Universal Kernel

Troisième partie · En montant la tourChapitre 10

La table en deux, en sept et à l’infini

The table at two, seven and infinity

Read from the draft of 3 October 2026

n = 2C442pairs ±v: 21 (D8)n = 3C356pairs ±v: 28 (S3)n = 4C4C484pairs ±v: 21 (D8), 21 (D8)n = 51168pairs ±v: 84 (C2)n = 6C31C3280pairs ±v: 28 (S3), 84 (C2), 28 (S3)n = 711336pairs ±v: 168 (1)n = 8C41C41C4462pairs ±v: 21 (D8), 84 (C2), 21 (D8), 84 (C2), 21 (D8)θ = 1 + 42q2 + 56q3 + 84q4 + 168q5 + 280q6 + 336q7 + 462q8 + ⋯
Plate 10.1The shells of Klein’s lattice, norms 2 to 8: each shell split into its G0G_0-orbits of vectors with their stabilizer classes, and beneath, the classes of the pairs ±v\pm v, all computed from Elkies’ basis.
  1. 10.1
  2. 10.2
  3. 10.3
  4. 10.4
  5. 10.5
  6. 10.6
  7. 10.7
  8. 10.8
  9. 10.9
  10. 10.10

Can the whole seam table be read from one lattice, and what does each place where the group of order 168 lives add to it or forget?

The seam table was built from finite theories. The completions of the last chapter give each of its objects more ways of being met, one at each place where the group of order 168 lives: at the prime 2 through the Frobenius of the field with eight elements, at the prime 7 through the points of the projective line over F49\F_{49} that are not defined over F7\F_7, and at the complex place through the cells of Thurston’s manifold.

Klein’s lattice reads the whole table. Every object of the group is a datum of it; every built seam of the first block is induced by its residues at 2, 7 and ∞\infty, except on the objects with stabilizers C2C_2 and 1, where the Galois involution stands in the way; and the Fano plane of Part One is its reduction at αˉ\bar\alpha.

Then the Frobenius at 2 turns out coherent across the theories that see it, the cells of the link complement give the table a seventh column, the octonion lattices of the Singer row are found among its cusps and tetrahedra, and the object of size 42 with cyclic stabilizer gains an incarnation carrying the Frobenius at 7, with two theories supplying seams on it whose agreement depends on the marking.

The central result · Which seams the lattice induces

Read the Fano plane as P(L∞/αˉL∞)\Proj(L_\infty/\bar\alpha L_\infty) and the sky through the arithmetic marking.

(1) Every incarnation of the first block of the seam table, and each of the five incarnations of the object of size 28, is a G0G_0-set of data of the lattice at one place: configurations in P(L∞/αˉL∞)\Proj(L_\infty/\bar\alpha L_\infty) and its incidence graph, the link of v0v_0, at 2; configurations on the sky, the link of Λ∞\Lambda_\infty, at 7; configurations in P(L∞⊗C)\Proj(L_\infty\otimes\C) with the invariant quartic at ∞\infty; data in G0G_0 itself; or data in the graph on the pairs of norm 3 with ∣h∣2=7|h|^2=7, which is the Coxeter graph.

(2) For the six rigid objects every seam is induced. (3) For C3C_3, C4C_4, C7C_7, A4aA_4^a, A4bA_4^b, V4aV_4^a and V4bV_4^b the natural automorphisms of the native data realize all of N(H)/HN(H)/H: −1-1 on the vectors of norm 3, the roots and the tetrahedra, Gal(L/E)\mathrm{Gal}(L/E) on the cyclotomic structures, and the exchange and rotation of the root pairs of a frame. So once one seam to an incarnation is induced, all are.

(4) For G/C2G/C_2 at most two of the four seams between two incarnations stable under an antilinear isometry are induced, and they differ by the involution of quotient class C4C_4. (5) For the regular object the natural automorphisms of such an incarnation form a group of order at most 6. The lattice reads its own regular data at every place, but no native datum was found that reads both as the frames of the Fano plane and as the ordered triples of points of P1(F7)\Proj^1(\F_7).

Proof

(1) by the identifications of the theorems below and of the last chapter; the incidence graph of the Fano plane is the link of v0v_0. (2) A rigid object has one seam between any two incarnations, and the residue maps named are seams. (3) −1-1 commutes with G0G_0 and with every antilinear isometry and moves a vector of norm 2 or 3 and a tetrahedron within its orbit; the frame operations are defined by orthogonality alone and generate S3S_3. (4) and (5) follow from the forms of the objects relative to the lattice’s arithmetic symmetry group, with the computed instances; the corollary “The two exceptional rows” below gives the reason.

Status

The fifteen objects in Klein’s lattice, the stabilizers of its shells and pairs, the kernels of the residue maps, the arithmetic marking and the readings of the sky at 7 are exact computations in EE, F2\F_2 and F7\F_7. The bound on natural automorphisms is proved; the forms of the fifteen objects relative to the arithmetic symmetry group of order 672, the naturality of the book’s named seams and the loops through the places are computed, and they supersede the explanation the theorem on induced seams first gave for its two exceptional rows: a residual symmetry there, not an obstruction, while the obstruction occurs at C3C_3 and C4C_4. The search for a native datum that reads both as the frames of the Fano plane and as the ordered triples of the line found none; that is the outcome of a search, not a proof that none exists. Two statements of Allcock and Kato are corrected: the vectors of norm 6 number 280, and the isometries have two orbits on them.

The coherence of the Frobenius at two is proved, its flexes and tangents computed exactly in Q(ζ)\Q(\zeta) and F8\F_8. The new objects of the Weil lattice, the signs at each place and the multiplier convention are computed; the two parents at 7 at the level of lattices are compared by Brauer characters and a Schur index argument. The column of MM and the eight lattices are proved with machine checks, and so are the two rules on the Coxeter graph, the agreement for μA\mu_A checked on all 42 edges. That the Frobenius at 7 acts on the imaginary points as the nontrivial element of N(T)/TN(T)/T for the non-split torus, and the octavian orders as the planes of PG(3,2)\mathrm{PG}(3,2) through Kirmse’s point, are reformulations of classical facts (Digne and Michel; Coxeter), and no novelty is claimed for them.

Les couches du réseau de KleinThe shells of Klein’s lattice

n = 2C442pairs ±v: 21 (D8)n = 3C356pairs ±v: 28 (S3)n = 4C4C484pairs ±v: 21 (D8), 21 (D8)n = 51168pairs ±v: 84 (C2)n = 6C31C3280pairs ±v: 28 (S3), 84 (C2), 28 (S3)n = 711336pairs ±v: 168 (1)n = 8C41C41C4462pairs ±v: 21 (D8), 84 (C2), 21 (D8), 84 (C2), 21 (D8)θ = 1 + 42q2 + 56q3 + 84q4 + 168q5 + 280q6 + 336q7 + 462q8 + ⋯
Plate 10.1The shells of Klein’s lattice, norms 2 to 8: each shell split into its G0G_0-orbits of vectors with their stabilizer classes, and beneath, the classes of the pairs ±v\pm v, all computed from Elkies’ basis.

