On the classification of $(\mathfrak{g},K)$-modules generated by nearly holomorphic Hilbert-Siegel modular forms and projection operators

The arithmeticity of special $L$ values is a central problem in modern number theory. In the motivic setting, Deligne [Del79] conjectured the algebraicity of critical $L$ values up to the period. For the critical values attached to scalar valued Hilbert-Siegel modular forms and Hermitian modular forms, Shimura proved the arithmeticity of them up to suitable periods in [Shi00] by using of nearly holomorphic modular forms. The period can be expressed by Petersson inner product times some power of $\pi$. Recently, in [HPSS21], Pitale, Saha, Schmidt and the author prove the arithmeticity of them attached to vector valued Siegel modular forms under the parity condition of weights. The purpose of this paper is to prepare to remove the parity condition by investigating the $(\mathfrak{g},K)$-modules generated by nearly holomorphic Hilbert-Siegel modular forms.

Let $F$ be a totally real field with degree $d$ and $\mathbf{a}$ the set of embeddings of $F$ into $\mathbb{R}$. Put $G_{n}=\mathop{\mathrm{Res}}_{F/\mathbb{Q}}\mathrm{Sp}_{2n}$. Here $\mathop{\mathrm{Res}}$ is the Weil restriction and $\mathrm{Sp}_{2n}$ is the symplectic group of rank $n$. Let $\mathfrak{H}_{n}$ be the Siegel upper half space of degree $n$. Put $\mathfrak{g}_{n}=\mathrm{Lie}(G_{n}(\mathbb{R}))\otimes_{\mathbb{R}}\mathbb{C}$. We denote by $K_{n,\infty}$ and $\mathcal{Z}_{n}$ the stabilizer of $\mathbf{i}=(\sqrt{-1}\,\mathbf{1}_{n},\ldots,\sqrt{-1}\,\mathbf{1}_{n})\in\mathfrak{H}_{n}^{d}$ and the center of the universal enveloping algebra $\mathcal{U}(\mathfrak{g}_{n})$, respectively. Let $K_{n,\mathbb{C}}$ be the complexification of $K_{n,\infty}$. Set $\mathfrak{k}_{n}=\mathrm{Lie}(K_{n,\infty})\otimes_{\mathbb{R}}\mathbb{C}$. We then have the well-known decomposition:

$$\mathfrak{g}_{n}=\mathfrak{k}_{n}\oplus\mathfrak{p}_{n,+}\oplus\mathfrak{p}_{n,-}.$$

Here $\mathfrak{p}_{n,+}$ (resp. $\mathfrak{p}_{n,-}$) is the Lie subalgebra of $\mathfrak{g}_{n}$ corresponding to the holomorphic tangent space (resp. anti-holomorphic tangent space) of $\mathfrak{H}_{n}^{d}$ at $\mathbf{i}$. We take a Cartan subalgebra of $\mathfrak{k}_{n}$. Then it is a Cartan subalgebra of $\mathfrak{g}_{n}$. The root system $\Phi$ of $\mathfrak{sp}_{2n}(\mathbb{C})$ is

$$\Phi=\{\;\pm(e_{i}+e_{j}),\;\pm(e_{k}-e_{\ell}),\quad 1\leq i\leq j\leq n,1\leq k<\ell\leq n\;\}.$$

We consider the set

$$\Phi^{+}=\{\;-(e_{i}+e_{j}),\;e_{k}-e_{\ell},\quad 1\leq i\leq j\leq n,1\leq k<\ell\leq n\;\}$$

to be a positive root system. Let $\rho$ be half the sum of positive roots. Note that $\mathfrak{g}_{n}=\bigoplus_{v\in\mathbf{a}}\mathfrak{sp}_{2n}(\mathbb{C})$. We say that a weight $\lambda=(\lambda_{1,v},\ldots,\lambda_{n,v})_{v\in\mathbf{a}}$ which lies in $\bigoplus_{v\in\mathbf{a}}\mathbb{C}^{n}$ is $\mathfrak{k}_{n}$-dominant if $\lambda_{i,v}-\lambda_{i+1,v}\in\mathbb{Z}_{\geq 0}$ for any $1\leq i\leq n-1$ and $v\in\mathbf{a}$. We also say that a $\mathfrak{k}_{n}$-dominant integral weight $\lambda=(\lambda_{1,v},\ldots,\lambda_{n,v})_{v\in\mathbf{a}}$ is anti-dominant if $\lambda_{n}\geq n$. For any $\mathfrak{k}_{n}$-dominant integral weight $\lambda$, there exist the (parabolic) Verma module $N(\lambda)$ with respect to a parabolic subalgebra $\mathfrak{p}_{n,-}\oplus\mathfrak{k}_{n}$ and a unique irreducible highest weight $(\mathfrak{g}_{n},K_{n,\infty})$-module $L(\lambda)$ of highest weight $\lambda$. Then, $L(\lambda)$ is the unique irreducible quotient of $N(\lambda)$. For a $(\mathfrak{g}_{n},K_{n,\infty})$-module $\pi$, the symbol $\pi^{\vee}$ denotes the contragredient of $\pi$ in the sense of [Hum08].

