Annals of Mathematics

Higher Explicit Reciprocity Laws

Author(s): A. Wiles
Source: The Annals of Mathematics, Second Series, Vol. 107, No. 2 (Mar., 1978), pp. 235-254
Published by: Annals of Mathematics
Stable URL: http://www.jstor.org/stable/1971143
Accessed: 22/10/2010 08:07

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/action/showPublisher?publisherCode=annals.

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact [email protected].

Annals of Mathematics is collaborating with JSTOR to digitize, preserve and extend access to The Annals of
Mathematics.

http://www.jstor.org
Annals of Mathematics,107 (1978),235-254

laws
Higherexplicitreciprocity
By A. WILES

Introduction

Various explicit reciprocitylaws have been given throughout the

historyof class fieldtheory. Perhaps the most strikingof these is due to
Artin and Hasse [1]. Let p be an odd prime number,and Qp the fieldof
p-adic numbers. For each integer n > 0, let C,,denote a primitive p"+'-th root
of unity,and put ,n= QP(Cn). Let (a, h denote the Hilbert normresidue
symbolin $Di relativeto the p`+'-th rootsof unity. (We use the definition of
the symbolas may be found in [10].) Write 7Cn = 1 - C?. Then Artinand
Hasse's explicitreciprocitylaw asserts that,foreverya -1 modwn, we have
1) (~,a=) T,(log a)/pn+l

(2) (a, ic "n) = log xU )/pfl

where T? denotes the trace map from JD.to Qpand log a means -,S' 2/2+?13/3...
wherea = 1 + fi. Motivatedby some problemsin the theoryof cyclotomic
fields,Iwasawa [71 proved an importantgeneralizationof these formulae.
An attemptto produceanalogues of theseresultsforcertainfieldsof division
points of elliptic curves was the startingpoint for [5]. Let (n denote the
elementsof (n whose normto Qpis a power of p. Take 9 to be any element
of (D'. Let m be any integer >2n + 1, and fix,3' in (D' whosenormto J(Dis
equal to S. Let b be the orderoff' with respect to 7cm.Then we can write
r5)'(non-uniquely)as a p-adic expansionin the powers of 7r,,with coefficients
in the ring ZP of p-adic integers in Q,, where the firstnon-zeroterm is a
denotetheelementof (D obtainedby formal-
Zr-multipleof Orb . Let di8'/d7Cm
lydifferentiating such an expansionwithrespectto Cm. Iwasawa established
the explicitformula
( 3) (?t fR~n
na,fl),- Z

where
[ac,8]n-
,,
\+
CM d,' log a
L+,TM(m
/3' d~ I
$01.00
0003-476X/78/0107-0002
(? 1978 by PrincetonUniversityPress
For copyinginformation,see inside back cover.
236 A. WILES

for all a in D,, with a _ 1 mod iC..

Lubin and Tate [9] have shownhow to constructanalogues of the fields
withany finiteextensionK of Q, as base field. More precisely,fix such
JD,,
a finiteextensionK of Qp,and fix a local parameter iz in the ring OK of
integersof K. Let G be any Lubin-Tateformalgroupassociated with r and
OK (cf. [9]). Then we obtain the analogue of OD,
by adjoiningto K the points
of orderdividingw'1;on G. For simplicity,we again denote this fieldby
Dn. Let ,nbe the maximal ideal of the ring of integers of Bins
and <>" the
Then, as is explained in detail in Section 2, we
multiplicativegroup of 1Dyf.
obtain a pairing
( 4) ( x (Dx~n G,,n+,;
)0n:G(P,,)
here G(Q") is the group of points of G in p,,nand G,,.+,is the subgroup of
pointsof orderdividingzl%+. The aim of the presentpaper is to prove an-
alogues for (4) of the explicit reciprocitylaws described above. For any
a e OK, let [a], be the correspondingendomorphismof the formal group G.
Let X, denote the logarithmmap of G. For each n ? 0 fixa generatorve,of
G="+?as an OK-module such that for any m > n, [vmn](V) = vn. Let JDn
denotethe elementsof Di whose norm to K is a power of iz. Take f to be
any elementof Ji,and let m be any integer >2n + 1. Fix ,6' in ?D'whose
normto (D. is equal to ,8 (such a ,6' exists, cf. Lemma18), and definedfl'/dvm
as previouslyexcept that now vmreplaces7zmand thecoefficients of a suitable
expansionlie in (K instead of Z,. (For the precise definitionsee Section 5.)
Finally, let Tmdenote the trace map from(D.nto K. Then our main result
is the following.
THEOREM 1. For any a e G(Qn),f e (D'

(it ) Tm(vm) dvm G

In principlethe proofclosely followsthat of Iwasawa in [7], but it also

relies heavily on the theoryof Lubin and Tate. I would like to offermy
sincerethanksto J. Coates for his constanthelp and encouragementwhile I
was workingon this paper, and to N. Katz and S. Lang for manyhelpful
suggestions.
It was remarkedon page 247 of [5] that the original motivationfor
studyingthe explicitreciprocitylaws was in order to prove an analogue for
the ellipticunits of a theoremof Iwasawa which gave a relationbetween
the cyclotomicunits and the p-adicL-functions. The proofof this analogue,
which is joint work of the author with Coates, will be appearing in the
Australian Journal of Mathematics. Although the reciprocitylaw is not
 HIGHER RECIPROCITY LAWS 237

