weil reciprocity in analytic geometry
TRANSCRIPT
Weil Reciprocity in Analytic GeometryTynan Ochse
3. Rigid difficulties
4. Current Work + Hopes for the Future
Theory Algebraic Complex analytic Rigid analytic Basic object Spec(A) Dn Spa(A)
Functions polynomial holomorphic “holomorphic”
Cohomology etale Deligne-Beilinson TBD
I K (C ) the function field of C
I D is a closed disk
I D is an open disk
(Algebraic) Weil Reciprocity
Version 1: Let C be a curve over an algebraically closed field K . For a divisor D and function f ∈ K (C ), define
f (D) = ∏ P
f (P)vP(D)
Then for any f , g ∈ K (C ) (with disjoint support), one has
f ((g)) = g((f ))
Version 2: Let C be a curve over an algebraically closed field K . Given invertible functions f and g in K (C ), we define a local symbol (f , g)P at each point P on C as follows:
(f , g)P = (−1)vP(f )·vP(g) f vP(g)
g vP(f ) (P)
where in this context vP is the order of a function at a point P. One then has that ∏
P∈C (f , g)P = 1
Reciprocity Laws Lightning Round
(−1) p−1
2 q−1
−1 else
Theorem (Gauss)
a, b ∈ Z[i ] distinct primes with a ≡ b ≡ 1 (mod 2 + 2i). Then
(−1) N(a)−1
4 N(b)−1
4
[ a
b
][ b
a
] = 1
For all a, b ∈ Z[ √ −1] not dividing 2 there is a unique[
a b
Theorem (Eisenstein)
l odd prime, α ∈ Z[ζl ] congruent some rational integer mod (1− ζl)2. Then (α
a
) l
for all a ∈ Z prime to l .
For α - p in Z[ζl ], there is a unique l th root of unity such that( α
p
Theorem (Hilbert)
k an algebraic number field. Then ∀a, b ∈ k×, one has∏ p
( a, b
Weil Reciprocity is King
The result is still true over non-algebraically closed fields: Let C = P1 and f , g polynomials in K (C ). Write f =
∏n i=1 fi and
g = ∏m
i=1 gi . Set Fi = k[x ]/fi and Gi = k[x ]/gi . Then
m∏ i=1
i=1
pn−1 p−1
(f mod g) pd−1
2 = (−1) ed(p−1)
2
2 = ( α p
Gm ×Gm → Gm
en : (Pic0C )[n]× (Pic0C )[n]→ µn ⊂ k×
Pic0C is the degree 0 divisors on C (isomorphism classes of line bundles with vanishing first Chern class).
en(D1,D2) = f1(D2) f2(D1) is not a-priori well-defined... but Weil
reciprocity assures it is.
If C = E is an elliptic curve, then E ∼= Pic0(E ).
en : E [n]× E [n]→ µn
Poincare duality in etale cohomology is a pairing
H i et(XK ,Z/nZ)× H2n−i
et (XK ,Z/nZ)→ µn
H1 et(E ,Z/lZ)× H1
Weil reciprocity = tame symbol = Poincare duality
.
Gm(K )×Gm(K )→ Gm(k)
Hence an isomorphism
We note Gm(K ) ∼= K0, and then dualize:
1 DGm(K )dR DGm(K ) ∼= BGm(K ) B1 K 0
d log
Qcoh(LocSysGm(D∗)) Qcoh(Gm(K )dR) Dmod(Gm(K ))∼ ∼
Variant in Characteristic p
Suppose k is finite and Γ a finite abelian group. The Tame symbol gives an isomorphism
Homgp(K×,BΓ) ∼= Homsch(D∗,BΓ)
The RHS is identified with Γ-torsors on D, i.e. Hom(πet 1 D∗, Γ).
On the other hand, the Lang isogeny gives an isomorphism
Hom(K×,BΓ) ∼= Hom(K×, Γ)
Take aways...