For an automorphic form $\varphi$ on $G_{n}(\mathbb{A}_{\mathbb{Q}})$, we say that $\varphi$ is nearly holomorphic if $\varphi$ is $\mathfrak{p}_{n,-}$-finite, i.e., $\mathcal{U}(\mathfrak{p}_{n,-})\cdot\varphi$ is finite-dimensional. The goal of this paper is to classify the indecomposable $(\mathfrak{g}_{n},K_{n,\infty})$-modules generated by nearly holomorphic automorphic forms.

This result is a generalization of [PSS21]. The key idea of proof is the harmonic analysis of the space of nearly holomorphic automorphic forms on $G_{n}(\mathbb{A}_{\mathbb{Q}})$, which is investigated in [Hor20b].

Fix a weight $\rho$ and a congruence subgroup $\Gamma$. Let $N_{\rho}(\Gamma)$ be the space of nearly holomorphic Hilbert-Siegel modular forms of weight $\rho$ with respect to $\Gamma$. For an infinitesimal character $\chi$ of $\mathcal{Z}_{n}$, we can define the projection operator $\mathfrak{p}_{\chi}\in\mathrm{End}(N_{\rho}(\Gamma))$ associated to $\chi$. Then, the projection operator $\mathfrak{p}_{\chi}$ commutes with the $\mathrm{Aut}(\mathbb{C})$ action as follows:

For a $\mathfrak{k}_{n}$-dominant integral weight $\lambda$ and $v\in\mathbf{a}$, put $j_{v}(\lambda)=\#\{j\mid\lambda_{1,v}\equiv\lambda_{j,v}\,(\mathrm{mod}\,2)\}$. Set

$$\wedge^{j_{v}(\lambda)}=\wedge^{j_{v}(\lambda)}\mathrm{std}_{\mathrm{GL}_{n}(\mathbb{C})},\quad\rho_{v}=\mathrm{det}^{\lambda_{1,v}-1}\otimes\wedge^{j_{v}}\mathrm{std}_{\mathrm{GL}_{n}(\mathbb{C})},\quad\text{and}\quad\rho=\bigotimes_{v\in\mathbf{a}}\rho_{v},$$

where $\mathrm{std}_{\mathrm{GL}_{n}(\mathbb{C})}$ is the standard representation of $\mathrm{GL}_{n}(\mathbb{C})$ and $\wedge^{j_{v}(\lambda)}\mathrm{std}_{\mathrm{GL}_{n}(\mathbb{C})}$ is the $j_{v}(\lambda)$-th exterior product of $\mathrm{std}_{\mathrm{GL}_{n}(\mathbb{C})}$.

The following is the analogue of holomorphic projection.

We then characterize the nearly holomorphic Hilbert-Siegel modular forms which generate a holomorphic discrete series representation in terms of projections $\mathfrak{p}_{\chi}$ under a mild assumption. This gives a generalization of Shimura’s holomorphic projection.

We denote by $\mathrm{Mat}_{m,n}$ the set of $m\times n$-matrices. Put $\mathrm{Mat}_{n}=\mathrm{Mat}_{n,n}$ with the unit $\mathbf{1}_{n}$. Let $\mathrm{GL}_{n}$ and $\mathrm{Sp}_{2n}$ be the algebraic groups defined by

$$\mathrm{GL}_{n}(R)=\{g\in\mathrm{Mat}_{n}\mid\det g\in R^{\times}\}$$

and

$$\mathrm{Sp}_{2n}(R)=\left\{g\in\mathrm{GL}_{2n}(R)\,\middle|\,{{}^{t}{g}}\begin{pmatrix}0_{n}&-\mathbf{1}_{n}\\ \mathbf{1}_{n}&0_{n}\end{pmatrix}g=\begin{pmatrix}0_{n}&-\mathbf{1}_{n}\\ \mathbf{1}_{n}&0_{n}\end{pmatrix}\right\}$$

for a ring $R$, respectively. Set $\mathrm{Sym}_{n}=\{g\in\mathrm{Mat}_{n}\mid{{}^{t}g}=g\}$. Let $B_{n}$ be the subgroup of $\mathrm{Sp}_{2n}$ defined by

$$B_{n}=\left\{\begin{pmatrix}a&*\\ 0&{{}^{t}a^{-1}}\end{pmatrix}\,\middle|\,\text{$a$ is a upper triangular matrix.}\right\}.$$

