Professional Documents
Culture Documents
Received by the editors June 29, 2010 and, in revised form, March 28, 2011, and April 11,
2011.
2010 Mathematics Subject Classification. Primary 11R23, 11R42, 11S40.
The authors acknowledge financial support provided by DFG and NSERC.
1 Where 0 is denoted 0̃.
2011
c American Mathematical Society
Reverts to public domain 28 years from publication
1015
1016 JÜRGEN RITTER AND ALFRED WEISS
A≤U≤G
Here verA
U : Λ∧ U → Λ∧ A extends the group transfer U → U ab → A to a ring
homomorphism between their Iwasawa algebras in the customary way6 . Theorem
2 is proved in §2.
Let S denote the set of all non-archimedean primes of S and let k ⊆ f ⊂ K with
[f : k] < ∞. The pseudomeasure λf = λf,S of [Se] is associated to the maximal
abelian S -ramified extension fS of f . We extend this notation to intermediate
G(F/f )
fields F of fS /f∞ with [F : f∞ ] < ∞ by defining λF/f = deflG(f /f ) λf in the
S
sense of Lemma 4.1. Now we can state
Theorem 3. Using notation as in Theorem 2,
μQ (U/A)verAU ab (λK [U,U ] /K U ) ≡ 0 mod trQ (Λ∧ A).
A≤U≤G
Added in proof. While this paper was with the referee, the authors learned about
D. Burns’s paper On main conjectures in non-commutative Iwasawa theory and
related conjectures (December 2010; submitted for publication) as well as the one
by M. Kakde, The main conjecture of Iwasawa theory for totally real fields
(arXiv:1008.0142 [math.NT]). The first describes various applications of the THE-
OREM to the equivariant Tamagawa number and other conjectures; it also shows
how it implies the main conjecture of non-commutative Iwasawa theory for Tate
motives over compact l-adic Lie extensions of totally real fields of arbitrary finite
rank [FK]. The second gives an alternative proof of the THEOREM with respect
to the localized K1 of [FK].
1. “Proof” of Theorem 1
We begin by quickly recalling some basic definitions.
(1) 0: The Galois group sequence X G(M/k) G is turned into a canon-
ical ΛG-module sequence which is the bottom row of
ΛG = ΛG
ψ ˇ↓ ψ ˇ↓
aug
X Y → ΛG Zl .
In it Y has projective dimension ≤ 1 and ψ has aug ◦ ψ = 0.
Clearly, the cokernels of ψ and ψ are torsion of finite projective dimen-
sion which permits us to set 0 = [coker ψ] − [coker ψ ] in K0 T (ΛG); 0 is
independent of the choice of ψ , ψ. (See [RW1, §1] and [RW2, pp. 562–563].)
(2) LK/k : Choosing a topological generator γk of Γk gives a u ∈ 1 + lZl so
that ζlγ∞k = ζlu∞ . By [Ca], [DR], [Gr], and [Wi] the S-truncated l-adic
Artin L-functions Ll,S (s, χ) of K/k for characters χ of irreducible Ql c -
representations of G with open kernel admit a representation Ll,S (1−s, χ) =
Gχ,S (us −1)
Hχ (us −1) with Gχ,S (T ) ∈ Zl [[T ]] and Hχ (T ) = 1 or = χ(γk )(1 + T ) − 1
c
G (γ −1) ×
according to res H G χ = 1 or = 1. Set LK/k (χ) = Hχ (γk −1) in (Q Γk ) =
8 χ,S k c
(Ql c ⊗Ql QΓk )× ; this is independent of the choice of γk . (See [RW2, pp. 563,
571].)
7 As the referee has pointed out to us, the descent property discussed in the appendix could
also be deduced from interpreting 0 in terms of the compactly supported étale cohomology of
Zl (1).
8 Z c denotes the integers in a fixed algebraic closure Q c of Q and H = ker(G Γ ).
l l l k
ON THE “MAIN CONJECTURE” OF EQUIVARIANT IWASAWA THEORY 1019
(3) Hom∗ (Rl (G), (Qc Γk )× ): Here Rl (G) is the character ring of all the χ
above and Hom∗ consists of all f : Rl (G) → (Qc Γk )× satisfying f (χσ ) =
f (χ)σ for σ ∈ G(Ql c /Ql ) and f (χ ⊗ ρ) = ρ (f (χ)) for all irreducible charac-
ters ρ with H ⊂ ker ρ, where ρ is the Ql c -automorphism of Qc Γk induced
by ρ (γk ) = ρ(γk )γk . (Compare also [RW2, p. 558].)
(4) HOM: This notation adds the further condition f (χ)l ≡ Ψ(f (ψl χ))
mod lΛc Γk to Hom∗ , with Λc Γk = Zl c ⊗Zl ΛΓk , ψl the l-th Adams operation
on Rl (G) (so (ψl χ)(g) = χ(g l ) for g ∈ G) and Ψ the ring endomorphism of
ΛΓk induced by γk → γkl . (See [RW3, p. 37].)
We are now in a position to explain the main ingredients of the proof of Theorem
1.
Proposition 1.1. The THEOREM holds if and only if LK/k ∈ DetK1 (Λ∧ G).
Proof. For pro-l extensions K/k this is Theorem A in [RW3] which is reduced via
the classical Main Conjecture, as in [RW3, §1], to the stronger form of Theorem
B in [RW3, §6]. It is here that the logarithm L appears first; it is defined by the
commutative square (see [RW3, Proposition 11])
L
K1 (Λ∧ G) → T (Q∧ G)
(1.D1) Det ↓ Tr ↓
Hom∗ (Rl (G), Qc∧ Γk ),
L
HOM(Rl (G), (Λc∧ Γk )× ) →
with T (R) = R/[R, R] for any ring R, where [R, R] is the additive subgroup gen-
erated by all Lie commutators [a, b] = ab − ba, a, b ∈ R, and with the isomorphism
l
‘Tr’ induced by the reduced trace of Q∧ G. Above, (Lf )(χ) = 1l log Ψ(ff (χ)
(ψl χ)) . Note
that every f ∈ DetK1 (Λ∧ G) satisfies the extra congruence for HOM; note also that
LK/k ∈ HOM [RW3, pp. 34, 42].
The second basic tool in the proof of [RW3, Theorem B] is the Wall congruence
(i.e., the special case |Q| = l of Theorem 2); see [RW3, Lemma 12]. It also plays
an important role later.
The generalization to arbitrary extensions K/k is carried out in [RW4, Theorems
(A) and (B)] by the induction techniques there.
The logarithm L is called integral when it takes values in T (Λ∧ G) ⊆ T (Q∧ G). We
should stress that (1.D1) is available to define L for arbitrary groups G = G(K/k).
This also applies to much of [RW3], in which the pro-l hypothesis is needed only
in §1, Theorem 8, Proposition 11, and §6 of [RW3] (where it is explicitly stated)9 .
The main exception is the integrality property, L(K1 (Λ∧ G)) ⊆ T (Λ∧ G), which
nevertheless holds when G is l-elementary. This will be discussed in §7.
As a direct consequence, there is a unique element tK/k ∈ T (Q∧ G), the loga-
rithmic pseudomeasure of K/k, such that Tr(tK/k ) = L(LK/k ). In the l-elementary
case, by (1.D1) and (i) of Lemma 7.1 in §7, LK/k ∈ DetK1 (Λ∧ G) implies tK/k ∈
T (Λ∧ G), which is the easy implication in Proposition 1.3 below.
Proposition 1.2. If LK/k ∈ DetK1 (Λ∧ G) holds whenever K/k is l-elementary,
then LK/k ∈ DetK1 (Λ∧ G) for arbitrary extensions K/k.
Proof. This is [RW4, Theorem (C)].
9μ = 0 is not needed for Theorem 9 of [RW3], but only for its corollary.
1020 JÜRGEN RITTER AND ALFRED WEISS
Proposition 1.3. If K/k is l-elementary, then ‘LK/k ∈ DetK1 (Λ∧ G)’ and ‘tK/k ∈
T (Λ∧ G)’ are equivalent.
Remark 1.1. For the convenience of the reader, it should perhaps be added that
Λ• G = Σ−1 ΛG with Σ = ΛΓ \ l · ΛΓ for any central open subgroup Γ Zl of an
arbitrary G = G(K/k). Note that Σ−1 ΛΓ has the unique maximal ideal lΣ−1 ΛΓ.
So it suffices to show that every element c ∈ Σ−1 ΛG, which is (left) regular modulo
·c
l, is a unit of Σ−1 ΛG. For this consider right multiplication Σ−1 ΛG −→ Σ−1 ΛG
by c. Since Σ−1 ΛG/l is a finite-dimensional Σ−1 ΛΓ/l-vector space, c mod l has a
(left) inverse in Σ−1 ΛG/l, and hence c has a left inverse b in Σ−1 ΛG by Nakayama’s
lemma. Since b mod l is now also (left) regular modulo l, the same argument
provides a ∈ Σ−1 ΛG with ab = 1. Then a = abc = c, so c is a unit.
