by
Mar elMohyla& GaborWiese
Abstra t. The arti le motivates,presentsand des ribeslarge omputer al ulations
on- erningtheasymptoti behaviourofarithmeti propertiesof oe ienteldsofmodularforms.
Theobservationssuggest ertainpatterns,whi hdeservefurtherstudy.
Résumé. Le but de et arti le est de motiver, présenter et dé rirede nombreux al uls
menés sur ordinateur on ernant le omportement asymptotique de propriétés arithmétiques
des orpsdes oe ientsdeformesmodulaires.Lesobservationssuggèrent plusieursquestions
quiméritentd'êtreétudiéesultérieurement.
1. Introdu tion
A re ent breakthrough in Arithmeti Geometry is the proof of the Sato-Tate onje ture by
Barnet-Lamb, Clozel, Geraghty, Harris, Shepherd-Barron and Taylor ([BLGHT℄, [CHT℄,
[HSHT℄, [T℄). It states that the normalised He ke eigenvalues
a
p
(f )
2p
(k−1)/2
on a holomorphi newformf
ofweightk ≥ 2
(andtrivialDiri hlet hara ter(1)
)areequidistributedwithrespe t
toa ertainmeasure(theso- alledSato-Tatemeasure),when
p
runsthroughthesetofprimenumbers. Thename horizontalSato-Tateis sometimesusedfor this situation.
The reversed situation, to be referred to as verti al horizontal Sato-Tate, was su essfully
treated by Serre in [S℄. He xes a prime
p
and allows any sequen e of positive integers(Nn, kn)
withevenkn
andp ∤ Nn
su hthatNn
+ kn
tendstoinnityandprovesthata
p
(f )
2p
(k−1)/2
2000 Mathemati sSubje tClassi ation. 11F33(primary);11F11,11Y40.
Key words and phrases. Modularforms, oe ient elds,asymptoti behaviour, ongruen es, He ke
algebras.
G.W.wouldliketothankGabrielChênevertforinspiringdis ussionsofvariantsofSato-Tate. Morethanksare
duetoPierreParent,ChristopheRitzenthaler,RenéS hoofandMarkWatkinsforinterestingdis ussionsand
suggestions. G.W.a knowledgespartialsupportbytheEuropeanResear hTrainingNetworkGaloisTheory
and Expli it Methods MRTN-CT-2006-035495 and by the Sonderfors hungsberei h Transregio 45 Periods,
modulispa esandarithmeti ofalgebrai varieties oftheDeuts heFors hungsgemeins haft.
(1)
isequidistributed with respe t to a ertain measuredepending on
p
(whi h is related to theSato-Tatemeasure),when
f
runsthroughallthenewforms inanyofthespa es of uspformsoflevel
Γ
0
(N
n
)
and weightk
n
. A orollaryisthat forxed positive even weightk
,theset(1.1)
{[Q
f
: Q] | f
newformof levelΓ
0
(N
n
)
and weightk}
is unbounded for any sequen e
N
n
tending to innity. Here,Q
f
denotes the number eldobtainedfrom
Q
by adjoining allHe ke eigenvalues onf
.Inthisarti leweperformarst omputationalstudytowardsa(weak)arithmeti analogueof
verti alSato-Tate,wherethenamearithmeti referstotakinganitepla eof
Q
,asopposedtotheinnitepla eusedinusualSato-Tate(theassertionofusualSato-Tate on ernstheHe ke
eigenvalues as real numbers). An arithmeti analog of horizontal Sato-Tate is Chebotarev's
density theorem. Consider, for example, a normalised uspidal He ke eigenform
f
withat-ta hed Galois representation
ρ
f
: Gal(Q/Q) → GL
2
(Z
ℓ
)
.(2)
Fix some
x ∈ Z
ℓ
and letn ∈ N
.Let
G
be the image of the omposite representationGal(Q/Q)
ρ
f
−→ GL
2
(Z
ℓ) ։ GL2
(Z/ℓ
n
Z)
andlet
d(x)
bethenumberofelements inG
withtra eequal tox
moduloℓ
n
. Thenthe
den-sity of the set
{p | |a
p
− x|
ℓ
≤ ℓ
−n
}
isequal to
d(x)
|G|
byChebotarev'sdensity theorem; hen e, thesituation is ompletely lear. Whereasat theinnite pla e horizontal Sato-Tateseemstobe more di ultthan verti al Sato-Tate, the situation appearsto be reversed for arithmeti
analogs. We are not goingto proposesu h an analogue. But, we aregoing to study related
questions by means of omputer al ulations. For instan e, as a motivation let us onsider
theset
(1.2)
{[F
ℓ,f
: F
ℓ
] | f ∈ S
k
(N
n
; F
ℓ
)
normalisedHe ke eigenform}
inanalogyto Equation1.1 . Here,
S
k
(N
n
; F
ℓ
)
denotestheF
ℓ
-ve torspa eof uspidalmodularforms over
F
ℓ
(see Se tion 2 for denitions) andF
ℓ,f
is dened by adjoining toF
ℓ
all He keeigenvalues on
f
. It is easy to onstru t sequen es(N
n
, k
n
)
for whi h theset inquestion isinnite(seee.g. [DiWi℄and[W℄), butitdoesnot seemsimple toobtain allnaturalnumbers
asdegrees. Mostimportantly,itseemsto beunknownwhetherthis setis innitewhen
(Nn)
isthesequen eof prime numbers,
k = 2
andℓ > 2
.Con erningproperties of modularforms inpositive hara teristi , thereis other, mu h more
subtleinformation thanjustthedegrees of
F
ℓ,f
tobestudied,e.g. ongruen esbetweenmod-ularforms. Inorder to take thefull information into a ount, inthis arti le we examine the
properties of the
F
ℓ
-He ke algebrasT
k
(Nn)
onS
k
(Nn; F
ℓ
)
asymptoti ally for xed weightk
(mostly
2
) and running levelN
n
(mostly the set ofprime numbers) by means ofexperimen-tation. Morepre isely,weinvestigate three quantities:
(a) The deviation of
T
k
(N
n
)
from being semisimple. In Se tion 3 , we in lude a propositionrelating nonsemisimpli ity to ongruen es, rami ation and ertain indi es. Our
exper-iments suggest that for odd primes
ℓ
, the He ke algebraT
k
(N
n
)
tends to be lose to(2)
semisimple, whereas the situation seems to be ompletely dierent for
p = 2
(seeSe -tion 4.1).
