6fbd3ac Δ f6bf760 Δ 56bc728 Δ b1f9bc4 Δ b34550f Δ 3ff9228 Δ 0fce9d8
editors: nobody









hypersetup{pdftitle={Modularity of elliptic curves over imaginary quadratic fields},pdfauthor={OpenAI}}

















































# Modularity of elliptic curves over imaginary quadratic fields


maketitle

Abstract: 
We prove that every elliptic curve over an imaginary quadratic field is
modular: it is realized by an automorphic representation of GL_2 of the
field with matching local parameters at every place and matching
L-function. The method is a geometric form of Wiles's prime switch. A
single dihedral family of cyclic covers of a quadratic twist of the curve
produces four two-dimensional Galois representations, pairwise congruent
modulo the primes 3 and a large prime p, with an elliptic curve at one
end. Automorphy enters at that end through known results and is carried to
the original curve by three modularity-lifting steps, then descended from a
cyclotomic extension to the field itself. The complex-multiplication case is
treated separately by automorphic induction.


tableofcontents

# Introduction

## The theorem

Let K be an imaginary quadratic number field and E/K an elliptic curve.
The modularity conjecture for E asks for an automorphic representation of
GL_2(A_K) whose attached Galois representations are the cohomology of
E. Concretely, at a place of good reduction this means that the Hecke
polynomial recovers the point count of the reduction; at a bad place it
means that the full local parameter, including inertia and monodromy, is
recovered; at the single complex place it means a fixed parameter dictated by
the Hodge decomposition. We prove that such a representation always exists.

Throughout, H^1(X)=H^1_{et}(X_{overlineoverline L}L,Q_l) for a variety over a
number field L, with coefficients made explicit when they matter, and
G_L=Gal(overlineoverline L/L).L/L). The cyclotomic character is eps_l, normalized
to have Hodge--Tate weight -1, so an elliptic curve has labeled
Hodge--Tate weights 0,1 on its H^1 and determinant eps_l^{-1}. For a
regular algebraic cuspidal representation pi of GL_2(A_L) and an
isomorphism iota:overlineQ_ltoC, we write r_iota(pi) for the
associated semisimple l-adic representation. The normalization of local
Langlands used on the cohomological side is
rec^T_{L_v}(pi_v)=rec_{L_v}(pi_v ⊗ |det|_v^{-1/2}); see
Section~sec:conv.


Let K be an imaginary quadratic field and E/K an elliptic curve. There is
an isobaric automorphic representation pi_E of GL_2(A_K) with trivial
central character and weight zero such that for every prime l and every
isomorphism iota:overlineQ_ltoC,
[
 r_iota(pi_E)~= H^1_{et}(E_{overlineoverline K}K,overlineQ_l).
]
At every finite place vnmid l the Frobenius-semisimplified
Weil--Deligne parameter of the right side equals
rec_{K_v}(pi_{E,v} ⊗ |det|_v^{-1/2}), and the unitary parameter at
the complex place is z|->diag(z/|z|,overlineoverline z/|z|).z/|z|). In particular
every local factor agrees, so L(E/K,s)=L(pi_E,s-1/2). If E has no
complex multiplication then pi_E is cuspidal; if E has complex
multiplication it is an automorphic induction of a Hecke character, or an
isobaric sum when the CM field equals K.


The phrase ``with matching local parameters at every place'' together with
the equality of L-functions is exactly what modularity means: the Hasse
--Weil L-function of the curve, an arithmetic object counting points over
all finite fields, is the L-function of an automorphic form, an analytic
object with a functional equation and a spectral meaning.

## Why it is hard

Over Q, the modularity of every elliptic curve was completed by
Wiles--Taylor and Breuil--Conrad--Diamond--Taylor. A central tool is
modularity lifting. One knows automorphy of a residual Galois
representation overlineoverline r,r, and one proves that every lift r of
overlineoverline rr with the right local behavior is also automorphic; applied to
the Galois representation of E this gives modularity of E. Lifting
theorems require residual hypotheses, and Wiles's 3--5 switch handles
curves for which the residual representation of E itself is unsuitable, by
passing through a resolvent auxiliary curve whose own modularity is already
known.

Over an imaginary quadratic field three new difficulties appear.
[label=(arabic*)]
 The relevant cohomology of Bianchi modular forms sits in more than one
degree, so constructing Galois representations and running a patching or
deformation argument both require extra work. Harris--Soudry--Taylor,
Taylor, Harris--Lan--Taylor--Thorne, Scholze, and Calegari--Geraghty supply
these constructions, and potential modularity over CM fields is available
(Allen--Calegari--Caraiani--Gee--Helm--Le Hung--Newton--Scholze--Taylor
--Thorne). But potential modularity may enlarge the field, whereas we must
stay over K.
 The known lifting theorem in the small characteristics available for
starters --- Caraiani--Newton --- imposes residual conditions and local
conditions that E itself need not satisfy. Their theorem already gives
modularity for a positive proportion of curves, and for all curves when
X_0(15)(K) has rank zero, but not for all curves.
 There is no obvious resolvent auxiliary curve directly attached to
E in the way Wiles's auxiliary curve is attached over Q.


Our contribution is to remove the restriction on the curve and on the
field. We do not build a single auxiliary curve; we build a whole family of
covers, so that the four representations participating in the modularity
chain are simultaneous eigenspaces of one geometric object.

## The geometric prime switch

Fix a non-CM elliptic curve E/K and write it as y^2=h(x) with origin the
point at infinity O. Fix an odd integer n=3p, where p>100 is a prime
chosen (via Serre's open image theorem) so that the image of G_K on E[p]
contains SL_2(F_p). Put F=K(mu_n). For t in a nonempty Zariski-open
subset UsubsetA^1_K we build a genus-n cyclic cover
[
 C_n(t)→ B_t
]
of the quadratic twist B_t:h(t)y^2=h(x) of E, branched at two points,
carrying an involution w that inverts the deck rotations. For each
character theta of the rotation group mu_n, the theta-eigenspace
[
 V_theta=H^1(C_n(t))_theta
]
in integral cohomology is a rank-two compatible system with determinant
eps_l^{-1} and labeled Hodge--Tate weights 0,1. Choose characters
eta of order 3 and psi of order p, and put chi=etapsi of
order 3p. The point of the construction is the chain of congruences

 {V_eta}_{H^1(A)}
  equiv pmod{p} V_chi
  equiv pmod{3} V_psi
  equiv pmod{p} {V_1}_{H^1(B_t)},

each congruence being an equality of semisimplified residual Galois
representations. The order-p part of a character becomes trivial modulo
p, and the order-3 part becomes trivial modulo 3, so the primary
factors eta,psi,chi collapse onto their neighbours. The order-three
system V_eta is H^1 of an elliptic curve A/F, while V_1 is H^1 of
the twist B_t.

Automorphy enters at the left. The prime 5 is used only to certify that
the auxiliary elliptic curve A is modular: A is chosen (through the same
one-parameter family) with a large mod-5 image and non-CM, so the
elliptic-curve starting theorem of Caraiani--Newton applies. Automorphy then
propagates rightwards along (intro:chain) by three applications of the
Caraiani--Newton lifting theorem, at primes p,3,p respectively. At the end
we know H^1(B_t) is automorphic; because B_t is a quadratic twist of
E, twisting by the corresponding Hecke character gives automorphy of
H^1(E). Finally we descend the automorphic form from the cyclotomic field
F to the original field K, and handle the CM case separately.

Two geometric specializations of the cover family make this work.
 At one degeneration the Jacobian of the cover has compact-type limit
E^n. Because E^n is a product of copies of E, its cohomology --- and
hence the cohomology of every eigenspace --- has its reduction type (good
ordinary, good supersingular, or totally multiplicative) read off from E.
A local-constancy argument over the local field of coefficients transfers
this to all nearby fibers at once. This gives simultaneous local conditions
on all four systems for one choice of t.
 At the other degeneration explicit vanishing cycles cut out opposite
unipotent elements on each nontrivial eigenspace. These unipotents give
absolute irreducibility of the residual representations and produce
suitable Frobenius elements, without any assumption that E and its
conjugate are independent. They are also what forces the auxiliary curve to
be non-CM.

A Hilbert-irreducibility argument over the restriction of scalars
Res_{K/Q}U then finds a single rational point t satisfying prescribed
local conditions together with the residual hypotheses for both conjugates
of K at once. This gives decomposed genericity at every place above a
split rational prime.

We emphasize what is genuinely new here relative to Caraiani--Newton: the
character pieces appear in one integral cohomology lattice, an elliptic curve
occurs at each end of the chain, and the local-constancy and
two-conjugate-specialization arguments are what make each lifting step
applicable to a prescribed curve over a prescribed imaginary quadratic
field.

## Related work

The arithmetic groups of imaginary quadratic fields and their action on
hyperbolic three-space go back to Bianchi, with presentations and cohomology
developed by Swan and Harder. The link with elliptic curves is part of the
Langlands program; Langlands's problem list discusses automorphy of local
representations of elliptic curves over global fields. Computations
relating elliptic curves and automorphic forms over imaginary quadratic
fields are due to Grunewald--Helling--Mennicke, Cremona, and
Cremona--Whitley.

Over Q, Wiles and Taylor--Wiles proved modularity of semistable curves and
Breuil--Conrad--Diamond--Taylor of all curves. Wiles's 3--5 switch
introduced the idea that a residual representation failing the initial
hypotheses can be replaced by one coming from an auxiliary elliptic curve
with prescribed torsion and local behavior. For imaginary quadratic fields,
Harris--Soudry--Taylor and Taylor constructed Galois representations
attached to Bianchi forms in important cases; Berger--Harcos obtained
Frobenius compatibility for cyclotomic central character away from finitely
many primes; Harris--Lan--Taylor--Thorne and Scholze constructed the general
representations; Calegari--Geraghty developed patching beyond the
Taylor--Wiles method. Potential automorphy over CM fields is due to
Allen--Calegari--Caraiani--Gee--Helm--Le Hung--Newton--Scholze--Taylor--Thorne
and, for genus-one curves over quadratic extensions of totally real fields,
to Boxer--Calegari--Gee--Pilloni. Allen--Khare--Thorne gave residual
modularity results in small characteristic and deduced modularity for a
positive proportion of curves over each CM field not containing zeta_5.
Caraiani--Newton gave a lifting theorem allowing small coefficient primes and
ramified CM fields, and deduced modularity of all curves over imaginary
quadratic K with X_0(15)(K) of rank zero, of 100% of curves ordered by
height over each such K not containing zeta_5, and under residual-image
hypotheses in characteristic 3 or 5. Our Theorem~thm:main removes
the last restrictions; the Caraiani--Newton lifting theorem and modularity
criterion remain essential inputs.

