Skip to content

Commit 6517eff

Browse files
committed
Second-round review corrections
- Drop hK : FunctionField Fq K from exists_companion and the family corollary: it follows from the finite-type hypothesis plus the RatFunc Fq embedding, so it was a redundant hypothesis. Fq is passed by name where the family corollary invokes exists_companion, since no explicit argument determines it any longer. - AUDIT: restore both inductive conditions on the Kummer construction (the condensed version pinned only the later pi_j and did not fix v_i(b_m)); retract the companion-matrix witness as overstated (degree-one places only, lambda-unit roots only at lambda_0) while keeping the degeneracy finding; E-rationality is definitionally necessary for the compatibility notion here; n = 0 and one-prime items reclassified as reformulations; Ramakrishna scoped to a density-one set of primes; finite-level Frobenius API recorded; primary-source verification status updated. - Frobenius convention noted where VII.6 was said to apply directly: Mathlib's Frobenius is arithmetic, Lafforgue's polynomials use geometric. - Unit-ness scoped away from p with slope bounds at p; the (sigma_lambda)^r descent restated as the n-th power in this file's notation; lisse scope added to the integrality remark; 'constant field' corrected to the chosen finite base field; 'need not be either' fix in the Setup comment. - Design debt recorded: IsUnramifiedAt generalisation over the target ring, FrobeniusChoice restriction to the complement of S.
1 parent 0acf855 commit 6517eff

3 files changed

Lines changed: 94 additions & 55 deletions

File tree

AUDIT.md

Lines changed: 57 additions & 30 deletions
Original file line numberDiff line numberDiff line change
@@ -50,7 +50,12 @@ infinite, and every result taking one is then vacuously true.
5050
`IsArithFrobAt.exists_of_isInvariant` fails at the infinite level on three
5151
counts, not one: a finite residue field at the chosen prime, `[Finite G]`, and
5252
`[Algebra.IsInvariant R S G]`. Existence of Frobenius elements in the absolute
53-
Galois group therefore has to be assumed as data.
53+
Galois group therefore has to be assumed as data. The same file carries a fuller
54+
finite-level theory: a canonical `arithFrobAt`, conjugacy of the canonical
55+
Frobenii at primes over one base prime (`isConj_arithFrobAt`, via
56+
`exists_primesOver_isConj`), and uniqueness under unramifiedness
57+
(`AlgHom.IsArithFrobAt.eq_of_isUnramifiedAt`), all under the same finiteness and
58+
invariance hypotheses, so none of it transfers either.
5459

5560
`isPretransitive_of_isGaloisGroup` requires `[Finite G]` and
5661
`[IsGaloisGroup G A B]`, so transitivity of the action on primes above a place is
@@ -70,8 +75,10 @@ eigenvalues outside `E`. Absolute irreducibility, finite-order determinant and
7075
`E`-rationality of the Frobenius characteristic polynomials were all added. The
7176
first two are Lafforgue's own hypotheses; the third is an artefact of fixing `E`
7277
in advance here, where Lafforgue obtains the number field as part of the theorem.
73-
None of the three is necessary for a compatible system to exist, since reducible
74-
ones do.
78+
Neither of the first two is necessary for a compatible family to exist, since
79+
reducible compatible systems do; the third cannot be dropped even in principle,
80+
because the notion of compatibility here builds `E`-rationality of the
81+
polynomials into the definition.
7582
Once those hypotheses were added the statement was still stronger than the
7683
theorem it named, in two independent ways. The conclusion placed companions over
7784
`E_λ` itself, where Theorem VII.6(v) gives the companion over a finite extension
@@ -87,31 +94,41 @@ was never actually assumed, since declaring a structure with an existence field
8794
does not assume it, so the compatibility condition could hold vacuously; it is
8895
now passed as data. Separately, the base ring was an arbitrary Dedekind domain
8996
with the right fraction field, and fields and discrete valuation rings qualify,
90-
so the place set could be empty or a single point. The conclusion was then
91-
satisfiable cheaply, since the roots of the prescribed polynomial are `λ`-adic
92-
units, `GL_n(O_λ)` is profinite and `` is procyclic, so a companion-matrix
93-
representation of the constant-field quotient satisfies everything. What made
94-
that sufficient is that the conclusion of the version in question imposed no
95-
irreducibility on the members of the family. It would not satisfy the present
96-
conclusion: a representation factoring through a procyclic group has image
97-
generated by a single matrix `C`, so its span lies in `M[C]` and has dimension at
98-
most `n`, short of the `` that `SpanFull` demands once `n > 1`. Requiring the
99-
base ring to be a finite-type algebra over the constant field makes its spectrum
97+
so the place set could be empty or a single point. In the empty case every condition in the
98+
definition quantifies over places that do not exist, and the conclusion was
99+
trivially satisfiable; a single place leaves the compatibility conditions
100+
constraining one Frobenius conjugacy class instead of infinitely many. An
101+
earlier version of this file claimed an explicit cheap witness in the one-place
102+
case, a companion-matrix representation of the constant-field quotient. That was
103+
overstated: the construction matches the prescribed polynomial only at a place
104+
of degree one, since a place of degree `d` sends the generator to the `d`-th
105+
power of the chosen matrix, and it needs the roots of the polynomial to be units
106+
at the target `λ`, which continuity supplies only at `λ₀`. Nor would any witness
107+
of that shape survive the present conclusion: a representation factoring through
108+
a procyclic group has image generated by a single matrix `C`, so its span lies
109+
in `M[C]` and has dimension at most `n`, short of the `` that `SpanFull`
110+
demands once `n > 1`. Requiring the
111+
base ring to be a finite-type algebra over the chosen finite base field makes its spectrum
100112
a smooth affine curve carrying cofinitely many of the places, and closes this.
101113

