Annals of Mathematics, 141 (1995), 553-572 Ring-theoretic properties of certain Hecke algebras By RICHARD TAYLOR AND ANDREW WILES Introduction The purpose of this article is to provide a key ingredient of [W2] by es- tablishing that certain minimal Hecke algebras considered there are complete intersections. As is recorded in [W2], a method going back to Mazur [M] al- lows one to show that these algebras are Gorenstein, but for the complete intersection property a new approach is required. The methods of this paper are related to those of Chapter 3 of [W2]. The methods of Section 3 of this paper are based on a previous approach of one of us (A.W.). We would like to thank Henri Darmon, Fred Diamond and Gerd Faltings for carefully reading the first version of this article. Gerd Faltings has also sug- gested a simplification of our argument as well as of the argument of Chapter 3 of [W2] and we would like to thank him for allowing us to reproduce these in the appendix to this paper. R. T. would like to thank A. W. for his invitation to collaborate and for sharing his many insights into the questions considered. R. T. would also like to thank Princeton University, Université de Paris 7 and Harvard University for their hospitality during this collaboration. A. W. was supported by an NSF grant. 1. Notation Let p denote an odd prime, let O denote the ring of integers of a finite extension K/Qp, let A denote its maximal ideal and let k = O/d. If L is a perfect field Gz, will denote its absolute Galois group and if the characteristic of L is not p then ¢ : Gz, + ZX will denote the p-adic cyclotomic character. If L is a number field and g a prime of its ring of integers then Gp will denote a decomposition group at g and I, the corresponding inertia group. We shall denote by Froby the arithmetic Frobenius element of Gp/Ip. If Gis a group and M a G-module we let MG and Mg denote respectively the invariants and coinvariants of G on M. If p is a representation of G into the automorphisms of some abelian group we shall let Vp denote the underlying554 RICHARD TAYLOR AND ANDREW WILES G-module. If H is a normal subgroup of G then we shall let p¥ and py denote the representation of G/H on, respectively, Vj! and Vp,n. We shall also fix a continuous representation 3B: Gq — Gla(k) with the following properties. © Dis modular in the sense that it is a mod p representation associated to some modular newform of some weight and level. The restriction of 7 to the group Gal(Q/Q(\/(—1)®-)/2p)) is absolutely irreducible. If c denotes complex conjugation then det 2(c) = The restriction of 7 to the decomposition group at p either has the form dh * 0 de with 1 and ye distinct characters and with 2 unramified; or is induced from a character x of the unramified quadratic extension of Qp whose restriction to the inertia group is the fundamental character of level 2, Ip -» FX. Ifl#p then ~ etree lw ( % :) = 1 * -wrue(} 3) — or p |g, is absolutely irreducible and in the case |;, is absolutely reducible, 1 # —1modp. (This implies that 7 |¢, is either unramified or of type A, B or C as defined in Chapter 1 of [W2]. On the other hand if |g, is of type A, B or C then some twist satisfies the condition above.) In the case that B |g,~ we fix the pair of characters q1, #2. Note vy 0 that in some cases this may involve making a choice. We let @ denote a finite set of primes q with the properties © pis unramified at q,RING-THEORETIC PROPERTIES OF CERTAIN HECKE ALGEBRAS 555 = 1modp, © p(Frob,) has distinct eigenvalues, denoted ay and fy. Much of our notation involves a subscript Q to denote dependence on Q; when- ever Q = 0 we may simply drop it from the notation. For q € Q we shall let Ag denote the Sylow p-subgroup of (Z/qZ)* and let 64 denote a generator. We write Ag for the product of the A, with q € Q, and let ag denote the kernel of the map O[Ag] — O which sends every element of Ag to 1. Let xq denote the character GQ ~ Gal(Q(6q)/Q) and let x9 = Tyeq Xa: We denote by Ng the product of the following quantities: Z/qZ)* -» dq, © the conductor of p; © the primes in Q; © p, if D is not flat (i.e. B does not arise from the action of Gp on the Q,- points of some finite flat group scheme over Zp) or if det [1,4 €. (We remark that if |g, is flat but det 7 |), € then 7 |g, arises from an étale group scheme over Zp. We also note that in [W2] the term flat is not used when the group scheme is ordinary.) We let Ig denote the inverse image under T'9(Ng) > (Z/N@Z)* of the product of the following subgroups: © the Sylow p-subgroup of (Z/MZ)*, where M denotes the