The geometric ingredients have antecedents: cyclic covers of elliptic
curves with two branch points and dihedral symmetry occur in real
multiplication constructions of Tautz--Top--Verberkmoes, in Ellenberg's work
on endomorphism algebras of Jacobians, and in Mestre's constructions. Darmon
studied congruences removing prime-power parts of monodromy and a conditional
modularity argument by successive lifts for rigid local systems. The
difference operator 2-zeta-zeta^{-1} in the monodromy computation is the
same one appearing in rank-two monodromy studied by Darmon--Mestre. Taylor's
use of auxiliary abelian varieties with real multiplication is a
higher-dimensional form of the prime switch.

## Organization

Section~sec:conv fixes conventions. Sections~sec:geom and
sec:mono build the cover family and compute its integral homology and
monodromy. Section~sec:sys constructs the four rank-two systems and
their congruences. Section~sec:inputs states the automorphy inputs in
the precise form used and reduces the remaining work to conditions on t.
Section~sec:local produces the local reduction conditions. Section
sec:res produces a single t satisfying all residual and local
conditions at once. Section~sec:auto runs the three lifts, descends to
K, and treats the CM case.

# Conventions

We write eps_l for the cyclotomic character with Hodge--Tate weight -1.
A bar over a characteristic-zero representation denotes the semisimplification
of the reduction of a chosen stable lattice; this semisimplification is
independent of the lattice. An elliptic curve thus has labeled Hodge--Tate
weights 0,1 on H^1 and determinant eps_l^{-1}.

For a regular algebraic cuspidal representation pi of GL_2(A_L) over a
CM field and iota:overlineQ_ltoC, the representation r_iota(pi) is
semisimple. We use the Tate-normalized local comparison

 rec^T_{L_v}(pi_v):=rec_{L_v}(pi_v ⊗ |det|_v^{-1/2}),

with geometric-Frobenius reciprocity. The algebraic weight coming from
trivial coefficients is called weight zero. At a complex place the generic
weight-zero parameter, in unitary normalization, is

 z⟼diag(z/|z|,overlineoverline z/|z|),z/|z|), |z|=(zoverlineoverline z)^z)^{1/2}.

Its Tate-normalized parameter has Hodge types (1,0) and (0,1). If the
finite parameters realize an elliptic curve E, then

 L(E/L,s)=L(pi,s-1/2).


For a local field, a Galois representation is potentially crystalline
if it becomes crystalline after a finite extension, and is potentially
ordinary of weight zero if after a finite extension it has an invariant line
on which inertia acts by an unramified character and a quotient on which
inertia acts by eps_l^{-1} times an unramified character; this is the
Hodge--Tate ordering (0,1). Local fields used to verify these properties may
be enlarged independently at different places.

A continuous representation overlineoverline r:G_Lr:G_LtoGL_2(overlineF_l) over a
number field L is decomposed generic if there is a rational prime
qne l splitting completely in L such that for every vmid q it is
unramified at v and the Frobenius eigenvalues alpha_v,beta_v satisfy
alpha_v/beta_vne q and beta_v/alpha_vne q in overlineF_l. Our
construction will take qequiv1pmod l, so it suffices to have
alpha_vnebeta_v at every vmid q.

# A cyclic-cover family and its two degenerations

Fix an imaginary quadratic field K, a non-CM elliptic curve E/K written
E:y^2=h(x) with h a separable cubic, and the origin O=[0:1:0] at
infinity. Fix an odd integer nge3 for the moment. We build a family of
cyclic covers of quadratic twists of E and identify two boundary fibers.
At the first the Jacobian becomes a product of copies of E; at the second
the cover acquires nodes whose vanishing cycles give explicit monodromy. All
degenerations here are over a characteristic-zero parameter curve.

## The family, its dihedral symmetry, and geometric connectedness

For t with h(t)ne0 put

 B_t: h(t)y^2=h(x), R_t=(t,1), S_t=[n]R_t,

the y-coordinate normalized so that R_t is the point (t,1) on the
twist. Let UsubsetA^1_k over ksupset K(mu_n) be the open set where
O,R_t,-R_t,S_t,-S_t are pairwise distinct. This is nonempty: over the double
cover t=x(R) parametrized by Rin E, the map
B_{x(R)}to E, (x,y)|->(x,y(R)y), identifies R_t with the moving
point R, and each excluded equality forces R to be torsion, hence excludes
only finitely many points. We may delete further finitely many points of U
whenever a statement requires it.


For each divisor m>1 of n there is a smooth proper family of
geometrically connected genus-m curves cC_mto U. It is the
normalization of the smooth proper elliptic family Bto U in the
function-field extension described fiberwise by the Kummer equation

 Y^m=f_t(Z),
 div(f_t)=(S_t)-(-S_t)-n(R_t)+n(-R_t), f_t(O)=1.

The family carries an involution

 w:(Z,Y)⟼(-Z,Y^{-1})

and, over a field containing mu_m, rotations
u_zeta:(Z,Y)|->(Z,zeta Y) with wu_zeta w=u_{zeta^{-1}}. For
dmid m the map Y|-> Y^{m/d} exhibits C_{m,t}to C_{d,t}; for d=1
the quotient is B_t.


Proof. 
On an elliptic curve a degree-zero divisor is principal if and only if its
sum in the group law is O. The divisor in (geom:kummer) has sum
S_t-(-S_t)-nR_t+n(-R_t)=2S_t-2nR_t=O, so it cuts out a rational function,
pinned down by f_t(O)=1. Working over k(U) and spreading out gives such a
function on the family. Its divisor changes sign under Z|->-Z, so
normalizing at O yields

 f_t(-Z)=f_t(Z)^{-1}.

This also shows f_t has no vertical divisor: every fiber component is
preserved by inversion, so its valuation is its own negative. Hence the
prescribed disjoint sections are the whole divisor.

The valuation of f_t at S_t is 1, so the Kummer extension has degree
m even geometrically: it is totally ramified at S_t and -S_t and
unramified elsewhere, because the remaining nonzero valuations pm n are
divisible by m. Riemann--Hurwitz gives 2g(C_{m,t})-2=2(m-1), so g=m.
The same local computation shows the normalization is smooth over U: a
zero or pole of order divisible by m drops out of the Kummer equation, and
at a section of order pm1 one takes a root of an 'etale unit and obtains
the mth-root cover of a transverse parameter. Normalization is finite here,
so the resulting family is proper. Equation~(geom:inverse) gives w,
and the quotient statements follow from the Kummer equations.


We pull this family back to the open set of E where R moves and
O,pm R,pm[n]R are distinct; the base elliptic curve there is the constant
curve E. We write C_{m,R} for the fiber and f_R for its normalized
function. The normalization at O makes the pullback identification exact:
it introduces no scalar into Y^m=f_R.

## Separating colliding sections

The two boundary parameters we use are

 R=O,
  and 
 R=R_*, [n]R_*in E[2]∖{O},
  O,pm R_*,[n]R_* distinct.

The second exists after extending scalars and will be used over C.


In the constant-E family, blow up the surface at the collision point over
either boundary parameter in (geom:boundary) and normalize the result in
Y^m=f_R. This is a regular semistable model near that parameter. Its
special fiber has m geometric nodes over the node of the blown-up elliptic
surface, each with smoothing parameter of order one in a base parameter. The
component over the exceptional projective line is a geometrically connected
rational curve.


Proof. 
At R=O use parameters r on the parameter curve and z on E, both with
the same tangent coordinate at O. The sections R,-R,[n]R,-[n]R have
first-order equations z=r,-r,nr,-nr; their strict transforms meet the
exceptional line at 1,-1,n,-n, and the node joining that line to the main
component is at ∞. At R_* only [n]R and -[n]R collide, with
distinct tangent slopes, so one blowup separates them from each other and from
the node.

The involution Z|->-Z preserves both components of each blown-up
fiber, so by (geom:inverse) the valuation of f_R on each component is
zero; no horizontal divisor section passes through the node. Since the
surface is regular, f_R is a unit there, and an mth root exists 'etale.
The node downstairs has local equation xy=b, so its m lifts have the same
equation and smoothing parameter b rather than a higher power. At the
separated sections the normalization is smooth as in Lemma~geom:family.
On the exceptional line the cover is ramified exactly at the two sections of
multiplicities pm1, is connected, and has genus zero by Riemann--Hurwitz.


[First degeneration]
Suppose k contains mu_m. At R=O the special fiber of
Lemma~geom:semistable is a rational curve meeting m copies of E,
each at one point, all defined over k. Rotations cyclically permute the
elliptic components. The Jacobian family of C_{m,R} extends to an abelian
scheme on a Zariski neighborhood of O in the parameter curve, with fiber
E^m at O.


Proof. 
Let v=z/r on the exceptional line. Its normalized function is

 f_{exc}(v)=(v-n)/(v+n)((v+1)/(v-1))^{n}.

The divisor gives this up to a scalar; the section Z=O meets the line at
v=0 where the value is 1, and since n is odd (geom:exc) is also
1 at 0, so the scalar is 1.

On the main elliptic component f_R has no divisor and is constant, equal to
f_{exc}(∞)=1 at the node. Thus the cover of the main
component is the disjoint union of the m copies of E indexed by
Yinmu_m. On the exceptional line the substitution
Y'=Y((v-1)/(v+1))^{n/m} gives Y'^m=(v-n)/(v+n); its m
points over v=∞ are defined over k and attach one to each elliptic
component.

The dual graph is a tree, so the fiber is of compact type: the
identity-component Picard variety is the product E^m of the component
Jacobians. The marked point (O,1) extends to a section through the smooth
locus, and the relative Picard theorem for nodal curves of compact type makes
the Jacobian family an abelian scheme across this fiber (see
cite[Section~9.4, Theorem~1, and Section~9.2, Proposition~10]{BLR}). Shrinking
the parameter neighborhood removes the other singular fibers.



Proposition~geom:compact concerns an abelian scheme over an open subset
of the characteristic-zero parameter curve E_k, and after completing k it
is a family over that local field. No integral model of the Kummer cover in
residue characteristic dividing m is needed; the arithmetic reduction types
of nearby Jacobians are established separately.