1. implies interesting modularity relations,
2. produces a well-defined pairing on algebraic group schemes, and, up to *insert theory here,*
3. is a crucial piece of geometric local class field theory
Now let X be an open Riemann surface.
(f , g)x = (−1)vx (f )vx (g) g vx (f )
f vx (g) (x)
( −1
2πi
∫ γ
log(f )d log(g)
) · g vx (f )
General idea is to compute the residue of some line bundle on X (monodromy). Monodromy away from ∂X is automatically trivial, so suffices to consider holomorphic functions defined near ∂X .
Robba Rings
O the structure sheaf on X , i : X → X .
Define RX as the subsheaf of i∗O consisting of germs of holomorphic functions that have a value on ∂X .
RX = ∏ π0(∂X )RX ,γ
For each γ ∈ π0(∂X ), have a local symbol (f , g)γ given by the monodromy formula.
Theorem (Deligne)
For all f ∈ H0(∂X ,R∗), g ∈ H0(X ,R∗) one has
1 = (f , g)∂X = ∏
} ∼= { π1(X )→ Gm}
Given functions f and g , and s ∈ π1(X ), we get a number
χf ,g (s) = (f , g)s ∈ Gm(C)
π1(X ) has a single relation ∏
i [ai , bi ] ∏
j γj = 1.
χf ,g ( ∏ i
[ai , bi ] ∏ j
C∗ is commutative, so χf ,g ( ∏
j γj) = ∏
0 Z(1) O O∗ 0 exp
0 G (C) G (O) 1 ⊗ Lie(G ) 0 dlog(g)
Hence a morphism
] (1)
G (C)[1] ∼−→ [ G (O)→ Lie(G )⊗ 1
] Yields
Passing to cohomology gives
H i (X ,O∗)⊗ H j(X ,G (O)) Ui,j−−→ H i+j+1(X ,G (C)(1))
For i = j = 0, this map attaches a G (C)(1)-torsor (f , g) to functions f and g .
We also have a residue map
Res : H1(X\{x},G (C)(1))→ A
If f is invertible on X\{x}, and g is a function on X\{x} with values in G , then set
Res U0,0 = (f , g)x ∈ G (C)
Let γ be a closed curve about x . Then (f , g)x is the monodromy of (f , g) about γ.
We can also specialize to various G : when G = Ga,
(f , g)x = Res(g · d log(f ))
And when G = Gm,
f vx (g) (x)
In general, provided g admits a logarithm in Lie(G ), one has
(f , g)x = exp(Res(g · d log(f )))
A choice of log(f ) defines a trivialization of (f , g), denoted (log(f ), g). Let γ be an oriented curve about x and fix y ∈ γ. Then a local horizontal section of (f , g) can be written as
(log(f ), g) · exp( −1
p−1 X → 0
And define Hq D(X ,Z(p)) = Hq(X ,Z(p)).
Just like in the algebraic case, the cup product on Deligne-Beilinson cohomology
H1 D(X ,Z(1))× H1
Rigid Geometry
Roughly, a rigid analytic space X is like a complex analytic space, but defined over a non-archimedian field k.
Non-archimedean means
2. |xy | = |x | · |y | 3. |x + y | ≤ max{|x |, |y |}
Definition The Tate algebra of dimension n is the subring kt1, · · · , tn of k[[t1, · · · , tn]] defined by
Tn(k) = { ∑ i
i→∞ |ai | = 0}
Definition An affinoid algebra A is an algebra isomorphic to Tn/I for some n and some closed ideal I . An affinoid is Spa(A) for some affinoid algebra A.
Definition A rigid analytic space over k is a locally ringed space (X ,OX ) with an admissible cover by affinoids.
Examples
2. Gm,k =
t , t qn )
3. Gm,k/qZ, q ∈ k∗, |q| < 1 (Tate elliptic curve).
4. XFF = ⊕
Rigid Difficulties p1
2. Covers by admissible opens (G-topology)
3. Want theory to be familiar (coherent sheaves, sheaf cohomology, GAGA) and robust (base changes, stalks, quasicoherent sheaves).