conductor of Z; © for each q € Q the unique maximal subgroup of (Z/qZ)* of order prime top. Let T(T'g) denote the Z-subalgebra of the complex endomorphisms of the space of weight 2 cusp forms on Ig generated by the Hecke operators Tj and (I) for | + pNg, by Ug for q € Q and by Uy if p| Ng. Let m denote the ideal of T(Pg) @z O generated by A, by trp(Froby) — T; and det 7(Froby) — I(1) for 1 + pNg, by Uy — aq for q € Q and by Uy — Y2(Frobp) if p | Ng. Note that if Q # 0 this definition only makes sense if O is sufficiently large that k contains the eigenvalues of p(Froby) for all q € Q. It is a deep result following from the work of many mathematicians that m is a proper ideal (see [D]), and thus maximal. We let Tg denote the localisation of T(Tg) ®z O at m. Note that Tg is reduced because the operators T; for 1 { Ng act semi-simply on the space of cusp forms for Ig and the U; for q € Q act semi-simply on all common eigenspaces for the T} for which the corresponding p-adic representation 7 is556 RICHARD TAYLOR AND ANDREW WILES either ramified at q or for which r(Frob,) has distinct eigenvalues. There is a natural map O[Ag] > Ta, which sends € Ag to (y) where y € Z, y = rmodq for all g € Q and y = 1mod No. It follows from the discussion after Theorem 2.1 of [W2] or from the work of Carayol [C2] that there is a continuous representation 3°? : Ga — GL2(Ta); such that if | Nop then om is unramified at 1 and we have trp§°¢(Froby) = Tj and det pi§4(Frob;) = (I). In particular the reduction of p°t modulo the maximal ideal of Tq is . From [C1] we can deduce the following: @ Ifq € Q then p§ |c,= $1 © 42 where ¢} is unramified and $1 (Froby) = Uy, and where $2 |1,= Xq |Iy- © If #p and p|;, is nontrivial but unipotent then pg° |;, is unipotent. If! ¢ QU {p} and either J |,= x 1 or P |g, is absolutely irreducible then p5°¢(Ii) + p(t). © det p3°4 = xqed where ¢ is a character of order prime to p. Moreover if 7 |a, is flat and if det |1,= € then p { Ng so that p3°* |a, is flat (ie. the reduction modulo every ideal of finite index is flat). If B Top i is not flat or if det J |1,# € then p| Na, P la,~ and Up is a unit in Tg. It F) 0 de follows from Theorem 2 of [W1] (or more directly in the case 1 |/,# € from xe 0 x2 and y1(Jp) has order prime to p. In the case that x1 is unramified we know further that x1 = x2 (see Proposition 1.1 of [W2]) and that this character has finite order. It will be convenient to introduce the twist ply = 8°? ® x9/? of 5°4, In particular we see that det p@ is valued in O*. ‘The main theorem of this paper is as follows. Recall that we may write T for Ty. Proposition 12.9 of [G]) that pB°? |a,~ , where x2 is unramified TueoreM 1. The ring T is a complete intersection. ‘We note that if ©! is the ring of integers of a finite extension K’/K then the ring constructed using ©’ in place of O is just Tg @0 O’. Also T is a complete intersection if and only if T @o ©’ is (by, for instance, Corollary 2.8 on page 209 of [K2]). Thus we may and we shall assume that O is sufficiently large that the eigenvalues of every element of p are rational over k and that there is a homomorphism x : T + O. In particular the definition of Tg makes sense for all Q. There is an induced map mg : Tg > T + O. The mapRING-THEORETIC PROPERTIES OF CERTAIN HECKE ALGEBRAS 557 Ta — T takes the operators Tj and (l) to themselves and the operator Ug to the unique root of U? — T,U + q(g) in T above ag. We let gq denote the kernel of 7g and 7g denote the ideal 79(annp,(Q)). Then it is known that 00 > #99/9% > #O/ng with equality if and only if Tg is a complete intersection (see the appendix of [W2] or [L]; we are using the fact that Tg is reduced). 2. Generalisation of a result of de Shalit In this section we shall use the methods of de Shalit (see [de Sh}) to prove the following theorem. THEOREM 2. The ring Tq is a free O[Ag] module of O[Ag]-rank equal to the O-rank of T. By Lemma 3 of [DT] we may choose a prime R with the following prop- erties: © Rt 6Nop; « R#l1modp; © 2(Froba) has distinct eigenvalues ap and Bp; © (1+ R)* det @(Frobr) # R(trp(Frob,))?. Let I'g- be defined in the same way as Tg but with (Z/gZ)* replacing its maximal subgroup of order prime to p in the definition for each q € Q. Let %) =TgNPi(R) and let M_ =I'g_NTi(R). The purpose of introducing the auxiliary prime R is to make these groups act freely on the upper half complex plane. Let T’(I%) denote the Z-subalgebra of the complex endomorphism ring of the space of weight-two modular (not necessarily cusp) forms generated by the operators T; and (l) for | | NR and by