used explicitlyin theproof,neverthelessthe methodof proofwas suggested

by the result in this paper. See also the forthcoming
work of Coleman in
this connection.
1. Preliminaries on Lubin-Tate groups
For the basic facts about Lubin-Tate formalgroups, see [9] or Serre's
article in [3]. Let K be a fixedfiniteextensionof Q. If L is any extension
of K we let C. denote the ring of integersof L and PL the maximal ideal of
OL. We shall always be consideringone parameter commutativeformal
group laws whichare definedover OK. Thus if G is such a group law, G is
given by an elementof OK[[X, Y]]. We write G(PL)for the set p.Lendowed
with the group law of G. Additionand subtractionvia G will be denotedby
[+G] and [-a] respectively.
In this paper we only consider Lubin-Tate groups. Recall that these
arise in thefollowingmanner. Let z be any local parameterin K, and write
XTfor the set ofall powerseriesf(X) _X mod(X),whereq is thecardinality
of the residue fieldof K. Lubin and Tate showed that to each fe T,, there
is a unique formalgroup law Gf with coefficients in OK for whichf is an
endomorphism (i.e., f(Gf(X, Y)) = Gf(f(X), f( Y))). Further,foreach a e OR
thereexists a unique [a]f e OK[[X]I such that [a]f is an endomorphism of Gf
and [a]f_ aX moddegree 2. The map a v-H[a]f thengives an isomorphismof
O with End(Gf) under which [z]f = f. We may therefore consider G(4L) as
an OK-module forany extensionL of K. If f and g are bothmembersof WF,,
thenthe correspondinggroup laws are isomorphicover OK* In makingcom-
putationsit will be convenientto work with the special Lubin-Tategroup
correspondingto the formal power series f(X) = ZX + Xq. We always
write e for this formalgroup.
We suppose for the rest of this paper that X has been fixed,and let G
be any Lubin-Tate group whichcorrespondsto an elementof T,. The next
theoremis a summaryof some of the main results of Lubin-Tate theory.
Let K be a fixedalgebraic closureof K. For each n > 0, let Gzn
be thekernel
of ["]G on G(pk) and put
G0.= Un|nGorn
THEOREM2 (Lubin-Tate).
(i) G0.is isomorphicto KIOK as an OK-module.
(ii) For each n >0, OJn = K(Grn+?)is a totallyramifiedabelian extension
of K of degreeql'(q - 1), from which X is a norm.
(iii) If u is a unit in Ox, thenthe Artin symbolau = (u, OQIK) acts on
via [u-']G.
Go,,
238 A. WILES

One special example of a Lubin-Tate group is the formalmultiplicative

group G., given by Gm(X, Y) = X + Y + XY. This corresponds to the case
K= Qp, 7r = p, [p](X) = (1 + X)P -1. The elementsofG,.+ are, ofcourse,
In-1} where t,, runs over the pn+1-th roots of unity.
Let %, be the logarithmmap of the formal group G. Thus \a is the
unique isomorphism, over K, ofG withtheformaladditivegroupG,, normal-
ized so that XG(O) = 1. For a discussion of the main propertiesof XG, see
Frohlich[6]. For the convenienceof the reader we include here some well
known lemmas on the logarithmof the formal group. Note that X\ is
ORK-linear,because if [a]G is an endomorphismof G, thenX, o [a]Gs o X(z) is an
endomorphism of Ga, and so must be equal to az. (The onlyendomorphisms
of Ga are the multiplicationsby a constant.) It is well knownthat XV(z) is
an integralpower series. However we need a strongerversion of this for
Lubin-Tategroups. We begin with a preliminarylemma. Let ord2 be the
order valuation on K normalizedso that ordQ(7r)= 1. Let K[[z1, ..., znj]
denotethe power series ringin n variables and let A, be the ideal containing
all power series of weight > t. Thus A, is the ideal generated by
{zil ... in + *
Zn: i+ = t}. Then we have a naturally definedlinear topology
on K[[z1, Be.,Zn]] as an 0,K[[zI, *, zj]I-modulewherebythe neighborhoods
of zero are given by {w(OK[[z1, **, zj] + A,}%1^. Thus a sequence of power
seriesfi convergesto f in this topologyif for every fixedk1,***k, > 0, the
coefficientof zk' ...**zk in fi convergesp-adicallyto that of f.
LEMMA 3. For a Lubin-Tate formal group G, we have

XG(Z) = limn0. - [7n ]G(Z)

Proof. Put hn(Z) = [I]G(Z)/Wr. Let k be a fixedinteger> 0. Then to

of zk in
show the existence of the limit we must show that the coefficient
hm(z) - hn(z) tends to zero as m, n tend to infinity. So suppose that m > n,
and put g.(z) = [WnIG(Z). Plainly

(6) hm(z) - hn(z) = ([wm n]G(g.(z)) - Wm-ngn(Z))

Sincegn(z) =i~j2 ?1
a(n)17'zi, where a~!) e and aln)= 1, it is easy to see
O,,
thattheright hand side of(6) is in the(9X-module spannedby{tzj: i j> n+11.
So as n -p oo the coefficient of zk tends to zero, and hence the limit exists.
To prove that XG(z) = f(z), where

f(z) = liMn,, Iw l(z)

7Cn
 HIGHER RECIPROCITY LAWS 239

it is enoughto showthatf(zl[+G~zS) =
fzzj) + f(z2). (Here[+G] denotesad-
ditionvia theformalgroupG.) But if z1[+g]z2 = bjzlzi, then

g1(z1[+ G]z2)= .4;g.(z) +? gt,(Z2)+X ,i+j2 bj;(g-(zJ))'(g.(z2))'