## The second degeneration and its nodes

Choose an oriented period basis a,b of E(C) and take R_*=a/(2n), so
[n]R_*=a/2 is a nonzero 2-torsion point. The first-order equation below
records why a loop about R_* contributes one twist at each node.


For R_*=a/(2n) the special fiber of Lemma~geom:semistable consists
of a connected unramified degree-m cover E'to E and a rational
component, joined at m distinct nodes. In analytic coordinates near each
node the family has equation XU=-2nt, where t=R-R_* is an additive local
parameter.


Proof. 
Write Q=a/2=[n]R_* and use additive coordinates t=R-R_* and z=Z-Q. The
colliding sections are z=nt and z=-nt, the other divisor sections avoid
this neighborhood, and
[
 f_R=g(z,t),(z-nt)/(z+nt)
]
with g a unit. Taking a local mth root of g and V=Y/g^{1/m}, the
Kummer equation becomes

 (z+nt)(V^m-1)=-2nt.

For each V=zetainmu_m, U=V^m-1 is a local coordinate; setting
X=z+nt gives the asserted node equation. This is the local chart of the
normalization of the blowup of Lemma~geom:semistable, with X=0 the
exceptional component and U=0 the main component.

On the main component the limiting divisor is n(-R_*)-n(R_*), whose
unramified Kummer cover is governed by the m-torsion class
[
 frac nm((-R_*)-(R_*))
 ↔
 -(2n)/(m)R_*=-(a)/(m)quadin E(C).
]
This class has exact order m, so the cover is connected; being unramified of
degree m over an elliptic curve, it is itself an elliptic curve E'. The
exceptional cover is rational by Lemma~geom:semistable, and its m
attachments are exactly these nodes.


The first degeneration gives a split product fiber E^m together with a
neighborhood over which nearby smooth fibers can be compared to it; the
second gives m nodes of smoothing order one whose vanishing cycles we count
next in a single integral basis.

# Integral homology and explicit monodromy

Fix the complex elliptic curve E and the constant-E family of
Section~sec:geom. Let nge3 be odd and m>1 divide n. Write
Gamma_m=mu_m and let u:(Z,Y)|->(Z,e^{2pi i/m}Y) generate the deck
group; throughout Z[Gamma_m] is the group ring, not a cyclotomic integer
ring. We build an integral basis permuted by u and compute two monodromy
operators in that one basis. Their reduction on character spaces gives the
residual irreducibility used later.

## A basis from cutting the torus

Choose an oriented period basis a,b and identify the underlying oriented
torus with (R/Z)^2, with a horizontal and b vertical. Consider the
path of parameters

 R_s=(s)/(n)a, 0<s<1/2,

whose branch points are mp sa. Let I_s be the horizontal arc from -sa
to sa through O.


For parameters (mon:path) the cover C_{m,R_s}to E is trivial over
E∖ I_s; it is obtained by cutting m copies of E along I_s and
joining adjacent banks cyclically. Crossing the slit changes the sheet index by
pm1 according to the crossing orientation and the initial indexing.


Proof. 
For small s enclose the colliding sections in a small disk about O. By
Proposition~geom:compact the model at R=O restricts outside that disk
to m disjoint copies of its complement. The inclusion
E∖int(D)into E∖ I_s is a homotopy
equivalence, so the unramified cover of E∖ I_s is trivial. As s
varies in (0,1/2) the arcs I_s move by isotopy and the cover's
topological type is constant. The monodromy around either branch point is a
generator of Gamma_m, since its valuation in f_R is pm1. This gives
the stated gluing.


On each sheet let a_i be the horizontal loop at height 1/2 and b_i the
vertical loop at position 1/2, iinZ/mZ. They avoid the slit, meet once
on their common sheet, and are oriented so that

 < a_i,b_j>=delta_{ij},
 < a_i,a_j>=< b_i,b_j>=0,

and u sends a_i,b_i to a_{i+1},b_{i+1}.


The cycles a_i,b_i form an integral symplectic basis of
H_1(C_{m,R_s},Z). Consequently, for every smooth fiber,

 H_1(C_m,Z)~=Z[Gamma_m]^2,
  H^1(C_m,Z)~=Z[Gamma_m]^2.

After comparison, H^1_{et}(C_m,Z_l) is free of rank two over
Z_l[Gamma_m] for every prime l, including lmid m.


Proof. 
The genus is m, so H_1 has rank 2m. By (mon:intersection) the
2m displayed classes are independent and span a sublattice with unimodular
pairing; if its index in H_1 were d, the intersection determinant would
differ by d^2, but both are 1, so d=1. The deck action gives the first
isomorphism.

The dual of a regular module is regular: the coefficient-of-identity pairing
Z[Gamma_m]timesZ[Gamma_m]toZ, (x,y)|->[1](xy), is perfect and,
with the pullback action on the dual, identifies
Hom_{Z}(Z[Gamma_m],Z) with Z[Gamma_m]. This proves the
cohomological assertion without dividing by m. Topological local
triviality transports the result through the smooth parameter open, and
comparison gives the 'etale statement.



The compact-type degeneration gives a second view: the Jacobian extends to an
abelian scheme with special fiber E^m on which Gamma_m permutes the
factors regularly, and the integral homology of that fiber is the regular
module twice. The slit construction makes this concrete by identifying cycles
that continue toward the other degeneration.


## Vanishing cycles and a common monodromy basis

Let sto1/2 in (mon:path). The branch points meet at a/2 while the
slit fills all but a shrinking gap in the horizontal circle. Take a small disk
about a/2 containing both branch points; its boundary has m closed lifts
delta_i, the neck circles at the m nodes of Lemma~geom:collision,
hence the vanishing cycles.


After a cyclic indexing of the lifts,

 [delta_i]=pm(a_i-a_{i+1}), iinZ/mZ,

or the same formula with i-1 in place of i+1.


Proof. 
Arrange the boundary circle to cross the slit transversely twice. A lifted
boundary circle has its upper arc on one sheet and its lower arc on an adjacent
sheet; index the upper sheet by i, so the lower is i+1 or i-1 by
Lemma~mon:slit. The circle is disjoint from every a_j, and crosses the
two corresponding vertical loops once each with opposite signs and no other
b_j. These intersections determine its class in the integral symplectic basis
of Proposition~mon:integral, giving (mon:delta).


[Monodromy transvections]
In one integral basis a_i,b_i as above, the geometric monodromy of the
constant-E family contains operators T_a,T_b with

 T_a(a_j)&=a_j,&
 T_a(b_j)&=b_j+sigma_a(2a_j-a_{j-1}-a_{j+1}),
 T_b(b_j)&=b_j,&
 T_b(a_j)&=a_j+sigma_b(2b_j-b_{j-1}-b_{j+1}),

where sigma_a,sigma_bin{1,-1} are independent of j. Equivalently,
their off-diagonal entries in the free Z[Gamma_m]-module are
sigma_a(2-u-u^{-1}) and sigma_b(2-u-u^{-1}). The geometric monodromy of the
family over U contains these actions under the pullback identification.


Proof. 
The Picard--Lefschetz formula for a positively oriented loop about a
semistable fiber is the product of the Dehn twists at its nodes (see
cite[Expos'e~XV]{SGA7II}). Every node has smoothing order one by
(geom:node), so every twist has exponent one. The vanishing cycles are
disjoint, so their mutual intersection pairings vanish, and a class gamma
maps to

 gamma⟼gamma+sigmasum_i<gamma,delta_i>[delta_i]

with one sign sigma fixed by the intersection convention. Substituting
(mon:delta) gives zero on the a_j and, up to the sign,

 (a_j-a_{j+1})-(a_{j-1}-a_j)=2a_j-a_{j-1}-a_{j+1}

on b_j, proving (mon:Ta). Replacing the horizontal path by
R_s=(s/n)b gives (mon:Tb), initially in the basis from the vertical
slit.

It remains to compare the two bases, since separate triangular matrices in
unrelated bases would not suffice. Choose both starting parameters close to
O. The horizontal loop at y=1/2 and the vertical loop at x=1/2 avoid one
fixed small disk about O and both short slits. Outside this disk the covering
extends across R=O as m disjoint sheets; trivialize it over a small simply
connected parameter disk containing O and use this trivialization to label the
lifts of the two fixed loops. Parallel transport between the starting
parameters then identifies both sets of a_i,b_i with the same basis. There is
no monodromy ambiguity from looping O, because the sheet trivialization
extends over O. A different initial labeling only imposes a common cyclic
shift, leaving (mon:Ta)--(mon:Tb) unchanged.

Both loops therefore give these operators in a single geometric monodromy
group. Pullback along the open parameter curve of Eto U maps its fundamental
group into that of U, so the operators occur in the latter image too, and
deleting further finitely many points does not remove them.


The operator 2-u-u^{-1} is the cyclic difference operator coming from the
adjacent-sheet vanishing classes. The next section uses the integral module of
Proposition~mon:integral to specialize rotation characters, and the two
matrices above to keep the resulting residual representations irreducible.

# Rank-two systems and the congruent chain

The cyclic covers now supply the Galois representations that transfer
automorphy. The involution pairs opposite rotation characters, making each
character space two-dimensional with the determinant and Hodge--Tate weights of
an elliptic curve. The integral regular-module computation then yields
congruences even at primes dividing the covering degree.

## Standing setup

Return to the imaginary quadratic field K and a non-CM elliptic curve E/K.
By Serre's open image theorem cite[S4.4, Th'eor`eme~3]{SerreOpenImage}
choose a prime p>100 such that the image of G_K on E[p] contains
SL_2(F_p), and set

 n=3p, F=K(mu_n).

Both K/Q and Q(mu_n)/Q are abelian, so F/Q is abelian; F is a CM
field, since complex conjugation is central with totally real fixed field and
F has no real embedding (it contains K). Moreover

 zeta_5notin F.

Indeed Q(mu_n) is unramified at 5, so the ramification index of F at
5 is at most [K:Q]=2, whereas a field containing Q(zeta_5) has
ramification index divisible by 4 at 5.

Apply Lemma~geom:family with this n: let UsubsetA^1_K be its
parameter open and B_t the quadratic twist of E. We shrink U whenever a
smooth proper model of one of the finitely many families is needed. Until
Section~sec:res the point tin U(K) is arbitrary.