Throughout, E=Q(−7)E=\Q(\sqrt{-7}), OE=Z[α]\mathcal O_E=\Z[\alpha] with α=(−1+−7)/2\alpha=(-1+\sqrt{-7})/2, L=Q(ζ)L=\Q(\zeta), L∞L_\infty is Klein’s lattice with Mumford’s form hh, and G0G_0 is its group of isometries of determinant one. A native datum is a G0G_0-orbit of data built from L∞L_\infty and hh: vectors and finite sets of vectors, sublattices, elements of G0G_0, or cells of Δ2×T7\Delta_2\times T_7 at the vertex (v0,Λ∞)(v_0,\Lambda_\infty). Its residue at 2 is the pair of its residues modulo (α)(\alpha) and (αˉ)(\bar\alpha); at 7, its residue modulo −7\sqrt{-7}; at ∞\infty, the point [v][v] of P(L∞⊗C)\Proj(L_\infty\otimes\C). A residue map is a description, its kernel at a datum is the stabilizer of the residue, and it forgets nothing exactly when it is a seam.

The shells come first. The vectors of norms 2 to 8 number 42, 56, 84, 168, 280, 336 and 462, as Elkies’ theta series 1+42q2+56q3+84q4+168q5+280q6+336q7+462q8+⋯1+42q^2+56q^3+84q^4+168q^5+280q^6+336q^7+462q^8+\cdots says. Their G0G_0-orbits have stabilizers C4C_4 (the roots, and α\alpha, αˉ\bar\alpha and 2 times roots), C3C_3 (the vectors of norm 3, and α\alpha and αˉ\bar\alpha times them) or trivial, and the pairs ±v\pm v have stabilizers D8D_8, S3S_3, C2C_2 or trivial. At norm 6 there are 280 vectors, the 112 imprimitive ones and one regular orbit of 168 primitive ones; a remark of Allcock and Kato counts 56+5656+56, so the group of linear and antilinear isometries has two orbits there, not one.

Proposition(Stabilizers of vectors and of their pairs) proved

(1) The stabilizer in G0G_0 of a nonzero vector of L∞⊗CL_\infty\otimes\C is trivial or of class C3C_3 or C4C_4; no vector has stabilizer of class C2C_2. (2) The stabilizer of a point of P(L∞⊗E)\Proj(L_\infty\otimes E), in particular of a pair ±v\pm v of lattice vectors, is trivial or of class C2C_2, S3S_3 or D8D_8. (3) All seven occur in L∞L_\infty: vectors of norm 5, 3, 2 have stabilizers 1, C3C_3, C4C_4, their pairs C2C_2, S3S_3, D8D_8, and pairs of norm 7 have trivial stabilizer.

Proof

(1) If the stabilizer KK of vv contains an involution tt, then vv spans the +1+1-eigenline of tt, its centre, whose stabilizer is the centralizer C(t)≅D8C(t)\cong D_8; it acts on the line by a character with kernel KK, so KK is D8D_8, C4C_4 or a Klein four-group. The character of L∞⊗CL_\infty\otimes\C contains the trivial one (3−5+2)/8=0(3-5+2)/8=0 times on D8D_8 and (3−3)/4=0(3-3)/4=0 times on a Klein four-group, so K≅C4K\cong C_4. If ∣K∣|K| is odd it lies in C3C_3, C7C_7 or 7:37{:}3, and an element of order 7 has eigenvalues ζ,ζ2,ζ4\zeta,\zeta^2,\zeta^4 or their conjugates and fixes no vector. (2) A finite group fixing [v][v], vv defined over EE, acts on EvEv through the roots of unity ±1\pm1 of EE, so the stabilizer of [v][v] contains that of vv with index at most 2: a centre gives D8D_8, the eigenline of an element of order 3 its normalizer S3S_3, and a trivial stabilizer at most C2C_2. (3) is computed.

Les quinze objets dans le réseauThe fifteen objects in the lattice

native datumsizeclassvectors of norm 51681pairs of norm 584C2vectors of norm 356C3roots42C4ordered root pairs of an a-frame42V4aordered root pairs of a b-frame42V4bpairs of norm 328S3cyclotomic structures24C7root pairs21D8a-tetrahedra14A4ab-tetrahedra14A4bMumford sublattices87:3a-frames7S4ab-frames7S4bthe lattice1G
Plate 10.2The fifteen objects as data of Klein’s lattice: each native datum with its orbit size and the class of its stabilizer, computed in G0G_0 and read through the arithmetic marking, which fixes the letters aa and bb.

Name the data as Allcock and Kato do, with the Fano plane read as P(L∞/αˉL∞)\Proj(L_\infty/\bar\alpha L_\infty). A frame is a triple of mutually orthogonal root pairs; an aa-frame is one whose six roots have one residue modulo αˉ\bar\alpha, a bb-frame one whose six roots have one residue modulo α\alpha. A tetrahedron is a set of four vectors of norm 3 with pairwise inner product −1-1, of class aa or bb in the same way. A cyclotomic structure is an element g∈G0g\in G_0 with g+g2+g4=αg+g^2+g^4=\alpha, that is, an OE\mathcal O_E-algebra embedding Z[ζ]→End⁡(L∞)\Z[\zeta]\to\operatorname{End}(L_\infty) with ζ↦g\zeta\mapsto g; and for a Sylow 7-subgroup P=⟨g⟩P=\langle g\rangle the Mumford sublattice is MP=L∞(1−g)M_P=L_\infty(1-g).

Each root pair is orthogonal to exactly four others, so the frames are fourteen, seven of each class. The tetrahedra are twenty-eight, each summing to 0, and if TT is one then −T-T is the other tetrahedron with the same residue, in the same orbit. The cyclotomic structures are exactly the 24 elements of trace α\alpha. The Mumford sublattice does not depend on the generator, and the eight of them are the eight neighbours of L∞L_\infty in T7T_7: each has Gram determinant 7, no roots, 14 vectors of norm 3 and 42 of norm 7, a copy of Mumford’s lattice with OL\mathcal O_L among them, and modulo −7\sqrt{-7} it is the plane orthogonal to the isotropic point its Sylow subgroup fixes.

At the vertex (v0,Λ∞)(v_0,\Lambda_\infty) the cells are data too: the fourteen neighbours at (α)(\alpha) have stabilizers S4aS_4^a and S4bS_4^b, seven of each; the 21 chambers D8D_8; the eight neighbours at 7 7:37{:}3; the 112 squares, a neighbour at 2 with one at 7, two orbits with stabilizer C3C_3; and the 168 prisms, a chamber at 2 with an edge at 7, one regular orbit.

Theorem(The fifteen objects in Klein’s lattice) computed

Each of the following is a single G0G_0-orbit whose stabilizers form the class named: vectors of norm 5 (1, size 168); pairs of norm 5 (C2C_2, 84); vectors of norm 3 (C3C_3, 56); roots (C4C_4, 42); ordered pairs of distinct root pairs of an aa-frame or of a bb-frame (V4aV_4^a, V4bV_4^b, 42 each); pairs of norm 3 (S3S_3, 28); cyclotomic structures (C7C_7, 24); root pairs (D8D_8, 21); aa- and bb-tetrahedra (A4aA_4^a, A4bA_4^b, 14 each); Mumford sublattices (7:37{:}3, 8); aa- and bb-frames (S4aS_4^a, S4bS_4^b, 7 each); and L∞L_\infty (GG). So every object of the group of order 168 is a native datum of L∞L_\infty, with the labels aa and bb of the seam table through the marking by P(L∞/αˉL∞)\Proj(L_\infty/\bar\alpha L_\infty).