U; for || NR. Let my denote the maximal ideal of T’(Ig) ®@z O generated by the following elements: or eT, — trp(Froby) and U(J) — det a(Froby) for 1 | NoRp; © Uy — ag for ge Qorq=R; ¢ Uj — try, (Froby) if || No and | # p; © Up— 2(Frob,) if p | No; © T,— tep;,(Brob,) if p | No.558, RICHARD TAYLOR AND ANDREW WILES Let TQ denote the localisation of T'(TQ) ®z O at mp. Let Y% denote the quotient of the upper half complex plane by Ig and let X% denote its standard compactification. Complex conjugation c acts continuously on these Riemann surfaces. We let H1(¥4,O)* and H1(X4,O)* denote the +1 eigenspaces of con H1(¥4,0) and H'(X%,0). All these definitions go over verbatim, but with Q— replacing Q. Lemma 1. To = TQ and To_ = T. The proof is a standard argument which we will only sketch. First observe that because 7 is irreducible T/y and ‘T/z_ can be defined using the ring gen- erated by the Hecke operators on the spaces of weight-two cusp forms S2(TQ) and $2(Ip_) (rather than spaces of all modular forms). The same arguments as in the proof of Proposition 2.15 of [W2] show that we can drop the Hecke operators Tp if p | Ng and the Hecke operators U; for | # p and || Ng from the definition without changing the Hecke algebra. Next we show that we need only consider the algebras generated in the endomorphisms of $2(I'g)? C S2(I'y) and Sp(P)?*"" $2(Ip_). This follows from the facts below: ¢ As R # 1modp and detj is unramified at R, no component of Ti nor of To can correspond to an eigenform with a nontrivial action of (Z/RZ)*. * As ap/Sp # R*" in k, no component of Ty nor of T/_ can correspond to an eigenform which is special at R (i.e. an eigenform which corresponds to a cuspidal automorphic representation of GL2(A) whose component at R is special). @ As for each prime q € Q, aq/6, # q*" in k, no component of Tiy_ can correspond to an eigenform which is special at q. The ring generated by the Hecke operators T; and (l) for | + pNQR, by Uy for q € QU {R} and by Up if p | Ng on $p(PQ)? is isomorphic to T(CQ)lurl/ (wk — Trur + R(R)). In fact Up acts by the matrix (ty 3) on $2(I'g)?. Similarly the ring generated by the Hecke operators T} and (J) for | { pNQR, by Ug for q € QU {R} and by Up if p | Ng on S2(P)2*2" is isomorphic to T(D)|ug : ¢ € QU {R}]/(u2 — Tyug +4(a) : ¢ € QU {R}). Tensoring with O and localising at the appropriate maximal ideal we get the desired isomorphism. We have to use the fact that uz — Tru + R(R) has two roots in Tg with distinct reductions modulo the maximal ideal and the similar facts over T for uz — Tyug+ q(q) with ¢ € QU {R}.RING-THEORETIC PROPERTIES OF CERTAIN HECKE ALGEBRAS 559 Because 7 is irreducible we see that H1(¥},O)mj, = H'(XQ,O)mj. and that H1(Y}_,O)q,_ = H'(XQ-,O)n,_- By Corollary 1 of Theorem 2.1 of [W2] we see that H1(XQ, Vn, are free rank-one Tg-modules and that H'(X4_,0)4,_ are free rank-one T/p_-modules. Hence it will suffice to prove the following proposition. PROPOSITION 1. H1(¥),O) is a free O[Ag|-module, with O[Ag]-rank equal to the O-rank of H(¥5_,O)~. Because H1(¥4_,K) = H1(¥}, K)4@ we need only show that H1(¥4, 0) is a free O[Ag]-module. Because R > 5, I"j_ acts freely on the upper half complex plane and so may be identified with the fundamental group of Y¥4_. In particular we see that [/p_ is a free group. Similarly [% acts freely on the upper half complex plane and we get identifications H'(¥$,0) = H'(9,0) £ H'(P9_, O[Ag]), the latter arising from Shapiro’s lemma. Under these identifications com- plex conjugation goes over to the involution induced by conjugation by € = -1 0 o1 action of con YZ is induced by the map z ++ —Z of the upper half complex plane to itself.) Because I'¥p_ is a free group, the group of cocycles Z'(I%_,O[Ag]) is a free O[Ag|-module. (If v1,..-,%a are free generators of ¥j_ then we have an isomorphism and trivial action on the coefficients. (This follows because the 2(TQ_,0[Ag]) > OlAgl* Wr (W(n1)s-+-1H(%a))-) On the other hand € acts trivially on Ag and so the coboundaries are contained in Z\(P'y_,O[Ag])*. Thus H"(¥4,0)- ¥ Z(T'_,O0[Ag])~ is a free O[Ag]- module, as desired. Before leaving this section we note the following corollaries of Theorem 2. Corouary 1. If q ¢ Q then Toutq/(6q — 1) “+ Te. Moreover, (6, — UTaqutag) and (1+ 6; +--+ + f°?) Tautqy are annihilators of each other in Tqutq}- CorOLLary 2. If for some Q the ring Tq is a complete intersection then so is T. The proof is by showing that under the assumption that Tgu(q) is a complete intersection, so is Tg. The argument in Section 2 of [K1] shows560 RICHARD TAYLOR AND ANDREW WILES that if a complete local noetherian ring S is a complete intersection, if f € S and if Homs($/(f), 8) = Homs($/(f), Anns(f)) is a free $/(f) module then ‘S/(f) is a complete intersection. We apply this result with $ = Tgu{q} and f = 6,—1. The final condition is met because the annihilator of 6, — 1 in Tu{q} is a free rank-one Tg module by the last corollary. COROLLARY 3. ng = n#Ag.