Arxiv:1512.06894V1

Home , Christopher Skinner

arXiv:1512.06894v1 [math.NT] 21 Dec 2015 (1.1.a) seas Tt6 n Br1 I-I))i h following: the 1.1.1 is Conjecture (I)-(IV)]) [Bir71, and [Tat66] also (see h ic n wnetnDe ojcuefor Conjecture Swinnerton-Dyer and Birch The ecl h oml 11a n()the (b) in (1.1.a) formula the call We 1.1. qaefe) Let square-free). htord that oyai Kl0 o9a o9b seas Go1)impli [Gro91]) also (see Kol91b] Kol91a, [Kol90, Kolyvagin mod 1.2. H IC N WNETNDE OML O ELLIPTIC FOR FORMULA SWINNERTON-DYER AND BIRCH THE (b) (a) h ic n wnetnDe conjecture. Swinnerton-Dyer and Birch The anresult. Main p p rv the prove Abstract. hsfruaas od for holds also formula This at ( gw ubrof number agawa Let h rank the egtpiigon height-pairing h re ftezr at zero the of order The aosrepresentation Galois a s ≥ =1 p Z . fgo euto uhthat such reduction good of 5 L Reg( bssfrtedffrnil fteNeo oe of Néron model the of differentials the for -basis ( E/ IIA ECE,CRSOHRSINR N I WAN XIN AND SKINNER, CHRISTOPHER JETCHEV, DIMITAR p r E/ Q pr fteBrhadSinro-yrfruafor formula Swinnerton-Dyer and Birch the of -part ord Let fteMrelWi group Mordell–Weil the of s , p Q BrhadSinro-yrConjecture) Swinnerton-Dyer and (Birch r UVSO NLTCRN ONE RANK ANALYTIC OF CURVES Let p .I hscs h oko rs n air[Z6 n of and [GZ86] Zagier and Gross of work the case this In 1. = ) ! ) ≥ E/ · etergltrof regulator the be Ω Ω Q eapieo odrdcin(i.e., reduction good of prime a be 3 L E/ E E E E ( easmsal litccresc htord that such curve elliptic semistable a be L · r ( · taprime a at Q ) Reg( Q ′ Reg( ( ( E, E, ) ), easmsal litccreo conductor of curve elliptic semistable a be p ρ 1) E/ E,p 1) provided 3 = s X E/ Q 1 = 1. Gal( : ( ) Q E/ E Introduction ) ℓ S oml for formula BSD [ ord = p fteHasse–Weil the of and , Q = sirreducible: is ] ) Q (# h aeSaaeihof Tate-Shafarevich the / # 1 p E a Q Ω   p X E E ( ( ) E Q # E ( ( ( Q if 0 = ) X Q ) → E/ = E ) ) tedsrmnn fteNéron-Tate the of discriminant (the tor ( . R ssae yTt Tt4 oj 4] Conj. [Tat74, Tate by stated as , Q E/ Aut( E ) ) ( 2 Q R E ) ) Y E ℓ | ℓ E Y a uesnua reduction supersingular has ≤∞ ω Let . L [ E c p -function ℓ c | )i reuil.Suppose irreducible. is ]) ( . ℓ sta rk that es E for E/ E/ ( E/ E/ s over =1 Q Q ω Q p Q L ) E E ) ,   ( ea litccurve. elliptic an be ∤ o ahprime each for , ,s E, Néron differentiala Z . L N c ) ℓ ( Then . Z ( .We 1. = ) uhta the that such ) ,s E, E/ E ( Q Q ) N and 1 = ) ) seulto equal is h Tam- the (so N is 2 DIMITARJETCHEV,CHRISTOPHERSKINNER,ANDXINWAN

#X(E/Q) < . In particular, part (a) of Conjecture 1.1.1 holds for E. In this paper we prove that∞ the p-part of the BSD formula (1.1.a) holds for E: Theorem 1.2.1 (p-part of the Birch and Swinnerton-Dyer formula). If p 5, then ≥

L′(E, 1) (1.2.a) ordp = ordp #X(E/Q) cℓ(E/Q) . Reg(E/Q) ΩE   ℓ∤ · Y∞   If p = 3, then (1.2.a) holds provided ap(E) = 0 when E has supersingular reduction at p.

L′(E/Q,1) It is a consequence of the Gross–Zagier formula that Q× (see [GZ86, Reg(E/Q) ΩE ∈ Thm.7.3]), so the p-adic valuation of the left-hand side of (1.2.a)· makes sense. Particular cases of Theorem 1.2.1 have been obtained by Zhang [Zha14] and by Berti, Bertolini, and Venerucci [BBV15], for p 5 a prime of good ordinary reduction. In particular, [Zha14, Thm. 7.3] also has the≥ extra assumption that for any ℓ N for which ℓ 1 mod p, ρ is ramiﬁed at ℓ (equivalently, that p ∤ c (E/Q)), and [BBV15,|| ≡± E,p ℓ Thm. A] also assumes that p is not anomalous (that is, p ∤ #E(Fp)) and that p does not divide any of the Tamagawa factors cℓ(E/Q). In contrast, Theorem 1.2.1 is general – including both the ordinary and supersingular cases, as well as the cases where there are Tamagawa numbers divisible by p – aside from the extra hypothesis that ap(E) = 0 when E has supersingular reduction at p if p = 3.

If p is a prime of supersingular reduction for E and ap(E) = 0 (which is always the case for supersingular p 5), then the conclusion of Theorem 1.2.1 follows from combining the work of Kobayashi≥ on the p-adic Gross–Zagier formula and the non-vanishing of the p-adic height of the Heegner point [Kob13, Thm. 1.1 and Cor. 4.9] with Wan’s recent work on Kobayashi’s supersingular variant of the Iwasawa main conjecture for E [Wan14a]; cf. [Kob13, Cor. 1.3]. This argument does not apply in the case where E has ordinary reduction at p as the p-adic height of the Heegner point is then not known to be non-zero. The proof of Theorem 1.2.1 in this paper treats the ordinary and supersingular cases the same and also applies to GL2-type modular abelian varieties. If E has complex multiplication (in which case E is not semistable and N is not square- free), the equality (1.2.a) similarly follows from the p-adic Gross–Zagier formula together with the non-vanishing of the p-adic height of the Heegner point and the Iwasawa main conjectures for E; see [Kob13, Cor. 1.4].

When ords=1L(E,s) = 0 (that is, L(E, 1) = 0) then the p-part of the Birch and Swinnerton-Dyer conjectural formula is also known.6 In fact, the equality in this case is an important ingredient in our proof of Theorem 1.2.1 as well as the proofs of all the results described above. For more on what is known in this case see Theorem 7.2.1 below.

1.3. Outline of the proof. Theorem 1.2.1 is proved in this paper by separately estab- lishing the upper and lower bounds predicted by (1.1.a) for the order #X(E/Q)[p∞] of the p-primary part of the Tate–Shafarevich group. THEBSDFORMULAFORTHERANKONECASE 3

To prove the exact lower bounds we use anticyclotomic Iwasawa theory. For a suitable imaginary quadratic field ′ = Q(√D′) of discriminant D′ < 0, using arguments similar to [Gre99, 4] we prove a controlK theorem, Theorem 3.3.1, that compares a specialization § of a certain Λ-cotorsion Selmer group (here Λ = Zp[[Gal( ′ / ′)]] is the Iwasawa algebra K∞ K for the anticyclotomic Zp-extension ′ of ′) to a certain Zp-cotorsion Selmer group K∞ K that is closely related to the p-primary part X(E/ ′)[p∞] of the Tate–Shafarevich group. This comparison is stated as an explicit formula involvingK the Tamagawa numbers of E at primes that split in ′. The Λ-cotorsion Selmer group is the Selmer group related via an anticyclotomic mainK conjecture to a p-adic L-function recently constructed for modular curves by Bertolini, Darmon and Prasanna [BDP13] and extended to Shimura curves by Brooks [Bro14] and Liu, Zhang, and Zhang [LZZ13]. This p-adic L-function is also a specialization of a two-variable p-adic L-function constructed by Hida, as explained in Section 5. We note that the anticyclotomic main conjecture that we use is different from the classical anticyclotomic main conjecture of Perrin-Riou formulated in [PR87] (see also [How04, p.3]): the p-adic L-functions in the two cases have different ranges of interpolation. Using an extension of the methods of [SU13], Wan [Wan13, Wan14a] has proved that this p-adic L-function divides the characteristic ideal of the Λ-cotorsion Selmer group up to a power of p (see Section 5 for the definitions of the relevant p- adic L-functions and Section 6 for the precise statement of the main conjecture). To extend this to an unambiguous divisibility in Λ and to pass from this divisibility to lower bounds on #X(E/ ′)[p∞] in terms of the index of a suitable Heegner point, we need two key ingredients:K 1) a recent result of Burungale [Bur14] on the vanishing of the corresponding analytic µ-invariant, and 2) a central value formula due to Brooks [Bro14, Thm.1.1] generalizing a recent formula of Bertolini, Darmon and Prasanna relating the p-adic L-function at a point outside of the range of interpolation to the p-adic logarithm of a Heegner point z ′ E( ′). The result is the inequality K ∈ K

2 (1.3.a) ordp(#X(E/ ′)[p∞]) ordp [E( ′) : Z z ′ ] cw(E/ ′) K ≥  K · K K  w|N ′  w a splitY prime of K    The Heegner point z ′ appearing in this formula (and also appearing in the formula of K + Brooks) comes from a parameterization of E by a Shimura curve XN +,N − , N N − = N, + of level N attached to a quaternion algebra B = BN − of discriminant N −. In order to appeal to the known results about the anticyclotomic main conjecture of interest, it is necessary to take N − > 1 (so this is not the classical Heegner point setting). To pass from the inequality (1.3.a) to one where the right-hand side is replaced with an L-value, we combine the inequality with the general Gross–Zagier formula for the point z′ , due to Zhang [Zha01a] and Yuan, Zhang, and Zhang [YZZ13]. The result is the exactK lower bound for #X(E/ )[p ] predicted by the BSD formula for E/ (see K′ ∞ K′ Conjecture 7.1.1). We note that the Tamagawa numbers at the non-split primes of ′ now appear, coming into the Gross–Zagier formula as the ratio of the degree of theK usual modular paramaterization and the degree of the Shimura curve parameterization (this is essentially due to Ribet and Takahashi [RT97]). Finally, to obtain the expected lower bound on #X(E/Q)[p∞] we express both sides of the resulting inequality for 4 DIMITARJETCHEV,CHRISTOPHERSKINNER,ANDXINWAN

D′ #X(E/ ′)[p∞] in terms of E/Q and its ′-twist E /Q. We then exploit the fact that K K D′ Kato has proved that the predicted upper bound holds for #X(E /Q)[p∞] (as ′ is ′ K chosen so that L(ED , 1) = 0). 6 Upper bounds on #X(E/Q)[p∞] are typically achieved by an Euler system argument via the method developed by Kolyvagin [Kol90], which applies since E/Q is assumed to have analytic rank one. Kolyvagin’s method uses the Euler system constructed from Heegner points that are the images of CM moduli via the usual modular parameterization X0(N) E. As explained in [Jet08], this has a drawback in the sense that it will only give the→ precise upper bounds on the p-primary part of the Tate-Shafarevich group (for E over a suitable imaginary quadratic field) if at most one Tamagawa number of E is divisible by p. To get around this problem we again consider a parameterization of E by a + general Shimura curve XN +,N − , N N − = N. As explained in Section 7.4.2, it is possible to choose this parametrization so that no Tamagawa number at a prime dividing N + is divisible by p. We then choose an imaginary quadratic field ′′ = Q(√D′′), D′′ < 0, such that each prime dividing N + splits in and each primeK dividing N is inert in K′′ − ′′ (this will generally be a different field than the ′ used to establish the lower bound). KKolyvagin’s method applied to the Heegner pointsK obtained from the parameterization by X + − and the field then yields an inequality in the direction opposite of (1.3.a): N ,N K′′ 2 (1.3.b) ordp(#X(E/ ′′)[p∞]) ordp [E( ′′) : Z z ′′ ] . K ≤ K · K Appealing to the general Gross–Zagier formula for z ′′ we then get the upper bound on #X(E/ )[p ] predicted by the BSD formula forKE/ . To pass from this to the K′′ ∞ K′′ expected upper bound for #X(E/Q)[p∞] we make use of the fact that the predicted D′′ lower bound for #X(E /Q)[p∞] is known. This lower bound follows from the proved ′′ cases of the cyclotomic main conjectures for ED (see [SU13, Thm.2(a)] and [Ski14b, Thm.C] for the ordinary case and [Wan14b, Thm.1.3] for the supersingular case).

′′ It is only at the ﬁnal step, where we invoke the p-part of the BSD formula for L(ED , 1), that we need to assume that ap(E) = 0if E has supersingular reduction at p, which is only a real condition when p = 3. Furthermore, most of our arguments apply more generally to the situation where E is replaced by a newform of weight 2, square-free level, and trivial character (see also Section 7.4.4 for additional comments on the general case).

1.4. Organization of the paper. In section 2 we recall some relevant background on Galois representations, local conditions and Selmer modules (generalizations of classical Selmer groups), as well as the speciﬁc cases arising from newforms and modular abelian varieties. In Section 3 we prove the control theorems for the relevant anticyclotomic Selmer groups. In Section 4 we include the relevant background on quaternion algebras, Shimura curves, CM points, and the Kolyvagin system coming from a Shimura curve, and we recall the upper bounds on #X(E/ )[p∞] obtained from Kolyvagin’s argument in the setting of Shimura curves. SectionK 5 is about p-adic L-functions and various comparisons. In it we recall the p-adic anticyclotomic L-function constructed THEBSDFORMULAFORTHERANKONECASE 5 by Brooks [Bro14] and compare it to a specialization of Hida’s two variable p-adic L- function [Hid88]. Section 5 also includes the statement of Burungale’s result on the vanishing of the analytic anticyclotomic µ-invariants and the statement of Brooks’ result expressing a certain value of the anticyclotomic p-adic L-function in terms of a p-adic logarithm of a Heegner point. In Section 6 we discuss the relevant anticyclotomic main conjectures and recent progress on proving them. We complete the proof of our main result, Theorem 1.2.1, in Section 7.

2. Preliminaries

Let Q C be the algebraic closure of Q. For any number field F Q, let G = ⊂ ⊂ F Gal(Q/F ). For each place v of F , fix an algebraic closure F v of Fv and an F -embedding ι : Q ֒ F to get an identification of G = Gal(F /F ) with a decomposition group for v → v Fv v v v in GF . For each finite place v, let Iv GFv be the inertia subgroup and Frobv GFv /Iv 1 ⊂ ∈ a geometric Frobenius element. Let Fv be the residue field of v and Fv an algebraic closure of Fv. Then there is a canonical identification GFv /Iv = Gal(Fv/Fv).

Throughout, let p 3 be a ﬁxed prime and let ǫ: GQ Zp× be the p-adic cyclotomic character. ≥ →

2.1. Galois representations. Let F be a number field (for much of this paper F will be either Q or an imaginary quadratic field ). By a p-adic Galois representation of G K F we will always mean a finite-dimensional vector space V over a finite extension L/Qp that is equipped with a continuous L-linear GF -action. Such a representation V will be understood to come with a scalar field L. We will always assume that

(geom) V is geometric in the sense introduced by Fontaine and Mazur [FM95]: V is unramiﬁed away from a ﬁnite set of places and potentially semistable at all places w p of F . We will further assume that |

(pure) V is pure.

In particular, this means that there is some integer m such that for any finite place w of F at which V is unramified, all the eigenvalues of Frobw are Weil numbers of absolute m/2 value (#Fw) . More generally, (pure) means that for all finite places w, the Frobenius

1Throughout this paper we take geometric normalizations (e.g., for Frobenius elements, for the reciprocity maps of class ﬁeld theory, for Hodge–Tate weights, for L-functions of Galois representations). 6 DIMITARJETCHEV,CHRISTOPHERSKINNER,ANDXINWAN

2 semi-simplification of the Weil–Deligne representation W Dw(V ) associated with V GFw is pure of weight m in the sense defined by Taylor and Yoshida [TY07]. | Let L be the ring of integers of L. We will generally choose a G -stable - O ⊂ F O lattice T V and let W = V/T . The latter is a discrete -divisible GF -module; it is canonically⊂ identified with T L/ . Note that the isomorphismO class of T , and hence also that of W , is not necessarily⊗O uniquelyO determined. In what follows we will often fix such a triple (V,T,W ). Let m be the maximal ideal and κ = /m the residue field. Let V be the semi- ⊂ O O simplification of the finite-dimensional κ-representation T/mT . Then V is uniquely- determined up to isomorphism; it is independent of the choice of T . Furthermore, if V is irreducible, then all lattices T V are homothetic (and so the isomorphism of class of ⊂ T is unique). Note that while our definition of V commutes with extension of the scalars κ, the property of being irreducible may not.

2.2. Local conditions. Let F be a number ﬁeld and let (V,T,W ) be as in Section 2.1. Let M be an -module with a continuous GF -action. By a local condition for M at a O 1 place w of F we mean a subgroup of the local cohomology group H (Fw, M). We discuss several local conditions that will be used throughout.

2.2.1. The unramified local condition. For a finite place w of F , the unramified local condition is defined as H1 (F , M) = ker H1(F , M) H1(I , M) , ur w { w → w } where I G is the inertia group at w. Note that we also have w ⊂ Fw 1 1 Iw Hur(Fw, M) = H (Fw, M ).

2.2.2. The finite local condition. Following Bloch and Kato [BK90], for a finite place w of F we define the finite local condition for V to be 1 Hur(Fw,V ) w ∤ p 1 1 1 ∞ H (Fw,V )= ker H (F ,V ) H (F ,V B ) w p f  { w → w ⊗Qp cris } | 0 w . |∞  2For w | p this was defined by Fontaine via p-adic Hodge theory: Let

GE Dpst(V )= (V ⊗ Bst) ,

E/F[w with E running over all ﬁnite extensions of Fw and Bst being Fontaine’s ring of semistable p-adic periods. ur This is a free L ⊗Qp Fw -module of rank two with an induced action of the monodromy operator N and

Frobenius ϕ of Bst. The Weil-Deligne representation associated to V |GFw by Fontaine is W Dw(V ) = ur ur Dpst(V ) ⊗L⊗Qp Fw Qp (chose any embedding Fw ֒→ Qp) with the induced action of N. The action of ur the Weil group WFw ⊂ GFw is deﬁned by twisting its L-linear, Fw -semilinear action rsl on Dpst(V ). An ν(g) element g ∈ WFw acts on W Dw(V ) as rsl(g)ϕ , where ν : WFw → Z is the normalized valuation map. THEBSDFORMULAFORTHERANKONECASE 7

The ﬁnite local condition for V can be propagated to T and W via the exact sequence 0 T V W 0. The resulting local conditions are → → → → H1 (F , T ) = preimage of H1 (F ,V ) under the map H1(F , T ) H1(F ,V ) f w f w w → w and H1 (F ,W ) = the image of H1 (F ,V ) under the map H1(F ,V ) H1(F ,W ). f w f w w → v Note that for w ∤ p we have H1 (F , T ) H1 (F , T ) and H1 (F ,W ) H1 (F ,W ) ur w ⊂ f w f w ⊂ ur w and that neither inclusion need be an equality.