102114
Three claims were described as doing more than they do.
103115
`charpoly_eq_of_isConj` was said to establish well-definedness of the
104116
compatibility condition, which needs inertia rather than conjugacy, and a
105117
docstring continued to assert that claim, and to cite the deleted axiom, after
106-
the code had been corrected. Unit-ness of the Frobenius roots was attributed to
107-
purity, when it is part (iii) of Deligne's Conjecture 1.2.10, labelled (c) in
118+
the code had been corrected. Unit-ness of the Frobenius roots away from `p` was attributed to purity, when it is part (iii) of Deligne's Conjecture 1.2.10, labelled (c) in
108119
Drinfeld's abbreviated restatement, proved for curves by Lafforgue, and a conclusion separate from the weight condition; `(3 + 4i)/5` has
109120
absolute value `1` at every archimedean place and is not a `5`-adic unit. The
110121
equivalence between the span condition and absolute irreducibility was stated
111122
without its scope, which is a field and `n ≥ 1`.
112123

113-
Two smaller defects: `n : ℕ` admitted `n = 0`, and unramifiedness quantified over
114-
one prime above a place where every prime is wanted.
124+
Two smaller items were reformulations rather than corrections of wrong
125+
mathematics. Positivity of `n` became a hypothesis of the main theorem; the
126+
definitions still compile at `n = 0`, where `GL (Fin 0)` is trivial and the
127+
predicates hold vacuously, harmless where they are used. And unramifiedness,
128+
which had quantified over one prime above a place, now quantifies over every
129+
prime; the two readings are equivalent once a prime above `v` exists, since the
130+
primes are conjugate and `ker ρ` is normal, but the every-prime form does not
131+
lean on the transitivity that is unavailable here.
115132