- We remind the reader that ng is defined at the end of Section 1 and that 1 = ng. The proof of the corollary involves showing that nguq} = Na#Ag if q¢Q. Write Q’ = QU {q} and let denote the natural surjection Tg + Ta. It suffices to prove that in Tg we have the equation Anngg = (1+ 5, + ++ 6f4*)6-!(Anng, (pq). However 0-!(Anny,((9q)) is just the set of elements t of Tg for which tgg C (6, — 1)Tq. The inclusion Anngg: > (146,+---+684""")6-1(Anng, (9) is now clear. Conversely if t € Annggy then ¢ annihilates ker@ and hence t = (1 + 6, +---+6¢!")s. Then we see that sq: C (by — 1)Tqy, ie. that s € @-!(Annp,(~Q))- The other inclusion now follows. Coro.tary 4. #(agTQ/eQaQT@)#(O/n) = #(O/ng)- To prove this corollary note that ag/a2 ~ @ycq O/#Ag. Thus it follows from Theorem 2 that agTg/oTo ~ Tq®ojag]40/a%) ~ Bye T/#Ag and so we deduce that agT/pqaqT 9 ¥ @yeq O/#Ag. This corollary now follows from the last one. 3. Some algebra In this section we shall establish certain criteria for rings to be complete intersections. We rely on the numerical criterion established in the appendix of [W2]. In that appendix there is a Gorenstein hypothesis which can be checked in the cases where we apply the results of this section. However in order to state the results of this section in somewhat greater generality we shall refer to the paper [L], where the Gorenstein hypothesis in the appendix of [W2] is removed, rather than to the appendix of [W2] directly. Fix a finite flat reduced local © algebra T with a section x: T +» O. We consider complete local noetherian © algebras R together with maps R -» T. We denote by Jr the kernel of the map R -» T, by mp the induced map R -» O, by gp the kernel of wp and by nr the image under zp of the annihilator in R of gp. We let Ua = (91 Ir)/pRIR- If S +» R-» T then g3 — oR, Js is the pre-image of Jp and so Vs +» Up. We have an exact sequence (0) > Vr Jr/prJa— r/o r/e% — (0).RING-THEORETIC PROPERTIES OF CERTAIN HECKE ALGEBRAS. 561 From this we deduce the following facts. © #UR < 00. (To see this it suffices to show that (Vr)p, = (0). However, as T is reduced, (Jr)px = (@R)p_ and so the map (Jr/prIR)pp (9R/@%)pr is an isomorphism. The result follows.) © #VR#(0R/eR) = #(er/94)#(Ja/oRIR)- LEMMA 2. Suppose the inequalities #(er/o2) < #(O/nr)#VpR and #(O/nr)#(Jr/orJr) < #(O/nk) < 00 hold and suppose R is a finite flat O-algebra. Then R is a complete intersection. To show this first note that we have the inequalities #(0R/R) #(er/0%)#(Jr/orIr)/#VR #(9r/9%)#(O/nr)/(#¥R#(O/nr)) #(O/nr). Now applying the criterion of [L] we see that R is a complete intersection. < < Lemma 3. The following inequality exists: #(O/nr) < #(Jr/eRIn)#(O/m7). As Fittr(Jr) C Anna(Jp) we see that Fitto(Jr/grJr) C mrAnna(Jr). On the other hand it is easy to see that Anna(pr) > {s € R| ser C Jr}Anna(Jp)- Applying 7p we see that nr > nrFitto(Jr/prIr), and the lemma follows. Lemma 4. If R is a complete intersection which is finite and flat over O and nr # (0) then #(gr/p%) < #(O/nr)#¥r. If R is a power series ring the same result is true without the assumptions that it is finite over O and that nr # (0). For the first part we see that, as R is a complete intersection, #(pr/}) = #(O/nr) (see [L]). Thus, #Vn#(O/ne) = #(01r/9%)#Ur/orIR) = #(0r/e7)#(O/ne)/#(O/nr)- The first result follows. For the second result note that we can factor R —» T' as R-~ R’ + T with R’ a complete intersection which is finite and flat over O and for which gp:/p%y ~ g/g} (by the proof of Lemma 9 of [L]). Then #O/npr = #on//9% < « and so #pr/—% < #(O/nr)#VR. However Vp —» Wp, so the result follows. ‘We now return to the notation of the first section and let Yq denote Wp, and Ja denote Jng.562 RICHARD TAYLOR AND ANDREW WILES PROPOSITION 2. Suppose that for a series of sets Qn ideals In in Tq, with the following properties exist: 1. In is contained in my, and Tq,/In has finite cardinality. 2. Inst T C InT and Ay, InT = (0). 3. There is a surjective map of O-algebras T,4;/In+i -* TQn/In such that the diagram TQn41/Int1 —> TQ,/In | | T/Ing1T ——> T/InT commutes. (Note that this map is not assumed to take a given Hecke operator to itself.) 