Strengthenings of Mazur’s Conjecture for Higher Heegner Points Xiaoyu Zhang Abstract . We establish quantitative strengthenings of Mazur’s conjecture regarding the non-torsion property of higher Heegner points on modular and Shimura curves, confirming both a vertical version for sufficiently large powers n n and a horizontal version for primes p ≫ 1 p\gg 1 . Distinct from previous strategies by Cornut and Vatsal that relied on Ratner’s theorems on unipotent flows and required restrictive hypotheses on level structures, our approach circumvents these constraints by exploiting the interplay between Galois orbits and Hecke orbits. By quantifying the relative size of these orbits, we reduce the problem to ergodic results concerning the equidistribution of Hecke operators and the classification of joining measures. This method allows for the analysis of simultaneous supersingular reductions without requiring the surjectivity of reduction maps. [columns=3] [columns=2,options= -s index_style_xyz.ist] 2020 Mathematics Subject Classification 11G15, 14K22, 22D40 1. Introduction Heegner points on modular and Shimura curves encode deep arithmetic information, playing a pivotal role in Iwasawa theory, the Birch–Swinnerton-Dyer conjecture, and the Gross–Zagier formula. Let L be an imaginary quadratic field of discriminant D_{L/\mathbb{Q}} , and let N be a positive integer such that every prime factor of N splits in L . There exists an ideal \mathcal{N}\subset\mathcal{O}_{L} satisfying \mathcal{O}_{L}/\mathcal{N}\simeq\mathbb{Z}/N\mathbb{Z} . The modular curve X_{0}(N) parameterizes isogenies of elliptic curves E_{1}\to E_{2} with a cyclic kernel of order N . For each positive integer n , let \mathcal{O}_{n}:=\mathbb{Z}+n\mathcal{O}_{L} denote the order of conductor n in L . Set \mathcal{N}_{n}:=\mathcal{O}_{n}\cap\mathcal{N} and define the Heegner point on X_{0}(N) of conductor n by x[n]:=[\mathbb{C}/\mathcal{O}_{n}\to\mathbb{C}/\mathcal{N}_{n}^{-1}]\in X_{0}(N). It is well known that x[n] is defined over the ring class field L[n] of L of conductor n . Let J_{0}(N) be the Jacobian of X_{0}(N) . For any abelian variety A/\mathbb{Q} arising as a quotient of J_{0}(N) , we denote the induced morphism by \pi\colon X_{0}(N)\to J_{0}(N)\to A. For each n , we set y[n]:=\pi(x[n]) . Fix a prime number p and let L[p^{\infty}]=\bigcup_{n}L[p^{n}] . Let H[p^{\infty}] denote the maximal anticyclotomic \mathbb{Z}_{p} -extension of L ; note that L[p^{\infty}]/H[p^{\infty}] is a finite extension. Mazur conjectured ( [ Maz83 ] ) a non-torsion property for these higher Heegner points, a result subsequently established by Cornut ( [ Cor02 ] ): Theorem 1.1 . There exists an integer n\geq 0 such that \mathrm{Tr}_{L[p^{\infty}]/H[p^{\infty}]}(y[p^{n}])\notin A(H[p^{\infty}])_{\mathrm{tors}}. While Cornut’s theorem proves the existence of non-torsion points, it leaves open the quantitative question of their distribution: for how many integers n is the trace \mathrm{Tr}_{L[p^{\infty}]/H[p^{\infty}]}(y[p^{n}]) non-torsion? In particular, are there infinitely many such n ? In this article, we prove the following strengthening of Mazur’s conjecture under mild hypotheses on p : Theorem 1.2 (Corollary 9.2 ) . Suppose that p is coprime to ND_{L/\mathbb{Q}} . For n\gg 1 , we have \mathrm{Tr}_{L[p^{\infty}]/H[p^{\infty}]}(y[p^{n}])\notin A(H[p^{\infty}])_{\mathrm{tors}}. We refer the reader to Corollary 9.2 for the precise statement. While Mazur’s conjecture naturally extends to Shimura curves, Theorem 1.2 was previously established in that setting by Cornut and Vatsal [ CV07 , Theorem 4.10] only under additional hypotheses on the level structure (assumption (H2) in loc.cit. ). Their proof ( [ CV05 , CV07 ] ) relies on a delicate analysis of the torus action on adelic points of units in the quaternion algebra. Our approach, by contrast, is distinct and circumvents these restrictions. We also formulate and prove a horizontal version of Mazur’s conjecture. Let L[\infty]=\bigcup_{k}L[k] and fix a set of representatives \mathcal{T}\subset\mathrm{Gal}(L[\infty]/L) for the finite quotient \mathrm{Gal}(L[1]/L) of \mathrm{Gal}(L[\infty]/L) . Theorem 1.3 (Corollary 8.4 , Horizontal Mazur’s conjecture) . For any prime p\gg 1 , we have \sum_{\sigma\in\mathcal{T}}\sigma(y[p])\notin A(L[\infty])_{\mathrm{tors}}. Both theorems are consequences of a more general result. Let n=\prod_{i=1}^{s}p_{i}^{e_{i}} be a positive integer coprime to ND_{L/\mathbb{Q}} . Define the density term d(n):=\prod_{i=1}^{s}d(p_{i}^{e_{i}}),\qquad d(p^{e})=\begin{cases}1,&\text{if $p$ is inert in $L$},\\ \frac{p-1}{p+1},&\text{if $p$ splits in $L$}.\end{cases} Fix a prime \ell\gg 1 inert in L 1 1 1 The condition that \ell is inert in L ensures that all Heegner points x[n] have supersingular reduction at \ell . (satisfying auxiliary conditions detailed in § 8 ) and a finite subset \mathcal{T}\subset\mathrm{Gal}(L[\infty]/L) . Theorem 1.4 (Corollary 8.2 ) . There exists a constant c(\pi,\ell,\mathcal{T})\in(0,1) , depending on the morphism \pi , the prime \ell , and the set \mathcal{T} , such that for any n\gg 1 coprime to \ell ND_{L/\mathbb{Q}} satisfying 1-d(n)0 . In Theorem 8.1 , we characterize the following simultaneous supersingular reduction map for n\gg 1 : \mathbf{R}_{\mathcal{T}}\colon\{\nu(x[n])|\nu\in\mathfrak{G}_{n}\}\to\prod_{\sigma\in\mathcal{T}}X_{0}(N)^{\mathrm{ss}}(\overline{\mathbb{F}}_{\ell}),\quad x\mapsto(\sigma(x)^{(\ell)})_{\sigma\in\mathcal{T}}. Here \sigma(x)^{(\ell)} denotes the reduction modulo \ell of \sigma(x) . Under a suitable commensurability condition on \mathcal{T} , we show that \mathbf{R}_{\mathcal{T}} is surjective for n\gg 1 . The strategy of the proof for Theorem 8.1 relies on the interplay between the Hecke orbit T_{n}(x[1]) and the Galois orbit \mathfrak{G}_{n}(x[n]) . We demonstrate in Theorem 7.2 that \mathfrak{G}_{n}(x[n]) is contained in T_{n}(x[1]) . Furthermore, their relative sizes are quantified by d(n) : as 1-d(n) decreases, the Galois orbit occupies a larger