Abstract
We analyze the “eigenbundle” (localization bundle) of certain Hilbert modules over bounded symmetric domains of rank r, giving rise to complex-analytic fibre spaces which are stratified of length \(r+1.\) The fibres are described in terms of Kähler geometry as line bundle sections over flag manifolds, and the metric embedding is determined by taking derivatives of reproducing kernel functions. Important examples are the determinantal ideals defined by vanishing conditions along the various strata of the stratification.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
Jörg Eschmeier was a master in the interplay between operator theory and multi-variable complex analysis, in particular in its modern sheaf-theoretic form. An interesting concept in this respect is the “eigenbundle” of a Hilbert module arising from a commuting tuple of non-selfadjoint operators \(T_1,\ldots ,T_d.\) Just as the spectral theorem is the basic tool for the analysis of self-adjoint operators, the eigenbundle plays a similar role in the non-selfadjoint case, naturally involving multi-variable complex analysis instead of “real” measure theory.
In the original approach by Cowen-Douglas [6] for Hilbert modules H of holomorphic functions on a bounded domain \(D\subset {{\textbf{C}}}^d,\) the eigenbundle, denoted by is a genuine holomorphic vector bundle on D whose hermitian geometry determines the underlying operator tuple. In more general situations [8, 9] the eigenbundle will have singularities along certain subvarieties of D, making the geometric aspects more complicated. For example, if \(H={\overline{I}}\) is the Hilbert closure of a prime ideal I of polynomials, whose vanishing locus X is smooth, then by a result of Duan-Guo [8] the eigenbundle has rank 1 on the regular set \(D{\setminus }X,\) whereas is isomorphic to the (dual) normal bundle of X. Thus we have a “stratification” of length 2. In this paper, we study K-invariant Hilbert modules H over bounded symmetric domains \(D=G/K\) of rank r, leading to singular vector bundles which are stratified of length \(r+1.\) This study was initiated in [23] where the eigenbundle of certain polynomial ideals \(J^\lambda ,\) for a given partition \(\lambda \) of length \(\le r,\) was determined explicitly. These ideals are not prime ideals except for the “fundamental” partitions \(\lambda =(1,\ldots ,1,0,\ldots ,0).\) The main result of [23] describes the fibres of the eigenbundle using representation theory of the compact Lie group K.
The current paper extends and generalizes this analysis. For the partition ideals \(J^\lambda \) and Hilbert closures \(H={\overline{J}}^\lambda \) we construct a Hilbert space embedding of the eigenbundle by taking certain derivatives of the reproducing kernel function of H. This is important to study the hermitian structure and is related to the “jet construction” introduced in [9]. Moreover, we give a holomorphic characterization of the eigenbundle in terms of holomorphic sections of line bundles over a flag manifold. Such a geometric characterization may also hold in more general situations. Beyond the setting of the partition ideals we consider arbitrary K-invariant polynomial ideals, in particular the so-called “determinantal” ideals which have a direct geometric meaning.