4. limT9, /In is a power series ring. Then for n sufficiently large, Tq, is a complete intersection, and hence T is a complete intersection (by Corollary 2 of Theorem 2). Let P denote limT9, /In. We get a natural map P +» T and can choose maps P + Tg, compatible with the maps Tg, » T and Tg, + Ta,/In- Because In C mip, qn We see that the map P — Ta, is surjective. We have a sequence Wp > Won — (Son + In) 9 (Gy, + In))/ (Ton Qn + In) (Note that although the maps P + Tg, and Vp + Wg, are not compatible as n varies, the composite map above is.) Moreover Up = lim((Jg, + In) (68, + In))/(Jon@Qn + In) (by the fact that Tg, /In is finite for all n) and so as Wp is finite, the map Up — ((Jg, + In) (8, + In))/(JQn@Qn + In) is injective for n sufficiently large. Thus for n sufficiently large Vp Wg,. We deduce the inequality #(0/6°) < HO/n#¥P = #O/N#¥Q,, where the first inequality follows from Lemma 4. The proposition follows on application of Corollary 4 of Theorem 2 and Lemma 2. PROPOSITION 3.’ Suppose that there are an integer r and a series of sets Qm with the following properties: 1. Ifq€ Qm then q=1modp™. 2. p is unramified at q and p(Frobg) has distinct eigenvalues.RING-THEORETIC PROPERTIES OF CERTAIN HECKE ALGEBRAS 563 3. #Qm =r. 4, Ta, can be generated as an O-algebra by r elements. Then T is a complete intersection. To prove this proposition it is useful to have the following definition. By a level n structure we mean a quadruple B = (A,a, 3,7), where « Ais an O-algebra, © a: Olff,...,T]] + A, © B: OffS1,...,Sr]]/(p, (Si +1)?" —1,.-.,(Sp+1)"" -1) + A makes Aa free module over O[[Si,..., Srl]/(p", (Si + 1)?" —1,..., (Sr +1)" - 1), @ and 7: A/(St,..-,S¢) 3 T/p™. Tf B is a structure of level n and n! 2 satisfy the requirements of Proposition 2 (using n > 2 to ensure that In C md, ). We need only check that lim Byn(n)n i8 a power series ring. On the one hand it is a finite free O[[S1,---, S-]]-module, and so has Krull dimension r + 1. On the other hand it is a quotient of O[[Ti,...,T;]] and so must in fact equal O[[71, ... Tr]. 4, Galois cohomology It remains to find a sequence of sets Qm with the properties of Propo- sition 3. We must recall some definitions in Galois cohomology. We define H}(Quad’?). 1. If ¢ p then H}(Qi,ad°p) = H(F, (ad°p)"") = ker(H"(Qi, ad’) > H}(h,ad°p)).564 RICHARD TAYLOR AND ANDREW WILES 2. If B |a, is flat and det |;,= € then we let H}(Qp, ad°p) denote those elements in H*(Qp,ad°p) C Extiig,)(Vp, Vp) which correspond to exten- sions which can be realised as the Q,-points on the generic fibre of a finite flat group scheme over Zp. LIEB lan a ri ) with v; |1,# € then we let H}(Qp,ad°p) denote 2 the kernel of H1(Qp,ad°B) > H}(Ip, (ad°7)/Homg(V5/F, F)), where F denotes the line in Vj where Gp acts by the character 1. y+ (3 4 we let H}(Qp,ad°p) denote the kernel of the map H1(Qp,ad°p) > H}(Qp, (ad°p)/Hom,(V5/F, F)), where F denotes the line in Vz where G, acts by the character 11. s . Finally if p \a,~ with 4 |),= € but 7 is not flat then We define H}(Q, ad%9) as the inverse image under H*(Q,ad°p) — TT #*(Qr,ad°A) 1gQ of Theo H}(Qr, ad). We also define H}(Qi, ad°(1)) to be the annihilator of H}(Qy, ad°p) under the pairing of Tate local duality H'(Qi,ad°7) x H?(Qi,ad°a(1)) + &. We then deine By. [Q, eB) to be the inverse image under BYR, ad’) — FE PNR, ad'A2)) (GQ of Tig 4} (Qi, ad°A(1)). Lema 5. dim, H}(Q,ad°p) < dim, H}.(Q,ad°p(1)) + #Q. To see this we apply Proposition 1.6 of [W2]. For | ¢ Q and | # p we see that hj = 1 because the index of H}(Qu, ad°p) in H1(Qi,ad°9) is equal to #H\(1h,ad°p)°F: which in turn equals #((ad"a(—1))n)Ft = #((0d°7()" Jay, = #H°Q,a€°9(2)). For q € Q we have that hg = #H°(Q,, ad°p(1)) = #k. It remains to check that Aphoo < 1. In the case that 7 |G, is not flat or det |1,# € this is proved in parts (iii) and (iv) of Proposition 1.9 of (W2]. Thus suppose that 7 |g, is flat and det |;,=«. We must show that dim, H}(Qp,ad°p) < 1+ dim, H°(Qp, ad°p) (= Lif |a, is indecomposable and = 2 otherwise).RING-THEORETIC PROPERTIES OF CERTAIN HECKE ALGEBRAS 565 Following [FL] let M/ denote the abelian category of k vector spaces M with a distinguished subspace M! and a k-linear isomorphism ¢ : M/M! @ M1 M. Then there are equivalences of categories among: ° mM”; finite flat group schemes A/Zp with an action of hj © K{G,]-modules which are isomorphic as modules over Fp[Gp] to the Qp points of some finite flat group scheme over Zp. See Section 9 of [FL] for details. The only point here is that an action of k on the generic fibre of a finite flat group scheme over Zp extends uniquely to an action on the whole scheme. Let M(p) denote the