Invent. math.

lll, 309 330 (1993)

/nvo/tiol/~
mathematicae
9 Springer-Verlag 1993

Riemann's existence problem for a p-adic field

W. Liitkebohmert
Universit~it Ulm, Abteilung Reine Mathematik, Helmholtzstrasse 18, W-7900 Ulm,
Federal Republic of Germany
Oblatum ii-1II-1992 & 7-VII-1992

Riemann posed the problem if any finite 6tale morphism from a Riemann surface to
an affine algebraic curve over C is induced by an algebraic morphism. Nowadays it
is well-known that this question has an affirmative answer. In the classical paper
[ G R 1] Grauert and Remmert settled this problem in higher dimension. It could be
solved due to Serre's GAGA-principle by an extension theorem for finite analytic
6tale coverings. A crucial point in their proof is the elementary fact that any 6tale
covering of a pointed disc extends to a finite covering of the whole disc. The proof
of Grauert and Remmert mainly consists in showing that a similar result holds for
finite analytic 6tale coverings of the complement o f a subvariety in a complex space.
Some years later Grothendieck gave a new proof by using Hironaka's resolution of
singularities; so he could avoid the technicalities of [GR1] by reducing the
extension problem to the case where the varieties are smooth.
In this article we treat a similar question for rigid-analytic coverings in the sense
of Tate I T ] . We will show that this problem has an affirmative answer without
further restrictions on the covering in the case where the base field has character-
istic zero and residue characteristic p. Following Grothendieck's proof we can
reduce the problem to the following theorem (ifp is zero one has to replace [p[ by 1)
which was announced by Gabber in 1982:

Let (o : X --+ A(r, R) = {z 6 K*, r < ]zl < R} be a finite Otale covering of degree
d. Then there exists a number b = b(d, p) ~ N depending only on the degree d of q~ and
on the residue characteristic p such that ~p is o f Kummer type over A(r', R') where
r' = r/[pl b and where R' = ]p[bR.

From this result it follows immediately that any finite 6tale covering of
a pointed disc extends to a finite (ramified) covering of the whole disc. In complex
analysis this problem is solved by a topological argument; in the p-adic case one
uses a number-theoretical argument.
As we will point out in 2.10, in positive characteristic there exist examples of
non-extendable finite 6tale coverings of a pointed disc. But we will show Riemann's
existence problem has an affirmative answer in positive characteristic if one sticks
to Galois coverings whose order is prime to the characteristic. Such a result is of
310 w. Lfitkebohmert

interest for Drinfeld's work, since it implies a comparison theorem between the
rigid 6tale cohomology and the algebraic ~tale cohomology of an algebraic variety
over a p-adic field. Since the resolution of singularities is not known in positive
characteristic, we will solve the extension problem in codimension /> 3 by an
extension theorem for coherent sheaves.

1 Diseriminants over stable fields

Let K be a field equipped with a non-archimedean absolute value I" 1.We denote by
(gK its valuation ring and by/~ its residue field. Recall that K is called stable if any
finite extension L/K equipped with its spectral norm is K-cartesian; cf. [BGR,
3.6.1/1]. The latter means that there exists an orthogonal basis of L over K; cf.
[BGR, 2.4.1/1]. Due to [BGR, 3.6.2/5] the latter is equivalent to the condition
F/ = Ei elf/where n = I-L:K ] is the degree of L over K, where e i IlL ri:f K r] is the
=

index of the valuation groups and wheref~ = [/~i : K ] is the degree of the residue
extension for the different values 1. li extending the given valuation of K to L for
i= l , . . . , t .
Stability of a complete field K can be checked by looking at finite separable
extensions L/K; cf. [BGR, 3.6.2/2]. For example a complete field K is stable if it is
discretely valued [BGR, 3.6.2/1] or spherically complete [BGR, 3.6.6/15] or
algebraically closed (just by definition). Any finite extension of a complete stable
field is again (complete) stable; cf. [BGR, 3.6.2/7]. The completion/~ of a stable
field K is stable too; cf. [BGR, 3.6.2/3]. In particular, if K is stable and ifL/K is
a finite field extension of K provided with its spectral norm, the completion L is the
norm-direct product
/2=i,• •
of finite extensions s of/~. Each s is K-cartesian. Moreover we have
do/. | (9s = 0 s = 0 s x ... x (f~s

An important theorem in rigid-analytic geometry is the result of Grauert-

Remmert and Gruson, respectively, that the field of fractions Q(K ( 31 . . . . . in)) is
stable if K is stable; cf. [BGR, 5.3.2/1].
If K is stable and if L/K is a finite separable field extension, the valuation ring
(.0L is a finite free Or-module; say with basis {El . . . . . f'n}. We have the notion of
the different ~L/K which is a fractional ideal on d0L whose inverse is given by
(~L/K) -1 = {X e L ; trl./K(xCL) c (~r}
which is generated by the dual basis of { ~'1 . . . . . ~', } with respect to the trace trL/K.
Moreover we have the notion of a discriminant
bL/K = (det (trL/r(f~fs))
where {fl, 9 9 9 f,} is a basis of OL over Cr. If L is a direct product
L=Ltx... xLt
of finite separable field extensions L~/K, the discriminant bL/K is defined as the
product
bL/K = bL1/K" 9 . . 9 bLt/g 9
Riemann's existence problem for a p-adic field 31 l

As in Serre's b o o k [S, chap. I I I ] , one verifies the following rules for the different
and the discriminant:
L e m m a 1.1 Let K be a stable field and let L/K be a finite separable extension of L.

(1) bL] K = NL/K(~L/K) G(-9K.

(2) Let L = L,1 x . . . x { , be the direct product decomposition of the completion of
L with respect to the spectral norm, then
~L/K = 9 s 1 6 3= ~s163 ... x ~s163
bL/K = b s 1 6 3 bs ... "bs163

(3) Let K be complete, in addition. Let M / L and L/K be finite separable field
extensions. Then

bM/K ---- N L / K ( b M I L ) . ( b L I K ) [M:L] .

(4) Let L = K(~) be a finite separable field extension with (-QL = (~K[ ~] and let
f ( C ) ~ K [ ~ ] be the minimal polynomial of ~. Then
~L/K = ( f ' ( ~ ) ) and bL/K = NL/K(f'(cO)
where f'(~) is the derivative off(C).
L e m m a 1.2 Let d ~ N be an integer. Then there exists for any prime p an integer
b = b(d, p) e N with the following property:
Let K be a stable field of characteristic 0 and residue characteristic p. Let L / K be
a finite separable extension of degree < d; i.e., L is a reduced finite K-algebra of
degree < d. Then the absolute value of the discriminant bL/r is bounded below by IP ]b.
Proof Due to 1.1(2) we m a y assume that K is complete. After replacing d by d! we
m a y assume that L / K is a galois field extension. By the usual techniques of
algebraic n u m b e r theory it suffices to treat the following cases by using the rules of
1.1:
(1) L/K is totally ramified with separable residue exstension.
(2) L / K is of degree p with radical residue extension of degree p.
The first case was already done by Hensel; cf. Serre IS, p. 58]. F o r the second
case we have an integral basis of the type

