ( g ∗ y ) (x*g)*y=x*(g*y) ( x ∗ g ) ∗ y = x ∗ ( g ∗ y ) for all x , y ∈ A N x,y \in A_N x , y ∈ A N and all g ∈ X g \in X g ∈ X , then ∗ * ∗ is associative. Equivalently, it suffices to verify Λ x ∗ g = Λ x ∘ Λ g \Lambda_{x*g}=\Lambda_x \circ\Lambda_g Λ x ∗ g = Λ x ∘ Λ g for g ∈ X g \in X g ∈ X .
Proof. If g g g and h h h satisfy the identity, then for all x , y x,y x , y
x ∗ ( ( g ∗ h ) ∗ y ) = x ∗ ( g ∗ ( h ∗ y ) ) = ( x ∗ g ) ∗ ( h ∗ y ) = ( ( x ∗ g ) ∗ h ) ∗ y = ( x ∗ ( g ∗ h ) ) ∗ y , x*((g*h)*y)=x*(g*(h*y))=(x*g)*(h*y)=((x*g)*h)*y=(x*(g*h))*y, x ∗ (( g ∗ h ) ∗ y ) = x ∗ ( g ∗ ( h ∗ y )) = ( x ∗ g ) ∗ ( h ∗ y ) = (( x ∗ g ) ∗ h ) ∗ y = ( x ∗ ( g ∗ h )) ∗ y , so g ∗ h g*h g ∗ h satisfies it as well. The set of elements satisfying the identity is thus closed under ∗ * ∗ and contains X X X , hence it is A N A_N A N .
With k k k generators the test needs about k N 2 kN^2 k N 2 evaluations of the digit function instead of N 3 N^3 N 3 .
Carry arrows from linear algebra
In a stacked diagram the associativity conditions of this chapter separate layer by layer: the conditions for layer h h h involve only its own codes and the upper block, which has already been built and is associative. Consider the building block with an abelian ground layer K = Z n K=\mathbb{Z}_n K = Z n , an upper block Q Q Q and a fixed action code Φ \Phi Φ (each φ A \varphi_A φ A an automorphism, i.e. multiplication by a unit). Then the cocycle condition is linear in the digits of the carry code F F F :
d n ( B + q A , F ) + d n ( C + q d q ( B + q A , Q ) , F ) − d n ( d n ( C + q B , F ) + n A , Φ ) − d n ( d q ( C + q B , Q ) + q A , F ) ≡ 0 ( m o d n ) , (41) \begin{aligned}
d_n(B+qA,F)+d_n(C+q\,d_q(B+qA,Q),F)\\
{}-d_n(d_n(C+qB,F)+nA,\Phi)-d_n(d_q(C+qB,Q)+qA,F)\equiv0 \pmod n, \tag*{(41)}
\end{aligned} d n ( B + q A , F ) + d n ( C + q d q ( B + q A , Q ) , F ) − d n ( d n ( C + q B , F ) + n A , Φ ) − d n ( d q ( C + q B , Q ) + q A , F ) ≡ 0 ( mod n ) , (41)
for all A , B , C ∈ A q A,B,C \in A_q A , B , C ∈ A q . Together with the normalisation d n ( m , F ) = 0 d_n(m,F)=0 d n ( m , F ) = 0 for m < q m<q m < q or d q ( 0 , m ) = 0 d_q(0,m)=0 d q ( 0 , m ) = 0 , this is a homogeneous linear system of q 3 q^3 q 3 equations modulo n n n in the ( q − 1 ) 2 (q-1)^2 ( q − 1 ) 2 unknown digits d n ( B + q A , F ) d_n(B+qA,F) d n ( B + q A , F ) , A , B ≠ 0 A,B\ne0 A , B = 0 . (Since φ A \varphi_A φ A is additive, d n ( d n ( C + q B , F ) + n A , Φ ) = λ A d n ( C + q B , F ) d_n(d_n(C+qB,F)+nA,\Phi)=\lambda_A d_n(C+qB,F) d n ( d n ( C + q B , F ) + n A , Φ ) = λ A d n ( C + q B , F ) modulo n n n .) For abelian K K K , condition 4 of the admissibility list holds automatically, so Eq. (41) is the only condition on F F F .
Define the digit-wise sum of two carry codes,
F ⊕ F ′ = ∑ m = 0 q 2 − 1 n m d n ( 0 , d n ( m , F ) + d n ( m , F ′ ) ) . (42) F\oplus F'=\sum_{m=0}^{q^2-1}n^m d_n\left(0,d_n(m,F)+d_n(m,F')\right). \tag*{(42)} F ⊕ F ′ = m = 0 ∑ q 2 − 1 n m d n ( 0 , d n ( m , F ) + d n ( m , F ′ ) ) . (42)
Proposition 3.23. With Q Q Q , K = Z n K=\mathbb{Z}_n K = Z n and Φ \Phi Φ fixed:
The admissible carry codes form a group Z 2 Z^2 Z 2 under ⊕ \oplus ⊕ , the solution space of Eq. (41) . Every admissible code is F = c 1 F 1 ⊕ ⋯ ⊕ c r F r F=c_1F_1\oplus\cdots\oplus c_rF_r F = c 1 F 1 ⊕ ⋯ ⊕ c r F r for a basis F 1 , … , F r F_1,\ldots,F_r F 1 , … , F r of solutions and coefficients c i ∈ A n c_i\in A_n c i ∈ A n .
For every unary code g g g on A q → A n A_q\to A_n A q → A n with g ( 0 ) = 0 g(0)=0 g ( 0 ) = 0 , the carry code δ g \delta g δ g with digits
d n ( B + q A , δ g ) = d n ( 0 , d n ( d n ( B , g ) + n A , Φ ) − d n ( d q ( B + q A , Q ) , g ) + d n ( A , g ) ) d_n(B+qA,\delta g)=d_n\left(0,d_n\left(d_n(B,g)+nA,\Phi\right)-d_n\left(d_q(B+qA,Q),g\right)+d_n(A,g)\right) d n ( B + q A , δ g ) = d n ( 0 , d n ( d n ( B , g ) + n A , Φ ) − d n ( d q ( B + q A , Q ) , g ) + d n ( A , g ) ) is admissible. The set of these codes is a subgroup B 2 ⊆ Z 2 B^2\subseteq Z^2 B 2 ⊆ Z 2 .
F F F and F ⊕ δ g F\oplus\delta g F ⊕ δ g define isomorphic groups: the relabelling x ( 0 ) ↦ x ( 0 ) + g ( x ( 1 ) ) x^{(0)}\mapsto x^{(0)}+g(x^{(1)}) x ( 0 ) ↦ x ( 0 ) + g ( x ( 1 ) ) of the ground layer transforms one into the other.
Hence one carry code per class of H 2 = Z 2 / B 2 H^2=Z^2/B^2 H 2 = Z 2 / B 2 suffices to produce all groups of the diagram.
The action codes Φ \Phi Φ are constructed in the same spirit, by choosing images of the generators of Q Q Q in Aut ( K ) \operatorname{Aut}(K) Aut ( K ) that satisfy the relations of Q Q Q . When K K K is non-abelian Eq. (41) is no longer linear; such layers are usually introduced as single nodes built with Proposition 3.21 .
Ordinary carries
The carry arrow takes its name from positional addition. Take a ground layer K = Z p K=\mathbb{Z}_p K = Z p under an upper layer Q = Z p Q=\mathbb{Z}_p Q = Z p with trivial action, and the carry code with digits
d p ( B + p A , F ) = t c p ( A , B ) , t ∈ A p , d_p(B+pA,F)=t\,c_p(A,B),\qquad t\in A_p, d p ( B + p A , F ) = t c p ( A , B ) , t ∈ A p ,
where c p c_p c p is the carry of Proposition 1.22, c p ( A , B ) = B ( A + B − 3 p − 2 2 , p − 1 2 ) c_p(A,B)=\mathcal{B}\left(A+B-\frac{3p-2}{2},\frac{p-1}{2}\right) c p ( A , B ) = B ( A + B − 2 3 p − 2 , 2 p − 1 ) . By Proposition 1.23 it satisfies Eq. (41) for every t t t , so all these diagrams are groups. For t = 0 t=0 t = 0 the diagram has no arrow and gives C p ⊕ C p C_p\oplus C_p C p ⊕ C p . For t ≠ 0 t\ne0 t = 0 the element with upper digit 1 and ground digit 0 has p p p -th power with upper digit 0 and ground digit t ≠ 0 t\ne0 t = 0 , so it has order p 2 p^2 p 2 and the group is C p 2 C_{p^2} C p 2 ; for t = 1 t=1 t = 1 the law is ordinary addition modulo p 2 p^2 p 2 , with the ground layer playing the role of the higher digit. The codes t c p tc_p t c p represent all classes of H 2 H^2 H 2 , which has order p p p : none of them with t ≠ 0 t\ne0 t = 0 lies in B 2 B^2 B 2 , since a coboundary gives a split, hence non-cyclic, group. An exhaustive check confirms the counts: for p = 2 p=2 p = 2 there are 2 admissible normalised carry codes and ∣ H 2 ∣ = 2 \lvert H^2\rvert=2 ∣ H 2 ∣ = 2 ; for p = 3 p=3 p = 3 there are 9, with ∣ B 2 ∣ = 3 \lvert B^2\rvert=3 ∣ B 2 ∣ = 3 and ∣ H 2 ∣ = 3 \lvert H^2\rvert=3 ∣ H 2 ∣ = 3 , six of them giving C 9 C_9 C 9 and three giving C 3 ⊕ C 3 C_3\oplus C_3 C 3 ⊕ C 3 .
Examples: orders 8 and 16
Example 3.24 (Order 8 without search). For three Z 2 \mathbb{Z}_2 Z 2 layers the upper block is Z 2 2 \mathbb{Z}_2^2 Z 2 2 or Z 4 \mathbb{Z}_4 Z 4 , and there are 9 unknown carry digits into the ground layer. Solving Eq. (41) modulo 2 gives dim Z 2 = 4 \dim\mathbb{Z}^2=4 dim Z 2 = 4 over Z 2 \mathbb{Z}_2 Z 2 and dim Z 2 = 3 \dim\mathbb{Z}^2=3 dim Z 2 = 3 over Z 4 \mathbb{Z}_4 Z 4 , i.e. 16 + 8 = 24 16+8=24 16 + 8 = 24 admissible diagrams. This is exactly the number of associative diagrams found by testing all 1024 of them. Moreover dim B 2 = 1 \dim B^2=1 dim B 2 = 1 and 2, so that ∣ H 2 ∣ = 8 \lvert H^2\rvert=8 ∣ H 2 ∣ = 8 and 2 respectively.
Example 3.25 (Boolean carry codes of order 8). Over the upper block Z 2 2 \mathbb{Z}_2^2 Z 2 2 write A = a 0 + 2 a 1 A=a_0+2a_1 A = a 0 + 2 a 1 and B = b 0 + 2 b 1 B=b_0+2b_1 B = b 0 + 2 b 1 with a i = d 2 ( i , A ) a_i=d_2(i,A) a i = d 2 ( i , A ) , b i = d 2 ( i , B ) b_i=d_2(i,B) b i = d 2 ( i , B ) , and take the bilinear carry codes
d 2 ( B + 4 A , F ) = d 2 ( 0 , c 00 a 0 b 0 + c 01 a 0 b 1 + c 10 a 1 b 0 + c 11 a 1 b 1 ) , c i j ∈ A 2 . d_2(B+4A,F)=d_2(0,c_{00}a_0b_0+c_{01}a_0b_1+c_{10}a_1b_0+c_{11}a_1b_1),\qquad c_{ij}\in A_2. d 2 ( B + 4 A , F ) = d 2 ( 0 , c 00 a 0 b 0 + c 01 a 0 b 1 + c 10 a 1 b 0 + c 11 a 1 b 1 ) , c ij ∈ A 2 . All 16 are admissible and they represent all 8 classes of H 2 H^2 H 2 (two codes per class). Listing c 00 c 01 c 10 c 11 c_{00}c_{01}c_{10}c_{11} c 00 c 01 c 10 c 11 :
Table 5 (PDF p. 44).
The list is symmetric under exchanging the two operands and under exchanging the two bits. A single product of bits is a carry: a 0 b 0 a_0b_0 a 0 b 0 is the carry of Z 4 Z_4 Z 4 in the first bit, giving C 4 × C 2 C_4\times C_2 C 4 × C 2 , while the mixed product a 1 b 0 a_1b_0 a 1 b 0 gives D 4 D_4 D 4 . The quaternion group needs three products, a 0 b 0 + a 1 b 0 + a 1 b 1 a_0b_0+a_1b_0+a_1b_1 a 0 b 0 + a 1 b 0 + a 1 b 1 . The code a 0 b 1 + a 1 b 0 a_0b_1+a_1b_0 a 0 b 1 + a 1 b 0 is the coboundary δ g \delta g δ g of g ( A ) = a 0 a 1 g(A)=a_0a_1 g ( A ) = a 0 a 1 and gives the elementary Abelian group. Each product a i b j a_ib_j a i b j is a Boolean polynomial in the sense of Corollary 1.21, so these carries are exactly the Boolean forms of degree two. An exhaustive search over all 2 16 2^{16} 2 16 carry tables confirms the classification: 8 classes, of types C 2 3 \mathsf{C}_2^3 C 2 3 (one class), C 4 × C 2 C_4\times C_2 C 4 × C 2 (three), D 4 D_4 D 4 (three) and Q 8 Q_8 Q 8 (one).
Example 3.26 (Order 16). With four Z 2 \mathbb{Z}_2 Z 2 layers there are 24 × 2 49 ≈ 1.35 × 10 16 24\times2^{49}\approx1.35\times10^{16} 24 × 2 49 ≈ 1.35 × 1 0 16 normalised diagrams with an associative upper block, far too many to test. Solving Eq. (41) over each of the 24 associative diagrams of order 8 gives 6912 associative diagrams of order 16:
Table 6.
Indeed 2 ⋅ 2 10 + 10 ⋅ 2 8 + 6 ⋅ 2 8 + 2 ⋅ 2 7 + 4 ⋅ 2 7 = 6912 2\cdot2^{10}+10\cdot2^8+6\cdot2^8+2\cdot2^7+4\cdot2^7=6912 2 ⋅ 2 10 + 10 ⋅ 2 8 + 6 ⋅ 2 8 + 2 ⋅ 2 7 + 4 ⋅ 2 7 = 6912 . Taking one carry code per class of H 2 H^2 H 2 for each of the five groups of order 8 yields 64 + 8 + 8 + 4 + 2 = 86 64+8+8+4+2=86 64 + 8 + 8 + 4 + 2 = 86 tables, all associative, which realise all 14 groups of order 16. This is to be expected: every group of order 16 has a central subgroup of order 2, so one final Z 2 \mathbb{Z}_2 Z 2 layer with a carry arrow reaches all of them.
The ternary codes L ( R ) L(R) L ( R ) and R ′ ( R ) R'(R) R ′ ( R ) remain useful as a final check of a construction, or, through the defect (38) , when exploring structures beyond groups.
Algorithm
Given a diagram, the code of the group is obtained as follows.
Choose the layers p M − 1 , … , p 0 p_{M-1},\ldots,p_0 p M − 1 , … , p 0 (top to ground) and the node codes K h K_h K h .
For each hollow layer choose the action code Φ h \Phi_h Φ h and the carry code F h F_h F h ; draw an arrow from every upper layer on whose digit they depend.
Impose the normalisations φ 0 = id \varphi_0=\operatorname{id} φ 0 = id and f ( 0 , ⋅ ) = f ( ⋅ , 0 ) = 0 f(0,\cdot)=f(\cdot,0)=0 f ( 0 , ⋅ ) = f ( ⋅ , 0 ) = 0 .
For every pair ( a , b ) (a,b) ( a , b ) compute y ( h ) y^{(h)} y ( h ) from the top layer down with Eq. (34) and assemble a ∗ b a*b a ∗ b with Eq. (35) .
Check associativity (or conditions 3–5 layer by layer); for abelian layers construct the carry codes directly as solutions of Eq. (41) , one per class of H 2 H^2 H 2 (§ 9 ).
Compute R R R from Eq. (32) if the explicit code is required, and identify the group through invariants such as the centre and the numbers of elements of each order.
Exercises
Exercise 3.1. Construct Z 8 \mathbb{Z}_8 Z 8 with three Z 2 \mathbb{Z}_2 Z 2 layers and carry arrows only, and write the carry codes F 1 F_1 F 1 and F 0 F_0 F 0 explicitly.
Exercise 3.2. Write the codes Φ \Phi Φ of D n D_n D n for n = 3 , 4 , 5 n = 3, 4, 5 n = 3 , 4 , 5 and verify the formula for y ( 0 ) y^{(0)} y ( 0 ) .
Exercise 3.3. Show that the recipe for the Heisenberg group satisfies the cocycle condition and compute its centre from Eq. (27) .
Exercise 3.4. Verify that Q 16 Q_{16} Q 16 is obtained from the generalised quaternion recipe and compare its diagram with that of D 8 D_8 D 8 .
The main theorems of group theory in digital form
In this chapter the classical theorems of finite group theory are formulated in terms of the digit function, of codes and of the layers and arrows of the diagrammatic method (Chapter 3). In each case the statement is digital: subgroups are ground blocks of mixed-radix labellings, cosets are fibres of layer projectors, normal subgroups are closed layers, quotients are layer codes, actions are codes of binary operators, and counting arguments are statements about base-p p p digits. The proofs follow the classical arguments, which become statements about digits.
The digital setting
A group of order N N N is given by its code R ∈ [ 0 , N N 2 ) R \in[0,N^{N^2}) R ∈ [ 0 , N N 2 ) ,
a ∗ b = d N ( a N + b , R ) , a , b ∈ A N , (43) a \ast b = d_N(aN+b,R), \qquad a,b \in\mathcal{A}_N, \tag*{(43)} a ∗ b = d N ( a N + b , R ) , a , b ∈ A N , (43)
and we always label the neutral element by 0. The inverse of a a a is denoted a ˉ \bar{a} a ˉ . In this chapter the order of the group is written N N N rather than p p p , because p p p is reserved for the prime of the Sylow theorems and p h p_h p h for the radices of a labelling; otherwise the notation is that of the preceding chapters, with rows L a L_a L a and alphabets A N \mathcal{A}_N A N .
Digital Kronecker delta. For integers k k k and s ≥ 2 s \ge2 s ≥ 2 one has d s ( k , 1 ) = ⌊ s − k ⌋ − s ⌊ s − k − 1 ⌋ = δ k , 0 d_s(k,1)=\lfloor s^{-k}\rfloor-s\lfloor s^{-k-1}\rfloor=\delta_{k,0} d s ( k , 1 ) = ⌊ s − k ⌋ − s ⌊ s − k − 1 ⌋ = δ k , 0 . Hence, for u , v ∈ A s u,v \in\mathcal{A}_s u , v ∈ A s ,
δ u , v = d s ( u − v , 1 ) . (44) \delta_{u,v}=d_s(u-v,1). \tag*{(44)} δ u , v = d s ( u − v , 1 ) . (44)
With (44) every counting statement becomes a sum of digits; for instance the number of elements x x x satisfying an equation u ( x ) = v ( x ) u(x)=v(x) u ( x ) = v ( x ) is ∑ x d s ( u ( x ) − v ( x ) , 1 ) \sum_x d_s(u(x)-v(x),1) ∑ x d s ( u ( x ) − v ( x ) , 1 ) .
Labellings and isomorphism. A labelling is a permutation code λ \lambda λ of A N \mathcal{A}_N A N with λ ( 0 ) = 0 \lambda(0)=0 λ ( 0 ) = 0 . The relabelled code is
R λ = ∑ m = 0 N 2 − 1 N m λ ( d N ( λ − 1 ( d N ( 1 , m ) ) N + λ − 1 ( d N ( 0 , m ) ) , R ) ) , (45) R^\lambda=\sum_{m=0}^{N^2-1}N^m\lambda\left(d_N\left(\lambda^{-1}(d_N(1,m))N+\lambda^{-1}(d_N(0,m)),R\right)\right), \tag*{(45)} R λ = m = 0 ∑ N 2 − 1 N m λ ( d N ( λ − 1 ( d N ( 1 , m )) N + λ − 1 ( d N ( 0 , m )) , R ) ) , (45)
and two codes R , R ′ R,R' R , R ′ on N N N symbols define isomorphic groups if and only if R ′ = R λ R'=R^\lambda R ′ = R λ for some labelling λ \lambda λ . Given a factorisation N = p 0 p 1 ⋯ p M − 1 N=p_0p_1\cdots p_{M-1} N = p 0 p 1 ⋯ p M − 1 , the labelled elements are expanded in mixed radix, with layer projectors x ( h ) x^{(h)} x ( h ) and upper blocks U h ( x ) U_h(x) U h ( x ) as in Eqs. (24) and (25) .
Homomorphisms. A map ψ : A N → A N ′ \psi: \mathcal{A}_N \to\mathcal{A}_{N'} ψ : A N → A N ′ is given by its code Ψ ∈ [ 0 , N ′ N ) \Psi\in[0, N'^{N}) Ψ ∈ [ 0 , N ′ N ) , ψ ( x ) = d N ′ ( x , Ψ ) \psi(x) = d_{N'}(x,\Psi) ψ ( x ) = d N ′ ( x , Ψ ) . It is a homomorphism from R R R to R ′ R' R ′ if and only if
d N ′ ( d N ( a N + b , R ) , Ψ ) = d N ′ ( d N ′ ( a , Ψ ) N ′ + d N ′ ( b , Ψ ) , R ′ ) ∀ a , b . (46) d_{N'}\left(d_N(aN+b,R),\Psi\right) = d_{N'}\left(d_{N'}(a,\Psi)N' + d_{N'}(b,\Psi),R'\right) \qquad\forall a,b. \tag*{(46)} d N ′ ( d N ( a N + b , R ) , Ψ ) = d N ′ ( d N ′ ( a , Ψ ) N ′ + d N ′ ( b , Ψ ) , R ′ ) ∀ a , b . (46)
Actions. An action of G G G on A s \mathcal{A}_s A s is given by a code Θ ∈ [ 0 , s s N ) \Theta\in[0,s^{sN}) Θ ∈ [ 0 , s s N ) , θ a ( x ) = d s ( x + s a , Θ ) \theta_a(x) = d_s(x+sa,\Theta) θ a ( x ) = d s ( x + s a , Θ ) , such that θ 0 = id \theta_0 = \operatorname{id} θ 0 = id and
d s ( d s ( x + s b , Θ ) + s a , Θ ) = d s ( x + s d N ( a N + b , R ) , Θ ) ∀ a , b , x . (47) d_s\left(d_s(x+sb,\Theta)+sa,\Theta\right) = d_s\left(x+s d_N(aN+b,R),\Theta\right) \qquad\forall a,b,x. \tag*{(47)} d s ( d s ( x + s b , Θ ) + s a , Θ ) = d s ( x + s d N ( a N + b , R ) , Θ ) ∀ a , b , x . (47)
Thus an action is a binary operator A N × A s → A s \mathcal{A}_N \times\mathcal{A}_s \to\mathcal{A}_s A N × A s → A s that is compatible with R R R , exactly as the action codes Φ \Phi Φ of the arrows of Chapter 3.
Adapted labellings. Let H H H be a subgroup of order n n n and let q = N / n q=N/n q = N / n . A labelling λ \lambda λ is adapted to H H H (with radices ( n , q ) (n,q) ( n , q ) ) if λ ( H ) = A n \lambda(H)=\mathcal{A}_n λ ( H ) = A n and the left cosets of H H H are mapped onto the fibres of the upper layer:
λ ( g H ) = { x ∈ A N : ⌊ x / n ⌋ = ⌊ λ ( g ) / n ⌋ } . \lambda(gH)=\{x \in\mathcal{A}_N : \lfloor x/n\rfloor= \lfloor\lambda(g)/n\rfloor\}. λ ( g H ) = { x ∈ A N : ⌊ x / n ⌋ = ⌊ λ ( g ) / n ⌋} .
In other words H H H is the ground block, and two elements lie in the same left coset if and only if their upper digits coincide.
Table 1 (PDF p. 48).
Cayley’s theorem
Theorem 4.1 (Cayley, digital form). Let R R R be the code of a group of order N N N and let L a = d N N ( a , R ) L_a=d_{N^N}(a,R) L a = d N N ( a , R ) be its rows, so that R = ∑ a L a N N a R=\sum_a L_aN^{Na} R = ∑ a L a N N a . Then each L a L_a L a is a permutation code of A N \mathcal{A}_N A N ,
a = d N ( 0 , L a ) , L a ∗ b = L a ∘ L b , a=d_N(0,L_a), \qquad L_{a*b}=L_a\circ L_b, a = d N ( 0 , L a ) , L a ∗ b = L a ∘ L b , and a ↦ L a a \mapsto L_a a ↦ L a is an injective homomorphism into the group of permutation codes of A N \mathcal{A}_N A N under composition (40) . Every group of order N N N is therefore isomorphic to a group of N N N unary codes 1 R N 1R_N 1 R N , and its code is recovered from the set of its rows.
Proof. By (39) , d N ( b , L a ) = a ∗ b d_N(b,L_a)=a*b d N ( b , L a ) = a ∗ b . Since a ∗ b = a ∗ b ′ a*b=a*b' a ∗ b = a ∗ b ′ implies b = b ′ b=b' b = b ′ , the digits of L a L_a L a are a permutation of A N \mathcal{A}_N A N . Taking b = 0 b=0 b = 0 gives d N ( 0 , L a ) = a d_N(0,L_a)=a d N ( 0 , L a ) = a , so different elements have different rows. Associativity gives L a ∗ b = L a ∘ L b L_{a*b}=L_a\circ L_b L a ∗ b = L a ∘ L b by Proposition 3.20 .
The converse, the construction of associative codes from closed sets of permutation codes, is Proposition 3.21 .
Lagrange's theorem
Theorem 4.2 (Lagrange, digital form). For every subgroup H H H of order n n n of a group of order N N N there is a labelling adapted to H H H . Consequently N = n q N=nq N = n q : every subgroup is the ground block of a mixed-radix labelling with radices ( n , q ) (n,q) ( n , q ) , its left cosets are the fibres of the upper layer, and the index [ G : H ] = q [G:H]=q [ G : H ] = q is the radix of the upper layer.
Proof. Choose a transversal t 0 = 0 , t 1 , … , t q − 1 t_0=0,t_1,\ldots,t_{q-1} t 0 = 0 , t 1 , … , t q − 1 of the left cosets and a bijection λ H : H → A n \lambda_H:H\to A_n λ H : H → A n with λ H ( 0 ) = 0 \lambda_H(0)=0 λ H ( 0 ) = 0 , and put λ ( t u ∗ h ) = λ H ( h ) + n u \lambda(t_u*h)=\lambda_H(h)+nu λ ( t u ∗ h ) = λ H ( h ) + n u . The left cosets are disjoint and cover G G G , and h ↦ t u ∗ h h\mapsto t_u*h h ↦ t u ∗ h is injective because the row L t u L_{t_u} L t u is a permutation code (Theorem 4.1 ). Hence every coset has n n n elements, λ \lambda λ is a bijection onto A n q A_{nq} A n q , and λ ( t u H ) \lambda(t_uH) λ ( t u H ) is the fibre { x : ⌊ x / n ⌋ = u } \{x:\lfloor x/n\rfloor=u\} { x : ⌊ x / n ⌋ = u } .
Corollary 4.3. The order of every element divides N N N . Equivalently, L a ∘ N = ∑ v N v L_a^{\circ N}=\sum_v N^v L a ∘ N = ∑ v N v (the identity code) for every row of R R R .
Proof. Apply Theorem 4.2 } to the cyclic subgroup generated by $a$, and Theorem 4.1`.
Corollary 4.4. If N N N is prime, no labelling has more than one layer, and R R R is a relabelling of the code of Z N \mathbb{Z}_N Z N , d N ( 0 , a + b ) d_N(0,a+b) d N ( 0 , a + b ) .
Theorem 4.2 is the reason why the radices of every diagram of a group divide its order, and why the tower labellings of Chapter 3 exist for every chain of subgroups.
Group actions
Theorem 4.5 (Orbit–stabiliser, digital form). Let Θ \Theta Θ be the code of an action of G G G on A s A_s A s , let x ∈ A s x\in A_s x ∈ A s , and let G x G_x G x and G x Gx G x be its stabiliser and its orbit. There is a labelling of G G G adapted to G x G_x G x , with radices ( ∣ G x ∣ , ∣ G x ∣ ) (|G_x|,|Gx|) ( ∣ G x ∣ , ∣ G x ∣ ) , whose upper layer projector is the orbit map:
θ g ( x ) = ι ( U 0 ( λ ( g ) ) ) \theta_g(x)=\iota\bigl(U_0(\lambda(g))\bigr) θ g ( x ) = ι ( U 0 ( λ ( g )) ) *for a bijection ι : A ∣ G x ∣ → G x \iota:A_{|G_x|}\to Gx ι : A ∣ G x ∣ → G x . In particular N = ∣ G x ∣ ⋅ ∣ G x ∣ N=|G_x|\cdot|Gx| N = ∣ G x ∣ ⋅ ∣ G x ∣ , and the upward arrows of this labelling (from the left operand into the upper layer) are the action itself: U 0 ( λ ( a ∗ g ) ) = ι − 1 ( θ a ( ι ( U 0 ( λ ( g ) ) ) ) ) U_0(\lambda(a*g))=\iota^{-1}\bigl(\theta_a(\iota(U_0(\lambda(g))))\bigr) U 0 ( λ ( a ∗ g )) = ι − 1 ( θ a ( ι ( U 0 ( λ ( g )))) ) . *
Proof. θ g ( x ) = θ g ′ ( x ) \theta_g(x)=\theta_{g'}(x) θ g ( x ) = θ g ′ ( x ) if and only if θ g − 1 ∗ g ′ ( x ) = x \theta_{g^{-1}*g'}(x)=x θ g − 1 ∗ g ′ ( x ) = x , i.e. g ′ ∈ g G x g'\in gG_x g ′ ∈ g G x . Thus the fibres of the orbit map are the left cosets of G x G_x G x , and Theorem 4.2 applied to G x G_x G x gives the labelling, with ι ( u ) = θ t u ( x ) \iota(u)=\theta_{t_u}(x) ι ( u ) = θ t u ( x ) . The last identity is (47) .
The number of points fixed by a a a is, by (44) ,
∣ Fix ( θ a ) ∣ = ∑ x = 0 s − 1 d s ( d s ( x + s a , Θ ) − x , 1 ) . (48) |\operatorname{Fix}(\theta_a)|=\sum_{x=0}^{s-1}d_s\bigl(d_s(x+sa,\Theta)-x,1\bigr). \tag*{(48)} ∣ Fix ( θ a ) ∣ = x = 0 ∑ s − 1 d s ( d s ( x + s a , Θ ) − x , 1 ) . (48)
Theorem 4.6 (Burnside’s lemma, digital form). The number of orbits of the action with code Θ \Theta Θ is
1 N ∑ a = 0 N − 1 ∑ x = 0 s − 1 d s ( d s ( x + s a , Θ ) − x , 1 ) . \frac{1}{N}\sum_{a=0}^{N-1}\sum_{x=0}^{s-1}d_s\bigl(d_s(x+sa,\Theta)-x,1\bigr). N 1 a = 0 ∑ N − 1 x = 0 ∑ s − 1 d s ( d s ( x + s a , Θ ) − x , 1 ) . Proof. The double sum counts the pairs ( a , x ) (a,x) ( a , x ) with θ a ( x ) = x \theta_a(x)=x θ a ( x ) = x , i.e. ∑ x ∣ G x ∣ = ∑ x N / ∣ G x ∣ \sum_x|G_x|=\sum_xN/|Gx| ∑ x ∣ G x ∣ = ∑ x N /∣ G x ∣ by Theorem 4.5, and each orbit contributes N N N to the last sum.
Lemma 4.7 (Ground-digit congruence). Let a group P P P of order p a p^a p a , p p p prime, act on A s A_s A s , and let Fix \operatorname{Fix} Fix be the set of fixed points. Then
d p ( 0 , s ) = d p ( 0 , ∣ Fix ∣ ) . d_p(0,s)=d_p(0,|\operatorname{Fix}|). d p ( 0 , s ) = d p ( 0 , ∣ Fix ∣ ) . Proof. By Theorem 4.5 every orbit has p k p^k p k points, with k ≥ 1 k\geq1 k ≥ 1 for non-fixed points. Hence s = ∣ Fix ∣ + ∑ i p k i s=|\operatorname{Fix}|+\sum_i p^{k_i} s = ∣ Fix ∣ + ∑ i p k i with k i ≥ 1 k_i\geq1 k i ≥ 1 , and the sum does not affect the digit d p ( 0 , ⋅ ) d_p(0,\cdot) d p ( 0 , ⋅ ) .
Conjugation. The conjugation action has the code Θ c \Theta_{\mathrm{c}} Θ c with d N ( x + N a , Θ c ) = a ∗ x ∗ a ˉ d_N(x+Na,\Theta_{\mathrm{c}})=a*x*\bar{a} d N ( x + N a , Θ c ) = a ∗ x ∗ a ˉ . An element x x x is central if and only if its row and its column coincide, so that
∣ Z ( G ) ∣ = ∑ x = 0 N − 1 ∏ b = 0 N − 1 d N ( d N ( x N + b , R ) − d N ( b N + x , R ) , 1 ) . (49) |Z(G)|=\sum_{x=0}^{N-1}\prod_{b=0}^{N-1}d_N\bigl(d_N(xN+b,R)-d_N(bN+x,R),1\bigr). \tag*{(49)} ∣ Z ( G ) ∣ = x = 0 ∑ N − 1 b = 0 ∏ N − 1 d N ( d N ( x N + b , R ) − d N ( b N + x , R ) , 1 ) . (49)
Theorem 4.8 (Class equation, digital form). Let x 1 , … , x r x_1,\ldots,x_r x 1 , … , x r represent the conjugacy classes with more than one element. Then
N = ∣ Z ( G ) ∣ + ∑ i = 1 r N ∣ C ( x i ) ∣ , ∣ C ( x ) ∣ = ∑ b d N ( d N ( x N + b , R ) − d N ( b N + x , R ) , 1 ) . N=|Z(G)|+\sum_{i=1}^{r}\frac{N}{|C(x_i)|},\qquad|C(x)|=\sum_b d_N\bigl(d_N(xN+b,R)-d_N(bN+x,R),1\bigr). N = ∣ Z ( G ) ∣ + i = 1 ∑ r ∣ C ( x i ) ∣ N , ∣ C ( x ) ∣ = b ∑ d N ( d N ( x N + b , R ) − d N ( b N + x , R ) , 1 ) . If N = p a N=p^a N = p a with a ≥ 1 a\geq1 a ≥ 1 , then d p ( 0 , ∣ Z ( G ) ∣ ) = 0 d_p(0,|Z(G)|)=0 d p ( 0 , ∣ Z ( G ) ∣ ) = 0 , so the centre is non-trivial.
Proof. Apply Theorem 4.5 to Θ c \Theta_{\mathrm{c}} Θ c , whose fixed points form the centre and whose stabilisers are the centralisers. For N = p a N=p^a N = p a , Lemma 4.7 gives d p ( 0 , ∣ Z ( G ) ∣ ) = d p ( 0 , N ) = 0 d_p(0,|Z(G)|)=d_p(0,N)=0 d p ( 0 , ∣ Z ( G ) ∣ ) = d p ( 0 , N ) = 0 , and 0 ∈ Z ( G ) 0\in Z(G) 0 ∈ Z ( G ) . □
Theorem 4.9 (Cauchy, digital form). If a prime p p p divides N N N , then G G G has an element of order p p p .
Proof. Read the integers x ∈ A N p x\in A_{Np} x ∈ A N p as strings of p p p base-N N N digits g i = d N ( i , x ) g_i=d_N(i,x) g i = d N ( i , x ) , and let X X X be the set of those x x x with g p − 1 ∗ g p − 2 ∗ ⋯ ∗ g 0 = 0 g_{p-1}*g_{p-2}*\cdots*g_0=0 g p − 1 ∗ g p − 2 ∗ ⋯ ∗ g 0 = 0 . Since g 0 g_0 g 0 is determined by the other digits, ∣ X ∣ = N p − 1 |X|=N^{p-1} ∣ X ∣ = N p − 1 . The cyclic shift of digits
σ ( x ) = ⌊ x / N ⌋ + N p − 1 d N ( 0 , x ) \sigma(x)=\lfloor x/N\rfloor+N^{p-1}d_N(0,x) σ ( x ) = ⌊ x / N ⌋ + N p − 1 d N ( 0 , x )
maps X X X into itself, because u ∗ v = 0 u*v=0 u ∗ v = 0 implies v ∗ u = 0 v*u=0 v ∗ u = 0 , and σ p = id \sigma^p=\operatorname{id} σ p = id , so Z p \mathbb{Z}_p Z p acts on X X X . Its fixed points are the repunits x = g ( N p − 1 ) / ( N − 1 ) x=g(N^p-1)/(N-1) x = g ( N p − 1 ) / ( N − 1 ) with g ∗ p = 0 g^{*p}=0 g ∗ p = 0 . By Lemma 4.7 , d p ( 0 , ∣ Fix ∣ ) = d p ( 0 , N p − 1 ) = 0 d_p(0,|\operatorname{Fix}|)=d_p(0,N^{p-1})=0 d p ( 0 , ∣ Fix ∣ ) = d p ( 0 , N p − 1 ) = 0 . Since 0 ∈ Fix 0\in\operatorname{Fix} 0 ∈ Fix , ∣ Fix ∣ ≥ p |\operatorname{Fix}|\geq p ∣ Fix ∣ ≥ p , and there is g ≠ 0 g\ne0 g = 0 with g ∗ p = 0 g^{*p}=0 g ∗ p = 0 . □
Transitive and sharply transitive actions
Orbits, stabilisers and the orbit–stabiliser theorem were introduced above (Theorem 4.5 ). Here they are used to measure how strongly a group action mixes points, and then to explain the first digits in the Mathieu laws of Chapter 13.
Definition 4.10. Let 1 ≤ k ≤ n 1\leq k\leq n 1 ≤ k ≤ n . The action is k k k -transitive if, for any two k k k -tuples of distinct points ( x 1 , … , x k ) (x_1,\ldots,x_k) ( x 1 , … , x k ) and ( y 1 , … , y k ) (y_1,\ldots,y_k) ( y 1 , … , y k ) , there is g ∈ G g\in G g ∈ G with g ( x i ) = y i g(x_i)=y_i g ( x i ) = y i for i = 1 , … , k i=1,\ldots,k i = 1 , … , k . It is sharply k k k -transitive if this g g g is always unique. A 1-transitive action is called transitive, and a sharply 1-transitive action is called regular.
In words: a transitive group can move any point to any other point. A k k k -transitive group can move any k k k points, in a given order, to any other k k k points, and a sharply k k k -transitive group can do so in exactly one way. A k k k -transitive action is l l l -transitive for every l ≤ k l\leq k l ≤ k , since an l l l -tuple of distinct points can be extended to a k k k -tuple.
Lemma 4.11. Let G G G be k k k -transitive on A n A_n A n , let x 1 , … , x k x_1,\ldots,x_k x 1 , … , x k be distinct points, and let G x 1 , … , x k G_{x_1,\ldots,x_k} G x 1 , … , x k be their pointwise stabiliser. Then
∣ G ∣ = n ( n − 1 ) ⋯ ( n − k + 1 ) ∣ G x 1 , … , x k ∣ , |G|=n(n-1)\cdots(n-k+1)|G_{x_1,\ldots,x_k}|, ∣ G ∣ = n ( n − 1 ) ⋯ ( n − k + 1 ) ∣ G x 1 , … , x k ∣ , and the following are equivalent:
G G G is sharply k k k -transitive;
G x 1 , … , x k = 1 G_{x_1,\ldots,x_k}=1 G x 1 , … , x k = 1 ; @page 50/200 starts_mid=0 ends_mid=1 Conjugation. The conjugation action has the code Θ c \Theta_{\mathrm{c}} Θ c with d N ( x + N a , Θ c ) = a ∗ x ∗ a ˉ d_N(x+Na,\Theta_{\mathrm{c}})=a*x*\bar{a} d N ( x + N a , Θ c ) = a ∗ x ∗ a ˉ . An element x x x is central if and only if its row and its column coincide, so that
∣ Z ( G ) ∣ = ∑ x = 0 N − 1 ∏ b = 0 N − 1 d N ( d N ( x N + b , R ) − d N ( b N + x , R ) , 1 ) . (50) |Z(G)|=\sum_{x=0}^{N-1}\prod_{b=0}^{N-1}d_N\bigl(d_N(xN+b,R)-d_N(bN+x,R),1\bigr). \tag*{(50)} ∣ Z ( G ) ∣ = x = 0 ∑ N − 1 b = 0 ∏ N − 1 d N ( d N ( x N + b , R ) − d N ( b N + x , R ) , 1 ) . (50) Theorem 4.8 (Class equation, digital form). Let x 1 , … , x r x_1,\ldots,x_r x 1 , … , x r represent the conjugacy classes with more than one element. Then
N = ∣ Z ( G ) ∣ + ∑ i = 1 r N ∣ C ( x i ) ∣ , ∣ C ( x ) ∣ = ∑ b d N ( d N ( x N + b , R ) − d N ( b N + x , R ) , 1 ) . N=|Z(G)|+\sum_{i=1}^{r}\frac{N}{|C(x_i)|},\qquad|C(x)|=\sum_b d_N\bigl(d_N(xN+b,R)-d_N(bN+x,R),1\bigr). N = ∣ Z ( G ) ∣ + i = 1 ∑ r ∣ C ( x i ) ∣ N , ∣ C ( x ) ∣ = b ∑ d N ( d N ( x N + b , R ) − d N ( b N + x , R ) , 1 ) . If N = p a N=p^a N = p a with a ≥ 1 a\geq1 a ≥ 1 , then d p ( 0 , ∣ Z ( G ) ∣ ) = 0 d_p(0,|Z(G)|)=0 d p ( 0 , ∣ Z ( G ) ∣ ) = 0 , so the centre is non-trivial.
Proof. Apply Theorem 4.5 to Θ c \Theta_{\mathrm{c}} Θ c , whose fixed points form the centre and whose stabilisers are the centralisers. For N = p a N=p^a N = p a , Lemma 4.7 gives d p ( 0 , ∣ Z ( G ) ∣ ) = d p ( 0 , N ) = 0 d_p(0,|Z(G)|)=d_p(0,N)=0 d p ( 0 , ∣ Z ( G ) ∣ ) = d p ( 0 , N ) = 0 , and 0 ∈ Z ( G ) 0\in Z(G) 0 ∈ Z ( G ) . □
Theorem 4.9 (Cauchy, digital form). If a prime p p p divides N N N , then G G G has an element of order p p p .
Proof. Read the integers x ∈ A N p x\in A_{Np} x ∈ A N p as strings of p p p base-N N N digits g i = d N ( i , x ) g_i=d_N(i,x) g i = d N ( i , x ) , and let X X X be the set of those x x x with g p − 1 ∗ g p − 2 ∗ ⋯ ∗ g 0 = 0 g_{p-1}*g_{p-2}*\cdots*g_0=0 g p − 1 ∗ g p − 2 ∗ ⋯ ∗ g 0 = 0 . Since g 0 g_0 g 0 is determined by the other digits, ∣ X ∣ = N p − 1 |X|=N^{p-1} ∣ X ∣ = N p − 1 . The cyclic shift of digits
σ ( x ) = ⌊ x / N ⌋ + N p − 1 d N ( 0 , x ) \sigma(x)=\lfloor x/N\rfloor+N^{p-1}d_N(0,x) σ ( x ) = ⌊ x / N ⌋ + N p − 1 d N ( 0 , x ) maps X X X into itself, because u ∗ v = 0 u*v=0 u ∗ v = 0 implies v ∗ u = 0 v*u=0 v ∗ u = 0 , and σ p = id \sigma^p=\operatorname{id} σ p = id , so Z p \mathbb{Z}_p Z p acts on X X X . Its fixed points are the repunits x = g ( N p − 1 ) / ( N − 1 ) x=g(N^p-1)/(N-1) x = g ( N p − 1 ) / ( N − 1 ) with g ∗ p = 0 g^{*p}=0 g ∗ p = 0 . By Lemma 4.7 , d p ( 0 , ∣ Fix ∣ ) = d p ( 0 , N p − 1 ) = 0 d_p(0,|\operatorname{Fix}|)=d_p(0,N^{p-1})=0 d p ( 0 , ∣ Fix ∣ ) = d p ( 0 , N p − 1 ) = 0 . Since 0 ∈ Fix 0\in\operatorname{Fix} 0 ∈ Fix , ∣ Fix ∣ ≥ p |\operatorname{Fix}|\geq p ∣ Fix ∣ ≥ p , and there is g ≠ 0 g\ne0 g = 0 with g ∗ p = 0 g^{*p}=0 g ∗ p = 0 . □
Transitive and sharply transitive actions {#sec:4.4.1} Orbits, stabilisers and the orbit–stabiliser theorem were introduced above (Theorem 4.5 ). Here they are used to measure how strongly a group action mixes points, and then to explain the first digits in the Mathieu laws of Chapter 13.
Definition 4.10. Let 1 ≤ k ≤ n 1\leq k\leq n 1 ≤ k ≤ n . The action is k k k -transitive if, for any two k k k -tuples of distinct points ( x 1 , … , x k ) (x_1,\ldots,x_k) ( x 1 , … , x k ) and ( y 1 , … , y k ) (y_1,\ldots,y_k) ( y 1 , … , y k ) , there is g ∈ G g\in G g ∈ G with g ( x i ) = y i g(x_i)=y_i g ( x i ) = y i for i = 1 , … , k i=1,\ldots,k i = 1 , … , k . It is sharply k k k -transitive if this g g g is always unique. A 1-transitive action is called transitive, and a sharply 1-transitive action is called regular.
In words: a transitive group can move any point to any other point. A k k k -transitive group can move any k k k points, in a given order, to any other k k k points, and a sharply k k k -transitive group can do so in exactly one way. A k k k -transitive action is l l l -transitive for every l ≤ k l\leq k l ≤ k , since an l l l -tuple of distinct points can be extended to a k k k -tuple.
Lemma 4.11. Let G G G be k k k -transitive on A n A_n A n , let x 1 , … , x k x_1,\ldots,x_k x 1 , … , x k be distinct points, and let G x 1 , … , x k G_{x_1,\ldots,x_k} G x 1 , … , x k be their pointwise stabiliser. Then
∣ G ∣ = n ( n − 1 ) ⋯ ( n − k + 1 ) ∣ G x 1 , … , x k ∣ , |G|=n(n-1)\cdots(n-k+1)|G_{x_1,\ldots,x_k}|, ∣ G ∣ = n ( n − 1 ) ⋯ ( n − k + 1 ) ∣ G x 1 , … , x k ∣ , and the following are equivalent:
G G G is sharply k k k -transitive;
G x 1 , … , x k = 1 G_{x_1,\ldots,x_k}=1 G x 1 , … , x k = 1 ;
g ↦ ( g ( x 1 ) , … , g ( x k ) ) g \mapsto(g(x_1), \ldots, g(x_k)) g ↦ ( g ( x 1 ) , … , g ( x k )) is a bijection from G G G onto the n ( n − 1 ) ⋯ ( n − k + 1 ) n(n-1)\cdots(n-k+1) n ( n − 1 ) ⋯ ( n − k + 1 ) k k k -tuples of distinct points.
Proof. For l ≤ k l \le k l ≤ k , the stabiliser G x 1 , … , x l − 1 G_{x_1,\ldots,x_{l-1}} G x 1 , … , x l − 1 is transitive on the n − l + 1 n-l+1 n − l + 1 points other than x 1 , … , x l − 1 x_1,\ldots,x_{l-1} x 1 , … , x l − 1 : given such a point y y y , k k k -transitivity gives g g g with g ( x i ) = x i g(x_i)=x_i g ( x i ) = x i for i < l i<l i < l and g ( x l ) = y g(x_l)=y g ( x l ) = y . Applying the orbit–stabiliser theorem at each of the k k k steps gives the order formula. Two elements g , g ′ g,g' g , g ′ have the same images of x 1 , … , x k x_1,\ldots,x_k x 1 , … , x k exactly when g − 1 g ′ ∈ G x 1 , … , x k g^{-1}g' \in G_{x_1,\ldots,x_k} g − 1 g ′ ∈ G x 1 , … , x k . The map in (3) is onto by k k k -transitivity, so (2) and (3) are equivalent, and (1) implies (2). Conversely, assume (2), and let g ( x i ) = g ′ ( x i ′ ) g(x_i)=g'(x'_i) g ( x i ) = g ′ ( x i ′ ) for some distinct x 1 ′ , … , x k ′ x'_1,\ldots,x'_k x 1 ′ , … , x k ′ . Choose h h h with h ( x i ) = x i ′ h(x_i)=x'_i h ( x i ) = x i ′ . Then h − 1 g − 1 g ′ h h^{-1}g^{-1}g'h h − 1 g − 1 g ′ h fixes every x i x_i x i , so it is 1 1 1 and g = g ′ g=g' g = g ′ .
Example 4.12.
The cyclic group generated by x ↦ d n ( 0 , x + 1 ) x \mapsto d_n(0,x+1) x ↦ d n ( 0 , x + 1 ) is regular on A n A_n A n .
Sym n \operatorname{Sym}_n Sym n is sharply n n n -transitive, and also sharply ( n − 1 ) (n-1) ( n − 1 ) -transitive, since the image of the last point is forced. Alt n \operatorname{Alt}_n Alt n , n ≥ 3 n \ge3 n ≥ 3 , is sharply ( n − 2 ) (n-2) ( n − 2 ) -transitive: of the two permutations with prescribed images of n − 2 n-2 n − 2 points, exactly one is even.
For a prime power q q q , the maps x ↦ a x + b x \mapsto ax+b x ↦ a x + b with a ≠ 0 a \ne0 a = 0 act sharply 2 2 2 -transitively on F q \mathbb{F}_q F q . The maps x ↦ ( a x + b ) / ( c x + d ) x \mapsto(ax+b)/(cx+d) x ↦ ( a x + b ) / ( c x + d ) with a d − b c ≠ 0 ad-bc \ne0 a d − b c = 0 form PGL ( 2 , q ) \operatorname{PGL}(2,q) PGL ( 2 , q ) , which is sharply 3 3 3 -transitive on the q + 1 q+1 q + 1 points of the projective line.
M 11 M_{11} M 11 and M 12 M_{12} M 12 are sharply 4 4 4 - and 5 5 5 -transitive. M 22 M_{22} M 22 , M 23 M_{23} M 23 and M 24 M_{24} M 24 are 3 3 3 -, 4 4 4 - and 5 5 5 -transitive but not sharply so: the pointwise stabiliser of 3 3 3 , 4 4 4 or 5 5 5 points, respectively, has order 48 48 48 (Proposition 13.1 ).
By a theorem of Jordan, for k ≥ 4 k \ge4 k ≥ 4 the only sharply k k k -transitive groups are Sym k \operatorname{Sym}_k Sym k , Sym k + 1 \operatorname{Sym}_{k+1} Sym k + 1 , Alt k + 2 \operatorname{Alt}_{k+2} Alt k + 2 and, for k = 4 k=4 k = 4 and k = 5 k=5 k = 5 , the groups M 11 M_{11} M 11 and M 12 M_{12} M 12 [5]. The two small Mathieu groups are thus the only highly sharply transitive groups that are not symmetric or alternating. By the classification of finite simple groups, every 6 6 6 -transitive group is symmetric or alternating.
Transitivity is visible in the labels of Theorem 8.34 in Chapter 13. If G G G is k k k -transitive and b 1 , … , b k b_1,\ldots,b_k b 1 , … , b k are the first base points, Lemma 4.11 gives r l = n − l + 1 r_l=n-l+1 r l = n − l + 1 for l ≤ k l \le k l ≤ k , so the first k k k digits of a label have the falling-factorial radices n , n − 1 , … , n − k + 1 n,n-1,\ldots,n-k+1 n , n − 1 , … , n − k + 1 . The group is sharply k k k -transitive exactly when the chain stops there, as for M 11 M_{11} M 11 and M 12 M_{12} M 12 in Proposition 13.1 . For M 22 M_{22} M 22 , M 23 M_{23} M 23 and M 24 M_{24} M 24 two more digits, with radices 16 16 16 and 3 3 3 , code the pointwise stabiliser of order 48 48 48 .
Remark 4.13 (Canonical labels). For a sharply k k k -transitive group on A n A_n A n , Lemma 4.11(3) gives a label that needs no choice of representatives. Put y l = g ( l − 1 ) y_l=g(l-1) y l = g ( l − 1 ) for l = 1 , … , k l=1,\ldots,k l = 1 , … , k , let c l c_l c l be the number of points y < y l y<y_l y < y l different from y 1 , … , y l − 1 y_1,\ldots,y_{l-1} y 1 , … , y l − 1 , and set
κ ( g ) = c 1 + n ( c 2 + ( n − 1 ) ( c 3 + ⋯ + ( n − k + 2 ) c k ) ) . \kappa(g)=c_1+n\bigl(c_2+(n-1)(c_3+\cdots+(n-k+2)c_k)\bigr). κ ( g ) = c 1 + n ( c 2 + ( n − 1 ) ( c 3 + ⋯ + ( n − k + 2 ) c k ) ) .
This is the falling-factorial rank of the word y 1 ⋯ y k y_1\cdots y_k y 1 ⋯ y k , and κ \kappa κ is a bijection G → A ∣ G ∣ G \to A_{|G|} G → A ∣ G ∣ ; for M 11 M_{11} M 11 this was checked on all 7920 7920 7920 elements. For the elements of Examples 13.4 and 13.5:
Table 7.
For instance κ = 2 + 11 ⋅ 9 + 110 ⋅ 4 + 990 ⋅ 6 = 6481 \kappa=2+11\cdot9+110\cdot4+990\cdot6=6481 κ = 2 + 11 ⋅ 9 + 110 ⋅ 4 + 990 ⋅ 6 = 6481 for the first row. In the transported law, 6481 ∗ 5931 = 346 6481*5931=346 6481 ∗ 5931 = 346 . The images of a product are not determined by the images of its factors: ( g ∘ h ) ( i ) = g ( h ( i ) ) (g\circ h)(i)=g(h(i)) ( g ∘ h ) ( i ) = g ( h ( i )) needs g g g at the points h ( 0 ) , … , h ( 3 ) = 2 , 10 , 9 , 6 h(0),\ldots,h(3)=2,10,9,6 h ( 0 ) , … , h ( 3 ) = 2 , 10 , 9 , 6 , and three of them are not among 0 , 1 , 2 , 3 0,1,2,3 0 , 1 , 2 , 3 . Finding these values means reconstructing g g g from its four images, which is what the chain of Theorem 8.34 does. So the labels κ \kappa κ are canonical, while the labels λ \lambda λ are the ones that are cheap to multiply.
Conjugacy classes
Theorem 4.8 gives the class equation for a group presented by a digital law. This subsection collects the general facts about conjugacy classes that the rest of the book uses, independently of the encoding: orbit–stabiliser for conjugation, the class equation, the number of classes, and the splitting of classes in a subgroup of index two. Power maps, rationality and algorithms follow in Chapter 5, and the classes of Sym n \operatorname{Sym}_n Sym n and Alt n \operatorname{Alt}_n Alt n in § 19.3 .
Definition 4.14. Let G G G be a group. For g , x ∈ G g,x \in G g , x ∈ G write g x = x − 1 g x g^{x}=x^{-1}gx g x = x − 1 g x . The conjugacy class of g g g is g G = { g x : x ∈ G } g^{G}=\{g^{x}:x \in G\} g G = { g x : x ∈ G } , its centraliser is C G ( g ) = { x ∈ G : g x = x g } C_{G}(g)=\{x \in G:gx=xg\} C G ( g ) = { x ∈ G : g x = xg } , and the centre is Z ( G ) = { g : g G = { g } } Z(G)=\{g:g^{G}=\{g\}\} Z ( G ) = { g : g G = { g }} . A function G → C G \to\mathbb{C} G → C constant on classes is a class function .
For permutation codes conjugation is a pure digit relocation: by Proposition 2.6 , S ∘ R ∘ S − 1 S \circ R \circ S^{-1} S ∘ R ∘ S − 1 has digit S ( R ( n ) ) S(R(n)) S ( R ( n )) in position S ( n ) S(n) S ( n ) . Conjugating a code therefore costs p p p digit look-ups, exactly as one composition (12) ; no multiplication table is needed.
Theorem 4.15 (Orbit–stabiliser for conjugation). The map C G ( g ) x ↦ g x C_G(g)x \mapsto g^x C G ( g ) x ↦ g x is a bijection from the right cosets of C G ( g ) C_G(g) C G ( g ) onto g G g^G g G . Hence ∣ g G ∣ = ∣ G : C G ( g ) ∣ |g^G|=|G:C_G(g)| ∣ g G ∣ = ∣ G : C G ( g ) ∣ , and ∣ g G ∣ |g^G| ∣ g G ∣ divides ∣ G ∣ |G| ∣ G ∣ .
Proof. g x = g y g^x=g^y g x = g y iff x − 1 g x = y − 1 g y x^{-1}gx=y^{-1}gy x − 1 g x = y − 1 g y iff y x − 1 yx^{-1} y x − 1 commutes with g g g iff C G ( g ) x = C G ( g ) y C_G(g)x=C_G(g)y C G ( g ) x = C G ( g ) y . So the map is well defined and injective; it is surjective by definition of g G g^G g G . Lagrange’s theorem gives the divisibility.
Corollary 4.16 (Class equation). If g 1 , … , g k g_1,\ldots,g_k g 1 , … , g k represent the noncentral classes, then
∣ G ∣ = ∣ Z ( G ) ∣ + ∑ i = 1 k ∣ G : C G ( g i ) ∣ , |G|=|Z(G)|+\sum_{i=1}^{k}|G:C_G(g_i)|, ∣ G ∣ = ∣ Z ( G ) ∣ + i = 1 ∑ k ∣ G : C G ( g i ) ∣ , each summand dividing ∣ G ∣ |G| ∣ G ∣ and exceeding 1.
Proof. The classes partition G G G ; the singleton classes are exactly the central elements. Apply Theorem 4.15 .
Theorem 4.8 is the same equation, read off the digits of a Cayley table.
Corollary 4.17. A nontrivial finite p p p -group has nontrivial centre. A group of order p 2 p^2 p 2 is abelian.
Proof. In the class equation of ∣ G ∣ = p a |G|=p^a ∣ G ∣ = p a , a ≥ 1 a\ge1 a ≥ 1 , every noncentral summand is a positive power of p p p , so p p p divides ∣ Z ( G ) ∣ ≥ 1 |Z(G)|\ge1 ∣ Z ( G ) ∣ ≥ 1 . If ∣ G ∣ = p 2 |G|=p^2 ∣ G ∣ = p 2 and Z ( G ) ≠ G Z(G)\ne G Z ( G ) = G , then G / Z ( G ) G/Z(G) G / Z ( G ) has order p p p , hence is cyclic, generated by g Z ( G ) gZ(G) g Z ( G ) ; then every element is g i z g^iz g i z with z z z central, and any two such commute, a contradiction.
Theorem 4.18 (Counting classes). The number k ( G ) k(G) k ( G ) of conjugacy classes satisfies
k ( G ) = 1 ∣ G ∣ ∑ g ∈ G ∣ C G ( g ) ∣ = ∣ { ( x , y ) ∈ G 2 : x y = y x } ∣ ∣ G ∣ . k(G)=\frac{1}{|G|}\sum_{g\in G}|C_G(g)|=\frac{|\{(x,y)\in G^2:xy=yx\}|}{|G|}. k ( G ) = ∣ G ∣ 1 g ∈ G ∑ ∣ C G ( g ) ∣ = ∣ G ∣ ∣ {( x , y ) ∈ G 2 : x y = y x } ∣ .
Proof. The second equality counts commuting pairs by their first coordinate. For the first, group the sum by classes: on a class C C C with representative g g g , ∑ x ∈ C ∣ C G ( x ) ∣ = ∣ C ∣ ∣ C G ( g ) ∣ = ∣ G ∣ \sum_{x\in C}|C_G(x)|=|C||C_G(g)|=|G| ∑ x ∈ C ∣ C G ( x ) ∣ = ∣ C ∣∣ C G ( g ) ∣ = ∣ G ∣ by Theorem 4.15 , because conjugate elements have conjugate, hence equipotent, centralisers. Summing ∣ G ∣ |G| ∣ G ∣ over the k ( G ) k(G) k ( G ) classes gives the claim.
The ratio k ( G ) / ∣ G ∣ k(G)/|G| k ( G ) /∣ G ∣ is the probability that two uniformly random elements commute. It is the first hint of a recurring theme: centraliser orders, not class sizes, are the natural currency of computation.
Theorem 4.19 (Splitting). Let H ≤ G H \le G H ≤ G have index 2 and let h ∈ H h \in H h ∈ H . Then h G ⊆ H h^G \subseteq H h G ⊆ H , and
if C G ( h ) ≰ H C_G(h)\not\le H C G ( h ) ≤ H , then h G = h H h^G=h^H h G = h H is a single H H H -class;
if C G ( h ) ≤ H C_G(h) \le H C G ( h ) ≤ H , then h G h^G h G is the disjoint union of two H H H -classes h H h^H h H and ( h t ) H (ht)^H ( h t ) H , for any t ∈ G ∖ H t \in G \setminus H t ∈ G ∖ H , each of size 1 2 ∣ h G ∣ \frac{1}{2}|h^G| 2 1 ∣ h G ∣ .
Proof. H H H is normal, so h G ⊆ H h^G \subseteq H h G ⊆ H . By Theorem 4.15 , ∣ h G ∣ = ∣ G ∣ / ∣ C G ( h ) ∣ |h^G| = |G|/|C_G(h)| ∣ h G ∣ = ∣ G ∣/∣ C G ( h ) ∣ and ∣ h H ∣ = ∣ H ∣ / ∣ C H ( h ) ∣ |h^H| = |H|/|C_H(h)| ∣ h H ∣ = ∣ H ∣/∣ C H ( h ) ∣ with C H ( h ) = C G ( h ) ∩ H C_H(h) = C_G(h) \cap H C H ( h ) = C G ( h ) ∩ H . If C G ( h ) ≰ H C_G(h) \nleq H C G ( h ) ≰ H , then C H ( h ) C_H(h) C H ( h ) has index 2 in C G ( h ) C_G(h) C G ( h ) , so ∣ h H ∣ = ∣ G ∣ / 2 ∣ C G ( h ) ∣ / 2 = ∣ h G ∣ |h^H| = \frac{|G|/2}{|C_G(h)|/2} = |h^G| ∣ h H ∣ = ∣ C G ( h ) ∣/2 ∣ G ∣/2 = ∣ h G ∣ . If C G ( h ) ≤ H C_G(h) \le H C G ( h ) ≤ H , then ∣ h H ∣ = 1 2 ∣ h G ∣ |h^H| = \frac{1}{2}|h^G| ∣ h H ∣ = 2 1 ∣ h G ∣ . Since G = H ∪ t H G = H \cup tH G = H ∪ t H , h G = h H ∪ ( h t ) H h^G = h^H \cup(ht)^H h G = h H ∪ ( h t ) H , and the two pieces have equal size because conjugation by t t t is a bijection carrying one onto the other; they are disjoint because their union has size ∣ h G ∣ = 2 ∣ h H ∣ |h^G| = 2|h^H| ∣ h G ∣ = 2∣ h H ∣ .
Example 4.20. In Alt 5 ≤ Sym 5 \operatorname{Alt}_5 \le\operatorname{Sym}_5 Alt 5 ≤ Sym 5 a 5-cycle has centraliser ⟨ g ⟩ \langle g\rangle ⟨ g ⟩ of order 5 in Sym 5 \operatorname{Sym}_5 Sym 5 , which lies in Alt 5 \operatorname{Alt}_5 Alt 5 , so its class of 24 elements splits into two Alt 5 \operatorname{Alt}_5 Alt 5 -classes of 12. A 3-cycle has centraliser of order 6 in Sym 5 \operatorname{Sym}_5 Sym 5 , which contains a transposition, so the 20 three-cycles remain one class. Hence the class equation of Alt 5 \operatorname{Alt}_5 Alt 5 is 60 = 1 + 15 + 20 + 12 + 12 60 = 1 + 15 + 20 + 12 + 12 60 = 1 + 15 + 20 + 12 + 12 ; see Theorem 11.9 for the general rule.
Counting colourings: Polya–Redfield and de Bruijn
Burnside’s lemma (Theorem 4.6 ) counts the orbits of a group on a set of points. Many counting problems ask instead for the orbits on the colourings of a set of positions, and then the fixed points can be counted cycle by cycle. The two theorems below [55, 56, 57, 58] are used for magmas (§ 7.4 ), necklaces and bracelets (Propositions 7.7 and 8.15 ), the group M 12 M_{12} M 12 (§ 11.11 ) and the symmetric groups (§ 19.3 ).
Let G G G act on a finite set X X X with ∣ X ∣ = s |X| = s ∣ X ∣ = s ; for a digital action, X = A s X = \mathcal{A}_s X = A s and the action has the code Θ \Theta Θ . Each g ∈ G g \in G g ∈ G induces a permutation of X X X . Write c k ( g ) c_k(g) c k ( g ) for its number of cycles of length k k k (fixed points are cycles of length 1) and c ( g ) = ∑ k c k ( g ) c(g) = \sum_k c_k(g) c ( g ) = ∑ k c k ( g ) . The cycle index of the action is
Z ( G ; x 1 , x 2 , … ) = 1 ∣ G ∣ ∑ g ∈ G ∏ k ≥ 1 x k c k ( g ) . Z(G; x_1, x_2, \ldots) = \frac{1}{|G|} \sum_{g \in G} \prod_{k \ge1} x_k^{c_k(g)}. Z ( G ; x 1 , x 2 , … ) = ∣ G ∣ 1 g ∈ G ∑ k ≥ 1 ∏ x k c k ( g ) .
If Y Y Y is a set of m m m colours, G G G acts on the set Y X Y^X Y X of colourings by ( g ⋅ f ) ( x ) = f ( g − 1 x ) (g \cdot f)(x) = f(g^{-1}x) ( g ⋅ f ) ( x ) = f ( g − 1 x ) . In digits, a colouring is a string of s s s base-m m m digits and g g g relocates the digits, just as conjugation relocates the digits of a permutation code (Proposition 2.6 ).
Theorem 4.21 (Pólya–Redfield). Let w : Y → R w : Y \to R w : Y → R take values in a commutative ring R R R and give the colouring f f f the weight w ( f ) = ∏ x ∈ X w ( f ( x ) ) w(f) = \prod_{x \in X} w(f(x)) w ( f ) = ∏ x ∈ X w ( f ( x )) , which is constant on orbits. Then
∑ O ∈ Y X / G w ( O ) = Z ( G ; p 1 , p 2 , … ) , p k = ∑ y ∈ Y w ( y ) k . \sum_{O \in Y^X/G} w(O) = Z(G; p_1, p_2, \ldots), \qquad p_k = \sum_{y \in Y} w(y)^k. O ∈ Y X / G ∑ w ( O ) = Z ( G ; p 1 , p 2 , … ) , p k = y ∈ Y ∑ w ( y ) k . In particular, the number of orbits of G G G on Y X Y^X Y X is
1 ∣ G ∣ ∑ g ∈ G m c ( g ) = Z ( G ; m , m , … ) . \frac{1}{|G|} \sum_{g \in G} m^{c(g)} = Z(G; m, m, \ldots). ∣ G ∣ 1 g ∈ G ∑ m c ( g ) = Z ( G ; m , m , … ) . Proof. A colouring is fixed by g g g if and only if it is constant on every cycle of g g g . Choosing one colour y Γ y_\Gamma y Γ for each cycle Γ \Gamma Γ gives such a colouring, of weight ∏ Γ w ( y Γ ) ∣ Γ ∣ \prod_\Gamma w(y_\Gamma)^{|\Gamma|} ∏ Γ w ( y Γ ) ∣Γ∣ , so the colourings fixed by g g g have total weight ∏ k p k c k ( g ) \prod_k p_k^{c_k(g)} ∏ k p k c k ( g ) . On the other hand, Theorem 4.5 for the action on Y X Y^X Y X gives
∑ g ∈ G ∑ g ⋅ f = f w ( f ) = ∑ f ∣ G f ∣ w ( f ) = ∑ O ∑ f ∈ O ∣ G ∣ ∣ O ∣ w ( f ) = ∣ G ∣ ∑ O w ( O ) . \sum_{g \in G} \sum_{g \cdot f=f} w(f) = \sum_f |G_f|w(f) = \sum_O \sum_{f \in O} \frac{|G|}{|O|}w(f) = |G| \sum_O w(O). g ∈ G ∑ g ⋅ f = f ∑ w ( f ) = f ∑ ∣ G f ∣ w ( f ) = O ∑ f ∈ O ∑ ∣ O ∣ ∣ G ∣ w ( f ) = ∣ G ∣ O ∑ w ( O ) . Divide by ∣ G ∣ |G| ∣ G ∣ ; the choice w ≡ 1 w \equiv1 w ≡ 1 gives the count.
Theorem 4.22 (de Bruijn). Let in addition a group H H H act on the set Y Y Y of m m m colours, and let G × H G \times H G × H act on Y X Y^X Y X by ( ( g , h ) ⋅ f ) ( x ) = h ( f ( g − 1 x ) ) ((g,h) \cdot f)(x) = h(f(g^{-1}x)) (( g , h ) ⋅ f ) ( x ) = h ( f ( g − 1 x )) . For h ∈ H h \in H h ∈ H and k ≥ 1 k \ge1 k ≥ 1 let fix k ( h ) = ∣ { y ∈ Y : h k y = y } ∣ \operatorname{fix}_k(h) = |\{y \in Y : h^k y = y\}| fix k ( h ) = ∣ { y ∈ Y : h k y = y } ∣ , the number of points of Y Y Y on cycles of h h h whose length divides k k k . Then the number of orbits of G × H G \times H G × H on Y X Y^X Y X is
1 ∣ G ∣ ∣ H ∣ ∑ g ∈ G ∑ h ∈ H ∏ k ≥ 1 fix k ( h ) c k ( g ) = 1 ∣ H ∣ ∑ h ∈ H Z ( G ; fix 1 ( h ) , fix 2 ( h ) , … ) . \frac{1}{|G||H|} \sum_{g \in G} \sum_{h \in H} \prod_{k \ge1} \operatorname{fix}_k(h)^{c_k(g)} = \frac{1}{|H|} \sum_{h \in H} Z(G; \operatorname{fix}_1(h), \operatorname{fix}_2(h), \ldots). ∣ G ∣∣ H ∣ 1 g ∈ G ∑ h ∈ H ∑ k ≥ 1 ∏ fix k ( h ) c k ( g ) = ∣ H ∣ 1 h ∈ H ∑ Z ( G ; fix 1 ( h ) , fix 2 ( h ) , … ) .
Proof. A colouring f f f is fixed by ( g , h ) (g,h) ( g , h ) if and only if f ( g x ) = h f ( x ) f(gx)=hf(x) f ( g x ) = h f ( x ) for all x x x . On a cycle x 0 , g x 0 , … , g k − 1 x 0 x_0,gx_0,\ldots,g^{k-1}x_0 x 0 , g x 0 , … , g k − 1 x 0 of g g g of length k k k the value y = f ( x 0 ) y=f(x_0) y = f ( x 0 ) determines f ( g i x 0 ) = h i y f(g^ix_0)=h^iy f ( g i x 0 ) = h i y , and the equation closes up exactly when h k y = y h^ky=y h k y = y ; conversely every such y y y gives a solution on the cycle. Hence ( g , h ) (g,h) ( g , h ) fixes ∏ k fix k ( h ) c k ( g ) \prod_k\operatorname{fix}_k(h)^{c_k(g)} ∏ k fix k ( h ) c k ( g ) colourings, and the claim is Burnside’s lemma Theorem 4.6 for G × H G\times H G × H .
For trivial H H H one has fix k ( h ) = m \operatorname{fix}_k(h)=m fix k ( h ) = m and Theorem 4.22 is the unweighted case of Theorem 4.21 . For H = Sym ( Y ) H=\operatorname{Sym}(Y) H = Sym ( Y ) it counts colourings up to renaming of the colours as well. The same argument, with g g g acting on a set of pairs and h h h on the values, counts operator tables up to relabelling (§ 7.4 ).
Example 4.23 (The square). D 8 D_8 D 8 acts on the vertices A 4 A_4 A 4 of a square by x ↦ x + 1 x\mapsto x+1 x ↦ x + 1 and x ↦ − x x\mapsto-x x ↦ − x modulo 4. The identity has cycle type 1 4 1^4 1 4 , the two quarter turns 4 4 4 , the half turn and the two reflections through the midpoints of opposite edges 2 2 2^2 2 2 , and the two reflections through opposite vertices 2 1 2 2\,1^2 2 1 2 . Hence
Z ( D 8 ) = 1 8 ( x 1 4 + 2 x 4 + 3 x 2 2 + 2 x 1 2 x 2 ) . Z(D_8)=\frac{1}{8}(x_1^4+2x_4+3x_2^2+2x_1^2x_2). Z ( D 8 ) = 8 1 ( x 1 4 + 2 x 4 + 3 x 2 2 + 2 x 1 2 x 2 ) .
and the number of orbits on m m m -colourings is ( m 4 + 2 m 3 + 3 m 2 + 2 m ) / 8 (m^4+2m^3+3m^2+2m)/8 ( m 4 + 2 m 3 + 3 m 2 + 2 m ) /8 , which is 6 for m = 2 m=2 m = 2 and 21 for m = 3 m=3 m = 3 . With the weights w ( white ) = 1 w(\text{white})=1 w ( white ) = 1 , w ( black ) = t w(\text{black})=t w ( black ) = t one has p k = 1 + t k p_k=1+t^k p k = 1 + t k and Theorem 4.21 gives 1 + t + 2 t 2 + t 3 + t 4 1+t+2t^2+t^3+t^4 1 + t + 2 t 2 + t 3 + t 4 : there are two inequivalent ways to colour two vertices black.
Homomorphisms and the isomorphism theorems
Theorem 4.24 (First isomorphism theorem, digital form). Let Ψ \Psi Ψ be the code of a homomorphism ψ \psi ψ from R R R to R ′ R' R ′ with kernel K K K of order k k k , and let q = N / k q=N/k q = N / k . There is a labelling of G G G adapted to K K K with radices ( k , q ) (k,q) ( k , q ) in which the upper layer is closed (it receives no arrows) and
ψ ( g ) = ι ( U 0 ( λ ( g ) ) ) \psi(g)=\iota(U_0(\lambda(g))) ψ ( g ) = ι ( U 0 ( λ ( g ))) for a bijection ι \iota ι of A q A_q A q onto the image. The code R { 1 } R_{\{1\}} R { 1 } of the closed upper layer (Proposition 3.15) is a relabelling of the code of ψ ( G ) \psi(G) ψ ( G ) . Conversely, every closed set of layers defines a homomorphism whose kernel is the ground block below it.
Proof. ψ ( g ) = ψ ( g ′ ) \psi(g)=\psi(g') ψ ( g ) = ψ ( g ′ ) if and only if g ′ ∈ g K g'\in gK g ′ ∈ g K , so by Theorem 4.2 the fibres of ψ \psi ψ are the fibres of the upper layer of a labelling adapted to K K K . Then U 0 ( λ ( a ∗ b ) ) U_0(\lambda(a*b)) U 0 ( λ ( a ∗ b )) is determined by ψ ( a ∗ b ) = ψ ( a ) ψ ( b ) \psi(a*b)=\psi(a)\psi(b) ψ ( a ∗ b ) = ψ ( a ) ψ ( b ) , i.e. by U 0 ( λ ( a ) ) U_0(\lambda(a)) U 0 ( λ ( a )) and U 0 ( λ ( b ) ) U_0(\lambda(b)) U 0 ( λ ( b )) : no arrow enters the upper layer. The converse is Proposition 3.15.
Hence normal subgroups, homomorphic images and closed sets of layers are the same objects, and G / K G/K G / K is the code of the closed upper layer.
Theorem 4.25 (Second isomorphism theorem, digital form). Let H ≤ G H\leq G H ≤ G and K ⊴ G K\unlhd G K ⊴ G , and take a labelling of G G G adapted to K K K with closed upper layer. The restriction of the upper projector U 0 U_0 U 0 to H H H is a homomorphism with kernel H ∩ K H\cap K H ∩ K whose image is the set of upper digits of H K HK H K . Hence the subcode of R { 1 } R_{\{1\}} R { 1 } on these digits is a relabelling of the code of the closed upper layer of a labelling of H H H adapted to H ∩ K H\cap K H ∩ K : H K / K ≅ H / ( H ∩ K ) HK/K\cong H/(H\cap K) H K / K ≅ H / ( H ∩ K ) .
Proof. The restriction of a homomorphism is a homomorphism; its kernel is H ∩ K H\cap K H ∩ K and its image is { U 0 ( λ ( h ) ) } = { U 0 ( λ ( h k ) ) } \{U_0(\lambda(h))\}=\{U_0(\lambda(hk))\} { U 0 ( λ ( h ))} = { U 0 ( λ ( hk ))} . Apply Theorem 4.24 to it.
Theorem 4.26 (Third isomorphism theorem, digital form). Let K ⊴ L K\unlhd L K ⊴ L be normal subgroups of G G G and take the tower labelling of the chain 1 < K < L < G 1<K<L<G 1 < K < L < G , with radices ( ∣ K ∣ , [ L : K ] , [ G : L ] ) (|K|,[L:K],[G:L]) ( ∣ K ∣ , [ L : K ] , [ G : L ]) . Then { 2 } \{2\} { 2 } and { 1 , 2 } \{1,2\} { 1 , 2 } are closed sets of layers, R { 1 , 2 } R_{\{1,2\}} R { 1 , 2 } is the code of G / K G/K G / K , the layer { 2 } \{2\} { 2 } is closed in R { 1 , 2 } R_{\{1,2\}} R { 1 , 2 } , and the code of { 2 } \{2\} { 2 } computed from R { 1 , 2 } R_{\{1,2\}} R { 1 , 2 } coincides with R { 2 } R_{\{2\}} R { 2 } computed from R R R : closed projections compose, π { 2 } = π { 2 } ∘ π { 1 , 2 } \pi_{\{2\}}=\pi_{\{2\}}\circ\pi_{\{1,2\}} π { 2 } = π { 2 } ∘ π { 1 , 2 } , i.e. ( G / K ) / ( L / K ) ≅ G / L (G/K)/(L/K)\cong G/L ( G / K ) / ( L / K ) ≅ G / L .
Proof. By Proposition 3.18 , the upper block above layer 0 (resp. 1) is closed because K K K (resp. L L L ) is normal. Since π { 2 } ( x ) \pi_{\{2\}}(x) π { 2 } ( x ) is a digit of π { 1 , 2 } ( x ) \pi_{\{1,2\}}(x) π { 1 , 2 } ( x ) and both are homomorphisms, the layer 2 receives no arrow from layer 1 in R { 1 , 2 } R_{\{1,2\}} R { 1 , 2 } , and the two codes coincide.
Theorem 4.27 (Correspondence theorem, digital form). Let K ◃ G K \triangleleft G K ◃ G and take a labelling adapted to K K K with closed upper layer. The subgroups of G G G containing K K K are exactly the unions of fibres { x : U 0 ( x ) ∈ S } \{x : U_0(x) \in S\} { x : U 0 ( x ) ∈ S } with S S S a subgroup of R { 1 } R_{\{1\}} R { 1 } ; this correspondence preserves inclusions, indices (the radices) and normality.
Proof. A subgroup containing K K K is a union of cosets of K K K , i.e. of fibres of U 0 U_0 U 0 , and since U 0 U_0 U 0 is a homomorphism the set of their upper digits is a subgroup of R { 1 } R_{\{1\}} R { 1 } ; conversely the preimage of a subgroup is a subgroup. Normality is preserved because U 0 U_0 U 0 is surjective.
Sylow theorems in digital form
Let p p p be a prime and let a a a be the number of trailing zero digits of N N N in base p p p :
d p ( k , N ) = 0 ( k < a ) , d p ( a , N ) ≠ 0 , m = N / p a , d p ( 0 , m ) = d p ( a , N ) . (51) d_p(k,N)=0\quad(k<a),\qquad d_p(a,N)\ne0,\qquad m=N/p^a,\qquad d_p(0,m)=d_p(a,N). \tag*{(51)} d p ( k , N ) = 0 ( k < a ) , d p ( a , N ) = 0 , m = N / p a , d p ( 0 , m ) = d p ( a , N ) . (51)
Lemma 4.28 (Lucas). For all integers n , k ≥ 0 n,k\ge0 n , k ≥ 0 ,
( n k ) ≡ ∏ i ≥ 0 ( d p ( i , n ) d p ( i , k ) ) ( m o d p ) . \binom{n}{k}\equiv\prod_{i\ge0}\binom{d_p(i,n)}{d_p(i,k)}\pmod p. ( k n ) ≡ i ≥ 0 ∏ ( d p ( i , k ) d p ( i , n ) ) ( mod p ) . In particular
( N p a ) ≡ d p ( a , N ) ≠ 0 ( m o d p ) . \binom{N}{p^a}\equiv d_p(a,N)\ne0\pmod p. ( p a N ) ≡ d p ( a , N ) = 0 ( mod p ) .
Proof. Modulo p p p , ( 1 + x ) p ≡ 1 + x p (1+x)^p\equiv1+x^p ( 1 + x ) p ≡ 1 + x p , hence ( 1 + x ) n = ∏ i ( 1 + x p i ) d p ( i , n ) (1+x)^n=\prod_i(1+x^{p^i})^{d_p(i,n)} ( 1 + x ) n = ∏ i ( 1 + x p i ) d p ( i , n ) . By uniqueness of base-p p p expansions, the coefficient of x k x^k x k only receives the contribution of the digits d p ( i , k ) d_p(i,k) d p ( i , k ) . For k = p a k=p^a k = p a the only non-zero digit of k k k is d p ( a , k ) = 1 d_p(a,k)=1 d p ( a , k ) = 1 , which leaves the factor d p ( a , N ) d_p(a,N) d p ( a , N ) .
Remark 4.29. By Kummer’s theorem the exponent of p p p in ( n k ) \binom{n}{k} ( k n ) is the number of carries in the base-p p p addition of k k k and n − k n-k n − k , i.e. the number of times the carry arrows of Z p 2 \mathbb{Z}_{p^2} Z p 2 , with f ( A , B ) = ⌊ ( A + B ) / p ⌋ f(A,B)=\lfloor(A+B)/p\rfloor f ( A , B ) = ⌊( A + B ) / p ⌋ , fire. For n = N n=N n = N and k = p a k=p^a k = p a there is no carry. This was checked for S 4 S_4 S 4 , A 5 A_5 A 5 , M 11 M_{11} M 11 , M 12 M_{12} M 12 and J 1 J_1 J 1 and every prime dividing their orders; for example ( 7920 5 ) ≡ d 5 ( 1 , 7920 ) = 4 ( m o d 5 ) \binom{7920}{5}\equiv d_5(1,7920)=4\pmod5 ( 5 7920 ) ≡ d 5 ( 1 , 7920 ) = 4 ( mod 5 ) for M 11 M_{11} M 11 .
Theorem 4.30 (Sylow I, digital form). For every prime p p p , G G G has a labelling with radices ( p , … , p , m ) (p,\ldots,p,m) ( p , … , p , m ) , with a a a radices equal to p p p , whose a a a lowest layers form a subgroup node of order p a p^a p a in which every upper block is closed, so that inside the node all arrows point downwards.
Proof. Let Ω \Omega Ω be the set of subsets of G G G with p a p^a p a elements, acted on by left multiplication. By Lemma 4.28, d p ( 0 , ∣ Ω ∣ ) ≠ 0 d_p(0,|\Omega|)\ne0 d p ( 0 , ∣Ω∣ ) = 0 , so some orbit has a size with non-zero ground digit. If X X X lies in this orbit and H H H is its stabiliser, Theorem 4.5 gives d p ( 0 , N / ∣ H ∣ ) ≠ 0 d_p(0,N/|H|)\ne0 d p ( 0 , N /∣ H ∣ ) = 0 , hence p a p^a p a divides ∣ H ∣ |H| ∣ H ∣ . On the other hand H x ⊆ X Hx\subseteq X H x ⊆ X for x ∈ X x\in X x ∈ X , so ∣ H ∣ ≤ p a |H|\le p^a ∣ H ∣ ≤ p a . Thus ∣ H ∣ = p a |H|=p^a ∣ H ∣ = p a , and Theorem 4.2 gives a labelling with radices ( p a , m ) (p^a,m) ( p a , m ) whose ground block is H H H . By Theorem 4.8 a group of order p b p^b p b , b ≥ 1 b\ge1 b ≥ 1 , has a central subgroup of order p p p ; by induction H H H has a chain 1 = H 0 < H 1 < ⋯ < H a = H 1=H_0<H_1<\cdots<H_a=H 1 = H 0 < H 1 < ⋯ < H a = H of subgroups that are normal in H H H , with all indices equal to p p p . Its tower labelling refines the ground block, and its upper blocks inside H H H are closed by Proposition 3.18 .
Theorem 4.31 (Sylow II, digital form). Let Q Q Q be a Sylow p p p -subgroup and take a labelling adapted to Q Q Q , with radices ( p a , m ) (p^a,m) ( p a , m ) . For every p p p -subgroup P P P there is an upper digit u ∈ A m u\in A_m u ∈ A m that is not moved by the upward arrow from P P P , i.e. U 0 ( λ ( a ∗ g ) ) = u U_0(\lambda(a*g))=u U 0 ( λ ( a ∗ g )) = u for all a ∈ P a\in P a ∈ P whenever U 0 ( λ ( g ) ) = u U_0(\lambda(g))=u U 0 ( λ ( g )) = u . Consequently P P P is contained in a conjugate of Q Q Q , and all Sylow p p p -subgroups are conjugate.
Proof. By Theorem 4.5 the upward arrow from the left operand is the action of G G G on the cosets of Q Q Q . Restricting it to P P P , Lemma 4.7 gives d p ( 0 , ∣ Fix ∣ ) = d p ( 0 , m ) ≠ 0 d_p(0,|\operatorname{Fix}|)=d_p(0,m)\ne0 d p ( 0 , ∣ Fix ∣ ) = d p ( 0 , m ) = 0 , so some coset g Q gQ g Q is fixed: P g Q = g Q PgQ=gQ P g Q = g Q , i.e. g ˉ ∗ P ∗ g ⊆ Q \bar{g}*P*g\subseteq Q g ˉ ∗ P ∗ g ⊆ Q .
Theorem 4.32 (Sylow III, digital form). If n p n_p n p is the number of Sylow p p p -subgroups, then
d p ( 0 , n p ) = 1 , n p ∣ m . d_p(0,n_p)=1,\qquad n_p\mid m. d p ( 0 , n p ) = 1 , n p ∣ m .
Proof. Let Q Q Q act by conjugation on the set of Sylow p p p -subgroups. If Q Q Q fixes Q ′ Q' Q ′ , then Q Q Q and Q ′ Q' Q ′ are Sylow subgroups of the normaliser of Q ′ Q' Q ′ , in which Q ′ Q' Q ′ is normal; by Theorem 4.31 they are conjugate there, so Q = Q ′ Q=Q' Q = Q ′ . Lemma 4.7 gives d p ( 0 , n p ) = d p ( 0 , 1 ) = 1 d_p(0,n_p)=d_p(0,1)=1 d p ( 0 , n p ) = d p ( 0 , 1 ) = 1 . By Theorem 4.5 , n p = N / ∣ N G ( Q ) ∣ n_p=N/|N_G(Q)| n p = N /∣ N G ( Q ) ∣ , which divides m m m because Q ≤ N G ( Q ) Q\leq N_G(Q) Q ≤ N G ( Q ) .
For A l t 5 \mathrm{Alt}_5 Alt 5 , n 2 = 5 n_2=5 n 2 = 5 , n 3 = 10 n_3=10 n 3 = 10 and n 5 = 6 n_5=6 n 5 = 6 , whose ground digits in bases 2, 3 and 5 are all equal to 1.
Proposition 4.33 (Labels of the Sylow subgroups). Let N = p a m N=p^am N = p a m with p ∤ m p\nmid m p ∤ m , let Q Q Q be a Sylow p p p -subgroup, and take the labelling of Theorem 4.30 , coarsened to radices ( p a , m ) (p^a,m) ( p a , m ) , so that Q = A p a Q=A_{p^a} Q = A p a and the upper digit u ∈ A m u\in A_m u ∈ A m of a label is its left coset of Q Q Q . Let g u g_u g u be the element of label p a u p^au p a u and put
Q u = { g u ∗ h ∗ g ˉ u : h ∈ A p a } , u ∈ A m . Q_u=\{g_u*h*\bar{g}_u:h\in A_{p^a}\},\qquad u\in A_m. Q u = { g u ∗ h ∗ g ˉ u : h ∈ A p a } , u ∈ A m . Then every Q u Q_u Q u is a Sylow p p p -subgroup, every Sylow p p p -subgroup is one of the Q u Q_u Q u , and Q u = Q v Q_u=Q_v Q u = Q v if and only if g ˉ v ∗ g u ∈ N G ( Q ) \bar{g}_v*g_u\in N_G(Q) g ˉ v ∗ g u ∈ N G ( Q ) . Each Sylow p p p -subgroup occurs ∣ N G ( Q ) : Q ∣ = m / n p |N_G(Q):Q|=m/n_p ∣ N G ( Q ) : Q ∣ = m / n p times in the list ( Q u ) u ∈ A m (Q_u)_{u\in A_m} ( Q u ) u ∈ A m .
Proof. Conjugation is an automorphism, so Q u = g u Q g ˉ u Q_u=g_uQ\bar{g}_u Q u = g u Q g ˉ u is a subgroup of order p a p^a p a . If Q ′ Q' Q ′ is a Sylow p p p -subgroup, Theorem 4.31 gives Q ′ = g Q g ˉ Q'=gQ\bar{g} Q ′ = g Q g ˉ for some g g g ; if u u u is the upper digit of λ ( g ) \lambda(g) λ ( g ) , then g = g u ∗ h ′ g=g_u*h' g = g u ∗ h ′ with h ′ ∈ Q h'\in Q h ′ ∈ Q , and h ′ Q h ˉ ′ = Q h'Q\bar{h}'=Q h ′ Q h ˉ ′ = Q gives Q ′ = Q u Q'=Q_u Q ′ = Q u . Next, Q u = Q v Q_u=Q_v Q u = Q v if and only if g ˉ v ∗ g u \bar{g}_v*g_u g ˉ v ∗ g u normalises Q Q Q . The elements g u g_u g u represent the m m m left cosets of Q Q Q , and g u g_u g u , g v g_v g v give the same conjugate exactly when they lie in the same left coset of N G ( Q ) N_G(Q) N G ( Q ) , which is a union of ∣ N G ( Q ) : Q ∣ |N_G(Q):Q| ∣ N G ( Q ) : Q ∣ cosets of Q Q Q . By Theorem 4.32 , ∣ N G ( Q ) : Q ∣ = ( N / n p ) / p a = m / n p |N_G(Q):Q|=(N/n_p)/p^a=m/n_p ∣ N G ( Q ) : Q ∣ = ( N / n p ) / p a = m / n p .
The proposition gives the Sylow subgroups in a labelling adapted to p p p . In any other labelling they are still finite sets of labels, computable from the law: for p p p dividing N N N only once they are the sets { 0 , x , x ∗ x , … } \{0,x,x*x,\ldots\} { 0 , x , x ∗ x , … } generated by the labels x x x of order p p p .
Example 4.34 (Sylow labels of A l t 5 \mathrm{Alt}_5 Alt 5 ). In the Lehmer law of A l t 5 \mathrm{Alt}_5 Alt 5 (Table 11.10 and Table 11.11 ) the Sylow 3- and 5-subgroups are the cyclic subgroups generated by the labels of order 3 and 5, and the Sylow 2-subgroups are the Klein groups { 0 , a , b , a ∗ b } \{0,a,b,a*b\} { 0 , a , b , a ∗ b } formed by two commuting labels of order 2. Computing them from the law gives:
Table 11.10. The Lehmer law of Alt 5 \operatorname{Alt}_5 Alt 5 , columns 0–29. The entry in row m m m and column k k k is m ∗ k m * k m ∗ k ; the second column is the word R ( 0 ) ⋯ R ( 4 ) R(0)\cdots R(4) R ( 0 ) ⋯ R ( 4 ) of R = μ − 1 ( m ) R=\mu^{-1}(m) R = μ − 1 ( m ) .
Table 11.11. The Lehmer law of Alt 5 \operatorname{Alt}_5 Alt 5 , columns 30–59, continuing Table 11.10 .
Table 4.34.
The five Sylow 2-subgroups contain the 15 labels of order 2, the ten Sylow 3-subgroups the 20 labels of order 3, and the six Sylow 5-subgroups the 24 labels of order 5; together with 0 these are the 60 labels, in accordance with the class equation 60 = 1 + 15 + 20 + 12 + 12 60=1+15+20+12+12 60 = 1 + 15 + 20 + 12 + 12 . Distinct Sylow subgroups for the same prime meet only in 0.
Series: Zassenhaus, Schreier and Jordan–Hölder
Definition 4.35 (Subnormal and composition labellings). A subnormal labelling is the tower labelling of a chain 1 = H 0 ⊴ H 1 ⊴ ⋯ ⊴ H M = G 1 = H_0 \trianglelefteq H_1 \trianglelefteq\cdots\trianglelefteq H_M = G 1 = H 0 ⊴ H 1 ⊴ ⋯ ⊴ H M = G in which each subgroup is normal in the next. Inside the block H k + 1 H_{k+1} H k + 1 (the labels with U k = 0 U_k = 0 U k = 0 ), layer k k k is closed, by Proposition 3.18 applied to H k + 1 H_{k+1} H k + 1 ; its code R ( k ) R^{(k)} R ( k ) on p k = [ H k + 1 : H k ] p_k = [H_{k+1} : H_k] p k = [ H k + 1 : H k ] symbols is the layer code , the code of H k + 1 / H k H_{k+1}/H_k H k + 1 / H k . A refinement factorises radices by inserting layers. A composition labelling is a subnormal labelling that admits no proper refinement, i.e. all of whose layer codes are simple: by Theorem 3.17 , each layer code has prime radix or a strongly connected arrow graph for every labelling.
Lemma 4.36 (Zassenhaus butterfly, digital form). Let A ′ ⊴ A ≤ G A' \trianglelefteq A \leq G A ′ ⊴ A ≤ G and B ′ ⊴ B ≤ G B' \trianglelefteq B \leq G B ′ ⊴ B ≤ G , and let D = ( A ′ ∩ B ) ( A ∩ B ′ ) D = (A' \cap B)(A \cap B') D = ( A ′ ∩ B ) ( A ∩ B ′ ) . Then D ⊴ A ∩ B D \trianglelefteq A \cap B D ⊴ A ∩ B , and the following three codes are relabellings of each other:
the closed upper layer of a labelling of A ′ ( A ∩ B ) A'(A \cap B) A ′ ( A ∩ B ) adapted to A ′ ( A ∩ B ′ ) A'(A \cap B') A ′ ( A ∩ B ′ ) ;
the closed upper layer of a labelling of B ′ ( A ∩ B ) B'(A \cap B) B ′ ( A ∩ B ) adapted to B ′ ( A ′ ∩ B ) B'(A' \cap B) B ′ ( A ′ ∩ B ) ;
the closed upper layer of a labelling of A ∩ B A \cap B A ∩ B adapted to D D D .
Proof. Put H = A ∩ B H = A \cap B H = A ∩ B and K = A ′ ( A ∩ B ′ ) K = A'(A \cap B') K = A ′ ( A ∩ B ′ ) . Classically, K ⊴ A ′ ( A ∩ B ) K \trianglelefteq A'(A \cap B) K ⊴ A ′ ( A ∩ B ) , H K = A ′ ( A ∩ B ) HK = A'(A \cap B) H K = A ′ ( A ∩ B ) and, by Dedekind’s modular law, H ∩ K = ( A ′ ∩ B ) ( A ∩ B ′ ) = D H \cap K = (A' \cap B)(A \cap B') = D H ∩ K = ( A ′ ∩ B ) ( A ∩ B ′ ) = D . Theorem 4.25 applied in A ′ ( A ∩ B ) A'(A \cap B) A ′ ( A ∩ B ) shows that the restriction of the upper projector to H H H is surjective with kernel D D D , so codes 1 and 3 coincide up to relabelling. Exchanging the roles of A A A and B B B gives codes 2 and 3. □ \square □
Example 4.37 (A butterfly in C 24 C_{24} C 24 ). We illustrate Lemma 4.36 in C 24 C_{24} C 24 , coded by A 24 A_{24} A 24 . For d ∣ 24 d \mid24 d ∣ 24 the subgroup ⟨ d ⟩ \langle d\rangle ⟨ d ⟩ has characteristic function 1 ⟨ d ⟩ ( x ) = B ( d d ( 0 , x ) ) \mathbf{1}_{\langle d\rangle}(x) = \mathcal{B}(d_{d}(0,x)) 1 ⟨ d ⟩ ( x ) = B ( d d ( 0 , x )) , so by Proposition 1.27 the intersection of ⟨ d ⟩ \langle d\rangle ⟨ d ⟩ and ⟨ e ⟩ \langle e\rangle ⟨ e ⟩ has characteristic function B ( d d ( 0 , x ) ) B ( d e ( 0 , x ) ) = B ( d lcm ( d , e ) ( 0 , x ) ) \mathcal{B}(d_{d}(0,x))\mathcal{B}(d_{e}(0,x)) = \mathcal{B}(d_{\operatorname{lcm}(d,e)}(0,x)) B ( d d ( 0 , x )) B ( d e ( 0 , x )) = B ( d lcm ( d , e ) ( 0 , x )) , while ⟨ d ⟩ ⟨ e ⟩ = ⟨ gcd ( d , e ) ⟩ \langle d\rangle\langle e\rangle= \langle\gcd(d,e)\rangle ⟨ d ⟩ ⟨ e ⟩ = ⟨ g cd( d , e )⟩ . For subgroups H H H , K K K of any group with a law on A n A_n A n one has
1 H K ( x ) = 1 − B ( ∑ h ∈ A n 1 H ( h ) 1 K ( h − 1 x ) ) , \mathbf{1}_{HK}(x) = 1 - \mathcal{B}\left(\sum_{h \in A_n} \mathbf{1}_H(h)\mathbf{1}_K(h^{-1}x)\right), 1 H K ( x ) = 1 − B ( h ∈ A n ∑ 1 H ( h ) 1 K ( h − 1 x ) ) , since the sum counts the factorisations x = h k x = hk x = hk with h ∈ H h \in H h ∈ H , k ∈ K k \in K k ∈ K .
Take A = ⟨ 2 ⟩ A = \langle2\rangle A = ⟨ 2 ⟩ , A ′ = ⟨ 8 ⟩ A' = \langle8\rangle A ′ = ⟨ 8 ⟩ , B = ⟨ 3 ⟩ B = \langle3\rangle B = ⟨ 3 ⟩ and B ′ = ⟨ 12 ⟩ B' = \langle12\rangle B ′ = ⟨ 12 ⟩ , of orders 12, 3, 8 and 2. Then A ∩ B = ⟨ 6 ⟩ A \cap B = \langle6\rangle A ∩ B = ⟨ 6 ⟩ , A ′ ∩ B = { 0 } A' \cap B = \{0\} A ′ ∩ B = { 0 } and A ∩ B ′ = ⟨ 12 ⟩ A \cap B' = \langle12\rangle A ∩ B ′ = ⟨ 12 ⟩ , and the two sides of the butterfly are
A ′ ( A ∩ B ) A ′ ( A ∩ B ′ ) = ⟨ 2 ⟩ ⟨ 4 ⟩ , B ′ ( A ∩ B ) B ′ ( A ′ ∩ B ) = ⟨ 6 ⟩ ⟨ 12 ⟩ . \frac{A'(A \cap B)}{A'(A \cap B')} = \frac{\langle2\rangle}{\langle4\rangle}, \qquad
\frac{B'(A \cap B)}{B'(A' \cap B)} = \frac{\langle6\rangle}{\langle12\rangle}. A ′ ( A ∩ B ′ ) A ′ ( A ∩ B ) = ⟨ 4 ⟩ ⟨ 2 ⟩ , B ′ ( A ′ ∩ B ) B ′ ( A ∩ B ) = ⟨ 12 ⟩ ⟨ 6 ⟩ . The four groups have orders 12, 6, 4 and 2, yet both quotients have order 2.
For a nonabelian case let G = ⟨ r , s ∣ r 4 = s 2 = 1 , s r s = r − 1 ⟩ G = \langle r, s \mid r^4 = s^2 = 1,\ srs = r^{-1}\rangle G = ⟨ r , s ∣ r 4 = s 2 = 1 , sr s = r − 1 ⟩ be dihedral of order 8, and take A = { 1 , r 2 , s , r 2 s } A = \{1, r^2, s, r^2s\} A = { 1 , r 2 , s , r 2 s } , A ′ = 1 A' = 1 A ′ = 1 , B = G B = G B = G and B ′ = ⟨ r ⟩ B' = \langle r\rangle B ′ = ⟨ r ⟩ . The two sides are A / ⟨ r 2 ⟩ A/\langle r^2\rangle A / ⟨ r 2 ⟩ and G / ⟨ r ⟩ G/\langle r\rangle G / ⟨ r ⟩ , both of order 2: with A ′ = 1 A' = 1 A ′ = 1 and B = G B = G B = G the butterfly reduces to the second isomorphism theorem.
Theorem 4.38 (Schreier refinement, digital form). Any two subnormal labellings of G G G have refinements with the same number of layers whose radices and layer codes coincide, up to a permutation of the layers and relabellings of the layer codes.
Proof. Let the chains be ( H k ) k = 0 M (H_k)_{k=0}^{M} ( H k ) k = 0 M and ( K l ) l = 0 L (K_l)_{l=0}^{L} ( K l ) l = 0 L . Insert the layers given by H k , l = H k ( H k + 1 ∩ K l ) H_{k,l} = H_k(H_{k+1} \cap K_l) H k , l = H k ( H k + 1 ∩ K l ) between H k H_k H k and H k + 1 H_{k+1} H k + 1 , and K l , k = K l ( K l + 1 ∩ H k ) K_{l,k} = K_l(K_{l+1} \cap H_k) K l , k = K l ( K l + 1 ∩ H k ) between K l K_l K l and K l + 1 K_{l+1} K l + 1 . Both refined labellings have M L ML M L layers (some of radix 1, which are omitted), and by Lemma 4.36 with A = H k + 1 A = H_{k+1} A = H k + 1 , A ′ = H k A' = H_k A ′ = H k , B = K l + 1 B = K_{l+1} B = K l + 1 , B ′ = K l B' = K_l B ′ = K l , the layer code between H k , l H_{k,l} H k , l and H k , l + 1 H_{k,l+1} H k , l + 1 is a relabelling of the layer code between K l , k K_{l,k} K l , k and K l , k + 1 K_{l,k+1} K l , k + 1 . □ \square □
Theorem 4.39 (Jordan–Hölder, digital form). All composition labellings of G G G have the same number of layers, the same multiset of radices, and the same multiset of layer codes up to relabelling.
Proof. A refinement of a composition labelling can only insert layers of radix 1. Hence Theorem 4.38 matches the layers of any two composition labellings bijectively, with equal radices and relabelled layer codes.
Example 4.40 (Digits as coordinates on composition factors). Let p p p be prime and G = C p N G=C_{p^N} G = C p N , coded by A p N A_{p^N} A p N . Its only composition series is G = G 0 > G 1 > ⋯ > G N = 0 G=G_0>G_1>\cdots>G_N=0 G = G 0 > G 1 > ⋯ > G N = 0 with G k = ⟨ p k ⟩ G_k=\langle p^k\rangle G k = ⟨ p k ⟩ , and by Lemma 1.2 (3)
1 G k ( x ) = ∏ j < k B ( d p ( j , x ) ) , 1_{G_k}(x)=\prod_{j<k}\mathcal{B}(d_p(j,x)), 1 G k ( x ) = j < k ∏ B ( d p ( j , x )) , that is, x ∈ G k x\in G_k x ∈ G k exactly when its lowest k k k digits vanish. For x , y ∈ G k x,y\in G_k x , y ∈ G k no carry reaches position k k k , and reduction modulo p N p^N p N does not change digits below position N N N , so d p ( k , x + y ) ≡ d p ( k , x ) + d p ( k , y ) ( m o d p ) d_p(k,x+y)\equiv d_p(k,x)+d_p(k,y)\pmod p d p ( k , x + y ) ≡ d p ( k , x ) + d p ( k , y ) ( mod p ) . Hence x ↦ d p ( k , x ) x\mapsto d_p(k,x) x ↦ d p ( k , x ) is a homomorphism G k → C p G_k\to C_p G k → C p with kernel G k + 1 G_{k+1} G k + 1 : the k k k -th digit is a coordinate on the composition factor G k / G k + 1 ≅ C p G_k/G_{k+1}\cong C_p G k / G k + 1 ≅ C p , and the N N N digits of x x x correspond to the N N N factors. A series with steps p a p^a p a , a ∣ N a\mid N a ∣ N , corresponds to base-p a p^a p a digits, and its refinement to the composition series is the splitting of each base-p a p^a p a digit into a a a base-p p p digits (Lemma 1.4 ). Cosets of G k G_k G k group integers by their lowest digits, whereas the nested intervals behind the digital expansion group them by their highest digits; carries are what distinguish the two pictures.
Corollary 4.41. G G G is simple if and only if its composition labellings have a single layer. G G G is solvable if and only if the radices of a composition labelling are the prime factors of N N N , counted with multiplicity; then every layer code is a relabelling of d p ( 0 , a + b ) d_p(0,a+b) d p ( 0 , a + b ) .
Example 4.42. The tower labelling of 1 < Z 2 < V < A 4 < S 4 1<Z_2<V<A_4<S_4 1 < Z 2 < V < A 4 < S 4 in § 8.8 is a composition labelling with radices 2 , 2 , 3 , 2 2,2,3,2 2 , 2 , 3 , 2 and layer codes Z 2 , Z 2 , Z 3 , Z 2 Z_2,Z_2,Z_3,Z_2 Z 2 , Z 2 , Z 3 , Z 2 . The tower of M 11 M_{11} M 11 is not subnormal: its upper layers are mixed with the lower ones, and by Corollary 4.41 every composition labelling of M 11 M_{11} M 11 has one layer of radix 7920 7920 7920 .
Exercises
For the law of C 4 C_4 C 4 (addition modulo 4 4 4 ), show that the left translations λ a \lambda_a λ a have the permutation codes 228 228 228 , 57 57 57 , 78 78 78 , 147 147 147 for a = 0 , 1 , 2 , 3 a=0,1,2,3 a = 0 , 1 , 2 , 3 , and verify λ a ∘ λ b = λ a + b \lambda_a\circ\lambda_b=\lambda_{a+b} λ a ∘ λ b = λ a + b on these codes.
Show that every group of prime order p p p has, up to relabelling, the single digital law m ∗ n = d p ( 0 , m + n ) m*n=d_p(0,m+n) m ∗ n = d p ( 0 , m + n ) .
Determine the numbers of Sylow 2 2 2 - and 3 3 3 -subgroups of Alt 4 \operatorname{Alt}_4 Alt 4 and of Sym 4 \operatorname{Sym}_4 Sym 4 .
Show that every group of order 15 15 15 is cyclic, and more generally that a group of order p q pq pq with primes p < q p<q p < q and p ∤ q − 1 p\nmid q-1 p ∤ q − 1 is cyclic.
Let Sym 4 \operatorname{Sym}_4 Sym 4 act on the 2 2 2 -element subsets of A 4 A_4 A 4 . Find the stabiliser of { 0 , 1 } \{0,1\} { 0 , 1 } and verify the orbit–stabiliser theorem.
Find a composition series of Sym 4 \operatorname{Sym}_4 Sym 4 and its composition factors.
In C 12 C_{12} C 12 refine the series 0 < ⟨ 6 ⟩ < ⟨ 2 ⟩ < C 12 0<\langle6\rangle<\langle2\rangle<C_{12} 0 < ⟨ 6 ⟩ < ⟨ 2 ⟩ < C 12 and 0 < ⟨ 4 ⟩ < C 12 0<\langle4\rangle<C_{12} 0 < ⟨ 4 ⟩ < C 12 to equivalent series, as Schreier’s theorem predicts.
Exercises
The rotation group of the cube acts on its six faces. Find its cycle index, show that the number of colourings of the faces with m m m colours up to rotation is ( m 6 + 3 m 4 + 12 m 3 + 8 m 2 ) / 24 (m^{6} + 3m^{4} + 12m^{3} + 8m^{2})/24 ( m 6 + 3 m 4 + 12 m 3 + 8 m 2 ) /24 , and compute it for m = 2 , 3 m = 2, 3 m = 2 , 3 .
Let Sym 4 \operatorname{Sym}_{4} Sym 4 act on the six 2 2 2 -element subsets of A 4 A_{4} A 4 , the possible edges of a graph on four vertices. Colour each subset by 0 or 1 and show that there are 11 graphs on four vertices up to isomorphism.
The Schreier–Sims algorithm and conjugacy classes
A permutation group of degree n n n can have order up to n ! n! n ! , yet almost every question about it (its order, membership, stabilisers, random elements, centralisers) can be answered in time polynomial in n n n and the number of generators. The tool is a stabiliser chain , computed by the Schreier–Sims algorithm. This chapter gives the algorithm with complete proofs, both in its deterministic and in its randomised form, and explains why it is exactly the digital tower labelling of Chapter 3 applied to the chain of point stabilisers: the Chain computations of Chapter 13 are instances of it. Throughout, permutations compose from right to left: ( g h ) ( b ) = g ( h ( b ) ) (gh)(b) = g(h(b)) ( g h ) ( b ) = g ( h ( b )) . We write g ⋅ b = g ( b ) g \cdot b = g(b) g ⋅ b = g ( b ) ; in a permutation code, g ( b ) g(b) g ( b ) is the digit of g g g at position b b b . Every numerical claim is checked by scripts/ss_verify.py, cross-checked against GAP 4.14.
The second half of the chapter uses the stabiliser chain to compute conjugacy classes: § 11.9 –§ 11.11 treat power maps, rationality and fusion, the algorithms for classes, and a complete example for M 12 M_{12} M 12 .
Bases, stabiliser chains and digital labels
Definition 5.1. Let G ≤ Sym n G \leq\operatorname{Sym}_n G ≤ Sym n . A sequence B = ( b 1 , … , b k ) B = (b_1,\ldots,b_k) B = ( b 1 , … , b k ) of points is a base if only the identity fixes all of them. Put G ( 1 ) = G G^{(1)} = G G ( 1 ) = G and G ( i + 1 ) = G b 1 , … , b i G^{(i+1)} = G_{b_1,\ldots,b_i} G ( i + 1 ) = G b 1 , … , b i , the pointwise stabiliser of b 1 , … , b i b_1,\ldots,b_i b 1 , … , b i ; so G = G ( 1 ) ≥ G ( 2 ) ≥ ⋯ ≥ G ( k + 1 ) = 1 G = G^{(1)} \geq G^{(2)} \geq\cdots\geq G^{(k+1)} = 1 G = G ( 1 ) ≥ G ( 2 ) ≥ ⋯ ≥ G ( k + 1 ) = 1 . The basic orbits are Δ i = G ( i ) ⋅ b i \Delta_i = G^{(i)} \cdot b_i Δ i = G ( i ) ⋅ b i . A generating set S S S of G G G is a strong generating set relative to B B B if ⟨ S ∩ G ( i ) ⟩ = G ( i ) \langle S \cap G^{(i)}\rangle= G^{(i)} ⟨ S ∩ G ( i ) ⟩ = G ( i ) for every i i i ; the pair ( B , S ) (B,S) ( B , S ) is a BSGS.
Theorem 5.2 (Factorisation). For each i i i choose a transversal U i = { u i , β : β ∈ Δ i } ⊆ G ( i ) U_i = \{u_{i,\beta} : \beta\in\Delta_i\} \subseteq G^{(i)} U i = { u i , β : β ∈ Δ i } ⊆ G ( i ) with u i , β ( b i ) = β u_{i,\beta}(b_i) = \beta u i , β ( b i ) = β . Then every g ∈ G g \in G g ∈ G has a unique expression
g = u 1 , β 1 u 2 , β 2 ⋯ u k , β k , β i ∈ Δ i , g = u_{1,\beta_1}u_{2,\beta_2}\cdots u_{k,\beta_k}, \qquad\beta_i \in\Delta_i, g = u 1 , β 1 u 2 , β 2 ⋯ u k , β k , β i ∈ Δ i , and
∣ G ∣ = ∏ i = 1 k ∣ Δ i ∣ . |G| = \prod_{i=1}^{k} |\Delta_i|. ∣ G ∣ = i = 1 ∏ k ∣ Δ i ∣. Proof. By the orbit–stabiliser theorem the cosets of G ( i + 1 ) G^{(i+1)} G ( i + 1 ) in G ( i ) G^{(i)} G ( i ) correspond to the points of Δ i \Delta_i Δ i , with x G ( i + 1 ) ↦ x ( b i ) xG^{(i+1)} \mapsto x(b_i) x G ( i + 1 ) ↦ x ( b i ) , so U i U_i U i is a left transversal and ∣ G ( i ) : G ( i + 1 ) ∣ = ∣ Δ i ∣ |G^{(i)} : G^{(i+1)}| = |\Delta_i| ∣ G ( i ) : G ( i + 1 ) ∣ = ∣ Δ i ∣ . Given g ∈ G ( i ) g \in G^{(i)} g ∈ G ( i ) , the point β i = g ( b i ) \beta_i = g(b_i) β i = g ( b i ) is forced, and u i , β i − 1 g ∈ G ( i + 1 ) u_{i,\beta_i}^{-1}g \in G^{(i+1)} u i , β i − 1 g ∈ G ( i + 1 ) . Induction on i i i gives existence and uniqueness, ending at G ( k + 1 ) = 1 G^{(k+1)} = 1 G ( k + 1 ) = 1 . Multiplying the indices gives ∣ G ∣ |G| ∣ G ∣ .
Corollary 5.3 (Digital chain label). Number each Δ i \Delta_i Δ i increasingly as Δ i = { δ i , 0 < ⋯ < δ i , ∣ Δ i ∣ − 1 } \Delta_i=\{\delta_{i,0}<\cdots<\delta_{i,|\Delta_i|-1}\} Δ i = { δ i , 0 < ⋯ < δ i , ∣ Δ i ∣ − 1 } and write β i = δ i , a i \beta_i=\delta_{i,a_i} β i = δ i , a i . Then
λ ( g ) = ∑ i = 1 k a i ∏ j < i ∣ Δ j ∣ \lambda(g)=\sum_{i=1}^{k}a_i\prod_{j<i}|\Delta_j| λ ( g ) = i = 1 ∑ k a i j < i ∏ ∣ Δ j ∣ is a bijection G → [ 0 , ∣ G ∣ ) G\to[0,|G|) G → [ 0 , ∣ G ∣ ) , and a i = D i ( λ ( g ) ) a_i=D_i(\lambda(g)) a i = D i ( λ ( g )) is its i i i -th mixed-radix digit with radices ∣ Δ 1 ∣ , … , ∣ Δ k ∣ |\Delta_1|,\ldots,|\Delta_k| ∣ Δ 1 ∣ , … , ∣ Δ k ∣ .
Proof. Theorem 5.2 makes g ↦ ( a 1 , … , a k ) g\mapsto(a_1,\ldots,a_k) g ↦ ( a 1 , … , a k ) a bijection onto ∏ i [ 0 , ∣ Δ i ∣ ) \prod_i[0,|\Delta_i|) ∏ i [ 0 , ∣ Δ i ∣ ) , and mixed-radix notation is a bijection from that product onto [ 0 , ∏ i ∣ Δ i ∣ ) [0,\prod_i|\Delta_i|) [ 0 , ∏ i ∣ Δ i ∣ ) .
This is the tower labelling of Chapter 3 for the chain of point stabilisers, and it is the label of Theorem 8.34 . For a sharply k k k -transitive group the radices are n , n − 1 , … , n − k + 1 n,n-1,\ldots,n-k+1 n , n − 1 , … , n − k + 1 and λ ( g ) \lambda(g) λ ( g ) is just a code of the images ( g ( b 1 ) , … , g ( b k ) ) (g(b_1),\ldots,g(b_k)) ( g ( b 1 ) , … , g ( b k )) , as for M 11 M_{11} M 11 and M 12 M_{12} M 12 in Proposition 13.1 . The script confirms that for M 12 M_{12} M 12 the label is a bijection from all 95040 elements onto [ 0 , 95040 ) [0,95040) [ 0 , 95040 ) with radices 12, 11, 10, 9, 8.
Orbits and Schreier trees
The basic orbits and transversals are computed by breadth-first search. Starting from { b } \{b\} { b } with u b = 1 u_b=1 u b = 1 , for every point β \beta β already reached and every generator s s s , if s ( β ) s(\beta) s ( β ) is new put u s ( β ) = s u β u_{s(\beta)}=su_\beta u s ( β ) = s u β . This visits each point once and costs O ( ∣ Δ ∣ ∣ S ∣ ) O(|\Delta||S|) O ( ∣Δ∣∣ S ∣ ) image computations. Storing all u β u_\beta u β costs O ( n ∣ Δ ∣ ) O(n|\Delta|) O ( n ∣Δ∣ ) memory; a Schreier tree stores only the edge label s s s by which β \beta β was first reached, and rebuilds u β u_\beta u β by walking to the root. The tree trades memory for time proportional to its depth, which is why depth-reducing generators are added in large computations such as those in Chapter 13.
Schreier’s lemma
Theorem 5.4 (Schreier). Let G = ⟨ S ⟩ G=\langle S\rangle G = ⟨ S ⟩ , let b b b be a point with orbit Δ \Delta Δ and transversal { u β } \{u_\beta\} { u β } , u b = 1 u_b=1 u b = 1 , and write x ‾ = u x ( b ) \overline{x}=u_x(b) x = u x ( b ) . Then G b G_b G b is generated by the Schreier generators
s u β ‾ − 1 s u β , β ∈ Δ , s ∈ S . \overline{su_\beta}^{-1}su_\beta,\qquad\beta\in\Delta,\ s\in S. s u β − 1 s u β , β ∈ Δ , s ∈ S .
Proof. Each Schreier generator h = s u β ‾ − 1 s u β h=\overline{su_\beta}^{-1}su_\beta h = s u β − 1 s u β fixes b b b , since
h ( b ) = s u β ‾ − 1 ( s ( β ) ) = b . h(b)=\overline{su_\beta}^{-1}(s(\beta))=b. h ( b ) = s u β − 1 ( s ( β )) = b .
Conversely let g = s m ⋯ s 2 s 1 ∈ G b g=s_m\cdots s_2s_1\in G_b g = s m ⋯ s 2 s 1 ∈ G b , a positive word (possible since s − 1 = s o ( s ) − 1 s^{-1}=s^{o(s)-1} s − 1 = s o ( s ) − 1 ). Put v 0 = 1 v_0=1 v 0 = 1 and v j = s j v j − 1 v_j=s_jv_{j-1} v j = s j v j − 1 , so each v j v_j v j is a transversal element. Then
g = v m ( v m − 1 s m v m − 1 ) ( v m − 1 − 1 s m − 1 v m − 2 ) ⋯ ( v 1 − 1 s 1 v 0 ) , g=v_m(v_m^{-1}s_mv_{m-1})(v_{m-1}^{-1}s_{m-1}v_{m-2})\cdots(v_1^{-1}s_1v_0), g = v m ( v m − 1 s m v m − 1 ) ( v m − 1 − 1 s m − 1 v m − 2 ) ⋯ ( v 1 − 1 s 1 v 0 ) ,
by telescoping. Each factor is a Schreier generator, and v m = g ‾ = u g ( b ) = u b = 1 v_m=\overline{g}=u_g(b)=u_b=1 v m = g = u g ( b ) = u b = 1 .
Schreier’s lemma alone gives a correct but impractical algorithm: the ∣ Δ ∣ ∣ S ∣ |\Delta||S| ∣Δ∣∣ S ∣ generators of G b G_b G b produce ∣ Δ ′ ∣ ∣ Δ ∣ ∣ S ∣ |\Delta'||\Delta||S| ∣ Δ ′ ∣∣Δ∣∣ S ∣ generators at the next level, and so on. Sims’s idea is to filter them through the part of the chain already built.
Sifting
Definition 5.5. Suppose partial data are given: points b 1 , … , b k b_1,\ldots,b_k b 1 , … , b k , sets S i S_i S i , and for each i i i the orbit Δ ~ i = ⟨ S i ⟩ ⋅ b i \widetilde{\Delta}_i=\langle S_i\rangle\cdot b_i Δ i = ⟨ S i ⟩ ⋅ b i with transversal U ~ i \widetilde{U}_i U i . To sift g g g from level i i i : for j = i , … , k j=i,\ldots,k j = i , … , k , if β = g ( b j ) ∉ Δ ~ j \beta=g(b_j)\notin\widetilde{\Delta}_j β = g ( b j ) ∈ / Δ j stop and return ( g , j ) (g,j) ( g , j ) ; otherwise replace g g g by u j , β − 1 g u_{j,\beta}^{-1}g u j , β − 1 g . If all levels are passed, return ( g , k + 1 ) (g,k+1) ( g , k + 1 ) . The returned element is the residue.
Proposition 5.6 (Membership). If ( B , S ) (B,S) ( B , S ) is a BSGS of G G G and x ∈ Sym n x\in\operatorname{Sym}_n x ∈ Sym n , then x ∈ G x\in G x ∈ G if and only if x x x sifts to the identity at level k + 1 k+1 k + 1 . A full sift costs O ( k n ) O(kn) O ( k n ) .
Proof. If x ∈ G x\in G x ∈ G , then at each step the current element lies in G ( j ) G^{(j)} G ( j ) , so x ( b j ) ∈ Δ j x(b_j)\in\Delta_j x ( b j ) ∈ Δ j and the step succeeds; at the end it lies in G ( k + 1 ) = 1 G^{(k+1)}=1 G ( k + 1 ) = 1 . Conversely, if the sift ends in the identity then x x x is the product of the transversal elements used, which lie in G G G .
The deterministic algorithm
Theorem 5.7 (Schreier–Sims criterion). Let b 1 , … , b k b_1,\ldots,b_k b 1 , … , b k and S 1 ⊇ S 2 ⊇ ⋯ ⊇ S k S_1\supseteq S_2\supseteq\cdots\supseteq S_k S 1 ⊇ S 2 ⊇ ⋯ ⊇ S k be given with ⟨ S 1 ⟩ = G \langle S_1\rangle=G ⟨ S 1 ⟩ = G , S i + 1 ⊆ S i ∩ G b i S_{i+1}\subseteq S_i\cap G_{b_i} S i + 1 ⊆ S i ∩ G b i , and no nonidentity element of S k S_k S k fixing b 1 , … , b k b_1,\ldots,b_k b 1 , … , b k pointwise. Put H i = ⟨ S i ⟩ H_i=\langle S_i\rangle H i = ⟨ S i ⟩ . If for every i i i every Schreier generator of H i H_i H i relative to b i b_i b i sifts to the identity from level i + 1 i+1 i + 1 , then H i = G ( i ) H_i=G^{(i)} H i = G ( i ) for all i i i , so ( B , S 1 ) (B,S_1) ( B , S 1 ) is a BSGS.
Proof. We show H i + 1 = ( H i ) b i H_{i+1}=(H_i)_{b_i} H i + 1 = ( H i ) b i for all i i i , by downward induction; together with H 1 = G H_1=G H 1 = G this gives H i = G ( i ) H_i=G^{(i)} H i = G ( i ) . For i = k i=k i = k , a Schreier generator of H k H_k H k that sifts to the identity from level k + 1 k+1 k + 1 is the identity, so ( H k ) b k = 1 (H_k)_{b_k}=1 ( H k ) b k = 1 by Theorem 5.4 ; and H k + 1 = 1 H_{k+1}=1 H k + 1 = 1 . Suppose H j + 1 = ( H j ) b j H_{j+1}=(H_j)_{b_j} H j + 1 = ( H j ) b j for all j > i j>i j > i . Then the data from level i + 1 i+1 i + 1 on form a BSGS of H i + 1 H_{i+1} H i + 1 , so by Proposition 5.6 an element sifts to the identity from level i + 1 i+1 i + 1 iff it lies in H i + 1 H_{i+1} H i + 1 . All Schreier generators of H i H_i H i do, hence ( H i ) b i ≤ H i + 1 (H_i)_{b_i}\leq H_{i+1} ( H i ) b i ≤ H i + 1 by Theorem 5.4 ; the reverse inclusion holds because S i + 1 ⊆ G b i ∩ H i S_{i+1}\subseteq G_{b_i}\cap H_i S i + 1 ⊆ G b i ∩ H i .
The algorithm enforces the hypothesis level by level from the bottom.
Choose b 1 , … , b k b_1,\ldots,b_k b 1 , … , b k so that each generator moves some b i b_i b i ; set S i S_i S i to the generators fixing b 1 , … , b i − 1 b_1,\ldots,b_{i-1} b 1 , … , b i − 1 ; compute the orbits.
Put i = k i=k i = k . While i ≥ 1 i\geq1 i ≥ 1 : for each β ∈ Δ ~ i \beta\in\widetilde{\Delta}_i β ∈ Δ i and s ∈ S i s\in S_i s ∈ S i sift the Schreier generator h = s u β ‾ − 1 s u β h=\overline{su_\beta}^{-1}su_\beta h = s u β − 1 s u β from level i + 1 i+1 i + 1 . If the residue r ≠ 1 r\neq1 r = 1 stops at level j j j , then (if j = k + 1 j=k+1 j = k + 1 ) append a new base point moved by r r r , add r r r to S i + 1 , … , S j S_{i+1},\ldots,S_j S i + 1 , … , S j , recompute those orbits, set i = j i=j i = j and restart the level. If all Schreier generators at level i i i sift to 1 1 1 , put i = i − 1 i=i-1 i = i − 1 .
The algorithm terminates because every addition strictly enlarges some ⟨ S l ⟩ \langle S_l\rangle ⟨ S l ⟩ , which can happen at most log 2 n ! \log_2 n! log 2 n ! times along the chain; at termination Theorem 5.7 applies. With stored transversals the running time is O ( n 2 ∣ S ∣ log 3 ∣ G ∣ + n 5 ) O(n^2|S|\log^3|G|+n^5) O ( n 2 ∣ S ∣ log 3 ∣ G ∣ + n 5 ) and the memory O ( n 2 log ∣ G ∣ + n ∣ S ∣ ) O(n^2\log|G|+n|S|) O ( n 2 log ∣ G ∣ + n ∣ S ∣ ) ; the base length satisfies k ≤ log 2 ∣ G ∣ k\leq\log_2|G| k ≤ log 2 ∣ G ∣ .