object of M corresponding to p. Then dim, M(p) = 2 and dim, M(p)! = 1 (since det |1,= €). We get an embedding Ext},(M(p), M(p)) @ H}(Qp,adp). We will show that 1. dim, Ext},(M(#), M(p)) = 2 if @ |, is indecomposable and = 3 other- wise. 2. The composite map Exth,(M(p), M(p)) + H1(Qp,adp) 4 H}(Qp,k) is nontrivial, where tr denotes the map induced by the trace. The lemma will then follow. For the first point it is explained in Lemma 4.4 of [R] how to calculate Ext;(M(p), M(p)). Let {eo,e1} be a basis of M(p) with e; € M(p)!. Let 4(e0,0) = ae + Ger and $(0,e1) = yeo + 6e1. Then Ext,(M(p), M(p)) can be identified as a k-vector space with Mz(k) modulo the subspace of matrices of the form Ce )GG E)-C5 O)G f)=(crtene “a’”): for any r,s,t € k. Thus dim, Ext\,(M(p),M(p)) = 2 if y # 0 and = 3 if 7 = 0. However = 0 if and only if M(p)* is a subobject of M(p) in M. This is true if and only if 7 has a one-dimensional quotient on which inertia acts by € which itself is true if and only if 2 |G, is decomposable. For the second point, consider the k{Gp|-module 7 @ 7 where T is the unramified representation 11 Frobp > ( an ) . Then 7 @ r is an extension of J by itself. Moreover its extension class maps to the element of H¥(Qp,k) = Hom(Q¥,k) which is trivial on Z¥ and takes p+ 2. Finally it is isomorphic to the action of G, on the Q» points of a finite flat group scheme over Zp, because this is true over an unramified extension.566 RICHARD TAYLOR AND ANDREW WILES LEMMA 6. Tg can be generated as an © algebra by dim, H3(Q,ad°p) elements. Let mg denote the maximal ideal of Tg. It will suffice to show that there is an embedding of k-vector spaces : Hom(mg/(m, ), k) + H9(Q,ad°A). We first define es Homy(mg/(md, d), ) — H*(Q, adp). If 8 is a nonzero clement of the left-hand group we may extend it uniquely to a map of local O-algebras 6 : Ta -» kle] where e? = 0. Let pp = 60 ply. We get an exact sequence (0) + Va > Von — Vs — (0); and hence a class (6) in Exthig, Ql (B,p) © H'(Q, ad). Because det pg is valued in k C Ale] we see that x(9) actually lies H1(Q, ad°p). We claim that resy(9) lies in H}(Qu,ad°) for | ¢ Q. This computation is very similar to some in [W2], but is not actually carried out there, so we give an argument here. First suppose that | # p and that either p | #(I1) or |g, is absolutely irreducible. In this case p(h) > p(I1) and det pQ |1, has order prime to I. Because either p { #p(I1) or p = 3 and adp(I;) = Aq we have that H1(p(Ii),ad°p) = (0) and so po |, B |1, @e kle]. The result follows in this case. Secondly suppose that |7, is unipotent and nontrivial. Then the same is true for pf |z, and then also for pp. However the Sylow p-subgroup of I; is pro-cyclic and so pg |z, must also be of the form 7 ®: k[é] and resyx(0) € H1(I;,ad°(f)) must vanish. In the case | = p, 9 is flat and det |1,= €, the claim is immediate from the definitions. In the case | = p and 3 |o,~ ( a b ) with th [z,4¢, use the fact that ply |1,~ ( ae ) where 1 denotes the Teichmiiller lifting of 1 |),. Finally in the case | = p, * Bla~ ( ra i‘ ) with 1 |1,=€ but 7 not flat, use the fact that , be * PQ lap~ ( 0 6 ) where 6 is an unramified character of order prime to p. It remains to show that « is injective. Suppose it is not. Then we could find a nonzero 8 such that po ~ Dy kle]. Thus tro is valued in O + kerd and in particular Tg is not generated as an O-algebra by trp. We will show this is not the case. If q € Q and 6 € Ag then we can find o € Gy, such thatRING-THEORETIC PROPERTIES OF CERTAIN HECKE ALGEBRAS 567 (Uq5)? — (trplg(o))(UaS) + det pig() = 0. (o will in fact lie above Frob,.) This polynomial has distinct roots in k and so both its roots in Tq lie in the sub- O-algebra T generated by the image of troy. Thus for all 6 € Ag, U,6 € T. Hence Uy € T, and as U, is a unit, 5 € T. Moreover for all | ¢ Q for which P is unramified we see that T;xq(Frob)~!/? € T and hence T; € T. If p| Ng then Upxq(Frobp)~¥/? is a root of the polynomial X?—(trpg(o))X +det pQ(7) for any element of Gp which lies above Frobp. For some o over Frobp this polynomial has two distinct roots in k and so Up € T. Thus T = Tg as we required. Finally we turn to the proof of the main theorem. As in [W2] (after equation (3.8)) we may find a set of primes Qm with the following properties: 1. if g € Qm then q = 1modp™; 2. ifq € Qm then J is unramified at q and p(Frob,) has distinct eigenvalues; 3. Hje(Q,ad°9(1)) > @gcQm H (Fas 0d°A(1)). As for each such q, H1(F,,ad°p) = k, we see that by shrinking Qm we may suppose that the latter map is an isomorphism. Then we have that #Qm = dim, H.