2 Hilbert Modules and Their Eigenbundle
Let D be a bounded domain in a finite dimensional complex vector space \(E\approx {{\textbf{C}}}^d.\) Denote by \({\mathcal {P}}_E\approx {{\textbf{C}}}[z_1,\ldots ,z_d]\) the algebra of all polynomials on E. A Hilbert space H of holomorphic functions f on D (supposed to be scalar-valued) is called a Hilbert module if for any polynomial \(p\in {\mathcal {P}}_E\) the multiplication operator \(T_pf:=pf\) leaves H invariant and is bounded. Using the adjoint operators \(T_p^*,\) the closed linear subspace
is called the joint eigenspace at \(\zeta \in D.\) Since \(T_pT_q=T_{pq}\) for polynomials p, q it suffices to consider linear functionals or just the coordinate functions. The disjoint union
becomes a subbundle of the trivial vector bundle \(D\times H,\) which is called the eigenbundle of H, although it is not locally trivial in general. One also requires that the fibres have finite dimension and their union is total in H. The map \(\phi \mapsto [\phi ]\) from to \(H/{\overline{{\mathcal {M}}_\zeta H}}\) is a Hilbert space isomorphism, with inverse map
where is the orthogonal projection. Thus becomes the “quotient module” for the submodule \({\overline{{\mathcal {M}}_\zeta H}}.\)
Classical examples of Hilbert modules are the Bergman space \(H^2(D)\) of square-integrable holomorphic functions, whose reproducing kernel is called the Bergman kernel, and the Hardy space \(H^2(\partial D)\) if D has a smooth boundary \(\partial D.\) For general Hilbert modules H, a reproducing kernel function is a sesqui-holomorphic function \({\mathcal {K}}(z,\zeta )\) on \(D\times D\) such that for each \(\zeta \in D\) the holomorphic function
belongs to H, and we have
for all \(\psi \in H\) and \(z\in D.\) Here \((\phi |\psi )_H\) is the inner product, anti-linear in \(\phi .\) Thus H is the closed linear span of the holomorphic functions \({\mathcal {K}}_\zeta ,\) where \(\zeta \in D\) is arbitrary. If \(\phi _\alpha \) is any orthonormal basis of H then
For each \(\zeta \in D\) we have as follows from the identity
for \(p\in {\mathcal {P}}_E\) and \(\psi \in H.\)
If the reproducing kernel \({\mathcal {K}}\) has no zeros (e.g., the Bergman kernel of a strongly pseudo-convex domain) then the eigenbundle is spanned by the functions \({\mathcal {K}}_\zeta \) and hence becomes a hermitian holomorphic line bundle. In more general cases the kernel function vanishes along certain analytic subvarieties of D and the eigenbundle is not locally trivial, its fibre dimension can jump along the varieties and we obtain a singular vector bundle on D, also called a “linearly fibered complex analytic space” [13]. Such singular vector bundles are important in Several Complex Variables since they are in duality with the category of coherent analytic module sheaves, whereas (regular) vector bundles correspond to locally free sheaves. In [4] the connection to coherent analytic module sheaves associated with H is made explicit.
An important class of Hilbert modules is given by the Hilbert closure \(H={\overline{I}}\) of a polynomial ideal \(I\subset {\mathcal {P}}_E.\) In this case the fibres (1.1) of the eigenbundle have finite dimension. More precisely, by [8] we have for the “localization bundle” over E, with fibre
at \(\zeta \in E.\) For any set of generators \(p_1,\ldots ,p_t\) of I the linear map
is surjective, showing that In [4, Lemma 2.3] it is shown that satisfies
for all i. This implies that the eigenbundle restricted to the open dense subset
of D is a holomorphic line bundle spanned by the reproducing kernel \({\mathcal {K}}_\zeta ,\ \zeta \in {\check{D}}.\)
The behavior of on the singular set \(D{\setminus }{\check{D}}\) is more complicated and has so far been studied mostly when the vanishing locus of the reproducing kernel is a smooth subvariety of D, for example given as a complete intersection of a regular sequence of polynomials. The case where I is a prime ideal whose vanishing locus X consists of smooth points has been studied by Duan-Guo [8]. They showed that for \(\zeta \in D{\setminus }X\)
is the 1-dimensional span of the reproducing kernel vector, whereas for \(\zeta \in X\)
is isomorphic to the normal space (more precisely its linear dual.) Thus we have a stratification of length 2. We consider a more complicated situation for bounded symmetric domains D of arbitrary rank r, where we have a stratification of length \(r+1\), the relevant varieties are not smooth and the ideal I is not prime in general.
