Lifting sections along torsors
Niels Borne
(j. w. Michel Emsalem [Lille]
Jakob Stix [Heidelberg])
Laboratoire Paul Painlevé Université Lille 1, France
Scuola Normale Superiore Pisa 10.10.2012
Outline
Anabelian conjectures Main conjecture Section conjecture
Cuspidalization conjecture Evaluation of units
Lifting Galois sections The torsors
Torsion packets Main result Sketch of a proof
Arithmetic first Chern class Stix’s result
Conclusion
Niels Borne (j. w. Michel Emsalem [Lille] Jakob Stix [Heidelberg]) Lifting sections along torsors
The Hom conjecture
Letk be a number field, and Galk its absolute Galois group. Let X/k be a smooth, geometrically connected curve. Denote byg the genus ofXk, andnits numbers of "holes". Letx be a geometric point ofXk. The algebraic fundamental group π1(Xk,x)'Γdg,nfits into a short exact sequence:
1→π1(Xk,x)→π1(X,x)→Galk →1 Conjecture (Grothendieck 1983, Tamagawa 1997, Mochizuki 1999)
If X,Y are hyperbolic curves the morphism Homdomk (X,Y)→HomopenGal
k(π1(X), π1(Y))/∼ is one to one (where∼denotes conjugacy byπ1(Yk)).
Section conjecture for proper curves
LetX/k be a geometrically connected scheme of finite type andx :SpecΩ→X a geometric point. The “fundamental exact sequence” is the extensionπ1(X/k):
1→π1(Xk,x)→π1(X,x)→Galk →1
By functoriality,x ∈X(k)induces a sectionsX(x), well defined up to conjugacy by an element ofπ1(Xk,x). One thus get:
sX :X(k)→Sπ1(X/k)
Conjecture (Section conjecture, Grothendieck 1983) If X a smooth, proper curve of genus greater than2over a field k of finite type overQ, the application sX is one to one.
Remark
The Mordell-Weil theorem easily implies that sX is into.
Niels Borne (j. w. Michel Emsalem [Lille] Jakob Stix [Heidelberg]) Lifting sections along torsors
Section conjecture and diophantine geometry
The Section Conjecture was initially thought as a strategy to show:
Conjecture (Mordell Conjecture 1922, Faltings 1983) Let k be a number field. If X/k is a smooth proper
geometrically connected curve of genus g≥2, then
#X(k)<∞.
Even today, no examples of suchX are known where 0<#Sπ1(X/k)<∞, and it is difficult to produce interesting examples whereSπ1(X/k)=∅.
However, Grothendieck’s idea to relate diophantine issues to fundamental groups (in a broader sense) has been very
successful lately (Kim’s proof of Siegel theorem, 2005, is based on this principle).
Packet of cuspidal sections
LetX/k be a smooth, proper, geometrically connected curve, andU =X\Dan affine dense open subset. Forx ∈D(k), denote byXcx a formal completion ofX atx and byUcx =Xcx\x. The morphism of extensions
1 //π1(Udk,x)
//π1(cUx) //
Galk //
1 1 //π1(Uk) //π1(U) //Galk //1 defines a mapSπ
1(cUx/k)→Sπ1(U/k).
The groupπ1(Udk,x)'Zb(1) =lim←−nµnis the inertia group atx.
One can show that the above sequence splits, so thatSπ
1(Ucx/k)
is a torsor under H1(Galk,bZ(1))'kc∗ :=lim←−nk∗/(k∗)n.
Niels Borne (j. w. Michel Emsalem [Lille] Jakob Stix [Heidelberg]) Lifting sections along torsors
Section conjecture for affine curves
U(k)` `
x∈D(k)Sπ
1(cUx/k) sU //
Sπ1(U/k)
X(k) sX //Sπ1(X/k)
Conjecture (Section conjecture for affine curves)
If U a smooth, hyperbolic curve (that is2−2g−r <0, where r = #D(k)) over a field k of finite type overQ, the application sU is one to one.