We note for later use that if dimL V = 2 and V is pure of weight diﬀerent from 0 and 1, 1 1 1 1 then for w ∤ p, Hf (Fw,V ) = Hur(Fw,V ) = 0, and so Hf (Fw,W )=0 and Hf (Fw, T ) = 1 0 0 Hf (Fw, T )tor. In fact, in this case we have H (Fw,V )=0=H (Fw,V ∨(1)) = 0, so 1 1 1 H (Fw,V ) = 0 and hence H (Fw, T ) and H (Fw,W ) both have ﬁnite order.

2.2.3. The anticyclotomic local condition. Suppose now that F = is an imaginary quadratic field. Suppose also that p splits in as p = vv. We defineK the anticyclotomic local condition for V as K 1 H ( v,V ) if w = v, 1 1 K H ( w,V )= H ( ,V ) if w ∤ p is split in , ac K  f Kw ∞ K 0 else. Note that this definition involves the choice of a prime v of above p.  K We propagate the anticyclotomic local condition via 0 T V W 0, getting anticyclotomic local conditions for T and W . In particular,→ → → → 1 H ( v,W )div if w = v, 1 1 K H ( w,W )= H ( ,W ) if w ∤ p is split in , ac K  f Kw ∞ K 0 else.  As noted in Section 2.2.2, if dim L V = 2 and V is pure of weight different from 0 or 1, then the conditions at w ∤ p agree in the split and non-split cases (the local condition is just 0). ∞

2.3. Selmer structures and Selmer modules. Following [MR04, Ch.2], a Selmer 1 1 structure on M is a choice of a local condition H (Fw, M) H (Fw, M) for each F 1 F ⊆ 1 place w of F such that for all but ﬁnitely many w, H (Fw, M) = Hur(Fw, M). A Selmer structure on M has an associated Selmer moduleF deﬁned as F 1 1 1 1 H (F, M) := ker H (F, M) H (Fw, M)/ H (Fw, M) , F → F ( w ) M where the sum is taken over all places w of F . 8 DIMITARJETCHEV,CHRISTOPHERSKINNER,ANDXINWAN

If is a Selmer structure on M, then we deﬁne the dual Selmer structure on F F ∗ M ∗ = Homcont(M, Qp/Zp(1)), the arithmetic dual of M, as 1 1 H ∗ (Fw, M ∗) = the annihilator of H (Fw, M) via local duality. F F

For the purpose of this paper, if S and S′ are two ﬁnite sets of ﬁnite places of for which S′ K S S′ = then S will denote the Selmer structure obtained from by replacing the local∩ conditions∅ atF the places in S with the trivial local conditions and theF local conditions 1 1 at the places in S′ with the relaxed local conditions (i.e., H S′ ( w, M) = H ( w, M) for S K K 1 F w S′ and H S′ ( w, M)=0 for w S). ∈ FS K ∈

2.3.1. A consequence of Poitou-Tate duality. If and are two Selmer structures on 1 1 F G M, we write if H (Fw, M) H (Fw, M) for every place w of F . If , there is a perfect bilinearF G pairingF ⊆ G F G 1 1 H (Fw, M) H ∗ (Fw, M ∗) 1G F1 Q/Z H (Fw, M) × H ∗ (Fw, M ∗) → F G that is induced from the Tate local pairing. The following theorem is a consequence of the Poitou–Tate global duality theorem (see [Rub00, Thm.1.7.3], [Mil86, Thm.I.4.10] and [Tat63, Thm.3.1]): Theorem 2.3.2. Let be two Selmer structures on M and consider the exact sequences F G

G 1 1 1 locF H (Fw, M) H (F, M) ֒ H (F, M) 1G 0 → F → G −−−→ w H (Fw, M)! M F and ∗ locF 1 1 1 G∗ H ∗ (Fw, M ∗) , H ∗ (F, M ∗) ֒ H ∗ (F, M ∗) F1 0 → G → F −−−→ ∗ w H (Fw, M ∗) ! G ∗ M where locG and locF∗ are the natural restriction maps and the sum is over all places F 1 G 1 ∗ w, for which H (Fw, M) ( H (Fw, M). The images of locG and locF∗ are orthogonal F G F G complements with respect to the pairing , obtained from the local Tate pairings h− −iw w on the local cohomology groups. X

2.3.3. The finite and anticyclotomic Selmer structures. Let (T,V,W ) be as in Section 2.1 and let M be one of T , V , or W . The finite (or Bloch-Kato) Selmer structure BK is defined by the finite local conditions F 1 1 H (Fw, M) = Hf (Fw, M). FBK THEBSDFORMULAFORTHERANKONECASE 9

Note that is just the finite Selmer structure on M . If F = is an imaginary FBK∗ ∗ K quadratic field in which p splits as p = vv¯, then the anticyclotomic Selmer structure ac is defined by the anticyclotomic local conditions F 1 1 H ac ( w, M) = Hac( w, M). F K K 2.3.4. Iwasawa-theoretic Selmer structures. Let be an imaginary quadratic field and K 2 let F denote either the anticyclotomic Zp-extension or the Zp-extension of . As before, ∞ K we assume that p splits in , i.e., p = vv. Let R = [[Gal(F / )]] be the associated K O ∞ K Iwasawa algebra and consider the R-module M = T R where R = Hom ,cont(R,L/ ). ⊗O 1 O O The module M is equipped with a G -action given by ρ Ψ− where Ψ: G R× is K ⊗ K → the character naturally defined by the projection G b Gal(F b/ ). K → ∞ K We define two Selmer structures ac and Gr on M that we refer to as the anticyclotomic and the Greenberg SelmerF structures,F respectively. The anticyclotomic Selmer 1 structure H ac ( w, M) is defined by F K 1 H ( v, M) if w = v 1 1 K H ac ( w, M)= Hur( w, M) if w ∤ p is split, F K  K ∞ 0 else. The Greenberg Selmer structure is defined by  1 1 H ( v, M) if w =v, ¯ H Gr ( w, M)= 1 K F K H ( , M) else. ( ur Kw Remark 2.3.5. When F is the anticyclotomic Zp-extension of , the anticyclotomic ∞ 1 K Selmer structure gives rise to a global Selmer module H ac ( , M) whose Pontrjagin dual 1 F K Xac(M) = Hom (H ac ( , M), L/ ) appears in the statement of the anticyclotomic main conjecture thatO F willK be relevantO to our argument (see Section 6.1 for more details). Remark 2.3.6. For the particular modules M = T R that we will consider later, the G -module T R can be viewed as a p-adic family⊗O of motivic p-adic Galois represen- K ⊗O tations. Moreover, the inductions of these representationb s to GQ satisfy the Panchishkin condition at p (see [Gre94, p. 211]). In this case, the Greenberg Selmer condition above can be identified with a special case of the Selmer conditions defined by Greenberg 2 [Gre94]. When F is the Zp-extension of , the Iwasawa–Greenberg conjecture [Gre94, ∞ K 1 Conj. 4.1] relates the characteristic ideal of XGr(M) = Hom (HGr( , M), L/ ) to the two-variable p-adic L-function constructed by Hida [Hid88] (seeO SectionK 5.2 forO details).

2.4. Newforms, their Galois representations, and modular abelian varieties. We introduce notation for and recall basic properties of newforms and their associated p-adic Galois representations.

2.4.1. Newforms. Let f S2k(Γ0(N)) be a normalized newform of weight 2k, level N, ∈ n and trivial Nebentypus. Let f = ∞ a (f)q be its q-expansion at the cusp . The n=1 n ∞ Fourier coeﬃcients an(f) generate a ﬁnite extension Q(f) C of Q. Fix an embedding P ⊂ 10 DIMITARJETCHEV,CHRISTOPHERSKINNER,ANDXINWAN

Q(f) ֒ Qp and let L Qp be a ﬁnite extension of Qp containing the image Q(f). Let be the→ ring of integers⊂ of L. O 2.4.2. Galois representations associated to newforms. Associated to f, L and the ﬁxed embedding Q(f) ֒ L is a two-dimensional L-space Vf with a continuous, absolutely → 3 irreducible L-linear GQ-action that is characterized by the equality of L-functions :

L(Vf∨,s)= L(f,s), 4 where Vf∨ is the L-dual of the GQ-representation Vf . This equality of L-functions can be refined as follows. Let π = ℓ πℓ be the cuspidal automorphic representation such that L(π,s 1/2) = L(f,s). Then⊗ ≤∞ the Frobenius-semisimplification of the Weil–Deligne − representation W Dℓ(V ∨) associated with each local Galois representation V ∨ G is, after f f | Qℓ extending scalars from L to Qp and fixing a Q(f)-isomorphism C ∼= Qp, the Weil–Deligne 1/2 representation associated with π − via the local Langlands correspondence. An ℓ ⊗|·|ℓ important feature of the Galois representation Vf is that it is geometric and pure of weight 1 2k. −

2.4.3. Modular abelian varieties. Associated to a newform f S2(Γ0(N)) of weight 2 is an isogeny class of abelian varieties whose endomorphism rings∈ contain an order in the ring of integers Z(f) of Q(f). Let Af be an abelian variety in this isogeny class such that Z(f) ֒ End A (such an A always exists). The p-adic Tate module T A is a free → Q f f p f Z(f) Z -module of rank two. Let p be a prime of Z(f) containing p and let TpA = ⊗ p f TpAf Z(f) Zp Z(f)p; this is the p-adic Tate-module of Af . Let VpAf = TpAf Z(f)p Q(f)p. ⊗ ⊗ ⊗ The quotient V A /T A T A Q(f) /Z(f) is naturally identiﬁed with A [p ]; p f p f ∼= p f Z(f)p p p f ∞ n ⊗n this identiﬁes A [p ] with TpA /p TpA for each n 1. For L = Q(f)p and = Z(f)p, f f f ≥ O VpAf is just the Vf of Section 2.4.2, and (VpAf , TpAf , Af [p∞]) is an example of a triple (V,T,W ) as in Section 2.1.

2.4.4. Selmer groups of newforms and modular abelian varieties. Let f S (Γ (N) be ∈ 2k 0 a newform and Vf an associated p-adic Galois representation as in Section 2.4.2. Let T V be an G -stable -lattice and let W = V /T . f ⊂ f Q O f f f For a number field F , let L(Vf∨/F,s) be the L-function of the GF -representation Vf∨. Then L(Vf∨/Q,s) is just the usual L-function L(f,s), and more generally L(Vf∨/F,s) is the value at s 1/2 of the L-function of the formal base change to GL2/F of the automorphic representation− π of Section 2.4.2. The Bloch–Kato conjectures connect the 5 1 central value L(Vf∨/F, k) with the order of the Selmer module H (F,Wf (1 k)). FBK − 3Following our convention of using geometric normalizations, the Euler factors of a Galois representation (or a Weil-Deligne representation) are defined via the action of geometric Frobenius on inertia ∨ invariants; Euler factors at p are defined using the Weil-Deligne representation associated with Vf |GQp . 4 We have adopted the conventions here so if f is associated with an elliptic curve E/Q, then Vf is just the Qp-Tate module of E (or an extension of scalars thereof). 5 ∨ ∗ Note the Tate-twist: if V = Vf (1 − k), then L(Vf /F,s) = L(V /F,s − k), so the central critical value is just L(V ∗, 0). THEBSDFORMULAFORTHERANKONECASE 11

Suppose that f has weight 2 and (Vf , Tf ,Wf ) = (VpAf , TpAf , Af [p∞]). Then for any 1 number ﬁeld F , H (F,Wf ) is just the usual p-adic Selmer group of Af : FBK 1 (2.4.a) H (F,Wf ) = Selp(Af /F ) = Selp(Af /F ) Z(f) Zp Z(f)p. FBK ⊗ ⊗ Here Selp(Af /F ) is the usual p-adic Selmer group of Af /F and Selp(Af /F ) is the usual p-adic Selmer group of Af /F .

3. An Anticyclotomic Control Theorem

Let /Q be an imaginary quadratic field such that p splits in : K K (split) p = vv.¯ Let τ Gal( /Q) be the nontrivial automorphism. Let be the anticyclotomic ∈ K K∞ Zp-extension of and let Γ = Gal( / ). K K∞ K Let (V,T,W ), and so also L and , be as in Section 2.1. Let Λ = [[Γ]] and put O O M = T Λ, Λ = Homcont(Λ, Qp/Zp). ⊗O 1 We equip M with an action of G via ρ Ψ− where the projection Ψ: G Γ is Kb b ⊗ K → viewed as a continuous Λ×-valued character of G . Given a finite set Σ of finite places w ∤ p, let K Σ 1 Xac(M) = Hom H Σ ( , M), L/ . O Fac K O We begin this section by listing a few assumptions on the Galois representation V and the related Selmer modules that will be assumed to be in force in all that follows. Then 1 under these assumptions we relate the order of the Selmer module H ac ( ,W ) to the order of the Shafarevich–Tate group X (W/ ), defined àla Bloch–KatoF K as BK K 1 1 XBK(W/ ) = H ( ,W )/ H ( ,W )div. K FBK K FBK K 1 We then prove a control theorem providing a connection between the order of H ac ( ,W ) Σ F K and the characteristic ideal of Xac(M). The latter is linked to p-adic L-functions via the anticyclotomic main conjectures discussed in Section 6. Finally, we deduce some consequences for the Selmer groups associated with modular forms and modular abelian varieties.

3.1. A few assumptions on (V,T,W ). In addition to (geom) and (pure) we will assume that

(sst) V is semistable as a representation of G w for all w p K | and that τ (τ-dual) V ∗ ∼= V , where V τ denotes the representation with the same underlying space as V but with the G -action composed with conjugation by (a lift of) τ. This last hypothesis forces K 12 DIMITARJETCHEV,CHRISTOPHERSKINNER,ANDXINWAN the weight of V to be 1 (that is, V is pure of weight 1). To slightly simplify some arguments we will additionally− assume that −

(2-dim) dimL V = 2, that (HT) no non-zero Hodge-Tate weight of V is 0(mod p 1), ≡ − and that (irred ) V is an irreducible κ-representation of G . K K We assume furthermore that 1 1 (corank 1) H ( ,W )div = L/ and Hf ( w,W ) = L/ , w p, FBK K ∼ O K ∼ O | and 1 loc։w 1 (sur) H ( ,W )div Hf ( w,W ), w p. FBK K K |

1 X 3.2. Relating H ac ( ,W ) to BK(W/ ). We will prove the following: F K K Proposition 3.2.1. One has 1 X 2 # H ac ( ,W )=# BK(W/ ) (#δv) , F K K · where δ = coker H1 ( , T ) locv H1 ( , T )/ H1( , T ) . In particular, H1 ( ,W ) v BK f v v tor ac has ﬁnite order. { F K → K K } F K

Proof. Let = . Consider the exact sequence F FBK 1 1 1 (3.2.a) 0 H v ( ,W ) H ( ,W ) Hf ( v,W ) → F K → F K → K and the dual exact sequence 1 1 1 1 (3.2.b) 0 H ( , T ∗) H v ( , T ∗) H ( v, T )/ Hf ( v, T ∗). → F K → F K → K K By the assumption (sur), (3.2.a) is surjective on the right. It then follows from Theo- 1 1 1 rem 2.3.2 that the image of the map H v ( ,W ∗) H ( v,W ∗)/ Hf ( v,W ∗) in (3.2.b) is 0, and so F K → K K 1 1 (3.2.c) H ( ,W ∗) = H v ( ,W ∗). F K F K By (3.2.a) and (sur) we have a map of short exact sequences

/ 1 1 / 1 / 1 / 0 H v ( ,W ) H ( ,W )div H ( ,W )div Hf ( v,W ) 0 F K ∩ F K F K K = / 1 / 1 / 1 / 0 H v ( ,W ) H ( ,W ) Hf ( v,W ) 0 F K F K K THEBSDFORMULAFORTHERANKONECASE 13

1 1 1 1 As H v ( ,W ) H ( ,W )div = ker H ( ,W )div Hf ( v,W ) , by applying the snake lemmaF toK the preceding∩ F K diagram we{ concludeF K that→ K } 1 X 1 1 # H v ( ,W )=# BK(W/ ) # ker H ( ,W )div Hf ( v,W ) (3.2.d) F K K · { F K → K } =#X (W/ ) #δ . BK K · v The last equality follows upon tensoring the short exact sequence 1 1 1 1 0 H ( , T )/ H ( , T )tor Hf ( v, T )/ H ( v, T )tor δv 0 → F K F K → K K → → with L/ . O It follows from V being pure of weight diﬀerent from 0 or 1 that we have an exact sequence 1 1 α 1 1 0 H v ( ,W ) H ac ( ,W ) H ( v,W )div/ Hf ( v,W ) → F K → F K −→ K K (cf. Section 2.2.2) with dual exact sequence

1 1 β 1 1 ∗ v 0 H( ac) ( ,W ∗) H ( ,W ∗) Hf ( v,W ∗)/ H ( v,W ∗)tor. → F K → F K −→ K K We then have 1 (3.2.e) # H ac ( ,W ) = # ker(α) #im(α) = # ker(α) #coker(β), F K · · where the second equality follows from Theorem 2.3.2. It then follows from (3.2.c) that

1 1 1 coker(β) = coker H ( ,W ∗) Hf ( v,W ∗)/ H ( v,W ∗)tor . { F K → K K } τ The hypotheses (τ-dual) and (irred ) imply that W ∗ ∼= T , so conjugating by the automorphism τ identiﬁes coker(β) withK

1 1 1 coker H ( , T ) Hf ( v, T )/ H ( v, T )tor = δv. { F K → K K } Hence combining (3.2.d) and (3.2.e) yields 1 1 X 2 # H ac ( ,W ) = # ker(α) #cokerβ = # H v ( ,W ) #δv =# BK(W/ ) (#δv) . F K · F K · K · 1 It follows from (corank 1) and (sur) that δv has finite order, whence H ac ( ,W ) has finite order. F K Remark 3.2.2. (a) Note that neither of the assumptions (sst) and (HT) is used in this 1 proof. (b) Clearly, (irred ) is not essential to the proof of the finiteness of H ac ( ,W ). K τ F K It is only used to ensure that W ∗ = T , so that δv can be identified with the cokernel 1 1 ∼ 1 coker H ( ,W ∗) Hf ( v,W ∗)/ H ( v,W ∗)tor . Under certain circumstances, such { F K → K τ K } as if T = TpAf , we have W ∗ = T without assuming (irred ). ∼ K

3.3. The anticyclotomic control theorem. Let S be a ﬁnite set of places of including all those at which V is ramiﬁed and let S S be the subset of those not dividingK p ⊂ p. Let Σ Sp. Fix a topological generator γ Γ. We identify [[T ]] with Λ = [[Γ]] via the continuous⊂ -algebra map sending 1 + T ∈ γ. We will proveO the followingO theorem: O 7→ 14 DIMITARJETCHEV,CHRISTOPHERSKINNER,ANDXINWAN

Σ Theorem 3.3.1 (Anticyclotomic Control Theorem). The Λ-module Xac(M) is Λ-torsion, Σ Σ and if fac(T ) is a generator of its characteristic Λ-ideal char(Xac(M)), then Σ 1 Σ # /fac(0) = # H ac ( ,W ) C (W ), O F K · where CΣ(W )=#H0( ,W ) # H0( ,W ) # H1 ( ,W ) # H1( ,W ). Kv · Kv · ur Kw · Kw w Sp Σ w Σ w∈Ysplit\ Y∈

Our proof of Theorem 3.3.1 follows the arguments of Greenberg in [Gre99, 4]. § 3.3.2. Surjectivity of the localization maps. For a ﬁnite set S of ﬁnite places of let Let ′ K 1 1 H ( w, M) H ( w,W ) ac (M; S′)= K and ac (W ; S′)= K . PF H1 ( , M) PF H1 ( ,W ) w S′ ac w w S′ ac w Y∈ F K Y∈ F K 1 Γ 1 S The key to our result relating H S ( , M) to H ( / ,W ) is understanding the ac K∞ K K images of F

1 S locS (3.3.a) H ( / , M) ac (M; S). K K −→ PF and

1 S locS (3.3.b) H ( / ,W ) ac (W ; S). K K −→ PF Here S/ is the maximal extension unramiﬁed at all ﬁnite places not in S. K K Proposition 3.3.3. The restriction maps (3.3.a) and (3.3.b) are surjective.