3 K-invariant Ideals on Bounded Symmetric Domains
Let \(D=G/K\) be an irreducible bounded symmetric domain of rank r, realized as the (spectral) unit ball of a hermitian Jordan triple (\(J^*\)-triple) E. Let \(\{uv^*w\}\in E\) denote the Jordan triple product of \(u,v,w\in E.\) The compact Lie group K acts by linear transformations on E preserving the Jordan triple product. For background on the Jordan theoretic approach towards symmetric domains, see [2, 5, 12, 17, 21]. Let \({\mathcal {P}}_E\) denote the algebra of all polynomials on E. Under the natural action
of \(k\in K\) on functions f on E the polynomial algebra \({\mathcal {P}}_E\) has a Peter–Weyl decomposition [11, 20]
into pairwise inequivalent irreducible K-modules \({\mathcal {P}}_E^\lambda .\) Here \({{\textbf{N}}}_+^r\) denotes the set of all partitions
of integers \(\lambda _1\ge \ldots \ge \lambda _r\ge 0.\) The polynomials in \({\mathcal {P}}_E^\lambda \) are homogeneous of degree \(|\lambda |:=\lambda _1+\ldots +\lambda _r.\) We often identify a partition \(\lambda \) with its Young diagram
For fixed \(\mu \in {{\textbf{N}}}_+^r\) denote by
the K-invariant projection onto \({\mathcal {P}}_E^\mu .\) As a consequence of Schur orthogonality we have [23, Lemma 3.1]
where \(\chi _\mu \) denotes the character of the K-representation on \({\mathcal {P}}_E^\mu .\)
An ideal \(I\subset {\mathcal {P}}_E\) is called K-invariant if \(k\cdot f\in I\) for all \(k\in K\) and \(f\in I.\) A similar definition applies to Hilbert modules of holomorphic funtions on a K-invariant domain. The formula (2.1) implies that a K-invariant ideal (resp., Hilbert module) is a direct sum (resp., Hilbert sum) of its Peter-Weyl subspaces \({\mathcal {P}}_E^\mu .\)
For a given partition \(\lambda \) denote by \(J^\lambda \subset {\mathcal {P}}_E\) the K-invariant ideal generated by \({\mathcal {P}}_E^\lambda .\) The first main result of [23] asserts that \(J^\lambda \) has the Peter-Weyl decomposition
where \(\lambda \le \mu \) means \(\lambda _i\le \mu _i\) for all i. This is equivalent to the inclusion \([\lambda ]\subset [\mu ]\) for the corresponding Young diagrams. As a consequence, \(J^\mu \subset J^\lambda \) if and only if \(\mu \ge \lambda .\) In this section we show that these “partition” ideals \(J^\lambda \) are fundamental for the study of general K-invariant ideals. Given a K-invariant ideal I, define \(I^\#:=\{\lambda \in {{\textbf{N}}}_+^r:\ J^\lambda \subset I\}.\)
Proposition 2.1
Let \(I\subset {\mathcal {P}}_E\) be a K-invariant ideal. Then there is a finite set \(\Lambda \subset I^\#\) of partitions such that
Proof
If \(f\in I\) then (2.1) shows that \(f^\mu \) also belongs to I for all \(\mu \in {{\textbf{N}}}_+^r.\) Let \(f_1,\ldots ,f_t\) be a finite set of generators of I. Then their non-zero K-homogeneous parts belong to I and form a finite set of generators. Thus we may assume that each \(f_s\in {\mathcal {P}}_E^{\lambda _s}\) for some partition \(\lambda _s.\) We claim that
Since \(f_s\in I\cap {\mathcal {P}}_E^{\lambda _s}\) is non-zero, I is K-invariant and \({\mathcal {P}}_E^{\lambda _s}\) is irreducible, it follows that \({\mathcal {P}}_E^{\lambda _s}\subset I\) and hence \(J^{\lambda _s}\subset I.\) Thus \(J\subset I.\) Conversely, each generator \(f_s\) of I belongs to J. Therefore \(I\subset J.\) \(\square \)
A subset \(\Lambda \subset I^\#\) is called “full” if
A subset \(A\subset I^\#\) is called “minimal” if \(\alpha \in A\) and \(\lambda <\alpha \) implies \(\lambda \notin I^\#,\) i.e.,
Lemma 2.2
Let \(A\subset I^\#\) be minimal and \(\Lambda \subset I^\#\) be full. Then \(A\subset \Lambda .\)
Proof
Suppose there exists \(\alpha \in A\) such that \(\alpha \notin \Lambda .\) Let \(p\in {\mathcal {P}}_E^\alpha .\) Then \(p\in J^\alpha \subset I=\sum \limits _{\lambda \in \Lambda }J^\lambda \) and therefore
for some \(f_\lambda \in J^\lambda .\) By (2.2) we have \(f_\lambda ^\alpha =0\) unless \(\alpha \ge \lambda .\) Since \(\alpha \notin \Lambda \) this implies
since \(\{\lambda \in \Lambda :\ \lambda <\alpha \}=\emptyset .\) This is a contradiction. \(\square \)
Corollary 2.3
Every minimal set \(A\subset I^\#\) is finite.