Rigid Difficulties p2
Now let X be a rigid analytic curve over a complete nonarchimedean field k . If one wants to produce an analogue of Deligne’s symbol, immediate questions are
1. What space of functions should we consider?
2. Do log, exp, ∫
, etc. even make sense in this setup?
3. Can the theory be set up to work uniformly for characteristic (0, 0), (0, p), and (p, p) situations?
I Characteritic (0, 0): R((t)) (|x | = q−ordt(x))
I Characteristic (0, p): Qp
I Characteristic (p, p): Fp((t)).
Robba Ring
Let Γ(r , 1) be the ring of convergent power series on the annulus {t | r < t < 1}.
Definition The Robba ring on D is
R = r<1
i→−∞ |ai |δi = 0
Replace D by the disk of radius p
−1 p−1 in characteristic p.
Rigid Analytic Tame Symbol
The goal is to produce a nondegenerate, perfect local symbols
R∗ ×R∗ (·,·)x−−−→ Gm
Unfortunately, log, exp and ∫
The upshot is one can extract monodromy-type information from 2-determinants [Garcia-Lopez, Clausen].
Linear algebra of Frechet/Tate-analytic spaces
Given f , g ∈ R∗X ,x , we get lines 2det(f ) and 2det(g) and two isomorphisms
2det(f )⊗ 2det(g) 2det(g)⊗ 2det(f )
RX = ∏ RX ,x and (f , g) =
∏ x(fx , gx)x
1 = (f , g) = ∏ x
Moreover, there exists a subquotient H of RX such that
H × H R/k ×R/k
Gm
log
exp(Res(fdg))
Local Global
Multiplicative R∗X ×R∗X → k∗ BunGm(X )× BunGm,c(X )→ · · · Additive RX ×RXdt → k BunGa(X )× BunGa,c,(X )→ · · ·
R∗X = Gm × Z× H+ × H−
Hopes for the Future
2. Rigid Class Field Theory
3. Matching the Rigid Tame Symbol with the cup product of some cohomology theory
Thank you!
3. Rigid difficulties
4. Current Work + Hopes for the Future
Theory Algebraic Complex analytic Rigid analytic Basic object Spec(A) Dn Spa(A)
Functions polynomial holomorphic “holomorphic”
Cohomology etale Deligne-Beilinson TBD
I K (C ) the function field of C
I D is a closed disk
I D is an open disk
(Algebraic) Weil Reciprocity
Version 1: Let C be a curve over an algebraically closed field K . For a divisor D and function f ∈ K (C ), define
f (D) = ∏ P
f (P)vP(D)
Then for any f , g ∈ K (C ) (with disjoint support), one has
f ((g)) = g((f ))
Version 2: Let C be a curve over an algebraically closed field K . Given invertible functions f and g in K (C ), we define a local symbol (f , g)P at each point P on C as follows:
(f , g)P = (−1)vP(f )·vP(g) f vP(g)
g vP(f ) (P)
where in this context vP is the order of a function at a point P. One then has that ∏
P∈C (f , g)P = 1
Reciprocity Laws Lightning Round
(−1) p−1
2 q−1
−1 else
Theorem (Gauss)
a, b ∈ Z[i ] distinct primes with a ≡ b ≡ 1 (mod 2 + 2i). Then
(−1) N(a)−1
4 N(b)−1
4
[ a
b
][ b
a
] = 1
For all a, b ∈ Z[ √ −1] not dividing 2 there is a unique[
a b
Theorem (Eisenstein)
l odd prime, α ∈ Z[ζl ] congruent some rational integer mod (1− ζl)2. Then (α
a
) l
for all a ∈ Z prime to l .