proportion of the Hecke orbit. Consequently, assuming 1-d(n)n_{0} coprime to the primes in \mathcal{S}^{\circ} and \mathfrak{N}D_{L/\mathbb{Q}} with 1-d(n)0} on \mathbf{X}^{\prime} , we write \mu^{\prime}_{j_{1},j_{2}} for its projection to \mathbf{X}_{j_{1}}^{\prime}\times\mathbf{X}_{j_{2}}^{\prime}\simeq(\mathbf{X}_{j_{1}}^{\prime})^{2} . Since the composition of \Pi^{\prime} with the projection of \mathbf{X}^{\prime} to \mathbf{X}_{j_{1}}^{\prime}\times\mathbf{X}_{j_{2}}^{\prime} is the diagonal map, by Proposition 4.1 , the measure \mu^{\prime}_{j_{1},j_{2}} is the normalized measure on \Gamma_{j_{1},j_{2}}^{2}\Delta_{j_{1},j_{2}}(\mathbf{G}_{0}^{1}(\mathbb{Q}_{\mathcal{S}^{\prime}\backslash\mathcal{S}}))g^{\prime} for some g^{\prime}\in\mathbf{G}_{j_{1}}^{1}(\mathbb{Q}_{\mathcal{S}^{\prime}\backslash\mathcal{S}})\times\mathbf{G}_{j_{2}}^{1}(\mathbb{Q}_{\mathcal{S}^{\prime}\backslash\mathcal{S}}) . Here \Delta_{j_{1},j_{2}}\colon\mathbf{G}_{0}^{1}(\mathbb{Q}_{\mathcal{S}^{\prime}\backslash\mathcal{S}})\to\mathbf{G}_{j_{1}}^{1}(\mathbb{Q}_{\mathcal{S}^{\prime}\backslash\mathcal{S}})\times\mathbf{G}_{j_{2}}^{1}(\mathbb{Q}_{\mathcal{S}^{\prime}\backslash\mathcal{S}}) is the diagonal map. Along the natural map \mathbf{X}^{\prime}\to\mathbf{X} , the measure \mu^{\prime} on \mathbf{X}^{\prime} is mapped to the measure \mu on \mathbf{X} . By the above discussion, we know that the projection \mu_{j_{1},j_{2}} of \mu to \mathbf{X}_{j_{1}}\times\mathbf{X}_{j_{2}} is the normalized measure on (\Gamma_{j_{1}}\times\Gamma_{j_{2}})\Delta_{j_{1},j_{2}}(\mathbf{G}_{0}(\mathbb{Q}_{\mathcal{S}^{\prime}\backslash\mathcal{S}}))g^{\prime} . It follows that the projection of H to the components (j_{1},j_{2}) is contained in \Delta_{j_{1},j_{2}}(\mathbf{G}_{0}(\mathbb{Q}_{\mathcal{S}^{\prime}\backslash\mathcal{S}}))(Z_{\mathbf{G}_{j_{1}}(\mathbb{Q}_{\mathcal{S}^{\prime}\backslash\mathcal{S}})}\times Z_{\mathbf{G}_{j_{2}}(\mathbb{Q}_{\mathcal{S}^{\prime}\backslash\mathcal{S}})}) . We deduce immediately that j_{1},j_{2}\in\mathcal{T}_{i^{\prime}}^{\prime} for some i^{\prime} . To finish the proof, we need to show that we can take g\in Z_{\mathbf{G}(\mathbb{Q}_{\mathcal{S}^{\prime}\backslash\mathcal{S}})} . For this, consider any j_{1},j_{2}\in\mathcal{T}_{i} . From the above discussion on \mu_{j_{1},j_{2}}^{\prime} , we know that the two projections of \Gamma Hg to the component j_{1} and j_{2} must be equal to each other, so we can take the two components g_{j_{1}} and g_{j_{2}} of g at j_{1} and j_{2} respectively to be equal to each other. In other words, we can assume g\in\prod_{i=1}^{k}\Delta^{\mathcal{T}_{i}}(\mathbf{G}_{0}^{1}(\mathbb{Q}_{\mathcal{S}^{\prime}\backslash\mathcal{S}})) . Thus we have \Gamma HZ_{\mathbf{G}(\mathbb{Q}_{\mathcal{S}^{\prime}\backslash\mathcal{S}})}=\Gamma HgZ_{\mathbf{G}(\mathbb{Q}_{\mathcal{S}^{\prime}\backslash\mathcal{S}})}, which finishes the proof of the theorem. ∎ 5. Shimura varieties of Hodge type We first fix some notation, following closely [ Kis10 , Kis17 , Wor13 , Zha25 ] . We fix a Shimura datum of Hodge type (G,X) with G a connected reductive group over \mathbb{Q} . We also fix a symplectic embedding (G,X)\hookrightarrow(\mathbf{GSp}_{L,\mathbb{Q}},S^{\pm}), where S^{\pm} is the Siegel upper half space of rank equal to \mathrm{rk}(L) . We can and will assume \mathrm{rk}(L)\geq 4 in the following. We fix a compact open subgroup U of G(\mathbb{A}_{f}) and write Sh_{U,\mathbb{C}}=Sh_{U,\mathbb{C}}(G,X) for the Shimura variety attached to the triple (G,X,U) , whose \mathbb{C} -points are given by (6) Sh_{U,\mathbb{C}}(\mathbb{C})=G(\mathbb{Q})\backslash(X\times G(\mathbb{A}_{f})/U), which has a canonical model defined over the reflex field F=F(G,X) . We denote this canonical model by Sh_{U,F} . Similarly, for a compact open subgroup \mathbf{U} of \mathbf{GSp}_{L}(\mathbb{A}_{f}) , we have a canonical model Sh_{\mathbf{U},\mathbb{Q}} over \mathbb{Q} . Suppose that U is contained in \mathbf{U} . Then we have a morphism of Shimura varieties Sh_{U,F}\to Sh_{\mathbf{U},F}=Sh_{\mathbf{U},\mathbb{Q}}\times_{\mathbb{Q}}E. For a place v of F over a rational prime \ell such that G is unramified at \ell and U=U^{\ell}U_{\ell} with U_{\ell} hyperspecial at \ell , Kisin ( [ Kis10 ] ) constructed an integral canonical model of Sh_{U,F} over \mathcal{O}_{(v)} , the localization of \mathcal{O}_{F} at v . We denote this integral model by Sh_{U,\mathcal{O}_{(v)}} . We assume henceforth that \mathbf{U}=\mathbf{U}^{\ell}\mathbf{U}_{\ell} with \mathbf{U}_{\ell} also hyperspecial at \ell . We write \mathcal{A}_{\mathbf{U},\mathcal{O}_{(v)}}\to Sh_{\mathbf{U},\mathcal{O}_{(v)}} for the universal abelian scheme over Sh_{\mathbf{U},\mathcal{O}_{(v)}} and we denote by \mathcal{A}_{K,\mathcal{O}_{(v)}}\to Sh_{U,\mathcal{O}_{(v)}} its pullback along Sh_{U,\mathcal{O}_{(v)}}\to Sh_{\mathbf{U},\mathcal{O}_{(v)}} . We fix a reductive model \mathcal{G} of G over \mathbb{Z}_{(\ell)} such that U_{\ell}=\mathcal{G}(\mathbb{Z}_{\ell}) . Then there is a finite set of tensors s^{(\ell)}=(s_{i}^{(\ell)})_{i}\subset L^{\otimes}_{\mathbb{Z}_{(\ell)}} such that \mathcal{G} is identified with the stabilizer in \mathrm{GL}(L_{\mathbb{Z}_{(\ell)}}) of these tensors s^{(\ell)} via our symplectic embedding G\hookrightarrow\mathbf{GSp}_{L,\mathbb{Q}} ( [ Kis10 , §2.3.1 and §2.3.2] ). 