Proof
By Proposition 2.1 there exists a full set \(\Lambda \subset I^\#\) which is finite. By Lemma 2.2 we have \(A\subset \Lambda .\) Hence A is finite. \(\square \)
Proposition 2.4
There exists a (finite) set \(A\subset I^\#\) which is both full and minimal.
Proof
For any finite subset \(\Lambda \subset I^\#\) we put
where \(|\lambda |:=\lambda _1+\ldots +\lambda _r.\) By Proposition 2.1 there exists a finite subset \(\Lambda \subset I^\#\) which is full. Put
Then there exists a full and finite set \(A\subset I^\#\) such that \(|A|=k.\) We claim that A is minimal. Suppose there exist \(\alpha \in A\) and \(\lambda \in I^\#\) with \(\lambda <\alpha .\) Then \(J^\alpha \subset J^\lambda \subset I\) which shows that the finite set \(\Lambda =(A{\setminus }\{\alpha \})\cup \{\lambda \}\) is still full. On the other hand, we have \(|\lambda |<|\alpha |\) and hence
This contradiction shows that A is minimal. \(\square \)
The arguments above show that there is a unique finite set \(I^\#_{min}\subset I^\#\) which is both full and minimal. We formulate this as
Proposition 2.5
Let \(I\subset {\mathcal {P}}_E\) be a K-invariant ideal. There exists a unique finite set \(\Lambda \subset {{\textbf{N}}}_+^r\) such that
and if \(\mu <\lambda \) for some \(\lambda \in \Lambda .\)
For example, the n-th power of the maximal ideal \({\mathcal {M}}_0\) has the form
since the polynomials in \({\mathcal {P}}_E^\lambda \) are homogeneous of degree \(|\lambda |.\) For \(n=1\) we have \({\mathcal {M}}_0=I^{1,0,\ldots ,0}\) since \({\mathcal {P}}_E^{1,0,\ldots ,0}=E^*\) is the linear dual space of E. More interesting examples will be studied in the next section.
As an application of Proposition 2.5 we determine for each K-invariant ideal I the “maximal fibre” of the eigenbundle.
Theorem 2.6
Let I be a K-invariant ideal, written in the “minimal” form (2.3). Then any Hilbert module closure \(H={\overline{I}}\) has the maximal fibre
at the origin.
Proof
We first show that for \(\lambda \in \Lambda .\) Let \(p\in {\mathcal {P}}_E^\lambda \) and \(\ell \in E^*\) a linear form on E. By [22] we have
If \(T_\ell ^*p\ne 0\) then \((T_\ell ^*p|q)\ne 0\) for some \(q\in J^\mu \) with \(\mu \in \Lambda .\) By (2.2) we have
Therefore
It follows that there exists j such that \(\lambda -\varepsilon _j=\nu .\) Therefore \(\lambda >\nu \ge \mu .\) Since both \(\lambda ,\mu \in \Lambda \) this contradicts the fact that \(\Lambda \) is minimal.