The group $B_{n}$ is a Borel subgroup of $\mathrm{Sp}_{2n}$ with the Levi decomposition $B_{n}=T_{n}N_{n}$. Here $T_{n}\subset B_{n}$ is the maximal diagonal torus of $\mathrm{Sp}_{2n}$. A parabolic subgroup $P$ of $\mathrm{Sp}_{2n}$ is called standard if $P$ contains $B_{n}$. Let $A_{P}$ be the split component of $P$ and $A_{P}^{\infty}$ the identity component of $A_{P}(\mathbb{R})$. We denote by $P_{i,n}$ and $Q_{i,n}$ the standard parabolic groups of $\mathrm{Sp}_{2n}$ with the Levi subgroups $\mathrm{GL}_{i}\times\mathrm{Sp}_{2(n-i)}$ and $(\mathrm{GL}_{1})^{i}\times\mathrm{Sp}_{2(n-i)}$, respectively. Set $P_{n}=P_{n,n}$. For a parabolic subgroup $P$, let $\delta_{P}$ be the modulus character of $P$.

For $n\in\mathbb{Z}_{\geq 1}$, set

$$\mathfrak{H}_{n}=\{z\in\mathrm{Sym}_{n}(\mathbb{C})\mid\text{$\mathrm{Im}(z)$ is positive definite}\}.$$

The space $\mathfrak{H}_{n}$ is called the Siegel upper half space of degree $n$. The Lie group $\mathrm{Sp}_{2n}(\mathbb{R})$ acts on $\mathfrak{H}_{n}$ by the rule

$$\begin{pmatrix}a&b\\ c&d\end{pmatrix}(z)=(az+b)(cz+d)^{-1},\qquad\begin{pmatrix}a&b\\ c&d\end{pmatrix}\in\mathrm{Sp}_{2n}(\mathbb{R}),\,z\in\mathfrak{H}_{n}.$$

Put

$$K_{n,\infty}=\left\{g=\begin{pmatrix}a&b\\ c&d\end{pmatrix}\in\mathrm{Sp}_{2n}(\mathbb{R})\,\middle|\,a=d,\,c=-b\right\}.$$

Then $K_{n,\infty}$ is the group of stabilizers of $\mathbf{i}=\sqrt{-1}\,\mathbf{1}_{n}\in\mathfrak{H}_{n}$. For simplicity the notation, the symbol $\mathbf{i}$ also denotes the element $(\sqrt{-1}\,\mathbf{1}_{n},\ldots,\sqrt{-1}\,\mathbf{1}_{n})\in\mathfrak{H}_{n}^{d}$. Since the action of $\mathrm{Sp}_{2n}(\mathbb{R})$ on $\mathfrak{H}_{n}$ is transitive, we have $\mathfrak{H}_{n}\cong\mathrm{Sp}_{2n}(\mathbb{R})/K_{n,\infty}$.

Let $F$ be a totally real field with degree $d$. Let $\mathbf{a}=\{\infty_{1},\ldots,\infty_{d}\}$ be the set of embeddings of $F$ into $\mathbb{R}$. We denote by $\mathbb{A}_{F}$ and $\mathbb{A}_{F,\mathrm{fin}}$ the adele ring of $F$ and the finite part of $\mathbb{A}_{F}$, respectively. For a place $v$, let $F_{v}$ be the $v$-completion of $F$. Put $F_{\infty}=\prod_{v\in\mathbf{a}}F_{v}$. For a non-archimedean place $v$, let $\mathcal{O}_{F_{v}}$ be the ring of integers of $F_{v}$.

Set $G_{n}=\mathop{\mathrm{Res}}_{F/\mathbb{Q}}\mathrm{Sp}_{2n}$ where $\mathop{\mathrm{Res}}$ is the Weil restriction. We define the standard parabolic subgroups $P_{i,n},Q_{i,n}$ and $B_{n}$ of $G_{n}$ by the Weil restriction of parabolic subgroups $P_{i,n},Q_{i,n}$ and $B_{n}$ of $\mathrm{Sp}_{2n}$, respectively. Let $W_{n}$ be the Weyl group of $\mathrm{Sp}_{2n}$. For an archimedean place $v$, set $K_{n,v}=K_{n,\infty}$. For the sake of simplicity, the symbol $K_{n,\infty}$ denotes the maximal compact subgroup $\prod_{v\in\mathbf{a}}K_{n,v}$ of $G_{n}(\mathbb{R})$. Let $K_{n,\mathbb{C}}$ be the complexification of $\prod_{v\in\mathbf{a}}K_{n,v}$. Put $\mathfrak{g}_{n}=\mathrm{Lie}(G_{n}(\mathbb{R}))\otimes_{\mathbb{R}}\mathbb{C}$ and $\mathfrak{k}_{n}=\mathrm{Lie}(\prod_{v\in\mathbf{a}}K_{n,v})\otimes_{\mathbb{R}}\mathbb{C}$. Set $K_{v}=\mathrm{Sp}_{2n}(\mathcal{O}_{F_{v}})$ for a non-archimedean place $v$. We denote by $\mathcal{Z}_{n}$ the center of the universal enveloping algebra $\mathcal{U}(\mathfrak{g}_{n})$. We then obtain the well-known decomposition