4 4 4 Here L_{\mathbb{Z}_{(\ell)}}^{\otimes} is the direct sum of all \mathbb{Z}_{(\ell)} -modules arising from L_{\mathbb{Z}_{(\ell)}} by taking a finite number of the following operations: duals, tensor products, symmetric powers and exterior powers. For each point x\in Sh_{U,\mathcal{O}_{(v)}}(\overline{\mathbb{Q}}) , Kisin constructed tensors s^{(\ell)}_{\ell,x}=(s^{(\ell)}_{i,\ell,x})_{i}\subset(H_{\mathrm{\acute{e}t}}^{1}(\mathcal{A}_{x},\mathbb{Z}_{\ell})\otimes_{\mathbb{Z}_{\ell}}B_{\mathrm{crys}})^{\otimes} ( [ Kis17 , §1.3.6] ). Here \mathcal{A}_{x} is the specialization of \mathcal{A}_{K,\mathcal{O}_{(v)}} to the point x . For each prime \ell^{\prime}\neq\ell , Kisin also constructed tensors s^{(\ell)}_{\ell^{\prime},x}=(s^{(\ell)}_{i,\ell^{\prime},x})\subset(H^{1}_{\mathrm{\acute{e}t}}(\mathcal{A}_{x},\mathbb{Q}_{\ell^{\prime}}))^{\otimes} ( loc.cit ). Similarly, for each point x\in Sh_{U,\mathcal{O}_{(v)}}(\overline{\mathbb{F}}_{\ell}) , Kisin constructed tensors s^{(\ell)}_{0,x}=(s^{(\ell)}_{i,0,x})_{i}\subset(H^{1}_{\mathrm{crys}}(\mathcal{A}_{x}/W(\overline{\mathbb{F}}_{\ell}))\otimes_{W(\overline{\mathbb{F}}_{\ell})}B_{\mathrm{crys}})^{\otimes} ( [ Kis17 , §1.3.10] ). For each prime \ell^{\prime}\neq\ell , there are tensors s^{(\ell)}_{\ell^{\prime},x}=(s^{(\ell)}_{i,\ell^{\prime},x})_{i}\subset H^{1}_{\mathrm{\acute{e}t}}(\mathcal{A}_{x},\mathbb{Q}_{\ell^{\prime}})^{\otimes} ( [ Kis17 , §1.3.6] ). Similarly, we have tensors s^{(\ell)}_{\ell,x}=(s^{(\ell)}_{i,\ell,x}) ( loc.cit ). We have a natural isomorphism of W(\overline{\mathbb{F}}_{\ell}) , resp. \mathbb{Q}_{\ell^{\prime}} -modules (7) L_{W(\overline{\mathbb{F}}_{\ell})}\simeq H^{1}_{\mathrm{crys}}(\mathcal{A}_{x}/W(\overline{\mathbb{F}}_{\ell})),\,\text{resp. }L_{\mathbb{Q}_{\ell^{\prime}}}\simeq H^{1}_{\mathrm{\acute{e}t}}(\mathcal{A},\mathbb{Q}_{\ell^{\prime}}), taking the tensors s^{(\ell)}_{i} to s^{(\ell)}_{i,0,x} , resp. s^{(\ell)}_{i,\ell^{\prime},x} for all i . These tensors for x\in Sh_{U,\mathcal{O}_{(v)}}(\overline{\mathbb{F}}_{\ell}) are constructed by considering a lifting \widetilde{x}\in Sh_{U,\mathcal{O}_{(v)}}(\overline{\mathbb{Q}}) of x and using \ell^{\prime} -adic comparison theorems ( [ Kis17 , Proposition 1.3.7] ). For x either in Sh_{U,\mathcal{O}_{(v)}}(\overline{\mathbb{Q}}) or in Sh_{U,\mathcal{O}_{(v)}}(\overline{\mathbb{F}}_{\ell}) , we define an algebraic subgroup I_{x} of \mathrm{Aut}_{\mathbb{Q}}(\mathcal{A}_{x}) , consisting of elements fixing the tensors s^{(\ell)}_{i,0,x} and s^{(\ell)}_{i,\ell^{\prime},x} for any \ell^{\prime}\neq\ell . Lemma 5.1 . Let x\in Sh_{U,\mathcal{O}_{(v)}}(\overline{\mathbb{F}}_{\ell}) . If \mathcal{A}_{x} is a supersingular abelian variety, then for any prime \ell^{\prime}\neq\ell , we have an isomorphism of algebraic groups over \mathbb{Q}_{\ell^{\prime}} :: I_{x/\mathbb{Q}_{\ell^{\prime}}}:=I_{x}\times_{\mathbb{Q}}\mathbb{Q}_{\ell^{\prime}}\simeq G_{\mathbb{Q}_{\ell^{\prime}}}. Proof. By definition, for any \mathbb{Q}_{\ell^{\prime}} -algebra R , I_{x/\mathbb{Q}_{\ell^{\prime}}}(R) consists of elements in \mathrm{Aut}_{\mathbb{Q}_{\ell^{\prime}}}(\mathcal{A}_{x})(R) fixing the tensors s^{(\ell)}_{i,0,x} and s^{(\ell)}_{i,\ell^{\prime},x} . By the assumption that \mathrm{rk}(L)\geq 4 , there exists a supersingular elliptic curve E(\ell) over \overline{\mathbb{F}}_{\ell} such that \mathcal{A}_{x} is isogenous to E(\ell)^{\mathrm{rk}(L)/2} (by the theorem of Deligne–Ogus–Shioda). In particular, we have an isomorphism of algebraic groups over \mathbb{Q} : \mathrm{Aut}_{\mathbb{Q}}(\mathcal{A}_{x})\simeq\mathrm{GL}_{\mathrm{rk}(L)/2}(B(\ell)), where B(\ell)=\mathrm{End}_{\overline{\mathbb{F}}_{\ell}}(E(\ell))^{\circ} . Since B(\ell) is unramified at \ell^{\prime}\neq\ell , we have an isomorphism B(\ell)\otimes_{\mathbb{Q}}\mathbb{Q}_{\ell^{\prime}}\simeq\mathrm{M}_{2}(\mathbb{Q}_{\ell^{\prime}}) and thus \mathrm{Aut}_{\mathbb{Q}_{\ell^{\prime}}}(\mathcal{A}_{x})\simeq\mathrm{GL}_{\mathrm{rk}(L)/2}(B(\ell)\otimes\mathbb{Q}_{\ell^{\prime}})\simeq\mathrm{GL}_{\mathrm{rk}(L)}(\mathbb{Q}_{\ell^{\prime}}). Since we have the symplectic embedding G\hookrightarrow\mathbf{GSp}_{L,\mathbb{Q}} , without loss of generality, we can assume that one of the tensors, say, s^{(\ell)}_{i_{0}} , corresponds to the symplectic pairing \langle-,-\rangle on L_{\mathbb{Z}_{(\ell)}} . So using the isomorphisms in ( 7 ), the stabilizer in \mathrm{Aut}_{\mathbb{Q}_{\ell^{\prime}}}(\mathcal{A}_{x}) of the tensor s^{(\ell)}_{i_{0},\ell^{\prime},x} is exactly \mathbf{GSp}_{L/\mathbb{Q}_{\ell^{\prime}}} . It follows that the stabilizer in \mathrm{Aut}_{\mathbb{Q}_{\ell^{\prime}}}(\mathcal{A}_{x}) of all the tensors (s^{(\ell)}_{i,\ell^{\prime},x})_{i} is exactly the group G_{\mathbb{Q}_{\ell^{\prime}}} . Thus we have I_{x/\mathbb{Q}_{\ell^{\prime}}}\simeq G_{\mathbb{Q}_{\ell^{\prime}}} . ∎ We deduce that I_{x} is a form of G . Moreover, if \mathcal{A}_{x} is supersingular, then I_{x}(\mathbb{R}) is a closed subgroup of \mathbf{G}_{\mathcal{A}_{x}}(\mathbb{R}) , which is compact. It follows that I_{x}(\mathbb{R}) is also compact. For a prime number \ell such that U=U^{\ell}U_{\ell} and a point x\in Sh_{U,\mathcal{O}_{(v)}}(K) , where K is an algebraically closed field, we write \mathcal{H}^{G,\ell}(\mathcal{A}_{x}) for the set of points x^{\prime}\in Sh_{U,\mathcal{O}_{(v)}}(K) such that there is a quasi-isogeny \phi\colon\mathcal{A}_{x}\to\mathcal{A}_{x^{\prime}} of degree prime to \ell sending the tensors s^{(\ell)}_{0,x} , resp. s^{(\ell)}_{\ell^{\prime},x} , to the tensors s^{(\ell)}_{0,x^{\prime}} , resp. s^{(\ell)}_{\ell^{\prime},x^{\prime}} (for any \ell^{\prime} ). For x\in Sh_{U,\mathcal{O}_{(v)}}(K) with K=\overline{\mathbb{Q}} or \overline{\mathbb{F}}_{\ell} , by [ Zha25 , (1) & (2)] and [ Kis17 , Corollary 1.4.2 & Propositions 2.1.3, 2.1.5] , we have a natural bijection \Theta_{x}\colon\mathcal{H}^{G,\ell}(\mathcal{A}_{x})\simeq I_{x}(\mathbb{Q})\backslash G(\mathbb{A}_{f}^{(\ell)})U/U. Here we implicitly identify I_{x}(\mathbb{Q}_{\ell^{\prime}}) with G(\mathbb{Q}_{\ell^{\prime}}) for \ell^{\prime}\neq\ell using Lemma 5.1 . Note that the right hand side is in natural bijection with (I_{x}(\mathbb{Q})\bigcap U^{p})\backslash G(\mathbb{Q}_{p})/U_{p} . For K=\overline{\mathbb{Q}} , the point x\in Sh_{U,\mathcal{O}_{(v)}}(K)\subset Sh_{U,\mathcal{O}_{(v)}}(\mathbb{C}) corresponds to an element (h_{x},g_{x})\in X\times G(\mathbb{A}_{f}) under the identification ( 6 ). Then for any g\in G(\mathbb{Q}_{p}) , the point \Theta_{x}^{-1}(g)\in Sh_{U,\mathcal{O}_{(v)}}(K)\subset Sh_{U,\mathcal{O}_{(v)}}(\mathbb{C}) corresponds to the element (h_{x},g_{x}g)\in X\times G(\mathbb{A}_{f}) under ( 6 ). Moreover, for x\in Sh_{U,\mathcal{O}_{(v)}}(\overline{\mathbb{F}}_{\ell}) and a lifting \widetilde{x}\in Sh_{U,\mathcal{O}_{(v)}}(\overline{\mathbb{Q}}) of x , we have the following commutative diagram ( [ Kis17 , Corollary 1.4.12] ) (8) {\lx@inpgf@ignorespaces I_{\widetilde{x}}(\mathbb{Q})\backslash G(\mathbb{A}_{f}^{(\ell)})U/U} {\lx@inpgf@ignorespaces\mathcal{H}^{G,\ell}(\mathcal{A}_{\widetilde{x}})} {\lx@inpgf@ignorespaces Sh_{U,\mathcal{O}_{(v)}}(\mathcal{O}_{\overline{\mathbb{Q}}})} {\lx@inpgf@ignorespaces I_{x}(\mathbb{Q})\backslash G(\mathbb{A}_{f}^{(\ell)})U/U} {\lx@inpgf@ignorespaces\mathcal{H}^{G,\ell}(\mathcal{A}_{x})} {\lx@inpgf@ignorespaces Sh_{U,\mathcal{O}_{(v)}}(\overline{\mathbb{F}}_{\ell})} \scriptstyle{\lx@inpgf@ignorespaces\Theta_{\widetilde{x}}} \scriptstyle{\lx@inpgf@ignorespaces\simeq} \scriptstyle{\lx@inpgf@ignorespaces\mathbf{R}_{\ell}} \scriptstyle{\lx@inpgf@ignorespaces\mathbf{R}_{\ell}} \scriptstyle{\lx@inpgf@ignorespaces\Theta_{x}} \scriptstyle{\lx@inpgf@ignorespaces\simeq} The left vertical arrow is the natural projection map induced by the inclusion I_{\widetilde{x}}(\mathbb{Q})\hookrightarrow I_{x}(\mathbb{Q}) . We consider a point x\in Sh_{U,\mathcal{O}_{(v)}}(\overline{\mathbb{F}}_{\ell}) with \mathcal{A}_{x} supersingular, U_{p}=I_{x}(\mathbb{Z}_{p}) . We write I_{x}^{1} for the derived subgroup of I_{x} and \mathrm{pr}\colon I_{x}\to I_{x}/I_{x}^{1}=:I_{x}^{\mathrm{ab}} for the projection map, where I_{x}^{\mathrm{ab}} is a torus over \mathbb{Q} . We assume (9) \mathrm{pr}(I_{x}(\mathbb{A}_{f}^{(\ell)}))=\mathrm{pr}(I_{x}(\widehat{\mathbb{Z}}^{(\ell)}))\mathrm{pr}(I_{x}(\mathbb{Q})\bigcap U_{\ell}). Then the inclusion I_{x}^{1}\hookrightarrow I_{x} induces a bijection (10) (I_{x}^{1}(\mathbb{Q})\bigcap U_{\ell})\backslash I_{x}^{1}(\mathbb{A}_{f}^{(\ell)})/I_{x}^{1}(\widehat{\mathbb{Z}}^{(\ell)})\simeq(I_{x}(\mathbb{Q})\bigcap U_{\ell})\backslash I_{x}(\mathbb{A}_{f}^{(\ell)})/I_{x}(\widehat{\mathbb{Z}}^{(\ell)}). Since I_{x}^{1}(\widehat{\mathbb{Z}}^{(\ell)}) contains the center of I_{x}^{1}(\mathbb{A}_{f}^{(\ell)}) , we have (11) (I_{x}^{1}(\mathbb{Q})\bigcap U_{\ell})\backslash\mathbf{P}I_{x}^{1}(\mathbb{A}_{f}^{(\ell)})/\mathbf{P}I_{x}^{1}(\widehat{\mathbb{Z}}^{(\ell)})\simeq(I_{x}^{1}(\mathbb{Q})\bigcap U_{\ell})\backslash I_{x}^{1}(\mathbb{A}_{f}^{(\ell)})/I_{x}^{1}(\widehat{\mathbb{Z}}^{(\ell)}). Putting all these together, we have a bijection (12) (I_{x}^{1}(\mathbb{Q})\bigcap U_{\ell})\backslash\mathbf{P}I_{x}^{1}(\mathbb{A}_{f}^{(\ell)})/\mathbf{P}I_{x}^{1}(\widehat{\mathbb{Z}}^{(\ell)})\simeq(I_{x}(\mathbb{Q})\bigcap U_{\ell})\backslash\mathbf{P}I_{x}(\mathbb{A}_{f}^{(\ell)})/\mathbf{P}I_{x}(\widehat{\mathbb{Z}}^{(\ell)}). We write I_{x}^{1,\mathrm{ad}} for the adjoint quotient of I_{x}^{1} . Then I_{x}^{1,\mathrm{ad}} is the derived subgroup of I_{x}^{\mathrm{ad}} . By [ Hum69 ] , we know that \mathrm{Aut}(I_{x}^{\mathrm{ad}}(\mathbb{Q}))=I_{x}^{\mathrm{ad}}(\mathbb{Q})/Z_{I_{x}^{\mathrm{ad}}}(\mathbb{Q})\rtimes\mathrm{Aut}_{\mathbb{Q}}(\mathfrak{D}_{x}^{\mathrm{ad}}), where \mathfrak{D}_{x}^{\mathrm{ad}} is the Dynkin diagram of I_{x}^{\mathrm{ad}} and \mathrm{Aut}_{\mathbb{Q}}(\mathfrak{D}_{x}^{\mathrm{ad}}) is the group of automorphisms over \mathbb{Q} of \mathfrak{D}_{x}^{1,\mathrm{ad}} . We fix a place v(\ell) of \overline{\mathbb{Q}} over \ell and a point x_{0}\in Sh_{U,F}(\overline{\mathbb{Q}}) such that x_{0} has supersingular good reduction at v(\ell) . We also fix a non-empty finite subset \mathcal{T}=\{\sigma_{1},\cdots,\sigma_{r}\} of Z_{I_{x_{0}}}(\mathbb{A}_{f}) . We identify \mathcal{T} with the set \{1,2,\cdots,r\} . Moreover, we write x_{0}^{(\ell)} for the reduction modulo v(\ell) of x_{0} . We write \Gamma=I_{x_{0}}(\mathbb{Q})\bigcap U_{\ell} , \Gamma^{1}=\Gamma\bigcap I_{x_{0}}^{1}(\mathbb{Q}) and for i=1,\cdots,r , \displaystyle\Gamma^{(\ell)} \displaystyle:=I_{x_{0}^{(\ell)}}(\mathbb{Q})\bigcap U_{\ell},\quad\Gamma_{i}^{(\ell)}:=\sigma_{i}^{-1}\Gamma\sigma_{i}, \displaystyle\Gamma^{(\ell),1} \displaystyle:=\Gamma\bigcap I_{x_{0}^{(\ell)}}^{1}(\mathbb{Q}),\quad\Gamma^{(\ell),1}_{i}:=\sigma_{i}^{-1}\Gamma^{1}\sigma_{i}. Then we have a partition \mathcal{T}=\bigsqcup_{i=1}^{k}\mathcal{T}_{i}, such that \sigma,\sigma^{\prime}\in\mathcal{T}_{i} if and only if \Gamma_{\sigma} and \Gamma_{\sigma^{\prime}} are commensurable. We have the following simultaneous reduction map of the p -adic Hecke orbit of \mathcal{A}_{x_{0}} : \mathbf{R}_{\mathcal{T}}^{G}\colon\mathcal{H}^{G,\ell}(\mathcal{A}_{x_{0}})\to\prod_{\sigma\in\mathcal{T}}\mathcal{H}^{G,\ell}(\mathcal{A}_{(\sigma x_{0})^{(\ell)}}),\quad\mathcal{A}_{x_{0}^{\prime}}\mapsto(\mathcal{A}_{(\sigma x_{0}^{\prime})^{(\ell)}})_{\sigma\in\mathcal{T}}. Moreover, assuming ( 9 ) and using ( 8 ), ( 10 ), ( 11 ), and ( 12 ), we have the following commutative diagram: (13) {\lx@inpgf@ignorespaces\mathcal{H}^{G,\ell}(\mathcal{A}_{x_{0}})} {\lx@inpgf@ignorespaces\prod_{\sigma\in\mathcal{T}}\mathcal{H}^{G,\ell}(\mathcal{A}_{(\sigma x_{0})^{(\ell)}})} {\lx@inpgf@ignorespaces\mathbf{P}\Gamma\backslash\mathbf{P}G(\mathbb{A}_{f}^{(\ell)})/\mathbf{P}U^{(\ell)}} {\lx@inpgf@ignorespaces\prod_{i=1}^{r}\mathbf{P}\Gamma_{i}^{(\ell)}\backslash\mathbf{P}G(\mathbb{A}_{f}^{(\ell)})/\mathbf{P}U^{(\ell)}} {\lx@inpgf@ignorespaces\mathbf{P}\Gamma^{1}\backslash\mathbf{P}G^{1}(\mathbb{A}_{f}^{(\ell)})/\mathbf{P}U^{(\ell),1}} {\lx@inpgf@ignorespaces\prod_{i=1}^{r}\mathbf{P}\Gamma^{(\ell),1}_{i}\backslash\mathbf{P}G^{1}(\mathbb{A}_{f}^{(\ell)})/\mathbf{P}U^{(\ell),1}} \scriptstyle{\lx@inpgf@ignorespaces\mathbf{R}^{G}_{\mathcal{T}}} \scriptstyle{\lx@inpgf@ignorespaces\simeq} \scriptstyle{\lx@inpgf@ignorespaces\simeq} \scriptstyle{\lx@inpgf@ignorespaces\mathrm{pr}_{\mathcal{T}}} \scriptstyle{\lx@inpgf@ignorespaces\simeq} \scriptstyle{\lx@inpgf@ignorespaces\simeq} \scriptstyle{\lx@inpgf@ignorespaces\mathrm{pr}_{\mathcal{T}}^{1}} Here U^{(\ell),1}=U^{(\ell)}\bigcap G^{1}(\mathbb{A}_{f}^{(\ell)}) , and the horizontal maps \mathrm{pr}_{\mathcal{T}} and \mathrm{pr}_{\mathcal{T}}^{1} are the simultaneous projection maps. For each \mathcal{T}_{i} , we define a map \widetilde{\Delta}^{\mathcal{T}_{i}}\colon\mathcal{H}^{G,\ell}(\mathcal{A}_{x_{0}^{(\ell)}})\to\prod_{\sigma\in\mathcal{T}_{i}}\mathcal{H}^{G,\ell}(\mathcal{A}_{(\sigma x_{0})^{(\ell)}}). This corresponds, under the commutative diagram ( 13 ), to the following twisted diagonal map: \Delta^{\mathcal{T}_{i}}\colon\mathbf{P}\Gamma\backslash\mathbf{P}G(\mathbb{A}_{f}^{(\ell)})/\mathbf{P}U^{(\ell)}\to\prod_{\sigma\in\mathcal{T}_{i}}\mathbf{P}\Gamma_{\sigma}\backslash\mathbf{P}G(\mathbb{A}_{f}^{(\ell)})/\mathbf{P}U^{(\ell)},\quad\overline{g}\mapsto(\overline{\sigma^{-1}g})_{\sigma\in\mathcal{T}_{i}}. By Theorem 4.2 , we have Theorem 5.2 . Assume the following: (1) ( 9 ) holds; (2) G^{1} is simple and simply connected; (3) there exist two primes \ell_{1},\ell_{2} different from \ell such that G(\mathbb{Q}_{\ell_{1}}) and G(\mathbb{Q}_{\ell_{2}}) are both non-compact and, modulo the center, they are both simple and non-commutative; (4) set \mathcal{S}=\{\ell\} , \mathcal{S}^{\prime}=\{\ell,\ell_{1},\ell_{2}\} , and \mathbf{G}_{0}=\cdots=\mathbf{G}_{r}=\mathbf{P}I_{x_{0}} ; then ( 3 ) holds. For any sequence (a_{n})_{n} in G(\mathbb{A}_{f}^{(\ell)}) such that \deg(a_{n})\to+\infty , the sequence (T_{a_{n}}(x_{0}))_{n} equidistributes on the following subset of \prod_{\sigma\in\mathcal{T}}\mathcal{H}^{G,\ell}(\mathcal{A}_{x_{0}^{(\ell)}}) : \prod_{i=1}^{k}\widetilde{\Delta}^{\mathcal{T}_{i}}(\mathcal{H}^{G,\ell}(\mathcal{A}_{x_{0}^{(\ell)}})). 6. Heegner points on Shimura curves In this section, we define Heegner points on Shimura curves. We first fix some notation. We fix an imaginary quadratic number field L , and an ideal \mathfrak{N} of \mathbb{Z} that is coprime to the discriminant D_{L/\mathbb{Q}} of L . We write \epsilon_{L}\colon\mathbb{A}_{\mathbb{Q}}^{\times}/\mathbb{Q}^{\times}\to\{\pm 1\} for the quadratic character attached to L/\mathbb{Q} . We assume the following weak Heegner hypothesis:: (wH) . \epsilon_{L}(\mathfrak{N})=1 . Under Hypothesis (wH) , there is a unique indefinite quaternion algebra B over \mathbb{Q} that is ramified exactly at the primes p of \mathbb{Q} that are inert in L with \mathrm{ord}_{p}(\mathfrak{N}) odd. We write \mathfrak{N}_{B} for the product of rational primes where B is ramified. We fix an isomorphism B\otimes_{\mathbb{Q}}\mathbb{R}\simeq\mathrm{M}_{2}(\mathbb{R}). We view G=B^{\times} as an algebraic group over \mathbb{Q} . We write \mathfrak{H}^{\pm}=\mathbb{C}\setminus\mathbb{R} for the union of the upper and lower half-planes. We let G(\mathbb{R})\simeq\mathrm{GL}_{2}(\mathbb{R}) act on \mathfrak{H}^{\pm} by fractional linear transformations. We then write U_{\infty} for the stabilizer in G(\mathbb{R}) of i\in\mathfrak{H}^{\pm} . Thus we have an identification G(\mathbb{R})/U_{\infty}\simeq\mathfrak{H}^{\pm}. We have a section of the natural projection map G(\mathbb{R})\to\mathfrak{H}^{\pm} : s\colon\mathfrak{H}^{\pm}\to G(\mathbb{R}),\quad x+iy\mapsto\begin{pmatrix}y&x\\ 0&1\end{pmatrix}. By construction, for any place v of \mathbb{Q} , if B is ramified at v , then L\otimes_{\mathbb{Q}}\mathbb{Q}_{v} is a field. Thus we have an embedding of \mathbb{Q} -algebras \kappa\colon L\hookrightarrow B. Moreover, there is a unique point w(\kappa)\in\mathfrak{H}^{\pm} that is fixed by \kappa(x) for any x\in L^{\times} . We normalize \kappa in such a way that for any x\in L^{\times} , we have \kappa(x)\begin{pmatrix}w(\kappa)\\ 1\end{pmatrix}=x\begin{pmatrix}w(\kappa)\\ 1\end{pmatrix}, where on the left \kappa(x) is viewed as a matrix in \mathrm{GL}_{2}(\mathbb{R}) while on the right x is viewed as a scalar in \mathbb{C}^{\times}=(L\otimes_{\mathbb{Q}}\mathbb{R})^{\times} ( [ How04 , §1.2] ). We fix a maximal order \mathcal{O}_{B} of B containing \kappa(\mathcal{O}_{L}) and write R=\kappa(\mathcal{O}_{L})+\kappa(\mathfrak{N})\mathcal{O}_{B} . Then R has reduced discriminant equal to \mathfrak{N} . Under the identification B(\mathbb{A}_{\mathbb{Q},f})^{\times}=G(\mathbb{A}_{\mathbb{Q},f}) , we write U for the image of (R\otimes_{\mathbb{Z}}\widehat{\mathbb{Z}})^{\times} in G(\mathbb{A}_{\mathbb{Q},f}) . The complex Shimura curve of level U is given by \displaystyle Sh_{U}(\mathbb{C}) \displaystyle=G(\mathbb{Q})\backslash G(\mathbb{A}_{\mathbb{Q}})/Z_{G}(\mathbb{A}_{\mathbb{Q}})U_{\infty}U\bigsqcup\{\text{cusps}\} \displaystyle=G(\mathbb{Q})\backslash\left(\mathfrak{H}^{\pm}\times G(\mathbb{A}_{\mathbb{Q},f})/Z_{G}(\mathbb{A}_{\mathbb{Q},f})U_{f}\right)\bigsqcup\{\text{cusps}\}. It has a canonical model Sh_{U,\mathbb{Q}} over \mathbb{Q} , which is smooth, connected, and projective. The set of cusps is empty if and only if B=\mathrm{M}_{2}(\mathbb{Q}) . It is well known that (G,\mathfrak{H}^{\pm}) is a Shimura datum of Hodge type ( [ Mum69 ] ). For (z,g)\in\mathfrak{H}^{\pm}\times G(\mathbb{A}_{\mathbb{Q},f}) , we write [(z,g)] for its image in Sh_{U}(\mathbb{C}) . We let the normalizer of U in G(\mathbb{A}_{\mathbb{Q},f}) act on Sh_{U}(\mathbb{C}) by the formula J(u)[(z,g)]:=[(z,gu^{-1})]. Write \mathrm{Nrm}\colon G(\mathbb{A}_{\mathbb{Q}})\to\mathbb{A}_{\mathbb{Q}}^{\times} for the reduced norm. Then the set \pi_{0}(Sh_{U}(\mathbb{C})) of connected components of Sh_{U}(\mathbb{C}) can be identified with \mathbb{Q}^{\times}\backslash\mathbb{A}_{\mathbb{Q}}^{\times}/\mathrm{Nrm}(Z_{G}(\mathbb{A}_{\mathbb{Q}})U_{\infty}U) . Write \mathbb{Q}_{U}/\mathbb{Q} for the abelian extension such that, under Artin reciprocity, there is an isomorphism \mathrm{Gal}(\mathbb{Q}_{U}/\mathbb{Q})\simeq\mathbb{Q}^{\times}\backslash\mathbb{A}_{\mathbb{Q}}^{\times}/\mathrm{Nrm}(Z_{G}(\mathbb{A}_{\mathbb{Q}})U_{\infty}U). In what follows, we identify these two groups and define a map \mathrm{GNrm}\colon G(\mathbb{A}_{\mathbb{Q},f})\to\mathrm{Gal}(\mathbb{Q}_{U}/\mathbb{Q}),\quad g\mapsto\mathrm{Nrm}(g)^{-1}. We let \mathrm{Gal}(\mathbb{Q}_{U}/\mathbb{Q}) act on \pi_{0}(Sh_{U}(\mathbb{C})) via \mathrm{GNrm} ( [ How04 , §1.2] ): {\lx@inpgf@ignorespaces Sh_{U}(\mathbb{C})} {\lx@inpgf@ignorespaces Sh_{U}(\mathbb{C})} {\lx@inpgf@ignorespaces\pi_{0}(Sh_{U}(\mathbb{C}))} {\lx@inpgf@ignorespaces\pi_{0}(Sh_{U}(\mathbb{C}))} \scriptstyle{\lx@inpgf@ignorespaces J(u)} \scriptstyle{\lx@inpgf@ignorespaces\mathrm{Nrm}} \scriptstyle{\lx@inpgf@ignorespaces\mathrm{Nrm}} \scriptstyle{\lx@inpgf@ignorespaces\mathrm{GNrm}(u)} We view T=L^{\times} as an algebraic group over \mathbb{Q} , which is also a maximal torus of G via the embedding \kappa . Then we put \mathrm{CM}_{L}=T(\mathbb{Q})\backslash G(\mathbb{A}_{\mathbb{Q},f})/Z_{G}(\mathbb{A}_{\mathbb{Q},f}), the set of CM points for L . We have a map \mathrm{CM}_{L}\to Sh_{U}(\mathbb{C}),\quad g\mapsto[(w(\kappa),g)]. The image of this map is the set of CM points on Sh_{U}(\mathbb{C}) . We let \mathrm{Gal}(L^{\mathrm{ab}}/L) act on the set of CM points by the formula (14) [(w(\kappa),g)]^{[s,L]}:=[(w(\kappa),s\cdot g)],\quad\forall s\in T(\mathbb{A}_{\mathbb{Q},f}). Here [-,L]\colon T(\mathbb{Q})\backslash T(\mathbb{A}_{\mathbb{Q},f})\simeq\mathrm{Gal}(L^{\mathrm{ab}}/L) is the Artin map. It is easy to see that Z_{G}(\mathbb{A}_{\mathbb{Q},f})=\mathbb{A}_{\mathbb{Q},f}^{\times} acts trivially on any CM point. For any CM point x=[(w(\kappa),g)] , the endomorphism ring of x is the preimage of R\otimes_{\mathbb{Z}}\widehat{\mathbb{Z}} under the map L\to B\otimes_{\mathbb{Z}}\widehat{\mathbb{Z}} sending h to g^{-1}hg . This endomorphism ring is an order of L of the form \mathcal{O}_{\mathfrak{C}_{x}}=\mathbb{Z}+\mathfrak{C}_{x}\mathcal{O}_{L} for some integral ideal \mathfrak{C}_{x}\subset\mathbb{Z} (the conductor of x ). We write T[\mathfrak{C}_{x}]=\kappa(\widehat{\mathcal{O}}_{\mathfrak{C}_{x}}^{\times})\subset T(\mathbb{A}_{\mathbb{Q},f}) . We write L[\mathfrak{C}] for the ring class field of L , i.e., the abelian extension of L corresponding to T[\mathfrak{C}]Z_{G}(\mathbb{A}_{\mathbb{Q},f}) via class field theory. This is a Galois extension of \mathbb{Q} , and it is the field of definition of x . For each rational prime \ell coprime to \mathfrak{N}D_{L/\mathbb{Q}} , we fix an isomorphism (15) R_{\ell}\simeq\mathrm{M}_{2}(\mathbb{Z}_{\ell}) such that \kappa(\mathcal{O}_{L,\ell}) is identified with the set of diagonal matrices \begin{pmatrix}x&0\\ 0&y\end{pmatrix} , respectively \begin{pmatrix}x&yu_{\ell}\\ y&x\end{pmatrix} for x,y\in\mathbb{Z}_{\ell} , if \ell splits in L , respectively if \ell is inert in L . Here u_{\ell} is a fixed non-square element in \mathbb{Z}_{\ell}^{\times} . This extends to an isomorphism B_{\ell}\simeq\mathrm{M}_{2}(F_{\ell}) . Fix a uniformizer \varpi_{\ell} of \mathbb{Q}_{\ell} , and for each integer k\geq 0 , define an element h[\ell^{k}] in B_{\ell} whose image in \mathrm{M}_{2}(\mathbb{Q}_{\ell}) is given by \begin{cases}\begin{pmatrix}\varpi_{\ell}^{k}&1\\ 0&1\end{pmatrix},&if $\ell$ splits in $L$;\\ \begin{pmatrix}\varpi_{\ell}^{k}&0\\ 0&1\end{pmatrix},&if $\ell$ is inert in $L$.\end{cases} We view h[\ell^{k}] as elements in G(\mathbb{A}_{\mathbb{Q},f}) via the embedding G(\mathbb{Q}_{\ell})\hookrightarrow G(\mathbb{A}_{\mathbb{Q},f}) . For any positive integer n=\prod_{i=1}^{r}\ell_{i}^{k_{i}} coprime to \mathfrak{N}D_{L/\mathbb{Q}} , we define h[n]:=\prod_{i=1}^{r}h[\ell_{i}^{k_{i}}]. We view h[n] also as elements in \mathrm{CM}_{L} and write x[n] for its image in Sh_{U}(\mathbb{C}) . It is well known that these x[n] are defined over the maximal abelian extension L^{\mathrm{ab}} of L . Moreover, the action of \mathrm{Gal}(L^{\mathrm{ab}}/L) on these x[n] agrees with the action in ( 14 ) ( [ How04 , Theorem 1.2.2] ). We decompose Sh_{U,\mathbb{Q}}\times_{\mathbb{Q}}\mathbb{Q}_{U} into geometric components Sh_{U,\mathbb{Q}}\times_{\mathbb{Q}}\mathbb{Q}_{U}=\bigsqcup_{i}Sh_{U,\mathbb{Q},i} , with each Sh_{U,\mathbb{Q},i} geometrically irreducible. Moreover, for any extension F/\mathbb{Q} such that Sh_{U,\mathbb{Q}}(F)\neq\emptyset , we must have \mathbb{Q}_{U}\subset F . We define J_{U,\mathbb{Q}}:=\mathrm{Res}^{\mathbb{Q}_{U}}_{\mathbb{Q}}(\mathrm{Jac}(Sh_{U,\mathbb{Q},i_{0}})) for some fixed component Sh_{U,\mathbb{Q},i_{0}} . Here \mathrm{Jac}(Sh_{U,\mathbb{Q},i_{0}}) is the Jacobian of Sh_{U,\mathbb{Q},i_{0}} . Then J_{U,\mathbb{Q}} is an abelian variety defined over \mathbb{Q} , which has good reduction away from \mathfrak{N} . There is a unique element \mathfrak{h}\in\mathrm{Pic}(Sh_{U,\mathbb{Q}}) (up to a constant multiple) whose degree on each geometric component Sh_{U,\mathbb{Q},i} is constant and which satisfies T_{a(m)}\mathfrak{h}=\mathrm{deg}(T_{a(m)})\mathfrak{h} for every Hecke operator T_{a(m)} with a positive integer m coprime to \mathfrak{N}D_{L/\mathbb{Q}} . This is called the Hodge class, and on each geometric component \mathfrak{h} is just the canonical divisor. We write \mathfrak{h}_{i} for the restriction of \mathfrak{h} to the component Sh_{U,\mathbb{Q},i} ; then we have a unique morphism defined over \mathbb{Q} Sh_{U,\mathbb{Q}}\to J_{U,\mathbb{Q}}, which, on complex points, takes x_{i}\in Sh_{U,\mathbb{Q}}(\mathbb{C}) to the divisor dx_{i}-\mathfrak{h}_{i}\in J_{U,\mathbb{Q}}(\mathbb{C}) . Let A/\mathbb{Q} be an abelian variety with a surjective morphism of abelian varieties \alpha\colon J_{U,\mathbb{Q}}\to A . Composed with the map Sh_{U,\mathbb{Q}}\to J_{U,\mathbb{Q}} , we have a morphism over \mathbb{Q} \pi\colon Sh_{U,\mathbb{Q}}\to J_{U,\mathbb{Q}}\to A. Since Sh_{U,\mathbb{Q}} is complete and connected, the morphism \pi is either finite or constant, the latter of which is impossible since \alpha is surjective. Thus \pi is a finite morphism ( cf. [ CV07 , Lemma 3.9] ). We write y[n]=\pi(x[n])\in A(L[n]). As in the introduction, we write L[\infty]=\bigcup_{k}L[k] . 