Conversely, suppose there exists which is orthogonal to \(\sum \nolimits _{\lambda \in \Lambda }{\mathcal {P}}_E^\lambda .\) By averaging over K we may assume that \(\phi \in {\mathcal {P}}_E^\mu \) for some \(\mu \notin \Lambda .\) We can write \(\phi \in H={\overline{I}}\) as
where \(\phi _\lambda \) belongs to the ideal \(J^\lambda .\) Since \(J^\lambda \) is generated by \({\mathcal {P}}_E^\lambda ,\) there exist \(g_\lambda ^i\in {\mathcal {M}}_0,\ a_\lambda ^i\in {{\textbf{C}}}\) and \(p_i^\lambda \in {\mathcal {P}}_E^\lambda \) such that
Since \(\phi \in {\mathcal {P}}_E^\mu \) applying the projection \(\pi ^\mu \) yields
It follows that
Therefore \(\phi =0.\) \(\square \)
As a special case of Theorem 2.6 we have
already proved in [23]
The description of the eigenbundle at non-zero points \(\zeta \) is more complicated and depends on the rank of \(\zeta \) (in a Jordan theoretic sense). For \(0\le \ell \le r\) define the Kepler manifold
The complexification \({\widehat{K}}\) of K is a complex subgroup of \({\textrm{GL}}(E)\) called the “structure group.” It acts transitively on each \({\check{E}}_\ell .\) An element \(c\in E\) satisfying \(\{cc^*c\}=2c\) is called a tripotent (triple idempotent). Let \(S_\ell \subset {\check{E}}_\ell \) denote the compact K-homogeneous manifold of all tripotents of rank \(\ell .\) It is shown in [23] that the eigenbundle restricted to each “stratum” \({\check{E}}_\ell \) is a homogeneous holomorphic vector bundle under the \({\widehat{K}}\)-action, induced by the fibre at any tripotent c of rank \(\ell .\) Hence it suffices to study the fibre at a tripotent c. In terms of the Peirce decomposition [16, 17]
where \(E_c^j:=\{z\in E:\ \{cc^*z\}=jz\},\) the tangent space at c is given by
Therefore the Peirce 0-space \(E_c^0\) can be identified with the normal space at c. We abbreviate \(E_c:=E_c^2,\ E^c:=E_c^0\) and note that \(E_c\) and \(E^c\) are irreducible \(J^*\)-subtriples of E having rank \(\ell \) and \(r-\ell ,\) respectively. (The Peirce 1-space \(E_c^1\) is also a Jordan subtriple, but not necessarily irreducible).
The principal tool to analyze the fibre is the normal projection map
onto the polynomial algebra \({\mathcal {P}}_W\) of the Peirce 0-space \(W:=E^c\) of a tripotent c of rank \(\ell .\) Since W is an irreducible \(J^*\)-triple of complementary rank \(r-\ell ,\) the polynomial algebra \({\mathcal {P}}_W\) has its own Peter-Weyl decomposition
with respect to the automorphism group \(K_W.\) Here we write partitions \(\alpha \in {{\textbf{N}}}_+^{r-\ell }\) in the form \(\alpha =(\alpha _{\ell +1},\ldots ,\alpha _r)\) with \(\alpha _{\ell +1}\ge \ldots \ge \alpha _r\ge 0.\) For any partition \(\lambda \) of length r we define the “truncated partition”
Let \(J_W^{\lambda ^*}\subset {\mathcal {P}}_W\) denote the ideal generated by \({\mathcal {P}}_W^{\lambda ^*}.\)
Proposition 2.7
For any K-invariant ideal I, written in the “minimal” form (2.3), and any tripotent c with Peirce 0-space W, the normal projection map (2.5) maps I into the \(K_W\)-invariant ideal
and hence induces a mapping
between the localization of I at c and the maximal fibre relative to W.
Proof
By [23, Theorem] the normal projection map satisfies
for any partition \(\lambda \in {{\textbf{N}}}_+^r.\) In other words, if \(f\in {\mathcal {P}}_E\) has only K-components for \(\mu \ge \lambda ,\) then \(\pi _c f\in {\mathcal {P}}_W\) has only \(K_W\)-components for partitions \(\alpha \in {{\textbf{N}}}_+^{r-\ell }\) satisfying \(\alpha \ge \lambda ^*.\) One can show that each such partition \(\alpha \) occurs. Taking the (finite) sum over \(\lambda \in \Lambda ,\) the first assertion follows. Since \((\pi _c f)(0)=f(c+0)=f(c)\) it follows that \(\pi _c\) maps the maximal ideal \({\mathcal {M}}_c\) to the maximal ideal \({\mathcal {M}}_{W,0}\subset {\mathcal {P}}_W\) relative to W. This implies the second assertion. \(\square \)
The main result of [23] asserts that for \(I=J^\lambda \) the map
is an isomorphism. Since (2.4) applied to W yields an isomorphism
by taking the lowest \(K_W\)-type \(\phi ^{\lambda ^*}=\pi _W^{\lambda ^*}\phi \) of \(\phi \in J_W^{\lambda ^*},\) (2.8) amounts to the isomorphism
where \(\pi _c^{\lambda ^*}f:=(\pi _c f)^{\lambda ^*}\) denotes the lowest \(K_W\)-type of \(\pi _c f\in J_W^{\lambda ^*}.\) It is likely that in general the map (2.6) is an isomorphism. This would reduce the description of the fibres to the combinatorial problem of finding a mimimal subset of the set \(\{\lambda ^*:\ \lambda \in \Lambda \}.\) This will be carried out separately.