Example 5.8 (Hand trace for Sym 4 \operatorname{Sym}_4 Sym 4 ). Orbits and transversals in this trace are built by breadth-first search over S S S itself, without adjoining S − 1 S^{-1} S − 1 , and the generators are tried in the order s s s , then t t t ; this is the convention of scripts/ssverify.py, and it produces the representatives below. Take S = { s = ( 0 1 ) , t = ( 0 1 2 3 ) } S=\{s=(0\ 1),\ t=(0\ 1\ 2\ 3)\} S = { s = ( 0 1 ) , t = ( 0 1 2 3 )} . Both move 0 0 0 , so b 1 = 0 b_1=0 b 1 = 0 , S 1 = S S_1=S S 1 = S , S 2 = ∅ S_2=\varnothing S 2 = ∅ . The orbit of 0 0 0 is reached as 0 → s 1 → t 2 → t 3 0\xrightarrow{s}1\xrightarrow{t}2\xrightarrow{t}3 0 s 1 t 2 t 3 , giving u 0 = 1 u_0=1 u 0 = 1 , u 1 = s u_1=s u 1 = s , u 2 = t s u_2=ts u 2 = t s , u 3 = t 2 s u_3=t^2s u 3 = t 2 s . The Schreier generator for ( β , x ) = ( 0 , t ) (\beta,x)=(0,t) ( β , x ) = ( 0 , t ) is u 1 − 1 t = s t u_1^{-1}t=st u 1 − 1 t = s t , which fixes 0 0 0 and acts on { 1 , 2 , 3 } \{1,2,3\} { 1 , 2 , 3 } as a 3-cycle; it does not sift (level 2 is empty), so it becomes the first generator of S 2 S_2 S 2 with new base point b 2 = 1 b_2=1 b 2 = 1 , orbit { 1 , 2 , 3 } \{1,2,3\} { 1 , 2 , 3 } . Continuing, a Schreier generator of H 2 H_2 H 2 produces a transposition of { 2 , 3 } \{2,3\} { 2 , 3 } at level 3 with b 3 = 2 b_3=2 b 3 = 2 and orbit { 2 , 3 } \{2,3\} { 2 , 3 } . Then every Schreier generator sifts through, and ∣ Sym 4 ∣ = 4 ⋅ 3 ⋅ 2 |\operatorname{Sym}_4|=4\cdot3\cdot2 ∣ Sym 4 ∣ = 4 ⋅ 3 ⋅ 2 . (The script records the exact base it finds; the order and orbit lengths are independent of such choices.)
The randomised algorithm
In practice one sifts random elements instead of all Schreier generators.
Lemma 5.9 (Error bound). If the current data are not a BSGS of G G G , then a uniformly random g ∈ G g \in G g ∈ G sifts to the identity with probability at most 1 2 \frac{1}{2} 2 1 .
Proof. With H i = ⟨ S i ⟩ H_i = \langle S_i\rangle H i = ⟨ S i ⟩ as before, the data form a BSGS exactly when H i + 1 = ( H i ) b i H_{i+1} = (H_i)^{b_i} H i + 1 = ( H i ) b i for all i i i (and H 1 = G H_1 = G H 1 = G ). Otherwise let i i i be the largest index with H i + 1 ≠ ( H i ) b i H_{i+1} \ne(H_i)^{b_i} H i + 1 = ( H i ) b i ; as in the proof of Theorem 5.7 the levels after i i i form a BSGS of H i + 1 H_{i+1} H i + 1 . Elements that pass levels 1 , … , i 1,\ldots,i 1 , … , i correspond bijectively to pairs (transversal word, residue r ∈ G ( i + 1 ) r \in G^{(i+1)} r ∈ G ( i + 1 ) ), so for uniform g g g , conditioned on reaching level i + 1 i+1 i + 1 , the residue is uniform in G ( i + 1 ) G^{(i+1)} G ( i + 1 ) . It sifts to the identity iff r ∈ H i + 1 r \in H_{i+1} r ∈ H i + 1 . But H i + 1 < ( H i ) b i ≤ G ( i + 1 ) H_{i+1} < (H_i)^{b_i} \le G^{(i+1)} H i + 1 < ( H i ) b i ≤ G ( i + 1 ) , so ∣ G ( i + 1 ) : H i + 1 ∣ ≥ 2 \lvert G^{(i+1)} : H_{i+1}\rvert\ge2 ∣ G ( i + 1 ) : H i + 1 ∣ ≥ 2 .
Theorem 5.10 (Randomised Schreier–Sims). Sift random elements, adding every nonidentity residue as in the deterministic algorithm, and stop after w w w consecutive elements sift to the identity. If the elements are uniform, the result is a BSGS with probability at least 1 − 2 − w 1 - 2^{-w} 1 − 2 − w . If ∣ G ∣ \lvert G\rvert ∣ G ∣ is known in advance, stop as soon as ∏ i ∣ Δ ~ i ∣ = ∣ G ∣ \prod_i \lvert\widetilde{\Delta}_i\rvert= \lvert G\rvert ∏ i ∣ Δ i ∣ = ∣ G ∣ ; the output is then certainly correct.
Proof. By Lemma 5.9 , while the data are incomplete each trial succeeds with probability at most 1 2 \frac{1}{2} 2 1 , independently; w w w successes in a row then have probability at most 2 − w 2^{-w} 2 − w . For the certified version: by Theorem 5.2 applied to the sifting data, the products of transversal elements form ∏ i ∣ Δ ~ i ∣ \prod_i \lvert\widetilde{\Delta}_i\rvert ∏ i ∣ Δ i ∣ distinct elements of G G G ; if this equals ∣ G ∣ \lvert G\rvert ∣ G ∣ every element sifts to the identity, which by Lemma 5.9 is impossible for incomplete data.
Random elements are produced by the product replacement algorithm: keep a list of group elements, repeatedly replace one entry by its product with another, and output a running product. Its output is not exactly uniform, which is why the certified stopping rule (known order) or a final deterministic verification is used whenever correctness matters. In Chapter 13 the order is always known and the Chain class stops by that rule.
What a BSGS gives
Once a BSGS is known, the following run in time polynomial in n n n .
Uniform random elements: choose β i ∈ Δ i \beta_i \in\Delta_i β i ∈ Δ i uniformly and independently and output u k , β k ⋯ u 1 , β 1 u_{k,\beta_k}\cdots u_{1,\beta_1} u k , β k ⋯ u 1 , β 1 ; by Theorem 5.2 this is exactly uniform. This is the reliable sampler required by the random class search of § 11.10 .
Stabilisers and transporters: G ( i ) G^{(i)} G ( i ) comes for free; for another point, change the base so that it comes first (base change by conjugation or by swapping adjacent base points), and u 1 , β u_{1,\beta} u 1 , β is an element mapping b 1 b_1 b 1 to β \beta β .
Worked results
Table 5.1 lists the outcome for five groups. The generators are:
S y m 4 \mathrm{Sym}_4 Sym 4 : ( 0 1 ) (0\ 1) ( 0 1 ) and ( 0 1 2 3 ) (0\ 1\ 2\ 3) ( 0 1 2 3 ) ;
A l t 5 \mathrm{Alt}_5 Alt 5 : ( 0 1 2 3 4 ) (0\ 1\ 2\ 3\ 4) ( 0 1 2 3 4 ) and ( 0 1 2 ) (0\ 1\ 2) ( 0 1 2 ) ;
P S L ( 2 , 7 ) \mathrm{PSL}(2,7) PSL ( 2 , 7 ) : x ↦ x + 1 x \mapsto x+1 x ↦ x + 1 and x ↦ − 1 / x x \mapsto-1/x x ↦ − 1/ x on F 7 ∪ { ∞ } \mathbb{F}_7 \cup\{\infty\} F 7 ∪ { ∞ } , with ∞ \infty ∞ coded as 7;
M 12 M_{12} M 12 : the reversal and the Mongean shuffle of § 11.11 ;
M 11 M_{11} M 11 : the strong generators of the stabiliser of the first base point of M 12 M_{12} M 12 , renumbered on 11 points, a group the algorithm itself produced.
The column Schreier counts the nontrivial Schreier generators sifted by the deterministic algorithm; rand. the random elements sifted with the rule w = 20 w=20 w = 20 ; rand.+∣ G ∣ |G| ∣ G ∣ those needed with the certified rule.
Table 5.1. Schreier–Sims runs (generated by scripts/ss_verify.py; orders and basic orbits cross-checked in GAP).
The basic orbits read off the transitivity directly: M 12 M_{12} M 12 has radices 12, 11, 10, 9, 8 (sharply 5-transitive), M 11 M_{11} M 11 has 11, 10, 9, 8 (sharply 4-transitive), and P S L ( 2 , 7 ) \mathrm{PSL}(2,7) PSL ( 2 , 7 ) has 8, 7, 3: it is 2-transitive on the projective line and the stabiliser of two points is the group x ↦ a x x \mapsto ax x ↦ a x , a a a a nonzero square, of order 3. The randomised runs need only a few dozen random elements, while the deterministic version sifts many more Schreier generators even for these small groups: that difference grows quickly with the degree. Finally, all 95040 elements of M 12 M_{12} M 12 sift to the identity, while random elements of S y m 12 \mathrm{Sym}_{12} Sym 12 are rejected, as expected since ∣ S y m 12 : M 12 ∣ = 5040 |\mathrm{Sym}_{12}:M_{12}|=5040 ∣ Sym 12 : M 12 ∣ = 5040 .
Power maps, real and rational classes, fusion
Conjugacy classes are the coarsest invariant that still sees the whole multiplication of a group: they index the columns of the character table, they control the centre, the p p p -structure and the fusion of subgroups, and almost every efficient algorithm on a finite group begins by fixing a list of class representatives. The basic theory (orbit–stabiliser, the class equation, counting classes and splitting in subgroups of index two) is in § 10.4.3 , and the classes of S y m n \mathrm{Sym}_n Sym n and A l t n \mathrm{Alt}_n Alt n are in § 19.3 . The rest of this chapter takes these as given, develops power maps, rationality and fusion with complete proofs, and then turns it all into algorithms. All numerical claims are verified by scripts/cc_verify.py, which works directly on permutation codes and is cross-checked against GAP 4.14.
Definition 5.11. For an integer m m m the m m m -th power map sends the class of g g g to the class of g m g^m g m ; it is well defined because ( g x ) m = ( g m ) x (g^x)^m=(g^m)^x ( g x ) m = ( g m ) x . A class is real if it contains g − 1 g^{-1} g − 1 together with g g g , and rational if it contains every generator g m g^m g m , gcd ( m , o ( g ) ) = 1 \gcd(m,o(g))=1 g cd( m , o ( g )) = 1 , of ⟨ g ⟩ \langle g\rangle ⟨ g ⟩ .
For codes, g m g^m g m is computed by binary powering with (12) , using O ( p log m ) O(p\log m) O ( p log m ) digit look-ups. For m m m coprime to o ( g ) o(g) o ( g ) the power map permutes the classes of elements of order o ( g ) o(g) o ( g ) ; for a prime q ∣ o ( g ) q \mid o(g) q ∣ o ( g ) it lowers the order. The power maps for the primes dividing ∣ G ∣ |G| ∣ G ∣ determine all others, because g m g^m g m depends only on m m o d o ( g ) m \bmod o(g) m mod o ( g ) and the maps compose.
Theorem 5.12 (Rationality). Let g ∈ G g \in G g ∈ G have order n n n and put K = { m ∈ ( Z / n Z ) × : g m ∈ g G } K = \{m \in(\mathbb{Z}/n\mathbb{Z})^\times: g^m \in g^G\} K = { m ∈ ( Z / n Z ) × : g m ∈ g G } . Then K K K is a subgroup of ( Z / n Z ) × (\mathbb{Z}/n\mathbb{Z})^\times ( Z / n Z ) × , isomorphic to N G ( ⟨ g ⟩ ) / C G ( g ) N_G(\langle g\rangle)/C_G(g) N G (⟨ g ⟩) / C G ( g ) , and the generators of ⟨ g ⟩ \langle g\rangle ⟨ g ⟩ fall into exactly φ ( n ) / ∣ K ∣ \varphi(n)/|K| φ ( n ) /∣ K ∣ classes of G G G . In particular g G g^G g G is rational iff ∣ N G ( ⟨ g ⟩ ) : C G ( g ) ∣ = φ ( n ) |N_G(\langle g\rangle) : C_G(g)| = \varphi(n) ∣ N G (⟨ g ⟩) : C G ( g ) ∣ = φ ( n ) , and real iff − 1 ∈ K -1 \in K − 1 ∈ K .
Proof. Every x ∈ N G ( ⟨ g ⟩ ) x \in N_G(\langle g\rangle) x ∈ N G (⟨ g ⟩) induces an automorphism g ↦ g x = g m ( x ) g \mapsto g^x = g^{m(x)} g ↦ g x = g m ( x ) of the cyclic group ⟨ g ⟩ \langle g\rangle ⟨ g ⟩ , and x ↦ m ( x ) x \mapsto m(x) x ↦ m ( x ) is a homomorphism N G ( ⟨ g ⟩ ) → ( Z / n Z ) × N_G(\langle g\rangle) \to(\mathbb{Z}/n\mathbb{Z})^\times N G (⟨ g ⟩) → ( Z / n Z ) × with kernel C G ( g ) C_G(g) C G ( g ) . Its image is K K K : if g m = g x g^m = g^x g m = g x then x x x normalises ⟨ g ⟩ = ⟨ g m ⟩ \langle g\rangle= \langle g^m\rangle ⟨ g ⟩ = ⟨ g m ⟩ . The group K K K acts on the φ ( n ) \varphi(n) φ ( n ) generators g m g^m g m by multiplication, freely; two generators g a , g b g^a, g^b g a , g b are G G G -conjugate iff g b = ( g a ) x g^b = (g^a)^x g b = ( g a ) x for some x x x , iff b / a ∈ K b/a \in K b / a ∈ K . So the generators split into the φ ( n ) / ∣ K ∣ \varphi(n)/|K| φ ( n ) /∣ K ∣ cosets of K K K .
If H ≤ G H \leq G H ≤ G , every class of H H H lies in a unique class of G G G ; the induced surjection from the H H H -classes covered by H H H to the G G G -classes meeting H H H is the fusion map . It is computed by applying class invariants (order, cycle type, power classes) to H H H -representatives, and resolving ties by an explicit conjugacy test. Fusion is what makes restriction of characters (next chapter on characters) computable.
Algorithms for conjugacy classes
Throughout, G = ⟨ s 1 , … , s r ⟩ G = \langle s_1, \ldots, s_r\rangle G = ⟨ s 1 , … , s r ⟩ is a permutation group of degree p p p given by codes.
Lemma 5.13. A subset X ⊆ G X \subseteq G X ⊆ G satisfies X G = X X^G = X X G = X iff x s i ∈ X x^{s_i} \in X x s i ∈ X for all x ∈ X x \in X x ∈ X and all i i i . Hence g G g^G g G is the closure of { g } \{g\} { g } under the r r r maps x ↦ x s i x \mapsto x^{s_i} x ↦ x s i .
Proof. In a finite group s i − 1 = s i o ( s i ) − 1 s_i^{-1} = s_i^{o(s_i)-1} s i − 1 = s i o ( s i ) − 1 , so every element of G G G is a positive word in the generators, and conjugation by a word is the composite of conjugations by its letters.
Theorem 5.14 (Completeness certificate). Let g 1 , … , g k ∈ G g_1, \ldots, g_k \in G g 1 , … , g k ∈ G be pairwise nonconjugate. Then { g i } \{g_i\} { g i } is a full set of class representatives iff ∑ i = 1 k ∣ G ∣ / ∣ C G ( g i ) ∣ = ∣ G ∣ \sum_{i=1}^{k} |G|/|C_G(g_i)| = |G| ∑ i = 1 k ∣ G ∣/∣ C G ( g i ) ∣ = ∣ G ∣ .
Proof. The classes g i G g_i^G g i G are disjoint subsets of G G G of sizes ∣ G ∣ / ∣ C G ( g i ) ∣ |G|/|C_G(g_i)| ∣ G ∣/∣ C G ( g i ) ∣ (Theorem 4.15 ). Their union is G G G iff their sizes add up to ∣ G ∣ |G| ∣ G ∣ .
This theorem converts the class problem into three subproblems: generate candidates, prove two candidates nonconjugate (an invariant that differs suffices), and compute centraliser orders. The three standard methods differ in how they do this.
Orbit enumeration. Repeatedly take an element not yet classified and close it under Lemma 5.13. Cost: r ∣ G ∣ r|G| r ∣ G ∣ conjugations of codes, O ( r p ∣ G ∣ ) O(rp|G|) O ( r p ∣ G ∣ ) digit look-ups, memory ∣ G ∣ |G| ∣ G ∣ codes. Exact and simple; feasible up to ∣ G ∣ ≈ 10 7 |G| \approx10^7 ∣ G ∣ ≈ 1 0 7 . This is how the M 12 M_{12} M 12 table below was computed.
Representatives and centralisers. Keep only representatives. Distinguish candidates by invariants (order, cycle type, fixed-point count, power classes), decide the remaining ties by a backtrack conjugacy test, and compute ∣ C G ( g ) ∣ |C_G(g)| ∣ C G ( g ) ∣ by Schreier–Sims (§ 11.7 ) applied to the centraliser, then stop by Theorem 5.14. Memory is polynomial in p p p ; this is the method for groups such as those in Chapter 13.
Random sampling. A uniform random element lies in g G g^G g G with probability 1 / ∣ C G ( g ) ∣ 1/|C_G(g)| 1/∣ C G ( g ) ∣ : classes with large centralisers are rare. Pure sampling is therefore bad at finding small classes, and it is completed by deterministic sources: the identity and the centre, powers g q g^q g q of found elements (which land in classes of larger centralisers), and elements of found centralisers.
Example 5.15. For M 12 M_{12} M 12 the identity is drawn with probability 1 / 95040 1/95040 1/95040 . With seed 1 uniform sampling met all 15 classes only after 30460 draws, essentially waiting for 1A; adjoining the identity and closing under the power maps of Table 5.2 removes this bottleneck, since 2A, 2B, 3A, 3B are the classes with the largest centralisers and all are powers of classes of orders 4, 6 and 10.
Table 5.2. Conjugacy classes of M 12 M_{12} M 12 in its action on A 12 A_{12} A 12 (verified by scripts/cc_verify.py and GAP).
A complete example: M 12 M_{12} M 12
Let M 12 ≤ Sym 12 M_{12} \leq\operatorname{Sym}_{12} M 12 ≤ Sym 12 be generated by the reversal x ↦ 11 − x x \mapsto11-x x ↦ 11 − x and the Mongean shuffle x ↦ min ( 2 x + 1 , 22 − 2 x ) x \mapsto\min(2x+1,22-2x) x ↦ min ( 2 x + 1 , 22 − 2 x ) on A 12 = { 0 , … , 11 } A_{12}=\{0,\ldots,11\} A 12 = { 0 , … , 11 } . Orbit enumeration gives order 95040 and 15 classes; GAP confirms that the group is 5-transitive and that the class sizes coincide with those of MathieuGroup(12). Table 5.2 lists the classes, named by order and, within an order, by decreasing centraliser and then by cycle type in this action; m P mP m P is the m m m -th power map.
The class equation reads
95040 = 1 + 396 + 495 + 1760 + 2640 + 2 ⋅ 2970 + 9504 + 7920 + 15840 + 2 ⋅ 11880 + 9504 + 2 ⋅ 8640. 95040=1+396+495+1760+2640+2\cdot2970+9504+7920+15840+2\cdot11880+9504+2\cdot8640. 95040 = 1 + 396 + 495 + 1760 + 2640 + 2 ⋅ 2970 + 9504 + 7920 + 15840 + 2 ⋅ 11880 + 9504 + 2 ⋅ 8640.
Several structural facts can be read off the table.
The two classes of order 11. For g ∈ 11 A g\in11\mathrm{A} g ∈ 11 A , ∣ C G ( g ) ∣ = 11 \lvert C_G(g)\rvert=11 ∣ C G ( g )∣ = 11 , so C G ( g ) = ⟨ g ⟩ C_G(g)=\langle g\rangle C G ( g ) = ⟨ g ⟩ . Squaring exchanges 11A and 11B, so 2 ∉ K 2\notin K 2 ∈ / K in Theorem 5.12 ; as 2 generates ( Z / 11 Z ) × (\mathbb{Z}/11\mathbb{Z})^\times ( Z /11 Z ) × and the ten generators of ⟨ g ⟩ \langle g\rangle ⟨ g ⟩ fall into the two classes, ∣ K ∣ = 5 \lvert K\rvert=5 ∣ K ∣ = 5 and K K K is the group of quadratic residues { 1 , 3 , 4 , 5 , 9 } \{1,3,4,5,9\} { 1 , 3 , 4 , 5 , 9 } . Thus ∣ N G ( ⟨ g ⟩ ) ∣ = 55 \lvert N_G(\langle g\rangle)\rvert=55 ∣ N G (⟨ g ⟩)∣ = 55 , the classes are non-real (− 1 -1 − 1 is a non-residue modulo 11) and inverse to each other, and g m ∈ 11 A g^m\in11\mathrm{A} g m ∈ 11 A iff m m m is a nonzero square mod 11. In practice: fix one 11-cycle h h h ; any element g g g of order 11 lies in h G h^G h G iff it is conjugate to h m h^m h m with m m m a residue, which is decided by the conjugacy test applied once.
All other classes are rational, by direct check of the definition; hence 13 classes are rational and real. By Theorem 5.12 , for instance, ∣ N G ( ⟨ g ⟩ ) ∣ = 8 ⋅ φ ( 8 ) = 32 \lvert N_G(\langle g\rangle)\rvert=8\cdot\varphi(8)=32 ∣ N G (⟨ g ⟩)∣ = 8 ⋅ φ ( 8 ) = 32 for g ∈ 8 A g\in8\mathrm{A} g ∈ 8 A .
ι 12 = 8842413667692 , \iota_{12} = 8842413667692, ι 12 = 8842413667692 ,
in agreement with the closed form (10) .
Cycle index and colourings. Table 5.2 is the cycle index of M 12 M_{12} M 12 on A 12 A_{12} A 12 : the class g G g^G g G contributes ∣ g G ∣ x λ / ∣ G ∣ |g^G|x^\lambda/|G| ∣ g G ∣ x λ /∣ G ∣ , where λ \lambda λ is its cycle type. The numbers c ( g ) c(g) c ( g ) of cycles of the fifteen classes are 12 , 6 , 8 , 6 , 4 , 6 , 4 , 4 , 2 , 4 , 4 , 2 , 2 , 2 , 2 12,6,8,6,4,6,4,4,2,4,4,2,2,2,2 12 , 6 , 8 , 6 , 4 , 6 , 4 , 4 , 2 , 4 , 4 , 2 , 2 , 2 , 2 , so Theorem 4.21 gives
1 95040 ∑ g G ∣ g G ∣ m c ( g ) = 14 ( m = 2 ) , 120 ( m = 3 ) . \frac{1}{95040}\sum_{g^G}|g^G|m^{c(g)} = 14\ (m=2),\qquad120\ (m=3). 95040 1 g G ∑ ∣ g G ∣ m c ( g ) = 14 ( m = 2 ) , 120 ( m = 3 ) .
The first number has a meaning. Since M 12 M_{12} M 12 is 5-transitive it has one orbit on k k k -subsets of A 12 A_{12} A 12 for each k ≤ 5 k\leq5 k ≤ 5 and each k ≥ 7 k\geq7 k ≥ 7 , which accounts for 12 of the 14 orbits on 2-colourings; the remaining two are the orbits on 6-subsets, the 132 hexads of the Steiner system S ( 5 , 6 , 12 ) S(5,6,12) S ( 5 , 6 , 12 ) and the other 792 six-subsets. Both numbers agree with direct orbit enumeration (scripts/polya_verify.py).
For conjugacy classes and power maps computed from the digital laws of sporadic groups, see Section 13.16.2.
Exercises
Finish the hand trace for Sym 4 \operatorname{Sym}_4 Sym 4 : list all Schreier generators at each level and check which ones sift.
Run the deterministic algorithm by hand on Alt 5 \operatorname{Alt}_5 Alt 5 with base ( 0 , 1 , 2 ) (0,1,2) ( 0 , 1 , 2 ) , and verify ∣ Alt 5 ∣ = 5 ⋅ 4 ⋅ 3 |\operatorname{Alt}_5|=5\cdot4\cdot3 ∣ Alt 5 ∣ = 5 ⋅ 4 ⋅ 3 .
Show that the base length of any G ≤ Sym n G\leq\operatorname{Sym}_n G ≤ Sym n satisfies 2 k ≤ ∣ G ∣ ≤ n k 2^k\leq|G|\leq n^k 2 k ≤ ∣ G ∣ ≤ n k .
Prove that the depth of a Schreier tree built by breadth-first search on S ∪ S − 1 S\cup S^{-1} S ∪ S − 1 is at most the diameter of the Cayley graph of the orbit, and give an example of depth ∣ Δ ∣ − 1 |\Delta|-1 ∣Δ∣ − 1 .
Using Corollary 5.3, write the unranking map [ 0 , 95040 ) → M 12 [0,95040)\to M_{12} [ 0 , 95040 ) → M 12 explicitly and compute the elements with labels 0, 1 and 95039.
Design a test that decides whether G = ⟨ S ⟩ ≤ Sym n G=\langle S\rangle\leq\operatorname{Sym}_n G = ⟨ S ⟩ ≤ Sym n is k k k -transitive from a BSGS with base ( 0 , 1 , … , k − 1 ) (0,1,\ldots,k-1) ( 0 , 1 , … , k − 1 ) .
Extend the algorithm to compute the order of the centraliser of an element by backtrack, and recover the centraliser orders of Table 5.2 .
Compute the classes, centralisers and power maps of Sym 5 \operatorname{Sym}_5 Sym 5 and Alt 5 \operatorname{Alt}_5 Alt 5 by hand and check Theorem 4.18 for both.
Show that the classes of the dihedral group of order 2 n 2n 2 n number ( n + 3 ) / 2 (n+3)/2 ( n + 3 ) /2 for odd n n n and n / 2 + 3 n/2+3 n /2 + 3 for even n n n , and list centraliser orders.
Exercises
Prove that a group in which every class has size at most 2 has derived subgroup of order at most 2.
Using Theorem 11.9 , list the split classes of Alt 7 \operatorname{Alt}_7 Alt 7 and Alt 8 \operatorname{Alt}_8 Alt 8 , and confirm k ( Alt 7 ) = 9 k(\operatorname{Alt}_7)=9 k ( Alt 7 ) = 9 , k ( Alt 8 ) = 14 k(\operatorname{Alt}_8)=14 k ( Alt 8 ) = 14 .
Prove that every class of Sym p \operatorname{Sym}_p Sym p is rational, and determine which classes of Alt p \operatorname{Alt}_p Alt p are real.
For G = PSL ( 2 , 7 ) G=\operatorname{PSL}(2,7) G = PSL ( 2 , 7 ) acting on the 8 points of the projective line over F 7 \mathbb{F}_7 F 7 , determine the six classes, and show that the two classes of order 7 are non-real, exactly as for 11 A , 11 B 11A,11B 11 A , 11 B in M 12 M_{12} M 12 .
Modify the orbit-enumeration algorithm so that it also returns, for each class, a conjugating element from a fixed representative to every class member, stored as a Schreier tree (§ 11.2 ).
Estimate the expected number of uniform draws needed to see every class of M 12 M_{12} M 12 and explain why it is dominated by 1 / ∣ G ∣ 1/|G| 1/∣ G ∣ .
For G = PSL ( 2 , 7 ) G=\operatorname{PSL}(2,7) G = PSL ( 2 , 7 ) in its action on the 8 points of the projective line over F 7 \mathbb{F}_7 F 7 , find the cycle index from the six classes and show that the number of orbits on 2-colourings of the points is 11. Deduce the number of orbits on 4-subsets.
Characters and character tables in digital form
The preceding chapters turned multiplication, conjugacy, stabilisers and subgroup chains into digital data. This chapter develops the full character theory needed to pass from those data to representation-theoretic invariants. The point of view is classical, but every construction is written so that it can be evaluated from a coded multiplication law on the alphabet A N \mathcal{A}_N A N .
Throughout, G = ( A N , ∗ ) G = (\mathcal{A}_N,\ast) G = ( A N , ∗ ) has code R R R , identity 0 0 0 , inverse a ˉ \bar{a} a ˉ , and
a ∗ b = d N ( a N + b , R ) . a \ast b = d_N(aN+b,R). a ∗ b = d N ( a N + b , R ) .
All vector spaces are finite-dimensional over C \mathbb{C} C .
Representations, characters and the group algebra
Definition 6.1. A representation of degree r r r is a map
ρ : A N ⟶ G L r ( C ) , ρ ( 0 ) = I r , ρ ( a ∗ b ) = ρ ( a ) ρ ( b ) . \rho:\mathcal{A}_N \longrightarrow\mathrm{GL}_r(\mathbb{C}), \qquad\rho(0)=I_r, \qquad\rho(a\ast b)=\rho(a)\rho(b). ρ : A N ⟶ GL r ( C ) , ρ ( 0 ) = I r , ρ ( a ∗ b ) = ρ ( a ) ρ ( b ) .
Its character is the class function
χ ρ ( a ) = tr ρ ( a ) . \chi_\rho(a)=\operatorname{tr}\rho(a). χ ρ ( a ) = tr ρ ( a ) .
Two representations are equivalent if σ ( a ) = T − 1 ρ ( a ) T \sigma(a)=T^{-1}\rho(a)T σ ( a ) = T − 1 ρ ( a ) T for a fixed invertible matrix T T T and all a a a . A nonzero representation is irreducible if it has no proper nonzero invariant subspace.
The law is encoded by finitely many matrix identities,
ρ ( d N ( a N + b , R ) ) = ρ ( a ) ρ ( b ) ( a , b ∈ A N ) , (52) \rho\bigl(d_N(aN+b,R)\bigr)=\rho(a)\rho(b) \qquad(a,b\in\mathcal{A}_N), \tag*{(52)} ρ ( d N ( a N + b , R ) ) = ρ ( a ) ρ ( b ) ( a , b ∈ A N ) , (52)
so a representation may be verified directly from the operation code.
Proposition 6.2 (Elementary character identities ). For characters χ , ψ \chi,\psi χ , ψ of G G G ,
χ ( 0 ) = deg χ , χ ( a ˉ ) = χ ( a ) ‾ , χ ( a ∗ b ) = χ ( b ∗ a ) , \chi(0)=\deg\chi, \qquad\chi(\bar{a})=\overline{\chi(a)}, \qquad\chi(a\ast b)=\chi(b\ast a), χ ( 0 ) = deg χ , χ ( a ˉ ) = χ ( a ) , χ ( a ∗ b ) = χ ( b ∗ a ) ,
χ ρ ⊕ σ = χ + ψ , χ ρ ⊗ σ = χ ψ . \chi_{\rho\oplus\sigma}=\chi+\psi, \qquad\chi_{\rho\otimes\sigma}=\chi\psi. χ ρ ⊕ σ = χ + ψ , χ ρ ⊗ σ = χ ψ .
Equivalent representations have the same character, and every character is constant on conjugacy classes.
Proof. The matrices ρ ( a ) \rho(a) ρ ( a ) have finite order, so their eigenvalues are roots of unity. Hence the eigenvalues of ρ ( a ˉ ) = ρ ( a ) − 1 \rho(\bar{a})=\rho(a)^{-1} ρ ( a ˉ ) = ρ ( a ) − 1 are the complex conjugates of those of ρ ( a ) \rho(a) ρ ( a ) . Trace is invariant under similarity, satisfies tr ( A B ) = tr ( B A ) \operatorname{tr}(AB)=\operatorname{tr}(BA) tr ( A B ) = tr ( B A ) , adds on block sums and multiplies on tensor products. Finally,
ρ ( g ∗ a ∗ g ˉ ) = ρ ( g ) ρ ( a ) ρ ( g ) − 1 \rho(g\ast a\ast\bar{g})=\rho(g)\rho(a)\rho(g)^{-1} ρ ( g ∗ a ∗ g ˉ ) = ρ ( g ) ρ ( a ) ρ ( g ) − 1
is similar to ρ ( a ) \rho(a) ρ ( a ) .
Let C G \mathbb{C}G C G be the group algebra of G G G , with basis { e a : a ∈ A N } \{e_a : a \in A_N\} { e a : a ∈ A N } and multiplication
e a e b = e a ∗ b . e_a e_b=e_{a*b}. e a e b = e a ∗ b .
A representation of G G G is the same thing as a left C G \mathbb{C}G C G -module: extend ρ \rho ρ linearly by
ρ ( ∑ a c a e a ) = ∑ a c a ρ ( a ) . \rho\left(\sum_a c_a e_a\right)=\sum_a c_a\rho(a). ρ ( a ∑ c a e a ) = a ∑ c a ρ ( a ) .
This reformulation makes the structure of the theory transparent. It also identifies characters as special matrix coefficients: if v 1 , … , v r v_1,\ldots,v_r v 1 , … , v r is a basis of the representation space, the coordinate functions ρ i j ( a ) \rho_{ij}(a) ρ ij ( a ) defined by
ρ ( a ) v j = ∑ i = 1 r ρ i j ( a ) v i \rho(a)v_j=\sum_{i=1}^{r}\rho_{ij}(a)v_i ρ ( a ) v j = i = 1 ∑ r ρ ij ( a ) v i
are ordinary functions on G G G , and the character is their diagonal sum. Many structural identities are best understood first for matrix coefficients and then specialised to traces.
A second basic construction is averaging over the group. If f : G → C f:G\to\mathbb{C} f : G → C is any function, define its average by
Av ( f ) = 1 N ∑ a = 0 N − 1 f ( a ) . \operatorname{Av}(f)=\frac{1}{N}\sum_{a=0}^{N-1}f(a). Av ( f ) = N 1 a = 0 ∑ N − 1 f ( a ) .
If V V V is a representation and T ∈ End C ( V ) T\in\operatorname{End}_{\mathbb{C}}(V) T ∈ End C ( V ) , the averaged operator
T # = 1 N ∑ a = 0 N − 1 ρ ( a ) T ρ ( a ‾ ) T^{\#}=\frac{1}{N}\sum_{a=0}^{N-1}\rho(a)T\rho(\overline{a}) T # = N 1 a = 0 ∑ N − 1 ρ ( a ) T ρ ( a )
is the canonical way to turn an arbitrary operator into a G G G -equivariant one.
Proposition 6.3 (Averaging and invariants). Let V V V be a representation of G G G .
The operator T # T^{\#} T # commutes with every ρ ( g ) \rho(g) ρ ( g ) .
If W ≤ V W\leq V W ≤ V is any subspace, then
V G = { v ∈ V : ρ ( g ) v = v for all g ∈ G } V^G=\{v\in V:\rho(g)v=v\text{ for all }g\in G\} V G = { v ∈ V : ρ ( g ) v = v for all g ∈ G } is the image of the averaging projector
P G = 1 N ∑ a = 0 N − 1 ρ ( a ) . P_G=\frac{1}{N}\sum_{a=0}^{N-1}\rho(a). P G = N 1 a = 0 ∑ N − 1 ρ ( a ) . The multiplicity of the trivial representation in V V V equals
dim V G = 1 N ∑ a = 0 N − 1 χ V ( a ) . \dim V^G=\frac{1}{N}\sum_{a=0}^{N-1}\chi_V(a). dim V G = N 1 a = 0 ∑ N − 1 χ V ( a ) .
Proof. For (1),
ρ ( g ) T # ρ ( g ‾ ) = 1 N ∑ a ρ ( g ∗ a ) T ρ ( g ∗ a ‾ ) , \rho(g)T^{\#}\rho(\overline{g})=\frac{1}{N}\sum_a\rho(g*a)T\rho(\overline{g*a}), ρ ( g ) T # ρ ( g ) = N 1 a ∑ ρ ( g ∗ a ) T ρ ( g ∗ a ) ,
and the map a ↦ g ∗ a a\mapsto g*a a ↦ g ∗ a permutes G G G , so the sum is unchanged. For (2), P G P_G P G clearly lands in V G V^G V G and fixes every vector already fixed by G G G . For (3), the trace of a projection equals the dimension of its image, while
tr ( P G ) = 1 N ∑ a tr ( ρ ( a ) ) = 1 N ∑ a χ V ( a ) . \operatorname{tr}(P_G)=\frac{1}{N}\sum_a\operatorname{tr}(\rho(a))=\frac{1}{N}\sum_a\chi_V(a). tr ( P G ) = N 1 a ∑ tr ( ρ ( a )) = N 1 a ∑ χ V ( a ) .
Maschke's theorem and complete reducibility
Theorem 6.4 (Maschke). Every finite-dimensional CG-module is a direct sum of irreducible submodules. Equivalently, every representation of a finite group is completely reducible.
Proof. Let V V V be a representation and let W ≤ V W \le V W ≤ V be a submodule. Choose any linear projection π : V → W \pi: V \to W π : V → W with π ∣ W = I W \left.\pi\right|_W = I_W π ∣ W = I W . Average it over the group:
Π ( v ) = 1 N ∑ a = 0 N − 1 ρ ( a ) π ( ρ ( a ‾ ) v ) . \Pi(v) = \frac{1}{N} \sum_{a=0}^{N-1} \rho(a)\pi\left(\rho(\overline{a})v\right). Π ( v ) = N 1 a = 0 ∑ N − 1 ρ ( a ) π ( ρ ( a ) v ) . For g ∈ G g \in G g ∈ G one has
Π ( ρ ( g ) v ) = 1 N ∑ a ρ ( a ) π ( ρ ( a ‾ ∗ g ) v ) = ρ ( g ) 1 N ∑ b ρ ( b ) π ( ρ ( b ‾ ) v ) = ρ ( g ) Π ( v ) , \Pi(\rho(g)v) = \frac{1}{N} \sum_a \rho(a)\pi\left(\rho(\overline{a} * g)v\right) = \rho(g)\frac{1}{N} \sum_b \rho(b)\pi\left(\rho(\overline{b})v\right) = \rho(g)\Pi(v), Π ( ρ ( g ) v ) = N 1 a ∑ ρ ( a ) π ( ρ ( a ∗ g ) v ) = ρ ( g ) N 1 b ∑ ρ ( b ) π ( ρ ( b ) v ) = ρ ( g ) Π ( v ) , where the change of variable b = g ‾ ∗ a b = \overline{g} * a b = g ∗ a uses the fact that left multiplication permutes A N \mathcal{A}_N A N . Thus Π \Pi Π is a G G G -equivariant projection onto W W W . Its kernel is a G G G -stable complement of W W W . Repeating the argument on the complement proves the theorem by induction on dim V \dim V dim V .
Averaging a Hermitian form in the same way gives the unitary version.
Corollary 6.5 (Unitary form). Every representation of a finite group is equivalent to a unitary representation.
Proof. Given any Hermitian form ⟨ ⋅ , ⋅ ⟩ \langle\cdot,\cdot\rangle ⟨ ⋅ , ⋅ ⟩ , define
⟨ v , w ⟩ G = 1 N ∑ a = 0 N − 1 ⟨ ρ ( a ) v , ρ ( a ) w ⟩ . \langle v,w\rangle_G = \frac{1}{N}\sum_{a=0}^{N-1}\langle\rho(a)v,\rho(a)w\rangle. ⟨ v , w ⟩ G = N 1 a = 0 ∑ N − 1 ⟨ ρ ( a ) v , ρ ( a ) w ⟩ . Then ⟨ ⋅ , ⋅ ⟩ G \langle\cdot,\cdot\rangle_G ⟨ ⋅ , ⋅ ⟩ G is positive definite and G G G -invariant.
Schur's lemma and Wedderburn theory
Lemma 6.6 (Schur). Let V V V and W W W be irreducible representations of G G G .
If T : V → W T : V \to W T : V → W satisfies T ρ V ( a ) = ρ W ( a ) T T\rho_V(a) = \rho_W(a)T T ρ V ( a ) = ρ W ( a ) T for all a a a , then either T = 0 T = 0 T = 0 or T T T is an isomorphism.
If V = W V = W V = W , then every intertwiner T : V → V T : V \to V T : V → V is a scalar multiple of the identity.
Proof. The kernel and image of an intertwiner are submodules, so irreducibility gives the first statement. For the second, choose an eigenvalue λ \lambda λ of T T T . Then the nonzero eigenspace of T − λ I T - \lambda I T − λ I is a submodule, hence all of V V V .
The group algebra is therefore semisimple. In particular every finite-dimensional representation has a decomposition
V ≅ m 1 V 1 ⊕ ⋯ ⊕ m k V k , V \cong m_1V_1 \oplus\cdots\oplus m_kV_k, V ≅ m 1 V 1 ⊕ ⋯ ⊕ m k V k ,
where V 1 , … , V k V_1,\ldots,V_k V 1 , … , V k are pairwise non-isomorphic irreducibles and the multiplicities m i m_i m i are uniquely determined. The uniqueness follows because
Hom G ( V i , V ) ≅ C m i \operatorname{Hom}_G(V_i,V) \cong\mathbb{C}^{m_i} Hom G ( V i , V ) ≅ C m i
by Schur's lemma: every copy of V i V_i V i contributes one independent projection onto that isotypic summand, while the other irreducible summands contribute nothing.
The most important module is the regular representation. It not only contains every irreducible representation but records the correct multiplicity of each one.
Theorem 6.7 (Decomposition of the regular module). If V 1 , … , V k V_1,\ldots,V_k V 1 , … , V k are the irreducible representations of G G G with degrees n 1 , … , n k n_1,\ldots,n_k n 1 , … , n k , then
C G ≅ n 1 V 1 ⊕ ⋯ ⊕ n k V k \mathbb{C}G \cong n_1V_1 \oplus\cdots\oplus n_kV_k C G ≅ n 1 V 1 ⊕ ⋯ ⊕ n k V k as left C G \mathbb{C}G C G -modules. In particular every irreducible representation occurs in the regular module with multiplicity equal to its degree.
Proof. Write
C G ≅ m 1 V 1 ⊕ ⋯ ⊕ m k V k . \mathbb{C}G \cong m_1V_1 \oplus\cdots\oplus m_kV_k. C G ≅ m 1 V 1 ⊕ ⋯ ⊕ m k V k . Applying Proposition 6.3 to the tensor product V i ∗ ⊗ C G V_i^* \otimes\mathbb{C}G V i ∗ ⊗ C G , or equivalently taking the inner product of characters with the regular character, shows that the multiplicity of V i V_i V i equals
dim Hom G ( V i , C G ) = n i . \dim\operatorname{Hom}_G(V_i,\mathbb{C}G)=n_i. dim Hom G ( V i , C G ) = n i . Indeed a G G G -map V i → C G V_i \to\mathbb{C}G V i → C G is determined by the image of a basis and these maps identify with the full dual space V i ∗ V_i^* V i ∗ under evaluation at the identity basis vector of the regular module.
Writing irreducible representations as V 1 , … , V k V_1,\ldots,V_k V 1 , … , V k with degrees n 1 , … , n k n_1,\ldots,n_k n 1 , … , n k , one obtains the precise structure theorem.
Theorem 6.8 (Wedderburn decomposition). There is an algebra isomorphism
C G ≅ M n 1 ( C ) ⊕ ⋯ ⊕ M n k ( C ) . \mathbb{C}G \cong M_{n_1}(\mathbb{C}) \oplus\cdots\oplus M_{n_k}(\mathbb{C}). C G ≅ M n 1 ( C ) ⊕ ⋯ ⊕ M n k ( C ) . In particular,
∣ G ∣ = dim C C G = ∑ i = 1 k n i 2 . |G|=\dim_{\mathbb{C}}\mathbb{C}G=\sum_{i=1}^{k}n_i^2. ∣ G ∣ = dim C C G = i = 1 ∑ k n i 2 . Proof. Let V = ⨁ i n i V i V=\bigoplus_i n_iV_i V = ⨁ i n i V i be the regular module. The action map
C G ⟶ End C ( V ) \mathbb{C}G \longrightarrow\operatorname{End}_{\mathbb{C}}(V) C G ⟶ End C ( V ) is injective because C G \mathbb{C}G C G acts faithfully on itself by left multiplication. Its image preserves each isotypic component and acts on the i i i th component as the full matrix algebra M n i ( C ) M_{n_i}(\mathbb{C}) M n i ( C ) by Schur’s lemma and the density argument for semisimple algebras. Comparing dimensions yields the decomposition.
Orthogonality relations
Let CF ( G ) \operatorname{CF}(G) CF ( G ) be the space of class functions on G G G . Define the inner product
⟨ α , β ⟩ G = 1 N ∑ a = 0 N − 1 α ( a ) β ( a ) ‾ . (53) \langle\alpha,\beta\rangle_G=\frac{1}{N}\sum_{a=0}^{N-1}\alpha(a)\overline{\beta(a)}. \tag*{(53)} ⟨ α , β ⟩ G = N 1 a = 0 ∑ N − 1 α ( a ) β ( a ) . (53)
Theorem 6.9 (Row orthogonality). If χ i \chi_i χ i and χ j \chi_j χ j are irreducible characters, then
⟨ χ i , χ j ⟩ G = δ i j . \langle\chi_i,\chi_j\rangle_G=\delta_{ij}. ⟨ χ i , χ j ⟩ G = δ ij . More generally, if
χ = ∑ i m i χ i , \chi=\sum_i m_i\chi_i, χ = i ∑ m i χ i , then
m i = ⟨ χ , χ i ⟩ G , ⟨ χ , χ ⟩ G = ∑ i m i 2 . m_i=\langle\chi,\chi_i\rangle_G,\qquad\langle\chi,\chi\rangle_G=\sum_i m_i^2. m i = ⟨ χ , χ i ⟩ G , ⟨ χ , χ ⟩ G = i ∑ m i 2 . Thus χ \chi χ is irreducible if and only if ⟨ χ , χ ⟩ G = 1 \langle\chi,\chi\rangle_G=1 ⟨ χ , χ ⟩ G = 1 .
Proof. Let V i , V j V_i,V_j V i , V j afford χ i , χ j \chi_i,\chi_j χ i , χ j . For a linear map A : V i → V j A:V_i\to V_j A : V i → V j set
P ( A ) = 1 N ∑ a ρ j ( a ) A ρ i ( a ˉ ) . \mathcal{P}(A)=\frac{1}{N}\sum_a \rho_j(a)A\rho_i(\bar{a}). P ( A ) = N 1 a ∑ ρ j ( a ) A ρ i ( a ˉ ) .
Then P ( A ) \mathcal{P}(A) P ( A ) is an intertwiner. By Schur’s lemma, P \mathcal{P} P is zero when i ≠ j i\ne j i = j and a rank-one projection onto the scalar endomorphisms when i = j i=j i = j . Computing its trace on matrix units gives
1 N ∑ a χ j ( a ) χ i ( a ) ‾ = δ i j . \frac{1}{N}\sum_a \chi_j(a)\overline{\chi_i(a)}=\delta_{ij}. N 1 a ∑ χ j ( a ) χ i ( a ) = δ ij .
The multiplicity formulas follow by expanding χ \chi χ in the orthonormal basis of irreducible characters.
Corollary 6.10 (Column orthogonality). Let g , h ∈ G g,h\in G g , h ∈ G , and let χ 1 , … , χ k \chi_1,\ldots,\chi_k χ 1 , … , χ k be the irreducible characters. Then
∑ i = 1 k χ i ( g ) χ i ( h ) ‾ = { ∣ C G ( g ) ∣ , g , h are conjugate , 0 , g , h are not conjugate . \sum_{i=1}^{k}\chi_i(g)\overline{\chi_i(h)}=
\begin{cases}
\lvert C_G(g)\rvert, & g,h\text{ are conjugate},\\
0, & g,h\text{ are not conjugate}.
\end{cases} i = 1 ∑ k χ i ( g ) χ i ( h ) = { ∣ C G ( g )∣ , 0 , g , h are conjugate , g , h are not conjugate . In particular, the irreducible characters form an orthonormal basis of CF ( G ) \operatorname{CF}(G) CF ( G ) , so the number of irreducible characters equals the number of conjugacy classes.
Proof. Let f g f_g f g be the indicator of the conjugacy class of g g g , divided by its class size. Then the class indicators form a basis of CF ( G ) \operatorname{CF}(G) CF ( G ) . Expand f g f_g f g in the irreducible-character basis and compare coefficients.
The regular character and central idempotents
Let ρ r e g \rho_{\mathrm{reg}} ρ reg be the regular representation on basis vectors e a e_a e a , defined by
ρ r e g ( g ) e a = e g ∗ a . \rho_{\mathrm{reg}}(g)e_a=e_{g*a}. ρ reg ( g ) e a = e g ∗ a .
Its character is
χ r e g ( g ) = { N , g = 0 , 0 , g ≠ 0. \chi_{\mathrm{reg}}(g)=
\begin{cases}
N, & g=0,\\
0, & g\ne0.
\end{cases} χ reg ( g ) = { N , 0 , g = 0 , g = 0.
Combining this with orthogonality yields the familiar degree formula.
Theorem 6.11 (Degree-sum formula). If χ 1 , … , χ k \chi_1,\ldots,\chi_k χ 1 , … , χ k are the irreducible characters of G G G , then
χ r e g = ∑ i = 1 k χ i ( 0 ) χ i , N = ∑ i = 1 k χ i ( 0 ) 2 . \chi_{\mathrm{reg}}=\sum_{i=1}^{k}\chi_i(0)\chi_i,\qquad
N=\sum_{i=1}^{k}\chi_i(0)^2. χ reg = i = 1 ∑ k χ i ( 0 ) χ i , N = i = 1 ∑ k χ i ( 0 ) 2 .
Proof. Taking inner products with χ j \chi_j χ j gives
⟨ χ r e g , χ j ⟩ G = χ j ( 0 ) , \langle\chi_{\mathrm{reg}},\chi_j\rangle_G=\chi_j(0), ⟨ χ reg , χ j ⟩ G = χ j ( 0 ) ,
so the expansion follows from Theorem 6.9 . Evaluating at the identity yields the sum of squares.
The primitive central idempotents of C G \mathbb{C}G C G are
e i = χ i ( 0 ) N ∑ a = 0 N − 1 χ i ( a ˉ ) e a . (54) e_i=\frac{\chi_i(0)}{N}\sum_{a=0}^{N-1}\chi_i(\bar{a})e_a. \tag*{(54)} e i = N χ i ( 0 ) a = 0 ∑ N − 1 χ i ( a ˉ ) e a . (54)
They satisfy e i e j = δ i j e i e_i e_j=\delta_{ij}e_i e i e j = δ ij e i , sum to 1 1 1 , and project a module onto its χ i \chi_i χ i -isotypic component. Thus the whole decomposition theory may be read directly from the character table.
Algebraic integrality and arithmetic consequences
Every character value is a sum of roots of unity, hence an algebraic integer. This simple observation has strong consequences.
Theorem 6.12 (Algebraic integrality). For every irreducible character χ \chi χ and every g ∈ G g \in G g ∈ G , the value χ ( g ) \chi(g) χ ( g ) is an algebraic integer.
Proof. In a unitary representation, ρ ( g ) \rho(g) ρ ( g ) has finite order, so its eigenvalues are roots of unity. Their sum is therefore an algebraic integer.
Theorem 6.13 (Central character formula). Let C C C be a conjugacy class with representative g g g , and put
K C = ∑ x ∈ C x ∈ Z ( C G ) . K_C = \sum_{x \in C} x \in Z(\mathbb{C}G). K C = x ∈ C ∑ x ∈ Z ( C G ) . On an irreducible module affording χ \chi χ , the element K C K_C K C acts as the scalar
ω χ ( C ) = ∣ C ∣ χ ( g ) χ ( 0 ) . \omega_\chi(C) = \frac{|C|\chi(g)}{\chi(0)}. ω χ ( C ) = χ ( 0 ) ∣ C ∣ χ ( g ) . In particular, each ω χ ( C ) \omega_\chi(C) ω χ ( C ) is an algebraic integer.
Proof. Since K C K_C K C is central, Schur’s lemma shows that it acts as a scalar. Taking traces on the irreducible module gives
∣ C ∣ χ ( g ) = ω χ ( C ) χ ( 0 ) . |C|\chi(g) = \omega_\chi(C)\chi(0). ∣ C ∣ χ ( g ) = ω χ ( C ) χ ( 0 ) . Because K C K_C K C acts through a matrix with algebraic-integer entries in a suitable integral basis of the group algebra, the scalar is an algebraic integer.
Corollary 6.14 (Degree divisibility). For every irreducible character χ \chi χ , the degree χ ( 0 ) \chi(0) χ ( 0 ) divides ∣ G ∣ |G| ∣ G ∣ .
Proof. Apply Theorem 6.13 to each class and sum the algebraic integers
∣ G ∣ χ ( 0 ) = ∑ C ∣ C ∣ χ ( g C ) χ ( g C ) ‾ χ ( 0 ) . \frac{|G|}{\chi(0)} = \sum_C \frac{|C|\chi(g_C)\overline{\chi(g_C)}}{\chi(0)}. χ ( 0 ) ∣ G ∣ = C ∑ χ ( 0 ) ∣ C ∣ χ ( g C ) χ ( g C ) . This number is both an algebraic integer and a rational number, hence an integer.
A second standard application is Burnside’s vanishing criterion: if χ \chi χ is irreducible and g ∈ G g \in G g ∈ G , then either χ ( g ) = 0 \chi(g) = 0 χ ( g ) = 0 or χ ( g ) \chi(g) χ ( g ) is a sum of roots of unity of absolute value strictly less than χ ( 0 ) \chi(0) χ ( 0 ) . This is one route to Burnside’s p a q b p^a q^b p a q b theorem and to many solvability criteria.
Induction, restriction and permutation characters
Let H ≤ G H \leq G H ≤ G . If χ \chi χ is a character of G G G , its restriction to H H H is denoted χ ↓ H \chi\downarrow_H χ ↓ H . If ψ \psi ψ is a character of H H H , its induced character on G G G is denoted ψ G \psi^G ψ G .
Theorem 6.15 (Induced-character formula). For g ∈ G g \in G g ∈ G ,
ψ G ( g ) = 1 ∣ H ∣ ∑ x ∈ G : x g x − 1 ∈ H ψ ( x g x − 1 ) . \psi^G(g) = \frac{1}{|H|}\sum_{\substack{x \in G : xgx^{-1} \in H}} \psi(xgx^{-1}). ψ G ( g ) = ∣ H ∣ 1 x ∈ G : xg x − 1 ∈ H ∑ ψ ( xg x − 1 ) . Proof. Choose left coset representatives for H H H in G G G and write the induced module in the standard tensor-product model C G ⊗ C H V \mathbb{C}G \otimes_{\mathbb{C}H} V C G ⊗ C H V . Computing the trace of g g g on the natural basis gives the formula.
Theorem 6.16 (Frobenius reciprocity). For every character χ \chi χ of G G G and every character ψ \psi ψ of H H H ,
⟨ χ ↓ H , ψ ⟩ H = ⟨ χ , ψ G ⟩ G . \langle\chi\mathord{\downarrow_H},\psi\rangle_H=\langle\chi,\psi^G\rangle_G. ⟨ χ ↓ H , ψ ⟩ H = ⟨ χ , ψ G ⟩ G . Proof. Substitute the induced-character formula into the inner product on G G G , then regroup the sum by H H H -conjugacy classes.
A transitive permutation action of G G G on a set Ω \Omega Ω with point stabiliser H H H affords the permutation character
1 H G ( g ) = ∣ Fix Ω ( g ) ∣ . 1_H^G(g)=|\operatorname{Fix}_{\Omega}(g)|. 1 H G ( g ) = ∣ Fix Ω ( g ) ∣.
This makes orbit counting and character theory two views of the same data. In the digital setting the subgroup chains of Chapter 5 and the class data of § 11.10 provide the subgroup orders, class intersections and fixed-point counts needed to compute such characters exactly. Pólya’s theorem is the same statement for colourings: if G G G acts on s s s points and m m m colours are available, then g ↦ m c ( g ) g\mapsto m^{c(g)} g ↦ m c ( g ) , with c ( g ) c(g) c ( g ) the number of cycles of g g g , is the permutation character of G G G on the m s m^s m s colourings (§ 10.4.4 ). The number of orbits of colourings is its inner product with the trivial character, and the multiplicity of an irreducible χ \chi χ is 1 ∣ G ∣ ∑ g m c ( g ) χ ( g ) \frac{1}{|G|}\sum_g m^{c(g)}\chi(g) ∣ G ∣ 1 ∑ g m c ( g ) χ ( g ) .
Operations on characters
The representation ring carries operations that are visible directly at the level of characters.
Proposition 6.17 (Duals, tensor products, symmetric and exterior squares). Let χ \chi χ be a character. Then
χ ∗ ( g ) = χ ( g − 1 ) = χ ( g ) ‾ , χ ρ ⊗ σ = χ ρ χ σ , \chi^*(g)=\chi(g^{-1})=\overline{\chi(g)},\qquad\chi_{\rho\otimes\sigma}=\chi_\rho\chi_\sigma, χ ∗ ( g ) = χ ( g − 1 ) = χ ( g ) , χ ρ ⊗ σ = χ ρ χ σ , and
χ Sym 2 ( g ) = χ ( g ) 2 + χ ( g 2 ) 2 , χ ∧ 2 ( g ) = χ ( g ) 2 − χ ( g 2 ) 2 . \chi_{\operatorname{Sym}^2}(g)=\frac{\chi(g)^2+\chi(g^2)}{2},\qquad\chi_{\wedge^2}(g)=\frac{\chi(g)^2-\chi(g^2)}{2}. χ Sym 2 ( g ) = 2 χ ( g ) 2 + χ ( g 2 ) , χ ∧ 2 ( g ) = 2 χ ( g ) 2 − χ ( g 2 ) . Proof. Choose a basis in which a unitary matrix representing g g g is diagonal with eigenvalues λ 1 , … , λ n \lambda_1,\ldots,\lambda_n λ 1 , … , λ n . Then Sym 2 \operatorname{Sym}^2 Sym 2 and ∧ 2 \wedge^2 ∧ 2 have eigenvalues λ i λ j \lambda_i\lambda_j λ i λ j with i ≤ j i\leq j i ≤ j and i < j i<j i < j respectively. Summing them gives the formulas.
These identities explain why power maps belong naturally beside the character table: once g ↦ g m g\mapsto g^m g ↦ g m is known on conjugacy classes, operations such as symmetric and exterior powers become explicit class functions.
Definition 6.18 (Frobenius–Schur indicator). For an irreducible character χ \chi χ , the Frobenius–Schur indicator is
ν ( χ ) = 1 N ∑ g ∈ G χ ( g 2 ) . \nu(\chi)=\frac{1}{N}\sum_{g\in G}\chi(g^2). ν ( χ ) = N 1 g ∈ G ∑ χ ( g 2 ) .
Theorem 6.19 (Frobenius–Schur). For an irreducible character χ \chi χ one has ν ( χ ) ∈ { 1 , 0 , − 1 } \nu(\chi)\in\{1,0,-1\} ν ( χ ) ∈ { 1 , 0 , − 1 } . More precisely:
ν ( χ ) = 1 \nu(\chi)=1 ν ( χ ) = 1 if and only if χ \chi χ is afforded by a real representation preserving a symmetric bilinear form;
ν ( χ ) = 0 \nu(\chi)=0 ν ( χ ) = 0 if and only if χ \chi χ is not real-valued;
ν ( χ ) = − 1 \nu(\chi)=-1 ν ( χ ) = − 1 if and only if χ \chi χ is real-valued but only afforded over the quaternions, equivalently preserves an alternating form.
Proof. The multiplicity of the trivial character in Sym 2 χ \operatorname{Sym}^2\chi Sym 2 χ minus its multiplicity in ∧ 2 χ \wedge^2\chi ∧ 2 χ is exactly the average of χ ( g 2 ) \chi(g^2) χ ( g 2 ) . Irreducibility forces the invariant bilinear form to be zero, symmetric or alternating, which yields the three cases.
Products, quotients and linear characters
The linear characters are the irreducible characters of degree 1. They form a group under pointwise multiplication and are naturally identified with
Hom ( G , C × ) ≅ Hom ( G / G ′ , C × ) , \operatorname{Hom}(G,\mathbb{C}^{\times}) \cong\operatorname{Hom}(G/G',\mathbb{C}^{\times}), Hom ( G , C × ) ≅ Hom ( G / G ′ , C × ) ,
where G ′ G' G ′ is the commutator subgroup. Thus the first column of the character table detects the abelianization. The common kernel of all linear characters is exactly G ′ G' G ′ .
If N ◃ G N \triangleleft G N ◃ G , then the irreducible characters of G / N G/N G / N are precisely the irreducible characters of G G G whose kernel contains N N N . Hence quotient tables sit visibly inside the table of G G G .
If G = H × K G = H \times K G = H × K , every irreducible character has the form
χ ⊗ ψ , χ ∈ Irr ( H ) , ψ ∈ Irr ( K ) , \chi\otimes\psi,\qquad\chi\in\operatorname{Irr}(H),\ \psi\in\operatorname{Irr}(K), χ ⊗ ψ , χ ∈ Irr ( H ) , ψ ∈ Irr ( K ) ,
and the character table of a direct product is the Kronecker product of the two smaller tables. This describes all abelian groups at once: if
G ≅ C n 1 × ⋯ × C n r , G \cong C_{n_1} \times\cdots\times C_{n_r}, G ≅ C n 1 × ⋯ × C n r ,
then every irreducible character is one-dimensional and is given by a product of roots of unity.
From class data to the character table
Characters are constant on conjugacy classes, so the table is indexed by the class list of Chapter 5. Let C 1 , … , C k C_1,\ldots,C_k C 1 , … , C k be the classes with representatives g 1 , … , g k g_1,\ldots,g_k g 1 , … , g k , and write
T = ( χ i ( g j ) ) 1 ≤ i , j ≤ k . T = \left(\chi_i(g_j)\right)_{1\le i,j\le k}. T = ( χ i ( g j ) ) 1 ≤ i , j ≤ k .
The row and column orthogonality relations become matrix identities. The centre of the group algebra has basis K C 1 , … , K C k KC_1,\ldots,KC_k K C 1 , … , K C k , and
K C i K C j = ∑ m = 1 k a i j m K C m KC_iKC_j = \sum_{m=1}^{k} a_{ijm}KC_m K C i K C j = m = 1 ∑ k a ij m K C m
defines the class multiplication constants.
Theorem 6.20 (Character table from the class algebra). The common eigenvectors of the commuting operators K C j KC_j K C j on Z ( C G ) Z(\mathbb{C}G) Z ( C G ) are the rows of the irreducible character table, scaled by the degrees. Equivalently, the irreducible characters are the simultaneous one-dimensional characters of the class algebra.
Proof. The class sums are central, so on every irreducible module they act by the central character scalars of Theorem 6.13 . These scalars determine the character values after division by class sizes and degrees. Conversely, the orthogonality relations reconstruct the irreducible rows from these eigenvalues.
This is the theoretical basis of the Burnside–Dixon–Schneider method: compute the classes and the class algebra, diagonalise enough central multiplication operators, and recover the irreducible rows by exact arithmetic and orthogonality.
Digital computation roadmap
The character tables of this book are all computed by the following workflow.
Compute the conjugacy classes, class sizes, centraliser orders and power maps by the methods of § 11.10 .
Obtain subgroup chains, transversals and permutation actions from § 11.7 .
Build permutation characters, induced characters and tensor constructions as explicit class functions.
Compute or infer the class multiplication constants in the centre of the group algebra.
Recover the irreducible rows using Theorem 6.20 , then verify orthogonality, degree sums, central characters, and power-map constraints.
Section 13.16 carries out the first steps of this workflow from the digital laws of sporadic groups: character degrees from rank-three actions, classes with power maps, and the complete character table of M 11 M_{11} M 11 .
Worked consequences and outlook
The abstract theory above already yields concrete information before any full table is written down. Degree-one characters determine the abelianization; kernels of irreducibles detect normal subgroups; direct-product decompositions factor the table; power maps control symmetric and exterior squares; and the Frobenius–Schur indicator distinguishes real, complex and quaternionic character types.
The next stage is systematic tabulation. For small groups one can list all irreducible rows explicitly. For families such as PSL ( 2 , q ) \operatorname{PSL}(2,q) PSL ( 2 , q ) and the Suzuki groups one can write uniform character formulas. For the sporadic and exceptional simple groups, the digital class algorithms supply the conjugacy data and the permutation actions that feed the class-algebra and induction machinery.
Exercises
Prove directly from Definition 6.1 that equivalent representations have the same character.
Verify the orthogonality relations for the character table of S 3 S_3 S 3 .
Show that the primitive central idempotents in (54) are orthogonal and sum to 1.
Compute the linear characters of a finite abelian group from its invariant factors.
Decompose the permutation character of a transitive action into irreducibles by Frobenius reciprocity.
Use the formulas for Sym 2 \operatorname{Sym}^{2} Sym 2 and ⋀ 2 \bigwedge^{2} ⋀ 2 to distinguish two irreducible characters with the same degree.
Prove Corollary 6.14 from Theorem 6.13 in full detail.
Explain how the class multiplication constants determine the character table.
Cyclic and Abelian groups
Cyclic groups
Definition 7.1. The cyclic group C N C_N C N is A N \mathcal{A}_N A N with a ∗ b = d N ( 0 , a + b ) a * b = d_N(0,a+b) a ∗ b = d N ( 0 , a + b ) , i.e. with code
R = ∑ a , b d N ( 0 , a + b ) N a N + b . R = \sum_{a,b} d_N(0,a+b)\,N^{aN+b}. R = a , b ∑ d N ( 0 , a + b ) N a N + b .
Definition 7.2. In a group ( A N , ∗ ) (\mathcal{A}_N,*) ( A N , ∗ ) write x ∗ 0 = 0 x^{*0}=0 x ∗ 0 = 0 and x ∗ ( n + 1 ) = x ∗ n ∗ x x^{*(n+1)}=x^{*n}*x x ∗ ( n + 1 ) = x ∗ n ∗ x . The order of x x x is the least n ≥ 1 n \ge1 n ≥ 1 with x ∗ n = 0 x^{*n}=0 x ∗ n = 0 . The group is cyclic if some a a a (a generator ) satisfies { a ∗ n } = A N \{a^{*n}\}=\mathcal{A}_N { a ∗ n } = A N ; we write G = ⟨ a ⟩ G=\langle a\rangle G = ⟨ a ⟩ .
Proposition 7.3. C N C_N C N is an Abelian group with identity 0 0 0 , inverse x ↦ d N ( 0 , N − x ) x \mapsto d_N(0,N-x) x ↦ d N ( 0 , N − x ) , and generator 1 1 1 , since 1 ∗ k = d N ( 0 , k ) 1^{*k}=d_N(0,k) 1 ∗ k = d N ( 0 , k ) . Its row codes are the cyclic shifts L a = ∑ b d N ( 0 , a + b ) N b L_a=\sum_b d_N(0,a+b)\,N^b L a = ∑ b d N ( 0 , a + b ) N b . Every cyclic group of order N N N is isomorphic to C N C_N C N via d N ( 0 , k ) ↦ a ∗ k d_N(0,k)\mapsto a^{*k} d N ( 0 , k ) ↦ a ∗ k .
Proof. By Lemma 1.3 , d N ( 0 , x + d N ( 0 , y + z ) ) = d N ( 0 , x + y + z ) = d N ( 0 , d N ( 0 , x + y ) + z ) d_N(0,x+d_N(0,y+z))=d_N(0,x+y+z)=d_N(0,d_N(0,x+y)+z) d N ( 0 , x + d N ( 0 , y + z )) = d N ( 0 , x + y + z ) = d N ( 0 , d N ( 0 , x + y ) + z ) ; commutativity, identity and inverses are clear. If a a a has order N N N , then a ∗ j ∗ a ∗ k = a ∗ d N ( 0 , j + k ) a^{*j}*a^{*k}=a^{*d_N(0,j+k)} a ∗ j ∗ a ∗ k = a ∗ d N ( 0 , j + k ) , so the map is an isomorphism.
Theorem 7.4 (Carry decomposition). Let N = p 0 q N=p_0q N = p 0 q with p 0 , q ≥ 1 p_0,q\ge1 p 0 , q ≥ 1 . Every x ∈ A N x\in\mathcal{A}_N x ∈ A N splits as x = d p 0 ( 0 , x ) + p 0 ⌊ x / p 0 ⌋ x=d_{p_0}(0,x)+p_0\lfloor x/p_0\rfloor x = d p 0 ( 0 , x ) + p 0 ⌊ x / p 0 ⌋ , and
d N ( 0 , x + y ) = d p 0 ( 0 , x + y ) + p 0 d q ( 0 , ⌊ x p 0 ⌋ + ⌊ y p 0 ⌋ + d p 0 ( 0 , x ) + d p 0 ( 0 , y ) p 0 ) . (55) d_N(0,x+y)=d_{p_0}(0,x+y)+p_0d_q\left(0,\left\lfloor\frac{x}{p_0}\right\rfloor+\left\lfloor\frac{y}{p_0}\right\rfloor+\frac{d_{p_0}(0,x)+d_{p_0}(0,y)}{p_0}\right). \tag*{(55)} d N ( 0 , x + y ) = d p 0 ( 0 , x + y ) + p 0 d q ( 0 , ⌊ p 0 x ⌋ + ⌊ p 0 y ⌋ + p 0 d p 0 ( 0 , x ) + d p 0 ( 0 , y ) ) . (55) Hence x ↦ d p 0 ( 0 , x ) x\mapsto d_{p_0}(0,x) x ↦ d p 0 ( 0 , x ) is a surjective homomorphism C N → C p 0 C_N\to C_{p_0} C N → C p 0 . Its kernel p 0 A q p_0\mathcal{A}_q p 0 A q is a subgroup isomorphic to C q C_q C q via x ↦ x / p 0 x\mapsto x/p_0 x ↦ x / p 0 , on which the law reads x ∗ y = p 0 d q ( 0 , x / p 0 + y / p 0 ) x*y=p_0d_q(0,x/p_0+y/p_0) x ∗ y = p 0 d q ( 0 , x / p 0 + y / p 0 ) .
Proof. Write x = x 0 + p 0 X x=x_0+p_0X x = x 0 + p 0 X and y = y 0 + p 0 Y y=y_0+p_0Y y = y 0 + p 0 Y with x 0 , y 0 ∈ A p 0 x_0,y_0\in\mathcal{A}_{p_0} x 0 , y 0 ∈ A p 0 , and let c = ⌊ ( x 0 + y 0 ) / p 0 ⌋ ∈ { 0 , 1 } c=\lfloor(x_0+y_0)/p_0\rfloor\in\{0,1\} c = ⌊( x 0 + y 0 ) / p 0 ⌋ ∈ { 0 , 1 } be the carry . Then x + y = d p 0 ( 0 , x 0 + y 0 ) + p 0 ( X + Y + c ) x+y=d_{p_0}(0,x_0+y_0)+p_0(X+Y+c) x + y = d p 0 ( 0 , x 0 + y 0 ) + p 0 ( X + Y + c ) , whose first term lies in A p 0 \mathcal{A}_{p_0} A p 0 . Reducing modulo N = p 0 q N=p_0q N = p 0 q gives d N ( 0 , x + y ) = d p 0 ( 0 , x + y ) + p 0 d q ( 0 , X + Y + c ) d_N(0,x+y)=d_{p_0}(0,x+y)+p_0d_q(0,X+Y+c) d N ( 0 , x + y ) = d p 0 ( 0 , x + y ) + p 0 d q ( 0 , X + Y + c ) . Finally, d q ( 0 , t ) = d q ( 0 , ⌊ t ⌋ ) d_q(0,t)=d_q(0,\lfloor t\rfloor) d q ( 0 , t ) = d q ( 0 , ⌊ t ⌋) for real t t t , because ⌊ t / q ⌋ = ⌊ ⌊ t ⌋ / q ⌋ \lfloor t/q\rfloor=\lfloor\lfloor t\rfloor/q\rfloor ⌊ t / q ⌋ = ⌊⌊ t ⌋ / q ⌋ , and ⌊ X + Y + ( x 0 + y 0 ) / p 0 ⌋ = X + Y + c \lfloor X+Y+(x_0+y_0)/p_0\rfloor=X+Y+c ⌊ X + Y + ( x 0 + y 0 ) / p 0 ⌋ = X + Y + c .
Theorem 7.6 (Subgroups of C N C_N C N ). For every divisor d d d of N N N , C N C_N C N has exactly one subgroup of order d d d , namely { k N / d : k ∈ A d } ≅ C d \{kN/d:k\in\mathcal{A}_d\}\cong C_d { k N / d : k ∈ A d } ≅ C d , and there are no others. The element k k k generates C N C_N C N if and only if gcd ( k , N ) = 1 \gcd(k,N)=1 g cd( k , N ) = 1 .
Proof. By Theorem 7.4 , { k N / d } \{kN/d\} { k N / d } is a subgroup of order d d d . Conversely, let H ≠ { 0 } H \ne\{0\} H = { 0 } be a subgroup and e e e its least positive element. For h ∈ H h \in H h ∈ H , Euclidean division h = t e + r h=te+r h = t e + r gives r = d N ( 0 , h − t e ) ∈ H r=d_N(0,h-te) \in H r = d N ( 0 , h − t e ) ∈ H with r < e r<e r < e , so r = 0 r=0 r = 0 ; the same argument with h = N ≡ 0 h=N \equiv0 h = N ≡ 0 shows e ∣ N e \mid N e ∣ N . Thus H = { k e } H=\{ke\} H = { k e } has order d = N / e d=N/e d = N / e . The subgroup generated by k k k is generated by gcd ( k , N ) \gcd(k,N) g cd( k , N ) .
Proposition 7.7 (Necklaces). The number of colourings of the positions A N \mathcal{A}_N A N with m m m colours, up to rotation x ↦ d N ( 0 , x + k ) x \mapsto d_N(0,x+k) x ↦ d N ( 0 , x + k ) , is
1 N ∑ d ∣ N φ ( d ) m N / d \frac{1}{N}\sum_{d\mid N}\varphi(d)m^{N/d} N 1 d ∣ N ∑ φ ( d ) m N / d
Proof. The orbits of x ↦ d N ( 0 , x + k ) x \mapsto d_N(0,x+k) x ↦ d N ( 0 , x + k ) are the cosets of ⟨ k ⟩ \langle k\rangle ⟨ k ⟩ , which has order N / gcd ( k , N ) N/\gcd(k,N) N / g cd( k , N ) by Theorem 7.6 ; so the rotation has g = gcd ( k , N ) g=\gcd(k,N) g = g cd( k , N ) cycles. Exactly φ ( N / g ) \varphi(N/g) φ ( N / g ) values of k k k have gcd ( k , N ) = g \gcd(k,N)=g g cd( k , N ) = g . By Theorem 4.21 the number of orbits is 1 N ∑ g ∣ N φ ( N / g ) m g \frac{1}{N}\sum_{g\mid N}\varphi(N/g)m^g N 1 ∑ g ∣ N φ ( N / g ) m g , which is the stated sum with d = N / g d=N/g d = N / g .
For N = 6 N=6 N = 6 and m = 2 m=2 m = 2 this is ( 64 + 8 + 2 ⋅ 4 + 2 ⋅ 2 ) / 6 = 14 (64+8+2\cdot4+2\cdot2)/6=14 ( 64 + 8 + 2 ⋅ 4 + 2 ⋅ 2 ) /6 = 14 . In digits, a colouring is a string of N N N base-m m m digits and the rotation is the cyclic shift of digits used in the proof of Cauchy’s theorem (Theorem 4.9); necklaces are strings up to cyclic shift.
Multiplicative cyclic groups
Theorem 7.8 (Fermat’s little theorem). Let p p p be prime and a ∈ Z a\in\mathbb{Z} a ∈ Z . Then p ∣ ( a p − a ) p\mid(a^p-a) p ∣ ( a p − a ) , hence d p ( 0 , a p ) = d p ( 0 , a ) d_p(0,a^p)=d_p(0,a) d p ( 0 , a p ) = d p ( 0 , a ) . If p ∤ a p\nmid a p ∤ a , then d p ( 0 , a p − 1 ) = 1 d_p(0,a^{p-1})=1 d p ( 0 , a p − 1 ) = 1 .
Proof. The classes 1 , … , p − 1 1,\ldots,p-1 1 , … , p − 1 are permuted by multiplication by a a a modulo p p p when p ∤ a p\nmid a p ∤ a , so
∏ k = 1 p − 1 a k ≡ ∏ k = 1 p − 1 k ( m o d p ) . \prod_{k=1}^{p-1}ak\equiv\prod_{k=1}^{p-1}k\pmod p. k = 1 ∏ p − 1 ak ≡ k = 1 ∏ p − 1 k ( mod p ) .
Cancelling ( p − 1 ) ! (p-1)! ( p − 1 )! gives a p − 1 ≡ 1 ( m o d p ) a^{p-1}\equiv1\pmod p a p − 1 ≡ 1 ( mod p ) . Multiplying by a a a yields a p ≡ a ( m o d p ) a^p\equiv a\pmod p a p ≡ a ( mod p ) . The case p ∣ a p\mid a p ∣ a is trivial.
Proposition 7.9 (Linear congruences). Let p p p be prime, let a ∈ Z a\in\mathbb{Z} a ∈ Z with p ∤ a p\nmid a p ∤ a , and let b , c ∈ A p b,c\in\mathcal{A}_p b , c ∈ A p . Then the congruence
d p ( 0 , a x + b ) = c d_p(0,ax+b)=c d p ( 0 , a x + b ) = c has the unique solution
x = d p ( 0 , a p − 2 ( c − b ) ) . x=d_p(0,a^{p-2}(c-b)). x = d p ( 0 , a p − 2 ( c − b )) . In particular, the multiplicative inverse of a a a modulo p p p is a − 1 = d p ( 0 , a p − 2 ) a^{-1}=d_p(0,a^{p-2}) a − 1 = d p ( 0 , a p − 2 ) .
Proof. By Theorem 7.8 , d p ( 0 , a p − 1 ) = 1 d_p(0,a^{p-1})=1 d p ( 0 , a p − 1 ) = 1 , so d p ( 0 , a d p ( 0 , a p − 2 ( c − b ) ) + b ) = d p ( 0 , a p − 1 ( c − b ) + b ) = c d_p(0,a\,d_p(0,a^{p-2}(c-b))+b)=d_p(0,a^{p-1}(c-b)+b)=c d p ( 0 , a d p ( 0 , a p − 2 ( c − b )) + b ) = d p ( 0 , a p − 1 ( c − b ) + b ) = c . Uniqueness follows because multiplication by the nonzero class a a a is a permutation of A p \mathcal{A}_p A p .
Definition 7.10. Let U ( p ) = A p ∖ { 0 } U(p)=\mathcal{A}_p\setminus\{0\} U ( p ) = A p ∖ { 0 } with multiplication modulo p p p :
x ⋅ y = d p ( 0 , x y ) . x\cdot y=d_p(0,xy). x ⋅ y = d p ( 0 , x y ) . This is the multiplicative group of units modulo p p p when every nonzero class is invertible.
Theorem 7.11 (Multiplicative cyclic groups). Let p p p be prime. Then U ( p ) U(p) U ( p ) is a group of order p − 1 p-1 p − 1 . If g g g is a primitive root modulo p p p , the map
x ↦ d p ( 0 , g x ) x\mapsto d_p(0,g^x) x ↦ d p ( 0 , g x ) is an isomorphism C p − 1 → U ( p ) C_{p-1}\to U(p) C p − 1 → U ( p ) . Thus U ( p ) U(p) U ( p ) is cyclic.
Proof. Closure and associativity come from integer multiplication and Lemma 1.3 . The identity is 1 1 1 , and inverses are given by Proposition 7.9. If g g g has order p − 1 p-1 p − 1 , then the displayed map is bijective and d p ( 0 , g p − 1 ( 0 , x + y ) ) = d p ( 0 , g x g y ) d_p(0,g^{p-1}(0,x+y))=d_p(0,g^xg^y) d p ( 0 , g p − 1 ( 0 , x + y )) = d p ( 0 , g x g y ) because g p − 1 ≡ 1 ( m o d p ) g^{p-1}\equiv1\pmod p g p − 1 ≡ 1 ( mod p ) .
Quadratic residues and the sign of a permutation
Let p p p be an odd prime and let U p = A p ∖ { 0 } U_p = \mathcal{A}_p \setminus\{0\} U p = A p ∖ { 0 } with the law x ⋅ y = d p ( 0 , x y ) x \cdot y = d_p(0,xy) x ⋅ y = d p ( 0 , x y ) , the multiplicative group modulo p p p . The squares Q p = { d p ( 0 , x 2 ) : x ∈ U p } Q_p = \{d_p(0,x^2) : x \in U_p\} Q p = { d p ( 0 , x 2 ) : x ∈ U p } are the quadratic residues modulo p p p , and the other elements of U p U_p U p are the quadratic nonresidues. This section shows that Q p Q_p Q p is the kernel of the sign of a permutation code, and obtains Gauss’s quadratic reciprocity law by comparing the two digital expansions of A p q \mathcal{A}_{pq} A pq . All numerical claims are checked by scripts/qr_verify.py.
Proposition 7.12 (Residues and Euler’s criterion). Q p Q_p Q p is a subgroup of index 2 2 2 of U p U_p U p . For a ∈ U p a \in U_p a ∈ U p one has d p ( 0 , a ( p − 1 ) / 2 ) ∈ { 1 , p − 1 } d_p(0,a^{(p-1)/2}) \in\{1,p-1\} d p ( 0 , a ( p − 1 ) /2 ) ∈ { 1 , p − 1 } , with the value 1 1 1 exactly when a ∈ Q p a \in Q_p a ∈ Q p . Hence the Legendre symbol
( a p ) = d p ( 0 , a ( p − 1 ) / 2 + 1 ) − 1 (56) \left(\frac{a}{p}\right) = d_p\left(0,a^{(p-1)/2}+1\right)-1 \tag*{(56)} ( p a ) = d p ( 0 , a ( p − 1 ) /2 + 1 ) − 1 (56)
equals 1 1 1 on Q p Q_p Q p and − 1 -1 − 1 on U p ∖ Q p U_p \setminus Q_p U p ∖ Q p , and a ↦ ( a / p ) a \mapsto(a/p) a ↦ ( a / p ) is a homomorphism U p → { ± 1 } U_p \to\{\pm1\} U p → { ± 1 } .
Proof. The squaring map x ↦ d p ( 0 , x 2 ) x \mapsto d_p(0,x^2) x ↦ d p ( 0 , x 2 ) is a homomorphism of the abelian group U p U_p U p with image Q p Q_p Q p . Its kernel consists of the x x x with p ∣ ( x − 1 ) ( x + 1 ) p \mid(x-1)(x+1) p ∣ ( x − 1 ) ( x + 1 ) , that is x ∈ { 1 , p − 1 } x \in\{1,p-1\} x ∈ { 1 , p − 1 } by Euclid’s lemma, so ∣ Q p ∣ = ( p − 1 ) / 2 |Q_p|=(p-1)/2 ∣ Q p ∣ = ( p − 1 ) /2 . By Lagrange’s theorem d p ( 0 , a p − 1 ) = 1 d_p(0,a^{p-1})=1 d p ( 0 , a p − 1 ) = 1 , so b = d p ( 0 , a ( p − 1 ) / 2 ) b=d_p(0,a^{(p-1)/2}) b = d p ( 0 , a ( p − 1 ) /2 ) satisfies d p ( 0 , b 2 ) = 1 d_p(0,b^2)=1 d p ( 0 , b 2 ) = 1 and therefore b ∈ { 1 , p − 1 } b \in\{1,p-1\} b ∈ { 1 , p − 1 } . If a = d p ( 0 , x 2 ) a=d_p(0,x^2) a = d p ( 0 , x 2 ) then b = d p ( 0 , x p − 1 ) = 1 b=d_p(0,x^{p-1})=1 b = d p ( 0 , x p − 1 ) = 1 . Conversely, the a a a with b = 1 b=1 b = 1 are roots of X ( p − 1 ) / 2 − 1 X^{(p-1)/2}-1 X ( p − 1 ) /2 − 1 over the field A p \mathcal{A}_p A p , of which there are at most ( p − 1 ) / 2 (p-1)/2 ( p − 1 ) /2 ; since Q p Q_p Q p already supplies ( p − 1 ) / 2 (p-1)/2 ( p − 1 ) /2 of them, b = 1 b=1 b = 1 holds exactly on Q p Q_p Q p . The power map a ↦ a ( p − 1 ) / 2 a \mapsto a^{(p-1)/2} a ↦ a ( p − 1 ) /2 is a homomorphism because U p U_p U p is abelian, and (56) sends b = 1 b=1 b = 1 to 1 1 1 and b = p − 1 b=p-1 b = p − 1 to d p ( 0 , p ) − 1 = − 1 d_p(0,p)-1=-1 d p ( 0 , p ) − 1 = − 1 .
Theorem 7.13 (Zolotarev). For a ∈ U p a \in U_p a ∈ U p let π a ∈ Sym ( A p ) \pi_a \in\operatorname{Sym}(\mathcal{A}_p) π a ∈ Sym ( A p ) be the permutation x ↦ d p ( 0 , a x ) x \mapsto d_p(0,ax) x ↦ d p ( 0 , a x ) . Then sgn π a = ( a / p ) \operatorname{sgn}\pi_a=(a/p) sgn π a = ( a / p ) .
Proof. Let k k k be the order of a a a in U p U_p U p . The permutation π a \pi_a π a fixes 0 0 0 , and on U p U_p U p its cycles are the cosets x ⟨ a ⟩ x\langle a\rangle x ⟨ a ⟩ , each a k k k -cycle; by Lagrange’s theorem there are ( p − 1 ) / k (p-1)/k ( p − 1 ) / k of them, so sgn π a = ( − 1 ) ( k − 1 ) ( p − 1 ) / k \operatorname{sgn}\pi_a=(-1)^{(k-1)(p-1)/k} sgn π a = ( − 1 ) ( k − 1 ) ( p − 1 ) / k . By Proposition 7.12, a ∈ Q p a \in Q_p a ∈ Q p if and only if d p ( 0 , a ( p − 1 ) / 2 ) = 1 d_p(0,a^{(p-1)/2})=1 d p ( 0 , a ( p − 1 ) /2 ) = 1 , that is, if and only if k k k divides ( p − 1 ) / 2 (p-1)/2 ( p − 1 ) /2 , i.e. ( p − 1 ) / k (p-1)/k ( p − 1 ) / k is even. If k k k is odd, ( p − 1 ) / k (p-1)/k ( p − 1 ) / k is even and both sides equal 1 1 1 . If k k k is even, sgn π a = ( − 1 ) ( p − 1 ) / k \operatorname{sgn}\pi_a=(-1)^{(p-1)/k} sgn π a = ( − 1 ) ( p − 1 ) / k , which is 1 1 1 exactly when ( p − 1 ) / k (p-1)/k ( p − 1 ) / k is even.
In the language of Chapter 2, π a \pi_a π a is the permutation code with digits d p ( 0 , a x ) d_p(0,ax) d p ( 0 , a x ) , x ∈ A p x \in\mathcal{A}_p x ∈ A p , and ( a / p ) (a/p) ( a / p ) is its sign. Thus the Legendre symbol is the composite U p → Sym ( A p ) → { ± 1 } U_p \to\operatorname{Sym}(\mathcal{A}_p) \to\{\pm1\} U p → Sym ( A p ) → { ± 1 } of the regular action of U p U_p U p on A p \mathcal{A}_p A p and the sign; its kernel Q p Q_p Q p is the preimage of the alternating group.
Example 7.14. For p = 7 p=7 p = 7 one has Q 7 = { 1 , 2 , 4 } Q_7=\{1,2,4\} Q 7 = { 1 , 2 , 4 } . The permutation π 2 = ( 1 2 4 ) ( 3 6 5 ) \pi_2=(1\ 2\ 4)(3\ 6\ 5) π 2 = ( 1 2 4 ) ( 3 6 5 ) is even, while π 3 = ( 1 3 2 6 4 5 ) \pi_3=(1\ 3\ 2\ 6\ 4\ 5) π 3 = ( 1 3 2 6 4 5 ) is a 6 6 6 -cycle and is odd, in agreement with ( 2 / 7 ) = 1 (2/7)=1 ( 2/7 ) = 1 and ( 3 / 7 ) = − 1 (3/7)=-1 ( 3/7 ) = − 1 . In digits, d 7 ( 0 , 2 3 + 1 ) − 1 = d 7 ( 0 , 9 ) − 1 = 1 d_7(0,2^3+1)-1=d_7(0,9)-1=1 d 7 ( 0 , 2 3 + 1 ) − 1 = d 7 ( 0 , 9 ) − 1 = 1 and d 7 ( 0 , 3 3 + 1 ) − 1 = d 7 ( 0 , 28 ) − 1 = − 1 d_7(0,3^3+1)-1=d_7(0,28)-1=-1 d 7 ( 0 , 3 3 + 1 ) − 1 = d 7 ( 0 , 28 ) − 1 = − 1 .
Proposition 7.15 (Gauss’s lemma). For a ∈ U p a \in U_p a ∈ U p ,
( a p ) = ( − 1 ) r , r = ∑ j = 1 ( p − 1 ) / 2 ⌊ 2 d p ( 0 , a j ) p ⌋ \left(\frac{a}{p}\right)=(-1)^r,\qquad r=\sum_{j=1}^{(p-1)/2}\left\lfloor\frac{2d_p(0,aj)}{p}\right\rfloor ( p a ) = ( − 1 ) r , r = j = 1 ∑ ( p − 1 ) /2 ⌊ p 2 d p ( 0 , aj ) ⌋
so that r r r counts the j ≤ ( p − 1 ) / 2 j \le(p-1)/2 j ≤ ( p − 1 ) /2 with d p ( 0 , a j ) > p / 2 d_p(0,aj)>p/2 d p ( 0 , aj ) > p /2 .
Proof. Put T = { 1 , … , ( p − 1 ) / 2 } T=\{1,\ldots,(p-1)/2\} T = { 1 , … , ( p − 1 ) /2 } , a transversal of the subgroup { 1 , p − 1 } \{1,p-1\} { 1 , p − 1 } of U p U_p U p . For j ∈ T j \in T j ∈ T write d p ( 0 , a j ) = d p ( 0 , ε j k j ) d_p(0,aj)=d_p(0,\varepsilon_jk_j) d p ( 0 , aj ) = d p ( 0 , ε j k j ) with k j ∈ T k_j \in T k j ∈ T and ε j ∈ { 1 , p − 1 } \varepsilon_j \in\{1,p-1\} ε j ∈ { 1 , p − 1 } ; then ε j = p − 1 \varepsilon_j=p-1 ε j = p − 1 exactly when d p ( 0 , a j ) > p / 2 d_p(0,aj)>p/2 d p ( 0 , aj ) > p /2 , i.e. when the j j j -th term of r r r is 1 1 1 . If k j = k j ′ k_j=k_{j'} k j = k j ′ then a j ≡ ± a j ′ aj\equiv\pm aj' aj ≡ ± a j ′ , hence j ≡ ± j ′ ( m o d p ) j\equiv\pm j'\pmod p j ≡ ± j ′ ( mod p ) and j = j ′ j=j' j = j ′ ; so j ↦ k j j\mapsto k_j j ↦ k j is a permutation of T T T . Multiplying over j ∈ T j\in T j ∈ T gives a ( p − 1 ) / 2 ∏ j ∈ T j ≡ ( − 1 ) r ∏ j ∈ T j ( m o d p ) a^{(p-1)/2}\prod_{j\in T}j\equiv(-1)^r\prod_{j\in T}j\pmod p a ( p − 1 ) /2 ∏ j ∈ T j ≡ ( − 1 ) r ∏ j ∈ T j ( mod p ) , and cancelling and applying Proposition 7.12 proves the claim.
Remark 7.16 (Gauss’s lemma is a transfer). The element a a a permutes the cosets { t , p − t } \{t,p-t\} { t , p − t } , t ∈ T t \in T t ∈ T , of { 1 , p − 1 } ≅ C 2 \{1,p-1\} \cong C_2 { 1 , p − 1 } ≅ C 2 , and the signs ε j \varepsilon_j ε j record which representative is hit. The map a ↦ ∏ j ε j a \mapsto\prod_j\varepsilon_j a ↦ ∏ j ε j is therefore the transfer of U p U_p U p into { 1 , p − 1 } \{1,p-1\} { 1 , p − 1 } computed with the transversal T T T , and Gauss’s lemma states that this transfer is the Legendre symbol. The same construction for a Sylow subgroup of a general finite group is the transfer homomorphism behind Burnside’s normal p p p -complement theorem.
Theorem 7.17 (Quadratic reciprocity). For distinct odd primes p p p and q q q ,
( p q ) ( q p ) = ( − 1 ) p − 1 2 ⋅ q − 1 2 . \left(\frac{p}{q}\right)\left(\frac{q}{p}\right)=(-1)^{\frac{p-1}{2}\cdot\frac{q-1}{2}}. ( q p ) ( p q ) = ( − 1 ) 2 p − 1 ⋅ 2 q − 1 .
Proof. Put X = A p × A q X=\mathcal{A}_p\times\mathcal{A}_q X = A p × A q and consider three bijections between X X X and A p q \mathcal{A}_{pq} A pq :
φ ( u , v ) = u + p v , ψ ( u , v ) = v + q u , χ − 1 ( x ) = ( d p ( 0 , x ) , d q ( 0 , x ) ) . \varphi(u,v)=u+pv,\qquad\psi(u,v)=v+qu,\qquad\chi^{-1}(x)=\left(d_p(0,x),d_q(0,x)\right). φ ( u , v ) = u + p v , ψ ( u , v ) = v + q u , χ − 1 ( x ) = ( d p ( 0 , x ) , d q ( 0 , x ) ) .
The first two are the digital expansions of A p q \mathcal{A}_{pq} A pq with radices ( p , q ) (p,q) ( p , q ) and ( q , p ) (q,p) ( q , p ) (Lemma 1.2 ), and χ − 1 \chi^{-1} χ − 1 is a bijection by the Chinese remainder theorem. Then
χ − 1 φ ( u , v ) = ( u , d q ( 0 , u + p v ) ) , χ − 1 ψ ( u , v ) = ( d p ( 0 , v + q u ) , v ) . \chi^{-1}\varphi(u,v)=\left(u,d_q(0,u+pv)\right),\qquad\chi^{-1}\psi(u,v)=\left(d_p(0,v+qu),v\right). χ − 1 φ ( u , v ) = ( u , d q ( 0 , u + p v ) ) , χ − 1 ψ ( u , v ) = ( d p ( 0 , v + q u ) , v ) .
The permutation χ − 1 φ \chi^{-1}\varphi χ − 1 φ of X X X preserves each row { u } × A q \{u\}\times\mathcal{A}_q { u } × A q and acts on it as v ↦ d q ( 0 , u + p v ) v\mapsto d_q(0,u+pv) v ↦ d q ( 0 , u + p v ) , which is π d q ( 0 , p ) \pi_{d_q(0,p)} π d q ( 0 , p ) on A q \mathcal{A}_q A q followed by the translation v ↦ d q ( 0 , u + v ) v\mapsto d_q(0,u+v) v ↦ d q ( 0 , u + v ) . The translation is even, since all its cycles have the odd length q q q or 1 1 1 , so by Theorem 7.13 each row contributes ( p / q ) (p/q) ( p / q ) and sgn ( χ − 1 φ ) = ( p / q ) p = ( p / q ) \operatorname{sgn}(\chi^{-1}\varphi)=(p/q)^p=(p/q) sgn ( χ − 1 φ ) = ( p / q ) p = ( p / q ) . In the same way χ − 1 ψ \chi^{-1}\psi χ − 1 ψ preserves the columns and sgn ( χ − 1 ψ ) = ( q / p ) q = ( q / p ) \operatorname{sgn}(\chi^{-1}\psi)=(q/p)^q=(q/p) sgn ( χ − 1 ψ ) = ( q / p ) q = ( q / p ) . Hence ψ − 1 φ = ( χ − 1 ψ ) − 1 ( χ − 1 φ ) \psi^{-1}\varphi=(\chi^{-1}\psi)^{-1}(\chi^{-1}\varphi) ψ − 1 φ = ( χ − 1 ψ ) − 1 ( χ − 1 φ ) has sign ( p / q ) ( q / p ) (p/q)(q/p) ( p / q ) ( q / p ) .
On the other hand ψ − 1 φ \psi^{-1}\varphi ψ − 1 φ is conjugate to the permutation φ ψ − 1 : v + q u ↦ u + p v \varphi\psi^{-1}:v+qu\mapsto u+pv φ ψ − 1 : v + q u ↦ u + p v of A p q \mathcal{A}_{pq} A pq , which reads the two digits of an element in the opposite order (the transposition of a p × q p\times q p × q array). A pair v + q u < v ′ + q u ′ v+qu<v'+qu' v + q u < v ′ + q u ′ is inverted exactly when u < u ′ u<u' u < u ′ and v > v ′ v>v' v > v ′ , so φ ψ − 1 \varphi\psi^{-1} φ ψ − 1 has ( p 2 ) ( q 2 ) = p q p − 1 2 q − 1 2 \binom{p}{2}\binom{q}{2}=pq\frac{p-1}{2}\frac{q-1}{2} ( 2 p ) ( 2 q ) = pq 2 p − 1 2 q − 1 inversions, and this number has the parity of p − 1 2 ⋅ q − 1 2 \frac{p-1}{2}\cdot\frac{q-1}{2} 2 p − 1 ⋅ 2 q − 1 . □
Example 7.18. For p = 3 p=3 p = 3 and q = 5 q=5 q = 5 , Q 5 = { 1 , 4 } Q_5=\{1,4\} Q 5 = { 1 , 4 } and Q 3 = { 1 } Q_3=\{1\} Q 3 = { 1 } , so ( 3 / 5 ) = ( 5 / 3 ) = ( 2 / 3 ) = − 1 (3/5)=(5/3)=(2/3)=-1 ( 3/5 ) = ( 5/3 ) = ( 2/3 ) = − 1 and the product is 1 = ( − 1 ) 1 ⋅ 2 1=(-1)^{1\cdot2} 1 = ( − 1 ) 1 ⋅ 2 . The transposition φ ψ − 1 \varphi\psi^{-1} φ ψ − 1 of the 3 × 5 3\times5 3 × 5 array has ( 3 2 ) ( 5 2 ) = 30 \binom{3}{2}\binom{5}{2}=30 ( 2 3 ) ( 2 5 ) = 30 inversions and is even.
Remark 7.19 (Where quadratic residues appear in finite groups). For g g g of prime order p p p in a finite group G G G , g k g^k g k is conjugate to g g g exactly when k k k lies in the image of N G ( ⟨ g ⟩ ) → Aut ⟨ g ⟩ ≅ U p N_G(\langle g\rangle)\to\operatorname{Aut}\langle g\rangle\cong U_p N G (⟨ g ⟩) → Aut ⟨ g ⟩ ≅ U p . When this image is Q p Q_p Q p , the elements of order p p p form two classes, interchanged by g ↦ g − 1 g\mapsto g^{-1} g ↦ g − 1 if and only if ( − 1 / p ) = − 1 (-1/p)=-1 ( − 1/ p ) = − 1 , and the character values on them lie in Q ( ε p ) \mathbb{Q}(\sqrt{\varepsilon p}) Q ( εp ) , ε = ( − 1 / p ) \varepsilon=(-1/p) ε = ( − 1/ p ) , the field of the Gauss period ∑ k ∈ Q p ζ p k = 1 2 ( − 1 + ε p ) \sum_{k\in Q_p}\zeta_p^k=\frac{1}{2}(-1+\sqrt{\varepsilon p}) ∑ k ∈ Q p ζ p k = 2 1 ( − 1 + εp ) . This is the case for the classes of order 11 in M 12 M_{12} M 12 (§ 11.11 ) and for the unipotent classes of P S L ( 2 , p ) PSL(2,p) P S L ( 2 , p ) (§ 20.3 ), whose two characters of degree ( p + ε ) / 2 (p+\varepsilon)/2 ( p + ε ) /2 take the values 1 2 ( − 1 ± ε p ) \frac{1}{2}(-1\pm\sqrt{\varepsilon p}) 2 1 ( − 1 ± εp ) . Quadratic residues also build the Mathieu groups combinatorially: the quadratic-residue codes of length 11 over A 3 A_3 A 3 and of length 23 over A 2 A_2 A 2 , extended by a parity digit, are the ternary and binary Golay codes, with automorphism groups 2. M 12 2.M_{12} 2. M 12 and M 24 M_{24} M 24 .
Direct sums
Definition 7.20. The direct sum C p ℓ 0 ⊕ ⋯ ⊕ C p ℓ − 1 C_{p^{\ell_0}}\oplus\cdots\oplus C_{p^{\ell-1}} C p ℓ 0 ⊕ ⋯ ⊕ C p ℓ − 1 is the direct product ((19) ) of cyclic groups:
m ⊕ n = ∑ h = 0 ℓ − 1 d p h ( 0 , D h ( m ) + D h ( n ) ) Π h . (57) m\oplus n=\sum_{h=0}^{\ell-1}d_{p^h}\left(0,D_h(m)+D_h(n)\right)\Pi_h. \tag*{(57)} m ⊕ n = h = 0 ∑ ℓ − 1 d p h ( 0 , D h ( m ) + D h ( n ) ) Π h . (57)
Theorem 7.21 (Digital Chinese remainder theorem). The map φ ( x ) = d r ( 0 , x ) + r d s ( 0 , x ) \varphi(x)=d_r(0,x)+r\,d_s(0,x) φ ( x ) = d r ( 0 , x ) + r d s ( 0 , x ) is a homomorphism C r s → C r ⊕ C s C_{rs}\to C_r\oplus C_s C r s → C r ⊕ C s . It is an isomorphism if and only if gcd ( r , s ) = 1 \gcd(r,s)=1 g cd( r , s ) = 1 , and otherwise C r s ≇ C r ⊕ C s C_{rs}\not\cong C_r\oplus C_s C r s ≅ C r ⊕ C s .
Proof. By Lemma 1.3 both digits of φ \varphi φ are additive, since r r r and s s s divide r s rs r s . The kernel consists of the x ∈ A r s x\in\mathcal{A}_{rs} x ∈ A r s divisible by lcm ( r , s ) \operatorname{lcm}(r,s) lcm ( r , s ) , so it is trivial exactly when lcm ( r , s ) = r s \operatorname{lcm}(r,s)=rs lcm ( r , s ) = r s . If gcd ( r , s ) > 1 \gcd(r,s)>1 g cd( r , s ) > 1 , every element of C r ⊕ C s C_r\oplus C_s C r ⊕ C s has order dividing lcm ( r , s ) < r s \operatorname{lcm}(r,s)<rs lcm ( r , s ) < r s , while 1 ∈ C r s 1\in C_{rs} 1 ∈ C r s has order r s rs r s . □ \square □
Coprime factors and explicit isomorphisms
Proposition 7.22. For u ∈ Z u\in\mathbb{Z} u ∈ Z , the map x ↦ d N ( 0 , u x ) x\mapsto d_N(0,ux) x ↦ d N ( 0 , ux ) is a permutation of A N \mathcal{A}_N A N if and only if gcd ( u , N ) = 1 \gcd(u,N)=1 g cd( u , N ) = 1 . If p p p is prime, multiplication by p − 1 p-1 p − 1 is additive inversion: d p ( 0 , ( p − 1 ) x ) = d p ( 0 , − x ) d_p(0,(p-1)x)=d_p(0,-x) d p ( 0 , ( p − 1 ) x ) = d p ( 0 , − x ) .
Proof. The map has a two-sided inverse x ↦ d N ( 0 , v x ) x\mapsto d_N(0,vx) x ↦ d N ( 0 , v x ) exactly when u v ≡ 1 ( m o d N ) uv\equiv1\pmod N uv ≡ 1 ( mod N ) , which is equivalent to gcd ( u , N ) = 1 \gcd(u,N)=1 g cd( u , N ) = 1 . The second assertion is immediate from p − 1 ≡ − 1 ( m o d p ) p-1\equiv-1\pmod p p − 1 ≡ − 1 ( mod p ) . □ \square □
Theorem 7.23 (Coprime cyclic factors). If r r r and s s s are coprime, then C r ⊕ C s ≅ C r s C_r\oplus C_s\cong C_{rs} C r ⊕ C s ≅ C r s . More precisely, for the direct-sum law on A r s \mathcal{A}_{rs} A r s with radices r , s r,s r , s , the element g = 1 + r g=1+r g = 1 + r has order r s rs r s , and
Ψ ( d r s ( 0 , k ) ) = g ⊕ k (58) \Psi\left(d_{rs}(0,k)\right)=g^{\oplus k} \tag*{(58)} Ψ ( d r s ( 0 , k ) ) = g ⊕ k (58) is an isomorphism C r s → C r ⊕ C s C_{rs}\to C_r\oplus C_s C r s → C r ⊕ C s .
Proof. The two mixed-radix digits of g ⊕ k g^{\oplus k} g ⊕ k are d r ( 0 , k ) d_r(0,k) d r ( 0 , k ) and d s ( 0 , k ) d_s(0,k) d s ( 0 , k ) : its lower digit starts at 1 and its upper digit starts at 1. Thus g ⊕ k = 0 g^{\oplus k}=0 g ⊕ k = 0 if and only if r ∣ k r\mid k r ∣ k and s ∣ k s\mid k s ∣ k , hence if and only if r s ∣ k rs\mid k r s ∣ k . So g g g has order r s rs r s , and the displayed map is a bijective homomorphism. This gives a constructive proof of Theorem 7.21. □ \square □
Example 7.24 (The two laws of order six). The laws of C 6 C_6 C 6 and C 3 ⊕ C 2 C_3\oplus C_2 C 3 ⊕ C 2 on A 6 \mathcal{A}_6 A 6 are respectively
x + 6 y = d 3 ( 0 , x + y ) + 3 d 2 ( 0 , ⌊ x / 3 ⌋ + ⌊ y / 3 ⌋ + d 3 ( 0 , x ) + d 3 ( 0 , y ) 3 ) , x+_6y=d_3(0,x+y)+3d_2\left(0,\left\lfloor x/3\right\rfloor+\left\lfloor y/3\right\rfloor+\frac{d_3(0,x)+d_3(0,y)}{3}\right), x + 6 y = d 3 ( 0 , x + y ) + 3 d 2 ( 0 , ⌊ x /3 ⌋ + ⌊ y /3 ⌋ + 3 d 3 ( 0 , x ) + d 3 ( 0 , y ) ) , x ⊕ y = d 3 ( 0 , x + y ) + 3 d 2 ( 0 , ⌊ x / 3 ⌋ + ⌊ y / 3 ⌋ ) . x\oplus y=d_3(0,x+y)+3d_2\left(0,\left\lfloor x/3\right\rfloor+\left\lfloor y/3\right\rfloor\right). x ⊕ y = d 3 ( 0 , x + y ) + 3 d 2 ( 0 , ⌊ x /3 ⌋ + ⌊ y /3 ⌋ ) . The first has a carry and the second does not. Nevertheless 4 = 1 + 3 4=1+3 4 = 1 + 3 has order 6 in the second law, and the explicit isomorphism is k ↦ 4 ⊕ k k\mapsto4^{\oplus k} k ↦ 4 ⊕ k ; the elements occur in the order 0 , 4 , 2 , 3 , 1 , 5 0,4,2,3,1,5 0 , 4 , 2 , 3 , 1 , 5 .
Proposition 7.25 (Prime-power reduction). If m 1 , … , m t m_1,\ldots,m_t m 1 , … , m t are pairwise coprime, then C m 1 ⊕ ⋯ ⊕ C m t ≅ C m 1 ⋯ m t C_{m_1}\oplus\cdots\oplus C_{m_t}\cong C_{m_1\cdots m_t} C m 1 ⊕ ⋯ ⊕ C m t ≅ C m 1 ⋯ m t . Consequently, every direct sum of cyclic groups is isomorphic to a direct sum in which each summand has prime-power order.
Proof. Apply Theorem 7.23 repeatedly. To obtain the final statement, factor each cyclic order into its pairwise-coprime prime-power factors and split it in the reverse direction. □ \square □
Theorem 7.26 (Fundamental theorem, digital form). Every finite Abelian group is isomorphic to a unique direct sum (7.3) with radices p 0 ∣ p 1 ∣ ⋯ ∣ p N − 1 p_0\mid p_1\mid\cdots\mid p_{N-1} p 0 ∣ p 1 ∣ ⋯ ∣ p N − 1 and p 0 ≥ 2 p_0\ge2 p 0 ≥ 2 (invariant factors); equivalently, by Theorem 7.21, to a direct sum of cyclic groups of prime-power order. The number of Abelian groups of order ∏ i ℓ i e i \prod_i\ell_i^{e_i} ∏ i ℓ i e i is ∏ i π ( e i ) \prod_i\pi(e_i) ∏ i π ( e i ) , where π \pi π is the partition function.
Proof. Let G = ( A n , ∗ ) G=(\mathcal{A}_n,*) G = ( A n , ∗ ) be Abelian of order n = ∏ i ℓ i e i n=\prod_i\ell_i^{e_i} n = ∏ i ℓ i e i ; powers x ∗ k x^{*k} x ∗ k with k < 0 k<0 k < 0 denote powers of the inverse.
Step 0: x ∗ n = 0 x^{*n}=0 x ∗ n = 0 . The row code L x L_x L x is a permutation of A n \mathcal{A}_n A n , so { x ∗ g : g ∈ A n } = A n \{x*g:g\in\mathcal{A}_n\}=\mathcal{A}_n { x ∗ g : g ∈ A n } = A n . Multiplying all elements in both forms and using commutativity gives x ∗ n ∗ ∏ g g = ∏ g g x^{*n}*\prod_g g=\prod_g g x ∗ n ∗ ∏ g g = ∏ g g , hence x ∗ n = 0 x^{*n}=0 x ∗ n = 0 .
Step 1: primary decomposition. For each prime ℓ = ℓ i \ell=\ell_i ℓ = ℓ i let G ℓ = { x : x ℓ e i = 0 } G_\ell=\{x:x^{\ell^{e_i}}=0\} G ℓ = { x : x ℓ e i = 0 } , a subgroup because G G G is Abelian, and let m ℓ = n / ℓ e i m_\ell=n/\ell^{e_i} m ℓ = n / ℓ e i . The m ℓ m_\ell m ℓ have no common prime factor, so there are integers u ℓ u_\ell u ℓ with ∑ ℓ u ℓ m ℓ = 1 \sum_\ell u_\ell m_\ell=1 ∑ ℓ u ℓ m ℓ = 1 . By Step 0, x ∗ u ℓ m ℓ ∈ G ℓ x^{*u_\ell m_\ell}\in G_\ell x ∗ u ℓ m ℓ ∈ G ℓ and x = ∏ ℓ x ∗ u ℓ m ℓ x=\prod_\ell x^{*u_\ell m_\ell} x = ∏ ℓ x ∗ u ℓ m ℓ , so the map ( x ℓ ) ℓ ↦ ∏ ℓ x ℓ (x_\ell)_\ell\mapsto\prod_\ell x_\ell ( x ℓ ) ℓ ↦ ∏ ℓ x ℓ from ⨁ ℓ G ℓ \bigoplus_\ell G_\ell ⨁ ℓ G ℓ to G G G is a surjective homomorphism. It is injective: if ∏ ℓ x ℓ = 0 \prod_\ell x_\ell=0 ∏ ℓ x ℓ = 0 , raising to the power m ℓ m_\ell m ℓ kills every factor except x ℓ x_\ell x ℓ , so x ℓ ∗ m ℓ = 0 x_\ell^{*m_\ell}=0 x ℓ ∗ m ℓ = 0 ; since also x ℓ ∗ ℓ e i = 0 x_\ell^{*\ell^{e_i}}=0 x ℓ ∗ ℓ e i = 0 and gcd ( m ℓ , ℓ e i ) = 1 \gcd(m_\ell,\ell^{e_i})=1 g cd( m ℓ , ℓ e i ) = 1 , we get x ℓ = 0 x_\ell=0 x ℓ = 0 . Thus G ≅ ⨁ ℓ G ℓ G\cong\bigoplus_\ell G_\ell G ≅ ⨁ ℓ G ℓ , and every element of G ℓ G_\ell G ℓ has order a power of ℓ \ell ℓ .
Step 2: an Abelian group A A A in which every element has ℓ \ell ℓ -power order is a direct sum of cyclic groups. Induct on ∣ A ∣ |A| ∣ A ∣ . Let a a a have maximal order ℓ k \ell^k ℓ k , so y ∗ ℓ k = 0 y^{*\ell^k}=0 y ∗ ℓ k = 0 for all y ∈ A y\in A y ∈ A , and let K K K be a subgroup maximal with K ∩ ⟨ a ⟩ = { 0 } K\cap\langle a\rangle=\{0\} K ∩ ⟨ a ⟩ = { 0 } . Suppose ⟨ a ⟩ K ≠ A \langle a\rangle K\ne A ⟨ a ⟩ K = A . Some y ∉ ⟨ a ⟩ K y\notin\langle a\rangle K y ∈ / ⟨ a ⟩ K has a power y ∗ ℓ j ∈ ⟨ a ⟩ K y^{*\ell^j}\in\langle a\rangle K y ∗ ℓ j ∈ ⟨ a ⟩ K ; taking j j j minimal and x = y ∗ ℓ j − 1 x=y^{*\ell^{j-1}} x = y ∗ ℓ j − 1 gives x ∉ ⟨ a ⟩ K x\notin\langle a\rangle K x ∈ / ⟨ a ⟩ K and x ∗ ℓ = a ∗ t ∗ c x^{*\ell}=a^{*t}*c x ∗ ℓ = a ∗ t ∗ c with c ∈ K c\in K c ∈ K . Then 0 = x ∗ t ℓ k − 1 = a ∗ t ℓ k − 1 ∗ c ∗ ℓ k − 1 0=x^{*t\ell^{k-1}}=a^{*t\ell^{k-1}}*c^{*\ell^{k-1}} 0 = x ∗ t ℓ k − 1 = a ∗ t ℓ k − 1 ∗ c ∗ ℓ k − 1 , so a ∗ t ℓ k − 1 ∈ K ∩ ⟨ a ⟩ = { 0 } a^{*t\ell^{k-1}}\in K\cap\langle a\rangle=\{0\} a ∗ t ℓ k − 1 ∈ K ∩ ⟨ a ⟩ = { 0 } , hence ℓ ∣ t \ell\mid t ℓ ∣ t , say t = ℓ s t=\ell s t = ℓ s . The element z = x ∗ a ∗ ( − s ) z=x*a^{*(-s)} z = x ∗ a ∗ ( − s ) satisfies z ∉ ⟨ a ⟩ K z\notin\langle a\rangle K z ∈ / ⟨ a ⟩ K and z ∗ ℓ = c ∈ K z^{*\ell}=c\in K z ∗ ℓ = c ∈ K . Let K ′ K' K ′ be generated by K K K and z z z . If a ∗ r = c ′ ∗ z ∗ j a^{*r}=c'*z^{*j} a ∗ r = c ′ ∗ z ∗ j with c ′ ∈ K c'\in K c ′ ∈ K , then either ℓ ∣ j \ell\mid j ℓ ∣ j , so z ∗ j ∈ K z^{*j}\in K z ∗ j ∈ K and a ∗ r ∈ K ∩ ⟨ a ⟩ = { 0 } a^{*r}\in K\cap\langle a\rangle=\{0\} a ∗ r ∈ K ∩ ⟨ a ⟩ = { 0 } , or gcd ( j , ℓ ) = 1 \gcd(j,\ell)=1 g cd( j , ℓ ) = 1 , and writing 1 = α j + β ℓ 1=\alpha j+\beta\ell 1 = α j + β ℓ gives z = ( z ∗ j ) α ∗ ( z ∗ ℓ ) β ∈ ⟨ a ⟩ K z=(z^{*j})^\alpha*(z^{*\ell})^\beta\in\langle a\rangle K z = ( z ∗ j ) α ∗ ( z ∗ ℓ ) β ∈ ⟨ a ⟩ K , a contradiction. So K ′ ∩ ⟨ a ⟩ = { 0 } K'\cap\langle a\rangle=\{0\} K ′ ∩ ⟨ a ⟩ = { 0 } with K ′ ⊋ K K'\supsetneq K K ′ ⊋ K , against maximality. Hence A = ⟨ a ⟩ K A=\langle a\rangle K A = ⟨ a ⟩ K with trivial intersection, so A ≅ ⟨ a ⟩ ⊕ K ≅ C ℓ k ⊕ K A\cong\langle a\rangle\oplus K\cong C_{\ell^k}\oplus K A ≅ ⟨ a ⟩ ⊕ K ≅ C ℓ k ⊕ K by Proposition 7.3 , and the induction hypothesis applies to K K K .
Digital form of Steps 1–2. Steps 1 and 2 give elements g h g_h g h of prime-power orders q h q_h q h , h < N h<N h < N , with ∏ h q h = n \prod_h q_h=n ∏ h q h = n , and the labelling
Ψ ( m ) = g 0 ∗ D 0 ( m ) ∗ g 1 ∗ D 1 ( m ) ∗ ⋯ ∗ g N − 1 ∗ D N − 1 ( m ) , m ∈ A n , \Psi(m)=g_0^{*D_0(m)}*g_1^{*D_1(m)}*\cdots*g_{N-1}^{*D_{N-1}(m)},\qquad m\in\mathcal{A}_n, Ψ ( m ) = g 0 ∗ D 0 ( m ) ∗ g 1 ∗ D 1 ( m ) ∗ ⋯ ∗ g N − 1 ∗ D N − 1 ( m ) , m ∈ A n ,
is an isomorphism from the direct-sum law (57) with radices q h q_h q h onto G G G : mixed-radix digits add modulo q h q_h q h under ⊕ \oplus ⊕ , and g h ∗ q h = 0 g_h^{*q_h}=0 g h ∗ q h = 0 . Thus G G G is a direct sum of cyclic groups of prime-power order.
Step 3: invariant factors. For each prime ℓ \ell ℓ list the exponents of the ℓ \ell ℓ -power radices in decreasing order, padding with zeros. Let p N ′ − 1 p_{N'-1} p N ′ − 1 be the product over ℓ \ell ℓ of the largest ℓ \ell ℓ -powers, p N ′ − 2 p_{N'-2} p N ′ − 2 the product of the second largest, and so on, discarding factors equal to 1 1 1 . By Proposition 7.25 each p h p_h p h is the order of a cyclic summand obtained by combining coprime prime-power summands, and by construction p 0 ∣ p 1 ∣ ⋯ ∣ p N ′ − 1 p_0\mid p_1\mid\cdots\mid p_{N'-1} p 0 ∣ p 1 ∣ ⋯ ∣ p N ′ − 1 with p 0 ≥ 2 p_0\ge2 p 0 ≥ 2 .
Step 4: uniqueness. For a prime ℓ \ell ℓ and j ≥ 0 j\ge0 j ≥ 0 let
c ℓ ( j ) = ∣ { x : x ∗ ℓ j = 0 } ∣ = ∑ x ∈ A n d n ( x ∗ ℓ j , 1 ) , c_\ell(j)=|\{x:x^{*\ell^j}=0\}|=\sum_{x\in\mathcal{A}_n}d_n(x^{*\ell^j},1), c ℓ ( j ) = ∣ { x : x ∗ ℓ j = 0 } ∣ = x ∈ A n ∑ d n ( x ∗ ℓ j , 1 ) ,
an isomorphism invariant, computed with the digital Kronecker delta d n ( u , 1 ) = δ u , 0 d_n(u,1)=\delta_{u,0} d n ( u , 1 ) = δ u , 0 for u ∈ A n u\in\mathcal{A}_n u ∈ A n (equivalently B ( u ) \mathcal{B}(u) B ( u ) , Proposition 1.16 ). In C ℓ k C_{\ell^k} C ℓ k the equation ℓ j x = 0 \ell^j x=0 ℓ j x = 0 has ℓ min ( j , k ) \ell^{\min(j,k)} ℓ m i n ( j , k ) solutions, by Theorem 7.6 ; in C q C_q C q with ℓ ∤ q \ell\nmid q ℓ ∤ q it has one solution; and solution counts multiply over direct summands. Hence, if the ℓ \ell ℓ -power summands have exponents k 1 , … , k r k_1,\ldots,k_r k 1 , … , k r ,
c ℓ ( j ) = ℓ ∑ i min ( j , k i ) , c ℓ ( j + 1 ) c ℓ ( j ) = ℓ ∣ { i : k i > j } ∣ , c_\ell(j)=\ell^{\sum_i\min(j,k_i)},\qquad\frac{c_\ell(j+1)}{c_\ell(j)}=\ell^{|\{i:k_i>j\}|}, c ℓ ( j ) = ℓ ∑ i m i n ( j , k i ) , c ℓ ( j ) c ℓ ( j + 1 ) = ℓ ∣ { i : k i > j } ∣ ,
so the multiset { k i } \{k_i\} { k i } is determined by G G G . This proves uniqueness of the prime-power decomposition. In an invariant-factor decomposition the exponent of ℓ \ell ℓ in p h p_h p h is nondecreasing in h h h , so splitting each p h p_h p h into prime powers recovers the sorted exponent lists, and conversely those lists determine the p h p_h p h as in Step 3. Hence the invariant factors are unique too.
Step 5: counting. By Steps 1 and 4, the isomorphism class of G G G is determined by, and determines, for each i i i a multiset of positive exponents with sum e i e_i e i , since ∣ G ℓ i ∣ = ℓ i e i |G_{\ell_i}|=\ell_i^{e_i} ∣ G ℓ i ∣ = ℓ i e i . Every such choice occurs. These multisets are the partitions of e i e_i e i , which gives ∏ i π ( e i ) \prod_i\pi(e_i) ∏ i π ( e i ) classes. □ \square □
Example 7.27 (Order 1176). Since
1176 = 2 3 ⋅ 3 ⋅ 7 2 , π ( 3 ) π ( 1 ) π ( 2 ) = 3 ⋅ 1 ⋅ 2 = 6 , 1176=2^3\cdot3\cdot7^2,\qquad\pi(3)\pi(1)\pi(2)=3\cdot1\cdot2=6, 1176 = 2 3 ⋅ 3 ⋅ 7 2 , π ( 3 ) π ( 1 ) π ( 2 ) = 3 ⋅ 1 ⋅ 2 = 6 ,
@page 86/200 starts_mid=0 ends_mid=1 Step 1: primary decomposition. For each prime ℓ = ℓ i \ell=\ell_i ℓ = ℓ i let G ℓ = { x : x ℓ e i = 0 } G_\ell=\{x:x^{\ell^{e_i}}=0\} G ℓ = { x : x ℓ e i = 0 } , a subgroup because G G G is Abelian, and let m ℓ = n / ℓ e i m_\ell=n/\ell^{e_i} m ℓ = n / ℓ e i . The m ℓ m_\ell m ℓ have no common prime factor, so there are integers u ℓ u_\ell u ℓ with ∑ ℓ u ℓ m ℓ = 1 \sum_\ell u_\ell m_\ell=1 ∑ ℓ u ℓ m ℓ = 1 . By Step 0, x ∗ u ℓ m ℓ ∈ G ℓ x^{*u_\ell m_\ell}\in G_\ell x ∗ u ℓ m ℓ ∈ G ℓ and x = ∏ ℓ x ∗ u ℓ m ℓ x=\prod_\ell x^{*u_\ell m_\ell} x = ∏ ℓ x ∗ u ℓ m ℓ , so the map ( x ℓ ) ℓ ↦ ∏ ℓ x ℓ (x_\ell)_\ell\mapsto\prod_\ell x_\ell ( x ℓ ) ℓ ↦ ∏ ℓ x ℓ from ⨁ ℓ G ℓ \bigoplus_\ell G_\ell ⨁ ℓ G ℓ to G G G is a surjective homomorphism. It is injective: if ∏ ℓ x ℓ = 0 \prod_\ell x_\ell=0 ∏ ℓ x ℓ = 0 , raising to the power m ℓ m_\ell m ℓ kills every factor except x ℓ x_\ell x ℓ , so x ℓ ∗ m ℓ = 0 x_\ell^{*m_\ell}=0 x ℓ ∗ m ℓ = 0 ; since also x ℓ ∗ ℓ e i = 0 x_\ell^{*\ell^{e_i}}=0 x ℓ ∗ ℓ e i = 0 and gcd ( m ℓ , ℓ e i ) = 1 \gcd(m_\ell,\ell^{e_i})=1 g cd( m ℓ , ℓ e i ) = 1 , we get x ℓ = 0 x_\ell=0 x ℓ = 0 . Thus G ≅ ⨁ ℓ G ℓ G\cong\bigoplus_\ell G_\ell G ≅ ⨁ ℓ G ℓ , and every element of G ℓ G_\ell G ℓ has order a power of ℓ \ell ℓ .
