Can the order of a group be read as a product of local factors, one for each place, the way Siegel’s mass formula reads the order of the symmetry group of E8?
Siegel’s mass formula is a classical product formula for lattices. Two lattices lie in one genus when they are isometric over every completion; the mass of the genus is the sum of 1/∣AutL∣ over its classes; and the formula of Smith, Minkowski and Siegel writes the mass as a product of local factors, one for each place. When a genus has a single class, the left side is 1/∣AutL∣, and the order of a symmetry group comes out of local data. For the even unimodular lattices of rank 8 the only class is E8, and the order of its group, 696729600, comes out as 240×24×240×504, one factor at a time from Bernoulli numbers.
The program reads this as a seam statement: each local factor is the order of a finite group, the group of the lattice’s residue at that place, its life there, so a group’s size would be the product of its lives. For 168 it asked whether Klein’s lattice is alone in its genus, so that its 336 symmetries would come out of data attached to −7. It is not. The genus of unimodular hermitian lattices of rank 3 over Z[α] has two classes, Klein’s lattice and the standard lattice Z[α]3, and its mass is 421=481+3361. The second class is made of Klein’s own frames, and its group is a stabilizer inside Klein’s.
In the strict sense, one global group the product of its lives, the formula holds exactly when the genus has one class, and the draft realizes it on the genera beside Klein’s: Mumford’s lattice Z[ζ7], whose 42 symmetries are the inverse of its mass; the Witting lattice, where the two lives of PSp(4,3)≅PSU(4,2) are two local factors of one mass; E8 over Z[α]; and the icosians, with the two lives of A5. The theorems are classical; reading them as statements about lives is the new part.
The central result · A product formula
Local lives multiply to the whole group. The mass of a genus of definite hermitian lattices is a product over all places, and each local factor is the normalized order of a finite group of Lie type, the group of the lattice’s residue at that place: its life there. In the strict sense, that one global group is the product of its lives, this holds exactly when the genus has one class.
For 168 the source is not alone: Klein’s genus has two classes, 421=481+3361, and the second is built from Klein’s frames. The strict form holds for Mumford’s genus beside it, 421=1/∣AutZ[ζ7]∣, and for the Witting lattice, E8 over Z[α] and the icosians.
Status
A reading of classical theorems, the mass formulas of Smith, Minkowski and Siegel, of Hashimoto and Koseki for hermitian lattices, and of Prasad for arithmetic groups, with every case below computed in exact arithmetic or proved: Klein’s genus has exactly two classes, by the mass formula and Kneser’s neighbours and again from the Z-lattices of its trace forms; Mumford’s genus has one; the Witting lattice is alone in its genus; E8 over Z[α] and the icosians have one class each. The program’s hope for 168, that Klein’s lattice would be alone and its order a product of lives, is answered in the negative.
Still open: the genera of the other lattices of the type law, Valentiner’s (2160 symmetries) and that of PSL(2,8) (1008); whether the mass can tell apart the two lives that a ramified prime offers, which at 7 have the same order; and a column of genus-mates for the seam table, such as the standard lattice beside Klein’s. The draft also records, as a reading and not a computation, Cartwright and Steger’s two classes of fake projective planes over Q(−7) as the two kinds of vertex of the tree at 7.
La masse d’un genreThe mass of a genus
Plate 19.1The even unimodular lattices of rank 8: one class, E8, whose mass is the inverse of the order of its group, 696729600.
Let L be a positive definite lattice, over Z or over the integers of an imaginary quadratic field. Its genus is the set of lattices isometric to L over the completion at every prime; it holds finitely many isometry classes, and each has a finite group. The draft reads the mass formula as a seam statement: each local factor is the order of a finite group of Lie type, the group of the lattice’s residue at that place, its life there.
Siegel’s formula computes the mass without listing the classes. For the even unimodular lattices of rank 8 the genus has one class, E8, so its mass is 1/∣W(E8)∣=1/696729600, the inverse of the order of the group of E8.