For α - p in Z[ζl ], there is a unique l th root of unity such that( α
p
Theorem (Hilbert)
k an algebraic number field. Then ∀a, b ∈ k×, one has∏ p
( a, b
Weil Reciprocity is King
The result is still true over non-algebraically closed fields: Let C = P1 and f , g polynomials in K (C ). Write f =
∏n i=1 fi and
g = ∏m
i=1 gi . Set Fi = k[x ]/fi and Gi = k[x ]/gi . Then
m∏ i=1
i=1
pn−1 p−1
(f mod g) pd−1
2 = (−1) ed(p−1)
2
2 = ( α p
Gm ×Gm → Gm
en : (Pic0C )[n]× (Pic0C )[n]→ µn ⊂ k×
Pic0C is the degree 0 divisors on C (isomorphism classes of line bundles with vanishing first Chern class).
en(D1,D2) = f1(D2) f2(D1) is not a-priori well-defined... but Weil
reciprocity assures it is.
If C = E is an elliptic curve, then E ∼= Pic0(E ).
en : E [n]× E [n]→ µn
Poincare duality in etale cohomology is a pairing
H i et(XK ,Z/nZ)× H2n−i
et (XK ,Z/nZ)→ µn
H1 et(E ,Z/lZ)× H1
Weil reciprocity = tame symbol = Poincare duality
.
Gm(K )×Gm(K )→ Gm(k)
Hence an isomorphism
We note Gm(K ) ∼= K0, and then dualize:
1 DGm(K )dR DGm(K ) ∼= BGm(K ) B1 K 0
d log
Qcoh(LocSysGm(D∗)) Qcoh(Gm(K )dR) Dmod(Gm(K ))∼ ∼
Variant in Characteristic p
Suppose k is finite and Γ a finite abelian group. The Tame symbol gives an isomorphism
Homgp(K×,BΓ) ∼= Homsch(D∗,BΓ)
The RHS is identified with Γ-torsors on D, i.e. Hom(πet 1 D∗, Γ).
On the other hand, the Lang isogeny gives an isomorphism
Hom(K×,BΓ) ∼= Hom(K×, Γ)
Take aways...
1. implies interesting modularity relations,
2. produces a well-defined pairing on algebraic group schemes, and, up to *insert theory here,*
3. is a crucial piece of geometric local class field theory
Now let X be an open Riemann surface.
(f , g)x = (−1)vx (f )vx (g) g vx (f )
f vx (g) (x)
( −1
2πi
∫ γ
log(f )d log(g)
) · g vx (f )
General idea is to compute the residue of some line bundle on X (monodromy). Monodromy away from ∂X is automatically trivial, so suffices to consider holomorphic functions defined near ∂X .
Robba Rings
O the structure sheaf on X , i : X → X .
Define RX as the subsheaf of i∗O consisting of germs of holomorphic functions that have a value on ∂X .
RX = ∏ π0(∂X )RX ,γ
For each γ ∈ π0(∂X ), have a local symbol (f , g)γ given by the monodromy formula.
Theorem (Deligne)
For all f ∈ H0(∂X ,R∗), g ∈ H0(X ,R∗) one has
1 = (f , g)∂X = ∏
} ∼= { π1(X )→ Gm}
Given functions f and g , and s ∈ π1(X ), we get a number
χf ,g (s) = (f , g)s ∈ Gm(C)
π1(X ) has a single relation ∏
i [ai , bi ] ∏
j γj = 1.
χf ,g ( ∏ i
[ai , bi ] ∏ j
C∗ is commutative, so χf ,g ( ∏
j γj) = ∏
0 Z(1) O O∗ 0 exp
0 G (C) G (O) 1 ⊗ Lie(G ) 0 dlog(g)
Hence a morphism
] (1)
G (C)[1] ∼−→ [ G (O)→ Lie(G )⊗ 1
] Yields
Passing to cohomology gives
H i (X ,O∗)⊗ H j(X ,G (O)) Ui,j−−→ H i+j+1(X ,G (C)(1))
For i = j = 0, this map attaches a G (C)(1)-torsor (f , g) to functions f and g .
We also have a residue map
Res : H1(X\{x},G (C)(1))→ A
If f is invertible on X\{x}, and g is a function on X\{x} with values in G , then set
Res U0,0 = (f , g)x ∈ G (C)
Let γ be a closed curve about x . Then (f , g)x is the monodromy of (f , g) about γ.