Step 2: an Abelian group A A A in which every element has ℓ \ell ℓ -power order is a direct sum of cyclic groups. Induct on ∣ A ∣ |A| ∣ A ∣ . Let a a a have maximal order ℓ k \ell^k ℓ k , so y ∗ ℓ k = 0 y^{*\ell^k}=0 y ∗ ℓ k = 0 for all y ∈ A y\in A y ∈ A , and let K K K be a subgroup maximal with K ∩ ⟨ a ⟩ = { 0 } K\cap\langle a\rangle=\{0\} K ∩ ⟨ a ⟩ = { 0 } . Suppose ⟨ a ⟩ K ≠ A \langle a\rangle K\ne A ⟨ a ⟩ K = A . Some y ∉ ⟨ a ⟩ K y\notin\langle a\rangle K y ∈ / ⟨ a ⟩ K has a power y ∗ ℓ j ∈ ⟨ a ⟩ K y^{*\ell^j}\in\langle a\rangle K y ∗ ℓ j ∈ ⟨ a ⟩ K ; taking j j j minimal and x = y ∗ ℓ j − 1 x=y^{*\ell^{j-1}} x = y ∗ ℓ j − 1 gives x ∉ ⟨ a ⟩ K x\notin\langle a\rangle K x ∈ / ⟨ a ⟩ K and x ∗ ℓ = a ∗ t ∗ c x^{*\ell}=a^{*t}*c x ∗ ℓ = a ∗ t ∗ c with c ∈ K c\in K c ∈ K . Then 0 = x ∗ t ℓ k − 1 = a ∗ t ℓ k − 1 ∗ c ∗ ℓ k − 1 0=x^{*t\ell^{k-1}}=a^{*t\ell^{k-1}}*c^{*\ell^{k-1}} 0 = x ∗ t ℓ k − 1 = a ∗ t ℓ k − 1 ∗ c ∗ ℓ k − 1 , so a ∗ t ℓ k − 1 ∈ K ∩ ⟨ a ⟩ = { 0 } a^{*t\ell^{k-1}}\in K\cap\langle a\rangle=\{0\} a ∗ t ℓ k − 1 ∈ K ∩ ⟨ a ⟩ = { 0 } , hence ℓ ∣ t \ell\mid t ℓ ∣ t , say t = ℓ s t=\ell s t = ℓ s . The element z = x ∗ a ∗ ( − s ) z=x*a^{*(-s)} z = x ∗ a ∗ ( − s ) satisfies z ∉ ⟨ a ⟩ K z\notin\langle a\rangle K z ∈ / ⟨ a ⟩ K and z ∗ ℓ = c ∈ K z^{*\ell}=c\in K z ∗ ℓ = c ∈ K . Let K ′ K' K ′ be generated by K K K and z z z . If a ∗ r = c ′ ∗ z ∗ j a^{*r}=c'*z^{*j} a ∗ r = c ′ ∗ z ∗ j with c ′ ∈ K c'\in K c ′ ∈ K , then either ℓ ∣ j \ell\mid j ℓ ∣ j , so z ∗ j ∈ K z^{*j}\in K z ∗ j ∈ K and a ∗ r ∈ K ∩ ⟨ a ⟩ = { 0 } a^{*r}\in K\cap\langle a\rangle=\{0\} a ∗ r ∈ K ∩ ⟨ a ⟩ = { 0 } , or gcd ( j , ℓ ) = 1 \gcd(j,\ell)=1 g cd( j , ℓ ) = 1 , and writing 1 = α j + β ℓ 1=\alpha j+\beta\ell 1 = α j + β ℓ gives z = ( z ∗ j ) α ∗ ( z ∗ ℓ ) β ∈ ⟨ a ⟩ K z=(z^{*j})^\alpha*(z^{*\ell})^\beta\in\langle a\rangle K z = ( z ∗ j ) α ∗ ( z ∗ ℓ ) β ∈ ⟨ a ⟩ K , a contradiction. So K ′ ∩ ⟨ a ⟩ = { 0 } K'\cap\langle a\rangle=\{0\} K ′ ∩ ⟨ a ⟩ = { 0 } with K ′ ⊋ K K'\supsetneq K K ′ ⊋ K , against maximality. Hence A = ⟨ a ⟩ K A=\langle a\rangle K A = ⟨ a ⟩ K with trivial intersection, so A ≅ ⟨ a ⟩ ⊕ K ≅ C ℓ k ⊕ K A\cong\langle a\rangle\oplus K\cong C_{\ell^k}\oplus K A ≅ ⟨ a ⟩ ⊕ K ≅ C ℓ k ⊕ K by Proposition 7.3 , and the induction hypothesis applies to K K K .
Digital form of Steps 1–2. Steps 1 and 2 give elements g h g_h g h of prime-power orders q h q_h q h , h < N h<N h < N , with ∏ h q h = n \prod_h q_h=n ∏ h q h = n , and the labelling
Ψ ( m ) = g 0 ∗ D 0 ( m ) ∗ g 1 ∗ D 1 ( m ) ∗ ⋯ ∗ g N − 1 ∗ D N − 1 ( m ) , m ∈ A n , \Psi(m)=g_0^{*D_0(m)}*g_1^{*D_1(m)}*\cdots*g_{N-1}^{*D_{N-1}(m)},\qquad m\in\mathcal{A}_n, Ψ ( m ) = g 0 ∗ D 0 ( m ) ∗ g 1 ∗ D 1 ( m ) ∗ ⋯ ∗ g N − 1 ∗ D N − 1 ( m ) , m ∈ A n ,
is an isomorphism from the direct-sum law (57) with radices q h q_h q h onto G G G : mixed-radix digits add modulo q h q_h q h under ⊕ \oplus ⊕ , and g h ∗ q h = 0 g_h^{*q_h}=0 g h ∗ q h = 0 . Thus G G G is a direct sum of cyclic groups of prime-power order.
Step 3: invariant factors. For each prime ℓ \ell ℓ list the exponents of the ℓ \ell ℓ -power radices in decreasing order, padding with zeros. Let p N ′ − 1 p_{N'-1} p N ′ − 1 be the product over ℓ \ell ℓ of the largest ℓ \ell ℓ -powers, p N ′ − 2 p_{N'-2} p N ′ − 2 the product of the second largest, and so on, discarding factors equal to 1 1 1 . By Proposition 7.25 each p h p_h p h is the order of a cyclic summand obtained by combining coprime prime-power summands, and by construction p 0 ∣ p 1 ∣ ⋯ ∣ p N ′ − 1 p_0\mid p_1\mid\cdots\mid p_{N'-1} p 0 ∣ p 1 ∣ ⋯ ∣ p N ′ − 1 with p 0 ≥ 2 p_0\ge2 p 0 ≥ 2 .
Step 4: uniqueness. For a prime ℓ \ell ℓ and j ≥ 0 j\ge0 j ≥ 0 let
c ℓ ( j ) = ∣ { x : x ∗ ℓ j = 0 } ∣ = ∑ x ∈ A n d n ( x ∗ ℓ j , 1 ) , c_\ell(j)=|\{x:x^{*\ell^j}=0\}|=\sum_{x\in\mathcal{A}_n}d_n(x^{*\ell^j},1), c ℓ ( j ) = ∣ { x : x ∗ ℓ j = 0 } ∣ = x ∈ A n ∑ d n ( x ∗ ℓ j , 1 ) ,
an isomorphism invariant, computed with the digital Kronecker delta d n ( u , 1 ) = δ u , 0 d_n(u,1)=\delta_{u,0} d n ( u , 1 ) = δ u , 0 for u ∈ A n u\in\mathcal{A}_n u ∈ A n (equivalently B ( u ) \mathcal{B}(u) B ( u ) , Proposition 1.16 ). In C ℓ k C_{\ell^k} C ℓ k the equation ℓ j x = 0 \ell^j x=0 ℓ j x = 0 has ℓ min ( j , k ) \ell^{\min(j,k)} ℓ m i n ( j , k ) solutions, by Theorem 7.6 ; in C q C_q C q with ℓ ∤ q \ell\nmid q ℓ ∤ q it has one solution; and solution counts multiply over direct summands. Hence, if the ℓ \ell ℓ -power summands have exponents k 1 , … , k r k_1,\ldots,k_r k 1 , … , k r ,
c ℓ ( j ) = ℓ ∑ i min ( j , k i ) , c ℓ ( j + 1 ) c ℓ ( j ) = ℓ ∣ { i : k i > j } ∣ , c_\ell(j)=\ell^{\sum_i\min(j,k_i)},\qquad\frac{c_\ell(j+1)}{c_\ell(j)}=\ell^{|\{i:k_i>j\}|}, c ℓ ( j ) = ℓ ∑ i m i n ( j , k i ) , c ℓ ( j ) c ℓ ( j + 1 ) = ℓ ∣ { i : k i > j } ∣ ,
so the multiset { k i } \{k_i\} { k i } is determined by G G G . This proves uniqueness of the prime-power decomposition. In an invariant-factor decomposition the exponent of ℓ \ell ℓ in p h p_h p h is nondecreasing in h h h , so splitting each p h p_h p h into prime powers recovers the sorted exponent lists, and conversely those lists determine the p h p_h p h as in Step 3. Hence the invariant factors are unique too.
Step 5: counting. By Steps 1 and 4, the isomorphism class of G G G is determined by, and determines, for each i i i a multiset of positive exponents with sum e i e_i e i , since ∣ G ℓ i ∣ = ℓ i e i |G_{\ell_i}|=\ell_i^{e_i} ∣ G ℓ i ∣ = ℓ i e i . Every such choice occurs. These multisets are the partitions of e i e_i e i , which gives ∏ i π ( e i ) \prod_i\pi(e_i) ∏ i π ( e i ) classes. □ \square □
Example 7.27 (Order 1176). Since
1176 = 2 3 ⋅ 3 ⋅ 7 2 , π ( 3 ) π ( 1 ) π ( 2 ) = 3 ⋅ 1 ⋅ 2 = 6 , 1176=2^3\cdot3\cdot7^2,\qquad\pi(3)\pi(1)\pi(2)=3\cdot1\cdot2=6, 1176 = 2 3 ⋅ 3 ⋅ 7 2 , π ( 3 ) π ( 1 ) π ( 2 ) = 3 ⋅ 1 ⋅ 2 = 6 ,
the fundamental theorem gives six Abelian isomorphism classes. A prime-power list is
C 8 ⊕ C 3 ⊕ C 49 , C 4 ⊕ C 2 ⊕ C 3 ⊕ C 49 , C 2 3 ⊕ C 3 ⊕ C 49 , C_{8} \oplus C_{3} \oplus C_{49},\quad C_{4} \oplus C_{2} \oplus C_{3} \oplus C_{49},\quad C_{2}^{3} \oplus C_{3} \oplus C_{49}, C 8 ⊕ C 3 ⊕ C 49 , C 4 ⊕ C 2 ⊕ C 3 ⊕ C 49 , C 2 3 ⊕ C 3 ⊕ C 49 , C 8 ⊕ C 3 ⊕ C 7 2 , C 4 ⊕ C 2 ⊕ C 3 ⊕ C 7 2 , C 2 3 ⊕ C 3 ⊕ C 7 2 . C_{8} \oplus C_{3} \oplus C_{7}^{2},\quad C_{4} \oplus C_{2} \oplus C_{3} \oplus C_{7}^{2},\quad C_{2}^{3} \oplus C_{3} \oplus C_{7}^{2}. C 8 ⊕ C 3 ⊕ C 7 2 , C 4 ⊕ C 2 ⊕ C 3 ⊕ C 7 2 , C 2 3 ⊕ C 3 ⊕ C 7 2 .
By Proposition 7.25, factors of coprime order may be combined; for instance C 1176 ≅ C 49 ⊕ C 24 ≅ C 8 ⊕ C 3 ⊕ C 49 C_{1176} \cong C_{49} \oplus C_{24} \cong C_{8} \oplus C_{3} \oplus C_{49} C 1176 ≅ C 49 ⊕ C 24 ≅ C 8 ⊕ C 3 ⊕ C 49 .
Example 7.28. (Abelian groups of order at most 32). Every order not listed below has only the cyclic group.
Table 7.28.
(Here the summands are written with decreasing radices.)
Exercises
Show that the subgroups of C n C_n C n are the sets { 0 , d , 2 d , … } \{0,d,2d,\ldots\} { 0 , d , 2 d , … } with d ∣ n d \mid n d ∣ n (compare Theorem 7.6 ), and that C n C_n C n has exactly φ ( d ) \varphi(d) φ ( d ) elements of order d d d .
Find a generator of the multiplicative group ( Z / p ) × (\mathbb{Z}/p)^{\times} ( Z / p ) × for p = 7 , 11 , 13 p=7,11,13 p = 7 , 11 , 13 , and write the corresponding discrete-logarithm relabelling of A p − 1 A_{p-1} A p − 1 .
Show that m = a + 3 b ↦ d 12 ( 0 , 4 a + 9 b ) m=a+3b \mapsto d_{12}(0,4a+9b) m = a + 3 b ↦ d 12 ( 0 , 4 a + 9 b ) (a ∈ A 3 a\in A_{3} a ∈ A 3 , b ∈ A 4 b\in A_{4} b ∈ A 4 ) is an isomorphism C 3 ⊕ C 4 → C 12 C_{3}\oplus C_{4}\to C_{12} C 3 ⊕ C 4 → C 12 , and compare with Theorem 7.21 .
Show that C m ⊕ C n ≅ C m n C_m\oplus C_n\cong C_{mn} C m ⊕ C n ≅ C mn if and only if gcd ( m , n ) = 1 \gcd(m,n)=1 g cd( m , n ) = 1 .
List the Abelian groups of order 72 up to isomorphism.
Show that ( Z / 2 k ) × ≅ C 2 ⊕ C 2 k − 2 (\mathbb{Z}/2^{k})^{\times}\cong C_{2}\oplus C_{2^{k-2}} ( Z / 2 k ) × ≅ C 2 ⊕ C 2 k − 2 for k ≥ 3 k\geq3 k ≥ 3 , generated by − 1 -1 − 1 and 5, and check the case k = 4 k=4 k = 4 explicitly.
Use Theorem 7.13 to show that ( − 1 / p ) = ( − 1 ) ( p − 1 ) / 2 \left(-1/p\right)=(-1)^{(p-1)/2} ( − 1/ p ) = ( − 1 ) ( p − 1 ) /2 : the permutation π p − 1 : x ↦ d p ( 0 , − x ) \pi_{p-1}:x\mapsto d_p(0,-x) π p − 1 : x ↦ d p ( 0 , − x ) is a product of ( p − 1 ) / 2 (p-1)/2 ( p − 1 ) /2 transpositions.
Apply Proposition 7.15 with a = 2 a=2 a = 2 to show that r = p − 1 2 − ⌊ p / 4 ⌋ r=\frac{p-1}{2}-\lfloor p/4\rfloor r = 2 p − 1 − ⌊ p /4 ⌋ and deduce ( 2 / p ) = ( − 1 ) ( p 2 − 1 ) / 8 \left(2/p\right)=(-1)^{(p^{2}-1)/8} ( 2/ p ) = ( − 1 ) ( p 2 − 1 ) /8 .
Let C p C_p C p act by rotation on the words of length p p p over A a A_a A a , coded by A a p A_{a^p} A a p . Use Burnside’s lemma to show that there are ( a p + ( p − 1 ) a ) / p (a^p+(p-1)a)/p ( a p + ( p − 1 ) a ) / p orbits, and deduce Fermat’s theorem d p ( 0 , a p ) = d p ( 0 , a ) d_p(0,a^p)=d_p(0,a) d p ( 0 , a p ) = d p ( 0 , a ) .
Show that the product of all elements of a finite abelian group is the identity unless the group has exactly one element of order 2 2 2 , in which case it is that element. Applying this to U p U_p U p , prove Wilson’s theorem d p ( 0 , ( p − 1 ) ! ) = p − 1 d_p(0,(p-1)!) = p-1 d p ( 0 , ( p − 1 )!) = p − 1 .
For distinct odd primes p , q p,q p , q , count the lattice points ( j , k ) (j,k) ( j , k ) with 1 ≤ j ≤ p − 1 2 1 \le j \le\frac{p-1}{2} 1 ≤ j ≤ 2 p − 1 , 1 ≤ k ≤ q − 1 2 1 \le k \le\frac{q-1}{2} 1 ≤ k ≤ 2 q − 1 on either side of the line q x = p y qx=py q x = p y to show ∑ j = 1 ( p − 1 ) / 2 ⌊ q j / p ⌋ + ∑ k = 1 ( q − 1 ) / 2 ⌊ p k / q ⌋ = p − 1 2 ⋅ q − 1 2 \sum_{j=1}^{(p-1)/2} \lfloor qj/p \rfloor+ \sum_{k=1}^{(q-1)/2} \lfloor pk/q \rfloor= \frac{p-1}{2} \cdot\frac{q-1}{2} ∑ j = 1 ( p − 1 ) /2 ⌊ q j / p ⌋ + ∑ k = 1 ( q − 1 ) /2 ⌊ p k / q ⌋ = 2 p − 1 ⋅ 2 q − 1 . Show that ( q / p ) = ( − 1 ) ∑ j = 1 ( p − 1 ) / 2 ⌊ q j / p ⌋ (q/p) = (-1)^{\sum_{j=1}^{(p-1)/2} \lfloor qj/p \rfloor} ( q / p ) = ( − 1 ) ∑ j = 1 ( p − 1 ) /2 ⌊ q j / p ⌋ (Eisenstein), and obtain a second proof of Theorem 7.17 .
Prove Kummer’s theorem: the exponent of p p p in ( a + b a ) \binom{a+b}{a} ( a a + b ) is the number of carries in the base-p p p addition of a a a and b b b . Compare the carries with the composition factors in Example 4.40.
Nonabelian groups
Semidirect products
Theorem 8.1. Let G 0 = ( A p 0 , ∗ 0 ) G_0=(\mathcal{A}_{p_0},\ast_0) G 0 = ( A p 0 , ∗ 0 ) and G 1 = ( A p 1 , ∗ 1 ) G_1=(\mathcal{A}_{p_1},\ast_1) G 1 = ( A p 1 , ∗ 1 ) be groups with identity 0 0 0 , and let c ↦ ϕ c c\mapsto\phi_c c ↦ ϕ c be a homomorphism G 1 → Aut ( G 0 ) G_1\to\operatorname{Aut}(G_0) G 1 → Aut ( G 0 ) . With m h = D h ( m ) m_h=D_h(m) m h = D h ( m ) (radices p 0 , p 1 p_0,p_1 p 0 , p 1 ), the law
m ∗ n = ( m 0 ∗ 0 ϕ m 1 ( n 0 ) ) + p 0 ( m 1 ∗ 1 n 1 ) (59) m\ast n=(m_0\ast_0\phi_{m_1}(n_0))+p_0(m_1\ast_1 n_1) \tag*{(59)} m ∗ n = ( m 0 ∗ 0 ϕ m 1 ( n 0 )) + p 0 ( m 1 ∗ 1 n 1 ) (59) makes A p 0 p 1 \mathcal{A}_{p_0p_1} A p 0 p 1 a group G 0 ⋊ ϕ G 1 G_0\rtimes_\phi G_1 G 0 ⋊ ϕ G 1 with identity 0 0 0 . It is the direct product if and only if ϕ \phi ϕ is trivial.
Proof. Both ( m ∗ n ) ∗ k (m\ast n)\ast k ( m ∗ n ) ∗ k and m ∗ ( n ∗ k ) m\ast(n\ast k) m ∗ ( n ∗ k ) have upper digit m 1 ∗ 1 n 1 ∗ 1 k 1 m_1\ast_1n_1\ast_1k_1 m 1 ∗ 1 n 1 ∗ 1 k 1 and lower digits m 0 ∗ 0 ϕ m 1 ( n 0 ) ∗ 0 ϕ m 1 ∗ 1 n 1 ( k 0 ) m_0\ast_0\phi_{m_1}(n_0)\ast_0\phi_{m_1\ast_1n_1}(k_0) m 0 ∗ 0 ϕ m 1 ( n 0 ) ∗ 0 ϕ m 1 ∗ 1 n 1 ( k 0 ) and m 0 ∗ 0 ϕ m 1 ( n 0 ∗ 0 ϕ n 1 ( k 0 ) ) m_0\ast_0\phi_{m_1}(n_0\ast_0\phi_{n_1}(k_0)) m 0 ∗ 0 ϕ m 1 ( n 0 ∗ 0 ϕ n 1 ( k 0 )) ; these agree because ϕ m 1 \phi_{m_1} ϕ m 1 is an automorphism and ϕ m 1 ∗ 1 n 1 = ϕ m 1 ∘ ϕ n 1 \phi_{m_1\ast_1n_1}=\phi_{m_1}\circ\phi_{n_1} ϕ m 1 ∗ 1 n 1 = ϕ m 1 ∘ ϕ n 1 . Since ϕ 0 = id \phi_0=\operatorname{id} ϕ 0 = id and ϕ c ( 0 ) = 0 \phi_c(0)=0 ϕ c ( 0 ) = 0 , 0 0 0 is an identity. The element with digits ( ϕ m 1 − 1 ( m 0 − 1 ) , m 1 − 1 ) (\phi_{m_1^{-1}}(m_0^{-1}),m_1^{-1}) ( ϕ m 1 − 1 ( m 0 − 1 ) , m 1 − 1 ) is a right inverse of m m m , and a finite monoid with right inverses is a group. □
Dicyclic groups
Theorem 8.2 (Dicyclic groups). Let q ≥ 1 q\ge1 q ≥ 1 . On A 4 q \mathcal{A}_{4q} A 4 q define, with ϵ ( m ) = d 2 ( 0 , m 2 q ) \epsilon(m)=d_2\left(0,\frac{m}{2q}\right) ϵ ( m ) = d 2 ( 0 , 2 q m ) ,
m ∗ n = d 2 q ( 0 , m + ( − 1 ) ϵ ( m ) n + q ϵ ( m ) ϵ ( n ) ) + 2 q d 2 ( 0 , ϵ ( m ) + ϵ ( n ) ) . (60) m\ast n=d_{2q}\left(0,m+(-1)^{\epsilon(m)}n+q\epsilon(m)\epsilon(n)\right)+2q\,d_2\left(0,\epsilon(m)+\epsilon(n)\right). \tag*{(60)} m ∗ n = d 2 q ( 0 , m + ( − 1 ) ϵ ( m ) n + q ϵ ( m ) ϵ ( n ) ) + 2 q d 2 ( 0 , ϵ ( m ) + ϵ ( n ) ) . (60) Then A 4 q \mathcal{A}_{4q} A 4 q is a group, denoted Q 4 q Q_{4q} Q 4 q . It is Abelian only for q = 1 q=1 q = 1 , and for q = 2 q=2 q = 2 it is the quaternion group Q 8 Q_8 Q 8 .
Proof. Write m = a + 2 q α m=a+2q\alpha m = a + 2 q α and n = b + 2 q β n=b+2q\beta n = b + 2 q β with a , b ∈ A 2 q a,b\in\mathcal{A}_{2q} a , b ∈ A 2 q and α , β ∈ A 2 \alpha,\beta\in\mathcal{A}_2 α , β ∈ A 2 . Then (60) becomes
( a , α ) ∗ ( b , β ) = ( a + ( − 1 ) α b + q α β m o d 2 q , α + β m o d 2 ) . (a,\alpha)\ast(b,\beta)=\left(a+(-1)^\alpha b+q\alpha\beta\mathbin{\bmod}2q,\ \alpha+\beta\mathbin{\bmod}2\right). ( a , α ) ∗ ( b , β ) = ( a + ( − 1 ) α b + q α β mod 2 q , α + β mod 2 ) .
This is the standard semidirect-type model of the dicyclic group with presentation generated by x = ( 1 , 0 ) x=(1,0) x = ( 1 , 0 ) and y = ( 0 , 1 ) y=(0,1) y = ( 0 , 1 ) , satisfying x 2 q = 0 x^{2q}=0 x 2 q = 0 , y 2 = x q y^2=x^q y 2 = x q , and y x y − 1 = x − 1 yxy^{-1}=x^{-1} y x y − 1 = x − 1 . Associativity follows by direct substitution from the law, or from that presentation; the identity is ( 0 , 0 ) (0,0) ( 0 , 0 ) and ( a , α ) − 1 = ( ( ( − 1 ) α ( q α − a ) m o d 2 q , α ) (a,\alpha)^{-1}=(((-1)^\alpha(q\alpha-a)\mathbin{\bmod}2q,\alpha) ( a , α ) − 1 = ((( − 1 ) α ( q α − a ) mod 2 q , α ) . If q = 1 q=1 q = 1 , the law reduces to C 2 ⊕ C 2 C_2\oplus C_2 C 2 ⊕ C 2 ; if q ≥ 2 q\ge2 q ≥ 2 , then ( 0 , 1 ) ∗ ( 1 , 0 ) = ( 2 q − 1 , 1 ) (0,1)\ast(1,0)=(2q-1,1) ( 0 , 1 ) ∗ ( 1 , 0 ) = ( 2 q − 1 , 1 ) while ( 1 , 0 ) ∗ ( 0 , 1 ) = ( 1 , 1 ) (1,0)\ast(0,1)=(1,1) ( 1 , 0 ) ∗ ( 0 , 1 ) = ( 1 , 1 ) . □
Example 8.3 (Quaternion group). For q = 2 q=2 q = 2 , Theorem 8.2 gives Q 8 Q_8 Q 8 . The element x = ( 1 , 0 ) x=(1,0) x = ( 1 , 0 ) has order 4 4 4 , the element y = ( 0 , 1 ) y=(0,1) y = ( 0 , 1 ) satisfies y 2 = x 2 y^2=x^2 y 2 = x 2 , and every noncentral element outside ⟨ x ⟩ \langle x\rangle ⟨ x ⟩ has order 4 4 4 .
Example 8.4. (The quaternion group Q 8 Q_8 Q 8 as a Cayley table.) For q = 2 q=2 q = 2 the law (60) on A 8 A_8 A 8 reads
m ∗ n = d 4 ( 0 , m + ( − 1 ) ϵ ( m ) n + 2 ϵ ( m ) ϵ ( n ) ) + 4 d 2 ( 0 , ϵ ( m ) + ϵ ( n ) ) , ϵ ( m ) = d 2 ( 0 , m 4 ) . m \ast n=d_4\left(0,m+(-1)^{\epsilon(m)}n+2\epsilon(m)\epsilon(n)\right)+4d_2\left(0,\epsilon(m)+\epsilon(n)\right), \qquad\epsilon(m)=d_2\left(0,\frac{m}{4}\right). m ∗ n = d 4 ( 0 , m + ( − 1 ) ϵ ( m ) n + 2 ϵ ( m ) ϵ ( n ) ) + 4 d 2 ( 0 , ϵ ( m ) + ϵ ( n ) ) , ϵ ( m ) = d 2 ( 0 , 4 m ) . With x = 1 x=1 x = 1 and y = 4 y=4 y = 4 every element is uniquely m = x m 0 y ϵ ( m ) m=x^{m_0}y^{\epsilon(m)} m = x m 0 y ϵ ( m ) , where m 0 = d 4 ( 0 , m ) m_0=d_4(0,m) m 0 = d 4 ( 0 , m ) , and y 2 = x 2 y^2=x^2 y 2 = x 2 , y x y − 1 = x − 1 yxy^{-1}=x^{-1} y x y − 1 = x − 1 . Identifying i = 1 i=1 i = 1 , j = 4 j=4 j = 4 , k = 5 k=5 k = 5 and − 1 = 2 -1=2 − 1 = 2 gives the labels
1 ↦ 0 , i ↦ 1 , − 1 ↦ 2 , − i ↦ 3 , j ↦ 4 , k ↦ 5 , − j ↦ 6 , − k ↦ 7 , 1\mapsto0,\quad i\mapsto1,\quad-1\mapsto2,\quad-i\mapsto3,\quad j\mapsto4,\quad k\mapsto5,\quad-j\mapsto6,\quad-k\mapsto7, 1 ↦ 0 , i ↦ 1 , − 1 ↦ 2 , − i ↦ 3 , j ↦ 4 , k ↦ 5 , − j ↦ 6 , − k ↦ 7 , and Table 8.1 is the multiplication table of the quaternion units. The element 2 2 2 is the only involution and generates the centre; the six elements 1 , 3 , 4 , 5 , 6 , 7 1,3,4,5,6,7 1 , 3 , 4 , 5 , 6 , 7 all have order 4 4 4 .
Table 8.1. Cayley table of Q 8 Q_8 Q 8 on A 8 A_8 A 8 given by the digital law (60) with q = 2 q=2 q = 2 ; the entry in row m m m , column n n n is m ∗ n m\ast n m ∗ n . The rules separate the cosets { 0 , … , 3 } \{0,\ldots,3\} { 0 , … , 3 } and { 4 , … , 7 } \{4,\ldots,7\} { 4 , … , 7 } of ⟨ x ⟩ \langle x\rangle ⟨ x ⟩ , i.e. the value of ϵ ( m ) \epsilon(m) ϵ ( m ) .
Example 8.5. (Computing an entry of Table 8.1 .) We compute 5 ∗ 6 5\ast6 5 ∗ 6 directly from the law, using only the digit function (1) .
Signs.
ϵ ( 5 ) = d 2 ( 0 , 5 4 ) = ⌊ 5 4 ⌋ − 2 ⌊ 5 8 ⌋ = 1 − 0 = 1 , ϵ ( 6 ) = d 2 ( 0 , 6 4 ) = ⌊ 6 4 ⌋ − 2 ⌊ 6 8 ⌋ = 1 − 0 = 1. \epsilon(5)=d_2\left(0,\frac{5}{4}\right)=\left\lfloor\frac{5}{4}\right\rfloor-2\left\lfloor\frac{5}{8}\right\rfloor=1-0=1,\qquad\epsilon(6)=d_2\left(0,\frac{6}{4}\right)=\left\lfloor\frac{6}{4}\right\rfloor-2\left\lfloor\frac{6}{8}\right\rfloor=1-0=1. ϵ ( 5 ) = d 2 ( 0 , 4 5 ) = ⌊ 4 5 ⌋ − 2 ⌊ 8 5 ⌋ = 1 − 0 = 1 , ϵ ( 6 ) = d 2 ( 0 , 4 6 ) = ⌊ 4 6 ⌋ − 2 ⌊ 8 6 ⌋ = 1 − 0 = 1. Lower digit of the product. Since ϵ ( 5 ) = 1 \epsilon(5)=1 ϵ ( 5 ) = 1 the sign is ( − 1 ) 1 = − 1 (-1)^1=-1 ( − 1 ) 1 = − 1 and the correction is 2 ⋅ 1 ⋅ 1 = 2 2\cdot1\cdot1=2 2 ⋅ 1 ⋅ 1 = 2 , so
d 4 ( 0 , 5 − 6 + 2 ) = d 4 ( 0 , 1 ) = 1. d_4(0,5-6+2)=d_4(0,1)=1. d 4 ( 0 , 5 − 6 + 2 ) = d 4 ( 0 , 1 ) = 1. Upper digit of the product.
d 2 ( 0 , ϵ ( 5 ) + ϵ ( 6 ) ) = d 2 ( 0 , 2 ) = 2 − 2 ⌊ 1 ⌋ = 0. d_2(0,\epsilon(5)+\epsilon(6))=d_2(0,2)=2-2\lfloor1\rfloor=0. d 2 ( 0 , ϵ ( 5 ) + ϵ ( 6 )) = d 2 ( 0 , 2 ) = 2 − 2 ⌊ 1 ⌋ = 0. Assembly. 5 ∗ 6 = 1 + 4 ⋅ 0 = 1 5\ast6=1+4\cdot0=1 5 ∗ 6 = 1 + 4 ⋅ 0 = 1 , which is the entry in row 5, column 6 of the table.
Check via the normal form. Since 5 = x y 5=xy 5 = x y , 6 = x 2 y 6=x^2y 6 = x 2 y and y x 2 y − 1 = x − 2 = x 2 yx^2y^{-1}=x^{-2}=x^2 y x 2 y − 1 = x − 2 = x 2 ,
5 ∗ 6 = x y x 2 y = x ( y x 2 y − 1 ) y 2 = x ⋅ x 2 ⋅ x 2 = x 5 = x = 1. 5\ast6=xy\,x^2y=x\left(yx^2y^{-1}\right)y^2=x\cdot x^2\cdot x^2=x^5=x=1. 5 ∗ 6 = x y x 2 y = x ( y x 2 y − 1 ) y 2 = x ⋅ x 2 ⋅ x 2 = x 5 = x = 1. In quaternion notation this is k ⋅ ( − j ) = − ( k j ) = i k\cdot(-j)=-(kj)=i k ⋅ ( − j ) = − ( k j ) = i . Non-commutativity. In the reverse order both signs are again 1 and
6 ∗ 5 = d 4 ( 0 , 6 − 5 + 2 ) + 4 d 2 ( 0 , 2 ) = d 4 ( 0 , 3 ) + 0 = 3 ≠ 1 , 6\ast5=d_4(0,6-5+2)+4d_2(0,2)=d_4(0,3)+0=3\ne1, 6 ∗ 5 = d 4 ( 0 , 6 − 5 + 2 ) + 4 d 2 ( 0 , 2 ) = d 4 ( 0 , 3 ) + 0 = 3 = 1 , that is, ( − j ) ⋅ k = − ( j k ) = − i (-j)\cdot k=-(jk)=-i ( − j ) ⋅ k = − ( j k ) = − i .
Example 8.6 (The dicyclic group Q 12 Q_{12} Q 12 ). Take q = 3 q=3 q = 3 in Theorem 8.2 . The law on A 12 A_{12} A 12 is
m ∗ n = d 6 ( 0 , m + ( − 1 ) ϵ ( m ) n + 3 ϵ ( m ) ϵ ( n ) ) + 6 d 2 ( 0 , ϵ ( m ) + ϵ ( n ) ) , ϵ ( m ) = d 2 ( 0 , m 6 ) . (61) m * n = d_{6}\left(0,m+(-1)^{\epsilon(m)}n+3\epsilon(m)\epsilon(n)\right)+6d_{2}\left(0,\epsilon(m)+\epsilon(n)\right), \qquad\epsilon(m)=d_{2}\left(0,\frac{m}{6}\right). \tag*{(61)} m ∗ n = d 6 ( 0 , m + ( − 1 ) ϵ ( m ) n + 3 ϵ ( m ) ϵ ( n ) ) + 6 d 2 ( 0 , ϵ ( m ) + ϵ ( n ) ) , ϵ ( m ) = d 2 ( 0 , 6 m ) . (61) With x = 1 x=1 x = 1 and y = 6 y=6 y = 6 every element is uniquely m = x m 0 y ϵ ( m ) m=x^{m_{0}}y^{\epsilon(m)} m = x m 0 y ϵ ( m ) , where m 0 = d 6 ( 0 , m ) m_{0}=d_{6}(0,m) m 0 = d 6 ( 0 , m ) , and x 6 = 0 x^{6}=0 x 6 = 0 , y 2 = x 3 y^{2}=x^{3} y 2 = x 3 , y x y − 1 = x − 1 yxy^{-1}=x^{-1} y x y − 1 = x − 1 . The resulting Cayley table is Table 8.2 . The elements 1 , 5 1,5 1 , 5 have order 6 6 6 , the elements 2 , 4 2,4 2 , 4 have order 3 3 3 , the element 3 3 3 is the unique involution and generates the centre, and the six elements 6 , … , 11 6,\ldots,11 6 , … , 11 outside ⟨ x ⟩ \langle x\rangle ⟨ x ⟩ all have order 4 4 4 . The group is not a semidirect product C 6 × C 2 C_{6}\times C_{2} C 6 × C 2 , because its only involution is central and lies in ⟨ x ⟩ \langle x\rangle ⟨ x ⟩ .
Table 8.2. Cayley table of Q 12 Q_{12} Q 12 on A 12 A_{12} A 12 given by the digital law (61) ; the entry in row m m m , column n n n is m ∗ n m*n m ∗ n . The rules separate the cosets { 0 , … , 5 } \{0,\ldots,5\} { 0 , … , 5 } and { 6 , … , 11 } \{6,\ldots,11\} { 6 , … , 11 } of ⟨ x ⟩ \langle x\rangle ⟨ x ⟩ , i.e. the value of ϵ ( m ) \epsilon(m) ϵ ( m ) .
Example 8.7 (Computing an entry of Table 8.2 ). We compute 8 ∗ 10 8*10 8 ∗ 10 directly from (61) , using only the digit function (1) .
Signs.
ϵ ( 8 ) = d 2 ( 0 , 8 6 ) = ⌊ 8 6 ⌋ − 2 ⌊ 8 12 ⌋ = 1 − 0 = 1 , \epsilon(8)=d_{2}\left(0,\frac{8}{6}\right)=\left\lfloor\frac{8}{6}\right\rfloor-2\left\lfloor\frac{8}{12}\right\rfloor=1-0=1, ϵ ( 8 ) = d 2 ( 0 , 6 8 ) = ⌊ 6 8 ⌋ − 2 ⌊ 12 8 ⌋ = 1 − 0 = 1 , ϵ ( 10 ) = d 2 ( 0 , 10 6 ) = ⌊ 10 6 ⌋ − 2 ⌊ 10 12 ⌋ = 1 − 0 = 1. \epsilon(10)=d_{2}\left(0,\frac{10}{6}\right)=\left\lfloor\frac{10}{6}\right\rfloor-2\left\lfloor\frac{10}{12}\right\rfloor=1-0=1. ϵ ( 10 ) = d 2 ( 0 , 6 10 ) = ⌊ 6 10 ⌋ − 2 ⌊ 12 10 ⌋ = 1 − 0 = 1. Lower digit of the product. The sign is ( − 1 ) 1 = − 1 (-1)^{1}=-1 ( − 1 ) 1 = − 1 and the correction is 3 ⋅ 1 ⋅ 1 = 3 3\cdot1\cdot1=3 3 ⋅ 1 ⋅ 1 = 3 , so
d 6 ( 0 , 8 − 10 + 3 ) = d 6 ( 0 , 1 ) = ⌊ 1 ⌋ − 6 ⌊ 1 6 ⌋ = 1. d_{6}(0,8-10+3)=d_{6}(0,1)=\lfloor1\rfloor-6\left\lfloor\frac{1}{6}\right\rfloor=1. d 6 ( 0 , 8 − 10 + 3 ) = d 6 ( 0 , 1 ) = ⌊ 1 ⌋ − 6 ⌊ 6 1 ⌋ = 1. By Lemma 1.3 one may reduce first: m 0 = 2 m_{0}=2 m 0 = 2 , n 0 = 4 n_{0}=4 n 0 = 4 , and d 6 ( 0 , 2 − 4 + 3 ) = 1 d_{6}(0,2-4+3)=1 d 6 ( 0 , 2 − 4 + 3 ) = 1 .
Upper digit of the product.
d 2 ( 0 , ϵ ( 8 ) + ϵ ( 10 ) ) = d 2 ( 0 , 2 ) = 2 − 2 ⌊ 1 ⌋ = 0. d_{2}(0,\epsilon(8)+\epsilon(10))=d_{2}(0,2)=2-2\lfloor1\rfloor=0. d 2 ( 0 , ϵ ( 8 ) + ϵ ( 10 )) = d 2 ( 0 , 2 ) = 2 − 2 ⌊ 1 ⌋ = 0. Assembly. 8 ∗ 10 = 1 + 6 ⋅ 0 = 1 8*10=1+6\cdot0=1 8 ∗ 10 = 1 + 6 ⋅ 0 = 1 , which is the entry in row 8, column 10 of the table.
Check via the normal form. Since 8 = x 2 y 8=x^{2}y 8 = x 2 y , 10 = x 4 y 10=x^{4}y 10 = x 4 y , y x 4 y − 1 = x − 4 yx^{4}y^{-1}=x^{-4} y x 4 y − 1 = x − 4 and y 2 = x 3 y^{2}=x^{3} y 2 = x 3 ,
8 ∗ 10 = x 2 y x 4 y = x 2 ( y x 4 y − 1 ) y 2 = x 2 − 4 + 3 = x = 1. 8*10=x^{2}yx^{4}y=x^{2}\left(yx^{4}y^{-1}\right)y^{2}=x^{2-4+3}=x=1. 8 ∗ 10 = x 2 y x 4 y = x 2 ( y x 4 y − 1 ) y 2 = x 2 − 4 + 3 = x = 1. Non-commutativity. In the reverse order both signs are again 1 and
10 ∗ 8 = d 6 ( 0 , 10 − 8 + 3 ) + 6 d 2 ( 0 , 2 ) = d 6 ( 0 , 5 ) + 0 = 5 ≠ 1. 10*8=d_{6}(0,10-8+3)+6d_{2}(0,2)=d_{6}(0,5)+0=5\ne1. 10 ∗ 8 = d 6 ( 0 , 10 − 8 + 3 ) + 6 d 2 ( 0 , 2 ) = d 6 ( 0 , 5 ) + 0 = 5 = 1.
Dihedral and metacyclic groups
Theorem 8.8 (Metacyclic groups). Let q , s ≥ 2 q,s \ge2 q , s ≥ 2 and let r r r satisfy gcd ( r , q ) = 1 \gcd(r,q)=1 g cd( r , q ) = 1 and r s ≡ 1 ( m o d q ) r^{s} \equiv1 \pmod q r s ≡ 1 ( mod q ) . With m 1 = d s ( 0 , m / q ) m_{1}=d_{s}(0,m/q) m 1 = d s ( 0 , m / q ) , the law
m ∗ n = d q ( 0 , m + r m 1 n ) + q d s ( 0 , m 1 + n 1 ) (62) m \ast n=d_{q}(0,m+r^{m_{1}}n)+q\,d_{s}(0,m_{1}+n_{1}) \tag*{(62)} m ∗ n = d q ( 0 , m + r m 1 n ) + q d s ( 0 , m 1 + n 1 ) (62) makes A q s A_{qs} A q s a group C q ⋊ r C s C_{q}\rtimes_{r} C_{s} C q ⋊ r C s . It is nonabelian if and only if r ≠ 1 ( m o d q ) r\ne1 \pmod q r = 1 ( mod q ) .
Proof. Take ϕ c ( x ) = d q ( 0 , r c x ) \phi_{c}(x)=d_{q}(0,r^{c}x) ϕ c ( x ) = d q ( 0 , r c x ) in Theorem 8.1 : it is an automorphism of C q C_{q} C q because r r r is a unit, and c ↦ ϕ c c\mapsto\phi_{c} c ↦ ϕ c is a homomorphism from C s C_{s} C s because r s ≡ 1 r^{s}\equiv1 r s ≡ 1 . By Lemma 1.3 , m ≡ m 0 m\equiv m_{0} m ≡ m 0 and n ≡ n 0 ( m o d q ) n\equiv n_{0}\pmod q n ≡ n 0 ( mod q ) , so (59) becomes (62) . Finally 1 ∗ q = 1 + q 1\ast q=1+q 1 ∗ q = 1 + q while q ∗ 1 = d q ( 0 , r ) + q q\ast1=d_{q}(0,r)+q q ∗ 1 = d q ( 0 , r ) + q .
Corollary 8.9 (Dihedral groups). For q ≥ 2 q\ge2 q ≥ 2 , r = q − 1 r=q-1 r = q − 1 , s = 2 s=2 s = 2 , the law
m ∗ n = d q ( 0 , m + n ( q − 1 ) d 2 ( 0 , m / q ) ) + q d 2 ( 0 , d 2 ( 0 , m q ) + d 2 ( 0 , n q ) ) (63) m\ast n=d_{q}\left(0,m+n(q-1)^{d_{2}(0,m/q)}\right)+q\,d_{2}\left(0,d_{2}\left(0,\frac{m}{q}\right)+d_{2}\left(0,\frac{n}{q}\right)\right) \tag*{(63)} m ∗ n = d q ( 0 , m + n ( q − 1 ) d 2 ( 0 , m / q ) ) + q d 2 ( 0 , d 2 ( 0 , q m ) + d 2 ( 0 , q n ) ) (63) defines the dihedral group D 2 q D_{2q} D 2 q of order 2 q 2q 2 q . It is nonabelian if and only if q ≥ 3 q\ge3 q ≥ 3 ; D 2 ≅ C 2 D_{2}\cong C_{2} D 2 ≅ C 2 and D 4 ≅ C 2 ⊕ C 2 D_{4}\cong C_{2}\oplus C_{2} D 4 ≅ C 2 ⊕ C 2 .
Remark 8.10. The dihedral law is exactly the semidirect product C q ⋊ C 2 C_{q}\rtimes C_{2} C q ⋊ C 2 in which the nontrivial element of C 2 C_{2} C 2 acts on C q C_{q} C q by inversion. Equivalently, writing m = a + q α m=a+q\alpha m = a + q α and n = b + q β n=b+q\beta n = b + q β with a , b ∈ A q a,b\in A_{q} a , b ∈ A q and α , β ∈ A 2 \alpha,\beta\in A_{2} α , β ∈ A 2 , one has ( a , α ) ∗ ( b , β ) = ( a + ( − 1 ) α b mod q , α + β mod 2 ) (a,\alpha)\ast(b,\beta)=(a+(-1)^{\alpha}b\mathbin{\operatorname{mod}}q,\alpha+\beta\mathbin{\operatorname{mod}}2) ( a , α ) ∗ ( b , β ) = ( a + ( − 1 ) α b mod q , α + β mod 2 ) . This makes transparent why D 6 ≅ Sym 3 D_{6}\cong\operatorname{Sym}_{3} D 6 ≅ Sym 3 and why the law is noncommutative as soon as q ≥ 3 q\ge3 q ≥ 3 .
Proposition 8.11 (Presentations and generators). In D 2 q D_{2q} D 2 q let r = ( 1 , 0 ) r=(1,0) r = ( 1 , 0 ) and s = ( 0 , 1 ) s=(0,1) s = ( 0 , 1 ) in the two-digit model of the preceding remark. Then
r q = s 2 = 1 , s r s − 1 = r − 1 , r^{q}=s^{2}=1,\qquad srs^{-1}=r^{-1}, r q = s 2 = 1 , sr s − 1 = r − 1 , and every element is uniquely r a r^{a} r a or r a s r^{a}s r a s with a ∈ A q a\in A_{q} a ∈ A q . Hence
D 2 q = ⟨ r , s ∣ r q = s 2 = 1 , s r s − 1 = r − 1 ⟩ = ⟨ r , s ∣ r q = s 2 = ( r s ) 2 = 1 ⟩ . D_{2q}=\langle r,s\mid r^{q}=s^{2}=1,\ srs^{-1}=r^{-1}\rangle
=\langle r,s\mid r^{q}=s^{2}=(rs)^{2}=1\rangle. D 2 q = ⟨ r , s ∣ r q = s 2 = 1 , sr s − 1 = r − 1 ⟩ = ⟨ r , s ∣ r q = s 2 = ( r s ) 2 = 1 ⟩ . In Q 4 q Q_{4q} Q 4 q , with the generators x = ( 1 , 0 ) x=(1,0) x = ( 1 , 0 ) and y = ( 0 , 1 ) y=(0,1) y = ( 0 , 1 ) of Theorem 8.2 ,
Q 4 q = ⟨ x , y ∣ x 2 q = 1 , y 2 = x q , y x y − 1 = x − 1 ⟩ . Q_{4q}=\langle x,y\mid x^{2q}=1,\ y^{2}=x^{q},\ yxy^{-1}=x^{-1}\rangle. Q 4 q = ⟨ x , y ∣ x 2 q = 1 , y 2 = x q , y x y − 1 = x − 1 ⟩ .
Proof. The displayed digit laws give r a = ( a , 0 ) r^{a}=(a,0) r a = ( a , 0 ) , r a s = ( a , 1 ) r^{a}s=(a,1) r a s = ( a , 1 ) , x a = ( a , 0 ) x^{a}=(a,0) x a = ( a , 0 ) and x a y = ( a , 1 ) x^{a}y=(a,1) x a y = ( a , 1 ) directly. Substitution in the respective laws verifies the relations; normal forms give the stated orders and uniqueness.
Remark 8.12 (Geometric form). D 2 q D_{2q} D 2 q is the symmetry group of a regular q q q -gon: r r r is rotation by one vertex and s s s is a reflection. Thus D 6 D_{6} D 6 is the symmetry group of an equilateral triangle and D 8 D_{8} D 8 that of a square. The first non-Abelian dicyclic group is Q 8 Q_{8} Q 8 , not D 8 D_{8} D 8 : the two groups have the same order but different element-order structure.
Example 8.13 (The metacyclic group C 7 ⋊ 2 C 3 C_{7}\rtimes_{2}C_{3} C 7 ⋊ 2 C 3 ). Take q = 7 q=7 q = 7 , s = 3 s=3 s = 3 and r = 2 r=2 r = 2 in Theorem 8.8 ; the hypotheses hold because gcd ( 2 , 7 ) = 1 \gcd(2,7)=1 g cd( 2 , 7 ) = 1 and 2 3 = 8 ≡ 1 ( m o d 7 ) 2^{3}=8\equiv1\pmod7 2 3 = 8 ≡ 1 ( mod 7 ) . Writing out m 1 = d 3 ( 0 , m / 7 ) m_{1}=d_{3}(0,m/7) m 1 = d 3 ( 0 , m /7 ) and n 1 = d 3 ( 0 , n / 7 ) n_{1}=d_{3}(0,n/7) n 1 = d 3 ( 0 , n /7 ) , the law (62) on A 21 A_{21} A 21 is
m ∗ n = d 7 ( 0 , m + 2 d 3 ( 0 , m / 7 ) n ) + 7 d 3 ( 0 , d 3 ( 0 , m 7 ) + d 3 ( 0 , n 7 ) ) . (64) m\ast n=d_{7}\left(0,m+2^{d_{3}(0,m/7)}n\right)+7d_{3}\left(0,d_{3}\left(0,\frac{m}{7}\right)+d_{3}\left(0,\frac{n}{7}\right)\right). \tag*{(64)} m ∗ n = d 7 ( 0 , m + 2 d 3 ( 0 , m /7 ) n ) + 7 d 3 ( 0 , d 3 ( 0 , 7 m ) + d 3 ( 0 , 7 n ) ) . (64) With x = 1 x=1 x = 1 and y = 7 y=7 y = 7 every element is uniquely m = x m 0 y m 1 m=x^{m_{0}}y^{m_{1}} m = x m 0 y m 1 , where m 0 = d 7 ( 0 , m ) m_{0}=d_{7}(0,m) m 0 = d 7 ( 0 , m ) , and x y x − 1 = x 2 xyx^{-1}=x^{2} x y x − 1 = x 2 . The resulting Cayley table is Table 8.3 . The six elements 1 , … , 6 1,\ldots,6 1 , … , 6 have order 7 7 7 , the fourteen elements 7 , … , 20 7,\ldots,20 7 , … , 20 have order 3 3 3 , and the centre is trivial: this is the Frobenius group of order 21 21 21 , the smallest non-Abelian group of odd order.
Table 8.3. Cayley table of C 7 ⋊ 2 C 3 C_7 \rtimes_2 C_3 C 7 ⋊ 2 C 3 on A 21 A_{21} A 21 given by the digital law (64) ; the entry in row m m m , column n n n is m ∗ n m \ast n m ∗ n . The rules separate the cosets { 0 , … , 6 } \{0,\ldots,6\} { 0 , … , 6 } , { 7 , … , 13 } \{7,\ldots,13\} { 7 , … , 13 } , { 14 , … , 20 } \{14,\ldots,20\} { 14 , … , 20 } of ⟨ x ⟩ \langle x\rangle ⟨ x ⟩ , i.e. the value of the upper digit m 1 m_1 m 1 .
Example 8.14 (Computing an entry of Table 8.3 ). We compute 12 ∗ 17 12 \ast17 12 ∗ 17 directly from (64) , using only the digit function (1) .
Upper digits.
m 1 = d 3 ( 0 , 12 7 ) = ⌊ 12 7 ⌋ − 3 ⌊ 12 21 ⌋ = 1 − 0 = 1 , n 1 = d 3 ( 0 , 17 7 ) = ⌊ 17 7 ⌋ − 3 ⌊ 17 21 ⌋ = 2 − 0 = 2. m_1=d_3\left(0,\frac{12}{7}\right)=\left\lfloor\frac{12}{7}\right\rfloor-3\left\lfloor\frac{12}{21}\right\rfloor=1-0=1,\qquad n_1=d_3\left(0,\frac{17}{7}\right)=\left\lfloor\frac{17}{7}\right\rfloor-3\left\lfloor\frac{17}{21}\right\rfloor=2-0=2. m 1 = d 3 ( 0 , 7 12 ) = ⌊ 7 12 ⌋ − 3 ⌊ 21 12 ⌋ = 1 − 0 = 1 , n 1 = d 3 ( 0 , 7 17 ) = ⌊ 7 17 ⌋ − 3 ⌊ 21 17 ⌋ = 2 − 0 = 2. Twisting factor. r m 1 = 2 1 = 2 r^{m_1}=2^1=2 r m 1 = 2 1 = 2 .
Lower digit of the product.
d 7 ( 0 , 12 + 2 ⋅ 17 ) = d 7 ( 0 , 46 ) = ⌊ 46 7 ⌋ − 7 ⌊ 46 7 ⌋ = 46 − 42 = 4. d_7(0,12+2\cdot17)=d_7(0,46)=\left\lfloor\frac{46}{7}\right\rfloor-7\left\lfloor\frac{46}{7}\right\rfloor=46-42=4. d 7 ( 0 , 12 + 2 ⋅ 17 ) = d 7 ( 0 , 46 ) = ⌊ 7 46 ⌋ − 7 ⌊ 7 46 ⌋ = 46 − 42 = 4. By Lemma 1.3 one may reduce first: m 0 = 5 m_0=5 m 0 = 5 , n 0 = 3 n_0=3 n 0 = 3 , and d 7 ( 0 , 5 + 2 ⋅ 3 ) = d 7 ( 0 , 11 ) = 4 d_7(0,5+2\cdot3)=d_7(0,11)=4 d 7 ( 0 , 5 + 2 ⋅ 3 ) = d 7 ( 0 , 11 ) = 4 .
Upper digit of the product.
d 3 ( 0 , m 1 + n 1 ) = d 3 ( 0 , 3 ) = 3 − 3 ⌊ 1 ⌋ = 0. d_3(0,m_1+n_1)=d_3(0,3)=3-3\left\lfloor1\right\rfloor=0. d 3 ( 0 , m 1 + n 1 ) = d 3 ( 0 , 3 ) = 3 − 3 ⌊ 1 ⌋ = 0. Assembly. 12 ∗ 17 = 4 + 7 ⋅ 0 = 4 12\ast17=4+7\cdot0=4 12 ∗ 17 = 4 + 7 ⋅ 0 = 4 , which is the entry in row 12, column 17 of the table.
Check via the normal form. Since 12 = x 5 y 12=x^5y 12 = x 5 y , 17 = x 3 y 2 17=x^3y^2 17 = x 3 y 2 and y x 3 y − 1 = x 6 yx^3y^{-1}=x^6 y x 3 y − 1 = x 6 ,
12 ∗ 17 = x 5 y x 3 y 2 = x 5 ( y x 3 y − 1 ) y 3 = x 11 = x 4 = 4. 12\ast17=x^5y\,x^3y^2=x^5\left(yx^3y^{-1}\right)y^3=x^{11}=x^4=4. 12 ∗ 17 = x 5 y x 3 y 2 = x 5 ( y x 3 y − 1 ) y 3 = x 11 = x 4 = 4. Non-commutativity. In the reverse order m 1 = 2 m_1=2 m 1 = 2 , so r m 1 = 4 r^{m_1}=4 r m 1 = 4 and
17 ∗ 12 = d 7 ( 0 , 17 + 4 ⋅ 12 ) + 7 d 3 ( 0 , 2 + 1 ) = d 7 ( 0 , 65 ) + 0 = 2 ≠ 4. 17\ast12=d_7(0,17+4\cdot12)+7d_3(0,2+1)=d_7(0,65)+0=2\ne4. 17 ∗ 12 = d 7 ( 0 , 17 + 4 ⋅ 12 ) + 7 d 3 ( 0 , 2 + 1 ) = d 7 ( 0 , 65 ) + 0 = 2 = 4.
Proposition 8.15 (Bracelets). Let D 2 q D_{2q} D 2 q act on the vertices A q A_q A q of a regular q q q -gon, generated by x ↦ x + 1 x \mapsto x+1 x ↦ x + 1 and x ↦ − x x \mapsto-x x ↦ − x modulo q q q , and let N q ( m ) = 1 q ∑ d ∣ q φ ( d ) m q / d N_q(m)=\frac{1}{q}\sum_{d\mid q}\varphi(d)m^{q/d} N q ( m ) = q 1 ∑ d ∣ q φ ( d ) m q / d be the number of necklaces (Proposition 7.7). The number of m m m -colourings of the vertices up to D 2 q D_{2q} D 2 q is
1 2 N q ( m ) + { 1 2 m ( q + 1 ) / 2 , q odd , 1 4 m q / 2 ( m + 1 ) , q even . \frac{1}{2}N_q(m)+
\begin{cases}
\frac{1}{2}m^{(q+1)/2}, & q\ \text{odd},\\
\frac{1}{4}m^{q/2}(m+1), & q\ \text{even}.
\end{cases} 2 1 N q ( m ) + { 2 1 m ( q + 1 ) /2 , 4 1 m q /2 ( m + 1 ) , q odd , q even .
Proof. The q q q rotations contribute 1 2 q ∑ r o t m c = 1 2 N q ( m ) \frac{1}{2q}\sum_{\mathrm{rot}}m^c=\frac{1}{2}N_q(m) 2 q 1 ∑ rot m c = 2 1 N q ( m ) . For odd q q q each of the q q q reflections fixes one vertex and swaps the others in pairs, so it has ( q + 1 ) / 2 (q+1)/2 ( q + 1 ) /2 cycles, contributing q m ( q + 1 ) / 2 / ( 2 q ) q\,m^{(q+1)/2}/(2q) q m ( q + 1 ) /2 / ( 2 q ) . For even q q q , half of the reflections fix two vertices ( q / 2 + 1 (q/2+1 ( q /2 + 1 cycles) and half fix none ( q / 2 (q/2 ( q /2 cycles), contributing q 2 ( m q / 2 + 1 + m q / 2 ) / ( 2 q ) \frac{q}{2}(m^{q/2+1}+m^{q/2})/(2q) 2 q ( m q /2 + 1 + m q /2 ) / ( 2 q ) . Apply Theorem 4.21 .
For q = 4 , 6 q=4,6 q = 4 , 6 and m = 2 m=2 m = 2 one gets 6 and 13 bracelets, and for q = 4 q=4 q = 4 , m = 3 m=3 m = 3 one gets 21, as in the example of the square after Theorem 4.22 .
Characters of metacyclic and dicyclic groups
Throughout this section ζ n = e 2 π i / n \zeta_n=e^{2\pi i/n} ζ n = e 2 π i / n , ω = ζ 3 \omega=\zeta_3 ω = ζ 3 , and characters are ordinary complex characters. Let G = C q ⋊ r C s G=C_q\rtimes_r C_s G = C q ⋊ r C s be the group of Theorem 8.8 and put x = 1 x=1 x = 1 , y = q y=q y = q . Since r 0 = 1 r^0=1 r 0 = 1 , the law (62) gives x a = a x^a=a x a = a and x a y α = a + q α x^a y^\alpha=a+q\alpha x a y α = a + q α for a ∈ A q a\in A_q a ∈ A q and α ∈ A s \alpha\in A_s α ∈ A s , so the two digits
m 0 = d q ( 0 , m ) , m 1 = d s ( 0 , m / q ) m_0=d_q(0,m),\qquad m_1=d_s(0,m/q) m 0 = d q ( 0 , m ) , m 1 = d s ( 0 , m / q )
of m m m are exactly the exponents in the normal form m = x m 0 y m 1 m=x^{m_0}y^{m_1} m = x m 0 y m 1 . Moreover y x y − 1 = x r yxy^{-1}=x^r y x y − 1 = x r , and A = ⟨ x ⟩ = A q A=\langle x\rangle=A_q A = ⟨ x ⟩ = A q is a cyclic normal subgroup with cyclic quotient G / A ≅ C s G/A\cong C_s G / A ≅ C s . The characters of A A A are λ j ( x a ) = ζ q j a \lambda_j(x^a)=\zeta_q^{ja} λ j ( x a ) = ζ q j a , j ∈ Z / q j\in\mathbb{Z}/q j ∈ Z / q , and conjugation by y y y acts on them by λ j y ( x a ) = λ j ( y x a y − 1 ) = λ j ( x r a ) \lambda_j^y(x^a)=\lambda_j(yx^ay^{-1})=\lambda_j(x^{ra}) λ j y ( x a ) = λ j ( y x a y − 1 ) = λ j ( x r a ) . Hence the cyclic group ⟨ r ⟩ ≤ ( Z / q ) × \langle r\rangle\le(\mathbb{Z}/q)^\times ⟨ r ⟩ ≤ ( Z / q ) × acts on Z / q \mathbb{Z}/q Z / q by multiplication; for an orbit O = j ⟨ r ⟩ O=j\langle r\rangle O = j ⟨ r ⟩ write e O = ∣ O ∣ e_O=|O| e O = ∣ O ∣ , which divides s s s , and t O = s / e O t_O=s/e_O t O = s / e O .
Theorem 8.16 (Characters of metacyclic groups). For every orbit O O O of ⟨ r ⟩ \langle r\rangle ⟨ r ⟩ on Z / q \mathbb{Z}/q Z / q and every μ ∈ C \mu\in\mathbb{C} μ ∈ C with μ t O = 1 \mu^{t_O}=1 μ t O = 1 , the function
χ O , μ ( m ) = { μ m 1 / e O ∑ k ∈ O ζ q k m 0 , e O ∣ m 1 , 0 , e O ∤ m 1 , (65) \chi_{O,\mu}(m)=
\begin{cases}
\mu^{m_1/e_O}\sum_{k\in O}\zeta_q^{km_0}, & e_O\mid m_1,\\
0, & e_O\nmid m_1,
\end{cases} \tag*{(65)} χ O , μ ( m ) = { μ m 1 / e O ∑ k ∈ O ζ q k m 0 , 0 , e O ∣ m 1 , e O ∤ m 1 , (65) is an irreducible character of C q ⋊ r C s C_q\rtimes_r C_s C q ⋊ r C s of degree e O e_O e O . These characters are pairwise distinct and exhaust Irr ( C q ⋊ r C s ) \operatorname{Irr}(C_q\rtimes_r C_s) Irr ( C q ⋊ r C s ) . In particular
k ( C q ⋊ r C s ) = ∑ O s e O , k(C_q\rtimes_r C_s)=\sum_O\frac{s}{e_O}, k ( C q ⋊ r C s ) = O ∑ e O s , and the linear characters are the χ { j } , μ \chi_{\{j\},\mu} χ { j } , μ with ( r − 1 ) j ≡ 0 ( m o d q ) (r-1)j\equiv0\pmod q ( r − 1 ) j ≡ 0 ( mod q ) .
Proof. Fix j ∈ O j\in O j ∈ O , e = e O e=e_O e = e O , t = t O t=t_O t = t O , and let S = { m : e ∣ m 1 } = ⟨ x , y e ⟩ S=\{m:e\mid m_1\}=\langle x,y^e\rangle S = { m : e ∣ m 1 } = ⟨ x , y e ⟩ , a normal subgroup of index e e e containing A A A . By (62) , x a y e β ∗ x b y e β ′ = x a + r e β b y e ( β + β ′ ) x^a y^{e\beta}*x^b y^{e\beta'}=x^{a+r^{e\beta}b}y^{e(\beta+\beta')} x a y e β ∗ x b y e β ′ = x a + r e β b y e ( β + β ′ ) , and ζ q j r e β b = ζ q j b \zeta_q^{jr^{e\beta}b}=\zeta_q^{jb} ζ q j r e β b = ζ q j b because r e j = j r^e j=j r e j = j . As μ t = 1 \mu^t=1 μ t = 1 , the exponent β \beta β may be read modulo t t t , so λ ( x a y e β ) = ζ q j a μ β \lambda(x^a y^{e\beta})=\zeta_q^{ja}\mu^\beta λ ( x a y e β ) = ζ q j a μ β is a linear character of S S S extending λ j \lambda_j λ j . Induce it over the transversal { y c : 0 ≤ c < e } \{y^c:0\le c<e\} { y c : 0 ≤ c < e } . Since y c x a y e β y − c = x r c a y e β y^c x^a y^{e\beta}y^{-c}=x^{r^ca}y^{e\beta} y c x a y e β y − c = x r c a y e β ,
Ind S G λ ( x a y e β ) = μ β ∑ c = 0 e − 1 ζ q j r c a = μ β ∑ k ∈ O ζ q k a , \operatorname{Ind}_S^G\lambda(x^a y^{e\beta})=\mu^\beta\sum_{c=0}^{e-1}\zeta_q^{jr^ca}=\mu^\beta\sum_{k\in O}\zeta_q^{ka}, Ind S G λ ( x a y e β ) = μ β c = 0 ∑ e − 1 ζ q j r c a = μ β k ∈ O ∑ ζ q k a ,
while the induced character vanishes off the normal subgroup S S S ; this is (65) . Its restriction to S S S is the sum of the e e e conjugates λ y \lambda^{y} λ y , which are distinct because their restrictions λ r c j \lambda^{r^{c_j}} λ r c j to A A A are distinct. Frobenius reciprocity gives ⟨ χ O , μ , χ O , μ ⟩ = ⟨ λ , χ O , μ ∣ S ⟩ S = 1 \langle\chi_{O,\mu},\chi_{O,\mu}\rangle=\langle\lambda,\chi_{O,\mu}|_S\rangle_S=1 ⟨ χ O , μ , χ O , μ ⟩ = ⟨ λ , χ O , μ ∣ S ⟩ S = 1 , so χ O , μ \chi_{O,\mu} χ O , μ is irreducible. The restriction ∑ k ∈ O λ k \sum_{k\in O}\lambda_k ∑ k ∈ O λ k of χ O , μ \chi_{O,\mu} χ O , μ to A A A determines O O O , and χ O , μ ( y e ) = e μ \chi_{O,\mu}(y^e)=e\mu χ O , μ ( y e ) = e μ determines μ \mu μ . Finally ∑ O e O 2 = ∑ O s e O = s q = ∣ G ∣ \sum_O e_O^2=\sum_O s e_O=sq=|G| ∑ O e O 2 = ∑ O s e O = s q = ∣ G ∣ , so no irreducible character is missing.
The formula is digital in the strict sense: χ O , μ ( m ) \chi_{O,\mu}(m) χ O , μ ( m ) is computed from the two digits m 0 m_0 m 0 , m 1 m_1 m 1 of m m m alone, without multiplying in the group.
Corollary 8.17. Let q q q be prime and let r r r have multiplicative order exactly s s s modulo q q q . Then C q ⋊ r C s C_q\rtimes_r C_s C q ⋊ r C s has the s s s linear characters m ↦ μ m 1 m\mapsto\mu^{m_1} m ↦ μ m 1 (μ s = 1 \mu^s=1 μ s = 1 ) and ( q − 1 ) / s (q-1)/s ( q − 1 ) / s characters of degree s s s , which vanish off A A A and take the values ∑ k ∈ j ⟨ r ⟩ ζ q k m 0 \sum_{k\in j\langle r\rangle}\zeta_q^{km_0} ∑ k ∈ j ⟨ r ⟩ ζ q k m 0 on A A A . Hence k ( C q ⋊ r C s ) = s + ( q − 1 ) / s k(C_q\rtimes_r C_s)=s+(q-1)/s k ( C q ⋊ r C s ) = s + ( q − 1 ) / s .
Proof. The orbits are { 0 } \{0\} { 0 } and the ( q − 1 ) / s (q-1)/s ( q − 1 ) / s cosets of ⟨ r ⟩ \langle r\rangle ⟨ r ⟩ in ( Z / q ) × (\mathbb{Z}/q)^\times ( Z / q ) × , each of size s s s .
Example 8.18. For C 5 ⋊ 2 C 4 C_5\rtimes_2 C_4 C 5 ⋊ 2 C 4 (order 20) the nonzero residues form one orbit, so there is a single character of degree 4; it equals 4 at 0, ζ 5 + ζ 5 2 + ζ 5 3 + ζ 5 4 = − 1 \zeta_5+\zeta_5^2+\zeta_5^3+\zeta_5^4=-1 ζ 5 + ζ 5 2 + ζ 5 3 + ζ 5 4 = − 1 on A ∖ { 0 } A\setminus\{0\} A ∖ { 0 } and 0 elsewhere. For C 7 ⋊ 2 C 3 C_7\rtimes_2 C_3 C 7 ⋊ 2 C 3 (order 21) the orbits { 1 , 2 , 4 } \{1,2,4\} { 1 , 2 , 4 } and { 3 , 5 , 6 } \{3,5,6\} { 3 , 5 , 6 } give two characters of degree 3 whose values on x x x are the quadratic Gauss periods ζ 7 + ζ 7 2 + ζ 7 4 = − 1 + i 7 2 \zeta_7+\zeta_7^2+\zeta_7^4=\frac{-1+i\sqrt{7}}{2} ζ 7 + ζ 7 2 + ζ 7 4 = 2 − 1 + i 7 and its complex conjugate. See Table 8.6 and Table 8.7 .
Table 8.6. Character table of C 5 ⋊ 2 C 4 C_5 \rtimes_2 C_4 C 5 ⋊ 2 C 4 (q = 5 q = 5 q = 5 , r = 2 r = 2 r = 2 , s = 4 s = 4 s = 4 ). Columns are indexed by class representatives x a y α = a + ∣ A ∣ α x^a y^\alpha= a + |A|\alpha x a y α = a + ∣ A ∣ α ; ν \nu ν is the Frobenius–Schur indicator.
Table 8.7. Character table of C 7 ⋊ 2 C 3 C_7 \rtimes_2 C_3 C 7 ⋊ 2 C 3 (q = 7 q = 7 q = 7 , r = 2 r = 2 r = 2 , s = 3 s = 3 s = 3 ). Columns are indexed by class representatives x a y α = a + ∣ A ∣ α x^a y^\alpha= a + |A|\alpha x a y α = a + ∣ A ∣ α ; ν \nu ν is the Frobenius–Schur indicator.
Corollary 8.19 (Characters of dihedral groups). Let q ≥ 3 q\ge3 q ≥ 3 and D 2 q = C q ⋊ q − 1 C 2 D_{2q}=C_q\rtimes_{q-1}C_2 D 2 q = C q ⋊ q − 1 C 2 as in Corollary 8.9. Its irreducible characters are the linear characters m ↦ ( ± 1 ) m 1 m\mapsto(\pm1)^{m_1} m ↦ ( ± 1 ) m 1 , together with m ↦ ( − 1 ) m 0 ( ± 1 ) m 1 m\mapsto(-1)^{m_0}(\pm1)^{m_1} m ↦ ( − 1 ) m 0 ( ± 1 ) m 1 when q q q is even, and the characters of degree 2
ψ j ( m ) = { 2 cos ( 2 π j m 0 / q ) , m 1 = 0 , 0 , m 1 = 1 , 1 ≤ j < q / 2. \psi_j(m)=\begin{cases}2\cos(2\pi j m_0/q),&m_1=0,\\0,&m_1=1,\end{cases}\qquad1\le j<q/2. ψ j ( m ) = { 2 cos ( 2 π j m 0 / q ) , 0 , m 1 = 0 , m 1 = 1 , 1 ≤ j < q /2. Thus k ( D 2 q ) = ( q + 3 ) / 2 k(D_{2q})=(q+3)/2 k ( D 2 q ) = ( q + 3 ) /2 for odd q q q and ( q + 6 ) / 2 (q+6)/2 ( q + 6 ) /2 for even q q q , and every irreducible character of D 2 q D_{2q} D 2 q has Frobenius–Schur indicator + 1 +1 + 1 .
Proof. Here r = − 1 r=-1 r = − 1 , whose orbits on Z / q \mathbb{Z}/q Z / q are { 0 } \{0\} { 0 } , { q / 2 } \{q/2\} { q /2 } for even q q q , and { j , − j } \{j,-j\} { j , − j } for 1 ≤ j < q / 2 1\le j<q/2 1 ≤ j < q /2 ; apply Theorem 8.16 . Every element x a y x^a y x a y is an involution, so the indicator ν ( ψ j ) = 1 2 q ∑ g ψ j ( g 2 ) \nu(\psi_j)=\frac{1}{2q}\sum_g\psi_j(g^2) ν ( ψ j ) = 2 q 1 ∑ g ψ j ( g 2 ) equals 1 2 q ( ∑ a 2 cos ( 4 π j a / q ) + q ψ j ( 0 ) ) = 1 \frac{1}{2q}\left(\sum_a2\cos(4\pi ja/q)+q\psi_j(0)\right)=1 2 q 1 ( ∑ a 2 cos ( 4 π j a / q ) + q ψ j ( 0 ) ) = 1 , the first sum vanishing because q ∤ 2 j q\nmid2j q ∤ 2 j . The linear characters are real, hence have indicator 1.
The dicyclic group is not a split extension of ⟨ x ⟩ \langle x\rangle ⟨ x ⟩ for q ≥ 2 q\ge2 q ≥ 2 , so Theorem 8.16 does not apply verbatim, but the same induction argument does.
Proposition 8.20 (Characters of dicyclic groups). Let q ≥ 2 q\ge2 q ≥ 2 and write m ∈ Q 4 q m\in Q_{4q} m ∈ Q 4 q as m = x m 0 y m 1 m=x^{m_0}y^{m_1} m = x m 0 y m 1 with m 0 = d 2 q ( 0 , m ) m_0=d_{2q}(0,m) m 0 = d 2 q ( 0 , m ) and m 1 = d 2 ( 0 , m / 2 q ) m_1=d_2(0,m/2q) m 1 = d 2 ( 0 , m /2 q ) , where x = 1 x=1 x = 1 and y = 2 q y=2q y = 2 q as in Theorem 8.2 . Then Irr ( Q 4 q ) \operatorname{Irr}(Q_{4q}) Irr ( Q 4 q ) consists of
the four linear characters λ ( m ) = ε m 0 δ m 1 \lambda(m)=\varepsilon^{m_0}\delta^{m_1} λ ( m ) = ε m 0 δ m 1 with ε = ± 1 \varepsilon=\pm1 ε = ± 1 and δ 2 = ε q \delta^2=\varepsilon^q δ 2 = ε q ;
the q − 1 q-1 q − 1 characters of degree 2
ψ j ( m ) = { 2 cos ( π j m 0 / q ) , m 1 = 0 , 0 , m 1 = 1 , 1 ≤ j ≤ q − 1. \psi_j(m)=\begin{cases}2\cos(\pi j m_0/q),&m_1=0,\\0,&m_1=1,\end{cases}\qquad1\le j\le q-1. ψ j ( m ) = { 2 cos ( π j m 0 / q ) , 0 , m 1 = 0 , m 1 = 1 , 1 ≤ j ≤ q − 1. Hence k ( Q 4 q ) = q + 3 k(Q_{4q})=q+3 k ( Q 4 q ) = q + 3 . The indicator of ψ j \psi_j ψ j is ( − 1 ) j (-1)^j ( − 1 ) j ; a linear character has indicator 1 if it is real and 0 if δ = ± i \delta=\pm i δ = ± i , which happens exactly when q q q is odd and ε = − 1 \varepsilon=-1 ε = − 1 .
Proof. The subgroup A = ⟨ x ⟩ = A 2 q A=\langle x\rangle=A_{2q} A = ⟨ x ⟩ = A 2 q is cyclic of index 2 and y x y − 1 = x − 1 yxy^{-1}=x^{-1} y x y − 1 = x − 1 by Proposition 8.11. For j ∈ Z / 2 q ∖ { 0 , q } j\in\mathbb{Z}/2q\setminus\{0,q\} j ∈ Z /2 q ∖ { 0 , q } the characters λ j \lambda_j λ j and λ j ∗ = λ − j \lambda_j^*=\lambda_{-j} λ j ∗ = λ − j differ, so ψ j = Ind A G λ j \psi_j=\operatorname{Ind}_A^G\lambda_j ψ j = Ind A G λ j is irreducible by Frobenius reciprocity, with the stated values. Since ψ j = ψ 2 q − j \psi_j=\psi_{2q-j} ψ j = ψ 2 q − j and the restrictions λ j + λ − j \lambda_j+\lambda_{-j} λ j + λ − j are distinct for 1 ≤ j ≤ q − 1 1\leq j\leq q-1 1 ≤ j ≤ q − 1 , this gives q − 1 q-1 q − 1 characters of degree 2. A linear character is determined by ε = λ ( x ) \varepsilon=\lambda(x) ε = λ ( x ) and δ = λ ( y ) \delta=\lambda(y) δ = λ ( y ) ; the relations of Proposition 8.11 hold if and only if ε 2 = 1 \varepsilon^2=1 ε 2 = 1 and δ 2 = ε q \delta^2=\varepsilon^q δ 2 = ε q , which gives four characters. As 4 ⋅ 1 + ( q − 1 ) ⋅ 4 = 4 q 4\cdot1+(q-1)\cdot4=4q 4 ⋅ 1 + ( q − 1 ) ⋅ 4 = 4 q , the list is complete. For the indicators note that ( x a y ) 2 = x q (x^ay)^2=x^q ( x a y ) 2 = x q by (60) , so
ν ( ψ j ) = 1 4 q ( ∑ a ∈ A 2 q ψ j ( x 2 a ) + 2 q ψ j ( x q ) ) = 0 + 2 q ⋅ 2 cos ( π j ) 4 q = ( − 1 ) j , \nu(\psi_j)=\frac{1}{4q}\left(\sum_{a\in A_{2q}}\psi_j(x^{2a})+2q\psi_j(x^q)\right)=\frac{0+2q\cdot2\cos(\pi j)}{4q}=(-1)^j, ν ( ψ j ) = 4 q 1 a ∈ A 2 q ∑ ψ j ( x 2 a ) + 2 q ψ j ( x q ) = 4 q 0 + 2 q ⋅ 2 cos ( π j ) = ( − 1 ) j ,
the sum vanishing because q ∤ j q\nmid j q ∤ j . A linear character has indicator 1 if λ 2 = 1 \lambda^2=1 λ 2 = 1 and 0 otherwise. □
Example 8.21 (D 8 D_8 D 8 versus Q 8 Q_8 Q 8 ). By Corollary 8.19 and Proposition 8.20 , D 8 D_8 D 8 and Q 8 Q_8 Q 8 both have four linear characters and one character of degree 2, and their character tables coincide. The indicators separate them: the degree-2 character of D 8 D_8 D 8 has ν = 1 \nu=1 ν = 1 , while ψ 1 \psi_1 ψ 1 of Q 8 Q_8 Q 8 has ν = − 1 \nu=-1 ν = − 1 , so it is afforded by no real representation. More generally Q 4 q Q_{4q} Q 4 q has exactly ⌊ q / 2 ⌋ \lfloor q/2\rfloor ⌊ q /2 ⌋ irreducible characters with indicator − 1 -1 − 1 .
Table 8.4 lists the character data of nine examples and Table 8.5 , Table 8.6 and Table 8.7 display three complete character tables. All of them are generated from the digit laws by scripts/nonab_char_verify.py, which builds each group, computes its classes and the characters of Theorem 8.16 and Proposition 8.20 , verifies the orthogonality relations exactly in Z [ ζ N ] \mathbb{Z}[\zeta_N] Z [ ζ N ] , and checks every character against GAP’s Irr.
Table 8.4. Character data of the metacyclic, dihedral and dicyclic examples: law parameters, class number k ( G ) k(G) k ( G ) , number ∣ G : G ′ ∣ |G:G'| ∣ G : G ′ ∣ of linear characters, degree multiset and numbers of characters with Frobenius–Schur indicator + 1 , 0 , − 1 +1,0,-1 + 1 , 0 , − 1 .
Table 8.5. Character table of Q 16 Q_{16} Q 16 (q = 4 q=4 q = 4 ). Columns are indexed by class representatives x a y α = a + ∣ A ∣ α x^ay^\alpha=a+|A|\alpha x a y α = a + ∣ A ∣ α ; ν \nu ν is the Frobenius–Schur indicator.
Wreath products
Theorem 8.22. Let G = ( A N , ∗ ) G = (A_N, *) G = ( A N , ∗ ) and H = ( A s , ∗ H ) H = (A_s, *_{H}) H = ( A s , ∗ H ) be groups and h ↦ S h h \mapsto S_h h ↦ S h a homomorphism H → Sym ℓ H \to\operatorname{Sym}_{\ell} H → Sym ℓ . For m ∈ A N ℓ s m \in A_{N^\ell s} m ∈ A N ℓ s put m ♮ = d s ( 0 , m / N ℓ ) m^\natural= d_s(0,m/N^\ell) m ♮ = d s ( 0 , m / N ℓ ) (control digit) and π m = S m ♮ \pi_m = S_{m^\natural} π m = S m ♮ . Then
m ∗ n = ∑ i = 0 ℓ − 1 ( d N ( i , m ) ∗ d N ( π m − 1 ( i ) , n ) ) N i + N ℓ ( m ♮ ∗ H n ♮ ) (66) m * n = \sum_{i=0}^{\ell-1} \left(d_N(i,m) * d_N\left(\pi_m^{-1}(i),n\right)\right)N^i + N^\ell\left(m^\natural*_H n^\natural\right) \tag*{(66)} m ∗ n = i = 0 ∑ ℓ − 1 ( d N ( i , m ) ∗ d N ( π m − 1 ( i ) , n ) ) N i + N ℓ ( m ♮ ∗ H n ♮ ) (66) is a group, the wreath product G ≀ H G \wr H G ≀ H with fibre G G G and control H H H , of order N ℓ s N^\ell s N ℓ s .
Proof. This is Theorem 8.1 with G 0 = G ℓ G_0 = G^\ell G 0 = G ℓ , the direct product of Eq. (19) . Define
ϕ h ( f ) = ∑ i d N ( S h − 1 ( i ) , f ) N i . \phi_h(f) = \sum_i d_N\left(S_h^{-1}(i),f\right)N^i. ϕ h ( f ) = i ∑ d N ( S h − 1 ( i ) , f ) N i .
Each ϕ h \phi_h ϕ h permutes coordinates, so it is an automorphism, and ϕ h ∘ ϕ k = ϕ h ∗ H k \phi_h \circ\phi_k = \phi_{h *_H k} ϕ h ∘ ϕ k = ϕ h ∗ H k because S h ∗ H k − 1 = ( S h ∘ S k ) − 1 S_{h *_H k}^{-1} = (S_h \circ S_k)^{-1} S h ∗ H k − 1 = ( S h ∘ S k ) − 1 .
Remark 8.23. The transfer acts only on the right factor. If π m \pi_m π m also permuted the digits of m m m , the law would fail to be associative. The control element is recovered from m ∗ n m * n m ∗ n through
d N ( i , m ∗ n ) = d N ( i , m ) ∗ d N ( π m − 1 ( i ) , n ) . d_N(i,m * n) = d_N(i,m) * d_N\left(\pi_m^{-1}(i),n\right). d N ( i , m ∗ n ) = d N ( i , m ) ∗ d N ( π m − 1 ( i ) , n ) .
Thus the control state determines exactly which coordinate of the right fibre enters each coordinate of the product. This is the correct digital form of the transfer mechanism in the original notes.
Wreath products and Pólya counting
The wreath product is the group in which Pólya’s theory lives: G ≀ H G \wr H G ≀ H is the symmetry group of ℓ \ell ℓ blocks, each carrying a pattern that may be moved by G G G , where the blocks themselves may be permuted by H H H . This section proves that its cycle index is the composition of the cycle indices of G G G and H H H (Theorem 8.25 ), derives the corresponding orbit counts (Corollary 8.26 and Theorem 8.27), and shows how the groups of de Bruijn’s theorem sit inside a wreath product (Proposition 8.28 ). The notation of § 10.4.4 is used throughout.
The action on blocks
Let G G G act on a finite set X X X with ∣ X ∣ = r |X|=r ∣ X ∣ = r , let H H H be a group with a homomorphism h ↦ S h ∈ Sym ℓ h\mapsto S_h\in\operatorname{Sym}_{\ell} h ↦ S h ∈ Sym ℓ , and write the elements of G ≀ H G\wr H G ≀ H as pairs ( f , h ) (f,h) ( f , h ) with f = ( f 0 , … , f ℓ − 1 ) ∈ G ℓ f=(f_0,\ldots,f_{\ell-1})\in G^\ell f = ( f 0 , … , f ℓ − 1 ) ∈ G ℓ . In the digital form of Theorem 8.22, G = ( A N , ∗ ) G=(\mathcal{A}_N,*) G = ( A N , ∗ ) , the pair ( f , h ) (f,h) ( f , h ) is the number m m m with digits f i = d N ( i , m ) f_i=d_N(i,m) f i = d N ( i , m ) and control digit m ♮ = h m^\natural=h m ♮ = h , and (8.8) is the product
( f , h ) ( f ′ , h ′ ) = ( f ⋅ ( S h f ′ ) , h h ′ ) , ( S h f ′ ) i = f S h − 1 ( i ) ′ . (67) (f,h)(f',h')=\bigl(f\cdot(S_hf'),hh'\bigr),\qquad(S_hf')_i=f'_{S_{h}^{-1}(i)}. \tag*{(67)} ( f , h ) ( f ′ , h ′ ) = ( f ⋅ ( S h f ′ ) , h h ′ ) , ( S h f ′ ) i = f S h − 1 ( i ) ′ . (67)
The imprimitive action of G ≀ H G\wr H G ≀ H on the r ℓ r^\ell r ℓ points X × A ℓ X\times A_\ell X × A ℓ , the set X X X repeated in ℓ \ell ℓ blocks, is
( f , h ) ⋅ ( x , i ) = ( f S h ( i ) ⋅ x , S h ( i ) ) . (68) (f,h)\cdot(x,i)=\bigl(f_{S_h(i)}\cdot x,S_h(i)\bigr). \tag*{(68)} ( f , h ) ⋅ ( x , i ) = ( f S h ( i ) ⋅ x , S h ( i ) ) . (68)
That this is an action follows from (8.9): applying ( f ′ , h ′ ) (f',h') ( f ′ , h ′ ) and then ( f , h ) (f,h) ( f , h ) sends ( x , i ) (x,i) ( x , i ) to ( f S h S h ′ ( i ) f S h ′ ( i ) ′ ⋅ x , S h S h ′ ( i ) ) \bigl(f_{S_hS_{h'}(i)}f'_{S_{h'}(i)}\cdot x,S_hS_{h'}(i)\bigr) ( f S h S h ′ ( i ) f S h ′ ( i ) ′ ⋅ x , S h S h ′ ( i ) ) , and the coordinate of f ⋅ ( S h f ′ ) f\cdot(S_hf') f ⋅ ( S h f ′ ) at S h S h ′ ( i ) S_hS_{h'}(i) S h S h ′ ( i ) is f S h S h ′ ( i ) f S h ′ ( i ) ′ f_{S_hS_{h'}(i)}f'_{S_{h'}(i)} f S h S h ′ ( i ) f S h ′ ( i ) ′ . When G = ( A N , ∗ ) G=(\mathcal{A}_N,*) G = ( A N , ∗ ) acts on X = A N X=\mathcal{A}_N X = A N by left multiplication, the action of the number m m m is
( x , i ) ⟼ ( d N ( π m ( i ) , m ) ∗ x , π m ( i ) ) , π m = S m ♮ , (x,i)\longmapsto\bigl(d_N(\pi_m(i),m)*x,\pi_m(i)\bigr),\qquad\pi_m=S_{m^\natural}, ( x , i ) ⟼ ( d N ( π m ( i ) , m ) ∗ x , π m ( i ) ) , π m = S m ♮ ,
so the control digit moves the block and the digit of m m m at the new block acts on the point inside it.
Lemma 8.24 (Cycles of an element). Let ( f , h ) ∈ G ≀ H (f,h)\in G\wr H ( f , h ) ∈ G ≀ H , and let C = ( i 0 , i 1 , … , i k − 1 ) C=(i_0,i_1,\ldots,i_{k-1}) C = ( i 0 , i 1 , … , i k − 1 ) be a cycle of S h S_h S h of length k k k , i t = S h t ( i 0 ) i_t=S_h^t(i_0) i t = S h t ( i 0 ) . Put g C = f i 0 f i k − 1 ⋯ f i 1 ∈ G g_C=f_{i_0}f_{i_{k-1}}\cdots f_{i_1}\in G g C = f i 0 f i k − 1 ⋯ f i 1 ∈ G , the cycle product.
The cycles of ( f , h ) (f,h) ( f , h ) on X × C X\times C X × C correspond bijectively to the cycles of g C g_C g C on X X X , and a cycle of length j j j of g C g_C g C corresponds to a cycle of length j k jk j k .
For fixed h h h , the map f ↦ ( g C ) C f\mapsto(g_C)_C f ↦ ( g C ) C from G ℓ G^\ell G ℓ to G c ( S h ) G^{c(S_h)} G c ( S h ) is surjective and each of its fibres has ∣ G ∣ ℓ − c ( S h ) |G|^{\ell-c(S_h)} ∣ G ∣ ℓ − c ( S h ) elements.
Proof. (1) By (8.10), ( f , h ) t ⋅ ( x , i 0 ) = ( f i t ⋯ f i 1 x , i t ) (f,h)^t\cdot(x,i_0)=(f_{i_t}\cdots f_{i_1}x,i_t) ( f , h ) t ⋅ ( x , i 0 ) = ( f i t ⋯ f i 1 x , i t ) , and for t = k t=k t = k this is ( g C x , i 0 ) (g_Cx,i_0) ( g C x , i 0 ) . Hence the cycle of ( x , i 0 ) (x,i_0) ( x , i 0 ) meets the fibre over i 0 i_0 i 0 exactly in the g C g_C g C -cycle of x x x ; if that has length j j j , the cycle has length j k jk j k . Every cycle on X × C X\times C X × C meets this fibre. (2) Distinct cycles of S h S_h S h involve disjoint sets of coordinates, and every coordinate lies on one cycle. Within a cycle, each coordinate occurs once in g C g_C g C , so when all other coordinates are fixed the map f i ↦ g C f_i\mapsto g_C f i ↦ g C is a bijection G → G G\to G G → G . Hence each g C g_C g C is attained by ∣ G ∣ k − 1 |G|^{k-1} ∣ G ∣ k − 1 choices of the coordinates on C C C , and the fibre over ( g C ) C (g_C)_C ( g C ) C has ∏ C ∣ G ∣ k C − 1 = ∣ G ∣ ℓ − c ( S h ) \prod_C|G|^{k_C-1}=|G|^{\ell-c(S_h)} ∏ C ∣ G ∣ k C − 1 = ∣ G ∣ ℓ − c ( S h ) elements. □
The cycle index
Theorem 8.25 (Cycle index of a wreath product). Let Z ( H ) = Z ( H ; y 1 , y 2 , … ) Z(H)=Z(H;y_1,y_2,\ldots) Z ( H ) = Z ( H ; y 1 , y 2 , … ) be the cycle index of H H H acting on A ℓ A_\ell A ℓ through S S S , and put
z k = Z ( G ; x k , x 2 k , x 3 k , … ) = 1 ∣ G ∣ ∑ g ∈ G ∏ j ≥ 1 x j k c j ( g ) . z_k=Z(G;x_k,x_{2k},x_{3k},\ldots)=\frac{1}{|G|}\sum_{g\in G}\prod_{j\geq1}x_{jk}^{c_j(g)}. z k = Z ( G ; x k , x 2 k , x 3 k , … ) = ∣ G ∣ 1 g ∈ G ∑ j ≥ 1 ∏ x j k c j ( g ) . Then the cycle index of G ≀ H G\wr H G ≀ H in its action on X × A ℓ X\times A_\ell X × A ℓ is the composition
Z ( G ≀ H ; x 1 , x 2 , … ) = Z ( H ; z 1 , z 2 , z 3 , … ) = 1 ∣ H ∣ ∑ h ∈ H ∏ k ≥ 1 z k c k ( S h ) . Z(G\wr H;x_1,x_2,\ldots)=Z(H;z_1,z_2,z_3,\ldots)=\frac{1}{|H|}\sum_{h\in H}\prod_{k\geq1}z_k^{c_k(S_h)}. Z ( G ≀ H ; x 1 , x 2 , … ) = Z ( H ; z 1 , z 2 , z 3 , … ) = ∣ H ∣ 1 h ∈ H ∑ k ≥ 1 ∏ z k c k ( S h ) .
Proof. By Lemma 8.24 (1), the monomial of ( f , h ) (f,h) ( f , h ) is ∏ C ∏ j x j k C c j ( g C ) \prod_C\prod_j x_{jk_C}^{c_j(g_C)} ∏ C ∏ j x j k C c j ( g C ) , where C C C runs over the cycles of S h S_h S h and k C k_C k C is the length of C C C . Since ∣ G ≀ H ∣ = ∣ G ∣ ℓ ∣ H ∣ |G\wr H|=|G|^\ell|H| ∣ G ≀ H ∣ = ∣ G ∣ ℓ ∣ H ∣ , Lemma 8.24 (2) gives
Z ( G ≀ H ) = 1 ∣ G ∣ ℓ ∣ H ∣ ∑ h ∣ G ∣ ℓ − c ( S h ) ∏ C ∑ g ∈ G ∏ j x j k C c j ( g ) = 1 ∣ H ∣ ∑ h ∏ C z k C , Z(G\wr H)=\frac{1}{|G|^\ell|H|}\sum_h |G|^{\ell-c(S_h)}\prod_C\sum_{g\in G}\prod_j x_{jk_C}^{c_j(g)}=\frac{1}{|H|}\sum_h\prod_C z_{k_C}, Z ( G ≀ H ) = ∣ G ∣ ℓ ∣ H ∣ 1 h ∑ ∣ G ∣ ℓ − c ( S h ) C ∏ g ∈ G ∑ j ∏ x j k C c j ( g ) = ∣ H ∣ 1 h ∑ C ∏ z k C ,
and ∏ C z k C = ∏ k z k c k ( S h ) \prod_C z_{k_C}=\prod_k z_k^{c_k(S_h)} ∏ C z k C = ∏ k z k c k ( S h ) .
The substitution x j ↦ x j k x_j\mapsto x_{jk} x j ↦ x j k in the k k k -th slot is the reason for the name composition : a k k k -cycle of blocks lengthens every cycle inside a block by the factor k k k .
Corollary 8.26 (Colourings of the blocks). Let Y Y Y be a set of m m m colours with weights w w w and p k = ∑ y w ( y ) k p_k=\sum_y w(y)^k p k = ∑ y w ( y ) k . The weighted number of orbits of G ≀ H G\wr H G ≀ H on the colourings of X × A ℓ X\times A_\ell X × A ℓ is Z ( H ; ω 1 , ω 2 , … ) Z(H;\omega_1,\omega_2,\ldots) Z ( H ; ω 1 , ω 2 , … ) , where ω k = Z ( G ; p k , p 2 k , p 3 k , … ) \omega_k=Z(G;p_k,p_{2k},p_{3k},\ldots) ω k = Z ( G ; p k , p 2 k , p 3 k , … ) . In particular, with N K ( t ) = Z ( K ; t , t , … ) N_K(t)=Z(K;t,t,\ldots) N K ( t ) = Z ( K ; t , t , … ) the number of orbits of a group K K K on t t t -colourings of its set, the number of orbits is N H ( N G ( m ) ) N_H(N_G(m)) N H ( N G ( m )) .
Proof. Theorem 4.21 says that the weighted count is the cycle index at x j = p j x_j=p_j x j = p j , which turns z k z_k z k into ω k \omega_k ω k . For w ≡ 1 w\equiv1 w ≡ 1 all p j p_j p j equal m m m , so every z k z_k z k equals N G ( m ) N_G(m) N G ( m ) and Z ( H ; N G ( m ) , N G ( m ) , … ) = N H ( N G ( m ) ) Z(H;N_G(m),N_G(m),\ldots)=N_H(N_G(m)) Z ( H ; N G ( m ) , N G ( m ) , … ) = N H ( N G ( m )) .
The unweighted formula says that a colouring of the blocks is the same as an ℓ \ell ℓ -tuple of patterns, each pattern being a G G G -orbit of colourings of X X X , and H H H then permutes the tuple. The next theorem proves this reading directly and for arbitrary G G G -sets.
Product action
If G G G acts on a set Ω \Omega Ω , then G ≀ H G\wr H G ≀ H acts on Ω ℓ \Omega^\ell Ω ℓ by the product action
( ( f , h ) ⋅ u ) j = f j ⋅ u S h − 1 ( j ) . (69) ((f,h)\cdot u)_j=f_j\cdot u_{S_h^{-1}(j)}. \tag*{(69)} (( f , h ) ⋅ u ) j = f j ⋅ u S h − 1 ( j ) . (69)
For Ω = Y X \Omega=Y^X Ω = Y X this is the action on colourings of X × A ℓ X\times A_\ell X × A ℓ : a colouring c c c corresponds to u i = c ( ⋅ , i ) u_i=c(\mathord\cdot,i) u i = c ( ⋅ , i ) , and for γ = ( f , h ) \gamma=(f,h) γ = ( f , h ) one has γ − 1 ( y , j ) = ( f j − 1 y , S h − 1 ( j ) ) \gamma^{-1}(y,j)=(f_j^{-1}y,S_h^{-1}(j)) γ − 1 ( y , j ) = ( f j − 1 y , S h − 1 ( j )) by (68) , so ( γ c ) ( y , j ) = u S h − 1 ( j ) ( f j − 1 y ) = ( f j ⋅ u S h − 1 ( j ) ) ( y ) (\gamma c)(y,j)=u_{S_h^{-1}(j)}(f_j^{-1}y)=(f_j\cdot u_{S_h^{-1}(j)})(y) ( γ c ) ( y , j ) = u S h − 1 ( j ) ( f j − 1 y ) = ( f j ⋅ u S h − 1 ( j ) ) ( y ) .
Theorem 8.27 (Pólya counting for the product action). Let ω \omega ω be the number of orbits of G G G on Ω \Omega Ω . Then the number of orbits of G ≀ H G\wr H G ≀ H on Ω ℓ \Omega^\ell Ω ℓ is N H ( ω ) = 1 ∣ H ∣ ∑ h ∈ H ω c ( S h ) N_H(\omega)=\frac{1}{|H|}\sum_{h\in H}\omega^{c(S_h)} N H ( ω ) = ∣ H ∣ 1 ∑ h ∈ H ω c ( S h ) .
Proof. A tuple u u u is fixed by ( f , h ) (f,h) ( f , h ) if and only if u j = f j ⋅ u S h − 1 ( j ) u_j=f_j\cdot u_{S_h^{-1}(j)} u j = f j ⋅ u S h − 1 ( j ) for all j j j . Along a cycle C = ( j 0 , … , j k − 1 ) C=(j_0,\ldots,j_{k-1}) C = ( j 0 , … , j k − 1 ) of S h S_h S h this gives u j t = f j t u j t − 1 u_{j_t}=f_{j_t}u_{j_{t-1}} u j t = f j t u j t − 1 , hence u j 0 = g C ⋅ u j 0 u_{j_0}=g_C\cdot u_{j_0} u j 0 = g C ⋅ u j 0 with the cycle product g C g_C g C of Lemma 8.24 , and the value u j 0 ∈ Fix Ω ( g C ) u_{j_0}\in\operatorname{Fix}_{\Omega}(g_C) u j 0 ∈ Fix Ω ( g C ) determines u u u on C C C . Conversely every such value gives a fixed tuple. So ( f , h ) (f,h) ( f , h ) fixes ∏ C ∣ Fix Ω ( g C ) ∣ \prod_C|\operatorname{Fix}_{\Omega}(g_C)| ∏ C ∣ Fix Ω ( g C ) ∣ tuples. By Lemma 8.24 (2) and Burnside’s lemma (Theorem 4.6 ),
1 ∣ G ∣ ℓ ∑ f ∈ G ℓ ∏ C ∣ Fix Ω ( g C ) ∣ = ∏ C 1 ∣ G ∣ ∑ g ∈ G ∣ Fix Ω ( g ) ∣ = ω c ( S h ) . \frac{1}{|G|^\ell}\sum_{f\in G^\ell}\prod_C|\operatorname{Fix}_{\Omega}(g_C)|=\prod_C\frac{1}{|G|}\sum_{g\in G}|\operatorname{Fix}_{\Omega}(g)|=\omega^{c(S_h)}. ∣ G ∣ ℓ 1 f ∈ G ℓ ∑ C ∏ ∣ Fix Ω ( g C ) ∣ = C ∏ ∣ G ∣ 1 g ∈ G ∑ ∣ Fix Ω ( g ) ∣ = ω c ( S h ) .
Burnside’s lemma for G ≀ H G\wr H G ≀ H now gives 1 ∣ H ∣ ∑ h ω c ( S h ) \frac{1}{|H|}\sum_h\omega^{c(S_h)} ∣ H ∣ 1 ∑ h ω c ( S h ) .
For Ω = Y X \Omega=Y^X Ω = Y X one has ω = N G ( m ) \omega=N_G(m) ω = N G ( m ) , which is the second proof of Corollary 8.26. The two proofs are two readings of one computation: the cycle product g C g_C g C acts on a block, and the fixed colourings of that block are counted either by the cycles of g C g_C g C or by its fixed points.
De Bruijn’s group inside a wreath product
Proposition 8.28. Let G G G act on X X X and H H H on Y Y Y , with ℓ = ∣ X ∣ \ell= |X| ℓ = ∣ X ∣ . The map
( g , h ) ↦ ( ( h , h , … , h ) , g ) (g,h) \mapsto((h,h,\ldots,h),g) ( g , h ) ↦ (( h , h , … , h ) , g ) is an injective homomorphism G × H → H ≀ G G \times H \to H \wr G G × H → H ≀ G , where G G G permutes the ℓ \ell ℓ blocks through its action on X X X , and it carries the action of Theorem 4.22 on Y X Y^X Y X to the restriction of the product action of H ≀ G H \wr G H ≀ G on Y X = Y ℓ Y^X = Y^\ell Y X = Y ℓ .
Proof. A constant tuple is unchanged by any permutation of its coordinates, so by (67)
( ( h , … , h ) , g ) ( ( h ′ , … , h ′ ) , g ′ ) = ( ( h h ′ , … , h h ′ ) , g g ′ ) . ((h,\ldots,h),g)((h',\ldots,h'),g') = ((hh',\ldots,hh'),gg'). (( h , … , h ) , g ) (( h ′ , … , h ′ ) , g ′ ) = (( h h ′ , … , h h ′ ) , g g ′ ) . In the product action (69) the element acts by u ↦ ( h ⋅ u g − 1 ( x ) ) x u \mapsto(h \cdot u_{g^{-1}(x)})_x u ↦ ( h ⋅ u g − 1 ( x ) ) x , which is ( ( g , h ) ⋅ u ) ( x ) = h ( u ( g − 1 x ) ) ((g,h)\cdot u)(x) = h(u(g^{-1}x)) (( g , h ) ⋅ u ) ( x ) = h ( u ( g − 1 x )) .
Thus de Bruijn’s group is the subgroup of H ≀ G H \wr G H ≀ G whose fibre element is the same at every position. The full wreath product allows an independent element of H H H at each position and therefore has fewer orbits: by Theorem 8.27 it has N G ( ω H ) N_G(\omega_H) N G ( ω H ) , where ω H \omega_H ω H is the number of H H H -orbits on Y Y Y . For G = H = C 2 G = H = C_2 G = H = C 2 acting on X = Y = A 2 X = Y = A_2 X = Y = A 2 the subgroup has 2 orbits on the four functions X → Y X \to Y X → Y (Exercise 11) and the wreath product has one.
Examples
Example 8.29 (C 2 ≀ C 2 = D 8 C_2 \wr C_2 = D_8 C 2 ≀ C 2 = D 8 ). With Z ( C 2 ) = 1 2 ( x 1 2 + x 2 ) Z(C_2) = \frac{1}{2}(x_1^2+x_2) Z ( C 2 ) = 2 1 ( x 1 2 + x 2 ) on two points one has z 1 = 1 2 ( x 1 2 + x 2 ) z_1 = \frac{1}{2}(x_1^2+x_2) z 1 = 2 1 ( x 1 2 + x 2 ) and z 2 = 1 2 ( x 2 2 + x 4 ) z_2 = \frac{1}{2}(x_2^2+x_4) z 2 = 2 1 ( x 2 2 + x 4 ) , so
Z ( C 2 ≀ C 2 ) = 1 2 ( z 1 2 + z 2 ) = 1 8 ( x 1 4 + 2 x 1 2 x 2 + 3 x 2 2 + 2 x 4 ) . Z(C_2 \wr C_2) = \frac{1}{2}(z_1^2+z_2) = \frac{1}{8}(x_1^4+2x_1^2x_2+3x_2^2+2x_4). Z ( C 2 ≀ C 2 ) = 2 1 ( z 1 2 + z 2 ) = 8 1 ( x 1 4 + 2 x 1 2 x 2 + 3 x 2 2 + 2 x 4 ) .
the cycle index of the square in the example following Theorem 4.22 : the blocks are the two diagonals { 0 , 2 } \{0,2\} { 0 , 2 } and { 1 , 3 } \{1,3\} { 1 , 3 } , the fibre group swaps the two vertices of a diagonal, and the control swaps the diagonals. The number of orbits on m m m -colourings is N C 2 ( N C 2 ( m ) ) = q ( q + 1 ) / 2 N_{C_2}(N_{C_2}(m)) = q(q+1)/2 N C 2 ( N C 2 ( m )) = q ( q + 1 ) /2 with q = m ( m + 1 ) / 2 q = m(m+1)/2 q = m ( m + 1 ) /2 , which is 6 for m = 2 m = 2 m = 2 and 21 for m = 3 m = 3 m = 3 .
Example 8.30 (The hyperoctahedral group). The group C 2 ≀ Sym n C_2 \wr\operatorname{Sym}_n C 2 ≀ Sym n of order 2 n n ! 2^n n! 2 n n ! acts on 2 n 2n 2 n points, n n n blocks of two. By Corollary 8.26 and Corollary 11.12, the number of orbits on m m m -colourings is N Sym n ( q ) = ( q + n − 1 n ) N_{\operatorname{Sym}_n}(q) = \binom{q+n-1}{n} N Sym n ( q ) = ( n q + n − 1 ) with q = N C 2 ( m ) = m ( m + 1 ) / 2 q = N_{C_2}(m) = m(m+1)/2 q = N C 2 ( m ) = m ( m + 1 ) /2 : the orbits are the multisets of n n n unordered pairs of colours. For n = 3 n = 3 n = 3 and m = 2 m = 2 m = 2 there are ( 5 3 ) = 10 \binom{5}{3} = 10 ( 3 5 ) = 10 . More generally, for every G G G the generating function of the wreath products with symmetric groups is
∑ n ≥ 0 Z ( G ≀ Sym n ) t n = exp ( ∑ k ≥ 1 t k k Z ( G ; x k , x 2 k , … ) ) . \sum_{n \ge0} Z(G \wr\operatorname{Sym}_n)t^n = \exp\left(\sum_{k \ge1}\frac{t^k}{k}Z(G;x_k,x_{2k},\ldots)\right). n ≥ 0 ∑ Z ( G ≀ Sym n ) t n = exp ( k ≥ 1 ∑ k t k Z ( G ; x k , x 2 k , … ) ) .
by Theorem 8.25 and the exponential formula of § 19.3 .
Example 8.31 (Sylow subgroups of symmetric groups). For a prime p p p , C p ≀ C p C_p \wr C_p C p ≀ C p is a Sylow p p p -subgroup of Sym p 2 \operatorname{Sym}_{p^2} Sym p 2 . With Z ( C p ; x ) = 1 p ( x 1 p + ( p − 1 ) x p ) Z(C_p;x) = \frac{1}{p}(x_1^p+(p-1)x_p) Z ( C p ; x ) = p 1 ( x 1 p + ( p − 1 ) x p ) one gets z 1 = 1 p ( x 1 p + ( p − 1 ) x p ) z_1 = \frac{1}{p}(x_1^p+(p-1)x_p) z 1 = p 1 ( x 1 p + ( p − 1 ) x p ) and z p = 1 p ( x p p + ( p − 1 ) x p 2 ) z_p = \frac{1}{p}(x_p^p+(p-1)x_{p^2}) z p = p 1 ( x p p + ( p − 1 ) x p 2 ) , and the number of orbits on m m m -colourings of the p 2 p^2 p 2 points is 1 p ( q p + ( p − 1 ) q ) \frac{1}{p}(q^p+(p-1)q) p 1 ( q p + ( p − 1 ) q ) with q = 1 p ( m p + ( p − 1 ) m ) q = \frac{1}{p}(m^p+(p-1)m) q = p 1 ( m p + ( p − 1 ) m ) . For p = 3 p = 3 p = 3 and m = 2 m = 2 m = 2 this is 24 orbits, and 451 for m = 3 m = 3 m = 3 . Iterating, the number of orbits of the n n n -fold wreath power C p ≀ ⋯ ≀ C p C_p \wr\cdots\wr C_p C p ≀ ⋯ ≀ C p on p n p^n p n points is obtained by applying q ↦ 1 p ( q p + ( p − 1 ) q ) q \mapsto\frac{1}{p}(q^p+(p-1)q) q ↦ p 1 ( q p + ( p − 1 ) q ) to m m m a total of n n n times.
caption: Orbits of wreath products on colourings. Each group was generated explicitly in scripts/polya_verify.py; the order, the identity of Theorem 8.25 at random points, and the counts N H ( N G ( m ) ) N_H(N_G(m)) N H ( N G ( m )) were checked.
:::
Transitive actions and stabiliser chains
A transitive action labels each group element by a point and an element of the point stabiliser. Iterating this along a chain of point stabilisers gives mixed-radix labels whose radices are orbit sizes. These laws are used for the symmetric and alternating groups in Chapter 11 and for the sporadic groups in Chapter 13.
Theorem 8.32 (Transitive law). Let G G G act transitively on A n A_n A n , and let K K K be the stabiliser of 0 0 0 , coded by a group law ( A k , ∗ K ) (A_k,*_K) ( A k , ∗ K ) with identity 0 0 0 . Choose s t ∈ G s_t \in G s t ∈ G with s t ( 0 ) = t s_t(0)=t s t ( 0 ) = t and s 0 = 1 s_0=1 s 0 = 1 . Code g = s t h g=s_t h g = s t h , with h ∈ K h\in K h ∈ K of code κ \kappa κ , by m = t + n κ m=t+n\kappa m = t + nκ . Then the law
( t , h ) ∗ ( u , k ) = ( t ′ , s t ′ − 1 s t h s u k ) , t ′ = s t ( h ( u ) ) , (70) (t,h) * (u,k) = (t',s_{t'}^{-1}s_t h s_u k),\qquad t'=s_t(h(u)), \tag*{(70)} ( t , h ) ∗ ( u , k ) = ( t ′ , s t ′ − 1 s t h s u k ) , t ′ = s t ( h ( u )) , (70) makes A n k A_{nk} A nk a group isomorphic to G G G , with identity $0.
Proof. Every g ∈ G g\in G g ∈ G can be written uniquely as g = s g ( 0 ) ⋅ ( s g ( 0 ) − 1 g ) g=s_{g(0)}\cdot(s_{g(0)}^{-1}g) g = s g ( 0 ) ⋅ ( s g ( 0 ) − 1 g ) , with the second factor in K K K . Hence g ↦ ( g ( 0 ) , s g ( 0 ) − 1 g ) g\mapsto(g(0),s_{g(0)}^{-1}g) g ↦ ( g ( 0 ) , s g ( 0 ) − 1 g ) is a bijection G → A n × K G\to A_n\times K G → A n × K , and ∣ G ∣ = n k |G|=nk ∣ G ∣ = nk . For g = s t h g=s_t h g = s t h and g ′ = s u k g'=s_u k g ′ = s u k we have g g ′ ( 0 ) = s t h ( s u ( 0 ) ) = s t ( h ( u ) ) = t ′ gg'(0)=s_t h(s_u(0))=s_t(h(u))=t' g g ′ ( 0 ) = s t h ( s u ( 0 )) = s t ( h ( u )) = t ′ , so g g ′ = s t ′ ⋅ ( s t ′ − 1 s t h s u k ) gg'=s_{t'}\cdot(s_{t'}^{-1}s_t h s_u k) g g ′ = s t ′ ⋅ ( s t ′ − 1 s t h s u k ) . Thus (8.12) is the multiplication of G G G transported along this bijection. Since s 0 = 1 s_0=1 s 0 = 1 and 0 0 0 codes 1 ∈ K 1\in K 1 ∈ K , the code 0 0 0 is the identity.
Remark 8.33. If K K K is normal in G G G and the representatives s t s_t s t form a subgroup T T T , then h ( u ) = u h(u)=u h ( u ) = u for h ∈ K h\in K h ∈ K , so s t ′ = s t s u s_{t'}=s_ts_u s t ′ = s t s u and (8.12) reduces to ( t , h ) ∗ ( u , k ) = ( t ′ , s u − 1 h s u k ) (t,h)*(u,k)=(t',s_u^{-1}hs_u k) ( t , h ) ∗ ( u , k ) = ( t ′ , s u − 1 h s u k ) , the multiplication of the semidirect product G = K ⋊ T G=K\rtimes T G = K ⋊ T of Theorem 8.1 .
Theorem 8.34 (Chain law). Let G G G act faithfully on A n A_n A n , and let ( b 1 , … , b k ) (b_1,\ldots,b_k) ( b 1 , … , b k ) be a base: only the identity fixes every b l b_l b l . Put G 0 = G G_0=G G 0 = G , let G l G_l G l be the pointwise stabiliser of b 1 , … , b l b_1,\ldots,b_l b 1 , … , b l , let Δ l = G l − 1 ( b l ) \Delta_l=G_{l-1}(b_l) Δ l = G l − 1 ( b l ) and r l = ∣ Δ l ∣ r_l=|\Delta_l| r l = ∣ Δ l ∣ . List Δ l \Delta_l Δ l as δ l , 0 = b l , δ l , 1 , … , δ l , r l − 1 \delta_{l,0}=b_l,\delta_{l,1},\ldots,\delta_{l,r_l-1} δ l , 0 = b l , δ l , 1 , … , δ l , r l − 1 , and choose s l , t ∈ G l − 1 s_{l,t}\in G_{l-1} s l , t ∈ G l − 1 with s l , t ( b l ) = δ l , t s_{l,t}(b_l)=\delta_{l,t} s l , t ( b l ) = δ l , t and s l , 0 = 1 s_{l,0}=1 s l , 0 = 1 . Then every g ∈ G g\in G g ∈ G has a unique expression
g = s 1 , t 1 ∘ s 2 , t 2 ∘ ⋯ ∘ s k , t k , t l ∈ A r l , g=s_{1,t_1}\circ s_{2,t_2}\circ\cdots\circ s_{k,t_k},\qquad t_l\in A_{r_l}, g = s 1 , t 1 ∘ s 2 , t 2 ∘ ⋯ ∘ s k , t k , t l ∈ A r l , so ∣ G ∣ = r 1 ⋯ r k |G|=r_1\cdots r_k ∣ G ∣ = r 1 ⋯ r k . The digits are read off from the top: with g 1 = g g_1=g g 1 = g , t l t_l t l is the position of g l ( b l ) g_l(b_l) g l ( b l ) in Δ l \Delta_l Δ l and g l + 1 = s l , t l − 1 ∘ g l g_{l+1}=s_{l,t_l}^{-1}\circ g_l g l + 1 = s l , t l − 1 ∘ g l . The code
λ ( g ) = t 1 + r 1 ( t 2 + r 2 ( t 3 + ⋯ + r k − 1 t k ) ) \lambda(g)=t_1+r_1\bigl(t_2+r_2\bigl(t_3+\cdots+r_{k-1}t_k\bigr)\bigr) λ ( g ) = t 1 + r 1 ( t 2 + r 2 ( t 3 + ⋯ + r k − 1 t k ) ) is a bijection G → A ∣ G ∣ G\to A_{|G|} G → A ∣ G ∣ with λ ( 1 ) = 0 \lambda(1)=0 λ ( 1 ) = 0 , and m ∗ n = λ ( λ − 1 ( m ) ∘ λ − 1 ( n ) ) m*n=\lambda\bigl(\lambda^{-1}(m)\circ\lambda^{-1}(n)\bigr) m ∗ n = λ ( λ − 1 ( m ) ∘ λ − 1 ( n ) ) is a digital law of G G G .
Proof. G l − 1 G_{l-1} G l − 1 acts transitively on Δ l \Delta_l Δ l , and the stabiliser of b l b_l b l in G l − 1 G_{l-1} G l − 1 is G l G_l G l . Relabel Δ l \Delta_l Δ l by positions, so that b l b_l b l becomes 0 0 0 , and apply Theorem 8.32 to G l − 1 G_{l-1} G l − 1 with K = G l K=G_l K = G l . This writes each element of G l − 1 G_{l-1} G l − 1 uniquely as s l , t h s_{l,t}h s l , t h with h ∈ G l h\in G_l h ∈ G l , coded by t + r l κ t+r_l\kappa t + r l κ . Induction on l l l , ending with G k = 1 G_k=1 G k = 1 because the b l b_l b l form a base, gives the unique expression and the mixed-radix code. Since s l , 0 = 1 s_{l,0}=1 s l , 0 = 1 for all l l l , the identity has code 0.
Simple groups and the classification
The groups of this chapter are assembled from smaller ones: a semidirect product adds a layer of digits on top of a normal subgroup, and a wreath product stacks a whole alphabet of copies. A group that cannot be assembled in this way is simple . In digital terms, Corollary 4.41 says that G G G is simple if and only if its composition labellings have a single layer, and that G G G is solvable if and only if the radices of a composition labelling are the prime factors of N = ∣ G ∣ N = |G| N = ∣ G ∣ . By Theorem 4.39 the radices and layer codes of a composition labelling do not depend on the labelling, so the simple groups are the building blocks of all finite groups. This section uses Chapters 4 and 6, and it anticipates the later chapters, where a digital law is given for every finite simple group.
The classification
Theorem 8.35 (Classification of the finite simple groups). Every finite simple group is isomorphic to one of the following:
a cyclic group C p C_p C p of prime order (Chapter 7);
an alternating group Alt n \operatorname{Alt}_n Alt n , n ≥ 5 n \ge5 n ≥ 5 (Chapter 11);
a group of Lie type, the Tits group 2 F 4 ( 2 ) ′ {}^{2}F_4(2)' 2 F 4 ( 2 ) ′ being counted among them (Chapter 12);
one of the 26 sporadic groups (Chapter 13).
The proof is spread over several hundred articles and is not reproduced here. What the following chapters add is a digital law ( A N , ∗ ) (A_N, \ast) ( A N , ∗ ) for every entry of the list. The smallest non-Abelian simple group is Alt 5 \operatorname{Alt}_5 Alt 5 , of order 60, and every non-Abelian group in the list has even order. That this is no accident is the content of the next theorem.
Groups of odd order
Theorem 8.36 (Feit–Thompson). Every group of odd order is solvable.
By Corollary 4.41, Theorem 8.36 says that every law ( A N , ∗ ) (A_N, \ast) ( A N , ∗ ) with N N N odd can be relabelled into a mixed radix ( p 1 , … , p r ) (p_1, \ldots, p_r) ( p 1 , … , p r ) of primes, N = p 1 ⋯ p r N = p_1 \cdots p_r N = p 1 ⋯ p r , in which every truncation is a homomorphism and every layer code is a relabelling of d p i ( 0 , a + b ) d_{p_i}(0, a + b) d p i ( 0 , a + b ) . The original proof occupies a whole issue of the Pacific Journal of Mathematics [59], and it has been checked in full by computer in the Coq proof assistant [60]. Its proof is not reproduced here either; instead we record what the digital language sees of the theorem, and where it stops.
Write N = 2 q + 1 N = 2q + 1 N = 2 q + 1 . The number 2 q 2q 2 q enters first through inversion.
Proposition 8.37 (Inversion as negation). If N N N is odd, G G G has a labelling λ : G → A N \lambda: G \to A_N λ : G → A N with λ ( 1 ) = 0 \lambda(1) = 0 λ ( 1 ) = 0 and λ ( g − 1 ) = N − λ ( g ) \lambda(g^{-1}) = N - \lambda(g) λ ( g − 1 ) = N − λ ( g ) for g ≠ 1 g \ne1 g = 1 . In such a labelling the inverse of m ≠ 0 m \ne0 m = 0 is N − m N - m N − m , as in C N C_N C N , and m ↦ m ∗ m m \mapsto m \ast m m ↦ m ∗ m is a bijection of A N A_N A N .
Proof. A group of odd order has no element of order 2, so g ≠ g − 1 g \ne g^{-1} g = g − 1 for g ≠ 1 g \ne1 g = 1 , and inversion splits G ∖ { 1 } G \setminus\{1\} G ∖ { 1 } into q q q pairs { g , g − 1 } \{g,g^{-1}\} { g , g − 1 } . Label the j j j -th pair by j j j and N − j N-j N − j , 1 ≤ j ≤ q 1 \le j \le q 1 ≤ j ≤ q . Squaring is injective, because g 2 = h 2 g^2 = h^2 g 2 = h 2 gives g = ( g 2 ) q + 1 = ( h 2 ) q + 1 = h g = (g^2)^{q+1} = (h^2)^{q+1} = h g = ( g 2 ) q + 1 = ( h 2 ) q + 1 = h , using g N = 1 g^N = 1 g N = 1 . □
Lemma 8.38 (Burnside). If N N N is odd, the trivial character is the only real-valued irreducible character of G G G .
Proof. The number of real-valued irreducible characters equals the number of conjugacy classes C C C with C = C − 1 C=C^{-1} C = C − 1 , since complex conjugation permutes the rows and the columns of the character table with the same number of fixed points [3, Problem 3.16]. Suppose x g x − 1 = g − 1 xgx^{-1}=g^{-1} xg x − 1 = g − 1 . Then x 2 x^2 x 2 commutes with g g g , and since x x x has odd order, x x x is a power of x 2 x^2 x 2 ; hence x x x commutes with g g g and g = g − 1 g=g^{-1} g = g − 1 , so g = 1 g=1 g = 1 . The only real class is { 1 } \{1\} { 1 } .
Proposition 8.39 (The decomposition of q q q ). Let N = 2 q + 1 N=2q+1 N = 2 q + 1 be odd. The non-trivial irreducible characters of G G G fall into s s s pairs { χ i , χ i ‾ } \{\chi_i,\overline{\chi_i}\} { χ i , χ i } , and with d i = χ i ( 0 ) d_i=\chi_i(0) d i = χ i ( 0 ) ,
q = ∑ i = 1 s d i 2 q=\sum_{i=1}^{s}d_i^2 q = i = 1 ∑ s d i 2 where each d i d_i d i is an odd divisor of N N N . The number of conjugacy classes is k = 2 s + 1 k=2s+1 k = 2 s + 1 , and k ≡ N ( m o d 16 ) k\equiv N\pmod{16} k ≡ N ( mod 16 ) .
Proof. By Lemma 8.38, χ ≠ χ ‾ \chi\ne\overline{\chi} χ = χ for χ ≠ 1 \chi\ne1 χ = 1 , which gives the pairs. Theorem 6.11 gives N = 1 + 2 ∑ i d i 2 N=1+2\sum_i d_i^2 N = 1 + 2 ∑ i d i 2 , and d i d_i d i divides N N N by Corollary 6.14, so it is odd. The number of classes equals the number of irreducible characters, 1 + 2 s 1+2s 1 + 2 s . Odd squares are ≡ 1 ( m o d 8 ) \equiv1\pmod{8} ≡ 1 ( mod 8 ) , so q ≡ s ( m o d 8 ) q\equiv s\pmod{8} q ≡ s ( mod 8 ) and N = 2 q + 1 ≡ 2 s + 1 = k ( m o d 16 ) N=2q+1\equiv2s+1=k\pmod{16} N = 2 q + 1 ≡ 2 s + 1 = k ( mod 16 ) .
Example 8.40 (Order 21). For N = 21 N=21 N = 21 , q = 10 q=10 q = 10 . The cyclic group C 21 C_{21} C 21 gives q = 1 2 + ⋯ + 1 2 q=1^2+\cdots+1^2 q = 1 2 + ⋯ + 1 2 (s = 10 s=10 s = 10 , k = 21 k=21 k = 21 ). The metacyclic group C 7 ⋊ 2 C 3 C_7\rtimes_2 C_3 C 7 ⋊ 2 C 3 of Section 8.4 has degrees 1 , 1 , 1 , 3 , 3 1,1,1,3,3 1 , 1 , 1 , 3 , 3 , so q = 1 2 + 3 2 q=1^2+3^2 q = 1 2 + 3 2 , s = 2 s=2 s = 2 and k = 5 k=5 k = 5 . In both cases k ≡ 21 ( m o d 16 ) k\equiv21\pmod{16} k ≡ 21 ( mod 16 ) .
The commutator subgroup G ′ G' G ′ is the natural normal subgroup to construct. In the labelling of Proposition 8.37 it is generated by the labels ( N − m ) ∗ ( N − n ) ∗ m ∗ n (N-m)*(N-n)*m*n ( N − m ) ∗ ( N − n ) ∗ m ∗ n , with N − 0 N-0 N − 0 read as 0 0 0 , so it can be computed from the law alone, and ∣ G : G ′ ∣ |G:G'| ∣ G : G ′ ∣ is the number of linear characters (Chapter 6).
Theorem 8.41 (Equivalent forms of the Odd Order Theorem). The following statements are equivalent.
Every group of odd order is solvable.
Every non-trivial group of odd order satisfies G ′ ≠ G G'\ne G G ′ = G .
For every non-trivial group of odd order N = 2 q + 1 N=2q+1 N = 2 q + 1 , some d i d_i d i in Proposition 8.39 equals 1 1 1 .
Proof. (1)⇒ \Rightarrow ⇒ (2) because a solvable non-trivial group has a non-trivial Abelian quotient. (2)⇒ \Rightarrow ⇒ (1): every subgroup of a group of odd order has odd order, so the derived series G ≥ G ′ ≥ G ′ ′ ≥ ⋯ G\ge G'\ge G''\ge\cdots G ≥ G ′ ≥ G ′′ ≥ ⋯ decreases strictly until it reaches 1 1 1 . (2)⇔ \Leftrightarrow ⇔ (3): the non-trivial linear characters are non-real by Lemma 8.38, so ∣ G : G ′ ∣ = 1 + 2 # { i : d i = 1 } |G:G'|=1+2\#\{i:d_i=1\} ∣ G : G ′ ∣ = 1 + 2# { i : d i = 1 } , which exceeds 1 1 1 exactly when some d i = 1 d_i=1 d i = 1 .
What the order alone cannot see
It is tempting to look for the normal subgroup in the digits of N = 2 q + 1 N=2q+1 N = 2 q + 1 in radix 2 q 2q 2 q . This cannot work.
Proposition 8.42 (Casting out 2 q 2q 2 q ). For every x ≥ 0 x\ge0 x ≥ 0 , x ≡ ∑ k d N ( k , x ) ( m o d N − 1 ) x\equiv\sum_k d_N(k,x)\pmod{N-1} x ≡ ∑ k d N ( k , x ) ( mod N − 1 ) . If R R R is the code of any law on A N \mathcal{A}_N A N whose Cayley table is a Latin square (in particular of any group), and N = 2 q + 1 N=2q+1 N = 2 q + 1 , then d 2 q ( 0 , R ) = q d_{2q}(0,R)=q d 2 q ( 0 , R ) = q , whatever the order in which the entries are placed in R R R .
Proof. Since N ≡ 1 ( m o d N − 1 ) N\equiv1\pmod{N-1} N ≡ 1 ( mod N − 1 ) , x = ∑ k d N ( k , x ) N k = ∑ k d N ( k , x ) x=\sum_k d_N(k,x)N^k=\sum_k d_N(k,x) x = ∑ k d N ( k , x ) N k = ∑ k d N ( k , x ) . The digits of R R R are the N 2 N^2 N 2 entries of the table; each row is a permutation of A N \mathcal{A}_N A N with sum N ( N − 1 ) / 2 = N q N(N-1)/2=Nq N ( N − 1 ) /2 = N q , so the digit sum is N 2 q = ( 4 q 2 + 4 q + 1 ) q ≡ q ( m o d 2 q ) N^2q=(4q^2+4q+1)q\equiv q\pmod{2q} N 2 q = ( 4 q 2 + 4 q + 1 ) q ≡ q ( mod 2 q ) .
The only digit of R R R in radix 2 q 2q 2 q that is determined by digits in radix N N N is therefore the same for all groups of order N N N . More simply, gcd ( N , 2 q ) = 1 \gcd(N,2q)=1 g cd( N , 2 q ) = 1 , so no non-trivial subgroup has an order dividing 2 q 2q 2 q : normal subgroups correspond to factorisations N = a b N=ab N = ab and truncations in radix ( a , b ) (a,b) ( a , b ) , not to additive decompositions of N N N .