## Character spaces

Fix an embedding overlineQintoC and put M=Q(mu_n). For each prime
l choose j_l:overlineQintooverlineQ_l and an isomorphism
iota_l:overlineQ_ltoC whose composite with j_l is the fixed complex
embedding. Every rotation character is a homomorphism
theta:mu_n(F)to M^times. For the fiber C_n=C_n(t) define

 V_{theta,l}=bigl{xin H^1_{et}(C_{n,overlineoverline F}F,overlineQ_l):
 u^*x=j_l(theta(u))x for all uinmu_n(F)bigr},

where u denotes the deck transformation with multiplier u. All deck
transformations are defined over F, so these are G_F-representations; write
V_theta for the system of realizations. In any congruence at l all
characters use the same j_l, which is essential when comparing reductions.


For every tin U(K) and every rotation character theta, the system
V_{theta,l} satisfies:
[label=(arabic*)]
 it is semisimple of dimension two with determinant eps_l^{-1};
 at every place of F above l it is potentially semistable with
labeled Hodge--Tate weights {0,1};
 there is a finite set of places S of F, independent of l, such
that V_{theta,l} is unramified outside S at places of residue
characteristic ne l, with Frobenius characteristic polynomial the image under
j_l of a polynomial in M[T] independent of l.



Proof. 
The rank-two regular module of Proposition~mon:integral, followed by
comparison between singular and l-adic cohomology, gives dim V_{theta,l}=2.

Let <,,> be the cup-product pairing on H^1(C_n) into
overlineQ_l(-1). It pairs the theta-space perfectly with the
theta^{-1}-space, and the involution w, defined over F and preserving cup
product, exchanges them. Thus
[
 [x,y]_theta=< x,w^*y>, x,yin V_{theta,l},
]
is a perfect G_F-equivariant pairing into overlineQ_l(-1), alternating
because < y,w^*x>=< w^*y,x>=-< x,w^*y>. A
perfect alternating pairing on a two-dimensional space trivializes the
determinant against the target line, so det V_{theta,l}=eps_l^{-1}.

For the Hodge numbers, fix an embedding of F into C and let a_theta be
the multiplicity of theta in the holomorphic differentials of the resulting
curve. The holomorphic involution gives a_theta=a_{theta^{-1}}, and the
Hodge decomposition together with complex conjugation identifies the dimension
of the full theta-space with a_theta+a_{theta^{-1}}. As this dimension is
two, a_theta=1 for every embedding and character; the comparison theorems
then give labeled weights 0,1. Potential semistability comes from semistable
reduction and the semistable comparison theorem for smooth proper varieties
[cite], and passes to character summands.

Outside a finite set C_n has good reduction and the rotations extend to its
smooth proper model; smooth proper base change gives unramifiedness. The
projector onto the theta-space is
e_theta=1/nsum_{uinmu_n(F)}j_l(theta(u))^{-1}u^*, used only in
characteristic zero. The Lefschetz trace formula for a rotation composed with a
positive power of Frobenius shows the trace on H^1(C_n) is independent of
l; applying the projector gives traces in M compatible under the j_l.
Together with the known determinant this gives the common characteristic
polynomial.

Finally H^1(C_n) is the first cohomology of its Jacobian, so Faltings'
semisimplicity theorem cite[S5, Satz~3]{Faltings} makes it a semisimple
G_F-representation. Extension of scalars and passage to a stable summand
preserve semisimplicity.


## The two elliptic-curve realizations

Characters of order one and three connect the systems to elliptic curves. For
the order-three character the involution produces a genus-one quotient rather
than only an abstract rank-two summand.


If theta has order m>1 dividing n, the system V_theta is realized by
the corresponding character space in H^1(C_m). For the trivial character,
V_{1,l}~= H^1_{et}(B_{t,overlineoverline F}F,overlineQ_l). Let
D_t=C_3(t)/< w> and A_t=Jac(D_t). Then D_t has genus one,
A_t is an elliptic curve over F, and for either nontrivial order-three
character eta there are G_F-equivariant isomorphisms
[
 V_{eta,l}~= H^1_{et}(A_{t,overlineoverline F}F,overlineQ_l)
  for every l.
]


Proof. 
The quotient C_nto C_m sends Y to Y^{n/m}. Pullback is injective on
rational cohomology since its composite with the trace is multiplication by the
degree, and its image is the fixed space of the kernel of mu_ntomu_m; the
rank computation identifies the order-m character summand with
V_{theta,l}. The same argument for C_nto B_t identifies the trivial
character space.

On H^1(C_3) the trivial rotation space is H^1(B_t), where w acts by
[-1]^*=-1, and w exchanges the two nontrivial rotation spaces, each of
dimension two. Their direct sum therefore contains a two-dimensional
w-invariant subspace. Rational cohomology of a finite quotient is the
invariant space, so H^1(D_t)~= H^1(C_3)^{w=1} has dimension two: D_t has
genus one and its Jacobian A_t is an elliptic curve, with or without an
F-rational point. Projection of x+w^*x onto the eta-space is x, so this
projection is a G_F-equivariant isomorphism from the invariant space to
V_{eta,l}. Finally H^1(D_t)~= H^1(Jac(D_t)) depends on no choice of
geometric base point, since translations act trivially on cohomology.


## Evaluation of the integral group ring

A character projector carries the denominator n and cannot by itself give a
congruence at 3 or p. Instead we use the integral freeness of
Proposition~mon:integral: evaluating a free group-ring module at a
character yields a lattice before reduction, and two characters with equal
residual values then yield equal residual representations.


Fix tin U(K) and a prime l. Let L/Q_l be a finite extension containing
j_l(M), with ring of integers cO, maximal ideal m,m, and residue
field k. For a rotation character theta the module
[
 T_
 =H^1_{et}(C_{n,overlineoverline F}F,Z_l) ⊗ _{Z_l[mu_n],,j_ltheta}cO
]
is a free rank-two G_F-stable lattice realizing V_{theta,l} after
extension to overlineQ_l. If j_ltheta and j_ltheta' have the same
reduction as characters mu_nto k^times, then
T_/mT_~=
T_{theta'}/mT_{theta'} as k[G_F]-modules.


Proof. 
By Proposition~mon:integral and comparison,
H^1_{et}(C_{n,overlineoverline F}F,Z_l)~=Z_l[mu_n]^2 as group-ring modules,
and the Galois action commutes with the group ring because the rotations are
defined over F. Tensoring by evaluation at j_ltheta gives a free rank-two
cO-module with continuous Galois action; in characteristic zero the group
algebra is semisimple and its quotient by character evaluation is the
corresponding summand, so the rational representation is V_{theta,l}. After
reduction the tensor product becomes
H^1_{et}(C_{n,overlineoverline F}F,Z_l) ⊗ _{Z_l[mu_n],bartheta}k, which
depends only on the residual evaluation bartheta. This holds without
assuming l prime to n.


Choose a primitive character chi of order n=3p with primary decomposition
chi=psieta, ord(psi)=p, ord(eta)=3. For a system V_theta write
overlineoverline V_V_{theta,l} for the semisimplified reduction of a stable lattice,
scalars extended to overlineF_l; this is independent of the lattice.


For every tin U(K),

 overlineoverline V_V_{eta,p}~=overlineoverline V_V_{chi,p},
 
 overlineoverline V_V_{chi,3}~=overlineoverline V_V_{psi,3},
 
 overlineoverline V_V_{psi,p}~=overlineoverline V_V_{1,p},

holding with the elliptic-curve realizations of Proposition~sys:elliptic
at the two ends.


Proof. 
A root of unity of l-power order reduces to 1 in characteristic l. Thus
chi and eta agree modulo p, chi and psi agree modulo 3, and
psi and 1 agree modulo p. Lemma~sys:integral gives the three
congruences, and the rational isomorphisms with smaller covers and elliptic
curves in Proposition~sys:elliptic preserve semisimplified reduction even
if their natural integral pullbacks have index divisible by l.


The direction is now fixed: automorphy will pass from V_eta to V_chi at
p, to V_psi at 3, and to V_1 at p. The congruences place no
condition on t; the local and residual arguments choose one t for which all
three lifts apply.

# Automorphy inputs

The congruences of Corollary~sys:congruences tie the cohomology of the
auxiliary elliptic curve A to that of the twist B_t of E. To turn them
into automorphy we use two results of Caraiani--Newton, stated here in the
exact form needed, and then isolate what the parameter t must still
satisfy. All references to [cite] use the numbering of its third arXiv
version.

The residual hypothesis below uses the notion of decomposed genericity
introduced in Section~sec:conv.

We also record the automorphic condition paired with potential ordinarity of
Galois representations in the convention of Section~sec:conv. It is
called iota-ordinarity of weight zero and is the ordinary Hecke
condition used in cite[Theorem~5.2]{CN}. It is imposed at a sufficiently deep
level, so it allows ramified finite-order characters. We obtain it only for
quadratic twists of Steinberg, via the local computation of
cite[Lemma~6.1.3(4)]{CN}.

[Caraiani--Newton lifting, specialized form]
Let L be a CM number field and l an odd prime other than 5. Let
r:G_LtoGL_2(overlineQ_l) be continuous, unramified outside finitely many
places, with determinant eps_l^{-1}. Suppose that:
[label=(roman*)]
 r is potentially semistable with labeled Hodge--Tate weights 0,1 at
every vmid l;
 overlineoverline rr is decomposed generic;
 overlineoverline r|_r|_{G_{L(zeta_l)}} is absolutely irreducible.

Suppose there are a cuspidal weight-zero regular algebraic representation
pi of PGL_2(A_L) and an isomorphism iota:overlineQ_ltoC with
r_iota(pi)bar~=overlineoverline r.r. At each vmid l assume one of:
[label=(arabic*)]
 both local Galois representations are potentially crystalline; one is
potentially ordinary of weight zero if and only if the other is; and
rec_{L_v}(pi_v) has zero monodromy;
 neither local Galois representation is potentially crystalline, and
pi is iota-ordinary of weight zero at v.

Then r~= r_iota(Pi) for a cuspidal weight-zero regular algebraic
representation Pi of PGL_2(A_L).


This is the sufficient-hypothesis specialization of cite[Theorem~5.2]{CN}.
Its inverse-cyclotomic determinant hypothesis permits l=3, including residual
image SL_2(F_3) cite[Remark~5.2.2]{CN}. No residual adequacy is assumed.
Our three applications below have l=3 or l=p>100.