ON THE “MAIN CONJECTURE” OF EQUIVARIANT IWASAWA THEORY 1021
2. Proof of Theorem 2
Fix a set of coset representatives rq of A in G, whence G = ˙ q∈Q rq A, q = rq A,
and rq1 rq2 = rq1 q2 aq1 ,q2 with aq1 ,q2 ∈ A a 2-cocycle, so aq1 ,q2 q3 aq2 ,q3 = aq1 q2 ,q3 aqq31 ,q2 .
Further, let Σ = Sym(Q) denote the symmetric group on the elements of Q. It
carries the natural (right) Q-action
π q (q1 ) = π(q1 q −1 )q, q, q1 ∈ Q, π ∈ Σ,
satisfying (π1 π2 ) =q
π1q π2q .
For V ≤ Q, the set of fixed points ΣV of V in Σ is thus
a subgroup of Σ. Note that π ∈ ΣV has π(qv) = π(q)v for all q ∈ Q, v ∈ V .
Lemma 2.1. Let U be a subgroup of G containing A, set V = U/A, and fix a
section κ : Q/V → Q, so (κs)V = sV for s ∈ Q/V . Let = q∈Q rq eq , with
eq ∈ Λ∧ A, be a unit in Λ∧ G = q∈Q rq · Λ∧ A. Then
κ(s)
verA U
U res G = sgn(π) aπ(q)q−1 ,q verA
U (eπ(κs)κ(s)−1 ) .
π∈ΣV q∈Q s∈Q/V
Proof. Writing Λ∧ G = s∈Q/V rκ(s) Λ∧ U , then
rv a−1
κ(s1 )
rκ(s1 ) = rκ(s2 ) κ(s2 ),v aκ(s2 )vκ(s1 ) ,κ(s1 ) · eκ(s )vκ(s
−1 −1
2 1)
s2 ∈Q/V v∈V
U : Λ∧ U →
(with the term in parentheses in Λ∧ U ). The ring homomorphism verA
× det
Λ∧ A induces the map verA U : K1 (Λ∧ U ) → K1 (Λ∧ A) = (Λ∧ A) , and we compute
A U A
verU res G by applying verU to the matrix of the action of on the right Λ∧ U -
module Λ∧ G to get
verA U
U res G
−1 κ(s)
= sgn(σ) U (rv aκ(σs),v aκ(σs)vκ(s)−1 ,κ(s) · eκ(σs)vκ(s)−1 )
verA
σ∈Sym(Q/V ) s∈Q/V v∈V
−1 κ(s)
= sgn(σ) U rf (s) aκ(σs),f (s) aκ(σs)f (s)κ(s)−1 ,κ(s) · eκ(σs)f (s)κ(s)−1
verA
σ∈Sym(Q/V ) f s∈Q/V
Proof. We first observe that this holds for all a ∈ A because vrV (as ) = v∈V asv , so
q
q ∈Q a is independent of V ≤ Q , and
s
s∈Q /V vrV (a ) = V ≤Q μQ (V ) = 0.
It therefore suffices to prove additivity of the left side of the claimed congruence,
i.e.,
μQ (V ) vrV ((e0 + e1 )s )
V ≤Q s∈Q /V
≡ μQ (V ) vrV (es0 ) + μQ (V ) vrV (es1 )mod trQ (Λ∧ A).
V ≤Q s∈Q /V V ≤Q s∈Q /V
We proceed by induction on |Q |; the case Q = 1 is trivial.
Let F = F (Q ) denote the set of maps f from Q to F2 , with Q -action (f q )(x) =
f (x(q )−1 ) for all x ∈ Q . Then
(e0 + e1 )s = esf (s) ,
s∈Q /V f ∈F V s∈Q /V
−1
10 tr
Q /V (e) = s∈Q /V es , V ≤ Q , e ∈ (Λ∧ A)V .
1024 JÜRGEN RITTER AND ALFRED WEISS
˙ due to f y −1 = f . Thus
with =
μ̃(V, f ) = μQ (V ) vrV (esf (s) ) = μW (V ) vrV (es∗ );
s∈Q /V s∈W/V
hence
−1
trW/NW (V ) μ̃(V, f ) = μ̃(V, f )t
t∈W/NW (V )
−1 −1
=
˙ μ̃(V t , f ) = μW (V t ) vrV t−1 (es∗ )
t∈W/NW (V ) t∈W/NW (V ) s∈W/V t−1
˙ by f t−1 = f .
with =
We substitute this into our sum (2.d) to get
trQ /W ( trW/NW (V ) (μ̃(V, f ))
V W
= trQ /W ( μW (V ) vrV (es∗ ))
V ≤W s∈W/V
Proof. First, sgn(π q ) = sgn(π) holds since {xi : i mod b} a cycle of π implies
{xi q : i mod b} is a cycle of π q . Second,
q
a−1
1
( aπ(q1 )q−1 ,q1 )q = aπ(q )q−1 ,q = aπ(q1 )q−1 ,q1 q π(q1 ),q aq1 ,q
1 1 1 1 1
q1 q1 q1 q1 q1
2
3
= aπ(q1 )q−1 ,q1 q = aπ(q1 q−1 )qq−1 ,q1 = aπq (q1 )q−1 ,q1
1 1 1
q1 q1 q1
1 2 3
with = due to the cocycle relation, = to π permuting the q1 , and = to the
substitution q1 q1 q −1 .
holds for all π ∈ Σ. Namely, assuming (2.f), its left side can be written as trStQ (π) (α)
for some α ∈ Λ∧ A. Since r̃(π) ∈ (Λ∧ A)StQ (π) , it follows that
trQ/StQ (π) (r̃(π)trStQ (π) (α)) = trQ (r̃(π)α)
is the orbit sum of π in (2.e), by Claim 2.B.
We next observe that (2.f) is a consequence of Lemma 2.2. To see this, let
π ∈ ΣV be given and set Q = StQ (π) and e = x∈Q/Q exπ(x)x−1 ∈ Λ∧ A, where
Q = ˙ xQ . Setting Q = ˙ yV , then Q = ˙ xyV and the V -term in (2.f) is
x y x,y
μQ (V ) vrV (exy
π(xy)(xy)−1 ) =
˙ μQ (V ) vrV (exy
π(x)x−1 ) = μQ (V )
vrV (ey )
x,y x,y y∈Q /V
Note that G(LS /F)ab = G(FS /F) and, for F ⊆ F , that G(LS /F ) is an open
subgroup of G(LS /F). This yields the transfer map G(FS /F) → G(FS /F ), and
in particular,
verF L L L F
K : GS → G(FS /F), verF : G(FS /F) → HS , verK = verF ◦ verK : GS → HS .
We recall that the Möbius function μ = μΣ of the poset of subgroups of the finite
group Σ is defined by
μ(1) = 1, μ(Σ ) = − μ(Σ ) for 1 = Σ ≤ Σ.
1≤Σ <Σ
For K ⊆ F ⊆ L and g ∈ GS set λ̃F = 2−[F :Q] λF ,S and λ̃gF = (1 − gF )λ̃F , where
gF = verF
K g. Moreover, denote the Galois group of the cyclotomic Zp -extension
F∞ /F by ΓF .
Theorem 5. There exists g ∈ GS so that gF has non-trivial image in ΓF for
K ⊆ F ⊆ L and the image of
μΣ (G(L/F))verL
F (λ̃gF ),
K⊆F ⊆L
Σ
under Zp [[HS ]] → Zp [[HS+ ]], is in the trace ideal trΣ (Zp [[HS+ ]]) of Zp [[HS+ ]] .
The organization of the proof parallels [RW7]. The new ingredient is the identi-
fication of the congruence (3.4) of Lemma 3.4 as the difference of constant terms
of q-expansions at two cusps of a Hilbert modular form of Eisenstein type. This
modular form is exhibited in §3.2 and then studied via the q-expansion principle of
Deligne and Ribet; the hypothesis of Lemma 3.4 is deduced, in Lemma 3.6, from a
property of Möbius coefficients in [HIO]. The proof, in §3.3, of the main result of
this section, Theorem 5, is then a computation of constant terms of q-expansions
at the “special cusps” of Lemma 3.5.
Our special cusps are a simple device to avoid comparing constant coefficients
of q-expansions of F and F|k Uβ at arbitrary cusps in [RW7, Lemma 6]. Having
overlooked the need for this comparison in [RW7] implies that we now have its
Theorem only for special g = gK ∈ GS , i.e., those in Lemma 3.7: this can be
deduced from Theorem 5 for odd p by [RW7, Lemma 5]. However, Theorem 5 is
better suited for the application to equivariant Iwasawa theory; see Remark 4.1.
3.1. A sufficient condition for a pseudomeasure congruence. For a coset x
of an open subgroup U of G(FS /F) set δ (x) (g) = 1 or 0 according as g ∈ x or not.
Then, for even integers k ≥ 1, define ζ̃F (1 − k, δ (x) ) = 2−[F :Q] ζF ,S (1 − k, δ (x) ) ∈ Q
to be 2−[F :Q] times the value at 1 − k of the partial ζ-function for the set of integral
ideals a of F prime to S with Artin symbol (a, FS /F) in x. Note that the definition
of ζ̃F (1 − k, δ (x) ) extends linearly to locally constant functions ε on G(FS /F) with
values in a Q-vector space and gives values ζ̃F (1 − k, ε) in that vector space, as
usual.