$$\mathfrak{g}_{n}=\mathfrak{k}_{n}\oplus\mathfrak{p}_{n,+}\oplus\mathfrak{p}_{n,-}$$

where $\mathfrak{p}_{n,+}$ (resp. $\mathfrak{p}_{n,-}$) is the Lie subalgebra of $\mathfrak{g}_{n}$ corresponding to the holomorphic tangent space (resp. anti-holomorphic tangent space) of $\mathfrak{H}_{n}^{d}$ at $\mathbf{i}$. It is well-known that the Lie algebras $\mathfrak{g}_{n}$ and $\mathfrak{k}_{n}$ have the same Cartan subalgebra. We fix such a Cartan subalgebra. Then the root system of $\mathfrak{sp}_{2n}(\mathbb{C})$ is

$$\Phi=\{\;\pm(e_{i}+e_{j}),\;\pm(e_{k}-e_{\ell}),\quad 1\leq i\leq j\leq n,1\leq k<\ell\leq n\;\}.$$

We consider the set

$\displaystyle\Phi^{+}=\{\;-(e_{i}+e_{j}),\;e_{k}-e_{\ell},\quad 1\leq i\leq j\leq n,1\leq k<\ell\leq n\;\}$

to be a positive root system. Let $\rho$ be half the sum of positive roots. Put $\rho_{i,n}=n-(i-1)/2$ and $\rho_{n}=\rho_{n,n}$. This corresponds to half the sum of roots in the unipotent subgroup of $P_{i,n}$. For $\lambda=(\lambda_{1,v},\ldots,\lambda_{n,v})_{v}\in\bigoplus_{v\in\mathbf{a}}\mathbb{C}^{n}$, we say that $\lambda$ is a weight if $\lambda_{i,v}-\lambda_{i+1,v}\in\mathbb{Z}$ for any $v$ and $1\leq i\leq n-1$. The weight $\lambda$ is $\mathfrak{k}_{n}$-dominant if $\lambda_{i,v}-\lambda_{i+1,v}\geq 0$ for any $v$ and $1\leq i\leq n-1$. For a $\mathfrak{k}_{n}$-dominant weight $\lambda$, let $\rho_{\lambda}$ be an irreducible finite-dimensional representation of $\mathfrak{k}_{n}$. When $\lambda$ is integral, i.e., any entry of $\lambda$ is an integer, we identify $\rho_{\lambda}$ as the derivative of an irreducible finite-dimensional representation of $K_{n,\mathbb{C}}$ with highest weight $\lambda$. We then write the representation of $K_{n,\mathbb{C}}$ by the same $\rho_{\lambda}$.

We fix a non-trivial additive character $\psi=\bigotimes_{v}\psi_{v}$ of $F\backslash\mathbb{A}_{F}$ as follows: If $F=\mathbb{Q}$, let

	$\displaystyle\psi_{p}(x)$	$\displaystyle=\exp(-2\pi\sqrt{-1}\,y),\qquad x\in\mathbb{Q}_{p},$
	$\displaystyle\psi_{\infty}(x)$	$\displaystyle=\exp(2\pi\sqrt{-1}\,x),\qquad x\in\mathbb{R},$

where $y\in\cup_{m=1}^{\infty}p^{-m}\mathbb{Z}$ such that $x-y\in\mathbb{Z}_{p}$. In general, for an archimedean place $v$ of $F$, put $\psi_{v}=\psi_{\infty}$ and for a non-archimedean place $v$ with the rational prime $p$ divisible by $v$, put $\psi_{v}(x)=\psi_{p}(\mathrm{Tr}_{F_{v}/\mathbb{Q}_{p}}(x))$.

For a function $f$ on a group $G$, let $r$ be the right translation, i.e., $r(g)f(h)=f(hg)$ for any $g,h\in G$. For a subset $H$ of $G$, we denote by $f|_{H}$ the restriction of $f$ to $H$. Let $G$ be a Lie group with the Lie algebra $\mathfrak{g}$. For a smooth function $f$ on $G$ and $X\in\mathfrak{g}$, put

$$X\cdot f(g)=\left.\frac{d}{dt}\right|_{t=0}f(g\exp(tX)),\qquad g\in G.$$

For the action of $G_{n}(\mathbb{A}_{\mathbb{Q}})$, we mean the $G_{n}(\mathbb{A}_{\mathbb{Q},\mathrm{fin}})\times(\mathfrak{g}_{n},K_{n,\infty})$-action.