[Elliptic starting point]
Let L be a CM number field not containing zeta_5 and let A/L be a non-CM
elliptic curve. If its mod-5 representation is decomposed generic and is
absolutely irreducible on G_{L(zeta_5)}, then H^1(A) is realized by a
cuspidal weight-zero regular algebraic representation of PGL_2(A_L).


This is cite[Theorem~6.1 and Lemma~6.1.3]{CN}, dualizing the Tate-module
convention there if necessary; absolute irreducibility and decomposed
genericity in dimension two are preserved by duality.

For the comparison after modularity of an elliptic curve is known, we use:

[Compatibility for modular elliptic curves]
Let D/L be a non-CM elliptic curve over a CM field L. Suppose a regular
algebraic cuspidal representation pi realizes H^1(D) at one coefficient
prime. Then pi has weight zero and trivial central character, realizes
H^1(D) at every coefficient prime and every coefficient isomorphism, and
[
 iotaWD(H^1(D,overlineQ_l)|_{G_{L_v}})^{Fss}
 ~=rec^T_{L_v}(pi_v)(vnmid l).
]
If D has potentially multiplicative reduction at vmid l, then pi is
iota-ordinary of weight zero there.


This is cite[Lemma~6.1.3]{CN}. For the intermediate systems, which need not
come from elliptic curves, we use the general existence and good-place
compatibility of r_iota(pi) from [cite], together with Varma's
equality of the underlying semisimplified Weil representations at vnmid l
cite[Theorem~1]{Varma}. The local analysis of Section~sec:local supplies
the extra facts needed in Theorem~aut:lift.

We can now state what remains for a single parameter tin U(K). By
Proposition~sys:properties, the determinant and Hodge--Tate conditions
hold for every t. At places above 3p we will make the Jacobian of
C_n(t) potentially good ordinary, potentially good with all Newton slopes
1/2, or potentially totally toric, according to the reduction type of E.
This makes all four systems respectively potentially crystalline and
potentially ordinary, potentially crystalline but not potentially ordinary, or
not potentially crystalline. Section~sec:local obtains these from
nonempty local open conditions on t. Section~sec:res then picks a
parameter in those opens for which the residual hypotheses hold. The pairs
(m,l)=(3,5),(3,p),(p,3) give the elliptic starting point and the first two
lifts; the last lift uses the large mod-p image of E. The two conjugates of
K must be treated together to get decomposed genericity. Once these
properties hold, Section~sec:auto carries automorphy along the three
congruences and descends to K.

# Prescribing the local reduction type

The degeneration at the origin lets us impose the same local behavior on every
character summand of the covering Jacobian. We first explain why the reduction
type of an abelian variety is controlled, locally in a characteristic-zero
family, by a sufficiently large finite torsion module, and then apply this to
the Jacobian family and translate the reduction into conditions on the
two-dimensional systems.

## Finite torsion controls the reduction

Throughout, a local field is a finite extension of Q_ell; here ell is
the residue characteristic of the base field, and d is an auxiliary prime
different from ell. For an abelian variety A write
V_d(A)=T_d(A) ⊗ _{Z_d}Q_d.

[Local constancy of finite torsion]
Let L be a local field, X a smooth L-variety, A X an
abelian scheme. For x_0in X(L) and Nge1 there is an analytic
neighborhood N of x_0 in X(L) such that
A_x[N](A_x[N](overlineoverline L)L)~=A_{x_0}[N](overlineoverline L)L) as
G_L-modules for all xinN.


Proof. 
Since X has characteristic zero, A[N]A[N] is finite 'etale over
X; equivalently, full level-N structures form a finite 'etale torsor
under the constant group GL_{2g}(Z/NZ), with g the relative dimension.
Take a finite Galois extension L'/L splitting the fiber at x_0. The inverse
function theorem for an 'etale morphism gives, through each point of the fiber,
a local section over a neighborhood of x_0 in X(L'); shrinking, these
sections enumerate every point of every nearby fiber. Intersecting finitely
many Galois conjugates and shrinking again keeps the neighborhood Galois-stable
and preserves the Galois permutations; the level-group action is preserved the
same way. Restricting to X(L) identifies the fibers as Galois sets with
level-group action, hence the torsion modules.


Finite torsion will first force semistability, then identify a Frobenius
polynomial. For the second step we use the full Tate module, which
keeps both toric and lattice parts of the semistable representation.

[Finitely many semistable Frobenius polynomials]
Fix a local field L with residue field of cardinality Q, an integer
gge1, and a prime d different from the residue characteristic. For a
semistable abelian variety A/L of dimension g let
P_A(T)=det(T-Frob_Lmid V_d(A)) for any lift of arithmetic Frobenius.
Then P_A(T)inZ[T], is independent of the lift and of d, and ranges over
a finite set depending only on g,Q. It determines the toric rank of A; if
that rank is zero it also determines the Newton polygon of the good reduction.


Proof. 
Let r be the toric rank. The semistable filtration identifies the graded
pieces of V_d(A) with the rational Tate module of a rank-r torus, the
rational Tate module of a (g-r)-dimensional abelian variety with good
reduction, and a rank-r lattice cite[Expos'e IX]{SGA7I}. The character
lattice of the torus and the lattice quotient act through finite unramified
characters, so the arithmetic Frobenius eigenvalues on those pieces are Q
times roots of unity and roots of unity respectively; the middle piece has the
2(g-r) eigenvalues of an abelian variety over F_Q, all of complex absolute
value Q^{1/2}.

Each graded piece has an integral characteristic polynomial independent of
d: for the toric and lattice pieces this follows from the finite-order
integral actions and the toric Tate twist, for the abelian piece it is the
usual finite-field Frobenius polynomial. Inertia acts trivially on the graded
pieces, so the product is independent of the Frobenius lift. Every root has
absolute value at most Q, so the coefficient of T^{2g-j} has absolute value
at most binom{2g}{j}Q^j; integral polynomials of fixed degree with these
bounds are finite in number.

Exactly r roots have absolute value 1, exactly r have absolute value
Q, and the rest have Q^{1/2}; as Q>1 these three values distinguish r.
If r=0, semistability is good reduction and the normalized valuations of the
roots of P_A are the Newton slopes.


[Local constancy of the semistable type]
Let L_0 be a local field, X a smooth L_0-variety, A X an
abelian scheme of relative dimension g. For x_0in X(L_0) there exist a
finite extension L/L_0 and an analytic neighborhood N
X(L_0) of x_0 such that for every xinN, A_x is
semistable over L with the same toric rank as A_{x_0} over L,
and if A_{x_0} has good reduction over L then A_x has
the same Newton polygon over L.

In particular, if A_{x_0}~= E_0^g, every nearby fiber has, after
the same finite extension, the same one of: good ordinary reduction; good
reduction with all Newton slopes 1/2; semistable reduction of toric rank g.


Proof. 
Choose an odd prime d different from the residue characteristic and a finite
L/L_0 over which A_{x_0} is semistable with rational d-torsion.
Lemma~loc:torsion after base change to L shows nearby fibers also have
rational d-torsion over L.

Nearby fibers are semistable over L. Indeed, an inertia element acts on their
d-adic Tate module as 1+dB with B integral, so each eigenvalue lambda
satisfies v_d(lambda-1)ge1, v_d(d)=1. By the local monodromy theorem
lambda is a root of unity; for odd d the only such root of unity with
v_d(lambda-1)ge1 is 1 (a root with nontrivial prime-to-d part has
v_d(lambda-1)=0, and a primitive d^ath root has
v_d(lambda-1)=1/(d^{a-1}(d-1))<1). Thus inertia acts unipotently and
Grothendieck's criterion gives semistability cite[Expos'e IX]{SGA7I}.

Let P be the finite set of Lemma~loc:frob for dimension g
and residue cardinality of L; choose j with reduction mod d^j injective
on P. A second application of Lemma~loc:torsion gives, on a
smaller neighborhood, A_x[d^j](A_x[d^j](overlineoverline L)L)~=
A_{x_0}[d^j](overlineoverline L)L) as G_L-modules, so the full Frobenius
polynomials are congruent mod d^j and hence equal. Lemma~loc:frob now
gives the assertions on toric rank and Newton polygon; pull the neighborhood
back to X(L_0).

For the final assertion, choose L so E_0 has good or split multiplicative
reduction. In the good case E_0 is ordinary or supersingular, the latter with
slopes 1/2; in the multiplicative case E_0^g has toric rank g. Enlarging
L to rationalize d-torsion preserves these, so the argument applies.


## Monodromy of a totally toric abelian variety

The third possibility of Proposition~loc:const must force nonzero
monodromy on each character summand, including when the coefficient
prime is the residue characteristic. The needed rank statement is a
consequence of the valuation pairing in toric uniformization, and we treat the
residue characteristic as well, through Kummer extensions.


Let L/Q_ell be finite and A/L an abelian variety of dimension g that
acquires semistable reduction of toric rank g after a finite extension. For
every prime d the monodromy operator on the Weil--Deligne representation of
H^1_{et}(A_{overlineoverline L}L,Q_d) has rank g; for d=ell the
Weil--Deligne representation is the one attached to the potentially semistable
representation by ell-adic Hodge theory.


Proof. 
Monodromy rank is unchanged by finite extension, so assume A has split toric
uniformization A^{an}~=(Gm^g)^{an}/Λ with
Λ~=Z^g rational over L. In bases the periods form a matrix
(q_{ij})in(L^times)^{gtimes g}, and the valuation pairing is nondegenerate:

 (v_L(q_{ij}))_{1le i,jle g} is nonsingular.

This is the toric case of the monodromy pairing cite[Expos'e
IX]{SGA7I}; its de Rham realization is in [cite].

For dneell the same pairing describes the monodromy map from the lattice
quotient to the toric subspace, and nondegeneracy gives rank g. We prove rank
g for d=ell. Toric uniformization gives an exact sequence

 0→Q_ell(1)^g→ V_ell(A)
   →Q_ell^g→0,

whose extension classes are the Kummer classes of the q_{ij}; it is
semistable by [cite]. Let L_{ur} be the maximal
unramified subfield and D_{st} Fontaine's functor. Monodromy is zero
on the two outer terms, so it induces nu:D_{st}(Q_ell^g)to
D_{st}(Q_ell(1)^g). Frobenius on the two spaces is sigma and
ell^{-1}sigma with sigma the Frobenius of L_{ur}/Q_ell, so
Nphi=ellphi N makes kernusubset L_{ur}^g
sigma-stable.