(b) The average residue degree of
T
k
(Nn)
. That is the arithmeti mean of the degrees ofF
ℓ,f
for allf
inS
k
(N
n
; F
ℓ
)
. Our omputations (see Se tion 4.2 ) strongly suggest that this quantity is unbounded. More pre isely, we seem to observe a ertain asymptotibehaviour,whi h we formulate asaquestion.
( ) The maximum residue degree of
T
k
(N
n
)
. That is the maximum of the degrees ofF
ℓ,f
for all
f
inS
k
(N
n
; F
ℓ
)
. Our experiments suggest that this quantity is 'asymptoti ally'proportional to the dimension of
S
k
(N
n
; F
ℓ
)
.InSe tion4wedes ribeour omputationsandderive ertainquestionsfromourobservations.
However, we do not attempt to propose any heuristi explanations in this arti le. This will
havetobethe subje tofsubsequent studies,buildingonrenedandextended omputations.
We see (b ) and ( ) as strong eviden e for the innity of the set in Equation 1.2, when
N
n
runs through the primes. Generally speaking, there appears to be some regularity in the
otherwise quite errati behaviour of the examined quantities, lending some support to the
hopeof ndinga formulationof an arithmeti analogue ofverti alSato-Tate.
2. Ba kground and notation
We start by introdu ing the ne essary notation and explaining the ba kground. For fa ts
on modular forms, we refer to [DI℄. Let us x an interger
k
and a ongruen e subgroupΓ ⊆ SL
2
(Z)
. DenotebyS
k
(Γ)
the omplexve torspa eofholomorphi uspformsofweightk
for
Γ
. DeneT := T
k
(Γ)
to be theZ
-He ke algebra of weightk
forΓ
, i.e. the subring ofEndC(S
k
(Γ))
spanned by all He ke operatorsTn
forn ∈ N
. IfΓ = Γ0(N )
we simply writeT
k
(N )
. Weuse similarnotation in other ontexts, too. It isan important theoremthatT
is freeof niterankasaZ
-module, hen ehasKrulldimension one asa ring, andthatthemap(2.3)
HomZ(T, C) → S
k
(Γ), φ 7→
∞
X
n=1
φ(T
n
)q
n
withq = q(z) = e
2πiz
denes an isomorphism of
C
-ve tor spa es, whi h is ompatible withthenatural He ke a tion. For anyring
A
,deneS
k
(Γ; A) := HomZ(T, A)
equipped withthenatural He kea tion (i.e.
T
-a tion), sothatwehaveS
k
(Γ) ∼
= Sk
(Γ; C)
. Wethink ofelementsin
S
k
(Γ; A)
intermsofformalq
-expansions, i.e.asformalpowerseriesinA[[q]]
,byananalogof Eq. 2.3 . Note that normalised He ke eigenforms, i.e. those
f =
P∞
n=1
a
n
q
n
∈ S
k
(Γ; A)
that satisfy
a
1
= 1
andT
n
f = a
n
f
,pre isely orrespond to ring homomorphismsφ : T → A
with
φ(Tn) = an
. WhenA
is an integral domain, a normalised eigenfun tion gives rise toa prime ideal
p
ofT
, namely the kernel ofφ
. We may think ofT/p
as thesmallest subringof
A
generated bythea
n
forn ∈ N
: the oe ient ring off
inA
. Note thatAut(A)
a tsonS
k
(Γ; A)
by omposingφ : T → A
withσ ∈ Aut(A)
. Obviously,this a tion doesnot hangeWexaprimenumber
p
. WeputT := b
b
T
k
(Γ) := T
k
(Γ) ⊗Z
Z
p
,T
b
η
:= T
k
(Γ) ⊗Z
Q
p
= b
T ⊗Z
p
Q
p
and
T := T
k
(Γ) := T
k
(Γ)⊗
Z
F
p
= b
T⊗
Z
p
F
p
. NotetheisomorphismsS
k
(Γ; Z
p
) ∼
= HomZ
p
(b
T, Z
p
)
,S
k
(Γ; Q
p
) ∼
= HomZ
p
(b
T, Q
p
)
andS
k
(Γ; Fp) ∼
= HomZ
p
(b
T, Fp) ∼
= HomF
p
(T, Fp)
.The
Gal(Q
p
/Qp)
- onjuga y lassesofnormalisedeigenformsinS
k
(Γ; Q
p
)
(bywhi hwe meanthe lasses for the
Gal(Q
p
/Q
p
)
-a tion des ribed above) arein bije tion withtheprime (and automati allymaximal)idealsofT
b
η
andalsoinbije tionwiththeminimalprimeidealsofT
b
,whosesetisdenotedby
MinSpec(b
T)
. These ondbije tionisexpli itlygivenbytakingpreim-agesfortheinje tion
T ֒→ b
b
Tη
. NotethatT
b
hasKrulldimensionone,meaningthatanyprimeideal is either minimal (i.e. not ontaining any smaller prime ideal) or maximal. Moreover,
the
Gal(F
p
/F
p
)
- onjuga y lasses of normalisedeigenforms inS
k
(Γ; F
p
)
areinbije tion withSpec(T) = MaxSpec(T)
. Furthermore,Spec(T)
is innatural bije tion withMaxSpec(b
T)
un-der takingpreimages for the natural proje tion
T ։ T
b
. By a resultin ommutative algebra,wehave dire tprodu tde ompositions
b
T =
Y
m∈MaxSpec(b
T)
b
Tm, T =
Y
m∈MaxSpec(b
T)
Tm
andT
b
η
=
Y
p∈Spec(b
T
η
)
b
T
η,p
,
where thefa torsarethe lo alisations at theprime idealsindi atedbythesubs ripts.