Let N = NF ,p : G(FS /F) → Zp × be that continuous character whose value on
(a, FS /F) for an integral ideal a of F prime to S is its absolute norm NF a. For
g ∈ G(FS /F), k ≥ 1, and ε a locally constant Qp -valued function on G(FS /F) we
define, following [DR],
Δ̃g (1 − k, ε) = ζ̃F (1 − k, ε) − N (g)k ζ̃F (1 − k, εg ) ∈ Qp ,
where εg (g ) = ε(gg ) for g ∈ G(FS /F).
ON THE “MAIN CONJECTURE” OF EQUIVARIANT IWASAWA THEORY 1027
for all admissible U , where N here also denotes the homomorphism G(FS /F)/U →
(Zp /pm(U) )× induced by our previous N . Moreover, if λ̃F is 2−[F :Q] times the
pseudomeasure of [Se], then (1 − h)λ̃F = λ̃h .
Proof. The lemma follows from the theorem above; see [RW7, Proposition 2].
−1
Lemma 3.3. (1) Let V be an admissible open subgroup of HS . If U ≤ (verL F) (V ),
then mF (U ) ≥ mL (V ) − eF where [L : F] = p .eF
to check that ζ̃F (1 − k, χ) = ζ̃F σ (1 − k, χs ), and this follows from the compati-
bility of the Artin L-functions with inflation and induction. Indeed, inflating χ
from G(FS /F)/U to G(FS /F) and further to G(LS /F) and then inducing up to
G(LS /K), and analogously with χ, F, U replaced by χs , F σ , U s (note that U s is
well-defined), we have
G(L /K) G(L /F ) G(L /K) G(L /F σ )
ind G(LSS /F ) inflG(FSS /F )/U (χ) = ind G(LSS /F σ ) inflG(FSs /F σ )/U (χs ).
S
for all even locally constant Z(p) -valued functions εL on HS , implies that
def
sg = μΣ (ΣF )verL
F (λ̃gF )
F
has image, under the map Zp [[HS ]] → Zp [[HS+ ]], in trΣ (Zp [[HS+ ]]).
Proof. For the proof of the lemma we first recall that a locally constant function
εL on HS is even if εL (cw h) = εL (h) for all h ∈ HS and all ‘Frobenius elements’
cw at the archimedean primes w of L, i.e., at the restrictions cw ∈ HS of complex
conjugation with respect to the embeddings LS → C inducing w on L. We denote
by C the group generated by the cw ’s, so HS+ = HS /C.
We next observe that sg ∈ Zp [[HS ]]Σ . Namely, verL s L s
F (λ̃gF ) = verF σ (λ̃gF ) =
L
verF σ (λ̃gF σ ), by (2) of Lemma 3.3. Moreover, μΣ (ΣF ) = μΣ (ΣF σ ).
Turning finally to the image of sg under the map Zp [[HS ]] → Zp [[HS+ ]], we first
replace the diagram in [RW7, p. 718] by the diagram below, in which N = ker verL F:
F ↓
verL ↓
Zp [[HS ]] → lim
←
Zp [HS /V ]/pmL (V )−eF .
Σ − stable V
Recall here that the right vertical map takes (xU )U to (yV )V by means of
ver
Ξ: Zp [G(FS /F)/U ]/pmF (U) −→ Zp [HS /V ]/pmF (U) → Zp [HS /V ]/pmL (V )−eF ,
−1
whenever U ≤ (verL F) (V ).13
Since the mL (V )’s are unbounded, there are admissible open Σ-stable V ≤ HS
with mL (V ) − eF ≥ eK ( ∀ F ). For any such V then Zp [HS /V ]/pmL (V )−eF maps
onto Zp [HS /V ]/|Σ| and we write the image of sg here as c y. Because Σ
y∈HS /V y
fixes sg , cyσ = cy for all σ. Since σ∈Σ mod StΣ (y) cyσ y σ = cy y σ , it follows that
sg will be in trΣ (Zp [HS /V ]) + |Σ|Zp [HS /V ] provided that
cy ≡ 0 mod |StΣ (y)|.
13 Note that ‘ver’ is the Zp -linear map induced by the group homomorphism obtained by
verL
factoring G(FS /F ) →F HS → HS /V through G(FS /F ) → G(FS /F )/U .
ON THE “MAIN CONJECTURE” OF EQUIVARIANT IWASAWA THEORY 1029
(x)
where δF is the characteristic function of the coset x ⊆ G(FS /F).
verL −1
Since G(FS /F) →F HS → HS /V has kernel U = (verL
F) (V ), either y is not in
L L (F )
the image of verF or y = verF (x ) for a unique x (F )
∈ G(FS /F)/U . Note
(y) (x(F ) )
that δL verL F is = 0 in the first case and = δF in the second, when also
NF (x(F ) )−|ΣF |k = NL (verL
F (x
(F ) −k
)) . Thus, Möbius-summing over F, we obtain
(y) −k
cy = ( μΣ (ΣF )Δ̃gF (1 − |ΣF |k, δL verL
F ) )NL (y) .
F
Our hypothesis (3.4) now implies that sg is in trΣ (Zp [HS /V ]) + |Σ|Zp [HS /V ] for
(y)
all V ≥ C (recall that δL is even when V ≥ C and y ∈ HS /V ). Since sg is fixed
by Σ and |Σ|Zp [HS /V ] ⊆ trΣ (Zp [HS /V ]), it follows that sg ∈ trΣ (Zp [HS /V ]).
Σ
The argument for F is the same with K, f, S replaced by F, foF , SF and shows
that ψF (αF ) acts on p-power roots of unity by (1 + f )−[F :Q] . Note also that
ψF (αF ) = verF
K (g) by the usual relation between inclusion and transfer. Setting
γ = αK , we then get (a) and (b).
This completes the proof of Lemma 3.5.
The next three results, Lemmas 6 and 7 and Proposition 8 of [RW7], concern
Hilbert modular forms with emphasis on their q-expansions. [RW7, Lemma 6]
constructs a Hecke operator Uβ on Mk (Γ00 (f), C), [RW7, Lemma 7] discusses re-
striction res K
F : Mk (Γ00 (foF ), C) → M[F :K]k (Γ00 (f), C) for field extensions F/K,
and [RW7, Proposition 8] is our bridge to [DR].
With k, εL , and f as at the beginning of the section and any g ∈ GS , we next
exhibit a Hilbert modular form E in M|Σ|k (Γ00 (f), C) with the constant term of
F μΣ (ΣF )Δ̃gF (1 − |ΣF |k, εL verF ) (compare
L
its standard q-expansion equal to
(3.4)).
First, by [DR, (6.1)], in the form of [RW7, Proposition 8], there are modular
forms
def
EF = G|ΣF |k,εL verLF ∈ M|ΣF |k (Γ00 (foF ), C)
of weight |ΣF |k with standard q-expansion
|ΣF |k−1 μ
ζ̃F (1 − |ΣF |k, εL verL
F) + εL verL
F (a)NF (a) qF .
μ0 μ∈a⊆oF
μ∈oF a prime to S
Appealing to [RW7, Lemmas 7 and 6], we apply res K F and the Hecke operator U[F :K]
to the modular form EF displayed above and obtain, for each F, the new modular
form
EF = (res K
F EF )|[F :K]||ΣF |k U[F :K]
where, for K ⊆ F ⊆ L, [α]F denotes the set of all pairs (αF , aF ) satisfying
0 αF ∈ aF ⊆ oF , aF prime to S, trF /K (αF ) = [F : K]α.
Here we have used (εL ◦ verL
F )(aF ) = (εL verF )(aF , FS /F) = εL (aF oL ).
L
Lemma 3.6. Assume that k is an even positive integer and εL is an even locally
constant Zp -valued function on HS . Then
μΣ (ΣF ) εL (aF oL )NF (aF )[L:F ]k−1 ≡ 0 mod |StΣ (εL )|Z(p) .
K⊆F ⊆L [α]F
Here, [StΣ (εL )∩StΣ (β, b)\StΣ (εL )] is a set of right coset representatives of StΣ (εL )∩
StΣ (β, b) in StΣ (εL ). Note that the sum (3.α ) is the sum of all such orbit sums (i).
Because of
• ΣF ≤ StΣ (β, b)σ ⇐⇒ ΣF σ−1 ≤ StΣ (β, b) (as StΣ (β, b)σ = (StΣ (β, b))σ
−1
and ΣσF = ΣF σ−1 ),
• μΣ (ΣF ) = μΣ (ΣF σ−1 ) (a direct consequence of the definition of the Möbius
function), and
• |ΣF | = |ΣF σ−1 |,
the inner sums of (i) are independent of σ. Hence, if we can show
− 1
(ii) μΣ (ΣF )NL (b) |ΣF | ≡ 0 mod |StΣ (εL ) ∩ StΣ (β, b)|Z(p) ,
F so ΣF ≤StΣ (β,b)
1032 JÜRGEN RITTER AND ALFRED WEISS
then sum (i) is ≡ 0 mod |StΣ (εL )|Z(p) and the proof of the proposition will be
complete.