Definition(Genus and mass)
Two positive definite hermitian lattices over OE,E imaginary quadratic, lie in the same genus if they become isometric over OE⊗Zp for every prime p. The mass of the genus of L is ∑i1/∣Aut(Li)∣, summed over representatives Li of its isometry classes. The same definitions apply to positive definite Z-lattices, with Aut=O.
Facteur par facteurFactor by factor
Plate 19.2Siegel’s mass for E8 one factor at a time: from B4,B2,B4 and B6 the factors 2401,241,2401 and 5041, whose product is 1/∣W(E8)∣.
With B2=61,B4=−301 and B6=421, the four factors for rank 8 are ∣B4∣/8=2401,∣B2∣/4=241,∣B4∣/8=2401 and ∣B6∣/12=5041, and their product is the mass:
∣W(E8)∣=240×24×240×504=696729600,
the order recorded in Chapter 8 for the Weyl group. In product form the Euler factor at a prime p is the normalized order ∣SO8+(Fp)∣/p28 of the special orthogonal group of the split quadratic space E8/pE8: the life of the group at p. At p=2 the global group fills it: W(E8) maps onto O8+(2), of order 348364800, with kernel {±1}. At every odd prime it is a proper subgroup of its life.
Example(The lattice E8)
E8 is the only even unimodular lattice of rank 8, so the mass of its genus is 1/∣W(E8)∣, and the Bernoulli form of the mass formula reads 6967296001=8∣B4∣⋅4∣B2∣⋅8∣B4∣⋅12∣B6∣=240⋅24⋅240⋅5041. In the form of Conway and Sloane it is the standard mass 28/696729600 times the type factor 2−8 at 2; in product form its Euler factor at p,(1−p−2)(1−p−4)2(1−p−6), is ∣SO8+(Fp)∣/p28.
Les facteurs sont des viesThe factors are lives
Plate 19.3Two lives of the group of order 168, two factors of the mass of Klein’s genus: the Heawood graph, the link of a vertex of the building over Q2, and P1(F7), the link of a vertex of the tree over Q7.
For hermitian lattices over E=Q(−d) the mass formula of Hashimoto and Koseki is a product over the primes in which each factor is the normalized order of the isometry group of the residue. At a split prime p=ppˉ the residues L/pL and L/pˉL are dual spaces paired by h, and the group is GLm(Fp); for Klein’s lattice at 2 they are the points and the lines of the Fano plane. At an inert prime the residue is a hermitian space over Fp2. At an odd ramified prime it is the quadratic space L/−dL; for Klein’s lattice at 7 it is the ternary space whose conic is the sky. At the archimedean place the factor is the volume of U(m).
So the factors are the lives of the lattice’s group, and the group itself sits at one point of the product of its local geometries: the stabilizer of the point given by L∞ in the building at 2 and the tree at 7 is Aut(L∞) modulo its centre, acting on the two links by the two lives of 168. Klein’s group fills its life at 2: it maps onto GL3(F2) with kernel {±1}. At 7 it has index 2 in O3(F7), the missing coset being the non-square Möbius maps of the outer automorphism, and at the inert prime 3 index 72 in U3(F3).
Theorem(The mass as a product of lives) proved
Let E=Q(−d) have discriminant dE and quadratic character χ, let t be the number of primes dividing dE, and let L lie in the genus of OEm with the form ∑ixiyˉi. Then
mass(L)=2∣dE∣m(m+1)/4j=1∏m(2π)j(j−1)!p∏αp−1,
where αp=∣GLm(Fp)∣/pm2 if p splits in E,αp=∣Um(Fp)∣/pm2 if p is inert, and αp=∣Om(Fp)∣/pm(m−1)/2 if p is odd and ramified and m is odd. For m odd, mass(L)=21−t∏j=1m∣Bj,χj∣/2j, with χj=χ for odd j and the trivial character for even j. (Hashimoto and Koseki; Gan and Yu prove the product form for every genus of unimodular lattices.)