Proof

By computation in exact arithmetic in EE, F2\F_2 and F7\F_7: all 179 subgroups of G0G_0 enumerated and each stabilizer identified among them, the vectors of norm at most 8 enumerated by the Fincke–Pohst method, and the antilinear isometries found as isometries from the conjugate lattice.

Ce que chaque place oublieWhat each place forgets

native datumown classat 2at 7at ∞vectors of norm 51S3C3C2pairs of norm 5C2S3S3C2vectors of norm 3C3S3C3S3rootsC4D8C4D8pairs of norm 3S3S3S3S3root pairsD8D8D8D8a-tetrahedraA4aS4aA4aS4aa-framesS4aS4aS4aS4a
Plate 10.3What each place forgets: for each native datum, the class of the stabilizer of its residue at 2, at 7 and at ∞\infty, computed; a gold entry is a faithful reading, the residue keeping the datum’s own class.

Each place forgets what its residue cannot see. At 2, where −1≡1-1\equiv1, the residue of a vector forgets its sign, so the roots, the vectors of norms 3 and 5 and the tetrahedra have kernels D8D_8, S3S_3, S3S_3 and S4S_4. At 7 every native datum is read faithfully except those of norm 5. At ∞\infty the projective residue forgets scalars: the pairs, the frames, the ordered frame pairs and the cyclotomic structures are read faithfully, the vectors and the tetrahedra are not. Reading a vector of norm nn instead as a map of degree nn from the Klein quartic to the elliptic curve with complex multiplication by OE\mathcal O_E, as Elkies does, is faithful. A Mumford sublattice has trivial residue at 2 and at ∞\infty, its index 7 being prime to 2.

No single place reads everything, but the places together read more. The vectors of norm 5 are read faithfully by 2 and 7 together. The ordered triangles of aa-frames, and the pairs of a root pair with an orthogonal pair of norm 3, are read faithfully at each place: the stabilizer of (±u,±w)(\pm u,\pm w) is D8∩S3D_8\cap S_3, of order at most 2, and it contains sus_u, which is −1-1 on u⊥∋wu^\perp\ni w.

Theorem(What each place forgets) computed

The kernels of the residue maps, as the class of the stabilizer of the residue at 2, at 7 and at ∞\infty: vectors of norm 5, S3S_3, C3C_3, C2C_2; pairs of norm 5, S3S_3, S3S_3, C2C_2; vectors of norm 3, S3S_3, C3C_3, S3S_3; roots, D8D_8, C4C_4, D8D_8; tetrahedra, S4S_4, A4A_4, S4S_4; Mumford sublattices, GG, 7:37{:}3, GG. Ordered frame pairs, pairs of norm 3, cyclotomic structures, root pairs, frames and L∞L_\infty are read faithfully at all three places. The letters aa and bb are preserved throughout.

Le marquage arithmétiqueThe arithmetic marking

0123456∞A(v) = {1, 3, 4, 6}, v a rootin the orbit of {0, 1, 2, 3}: 42 setsA(T) = {0, 1, 2, 5}, T an a-tetrahedronin the orbit of {0, 1, 2, 5}the ordered secant (2, ∞) of a vectorof norm 3: det(ã, b̃, w) a square
Plate 10.4The sky as Klein’s lattice reads it at 7: the eight isotropic points of L∞/−7L∞L_\infty/\sqrt{-7}L_\infty, labelled by the arithmetic marking, with a root’s set A(v)A(v), a four-set in the orbit of {0,1,2,3}\{0,1,2,3\}, and an aa-tetrahedron‘s set A(T)A(T), in the orbit of {0,1,2,5}\{0,1,2,5\}, computed from square classes.

The sky is the conic of isotropic points of the ternary quadratic space L∞/−7L∞L_\infty/\sqrt{-7}L_\infty, the link of Λ∞\Lambda_\infty in T7T_7. Of the bijections from it onto P1(F7)\Proj^1(\F_7) that carry G0G_0 onto the Möbius group, half carry the elements of trace α\alpha into the class of z↦z+1z\mapsto z+1; under these the lattice has the character of Klein’s representation, the marking of the quartic column. So the book’s labelling is the arithmetic one read at αˉ\bar\alpha: the Fano plane of Part One is the reduction of Klein’s lattice at the prime of EE that does not contain the character value of 7A7A. Two earlier statements agree with this independently of the lattice: the cyclic labellings for {0,4,6}\{0,4,6\} go onto 7A7A, and at the prime (2,ζ3+ζ+1)(2,\zeta^3+\zeta+1) of Q(ζ)\Q(\zeta), which lies over (α)(\alpha), the points of the Frobenius-fixed plane have class S4bS_4^b.

Read at 7, the column of P1(F7)\Proj^1(\F_7) is the lattice’s too. Write bb for h mod −7h\bmod\sqrt{-7} and SS for one of the two G0G_0-orbits of nonzero isotropic vectors, each of 24. The secant of a pair of norm 3, the two isotropic points orthogonal to its residue, is a bijection onto the 2-subsets. For a root vv, the set A(v)A(v) of isotropic points cc with b(c~,v)b(\tilde c,v) a nonzero square, c~∈S\tilde c\in S on cc, is a bijection onto the orbit of {0,1,2,3}\{0,1,2,3\}, and −v-v gives the complement. A vector ww of norm 3 orders its secant {a,b}\{a,b\} so that det⁡(a~,b~,w)\det(\tilde a,\tilde b,w) is a nonzero square, a bijection onto the 56 ordered pairs, −w-w giving the reversed pair. A tetrahedron TT goes to the set A(T)A(T) of the first points of the ordered secants of its four vectors, the aa-tetrahedra onto the orbit of {0,1,2,5}\{0,1,2,5\} and the bb-tetrahedra onto that of {0,1,2,4}\{0,1,2,4\}; and a Mumford sublattice goes to the isotropic point fixed by its Sylow subgroup. The square classes do not depend on the representatives, which change by squares.

Proposition(The arithmetic marking) computed

(1) There are 336 bijections from the sky onto P1(F7)\Proj^1(\F_7) that carry the action of G0G_0 onto the Möbius group PSL⁡(2,7)\PSL(2,7), a torsor for PGL⁡(2,7)\PGL(2,7); exactly 168 of them, one PSL⁡(2,7)\PSL(2,7)-orbit, carry the elements of trace α\alpha into the class of z↦z+1z\mapsto z+1.

(2) With these, the element of G0G_0 acting as z↦z+1z\mapsto z+1 has characteristic polynomial x3+x+1x^3+x+1 on L∞/αL∞L_\infty/\alpha L_\infty and x3+x2+1x^3+x^2+1 on L∞/αˉL∞L_\infty/\bar\alpha L_\infty, and the marking μA\mu_A of Part One sends z↦z+1z\mapsto z+1 to an element with x3+x2+1x^3+x^2+1.

(3) Hence μA\mu_A is, up to inner automorphisms, the marking of G0G_0 on P(L∞/αˉL∞)\Proj(L_\infty/\bar\alpha L_\infty), the marking on P(L∞/αL∞)\Proj(L_\infty/\alpha L_\infty) differs from it by the outer automorphism, and S4aS_4^a is the class of the stabilizers of the points of P(L∞/αˉL∞)\Proj(L_\infty/\bar\alpha L_\infty). (4) For a cyclotomic structure gg and a point p0p_0, p0gˉx↦xp_0\bar g^x\mapsto x is a cyclic labelling of P(L∞/αˉL∞)\Proj(L_\infty/\bar\alpha L_\infty) for {0,4,6}\{0,4,6\}, whose Singer collineation gˉ\bar g lies in 7A7A; over (α)(\alpha) the same labelling is for {0,1,3}\{0,1,3\}.