Proof. Let M ∗ = M ∨(1), which is just T Λ with G -action given by ρf Ψ. Recall that ⊗O K ⊗ 1 1 H S ( , M ∗)= c H S ( , M ∗) : loc c = 0 . ( )v¯ v¯ Fac K { ∈ Fac K } By Theorem 2.3.2 the dual of the cokernel of (3.3.a) is identified with a quotient of 1 H S ( , M ∗). Therefore, to prove the desired surjectivity of (3.3.a), it suffices to show ( ac)v¯ F 1 K that H S ( , M ) = 0. ( )v¯ ∗ Fac K Note that M /(γ 1)M ∼ T . We claim that the natural injection ∗ − ∗ → 1 S 1 S 1 S ( H ( / , M ∗)/(γ 1) H ( / , M ∗) ֒ H ( / , T K K − K K → K K induces an injection 1 S 1 S 1 .( H S ∗ ( / , M ∗)/(γ 1) H S ∗ ( / , M ∗) ֒ H S ( , T ( ) ( ) ( )v¯ Fac K K − Fac K K → Fac K 1 S 1 For this, suppose c H S ∗ ( / , M ∗) has trivial image in H S ( , T ). Then c = ( ac) ( ac)v¯ ∈ F 1 SK K F K1 (γ 1)d for some d H ( / , M ∗) such that (γ 1)d = 0 in H ( v¯, M ∗). But − ∈ K K 1 − K0 the kernel of multiplication by γ 1 on H ( v¯, M ∗) is the image of H ( v¯, T ). The vanishing of the latter follows from−H0( ,V )K = 0 (which is true as V is pureK of weight Kv¯ THEBSDFORMULAFORTHERANKONECASE 15 different from 0 or 1: V GK = 0 would imply that 1 was an eigenvalue of Frobenius on N=0 6 W Dv¯(V ) , which would contradict purity if the weight is not 0 or 1). Next we note that the canonical isomorphism T L/ = W induces an injection ⊗O O ∼ 1 1 .( H S ( , T ) L/ ֒ H S ( ,W ( )v¯ ( )v¯ Fac K ⊗O O → Fac K It follows from Proposition 3.2.1 that the right-hand side is finite (reversing the roles of 1 v andv ¯ and using that H (Kw,W ) is finite for w ∤ p (cf. Section 2.2.2). As (irred ) 1 1 K implies H S ( , T ) is torsion-free, it follows from the finiteness of H S ( ,W ) that ( ac)v¯ ( ac)v¯ 1 F K F K H S ( , T ) = 0. Hence ( )v¯ Fac K 1 1 H S ( , M )/(γ 1) H S ( , M ) = 0, ( )v¯ ∗ ( )v¯ ∗ Fac K − Fac K 1 and so, by Nakayama’s lemma, H S ( , M ∗) = 0. This completes the proof of the ( ac)v¯ K surjectivity of (3.3.a). The proof ofF the surjectivity of (3.3.b) is similar: Poitou–Tate 1 duality identifies the cokernel with a quotient of H S ( , T ), which we have already ( ac)v¯ K seen to be 0. F

1 S 1 3.3.4. Trivial coinvariants. We now show that H ( / , M)Γ and H Σ ( , M)Γ both K K ac K vanish: F 1 S 1 Lemma 3.3.5. We have H ( / , M)Γ = 0 and H Σ ( , M)Γ = 0. K K Fac K Proof. The long exact sequence on Galois cohomology associated to the short exact se- γ 1 1 S 2 S .( quence 0 W M − M 0 yields an injection H ( / , M)Γ ֒ H ( / ,W → → −−→ →2 0 K K → K K The local cohomology group H ( w,W ) is dual to H ( w, T ), and the latter is 0; for w p this was explained in the proofK of Proposition 3.3.3,K and for w ∤ p this was explained in| Section 2.2.2. Consequently (using the notation of [Mil86, 4]), § H2( S/ ,W )= X2 ( ,W )= c H2( S / ,W ): loc c = 0 w S . K K S K { ∈ K K w ∀ ∈ } By Poitou-Tate duality [Mil86, Thm.4.10(a)], this group is dual to X1 ( , T )= c H1( S / , T ): loc c = 0 w S , S K { ∈ K K w ∀ ∈ } X1 and the latter is trivial. Indeed, (irred ) implies that S( , T ) is torsion-free while K K (corank 1) and (sur) imply that X1 ( , T ) is torsion and thus, H1( S / , M) = 0. S K K K Γ 1 To show that H Σ ( , M)Γ = 0, consider the exact sequence Fac K 1 1 S 0 H Σ ( , M) H ( / , M) ac (M; S Σ) 0. → Fac K → K K → PF \ → The exactness on the right is a consequence of Proposition 3.3.3. Multiplying by γ 1, we obtain from the snake lemma the exact sequence − 1 S 1 S Γ Γ 1 1 S H ( / ,W ) = H ( / , M) ac (M; S Σ) H Σ ( , M)Γ H ( / , M)Γ. K K K K → PF \ → Fac K → K K 1 S Γ The map H ( / ,W ) ac (M; S Σ) is the composite of the restriction map 1 S K K → PF \ H ( / ,W ) ac (W ; S Σ), which is surjective by Proposition 3.3.3, and the map K K → PF \ Γ 1 ac (W ; S Σ) ac (M; S Σ) , which is also surjective (as the maps H ( w,W ) PF \ → PF \ K → 16 DIMITARJETCHEV,CHRISTOPHERSKINNER,ANDXINWAN

1 Γ 1 1 S H ( w, M) are surjective). It follows that H Σ ( , M)Γ injects into H ( / , M)Γ. K ac K K K But we have already shown the latter to be trivial.F

3.3.6. Computing # ker(r). We now calculate the order of the kernel of the map Γ r : ac (W ; S Σ) ac (M; S Σ) . PF \ → PF \ Proposition 3.3.7. The kernel of r has order (3.3.c) # ker(r)=#H0( ,W ) # H0( ,W ) c(p)(W ), Kv · Kv · w w Sp Σ w∈Ysplit\ (p) 1 1 1 where cw (W ) := [Hur( w,W ) : Hf (Kw,W )] = # Hur( w,W ) are the p-parts of the local Tamagawa numbers. K K

Proof. Let w be a place of . By the long exact sequence on Galois cohomology associated to the short sequence K (γ 1) 0 W M × − M 0, → → −−−−−→ → the kernel of the restriction map r : H1( ,W ) H1( , M)Γ is the image of w Kw → Kw M GKw /(γ 1)M GKw under the coboundary map. Let ℓ be the prime below w. Un- like the cyclotomic− case over Q treated in [Gre99, 3] where every prime w is ﬁnitely § decomposed in the Zp-extension, we need to consider the cases of ℓ being split or non-split in separately. K Case 1(a): w ∤ p, W is ramiﬁed at w and ℓ is split in . We have a commutative diagram K / 1 / 1 / 1 / 0 / H ( ,W ) / H ( ,W ) / H (I ,W )GKw / 0 ur Kw Kw w

0 / H1 ( , M) / H1( , M) / H1(I , M) ur Kw Kw w The kernel of the right-most map is just the image of (M Iw /(γ 1)M Iw )GKw . Since − Ψ is not ramiﬁed at w, M Iw is (γ 1)-divisible, hence this kernel is trivial. It follows 1 1 − 1 1 that the map H ( w,W )/ Hur( w,W ) H ( w, M)/ Hur( w, M) is injective and hence K K 1 → K1 K 1 1 that the kernel of the map rw : H ( w,W )/ Hf ( w,W ) H ( w, M)/ Hur( w, M), is 1 1 K K → K K isomorphic to Hur( w,W )/ Hf ( w,W ). But the order of the latter is exactly the p-part (p) K K cw (W ) of the Tamagawa number at w.

Case 1(b): w ∤ p, W is ramified at w and ℓ is not split in . In this case ℓ is inert or ramified in , and by the definition of the anticyclotomicK Selmer structure 1 K 1 1 ։ both H ac (Kw,W ) and H ac (Kw, M) are trivial. Hence ker(rw) = ker H ( w,W ) 1 F F G G { K1 .( H ( w, M)[γ 1]) and so equals the image of M Kw /(γ 1)M Kw ֒ H ( w,W K − } − GKw → K However, since w is not split, Ψ is trivial on G w and hence, M is divisible by γ 1. K − It follows that ker(rw) = 0. THEBSDFORMULAFORTHERANKONECASE 17

Case 2(a): w ∤ p, W is unramiﬁed at w and ℓ is split in . Similarly to Case 1 1 K 1(a), we obtain ker(rw) ∼= Hur(Kw,W )/ Hf (Kw,W ) which is trivial since the two local conditions coincide.

Case 2(b): w ∤ p, W is unramiﬁed at w and ℓ is not split in . Exactly the same K argument as employed for Case 1(b) shows that ker(rw) = 0.

1 1 Case 3(a): w =v ¯. We have ker(rv) = H ( v,W )/ H ( v,W )div. By Tate local 1 1 1 K K 1 duality H ( v¯,W )/ H ( v¯,W )div is dual to H ( v¯, T )tor, which is just ker H ( v¯, T ) 1 K 0 K 0 K0 { K → H ( v¯,V ) ∼= H ( v¯,W )/ H ( v¯,W )div. But H ( v¯,V ) = 0 (as noted in the proof of PropositionK } 3.3.3,K this is a consequenceK of being pureK of weight diﬀerent from 0 or 1), so H0( ,W ) = 0 and # ker(r )=#H0( ,W ). Kv¯ div v¯ Kv¯ Case 3(b): w = v. In this case the map is r : H1( ,W ) H1( , M)Γ and we have v Kv → Kv .( ker(r ) = M GKv /(γ 1)M GKv ֒ H1( ,W v ∼ − → Kv

-Let Pv = ker Ψ G and Γv = G v /Pv ֒ Γ, where Ψ is as in the beginning of Sec | Kv K → tion 3. Then Γv has finite index in Γ, and the image of Iv in Γv also has finite index. Let γ Γ be a topological generator. Let T = Hom (T, Z ) and let v ∈ v ∨ Zp p TP∨ be its Pv-coinvariants. Then M = T Λ = Homcont(T ∨ Λ, Qp/Zp) and so v ⊗O ∼ ⊗O Pv GKv M = Homcont(TP∨ Λ, Qp/Zp). If#TP∨v is finite, then M is therefore isomorphic ∼ v O ⊗ pt b to a submodule of HomZp (TP∨ Λ/(γv 1), Qp/Zp), which has finite order. Here t> 0 v O pt ⊗ − GKv is such that γv acts trivially on TP∨v . This shows that M has finite order if #TP∨v is finite, in which case

#M GKv /(γ 1)M GKv =#M GKv [γ 1]=#M[γ 1]GKv = # H0( ,W ). − − − Kv

It remains to show that #TP∨v is finite, which we will do by arguing by contradiction. Pv Pv Assume that #T ∨ is not finite. Then T = 0 and hence V = 0. As V is two- Pv 6 6 dimensional and semistable, there are two cases to consider: (i) V is crystalline and (ii) V is non-crystalline and hence a non-split extension of the form 0 L(ǫα) → → V L(α) 0 with α unramified. In case (i), V Pv would have to be, possibly after a finite→ extension→ of scalars, a sum of one-dimensional crystalline representations of weight 1. Such a crystalline character must be of the form ǫaα for some integer a and some − unramified character α, while the condition of being weight 1 means that α(Frobv) is 1/2+a a− a Weil number of absolute value p− . However, since ǫ α factors through a pro-p- a a group, so must ǫ Iv = ǫ α Iv . But this only happens if a 0 (mod p 1). As a is a Hodge-Tate weight| of V , it| follows from (HT) that we must≡ have a = 0.− Then α must factor through the quotient of Γv by the image of Iv. This quotient is finite, so α must have finite order, contradicting it being a character of weight 1. On the other hand, if V is as in case (ii), then V is also a non-split extension of−P and so if V Pv = 0, v 6 then again ǫα, α unramified, must factor through Γv, which we have already seen to be impossible. This contradiction completes the last case and hence proves the lemma. 18 DIMITARJETCHEV,CHRISTOPHERSKINNER,ANDXINWAN

Remark 3.3.8. The proof of Case 3(b) above is the only place where the hypothesis (HT) is invoked.

3.3.9. An application of the snake lemma. There is a commutative diagram

/ 1 / 1 S / / 0 H Σ ( ,W ) H ( / ,W ) ac (W ; S Σ) 0 Fac K K K PF \ s h r / 1 Γ / 1 S Γ / Γ 0 H Σ ( , M) H ( / , M) ac (M; S Σ) Fac K K K PF \ (note that exactness of the top row follows from Proposition 3.3.3), and the snake lemma yields an exact sequence 0 ker(s) ker(h) ker(r) coker(s) coker(h) coker(r). → → → → → → However, coker(h) = 0 and hence there is an exact sequence 0 ker(s) ker(h) ker(r) coker(s) 0. → → → → → It follows that 1 Γ # H Σ ( , M) #coker(s) # ker(r) Fac K 1 = = . # H Σ ( ,W ) # ker(s) # ker(h) Fac K The order of ker(r) was computed in Proposition 3.3.7 while ker(h)= M GK /(γ 1)M GK = M GK = H0( ,W ) − K which vanishes by (irred ). It follows that K 1 Γ 1 0 0 (p) (3.3.d) # H Σ ( , M) = # H Σ ( ,W ) # H ( v,W ) # H ( v,W ) cw (W ). Fac K Fac K · K · K · w Sp Σ w∈Ysplit\

It also follows from the surjectivity of (3.3.b) that there is an exact sequence 1 1 0 H ac ( ,W ) H Σ ( ,W ) ac (W ; Σ) 0. → F K → Fac K → PF → As H1 ( ,W ) = 0 (see Section 2.2.2), we have f Kw 1 1 1 # H Σ ( ,W )=#H ac ( ,W ) # H ( w,W ). Fac K F K · K w Σ Y∈ Combining this with (3.3.d) yields 1 Γ 1 Σ (3.3.e) # H Σ ( , M) = # H ac ( ,W ) C (W ). Fac K F K ·

Σ 3.3.10. Proof of Theorem 3.3.1: Torsionness of Xac(M). It follows from Proposition 1 Γ Σ 1 Γ 3.2.1 and (3.3.e) that H Σ ( , M) has ﬁnite order. Hence Xac(M)Γ = Hom (H Σ ( , M) , L/ ) ac O ac F K Σ F K O has ﬁnite order. It follows easily that Xac(M) must therefore be Λ-torsion. THEBSDFORMULAFORTHERANKONECASE 19

Σ 3.3.11. Proof of Theorem 3.3.1: determining # /fac(0). To complete the proof of the O1 1 Γ control theorem, we use the relation between H ( ,W ) and H Σ ( , M) to relate ac ac 1 F K F K # H ac ( ,W ) to fac(0). The key to the comparison is the following proposition on the F K Σ non-existence of pseudo-null submodules of the Λ-module Xac(M): 1 Proposition 3.3.12. The Selmer module H Σ ( , M) has no proper Λ-submodule of ac ΣF K ﬁnite order. Equivalently, the Λ-module Xac(M) has no non-trivial pseudo-null Λ- submodule.

1 Σ Proof. Recall that H Σ ( , M)Γ by Lemma 3.3.5. If X X (M) is a Λ-submodule of ac ac F K ⊂ 1 finite order, then its dual X∗ is a finite order quotient of H Σ ( , M), so X∗/(γ 1)X∗ ac 1 F K − is a quotient of H Σ ( , M)Γ and therefore 0. But X∗ is a finite Λ-module and so, by Fac K Nakayama’s Lemma, X∗ = 0 and hence X = 0.

We can now establish the comparison result: Lemma 3.3.13. We have

Σ Σ 1 Σ # /fac(0) = #Λ/(T,fac(T )) = # H ac ( ,W ) C (W ). O F K ·

Σ 1 Γ Proof. Let X = X (M). Then #XΓ = # H Σ ( , M) . Since X is Λ-torsion, X is ac ac r F K pseudoisomorphic to a Λ-module Y = Λ/(fi) for distinguished polynomials fi. By i=1 Proposition 3.3.12, X has no pseudo-nullY submodule so we obtain an exact sequence 0 X Y K 0, where K is a pseudo-null Λ-module. Applying the snake lemma to→ the commutative→ → → diagram

0 / X / Y / K / 0

γ 1 γ 1 γ 1 − − − 0 / X / Y / K / 0, we obtain #X =#Y = #Λ/(T,f (T )) = # /(f (0)) = #Λ/(T,f Σ (T )), Γ Γ i O i ac Σ Y Y where fac(T ) f1(T ) fr(T ) is a generator of the characteristic ideal of X. We thus get that ∼ · · · Σ 1 1 Σ # /fac(0) = # H Σ ( ,W )=#H ac ( ,W ) C (W ), O Fac K F K · which proves the claim.

This completes the proof of Theorem 3.3.1. 20 DIMITARJETCHEV,CHRISTOPHERSKINNER,ANDXINWAN

2 3.4. Comparison with Selmer groups over Zp-extensions. In order to deduce what we will need from the existing theorems in Iwasawa theory, it is necessary to consider Selmer groups for Galois extensions larger than / . Let K / be the composite of K∞ K 2 ∞ K all Zp-extensions of in Q. Let Γ = Gal(K / ) ∼= Zp. The Galois group Gal( /Q) K K ∞ K + K acts on Γ , which decomposes under this action as Γ =Γ Γ− with the superscript K K ⊕ denoting the subgroup on which τ Gal( /Q) acts at K1; thenK Γ+ is the kernel of the± natural surjection Γ ։ Γ. Let Λ∈ = K[[Γ ]]. Let Ψ±: G ։ ΓK be the natural K K O K K K K surjection; this is also a continuous Λ×-valued character of G . Let = T Λ∗ 1 K K 1 M ⊗O K with G -action given by ρ Ψ− . Consider the Selmer module H Σ ( , ) and its K ⊗ K Gr K M Σ 1 F Pontrjagin dual XGr( ) = Hom (H Σ ( , ), L/ ) which is a finite Λ -module. Let M O FGr K M O K Σ Σ char XGr( ) Λ be its Λ -characteristic ideal. Since Λ is a UFD, char XGr( ) M ⊂ K K Σ K M is a principal ideal, and it is nonzero if and only if XGr( ) is a torsion Λ -module. M K Let γ Γ± = Zp be fixed topological generators; we assume that γ γ under the ± ∈ K ∼ − 7→ projection Γ ։ Γ. We have M = [γ+ 1] and, by (irred ), K M − K H1( S / , M) ∼ H1( S/ , )[γ 1], K K → K K M + − which induces maps 1 1 Σ Σ Σ .(a) H Σ ( , M) ֒ H Σ ( , )[γ+ 1] and XGr( )/(γ+ 1)XGr( ) Xac(M.3.4) Fac K → FGr K M − M − M → Lemma 3.4.1. Suppose Σ contains all the finite places w ∤ p at which V is ramified. Then the maps in (3.4.a) have finite cokernel and kernel, respectively.