4 Determinantal Ideals
While the ideals \(J^\lambda \) are defined in terms of representation theory and make no sense beyond the Jordan theoretic setting, we now introduce K-invariant ideals which are defined by vanishing conditions along certain subvarieties of E. The resulting analysis could shed some light on more general situations. For any ideal \(I\subset {\mathcal {P}}_E\) define the vanishing locus
A closed subset \(X\subset E\) of the form \(X={\mathcal {V}}^I\) for some ideal \(I\subset {\mathcal {P}}_E\) is called an algebraic variety. Conversely, for an algebraic variety \(X\subset E\) consider the vanishing ideal
Here \({\mathcal {M}}_\zeta \subset {\mathcal {P}}_E\) denotes the maximal ideal of all polynomials vanishing at \(\zeta \in E.\) By Hilbert’s basis theorem, \({\mathcal {M}}_X\) has a finite set of generators. Hence any algebraic variety is the vanishing locus of finitely many polynomials. An ideal I is prime if and only if the algebraic variety \({\mathcal {V}}^I\) is irreducible. Conversely, an algebraic variety X is irreducible if and only if \({\mathcal {M}}_X\) is a prime ideal. In general, we have
and, by Hilbert’s Nullstellensatz,
We say that \(f\in {\mathcal {P}}_E\) has order of vanishing \({\textrm{ord}}_\zeta (f)\ge n\) at \(\zeta \in E\) if \(f\in {\mathcal {M}}_\zeta ^n.\) Equivalently, \({\textrm{ord}}_\zeta (f)>n\) if the n-th Taylor polynomial of f at \(\zeta \) vanishes. Given an irreducible algebraic variety \(X\subset E\) one defines the n-th symbolic power
consisting of all polynomials which vanish of order \(\ge n\) on X. These are “primary ideals” associated to the prime ideal
The algebraic power \({\mathcal {M}}_X^n\) is contained in \({\mathcal {M}}_X^{(n)}\) but is generally smaller if \(n>1.\) For a thorough discussion of such matters, cf. [7].
For any algebraic variety X let \({\check{X}}\) denote the open dense subset of all smooth (regular) points. The complement \(X{\setminus }{\check{X}}\) is the singular set of X. A nested sequence \(X_0\subset X_1\subset \;\subset X_{r-1}\) of algebraic varieties is called a stratification if for each \(\ell \le 1\) the algebraic variety \(X_\ell \) has the singular set \(X_{\ell -1},\) i.e., the smoooth points
We assume that the lowest stratum \(X_0={\check{X}}_0\) is smooth, so that \(X_{-1}=\emptyset .\) We put \({\check{X}}_r:=E{\setminus }X_{r-1}\) as an open dense subset of E. The sets \({\check{X}}_\ell \) for \(0\le \ell \le r\) are called the strata of the stratification. The \(\ell \)-th stratum \({\check{X}}_\ell \) has the closure
Thus \(X_0\) is the only closed stratum. Now consider an r-tuple
of integers \(n_1\ge n_2\ge \ldots \ge n_r\ge 0,\) and define the joint symbolic power
consisting of all polynomials which vanish of order \(\ge n_j\) along the subvariety \(X_{j-1}.\) As special cases we have
and in particular, for the prime ideal (if \(X_j\) is irreducible)
An irreducible \(J^*\)-triple E has a canonical stratification
where \({\widehat{E}}_j\) is the set of all elements \(\zeta \in E\) of rank \(\le j,\) called the j-th Kepler variety (in a Jordan theoretic setting). The smooth points of \({\widehat{E}}_j\) form the Kepler manifold \({\check{E}}_j\) defined above. Hence the singular set of \({\widehat{E}}_j\) is \({\widehat{E}}_{j-1}\) so that the condition (3.1) is satisfied. As a special case of (3.3) define the joint symbolic power
associated with the decreasing tuple (3.2). Thus \({\mathcal {M}}_E^{({{\varvec{n}}})}\) consists of all polynomials on E which vanish of order \(\ge n_j\) along the subvariety \({\widehat{E}}_{j-1}.\) These ideals are called “determinantal ideals” since the Kepler varieties are defined by vanishing conditions for Jordan theoretic determinants and minors. Since the Kepler varieties \({\widehat{E}}_j\) are K-invariant, \({\mathcal {M}}_E^{({{\varvec{n}}})}\) is a K-invariant ideal. As such, it is a sum of certain “partition” ideals \(J^\lambda .\) Our next result makes this precise.