We recall the differential operators on $\mathfrak{H}_{n}$. For details, see [Shi00, §12]. Fix a basis on $\mathrm{Sym}_{n}(\mathbb{C})$ by $\{(1+\delta_{i,j})^{-1}(e_{i,j}+e_{j,i})\mid 1\leq i\leq j\leq n\}$. We denote the basis by $\{\varepsilon_{\nu}\}$. For $u\in\mathrm{Sym}_{n}(\mathbb{C})$, write $u=\sum_{\nu}u_{\nu}\varepsilon_{\nu}$ with $u_{\nu}\in\mathbb{C}$ and for $z\in\mathfrak{H}_{n}$, write $z=\sum_{\nu}z_{\nu}\varepsilon_{\nu}$ with $z_{\nu}\in\mathbb{C}$. For a non-negative integer $e$ and a finite-dimensional vector space $V$, let $S_{e}(\mathrm{Sym}_{n}(\mathbb{C}),V)$ be the space of $V$-valued homogeneous polynomial maps of degree $e$ on $\mathrm{Sym}_{n}(\mathbb{C})$ and $\mathrm{Ml}_{e}(\mathrm{Sym}_{n}(\mathbb{C}),V)$ the space of $e$-multilinear maps on $\mathrm{Sym}_{n}(\mathbb{C})^{e}$ to $V$. Note that $S_{e}(\mathrm{Sym}_{n}(\mathbb{C}),V)$ can be viewed as the space of symmetric elements of $\mathrm{Ml}_{e}(\mathrm{Sym}_{n}(\mathbb{C}),V)$. For a representation $\rho$ of $\mathrm{GL}_{n}(\mathbb{C})$ on $V$, we define representations $\rho\otimes\tau^{e}$ and $\rho\otimes\sigma^{e}$ on $\mathrm{Ml}_{e}(\mathrm{Sym}_{n}(\mathbb{C}),V)$ by

$$((\rho\otimes\tau^{e})(a)h)(u_{1},\ldots,u_{e})=\rho(a)h({{}^{t}a}u_{1}a,\ldots,{{}^{t}a}u_{e}a)$$

and

$$((\rho\otimes\sigma^{e})(a)h)(u_{1},\ldots,u_{e})=\rho(a)h({a}^{-1}u_{1}{{}^{t}a^{-1}},\ldots,{a}^{-1}u_{e}{{}^{t}a^{-1}}),$$

respectively. Here, $h\in\mathrm{Ml}_{e}(\mathrm{Sym}_{n}(\mathbb{C}),V)$, $a\in\mathrm{GL}_{n}(\mathbb{C})$ and $(u_{1},\ldots,u_{e})\in\mathrm{Sym}_{n}(\mathbb{C})^{e}$. The symbols $\rho\otimes\tau^{e}$ and $\rho\otimes\sigma^{e}$ also denote the restrictions to the representations space $S_{e}(\mathrm{Sym}_{n}(\mathbb{C}),V)$.

For $f\in C^{\infty}(\mathfrak{H}_{n},V)$, we define functions $Df,\overline{D}f,Cf,E,f$ on $C^{\infty}(\mathfrak{H}_{n},S_{1}(\mathrm{Sym}_{n}(\mathbb{C}),V))$ by

	$\displaystyle((Df)(z))(u)=\sum_{\nu}u_{\nu}\frac{\partial f}{\partial z_{\nu}}(z)$	$\displaystyle,\qquad((\overline{D}f)(z))(u)=\sum_{\nu}u_{\nu}\frac{\partial f}{\partial\overline{z_{\nu}}}(z),$
	$\displaystyle((Cf)(z))(u)=4((Df)(z))(yuy)$	$\displaystyle,\qquad((Ef)(z))(u)=4((\overline{D}f)(z))(yuy).$

Here, $u=\sum_{\nu}u_{\nu}\varepsilon_{\nu}\in\mathrm{Sym}_{n}(\mathbb{C}),z=\sum_{\nu}z_{\nu}\varepsilon_{\nu}\in\mathfrak{H}_{n}$ and $y=\mathrm{Im}(z)$. For $f\in C^{\infty}(\mathfrak{H}_{n},V)$, we say that $f$ is nearly holomorphic if there exists $e$ such that $E^{e}f=0$.

Let $F$ be the fixed totally real field. For an integral ideal $\mathfrak{n}$ of $F$, set

$$\Gamma(\mathfrak{n})=\left\{\gamma\in\mathrm{Sp}_{2n}(\mathcal{O}_{F})\,\middle|\,\gamma-\mathbf{1}_{2n}\in\mathrm{Mat}_{2n}(\mathfrak{n})\right\}.$$

The group $\Gamma(\mathfrak{n})$ is called the principal congruence subgroup of $G_{n}(\mathbb{Q})$ of level $\mathfrak{n}$. We say that a subgroup $\Gamma$ of $G_{n}(\mathbb{Q})$ is a congruence subgroup if there exists an integral ideal $\mathfrak{n}$ such that $\Gamma$ contains $\Gamma(\mathfrak{n})$ and $[\Gamma\colon\Gamma(\mathfrak{n})]<\infty$. In this subsection, we regard $G_{n}(\mathbb{Q})$ as a subgroup of $G_{n}(\mathbb{R})=\prod_{v\in\mathbf{a}}\mathrm{Sp}_{2n}(F_{v})$ by $\gamma\longmapsto(\infty_{1}(\gamma),\ldots,\infty_{d}(\gamma))$. Similarly, we regard a congruence subgroup $\Gamma$ of $G_{n}(\mathbb{Q})$ as a subgroup of $G_{n}(\mathbb{R})$.