Proof. The identification M = [γ 1] yields a short exact sequence M + − .[GKv /(γ 1) GKv ֒ H1( , M) ։ H1( , )[γ 1 M + − M → Kv Kv M + − Since the image of G v¯ in Γ has finite index, the argument used in the Case 3(b) of K K the proof of Proposition 3.3.7 to show that M GKv has finite order can be easily adapted to prove that GKv has finite order by replacing P with ker Ψ . It follows that M v K|GKv # GKv /(γ 1) GKv =# GKv [γ 1] < . M + − M M + − ∞ 1 1 Since the cokernel of the map H Σ ( , M) ֒ H Σ ( , )[γ+ 1] is a quotient of Fac K → FGr K M − GKv /(γ 1) GKv , it is therefore finite, and so, too, is the kernel of the dual map. M + − M Corollary 3.4.2. Suppose Σ contains all the finite place w ∤ p at which V is ramified. Then char XΣ (M) char XΣ ( ) mod (γ 1). ac ⊂ Gr M + − Proof. Let F Σ ( ) be the Λ -fitting ideal of XΣ ( ) and F Σ(M) the Λ-fitting ideal Gr M K Gr M ac of XΣ (M). Then F Σ ( ) char XΣ ( ) . Since the kernel of the surjection map ac Gr M ⊂ Gr M XΣ ( )/(γ 1)XΣ ( ) ։ XΣ (M) has finite order and since the source has Λ-Fitting Gr M + − Gr M ac ideal equal to F Σ ( ) modulo (γ 1), there is some c > 0 such that F Σ(M)mc Gr M + − ac Λ ⊂ F Σ ( ) mod (γ 1), where m = (γ 1, m) is the maximal ideal of Λ. It follows that Gr M + − Λ − F Σ(M)mc char XΣ ( ) mod (γ 1). ac Λ ⊂ Gr M + − THEBSDFORMULAFORTHERANKONECASE 21

Σ Since the right-hand side is a principal ideal, it follows that Fac(M) is contained in char XΣ ( ) mod (γ 1). And since char XΣ (M) is the smallest principal ideal Gr M + − ac containing F Σ(M), we must also have that char XΣ (M) is contained in char XΣ ( ) ac ac Gr mod (γ 1). M + −

3.5. Applications to newforms and modular abelian varieties Af . We return to the notation of Section 2.4. Let f S2k(Γ0(N)) be a newform, and let (Vf , Tf ,Wf ) be as in Section 2.4.2. ∈ Let (V,T,W ) = (V (1 k), T (1 k),W (1 k)). f − f − f − Note that V ∗ ∼= V . It is a theorem of Saito [Sai97], building on work of Deligne, Langlands, Carayol, and others, that V is geometric and pure of weight 1. So (geom), (pure), (τ-dual), and (2-dim) all hold for V . Furthermore, it follows that− if (p-sst) ord (N) 1, p ≤ then V is semistable at p, that is, (sst) also holds. As the Hodge-Tate weights of V are k 1 and k, (HT) holds if and only if neither k nor k 1 (when k = 1) are divisible by−p 1; in− particular, (HT) always holds for 2k = 2. − 6 − Suppose now that f has weight 2 and that

(V,T,W ) = (Vf , Tf ,Wf ) = (VpAf , TpAf , Af [p∞]) as in Section 2.4.3. The Mordell–Weil group Af (F ) is a Z(f)-module and

rankZAf (F ) = [Q(f) : Q] rank (Af (F ) Z(f) ). · O ⊗ O Suppose

(rank 1) rank (Af ( ) Z(f) ) = 1 O K ⊗ O or, equivalently, rankZAf ( ) = [Q(f) : Q]. The Tate-Shafarevich group X(Af / ) is also a Z(f)-module and satisfiesK X(A / )[p ] = X(A / )[p ]. SupposeK f K ∞ ⊗Z(f) O f K ∞ (X p-finite) #X(A / )[p∞] < . f K ∞ Note that is this case XBK (Wf / ) = X(Af / )[p∞]. Under the assumptions (rank 1) and (X p-finite), it follows from theK exact sequenceK

,[∞FES) A (F ) L/ ֒ Selp∞ (A /F ) ։ X(A /F )[p) f ⊗Z(f) O → f f 1 that (corank 1) holds for H ( ,Wf ) = Selp∞ (Af /F ). If furthermore FBK K (p-irred) Af [p] is an irreducible G -representation, K then A ( ) ∼ H1 ( , T ) = . f K ⊗Z(f) O → f K ∼ O If w p is a place of then A ( ) is a finitely generated Z -module of rank equal | K f Kw p to [ w : Qp] [Z(f) : Z] (there is a natural Z(f)-injection of Z(f) w into the K · ⊗ OK compact Lie group Af ( w), with image having finite index; this is just the exp map from a neighborhood ofK zero in the tangent space at the origin). In particular, if p 22 DIMITARJETCHEV,CHRISTOPHERSKINNER,ANDXINWAN splits in (so Q ∼ ), we have that A ( ) L/ = L/ . Furthermore, since K p → Kw f Kw ⊗Z(f) O ∼ O A ( ) A ( ), if (rank 1) holds, then we also have that f K ⊂ f Kw A( ) L/ = L/ ։ A( ) L/ = L/ . K ⊗Z(f) O ∼ O Kw ⊗Z(f) O ∼ O That is, (sur) also holds. This shows that if (split), (rank 1), (X p-finite), and (p-irred) hold, then so do (corank 1), (sur), and (irred ). K ∼ 1 Still assuming that p splits in , we also have Af ( v) Z(f) Zp Hf ( v, T ) is a K K ⊗ ⊗ O → K finite -module of rank one. Let A ( ) = A ( )/A (K ) (this is then a free O f Kv /tor f Kv f v tor ∼ 1 1 Z(f) Zp-module), so Af ( v)/tor Z(f) Zp Hf ( v, T )/ H ( v, T )tor. Then ⊗ K ⊗ ⊗ O → K K #δ = #coker H1 ( , T ) locv H1 ( , T )/ H1( , T ) (3.5.a) v { f K → f Kv Kv tor} = #[Af ( v)/tor Z(f) Zp : Af ( ) Z(f) ]. K ⊗ ⊗ O K ⊗ O Let P Af ( ) be any point of infinite order. The -module P Af ( ) Z(f) generated∈ byKP has finite index, and it follows that O O · ⊂ K ⊗ O

[Af ( v)/tor Z(f) Zp : P ] #δ = K ⊗ ⊗ O O · . v [A ( ) : P ] f K ⊗Z(f) O O · In particular, it then follows from Proposition 3.2.1 that

2 [Af ( v)/tor Z(f) Zp : P ] 1 X K ⊗ ⊗ O O · (3.5.b) # H ac ( ,Wf )=# BK(Wf / ) . F K K · [A ( ) : P ]2 f K ⊗Z(f) O O · Suppose now that (good) p ∤ N.

Then Af has good reduction at p and so extends to an abelian scheme over Z(p). Let 1 1 1 Af (Qp) Af (Qp) be the kernel of reduction modulo p. LetΩ (Af /Zp)∨ = HomZp (Ω (Af /Zp), Zp). ⊂ 1 1 The formal group logarithm deﬁnes a Z(f) Zp-isomorphism log : Af (Qp) pΩ (Af /Zp)∨, ⊗ → 1 .∨(which extends to an injective Z(f) Zp-homomorphism log : Af (Qp)/tor ֒ Ω (Af /Zp ⊗ 1 → Recall that = Z(f)p for a chosen prime p p of Z(f). Let ωf Ω (Af /Zp) Zp be an -basis elementO such that the action of each| α Z(f) on A ∈induces α ω ⊗= αO ω O ∈ f ∗ f · f (multiplication by the scalar α ). Then composition of log with evaluation on ωf deﬁnes an -homomorphism ∈ O O log : A (Q ) ωf f p /tor ⊗Zp O → O that maps A1(Q ) surjectively onto p . By the choice of ω , the map log factors p ⊗Zp O O f ωf ֒ through Af (Qp)/tor Z(f) Zp . The induced homomorphism Af (Qp)/tor Z(f) Zp ⊗ ⊗1 O ⊗ ⊗ O → O is injective and maps Af (Qp) Z(f) Zp isomorphically onto p ; this follows from 1 ⊗ ⊗ O O Af (Qp)/tor Z(f) Zp and Af (Qp) Z(f) Zp both being free -modules of rank one ⊗ ⊗ O ⊗ ⊗ O O and the surjective mapping of the latter onto p . As = Q (since p splits in ) it O Kv p K THEBSDFORMULAFORTHERANKONECASE 23 easily follows from this that # /(log P ) O ωf [Af ( v)/tor Z(f) Zp : P ]= K ⊗ ⊗ O O · # /(log (A (Q ) )) O ωf f p /tor ⊗Zp O 1 # /(logω P ) #(Af (Qp)/tor/Af (Qp) Z(f) Zp ) = O f · ⊗ ⊗ O # /p O O1 # /(logω P ) #(Af (Qp)/A (Qp) Z(f) ) = O f · f ⊗ O . # /p #(A (Q ) ) O O · f p tor ⊗Z(f) O Since reduction modulo p yields an isomorphism A (Q )/A1 (Q ) ∼ A (F ), we have f p f p → f p A (Q )/A1 (Q ) ∼ A (F ) = A [p ](F ). The latter group is trivial f p f p ⊗Z(f) O → f p ⊗Z(f) O f ∞ p unless f is ordinary with respect to p (that is, p ∤ ap) in which case it is isomorphic to /(1 a + p). Hence O − p #A (Q )/A1 (Q ) =# /(1 a + p). f p f p ⊗Z(f) O O − p 0 0 Also, Af (Qp)tor Zp = H (Qp, Af [p∞]) so Af (Qp) Z(f) Zp = H (Qp,Wf ), hence ⊗ ⊗ ⊗ O #A (Q ) Z = # H0(Q ,W ), f p tor ⊗ p p f

Putting this together with the preceding formula for [Af ( v)/tor Z(f) Zp : P ] we ﬁnd K ⊗ ⊗ O O · # /(log P ) # /(1 a + p) O ωf · O − p [Af ( v)/tor Z(f) Zp : P ]= K ⊗ ⊗ O O · # /p # H0( ,W ) O O · Kv f 1 ap+p # /(( − ) logω P ) = O p f . #H0( ,W ) Kv f Combining this last equality with (3.5.a) we get

1 ap+p − # /(( p ) logωf P ) #δv = O , [A ( ) : P ] # H0( ,W ) f K ⊗Z(f) O O · · Kv f which, when substituted into (3.5.b), yields

1 ap+p 2 # /(( − ) log P ) 1 X O p ωf (3.5.c) # H ac ( ,Wf )=# BK(Wf / ) 0 F K K · [A ( ) : P ] # H ( ,W ) f K ⊗Z(f) O O · · Kv f !

In the special case that Af = E is an elliptic curve (i.e., f has rational coeﬃcients), p is a prime of good reduction, = Z , and P E( ) has inﬁnite order, we can rewrite O p ∈ K the formula for #δv as

1 ap(E)+p − #Zp/(( p ) logωE P ) #δv = , [E( ) : Z P ] # H0( , E[p ]) K · p · Kv ∞ 24 DIMITARJETCHEV,CHRISTOPHERSKINNER,ANDXINWAN where [ , ] denotes the p-part of the index and we have take for ω the N´eron diﬀer- − − p f ential ω Ω1(E/Z ). In this case (3.5.c) then becomes E ∈ (p) 2 1 ap(E)+p #Z /(( − ) log P ) 1 X p p ωE (3.5.d) # H ac ( , E[p∞])=# (E/ )[p∞] 0 . F K K · [E( ) : Z P ] # H ( , E[p ]) K · p · Kv ∞   4. CM Points, CM Periods and Upper Bounds on X

In this section we recall the definition of Heegner points on certain Shimura curves and their Jacobians. The Shimura curves considered are moduli spaces for false elliptic curves and the Heegner points correspond to false elliptic curves with complex multiplication. These Heegner points give rise to cohomology classes for the Tate module of an optimal quotient of the Jacobian of the Shimura curve. The Euler system method of Kolyvagin yields an upper bound on the order of the Tate–Shafarevich group of this optimal quotient in terms of the indices of certain of these classes. This upper bound is recalled in Section [Jet08]. It is an essential ingrediant in our proof of the main theorem of this paper. For use in comparing p-adic L-functions in Section 5, we also explain how the complex and p-adic periods of false CM elliptic curves are identified with complex and p-adic periods of (true) CM elliptic curves. Let be an imaginary quadratic field such that (split) holds. Let D < 0 be the discriminantK of . − K K

4.1. The Heegner hypothesis. Let N be an integer. We will also assume that N satisfies at least one of the two Heegner-type hypotheses recalled below. The first of these is: + + N = N N − with (N ,N −) = 1; • ℓ N + if and only if ℓ splits in ; (H) • | K ℓ N − if and only if ℓ is inert in ; • | K N − is squarefree with an even number of prime factors. • Implicit in the second of these assumptions is that the discriminant D < 0 of satisfies − K K (coprime) (N, D ) = 1. K If N − = 1, then (H) is just the usual Heegner hypothesis. For some of the arguments that follow, we also need a more general Heegner-type hypothesis: + + N = N N − with (N ,N −) = 1; • ℓ N + if and only if ℓ is split or ramified in ; (gen-H) • | K ℓ N − if and only if ℓ is inert in ; • | K N − is squarefree with an even number of prime factors. • THEBSDFORMULAFORTHERANKONECASE 25

Note that the only difference between the hypotheses (H) and (gen-H) is in the latter the primes dividing N + are allowed to be ramified in (so (coprime) may not hold). K Let f S (Γ (N) be a newform and π the associated cuspidal representation of ∈ 2k 0 GL2(A) as in Section 2.4.2. The third and forth assumption of either Heegner-type hypothesis implies that the epsilon factor ǫ(π, ,s) of the base change of π to GL2(A ) satisfies K K (sign 1) ǫ(π, , 1/2) = 1. − K −

4.2. The quaternion algebra B and the Shimura curve XN +,N − . Suppose N satisﬁes (gen-H). Let B be the indeﬁnite quaternion algebra of discriminant N −. Let + B be a maximal order of B and let R B be an Eichler order of level N . Fix an isomorphismO of R-algebras ⊂ O

(4.2.a) B R ∼ M (R) ⊗ → 2 N − and an isomorphism of A∞ -algebras N − N (4.2.b) B A∞ ∼ M (A −) ⊗ → 2 f N − a b N − + that identifies R Z with the order c d M2(Z ) : N c . Here, as usual, N − N −⊗ { ∈ | } A∞ (resp. Z ) denotes the restricted product over Qℓ (resp. Zℓ), ℓ ∤ N −. b b In order to compare later statements with results in [Bro14], we assume (without loss of b generality) that the isomorphism (4.2.a) and the fixed isomorphisms R Z = M (Z ) for ⊗ ℓ ∼ 2 ℓ ℓ N +p all arise from the choice of a real quadratic field M = Q(√p ) B, p ∤ pND, | 0 ⊂ 0 in which the primes ℓ pN + split and from an identification | ι : B M ∼ M (M). M ⊗ → 2 ֒ In particular, the chosen isomorphisms are induced from ιM by fixing an inclusion M R and a prime of M above each ℓ pN +. Furthermore, we fix an idempotent e → 1 | ∈ B M [ ] as in [Bro14, p.7] such that O ⊗ O 2p0 1 0 ιM (e) = ( 0 0 ) .

Let G be the algebraic group over Q such that G(S) = (B S)× for each Q-algebra S. Using the identiﬁcation (4.2.a) we deﬁne a homomorphism,⊗ x y h0 : ResC/R(Gm) G/R, x + iy y x , → 7→ − and let X be the G(R)-conjugacy class of h0. The set X has a natural complex structure and the map X ∼ h := C R, Ad(g)h g(i), is a G(R)-equivariant holomorphic → ± − 0 7→ isomorphism. The action of G(R) on h± is via (4.2.a) and the usual action of GL2(R).

Let XN +,N − be the Shimura curve associated with the Shimura datum (G, X) and the open compact subgroup K = (R Z)× B× = (B A∞)× = G(A∞). This curve has a canonical model over Q with complex⊗ ⊂ uniformization⊗ b b (4.2.c) X + − (C)= B× (X B×/K) N ,N \ × b 26 DIMITARJETCHEV,CHRISTOPHERSKINNER,ANDXINWAN

It even has the structure of a coarse moduli space [KM85], [Buz97], [Hel07]; the solution of the moduli problem yields a smooth regular model over Z[1/N −] which, in particular, is smooth at p if (good) p ∤ N holds.

To describe the moduli problem represented by XN +,N − we recall that a false elliptic curve over a scheme S is a pair (A, ι) with A an abelian surface over S and (ι : ֒ End (A) an injective homomorphism. A full level-N + structure on (A, i OB → S is an isomorphism of group schemes t: A[N +] ∼ (Z/NZ) commuting with the → OB ⊗ S B-action on both sides, and a K-level structure is a K-equivalence class of such full O + 1 level N structures. Then XN +,N − represents the course moduli scheme over Z[ DN ] for the moduli problem classifying isomorphism classes of triples (A, ι, t) where (A, ι) is a false elliptic curve over S and t is a K-level structure 6. In terms of the complex uniformization, [h, 1], h X, represents the isomorphism class of the triple (Ah, ιcan,tcan), where ∈

Ah = ( B R) / B, where the complex structure on B R is deﬁned by right • multiplicationO ⊗ byOh(z), z C; O ⊗ ∈ ιcan is the action of B arising from left multiplication; • t is the K-equivalenceO class of the canonical isomorphism A [N +]= 1 / . • can h N + OB OB

When N − = 1 (so B = M2(Q)) XN +,N is not proper, but a proper model XN∗ +.N − is − obtained by adding cusps; this extends to a regular model over Z(p), which is still smooth if (good) holds. If N = 1, then X + − is already proper, but to unify notation we − 6 N ,N also write XN∗ +.N − for XN +,N − in this case.

Let ℓ0 ∤ Npp0 be a prime (to be chosen later). Associated with ℓ0 is the Hecke correspondence Tℓ0 for XN∗ +,N − . The degree of this correspondence is ℓ0 + 1, so for any x X the divisor ∈ N∗ +,N − 0 (T ℓ 1)[x] Div (X∗ + − ) ℓ0 − 0 − ∈ N ,N has degree 0.