116133
One error was not about the mathematics at all. An action of `G_K` on the
117134
integral closure of `A` in ``, and the `SMulCommClass` accompanying it, were
@@ -133,13 +150,17 @@ It is tempting to drop it. Continuity forces the image into a compact subgroup o
133150
of the maximal ideal factors through a finite extension ramified at finitely many
134151
places. That gives a finite ramification set at every finite level, but the
135152
level-wise sets need not stabilise, and for `n ≥ 2` they need not. For
136-
`ℓ ≠ char K`, a Kummer class built from `b_m = ∏_{i ≤ m} π_i^{ℓ^i}`, with places
137-
`v_i` and elements `π_i` chosen so that `v_i(π_i) = 1` and each later `π_j` is a
138-
unit at the earlier `v_i`, gives a continuous upper-triangular
139-
`ρ = (χ_ℓ, c; 0, 1) : G_K → GL₂(ℤ_ℓ)` ramified at every `v_i`. Ramakrishna,
153+
`ℓ ≠ char K`, a Kummer class built from `b_m = ∏_{i ≤ m} π_i^{ℓ^i}`, with the places `v_i` and elements `π_i` chosen
154+
inductively so that `v_i(π_i) = 1` and `v_i(π_j) = 0` for every `j ≠ i`, gives a
155+
continuous upper-triangular `ρ = (χ_ℓ, c; 0, 1) : G_K → GL₂(ℤ_ℓ)` ramified at
156+
every `v_i`; a block sum with the trivial representation extends this to every
157+
`n ≥ 2`. Controlling the earlier `π_j` at `v_i` as well as the later ones is
158+
what makes `v_i(b_m) = ℓ^i` exactly, which is nonzero mod `ℓ^m` for every
159+
`m > i`. Ramakrishna,
140160
*Infinitely ramified Galois representations*, Ann. of Math. 151 (2000), 793–815,
141-
constructs over `` surjective `GL₂(ℤ_ℓ)`-valued representations ramified at
142-
infinitely many primes, so even full image is compatible with infinite
161+
constructs over ``, for every prime `ℓ ≥ 5` in a set of density one,
162+
surjective `GL₂(ℤ_ℓ)`-valued representations ramified at infinitely many primes,
163+
so even full image is compatible with infinite
143164
ramification;
144165
Khare–Rajan, Int. Math. Res. Not. 2001, no. 12, 601–607, show that for continuous
145166
semisimple representations of the absolute Galois group of a number field the
@@ -173,13 +194,19 @@ Three notions defined here have close Mathlib counterparts that are not used.
173194
`LambdaAdicRep` is a `ContinuousMonoidHom`; the recurring conjunction
174195
`Q.IsPrime ∧ Q.under A = v.asIdeal` restates `Ideal.LiesOver`, the idiom
175196
`RingTheory/Frobenius.lean` itself uses; and `Field.absoluteGaloisGroup` is the
176-
fixed-closure form of `AbsGal`. Each is equivalent to what is written, and each
177-
would be the better choice in a version aimed at upstreaming.
178-
179-
Lafforgue's Theorem VII.6 and Deligne's Conjecture 1.2.10 were verified through
180-
Drinfeld's verbatim quotations and bibliography rather than the originals. The
181-
internal part-numbering of the conjecture in Weil II was not checked against
182-
Deligne's own text.
197+
fixed-closure form of `AbsGal`. Each is equivalent to what is written, and each would be the better choice in a version aimed at upstreaming. Two further
198+
items: `IsUnramifiedAt` is hard-wired to representations over `E_λ`, so the same
199+
inertia condition is written a second time, inline, in the conclusion of
200+
`exists_companion`, where a version stated for an arbitrary `G →* GL n R` would
201+
let hypothesis, conclusion and the family corollary share one predicate; and
202+
`FrobeniusChoice` supplies a lift at every place where only the places outside
203+
`S` are used.
204+
205+
The literature attributions, including Deligne's part-numbering in Weil II and
206+
the hypotheses and conclusions of Lafforgue's Theorem VII.6, have been checked
207+
against the primary sources. Two asides rest on standard usage rather than the
208+
original papers: the attribution of the matrix-span criterion to Burnside, and
209+
the remark about the Fontaine–Mazur conjecture.
183210

184211
Every mathematical claim in this repository that is not a compiled Lean theorem
185212
is a claim, not a fact. The compiled theorems are listed in the README.