which reduces to a basis of the residue extension. N o w C satisfies an equation

f ( ~ ) = 0 where
f(~)=4"+al~P-1 +... +ap
with lal[ < 1. Then the derivative
f,(r = p ~ p - t + (p _ 1)al~p-2 + . . . + ap-x

has absolute value

If'(r = M a x { l ( p - i)a,I; i = 1. . . . . p} ,
312 W. Lfitkebohmert

since 4 ~ 4 i . . . . . 4 p-1 induces a basis over the residue field. Thus we see
If'(4)[ => IP[ and the claim follows from 1.1(4). []
In the next section we will use these considerations in a geometric situation.
Definition 1.3 Let K be a p-adic field; i.e., a complete field with respect to
a non-archimedean absolute value 1.1. Assume that K is algebraically closed; in
particular K is stable. Let m be the maximal ideal of (9K. Let X be a smooth rigid
space and let
~ : x - - , D(R) := {z ~A~; Izl _-< R}
be a finite flat morphism of degree d which is generically 6tale. For p e l K * [ set
X ( p ) = ~0-1(D(p)) and

C~ = { f e Cx(X(p)); If[ < 1}

C~ = {feC~(R)(lD(p)); If[ _-< 1}.
Then C~176 is a finitely generated and torsionfree module over
(9~176 = K [ T/p] of degree d. Thus it is a f r e e / { [ T/p]-module of
rank d and, hence, (9~ is a free (9~ of rank d; cf. [BGR, 6.4.2].
Then denote by b(p) the absolute value of the discriminant of (P~ over
C~ Recall that the field of fractions of C ( D ( p ) ) is stable; [BGR, 5.3.2/1].
Remark 1.4 If n is the number of zeros (with multiplicities) of the discriminant of
(S(X(p2)) over (9(D(p2)), then b(pl) > (p~/p2)"b(p2) for PI ~ P2 ~ R.
Corollary 1.5 Keep the situation of 1.3 and assume that K is of characteristic zero
and of residue characteristic p. Then b (p) > IP [bwhere b = b(d, p) is the bound of 1.2.
For later use we provide a calculation in a special case. The following lemma is
well-known; see [BGR, 9.7.1/1].
L e m m a 1.6 Let K be a p-adic field with residue characteristic p. For el, e2 e [K*I
with el < e2 consider the annulus

a ( g l , g2) = {4 E~m,K; gl ~ 141 < ~2} 9

An invertible function (o on A(el, e2) c a n be written in the form
q~(4) = ~ ~oi4i = c ' 4 a ' ( 1 + h)
ie;r

with c e K * and ]h[ < 1 where

h(4)= ~ h~'e(gK<(4/e2),(4/ei)-i>.
ieZ-{O}

The integer d = ord(~p) is called the order of ~o. The coefficients of h satisfy the
conditions
[hi[p i < l for all ie77 and for all p with ~i < P < e2.
If there is given a finite btale covering of degree d
(p:A(gl, g2)--+A(g~, l/ca), q~(~) = ~a(1 + h) with Ihl < 1
Riemann's existence problem for a p-adic field 313

its derivative
~o' = Y. i~o,~ ~-1 = d U - l ( 1 + h) + U h '
iEZ

is invertible where h'(~) = Y ihi~ i- 1. So it has a well-defined order

a=ord(~p')=d- 1 +v.
(1) In the case v = 0 one has
tihi[p~ < [d[ .['or all i ~ 7l and for all p with ~l < P < ~2.
(2) In the case v =# 0 one has
[ d l < l v h v l p ~ for all p with ~l < p < e2 .
Remark. The case (2) can only occur if ]d] < [v[. The latter is possible only if
p divides d.
L e m m a 1.7. Let K be an algebraically closed p-adic field with residue characteristic
p. Let el, ~2 e i K * [ and consider a finite btale morphism

~o:X=A(~,~2)~ Y=A(~,,e2), ~-~a'(1 +h),

where
h(~)= ~ h~ ~
vEZ - {0}

is a holomorphic function on X with sup-norm [hi < 1. Let

a=ord(~o')=d- 1 +v
be the order o f the derivative (p'. For p ~ [K*[, let b(p) be the absolute value o f the
discriminant o f the restriction

r : X(p) = A(~/p, , ~ ) ~ Y(p) = A ( p , p), ~ ~-~ ~a.(1 + h ) .

Then b(p) is given by the following formulas
(1) I f ~r = d - 1, then b(p) = [dl a.
(2) l f a = d - 1 + v with v ~ O, then b(p) = Ivh~l d. IPl ~ .
In any case the absolute value o f the discriminant is proportional to p L
Proof. Let ~ resp. r / b e the coordinate on X resp. on Y. So we have

The differentials are related by the formula

dr/= (~i~p~-l)d~ = (d + v)(p.+~ ~ ( 1 + g ) d ~ ,
where g is a function on X with sup-norm [gl < 1. N o w x and y which are defined
via
= p l / a x and ~l = PY
yield integral coordinates on X ( p ) resp. Y(p). Their differentials are related by the
formula
pdy = (d + v)q)a+~ pold x~(1 + g)pl/adx
314 W. L(itkebohmert

and, hence, since a = d + v - 1, we get

dy = (d + V)~pd+v p~/d x~(1 + g ) d x .

Thus we see that the different is generated by (d + v) (Pd+vp~/d and, hence, the
d i s c r i m i n a n t is generated by its d-th p o w e r so that
b ( p ) = I(d + v)fpd+vld'p v.

In the case v = 0 we have I~0dl = 1. In the case v :# 0 we have ~Pd+~ = h~ a n d

Id + v[ = Iv[, since the case v :# 0 can only occur if Idl < Ivl. So we are d o n e in b o t h
cases. []

2 Etale coverings of p-adic discs and annuli

In the following let K be an algebraically closed p-adic field with an a b s o l u t e value

[. ]. T h e m a i n results of this section are the following theorems.

Theorem 2.1 Let d E N be an integer. Then there exists for any prime p an integer
b = b(d, p) ~ N with the following property: Let K be an algebraically closed p-adic
field of characteristic zero and of residue characteristic p. Then any btale finite
covering
~ o : X --* l D ( e ) = { z e A l ; Izl =< R}
of degree d is split over D ( R ' ) where R' = Ip[bR.
Theorem 2.2 Let d ~ N be an integer. Then there exists for any prime p an integer
b = b(d, p) e N with the following property: Let K be an algebraically closed p-adic
field of characteristic zero and of residue characteristic p. Then any btale finite
covering
qo : X--* A ( r , e ) = { z ~ A 1 ; r=< lzl _-<e }
of degree d is of Kummer type over A(r', R') where r' = r/[p[ b and where R' = Ip[bR;
i.e., it is isomorphic to a sum of morphisms of type

A(~ ~x/S) --* A(s, S), ~ ~--* ~a.