of z{o?zo in thelast termof thisexpression

But forfixedio,j0, thecoefficient
tends to zero as n a-. oo. So the result followson taking the limitof both
sides as n _ Oo.
LEMMA4. For a Lubin-Tategroup G, ag(Z) is in the0K[[z]I-module
which is spanned by {z' '/zri}=O.
Proof. We needonlyprovethe resultforthe specialformalgroup X,
becauseifMa is an isomorphism fromG toe suchthatqSg(z) z moddegree2,
then aG = %B of. Let M be the0,S[[z]]-module
whichis spannedby {z$'1r'}f=o0.
Thenonesees thatM is closedin the inducedtopologyand henceit is suf-
ficientby Lemma 3 to show that [ir*]f(z)/7r'lies in M for each n. This is
equivalent to provingthat [7r"]e(z) lies in theSK-modulewhichis spannedby
{17riZt:i 2 O t > O0i + 0q(log(t)> )
where0(x) denotestheintegralpartof x. But thisis nowprovedby induc-
tionsince
[1+]()= [iro](Z) ? ([7rn](Z))f
We remarkthatLemma4 mayalso be deducedfromthefactthatoneof the
formalgrouplaws associatedto 7r and OK has the logarithm
X+7r-'XI+
7r-Xg2+ (see forexample [2]).
LEMMA5. Let G be a Lubin-Tate group and eG(z) the inverseof XG(Z).
Then eG(z)convergeson D = {z e pk:ord(z) > 1/(q- 1)} and eG(z)and XG(Z)
induce inverse isomorphismsbetweenGJ(D) and G(D).
Proof. Thelemmaand theproofare a straightforward generalization
of a resultof Serre(cf. [6],p. 109). Let y be any elementof D. We define
xv(z)to be XG(YZ)IY. ThenX1(z)is seen to have integralcoefficients
by the
preceding lemma. So we mayfindan integralpowerseriesE,(z) suchthat
E,(x,(z)) = z. Replacing z by z/y we obtain that E,(XG(z)ly)= z/y, and
hencethat E,(z) = eG(yz)IY. Thus eG(yz)Iy has integralcoefficients
and so
convergesforord,,(z)>0. As thisholdsforanyy e D, eG(z)convergesforall
z e D. It followsalso that if y e D theneG(Y) e D, and hencethateG(z) and
'G(Z) induceinverseisomorphisms betweenG6(D) and G(D). For a Lubin-
Tate group G we may deduce the followingcorollaries,where S,, denotes
the ring of integersof (,,.
240 A. WILES

COROLLARY 1. G(wr+2Cn)
C [Cn+']GG(P,).

COROLLARY2. wr+1n C_ Zr X.(pn) for all r > 0.

2. Norm residue symbol

As in the previous section,we suppose that we have fixeda local para-
meter7 for 0K, and we take G to be any Lubin-Tate group associated with
w. We now study an analogue of the Hilbert norm residue symbol for the
fieldsD,, K(G ?+1). The definition of this symbolhas previouslybeen pro-
posed by Fr6hlich[6].
Fix an integern ?0. Let pnbe the maximalideal of the ringof integers
of (D.. The generalizednormresidue symbolis a pairing
(7) ( , G(Pn)X (D >G n+1y
whichis definedas follows. If 8 e (Dx, let up be the element of the Galois
group over On of the maximalabelian extensionof (D. whichis attached to
,8 by local class fieldtheory. If a e G(p), choose 7 in the maximal ideal of
the ring of integersof the algebraic closure of K such that [n+l']G(Y) = a.
(Such a 'yexists because G(pR)is divisible as an OK-module;see [6], p. 107.)
Then we define(a, ),G= p-G]. It is obvious that thisdefinitionis inde-
pendentof the choice of a.
Let (D, = U,,, (Dny and let x: G((DO/K) -~ be the charactergiving the
action of the Galois group of (D,,/Kon Go. = Un0 Grn+I,i.e., K is definedby
ur = [K(z)](uz)for all u e G,. Note that K depends onlyon the isomorphism
class of the Lubin-Tate group G. In analogy with the Hilbertnormresidue
symbolwe have the relation
(8) (ar, 37)cn= [K(z)](ay,8)6 for all z E G($Dco/K)
It is a consequence of the formulaa, r-rz'-; see for example Theorem
11.5 of Tate's article in [3].
Suppose next that G, and G2are isomorphicLubin-Tategroups,and take
9: G1?",' G2to be an isomorphismdefinedover by. Thenthe symbolsfor G,
and G2are related by
(9) (a, 3)n2 -
0((0'(a) f)G')

Recall that e is the formal group correspondingto the power series

wX Xq Gr.

LEMMA6. For any Lubin-Tate group G, we have

(i) (a,cf)nis bilinear, and OK-linear in a;
(ii) (a, i),,= 0 if and onlyif fiis a normfrom$(Dr), where[i1;'J(Y)=a;
(iii) [m-nI(a,/!3)m = (a, Nmn(fl)), if a e G(PQ),9 e IDxe.
 HIGHER RECIPROCITY LAWS 241

For thespecial Lubin-Tate group $, we also have

(iv) (a, a),, = 0 for all non-zeroa e $(p).
Proof. The onlyassertionwhichdoes not followformallyfromthe de-
finitionis (iv). Let f,,(X) = [7r'1n'G(X). Then, if " is any root of f,,(X) = a,
the extension P,,(7)/P,, is obviouslyindependentof the choiceof -r. Thus, if
we factor f.(X) - a into irreducible polynomials over D,,, say
fn(X) - a = II,, f,6j(X), then,for each j, we mustobtain P,,r()by adjoin-
ing a single root of f,,,j(X) to D,,. Therefore,if c,,,jdenotes the constant
term of f,,,j(X), then -cad; is a norm from Pn,,(Y) for each j. Hence
a= - IIjI ,'J (-cn,j) is also a norm from P,,(7),and so (ii) implies (iv). (Note
that -1 is a normas p is odd.)
One explicitreciprocitylaw is immediatefromTheorem2. Let vnbe a
generatorof G,"+, as an OK-module.Let N,, and T, denote the normand
trace fromOnto K. Then it followseasily fromassertion(iii) of Theorem2
that

(10) L-)G
(V"J fl) = (Ni - 1)j (v")

foreach unit 8 =_1 modp,,in OneThis formulamakessense because by local

class fieldtheoryN,,1 - 1 _=0modi"'. If K/Q, is unramified,this law
may be rewritten

(ll) ~~~(VI,8)n=[G~~ Tn(log/!)](Vn)

because N,j-1 - 1 0_ modpf+iand hence

NnS-l - 1 _ log(Nj3-) - T,,(log b) mod p2n+2.