Let J(XN∗ +,N − )/Q be the Jacobian of XN∗ +,N − . We deﬁne a ﬁnite morphism

(4.2.d) ι + − : X∗ + − J(XNs), x (T ℓ 1)[x]. N ,N N ,N → 7→ ℓ0 − 0 − Remark 4.2.1. To compare with formulas in [CST14] (and ultimately with those in [YZZ13]), we also consider a diﬀerent embedding. Let δ + − Pic(X ) be deﬁned N ,N ∈ N∗ +,N − as follows. If N − = 1, then let m be an integer that annihilates the cuspidal subgroup

6 − When N = 1 (so B = M2(Q) and G = GL2) the usual moduli interpretation of the modular curve X1,N = X0(N) classiﬁes elliptic curves E together with a Γ0(N)-equivalence class of isomorphisms ∼ 2 α: E[N] → (Z/NZ)S . The two moduli problems are isomorphic, with the class of (E, α) being identiﬁed with the class of A = E × E together with the obvious action of OB = M2(Z) and the K-level structure α × α. THEBSDFORMULAFORTHERANKONECASE 27

0 of Pic (XN∗ +,N − ) (which is finite by the the theorem of Manin-Drinfeld) and δN +,N − = m[ ], where [ ] is the divisor of the cusp at infinity. If N = 1, then let δ + − = ∞ ∞ − 6 N ,N ξ(X^ ) be the Hodge class defined in [Zha01b, 6.2] and let m 1 be its degree. In N∗ +,N − § ≥ all cases, the action of any Hecke correspondences on δN +,N − is just multiplication by the degree of the correspondence (in particular, the action is Eisenstein). We define a finite morphism (over Q):

(4.2.e) ˜ι + − : X∗ + − J(X∗ + − ), x m[x] δ + − . N ,N N ,N → N ,N 7→ − N ,N As δN +,N − is Eisenstein we then have

(4.2.f) (T ℓ 1) ˜ι + − (x)= m ι + − (x). ℓ0 − 0 − · N ,N · N ,N

CM points on X∗ + − . Let ι : ֒ B be an optimal embedding with respect .4.3 N ,N K K → to R, in the sense that ι ( ) R = . Each of the hypotheses (H) and (gen-H) ensures that such an embeddingK exist.K ∩ OK + There exists a unique point h X such that h h and ι ( ×) ﬁxes h. Moreover, ∈ ∈ K K the subgroup of B× ﬁxing h is just ι ( ×). Replacing the choice if ι with ι τ if necessary (this does not change the pointK K h), we may assume that theK homomorphismK ◦ 1 h: C G(R) is such that h(i)= ι (√ D) and that the action of ι (k), k ×, → K − ⊗ √D K ∈ K h on 1 is just multiplication by k.

The set of CM points of XN∗ +,N − is, in terms of the complex uniformization (4.2.c),

CM(X∗ + − )= [h, b] X + − (C) : b B× . N ,N { ∈ N ,N ∈ } This set does not depend on ι since any two embeddings ֒ B are B -conjugate. In b × addition, the ﬁxed optimal embeddingK ι induces a bijectionK → K

(4.3.a) CM(X∗ + − ) × G(A )/R×. N ,N ≃ K \ f ab Shimura’s reciprocity law shows that CM(X + − ) X + − ( ) and that the action N ,N Nb ,N ab ⊂ K recK ab of Gal( / ) is described in terms of the reciprocity map × A×,f Gal( / ) by K K K \ K → K K rec (t)[h, b] = [h, ι (t)b]. K K In particular, since ι is an optimal embedding, the point x = [h, 1] is deﬁned over the Hilbert class ﬁeld H Kof . Let K N +,N − σ x = ιN +,N − (x) J(XN∗ +,N − )( ). K ∈ K σ Gal(H/ ) ∈ X K N +,N − Note that x depends on the auxiliary prime ℓ0 (to be chosen later). K Let f S2(Γ0(N) be a a newform. Then the modular abelian variety Af is a quotient of J(X ∈ ). Let π : J(X ) A be such a quotient map (in later applications N∗ +,N − N∗ +,N − → f we will choose a π with nice properties at p). We then obtain a Heegner point N +,N − N +,N − y = π(x ) Af (K). K K ∈ 28 DIMITARJETCHEV,CHRISTOPHERSKINNER,ANDXINWAN

Let p be a prime of Z(f) containing p. Suppose that (ℓ -good) p ∤ (a(ℓ ) ℓ 1). 0 0 − 0 − Then N+,N − 1 N +,N − z = y A ( ) Z(f)p a(ℓ ) ℓ 1 ∈ f K ⊗Z(f) K 0 − 0 − K is independent of the choice of ℓ0. Remark 4.3.1. If (irred ) holds for the residual representation modulo the prime p, then K there is a positive proportion of primes ℓ0 ∤ Npp0 for which (ℓ0-good) holds.

4.4. Upper bounds on #X(E/ )[p∞]. Suppose now that Af = E is an elliptic equivalently (equivalenty, that Q(f)=KQ). Here, we recall a consequence of Kolyvagin’s theorem on the structure of X(E/ )[p∞] extended to the case of elliptic curves quotients of Jacobians of Shimura curves.K The following result is a direct consequence of [Nek07, Thm.3.2]:

Theorem 4.4.1. Suppose that E[p] is an irreducible GQ-representation. Suppose that hypothesis (H) holds for the imaginary quadratic ﬁeld and the conductor N of E. Then K + − 2mN ,N (4.4.a) #X(E/ )[p∞] p 0 , K ≤ N +,N − N +,N − where m0 = ordp[E( ) : Zy ]. K K Remark 4.4.2. In fact, one can deduce from the work of [Kol91b] (see also [McC88]) and its generalizations to Shimura curves by Nekov´aˇr[Nek07] a more precise result: + − + − 2(mN ,N mN ,N ) (4.4.b) #X(E/ )[p∞]= p 0 − ∞ , K + − where mN ,N is the maximal non-negative integer m such that pm divides all the cohomology∞ classes constructed by Kolyvagin (see [Jet08] for more details in the case of modular curves).

4.5. Comparisons of CM periods. In the following we recall the complex and p-adic periods of CM elliptic curves and explain how they can be identiﬁed with periods of ‘false’ CM elliptic curves.

4.5.1. Complex and p-adic periods of CM elliptic curves. Suppose that p splits in . Let K H/ be the Hilbert class field of and let F be a finite extension of H and let E0/F beK an elliptic curve with complexK multiplication (CM) by an order O in the imaginary quadratic field . We assume that O Z is a maximal Z -order. Fix a complex K ⊗ (p) (p) uniformization E0(C) ∼= C/a, where a is a non-zero ideal of O = End(E0). By enlarging F if necessary we may assume that E0 has good reduction at all primes above p and so extends to an elliptic scheme / . We are interested in Ω1( / ), which is a E0 OF,(p) E0 OF,(p) free -module of rank one. Write Ω1( / )= ω for some differential OF,(p) E0 OF,(p) OF,(p) · E0 ω Ω1( / ). E0 ∈ E0 OF,(p) THEBSDFORMULAFORTHERANKONECASE 29

We define the complex period of E0 as follows. We consider C to be an F -algebra via the fixed inclusion ι : Q ֒ C. Then ∞ → 1 1 Ω ( 0/ F,(p)) C =Ω (E0/C). E O ⊗OF,(p) As the invariant differential ωC = dz on the complex torus E0(C) = C/a gives a generator 1 ∼ for Ω (E0/C), there exists a scalar Ω C× such that ωE0 =Ω (2πi) ωC. This agrees with the definition of the complex period∞ ∈ of the CM elliptic cu∞rve· that· is given in, e.g., [BDP13, p.1132].

We define the p-adic period of E0 in much the same way, following [BDP13, p.1134]. Recall that p = vv¯ splits in . We take the place v to be that determined by the fixed embedding ι : Q ֒ Q andK ι : Q ֒ = Q to just be ι . Denote also by v the p → p v → Kv p p place of F determined by the embedding ιv; so ιv identifies Fv with a finite extension of = Q in Q . Let F ur be the maximal unramified extension of F in Q and let Kv p p v v p R be the ring of integers of F ur. Considering the good integral model /R, let be v E0 E0 the formal completion over the identity section. As we are working over R, is non- E0 canonically isomorphic to Gm. We fix such an isomorphism 0 ∼ Gm. The latterb is E0 → equivalent to fixing an isomorphism of p-divisible groups [p ] = [p ] = µb∞ , which E0 ∞ E0 v∞ ∼ p is uniquely determined up tob the action of an element of Zp×. Hereb pvbis the prime ideal of O corresponding to v. The pullback of dt/t under the the fixed isomorphism 0 ∼ Gm is an element E → 1 1 ωcan Ω ( 0/R)=Ω ( / F,(p)) R. b b ∈ E E O ⊗OF,(p) We then define the p-adic period Ω R by ω =Ω ω . p ∈ × E0 p · can

Remark 4.5.2. Both Ω and Ωp depend on the choice of ωE0 . However, another choice ∞ of ωE0 only replaces Ω and Ωp with their multiples by the same scalar in × . The ∞ OF,(p) definition of Ω also depends on the choice of the complex uniformization. Composing with multiplication∞ by a non-zero scalar α C : ∈ × E (C) ∼ C/a ∼ C/αa, 0 → → the complex period Ω gets multiplied by the scalar α. Similarly, Ωp also depends on ∞ the choice of the isomorphism = G ; changing this isomorphism replaces Ω with a E0 ∼ m p Zp×-multiple. b b 4.5.3. Complex and p-adic periods of “false” CM elliptic curves. To understand the complex and p-adic periods that appear in the formulas in [Bro14], let A be a “false elliptic curve” as in Section 4.2. We take A and its endomorphisms to all be defined over a finite extension F of containing the real quadratic field M. We assume that A has good K reduction at all places of F over p and we consider a good integral model / F,(p). Then 1 A O Ω ( / F,(p)) is free of rank two over F,(p). Let e B M,(p) be an idempotent as A O O ∈ O1 ⊗ O 1 in Section 4.2 and let ωA be an F,(p)-generator of eΩ (A/ F,(p)), i.e., eΩ ( / F,(p))= O 1 O 1 A O F,(p) ωA. To define a complex period, consider Ω (A/C)=Ω ( / F,(p)) C. If O · A O ⊗OF,(p) we have a canonical basis element ω eΩ1(A/C), we will then be able to compare A,C ∈ it against ωA to define a complex period. We would similarly get a p-adic period from 30 DIMITARJETCHEV,CHRISTOPHERSKINNER,ANDXINWAN comparison with a canonical R-basis element of eΩ1( /R). In the next section, we explain how to obtain such canonical elements when AAis a false elliptic curve obtained from a CM elliptic curve via Serre’s tensor product construction.

Remark 4.5.4. For the above, the only hypotheses on the real quadratic ﬁeld M that is needed is that B M is split and p splits in M. No additional hypotheses are needed on the idempotent⊗e. The extra hypotheses imposed in Section 4.2 are assumed in the construction of Brooks [Bro14] where they are used to obtain a certain explicit formula for the Maass–Shimura operators; they follow closely the choices made by Mori [Mor11] and Hashimoto [Has95]. For the comparison between periods of CM elliptic curves and false elliptic curves in the next section we do not need to make these extra assumptions.

4.5.5. Serre’s tensor product construction. To relate the periods of a CM elliptic curve to the periods of a false elliptic curve, we use Serre’s tensor product construction (see [CCO14, 1.7.4]) which we brieﬂy recall. If E0/F is a CM elliptic curve as in Section 4.5.1 with complex§ multiplication by , then by Serre’s tensor product construction applied OK to B E0, there exists a false elliptic scheme / F,(p) such that (R)= B O ⊗OK A O A O ⊗OK 0(R) for any F,(p)-algebra R; we let the false elliptic curve A/F be the generic ﬁber of E . We will nowO relate the complex and p-adic periods of A to the complex and p-adic A periods of E0, respectively. We have

1 1 Ω ( / F,(p))= B Ω ( 0/ F,(p))= B F,(p) F,(p) ωE0 . A O O ⊗OK E O O ⊗OK O ⊗OF,(p) O · We continue to write e for the image in B K F,(p) of the idempotent e B F,(p), 1 O ⊗O O ∈1 O ⊗O and let ωA = e ωE0 eΩ ( / F,(p)). Then ωA is an F,(p)-basis of eΩ ( / F,(p)). On the other hand,⊗ ∈ A O O A O

1 1 Ω (A/C)= B Ω (E0/C) = ( B F,(p)) C ωC, O ⊗OK O ⊗OK O ⊗OF,(p) ·

1 where ωC Ω (E0/C) is the 1-form defined in Section 4.5.1. Then ωA,C = e dz ∈ 1 ⊗ is a C-basis of eΩ (A/C). The complex period ΩA, C is then defined via ωA = ∞ ∈ ΩA, (2πi) ωA,C. It follows that ΩA, =Ω . ∞ · · ∞ ∞ The comparison of the p-adic periods of E0 and A is similar. We first note that over R we have

0 0 ∞ [p∞] = B K 0[p∞] = B K 0[pv∞] = ( B K Zp) Zp µp , A O ⊗O E O ⊗O E ∼ O ⊗O ⊗ where Zp is a -module via ֒ ,p = Zp. It follows that the pullback of e dt/t OK OK → OK 1 ⊗ is just ωA,can = e ωcan eΩ( /R)= e( B Zp) Zp Ω ( 0/R). Here we have used ⊗ ∈ A O ⊗OK ⊗ E the ﬁxed embedding M ֒ Qp to identify e with an element of B Zp. The p-adic period Ω R is then→ deﬁned via ω =Ω ω . Clearly, ΩO ⊗=Ω . A,p ∈ × A A,p · A,can A,p p THEBSDFORMULAFORTHERANKONECASE 31

5. Anticyclotomic p-adic L-functions

In this section we recall the p-adic L-functions that appear in the Iwasawa-Greenberg 1 1 main conjectures for the Selmer groups H Σ ( , M) and H Σ ( , ) deﬁned in Sec- ac K ac K M tions 2.3.4 and 3.4, respectively. These conjecturesF and theirF current status are recalled in the section following this one. We continue with most of the notation and conventions of the preceding discussion.

5.1. Anticyclotomic p-adic L-functions. Let be an imaginary quadratic field of K discriminant D < 0 such that p splits in . Let Σcc be the set of those continuous − K K characters ψ : G Qp× that satisfy K → τ 1 ψ = ψ− ; • ψ factors through Γ; • ψ is crystalline at both v andv ¯; • ψ has Hodge–Tate weights n< 0 and n> 0 at v andv ¯, respectively. • − alg Such a character ψ is the p-adic avatar of an algebraic Hecke character ψ of A× as ab K follows. Let rec : A× G be the reciprocity map of class field theory, normalized so K → that uniformizers mapK to geometricK Frobenius elements. Then ψalg is given by alg n n n n ψ (x)= ψ(rec (x))xv− xv¯ x x¯− . K ∞ ∞ To make sense of this expression we have to explain how this is to be seen as a C×- n n valued character. The quantity ψ(rec (x))xv− xv¯, which is a priori Qp×-valued (since K v and v¯ are both just Qp), actually takes values in Q× Qp; we use the chosen K K ιv ⊂ embedding Q = ֒ v = Qp to identify Q× as a subgroup of Qp×. We then use the K → K n n fixed embedding Q C to see ψ(rec (x))xv− xv¯ as C-valued and to identify R with ⊂alg K alg K⊗ n n C. The character ψ then has infinity type (n, n) in the sense that ψ (z)= z z¯− . − ∞ Let f S2(Γ0(N)) be a newform and let and L be as in Section 2.4.1. We assume that L ∈is so large that it contains the HilbertO class field of . We also assume that (gen-H) holds for and N. Let B be the quaternion algebra ofK discriminant N and let K − + be the orders as in Section 4.2. The Jacquet–Langlands correspondence OB,N ⊂ OB implies the existence of a classical holomorphic quaternionic modular form fB (that is, B + a modular form for the arithmetic subgroup Γ0 (N ) defined by the units of the Eichler order + ) of weight 2 and trivial character and having the same Hecke eigenvalues OB,N as f at all primes ℓ ∤ N . Let p Z(f) be the prime determined by the fixed embedding − ⊂ Q(f) ֒ Q (so L is a finite extension of Q(f)p). We now require that (good) hold; that → p is, p ∤ N. Then the form fB can be normalized so that it is defined over Z(f)(p) – that is, 1 it is identified with a global section ωf of Ω (XN∗ +,N − /Z(f)(p)) – and non-zero modulo p.

Let R be the completion of the ring of integers of the subfield of Qp generated by the p-adic field L from Section 2.4 and the maximal unramified extension Qur Q of Q . p ⊂ p p Note that R is a complete DVR. As explained in [BDP13], [Bro14], and [Bur14], under 32 DIMITARJETCHEV,CHRISTOPHERSKINNER,ANDXINWAN

the hypotheses (H) and (good) there exists Lp(f) ΛR := Λ R = R[[Γ]] such that for ∈ ⊗O ψ Σ with n 0 (mod p 1), the image L (f, ψ) := ψ(L (f)) R of L (f) under ∈ cc ≡ − p p ∈ p the R-linear extension of the character ψ to a continuous homomorphismb ψ : ΛR R satisﬁes b →

alg b 2 C(f, ψ) 4n L(f, ψ , 1) (5.1.a) Lp(f, ψ)= Ev¯(f, ψ) t Ωp 4n . · K · α(f,fB)W (f, ψ) · · Ω ∞ The various factors appearing in this expression are:

alg 1 alg 2 1 Ev¯(f, ψ) = (1 ψ (̟v¯)app− + ψ (̟v¯) p− ), with ̟v¯ v¯ a uniformizer; • t is a power of− 2 that depends only on (unimportant since∈ Kp is assumed odd); • K K 1 2n 1 ℓ 1 C(f, ψ) = π − Γ(n)Γ(n + 1)w D − , where w is the number of • 4 K K ℓ + 1 K ℓ N − p Y| roots of unity in ; K alg n 2n W (f, ψ) = Wf N /Q(b)ψ (xb)N bN− , where b and bN are as in • K , ⊂ OK ∈ OK [Bro14, Prop. 8.3], xb A∞ × is such that ordw(xb,w) = ordw(b) for all ﬁnite places w of , and W ∈ C K is a complex number of norm one as in the remarks K f ∈ × following the proof of Lemma 8.4 of [Bro14]; both b and bN can be taken to be coprime to p; f,f Γ0(N) α(f,fB) = h i is a ratio of Petersson norms, which are normalized so fB ,fB B + • h iΓ0 (N ) dxdy that g, g Γ = Γ h g(z)g(z) y2 ; in [Bro14] this ratio was shown to belong to L; h i \ Ωp R× andR Ω C× are, respectively, the p-adic and complex periods of a • ‘false∈ elliptic curve’∞ ∈ with CM by as deﬁned at the start of [Bro14, 8.4]; as explained in Section 4.5.5 theseO canK be taken to be the respective p-adic§ and complex periods of an elliptic curve with complex multiplication by ; alg alg OK L(f, ψ ,s)= L(π ψ ,s 1/2) = L(V ∨ ψ,s), where π is the base change of • K× − f ⊗ K π to GL2(A ) and, following our earlier convention, we use geometric conventions K for the L-function of the G -representation V ∨ ψ. K f ⊗ Remark 5.1.1.