We define the factor of automorphy $j\colon G_{n}(\mathbb{R})\times\mathfrak{H}_{n}^{d}\longrightarrow\mathrm{GL}_{n}(\mathbb{C})^{d}$ by

$$j(g,z)=(c_{v}z_{v}+d_{v})_{v}\in\prod_{v\in\mathbf{a}}\mathrm{GL}_{n}(\mathbb{C})=\mathrm{GL}_{n}(\mathbb{C})^{d},\qquad g=\left(\begin{pmatrix}a_{v}&b_{v}\\ c_{v}&d_{v}\end{pmatrix}\right)_{v}\in G_{n}(\mathbb{R}),\quad z=(z_{v})_{v}\in\mathfrak{H}_{n}^{d}.$$

For a representation $\rho$ of $K_{n,\mathbb{C}}$ on $V$, set $j_{\rho}=\rho\circ j$. For $g\in G_{n}(\mathbb{R})$, we define the slash operator $|_{\rho}g$ on $C^{\infty}(\mathfrak{H}_{n}^{d},V)$ by

$$(f|_{\rho}g)(z_{1},\ldots,z_{d})=j_{\rho}(g,z)^{-1}f(\gamma(z_{1},\ldots,z_{d})),$$

for $f\in C^{\infty}(\mathfrak{H}_{n}^{d},V)$ and $(z_{1},\ldots,z_{d})\in\mathfrak{H}^{d}_{n}$. Let $\Gamma$ be a congruence subgroup of $G_{n}(\mathbb{Q})$. Suppose that a function $f\in C^{\infty}(\mathfrak{H}_{n}^{d},V)$ satisfies the automorphy $f|_{\rho}\gamma=f$ for any $\gamma\in\Gamma$. Then, $f$ has the Fourier expansion

$$(f|_{\rho}\gamma)(z)=\sum_{h\in\mathrm{Sym}_{n}(F)}c_{f}(h,y,\gamma)\mathbf{e}(\mathrm{tr}({hz})),\qquad z\in\mathfrak{H}_{n}^{d},\,y=\mathrm{Im}(z)$$

where $\mathbf{e}(\mathrm{tr}(hz))=\exp(2\pi\sqrt{-1}\,\sum_{j=1}^{h}\mathrm{tr}(\infty_{j}(h)z_{j}))$ for $(z_{1},\ldots,z_{d})\in\mathfrak{H}_{n}^{d}$ and $h\in\mathrm{Sym}_{n}(F)$. We consider the following condition: If $c_{f}(h,y,\gamma)\neq 0$, the matrix $h$ is positive semi-definite. We call this condition the cusp condition. We say that a $V$-valued $C^{\infty}$-function $f$ on $\mathfrak{H}_{n}^{d}$ is a nearly holomorphic Hilbert-Siegel modular form of weight $\rho$ with respect to $\Gamma$ if $f$ satisfies the following conditions:

• •

  $f$ is a nearly holomorphic function.

• •

  $f|_{\rho}\gamma=f$ for all $\gamma\in\Gamma$.

• •

  $f$ satisfies the cusp condition.

We denote by $N_{\rho}(\Gamma)$ the space of nearly holomorphic Hilbert-Siegel modular forms of weight $\rho$ with respect to $\Gamma$. In the following, for modular forms, we mean a (nearly holomorphic) Hilbert-Siegel modular forms. By Köecher principle, we can remove the cusp condition if $n>1$ or $F\neq\mathbb{Q}$. For the proof, see [Hor20a, Proposition 4.1] for $n>1$. We can give the same proof for the case of $F\neq\mathbb{Q}$. For simplicity, if $\rho=\det^{k}$, we say that a modular form of weight $\det^{k}$ is a modular form of weight $k$.

Let $f$ be a nearly holomorphic modular form of weight $\rho$ with respect to $\Gamma$. Take a model $V$ of $\rho$ and fix a rational structure of $V$. Then, Shimura introduced the $\mathrm{Aut}(\mathbb{C})$-action on $f$. For details, see [Shi00, §14.11] and [HPSS21, §3.3]. For $\sigma\in\mathrm{Aut}(\mathbb{C})$, we denote by ${{}^{\sigma}f}$ the action of $\sigma$ on $f$. For a weight $\rho=\bigotimes_{v\in\mathbf{a}}\rho_{v}$, put ${{}^{\sigma}\rho}=\bigotimes_{v\in\mathbf{a}}\rho_{\sigma\circ v}$. The following theorem is proved in [Shi00, Theorem 14.12].

