Green forms and the arithmetic Siegel-Weil formula (Q1741568)

Around 1997, Kudla started to analyze the relation of the intersection numbers of certain algebraic cycles on Shimura varieties associated to orthogonal groups of signature \((p,2)\) to special values (of derivatives) of Siegel Eisenstein series [\textit{S. S. Kudla}, Ann. Math. (2) 146, No. 3, 545--646 (1997; Zbl 0990.11032)]. Since then, this has grown out to a whole program concerning results and conjectures on special cycles in arithmetic Chow groups. Now, as the authors state, this paper is a contribution to the archimedean aspects of this theory. Building upon previous work of the first author [Adv. Math. 329, 555--589 (2018; Zbl 1387.58007)], they use Quillen's formalism of superconnections as developed by Bismut-Gillet-Soulé to construct the natural Green forms for special cycles defined by Kudla in a different way, in all codimensions, on orthogonal and unitary Shimura varieties. They show that these forms have good functorial properties and are compatible with star products. Next, specializing to compact Shimura varieties of type \(\mathrm{GSpin}(p,2)\) or \(\mathrm{U}(p,1)\), they study these Green forms using the theta correspondence, relating them to Siegel Eisenstein series: Their main theorem comes out as an explicit formula for the local archimedean height of a special cycle in terms of a Fourier coefficient of a special derivative of an Eisenstein series. And, as the authors assure, this result provides compelling evidence for Kudla's conjectural identity, termed the arithmetic Siegel-Weil formula, between special derivatives of Eisenstein series and generating series of arithmetic heights of special cycles. More precisely, they show that the non-holomorphic terms on both sides are equal for these Shimura varieties. Their methods combine Quillen's extension of Chern-Weil theory with the theory of the theta correspondence. This allows to use representation theoretic arguments and the Siegel-Weil formula when computing archimedean local heights. In this way, they intend to give a conceptual explanation for the equality of non-holomorphic terms in Kudla's conjectural identities. The long paper has five sections. For an overview it may be worthwhile to reproduce nearly verbatim some parts of the introducing Section 1.1. Throughout the paper, one treats \(\mathrm{GSpin}(p,2)\) and \(\mathrm{U}(p,q)\) Shimura varieties in parallel; these are referred to as the orthogonal and unitary cases, respectively. One has - \(F\) a totally real field of degree \([F:\mathbb{Q}] = d\), - \(E\) be a CM extension of \(F\) equipped with a fixed CM type, - \(\mathbb{V}\) is a quadratic space over \(F\) in the orthogonal case (resp. a Hermitian space over \(E\) in the unitary case). We assume that there is one archimedean place \(\sigma_1\) such that \(\mathbb{V}_{\sigma_1}\) satisfies the signature condition signature \((\mathbb{V}_{\sigma_1}) = (p,2), p>0\) in the orthogonal case (resp. \(=(p,q), p,q>0\) in the unitary case) and \(\mathbb{V}\) is positive definite at all other archimedean places. Take \(\mathbf{H} = \mathrm{Res}_{F/\mathbb{Q} }\mathrm{GSpin}(\mathbb{V})\) in the orthogonal case, (resp. \(= \mathrm{Res}_{F/\mathbb{Q}}\mathrm{U}(\mathbb{V})\) in the unitary case) and \(\mathbb{D}\) the hermitian symmetric domain attached to \(\mathbf{H}(\mathbb{R})\), concretely, \(\mathbb{D}\) parametrizes oriented negative-definite real