LambdaAdicSlice/Companions.lean

Lines changed: 21 additions & 16 deletions
Original file line numberDiff line numberDiff line change
@@ -25,9 +25,10 @@ section Setup
2525
-- places carried by the chosen model: for `A = 𝓞 K` they are the finite places of
2626
-- a number field, and for a finite-type Dedekind model of a function field they
2727
-- are the closed points of the corresponding affine curve, hence all but finitely
28-
-- many places of the proper curve. For an arbitrary Dedekind subring they are
29-
-- neither, since Mathlib's `IsDedekindDomain` admits localisations, discrete
30-
-- valuation rings and fields. `exists_companion` imposes the finite-type
28+
-- many places of the proper curve. For an arbitrary Dedekind domain with
29+
-- fraction field `K` they need not be either, since Mathlib's `IsDedekindDomain`
30+
-- admits localisations, discrete valuation rings and fields.
31+
-- `exists_companion` imposes the finite-type
3132
-- hypothesis that pins this down.
3233
variable (A K : Type*) [CommRing A] [IsDedekindDomain A] [Field K]
3334
[Algebra A K] [IsFractionRing A K]
@@ -365,16 +366,19 @@ place `λ` at a time.
365366
366367
This is a slice of VII.6, not the whole of it. VII.6 begins from an irreducible
367368
lisse sheaf of finite-order determinant and constructs `E` and the polynomials
368-
`P_v`; here both are supplied as data. VII.6 also proves purity, integrality of
369-
the Frobenius roots, and a further descent statement, none of which is
370-
formalised.
369+
`P_v`; here both are supplied as data. VII.6 also proves purity, unit-ness of the Frobenius
370+
roots away from `p` with slope bounds at the places over `p`, and a further
371+
statement descending the `n`-th power of the companion to `E_λ` itself, none of
372+
which is formalised.
371373
372374
`A` is a finite-type Dedekind `𝔽_q`-algebra with fraction field the function
373375
field `K`, so `Spec A` is a smooth affine curve over `𝔽_q` and its closed points
374-
are all but finitely many places of `K`. Without the finite-type hypothesis `A`
375-
could be a discrete valuation ring or `K` itself, `HeightOneSpectrum A` could be
376-
a single point or empty, and the statement would be satisfiable by an
377-
everywhere-unramified companion-matrix construction.
376+
are all but finitely many places of `K`. Without the finite-type hypothesis `A` could be a
377+
discrete valuation ring or `K` itself and `HeightOneSpectrum A` a single point
378+
or empty, collapsing the compatibility conditions. Finiteness of `K` over
379+
`Fq(t)` is not assumed: the finite-type hypothesis and the compatible embedding
380+
of `RatFunc Fq` already force it, which is why no `FunctionField` hypothesis
381+
appears.
378382
379383
Hypotheses on `ρ₀`: absolute irreducibility (`SpanFull`), finite-order
380384
determinant, unramified at every `v ∉ S`, and Frobenius characteristic
@@ -387,8 +391,11 @@ varies).
387391
Conclusion: a companion over a finite extension of `E_λ`, unramified outside `S`,
388392
with the same `P v`, itself absolutely irreducible with finite-order determinant.
389393
390-
The curve here is the smooth affine `U = Spec A` with `S` removed, to which
391-
VII.6 applies directly. The places of a proper model lying outside `Spec A` are
394+
The curve here is the smooth affine `U = Spec A` with `S` removed, to which VII.6
395+
applies after one change of convention: the Frobenius here is arithmetic,
396+
`x ↦ x ^ #(A/v)`, where Lafforgue writes the characteristic polynomials for
397+
geometric Frobenius, its inverse, and the polynomials of an invertible matrix
398+
and of its inverse determine each other. The places of a proper model lying outside `Spec A` are
392399
not points of `U`, so nothing is claimed about them; that is a statement of scope,
393400
not a shortfall. What is deliberately not formalised is any descent of the
394401
coefficient field to `E_λ`.
@@ -398,7 +405,6 @@ theorem exists_companion
398405
{Fq : Type*} [Field Fq] [Finite Fq]
399406
[Algebra Fq A] [Algebra Fq K] [IsScalarTower Fq A K] [Algebra.FiniteType Fq A]
400407
[Algebra (RatFunc Fq) K] [IsScalarTower Fq (RatFunc Fq) K]
401-
(hK : FunctionField Fq K)
402408
(hn : 0 < n)
403409
(Frob : FrobeniusChoice A K Kbar)
404410
(S : Finset (HeightOneSpectrum A))
@@ -449,7 +455,6 @@ theorem exists_companion_family
449455
{Fq : Type*} [Field Fq] [Finite Fq]
450456
[Algebra Fq A] [Algebra Fq K] [IsScalarTower Fq A K] [Algebra.FiniteType Fq A]
451457
[Algebra (RatFunc Fq) K] [IsScalarTower Fq (RatFunc Fq) K]
452-
(hK : FunctionField Fq K)
453458
(hn : 0 < n)
454459
(Frob : FrobeniusChoice A K Kbar)
455460
(S : Finset (HeightOneSpectrum A))
@@ -478,8 +483,8 @@ theorem exists_companion_family
478483
(algebraMap E (Completion E lam)))) ∧
479484
SpanFull (F lam h).toHom ∧ DetFiniteOrderHom (F lam h).toHom := by
480485
choose F hF using fun (lam : CoeffPlace E) (h : resChar E lam ≠ ringChar K) =>
481-
exists_companion A K Kbar E n hK hn Frob S P hchar₀ rho₀ hspan hdet hunram
482-
hrat lam h
486+
exists_companion (Fq := Fq) A K Kbar E n hn Frob S P hchar₀ rho₀ hspan
487+
hdet hunram hrat lam h
483488
exact ⟨F, hF⟩
484489