For (ii), we first shorten the notation by setting P = StΣ (β, b) ≤ Σ and
− 1
r = NL (b) |StΣ (β,b)| which is a unit in Z(p) because b = aM oL if ΣM = P (com-
pare with the proof of (3.α ) above). This turns the left-hand side of (ii) into
[P :P ]
1≤P ≤P μP (P )r , as obviously μΣ (P ) = μP (P ). Applying now Claim 3.A
below we obtain
μP (P )r [P :P ] ≡ 0 mod |P |Z(p) ,
1≤P ≤P
3.3. Conclusion of the proof: Special cusps. By what has been said at the
end of the previous subsection we need to compute the constant coefficients of E(γ)
and E(1) − E(γ). We do this for the γ’s of Lemma 3.5 and then prove Theorem 5.
Lemma 3.7. Setting g = j(γ), with γ as in Lemma 3.5, the constant term of E(γ)
is
NK (g)|Σ|k · ( μΣ (ΣF )ζ̃F (1 − |ΣF |k, (εL verL
F )gF ) ).
F
at the cusp determined by γ ∈ K̂× . By (2) of [RW7, Lemma 7], this constant term
of res K ×
F EF is equal to the one of EF at the cusp determined by γ ∈ F̂ , whence,
by (2) of [RW7, Proposition 8], equals
NF ((γ)F )|ΣF |k ζ̃F (1 − |ΣF |k, (εL verL
F )gF ) = NK ((γ))
|Σ|k
ζ̃F (1 − |ΣF |k, (εL verL
F )gF ),
hence E(γ) has the required constant term, because NK (γp )−1 NK ((γ)) = NK,p (g),
by (3) of [RW7, Proposition 8].
This completes the proof of Lemma 3.7.
We can now complete the proof of Theorem 5. It follows from equation (3.0) and
Lemma 3.7 that for such γ the constant term of E(1) − E(γ) is
|Σ|k
μΣ (ΣF )[ζ̃F (1 − |ΣF |k, εL verL
F ) − NK (g) ζ̃F (1 − |ΣF |k, (εL verL
F )gF )]
F
= μΣ (ΣF )Δ̃gF (1 − |ΣF |k, εL verL
F ),
F
H+
has image in trQ (Zl [[HS+ ]]), under deflHSS : Zl [[H S ]] → Zl [[HS+ ]].
We first remove the ˜ on λgf in (4.a). Recall that λ̃f = 2−[f :Q] λf implies
λ̃gf = 2−[f :Q] λgf . So it suffices to show that
μQ (G(L/f ))2−[f :Q] ≡ μQ (G(L/f ))2−[L:Q] mod |Q|,
and this congruence in turn is a consequence of |Q| being a power of l = 2 and
μQ (G(L/f ))(2−[f :Q] − 2−[L:Q] ) = 2−[L:Q] μQ (G(L/f )) (2[L:f ]−1 )[f :Q] − 1 ,
2[L:f ]−1 ≡ 1 mod l by 2l ≡ 2 mod l, so (2[L:f ]−1 )[f :Q] ≡ 1 mod l[f : Q],
so it now suffices to show that μQ (G(L/f ))l[f : Q] is divisible by |Q| = [L : k]. Due
to [HIO, Corollary 3.5], μQ (G(L/f ))l is divisible by [L : f ], whence μQ (G(L/f ))l[f :
Q] by [L : f ][f : Q] = [L : Q].
f
Next, setting h = verL k g = verf verk g = verf gf , we have (1 − gf )λf = λgf ∈
L L
S H S H
because verL
f = verG(LS /f )ab = verG(f S /f ) . Thus
μQ (G(L/f ))verL f ((1 − gf )λf ) = (1 − h) μQ (G(L/f ))verL
f (λf )
k⊆f ⊆L k⊆f ⊆L
implies that every term of the second sum is in Mh−1 Zl [[H S ]]. Applying deflA
H S =
H+
deflA
H+
deflHSS to (4.a) then yields that
S
(4.b) k⊆f ⊆L μQ (G(L/f ))deflA L
H S verf (λf )
H+
is in deflA
H+
(trQ (Mh−1
+ Zl [[H S ]])) = trQ (Λ• A), where h
+
= deflHSS h.
S
In order to derive Theorem 3 from this, we set U = G(K/f ) and show that
H ab
S
(4.c) deflA L A
H S verf = deflH S verG(f = verA U
U ab deflG(f S /f ) .
S /f )
5. Proof of Theorem 4
We fix some notation. A is an abelian normal open subgroup of G = G(K/k) =
z×G[l] containing z with factor group Q = G/A, and Γ, C are central subgroups
of G contained in A, with Γ Zl open and C of exponent l. Further, U ≤ G is
open and contains C and z.
Recall that ResU G satisfies diagram (1.D2) and is discussed in §7 and [RW8,
Proposition A]. In particular, we know that ‘Res’ is additive and transitive and that
it preserves integrality. Also, if g ∈ G, then conjugation by g canonically induces
maps Λ∧ U → Λ∧ U g and T (Λ∧ U ) → T (Λ∧ U g ), the latter by τU (u)g = τU g (ug ) for
u ∈ U .16
Lemma 5.1. (1) If [G : U ] = l and α ∈ Λ∧ Γ, g ∈ G, we have
α x∈G/U τU (g x ) if g ∈ U,
ResU (ατ G (g)) =
G
Ψ(α)τU (g l ) if g ∈
/ U,
where Ψ : Λ∧ Γ → Λ∧ Γ is the continuous Zl -linear ring homomorphism induced by
γ → γ l for γ ∈ Γ.
(2) For g ∈ G, c ∈ C, and α ∈ Λ∧ Γ,
G (ατG (g(c − 1))) = α
ResU τ̇U (g x (c − 1))
x∈G/U
with x ∈ G/U meaning G = ˙ x xU and τ̇U : Λ∧ G → T (Λ∧ U ) defined by extending
τU to take the value 0 outside of Λ∧ U .
(3) Let V ≤ U be open subgroups of G containing z and let g ∈ G, w ∈ T (Λ∧ G).
g
Then (ResVU w)g = ResVU g (wg ).
Proof. For α ∈ Γ the claimed formula (1) is already given at the bottom of [RW8,
p. 127]; additivity and continuitythen imply it in general.
For (2), we first remark that x∈G/U τ̇ (g x (c − 1)) is independent of a special
choice of coset representatives x of U in G and on replacing g by g s for s ∈ G. We
proceed by induction on [G : U ]. If U < G, choose U ≤ G < G, [G : G ] = l, so
G G. Then
G (ατG (g(c − 1))) = ResG ResG (ατG (g(c − 1)))
ResU U G
α x
x ∈G/G τG (g (c − 1)) if g ∈ G ,
= ResU G
Ψ(α)τG ((g c) − (g ) )
x l x l
ifg ∈
/G
x ∈G/G ResG ατG (g (c − 1)) if g ∈ G ,
U x
=
0 if g ∈/ G .
In the first case, the induction hypothesis turns this into
α τ̇U (g x x (c − 1)) = α τ̇U (g x (c − 1)),
x ∈G/G x ∈G /U x∈G/U
as required. In the second case, when g ∈ / G , we must show that the right-hand
side of the assertion is zero, which however is clear since g x ∈ U implies g x ∈ G ;
hence g ∈ G . This finishes the proof of (2).
16 τ
U : Λ∧ U → T (Λ∧ U ) is the natural map.
ON THE “MAIN CONJECTURE” OF EQUIVARIANT IWASAWA THEORY 1037
!
in which the equality = still needs to be verified. However, here we are in the
index l case, so V g V1g , and we can apply (1).
Lemma 5.1 is established.
Lemma 5.2. Denote the map G → G/C by and define a = ker(Λ∧ G → Λ∧ G),
b = ker(Λ∧ A → Λ∧ A), so a Λ∧ G Λ∧ G and b Λ∧ A Λ∧ A are exact.
Then
(i) bQ /trQ b → (Λ∧ A)Q /trQ (Λ∧ A) is injective,
(ii) τG (a) T (Λ∧ G) T (Λ∧ G) is exact and ResAG τG (a) = trQ b.
Proof. Applying Tate cohomology to the sequence defining b, we see that the
claimed injectivity (i) is a consequence of H −1 (Q, Λ∧ A) = 0. In order to see
this vanishing, choose a central open Γ Zl of G contained in A and pick Q-orbit
representatives ai Γ of A/Γ. Set Qi = stabQ (ai Γ). For qi ∈ Qi , aiqi −1 ∈ Γ has finite
order; hence aqi i = ai , i.e., Qi = stabQ (ai ). Hence we have a set of representatives
of Γ in A consisting of Q-orbits Qi \Q for some Qi ≤ Q and consequently, [RW3,
Lemma 5],
H −1 (Q, Λ∧ A) = H −1 (Q, ind Q
Qi (Λ∧ Γ)) = H −1 (Qi , Λ∧ Γ) = 0,
i i
as Λ∧ Γ has Zl -torsion = 0.