Denition 2.1. We say that two
p
1
, p
2
∈ MinSpec(b
T)
are ongruent if they lie in thesame maximalideal
m
∈ MaxSpec(b
T)
. Forp
∈ MinSpec(b
T)
, we allT/p
b
thelo al oe ientring and
L
p
:= Frac(b
T/p)
the lo al oe ient eld. We say thatp
∈ MinSpec(b
T)
is ramiedif
L
p
isa ramied extension ofQ
p
. We denote byi
p
the index ofT/p
b
in thering of integersof
Lp
. The residue eldT/m ∼
b
= T/m
will be denoted byFm
and will be alled the residualoe ient eld.
Wenowestablishthe onne tionwiththe usualunderstanding ofthetermsinthedenition.
Let
Z ⊂ Q ⊂ C
be the algebrai integers andthe algebrai numbers, respe tively. AsT
isofnite
Z
-rank, theset of normalisedeigenforms inS
k
(Γ)
isthe same astheset of normalisedeigenforms in
S
k
(Γ; Z)
. FixhomorphismsZ
ι
//
π
=
=
=
=
=
=
=
=
=
Zp
π
F
p
giving rise toS
k(Γ; Z)
ι
//
π
%% %%K
K
K
K
K
K
K
K
K
K
K
S
k(Γ; Zp
)
π
S
k
(Γ; F
p
)
.
From this perspe tive, a holomorphi normalised He ke eigenform
f =
P∞
n=1
a
n
q
n
∈ S
k
(Γ)
gives rise to an eigenform in
S
k
(Γ; F
p
)
, alled the redu tion off
modulop
, whose formalq
-expansionis
π(f ) :=
P∞
n=1
π(an)q
n
∈ Fp[[q]]
. Theredu tion orrespondstom
∈ MaxSpec(b
T)
and to
m
∈ Spec(T)
(we use the same symbol due to the natural bije tion between thetwo sets). The form
f
also gives rise to an eigenform inS
k
(Γ; Q
p
)
, whi h orresponds top
f
∈ MinSpec(b
T)
and top
f
∈ Spec(b
T
η
)
(the same symbolis used again due to the naturalLet
g =
P∞
n=1
b
n
q
n
beanotherholomorphi normalisedHe keeigenform. Ifπ(f ) = π(g)
,then learlyp
f
⊂ m
andp
g
⊂ m
,i.e.p
f
andp
g
are ongruent. Conversely,letp
f
, p
g
∈ MinSpec(b
T)
su h that
p
f
⊂ m
andp
g
⊂ m
for somem
∈ MaxSpec(b
T)
, so thatp
f
andp
g
are ongruent.The ideals
p
f
andp
g
orrespond toGal(Q
p
/Q
p
)
- onjuga y lasses inS
k
(Γ; Q
p
)
and we anhoose
f, g ∈ S
k
(Γ; Z
p
)
orresponding top
f
andp
g
withπ(f ) = π(g)
. We illustrate thesituation by thediagram
b
T/p
f
(( ((R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
_
b
T
88 88q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
&& &&M
M
M
M
M
M
M
M
M
M
M
M
M
M
M
f
,,
g
22
Z
p
π
// //
F
p
oo
? _
T/m = Fm.
b
T/p
g
66 66
l
l
l
l
l
l
l
l
l
l
l
l
l
l
l
l
l
l
?
OO
Notethat
f, g
an alreadybefoundinS
k
(Γ; Z) ⊂ S
k
(Γ)
. Thisjustiesour usageof thetermongruen e.
Moreover, the lo al oe ient ring
T/p
b
an be identied withZ
p,f
:= Z
p
[ι(a
n
)|n ∈ N]
andits fra tion eld
Lp
withQ
p,f
:= Qp(ι(an)|n ∈ N)
, when eip
is the index ofZ
p,f
in itsnormalisation. Furthermore, the residual oe ient eld, i.e.
Fm
= T/m
, an be interpretedas
F
p
[π(a
n
)|n ∈ N]
. The relation to thearithmeti of the oe ient eldQ
f
:= Q(a
n
|n ∈ N)
and the oe ient ring
Z
f
:= Z[an|n ∈ N]
is apparent.In order to on lude this ba kground se tion, we point out that in the ase
k = 2
, theoe ient ring
Z
f
is theendomorphism ring of theabelian varietyA
f
atta hed tof
. Fromthatpointof view,thefollowing analysis analso be interpretedasastudy ofthearithmeti
of the endomorphismalgebrasof
GL
2
-abelian varieties.3. Semisimpli ityof He ke algebras
We re allthata nitedimensional ommutative
K
-algebra,whereK
is aeld, issemisimpleif and only ifit is isomorphi to a dire t produ tof elds (whi h are ne essarily nite eld
extensionsof
K
).In this se tion we rst study the semisimpli ity of the He ke algebra
T
b
η
. In the ase whenit is semisimple, we relate the non-semisimpli ity of the mod
p
He ke algebraT
to threephonomena: ongruen es between
Gal(Q
p
/Qp)
- onjuga y lasses of newforms, rami ationat
p
of the oe ient elds of newforms and thep
-index of the lo al oe ient ring in theringof integersof thelo al oe ient eld.
Let
f =
P∞
n=1
a
n
(f )q
n
∈ S
k
(Γ
1
(M ))
new
be a normalised He ke eigenform and letm
be anypositiveinteger. Wedene the
C
-ve torspa eV
f
(m)
tobe thespanof{f (q
d
) | d | m}
d
runs throughall positive divisorsofm
(in luding1
andm
). Newform theorystatesthatS
k
(Γ
1
(N )) ∼
=
M
m|N
M
f ∈S
k
(Γ
1
(N/m))
new
V
f
(m).