We can also specialize to various G : when G = Ga,
(f , g)x = Res(g · d log(f ))
And when G = Gm,
f vx (g) (x)
In general, provided g admits a logarithm in Lie(G ), one has
(f , g)x = exp(Res(g · d log(f )))
A choice of log(f ) defines a trivialization of (f , g), denoted (log(f ), g). Let γ be an oriented curve about x and fix y ∈ γ. Then a local horizontal section of (f , g) can be written as
(log(f ), g) · exp( −1
p−1 X → 0
And define Hq D(X ,Z(p)) = Hq(X ,Z(p)).
Just like in the algebraic case, the cup product on Deligne-Beilinson cohomology
H1 D(X ,Z(1))× H1
Rigid Geometry
Roughly, a rigid analytic space X is like a complex analytic space, but defined over a non-archimedian field k.
Non-archimedean means
2. |xy | = |x | · |y | 3. |x + y | ≤ max{|x |, |y |}
Definition The Tate algebra of dimension n is the subring kt1, · · · , tn of k[[t1, · · · , tn]] defined by
Tn(k) = { ∑ i
i→∞ |ai | = 0}
Definition An affinoid algebra A is an algebra isomorphic to Tn/I for some n and some closed ideal I . An affinoid is Spa(A) for some affinoid algebra A.
Definition A rigid analytic space over k is a locally ringed space (X ,OX ) with an admissible cover by affinoids.
Examples
2. Gm,k =
t , t qn )
3. Gm,k/qZ, q ∈ k∗, |q| < 1 (Tate elliptic curve).
4. XFF = ⊕
Rigid Difficulties p1
2. Covers by admissible opens (G-topology)
3. Want theory to be familiar (coherent sheaves, sheaf cohomology, GAGA) and robust (base changes, stalks, quasicoherent sheaves).
Rigid Difficulties p2
Now let X be a rigid analytic curve over a complete nonarchimedean field k . If one wants to produce an analogue of Deligne’s symbol, immediate questions are
1. What space of functions should we consider?
2. Do log, exp, ∫
, etc. even make sense in this setup?
3. Can the theory be set up to work uniformly for characteristic (0, 0), (0, p), and (p, p) situations?
I Characteritic (0, 0): R((t)) (|x | = q−ordt(x))
I Characteristic (0, p): Qp
I Characteristic (p, p): Fp((t)).
Robba Ring
Let Γ(r , 1) be the ring of convergent power series on the annulus {t | r < t < 1}.
Definition The Robba ring on D is
R = r<1
i→−∞ |ai |δi = 0
Replace D by the disk of radius p
−1 p−1 in characteristic p.
Rigid Analytic Tame Symbol
The goal is to produce a nondegenerate, perfect local symbols
R∗ ×R∗ (·,·)x−−−→ Gm
Unfortunately, log, exp and ∫
The upshot is one can extract monodromy-type information from 2-determinants [Garcia-Lopez, Clausen].
Linear algebra of Frechet/Tate-analytic spaces
Given f , g ∈ R∗X ,x , we get lines 2det(f ) and 2det(g) and two isomorphisms
2det(f )⊗ 2det(g) 2det(g)⊗ 2det(f )
RX = ∏ RX ,x and (f , g) =
∏ x(fx , gx)x
1 = (f , g) = ∏ x
Moreover, there exists a subquotient H of RX such that
H × H R/k ×R/k
Gm
log
exp(Res(fdg))
Local Global
Multiplicative R∗X ×R∗X → k∗ BunGm(X )× BunGm,c(X )→ · · · Additive RX ×RXdt → k BunGa(X )× BunGa,c,(X )→ · · ·
R∗X = Gm × Z× H+ × H−
Hopes for the Future
2. Rigid Class Field Theory
3. Matching the Rigid Tame Symbol with the cup product of some cohomology theory
Thank you!