For the first claim in (ii) we only need to check exactness at the middle, or, more
precisely, that deflG G τG (v) = 0 implies τG (v) ∈ τG (a). Now, deflG (v) =
G
i [w i , xi ]
with wi , xi ∈ Λ∧ G implies that v − i [wi , xi ] ∈ ker deflG G = a, so apply τG and
arrive at τG (v) ∈ τG (a).
Regarding the second claim of (ii), the elements of a are Λ∧ Γ-linear combinations
of τG (g(c − 1)). By (2) of Lemma 5.1, ResA G of this equals x∈G/A τ̇A (g (c − 1)) =
x
Lemma 5.3. Using notation as in Lemma 5.2, we have L(1 + trQ b) ⊂ trQ (Λ∧ A).
Proof. Let β = a∈A,c∈C βa,c a(c − 1) be an element in b, where βa,c ∈ Λ∧ Γ for
some central open Γ Zl contained in A. Now
1 (1 + trQ β)l
L(1 + trQ β) = log ,
l Ψ(1 + trQ β)
by the argument given in [RW3, pp. 39–40], which also works in the situation when
G is l-elementary and the unit u to which L is applied (see [RW3, p. 39, (∗)]) is in
Λ∧ A rather than in ΛΓ (the ring O there is Zl here, so the Frobenius automorphism
Fr is trivial). The point is that [RW3, p. 40, (∗∗)] applies to an abelian situation
and so we still need only consider degree 1 characters.
1038 JÜRGEN RITTER AND ALFRED WEISS
(1+tr β)l
Thus L(1 + trQ β) ∈ trQ (Λ∧ A) if Ψ(1+trQQ β) ≡ 1 mod ltrQ (Λ∧ A). Since (1 +
trQ β)l ≡ 1 + (trQ β)l mod ltrQ (Λ∧ A), it suffices to show that (trQ β)l ≡ Ψ(trQ β)
mod ltrQ (Λ∧ A).
Now, as Ψ(a) = al for a ∈ A [RW3, p. 33], Ψ(trQ β) = a,c Ψ(βa,c )(trQ ((ac)l ) −
trQ (al )) = 0, since (ac)l = alcl = al , and we are left to check
that (trQ β)l ≡ 0
mod ltrQ (Λ∧ A). But trQ β = c κc (c − 1) with κc = trQ ( a βa,c a) ∈ trQ (Λ∧ A),
so
(trQ β)l ≡ κlc (c − 1)l mod ltrQ (Λ∧ A)
c
as trQ (Λ∧ A) is an ideal in (Λ∧ A)Q and c ∈ (Λ∧ A)Q . So (c − 1)l ≡ 0 mod l(Λ∧ A)Q
establishes Lemma 5.3.
K1 (Λ∧ G) → K1 (Λ∧ G)
in which the (natural) vertical maps are surjective, since Λ∧ G, Λ∧ G are semi-local
rings. Moreover, the top horizontal map is surjective as well, because ker(Λ∧ G →
Λ∧ G) ⊂ rad(Λ∧ G).
By Proposition 1.3, tG ∈ T (Λ∧ G) implies LK C /k = Det θ with θ ∈ K1 (Λ∧ G).
Observe that res A
G
θ = λA , because
Det(res A
G
θ) = res A
G
(Detθ) = res A
G
(LK C /k ) = LK C /K A = DetλA
and SK1 (Λ∧ A) = 1, as Λ∧ A is commutative semilocal [CR, (45.12)].
The above square gives a ϑ ∈ K1 (Λ∧ G) with deflG G ϑ = θ. Define ϑ ∈ K1 (Λ∧ A)
A A
by res G ϑ = λA ϑ ; hence deflA ϑ = 1.
Further define ξ = L(ϑ), ξ = L(ϑ ). Then, using diagrams (1.D1) and (1.D2),
A
ξ ∈ T (Λ∧ G) has deflG A
G ξ = tG , ResG ξ = tA + ξ , and deflA ξ = 0.
The exact sequences displayed above give rise to the commutative diagram
τG (a) T (Λ∧ G) T (Λ∧ G)
↓ G ↓
ResA A
ResG ↓
b Λ∧ A Λ∧ A
with top sequence exact by Lemma 5.1. We need to modify ξ by adding an element
α ∈ τG (a) (so without changing deflGG
ξ) to arrange that ResA
G (ξ + α) = tA ; i.e., we
need to prove that ξ ∈ ResG (τG (a)).
A
ON THE “MAIN CONJECTURE” OF EQUIVARIANT IWASAWA THEORY 1039
ab
Write deflU U res G ϑ = λU ab ϑU for A ≤ U ≤ G. Then ϑU ∈ K1 (Λ∧ U ) =
U ab
ab
(ϑU ) = deflU ab
U/[U,U ]C U/[U,U ]C U/[U,U ]C
λU/[U,U ]C deflU ab deflU U
U res G ϑ = deflU deflU U
U res G ϑ
U/[U,U ]C U/[U,U ]C
= deflU res U G
G deflG ϑ = deflU res U
G θ = λU/[U,U ]C ,
of Theorem 3 gives ϑ ≡ 1 mod trQ (Λ∧ A), as λA is a unit in (Λ∧ A)Q and trQ (Λ∧ A)
an ideal. Therefore
ϑ ∈ (1 + bQ ) ∩ (1 + trQ (Λ∧ A)) = 1 + (bQ ∩ trQ (Λ∧ A)) = 1 + trQ b,
with the last equality due to Lemma 5.2. This proves (5.a).
Turning back to the proof of Theorem 4, we know that ξ = L(ϑ ) is in bQ . By
(5.a) and Lemma 5.3, we also have ξ ∈ trQ (Λ∧ A), hence ξ ∈ bQ ∩ trQ (Λ∧ A) =
trQ (b) = ResA
G (τG (a)), by Lemma 5.2.
Thus the proof of Theorem 4 is complete.
Lemma 6.1. Let C have order l and let U ≥ A satisfy C ∩ [U, U ] = 1. Denote
the normalizer of U in G by N = NG (U ) and let Y be a set of representatives of
N/U -orbits in U/ΓC[U, U ]. Then trN/U (cab
U ) has Λ∧ Γ-basis
Proof. For the proof, we use C ∩ [U, U ] = 1 = Γ ∩ [U, U ] to identify C, Γ with their
images in U ab (hence c with c̃). We investigate the N/U -structure of 0 → cab U →
Λ∧ U ab → Λ∧ (U ab /C) → 0 via the Λ∧ Γ-bases coming from the N/U -action on
C U ab /Γ U ab /ΓC by [RW3, Lemma 5].
Now Y is a set of representatives of N/U -orbits on U ab /ΓC. If ŷ is a preimage
of y ∈ Y under U ab /Γ → U ab /ΓC, then stabN/U (y) either fixes ŷ (which will be
referred to as Case 1) or moves ŷ (which will be referred to as Case 2); moreover this
case distinction is independent of the choice of ŷ. This permits us to analyze the
map Λ∧ U ab → Λ∧ (U ab /C) one y ∈ Y at a time in terms of the map of N/U -sets
from the preimage of the N/U -orbit of y to the N/U -orbit of y it induces. This
is because of the permutation Λ∧ Γ-basis given by choosing preimages ỹ of ŷ under
U ab → U ab /Γ with stabN/U (ỹ) = stabN/U (ŷ), as in the proof of Lemma 5.2.
Thus, in Case 1, the preimage of the N/U -orbit of y is ˙ c∈C (N/U -orbit of ŷc),
so l copies of stabN/UN/U (y)
as N/U -sets, and the map is ŷ n c → y n for n ∈ N, c ∈
C. So the kernel on Λ∧ Γ-permutation modules has Λ∧ Γ-basis {ỹ n (c − 1) : n ∈
N/U
stabN/U (y) , 1 = c ∈ C}.
Similarly, in Case 2, the preimage of the N/U -orbit of y is the N/U -orbit of ŷ:
here ŷ z = ŷγy (z), with γy a homomorphism stabN/U (y) C, has stabN/U (ŷ) as
its kernel. Now the kernel on Λ∧ Γ-permutation modules has Λ∧ Γ-basis {ỹ n − ỹ :
n ∈ stabN/U
N/U (ŷ)
}.
Hence cabU has Λ∧ Γ-basis the union of these over y ∈ Y , and trN/U (cU ) has the
ab
claimed Λ∧ Γ-basis since Y1 consists of the y ∈ Y in Case 1 and trN/U (ỹ n − ỹ) = 0
for all y ∈ Y in Case 2. This proves the lemma.
ab
Lemma 6.2. If v ∈ T (Λ∧ G) has deflU U
U ResG v = 0 for all subgroups U of G
containing A, then v = 0.
Thus L(res U U U U
G ω) = ResG (Lω) = 0 implies L(Det(res G ω)) = Tr(L(res G ω)) = 0;
U l U
hence Det(res G ω)(χ1 ) = Ψ((Det(res G ω))(ψl χ1 ) for all characters χ1 of U , by the
definition of L.
If χ1 is an irreducible character of U with kernel containing Γ, then χ1 = β ⊗
with β, irreducible characters of U with kernels containing U [l], zΓ, respectively.