The n u m b e r b = b(d, p) in 2.1 is given as a sum of the n u m b e r p r o v i d e d in 1.2 a n d
the smallest r E N with d < pr. In 2.2 one can choose the n u m b e r b = b(d, p) as the
sum of twice the n u m b e r a p p e a r i n g in 1.2 a n d of the n u m b e r r(1 + I/p) where r is as
above.

Notations/Lemma 2.3 Let X be a smooth rigid space. Let

~o:X--* A ( r , R )
be a finite and flat rigid morphism which is generically ~tale. Then there exist numbers
co=r<cl <c2< ... <c,<c,+l=R
such that the restriction
q~i = ~ola, : Ai = X x a~,, R) A~ - 1, ci) ~ A~ 1. c~)
Riemann's existence problem for a p-adic field 315

splits into a disjoint sum o.f ~tale maps o f the type

(P(~Lj) = ((i,j) d.... (1 + hi, j) with di, j ~ 1 and ]hi, jl < 1

where

A~ ei) = {2 ~Gm, K, Ci-1 < Izl < ci} ,

Ai = Ai, 1 ~ . 9 9 u Ai.~i) is a disjoint union o f open annuli,

Xi = ~o-l({z; Izl = c i } ) = Xi. 1 u . . . w X~.x~i~

is the decomposition into connected components. Furthermore set

aij = o r d ( ( ~ l A i . j ) ' ) = di, j - 1 + vl,j

ai=ai, 1 + ... +~i,~li)=d-6(i)+ vi

Vi = VI, 1 "~ . . . + Vi, 6(i)

the total (norm) order o f the derivative o f tp over Ai.

P r o o f This follows easily from the stable reduction theorem. Namely, due to
[-GR2] the affinoid space X admits a formal model :T over the formal spectrum
Spf(R) of the valuation ring R with geometrically reduced special fibre ,T | k. Let
~" | k be a compactification of,T | k which is normal at infinity. Since the residue
field k of R is perfect, the special fibre is smooth at infinity. After increasing the ideal
of definition of R, we may assume that ~ | k is induced by a projective Ro-scheme
~ o . Let ~'o be an affine open subscheme of,To which is smooth over Ro and which
contains infinity. There exists a smooth affine formal scheme ~ over Spf(R) which
lifts ~'o. N o w we may view f o as a result of a gluing procedure between ~ o and
~ o . Since the intersection of .To amd ok'o is affine, due to the lifting property of
smoothness the gluing m a p lifts to a formal gluing datum between ,T and ~ . Thus
we obtain a proper flat curve ,T over Spf(R) which contains Y" as a dense open
subscheme and which has s m o o t h rigid fibre, since X is smooth over K. N o w Y" is
algebraic, so the generic fibre .T @ K is a smooth projective curve over K. Due to
[ D M ] or [BL, 7.1] there exists a semi-stable model over Spf(R) which we call
again such that ~ is a dense open subscheme of S . Let Y~o~. . . . . . T~ be the
irreducible components of .To and let c p be the supnorm of ~0 with respect to
a dense open subscheme o f ~ for p = 1. . . . . r; see [BL, Sect. 3]. Thus we see that
the set of rigid points x e X with Icp(x)l =~ c o for p = 1. . . . . r is contained in
formal fibres of ~ . Due to [BL, 2.2 and 2.3] these formal fibres are isomorphic to
open discs or to annuli. Since r has no zeros, q~ is of the type as asserted after
arranging the numbers c ~. . . . , c ~ in increasing order. Since there are only finitely
m a n y points where (p is not 6tale, the claim follows by adding to the critical values
the absolute values which (p attains at these points. []

Remark. The numbers a~,j do not depend on the chosen coordinate ~,j respecting
the orientation of the target A(r, R). Also it does not depend on the coordinate of
A(r, R ) if one respects the orientation.
If one chooses the opposite orientation, the numbers a~,j have to be trans-
formed into ( d i , j - 1 - vi, j).
316 W. L/itkebohmert

L e m m a 2.4 Keep the situation of 2.3. I f the canonical divisor on X is trivial and if ~o is
btale over the boundaries Ai and Ai+ l, the following formula holds for i = 1. . . . . n
a,- at+l = 2. [g(Xt) + (6(i + 1) - x(i))] - deg(dqglXt)
where 9 ( X i ) is the genus of Xt (see below) and where
x(i) = number of connected components of Xi,
6(i) = number o f connected components o f At.
Proof. We need a compactification Xt of Xt which is obtained in the following way.
ForeEJK*lwithct_l<c~--eandct+e<c,-+lset
X~ = qg-l( A(ci -- ~.,C i "Jv e ) ) .

Then attach closed discs along the annuli A .~,,j.resp. A [+ 1, ~ on the b o u n d a r y of X~ in

order to get a proper curve. So we compactified Xt by X~ by adding open discs to
Xi. I n each disc choose a point which does not belong to X~. Then approximate
q~ by a rational function over this compactification X~ which has poles at most in
the chosen points. M o r e o v e r make the approximation q~ so strong such that the
n o r m of q~ - ~b is less than the norms of cp' over any A,,j~9and over any A ~+ L j . N o w
we have for the degree of the canonical divisor

2g(X~) - 2x(i) = deg(dq0);

note that x(i) is the n u m b e r of connected components of Xi. Just by definition we

have g(X~) = g(Xt). M o r e o v e r d<b and &o have the same numbers of zeros on Xt,
since the canonical divisor is trivial and since dq0 and &p have the same n o r m at the
b o u n d a r y o f X t ; cf. [BL, 3.1 and its remark]. Thus it remains to compute the degree
over the discs. Since the canonical divisor is trivial over a disc, the latter can be
c o m p u t e d by the n o r m degree ofdcb over the annuli A?,j resp. A~+ ~,j. Since we have
to respect the orientation, the order of dq0 over the disc which appear by gluing
over A~+I,j is the negative of the order of +p' minus 2. Thus we see how the
s u m m a n d ~(i + 1) appears. []
L e m m a 2.5 Keep the situation o f 2.3 and assume that ~p is &ale over A~ e,+ 1)-
Then
(a) (6(i+ 1)-x(i))>=0 for i = O. . . . . n
(b) a t > a i + l for i = 1. . . . . n
(c) trt=ai+l,~g(Xi)=O and J ( i + 1 ) = x ( i )
(d) ai >=vt for i = 1. . . . . n + 1
(e) ~ri = at + 1 ~ v+ > vt + 1
Proof. (a) 6(i + 1) ~ x(i) because the m a p ~o is finite.
(b) follows from (a) and 2.4, since deg(dq~lXt) = 0 for ~p is 6tale.
(c) follows from (a) and 2.4.
(d) O"t 2 j a i , j = ~ ' 4 ( d t , j - 1 + vt, j) = deg(tp) - 60) + vi > vt
=-

(e) deg(cp) - 60) + vt = at = tri+l = deg(tp) - 6(i + 1) + v~+l.