Proof

An isomorphism PSL⁡(2,7)→GL⁡(3,2)\PSL(2,7)\to\GL(3,2) is determined up to inner automorphisms by the characteristic polynomial of the image of z↦z+1z\mapsto z+1; inner automorphisms preserve it, and the outer one, inverse transpose, replaces x3+x+1x^3+x+1 by its reciprocal. The elements of trace α\alpha act on L∞/αˉL∞L_\infty/\bar\alpha L_\infty with trace α mod αˉ=1\alpha\bmod\bar\alpha=1, and g+g2+g4=αg+g^2+g^4=\alpha gives gˉ3=gˉ2+1\bar g^3=\bar g^2+1 modulo αˉ\bar\alpha, so pp, pgˉ2p\bar g^2, pgˉ3p\bar g^3 are collinear for every pp. The bijections were checked over all 8!8!.

Le Frobenius en deuxThe Frobenius at two

(1:0:0)(0:0:1)(0:1:0)τthe flexes of the klein quarticw0w1w2w3w4w5w6F8 = F2[w]/(w3 + w + 1)on its 24 coordinatizations, post-composition: τp ↦ (f ↦ f(p))124twisting element124equivariant twistΦ(y) = c−1y
Plate 10.5The Frobenius at 2, twice: the flex-tangent map τ\tau turning the coordinate flex triangle of the Klein quartic, and v↦v2v\mapsto v^2 on F8×\F_8^\times; a thread joins a flex to its coordinatization by F8\F_8, and the powers are 2 for the twisting element and 4 for its equivariant twist.

The prime 2 enters twice. The Klein quartic has good reduction there, its 24 points over F8\F_8 the reductions of its flexes, and F8\F_8 carries a Frobenius; the Fano plane is P(F8)\Proj(\F_8) over F2\F_2, and the octonion completion is built from the maps v↦wxv2v\mapsto w^xv^2. A symmetry that normalizes the group gives two objects: if σ∘μ(x)∘σ−1=μ(cxc−1)\sigma\circ\mu(x)\circ\sigma^{-1}=\mu(cxc^{-1}) on a marked set, then Φ=μ(c)−1∘σ\Phi=\mu(c)^{-1}\circ\sigma is an automorphism, the twisting element cc is unique when GG has trivial centre, and Φ(y)=c−1y\Phi(y)=c^{-1}y where σ\sigma fixes yy.

On the quartic, σ2 ⁣:ζ↦ζ2\sigma_2\colon\zeta\mapsto\zeta^2 satisfies σ2(ρ(x))=ρ(h−1xh)\sigma_2(\rho(x))=\rho(h^{-1}xh), so the twisting element c=h−1c=h^{-1} has power 2, and Φ=ρ(h)∘σ2\Phi=\rho(h)\circ\sigma_2 is the flex-tangent map τ\tau on the flexes, of power 4, and the identity on the bitangents, centres and flex triangles. The vectors fixed by the twisted Frobenius modulo a prime above 2 form a three-dimensional F2\F_2-space V0V_0, the two primes giving Fano planes whose points have classes S4bS_4^b and S4aS_4^a, and each point of the quartic over F8\F_8 coordinatizes the dual plane, carrying τ\tau to post-composition with the Frobenius of F8\F_8. In the octonion completion the Frobenius conjugates MxM_x to M2xM_{2x} and multiplication by ww conjugates it to Mx−1M_{x-1}; together they generate the normalizer of order 21 of the Singer group, and its Frobenius has power 2.

In the lattice’s own coordinates the Frobenius needs no twist. The eigenlines eλe_\lambda, λ∈{ζ,ζ2,ζ4}\lambda\in\{\zeta,\zeta^2,\zeta^4\}, of the 24 cyclotomic structures are pairwise orthogonal and are the 24 flexes; the tangent at eλe_\lambda is eλ∨eλ2e_\lambda\vee e_{\lambda^2}, and σ2\sigma_2 applied to coordinates carries eλe_\lambda to eλ2e_{\lambda^2}. So τ=σ2\tau=\sigma_2 on the flexes, and the polarity e↦e⊥e\mapsto e^\perp of hh is T∘τT\circ\tau, TT sending a flex to its tangent: the three seams from the flexes to the flex tangents, TT, T∘τT\circ\tau and T∘τ2T\circ\tau^2, form a torsor for ⟨τ⟩≅C3\langle\tau\rangle\cong C_3, and the lattice supplies the polarity. The automorphism group N(C7)/C7≅C3N(C_7)/C_7\cong C_3 of the object of size 24 is Gal(L/E)\mathrm{Gal}(L/E), the decomposition group of the prime P=(2,ζ3+ζ+1)\mathfrak P=(2,\zeta^3+\zeta+1) over (α)(\alpha), which is inert in LL, and at the totally ramified −7\sqrt{-7} the inertia group. Modulo P\mathfrak P the eigenlines become the 24 points of P2(F8)\Proj^2(\F_8) off the lines of P2(F2)\Proj^2(\F_2), and σ2\sigma_2 the Frobenius x↦x2x\mapsto x^2. In Klein’s coordinates, whose matrices are not defined over EE, the twisting element is the cost of that choice.

Theorem(The Frobenius at two is coherent) proved

Under every seam between incarnations of the object of size 24 in the Klein quartic and in the Fano plane coordinatized by F8\F_8, the vertex link of the octonion completion, the GG-equivariant Frobenius automorphisms correspond: both are τ\tau, of power 4. The twisting elements, h−1h^{-1} for the quartic and the Frobenius for the completion, both have power 2, and the seam from a point of the quartic over F8\F_8 to its coordinatization of the dual plane realizes the correspondence directly.

Proof

The pairs of twisting element and twist were computed on the quartic and in the completion, and the power of an automorphism does not depend on the seam. The relation between them, a twist of power 4 against a twisting element of power 2, is the general Φ(y)=c−1y\Phi(y)=c^{-1}y, and it takes the same form in both theories: the flex-tangent map τ\tau, a fact of the projective geometry of the quartic, is the Frobenius at 2 made equivariant.

Les sutures naturelles du réseauThe natural seams of the lattice

D8 = N(C2), the half-turn at its centreC40123456∞{0, ∞} {1, 6}{2, 3} {4, 5}the coxeter pairingV4a0123456∞{0, ∞} {4, 5}{1, 6} {2, 3}V4b0123456∞{0, ∞} {2, 3}{1, 6} {4, 5}
Plate 10.6The three involutions of the object of size 84: in N(C2)≅D8N(C_2)\cong D_8, the half-turn at its centre and its three subgroups of order 4; on the line, the three ways of re-pairing the four pairs that z↦−1/zz\mapsto-1/z swaps, with their quotient classes C4C_4, V4aV_4^a and V4bV_4^b. The Galois involution exchanges the last two, so only the first is natural: two of the four seams between incarnations of one form.