(a) To see the interpolated values Lp(f, ψ) as belonging to R, one first recognizes many of the quantities on the right-hand side of (5.1.a) as algebraic, that is, as belonging to Q C. These are then viewed as belonging to Qp via an embedding .extending the fixed⊂ inclusion Q(f) ֒ L (b) To compare the interpolation formula→ (5.1.a) with that in [Bro14] and [Bur14], 1 alg one should take χ− = ψ AK in the formulas in loc. cit. Note that χ has infinity type ( 2 j, j) with|j · |= n + 1 (this is infinity type (2 + j, j) with the − − 1 alg − conventions of loc. cit.) and L(f,χ− , 0) = L(f, ψ , 1). (c) In [BDP13] and [Bro14] the p-adic L-function is only constructed as a continuous function (though the construction clearly gives a measure). The measure is made explicit, for example, in [Bur14]. THEBSDFORMULAFORTHERANKONECASE 33

In this paper we make use of two important results about Lp(f), due to Burungale and Brooks, respectively. The ﬁrst of these is the vanishing of the µ-invariant of Lp(f) and the second is a formula for the value Lp(f, 1) under the trivial character (which does not belong to Σcc). We recall these results in the following two sections.

5.1.2. Vanishing of the anticyclotomic µ-invariant. The choice of topological generator γ Γ determines a continuous isomorphism Λ ∼ R[[T ]] such that γ 1 T ; we use ∈ R → − 7→ this to identify each λ ΛR with a power series λ(T ). Each element λ ΛR has a ∈ µ(λ) ∈ unique expression λ(T )= ̟R pλ(T )uλ(T ) with ̟R R a ﬁxed uniformizer, µ(λ) Z a non-negative integer, p a monic polynomial of minimal∈ degree, and u R[[T ]] . The∈ λ λ ∈ × integer µ(λ) - the µ-invariant of λ - and the degree of pλ are independent of the choices of γ and ̟R. Under the additional assumption that (-free) N is squarefree

Burungale [Bur14] has shown that the µ-invariant of Lp(f) vanishes: Proposition 5.1.3 ([Bur14, Thm. B]). If p 3 and (irred ), (split), (H), (good), and ≥ K (-free) hold, then µ(Lp(f)) = 0.

Let Σ be a ﬁnite set of places of that do not divide p. We also deﬁne an incomplete Σ K p-adic L-function Lp (f) by removing the Euler factors at those places in Σ:

Σ 1 1 L (f)= L (f) P (ǫ− Ψ− (Frob )) R[[Γ]]. p p × w w ∈ w Σ Y∈ Then LΣ(f, ψ) := ψ(LΣ(f)), ψ Σ , satisﬁes the interpolation formula (5.1.a) but with p p ∈ cc L(f, ψalg, 1) replaced with the incomplete L-value LΣ(f, ψalg, 1) on the right-hand side. b Remark 5.1.4. If w = (ℓ) is an inert place in , then Ψ(Frobw) = 1 and in this case 1 1 1 1 K 1 1 Pw(ǫ− Ψ− (Frobw)) = (1 aℓ(f)ℓ− + ℓ− )(1 + aℓ(f)ℓ− + ℓ− ), which can contribute Σ − to the µ-invariant of Lp (f). In particular, the µ-invariant of the incomplete L-function Σ Lp (f) may be greater than that of Lp(f). This does not happen for the usual cyclotomic p-adic L-function.

5.1.5. A formula for Lp(f, 1). We recall Brooks’ formula [Bro14, Prop. 8.13] (see also [Ski14a, Prop.2.6.1]) for the value Lp(f, 1) of Lp(f) at the trivial character (which is outside the range of interpolation). In order to this, suppose that N and satisfy (gen-H). Recall that there is a logarithm map K

log : J(X∗ + − )( ) ωf N ,N Kv ⊗Zp O → Kv such that d log = ω Ω1(J(X )/Z ) = Ω1(J(X / ). Recall that ωf f ∈ N∗ +,N − p ⊗Zp O N∗ +,N 1 O ω Ω1(J(X )/Z ) is the -basis element associated with f , as deﬁned f ∈ N∗ +,N − p ⊗Zp O O B above. Let ℓ ∤ pN be a prime that splits in and such that 1 a + ℓ = 0 (there are 0 K − ℓ0 0 6 34 DIMITARJETCHEV,CHRISTOPHERSKINNER,ANDXINWAN

N +,N − a positive proportion of such ℓ0). In Section 4.3 we deﬁned a Heegner point x J(X )( ). Then Brooks’ formula is: K ∈ N∗ +,N − K Proposition 5.1.6 ([Bro14, Prop. 8.13]). Suppose (-free) holds and N and satisfy (gen-H). Then K 2 1 1+ p a + − 2 L (f, 1) = p log xN ,N , p 2 − ωf (1 aℓ0 + ℓ0) · p K − where the equality is up to a p-adic unit.

Brooks’ formula is actually given in terms of the logarithm of a point N +,N − σ x˜ = ǫf [x] J(XN∗ +,N − )( ′) Z(f), K ∈ K ⊗ σ Gal(H/ ) ∈ X K N +,N − where [x] is as in the deﬁnition of x but ǫf is a certain projector in a ring cor- N +K,N − N −,N − N +,N − respondences. The relation with x is ǫf x = (1 aℓ0 + ℓ0) x˜ from which we ﬁnd K · K − · K N +,N − N +,N − N +,N − logωf x = logωf ǫf x = (1 aℓ0 + ℓ0) logωf x˜ . K · K − · K Additional comments on the comparison of these cohomological trivializations can be found in [Ski14a, p.13]. We now rewrite Brook’s formula in a form more directly comparable to the expression (3.5.c) at the end of Section 3. We do this in the case that V is an irreducible κ- representation of GQ. Then we can – and do – take Af and the quotient map π : J(X ) A to be (Z(f), p)-optimal in the sense of [Zha14]. This means that N∗ +,N − → f π is the composition of an optimal quotient π : J(X ) A and an isogeny 0 N∗ +,N − → 0 φ : A A such that the image of the map T A T A induced by φ is not 0 → f p 0 → p f contained in pTpAf . We note that (irred ) also holds and that we may choose ℓ0 ∤ pN K such that ℓ0 splits in and (ℓ0-good) holds. Then the formula in Proposition 5.1.6 can be expressed as: K Proposition 5.1.7. Suppose (good) and (irred ) hold. Then with the above choice of A and π, up to a unit in we have that L (f, K1) equals f O p 2 2 1+ p ap N +,N − − logωA z , p f K up to a p-adic unit.

Proof. To prove this proposition it suffices to prove that π (ω ) is an -multiple of ∗ Af O× ωf . By the hypothesis (good), each of J(XN +,N − ), A0, and Af has good reduction at p. Let , , and be their respective Néron models over Z ; these are abelian schemes. J A0 A p The maps π, π0, and φ extend to morphisms of these Néron models. Furthermore, as π0 is an optimal quotient (so the kernel of π0 is also an abelian variety) and since p 1 > 1, the image of π :Ω1( /Z ) Ω1( /Z ) is a Z -direct summand; this follows− from 0∗ A0 p → J p p THEBSDFORMULAFORTHERANKONECASE 35

[BLR90, Thm. 4, p. 187] (cf. [AU96, Cor. A.1]). So it suffices to show that φ∗(ωAf ) is part of an -basis of Ω1( /Z ) . O A0 p ⊗Zp O We have that 1 1 0 Ω ( /Z ) /p =Ω ( /F ) /p = Lie ( [p] )∨ /p. Af p ⊗Zp O Af p ⊗Zp O Fp Af ⊗Zp O eq Using the prime factorization pZ(f)= q p q , the right-hand side can be written as | eq 0 Lie ( Q[q ] )∨ /p. Fp Af ⊗Zp O q p Y| By the choice of ω , the image of ω in Ω1( /Z ) /p is identified with a /p- Af Af Af p ⊗Zp O O basis element of the p-summand (that is, for q = p). So it suffices to prove that φ∗ induces an injection

ep 0 0 .Lie ( [p ] )∨ /p ֒ Lie ( [p] )∨ /p Fp Af ⊗Zp O → Fp A0 ⊗Zp O For this, we note that since the image of TpA0 is not contained in pTpAf , and since Af [p] is an irreducible κ-representation of GQ, the induced homomorphism A [p]= T A /pT A T A /pep T A = A [pep ] 0 p 0 p 0 → p f p f f ep is surjective. Hence the morphism 0[p] f [p ] is surjective and so, too, is the induced map of the Lie algebras of theirA connected→ A subgroups.

5.2. The two-variable Rankin–Selberg p-adic L-function. The p-adic L-function Σ Σ Lp (f) is the specialization of a p-adic L-function p (f) Λ ,R := R[[Γ ]] first con- L ∈ K K structed by Hida [Hid88]. In what follows we recall this p-adic L-function and explain Σ its relation to Lp (f). Let Λ+ = [[Γ+]]. Let O K u = ǫ(γ+). + We say that a continous -homomorphism λ : Λ Qp is algebraic of weight n if n O v → λ(γ+)= u− for some integer n 0. Let K K be the maximal subfield unramified v ≥ ∞ ⊂ ∞ at v, and let Γv = Gal(K / ). Then Γv ∼= Zp. Let = [[Γv]]. The composite of the ∞ K+ I O + canonical homomorphisms Γ ֒ Γ ։ Γv makes into a Λ -homomorphism. We say → K I that a continuous -homomorphismK λ : Q is algebraic of weight k = 2n + 1 1 if O I → p ≥ its restriction to Λ+ is algebraic of weight n. ab Let Θv : A× Γv be the composition of the reciprocity map rec : A× G of class K → ab K K → K field theory with the canonical projection G ։ Γv. For each non-zero fractional ideal , K a of , we let xa A∞ × be a finite idele of such that ordw(xa,v) = ordw(a) for all finiteK places w of ∈. ThenK the formal q-expansionK K

∞ n

b b f = (n)q [[q]], (n)= Θ (xa), ∈I v n=1 a⊂OK,N(a)=n X (a,Xpv¯)=1, 36 DIMITARJETCHEV,CHRISTOPHERSKINNER,ANDXINWAN

is an ordinary -adic eigenform of tame level D , in the sense that if λ : Qp is an algebraicI homomorphism of weight k = 2nK+ 1 with n 0 (mod p I1), → then n ≡ − fλ = n∞=1 λ(b(n))q is the q-expansion of a p-ordinary p-stabilized newform of weight k and level D p. The form fλ can be identified as follows. The condition that λ be P K algebraic of weight k implies that ρ = λ , viewed as a continous Q×-valued character λ |Γv p of G , has Hodge-Tate weights 0 and k 1 at v andv ¯, respectively. Associated with ρλ K alg − is an algebraic Hecke character ρλ with infinity type (0, 1 k) that is unramified at v − 0 andv ¯. The ordinary eigenform fλ is the ordinary p-stabilization of the newform fλ of weight k and level D associated with ρalg (so, in particular, L(f 0,s)= L(ρalg,s)). K λ λ λ Hida + Let Λ = [[1 + pZp]] (this is denoted Λ in [Hid88]). We give Λ the structure of a Hida O 1 1 + Λ -algebra by 1 + Zp u u− γ+− Λ ; this is an isomorphism. We also endow ∋ Hida7→ ∈ 2 with the structure of a Λ -algebra by 1 + Zp u uγ+− . With the latter structure,I our definition of an algebraic homomorphism∋ 7→ of weight∈k I is consistent with that in [Hid88]. Note that is a finite, flat ΛHida-algebra. I For a p-adic -algebra A, let (A) = Hom alg,cts(A, Qp) be the set of all continuous -algebra homomorphisms.O HidaX [Hid88, Thm.O− 5.1b] has shown that there is an element DO in the fraction ring of Λ+ that, when viewed as a p-adic analytic function on ( ) (Λ+), has the followingI⊗O properties. The function D is finite at each point X I ×X + (λ, λ′) ( ) (Λ ) withbλ and λ′ algebraic homomorphisms of respective weights k and n and∈X satisfyingI ×X 1 n k 1, and if k 1 (mod 2(p 1)) and n 0 (mod p 1), then ≤ ≤ − ≡ − ≡ −

k/2 n E(1 + n) (1 + n,fλ,f) (5.2.a) D(λ, λ′) = (D N) N Γ(n)Γ(n + 1)Wf D , K S(λ) Ω(fλ,f,n) where

k 1 alg 2 k 2 alg 2 S(λ) = (1 p − ρλ (̟v)− )(1 p − ρλ (̟v)− ); • − alg s − alg 1 s E(s) = (1 ρλ (̟v)ap(f)p− + ρλ (̟v)p − ); • − pND s alg (f ,f,s)= ζ K (k + 2 2s 2) ∞ a (f )a n = E(s)L(f, (ρ ),s); • D λ − − × n=1 n λ n − λ Ω(f ,f,n) = (2πi)2n 1(2i)k+1π2 f 0,f 0 . • λ − h λ λPiΓ1(DK)

Let R = R = R[[Γv]] and let Lp−( ) be the anticyclotomic p-adic L-function I I⊗O K ∈ I of the imaginary quadratic field K. The p-adic L-function L ( ) is a specialization p− K of Katz’s two-variableb p-adic L-function and satisfies the following interpolation property: for an integer k 0 such that k 1 (mod 2(p 1)) and λ ( ) an algebraic homomorphism of weight≥ k, ≡ − ∈ X I k 2 2k 2 w (2π) − Γ(k) alg alg, τ L−( , λ) := λ(L−( )) = S(λ)Ω − K L(ρ ρ − , 1). p p p (k 1)/2 2k 2 λ λ K K 2D − Ω − K ∞ Here the superscript ‘-τ’ denotes composing the inverse character with the action on A× K of the nontrivial automorphism τ of K. The complex and p-adic periods Ω and Ωp are the periods of an elliptic curve with CM by and can be fixed to be the same as those OK THEBSDFORMULAFORTHERANKONECASE 37 appearing in (5.1.a). Appealing to the well-known relation 0 0 2 2k 1 k τ fλ ,fλ Γ1(DK) = Γ(k)D 2− π− − ress=k (s,fλ,fλ ) h i K · D 2 2k 1 k 2πh alg alg, τ = Γ(k)D 2− π− − K L(ρλ ρλ − , 1), K · w D1/2 K K where h is the class number of , the interpolation formula for Lp−( ) can be rewritten as K K K 2k 2 3k 4 2 2k 2 π − 2 − 0 0 (5.2.b) L−( , λ)= w S(λ)Ω − f ,f . p p k/2+1 2k 2 λ λ Γ1(DK) K K h D Ω − h i K K ∞ alg alg, τ Note that ρ ρ − is an anticyclotomic character with infinity type (k 1, 1 k). λ λ − − Suppose (units) p ∤ w . K We put h (f)= K L−( ) 1 D. Lp w p K ⊗ K We consider p(f) as an element of the fraction field of Λ = [[Γ ]] via the isomor- L + K O K + phism Γ ∼ Γv Γ that is the direct sum of the canonical projections to Γv and Γ , respectively.K → Then⊕ forK a finite set Σ of finite places of K not dividing p we put K

Σ 1 p (f)= p(f) Pw(ǫ− Ψ (frobw)). L L × K w Σ Y∈ If Σ contains all places dividing ND that do not divide p, then Wan has shown that Σ K Σ Σ p (f) Λ [Wan13]. For such Σ we let p (f)− Λ be the image of p (f) under the quotientL ∈ mapK Λ ։ Λ induced by the canonicalL projection∈ Γ ։ Γ. L K K

5.3. Relating Σ(f) to LΣ(f). Let ψ Σ . By abuse of notation we also denote by Lp − p ∈ cc ψ the homomorphism of Λ obtained by composition with the projection Λ ։ Λ and Σ ΣK K we put p (f, ψ)− = ψ( p (f)). Then the homomorphism ψ (Λ ) corresponds via L L ∈ X K b + + n the isomorphism Λ ∼ Λ to the point (λ, λ′) ( ) (Λ ) with λ Γv = ψnǫ− K →Ib ⊗O ∈Xn I ×Xb | (and so algebraic of weight 2n + 1) and with λ′ Γ+ = ǫ (and so algebraic of weight n). | K Here n is such that ψalg hasb inﬁnity type (n, n). From the interpolation formulas for D(λ, λ ) and L ( , λ) we then ﬁnd − ′ p− K Σ Σ 2n+1/2 1 1 2n 2n 1 p (f, ψ)− = ψ( p (f)) = w N D− Γ(n)Γ(n + 1)Wf i− 2 π − L L K K LΣ(f, ψalg, 1) b E (f, ψ)2Ω4n n . × v¯ p Ω4n In particular, for ψ Σ , ∈ cc Σ (5.3.a) (f, ψ)− = c(f, ψ)L (f, ψ), Lp p 38 DIMITARJETCHEV,CHRISTOPHERSKINNER,ANDXINWAN where

1 3/2 3n+1/2 ℓ + 1 2n+2 1 alg 2n c(f, ψ)= tK− D− N 2 i− N /Q(b)ψ (xp)bN− α(f,fb). K ℓ 1 K ℓ N − Y| − Σ The following lemma allows us to pass from (5.3.a) to a relation between p (f, ψ)− and Σ L Lp (f).