The arithmetic of Proposition 8.39 does not decide the question either. A perfect group of odd order would need a decomposition q = ∑ d i 2 q=\sum d_i^2 q = ∑ d i 2 with every d i d_i d i an odd divisor of N N N , 3 ≤ d i ≤ N 3\le d_i\le\sqrt{N} 3 ≤ d i ≤ N , and such decompositions exist:
Table 1 (PDF p. 104).
For N = 105 N=105 N = 105 the decomposition gives k = 9 ≡ 105 ( m o d 16 ) k=9\equiv105\pmod{16} k = 9 ≡ 105 ( mod 16 ) , so even Burnside’s congruence is satisfied. Yet no group of order 105 = 3 ⋅ 5 ⋅ 7 105=3\cdot5\cdot7 105 = 3 ⋅ 5 ⋅ 7 is perfect: if neither the Sylow 5-subgroup nor the Sylow 7-subgroup were normal there would be 21 of the first and 15 of the second (§ 10.6 ), hence 84 + 90 > 105 84+90>105 84 + 90 > 105 elements of order 5 or 7. Excluding the decompositions of q q q requires information about subgroups, not only about N N N and the degrees; this is the local analysis of the Feit–Thompson proof.
Cases within reach, and an arithmetic question
Two classical special cases have short proofs in the present language. If N N N is square-free, all Sylow subgroups are cyclic, G G G is metacyclic by a theorem of Hölder and Burnside [1], and its law is the two-digit rule of § 14.4 ; it is solvable by inspection. If N = p a q b N=p^aq^b N = p a q b , Burnside’s vanishing criterion of Chapter § 12 gives solvability for all N N N of this form, odd or even.
The last step of the Feit–Thompson proof is arithmetic, and it is digital in nature. For distinct primes p p p and q q q , the number ( p q − 1 ) / ( p − 1 ) = ∑ k = 0 q − 1 p k (p^q-1)/(p-1)=\sum_{k=0}^{q-1}p^k ( p q − 1 ) / ( p − 1 ) = ∑ k = 0 q − 1 p k is the repunit whose q q q digits in base p p p are all equal to 1: d p ( k , ( p q − 1 ) / ( p − 1 ) ) = 1 d_p(k,(p^q-1)/(p-1))=1 d p ( k , ( p q − 1 ) / ( p − 1 )) = 1 for 0 ≤ k < q 0\le k<q 0 ≤ k < q . Feit and Thompson [61] conjectured that this repunit never divides the repunit with p p p digits in base q q q ; a proof would shorten the final chapter of their argument considerably. The conjecture is known for q = 2 q=2 q = 2 and q = 3 q=3 q = 3 . The stronger statement that the two repunits are coprime is false: for p = 17 p=17 p = 17 and q = 3313 q=3313 q = 3313 they share the factor 2 p q + 1 = 112643 2pq+1=112643 2 pq + 1 = 112643 [62].
Exercises
Compute the Cayley table of D 6 D_6 D 6 from (63) and give an explicit isomorphism onto Sym 3 = { 5 , 7 , 11 , 15 , 19 , 21 } \operatorname{Sym}_3=\{5,7,11,15,19,21\} Sym 3 = { 5 , 7 , 11 , 15 , 19 , 21 } .
Show that C q ⋊ r C s ≅ C q ⋊ r ′ C s C_q\rtimes_r C_s\cong C_q\rtimes_{r'}C_s C q ⋊ r C s ≅ C q ⋊ r ′ C s whenever r ′ ≡ r k ( m o d q ) r'\equiv r^k\pmod q r ′ ≡ r k ( mod q ) with gcd ( k , s ) = 1 \gcd(k,s)=1 g cd( k , s ) = 1 .
Show that the centre of D 2 q D_{2q} D 2 q is trivial for odd q q q and is { 0 , q / 2 } \{0,q/2\} { 0 , q /2 } for even q q q .
Show that Q 8 Q_8 Q 8 has a unique element of order 2 and that all its subgroups are normal; deduce Q 8 ≇ D 8 Q_8\not\cong D_8 Q 8 ≅ D 8 .
Let q q q be prime and let r r r have order s > 1 s>1 s > 1 modulo q q q . Show that C q ⋊ r C s C_q\rtimes_r C_s C q ⋊ r C s has trivial centre.
Using Theorem 8.16 , show that C 13 × 3 C 3 C_{13}\times_3 C_3 C 13 × 3 C 3 has 7 conjugacy classes and character degrees 1 , 1 , 1 , 3 , 3 , 3 , 3 1,1,1,3,3,3,3 1 , 1 , 1 , 3 , 3 , 3 , 3 .
Show that the group of Theorem 8.22 has order N s N^s N s , and identify C 2 ≀ C 2 C_2\wr C_2 C 2 ≀ C 2 (with C 2 C_2 C 2 permuting two coordinates) with D 8 D_8 D 8 .
Exercises
Compute the cycle index of C 3 ≀ C 2 C_3 \wr C_2 C 3 ≀ C 2 acting on six points and show that it has 10 orbits on 2-colourings.
Show that the number of orbits of C 2 ≀ Sym n C_2 \wr\operatorname{Sym}_n C 2 ≀ Sym n on the m m m -colourings of its 2 n 2n 2 n points is ( q + n − 1 n ) \binom{q+n-1}{n} ( n q + n − 1 ) with q = m ( m + 1 ) / 2 q=m(m+1)/2 q = m ( m + 1 ) /2 .
For a prime p p p , show that C p ≀ C p C_p \wr C_p C p ≀ C p has 1 p ( q p + ( p − 1 ) q ) \frac{1}{p}(q^p+(p-1)q) p 1 ( q p + ( p − 1 ) q ) orbits on the m m m -colourings of its p 2 p^2 p 2 points, where q = 1 p ( m p + ( p − 1 ) m ) q=\frac{1}{p}(m^p+(p-1)m) q = p 1 ( m p + ( p − 1 ) m ) , and evaluate it for p = 3 p=3 p = 3 , m = 2 m=2 m = 2 .
Let C 2 × C 2 C_2 \times C_2 C 2 × C 2 act on the four functions A 2 → A 2 A_2 \to A_2 A 2 → A 2 as in Theorem 4.22 , the first factor swapping the two positions and the second complementing the values. Count the orbits with Theorem 4.22 , and compare with the orbits of C 2 ≀ C 2 C_2 \wr C_2 C 2 ≀ C 2 on the same set (Proposition 8.28 ).
Compute d 20 ( 0 , R ) d_{20}(0,R) d 20 ( 0 , R ) for the codes of C 21 C_{21} C 21 and C 7 ⋊ 2 C 3 C_7 \rtimes_2 C_3 C 7 ⋊ 2 C 3 , and check Proposition 8.42 .
List all odd N < 400 N<400 N < 400 for which q = ( N − 1 ) / 2 q=(N-1)/2 q = ( N − 1 ) /2 is a sum of squares of odd divisors 3 ≤ d ≤ N 3\le d\le\sqrt{N} 3 ≤ d ≤ N of N N N .
Construct the labelling of Proposition 8.37 for C 7 ⋊ 2 C 3 C_7 \rtimes_2 C_3 C 7 ⋊ 2 C 3 and compute G ′ G' G ′ from the law.
Show that every group of order 105 has a normal subgroup of order 35.
Small finite groups
Theorem 9.1 (Lagrange). If H H H is a subgroup of a group ( A N , R ) (\mathcal{A}_N,R) ( A N , R ) , then ∣ H ∣ |H| ∣ H ∣ divides N N N .
Proof. The left cosets λ a ( H ) \lambda_a(H) λ a ( H ) are images of H H H under the bijections λ a \lambda_a λ a (Theorem 2.15 ), so all have ∣ H ∣ |H| ∣ H ∣ elements, and two cosets are equal or disjoint.
Corollary 9.2. If p p p is prime, every group of order p p p is isomorphic to C p C_p C p .
Proof. Any a ≠ 0 a \ne0 a = 0 generates a subgroup of order greater than 1 dividing p p p , hence the whole group, and k ↦ a k k \mapsto a^k k ↦ a k is an isomorphism from C p C_p C p .
Example 9.3 (Order 4). There are two groups: C 4 C_4 C 4 , with a ∗ b = d 4 ( 0 , a + b ) a * b = d_4(0,a+b) a ∗ b = d 4 ( 0 , a + b ) , and V 4 = C 2 ⊕ C 2 V_4 = C_2 \oplus C_2 V 4 = C 2 ⊕ C 2 , with a ∗ b = d 2 ( 0 , a + b ) + 2 d 2 ( 0 , d 2 ( 1 , a ) + d 2 ( 1 , b ) ) a * b = d_2(0,a+b) + 2d_2(0,d_2(1,a)+d_2(1,b)) a ∗ b = d 2 ( 0 , a + b ) + 2 d 2 ( 0 , d 2 ( 1 , a ) + d 2 ( 1 , b )) (digitwise addition mod 2).
Example 9.4 (Order 6). There are two groups: C 6 ≅ C 3 ⊕ C 2 C_6 \cong C_3 \oplus C_2 C 6 ≅ C 3 ⊕ C 2 (Theorem 7.21 ), and the nonabelian D 6 ≅ Sym 3 D_6 \cong\operatorname{Sym}_3 D 6 ≅ Sym 3 (Corollary 8.9 with q = 3 q=3 q = 3 ).
Classification of the groups of order at most 15
This section lists every isomorphism type of group of order n ≤ 15 n \le15 n ≤ 15 , each realised on the alphabet A n \mathcal{A}_n A n with identity 0 by one of the digital laws already constructed: the cyclic law, the digitwise mixed-radix law of direct sums, the metacyclic law ((62) ), the dihedral law ((63) ), the dicyclic law ((60) ), and the semidirect law ((59) ). The only group needing a new automorphism is Alt 4 \operatorname{Alt}_4 Alt 4 .
The alternating group Alt 4 ≅ V 4 ⋊ C 3 \operatorname{Alt}_4 \cong V_4 \rtimes C_3 Alt 4 ≅ V 4 ⋊ C 3
Proposition 9.5. Write m , n ∈ A 12 m,n \in\mathcal{A}_{12} m , n ∈ A 12 in the mixed radix ( 2 , 2 , 3 ) (2,2,3) ( 2 , 2 , 3 ) ,
a 0 = d 2 ( 0 , m ) , a 1 = d 2 ( 1 , m ) , c = d 3 ( 0 , m 4 ) , a_0=d_2(0,m), \quad a_1=d_2(1,m), \quad c=d_3\left(0,\frac{m}{4}\right), a 0 = d 2 ( 0 , m ) , a 1 = d 2 ( 1 , m ) , c = d 3 ( 0 , 4 m ) , b 0 = d 2 ( 0 , n ) , b 1 = d 2 ( 1 , n ) , e = d 3 ( 0 , n 4 ) , b_0=d_2(0,n), \quad b_1=d_2(1,n), \quad e=d_3\left(0,\frac{n}{4}\right), b 0 = d 2 ( 0 , n ) , b 1 = d 2 ( 1 , n ) , e = d 3 ( 0 , 4 n ) , and let f k = d 3 ( 0 , k 2 ) f_k=d_3(0,k^2) f k = d 3 ( 0 , k 2 ) , so that f k = 0 f_k=0 f k = 0 if 3 ∣ k 3 \mid k 3 ∣ k and f k = 1 f_k=1 f k = 1 otherwise. Then the law
m ∗ n = d 2 ( 0 , a 0 + f c + 2 b 0 + f c b 1 ) + 2 d 2 ( 0 , a 1 + f c b 0 + f c + 1 b 1 ) + 4 d 3 ( 0 , c + e ) (71) m * n=d_2(0,a_0+f_{c+2}b_0+f_cb_1)+2d_2(0,a_1+f_cb_0+f_{c+1}b_1)+4d_3(0,c+e) \tag*{(71)} m ∗ n = d 2 ( 0 , a 0 + f c + 2 b 0 + f c b 1 ) + 2 d 2 ( 0 , a 1 + f c b 0 + f c + 1 b 1 ) + 4 d 3 ( 0 , c + e ) (71) makes A 12 \mathcal{A}_{12} A 12 a group isomorphic to Alt 4 ≅ V 4 ⋊ C 3 \operatorname{Alt}_4 \cong V_4 \rtimes C_3 Alt 4 ≅ V 4 ⋊ C 3 . The two lower layers carry the Klein group V 4 = C 2 ⊕ C 2 V_4=C_2\oplus C_2 V 4 = C 2 ⊕ C 2 (digitwise addition modulo 2), the top layer carries C 3 C_3 C 3 , and the coefficients f f f are the action of C 3 C_3 C 3 on V 4 V_4 V 4 : Eq. (71) is m ∗ n = ( d 4 ( 0 , m ) ⊕ ϕ c ( d 4 ( 0 , n ) ) ) + 4 d 3 ( 0 , c + e ) m*n=(d_4(0,m)\oplus\phi^c(d_4(0,n)))+4d_3(0,c+e) m ∗ n = ( d 4 ( 0 , m ) ⊕ ϕ c ( d 4 ( 0 , n ))) + 4 d 3 ( 0 , c + e ) , where ⊕ \oplus ⊕ is the bitwise exclusive-or on A 4 \mathcal{A}_4 A 4 and
ϕ = ( 0 1 2 3 0 2 3 1 ) \phi=\begin{pmatrix}0&1&2&3\\0&2&3&1\end{pmatrix} ϕ = ( 0 0 1 2 2 3 3 1 ) is the linear map ( v 0 , v 1 ) ↦ ( v 1 , v 0 + v 1 ) (v_0,v_1)\mapsto(v_1,v_0+v_1) ( v 0 , v 1 ) ↦ ( v 1 , v 0 + v 1 ) of F 2 2 \mathbb{F}_2^2 F 2 2 written on the code v 0 + 2 v 1 v_0+2v_1 v 0 + 2 v 1 .
Proof. On the coordinates ( v 0 , v 1 ) (v_0,v_1) ( v 0 , v 1 ) the map ϕ \phi ϕ has matrix M = ( 0 1 1 1 ) M=\begin{pmatrix}0&1\\1&1\end{pmatrix} M = ( 0 1 1 1 ) over F 2 \mathbb{F}_2 F 2 . Since M 2 = M + I M^2=M+I M 2 = M + I , its powers are
M c = ( F c − 1 F c F c F c + 1 ) M^c=\begin{pmatrix}F_{c-1}&F_c\\F_c&F_{c+1}\end{pmatrix} M c = ( F c − 1 F c F c F c + 1 )
with F k F_k F k the Fibonacci numbers, and modulo 2 these have period 3 with values 0, 1, 1, that is, F k ≡ f k F_k\equiv f_k F k ≡ f k and F c − 1 ≡ f c + 2 F_{c-1}\equiv f_{c+2} F c − 1 ≡ f c + 2 . Hence the two lower digits of (71) are d 4 ( 0 , m ) ⊕ ϕ c ( d 4 ( 0 , n ) ) d_4(0,m)\oplus\phi^c(d_4(0,n)) d 4 ( 0 , m ) ⊕ ϕ c ( d 4 ( 0 , n )) and the two forms agree. As M 3 = I M^3=I M 3 = I and M M M is invertible, c ↦ ϕ c c\mapsto\phi^c c ↦ ϕ c is a homomorphism C 3 → Aut ( V 4 ) C_3\to\operatorname{Aut}(V_4) C 3 → Aut ( V 4 ) , and Theorem 8.1 gives a group of order 12. Its normal subgroup A 4 A_4 A 4 (the elements with c = 0 c=0 c = 0 ) is a Klein group. The eight elements with c ≠ 0 c\ne0 c = 0 have order 3, because M 2 + M + I = 0 M^2+M+I=0 M 2 + M + I = 0 over F 2 \mathbb{F}_2 F 2 . A group of order 12 with eight elements of order 3 is Alt 4 \operatorname{Alt}_4 Alt 4 by Theorem 9.10 below.
Remark 9.6 (The same law as a code). For a group this small the action can also be stored in a single code, by the coding principle. Writing the three permutations ϕ 0 , ϕ 1 , ϕ 2 \phi^0,\phi^1,\phi^2 ϕ 0 , ϕ 1 , ϕ 2 one after the other in base 4 (digits 0123, 0231, 0312 from position 0),
Φ = ∑ c = 0 2 ∑ v = 0 3 ϕ c ( v ) 4 4 c + v = 10254564 , ϕ c ( v ) = d 4 ( 4 c + v , Φ ) . \Phi=\sum_{c=0}^{2}\sum_{v=0}^{3}\phi^c(v)4^{4c+v}=10254564,\qquad\phi^c(v)=d_4(4c+v,\Phi). Φ = c = 0 ∑ 2 v = 0 ∑ 3 ϕ c ( v ) 4 4 c + v = 10254564 , ϕ c ( v ) = d 4 ( 4 c + v , Φ ) .
With the exclusive-or written digitally, u ⊕ w = ∑ j = 0 1 2 j d 2 ( 0 , d 2 ( j , u ) + d 2 ( j , w ) ) u\oplus w=\sum_{j=0}^{1}2^j d_2(0,d_2(j,u)+d_2(j,w)) u ⊕ w = ∑ j = 0 1 2 j d 2 ( 0 , d 2 ( j , u ) + d 2 ( j , w )) , the law becomes
m ∗ n = ∑ j = 0 1 2 j d 2 ( 0 , d 2 ( j , m ) + d 2 ( j , w ) ) + 4 d 3 ( 0 , c + e ) , w = d 4 ( 4 c + d 4 ( 0 , n ) , Φ ) . m*n=\sum_{j=0}^{1}2^j d_2(0,d_2(j,m)+d_2(j,w))+4d_3(0,c+e),\qquad w=d_4(4c+d_4(0,n),\Phi). m ∗ n = j = 0 ∑ 1 2 j d 2 ( 0 , d 2 ( j , m ) + d 2 ( j , w )) + 4 d 3 ( 0 , c + e ) , w = d 4 ( 4 c + d 4 ( 0 , n ) , Φ ) .
This form looks the action up in a table instead of computing it; Eq. (71) shows its arithmetic, which is what generalises.
Classification theorems
Theorem 9.7 (Order p 2 p^2 p 2 ). If ∣ G ∣ = p 2 |G|=p^2 ∣ G ∣ = p 2 , then G ≅ C p 2 G\cong C_{p^2} G ≅ C p 2 or G ≅ C p ⊕ C p G\cong C_p\oplus C_p G ≅ C p ⊕ C p .
Proof. The class equation gives Z ( G ) ≠ { 0 } Z(G)\ne\{0\} Z ( G ) = { 0 } . If ∣ Z ( G ) ∣ = p |Z(G)|=p ∣ Z ( G ) ∣ = p , then G / Z ( G ) G/Z(G) G / Z ( G ) is cyclic, which forces G G G Abelian, a contradiction. Hence G G G is Abelian, and the classification of finite Abelian groups gives the two types.
Theorem 9.8 (Order p q pq pq ). Let p < q p<q p < q be primes. If p ∤ q − 1 p\nmid q-1 p ∤ q − 1 , then G ≅ C p q G\cong C_{pq} G ≅ C pq . If p ∣ q − 1 p\mid q-1 p ∣ q − 1 , then G G G is either C p q C_{pq} C pq or the nonabelian group C q ⋊ r C p C_q\rtimes_r C_p C q ⋊ r C p of Theorem 8.8 , where r r r has multiplicative order p p p modulo q q q ; this group does not depend on the choice of r r r .
Proof. By Sylow’s theorems n q ≡ 1 ( m o d q ) n_q\equiv1\pmod q n q ≡ 1 ( mod q ) and n q ∣ p < q n_q\mid p<q n q ∣ p < q , so n q = 1 n_q=1 n q = 1 , the Sylow subgroup C q C_q C q is normal, and G ≅ C q ⋊ ϕ C p G\cong C_q\rtimes_\phi C_p G ≅ C q ⋊ ϕ C p with ϕ : C p → Aut ( C q ) ≅ C q − 1 \phi:C_p\to\operatorname{Aut}(C_q)\cong C_{q-1} ϕ : C p → Aut ( C q ) ≅ C q − 1 . If p ∤ q − 1 p\nmid q-1 p ∤ q − 1 , ϕ \phi ϕ is trivial. Otherwise the nontrivial homomorphisms send a generator to the elements r , r 2 , … , r p − 1 r,r^2,\ldots,r^{p-1} r , r 2 , … , r p − 1 of order p p p ; replacing the generator of C p C_p C p by its k k k th power turns r r r into r k r^k r k , so all of them give isomorphic groups.
For p = 2 p=2 p = 2 the nonabelian case is the dihedral group D 2 q D_{2q} D 2 q of Corollary 8.9. This covers n = 6 , 10 , 14 n=6,10,14 n = 6 , 10 , 14 . Since 3 ∤ 4 3\nmid4 3 ∤ 4 , every group of order 15 is cyclic.
Theorem 9.9 (Order 8). There are exactly five groups of order 8: C 8 C_8 C 8 , C 4 ⊕ C 2 C_4\oplus C_2 C 4 ⊕ C 2 , C 2 ⊕ C 2 ⊕ C 2 C_2\oplus C_2\oplus C_2 C 2 ⊕ C 2 ⊕ C 2 , D 8 = C 4 ⋊ 3 C 2 D_8=C_4\rtimes_3 C_2 D 8 = C 4 ⋊ 3 C 2 and Q 8 Q_8 Q 8 .
Proof. The Abelian types come from the partitions of 3. Let G G G be nonabelian. Then no element has order 8, and not every element has order ≤ 2 \le2 ≤ 2 (otherwise G G G is Abelian). Take a a a of order 4 and b ∉ ⟨ a ⟩ b\notin\langle a\rangle b ∈ / ⟨ a ⟩ . The subgroup ⟨ a ⟩ \langle a\rangle ⟨ a ⟩ has index 2, so it is normal, and conjugation by b b b is an automorphism of C 4 C_4 C 4 . It is not the identity, since then G G G would be Abelian, so b a b − 1 = a − 1 bab^{-1}=a^{-1} ba b − 1 = a − 1 . Next b 2 ∈ ⟨ a ⟩ b^2\in\langle a\rangle b 2 ∈ ⟨ a ⟩ , and b 2 ≠ a ± 1 b^2\ne a^{\pm1} b 2 = a ± 1 because b b b would then have order 8, so b 2 ∈ { 0 , a 2 } b^2\in\{0,a^2\} b 2 ∈ { 0 , a 2 } . If some b ∉ ⟨ a ⟩ b\notin\langle a\rangle b ∈ / ⟨ a ⟩ has b 2 = 0 b^2=0 b 2 = 0 , the relations are those of D 8 D_8 D 8 (Proposition 8.11). Otherwise b 2 = a 2 b^2=a^2 b 2 = a 2 , which gives Q 8 Q_8 Q 8 . The two groups are not isomorphic: D 8 D_8 D 8 has five involutions and Q 8 Q_8 Q 8 has one.
Theorem 9.10 (Order 12). There are exactly five groups of order 12: C 12 C_{12} C 12 , C 6 ⊕ C 2 C_{6} \oplus C_{2} C 6 ⊕ C 2 , D 12 = C 6 ⋊ 5 C 2 D_{12} = C_{6} \rtimes_{5} C_{2} D 12 = C 6 ⋊ 5 C 2 , Alt 4 \operatorname{Alt}_{4} Alt 4 (Proposition 9.5 ) and Q 12 = C 3 ⋊ 2 C 4 Q_{12} = C_{3} \rtimes_{2} C_{4} Q 12 = C 3 ⋊ 2 C 4 .
Proof. By Sylow’s theorems n 3 ∈ { 1 , 4 } n_{3} \in\{1,4\} n 3 ∈ { 1 , 4 } . If n 3 = 4 n_{3}=4 n 3 = 4 , the four Sylow 3-subgroups contain eight elements of order 3. The remaining four elements form the unique, hence normal, Sylow 2-subgroup S S S , so G ≅ S ⋊ ϕ C 3 G \cong S \rtimes_{\phi} C_{3} G ≅ S ⋊ ϕ C 3 . Here ϕ \phi ϕ is nontrivial, since otherwise the Sylow 3-subgroup would be normal. Because Aut ( C 4 ) ≅ C 2 \operatorname{Aut}(C_{4}) \cong C_{2} Aut ( C 4 ) ≅ C 2 has no element of order 3, S = V 4 S=V_{4} S = V 4 . The two nontrivial choices ϕ ± 1 \phi^{\pm1} ϕ ± 1 differ by inverting the generator of C 3 C_{3} C 3 , so G ≅ Alt 4 G \cong\operatorname{Alt}_{4} G ≅ Alt 4 .
If n 3 = 1 n_{3}=1 n 3 = 1 , then G ≅ C 3 ⋊ ψ S G \cong C_{3} \rtimes_{\psi} S G ≅ C 3 ⋊ ψ S with S ∈ { C 4 , V 4 } S \in\{C_{4},V_{4}\} S ∈ { C 4 , V 4 } and ψ : S → Aut ( C 3 ) ≅ C 2 \psi:S \to\operatorname{Aut}(C_{3}) \cong C_{2} ψ : S → Aut ( C 3 ) ≅ C 2 . For S = C 4 S=C_{4} S = C 4 , the trivial ψ \psi ψ gives C 12 C_{12} C 12 and the surjective one gives C 3 ⋊ 2 C 4 = Q 12 C_{3} \rtimes_{2} C_{4}=Q_{12} C 3 ⋊ 2 C 4 = Q 12 . For S = V 4 S=V_{4} S = V 4 , the trivial ψ \psi ψ gives C 6 ⊕ C 2 C_{6} \oplus C_{2} C 6 ⊕ C 2 . The three surjective ψ \psi ψ are permuted transitively by Aut ( V 4 ) ≅ Sym 3 \operatorname{Aut}(V_{4}) \cong\operatorname{Sym}_{3} Aut ( V 4 ) ≅ Sym 3 , so they give one group, D 6 × C 2 ≅ D 12 D_{6} \times C_{2} \cong D_{12} D 6 × C 2 ≅ D 12 . The element-order profiles of Table 9.2 show that the five groups are pairwise nonisomorphic.
Table 9.2. Element-order profiles separating the groups of orders 8 and 12.
Remark 9.11. The group C 3 ⋊ 2 C 4 C_{3} \rtimes_{2} C_{4} C 3 ⋊ 2 C 4 is the dicyclic group Q 4 ⋅ 3 Q_{4\cdot3} Q 4 ⋅ 3 of Theorem 8.2 : both are nonabelian, of order 12, with a normal C 3 C_{3} C 3 and a cyclic Sylow 2-subgroup. The dicyclic law (60) with q = 3 q=3 q = 3 and the metacyclic law (62) with ( q , s , r ) = ( 3 , 4 , 2 ) (q,s,r)=(3,4,2) ( q , s , r ) = ( 3 , 4 , 2 ) are two digital codings of the same group.
Catalogue
Table 9.1. All groups of order n ≤ 15 n \le15 n ≤ 15 up to isomorphism (28 groups in total).
Groups of orders 16 to 23
There are 31 groups of order 16 ≤ n ≤ 23 16 \le n \le23 16 ≤ n ≤ 23 : fourteen of order 16, five each of orders 18 and 20, two each of orders 21 and 22, and the cyclic groups of prime orders 17, 19 and 23. Each group carries its identifier [ n , k ] [n,k] [ n , k ] , meaning SmallGroup(n,k) in the Small Groups Library [13]. All identifiers in this section were certified by computing IdGroup in GAP 4.14.0 [30] on the right-regular permutation representation of each digit table.
Semidirect products by a cyclic group
Corollary 9.12. Let G 0 = ( A p 0 , ∗ 0 ) G_0=(\mathcal{A}_{p_0},\ast_0) G 0 = ( A p 0 , ∗ 0 ) be a group with identity 0, let s ≥ 1 s \ge1 s ≥ 1 , and let α ∈ Aut ( G 0 ) \alpha\in\operatorname{Aut}(G_0) α ∈ Aut ( G 0 ) satisfy α s = id \alpha^s=\operatorname{id} α s = id . With m 1 = d s ( 0 , m / p 0 ) m_1=d_s(0,m/p_0) m 1 = d s ( 0 , m / p 0 ) and n 1 = d s ( 0 , n / p 0 ) n_1=d_s(0,n/p_0) n 1 = d s ( 0 , n / p 0 ) , the law
m ∗ n = ( d p 0 ( 0 , m ) ∗ 0 α m 1 ( d p 0 ( 0 , n ) ) ) + p 0 d s ( 0 , m 1 + n 1 ) (72) m \ast n = \left(d_{p_0}(0,m)\ast_0\alpha^{m_1}(d_{p_0}(0,n))\right)+p_0d_s(0,m_1+n_1) \tag*{(72)} m ∗ n = ( d p 0 ( 0 , m ) ∗ 0 α m 1 ( d p 0 ( 0 , n )) ) + p 0 d s ( 0 , m 1 + n 1 ) (72) makes A p 0 s \mathcal{A}_{p_0s} A p 0 s a group G 0 ⋊ α C s G_0\rtimes_\alpha C_s G 0 ⋊ α C s . For α = id \alpha=\operatorname{id} α = id it is the direct product G 0 × C s G_0 \times C_s G 0 × C s .
Proof. Apply Theorem 8.1 with G 1 = C s G_1=C_s G 1 = C s and ϕ c = α c \phi_c=\alpha^c ϕ c = α c . This is a homomorphism C s → Aut ( G 0 ) C_s \to\operatorname{Aut}(G_0) C s → Aut ( G 0 ) because α s = id \alpha^s=\operatorname{id} α s = id .
Three new automorphisms are needed:
σ = ( 0 1 2 3 0 2 1 3 ) \sigma=\begin{pmatrix}0&1&2&3\\0&2&1&3\end{pmatrix} σ = ( 0 0 1 2 2 1 3 3 ) on V 4 V_4 V 4 , which exchanges the two binary digits, so σ 2 = id \sigma^2=\operatorname{id} σ 2 = id . With s = 4 s=4 s = 4 it gives V 4 ⋊ σ C 4 ≅ ( C 4 ⊕ C 2 ) ⋊ C 2 V_4\rtimes_\sigma C_4 \cong(C_4\oplus C_2)\rtimes C_2 V 4 ⋊ σ C 4 ≅ ( C 4 ⊕ C 2 ) ⋊ C 2 , the group [ 16 , 3 ] [16,3] [ 16 , 3 ] .
τ ( a + 4 b ) = d 4 ( 0 , a + 2 b ) + 4 b \tau(a+4b)=d_4(0,a+2b)+4b τ ( a + 4 b ) = d 4 ( 0 , a + 2 b ) + 4 b on C 4 ⊕ C 2 C_4\oplus C_2 C 4 ⊕ C 2 , coded on A 8 \mathcal{A}_8 A 8 . It is additive and bijective, and τ 2 = id \tau^2=\operatorname{id} τ 2 = id because 4 b ≡ 0 ( m o d 4 ) 4b\equiv0\pmod{4} 4 b ≡ 0 ( mod 4 ) . With s = 2 s=2 s = 2 it gives C 4 ∘ D 8 C_4\circ D_8 C 4 ∘ D 8 , the group [ 16 , 13 ] [16,13] [ 16 , 13 ] .
ν ( a + 3 b ) = d 3 ( 0 , − a ) + 3 d 3 ( 0 , − b ) \nu(a+3b)=d_3(0,-a)+3d_3(0,-b) ν ( a + 3 b ) = d 3 ( 0 , − a ) + 3 d 3 ( 0 , − b ) , inversion on C 3 ⊕ C 3 C_3\oplus C_3 C 3 ⊕ C 3 . With s = 2 s=2 s = 2 it gives the generalised dihedral group ( C 3 ⊕ C 3 ) ⋊ − 1 C 2 (C_3\oplus C_3)\rtimes_{-1} C_2 ( C 3 ⊕ C 3 ) ⋊ − 1 C 2 , the group [ 18 , 4 ] [18,4] [ 18 , 4 ] .
Remark 9.13. In ( C 4 ⊕ C 2 ) ⋊ τ C 2 (C_4\oplus C_2)\rtimes_\tau C_2 ( C 4 ⊕ C 2 ) ⋊ τ C 2 the element x = 1 x=1 x = 1 is central of order 4. The elements y = 4 y=4 y = 4 and t = 8 t=8 t = 8 are involutions with t y t − 1 = τ ( y ) = y ∗ x 2 tyt^{-1}=\tau(y)=y\ast x^2 t y t − 1 = τ ( y ) = y ∗ x 2 , so ( y ∗ t ) 2 = x 2 (y\ast t)^2=x^2 ( y ∗ t ) 2 = x 2 . Hence ⟨ y , t ⟩ ≅ D 8 \langle y,t\rangle\cong D_8 ⟨ y , t ⟩ ≅ D 8 meets ⟨ x ⟩ \langle x\rangle ⟨ x ⟩ in ⟨ x 2 ⟩ \langle x^2\rangle ⟨ x 2 ⟩ , and the group is the central product C 4 ∘ D 8 ≅ C 4 ∘ Q 8 C_4\circ D_8\cong C_4\circ Q_8 C 4 ∘ D 8 ≅ C 4 ∘ Q 8 (the Pauli group).
Order 16
Theorem 9.14 (Order 16). There are exactly 14 groups of order 16 [12]: five Abelian and nine nonabelian. They are realised by the digital laws of Table 9.3 , and the invariants listed there show that they are pairwise nonisomorphic.
Table 9.3. The 14 groups of order 16. Here Z Z Z is the centre, G ′ G' G ′ the derived subgroup and G ( 2 ) = { g ∗ g } G^{(2)}=\{g*g\} G ( 2 ) = { g ∗ g } the set of squares. Metacyclic parameters are ( q , s , r ) (q,s,r) ( q , s , r ) .
Partial proof. The Abelian types correspond to the five partitions of 4. A nonabelian G G G of order 16 has exponent 4 or 8. Suppose the exponent is 8; let a a a have order 8 and b ∉ ⟨ a ⟩ b\notin\langle a\rangle b ∈ / ⟨ a ⟩ . Then ⟨ a ⟩ \langle a\rangle ⟨ a ⟩ is normal of index 2, and b a b − 1 = a r bab^{-1}=a^r ba b − 1 = a r with r ∈ { 3 , 5 , 7 } r\in\{3,5,7\} r ∈ { 3 , 5 , 7 } . Moreover b 2 ∈ ⟨ a ⟩ b^2\in\langle a\rangle b 2 ∈ ⟨ a ⟩ commutes with b b b , so it is fixed by x ↦ x r x\mapsto x^r x ↦ x r . Since r 2 ≡ 1 ( m o d 8 ) r^2\equiv1\pmod{8} r 2 ≡ 1 ( mod 8 ) , replacing b b b by b a k ba^k b a k changes b 2 b^2 b 2 into b 2 a ( r + 1 ) k b^2a^{(r+1)k} b 2 a ( r + 1 ) k .
r = 3 r=3 r = 3 : the fixed points are { 0 , a 4 } \{0,a^4\} { 0 , a 4 } and a 4 k a^{4k} a 4 k reaches both, so one may take b 2 = 0 b^2=0 b 2 = 0 . This is S D 16 SD_{16} S D 16 .
r = 5 r=5 r = 5 : the fixed points are the even powers and a 6 k a^{6k} a 6 k reaches all of them, so again b 2 = 0 b^2=0 b 2 = 0 . This is M 16 M_{16} M 16 .
r = 7 r=7 r = 7 : a 8 k = 0 a^{8k}=0 a 8 k = 0 , so b 2 ∈ { 0 , a 4 } b^2\in\{0,a^4\} b 2 ∈ { 0 , a 4 } is an invariant. The two cases give D 16 D_{16} D 16 and Q 16 Q_{16} Q 16 .
The five nonabelian groups of exponent 4 are [ 16 , 3 ] [16,3] [ 16 , 3 ] , [ 16 , 4 ] [16,4] [ 16 , 4 ] , [ 16 , 11 ] [16,11] [ 16 , 11 ] , [ 16 , 12 ] [16,12] [ 16 , 12 ] and [ 16 , 13 ] [16,13] [ 16 , 13 ] ; for the proof that there are no others we refer to [12]. Each law in Table 9.3 has been checked to satisfy the group axioms, and the invariant columns are pairwise distinct.
Orders 17 to 23
By Corollary 9.2, the only groups of orders 17, 19 and 23 are cyclic. By Theorem 9.8 , order 21 = 3 ⋅ 7 21=3\cdot7 21 = 3 ⋅ 7 gives C 21 C_{21} C 21 and C 7 ⋊ 2 C 3 C_7\rtimes_2 C_3 C 7 ⋊ 2 C 3 , since 3 ∣ 6 3\mid6 3 ∣ 6 and 2 has order 3 modulo 7. Order 22 = 2 ⋅ 11 22=2\cdot11 22 = 2 ⋅ 11 gives C 22 C_{22} C 22 and D 22 D_{22} D 22 .
Theorem 9.15 (Order 18). There are exactly five groups of order 18: C 18 C_{18} C 18 , C 6 ⊕ C 3 C_6\oplus C_3 C 6 ⊕ C 3 , D 18 D_{18} D 18 , C 3 × Sym 3 ≅ C 3 ⋊ 2 C 6 C_3\times\operatorname{Sym}_3\cong C_3\rtimes_2 C_6 C 3 × Sym 3 ≅ C 3 ⋊ 2 C 6 and ( C 3 ⊕ C 3 ) ⋊ ν C 2 (C_3\oplus C_3)\rtimes_{\nu} C_2 ( C 3 ⊕ C 3 ) ⋊ ν C 2 .
Proof. By Sylow’s theorems n 3 ∣ 2 n_3\mid2 n 3 ∣ 2 and n 3 ≡ 1 ( m o d 3 ) n_3\equiv1\pmod3 n 3 ≡ 1 ( mod 3 ) , so the Sylow 3-subgroup P P P is normal and G ≅ P ⋊ ψ C 2 G\cong P\rtimes_{\psi}C_2 G ≅ P ⋊ ψ C 2 , where ψ ( 1 ) ∈ Aut ( P ) \psi(1)\in\operatorname{Aut}(P) ψ ( 1 ) ∈ Aut ( P ) has order at most 2. Conjugate choices give isomorphic groups. If P = C 9 P=C_9 P = C 9 , then Aut ( C 9 ) ≅ C 6 \operatorname{Aut}(C_9)\cong C_6 Aut ( C 9 ) ≅ C 6 has the single involution x ↦ − x x\mapsto-x x ↦ − x , which gives C 18 C_{18} C 18 and D 18 D_{18} D 18 . If P = C 3 ⊕ C 3 P=C_3\oplus C_3 P = C 3 ⊕ C 3 , then an involution of GL 2 ( F 3 ) \operatorname{GL}_2(\mathbb{F}_3) GL 2 ( F 3 ) is diagonalizable with eigenvalues ± 1 \pm1 ± 1 , so it is conjugate to diag ( 1 , − 1 ) \operatorname{diag}(1,-1) diag ( 1 , − 1 ) or to − I -I − I . The identity gives C 6 ⊕ C 3 C_6\oplus C_3 C 6 ⊕ C 3 , diag ( 1 , − 1 ) \operatorname{diag}(1,-1) diag ( 1 , − 1 ) gives C 3 × Sym 3 C_3\times\operatorname{Sym}_3 C 3 × Sym 3 , and − I = ν -I=\nu − I = ν gives the generalised dihedral group.
Theorem 9.16 (Order 20). There are exactly five groups of order 20: C 20 C_{20} C 20 , C 10 ⊕ C 2 C_{10}\oplus C_2 C 10 ⊕ C 2 , D 20 D_{20} D 20 , Q 20 = C 5 ⋊ 4 C 4 Q_{20}=C_5\rtimes_4 C_4 Q 20 = C 5 ⋊ 4 C 4 and the Frobenius group F 20 = C 5 ⋊ 2 C 4 F_{20}=C_5\rtimes_2 C_4 F 20 = C 5 ⋊ 2 C 4 .
Proof. Since n 5 ∣ 4 n_5\mid4 n 5 ∣ 4 and n 5 ≡ 1 ( m o d 5 ) n_5\equiv1\pmod5 n 5 ≡ 1 ( mod 5 ) , C 5 C_5 C 5 is normal and G ≅ C 5 ⋊ ψ S G\cong C_5\rtimes_{\psi} S G ≅ C 5 ⋊ ψ S with S ∈ { C 4 , V 4 } S\in\{C_4,V_4\} S ∈ { C 4 , V 4 } and ψ : S → Aut ( C 5 ) ≅ C 4 = ⟨ 2 ⟩ \psi:S\to\operatorname{Aut}(C_5)\cong C_4=\langle2\rangle ψ : S → Aut ( C 5 ) ≅ C 4 = ⟨ 2 ⟩ . For S = C 4 S=C_4 S = C 4 , the generator acts by r ∈ { 1 , 2 , 3 , 4 } r\in\{1,2,3,4\} r ∈ { 1 , 2 , 3 , 4 } : r = 1 r=1 r = 1 gives C 20 C_{20} C 20 , r = 4 r=4 r = 4 gives Q 20 Q_{20} Q 20 , and r = 2 , 3 r=2,3 r = 2 , 3 give the same group F 20 F_{20} F 20 , because 3 ≡ 2 3 3\equiv2^3 3 ≡ 2 3 amounts to replacing the generator by its cube. For S = V 4 S=V_4 S = V 4 , ψ \psi ψ has image of order at most 2. The trivial ψ \psi ψ gives C 10 ⊕ C 2 C_{10}\oplus C_2 C 10 ⊕ C 2 ; the three surjections onto C 2 C_2 C 2 are permuted by Aut ( V 4 ) \operatorname{Aut}(V_4) Aut ( V 4 ) and give D 10 × C 2 ≅ D 20 D_{10}\times C_2\cong D_{20} D 10 × C 2 ≅ D 20 .
Remark 9.17. The law C 5 ⋊ 4 C 4 C_5\rtimes_4 C_4 C 5 ⋊ 4 C 4 is a second digital coding of the dicyclic group Q 4 ⋅ 5 Q_{4\cdot5} Q 4 ⋅ 5 of Theorem 8.2 . Likewise C 10 ⋊ 9 C 2 ≅ D 10 × C 2 C_{10}\rtimes_9 C_2\cong D_{10}\times C_2 C 10 ⋊ 9 C 2 ≅ D 10 × C 2 and C 3 ⋊ 2 C 6 ≅ C 3 × Sym 3 C_3\rtimes_2 C_6\cong C_3\times\operatorname{Sym}_3 C 3 ⋊ 2 C 6 ≅ C 3 × Sym 3 ; all three isomorphisms have been checked explicitly on the digit tables.
Groups of orders 24 to 31
There are 34 groups of order 24 ≤ n ≤ 31 24\le n\le31 24 ≤ n ≤ 31 : fifteen of order 24, five of order 27, four each of orders 28 and 30, two each of orders 25 and 26, and the cyclic groups of prime orders 29 and 31. All identifiers in this section were certified by computing IdGroup in GAP 4.14.0 [30] on the right-regular permutation representation of each digit table.
Table 4. All 17 groups of orders 17 to 23, with the invariants of Table 9.3 .
Three further digital laws
Proposition 9.18 (The symmetric group Sym 4 \operatorname{Sym}_4 Sym 4 ). On V 4 = ( A 4 , ⊕ ) V_4=(\mathcal{A}_4,\oplus) V 4 = ( A 4 , ⊕ ) let ρ = ( 0 1 2 3 0 2 3 1 ) \rho=\begin{pmatrix}0&1&2&3\\0&2&3&1\end{pmatrix} ρ = ( 0 0 1 2 2 3 3 1 ) and σ = ( 0 1 2 3 0 2 1 3 ) \sigma=\begin{pmatrix}0&1&2&3\\0&2&1&3\end{pmatrix} σ = ( 0 0 1 2 2 1 3 3 ) . Code D 6 D_6 D 6 by c = a + 3 γ c=a+3\gamma c = a + 3 γ as in Corollary 8.9, and put φ c = ρ a ∘ σ γ \varphi_c=\rho^a\circ\sigma^\gamma φ c = ρ a ∘ σ γ . With m 1 = ⌊ m / 4 ⌋ m_1=\lfloor m/4\rfloor m 1 = ⌊ m /4 ⌋ and n 1 = ⌊ n / 4 ⌋ n_1=\lfloor n/4\rfloor n 1 = ⌊ n /4 ⌋ , the law
m ∗ n = ( d 4 ( 0 , m ) ⊕ φ m 1 ( d 4 ( 0 , n ) ) ) + 4 ( m 1 ∗ D 6 n 1 ) (73) m*n=\left(d_4(0,m)\oplus\varphi_{m_1}(d_4(0,n))\right)+4(m_1*_{D_6}n_1) \tag*{(73)} m ∗ n = ( d 4 ( 0 , m ) ⊕ φ m 1 ( d 4 ( 0 , n )) ) + 4 ( m 1 ∗ D 6 n 1 ) (73) makes A 24 \mathcal{A}_{24} A 24 a group isomorphic to Sym 4 \operatorname{Sym}_4 Sym 4 .
Proof. Both ρ \rho ρ and σ \sigma σ permute the nonzero elements of V 4 V_4 V 4 , so they are automorphisms, and σ ρ σ − 1 = ρ − 1 \sigma\rho\sigma^{-1}=\rho^{-1} σ ρ σ − 1 = ρ − 1 . Hence ( a , γ ) ↦ ρ a σ γ (a,\gamma)\mapsto\rho^a\sigma^\gamma ( a , γ ) ↦ ρ a σ γ is a homomorphism, in fact an isomorphism D 6 → Aut ( V 4 ) ≅ Sym 3 D_6\to\operatorname{Aut}(V_4)\cong\operatorname{Sym}_3 D 6 → Aut ( V 4 ) ≅ Sym 3 , and Theorem 8.1 gives V 4 ⋊ Aut ( V 4 ) V_4\rtimes\operatorname{Aut}(V_4) V 4 ⋊ Aut ( V 4 ) . In Sym 4 \operatorname{Sym}_4 Sym 4 the double transpositions together with the identity form a normal Klein subgroup. The stabiliser of a point is a complement isomorphic to Sym 3 \operatorname{Sym}_3 Sym 3 , and it permutes the three double transpositions faithfully, so it induces all of Aut ( V 4 ) \operatorname{Aut}(V_4) Aut ( V 4 ) . Hence Sym 4 ≅ V 4 ⋊ Aut ( V 4 ) \operatorname{Sym}_4\cong V_4\rtimes\operatorname{Aut}(V_4) Sym 4 ≅ V 4 ⋊ Aut ( V 4 ) .
Proposition 9.19 (SL ( 2 , 3 ) ≅ Q 8 ⋊ C 3 \operatorname{SL}(2,3)\cong Q_8\rtimes C_3 SL ( 2 , 3 ) ≅ Q 8 ⋊ C 3 ). In the coding of Q 8 Q_8 Q 8 given by Theorem 8.2 with q = 2 q=2 q = 2 , let
α = ( 0 1 2 3 4 5 6 7 0 4 2 6 5 1 7 3 ) . \alpha=\begin{pmatrix}0&1&2&3&4&5&6&7\\0&4&2&6&5&1&7&3\end{pmatrix}. α = ( 0 0 1 4 2 2 3 6 4 5 5 1 6 7 7 3 ) . Then α ∈ Aut ( Q 8 ) \alpha\in\operatorname{Aut}(Q_8) α ∈ Aut ( Q 8 ) and α 3 = id \alpha^3=\operatorname{id} α 3 = id . It cycles 1 ↦ 4 ↦ 5 ↦ 1 1\mapsto4\mapsto5\mapsto1 1 ↦ 4 ↦ 5 ↦ 1 , and hence permutes cyclically the three cyclic subgroups of order 4. The law (72) with G 0 = Q 8 G_0=Q_8 G 0 = Q 8 , s = 3 s=3 s = 3 and this α \alpha α makes A 24 \mathcal{A}_{24} A 24 a group isomorphic to SL ( 2 , 3 ) \operatorname{SL}(2,3) SL ( 2 , 3 ) .
Proof. α \alpha α is the extension of x = 1 ↦ 4 x=1\mapsto4 x = 1 ↦ 4 , y = 4 ↦ 5 = x ∗ y y=4\mapsto5=x*y y = 4 ↦ 5 = x ∗ y to Q 8 = ⟨ x , y ⟩ Q_8=\langle x,y\rangle Q 8 = ⟨ x , y ⟩ ; one checks on the digit table that it is a bijective homomorphism with α 3 = id \alpha^3=\operatorname{id} α 3 = id , and Corollary 9.12 applies. In SL ( 2 , 3 ) \operatorname{SL}(2,3) SL ( 2 , 3 ) the elements of order dividing 4 form a normal quaternion subgroup, and an element of order 3 acts on it nontrivially. The elements of order 3 in Aut ( Q 8 ) ≅ Sym 4 \operatorname{Aut}(Q_8)\cong\operatorname{Sym}_4 Aut ( Q 8 ) ≅ Sym 4 are conjugate up to inversion, so every nontrivial Q 8 ⋊ C 3 Q_8\rtimes C_3 Q 8 ⋊ C 3 is SL ( 2 , 3 ) \operatorname{SL}(2,3) SL ( 2 , 3 ) .
Proposition 9.20 (The group C 3 ⋊ D 8 C_3\rtimes D_8 C 3 ⋊ D 8 ). With m 1 = ⌊ m / 3 ⌋ m_1=\lfloor m/3\rfloor m 1 = ⌊ m /3 ⌋ , n 1 = ⌊ n / 3 ⌋ n_1=\lfloor n/3\rfloor n 1 = ⌊ n /3 ⌋ and ∗ D 8 *_{D_8} ∗ D 8 the law (63) for q = 4 q=4 q = 4 ,
m ∗ n = d 3 ( 0 , m + ( − 1 ) m 1 n 1 ) + 3 ( m 1 ∗ D 8 n 1 ) (74) m*n=d_3(0,m+(-1)^{m_1}n_1)+3(m_1*_{D_8}n_1) \tag*{(74)} m ∗ n = d 3 ( 0 , m + ( − 1 ) m 1 n 1 ) + 3 ( m 1 ∗ D 8 n 1 ) (74) makes A 24 \mathcal{A}_{24} A 24 a group C 3 ⋊ D 8 C_3\rtimes D_8 C 3 ⋊ D 8 . In it the Klein subgroup { 0 , 2 , 4 , 6 } \{0,2,4,6\} { 0 , 2 , 4 , 6 } of D 8 D_8 D 8 centralises C 3 C_3 C 3 , and the other four elements of D 8 D_8 D 8 invert it.
Proof. In (63) with q = 4 q=4 q = 4 , the lower digit of m 1 ∗ D 8 n 1 m_1 *_{D_8} n_1 m 1 ∗ D 8 n 1 is congruent to a 1 ± a 2 a_1 \pm a_2 a 1 ± a 2 modulo 4 4 4 , so its parity is the sum of the parities. Since the code c = a + 4 γ c=a+4\gamma c = a + 4 γ has the parity of a a a , the map c ↦ ( − 1 ) c c\mapsto(-1)^c c ↦ ( − 1 ) c is a homomorphism D 8 → Aut ( C 3 ) D_8\to\operatorname{Aut}(C_3) D 8 → Aut ( C 3 ) . Theorem 8.1 then applies.
Remark 9.21. The exceptional laws were checked against explicit realisations. Law (73) was compared with the permutations of A 4 A_4 A 4 , the law of Proposition 9.19 with the 24 matrices of determinant 1 over F 3 \mathbb{F}_3 F 3 , and the Heisenberg law of Theorem 9.23 with the unitriangular 3 × 3 3\times3 3 × 3 matrices over F 3 \mathbb{F}_3 F 3 . In each case an explicit isomorphism was found by search over the images of a generating set.
Order 24
Theorem 9.22 (Order 24). There are exactly 15 groups of order 24, realised by the digital laws of Table 9.5 .
Table 9.5. The 15 groups of order 24, with the invariants of Table 9.3 .
Proof. Let P P P be a Sylow 3-subgroup, so that n 3 ∈ { 1 , 4 } n_3\in\{1,4\} n 3 ∈ { 1 , 4 } .
Case n 3 = 1 n_3=1 n 3 = 1 . Then G ≅ C 3 ⋊ φ S G\cong C_3\rtimes_\varphi S G ≅ C 3 ⋊ φ S with ∣ S ∣ = 8 |S|=8 ∣ S ∣ = 8 and φ : S → Aut ( C 3 ) ≅ C 2 \varphi:S\to\operatorname{Aut}(C_3)\cong C_2 φ : S → Aut ( C 3 ) ≅ C 2 . Such a φ \varphi φ is determined by its kernel, a subgroup of index at most 2, and kernels related by Aut ( S ) \operatorname{Aut}(S) Aut ( S ) give isomorphic groups. Running through the five groups of Theorem 9.9 :
S = C 8 S=C_8 S = C 8 : kernel S S S gives C 24 C_{24} C 24 , and kernel C 4 C_4 C 4 gives C 3 ⋊ 2 C 8 C_3\rtimes_2 C_8 C 3 ⋊ 2 C 8 .
S = C 4 ⊕ C 2 S=C_4\oplus C_2 S = C 4 ⊕ C 2 : kernel S S S gives C 12 ⊕ C 2 C_{12}\oplus C_2 C 12 ⊕ C 2 . The two cyclic subgroups of index 2 are exchanged by Aut ( S ) \operatorname{Aut}(S) Aut ( S ) and give C 4 × Sym 3 C_4\times\operatorname{Sym}_3 C 4 × Sym 3 , and the Klein kernel gives C 2 × Q 12 C_2\times Q_{12} C 2 × Q 12 .
S = C 2 3 S=C_2^3 S = C 2 3 : kernel S S S gives C 6 ⊕ C 2 ⊕ C 2 C_6\oplus C_2\oplus C_2 C 6 ⊕ C 2 ⊕ C 2 . The seven subgroups of index 2 are permuted transitively by GL 3 ( F 2 ) \operatorname{GL}_3(\mathbb{F}_2) GL 3 ( F 2 ) and give C 2 × D 12 ≅ C 2 2 × Sym 3 C_2\times D_{12}\cong C_2^2\times\operatorname{Sym}_3 C 2 × D 12 ≅ C 2 2 × Sym 3 .
S = D 8 S=D_8 S = D 8 : kernel S S S gives C 3 × D 8 C_3\times D_8 C 3 × D 8 , and the rotation subgroup C 4 C_4 C 4 gives D 24 D_{24} D 24 . The two Klein subgroups are exchanged by an outer automorphism and give C 3 ⋊ D 8 C_3\rtimes D_8 C 3 ⋊ D 8 (Proposition 9.20 ).
S = Q 8 S=Q_8 S = Q 8 : kernel S S S gives C 3 × Q 8 C_3\times Q_8 C 3 × Q 8 . The three cyclic subgroups of order 4 are permuted by Aut ( Q 8 ) \operatorname{Aut}(Q_8) Aut ( Q 8 ) and give C 3 ⋊ Q 8 ≅ Q 24 C_3\rtimes Q_8\cong Q_{24} C 3 ⋊ Q 8 ≅ Q 24 .
This case yields eleven groups.
Case n 3 = 4 n_3=4 n 3 = 4 . Conjugation on the four Sylow 3-subgroups gives a homomorphism G → Sym 4 G\to\operatorname{Sym}_4 G → Sym 4 whose kernel K K K lies in N G ( P ) N_G(P) N G ( P ) , a subgroup of order 6. If 3 divided ∣ K ∣ |K| ∣ K ∣ , the unique Sylow 3-subgroup of K K K would be characteristic in K K K , hence normal in G G G , contradicting n 3 = 4 n_3=4 n 3 = 4 . So ∣ K ∣ ∈ { 1 , 2 } |K|\in\{1,2\} ∣ K ∣ ∈ { 1 , 2 } . If K = 1 K=1 K = 1 , then G ≅ Sym 4 G\cong\operatorname{Sym}_4 G ≅ Sym 4 . If ∣ K ∣ = 2 |K|=2 ∣ K ∣ = 2 , then K K K is central and G / K G/K G / K has order 12 and four Sylow 3-subgroups, so G / K ≅ Alt 4 G/K\cong\operatorname{Alt}_4 G / K ≅ Alt 4 by Theorem 9.10 . The preimage S S S of the Klein subgroup of Alt 4 \operatorname{Alt}_4 Alt 4 is a normal Sylow 2-subgroup of order 8, so G ≅ S ⋊ C 3 G\cong S\rtimes C_3 G ≅ S ⋊ C 3 , with nontrivial action because P P P is not normal. Of the groups of order 8, only C 2 3 C_2^3 C 2 3 and Q 8 Q_8 Q 8 have automorphisms of order 3; the automorphism groups of C 8 C_8 C 8 , C 4 ⊕ C 2 C_4\oplus C_2 C 4 ⊕ C 2 and D 8 D_8 D 8 are 2-groups.
For C 2 3 C_2^3 C 2 3 , the elements of order 3 in GL 3 ( F 2 ) \operatorname{GL}_3(\mathbb{F}_2) GL 3 ( F 2 ) form one conjugacy class, each fixing a line and acting irreducibly on a complementary plane. This gives C 2 × Alt 4 C_2\times\operatorname{Alt}_4 C 2 × Alt 4 .
For Q 8 Q_8 Q 8 , Proposition 9.19 gives SL ( 2 , 3 ) \operatorname{SL}(2,3) SL ( 2 , 3 ) .
This gives 11 + 1 + 2 = 15 11+1+2=15 11 + 1 + 2 = 15 groups. The invariant columns of Table 9.5 are pairwise distinct within each order, so these groups are pairwise nonisomorphic.
Orders 25 to 31
By Corollary 9.2, the only groups of orders 29 and 31 are cyclic. By Theorem 9.7, order 25 gives C 25 C_{25} C 25 and C 5 ⊕ C 5 C_5\oplus C_5 C 5 ⊕ C 5 . By Theorem 9.8, order 26 gives C 26 C_{26} C 26 and D 26 D_{26} D 26 .
Theorem 9.23 (Order 27). Let β ( a + 3 b ) = d 3 ( 0 , a + b ) + 3 b \beta(a+3b)=d_3(0,a+b)+3b β ( a + 3 b ) = d 3 ( 0 , a + b ) + 3 b on C 3 ⊕ C 3 C_3\oplus C_3 C 3 ⊕ C 3 , i.e. the matrix ( 0 1 1 1 ) \begin{pmatrix}0&1\\1&1\end{pmatrix} ( 0 1 1 1 ) , so that β 3 = id \beta^3=\operatorname{id} β 3 = id . There are exactly five groups of order 27: C 27 C_{27} C 27 , C 9 ⊕ C 3 C_9\oplus C_3 C 9 ⊕ C 3 , C 3 3 C_3^3 C 3 3 , the Heisenberg group He 3 = ( C 3 ⊕ C 3 ) ⋊ β C 3 \operatorname{He}_3=(C_3\oplus C_3)\rtimes_\beta C_3 He 3 = ( C 3 ⊕ C 3 ) ⋊ β C 3 of exponent 3, and C 9 ⋊ 4 C 3 C_9\rtimes_4 C_3 C 9 ⋊ 4 C 3 of exponent 9 [12].
Partial proof. The Abelian types correspond to the partitions of 3. A nonabelian G G G has ∣ Z ( G ) ∣ = 3 |Z(G)|=3 ∣ Z ( G ) ∣ = 3 and G / Z ( G ) ≅ C 3 ⊕ C 3 G/Z(G)\cong C_3\oplus C_3 G / Z ( G ) ≅ C 3 ⊕ C 3 . Suppose G G G has an element a a a of order 9, and take b ∉ ⟨ a ⟩ b\notin\langle a\rangle b ∈ / ⟨ a ⟩ . Then ⟨ a ⟩ \langle a\rangle ⟨ a ⟩ is normal of index 3, and b a b − 1 = a r bab^{-1}=a^r ba b − 1 = a r with r ∈ { 4 , 7 } r\in\{4,7\} r ∈ { 4 , 7 } . Since 7 ≡ 4 2 ( m o d 9 ) 7\equiv4^2\pmod9 7 ≡ 4 2 ( mod 9 ) , replacing b b b by b 2 b^2 b 2 gives r = 4 r=4 r = 4 . The power b 3 ∈ ⟨ a ⟩ b^3\in\langle a\rangle b 3 ∈ ⟨ a ⟩ is fixed by x ↦ x 4 x\mapsto x^4 x ↦ x 4 , so it lies in ⟨ a 3 ⟩ \langle a^3\rangle ⟨ a 3 ⟩ . Replacing b b b by b a k ba^k b a k multiplies b 3 b^3 b 3 by a 3 k a^{3k} a 3 k , so one may take b 3 = 0 b^3=0 b 3 = 0 , which gives C 9 ⋊ 4 C 3 C_9\rtimes_4 C_3 C 9 ⋊ 4 C 3 . If G G G has exponent 3, it is the Heisenberg group; for this case we refer to [12].
Theorem 9.24 (Order 28). There are exactly four groups of order 28: C 28 C_{28} C 28 , C 14 ⊕ C 2 C_{14}\oplus C_2 C 14 ⊕ C 2 , D 28 D_{28} D 28 and Q 28 = C 7 ⋊ 6 C 4 Q_{28}=C_7\rtimes_6 C_4 Q 28 = C 7 ⋊ 6 C 4 .
Proof. Since n 7 ∣ 4 n_7\mid4 n 7 ∣ 4 and n 7 ≡ 1 ( m o d 7 ) n_7\equiv1\pmod7 n 7 ≡ 1 ( mod 7 ) , C 7 C_7 C 7 is normal and G ≅ C 7 ⋊ ψ S G\cong C_7\rtimes_\psi S G ≅ C 7 ⋊ ψ S with S ∈ { C 4 , V 4 } S\in\{C_4,V_4\} S ∈ { C 4 , V 4 } and ψ : S → Aut ( C 7 ) ≅ C 6 \psi:S\to\operatorname{Aut}(C_7)\cong C_6 ψ : S → Aut ( C 7 ) ≅ C 6 , whose image has order at most 2. For S = C 4 S=C_4 S = C 4 , the trivial ψ \psi ψ gives C 28 C_{28} C 28 and the surjection onto { ± 1 } \{\pm1\} { ± 1 } gives Q 28 Q_{28} Q 28 . For S = V 4 S=V_4 S = V 4 , the trivial ψ \psi ψ gives C 14 ⊕ C 2 C_{14}\oplus C_2 C 14 ⊕ C 2 , and the three nontrivial ψ \psi ψ are permuted by Aut ( V 4 ) \operatorname{Aut}(V_4) Aut ( V 4 ) and give D 14 × C 2 ≅ D 28 D_{14}\times C_2\cong D_{28} D 14 × C 2 ≅ D 28 .
Theorem 9.25 (Order 30). The groups of order 30 are exactly the four metacyclic groups C 15 ⋊ r C 2 C_{15}\rtimes_r C_2 C 15 ⋊ r C 2 with r ∈ { 1 , 4 , 11 , 14 } r\in\{1,4,11,14\} r ∈ { 1 , 4 , 11 , 14 } : C 30 C_{30} C 30 , C 3 × D 10 C_3\times D_{10} C 3 × D 10 ( r = 4 ) (r=4) ( r = 4 ) , C 5 × Sym 3 C_5\times\operatorname{Sym}_3 C 5 × Sym 3 ( r = 11 ) (r=11) ( r = 11 ) and D 30 D_{30} D 30 ( r = 14 ) (r=14) ( r = 14 ) .
Proof. We have n 3 ∈ { 1 , 10 } n_3\in\{1,10\} n 3 ∈ { 1 , 10 } and n 5 ∈ { 1 , 6 } n_5\in\{1,6\} n 5 ∈ { 1 , 6 } . If both were large, there would be 20 elements of order 3 and 24 of order 5, more than 30. Hence P 3 P_3 P 3 or P 5 P_5 P 5 is normal, so P 3 P 5 P_3P_5 P 3 P 5 is a subgroup of order 15. It is cyclic by Theorem 9.8, since 3 ∤ 4 3\nmid4 3 ∤ 4 , and normal because it has index 2. Thus G ≅ C 15 ⋊ r C 2 G\cong C_{15}\rtimes_r C_2 G ≅ C 15 ⋊ r C 2 with r 2 ≡ 1 ( m o d 15 ) r^2\equiv1\pmod{15} r 2 ≡ 1 ( mod 15 ) , i.e. r ≡ ± 1 r\equiv\pm1 r ≡ ± 1 modulo 3 and modulo 5, which gives r ∈ { 1 , 4 , 11 , 14 } r\in\{1,4,11,14\} r ∈ { 1 , 4 , 11 , 14 } . The four groups have centers of orders 30, 3, 5 and 1, so they are distinct.
Remark 9.26. Several of these laws are second digital codings of groups met before. One has C 7 ⋊ 6 C 4 ≅ Q 4 ⋅ 7 C_7\rtimes_6 C_4\cong Q_{4\cdot7} C 7 ⋊ 6 C 4 ≅ Q 4 ⋅ 7 (dicyclic law), C 14 ⋊ 13 C 2 ≅ D 14 × C 2 C_{14}\rtimes_{13}C_2\cong D_{14}\times C_2 C 14 ⋊ 13 C 2 ≅ D 14 × C 2 , C 15 ⋊ 11 C 2 ≅ D 6 × C 5 C_{15}\rtimes_{11}C_2\cong D_6\times C_5 C 15 ⋊ 11 C 2 ≅ D 6 × C 5 , C 15 ⋊ 4 C 2 ≅ D 10 × C 3 C_{15}\rtimes_4 C_2\cong D_{10}\times C_3 C 15 ⋊ 4 C 2 ≅ D 10 × C 3 , and the dicyclic Q 24 Q_{24} Q 24 is isomorphic to C 3 × Q 8 C_3\times Q_8 C 3 × Q 8 with kernel ⟨ 1 ⟩ \langle1\rangle ⟨ 1 ⟩ . All of these isomorphisms were found explicitly on the digit tables. The invariant columns of Table 9.6 are pairwise distinct within each order, so these groups are pairwise nonisomorphic.
Table 9.6. All 19 groups of orders 25 to 31, with the invariants of Table Table 9.3 .
Groups of order 32
There are 51 groups of order 32 [13]: 7 Abelian, 26 of nilpotency class 2, 15 of class 3 and 3 of maximal class 4. This section gives a digital law for each of them. Besides the laws already available, two new constructions are needed: central extensions given by a 2-cocycle, and a semidirect product with the nonabelian complement Q 8 Q_8 Q 8 . Every law below was checked to satisfy the group axioms, and its identifier [32, k k k ] was certified by computing IdGroup in GAP 4.14.0 [30] on its right-regular permutation representation.
Central extensions
Theorem 9.27 (Central extensions). Let H = ( A h , ∗ H ) H = (\mathcal{A}_h, *_H) H = ( A h , ∗ H ) be a group with identity 0, let z ≥ 1 z \ge1 z ≥ 1 , and let f : A h × A h → A z f : \mathcal{A}_h \times\mathcal{A}_h \to\mathcal{A}_z f : A h × A h → A z satisfy f ( 0 , b ) = f ( a , 0 ) = 0 f(0,b) = f(a,0) = 0 f ( 0 , b ) = f ( a , 0 ) = 0 and the cocycle identity
f ( a , b ) + f ( a ∗ H b , c ) ≡ f ( b , c ) + f ( a , b ∗ H c ) ( m o d z ) . f(a,b) + f(a *_H b,c) \equiv f(b,c) + f(a,b *_H c) \pmod{z}. f ( a , b ) + f ( a ∗ H b , c ) ≡ f ( b , c ) + f ( a , b ∗ H c ) ( mod z ) . With m 0 = d z ( 0 , m ) m_0 = d_z(0,m) m 0 = d z ( 0 , m ) and m 1 = ⌊ m / z ⌋ m_1 = \lfloor m/z \rfloor m 1 = ⌊ m / z ⌋ , the law
m ∗ n = d z ( 0 , m 0 + n 0 + f ( m 1 , n 1 ) ) + z ( m 1 ∗ H n 1 ) (75) m * n = d_z(0,m_0+n_0+f(m_1,n_1)) + z\,(m_1 *_H n_1) \tag*{(75)} m ∗ n = d z ( 0 , m 0 + n 0 + f ( m 1 , n 1 )) + z ( m 1 ∗ H n 1 ) (75) makes A z h \mathcal{A}_{zh} A z h a group. The lower digits form a central subgroup isomorphic to C z C_z C z , with quotient isomorphic to H H H , and every central extension of H H H by C z C_z C z is isomorphic to a group of this form.
Proof. Both ( m ∗ n ) ∗ k (m*n)*k ( m ∗ n ) ∗ k and m ∗ ( n ∗ k ) m*(n*k) m ∗ ( n ∗ k ) have upper digit m 1 ∗ H n 1 ∗ H k 1 m_1 *_H n_1 *_H k_1 m 1 ∗ H n 1 ∗ H k 1 . Their lower digits are m 0 + n 0 + k 0 + f ( m 1 , n 1 ) + f ( m 1 ∗ H n 1 , k 1 ) m_0+n_0+k_0+f(m_1,n_1)+f(m_1 *_H n_1,k_1) m 0 + n 0 + k 0 + f ( m 1 , n 1 ) + f ( m 1 ∗ H n 1 , k 1 ) and m 0 + n 0 + k 0 + f ( n 1 , k 1 ) + f ( m 1 , n 1 ∗ H k 1 ) m_0+n_0+k_0+f(n_1,k_1)+f(m_1,n_1 *_H k_1) m 0 + n 0 + k 0 + f ( n 1 , k 1 ) + f ( m 1 , n 1 ∗ H k 1 ) modulo z z z , which agree by the cocycle identity. By the normalisation, 0 is an identity and the elements with m 1 = 0 m_1 = 0 m 1 = 0 commute with every element. The element with digits ( d z ( 0 , − m 0 − f ( m 1 , m 1 − 1 ) ) , m 1 − 1 ) (d_z(0,-m_0-f(m_1,m_1^{-1})),m_1^{-1}) ( d z ( 0 , − m 0 − f ( m 1 , m 1 − 1 )) , m 1 − 1 ) is a right inverse of m m m , so the finite monoid is a group. Conversely, given a central extension 1 → C z → G → H → 1 1 \to C_z \to G \to H \to1 1 → C z → G → H → 1 , choose a section s s s with s ( 0 ) = 1 s(0) = 1 s ( 0 ) = 1 . Then s ( a ) s ( b ) = f ( a , b ) s ( a ∗ H b ) s(a)s(b) = f(a,b)s(a *_H b) s ( a ) s ( b ) = f ( a , b ) s ( a ∗ H b ) defines a normalised cocycle, and ( c , a ) ↦ c s ( a ) (c,a) \mapsto cs(a) ( c , a ) ↦ cs ( a ) is an isomorphism from (9.5) onto G G G . ◻
The group C4:Q8
Proposition 9.28. With m 1 = ⌊ m / 4 ⌋ m_1=\lfloor m/4\rfloor m 1 = ⌊ m /4 ⌋ , n 1 = ⌊ n / 4 ⌋ n_1=\lfloor n/4\rfloor n 1 = ⌊ n /4 ⌋ , ϵ ( c ) = d 2 ( 0 , c / 4 ) \epsilon(c)=d_2(0,c/4) ϵ ( c ) = d 2 ( 0 , c /4 ) and ∗ Q 8 \ast_{Q_8} ∗ Q 8 the law (60) with q = 2 q=2 q = 2 , the law
m ∗ n = d 4 ( 0 , m + ( − 1 ) ϵ ( m 1 ) n ) + 4 ( m 1 ∗ Q 8 n 1 ) (76) m\ast n=d_4\left(0,m+(-1)^{\epsilon(m_1)}n\right)+4(m_1\ast_{Q_8}n_1) \tag*{(76)} m ∗ n = d 4 ( 0 , m + ( − 1 ) ϵ ( m 1 ) n ) + 4 ( m 1 ∗ Q 8 n 1 ) (76) makes A 32 \mathcal{A}_{32} A 32 a group C_4\rtimes Q_8\cong\[32,35\] . In it the cyclic subgroup { 0 , 1 , 2 , 3 } \{0,1,2,3\} { 0 , 1 , 2 , 3 } of Q 8 Q_8 Q 8 centralises C 4 C_4 C 4 , and the other four elements of Q 8 Q_8 Q 8 invert it.
Proof. In (60) the upper digit ϵ \epsilon ϵ is added modulo 2, so c ↦ ( − 1 ) ϵ ( c ) c\mapsto(-1)^{\epsilon(c)} c ↦ ( − 1 ) ϵ ( c ) is a homomorphism Q 8 → Aut ( C 4 ) Q_8\to\operatorname{Aut}(C_4) Q 8 → Aut ( C 4 ) . Theorem 8.1 then applies.
Groups with a cyclic subgroup of index 2
Proposition 9.29. The groups of order 32 with a cyclic subgroup of index 2 are C 32 C_{32} C 32 , C 16 ⊕ C 2 C_{16}\oplus C_2 C 16 ⊕ C 2 , M_{32}=C_{16}\rtimes_9 C_2=\[32,17\] , D_{32}=\[32,18\] , SD_{32}=C_{16}\rtimes_7 C_2=\[32,19\] and Q_{32}=\[32,20\] . The last three are exactly the groups of maximal class.
Proof. Let a a a have order 16 and b ∉ ⟨ a ⟩ b\notin\langle a\rangle b ∈ / ⟨ a ⟩ , in a nonabelian G G G . Then b a b − 1 = a r bab^{-1}=a^r ba b − 1 = a r with r ∈ { 7 , 9 , 15 } r\in\{7,9,15\} r ∈ { 7 , 9 , 15 } , the units of order 2 modulo 16, and b 2 ∈ ⟨ a ⟩ b^2\in\langle a\rangle b 2 ∈ ⟨ a ⟩ is fixed by x ↦ x r x\mapsto x^r x ↦ x r . Since r 2 ≡ 1 ( m o d 16 ) r^2\equiv1\pmod{16} r 2 ≡ 1 ( mod 16 ) , replacing b b b by b a k ba^k b a k turns b 2 b^2 b 2 into b 2 a ( r + 1 ) k b^2a^{(r+1)k} b 2 a ( r + 1 ) k .
r = 7 r=7 r = 7 : the fixed points are { 0 , a 8 } \{0,a^8\} { 0 , a 8 } and a 8 k a^{8k} a 8 k reaches both, so b 2 = 0 b^2=0 b 2 = 0 . This is S D 32 SD_{32} S D 32 .
r = 9 r=9 r = 9 : the fixed points are the even powers and a 10 k a^{10k} a 10 k reaches all of them, so b 2 = 0 b^2=0 b 2 = 0 . This is M 32 M_{32} M 32 .
r = 15 r=15 r = 15 : a 16 k = 0 a^{16k}=0 a 16 k = 0 , so b 2 ∈ { 0 , a 8 } b^2\in\{0,a^8\} b 2 ∈ { 0 , a 8 } is invariant. The two cases give D 32 D_{32} D 32 and Q 32 Q_{32} Q 32 .
The identifiers were certified with IdGroup for the laws (62) with ( 16 , 2 , 9 ) (16,2,9) ( 16 , 2 , 9 ) , ( 16 , 2 , 7 ) (16,2,7) ( 16 , 2 , 7 ) , the law (63) with q = 16 q=16 q = 16 , and the law (60) with q = 8 q=8 q = 8 . The nilpotency classes in Table 9.7 show that \[32,18\] , \[32,19\] and \[32,20\] are the only groups of class 4.
Table 9.7. Groups of order 32, part I. Structure descriptions follow GAP, with “.” denoting a nonsplit extension.
Catalogue
Theorem 9.30 (Order 32). There are exactly 51 groups of order 32 [13]. Each is realised on A 32 \mathcal{A}_{32} A 32 by one of the following digital laws (Table 9.7 –Table 9.11 ):
the seven Abelian groups by direct sums;
39 groups by N ⋊ α C s N\rtimes_\alpha C_s N ⋊ α C s (Corollary 9.12 ), with N N N a law of order 32 / s 32/s 32/ s from the earlier sections and α \alpha α listed in Table 9.9 and Table 9.10 ;
Q 32 Q_{32} Q 32 by the dicyclic law (60) with q = 8 q=8 q = 8 ;
C 4 ⋊ Q 8 C_4\rtimes Q_8 C 4 ⋊ Q 8 by (76) ;
\[32,8\] , \[32,15\] and \[32,32\] by the central-extension law (75) with z = 2 z=2 z = 2 , over H=\[16,3\] , \[16,4\] and \[16,13\] respectively, with the cocycles of Table 9.11 .
Proof. For the number 51 we refer to the classification of the groups of order 2 5 2^5 2 5 [13]. Each law is an instance of Theorem 8.1 , Corollary 9.12 , Theorem 8.2 , Proposition 9.28 or Theorem 9.27 , so it defines a group. The data were obtained as follows.
For a normal subgroup N N N with cyclic quotient of order s s s and a complement ⟨ t ⟩ \langle t\rangle ⟨ t ⟩ , α \alpha α is conjugation by t t t , transported to the digital coding of N N N by an explicit isomorphism.
Table 9.9. Automorphisms for the laws N ⋊ α C s N \rtimes_{\alpha} C_s N ⋊ α C s of order 32, part I.
Table 9.10. Automorphisms for the laws N ⋊ α C 2 N\mathbin{\rtimes_\alpha} C_2 N ⋊ α C 2 of order 32, part II.
Table 9.11. Cocycles f ( a , ⋅ ) f(a,\mathord{\cdot}) f ( a , ⋅ ) for the central-extension laws of order 32.
In every case IdGroup applied to the resulting digit table returns [ 32 , k ] [32,k] [ 32 , k ] .
Remark 9.31. The five groups [ 32 , 8 ] [32,8] [ 32 , 8 ] , [ 32 , 15 ] [32,15] [ 32 , 15 ] , [ 32 , 20 ] [32,20] [ 32 , 20 ] , [ 32 , 32 ] [32,32] [ 32 , 32 ] and [ 32 , 35 ] [32,35] [ 32 , 35 ] are exactly those with no normal subgroup with cyclic quotient that has a complement. The group [ 32 , 15 ] [32,15] [ 32 , 15 ] is metacyclic, but none of its cyclic normal subgroups with cyclic quotient is complemented. In Table 9.11 the row of a a a lists f ( a , 0 ) f ( a , 1 ) ⋯ f ( a , 15 ) f(a,0)f(a,1)\cdots f(a,15) f ( a , 0 ) f ( a , 1 ) ⋯ f ( a , 15 ) .
Exercises
Show that every group of order p 2 p^2 p 2 , p p p prime, is Abelian.
Show that a group of order 2 p 2p 2 p , p p p an odd prime, is isomorphic to C 2 p C_{2p} C 2 p or to D 2 p D_{2p} D 2 p .
Count the elements of each order in the five groups of order 8 and show that these counts distinguish them.
Show that Alt 4 \operatorname{Alt}_4 Alt 4 has no subgroup of order 6.
Show that there are exactly five groups of order 12, namely C 12 C_{12} C 12 , C 2 ⊕ C 6 C_2 \oplus C_6 C 2 ⊕ C 6 , Alt 4 \operatorname{Alt}_4 Alt 4 , D 12 D_{12} D 12 and Q 12 Q_{12} Q 12 .
Using the small groups library of GAP [30, 13], confirm that there are 14, 15 and 51 groups of orders 16, 24 and 32, and compare with the tables of this chapter.
Table 9.8. Groups of order 32, part II.
Crystallographic groups
This chapter attaches finite groups and digital laws to the 17 plane groups and the 230 space groups of crystallography [39]. § 17.1 explains why the point group is not enough; § 17.2 and § 17.5 give the digital laws of the quotients modulo a sublattice; § 17.4 and § 17.6 show that these finite groups separate the crystallographic groups; and § 17.7 identifies them as abstract finite groups.
Quotients modulo a sublattice
The 17 plane groups and the 230 space groups of crystallography are infinite: each contains its lattice T ≅ Z n T \cong\mathbb{Z}^{n} T ≅ Z n (n = 2 , 3 n=2,3 n = 2 , 3 ) of translations as a normal subgroup, and only the quotient P = G / T P=G/T P = G / T , the point group, is finite. Passing to P P P forgets the glide reflections and the screw rotations, whose translation parts do not lie in T T T : the 17 plane groups have only 10 point groups (9 abstract groups), and the 230 space groups only 32 crystal classes (18 abstract groups). The finite quotients G / m T G/mT G / m T keep this information, and they carry digital laws of the kind used throughout the book.
The digital law of a plane group
In a lattice basis every element of G G G is an affine map { A ∣ v A + t } : x ↦ A x + v A + t \{A\mid v_A+t\}:x\mapsto Ax+v_A+t { A ∣ v A + t } : x ↦ A x + v A + t with A ∈ P ⊂ GL ( n , Z ) A\in P\subset\operatorname{GL}(n,\mathbb{Z}) A ∈ P ⊂ GL ( n , Z ) , t ∈ Z n t\in\mathbb{Z}^{n} t ∈ Z n , and a fixed fractional translation v A v_A v A for each A A A . For m ≥ 1 m\ge1 m ≥ 1 the subgroup m T mT m T is normal in G G G and ∣ G / m T ∣ = m n ∣ P ∣ |G/mT|=m^n|P| ∣ G / m T ∣ = m n ∣ P ∣ . Numbering P = { A 0 = 1 , A 1 , … } P=\{A_0=1,A_1,\ldots\} P = { A 0 = 1 , A 1 , … } , in the plane we label { A i ∣ v i + t } \{A_i\mid v_i+t\} { A i ∣ v i + t } , t ∈ ( Z / m ) 2 t\in(\mathbb{Z}/m)^2 t ∈ ( Z / m ) 2 , by
k = i m 2 + t 1 m + t 2 , (77) k=i m^2+t_1m+t_2, \tag*{(77)} k = i m 2 + t 1 m + t 2 , (77)
and the product is
{ A i ∣ v i + t } { A j ∣ v j + s } = { A i A j ∣ v i j + A i s + t + c ( i , j ) } , c ( i , j ) = A i v j + v i − v i j ∈ Z n , (78) \{A_i\mid v_i+t\}\{A_j\mid v_j+s\}=\{A_iA_j\mid v_{ij}+A_i s+t+c(i,j)\},\qquad c(i,j)=A_i v_j+v_i-v_{ij}\in\mathbb{Z}^{n}, \tag*{(78)} { A i ∣ v i + t } { A j ∣ v j + s } = { A i A j ∣ v ij + A i s + t + c ( i , j )} , c ( i , j ) = A i v j + v i − v ij ∈ Z n , (78)
where v i j v_{ij} v ij is the fractional translation of A i A j A_iA_j A i A j and translations are reduced modulo m m m . The carry c c c is a 2-cocycle representing the class of the extension 1 → T → G → P → 1 1\to T\to G\to P\to1 1 → T → G → P → 1 in H 2 ( P , T ) H^2(P,T) H 2 ( P , T ) ; it vanishes for the symmorphic groups. If ∣ P ∣ |P| ∣ P ∣ divides m m m , the exact sequence H 2 ( P , T ) → m H 2 ( P , T ) → H 2 ( P , T / m T ) H^2(P,T)\xrightarrow{m}H^2(P,T)\to H^2(P,T/mT) H 2 ( P , T ) m H 2 ( P , T ) → H 2 ( P , T / m T ) together with ∣ P ∣ H 2 ( P , T ) = 0 |P|H^2(P,T)=0 ∣ P ∣ H 2 ( P , T ) = 0 shows that the class survives in H 2 ( P , T / m T ) H^2(P,T/mT) H 2 ( P , T / m T ) , so the extension 1 → T / m T → G / m T → P → 1 1\to T/mT\to G/mT\to P\to1 1 → T / m T → G / m T → P → 1 determines G G G .
A worked product: p m pm p m and p g pg p g
Both groups have the point group { 1 , A 1 } \{1,A_1\} { 1 , A 1 } , A 1 = diag ( − 1 , 1 ) A_1=\operatorname{diag}(-1,1) A 1 = diag ( − 1 , 1 ) , the mirror x ↦ ( − x 1 , x 2 ) x\mapsto(-x_1,x_2) x ↦ ( − x 1 , x 2 ) ; in p g pg p g it carries v 1 = ( 0 , 1 2 ) v_1=(0,\frac{1}{2}) v 1 = ( 0 , 2 1 ) , a glide. In G / 6 T G/6T G /6 T the label of { A 1 ∣ v 1 } \{A_1\mid v_1\} { A 1 ∣ v 1 } is k = 1 ⋅ 6 2 = 36 k=1\cdot6^2=36 k = 1 ⋅ 6 2 = 36 by (77) . In p m pm p m we have c ( 1 , 1 ) = 0 c(1,1)=0 c ( 1 , 1 ) = 0 and 36 ∗ 36 = 0 36 * 36=0 36 ∗ 36 = 0 , the identity. In p g pg p g , c ( 1 , 1 ) = A 1 v 1 + v 1 = ( 0 , 1 ) c(1,1)=A_1v_1+v_1=(0,1) c ( 1 , 1 ) = A 1 v 1 + v 1 = ( 0 , 1 ) and 36 ∗ 36 = 1 36 * 36=1 36 ∗ 36 = 1 , the label of the translation { 1 ∣ ( 0 , 1 ) } \{1 \mid(0,1)\} { 1 ∣ ( 0 , 1 )} : the glide squares to a lattice translation, and the carry records exactly this. Already modulo 2 T 2T 2 T the two groups differ: p m / 2 T ≅ C 2 2 pm/2T \cong C_2^2 p m /2 T ≅ C 2 2 , whereas p g / 2 T ≅ C 4 × C 2 pg/2T \cong C_4 \times C_2 p g /2 T ≅ C 4 × C 2 .
Separating the plane groups
Table 10.1 counts, for small m m m , the isomorphism types among the 17 quotients G / m T G/mT G / m T , as detected by the distribution of element orders, the order of the centre, the number of conjugacy classes and the order of the derived subgroup. For m = 6 m=6 m = 6 all 17 are distinct: the factor 2 detects the glides, and the factor 3 separates p 3 m 1 p3m1 p 3 m 1 from p 31 m p31m p 31 m , which differ only in the position of the mirrors relative to the hexagonal lattice. For m = 1 m=1 m = 1 the quotients are the point groups.
Table 10.1. Plane groups modulo m T mT m T : isomorphism types among the 17 quotients G / m T G/mT G / m T .
Digital laws of the 230 space groups
The law (78) carries over to three dimensions without change. Let { R i ∣ t i } \{R_i \mid t_i\} { R i ∣ t i } , 0 ≤ i < r 0 \le i < r 0 ≤ i < r , be the symmetry operations of the conventional cell, centring translations included, with translations written in units of 1 24 \frac{1}{24} 24 1 and 0 ≤ t i < 24 0 \le t_i < 24 0 ≤ t i < 24 , and let L = Z 3 L=\mathbb{Z}^3 L = Z 3 be the conventional lattice. The element { R i ∣ t i + 24 s } \{R_i \mid t_i+24s\} { R i ∣ t i + 24 s } of G / m L G/mL G / m L , s ∈ ( Z / m ) 3 s \in(\mathbb{Z}/m)^3 s ∈ ( Z / m ) 3 , gets the label
k = i m 3 + s 1 m 2 + s 2 m + s 3 , (79) k=i\,m^3+s_1m^2+s_2m+s_3, \tag*{(79)} k = i m 3 + s 1 m 2 + s 2 m + s 3 , (79)
and the product is
{ R i ∣ t i + 24 s } { R j ∣ t j + 24 s ′ } = { R k ∣ t k + 24 ( R i s ′ + s + c ( i , j ) ) } , c ( i , j ) = 1 24 ( R i t j + t i − t k ) ∈ Z 3 , (80) \{R_i \mid t_i+24s\}\{R_j \mid t_j+24s'\}=\{R_k \mid t_k+24(R_i s'+s+c(i,j))\},
\qquad
c(i,j)=\frac{1}{24}(R_i t_j+t_i-t_k)\in\mathbb{Z}^3, \tag*{(80)} { R i ∣ t i + 24 s } { R j ∣ t j + 24 s ′ } = { R k ∣ t k + 24 ( R i s ′ + s + c ( i , j ))} , c ( i , j ) = 24 1 ( R i t j + t i − t k ) ∈ Z 3 , (80)
where k k k is the operation with rotation R i R j R_iR_j R i R j and translation congruent to R i t j + t i R_it_j+t_i R i t j + t i modulo 24. Thus every one of the 230 space groups, and every modulus m m m , gives a digital law on { 0 , … , m 3 r − 1 } \{0,\ldots,m^3r-1\} { 0 , … , m 3 r − 1 } : the digit i i i names a symmetry operation, the base-m m m digits s 1 s 2 s 3 s_1s_2s_3 s 1 s 2 s 3 a translation, and the carry table c c c , of size r × r r \times r r × r , is the whole crystallographic content of the law. In multiply.py, inverse.py and conjugate.py these laws are called SGN/m; for example SG62/2 is P n m a Pnma P nma modulo 2 L 2L 2 L , a group of order 64, and SG225/2 is F m 3 ˉ m Fm\bar{3}m F m 3 ˉ m modulo 2 L 2L 2 L , of order 1536.
By hand: P 2 1 / c P2_1/c P 2 1 / c (No. 14). The operation lists are the triplets of the International Tables, kept in data/space_group_ops.txt; nothing else is needed. For P 2 1 / c P2_1/c P 2 1 / c (setting P 12 1 / c 1 P 1 2_1/c 1 P 1 2 1 / c 1 ) they are
0 : x , y , z , 1 : − x , y + 1 2 , − z + 1 2 , 2 : − x , − y , − z , 3 : x , − y + 1 2 , z + 1 2 , \begin{aligned}
0 &: x,y,z, \qquad&1 &: -x,y+\frac{1}{2},-z+\frac{1}{2},\\
2 &: -x,-y,-z, \qquad&3 &: x,-y+\frac{1}{2},z+\frac{1}{2},
\end{aligned} 0 2 : x , y , z , : − x , − y , − z , 1 3 : − x , y + 2 1 , − z + 2 1 , : x , − y + 2 1 , z + 2 1 ,
{#eq:10.5}
that is, R 0 = I R_0=I R 0 = I , R 1 = diag ( − 1 , 1 , − 1 ) R_1=\operatorname{diag}(-1,1,-1) R 1 = diag ( − 1 , 1 , − 1 ) , R 2 = − I R_2=-I R 2 = − I , R 3 = diag ( 1 , − 1 , 1 ) R_3=\operatorname{diag}(1,-1,1) R 3 = diag ( 1 , − 1 , 1 ) , t 0 = t 2 = 0 t_0=t_2=0 t 0 = t 2 = 0 and t 1 = t 3 = ( 0 , 1 2 , 1 2 ) t_1=t_3=(0,\frac{1}{2},\frac{1}{2}) t 1 = t 3 = ( 0 , 2 1 , 2 1 ) in units of 1 24 \frac{1}{24} 24 1 . Composing the sixteen pairs gives the table of ( k , c ( i , j ) ) (k,c(i,j)) ( k , c ( i , j )) :
The entry c ( 1 , 1 ) = ( 0 , 1 , 0 ) c(1,1)=(0,1,0) c ( 1 , 1 ) = ( 0 , 1 , 0 ) says that the screw 2 1 2_1 2 1 squared is the translation by b b b ; with all carries zero one would have the symmorphic group P 2 / m P2/m P 2/ m . Modulo 2 L 2L 2 L there are 32 labels k = 8 i + 4 s 1 + 2 s 2 + s 3 k=8i+4s_1+2s_2+s_3 k = 8 i + 4 s 1 + 2 s 2 + s 3 . For instance 13 is i = 1 i=1 i = 1 , s = ( 1 , 0 , 1 ) s=(1,0,1) s = ( 1 , 0 , 1 ) , the map − x + 1 -x+1 − x + 1 , y + 1 2 y+\frac{1}{2} y + 2 1 , − z + 3 2 -z+\frac{3}{2} − z + 2 3 , and 27 is i = 3 i=3 i = 3 , s = ( 0 , 1 , 1 ) s=(0,1,1) s = ( 0 , 1 , 1 ) ; then k = 2 k=2 k = 2 , c ( 1 , 3 ) = ( 0 , 1 , 0 ) c(1,3)=(0,1,0) c ( 1 , 3 ) = ( 0 , 1 , 0 ) , and R 1 ( 0 , 1 , 1 ) + ( 1 , 0 , 1 ) + ( 0 , 1 , 0 ) = ( 1 , 2 , 0 ) ≡ ( 1 , 0 , 0 ) R_1(0,1,1)+(1,0,1)+(0,1,0)=(1,2,0)\equiv(1,0,0) R 1 ( 0 , 1 , 1 ) + ( 1 , 0 , 1 ) + ( 0 , 1 , 0 ) = ( 1 , 2 , 0 ) ≡ ( 1 , 0 , 0 ) , so 13 ⋅ 27 = 8 ⋅ 2 + 4 = 20 13\cdot27=8\cdot2+4=20 13 ⋅ 27 = 8 ⋅ 2 + 4 = 20 (multiply.py SG14/2 13 27 --check). Likewise 13 ⋅ 13 = 2 13\cdot13=2 13 ⋅ 13 = 2 , a translation by b b b , so 13 has order 4.
Separating the space groups
For the 230 space-group types, in the standard settings of the International Tables, we use the lattice L ⊆ T L\subseteq T L ⊆ T of the conventional cell, which is normal in G G G , and the quotients G / m L G/mL G / m L of order m 3 ∣ G / L ∣ m^3|G/L| m 3 ∣ G / L ∣ . For g = { R ∣ t } g=\{R\mid t\} g = { R ∣ t } with R R R of order k k k we have g k = { 1 ∣ ( 1 + R + ⋯ + R k − 1 ) t } g^k=\{1\mid(1+R+\cdots+R^{k-1})t\} g k = { 1 ∣ ( 1 + R + ⋯ + R k − 1 ) t } , so the order of every element of G / m L G/mL G / m L is obtained in closed form. So is its centralizer: { R ∣ t } \{R\mid t\} { R ∣ t } and { S ∣ u } \{S\mid u\} { S ∣ u } commute if and only if R S = S R RS=SR R S = S R and ( R − 1 ) u ≡ ( S − 1 ) t (R-1)u\equiv(S-1)t ( R − 1 ) u ≡ ( S − 1 ) t modulo m L mL m L , and for fixed g g g and S S S the admissible u u u form either the empty set or a coset of the kernel of R − 1 R-1 R − 1 on L / m L L/mL L / m L . No multiplication table is needed, even for ∣ G / 12 L ∣ = 331776 |G/12L|=331776 ∣ G /12 L ∣ = 331776 . Table 10.2 counts the types separated by the distribution of element orders, by the joint distribution of ( ord g , ∣ C ( g ) ∣ ) (\operatorname{ord}g,|C(g)|) ( ord g , ∣ C ( g ) ∣ ) (both are isomorphism invariants of G / m L G/mL G / m L ; the second contains the order of the centre and the number of conjugacy classes), and by the latter refined by ( det R , tr R ) (\det R,\operatorname{tr}R) ( det R , tr R ) , which also uses the action of the point group. Two space groups are isomorphic if and only if they are affinely conjugate (Bieberbach), so the 11 enantiomorphic pairs, such as P 3 1 21 P3_121 P 3 1 21 and P 3 2 21 P3_221 P 3 2 21 , have isomorphic quotients for every m m m : no invariant of the abstract group separates more than 219 types, and the remaining distinction requires an orientation of L / m L L/mL L / m L .
Table 10.2. Space groups modulo m L mL m L : types separated among the 230 (at most 219 up to isomorphism).
For m = 12 m=12 m = 12 the joint distribution separates 213 types. Besides the enantiomorphic pairs it leaves five sets, which account for the remaining 6: { P c a 2 1 , P n a 2 1 } \{Pca2_1,Pna2_1\} { P c a 2 1 , P na 2 1 } (Nos. 29, 33), { P n c 2 , P b a 2 , P n n 2 } \{Pnc2,Pba2,Pnn2\} { P n c 2 , P ba 2 , P nn 2 } (30, 32, 34), { P n n n , P b a n } \{Pnnn,Pban\} { P nnn , P ban } (48, 50), { P ‾ 4 b 2 , P ‾ 4 n 2 } \{\overline{P}4b2,\overline{P}4n2\} { P 4 b 2 , P 4 n 2 } (117, 118) and { P 6 3 c m , P 6 3 m c } \{P6_3cm,P6_3mc\} { P 6 3 c m , P 6 3 m c } (185, 186); the same sets are unseparated for m = 4 m=4 m = 4 and m = 6 m=6 m = 6 . The full multiplication tables (space_residual.py) separate most of them by finer invariants: P c a 2 1 Pca2_1 P c a 2 1 and P n a 2 1 Pna2_1 P na 2 1 already modulo 2 L 2L 2 L , where the subgroup generated by the squares has order 4, resp. 8; P n n n Pnnn P nnn and P b a n Pban P ban modulo 4 L 4L 4 L , where the third term of the lower central series has order 8, resp. 4; P ‾ 4 b 2 \overline{P}4b2 P 4 b 2 and P ‾ 4 n 2 \overline{P}4n2 P 4 n 2 modulo 4 L 4L 4 L (∣ γ 3 ∣ = 8 |\gamma_3|=8 ∣ γ 3 ∣ = 8 , resp. 16); P n c 2 Pnc2 P n c 2 from P b a 2 Pba2 P ba 2 and P n n 2 Pnn2 P nn 2 modulo 4 L 4L 4 L (∣ γ 3 ∣ = 2 |\gamma_3| = 2 ∣ γ 3 ∣ = 2 , resp. 4 4 4 ); and P 6 3 c m P6_3cm P 6 3 c m and P 6 3 m c P6_3mc P 6 3 m c modulo 6 L 6L 6 L , by the fibres of the squaring map. For P b a 2 Pba2 P ba 2 and P n n 2 Pnn2 P nn 2 (Nos. 32, 34) none of these invariants differs for m ≤ 8 m \le8 m ≤ 8 , but an exhaustive backtracking search over images of generators shows that their quotients modulo 4 L 4L 4 L are not isomorphic (for m = 2 m = 2 m = 2 the search does find isomorphisms). Hence the finite quotients G / m L G/mL G / m L with m ∈ { 2 , 4 , 6 , 12 } m \in\{2, 4, 6, 12\} m ∈ { 2 , 4 , 6 , 12 } (modulo 12 L 12L 12 L for the bulk of the types, modulo 2 L 2L 2 L , 4 L 4L 4 L and 6 L 6L 6 L for the residual sets) separate all 219 isomorphism types of space groups, and only the orientation of the enantiomorphic pairs is invisible to them. The last step, an exhaustive search in groups of order 256, is confirmed independently by GAP (space_gap.py): modulo 2 L 2L 2 L the quotients of P n c 2 Pnc2 P n c 2 , P b a 2 Pba2 P ba 2 and P n n 2 Pnn2 P nn 2 are all the small group [32, 22], whereas modulo 4 L 4L 4 L they are [256, 2662], [256, 1216] and [256, 557]; modulo 8 L 8L 8 L , of order 2048, IsomorphismGroups also finds no isomorphism between the quotients of P b a 2 Pba2 P ba 2 and P n n 2 Pnn2 P nn 2 .
The computations are made by wallpaper_law.py and space_quotients.py (Chapter 15); the 17 laws (78) are available in multiply.py, inverse.py and conjugate.py under the names pg/6, p4g/6, . . . , with any modulus (pg/4, . . . ).
The quotients as abstract finite groups
Extensions. For every m m m the quotient is an extension
1 → T / m T → G / m T → P → 1 (81) 1 \to T/mT \to G/mT \to P \to1 \tag*{(81)} 1 → T / m T → G / m T → P → 1 (81)
of the abelian group T / m T ≅ ( Z / m ) n T/mT \cong(\mathbb{Z}/m)^n T / m T ≅ ( Z / m ) n by the point group P P P (for a centred space group and the conventional lattice L L L , of the group of all translations modulo m L mL m L by P P P ). If G G G is symmorphic, (81) splits and G / m T ≅ ( Z / m ) n ⋊ P G/mT \cong(\mathbb{Z}/m)^n \rtimes P G / m T ≅ ( Z / m ) n ⋊ P , with P P P acting through its integer matrices reduced modulo m m m . If G G G is not symmorphic and ∣ P ∣ |P| ∣ P ∣ divides m m m , the extension does not split, since the class of G G G is non-zero and survives in H 2 ( P , T / m T ) H^2(P,T/mT) H 2 ( P , T / m T ) (Section 10.2). For smaller m m m the class may die: modulo 2 T 2T 2 T the glide group p g pg p g gives C 4 × C 2 C_4 \times C_2 C 4 × C 2 , a non-split extension of C 2 2 C_2^2 C 2 2 by C 2 C_2 C 2 , but p 3 m 1 p3m1 p 3 m 1 and p 31 m p31m p 31 m both give S 4 S_4 S 4 , and they are told apart only modulo 6 T 6T 6 T .
Dicyclic and metacyclic quotients. None of these groups is dicyclic: a dicyclic group has a unique involution, whereas every quotient in Table 10.3 has at least 3 involutions, and every quotient G / m L G/mL G / m L of a space group with m m m even contains L / 2 L L/2L L /2 L -translations forming C 2 3 C_2^3 C 2 3 , hence at least 7 involutions. Metacyclic groups are rare too. Every subgroup of a metacyclic group is generated by two elements, so no quotient G / m L G/mL G / m L of a space group with m ≥ 2 m \ge2 m ≥ 2 is metacyclic: it contains ( Z / m ) 3 (\mathbb{Z}/m)^3 ( Z / m ) 3 . In the plane, GAP finds exactly p 1 p1 p 1 , p g pg p g and c m cm c m metacyclic modulo 2 T 2T 2 T , and p 1 p1 p 1 and p g pg p g modulo 6 T 6T 6 T ; for example p g / 6 T ≅ C 6 × ( C 3 × C 4 ) pg/6T \cong C_6 \times(C_3 \times C_4) p g /6 T ≅ C 6 × ( C 3 × C 4 ) .
Table 10.3. The plane groups G G G , their point groups P P P and their quotients modulo 2 T 2T 2 T and 6 T 6T 6 T .
The tables. Tables 10.3 and 10.4 identify the quotients with GAP (space_table.py): [ N , i ] [N,i] [ N , i ] is the group SmallGroup(N , i N,i N , i ) of the SmallGroups library, and the structure is GAP’s StructureDescription, in which D 2 k D_{2k} D 2 k is the dihedral group of order 2 k 2k 2 k , A × B A \times B A × B a split extension and A ⋅ B A \cdot B A ⋅ B a non-split one. A structure description names a group only up to such products, so the library number is the precise identification; it is not available for order 1536 (4 face-centred cubic types), marked —. Each row is also a digital law: multiply.py pg/2, multiply.py SG62/2, and so on.
Table 10.4. 4: The 230 space groups G G G (International Tables number and symbol), their point groups P P P and their quotients modulo 2 L 2L 2 L , L L L the conventional lattice.
Separating the space groups
Table 8.
Table 10.
Separating the space groups
Table 1 (PDF p. 127).
Table 9.
Exercises
Let R R R be a rotation of order n n n of R 2 \mathbb{R}^2 R 2 or R 3 \mathbb{R}^3 R 3 that maps a lattice L L L onto itself. Show that n ∈ { 1 , 2 , 3 , 4 , 6 } n \in\{1,2,3,4,6\} n ∈ { 1 , 2 , 3 , 4 , 6 } (the crystallographic restriction).
Let G G G be a space group with lattice L = Z 3 L = \mathbb{Z}^3 L = Z 3 and conventional operations { R i ∣ t i } \{R_i \mid t_i\} { R i ∣ t i } , 0 ≤ i < r 0 \le i < r 0 ≤ i < r . Show that m L mL m L is a normal subgroup of G G G for every m ≥ 1 m \ge1 m ≥ 1 , that ∣ G / m L ∣ = r m 3 |G/mL| = r m^3 ∣ G / m L ∣ = r m 3 , and that (79) is a bijection from G / m L G/mL G / m L onto [ 0 , r m 3 ) [0,r m^3) [ 0 , r m 3 ) .
Show that c ( i , j ) c(i,j) c ( i , j ) in (80) lies in Z 3 \mathbb{Z}^3 Z 3 . If R i R j = R l R_iR_j = R_l R i R j = R l , show that the product of the labels i m 3 + s i m^3+s i m 3 + s and j m 3 + s ′ j m^3+s' j m 3 + s ′ (with s , s ′ s,s' s , s ′ read as vectors in ( Z / m ) 3 (\mathbb{Z}/m)^3 ( Z / m ) 3 ) is the label l m 3 + s ′ ′ l m^3+s'' l m 3 + s ′′ with s ′ ′ ≡ R i s ′ + s + c ( i , j ) ( m o d m ) s'' \equiv R_i s' + s + c(i,j) \pmod m s ′′ ≡ R i s ′ + s + c ( i , j ) ( mod m ) .
For P 1 ‾ P\overline{1} P 1 (No. 2), with operations { I ∣ 0 } \{I \mid0\} { I ∣ 0 } and { − I ∣ 0 } \{-I \mid0\} { − I ∣ 0 } , show that G / m L ≅ ( Z / m ) 3 ⋊ C 2 G/mL \cong(\mathbb{Z}/m)^3 \rtimes C_2 G / m L ≅ ( Z / m ) 3 ⋊ C 2 , the inversion acting by s ↦ − s s \mapsto-s s ↦ − s , and that G / 2 L ≅ C 2 4 G/2L \cong C_2^4 G /2 L ≅ C 2 4 .
For P 2 1 / c P2_1/c P 2 1 / c (No. 14), with the four operations 0 0 0 : x , y , z x,y,z x , y , z ; 1 1 1 : − x , y + 1 2 , − z + 1 2 -x,y+\frac{1}{2},-z+\frac{1}{2} − x , y + 2 1 , − z + 2 1 ; 2 2 2 : − x , − y , − z -x,-y,-z − x , − y , − z ; 3 3 3 : x , − y + 1 2 , z + 1 2 x,-y+\frac{1}{2},z+\frac{1}{2} x , − y + 2 1 , z + 2 1 , compute c ( 1 , 1 ) c(1,1) c ( 1 , 1 ) , c ( 1 , 2 ) c(1,2) c ( 1 , 2 ) , c ( 2 , 1 ) c(2,1) c ( 2 , 1 ) and c ( 3 , 3 ) c(3,3) c ( 3 , 3 ) . Deduce that the screw rotation { R 1 ∣ t 1 } \{R_1 \mid t_1\} { R 1 ∣ t 1 } and the glide reflection { R 3 ∣ t 3 } \{R_3 \mid t_3\} { R 3 ∣ t 3 } have order 2 m 2m 2 m in G / m L G/mL G / m L , while the inversion { R 2 ∣ t 2 } \{R_2 \mid t_2\} { R 2 ∣ t 2 } has order 2 2 2 .
There are 230 space groups up to conjugation by orientation-preserving affine maps, and 11 enantiomorphic pairs among them. Using Bieberbach’s theorem that two space groups are isomorphic if and only if they are conjugate by an affine map, show that they fall into 219 isomorphism classes.
Show that a finite subgroup of O ( 2 ) O(2) O ( 2 ) that preserves a lattice is one of the ten groups C 1 , C 2 , C 3 , C 4 , C 6 , D 1 , D 2 , D 3 , D 4 , D 6 C_1,C_2,C_3,C_4,C_6,D_1,D_2,D_3,D_4,D_6 C 1 , C 2 , C 3 , C 4 , C 6 , D 1 , D 2 , D 3 , D 4 , D 6 , the point groups of the 17 wallpaper groups.
The symmetric and alternating groups
This chapter treats the symmetric group Sym n \operatorname{Sym}_n Sym n and the alternating group Alt n \operatorname{Alt}_n Alt n . It begins with permutations as digital codes, generators and presentations, conjugacy classes, the normal structure of Sym n \operatorname{Sym}_n Sym n and Alt n \operatorname{Alt}_n Alt n , including the simplicity of Alt n \operatorname{Alt}_n Alt n for n ≥ 5 n \ge5 n ≥ 5 , and their actions and maximal subgroups. It then gives digital laws for them, obtained from normal forms of permutations, and ends with generated tables of numerical data for n ≤ 8 n \le8 n ≤ 8 . Since Alt n \operatorname{Alt}_n Alt n is simple for n ≥ 5 n \ge5 n ≥ 5 , no such law can be graded; the other families of nonabelian finite simple groups are treated in Chapters 12 and 13.
Permutations and digital codes
As in Proposition 2.2 , a permutation R R R of A n \mathcal{A}_n A n is stored as its code R = ∑ i = 0 n − 1 R ( i ) n i R = \sum_{i=0}^{n-1} R(i)n^i R = ∑ i = 0 n − 1 R ( i ) n i , so that d n ( i , R ) = R ( i ) d_n(i,R)=R(i) d n ( i , R ) = R ( i ) ; identity, inverse and composition are given by (10) , (11) and (12) . A word c ∈ [ 0 , n n ) c \in[0,n^n) c ∈ [ 0 , n n ) is the code of a permutation exactly when its n n n digits are pairwise distinct, so precisely n ! n! n ! of the n n n^n n n words of length n n n are permutation codes. Every numerical claim of this chapter is checked by scripts/sn_verify.py and cross-checked against GAP 4.14; in particular the script confirms (10) for n ≤ 8 n \le8 n ≤ 8 , and decoding, (11) , (12) and the multiplicativity of sgn \operatorname{sgn} sgn exhaustively for n ≤ 5 n \le5 n ≤ 5 .
Definition 11.1 (Lehmer label). For R ∈ Sym n R \in\operatorname{Sym}_n R ∈ Sym n and 0 ≤ i < n 0 \le i < n 0 ≤ i < n put a i ( R ) = # { j > i : R ( j ) < R ( i ) } ∈ [ 0 , n − i ) a_i(R)=\#\{j>i:R(j)<R(i)\}\in[0,n-i) a i ( R ) = # { j > i : R ( j ) < R ( i )} ∈ [ 0 , n − i ) , and
λ ( R ) = ∑ i = 0 n − 1 a i ( R ) ∏ j < i ( n − j ) . \lambda(R)=\sum_{i=0}^{n-1}a_i(R)\prod_{j<i}(n-j). λ ( R ) = i = 0 ∑ n − 1 a i ( R ) j < i ∏ ( n − j ) .
Theorem 11.2 (Lehmer label and sign). The map λ \lambda λ is a bijection Sym n → [ 0 , n ! ) \operatorname{Sym}_n \to[0,n!) Sym n → [ 0 , n !) whose mixed-radix digits , with radices n , n − 1 , … , 1 n,n-1,\ldots,1 n , n − 1 , … , 1 , are a 0 ( R ) , … , a n − 1 ( R ) a_0(R),\ldots,a_{n-1}(R) a 0 ( R ) , … , a n − 1 ( R ) . Moreover ∑ i a i ( R ) \sum_i a_i(R) ∑ i a i ( R ) is the number inv ( R ) \operatorname{inv}(R) inv ( R ) of inversions of R R R , and sgn R = ( − 1 ) inv ( R ) \operatorname{sgn}R=(-1)^{\operatorname{inv}(R)} sgn R = ( − 1 ) inv ( R ) .
Proof. Given digits a i ∈ [ 0 , n − i ) a_i\in[0,n-i) a i ∈ [ 0 , n − i ) , the permutation is recovered uniquely: R ( 0 ) R(0) R ( 0 ) is the element of rank a 0 a_0 a 0 in A n \mathcal{A}_n A n , R ( 1 ) R(1) R ( 1 ) the element of rank a 1 a_1 a 1 among the remaining n − 1 n-1 n − 1 , and so on. Hence R ↦ ( a 0 , … , a n − 1 ) R\mapsto(a_0,\ldots,a_{n-1}) R ↦ ( a 0 , … , a n − 1 ) is a bijection onto ∏ i [ 0 , n − i ) \prod_i[0,n-i) ∏ i [ 0 , n − i ) , and mixed-radix notation is a bijection of that product onto [ 0 , n ! ) [0,n!) [ 0 , n !) . Each inversion i < j i<j i < j , R ( i ) > R ( j ) R(i)>R(j) R ( i ) > R ( j ) is counted once, in a i a_i a i . For the sign, let s i = ( i i + 1 ) s_i=(i\ i+1) s i = ( i i + 1 ) . The one-line word of R ∘ s i R\circ s_i R ∘ s i is that of R R R with the entries in positions i , i + 1 i,i+1 i , i + 1 exchanged, so inv \operatorname{inv} inv changes by exactly 1; and composing with a transposition merges two cycles or splits one, so by Theorem 2.8 the sign also changes. Sorting the word by adjacent exchanges reaches the identity, which has inv = 0 \operatorname{inv}=0 inv = 0 and sign 1. ∎
Example 11.3. For n = 3 n = 3 n = 3 and R = ( 0 1 2 ) R = (0\ 1\ 2) R = ( 0 1 2 ) we have R ( 0 ) , R ( 1 ) , R ( 2 ) = 1 , 2 , 0 R(0), R(1), R(2) = 1, 2, 0 R ( 0 ) , R ( 1 ) , R ( 2 ) = 1 , 2 , 0 , so the code is 1 + 2 ⋅ 3 + 0 ⋅ 9 = 7 1 + 2 \cdot3 + 0 \cdot9 = 7 1 + 2 ⋅ 3 + 0 ⋅ 9 = 7 . The Lehmer digits are a 0 = 1 a_0 = 1 a 0 = 1 , a 1 = 1 a_1 = 1 a 1 = 1 , a 2 = 0 a_2 = 0 a 2 = 0 , so λ ( R ) = 1 + 1 ⋅ 3 = 4 \lambda(R) = 1 + 1 \cdot3 = 4 λ ( R ) = 1 + 1 ⋅ 3 = 4 , and inv ( R ) = 2 \operatorname{inv}(R) = 2 inv ( R ) = 2 shows again that a 3-cycle is even.
Proposition 11.4 (Cycle counts). The number c ( n , k ) c(n,k) c ( n , k ) of R ∈ Sym n R \in\operatorname{Sym}_n R ∈ Sym n with exactly k k k cycles (fixed points included) satisfies c ( n , k ) = c ( n − 1 , k − 1 ) + ( n − 1 ) c ( n − 1 , k ) c(n,k) = c(n-1,k-1) + (n-1)c(n-1,k) c ( n , k ) = c ( n − 1 , k − 1 ) + ( n − 1 ) c ( n − 1 , k ) . Since sgn \operatorname{sgn} sgn is a surjective homomorphism for n ≥ 2 n \ge2 n ≥ 2 , ∣ Alt n ∣ = n ! / 2 |\operatorname{Alt}_n| = n!/2 ∣ Alt n ∣ = n ! /2 .
Proof. Delete the point n − 1 n-1 n − 1 from the cycle notation of R R R . Either it was a fixed point, leaving a permutation of A n − 1 A_{n-1} A n − 1 with k − 1 k-1 k − 1 cycles, or it followed one of the n − 1 n-1 n − 1 other points in its cycle, and it can be reinserted after any of them in a permutation of A n − 1 A_{n-1} A n − 1 with k k k cycles. □
For n = 7 n = 7 n = 7 the script finds c ( 7 , k ) = 720 , 1764 , 1624 , 735 , 175 , 21 , 1 c(7,k) = 720, 1764, 1624, 735, 175, 21, 1 c ( 7 , k ) = 720 , 1764 , 1624 , 735 , 175 , 21 , 1 for k = 1 , … , 7 k = 1,\ldots,7 k = 1 , … , 7 , in agreement with the recurrence; the Lehmer label is checked to be a bijection, and sgn R = ( − 1 ) inv ( R ) \operatorname{sgn} R = (-1)^{\operatorname{inv}(R)} sgn R = ( − 1 ) inv ( R ) , for all n ≤ 7 n \le7 n ≤ 7 .
Generators and presentations
Write s i = ( i i + 1 ) s_i = (i\ i+1) s i = ( i i + 1 ) for 0 ≤ i ≤ n − 2 0 \le i \le n-2 0 ≤ i ≤ n − 2 and ℓ ( R ) \ell(R) ℓ ( R ) for the length of a shortest word in the s i s_i s i equal to R R R .
Theorem 11.5 (Generators of Sym n \operatorname{Sym}_n Sym n ). Sym n = ⟨ s 0 , … , s n − 2 ⟩ = ⟨ ( 0 1 ) , ( 0 1 ⋯ n − 1 ) ⟩ \operatorname{Sym}_n = \langle s_0,\ldots,s_{n-2}\rangle= \langle(0\ 1),(0\ 1\cdots n-1)\rangle Sym n = ⟨ s 0 , … , s n − 2 ⟩ = ⟨( 0 1 ) , ( 0 1 ⋯ n − 1 )⟩ . Moreover ℓ ( R ) = inv ( R ) \ell(R) = \operatorname{inv}(R) ℓ ( R ) = inv ( R ) , the maximum n ( n − 1 ) / 2 n(n-1)/2 n ( n − 1 ) /2 is attained only by i ↦ n − 1 − i i \mapsto n-1-i i ↦ n − 1 − i , and
∑ R ∈ Sym n q ℓ ( R ) = ∏ i = 1 n ( 1 + q + ⋯ + q i − 1 ) . \sum_{R\in\operatorname{Sym}_n} q^{\ell(R)} = \prod_{i=1}^{n}(1+q+\cdots+q^{i-1}). R ∈ Sym n ∑ q ℓ ( R ) = i = 1 ∏ n ( 1 + q + ⋯ + q i − 1 ) .
Proof. By the proof of Theorem 11.2 , right multiplication by s i s_i s i changes inv \operatorname{inv} inv by ± 1 \pm1 ± 1 , so ℓ ( R ) ≥ inv ( R ) \ell(R) \ge\operatorname{inv}(R) ℓ ( R ) ≥ inv ( R ) . If R ≠ 1 R \ne1 R = 1 it has a descent R ( i ) > R ( i + 1 ) R(i) > R(i+1) R ( i ) > R ( i + 1 ) , and inv ( R ∘ s i ) = inv ( R ) − 1 \operatorname{inv}(R \circ s_i) = \operatorname{inv}(R)-1 inv ( R ∘ s i ) = inv ( R ) − 1 ; induction gives ℓ ( R ) ≤ inv ( R ) \ell(R) \le\operatorname{inv}(R) ℓ ( R ) ≤ inv ( R ) , and in particular the s i s_i s i generate. With c = ( 0 1 ⋯ n − 1 ) c = (0\ 1\cdots n-1) c = ( 0 1 ⋯ n − 1 ) , Proposition 2.6 gives c k ( 0 1 ) c − k = ( k k + 1 ) c^k(0\ 1)c^{-k} = (k\ k+1) c k ( 0 1 ) c − k = ( k k + 1 ) , so the second pair generates as well. All pairs are inverted only by the reversal. Finally inv = ∑ i a i \operatorname{inv} = \sum_i a_i inv = ∑ i a i with the digits a i ∈ [ 0 , n − i ) a_i \in[0,n-i) a i ∈ [ 0 , n − i ) independent by Theorem 11.2 , which gives the product. □
Theorem 11.6 (Coxeter presentation).
Sym n ≅ Γ n = ⟨ σ 1 , … , σ n − 1 ∣ σ i 2 , ( σ i σ i + 1 ) 3 , ( σ i σ j ) 2 ( ∣ i − j ∣ ≥ 2 ) ⟩ , \operatorname{Sym}_n \cong\Gamma_n = \langle\sigma_1,\ldots,\sigma_{n-1} \mid\sigma_i^2,\;(\sigma_i\sigma_{i+1})^3,\;(\sigma_i\sigma_j)^2\ (|i-j|\ge2)\rangle, Sym n ≅ Γ n = ⟨ σ 1 , … , σ n − 1 ∣ σ i 2 , ( σ i σ i + 1 ) 3 , ( σ i σ j ) 2 ( ∣ i − j ∣ ≥ 2 )⟩ , via σ i ↦ s i − 1 \sigma_i \mapsto s_{i-1} σ i ↦ s i − 1 .
Proof. The relations hold in Sym n \operatorname{Sym}_n Sym n (s i s i + 1 s_i s_{i+1} s i s i + 1 is a 3-cycle and disjoint transpositions commute), so by Theorem 11.5 there is an epimorphism Γ n → Sym n \Gamma_n \to\operatorname{Sym}_n Γ n → Sym n , and it suffices to show ∣ Γ n ∣ ≤ n ! |\Gamma_n| \le n! ∣ Γ n ∣ ≤ n ! . We induct on n n n , the case n = 1 n=1 n = 1 being trivial. Let H = ⟨ σ 1 , … , σ n − 2 ⟩ H = \langle\sigma_1,\ldots,\sigma_{n-2}\rangle H = ⟨ σ 1 , … , σ n − 2 ⟩ ; it is a quotient of Γ n − 1 \Gamma_{n-1} Γ n − 1 , so ∣ H ∣ ≤ ( n − 1 ) ! |H| \le(n-1)! ∣ H ∣ ≤ ( n − 1 )! . Put w k = σ n − 1 σ n − 2 ⋯ σ k w_k = \sigma_{n-1}\sigma_{n-2}\cdots\sigma_k w k = σ n − 1 σ n − 2 ⋯ σ k for 1 ≤ k ≤ n 1 \le k \le n 1 ≤ k ≤ n , with w n = 1 w_n = 1 w n = 1 . We claim that the union of the cosets H w 1 , … , H w n Hw_1,\ldots,Hw_n H w 1 , … , H w n is stable under right multiplication by every σ j \sigma_j σ j ; since it contains 1 1 1 , it is then all of Γ n \Gamma_n Γ n , and ∣ Γ n ∣ ≤ n ( n − 1 ) ! |\Gamma_n| \le n(n-1)! ∣ Γ n ∣ ≤ n ( n − 1 )! . Indeed: if j < k − 1 j < k-1 j < k − 1 , σ j \sigma_j σ j commutes with every factor of w k w_k w k , so w k σ j = σ j w k ∈ H w k w_k\sigma_j = \sigma_j w_k \in Hw_k w k σ j = σ j w k ∈ H w k ; if j = k − 1 j = k-1 j = k − 1 , w k σ j = w k − 1 w_k\sigma_j = w_{k-1} w k σ j = w k − 1 ; if j = k ≤ n − 1 j = k \le n-1 j = k ≤ n − 1 , w k σ j = w k + 1 w_k\sigma_j = w_{k+1} w k σ j = w k + 1 . If j > k j > k j > k , move σ j \sigma_j σ j left past σ j − 2 , … , σ k \sigma_{j-2},\ldots,\sigma_k σ j − 2 , … , σ k and use the braid relation σ j σ j − 1 σ j = σ j − 1 σ j σ j − 1 \sigma_j\sigma_{j-1}\sigma_j = \sigma_{j-1}\sigma_j\sigma_{j-1} σ j σ j − 1 σ j = σ j − 1 σ j σ j − 1 , a consequence of the first two relations; the resulting σ j − 1 \sigma_{j-1} σ j − 1 commutes with σ n − 1 , … , σ j + 1 \sigma_{n-1},\ldots,\sigma_{j+1} σ n − 1 , … , σ j + 1 , so w k σ j = σ j − 1 w k ∈ H w k w_k\sigma_j = \sigma_{j-1}w_k \in Hw_k w k σ j = σ j − 1 w k ∈ H w k . □
Theorem 11.7 (Generators and presentation of Alt n \operatorname{Alt}_n Alt n ). Let n ≥ 3 n \ge3 n ≥ 3 and x i = ( 0 1 i + 1 ) x_i = (0\ 1\ i+1) x i = ( 0 1 i + 1 ) for 1 ≤ i ≤ n − 2 1 \le i \le n-2 1 ≤ i ≤ n − 2 . Then Alt n = ⟨ x 1 , … , x n − 2 ⟩ \operatorname{Alt}_n = \langle x_1,\ldots,x_{n-2}\rangle Alt n = ⟨ x 1 , … , x n − 2 ⟩ , the 3-cycles generate Alt n \operatorname{Alt}_n Alt n , and so do ( 0 1 2 ) (0\ 1\ 2) ( 0 1 2 ) together with ( 0 1 ⋯ n − 1 ) (0\ 1\cdots n-1) ( 0 1 ⋯ n − 1 ) for n n n odd, or with ( 1 2 ⋯ n − 1 ) (1\ 2\cdots n-1) ( 1 2 ⋯ n − 1 ) for n n n even. The x i x_i x i satisfy x i 3 = ( x i x j ) 2 = 1 x_i^3 = (x_i x_j)^2 = 1 x i 3 = ( x i x j ) 2 = 1 (i ≠ j i \ne j i = j ), and these relations present Alt n \operatorname{Alt}_n Alt n (Carmichael).
Proof. An even permutation is a product of pairs of transpositions, and ( a b ) ( a c ) = ( a c b ) (a\,b)(a\,c)=(a\,c\,b) ( a b ) ( a c ) = ( a c b ) , ( a b ) ( c d ) = ( ( a b ) ( b c ) ) ( ( b c ) ( c d ) ) (a\,b)(c\,d)=((a\,b)(b\,c))((b\,c)(c\,d)) ( a b ) ( c d ) = (( a b ) ( b c )) (( b c ) ( c d )) , so the 3-cycles generate Alt n \operatorname{Alt}_n Alt n . Let H = ⟨ x i ⟩ H=\langle x_i\rangle H = ⟨ x i ⟩ ; it contains every 3-cycle through 0 and 1. For distinct a , b ≥ 2 a,b\ge2 a , b ≥ 2 , conjugating ( 0 1 a ) (0\,1\,a) ( 0 1 a ) by ( 0 1 b ) ± 1 (0\,1\,b)^{\pm1} ( 0 1 b ) ± 1 gives ( 1 b a ) (1\,b\,a) ( 1 b a ) and ( 0 a b ) (0\,a\,b) ( 0 a b ) , so H H H contains every 3-cycle meeting { 0 , 1 } \{0,1\} { 0 , 1 } ; and ( 0 a b ) ( 0 b d ) = ( a b d ) (0\,a\,b)(0\,b\,d)=(a\,b\,d) ( 0 a b ) ( 0 b d ) = ( a b d ) gives the rest. For the two-generator sets: if n n n is odd, conjugating ( 0 1 2 ) (0\,1\,2) ( 0 1 2 ) by powers of the even n n n -cycle gives t k = ( k k + 1 k + 2 ) t_k=(k\,k+1\,k+2) t k = ( k k + 1 k + 2 ) ; conjugating ( 0 2 3 ) = t 1 ( 0 1 2 ) t 1 − 1 (0\,2\,3)=t_1(0\,1\,2)t_1^{-1} ( 0 2 3 ) = t 1 ( 0 1 2 ) t 1 − 1 by t 0 t_0 t 0 gives ( 0 1 3 ) − 1 (0\,1\,3)^{-1} ( 0 1 3 ) − 1 , and for m ≥ 3 m\ge3 m ≥ 3 conjugation by t m − 1 t_{m-1} t m − 1 , which fixes 0, 1, sends ( 0 1 m ) (0\,1\,m) ( 0 1 m ) to ( 0 1 m + 1 ) (0\,1\,m+1) ( 0 1 m + 1 ) , so all x i x_i x i are reached. If n n n is even, conjugating ( 0 1 2 ) (0\,1\,2) ( 0 1 2 ) by powers of the even ( n − 1 ) (n-1) ( n − 1 ) -cycle gives ( 0 m m + 1 ) (0\,m\,m+1) ( 0 m m + 1 ) for 1 ≤ m ≤ n − 2 1\le m\le n-2 1 ≤ m ≤ n − 2 , and conjugating ( 0 1 m ) (0\,1\,m) ( 0 1 m ) by ( 0 m m + 1 ) − 1 (0\,m\,m+1)^{-1} ( 0 m m + 1 ) − 1 gives ( 0 1 m + 1 ) − 1 (0\,1\,m+1)^{-1} ( 0 1 m + 1 ) − 1 . The relations hold since x i x j = ( 0 i + 1 ) ( 1 j + 1 ) x_i x_j=(0\,i+1)(1\,j+1) x i x j = ( 0 i + 1 ) ( 1 j + 1 ) for i ≠ j i\ne j i = j . That they suffice is Carmichael’s theorem, which we do not reprove.