Which seams does the lattice prefer? The 336 antilinear isometries of L∞L_\infty, the bijections with c(λx)=λˉc(x)c(\lambda x)=\bar\lambda c(x) and h(cx,cy)=h(x,y)‾h(cx,cy)=\overline{h(x,y)}, form a coset of {±1}×G0\{\pm1\}\times G_0, and each induces the outer automorphism of G0G_0. Complex conjugation b0b_0 of LL preserves L∞=(1−ζ)−1Z[ζ]L_\infty=(1-\zeta)^{-1}\Z[\zeta] and is one of them. The arithmetic symmetry group of the lattice, its linear and antilinear isometries normalizing G0G_0, is A={±1}×A1A=\{\pm1\}\times A_1 with A1=G0⋊⟨b0⟩≅PGL⁡(2,7)A_1=G_0\rtimes\langle b_0\rangle\cong\PGL(2,7), of order 672. It acts on the sky through PGL⁡(2,7)\PGL(2,7) with kernel {±1}\{\pm1\}, b0b_0 going to an involution outside PSL⁡(2,7)\PSL(2,7); on the incidence graph of the Fano plane, the link at 2, by collineations and a polarity; and on Klein’s plane by ρ(G)\rho(G) and complex conjugation of Klein’s coordinates. A seam between two incarnations built from the lattice is natural when it commutes with every element of AA that preserves both. By the stabilizer principle applied to AA, an incarnation stable under a group A′A' between G0G_0 and AA has a form, the class of its A′A'-stabilizer among the complements of N(H)/HN(H)/H; natural seams exist exactly between incarnations of the same form, and then they form a torsor under the natural automorphisms, the part of N(H)/HN(H)/H fixed by the stabilizer. A form is projective when −1-1 acts trivially and a vector form otherwise. For Klein’s lattice: C3C_3 and C4C_4 have four forms each, two of them projective; C7C_7 has one, on which all of C3C_3 is natural; the rigid rows have one; V4aV_4^a, V4bV_4^b, A4aA_4^a, A4bA_4^b, relative to {±1}×G0\{\pm1\}\times G_0, have one projective form, on which all of N(H)/HN(H)/H is natural; the object of size 84 has two forms, on which only the involution of quotient class C4C_4 is natural; and the regular object has a projective form with natural automorphisms S3S_3 and a vector form with {±1}\{\pm1\}.

The book’s named seams sort accordingly. Natural for all of AA: the seam by the multiplier 2 from ordered pairs to 3A3A, the seam by the multiplier ι\iota from imaginary points to 4A4A, the tangent and residual-point seams between flexes and flex tangents, hence τ\tau and the polarity, the seam from centres with a bitangent to the arcs of the Coxeter graph, the residue maps at 7 of the pairs of norm 3, the root pairs and the Mumford sublattices, and every seam of a rigid object stable under AA, among them the ten among the five incarnations of the twenty-eight. Natural only for {±1}×G0\{\pm1\}\times G_0: the square-class seams on the line, the seam by ω\omega on the tangent, the point rule and the bracket rule on the Coxeter edges, the reduction of contact points at the prime over the Bianchi prime, and the involutions of quotient classes V4aV_4^a and V4bV_4^b of the object of size 84. Natural for a subgroup of index two not containing −1-1: the ordered secant of a vector of norm 3, natural for A1A_1, and the four-set A(v)A(v) of a root, natural for the stabilizer ⟨G0,−b0⟩\langle G_0,-b_0\rangle of the orbit SS it uses. Every seam of the last two kinds joins incarnations of different forms and is natural for the stabilizer of the convention it uses.

Loops through the places close up inside these groups. Read at 2 through the incidence graph, at ∞\infty through Klein’s plane and at 7 through the sky, the holonomies of the loops 2→∞→7→22\to\infty\to7\to2 of natural seams form exactly the natural automorphisms of the incarnation at 2 for C2C_2, C3C_3, C7C_7 and 1, of orders 2, 2, 3 and 6; for C4C_4 no natural seam joins the directed 8-cycles at 2 to the eigenvectors for ii at ∞\infty, and the cyclic forms close up through 7 and G0G_0 instead. The twisted natural self-seams of an incarnation form a group of order 336 ∣Aut⁡A∣336\,|\Aut_A| in which the untwisted ones have index 2: the Galois class, realized at 2 by the polarity, at ∞\infty by complex conjugation and at 7 by an odd Möbius map, as the type law says for a split, a complex and a ramified place. Not every natural seam has a construction. There are six natural seams from the pairs of norm 7 to the orbit of (1:2:5)(1{:}2{:}5), and no native datum with that orbit as residue was found; every point of Klein’s plane fixed by complex conjugation and by no element of ρ(G)\rho(G) has an orbit of the one projective form of the regular object, so these orbits form a continuum, any two joined by six natural seams. A statement that every natural seam is induced can therefore hold only for a notion of induced seam that contains the stabilizer principle for AA itself, and then it says nothing. The general theory of natural seams of an arithmetic source is the subject of the chantier on reciprocity.

Corollary(The two exceptional rows) computed

In the theorem on the seams the lattice induces, the objects G/C2G/C_2 and G/1G/1 are the rows on which AA acts nontrivially on N(H)/HN(H)/H, so that even between incarnations of one form only some seams are natural: two of four for G/C2G/C_2, six (projective) or two (vector) of 168 for G/1G/1. Their exception is a residual symmetry, a torsor that is never a point, not a cohomological obstruction: each has one projective form. The obstruction does occur, at the rows C3C_3 and C4C_4, which the theorem counts as induced because it allows auxiliary choices: points of contact and eigenvectors for ii have the dihedral form, ordered pairs, imaginary points and the classes 3A3A, 4A4A the cyclic one.

Proof

The forms and their natural automorphisms were enumerated as the complements of N(H)/HN(H)/H in NA(H)/HN_A(H)/H and their fixed groups, and the forms of the book’s incarnations by computing the AA-stabilizer of a point of each. For C2C_2, N(H)/H≅C2×C2N(H)/H\cong C_2\times C_2 and the Galois element exchanges the involutions of quotient classes V4aV_4^a and V4bV_4^b, so it is an induced module with trivial H1H^1; for H=1H=1 the complements of G0G_0 in A1A_1 are generated by the outer involutions, all conjugate, with centralizer S3S_3. The forms can also be seen: an antiholomorphic element fixing a point of Klein’s plane where an element tt of order 3 or 4 acts by ε\varepsilon carries it to a point where tt acts by εˉ\bar\varepsilon, so it inverts tt and the form is dihedral; the fixed points of a torus on the line are fixed by the whole torus of PGL⁡(2,7)\PGL(2,7), which centralizes tt, so those forms are cyclic. The earlier bound, by which the automorphisms commuting with one antilinear isometry form all of N(H)/HN(H)/H for C3C_3, C4C_4, C7C_7, only the involution of class C4C_4 for C2C_2, and a group of order 3, 4 or 6 for the regular object, is the instance A′=⟨G0,c⟩A'=\langle G_0,c\rangle.

Le revêtement double dans le réseau de WeilThe double cover in the Weil lattice

orbits of sl(2,7) on the shells, by sizenorm 12401 × 16 + 2 × 112no orbit of 48norm 221602 × 16 + 7 × 112 + 4 × 336no orbit of 48norm 367203 × 112 + 19 × 336no orbit of 4816: the odd lift of 7:3112: the odd lift of C3336: trivial
Plate 10.7The shells of norms 1, 2 and 3 of the lattice E8E_8 over the tetrahedra of class aa, split into their orbits under SL⁡(2,7)\SL(2,7), computed from its signed permutations of the eight vectors vcv_c: every orbit has 16, 112 or 336 elements, and none has 48.