(Q,ad°p(1)). Also H}, (Q,ad°A(1)) is the kernel of the map in 3 above and so is trivial. Thus by Lemma 5 we see that dim, HQ,,(Q, ad°p) < #Qm and so Tg,, can be generated by #Qm = dim, Hj.(Q, ad°7(1)) elements. The main theorem now follows from Proposition 3. CAMBRIDGE UniversiTy, CAMBRIDGE, U.K. PRINCETON Untversity, PRINCETON, NJ, U.S.A. REFERENCES [Cl] H. Caravor, Sur les représentations p-adiques associées aux formes modulaires de Hilbert, Ann. Sci. Ec. Norm. Super. 19 (1986), 409-468. ', Formes modulaires et représentations Galoisiennes & valeurs dans un an- neau local complet, in p-adic Monodromy and the Birch-Swinnerton-Dyer Conjecture (eds. B. Mazur and G. Stevens), Contemp. Math. 165 (1994) [D] _F. Diaonp, The refined conjecture of Serre, to appear in Proc. 1993 Hong Kong Conf. on Modular Forms and Elliptic Curves. [DT] __ F. Diamonp and R. Tavtoa, Lifting modular mod | representations, Duke Math. J. ‘74 (1994), 253-269. (FL] J.-M. Fontaine and G. Larartie, Construction de représentations p-adiques, Ann. Sci. Ec. Norm. Super. 15 (1982), 547-608. [CG] _B. Gross, A tameness criterion for Galois representations associated to modular forms mod p, Duke Math. J. 61 (1990), 445-517. {K1] _E. Kunz, Almost complete intersections are not Gorenstein, J. of Alg. 28 (1974), 111-115. , Introduction to Commutative Algebra and Algebraic Geometry, Birkhauser, (c2] (k2] 1985.568, RICHARD TAYLOR AND ANDREW WILES {L] __H. Lensrra, Complete intersections and Gorenstein rings, to appear in Proc. 1993 Hong Kong Conf. on Modular Forms and Elliptic Curves. (M] _B. Mazur, Modular curves and the Eisenstein ideal, Publ. Math. IHES 47 (1977), 133-186. [de Sh] E. pe Suatrr, On certain Galois representations related to the modular curve X1(p), to appear in Comp. Math. [R] _R. Ramaxrisuna, On a variation of Mazur's deformation functor, Comp. Math. 87 (1993), 269-286. [Wi] A. Wites, On ordinary )-adic representations associated to modular forms, Invent. Math, 94 (1988), 529-573. , Modular elliptic curves and Fermat's Last Theorem, Ann. of Math. 141 (1995), prior paper, this issue. [wa] (Received October 7, 1994) Appendix The purpose of this appendix is to explain certain simplifications to some of the arguments of Chapter 3 of [W2] and to Section 3 of this paper. These simplifications were found by G. Faltings and we would like to thank him for allowing us to include them here. We should make it clear that the arguments of this appendix (just like those of Chapter 3 of [W2] and Section 3 of this paper) apply only to proving Conjecture 2.16 of [W2] for the minimal Hecke ring and minimal deformation problem. In order to prove Theorem 3.3 of [W2] one needs to invoke Theorem 2.17 and the arguments of Chapter 2 of [W2]. We will keep the notation and assumptions of the main body of this pa- per. Let Q denote a finite set of primes as described in Section 1 of this paper. By a deformation of 7 of type Q we mean a complete noetherian local O-algebra A with residue field k together with an equivalence class of contin- uous representations p : Gq —> GL2(A) with the following properties: © pmodm4 « “det p is a character of finite order prime to p; @ if ¢ QU {p} and p |y, is semisimple then p(i) ~ pL); . - 1 1+ 1 QU Gp) and ali ( j + ) sien in (5 :) © if pis flat and det |;,= then p is flat; © if either D is not flat or if det |),4 € then p |c,~ ( A A ) where $2 2 is unramified and ¢2mod ma = 2.RING-THEORETIC PROPERTIES OF CERTAIN HECKE ALGEBRAS 569 As in Chapter 1 of [W2] there is a universal lift pg” : Gq — GLa(Rg) of type Q. Recall that the universal property is for lifts up to conjugation. Moreover, one checks (cf. the second paragraph of the proof of Lemma 6) that there is a natural isomorphism Homg(mrg/(A, mq), k) ¥ Ho(Q, ad°2). There is also a natural map Ra —+ Tq so that py’ pushes forward to a conjugate of py. Recall that if Q = 0 we shall often drop it from the notation. In this appendix we shall reprove the following result which combines Theorem 1 of this paper and Theorem 3.1(ii) of [W2]. THEOREM. RT and these rings are complete intersections. We note that if O’ is the ring of integers of a finite extension K’/K then the rings TQ and Rg constructed using O' in place of O are just Tg @o O' and Rg @o O'. Also Tg is a complete intersection if and only if Tg @o O! is (by, for instance, Corollary 2.8 on page 209 of [K2]). Thus we may and we shall assume that O is sufficiently large that the eigenvalues of every element of p(GQ) are rational over k. We recall that in the penultimate paragraph of Section 4 of this paper we showed that Tg is generated as an O-algebra by trp9(Gq). Thus we see that the map Rg — TQ is a surjection. We need the following result. oi 0 0 oe both of these characters factor through Xq : Iq» Aq. Lema. If € Q then pi |c,~ ( where $1 |1,= $2 |7,) and It suffices to check the first assertion. As 7 is unramified at q, pf |e, factors through ZxZ,(1), where Z is topologically generated by some lift f of Froby, Zp(1) is topologically generated by some element o and where fo f~? 