The script checks all generating sets and relations for n ≤ 7 n\le7 n ≤ 7 (for Alt n \operatorname{Alt}_n Alt n , n ≤ 8 n\le8 n ≤ 8 ), that ℓ = inv \ell=\operatorname{inv} ℓ = inv on all of Sym n \operatorname{Sym}_n Sym n by breadth-first search in the Cayley graph, and the length distribution. In GAP, coset enumeration of the two presentations returns the orders n ! n! n ! and n ! / 2 n!/2 n ! /2 for all n ≤ 8 n\le8 n ≤ 8 .
Conjugacy classes
This section specialises § 10.4.3 to Sym n \operatorname{Sym}_n Sym n and Alt n \operatorname{Alt}_n Alt n . By Proposition 2.6 , the code of σ ∘ R ∘ σ − 1 \sigma\circ R\circ\sigma^{-1} σ ∘ R ∘ σ − 1 has digit σ ( R ( i ) ) \sigma(R(i)) σ ( R ( i )) in position σ ( i ) \sigma(i) σ ( i ) : conjugation relabels the cycle notation of R R R through σ \sigma σ . For a partition μ = 1 m 1 2 m 2 ⋯ \mu=1^{m_1}2^{m_2}\cdots μ = 1 m 1 2 m 2 ⋯ of n n n put z μ = ∏ i i m i m i ! z_\mu=\prod_i i^{m_i}m_i! z μ = ∏ i i m i m i ! .
Theorem 11.8 (Classes of Sym n \operatorname{Sym}_n Sym n ). Two permutations are conjugate in Sym n \operatorname{Sym}_n Sym n if and only if they have the same cycle type. The class of type μ \mu μ has centraliser order z μ z_\mu z μ and size n ! / z μ n!/z_\mu n ! / z μ , and k ( Sym n ) = p ( n ) k(\operatorname{Sym}_n)=p(n) k ( Sym n ) = p ( n ) .
Proof. Relabelling preserves cycle lengths; conversely, if R , S R,S R , S have the same type, writing the cycles of S S S under those of R R R , cycle by cycle, defines σ \sigma σ with σ R σ − 1 = S \sigma R\sigma^{-1}=S σ R σ − 1 = S . A permutation commutes with R R R exactly when the relabelling maps the cycle notation of R R R to itself: it may permute the m i m_i m i cycles of each length i i i and rotate each of them, in ∏ i m i ! i m i \prod_i m_i!i^{m_i} ∏ i m i ! i m i ways. Theorem 4.15 gives the class size.
Theorem 11.9 (Classes of Alt n \operatorname{Alt}_n Alt n ). Let n ≥ 2 n\ge2 n ≥ 2 and let μ \mu μ be a partition with n − ℓ ( μ ) n-\ell(\mu) n − ℓ ( μ ) even, ℓ ( μ ) \ell(\mu) ℓ ( μ ) its number of parts, so that the Sym n \operatorname{Sym}_n Sym n -class C μ C_\mu C μ lies in Alt n \operatorname{Alt}_n Alt n . If the parts of μ \mu μ are distinct and odd, C μ C_\mu C μ splits into two Alt n \operatorname{Alt}_n Alt n -classes of size n ! / ( 2 z μ ) n!/(2z_\mu) n ! / ( 2 z μ ) ; otherwise it is one Alt n \operatorname{Alt}_n Alt n -class. Hence k ( Alt n ) k(\operatorname{Alt}_n) k ( Alt n ) is the number of such μ \mu μ plus the number of partitions of n n n into distinct odd parts.
Proof. For R ∈ C μ R\in C_\mu R ∈ C μ , C Alt n ( R ) = C Sym n ( R ) ∩ Alt n C_{\operatorname{Alt}_n}(R)=C_{\operatorname{Sym}_n}(R)\cap\operatorname{Alt}_n C Alt n ( R ) = C Sym n ( R ) ∩ Alt n has index 1 or 2 in C Sym n ( R ) C_{\operatorname{Sym}_n}(R) C Sym n ( R ) . If the centraliser contains an odd permutation, the Alt n \operatorname{Alt}_n Alt n -class of R R R has size ( n ! / 2 ) / ( z μ / 2 ) = ∣ C μ ∣ (n!/2)/(z_\mu/2)=|C_\mu| ( n ! /2 ) / ( z μ /2 ) = ∣ C μ ∣ . Otherwise it has size ∣ C μ ∣ / 2 |C_\mu|/2 ∣ C μ ∣/2 , and conjugation by any odd permutation exchanges the two halves. An even part gives an odd centralising element, the cycle itself; two equal odd parts i i i give one too, the product of the i i i transpositions exchanging the two cycles. If the parts are distinct and odd, the centraliser is generated by the cycles of R R R , all even.
Corollary 11.10 . For n ≥ 5 n\ge5 n ≥ 5 all 3-cycles are conjugate in Alt n \operatorname{Alt}_n Alt n ; for n = 3 , 4 n=3,4 n = 3 , 4 they form two classes.
Proof. The type 3 1 n − 3 3\,1^{n-3} 3 1 n − 3 has distinct parts exactly when n ≤ 4 n\le4 n ≤ 4 .
Example 11.11. In Sym 5 \operatorname{Sym}_5 Sym 5 the types 1 5 1^5 1 5 , 2 1 3 2\,1^3 2 1 3 , 2 2 1 2^2 1 2 2 1 , 3 1 2 3\,1^2 3 1 2 , 3 2 3\,2 3 2 , 4 1 4\,1 4 1 , 5 5 5 have z μ = 120 , 12 , 8 , 6 , 6 , 4 , 5 z_\mu=120,12,8,6,6,4,5 z μ = 120 , 12 , 8 , 6 , 6 , 4 , 5 and class sizes 1 , 10 , 15 , 20 , 20 , 30 , 24 1,10,15,20,20,30,24 1 , 10 , 15 , 20 , 20 , 30 , 24 . In Alt 5 \operatorname{Alt}_5 Alt 5 only 5 splits, and the class equation is 60 = 1 + 15 + 20 + 12 + 12 60=1+15+20+12+12 60 = 1 + 15 + 20 + 12 + 12 .
The script computes the conjugacy classes of S y m n \mathrm{Sym}_n Sym n (n ≤ 7 n \le7 n ≤ 7 ) and A l t n \mathrm{Alt}_n Alt n (n ≤ 8 n \le8 n ≤ 8 ) as orbits, checks the digit relocation for n ≤ 5 n \le5 n ≤ 5 , the class sizes and the splitting rule, and finds k ( A l t n ) = 1 , 3 , 4 , 5 , 7 , 9 , 14 k(\mathrm{Alt}_n) = 1, 3, 4, 5, 7, 9, 14 k ( Alt n ) = 1 , 3 , 4 , 5 , 7 , 9 , 14 for n = 2 , … , 8 n = 2,\ldots,8 n = 2 , … , 8 . GAP agrees, and gives k ( A l t n ) = 18 , 24 , 31 , 43 k(\mathrm{Alt}_n) = 18,24,31,43 k ( Alt n ) = 18 , 24 , 31 , 43 for n = 9 , … , 12 n = 9,\ldots,12 n = 9 , … , 12 and k ( S y m n ) = p ( n ) k(\mathrm{Sym}_n) = p(n) k ( Sym n ) = p ( n ) for n ≤ 12 n \le12 n ≤ 12 .
Cycle index. By Theorem 11.8 there are n ! / z μ n!/z_\mu n ! / z μ permutations of type μ = 1 m 1 2 m 2 ⋯ \mu= 1^{m_1}2^{m_2}\cdots μ = 1 m 1 2 m 2 ⋯ , with z μ = ∏ k k m k m k ! z_\mu= \prod_k k^{m_k}m_k! z μ = ∏ k k m k m k ! . Hence the cycle index of S y m n \mathrm{Sym}_n Sym n on A n \mathcal{A}_n A n (§ 10.4.4 ) is, writing x μ = ∏ k x k m k x^\mu= \prod_k x_k^{m_k} x μ = ∏ k x k m k ,
Z ( S y m n ) = ∑ μ ⊢ n x μ z μ , ∑ n ≥ 0 Z ( S y m n ) t n = exp ( ∑ k ≥ 1 x k t k k ) . Z(\mathrm{Sym}_n)=\sum_{\mu\vdash n}\frac{x^\mu}{z_\mu},\qquad\sum_{n\ge0}Z(\mathrm{Sym}_n)t^n=\exp\left(\sum_{k\ge1}\frac{x_kt^k}{k}\right). Z ( Sym n ) = μ ⊢ n ∑ z μ x μ , n ≥ 0 ∑ Z ( Sym n ) t n = exp ( k ≥ 1 ∑ k x k t k ) .
the second formula because
z μ − 1 t n x μ = ∏ k 1 m k ! ( x k t k k ) m k . z_\mu^{-1}t^nx^\mu=\prod_k\frac{1}{m_k!}\left(\frac{x_kt^k}{k}\right)^{m_k}. z μ − 1 t n x μ = k ∏ m k ! 1 ( k x k t k ) m k .
By Theorem 11.9 ,
Z ( A l t n ) = ∑ μ ⊢ n n − ℓ ( μ ) e v e n 2 x μ z μ . Z(\mathrm{Alt}_n)=\sum_{\substack{\mu\vdash n\\ n-\ell(\mu)\ \mathrm{even}}}\frac{2x^\mu}{z_\mu}. Z ( Alt n ) = μ ⊢ n n − ℓ ( μ ) even ∑ z μ 2 x μ .
Corollary 11.12. For all n , m ≥ 1 n,m\ge1 n , m ≥ 1 ,
∑ μ ⊢ n m ℓ ( μ ) z μ = ( m + n − 1 n ) . \sum_{\mu\vdash n}\frac{m^{\ell(\mu)}}{z_\mu}=\binom{m+n-1}{n}. μ ⊢ n ∑ z μ m ℓ ( μ ) = ( n m + n − 1 ) .
Proof. Two colourings of A n \mathcal{A}_n A n with m m m colours lie in one S y m n \mathrm{Sym}_n Sym n -orbit if and only if they use each colour equally often, so the orbits are the multisets of size n n n from m m m colours, and there are ( m + n − 1 n ) \binom{m+n-1}{n} ( n m + n − 1 ) of them. By Theorem 4.21 their number is 1 n ! ∑ σ m c ( σ ) = ∑ μ m ℓ ( μ ) / z μ \frac{1}{n!}\sum_{\sigma}m^{c(\sigma)}=\sum_{\mu}m^{\ell(\mu)}/z_\mu n ! 1 ∑ σ m c ( σ ) = ∑ μ m ℓ ( μ ) / z μ .
For n = 4 n=4 n = 4 , m = 2 m=2 m = 2 : 16 24 + 8 2 ⋅ 1 2 + 4 8 + 4 3 + 2 4 = 5 = ( 5 4 ) \frac{16}{24}+\frac{8}{2}\cdot\frac{1}{2}+\frac{4}{8}+\frac{4}{3}+\frac{2}{4}=5=\binom{5}{4} 24 16 + 2 8 ⋅ 2 1 + 8 4 + 3 4 + 4 2 = 5 = ( 4 5 ) , where the terms belong to the types 1 4 1^4 1 4 , 2 1 2 2\,1^2 2 1 2 , 2 2 2^2 2 2 , 3 1 3\,1 3 1 , 4 4 4 with z μ = 24 , 4 , 8 , 3 , 4 z_\mu=24,4,8,3,4 z μ = 24 , 4 , 8 , 3 , 4 . With weights, ∑ μ p μ / z μ = h n \sum_\mu p_\mu/z_\mu=h_n ∑ μ p μ / z μ = h n , the complete homogeneous symmetric function; this is the Frobenius characteristic of the trivial character (compare § 19.7 ).
Normal structure and the simplicity of A l t n \mathrm{Alt}_n Alt n
A normal subgroup is a union of conjugacy classes containing 1 1 1 whose size divides the group order. Conversely such a union N N N is a subgroup as soon as r h ∈ N rh\in N r h ∈ N for one representative r r r of each class in N N N and all h ∈ N h\in N h ∈ N : if r ′ = x r x − 1 r'=xrx^{-1} r ′ = x r x − 1 , then r ′ h = x r ( x − 1 h x ) x − 1 ∈ x N x − 1 = N r'h=xr(x^{-1}hx)x^{-1}\in xNx^{-1}=N r ′ h = x r ( x − 1 h x ) x − 1 ∈ x N x − 1 = N . This gives a finite test, which the script applies to all unions of classes.
Theorem 11.13 (Simplicity of A l t n \mathrm{Alt}_n Alt n ). A l t n \mathrm{Alt}_n Alt n is simple for n ≥ 5 n\ge5 n ≥ 5 .
Proof. For n = 5 n=5 n = 5 no partial sum of 1 + 15 + 20 + 12 + 12 1+15+20+12+12 1 + 15 + 20 + 12 + 12 that contains 1 1 1 divides 60 60 60 except 1 1 1 and 60 60 60 . Let n ≥ 5 n\ge5 n ≥ 5 and 1 ≠ N ⊴ A l t n 1\ne N\trianglelefteq\mathrm{Alt}_n 1 = N ⊴ Alt n , and pick σ ∈ N \sigma\in N σ ∈ N and a a a with b = σ ( a ) ≠ a b=\sigma(a)\ne a b = σ ( a ) = a . Choose c ∉ { a , b , σ ( b ) } c\notin\{a,b,\sigma(b)\} c ∈ / { a , b , σ ( b )} and put τ = ( a b c ) \tau=(a\,b\,c) τ = ( a b c ) . Then ρ = ( τ σ τ − 1 ) σ − 1 ∈ N \rho=(\tau\sigma\tau^{-1})\sigma^{-1}\in N ρ = ( τ σ τ − 1 ) σ − 1 ∈ N , and also ρ = τ ⋅ ( σ τ σ − 1 ) − 1 \rho=\tau\cdot(\sigma\tau\sigma^{-1})^{-1} ρ = τ ⋅ ( σ τ σ − 1 ) − 1 with σ τ σ − 1 = ( b σ ( b ) σ ( c ) ) ≠ τ \sigma\tau\sigma^{-1}=(b\,\sigma(b)\,\sigma(c))\ne\tau σ τ σ − 1 = ( b σ ( b ) σ ( c )) = τ because σ ( b ) ≠ c \sigma(b)\ne c σ ( b ) = c . So ρ ≠ 1 \rho\ne1 ρ = 1 moves only points of Y = { a , b , c , σ ( b ) , σ ( c ) } Y=\{a,b,c,\sigma(b),\sigma(c)\} Y = { a , b , c , σ ( b ) , σ ( c )} . Enlarge Y Y Y to 5 points; N ∩ A l t ( Y ) N\cap\mathrm{Alt}(Y) N ∩ Alt ( Y ) is normal in A l t ( Y ) ≅ A l t 5 \mathrm{Alt}(Y)\cong\mathrm{Alt}_5 Alt ( Y ) ≅ Alt 5 and contains ρ \rho ρ , so it is A l t ( Y ) \mathrm{Alt}(Y) Alt ( Y ) and N N N contains a 3-cycle. By Corollary 11.10 it contains all 3-cycles, which generate A l t n \mathrm{Alt}_n Alt n by Theorem 11.7 .
Theorem 11.14 (Normal subgroups of S y m n \mathrm{Sym}_n Sym n ). For n ≥ 5 n\ge5 n ≥ 5 the normal subgroups of S y m n \mathrm{Sym}_n Sym n are 1 1 1 , A l t n \mathrm{Alt}_n Alt n , S y m n \mathrm{Sym}_n Sym n ; S y m 4 \mathrm{Sym}_4 Sym 4 has in addition V 4 = { 1 , ( 0 1 ) ( 2 3 ) , ( 0 2 ) ( 1 3 ) , ( 0 3 ) ( 1 2 ) } V_4=\{1,(0\,1)(2\,3),(0\,2)(1\,3),(0\,3)(1\,2)\} V 4 = { 1 , ( 0 1 ) ( 2 3 ) , ( 0 2 ) ( 1 3 ) , ( 0 3 ) ( 1 2 )} . Moreover S y m n ′ = A l t n \mathrm{Sym}'_n=\mathrm{Alt}_n Sym n ′ = Alt n for n ≥ 2 n\ge2 n ≥ 2 , A l t 4 ′ = V 4 \mathrm{Alt}'_4=V_4 Alt 4 ′ = V 4 , and A l t n \mathrm{Alt}_n Alt n is perfect for n ≥ 5 n\ge5 n ≥ 5 .
Proof. Let N ⊴ Sym n N \trianglelefteq\operatorname{Sym}_n N ⊴ Sym n , n ≥ 5 n \ge5 n ≥ 5 . By Theorem 11.13 , N ∩ Alt n N \cap\operatorname{Alt}_n N ∩ Alt n is 1 1 1 or Alt n \operatorname{Alt}_n Alt n . In the second case N ∈ { Alt n , Sym n } N \in\{\operatorname{Alt}_n,\operatorname{Sym}_n\} N ∈ { Alt n , Sym n } . In the first, ∣ N ∣ ≤ 2 |N| \le2 ∣ N ∣ ≤ 2 , and a normal subgroup of order 2 2 2 would be central; but by Theorem 11.8 only the identity class has z μ = n ! z_\mu= n! z μ = n ! , so Z ( Sym n ) = 1 Z(\operatorname{Sym}_n) = 1 Z ( Sym n ) = 1 for n ≥ 3 n \ge3 n ≥ 3 . For the derived subgroups, Sym n / Alt n \operatorname{Sym}_n / \operatorname{Alt}_n Sym n / Alt n is abelian, and [ ( a b ) , ( b c ) ] = ( ( a b ) ( b c ) ) 2 [(a\,b),(b\,c)] = ((a\,b)(b\,c))^2 [( a b ) , ( b c )] = (( a b ) ( b c ) ) 2 is a 3 3 3 -cycle, so Sym n ′ = Alt n \operatorname{Sym}_n' = \operatorname{Alt}_n Sym n ′ = Alt n ; a simple nonabelian group is perfect. The case n = 4 n = 4 n = 4 is read off Table 11.1 . □
Table 11.1. Classes and normal subgroups of Sym n \operatorname{Sym}_n Sym n and Alt n \operatorname{Alt}_n Alt n (generated by scripts/sn_verify.py; cross-checked in GAP).
The script finds every normal subgroup of Sym n \operatorname{Sym}_n Sym n (n ≤ 7 n \le7 n ≤ 7 ) and Alt n \operatorname{Alt}_n Alt n (n ≤ 8 n \le8 n ≤ 8 ) by the union-of-classes test and computes Sym n ′ \operatorname{Sym}_n' Sym n ′ and Alt n ′ \operatorname{Alt}_n' Alt n ′ for n ≤ 6 n \le6 n ≤ 6 ; GAP confirms the normal subgroups for n ≤ 8 n \le8 n ≤ 8 , the derived subgroups, and the simplicity of Alt 5 , … , Alt 12 \operatorname{Alt}_5,\ldots,\operatorname{Alt}_{12} Alt 5 , … , Alt 12 .
Actions and maximal subgroups
An action of a group G G G on A n \mathcal{A}_n A n is a homomorphism G → Sym n G \to\operatorname{Sym}_n G → Sym n , so its image is a set of codes. The action is transitive if A n \mathcal{A}_n A n is one orbit. A block is a nonempty Δ ⊆ A n \Delta\subseteq\mathcal{A}_n Δ ⊆ A n such that, for every g ∈ G g \in G g ∈ G , either g Δ = Δ g\Delta= \Delta g Δ = Δ or g Δ ∩ Δ = ∅ g\Delta\cap\Delta= \varnothing g Δ ∩ Δ = ∅ ; singletons and A n \mathcal{A}_n A n are the trivial blocks, and the images of a block form a block system . A transitive action is primitive if it has only trivial blocks.
Proposition 11.15 (Coset actions). Let G G G act transitively on Ω \Omega Ω , let ω ∈ Ω \omega\in\Omega ω ∈ Ω and H = G ω H = G_\omega H = G ω . Then g ω ↦ g H g\omega\mapsto gH g ω ↦ g H identifies Ω \Omega Ω with the action on G / H G/H G / H , and K ↦ K ω K \mapsto K\omega K ↦ K ω is an inclusion-preserving bijection from the subgroups H ≤ K ≤ G H \le K \le G H ≤ K ≤ G onto the blocks containing ω \omega ω . In particular the action is primitive if and only if H H H is maximal in G G G .
Proof. The first statement is the orbit–stabiliser theorem. If H ≤ K H \le K H ≤ K and g k 1 ω = k 2 ω gk_1\omega= k_2\omega g k 1 ω = k 2 ω with k i ∈ K k_i \in K k i ∈ K , then k 2 − 1 g k 1 ∈ H ≤ K k_2^{-1}gk_1 \in H \le K k 2 − 1 g k 1 ∈ H ≤ K , so g ∈ K g \in K g ∈ K and g K ω = K ω gK\omega= K\omega g K ω = K ω : thus K ω K\omega K ω is a block, and its setwise stabiliser is K K K , since g K ω = K ω gK\omega= K\omega g K ω = K ω gives g ω = k ω g\omega= k\omega g ω = k ω and k − 1 g ∈ H k^{-1}g \in H k − 1 g ∈ H . Conversely, if Δ ∋ ω \Delta\ni\omega Δ ∋ ω is a block with setwise stabiliser G Δ ≥ H G_\Delta\ge H G Δ ≥ H and δ = g ω ∈ Δ \delta= g\omega\in\Delta δ = g ω ∈ Δ , then g Δ ∩ Δ ∋ δ g\Delta\cap\Delta\ni\delta g Δ ∩ Δ ∋ δ , so g ∈ G Δ g \in G_\Delta g ∈ G Δ ; hence Δ = G Δ ω \Delta= G_\Delta\omega Δ = G Δ ω . □
By Proposition 11.15 , a maximal subgroup M M M of Sym n \operatorname{Sym}_n Sym n or Alt n \operatorname{Alt}_n Alt n is of one of three kinds: intransitive , imprimitive (transitive, preserving a nontrivial block system), or primitive . The first two kinds are described completely by the next results.
Lemma 11.16 (Transpositions). A primitive subgroup of Sym n \operatorname{Sym}_n Sym n that contains a transposition is Sym n \operatorname{Sym}_n Sym n .
Proof. Write x ∼ y x \sim y x ∼ y if x = y x = y x = y or ( x y ) ∈ M (xy) \in M ( x y ) ∈ M . Since ( x y ) ( y z ) ( x y ) = ( x z ) (xy)(yz)(xy) = (xz) ( x y ) ( y z ) ( x y ) = ( x z ) , this is an equivalence relation, and since g ( x y ) g − 1 = ( g ( x ) g ( y ) ) g(xy)g^{-1} = (g(x)g(y)) g ( x y ) g − 1 = ( g ( x ) g ( y )) by Proposition 2.6 , its classes form a block system. One class has at least two points, so by primitivity there is one class, M M M contains every transposition, and M = Sym n M = \operatorname{Sym}_n M = Sym n by Theorem 11.5 . □
Theorem 11.17 (Intransitive maximal subgroups). Let A = [ 0 , k ) A=[0,k) A = [ 0 , k ) , B = [ k , n ) B=[k,n) B = [ k , n ) with 1 ≤ k < n / 2 1\le k<n/2 1 ≤ k < n /2 . Then Sym ( A ) × Sym ( B ) ≅ Sym k × Sym n − k \operatorname{Sym}(A)\times\operatorname{Sym}(B)\cong\operatorname{Sym}_k\times\operatorname{Sym}_{n-k} Sym ( A ) × Sym ( B ) ≅ Sym k × Sym n − k is maximal in Sym n \operatorname{Sym}_n Sym n . If n = 2 k n=2k n = 2 k it is not maximal, being properly contained in the stabiliser of the partition { A , B } \{A,B\} { A , B } .
Proof. Let H = Sym ( A ) × Sym ( B ) < M H=\operatorname{Sym}(A)\times\operatorname{Sym}(B)<M H = Sym ( A ) × Sym ( B ) < M and g ∈ M ∖ H g\in M\setminus H g ∈ M ∖ H , so g ( A ) ≠ A g(A)\ne A g ( A ) = A and there is a ∈ A a\in A a ∈ A with g ( a ) ∈ B g(a)\in B g ( a ) ∈ B . Since ∣ B ∣ > ∣ A ∣ \lvert B\rvert>\lvert A\rvert ∣ B ∣ > ∣ A ∣ , g ( B ) ⊈ A g(B)\not\subseteq A g ( B ) ⊆ A , so there is b ∈ B b\in B b ∈ B with g ( b ) ∈ B g(b)\in B g ( b ) ∈ B . Then ( g ( a ) g ( b ) ) ∈ H (g(a)\ g(b))\in H ( g ( a ) g ( b )) ∈ H , and g − 1 ( g ( a ) g ( b ) ) g = ( a b ) ∈ M g^{-1}(g(a)\ g(b))g=(a\ b)\in M g − 1 ( g ( a ) g ( b )) g = ( a b ) ∈ M . Conjugating ( a b ) (a\ b) ( a b ) by H H H gives every transposition meeting both A A A and B B B ; with those of H H H these are all transpositions, so M = Sym n M=\operatorname{Sym}_n M = Sym n . For n = 2 k n=2k n = 2 k , the involution i ↦ i + k ( m o d n ) i\mapsto i+k\pmod n i ↦ i + k ( mod n ) exchanges A A A and B B B and normalises H H H .
Definition 11.18. For n = a b n=ab n = ab with a , b ≥ 2 a,b\ge2 a , b ≥ 2 let Σ = { Σ 0 , … , Σ b − 1 } \Sigma=\{\Sigma_0,\ldots,\Sigma_{b-1}\} Σ = { Σ 0 , … , Σ b − 1 } , Σ j = [ j a , ( j + 1 ) a ) \Sigma_j=[ja,(j+1)a) Σ j = [ j a , ( j + 1 ) a ) , and let W a , b ≤ Sym n W_{a,b}\le\operatorname{Sym}_n W a , b ≤ Sym n be the stabiliser of the partition Σ \Sigma Σ . Then W a , b ≅ Sym a ≀ Sym b W_{a,b}\cong\operatorname{Sym}_a\wr\operatorname{Sym}_b W a , b ≅ Sym a ≀ Sym b has order ( a ! ) b b ! (a!)^b b! ( a ! ) b b ! .
Theorem 11.19 (Imprimitive maximal subgroups). For n = a b n=ab n = ab with a , b ≥ 2 a,b\ge2 a , b ≥ 2 , W a , b W_{a,b} W a , b is maximal in Sym n \operatorname{Sym}_n Sym n .
Proof. First, the only block systems of W = W a , b W=W_{a,b} W = W a , b are the trivial ones and Σ \Sigma Σ . Let x ∈ Σ ( x ) ∈ Σ x\in\Sigma(x)\in\Sigma x ∈ Σ ( x ) ∈ Σ . The orbits of W x W_x W x are { x } \{x\} { x } , Σ ( x ) ∖ { x } \Sigma(x)\setminus\{x\} Σ ( x ) ∖ { x } and A n ∖ Σ ( x ) \mathcal{A}_n\setminus\Sigma(x) A n ∖ Σ ( x ) , all nonempty, and a block through x x x is a union of W x W_x W x -orbits. Of the four candidates { x } ∪ ( A n ∖ Σ ( x ) ) \{x\}\cup(\mathcal{A}_n\setminus\Sigma(x)) { x } ∪ ( A n ∖ Σ ( x )) is not a block: a transposition ( x y ) (x\ y) ( x y ) with y ∈ Σ ( x ) y\in\Sigma(x) y ∈ Σ ( x ) maps it to a set that meets it and is different. Now let W < M ≤ Sym n W<M\le\operatorname{Sym}_n W < M ≤ Sym n . Every block of M M M is a block of W W W , and M M M does not preserve Σ \Sigma Σ , because W W W is its full stabiliser; so M M M is primitive. It contains the transposition ( 0 1 ) ∈ W (0\ 1)\in W ( 0 1 ) ∈ W , so M = Sym n M=\operatorname{Sym}_n M = Sym n by Lemma 11.16 .
In particular the point stabiliser Sym n − 1 \operatorname{Sym}_{n-1} Sym n − 1 (k = 1 k=1 k = 1 ) is maximal, and the natural action of Sym n \operatorname{Sym}_n Sym n on A n \mathcal{A}_n A n is primitive. The primitive maximal subgroups are described by the O’Nan–Scott theorem together with the work of Liebeck, Praeger and Saxl, which we state without proof.
Theorem 11.20 (O’Nan–Scott; Liebeck–Praeger–Saxl). Let G ∈ { Sym n , Alt n } G\in\{\operatorname{Sym}_n,\operatorname{Alt}_n\} G ∈ { Sym n , Alt n } and let M M M be a maximal subgroup of G G G not containing Alt n \operatorname{Alt}_n Alt n . Then M M M is one of: (i) ( Sym k × Sym n − k ) ∩ G (\operatorname{Sym}_k\times\operatorname{Sym}_{n-k})\cap G ( Sym k × Sym n − k ) ∩ G with 1 ≤ k < n / 2 1\le k<n/2 1 ≤ k < n /2 ; (ii) W a , b ∩ G W_{a,b}\cap G W a , b ∩ G with n = a b n=ab n = ab , a , b ≥ 2 a,b\ge2 a , b ≥ 2 ; (iii) a primitive group of affine, diagonal, product or almost simple type. Every subgroup in (i) and (ii) is maximal in G G G , with the single exception of W 2 , 4 ∩ Alt 8 W_{2,4}\cap\operatorname{Alt}_8 W 2 , 4 ∩ Alt 8 , which lies in a primitive subgroup AGL ( 3 , 2 ) \operatorname{AGL}(3,2) AGL ( 3 , 2 ) of order 1344.
Theorems 11.17 and 11.19 prove the case G = Sym n G=\operatorname{Sym}_n G = Sym n of the last statement. For codes, maximality is decided by a finite computation: ⟨ H , g ⟩ \langle H,g\rangle ⟨ H , g ⟩ depends only on the double coset H g H HgH H g H , so H H H is maximal in G G G if and only if ⟨ H , g ⟩ = G \langle H,g\rangle=G ⟨ H , g ⟩ = G for one g g g in each double coset H g H ≠ H HgH\ne H H g H = H . The script enumerates double cosets by breadth-first search under left and right multiplication by the generators of H H H , and confirms for 4 ≤ n ≤ 8 4\le n\le8 4 ≤ n ≤ 8 that the subgroups of (i) and (ii) and their intersections with Alt n \operatorname{Alt}_n Alt n are maximal, that Sym k × Sym k \operatorname{Sym}_k\times\operatorname{Sym}_k Sym k × Sym k is not, and that W 2 , 4 ∩ Alt 8 W_{2,4}\cap\operatorname{Alt}_8 W 2 , 4 ∩ Alt 8 is not, one of its overgroups having order 1344. GAP computes all classes of maximal subgroups for 4 ≤ n ≤ 8 4\le n\le8 4 ≤ n ≤ 8 ; Table 11.2 lists their orders by type, and the intransitive and imprimitive columns agree with Theorem 11.20 . Repeated entries are distinct conjugacy classes, for instance the two classes of PSL ( 2 , 7 ) \operatorname{PSL}(2,7) PSL ( 2 , 7 ) in Alt 7 \operatorname{Alt}_7 Alt 7 .
Table 11.2. Orders of the maximal subgroups of Sym n \operatorname{Sym}_n Sym n and Alt n \operatorname{Alt}_n Alt n , 4 ≤ n ≤ 8 4 \le n \le8 4 ≤ n ≤ 8 , one entry per conjugacy class, by type (computed in GAP 4.14 and checked by scripts/sn_verify.py).
Automorphisms, the outer automorphism of Sym 6 \operatorname{Sym}_6 Sym 6 , and the exceptional isomorphisms
For g ∈ G g\in G g ∈ G , conjugation x ↦ g x g − 1 x\mapsto gxg^{-1} x ↦ g x g − 1 is an automorphism of G G G ; these form the normal subgroup Inn ( G ) ≅ G / Z ( G ) \operatorname{Inn}(G)\cong G/Z(G) Inn ( G ) ≅ G / Z ( G ) of Aut ( G ) \operatorname{Aut}(G) Aut ( G ) . The quotient Out ( G ) = Aut ( G ) / Inn ( G ) \operatorname{Out}(G)=\operatorname{Aut}(G)/\operatorname{Inn}(G) Out ( G ) = Aut ( G ) / Inn ( G ) measures the failure of every automorphism to be inner. For symmetric and alternating groups the only exceptional degree is 6.
Theorem 11.21 (Automorphisms of Sym n \operatorname{Sym}_n Sym n ). For n ≠ 6 n\ne6 n = 6 every automorphism of Sym n \operatorname{Sym}_n Sym n is inner. For n = 6 n=6 n = 6 , ∣ Aut ( Sym 6 ) : Inn ( Sym 6 ) ∣ = 2 \lvert\operatorname{Aut}(\operatorname{Sym}_6):\operatorname{Inn}(\operatorname{Sym}_6)\rvert=2 ∣ Aut ( Sym 6 ) : Inn ( Sym 6 )∣ = 2 .
Proof. An automorphism preserves orders and class sizes, hence permutes conjugacy classes. For n ≠ 4 , 6 n \ne4,6 n = 4 , 6 , transpositions are characterised among involutions as the unique class of size ( n 2 ) \binom{n}{2} ( 2 n ) : the class of cycle type 2 m 1 n − 2 m 2^m1^{n-2m} 2 m 1 n − 2 m has size n ! / ( 2 m m ! ( n − 2 m ) ! ) n!/(2^m m!(n-2m)!) n ! / ( 2 m m ! ( n − 2 m )!) , and this equals ( n 2 ) \binom{n}{2} ( 2 n ) only for m = 1 m=1 m = 1 , except at n = 6 n=6 n = 6 , where m = 3 m=3 m = 3 also gives 15. Thus an automorphism of Sym n \operatorname{Sym}_n Sym n permutes the transpositions. Two transpositions commute exactly when they are disjoint, so the set of transpositions through a fixed point is characterised as a maximal pairwise noncommuting family of size n − 1 n-1 n − 1 . Hence the automorphism permutes the points and is conjugation by the resulting permutation.
For n = 4 n=4 n = 4 , the transpositions again form a distinguished class, of size 6, whereas the other involution class has size 3. Thus an automorphism permutes the six transpositions. The three transpositions through a fixed point are exactly the maximal pairwise noncommuting triples, so the four points are again recovered from the transposition graph, and the automorphism is inner. At n = 6 n=6 n = 6 , the transposition and triple-transposition classes both have size 15, so one further possibility remains.
Identify the transpositions of Sym 6 \operatorname{Sym}_6 Sym 6 with the edges of K 6 K_6 K 6 . A syntheme is a perfect matching, hence a set of three disjoint transpositions; there are 15 synthemes. A pentad is a set of five synthemes partitioning the 15 edges; there are six pentads. Relabelling the vertices gives an action of Sym 6 \operatorname{Sym}_6 Sym 6 on these six pentads.
Proposition 11.22 (The pentad action). The pentad action is faithful and transitive. A transposition of Sym 6 \operatorname{Sym}_6 Sym 6 acts on the six pentads as a triple transposition. Thus the induced automorphism of Sym 6 \operatorname{Sym}_6 Sym 6 is outer.
Proof. A permutation fixing every pentad fixes the incidence relation between edges and synthemes, hence every edge of K 6 K_6 K 6 , and is the identity. The six pentads are equivalent under vertex relabelling, so the action is transitive. The verifier constructs an explicit list of the six pentads and checks that a transposition swaps three disjoint pairs. Therefore the action gives an automorphism that carries the transposition class to the triple-transposition class, whereas inner automorphisms preserve every conjugacy class.
Theorem 11.23 (Automorphisms of Alt n \operatorname{Alt}_n Alt n ). Let n ≥ 4 n \ge4 n ≥ 4 . For n ≠ 6 n \ne6 n = 6 every automorphism of Alt n \operatorname{Alt}_n Alt n is induced by conjugation in Sym n \operatorname{Sym}_n Sym n , so Aut ( Alt n ) ≅ Sym n \operatorname{Aut}(\operatorname{Alt}_n) \cong\operatorname{Sym}_n Aut ( Alt n ) ≅ Sym n and Out ( Alt n ) ≅ C 2 \operatorname{Out}(\operatorname{Alt}_n) \cong C_2 Out ( Alt n ) ≅ C 2 . For n = 6 n=6 n = 6 , ∣ Aut ( Alt 6 ) ∣ = 1440 |\operatorname{Aut}(\operatorname{Alt}_6)|=1440 ∣ Aut ( Alt 6 ) ∣ = 1440 and ∣ Out ( Alt 6 ) ∣ = 4 |\operatorname{Out}(\operatorname{Alt}_6)|=4 ∣ Out ( Alt 6 ) ∣ = 4 .
Proof. For n ≥ 5 n \ge5 n ≥ 5 , Alt n \operatorname{Alt}_n Alt n is simple by Theorem 11.13 , and automorphisms permute its conjugacy classes. Away from degree 6, the class of 3-cycles is characterised by its size and centraliser
(end_mid=1) structure, so an automorphism permutes the point stabilisers in the natural action and is induced by a permutation of the points. Conjugation embeds Sym n \operatorname{Sym}_n Sym n in Aut ( Alt n ) \operatorname{Aut}(\operatorname{Alt}_n) Aut ( Alt n ) , and the preceding argument proves equality. For n = 4 n = 4 n = 4 , Table 11.3 records ∣ Aut ( Alt 4 ) ∣ = 24 |\operatorname{Aut}(\operatorname{Alt}_4)| = 24 ∣ Aut ( Alt 4 ) ∣ = 24 and ∣ Inn ( Alt 4 ) ∣ = ∣ Alt 4 / Z ( Alt 4 ) ∣ = 12 |\operatorname{Inn}(\operatorname{Alt}_4)| = |\operatorname{Alt}_4/Z(\operatorname{Alt}_4)| = 12 ∣ Inn ( Alt 4 ) ∣ = ∣ Alt 4 / Z ( Alt 4 ) ∣ = 12 , so ∣ Out ( Alt 4 ) ∣ = 2 |\operatorname{Out}(\operatorname{Alt}_4)| = 2 ∣ Out ( Alt 4 ) ∣ = 2 . At degree 6, the outer automorphism of Sym 6 \operatorname{Sym}_6 Sym 6 restricts to an additional automorphism of Alt 6 \operatorname{Alt}_6 Alt 6 ; the asserted orders are verified in Table 11.3 .
Table 11.3. Automorphism orders and exceptional actions for Section 11.6.
Theorem 11.24 (Exceptional isomorphisms). There are isomorphisms
PSL ( 2 , 4 ) ≅ PSL ( 2 , 5 ) ≅ Alt 5 , PSL ( 2 , 9 ) ≅ Alt 6 , \operatorname{PSL}(2,4) \cong\operatorname{PSL}(2,5) \cong\operatorname{Alt}_5,\qquad\operatorname{PSL}(2,9) \cong\operatorname{Alt}_6, PSL ( 2 , 4 ) ≅ PSL ( 2 , 5 ) ≅ Alt 5 , PSL ( 2 , 9 ) ≅ Alt 6 , PSp ( 4 , 2 ) ≅ Sym 6 , PSL ( 4 , 2 ) ≅ Alt 8 . \operatorname{PSp}(4,2) \cong\operatorname{Sym}_6,\qquad\operatorname{PSL}(4,2) \cong\operatorname{Alt}_8. PSp ( 4 , 2 ) ≅ Sym 6 , PSL ( 4 , 2 ) ≅ Alt 8 .
Proof. The order formula ∣ PSL ( 2 , q ) ∣ = q ( q 2 − 1 ) / gcd ( 2 , q − 1 ) |\operatorname{PSL}(2,q)| = q(q^2 - 1)/\gcd(2,q - 1) ∣ PSL ( 2 , q ) ∣ = q ( q 2 − 1 ) / g cd( 2 , q − 1 ) gives order 60 for q = 4 , 5 q = 4,5 q = 4 , 5 and order 360 for q = 9 q = 9 q = 9 . For q = 4 q = 4 q = 4 and q = 5 q = 5 q = 5 , the natural action on the projective line P 1 ( F q ) \mathbb{P}^1(\mathbb{F}_q) P 1 ( F q ) has degrees 5 and 6, and scripts/sn_verify.py checks that the resulting faithful images are respectively Alt 5 \operatorname{Alt}_5 Alt 5 and a transitive subgroup of Alt 6 \operatorname{Alt}_6 Alt 6 isomorphic to Alt 5 \operatorname{Alt}_5 Alt 5 . For PSL ( 2 , 9 ) \operatorname{PSL}(2,9) PSL ( 2 , 9 ) , the verifier takes the action on the cosets of a subgroup of order 60, obtaining a faithful degree-6 action with image Alt 6 \operatorname{Alt}_6 Alt 6 .
Likewise ∣ Sp ( 4 , 2 ) ∣ = 720 |\operatorname{Sp}(4,2)| = 720 ∣ Sp ( 4 , 2 ) ∣ = 720 and ∣ GL ( 4 , 2 ) ∣ = 20160 |\operatorname{GL}(4,2)| = 20160 ∣ GL ( 4 , 2 ) ∣ = 20160 . The verifier takes the actions on the cosets of subgroups of orders 120 and 2520, obtaining faithful permutation representations of degrees 6 and 8 whose images are Sym 6 \operatorname{Sym}_6 Sym 6 and Alt 8 \operatorname{Alt}_8 Alt 8 . The corresponding automorphism orders and exceptional actions are listed in Table 11.3 .
The exceptional degree 6 is visible twice: ∣ Out ( Sym 6 ) ∣ = 2 |\operatorname{Out}(\operatorname{Sym}_6)| = 2 ∣ Out ( Sym 6 ) ∣ = 2 and ∣ Out ( Alt 6 ) ∣ = 4 |\operatorname{Out}(\operatorname{Alt}_6)| = 4 ∣ Out ( Alt 6 ) ∣ = 4 . The verification script constructs the pentad action explicitly and cross-checks the automorphism-group orders and exceptional isomorphisms in GAP.
The conjugacy classes of Sym n \operatorname{Sym}_n Sym n are indexed by partitions μ ⊢ n \mu\vdash n μ ⊢ n (§ 19.3 ); the same set also indexes the irreducible complex characters. This section constructs the corresponding modules and gives three complementary methods for their dimensions and character values [14, 16, 17].
Characters of Sym(n)
Specht modules
Let λ = ( λ 1 , … , λ r ) ⊢ n \lambda= (\lambda_1,\ldots,\lambda_r) \vdash n λ = ( λ 1 , … , λ r ) ⊢ n . Its Young diagram is [ λ ] = { ( i , j ) : 1 ≤ i ≤ r , 1 ≤ j ≤ λ i } [\lambda] = \{(i,j): 1 \leq i \leq r,\ 1 \leq j \leq\lambda_i\} [ λ ] = {( i , j ) : 1 ≤ i ≤ r , 1 ≤ j ≤ λ i } . A λ \lambda λ -tableau is a bijective filling of [ λ ] [\lambda] [ λ ] by { 1 , … , n } \{1,\ldots,n\} { 1 , … , n } . Two tableaux are row-equivalent when every row contains the same entries; a row-equivalence class is a tabloid { t } \{t\} { t } . The permutation module M λ M^\lambda M λ has the λ \lambda λ -tabloids as an orthonormal basis.
For a tableau t t t , let C t C_t C t be the subgroup preserving each column setwise and put
e t = ∑ π ∈ C t sgn ( π ) π { t } ∈ M λ . e_t = \sum_{\pi\in C_t} \operatorname{sgn}(\pi)\,\pi\{t\} \in M^\lambda. e t = π ∈ C t ∑ sgn ( π ) π { t } ∈ M λ .
The Specht module is S λ = ⟨ e t : t a λ -tableau ⟩ S^\lambda= \langle e_t : t\text{ a }\lambda\text{-tableau}\rangle S λ = ⟨ e t : t a λ -tableau ⟩ . Since g e t = e g t g e_t = e_{gt} g e t = e g t , it is a Sym n \operatorname{Sym}_n Sym n -submodule.
Theorem 11.25 (Standard basis and classification). Over a field of characteristic zero, the polytabloids e t e_t e t for standard tableaux t t t form a basis of S λ S^\lambda S λ . The modules S λ S^\lambda S λ , λ ⊢ n \lambda\vdash n λ ⊢ n , are pairwise nonisomorphic irreducible modules and form a complete list of the irreducible Sym n \operatorname{Sym}_n Sym n -modules.
Proof. The straightening algorithm, using the column antisymmetry and Garnir relations, writes every polytabloid as a linear combination of standard polytabloids; triangularity with respect to dominance proves their independence. James’s submodule theorem says that every submodule of M λ M^\lambda M λ either contains S λ S^\lambda S λ or lies in its orthogonal complement. Since the invariant form is nonzero on S λ S^\lambda S λ in characteristic zero, S λ S^\lambda S λ is irreducible. The dominance lemma shows that S λ ≅ S μ S^\lambda\cong S^\mu S λ ≅ S μ forces λ = μ \lambda= \mu λ = μ . The number of partitions of n n n equals the number of conjugacy classes of Sym n \operatorname{Sym}_n Sym n (§ 19.3 ), which is the number of irreducible characters, so these modules exhaust the irreducibles. Full details of straightening and the submodule theorem are in [14].
Hook lengths
For c = ( i , j ) ∈ [ λ ] c = (i,j) \in[\lambda] c = ( i , j ) ∈ [ λ ] , its hook consists of c c c , the cells to its right, and the cells below it; write h ( c ) h(c) h ( c ) for its size.
Theorem 11.26 (Frame–Robinson–Thrall hook-length formula). The number f λ f^\lambda f λ of standard tableaux of shape λ \lambda λ , and hence the degree of the character χ λ \chi^\lambda χ λ afforded by S λ S^\lambda S λ , is
f λ = dim S λ = n ! ∏ c ∈ [ λ ] h ( c ) . f^\lambda= \dim S^\lambda= \frac{n!}{\prod_{c \in[\lambda]} h(c)}. f λ = dim S λ = ∏ c ∈ [ λ ] h ( c ) n ! .
Proof. A standard tableau has n n n in a removable corner, so f λ = ∑ c f λ − c f^\lambda= \sum_c f^{\lambda-c} f λ = ∑ c f λ − c over the removable corners c c c of [ λ ] [\lambda] [ λ ] , and f ∅ = 1 f^\varnothing= 1 f ∅ = 1 . The formula therefore follows by induction once F ( λ ) = n ! / ∏ c h ( c ) F(\lambda) = n!/\prod_c h(c) F ( λ ) = n ! / ∏ c h ( c ) is shown to satisfy the same recurrence F ( λ ) = ∑ c F ( λ − c ) F(\lambda) = \sum_c F(\lambda-c) F ( λ ) = ∑ c F ( λ − c ) . That identity is the nontrivial step; it is proved in [18], and the hook-walk proof of Greene, Nijenhuis and Wilf is given in [16].
Example 11.27. For λ = ( 3 , 2 ) \lambda= (3,2) λ = ( 3 , 2 ) the hook lengths are 4 , 3 , 1 , 2 , 1 4,3,1,2,1 4 , 3 , 1 , 2 , 1 , so f ( 3 , 2 ) = 5 ! / ( 4 ⋅ 3 ⋅ 1 ⋅ 2 ⋅ 1 ) = 5 f^{(3,2)} = 5!/(4\cdot3\cdot1\cdot2\cdot1) = 5 f ( 3 , 2 ) = 5 ! / ( 4 ⋅ 3 ⋅ 1 ⋅ 2 ⋅ 1 ) = 5 .
Put p k ( x ) = ∑ i x i k p_k(x)=\sum_i x_i^k p k ( x ) = ∑ i x i k , p μ = ∏ j p μ j p_\mu=\prod_j p_{\mu_j} p μ = ∏ j p μ j , δ = ( r − 1 , r − 2 , … , 0 ) \delta=(r-1,r-2,\ldots,0) δ = ( r − 1 , r − 2 , … , 0 ) , and Δ ( x ) = ∏ i < j ( x i − x j ) \Delta(x)=\prod_{i<j}(x_i-x_j) Δ ( x ) = ∏ i < j ( x i − x j ) . If r ≥ ℓ ( λ ) r\geq\ell(\lambda) r ≥ ℓ ( λ ) , Frobenius’s coefficient formula is
χ λ ( μ ) = [ x λ + δ ] Δ ( x ) p μ ( x ) . (82) \chi^\lambda(\mu)=[x^{\lambda+\delta}]\Delta(x)p_\mu(x). \tag*{(82)} χ λ ( μ ) = [ x λ + δ ] Δ ( x ) p μ ( x ) . (82)
Equivalently, the Schur function s λ s_\lambda s λ is the Frobenius characteristic of S λ S^\lambda S λ , and [17]
p μ = ∑ λ ⊢ n χ λ ( μ ) s λ . p_\mu=\sum_{\lambda\vdash n}\chi^\lambda(\mu)s_\lambda. p μ = λ ⊢ n ∑ χ λ ( μ ) s λ .
Example 11.28. With λ = ( 2 , 1 ) \lambda=(2,1) λ = ( 2 , 1 ) , μ = ( 2 , 1 ) \mu=(2,1) μ = ( 2 , 1 ) and two variables, the coefficient of x 1 3 x 2 x_1^3x_2 x 1 3 x 2 in ( x 1 − x 2 ) ( x 1 2 + x 2 2 ) ( x 1 + x 2 ) (x_1-x_2)(x_1^2+x_2^2)(x_1+x_2) ( x 1 − x 2 ) ( x 1 2 + x 2 2 ) ( x 1 + x 2 ) is 0 0 0 . Hence χ ( 2 , 1 ) ( 2 , 1 ) = 0 \chi^{(2,1)}(2,1)=0 χ ( 2 , 1 ) ( 2 , 1 ) = 0 .
Murnaghan–Nakayama
A border strip is a connected skew diagram containing no 2 × 2 2\times2 2 × 2 square; its height is one less than its number of rows.
Theorem 11.29 (Murnaghan–Nakayama rule). For μ = ( μ 1 , … , μ s ) ⊢ n \mu=(\mu_1,\ldots,\mu_s)\vdash n μ = ( μ 1 , … , μ s ) ⊢ n ,
χ λ ( μ ) = ∑ ν ( − 1 ) ht ( λ / ν ) χ ν ( μ 2 , … , μ s ) , (83) \chi^\lambda(\mu)=\sum_\nu(-1)^{\operatorname{ht}(\lambda/\nu)}\chi^\nu(\mu_2,\ldots,\mu_s), \tag*{(83)} χ λ ( μ ) = ν ∑ ( − 1 ) ht ( λ / ν ) χ ν ( μ 2 , … , μ s ) , (83) where the sum is over ν ⊆ λ \nu\subseteq\lambda ν ⊆ λ for which λ / ν \lambda/\nu λ / ν is a border strip of size μ 1 \mu_1 μ 1 , and $\chi^\varnothing(\varnothing)=1.
Proof. Take r ≥ n r\geq n r ≥ n variables. For β 1 > ⋯ > β r ≥ 0 \beta_1>\cdots>\beta_r\geq0 β 1 > ⋯ > β r ≥ 0 put a β = det ( x i β j ) a_\beta=\det(x_i^{\beta_j}) a β = det ( x i β j ) , so that a δ = Δ ( x ) a_\delta=\Delta(x) a δ = Δ ( x ) and s ν = a ν + δ / a δ s_\nu=a_{\nu+\delta}/a_\delta s ν = a ν + δ / a δ . The polynomial Δ ( x ) p μ \Delta(x)p_\mu Δ ( x ) p μ is alternating, hence a linear combination of the a κ + δ a_{\kappa+\delta} a κ + δ with κ ⊢ n \kappa\vdash n κ ⊢ n ; since x λ + δ x^{\lambda+\delta} x λ + δ occurs in a κ + δ a_{\kappa+\delta} a κ + δ only for κ = λ \kappa=\lambda κ = λ , with coefficient $1, formula (82) is equivalent to p μ = ∑ λ ⊢ n χ λ ( μ ) s λ p_\mu=\sum_{\lambda\vdash n}\chi^\lambda(\mu)s_\lambda p μ = ∑ λ ⊢ n χ λ ( μ ) s λ .
Now p k a β = ∑ i = 1 r a β + k e i p_ka_\beta=\sum_{i=1}^r a_{\beta+ke_i} p k a β = ∑ i = 1 r a β + k e i . A term with a repeated entry vanishes. Otherwise, sorting β + k e i \beta+ke_i β + k e i into decreasing order moves the entry β i + k \beta_i+k β i + k past the h h h entries β j \beta_j β j with β i < β j < β i + k \beta_i<\beta_j<\beta_i+k β i < β j < β i + k , so the term equals ( − 1 ) h a λ + δ (-1)^ha_{\lambda+\delta} ( − 1 ) h a λ + δ , where λ \lambda λ has beta-set { β j } j ≠ i ∪ { β i + k } \{\beta_j\}_{j\neq i}\cup\{\beta_i+k\} { β j } j = i ∪ { β i + k } . With β = ν + δ \beta=\nu+\delta β = ν + δ , the skew diagram λ / ν \lambda/\nu λ / ν is then a border strip of size k k k added to ν \nu ν , of height h h h , and every such strip arises exactly once. Hence
p k s ν = ∑ λ ( − 1 ) ht ( λ / ν ) s λ , p_ks_\nu=\sum_\lambda(-1)^{\operatorname{ht}(\lambda/\nu)}s_\lambda, p k s ν = λ ∑ ( − 1 ) ht ( λ / ν ) s λ ,
the sum over λ ⊇ ν \lambda\supseteq\nu λ ⊇ ν with λ / ν \lambda/\nu λ / ν a border strip of size k k k . Write μ ˉ = ( μ 2 , … , μ s ) \bar\mu=(\mu_2,\ldots,\mu_s) μ ˉ = ( μ 2 , … , μ s ) , so p μ = p μ 1 p μ ˉ = ∑ ν χ ν ( μ ˉ ) p μ 1 s ν p_\mu=p_{\mu_1}p_{\bar\mu}=\sum_\nu\chi^\nu(\bar\mu)p_{\mu_1}s_\nu p μ = p μ 1 p μ ˉ = ∑ ν χ ν ( μ ˉ ) p μ 1 s ν . Comparing coefficients of s λ s_\lambda s λ gives (83) ; see also [16, 17]. □ \square □
Example 11.30. For λ = ( 3 , 2 ) \lambda=(3,2) λ = ( 3 , 2 ) and cycle type ( 3 , 2 ) (3,2) ( 3 , 2 ) , the valid successive border-strip removals have signed total 1 1 1 ; hence χ ( 3 , 2 ) ( 3 , 2 ) = 1 \chi^{(3,2)}(3,2)=1 χ ( 3 , 2 ) ( 3 , 2 ) = 1 .
Remark 11.31 (Checks). The exact-arithmetic script scripts/sn_char_verify.py independently implements Frobenius’s coefficient formula and the Murnaghan–Nakayama recursion, compares every value for n ≤ 7 n\leq7 n ≤ 7 , verifies row orthogonality and ∑ λ χ λ ( 1 ) 2 = n ! \sum_\lambda\chi^\lambda(1)^2=n! ∑ λ χ λ ( 1 ) 2 = n ! , and writes Table 11.4 (PDF p. 141) . It also compares the hook formula with the corner-removal recurrence for standard tableaux for every partition of n ≤ 12 n\leq12 n ≤ 12 . For n ≤ 5 n\leq5 n ≤ 5 it realises S λ S^\lambda S λ as the span of the Specht polynomials F t F_t F t , the product over the columns of t t t of the Vandermonde determinants in the variables indexed by that column; this module is isomorphic to S λ S^\lambda S λ [16]. It checks that the F t F_t F t for standard t t t are linearly independent, that their span is stable under a transposition and an n n n -cycle, and that the trace of a permutation of each cycle type equals the Murnaghan–Nakayama value. When GAP is available, every value for n ≤ 7 n\leq7 n ≤ 7 is compared with CharacterTable("Symmetric", n) from the GAP character table library [30]. The script is invoked by scripts/sn_verify.py.
Table 11.4 (PDF p. 141). The irreducible character table of Sym 5 \operatorname{Sym}_5 Sym 5 . Rows and columns are indexed by partitions of 5; the first column is the degree.
Characters of Alt n \operatorname{Alt}_n Alt n
Throughout, n ≥ 2 n \ge2 n ≥ 2 , λ ′ \lambda' λ ′ denotes the conjugate (transposed) partition of λ ⊢ n \lambda\vdash n λ ⊢ n , and ℓ ( μ ) \ell(\mu) ℓ ( μ ) is the number of parts of μ \mu μ . Since Alt n \operatorname{Alt}_n Alt n has index 2 in Sym n \operatorname{Sym}_n Sym n , its irreducible characters are obtained from those of § 19.7 by restriction. The only questions are which restrictions stay irreducible and how the others split; the answer is governed by the self-conjugate partitions and by the classes of Sym n \operatorname{Sym}_n Sym n that split in Alt n \operatorname{Alt}_n Alt n (§ 19.3 ) [15, 10].
Lemma 11.32 (Sign twist). S λ ′ ≅ S λ ⊗ sgn S^{\lambda'} \cong S^\lambda\otimes\operatorname{sgn} S λ ′ ≅ S λ ⊗ sgn . Hence χ λ ′ ( g ) = sgn ( g ) χ λ ( g ) \chi^{\lambda'}(g)=\operatorname{sgn}(g)\chi^\lambda(g) χ λ ′ ( g ) = sgn ( g ) χ λ ( g ) for all g ∈ Sym n g \in\operatorname{Sym}_n g ∈ Sym n .
Proof. The module isomorphism is proved in [14]. For the characters, use (83) and induction on ℓ ( μ ) \ell(\mu) ℓ ( μ ) . Transposition is a bijection between the border strips of size k k k in λ \lambda λ and in λ ′ \lambda' λ ′ . A strip with r r r rows and c c c columns has r + c − 1 = k r+c-1=k r + c − 1 = k , so its transpose has height k − 1 − ht k-1-\operatorname{ht} k − 1 − ht . Each term of (83) for λ ′ \lambda' λ ′ is therefore ( − 1 ) μ 1 − 1 sgn ( μ ˉ ) (-1)^{\mu_1-1}\operatorname{sgn}(\bar{\mu}) ( − 1 ) μ 1 − 1 sgn ( μ ˉ ) times the corresponding term for λ \lambda λ , where μ ˉ = ( μ 2 , … ) \bar{\mu}=(\mu_2,\ldots) μ ˉ = ( μ 2 , … ) . Since ( − 1 ) μ 1 − 1 sgn ( μ ˉ ) = ( − 1 ) n − ℓ ( μ ) = sgn ( μ ) (-1)^{\mu_1-1}\operatorname{sgn}(\bar{\mu})=(-1)^{n-\ell(\mu)}=\operatorname{sgn}(\mu) ( − 1 ) μ 1 − 1 sgn ( μ ˉ ) = ( − 1 ) n − ℓ ( μ ) = sgn ( μ ) , the claim follows.
Theorem 11.33 (Restriction to Alt n \operatorname{Alt}_n Alt n ). Let λ ⊢ n \lambda\vdash n λ ⊢ n .
If λ ≠ λ ′ \lambda\ne\lambda' λ = λ ′ , then χ λ ∣ Alt n = χ λ ′ ∣ Alt n \chi^\lambda|_{\operatorname{Alt}_n}=\chi^{\lambda'}|_{\operatorname{Alt}_n} χ λ ∣ Alt n = χ λ ′ ∣ Alt n is irreducible.
If λ = λ ′ \lambda=\lambda' λ = λ ′ , then χ λ ∣ Alt n = χ + λ + χ − λ \chi^\lambda|_{\operatorname{Alt}_n}=\chi^\lambda_++\chi^\lambda_- χ λ ∣ Alt n = χ + λ + χ − λ , where χ ± λ \chi^\lambda_\pm χ ± λ are distinct irreducible characters interchanged by conjugation by any odd permutation.
Each irreducible character of Alt n \operatorname{Alt}_n Alt n occurs exactly once in this list: one character for each pair { λ , λ ′ } \{\lambda,\lambda'\} { λ , λ ′ } with λ ≠ λ ′ \lambda\ne\lambda' λ = λ ′ , and two for each λ = λ ′ \lambda=\lambda' λ = λ ′ .
Proof. Write G = Sym n G=\operatorname{Sym}_n G = Sym n and H = Alt n H=\operatorname{Alt}_n H = Alt n , and fix τ ∈ G ∖ H \tau\in G\setminus H τ ∈ G ∖ H . Since Ind H G Res H G χ = χ ⋅ Ind H G 1 = χ ( 1 + sgn ) \operatorname{Ind}^G_H\operatorname{Res}^G_H\chi=\chi\cdot\operatorname{Ind}^G_H1=\chi(1+\operatorname{sgn}) Ind H G Res H G χ = χ ⋅ Ind H G 1 = χ ( 1 + sgn ) , Frobenius reciprocity and Lemma 11.32 give
⟨ χ λ ∣ H , χ μ ∣ H ⟩ H = ⟨ χ λ , χ μ ⟩ G + ⟨ χ λ , χ μ ′ ⟩ G . \langle\chi^\lambda|_H,\chi^\mu|_H\rangle_H=\langle\chi^\lambda,\chi^\mu\rangle_G+\langle\chi^\lambda,\chi^{\mu'}\rangle_G. ⟨ χ λ ∣ H , χ μ ∣ H ⟩ H = ⟨ χ λ , χ μ ⟩ G + ⟨ χ λ , χ μ ′ ⟩ G .
With μ = λ \mu=\lambda μ = λ , the norm of χ λ ∣ H \chi^\lambda|_H χ λ ∣ H is 1 if λ ≠ λ ′ \lambda\ne\lambda' λ = λ ′ and 2 if λ = λ ′ \lambda=\lambda' λ = λ ′ . In the first case χ λ ∣ H \chi^\lambda|_H χ λ ∣ H is irreducible, and it equals χ λ ′ ∣ H \chi^{\lambda'}|_H χ λ ′ ∣ H because sgn \operatorname{sgn} sgn is trivial on H H H . In the second case χ λ ∣ H = ψ 1 + ψ 2 \chi^\lambda|_H=\psi_1+\psi_2 χ λ ∣ H = ψ 1 + ψ 2 with distinct irreducible ψ i \psi_i ψ i . By reciprocity, χ λ \chi^\lambda χ λ is a constituent of Ind H G ψ 1 \operatorname{Ind}^G_H\psi_1 Ind H G ψ 1 , whose degree is 2 ψ 1 ( 1 ) = χ λ ( 1 ) 2\psi_1(1)=\chi^\lambda(1) 2 ψ 1 ( 1 ) = χ λ ( 1 ) ; hence Ind H G ψ 1 = χ λ \operatorname{Ind}^G_H\psi_1=\chi^\lambda Ind H G ψ 1 = χ λ . Restricting back gives ψ 1 + ψ 1 τ = ψ 1 + ψ 2 \psi_1+\psi_1^\tau=\psi_1+\psi_2 ψ 1 + ψ 1 τ = ψ 1 + ψ 2 , so ψ 2 = ψ 1 τ \psi_2=\psi_1^\tau ψ 2 = ψ 1 τ . The displayed identity also shows that the restrictions of χ λ \chi^\lambda χ λ and χ μ \chi^\mu χ μ have no common constituent unless μ ∈ { λ , λ ′ } \mu\in\{\lambda,\lambda'\} μ ∈ { λ , λ ′ } . Finally, every ψ ∈ Irr ( H ) \psi\in\operatorname{Irr}(H) ψ ∈ Irr ( H ) is a constituent of χ ∣ H \chi|_H χ ∣ H for any irreducible constituent χ \chi χ of Ind H G ψ \operatorname{Ind}^G_H\psi Ind H G ψ .
For λ = λ ′ \lambda=\lambda' λ = λ ′ let k k k be the number of diagonal cells ( i , i ) ∈ [ λ ] (i,i)\in[\lambda] ( i , i ) ∈ [ λ ] , and let h ( λ ) = ( h 1 , … , h k ) h(\lambda)=(h_1,\ldots,h_k) h ( λ ) = ( h 1 , … , h k ) be the lengths of the hooks at ( 1 , 1 ) , … , ( k , k ) (1,1),\ldots,(k,k) ( 1 , 1 ) , … , ( k , k ) .
Lemma 11.34 (Diagonal hooks). The map λ ↦ h ( λ ) \lambda\mapsto h(\lambda) λ ↦ h ( λ ) is a bijection from the self-conjugate partitions of n n n to the partitions of n n n into distinct odd parts.
Proof. If λ = λ ′ \lambda=\lambda' λ = λ ′ , the hook at ( i , i ) (i,i) ( i , i ) has arm and leg both equal to a i = λ i − i a_i=\lambda_i-i a i = λ i − i , so h i = 2 a i + 1 h_i=2a_i+1 h i = 2 a i + 1 is odd. The numbers a 1 > a 2 > ⋯ > a k a_1>a_2>\cdots>a_k a 1 > a 2 > ⋯ > a k are strictly decreasing, and the k k k diagonal hooks partition [ λ ] [\lambda] [ λ ] , so h ( λ ) h(\lambda) h ( λ ) is a partition of n n n into distinct odd parts. Conversely, distinct odd parts h 1 > ⋯ > h k h_1>\cdots>h_k h 1 > ⋯ > h k determine strictly decreasing arms ( h i − 1 ) / 2 (h_i-1)/2 ( h i − 1 ) /2 , and nesting the corresponding symmetric hooks along the diagonal reconstructs the unique self-conjugate λ \lambda λ with h ( λ ) = ( h 1 , … , h k ) h(\lambda)=(h_1,\ldots,h_k) h ( λ ) = ( h 1 , … , h k ) .
By § 19.3 , the classes of Sym n \operatorname{Sym}_n Sym n that split in Alt n \operatorname{Alt}_n Alt n are exactly those whose cycle types have distinct odd parts. By Lemma 11.34 they are as numerous as the self-conjugate partitions, so the count of irreducible characters in Theorem 11.33 (3) agrees with the number of conjugacy classes of Alt n \operatorname{Alt}_n Alt n , as it must.
Theorem 11.35 (Frobenius: the split characters). Let λ = λ ′ \lambda=\lambda' λ = λ ′ , h ( λ ) = ( h 1 , … , h k ) h(\lambda)=(h_1,\ldots,h_k) h ( λ ) = ( h 1 , … , h k ) , q λ = h 1 h 2 ⋯ h k q_\lambda=h_1h_2\cdots h_k q λ = h 1 h 2 ⋯ h k , and ϵ λ = ( − 1 ) ( n − k ) / 2 \epsilon_\lambda=(-1)^{(n-k)/2} ϵ λ = ( − 1 ) ( n − k ) /2 . The Sym n \operatorname{Sym}_n Sym n -class of cycle type h ( λ ) h(\lambda) h ( λ ) splits into two Alt n \operatorname{Alt}_n Alt n -classes h + h^+ h + and h − h^- h − , which can be labelled so that
χ ± λ ( h + ) = 1 2 ( ϵ λ ± ϵ λ q λ ) , χ ± λ ( h − ) = 1 2 ( ϵ λ ∓ ϵ λ q λ ) . \chi^\lambda_\pm(h^+)=\frac{1}{2}\left(\epsilon_\lambda\pm\sqrt{\epsilon_\lambda q_\lambda}\right),\qquad\chi^\lambda_\pm(h^-)=\frac{1}{2}\left(\epsilon_\lambda\mp\sqrt{\epsilon_\lambda q_\lambda}\right). χ ± λ ( h + ) = 2 1 ( ϵ λ ± ϵ λ q λ ) , χ ± λ ( h − ) = 2 1 ( ϵ λ ∓ ϵ λ q λ ) . On every other conjugacy class C C C of Alt n \operatorname{Alt}_n Alt n , χ + λ ( C ) = χ − λ ( C ) = 1 2 χ λ ( C ) \chi^\lambda_+(C)=\chi^\lambda_-(C)=\frac{1}{2}\chi^\lambda(C) χ + λ ( C ) = χ − λ ( C ) = 2 1 χ λ ( C ) .
Proof. The class of type h ( λ ) h(\lambda) h ( λ ) splits by Lemma 11.34 . If a class C C C of Alt n \operatorname{Alt}_n Alt n is a full class of Sym n \operatorname{Sym}_n Sym n , then τ C τ − 1 = C \tau C\tau^{-1}=C τ C τ − 1 = C , so Theorem 11.33 (2) gives χ − λ ( C ) = χ + λ ( C ) \chi^\lambda_-(C)=\chi^\lambda_+(C) χ − λ ( C ) = χ + λ ( C ) , and both equal 1 2 χ λ ( C ) \frac{1}{2}\chi^\lambda(C) 2 1 χ λ ( C ) . That the two characters also agree on the split classes whose type is not h ( λ ) h(\lambda) h ( λ ) , together with the values on h ± h^\pm h ± , is Frobenius’s theorem; see [10, §5.1] and [15]. The two values on h ± h^\pm h ± sum to χ λ ( h ( λ ) ) = ϵ λ \chi^\lambda(h(\lambda))=\epsilon_\lambda χ λ ( h ( λ )) = ϵ λ , as Theorem 11.33 (2) requires.
Fields of values
Corollary 11.36 (Fields of values). For λ ≠ λ ′ \lambda\ne\lambda' λ = λ ′ the character χ λ ∣ Alt n \left.\chi^\lambda\right|_{\operatorname{Alt}_n} χ λ Alt n is rational-valued. For λ = λ ′ \lambda=\lambda' λ = λ ′ , the characters χ ± λ \chi^\lambda_\pm χ ± λ are real-valued if and only if ϵ λ = 1 \epsilon_\lambda=1 ϵ λ = 1 , that is n ≡ k ( m o d 4 ) n\equiv k\pmod{4} n ≡ k ( mod 4 ) , and rational-valued if and only if ϵ λ q λ \epsilon_\lambda q_\lambda ϵ λ q λ is a perfect square. Otherwise their values generate Q ( ϵ λ q λ ) \mathbb{Q}\left(\sqrt{\epsilon_\lambda q_\lambda}\right) Q ( ϵ λ q λ ) .
Example 11.37 (Alt 4 \operatorname{Alt}_4 Alt 4 ). The only self-conjugate partition of 4 4 4 is ( 2 , 2 ) (2,2) ( 2 , 2 ) , with h = ( 3 , 1 ) h=(3,1) h = ( 3 , 1 ) , k = 2 k=2 k = 2 , ϵ = − 1 \epsilon=-1 ϵ = − 1 and q = 3 q=3 q = 3 . Hence χ ( 2 , 2 ) \chi^{(2,2)} χ ( 2 , 2 ) , of degree 2 2 2 , restricts to two linear characters, with values 1 2 ( − 1 ± − 3 ) \frac{1}{2}(-1\pm\sqrt{-3}) 2 1 ( − 1 ± − 3 ) , that is ω \omega ω and ω ‾ \overline{\omega} ω with ω = e 2 π i / 3 \omega=e^{2\pi i/3} ω = e 2 π i /3 , on the two classes of 3 3 3 -cycles. The pairs { ( 4 ) , ( 1 4 ) } \{(4),(1^4)\} {( 4 ) , ( 1 4 )} and { ( 3 , 1 ) , ( 2 , 1 2 ) } \{(3,1),(2,1^2)\} {( 3 , 1 ) , ( 2 , 1 2 )} give the trivial character and the character of degree 3 3 3 . The degrees 1 , 1 , 1 , 3 1,1,1,3 1 , 1 , 1 , 3 satisfy 1 + 1 + 1 + 9 = 12 = ∣ Alt 4 ∣ 1+1+1+9=12=|\operatorname{Alt}_4| 1 + 1 + 1 + 9 = 12 = ∣ Alt 4 ∣ .
Example 11.38 (Alt 5 \operatorname{Alt}_5 Alt 5 , Alt 7 \operatorname{Alt}_7 Alt 7 , Alt 9 \operatorname{Alt}_9 Alt 9 ). For n = 5 n=5 n = 5 , λ = ( 3 , 1 2 ) \lambda=(3,1^2) λ = ( 3 , 1 2 ) has h = ( 5 ) h=(5) h = ( 5 ) , ϵ = 1 \epsilon=1 ϵ = 1 and q = 5 q=5 q = 5 , so the two characters of degree 3 3 3 take the values 1 2 ( 1 ± 5 ) \frac{1}{2}(1\pm\sqrt{5}) 2 1 ( 1 ± 5 ) on the two classes of 5 5 5 -cycles (Table 11.5 ). For n = 7 n=7 n = 7 , λ = ( 4 , 1 3 ) \lambda=(4,1^3) λ = ( 4 , 1 3 ) has h = ( 7 ) h=(7) h = ( 7 ) and ϵ = − 1 \epsilon=-1 ϵ = − 1 , giving the non-real values 1 2 ( − 1 ± − 7 ) \frac{1}{2}(-1\pm\sqrt{-7}) 2 1 ( − 1 ± − 7 ) . For n = 9 n=9 n = 9 , λ = ( 5 , 1 4 ) \lambda=(5,1^4) λ = ( 5 , 1 4 ) has h = ( 9 ) h=(9) h = ( 9 ) , ϵ = 1 \epsilon=1 ϵ = 1 and q = 9 q=9 q = 9 , so 1 2 ( 1 ± 3 ) ∈ { 2 , − 1 } \frac{1}{2}(1\pm3)\in\{2,-1\} 2 1 ( 1 ± 3 ) ∈ { 2 , − 1 } : the two constituents are rational-valued although χ λ \chi^\lambda χ λ splits.
Table 11.5. The character table of Alt 5 \operatorname{Alt}_5 Alt 5 , obtained from Table 11.4 (PDF p. 141) by Theorems 11.33 and 11.35 . A row ( λ ) ∼ ( λ ′ ) (\lambda)\sim(\lambda') ( λ ) ∼ ( λ ′ ) is the common restriction of χ λ \chi^\lambda χ λ and χ λ ′ \chi^{\lambda'} χ λ ′ ; 5 ± 5_\pm 5 ± are the two classes of 5 5 5 -cycles.
Digital laws of Sym n \operatorname{Sym}_n Sym n and $\operatorname{Alt}_n`
The Lehmer laws
The Lehmer label of Definition 11.1 is a bijection Sym n → A n ! \operatorname{Sym}_n \to\mathcal A_{n!} Sym n → A n ! (Theorem 11.2 ), so transporting composition along it gives a digital law of Sym n \operatorname{Sym}_n Sym n . This section makes that law explicit, derives from it a digital law of Alt n \operatorname{Alt}_n Alt n on A n ! / 2 \mathcal A_{n!}/2 A n ! /2 , and shows that the Coxeter generators, the sign, the cosets of Alt n \operatorname{Alt}_n Alt n and conjugation by a transposition all act directly on the digits.
Throughout, n ≥ 2 n \ge2 n ≥ 2 , composition is right to left as in (12) , and Π i = ∏ j < i ( n − j ) = n ! / ( n − i ) ! \Pi_i=\prod_{j<i}(n-j)=n!/(n-i)! Π i = ∏ j < i ( n − j ) = n ! / ( n − i )! for 0 ≤ i ≤ n 0 \le i \le n 0 ≤ i ≤ n . For m ∈ A n ! m \in\mathcal A_{n!} m ∈ A n ! let d 0 ( m ) , … , d n − 1 ( m ) d_0(m),\ldots,d_{n-1}(m) d 0 ( m ) , … , d n − 1 ( m ) be its mixed-radix digits for the radices n , n − 1 , … , 1 n,n-1,\ldots,1 n , n − 1 , … , 1 , so that m = ∑ i d i ( m ) Π i m=\sum_i d_i(m)\Pi_i m = ∑ i d i ( m ) Π i with 0 ≤ d i ( m ) < n − i 0 \le d_i(m)<n-i 0 ≤ d i ( m ) < n − i ; in particular d n − 1 ( m ) = 0 d_{n-1}(m)=0 d n − 1 ( m ) = 0 . Write s ( m ) = ∑ i d i ( m ) s(m)=\sum_i d_i(m) s ( m ) = ∑ i d i ( m ) for the digit sum. By Theorem 11.2 , d i ( λ ( R ) ) = a i ( R ) d_i(\lambda(R))=a_i(R) d i ( λ ( R )) = a i ( R ) . For a statement P P P , [ P ] [P] [ P ] is 1 if P P P holds and 0 otherwise, and w 0 w_0 w 0 denotes the reversal i ↦ n − 1 − i i\mapsto n-1-i i ↦ n − 1 − i .
Definition 11.40 (Lehmer law). For m , k ∈ A n ! m,k\in\mathcal A_{n!} m , k ∈ A n ! put m ∗ L k = λ ( λ − 1 ( m ) ∘ λ − 1 ( k ) ) m*_{L}k=\lambda\left(\lambda^{-1}(m)\circ\lambda^{-1}(k)\right) m ∗ L k = λ ( λ − 1 ( m ) ∘ λ − 1 ( k ) ) .
Theorem 11.41 (Lehmer law of Sym n \operatorname{Sym}_n Sym n ).
( A n ! , ∗ L ) (\mathcal A_{n!},*_{L}) ( A n ! , ∗ L ) is a group with identity 0, and λ \lambda λ is an isomorphism from Sym n \operatorname{Sym}_n Sym n onto ( A n ! , ∗ L ) (\mathcal A_{n!},*_{L}) ( A n ! , ∗ L ) .
Encoding. For a permutation code R R R ,
λ ( R ) = ∑ 0 ≤ i < j < n [ d n ( j , R ) < d n ( i , R ) ] Π i . \lambda(R)=\sum_{0\le i<j<n}[d_n(j,R)<d_n(i,R)]\Pi_i. λ ( R ) = 0 ≤ i < j < n ∑ [ d n ( j , R ) < d n ( i , R )] Π i . Decoding. λ − 1 ( m ) \lambda^{-1}(m) λ − 1 ( m ) is the permutation R R R with R ( i ) R(i) R ( i ) equal to the element of rank d i ( m ) d_i(m) d i ( m ) , counting from 0, in A n ∖ { R ( 0 ) , … , R ( i − 1 ) } \mathcal A_n\setminus\{R(0),\ldots,R(i-1)\} A n ∖ { R ( 0 ) , … , R ( i − 1 )} .
Inverse. The inverse of m m m in ( A n ! , ∗ L ) (\mathcal A_{n!},*_{L}) ( A n ! , ∗ L ) is λ ( ( λ − 1 ( m ) ) − 1 ) \lambda((\lambda^{-1}(m))^{-1}) λ (( λ − 1 ( m ) ) − 1 ) , with the inverse permutation given by (11) .
λ ( s i ) = Π i \lambda(s_i)=\Pi_i λ ( s i ) = Π i for 0 ≤ i ≤ n − 2 0\le i\le n-2 0 ≤ i ≤ n − 2 , λ ( w 0 ) = n ! − 1 \lambda(w_0)=n!-1 λ ( w 0 ) = n ! − 1 , and sgn λ − 1 ( m ) = ( − 1 ) s ( m ) \operatorname{sgn}\lambda^{-1}(m)=(-1)^{s(m)} sgn λ − 1 ( m ) = ( − 1 ) s ( m ) .
Proof. Part (1) is transport of structure along the bijection λ \lambda λ , and λ ( id ) = 0 \lambda(\operatorname{id})=0 λ ( id ) = 0 because the identity has no inversions. Part (2) restates Definition 11.1 , since R ( i ) = d n ( i , R ) R(i)=d_n(i,R) R ( i ) = d n ( i , R ) . Part (3) is the decoding in the proof of Theorem 11.2 , and (4) is again transport of structure. For (5): the word of s i s_i s i has a single inversion, at position i i i , so a i ( s i ) = 1 a_i(s_i)=1 a i ( s i ) = 1 and all other digits vanish. Every digit of w 0 w_0 w 0 is maximal, a i ( w 0 ) = n − 1 − i a_i(w_0)=n-1-i a i ( w 0 ) = n − 1 − i , and ( n − 1 − i ) Π i = Π i + 1 − Π i (n-1-i)\Pi_i=\Pi_{i+1}-\Pi_i ( n − 1 − i ) Π i = Π i + 1 − Π i telescopes to Π n − Π 0 = n ! − 1 \Pi_n-\Pi_0=n!-1 Π n − Π 0 = n ! − 1 . The sign formula is Theorem 11.2 . □
Theorem 11.42 (Coxeter generators act on the digits). Let m ∈ A n ! m\in\mathcal A_{n!} m ∈ A n ! , d j = d j ( m ) d_j=d_j(m) d j = d j ( m ) and 0 ≤ i ≤ n − 2 0\le i\le n-2 0 ≤ i ≤ n − 2 .
Right multiplication by λ ( s i ) = Π i \lambda(s_i)=\Pi_i λ ( s i ) = Π i changes only the digits in positions i i i and i + 1 i+1 i + 1 :
m ∗ L Π i = { m + ( d i + 1 + 1 − d i ) Π i + ( d i − d i + 1 ) Π i + 1 , d i ≤ d i + 1 , m + ( d i + 1 − d i ) Π i + ( d i − 1 − d i + 1 ) Π i + 1 , d i > d i + 1 . m*_{L}\Pi_i=
\begin{cases}
m+(d_{i+1}+1-d_i)\Pi_i+(d_i-d_{i+1})\Pi_{i+1}, & d_i\le d_{i+1},\\
m+(d_{i+1}-d_i)\Pi_i+(d_i-1-d_{i+1})\Pi_{i+1}, & d_i>d_{i+1}.
\end{cases} m ∗ L Π i = { m + ( d i + 1 + 1 − d i ) Π i + ( d i − d i + 1 ) Π i + 1 , m + ( d i + 1 − d i ) Π i + ( d i − 1 − d i + 1 ) Π i + 1 , d i ≤ d i + 1 , d i > d i + 1 .
that is, ( d i , d i + 1 ) (d_i,d_{i+1}) ( d i , d i + 1 ) is replaced by ( d i + 1 + 1 , d i ) (d_{i+1}+1,d_i) ( d i + 1 + 1 , d i ) or by ( d i + 1 , d i − 1 ) (d_{i+1},d_i-1) ( d i + 1 , d i − 1 ) .
Let R = λ − 1 ( m ) R=\lambda^{-1}(m) R = λ − 1 ( m ) , p = R − 1 ( i ) p=R^{-1}(i) p = R − 1 ( i ) and q = R − 1 ( i + 1 ) q=R^{-1}(i+1) q = R − 1 ( i + 1 ) . Left multiplication changes a single digit by ± 1 \pm1 ± 1 :
Π i ∗ L m = { m + Π p , p < q , m − Π q , p > q . \Pi_i*_{L}m=
\begin{cases}
m+\Pi_p, & p<q,\\
m-\Pi_q, & p>q.
\end{cases} Π i ∗ L m = { m + Π p , m − Π q , p < q , p > q .
In particular s ( m ∗ L Π i ) = s ( m ) ± 1 s(m*_{L}\Pi_i)=s(m)\pm1 s ( m ∗ L Π i ) = s ( m ) ± 1 and s ( Π i ∗ L m ) = s ( m ) ± 1 s(\Pi_i*_{L}m)=s(m)\pm1 s ( Π i ∗ L m ) = s ( m ) ± 1 .
Proof. (1) The word of R ∘ s i R\circ s_i R ∘ s i is that of R R R with the entries in positions i i i and i + 1 i+1 i + 1 exchanged. For j ∉ { i , i + 1 } j\notin\{i,i+1\} j ∈ / { i , i + 1 } the entry in position j j j and the set of entries after it are unchanged, so a j a_j a j is unchanged. The new entry in position i i i is R ( i + 1 ) R(i+1) R ( i + 1 ) , and it exceeds a i + 1 + [ R ( i ) < R ( i + 1 ) ] a_{i+1}+[R(i)<R(i+1)] a i + 1 + [ R ( i ) < R ( i + 1 )] later entries; the new entry in position i + 1 i+1 i + 1 is R ( i ) R(i) R ( i ) , which exceeds a i − [ R ( i + 1 ) < R ( i ) ] a_i-[R(i+1)<R(i)] a i − [ R ( i + 1 ) < R ( i )] later entries. Finally R ( i ) < R ( i + 1 ) R(i)<R(i+1) R ( i ) < R ( i + 1 ) if and only if a i ≤ a i + 1 a_i\le a_{i+1} a i ≤ a i + 1 : if R ( i ) < R ( i + 1 ) R(i)<R(i+1) R ( i ) < R ( i + 1 ) , every later entry below R ( i ) R(i) R ( i ) is also below R ( i + 1 ) R(i+1) R ( i + 1 ) , so a i ≤ a i + 1 a_i\le a_{i+1} a i ≤ a i + 1 ; if R ( i ) > R ( i + 1 ) R(i)>R(i+1) R ( i ) > R ( i + 1 ) , then a i ≥ 1 + a i + 1 a_i\ge1+a_{i+1} a i ≥ 1 + a i + 1 . (2) The word of s i ∘ R s_i\circ R s i ∘ R is that of R R R with the values i i i and i + 1 i+1 i + 1 exchanged. Since these values are adjacent, the only comparison that changes is the one between positions p p p and q q q . If p < q p<q p < q the entry in position p p p gains one smaller later entry, so a p a_p a p increases by 1; if p > q p>q p > q the entry in position q q q loses one, so a q a_q a q decreases by 1.
Proposition 11.43 (Sign and cosets). Let N = n ! / 2 = Π n − 2 N=n!/2=\Pi_{n-2} N = n ! /2 = Π n − 2 .
λ ( Alt n ) = { m ∈ A n ! : s ( m ) even } \lambda(\operatorname{Alt}_n)=\{m\in A_n!:s(m)\ \text{even}\} λ ( Alt n ) = { m ∈ A n ! : s ( m ) even } .
m ∗ L N = m + N m*_{L}N=m+N m ∗ L N = m + N if m < N m<N m < N , and m ∗ L N = m − N m*_{L}N=m-N m ∗ L N = m − N if m ≥ N m\ge N m ≥ N . Thus right multiplication by the label of s n − 2 s_{n-2} s n − 2 exchanges the two halves of A n ! A_n! A n ! .
For m < N m<N m < N , exactly one of m m m and m + N m+N m + N labels an even permutation, namely m + ε ( m ) N m+\varepsilon(m)N m + ε ( m ) N , where ε ( m ) = s ( m ) m o d 2 \varepsilon(m)=s(m)\bmod2 ε ( m ) = s ( m ) mod 2 .
Proof. (1) is Theorem 11.41 (5). For (2), apply Theorem 11.42 (1) with i = n − 2 i=n-2 i = n − 2 : here d n − 1 = 0 d_{n-1}=0 d n − 1 = 0 , and d n − 2 ∈ { 0 , 1 } d_{n-2}\in\{0,1\} d n − 2 ∈ { 0 , 1 } equals 1 exactly when m ≥ N m\ge N m ≥ N . For (3), if m < N m<N m < N then m + N m+N m + N differs from m m m only in the digit d n − 2 d_{n-2} d n − 2 , which changes from 0 to 1, so s ( m + N ) = s ( m ) + 1 s(m+N)=s(m)+1 s ( m + N ) = s ( m ) + 1 .
Lemma 11.44 (Deleting a parity digit). Let U U U be a finite digit set and let θ : U × A 2 → Sym n \theta:U\times A_2\to\operatorname{Sym}_n θ : U × A 2 → Sym n be a bijection such that sgn θ ( u , e ) = ( − 1 ) e + q ( u ) \operatorname{sgn}\theta(u,e)=(-1)^{e+q(u)} sgn θ ( u , e ) = ( − 1 ) e + q ( u ) for some q : U → A 2 q:U\to A_2 q : U → A 2 . Then u ↦ θ ( u , q ( u ) ) u\mapsto\theta(u,q(u)) u ↦ θ ( u , q ( u )) is a bijection from U U U onto Alt n \operatorname{Alt}_n Alt n . Transporting composition along it gives a digital law of Alt n \operatorname{Alt}_n Alt n , obtained from the law of Sym n \operatorname{Sym}_n Sym n by deleting the radix-2 parity digit and restoring it as e = q ( u ) e=q(u) e = q ( u ) before multiplying.
Proof. For each u ∈ U u\in U u ∈ U exactly one of θ ( u , 0 ) \theta(u,0) θ ( u , 0 ) and θ ( u , 1 ) \theta(u,1) θ ( u , 1 ) is even, namely θ ( u , q ( u ) ) \theta(u,q(u)) θ ( u , q ( u )) . So the map picks exactly one element from each two-element fibre over U U U , and its image is Alt n \operatorname{Alt}_n Alt n . The law is transport of structure.
Theorem 11.45 (Lehmer law of Alt n \operatorname{Alt}_n Alt n ). Let N = n ! / 2 N=n!/2 N = n ! /2 and, for R ∈ Alt n R\in\operatorname{Alt}_n R ∈ Alt n ,
μ ( R ) = λ ( R ) m o d N = ∑ i < n − 2 a i ( R ) Π i . \mu(R)=\lambda(R)\bmod N=\sum_{i<n-2}a_i(R)\Pi_i. μ ( R ) = λ ( R ) mod N = i < n − 2 ∑ a i ( R ) Π i .
Then μ \mu μ is a bijection Alt n → A N \operatorname{Alt}_n\to A_N Alt n → A N with inverse μ − 1 ( m ) = λ − 1 ( m + ε ( m ) N ) \mu^{-1}(m)=\lambda^{-1}(m+\varepsilon(m)N) μ − 1 ( m ) = λ − 1 ( m + ε ( m ) N ) . The transported product m ∗ L k = μ ( μ − 1 ( m ) ∘ μ − 1 ( k ) ) m*_{L}k=\mu(\mu^{-1}(m)\circ\mu^{-1}(k)) m ∗ L k = μ ( μ − 1 ( m ) ∘ μ − 1 ( k )) is a digital law of Alt n \operatorname{Alt}_n Alt n with identity 0, and
m ∗ L k = ( ( m + ε ( m ) N ) ∗ ( k + ε ( k ) N ) ) m o d N . m*_{L}k=((m+\varepsilon(m)N)*(k+\varepsilon(k)N))\bmod N. m ∗ L k = (( m + ε ( m ) N ) ∗ ( k + ε ( k ) N )) mod N .
Proof. Since a n − 1 ( R ) = 0 a_{n-1}(R)=0 a n − 1 ( R ) = 0 and ∑ i < n − 2 a i ( R ) Π i < Π n − 2 = N \sum_{i<n-2}a_i(R)\Pi_i<\Pi_{n-2}=N ∑ i < n − 2 a i ( R ) Π i < Π n − 2 = N , we have
λ ( R ) = ∑ i < n − 2 a i ( R ) Π i + a n − 2 ( R ) N , \lambda(R)=\sum_{i<n-2}a_i(R)\Pi_i+a_{n-2}(R)N, λ ( R ) = i < n − 2 ∑ a i ( R ) Π i + a n − 2 ( R ) N ,
which gives the two expressions for μ \mu μ . By Proposition 11.43(3), each fibre { m , m + N } \{m,m+N\} { m , m + N } of reduction modulo N N N contains exactly one even label, namely m + ε ( m ) N m+\varepsilon(m)N m + ε ( m ) N ; hence μ \mu μ is a bijection with the stated inverse. The label λ ( μ − 1 ( m ) ∘ μ − 1 ( k ) ) \lambda\left(\mu^{-1}(m)\circ\mu^{-1}(k)\right) λ ( μ − 1 ( m ) ∘ μ − 1 ( k ) ) equals ( m + ε ( m ) N ) ∗ ( k + ε ( k ) N ) (m+\varepsilon(m)N)\mathbin{*}(k+\varepsilon(k)N) ( m + ε ( m ) N ) ∗ ( k + ε ( k ) N ) , and reducing it modulo N N N gives μ \mu μ of the product.
Proposition 11.46 (Conjugation by a transposition). Let n ≥ 3 n\ge3 n ≥ 3 , N = n ! / 2 N=n!/2 N = n ! /2 and τ = s n − 2 \tau=s_{n-2} τ = s n − 2 . For m ∈ A n m\in A_n m ∈ A n let R = μ − 1 ( m ) R=\mu^{-1}(m) R = μ − 1 ( m ) , p = R − 1 ( n − 2 ) p=R^{-1}(n-2) p = R − 1 ( n − 2 ) and q = R − 1 ( n − 1 ) q=R^{-1}(n-1) q = R − 1 ( n − 1 ) . The automorphism of ( A N , ⋆ L ) (A_N,\star_L) ( A N , ⋆ L ) induced by conjugation by τ \tau τ is
m ⟼ μ ( τ R τ ) = { ( m + Π p ) m o d N , p < q , ( m − Π q ) m o d N , p > q . m\longmapsto\mu(\tau R\tau)=
\begin{cases}
(m+\Pi_p)\bmod N, & p<q,\\
(m-\Pi_q)\bmod N, & p>q.
\end{cases} m ⟼ μ ( τ R τ ) = { ( m + Π p ) mod N , ( m − Π q ) mod N , p < q , p > q .
Proof. By Theorem 11.42 (2) with i = n − 2 i=n-2 i = n − 2 , λ ( τ R ) = λ ( R ) + Π p \lambda(\tau R)=\lambda(R)+\Pi_p λ ( τ R ) = λ ( R ) + Π p or λ ( R ) − Π q \lambda(R)-\Pi_q λ ( R ) − Π q . By Proposition 11.43(2), λ ( τ R τ ) = λ ( τ R ) ± N \lambda(\tau R\tau)=\lambda(\tau R)\pm N λ ( τ R τ ) = λ ( τ R ) ± N . Since λ ( R ) ≡ m ( m o d N ) \lambda(R)\equiv m\pmod N λ ( R ) ≡ m ( mod N ) , reducing modulo N N N gives the formula.
Proposition 11.47 (Chain property). Let 0 ≤ i < n 0\le i<n 0 ≤ i < n and R , R ′ ∈ S y m n R,R'\in\mathrm{Sym}_n R , R ′ ∈ Sym n . The digit prefixes ( d 0 , … , d i ) (d_0,\ldots,d_i) ( d 0 , … , d i ) of λ ( R ) \lambda(R) λ ( R ) and λ ( R ′ ) \lambda(R') λ ( R ′ ) agree if and only if R ( j ) = R ′ ( j ) R(j)=R'(j) R ( j ) = R ′ ( j ) for all j ≤ i j\le i j ≤ i , that is, if and only if R ′ ∈ R S y m n ( i ) R'\in R\mathrm{Sym}_n^{(i)} R ′ ∈ R Sym n ( i ) , where S y m n ( i ) \mathrm{Sym}_n^{(i)} Sym n ( i ) is the pointwise stabiliser of { 0 , … , i } \{0,\ldots,i\} { 0 , … , i } . Thus λ \lambda λ is the stabiliser-chain labelling for the base ( 0 , 1 , … , n − 1 ) (0,1,\ldots,n-1) ( 0 , 1 , … , n − 1 ) , in which digit d i d_i d i selects the coset of S y m n ( i ) \mathrm{Sym}_n^{(i)} Sym n ( i ) inside R S y m n ( i − 1 ) R\mathrm{Sym}_n^{(i-1)} R Sym n ( i − 1 ) by the rank of R ( i ) R(i) R ( i ) among the letters not yet used; compare Theorem 8.34.
Proof. For j ≤ i j\le i j ≤ i , the digit a j ( R ) a_j(R) a j ( R ) depends only on R ( j ) R(j) R ( j ) and on the set { R ( j ′ ) : j ′ > j } = A n ∖ { R ( 0 ) , … , R ( j ) } \{R(j'):j'>j\}=A_n\setminus\{R(0),\ldots,R(j)\} { R ( j ′ ) : j ′ > j } = A n ∖ { R ( 0 ) , … , R ( j )} , hence only on R ( 0 ) , … , R ( i ) R(0),\ldots,R(i) R ( 0 ) , … , R ( i ) . Conversely, the decoding of Theorem 11.41 (3) recovers R ( 0 ) , … , R ( i ) R(0),\ldots,R(i) R ( 0 ) , … , R ( i ) from d 0 , … , d i d_0,\ldots,d_i d 0 , … , d i . Finally R ′ = R h R'=Rh R ′ = R h with h h h fixing 0 , … , i 0,\ldots,i 0 , … , i exactly when R ′ ( j ) = R ( j ) R'(j)=R(j) R ′ ( j ) = R ( j ) for j ≤ i j\le i j ≤ i .
Remark 11.48 (Cost and comparison). Computing m ⋆ L k m\star_L k m ⋆ L k takes two decodings, one composition ((12) ) and one encoding. Each decoding and encoding costs O ( n 2 ) O(n^2) O ( n 2 ) elementary operations as written, or O ( n log n ) O(n\log n) O ( n log n ) with a binary indexed tree for the rank queries. The law ⋆ L \star_L ⋆ L adds two digit sums and one reduction modulo N N N . The generator rules of Theorem 11.42 cost O ( 1 ) O(1) O ( 1 ) digit operations on the right and O ( n ) O(n) O ( n ) on the left, since p p p and q q q must be located. § 19.12.2 constructs a different law of A l t n \mathrm{Alt}_n Alt n (Theorem 11.53 ). Both laws realise A l t n \mathrm{Alt}_n Alt n and are therefore isomorphic, but they label the elements differently. The Lehmer law is the one in which the sign, the coset of A l t n \mathrm{Alt}_n Alt n and the Coxeter generators can be read off, or applied, digit by digit.
Example 11.49 (Sym3 _3 3 and Alt3 _3 3 ). For n = 3 n=3 n = 3 , Π 0 , Π 1 , Π 2 = 1 , 3 , 6 \Pi_0,\Pi_1,\Pi_2=1,3,6 Π 0 , Π 1 , Π 2 = 1 , 3 , 6 and m = d 0 + 3 d 1 m=d_0+3d_1 m = d 0 + 3 d 1 . The labels 0 , … , 5 0,\ldots,5 0 , … , 5 are the words 012, 102, 201, 021, 120, 210, so λ ( s 0 ) = 1 \lambda(s_0)=1 λ ( s 0 ) = 1 , λ ( s 1 ) = 3 \lambda(s_1)=3 λ ( s 1 ) = 3 and λ ( w 0 ) = 5 \lambda(w_0)=5 λ ( w 0 ) = 5 . Then 1 ∗ 3 = 4 1*3=4 1 ∗ 3 = 4 while 3 ∗ 1 = 2 3*1=2 3 ∗ 1 = 2 , and 2 ∗ 4 = 0 2*4=0 2 ∗ 4 = 0 , so the two 3-cycles 201 and 120 are mutually inverse; Table 11.6 gives the complete law. The even labels are 0, 2, 4, those with even digit sum, and μ \mu μ sends them to 0, 2, 1. For instance μ ( 120 ) = 4 m o d 3 = 1 \mu(120)=4\bmod3=1 μ ( 120 ) = 4 mod 3 = 1 , and conversely ε ( 1 ) = 1 \varepsilon(1)=1 ε ( 1 ) = 1 gives μ − 1 ( 1 ) = λ − 1 ( 1 + 3 ) \mu^{-1}(1)=\lambda^{-1}(1+3) μ − 1 ( 1 ) = λ − 1 ( 1 + 3 ) .
Table 11.6. The Lehmer law of Sym 3 \operatorname{Sym}_3 Sym 3 . The entry in row m m m and column k k k is m ∗ L k m *_{\mathrm{L}} k m ∗ L k ; the second column is the word R ( 0 ) R ( 1 ) R ( 2 ) R(0)R(1)R(2) R ( 0 ) R ( 1 ) R ( 2 ) of R = λ − 1 ( m ) R = \lambda^{-1}(m) R = λ − 1 ( m ) .
Tables 11.8, 11.9, 11.10 and 11.11 give the Lehmer laws of Alt4 _4 4 , Sym4 _4 4 and Alt5 _5 5 in full. They are written by scripts/sn_cayley.py, which first checks that label 0 is the identity, that each table is a Latin square, that the law is associative on all triples, and that every entry agrees with multiply.py and with direct composition of the decoded permutations.
The rotation laws
Throughout, n ≥ 3 n \ge3 n ≥ 3 , permutations act on A n \mathcal{A}_n A n , and ( σ ∘ τ ) ( i ) = σ ( τ ( i ) ) (\sigma\circ\tau)(i) = \sigma(\tau(i)) ( σ ∘ τ ) ( i ) = σ ( τ ( i )) , the right-to-left composition of (12) .
Definition 11.50. For 2 ≤ k ≤ n 2 \le k \le n 2 ≤ k ≤ n the prefix rotation ρ k \rho_k ρ k is the k k k -cycle ( 0 1 ⋯ k − 1 ) (0\ 1\ \cdots\ k-1) ( 0 1 ⋯ k − 1 ) of A n \mathcal{A}_n A n . Its powers are
ρ k ℓ ( i ) = i + B ( ⌊ i / k ⌋ ) ( d k ( 0 , i + ℓ ) − i ) , i ∈ A n . (84) \rho_k^\ell(i)=i+\mathcal{B}\left(\left\lfloor i/k\right\rfloor\right)\left(d_k(0,i+\ell)-i\right),\qquad i\in\mathcal{A}_n. \tag*{(84)} ρ k ℓ ( i ) = i + B ( ⌊ i / k ⌋ ) ( d k ( 0 , i + ℓ ) − i ) , i ∈ A n . (84)
Since B ( ⌊ i / k ⌋ ) = 1 \mathcal{B}(\lfloor i/k\rfloor)=1 B (⌊ i / k ⌋) = 1 exactly when i < k i<k i < k , the letters i ≥ k i\ge k i ≥ k are fixed and the first k k k letters are shifted cyclically by ℓ \ell ℓ .
Lemma 11.51 (Rotation normal form). Every σ ∈ Sym n \sigma\in\operatorname{Sym}_n σ ∈ Sym n has a unique expression
σ = ρ n ℓ n − 1 ∘ ⋯ ∘ ρ 3 ℓ 2 ∘ ρ 2 ℓ 1 , ℓ h ∈ A h + 1 . \sigma=\rho_n^{\ell_{n-1}}\circ\cdots\circ\rho_3^{\ell_2}\circ\rho_2^{\ell_1},\qquad\ell_h\in\mathcal{A}_{h+1}. σ = ρ n ℓ n − 1 ∘ ⋯ ∘ ρ 3 ℓ 2 ∘ ρ 2 ℓ 1 , ℓ h ∈ A h + 1 .
Table 11.8. The Lehmer law of Alt 4 \operatorname{Alt}_4 Alt 4 (Theorem 11.45). The entry in row m m m and column k k k is m ∗ k m * k m ∗ k ; the second column is the word R ( 0 ) ⋯ R ( 3 ) R(0)\cdots R(3) R ( 0 ) ⋯ R ( 3 ) of R = μ − 1 ( m ) R=\mu^{-1}(m) R = μ − 1 ( m ) .
Table 11.9. The Lehmer law of Sym 4 \operatorname{Sym}_4 Sym 4 (Theorem 11.41 ). The entry in row m m m and column k k k is m ∗ L k m \mathbin{*_{\mathrm L}} k m ∗ L k ; the second column is the word R ( 0 ) ⋯ R ( 3 ) R(0)\cdots R(3) R ( 0 ) ⋯ R ( 3 ) of R = λ − 1 ( m ) R=\lambda^{-1}(m) R = λ − 1 ( m ) .
The exponents are read off from the top down: with π n = σ \pi_n=\sigma π n = σ , for k = n , n − 1 , … , 2 k=n,n-1,\ldots,2 k = n , n − 1 , … , 2 ,
ℓ k − 1 = d k ( 0 , π k ( k − 1 ) + 1 ) , π k − 1 ( i ) = d k ( 0 , π k ( i ) − ℓ k − 1 ) ( i < k − 1 ) . (85) \ell_{k-1}=d_k(0,\pi_k(k-1)+1),\qquad\pi_{k-1}(i)=d_k(0,\pi_k(i)-\ell_{k-1})\quad(i<k-1). \tag*{(85)} ℓ k − 1 = d k ( 0 , π k ( k − 1 ) + 1 ) , π k − 1 ( i ) = d k ( 0 , π k ( i ) − ℓ k − 1 ) ( i < k − 1 ) . (85)
Moreover sgn σ = ( − 1 ) ℓ 1 + ℓ 3 + ℓ 5 + ⋯ \operatorname{sgn}\sigma=(-1)^{\ell_1+\ell_3+\ell_5+\cdots} sgn σ = ( − 1 ) ℓ 1 + ℓ 3 + ℓ 5 + ⋯ .
Proof. Since ρ k ℓ ( k − 1 ) = d k ( 0 , ℓ − 1 ) \rho_k^\ell(k-1)=d_k(0,\ell-1) ρ k ℓ ( k − 1 ) = d k ( 0 , ℓ − 1 ) , the k k k powers of ρ k \rho_k ρ k send k − 1 k-1 k − 1 to distinct letters of A k A_k A k . Let π k \pi_k π k be a permutation of A k A_k A k (fixing the letters ≥ k \ge k ≥ k ). Exactly one ℓ ∈ A k \ell\in A_k ℓ ∈ A k satisfies ρ k − ℓ ( π k ( k − 1 ) ) = k − 1 \rho_k^{-\ell}(\pi_k(k-1))=k-1 ρ k − ℓ ( π k ( k − 1 )) = k − 1 , namely ℓ = d k ( 0 , π k ( k − 1 ) + 1 ) \ell=d_k(0,\pi_k(k-1)+1) ℓ = d k ( 0 , π k ( k − 1 ) + 1 ) , and then π k − 1 = ρ k − ℓ ∘ π k \pi_{k-1}=\rho_k^{-\ell}\circ\pi_k π k − 1 = ρ k − ℓ ∘ π k is a permutation of A k − 1 A_{k-1} A k − 1 . Its values are those in (11.4), because ρ k − ℓ ( j ) = d k ( 0 , j − ℓ ) \rho_k^{-\ell}(j)=d_k(0,j-\ell) ρ k − ℓ ( j ) = d k ( 0 , j − ℓ ) for j < k j<k j < k . Induction from k = n k=n k = n down to k = 2 k=2 k = 2 , where π 1 \pi_1 π 1 is the identity, gives existence and uniqueness; as a check, there are 2 ⋅ 3 ⋯ n = n ! 2\cdot3\cdots n=n! 2 ⋅ 3 ⋯ n = n ! exponent tuples. A k k k -cycle has sign ( − 1 ) k − 1 (-1)^{k-1} ( − 1 ) k − 1 , so sgn σ = ( − 1 ) ∑ h h ℓ h \operatorname{sgn}\sigma=(-1)^{\sum_h h\ell_h} sgn σ = ( − 1 ) ∑ h h ℓ h , and only the odd h h h contribute.
Theorem 11.52 (Rotation law of Sym n \operatorname{Sym}_n Sym n ). For σ ∈ Sym n \sigma\in\operatorname{Sym}_n σ ∈ Sym n let ℓ h ( σ ) \ell_h(\sigma) ℓ h ( σ ) be the exponents of Lemma 11.51, and put
r ( σ ) = ∑ h = 1 n − 1 ℓ h ( σ ) h ! ∈ A n ! . r(\sigma)=\sum_{h=1}^{n-1}\ell_h(\sigma)\,h!\in A_n!. r ( σ ) = h = 1 ∑ n − 1 ℓ h ( σ ) h ! ∈ A n ! . Then r r r is a bijection, and
x ∗ ρ y = r ( r − 1 ( x ) ∘ r − 1 ( y ) ) x*_{\rho}y=r\left(r^{-1}(x)\circ r^{-1}(y)\right) x ∗ ρ y = r ( r − 1 ( x ) ∘ r − 1 ( y ) ) makes A n ! A_n! A n ! a group isomorphic to Sym n \operatorname{Sym}_n Sym n , with identity 0 0 0 . The digit of place value h ! h! h ! is ℓ h \ell_h ℓ h . In particular r ( ρ 2 ) = 1 r(\rho_2)=1 r ( ρ 2 ) = 1 and
x ∗ ρ 1 = { x + 1 , x even , x − 1 , x odd , x*_{\rho}1=
\begin{cases}
x+1,&x\ \text{even},\\
x-1,&x\ \text{odd},
\end{cases} x ∗ ρ 1 = { x + 1 , x − 1 , x even , x odd , so right multiplication by the transposition ρ 2 \rho_2 ρ 2 toggles the radix-2 digit.
Proof. Lemma 11.51 identifies Sym n \operatorname{Sym}_n Sym n bijectively with A 2 × A 3 × ⋯ × A n \mathcal{A}_2 \times\mathcal{A}_3 \times\cdots\times\mathcal{A}_n A 2 × A 3 × ⋯ × A n , and mixed-radix evaluation with place values 1 ! , 2 ! , … , ( n − 1 ) ! 1!,2!,\ldots,(n-1)! 1 ! , 2 ! , … , ( n − 1 )! identifies this product with A n ! \mathcal{A}_{n!} A n ! (Lemma 1.6 ). The law is transport of structure. Since ρ 2 \rho_2 ρ 2 is the rightmost factor of the normal form, right multiplication by ρ 2 \rho_2 ρ 2 replaces ℓ 1 \ell_1 ℓ 1 by ℓ 1 + 1 ( m o d 2 ) \ell_1+1\pmod{2} ℓ 1 + 1 ( mod 2 ) and leaves the other exponents unchanged. □
Theorem 11.53 (Rotation law of Alt n \operatorname{Alt}_n Alt n ). Let n ≥ 3 n\geq3 n ≥ 3 . Write a ∈ A n ! / 2 a\in\mathcal{A}_{n!/2} a ∈ A n ! /2 in the mixed radix ( 3 , 4 , … , n ) (3,4,\ldots,n) ( 3 , 4 , … , n ) and put
ℓ h = D h − 2 ( a ) ( h = 2 , … , n − 1 ) , ℓ 1 = d 2 ( 0 , ℓ 3 + ℓ 5 + ⋯ ) . (86) \ell_h=D_{h-2}(a)\quad(h=2,\ldots,n-1),\qquad\ell_1=d_2(0,\ell_3+\ell_5+\cdots). \tag*{(86)} ℓ h = D h − 2 ( a ) ( h = 2 , … , n − 1 ) , ℓ 1 = d 2 ( 0 , ℓ 3 + ℓ 5 + ⋯ ) . (86) Let σ a = ρ n ℓ n − 1 ∘ ⋯ ∘ ρ 2 ℓ 1 \sigma_a=\rho_n^{\ell_{n-1}}\circ\cdots\circ\rho_2^{\ell_1} σ a = ρ n ℓ n − 1 ∘ ⋯ ∘ ρ 2 ℓ 1 , evaluated with (84) . Then a ↦ σ a a\mapsto\sigma_a a ↦ σ a is a bijection A n ! / 2 → Alt n \mathcal{A}_{n!/2}\to\operatorname{Alt}_n A n ! /2 → Alt n , and
a = ∑ h = 2 n − 1 ℓ h h ! 2 = ⌊ r ( σ a ) 2 ⌋ . a=\sum_{h=2}^{n-1}\ell_h\frac{h!}{2}=\left\lfloor\frac{r(\sigma_a)}{2}\right\rfloor. a = h = 2 ∑ n − 1 ℓ h 2 h ! = ⌊ 2 r ( σ a ) ⌋ . For a , b ∈ A n ! / 2 a,b\in\mathcal{A}_{n!/2} a , b ∈ A n ! /2 let u h = ℓ h ( σ a ∘ σ b ) u_h=\ell_h(\sigma_a\circ\sigma_b) u h = ℓ h ( σ a ∘ σ b ) be the exponents given by (85) . Then
a ⋆ ρ b = ∑ h = 2 n − 1 u h h ! 2 (87) a\mathbin{\star_\rho}b=\sum_{h=2}^{n-1}u_h\frac{h!}{2} \tag*{(87)} a ⋆ ρ b = h = 2 ∑ n − 1 u h 2 h ! (87) makes A n ! / 2 \mathcal{A}_{n!/2} A n ! /2 a group isomorphic to Alt n \operatorname{Alt}_n Alt n , with identity 0 0 0 .
Proof. By Lemma 11.51, sgn σ = ( − 1 ) ℓ 1 + ℓ 3 + ℓ 5 + ⋯ \operatorname{sgn}\sigma=(-1)^{\ell_1+\ell_3+\ell_5+\cdots} sgn σ = ( − 1 ) ℓ 1 + ℓ 3 + ℓ 5 + ⋯ . Apply Lemma 11.44 with U = A 3 × ⋯ × A n U=\mathcal{A}_3\times\cdots\times\mathcal{A}_n U = A 3 × ⋯ × A n , parity digit e = ℓ 1 e=\ell_1 e = ℓ 1 and q = ℓ 3 + ℓ 5 + ⋯ ( m o d 2 ) q=\ell_3+\ell_5+\cdots\pmod{2} q = ℓ 3 + ℓ 5 + ⋯ ( mod 2 ) . Mixed-radix evaluation of the remaining digits has place values h ! / 2 h!/2 h ! /2 , which gives the bijection and the first expression for a a a . Since r ( σ a ) = ℓ 1 + 2 a r(\sigma_a)=\ell_1+2a r ( σ a ) = ℓ 1 + 2 a with ℓ 1 ∈ A 2 \ell_1\in\mathcal{A}_2 ℓ 1 ∈ A 2 , also a = ⌊ r ( σ a ) / 2 ⌋ a=\lfloor r(\sigma_a)/2\rfloor a = ⌊ r ( σ a ) /2 ⌋ . The product σ a ∘ σ b \sigma_a\circ\sigma_b σ a ∘ σ b is even, so u 1 u_1 u 1 is its restored parity digit, and (87) is its label. For a = 0 a=0 a = 0 all exponents vanish and σ 0 \sigma_0 σ 0 is the identity. □
Example 11.54. For n = 5 n=5 n = 5 the labels a ∈ A 60 a\in\mathcal{A}_{60} a ∈ A 60 have digits ℓ 2 ∈ A 3 \ell_2\in\mathcal{A}_3 ℓ 2 ∈ A 3 , ℓ 3 ∈ A 4 \ell_3\in\mathcal{A}_4 ℓ 3 ∈ A 4 , ℓ 4 ∈ A 5 \ell_4\in\mathcal{A}_5 ℓ 4 ∈ A 5 , with a = ℓ 2 + 3 ℓ 3 + 12 ℓ 4 a=\ell_2+3\ell_3+12\ell_4 a = ℓ 2 + 3 ℓ 3 + 12 ℓ 4 and ℓ 1 = d 2 ( 0 , ℓ 3 ) \ell_1=d_2(0,\ell_3) ℓ 1 = d 2 ( 0 , ℓ 3 ) . The first labels, with σ a \sigma_a σ a written as σ a ( 0 ) ⋯ σ a ( 4 ) \sigma_a(0)\cdots\sigma_a(4) σ a ( 0 ) ⋯ σ a ( 4 ) , are
The labels 0, 1, 2 form the subgroup Alt 3 \operatorname{Alt}_3 Alt 3 , and the labels 0, … \ldots … , 11 the subgroup Alt 4 \operatorname{Alt}_4 Alt 4 fixing the letter 4. A computer check of (87) confirms associativity for n ≤ 5 n\leq5 n ≤ 5 and gives the conjugacy class sizes 1, 12, 12, 15, 20 for n = 5 n=5 n = 5 and 1, 40, 40, 45, 72, 72, 90 for n = 6 n=6 n = 6 , those of the simple groups Alt 5 \operatorname{Alt}_5 Alt 5 and Alt 6 \operatorname{Alt}_6 Alt 6 .
A third chain labelling comes directly from Theorem 8.32 , with cyclic shifts as coset representatives.
Corollary 11.55 (Symmetric and alternating groups). Let r ( x ) = d n ( 0 , x + 1 ) r(x)=d_n(0,x+1) r ( x ) = d n ( 0 , x + 1 ) on A n \mathcal{A}_n A n . For Sym n \operatorname{Sym}_n Sym n , and for Alt n \operatorname{Alt}_n Alt n with n n n odd, take s t = r t s_t=r^t s t = r t . Then (70) becomes
( t , h ) ∗ ( u , k ) = ( d n ( 0 , t + h ( u ) ) , h ( u ) ∘ k ) , h ( u ) ( x ) = d n ( 0 , h ( x + u ) − h ( u ) ) . (88) (t,h)\ast(u,k)=\left(d_n(0,t+h(u)),h^{(u)}\circ k\right),\qquad h^{(u)}(x)=d_n(0,h(x+u)-h(u)). \tag*{(88)} ( t , h ) ∗ ( u , k ) = ( d n ( 0 , t + h ( u )) , h ( u ) ∘ k ) , h ( u ) ( x ) = d n ( 0 , h ( x + u ) − h ( u )) . (88) The stabiliser of 0 is Sym n − 1 \operatorname{Sym}_{n-1} Sym n − 1 (respectively Alt n − 1 \operatorname{Alt}_{n-1} Alt n − 1 ) acting on { 1 , … , n − 1 } \{1,\ldots,n-1\} { 1 , … , n − 1 } . Relabelling x ↦ x − 1 x\mapsto x-1 x ↦ x − 1 and repeating gives a mixed-radix code
m = t n + n ( t n − 1 + ( n − 1 ) ( t n − 2 + ⋯ ) ) , t j ∈ A j , m=t_n+n\left(t_{n-1}+(n-1)(t_{n-2}+\cdots)\right),\qquad t_j\in\mathcal{A}_j, m = t n + n ( t n − 1 + ( n − 1 ) ( t n − 2 + ⋯ ) ) , t j ∈ A j , with radices n , n − 1 , … , 2 n,n-1,\ldots,2 n , n − 1 , … , 2 for Sym n \operatorname{Sym}_n Sym n and n , n − 1 , … , 3 n,n-1,\ldots,3 n , n − 1 , … , 3 for Alt n \operatorname{Alt}_n Alt n . For Alt n \operatorname{Alt}_n Alt n with n n n even, take s t = r t s_t=r^t s t = r t for t t t even and s t = r t ∘ ( 1 2 ) s_t=r^t\circ(1\,2) s t = r t ∘ ( 1 2 ) for t t t odd.
Proof. For s t = r t s_t=r^t s t = r t we get t ′ = t + h ( u ) t'=t+h(u) t ′ = t + h ( u ) and s t ′ − 1 t s t ′ = r − h ( u ) h r u s_{t'}^{-1}t s_{t'}=r^{-h(u)}h r^u s t ′ − 1 t s t ′ = r − h ( u ) h r u , which sends x x x to h ( x + u ) − h ( u ) h(x+u)-h(u) h ( x + u ) − h ( u ) . For odd n n n the n n n -cycle r r r is even, so s t ∈ Alt n s_t\in\operatorname{Alt}_n s t ∈ Alt n . For even n n n the given s t s_t s t are even and satisfy s t ( 0 ) = t s_t(0)=t s t ( 0 ) = t , because ( 1 2 ) (1\ 2) ( 1 2 ) fixes 0. Now apply Theorem 8.32 .
Remark 11.56. Corollary 11.55 also uses rotations as coset representatives, but it peels off the stabiliser of a point and needs an extra transposition when n n n is even. The prefix rotations together with the parity digit ℓ 1 \ell_1 ℓ 1 give one formula for all n n n .
Remark 11.57 (Why the decoding is a recursion). For n ≥ 5 n\ge5 n ≥ 5 the group Alt n \operatorname{Alt}_n Alt n is simple, so no layer projector of a labelling of Alt n \operatorname{Alt}_n Alt n onto a layer of size strictly between 1 and n ! / 2 n!/2 n ! /2 is a homomorphism: every layer is coupled to the layers above it. Correspondingly the decoding (85) runs from the top layer down, each exponent depending on those above it, much as carries propagate in ordinary addition.
Comparison and checks
Remark 11.58 (Two chain labellings). The Lehmer and rotation labels are two stabiliser-chain coordinate systems on the same groups.
Table 11.9.3.
In both cases Lemma 11.44 produces the Alt n \operatorname{Alt}_n Alt n law by deleting the radix-2 digit. The relabellings
Φ = r ∘ λ − 1 : ( A n ! , ∗ L ) → ( A n ! , ∗ ρ ) , Φ A = ⌊ r ∘ μ − 1 / 2 ⌋ : ( A n ! / 2 , ⋆ L ) → ( A n ! / 2 , ⋆ ρ ) \Phi=r\circ\lambda^{-1}:(A_{n!},\ast_L)\to(A_{n!},\ast_\rho),\qquad\Phi_A=\lfloor r\circ\mu^{-1}/2\rfloor:(A_{n!/2},\star_L)\to(A_{n!/2},\star_\rho) Φ = r ∘ λ − 1 : ( A n ! , ∗ L ) → ( A n ! , ∗ ρ ) , Φ A = ⌊ r ∘ μ − 1 /2 ⌋ : ( A n ! /2 , ⋆ L ) → ( A n ! /2 , ⋆ ρ )
are explicit group isomorphisms. The Lehmer coordinates make the Coxeter generators, the sign and the cosets of Alt n \operatorname{Alt}_n Alt n transparent. The rotation coordinates make the nested subgroups Sym 2 < Sym 3 < ⋯ < Sym n \operatorname{Sym}_2<\operatorname{Sym}_3<\cdots<\operatorname{Sym}_n Sym 2 < Sym 3 < ⋯ < Sym n and Alt 3 < ⋯ < Alt n \operatorname{Alt}_3<\cdots<\operatorname{Alt}_n Alt 3 < ⋯ < Alt n transparent: they are initial segments of labels.
Remark 11.59 (Checks). The script scripts/sn_law_verify.py checks both labellings exhaustively for n ≤ 6 n\le6 n ≤ 6 . For the Lehmer laws it verifies encoding and decoding, identities and inverses, both generator rules, the sign and coset rule, the Alt n \operatorname{Alt}_n Alt n product formula, the conjugation rule and the chain property. For the rotation laws it verifies Lemma 11.51, Theorems 11.52 and 11.53 , the sign formula, the identity a = ⌊ r ( σ a ) / 2 ⌋ a=\lfloor r(\sigma_a)/2\rfloor a = ⌊ r ( σ a ) /2 ⌋ , the parity toggle and the relabelling between the two Alt n \operatorname{Alt}_n Alt n laws. For n ≤ 5 n\le5 n ≤ 5 it forms the full Cayley tables and checks associativity. When GAP is available the right regular representations of all four laws are checked with IsomorphismGroups against SymmetricGroup(n) and AlternatingGroup(n) for $n\le 6. The script writes Table 11.6 and Table 11.7 and is called by scripts/sn_verify.py. GAP 4.14.0 [30] confirms the shift labelling of Corollary 11.55 on small cases: the codes of Alt 5 \operatorname{Alt}_5 Alt 5 and Alt 6 \operatorname{Alt}_6 Alt 6 have IdGroup [ 60 , 5 ] [60,5] [ 60 , 5 ] and [ 360 , 118 ] [360,118] [ 360 , 118 ] , and the code of Alt 7 \operatorname{Alt}_7 Alt 7 gives a simple group isomorphic to Alt 7 \operatorname{Alt}_7 Alt 7 .
Table 11.7. Data of the Lehmer laws of Sym n \operatorname{Sym}_n Sym n and Alt n \operatorname{Alt}_n Alt n for 3 ≤ n ≤ 8 3 \le n \le8 3 ≤ n ≤ 8 : generator labels Π i = n ! / ( n − i ) ! \Pi_i = n!/(n-i)! Π i = n ! / ( n − i )! , the label n ! − 1 n!-1 n ! − 1 of the reversal w 0 w_0 w 0 and its sign ( − 1 ) n ( n − 1 ) / 2 (-1)^{n(n-1)/2} ( − 1 ) n ( n − 1 ) /2 , the order of Alt n \operatorname{Alt}_n Alt n , and the Alt n \operatorname{Alt}_n Alt n -label of the 3-cycle ( 0 1 2 ) (0\ 1\ 2) ( 0 1 2 ) .
Worked products, inverses and conjugates
This section computes by hand a product in the Lehmer laws of Sym 5 \operatorname{Sym}_5 Sym 5 and of Alt 5 \operatorname{Alt}_5 Alt 5 , an inverse and a conjugate, in the manner of Chapter 14. Labels are decoded with Proposition 1.8 and Theorem 11.2 , and permutations are composed from right to left, as in (12) .
Example 11.60 (A product in Sym 5 \operatorname{Sym}_5 Sym 5 ). We compute 68 ∗ 101 68 * 101 68 ∗ 101 in the Lehmer law of Sym 5 \operatorname{Sym}_5 Sym 5 (Theorem 11.41 ), whose place values are Π i = 1 , 5 , 20 , 60 , 120 \Pi_i = 1, 5, 20, 60, 120 Π i = 1 , 5 , 20 , 60 , 120 . By Proposition 1.8 ,
68 = 13 ⋅ 5 + 3 , 13 = 3 ⋅ 4 + 1 , 3 = 1 ⋅ 3 + 0 , 1 = 0 ⋅ 2 + 1 , 0 = 0 ⋅ 1 + 0 , 68 = 13 \cdot5 + 3,\qquad13 = 3 \cdot4 + 1,\qquad3 = 1 \cdot3 + 0,\qquad1 = 0 \cdot2 + 1,\qquad0 = 0 \cdot1 + 0, 68 = 13 ⋅ 5 + 3 , 13 = 3 ⋅ 4 + 1 , 3 = 1 ⋅ 3 + 0 , 1 = 0 ⋅ 2 + 1 , 0 = 0 ⋅ 1 + 0 ,
so 68 68 68 has digits ( a 0 , … , a 4 ) = ( 3 , 1 , 0 , 1 , 0 ) (a_0,\ldots,a_4)=(3,1,0,1,0) ( a 0 , … , a 4 ) = ( 3 , 1 , 0 , 1 , 0 ) . Reading each digit as a rank among the values not yet used (Theorem 11.2 ) gives R = λ − 1 ( 68 ) = ( 3 , 1 , 0 , 4 , 2 ) R=\lambda^{-1}(68)=(3,1,0,4,2) R = λ − 1 ( 68 ) = ( 3 , 1 , 0 , 4 , 2 ) in one-line notation. In the same way 101 101 101 has digits ( 1 , 0 , 2 , 1 , 0 ) (1,0,2,1,0) ( 1 , 0 , 2 , 1 , 0 ) and S = λ − 1 ( 101 ) = ( 1 , 0 , 4 , 3 , 2 ) S=\lambda^{-1}(101)=(1,0,4,3,2) S = λ − 1 ( 101 ) = ( 1 , 0 , 4 , 3 , 2 ) . Composing from right to left, ( R ∘ S ) ( i ) = R ( S ( i ) ) (R\circ S)(i)=R(S(i)) ( R ∘ S ) ( i ) = R ( S ( i )) , gives R ∘ S = ( 1 , 3 , 2 , 4 , 0 ) R\circ S=(1,3,2,4,0) R ∘ S = ( 1 , 3 , 2 , 4 , 0 ) , whose inversion counts are ( 1 , 2 , 1 , 1 , 0 ) (1,2,1,1,0) ( 1 , 2 , 1 , 1 , 0 ) . Hence
68 ∗ 101 = λ ( R ∘ S ) = 1 ⋅ 1 + 2 ⋅ 5 + 1 ⋅ 20 + 1 ⋅ 60 = 91. 68 * 101=\lambda(R\circ S)=1\cdot1+2\cdot5+1\cdot20+1\cdot60=91. 68 ∗ 101 = λ ( R ∘ S ) = 1 ⋅ 1 + 2 ⋅ 5 + 1 ⋅ 20 + 1 ⋅ 60 = 91.