The proofof the general reciprocitylaw will eventuallybe reduced to this

case. We note that in the special case X = p we maytake G to be the formal
multiplicativegroupand (11) is thenequivalentto law (1) in the introduction.

3. The map f,,,

In order to go further,it is importantto introducea map e,,,first
studied in the cyclotomiccase by Iwasawa [8]. Let XG: G --+G. be the
logarithmof G. Then XG converges for each w e G(pJ)and is OK-linearon
G(pn)(cf. ? 1). Hence 'JG(A) is an open OK-submodule of the locally compact
group D+. We recall that D,, is self-dual as a vector space over K with
respectto the trace. This induces a non-degenerateK-linearpairing
PD,x On, - K/OK
242 A. WILES

given by (x, y) - T.(xy) modOK. Let A be any closed OK-modulein Dn.

Then we denoteby A' the annihilatorof A in oD, relative to this pairing.
We may thenidentifyOn/A'with Home,(A, K/IOK).For let p: A -- KI@K be
an OK-homomorphism. Then p can be extendedto a K-linear homomorphism
K since as O-modules A is freeand K is infective. Thereforeforsome
ye on we have
p(x) = Tn(xy) for all x E On S
and we may associate to p the coset of y in (D/A'. We define
(12) 1% = XG(P.)

and so we may identifyoni/n KICK)- We observe that

with HomK(?XJG(PJl),
I, depends only on the isomorphism class of G. Let Nmnand Tmn denote
respectivelythe normand trace from(m to (D,. We defineD* by
(13) (D* = N2,+ln(Dx -
We fixonce and for all a generator us, of $~?~ as an OK-module. For any
Lubin-Tategroup G which correspondsto an element of T,, let q5,be the
unique isomorphismfrom G to e which is definedover OK and such that
0G(Z) z mod degree 2. Then 051(un)is a generatorof G which we will
henceforthdenote by v". The followingcorrespondsto Proposition14 of
[8].
PROPOSITION7. Let v. be thefixedgenerator of G,,+, as an OK-Module
which is describedabove. Then there is a unique map *n $(D*N7t/
such that
(14) (a, l~ =[G
for all a e G(p") and 8 e (*. This map is a group homomorphismand
ep"(flr) = K(z}C)+"(fl)r for all z e G(Q1/K). It depends only on the isomorphism
class of G.
Proof. Let i: Grtl -p K/OK be the unique OK-linearhomomorphism such
thati(v^) = 17-(n+')
modOK. Thenby composingthepairing(7) withthemap i
we obtain a pairing
G(Q,) x On - K/OK.

The annihilatorof G in this pairing is (D* (as (vn,fl)~n

is zero if and only
if fi is a normfrom P"(v2,+l) = D2,+1),and so thereis an inducedpairing
G(Qp)/G,,+, x (D - K/OK

The kernelof the logarithmmap X, on G(p%) is precisely G-,+,,so we may

considerthis as a pairing
 HIGHER RECIPROCITY LAWS 243

%G(Pn) x on - K/OK.,
fromwhichwe obtain a homomorphism

In HomK(\G(%), K/OK)
given by f v {X%(a) i((a, f)l)}. The pairing definedat the beginningof
thissection(cf. (12)) permitsus to identify
Hom0K(XG(n), K/OK)with oD/Yn-
Using this identificationwe thendefine
*,(,R) =-- z"+1 'lS) mod n

for fie on*. That "j) is in Xn and that *, satisfiesequation (14) follows
immediatelyfrom the construction. Clearly 'n(fl) is uniquely definedby
equation (14). In particular, this implies that n,,must be a group homo-
morphism.The finalassertionsare formalconsequencesof (8) and (9). This
completesthe proof.
4. Explicit laws
We have seen that the map Ane,of Proposition7 dependsonlyon the iso-
morphismclass of the Lubin-Tategroup G. Thus withoutloss of generality,
we can henceforthtake G to be the special Lubin-Tate group $ associated
with the power series 7X + Xq. We thereforedrop the index $ fromour
notationwhen thereis no risk of confusion.
Our strategyfor provingthe explicitformulafor *,n given in Theorem
1 is the following. We firstestablish two explicitformulaefor(a, fi), when
a is subject to the hypothesisthat ord,(a) is very large (see Lemmas 8 and
9). We thennotethat thegeneralcase can always be reducedto this. Indeed,
if m > n and fie V, we can always choose fi' in PD'1such that Nmn(fl') = S.
By (iii) of Lemma 6, we thenhave
(a, f)n = ([TmC ](a), /3)m-

But, by choosing m sufficientlylarge, we can always guarantee that

ord,([Vrm-1](a))is large, and so we can express any symbolin termsof one
in whichthe argumenton the left has large valuation. This techniquewas
firstused by Iwasawa [7] in the cyclotomiccase. Wherewe differfromhis
developmentis that, whereas Iwasawa used law (2) of Artinand Hasse to
prove the two weak reciprocitylaws, we are able to prove these laws by
reducingthemto the result of Lubin-Tate (10). In particular this permits
us later to deduce the analogue of law (2) of Artin-Hassefromour formula
for And
Let qS(x)denotethe integralpart of x, and let e be the ramification
index
of K over Q,. Recall that us,is a generatorof $~, as an OK-module.
244 A. WILES

LEMMA 8. Suppose that O(n/2) + 1 0 mode. Then, for all a e(p)

such that ord,(a) > O5(n/2)
+ 1 + e, we have

(a, u-), =, T ( U(.))n )] (u)-

Proof. We divide the proofinto fourparts:
Step 1. Put -t= a + un. The conditionon a has been chosen to enable
us to prove the identity

(15) (7L-Ing,1 + 1) 0-

Let (zi) denote the ideal wi~O and write f(n) for O(n/2)+ 1 + e. By the
bilinearityof the pairing, to establish (15) it is sufficientto show that
L-]-u e [r]c4n4), and 1 + a/us,; e ($Dx)Pd, for some r and d satisfying
r r+ ed > n + 1. If we set X[ -] Y = f(X, Y), thenusingtheTaylorexpansion
of f(X, uq,) around X== ,
a U.) mod(Gzf
-[-]U,= ax (U

and hence we may take r = f(n) - 1 by the second corollary to Lemma 5.