Note that ψln χ1 = ψln β ⊗ ψln = β ⊗ 1 = β; hence
n
Det(res U
G ω)(χ1 )
l
= Ψn (Det(res U n n U
G ω)(ψl χ1 )) = Ψ ((res G (Detω))(β))
n G
= Ψ ((Detω)(ind U β))
ON THE “MAIN CONJECTURE” OF EQUIVARIANT IWASAWA THEORY 1041
is torsion. This holds because U G l-elementary implies that β = res U G β where
β = inflGab β , hence ind U β = β ind U 1 and ind U 1 = i=1 αi with αi the irre-
G G G G l
G
ducible characters of G having res U G αi = 1; now αi = inflGab αi so
l l
(Detω)(ind G U β) = (Detω)( i=1 β αi ) =
G
i=1 (Detω)(inflGab (β αi ))
l Gab
= i=1 Det(deflG ω)(β αi )
is torsion. Thus (Det(res UG ω))(χ1 ) is torsion for all such χ1 and U ≥ A of index l
in G.
Now if χ is one of the finitely many irreducible characters of G with Γ ⊆
ker(χ), then (by Clifford theory) either χ = ind G U χ1 with U ≥ A of index l
when (Detω)(χ) = (res U G (Detω))(χ 1 ) is torsion or χ = inflG
Gab α when (Detω)(χ) =
Gab
Det(deflG ω)(α) is again torsion. Every irreducible character of G has the form
χ ⊗ ρ with such a χ and ρ of type W 18 ; hence (Detω)(χ ⊗ ρ) = ρ ((Detω)(χ))
is torsion of order at most that of (Detω)(χ). Thus Detω is a torsion element in
HOM(Rl (G), (Λ∧ Γk )× ) and so Tr(v) = Tr(Lω) = L(Detω) = 0 implies v = 0.
Remark 6.1. When G is abelian pro-l, Proposition 5.1 of [RW6] gives a description
of the kernel of L on (Λ∧ G)× . This can be (and originally was) used to prove the
lemma for pro-l groups. The present proof for l-elementary groups is shorter than
extending that proposition from Λ∧ to Λβ∧ (for the notation compare the proof of
Lemma 7.1).
We next state
ab
Lemma 6.3 (Uniqueness Principle). If ξ ∈ T (Λ∧ G) has deflU U
U ResG ξ = tU ab
for all subgroups U of G containing a fixed abelian normal open subgroup A of G,
then ξ = tG . In particular, tG ∈ T (Λ∧ G).
Proof. This holds because ln tG is integral for large enough natural n (see [RW4,
ab
Proposition 9]). Setting v = ln tG − ln ξ, we see that deflU U ResG v = l tU ab −
U n
Proof. Now we start the proof of the THEOREM. If it were false, there would exist
a counterexample K/k, with tK/k not in T (Λ∧ G), for which the Galois group G
would have commutator subgroup [G, G] of minimal order; among these groups G
there would be one with centre Z(G) of minimal index [G : Z(G)].
Since [G, G] = 1, by [RW3, Theorem 9], and [G, G] is an l-group, as G is l-
elementary, we may choose a central subgroup C ≤ [G, G] of order l in G and then
a maximal abelian normal subgroup A of G, necessarily containing C and z. We
also fix a central open Γ Zl inside A. Since |[G, G]| < |[G, G]|, Theorem 4 guar-
antees the existence of
(6.1) ξ ∈ T (Λ∧ G) with deflG A
G ξ = tG and ResG ξ = tA ,
where, as before, denotes the canonical surjection G G/C. To defeat the coun-
terexample G it suffices, by the Uniqueness Principle (Lemma 6.3), to find such a
18 I.e., res H
G ρ = 1.
1042 JÜRGEN RITTER AND ALFRED WEISS
ab
ξ so that deflUU ResG ξ = tU ab for all subgroups U ≥ A of G. Observe that this al-
U
ab
U ab U/C
ready holds for U with [U, U ] ≥ C: for then deflU U
U ResG ξ = deflU/C deflU ResU
Gξ =
ab ab
deflUU
U
ResG deflG U U
G ξ = deflU ResG tG = tU ab .
On the other hand, for U ≥ A with [U, U ] ≥ C, hence C ∩ [U, U ] = 1, then
|[U, U ]| < |[G, G]| implies19 tU ∈ T (Λ∧ U ), by our hypothesis on G, permitting us
to define
ξU = ResUG ξ − tU ∈ T (Λ∧ U )
and to define the support of ξ by
supp(ξ) = {U ≥ A : C ∩ [U, U ] = 1 and ξU = 0}.
Note that supp(ξ) = ∅ for our counterexample G by the Uniqueness Principle
(Lemma 6.3).
To investigate U ∈ supp(ξ), we state
Claim 6.A. (a) If A ≤ V ≤ U and C ∩ [U, U ] = 1, then ResVU ξU = ξV and ξVg = ξV g
for g ∈ G.
(b) G acts on supp(ξ) by conjugation.
(c) A ∈
/ supp(ξ).
Proof. Recall that tU = ResU
G tG . Now (a) results from
and g g
ξVg = ResVG (ξ − tG )g = ResVGg (ξ g − tgG ) = ResVG (ξ − tG ) = ξV g ,
by (3) of Lemma 5.1, which at the same time implies (b); (c) follows from (6.1) and
the definition of ‘supp’.
Moreover, we let, as in Lemma 6.1, N = NG (U ) be the normalizer of U in G.
Define cU and cab
U by the exact sequences
0 → cU → Λ∧ (N/[U, U ]) → Λ∧ (U/C[U, U ]) → 0,
0 → cab
U → Λ∧ (U ) → Λ∧ (U/C[U, U ]) → 0.
ab
ab
U ξU ∈ trN/U (cU ).
Claim 6.B. If U ∈ supp(ξ), then deflU ab
Proof. We first note that tN/[U,U] is in T (Λ∧ (N/[U, U ])): for the commutator sub-
group [N, N ]/[U, U ] of N/[U, U ] has smaller order than [G, G] unless [U, U ] = 1 and
[N, N ] = [G, G], in which case A ∈ / supp(ξ) implies N < G, because A is maximal
abelian normal in G; but then Z(G) ≤ N implies [N/[U, U ] : Z(N/[U, U ])] ≤ [N :
Z(N )] ≤ [N : Z(G)] < [G : Z(G)], contrary to the minimality hypothesis on G.
N/[U,U]
Writing deflN G ξ = tN/[U,U] + zU , with zU ∈ T (Λ∧ (N/[U, U ])), and
ResN
τ = τN/[U,U] , we consider the commutative diagram
T (Λ∧ N ) T (Λ∧ N )
ab
= deflU U N
U ResN (ResG ξ)
ab N/[U,U]
= ResU
N/[U,U] deflN (ResN
G ξ)
ab ab
= ResU U
N/[U,U] (zU + tN/[U,U] ) = ResN/[U,U] zU + tU ab
ab ab ab
U ξU = ResN/[U,U] zU ∈ ResN/[U,U] τ (cU ). Combining with the previous
implies deflU U U
Now continuing with the proof, it follows that the THEOREM holds if we can
modify ξ, subject to (6.1) holding, so that supp(ξ) is empty. Since this is not
possible for our G, by hypothesis, there must exist a ξ for which supp(ξ) has
minimal cardinality = 0.
Since supp(ξ) is non-empty, it contains a U with minimal [U : A]. By Lemma
6.1 we may write
ab
deflU y∈Y1 , 1=c∈C α(y, c)trN/U (ỹ(c̃ − 1))
(6.2) U ξU =
Claim 6.C. Assume that U/A is cyclic. Set, in the notation of (6.3),
ξ = α(y, c)τG (y (c − 1)) in T (Λ∧ G),
y∈Y1 , yA =U/A, 1=c∈C
ab
with preimages y ∈ U of ỹ under deflU
U . Then
ab ab
U
(i) deflU U
U ResG ξ = deflU ξU , and
1
(ii) if A ≤ U1 ≤ G, then ResU
G ξ = 0 implies ∃ g ∈ G : U ≤ U1 .
g
Proof. Recall that the y in the ξ -sum have yA = U/A. Applying (2) of Lemma
5.1 gives
ResUGξ = α(y, c) τ̇U ((y )x (c − 1)).
y∈Y1 , yA =U/A, 1=c x∈G/U
x
Note that (y ) ∈ U implies (yA) ∈ U/A; hence (U/A)x = U/A, i.e., x ∈ N . Now
x
we have
ResU
Gξ = α(y, c) τU ((y )x (c − 1));
y,c x∈N/U
ab
hence applying deflU
gives (i).
U
For (ii) note that y still has yA = U/A, but we now apply (2) of Lemma 5.1
ResG τG (y (c − 1)) in this sum must be = 0,
U1
with U replaced by U1 . Some term
by hypothesis; but this term is τ̇
x1 ∈G/U U1 ((y ) x1
(c − 1)), so we must have a
x1
non-zero term here, i.e., (y ) ∈ U1 for some x1 . Now (U/A)x1 ⊆ U1 /A implies
U x1 ⊆ U1 .
def
We apply Claim 6.C and set ξ = ξ −ξ . Then ResA
G ξ = tA , by (ii) with U1 = A;
moreover, due to the appearance of the elements 1 = c ∈ C in ξ , deflG ξ = tG/C ;
G/C
7. l-elementary groups
Recall that G = G(K/k) = z × G[l] is l-elementary.