Deux classesTwo classes
Plate 19.4Klein’s genus: Klein’s lattice and the standard lattice, its two classes, with masses 3361 and 481 making up 421; Klein’s fourteen neighbours at (α) are the points and lines of its Fano plane, each the standard lattice spanned by αˉ−1 times a frame.
Over E=Q(−7) all positive definite unimodular hermitian lattices of rank 3 lie in one genus. Its mass, from the Bernoulli form with B1,χ=−1,B2=61 and B3,χ=748, is 21⋅241⋅78=421. Klein’s lattice contributes 3361, so it is not alone. Kneser’s neighbours at the split prime (α), started at the standard lattice, find exactly two classes, the standard lattice with 48 symmetries and Klein’s, and 481+3361 is already the whole mass.
The second class is not a new group. Klein’s fourteen neighbours at (α), one for each point and each line of its Fano plane P(L∞/αL∞), are all standard: αˉ times the six unit vectors of such a neighbour is a frame of L∞, three mutually orthogonal root pairs, and its group is the stabilizer of that frame in Klein’s group, {±1}×S4, the objects S4a and S4b. The same mass comes out of Z-lattices: the trace forms of the two classes are B⊕3 and the Barnes lattice P6, in a genus of three classes whose OE-structures number 8,2 and 0, and 3848+6722=421.
Theorem(Klein’s genus has two classes) computed
The genus of OE3 has exactly two classes, the standard lattice and Klein’s lattice L∞, and mass=421=481+3361, with ∣Aut(OE3)∣=48 and ∣Aut(L∞)∣=336. The fourteen Kneser neighbours of L∞ at (α), seven for the points and seven for the lines of the Fano plane P(L∞/αL∞), are standard; for each, αˉ times its six vectors of norm 1 is a frame of L∞, a bijection onto the fourteen frames; and its group is the stabilizer of the frame in Aut(L∞), of order 48, the stabilizer of a point or of a line of the Fano plane.
Proof
The mass is the Bernoulli form with m=3 and t=1. Kneser’s method of neighbours at the split prime (α), started at OE3 and closed when no new class appears, gives the two classes; since 481+3361 is already the whole mass, there is no other. The neighbours, frames and stabilizers were computed from the 336 automorphisms, the 42 roots and the 14 frames of L∞. The class number agrees with the list of single-class genera of Jürgens and Zimmermann, which has no genus of rank 3 over Q(−7), and with Hoffmann’s result, recorded by Elkies, that L∞ is the only indecomposable unimodular lattice of rank 3 over OE.
Les voisins à septThe neighbours at seven
Plate 19.5The quotient of the tree at 7: the standard lattice, Z[ζ7] and Klein’s lattice, with their groups modulo ±1 and their eight neighbours each, coloured by class; counting edges from both ends gives the mass relation, and Z[ζ7]‘s neighbours, 7+1, divide Klein’s mass.
Change the lattice at 7 instead. Mumford’s lattice Z[ζ7], with the form tr(xyˉ), is of type 2 at 7 and unimodular elsewhere, and its genus has exactly one class: its mass is 421=1/∣AutZ[ζ7]∣, with 42=∣7:3∣⋅∣μE∣. Here the product formula holds in the strict sense. Its eight self-dual neighbours in the tree at 7 are one Klein lattice and seven standard ones, and the eight points of the sky, the link of its vertex, divide Klein’s mass accordingly: 421=3361+7, the orbits of the Frobenius group 7:3 on P1(F7).