Let $M_{\rho}(\Gamma)$ be the space of holomorphic functions in $N_{\rho}(\Gamma)$. Set $N^{p}_{\rho}(\Gamma)=N_{\rho}^{(p_{v})_{v}}(\Gamma)=\{f\in N_{\rho}(\Gamma)\mid\text{$E_{v}^{p_{v}+1}f=0$ for any $v\in\mathbf{a}$}\}$. The, $N^{0}_{\rho}(\Gamma)=M_{\rho}(\Gamma)$. Let $\rho=\bigotimes_{v}\rho_{v}$ be a character of $K_{n,\mathbb{C}}$ with the weight $(k_{v})_{v\in\mathbf{a}}$. Take non-negative integers $p_{v}$ satisfies $k_{v}>n+p_{v}$ or $k_{v}<n+(3-p_{v})/2$ for any $v\in\mathbf{a}$. Put $p=(p_{v})_{v}$. Then, in [Shi00, §15.3], Shimura introduced a projection $\mathfrak{A}\colon N_{\rho}^{p}(\Gamma)\longrightarrow M_{\rho}(\Gamma)$. The projection $\mathfrak{A}$ is called the holomorphic projection. By Shimura [Shi00, Proposition 15.3], it commutes with the $\mathrm{Aut}(\mathbb{C})$ actions as follows:

In [HPSS21, §3.4], we define other projection operators $\mathfrak{p}_{\chi}$ associated to infinitesimal characters $\chi$ of $\mathcal{Z}_{n}$. This can be viewed as a generalization of the holomorphic projection $\mathfrak{A}$. In this paper, we study the image of $\mathfrak{p}_{\chi}$ in terms of $(\mathfrak{g}_{n},K_{n,\infty})$-modules.

Let $P=MN$ be a standard parabolic subgroup of $G_{n}$. For a smooth function $\phi:N(\mathbb{A}_{\mathbb{Q}})M(\mathbb{Q})\backslash G_{n}(\mathbb{A}_{\mathbb{Q}})\longrightarrow\mathbb{C}$, we say that $\phi$ is automorphic if it satisfies the following conditions:

• •

  $\phi$ is right $K_{n}$-finite.

• •

  $\phi$ is $\mathcal{Z}_{n}$-finite.

• •

  $\phi$ is slowly increasing.

We denote by $\mathcal{A}(P\backslash G_{n})$ the space of automorphic forms on $N(\mathbb{A}_{\mathbb{Q}})M(\mathbb{Q})\backslash G_{n}(\mathbb{A}_{\mathbb{Q}})$. For simplicity, we write $\mathcal{A}(G_{n})$ when $P=G_{n}$. The space $\mathcal{A}(P\backslash G_{n})$ is stable under the action of $G_{n}(\mathbb{A}_{\mathbb{Q}})$.

For parabolic subgroups $P$ and $Q$ of $G_{n}$, we say that $P$ and $Q$ are associate if the split components $A_{P}$ and $A_{Q}$ are $G_{n}(\mathbb{Q})$-conjugate. We denote by $\{P\}$ the associated class of the parabolic subgroup $P$. For a locally integrable function $\varphi$ on $N_{P}(\mathbb{Q})\backslash G_{n}(\mathbb{A}_{\mathbb{Q}})$, set

$$\varphi_{P}(g)=\int_{N_{P}(\mathbb{Q})\backslash N_{P}(\mathbb{A}_{\mathbb{Q}})}\varphi(ng)\,dn$$

where $P=M_{P}N_{P}$ is the Levi decomposition of $P$ and the Haar measure $dn$ is normalized by

$$\int_{N_{P}(\mathbb{Q})\backslash N_{P}(\mathbb{A}_{\mathbb{Q}})}\,dn=1.$$

The function $\varphi_{P}$ is called the constant term of $\varphi$ along $P$. If $\varphi$ lies in $\mathcal{A}(P\backslash G_{n})$, $\varphi_{Q}$ is an automorphic form on $N_{Q}(\mathbb{A}_{\mathbb{Q}})M_{Q}(\mathbb{Q})\backslash G_{n}(\mathbb{A}_{\mathbb{Q}})$ for a parabolic subgroup $Q\subset P$. We call $\varphi$ cuspidal if $\varphi_{Q}$ is zero for any standard parabolic subgroup $Q$ of $G$ with $Q\subsetneq P$. We denote by $\mathcal{A}_{\mathrm{cusp}}(P\backslash G_{n})$ the space of cusp forms in $\mathcal{A}(P\backslash G_{n})$. For a character $\xi$ of the split component $A_{P}^{\infty}$, put

$$\mathcal{A}(P\backslash G)_{\xi}=\{\varphi\in\mathcal{A}(P\backslash G_{n})\mid\text{$\varphi(ag)=a^{\xi+\rho_{P}}\varphi(g)$ for any $g\in G_{n}(\mathbb{A}_{\mathbb{Q}})$ and $a\in A_{P}^{\infty}$}\}.$$