Thisis anisomorphism of He ke modules. TheHe ke operators
Tn
for(n, m) = 1
restri tedto
V
f
(m)
are s alar matri es witha
n
(f )
as diagonal entries. We now des ribe the He keoperator
T
ℓ
onV
f
(m)
for a primeℓ
. Suppose thatℓ
r
|| m
. Let
ǫ
be theDiri hlet hara terof
f
. Consider the(r + 1) × (r + 1)
-matrixA := A
f
(m, ℓ) :=
a
ℓ
(f )
1
0 0 . . . 0
−δǫ(ℓ)ℓ
k−1
0
1 0 . . . 0
0
0
0 1 . . . 0
. . . . . .0
. . . 0 0
0
1
0
. . . 0 0
0
0
,
where
δ = 0
ifℓ | (N/m)
andδ = 1
otherwise. TheHe ke operatorT
ℓ
isthengiven onV
f
(m)
(fora ertainnatural basis)byadiagonal blo kmatrix having onlyblo ks equalto
A
onthediagonal, where ea h blo k on thediagonal orresponds to a divisor of
m/ℓ
r
. Let
T
be theHe kealgebraof
S
k
(Γ
1
(N ))
(asinSe tion 2). ThealgebraTQ
:= T ⊗Z
Q
issemisimpleifandonlyif
TC
:= T ⊗Z
C
issemisimple(ifandonlyifb
T
η
issemisimple). Bytheabovedis ussion,T
C
is semisimpleifand onlyifall thematri esA
f
(m, ℓ)
thatappeararediagonalisable. Proposition 3.1. Assume the notationabove,M = N/m
andk ≥ 2
. Moreover, ifk ≥ 3
assumeTate's onje ture (see [CE ℄, Se tion 1).
(a) Assume
ℓ ∤ M
. ThenA
f
(m, ℓ)
isdiagonalisable if andonlyifr ≤ 2
.(b) Assume that
ℓ | M
and that eitherℓ || M
or thatǫ
annot be dened modM/ℓ
. ThenA
f
(m, ℓ)
is diagonalisableif and onlyifr ≤ 1
.( ) Assumethat
ℓ
2
| M
andthat
ǫ
anbedenedmoduloM/ℓ
. ThenA
f
(m, ℓ)
isdiagonalisableifand onlyif
r = 0
.Proof. (a)Assume
r ≥ 1
(otherwise theresultistrivial)and allB
thetopleft2 × 2
-blo kof
A = A
f
(m, ℓ)
. The hara teristi polynomial ofB
isg(X) = X
2
− a
ℓ
(f )X + ǫ(ℓ)ℓ
k−1
.We have
g(0) 6= 0
andg(X)
has dis riminanta
ℓ
(f )
2
− 4ǫ(ℓ)ℓ
k−1
, whi h is non-zero, sin e
|a
ℓ
(f )| = 2|ℓ|
(k−1)/2
would ontradi t [CE℄, Theorem 4.1. Consequently,A
is diagonalisableifand only ifapartfrom
B
there isat most one morerow and olumn.In ases (b)and( ), notethat
A
isinJordanform. Theresultisnowimmediate, sin ea
ℓ
(f )
isnon-zero for (b)andzero for ( )(see [DS℄,1.8).
Wehave theimmediate orollary (whi h isTheorem 4.2in[CE℄).
Corollary 3.2. Let
N
be ubefree andk ≥ 2
. Ifk ≥ 3
assumeTate's onje ture (see[CE℄,Se tion 1). Then the He ke algebras
T
k
(Γ
1
(N )) ⊗ Q
andT
b
k
(Γ
1
(N ))
η
as well asT
k
(N ) ⊗ Q
Theprin ipalresultofthisse tionisthefollowing propositiononthestru tureoftheresidual
He ke algebra. We assume thenotation laid out in Se tion 2, inparti ular, we work with a
general ongruen e subgroup
Γ
.Proposition 3.3. Assume that
T
b
η
is semisimple (see, e.g., Corollary 3.2), i.e.b
T
η
∼
=
Y
p∈MinSpec(b
T)
Lp.
Then the residual He ke algebra
T
issemisimple if and onlyif all of the followingthreeon-ditionsare satised:
(i) No two
p
1, p2
∈ MinSpec(b
T)
are ongruent.(ii) None of the
p
∈ MinSpec(b
T)
is ramied.(iii) Forall
p
∈ MinSpec(b
T)
, the indexip
= 1
.Proof. We rst prove that(i),(ii)and (iii)imply the semisimpli ityof
T
.Thefa t thatthere isno ongruen e means thatinevery
m
∈ MaxSpec(b
T)
there isa uniquep
∈ MinSpec(b
T
m
)
. AsT
b
m
⊗
Z
p
Q
p
∼
= Lp
,itfollows thatT
b
m
isasubring ofL
p
. Dueto (iii),T
b
m
isthe ringofintegersof
Lp
. Sin eby(ii)Lp
isunramied, we getthatTm
istheresidueeldof the integersof
Lp
. Thisshows thatT
isa produ tofnite elds,i.e.semisimple.Nowweprovethe onversedire tion andassumethat
T
issemisimple. Letm
∈ MaxSpec(b
T)
.Let
p
1
, p
2
, . . . , p
m
∈ MinSpec(b
T)
bethedistin tminimalprimes ontainedinm
. ThenT
b
m
⊗
Z
p
Qp
∼
= Lp
1
× · · · × Lp
m
. Dueto thenon-degenerationTm
∼
= b
Tm⊗Z
p
Fp
∼
= Fp
n
forsomen
. Sin edimQ
p
Tm
b
⊗Z
p
Q
p
= n
,we have[Lp
i
: Q
p
] ≤ n
fori = 1, . . . , m
.Let
O
i
be the ring of integers ofL
p
i
fori = 1, . . . , m
. It ontainsT
b
m
/p
i
with indexi
p
i
. Tensoring theexa t sequen eofZ
p
-modules0 → b
Tm/p
i
→ O
i
→ O
i
/(b
Tm/p
i
) → 0
with
F
p
overZ
p
we obtaintheexa t sequen eofF
p
-ve tor spa es:F
p
n
→ O
i
⊗Z
p
F
p
→ (O
i
/(b
Tm/p
i
)) ⊗Z
p
F
p
→ 0.