Lemma 5.3.1. There exist c R 1 × and Λ such that c(f, ψ) = cψ( ) for all ∈ p U ∈ × U ψ Σcc with n 0 (mod p 1). h i ∈ ≡ − b Proof. We let

1 3/2 1/2 ℓ + 1 2 1 1 c = t− D− N 2 i− N /Q(b)α(f,fB) R[ ]×. K K ℓ 1 K ∈ p ℓ N − Y| −

We can assume that xb is chosen so that xb,v =1= xb,v¯. Let γb Γ be the image of alg ∈ rec (xb). Then ψ (xb) = ψ(γb). Let γv Γ be a topological generator of the image of theK inertia group I . For ψ to be crystalline∈ at v with Hodge-Tate weight n means v − that ψ(γ )= ǫ(γ )n. Let a Z such that ǫ(γ )=(1+ p)av and put v v v ∈ p× v

(2 logp(2)+3 logp(N)+logp(bN ))/av = γv γ Γ Λ×. U b ∈ ⊂

Here logp is the Iwasawa p-adic logarithm. Also, we are viewing bN ×,(p) as an ∈ OK element of Zp× via the identification Zp = ,v (which comes from the hypothesis that p OK splits in ). Then if n 0 (mod p 1), satisfies K ≡ − U ψ( )=(1+ p)n(2 logp(2)+3 logp(N)+logp(bN ))ψalg(x ) = 22nN 3nbn ψalg(x ), U b N b and so cψ( )= c(f, ψ). b U b Combining Lemma 5.3.1 with (5.3.a) we find that Σ(f, ψ) = ψ(c LΣ(f)) for all Lp − U p ψ Σcc with n 0 (mod p 1). If Σ contains all the places dividing ND not ∈ ≡ − K dividing p, then Σ(f, ψ) = ψ( Σ(f) ) for all these ψ. Since the setb of kernels of the Lp − Lp − homomorphisms ψ of ΛR for such ψ are Zariski dense in Spec(ΛR), we conclude that we then have Σ(f) = c L (f) inb Λ . Hence: Lp − U p R b Corollary 5.3.2. If Σ contains all places of diving ND that do not divide p, then K K (LΣ(f)) = ( Σ(f) ) in Λ 1 . p Lp − R p h i Remark 5.3.3. It is possible to define Σ(f) for all finite sets Σ of places of not Lp − K dividing p and to directly show, again making use of (5.3.a) and Lemma 5.3.1, that the conclusion of Corollary 5.3.2 holds for all such Σ. We will not need this. THEBSDFORMULAFORTHERANKONECASE 39

6. Main Conjectures

1 Recall that in Section 2.3.4 we defined the Iwasawa-theoretic Selmer groups H Σ ( , M) ac 1 F K and H Σ ( , ). Let FGr K M Σ 1 Σ 1 Xac(M) = Hom (H Σ ( , M), L/ ) and XGr( ) = Hom (H Σ ( , ), L/ ) O Fac K O M O FGr K M O be their Pontrjagin duals. We now recall the Iwasawa-Greenberg main conjectures for these groups together with some recent results towards proving these conjectures. To do this, recall that Λ = [[Γ]] and Λ = [[Γ ]] and that R is the valuation ring of O K O K the completion of the maximal unramified extension of L. Let ΛR = R[[Γ]] and Λ ,R = Σ Σ K R[[Γ ]]. The groups Xac(M) and XGr( ) are, respectively, finite Λ- and Λ -modules. K M K

6.1. The Iwawasa–Greenberg main conjectures. Let Σ be a ﬁnite set of ﬁnite places of that do not divide p. The main conjectures are easy to state: K Σ Conjecture 6.1.1 (Main Conjecture for ). Let p (f) be Hida’s two variable p-adic L-function recalled in Section 5.2. Then M L Σ Σ charΛK XGr( ) Λ ,R = ( p (f)) Λ ,R. M K L ⊂ K

Σ Note that implicit in the statement of this conjecture is that p (f) belongs to Λ ,R and not just its field of fractions. In Section 5.2 this is explainedL to be known to holdK at least when Σ is sufficiently large. Σ Conjecture 6.1.2 (Main Conjecture for M). Let Lp (f) ΛR be the p-adic L-function defined in Section 5.1. Then ∈ char XΣ (M) Λ = (LΣ(f)) Λ . Λ ac R p ⊂ R Remark 6.1.3. Note that we always have equalities of ideals Σ Σ charΛK,R XGr( ) ΛK Λ ,R = charΛK XGr( ) Λ ,R M ⊗ K M K and char XΣ (M) Λ = char XΣ (M) Λ . ΛR ac ⊗Λ R Λ ac R Recently substantial progress has been made toward these conjectures in [Wan13] and [Wan14a] by following the strategy from [SU13] of exploiting congruences between suitable Eisenstein series and cuspforms, this time on GU(3, 1). Theorem 6.1.4 ([Wan13, Thm.1.1],[Wan14a, Thm.1.1]). Suppose (irred ), (split), (gen-H), (good), and (-free) hold. Suppose also that there is at least one primeK divisor of N non-split in and that Σ contains all places dividing ND . Then K K Σ Σ charΓK XGr( ) Λ ,R [1/p] ( p (f)) Λ ,R [1/p]. M K ⊆ L ⊂ K Combining this with Corollary 3.4.2 and Corollary 5.3.2, we conclude: 40 DIMITARJETCHEV,CHRISTOPHERSKINNER,ANDXINWAN

Theorem 6.1.5. Suppose (irred ), (split), (gen-H), (good), and (-free) hold. Suppose also that there is at least one primeK divisor of N non-split in and that Σ contains all places dividing ND . Then K K char XΣ (M) Λ [1/p] (LΣ(f)) Λ [1/p]. Λ ac R ⊂ p ⊂ R This theorem can be strengthened upon combination with Proposition 3.3.3 (in particular, the surjectivity of (3.3.a)) and Burungale’s µ = 0 result of Proposition 5.1.3: Theorem 6.1.6. Suppose (irred ), (split), (gen-H), (good), and (-free) hold. Suppose also that there is at least one primeK divisor of N non-split in . Then K char XΣ (M) Λ (LΣ(f)) Λ . Λ ac R ⊂ p ⊂ R Proof. Let Σ1 Σ2 be two ﬁnite sets of places of dividing p and with Σ2 containing all the places dividing⊂ ND . Then the surjectivityK of (3.3.a) yields K 1 Σ2 Σ1 H ( w, M) charΛ Xac (M) = charΛ Xac (M) charΛ Hom ( 1 K , L/ )  O H ( w, M) O  w Σ2 Σ1 ac ∈Y\ F K Σ1  1 1  = charΛ Xac (M) Pw(ǫ− Ψ− (Frobw)) . w Σ2 Σ1 ∈Y\ Σ2 Comparing this with the deﬁnition of Lp (f) we then see that the hypothesis that Σ contain the places dividing ND can be removed from Theorem 6.1.5. In the case where Σ= the resulting inclusion ofK ideals can then be improved to an inclusion in Λ in light ∅ R of the µ = 0 result of Proposition 5.1.3. The inclusion in ΛR for any Σ then follows. Remark 6.1.7. As a consequence of our results below, in some cases we will be able to improve the inclusion in Theorem 6.1.6 to an equality. That is, we will prove Conjecture 6.1.2 in these cases.

1 6.2. Consequences for the order of H ac ( ,W ). We can now record the key result F K 1 connecting the value Lp(f, 1) with the order of H ac ( ,W ). F K Proposition 6.2.1. Suppose (irred ), (split), (gen-H), (good), and (-free) hold. Sup- pose also that there is at least one primeK divisor of N non-split in . Suppose further that (corank 1) and (sur) also hold. Then L (f, 1) = 0 and K p 6 1 # /(Lp(f, 1)) # H ac ( ,W ) C(W ), O | F K · where C(W )= C∅(W ) is as in Theorem 3.3.1.

Proof. Let (f (T )) = char (X (M)) with f (T ) [[T ]] = Λ, where the isomorphism ac Λ ac ac ∈ O ∼ identiﬁes 1 + T with the chosen topological generator γ. Then by Theorem 6.1.6, Lp(f) divides F (T ) in R[[T ]]. In particular F (0) is an R-multiple of Lp(f, 1). On the other hand, by Theorem 3.3.1 1 # /(fac(0)) = #H ac ( ,W ) C(W ) < . O F K · ∞ THEBSDFORMULAFORTHERANKONECASE 41

(The ﬁniteness of the right-hand side was established in Proposition 3.2.1.) The proposition follows.

7. Proof of the Main Theorem

In this section we piece together the results from the previous sections to prove the main result of the paper, Theorem 1.2.1. We therefore take E/Q to be an elliptic curve as in that theorem. In particular, we assume:

E is a semistable elliptic curve with square-free conductor N; • E has good reduction at the the prime p (i.e., p ∤ N); • if p = 3 and E has supersingular reduction at p, then a (E) = 0; • p the residual representationρ ¯E,p : GQ Aut(E[p]) is irreducible; • ord L(E,s) = 1. → • s=1

The results of Gross–Zagier and Kolyvagin then imply that rkZ E(Q)=1 and X(E/Q) is ﬁnite. To prove Theorem 1.2.1 we must show that the same power of p appears in both sides of (1.1.a). Since Conjecture 1.1.1 is isogeny invariant – more precisely, the ratio of both sides of (1.1.a), when both are ﬁnite, is an invariant of the isogeny class of E (cf. [Tat66, Thm. 2.1]) – we may further assume:

E admits an optimal quotient map π : J(X0(N)) E (that is, such that the • kernel of π is connected). →

7.1. The Birch and Swinnerton-Dyer Conjecture. We will eventually deduce both the conjectured upper and lower bounds on #X(E/Q)[p∞] from corresponding upper and lower bounds for #X(E/ )[p∞] for suitable imaginary quadratic fields . For this reason we find it helpful to recallK the general Birch and Swinerton-Dyer conjectureK for an elliptic curve over a number field. As stated by Tate [Tat66, (A) and (B)], this is:

Conjecture 7.1.1 (general Birch and Swinnerton-Dyer Conjecture). Let F be a number field and let /F be an elliptic curve over F . E (a) The Hasse–Weil L-function L( /F,s) has an analytic continuation to the entire E complex plane and ords=1L( /F,s)=rkZ (F ); (b) The Tate-Shafarevich group EX( /F ) has finiteE order, and E (r) X L ( /F, 1) # ( /F ) v∤ cv( /F ) (7.1.a) = E · ∞ E , E 1/2 2 r! Ω /F Reg( /F ) ∆F − # (F )tors · E · E · | | E Q 0 where r = ords=1L( /F,s), cv( /F ) = [ (Fv) : (Fv)] is the Tamagawa number at v for a finite placeE v of F , Reg(E /F ) Eis the regulatorE of the Néron-Tate height E pairing on (F ), ∆F is the discriminant of F , and Ω /F C× is the period E E ∈ 42 DIMITARJETCHEV,CHRISTOPHERSKINNER,ANDXINWAN

deﬁned by

(7.1.b) Ω /F = NF/Q(aω) ω 2 ω ω . E · (Fv) | | · · (Fv ) ∧ v ZE v ZE ! vY-real|∞ v-complexY|∞

1 Here ω Ω ( ˜/ F ) is any non-zero differential on the Néron model ˜ of over ∈ E O 1 E E F , and aω F is the fractional ideal such that aω ω =Ω ( ˜/ F ). Also, for a O ⊂ 0 · E O finite place v, (Fv) (Fv) denotes the subgroup of local points that specialize to the identityE component⊂E of the Néron model of at the place v. E

When F = Q, we generally write Ω for Ω /Q. E E

7.2. The Birch–Swinnertion-Dyer formula for rank zero elliptic curves. To pass from the expected upper and lower bounds for #X(E/ )[p ] for suitable imag- K ∞ inary quadratic fields = √D (D < 0) to the expected upper or lower bound for K #X(E/Q)[p∞], we will need to appeal to known results for the BSD formula for the -twists7 ED of E. The fields will always be chosen so that the groups ED(Q) have rankK 0. We therefore recall theK known results about the p-part of the BSD formula for rank 0 curves. In the theorem below we summarize the already known results on the Birch and Swinnerton-Dyer conjecture for both ordinary and supersingular elliptic curves of analytic rank zero. The inequality of part (i) of the theorem is a consequence of Kato’s groundbreaking work on an Euler system for elliptic curves (see [PR03, Thm.4.8]). The equality of part (ii) for the ordinary case is proved in [SU13, Thm.2], [Ski14b, Thm.C] as a consequence of the proof Iwasawa main conjecture for GL2, and that of part (iii) for the supersingular case is a consequence of the proof of Kobayashi’s main conjecture [Wan14b, Cor.4.8]. Theorem 7.2.1. Let /Q be an elliptic curve with good or multiplicative reduction at E the odd prime p and suppose that ρ ,p : Gal(Q/Q) Aut( [p]) is irreducible. Suppose E → E L( , 1) = 0. E 6 (i) One has L( /Q, 1) (7.2.a) ordp(#X( /Q)[p∞]) ordp E . E ≤ Ω ℓ cℓ( ) E · E (ii) If has good ordinary or multiplicative reductionQ at p and there exists a prime E q of multiplicative reduction for E at which the representation ρ ,p is ramified, E

7If E has Weierstrass equation y2 = x3 + Ax + B, then the K-twist of E is the curve ED/Q having Weierstrass equation Dy2 = x3 + Ax + B. If f is the newform associated with E, then the newform D associated with E is just the newform fK associated with the twist of f by the quadratic character χK of K. THEBSDFORMULAFORTHERANKONECASE 43

then L( /Q, 1) (7.2.b) ordp(#X( /Q)[p∞]) = ordp E . E Ω ℓ cℓ( ) E · E (iii) If is semistable or a twist of a semistable ellipticQ curve by a quadratic character thatE is unramiﬁed at the primes dividing the conductor of the semistable curve and if has supersingular reduction at p with a ( ) = 0, then E p E L( /Q, 1) (7.2.c) ordp(#X( /Q)[p∞]) = ordp E . E Ω ℓ cℓ( ) E · E Remark 7.2.2. Note that the condition that ap( ) = 0Q in part (iii) of this theorem is superﬂuous if p 5. E ≥ Remark 7.2.3. Part (iii) of the above theorem is slightly more general than the result cited in [Wan14b], where the elliptic curve is assumed to be semistable. However the same proof extends to the case of the quadratic twist of a semistable curve. The only reason for putting the semistable assumption there is to avoid local triple product integrals for supercuspidal representations at split primes in [Wan13], [Wan14a], which is excluded under the assumption of the above theorem. The results of Hsieh and Hung on non- vanishing (or non-vanishing modulo p) of L-values and vanishing of µ-invariants are also valid in the more general setting of the quadratic twist.

7.3. Comparison of Tamagawa numbers and periods for quadratic twists. We will derive both upper and lower bounds on #X(E/Q)[p∞] from corresponding bounds on #X(E/ )[p∞] for suitable choices of imaginary quadratic ﬁelds = Q(√ D). In order to deriveK from this the exact upper and lower bounds predictedK by Conjecture− 1.1.1, we recall here the relations between the Tamagawa numbers and periods of E over and the Tamagawa numbers and periods of E and its quadratic twist8 ED. These are Kdiscussed in detail in [SZ14].

7.3.1. Tamagawa numbers. We recall [SZ14, Cor.9.2.] that

(7.3.a) ord c (E/ ) = ord c (E/Q) c (ED/Q) . p w K p ℓ · ℓ w ! ! Y Yℓ Yℓ On the left-hand side w is running over ﬁnite places of the imaginary quadratic ﬁeld = Q(√D) of discriminant D < 0, and on the right-hand side ℓ is running over all primes.K Here we have also used that E has good reduction at p and so has Tamagawa number 1 at any place dividing p.

8If E has Weierstrass equation y2 = x3 + Ax + B, then by ED we mean the elliptic curve over Q having Weierstrass equation −Dy2 = x3 + Ax + B. This is just the K-twist of E. If f is the newform D associated with E, then the newform associated with E is just the newform fK associated with the twist of f by the quadratic character of K. 44 DIMITARJETCHEV,CHRISTOPHERSKINNER,ANDXINWAN

7.3.2. Periods and comparisons. For the elliptic curve E/Q, one deﬁnes its real period to be

Ω = ω . E | | ZE(R) Here ωE is a Néron differential (a Z-basis of the module of differentials of the Néron model of E over Z). If f S2(Γ0(N)) is the normalized cuspidal eigenform associated with E (that is, satisfying∈L(f,s)= L(E,s)) then, as explained in [SZ14, 9.2], one also + § defines canonical periods Ω , Ω− C ; these are defined up to Z× -multiples (this can f f ∈ × (p) be refined to Z×-multiples, but we do not need this here). Furthermore, as recalled in [SZ14, 9.3], one also has the congruence (or Hida) period §

cong f,f × Ωf = h i Qp , ηf ∈ where f,f is the Petersson norm and η is the congruence number of f recalled in h i f loc. cit.; to make sense of this deﬁnition we use our chosen isomorphism Qp ∼= C. Recall that Ωf is only well-deﬁned up to Zp×-multiple. We next recall the relations between these various periods. It is explained in [Ski14b, 3.3] (see also [GV00, Prop. 3.1]) that § + + (7.3.b) ΩE = 2πiΩ and Ω D = 2πiΩ , − f E − fK up to Z(×p)-multiples. By [SZ14, Lem. 9.5] the congruence period can be chosen to satisfy

cong 2 + (7.3.c) Ωf = i(2π) Ωf Ωf−. Let f be the newform associated with the twist of f by the quadratic character of the K imaginary quadratic ﬁeld . Under the hypothesis thatρ ¯E,p is irreducible, it is shown in [SZ14, Lem. 9.6] that ifKp ∤ D then

(7.3.d) Ω± = √ D Ω∓ , f − · fK up to a Z -multiple. Note that the statement of loc. cit. omits the factor √ D (which p× − equals the Gauss sum τ(χ ) up to sign); this is because it is assumed there that p splits in (also assumed here)K and so √ D Z . In particular, combining (7.3.b), (7.3.c), K − ∈ p× and (7.3.d) we ﬁnd:

cong (7.3.e) Ω = D Ω Ω D . f | | · E E p Remark 7.3.3. The comparison between the periods ΩE of ΩED of the elliptic curves and Ω+ and Ω+ is often done in terms of what is known as the Manin constants, as f fK explained in [GV00, 3]. It is known by a result of Mazur that if p ∤ 2ND, then p does not divide either of the§ Manin constants (see, e.g., [Jet08, 1]). § THEBSDFORMULAFORTHERANKONECASE 45

7.4. The ﬁnal argument. Let f S (Γ (N)) be the Hecke eigenform associated to E. ∈ 2 0 Then (V,T,W ) = (V , T ,W ) = (V E, T E, E[p ]), = Z , and V = E[p]. Note that f f f p p ∞ O p ∼ there exists at least one prime q N such that the mod p representationρ ¯E,p is ramiﬁed at q. If not, then Ribet’s level lowering| theorem [Rib90, Thm.1.1] yields a cuspform g of weight 2 and level 1 with mod p residual representation isomorphic toρ ¯E,p (we apply Ribet’s theorem to remove, one-by-one, the primes dividing N). This is a contradiction as there are no cuspidal eigenforms of weight 2 and level 1. Let N = q1 ...qr with q1 = q.