485490
end CompanionFamily

README.md

Lines changed: 16 additions & 9 deletions
Original file line numberDiff line numberDiff line change
@@ -25,7 +25,7 @@ with fraction field `K`, so that `Spec A` is a smooth affine curve over `𝔽_q`
2525
the height-one primes of `A` are its closed points, all but finitely many of the
2626
places of `K`; a number field `E` of coefficients, with `λ` ranging over its
2727
finite places, completion `E_λ` and residue characteristic `ℓ(λ)`; a finite set
28-
`S` of places of `A`; an integer `n ≥ 1`; and a choice of Frobenius element
28+
`S` of places of `A`; an integer `n`; and a choice of Frobenius element
2929
`Frob_v ∈ G_K` at each place `v`.
3030

3131
A family `(ρ_λ)` of continuous homomorphisms `ρ_λ : G_K → GL_n(E_λ)` is a
@@ -35,8 +35,8 @@ representation `ρ_λ` is unramified at `v` and the characteristic polynomial of
3535
`ρ_λ(Frob_v)` is the image of `P_v`. The polynomial is chosen before `λ`, not
3636
after; that quantifier order is the content of the definition.
3737

38-
The companion theorem, as formalised here, reads as follows. Let `ℓ(λ₀) ≠ p` and
39-
let `ρ₀ : G_K → GL_n(E_{λ₀})` be continuous, absolutely irreducible, of
38+
The companion theorem, as formalised here, reads as follows. Let `n ≥ 1` and
39+
`ℓ(λ₀) ≠ p`, and let `ρ₀ : G_K → GL_n(E_{λ₀})` be continuous, absolutely irreducible, of
4040
finite-order determinant, unramified at every `v ∉ S`, and such that the
4141
characteristic polynomial of `ρ₀(Frob_v)` is the image of `P_v` for a fixed
4242
family `P_v ∈ E[T]`. Then for every finite place `λ` of `E` with `ℓ(λ) ≠ p` there
@@ -50,8 +50,10 @@ follows by choice, and `exists_companion_family` is that derivation.
5050

5151
This is the companion-existence part of Lafforgue's Theorem VII.6(v), not the
5252
whole of VII.6. VII.6 starts from an irreducible lisse sheaf with finite-order
53-
determinant and constructs `E` and the `P_v`, which are hypotheses here; it also
54-
proves purity, integrality, and a further descent statement about `(σ_λ)^r`. In
53+
determinant and constructs `E` and the `P_v`, which are hypotheses here; it also proves
54+
purity, unit-ness of the Frobenius roots away from `p` with slope bounds at the
55+
places over `p`, and a further statement descending the `n`-th power of the
56+
companion to `E_λ` itself. In
5557
the other direction, the curve here is the affine `Spec A` with `S` removed,
5658
where VII.6 treats a smooth curve as an open subscheme of a projective one.
5759

@@ -72,9 +74,11 @@ the companion factors, the trace is continuous and conjugacy-invariant, and `E_
7274
is closed in `M`. So `[M : E_λ] = n!` suffices, and is not claimed minimal.
7375

7476
Drinfeld's Theorem 1.1 carries a hypothesis that the roots of the characteristic
75-
polynomials are `λ`-adic units. For curves that is a consequence of absolute
76-
irreducibility and finite-order determinant rather than an extra assumption, so
77-
nothing has been dropped, but the mechanism is not purity. Being a unit away from
77+
polynomials are `λ`-adic units. For curves that is a consequence rather than an
78+
extra assumption: unramifiedness on the curve makes the representation a lisse
79+
sheaf there, and Lafforgue proves unit-ness for an irreducible lisse sheaf with
80+
finite-order determinant. Nothing has been dropped, and the mechanism is not
81+
purity. Being a unit away from
7882
`p` is part (iii) of Deligne's Conjecture 1.2.10, where the label (c) is
7983
Drinfeld's numbering in his abbreviated restatement rather than Deligne's. It is
8084
a conclusion separate from the weight condition, which is part (i) there, and for
@@ -171,7 +175,10 @@ here.
171175

172176
`HeightOneSpectrum A` indexes an affine model, which is a statement of scope
173177
rather than a shortfall: `U = Spec A` with `S` removed is a smooth affine curve
174-
and Lafforgue's theorem applies to it directly. What follows from it is that the
178+
and Lafforgue's theorem applies to it, up to one change of convention: the
179+
Frobenius used here is arithmetic, where Lafforgue writes the characteristic
180+
polynomials for geometric Frobenius, its inverse, and the two determine each
181+
other. What follows from it is that the
175182
places of a proper model outside `Spec A` are not points of `U`, so "unramified
176183
outside `S`" says nothing about them, and the group here is not `π₁^ét(X ∖ S)`
177184
for projective `X`.

0 commit comments

Comments
 (0)