Lemma 7.1. (i) The logarithm L : K1 (Λ∧ G) → T (Q∧ G) of diagram (1.D1) has
image in T (Λ∧ G).
ab
(ii) Let a be the kernel of deflG
G : Λ∧ G Λ∧ Gab . Then the commutative
diagram
1 + a (Λ∧ G)× (Λ∧ Gab )×
L↓ Lab ↓
Proof. For (i), abbreviate G[l] as U . Each Ql c -irreducible character β of z induces
a Zl -algebra homomorphism Zl [z] → Zl c with image Zl [β], hence surjective ring
homomorphisms
def
Ql [z] Ql (β), Λ∧ G Zl [β] ⊗Zl Λ∧ U = Λβ∧ U, Q∧ G Qβ∧ U.
Applying the functors K1 and T gives the southeast and southwest arrows of the
diagram (with Γ, RG, RU short for Γk , Rl (G), Rl (U ), respectively):
K1 (Λ∧ G) −→ T (Q∧ G)
K1 (Λβ
∧ U) → T (Qβ
∧ U)
| ↓ ↓ |
↓ ↓
HOM(β) (RU,(Λc∧ Γ)× ) → Hom(β) (RU,Qc∧ Γ)
HOM(RG,(Λc∧ Γ)× ) −→ Hom∗ (RG,Qc∧ Γ)
with large square from diagram (1.D1) of §1 and small square [RW3, 2. of Proposi-
tion 11] with unramified coefficients Zl [β], which are abbreviated by the superscript
β. The northwest and northeast arrows f → f β are defined by f β () = f (β ⊗ ).
To see that the left quadrilateral commutes20 let H = H ∩ U (recalling that
H is the kernel of G → Γk ), hence H = z × H , and let β(x) denote the image
of x ∈ Λ∧ G in Λβ∧ U . We must check that (Detx)β () = (Detβ(x))(), i.e.,21
detQc∧ Γk (x | Vβ⊗ ) = det(β(x) | V ). Here,
V = HomQl c [H ] (V , Ql c ⊗Ql (β) Qβ∧ U ) = HomQl c [H ] (V , Qc∧ U )
and
Vβ⊗ = HomQl c [H] (Vβ⊗ , Qc∧ G)
= HomQl c [ z ]⊗Ql c Ql c [H ] (Vβ ⊗Ql c V , Ql c [z] ⊗Ql c Qc∧ U )
= HomQl c [H ] (Vβ ⊗Ql c V , Vβ ⊗Ql c Qc∧ U ).
Then h → 1 ⊗ h is an isomorphism V → Vβ⊗ of vector spaces over Qc∧ Γk and
one checks that (1 ⊗ h)x = 1 ⊗ h · β(x).
The same argument, with T, Tr rather than K1 , Det, yields the commutativity of
the right quadrilateral, and the commutativity of the bottom quadrilateral follows
from the formula [RW3, p. 37] for L by ψl (β) = β Fr . The diagram now implies that
the top quadrilateral commutes.
Recall that, [RW3, Proposition 11], the logarithm Lβ : K1 (Λβ∧ U ) → T (Qβ∧ U ) is
integral for all β. It thus suffices to show that if x ∈ T (Q∧ G) has image in T (Λβ∧ U )
under the southwest arrow for every β, then x ∈ T (Λ∧ G).
Letting β run through a set of representatives of the G(Ql c /Ql )-action on the
Ql -irreducible characters of z, we get an isomorphism Zl [z] → β Zl [β]. This
c
induces isomorphisms T (Λ∧ G) → β T (Λβ∧ U ) and T (Q∧ G) → β T (Qβ∧ U ). The
first of these provides an x ∈ T (Λ∧ G) with the same images as x for all β, and the
second gives x = x ∈ T (Λ∧ G).
We now prove (ii). The exact sequence defining a gives the top row since a ⊆
rad(Λ∧ G), as [G, G] is an l-group. The bottom row is exact by (ii) of Lemma 5.2. To
20 This can also be obtained from [RW3, Theorem 1].
21 Compare also [RW2, Proposition 6] and [RW3, Lemma 2].
1046 JÜRGEN RITTER AND ALFRED WEISS
ab
see the vertical surjectivity, write u for the kernel of deflU
U : Λ∧ U Λ∧ U ab (with
β β
U = G[l]); also write uβ for the kernel of Λ∧ U Λ∧ U ab (with β as before). Then
the map 1 + uβ → τU (uβ ) induced by Lβ : (Λβ∧ U )× → T (Λβ∧ U ) is surjective [RW3,
2b. of Proposition 11]. Identifying Λ∧ G and β Λβ∧ U as in the last paragraph of
the proof of (i) (also for the abelianizations) and assembling our asserted diagram
in terms of the β-decomposition, noting that the commutativity of the square below
follows from that of the top quadrangle in (i), we deduce that 1 + a → τG (a) is also
surjective:
β
(Λ∧ G)× → (Λβ∧ U )×
L↓ Lβ ↓
β
T (Λ∧ G) → T (Λβ∧ U ).
def
where χ ∈ Rl (G ), ψl denotes the l-th Adams operation, χ = ψl (ind G G χ ) −
ind G (ψl χ ), and k = K G . The diagram below, and so diagram (1.D2), com-
G
mutes:
HOM(Rl (G),(Λc∧ Γk )× ) Hom∗ (Rl (G),Qc∧ Γk )
Det L Tr
K1 (Λ∧ G) → → ← T (Q∧ G)
G ↓
res G G ↓
res G G ↓
ResG G ↓
ResG
HOM(Rl (G ),(Λc∧ Γk )× ) Hom (Rl (G ),Qc∧ Γk )
∗
T (Q∧ G ).
Det L Tr
K1 (Λ∧ G ) → → ←
Moreover, ResG G : T (Q∧ G) → T (Q∧ G ) takes T (Λ∧ G) to T (Λ∧ G ) and is transi-
G G G
tive; i.e., ResG = ResG ResG whenever z ≤ G ≤ G ≤ G.
⊂ G for a suitable power lr0 of l; thus
r0
Proof. We first observe that Gl
ψlr0 −1 χ = ψlr0 −1 (ψl (ind G
G χ ) − ind G (ψl χ )) = 0,
G
for all χ ∈ Rl (G ) (compare [RW8, p. 119]): this again follows from
(7.1) χ(g) = t with m(gt )=1 χ (g lt ),
with m(g) = min{r ≥ 0 : g l ∈ G } and G = ˙ t tG . From this, ResG
r
G is well-defined
and the middle square commutes just as in [RW8, p. 120].
The map ResGG on the very right is defined by transporting
∗ ∗
G : Hom (Rl (G), Q∧ Γk ) → Hom (Rl (G ), Q∧ Γk )
ResG c c
We close this section with adjusting the arguments in §1 for the proof of Propo-
sition 1.3 for pro-l groups to l-elementary groups G.
Lemma 7.3. (i) If tK/k ∈ T (Λ∧ G), then there is a unique torsion w ∈ HOM(Rl (G),
ab
(Λc∧ Γk )× ) with wLK/k ∈ DetK1 (Λ∧ G) and deflGG w = 1.
(ii) Moreover, if G has an abelian subgroup A of index l, then
w = 1 ⇐⇒ verA
Gab λK [G,G] /k ≡ λK/k mod trQ (Λ∧ G )
where k = K A and Q = G/A.
Proof. For (i), using the diagram in (ii) of Lemma 7.1 to replace the one in the
proof of [RW5, Proposition 2.2], there exists a y ∈ (Λ∧ G)× so L(y) = tK/k and
ab
deflG G w = 1. Following the proof of [RW5, Proposition 2.4], one defines w by
ab
wLK/k = Dety and checks that deflG G w = 1 and L(w) = 0. This implies w
is torsion by the indicated argument from [RW3, p. 46] by observing that, while
ψln χ is only a character β of G/G[l] for large n, β = inflG Gab β still implies that
Gab
w(β) = (deflG w)(β ) = 1. The argument for the uniqueness of w still works
because [RW3, Lemma 12] is already proved for Λβ∧ (G[l]), in the notation of Lemma
7.1.
More precisely, in the notation of the proof of (ii) of Lemma 7.1, let x ∈ 1 + a
have Detx torsion. By the commutativity of the left quadrangle in the proof of (i) of
Lemma 7.1, β(x) ∈ 1 + uβ has Detβ(x) torsion, hence we have Detβ(x) = 1, and so
it suffices to observe that HOM(Rl (G), (Λc∧ Γk )× ) → β HOMβ (Rl (U ), (Λc∧ Γk )× )
is injective.