7. Relation between Hecke orbits and Galois orbits In this section, we study the relationship between Hecke orbits and Galois orbits. We keep the notation as in § 6 . We fix a prime \ell ( \mathcal{S}=\{\ell\} in the notation of § 3 ) that is inert in L , and a place v(\ell) of \overline{\mathbb{Q}} over \ell . We fix the point x_{0}=x[1]\in Sh_{U,\mathbb{Q}}(\overline{\mathbb{Q}}) and assume that it has supersingular good reduction at v(\ell) . We fix a non-empty finite subset \mathcal{T}=\{\sigma_{1},\cdots,\sigma_{r}\} of Z_{I_{x_{0}}}(\mathbb{Q})\backslash Z_{I_{x_{0}}}(\mathbb{A}_{f})=I_{x_{0}}(\mathbb{Q})\backslash I_{x_{0}}(\mathbb{A}_{f})=L^{\times}\backslash\widehat{L}^{\times}. We identify \mathcal{T} with its image in \mathrm{Gal}(L[\infty]/L) via the Artin reciprocity map: (16) \mathrm{Art}\colon\mathrm{Gal}(L[\infty]/L)\simeq\widehat{L}^{\times}/L^{\times}. We have a partition \mathcal{T}=\bigsqcup_{i=1}^{k}\mathcal{T}_{i} as in § 5 . We set \mathbf{G}_{0}=I_{x_{0}^{(\ell)}} , an algebraic group over \mathbb{Q} . We fix isomorphisms (17) \mathbf{G}_{0}(\mathbb{Z}_{q})\simeq G(\mathbb{Z}_{q}),\quad q\neq\ell. Consider a positive integer n coprime to \ell\mathfrak{N}D_{L/\mathbb{Q}} . We first consider Galois orbits. There are the following cases: (1) n=q^{e} is a prime power with q a prime number that splits in L . We fix a set S_{n} of representatives in \mathbb{Z}_{q}^{\times} of the quotient set (\mathbb{Z}_{q}/n\mathbb{Z}_{q})^{\times} . Note that S_{n} can be viewed as a set of representatives of the quotient (\mathcal{O}_{L,q}/n\mathcal{O}_{L,q})^{\times}/(\mathbb{Z}_{q}/n\mathbb{Z}_{q})^{\times} because \mathcal{O}_{L,q}\simeq\mathbb{Z}_{q}^{2} as \mathbb{Z}_{q} -algebras. Moreover, \#S_{n}=q^{e-1}(q-1) . Under the identification ( 15 ), we define a subset \mathfrak{G}_{n} of \mathrm{Gal}(L[\infty]/L) such that under the Artin map ( 16 ), we have \mathrm{Art}(\mathfrak{G}_{n})=L^{\times}\cdot\left\{\begin{pmatrix}a&0\\ 0&1\end{pmatrix}|a\in S_{n}\right\}. (2) n=q^{e} is a prime power with q a prime number that is inert in L . We fix a set S_{n} of representatives in \mathcal{O}_{L,q}^{\times} of the quotient (\mathcal{O}_{L,q}/n\mathcal{O}_{L,q})^{\times}/(\mathbb{Z}_{q}/n\mathbb{Z}_{q})^{\times} . Moreover, \#S_{n}=q^{e-1}(q+1) . We define a subset \mathfrak{G}_{n} of \mathrm{Gal}(L[\infty]/L) such that \mathrm{Art}(\mathfrak{G}_{n})=L^{\times}\cdot S_{n}. Explicitly, we can take S_{n} to be (under the identification ( 15 )):: S_{n}=\left\{\begin{pmatrix}1&bu_{q}\\ b&1\end{pmatrix}|b=1,\cdots,q^{e}\right\}\bigsqcup\left\{\begin{pmatrix}pau_{q}&u_{q}\\ 1&pau_{q}\end{pmatrix}|a=1,\cdots,q^{e-1}\right\}. (3) In general, for a prime factorization n=\prod_{i=1}^{s}p_{i}^{e_{i}} , we define S_{n}=\prod_{i=1}^{s}S_{p_{i}^{e_{i}}} , viewed as a subset of \widehat{\mathcal{O}}_{L}^{\times} . We then define a subset \mathfrak{G}_{n} of \mathrm{Gal}(L[\infty]/L) such that \mathrm{Art}(\mathfrak{G}_{n})=L^{\times}\cdot S_{n}. We next consider Hecke orbits. As above, there are the following cases:: (1) n=q^{e} is a prime power. We define a(n):=\begin{pmatrix}n&0\\ 0&1\end{pmatrix}\in\mathrm{GL}_{2}(\mathbb{Q}_{q})\simeq G(\mathbb{Q}_{q})\hookrightarrow G(\mathbb{A}_{f}^{(\ell)}). By definition, the double coset decomposes into right cosets G(\mathbb{Z}_{q})a(n)G(\mathbb{Z}_{q})=\bigsqcup_{g\in S_{n}^{\prime}}a(n)G(\mathbb{Z}_{q}), where S_{n}^{\prime} is a set of representatives in G(\mathbb{Z}_{q}) of the quotient G(\mathbb{Z}_{q})/(G(\mathbb{Z}_{q})\bigcap a(n)G(\mathbb{Z}_{q})a(n)^{-1}) . Note that \#S_{n}^{\prime}=q^{e-1}(q+1)=\deg(a(n)) ( [ Shi71 , Theorem 3.24] ). We have Lemma 7.1 . For any distinct g,g^{\prime}\in S_{n} , the right cosets gh[n]G(\mathbb{Z}_{q}) and g^{\prime}h[n]G(\mathbb{Z}_{q}) are distinct right cosets in the decomposition G(\mathbb{Z}_{q})a(n)G(\mathbb{Z}_{q}) (here we use the identifications ( 17 )). Proof. We first show that the two cosets are distinct. If q splits in L , write g=\begin{pmatrix}a&0\\ 0&1\end{pmatrix} and g^{\prime}=\begin{pmatrix}a^{\prime}&0\\ 0&1\end{pmatrix} . Then we have gh[n]G(\mathbb{Z}_{q})=\begin{pmatrix}a\varpi_{q}^{e}&a\\ 0&1\end{pmatrix}G(\mathbb{Z}_{q})=\begin{pmatrix}n&a\\ 0&1\end{pmatrix}G(\mathbb{Z}_{q}),\quad g^{\prime}h[n]G(\mathbb{Z}_{q})=\begin{pmatrix}n&a^{\prime}\\ 0&1\end{pmatrix}G(\mathbb{Z}_{q}). Since \begin{pmatrix}n&a\\ 0&1\end{pmatrix}^{-1}\begin{pmatrix}n&a^{\prime}\\ 0&1\end{pmatrix}=\begin{pmatrix}1&(a^{\prime}-a)/n\\ 0&1\end{pmatrix}\notin G(\mathbb{Z}_{q}), these right cosets are indeed distinct. If q is inert in L , write g^{-1}g^{\prime}=\begin{pmatrix}a&b\\ c&d\end{pmatrix} . By definition of S_{n} , one checks that b/n\notin\mathbb{Z}_{q} . Then we have (gh[n])^{-1}(g^{\prime}h[n])=\begin{pmatrix}a&b/n\\ cn&d\end{pmatrix}\notin G(\mathbb{Z}_{q}). It follows again that these two right cosets are distinct. Next we show that both cosets lie in G(\mathbb{Z}_{q})a(n)G(\mathbb{Z}_{q}) . By the Cartan decomposition for G(\mathbb{Q}_{q}) , we can characterize G(\mathbb{Z}_{q})a(n)G(\mathbb{Z}_{q}) as the subset of G(\mathbb{Q}_{q}) , consisting of matrices A=\begin{pmatrix}a&b\\ c&d\end{pmatrix} with a,b,c,d\in\mathbb{Z}_{q} such that \det(A)\in n\mathbb{Z}_{q}^{\times} and a,b,c,d are not simultaneous divisible by q . Using this criterion, it follows easily that gh[n]G(\mathbb{Z}_{q}) and g^{\prime}h[n]G(\mathbb{Z}_{q}) both lie in G(\mathbb{Z}_{q})a(n)G(\mathbb{Z}_{q}) . ∎ In particular, if q is inert in L , then we can take S_{n}^{\prime} to be S_{n} ; if q splits in L , we can take S_{n}^{\prime} to contain S_{n} . (2) In general, for a prime factorization n=\prod_{i=1}^{s}p_{i}^{e_{i}} , we have ( [ Shi71 , Theorem 3.24] ) T_{a(n)}=\prod_{i=1}^{s}T_{a(p_{i}^{e_{i}})}. It follows from the previous case that for any distinct g,g^{\prime}\in S_{n} , the right cosets gh[n]U and g^{\prime}h[n]U are distinct, and both lie in Uh[n]U (note that for any p_{i} , U_{p_{i}}=G(\mathbb{Z}_{p_{i}}) under the identification ( 17 ) because n is coprime to \ell\mathfrak{N}D_{L/\mathbb{Q}} ). For a positive integer n=\prod_{i=1}^{s}p_{i}^{e_{i}} coprime to \ell\mathfrak{N}D_{L/\mathbb{Q}} , we define a rational number d(n):=\frac{\#S_{n}}{\#S_{n}^{\prime}}=\frac{\#S_{n}}{\deg(a(n))}\in(0,1]. In particular, d(n)=1 if and only if all the prime factors p_{i} are inert in L . In summary, we have obtained:: Theorem 7.2 . For any n coprime to \ell\mathfrak{N}D_{L/\mathbb{Q}} , the Hecke orbit T_{n}(x[1]^{(\ell)}) contains the Galois orbit \mathfrak{G}_{n}(x[n]^{(\ell)}) . Moreover, we have d(n)=\frac{\#\mathfrak{G}_{n}}{\deg(a(n))} . 