planes (resp. negative definite complex \(q\)-dimensional subspaces) in \(\mathbb{V}_{\sigma_1}\) in the orthogonal (resp. unitary) case. For a fixed compact open subgroup \(K\subset \mathbf{H}(\mathbb{A}_f)\), one has - \(X_{\mathbb{V},K}\) the corresponding Shimura variety, which has a canonical model over \(\sigma_1(F)\) (resp. \(\sigma_1(F)\)) and is a finite disjoint union \(\coprod \Gamma\backslash \mathbb{D}\) of quotients of \(\mathbb{D}\) by certain arithmetic subgroups \(\Gamma\subset \mathbf{H}(\mathbb{Q})\), - \(\mathcal{X}_K\) the variety obtained by viewing the canonical model of \(X_{\mathbb{V},K}\) as a variety over Spec\((\mathbb{Q})\) and its complex points - \(\mathcal{X}_K(\mathbb{C}) = \coprod X_{\mathbb{V}[k],K}\) are a finite disjoint union of Shimura varieties attached to \(\mathbb{V}[1] := \mathbb{V}\) and its \textit{nearby} spaces \(\mathbb{V}[k]\) to be explained in Section 4 of the paper. Now, as the central objects of the paper (generalizing the construction of Heegner points on modular curves and Hirzebruch-Zagier cycles on Hilbert modular surfaces), the variety \(\mathcal{X}_K\) is equipped with a family of rational \textit{special cycles} \({Z(T,\varphi_f)}\) as defined by \textit{S. S. Kudla} [Duke Math. J. 86, No. 1, 39--78 (1997; Zbl 0879.11026)], that are parametrized by pairs \((T,\varphi_f)\) consisting of a matrix \(T\in \mathrm{Sym}_r(F)\) (resp. \(T\in \mathrm{Her}_r(E)\)) and a \(K\)-invariant Schwartz function \(\varphi_f\in \mathcal{S}(\mathbb{V}(\mathbb{A}_f)^r)^K\). The irreducible components of the cycle \({Z(T,\varphi_f)}\) on \(X_{\mathbb{V},K}\) (say) admit a complex uniformization by certain complex submanifolds \(\mathbb{D}_\mathbf{v}\) of \(\mathbb{D}\) defined as follows: for a collection of vectors \(\mathbf{v} = (v_1,\dots,v_r) \in \mathbb{V}^r\) satisfying \[ T(\mathbf{v}) := ((1/2) Q(v_i,v_j))_{i,j=1,\dots,r} = T \] take \[ \mathbb{D}_\mathbf{v} := \{z\in \mathbb{D}| z\bot v_i \,\mathrm{for all}\, \,i=1,\dots,r\}. \] so that \(\mathbb{D}_\mathbf{v}\), if non-empty, is a hermitian symmetric subdomain of \(\mathbb{D}\) of codimension \(\tilde{r} := \mathrm{rk}(T )\) in the orthogonal case (resp. \(\tilde{r} := q \cdot \mathrm{rk}(T )\) in the unitary case). As a first step towards defining a Green current for \({Z(T,\varphi_f)}\), construct a current \(\mathfrak{g}^o(\mathbf{v})\) on \(\mathbb{D}\) satisfying the equation \[ dd^c\mathfrak{g}^o(\mathbf{v}) + \delta_{\mathbb{D}_\mathbf{v}}\wedge \Omega_{\mathcal{E}^\vee}^{r-\mathrm{rk}(T)} = [\varphi^o_{\mathrm{KM}}(\mathbf{v})]. \] Here \(\delta_{\mathbb{D}_\mathbf{v}}\) is the current defined by integration along \(\mathbb{D}_\mathbf{v}\) and \(\Omega_{\mathcal{E}^\vee}\) is the top Chern form of the dual of the tautological bundle \(\mathcal{E}\), to be explained in Section 2.2. The form \(\varphi^o_{\mathrm{KM}}(\mathbf{v}) := e^{2\pi \mathrm{tr}(T(\mathbf{v}))}\varphi_{KM}(\mathbf{v})\) is, up to a normalizing factor, the Schwartz form \(\varphi_{KM}(\mathbf{v})\) introduced by \textit{S. S. Kudla} and \textit{J. J. Millson} [Publ. Math., Inst. Hautes Étud. Sci. 71, 121--172 (1990; Zbl 0722.11026)]. In recent work from 2018, already mentioned above, the first author introduced a superconnection \(\nabla_\mathbf{v}\) on \(\mathbb{D}\) and showed that the component of degree \((\tilde{r},\tilde{r})\) of the corresponding Chern form agrees with \(\varphi^o_{\mathrm{KM}}(\mathbf{v})\). This