Thus we see by using (c)
vt-- V~+l = 6 ( i ) - 6(i + 1) = 6 ( i ) - - x(i) > 0
since q3 is finite. []
Riemann's existence problem for a p-adic field 317

L e m m a 2.6. Keep the situation of 2.3 for a morphism

( p : X ~ ID(R) = A(0, R)

Let b(p) be the absolute value of the discriminant of q~l~{p) as in 1.3. Then the absolute
value of the discriminant is proportional to

b ( p ) . . . p ~' .for ci-l<p<ei a n d f o r i = l . . . . , ( n + 1).

The increase of the discriminant can be estimated by

b(Ci)/~(Ci- 1) ~ (Ci/Ci- 1)v'-

Proof. The map ~o : Ai, j --* A~ ci) is of the type

(P(r = ((i.j) d.... (1 + hi,t) with Ih~,~l < 1 .

Due to 1.7 we k n o w that b(p) is proportional to pV, in the asserted range; say

b ( p ) = 71"(P/CiF' for ci-1 < p < c~.

Due to 1.4, we have

b(ci)_-<Vi and 3'i'(ci-1/ci) ~' <=b(ci-1).

Therefore we obtain

b(ci).(ci_ l/c3~' <=~(ci_ ~) .

Thus the increase of b(p) from p = ci to p = ci_ ~ can be estimated by

b(ci) <=~(ci- ~)" (cJcl- 1Y' 9 []

Proof of Theorem 2.1 Let ~o : X ~ ID(R) be &ale finite of order d. Let

C0 = 0 < C1 < C2 < . . . < C n "~ Cn+ t -~- R

be the critical values; cf. 2.3. Keeping the notation of 2.3, let a~ be the total order of
~o[ over A~ cg). Since ~o is &ale at the origin we have a~ = 0. D u e to 2.5 we
know

0---~-0"1 ~ 0 " 2 ~ . . . ~qe+l=0>ae+2~... ~O-n+ 1 .

Then we conclude from 2.5

g(X~)=O and x ( i ) = 6 ( i + 1) for i<f.

Then the graph of coincidence of the components in the reduction is tree-like. So

we see

(P-I({z eID(R); Izl < r162 = D1 W . . . w Oar

where Do are open discs. Furthermore we can estimate the ratio (c~+ x/R) by using
the universal b o u n d provided by 1.5 and the estimate for the discriminant given by
2.6. Since v~ <=ai < 0 for i => (f + 2) we get
318 W. Lfitkebohmert

b(R) = b(c~+l)" I-[ (~)(ci)/~)(ci-1))

i>(r

=< b(c~+ 1)" 1-J (ci/ci-l) ''

i>(d+2)

_--<b(C{+ 1)" F[ ( c'/ci-~)-~

i__>(F+2)

< b(c:+ 1)" (cr + 1/R).

N o w we have the universal b o u n d due to 1.5, so we get
Ipl b <=b(cf, + l)'(c~ + l / R ) <=(c~ + I / R ) .
Thus it remains to study &ale finite maps from a disc to a disc
~o:ID--,IO, q,(r y~ a,,~'.
v_>_l

Since q) is finite of degree d we have

laat >]a,] for all yeN.

Since ~0' is unit, the coefficient al satisfies

lall>lvavl for all v>2.

After shrinking by the factor Id[ we get what we want. []

Next we want to show T h e o r e m 2.2. Let us start with a simple lemma which is
well-known.

L e m m a 2.7 Let R be a complete ring with an ultrametric valuation of height 1. Let

h ~ R be an element with

Ihl < IPl P x / ~ .

Then there exists an element ~ ~ R with
(1 + h) = (1 + #) p.
P r o o f o f Theorem 2.2 We proceed by a similar m e t h o d as before. F o r technical
reasons we (may) assume that the annulus is symmetric to 1; i.e., R -- 1/r. So let

q~ : X ~ A(r, l/r)
be &ale finite of degree d. Due to 1.3, by adding some values, we may assume that
the critical values are of the following type

c - ( n + l~ = r < c _ . < .... <Co= 1 < cl < ... < cn < c . + l = 1/r
c-l.c~ = 1 for i -- 1. . . . , (n + 1).

F o r the p r o o f we will use the following m a p

O:A(r, 1/r)~ID(1/r), O ( z ) = z + 1/z.
which is 6tale outside z = _+ 1. Then we will consider the composition
Fp = t~o ~o : X --+ A(r, l/r) --+ lD(1/r).
Riemann's existence problem for a p-adic field 319

We are interested in the critical values of ~ which are > 1. Obviously they are
given by Co,. 9 9 c, + 1. Next we want to calculate the norm-orders of the derivat-
ives ~' Id, which we want to relate to the orders of (o'14, and of ~p(~_. F o r doing this
we have to transform the orientation of the coordinates on the annuli above A - i -
F o r example consider a map

~ ( 4 ) = 4~.(1 + h).
O n the b o u n d a r y A + we have to choose the coordinate ~ and we have to compute
the norm-order of

,5+ (4) = ~ ( 4 ) + ~p(r - 1 9

The derivative is given by

~+(4) = q,'(~)- ~o'(~). ~o(4) - 2 .

Since we are interested in the case l~l > 1, the dominating term is given by (p'(~)
because I(P(~)I > 1 on this part. Therefore we obtain

ord((~tA +)') = ord((91a+ )').

O n the b o u n d a r y A_ we have to choose the coordinate q = 1/4. Then we have to
compute the order of the derivative of
~o_ (q) = qo(1/q) + (p(1/q) -1

It is given by
~'_ (q) = [ ~ p ' ( 1 / q ) - ~p'(1/q). ~p(1)/)-2] 9( - - q-z).

On the b o u n d a r y A_ we have Iq~(1/q)] < 1. So the dominating term is given by

d / d ~ [-(p(]/q)-- 1] = d / d q [ q d . (1 + h(1/q))- 1]

= d q a - l ( l + h(1/q)) -1 + qd'(1 + h ( 1 / q ) ) - Z h ' ( l / q ) ' q -2

Thus the norm-order of the derivative is given by

ord(b2)=d-(v- l)-2=2d-2-(d- 1 +v)

which is even true if a_ = (d - 1) or equivalently v = 0. So we have

ord((~lA_)') = 2 d - 2 -- ord((~OlA )').

Henceforth we obtain for the total order of the derivative of c~ over Ai

3-~ = ~ ord((~lA~j)') + ~ ord((~IA-i,j,)')

1 j,
= ~ ord((~plAi,j)') + ~ [2d-l,s, - 2 - ord((~olA _~j,)')]
j j'
= a/- a-i + 2d- 26( - i)

= (d - b(i) + vi) - (d - b( - i) + v _ i ) + 2 d - 2 6 ( - i)

= 2d -- [6(i) + ~( --i)] + [vi - v-i]

= d - 6(0 + ~
320 W. Lfitkebohmert

where
d = 2d, 6(i) = 6(0 + 6( - i), vi = vi - v - i .
Due to 1.7 the absolute value of the discriminant is proportional to
b(~3, p ) ~ p ~ forpwithci_l<p<cl for i=l ..... n+l.
By the following lemma we will see
~<0 for i=l,...,n+l.