It also followseasily froma considerationof the range of convergenceof
e l/p)logx that we may take d = (f(n)/e) - 1. This then proves (15).
Step 2. We now wish to use (15) to prove a formulafor (7, u.),,. The
formula of Lubin-Tate, (10), gives an explicit value for (u., 1 + a/us),
namelythat
1+ 1) = L7n+'
(N" (1 + C -1)il (U") .

We assert that N,(1 + a/us)-l - 1 - T,(a/u) modW2('rl) and so we may

rewritethis as

(16) (a, 1 + = L-+1T"(-)](u,) .

To prove the assertionlet N.(1 + a/u,)-l 1 + z withz _ 0 modzw+'. Then

z log(l + z)- T.(log(1 + c mod +

and the result followson expandingthe logarithm. Recalling property(iv)

of Lemma 6 we have that

0 = (0 Y)n = (7, UJ)J+]Q', 1 +a

UC n

Combiningthis with (15) and (16) gives-the equality

 HIGHER RECIPROCITY LAWS 245

(17) (, Us), = n Tn a_ (u.)

Step 3. We proceednow to obtainanotherexpressionfor(/, uJ),,) using

the definingpropertyof the map (cf. Proposition7). Recall that V(z) is
An

an integralpower series, and so by Taylor's theorem,

+ a%'(u") mod2f
%(U,(u) (T)

for all a e cf
(wf). However ,(U.) = 0 and irCf( _ Z'l(pn) by Corollary 2 to
Lemma 5, so by Proposition7 we have
(18) (0, UA)= [
Step 4. Comparingthis expression for (-, U.)t, with (17) we obtain the
identity
(19) [Tn(aV(ufl)'+nl(uf))J(ufl)
= To (U .)

which is valid for all a e 7rf(). As V(u.) is a unit, we may replace a by

a/V'(u.) in (19), and so obtain

(20) [ LCt?(a Un))](Wn) = 17n Tn tun a X)(U

Finally,as %(a) a modZ the left hand side of the above expressionis
seen to be (a, U,),. This completesthe proofof the lemma.
LEMMA9. For all un e $(pj) such thatord,(u') > 0(n/2)+ 2, and for all
j > 1, we have
1)r = [-U](,r, (
(un, sun - un)n)
where A denotes summation in the formal group, and s is any (q - 1)-th
rootof unity.
Proof. Let us now write g(n) for 0(n/2)+ 1. Put w = suu(j - 1) and
v= 4[+]w. Since ssu' e (zg(fl)+t) it is clear that v[-]s8+lu'+i 0 mod(wz"+3).
So as ~(+3) C [wz"+]$(p) by Corollary 1 to Lemma 5, we have v =
t t ][ +11(a) for some a e $(p,). Recalling (iv) of Lemma 6, we deduce
that
(e suj - 1)n = (e e W)n = (V, w)n = (V, suj - 1)n[+](v, U)n.

(Note that as (a, ), = 0 for all a we may write (v, Un), for (v, sun)n)
As v = s8+1ui+i[+w[:z+](a),it followsthat

(Sovnsugreuri)n =
(obta n+i sun - 1)[+j(S+1Ui+j Ui
we
Solving recursively, obtain
246 A. WILES

whereE denotessummationon the formalgroup. The sum on the rightis

finitebecause (sru', ua") 0 for all sufficiently large t and any s (since then
e [7cf]$(pJ)). But
G

(sruiri, r--+(T-l)) = [i + (r. - 1)j](6rUiri, U")n = [_j](6rUi+rj, u)n

because [i + rj](nTut?ri, u'?) (es24+ri, nS)I+r) = 0. The lemmanow follows.

5. The map An
In this sectionwe definea map a. whichgeneralizes a similarmap used
by Iwasawa in [7]. This is an essential part of our descriptionof Gnu As
previouslylet G be any Lubin-Tate group associated to i1 and K, and let v,
be a primitivewn+'-division point on G. We continuethe conventionin the
choice of vnwhich was made in Section 3. Let 9 e $D', and select a power
series ,8(z) =b. r(z), where b = ordj(,8) and r(z) e DK[[z]] satisfiesr(O) ZL 0,
such that fl(v")= F. (Here ordndenotesthe ordervaluation in On for which
ord,(v) -1.) Then we define

\4X3i V ) (3 zv
As D(z)is not uniquelydeterminedby,(, 8i(fl)is manyvalued. However, it
is easily seen that this set of values depends only on the isomorphismclass
of G. For simplicitywe thereforerestrictattention to the special formal
group e and continueto use the notationof Section 4.
LEMMA 10. (i) 8J,(f)e p;-land is well-definedmod@, where Z"n is the
differentof On over K.
(ii) an inducesa homomorphism: D<x _ paimodZ,.
(iii) &n(fl?) = K(a)&(l)~ for all ,le x, a e G($D../K).
$
(iv) For m > n, and for any unit rlin Ox,
&m(27) - Wm-n&^,,() modZm .

Proof. It is clear that e 1. Suppose that ord"(,f)

p;') b, and that
,8,(z)and 2(Z) satisfy l1(u,)= 92(u) = 9,andZ-b1e1(Z), Z-b,32(Z) e O4[Z]]. Then
by the Weierstrass preparation theorem {fl1(z) - 82(z)}z-b is divisible by gj(z),
where gj(z) is the minimumpolynomialfor u". (Recall that u, is the chosen
generatorof d7n+l) Say l1(z)- f2(Z) =b gj(z)t(z)wheret(z)e .K[[z]]. Dif-
ferentiatingand evaluating at z = us gives,
f31(un)- 823(u.) - Ung (un)t(uu) .