8. Variants of Mazur’s conjecture-1 Recall that we have fixed a moduli parametrization \pi\colon Sh_{U,\mathbb{Q}}\to A . This is a finite morphism ( cf. [ CV07 , Theorem 3.9] or [ Zha25 , §9] ). By [ Cor02 , Lemma 4.1] , we know that A(E[\infty])_{\mathrm{tors}} is finite, say, of cardinality t_{1} . We write t_{2} for an upper bound for the cardinalities of the fibers of the map \pi\colon Sh_{U}\to A , which is finite since \pi is a finite morphism. We write Sh_{U}^{\mathrm{ss}}(\mathbb{F}_{v(\ell)}) for the supersingular locus of Sh_{U}(\mathbb{F}_{v(\ell)}) , where \mathbb{F}_{v(\ell)}=\overline{\mathbb{F}}_{\ell} is the residue field of \mathcal{O}_{\overline{\mathbb{Q}}} at the place v(\ell) . Then we have bijections ( [ CV05 , Corollary 3.13 & §3.2.5] ) Sh_{U}^{\mathrm{ss}}(\overline{\mathbb{F}}_{\ell})\simeq I_{x_{0}}(\mathbb{Q})\backslash G(\mathbb{A}_{f}^{\ell})U/U\simeq\mathcal{H}^{G,\ell}(\mathcal{A}_{x[1]^{(\ell)}}). By Eichler’s mass formula, we know that as \ell\to+\infty , \#Sh_{U}^{\mathrm{ss}}(\overline{\mathbb{F}}_{\ell})\to+\infty . We choose \ell inert in L such that (18) \#Sh_{U}^{\mathrm{ss}}(\overline{\mathbb{F}}_{\ell})>(t_{1}\times r!)^{2\dim(A)}\times t_{2}. In what follows, we fix x_{0}=x[1] and assume that x[1] has supersingular reduction x[1]^{(\ell)} at v(\ell) . For each element g\in I_{x[1]}(\mathbb{Q})\backslash G(\mathbb{A}_{f}^{(\ell)})U , we write c_{g}=c_{g}(\ell) for the cardinality of the stabilizer of g in U (where U acts on the right on I_{x[1]^{(\ell)}}(\mathbb{Q})\backslash G(\mathbb{A}_{f}^{(\ell)})U ). This is finite because I_{x[1]^{(\ell)}}(\mathbb{R}) is compact modulo its center. Choose a set of representatives \{g_{1},\cdots,g_{s}\} of I_{x[1]^{(\ell)}}(\mathbb{Q})\backslash G(\mathbb{A}_{f}^{(\ell)})U/U and set (19) c(\pi,\ell,\mathcal{T}):=\min_{i}\left(\frac{c_{g_{i}}^{-1}}{\sum_{j=1}^{s}c_{g_{j}}^{-1}}\right)^{r}\in(0,1). Here k is the cardinal of the partition \mathcal{T}=\bigsqcup_{i=1}^{k}\mathcal{T}_{i} as in § 5 . We have the following result. Theorem 8.1 . Let \ell and \pi\colon Sh_{U}\to A be as above. For any integer n\gg 1 coprime to \ell\mathfrak{N}D_{L/\mathbb{Q}} with 1-d(n)(t_{1}\times r!)^{2\dim(A)}\times t_{2} . So there exists x,x^{\prime}\in Sh_{U}^{\mathrm{ss}}(\overline{\mathbb{F}}_{\ell}) such that \pi(x)-\pi(x^{\prime})\in A(\overline{\mathbb{F}}_{\ell}) has order not dividing t_{1}\times r! . On the other hand, we have assumed that \sum_{\sigma\in\mathcal{T}}\sigma(y[n]) is torsion; hence both \sum_{\sigma\in\mathcal{T}}\sigma(\nu(y[n])) and \sum_{\sigma\in\mathcal{T}}\sigma(\nu^{\prime}(y[n])) are torsion, and so are their reductions modulo v(\ell) , namely \sum_{\sigma\in\mathcal{T}}\sigma(\nu(y[n]))^{(\ell)} and \sum_{\sigma\in\mathcal{T}}\sigma(\nu^{\prime}(y[n]))^{(\ell)} . In particular, the following element in A(\overline{\mathbb{F}}_{\ell}) , \sum_{\sigma\in\mathcal{T}}\sigma(\nu(y[n]))^{(\ell)}-\sum_{\sigma\in\mathcal{T}}\sigma(\nu^{\prime}(y[n]))^{(\ell)}=\#\mathcal{T}_{i_{0}}(\pi(x)-\pi(x^{\prime})) is a torsion element, whose order necessarily divides t_{1}=\#A(L[\infty])_{\mathrm{tors}} . So \pi(x)-\pi(x^{\prime}) is a torsion element, whose order necessarily divides t_{1}\times r! , 5 5 5 More precisely, we can deduce that the order of \pi(x)-\pi(x^{\prime}) divides t_{1}\times\#\mathcal{T}_{i_{0}} . But for our argument in the proof, the multiple t_{1}\times r! of t_{1}\times\#\mathcal{T}_{i_{0}} suffices. which is a contradiction to the preceding paragraph. This finishes the proof of the corollary. ∎ We then deduce some consequences from the above. We fix a prime p and write L[p^{\infty}]=\bigcup_{k}L[p^{k}] . This is a finite extension over the maximal anticyclotomic \mathbb{Z}_{p} -extension H[p^{\infty}] of L . We take \mathcal{T} to be a set of representatives in \widehat{L}^{\times} of the subgroup \mathrm{Gal}(L[p^{\infty}]/H[p^{\infty}]) of \mathrm{Gal}(L[p^{\infty}]/L) (under the Artin map ( 16 )). Then we have Corollary 8.3 . Let \ell and \pi\colon Sh_{U}\to A be as above. Suppose that p is either inert in L or splits in L with \frac{2}{p+1}\ell\mathfrak{N}D_{L/\mathbb{Q}} . Moreover, 1-d(p)<\frac{2}{p} . For p\gg 1 , 1-d(p)\ll 1 . In particular, we can take p such that 1-d(p)0 such that for any n>n_{0} coprime to all primes in \mathcal{S}^{\circ} and \mathfrak{N}D_{L/\mathbb{Q}} with 1-d(n)1-c , we have (21) \frac{\#(S_{n})^{2}}{\#(S_{n}^{\prime})^{2}}=d(n)^{2}>(1-c)^{2}. We choose n_{0}=n_{0}(\ell) such that for any n>n_{0} coprime to all primes in \mathcal{S}^{\circ} and \mathfrak{N}D_{L/\mathbb{Q}} , we have (22) \frac{f(n)}{\#(S_{n}^{\prime})^{2}}<(1-c)^{2}. Combining ( 21 ) and ( 22 ), it follows that there exists (\underline{h},\underline{h}^{\prime})\in\mathbf{R}_{\mathcal{T}}(\mathfrak{G}_{n}(x[n]))^{2} that is not in X(\ell) . In other words, we can find \nu,\nu^{\prime}\in\mathfrak{G}_{n} such that \sum_{\sigma\in\mathcal{T}}(\sigma(\nu(y[n]))-\sigma(\nu^{\prime}(y[n])))\notin A(L[\infty])_{\mathrm{tors}}. In particular, either \sum_{\sigma\in\mathcal{T}}\sigma(\nu(y[n])) or \sum_{\sigma\in\mathcal{T}}\sigma(\nu^{\prime}(y[n])) is non-torsion in A(L[\infty]) . In either case, \sum_{\sigma\in\mathcal{T}}\sigma(y[n]) is non-torsion, which finishes the proof of the theorem. ∎ An immediate corollary is the following strengthening of Corollary 8.3 . Corollary 9.2 . Let \pi\colon Sh_{U}\to A be as above. Fix a prime p coprime to \mathfrak{N}D_{L/\mathbb{Q}} . Then for k\gg 1 , we have \sum_{\sigma\in\mathcal{T}}\sigma(y[p^{k}])\notin A(L[\infty])_{\mathrm{tors}}. Proof. For \mathcal{S}^{\circ} as above, we apply Theorem 9.1 to \mathcal{S}^{\circ}\backslash\{p\} instead of \mathcal{S}^{\circ} . ∎