Example 11.61 (A product in Alt 5 \operatorname{Alt}_5 Alt 5 ). We compute 23 ∗ 41 23 * 41 23 ∗ 41 in the Lehmer law of Alt 5 \operatorname{Alt}_5 Alt 5 (Theorem 11.45), with N = 60 N=60 N = 60 . A label m ∈ A 60 m\in A_{60} m ∈ A 60 is decoded through its Lehmer digits in Sym 5 \operatorname{Sym}_5 Sym 5 : if ε ( m ) \varepsilon(m) ε ( m ) is the parity of their sum, then m + 60 ε ( m ) m+60\varepsilon(m) m + 60 ε ( m ) is the Lehmer label of an even permutation, and a product is encoded by reducing its Lehmer label modulo 60 60 60 . Here 23 23 23 has digits ( 3 , 0 , 1 , 0 , 0 ) (3,0,1,0,0) ( 3 , 0 , 1 , 0 , 0 ) with sum 4 4 4 , so ε ( 23 ) = 0 \varepsilon(23)=0 ε ( 23 ) = 0 and R = λ − 1 ( 23 ) = ( 3 , 0 , 2 , 1 , 4 ) R=\lambda^{-1}(23)=(3,0,2,1,4) R = λ − 1 ( 23 ) = ( 3 , 0 , 2 , 1 , 4 ) . Similarly 41 41 41 has digits ( 1 , 0 , 2 , 0 , 0 ) (1,0,2,0,0) ( 1 , 0 , 2 , 0 , 0 ) with sum 3 3 3 , so ε ( 41 ) = 1 \varepsilon(41)=1 ε ( 41 ) = 1 and S = λ − 1 ( 101 ) = ( 1 , 0 , 4 , 3 , 2 ) S=\lambda^{-1}(101)=(1,0,4,3,2) S = λ − 1 ( 101 ) = ( 1 , 0 , 4 , 3 , 2 ) . Then R ∘ S = ( 0 , 3 , 4 , 1 , 2 ) R\circ S=(0,3,4,1,2) R ∘ S = ( 0 , 3 , 4 , 1 , 2 ) has digits ( 0 , 2 , 2 , 0 , 0 ) (0,2,2,0,0) ( 0 , 2 , 2 , 0 , 0 ) , so
λ ( R ∘ S ) = 2 ⋅ 5 + 2 ⋅ 20 = 50 , 23 ∗ 41 = 50 mod 60 = 50. \lambda(R\circ S)=2\cdot5+2\cdot20=50,\qquad23*41=50\ \operatorname{mod}60=50. λ ( R ∘ S ) = 2 ⋅ 5 + 2 ⋅ 20 = 50 , 23 ∗ 41 = 50 mod 60 = 50.
Example 11.62 (An inverse in Sym 5 \operatorname{Sym}_5 Sym 5 ). Let R = λ − 1 ( 68 ) = ( 3 , 1 , 0 , 4 , 2 ) R=\lambda^{-1}(68)=(3,1,0,4,2) R = λ − 1 ( 68 ) = ( 3 , 1 , 0 , 4 , 2 ) be as in Example 11.60. Since R − 1 ( R ( i ) ) = i R^{-1}(R(i))=i R − 1 ( R ( i )) = i , the word of R − 1 R^{-1} R − 1 lists the positions of 0 , 1 , … , 4 0,1,\ldots,4 0 , 1 , … , 4 in the word of R R R , so R − 1 = ( 2 , 1 , 4 , 0 , 3 ) R^{-1}=(2,1,4,0,3) R − 1 = ( 2 , 1 , 4 , 0 , 3 ) . Its inversion counts are ( 2 , 1 , 2 , 0 , 0 ) (2,1,2,0,0) ( 2 , 1 , 2 , 0 , 0 ) , hence 68 − 1 = λ ( R − 1 ) = 2 ⋅ 1 + 1 ⋅ 5 + 2 ⋅ 20 = 47 68^{-1}=\lambda(R^{-1})=2\cdot1+1\cdot5+2\cdot20=47 6 8 − 1 = λ ( R − 1 ) = 2 ⋅ 1 + 1 ⋅ 5 + 2 ⋅ 20 = 47 . The script inverse.py reaches the same label inside the law, without decoding (Remark 14.15): the powers of 68 68 68 return to 68 68 68 after 4 4 4 steps, so 68 68 68 has order 4 4 4 and 68 − 1 = 68 3 = 47 68^{-1}=68^3=47 6 8 − 1 = 6 8 3 = 47 .
Example 11.63 (A conjugate in Sym 5 \operatorname{Sym}_5 Sym 5 ). With R R R and S S S as in Example 11.60, the conjugate R ∘ S ∘ R − 1 R\circ S\circ R^{-1} R ∘ S ∘ R − 1 maps R ( i ) R(i) R ( i ) to R ( S ( i ) ) R(S(i)) R ( S ( i )) . It is therefore obtained from the cycles of S S S by applying R R R to every entry: S = ( 01 ) ( 24 ) S=(01)(24) S = ( 01 ) ( 24 ) gives R ∘ S ∘ R − 1 = ( 31 ) ( 02 ) R\circ S\circ R^{-1}=(31)(02) R ∘ S ∘ R − 1 = ( 31 ) ( 02 ) , with word ( 2 , 3 , 0 , 1 , 4 ) (2,3,0,1,4) ( 2 , 3 , 0 , 1 , 4 ) and inversion counts ( 2 , 2 , 0 , 0 , 0 ) (2,2,0,0,0) ( 2 , 2 , 0 , 0 , 0 ) . Hence
68 ∗ 101 ∗ 68 − 1 = 2 ⋅ 1 + 2 ⋅ 5 = 12 , 68*101*68^{-1}=2\cdot1+2\cdot5=12, 68 ∗ 101 ∗ 6 8 − 1 = 2 ⋅ 1 + 2 ⋅ 5 = 12 ,
which is the output of conjugate.py Sym(5) 68 101. The other convention gives 68 − 1 ∗ 101 ∗ 68 = 65 68^{-1}*101*68=65 6 8 − 1 ∗ 101 ∗ 68 = 65 (option --right).
For larger degrees the same computations are carried out by the scripts. The following outputs were obtained with --check and agree with direct composition of the decoded permutations.
Table 11.
Remark 11.64 (Checks). The products, inverses and conjugates of this section are reproduced by multiply.py, inverse.py and conjugate.py, run from the directory scripts as in Remarks 14.14 and 14.15. The group names are Sym ( n ) \operatorname{Sym}(n) Sym ( n ) and Alt ( n ) \operatorname{Alt}(n) Alt ( n ) . In zsh, quote a name containing parentheses, for example ’Sym(20)’, or use the aliases Sym20, S20, Alt12 and A12. With --check, multiply.py compares each product of Sym n \operatorname{Sym}_n Sym n or Alt n \operatorname{Alt}_n Alt n with a separately coded ranking of permutations. Labels of any size are accepted; decimal labels with more than 4300 digits, which occur for n n n larger than about 1560, require raising the default limit of Python on the conversion of long decimal strings.
Remark 11.65 (Conjugates). The conjugate of a label n n n by a label m m m , in any group of the archive, is computed by
python3 conjugate.py GROUP m n [--right] [--check]
which prints m ∗ n ∗ m − 1 m \ast n \ast m^{-1} m ∗ n ∗ m − 1 , the image of n n n under the inner automorphism x ↦ m x m − 1 x \mapsto mxm^{-1} x ↦ m x m − 1 ; with --right it prints m − 1 ∗ n ∗ m m^{-1} \ast n \ast m m − 1 ∗ n ∗ m . The script uses only the law of multiply.py and the inverse of inverse.py, so it applies without change to every group for which these scripts are available. With --check it verifies m ∗ m − 1 = m − 1 ∗ m = e m \ast m^{-1} = m^{-1} \ast m = e m ∗ m − 1 = m − 1 ∗ m = e , the identity ( m ∗ n ∗ m − 1 ) ∗ m = m ∗ n (m \ast n \ast m^{-1}) \ast m = m \ast n ( m ∗ n ∗ m − 1 ) ∗ m = m ∗ n , and that n n n and its conjugate have the same order.
Generated tables
The numbers in this section depend only on the partitions of n n n . A conjugacy class of Sym n \operatorname{Sym}_n Sym n is given by a cycle type λ = ( 1 m 1 2 m 2 ⋯ ) \lambda= (1^{m_1}2^{m_2}\cdots) λ = ( 1 m 1 2 m 2 ⋯ ) with ℓ ( λ ) \ell(\lambda) ℓ ( λ ) parts (§ 19.3 ). For g g g in the class C λ C_\lambda C λ ,
∣ C λ ∣ = n ! z λ , z λ = ∏ i i m i m i ! , o ( g ) = lcm ( λ 1 , λ 2 , … ) , sgn g = ( − 1 ) n − ℓ ( λ ) , |C_\lambda| = \frac{n!}{z_\lambda}, \qquad z_\lambda= \prod_i i^{m_i}m_i!, \qquad o(g) = \operatorname{lcm}(\lambda_1,\lambda_2,\ldots), \qquad\operatorname{sgn} g = (-1)^{n-\ell(\lambda)}, ∣ C λ ∣ = z λ n ! , z λ = i ∏ i m i m i ! , o ( g ) = lcm ( λ 1 , λ 2 , … ) , sgn g = ( − 1 ) n − ℓ ( λ ) ,
where z λ z_\lambda z λ is the order of the centraliser of g g g . The class lies in Alt n \operatorname{Alt}_n Alt n when n − ℓ ( λ ) n-\ell(\lambda) n − ℓ ( λ ) is even, and then it splits into two classes of Alt n \operatorname{Alt}_n Alt n exactly when the parts of λ \lambda λ are distinct and odd. The irreducible character χ λ \chi^\lambda χ λ of Sym n \operatorname{Sym}_n Sym n has degree f λ = n ! / ∏ x h ( x ) f^\lambda= n!/\prod_x h(x) f λ = n ! / ∏ x h ( x ) , the product running over the hook lengths of the diagram of $\lambda ({xref}Section 11.7). On $\operatorname{Alt}_n$, $\chi^\lambda$ and $\chi^{\lambda'}$ restrict to the same irreducible character when $\lambda \ne \lambda'$, and $\chi^\lambda$ splits into two characters of degree $f^\lambda/2$ when $\lambda = \lambda'$ ({xref}Section 11.8`). We write k ( G ) k(G) k ( G ) for the number of conjugacy classes of G G G .
In Table 11.12 , e ( G ) e(G) e ( G ) is the exponent of G G G , g ( n ) g(n) g ( n ) is Landau’s function, the largest order of an element of Sym n \operatorname{Sym}_n Sym n , g A ( n ) g_A(n) g A ( n ) is the largest order of an element of Alt n \operatorname{Alt}_n Alt n , and t ( G ) t(G) t ( G ) is the number of x ∈ G x \in G x ∈ G with x 2 = 1 x^2 = 1 x 2 = 1 , the identity included. Every irreducible representation of Sym n \operatorname{Sym}_n Sym n is realised over Q \mathbb{Q} Q , so every Frobenius–Schur indicator is 1 and t ( Sym n ) t(\operatorname{Sym}_n) t ( Sym n ) is the sum of the degrees in Table 11.13 . It also satisfies t ( Sym n ) = t ( Sym n − 1 ) + ( n − 1 ) t ( Sym n − 2 ) t(\operatorname{Sym}_n) = t(\operatorname{Sym}_{n-1}) + (n-1)t(\operatorname{Sym}_{n-2}) t ( Sym n ) = t ( Sym n − 1 ) + ( n − 1 ) t ( Sym n − 2 ) , since a solution either fixes the letter n − 1 n-1 n − 1 or exchanges it with one of the other n − 1 n-1 n − 1 letters.
Table 11.12. Exponents, largest element orders and solutions of x 2 = 1 x^2 = 1 x 2 = 1 .
Table 11.13. Character degrees of Sym n \operatorname{Sym}_n Sym n and Alt n \operatorname{Alt}_n Alt n ; exponents are multiplicities.
Remark 11.66 (Checks). The script scripts/sn_tables.py writes Table 11.12 , Table 11.13 and Table 11.14 , and a data file with the same data for n ≤ 10 n \le10 n ≤ 10 . It writes them only after the following checks pass for every n ≤ 10 n \le10 n ≤ 10 . The class sizes add up to n ! n! n ! , and the number of partitions agrees with Euler’s pentagonal recurrence. The hook formula agrees with the branching rule f λ = ∑ f λ − □ f^\lambda= \sum f^{\lambda-\square} f λ = ∑ f λ − □ , and the squared degrees add up to n ! n! n ! for Sym n \operatorname{Sym}_n Sym n and to n ! / 2 n!/2 n ! /2 for Alt n \operatorname{Alt}_n Alt n . The numbers of classes and of irreducible characters of Alt n \operatorname{Alt}_n Alt n agree, and there are as many self-conjugate partitions as partitions into distinct odd parts. The three values of t ( Sym n ) t(\operatorname{Sym}_n) t ( Sym n ) , from the classes, from the recurrence and as the sum of the degrees, agree, and g ( n ) g(n) g ( n ) equals the largest product of powers of distinct primes with sum at most n n n . For n ≤ 8 n \le8 n ≤ 8 the script also enumerates Sym n \operatorname{Sym}_n Sym n : it recomputes cycle types, orders, signs, exponents and t t t , and checks the splitting of each class of Alt n \operatorname{Alt}_n Alt n by computing its orbit under conjugation by the 3-cycles ( 0 1 i ) (0\ 1\ i) ( 0 1 i ) . It compares the class numbers with Table 11.1 . When GAP is available it compares the degrees with CharacterTable("Symmetric",n) and CharacterTable("Alternating",n) for n ≤ 10 n \le10 n ≤ 10 .
Table 11.14. The 22 conjugacy classes of Sym 8 \operatorname{Sym}_8 Sym 8 .
Exercises
Show that k ( Sym 6 ) = 11 k(\operatorname{Sym}_6) = 11 k ( Sym 6 ) = 11 and list the classes of Sym 6 \operatorname{Sym}_6 Sym 6 with their sizes.
Show that Sym 7 \operatorname{Sym}_7 Sym 7 has 8 classes of even permutations, that only the class of 7-cycles splits in Alt 7 \operatorname{Alt}_7 Alt 7 , and deduce k ( Alt 7 ) = 9 k(\operatorname{Alt}_7) = 9 k ( Alt 7 ) = 9 .
Use the hook length formula to show that the irreducible characters of Sym 5 \operatorname{Sym}_5 Sym 5 have degrees 1, 1, 4, 4, 5, 5, 6, and check that the squares add up to 120.
Show that Alt n \operatorname{Alt}_n Alt n is generated by its 3-cycles for n ≥ 3 n \ge3 n ≥ 3 .
Show that the classes of cycle types 2 2 2 and 2 3 2^3 2 3 , 3 3 3 and 3 2 3^2 3 2 , and 6 6 6 and 3 2 1 3\ 2\ 1 3 2 1 of Sym 6 \operatorname{Sym}_6 Sym 6 have equal sizes in pairs, as required by the outer automorphism.
Show that the number of permutations of A n A_n A n with exactly k k k inversions is the coefficient of t k t^k t k in ∏ i = 1 n ( 1 + t + ⋯ + t i − 1 ) \prod_{i=1}^{n}(1+t+\cdots+t^{i-1}) ∏ i = 1 n ( 1 + t + ⋯ + t i − 1 ) .
Finite simple groups of Lie type
This chapter gives digital laws for the finite simple groups of Lie type: the classical groups, the Suzuki and Ree groups, and the exceptional groups up to E 8 ( q ) E_8(q) E 8 ( q ) . Simplicity has a direct meaning for digital laws. A layer is drawn as a solid node exactly when its layer projector is a homomorphism, and then its kernel is a normal subgroup. A simple group has no proper nontrivial normal subgroups, so no labelling of it can be graded, and it cannot be built by stacking hollow layers on solid ones. Its laws must come from another principle. For the alternating groups of Chapter 11 it was a normal form of permutations; here it is a normal form of matrices, mostly from the Bruhat decomposition, which is uniform in q q q . After the field arithmetic, the smallest family PSL ( 2 , p ) \operatorname{PSL}(2,p) PSL ( 2 , p ) is treated in § 20.3 .
Dynkin diagrams, Weyl data and digital laws
A Dynkin diagram is the compact combinatorial record of the root data that organise this chapter. Its vertices are the simple roots, and its edges encode the Cartan integers. Thus it determines the Cartan matrix, the Coxeter presentation of the Weyl group, the positive roots and, in particular, the Bruhat cells used below. It does not replace the matrix calculations: it predicts their finite combinatorics and says which calculations belong to the same family.
For a split group of rank r r r , let Φ + \Phi^+ Φ + be its positive roots and let W W W be its Weyl group. The Borel subgroup has the form B = U T B=UT B = U T , with one field coordinate for each positive-root subgroup in U U U and r r r multiplicative coordinates in T T T (subject to the central quotient when one passes to a simple group). The Bruhat decomposition
G = ∐ w ∈ W B w B G=\coprod_{w\in W}BwB G = w ∈ W ∐ B w B
therefore gives the architecture of a digital label: first a Weyl digit selecting w w w , then coordinates for the two unipotent pieces and the torus. The explicit normal forms of § 20.4 –§ 22.4 are implementations of this architecture, with the elimination procedure determining the digits rather than merely counting them.
The classical diagrams explain the matrix restrictions. Type A r A_r A r has Weyl group Sym r + 1 \operatorname{Sym}_{r+1} Sym r + 1 , so the pivot permutations in Gaussian elimination are ordinary permutations. Types B r B_r B r and C r C_r C r have signed permutations, and type D r D_r D r has the even signed permutations; these are exactly the permutations preserving the antidiagonal forms in § 20.5 . The explicit Coxeter generators of Chapter 11 are therefore the type-A A A prototype of the Weyl digits used here.
Twisting has an equally simple diagrammatic meaning. A symmetry of the diagram may be composed with the Frobenius map t ↦ t q t\mapsto t^q t ↦ t q : this gives 2 A r {}^{2}A_r 2 A r , 2 D r {}^{2}D_r 2 D r , 3 D 4 {}^{3}D_4 3 D 4 and 2 E 6 {}^{2}E_6 2 E 6 . In the corresponding digital laws one chooses coordinates constant on the resulting root orbits, with the field relation prescribed by the same Frobenius action. The Suzuki and Ree constructions are the exceptional analogue: a special isogeny interchanges the long- and short-root data in characteristics 2 or 3, as is made explicit in § 20.7 , § 20.8 and § 21.3 .
Table 12.1. Dynkin data behind the digital laws of Chapter 12. A double or triple edge is directed toward the short root.
This table is complete for the families constructed in the chapter. The small exceptional simple group usually called the Tits group is not an omission: it is constructed in § 21.3 as 2 F 4 ( 2 ) ′ {}^{2}F_4(2)' 2 F 4 ( 2 ) ′ . Thus the diagram is a map of the existing digital laws, not a promise of laws deferred to another chapter.
Finite fields in digital form
The groups of Lie type are matrix groups over finite fields, so their digital laws rest on a digital form of the field arithmetic. Addition turns out to be the carry-free sum, and multiplication is a quadratic digital operator.
Definition 12.1 (Polynomial coordinates). Let p p p be prime, q = p r q=p^r q = p r , and let f ( X ) = X r + ∑ j < r c j X j f(X)=X^r+\sum_{j<r}c_jX^j f ( X ) = X r + ∑ j < r c j X j be irreducible over F p \mathbb{F}_p F p , so that F q = F p [ θ ] / ( f ( θ ) ) \mathbb{F}_q=\mathbb{F}_p[\theta]/(f(\theta)) F q = F p [ θ ] / ( f ( θ )) . The element a = ∑ j < r a j θ j a=\sum_{j<r}a_j\theta^j a = ∑ j < r a j θ j is labelled by [ a ] = ∑ j < r a j p j ∈ A q [a]=\sum_{j<r}a_jp^j\in A_q [ a ] = ∑ j < r a j p j ∈ A q . The structure constants C i j k ∈ A p C_{ij}^k\in A_p C ij k ∈ A p , i , j , k < r i,j,k<r i , j , k < r , are the digits of θ i + j = ∑ k < r C i j k θ k \theta^{i+j}=\sum_{k<r}C_{ij}^k\theta^k θ i + j = ∑ k < r C ij k θ k , and E j k ∈ A p E_j^k\in A_p E j k ∈ A p are the digits of θ p j = ∑ k < r E j k θ k \theta^{pj}=\sum_{k<r}E_j^k\theta^k θ p j = ∑ k < r E j k θ k .
Proposition 12.2 (Digital field arithmetic). For labels x = [ a ] x=[a] x = [ a ] and y = [ b ] y=[b] y = [ b ] in A q A_q A q ,
[ a + b ] = ∑ k < r p k d p ( 0 , d p ( k , x ) + d p ( k , y ) ) = x ⊕ p y , [ − a ] = ∑ k < r p k d p ( 0 , p − d p ( k , x ) ) , [a+b]=\sum_{k<r}p^k d_p(0,d_p(k,x)+d_p(k,y))=x\oplus_p y,\qquad[-a]=\sum_{k<r}p^k d_p(0,p-d_p(k,x)), [ a + b ] = k < r ∑ p k d p ( 0 , d p ( k , x ) + d p ( k , y )) = x ⊕ p y , [ − a ] = k < r ∑ p k d p ( 0 , p − d p ( k , x )) , [ a b ] = ∑ k < r p k d p ( 0 , ∑ i , j < r d p ( i , x ) d p ( j , y ) C i j k ) , [ a p ] = ∑ k < r p k d p ( 0 , ∑ j < r d p ( j , x ) E j k ) . [ab]=\sum_{k<r}p^k d_p\left(0,\sum_{i,j<r}d_p(i,x)d_p(j,y)C_{ij}^k\right),\qquad[a^p]=\sum_{k<r}p^k d_p\left(0,\sum_{j<r}d_p(j,x)E_j^k\right). [ ab ] = k < r ∑ p k d p ( 0 , i , j < r ∑ d p ( i , x ) d p ( j , y ) C ij k ) , [ a p ] = k < r ∑ p k d p ( 0 , j < r ∑ d p ( j , x ) E j k ) .
Proof. The digits of x x x are the coordinates a j a_j a j . Addition and negation are coordinatewise modulo p p p , which is the carry-free sum ⊕ p \oplus_p ⊕ p of Section 1.3. By distributivity a b = ∑ i , j a i b j θ i + j = ∑ k ( ∑ i , j a i b j C i j k ) θ k ab=\sum_{i,j}a_ib_j\theta^{i+j}=\sum_k\left(\sum_{i,j}a_ib_jC_{ij}^k\right)\theta^k ab = ∑ i , j a i b j θ i + j = ∑ k ( ∑ i , j a i b j C ij k ) θ k , and each coefficient is reduced modulo p p p by d p ( 0 , ⋅ ) d_p(0,\cdot) d p ( 0 , ⋅ ) . Finally a p = ∑ j a j θ p j a^p=\sum_j a_j\theta^{pj} a p = ∑ j a j θ p j because the Frobenius map is additive and fixes F p \mathbb{F}_p F p .
Remark 12.3 (Zech logarithms). If γ \gamma γ is a primitive element, every nonzero element is γ e \gamma^e γ e with e ∈ A q − 1 e\in A_{q-1} e ∈ A q − 1 , and multiplication is the cyclic law γ a γ b = γ d q − 1 ( 0 , a + b ) \gamma^a\gamma^b=\gamma^{d_{q-1}(0,a+b)} γ a γ b = γ d q − 1 ( 0 , a + b ) . Addition is recovered from the Zech logarithm Z Z Z , defined by 1 + γ t = γ Z ( t ) 1+\gamma^t=\gamma^{Z(t)} 1 + γ t = γ Z ( t ) whenever γ t ≠ − 1 \gamma^t\ne-1 γ t = − 1 :
γ a + γ b = γ d q − 1 ( 0 , a + Z ( d q − 1 ( 0 , b − a ) ) ) , \gamma^a+\gamma^b=\gamma^{d_{q-1}(0,a+Z(d_{q-1}(0,b-a)))}, γ a + γ b = γ d q − 1 ( 0 , a + Z ( d q − 1 ( 0 , b − a ))) ,
and γ a + γ b = 0 \gamma^a+\gamma^b=0 γ a + γ b = 0 when γ b − a = − 1 \gamma^{b-a}=-1 γ b − a = − 1 , that is, when b = a b=a b = a for p = 2 p=2 p = 2 and b − a ≡ ( q − 1 ) / 2 b-a\equiv(q-1)/2 b − a ≡ ( q − 1 ) /2 for odd p p p . The table of Z Z Z can be stored as a single code, as the action of C 3 C_3 C 3 was for Alt 4 \operatorname{Alt}_4 Alt 4 in
Section 9.1. Polynomial coordinates show the arithmetic; Zech coordinates turn multiplication into the cyclic law.
Example 12.4. For F 4 = F 2 [ θ ] / ( θ 2 + θ + 1 ) \mathbb{F}_4=\mathbb{F}_2[\theta]/(\theta^2+\theta+1) F 4 = F 2 [ θ ] / ( θ 2 + θ + 1 ) the labels are 0 , 1 , 2 = θ , 3 = 1 + θ 0,1,2=\theta,3=1+\theta 0 , 1 , 2 = θ , 3 = 1 + θ , addition is the bitwise exclusive-or, and θ 2 = 1 + θ \theta^2=1+\theta θ 2 = 1 + θ gives C 00 = ( 1 , 0 ) C_{00}=(1,0) C 00 = ( 1 , 0 ) , C 01 = C 10 = ( 0 , 1 ) C_{01}=C_{10}=(0,1) C 01 = C 10 = ( 0 , 1 ) , C 11 = ( 1 , 1 ) C_{11}=(1,1) C 11 = ( 1 , 1 ) . The nonzero labels multiply as 2 ⋅ 2 = 3 2\cdot2=3 2 ⋅ 2 = 3 , 2 ⋅ 3 = 1 2\cdot3=1 2 ⋅ 3 = 1 , 3 ⋅ 3 = 2 3\cdot3=2 3 ⋅ 3 = 2 ; with γ = θ \gamma=\theta γ = θ the powers γ 0 , γ 1 , γ 2 \gamma^0,\gamma^1,\gamma^2 γ 0 , γ 1 , γ 2 have labels 1 , 2 , 3 1,2,3 1 , 2 , 3 , and Z ( 1 ) = 2 Z(1)=2 Z ( 1 ) = 2 , Z ( 2 ) = 1 Z(2)=1 Z ( 2 ) = 1 . The table below lists two further cases (labels of γ 0 , … , γ q − 2 \gamma^0,\ldots,\gamma^{q-2} γ 0 , … , γ q − 2 , and Z ( 0 ) , … , Z ( q − 2 ) Z(0),\ldots,Z(q-2) Z ( 0 ) , … , Z ( q − 2 ) , with − - − where γ t = − 1 \gamma^t=-1 γ t = − 1 ).
The formulas of Proposition 12.2 and Remark 12.3 were checked by computer against the full addition and multiplication tables of F 4 \mathbb{F}_4 F 4 , F 8 \mathbb{F}_8 F 8 and F 9 \mathbb{F}_9 F 9 .
The groups P S L ( 2 , p ) PSL(2,p) P S L ( 2 , p )
Theorem 12.5 (P S L ( 2 , p ) PSL(2,p) P S L ( 2 , p ) ). Let p ≥ 3 p\geq3 p ≥ 3 be prime, h = ( p − 1 ) / 2 h=(p-1)/2 h = ( p − 1 ) /2 and N = p h ( p + 1 ) = ∣ P S L ( 2 , p ) ∣ N=ph(p+1)=|PSL(2,p)| N = p h ( p + 1 ) = ∣ P S L ( 2 , p ) ∣ , and write x − 1 = d p ( 0 , x p − 2 ) x^{-1}=d_p(0,x^{p-2}) x − 1 = d p ( 0 , x p − 2 ) for x ≠ 0 x\neq0 x = 0 . Decode m ∈ A N m\in\mathcal{A}_N m ∈ A N to a matrix M ( m ) = ( a b c d ) M(m)=\begin{pmatrix}a&b\\c&d\end{pmatrix} M ( m ) = ( a c b d ) over F p \mathbb{F}_p F p by
{ a = ⌊ m / p ⌋ + 1 , b = d p ( 0 , m ) , c = 0 , d = a − 1 , m < p h , a = d p ( 1 , m ′ ) , c = ⌊ m ′ / p 2 ⌋ + 1 , d = d p ( 0 , m ′ ) , b = d p ( 0 , ( a d − 1 ) c − 1 ) , m ′ = m − p h ≥ 0. \begin{cases}
a=\lfloor m/p\rfloor+1,\ b=d_p(0,m),\ c=0,\ d=a^{-1}, & m<ph,\\
a=d_p(1,m'),\ c=\lfloor m'/p^2\rfloor+1,\ d=d_p(0,m'),\ b=d_p(0,(ad-1)c^{-1}),\quad m'=m-ph\geq0.
\end{cases} { a = ⌊ m / p ⌋ + 1 , b = d p ( 0 , m ) , c = 0 , d = a − 1 , a = d p ( 1 , m ′ ) , c = ⌊ m ′ / p 2 ⌋ + 1 , d = d p ( 0 , m ′ ) , b = d p ( 0 , ( a d − 1 ) c − 1 ) , m ′ = m − p h ≥ 0. m < p h , To encode a matrix of determinant 1 1 1 , first replace it by its negative if c > h c>h c > h , or if c = 0 c=0 c = 0 and a > h a>h a > h , and then invert these formulas. Then the law
m ∗ n = code ( M ( m ) M ( n ) m o d p ) (89) m*n=\operatorname{code}(M(m)M(n)\bmod p) \tag*{(89)} m ∗ n = code ( M ( m ) M ( n ) mod p ) (89) makes A N \mathcal{A}_N A N a group isomorphic to P S L ( 2 , p ) PSL(2,p) P S L ( 2 , p ) , with identity 0 0 0 . For p ≥ 5 p\geq5 p ≥ 5 this group is nonabelian and simple.
Proof. Since P S L ( 2 , p ) = S L ( 2 , p ) / { ± 1 } PSL(2,p)=SL(2,p)/\{\pm1\} P S L ( 2 , p ) = S L ( 2 , p ) / { ± 1 } , each element is a pair { M , − M } \{M,-M\} { M , − M } . If c ≠ 0 c\neq0 c = 0 , exactly one of c , − c c,-c c , − c lies in { 1 , … , h } \{1,\ldots,h\} { 1 , … , h } ; if c = 0 c=0 c = 0 , then a ≠ 0 a\neq0 a = 0 and exactly one of a , − a a,-a a , − a lies there. So each element has a unique normalised representative. A normalised matrix with c = 0 c=0 c = 0 is determined by a ∈ { 1 , … , h } a\in\{1,\ldots,h\} a ∈ { 1 , … , h } and b ∈ A p b\in\mathcal{A}_p b ∈ A p , since d = a − 1 d=a^{-1} d = a − 1 . One with c ≠ 0 c\neq0 c = 0 is determined by a , d ∈ A p a,d\in\mathcal{A}_p a , d ∈ A p and c ∈ { 1 , … , h } c\in\{1,\ldots,h\} c ∈ { 1 , … , h } , since b = ( a d − 1 ) c − 1 b=(ad-1)c^{-1} b = ( a d − 1 ) c − 1 . This gives p h + p 2 h = N ph+p^2h=N p h + p 2 h = N codes, and the decoding formulas invert the encoding. Hence (89) is the multiplication of P S L ( 2 , p ) PSL(2,p) P S L ( 2 , p ) transported to A N \mathcal{A}_N A N , and M ( 0 ) M(0) M ( 0 ) is the identity matrix. Simplicity for p ≥ 5 p\geq5 p ≥ 5 is classical [19].
Remark 12.6 (Certification). GAP 4.14.0 [30] confirms these laws on several cases. For (89) , IdGroup returns [ 60 , 5 ] [60,5] [ 60 , 5 ] for p = 5 p=5 p = 5 and [ 168 , 42 ] [168,42] [ 168 , 42 ] for p = 7 p=7 p = 7 . For p = 11 , 13 , 17 p=11,13,17 p = 11 , 13 , 17 , the right multiplications by the codes of ( 1 1 0 1 ) \begin{pmatrix}1&1\\0&1\end{pmatrix} ( 1 0 1 1 ) and ( 1 0 1 0 ) \begin{pmatrix}1&0\\1&0\end{pmatrix} ( 1 1 0 0 ) generate a simple group of order N N N isomorphic to P S L ( 2 , p ) PSL(2,p) P S L ( 2 , p ) . In particular A 60 \mathcal{A}_{60} A 60 carries two different laws for A l t 5 ≅ P S L ( 2 , 5 ) Alt_5\cong PSL(2,5) A l t 5 ≅ P S L ( 2 , 5 ) : (89) here, and the transitive law of Corollary 11.5.
Remark 12.7 (The classification). By the classification of finite simple groups, every nonabelian finite simple group is alternating, of Lie type, or one of 26 sporadic groups. The alternating groups are covered by Section 11.9. For P S L ( 2 , q ) PSL(2,q) P S L ( 2 , q ) with q = p f q=p^f q = p f , the law (89) carries over with arithmetic in F q \mathbb{F}_q F q : elements of F q \mathbb{F}_q F q are f f f -digit words in base p p p , multiplied modulo an irreducible polynomial. For the other groups of Lie type, the Bruhat decomposition provides a normal form, so laws uniform in the parameters exist, although they are much longer. Theorem 8.32 also applies to each sporadic group through any of its transitive actions. This gives a law for each one, but no uniform formula for the sporadic groups as a family. The Mathieu groups are treated in Chapter 13.
The groups P S L n ( q ) PSL_n(q) P S L n ( q )
Let n ≥ 2 n \ge2 n ≥ 2 , let F q \mathbb{F}_q F q carry polynomial coordinates, and let d = gcd ( n , q − 1 ) d = \operatorname{gcd}(n,q-1) d = gcd ( n , q − 1 ) . The centre of S L n ( q ) SL_n(q) S L n ( q ) consists of the scalar matrices ζ I \zeta I ζ I with ζ ∈ μ d = { ζ ∈ F q : ζ n = 1 } \zeta\in\mu_d = \{\zeta\in\mathbb{F}_q : \zeta^n = 1\} ζ ∈ μ d = { ζ ∈ F q : ζ n = 1 } , and P S L n ( q ) = S L n ( q ) / μ d PSL_n(q) = SL_n(q)/\mu_d P S L n ( q ) = S L n ( q ) / μ d has order
N n , q = q n ( n − 1 ) / 2 d ∏ j = 2 n ( q j − 1 ) . N_{n,q} = \frac{q^{n(n-1)/2}}{d} \prod_{j=2}^{n}(q^j-1). N n , q = d q n ( n − 1 ) /2 j = 2 ∏ n ( q j − 1 ) .
Definition 12.8. For an n × n n \times n n × n matrix A A A over F q \mathbb{F}_q F q put κ ( A ) = ∑ i , j < n [ A i j − δ i j ] q i + n j \kappa(A) = \sum_{i,j<n}[A_{ij}-\delta_{ij}]q^{i+nj} κ ( A ) = ∑ i , j < n [ A ij − δ ij ] q i + nj , the digital word of A − I A-I A − I . The canonical representative A ^ \widehat{A} A of A ∈ S L n ( q ) A \in SL_n(q) A ∈ S L n ( q ) is the matrix ζ A \zeta A ζ A , ζ ∈ μ d \zeta\in\mu_d ζ ∈ μ d , with the smallest key. Let K = { κ ( A ^ ) : A ∈ S L n ( q ) } K = \{\kappa(\widehat{A}) : A \in SL_n(q)\} K = { κ ( A ) : A ∈ S L n ( q )} and rk ( A ) = # { k ∈ K : k < κ ( A ^ ) } \operatorname{rk}(A) = \#\{k \in K : k < \kappa(\widehat{A})\} rk ( A ) = # { k ∈ K : k < κ ( A )} .
Theorem 12.9 (Digital law of P S L n ( q ) PSL_n(q) P S L n ( q ) ). Let M m M_m M m be the canonical representative of rank m ∈ A N n , q m \in\mathcal{A}_{N_{n,q}} m ∈ A N n , q . Then
m ∗ n = rk ( M m M n ) , (90) m * n = \operatorname{rk}(M_m M_n), \tag*{(90)} m ∗ n = rk ( M m M n ) , (90) with the matrix product computed by Proposition 12.2 , makes A N n , q \mathcal{A}_{N_{n,q}} A N n , q a group isomorphic to P S L n ( q ) PSL_n(q) P S L n ( q ) , with identity 0 0 0 .
Proof. The key κ \kappa κ is injective, since it is the digital word of A − I A-I A − I . Each coset μ d A \mu_d A μ d A has d d d elements with distinct keys, so it has exactly one canonical representative, and rk \operatorname{rk} rk is a bijection from P S L n ( q ) PSL_n(q) P S L n ( q ) to A N n , q \mathcal{A}_{N_{n,q}} A N n , q . Since rk ( A B ) \operatorname{rk}(AB) rk ( A B ) depends only on the cosets of A A A and B B B , (90) is the multiplication of P S L n ( q ) PSL_n(q) P S L n ( q ) transported to A N n , q \mathcal{A}_{N_{n,q}} A N n , q . The identity has key κ ( I ) = 0 \kappa(I)=0 κ ( I ) = 0 , the smallest possible, so its rank is 0 0 0 .
Example 12.10. For P S L 3 ( 2 ) PSL_3(2) P S L 3 ( 2 ) , of order 168, (90) was checked by computer on all triples: it is associative, and its conjugacy classes have sizes 1, 21, 24, 24, 42, 56, those of the simple group of order 168. For P S L 3 ( 3 ) PSL_3(3) P S L 3 ( 3 ) (order 5616, d = 1 d=1 d = 1 ) and P S L 3 ( 4 ) PSL_3(4) P S L 3 ( 4 ) (order 20160, d = 3 d=3 d = 3 , over the non-prime field of Example 12.4) the number of canonical representatives equals N n , q N_{n,q} N n , q , the identity has label 0, and associativity holds on 3000 random triples. Apart from P S L 2 ( 2 ) PSL_2(2) P S L 2 ( 2 ) and P S L 2 ( 3 ) PSL_2(3) P S L 2 ( 3 ) , all the groups P S L n ( q ) PSL_n(q) P S L n ( q ) are simple.
Remark 12.11 (Towards Bruhat coordinates). The rank label is universal, but decoding it requires the ordered set K K K . Dense coordinates come from the Bruhat decomposition S L n ( q ) = ⋃ w ∈ Sym n B w B SL_n(q) = \bigcup_{w \in\operatorname{Sym}_n} BwB S L n ( q ) = ⋃ w ∈ Sym n B w B , with B B B the upper triangular matrices of determinant 1, where the cell of w w w has ∣ B ∣ q ℓ ( w ) |B|q^{\ell(w)} ∣ B ∣ q ℓ ( w ) elements and ℓ ( w ) \ell(w) ℓ ( w ) is the number of inversions of w w w . Hence
N n , q = ∣ T ∣ ∣ U ∣ ∑ w ∈ Sym n q ℓ ( w ) , ∣ U ∣ = q n ( n − 1 ) / 2 , ∣ T ∣ = ( q − 1 ) n − 1 d ; N_{n,q} = |T|\,|U| \sum_{w \in\operatorname{Sym}_n} q^{\ell(w)}, \qquad|U| = q^{n(n-1)/2}, \qquad|T| = \frac{(q-1)^{n-1}}{d}; N n , q = ∣ T ∣ ∣ U ∣ w ∈ Sym n ∑ q ℓ ( w ) , ∣ U ∣ = q n ( n − 1 ) /2 , ∣ T ∣ = d ( q − 1 ) n − 1 ;
for n = 3 n=3 n = 3 the sum is 1 + 2 q + 2 q 2 + q 3 1+2q+2q^2+q^3 1 + 2 q + 2 q 2 + q 3 , which gives 168, 5616 and 20160 for q = 2 , 3 , 4 q=2,3,4 q = 2 , 3 , 4 . Inside a cell an element is described by w w w , labelled by the rotation normal form of Lemma 11.51, by n − 1 n-1 n − 1 torus digits, and by ℓ ( w ) + n ( n − 1 ) / 2 \ell(w)+n(n-1)/2 ℓ ( w ) + n ( n − 1 ) /2 field digits, and Gaussian elimination reads these off from the pivots. For n = 2 n=2 n = 2 , Theorem 12.5 already gives such a dense labelling. Theorems 12.12 and 12.13 give the general case.
Theorem 12.12 (Bruhat coordinates). For w ∈ Sym n w \in\operatorname{Sym}_n w ∈ Sym n let w ˙ \dot{w} w ˙ be the permutation matrix with w ˙ w ( j ) , j = 1 \dot{w}_{w(j),j}=1 w ˙ w ( j ) , j = 1 , and let U w U_w U w be the set of upper unitriangular matrices whose off-diagonal entries vanish outside the inversion positions ( i , i ′ ) (i,i') ( i , i ′ ) , i < i ′ i<i' i < i ′ , w − 1 ( i ) > w − 1 ( i ′ ) w^{-1}(i)>w^{-1}(i') w − 1 ( i ) > w − 1 ( i ′ ) . Every g ∈ S L n ( q ) g \in SL_n(q) g ∈ S L n ( q ) has a unique expression
g = u w ˙ t v , u ∈ U w , t diagonal with det t = sgn w , v upper unitriangular . (91) g = u\dot{w}tv,\qquad u \in U_w,\quad t\text{ diagonal with }\det t=\operatorname{sgn}w,\quad v\text{ upper unitriangular}. \tag*{(91)} g = u w ˙ t v , u ∈ U w , t diagonal with det t = sgn w , v upper unitriangular . (91) The permutation w w w and X = w ˙ t v X=\dot{w}tv X = w ˙ t v are found by elimination: for j = 0 , 1 , … , n − 1 j=0,1,\ldots,n-1 j = 0 , 1 , … , n − 1 , w ( j ) w(j) w ( j ) is the largest row index not among w ( 0 ) , … , w ( j − 1 ) w(0),\ldots,w(j-1) w ( 0 ) , … , w ( j − 1 ) with X w ( j ) , j ≠ 0 X_{w(j),j}\ne0 X w ( j ) , j = 0 , after which, for every
such unused row i < w ( j ) i < w(j) i < w ( j ) , the multiple X i j X w ( j ) , j − 1 X_{ij}X^{-1}_{w(j),j} X ij X w ( j ) , j − 1 of row w ( j ) w(j) w ( j ) is subtracted from row i i i . Then u = g X − 1 u = gX^{-1} u = g X − 1 , t i i = X w ( i ) , i t_{ii} = X_{w(i),i} t ii = X w ( i ) , i and v = t − 1 w ˙ − 1 X v = t^{-1}\dot{w}^{-1}X v = t − 1 w ˙ − 1 X .
Proof. Uniqueness of w w w . Left multiplication by an invertible upper triangular matrix replaces each row by a combination of itself and the rows below it, and right multiplication replaces each column by a combination of itself and the columns to its left. Hence the ranks r i j ( g ) r_{ij}(g) r ij ( g ) of the submatrices formed by the rows ≥ i \ge i ≥ i and the columns ≤ j \le j ≤ j are unchanged. For w ˙ \dot{w} w ˙ they are r i j = # { j ′ ≤ j : w ( j ′ ) ≥ i } r_{ij} = \#\{j' \le j : w(j') \ge i\} r ij = # { j ′ ≤ j : w ( j ′ ) ≥ i } , which determine w w w . Since u u u and t v tv t v are upper triangular, w w w is determined by g g g .
Uniqueness of the rest. U w = U ∩ w ˙ U − w ˙ − 1 U_w = U \cap\dot{w}U^-\dot{w}^{-1} U w = U ∩ w ˙ U − w ˙ − 1 , with U − U^- U − the lower unitriangular matrices, is a group. If u w ˙ b = u ′ w ˙ b ′ u\dot{w}b = u'\dot{w}b' u w ˙ b = u ′ w ˙ b ′ with b , b ′ b,b' b , b ′ upper triangular, then w ˙ − 1 ( u ′ − 1 u ) w ˙ = b ′ b − 1 \dot{w}^{-1}(u'^{-1}u)\dot{w} = b'b^{-1} w ˙ − 1 ( u ′ − 1 u ) w ˙ = b ′ b − 1 is both lower unitriangular and upper triangular, hence I I I . So u = u ′ u = u' u = u ′ and b = b ′ b = b' b = b ′ , and b = t v b = tv b = t v is unique.
Existence. At step j j j the unused rows below w ( j ) w(j) w ( j ) are zero in column j j j and those above it are cleared, while the used rows w ( k ) w(k) w ( k ) , k < j k < j k < j , are left alone. A row cleared at step j j j is w ( k ) w(k) w ( k ) with k > j k > j k > j and so is zero in column j j j ; adding multiples of it later does not change column j j j . At the end, column j j j of X X X is nonzero at row w ( j ) w(j) w ( j ) and at rows w ( k ) w(k) w ( k ) , k < j k < j k < j , only, so X = w ˙ b X = \dot{w}b X = w ˙ b with b b b upper triangular. Each operation adds a multiple of row w ( j ) w(j) w ( j ) to a row w ( k ) w(k) w ( k ) above it with k > j k > j k > j , that is, at an inversion position, so X = L g X = Lg X = Lg with L ∈ U w L \in U_w L ∈ U w , and u = L − 1 ∈ U w u = L^{-1} \in U_w u = L − 1 ∈ U w . Finally det b = det g / det w ˙ = sgn w \det b = \det g/\det\dot{w} = \operatorname{sgn} w det b = det g / det w ˙ = sgn w .
Theorem 12.13 (Dense digital law of P S L n ( q ) PSL_n(q) P S L n ( q ) ). Let γ \gamma γ be a primitive element of F q \mathbb{F}_q F q , h = ( q − 1 ) / d h = (q - 1)/d h = ( q − 1 ) / d , and order Sym n \operatorname{Sym}_n Sym n as w 0 , … , w n ! − 1 w_0,\ldots,w_{n!-1} w 0 , … , w n ! − 1 by the rotation label of Lemma 11.51. Put
O k = h ( q − 1 ) n − 2 q n ( n − 1 ) / 2 ∑ k ′ < k q ℓ ( w k ′ ) . O_k = h(q-1)^{n-2}q^{n(n-1)/2}\sum_{k'<k}q^{\ell(w_{k'})}. O k = h ( q − 1 ) n − 2 q n ( n − 1 ) /2 k ′ < k ∑ q ℓ ( w k ′ ) . For g g g with coordinates (91) , w = w k w = w_k w = w k , write t s s = γ e s t_{ss} = \gamma^{e_s} t ss = γ e s for s < n − 1 s < n - 1 s < n − 1 , normalised by e s ↦ d q − 1 ( 0 , e s − h ⌊ e 0 / h ⌋ ) e_s \mapsto d_{q-1}(0,e_s-h\lfloor e_0/h\rfloor) e s ↦ d q − 1 ( 0 , e s − h ⌊ e 0 / h ⌋) so that e 0 ∈ A h e_0 \in A_h e 0 ∈ A h . Let V V V and U w U_w U w be the base-q q q words of the labels of the entries of v v v above the diagonal and of u u u at the inversion positions, each listed column by column. Then
λ ( g ) = O k + e 0 + h ( ∑ s = 1 n − 2 e s ( q − 1 ) s − 1 + ( q − 1 ) n − 2 ( V + q n ( n − 1 ) / 2 U w ) ) (92) \lambda(g) = O_k + e_0 + h\left(\sum_{s=1}^{n-2}e_s(q-1)^{s-1} + (q-1)^{n-2}\left(V+q^{n(n-1)/2}U_w\right)\right) \tag*{(92)} λ ( g ) = O k + e 0 + h ( s = 1 ∑ n − 2 e s ( q − 1 ) s − 1 + ( q − 1 ) n − 2 ( V + q n ( n − 1 ) /2 U w ) ) (92) is constant on the cosets of μ d \mu_d μ d and is a bijection P S L n ( q ) → A N n , q PSL_n(q) \to A_{N_{n,q}} P S L n ( q ) → A N n , q with λ ( I ) = 0 \lambda(I)=0 λ ( I ) = 0 . Hence
m ∗ n = λ ( λ − 1 ( m ) λ − 1 ( n ) ) (93) m \ast n = \lambda\left(\lambda^{-1}(m)\lambda^{-1}(n)\right) \tag*{(93)} m ∗ n = λ ( λ − 1 ( m ) λ − 1 ( n ) ) (93) makes A N n , q A_{N_{n,q}} A N n , q a group isomorphic to P S L n ( q ) PSL_n(q) P S L n ( q ) . Decoding m m m is digital: the cell index is k = ∑ k ′ = 1 n ! − 1 ⌊ ( m − O k ′ + N n , q ) / N n , q ⌋ k = \sum_{k'=1}^{n!-1}\lfloor(m-O_{k'}+N_{n,q})/N_{n,q}\rfloor k = ∑ k ′ = 1 n ! − 1 ⌊( m − O k ′ + N n , q ) / N n , q ⌋ , and the coordinates are the mixed-radix digits of m − O k m-O_k m − O k .
Proof. A central scalar ζ = γ c h \zeta=\gamma^{ch} ζ = γ c h changes only t t t , adding c h ch c h to every exponent; so each coset has exactly one representative with e 0 ∈ A h e_0\in A_h e 0 ∈ A h , and λ \lambda λ is constant on cosets. In the cell of w w w the coordinates range freely over h ( q − 1 ) n − 2 q ℓ ( w ) h(q-1)^{n-2}q^{\ell(w)} h ( q − 1 ) n − 2 q ℓ ( w ) values (the last diagonal entry of t t t being fixed by det t = sgn w \det t=\operatorname{sgn}w det t = sgn w ), and by Lemma 1.6 the bracket in (92) is a bijection onto [ 0 , O k + 1 − O k ) [0,O_{k+1}-O_k) [ 0 , O k + 1 − O k ) . By Theorem 12.12 the cells partition S L n ( q ) SL_n(q) S L n ( q ) , so the offsets give a bijection onto A N n , q A_{N_{n,q}} A N n , q , consistent with Remark 12.11. Each term of the sum for k k k is 1 1 1 exactly when m ≥ O k ′ m\ge O_{k'} m ≥ O k ′ . The identity lies in the cell of w 0 = id w_0=\operatorname{id} w 0 = id , with u = t = v = I u=t=v=I u = t = v = I , so λ ( I ) = 0 \lambda(I)=0 λ ( I ) = 0 .
Example 12.14. For P S L 3 ( 3 ) PSL_3(3) P S L 3 ( 3 ) the cells, with w w w written as w ( 0 ) w ( 1 ) w ( 2 ) w(0)w(1)w(2) w ( 0 ) w ( 1 ) w ( 2 ) , are
Table 12.
The law (93) was checked by computer for P S L 2 ( 5 ) ≅ A l t 5 PSL_2(5) \cong\mathrm{Alt}_5 P S L 2 ( 5 ) ≅ Alt 5 , P S L 2 ( 9 ) ≅ A l t 6 PSL_2(9) \cong\mathrm{Alt}_6 P S L 2 ( 9 ) ≅ Alt 6 , P S L 3 ( 2 ) PSL_3(2) P S L 3 ( 2 ) , P S L 3 ( 3 ) PSL_3(3) P S L 3 ( 3 ) , P S L 3 ( 4 ) PSL_3(4) P S L 3 ( 4 ) and P S L 4 ( 2 ) PSL_4(2) P S L 4 ( 2 ) . Encoding after decoding returns every label (on 5000 random labels for P S L 4 ( 2 ) PSL_4(2) P S L 4 ( 2 ) ), every decoded matrix has determinant 1, central scalars do not change the label, 0 is the identity, and associativity holds on 1000 random triples. P S L 3 ( 4 ) PSL_3(4) P S L 3 ( 4 ) and P S L 4 ( 2 ) PSL_4(2) P S L 4 ( 2 ) have the same order 20160 but are not isomorphic.
The other classical families
The symplectic, orthogonal and unitary groups are the isometry groups of forms. Written in a basis in which the form is antidiagonal, their upper triangular elements form a Borel subgroup, and the elimination of Theorem 12.12 applies to them without change. What changes is which permutations occur.
Definition 12.16 (Standard forms). Let r ( i ) = N − 1 − i r(i) = N - 1 - i r ( i ) = N − 1 − i on A N A_N A N , and let J J J be the N × N N \times N N × N matrix with J i , r ( i ) = ϵ i J_{i,r(i)} = \epsilon_i J i , r ( i ) = ϵ i and all other entries 0.
S p 2 n ( q ) Sp_{2n}(q) S p 2 n ( q ) : N = 2 n N = 2n N = 2 n , ϵ i = 1 \epsilon_i = 1 ϵ i = 1 for i < n i < n i < n and ϵ i = − 1 \epsilon_i = -1 ϵ i = − 1 for i ≥ n i \ge n i ≥ n , and G = { g : g T J g = J } G = \{g : g^{\mathsf T} Jg = J\} G = { g : g T J g = J } .
Ω 2 n + 1 ( q ) \Omega_{2n+1}(q) Ω 2 n + 1 ( q ) and Ω 2 n + ( q ) \Omega^+_{2n}(q) Ω 2 n + ( q ) , q q q odd: all ϵ i = 1 \epsilon_i = 1 ϵ i = 1 , and G G G is the group generated by the upper and lower unitriangular matrices with g T J g = J g^{\mathsf T} Jg = J g T J g = J .
Ω 2 n + ( q ) \Omega^+_{2n}(q) Ω 2 n + ( q ) , q q q even: all ϵ i = 1 \epsilon_i = 1 ϵ i = 1 , Q ( x ) = ∑ i < n x i x r ( i ) Q(x) = \sum_{i<n} x_i x_{r(i)} Q ( x ) = ∑ i < n x i x r ( i ) , whose polar form has matrix J J J , and G G G is the group generated by the upper and lower unitriangular matrices g g g with Q ( g x ) = Q ( x ) Q(gx) = Q(x) Q ( g x ) = Q ( x ) for all x x x .
S U N ( q ) SU_N(q) S U N ( q ) : matrices over F q 2 \mathbb{F}_{q^2} F q 2 , all ϵ i = 1 \epsilon_i = 1 ϵ i = 1 , x ‾ = x q \overline{x} = x^q x = x q , and G = { g : det g = 1 , g T J g ‾ = J } G = \{g : \det g = 1,\ g^{\mathsf T} J\overline{g} = J\} G = { g : det g = 1 , g T J g = J } .
Put n = ⌊ N / 2 ⌋ n = \lfloor N/2 \rfloor n = ⌊ N /2 ⌋ . A permutation w w w commutes with r r r if w ( r ( j ) ) = r ( w ( j ) ) w(r(j)) = r(w(j)) w ( r ( j )) = r ( w ( j )) for all j j j ; its number of sign changes is s ( w ) = # { i < n : w ( i ) ≥ N − n } s(w) = \#\{i < n : w(i) \ge N - n\} s ( w ) = # { i < n : w ( i ) ≥ N − n } .
Theorem 12.17 (Bruhat cells of the classical groups). Let G G G be one of the groups of Definition 12.16 , B G B_G B G its upper triangular elements and U G U_G U G its upper unitriangular elements. For g ∈ G g \in G g ∈ G , write g = u w ˙ t v g = u\dot{w}tv g = u w ˙ t v as in Theorem 12.12 . Then w w w commutes with r r r and u ∈ G ∩ U w u \in G \cap U_w u ∈ G ∩ U w . The permutations that occur form the Weyl group W G W_G W G :
for S p 2 n Sp_{2n} S p 2 n , Ω 2 n + 1 \Omega_{2n+1} Ω 2 n + 1 and S U N SU_N S U N , all permutations commuting with r r r (the signed permutations of A n A_n A n , 2 n n ! 2^n n! 2 n n ! of them);
for Ω 2 n + \Omega^+_{2n} Ω 2 n + , those with an even number of sign changes (2 n − 1 n ! 2^{n-1}n! 2 n − 1 n ! of them).
The cell of w w w is ( G ∩ U w ) × ( G ∩ w B ) , (G \cap U_w) \times(G \cap wB), ( G ∩ U w ) × ( G ∩ w B ) , where B B B is the group of upper triangular matrices, and has ∣ G ∩ U w ∣ ∣ B G ∣ |G \cap U_w||B_G| ∣ G ∩ U w ∣∣ B G ∣ elements.
Proof. These groups have a ( B , N ) (B,N) ( B , N ) -pair with B = B G B = B_G B = B G and N N N the monomial matrices of G G G [21, 22], and the refined Bruhat decomposition writes g = u ′ n b ′ g = u'n b' g = u ′ n b ′ uniquely with b ′ ∈ B G b' \in B_G b ′ ∈ B G , n ∈ N n \in N n ∈ N monomial and u ′ ∈ U G ∩ n U G n − 1 u' \in U_G \cap nU_Gn^{-1} u ′ ∈ U G ∩ n U G n − 1 . Write n = w ˙ δ n = \dot{w}\delta n = w ˙ δ with δ \delta δ diagonal. Conjugation by δ \delta δ preserves lower unitriangularity, so w ˙ − 1 u ′ w ˙ \dot{w}^{-1}u'\dot{w} w ˙ − 1 u ′ w ˙ is lower unitriangular, that is, u ′ ∈ U w u' \in U_w u ′ ∈ U w . Thus g = u ′ w ˙ ( δ b ′ ) g = u'\dot{w}(\delta b') g = u ′ w ˙ ( δ b ′ ) is a factorisation as in Theorem 12.12 (over F q 2 \mathbb{F}_{q^2} F q 2 for S U N SU_N S U N ), and by uniqueness u = u ′ ∈ G u = u' \in G u = u ′ ∈ G and w w w is the permutation of n n n . A monomial isometry of an antidiagonal form maps each pair of basis vectors { e j , e r ( j ) } \{e_j,e_{r(j)}\} { e j , e r ( j ) } to such a pair, so w w w commutes with r r r . The description of W G W_G W G is that of the Weyl groups of types B n B_n B n , C n C_n C n , D n D_n D n and of the unitary groups [21]. Finally x = u − 1 g ∈ G ∩ w ˙ B x=u^{-1}g\in G\cap\dot wB x = u − 1 g ∈ G ∩ w ˙ B , and G ∩ w ˙ B = n B G G\cap\dot wB=nB_G G ∩ w ˙ B = n B G has ∣ B G ∣ |B_G| ∣ B G ∣ elements.
Theorem 12.18 (Digital law of a classical group). Let G G G and W G W_G W G be as above, let Z Z Z be the group of scalar matrices in G G G , and order W G W_G W G as w 0 = i d , w 1 , … w_0=\mathrm{id},w_1,\ldots w 0 = id , w 1 , … by the rotation label of Lemma 11.51. Fix bijections ρ w : G ∩ U w → A ∣ G ∩ U w ∣ \rho_w:G\cap U_w\to A_{\lvert G\cap U_w\rvert} ρ w : G ∩ U w → A ∣ G ∩ U w ∣ and β w : ( G ∩ w ˙ B ) / Z → A ∣ B G ∣ / ∣ Z ∣ \beta_w:(G\cap\dot wB)/Z\to A_{\lvert B_G\rvert/\lvert Z\rvert} β w : ( G ∩ w ˙ B ) / Z → A ∣ B G ∣ / ∣ Z ∣ sending I I I , respectively the coset of n n n with w = i d w=\mathrm{id} w = id , to 0 0 0 . With O k = ∣ B G ∣ ∣ Z ∣ ∑ k ′ < k ∣ G ∩ U w k ′ ∣ O_k=\frac{\lvert B_G\rvert}{\lvert Z\rvert}\sum_{k'<k}\lvert G\cap U_{w_{k'}}\rvert O k = ∣ Z ∣ ∣ B G ∣ ∑ k ′ < k ∣ G ∩ U w k ′ ∣ , the label
λ ( g ) = O k + ρ w k ( u ) + ∣ G ∩ U w k ∣ β w k ( Z u − 1 g ) ( w = w k ) \lambda(g)=O_k+\rho_{w_k}(u)+\lvert G\cap U_{w_k}\rvert\beta_{w_k}(Zu^{-1}g)\qquad(w=w_k) λ ( g ) = O k + ρ w k ( u ) + ∣ G ∩ U w k ∣ β w k ( Z u − 1 g ) ( w = w k ) is a bijection G / Z → A ∣ G / Z ∣ G/Z\to A_{\lvert G/Z\rvert} G / Z → A ∣ G / Z ∣ with λ ( I ) = 0 \lambda(I)=0 λ ( I ) = 0 , and m ∗ n = λ ( λ − 1 ( m ) λ − 1 ( n ) ) m*n=\lambda(\lambda^{-1}(m)\lambda^{-1}(n)) m ∗ n = λ ( λ − 1 ( m ) λ − 1 ( n )) is a digital law of G / Z G/Z G / Z .
Proof. Scalars commute with everything and lie in B G B_G B G , so multiplying g g g by z ∈ Z z\in Z z ∈ Z leaves w w w and u u u unchanged and multiplies u − 1 g u^{-1}g u − 1 g by z z z . Hence λ \lambda λ is well defined on G / Z G/Z G / Z . By Theorem 12.17 each cell is the product of its two factors, and the offsets enumerate the cells, so λ \lambda λ is a bijection. The identity lies in the cell of w 0 w_0 w 0 with u = I u=I u = I .
The tables behind ρ w \rho_w ρ w and β w \beta_w β w have at most ∣ U G ∣ \lvert U_G\rvert ∣ U G ∣ and ∣ B G ∣ \lvert B_G\rvert ∣ B G ∣ entries, far fewer than ∣ G ∣ \lvert G\rvert ∣ G ∣ ; they are the digital analogue of the root coordinates of U G U_G U G and of the torus coordinates of B G B_G B G .
Example 12.19. For the groups below, U G U_G U G was found by an exhaustive search over upper unitriangular matrices, G G G was generated by elements of U G U_G U G and their images under g ↦ ( g r ( i ) , r ( j ) ) g\mapsto(g_{r(i),r(j)}) g ↦ ( g r ( i ) , r ( j ) ) , and every element was put through the elimination. In each case the order is the expected one, every w w w commutes with r r r , every u u u lies in G G G , and the cells have the sizes ∣ G ∩ U w ∣ ∣ B G ∣ \lvert G\cap U_w\rvert\lvert B_G\rvert ∣ G ∩ U w ∣ ∣ B G ∣ listed. An entry v × k v^{\times k} v × k records k k k cells of the same size v v v .
Table 12.19.
For Ω 4 + ( 3 ) \Omega_4^+(3) Ω 4 + ( 3 ) exactly the four permutations with s ( w ) ∈ { 0 , 2 } s(w)\in\{0,2\} s ( w ) ∈ { 0 , 2 } occur. For S U N SU_N S U N the factors q q q and q 2 q^2 q 2 mix, because the root groups of a unitary group have orders q q q , q 2 q^2 q 2 or q 3 q^3 q 3 .
Remark 12.20. The simple groups P S p 4 ( 3 ) = S p 4 ( 3 ) / { ± I } PSp_4(3)=Sp_4(3)/\{\pm I\} P S p 4 ( 3 ) = S p 4 ( 3 ) / { ± I } , Ω 5 ( 3 ) \Omega_5(3) Ω 5 ( 3 ) and P S U 4 ( 2 ) = S U 4 ( 2 ) PSU_4(2)=SU_4(2) P S U 4 ( 2 ) = S U 4 ( 2 ) all have order 25920 and are isomorphic, so Theorem 12.18 gives three different digital laws of one group, from forms of three different types. S U 3 ( 3 ) SU_3(3) S U 3 ( 3 ) is simple of order 6048, while S p 4 ( 2 ) ≅ Sym 6 Sp_4(2)\cong\operatorname{Sym}_6 S p 4 ( 2 ) ≅ Sym 6 and Ω 4 + ( 3 ) \Omega_4^+(3) Ω 4 + ( 3 ) are not simple. For q q q even, Ω 2 n + ( q ) \Omega_{2n}^+(q) Ω 2 n + ( q ) is defined through the quadratic form Q Q Q of Definition 12.16 , since J J J alone only determines the larger group S p 2 n ( q ) Sp_{2n}(q) S p 2 n ( q ) . The rows for q = 2 , 4 q=2,4 q = 2 , 4 in Example 12.19 are of this kind: Ω 6 + ( 2 ) \Omega_6^+(2) Ω 6 + ( 2 ) is isomorphic to the alternating group of degree 8 and is simple, while Ω 4 + ( q ) ≅ S L 2 ( q ) × S L 2 ( q ) \Omega_4^+(q)\cong SL_2(q)\times SL_2(q) Ω 4 + ( q ) ≅ S L 2 ( q ) × S L 2 ( q ) is not. In odd dimension and characteristic 2 the orthogonal groups are isomorphic to S p 2 n ( q ) Sp_{2n}(q) S p 2 n ( q ) , so no further case is needed. The groups Ω 2 n − ( q ) \Omega_{2n}^-(q) Ω 2 n − ( q ) of type 2 D n {}^2D_n 2 D n are treated in Section 12.6.
The twisted orthogonal groups
Let q q q be odd. The orthogonal space of minus type in dimension 2 n 2n 2 n is H 1 ⊥ ⋯ ⊥ H n − 1 ⊥ A H_1 \perp\cdots\perp H_{n-1} \perp A H 1 ⊥ ⋯ ⊥ H n − 1 ⊥ A , with hyperbolic planes H i H_i H i and an anisotropic plane A A A . Identifying A A A with F q 2 \mathbb{F}_{q^2} F q 2 , its quadratic form is the norm z z ˉ z\bar{z} z z ˉ , z ˉ = z q \bar{z}=z^q z ˉ = z q . Over F q 2 \mathbb{F}_{q^2} F q 2 the plane A A A splits, so Ω 2 n − ( q ) \Omega^-_{2n}(q) Ω 2 n − ( q ) sits inside Ω 2 n + ( q 2 ) \Omega^+_{2n}(q^2) Ω 2 n + ( q 2 ) as the fixed points of a twisted Frobenius map, just as S U N ( q ) SU_N(q) S U N ( q ) sits inside S L N ( q 2 ) SL_N(q^2) S L N ( q 2 ) . This lets the elimination of Theorem 12.12 be used over F q 2 \mathbb{F}_{q^2} F q 2 without change.
Definition 12.21 (Twisted standard form). Let N = 2 n N=2n N = 2 n , let J J J be the antidiagonal matrix with all ϵ i = 1 \epsilon_i=1 ϵ i = 1 , let τ \tau τ be the transposition of n − 1 n-1 n − 1 and n n n , and put F ( g ) = ( g τ ( i ) , τ ( j ) ) i , j F(g)=(g_{\tau(i),\tau(j)})_{i,j} F ( g ) = ( g τ ( i ) , τ ( j ) ) i , j for matrices over F q 2 \mathbb{F}_{q^2} F q 2 . Let G − = { g ∈ S L N ( q 2 ) : g T J g = J , F ( g ) = g } G^-=\{g\in SL_N(q^2):g^T Jg=J,\ F(g)=g\} G − = { g ∈ S L N ( q 2 ) : g T J g = J , F ( g ) = g } , and let Ω 2 n − ( q ) \Omega^-_{2n}(q) Ω 2 n − ( q ) be the subgroup generated by the upper and lower unitriangular elements of G − G^- G − .
Lemma 12.22. The vectors x ∈ F q 2 N x\in\mathbb{F}_{q^2}^N x ∈ F q 2 N with x τ ( i ) = x ˉ i x_{\tau(i)}=\bar{x}_i x τ ( i ) = x ˉ i form a 2 n 2n 2 n -dimensional F q \mathbb{F}_q F q -space V F V^F V F , on which Q ( x ) = 1 2 x T J x Q(x)=\frac{1}{2}x^T Jx Q ( x ) = 2 1 x T J x takes the values
Q ( x ) = ∑ i < n − 1 x i x r ( i ) + x n − 1 x ‾ n − 1 . Q(x)=\sum_{i<n-1}x_i x_{r(i)}+x_{n-1}\overline{x}_{n-1}. Q ( x ) = i < n − 1 ∑ x i x r ( i ) + x n − 1 x n − 1 .
This is a quadratic space of minus type, and G − G^- G − acts on it as S O 2 n − ( q ) SO^-_{2n}(q) S O 2 n − ( q ) .
Proof. The coordinates x i x_i x i with i ∉ { n − 1 , n } i\notin\{n-1,n\} i ∈ / { n − 1 , n } lie in F q \mathbb{F}_q F q , while x n − 1 ∈ F q 2 x_{n-1}\in\mathbb{F}_{q^2} x n − 1 ∈ F q 2 is free and x n = x ‾ n − 1 x_n=\overline{x}_{n-1} x n = x n − 1 . The outer coordinates give n − 1 n-1 n − 1 hyperbolic planes and the middle pair the norm form, which is anisotropic. A matrix with F ( g ) = g F(g)=g F ( g ) = g commutes with the semilinear map x ↦ ( x τ ( i ) ‾ ) i x\mapsto(\overline{x_{\tau(i)}})_i x ↦ ( x τ ( i ) ) i , so it preserves V F V^F V F . Since V F V^F V F spans F q 2 N \mathbb{F}_{q^2}^N F q 2 N , isometries of V F V^F V F of determinant 1 correspond exactly to the elements of G − G^- G − .
The root elements of Ω N + \Omega^+_N Ω N + are x i j ( c ) = I + c E i j − c E r ( j ) , r ( i ) x_{ij}(c)=I+cE_{ij}-cE_{r(j),r(i)} x ij ( c ) = I + c E ij − c E r ( j ) , r ( i ) for i < j i<j i < j , j ≠ r ( i ) j\ne r(i) j = r ( i ) . Those fixed by F F F are x i j ( c ) x_{ij}(c) x ij ( c ) with c ∈ F q c\in\mathbb{F}_q c ∈ F q and i , j ∉ { n − 1 , n } i,j\notin\{n-1,n\} i , j ∈ / { n − 1 , n } , and the folded root elements
y i ( c ) = x i , n − 1 ( c ) x i , n ( c ˉ ) , c ∈ F q 2 , i < n − 1 , y_i(c)=x_{i,n-1}(c)x_{i,n}(\bar{c}),\qquad c\in\mathbb{F}_{q^2},\ i<n-1, y i ( c ) = x i , n − 1 ( c ) x i , n ( c ˉ ) , c ∈ F q 2 , i < n − 1 ,
which are fixed by F F F because c c ˉ ∈ F q c\bar{c}\in\mathbb{F}_q c c ˉ ∈ F q . A folded root group has q 2 q^2 q 2 elements and contributes one F q 2 \mathbb{F}_{q^2} F q 2 -digit.
Theorem 12.23 (Bruhat cells of Ω 2 n − ( q ) \Omega^-_{2n}(q) Ω 2 n − ( q ) ). Let g ∈ Ω 2 n − ( q ) g\in\Omega^-_{2n}(q) g ∈ Ω 2 n − ( q ) and write g = u w t v ˙ g=uwtv\dot{\ } g = u w t v ˙ as in Theorem 12.12 , over F q 2 \mathbb{F}_{q^2} F q 2 . Then w w w commutes with r r r and τ \tau τ and has an even number of sign changes, u ∈ Ω 2 n − ( q ) ∩ U w u\in\Omega^-_{2n}(q)\cap U_w u ∈ Ω 2 n − ( q ) ∩ U w , and F ( t v ) = t v F(tv)=tv F ( t v ) = t v . The outer diagonal entries of t t t lie in F q × \mathbb{F}_q^\times F q × , and the middle ones are z , z ˉ z,\bar{z} z , z ˉ with z z ˉ = 1 z\bar{z}=1 z z ˉ = 1 . The Weyl group W G W_G W G of such permutations is determined by the action on the n − 1 n-1 n − 1 outer pairs, so W G ≅ W ( B n − 1 ) W_G\cong W(B_{n-1}) W G ≅ W ( B n − 1 ) , of order 2 n − 1 ( n − 1 ) ! 2^{n-1}(n-1)! 2 n − 1 ( n − 1 )! : the middle pair is swapped exactly when the outer pairs have an odd number of sign changes.
Proof. The generators of Ω 2 n − ( q ) \Omega^-_{2n}(q) Ω 2 n − ( q ) are unipotent elements of Ω N + ( q 2 ) \Omega^+_N(q^2) Ω N + ( q 2 ) , so Theorem 12.17 applies over F q 2 \mathbb{F}_{q^2} F q 2 : w w w commutes with r r r , has an even number of sign changes, and u ∈ U w u\in U_w u ∈ U w is an isometry. An upper triangular isometry has middle block ( 0 d c 0 ) \left(\begin{smallmatrix}0&d\\ c&0\end{smallmatrix}\right) ( 0 c d 0 ) , and the condition on this block gives 2 c d = 0 2cd=0 2 c d = 0 , so c = 0 c=0 c = 0 because q q q is odd. Hence F F F maps upper triangular isometries to upper triangular isometries. The matrix w w w is an isometry, so t v = w − 1 u − 1 g tv=w^{-1}u^{-1}g t v = w − 1 u − 1 g is an upper triangular isometry. Applying F F F to g = u w t v g=uwtv g = u w t v gives g = F ( u ) w ′ F ( t ) F ( v ) g=F(u)w'F(t)F(v) g = F ( u ) w ′ F ( t ) F ( v ) with w ′ = τ w τ w'=\tau w\tau w ′ = τ w τ , which is again a factorisation as in Theorem 12.12 . By uniqueness w ′ = w w'=w w ′ = w , F ( u ) = u F(u)=u F ( u ) = u and F ( t v ) = t v F(tv)=tv F ( t v ) = t v . So u ∈ G − u\in G^- u ∈ G − ; being unipotent of odd order, u u u lies in Ω 2 n − ( q ) \Omega^-_{2n}(q) Ω 2 n − ( q ) , since S O 2 n − ( q ) / Ω 2 n − ( q ) SO^-_{2n}(q)/\Omega^-_{2n}(q) S O 2 n − ( q ) / Ω 2 n − ( q ) has order 2. A diagonal isometry fixed by F F F satisfies t i t r ( i ) = 1 t_i t_{r(i)}=1 t i t r ( i ) = 1 and t τ ( i ) = t i t_{\tau(i)}=t_i t τ ( i ) = t i , which gives the description of t t t . A permutation commuting with r r r and τ \tau τ is a signed permutation of the n − 1 n-1 n − 1 outer pairs together with a choice on the middle pair, and the parity condition fixes that choice. That every element of W G W_G W G occurs follows from the ( B , N ) (B,N) ( B , N ) -pair of the twisted group [21].
Corollary 12.24 (Digital law of P Ω 2 n − ( q ) \mathrm{P}\Omega^-_{2n}(q) P Ω 2 n − ( q ) ). The construction of Theorem 12.18 , with W G ≅ W ( B n − 1 ) W_G \cong W(B_{n-1}) W G ≅ W ( B n − 1 ) ordered by the rotation label and Z Z Z the scalars in Ω 2 n − ( q ) \Omega^-_{2n}(q) Ω 2 n − ( q ) , gives a digital law of P Ω 2 n − ( q ) \mathrm{P}\Omega^-_{2n}(q) P Ω 2 n − ( q ) with identity 0. In its torus part the outer exponents are ( q − 1 ) (q-1) ( q − 1 ) -digits, and the middle entry is z = δ a z=\delta^a z = δ a with one ( q + 1 ) (q+1) ( q + 1 ) -digit a ∈ A q + 1 a\in\mathcal{A}_{q+1} a ∈ A q + 1 , where δ = γ q − 1 \delta=\gamma^{q-1} δ = γ q − 1 for a primitive element γ \gamma γ of F q 2 \mathbb{F}_{q^2} F q 2 .
Proof. The proof of Theorem 12.18 uses only the product structure of the cells, which Theorem 12.23 provides. The norm-one elements of F q 2 ∗ \mathbb{F}_{q^2}^{*} F q 2 ∗ form the cyclic group of order q + 1 q+1 q + 1 generated by γ q − 1 \gamma^{q-1} γ q − 1 .
The ( q + 1 ) (q+1) ( q + 1 ) -digit is the digital trace of the twist: it accounts for the factor q n + 1 q^n+1 q n + 1 in ∣ O 2 n − ( q ) ∣ = 2 q n ( n − 1 ) ( q n + 1 ) ∏ i = 1 n − 1 ( q 2 i − 1 ) |\mathrm{O}_{2n}^-(q)|=2q^{n(n-1)}(q^n+1)\prod_{i=1}^{n-1}(q^{2i}-1) ∣ O 2 n − ( q ) ∣ = 2 q n ( n − 1 ) ( q n + 1 ) ∏ i = 1 n − 1 ( q 2 i − 1 ) , where the plus type has q n − 1 q^n-1 q n − 1 .
Example 12.25. The groups Ω 4 − ( 3 ) \Omega^-_4(3) Ω 4 − ( 3 ) and Ω 4 − ( 5 ) \Omega^-_4(5) Ω 4 − ( 5 ) were generated by computer from the elements y 0 ( 1 ) y_0(1) y 0 ( 1 ) , y 0 ( θ ) y_0(\theta) y 0 ( θ ) (with θ ∉ F q \theta\notin\mathbb{F}_q θ ∈ / F q ) and their images under g ↦ ( g r ( i ) , r ( j ) ) g\mapsto(g_{r(i),r(j)}) g ↦ ( g r ( i ) , r ( j ) ) . They have orders 360 and 7800, so they are P S L 2 ( 9 ) ≅ A l t 6 \mathrm{PSL}_2(9)\cong\mathrm{Alt}_6 PSL 2 ( 9 ) ≅ Alt 6 and P S L 2 ( 25 ) \mathrm{PSL}_2(25) PSL 2 ( 25 ) , and − I -I − I does not belong to them. For every element the elimination gives w ∈ { id , w 0 } w\in\{\operatorname{id},w_0\} w ∈ { id , w 0 } with w 0 ( j ) = r ( j ) w_0(j)=r(j) w 0 ( j ) = r ( j ) , and u u u in the group. The cells have ∣ B G ∣ |B_G| ∣ B G ∣ and q 2 ∣ B G ∣ q^2|B_G| q 2 ∣ B G ∣ elements, with ∣ B G ∣ = 36 |B_G|=36 ∣ B G ∣ = 36 and 300; the factor q 2 q^2 q 2 is the folded root group. Every upper triangular element has outer diagonal entries in F q \mathbb{F}_q F q and middle entries of norm 1. For Ω 4 − ( 3 ) \Omega^-_4(3) Ω 4 − ( 3 ) , generated by its 14 root elements of the two kinds and their reversals, 1500 random products of 30 generators were checked. Every w w w commutes with r r r and τ \tau τ , every u u u lies in the group, and all 8 elements of W ( B 2 ) W(B_2) W ( B 2 ) occur. Together with § 19.12.2 and Theorem 12.13 , the book now gives three independent digital laws of A l t 6 \mathrm{Alt}_6 Alt 6 .
Remark 12.26. In even characteristic the twisted orthogonal groups must be defined by a quadratic form, since the polar bilinear form alone defines a symplectic group. The middle block then cannot be made triangular by the argument above, because the condition 2 c d = 0 2cd=0 2 c d = 0 is empty, and a separate treatment is needed.
The Suzuki groups
The Suzuki groups 2 B 2 ( q ) {}^{2}B_2(q) 2 B 2 ( q ) are the first exceptional family, and the smallest twisted groups of rank one. Their Bruhat decomposition has only two cells, and it gives a digital law with no tables at all.
Definition 12.27. Let q = 2 2 m + 1 q=2^{2m+1} q = 2 2 m + 1 and x θ = x 2 m + 1 x^\theta=x^{2^{m+1}} x θ = x 2 m + 1 , so that ( x θ ) θ = x 2 (x^\theta)^\theta=x^2 ( x θ ) θ = x 2 . For a , b , λ ∈ F q a,b,\lambda\in\mathbb{F}_q a , b , λ ∈ F q , λ ≠ 0 \lambda\ne0 λ = 0 , put
S ( a , b ) = ( 1 0 0 0 a 1 0 0 b a θ 1 0 a 2 + θ + a b + b θ a 1 + θ + b a 1 ) , M ( λ ) = diag ( λ 1 + 2 m , λ 2 m , λ − 2 m , λ − 1 − 2 m ) , S(a,b)=\begin{pmatrix}
1&0&0&0\\
a&1&0&0\\
b&a^\theta&1&0\\
a^{2+\theta}+ab+b^\theta&a^{1+\theta}+b&a&1
\end{pmatrix},
\qquad
M(\lambda)=\operatorname{diag}(\lambda^{1+2^m},\lambda^{2^m},\lambda^{-2^m},\lambda^{-1-2^m}), S ( a , b ) = 1 a b a 2 + θ + ab + b θ 0 1 a θ a 1 + θ + b 0 0 1 a 0 0 0 1 , M ( λ ) = diag ( λ 1 + 2 m , λ 2 m , λ − 2 m , λ − 1 − 2 m ) , and let T T T be the antidiagonal matrix with entries 1. The Suzuki group S z ( q ) Sz(q) S z ( q ) is the group generated by these matrices; it has order q 2 ( q 2 + 1 ) ( q − 1 ) q^2(q^2+1)(q-1) q 2 ( q 2 + 1 ) ( q − 1 ) [5].
Proposition 12.28 (Borel law). The matrices S ( a , b ) S(a,b) S ( a , b ) form a group U U U of order q 2 q^2 q 2 , with
S ( a , b ) S ( c , d ) = S ( a + c , b + d + a θ c ) , S ( a , b ) − 1 = S ( a , b + a 1 + θ ) , S(a,b)S(c,d)=S(a+c,b+d+a^\theta c),\qquad S(a,b)^{-1}=S(a,b+a^{1+\theta}), S ( a , b ) S ( c , d ) = S ( a + c , b + d + a θ c ) , S ( a , b ) − 1 = S ( a , b + a 1 + θ ) , and B = { S ( a , b ) M ( λ ) } B=\{S(a,b)M(\lambda)\} B = { S ( a , b ) M ( λ )} is a group of order q 2 ( q − 1 ) q^2(q-1) q 2 ( q − 1 ) . On labels [ a ] + [ b ] [a]+[b] [ a ] + [ b ] , the law of U U U is ( a , b ) ∗ ( c , d ) = ( a ⊕ 2 c , b ⊕ 2 d ⊕ 2 [ a θ c ] ) (a,b)\ast(c,d)=(a\oplus_2 c,b\oplus_2d\oplus_2[a^\theta c]) ( a , b ) ∗ ( c , d ) = ( a ⊕ 2 c , b ⊕ 2 d ⊕ 2 [ a θ c ]) . Here ⊕ 2 \oplus_2 ⊕ 2 is the carry-free sum, a θ a^\theta a θ is a digit-linear map (Proposition 12.2 ), and the product a θ c a^\theta c a θ c is computed by the field law.
Proof. Row 2 of S ( a , b ) S(a,b) S ( a , b ) times column 0 of S ( c , d ) S(c,d) S ( c , d ) gives b + a θ c + d b+a^{\theta}c+d b + a θ c + d , and row 1 times column 0 gives a + c a+c a + c . The entries ( 1 , 0 ) (1,0) ( 1 , 0 ) , ( 2 , 0 ) (2,0) ( 2 , 0 ) and ( 3 , 2 ) (3,2) ( 3 , 2 ) of the product are a + c a+c a + c , b + a θ c + d b+a^{\theta}c+d b + a θ c + d and a + c a+c a + c . Put a ′ = a + c a'=a+c a ′ = a + c and b ′ = b + d + a θ c b'=b+d+a^{\theta}c b ′ = b + d + a θ c , and use ( x + y ) θ = x θ + y θ (x+y)^{\theta}=x^{\theta}+y^{\theta} ( x + y ) θ = x θ + y θ , ( x θ ) = x 2 (x^{\theta})=x^2 ( x θ ) = x 2 and characteristic 2. Entry ( 2 , 1 ) (2,1) ( 2 , 1 ) is a θ + c θ = a ′ θ a^{\theta}+c^{\theta}=a'^{\theta} a θ + c θ = a ′ θ . Entry ( 3 , 1 ) (3,1) ( 3 , 1 ) is a 1 + θ + b + a c θ + c 1 + θ + d a^{1+\theta}+b+ac^{\theta}+c^{1+\theta}+d a 1 + θ + b + a c θ + c 1 + θ + d , and a ′ 1 + θ + b ′ = a 1 + θ + a c θ + a θ c + c 1 + θ + b + d + a θ c a'^{1+\theta}+b'=a^{1+\theta}+ac^{\theta}+a^{\theta}c+c^{1+\theta}+b+d+a^{\theta}c a ′ 1 + θ + b ′ = a 1 + θ + a c θ + a θ c + c 1 + θ + b + d + a θ c is the same. Entry ( 3 , 0 ) (3,0) ( 3 , 0 ) is
a 2 + θ + a b + b θ + ( a 1 + θ + b ) c + a d + c 2 + θ + c d + d θ . a^{2+\theta}+ab+b^{\theta}+(a^{1+\theta}+b)c+ad+c^{2+\theta}+cd+d^{\theta}. a 2 + θ + ab + b θ + ( a 1 + θ + b ) c + a d + c 2 + θ + c d + d θ .
On the other side, a ′ 2 + θ = a 2 + θ + a 2 c θ + a θ c 2 + c 2 + θ a'^{2+\theta}=a^{2+\theta}+a^2c^{\theta}+a^{\theta}c^2+c^{2+\theta} a ′ 2 + θ = a 2 + θ + a 2 c θ + a θ c 2 + c 2 + θ , a ′ b ′ = a b + a d + a 1 + θ c + b c + c d + a θ c 2 a'b'=ab+ad+a^{1+\theta}c+bc+cd+a^{\theta}c^2 a ′ b ′ = ab + a d + a 1 + θ c + b c + c d + a θ c 2 and b ′ θ = b θ + d θ + a 2 c θ b'^{\theta}=b^{\theta}+d^{\theta}+a^2c^{\theta} b ′ θ = b θ + d θ + a 2 c θ . Their sum is the expression above, because the terms a 2 c θ a^2c^{\theta} a 2 c θ and a θ c 2 a^{\theta}c^2 a θ c 2 occur twice. The remaining entries are 0 or 1 because both factors are lower unitriangular. The inverse follows by taking c = a c=a c = a , since a θ a + a 1 + θ = 0 a^{\theta}a+a^{1+\theta}=0 a θ a + a 1 + θ = 0 . Write M ( λ ) = diag ( d 0 , … , d 3 ) M(\lambda)=\operatorname{diag}(d_0,\ldots,d_3) M ( λ ) = diag ( d 0 , … , d 3 ) . The entry ( i , j ) (i,j) ( i , j ) of M ( λ ) − 1 S ( a , b ) M ( λ ) M(\lambda)^{-1}S(a,b)M(\lambda) M ( λ ) − 1 S ( a , b ) M ( λ ) is S ( a , b ) i j d j / d i S(a,b)_{ij}d_j/d_i S ( a , b ) ij d j / d i , and d 0 / d 1 = λ d_0/d_1=\lambda d 0 / d 1 = λ , d 0 / d 2 = λ 1 + θ d_0/d_2=\lambda^{1+\theta} d 0 / d 2 = λ 1 + θ , d 1 / d 2 = λ θ d_1/d_2=\lambda^{\theta} d 1 / d 2 = λ θ , d 0 / d 3 = λ 2 + θ d_0/d_3=\lambda^{2+\theta} d 0 / d 3 = λ 2 + θ , d 1 / d 3 = λ 1 + θ d_1/d_3=\lambda^{1+\theta} d 1 / d 3 = λ 1 + θ and d 2 / d 3 = λ d_2/d_3=\lambda d 2 / d 3 = λ . Comparing entries, using λ θ 2 = λ 2 \lambda^{\theta^2}=\lambda^2 λ θ 2 = λ 2 , gives M ( λ ) − 1 S ( a , b ) M ( λ ) = S ( λ a , λ 1 + θ b ) M(\lambda)^{-1}S(a,b)M(\lambda)=S(\lambda a,\lambda^{1+\theta}b) M ( λ ) − 1 S ( a , b ) M ( λ ) = S ( λa , λ 1 + θ b ) . So the M ( λ ) M(\lambda) M ( λ ) normalise U U U , and B = U M B=UM B = U M .
Theorem 12.29 (Digital law of S z ( q ) S_z(q) S z ( q ) ). Let γ \gamma γ be a primitive element of F q \mathbb{F}_q F q , N B = q 2 ( q − 1 ) N_B=q^2(q-1) N B = q 2 ( q − 1 ) , and label the Borel subgroup by
L B ( S ( a , b ) M ( γ e ) ) = e + ( q − 1 ) [ a ] + q [ b ] ∈ A N B . L_B(S(a,b)M(\gamma^e))=e+(q-1)[a]+q[b]\in A_{N_B}. L B ( S ( a , b ) M ( γ e )) = e + ( q − 1 ) [ a ] + q [ b ] ∈ A N B . For g ∈ S z ( q ) g\in S_z(q) g ∈ S z ( q ) :
if g 03 = 0 g_{03}=0 g 03 = 0 , then g ∈ B g\in B g ∈ B and λ ( g ) = L B ( g ) \lambda(g)=L_B(g) λ ( g ) = L B ( g ) , where γ e = g 11 θ \gamma^e=g_{11}^{\theta} γ e = g 11 θ , a = g 10 / g 00 a=g_{10}/g_{00} a = g 10 / g 00 and b = g 20 / g 00 b=g_{20}/g_{00} b = g 20 / g 00 ;
if g 03 ≠ 0 g_{03}\ne0 g 03 = 0 , put a = g 13 / g 03 a=g_{13}/g_{03} a = g 13 / g 03 , b = g 23 / g 03 b=g_{23}/g_{03} b = g 23 / g 03 and g ′ = T S ( a , b ) − 1 g g'=TS(a,b)^{-1}g g ′ = T S ( a , b ) − 1 g . Then g ′ ∈ B g'\in B g ′ ∈ B and λ ( g ) = N B + L B ( g ′ ) + N B ( [ a ] + q [ b ] ) \lambda(g)=N_B+L_B(g')+N_B([a]+q[b]) λ ( g ) = N B + L B ( g ′ ) + N B ([ a ] + q [ b ]) .
Then λ \lambda λ is a bijection S z ( q ) → A q 2 ( q 2 + 1 ) ( q − 1 ) S_z(q)\to A_{q^2(q^2+1)(q-1)} S z ( q ) → A q 2 ( q 2 + 1 ) ( q − 1 ) with λ ( I ) = 0 \lambda(I)=0 λ ( I ) = 0 , and m ∗ n = λ ( λ − 1 ( m ) λ − 1 ( n ) ) m*n=\lambda(\lambda^{-1}(m)\lambda^{-1}(n)) m ∗ n = λ ( λ − 1 ( m ) λ − 1 ( n )) is a digital law of S z ( q ) S_z(q) S z ( q ) . The case distinction is the selector B ( [ g 03 ] ) \mathcal{B}([g_{03}]) B ([ g 03 ]) .
Proof. If g = S ( a , b ) T g ′ g=S(a,b)Tg' g = S ( a , b ) T g ′ with g ′ ∈ B g'\in B g ′ ∈ B , then g ′ g' g ′ is lower triangular, so T g ′ Tg' T g ′ has last column g 33 ′ e 0 g'_{33}e_0 g 33 ′ e 0 . Hence the last column of g g g is g 33 ′ ( 1 , a , b , ∗ ) T g'_{33}(1,a,b,*)^{\mathsf T} g 33 ′ ( 1 , a , b , ∗ ) T , so g 03 = g 33 ′ ≠ 0 g_{03}=g'_{33}\ne0 g 03 = g 33 ′ = 0 and a , b a,b a , b are read off as stated. Elements of B B B are lower triangular, so g 03 = 0 g_{03}=0 g 03 = 0 ; thus B B B and U T B UTB U T B are disjoint. If S ( a , b ) T g ′ = S ( c , d ) T g ′ ′ S(a,b)Tg'=S(c,d)Tg'' S ( a , b ) T g ′ = S ( c , d ) T g ′′ , then a , b a,b a , b and c , d c,d c , d are read off from the same column, so S ( a , b ) = S ( c , d ) S(a,b)=S(c,d) S ( a , b ) = S ( c , d ) and g ′ = g ′ ′ g'=g'' g ′ = g ′′ . Therefore ∣ B ⊔ U T B ∣ = N B + q 2 N B = ∣ S z ( q ) ∣ |B\mathbin{\sqcup}UTB|=N_B+q^2N_B=|S_z(q)| ∣ B ⊔ U T B ∣ = N B + q 2 N B = ∣ S z ( q ) ∣ , and S z ( q ) = B ⊔ U T B S_z(q)=B\mathbin{\sqcup}UTB S z ( q ) = B ⊔ U T B . For g = S ( a , b ) M ( λ ) g=S(a,b)M(\lambda) g = S ( a , b ) M ( λ ) one has g 00 = λ 1 + 2 m g_{00}=\lambda^{1+2m} g 00 = λ 1 + 2 m , g 10 = a λ 1 + 2 m g_{10}=a\lambda^{1+2m} g 10 = a λ 1 + 2 m , g 20 = b λ 1 + 2 m g_{20}=b\lambda^{1+2m} g 20 = b λ 1 + 2 m and g 11 = λ 2 m g_{11}=\lambda^{2^m} g 11 = λ 2 m , and ( λ 2 m ) θ = λ (\lambda^{2^m})^{\theta}=\lambda ( λ 2 m ) θ = λ . By Lemma 1.6 the labels fill A N B A_{N_B} A N B on B B B and [ N B , N B ( q 2 + 1 ) ) [N_B,N_B(q^2+1)) [ N B , N B ( q 2 + 1 )) on U T B UTB U T B . The identity is S ( 0 , 0 ) M ( γ 0 ) S(0,0)M(\gamma^0) S ( 0 , 0 ) M ( γ 0 ) , with label 0.
Remark 12.30. This is the transitive law of Theorem 8.32 for the 2-transitive action of S z ( q ) S_z(q) S z ( q ) on the q 2 + 1 q^2+1 q 2 + 1 points of the Suzuki ovoid, with stabiliser B B B and coset representatives I I I and S ( a , b ) T S(a,b)T S ( a , b ) T . Here, however, the coset of an element is read off directly from its last column, by field divisions.
Example 12.31. For S z ( 8 ) S_z(8) S z ( 8 ) take m = 1 m=1 m = 1 , x θ = x 4 x^{\theta}=x^4 x θ = x 4 , and F 8 \mathbb{F}_8 F 8 with X 3 + X + 1 X^3+X+1 X 3 + X + 1 and γ = θ \gamma=\theta γ = θ as in Example 12.4. By computer, the 29120 labels decode to distinct matrices, and these are exactly the group generated by S ( 1 , 0 ) S(1,0) S ( 1 , 0 ) , S ( 0 , 1 ) S(0,1) S ( 0 , 1 ) , S ( γ , 0 ) S(\gamma,0) S ( γ , 0 ) , M ( γ ) M(\gamma) M ( γ ) and T T T . Encoding after decoding returns every label, 0 is the identity, and associativity holds on 3000 random triples. The element orders are 1, 2, 4, 5, 7, 13, occurring 1, 455, 3640, 5824, 12480 and 6720 times; the 455=( q 2 + 1 ) ( q − 1 ) (q^2+1)(q-1) ( q 2 + 1 ) ( q − 1 ) involutions agree with the structure of S z ( 8 ) S_z(8) S z ( 8 ) .
Remark 12.32. The small Ree groups 2 G 2 ( q ) {}^{2}G_2(q) 2 G 2 ( q ) , q = 3 2 m + 1 q=3^{2m+1} q = 3 2 m + 1 , follow the same pattern: they act 2-transitively on q 3 + 1 q^3+1 q 3 + 1 points, the Borel subgroup has order q 3 ( q − 1 ) q^3(q-1) q 3 ( q − 1 ) , and the group consists of two Bruhat cells in a 7-dimensional representation.
The small Ree groups
The small Ree groups 2 G 2 ( q ) {}^{2}G_{2}(q) 2 G 2 ( q ) , q = 3 2 m + 1 q = 3^{2m+1} q = 3 2 m + 1 , are the second twisted family of rank one. Like the Suzuki groups they consist of two Bruhat cells, but their unipotent radical has three field digits and nilpotency class 3. We use the 7-dimensional matrices of Kemper, Lübeck and Magaard [25].
Definition 12.33. Let q = 3 2 m + 1 q = 3^{2m+1} q = 3 2 m + 1 and x θ = x 3 m x^{\theta} = x^{3^{m}} x θ = x 3 m , so that x ↦ x 3 θ x \mapsto x^{3\theta} x ↦ x 3 θ is the inverse of x ↦ x θ x \mapsto x^{\theta} x ↦ x θ on F q \mathbb{F}_{q} F q . For t , u , v ∈ F q t,u,v \in\mathbb{F}_{q} t , u , v ∈ F q let
x S ( t , u , v ) = ( 1 t θ − u θ ( t u ) θ − v θ f 1 f 2 f 3 0 1 t u θ + t θ + 1 − t 2 θ + 1 − v θ f 4 f 5 0 0 1 t θ − t 2 θ v θ + ( t u ) θ f 6 0 0 0 1 t θ u θ ( t u ) θ − v θ 0 0 0 0 1 − t u θ + t θ + 1 0 0 0 0 0 1 − t θ 0 0 0 0 0 0 1 ) , x_{S}(t,u,v)=
\begin{pmatrix}
1 & t^{\theta} & -u^{\theta} & (tu)^{\theta}-v^{\theta} & f_{1} & f_{2} & f_{3}\\
0 & 1 & t & u^{\theta}+t^{\theta+1} & -t^{2\theta+1}-v^{\theta} & f_{4} & f_{5}\\
0 & 0 & 1 & t^{\theta} & -t^{2\theta} & v^{\theta}+(tu)^{\theta} & f_{6}\\
0 & 0 & 0 & 1 & t^{\theta} & u^{\theta} & (tu)^{\theta}-v^{\theta}\\
0 & 0 & 0 & 0 & 1 & -t & u^{\theta}+t^{\theta+1}\\
0 & 0 & 0 & 0 & 0 & 1 & -t^{\theta}\\
0 & 0 & 0 & 0 & 0 & 0 & 1
\end{pmatrix}, x S ( t , u , v ) = 1 0 0 0 0 0 0 t θ 1 0 0 0 0 0 − u θ t 1 0 0 0 0 ( t u ) θ − v θ u θ + t θ + 1 t θ 1 0 0 0 f 1 − t 2 θ + 1 − v θ − t 2 θ t θ 1 0 0 f 2 f 4 v θ + ( t u ) θ u θ − t 1 0 f 3 f 5 f 6 ( t u ) θ − v θ u θ + t θ + 1 − t θ 1 , with
f 1 = − u − t 3 θ + 1 − ( t v ) θ , f_{1}=-u-t^{3\theta+1}-(tv)^{\theta}, f 1 = − u − t 3 θ + 1 − ( t v ) θ , f 2 = − v − ( u v ) θ − t 3 θ + 2 − t θ u 2 θ , f_{2}=-v-(uv)^{\theta}-t^{3\theta+2}-t^{\theta}u^{2\theta}, f 2 = − v − ( uv ) θ − t 3 θ + 2 − t θ u 2 θ , f 3 = t θ v − u θ + 1 + t 4 θ + 2 − v 2 θ − t 3 θ + 1 u θ − ( t u v ) θ , f_{3}=t^{\theta}v-u^{\theta+1}+t^{4\theta+2}-v^{2\theta}-t^{3\theta+1}u^{\theta}-(tuv)^{\theta}, f 3 = t θ v − u θ + 1 + t 4 θ + 2 − v 2 θ − t 3 θ + 1 u θ − ( t uv ) θ , f 4 = − u 2 θ + t θ + 1 u θ + t v θ , f_{4}=-u^{2\theta}+t^{\theta+1}u^{\theta}+tv^{\theta}, f 4 = − u 2 θ + t θ + 1 u θ + t v θ , f 5 = v + t u − t 2 θ + 1 u θ − ( u v ) θ − t 3 θ + 2 − t θ + 1 v θ , f_{5}=v+tu-t^{2\theta+1}u^{\theta}-(uv)^{\theta}-t^{3\theta+2}-t^{\theta+1}v^{\theta}, f 5 = v + t u − t 2 θ + 1 u θ − ( uv ) θ − t 3 θ + 2 − t θ + 1 v θ , f 6 = u + t 3 θ + 1 − ( t v ) θ − t 2 θ u θ . f_{6}=u+t^{3\theta+1}-(tv)^{\theta}-t^{2\theta}u^{\theta}. f 6 = u + t 3 θ + 1 − ( t v ) θ − t 2 θ u θ . Let h ( λ ) = diag ( λ θ , λ 1 − θ , λ 2 θ − 1 , 1 , λ 1 − 2 θ , λ θ − 1 , λ − θ ) h(\lambda)=\operatorname{diag}(\lambda^{\theta},\lambda^{1-\theta},\lambda^{2\theta-1},1,\lambda^{1-2\theta},\lambda^{\theta-1},\lambda^{-\theta}) h ( λ ) = diag ( λ θ , λ 1 − θ , λ 2 θ − 1 , 1 , λ 1 − 2 θ , λ θ − 1 , λ − θ ) for λ ∈ F q × \lambda\in\mathbb{F}_{q}^{\times} λ ∈ F q × , and let n n n be the antidiagonal matrix with entries − 1 -1 − 1 . The Ree group 2 G 2 ( q ) {}^{2}G_{2}(q) 2 G 2 ( q ) is generated by these matrices; it has order q 3 ( q 3 + 1 ) ( q − 1 ) q^{3}(q^{3}+1)(q-1) q 3 ( q 3 + 1 ) ( q − 1 ) [25, 21].
Proposition 12.34 (Unipotent law). The matrices x S ( t , u , v ) x_{S}(t,u,v) x S ( t , u , v ) form a group U U U of order q 3 q^{3} q 3 , with
x S ( t 1 , u 1 , v 1 ) x S ( t 2 , u 2 , v 2 ) = x S ( t 1 + t 2 , u 1 + u 2 − t 1 t 2 θ , v 1 + v 2 − t 1 2 t 2 θ − t 2 u 1 + t 1 t 2 3 θ + 1 ) . x_{S}(t_{1},u_{1},v_{1})x_{S}(t_{2},u_{2},v_{2})=x_{S}(t_{1}+t_{2},u_{1}+u_{2}-t_{1}t_{2}^{\theta},v_{1}+v_{2}-t_{1}^{2}t_{2}^{\theta}-t_{2}u_{1}+t_{1}t_{2}^{3\theta+1}). x S ( t 1 , u 1 , v 1 ) x S ( t 2 , u 2 , v 2 ) = x S ( t 1 + t 2 , u 1 + u 2 − t 1 t 2 θ , v 1 + v 2 − t 1 2 t 2 θ − t 2 u 1 + t 1 t 2 3 θ + 1 ) . and B = U h ( F q × ) B=U h(\mathbb{F}_{q}^{\times}) B = U h ( F q × ) is the group of upper triangular elements, of order q 3 ( q − 1 ) q^{3}(q-1) q 3 ( q − 1 ) . The coordinates are read off as t = x 12 t=x_{12} t = x 12 , u = ( x 46 − t θ + 1 ) 3 θ u=(x_{46}-t^{\theta+1})^{3\theta} u = ( x 46 − t θ + 1 ) 3 θ and v = ( x 25 − ( t u ) θ ) 3 θ v=(x_{25}-(tu)^{\theta})^{3\theta} v = ( x 25 − ( t u ) θ ) 3 θ .
Proof. The product formula is [25, p. 3]; it was also checked by computer on random pairs for q = 27 q=27 q = 27 . The reading formulas come from the entries x 12 = t x_{12}=t x 12 = t , x 46 = u θ + t θ + 1 x_{46}=u^{\theta}+t^{\theta+1} x 46 = u θ + t θ + 1 and x 25 = v θ + ( t u ) θ x_{25}=v^{\theta}+(tu)^{\theta} x 25 = v θ + ( t u ) θ , so distinct triples give distinct matrices.
Theorem 12.35 (Digital law of 2 G 2 ( q ) {}^{2}G_{2}(q) 2 G 2 ( q ) ). Let γ \gamma γ be a primitive element of F q \mathbb{F}_{q} F q , N B = q 3 ( q − 1 ) N_{B}=q^{3}(q-1) N B = q 3 ( q − 1 ) , and
L B ( x S ( t , u , v ) h ( γ e ) ) = e + ( q − 1 ) ( [ t ] + q [ u ] + q 2 [ v ] ) ∈ A N B . L_{B}(x_{S}(t,u,v)h(\gamma^{e}))=e+(q-1)([t]+q[u]+q^{2}[v])\in A_{N_{B}}. L B ( x S ( t , u , v ) h ( γ e )) = e + ( q − 1 ) ([ t ] + q [ u ] + q 2 [ v ]) ∈ A N B . For g ∈ 2 G 2 ( q ) g\in{}^{2}G_{2}(q) g ∈ 2 G 2 ( q ) :
if g 60 = 0 g_{60}=0 g 60 = 0 , then g ∈ B g\in B g ∈ B and λ ( g ) = L B ( g ) \lambda(g)=L_{B}(g) λ ( g ) = L B ( g ) , where γ e = g 00 3 θ \gamma^{e}=g_{00}^{3\theta} γ e = g 00 3 θ and t , u , v t,u,v t , u , v are read off from g h ( γ e ) − 1 g h(\gamma^{e})^{-1} g h ( γ e ) − 1 by Proposition 12.34 12.34 12.34 ;
t = ( − g 50 / c ) 3 θ , u = ( g 40 / c − t θ + 1 ) 3 θ , v = ( ( t u ) θ − g 30 / c ) 3 θ , t=\left(-g_{50}/c\right)^{3\theta},\qquad u=\left(g_{40}/c-t^{\theta+1}\right)^{3\theta},\qquad v=\left((tu)^{\theta}-g_{30}/c\right)^{3\theta}, t = ( − g 50 / c ) 3 θ , u = ( g 40 / c − t θ + 1 ) 3 θ , v = ( ( t u ) θ − g 30 / c ) 3 θ , and g ′ = n − 1 x S ( t , u , v ) − 1 g g'=n^{-1}x_{S}(t,u,v)^{-1}g g ′ = n − 1 x S ( t , u , v ) − 1 g . Then g ′ ∈ B g'\in B g ′ ∈ B and λ ( g ) = N B + L B ( g ′ ) + N B ( [ t ] + q [ u ] + q 2 [ v ] ) \lambda(g)=N_{B}+L_{B}(g')+N_{B}([t]+q[u]+q^{2}[v]) λ ( g ) = N B + L B ( g ′ ) + N B ([ t ] + q [ u ] + q 2 [ v ]) .
Then λ \lambda λ is a bijection 2 G 2 ( q ) → A q 3 ( q + 1 ) ( q − 1 ) {}^{2}G_2(q) \to A_{q^3(q+1)(q-1)} 2 G 2 ( q ) → A q 3 ( q + 1 ) ( q − 1 ) with λ ( I ) = 0 \lambda(I)=0 λ ( I ) = 0 , and m ∗ n = λ ( λ − 1 ( m ) λ − 1 ( n ) ) m*n=\lambda(\lambda^{-1}(m)\lambda^{-1}(n)) m ∗ n = λ ( λ − 1 ( m ) λ − 1 ( n )) is a digital law of 2 G 2 ( q ) {}^{2}G_2(q) 2 G 2 ( q ) , selected by B ( [ g 60 ] ) B([g_{60}]) B ([ g 60 ]) .
Proof. If g = x S ( t , u , v ) n g ′ g=x_S(t,u,v)n g' g = x S ( t , u , v ) n g ′ with g ′ ∈ B g'\in B g ′ ∈ B , then n g ′ ng' n g ′ has first column − g 00 ′ e 6 -g'_{00}e_6 − g 00 ′ e 6 . So the first column of g g g is − g 00 ′ -g'_{00} − g 00 ′ times the last column ( f 3 , f 5 , f 6 , ( t u ) θ − v θ , u θ + t θ + 1 , − t θ , 1 ) T (f_3,f_5,f_6,(tu)^\theta-v^\theta,u^\theta+t^{\theta+1},-t^\theta,1)^{\mathsf T} ( f 3 , f 5 , f 6 , ( t u ) θ − v θ , u θ + t θ + 1 , − t θ , 1 ) T of x S ( t , u , v ) x_S(t,u,v) x S ( t , u , v ) . Hence g 60 = − g 00 ′ ≠ 0 g_{60}=-g'_{00}\ne0 g 60 = − g 00 ′ = 0 , and t , u , v t,u,v t , u , v are recovered by the stated formulas. Elements of B B B are upper triangular, so g 60 = 0 g_{60}=0 g 60 = 0 ; thus B B B and U n B UnB U n B are disjoint, and the decomposition g = x S ( t , u , v ) n g ′ g=x_S(t,u,v)ng' g = x S ( t , u , v ) n g ′ is unique. Therefore ∣ B ⊔ U n B ∣ = N B ( 1 + q 3 ) = ∣ 2 G 2 ( q ) ∣ |B\sqcup UnB|=N_B(1+q^3)=|{}^{2}G_2(q)| ∣ B ⊔ U n B ∣ = N B ( 1 + q 3 ) = ∣ 2 G 2 ( q ) ∣ , and 2 G 2 ( q ) = B ⊔ U n B {}^{2}G_2(q)=B\sqcup UnB 2 G 2 ( q ) = B ⊔ U n B . On B B B the diagonal of g g g is that of h ( λ ) h(\lambda) h ( λ ) , so g 00 = λ θ g_{00}=\lambda^\theta g 00 = λ θ and λ = g 00 3 θ \lambda=g_{00}^{3\theta} λ = g 00 3 θ . By Lemma 1.6 the labels fill A N B A_{N_B} A N B on B B B and [ N B , N B ( q 3 + 1 ) ) [N_B,N_B(q^3+1)) [ N B , N B ( q 3 + 1 )) on U n B UnB U n B . The identity has label 0.
Example 12.36. For q = 3 q=3 q = 3 (m = 0 m=0 m = 0 , θ = 1 \theta=1 θ = 1 ) the group 2 G 2 ( 3 ) ≅ P Γ L ( 2 , 8 ) {}^{2}G_2(3)\cong P\Gamma L(2,8) 2 G 2 ( 3 ) ≅ P Γ L ( 2 , 8 ) , of order 1512, is not simple, but it tests the construction completely. By computer, the group generated by x S ( 1 , 0 , 0 ) x_S(1,0,0) x S ( 1 , 0 , 0 ) , x S ( 2 , 0 , 0 ) x_S(2,0,0) x S ( 2 , 0 , 0 ) , x S ( 0 , 1 , 0 ) x_S(0,1,0) x S ( 0 , 1 , 0 ) , x S ( 0 , 0 , 1 ) x_S(0,0,1) x S ( 0 , 0 , 1 ) , h ( 2 ) h(2) h ( 2 ) and n n n has 1512 elements; they are exactly the decoded labels, encoding after decoding returns every label, and associativity holds on 200 random triples. For the simple group 2 G 2 ( 27 ) {}^{2}G_2(27) 2 G 2 ( 27 ) (θ = 3 \theta=3 θ = 3 , F 27 \mathbb{F}_{27} F 27 from X 3 + 2 X + 1 X^3+2X+1 X 3 + 2 X + 1 ), of order 10 073 444 472, the unipotent law was checked on 300 random pairs. On 800 random products of 40 generators, encoding followed by decoding returned the element exactly. On 800 random pairs of labels the product was encoded and reconstructed exactly, and associativity held on 300 random triples.
Remark 12.37. The unipotent law has three coupled layers. The t t t -digit is the carry-free sum ⊕ 3 \oplus_3 ⊕ 3 ; the u u u -digit receives the quadratic carry − t 1 t 2 3 θ -t_1t_2^{3\theta} − t 1 t 2 3 θ ; and the v v v -digit receives a carry of degree up to 3 θ + 2 3\theta+2 3 θ + 2 from both lower digits. This is a stacked diagram of three solid layers, one more than for the Suzuki groups, matching nilpotency class 3. The large Ree groups 2 F 4 ( q ) {}^{2}F_4(q) 2 F 4 ( q ) have rank 2, so the two-cell argument must be replaced by the cell-wise law of Theorem 12.18 .
The group G 2 ( q ) G_2(q) G 2 ( q )
The group G 2 ( q ) G_2(q) G 2 ( q ) is the first exceptional group of untwisted type. It is the automorphism group of the split octonions, and in its 7-dimensional representation it lies in Ω 7 ( q ) \Omega_7(q) Ω 7 ( q ) with an upper triangular Borel subgroup. The cell-wise law of Theorem 12.18 therefore applies, with a Weyl group of order 12.
Definition 12.38 (Split octonions). The Zorn algebra O \mathbb{O} O over F q \mathbb{F}_q F q consists of the quadruples ( a , v , w , b ) (a,v,w,b) ( a , v , w , b ) with a , b ∈ F q a,b\in\mathbb{F}_q a , b ∈ F q and v , w ∈ F q 3 v,w\in\mathbb{F}_q^3 v , w ∈ F q 3 , with product
( a , v , w , b ) ( a ′ , v ′ , w ′ , b ′ ) = ( a a ′ + v ⋅ w ′ , a v ′ + b ′ v + w × w ′ , a ′ w + b w ′ − v × v ′ , b b ′ + w ⋅ v ′ ) . (a,v,w,b)(a',v',w',b')=(aa'+v\cdot w',av'+b'v+w\times w',a'w+bw'-v\times v',bb'+w\cdot v'). ( a , v , w , b ) ( a ′ , v ′ , w ′ , b ′ ) = ( a a ′ + v ⋅ w ′ , a v ′ + b ′ v + w × w ′ , a ′ w + b w ′ − v × v ′ , b b ′ + w ⋅ v ′ ) . It is an alternative algebra with identity 1 = ( 1 , 0 , 0 , 1 ) 1=(1,0,0,1) 1 = ( 1 , 0 , 0 , 1 ) and norm N ( a , v , w , b ) = a b − v ⋅ w N(a,v,w,b)=ab-v\cdot w N ( a , v , w , b ) = ab − v ⋅ w . Put G 2 ( q ) = Aut ( O ) G_2(q)=\operatorname{Aut}(\mathbb{O}) G 2 ( q ) = Aut ( O ) , which acts on the trace-zero part. Order the basis of the trace-zero part as
w 3 , v 1 , v 2 , m , w 2 , w 1 , v 3 , m = ( 1 , 0 , 0 , − 1 ) . w_3,\ v_1,\ v_2,\ m,\ w_2,\ w_1,\ v_3,\qquad m=(1,0,0,-1). w 3 , v 1 , v 2 , m , w 2 , w 1 , v 3 , m = ( 1 , 0 , 0 , − 1 ) . In this basis the polar form of N N N is a multiple of the antidiagonal matrix J J J , so G 2 ( q ) ⊂ Ω 7 ( q ) G_2(q)\subset\Omega_7(q) G 2 ( q ) ⊂ Ω 7 ( q ) for q q q odd [21, 5].
Lemma 12.39 (Root elements). Let q q q be odd. The torus diag ( 1 , t 1 , t 2 , t 3 , t 1 − 1 , t 2 − 1 , t 3 − 1 , 1 ) \operatorname{diag}(1,t_1,t_2,t_3,t_1^{-1},t_2^{-1},t_3^{-1},1) diag ( 1 , t 1 , t 2 , t 3 , t 1 − 1 , t 2 − 1 , t 3 − 1 , 1 ) , t 1 t 2 t 3 = 1 t_1t_2t_3=1 t 1 t 2 t 3 = 1 , acts by automorphisms. With the cocharacter ( 3 , 1 , − 4 ) (3,1,-4) ( 3 , 1 , − 4 ) its weights on the derivation algebra of O \mathbb{O} O are regular, and they split the 14-dimensional algebra into the torus part and 12 one-dimensional root spaces F q D α \mathbb{F}_qD_\alpha F q D α . The root elements x α ( c ) = exp ( c D α ) x_\alpha(c)=\exp(cD_\alpha) x α ( c ) = exp ( c D α ) are automorphisms, and for positive α \alpha α they are upper unitriangular in the basis above.
Proof. The torus multiplies v i v_i v i by t i t_i t i and w i w_i w i by t i − 1 t_i^{-1} t i − 1 . The products v ⋅ w ′ v\cdot w' v ⋅ w ′ and w ⋅ v ′ w\cdot v' w ⋅ v ′ have weight 0, while w × w ′ w\times w' w × w ′ and v × v ′ v\times v' v × v ′ have the weights of v v v and w w w because t 1 t 2 t 3 = 1 t_1t_2t_3=1 t 1 t 2 t 3 = 1 . Under the cocharacter the basis vectors w 3 , v 1 , v 2 , m , w 2 , w 1 , v 3 w_3,v_1,v_2,m,w_2,w_1,v_3 w 3 , v 1 , v 2 , m , w 2 , w 1 , v 3 have the strictly decreasing values 4 , 3 , 1 , 0 , − 1 , − 3 , − 4 4,3,1,0,-1,-3,-4 4 , 3 , 1 , 0 , − 1 , − 3 , − 4 . A root vector of positive value therefore has nonzero entries only above the diagonal. Each D α D_\alpha D α is nilpotent of degree at most 3, so exp ( c D α ) \exp(cD_\alpha) exp ( c D α ) needs only the denominators 1 and 2. The derivation equations D ( x y ) = D ( x ) y + x D ( y ) D(xy)=D(x)y+xD(y) D ( x y ) = D ( x ) y + x D ( y ) are linear, and their solution spaces were found by Gaussian elimination; the statement was verified by computer for q = 5 , 7 q=5,7 q = 5 , 7 and is the standard root decomposition of G 2 G_2 G 2 [21].
Theorem 12.40 (Bruhat cells of G 2 ( q ) G_2(q) G 2 ( q ) ). Let q q q be odd and g ∈ G 2 ( q ) g\in G_2(q) g ∈ G 2 ( q ) , and write g = u w ˙ t v g=u\dot{w}tv g = u w ˙ t v as in Theorem 12.12 . Then w w w lies in the set W ( G 2 ) W(G_2) W ( G 2 ) of 12 permutations of A 7 A_7 A 7 that commute with r r r and are realised by monomial automorphisms, and u ∈ G 2 ( q ) ∩ U w u\in G_2(q)\cap U_w u ∈ G 2 ( q ) ∩ U w . Moreover u − 1 g ∈ G 2 ( q ) u^{-1}g\in G_2(q) u − 1 g ∈ G 2 ( q ) , and the cell of w w w is ( G 2 ( q ) ∩ U w ) × ( G 2 ( q ) ∩ w ˙ B ) (G_2(q)\cap U_w)\times(G_2(q)\cap\dot{w}B) ( G 2 ( q ) ∩ U w ) × ( G 2 ( q ) ∩ w ˙ B ) . W ( G 2 ) W(G_2) W ( G 2 ) is the dihedral group of order 12 inside the group W ( B 3 ) W(B_3) W ( B 3 ) of order 48 from Theorem 12.17 .
Proof. G 2 ( q ) G_2(q) G 2 ( q ) has a split ( B , N ) (B,N) ( B , N ) -pair with B B B its upper triangular elements and N N N its monomial elements [21]. The proof of Theorem 12.17 applies word for word: the refined Bruhat decomposition g = u ′ n b ′ g=u'n b' g = u ′ n b ′ in G 2 ( q ) G_2(q) G 2 ( q ) is a factorisation as in Theorem 12.12 , so by uniqueness u = u ′ ∈ G 2 ( q ) u=u'\in G_2(q) u = u ′ ∈ G 2 ( q ) and w w w is the permutation of n n n . Since G 2 ( q ) ⊆ Ω 7 ( q ) G_2(q)\subseteq\Omega_7(q) G 2 ( q ) ⊆ Ω 7 ( q ) , the permutation w w w commutes with r r r .
Corollary 12.41 (Digital law of G 2 ( q ) G_2(q) G 2 ( q ) ). Order W ( G 2 ) W(G_2) W ( G 2 ) by the rotation label. The construction of Theorem 12.18 , with Z = { I } Z=\{I\} Z = { I } , gives a digital law of G 2 ( q ) G_2(q) G 2 ( q ) on A q 6 ( q 6 − 1 ) ( q 2 − 1 ) A_{q^6(q^6-1)(q^2-1)} A q 6 ( q 6 − 1 ) ( q 2 − 1 ) with identity 0. Membership of a 7 × 7 7\times7 7 × 7 matrix in G 2 ( q ) G_2(q) G 2 ( q ) is itself digital: it means that the 64 products of basis vectors are preserved, each computed by the field law.
Proof. Theorem 12.40 provides the product structure of the cells used in the proof of Theorem 12.18 , and the centre of G 2 ( q ) G_2(q) G 2 ( q ) is trivial.
Example 12.42. For q = 5 q=5 q = 5 and q = 7 q=7 q = 7 the computer gave the following. The Zorn product is alternative on random triples. The derivation algebra has dimension 14, with 12 one-dimensional root spaces, 6 of them positive, and the positive root elements are upper unitriangular automorphisms. For the two simple roots, x α ( 1 ) x − α ( c ) x α ( 1 ) x_\alpha(1)x_{-\alpha}(c)x_\alpha(1) x α ( 1 ) x − α ( c ) x α ( 1 ) is monomial for a suitable c c c . These two monomial automorphisms generate exactly 12 permutations, all commuting with r r r . For each of the 12 permutations w w w , 20 random elements u n w t u ′ u n_w t u' u n w t u ′ were reduced: the elimination returned w w w , and both u u u and u − 1 g u^{-1}g u − 1 g preserve the octonion product. On 300 random products of 40 root elements, the elimination always gave a permutation in W ( G 2 ) W(G_2) W ( G 2 ) , with u u u and u − 1 g u^{-1}g u − 1 g in G 2 ( q ) G_2(q) G 2 ( q ) .
Remark 12.43. In characteristic 3 the Ree groups of § 20.8 are the fixed points of an exceptional endomorphism of G 2 ( q ) G_2(q) G 2 ( q ) , and the matrices x ( s ( t , u , v ) ) x(s(t,u,v)) x ( s ( t , u , v )) are products of root elements of G 2 ( q ) G_2(q) G 2 ( q ) . The next untwisted and twisted cases are 3 D 4 ( q ) {}^{3}D_4(q) 3 D 4 ( q ) , whose natural module has dimension 8 over F q 3 \mathbb{F}_{q^3} F q 3 , and F 4 ( q ) F_4(q) F 4 ( q ) , with a 26-dimensional module.
The triality groups 3 D 4 ( q ) {}^{3}D_4(q) 3 D 4 ( q )
Let K = F q 3 K=\mathbb{F}_{q^3} K = F q 3 and σ ( z ) = z q \sigma(z)=z^q σ ( z ) = z q . The triality group 3 D 4 ( q ) {}^{3}D_4(q) 3 D 4 ( q ) is the twisted group obtained from D 4 ( K ) ≅ Ω 8 + ( K ) D_4(K)\cong\Omega_8^+(K) D 4 ( K ) ≅ Ω 8 + ( K ) by combining the order-three symmetry of the D 4 D_4 D 4 diagram with σ \sigma σ [21]. Its relative root system is G 2 G_2 G 2 , so it has 12 Bruhat cells as in § 20.9 . The new feature is that the root digits have two widths: short root groups are parametrised by K K K , which is three F q \mathbb{F}_q F q -digits, and long root groups by F q \mathbb{F}_q F q .
Definition 12.44 (Triality model). On K 8 K^8 K 8 take the quadratic form Q ( x ) = x 1 x 8 + x 2 x 7 + x 3 x 6 + x 4 x 5 Q(x)=x_1x_8+x_2x_7+x_3x_6+x_4x_5 Q ( x ) = x 1 x 8 + x 2 x 7 + x 3 x 6 + x 4 x 5 , whose polar form has antidiagonal matrix J J J . Write i ′ = 9 − i i'=9-i i ′ = 9 − i , and for i < j i<j i < j with
j ≠ i ′ j \ne i' j = i ′ put x i j ( t ) = I + t ( E i j − E j ′ i ′ ) x_{ij}(t)=I+t(E_{ij}-E_{j'i'}) x ij ( t ) = I + t ( E ij − E j ′ i ′ ) . The simple roots of D 4 D_4 D 4 correspond to ( 1 , 2 ) (1,2) ( 1 , 2 ) , ( 2 , 3 ) (2,3) ( 2 , 3 ) , ( 3 , 4 ) (3,4) ( 3 , 4 ) and ( 3 , 5 ) (3,5) ( 3 , 5 ) , and ( 2 , 3 ) (2,3) ( 2 , 3 ) is the central node. Put
x R ( ξ ) = x 12 ( ξ ) x 34 ( ξ σ ) x 35 ( ξ σ 2 ) ( ξ ∈ K ) , x S ( η ) = x 23 ( η ) ( η ∈ F q ) . x_R(\xi)=x_{12}(\xi)x_{34}(\xi^\sigma)x_{35}(\xi^{\sigma^2}) \quad(\xi\in K), \qquad x_S(\eta)=x_{23}(\eta) \quad(\eta\in\mathbb{F}_q). x R ( ξ ) = x 12 ( ξ ) x 34 ( ξ σ ) x 35 ( ξ σ 2 ) ( ξ ∈ K ) , x S ( η ) = x 23 ( η ) ( η ∈ F q ) .
and let x − R ( ξ ) = x R ( ξ ) T x_{-R}(\xi)=x_R(\xi)^T x − R ( ξ ) = x R ( ξ ) T and x − S ( η ) = x S ( η ) T x_{-S}(\eta)=x_S(\eta)^T x − S ( η ) = x S ( η ) T . Then 3 D 4 ( q ) {}^3D_4(q) 3 D 4 ( q ) is the group generated by these four families. Further put n R = x R ( 1 ) x − R ( − 1 ) x R ( 1 ) n_R=x_R(1)x_{-R}(-1)x_R(1) n R = x R ( 1 ) x − R ( − 1 ) x R ( 1 ) and n S = x S ( 1 ) x − S ( − 1 ) x S ( 1 ) n_S=x_S(1)x_{-S}(-1)x_S(1) n S = x S ( 1 ) x − S ( − 1 ) x S ( 1 ) .