Sin e the mapon theleft isa ring homomorphism,itis inje tive. Now
dimF
p
O
i
⊗Z
p
F
p
≤ n
implies that
O
i
is unramied and thati
p
i
= 1
. Thus[L
p
i
: Q
p
] = n
fori = 1, . . . , m
and,hen e,
m = 1
, on ludingtheproof.4. Observations and Questions
In this se tion, we explain and expose our omputer experiments and we ask some
ques-tions suggestedbyour studies. All omputer al ulationswereperformedusingMagma (see
4.1. Semisimpli ityof the residualHe ke algebra. Alo alnite-dimensional
om-mutative
F
p
-algebraA
issemisimpleifand onlyifitissimple,whi hisequivalenttoA
beingeld. Wetakethedimensionofthe maximalideal
m
ofA
asameasureforthedeviationofA
frombeing semisimple. Inparti ular,
A
isa eldifand onlyifm
hasdimension0
.Forgiven prime
p
,levelN
and weightk
we study thesumof the residue degrees ofall primeideals:
a
(p)
N,k
=
X
m∈Spec(T
k
(N ))
[Fm
: F
p
].
Clearly,a
(p)
N,k
islessthanorequaltotheF
p
-dimensionofS
k
(N ; F
p
)
. Hen e,T
k
(N )
is semisim-pleifandonly ifa
(p)
N,k
isequal to this dimension.Weintendtostudytheasymptoti behaviourofthefun tion
a
(p)
N,k
foraxedprimep
andxed weightk
asafun tionofthelevelN
. Forsimpli ity,weletN
runthroughtheprimenumbersonly in order to avoid ontributions from lower levels via the degenera y maps. We should
point outthatthere anbe ontributions fromlowerweights: aneigenform in
S
k
(N ; F
p
)
alsolivesin
S
k+n(p−1)
(N ; F
p
)
foralln ≥ 0
bymultipli ation bytheHasseinvariant. Notethatforp > 2
andk = 2
,aswell asforp > 3
andk = 4
there isno su h ontribution.Our omputational ndings are best illustrated by plotting graphs. In ea h of the
follow-ing plots, the prime
p
and the weightk
are xed and on thex
-axis we plotd(N ) :=
dim
F
p
S
k
(N ; F
p
)
and on the
y
-axis the fun tiona
(p)
N,k
as a fun tion ofN
, i.e. ea hN
gives riseto a dot at theappropriate pla e. Thestraight line inthegraphs wasdetermined asthelinearfun tion
α · d(N )
whi h bestts thedata (a ordingto gnuplot and theleastsquaresmethod).
In the weights that we onsidered we observed a behaviour for
p = 2
whi h seems to beompletelydierentfromthebehaviouratallotherprimes. Wemadeplotsforalloddprimes
lessthan
100
and weight2
and present a sele tion here. The graphs that we leave out lookverysimilar. Figure1
0
20
40
60
80
100
120
140
160
0
20
40
60
80
100
120
140
160
Sum of Residue Degrees
Dimension k = 2, p = 3
x * 0.990097
Figure20
20
40
60
80
100
120
140
160
0
20
40
60
80
100
120
140
160
Sum of Residue Degrees
Dimension k = 2, p = 5
Figure3
0
20
40
60
80
100
120
140
160
0
20
40
60
80
100
120
140
160
Sum of Residue Degrees
Dimension k = 2, p = 7
x * 0.995265
Figure40
20
40
60
80
100
0
20
40
60
80
100
Sum of Residue Degrees
Dimension k = 2, p = 19
x * 0.997344
Figure50
20
40
60
80
100
120
140
160
0
20
40
60
80
100
120
140
160
Sum of Residue Degrees
Dimension k = 2, p = 31
x * 0.999248
Figure60
10
20
30
40
50
60
70
80
0
10
20
30
40
50
60
70
80
Sum of Residue Degrees
Dimension k = 2, p = 41
x * 0.998996
Figure70
10
20
30
40
50
60
70
80
0
10
20
30
40
50
60
70
80
Sum of Residue Degrees
Dimension k = 2, p = 53
x * 0.999397
Figure80
10
20
30
40
50
60
0
10
20
30
40
50
60
Sum of Residue Degrees
Dimension k = 2, p = 73
Figure9
0
10
20
30
40
50
60
0
10
20
30
40
50
60
Sum of Residue Degrees
Dimension k = 2, p = 83
x * 0.999396
Figure100
20
40
60
80
100
120
140
0
20
40
60
80
100
120
140
Sum of Residue Degrees
Dimension k = 2, p = 97
x * 0.999889
InFigures110 thelevelsrangeoverallprimesup toa ertainbound(whi h isnot thesame
for all
p
). We observe that non-semisimpli ity seemsto be a rather rare phenomenon whi hbe omes rarer for growing
p
,asone might have guessed. In thenext gures, we analyse theases
p = 3, 5
still for weight2
more losely by letting the levels range through all primesbetween
3000
and10009
subdivided into four onse utive intervals.Figure11
240
260
280
300
320
340
360
380
240
260
280
300
320
340
360
380
Sum of Residue Degrees
Dimension k = 2, p = 3
x * 0.995086
Figure12380
400
420
440
460
480
500
520
540
380
400
420
440
460
480
500
520
540
Sum of Residue Degrees
Dimension k = 2, p = 3
x * 0.996909
Figure13540
560
580
600
620
640
660
680
540
560
580
600
620
640
660
680
Sum of Residue Degrees
Dimension k = 2, p = 3
x * 0.997720
Figure14680
700
720
740
760
780
800
820
840
680
700
720
740
760
780
800
820
840
Sum of Residue Degrees
Dimension k = 2, p = 3
Figure15
240
260
280
300
320
340
360
380
240
260
280
300
320
340
360
380
Sum of Residue Degrees
Dimension k = 2, p = 5
x * 0.997053
Figure16380
400
420
440
460
480
500
520
540
380
400
420
440
460
480
500
520
540
Sum of Residue Degrees
Dimension k = 2, p = 5
x * 0.997917
Figure17520
540
560
580
600
620
640
660
680
520
540
560
580
600
620
640
660
680
Sum of Residue Degrees
Dimension k = 2, p = 5
x * 0.998676
Figure18680
700
720
740
760
780
800
820
840
680
700
720
740
760
780
800
820
840
Sum of Residue Degrees
Dimension k = 2, p = 5
x * 0.998808
One anobservethattheslopeofthebestttinglinethroughtheoriginseemstobein reasing
with growing dimension. Although we only omputed relatively litte data, we in lude two
examplesfor weight
4
. Theydo not suggestanysigni ant dieren e to theweight2
ase.Figure19
0
50
100
150
200
0
50
100
150
200
Sum of Residue Degrees
Dimension k = 4, p = 5
x * 0.993821
Figure200
50
100
150
200
0
50
100
150
200
Sum of Residue Degrees
Dimension k = 4, p = 7
x * 0.994612
Question 4.1. Fix an odd prime
p
and an evenk ≥ 2
. Leta(N ) := a
(p)
k,N
andd(N ) :=
dim
F
p
S
k
(N ; Fp
)
. Does the following statement hold?For all
ǫ > 0
there isC
ǫ
> 0
su h that for all primesN
the inequalitya(N ) > (1 − ǫ)d(N ) − C
ǫ
holds.