To verify (ii), we can follow the proof of [RW5, equivalence of (1) and (2)
in Proposition 3.2], except that we still need to show that the only torsion unit
e ∈ Λ∧ A congruent to 1 mod trQ (Λ∧ A) is 1, even when G is l-elementary. Decom-
posing the torsion subgroup H of G as H = z × H , we have A = Γ × H with
Γ Γk and Λ∧ A = β (Λβ∧ Γ)[H ]. Now β(e) ≡ 1 mod trQ ((Λβ∧ Γ)[H ]). Hence
β(e) = 1 by Higman’s theorem for (Λβ∧ Γ)[H ]; see [RW3, p. 47]. This holds for all
β; hence e = 1.
Appendix: Deflating 0
The proof of [RW2, Proposition 12(a)] refers to [RWt, Proposition 4.8] which,
however, requires Leopoldt’s conjecture (see [RWt, Lemma 3.4]). We recall the
statement made in [RW2] (suppressing the index ∞ on K and G as well as the ˜
on 0S ):
If N is a finite normal subgroup of G with factor group G and
fixed field K = K N , then deflGG : K0 T (ΛG) → K0 T (ΛG) takes
0S = 0K/k,S to 0S = 0K/k,S .
def
The refinement 0K/k,S of the Iwasawa module X = XK/k,S = G(M/K), with
M the maximal abelian S-ramified l-extension of K, is described in [RW1, §1].
Here is a direct argument for the above claim.
Let M be the maximal abelian S-ramified l-extension of K; hence G(M /K) is the
biggest abelian pro-l quotient of G(M/K). Consider the diagram below, where X̃
is the pushout of i and ver and where the transfer G(M/K) → X N factors through
1048 JÜRGEN RITTER AND ALFRED WEISS
G(M /K) since X N is abelian pro-l. The bottom row is called the deflation of the
top one in [RWt, §3.2].
G(M/K) G(M/k) G(K/k) = G
i
X = G(M /K) G(M /K) G
ver ↓ ↓
XN X̃ G.
By the Appendix 4.A analogue of Lemma 3.2 in [RWt] the translation functor turns
the bottom two rows into22
X Y ΔG
() ↓ ↓
XN YN (ΔG)N
by replacing the bottom one by the equivalent extension given by the ΛG-analogue
of Lemma 3.3 [RWt]. Here we should note that Y has projective dimension ≤ 1 over
ΛG [RW1, Theorem 1]; hence Y has projective dimension ≤ 1 over Zl [N ] [RW1, 2.
of proof of Proposition 4], which does not need M to be finite.
Suppose that we know that ver : G(M /K) → X N is an isomorphism, hence that
the extensions in () are equivalent. If we use, [RW1, §1], X Y → ΛG Zl to
compute 0, then taking N -coinvariants computes the analogous deflG G
(0) for the
bottom row of () (cf. the analogy with [RWt, Lemma 4B.1, p. 41]). By diagram
(), the same procedure for the top row computes 0. Thus deflG G (0) = 0.
ver
Concerning G(M /K) −→ X N , let L be the maximal S-ramified Galois extension
of k or, equivalently, of k∞ . Denote the corresponding Galois groups by G and
H, respectively; so H G Γk is exact. Moreover, set V = G(L/K) and
U = G(L/K), whence V U N .
Assume that we already know scdl (H) = 2. Then we proceed as follows. As
U is open in H, it follows that also scdl (U ) = 2 by [NSW, pp. 139–140]. The
proof of [ (i) =⇒ (ii) ] of [NSW, Theorem 3.6.4, p. 160] gives the isomorphism
cor ver
H 2 (V, Z)(l)N −→ H 2 (U, Z)(l) (see [NSW, 3.3.8, p. 142]) and so (U ab )l −→ (V ab )N
l .
ab ab
Since (U )l = G(M /K) = X, (V )l = G(M/K) = X finishes the proof.
Hence it remains to show scdl (H) = 2. Now, scdl (H) ≤ 2 is a consequence of
the weak Leopoldt conjecture (see [NSW, 10.3.26, p. 549]) and then scdl (H) = 2
results from the remark following it, of which we add a proof: Assume scdl (H) ≤ 1.
Then cdl (H) ≤ 1 and cdl (Γk ) = 1; hence 2 = cdl (G) ≤ cdl (Γk ) + cdl (H) implies
cdl (H) = 1. Note that cdl (G) = 2 by [NSW, 10.9.3, p. 587]. Denoting a Sylow-l
subgroup of H by Hl , we have cdl (Hl ) = 1 = scdl (Hl ) [NSW, 3.3.6, p. 141] and thus
H 2 (Hl , Z)(l) = 0 [NSW, 3.3.4, p. 139]. Hence H 1 (Hl , Ql /Zl ) = 0 and so Hl = 1,
contradicting cdl (Hl ) = 1.
References
[Ca] Cassou-Noguès, P., Valeurs aux entiers négatifs des fonctions zêta et fonctions zêta p-
adiques. Invent. Math. 51 (1979), 29–59. MR524276 (80h:12009b)
[CR] Curtis, C. W. and Reiner, I., Methods of Representation Theory, vol. 2. John Wiley &
Sons (1987). MR892316 (88f:20002)
[DR] Deligne, P., and Ribet, K., Values of abelian L-functions at negative integers over totally
real fields. Invent. Math. 59 (1980), 227–286. MR579702 (81m:12019)
[FK] Fukaya, T., and Kato, K., A formulation of conjectures on p-adic zeta functions in non-
commutative Iwasawa theory. Proceedings of the St. Petersburg Mathematical Society,
vol. XII (N. N. Uraltseva, ed.), AMS Translations – Series 2, 219 (2006), 1–86. MR2276851
(2007k:11200)
[Gr] Greenberg, R., On p-adic Artin L-functions. Nagoya Math. J. 89 (1983), 77–87.
MR692344 (85b:11104)
[Ha1] Hara, T. Iwasawa theory of totally real fields for certain non-commutative p-extensions.
Journal of Number Theory 130 (2010), 1068–1097. MR2600423 (2011c:11166)
[Ha2] , Inductive construction of the p-adic zeta functions for non-commutative p-
extensions of totally real fields with exponent p. arXiv:0908.2178v1 [math.NT].
[HIO] Hawkes, T., Isaacs, I. M., and Özaydin, M., On the Möbius function of a finite group.
Rocky Mountain J. of Mathematics 19 (1989), 1003–1033. MR1039540 (90k:20046)
[Kk] Kakde, M., Proof of the main conjecture of noncommutative Iwasawa theory for totally
real number fields in certain cases. J. Algebraic Geometry, posted on April 5, 2011,
PII S 1056-3911(2011)00539-0 (to appear in print).
[Kt1] Kato, K., Iwasawa theory and generalizations. Proc. ICM, Madrid, Spain, 2006; European
Math. Soc. (2007), 335–357. MR2334196 (2008e:11133)
[Kt2] , Iwasawa theory of totally real fields for Galois extensions of Heisenberg type.
Preprint (“Very preliminary version”, 2007).
[La] Lau, I., Algebraic contributions to equivariant Iwasawa theory. Ph.D. thesis, Universität
Augsburg (2010).
[NSW] Neukirch, J., Schmidt, A., and Wingberg, K., Cohomology of Number Fields. Springer
Grundlehren der math. Wiss. 323 (2000). MR1737196 (2000j:11168)
[RWt] Ritter, J., and Weiss, A., The Lifted Root Number Conjecture and Iwasawa theory. Mem-
oirs of the AMS 157/748 (2002). MR1894887 (2003d:11164)
[RW1] , Towards equivariant Iwasawa theory. Manuscripta Math. 109 (2002), 131–146.
MR1935024 (2003i:11161)
[RW2] , Towards equivariant Iwasawa theory, II. Indag. Mathemat. 15 (2004), 549–572.
MR2114937 (2006d:11132)
[RW3] , Towards equivariant Iwasawa theory, III. Math. Ann. 336 (2006), 27–49.
MR2242618 (2007d:11123)
[RW4] , Towards equivariant Iwasawa theory, IV. Homology, Homotopy and Applications
7 (2005), 155–171. MR2205173 (2006j:11151)
[RW5] , Non-abelian pseudomeasures and congruences between abelian Iwasawa L-
functions. Pure and Applied Math. Quarterly 4 (2008), 1085–1106. MR2441694
(2010b:11149)
[RW6] , The integral logarithm in Iwasawa theory: An exercise. Journal de Théorie des
Nombres de Bordeaux 22 (2010), 197–207. MR2675880 (2011f:11144)
[RW7] , Congruences between abelian pseudomeasures. Math. Res. Lett. 15 (2008), 715–
725. MR2424908 (2009m:11183)
[RW8] , Equivariant Iwasawa Theory: An Example. Documenta Math. 13 (2008), 117–
129. MR2420909 (2009h:11183)
[RoW] Roblot, X.-F., and Weiss, A., Numerical evidence toward a 2-adic equivariant “main
conjecture”. Experimental Mathematics 20 (2011), 169-176.
[Se] Serre, J.-P., Sur le résidu de la fonction zêta p-adique d’un corps de nombres., C. R. Acad.
Sci. Paris 287 (1978), Série A, 183–188. MR0506177 (58:22024)
[Wi] Wiles, A., The Iwasawa conjecture for totally real fields. Annals of Math. 131 (1990),
493–540. MR1053488 (91i:11163)
1050 JÜRGEN RITTER AND ALFRED WEISS