Suppose nu has rank <g. Galois descent gives nonzero
a=(a_j)inQ_ell^gcapkernu. Pull (loc:tate) back along the line
spanned by a in its trivial quotient: the result is a semistable extension of
Q_ell by Q_ell(1)^g with zero monodromy, hence crystalline. By
Kummer theory the Bloch--Kato crystalline subspace H_f^1(L,Q_ell(1)) is the
kernel of the valuation map cite[Example~3.9]{BlochKato}, and applying this in
each coordinate gives sum_j a_j v_L(q_{ij})=0 for 1le ile g, contrary to
(loc:val). Hence N has rank g on V_ell(A), and dualizing gives rank
g on H^1_{et}(A_{overlineoverline L}L,Q_ell).


## Open conditions on the original parameter

Return to E/K, the odd integer n=3p, and F=K(mu_n). By
Proposition~geom:compact the Jacobian of the constant-E cover C_n
extends over a neighborhood of O in the parameter curve E, with fiber
E^n at O, so Proposition~loc:const applies at every finite completion
of F. The next statement re-expresses the resulting neighborhoods in the
original parameter tin UsubsetA^1_K, to combine with Hilbert
irreducibility.


For every place v_0mid3p of K there is a nonempty open
Ω_{v_0}subset U(K_{v_0}) with the following property. Suppose
tin U(K) lies in Ω_{v_0} for all v_0mid3p, and set
J_t=Jac(C_{n,t}). At each place vmid3p of F, J_t has the same one of
the following potential reduction types as E^n:
[label=(arabic*)]
 good ordinary reduction;
 good reduction with all Newton slopes 1/2;
 semistable reduction of toric rank n.

For each v the extensions giving these properties may be chosen in advance,
independently of t in the opens.


Proof. 
Fix v_0mid3p. For each vmid v_0 apply Proposition~loc:const to the
extended Jacobian family over F_v at the point O of its parameter curve,
obtaining a neighborhood of O in E(F_v) and a fixed finite extension of
F_v on which the reduction type holds. The inverse images of these finitely
many neighborhoods in E(K_{v_0}) intersect in a neighborhood of O.

Choose R_0ne O in this intersection, close to O and avoiding the finitely
many excluded parameters and the ramification points of x:EtoPbf^1. The
inverse function theorem gives a nonempty neighborhood
Ω_{v_0}subset U(K_{v_0}) of x(R_0) and an analytic lift
R(t)=(t,y(t))in E(K_{v_0}), y(t)^2=h(t), in the chosen intersection;
shrink so y(t)ne0 and all excluded parameters are avoided.

The map B_tto E, (x,y)|->(x,y(t)y), is an isomorphism over
K_{v_0} carrying (t,1) to R(t); it sends the defining function of C_n(t)
to that of the constant-E cover (their divisors agree and both are 1 at
O), so it identifies the covers and Jacobians at every vmid v_0. The
reduction conclusions follow from the choice of neighborhoods; do this
independently at the finitely many v_0mid3p.


## Local representations of the character summands

For a prime ell and rotation character theta, let V_{theta,ell} be
the ell-adic realization of Proposition~sys:properties. Recall from
Section~sec:conv that an ordinary filtration of weight zero has an
unramified character on its line and eps_ell^{-1} times an unramified
character on its quotient, after a finite extension; in the ordinary case below
the line is preserved already before the extension.


Choose t satisfying Proposition~loc:opens. Let vmid3p be a place of
F of residue characteristic ell. Then simultaneously for every rotation
character theta:
[label=(arabic*)]
 if E has potentially good ordinary reduction at v, then
V_{theta,ell}|_{G_{F_v}} is potentially crystalline and potentially
ordinary of weight 0;
 if E has potentially good supersingular reduction at v, then
V_{theta,ell}|_{G_{F_v}} is potentially crystalline and not potentially
ordinary of weight 0;
 if E has potentially multiplicative reduction at v, then the monodromy
of V_{theta,d}|_{G_{F_v}} has rank one for every coefficient prime d,
including d=ell; in particular V_{theta,ell}|_{G_{F_v}} is not
potentially crystalline.



Proof. 
Choose a finite L/F_v on which J_t has the reduction of
Proposition~loc:opens. In the two good-reduction cases, crystalline
comparison makes H^1_{et}(J_{t,overlineoverline L}L,Q_ell), and each character
summand, crystalline over L.

If the reduction is ordinary, the connected--'etale sequence of the ordinary
ell-divisible group, dualized to H^1, gives a filtration with an
unramified sub and a quotient that is eps_ell^{-1} times unramified. This
filtration is functorial, hence stable under the rotations, and restricts to
each V_{theta,ell}. By Proposition~sys:properties this summand has
dimension two and labeled Hodge--Tate weights 0,1; the unramified part is
weight 0 and the quotient weight 1, so both parts are one-dimensional,
giving the required ordinary filtration. After a finite Galois extension, the
weight-zero line is determined by its inertia action, since eps_ell has
infinite image on inertia; conjugation by G_{F_v} preserves it, so it descends
to the original decomposition group.

If all slopes are 1/2, every summand has that slope, and slopes are
unchanged by finite extension. An ordinary filtration of weight 0 would give
a crystalline rank-one subrepresentation of parallel Hodge--Tate weight zero,
whose slope is forced to be zero by weak admissibility, contradiction. So no
potential ordinary filtration exists.

If J_t has potential toric rank n, Lemma~loc:toric gives monodromy of
rank n on the 2n-dimensional H^1 at every coefficient prime. Monodromy
commutes with rotations, so preserves the n two-dimensional summands; since
N^2=0, each summand has rank at most one, the ranks sum to n, and there are
n summands, so every rank is one. Nonzero monodromy survives finite extension
and excludes potential crystallinity.


# A single parameter with all residual properties

We now choose one parameter for the whole congruence chain. The geometric
monodromy gives irreducibility and suitable Frobenius elements; Hilbert
irreducibility lets us keep them after imposing the local conditions of
Section~sec:local. To get decomposed genericity we must control every
place above a split rational prime, so we specialize the two conjugate
parameter families together.

Throughout, K,E,p,n,F,U are as in Section~sec:sys, so p>100, n=3p,
F=K(mu_n). The pairs of covering degree and residual characteristic needing
geometric monodromy are

 (m,ell)=(3,p),(3,5),(p,3).

In each pair ellnmid m, so residual cohomology decomposes into
rotation-character spaces after extending its finite coefficient field.

## Consequences of the two monodromy transformations


For each pair in (res:pairs) the geometric monodromy of
H^1(C_m,overlineF_ell) on every nontrivial rotation-character space is
absolutely irreducible, its image contains opposite nonidentity root
unipotents on that space, and one element of the geometric monodromy acts
regular semisimply on all these nontrivial character spaces simultaneously.
These statements also hold for the conjugate family over K.


Proof. 
Work first with homology; duality transfers the conclusions. On the character
space with generator eigenvalue lambdainoverlineF_ell^times the
transformations of Theorem~mon:transvections are
[
 T_a=1&sigma c_lambda0&1,
 T_b=1&0tau c_lambda&1,
 c_lambda=2-lambda-lambda^{-1},
]
with sigma,tauin{1,-1}. Since c_lambda=-(lambda-1)^2/lambda, it is
nonzero when lambdane1. The two matrices have different unique invariant
lines, even over the algebraic closure, so their common representation is
absolutely irreducible.

The product T_aT_b^k has determinant one and trace

 2+sigmatau,k,c_lambda^2.

In odd characteristic a determinant-one matrix is regular semisimple exactly
when its trace is neither 2 nor -2. If m=3 then
lambda+lambda^{-1}=-1, so c_lambda=3 for both nontrivial characters; take
k=2, so the two possible traces in (res:trace) are 20 and -16.
Modulo 5 these are 0 and 4, and modulo p>100 they are 20 and -16;
in neither case do they equal pm2.

For m=p and ell=3 take k=1; failure of regular semisimplicity would
force c_lambda^2=4 or -4, since c_lambdane0 and the excluded traces are
pm2. Over F_3 the elements 4 and -4 are pm1, so
c_lambda^2=pm1; equivalently c_lambda=pm1 or pm i. Multiplying
c_lambda=2-lambda-lambda^{-1} through by lambda^2 gives, in each case, a
degree-four polynomial over F_3 satisfied by lambda. Hence
lambda has degree dle4 over F_3 and its order divides
3^d-1le80, contradicting the fact that its order is the prime p>100.
Thus the same product works in this case too.

The transformations come from loops in the constant-elliptic-curve parameter
space, which by Theorem~mon:transvections is a pullback of the family over
U; its monodromy image lies in the geometric monodromy over U. Duality
preserves absolute irreducibility, nonidentity unipotents, and regular
semisimplicity, giving the cohomological claims. The geometric argument works
for every complex elliptic curve, hence for the conjugate of E.


## The original elliptic curve and its conjugate

The trivial character needs a separate argument, because the geometric family
varies only a quadratic twist of E; its large residual image was built into
p. The count below needs no independence hypothesis between E and its
conjugate.


For every tin U(K) the representation overlineoverline V_V_{1,p} is absolutely
irreducible on G_{F(zeta_p)} and decomposed generic over F.


Proof. 
Let Gamma be the image of G_K on E[p]. Since F/K is abelian, the image
of G_F contains [Gamma,Gamma]. As Gamma contains SL_2(F_p) and this
group is perfect for p>3, its commutator subgroup is SL_2(F_p) (one
inclusion from perfectness, the other from the determinant). The determinant is
trivial on G_F because mu_psubset F, so

 rho_{E,p}(G_F)=SL_2(F_p).

The same equality holds for the conjugate E^c, c the nontrivial automorphism
of K/Q, because F/Q is normal.

Consider the joint image JsubsetSL_2(F_p)timesSL_2(F_p) of G_F on
E[p] and E^c[p]; both projections are surjective. In SL_2(F_p) there
are exactly p^2 matrices of trace 2: writing
(a&bc&d), trace and
determinant give d=2-a and bc=-(a-1)^2; a=1 yields 2p-1 choices of
(b,c), and each of the other p-1 values of a yields p-1 choices, for a
total p^2. Negating gives the same count for trace -2.