The three factors of x R ( ξ ) x_R(\xi) x R ( ξ ) belong to the three outer nodes, which triality permutes, and their parameters are the three conjugates of ξ \xi ξ . This is Steinberg’s description of the twisted group by its fixed root subgroups [21]. Signs of the simple root elements can be changed by conjugating with a diagonal matrix, so the sign convention does not matter.
Lemma 12.45 (Borel digits). Let U = ⟨ x R ( K ) , x S ( F q ) ⟩ U=\langle x_R(K),x_S(\mathbb{F}_q)\rangle U = ⟨ x R ( K ) , x S ( F q )⟩ and let T T T be generated by h R ( ξ ) = n R ( ξ ) n R ( 1 ) − 1 h_R(\xi)=n_R(\xi)n_R(1)^{-1} h R ( ξ ) = n R ( ξ ) n R ( 1 ) − 1 and h S ( η ) = n S ( η ) n S ( 1 ) − 1 h_S(\eta)=n_S(\eta)n_S(1)^{-1} h S ( η ) = n S ( η ) n S ( 1 ) − 1 , where n R ( ξ ) = x R ( ξ ) x − R ( − ξ − 1 ) x R ( ξ ) n_R(\xi)=x_R(\xi)x_{-R}(-\xi^{-1})x_R(\xi) n R ( ξ ) = x R ( ξ ) x − R ( − ξ − 1 ) x R ( ξ ) and similarly for n S n_S n S . Then U U U is upper unitriangular and T T T is diagonal. In the G 2 G_2 G 2 order of positive relative roots
R , S , R + S , 2 R + S , 3 R + S , 3 R + 2 S , R,\quad S,\quad R+S,\quad2R+S,\quad3R+S,\quad3R+2S, R , S , R + S , 2 R + S , 3 R + S , 3 R + 2 S , the root groups are parametrised by K , F q , K , K , F q , F q K,\mathbb{F}_q,K,K,\mathbb{F}_q,\mathbb{F}_q K , F q , K , K , F q , F q . Hence ∣ U ∣ = q 12 |U|=q^{12} ∣ U ∣ = q 12 , ∣ T ∣ = ( q 3 − 1 ) ( q − 1 ) |T|=(q^3-1)(q-1) ∣ T ∣ = ( q 3 − 1 ) ( q − 1 ) and B = U T B=UT B = U T has order q 12 ( q 3 − 1 ) ( q − 1 ) q^{12}(q^3-1)(q-1) q 12 ( q 3 − 1 ) ( q − 1 ) .
Proof. The roots R R R , R + S R+S R + S and 2 R + S 2R+S 2 R + S are images of orbits of length three of D 4 D_4 D 4 roots, so their fixed root elements carry one parameter of K K K . The other three positive relative roots come from single roots fixed by triality and carry a parameter of F q \mathbb{F}_q F q [21]. The orders follow, with
q 3 ⋅ q ⋅ q 3 ⋅ q 3 ⋅ q ⋅ q = q 12 . q^3\cdot q\cdot q^3\cdot q^3\cdot q\cdot q=q^{12}. q 3 ⋅ q ⋅ q 3 ⋅ q 3 ⋅ q ⋅ q = q 12 .
Proposition 12.46 (Unipotent law of 3 D 4 ( q ) {}^3D_4(q) 3 D 4 ( q ) ). Write N ( x ) = x 1 + σ + σ 2 N(x)=x^{1+\sigma+\sigma^2} N ( x ) = x 1 + σ + σ 2 and Tr ( x ) = x + x σ + x σ 2 \operatorname{Tr}(x)=x+x^\sigma+x^{\sigma^2} Tr ( x ) = x + x σ + x σ 2 for x ∈ K x\in K x ∈ K . Put
x R + S ( ξ ) = x 13 ( ξ ) x 24 ( − ξ σ ) x 25 ( − ξ σ 2 ) , x 2 R + S ( ξ ) = x 14 ( ξ ) x 26 ( ξ σ ) x 15 ( ξ σ 2 ) , x_{R+S}(\xi)=x_{13}(\xi)x_{24}(-\xi^\sigma)x_{25}(-\xi^{\sigma^2}), \qquad x_{2R+S}(\xi)=x_{14}(\xi)x_{26}(\xi^\sigma)x_{15}(\xi^{\sigma^2}), x R + S ( ξ ) = x 13 ( ξ ) x 24 ( − ξ σ ) x 25 ( − ξ σ 2 ) , x 2 R + S ( ξ ) = x 14 ( ξ ) x 26 ( ξ σ ) x 15 ( ξ σ 2 ) , x 3 R + S ( η ) = x 16 ( η ) , x 3 R + 2 S ( η ) = x 17 ( η ) , x_{3R+S}(\eta)=x_{16}(\eta), \qquad x_{3R+2S}(\eta)=x_{17}(\eta), x 3 R + S ( η ) = x 16 ( η ) , x 3 R + 2 S ( η ) = x 17 ( η ) , and, for a , c , d ∈ K a,c,d\in K a , c , d ∈ K and b , e , f ∈ F q b,e,f\in\mathbb{F}_q b , e , f ∈ F q ,
u ( a , b , c , d , e , f ) = x R ( a ) x S ( b ) x R + S ( c ) x 2 R + S ( d ) x 3 R + S ( e ) x 3 R + 2 S ( f ) . u(a,b,c,d,e,f)=x_R(a)x_S(b)x_{R+S}(c)x_{2R+S}(d)x_{3R+S}(e)x_{3R+2S}(f). u ( a , b , c , d , e , f ) = x R ( a ) x S ( b ) x R + S ( c ) x 2 R + S ( d ) x 3 R + S ( e ) x 3 R + 2 S ( f ) . These q 12 q^{12} q 12 matrices are distinct and form the group U U U . Moreover
u ( a , b , c , d , e , f ) u ( a ′ , b ′ , c ′ , d ′ , e ′ , f ′ ) = u ( a ′ ′ , b ′ ′ , c ′ ′ , d ′ ′ , e ′ ′ , f ′ ′ ) , u(a,b,c,d,e,f)u(a',b',c',d',e',f')=u(a'',b'',c'',d'',e'',f''), u ( a , b , c , d , e , f ) u ( a ′ , b ′ , c ′ , d ′ , e ′ , f ′ ) = u ( a ′′ , b ′′ , c ′′ , d ′′ , e ′′ , f ′′ ) , where
a ′ ′ = a + a ′ , b ′ ′ = b + b ′ , c ′ ′ = c + c ′ − b a ′ , a''=a+a', \qquad b''=b+b', \qquad c''=c+c'-ba', a ′′ = a + a ′ , b ′′ = b + b ′ , c ′′ = c + c ′ − b a ′ , d ′ ′ = d + d ′ + a ′ c σ + a ′ σ c − b a ′ a ′ σ , d''=d+d'+a'c^\sigma+a'^\sigma c-ba'a'^\sigma, d ′′ = d + d ′ + a ′ c σ + a ′ σ c − b a ′ a ′ σ , e ′ ′ = e + e ′ + b N ( a ′ ) − Tr ( a ′ a ′ σ c σ 2 ) − Tr ( a ′ d σ ) , e''=e+e'+bN(a')-\operatorname{Tr}(a'a'^\sigma c^{\sigma^2})-\operatorname{Tr}(a'd^\sigma), e ′′ = e + e ′ + b N ( a ′ ) − Tr ( a ′ a ′ σ c σ 2 ) − Tr ( a ′ d σ ) , f ′ ′ = f + f ′ − b ′ e + b ( b − b ′ ) N ( a ′ ) + ( b ′ − b ) Tr ( a ′ a ′ σ c σ 2 ) − b Tr ( a ′ σ c σ 2 ) + b ′ Tr ( a ′ d σ ) + Tr ( a ′ c σ c σ 2 ) + Tr ( a ′ c σ c σ 2 ) + Tr ( a ′ c σ c σ 2 ) + Tr ( c ′ d σ ) . \begin{aligned}
f''={}&f+f'-b'e+b(b-b')N(a')+(b'-b)\operatorname{Tr}(a'a'^\sigma c^{\sigma^2})\\
&-b\operatorname{Tr}(a'^\sigma c^{\sigma^2})+b'\operatorname{Tr}(a'd^\sigma)\\
&+\operatorname{Tr}(a'c^\sigma c^{\sigma^2})+\operatorname{Tr}(a'c^\sigma c^{\sigma^2})+\operatorname{Tr}(a'c^\sigma c^{\sigma^2})+\operatorname{Tr}(c'd^\sigma).
\end{aligned} f ′′ = f + f ′ − b ′ e + b ( b − b ′ ) N ( a ′ ) + ( b ′ − b ) Tr ( a ′ a ′ σ c σ 2 ) − b Tr ( a ′ σ c σ 2 ) + b ′ Tr ( a ′ d σ ) + Tr ( a ′ c σ c σ 2 ) + Tr ( a ′ c σ c σ 2 ) + Tr ( a ′ c σ c σ 2 ) + Tr ( c ′ d σ ) . The inverse of u ( a , b , c , d , e , f ) u(a,b,c,d,e,f) u ( a , b , c , d , e , f ) is u ( a ˉ , b ˉ , c ˉ , d ˉ , e ˉ , f ˉ ) u(\bar a,\bar b,\bar c,\bar d,\bar e,\bar f) u ( a ˉ , b ˉ , c ˉ , d ˉ , e ˉ , f ˉ ) with
a ˉ = − a , b ˉ = − b , c ˉ = − c − a b , d ˉ = − d + a c σ + a σ c + b a a σ , \bar a=-a,\qquad\bar b=-b,\qquad\bar c=-c-ab,\qquad\bar d=-d+ac^\sigma+a^\sigma c+baa^\sigma, a ˉ = − a , b ˉ = − b , c ˉ = − c − ab , d ˉ = − d + a c σ + a σ c + ba a σ , e ˉ = − e + b N ( a ) + Tr ( a a σ c σ 2 ) − Tr ( a d σ ) , \bar e=-e+bN(a)+\operatorname{Tr}(aa^\sigma c^{\sigma^2})-\operatorname{Tr}(ad^\sigma), e ˉ = − e + b N ( a ) + Tr ( a a σ c σ 2 ) − Tr ( a d σ ) , f ˉ = − f − b e − b 2 N ( a ) − b Tr ( a a σ c σ 2 ) − Tr ( a c σ c σ 2 ) + Tr ( c d σ ) . \bar f=-f-be-b^2N(a)-b\operatorname{Tr}(aa^\sigma c^{\sigma^2})-\operatorname{Tr}(ac^\sigma c^{\sigma^2})+\operatorname{Tr}(cd^\sigma). f ˉ = − f − b e − b 2 N ( a ) − b Tr ( a a σ c σ 2 ) − Tr ( a c σ c σ 2 ) + Tr ( c d σ ) .
The coordinates of u ∈ U u \in U u ∈ U are read off from left to right: each one is the entry of u u u at the position of its first factor ( ( 1 , 2 ) , ( 2 , 3 ) , ( 1 , 3 ) , ( 1 , 4 ) , ( 1 , 6 ) , ( 1 , 7 ) ) ((1,2),(2,3),(1,3),(1,4),(1,6),(1,7)) (( 1 , 2 ) , ( 2 , 3 ) , ( 1 , 3 ) , ( 1 , 4 ) , ( 1 , 6 ) , ( 1 , 7 )) after the factors to its left have been removed.
Proof. Work in the polynomial ring over Z \mathbb{Z} Z in independent variables that stand for a a a , a σ a^\sigma a σ , a σ 2 a^{\sigma^2} a σ 2 , and so on. Multiply the two 8 × 8 8 \times8 8 × 8 matrices, then strip off the twelve D 4 D_4 D 4 root factors in order of height. Each stripped coordinate is the entry at the root position, because all other factors contribute only at greater heights. Computer algebra gives the twelve D 4 D_4 D 4 coordinates of the product as polynomials. In each triality orbit they are conjugate, with the signs used above, and the coordinates of the three fixed roots are invariant under the cyclic shift of variables. So the product is again of the form u ( ⋅ ) u(\cdot) u ( ⋅ ) , with the coordinates displayed. The same holds for the inverse. These are polynomial identities over Z \mathbb{Z} Z , so they remain true after specialising to K K K in every characteristic, and N N N and Tr \operatorname{Tr} Tr take values in F q \mathbb{F}_q F q . Hence the q 12 q^{12} q 12 matrices u ( ⋅ ) u(\cdot) u ( ⋅ ) form a group, which contains x R ( K ) x_R(K) x R ( K ) and x S ( F q ) x_S(\mathbb{F}_q) x S ( F q ) , so it contains U U U . By Lemma 12.45, ∣ U ∣ = q 12 |U| = q^{12} ∣ U ∣ = q 12 , so the two groups are equal. The signs in x R + S x_{R+S} x R + S were found in the same way, from the D 4 D_4 D 4 coordinates of x R ( a ) x S ( b ) x R ( c ) x S ( d ) x R ( e ) x S ( f ) x_R(a)x_S(b)x_R(c)x_S(d)x_R(e)x_S(f) x R ( a ) x S ( b ) x R ( c ) x S ( d ) x R ( e ) x S ( f ) , where all twelve coordinates are nonzero.
Proposition 12.47 (Torus and Weyl relations). Write n R = n R ( 1 ) n_R = n_R(1) n R = n R ( 1 ) and n S = n S ( 1 ) n_S = n_S(1) n S = n S ( 1 ) , and for x ∈ K × x \in K^\times x ∈ K × , y ∈ F q × y \in\mathbb{F}_q^\times y ∈ F q × put h ( x , y ) = h R ( x ) h S ( y ) h(x,y) = h_R(x)h_S(y) h ( x , y ) = h R ( x ) h S ( y ) . Then
h ( x , y ) = diag ( x , y x , x σ x σ 2 y , x σ 2 x σ , x σ x σ 2 , y x σ x σ 2 , 1 y , 1 x ) . h(x,y) = \operatorname{diag}\left(x,\frac{y}{x},\frac{x^\sigma x^{\sigma^2}}{y},\frac{x^{\sigma^2}}{x^\sigma},\frac{x^\sigma}{x^{\sigma^2}},\frac{y}{x^\sigma x^{\sigma^2}},\frac{1}{y},\frac{1}{x}\right). h ( x , y ) = diag ( x , x y , y x σ x σ 2 , x σ x σ 2 , x σ 2 x σ , x σ x σ 2 y , y 1 , x 1 ) . so h ( x , y ) h ( x ′ , y ′ ) = h ( x x ′ , y y ′ ) h(x,y)h(x',y') = h(xx',yy') h ( x , y ) h ( x ′ , y ′ ) = h ( x x ′ , y y ′ ) , and x = h 11 x = h_{11} x = h 11 , y = h 11 h 22 y = h_{11}h_{22} y = h 11 h 22 are read off from the diagonal. Moreover:
(a) For h = h ( x , y ) h = h(x,y) h = h ( x , y ) ,
h u ( a , b , c , d , e , f ) h − 1 = u ( x 2 y a , y 2 N ( x ) b , x 2 y N ( x ) c , x x σ x σ 2 d , N ( x ) y e , y f ) . h u(a,b,c,d,e,f) h^{-1} = u\left(\frac{x^2}{y}a,\frac{y^2}{N(x)}b,\frac{x^2y}{N(x)}c,\frac{x x^\sigma}{x^{\sigma^2}}d,\frac{N(x)}{y}e,yf\right). h u ( a , b , c , d , e , f ) h − 1 = u ( y x 2 a , N ( x ) y 2 b , N ( x ) x 2 y c , x σ 2 x x σ d , y N ( x ) e , y f ) . In particular h x R ( t ) h − 1 = x R ( χ R ( h ) t ) h x_R(t) h^{-1} = x_R(\chi_R(h)t) h x R ( t ) h − 1 = x R ( χ R ( h ) t ) and h x S ( t ) h − 1 = x S ( χ S ( h ) t ) h x_S(t) h^{-1} = x_S(\chi_S(h)t) h x S ( t ) h − 1 = x S ( χ S ( h ) t ) , where χ R ( h ) = x 2 y − 1 \chi_R(h) = x^2y^{-1} χ R ( h ) = x 2 y − 1 and χ S ( h ) = y 2 N ( x ) − 1 \chi_S(h) = y^2N(x)^{-1} χ S ( h ) = y 2 N ( x ) − 1 .
(b) n R − 1 h ( x , y ) n R = h ( y x − 1 , y ) n_R^{-1}h(x,y)n_R = h(yx^{-1},y) n R − 1 h ( x , y ) n R = h ( y x − 1 , y ) , n S − 1 h ( x , y ) n S = h ( x , N ( x ) y − 1 ) n_S^{-1}h(x,y)n_S = h(x,N(x)y^{-1}) n S − 1 h ( x , y ) n S = h ( x , N ( x ) y − 1 ) , n R 2 = h ( − 1 , 1 ) n_R^2 = h(-1,1) n R 2 = h ( − 1 , 1 ) and n S 2 = h ( 1 , − 1 ) n_S^2 = h(1,-1) n S 2 = h ( 1 , − 1 ) .
(c) n s x s ( t ) n s − 1 = x − s ( − t ) n_s x_s(t)n_s^{-1} = x_{-s}(-t) n s x s ( t ) n s − 1 = x − s ( − t ) for s ∈ { R , S } s \in\{R,S\} s ∈ { R , S } , and for t ≠ 0 t \ne0 t = 0 , x − s ( t ) = x s ( t − 1 ) n s h s ( − t ) x s ( t − 1 ) x_{-s}(t) = x_s(t^{-1})n_s h_s(-t)x_s(t^{-1}) x − s ( t ) = x s ( t − 1 ) n s h s ( − t ) x s ( t − 1 ) .
(d) The conjugations φ s ( v ) = n s v n s − 1 \varphi_s(v) = n_svn_s^{-1} φ s ( v ) = n s v n s − 1 and ψ s ( v ) = n s − 1 v n s \psi_s(v) = n_s^{-1}vn_s ψ s ( v ) = n s − 1 v n s , on elements of U U U whose s s s -coordinate is $0, are
φ R u ( 0 , b , c , d , e , f ) = u ( 0 , e , − d σ , c σ 2 , − b , f + b e − Tr ( c d σ ) ) , \varphi_R u(0,b,c,d,e,f) = u(0,e,-d^\sigma,c^{\sigma^2},-b,f+be-\operatorname{Tr}(cd^\sigma)), φ R u ( 0 , b , c , d , e , f ) = u ( 0 , e , − d σ , c σ 2 , − b , f + b e − Tr ( c d σ )) , ψ R u ( 0 , b , c , d , e , f ) = u ( 0 , − e , d σ , − c σ 2 , b , f + b e − Tr ( c d σ ) ) , \psi_R u(0,b,c,d,e,f) = u(0,-e,d^\sigma,-c^{\sigma^2},b,f+be-\operatorname{Tr}(cd^\sigma)), ψ R u ( 0 , b , c , d , e , f ) = u ( 0 , − e , d σ , − c σ 2 , b , f + b e − Tr ( c d σ )) , φ S u ( a , 0 , c , d , e , f ) = u ( c , 0 , − a , d − a c σ − a σ c , Tr ( a c σ c σ 2 ) − f , e + Tr ( a a σ c σ 2 ) ) , \varphi_S u(a,0,c,d,e,f) = u\left(c,0,-a,d-ac^\sigma-a^{\sigma}c,\operatorname{Tr}(ac^\sigma c^{\sigma^2})-f,e+\operatorname{Tr}(aa^\sigma c^{\sigma^2})\right), φ S u ( a , 0 , c , d , e , f ) = u ( c , 0 , − a , d − a c σ − a σ c , Tr ( a c σ c σ 2 ) − f , e + Tr ( a a σ c σ 2 ) ) , ψ S u ( a , 0 , c , d , e , f ) = u ( − c , 0 , a , d − a c σ − a σ c , f − Tr ( a c σ c σ 2 ) , − e − Tr ( a a σ c σ 2 ) ) . \psi_S u(a,0,c,d,e,f) = u\left(-c,0,a,d-ac^\sigma-a^\sigma c,f-\operatorname{Tr}(ac^\sigma c^{\sigma^2}),-e-\operatorname{Tr}(aa^\sigma c^{\sigma^2})\right). ψ S u ( a , 0 , c , d , e , f ) = u ( − c , 0 , a , d − a c σ − a σ c , f − Tr ( a c σ c σ 2 ) , − e − Tr ( a a σ c σ 2 ) ) . (e) u ( a , b , c , d , e , f ) = x R ( a ) u ( 0 , b , c , d , e , f ) u(a,b,c,d,e,f) = x_R(a)u(0,b,c,d,e,f) u ( a , b , c , d , e , f ) = x R ( a ) u ( 0 , b , c , d , e , f ) and u ( a , b , c , d , e , f ) = x S ( b ) u ( a , 0 , c + a b , d + b a a σ , e − b N ( a ) , f + 2 b 2 N ( a ) + b Tr ( a a σ c σ 2 ) ) u(a,b,c,d,e,f) = x_S(b)u( a,0,c+ab,d+b aa^\sigma,e-bN(a),f+2b^2N(a)+b\operatorname{Tr}(aa^\sigma c^{\sigma^2})) u ( a , b , c , d , e , f ) = x S ( b ) u ( a , 0 , c + ab , d + ba a σ , e − b N ( a ) , f + 2 b 2 N ( a ) + b Tr ( a a σ c σ 2 )) .
Hence T = { h ( x , y ) } T = \{h(x,y)\} T = { h ( x , y )} has order ( q 3 − 1 ) ( q − 1 ) (q^3-1)(q-1) ( q 3 − 1 ) ( q − 1 ) .
Proof. All statements except the formulas for ψ s \psi_s ψ s were obtained by exact computer algebra. The matrices are built in the ring of Laurent polynomials over Z \mathbb{Z} Z , with independent variables for the conjugates of x x x , a a a , c c c , d d d and for y y y , b b b , e e e , f f f , and products in U U U are decomposed by the stripping procedure of Proposition 12.46 . The coordinates in each triality orbit came out conjugate, as required. These are identities over Z \mathbb{Z} Z , so they hold after specialisation in every characteristic. For ψ s \psi_s ψ s : from (b), n s − 1 = h s ( − 1 ) n s n_s^{-1}=h_s(-1)n_s n s − 1 = h s ( − 1 ) n s , so ψ s ( v ) = h s ( − 1 ) φ s ( v ) h s ( − 1 ) \psi_s(v)=h_s(-1)\varphi_s(v)h_s(-1) ψ s ( v ) = h s ( − 1 ) φ s ( v ) h s ( − 1 ) . By (a), conjugation by h ( − 1 , 1 ) h(-1,1) h ( − 1 , 1 ) changes the signs of b b b , c c c , d d d , e e e , and conjugation by h ( 1 , − 1 ) h(1,-1) h ( 1 , − 1 ) changes those of a a a , c c c , e e e , f f f . Finally, the map ( x , y ) ↦ h ( x , y ) (x,y)\mapsto h(x,y) ( x , y ) ↦ h ( x , y ) is injective by the diagonal formula. □
For w ∈ W w\in W w ∈ W fix a reduced word w = s 1 ⋯ s k w=s_1\cdots s_k w = s 1 ⋯ s k in { R , S } \{R,S\} { R , S } . It is unique for w ≠ w 0 w\ne w_0 w = w 0 ; for the longest element take w 0 = ( R S ) 3 w_0=(RS)^3 w 0 = ( R S ) 3 . The other word ( S R ) 3 (SR)^3 ( S R ) 3 gives n ( R S ) 3 = n ( S R ) 3 t 0 n_{(RS)^3}=n_{(SR)^3}t_0 n ( R S ) 3 = n ( S R ) 3 t 0 for some t 0 ∈ T t_0\in T t 0 ∈ T , since the two monomial matrices have the same permutation. In fact t 0 = I t_0=I t 0 = I : the matrices n R , n S n_R,n_S n R , n S have integer entries, so t 0 t_0 t 0 is a diagonal matrix of signs over Z \mathbb{Z} Z , and its reduction modulo 3 was computed to be I I I . For a word s 1 ⋯ s k s_1\cdots s_k s 1 ⋯ s k with digits t i t_i t i , where t i ∈ K t_i\in K t i ∈ K if s i = R s_i=R s i = R and t i ∈ F q t_i\in\mathbb{F}_q t i ∈ F q if s i = S s_i=S s i = S , put
P ( t 1 , … , t k ) = x s 1 ( t 1 ) n s 1 x s 2 ( t 2 ) n s 2 ⋯ x s k ( t k ) n s k . P(t_1,\ldots,t_k)=x_{s_1}(t_1)n_{s_1}x_{s_2}(t_2)n_{s_2}\cdots x_{s_k}(t_k)n_{s_k}. P ( t 1 , … , t k ) = x s 1 ( t 1 ) n s 1 x s 2 ( t 2 ) n s 2 ⋯ x s k ( t k ) n s k .
Theorem 12.48 (Closed multiplication in 3 D 4 ( q ) {}^3D_4(q) 3 D 4 ( q ) ). Every g ∈ 3 D 4 ( q ) g\in{}^3D_4(q) g ∈ 3 D 4 ( q ) has a unique expression g = P ( t 1 , … , t k ) h ( x , y ) u ( a , b , c , d , e , f ) g=P(t_1,\ldots,t_k)h(x,y)u(a,b,c,d,e,f) g = P ( t 1 , … , t k ) h ( x , y ) u ( a , b , c , d , e , f ) , with the fixed word of its Bruhat cell. Right multiplication of such a g = P h v g=Phv g = P h v is given in closed form as follows.
By v ′ ∈ U v'\in U v ′ ∈ U : replace v v v by v v ′ vv' v v ′ (Proposition 12.46 ).
By h ′ ∈ T h'\in T h ′ ∈ T : replace v v v by h ′ − 1 v h ′ {h'}^{-1}vh' h ′ − 1 v h ′ (Proposition 12.47 (a)) and h h h by h h ′ hh' h h ′ .
By n s n_s n s : write v = x s ( t ) v ′ v=x_s(t)v' v = x s ( t ) v ′ by Proposition 12.47 (e). Put t ′ = χ s ( h ) t t'=\chi_s(h)t t ′ = χ s ( h ) t , v ′ ′ = ψ s ( v ′ ) v''=\psi_s(v') v ′′ = ψ s ( v ′ ) and h s = n s − 1 h n s h^s=n_s^{-1}hn_s h s = n s − 1 h n s . Then g n s = P x s ( t ′ ) n s h s v ′ ′ gn_s=P x_s(t')n_s h^s v'' g n s = P x s ( t ′ ) n s h s v ′′ , and:
(a) if appending s s s to the word of P P P gives the fixed word of w s ws w s , the result is
P ( t 1 , … , t k , t ′ ) h s v ′ ′ ; P(t_1,\ldots,t_k,t')h^s v''; P ( t 1 , … , t k , t ′ ) h s v ′′ ; (b) if the word of P P P ends with s s s , so that P = P ′ x s ( t k ) n s P=P'x_s(t_k)n_s P = P ′ x s ( t k ) n s , the result is
{ P ′ ⋅ h s ( − 1 ) h s ⋅ x s ( χ s ( h s ) − 1 t ′ ) v ′ ′ , if t ′ = 0 , P ′ x s ( t k − t ′ − 1 ) n s ⋅ h s ( − t ′ ) h s ⋅ x s ( − χ s ( h s ) − 1 t ′ − 1 ) v ′ ′ , if t ′ ≠ 0 ; \begin{cases}
P'\cdot h_s(-1)h^s\cdot x_s\left(\chi_s(h^s)^{-1}t'\right)v'', & \text{if }t'=0,\\
P'x_s\left(t_k-t'^{-1}\right)n_s\cdot h_s(-t')h^s\cdot x_s\left(-\chi_s(h^s)^{-1}t'^{-1}\right)v'', & \text{if }t'\ne0;
\end{cases} { P ′ ⋅ h s ( − 1 ) h s ⋅ x s ( χ s ( h s ) − 1 t ′ ) v ′′ , P ′ x s ( t k − t ′ − 1 ) n s ⋅ h s ( − t ′ ) h s ⋅ x s ( − χ s ( h s ) − 1 t ′ − 1 ) v ′′ , if t ′ = 0 , if t ′ = 0 ; (c) in the remaining cases, which involve only w 0 w_0 w 0 , rewrite P P P in the other reduced word of w 0 w_0 w 0 and apply (b). Whenever (a) or (b) produces the word ( S R ) 3 (SR)^3 ( S R ) 3 , rewrite it back into the fixed word ( R S ) 3 (RS)^3 ( R S ) 3 , so that the result is again in normal form.
The rewriting in (c) is closed. Write P ( t 1 , … , t 6 ) = u n s 1 ⋯ n s 6 P(t_1,\ldots,t_6)=u\,n_{s_1}\cdots n_{s_6} P ( t 1 , … , t 6 ) = u n s 1 ⋯ n s 6 with
u = x s 1 ( t 1 ) φ s 1 ( x s 2 ( t 2 ) ) φ s 1 φ s 2 ( x s 3 ( t 3 ) ) ⋯ ∈ U . u=x_{s_1}(t_1)\varphi_{s_1}(x_{s_2}(t_2))\varphi_{s_1}\varphi_{s_2}(x_{s_3}(t_3))\cdots\in U. u = x s 1 ( t 1 ) φ s 1 ( x s 2 ( t 2 )) φ s 1 φ s 2 ( x s 3 ( t 3 )) ⋯ ∈ U . Replace n ( R S ) 3 n_{(RS)^3} n ( R S ) 3 by n ( S R ) 3 t 0 n_{(SR)^3}t_0 n ( S R ) 3 t 0 , or the reverse. Then split u u u along the new word, using Proposition 12.47 (e) for the first letter and ψ \psi ψ to move the remainder past n s 1 n_{s_1} n s 1 , and so on. The remainder after six letters is moved to the right past the torus factor by (a).
Proof. Existence: by Theorem 12.51 , g ∈ ( U ∩ U w ) n w B g\in(U\cap U_w)n_wB g ∈ ( U ∩ U w ) n w B . Splitting the left factor letter by letter, as in (c), puts u n w u n_w u n w into the form P ( t ) v 0 P(t)v_0 P ( t ) v 0 with v 0 ∈ U v_0\in U v 0 ∈ U . The digit widths are 3 and 1 along the word, so the number of tuples is ∑ w q ℓ ( w ) ∣ B ∣ = ∣ 3 D 4 ( q ) ∣ \sum_w q^{\ell(w)}|B|=|{}^3D_4(q)| ∑ w q ℓ ( w ) ∣ B ∣ = ∣ 3 D 4 ( q ) ∣ , and the expression is unique. For rule 3, h x s ( t ) n s = x s ( t ′ ) h n s = x s ( t ′ ) n s h s h\,x_s(t)\,n_s=x_s(t')\,h\,n_s=x_s(t')n_s h^s h x s ( t ) n s = x s ( t ′ ) h n s = x s ( t ′ ) n s h s and v ′ n s = n s ψ s ( v ′ ) v'n_s=n_s\psi_s(v') v ′ n s = n s ψ s ( v ′ ) . In case (a) nothing more is needed. In case (b), with t ′ = 0 t'=0 t ′ = 0 , we have n s n s = h s ( − 1 ) n_s n_s=h_s(-1) n s n s = h s ( − 1 ) , and h s ( − 1 ) h_s(-1) h s ( − 1 ) commutes with x s x_s x s because χ s ( h s ( − 1 ) ) = 1 \chi_s(h_s(-1))=1 χ s ( h s ( − 1 )) = 1 . With t ′ ≠ 0 t'\ne0 t ′ = 0 , Proposition 12.47(b),(c) give
n s x s ( t ′ ) n s = x − s ( − t ′ ) h s ( − 1 ) = x s ( − t ′ − 1 ) n s h s ( − t ′ ) x s ( − t ′ − 1 ) , n_s x_s(t') n_s=x_{-s}(-t')h_s(-1)=x_s(-t'^{-1})n_s h_s(-t')x_s(-t'^{-1}), n s x s ( t ′ ) n s = x − s ( − t ′ ) h s ( − 1 ) = x s ( − t ′ − 1 ) n s h s ( − t ′ ) x s ( − t ′ − 1 ) ,
and x s ( ⋅ ) h s = h s x s ( χ s ( h s ) − 1 ) x_s(\cdot)h^s=h^s x_s(\chi_s(h^s)^{-1}) x s ( ⋅ ) h s = h s x s ( χ s ( h s ) − 1 ) . Here P ′ P' P ′ is the fixed word of w s w_s w s , because ℓ ( w s ) < 6 \ell(w_s)<6 ℓ ( w s ) < 6 . The conjugates in (c) lie in U U U because the word is reduced, so each φ s i \varphi_{s_i} φ s i is applied to an element whose s i s_i s i -coordinate is 0. □
Corollary 12.49 (Closed digital law of 3 D 4 ( q ) {}^{3}D_4(q) 3 D 4 ( q ) ). Fix a basis of K K K over F q \mathbb{F}_q F q , write [ ξ ] [\xi] [ ξ ] for the base-q q q integer of the coordinates of ξ \xi ξ , and fix primitive elements μ \mu μ of K K K and γ \gamma γ of F q \mathbb{F}_q F q . For g = P ( t ) h ( μ i , γ j ) u ( a , b , c , d , e , f ) g=P(t)h(\mu^i,\gamma^j)u(a,b,c,d,e,f) g = P ( t ) h ( μ i , γ j ) u ( a , b , c , d , e , f ) in the cell of w w w , let
λ ( g ) = O w + [ t ] + q ℓ ( w ) ( i + ( q 3 − 1 ) j + ( q 3 − 1 ) ( q − 1 ) ( [ a ] + q 3 [ b ] + q 4 [ c ] + q 7 [ d ] + q 10 [ e ] + q 11 [ f ] ) ) . \lambda(g)=O_w+[t]+q^{\ell(w)}\left(i+(q^3-1)j+(q^3-1)(q-1)\left([a]+q^3[b]+q^4[c]+q^7[d]+q^{10}[e]+q^{11}[f]\right)\right). λ ( g ) = O w + [ t ] + q ℓ ( w ) ( i + ( q 3 − 1 ) j + ( q 3 − 1 ) ( q − 1 ) ( [ a ] + q 3 [ b ] + q 4 [ c ] + q 7 [ d ] + q 10 [ e ] + q 11 [ f ] ) ) . where [ t ] [t] [ t ] packs the digits of t t t in base q q q (3 digits for each R R R -letter, 1 for each S S S -letter), and O w = ∣ B ∣ ∑ w ′ ≺ w q ℓ ( w ′ ) O_w=|B|\sum_{w'\prec w}q^{\ell(w')} O w = ∣ B ∣ ∑ w ′ ≺ w q ℓ ( w ′ ) for a fixed order ≺ \prec ≺ on W W W . Then λ \lambda λ is a bijection onto A ∣ 3 D 4 ( q ) ∣ \mathcal{A}_{|{}^{3}D_4(q)|} A ∣ 3 D 4 ( q ) ∣ with λ ( I ) = 0 \lambda(I)=0 λ ( I ) = 0 , and m ∗ n = λ ( λ − 1 ( m ) λ − 1 ( n ) ) m*n=\lambda(\lambda^{-1}(m)\lambda^{-1}(n)) m ∗ n = λ ( λ − 1 ( m ) λ − 1 ( n )) is a digital law whose product is computed by Theorem 12.48 . The only operations used are field arithmetic, σ \sigma σ , N N N , Tr \operatorname{Tr} Tr , and discrete logarithms for the torus digits, as in Theorem 12.35 .
Proof. This follows from the uniqueness in Theorem 12.48 and Lemma 1.6. To multiply, expand the second factor as P ( t ′ ) h ′ v ′ P(t')h'v' P ( t ′ ) h ′ v ′ and apply rules 3, 2 and 1 letter by letter. At most six steps of type 3 occur, each possibly with one rewriting of P P P . □
Remark 12.50. For q = 2 , 3 , 4 q=2,3,4 q = 2 , 3 , 4 (F 8 \mathbb{F}_8 F 8 , F 27 \mathbb{F}_{27} F 27 , F 64 \mathbb{F}_{64} F 64 ), each formula of Propositions 12.46 and 12.47 was checked against matrix products on 40 random inputs: product, inverse, torus action, the diagonal of h ( x , y ) h(x,y) h ( x , y ) , n s − 1 h n s n_s^{-1}h n_s n s − 1 h n s , n s 2 n_s^2 n s 2 , φ s \varphi_s φ s , ψ s \psi_s ψ s , splitting, n s x s n s − 1 = x − s ( − t ) n_sx_sn_s^{-1}=x_{-s}(-t) n s x s n s − 1 = x − s ( − t ) and the rank-one relation. All held. The rules of Theorem 12.48 were then applied to 144 random normal forms for each simple reflection, 12 in each of the 12 cells, and compared with the matrix product. Rule 3(a) and both subcases of rule 3(b) occurred, and the rewriting between ( R S ) 3 (RS)^3 ( R S ) 3 and ( S R ) 3 (SR)^3 ( S R ) 3 was used several hundred times in each direction for each q q q . A walk of 1500 right multiplications done only with the closed rules agreed with the matrix product, and every result used the fixed word of its cell. Finally, for the label of Corollary 12.49, random normal forms were recovered exactly from their matrices, λ − 1 λ \lambda^{-1}\lambda λ − 1 λ and λ λ − 1 \lambda\lambda^{-1} λ λ − 1 were the identity on random labels and on both sides of every cell boundary, λ ( I ) = 0 \lambda(I)=0 λ ( I ) = 0 , the closed product agreed with the label of the matrix product on 120 random pairs, and m ∗ n m*n m ∗ n was associative on 300 random triples with 0 as identity.
Theorem 12.51 (Bruhat cells of 3 D 4 ( q ) {}^{3}D_4(q) 3 D 4 ( q ) ). The monomial elements n R n_R n R and n S n_S n S generate, modulo T T T , a dihedral group W W W of order 12. Let g ∈ 3 D 4 ( q ) g\in{}^{3}D_4(q) g ∈ 3 D 4 ( q ) , and let w w w and u u u be the permutation and upper factor given by the elimination of Theorem 12.12 . Then the permutation of some n w ∈ { n R , n S } n_w\in\{n_R,n_S\} n w ∈ { n R , n S } , u ∈ U ∩ U w u\in U\cap U_w u ∈ U ∩ U w , and n w − 1 u − 1 g ∈ B n_w^{-1}u^{-1}g\in B n w − 1 u − 1 g ∈ B . Thus the cell of w w w is ( U ∩ U w ) n w B (U\cap U_w)n_wB ( U ∩ U w ) n w B , and
∣ U ∩ U w ∣ = q ℓ ( w ) , ℓ ( w ) ∈ { 0 , 1 , 3 , 4 , 4 , 5 , 7 , 8 , 8 , 9 , 11 , 12 } , |U\cap U_w|=q^{\ell(w)},\qquad\ell(w)\in\{0,1,3,4,4,5,7,8,8,9,11,12\}, ∣ U ∩ U w ∣ = q ℓ ( w ) , ℓ ( w ) ∈ { 0 , 1 , 3 , 4 , 4 , 5 , 7 , 8 , 8 , 9 , 11 , 12 } , where ℓ ( w ) \ell(w) ℓ ( w ) is the sum of the widths (3 or 1) of the positive roots made negative by w − 1 w^{-1} w − 1 .
Proof. The group has a split ( B , N ) (B,N) ( B , N ) -pair with N = ⟨ T , n R , n S ⟩ N=\langle T,n_R,n_S\rangle N = ⟨ T , n R , n S ⟩ and relative Weyl group W ( G 2 ) W(G_2) W ( G 2 ) [21]. So every g g g has a unique expression g = u ′ n w b g=u'n_wb g = u ′ n w b with u ′ ∈ U ∩ U w u'\in U\cap U_w u ′ ∈ U ∩ U w and b ∈ B b\in B b ∈ B . This is a factorisation of the form in Theorem 12.12 , so by uniqueness u = u ′ u=u' u = u ′ and w w w is the permutation of n w n_w n w . Reduced words in W ( G 2 ) W(G_2) W ( G 2 ) alternate the two simple reflections. Each new inverted root therefore adds 3 or 1 in alternation, giving the partial sums 3, 4, 7, 8, 11, 12 and 1, 4, 5, 8, 9, 12. □
Summing over the cells gives
∑ w ∈ W q ℓ ( w ) = ( 1 + q ) ( 1 + q 3 ) ( 1 + q 4 + q 8 ) , ∣ 3 D 4 ( q ) ∣ = q 12 ( q 8 + q 4 + 1 ) ( q 6 − 1 ) ( q 2 − 1 ) . \sum_{w\in W}q^{\ell(w)}=(1+q)(1+q^3)(1+q^4+q^8),\qquad|{}^3D_4(q)|=q^{12}(q^8+q^4+1)(q^6-1)(q^2-1). w ∈ W ∑ q ℓ ( w ) = ( 1 + q ) ( 1 + q 3 ) ( 1 + q 4 + q 8 ) , ∣ 3 D 4 ( q ) ∣ = q 12 ( q 8 + q 4 + 1 ) ( q 6 − 1 ) ( q 2 − 1 ) .
Corollary 12.52 (Digital law of 3 D 4 ( q ) {}^3D_4(q) 3 D 4 ( q ) ). The construction of Theorem 12.18 , with Z = { I } Z=\{I\} Z = { I } and the cells of Theorem 12.51 , gives a digital law of 3 D 4 ( q ) {}^3D_4(q) 3 D 4 ( q ) on A q 12 ( q 8 + q 4 + 1 ) ( q 6 − 1 ) ( q 2 − 1 ) A_{q^{12}(q^8+q^4+1)(q^6-1)(q^2-1)} A q 12 ( q 8 + q 4 + 1 ) ( q 6 − 1 ) ( q 2 − 1 ) with identity 0 0 0 . A label is given by a Weyl digit in A 12 A_{12} A 12 , then ℓ ( w ) \ell(w) ℓ ( w ) base- q q q digits for U ∩ U w U\cap U_w U ∩ U w , then the Borel label in A q 12 ( q 3 − 1 ) ( q − 1 ) A_{q^{12}(q^3-1)(q-1)} A q 12 ( q 3 − 1 ) ( q − 1 ) .
Proof. Theorem 12.51 supplies the product structure of each cell required in the proof of Theorem 12.18 . The centre of 3 D 4 ( q ) {}^3D_4(q) 3 D 4 ( q ) is trivial.
Example 12.53. For q = 2 q=2 q = 2 , with F 8 = F 2 [ X ] / ( X 3 + X + 1 ) \mathbb{F}_8=\mathbb{F}_2[X]/(X^3+X+1) F 8 = F 2 [ X ] / ( X 3 + X + 1 ) , the computer gave the following.
All generators preserve Q Q Q , and x R ( ξ ) x R ( ζ ) = x R ( ξ + ζ ) x_R(\xi)x_R(\zeta)=x_R(\xi+\zeta) x R ( ξ ) x R ( ζ ) = x R ( ξ + ζ ) for all ξ , ζ ∈ F 8 \xi,\zeta\in\mathbb{F}_8 ξ , ζ ∈ F 8 .
The line ⟨ e 1 ⟩ \langle e_1\rangle ⟨ e 1 ⟩ has an orbit of 2457 = ( q 8 + q 4 + 1 ) ( q 3 + 1 ) 2457=(q^8+q^4+1)(q^3+1) 2457 = ( q 8 + q 4 + 1 ) ( q 3 + 1 ) points. On this orbit the permutation group generated by x ± R ( 1 ) x_{\pm R}(1) x ± R ( 1 ) , x ± R ( X ) x_{\pm R}(X) x ± R ( X ) , x ± R ( X 2 ) x_{\pm R}(X^2) x ± R ( X 2 ) and x ± S ( 1 ) x_{\pm S}(1) x ± S ( 1 ) has order 211 341 312 = ∣ 3 D 4 ( 2 ) ∣ 211\,341\,312=|{}^3D_4(2)| 211 341 312 = ∣ 3 D 4 ( 2 ) ∣ .
∣ U ∣ = 4096 |U|=4096 ∣ U ∣ = 4096 , ∣ T ∣ = 7 |T|=7 ∣ T ∣ = 7 and ∣ B ∣ = 28 672 |B|=28\,672 ∣ B ∣ = 28 672 . The elements n R n_R n R and n S n_S n S are monomial and give 12 permutations.
The counts ∣ U ∩ U w ∣ |U\cap U_w| ∣ U ∩ U w ∣ are 1 , 2 , 8 , 16 , 16 , 32 , 128 , 256 , 256 , 512 , 2048 , 4096 1,2,8,16,16,32,128,256,256,512,2048,4096 1 , 2 , 8 , 16 , 16 , 32 , 128 , 256 , 256 , 512 , 2048 , 4096 , with sum 7371 = ∣ G : B ∣ 7371=|G:B| 7371 = ∣ G : B ∣ .
For each w w w , 15 random elements u n w t u ′ u n_w t u' u n w t u ′ were reduced, with u , u ′ ∈ U u,u'\in U u , u ′ ∈ U and t ∈ T t\in T t ∈ T . The elimination returned w w w , a factor in U U U , and n w − 1 u ′ − 1 g ∈ B n_w^{-1}u'^{-1}g\in B n w − 1 u ′ − 1 g ∈ B .
On 300 random products of 40 generators, which met all 12 cells, the same held.
Order the cells by ℓ ( w ) \ell(w) ℓ ( w ) and then by the fixed word: ∅ \varnothing ∅ , S S S , R R R , R S RS R S , S R SR S R , S R S SRS S R S , R S R RSR R S R , R S R S RSRS R S R S , S R S R SRSR S R S R , S R S R S SRSRS S R S R S , R S R S R RSRSR R S R S R , ( R S ) 3 (RS)^3 ( R S ) 3 . The offsets O w O_w O w of Corollary 12.49 are then 0 0 0 , 28 672 28\,672 28 672 , 86 016 86\,016 86 016 , 315 392 315\,392 315 392 , 774 144 774\,144 774 144 , 1 232 896 1\,232\,896 1 232 896 , 2 150 400 2\,150\,400 2 150 400 , 5 820 416 5\,820\,416 5 820 416 , 13 160 448 13\,160\,448 13 160 448 , 20 500 480 20\,500\,480 20 500 480 , 35 180 544 35\,180\,544 35 180 544 , 93 900 800 93\,900\,800 93 900 800 , and the last cell ends at 211 341 312 = ∣ 3 D 4 ( 2 ) ∣ 211\,341\,312=|{}^3D_4(2)| 211 341 312 = ∣ 3 D 4 ( 2 ) ∣ .
For this order, λ − 1 λ \lambda^{-1}\lambda λ − 1 λ and λ λ − 1 \lambda\lambda^{-1} λ λ − 1 were the identity on random labels and at both ends of every cell, and the closed product agreed with the matrix product on random pairs.
Remark 12.54. Proposition 12.46 is a closed six-layer carry law, the analogue of the three-layer law for 2 G 2 ( q ) {}^2G_2(q) 2 G 2 ( q ) in § 20.8 . The digits a a a and b b b add without carry. The digit c c c receives the carry − b a ′ -ba' − b a ′ ; d d d receives carries of degree 2 and 3; e e e receives the norm and trace carries of degree up to 4; and f f f receives carries of degree up to 5, mixing all lower layers. The carries of width-3 digits enter F q \mathbb{F}_q F q -digits only through N N N and Tr \operatorname{Tr} Tr . With Proposition 12.47 and Theorem 12.48 , the whole law becomes closed (Corollary 12.49 ). The formulas hold in every characteristic, and all the checks above were run for q = 2 , 3 , 4 q=2,3,4 q = 2 , 3 , 4 .
The groups F 4 ( q ) F_4(q) F 4 ( q )
The group F 4 ( q ) F_4(q) F 4 ( q ) is the automorphism group of the split Albert algebra, the 27-dimensional exceptional Jordan algebra. We use the Albert model over a field of odd characteristic, where the Jordan product 1 2 ( X Y + Y X ) \frac{1}{2}(XY+YX) 2 1 ( X Y + Y X ) is available. The computations reported below were done over F 5 \mathbb{F}_5 F 5 and F 7 \mathbb{F}_7 F 7 . Definition 12.69 gives a form of the same model over Z \mathbb{Z} Z , with the quadratic operator U X U_X U X in place of X ∘ Y X\circ Y X ∘ Y ; it is used in characteristics 2 and 3.
THE GROUPS F 4 ( q ) F_4(q) F 4 ( q )
Definition 12.55 (Albert model). Let q q q be odd. The split octonions O \mathbb{O} O over F q \mathbb{F}_q F q are the quadruples ( a , v , w , b ) (a,v,w,b) ( a , v , w , b ) , with a , b ∈ F q a,b \in\mathbb{F}_q a , b ∈ F q and v , w ∈ F q 3 v,w \in\mathbb{F}_q^3 v , w ∈ F q 3 , under the Zorn product
( a , v , w , b ) ( a ′ , v ′ , w ′ , b ′ ) = ( a a ′ + v ⋅ w ′ , a v ′ + b ′ v + w × w ′ , a ′ w + b w ′ − v × v ′ , b b ′ + w ⋅ v ′ ) . (a,v,w,b)(a',v',w',b') = (aa' + v \cdot w', av' + b'v + w \times w', a'w + bw' - v \times v', bb' + w \cdot v'). ( a , v , w , b ) ( a ′ , v ′ , w ′ , b ′ ) = ( a a ′ + v ⋅ w ′ , a v ′ + b ′ v + w × w ′ , a ′ w + b w ′ − v × v ′ , b b ′ + w ⋅ v ′ ) . with conjugation ( a , v , w , b ) ‾ = ( b , − v , − w , a ) \overline{(a,v,w,b)} = (b,-v,-w,a) ( a , v , w , b ) = ( b , − v , − w , a ) . Let J J J be the space of Hermitian matrices
X = ( α 1 x 3 x ‾ 2 x ‾ 3 α 2 x 1 x 2 x ‾ 1 α 3 ) , α i ∈ F q , x i ∈ O , X = \begin{pmatrix} \alpha_1 & x_3 & \overline{x}_2 \\ \overline{x}_3 & \alpha_2 & x_1 \\ x_2 & \overline{x}_1 & \alpha_3 \end{pmatrix}, \qquad\alpha_i \in\mathbb{F}_q,\ x_i \in\mathbb{O}, X = α 1 x 3 x 2 x 3 α 2 x 1 x 2 x 1 α 3 , α i ∈ F q , x i ∈ O , with coordinates ( α 1 , α 2 , α 3 , x 1 , x 2 , x 3 ) ∈ F q 27 (\alpha_1,\alpha_2,\alpha_3,x_1,x_2,x_3) \in\mathbb{F}_q^{27} ( α 1 , α 2 , α 3 , x 1 , x 2 , x 3 ) ∈ F q 27 and product X ∘ Y = 1 2 ( X Y + Y X ) X \circ Y = \frac{1}{2}(XY+YX) X ∘ Y = 2 1 ( X Y + Y X ) . For a root derivation e α e_\alpha e α of J J J (Proposition 12.56 ), put x α ( t ) = exp ( t e α ) = I + t e α + 1 2 t 2 e α 2 x_\alpha(t) = \exp(te_\alpha) = I + te_\alpha+ \frac{1}{2}t^2e_\alpha^2 x α ( t ) = exp ( t e α ) = I + t e α + 2 1 t 2 e α 2 , and let F 4 ( q ) = ⟨ x α ( t ) ⟩ ≤ G L 27 ( q ) F_4(q) = \langle x_\alpha(t)\rangle\leq GL_{27}(q) F 4 ( q ) = ⟨ x α ( t )⟩ ≤ G L 27 ( q ) .
Proposition 12.56 (Root data of the Albert model). For q = 5 q = 5 q = 5 and q = 7 q = 7 q = 7 :
∘ \circ ∘ is commutative and satisfies the Jordan identity ( X ∘ Y ) ∘ X ∘ 2 = X ∘ ( Y ∘ X ∘ 2 ) (X \circ Y) \circ X^{\circ2} = X \circ(Y \circ X^{\circ2}) ( X ∘ Y ) ∘ X ∘ 2 = X ∘ ( Y ∘ X ∘ 2 ) .
The derivation algebra Der ( J ) \operatorname{Der}(J) Der ( J ) has dimension 52. Its diagonal part, in the coordinates above, is a 4-dimensional torus h \mathfrak{h} h . The coordinate vectors are weight vectors for h \mathfrak{h} h , with 25 distinct weights; the weight 0 has multiplicity 3, and each of the other 24 weights has multiplicity 1.
Der ( J ) = h ⊕ ⨁ α F q e α \operatorname{Der}(J) = \mathfrak{h} \oplus\bigoplus_\alpha\mathbb{F}_q e_\alpha Der ( J ) = h ⊕ ⨁ α F q e α over 48 roots α \alpha α , each root space being 1-dimensional. A generic linear functional gives 24 positive roots and 4 simple roots, with Cartan matrix
( ⟨ α j , α i ∨ ⟩ ) = ( 2 − 1 0 − 2 − 1 2 0 0 0 0 2 − 1 − 1 0 − 1 2 ) . (\langle\alpha_j,\alpha_i^\vee\rangle) =
\begin{pmatrix}
2 & -1 & 0 & -2 \\
-1 & 2 & 0 & 0 \\
0 & 0 & 2 & -1 \\
-1 & 0 & -1 & 2
\end{pmatrix}. (⟨ α j , α i ∨ ⟩) = 2 − 1 0 − 1 − 1 2 0 0 0 0 2 − 1 − 2 0 − 1 2 . that is, the Dynkin diagram α 2 − α 1 ⟸ α 4 − α 3 \alpha_2-\alpha_1 \Longleftarrow\alpha_4-\alpha_3 α 2 − α 1 ⟸ α 4 − α 3 of type F 4 F_4 F 4 , with the arrow pointing to the short roots α 1 , α 2 \alpha_1,\alpha_2 α 1 , α 2 .
e α 3 = 0 e_\alpha^3 = 0 e α 3 = 0 for every root. Every x α ( t ) x_\alpha(t) x α ( t ) is an automorphism of J J J , and for positive α \alpha α it is upper unitriangular when the coordinates are ordered by decreasing value of the functional on their weights.
With e − α e_{-\alpha} e − α normalised so that h α = [ e α , e − α ] h_\alpha= [e_\alpha,e_{-\alpha}] h α = [ e α , e − α ] satisfies α ( h α ) = 2 \alpha(h_\alpha) = 2 α ( h α ) = 2 , the elements n α = x α ( 1 ) x − α ( − 1 ) x α ( 1 ) n_\alpha= x_\alpha(1)x_{-\alpha}(-1)x_\alpha(1) n α = x α ( 1 ) x − α ( − 1 ) x α ( 1 ) for simple α \alpha α are automorphisms of J J J and map the weight space of λ \lambda λ onto that of s α ( λ ) s_\alpha(\lambda) s α ( λ ) . The four reflections generate a group of order 1152, and the maximal length is 24.
Proof. By computer, over F 5 \mathbb{F}_5 F 5 and F 7 \mathbb{F}_7 F 7 . The structure constants of ∘ \circ ∘ were computed from the matrix product. The derivation equations D ( X ∘ Y ) = D ( X ) ∘ Y + X ∘ D ( Y ) D(X \circ Y) = D(X) \circ Y + X \circ D(Y) D ( X ∘ Y ) = D ( X ) ∘ Y + X ∘ D ( Y ) are linear in the 729 entries of D D D . The diagonal ones reduce to integer equations d m = d i + d j d_m = d_i + d_j d m = d i + d j , whose solution space over Q \mathbb{Q} Q has dimension 4; this gives the integral weights. The equations are homogeneous for these weights, so Der ( J ) \operatorname{Der}(J) Der ( J ) was computed one weight class at a time. Parts (iv) and (v) were checked on the full 27 × 27 27 \times27 27 × 27 matrices, the automorphism property on all pairs of basis vectors. The Weyl group was generated by its action on the 48 roots. □ \square □
Theorem 12.57 (Bruhat cells of F 4 ( q ) F_4(q) F 4 ( q ) ). Let U = ∏ α > 0 x α ( F q ) U = \prod_{\alpha>0} x_\alpha(\mathbb{F}_q) U = ∏ α > 0 x α ( F q ) , in a fixed order, let T T T be the torus, and let B = U T B = UT B = U T . Then F 4 ( q ) = ⨆ w ∈ W ( U ∩ U w ) n w B F_4(q) = \bigsqcup_{w \in W}(U \cap U_w)n_wB F 4 ( q ) = ⨆ w ∈ W ( U ∩ U w ) n w B over the 1152 elements of W W W , the cell of w w w has q ℓ ( w ) ∣ B ∣ q^{\ell(w)}|B| q ℓ ( w ) ∣ B ∣ elements, and
∑ w ∈ W q ℓ ( w ) = ( q 2 − 1 ) ( q 6 − 1 ) ( q 8 − 1 ) ( q 12 − 1 ) ( q − 1 ) 4 , ∣ B ∣ = q 24 ( q − 1 ) 4 , \sum_{w \in W} q^{\ell(w)} = \frac{(q^2-1)(q^6-1)(q^8-1)(q^{12}-1)}{(q-1)^4}, \qquad|B| = q^{24}(q-1)^4, w ∈ W ∑ q ℓ ( w ) = ( q − 1 ) 4 ( q 2 − 1 ) ( q 6 − 1 ) ( q 8 − 1 ) ( q 12 − 1 ) , ∣ B ∣ = q 24 ( q − 1 ) 4 ,
∣ F 4 ( q ) ∣ = q 24 ( q 2 − 1 ) ( q 6 − 1 ) ( q 8 − 1 ) ( q 12 − 1 ) . \lvert F_4(q)\rvert= q^{24}(q^2-1)(q^6-1)(q^8-1)(q^{12}-1). ∣ F 4 ( q )∣ = q 24 ( q 2 − 1 ) ( q 6 − 1 ) ( q 8 − 1 ) ( q 12 − 1 ) .
Proof. By Proposition 12.56, F 4 ( q ) F_4(q) F 4 ( q ) is the Chevalley group of type F 4 F_4 F 4 in its 27-dimensional representation. It has a split B N BN B N -pair, and every element of U U U is a unique product of root elements in any fixed order [21]; so ∣ U ∣ = q 24 \lvert U\rvert= q^{24} ∣ U ∣ = q 24 . The Bruhat decomposition and the cell sizes follow. The length distribution computed in Proposition 12.56 (v) gives exactly the displayed polynomial. Its degrees are 2, 6, 8, 12.
Corollary 12.58 (Cell-wise digital law of F 4 ( q ) F_4(q) F 4 ( q ) ). Fix an order on W W W , a reduced word for each w w w , an order on the positive roots, and a primitive element γ \gamma γ of F q \mathbb{F}_q F q . Write g = P ( t ) h v g=P(t)\,h\,v g = P ( t ) h v as in Theorem 12.59 , with h = ∏ i = 1 4 h α i ( γ k i ) h=\prod_{i=1}^{4}h_{\alpha_i}(\gamma^{k_i}) h = ∏ i = 1 4 h α i ( γ k i ) and v = ∏ α > 0 x α ( s α ) v=\prod_{\alpha>0}x_\alpha(s_\alpha) v = ∏ α > 0 x α ( s α ) , the product taken in order of height. Put
λ ( g ) = O w + [ t ] + q ℓ ( w ) ( ∑ i = 1 4 k i ( q − 1 ) i − 1 + ( q − 1 ) 4 [ s ] ) , O w = ∣ B ∣ ∑ w ′ ≺ w q ℓ ( w ′ ) . \lambda(g)=O_w+[t]+q^{\ell(w)}\left(\sum_{i=1}^{4}k_i(q-1)^{i-1}+(q-1)^4[s]\right),\qquad O_w=\lvert B\rvert\sum_{w'\prec w}q^{\ell(w')}. λ ( g ) = O w + [ t ] + q ℓ ( w ) ( i = 1 ∑ 4 k i ( q − 1 ) i − 1 + ( q − 1 ) 4 [ s ] ) , O w = ∣ B ∣ w ′ ≺ w ∑ q ℓ ( w ′ ) . where [ t ] [t] [ t ] and [ s ] [s] [ s ] are the base-q q q integers of the digit strings ( t α ) (t_\alpha) ( t α ) and ( s α ) (s_\alpha) ( s α ) . Then λ \lambda λ is a bijection from F 4 ( q ) F_4(q) F 4 ( q ) onto A ∣ F 4 ( q ) ∣ A_{\lvert F_4(q)\rvert} A ∣ F 4 ( q )∣ with λ ( I ) = 0 \lambda(I)=0 λ ( I ) = 0 , and m ∗ n = λ ( λ − 1 ( m ) λ − 1 ( n ) ) m*n=\lambda(\lambda^{-1}(m)\lambda^{-1}(n)) m ∗ n = λ ( λ − 1 ( m ) λ − 1 ( n )) is a digital law of F 4 ( q ) F_4(q) F 4 ( q ) .
Proof. The normal form is unique by Theorem 12.59 . The torus digits are unique because ( λ i ) ↦ ∏ h α i ( λ i ) (\lambda_i)\mapsto\prod h_{\alpha_i}(\lambda_i) ( λ i ) ↦ ∏ h α i ( λ i ) is injective, the group being simply connected. The root digits are unique because products in U U U are unique [21]. The packing is Lemma 1.6 , as in Corollary 12.52 .
For a simple root s ∈ { 1 , 2 , 3 , 4 } s\in\{1,2,3,4\} s ∈ { 1 , 2 , 3 , 4 } put n s = x s ( 1 ) x − s ( − 1 ) x s ( 1 ) n_s=x_s(1)x_{-s}(-1)x_s(1) n s = x s ( 1 ) x − s ( − 1 ) x s ( 1 ) , h s ( t ) = n s ( t ) n s − 1 h_s(t)=n_s(t)n_s^{-1} h s ( t ) = n s ( t ) n s − 1 with n s ( t ) = x s ( t ) x − s ( − t − 1 ) x s ( t ) n_s(t)=x_s(t)x_{-s}(-t^{-1})x_s(t) n s ( t ) = x s ( t ) x − s ( − t − 1 ) x s ( t ) , and let χ s ( h ) \chi_s(h) χ s ( h ) be defined by h x s ( t ) h − 1 = x s ( χ s ( h ) t ) hx_s(t)h^{-1}=x_s(\chi_s(h)t) h x s ( t ) h − 1 = x s ( χ s ( h ) t ) for h ∈ T h\in T h ∈ T . For w ∈ W w\in W w ∈ W fix a reduced word s 1 ⋯ s k s_1\cdots s_k s 1 ⋯ s k (the first one found by a breadth-first search), and put P ( t 1 , … , t k ) = x s 1 ( t 1 ) n s 1 ⋯ x s k ( t k ) n s k P(t_1,\ldots,t_k)=x_{s_1}(t_1)n_{s_1}\cdots x_{s_k}(t_k)n_{s_k} P ( t 1 , … , t k ) = x s 1 ( t 1 ) n s 1 ⋯ x s k ( t k ) n s k .
Theorem 12.59 (Multiplication in F 4 ( q ) F_4(q) F 4 ( q ) by rewriting). Every g ∈ F 4 ( q ) g\in F_4(q) g ∈ F 4 ( q ) has a unique expression g = P ( t ) h v g=P(t)\,h\,v g = P ( t ) h v , with the fixed word of its Bruhat cell, h ∈ T h\in T h ∈ T and v ∈ U v\in U v ∈ U . Right multiplication of g g g is computed as follows.
By v ′ ∈ U v'\in U v ′ ∈ U : replace v v v by v v ′ vv' v v ′ .
By h ′ ∈ T h'\in T h ′ ∈ T : replace v v v by h ′ − 1 v h ′ h'^{-1}vh' h ′ − 1 v h ′ and h h h by h h ′ hh' h h ′ .
By n s n_s n s : if ℓ ( w s ) < ℓ ( w ) \ell(ws)<\ell(w) ℓ ( w s ) < ℓ ( w ) and the word does not end in s s s , first rewrite P P P along the reduced word (fixed word of w s ) ⋅ s ws)\cdot s w s ) ⋅ s . Then let t t t be the s s s -coordinate of v v v , and put v ′ ′ = n s − 1 x s ( − t ) v n s ∈ U v''=n_s^{-1}x_s(-t)vn_s\in U v ′′ = n s − 1 x s ( − t ) v n s ∈ U , t ′ = χ s ( h ) t t'=\chi_s(h)t t ′ = χ s ( h ) t and h s = n s − 1 h n s h^s=n_s^{-1}hn_s h s = n s − 1 h n s .
(a) If ℓ ( w s ) > ℓ ( w ) \ell(ws)>\ell(w) ℓ ( w s ) > ℓ ( w ) , the result is P ( t 1 , … , t k , t ′ ) h s v ′ ′ P(t_1,\ldots,t_k,t')h^sv'' P ( t 1 , … , t k , t ′ ) h s v ′′ .
(b) Otherwise P = P ′ x s ( t k ) n s P=P'x_s(t_k)n_s P = P ′ x s ( t k ) n s , and the result is P ′ ⋅ h s ( − 1 ) h s ⋅ x s ( χ s ( h s ) − 1 t k ) v ′ ′ P'\cdot h_s(-1)h^s\cdot x_s(\chi_s(h^s)^{-1}t_k)v'' P ′ ⋅ h s ( − 1 ) h s ⋅ x s ( χ s ( h s ) − 1 t k ) v ′′ if t ′ = 0 t'=0 t ′ = 0 , and P ′ x s ( t k − t ′ − 1 ) n s ⋅ h s ( − t ′ ) h s ⋅ x s ( − χ s ( h s ) − 1 t ′ − 1 ) v ′ ′ P'x_s(t_k-t'^{-1})n_s\cdot h_s(-t')h^s\cdot x_s(-\chi_s(h^s)^{-1}t'^{-1})v'' P ′ x s ( t k − t ′ − 1 ) n s ⋅ h s ( − t ′ ) h s ⋅ x s ( − χ s ( h s ) − 1 t ′ − 1 ) v ′′ if t ′ ≠ 0 t'\ne0 t ′ = 0 .
If the resulting word is not the fixed word of its element, rewrite it along the fixed word.
To rewrite P P P , given on a reduced word W W W , along another reduced word W ′ W' W ′ of the same element, put u = P n W − 1 ∈ U u=P\,n_W^{-1}\in U u = P n W − 1 ∈ U , where n W n_W n W is the product of the n s n_s n s along W W W . Split u = x s 1 ′ ( t 1 ′ ) u 1 u=x_{s'_1}(t'_1)u_1 u = x s 1 ′ ( t 1 ′ ) u 1 , replace u 1 n s 1 ′ u_1n_{s'_1} u 1 n s 1 ′ by n s 1 ′ ( n s 1 ′ − 1 u 1 n s 1 ′ ) n_{s'_1}(n_{s'_1}^{-1}u_1n_{s'_1}) n s 1 ′ ( n s 1 ′ − 1 u 1 n s 1 ′ ) , and continue along W ′ W' W ′ . The remainder r ∈ U r\in U r ∈ U left after the last letter is moved past the torus factor, giving P ′ ( t ′ ) h ( h − 1 r h ) v P'(t')h(h^{-1}rh)v P ′ ( t ′ ) h ( h − 1 r h ) v , because n W = n W ′ n_W=n_{W'} n W = n W ′ .
Proof. Uniqueness is the Bruhat decomposition of Theorem 12.57 , with each cell parametrised by the digits along the fixed word. The rules use only the Chevalley relations
n s x s ( t ) n s − 1 = x − s ( − t ) , x − s ( t ) = x s ( t − 1 ) n s h s ( − t ) x s ( t − 1 ) , n_sx_s(t)n_s^{-1}=x_{-s}(-t),\qquad x_{-s}(t)=x_s(t^{-1})n_sh_s(-t)x_s(t^{-1}), n s x s ( t ) n s − 1 = x − s ( − t ) , x − s ( t ) = x s ( t − 1 ) n s h s ( − t ) x s ( t − 1 ) ,
n s 2 = h s ( − 1 ) , χ s ( h s ( − 1 ) ) = 1 , n_s^2=h_s(-1),\qquad\chi_s(h_s(-1))=1, n s 2 = h s ( − 1 ) , χ s ( h s ( − 1 )) = 1 ,
all proved over Z \mathbb{Z} Z in Proposition 12.72 . The first is part (4) with γ = α s \gamma=\alpha_s γ = α s ; the second is part (6), rewritten with n s − 1 h s ( λ ) n s = h s ( λ − 1 ) n_s^{-1}h_s(\lambda)n_s=h_s(\lambda^{-1}) n s − 1 h s ( λ ) n s = h s ( λ − 1 ) from part (5); the last two are part (5). Otherwise the argument is the one in the proof of Theorem 12.48 . The factor t ′ = χ s ( h ) t t'=\chi_s(h)t t ′ = χ s ( h ) t comes from h x s ( t ) h − 1 = x s ( χ s ( h ) t ) hx_s(t)h^{-1}=x_s(\chi_s(h)t) h x s ( t ) h − 1 = x s ( χ s ( h ) t ) , again part (5). The rewriting uses u = P n W − 1 = ∏ k x β k ( ε k t k ) u=P_{n_W}^{-1}=\prod_k x_{\beta_k}(\varepsilon_k t_k) u = P n W − 1 = ∏ k x β k ( ε k t k ) , which is part (7), and n W = n W ′ n_W=n_{W'} n W = n W ′ , which is part (8) for the words that occur: fixed words, and (fixed word of w s w_s w s )⋅ s \cdot s ⋅ s . An element of U U U with s s s -coordinate 0 is a product of root elements at positive roots γ ≠ α s \gamma\ne\alpha_s γ = α s , and n s − 1 x γ ( t ) n s = x s ( γ ) ( ± t ) n_s^{-1}x_\gamma(t)n_s=x_{s(\gamma)}(\pm t) n s − 1 x γ ( t ) n s = x s ( γ ) ( ± t ) with s ( γ ) > 0 s(\gamma)>0 s ( γ ) > 0 by part (4); so every conjugate formed above lies in U U U .
Example 12.60. For q = 5 q=5 q = 5 (with γ = 2 \gamma=2 γ = 2 ) and q = 7 q=7 q = 7 (with γ = 3 \gamma=3 γ = 3 ), the computer gave the following.
dim Der ( J ) = 52 \dim\operatorname{Der}(J)=52 dim Der ( J ) = 52 with 48 roots, and ∣ W ∣ = 1152 |W|=1152 ∣ W ∣ = 1152 . For all four simple roots and all t ≠ 0 t\ne0 t = 0 : h s ( t ) h_s(t) h s ( t ) is diagonal and multiplicative in t t t , n s 2 = h s ( − 1 ) n_s^2=h_s(-1) n s 2 = h s ( − 1 ) , χ s ( h s ( − 1 ) ) = 1 \chi_s(h_s(-1))=1 χ s ( h s ( − 1 )) = 1 , n s x s ( t ) n s − 1 = x − s ( − t ) n_sx_s(t)n_s^{-1}=x_{-s}(-t) n s x s ( t ) n s − 1 = x − s ( − t ) , and the rank-one relation holds.
The map ( λ 1 , … , λ 4 ) ↦ ∏ h α i ( λ i ) (\lambda_1,\ldots,\lambda_4)\mapsto\prod h_{\alpha_i}(\lambda_i) ( λ 1 , … , λ 4 ) ↦ ∏ h α i ( λ i ) is injective on all ( q − 1 ) 4 (q-1)^4 ( q − 1 ) 4 tuples, so ∣ T ∣ = ( q − 1 ) 4 |T|=(q-1)^4 ∣ T ∣ = ( q − 1 ) 4 .
On 40 random pairs of normal forms, the product given by Theorem 12.59 equalled the 27 × 27 27\times27 27 × 27 matrix product every time. Each pair had a random element of W W W , random digits, a random torus element and a random v ∈ U v\in U v ∈ U . Every result was in the fixed word of its cell; the results met 34 cells for q = 5 q=5 q = 5 and 29 for q = 7 q=7 q = 7 .
In all rewritings (576 for q = 5 q=5 q = 5 , 586 for q = 7 q=7 q = 7 , counting the products recomputed in the label check below), n W = n W ′ n_W=n_{W'} n W = n W ′ and every remainder lay in U U U .
For the same 40 products, the label of Corollary 12.58 lay in A ∣ F 4 ( q ) ∣ \mathcal{A}_{|F_4(q)|} A ∣ F 4 ( q ) ∣ . Decoding it gave back the same matrix, and λ ( λ − 1 ( m 1 ) λ − 1 ( m 2 ) ) \lambda(\lambda^{-1}(m_1)\lambda^{-1}(m_2)) λ ( λ − 1 ( m 1 ) λ − 1 ( m 2 )) agreed with the label of the matrix product. Moreover, 40 random integers m m m satisfied λ ( λ − 1 ( m ) ) = m \lambda(\lambda^{-1}(m))=m λ ( λ − 1 ( m )) = m , and λ ( I ) = 0 \lambda(I)=0 λ ( I ) = 0 . The offsets O w O_w O w sum to ∣ F 4 ( q ) ∣ |F_4(q)| ∣ F 4 ( q ) ∣ .
In this implementation, the coordinates of U U U are read off from the matrices by stripping root factors in order of height. Products and conjugations in U U U are computed with 27 × 27 27\times27 27 × 27 matrices, not with closed formulas.
A closed law on the unipotent radical of F 4 F_4 F 4
Write a positive root as ∑ c i α i \sum c_i\alpha_i ∑ c i α i with label c 1 c 2 c 3 c 4 c_1c_2c_3c_4 c 1 c 2 c 3 c 4 , in the numbering of Proposition 12.56 . There α 1 , α 2 \alpha_1,\alpha_2 α 1 , α 2 are short, α 3 , α 4 \alpha_3,\alpha_4 α 3 , α 4 are long, and the highest root is 4223 (Bourbaki’s 2342). Order the positive roots by height, and within a height lexicographically.
Proposition 12.61 (Commutator table of F 4 F_4 F 4 ). The root derivations of Proposition 12.56 can be normalised to a Chevalley basis. Put e γ = [ e α , e β ] / ( r + 1 ) e_\gamma=[e_\alpha,e_\beta]/(r+1) e γ = [ e α , e β ] / ( r + 1 ) for the first simple α i \alpha_i α i with β = γ − α i \beta=\gamma-\alpha_i β = γ − α i a root, where r r r is the largest integer with β − r α i \beta-r\alpha_i β − r α i a root, and normalise e − γ e_{-\gamma} e − γ by γ ( [ e γ , e − γ ] ) = 2 \gamma([e_\gamma,e_{-\gamma}])=2 γ ([ e γ , e − γ ]) = 2 . In this basis, for positive roots α \alpha α before β \beta β ,
[ x α ( t ) , x β ( u ) ] = x α ( t ) − 1 x β ( u ) − 1 x α ( t ) x β ( u ) [x_\alpha(t),x_\beta(u)]=x_\alpha(t)^{-1}x_\beta(u)^{-1}x_\alpha(t)x_\beta(u) [ x α ( t ) , x β ( u )] = x α ( t ) − 1 x β ( u ) − 1 x α ( t ) x β ( u ) is the identity, or the product listed in Table 12.2. Factors at different roots there commute. The coefficients are integers in { ± 1 , ± 2 } \{\pm1,\pm2\} { ± 1 , ± 2 } and only t u tu t u , t 2 u t^2u t 2 u and t u 2 tu^2 t u 2 occur. For β \beta β before α \alpha α use [ x β ( u ) , x α ( t ) ] = [ x α ( t ) , x β ( u ) ] − 1 [x_\beta(u),x_\alpha(t)]=[x_\alpha(t),x_\beta(u)]^{-1} [ x β ( u ) , x α ( t )] = [ x α ( t ) , x β ( u ) ] − 1 . These relations hold in F 4 ( K ) F_4(K) F 4 ( K ) for every field K K K .
Proof. The basis was normalised over F 5 \mathbb{F}_5 F 5 and F 7 \mathbb{F}_7 F 7 as stated. There N α β = ± ( r + 1 ) N_{\alpha\beta}=\pm(r+1) N α β = ± ( r + 1 ) and N − α , − β = − N α β N_{-\alpha,-\beta}=-N_{\alpha\beta} N − α , − β = − N α β for all roots α , β \alpha,\beta α , β with α + β \alpha+\beta α + β a root, so it reduces a Chevalley basis, and its lift to Z \mathbb{Z} Z is the family D γ D_\gamma D γ of Proposition 12.70 . The relations are Proposition 12.72 (2), proved there as identities over Z [ t , u ] \mathbb{Z}[t,u] Z [ t , u ] ; they therefore hold over every commutative ring, in particular in F 4 ( K ) F_4(K) F 4 ( K ) for every field K K K . They agree with the table first read off from 27 × 27 27 \times27 27 × 27 matrices mod 5 and mod 7. □ \square □
Table 12.2. The 68 nontrivial commutators of positive root elements of F 4 F_4 F 4 .
Theorem 12.62 (Collection law on U U U ). Write u ( t ) = ∏ γ x γ ( t γ ) u(t)=\prod_\gamma x_\gamma(t_\gamma) u ( t ) = ∏ γ x γ ( t γ ) over the positive roots in the fixed order. To multiply u ( t ) u(t) u ( t ) on the right by x γ ( c ) x_\gamma(c) x γ ( c ) , let x δ 1 ( v 1 ) ⋅ … ⋅ x δ m ( v m ) x_{\delta_1}(v_1)\cdot\ldots\cdot x_{\delta_m}(v_m) x δ 1 ( v 1 ) ⋅ … ⋅ x δ m ( v m ) be the factors after γ \gamma γ . Replace t γ t_\gamma t γ by t γ + c t_\gamma+c t γ + c , delete those factors, and then multiply on the right, recursively, by
x δ 1 ( v 1 ) [ x δ 1 ( v 1 ) , x γ ( c ) ] ⋯ x δ m ( v m ) [ x δ m ( v m ) , x γ ( c ) ] , x_{\delta_1}(v_1)[x_{\delta_1}(v_1),x_\gamma(c)]\cdots x_{\delta_m}(v_m)[x_{\delta_m}(v_m),x_\gamma(c)], x δ 1 ( v 1 ) [ x δ 1 ( v 1 ) , x γ ( c )] ⋯ x δ m ( v m ) [ x δ m ( v m ) , x γ ( c )] , with the commutators taken from Table 12.2. Then u ( t ) u ( s ) = u ( r ) u(t)u(s)=u(r) u ( t ) u ( s ) = u ( r ) , where r r r is obtained by multiplying by x γ ( s γ ) x_\gamma(s_\gamma) x γ ( s γ ) for all γ \gamma γ in order. Each r γ = t γ + s γ + c γ ( t , s ) r_\gamma=t_\gamma+s_\gamma+c_\gamma(t,s) r γ = t γ + s γ + c γ ( t , s ) , with c γ c_\gamma c γ an integer polynomial in the coordinates of lower roots, valid over every field.
Proof. For δ \delta δ after γ \gamma γ , x γ ( − c ) x δ ( v ) x γ ( c ) = x δ ( v ) [ x δ ( v ) , x γ ( c ) ] x_{\gamma}(-c)x_{\delta}(v)x_{\gamma}(c)=x_{\delta}(v)[x_{\delta}(v),x_{\gamma}(c)] x γ ( − c ) x δ ( v ) x γ ( c ) = x δ ( v ) [ x δ ( v ) , x γ ( c )] . The new factors lie at roots strictly higher than δ \delta δ , so the recursion terminates. Integrality follows from Proposition 12.61. As an identity of polynomial maps, u ( t ) u ( s ) = u ( r ( t , s ) ) u(t)u(s)=u(r(t,s)) u ( t ) u ( s ) = u ( r ( t , s )) is Proposition 12.72 (3), proved over Z [ t , s ] \mathbb{Z}[t,s] Z [ t , s ] . By the injectivity in Proposition 12.72 (1), the polynomials produced by the recursion are the certified ones.
Expanded, the carries of heights 2 and 3 are
c 0011 = s 0001 t 0010 , c 1001 = s 0001 t 1000 , c 1100 = s 0100 t 1000 . c_{0011}=s_{0001}t_{0010},\qquad c_{1001}=s_{0001}t_{1000},\qquad c_{1100}=s_{0100}t_{1000}. c 0011 = s 0001 t 0010 , c 1001 = s 0001 t 1000 , c 1100 = s 0100 t 1000 .
c 1101 = − s 0001 s 0100 t 1000 − s 0001 t 1100 − s 0100 t 1001 , c_{1101}=-s_{0001}s_{0100}t_{1000}-s_{0001}t_{1100}-s_{0100}t_{1001}, c 1101 = − s 0001 s 0100 t 1000 − s 0001 t 1100 − s 0100 t 1001 ,
c 2001 = − 2 s 0001 s 1000 t 1000 − s 0001 t 1000 2 − 2 s 1000 t 1001 , c_{2001}=-2s_{0001}s_{1000}t_{1000}-s_{0001}t_{1000}^{2}-2s_{1000}t_{1001}, c 2001 = − 2 s 0001 s 1000 t 1000 − s 0001 t 1000 2 − 2 s 1000 t 1001 ,
c 1011 = − s 0001 s 0010 t 1000 − s 0001 s 1000 t 0010 − s 0001 t 0010 t 1000 − s 0010 t 1001 − s 1000 t 0011 . c_{1011}=-s_{0001}s_{0010}t_{1000}-s_{0001}s_{1000}t_{0010}-s_{0001}t_{0010}t_{1000}-s_{0010}t_{1001}-s_{1000}t_{0011}. c 1011 = − s 0001 s 0010 t 1000 − s 0001 s 1000 t 0010 − s 0001 t 0010 t 1000 − s 0010 t 1001 − s 1000 t 0011 .
The number of monomials in each c γ c_{\gamma} c γ is as follows.
Table 12.1 (PDF p. 179).
In total there are 2552 monomials, against an a priori bound of 24 260 from weights alone.
Example 12.63. The collection law of Theorem 12.62 , run on root coordinates only, agreed with the 27 × 27 27\times27 27 × 27 matrix product on 200 random pairs for each of q = 5 q=5 q = 5 and q = 7 q=7 q = 7 . It was associative, with u ( 0 ) u(0) u ( 0 ) as identity, on 150 random triples for each q ∈ { 2 , 3 , 5 , 7 , 11 , 13 } q\in\{2,3,5,7,11,13\} q ∈ { 2 , 3 , 5 , 7 , 11 , 13 } . This includes characteristics 2 and 3, where the Albert model of Definition 12.55 is not used. In characteristic 2 the relations with coefficient ± 2 \pm2 ± 2 become trivial, which reflects the special isogeny of F 4 F_{4} F 4 in characteristic 2. Expanding all 24 carries symbolically took about 6 seconds.
Proposition 12.64 (Weyl and torus actions on root coordinates). Let A = ( A i j ) = ( ⟨ α j , α i ∨ ⟩ ) A=(A_{ij})=(\langle\alpha_j,\alpha_i^\vee\rangle) A = ( A ij ) = (⟨ α j , α i ∨ ⟩) be the Cartan matrix of Proposition 12.56 , and for a root γ = ∑ j c j α j \gamma=\sum_j c_j\alpha_j γ = ∑ j c j α j put ⟨ γ , α i ∨ ⟩ = ∑ j c j A i j \langle\gamma,\alpha_i^\vee\rangle=\sum_j c_jA_{ij} ⟨ γ , α i ∨ ⟩ = ∑ j c j A ij . In the Chevalley basis of Proposition 12.61:
For h = ∏ i = 1 4 h α i ( λ i ) h=\prod_{i=1}^{4}h_{\alpha_i}(\lambda_i) h = ∏ i = 1 4 h α i ( λ i ) , h x γ ( t ) h − 1 = x γ ( γ ( h ) t ) hx_\gamma(t)h^{-1}=x_\gamma(\gamma(h)t) h x γ ( t ) h − 1 = x γ ( γ ( h ) t ) with γ ( h ) = ∏ i λ i ⟨ γ , α i ∨ ⟩ \gamma(h)=\prod_i\lambda_i^{\langle\gamma,\alpha_i^\vee\rangle} γ ( h ) = ∏ i λ i ⟨ γ , α i ∨ ⟩ .
n s − 1 h n s = ∏ i h α i ( λ i ′ ) n_s^{-1}hn_s=\prod_i h_{\alpha_i}(\lambda_i') n s − 1 h n s = ∏ i h α i ( λ i ′ ) , where λ i ′ = λ i \lambda_i'=\lambda_i λ i ′ = λ i for i ≠ s i\ne s i = s and λ s ′ = λ s − 1 ∏ i ≠ s λ i − A i s \lambda_s'=\lambda_s^{-1}\prod_{i\ne s}\lambda_i^{-A_{is}} λ s ′ = λ s − 1 ∏ i = s λ i − A i s .
For every root γ \gamma γ , n s − 1 x γ ( t ) n s = x s ( γ ) ( η s , γ t ) n_s^{-1}x_\gamma(t)n_s=x_{s(\gamma)}(\eta_{s,\gamma}t) n s − 1 x γ ( t ) n s = x s ( γ ) ( η s , γ t ) and n s x γ ( t ) n s − 1 = x s ( γ ) ( ( − 1 ) ⟨ γ , α s ∨ ⟩ η s , γ t ) n_sx_\gamma(t)n_s^{-1}=x_{s(\gamma)}((-1)^{\langle\gamma,\alpha_s^\vee\rangle}\eta_{s,\gamma}t) n s x γ ( t ) n s − 1 = x s ( γ ) (( − 1 ) ⟨ γ , α s ∨ ⟩ η s , γ t ) , with η s , γ = ± 1 \eta_{s,\gamma}=\pm1 η s , γ = ± 1 . For positive γ ≠ α s \gamma\ne\alpha_s γ = α s , η s , γ = − 1 \eta_{s,\gamma}=-1 η s , γ = − 1 exactly for the following roots:
Table 12.2 (PDF p. 179).
All of these are integral data, valid over every field.
Proof. All three parts are identities over Z \mathbb{Z} Z or over Laurent polynomial rings, proved in Proposition 12.72 . Part (1) follows from Proposition 12.72 (5), applied to each factor h α i ( λ i ) h_{\alpha_i}(\lambda_i) h α i ( λ i ) ; part (2) is Proposition 12.72 (5); part (3), with its sign table, is Proposition 12.72 (4). The signs agree with those first found by matching 27 × 27 27 \times27 27 × 27 matrices over F 5 \mathbb{F}_5 F 5 and F 7 \mathbb{F}_7 F 7 .
Theorem 12.65 (Multiplication of F 4 ( q ) F_4(q) F 4 ( q ) on coordinates). Represent g = P ( t ) h v g=P(t)hv g = P ( t ) h v by the data ( w , t , λ , r ) (w,t,\lambda,r) ( w , t , λ , r ) : the fixed word w w w , its digits t t t , the torus parameters λ ∈ ( F q × ) 4 \lambda\in(\mathbb{F}_q^\times)^4 λ ∈ ( F q × ) 4 with h = ∏ h α i ( λ i ) h=\prod h_{\alpha_i}(\lambda_i) h = ∏ h α i ( λ i ) , and the root coordinates r r r of v ∈ U v\in U v ∈ U . Then every step of Theorem 12.59 is a formula in these data. Products in U U U are given by the collection law of Theorem 12.62 . Conjugation of U U U by h h h and of T T T by n s n_s n s is given by Proposition 12.64 (1) and (2), and χ s ( h ) = α s ( h ) \chi_s(h)=\alpha_s(h) χ s ( h ) = α s ( h ) . Conjugating an element of U U U with s s s -coordinate 0 by n s n_s n s means applying Proposition 12.64 (3) factor by factor and recollecting. Finally, in the rewriting step,
P ( t ) n W − 1 = ∏ k = 1 ℓ x β k ( ε k t k ) , β k = s 1 ⋯ s k − 1 ( α s k ) , P(t)n_W^{-1}=\prod_{k=1}^{\ell}x_{\beta_k}(\varepsilon_k t_k),\qquad\beta_k=s_1\cdots s_{k-1}(\alpha_{s_k}), P ( t ) n W − 1 = k = 1 ∏ ℓ x β k ( ε k t k ) , β k = s 1 ⋯ s k − 1 ( α s k ) , where ε k \varepsilon_k ε k is the product of the signs of Proposition 12.64 (3) along s k − 1 , … , s 1 s_{k-1},\ldots,s_1 s k − 1 , … , s 1 . The resulting law uses only field arithmetic, and discrete logarithms for the torus digits of the label. Its integer tables are Table 12.2, the Cartan matrix and the signs η s , γ \eta_{s,\gamma} η s , γ .
Proof. Write P ( t ) = x s 1 ( t 1 ) ( n s 1 x s 2 ( t 2 ) n s 1 − 1 ) ⋯ n W P(t)=x_{s_1}(t_1)\left(n_{s_1}x_{s_2}(t_2)n_{s_1}^{-1}\right)\cdots n_W P ( t ) = x s 1 ( t 1 ) ( n s 1 x s 2 ( t 2 ) n s 1 − 1 ) ⋯ n W . The roots β k \beta_k β k are the positive roots inverted by w − 1 w^{-1} w − 1 , because the word is reduced. Everything else is a direct translation of Theorem 12.59 , using Propositions 12.61 and 12.64 .
Example 12.66. The law of Theorem 12.65 was run with no matrices, on random normal forms. For q = 5 q=5 q = 5 and q = 7 q=7 q = 7 , 20 random pairs each were compared with the 27 × 27 27\times27 27 × 27 matrix product of the corresponding matrices; all agreed, and every result was in its fixed word. For q = 2 q=2 q = 2 , 3 and 11, 6 random triples each were associative, and the identity was a right identity. The runs used 1031 rewritings, 933 steps of type (a) and 595 of type (b) (153 with t ′ = 0 t'=0 t ′ = 0 ); the whole test took about 10 seconds. For q = 2 q=2 q = 2 and q = 3 q=3 q = 3 this is only a consistency check of the integral law, since the Albert model of Definition 12.55 is not available there.
Proposition 12.67 (An integral model in all characteristics). Let g Z \mathfrak{g}_{\mathbb{Z}} g Z be the free Z \mathbb{Z} Z -module with basis the e α e_\alpha e α for the 48 roots and h 1 , … , h 4 h_1,\ldots,h_4 h 1 , … , h 4 , with brackets
[ h i , e α ] = ⟨ α , α i ∨ ⟩ e α , [ e α , e − α ] = h α , [ e α , e β ] = N α β e α + β [h_i,e_\alpha]=\langle\alpha,\alpha_i^\vee\rangle e_\alpha,\qquad[e_\alpha,e_{-\alpha}]=h_\alpha,\qquad[e_\alpha,e_\beta]=N_{\alpha\beta}e_{\alpha+\beta} [ h i , e α ] = ⟨ α , α i ∨ ⟩ e α , [ e α , e − α ] = h α , [ e α , e β ] = N α β e α + β when α + β \alpha+\beta α + β is a root, and [ e α , e β ] = 0 [e_\alpha,e_\beta]=0 [ e α , e β ] = 0 otherwise. Here h α = ∑ i k i ( α ) h i h_\alpha=\sum_i k_i(\alpha)h_i h α = ∑ i k i ( α ) h i is the coroot of α \alpha α , and the integers N α β = ± ( r + 1 ) N_{\alpha\beta}=\pm(r+1) N α β = ± ( r + 1 ) are the structure constants of the Chevalley basis of Proposition 12.61. Then the matrices
X α ( t ) = I + t ad ( e α ) + t 2 1 2 ad ( e α ) 2 ∈ GL 52 ( K ) X_\alpha(t)=I+t\operatorname{ad}(e_\alpha)+t^2\frac{1}{2}\operatorname{ad}(e_\alpha)^2\in\operatorname{GL}_{52}(K) X α ( t ) = I + t ad ( e α ) + t 2 2 1 ad ( e α ) 2 ∈ GL 52 ( K ) are defined over every field K K K , and they generate the Chevalley group F 4 ( K ) F_4(K) F 4 ( K ) . The relations of Propositions 12.61 and 12.64 hold in this model in every characteristic. Hence the law of Theorem 12.65 is the multiplication of F 4 ( q ) F_4(q) F 4 ( q ) for all q q q , including characteristics 2 and 3.
Proof. The brackets define the Chevalley Z \mathbb{Z} Z -form of the Lie algebra of type F 4 F_4 F 4 [21]. Root strings have length at most three, so ad ( e α ) 3 = 0 \operatorname{ad}(e_\alpha)^3=0 ad ( e α ) 3 = 0 . The entries of ad ( e α ) 2 \operatorname{ad}(e_\alpha)^2 ad ( e α ) 2 are even, so 1 2 ad ( e α ) 2 \frac{1}{2}\operatorname{ad}(e_\alpha)^2 2 1 ad ( e α ) 2 is integral. The X α ( t ) X_\alpha(t) X α ( t ) are therefore the root elements of the adjoint Chevalley construction over Z \mathbb{Z} Z , reduced to K K K . The commutator constants and the signs η s , γ \eta_{s,\gamma} η s , γ depend only on the Chevalley basis, not on the representation. So the integral relations found in the 27-dimensional model also hold in this one, in every characteristic. The centre of F 4 F_4 F 4 is trivial, so the adjoint and simply connected groups coincide.
Example 12.68. The constants N α β N_{\alpha\beta} N α β lifted from the models over F 5 \mathbb{F}_5 F 5 and over F 7 \mathbb{F}_7 F 7 are the same, and each coroot has a unique expression in h 1 , … , h 4 h_1,\ldots,h_4 h 1 , … , h 4 . For the resulting integer 52 × 52 52 \times52 52 × 52 matrices, the identity [ ad ( e α ) , ad ( e β ) ] = ad ( [ e α , e β ] ) [\operatorname{ad}(e_\alpha),\operatorname{ad}(e_\beta)]=\operatorname{ad}([e_\alpha,e_\beta]) [ ad ( e α ) , ad ( e β )] = ad ([ e α , e β ]) was checked for all 48 2 48^2 4 8 2 ordered pairs of roots. For all 48 roots, ad ( e α ) 2 \operatorname{ad}(e_\alpha)^2 ad ( e α ) 2 is even and ad ( e α ) 3 = 0 \operatorname{ad}(e_\alpha)^3=0 ad ( e α ) 3 = 0 . After reduction mod q q q , the law of Theorem 12.65 agreed with the 52 × 52 52 \times52 52 × 52 matrix product on 25 random pairs for each q ∈ { 2 , 3 , 5 , 7 } q\in\{2,3,5,7\} q ∈ { 2 , 3 , 5 , 7 } , and every result was in its fixed word. In characteristics 2 and 3 the law is therefore compared with an actual model, not only tested for associativity.
A second test covers every Weyl cell. For each of the 1152 fixed words w w w and each s s s , take a random normal form g g g in the cell of w w w and compare g ⋅ x s ( t ) n s g\cdot x_s(t)n_s g ⋅ x s ( t ) n s , as given by the law, with the product in the adjoint model. On the 2304 pairs ( w , s ) (w,s) ( w , s ) where w s ws w s is longer than w w w , there is one test. On the 2304 pairs where it is shorter, tests are repeated until both the case t ′ = 0 t'=0 t ′ = 0 and the case t ′ ≠ 0 t'\ne0 t ′ = 0 have occurred. Matrices are compared on a block of k k k random vectors, so a single wrong result escapes with probability at most q − k q^{-k} q − k . The last column bounds the chance that any error escaped.
Table 1 (PDF p. 181).
For q = 4 q=4 q = 4 , 8 and 9 the field is F p [ x ] / ( f ) \mathbb{F}_p[x]/(f) F p [ x ] / ( f ) with f = x 2 + x + 1 f=x^2+x+1 f = x 2 + x + 1 , x 3 + x + 1 x^3+x+1 x 3 + x + 1 and x 2 + 2 x + 2 x^2+2x+2 x 2 + 2 x + 2 respectively, and every coefficient operation of the law and of the model is carried out in this field. All tests agreed. Thus every branch of the rank-one rules is correct on every Weyl cell for q ≤ 9 q\le9 q ≤ 9 , in characteristics 2, 3, 5 and 7. The code is in scripts/f4_core.py, scripts/f4_law.py and scripts/f4_adjoint.py; the field arithmetic and the versions over F q \mathbb{F}_q F q are in scripts/gf.py, scripts/f4_lawq.py and scripts/f4_adjointq.py. For q q q prime the versions over F q \mathbb{F}_q F q agree with the prime-field code on all products and cells compared.
Definition 12.69 (Integral Albert algebra). Let O Z = Z 8 \mathbb{O}_{\mathbb{Z}}=\mathbb{Z}^8 O Z = Z 8 with the Zorn product of Definition 12.55 , which has integer coefficients, with conjugation ( a , v , w , b ) = ( b , − v , − w , a ) (a,v,w,b)=(b,-v,-w,a) ( a , v , w , b ) = ( b , − v , − w , a ) , norm n ( a , v , w , b ) = a b − v ⋅ w n(a,v,w,b)=ab-v\cdot w n ( a , v , w , b ) = ab − v ⋅ w , trace t ( a , v , w , b ) = a + b t(a,v,w,b)=a+b t ( a , v , w , b ) = a + b and polar form n ( x , y ) = n ( x + y ) − n ( x ) − n ( y ) = t ( x y ˉ ) n(x,y)=n(x+y)-n(x)-n(y)=t(x\bar{y}) n ( x , y ) = n ( x + y ) − n ( x ) − n ( y ) = t ( x y ˉ ) . Let J Z = Z 27 J_{\mathbb{Z}}=\mathbb{Z}^{27} J Z = Z 27 , with coordinates X = ( α 1 , α 2 , α 3 ; x 1 , x 2 , x 3 ) X=(\alpha_1,\alpha_2,\alpha_3;x_1,x_2,x_3) X = ( α 1 , α 2 , α 3 ; x 1 , x 2 , x 3 ) as in Definition 12.55 . Define
T ( X ) = α 1 + α 2 + α 3 , T ( X , Y ) = ∑ i α i β i + ∑ i n ( x i , y i ) , N ( X ) = α 1 α 2 α 3 − ∑ i α i n ( x i ) + t ( x 1 x 2 x 3 ) , X # = ( α 2 α 3 − n ( x 1 ) , α 3 α 1 − n ( x 2 ) , α 1 α 2 − n ( x 3 ) ; x ˉ 2 x ˉ 3 − α 1 x 1 , x ˉ 3 x ˉ 1 − α 2 x 2 , x ˉ 1 x ˉ 2 − α 3 x 3 ) , X × Y = ( X + Y ) # − X # − Y # , U X Y = T ( X , Y ) X − X # × Y , 1 = ( 1 , 1 , 1 ; 0 , 0 , 0 ) . \begin{aligned}
T(X)&=\alpha_1+\alpha_2+\alpha_3, \qquad T(X,Y)=\sum_i\alpha_i\beta_i+\sum_i n(x_i,y_i),\\
N(X)&=\alpha_1\alpha_2\alpha_3-\sum_i\alpha_i n(x_i)+t(x_1x_2x_3),\\
X^\#&=(\alpha_2\alpha_3-n(x_1),\ \alpha_3\alpha_1-n(x_2),\ \alpha_1\alpha_2-n(x_3);\\
&\qquad\bar{x}_2\bar{x}_3-\alpha_1x_1,\ \bar{x}_3\bar{x}_1-\alpha_2x_2,\ \bar{x}_1\bar{x}_2-\alpha_3x_3),\\
X\times Y&=(X+Y)^\#-X^\#-Y^\#, \qquad U_XY=T(X,Y)X-X^\#\times Y,\\
1&=(1,1,1;0,0,0).
\end{aligned} T ( X ) N ( X ) X # X × Y 1 = α 1 + α 2 + α 3 , T ( X , Y ) = i ∑ α i β i + i ∑ n ( x i , y i ) , = α 1 α 2 α 3 − i ∑ α i n ( x i ) + t ( x 1 x 2 x 3 ) , = ( α 2 α 3 − n ( x 1 ) , α 3 α 1 − n ( x 2 ) , α 1 α 2 − n ( x 3 ) ; x ˉ 2 x ˉ 3 − α 1 x 1 , x ˉ 3 x ˉ 1 − α 2 x 2 , x ˉ 1 x ˉ 2 − α 3 x 3 ) , = ( X + Y ) # − X # − Y # , U X Y = T ( X , Y ) X − X # × Y , = ( 1 , 1 , 1 ; 0 , 0 , 0 ) .
For a commutative ring R R R put J R = J Z ⊗ R J_R=J_{\mathbb{Z}}\otimes R J R = J Z ⊗ R , with the same formulas.
Proposition 12.70 (A quadratic Albert model in every characteristic). Let J Z J_{\mathbb{Z}} J Z , N N N , # \# # , T T T and U U U be as in Definition 12.69.
Over Z \mathbb{Z} Z : N ( 1 ) = 1 N(1)=1 N ( 1 ) = 1 , 1 # = 1 1^\#=1 1 # = 1 , ( X # ) # = N ( X ) X (X^\#)^\#=N(X)X ( X # ) # = N ( X ) X , T ( X # , Y ) = ∂ Y N ( X ) T(X^\#,Y)=\partial_YN(X) T ( X # , Y ) = ∂ Y N ( X ) , 1 × X = T ( X ) 1 − X 1\times X=T(X)1-X 1 × X = T ( X ) 1 − X , U 1 = I U_1=I U 1 = I and U U X Y = U X U Y U X U_{U_XY}=U_XU_YU_X U U X Y = U X U Y U X . Over Z [ 1 2 ] \mathbb{Z}[\frac{1}{2}] Z [ 2 1 ] , with ∘ \circ ∘ as in Definition 12.55 , T ( X , Y ) = T ( X ∘ Y ) T(X,Y)=T(X\circ Y) T ( X , Y ) = T ( X ∘ Y ) , U X Y = 2 X ∘ ( X ∘ Y ) − X 2 ∘ Y U_XY=2X\circ(X\circ Y)-X^2\circ Y U X Y = 2 X ∘ ( X ∘ Y ) − X 2 ∘ Y , X # = X ∘ 2 − T ( X ) X + T ( X # ) 1 X^\#=X^{\circ2}-T(X)X+T(X^\#)1 X # = X ∘ 2 − T ( X ) X + T ( X # ) 1 and X ∘ 3 − T ( X ) X ∘ 2 + T ( X # ) X − N ( X ) 1 = 0 X^{\circ3}-T(X)X^{\circ2}+T(X^\#)X-N(X)1=0 X ∘ 3 − T ( X ) X ∘ 2 + T ( X # ) X − N ( X ) 1 = 0 .
The root derivations e α e_{\alpha} e α of Proposition 12.56 lift to integer matrices D α D_{\alpha} D α , and the lifts from F 5 \mathbb{F}_{5} F 5 and from F 7 \mathbb{F}_{7} F 7 coincide. For every root, D α 1 = 0 D_{\alpha}1=0 D α 1 = 0 , D α D_{\alpha} D α leaves T ( ⋅ , ⋅ ) T(\cdot,\cdot) T ( ⋅ , ⋅ ) and N N N infinitesimally invariant, D α 3 = 0 D_{\alpha}^{3}=0 D α 3 = 0 and D α 2 ∈ 2 M 27 ( Z ) D_{\alpha}^{2}\in2M_{27}(\mathbb{Z}) D α 2 ∈ 2 M 27 ( Z ) ; D α 2 ≠ 0 D_{\alpha}^{2}\ne0 D α 2 = 0 exactly for the 24 short roots. For the 816 ordered pairs with α + β \alpha+\beta α + β a root, [ D α , D β ] = N α β D α + β [D_{\alpha},D_{\beta}]=N_{\alpha\beta}D_{\alpha+\beta} [ D α , D β ] = N α β D α + β with the constants of Proposition 12.67 , signs included; [ D α , D β ] = 0 [D_{\alpha},D_{\beta}]=0 [ D α , D β ] = 0 if α + β \alpha+\beta α + β is neither a root nor 0; and h α = [ D α , D − α ] h_{\alpha}=[D_{\alpha},D_{-\alpha}] h α = [ D α , D − α ] is diagonal with [ h α , D α ] = 2 D α [h_{\alpha},D_{\alpha}]=2D_{\alpha} [ h α , D α ] = 2 D α .
The matrices x α ( t ) = I + t D α + t 2 D α ( 2 ) x_{\alpha}(t)=I+tD_{\alpha}+t^{2}D_{\alpha}^{(2)} x α ( t ) = I + t D α + t 2 D α ( 2 ) , D α ( 2 ) = 1 2 D α 2 D_{\alpha}^{(2)}=\frac{1}{2}D_{\alpha}^{2} D α ( 2 ) = 2 1 D α 2 , are polynomial in t t t with integer coefficients, fix 1, and preserve N N N , T ( ⋅ , ⋅ ) T(\cdot,\cdot) T ( ⋅ , ⋅ ) and U U U .
Over F 2 \mathbb{F}_{2} F 2 , T ( 1 ) = 1 T(1)=1 T ( 1 ) = 1 , so J = F 2 1 ⊕ ker T J=\mathbb{F}_{2}1\oplus\ker T J = F 2 1 ⊕ ker T , and ker T \ker T ker T is absolutely irreducible of dimension 26 under the group generated by the x α ( t ) x_{\alpha}(t) x α ( t ) . Over F 3 \mathbb{F}_{3} F 3 , 1 ∈ ker T 1\in\ker T 1 ∈ ker T , and ker T / F 3 1 \ker T/\mathbb{F}_{3}1 ker T / F 3 1 is absolutely irreducible of dimension 25.
Proof. By computer, in exact integer arithmetic. The identities in (i) and the invariance statements in (ii) and (iii) are polynomial identities of bounded degree, tested at random integer points; in (iii), for each root and each test point, at seven values of t t t , which determines N ( x α ( t ) X ) N(x_{\alpha}(t)X) N ( x α ( t ) X ) as a polynomial of degree at most 6 in t t t . The other statements in (ii) are exact matrix identities over Z \mathbb{Z} Z , and the signs of the constants were compared with the 52 × 52 52\times52 52 × 52 model. For (iv), the F p \mathbb{F}_{p} F p -span of all products of the generators x ± α i ( 1 ) x_{\pm\alpha_i}(1) x ± α i ( 1 ) , acting on the module, has dimension 26 2 = 676 26^{2}=676 2 6 2 = 676 for p = 2 p=2 p = 2 and 25 2 = 625 25^{2}=625 2 5 2 = 625 for p = 3 p=3 p = 3 , so the module is absolutely irreducible by Burnside’s theorem.
By (i) and [24], ( J R , U , 1 ) (J_R,U,1) ( J R , U , 1 ) is a quadratic Jordan algebra for every commutative ring R R R , and by (iii) the reductions of the x α ( t ) x_{\alpha}(t) x α ( t ) modulo p p p are automorphisms of J F q J_{\mathbb{F}_{q}} J F q for every q q q . This replaces the product 1 2 ( X Y + Y X ) \frac{1}{2}(XY+YX) 2 1 ( X Y + Y X ) , which is not defined in characteristic 2; it is not integral on J Z J_{\mathbb{Z}} J Z either. That the automorphism group of J K J_K J K , which is also the stabiliser of 1 and N N N , is the split group of type F 4 F_{4} F 4 in every characteristic is taken from [23], not checked.
Example 12.71. The cell sweep of Example 12.68 was repeated with the 27 × 27 27\times27 27 × 27 matrices x α ( t ) x_{\alpha}(t) x α ( t ) of Proposition 12.70 , reduced mod p p p , in place of the 52 × 52 52\times52 52 × 52 adjoint model, for every q ≤ 9 q\le9 q ≤ 9 in characteristics 2 and 3, and for q = 11 q=11 q = 11 and q = 16 q=16 q = 16 . This gives a second model of the group in these characteristics, independent of the adjoint one. The columns are as in Example 12.68. For q = 11 q=11 q = 11 and q = 16 q=16 q = 16 , only the 27-dimensional model was used. The F 16 \mathbb{F}_{16} F 16 sweep ran in four pieces with independent random streams; 95 cells were tested twice, and the counts include them.
Table 12.71.
All tests agreed. The code is in scripts/f4_jordan.py and scripts/f4_jordang.py; the sweep is verify-cells-q of scripts/f4_adjointq.py, unchanged.
Proposition 12.72 (Exact certificates for the unipotent, torus and Weyl data). Let D γ D_{\gamma} D γ , for γ \gamma γ a root, be the integer matrices of Proposition 12.70 , let x γ ( t ) = I + t D γ + t 2 D γ ( 2 ) x_{\gamma}(t)=I+tD_{\gamma}+t^{2}D_{\gamma}^{(2)} x γ ( t ) = I + t D γ + t 2 D γ ( 2 ) , and let X ( t ) = ∏ γ x γ ( t γ ) X(t)=\prod_{\gamma}x_{\gamma}(t_{\gamma}) X ( t ) = ∏ γ x γ ( t γ ) over the positive roots in the fixed order. Put n s = x α s ( 1 ) x − α s ( − 1 ) x α s ( 1 ) n_s=x_{\alpha_s}(1)x_{-\alpha_s}(-1)x_{\alpha_s}(1) n s = x α s ( 1 ) x − α s ( − 1 ) x α s ( 1 ) , n s ( λ ) = x α s ( λ ) x − α s ( − λ − 1 ) x α s ( λ ) n_s(\lambda)=x_{\alpha_s}(\lambda)x_{-\alpha_s}(-\lambda^{-1})x_{\alpha_s}(\lambda) n s ( λ ) = x α s ( λ ) x − α s ( − λ − 1 ) x α s ( λ ) and h s ( λ ) = n s ( λ ) n s − 1 h_s(\lambda)=n_s(\lambda)n_s^{-1} h s ( λ ) = n s ( λ ) n s − 1 . The following are identities of matrices over the rings indicated, and hence hold over every commutative ring.
The groups F 4 ( q ) F_4(q) F 4 ( q )
For every positive root γ \gamma γ there is an entry ( i γ , j γ ) (i_\gamma,j_\gamma) ( i γ , j γ ) with D γ [ i γ , j γ ] = ϵ γ = ± 1 D_\gamma[i_\gamma,j_\gamma]=\epsilon_\gamma=\pm1 D γ [ i γ , j γ ] = ϵ γ = ± 1 such that the following decoding, applied to M = X ( z ) M=X(z) M = X ( z ) over Z [ z ] \mathbb{Z}[z] Z [ z ] , returns r = z r=z r = z and ends at I I I : for γ \gamma γ in order, put r γ = ϵ γ M i γ j γ r_\gamma=\epsilon_\gamma M_{i_\gamma j_\gamma} r γ = ϵ γ M i γ j γ and replace M M M by x γ ( − r γ ) M x_\gamma(-r_\gamma)M x γ ( − r γ ) M . In particular t ↦ X ( t ) t\mapsto X(t) t ↦ X ( t ) is injective over every ring.
For all 552 ordered pairs of distinct positive roots, x α ( − a ) x β ( − b ) x α ( a ) x β ( b ) = X ( c ) x_\alpha(-a)x_\beta(-b)x_\alpha(a)x_\beta(b)=X(c) x α ( − a ) x β ( − b ) x α ( a ) x β ( b ) = X ( c ) over Z [ a , b ] \mathbb{Z}[a,b] Z [ a , b ] , where each c γ c_\gamma c γ is 0 or a single monomial C a i b j Ca^ib^j C a i b j . These are the relations of Table 12.2.
X ( t ) X ( s ) = X ( r ( t , s ) ) X(t)X(s)=X(r(t,s)) X ( t ) X ( s ) = X ( r ( t , s )) over Z [ t , s ] \mathbb{Z}[t,s] Z [ t , s ] , where r γ = t γ + s γ + c γ r_\gamma=t_\gamma+s_\gamma+c_\gamma r γ = t γ + s γ + c γ and the c γ c_\gamma c γ have 2552 monomials in total, distributed by height as in the table after Theorem 12.62 .
n s n_s n s has an integral inverse, and for all 48 roots γ \gamma γ and all s s s , over Z [ t ] \mathbb{Z}[t] Z [ t ] , n s − 1 x γ ( t ) n s = x s ( γ ) ( η s , γ t ) n_s^{-1}x_\gamma(t)n_s=x_{s(\gamma)}(\eta_{s,\gamma}t) n s − 1 x γ ( t ) n s = x s ( γ ) ( η s , γ t ) and n s x γ ( t ) n s − 1 = x s ( γ ) ( ( − 1 ) ⟨ γ , α s ∨ ⟩ η s , γ t ) n_sx_\gamma(t)n_s^{-1}=x_{s(\gamma)}((-1)^{\langle\gamma,\alpha_s^\vee\rangle}\eta_{s,\gamma}t) n s x γ ( t ) n s − 1 = x s ( γ ) (( − 1 ) ⟨ γ , α s ∨ ⟩ η s , γ t ) . The signs are those of Proposition 12.64 (3). In particular η s , α s = − 1 \eta_{s,\alpha_s}=-1 η s , α s = − 1 , so n s x α s ( t ) n s − 1 = x − α s ( − t ) n_sx_{\alpha_s}(t)n_s^{-1}=x_{-\alpha_s}(-t) n s x α s ( t ) n s − 1 = x − α s ( − t ) .
Over Z [ λ ± 1 ] \mathbb{Z}[\lambda^{\pm1}] Z [ λ ± 1 ] , h s ( λ ) h_s(\lambda) h s ( λ ) is diagonal with monomial entries, and h s ( λ ) x γ ( t ) h s ( λ ) − 1 = x γ ( λ ⟨ γ , α s ∨ ⟩ t ) h_s(\lambda)x_\gamma(t)h_s(\lambda)^{-1}=x_\gamma(\lambda^{\langle\gamma,\alpha_s^\vee\rangle}t) h s ( λ ) x γ ( t ) h s ( λ ) − 1 = x γ ( λ ⟨ γ , α s ∨ ⟩ t ) for all 48 roots. Moreover n s 2 = h s ( − 1 ) n_s^2=h_s(-1) n s 2 = h s ( − 1 ) , and over Z [ λ 1 ± 1 , … , λ 4 ± 1 ] \mathbb{Z}[\lambda_1^{\pm1},\ldots,\lambda_4^{\pm1}] Z [ λ 1 ± 1 , … , λ 4 ± 1 ] , n s − 1 h n s = ∏ i h α i ( λ i ) n_s^{-1}h n_s=\prod_i h_{\alpha_i}(\lambda_i) n s − 1 h n s = ∏ i h α i ( λ i ) with λ i \lambda_i λ i as in Proposition 12.64 (2).
Over Z [ u ± 1 ] \mathbb{Z}[u^{\pm1}] Z [ u ± 1 ] , x − α s ( u ) = x α s ( u − 1 ) h s ( − u − 1 ) n s x α s ( u − 1 ) x_{-\alpha_s}(u)=x_{\alpha_s}(u^{-1})h_s(-u^{-1})n_sx_{\alpha_s}(u^{-1}) x − α s ( u ) = x α s ( u − 1 ) h s ( − u − 1 ) n s x α s ( u − 1 ) .
For each of the 1152 fixed words s 1 ⋯ s ℓ s_1\cdots s_\ell s 1 ⋯ s ℓ and each k k k , the following identity holds over Z [ t ] \mathbb{Z}[t] Z [ t ] : n s 1 ⋯ n s k − 1 x α s k ( t ) ( n s 1 ⋯ n s k − 1 ) − 1 = x β k ( ϵ k t ) n_{s_1}\cdots n_{s_{k-1}}x_{\alpha_{s_k}}(t)(n_{s_1}\cdots n_{s_{k-1}})^{-1}=x_{\beta_k}(\epsilon_k t) n s 1 ⋯ n s k − 1 x α s k ( t ) ( n s 1 ⋯ n s k − 1 ) − 1 = x β k ( ϵ k t ) , where β k = s 1 ⋯ s k − 1 ( α s k ) \beta_k=s_1\cdots s_{k-1}(\alpha_{s_k}) β k = s 1 ⋯ s k − 1 ( α s k ) and ϵ k \epsilon_k ϵ k is the product of the signs ( − 1 ) ⟨ γ , α s j ∨ ⟩ η s j , γ (-1)^{\langle\gamma,\alpha_{s_j}^\vee\rangle}\eta_{s_j,\gamma} ( − 1 ) ⟨ γ , α s j ∨ ⟩ η s j , γ of (4) met along s k − 1 , … , s 1 s_{k-1},\ldots,s_1 s k − 1 , … , s 1 . The β k \beta_k β k are distinct and form the inversion set of w w w , and P ( t ) n W − 1 = ∏ k x β k ( ϵ k t k ) P(t)n_W^{-1}=\prod_k x_{\beta_k}(\epsilon_k t_k) P ( t ) n W − 1 = ∏ k x β k ( ϵ k t k ) .
For every fixed word W W W of an element w w w and every s s s with ℓ ( w s ) > ℓ ( w ) \ell(ws)>\ell(w) ℓ ( w s ) > ℓ ( w ) , n W n s = n W ′ n_Wn_s=n_{W'} n W n s = n W ′ over Z \mathbb{Z} Z , where W ′ W' W ′ is the fixed word of w s ws w s and n W n_W n W is the product of the n s n_s n s along $W.
Proof. By exact computation with integer and Laurent polynomials. Since D γ 3 = 0 D_\gamma^3=0 D γ 3 = 0 and D γ 2 = 2 D γ ( 2 ) D_\gamma^2=2D_\gamma^{(2)} D γ 2 = 2 D γ ( 2 ) , the matrix x γ ( − r ) x_\gamma(-r) x γ ( − r ) is the inverse of x γ ( r ) x_\gamma(r) x γ ( r ) . In (2) and (3) the decoding of (1) is applied to the left-hand side; it reads off the coordinates and ends exactly at I I I , and this proves the identity. The remaining statements are direct matrix identities. The data, namely the matrices, the entries ( i γ , j γ ) (i_\gamma,j_\gamma) ( i γ , j γ ) , the polynomials of (2) and (3), the signs and the weights, are stored in the certificates certificates/f4_ucert.json and certificates/f4_wcert.json. These are written by scripts/f4_ucert.py and scripts/f4_wcert.py and checked by scripts/f4_ucert_verify.py and scripts/f4_wcert_verify.py. The verifiers read only the certificates and import nothing else from the project; together they run in about a second. The last formula in (7) follows by telescoping, since P ( t ) = ∏ k ( N k − 1 x α s k ( t k ) N k − 1 − 1 ) ⋅ n W P(t)=\prod_k(N_{k-1}x_{\alpha_{s_k}}(t_k)N_{k-1}^{-1})\cdot n_W P ( t ) = ∏ k ( N k − 1 x α s k ( t k ) N k − 1 − 1 ) ⋅ n W with N k − 1 = n s 1 ⋯ n s k − 1 N_{k-1}=n_{s_1}\cdots n_{s_{k-1}} N k − 1 = n s 1 ⋯ n s k − 1 . Parts (7) and (8), 13824 conjugation identities and 2304 braid identities, are checked by scripts/f4_wordcert_verify.py against certificates/f4_wordcert.json in about three seconds.
Proposition 12.72 concerns the matrices x γ ( t ) x_\gamma(t) x γ ( t ) over a ring R R R . That over a field K K K they generate F 4 ( K ) F_4(K) F 4 ( K ) rests on Proposition 12.70 and [23]. Theorem 12.59 uses no other identities; its proof combines parts (3)–(8) by hand. What remains computational is only that the implementation follows the rules of the theorem. This is checked exactly, for every field at once, in Proposition 12.73 ; the random cell sweeps of Examples 12.68 and 12.71 remain as independent evidence.
Proposition 12.73 (The implemented law agrees with the model over every field). Let the digits d 1 , … , d 24 d_1,\ldots,d_{24} d 1 , … , d 24 , the torus coordinates λ 1 , … , λ 4 \lambda_1,\ldots,\lambda_4 λ 1 , … , λ 4 and the coordinates u β u_\beta u β (β ∈ Φ + \beta\in\Phi^+ β ∈ Φ + ) be independent variables, and let R = Z [ d i , λ i ± 1 , u β ] R=\mathbb{Z}[d_i,\lambda_i^{\pm1},u_\beta] R = Z [ d i , λ i ± 1 , u β ] . For each of the 1152 fixed words w w w and each simple
reflection s s s , run the implemented law (mul_root, convert and mul_n in scripts/f4_lawq.py) on the generic normal form with these coordinates: in branch (a) as it stands, in branch (b0) with u α s = 0 u_{\alpha_s}=0 u α s = 0 , and in branch (b1) over R [ u α s − 1 ] R[u_{\alpha_s}^{-1}] R [ u α s − 1 ] . Then the output is a normal form over this ring, and its image in the 27-dimensional representation of Proposition 12.70 equals the image of the input times n s n_s n s . The same holds for the right multiplication by a generic element h v hv h v of B B B in multiply. Consequently, over every finite field F q \mathbb{F}_q F q , the implemented product of any two normal forms is their product in F 4 ( q ) F_4(q) F 4 ( q ) .
Proof. The check is scripts/f4_symbolic_model.py. It runs the unchanged law code over Laurent polynomials and applies the model matrices x β ( t ) = I + t E β + t 2 E β 2 / 2 x_\beta(t)=I+tE_\beta+t^2E_\beta^2/2 x β ( t ) = I + t E β + t 2 E β 2 /2 of Proposition 12.70 to a vector of 27 further variables, so each comparison is an identity of 27 × 27 27\times27 27 × 27 matrices. It verifies the root step mul_root(u, α_s,c) = u x_{α_s}(c) for all s s s , the identity mul_n(f,s) = f n_s for all 6912 pairs of a cell and a branch (2304 each of (a), (b0), (b1)), and convert(f,w′) = f for the 2304 downward edges, where w ′ w' w ′ is the reduced word ending in s s s . All 9216 identities hold exactly. The right multiplication by h v hv h v does not depend on the cell and is checked once, with h h h and v v v generic, by scripts/f4_symbolic_tail.py. The code inverts only t ′ = χ s ( λ ) t t'=\chi_s(\lambda)t t ′ = χ s ( λ ) t and torus characters, always monomials with coefficient ± 1 \pm1 ± 1 , and every output coefficient is an integer (scripts/f4_symbolic_integrality.py); so the identities hold over R R R and R [ u α s − 1 ] R[u_{\alpha_s}^{-1}] R [ u α s − 1 ] .
For F q \mathbb{F}_q F q , specialise along a ring homomorphism to F q \mathbb{F}_q F q . The branch taken by the field code depends only on whether the length goes up and whether t = 0 t=0 t = 0 , and t ′ = 0 t'=0 t ′ = 0 exactly when t = 0 t=0 t = 0 since χ s ( λ ) ≠ 0 \chi_s(\lambda)\ne0 χ s ( λ ) = 0 . In branch (b1) the variable u α s u_{\alpha_s} u α s maps to a unit, so the identity over R [ u α s − 1 ] R[u_{\alpha_s}^{-1}] R [ u α s − 1 ] specialises; in branch (b0) it was checked with u α s = 0 u_{\alpha_s}=0 u α s = 0 . The other zero tests in the code only skip factors x β ( 0 ) = 1 x_\beta(0)=1 x β ( 0 ) = 1 . A downward step is convert followed by mul_n on the converted form, and both identities specialise at every point. Finally, multiply processes the second factor letter by letter, each letter ( s , t ) (s,t) ( s , t ) being x α s ( t ) n s x_{\alpha_s}(t)n_s x α s ( t ) n s (a root step followed by mul_n), and then multiplies by h v hv h v .
The large Ree groups 2 F 4 ( q ) ^2F_4(q) 2 F 4 ( q )
Let q = 2 2 m + 1 q=2^{2m+1} q = 2 2 m + 1 and θ = 2 m \theta=2^m θ = 2 m , so that 2 θ 2 = q 2\theta^2=q 2 θ 2 = q . The large Ree group 2 F 4 ( q ) ^2F_4(q) 2 F 4 ( q ) is the group of fixed points in F 4 ( q ) F_4(q) F 4 ( q ) of an automorphism σ \sigma σ whose square is the Frobenius map t ↦ t q t\mapsto t^q t ↦ t q [21]. We construct σ \sigma σ from the law of Theorem 12.59 , which by Proposition 12.73 is correct over every field, so no new group law is needed: the law of 2 F 4 ( q ) ^2F_4(q) 2 F 4 ( q ) is the law of F 4 ( q ) F_4(q) F 4 ( q ) restricted to the fixed points. What is new is the description of the fixed points in coordinates.
The large Ree groups 2 F 4 ( q ) {}^{2}F_{4}(q) 2 F 4 ( q )
Throughout this section roots are written ∑ c i α i \sum c_i\alpha_i ∑ c i α i as ( c 1 c 2 c 3 c 4 ) (c_1c_2c_3c_4) ( c 1 c 2 c 3 c 4 ) in the labelling of the implementation (scripts/f4_law.py). In it α 1 , α 2 \alpha_1,\alpha_2 α 1 , α 2 are short, α 3 , α 4 \alpha_3,\alpha_4 α 3 , α 4 are long, the double bond joins α 1 \alpha_1 α 1 and α 4 \alpha_4 α 4 , and α 2 ⊥ α 3 \alpha_2\perp\alpha_3 α 2 ⊥ α 3 . For a root α \alpha α put k ( α ) = 1 k(\alpha)=1 k ( α ) = 1 if α \alpha α is long and k ( α ) = 2 k(\alpha)=2 k ( α ) = 2 if it is short.
Proposition 12.75 (The root map and the special isogeny). There is a bijection ρ \rho ρ of the root system of F 4 F_4 F 4 that exchanges long and short roots, preserves positivity, satisfies ρ 2 = 1 \rho^2=1 ρ 2 = 1 , swaps α 1 ↔ α 4 \alpha_1\leftrightarrow\alpha_4 α 1 ↔ α 4 and α 2 ↔ α 3 \alpha_2\leftrightarrow\alpha_3 α 2 ↔ α 3 , and sends α \alpha α to the positive multiple of ∑ c i d i α π ( i ) \sum c_i d_i\alpha_{\pi(i)} ∑ c i d i α π ( i ) that is a root; here d = ( 1 , 1 , 2 , 2 ) d=(1,1,2,2) d = ( 1 , 1 , 2 , 2 ) is the symmetriser and π \pi π reverses the indices. Over a field K K K of characteristic 2 the assignment
ψ : x α ( t ) ↦ x ρ α ( t k ( α ) ) , n α i ↦ n ρ α i , h α i ( λ ) ↦ h ρ α i ( λ k ( α i ) ) \psi:x_\alpha(t)\mapsto x_{\rho\alpha}(t^{k(\alpha)}),\qquad n_{\alpha_i}\mapsto n_{\rho\alpha_i},\qquad h_{\alpha_i}(\lambda)\mapsto h_{\rho\alpha_i}(\lambda^{k(\alpha_i)}) ψ : x α ( t ) ↦ x ρ α ( t k ( α ) ) , n α i ↦ n ρ α i , h α i ( λ ) ↦ h ρ α i ( λ k ( α i ) ) is an endomorphism of F 4 ( K ) F_4(K) F 4 ( K ) , and the implemented law satisfies ψ ( a b ) = ψ ( a ) ψ ( b ) \psi(ab)=\psi(a)\psi(b) ψ ( ab ) = ψ ( a ) ψ ( b ) .