Here, $\rho_{P}$ is the character of $A_{P}^{\infty}$ corresponding to half the sum of roots of $N_{P}$ relative to $A_{P}$. We define $\mathcal{A}_{\mathrm{cusp}}(P\backslash G)_{\xi}$ similarly. Set

$$\mathcal{A}(P\backslash G_{n})_{Z}=\bigoplus_{\xi}\mathcal{A}(P\backslash G_{n})_{\xi},\qquad\mathcal{A}_{\mathrm{cusp}}(P\backslash G_{n})_{Z}=\bigoplus_{\xi}\mathcal{A}_{\mathrm{cusp}}(P\backslash G_{n})_{\xi}.$$

Here, $\xi$ runs over all the characters of $A_{P}^{\infty}$. Let $\mathfrak{a}_{P}$ be the real vector space generated by coroots associated to the root system of $G_{n}$ relative to $A_{P}$. Then, by [MW95, Lemma I.3.2], there exist canonical isomorphisms

(2.4.1)

$\displaystyle\mathbb{C}[\mathfrak{a}_{P}]\otimes\mathcal{A}(P\backslash G_{n})_{Z}\cong\mathcal{A}(P\backslash G),\qquad\mathbb{C}[\mathfrak{a}_{P}]\otimes\mathcal{A}_{\mathrm{cusp}}(P\backslash G_{n})_{Z}\cong\mathcal{A}_{\mathrm{cusp}}(P\backslash G_{n}).$

For a standard Levi subgroup $M$, set

$$M(\mathbb{A}_{\mathbb{Q}})^{1}=\bigcap_{\chi\in\mathop{\mathrm{Hom}}_{\mathrm{conti}}(M(\mathbb{A}_{\mathbb{Q}}),\mathbb{C}^{\times})}\mathrm{Ker}(|\chi|).$$

For a function $f$ on $G_{n}(\mathbb{A}_{\mathbb{Q}})$ and $g\in G_{n}(\mathbb{A}_{\mathbb{Q}})$, let $f_{g}$ be the function on $M_{P}(\mathbb{A}_{\mathbb{Q}})^{1}$ defined by $m\longmapsto m^{-\rho_{P}}f(mg)$. Put

$$\mathcal{A}(G_{n})_{\{P\}}=\left\{\varphi\in\mathcal{A}(G)\,\middle|\,\begin{matrix}\text{$\varphi_{Q,ak}$ is orthogonal to all cusp forms on $M_{Q}(\mathbb{A}_{Q})^{1}$}\\ \text{for any $a\in A_{Q},k\in K_{n}$, and $Q\not\in\{P\}$}\end{matrix}\right\}.$$

By [MW95, Lemma I.3.4], $\mathcal{A}(G_{n})_{\{G\}}$ is equal to $\mathcal{A}_{\mathrm{cusp}}(G_{n})$. More precisely, Langlands [Lan06] had proven the following result:

Let $M$ be a standard Levi subgroup of $G_{n}$ and $\tau$ an irreducible cuspidal automorphic representation of $M(\mathbb{A}_{\mathbb{Q}})$. We say that a cuspidal datum is a pair $(M,\tau)$ such that $M$ is a Levi subgroup of $G_{n}$ and that $\tau$ is an irreducible cuspidal automorphic representation of $M(\mathbb{A}_{\mathbb{Q}})$. Take $w\in W_{n}$. Put $M^{w}=wMw^{-1}$ and let $P^{w}=M^{w}N^{w}$ be the standard parabolic subgroup with Levi subgroup $M^{w}$. The irreducible cuspidal automorphic representation $\tau^{w}$ of $M^{w}(\mathbb{A}_{\mathbb{Q}})$ is defined by $\tau^{w}(m^{\prime})=\tau(w^{-1}m^{\prime}w)$ for $m^{\prime}\in M^{w}(\mathbb{A}_{\mathbb{Q}})$. Two cuspidal data $(M,\tau)$ and $(M^{\prime},\tau^{\prime})$ are called equivalent if there exists $w\in W(M)$ such that $M^{\prime}=M^{w}$ and that $\tau^{\prime}=\tau^{w}$. Here we put

$$W(M)=\left\{w\in W\,\middle|\,\begin{matrix}\text{$wMw^{-1}$ is a standard Levi subgroup of $G_{n}$}\\ \text{and $w$ has a minimal length in $wW_{M}$}\end{matrix}\right\}$$

where $W_{M}$ is the Weyl group of $M$.