We ontrast the situation, whi h seems very similar for every odd prime, with the one for
p = 2
andk = 2
. We do not onsider any higher weights due to the ontributions fromweight
2
,whi hwould`disturb'thesituation. ThefollowingplotstakeprimenumbersN
intoa ount thatlie insixdierent intervals upto
12000
.Figure21
0
20
40
60
80
100
0
20
40
60
80
100
120
140
160
Sum of Residue Degrees
Dimension k = 2, p = 2
x * 0.521382
Figure22200
220
240
260
280
420
430
440
450
460
470
480
490
500
510
Sum of Residue Degrees
Dimension k = 2, p = 2
x * 0.510937
Figure23240
260
280
300
320
520
540
560
580
600
Sum of Residue Degrees
Dimension k = 2, p = 2
x * 0.509345
Figure24280
300
320
340
360
380
400
620
640
660
680
700
720
Sum of Residue Degrees
Dimension k = 2, p = 2
Figure25
320
340
360
380
400
420
440
460
480
720
740
760
780
800
820
Sum of Residue Degrees
Dimension k = 2, p = 2
x * 0.508675
Figure26440
460
480
500
520
540
560
930
940
950
960
970
980
990
1000
Sum of Residue Degrees
Dimension k = 2, p = 2
x * 0.508181
Inspite oftheveryirregular behaviour, itisremarkablethattheslope ofthebesttting line
throughthe origin isalways justa little biggerthan
1
2
.At the moment we annot fullyexplain this behaviour. Contributions from weight one play
somerole. However,probablymoreimportant are ongruen es offormshavingAtkin-Lehner
eigenvalue
+1
with forms having eigenvalue−1
. As Johan Bosman observes in a remarkin[B℄,itfollows fromthe onne tednessof thespe trumoftheHe ke algebra([M ℄,10.6) for
k = 2
andprimelevels thatthere isalwaysat leastone su h ongruen e forp = 2
,wheneverthe
+1
- andthe−1
-eigenspa e are nonempty.We areledto ask thefollowing question.
Question 4.2. Fixanevenweight
k ≥ 2
. Leta(N ) := a
(2)
k,N
andd(N ) := dim
F
2
S
k
(N ; F
2
)
.Are there
1 > α ≥ β > 0
and onstantsC, D > 0
su h that the inequalityα · d(N ) + C > a(N ) > β · d(N ) − D
holds?
4.2. Average Residue Degree. We now study the average residue degree, whi h we
dene for givenlevel
N
,weightk
and primep
asb
(p)
N,k
=
1
# Spec(T
k
(N ))
X
m∈Spec(T
k
(N ))
[Fm
: F
p
] =
a
(p)
N,k
# Spec(T
k
(N ))
.
We made omputations for weight
2
and all primesp
lessthan100
, whereN
runs throughthe same ranges as previously. We again plot the dimension
d(N )
on thex
-axis and thefun tion
b
(p)
N,k
on they
-axis and the straight line is again the best tting fun tionα · d(N )
, although we believe thatthis isnot the right fun tion totake (seebelow). Weagain presenta sele tion of prime numbers as before, however, in luding
p = 2
from the beginning. TheFigure27
0
5
10
15
20
25
30
35
0
20
40
60
80
100
120
140
160
Average Residue Degree
Dimension k = 2, p = 2
x * 0.098129
Figure280
10
20
30
40
50
0
20
40
60
80
100
120
140
160
Average Residue Degree
Dimension k = 2, p = 3
x * 0.129624
Figure290
5
10
15
20
25
30
35
40
45
0
20
40
60
80
100
120
140
160
Average Residue Degree
Dimension k = 2, p = 5
x * 0.114418
Figure300
5
10
15
20
25
30
35
40
0
20
40
60
80
100
120
140
160
Average Residue Degree
Dimension k = 2, p = 7
x * 0.113730
Figure310
5
10
15
20
0
20
40
60
80
100
Average Residue Degree
Dimension k = 2, p = 19
x * 0.128238
Figure320
5
10
15
20
25
30
0
20
40
60
80
100
120
140
160
Average Residue Degree
Dimension k = 2, p = 31
Figure33
0
2
4
6
8
10
12
14
16
18
0
10
20
30
40
50
60
70
80
Average Residue Degree
Dimension k = 2, p = 53
x * 0.127727
Figure340
2
4
6
8
10
12
14
0
10
20
30
40
50
60
Average Residue Degree
Dimension k = 2, p = 73
x * 0.134275
Figure350
5
10
15
20
0
10
20
30
40
50
60
Average Residue Degree
Dimension k = 2, p = 83
x * 0.141422
Figure360
5
10
15
20
25
30
35
40
0
20
40
60
80
100
120
140
Average Residue Degree
Dimension k = 2, p = 97
x * 0.119130
Hereare againtwo examplesfor weight
4
.Figure 37
0
5
10
15
20
25
0
50
100
150
200
Average Residue Degree
Dimension k = 4, p = 5
x * 0.110425
Figure385
10
15
20
25
30
35
0
20
40
60
80
100
120
140
160
180
Average Residue Degree
Dimension k = 4, p = 11
x * 0.106811
Very roughly speaking the data suggest that the average residue degree grows with the
di-mension, asis ertainlyto be expe ted. Wealso ondu teda loseranalysisfortheprimes
2
,3
and5
. Forp = 2
we used all primes in dierent intervals up to12000
and obtained theseFigure39
15
20
25
30
35
40
45
50
55
60
420
430
440
450
460
470
480
490
500
510
Average Residue Degree
Dimension k = 2, p = 2
x * 0.062388
Figure4020
30
40
50
60
520
540
560
580
600
Average Residue Degree
Dimension k = 2, p = 2
x * 0.057421