The double cover has a lattice of its own. Let G~=SL⁡(2,7)\tilde G=\SL(2,7) and let E8aE_8^a be the hermitian lattice of the Weil quartet of the next chapter, on which G~\tilde G acts by the signed permutations of a cross of sixteen vectors ±vc\pm v_c, one pair over each point cc of P1(F7)\Proj^1(\F_7); its roots outside the cross lie over pairs of points, the two points fixed by the image of their stabilizer. Every G~\tilde G-orbit of nonzero vectors is a new object, since −I-I acts as −1-1. The odd lift of C7C_7 is never the stabilizer of a vector, so the object of size 48 is a forced gap of the vectors. Sets of vectors fill it: for a root rr over a pair not containing ∞\infty, the seven roots UrUr, UU the unipotent odd lift of C7C_7 fixing v∞v_\infty, have stabilizer exactly UU, and their orbit is the object of size 48.

So the four new objects are data of E8aE_8^a: the cross, the other two orbits of roots, the vectors of norm 2 or 3 with trivial stabilizer, and the seven-sets UrUr. With the fifteen objects of Klein’s lattice, L∞⊕E8aL_\infty\oplus E_8^a over Z[α]\Z[\alpha], an integral form of the Weil representation 3⊕43\oplus4, carries all nineteen objects of G~\tilde G.

The sign of −I-I is seen differently at each place. At 2, x−(−x)=2x=λλˉxx-(-x)=2x=\lambda\bar\lambda x lies in λE8a\lambda E_8^a and in λˉE8a\bar\lambda E_8^a, so −I-I lies in the kernel of every residue and no new object is read faithfully there. At 7, E8a/−7E8a≅F74E_8^a/\sqrt{-7}E_8^a\cong\F_7^4 has no invariant line, plane or hyperplane, and the residues of the cross lie on eight lines any four of which are independent: the module is Sym3V\mathrm{Sym}^3V, VV the natural module, and the cross is the twisted cubic over the eight points of P1(F7)\Proj^1(\F_7). There −I-I lies in no residue kernel, every residue of a new object is again a new object, and the roots, the vectors of norm 2 and the seven-sets UrUr are read faithfully. At ∞\infty the linear residue in the Weil quartet is faithful, and the projective one has kernel {±I}\{\pm I\} times the stabilizer.

Theorem(The new objects in the Weil lattice) computed

(1) On the vectors of E8aE_8^a of norms 1, 2 and 3, shells of 240, 2160 and 6720, the G~\tilde G-orbits are: at norm 1, one of 16, with stabilizer the odd lift of 7:37{:}3, and two of 112, with stabilizer the odd lift of C3C_3; at norm 2, two of 16, seven of 112 and four of 336; at norm 3, three of 112 and nineteen of 336. No vector has stabilizer the odd lift of C7C_7. (2) For a root rr over a pair not containing ∞\infty, the seven roots UrUr have stabilizer exactly UU, and their orbit is the object of size 48. (3) Hence the four new objects are data of E8aE_8^a, and L∞⊕E8aL_\infty\oplus E_8^a carries all nineteen objects of G~\tilde G.

Proof

The Weil representation and its cross were rebuilt from their formulas, the shells enumerated in doubled coordinates, and orbits and stabilizers computed in the 336 signed permutations. A stabilizer has odd order because −I-I moves every vector, and the vectors fixed by UU are the multiples of v∞v_\infty, so no vector has stabilizer UU. The stabilizer of UrUr contains UU and has odd order, so it is UU or the odd lift of the Borel subgroup; an element of order 3 of the latter moves the pair of rr to a pair that is not a translate of it.

La rencontre des parents, lue dans le corps composéWhere the parents meet, read in the compositum

0123456∞on L/√−7 L: eigenvalues 1, 2, 4the 4-eigenline: 3, multiplier 2the 2-eigenline: 1, multiplier 4the same for all 56 elements of order 3
Plate 10.8An element tt of order 3 of G0G_0 on the sky, computed: its eigenvalues on L∞/−7L∞L_\infty/\sqrt{-7}L_\infty are 1, 2 and 4, its 2- and 4-eigenlines are two points of the sky, and its Möbius map has multiplier 2 at the 4-eigenline and 4 at the 2-eigenline, for all 56 such elements.

Let F=Q(−7,−3)F=\Q(\sqrt{-7},\sqrt{-3}), ω=e2πi/3\omega=e^{2\pi i/3}, and p=(2+ζ6)=(3+ω)\mathfrak p=(2+\zeta_6)=(3+\omega) the prime of Q(−3)\Q(\sqrt{-3}) of Thurston’s manifold, at which ω≡4\omega\equiv4; let Q\mathfrak Q be the prime of FF over p\mathfrak p and over −7\sqrt{-7}, with residue field F7\F_7. An element tt of order 3 of G0G_0 fixes the pole of its bitangent and two contact points, and the contact point at which tt acts on the tangent by ω\omega is the ω\omega-eigenline of tt, defined over FF. The theorem says that the book’s three conventions, ω\omega on the tangent at the contact point, multiplier 2 at the first point of an ordered pair, and the prime p\mathfrak p for Thurston’s manifold, are one choice.

At 7 the two parents’ sources share more than the sky. Over Q(−7)\Q(\sqrt{-7}) Klein’s lattice reduces to Sym2V\mathrm{Sym}^2V, VV the natural module of G~\tilde G over F7\F_7, and E8aE_8^a to Sym3V\mathrm{Sym}^3V; the faithful irreducible 8, quaternionic with rational character, is realizable over Q(−7)\Q(\sqrt{-7}), its quaternion algebra being ramified exactly at 3 and ∞\infty, and a stable lattice for it reduces with factors VV and Sym5V\mathrm{Sym}^5V. Over Q(−3)\Q(\sqrt{-3}) the monomial lattice of the principal series reduces at p\mathfrak p with factors VV and Sym5V\mathrm{Sym}^5V, and the Bianchi parent’s lattice O2\mathcal O^2 reduces to VV, the second layer of its tree being sl2(F7)≅Sym2V\mathfrak{sl}_2(\F_7)\cong\mathrm{Sym}^2V. So the depths are exchanged: Klein’s depth 0 is the Bianchi tree’s second layer. They differ in their completions and at the archimedean place: compact U(3)\mathrm U(3) against PSL⁡(2,C)\PSL(2,\C), complex multiplication by OE\mathcal O_E against j=0j=0 on the cusp tori.

Two columns of the table become residues. The column of Thurston’s manifold, read as GG-sets, consists of residues at 7: the cusps are the Mumford sublattices, the edges the secants of the pairs of norm 3, the faces the ordered secants of the vectors of norm 3, the tetrahedra the sets A(T)A(T) of the 28 tetrahedra of the lattice, aa onto aa, and the complementary pairs of tetrahedra the frames. The octonion column is a residue at 2: the 30 Fano structures on the points of P(L∞/αˉL∞)\Proj(L_\infty/\bar\alpha L_\infty) fall into orbits of 1, 7, 14 and 8, sharing 7, 3, 1 and 0 lines with the plane, with classes GG, S4aS_4^a, A4bA_4^b, 7:37{:}3: Kirmse’s lattice, the octavian orders, and the lattices sharing one line or none. What the residues do not carry is the arithmetic of the Bianchi parent: the hyperbolic structure of MM, the cusp tori C/p\C/\mathfrak p with complex multiplication by Z[ω]\Z[\omega], and the tree of PSL⁡(2,O[1/p])\PSL(2,\mathcal O[1/\mathfrak p]).