allows to apply general results from Bismut, Gillet and Soulé [\textit{J. M. Bismut} et al., Commun. Math. Phys. 115, No. 1, 49--78, 79--126 (1988; Zbl 0651.32017)] to obtain an explicit natural form \(\nu^o(\mathbf{v})\) satisfying the transgression formula \[ dd^c \nu^o(\sqrt t \mathbf{v}) = -t \frac{d}{dt} \varphi^o_{\mathrm{KM}}(\sqrt t \mathbf{v}) , \quad t \in \mathbb{R}_{>0} \] For non-degenerate \(T\) (i.e., linear independent \(v_1,\dots,v_r\) and \(r = \mathrm{rk}(T)\)), consider \[ \mathfrak{g}^o(\mathbf{v}) := \int_1^\infty \nu^o(\sqrt t \mathbf{v})\frac{dt}{t} \] to get a Green form for \(\mathbb{D}_\mathbf{v}\) in the terminology of Bost, Gillet and Soulé [\textit{J. B. Bost} et al., J. Am. Math. Soc. 7, No. 4, 903--1027 (1994; Zbl 0973.14013), \S 1.1]. When \(T\) is degenerate, one has to use regularization of the integral and and more work to be found in Section 2.6.1. For special cycles \(Z (T,\varphi_f )\), define currents \(\mathfrak{g}(T,\mathbf{y},\varphi_f)\) on \(\mathcal{X}_K (\mathbb{C})\) as weighted sums of \(\mathfrak{g}^o(\mathbf{v})\) over vectors \(\mathbf{v}\) satisfying \(T(\mathbf{v}) = T\). These currents also depend on a parameter \(\mathbf{y} \in \mathrm{Sym}_r(F\otimes_{\mathbb{Q}} \mathbb{R})_{\gg 0}\) or \(\mathbf{y} \in \mathrm{Her}_r(E\otimes_{\mathbb{Q}} \mathbb{R})_{\gg 0}\) in the orthogonal or unitary cases, respectively. Hence, one has Theorem 1.1. The current \(\mathfrak{g}(T,\mathbf{y},\varphi_f)\) satisfies \[ dd^c\mathfrak{g}(T,\mathbf{y},\varphi_f) + \delta_{Z(T,\varphi_f)(\mathbb{C})} \wedge \Omega_{\mathcal{E}^\vee}^{r-\mathrm{rk}(T)} = \omega(T,\mathbf{y},\varphi_f) \] where \(\omega(T,\mathbf{y},\varphi_f)\) is the \(T\)-th coefficient in the \(q\)-expansion of the theta function attached to \(\varphi_{\mathrm{KM}}\otimes \varphi_f\). Moreover, if \(T_1\) and \(T_2\) are non-degenerate and \(Z(T_1,\varphi_f)\) and \(Z(T_2,\varphi_f)\) intersect properly, then \[ \mathfrak{g}(T_1,\mathbf{y}_1,\varphi_1) \ast \mathfrak{g}(T_2,\mathbf{y}_2,\varphi_2) \equiv \sum_{T=(\begin{smallmatrix} T_1&\ast\\ \ast&T_2 \end{smallmatrix})} \mathfrak{g}(T,(\begin{smallmatrix}\mathbf{y}_1&\ast\\ \ast&\mathbf{y}_2 \end{smallmatrix}),\varphi_1\otimes \varphi_2) \quad \mathrm{mod\,im}\,d + \mathrm{im}\,d^c. \] When \(T\) is non-degenerate, the first equation in the theorem is Green's equation for the cycle \(Z(T,\varphi_f)\) and hence \(\mathfrak{g}(T,\mathbf{y},\varphi_f)\) is a Green current (in fact, a Green form) for \(Z(T,\varphi_f)\). When \(T\) is degenerate, the cycle \(Z(T,\varphi_f)\) appears in the `wrong' codimension; following \textit{S. S. Kudla} [Duke Math. J. 86, No. 1, 39--78 (1997; Zbl 0879.11026)], this deficiency can be rectified by intersecting with a power of the tautological bundle and, as discussed in Section 5.4, solutions to the equation above correspond naturally to Green currents for this modified cycle. The main result of the paper computes local archimedean heights of special cycles in terms of Siegel Eisenstein series. The authors restrict to the case that \(\mathbb{V}\) is anisotropic, so that the corresponding Shimura variety is compact; They state that their assumption on the signature of \(\mathbb{V}\) ensures that this is the case whenever \(F \not= \mathbb{Q}\). They further assume that \(q = 1\) in the unitary case. Fix an integer \(r \leqslant p+1\), let \(s_0 := (p+1-r)/2\), and, for a Schwartz function \(\varphi_f \in \mathcal{S}(\mathbb{V(}(\mathbb{A}_f)^r )^K\), (as to be discussed in Section 5.2) consider the corresponding genus \(r\) Siegel Eisenstein series \[ E(\boldsymbol{\tau},\Phi_f,s) = \sum_T E_T(\boldsymbol{\tau},\Phi_f,s) \] of parallel scalar weight \(l = \mathrm{dim}_F(\mathbb{V})/2\) (resp. \(\mathrm{dim}_E(\mathbb{V})\)); here \(\boldsymbol{\tau} = \mathbf{x} + i\mathbf{y} \in \mathbb{H}_r^d \) where \(\mathbb{H}_r\) is the Siegel (resp. Hermitian) upper halfspace of genus \(r\). Let \[ E'_T(\boldsymbol{\tau},\Phi_f,s_0) := \frac{d}{ds}E_T(\boldsymbol{\tau},\Phi_f,s)|_{s=s_0} \] denote the derivative of its Fourier coefficient \(E_T(\boldsymbol{\tau},\Phi_f,s)\) at \(s = s_0\). Theorem 1.2. Suppose that \(\mathbb{V}\) is anisotropic and, in the unitary case, that \(q = 1\). Then for any \(T\) , there is an explicit constant \(\kappa(T,\Phi_f)\), given by Definition 5.7, such that \[ \frac{(-1^r\kappa_0}{2\mathrm{Vol}(X_{\mathbb{V},K},\Omega_\mathcal{E})} \int_{[\mathcal{X}_K(\mathbb{C})]}\mathfrak{g}(T,\mathbf{y},\varphi_f) \wedge \Omega_\mathcal{E}^{p+1-r} q^T = E'_T(\boldsymbol{\tau},\Phi_f,s_0) - \kappa(T,\Phi_f) q^T. \] Here \(q^T = e^{2\pi i\mathrm{tr}(T\tau)}\), and \(\kappa_0 =1\) if \(s_0 >0\) and \(\kappa_0 =2\) if \(s_0 =0\). In the special case that \(T\) is non-degenerate, one has a factorization \[ E_T(\boldsymbol{\tau},\Phi_f,s) = W_{T,\infty}(\boldsymbol{\tau},\Phi^l_\infty,s)\cdot W_{T,f}(e,\Phi_f,s) \] where the factors on the right are the products of the archimedean and nonarchimedean local Whittaker functionals, respectively. Let \[ E'_T(\boldsymbol{\tau},\Phi_f,s_0)_\infty = W'_{T,\infty}(\boldsymbol{\tau},\Phi^l_\infty,s_0)\cdot W_{T,f}(e,\Phi_f,s_0) \] denote the archimedean contribution to the special derivative. Then, if \(T\) is totally positive definite, Theorem 1.2 specializes to the identity \[ \begin{aligned} \frac{(-1)^r\kappa_0}{2\mathrm{Vol}(X_{\mathbb{V},K},\Omega_\mathcal{E})} & \int_{[\mathcal{X}_K(\mathbb{C})]}\mathfrak{g}(T,\mathbf{y},\varphi_f) \wedge \Omega_\mathcal{E}^{p+1-r} q^T = E'_T(\boldsymbol{\tau},\Phi_f,s_0)_\infty\\ & - E_T(\boldsymbol{\tau},\Phi_f,s_0) ((\iota d)/2(r\mathrm{log}\,\pi -\frac{\Gamma'(\iota m/2)}{\Gamma(\iota m/2)} + \frac{\iota}{2}\,\mathrm{log}\,N_{F/\mathbb{Q}}\,\mathrm{det}\,T), \end{aligned} \] here \(\iota = 1\) (resp. \(\iota =2\)) in the orthogonal (resp. unitary) case. If \(T\) is not totally positive definite, the last summand is zero. Theorem 1.2 reappears in a slightly more explicit form as Theorem 5.10 in Section 5 and is proven there. The proof relies heavily on the appropriate Siegel Weil formula. The last parts of Section 5 also report on prior work with similar results mainly on the case of codimension one cycles. Here the reviewer would like to add another item, namely his work with \textit{U. Kühn} [``On Kudla's Green function for signature (2,2)'', part I and II, \url{arXiv:1205.6417}; \url{arXiv:1209.3949}]. Finally, the authors place their results in the context of Kudla's conjectures on special cycles in arithmetic Chow groups. And they close their introductory Section 1.1 with the statement ``Put another way, our theorem reduces Kudla's conjecture to a relatively explicit conjectural formula for the analogue of the Faltings height (as in the above mentioned paper of Bost, Gillet, and Soulé) of a special cycle \(Z(T,\varphi_f)\) in terms of \(T\) and \(\varphi_f\), we discuss this point in more detail in Sect. 5.5.''