Figure4120
40
60
80
100
120
620
640
660
680
700
720
Average Residue Degree
Dimension k = 2, p = 2
x * 0.057158
Figure4220
40
60
80
100
120
140
160
180
200
720
740
760
780
800
820
Average Residue Degree
Dimension k = 2, p = 2
x * 0.055027
Figure4320
40
60
80
100
120
140
160
180
930
940
950
960
970
980
990
1000
Average Residue Degree
Dimension k = 2, p = 2
x * 0.055893
Hereare theplots for
p = 3, 5
and the primes between3000
and10009
subdivided into fourFigure44
20
40
60
80
100
120
240
260
280
300
320
340
360
380
Average Residue Degree
Dimension k = 2, p = 3
x * 0.096554
Figure4520
40
60
80
100
120
140
160
180
380
400
420
440
460
480
500
520
540
Average Residue Degree
Dimension k = 2, p = 3
x * 0.089212
Figure4630
40
50
60
70
80
540
560
580
600
620
640
660
680
Average Residue Degree
Dimension k = 2, p = 3
x * 0.081910
Figure4720
40
60
80
100
120
140
680
700
720
740
760
780
800
820
840
Average Residue Degree
Dimension k = 2, p = 3
x * 0.078536
Figure4820
30
40
50
60
70
80
240
260
280
300
320
340
360
380
Average Residue Degree
Dimension k = 2, p = 5
x * 0.090116
Figure4930
40
50
60
70
380
400
420
440
460
480
500
520
540
Average Residue Degree
Dimension k = 2, p = 5
Figure50
30
40
50
60
70
80
90
520
540
560
580
600
620
640
660
680
Average Residue Degree
Dimension k = 2, p = 5
x * 0.080942
Figure5140
50
60
70
80
90
680
700
720
740
760
780
800
820
840
Average Residue Degree
Dimension k = 2, p = 5
x * 0.077091
We observe that the slope of the best tting line goes slightly down with the dimension.
Thisstrongly suggeststhat takingastraight line doesnot seem to bequite orre t. Wealso
madelogarithmi plots, whi h we do not reprodu e here; they seemed to suggest to us that
abehaviour oftheform
b
(p)
N,k
∼ const · d(N )
α
isnot quite orre teither (thebest hoi eofα
seemsto be lose to
1
ina ordan e withthe previous dis ussion).These omputationssuggestthe following question.
Question 4.3. Fix a prime
p
and an even weightk ≥ 2
. Letb(N ) := b
(p)
k,N
andd(N ) :=
dim
F
p
S
k
(N ; F
p
)
. Do there exist onstantsC
1
, C
2
and0 < α ≤ β < 1
su h that the inequalityC
1
+ α
d(N )
log(d(N ))
≤ b(N ) ≤ C
2
+ β · d(N )
holds? Weremarkthatifa
(p)
N,k
behavesliked(N )
,assuggested byQuestion4.1, thenQuestion 4.3is equivalenttoaskingthat# Spec(T
k
(N ))
doesnotgrowfasterthana onstanttimeslog(d(N ))
.The phenomenon that for odd primes
p
most of the dots in the diagrams seem to lie on orvery lose to ertaindistinguished linesthroughthe originis naturalinviewof Question4.1:
theslope oftheline on or loseto whi h adot lies isjust
1
# Spec(T
k
(N ))
.
4.3. MaximumResidue Degree. Nowweturnour attention tothemaximum residue
degree, whi h wedene for given level
N
,weightk
andprimep
asc
(p)
N,k
= max{[Fm
: F
p
] | m ∈ Spec(T
k
(N ))}.
We made omputations for weight
2
and all primesp
less than100
, whereN
runs throughthe same ranges as previously. We again plot the dimension
d(N )
on thex
-axis and thefun tion
c
(p)
N,k
on they
-axis and the straight line is again the best tting fun tionα · d(N )
. Thistimewe believe thatthis might be the right fun tion to take. Here isagain a sele tionFigure52
0
10
20
30
40
50
60
70
0
20
40
60
80
100
120
140
160
Maximum Residue Degree
Dimension k = 2, p = 2
x * 0.277970
Figure530
10
20
30
40
50
60
70
80
0
20
40
60
80
100
120
140
160
Maximum Residue Degree
Dimension k = 2, p = 3
x * 0.390147
Figure540
10
20
30
40
50
60
70
80
90
0
20
40
60
80
100
120
140
160
Maximum Residue Degree
Dimension k = 2, p = 5
x * 0.377419
Figure550
10
20
30
40
50
60
70
80
90
0
20
40
60
80
100
120
140
160
Maximum Residue Degree
Dimension k = 2, p = 7
x * 0.387289
Figure560
10
20
30
40
50
0
20
40
60
80
100
Maximum Residue Degree
Dimension k = 2, p = 19
x * 0.391898
Figure570
10
20
30
40
50
60
70
80
0
20
40
60
80
100
120
140
160
Maximum Residue Degree
Dimension k = 2, p = 31
Figure58
0
10
20
30
40
50
0
10
20
30
40
50
60
70
80
Maximum Residue Degree
Dimension k = 2, p = 53
x * 0.384707
Figure590
5
10
15
20
25
30
35
0
10
20
30
40
50
60
Maximum Residue Degree
Dimension k = 2, p = 73
x * 0.416753
Figure600
5
10
15
20
25
30
35
0
10
20
30
40
50
60
Maximum Residue Degree
Dimension k = 2, p = 83
x * 0.405110
Figure610
10
20
30
40
50
60
70
80
0
20
40
60
80
100
120
140
Maximum Residue Degree
Dimension k = 2, p = 97
x * 0.393488
Hereareagaintwo examplesfor weight
4
.Figure62
0
10
20
30
40
50
60
70
80
90
0
50
100
150
200
Maximum Residue Degree
Dimension k = 4, p = 5
x * 0.362600