Proof. The properties of ρ \rho ρ are checked on the 48 roots by scripts/f42-psi.py. The same script runs the unchanged law over F 2 [ v a r s ] \mathbb{F}_2[\mathrm{vars}] F 2 [ vars ] and checks ψ ( u v ) = ψ ( u ) ψ ( v ) \psi(uv)=\psi(u)\psi(v) ψ ( uv ) = ψ ( u ) ψ ( v ) for generic u , v ∈ U u,v\in U u , v ∈ U ; signs ± 1 \pm1 ± 1 are irrelevant in characteristic 2. With the exponents the other way round (t t t on short and t 2 t^2 t 2 on long roots) 18 of the 24 coordinates disagree, so the convention is forced. scripts/f42-psi_n.py checks ψ ( f n s ) = ψ ( f ) n ρ s \psi(fn_s)=\psi(f)n_{\rho s} ψ ( f n s ) = ψ ( f ) n ρ s for a generic normal form f f f in every cell and on every branch, 6912 identities, with ψ \psi ψ applied directly to normal forms (word ↦ ρ ( word ) \mapsto\rho(\text{word}) ↦ ρ ( word ) , digits and torus coordinates raised to the power k k k ); and scripts/f42-psi_group.py checks right multiplication by a generic h v ∈ B h_v\in B h v ∈ B . The specialisation to a field is as in Proposition 12.73 . The existence of ψ \psi ψ is classical [21]; the computation fixes it in the coordinates of the law. □
Definition 12.76. For q = 2 2 m + 1 q=2^{2m+1} q = 2 2 m + 1 let σ = ψ ∘ F θ \sigma=\psi\circ F_\theta σ = ψ ∘ F θ , where F θ F_\theta F θ raises every coordinate to the power θ \theta θ . Thus σ ( x α ( t ) ) = x ρ α ( t θ k ( α ) ) \sigma(x_\alpha(t))=x_{\rho\alpha}(t^{\theta k(\alpha)}) σ ( x α ( t )) = x ρ α ( t θ k ( α ) ) , and σ 2 \sigma^2 σ 2 raises coordinates to the power 2 θ 2 = q 2\theta^2=q 2 θ 2 = q , so σ 2 = 1 \sigma^2=1 σ 2 = 1 on F 4 ( q ) F_4(q) F 4 ( q ) . Put 2 F 4 ( q ) = F 4 ( q ) σ {}^{2}F_4(q)=F_4(q)^\sigma 2 F 4 ( q ) = F 4 ( q ) σ , U σ = U ∩ 2 F 4 ( q ) U^\sigma=U\cap{}^{2}F_4(q) U σ = U ∩ 2 F 4 ( q ) and T σ = T ∩ 2 F 4 ( q ) T^\sigma=T\cap{}^{2}F_4(q) T σ = T ∩ 2 F 4 ( q ) .
The 24 positive roots fall into 12 orbits of ρ \rho ρ . Grouping them by the ray of 1 2 ( α + τ α ) \frac{1}{2}(\alpha+\tau\alpha) 2 1 ( α + τ α ) , where τ \tau τ is ρ \rho ρ rescaled to an isometry, gives 8 twisted positive roots: four classes of 2 roots and four classes of 4 roots (certificates/f42_twisted_roots.json).
Proposition 12.77 (Twisted root subgroups). For each class the following elements, with the roots taken in the order of the law, are σ \sigma σ -fixed for all t , u ∈ F q t,u\in\mathbb{F}_q t , u ∈ F q and every q = 2 2 m + 1 q=2^{2m+1} q = 2 2 m + 1 , and they are distinct for distinct parameters:
Table 12.77.
The first four rows give subgroups of order q q q , the last four of order q 2 q^2 q 2 .
Proof. The formulas were found by interpolating the σ \sigma σ -fixed elements of each class over F 8 \mathbb{F}_8 F 8 and choosing, for each exponent, the unique reading i + j θ i+j\theta i + j θ that also works over F 32 \mathbb{F}_{32} F 32 (scripts/f42_param.py). The proof for all q q q is scripts/f42_param_exact.py: it works over F 2 \mathbb{F}_2 F 2 with monomials t a + b θ t^{a+b\theta} t a + b θ , a , b ∈ 1 2 Z a,b\in\frac{1}{2}\mathbb{Z} a , b ∈ 2 1 Z . On exponents θ 2 \theta^2 θ 2 acts as 1 2 \frac{1}{2} 2 1 , since 2 θ 2 = q 2\theta^2=q 2 θ 2 = q and squaring is bijective on F q \mathbb{F}_q F q , so σ \sigma σ maps a monomial with exponent e e e on a root of factor k k k to one with exponent k θ e k\theta e k θ e , where θ ( a + b θ ) = b 2 + a θ \theta(a+b\theta)=\frac{b}{2}+a\theta θ ( a + b θ ) = 2 b + a θ . Evaluating at t , u ∈ F q t,u\in\mathbb{F}_q t , u ∈ F q is a ring homomorphism, and the identities σ ( g ) = g \sigma(g)=g σ ( g ) = g hold coordinate by coordinate. Distinctness holds because t t t and u u u appear unchanged as coordinates. □
Proposition 12.78 (A closed law on U σ U^\sigma U σ ). Order the eight classes by the position of their first root in the law, and let Φ ( p ) \Phi(p) Φ ( p ) , p ∈ F q 12 p \in\mathbb{F}_{q}^{12} p ∈ F q 12 , be the product of the eight elements of Proposition 12.77 in this order, with one parameter for each class of size 2 and two for each class of size 4. There are 12 polynomials L 1 , … , L 12 L_1,\ldots,L_{12} L 1 , … , L 12 in p , p ′ p,p' p , p ′ and their θ \theta θ -powers, with 2, 3, 2, 3, 14, 5, 119, 4, 42, 8, 23, 28 terms, such that Φ ( p ) Φ ( p ′ ) = Φ ( L ( p , p ′ ) ) \Phi(p)\Phi(p')=\Phi(L(p,p')) Φ ( p ) Φ ( p ′ ) = Φ ( L ( p , p ′ )) for every q q q . In particular Φ \Phi Φ is a bijection of F q 12 \mathbb{F}_{q}^{12} F q 12 onto a subgroup of U σ U^\sigma U σ of order q 12 q^{12} q 12 .
Proof. The inverse of Φ \Phi Φ peels off one class at a time: read t t t at the first root of the class, multiply on the left by the inverse of x α ( t ) x β ( ⋅ ) x_\alpha(t)x_\beta(\cdot) x α ( t ) x β ( ⋅ ) , which in characteristic 2 is the reversed product since every x γ ( c ) x_\gamma(c) x γ ( c ) is an involution, then read u u u at the third root and repeat. In the exponent ring of Proposition 12.77 , scripts/f42_ulaw.py checks Δ ( Φ ( p ) ) = p \Delta(\Phi(p))=p Δ ( Φ ( p )) = p with trivial remainder, and computes L = Δ ( Φ ( p ) Φ ( p ′ ) ) L=\Delta(\Phi(p)\Phi(p')) L = Δ ( Φ ( p ) Φ ( p ′ )) , again with trivial remainder. As a check, 200 random pairs over each of F 8 \mathbb{F}_8 F 8 and F 32 \mathbb{F}_{32} F 32 agree with the numeric product. The polynomials are stored in certificates/f42_ulaw_polys.json.
Proposition 12.79 (Twisted Weyl group and order). The centraliser W σ W^\sigma W σ of ρ \rho ρ in W ( F 4 ) W(F_4) W ( F 4 ) has order 16; it is generated by w 1 = s α 2 s α 3 w_1=s_{\alpha_2}s_{\alpha_3} w 1 = s α 2 s α 3 and w 2 = ( s α 1 s α 4 ) 2 w_2=(s_{\alpha_1}s_{\alpha_4})^2 w 2 = ( s α 1 s α 4 ) 2 and is dihedral. For w ∈ W σ w \in W^\sigma w ∈ W σ the set of positive roots made negative by w w w is a union of classes; let d ( w ) d(w) d ( w ) count the classes of size 2 once and those of size 4 twice. Then
∑ w ∈ W σ q d ( w ) = 1 + q + q 2 + 2 q 3 + q 4 + q 5 + 2 q 6 + q 7 + q 8 + 2 q 9 + q 10 + q 11 + q 12 \sum_{w\in W^\sigma}q^{d(w)}=1+q+q^2+2q^3+q^4+q^5+2q^6+q^7+q^8+2q^9+q^{10}+q^{11}+q^{12} w ∈ W σ ∑ q d ( w ) = 1 + q + q 2 + 2 q 3 + q 4 + q 5 + 2 q 6 + q 7 + q 8 + 2 q 9 + q 10 + q 11 + q 12 and q 12 ( q − 1 ) 2 q^{12}(q-1)^2 q 12 ( q − 1 ) 2 times this sum equals q 12 ( q 6 + 1 ) ( q 4 − 1 ) ( q 3 + 1 ) ( q − 1 ) q^{12}(q^6+1)(q^4-1)(q^3+1)(q-1) q 12 ( q 6 + 1 ) ( q 4 − 1 ) ( q 3 + 1 ) ( q − 1 ) , the order of 2 F 4 ( q ) {}^{2}F_4(q) 2 F 4 ( q ) [21]. For q = 2 q=2 q = 2 this is 35 942 400.
Proof. scripts/f42_order.py builds W ( F 4 ) W(F_4) W ( F 4 ) as a permutation group on the 48 roots, finds the 16 elements commuting with ρ \rho ρ , nine of them involutions, checks that every inversion set is a union of classes, and compares the polynomials coefficient by coefficient. That w 1 , w 2 w_1,w_2 w 1 , w 2 generate W σ W^\sigma W σ is checked in scripts/f42_nw_exact.py.
Proposition 12.80 (Normal form). The elements n w 1 = n α 2 n α 3 n_{w_1}=n_{\alpha_2}n_{\alpha_3} n w 1 = n α 2 n α 3 and n w 2 = ( n α 1 n α 4 ) 2 n_{w_2}=(n_{\alpha_1}n_{\alpha_4})^2 n w 2 = ( n α 1 n α 4 ) 2 lie in 2 F 4 ( q ) {}^{2}F_4(q) 2 F 4 ( q ) for every q q q , and T σ = { h ( b θ , a θ , a , b ) : a , b ∈ F q × } T^\sigma=\{h(b^\theta,a^\theta,a,b):a,b\in\mathbb{F}_q^\times\} T σ = { h ( b θ , a θ , a , b ) : a , b ∈ F q × } , where h ( λ ) = ∏ i h α i ( λ i ) h(\lambda)=\prod_i h_{\alpha_i}(\lambda_i) h ( λ ) = ∏ i h α i ( λ i ) , has order ( q − 1 ) 2 (q-1)^2 ( q − 1 ) 2 . Every element of 2 F 4 ( q ) {}^{2}F_4(q) 2 F 4 ( q ) can be written as u n w h v u n_w h v u n w h v with w ∈ W σ w\in W^\sigma w ∈ W σ , n w n_w n w the corresponding product of n w 1 n_{w_1} n w 1 and n w 2 n_{w_2} n w 2 , h ∈ T σ h\in T^\sigma h ∈ T σ and u , v ∈ U σ u,v\in U^\sigma u , v ∈ U σ , and products are computed with the law of Theorem 12.59.
Proof. The elements n w i n_{w_i} n w i have coordinates in F 2 \mathbb{F}_2 F 2 , on which F θ F_\theta F θ is trivial, so σ ( n w i ) = ψ ( n w i ) \sigma(n_{w_i})=\psi(n_{w_i}) σ ( n w i ) = ψ ( n w i ) ; that this equals n w i n_{w_i} n w i is checked exactly over F 2 \mathbb{F}_2 F 2 by scripts/f42_nw_exact.py. For the torus, σ ( h ( λ ) ) \sigma(h(\lambda)) σ ( h ( λ )) has π ( i ) \pi(i) π ( i ) -th coordinate λ i θ κ ( α i ) \lambda_i^{\theta\kappa(\alpha_i)} λ i θ κ ( α i ) , and for λ = ( b θ , a θ , a , b ) \lambda=(b^\theta,a^\theta,a,b) λ = ( b θ , a θ , a , b ) this gives back λ \lambda λ because b θ 2 2 = b q = b b^{\theta2^2}=b^q=b b θ 2 2 = b q = b and a 2 2 θ = a a^{2^{2\theta}}=a a 2 2 θ = a . The decomposition is the Bruhat decomposition of a twisted group [21], and its order count is Proposition 12.79 . As a check, over F 8 \mathbb{F}_8 F 8 (scripts/f42_group-q8.py) σ 2 = 1 \sigma^2=1 σ 2 = 1 on 200 random elements of F 4 ( 8 ) F_4(8) F 4 ( 8 ) , the elements n w 1 , n w 2 n_{w_1},n_{w_2} n w 1 , n w 2 reach exactly 16 cells, and 300 random products of length 30 in n w 1 , n w 2 n_{w_1},n_{w_2} n w 1 , n w 2 , T σ T^\sigma T σ and the root elements are all σ \sigma σ -fixed and lie in cells of W σ W^\sigma W σ .
Proposition 12.81 (Rank-one rules). Let X A X_A X A and X B X_B X B be the twisted root subgroups of the simple classes,
X A ( t ) = x ( 0010 ) ( t ) x ( 0100 ) ( t θ ) , X B ( t , u ) = x ( 0001 ) ( t ) x ( 1000 ) ( t θ ) x ( 1001 ) ( u ) x ( 2001 ) ( u 2 θ + t 2 θ + 1 ) , X_A(t)=x_{(0010)}(t)x_{(0100)}(t^\theta),\qquad X_B(t,u)=x_{(0001)}(t)x_{(1000)}(t^\theta)x_{(1001)}(u)x_{(2001)}(u^{2^\theta}+t^{2^\theta+1}), X A ( t ) = x ( 0010 ) ( t ) x ( 0100 ) ( t θ ) , X B ( t , u ) = x ( 0001 ) ( t ) x ( 1000 ) ( t θ ) x ( 1001 ) ( u ) x ( 2001 ) ( u 2 θ + t 2 θ + 1 ) ,
and put N ( t , u ) = t 2 θ + 1 + t θ u + u 2 θ N(t,u)=t^{2\theta+1}+t^\theta u+u^{2\theta} N ( t , u ) = t 2 θ + 1 + t θ u + u 2 θ . Then N ( t , u ) ≠ 0 N(t,u)\ne0 N ( t , u ) = 0 for ( t , u ) ≠ ( 0 , 0 ) (t,u)\ne(0,0) ( t , u ) = ( 0 , 0 ) , and for every q q q :
n w 1 X A ( t ) n w 1 − 1 = X A ( t − 1 ) n w 1 h ( 1 , t θ , t , 1 ) X A ( t − 1 ) ( t ≠ 0 ) , n_{w_1}X_A(t)n_{w_1}^{-1}=X_A(t^{-1})n_{w_1}h(1,t^\theta,t,1)X_A(t^{-1}) \qquad(t\ne0), n w 1 X A ( t ) n w 1 − 1 = X A ( t − 1 ) n w 1 h ( 1 , t θ , t , 1 ) X A ( t − 1 ) ( t = 0 ) ,
n w 2 X B ( t , u ) n w 2 − 1 = X B ( u 2 θ + t 2 θ + 1 N 2 θ , u N 2 θ ) n w 2 ⋅ h ( N , 1 , 1 , N 2 θ ) X B ( u 2 θ N 2 θ , t θ N ) ( ( t , u ) ≠ ( 0 , 0 ) ) . \begin{aligned}
n_{w_2}X_B(t,u)n_{w_2}^{-1}
&=X_B\left(\frac{u^{2\theta}+t^{2\theta+1}}{N^{2\theta}},\frac{u}{N^{2\theta}}\right)n_{w_2}\\
&\quad\cdot h(N,1,1,N^{2\theta})X_B\left(\frac{u^{2\theta}}{N^{2\theta}},\frac{t^\theta}{N}\right)
\qquad((t,u)\ne(0,0)).
\end{aligned} n w 2 X B ( t , u ) n w 2 − 1 = X B ( N 2 θ u 2 θ + t 2 θ + 1 , N 2 θ u ) n w 2 ⋅ h ( N , 1 , 1 , N 2 θ ) X B ( N 2 θ u 2 θ , N t θ ) (( t , u ) = ( 0 , 0 )) .
Both torus elements lie in T σ T^\sigma T σ , since ( N 2 θ ) θ = N q = N (N^{2\theta})^\theta=N^q=N ( N 2 θ ) θ = N q = N .
Proof. If t = 0 t=0 t = 0 then N = u 2 θ ≠ 0 N=u^{2\theta}\ne0 N = u 2 θ = 0 . If t ≠ 0 t\ne0 t = 0 , write u = t θ + 1 v u=t^{\theta+1}v u = t θ + 1 v ; since 2 θ ( θ + 1 ) = q + 2 θ 2\theta(\theta+1)=q+2\theta 2 θ ( θ + 1 ) = q + 2 θ , this gives N = t 2 θ + 1 ( 1 + v + v 2 θ ) N=t^{2\theta+1}(1+v+v^{2\theta}) N = t 2 θ + 1 ( 1 + v + v 2 θ ) . If v 2 θ = v + 1 v^{2\theta}=v+1 v 2 θ = v + 1 , raising to the power 2 θ 2\theta 2 θ gives v 2 q = v 2 θ + 1 = v v^{2q}=v^{2\theta}+1=v v 2 q = v 2 θ + 1 = v , that is v 2 = v v^2=v v 2 = v ; but v = 0 v=0 v = 0 and v = 1 v=1 v = 1 both give 1 + v + v 2 θ = 1 1+v+v^{2\theta}=1 1 + v + v 2 θ = 1 . So N ≠ 0 N\ne0 N = 0 .
The formulas were found from the law over F 32 \mathbb{F}_{32} F 32 and confirmed at every parameter over F 8 \mathbb{F}_8 F 8 and F 32 \mathbb{F}_{32} F 32 (scripts/f42_rank1.py, scripts/f42_rank1_tprime.py). Both identities live in the Levi subgroup of the two simple roots involved: ⟨ α 2 , α 3 ⟩ \langle\alpha_2,\alpha_3\rangle ⟨ α 2 , α 3 ⟩ of type A 1 × A 1 A_1\times A_1 A 1 × A 1 and ⟨ α 1 , α 4 ⟩ \langle\alpha_1,\alpha_4\rangle ⟨ α 1 , α 4 ⟩ of type B 2 B_2 B 2 . In characteristic 2 these are S L 2 × S L 2 SL_2\times SL_2 S L 2 × S L 2 and S p 4 Sp_4 S p 4 with their Chevalley root elements, since the Steinberg relations fix the signs [21]. scripts/f42_rank1_proof.py checks the B 2 B_2 B 2 commutator relation in the 4 × 4 4\times4 4 × 4 model, confirms the torus convention numerically (the inverse convention fails at 990 points over F 32 \mathbb{F}_{32} F 32 ), and proves both identities for every q q q . It works over F 2 \mathbb{F}_2 F 2 with monomials t a + b θ t^{a+b\theta} t a + b θ , u a + b θ u^{a+b\theta} u a + b θ , v a + b θ v^{a+b\theta} v a + b θ with ν = N − 1 \nu=N^{-1} ν = N − 1 , as in Proposition 12.77 , multiplies each entry of the difference of the two sides by a power N A + B θ N^{A+B\theta} N A + B θ to clear ν \nu ν , expands N N N , and gets 0 in all 16 entries; the S L 2 SL_2 S L 2 identity is proved the same way.
Propositions 12.78 and 12.81 give a multiplication of 2 F 4 ( q ) {}^{2}F_4(q) 2 F 4 ( q ) that does not use the law of F 4 F_4 F 4 at run time. Write an element as u n w h v u n_w h v u n w h v , where w w w is an alternating word in w 1 , w 2 w_1,w_2 w 1 , w 2 of length at most 8, u , v ∈ U σ u,v\in U^\sigma u , v ∈ U σ are given by their 12 coordinates and h = h ( b θ , a θ , a , b ) h=h(b^\theta,a^\theta,a,b) h = h ( b θ , a θ , a , b ) . Right multiplication by an element of U σ U^\sigma U σ is the law L L L . Right multiplication by h ′ ∈ T σ h'\in T^\sigma h ′ ∈ T σ replaces h h h by h h ′ hh' h h ′ and conjugates v v v , which scales each coordinate by a root character. To multiply by n w i n_{w_i} n w i , split v = X c ( p ) v ′ v=X_c(p)v' v = X c ( p ) v ′ with c c c the class of w i w_i w i , conjugate v ′ v' v ′ by n w i n_{w_i} n w i (twelve polynomial maps, computed once from the law), and move h h h past X c ( p ) X_c(p) X c ( p ) . If w i w_i w i lengthens w w w , conjugate X c ( p ) X_c(p) X c ( p ) by n w n_w n w into u u u and append w i w_i w i . Otherwise, if p = 0 p=0 p = 0 , drop the last letter; if p ≠ 0 p\ne0 p = 0 , apply the rank-one rule and stay in the cell of w w w . A general product applies these steps along the second factor. Splitting off X c ( p ) X_c(p) X c ( p ) needs inverses. In characteristic 2 the inverse of a product of root elements is the reversed product, and peeling it as in Proposition 12.78 , exactly for every q q q , gives X c ( t ) − 1 = X c ( t ) X_c(t)^{-1}=X_c(t) X c ( t ) − 1 = X c ( t ) for the classes of size 2 and
X c ( t , u ) − 1 = X c ( t , u + t θ + 1 ) ( c ⊃ ( 0001 ) or ( 0011 ) ) , X_c(t,u)^{-1}=X_c(t,u+t^{\theta+1}) \qquad(c\supset(0001)\text{ or }(0011)), X c ( t , u ) − 1 = X c ( t , u + t θ + 1 ) ( c ⊃ ( 0001 ) or ( 0011 )) ,
X c ( t , u ) − 1 = X c ( t , u + t 2 θ + 1 ) ( c ⊃ ( 1011 ) or ( 1101 ) ) . X_c(t,u)^{-1}=X_c(t,u+t^{2\theta+1}) \qquad(c\supset(1011)\text{ or }(1101)). X c ( t , u ) − 1 = X c ( t , u + t 2 θ + 1 ) ( c ⊃ ( 1011 ) or ( 1101 )) .
scripts/f42_twisted_law.py implements this for any q = 2 2 m + 1 q=2^{2m+1} q = 2 2 m + 1 . On 200 random pairs in arbitrary cells over each of F 8 \mathbb{F}_8 F 8 and F 32 \mathbb{F}_{32} F 32 , mapping the product back to F 4 F_4 F 4 agrees with the product computed in F 4 F_4 F 4 every time, at about 0.05 seconds per product.
Theorem 12.83 (Digital law of 2 F 4 ( q {}^{2}F_4(q 2 F 4 ( q )). Let q = 2 2 m + 1 q=2^{2m+1} q = 2 2 m + 1 . Order the 16 elements of W σ = ⟨ w 1 , w 2 ⟩ W^\sigma=\langle w_1,w_2\rangle W σ = ⟨ w 1 , w 2 ⟩ as 1, then for ℓ = 1 , … , 7 \ell=1,\ldots,7 ℓ = 1 , … , 7 the alternating words of length ℓ \ell ℓ beginning with w 1 w_1 w 1 and with w 2 w_2 w 2 , and finally w 0 = ( w 1 w 2 ) 4 w_0=(w_1w_2)^4 w 0 = ( w 1 w 2 ) 4 . Fix the reduced word just listed for each w w w . Every g ∈ 2 F 4 ( q ) g\in{}^{2}F_4(q) g ∈ 2 F 4 ( q ) has a unique expression
g = X c 1 ( p 1 ) n w i 1 ⋯ X c k ( p k ) n w i k h ( b θ , a θ , a , b ) Φ ( v ) , g=X_{c_1}(p_1)n_{w_{i_1}}\cdots X_{c_k}(p_k)n_{w_{i_k}}h(b^\theta,a^\theta,a,b)\Phi(v), g = X c 1 ( p 1 ) n w i 1 ⋯ X c k ( p k ) n w i k h ( b θ , a θ , a , b ) Φ ( v ) , where w = w i 1 ⋯ w i k w=w_{i_1}\cdots w_{i_k} w = w i 1 ⋯ w i k is that reduced word, c j c_j c j is the class of w i j w_{i_j} w i j (so p j ∈ F q p_j\in\mathbb{F}_q p j ∈ F q for w 1 w_1 w 1 and p j ∈ F q 2 p_j\in\mathbb{F}_q^2 p j ∈ F q 2 for w 2 w_2 w 2 ), a , b ∈ F q 2 a,b\in\mathbb{F}_q^2 a , b ∈ F q 2 and v ∈ F q 2 v\in\mathbb{F}_q^2 v ∈ F q 2 . Write d ( w ) d(w) d ( w ) for the number of letters w 1 w_1 w 1 plus twice the number of
letters w 2 w_2 w 2 . Then ∑ w q d ( w ) ( q − 1 ) 2 q 12 = ∣ 2 F 4 ( q ) ∣ \sum_w q^{d(w)}(q-1)^2q^{12}=|{}^{2}F_4(q)| ∑ w q d ( w ) ( q − 1 ) 2 q 12 = ∣ 2 F 4 ( q ) ∣ . With field elements coded as in § 20.2 , set
λ ( g ) = O w + v ˉ + q 12 ( ( b − 1 ) + ( q − 1 ) ( ( a − 1 ) + ( q − 1 ) p ˉ ) ) . \lambda(g)=O_w+\bar{v}+q^{12}\left((b-1)+(q-1)\left((a-1)+(q-1)\bar{p}\right)\right). λ ( g ) = O w + v ˉ + q 12 ( ( b − 1 ) + ( q − 1 ) ( ( a − 1 ) + ( q − 1 ) p ˉ ) ) .
where O w O_w O w is the number of elements in the cells before w w w , v ˉ = ∑ j v j q j \bar{v}=\sum_j v_jq^j v ˉ = ∑ j v j q j , and p ˉ \bar{p} p ˉ reads p 1 , … , p k p_1,\ldots,p_k p 1 , … , p k as a mixed-radix number with p 1 p_1 p 1 most significant. Each digit is t t t for w 1 w_1 w 1 and t + q u t+qu t + q u for w 2 w_2 w 2 , with radix q q q or q 2 q^2 q 2 . Then λ : 2 F 4 ( q ) → A ∣ 2 F 4 ( q ) ∣ \lambda:{}^{2}F_4(q)\to\mathcal{A}_{|{}^{2}F_4(q)|} λ : 2 F 4 ( q ) → A ∣ 2 F 4 ( q ) ∣ is a bijection with λ ( 1 ) = 0 \lambda(1)=0 λ ( 1 ) = 0 , and m ∗ n = λ ( λ − 1 ( m ) λ − 1 ( n ) ) m\ast n=\lambda\left(\lambda^{-1}(m)\lambda^{-1}(n)\right) m ∗ n = λ ( λ − 1 ( m ) λ − 1 ( n ) ) is a digital law of 2 F 4 ( q ) {}^{2}F_4(q) 2 F 4 ( q ) . Conversely, if O w ≤ m < O w + q d ( w ) ( q − 1 ) 2 q 12 O_w\leq m<O_w+q^{d(w)}(q-1)^2q^{12} O w ≤ m < O w + q d ( w ) ( q − 1 ) 2 q 12 , then the coordinates of λ − 1 ( m ) \lambda^{-1}(m) λ − 1 ( m ) are the mixed-radix digits of m − O w m-O_w m − O w with radices q q q (twelve times), q − 1 q-1 q − 1 , q − 1 q-1 q − 1 , and then the radices of p k , … , p 1 p_k,\ldots,p_1 p k , … , p 1 . That is, v j = D j ( m − O w ) v_j=D_j(m-O_w) v j = D j ( m − O w ) for j < 12 j<12 j < 12 , b − 1 = D 12 ( m − O w ) b-1=D_{12}(m-O_w) b − 1 = D 12 ( m − O w ) , a − 1 = D 13 ( m − O w ) a-1=D_{13}(m-O_w) a − 1 = D 13 ( m − O w ) , and p k − j p_{k-j} p k − j is read from D 14 + j ( m − O w ) D_{14+j}(m-O_w) D 14 + j ( m − O w ) , with t = d q ( 0 , x ) t=d_q(0,x) t = d q ( 0 , x ) and u = ⌊ x / q ⌋ u=\lfloor x/q\rfloor u = ⌊ x / q ⌋ for a digit x x x of radix q 2 q^2 q 2 . The product is computed in twisted coordinates by Remark 12.82, followed by the normal form.
Proof. The expression is the Bruhat decomposition of the split B N BN B N -pair of 2 F 4 ( q ) {}^{2}F_4(q) 2 F 4 ( q ) , with U σ ∩ n w U σ , − n w U^\sigma\cap n_wU^{\sigma,-n_w} U σ ∩ n w U σ , − n w factored along the reduced word into the root subgroups X c j X_{c_j} X c j [21]. The normal form is reached by peeling u ∈ U σ u\in U^\sigma u ∈ U σ along the word: write u = X c ( p ) u ′ u=X_c(p)u' u = X c ( p ) u ′ with the c c c -coordinate of u ′ u' u ′ zero, and move u ′ u' u ′ past n w i n_{w_i} n w i into U σ U^\sigma U σ . The count is ∑ w q d ( w ) = ( q 6 + 1 ) ( q 3 + 1 ) ( q 2 + 1 ) ( q + 1 ) \sum_wq^{d(w)}=(q^6+1)(q^3+1)(q^2+1)(q+1) ∑ w q d ( w ) = ( q 6 + 1 ) ( q 3 + 1 ) ( q 2 + 1 ) ( q + 1 ) , so λ \lambda λ is a bijection.
Theorem 12.84 (The Tits group and 2 F 4 ( 2 ) {}^{2}F_4(2) 2 F 4 ( 2 ) on A 2 p 0 A_{2p_0} A 2 p 0 ). Let q = 2 q=2 q = 2 , so the torus is trivial, and let p 0 = 17 971 200 p_0=17\,971\,200 p 0 = 17 971 200 . In the normal form of Theorem 12.83 set
f ( g ) = ∑ w i j = w 2 t j + ∑ c t c ( v ) ( m o d 2 ) , f(g)=\sum_{w_{ij}=w_2}t_j+\sum_c t_c(v)\pmod2, f ( g ) = w ij = w 2 ∑ t j + c ∑ t c ( v ) ( mod 2 ) , where p j = ( t j , u j ) p_j=(t_j,u_j) p j = ( t j , u j ) and c c c runs over the four classes of size 4, with t c ( v ) t_c(v) t c ( v ) the first coordinate of v v v in class c c c . Then f : 2 F 4 ( 2 ) → C 2 f:{}^{2}F_4(2)\to C_2 f : 2 F 4 ( 2 ) → C 2 is the unique nontrivial homomorphism, and its kernel is the Tits group T = 2 F 4 ( 2 ) ′ T={}^{2}F_4(2)' T = 2 F 4 ( 2 ) ′ . Deleting from λ \lambda λ the coordinate t c 0 ( v ) t_{c_0}(v) t c 0 ( v ) , for c 0 c_0 c 0 the class of ( 0001 ) (0001) ( 0001 ) , gives a bijection λ T : T → A p 0 \lambda_T:T\to A_{p_0} λ T : T → A p 0 with λ T ( 1 ) = 0 \lambda_T(1)=0 λ T ( 1 ) = 0 , because f ( g ) = 0 f(g)=0 f ( g ) = 0 determines that coordinate. Let τ = X c 0 ( 1 , 0 ) \tau=X_{c_0}(1,0) τ = X c 0 ( 1 , 0 ) . Then f ( τ ) = 1 f(\tau)=1 f ( τ ) = 1 and τ \tau τ has order 4. Every g g g is uniquely t τ e t\tau^e t τ e with t ∈ T t\in T t ∈ T and e ∈ { 0 , 1 } e\in\{0,1\} e ∈ { 0 , 1 } , and
Λ ( t τ e ) = λ T ( t ) + p 0 e \Lambda(t\tau^e)=\lambda_T(t)+p_0e Λ ( t τ e ) = λ T ( t ) + p 0 e is a bijection 2 F 4 ( 2 ) → A 2 p 0 {}^{2}F_4(2)\to A_{2p_0} 2 F 4 ( 2 ) → A 2 p 0 . With radices p 0 , 2 p_0,2 p 0 , 2 write m h = D h ( m ) m_h=D_h(m) m h = D h ( m ) and n h = D h ( n ) n_h=D_h(n) n h = D h ( n ) . Thus m 0 = d p 0 ( 0 , m ) m_0=d_{p_0}(0,m) m 0 = d p 0 ( 0 , m ) is the Tits code and m 1 = ⌊ m / p 0 ⌋ ∈ { 0 , 1 } m_1=\lfloor m/p_0\rfloor\in\{0,1\} m 1 = ⌊ m / p 0 ⌋ ∈ { 0 , 1 } is the coset digit. Let ∗ T \ast_T ∗ T be the digital law of T T T transported by λ T \lambda_T λ T , and let φ ( n 0 ) = λ T ( τ λ T − 1 ( n 0 ) τ − 1 ) \varphi(n_0)=\lambda_T(\tau\lambda_T^{-1}(n_0)\tau^{-1}) φ ( n 0 ) = λ T ( τ λ T − 1 ( n 0 ) τ − 1 ) . Then
m ∗ n = ( m 0 ∗ T φ m 1 ( n 0 ) ∗ T c m 1 n 1 ) + p 0 d 2 ( 0 , m 1 + n 1 ) , m\ast n=\left(m_0\ast_T\varphi^{m_1}(n_0)\ast_T c^{m_1n_1}\right)+p_0d_2(0,m_1+n_1), m ∗ n = ( m 0 ∗ T φ m 1 ( n 0 ) ∗ T c m 1 n 1 ) + p 0 d 2 ( 0 , m 1 + n 1 ) , where c = λ T ( τ 2 ) = 1 c=\lambda_T(\tau^2)=1 c = λ T ( τ 2 ) = 1 .
This is the semidirect law (59) with G 1 = C 2 G_1=C_2 G 1 = C 2 , m 1 ∗ 1 n 1 = d 2 ( 0 , m 1 + n 1 ) m_1\ast_1n_1=d_2(0,m_1+n_1) m 1 ∗ 1 n 1 = d 2 ( 0 , m 1 + n 1 ) and ϕ m 1 = φ m 1 \phi_{m_1}=\varphi^{m_1} ϕ m 1 = φ m 1 , corrected by the factor c m 1 n 1 c^{m_1n_1} c m 1 n 1 , which is present only when both coset digits are 1. The codes below p 0 p_0 p 0 form the Tits group, and φ \varphi φ is the outer automorphism induced by τ \tau τ .
Proof. Since U σ U^\sigma U σ , n w 1 n_{w_1} n w 1 and n w 2 n_{w_2} n w 2 generate 2 F 4 ( 2 ) {}^{2}F_4(2) 2 F 4 ( 2 ) , and f f f is defined on all of 2 F 4 ( 2 ) {}^{2}F_4(2) 2 F 4 ( 2 ) through the normal form, it suffices to show f ( g x ) = f ( g ) + f ( x ) f(gx)=f(g)+f(x) f ( g x ) = f ( g ) + f ( x ) for every g g g and every x ∈ U σ ∪ { n w 1 , n w 2 } x\in U^\sigma\cup\{n_{w_1},n_{w_2}\} x ∈ U σ ∪ { n w 1 , n w 2 } . Then f f f of a product of generators is the sum of their values, so f f f is a homomorphism. Write g = D n w v g=D_{n_w}v g = D n w v in normal form, where D D D is the digit part, so that f ( g ) = f ( D n w ) + f ( v ) f(g)=f(D_{n_w})+f(v) f ( g ) = f ( D n w ) + f ( v ) .
(i) For x ∈ U σ x\in U^\sigma x ∈ U σ , g x = D n w ( v x ) gx=D_{n_w}(vx) g x = D n w ( v x ) is again in normal form, so we only need f ∣ U σ f|_{U^\sigma} f ∣ U σ to be a homomorphism. It is enough to check f ( v x ) = f ( v ) + f ( x ) f(vx)=f(v)+f(x) f ( v x ) = f ( v ) + f ( x ) for all 4096 elements v v v and all 16 nonzero root-class elements x x x , which is 65 536 pairs.
(ii) For x = n w i x=n_{w_i} x = n w i , write v = X c i ( p ) v ′ v=X_{c_i}(p)v' v = X c i ( p ) v ′ with v ′ v' v ′ in the subgroup U i σ U_i^\sigma U i σ of elements whose c i c_i c i -coordinate is 0. This subgroup is normalised by n w i n_{w_i} n w i . Then v n w i = ( D n w X c i ( p ) n w i ) ( n w i − 1 v ′ n w i ) vn_{w_i}=(D_{n_w}X_{c_i}(p)n_{w_i})(n_{w_i}^{-1}v'n_{w_i}) v n w i = ( D n w X c i ( p ) n w i ) ( n w i − 1 v ′ n w i ) . By (i), f ( g n w i ) = f ( D n w X c i ( p ) n w i ) + f ( n w i − 1 v ′ n w i ) f(gn_{w_i}) = f(D_{n w}X_{c_i}(p)n_{w_i}) + f(n_{w_i}^{-1}v'n_{w_i}) f ( g n w i ) = f ( D n w X c i ( p ) n w i ) + f ( n w i − 1 v ′ n w i ) . An exhaustive check gives f ( n w i − 1 v ′ n w i ) = f ( v ′ ) f(n_{w_i}^{-1}v'n_{w_i}) = f(v') f ( n w i − 1 v ′ n w i ) = f ( v ′ ) for all 3072 elements v ′ ∈ U 1 ′ ∪ U 2 ′ v' \in U'_1 \cup U'_2 v ′ ∈ U 1 ′ ∪ U 2 ′ , and f ( D n w X c i ( p ) n w i ) = f ( D n w ) + f ( X c i ( p ) ) f(D_{n w}X_{c_i}(p)n_{w_i}) = f(D_{n w}) + f(X_{c_i}(p)) f ( D n w X c i ( p ) n w i ) = f ( D n w ) + f ( X c i ( p )) for all 52 650 triples ( w , D , p ) (w,D,p) ( w , D , p ) with i = 1 , 2 i = 1,2 i = 1 , 2 . Since n w i 2 = 1 n_{w_i}^2 = 1 n w i 2 = 1 and f ( n w i ) = 0 f(n_{w_i}) = 0 f ( n w i ) = 0 , this gives f ( g n w i ) = f ( g ) = f ( g ) + f ( n w i ) f(gn_{w_i}) = f(g) = f(g) + f(n_{w_i}) f ( g n w i ) = f ( g ) = f ( g ) + f ( n w i ) . These checks are scripts/f42_fhom_proof.py, with results in certificates/f42_fhom.json.
Since f ( τ ) = 1 f(\tau) = 1 f ( τ ) = 1 , the kernel of f f f has index 2. By Griess and Lyons [29], the Tits group T = 2 F 4 ( 2 ) ′ T = {}^{2}F_4(2)' T = 2 F 4 ( 2 ) ′ is the only nontrivial proper normal subgroup of 2 F 4 ( 2 ) {}^{2}F_4(2) 2 F 4 ( 2 ) . Hence it is the only subgroup of index 2, and it equals ker f \ker f ker f . The formula for m ∗ n m * n m ∗ n is the multiplication of the extension T ⋅ ⟨ τ ⟩ T \cdot\langle\tau\rangle T ⋅ ⟨ τ ⟩ with τ 2 ∈ T \tau^2 \in T τ 2 ∈ T .
Remark 12.85 (Checks and the non-split extension). scripts/f42_digital.py implements both theorems, and scripts/f42_fhom_proof.py carries out the exhaustive checks used in the proof of Theorem 12.84 . For q = 2 q = 2 q = 2 and q = 8 q = 8 q = 8 the number of codes equals ∣ 2 F 4 ( q ) ∣ \lvert{}^{2}F_4(q)\rvert ∣ 2 F 4 ( q )∣ (35 942 400 and 264 905 352 699 586 176 614 400). On random samples, code to element to code and element to code to element are identities, the law agrees with the law of F 4 F_4 F 4 , and it is associative. For q = 2 q = 2 q = 2 the code of τ \tau τ is p 0 p_0 p 0 and that of τ 2 \tau^2 τ 2 is 1. Products of codes below p 0 p_0 p 0 stay below p 0 p_0 p 0 , φ \varphi φ preserves A p 0 A_{p_0} A p 0 , and the formula for m ∗ n m * n m ∗ n holds in all sampled cases. Among 300 random elements of 2 F 4 ( 2 ) {}^{2}F_4(2) 2 F 4 ( 2 ) outside T T T , the orders found were 4, 8, 12, 16, 20, so no outer element had order 2 ⋅ o d d 2\mathord{\cdot}\mathrm{odd} 2 ⋅ odd . This agrees with the ATLAS of Group Representations [27]: elements of orders 1, 2, 3, 6, 10, 13 of 2 F 4 ( 2 ) {}^{2}F_4(2) 2 F 4 ( 2 ) are inner, so the coset T τ T\tau T τ contains no involution, and the extension T ⋅ 2 T \cdot2 T ⋅ 2 does not split [26]. Hence the decomposition of A 2 p 0 A_{2p_0} A 2 p 0 needs the factor c = λ T ( τ 2 ) c = \lambda_T(\tau^2) c = λ T ( τ 2 ) and is not a semidirect product.
Remark 12.86 (Status of the 2 F 4 {}^{2}F_4 2 F 4 section). Proved by exact computation for every q q q : the isogeny ψ \psi ψ in the coordinates of the law, the twisted root subgroups, the closed law on a subgroup of U σ U^\sigma U σ of order q 12 q^{12} q 12 , the elements n w 1 , n w 2 n_{w_1}, n_{w_2} n w 1 , n w 2 , and the order polynomial. Cited: the existence of the Bruhat decomposition for twisted groups and that ∣ U σ ∣ = q 12 \lvert U^\sigma\rvert= q^{12} ∣ U σ ∣ = q 12 , so that the subgroup of Proposition 12.78 is all of U σ U^\sigma U σ [21]; that 2 F 4 ( q ) {}^{2}F_4(q) 2 F 4 ( q ) is simple for q ≥ 8 q \ge8 q ≥ 8 , and that 2 F 4 ( 2 ) {}^{2}F_4(2) 2 F 4 ( 2 ) has the simple Tits group 2 F 4 ( 2 ) ′ {}^{2}F_4(2)' 2 F 4 ( 2 ) ′ of order 17 971 200 as a subgroup of index 2 [21, 5]. The rank-one rules of Proposition 12.81 are proved for every q q q . Multiplication entirely in twisted coordinates is implemented for every q q q and validated over F 8 \mathbb{F}_8 F 8 and F 32 \mathbb{F}_{32} F 32 (Remark 12.82). Digital laws of 2 F 4 ( q ) {}^{2}F_4(q) 2 F 4 ( q ) , of the Tits group and of 2 F 4 ( 2 ) {}^{2}F_4(2) 2 F 4 ( 2 ) on A 2 p 0 A_{2p_0} A 2 p 0 are in Theorems 12.83 and 12.84 .
The groups E 6 ( q ) E_6(q) E 6 ( q )
The group E 6 ( q ) E_6(q) E 6 ( q ) acts on a module of dimension 27 whose weights form a single Weyl orbit. On such a minuscule module every root element is I + t e β I + t e_\beta I + t e β with e β 2 = 0 e_\beta^2 = 0 e β 2 = 0 , so the construction works uniformly in q q q and needs none of the case distinctions that F 4 F_4 F 4 required in characteristics 2 and 3.
Proposition 12.87 (Integral model of E 6 E_6 E 6 ). Use Bourbaki labels: the simple roots α 1 , … , α 6 \alpha_1,\ldots,\alpha_6 α 1 , … , α 6 with α 1 α 3 α 4 α 5 α 6 \alpha_1\alpha_3\alpha_4\alpha_5\alpha_6 α 1 α 3 α 4 α 5 α 6 a chain and α 2 \alpha_2 α 2 joined to α 4 \alpha_4 α 4 . Let Λ \Lambda Λ be the orbit of ω 1 \omega_1 ω 1 under W W W , written in fundamental-weight coordinates, and let V = Z Λ V = \mathbb{Z}\Lambda V = Z Λ with basis v μ v_\mu v μ . Put e i v μ = v μ + α i e_i v_\mu= v_{\mu+\alpha_i} e i v μ = v μ + α i if μ i = − 1 \mu_i = -1 μ i = − 1 and f i v μ = v μ − α i f_i v_\mu= v_{\mu-\alpha_i} f i v μ = v μ − α i if μ i = 1 \mu_i = 1 μ i = 1 , and 0 otherwise. For a positive root β \beta β of height at least 2, choose i i i with β − α i \beta-\alpha_i β − α i a root and set e β = [ e i , e β − α i ] e_\beta= [e_i,e_{\beta-\alpha_i}] e β = [ e i , e β − α i ] and e − β = ± [ f i , e − ( β − α i ) ] e_{-\beta} = \pm[f_i,e_{-(\beta-\alpha_i)}] e − β = ± [ f i , e − ( β − α i ) ] , with the sign chosen so that [ e β , e − β ] = h β [e_\beta,e_{-\beta}] = h_\beta [ e β , e − β ] = h β , where h β v μ = ⟨ μ , β ∨ ⟩ v μ h_\beta v_\mu= \langle\mu,\beta^\vee\rangle v_\mu h β v μ = ⟨ μ , β ∨ ⟩ v μ . Then:
∣ Λ ∣ = 27 \lvert\Lambda\rvert= 27 ∣ Λ ∣ = 27 , there are 72 roots, every e β e_\beta e β has entries in { 0 , ± 1 } \{0,\pm1\} { 0 , ± 1 } , and e β 2 = 0 e_\beta^2 = 0 e β 2 = 0 ;
the 72 matrices e β e_\beta e β and the six h α i h_{\alpha_i} h α i span a Lie algebra of dimension 78;
with x β ( t ) = I + t e β x_\beta(t) = I + t e_\beta x β ( t ) = I + t e β : if α + β \alpha+\beta α + β is a root, then x α ( t ) x β ( u ) = x β ( u ) x α ( t ) x α + β ( N α β t u ) x_\alpha(t)x_\beta(u) = x_\beta(u)x_\alpha(t)x_{\alpha+\beta}(N_{\alpha\beta}tu) x α ( t ) x β ( u ) = x β ( u ) x α ( t ) x α + β ( N α β t u ) with N α β = ± 1 N_{\alpha\beta} = \pm1 N α β = ± 1 ; if α + β \alpha+\beta α + β is neither a root nor 0, then x α ( t ) x_\alpha(t) x α ( t ) and x β ( u ) x_\beta(u) x β ( u ) commute;
n α = x α ( 1 ) x − α ( − 1 ) x α ( 1 ) n_\alpha=x_\alpha(1)x_{-\alpha}(-1)x_\alpha(1) n α = x α ( 1 ) x − α ( − 1 ) x α ( 1 ) is a signed permutation matrix with n α v μ = ± v s α μ n_\alpha v_\mu=\pm v_{s_\alpha\mu} n α v μ = ± v s α μ . The n i = n α i n_i=n_{\alpha_i} n i = n α i satisfy the braid relations of W W W , and n i 2 = h i ( − 1 ) n_i^2=h_i(-1) n i 2 = h i ( − 1 ) , where h i ( c ) v μ = c μ i v μ h_i(c)v_\mu=c^{\mu_i}v_\mu h i ( c ) v μ = c μ i v μ ;
∣ W ∣ = 51 840 |W|=51\,840 ∣ W ∣ = 51 840 , and ∑ w ∈ W q ℓ ( w ) = ∏ d ∈ { 2 , 5 , 6 , 8 , 9 , 12 } ( q d − 1 ) / ( q − 1 ) \sum_{w\in W}q^{\ell(w)}=\prod_{d\in\{2,5,6,8,9,12\}}(q^d-1)/(q-1) ∑ w ∈ W q ℓ ( w ) = ∏ d ∈ { 2 , 5 , 6 , 8 , 9 , 12 } ( q d − 1 ) / ( q − 1 ) ;
the invariant cubic forms on V V V form a space of dimension 1, spanned by a form with 45 monomials;
z = ∏ i h i ( ω k i ) z=\prod_i h_i(\omega^{k_i}) z = ∏ i h i ( ω k i ) with k = ( 1 , 0 , 2 , 0 , 1 , 2 ) k=(1,0,2,0,1,2) k = ( 1 , 0 , 2 , 0 , 1 , 2 ) acts as the scalar ω \omega ω whenever ω 3 = 1 \omega^3=1 ω 3 = 1 .
Proof. Every statement is an identity between integer matrices, and scripts/e6_model.py verifies all of them exactly. In (3) the product x α ( t ) x β ( u ) x_\alpha(t)x_\beta(u) x α ( t ) x β ( u ) is a polynomial in t , u t,u t , u , because the e e e ’s are nilpotent. The relation is equivalent to [ e α , e β ] = N α β e α + β [e_\alpha,e_\beta]=N_{\alpha\beta}e_{\alpha+\beta} [ e α , e β ] = N α β e α + β together with e α e α + β = e β e α + β = e β e α e α + β = 0 e_\alpha e_{\alpha+\beta}=e_\beta e_{\alpha+\beta}=e_\beta e_\alpha e_{\alpha+\beta}=0 e α e α + β = e β e α + β = e β e α e α + β = 0 , which is checked for all 1440 ordered pairs. Reduction modulo p p p preserves integer identities, so (1)–(4) and (7) hold over every F q \mathbb{F}_q F q .
Let G G G be the subgroup of S L 27 ( q ) SL_{27}(q) S L 27 ( q ) generated by all x β ( t ) x_\beta(t) x β ( t ) . It is the simply connected Chevalley group of type E 6 E_6 E 6 [21]. Its center is Z = ⟨ z ⟩ Z=\langle z\rangle Z = ⟨ z ⟩ , of order g 3 = gcd ( 3 , q − 1 ) g_3=\gcd(3,q-1) g 3 = g cd( 3 , q − 1 ) , and E 6 ( q ) = G / Z E_6(q)=G/Z E 6 ( q ) = G / Z is simple. Let U U U be the group generated by the positive x β x_\beta x β , H H H the diagonal group of the h i h_i h i , and N ( w ) = { β > 0 : w − 1 β < 0 } N(w)=\{\beta>0:w^{-1}\beta<0\} N ( w ) = { β > 0 : w − 1 β < 0 } .
Theorem 12.88 (Digital law of E 6 ( q ) E_6(q) E 6 ( q ) ). Order W W W by a breadth-first search from 1 through the simple reflections, and give each w w w the reduced word w = s i 1 ⋯ s i ℓ w=s_{i_1}\cdots s_{i_\ell} w = s i 1 ⋯ s i ℓ along which it is first reached. Put n w = n i 1 ⋯ n i ℓ n_w=n_{i_1}\cdots n_{i_\ell} n w = n i 1 ⋯ n i ℓ . Order the positive roots by height. Fix a generator γ \gamma γ of F q × \mathbb{F}_q^\times F q × .
Every g ∈ G g\in G g ∈ G has a unique expression
g = ∏ β ∈ N ( w ) x β ( t β ) ⋅ n w ⋅ ∏ i = 1 6 h i ( γ e i ) ⋅ ∏ β > 0 x β ( t β ′ ) , g=\prod_{\beta\in N(w)}x_\beta(t_\beta)\cdot n_w\cdot\prod_{i=1}^{6}h_i(\gamma^{e_i})\cdot\prod_{\beta>0}x_\beta(t'_\beta), g = β ∈ N ( w ) ∏ x β ( t β ) ⋅ n w ⋅ i = 1 ∏ 6 h i ( γ e i ) ⋅ β > 0 ∏ x β ( t β ′ ) , with both products in height order, t β , t β ′ ∈ F q t_\beta,t'_\beta\in\mathbb{F}_q t β , t β ′ ∈ F q and e i ∈ A q − 1 e_i\in A_{q-1} e i ∈ A q − 1 . Multiplying g g g by z z z adds q − 1 3 k \frac{q-1}{3}k 3 q − 1 k to e e e . So when g 3 = 3 g_3=3 g 3 = 3 there is a unique representative of g Z gZ g Z with e 1 < q − 1 3 e_1<\frac{q-1}{3} e 1 < 3 q − 1 . Let u ˉ \bar u u ˉ , e ˉ \bar e e ˉ and v ˉ \bar v v ˉ be the mixed-radix numbers formed by ( t β ) β ∈ N ( w ) (t_\beta)_{\beta\in N(w)} ( t β ) β ∈ N ( w ) (radix q q q ), by e e e (radix ( q − 1 ) / g 3 (q-1)/g_3 ( q − 1 ) / g 3 for e 1 e_1 e 1 and q − 1 q-1 q − 1 for the others), and by ( t β ′ ) (t'_\beta) ( t β ′ ) (radix q q q ), each with its first entry least significant. Then
λ ( g Z ) = O w + u ˉ + q ℓ ( w ) ( e ˉ + ( q − 1 ) 6 g 3 v ˉ ) , O w = ( q − 1 ) 6 q 36 g 3 ∑ w ′ < w q ℓ ( w ′ ) . \lambda(gZ)=O_w+\bar u+q^{\ell(w)}\left(\bar e+\frac{(q-1)^6}{g_3}\bar v\right),\qquad O_w=\frac{(q-1)^6q^{36}}{g_3}\sum_{w'<w}q^{\ell(w')}. λ ( g Z ) = O w + u ˉ + q ℓ ( w ) ( e ˉ + g 3 ( q − 1 ) 6 v ˉ ) , O w = g 3 ( q − 1 ) 6 q 36 w ′ < w ∑ q ℓ ( w ′ ) . is a bijection E 6 ( q ) → A ∣ E 6 ( q ) ∣ E_6(q)\to A_{|E_6(q)|} E 6 ( q ) → A ∣ E 6 ( q ) ∣ with λ ( 1 ) = 0 \lambda(1)=0 λ ( 1 ) = 0 , and m ∗ n = λ ( λ − 1 ( m ) λ − 1 ( n ) ) m*n=\lambda(\lambda^{-1}(m)\lambda^{-1}(n)) m ∗ n = λ ( λ − 1 ( m ) λ − 1 ( n )) is a digital law of E 6 ( q ) E_6(q) E 6 ( q ) . The coordinates of a unipotent matrix M = ∏ β x β ( t β ) M=\prod_\beta x_\beta(t_\beta) M = ∏ β x β ( t β ) are read off by peeling in height order: t β = ± M ν μ t_\beta=\pm M_{\nu\mu} t β = ± M ν μ for any entry where e β e_\beta e β is nonzero, after the factors of smaller height have been removed on the left.
Proof. G G G has a split B N BN B N -pair with B = U H B=UH B = U H and Weyl group W W W , and n w n_w n w does not depend on the reduced word, by the braid relations of Proposition 12.87. The refined Bruhat decomposition then gives the unique expression [21]. Peeling works because a product of two or more e e e ’s raises height by at least 2. The map e ↦ ∏ i h i ( γ e i ) e\mapsto\prod_i h_i(\gamma^{e_i}) e ↦ ∏ i h i ( γ e i ) has kernel trivial modulo Z Z Z , since the weights of Λ \Lambda Λ span the weight lattice. Counting gives ∣ G / Z ∣ = q 36 ( q − 1 ) 6 ∑ w q ℓ ( w ) / g 3 = q 36 ∏ d ( q d − 1 ) / g 3 |G/Z|=q^{36}(q-1)^6\sum_w q^{\ell(w)}/g_3=q^{36}\prod_d(q^d-1)/g_3 ∣ G / Z ∣ = q 36 ( q − 1 ) 6 ∑ w q ℓ ( w ) / g 3 = q 36 ∏ d ( q d − 1 ) / g 3 , which is ∣ E 6 ( q ) ∣ |E_6(q)| ∣ E 6 ( q ) ∣ .
Remark 12.89 (Computing the law). To form g 1 g 2 g_1g_2 g 1 g 2 , right-multiply g 1 = u n w h v g_1=un_whv g 1 = u n w h v successively by u 2 u_2 u 2 , by the letters n i n_i n i of w 2 w_2 w 2 , by h 2 h_2 h 2 and by v 2 v_2 v 2 . Right multiplication by U U U or H H H only changes ( h , v ) (h,v) ( h , v ) . For n i n_i n i , split v = x α i ( t ) v ′ v=x_{\alpha_i}(t)v' v = x α i ( t ) v ′ and move v ′ v' v ′ through n i n_i n i . If w α i > 0 w\alpha_i>0 w α i > 0 , the factor n w x α i ( ⋅ ) n w − 1 n_wx_{\alpha_i}(\cdot)n_w^{-1} n w x α i ( ⋅ ) n w − 1 joins u u u and w w w becomes w s i ws_i w s i . If w α i < 0 w\alpha_i<0 w α i < 0 , write w = w ′ s i w=w's_i w = w ′ s i , split off the x w ′ α i x_{w'\alpha_i} x w ′ α i -component of u u u , and use n i 2 = h i ( − 1 ) n_i^2=h_i(-1) n i 2 = h i ( − 1 ) together with the identity x − α ( c ) = x α ( c − 1 ) n α h α ( − c ) x α ( c − 1 ) x_{-\alpha}(c)=x_\alpha(c^{-1})n_\alpha h_\alpha(-c)x_\alpha(c^{-1}) x − α ( c ) = x α ( c − 1 ) n α h α ( − c ) x α ( c − 1 ) for c ≠ 0 c\ne0 c = 0 from S L 2 SL_2 S L 2 .
Remark 12.90 (Checks). scripts/e6_law.py (prime q q q ) and scripts/e6_lawq.py (any q q q , with F q = F p [ x ] / ( f ) \mathbb{F}_q=\mathbb{F}_p[x]/(f) F q = F p [ x ] / ( f ) for f = x 2 + x + 1 f=x^2+x+1 f = x 2 + x + 1 , x 3 + x + 1 x^3+x+1 x 3 + x + 1 , x 2 + 1 x^2+1 x 2 + 1 when q = 4 , 8 , 9 q=4,8,9 q = 4 , 8 , 9 ) implement Theorem 12.88. For q = 2 , 3 , 4 , 5 , 7 , 8 , 9 , 13 q=2,3,4,5,7,8,9,13 q = 2 , 3 , 4 , 5 , 7 , 8 , 9 , 13 the number of codes equals ∣ E 6 ( q ) ∣ \lvert E_6(q)\rvert ∣ E 6 ( q )∣ (for q = 2 q=2 q = 2 this is 214 841 575 522 005 575 270 400 214\,841\,575\,522\,005\,575\,270\,400 214 841 575 522 005 575 270 400 ), and the identity has code 0. On random samples, codes survive decoding and re-encoding, products agree with the 27 × 27 27\times27 27 × 27 matrix products, and the law is associative. For q = 4 , 7 , 13 q=4,7,13 q = 4 , 7 , 13 , where g 3 = 3 g_3=3 g 3 = 3 , the element z z z is a nontrivial scalar and g g g and z g zg z g receive the same code.
The groups 2 E 6 ( q ) {}^{2}E_6(q) 2 E 6 ( q )
The group 2 E 6 ( q ) {}^{2}E_6(q) 2 E 6 ( q ) is the fixed-point group in E 6 ( q 2 ) E_6(q^2) E 6 ( q 2 ) of the graph automorphism ρ \rho ρ of the Dynkin diagram ( α 1 ↔ α 6 , α 3 ↔ α 5 ) (\alpha_1\leftrightarrow\alpha_6,\ \alpha_3\leftrightarrow\alpha_5) ( α 1 ↔ α 6 , α 3 ↔ α 5 ) , composed with the field automorphism t ↦ t q t\mapsto t^q t ↦ t q . The model of § 22.1 can be reused as it stands. The graph automorphism exchanges the 27-dimensional module with its dual, so on matrices it is an inverse transpose followed by conjugation by a signed permutation matrix.
Proposition 12.91 (The twist). Let n 0 = n w 0 n_0=n_{w_0} n 0 = n w 0 for the longest element w 0 ∈ W w_0\in W w 0 ∈ W . There is a diagonal matrix d d d with entries ± 1 \pm1 ± 1 such that P = n 0 d P=n_0d P = n 0 d has the following properties. Here τ ( g ) = P ( g T ) − 1 P − 1 \tau(g)=P(g^{T})^{-1}P^{-1} τ ( g ) = P ( g T ) − 1 P − 1 .
τ ( x β ( t ) ) = x ρ β ( c β t ) \tau(x_\beta(t))=x_{\rho\beta}(c_\beta t) τ ( x β ( t )) = x ρβ ( c β t ) for all 72 roots, with c β = ± 1 c_\beta=\pm1 c β = ± 1 . Moreover c β = 1 c_\beta=1 c β = 1 for the simple roots, their negatives and the roots fixed by ρ \rho ρ , and c β c ρ β = 1 c_\beta c_{\rho\beta}=1 c β c ρβ = 1 for all β \beta β .
P 2 = I P^2=I P 2 = I , and τ ( n i ) = n ρ i \tau(n_i)=n_{\rho i} τ ( n i ) = n ρ i for i = 1 , … , 6 i=1,\ldots,6 i = 1 , … , 6 .
ρ \rho ρ fixes 12 positive roots and exchanges the other 24 in 12 pairs { β , ρ β } \{\beta,\rho\beta\} { β , ρβ } . In each pair the two roots are orthogonal, and β + ρ β \beta+\rho\beta β + ρβ is not a root.
The centraliser W ρ W^\rho W ρ of ρ \rho ρ in W W W has order 1152. It is the Weyl group of type F 4 F_4 F 4 .
Proof. The identities in (1)–(3) are exact identities between integer matrices. The script scripts/e2e6_model.py checks each of them. The matrix d d d is found by propagating the sign conditions of (1) for the simple roots along the weight graph; they turn out to be consistent. Statement (4) is a direct enumeration of W W W . Since P P P is an integer matrix, (1)–(3) hold over every field.
Let G ⊂ S L 27 ( q 2 ) G\subset SL_{27}(q^2) G ⊂ S L 27 ( q 2 ) be the group of Section 12.13 over F q 2 \mathbb{F}_{q^2} F q 2 , and let σ ( g ) = τ ( g ) ( q ) \sigma(g)=\tau(g)^{(q)} σ ( g ) = τ ( g ) ( q ) , where ( q ) (q) ( q ) raises every entry to the power q q q . By Proposition 12.91, σ \sigma σ is an automorphism of order 2 that preserves U U U , H H H and every n w n_w n w with w ∈ W ρ w\in W^\rho w ∈ W ρ . Its fixed-point group G σ G^\sigma G σ is the simply connected group of type 2 E 6 {}^{2}E_6 2 E 6 over F q \mathbb{F}_q F q [21]. Its center Z σ Z^\sigma Z σ has order g 3 = gcd ( 3 , q + 1 ) g_3=\gcd(3,q+1) g 3 = g cd( 3 , q + 1 ) and is generated by z z z when g 3 = 3 g_3=3 g 3 = 3 , and 2 E 6 ( q ) = G σ / Z σ {}^{2}E_6(q)=G^\sigma/Z^\sigma 2 E 6 ( q ) = G σ / Z σ is simple.
Theorem 12.92 (Digital law of 2 E 6 ( q ) {}^{2}E_6(q) 2 E 6 ( q ) ). Order the 24 root classes of Proposition 12.91 (3) by height. Put X c ( t ) = x β ( t ) X_c(t)=x_\beta(t) X c ( t ) = x β ( t ) with t ∈ F q t\in\mathbb{F}_q t ∈ F q for a fixed root c = { β } c=\{\beta\} c = { β } . Put X c ( t ) = x β ( t ) x ρ β ( c β t q ) X_c(t)=x_\beta(t)x_{\rho\beta}(c_\beta t^q) X c ( t ) = x β ( t ) x ρβ ( c β t q ) with t ∈ F q 2 t\in\mathbb{F}_{q^2} t ∈ F q 2 for a pair c = { β , ρ β } c=\{\beta,\rho\beta\} c = { β , ρβ } , where β \beta β is the earlier root. Let γ \gamma γ generate F q 2 × \mathbb{F}_{q^2}^{\times} F q 2 × . Every g ∈ G σ g\in G^\sigma g ∈ G σ has a unique expression
g = ∏ c ∈ N σ ( w ) X c ( t c ) ⋅ n w ⋅ h ( e ) ⋅ ∏ c X c ( t c ′ ) , w ∈ W ρ , g=\prod_{c\in N^\sigma(w)}X_c(t_c)\cdot n_w\cdot h(e)\cdot\prod_c X_c(t'_c),\qquad w\in W^\rho, g = c ∈ N σ ( w ) ∏ X c ( t c ) ⋅ n w ⋅ h ( e ) ⋅ c ∏ X c ( t c ′ ) , w ∈ W ρ , where N σ ( w ) N^\sigma(w) N σ ( w ) is the set of classes inverted by w − 1 w^{-1} w − 1 and h ( e ) = ∏ i h i ( γ e i ) h(e)=\prod_i h_i(\gamma^{e_i}) h ( e ) = ∏ i h i ( γ e i ) with e = ( e 1 , ( q + 1 ) a 2 , e 3 , ( q + 1 ) a 4 , q e 3 , q e 1 ) e=(e_1,(q+1)a_2,e_3,(q+1)a_4,qe_3,qe_1) e = ( e 1 , ( q + 1 ) a 2 , e 3 , ( q + 1 ) a 4 , q e 3 , q e 1 ) , e 1 , e 3 ∈ A q 2 − 1 e_1,e_3\in A_{q^2-1} e 1 , e 3 ∈ A q 2 − 1 and a 2 , a 4 ∈ A q − 1 a_2,a_4\in A_{q-1} a 2 , a 4 ∈ A q − 1 . When g 3 = 3 g_3=3 g 3 = 3 , the coset g Z σ gZ^\sigma g Z σ has a unique representative with e 1 < ( q 2 − 1 ) / 3 e_1<(q^2-1)/3 e 1 < ( q 2 − 1 ) /3 . Let d ( w ) d(w) d ( w ) count the inverted fixed classes once and the inverted pairs twice. Let u ⃗ \vec{u} u , e ⃗ \vec{e} e and v ⃗ \vec{v} v be the mixed-radix numbers of the digits t c t_c t c (radix
q q q or q 2 q^{2} q 2 ), of ( e 1 , a 2 , e 3 , a 4 ) (e_{1},a_{2},e_{3},a_{4}) ( e 1 , a 2 , e 3 , a 4 ) (radices ( q 2 − 1 ) / g 3 (q^{2}-1)/g_{3} ( q 2 − 1 ) / g 3 , q − 1 q-1 q − 1 , q 2 − 1 q^{2}-1 q 2 − 1 , q − 1 q-1 q − 1 ) and of the t c t_{c} t c . A digit in F q \mathbb{F}_{q} F q is its position in the increasing list of the codes of the elements of the subfield. With T = ( q − 1 ) 2 ( q 2 − 1 ) 2 / g 3 T=(q-1)^{2}(q^{2}-1)^{2}/g_{3} T = ( q − 1 ) 2 ( q 2 − 1 ) 2 / g 3 and W ρ W^{\rho} W ρ in the order inherited from W W W ,
λ ( g Z σ ) = O w + u ˉ + q d ( w ) ( e ˉ + T v ˉ ) , O w = T q 36 ∑ w ′ < w q d ( w ′ ) . \lambda(gZ^{\sigma})=O_{w}+\bar{u}+q^{d(w)}(\bar{e}+T\bar{v}),\qquad O_{w}=Tq^{36}\sum_{w'<w}q^{d(w')}. λ ( g Z σ ) = O w + u ˉ + q d ( w ) ( e ˉ + T v ˉ ) , O w = T q 36 w ′ < w ∑ q d ( w ′ ) .
is a bijection 2 E 6 ( q ) → A ∣ 2 E 6 ( q ) ∣ {}^{2}E_{6}(q)\to A_{\lvert{}^{2}E_{6}(q)\rvert} 2 E 6 ( q ) → A ∣ 2 E 6 ( q )∣ with λ ( 1 ) = 0 \lambda(1)=0 λ ( 1 ) = 0 , and m ∗ n = λ ( λ − 1 ( m ) λ − 1 ( n ) ) m*n=\lambda(\lambda^{-1}(m)\lambda^{-1}(n)) m ∗ n = λ ( λ − 1 ( m ) λ − 1 ( n )) is a digital law. Products are computed in E 6 ( q 2 ) E_{6}(q^{2}) E 6 ( q 2 ) by Remark 12.89.
Proof. The factors of the normal form of Theorem 12.88 are unique and σ \sigma σ preserves each kind of factor, so for g ∈ G σ g\in G^{\sigma} g ∈ G σ every factor is σ \sigma σ -fixed, and w ∈ W ρ w\in W^{\rho} w ∈ W ρ . In a σ \sigma σ -fixed unipotent element the lowest-height coordinates are additive and are permuted by σ \sigma σ . Peeling class by class therefore gives t β ∈ F q t_{\beta}\in\mathbb{F}_{q} t β ∈ F q for a fixed root (c β = 1 c_{\beta}=1 c β = 1 ) and t ρ β = c β t β q t_{\rho\beta}=c_{\beta}t_{\beta}^{q} t ρβ = c β t β q for a pair, and each removed factor X c ( t ) X_{c}(t) X c ( t ) is itself σ \sigma σ -fixed. The σ \sigma σ -fixed torus elements are exactly the h ( e ) h(e) h ( e ) above, because σ ( h i ( c ) ) = h ρ i ( c q ) \sigma(h_{i}(c))=h_{\rho i}(c^{q}) σ ( h i ( c )) = h ρ i ( c q ) . Counting gives ∣ G σ / Z σ ∣ = q 36 ( q − 1 ) 2 ( q 2 − 1 ) 2 ∑ w ∈ W ρ q d ( w ) / g 3 \lvert G^{\sigma}/Z^{\sigma}\rvert=q^{36}(q-1)^{2}(q^{2}-1)^{2}\sum_{w\in W^{\rho}}q^{d(w)}/g_{3} ∣ G σ / Z σ ∣ = q 36 ( q − 1 ) 2 ( q 2 − 1 ) 2 ∑ w ∈ W ρ q d ( w ) / g 3 , which is ∣ 2 E 6 ( q ) ∣ \lvert{}^{2}E_{6}(q)\rvert ∣ 2 E 6 ( q )∣ by the twisted Bruhat decomposition [21].
Remark 12.93 (Checks). scripts/e2e6_law.py implements Theorem 12.92 . It is built on scripts/e6_lawq.py. The checks use q = 2 , 3 , 4 , 5 q=2,3,4,5 q = 2 , 3 , 4 , 5 , over F 4 , F 9 , F 16 , F 25 \mathbb{F}_{4},\mathbb{F}_{9},\mathbb{F}_{16},\mathbb{F}_{25} F 4 , F 9 , F 16 , F 25 . The last two fields are F 2 [ x ] / ( x 4 + x + 1 ) \mathbb{F}_{2}[x]/(x^{4}+x+1) F 2 [ x ] / ( x 4 + x + 1 ) and F 5 [ x ] / ( x 2 + x + 2 ) \mathbb{F}_{5}[x]/(x^{2}+x+2) F 5 [ x ] / ( x 2 + x + 2 ) . In each case the number of codes equals ∣ 2 E 6 ( q ) ∣ \lvert{}^{2}E_{6}(q)\rvert ∣ 2 E 6 ( q )∣ , and in particular q 36 ( q − 1 ) 2 ( q 2 − 1 ) 2 ∑ w q d ( w ) / g 3 q^{36}(q-1)^{2}(q^{2}-1)^{2}\sum_{w}q^{d(w)}/g_{3} q 36 ( q − 1 ) 2 ( q 2 − 1 ) 2 ∑ w q d ( w ) / g 3 agrees with the order formula. The identity has code 0; codes survive decoding and re-encoding; and decoded elements are σ \sigma σ -fixed, checked as g T P g ( q ) − 1 P = I g^{T}P^{-1}_{g^{(q)}}P=I g T P g ( q ) − 1 P = I . Products agree with matrix products in S L 27 ( q 2 ) SL_{27}(q^{2}) S L 27 ( q 2 ) , and the law is associative on samples. For q = 2 q=2 q = 2 and q = 5 q=5 q = 5 , where g 3 = 3 g_{3}=3 g 3 = 3 , the element z z z is σ \sigma σ -fixed and g g g and z g zg z g receive the same code.
The groups E 7 ( q ) E_{7}(q) E 7 ( q )
The group E 7 ( q ) E_{7}(q) E 7 ( q ) acts on a module of dimension 56 whose weights form a single Weyl orbit, the orbit of the fundamental weight ω 7 \omega_{7} ω 7 . As for E 6 E_{6} E 6 , every root element acts on this minuscule module by a matrix with entries 0 , ± 1 0,\pm1 0 , ± 1 and square zero, so x β ( t ) = I + t e β x_{\beta}(t)=I+te_{\beta} x β ( t ) = I + t e β and no division is needed in any characteristic. Two features are new. The module is self-dual, so the group lies in a symplectic group. The Weyl group, of order 2 903 040, is handled through a chain of parabolic subgroups instead of being enumerated.
Proposition 12.94 (Integral model of E 7 E_{7} E 7 ). Label the simple roots as in Bourbaki, with α 2 \alpha_{2} α 2 attached to α 4 \alpha_{4} α 4 . Let Λ \Lambda Λ be the free Z \mathbb{Z} Z -module with basis v μ v_{\mu} v μ , indexed by the 56 weights μ ∈ W ω 7 \mu\in W\omega_{7} μ ∈ W ω 7 . There are matrices e β e_{\beta} e β , β ∈ Φ \beta\in\Phi β ∈ Φ , on Λ \Lambda Λ such that:
every entry of e β e_{\beta} e β is 0 or ± 1 \pm1 ± 1 , and e β 2 = 0 e_{\beta}^{2}=0 e β 2 = 0 ;
for α ≠ ± β \alpha\ne\pm\beta α = ± β , [ e α , e β ] = N α β e α + β [e_{\alpha},e_{\beta}]=N_{\alpha\beta}e_{\alpha+\beta} [ e α , e β ] = N α β e α + β with N α β = ± 1 N_{\alpha\beta}=\pm1 N α β = ± 1 if α + β ∈ Φ \alpha+\beta\in\Phi α + β ∈ Φ , and [ e α , e β ] = 0 [e_{\alpha},e_{\beta}]=0 [ e α , e β ] = 0 otherwise;
[ e α , e − α ] v μ = ⟨ μ , α ∨ ⟩ v μ [e_{\alpha},e_{-\alpha}]v_{\mu}=\langle\mu,\alpha^{\vee}\rangle v_{\mu} [ e α , e − α ] v μ = ⟨ μ , α ∨ ⟩ v μ for every α > 0 \alpha>0 α > 0 , and the e β e_{\beta} e β together with these coroots span a Lie algebra of dimension 133;
n α = x α ( 1 ) x − α ( − 1 ) x α ( 1 ) n_{\alpha}=x_{\alpha}(1)x_{-\alpha}(-1)x_{\alpha}(1) n α = x α ( 1 ) x − α ( − 1 ) x α ( 1 ) is a signed permutation matrix sending v μ v_{\mu} v μ to ± v s α μ \pm v_{s_{\alpha}\mu} ± v s α μ ; the n i = n α i n_{i}=n_{\alpha_{i}} n i = n α i satisfy the braid relations of E 7 E_{7} E 7 , and n i 2 = h i ( − 1 ) n_{i}^{2}=h_{i}(-1) n i 2 = h i ( − 1 ) , where h i ( c ) v μ = c ⟨ μ , α i ∨ ⟩ v μ h_{i}(c)v_{\mu}=c^{\langle\mu,\alpha_{i}^{\vee}\rangle}v_{\mu} h i ( c ) v μ = c ⟨ μ , α i ∨ ⟩ v μ ;
up to a scalar there is exactly one bilinear form J J J pairing each v μ v_{\mu} v μ with v − μ v_{-\mu} v − μ and satisfying e β J + J e β = 0 e_{\beta}J+Je_{\beta}=0 e β J + J e β = 0 for all β \beta β , and it is alternating.
Proof. Every α \alpha α -string through a weight of a minuscule module has length at most 2, which gives e β 2 = 0 e_{\beta}^{2}=0 e β 2 = 0 . The remaining statements are integer identities and were checked directly. Item (2) was checked for all 126 ⋅ 124 = 15 624 126\cdot124=15\,624 126 ⋅ 124 = 15 624 ordered pairs, together with e α e α + β = e β e α + β = e β e α e α + β = 0 e_{\alpha}e_{\alpha+\beta}=e_{\beta}e_{\alpha+\beta}=e_{\beta}e_{\alpha}e_{\alpha+\beta}=0 e α e α + β = e β e α + β = e β e α e α + β = 0 when α + β ∈ Φ \alpha+\beta\in\Phi α + β ∈ Φ . Item (3) was checked for all 63 positive roots, and item (4) for all 63 matrices n α n_{\alpha} n α and all 21 pairs of simple reflections. Item (5) follows by solving the linear invariance conditions. Since e β J e β = − J e β 2 = 0 e_{\beta}Je_{\beta}=-Je_{\beta}^{2}=0 e β J e β = − J e β 2 = 0 , every x β ( t ) x_{\beta}(t) x β ( t ) preserves J J J . Reduction modulo p p p preserves integer identities, so (1)–(5) hold over every F q \mathbb{F}_{q} F q .
Let G G G be the subgroup of G L 56 ( q ) GL_{56}(q) G L 56 ( q ) generated by all x β ( t ) x_{\beta}(t) x β ( t ) . By Proposition 12.94(5), G ≤ S p 56 ( q ) G\leq Sp_{56}(q) G ≤ S p 56 ( q ) . It is the simply connected Chevalley group of type E 7 E_{7} E 7 [21]. The vector ( 0 , 1 , 0 , 0 , 1 , 0 , 1 ) (0,1,0,0,1,0,1) ( 0 , 1 , 0 , 0 , 1 , 0 , 1 ) is the only nonzero k ∈ { 0 , 1 } 7 k\in\{0,1\}^{7} k ∈ { 0 , 1 } 7 for which ∑ k i ⟨ μ , α i ∨ ⟩ \sum k_{i}\langle\mu,\alpha_{i}^{\vee}\rangle ∑ k i ⟨ μ , α i ∨ ⟩ has the same parity for all 56 weights, and that parity is odd. Hence z = h 2 ( − 1 ) h 5 ( − 1 ) h 7 ( − 1 ) = − I z=h_{2}(-1)h_{5}(-1)h_{7}(-1)=-I z = h 2 ( − 1 ) h 5 ( − 1 ) h 7 ( − 1 ) = − I . The center is Z = ⟨ z ⟩ Z=\langle z\rangle Z = ⟨ z ⟩ , of order g 2 = gcd ( 2 , q − 1 ) g_{2}=\gcd(2,q-1) g 2 = g cd( 2 , q − 1 ) , and E 7 ( q ) = G / Z E_{7}(q)=G/Z E 7 ( q ) = G / Z is simple. Let U U U be the group generated by the positive x β x_{\beta} x β , H H H the diagonal group of the h i h_{i} h i , and N ( w ) = { β > 0 : w − 1 β < 0 } N(w)=\{\beta>0:w^{-1}\beta<0\} N ( w ) = { β > 0 : w − 1 β < 0 } .
For 0 ≤ k ≤ 7 0\leq k\leq7 0 ≤ k ≤ 7 put W k = ⟨ s 1 , … , s k ⟩ W_{k}=\langle s_{1},\ldots,s_{k}\rangle W k = ⟨ s 1 , … , s k ⟩ , so that W 7 = W W_{7}=W W 7 = W and W 0 = 1 W_{0}=1 W 0 = 1 . The types are
E 7 ⊃ E 6 ⊃ D 5 ⊃ A 4 ⊃ A 2 A 1 ⊃ A 1 A 1 ⊃ A 1 ⊃ 1. E_{7}\supset E_{6}\supset D_{5}\supset A_{4}\supset A_{2}A_{1}\supset A_{1}A_{1}\supset A_{1}\supset1. E 7 ⊃ E 6 ⊃ D 5 ⊃ A 4 ⊃ A 2 A 1 ⊃ A 1 A 1 ⊃ A 1 ⊃ 1.
The stabiliser of ω k \omega_{k} ω k in W k W_{k} W k is W k − 1 W_{k-1} W k − 1 , so the set D k D_{k} D k of minimal coset representatives of W k / W k − 1 W_{k}/W_{k-1} W k / W k − 1 is in bijection with the orbit W k ω k W_{k}\omega_{k} W k ω k . In every case ω k \omega_{k} ω k is minuscule for W k W_{k} W k , and the orbit sizes are 56, 27, 16, 10, 3, 2, 2. Their product is 2 903 040 = ∣ W ∣ 2\,903\,040=|W| 2 903 040 = ∣ W ∣ . List D k D_{k} D k by a breadth-first search on W k ω k W_{k}\omega_{k} W k ω k starting from ω k \omega_{k} ω k , trying s 1 , … , s k s_{1},\ldots,s_{k} s 1 , … , s k in increasing order. If μ ′ = s i μ \mu'=s_{i}\mu μ ′ = s i μ is first reached from μ \mu μ , the word of μ ′ \mu' μ ′ is s i s_{i} s i followed by the word of μ \mu μ . Every w ∈ W w\in W w ∈ W then has a unique factorisation
w = d 7 d 6 ⋯ d 1 , d k ∈ D k , ℓ ( w ) = ∑ k ℓ ( d k ) . w=d_{7}d_{6}\cdots d_{1},\qquad d_{k}\in D_{k},\qquad\ell(w)=\sum_{k}\ell(d_{k}). w = d 7 d 6 ⋯ d 1 , d k ∈ D k , ℓ ( w ) = k ∑ ℓ ( d k ) .
and joining the words of d 7 , … , d 1 d_{7},\ldots,d_{1} d 7 , … , d 1 gives a reduced word w = s i 1 ⋯ s i ℓ w=s_{i_{1}}\cdots s_{i_{\ell}} w = s i 1 ⋯ s i ℓ . Put n w = n i 1 ⋯ n i ℓ n_{w}=n_{i_{1}}\cdots n_{i_{\ell}} n w = n i 1 ⋯ n i ℓ . Let j k j_{k} j k be the position of d k d_{k} d k in D k D_{k} D k , and order W W W lexicographically by ( j 7 , … , j 1 ) (j_{7},\ldots,j_{1}) ( j 7 , … , j 1 ) .
Theorem 12.95 (Digital law of E 7 ( q ) E_{7}(q) E 7 ( q ) ). Order the positive roots by height and fix a generator γ \gamma γ of F q × \mathbb{F}_{q}^{\times} F q × . Every g ∈ G g\in G g ∈ G has a unique expression
g = ∏ β ∈ N ( w ) x β ( t β ) ⋅ n w ⋅ ∏ i = 1 7 h i ( γ e i ) ⋅ ∏ β > 0 x β ( t β ′ ) , g=\prod_{\beta\in N(w)}x_{\beta}(t_{\beta})\cdot n_{w}\cdot\prod_{i=1}^{7}h_{i}(\gamma^{e_{i}})\cdot\prod_{\beta>0}x_{\beta}(t'_{\beta}), g = β ∈ N ( w ) ∏ x β ( t β ) ⋅ n w ⋅ i = 1 ∏ 7 h i ( γ e i ) ⋅ β > 0 ∏ x β ( t β ′ ) , *with both products in height order, t β , t β ′ ∈ F q t_{\beta},t'_{\beta}\in\mathbb{F}_{q} t β , t β ′ ∈ F q and e i ∈ A q − 1 e_{i}\in A_{q-1} e i ∈ A q − 1 . Multiplying g g g by z z z adds q − 1 2 \frac{q-1}{2} 2 q − 1 to e 2 e_{2} e 2 , e 5 e_{5} e 5 and e 7 e_{7} e 7 . So when g 2 = 2 g_{2}=2 g 2 = 2 there is a unique representative of g Z gZ g Z with e 2 < q − 1 2 e_{2}<\frac{q-1}{2} e 2 < 2 q − 1 . Let u ˉ \bar{u} u ˉ , e ˉ \bar{e} e ˉ and v ˉ \bar{v} v ˉ be the mixed-radix numbers formed by ( t β ) β ∈ N ( w ) (t_{\beta})_{\beta\in N(w)} ( t β ) β ∈ N ( w ) (radix q q q ), by e e e (radix ( q − 1 ) / g 2 (q-1)/g_{2} ( q − 1 ) / g 2 for e 2 e_{2} e 2 and q − 1 q-1 q − 1 for the others), and by ( t β ′ ) (t'_{\beta}) ( t β ′ ) (radix q q q ), each with its first entry least significant. Put T = ( q − 1 ) 7 / g 2 T=(q-1)^{7}/g_{2} T = ( q − 1 ) 7 / g 2 ,
P i = ∑ d ∈ D i q ℓ ( d ) , P k < ( j ) = ∑ j ′ < j q ℓ ( d k , j ′ ) . P_{i}=\sum_{d\in D_{i}}q^{\ell(d)},\qquad P_{k}^{<}(j)=\sum_{j'<j}q^{\ell(d_{k,j'})}. P i = d ∈ D i ∑ q ℓ ( d ) , P k < ( j ) = j ′ < j ∑ q ℓ ( d k , j ′ ) . where d k , j d_{k,j} d k , j is the j j j -th element of D k D_{k} D k . Then
λ ( g Z ) = O w + u ˉ + q ℓ ( w ) ( e ˉ + T v ˉ ) , \lambda(gZ)=O_{w}+\bar{u}+q^{\ell(w)}(\bar{e}+T\bar{v}), λ ( g Z ) = O w + u ˉ + q ℓ ( w ) ( e ˉ + T v ˉ ) , O w = T q 63 ∑ w ′ < w q ℓ ( w ′ ) = T q 63 ∑ k = 1 7 q ℓ ( d 7 ) + ⋯ + ℓ ( d k + 1 ) P k < ( j k ) ∏ i < k P i , O_{w}=Tq^{63}\sum_{w'<w}q^{\ell(w')}=Tq^{63}\sum_{k=1}^{7}q^{\ell(d_{7})+\cdots+\ell(d_{k+1})}P_{k}^{<}(j_{k})\prod_{i<k}P_{i}, O w = T q 63 w ′ < w ∑ q ℓ ( w ′ ) = T q 63 k = 1 ∑ 7 q ℓ ( d 7 ) + ⋯ + ℓ ( d k + 1 ) P k < ( j k ) i < k ∏ P i , is a bijection E 7 ( q ) → A ∣ E 7 ( q ) ∣ E_{7}(q)\to A_{|E_{7}(q)|} E 7 ( q ) → A ∣ E 7 ( q ) ∣ with λ ( 1 ) = 0 \lambda(1)=0 λ ( 1 ) = 0 , and m ∗ n = λ ( λ − 1 ( m ) λ − 1 ( n ) ) m*n=\lambda(\lambda^{-1}(m)\lambda^{-1}(n)) m ∗ n = λ ( λ − 1 ( m ) λ − 1 ( n )) is a digital law of E 7 ( q ) E_{7}(q) E 7 ( q ) .
Proof. G G G has a split B N BN B N -pair with B = U H B=UH B = U H and Weyl group W W W . By the braid relations of Proposition 12.94 , n w n_w n w does not depend on the reduced word, and the refined Bruhat decomposition gives the unique expression [21]. The map e ↦ ∏ h i ( γ e i ) e \mapsto\prod h_i(\gamma^{e_i}) e ↦ ∏ h i ( γ e i ) is injective on A q − 1 7 A^7_{q-1} A q − 1 7 , since the weights of Λ \Lambda Λ span the weight lattice. Modulo Z Z Z , the condition e 2 < q − 1 2 e_2 < \frac{q-1}{2} e 2 < 2 q − 1 picks exactly one of g g g and z g zg z g . The elements w ′ < w w' < w w ′ < w are of two kinds. For each k k k , there are those that agree with w w w in d 7 , … , d k + 1 d_7,\ldots,d_{k+1} d 7 , … , d k + 1 and have j k ′ < j k j'_k < j_k j k ′ < j k , with d k − 1 ′ , … , d 1 ′ d'_{k-1},\ldots,d'_1 d k − 1 ′ , … , d 1 ′ arbitrary. Since lengths add, the elements of this kind contribute q ℓ ( d 7 ) + ⋯ + ℓ ( d k + 1 ) P k < ( j k ) ∏ i < k P i q^{\ell(d_7)+\cdots+\ell(d_{k+1})}P_k^<(j_k)\prod_{i<k}P_i q ℓ ( d 7 ) + ⋯ + ℓ ( d k + 1 ) P k < ( j k ) ∏ i < k P i , which gives the second expression for O w O_w O w . Finally, ∏ i P i = ∑ w q ℓ ( w ) = ∏ d q d − 1 q − 1 \prod_i P_i=\sum_w q^{\ell(w)}=\prod_d \frac{q^d-1}{q-1} ∏ i P i = ∑ w q ℓ ( w ) = ∏ d q − 1 q d − 1 over d ∈ { 2 , 6 , 8 , 10 , 12 , 14 , 18 } d\in\{2,6,8,10,12,14,18\} d ∈ { 2 , 6 , 8 , 10 , 12 , 14 , 18 } . Hence ∣ G / Z ∣ = q 63 ( q − 1 ) 7 ∏ i P i / g 2 = q 63 ∏ d ( q d − 1 ) / g 2 = ∣ E 7 ( q ) ∣ |G/Z|=q^{63}(q-1)^7\prod_i P_i/g^2=q^{63}\prod_d(q^d-1)/g^2=|E_7(q)| ∣ G / Z ∣ = q 63 ( q − 1 ) 7 ∏ i P i / g 2 = q 63 ∏ d ( q d − 1 ) / g 2 = ∣ E 7 ( q ) ∣ .
The groups E 8 ( q ) E_8(q) E 8 ( q )
Remark 12.96 (Computing the law). Products are formed exactly as in Remark 12.89, using right multiplication by U U U , by H H H and by the letters n i n_i n i , with the identity
x − α ( c ) = x α ( c − 1 ) n α h α ( − c ) x α ( c − 1 ) ( c ≠ 0 ) . x_{-\alpha}(c)=x_\alpha(c^{-1})n_\alpha h_\alpha(-c)x_\alpha(c^{-1})\qquad(c\ne0). x − α ( c ) = x α ( c − 1 ) n α h α ( − c ) x α ( c − 1 ) ( c = 0 ) .
The Weyl part of a code is decoded one factor at a time. At level k k k , the digit j k j_k j k is found by bisection in the partial sums P k < P_k^< P k < , so W W W is never listed. Torus coordinates are read off from seven weights whose coordinate matrix has determinant ± 1 \pm1 ± 1 . Unipotent coordinates are peeled in height order.
Remark 12.97 (Checks). Proposition 12.94 and Theorem 12.95 are implemented, for prime q q q , in the script scripts/e7_law.py. Running python3 e7_law.py q --ns 100 --assoc 20 --center 20 for q = 2 , 3 , 5 , 7 q=2,3,5,7 q = 2 , 3 , 5 , 7 gives:
every identity of Proposition 12.94 holds with no failures, and the invariant forms of (5) form a one-dimensional space spanned by an alternating form;
the coset sizes are ( 56 , 27 , 16 , 10 , 3 , 2 , 2 ) (56,27,16,10,3,2,2) ( 56 , 27 , 16 , 10 , 3 , 2 , 2 ) ;
the length generating function of the chain equals ∏ d ( 1 + q + ⋯ + q d − 1 ) \prod_d(1+q+\cdots+q^{d-1}) ∏ d ( 1 + q + ⋯ + q d − 1 ) ;
the number of codes equals ∣ E 7 ( q ) ∣ |E_7(q)| ∣ E 7 ( q ) ∣ , and the identity has code 0;
there were no failures on random samples: 100 codes survive decoding and re-encoding, 100 products agree with the 56 × 56 56\times56 56 × 56 matrix products, and 20 triples satisfy associativity;
for q = 3 , 5 , 7 q=3,5,7 q = 3 , 5 , 7 , the element z z z is the scalar − I -I − I , and g g g and z g zg z g receive the same code on 20 samples.
The group E 8 E_8 E 8 has no minuscule module. Its smallest nontrivial module is the adjoint module, of dimension 248, whose nonzero weights are the 240 roots. Root elements no longer square to zero, so x β ( t ) x_\beta(t) x β ( t ) is quadratic in t t t . The only denominator is a factor 1 2 \frac{1}{2} 2 1 , and it cancels over Z \mathbb{Z} Z . The root lattice of E 8 E_8 E 8 equals its weight lattice, so the group is both simply connected and adjoint, and its centre is trivial. The Weyl group, of order 696729600, is again handled through a chain of parabolic subgroups. The first step of the chain is the orbit of the highest root.
Proposition 12.98 (Integral adjoint model of E 8 E_8 E 8 ). Label the simple roots as in Bourbaki, with α 2 \alpha_2 α 2 attached to α 4 \alpha_4 α 4 , and let A = ( ⟨ α i , α j ∨ ⟩ ) A=(\langle\alpha_i,\alpha_j^\vee\rangle) A = (⟨ α i , α j ∨ ⟩) be the Cartan matrix. Let Λ \Lambda Λ be the free Z \mathbb{Z} Z -module with basis e γ e_\gamma e γ (γ ∈ Φ \gamma\in\Phi γ ∈ Φ ) and H 1 , … , H 8 H_1,\ldots,H_8 H 1 , … , H 8 , of rank 240 + 8 = 248 240+8=248 240 + 8 = 248 . For α ∨ = ∑ i c i α i ∨ \alpha^\vee=\sum_i c_i\alpha_i^\vee α ∨ = ∑ i c i α i ∨ put H α = ∑ i c i H i H_\alpha=\sum_i c_iH_i H α = ∑ i c i H i . Choose N α β = ± 1 N_{\alpha\beta}=\pm1 N α β = ± 1 by the Frenkel–Kac sign cocycle on the root lattice, and let E β = ad e β E_\beta=\operatorname{ad}e_\beta E β = ad e β . Then:
the bracket given by [ e α , e β ] = N α β e α + β [e_{\alpha},e_{\beta}]=N_{\alpha\beta}e_{\alpha+\beta} [ e α , e β ] = N α β e α + β if α + β ∈ Φ \alpha+\beta\in\Phi α + β ∈ Φ , [ e α , e β ] = 0 [e_{\alpha},e_{\beta}]=0 [ e α , e β ] = 0 if α + β ∉ Φ ∪ { 0 } \alpha+\beta\notin\Phi\cup\{0\} α + β ∈ / Φ ∪ { 0 } , [ e α , e − α ] = H α [e_{\alpha},e_{-\alpha}]=H_{\alpha} [ e α , e − α ] = H α , [ H i , e γ ] = ⟨ γ , α i ∨ ⟩ e γ [H_i,e_{\gamma}]=\langle\gamma,\alpha_i^{\vee}\rangle e_{\gamma} [ H i , e γ ] = ⟨ γ , α i ∨ ⟩ e γ and [ H i , H j ] = 0 [H_i,H_j]=0 [ H i , H j ] = 0 is antisymmetric and satisfies the Jacobi identity, so Λ \Lambda Λ is a Z \mathbb{Z} Z -form of the Lie algebra of type E 8 E_8 E 8 ;
every E β E_{\beta} E β has integer entries and E β 3 = 0 E_{\beta}^{3}=0 E β 3 = 0 . Moreover E β 2 E_{\beta}^{2} E β 2 has exactly one nonzero entry, which sends e − β e_{-\beta} e − β to − 2 e β -2e_{\beta} − 2 e β . Hence
x β ( t ) = exp ( t E β ) = I + t E β + 1 2 t 2 E β 2 x_{\beta}(t)=\exp(tE_{\beta})=I+tE_{\beta}+\tfrac{1}{2}t^{2}E_{\beta}^{2} x β ( t ) = exp ( t E β ) = I + t E β + 2 1 t 2 E β 2
has integer entries, and x β ( s ) x β ( t ) = x β ( s + t ) x_{\beta}(s)x_{\beta}(t)=x_{\beta}(s+t) x β ( s ) x β ( t ) = x β ( s + t ) over Z [ s , t ] \mathbb{Z}[s,t] Z [ s , t ] ;
n i = x α i ( 1 ) x − α i ( − 1 ) x α i ( 1 ) n_i=x_{\alpha_i}(1)x_{-\alpha_i}(-1)x_{\alpha_i}(1) n i = x α i ( 1 ) x − α i ( − 1 ) x α i ( 1 ) sends e γ e_{\gamma} e γ to ± e s i γ \pm e_{s_i\gamma} ± e s i γ and preserves the span of the H j H_j H j . The n i n_i n i satisfy the braid relations of E 8 E_8 E 8 , and n i 2 = h i ( − 1 ) n_i^2=h_i(-1) n i 2 = h i ( − 1 ) , where h i ( c ) e γ = c ⟨ γ , α i ∨ ⟩ e γ h_i(c)e_{\gamma}=c^{\langle\gamma,\alpha_i^{\vee}\rangle}e_{\gamma} h i ( c ) e γ = c ⟨ γ , α i ∨ ⟩ e γ and h i ( c ) H j = H j h_i(c)H_j=H_j h i ( c ) H j = H j ;
over any field k k k , ∏ i h i ( c i ) = I \prod_i h_i(c_i)=I ∏ i h i ( c i ) = I with c i ∈ k × c_i\in k^{\times} c i ∈ k × only if c 1 = ⋯ = c 8 = 1 c_1=\cdots=c_8=1 c 1 = ⋯ = c 8 = 1 .
Proof. In a simply laced root system, 2 β + γ ∉ Φ 2\beta+\gamma\notin\Phi 2 β + γ ∈ / Φ for γ ≠ − β \gamma\ne-\beta γ = − β , so E β 2 e γ = 0 E_{\beta}^{2}e_{\gamma}=0 E β 2 e γ = 0 for those γ \gamma γ . Also E β 2 H i = 0 E_{\beta}^{2}H_i=0 E β 2 H i = 0 , and E β 2 e − β = E β H β = − ⟨ β , β ∨ ⟩ e β = − 2 e β E_{\beta}^{2}e_{-\beta}=E_{\beta}H_{\beta}=-\langle\beta,\beta^{\vee}\rangle e_{\beta}=-2e_{\beta} E β 2 e − β = E β H β = − ⟨ β , β ∨ ⟩ e β = − 2 e β . This gives the shape of E β 2 E_{\beta}^{2} E β 2 in (2). The remaining statements are integer identities and were checked directly. Antisymmetry was checked on all pairs of basis vectors. Jacobi was checked as the statement that ad x \operatorname{ad}x ad x is a derivation for the 16 generators x = e ± α i x=e_{\pm\alpha_i} x = e ± α i . This suffices, because the x x x for which ad x \operatorname{ad}x ad x is a derivation form a Lie subalgebra, and the e ± α i e_{\pm\alpha_i} e ± α i generate. For (4), ∏ i h i ( c i ) \prod_i h_i(c_i) ∏ i h i ( c i ) acts on e α j e_{\alpha_j} e α j by ∏ i c i A j i \prod_i c_i^{A_{ji}} ∏ i c i A j i . Since det A = 1 \det A=1 det A = 1 , this map ( k × ) 8 → ( k × ) 8 (k^{\times})^8\to(k^{\times})^8 ( k × ) 8 → ( k × ) 8 is an automorphism. Reduction modulo p p p preserves integer identities, so (1)--(4) hold over every F q \mathbb{F}_q F q .
Remark 12.99 (No separate model in characteristic 2). The factor 1 2 \tfrac{1}{2} 2 1 in x β ( t ) x_{\beta}(t) x β ( t ) multiplies E β 2 ∈ 2 M 248 ( Z ) E_{\beta}^{2}\in2M_{248}(\mathbb{Z}) E β 2 ∈ 2 M 248 ( Z ) . So the model of Proposition 12.98 is integral, and it is used unchanged in characteristic 2. For F 4 F_4 F 4 the Albert model of Definition 12.55 needs odd characteristic, and the adjoint model of Proposition 12.67 is needed for characteristics 2 and 3. For E 8 E_8 E 8 the adjoint model is the only model used, and it serves every characteristic.
Let G G G be the subgroup of G L 248 ( q ) GL_{248}(q) G L 248 ( q ) generated by all x β ( t ) x_{\beta}(t) x β ( t ) . It is the Chevalley group of type E 8 E_8 E 8 [21]. By Proposition 12.98(4) its centre is trivial, so E 8 ( q ) = G E_8(q)=G E 8 ( q ) = G is simple and no quotient is taken. Let U U U be the group generated by the positive x β x_{\beta} x β , H H H the diagonal group of the h i h_i h i , and N ( w ) = { β > 0 : w − 1 β < 0 } N(w)=\{\beta>0:w^{-1}\beta<0\} N ( w ) = { β > 0 : w − 1 β < 0 } .
For 0 ≤ k ≤ 8 0\le k\le8 0 ≤ k ≤ 8 put W k = ⟨ s 1 , … , s k ⟩ W_k=\langle s_1,\ldots,s_k\rangle W k = ⟨ s 1 , … , s k ⟩ , so that W 8 = W W_8=W W 8 = W and W 0 = 1 W_0=1 W 0 = 1 . The types are
E 8 ⊃ E 7 ⊃ E 6 ⊃ D 5 ⊃ A 4 ⊃ A 2 A 1 ⊃ A 1 A 1 ⊃ A 1 ⊃ 1 E_8\supset E_7\supset E_6\supset D_5\supset A_4\supset A_2A_1\supset A_1A_1\supset A_1\supset1 E 8 ⊃ E 7 ⊃ E 6 ⊃ D 5 ⊃ A 4 ⊃ A 2 A 1 ⊃ A 1 A 1 ⊃ A 1 ⊃ 1
The stabiliser of ω k \omega_k ω k in W k W_k W k is W k − 1 W_{k-1} W k − 1 , so the set D k D_k D k of minimal coset representatives of W k / W k − 1 W_k/W_{k-1} W k / W k − 1 is in bijection with the orbit W k ω k W_k\omega_k W k ω k . For k ≤ 7 k\le7 k ≤ 7 these are the minuscule orbits of Section 12.15. For k = 8 k=8 k = 8 , ω 8 \omega_8 ω 8 is the highest root, and W 8 = Φ W_8=\Phi W 8 = Φ . The orbit sizes are 240, 56, 27, 16, 10, 3, 2, 2, and their product is 696 729 600 = ∣ W ∣ =|W| = ∣ W ∣ . List each D k D_k D k by breadth-first search, as in Section 12.15. Every w ∈ W w\in W w ∈ W then has a unique factorisation
w = d 8 d 7 ⋯ d 1 , d k ∈ D k , ℓ ( w ) = ∑ k ℓ ( d k ) w=d_8d_7\cdots d_1,\qquad d_k\in D_k,\qquad\ell(w)=\sum_k\ell(d_k) w = d 8 d 7 ⋯ d 1 , d k ∈ D k , ℓ ( w ) = k ∑ ℓ ( d k )
and joining the words of d 8 , … , d 1 d_8,\ldots,d_1 d 8 , … , d 1 gives a reduced word w = s i 1 ⋯ s i ℓ w=s_{i_1}\cdots s_{i_\ell} w = s i 1 ⋯ s i ℓ . Put n w = n i 1 ⋯ n i ℓ n_w=n_{i_1}\cdots n_{i_\ell} n w = n i 1 ⋯ n i ℓ . Let j k j_k j k be the position of d k d_k d k in D k D_k D k , and order W W W lexicographically by ( j 8 , … , j 1 ) (j_8,\ldots,j_1) ( j 8 , … , j 1 ) .
Theorem 12.100 (Digital law of E 8 ( q ) E_8(q) E 8 ( q ) ). Order the positive roots by height and fix a generator γ \gamma γ of F q × \mathbb{F}_q^{\times} F q × . Every g ∈ G g\in G g ∈ G has a unique expression
g = ∏ β ∈ N ( w ) x β ( t β ) ⋅ n w ⋅ ∏ i = 1 8 h i ( γ e i ) ⋅ ∏ β > 0 x β ( t β ′ ) , g=\prod_{\beta\in N(w)}x_{\beta}(t_{\beta})\cdot n_w\cdot\prod_{i=1}^{8}h_i(\gamma^{e_i})\cdot\prod_{\beta>0}x_{\beta}(t'_{\beta}), g = β ∈ N ( w ) ∏ x β ( t β ) ⋅ n w ⋅ i = 1 ∏ 8 h i ( γ e i ) ⋅ β > 0 ∏ x β ( t β ′ ) ,
with both products in height order, t β , t β ′ ∈ F q t_{\beta},t'_{\beta}\in\mathbb{F}_{q} t β , t β ′ ∈ F q and e i ∈ A q − 1 e_i\in\mathcal{A}_{q-1} e i ∈ A q − 1 . Let u ˉ \bar{u} u ˉ , e ˉ \bar{e} e ˉ and v ˉ \bar{v} v ˉ be the mixed-radix numbers formed by ( t β ) β ∈ N ( w ) (t_{\beta})_{\beta\in N(w)} ( t β ) β ∈ N ( w ) (radix q q q ), by e e e (radix q − 1 q-1 q − 1 ) and by ( t β ′ ) (t'_{\beta}) ( t β ′ ) (radix q q q ), each with its first entry least significant. Put T = ( q − 1 ) 8 T=(q-1)^8 T = ( q − 1 ) 8 ,
P i = ∑ d ∈ D i q ℓ ( d ) , P k < ( j ) = ∑ j ′ < j q ℓ ( d k , j ′ ) . P_i=\sum_{d\in D_i}q^{\ell(d)},\qquad P_k^{<}(j)=\sum_{j'<j}q^{\ell(d_{k,j'})}. P i = d ∈ D i ∑ q ℓ ( d ) , P k < ( j ) = j ′ < j ∑ q ℓ ( d k , j ′ ) .
where d k , j ′ d_{k,j'} d k , j ′ is the j ′ j' j ′ -th element of D k D_k D k . Then
λ ( g ) = O w + u ˉ + q ℓ ( w ) ( e ˉ + T v ˉ ) , \lambda(g)=O_w+\bar{u}+q^{\ell(w)}(\bar{e}+T\bar{v}), λ ( g ) = O w + u ˉ + q ℓ ( w ) ( e ˉ + T v ˉ ) ,
O w = T q 120 ∑ w ′ < w q ℓ ( w ′ ) = T q 120 ∑ k = 1 8 q ℓ ( d 8 ) + ⋯ + ℓ ( d k + 1 ) P k < ( j k ) ∏ i < k P i O_w=Tq^{120}\sum_{w'<w}q^{\ell(w')}=Tq^{120}\sum_{k=1}^{8}q^{\ell(d_8)+\cdots+\ell(d_{k+1})}P_k^{<}(j_k)\prod_{i<k}P_i O w = T q 120 w ′ < w ∑ q ℓ ( w ′ ) = T q 120 k = 1 ∑ 8 q ℓ ( d 8 ) + ⋯ + ℓ ( d k + 1 ) P k < ( j k ) i < k ∏ P i
is a bijection E 8 ( q ) → A ∣ E 8 ( q ) ∣ E_8(q)\to\mathcal{A}_{|E_8(q)|} E 8 ( q ) → A ∣ E 8 ( q ) ∣ with λ ( 1 ) = 0 \lambda(1)=0 λ ( 1 ) = 0 , where
∣ E 8 ( q ) ∣ = q 120 ∏ d ∈ { 2 , 8 , 12 , 14 , 18 , 20 , 24 , 30 } ( q d − 1 ) , |E_8(q)|=q^{120}\prod_{d\in\{2,8,12,14,18,20,24,30\}}(q^d-1), ∣ E 8 ( q ) ∣ = q 120 d ∈ { 2 , 8 , 12 , 14 , 18 , 20 , 24 , 30 } ∏ ( q d − 1 ) ,
and m ∗ n = λ ( λ − 1 ( m ) λ − 1 ( n ) ) m*n=\lambda(\lambda^{-1}(m)\lambda^{-1}(n)) m ∗ n = λ ( λ − 1 ( m ) λ − 1 ( n )) is a digital law of E 8 ( q ) E_8(q) E 8 ( q ) .
Proof. G G G has a split B N BN B N -pair with B = U H B=UH B = U H and Weyl group W W W . By the braid relations of Proposition 12.98, n w n_w n w does not depend on the reduced word, and the refined Bruhat decomposition gives the unique expression [21]. The map e ↦ ∏ i h i ( γ e i ) e\mapsto\prod_i h_i(\gamma^{e_i}) e ↦ ∏ i h i ( γ e i ) is injective on A q − 1 8 \mathcal{A}_{q-1}^8 A q − 1 8 by Proposition 12.98(4). The formula for O w O_w O w follows as in the proof of Theorem 12.95. For each k k k , the w ′ w' w ′ < < < w w w that agree with w w w in d 8 , … , d k + 1 d_8,\ldots,d_{k+1} d 8 , … , d k + 1 and have j k ′ < j k j'_k<j_k j k ′ < j k contribute q ℓ ( d 8 ) + ⋯ + ℓ ( d k + 1 ) P k < ( j k ) ∏ i < k P i q^{\ell(d_8)+\cdots+\ell(d_{k+1})}P_k^{<}(j_k)\prod_{i<k}P_i q ℓ ( d 8 ) + ⋯ + ℓ ( d k + 1 ) P k < ( j k ) ∏ i < k P i . Finally, ∏ i P i = ∑ w q ℓ ( w ) = ∏ d ( q d − 1 ) / ( q − 1 ) \prod_iP_i=\sum_wq^{\ell(w)}=\prod_d(q^d-1)/(q-1) ∏ i P i = ∑ w q ℓ ( w ) = ∏ d ( q d − 1 ) / ( q − 1 ) over d ∈ { 2 , 8 , 12 , 14 , 18 , 20 , 24 , 30 } d\in\{2,8,12,14,18,20,24,30\} d ∈ { 2 , 8 , 12 , 14 , 18 , 20 , 24 , 30 } . There are eight degrees, so ∣ G ∣ = q 120 ( q − 1 ) 8 ∏ i P i = q 120 ∏ d ( q d − 1 ) = ∣ E 8 ( q ) ∣ |G|=q^{120}(q-1)^8\prod_iP_i=q^{120}\prod_d(q^d-1)=|E_8(q)| ∣ G ∣ = q 120 ( q − 1 ) 8 ∏ i P i = q 120 ∏ d ( q d − 1 ) = ∣ E 8 ( q ) ∣ .
Remark 12.101 (Computing the law). Products are formed as in Remarks 12.89 and 12.96, by right multiplication by U U U , by H H H and by the letters n i n_i n i . The Weyl digits j k j_k j k are again found by bisection in the partial sums P k < P_k^{<} P k < , so W W W is never listed. Products of 248 × 248 248\times248 248 × 248 matrices modulo q q q are computed in floating point. This is exact, because every entry of an unreduced product is at most 248 ( q − 1 ) 2 < 2 53 248(q-1)^2<2^{53} 248 ( q − 1 ) 2 < 2 53 . On e α j e_{\alpha_j} e α j , the torus part acts by γ f j \gamma^{f_j} γ f j with f = A e m o d ( q − 1 ) f=Ae\bmod(q-1) f = A e mod ( q − 1 ) . Since det A = 1 \det A=1 det A = 1 , e = A − 1 f m o d ( q − 1 ) e=A^{-1}f\bmod(q-1) e = A − 1 f mod ( q − 1 ) is read off from the eight simple root vectors.
Unipotent coordinates are peeled from the left in height order. In the adjoint model this needs a short argument, because x β ( t ) x_{\beta}(t) x β ( t ) is quadratic. Grade Λ \Lambda Λ by weight, giving the H i H_i H i weight 0 0 0 , and suppose that every remaining factor x β ′ ( t β ′ ) x_{\beta'}(t_{\beta'}) x β ′ ( t β ′ ) has ht β ′ ≥ ht β \operatorname{ht}\beta'\geq\operatorname{ht}\beta ht β ′ ≥ ht β . Then the quadratic term of x β x_{\beta} x β raises weights by 2 β 2\beta 2 β , and a product of terms from two or more factors raises them by height at least 2 ht β 2\operatorname{ht}\beta 2 ht β . So only the linear term t β E β t_{\beta}E_{\beta} t β E β contributes in weight shift β \beta β . Fix a root δ \delta δ such that E β e δ = ε b E_{\beta}e_{\delta}=\varepsilon b E β e δ = ε b for a single basis vector b b b and a sign ε \varepsilon ε . For example, take δ ≠ − β \delta\ne-\beta δ = − β with β + δ ∈ Φ \beta+\delta\in\Phi β + δ ∈ Φ , b = e β + δ b=e_{\beta+\delta} b = e β + δ and ε = N β δ \varepsilon=N_{\beta\delta} ε = N β δ ; for simple β \beta β , δ = − β \delta=-\beta δ = − β and b = H β b=H_{\beta} b = H β also work. Then t β t_{\beta} t β is ε \varepsilon ε times the ( b , e δ ) (b,e_{\delta}) ( b , e δ ) entry of the current matrix, and no other term of shift β \beta β meets that entry. Left multiplication by x β ( − t β ) x_{\beta}(-t_{\beta}) x β ( − t β ) removes the factor, and the argument repeats.
Remark 12.102 (Checks). Proposition 12.98 and Theorem 12.100 are implemented, for prime q q q , in the scripts scripts/e8_model.py and scripts/e8_law.py. Running python3 e8_model.py once, and python3 e8_law.py q --ns 100 --assoc 20 for q = 2 , 3 , 5 , 7 q=2,3,5,7 q = 2 , 3 , 5 , 7 , gives:
every identity of Proposition 12.98 holds with no failures, and the centre is trivial. Antisymmetry and the Jacobi identity are checked by e8_model.py; the identities for root elements, the n i n_i n i and the centre are checked again modulo q q q by e8_law.py;
the coset sizes are ( 240 , 56 , 27 , 16 , 10 , 3 , 2 , 2 ) (240,56,27,16,10,3,2,2) ( 240 , 56 , 27 , 16 , 10 , 3 , 2 , 2 ) , and ∣ W ∣ = 696729600 |W|=696729600 ∣ W ∣ = 696729600 ;
λ ( λ − 1 ( n ) ) = n \lambda(\lambda^{-1}(n))=n λ ( λ − 1 ( n )) = n for 100 random codes, ∗ \ast ∗ agrees with matrix multiplication on 100 random pairs, and ∗ \ast ∗ is associative on 20 random triples, all with no failures.
With the four values of q q q run in parallel on one machine, the wall-clock times for q = 2 , 3 , 5 , 7 q=2,3,5,7 q = 2 , 3 , 5 , 7 were 657, 486, 338 and 311 s, respectively.
Exercises
Show that ∣ P S L ( 2 , p ) ∣ = p ( p 2 − 1 ) / 2 |\mathrm{PSL}(2,p)|=p(p^2-1)/2 ∣ PSL ( 2 , p ) ∣ = p ( p 2 − 1 ) /2 for odd primes p p p , and compute it for p = 5 , 7 , 11 p=5,7,11 p = 5 , 7 , 11 .
Construct F 4 = F 2 [ t ] / ( t 2 + t + 1 ) \mathbb{F}_4=\mathbb{F}_2[t]/(t^2+t+1) F 4 = F 2 [ t ] / ( t 2 + t + 1 ) in digital coordinates as in § 20.2 , write its addition and multiplication tables on A 4 \mathcal{A}_4 A 4 , and show that t t t generates F 4 × \mathbb{F}_4^\times F 4 × .
Show that P S L ( 2 , 4 ) \mathrm{PSL}(2,4) PSL ( 2 , 4 ) and P S L ( 2 , 5 ) \mathrm{PSL}(2,5) PSL ( 2 , 5 ) both have order 60 and five conjugacy classes, and that both are isomorphic to A l t 5 \mathrm{Alt}_5 Alt 5 .
Show that ∣ S L 3 ( 2 ) ∣ = 168 = ∣ P S L ( 2 , 7 ) ∣ |\mathrm{SL}_3(2)|=168=|\mathrm{PSL}(2,7)| ∣ SL 3 ( 2 ) ∣ = 168 = ∣ PSL ( 2 , 7 ) ∣ .
Show that the Weyl group of S L n \mathrm{SL}_n SL n is S y m n \mathrm{Sym}_n Sym n , so that the Bruhat decomposition of S L n ( q ) \mathrm{SL}_n(q) SL n ( q ) has n ! n! n ! cells.
Use Table 12.1 to explain why the signed permutations occur for the symplectic and odd orthogonal groups, while only even signed permutations occur for the even orthogonal groups. State the corresponding Weyl-group orders.
Show that ∣ Sz ( 8 ) ∣ = 29120 \lvert\operatorname{Sz}(8)\rvert= 29120 ∣ Sz ( 8 )∣ = 29120 and ∣ G 2 ( 2 ) ∣ = 12096 \lvert G_{2}(2)\rvert= 12096 ∣ G 2 ( 2 )∣ = 12096 , using ∣ Sz ( q ) ∣ = q 2 ( q 2 + 1 ) ( q − 1 ) \lvert\operatorname{Sz}(q)\rvert= q^{2}(q^{2}+1)(q-1) ∣ Sz ( q )∣ = q 2 ( q 2 + 1 ) ( q − 1 ) and ∣ G 2 ( q ) ∣ = q 6 ( q 6 − 1 ) ( q 2 − 1 ) \lvert G_{2}(q)\rvert= q^{6}(q^{6}-1)(q^{2}-1) ∣ G 2 ( q )∣ = q 6 ( q 6 − 1 ) ( q 2 − 1 ) .
Sporadic groups
By the classification, there are 26 sporadic simple groups. They do not come in families with a parameter, so there is no uniform law like those of Chapter 12. Each of them has faithful transitive actions, however, and the chain law of Theorem 8.34 in § 14.12 gives a law in which a code is a mixed-radix word whose radices are orbit sizes. This chapter starts with the five Mathieu groups, whose small degrees make the chains short and explicit. It continues with the Conway groups, through the Leech lattice, and with the Fischer groups F i 22 \mathrm{Fi}_{22} Fi 22 , F i 23 \mathrm{Fi}_{23} Fi 23 and F i 24 ′ \mathrm{Fi}'_{24} Fi 24 ′ , through their 3-transpositions. It then treats six sporadic groups of small degree, from the library of primitive groups, and continues with J 3 J_3 J 3 , O ′ N \mathrm{O}'\mathrm{N} O ′ N , H N \mathrm{HN} HN , the Lyons group L y \mathrm{Ly} Ly , Thompson’s group T h \mathrm{Th} Th , through 3 D 4 ( 2 ) : 3 {}^{3}\mathrm{D}_4(2){:}3 3 D 4 ( 2 ) : 3 , and J 4 J_4 J 4 , through 2 11 : M 24 2^{11}{:}\mathrm{M}_{24} 2 11 : M 24 . For the last two, the Baby Monster B \mathbb{B} B and the Monster M \mathbb{M} M , the smallest permutation actions have about 1.4 ⋅ 10 10 1.4 \cdot10^{10} 1.4 ⋅ 1 0 10 and 10 20 10^{20} 1 0 20 points, so the coset digits are computed structurally, from the axes of the 196 884-dimensional representation, and not by orbit enumeration. Following the ATLAS [26], L n ( q ) L_n(q) L n ( q ) denotes PSL n ( q ) \operatorname{PSL}_n(q) PSL n ( q ) .
Dynkin diagrams at the sporadic boundary
Dynkin diagrams do not classify the sporadic groups: unlike the groups of Chapter 12, the 26 sporadic simple groups do not arise as a parameterised family of root data. Nevertheless, the diagrams remain useful in this chapter in three precise ways. They describe the Lie-type subgroups that occur as stabilisers, they organise the root lattices visible at the deep holes of the Leech lattice, and their Coxeter presentations reappear in the reflection-theoretic route toward the Monster and the bimonster.
The most concrete bridge is the Leech lattice Λ \Lambda Λ . It has no roots, but its deep holes carry affine Dynkin diagrams: the nearest lattice points to a deep hole form an extended ADE configuration of total rank 24. Up to affine isometry there are 23 types, corresponding to the 23 Niemeier lattices with roots. Consequently the Golay-code and Leech-lattice constructions below are not detached from Chapter 12: they place its simply laced diagrams at the boundary of the rootless lattice whose automorphism groups give C o 0 \mathrm{Co}_0 Co 0 , C o 1 \mathrm{Co}_1 Co 1 , C o 2 \mathrm{Co}_2 Co 2 and C o 3 \mathrm{Co}_3 Co 3 .
A second bridge is local structure. The chain laws in this chapter use transitive permutation actions and stabilisers, not Bruhat cells; however, many stabilisers are familiar Lie-type groups. For example, Table 13.6 uses L 2 ( 11 ) L_2(11) L 2 ( 11 ) for J 1 J_1 J 1 , U 3 ( 3 ) U_3(3) U 3 ( 3 ) for J 2 J_2 J 2 , G 2 ( 4 ) G_2(4) G 2 ( 4 ) for S u z \mathrm{Suz} Suz and 2 F 4 ( 2 ) ^{2}F_4(2) 2 F 4 ( 2 ) for R u \mathrm{Ru} Ru . Their Dynkin types identify the corresponding Weyl and root data, while the actual sporadic law remains the stabiliser-chain law of Theorem 8.34 .
Finally, Coxeter diagrams provide a useful signpost near the Monster without becoming a substitute for the verified law developed later in this chapter. The bimonster M ≀ C 2 \mathbb{M}\wr C_2 M ≀ C 2 is a quotient of the Coxeter group attached to the Y 555 Y_{555} Y 555 diagram after an additional relation. This explains why diagrammatic reflection ideas occur in the background of the Monster, whereas the present book deliberately evaluates the Baby Monster and Monster laws through their structural axis data and the mmgroup verification.
Thus Dynkin diagrams play three different roles in Chapter 13: affine diagrams label Leech-lattice hole geometry, finite diagrams label Lie-type local subgroups, and Coxeter diagrams supply historical and structural context for the Monster. None of these turns a sporadic group into a Lie-type family; the distinction is exactly why the mixed-radix chain laws of this chapter are needed.
The Mathieu groups
The Mathieu groups act on A n \mathcal{A}_n A n for n = 11 , 12 , 22 , 23 , 24 n = 11, 12, 22, 23, 24 n = 11 , 12 , 22 , 23 , 24 [26]. Their orders are
∣ M 11 ∣ = 7920 , ∣ M 12 ∣ = 95 040 , ∣ M 22 ∣ = 443 520 , \lvert M_{11}\rvert= 7920,\quad\lvert M_{12}\rvert= 95\,040,\quad\lvert M_{22}\rvert= 443\,520, ∣ M 11 ∣ = 7920 , ∣ M 12 ∣ = 95 040 , ∣ M 22 ∣ = 443 520 ,
∣ M 23 ∣ = 10 200 960 , ∣ M 24 ∣ = 244 823 040. \lvert M_{23}\rvert= 10\,200\,960,\quad\lvert M_{24}\rvert= 244\,823\,040. ∣ M 23 ∣ = 10 200 960 , ∣ M 24 ∣ = 244 823 040.
M 11 M_{11} M 11 is sharply 4-transitive, M 12 M_{12} M 12 is sharply 5-transitive, M 22 M_{22} M 22 is 3-transitive, M 23 M_{23} M 23 is 4-transitive and M 24 M_{24} M 24 is 5-transitive.
For the chain law of Theorem 8.34 , a product m ∗ n m * n m ∗ n costs one composition of permutations of A n \mathcal{A}_n A n and one pass through the chain, that is O ( k n ) O(kn) O ( k n ) operations.
Proposition 13.1 (Chains of the Mathieu groups). Let M 12 M_{12} M 12 and M 24 M_{24} M 24 be generated as follows. We use the underlying sets { 1 , … , 12 } \{1,\ldots,12\} { 1 , … , 12 } and { 1 , … , 24 } \{1,\ldots,24\} { 1 , … , 24 } , relabelled by i ↦ i − 1 i \mapsto i - 1 i ↦ i − 1 :
M 12 = ⟨ ( 1 2 ⋯ 11 ) , ( 3 7 11 8 ) ( 4 10 5 6 ) , ( 1 12 ) ( 2 11 ) ( 3 6 ) ( 4 8 ) ( 5 9 ) ( 7 10 ) ⟩ , M_{12} = \langle(1\ 2\ \cdots\ 11),\ (3\ 7\ 11\ 8)(4\ 10\ 5\ 6),\ (1\ 12)(2\ 11)(3\ 6)(4\ 8)(5\ 9)(7\ 10)\rangle, M 12 = ⟨( 1 2 ⋯ 11 ) , ( 3 7 11 8 ) ( 4 10 5 6 ) , ( 1 12 ) ( 2 11 ) ( 3 6 ) ( 4 8 ) ( 5 9 ) ( 7 10 )⟩ , M 24 = ⟨ ( 1 2 ⋯ 23 ) , ( 3 17 10 7 9 ) ( 4 13 14 19 5 ) ( 8 18 11 12 23 ) ( 15 20 22 21 16 ) , M_{24} = \langle(1\ 2\ \cdots\ 23),\ (3\ 17\ 10\ 7\ 9)(4\ 13\ 14\ 19\ 5)(8\ 18\ 11\ 12\ 23)(15\ 20\ 22\ 21\ 16), M 24 = ⟨( 1 2 ⋯ 23 ) , ( 3 17 10 7 9 ) ( 4 13 14 19 5 ) ( 8 18 11 12 23 ) ( 15 20 22 21 16 ) , ( 1 24 ) ( 2 23 ) ( 3 12 ) ( 4 16 ) ( 5 18 ) ( 6 10 ) ( 7 20 ) ( 8 14 ) ( 9 21 ) ( 11 17 ) ( 13 22 ) ( 15 19 ) ⟩ . \qquad(1\ 24)(2\ 23)(3\ 12)(4\ 16)(5\ 18)(6\ 10)(7\ 20)(8\ 14)(9\ 21)(11\ 17)(13\ 22)(15\ 19)\rangle. ( 1 24 ) ( 2 23 ) ( 3 12 ) ( 4 16 ) ( 5 18 ) ( 6 10 ) ( 7 20 ) ( 8 14 ) ( 9 21 ) ( 11 17 ) ( 13 22 ) ( 15 19 )⟩ . Let M 11 M_{11} M 11 be the stabiliser of 11 in M 12 M_{12} M 12 , M 23 M_{23} M 23 the stabiliser of 23 in M 24 M_{24} M 24 , and M 22 M_{22} M 22 the pointwise stabiliser of 22 and 23 in M 24 M_{24} M 24 , each acting on the remaining points. The Schreier–Sims algorithm gives the following bases and radices for Theorem 8.34 :
Table 13.1.
Proof. The products of the radices are the orders listed above, so the generated groups have the orders of the Mathieu groups. The radices also show the transitivity. For M 11 M_{11} M 11 and M 12 M_{12} M 12 every r l r_l r l equals n − l + 1 n-l+1 n − l + 1 and the chain ends there, so these actions are sharply 4- and 5-transitive. The bases, radices and orders were computed by the deterministic Schreier–Sims algorithm, as recorded in Remark 13.3.