L e m m a 2.8 K e e p the situation o f above and let ~ ~ {0. . . . , n}. Then

(a) ~<0 for i=1 ..... n+l

(b) vr for i=l,...,{.
Proof If vt + 1 > 0 then
0 < v.e+l = vt+l - v - ~ - i = a~+l + 6 ( f + 1) - a - ~ - i - 6 ( - ( - 1).
Henceforth we obtain

(1) a _ ~ _ , - a~+l _-< 6 ( ~ + 1) - 6( - ~ - 1).

Since r is &ale, the genus formula yields

(2) a - c - 1 - a~+l = 2 " [ g ( X ( c - t - 1 , c~+1)) + 6 ( ( + 1 ) - # ( X ( c _ ~ _ l , cr

where
X(c-t-1, el+l) = ~o-l(A~ c:+ 1))
# ( X ( c _ : _ 1, cr 1)) = n u m b e r of components of X(c_~_ 1, cr 1).
This formula can be proved in the same way as the formula of 2.4. Combining (1)
and (2) shows
- 6 ( ( + 1) - 6(- f - 1) > 2 . [ g ( X ( c _ e _ I , c ~ + I ) ) - #(X(c-~-l, ct+l))].
Since q~ is finite, we have
# ( X ( c _ t _ ~ , c f + ~ ) ) < m i n { 6 ( ( + 1), 6 ( - ( - 1)}.
Since g ( X ( c _ e _ a, ct+ 1)) > 0, this implies
g ( X ( c - l - 1 , ct+l)) = 0 and v/+l = 0 (and, hence, (a))
# ( X ( c _ ~ - l , c ~ + l ) ) = 6({ + 1) = 6 ( - E - 1).
D u e to (2) this implies

O" (_ 1 ~ O'f+ 1 .

Since a~ > ai+l for i = ( - E - 1). . . . . (E + 1) due to 2.5, we get

O'_f_ 1 ~ 0"_ t ~--- . . . ~-- O 'g ~ O'E+ 1 .

and, hence, by 2.5(e)

V-d-1 = Y-s ~" 9 9 9 ~ V d -'~ Y d + I

Riemann's existence problem for a p-adic field 321

and, hence,
0 = 171 = V2 . . . . . Vt'+I 9 []

Continuation of the proof of 2.2 N o w let { > 1 be the largest n u m b e r such that
i ; / = 0. In particular we have due to 2.8
0=vl=v2= ... =v/=0>Vi+l for all i>(.
As before in the proof of 2.1 we can estimate the increase of b(qS, p). Since 0 > vi+l
for i > (, we get from 2.6
b(q3, 1/r)= b(qb, ce+l). 1-[ (b((P,c,)/13(O, ci-1))
i > (,~+2)

<-- b((o,c~+l)" I~ (c,/c,-l);'

i>({+2)

< b(O, C(+1)" [I (<-1/<)

i>({+2)

< b(O, c~+,).(c~+ ~/c,+,).

N o w we have the universal bound due to 1.2. Since ~ splits over the b o u n d a r y into
two components of degree d, we get
Ipl 2b < b(q3, l/r) < b(q3, CE+I)*(Cg +I/Cn+ I) ~ (ce +l/R) .
Thus it remains to study what happens over the annulus A~ 11, c(e+ 1)). We
know ~7~= 0 for 1 < i _< E. As seen in the proof of 2.8 we have
g ( X (c_~_ 1, c~+ 1)) = 0
# ( X ( c - e - 1 , ce+l)) = 6(f + 1) = 6 ( - f - 1).
This condition implies that the configuration of components above
A~ 2b, Ipl2b/r) is tree-like where any slope is an annulus. Thus we see that any
connected c o m p o n e n t of ~0 -1 A~ ce) is an annulus. Then we are reduced to
study &ale finite maps from an annulus to an annulus

(o: A (~/~, ~ i ~ ) ~ A(~, 1/O

q~(~)=~(l +h) with Ihl<l and h(1)=O.
We must show that we can extract the d-th root of(1 + h) after shrinking by a fixed
amount. Write
d=p*d' where (p,d')=l.
Now
h= ~ hv~ ~'
v~Z-{O}

satisfies the inequalities

[hvl_-<e Ivl for vs7/.

By shrinking by the factor (I PI" ~ ) ' we conclude from 2.7 that the d-th root of
(1 + h) exists. Thus we can write
(1 + h ) = ( 1 +g)n.
322 W. Lfitkebohmert

Therefore we obtain a commutative diagram

A(~,fe,~) ~ A ( d~ , ~ , d~ ) , ~ w - ~ ( l + g(~)),

A(~, 1/~) -2 A(~, l/e), ~ ~ ~(1 + h(~)),

where Z is the K u m m e r m a p ~ ~ ~d.
Thus we are done with the p r o o f of Theorem 2.2. []
Corollary 2.9 Any finite btale covering q) : X--* (ID - {0}) extends to a (ramified)
finite covering of ID.
Proof The restriction of q~ to the open annulus A~ IplbR) is a covering of
K u m m e r type. []
In the case where the characteristic of K is positive the situation is different as
seen by the following example.
Example 2.10 Let K be a p-adic field of characteristic p > 0 with a discrete
(multiplicative) valuation g r o u p c z. Consider the rigid variety
E* = V(~P - ~ - h) = ~ x ~)*
where/~ is the affinie line with coordinate function ~ and where ID* is a punctured
disc with coordinate function q and where

i~N

Ikil=<l and ]kilq~c pZ and Iki+l]<lkil p for ieN,

then the projection q~ : E* --* ID* is an example of a finite &ale covering which does
not extend to a finite covering E --, 1D.
Proof Due to the choice of h the polynomial ~ e _ ~ _ h is irreducible, since
h cannot be written in the form e p - e for some e e J/d(lD*) as we will see below. So
E* is a geometrically irreducible s m o o t h rigid variety. N o w let us assume that
p extends to a finite covering E ~ ID. Due to the theory of Artin-Schreier there
exists an element e e (9(E) with J I ( E ) = dg(lD)(~) such that c~satisfies an equation
T p-T-q~*a=O
for some a e ~'(ID). But this is impossible. After replacing c~ by fc( for some
( e { 1. . . . . p - 1}, we may assume that h and a differ by an element of the type
(1) h - a = ep - e
for some e 9 J/g(ID*); [A, chap. V, w 11, n~ Example 2]. The Eq. (1) shows that
e can have only finitely m a n y poles. N o w look at the Laurent series expansions
a r o u n d the origin
e = 2 e~tl ~, h = 2 h, tl ~, a= 2 a~tl ~.
v~Z v E 7~ v ~ n

Thus we see for v = - p~ < n

(2) ki+ l = hpv = e f - ep~ = ef - ei+ l
Riemann's existence problem for a p-adic field 323

where
e_p, : 8 i.