Figure6310
20
30
40
50
60
70
80
0
20
40
60
80
100
120
140
160
180
Maximum Residue Degree
Dimension k = 4, p = 11
x * 0.377393
Thedata ertainly suggestthatthemaximumresiduedegreegrowswiththedimension. It is
remarkabletoseethattheslopesofthebestttinglinesallseemtobevery losetoea hother
withthesingleex eptionofthe ase
p = 2
,whi hmightbe ausedbythesamephenomenonas earlier. Also in this ase we ondu ted a loser analysis for the primes
2
,3
and5
. ForFigure64
60
80
100
120
140
160
180
200
220
240
420
430
440
450
460
470
480
490
500
510
Maximum Residue Degree
Dimension k = 2, p = 2
x * 0.282934
Figure65100
150
200
250
520
540
560
580
600
Maximum Residue Degree
Dimension k = 2, p = 2
x * 0.266686
Figure66100
150
200
250
300
620
640
660
680
700
720
Maximum Residue Degree
Dimension k = 2, p = 2
x * 0.289400
Figure67100
150
200
250
300
350
720
740
760
780
800
820
Maximum Residue Degree
Dimension k = 2, p = 2
x * 0.274635
Figure68150
200
250
300
350
400
450
930
940
950
960
970
980
990
1000
Maximum Residue Degree
Dimension k = 2, p = 2
x * 0.276529
Here arethe plots for
p = 3, 5
and the primes between3000
and10009
subdivided into fourFigure69
40
60
80
100
120
140
160
180
200
240
260
280
300
320
340
360
380
Maximum Residue Degree
Dimension k = 2, p = 3
x * 0.378322
Figure70100
150
200
250
300
380
400
420
440
460
480
500
520
540
Maximum Residue Degree
Dimension k = 2, p = 3
x * 0.374745
Figure71100
150
200
250
300
350
540
560
580
600
620
640
660
680
Maximum Residue Degree
Dimension k = 2, p = 3
x * 0.366550
Figure72150
200
250
300
350
400
680
700
720
740
760
780
800
820
840
Maximum Residue Degree
Dimension k = 2, p = 3
x * 0.374045
Figure7360
80
100
120
140
160
180
200
220
240
260
280
300
320
340
360
380
Maximum Residue Degree
Dimension k = 2, p = 5
x * 0.372593
Figure74100
150
200
250
300
380
400
420
440
460
480
500
520
540
Maximum Residue Degree
Dimension k = 2, p = 5
Figure75
100
150
200
250
300
350
520
540
560
580
600
620
640
660
680
Maximum Residue Degree
Dimension k = 2, p = 5
x * 0.370869
Figure76150
200
250
300
350
400
680
700
720
740
760
780
800
820
840
Maximum Residue Degree
Dimension k = 2, p = 5
x * 0.366508
We observethat,although thebestttinglineswere omputedusingdierentintervals,their
slopesarevery loseto ea h other. These omputationssuggestthefollowing question.
Question 4.4. Fix a prime
p
and an even weightk ≥ 2
. Letc(N ) := c
(p)
k,N
andd(N ) :=
dim
F
p
S
k
(N ; F
p
)
. Dothere exist onstantsC
1
, C
2
and0 < α ≤ β < 1
su h that the inequalityC
1
+ α · d(N ) ≤ c(N ) ≤ C
2
+ β · d(N )
holds?
Referen es
[BLGHT℄ T.Barnet-Lamb, D. Geraghty,M.HarrisandR. Taylor.A family of Calabi-Yauvarieties
andpotentialautomorphy II.Toappearin P.R.I.M.S.
[BCP℄ W. Bosma,J.Cannon,andC.Playoust.The MagmaalgebrasystemI: Theuser language. J.
Symb. Comp. 24(34)(1997),235265.
[B℄ J.Bosman.Modular formsapplied tothe omputationalinverseGalois problem.Preprint,2010.
[CHT℄ L. Clozel, M. Harris and R. Taylor.Automorphy for some l-adi lifts of automorphi mod l
representations.Pub.Math.IHES108(2008),1181.
[CE℄ Coleman, Robert F.; Edixhoven, Bas. On the semi-simpli ity of the
U
p
-operator on modular forms. Math.Ann.310(1998),no.1,119127.[DiWi℄ L. Dieulefait, G. Wiese. On Modular Forms and the Inverse Galois Problem. A epted for
publi ationin theTransa tionsoftheAMS.
[DS℄ Deligne, Pierre;Serre,Jean-Pierre.Formes modulaires de poids
1
.Ann. S i. É oleNorm. Sup. (4)7(1974),507530.[DI℄ F.DiamondandJ.Im.Modularformsandmodular urves,inSeminaronFermat'sLastTheorem
(Toronto, ON,19931994), 39133,Amer.Math. So .,Providen e,RI,1995.
[HSHT℄ M.Harris,N.Shepherd-BarronandR.Taylor.AfamilyofCalabi-Yauvarietiesandpotential
automorphy. AnnalsofMath.171(2010),779813.
[M℄ B.Mazur.Modular urvesandtheEisensteinideal.Inst.HautesÉtudesS i.Publ.Math. No.47
(1977),33186(1978).
[S℄ J.-P.Serre.Répartitionasymptotiquedesvaleurspropres del'opérateurde He ke
T
[T℄ R.Taylor.Automorphy forsomel-adi liftsofautomorphi modlrepresentations.IIPub.Math.
IHES108(2008),183239.
[W℄ G.Wiese.Onproje tivelineargroupsoverniteeldsasGaloisgroupsovertherationalnumbers.
In: 'Modular Forms onS hiermonnikoog'editedby BasEdixhoven, Gerardvander Geerand Ben
Moonen.CambridgeUniversityPress,2008,343350.
23o tobre 2010
Mar el Mohyla, Universität Duisburg-Essen,Institutfür ExperimentelleMathematik,Ellernstraÿe 29,
45326Essen,Germany
GaborWiese,UniversitätDuisburg-Essen,InstitutfürExperimentelleMathematik,Ellernstraÿe 29,45326