Part (i) now followson recallingthat (gn(Un))= (,)n and that u4/flis a unit.
Part (ii) followsimmediatelyfrom(i). For (iii) it is sufficient
to prove
 HIGHER RECIPROCITY LAWS 247

theresult when,8is an integer. In thiscase if ,3(z) is a suitable powerseries

for e, thenfl([K(a)](z))is a suitable power series for fla. So
a (/3a)1 fl'(Un)[K(a)]'(Us~)

Recall that A([sa)J(z)) = tr(a)x(z),which,on differentiating

and evaluating
at z = u,,, gives
(21) = K(a) (u")
r'(un)[ff(a)J'(u)
On substitutingthis in the above formula for &"(fla)we obtain (iii). To
prove part (iv) we proceedsimilarly. If g is a unit and 72f(z)is a suitable
powerseriesforg such that (Uf) Us = thenVim(Z) ,?([Wm-n](Z)) satisfies
".m(.u'm)=. This gives

am(7) 1 rn(Un)[lrn1 (Um)

XV(Um) 27

The result thenfollowsif we use in place of (21) the relation

(22) (Un)[7zm]P(um) =

6.
In thissectionwe introducea pairing<, > whichis based on the map d.,
and derive some of its basic properties. We begin with several elementary
lemmas. Recall that Z is the differentof $D/K, and so Z. = n+?tn-q"-

LEMMA11. Let (z() =z"'O. Then for all a e t(pj we have

T,((1)X(a))=- 0 modw"'
Proof. By Lemma4 it is sufficientto provethat T"((wz")aq'/i) 0 modzn+
forall i > 0. For i n - 1 this is immediatebecause the relative different
of (D,,/(Dis (an). For i > n aq e 40(7ri-f)O,
and the result follows because
q(]POO9n) C-X7r'K-

LEMMA 12. (i) +<Tm((7Cmj)(pn)) 0 modz"+' if m > 2n + 1.

(ii) X,+lTn(ZnA,(p _ 0 modzn+' if m >n.

Proof. Part (i) followsfromthe previous lemma. To prove part (ii) we

recall that ,(p26n) - 02@_ and Zm = wm7+'O
- So
=qn)) - Tm(,+1p0(9n) C2m+2
248 A. WILES

LEMMA13. (i)

Tm,n( for all m ? n;

(ii) For m = n + 1, 8 e DX, we have

&n(Nmsn(f3))_ -' TmsnQ8m(f3)) mod (acn)
Proof. As 1, ua, *.*, u-1 forman integral basis for over ,n,-and
D,,
as u. has minimalpolynomialXq + 7rX- u" over Jkn-l, by Euler's formulae

Let w. = 1 + quIn1/z.Then (22) gave the relation

(E)=wnX'(un1)
-

Hence

( zAs(b) 7n+lx,
u~n(Ut_ ) )8U,_ witn( 1a

Since un1-
ru, + us, we have the equality

oz 1- 1)
un_1 =_ 7+ q
wnun q w"
By the remarksat the beginningof the proof,the trace of this to ($__-, is 7:,
and so part (i) follows.
Both sides of the expressionin part (ii) are well definedmodZ. and so
in particularmod(z"). It is sufficient to prove (ii) when 8 = ur and when
,8 is a unit. For i8 = ur it followsfrompart (i). For f8a unit, we have by
parts (iii) and (iv) of Lemma 10,
7C8n(Nm,n(f)) = 0 dm(/3) 0 Kc(Jf)8m(/)? mod (~n+1)

where the sum is taken over all a e G($Dm/$n.). As i(a) _ 1 mod (zlt+1) for all
a e G(QDm/$n)the result follows.
Now forace G(pn), fie Ox, n > 0, we define

<KaRfX= 7rl ' L "() ))

Althoughthe value of <a, 8>, dependson the choiceof the value of ajs), it
is in certaincases well-defined
moduloa high power of z.
LEMMA14. Let m > n, f8e Ox. In each of thefollowing cases the value
of <a, f>m is well-defined mod (rz4+1):
(i) m > 2n + 1, aeE (pn),
 HIGHER RECIPROCITY LAWS 249

Proof. As '3m(f) is well determinedmodZm, the lemmafollowsimmedi-

ately fromLemma 12.
We note that it follows easily fromthe definitionthat, when it is well-
defined,the pairing<a, f>lXis bilinearand OK-linearin a.
LEMMA15. Let 1 > m > n, fie Dj,
D a e $(p). Then if y, m satisfy con-
dition (i) of theprecedinglemma,
<a, Nim(f3)>m <(a, >, mod(z+1) .

to provethisfor 1 = m + 1. In that case, we have

Proof. It is sufficient
tbm(NIm(f3))-Z:l 71Tzm(bl(/)) mod(7zr),
by Lemma 13 (ii). Applying(1/zm+l)Tmto both sides after multiplicationby
%(a), we obtain the result by Lemma 12.

7.