A consequence of the Section conjecture forX is:
Conjecture (Cuspidalization conjecture) Sπ1(U/k)→Sπ1(X/k)is surjective.
Kummer class of a unit
Mochizuki has observed than one can (almost) reconstruct the valuef(x)of a unitf ∈Γ(U,Gm)at a rational pointx ∈U(k) from the Kummer class off and associated sectionsx. Starting from Kummer sequence
1→µn→Gm
−−→×n Gm →1
one defines evaluation·(s)of a unit on a fixed sectionsas:
Γ(U,Gm) //H1(U,Z(1))b
H1(π1(U),Zb(1))
∼=
OO
s∗ //H1(Galk,Zb(1)) ∼= //ck∗ It is easy to check thatf(sx)is the image off(x)byk∗ →kc∗.
Niels Borne (j. w. Michel Emsalem [Lille] Jakob Stix [Heidelberg]) Lifting sections along torsors
Cuspidally central fundamental group
Mozichuki defines the cuspidally central fundamental group πcc1 (U)as the largest intermediate quotient
π1(U)π1cc(U)π1(X)
such thatK =ker(π1cc(U)→π1(X))is abelian, and the induced action ofπ1(Xk)onK by conjugacy is trivial.
In Mochizuki’s theory of evaluation of units on sections,π1cc(U) is almost as good asπ1(U)since:
Lemma (Mochizuki 2005)
The morphismH1(π1cc(U),Zb(1))→H1(π1(U),Zb(1))is an isomorphism.
Set up
Letk be a field of characteristic 0.
LetX/k be a smooth, proper, geometrically connected curve, E →X a torsor under a torusT, trivialized over an affine open subsetU =X\D, so that the chosen section induces an epimorphismπ1(U)π1(E). This gives rise to:
π1(U)π1(E)π1(X)
This will allow us to give a geometric interpretation of the intermediate quotient constructed by Mochizuki by profinite means, and to give a positive result - in a special situation - for lifting Galois sections alongπcc1 (U)→π1(X).
Our construction of the relevant torsor consist of two steps.
Niels Borne (j. w. Michel Emsalem [Lille] Jakob Stix [Heidelberg]) Lifting sections along torsors
The torsor E
DThe generalized Jacobian ofUis constructed from the pushout
diagram D //
X
Speck //XD
. This leads to an exact sequence
0→Gm→Gm,D→Pic0X
D/k →Pic0X/k →0. So, denoting bySD the torusGm,D/Gm, we get a naturalSD-torsor onX by
pullback: ED //
PicXD/k
X
O(∆)//PicX/k
The torsor F
DDenote byTD the torusGm,D, it fits into an exact sequence 0→Gm→TD →SD →0. We define a canonical line bundle FD/ED corresponding to the top row of:
ED //
PicXD/k
//Pick/k =BGm
X O(∆)//PicX/k
where we have to
consider here Picardstacks. The schemeFD/X is endowed with a canonical structure of aTD-torsor, trivialized overU.
Proposition
The morphismπ1(U)→π1(FD)is surjective, and induces an isomorphismπcc1 (U)'π1(FD).
Niels Borne (j. w. Michel Emsalem [Lille] Jakob Stix [Heidelberg]) Lifting sections along torsors
Torsion criterion for E
DDefinition (Baker-Poonen)
A reduced effective divisor D is a torsion packet if any degree0 divisor on Xk with support in Dk is torsion.
Remarks
1. Any rational point defines a torsion packet.
2. Let X be a modular curve. Manin-Drinfeld’s theorem states that the divisor of cusps D is a torsion packet.
Proposition
Let X/k be a smooth, proper, geometrically connected curve.