Since Ikll r c pz and ef ~ c vz, the ultrametric valuation gives

max{Ik/+ll, lc~I} =[ki+l -8~1=18i+1[.
In particular, we see lei+ 1 [ > Iki+ 1 I. Since l ki+ z [ < }ki+ 11p, the Eq. (2) implies by
the same trick

[e~§ = I~i+zl .
Thus we can conclude for large i e N
(3) I < + j l = 1811p' for all j~N.
The latter condition implies that the Lauerent series of e is not converging on
ID*. []
We want to mention that any finite 6tale map X ~ (D 2 - {0}) extends to
a finite &ale m a p X ~ ID2. But the method of its proof is different from the one
given in the case X --* ID*. Finally let us discuss the case of tame Galois coverings in
positive characteristic.
Theorem 2.11 Let K be an alyebraically closed p-adic field of residue characteristic
p and let G be a finite 9roup which is of order prime to p if p is positive.
(a) Any finite &ale 9alois coverin9 ~o 9X --* 1D(R) with yroup G is split.
(b) Any finite &ale 9alois coverin9 go" X ~ A(r, R) with 9roup G is of Kummer type.
Proof. Consider the situation of 2.3, Since ~o is galois of order prime to p, any d~,~ is
prime to p and, hence, all vi, j are zero. In particular, we obtain cri,j > 0 for all i,j
and, hence, cri > 0 for all i. So we obtain in case (a) by 2.5
0=cr~ >a2 >
~ 9 . . >
~_ Gn+l >0
and, hence, at = 0 for all i = 1. . . . , n + l.This means dl..i = 1 for all i,j. So
x(i) = 6(i) = 6(n + 1) = d
and, hence,
go-~(D~ = D o u ... ta D O

w h e r e go gives rise to a n i s o m o r p h i s m b e t w e e n D o a n d D O for a n y b. N o w we w a n t

to extend this result to the whole disc. Therefore consider the morphism of formal
schemes
: .~ = Spf(C~ -~ ~ = Spf((9~
which is a galois covering. F o r any closed point in the special fiber y e @o the
inverse image
9 = .... ,

consists of d points which are s m o o t h on f o by what we have proved above; use

[BL, 2.2]. Thus we see that 4~o is &ale. Since a finite 6tale m a p to the affine line like
q% which is galois of order prime to p splits into a disjoint union of isomorphisms,
the claim follows.
324 W. L/itkebohmert

In case (b) we can proceed as above; but we get only the result over the open
annulus A~ R). N a m e l y we obtain as above
- 5(i) + a(i + 1) = ai - ai+ ~ = 2 [ g ( X i ) + 6(i + 1) - x(i)]
and, hence,
-- 6 ( i ) -- (5(i + 1) = 2[0(x3 - x(i)].

Since x(i) < Min {6(i), 5(i + 1)}, we get g(XO = 0 and x(i) = cS(i) = 5(i + 1) for all
i. Therefore ~ p - l ( A ~ ,, c,+ 1)) is an annulus and, since the degree of (p is prime
to p, the m a p ~0]ao(. . . . . . . . . . ) is of K u m m e r type. N o w consider the map on the
boundaries
X + = (p-I(A(R,R))-+A+ = A(R,R)
X- =~o-l(A(r,r)) --+ A - = A(r,r).
By using formal models Y" -+ --+ .~49 as above one sees by using assertion (a) that
s + __+aj • is &ale and, hence, of K u m m e r type. This implies by combining it with
the result over the open annulus that ~o-l(A(r, R)) is an annulus and, since the
degree of ~o is prime to p, the m a p (P[a(,, R) is of K u m m e r type. []

3 The comparison theorem in c h a r a c t e r i s t i c zero

In the following let K be a p-adic field (i.e., a complete field with respect to
a non-archimedean absolute value I" l) of characteristic zero. The aim of this section
is to show the following result:
Theorem 3.1 Let K be a p-adic field of characteristic zero. Let X be a K-scheme
which is locally of finite type over K, and denote by X the rigid-analytic space
associated to X via the GAGA-functor.
Then the GAGA-functor from the category of finite &ale coverings of X to the
category of finite rigid-analytic ~tale coverings of X is an equivalence of categories.
F o r the p r o o f we follow mainly the notes of Grothendieck in [ S G A 1, Exp. XII,
Sect. 5 (pp. 332-343)] which uses Hironaka's result on resolution of singularities.
So we describe only the part which has to be changed. As in the notes of
Grothendieck it suffices to consider the case where X is affine and smooth and
where X is an open subscheme of a smooth projective scheme X ' such that the
complement
A = (X' - X)
is a divisor with n o r m a l crossings. Due to G A G A it suffices to show that any finite
rigid-analytic &ale covering
Y--+ X
extends to a finite (ramified) covering Y' --, X ' . For the p r o o f we start with the
following generalization of the main result 2.2 of the last section. In the following
let K be a p-adic field of characteristic zero and of residue characteristic p.
Lemma 3.2 Let S be a smooth connected affinoid space over K. Let
~o: Y ~ S x A ( r , R )
Riemann's existence problem for a p-adic field 325

be a finite btale covering of degree d. Let 0.i: S ---, Y be sections for i = 1. . . . . d which
describe the inverse image of the unit section

c'S~SxA(r,R).

Then ~o is of Kummer type over S x A(r', R') where r' = r/lp] h and where R' = Iplb R
with the number b = b(d, p) e N of 2.2.
I f Y' = (o-1 (S x A (r', R')) is connected, there is a unique choice of the coordinate
q on Y' such that qd is the coordinate ~ of A(r, R) and such that 0.'?q = 1.

Proof In the absolute case where S is a field, the assertion follows from 2.2 by using
Galois descent. N o w consider a continuous valuation v on S of height 1 which
extends the valuation of the base field K; cf._ [ B ] p. 12. Let V = Sp(K(v)) be the
spectrum of the valuation field K(v) of v, let K(v) be the completion of K(v), and let
P = Sp(/s be its spectrum. Let r be the coordinate of A(r, R). Due to the
absolute case there exists a function on l? x s Y' which gives rise to a coordinate on
the connected component of V x s Y ' containing 0-1 such that f/~(")= ~ with
0.* f / = 1. A priori the order d(v) of the K u m m e r type covering depends on v. Then
there exists a neighborhood U of v and an approximation ~ on U x s Y such that

I(q nw) - ~)lvkup < ([Pl P ~ ) ~

where d(v) = p~(~)d' with (p, d'(v)) = 1. Due to 2.7 there exists a function r/ on
U x s Y' such that r/dw) = ~. It is clear that 0.*r/= 1. Since we can construct an open
affinoid neighborhood for any continuous valuation of height 1, we obtain a finite
covering of S; cf. [B, 1.2.1]. Thus we see that, due to the connectedness of S, the
order d(v) depends only on 0.~. Since the choice of the coordinate is unique, the
claim follows by repeating the construction with the other connected components
of ~p-l(S x A(r', R')) and suitable sections 0.i. []

L e m m a 3.3 Let S be as in 3.2. Let r > 1 be an integer. Let Dr be the unit ball with
coordinates zl . . . . . z , Set A = V(Zl . . . z~) and set ID~ = D r - {0}. Let

qo " Y--+ X = S x Dr

be a finite covering of degree d_ which is btale outside S x A. Then q) extends to

a (ramified) finite covering O" Y ~ S x D L

Proof Set ~ = pb where b e n is the number of 2.2. For r = 1 the claim follows from
the last statement. First use the finite &ale base change S ' - , S by the universal
splitting space of the fibred product of Y over S x D ~ and the section
(id, e ) ' S ~ S x D ~. So the inverse image of (id, e) splits into sections over S'. The
canonical descent datum on q~ extends to a descent d a t u m on 0 ' because Y' is the
normalization of X ' in Y. Since finite 6tale descent is effective, 0 ' descends to
a morphism 0 defined over the given base. N o w assume r > 2. Set

A(~) = {z e lDq Izil < I~1 for i = 1. . . . , r} ,

r(e) = ~0-~(s • ~(~)).