Theorem(The multiplier convention is the Bianchi prime) computed

For each of the 56 elements tt of order 3 of G0G_0: (1) tt acts on L∞/−7L∞L_\infty/\sqrt{-7}L_\infty with eigenvalues 1,2,4, its 2- and 4-eigenlines are points of the sky, and its Möbius map has multiplier 2 at the 4-eigenline and 4 at the 2-eigenline; (2) the ω\omega-eigenline of tt over FF reduces modulo Q\mathfrak Q to the 4-eigenline, and modulo the prime over pˉ\bar{\mathfrak p} to the 2-eigenline.

Hence reducing the contact points at the prime of FF over the Bianchi prime p\mathfrak p carries the convention ω\omega on the tangent at the contact point to the convention multiplier 2 at the first point, and reducing at the prime over pˉ\bar{\mathfrak p} carries it to the opposite convention. The choices of ω\omega, of the multiplier 2 and of p\mathfrak p are one choice: an identification of the cube roots of unity in C\C with those in F7\F_7.

Proof

L∞/−7L∞≅Sym2VL_\infty/\sqrt{-7}L_\infty\cong\mathrm{Sym}^2V, and for s=diag(μ,μ−1)s=\mathrm{diag}(\mu,\mu^{-1}) the conic point of e12e_1^2 has eigenvalue μ2\mu^2 and multiplier μ−2\mu^{-2}; so the multiplier at an isotropic eigenline is the inverse of its eigenvalue, and the inverse of 4 is 2. The multiplier of tt on its bitangent at the ω\omega-eigenline is ω2/ω=ω\omega^2/\omega=\omega, and multipliers at fixed points do not depend on the identification of the sky with P1(F7)\Proj^1(\F_7). The eigenlines are integral over OE[ω]\mathcal O_E[\omega] and nonzero modulo Q\mathfrak Q, and their reductions were computed.

La colonne des cellules et les huit réseauxThe column of the cells and the eight lattices

0123456∞the tetrahedron {0, 4, 6, ∞} of class bis an affine plane: its points sum to 0affine planes among the 14 tetrahedra: 14origin ∞; p + q the fourth cuspof the tetrahedron through ∞, p, q
Plate 10.9The eight cusps as an affine space over F2\F_2 whose planes are the tetrahedra of class bb, computed from the Steiner system: origin ∞\infty, and p+qp+q the fourth cusp of the tetrahedron through ∞\infty, pp and qq. Through the seam from the eight octonion lattices that share no line with the table, these fourteen planes are the lattices that share one.

Every cell of MM is determined by its cusps, so every figure of cells and incidences is a figure of the projective line, and the cells give the table a seventh column: faces with one of their edges (1); tetrahedra with one of their edges (C2C_2); faces, edges with one end, tetrahedra with a face and tetrahedra with a cusp (C3C_3); four cusps spanning no tetrahedron (C4C_4); tetrahedra of class A4aA_4^a or A4bA_4^b with a pair of opposite edges (V4aV_4^a, V4bV_4^b); edges (S3S_3); cusps with one of the three parallel classes of edges of their cusp torus (C7C_7); bisections into two fours spanning no tetrahedron (D8D_8); the two classes of tetrahedra; cusps (7:37{:}3); complementary pairs (S4aS_4^a, S4bS_4^b); and MM. The parallel classes at ∞\infty are the edges {a,a+d}\{a,a+d\} with d=±1d=\pm1, ±2\pm2, ±3\pm3, the images of the three directions 1, ζ\zeta, ζ2\zeta^2 of the triangular lattice, and the stabilizer of ∞\infty permutes them cyclically.

Each class of tetrahedra is a Steiner system S(3,4,8)S(3,4,8): three cusps lie in exactly one tetrahedron of each class, and the cusps outside a tetrahedron are those of a tetrahedron of the same class. So each class makes the eight cusps an affine space of dimension 3 over F2\F_2, whose planes are its tetrahedra, and GG lies in the affine group GL⁡(3,2)⋉F23\GL(3,2)\ltimes\F_2^3 of the cusps in two ways, exchanged by the outer automorphism. The stabilizer of a point of MM is trivial or of class C2C_2, C3C_3, S3S_3, A4aA_4^a or A4bA_4^b, each occurring: the open cells partition MM, a tetrahedron’s stabilizer acts on it as its rotation group, a face’s rotates it, and an edge’s S3S_3 fixes the edge pointwise by its rotations and its midpoint by its involutions.

The octonions meet MM in the row of the Sylow normalizers. The eight lattices LCL_C whose Fano plane shares no line with the plane Π\Pi of the table have the normalizers of the Sylow 7-subgroups as stabilizers. For a Fano plane BB on the units, a multiplication with table BB is the table carried by a permutation of the units taking Π\Pi to BB, well defined up to sign changes, which preserve every LCL_C. Kirmse’s lattice turns out to be an order after all, for seven multiplications on the same units, though not for its own; and the eight lattices without a common line are the eight cusps of MM, the fourteen with one common line the fourteen tetrahedra of class A4bA_4^b, with incidence preserved. The class A4aA_4^a does not occur among the thirty lattices.

Proposition(The eight lattices) proved

(a) For each Sylow 7-subgroup PP of GG, exactly two of the thirty Fano planes on the units are PP-invariant: Π\Pi, and a plane ΠP\Pi_P sharing no line with it. Labelling the units by Z/7\Z/7 so that PP acts by translation and Π\Pi has the lines x+{0,1,3}x+\{0,1,3\}, ΠP\Pi_P has the lines x+{0,4,6}x+\{0,4,6\}, the mirror image of Π\Pi, and P↦LΠPP\mapsto L_{\Pi_P} is the seam from the Sylow 7-subgroups onto the eight lattices.

(b) The thirty lattices are the fifteen points and fifteen planes of PG(3,2)\mathrm{PG}(3,2), two lattices sharing as many lines as the corresponding elements: Kirmse’s lattice is a point P0P_0, the seven octavian orders the planes through P0P_0, the fourteen lattices sharing one line with Π\Pi the other points, and the eight lattices the planes not through P0P_0. (c) LAL_A is closed under the multiplication with table BB exactly when AA and BB share three lines, that is, when they are incident in PG(3,2)\mathrm{PG}(3,2). (d) For each of the fourteen lattices LQL_Q sharing one line with Π\Pi, exactly four of the eight lattices share three lines with it; these fourteen four-sets are the planes of an affine space of dimension 3 over F2\F_2 on the eight lattices, and the seam from the eight lattices to the cusps of MM carries them onto the fourteen tetrahedra of class A4bA_4^b.