7.4.1. Lower bounds on #X(E/Q)[p∞]. We ﬁrst choose an auxiliary imaginary quadratic ﬁeld = Q(√D ) of discriminant D < 0 such that K′ ′ ′

(a) N and ′ satisfy (gen-H); (b) q is eitherK inert or ramiﬁed in ; K′ (c) p splits in ′; ′ K (d) L(ED , 1) = 0. 6

It is easy to find ′ such that (a), (b) and (c) hold. Indeed, if r 2 (i.e., N has at least two prime divisors),K then we can even guarantee that (H) holds≥ by requiring that q be + inert in ′. In the case where N = q is prime, we take N − = 1 and N = q, which only satisfiesK (gen-H). We note that for any for which (a), (b), and (c) hold, the root K′ number w(E/ ′) of E/ ′ is 1 (cf. (sign 1)). As w(E/Q)= 1 by hypothesis (this is K K − − − D′ a consequence of ords=1L(E,s) being odd) and w(E/ ′)= w(E/Q)w(E /Q), we have ′ K w(ED ) = +1. As (a), (b), and (c) impose only finitely many congruence conditions on the discriminant of ′, it then follows from a result of Friedberg and Hoffstein [FH95, Thm. B] that canK be chosen so that (d) also holds. Note that the condition (d) means K′ that ords=1L(E/ ′,s) = 1. In particular, by the work of Gross–Zagier and Kolyvagin we know that E(K ) has rank one and that X(E/ ) is finite. K′ K′ To take a first step towards a lower bound on #X(E/ ′)[p∞] we want to appeal to the Proposition 6.2.1. So we first check that the hypothesesK of that proposition hold with = . The conditions (split), (gen-H), (good), and (-free) either follow immediately K K′ from our hypotheses on E or from the choice of ′. The condition (irred ) is an easy K K consequence of the hypotheses thatρ ¯E,p is irreducible and thatρ ¯E,p is ramified at the prime q = q1 N (see [Ski14a, Lem. 2.8.1]). Finally, note that (corank 1) and (sur) also hold as rk E(|| )=1and #X(E/ )[p ] < (cf. Section 3.5). K′ K′ ∞ ∞ By Proposition 6.2.1 (with = ) we have K K′ 1 (7.4.a) ordp (Lp(f, 1)) ordp # H ac ( ′, E[p∞]) C(E[p∞]) . ≤ F K · N +,N − Let z ′ = z ′ E( ′) be the Heegner point defined in Section 4.3. (Note that K K ∈ K the hypothesis (irred ) ensures that there are many good auxiliary primes ℓ0.) As the hypotheses of PropositionK 5.1.7 are also clearly satisfied, the left-hand side of (7.4.a) is

1+ p ap ordp(Lp(f, 1)) = 2 ordp − logω (z ′ ) . · p E K 46 DIMITARJETCHEV,CHRISTOPHERSKINNER,ANDXINWAN

As Lp(f, 1) = 0 (also by Proposition 6.2.1), this implies that z ′ has infinite order (of 6 K course, this is also a consequence of the general Gross–Zagier formula for z ′ ). Let K m ′ = [E( ′) : Z z ′ ]. K K · K This index is finite. The hypotheses of Section 3.5 are also all satisfied, so it follows from (3.5.d) and the definition of C(E[p∞]) that the left-hand side of (7.4.a) is 1 X ′ ordp # H ac ( ′, E[p∞] C(E[p∞]) = ordp # (E/ ′) 2 ordp(m ) F K · K − · K 1+ p ap + 2 ordp − logω (z ′ ) · p E K

+ ord c (E/ ′) . p  w K  w N + Y|   We then conclude from all this that

(7.4.b) ordp #X(E/ ′)[p∞] 2 ordp(m ′ ) ordp cw(E/ ′) . K ≥ · K −  K  w N + Y|   To pass from the inequality (7.4.b) to one involving the derivative L′(E/ ′, 1) we use 9 K the variant of the Gross–Zagier formula for the Heegner point z ′ [Zha14, p. 245]: up K to a unit in Zp×, L′(E/ ′, 1) δ(N, 1) z ′ , z ′ NT K K D′ congK = + h 2 i , | | Ωf δ(N ,N −) cE p where cE Z is the Manin constant of E (so p ∤ cE in this case; see Remark 7.3.3) and ∈ + δ(N, 1), δ(N ,N −) are deﬁned in loc. cit. In particular, as explained on the same page of loc. cit., δ(N, 1) δ + − := = c (E/ ′) N ,N δ(N +,N ) ℓ K − ℓ N − Y| up to a unit in Zp×. We can then rewrite the Gross–Zagier formula for z ′ as an equality up to p-adic unit: K

D′ L′(E, 1) L(E , 1) 2 (7.4.c) = m ′ cℓ(E/ ′). ΩE Reg(E/Q) · Ω D′ K · K E ℓ N − · Y| Here we have used that z ′ , z ′ NT = m ′ Reg(E/ ′), Reg(E/ ′) = Reg(E/Q) (as ′ h K K i K′ K K ED (Q) is ﬁnite), L (E/ ) = L (E, 1)L(ED , 1), and the period relation (7.3.e). From ′ K′ ′

9 Though stated in terms of zK′ in [Zha14], the formula in loc. cit. is a special case of a formula in [CST14], which is expressed in terms of a ‘Heegner point’ deﬁned using the map 1/m·˜ιN+ ,N− from (4.2.e) ′ in place of ιN+,N− . However, it is clear from (4.2.f) that replacing this point with zK only changes the formula by a p-adic unit. THEBSDFORMULAFORTHERANKONECASE 47

(7.4.c) we obtain

D′ L′(E, 1) L(E , 1) 2 ordp(m ′ ) = ordp ordp cℓ(E/ ′) . · K ΩE Reg(E/Q) · Ω D′ −  K  E ! ℓ N − · Y|   Combining this with (7.4.b) and (7.3.a) (with = ) we obtain K K′ D′ X L′(E, 1) L(E , 1) ordp(# (E/ ′)[p∞]) ordp D′ . K ≥ Ω Reg(E/Q) c (E/Q) · Ω D′ c (E /Q) E · ℓ ℓ E ℓ ℓ ! X X X D′ Q Q As (E/ ′)[p∞] ∼= (E/Q)[p∞] (E /Q)[p∞], from the above lower bound on K ⊕ D′ ordp(#X(E/ ′)[p∞]) and Theorem 7.2.1(i) (really (7.2.a) for = E ) we conclude that K E

L′(E, 1) (7.4.d) ord (#X(E/Q)[p∞]) ord . p ≥ p Ω Reg(E/Q) c (E/Q) E · ℓ ℓ That is, we have proved the exact conjectured lower bound onQ #X(E/Q)[p∞].

7.4.2. Upper bounds on #X(E/Q)[p∞]. Recall that N = q1 qr with q1 = q such that + · · · + p ∤ cq(E/Q). If r is odd, let N = q1 and N − = N/q1. If r is even, let N = 1 and N − = N. We choose a second auxiliary imaginary quadratic ﬁeld Q(√D′′) of discriminant D′′ < 0 such that

+ (a) the primes dividing N split in ′′; (b) the primes dividing N are all inertK in ; − K′′ (c) p splits in ′′; ′′ K (d) L(ED , 1) = 0. 6

Note that (H) holds for N and any ′′ satisfying (a), (b), and (c). As with the choice of K D′′ ′ above, the root number of the quadratic twist E is +1 and the result of Friedberg Kand Hoﬀstein ensures that can be chosen so that (d) also holds. K′′ N +,N − Note that by (d) we have ords=1L(E/ ′′,s) = 1. Let z ′′ = z ′′ E( ′′) be the K K K ∈ K Heegner point deﬁned in Section 4.3 and let m ′′ = [E( ′′) : Z z ′′ ]. From Theorem 4.4.1 we obtain K K · K

ordp(#X(E/ ′′)[p∞]) 2 ordp(m ′′ ). K ≤ · K From the Gross–Zagier formula for z ′′ we have, just as we did for z ′ , that K K D′′ L′(E, 1) L(E , 1) 2 ordp(m ′′ ) = ordp ordp cℓ(E/ ′′) . · K ΩE Reg(E/Q) · Ω D′′ −  K  E ! ℓ N − · Y| +   For our choice of ′′ there are no w N such that p cw(E/ ′′), and so the product on the left-hand sideK can be replaced| by a product over| all wK. Combining the above 48 DIMITARJETCHEV,CHRISTOPHERSKINNER,ANDXINWAN equality with the preceding inequality and appealing to (7.3.a) (for = ) we get K K′′ D′′ X L′(E, 1) L(E , 1) ordp(# (E/ ′′)[p∞]) ordp D′′ K ≤ Ω Reg(E/Q) c (E/Q) · Ω D′′ c (E /Q) E · ℓ ℓ E ℓ ℓ ! ′′ Appealing to part (ii) or (iii) of Theorem 7.2.1 (forQ = ED ) then yieldsQ E L′(E, 1) (7.4.e) ord (#X(E/Q)[p∞]) ord . p ≤ p Ω Reg(E/Q) c (E/Q) E · ℓ ℓ X That is, we have proved the exact conjectured upper boundQ on # (E/Q)[p∞].

7.4.3. The ﬁnal step: equality. Combining (7.4.d) and (7.4.e) gives

L′(E, 1) ord (#X(E/Q)[p∞]) = ord , p p Ω Reg(E/Q) c (E/Q) E · ℓ ℓ proving Theorem 1.2.1. Q

7.4.4. Some remarks on the hypotheses in Theorem 1.2.1. We make a few comments on how the various hypotheses on Theorem 1.2.1 intervene in its proof and some remarks on possible generalizations.

(i) The requirement that N be square-free is made in [Bro14] and so appears as a hypothesis of Proposition 5.1.6. However, this condition can likely be dropped in light of the main results of [LZZ13]. The formulas in the latter are not as explicit as the result in [Bro14] but can made so just as the general Gross–Zagier formula is made explicit in [CST14]. (ii) The requirement that N be square-free is also made in [Wan13] and [Wan14b], but as we have already indicated (see Remark 7.2.3) this can be replaced with the requirement that the cuspidal automorphic representation π = ℓ π associated with E is either a principal series or a Steinberg representation⊗ ≤∞ at each ℓ N. Further relaxing of this conditions requires a better understanding of certain local| triple product integrals. (iii) If p = 3 and E has supersingular reduction at p, then we have required that ap(E) = 0. This is only because the same hypothesis is made in [Wan14b] and ′′ so p-part of the BSD formula in the rank 0 case (needed for L(ED , 1) in the argument deducing the exact upper bound) is only known for supersingular p when ap(E) = 0. (iv) As much as possible, we have worked in the context of the Selmer groups of a general newform f S2(Γ0(N)) of weight 2 and trivial character. The ultimate restriction to the case∈ of an elliptic curve is made for two reasons: (1) the lack of a precise reference for the upper-bound on #X(Af / )[p∞] coming from the Euler system of Heegner points (the generalization of TheorK em 4.4.1), and (2) the lack of a general result about the p-part of the special value L(f , 1)/2πiΩ+ K fK (when L(f , 1) = 0) when f is not ordinary at p. We expect that both of these issues willK be addressed6 in forthcoming work. THEBSDFORMULAFORTHERANKONECASE 49

(v) The condition that p be a prime of good reduction can likely be relaxed to at least a prime of multiplicative reduction. The rank 0 special value formulas are proved, for example, in [Ski14b], and the results of [Cas15] and [LZZ13] allows p to be a prime of multplicative reduction. (vi) That p is odd is, of course, an essential hypothesis of many of the results used in the course of our proof.

Acknowledgements

We thank Ernest H. Brooks for a careful reading of an earlier draft of this paper and various helpful discussions. We are grateful to Christophe Cornut for his interest and multiple useful conversations on the problem. The work of the ﬁrst named author (D.J.) was supported by Swiss National Science Foundation professorship grant PP00P2-144658. The work of the second named author (C.S.) was partially supported by the grants DMS-0758379 and DMS-1301842 from the National Science Foundation and by the Simons Investigator grant #376203 from the Simons Foundation. He also gratefully acknowledges the hospitality of the California Institute of Technology, where part of this work was carried out during an extended visit in 2014.

References

[AU96] A. Abbes and E. Ullmo, A` propos de la conjecture de Manin pour les courbes elliptiques modulaires, Compositio Math. 103 (1996), no. 3, 269–286. [BBV15] A. Berti, M. Bertolini, and R. Venerucci, Congruences between modular forms and the Birch and Swinnerton-Dyer conjecture, preprint (2015). [BDP13] M. Bertolini, H. Darmon, and K. Prasanna, Generalized Heegner cycles and p-adic Rankin L-series, Duke Math. J. 162 (2013), no. 6, 1033–1148. [Bir71] B. J. Birch, Elliptic curves over Q: A progress report, 1969 Number Theory Institute (Proc. Sympos. Pure Math., Vol. XX, State Univ. New York, Stony Brook, N.Y., 1969), Amer. Math. Soc., Providence, R.I., 1971, pp. 396–400. [BK90] S. Bloch and K. Kato, L-functions and Tamagawa numbers of motives, The Grothendieck Festschrift, Vol. I, Progr. Math., vol. 86, Birkhäuser Boston, Boston, MA, 1990, pp. 333–400. [BLR90] S. Bosch, W. Lütkebohmert, and M. Raynaud, Néron models, Springer-Verlag, Berlin, 1990. MR 91i:14034 [Bro14] H. Brooks, Shimura curves and special values of p-adic l-functions, to appear in Intl. Math. Res. Notices (2014). [Bur14] A. Burungale, On the non-triviality of generalized Heegner cycles modulo p, II: Shimura curves, preprint (2014). [Buz97] K. Buzzard, Integral models of certain Shimura curves, Duke Math. J. 87 (1997), no. 3, 591– 612. [Cas15] F. Castella, On the exceptional specializations of big Heegner points, preprint (2015). [CCO14] C.-L. Chai, B. Conrad, and F. Oort, Complex multiplication and lifting problems, Mathematical Surveys and Monographs, vol. 195, American Mathematical Society, Providence, RI, 2014. [CST14] L. Cai, J. Shu, and Y. Tian, Explicit Gross-Zagier and Waldspurger formulae, Algebra Number Theory 8 (2014), no. 10, 2523–2572. [FH95] S. Friedberg and J. Hoffstein, Nonvanishing theorems for automorphic L-functions on GL(2), Ann. of Math. (2) 142 (1995), no. 2, 385–423. 50 DIMITARJETCHEV,CHRISTOPHERSKINNER,ANDXINWAN

[FM95] J.-M. Fontaine and B. Mazur, Geometric Galois representations, Elliptic curves, modular forms, & Fermat’s last theorem (Hong Kong, 1993), Ser. Number Theory, I, Int. Press, Cam- bridge, MA, 1995, pp. 41–78. [Gre94] R. Greenberg, Iwasawa theory and p-adic deformations of motives, Motives (Seattle, WA, 1991), Proc. Sympos. Pure Math., vol. 55, Amer. Math. Soc., Providence, RI, 1994, pp. 193– 223. [Gre99] , Iwasawa theory for elliptic curves, Arithmetic theory of elliptic curves (Cetraro, 1997), Lecture Notes in Math., vol. 1716, Springer, Berlin, 1999, pp. 51–144. [Gro91] B. H. Gross, Kolyvagin’s work on modular elliptic curves, L-functions and arithmetic (Durham, 1989), Cambridge Univ. Press, Cambridge, 1991, pp. 235–256. [GV00] R. Greenberg and V. Vatsal, On the Iwasawa invariants of elliptic curves, Invent. Math. 142 (2000), no. 1, 17–63. [GZ86] B. Gross and D. Zagier, Heegner points and derivatives of L-series, Invent. Math. 84 (1986), no. 2, 225–320. [Has95] K. Hashimoto, Explicit form of quaternion modular embeddings, Osaka J. Math. 32 (1995), no. 3, 533–546. [Hel07] D. Helm, On maps between modular Jacobians and Jacobians of Shimura curves, Israel J. Math. 160 (2007), 61–117. [Hid88] H. Hida, A p-adic measure attached to the zeta functions associated with two elliptic modular forms. II, Ann. Inst. Fourier (Grenoble) 38 (1988), no. 3, 1–83. [How04] B. Howard, The Heegner point Kolyvagin system, Compos. Math. 140 (2004), no. 6, 1439– 1472. [Jet08] D. Jetchev, Global divisibility of Heegner points and Tamagawa numbers, Compos. Math. 144 (2008), no. 4, 811–826. [KM85] N. M. Katz and B. Mazur, Arithmetic moduli of elliptic curves, Princeton University Press, Princeton, N.J., 1985. [Kob13] S. Kobayashi, The p-adic Gross-Zagier formula for elliptic curves at supersingular primes, Invent. Math. 191 (2013), no. 3, 527–629. [Kol90] V. A. Kolyvagin, Euler systems, The Grothendieck Festschrift, Vol. II, Birkhäuser Boston, Boston, MA, 1990, pp. 435–483. [Kol91a] , On the structure of Selmer groups, Math. Ann. 291 (1991), no. 2, 253–259. [Kol91b] V. A. Kolyvagin, On the structure of Shafarevich-Tate groups, Algebraic geometry (Chicago, IL, 1989), Springer, Berlin, 1991, pp. 94–121. [LZZ13] Y. Liu, S. Zhang, and W. Zhang, On p-adic Waldspurger formula, preprint (2013). [McC88] W. G. McCallum, On the Shafarevich-Tate group of the Jacobian of a quotient of a Fermat curve, Invent. Math. 53 (1988), no. 3, 637–666. [Mil86] J. S. Milne, Arithmetic duality theorems, Academic Press Inc., Boston, Mass., 1986. [Mor11] A. Mori, Power series expansions of modular forms and their interpolation properties, Int. J. Number Theory 7 (2011), no. 2, 529–577. [MR04] B. Mazur and K. Rubin, Kolyvagin systems, Mem. Amer. Math. Soc. 168 (2004), no. 799, viii+96. [Nek07] J. Nekováˇr, The Euler system method for CM points on Shimura curves, L-functions and Galois representations, London Math. Soc. Lecture Note Ser., vol. 320, Cambridge Univ. Press, Cambridge, 2007, pp. 471–547. [PR87] B. Perrin-Riou, Fonctions L p-adiques, théorie d’Iwasawa et points de Heegner, Bull. Soc. Math. France 115 (1987), no. 4, 399–456. [PR03] Bernadette Perrin-Riou, Arithmétique des courbes elliptiques àréduction supersingulière en p, Experiment. Math. 12 (2003), no. 2, 155–186. [Rib90] K. A. Ribet, On modular representations of Gal(Q/Q) arising from modular forms, Invent. Math. 100 (1990), no. 2, 431–476. THEBSDFORMULAFORTHERANKONECASE 51

[RT97] K. Ribet and S. Takahashi, Parametrizations of elliptic curves by Shimura curves and by classical modular curves, Proc. Nat. Acad. Sci. U.S.A. 94 (1997), no. 21, 11110–11114, Elliptic curves and modular forms (Washington, DC, 1996). [Rub00] K. Rubin, Euler Systems, Princeton University Press, Spring 2000, Annals of Mathematics Studies 147. [Sai97] T. Saito, Modular forms and p-adic Hodge theory, Invent. Math. 129 (1997), 607–620. [Ski14a] C. Skinner, A converse to a theorem of Gross, Zagier, and Kolyvagin, preprint (2014). [Ski14b] , Multiplicative reduction and the cyclotomic main conjecture for GL2, preprint (2014). [SU13] C. Skinner and E. Urban, The main conjecture for GL2, to appear in Invent. Math. (2013). [SZ14] C. Skinner and W. Zhang, Indivisibility of heegner points in the multiplicative case, preprint (2014). [Tat63] J. Tate, Duality theorems in Galois cohomology over number fields, Proc. Internat. Congr. Mathematicians (Stockholm, 1962), Inst. Mittag-Leffler, Djursholm, 1963, pp. 288–295. MR 31 #168 [Tat74] , The arithmetic of elliptic curves, Invent. Math. 23 (1974), 179–206. MR 54 #7380 [Tat66] , On the conjectures of Birch and Swinnerton-Dyer and a geometric analog, Séminaire Bourbaki, Vol. 9, Soc. Math. France, Paris, 1965/66, pp. Exp. No. 306, 415–440. [TY07] R. Taylor and T. Yoshida, Compatibility of local and global Langlands correspondences, J. Amer. Math. Soc. 20 (2007), no. 2, 467–493. [Wan13] X. Wan, Iwasawa main conjecture for Rankin–Selberg p-adic L-functions, preprint (2013). [Wan14a] , Iwasawa main conjecture for Rankin–Selberg p-adic l-functions: non-ordinary case, preprint (2014). [Wan14b] , Iwasawa main conjecture for supersingular elliptic curves, preprint (2014). [YZZ13] X. Yuan, S.-W. Zhang, and W. Zhang, The Gross-Zagier formula on Shimura curves, Annals of Mathematics Studies, vol. 184, Princeton University Press, Princeton, NJ, 2013. [Zha01a] S. Zhang, Heights of Heegner points on Shimura curves, Ann. of Math. (2) 153 (2001), no. 1, 27–147. [Zha01b] S. W. Zhang, Gross-Zagier formula for GL2, Asian J. Math. 5 (2001), no. 2, 183–290. [Zha14] W. Zhang, Selmer groups and divisiblity of Heegner points, preprint (2014).

Ecole´ Polytechnique Fed´ erale´ de Lausanne, FSB MATHGEOM GR-JET, Batimentˆ MA C3 605, CH-1015 Lausanne, Switzerland

E-mail address: [email protected]

Department of Mathematics, Princeton University, Fine Hall, Washington Road, Prince- ton, NJ 08544-1000, USA

E-mail address: [email protected]

Department of Mathematics, Columbia University, New York, NY 10027, USA

E-mail address: [email protected]