Let n o be the winding number along V(zp) for p = 1. . . . . r. T h e n set

Z ( ~ ) = V ( ( T~ - z ~ ) " ' , . . . . ( T , - z~) "~) = ~ • .

326 w. Lfitkebohmert

Due to the case r = 1 we obtain a cartesian diagram

r(~) ~ {z~ 4= 0} ~ z(~)
+ [] +
(S x A(O)c~ {z~ 4: 0} ~ (SxA(0)
for p = 1. . . . . r. These maps fit together to build a cartesian diagram
r(~) ~_. z(e)
+ [] l
( S x ( Z ( e ) - {0})) ~ (Sx~(~)).
This means that q~ extends to a ramified covering over S x ID'. []
Proof of Theorem 3.1 Consider a finite rigid-analytic 6tale covering
q~: Y ~ X
We have to show that ~o extends to a finite (ramified) covering
(p':Y'~X'.
The latter follows by induction on r from 3.3 and the existence of tubular
neighborhoods of smooth subvarieties in a smooth space; cf. I-K1, Theorem
1.18]. []
In the next section we will give a different proof in characteristic p > 0 which
uses an extension theorem on coherent sheaves and the resolution of singularities
in dimension 2. That method works in characteristic zero too; so one can avoid to
use Hironaka's resolution of singularities in its full generality. But nevertheless one
uses involved methods. One can also try to adapt the geometric reasoning of the
original paper by Grauert and Remmert to the p-adic case so that one might obtain
a more elementary proof.

4 The comparison theorem in positive characteristic

In this section we will solve Riemann's existence problem for finite 6tale galois
coverings of order prime to the characteristic of the base field. The main result is
the following theorem.
Theorem 4.1 Let K be a p-adic field of positive characteristic p and let G be a finite
group with order prime to p. Let X be a K-scheme which is locally of finite type over
K, and denote by X__the rigid-analytic space associated to X via the GAGA-functor.
Then the GAGA-functor from the category of finite btale galois coverings of
X with group G to the category offinite rigid-analytic btale galois coverings of X with
group G is an equivalence of categories.
For some parts we follow the method described in the last section. We may
assume that the base field is algebraically closed. As before it suffices to consider the
case where X is affine and smooth and where X is an open subset of a normal (not
necessarily smooth) projective scheme X'. Let
d' = (X' - X)
Riemann's existence problem for a p-adic field 327

be the complement of X in X'. Due to G A G A it suffices to show that any

rigid-analytic 6tale covering
~o: Y ~ X
extends to a finite (ramified) covering (p" Y' ~ X'. Providing A' with its reduced
structure, we know that there is a closed subscheme S' of X' of codimension 2 such
that X' and A' are smooth outside S'. So, due to 2.11 we can apply the same
reasoning as in Sect. 3 in order to show that the given covering extends to a finite
(ramified) covering
~.X~(x,- s,)
which we denote by the old symbols too. Since the normalization of Y is also finite
over Y, we may assume that Y is normal.
It suffices to see that the coherent (gx-module ~ = q~.Cy extends to a coherent
Cx,-module; here and in the following we always mean coherence in the analytic
context. Namely, if it extends to a coherent sheaf, the direct image ~ ' = l,g~/will be
a coherent (9x,-module where
z:(X'-S')~X' I

is the inclusion map. Then Yr is a coherent sheaf of Cx,-algebras and, hence,

Y-- = Sp(~ ') gives rise to a finite covering ~0' 9 Y' -~ X ' which extends the given q~.
As a first step we will show that ~ extends to a coherent sheaf on ( X ' - S")
where S" is of codimension 3 in X'; this will be done by using the desingularization
of two-dimensional schemes. The final step will be proved by a general result on
extension of coherent sheaves in the style of Frisch and Guenot [ F G ] .
For the first step we need the following result on desingularization of two-
dimensional schemes; cf. [L] and [Abl, (0.2)].
Let P be a noetherian, normal, excellent scheme of dimension 2. Let A be a closed,
reduced subscheme of dimension 1. Then there exists a blowing-up 7r: P' ~ P of an
ideal of codimension 2 such that P' is non-singular and such that ~ - I ( A ) with its
reduced structure is a divisor with normal crossings.
Now consider the scheme X ' and let S' = X' - X be the complement of X in X'
equipped with its reduced structure. Then consider a generic point t/of the singular
locus of X' or of the singular locus of A' and assume that r/is of codimension 2 in
X'. Then we apply the result on desingularization to the scheme P = Spec(G'x, 7)
and the divisor A',. So there exists a coherent ideal J of G'x,,~ with V ( J ) = {q}
such that the blowing-up n : P ' ~ P of J becomes non-singular and the inverse
image n - 1 A', has normal crossings on P'. We may assume that J is induced by
a coherent ideal f# of Cx, such that t/is the generic point of V(N). So we can look at
the blowing-up
rc" X"---~ X'
of f#. Since blowing-up commutes with flat base change, the singular locus of X"
does not meet the fibre 7r-l(q). Then consider the inverse image A" = ~ - I A ' . By
the same argument as above, we see that the singular locus of the irreducible
components of A" does not meet ~ - l ( q ) and that they meet in normal crossings
above 7z-l(r/). Summarizing we obtain the following result:
There exists a closed subset T' of X ' of codimension 3 such that X" is
non-singular above X ' - T' and such that A" has normal crossing above X' - T'.
328 w. Lfitkebohmert