Proof

(a) A PP-invariant plane consists of the seven translates of a line with six distinct differences, and the three-element perfect difference sets of Z/7\Z/7 are the translates of {0,1,3}\{0,1,3\} and of {0,4,6}\{0,4,6\}. The normalizer of PP fixes Π\Pi, so it fixes ΠP\Pi_P; its class 7:37{:}3 is that of the eight lattices, and the object of size 8 is rigid. (b) A doubly even code of length 8 and dimension 4 has fourteen words of weight 4, two of them meeting in 0 or 2 coordinates, so any three coordinates lie in exactly one; the isomorphism A8≅GL⁡(4,2)A_8\cong\GL(4,2) identifies the lines of PG(3,2)\mathrm{PG}(3,2) with the bisections and its points and planes with the two orbits of A8A_8 on the Steiner systems. (c) Relabel by a permutation carrying Π\Pi to BB; all 900 pairs were also checked. (d) In PG(3,2)\mathrm{PG}(3,2) the planes not through P0P_0 that pass through a point Q≠P0Q\neq P_0 are the four not containing the line P0QP_0Q, and in the dual space these four-sets are the planes of the affine space off the plane dual to P0P_0. The stabilizer of the four-set of LQL_Q is that of LQL_Q, of class A4bA_4^b, so its image is a tetrahedron of class A4bA_4^b.

L’objet de taille 42 à stabilisateur cycliqueThe object of size 42 with cyclic stabilizer

F49 = F7[ι], ι2 = −1the real row: P1(F7) without ∞z, multiplier −ιz7, multiplier ι42 imaginary points, one orbit of 42, stabilizer C4the Frobenius z ↦ z7: the reflection in the real row0123456∞0 → 1 → ∞ → 6: [a,c][c,b][b,a] a squareadvanced by z ↦ (−z − 1)/(z − 1), of order 4the reverse cycle: a non-square
Plate 10.10The object of size 42 twice. Left, F49=F7[ι]\F_{49}=\F_7[\iota] as a grid, the real row and the 42 imaginary points off it, and an element of order 4 fixing zz and its Frobenius image z7z^7, with multipliers ι\iota and −ι-\iota. Right, the harmonic pair {0,∞},{1,6}\{0,\infty\},\{1,6\} on the line and the directed 4-cycle 0→1→∞→60\to1\to\infty\to6 whose bracket product is a square.

The object G/C4G/C_4 has automorphism group D8/C4≅C2D_8/C_4\cong C_2 and many incarnations: the class 4A4A, the directed 4-cycles on quadrangles, the harmonic pairs of disjoint pairs and the four-sets in the orbit of {0,1,2,3}\{0,1,2,3\} on the line, the eigenvectors for ii in Klein’s plane, the edges of the Coxeter graph, the sets of four cusps of MM spanning no tetrahedron. The projective line supplies one more, whose automorphism is a Frobenius. Let F49=F7[ι]\F_{49}=\F_7[\iota] with ι2=−1\iota^2=-1, possible because −1-1 is not a square modulo 7; GG acts on P1(F49)\Proj^1(\F_{49}) by the same Möbius maps, and the 42 points outside P1(F7)\Proj^1(\F_7) are the imaginary points. They form one orbit with stabilizer class C4C_4; each element of order 4 fixes exactly two of them, zz and zˉ=z7\bar z=z^7, with multipliers ι\iota and −ι-\iota; and the Frobenius z↦z7z\mapsto z^7, which commutes with Möbius maps over F7\F_7 and moves every imaginary point, is the nontrivial automorphism of this incarnation. The diagonal matrices of SL⁡(2,7)\SL(2,7) map onto a group of class C3C_3, and the groups of order 8 generated by lifts of elements of order 4 onto the groups C4C_4: these are the split and the non-split tori. For the split torus the object is the ordered pairs of points, its two fixed points, and the automorphism exchanges them. For the non-split torus it is the imaginary points, and the automorphism, which again exchanges the two fixed points of the torus, is the Frobenius at 7.

The Coxeter graph, which lies between the Fano plane and the line, lets two theories supply seams on this object. In its antiflag model an edge is {(p,B),(q,B′)}\{(p,B),(q,B')\} with B∩B′B\cap B' the third point cc of the line pqpq. The point rule goes from each point of BB off pqpq to the third point of its line with pp, and from each point of B′B' off pqpq to the third point of its line with qq, tracing a directed 4-cycle on the quadrangle complementary to pqpq; the line rule does the same with the lines through pp and qq, on the four lines missing cc. The vertex seam sends the edge to a harmonic pair of disjoint pairs {{a,b},{c,d}}\{\{a,b\},\{c,d\}\}, and of its two directed 4-cycles a→c→b→da\to c\to b\to d and a→d→b→ca\to d\to b\to c exactly one has [a,c][c,b][b,a][a,c][c,b][b,a] a nonzero square, whatever the starting point and the coordinate vectors; the bracket rule takes the element of order 4 advancing it.

So the square classes of F7\F_7, which GG preserves and PGL⁡(2,7)\PGL(2,7) does not, tell the points of the Fano plane from its lines through the edges of the Coxeter graph: the seam system formed by the Coxeter edges, the harmonic pairs of pairs and 4A4A, with the vertex seam, the bracket rule and the point rule, is coherent when the marking is in the class of μA\mu_A, and its monodromy is the nontrivial automorphism otherwise; equivalently, a marking carries the stabilizers of the bisection {0,1,2,5}∣{3,4,6,∞}\{0,1,2,5\}\mid\{3,4,6,\infty\} to the stabilizers of points exactly when the bracket rule agrees with the point rule. The modular curve did the same for the inner class through its holomorphic structure. Neither statement is about a single theory: each singles out one class of markings by asking two theories to agree.

Theorem(The Coxeter edges and the marking) proved

Let the Coxeter graph be in its antiflag model, with GG acting through a marking μ\mu. (a) The point rule and the line rule, each followed by the element of order 4 that advances its cycle by one step, are seams from the edges to 4A4A, and the two rules give mutually inverse elements. (b) The bracket rule is a well-defined seam from the edges to 4A4A. (c) If μ\mu differs from μA\mu_A by an inner automorphism of GG, the bracket rule agrees with the point rule on every edge; if it differs by an outer automorphism, the bracket rule agrees with the line rule.

Proof

(a) The rules use only incidence and treat the two antiflags of an edge alike, so they are GG-maps, and each traces a 4-cycle. (b) In [a,c][c,b][b,a][a,c][c,b][b,a] each point occurs twice, so its square class does not depend on the vectors and is SL⁡(2,7)\SL(2,7)-invariant. Take a=0a=0 and b=∞b=\infty; harmonicity gives d=−cd=-c, and with c=uc=u the products along a→c→b→da\to c\to b\to d read from each start are uu, 2u2u, uu, 2u32u^3, all in the class of uu, while the reverse cycle gives −u-u, and −1-1 is not a square. (c) For μA\mu_A the agreement was checked on all 42 edges. An inner change of marking commutes with every construction; an outer one, by a Möbius map of non-square determinant, multiplies every bracket by a non-square and so reverses the bracket rule, while the point rule only relabels.

Klein’s lattice carries the first block of the table: every object a native datum, every rigid seam induced, and every seam of all but two rows, the two exceptions exactly where the Galois involution acts. The double cover’s objects need the Weil lattice beside it.

The next chapter takes up the double cover itself: its four new objects, their second theory in the Weil representation, the lattice E8E_8 that representation carries, and the octonion table’s mirror.

Introduced here
twisting element