04. As p(Frobg) has distinct eigenvalues it is easy to see that after conjugation we may assume that pij'™(f) = ( ‘ t ) where a # bmod mpg. We will show that pg'"(c) is a diagonal matrix with entries congruent to 1modmpg. We will in fact prove this mod ma for all n by induction on n. For n = 1 there is nothing to prove. So suppose this is true modulo mf, with n > 0. Then 0 0 pe ) (1p +N) mod mgt,570 RICHARD TAYLOR AND ANDREW WILES where each 4; = 1 mod mpg and where N = 0modm%,,. We see that mod mj", 2 Rea a a0 a0 wet 1+ N - 1+N)* (3 5)*(a8) (0)! , 9 (4 ype )O+aN) a 0 ( as io ) +N, and as a # bmodmag we deduce that N is diagonal mod mt} as required. Choosing $2 so that ¢2(f) = 6,modmpg, we can define a map Ay > RS to be ds [f,. This makes Rg into an O[Ag|-algebra. Using the last lemma and the universal properties of Rg and of R it is easy to see that Re/ag ~ R. It moreover follows from the discussion preceding Theorem 1 of this paper that the map Rg — To is a map of O[Ag]-algebras. The key observation is the following ring-theoretic proposition. The theo- rem follows on applying it to the rings Rg, and Ta,, for the sets Qn constructed in Section 4 of this paper. Note that there is a map O[[S1,..., S;]] ~ O[Ag,] with kernel the ideal (1 + 51)#4n —1,...,(1 + S,)#4e — 1), where Qn = {q1,-+-,r}- Note also that by the displayed isomorphism five lines before the theorem there exists a surjection of O-algebras O[[X1,-..,Xr]] -» Ran: MW PROPOSITION. Suppose r is a nonnegative integer and that there is a map of O-algebras R -» T with T finite and flat over O. Suppose for each positive integer n there are a map of O-algebras Rn —» Ty and a commutative diagram of O-algebras OllS1,...,S]] > Rn ~ RB J L Tt > T, where 1. there is a surjection of O-algebras O[[X1,....X-]] -» Rn: 2. (Si,---$;)Rn C ker(Rn -» R), 3. (Si,-+5$+)Tn = ker(In + T), 4. if by denotes the kernel of O[[S1,-..)Sr]] + In then by C ((1+ 51)" — 1,...(1 +5," -1) and Tp is a finite free O[[S1,...,Sr]]/bn-module. Then RT and these rings are complete intersections.RING-THEORETIC PROPERTIES OF CERTAIN HECKE ALGEBRAS 571 Reducing mod we see that it suffices to prove this result with k re- placing O everywhere. In this case we see that the last condition becomes bn C (SP",...,52"). Further we may replace R by its reduction modulo mgker(R — T), and so we may assume that R is finite over k. We may replace Ty by Tn/(Sf’,..., SP") and so assume that bn = (S?",..., 52"). Finally we may replace Ry, by its image in R ® Tp. Now define an n-structure to be a pair of k-algebras B -» A together with a commutative diagram of k-algebras K(ISi,...,5-]] J KLX1,..., Xe] B > R L tL A > T, such that 1. BORA, 2. (Si,...,S-)B C ker(B R), 3. ($1,...,5,)A = ker(A - 7), 4, Aisa finite free k[[S1,...,S,]]/(SP",...,.52")-module. Note that #B < (#T)?" #R and thus there are only finitely many isomor- phism classes of n-structures. If S is an n-structure and if m 1, 10) 2 SOD. Then set S/, = 8°")... ‘Thus we obtain a commutative diagram Ri... Ro > Rom J J t Th. Tw To» Ae yw572 RICHARD TAYLOR AND ANDREW WILES of k{[X1,..., Xr, S1,-.-, Sp])-algebras. Moreover, we have that © ALXi,...,Xsl] + Rh > Th, © T’, isa finite free k{[S1,...,S;]]/(SP",...,S2")-module, © Ri /(S1,..-, Se) » Rand Ti/(S1,...,$,) > 7. Let R’,, denote the quotient of k[[X1,...,X;]] by the intersection of the ideals ker(k[[X1,...,X;]] » Rh) and let T%, denote the quotient of k[[X1,...,Xr]] by the intersection of the ideals ker(k{[X1,...,X;]] -» Ty). Then we have a commutative diagram K([S1,--., Se] L KX, Xl] > Ry ~ oR J t Teo ~ 1, such that © Ry > Te: © Ti, isa finite free k([$1,...,5,]]-module, © Ri,/(S1,---55r) » Rand Th /(S1,...,5;) > 7. We deduce that Ti, has Krull dimension r and hence that the map k([X,...,Xr]] » Th has trivial kernel. That is, we have isomorphisms A{[X1,...,Xr]] S RL = Th. Thus RT. As T has Krull dimension 0 and T = k{[X1,..., Xr]]/(S1,--- 5p) we see that T’ is a complete intersection, and the proposition is proved. (Appendix received January 26, 1995) CamprinGe UNIVERSITY, CAMBRIDGE, U.K. PRINCETON UNIVERSITY, PRINCETON, NJ, U.S.A.