Because the base field is algebraically closed, X" is smooth above X ' - T' and
A" is a normal crossing of smooth divisors above X ' - T'. So we can apply the
same reasoning as in the last section to show that the given finite rigid-analytic
covering (p " Y ~ X extends to a finite covering
(p" 9Y'.~'-+ (X" - T"]
where T" = ~ - 1 T'. Since the m a p n : ( X " - T") ~ ( X ' - T') is projective, the
direct image n,~o" (fir,, is coherent on ( X ' - T'). This shows that ~o, (fly extends to
a coherent sheaf on ( X ' - T'). Since we m a y assume that Y'__[is normal, we know
that the closed subscheme where the depth of ~,+, ,,+n ~v,, is less than d i m ( X ' ) is of
codimension 3. N o w the claim follows from the following extension theorem for
coherent sheaves; cf. [ L u l , w Satz 1].
Theorem 4.2 Let X be an affinoid space o f pure dimension n + 3 and let S be a closed
subvariety o f X o f dimension < n. Let ~ be a coherent (fix - module on (X - S)
which has depth > (n + 3) at all points. Then the direct image t , ~ is a coherent
(fix - module on X where l : (X - S) + X is the inclusion.
Proof. There exists a finite m o r p h i s m
rc : X --~ ID n+3

such that
7 ~ - l ( I O n x (ID 3 -- IO3(~)) ~ ( X - S)

where ID3(e) is t.he open 3-dimensional ball a r o u n d the origin of a suitable radius
with e < 1. Then consider the sheaf
~r = ~ * ( ~ l ~ - ' ~ " • (~3_~3~)
which is a coherent sheaf on (ID" x (ID 3 - D3(e)). The depth condition implies that
ff has depth (n + 3) at each point and so that ff is a vector bundle on
(ID" x (D 3 - ID3 (e)). By the usual techniques of sheaf extension one shows that, for
any 6 with g < 6 < 1, the restriction map
Hq((ID n x (ID3 -- ]D3(~)), .C~) -~ Hq((ID" • (ID 3 - ID3(6)), ~ )

is bijective for q = 0, 1. Since e < 6 the above c o h o m o l o g y groups are finitely

generated (fi(ID"+3)-modules; as seen by the methods of Kiehl [K2, w By the
usual c o h o m o l o g y argument one deduces that there exists a morphism

such that the cokernel has support of dimension < (n + 2). So the problem of
extending ~ and, hence, of ~ becomes a problem of extending coherent subsheaves
3V on (X - S) of a coherent sheaf ~ on X. The latter is related to extension
problems for m e r o m o r p h i c functions and for subvarieties. Of course it requires
some effort, but this has been done in [ L u 2 ] by studying a rigid analytic version of
the p r o g r a m of Siu. []
Actually we do not need the full assertion of 4.2; it is enough to know that there
exists a coherent subsheaf of ~ of rank equal to the rank of ff which is generated by
its global sections. This is precisely the part we have shown in the p r o o f of 4.2.
Following an idea of Bingener, we will briefly indicate how to finish the p r o o f of 4. !
by that partial result.
Riemann's existence problem for a p-adic field 329

Indeed, for the remaining p a r t of the p r o o f we can stick to o p e n aMnoid parts of

X___~';call it X__~'again. T h e n look at the field extension given by the m e r o m o r p h i c
functions

K = ~x,(X') = ~llx,(X__) <- >L = o11~(s

The partial result, we have shown in the p r o o f 4.2, implies
[ L : K ] = deg(~o).

N o w consider the integral closure

B' = { ( e L ; ( integral over A ' = (9(X')} .

Since (?(X--) is excellent, B' is a finite module over A' = g?(X') = (C(X). Moreover,
if U = Sp(A) is an open affinoid subdomain of X__,then

B' | A = { [ ~ L |K Q( A ); ( integral over A }

because A'~A is a regular morphism; cf. [EGA, |V2, 6.14.4]. Since

[ L ' K ] = deg(rp), the inclusion
L| ~~ ~(r(~P ~ ( U ) )
r

is an isomorphism. So we o b t a i n the i s o m o r p h i s m

B' | A c- ~> 6)r(cp-l(U)) .

A'

Thus we see that the canonical m a p

g' = S p ( B ' ) --, Sp(A')

extends the given morphism ~o" Y--* X. []

Acknowledgements. I'd like to thank the I.H.E.S. for its hospitality and especially I'm indebted to
O. Gabber for many interesting discussions. Moreover I'm grateful to the referee for suggestions
how to improve the original version of Sect. 1.

References

[A] Bourbaki, N.: Alg~bre, Chap. 4 fi 7. Paris: Masson 1981

[Abl] Abhyankar, S.S.: Resolution of singularities of embedded algebraic surfaces. New
York: Academic Press 1966
[Ab2] Abhyankar, S.S: Resolution of singularities of algebraic surfaces. In: Algebraic
Geometry, pp. 1-1 l. London: Oxford University Press 1969
[B] Berkovich, V.: Spectral Theory and Analytic Geometry over Non-Archimedean
Fields. (Math. Surv. Monogr., vol 33) Providence, RI: Am. Math. Soc. 1985
[BGR] Bosch, S., G/intzer, U., Remmert, R.: Non-Archimedean Analysis. (Grundl. Math.
Wiss., Bd. 261) Berlin Heidelberg New York: Springer 1984
[BL] Bosch, S., Liitkebohmert, W.: Stable reduction and uniformization of abelian var-
ieties I. Math. Ann. 270, 349 379 (1985)
[D] Drinfeld, V.G.: Coverings of p-adic symmetric domains. Funct. Anal. Appl. 10,
107 115 (1976)
330 W. Lfitkebohmert

[DM] Deligne, P., Mumford, D.: The irreducibility of the space of curves of given genus.
Publ. Math., Inst. Hautes Etud. Sci. 36, 75-110 (1969)
[EGA] Grothendieck, A.: E16ments de G6om&rie Alg6briques. Publ. Math.
4, 8, 11, 17, 20, 24, 28, 32 (1961-1967)
[FG] Frisch, J., Guenot, J.: Prolongement de faisceaux analytiques coherents. Invent.
Math. 7, 321-343 (1969)
[GRI] Grauert, H., Remmert, R.: Komplexe R/iume. Math. Ann. 136, 245-318 (1958)
[GR2] Grauert, H., Remmert, R.: Uber die Methode der diskret bewerteten Ringe in der
nichtarchimedischen Analysis. Invent. Math. 2, 87 133 (1966)
[HI Hironaka, H.: Resolution of singularities of an algebraic variety over a field of
characteristic zero. Ann. Math. 79, I: 109 203; II: 205-326 (1964)
[KI] Kiehl, R.: Die De Rham Kohomologie algebraischer Mannigfaltigkeiten fiber einem
bewerteten K6rper. Publ. Math. 33, 5-20 (1967)
[K2] Kiehl, R.: Der Endlichkeitssatz ffir eigentliche Abbildungen in der nichtarchimedis-
chen Funktionentheorie, Invent. Math. 2, 191-214 (1967)
[L] Lipman, J.: Desingularization of two-dimensional schemes. Ann. Math. 107,
151-207 (1978)
[Lul] Liitkebohmert, W.: Vektorraumbfindel fiber nichtarchimedischen holomorphen
Rfiumen. Math. Z. 152, 127-143 (1977)
[Lu2] Lfitkebohmert, W.: Zur Theorie der Fortsetzung rigid-analytischer Strukturen.
Habilitationsschrift M/inster (1977)
IS] Serre, J.-P.: Local Fields. Berlin Heidelberg New York: Springer 1979
[SGAI] Grothendieck, A.: Rev6tements &ale et groupe fondamental. (Lect. Notes Math.,
vol. 224) Berlin Heidelberg New York: Springer 1970
[Siu] Siu, Y.-T.: Techniques of extension of analytic objects. (Lect. Notes Math., vol. 8)
New York: Dekker 1974
IT] Tate, J.: Rigid analytic spaces. Invent. Math. 12, 257-289 (1971)