The order of SL_2(F_p) is p(p^2-1), and each projection of J has
equal-sized fibers, so the proportion of J whose image in at least one factor
has trace pm2 is at most
[
 2·(2p^2)/(p(p^2-1))=(4p)/(p^2-1)<1.
]
Hence some element of G_F acts regular semisimply on both E[p] and
E^c[p].

Take a finite Galois extension of Q containing F and the fields of these
two residual actions. Conjugation by G_Q either preserves or exchanges the
two curves and conjugates their residual matrices, so the chosen element and all
its conjugates act regular semisimply on the representation of E. Chebotarev
gives rational primes q splitting completely in F with this property at
every vmid q of F. Excluding p and the finitely many bad primes and using
mu_psubset F, these satisfy qequiv1pmod p; distinct Frobenius
eigenvalues give the ratio condition of decomposed genericity.

The curve B_t is a quadratic twist of E; twisting multiplies residual
matrices by scalar signs, preserving absolute irreducibility and eigenvalue
ratios. Excluding the finitely many primes ramified in this twist leaves the
same Chebotarev argument. Since zeta_pin F and the dual of E[p] gives the
residual H^1-realization, Proposition~sys:elliptic proves both claims for
overlineoverline V_V_{1,p}.


## Simultaneous specialization

We have the required geometric elements and the trivial character is handled.
It remains to keep the geometric elements at one parameter while arranging their
Frobenius properties above a rational prime; restriction of scalars puts the two
conjugate families in one finite cover.


Let W be a nonempty open subset of A^d_Q, D/Q finite Galois, and H a
finite quotient of pi_1(W) through which the constant-field map factors as
pi_1(W)surj HsurjGal(D/Q). Prescribe nonempty open subsets of W(Q_r)
for finitely many rational primes r. Then there is xin W(Q) in all these
opens for which the specialization s_x:G_Qto H is surjective; for this x,

 s_x(G_D)=ker(HtoGal(D/Q)).



Proof. 
The quotient H defines a connected finite 'etale Galois cover of W.
Hilbert irreducibility with weak approximation gives a rational point in the
prescribed opens whose fiber is connected cite[Proposition~3.3.1 and
Theorem~3.5.3]{SerreHilbert}. Connectedness of the Galois fiber is equivalent to
surjectivity of s_x (geometric connectedness is not required). The composite
of s_x with HtoGal(D/Q) is the restriction map on absolute Galois
groups, so an element of the displayed kernel has a preimage in G_D.



Let S_K be a finite set of finite places of K containing all places above
3p and, for each v_0in S_K, prescribe a nonempty open
Ω_{v_0}subset U(K_{v_0}). There is a single tin U(K) in every
Ω_{v_0} such that, over F=K(mu_{3p}):
[label=(arabic*)]
 for each (m,ell) in (res:pairs) and each character theta of
order m, the representation overlineoverline V_V_{theta,ell} is absolutely
irreducible on G_{F(zeta_ell)} and decomposed generic;
 the representation overlineoverline V_V_{1,p} has the same two properties;
 the elliptic curve A_t=Jac(C_3(t)/< w>) is non-CM.

In particular, with the opens above 3p taken from Proposition~loc:opens,
this t also satisfies all conclusions of Corollary~loc:chars.


Proof. 
Put W=Res_{K/Q}U, an open subset of A^2_Q since UsubsetA^1_K. Choose
finite Galois D/Q containing F and zeta_ell for every residual
characteristic in (res:pairs). Let c be the nontrivial automorphism of
K/Q and write U^c, cC_m^c for the conjugate parameter open and family.
Then W_D~= U_Dtimes(U^c)_D. For all three pairs pull back the finite
local system H^1(C_m,F_ell) from the first factor and H^1(C_m^c,F_ell)
from the second.

Their joint monodromy has finite image on pi_1(W_D). Take, in pi_1(W), the
normal core of the common kernel; it is open normal, contained in pi_1(W_D),
and the resulting finite quotient H records every listed action and has
quotient Gal(D/Q).

Over C a loop in one parameter factor acts trivially on local systems pulled
back from the other, so the geometric monodromy of Lemma~res:geom is
available independently in both factors. For each pair (m,ell) choose an
element h_{m,ell} of the geometric monodromy image that is regular semisimple
on every nontrivial character space of both factors (product of the two factor
loops from the lemma); it maps trivially to Gal(D/Q). The quotient also
contains the individual opposite unipotents on every character space.

The local conditions are compatible with restriction of scalars: for a prime
r, W(Q_r)=prod_{v_0mid r}U(K_{v_0}), so the Ω_{v_0} give
nonempty open subsets of W(Q_r) for finitely many r (at unprescribed
factors impose no condition). Apply Lemma~res:finite: it gives
xin W(Q), i.e. tin U(K), satisfying all local conditions and inducing a
surjection G_Qsurj H. Over D this point has coordinates (t,t^c); smooth
proper base change identifies the specialized local systems with the residual
cohomology of C_m(t) and C_m^c(t^c).

By (res:kernel) every geometric monodromy element used above is the image
of an element of G_D. The opposite unipotents therefore survive on each
specialized character space over G_D, proving its absolute irreducibility; as
G_Dsubset G_{F(zeta_ell)}, this also holds on the latter group. Because
ellnmid m, taking rotation eigenspaces commutes with reduction after
adjoining the character values, so these spaces are reductions of
V_{theta,ell}; the rational identifications and independence of
semisimplified reduction of Section~sec:sys show they give exactly
overlineoverline V_V_{theta,ell}, even when the system is defined using C_n.

For decomposed genericity fix (m,ell) and a preimage gin G_D of
h_{m,ell}. An element sigmain G_Q preserves or exchanges C_m(t) and
C_m^c(t^c) according to its restriction to K, and permutes rotation characters
through its action on roots of unity; so sigma gsigma^{-1} on a character
space of the first cover is conjugate to g on some character space of one of
the two covers. Regular semisimplicity was imposed on all these spaces, so it
holds for every conjugate.

The kernel of the surjection G_Qsurj H defines a finite Galois extension of
Q containing D. Chebotarev applied to the conjugacy class of h_{m,ell}
gives rational primes q whose Frobenius class is that class; they split
completely in D because h_{m,ell} is trivial on D. Discard q=ell and
the finitely many primes where curves or residual representations ramify. For
every vmid q of F the Frobenius is represented by a conjugate of the
selected element and acts regular semisimply on each character space; the
residue degree is one since q splits in F, and qequiv1pmodell since
zeta_ellin D. Hence the ratio of the two Frobenius eigenvalues differs from
q in characteristic ell, at every vmid q: this is decomposed genericity.
A different q may be used for each pair. Assertion (2) follows from
Lemma~res:triv.

To exclude CM for A_t: for (m,ell)=(3,p) the specialization places two
opposite nonidentity unipotents in the residual image of V_{eta,p}, which by
Proposition~sys:elliptic is the residual cohomology of A_t (both are
irreducible, so their common semisimplification is the representation itself).
If A_t had CM, all its endomorphisms would be defined over an extension of F
of degree at most two; over that extension the characteristic-zero Galois image
lies in the commutative centralizer of the CM algebra, so its reduction is
abelian, and the full residual image would have an abelian normal subgroup of
index at most two. Each opposite unipotent has odd prime order p, so both
would lie in that abelian subgroup; but
[
 1&a0&11&0b&1
 ne1&0b&11&a0&1
 (abne0),
]
a contradiction. This proves (3). Finally, choosing the prescribed opens as in
Proposition~loc:opens, the Hilbert specialization was made inside them, so
the local conclusions of Corollary~loc:chars also hold.


# Propagation of automorphy, descent, and the CM case

Fix tin U(K) from Proposition~res:special, taking the local opens of
Proposition~loc:opens, and write B=B_t. The four systems
V_eta,V_chi,V_psi,V_1 have the congruences of
Corollary~sys:congruences, the local properties of
Corollary~loc:chars, and the residual properties of
Section~sec:res. We first explain how automorphy at one coefficient prime
gives what is needed at the next.

## Coefficient primes and local automorphic conditions


Suppose a weight-zero regular algebraic cuspidal representation pi of
GL_2(A_F) realizes one member of the compatible system V_theta. Then it
realizes every member, with the coefficient embeddings fixed in
Section~sec:sys.


Proof. 
Outside a finite set and away from the coefficient prime, the characteristic
polynomial of Frobenius on V_theta is independent of that prime, as an
algebraic polynomial with the fixed embeddings of character values; the
good-place characteristic polynomials of r_iota(pi) have the same
compatibility. Equality at the original prime identifies the polynomials at
every other prime. Both representations are semisimple: for V_theta this is
Faltings' theorem as in Proposition~sys:properties, and automorphic Galois
representations are semisimple. Chebotarev and Brauer--Nesbitt give the
isomorphism.


The next lift also needs the local automorphic component at the new prime. We
recover it by applying Varma's comparison at a different coefficient prime,
where it applies directly.


Suppose a weight-zero regular algebraic cuspidal representation pi of
PGL_2(A_F) realizes V_theta. Let lin{3,p} and vmid l, with
J_t=Jac(C_n(t)).
[label=(roman*)]
 If J_t has potentially good reduction at v, then
rec_{F_v}(pi_v) has zero monodromy.
 If J_t has semistable reduction of toric rank n at v, then pi_v
is a quadratic or trivial twist of Steinberg and pi is iota_l-ordinary of
weight zero at v.



Proof. 
Choose a coefficient prime a different from the residue characteristic l.
Lemma~auto:transport identifies r_{iota_a}(pi) with V_{theta,a}.
Varma's local-global compatibility identifies the underlying semisimplified
Weil representations of its local Weil--Deligne parameter and
rec^T_{F_v}(pi_v) cite[Theorem~1]{Varma}; we use this for the Weil
representations first, not assuming equality of monodromy operators.

In the potentially good case, pass to an extension where the Jacobian has good
reduction. All Frobenius eigenvalues on its H^1, hence on the character
summand, have complex absolute value the square root of the residue
cardinality; taking a positive power of Frobenius shows this also before the
extension. A rank-two Weil--Deligne representation with nonzero monodromy has
Frobenius eigenvalues differing by the norm, which cannot have equal absolute
value, so the automorphic parameter has zero monodromy. The Tate twist in
(eq:tate) does not change this.