The two genera, one vertex type of the tree each, have the same mass, 428=488+3368 by counting the edges between them, and in the product formula this is the equality of their factors at 7: the reductive quotients at the two kinds of vertex are O3 and O1×Sp2 over F7, and ∣O3(F7)∣=2∣PGL(2,7)∣=672=2∣SL(2,7)∣. The mass cannot tell the two lives of the sky apart. Modulo ±1 and with determinant one this is the relation 241+1681=211 that the neighbours at 7 found first, Prasad’s formula for Q(−7). Mumford’s group, whose vertex stabilizer is that 211, has Euler characteristic χ(Γ1)=211=2mass(Z[ζ7]), the factor at 2, the Fano building, being 1−27+7+321=1.
Theorem(Mumford’s genus has one class) computed
The genus of Mumford’s lattice OL=Z[ζ7], of type 2 at 7 and unimodular elsewhere, has exactly one class, and mass(OL)=421=1/∣Aut(OL)∣, ∣Aut(OL)∣=42=∣7:3∣⋅∣μE∣. The eight self-dual lattices between OL and its dual are one Klein lattice and seven standard ones, permuted by Aut(OL) in orbits of sizes 1 and 7;OE3 and L∞ each have exactly eight sublattices of type 2 and colength one, all isometric to OL; and the masses of the two genera are equal, 428=488+3368.
Proof
Let M lie in the genus of OL. An isometry from OL⊗Z7 onto M⊗Z7 carries a self-dual lattice between OL⊗Z7 and its dual to one between M⊗Z7 and its dual. The lattice U that is the latter at 7 and M elsewhere lies in the genus of OE3, so it is standard or Klein’s, and M is a sublattice of U of type 2 and colength one, hence isometric to OL. Every vertex of T7 has eight neighbours, and counting the edges between the two genera gives the identity of masses. The neighbour data were computed.
Deux vies, deux facteursTwo lives, two factors
Plate 19.6The genera the draft has computed, with their masses, gold where one class makes the mass the inverse of one group, and the lives each group fills; below, the two lattices of the type law whose genera are not yet computed.
Where a group has a double life, the formula shows both lives as factors of one mass. The Witting lattice W over Z[ω], the lattice of PSp(4,3)≅PSU(4,2) in the type law, is alone in its genus, so its 155520 symmetries are the inverse of its mass; reduction maps Aut(W) onto the unitary group GU4(2) of W/2W with kernel {±1}, and onto the symplectic group Sp4(3) of W/−3W with kernel μ3, so 155520=2∣GU4(2)∣=3∣Sp4(3)∣. The icosians do the same for A5: one ideal class, mass 601=1/∣A5∣, and 2I fills SL2(F4) at 2 and SL2(F5) at 5, the two lives of A5≅PSL(2,4)≅PSL(2,5).
In the four cases computed the global group fills its life, up to scalars and the outer automorphism, at 2 for E8, at 2 and 7 for Klein’s lattice, at 2 and −3 for the Witting lattice, and at 2 and 5 for the icosians; for the last three these are exactly the places of the double lives of PSL(2,7),PSp(4,3) and A5. It cannot happen at large primes, where the order of the life grows like a power of p.
Theorem(The Witting lattice) computed
Let W=E8ρ be the z∈Z[ω]4 whose reduction modulo θ=−3 lies in the tetracode, with h(x,y)=∑ixiyˉi. Then W∨=θ−1W,31Trh is E8, and ∣Aut(W)∣=155520. The genus of W has mass 1555201, so W is alone in its genus and the product of local factors returns 1/∣μ6⋅Sp(4,3)∣. Reduction maps Aut(W) onto GU4(2) with kernel {±1} and onto Sp4(3) with kernel μ3. So the two lives of PSp(4,3)≅PSU(4,2) are two local factors of one mass, and the global group fills both.
Quinze croixFifteen crosses
Plate 19.7E8 over Z[(1+−7)/2] modulo λ: the fifteen crosses as the points of PG(3,2), the cross of the sixteen vectors at the centre and the seven lines through it, labelled by their points in the Fano plane.