We confineour attentionto the formalgroup e as before. For conven-

ience we write (c, f)n = [[a, 4].I(u.), so that [a, f]n is a well determined
elementof +9KCOK.. Obviouslythe propertiesof (a, 38),given in Lemma 6
also hold for [a, 6]%. As was indicatedin Section4 we have explicitformulae
for (a, f)n,for certainvalues of n and providedord,(a) is sufficiently large.
Using these formulae we will show that under the same restrictions,
[a, 6]n = <a, 6>%. We then apply some of the formal properties of the pair-
ing <, >%whichwill give a simpleconnectionbetween these two symbolsof
much greater generality. The followinglemma is the crucial step in the
whole argument.
LEMMA 16. Suppose that 0(n/2)+ 1 _ 0 mode. Then for all a e t(p)
such that ord,(a) > q(n/2)+ 1 + e,
[a, flhn
= <a, >n%-
Proof. The conditionsin the lemmaare the same as those of Lemma 8.
In particularq(x) is the integralpart of x and e is the ramification
index of
K over Qp. First note that <a, %>n is well-definedmod nZ"+1 by Lemma 14 (ii).
As both sides are multiplicativein ,f,it is sufficient
to prove the result for
R = u. and for fi a unit. In the formercase one value of j(uj) is given by
1/% (U,,)UnYsSO

<a, ?1'8n = Tn(( X(a))

Since %(a) =a mod('z"+3),the equality now follows directlyfromLemma 8.
The units of Ox are generatedas a Z,-moduleby the (q - 1)-throotsof unity
250 A. WILES

and {1 - sun: j = 1, *c. co; s in the (q -1)-th roots of unity}. For f8a root
of unityof orderprime to p both sides of the equality are trivially zero.
By the OK-linearityin a of the two pairings it is sufficientto prove the
equality for a in the set {ui: ord4,(u) > O(n/2)+ 1 + e}. So suppose now
that u4 is in this set and fixj an integer ?1. Then by Lemma 9,
(23) [4u, 1 - Sun [ J]r=o [e4?ri, u]
The value of [SrU?ri, run], is given by Lemma 8 as,

(24) [eru4?rJ,uj1,
.T] = (1 i)rj

We may replace SrU+rj in (24) by SrUrX(Ui), and thenon substitutingthis in

(23) we obtain
1
(25) [uk, 1 - u ____
Ti ((U)n) SrUriX

But

anj - 'Sun) - J = _
X'(uj) 1 su4 X'(un)un
and so (25) gives us that [u', 1 -sufl = (<4, 1 - su> as required.
+- 1 _ 0 mode. Thenfor
LEMMA17. Let m ? 4(n + 1) + 2e and 5(m/2)
a e p 8eG xmwe have
D

[a, Nmn(f)]n = <a, fl>m.

Proof. We note that <a, /S>mis well-definedmod(ir"1) by Lemma 14 (i).

The firstconditionon m has been chosen so that
+ 1+ e.
m - 2n > q5(m/2)
For any a e $(pP),we have [wl](a)C . and [n-"](a) E 2m-2n So if we let
a' = [7m-n](a), the above inequality gives
(26) + 1 + e.
ordc(a') > Os(m/2)
By property(iii) of Lemma 6,
(a, Nmn(f8))n = [7m-"](a, f)m= (a', f)m.
But because of (26), Lemma 16 gives an explicit value for (a', S)m In fact
we deduce that
-[<aft 8>,](un) = [7rl-fl<aff
(af Nm,,n(,S))n RXJ]Un).
point on I, <Kaf3>m is an integer,and we
Since (a, Nmsn(fl))nis a wr+'-division
may rewritethis equation as
(a, = />m](Un)
Nm,n(,8))n [<a,
 HIGHER RECIPROCITY LAWS 251

This completesthe proofof the lemma.

We are now in a positionto prove our main theorem. We define
la= fnm.n Nm,n(vx)
An alternativedescriptionof PD'is given in the following.
LEMMA18. (D' is the subgroupof Ox consisting of all elementswhose
norm to K is a power of C.

Proof. Let u"(K) denote the group of unitsof K whichare 1 mod rZA.
Then Lubin-Tatetheoryshows that N%(D4x)= {7z}x u"+'(K) where {z} is the
cyclicgroup generatedby z. On the otherhand, by local class fieldtheory,
an element ,9 of Ox is contained in Nm,,Q((Dx)if and only if N%(OS)is contained
in Nm((Dx)(m > n). It is now clear that ,8 e (D' if and only if NQ(3) belongs
to {r}.
It follows from the above that Nm.,t(QD')= Ad. So for any ,3 e ID' there
exists an element /3' in (D' such that Nm.,(,/') = ,8. Any such /3' will be
denoted by Nln(,8).
THEOREM 19. For any a e ,(,)y, 8 e (D'
[x, 3]. =<a, Nmln(/)>m for any m > 2n + 1.
So for F e ',

*n()=8m'+1* Tmn(8m(Nln(/))) for any m > 2n + 1 .

Proof. By Lemma17 thefirststatementis valid forany m ! 4(n + 1) + 2e

such that 0(m/2)+ 1 = 0 mode. Then by Lemma 15 it also holds for any
m? 2n + 1. The secondpart thenfollowsfromthe uniquenessof the map ' n
satisfying(14).
We note also that Theorem1 follows from this as *n dependsonlyon
the isomorphismclass of G. (See Lemma 7.) The theoremalso shows that
an explicit value for *,&(un)is given by (1/lrm+l)Tm.,(8m(um)).By Lemma 13 (i)
this is equal to (1/z"+')%(u,)u,. We deduce the following:
COROLLARY. For all a e $(pn),

=
(a, un,)n 1+ T" (%, ) ,(a)) (U") .