1. The torsor ED →X is torsion if and only if D is a torsion packet.
2. The torsor FD →X is never torsion.
Lifting Galois sections to F
Dover the rationals
Theorem (Emsalem-Stix-B. 2012)
Let X/Qbe a smooth, proper, geometrically connected curve of genus≥2, and let D⊂X be a union of torsion packets. Then every Galois section s:GalQ→π1(X)lifts to a section
GalQ →π1(FD).
Corollary
Let X/Qbe a smooth, proper, geometrically connected of genus≥2, and let U=X \X(Q)be the complement of the set of allQ-rational points. Then every Galois section
s:GalQ →π1(X) lifts to a sectionGalQ →πcc1 (U).
Niels Borne (j. w. Michel Emsalem [Lille] Jakob Stix [Heidelberg]) Lifting sections along torsors
Cohomological translation
LetE →X a torsor under a torusT. The fiberEx ofE →X at a geometric pointx ofX identifies withTk, leading to a sequence
1→π1(Tk)→π1(E)→π1(X)→1
which is exact becauseX is an algebraicK(π,1). The same fact shows that the morphism H2(π1(X), π1(Tk))→H2(X,T(T)) is an isomorphism. Moreover the image of the classπ1(E/X)is the arithmetic first Chern class ofE/X, obtained from the cobord morphismc1:H1(X,T)→H2(X,T(T))associated to the Kummer sequences 0→T[n]→T −−→×n T →0.
To a sections:Galk →π1(X), one can thus associate a class s∗(c1(E/X))∈H2(Speck,T(T))
which cancels exactly whenslifts toπ1(E).
Over Q , a section kills the relative Brauer group
The image of a sections:Galk →π1(X)corresponds to a (pro-)coverXs →X, and it is easy to check that for a line bundleLonX
s∗(c1(L)) =0 ⇐⇒ Lis divisible onXs Moreover from the exact sequence:
0→Pic(Xs)→PicXs/k(k)→Br(k)→Br(Xs)
one deduces that if Br(Xs/k) =ker(Br(k)→Br(Xs))vanishes, then any line bundleLonX is divisible onXs.
Theorem (Stix 2010)
Let X/Qbe a smooth projective curve of positive genus such thatπ1(X/Q)admits a section. Then the relative Brauer group Br(X/Q)vanishes.
Niels Borne (j. w. Michel Emsalem [Lille] Jakob Stix [Heidelberg]) Lifting sections along torsors
Killing the obstruction over Q
H1(X,Gm) c
1
//
uukkkkkkk H2(X,Zb(1))
uukkkkkkk
s∗
H1(X,TD) c
1
//
uullllll H2(X,T(TD))
s∗
H1(X,SD)
H2(k,Zb(1))
uukkkkkkk
0 //H1(k,TD) =0 //H2(k,T(TD)) //T(H2(k,TD))
What is missing
I A version of Stix’s theorem over a number field.
I Canonical liftings for non cuspidal sections:
U(k)` `
x∈D(k)Sπ
1(Ucx/k) sU //
Sπ1(U/k)
//Sπ1(F/k)
yyssssssssss
X(k) sX //Sπ1(X/k)
If the section conjecture holds, then a non-cuspidal section inSπ1(X/k)should have acanonicallifting inSπ1(F/k).
I Algebraization of value of sections at units: the valuef(s) of a sections inSπ1(U/k)at a unitΓ(U,Gm)should be algebraic, i.e. should belong tok∗ ⊂ck∗.
Niels Borne (j. w. Michel Emsalem [Lille] Jakob Stix [Heidelberg]) Lifting sections along torsors
Appendix : a sketch of a proof of Stix result
Proposition
Let X/k be a smooth projective curve of positive genus such thatπ1(X/k)admits a section. Then:
1. if k/Qpis a finite extension, then#Br(X/k)is a power of p, 2. if k =RthenBr(X/R) =0.
Going back toX/Q, one uses Hasse-Brauer-Noether theorem that shows that:
Br(X/Q),→ker M
v
Br(XQv/Qv)
P
vinvv
−−−−−→Q/Z
!