In the totally toric case, Corollary~loc:chars gives nonzero monodromy on
V_{theta,a}, so its semisimplified Weil representation is a sum of two
characters differing by the norm. Under local Langlands for GL_2 the only
possibilities are the special parameter with nonzero monodromy and the
zero-monodromy parameter of a one-dimensional local representation; the latter is
not generic, and every local component of a cuspidal representation is generic.
So pi_v is a twist of Steinberg, and trivial central character makes the
twisting character have square one.

The local computation in the proof of cite[Lemma~6.1.3(4)]{CN}, invoking
cite[Lemma~5.6]{Geraghty}, shows a quadratic twist of Steinberg is
iota_l-ordinary of weight zero. It uses only the stated local representation
and trivial central character, so it applies also when the quadratic character is
ramified: the ordinary condition is imposed at a deep level, and a finite-order
twist multiplies normalized Hecke eigenvalues by units.


## The three lifts


The elliptic curve B/F is modular: H^1(B) is realized by a cuspidal
weight-zero regular algebraic representation of PGL_2(A_F).


Proof. 
The order-three system V_eta is H^1(A) for the elliptic curve A/F of
Proposition~sys:elliptic. It is non-CM by Proposition~res:special,
its mod-5 representation is decomposed generic and absolutely irreducible
over F(zeta_5) by the (m,l)=(3,5) part of that proposition, and
zeta_5notin F by (sys:nozeta). So Theorem~aut:seed supplies a
cuspidal weight-zero representation of PGL_2(A_F) realizing V_eta.

Apply Theorem~aut:lift in the order

{@{}cccc@{}}

Input & Coefficient prime & Lifted & Residual input

V_eta & p & V_chi & (m,l)=(3,p)
V_chi & 3 & V_psi & (m,l)=(p,3)
V_psi & p & V_1 & H^1(B)bmod p



In each row Corollary~sys:congruences identifies the residual
representations and Proposition~res:special gives decomposed genericity and
absolute irreducibility over the cyclotomic extension;
Proposition~sys:properties supplies the determinant, finite ramification,
potential semistability, and labeled Hodge--Tate weights; and
Lemma~auto:transport makes the previous automorphy available at the next
coefficient prime.

It remains to check the local hypotheses at that prime. At a place of potentially
good ordinary type, both systems are potentially crystalline and potentially
ordinary of weight zero; at a place of potentially good supersingular type, both
are potentially crystalline and neither is potentially ordinary of weight zero
(these are Corollary~loc:chars). In either case
Lemma~auto:autolocal(i) gives zero monodromy in the input automorphic
parameter. At a place of potentially totally toric type, neither Galois
representation is potentially crystalline and Lemma~auto:autolocal(ii)
gives automorphic ordinarity. So the relevant alternative of
Theorem~aut:lift holds at every place above the coefficient prime.

The theorem thus applies in all three rows. Its conclusion preserves cuspidality,
weight zero, and trivial central character, so each row feeds the next, and the
last gives automorphy of V_1=H^1(B).


## Descending to the original field

The character adjustment in the next argument identifies the descended
representation with the prescribed one.


Let L'/L be a cyclic extension of prime degree between CM fields and
r:G_LtoGL_2(overlineQ_l) continuous with absolutely irreducible restriction
to G_{L'}. If this restriction is realized by a cuspidal weight-zero regular
algebraic representation Pi of GL_2(A_{L'}), then r is realized by a
cuspidal weight-zero regular algebraic representation of GL_2(A_L).


Proof. 
For sigmainGal(L'/L) the representations Pi^sigma and Pi have the
same good-place Hecke polynomials, since the Galois representation extends to
G_L; by strong multiplicity one Pi is Galois invariant, and the cyclic
base-change theorem gives a cuspidal pi of GL_2(A_L) with base change
Pi [cite]. Every archimedean place is complex, so
local extensions are trivial and pi is regular algebraic of weight zero.

Good-place compatibility and Chebotarev give
(r_iota(pi)|_{G_{L'}})^{ss}~= r|_{G_{L'}}. The right side is
absolutely irreducible, so the left restriction is already irreducible and
isomorphic to it. After choosing this isomorphism the two representations agree
on G_{L'}; at gin G_L their matrices differ by a scalar (same conjugation on
the absolutely irreducible G_{L'}-representation), and Schur's lemma makes
these scalars a character delta:Gal(L'/L)tooverlineQ_l^times with
r_iota(pi)~= r ⊗ delta. The finite-order Hecke character
delta^{-1} removes the discrepancy, preserving cuspidality and weight zero,
with Galois-twisting compatibility at good places and hence everywhere by
Chebotarev.


Since F/K is abelian, choose a tower from K to F with cyclic prime-degree
steps; each intermediate field is abelian over Q and contains the imaginary
quadratic K, hence is CM. The representation H^1(B,overlineQ_p) stays
absolutely irreducible over F because its residual representation is
absolutely irreducible there. Apply Lemma~auto:descent down the tower to
Proposition~auto:overF: this gives automorphy of B/K. Finally B is the
quadratic twist of E by h(t), so twisting the automorphic representation by
the same quadratic Hecke character gives a cuspidal weight-zero pi_E
realizing H^1(E).

## Local parameters and the CM case

For non-CM E, Theorem~aut:compat shows pi_E has trivial central
character, realizes H^1(E) at every coefficient prime, and has the required
Frobenius-semisimplified Weil--Deligne parameter at each finite place (choosing
a coefficient prime different from the residue characteristic). At infinity a
cuspidal representation is generic, so the archimedean classification with
weight-zero infinitesimal character gives exactly (eq:inf); the
finite-dimensional alternative is not generic. Equations (eq:tate) and
(eq:Lshift) give the stated normalization.

Suppose now E has complex multiplication by an imaginary quadratic M. Its
endomorphisms are defined over KM (their faithful action on invariant
differentials has eigenvalues in M fixed by G_{KM}). The main theorem of
complex multiplication expresses the two one-dimensional constituents of
H^1(E)|_{G_{KM}} as realizations of algebraic Hecke characters with infinity
types given by the CM type [cite].

If Mnotsubset K, Galois exchanges the two characters and H^1(E) is induced
from either over KM/K; automorphic induction gives the required representation
of GL_2(A_K). If M=K, use the isobaric sum of the two characters. In either
case normalize by |det|^{1/2} so the Tate-normalized parameter realizes
H^1(E); complex multiplication and local class field theory, together with
local compatibility of quadratic automorphic induction in the first case, give
the finite parameters and infinity types. Their central character is trivial:
the determinant of H^1(E) is eps_l^{-1}, so undoing the Tate normalization
gives determinant one, hence trivial central character.

In both the CM and non-CM cases the finite factors also agree for the
residue-characteristic coefficient prime. For good reduction this is crystalline
comparison. For bad potentially good reduction, the Weil--Deligne inertia
action is finite and nontrivial with zero monodromy and determinant one; its
invariant space cannot be one-dimensional (a semisimple two-dimensional
determinant-one representation with a nonzero fixed vector is trivial), so it has
no invariants and local factor one, nontriviality of inertia following from the
crystalline good-reduction criterion [cite]. In the potentially
multiplicative case the Tate-curve description, with a quadratic twist, gives the
same special Weil--Deligne parameter at every coefficient prime. This completes
the proof of Theorem~thm:main.

{99}
 J.~Arthur and L.~Clozel, Simple Algebras, Base Change, and the Advanced Theory of the Trace Formula, Ann. of Math. Stud. 120, Princeton Univ. Press, 1989.
 L.~Bianchi, Sui gruppi di sostituzioni lineari con coefficienti appartenenti a corpi quadratici immaginar^{i}, Math. Ann. 40 (1892), 332--412.
 C.~Breuil, B.~Conrad, F.~Diamond, R.~Taylor, On the modularity of elliptic curves over Q: wild 3-adic exercises, J. Amer. Math. Soc. 14 (2001), 843--939.
 S.~Bosch, W.~L"utkebohmert, M.~Raynaud, N'eron Models, Ergeb. Math. Grenzgeb. 21, Springer, 1990.
 A.~Caraiani and J.~Newton, On the modularity of elliptic curves over imaginary quadratic fields, preprint arXiv:2301.10509v3 (2025).
 R.~Coleman and A.~Iovita, The Frobenius and monodromy operators for curves and abelian varieties, Duke Math. J. 97 (1999), 171--215.
 G.~Faltings, Endlichkeitss"atze f"ur abelsche Variet"aten "uber Zahlk"orpern, Invent. Math. 73 (1983), 349--366.
 D.~Geraghty, Modularity lifting theorems for ordinary Galois representations, Math. Ann. 373 (2019), 1341--1427.
 M.~Harris, K.-W.~Lan, R.~Taylor, J.~Thorne, On the rigid cohomology of certain Shimura varieties, Res. Math. Sci. 3 (2016), Paper No. 37.
 R.~Langlands, Base Change for GL(2), Ann. of Math. Stud. 96, Princeton Univ. Press, 1980.
 P.~Scholze, On torsion in the cohomology of locally symmetric varieties, Ann. of Math. 182 (2015), 945--1066.
 J.-P.~Serre, Topics in Galois Theory, 2nd ed., Res. Notes Math. 1, A K Peters, 2008.
 J.-P.~Serre, Propri'et'es galoisiennes des points d'ordre fini des courbes elliptiques, Invent. Math. 15 (1972), 259--331.
 A.~Grothendieck, M.~Raynaud, D.~S.~Rim, Groupes de monodromie en g'eom'etrie alg'ebrique I (SGA 7 I), Lecture Notes in Math. 288, Springer, 1972.
 P.~Deligne, N.~Katz (eds.), Groupes de monodromie en g'eom'etrie alg'ebrique II (SGA 7 II), Lecture Notes in Math. 340, Springer, 1973.
 G.~Shimura and Y.~Taniyama, Complex Multiplication of Abelian Varieties and Its Applications to Number Theory, Publ. Math. Soc. Japan 6, Math. Soc. Japan, 1961.
 T.~Tsuji, p-adic 'etale cohomology and crystalline cohomology in the semi-stable reduction case, Invent. Math. 137 (1999), 233--411.
 I.~Varma, Local-global compatibility for regular algebraic cuspidal automorphic representations when ellne p, Forum Math. Sigma 12 (2024), Paper No. e21.
 S.~Bloch and K.~Kato, L-functions and Tamagawa numbers of motives, in The Grothendieck Festschrift I, Progr. Math. 86, Birkh"auser, 1990, 333--400.