8.
We shall now derive the secondlaw of Artinand Hasse whichwas given
in theintroduction.(See (2).) We let G. be the formalmultiplicativegroup,
K = Q, and uz = p. Then Cm- 1 is a generator of (Gm)pm+land %Gm(Z) is
252 A. WILES

log(1 -4 z). Hence 3m(Cm- 1) is given by -C.7mW where 7Cm= 1 - rm. One
sees easily by direct computation that TmCnrmW') Cw;'. Hence, in this
case with f= 1, Theorem1 gives

(27) (a, - 1)Gm- __+ TX(_ log(1 + a))] (7 -)

If (, )"' denotesthe Hilbert normresidue symbol,definedas in [101,then we

have the simplerelation
(as, )[ - (a - 1, f)IOm + 1

when a-=1 modjps. So (27) is seen to be equivalent to formula(2) of Artin

and Hasse.
9.
Artinand Hasse proved one furtherlaw whichis stated for example in
L5].This law is equivalent to the identity
(CwS0 - [ T0(XG(aK)8009))](U0)
for all a e G(p%)/3 (DI, when G is the formalmultiplicativegroup Gm. A
proofin the special case of heightone Lubin-Tate groups may be foundin
[4]. We now prove thislaw forLubin-Tategroupsof any height. The proof
is very similarto the proofof the formulaforen. For the followinglemmas
we work onlywith the formalgroup s^
20. For the formcalgroup $, we have [4](w) -w for each
LEMMA
(q - )-1throotof unityC in (K,
Moreoverif d(X, Y) = X+ Y+ aiX i YJ
= i
is theformal group law of $, thenaij 0 unless + j 1 mod(q 1).
-

Proof. Recall that for each n > 0, uo is a generatorof _ and that

>
u._-for all n 1. Let C be a fixed(q 1)-throotof unity. As the
-
minimalpolynomialfor us over K is a polynomialin Xq-1, we can find
an e G(QP/K) such that Crn(un)
= u. Thus thereis an elementa of G(Qc,3/K)
such that a(u) = Cu for all n. By the last part of Theorem2 thereexists
a e OK such that a(u") = [a](u^) for all n > 0. Since aq-1 1, a is a (q - 1)-th
-

root of unity. Let p,,be the maximalideal of (0n- Then as

C [a](u") _ a us modpn ,
-

C=a modw,whence C = a. The finalassertion follows easily because we

have
p -$(X, Y) = d(pX, p Y) for every (q -
1)-throot of unity p .
LEMMA21. Let X(w) = w + E12 aiwi. Then ai = 0 unless
 HIGHER RECIPROCITY LAWS 253

Proof. This followsfromLemma 20 because 4-Xe(X) =%UX))

LEMMA22. (u4, sui-1) = [-ji] ' (ru4+rii, u0)0if i 2, j > 1.
(Here a denotessummationon theformal group,and s is a (q - 1)-throot
of unity)-
Proof. This is based on Lemma 9. Put w= s8u(suo-1) and v= s8u'[ + 1w.
Since u' e p2,it is clear from Lemma 20 that v[-]8+u+lui+j 0 mod to* So as
+ Thr](a)forsomea e d(po). The remainder
2q C [7I]r(po)we have v = c8+lui+j4

of the proofis now just as in Lemma 9.

23. For axe G(p2) and f8e PDxwe have
THEOREM

(28) (a, f)0[G

- T(XG(Z)8O(Q))] (vO)-

In particular *0 = (1/>)83on (o*.

Proof. Using (9) it is clear that we need onlyprove the theoremin the
special case G = d. Since X(u0) = 0, we see that X(pQ)= %(p') = Jp. This
shows that the righthand side of (28) is well-defined as 80(B)is uniquelyde-
terminedas an elementof po-modzp-1. Both sides are bilinearand OK-linear
in a. So we need only check the equality for a = u', i 2 2, and ,8 in some
generatingset of Ox. As noted previously,we may take this generating
set to consistof u0,1 - sui (j > 1), and the (q - 1)-throots of unity.
We know by the corollaryto Theorem19 that 0(u0)= 80(u0)and so (28)
holds for fi = u0. For /8a (q - 1)-throot of unitybothsides of (28) are zero.
It remainsto checkthe case a = u, 98= 1-Suoi for i > 2 and j 2 1. By the
precedinglemma
(29) (ui, 1 - Suo)0 = [-i]E,1 (6Sui+t, u0)0-

The value of (sruoj, u0)0is given by the corollaryto Theorem19,

(30) u0)0= [1To(

(Sruo+ri, x X( ruO+ri))](uo)

in (30) because by Lemma21 we have

We may replace %(,su'+rj)by sru6Xj%(u')
%(u')ou' modJpq. On substitutingthisvalue for (u'+ri,u0)0in (29) we obtain

(u", 1 su6) - [ ' ?rL1 To( 1 ruox(u0))I(u0)

The result now follows on observing,as in Lemma 16, that one value of
E' 1 truri. That '0 = (1/z)80on PD*follows
80( - ui) is given by (- j/X'(uo)uo)
immediatelyfromthe uniquenessof the map '0 satisfying(14).
HARVARDUNIVERSITY,CAMBRIDGE,MASSACHUSETTS
254 A. WILES

REFERENCES

[1] A. ARTINand H. HASSE,Die beiden Ergdnzungsitzezum Reziprozitdtsgesetz der In-ten

Potenzrestein K5rperder In-tenEinheitswurzeln,Abh. Math. Sem. Univ. Hamburg
6 (1928),146-162.
[2] P. CARTIER, Groupesde Lubin-Tategeneralises,Inv. Math. 35 (1976), 273-284.
[3] J. CASSELS and A. FR6HLICH, AlgebraicNumber Theory.Academic Press, 1967.
[4] J. COATESand A. WILES,Explicit reciprocity laws, Soc. Math. France, Asterisque 41-42
(1977),7-17.
[5] , On the conjectureof Birch and Swinnerton-Dyer, Inv. Math. 39 (1977),223-
253.
[6] A. FR6HLICH, Formal Groups, Lecture notes in Mathematics74, Springer,1968.
[7] K. IWASAWA, On explicit formulasfor the norm residue symbol,Math. Soc. Japan 20
(1968),151-164.
[8] , On some modulesin the theoryof cyclotomicfields,J. Math. Soc. Japan 16
(1964),42-82.
[9] J. LUBINand J. TATE,Formal complex multiplicationin local fields,Ann. of Math. 81
(1965),380-387.
[10] J-P. SERRE,Corps Locaux, Hermann,Paris, 1968.
(ReceivedJanuary14, 1977)
(Revised August 7, 1977)