The ring of Klein’s lattice also carries E8. The complex structure of the Weil representation of SL(2,7) is integral on the lattice E8a, a free module of rank 4 over Z[λ],λ=(1+−7)/2, with a basis of roots. Since 2=λλˉ, the 240 roots fall into fifteen crosses of sixteen, one in each nonzero residue modulo λ. The Z[λ]-linear isometries form 2⋅A7, of order 5040, two-transitive on the crosses as A7 is on the points of PG(3,2), and the stabilizer of the cross of the sixteen vectors ±vc is SL(2,7):5040=15×336. Modulo λ the seven lines through that cross have stabilizers S4a, and the quotient is the Fano plane of points.
The same construction gives a lattice alone in its genus. The unimodular lattice OE4 has sixteen sublattices M between −7OE4 and itself with a totally isotropic plane as quotient; on each, 71Trh is E8 and ∣Aut(M)∣=5040, and since each has 400 self-dual overlattices, mass(M)=10085⋅40016=50401. The genus has one class, with 5040=∣A7∣⋅∣μE∣, the order of the group of the hermitian E8 above; an isometry between the two lattices was not computed.
Theorem(E8 as a Hermitian lattice) computed
λ=(1+θ)/2 preserves E8a and E8b; each is a free Z[λ]-module of rank 4 with a basis of roots, and TrK/Qh=2⟨,⟩. The roots in each nonzero residue of E8/λE8≅F24 form a cross, and likewise modulo λˉ. The Z[λ]-linear isometries of E8a form a group U≅2⋅A7 of order 5040 with centre {±1}, two-transitive on the fifteen crosses modulo λ and transitive on the 240 roots, and the stabilizer of the cross of the sixteen vectors is SL(2,7).
Proof
No root lies in λE8, since ∣λy∣2=2∣y∣2≥2. Two roots in one residue differ by an element of λE8, so their inner product is an integer and they are equal, opposite or orthogonal; a residue holds at most sixteen roots, and 240 roots in fifteen residues fill each. The isometries were found from the images of a basis of roots; the image of U on the crosses has order 2520 and index 8 in GL(4,2)≅A8, hence is A7, and the extension does not split because −I is the only involution of the stabilizer SL(2,7).
Ce qui est cherchéWhat is sought
Plate 19.8What remains: the lattices of Valentiner and of PSL(2,8), lit, whose genera are not yet computed, below the genera the draft has.
Three things are sought. The other lattices of the type law: the genera of Valentiner’s lattice, 2160 symmetries over Q(−3,5), and of the lattice of PSL(2,8),1008 symmetries over the real cubic subfield of Q(ζ9), are not yet computed. The ramified places: at 7 the self-dual and the type-2 vertices have local factors of the same order, so the mass cannot tell the two lives of the sky apart, and whether this blindness holds at every ramified prime in odd rank is open. And the classes beside the source: for 168 the second class of Klein’s genus is made of Klein’s frames, and a column of such genus-mates for the seam table is proposed.
One more seam between genera is in the draft. The rank-4 unimodular genus over Z[α] and the genus of Z-lattices of the Barnes lattice P6 both have mass 10085 and three classes, with 384,672 and 1152 symmetries, and L↦(Λ2L)⋆ carries one onto the other, sending OE⊕L∞ to P6: the exceptional isomorphism SU4≃Spin6 read on lattices, a seam between two genera.
Open questionopen
Do the genera of Valentiner’s lattice and of the lattice of PSL(2,8) have one class, so that their groups are products of their lives? At every ramified prime in odd rank, are the local factors of the two kinds of vertex equal, so that the mass cannot tell the two lives there apart, as at 7?
What would settle the chantier is the genera of the remaining lattices of the type law computed, with their local factors identified as lives, and the blindness at ramified primes decided. For 168 the answer to the program’s question is in: Klein’s lattice is not alone, and the classes that share its mass are built from its frames, the objects S4a and S4b.
The product formula reads a group’s size from its lives at all places, beside the reciprocity law of Chapter 16, which reads its natural seams from the same source. The epilogue turns from one group to all curves at once.