An irreducible Jordan algebra E with unit element e has a unique determinant polynomial \(\Delta _e:E\rightarrow {{\textbf{C}}}\) normalized by \(\Delta _e(e)=1\) [12, 19]. For the matrix algebra \(E={{\textbf{C}}}^{r\times r}\) and the symmetric matrices \(E={{\textbf{C}}}_{{\textrm{sym}}}^{r\times r}\) this is the usual determinant. For the antisymmetric matrices \(E={{\textbf{C}}}_{{\textrm{asym}}}^{2r\times 2r}\) we obtain the Pfaffian determinant instead. The determinant polynomial \(\Delta _e\) has the semi-invariance property
for all \(k\in K\) and \(z\in E.\) The map \(\chi :K\rightarrow {{\textbf{T}}}\) defined by
is a character of K. It follows that for any \(k\in K\)
is a Jordan determinant normalized at ke.
Theorem 3.1
For each tuple (3.2) the joint symbolic power has the decomposition
where
Proof
By highest weight theory [21] the space \({\mathcal {P}}_E^\lambda \) is spanned by polynomials
where \(e_1,\ldots ,e_r\) is any frame of E and
denotes the Jordan theoretic minor for the tripotent \(e=e_{[m]}=e_1+\ldots +e_m\) and its Peirce 2-space \(E_e.\) Here \(P_e:E\rightarrow E_e\) is the Peirce 2-projection. Let \(\zeta \in {\widehat{E}}_j\) and \(m>j.\) By the spectral theorem applied to \(E_e\) there exist a frame \(c_1,\ldots ,c_m\) of \(E_e\) and \(a_i\in {{\textbf{C}}}\) such that
Choose \(k\in K_{E_e}\) such that \(ke_i=c_i\) for all \(i\le m.\) By [19, Theorem 1] we have
for all \(u\in E_e,\) where \(e_T:=\sum \limits _{i\in T}e_i.\) For \(z\in E\) it follows that
Since \(P_e\zeta \) has rank \(\le j\) it follows that at most j coefficients \(a_i\) are non-zero, and hence \(\prod \limits _{i\in T}a_i=0\) whenever the cardinality \(|T|>j.\) Hence in the sum only subsets T with \(|T|\le j\) occur. Since the polynomial \(\Delta _{e-e_T}\circ P_{e-e_T}k^{-1}P_e\) on E is homogeneous of degree \(m-|T|={\textrm{rank}}(E_{e-e_T})\) we obtain
It follows that
Hence the estimate \({\textrm{ord}}_\zeta (f)\ge \sum \nolimits _{m=j+1}^r\lambda _m\) holds for all \(f\in {\mathcal {P}}_E^\lambda \) and a fortiori for all \(f\in J^\lambda .\) This shows that \(J^\lambda \subset {\mathcal {M}}_E^{({{\varvec{n}}})}\) whenever \(\lambda \in {{\textbf{N}}}_{({{\varvec{n}}})}^r.\)
On the other hand, the tripotent \(c=e_{[j]}=e_1+\ldots +e_j\in {\check{E}}_j\) satisfies
It follows that
Thus if \(\lambda \) is a partition such that \(J^\lambda \subset {\mathcal {M}}_E^{({{\varvec{n}}})}\) then \(N^\lambda \in {\mathcal {M}}_E^{({{\varvec{n}}})}\) and (3.6) implies
for all \(0\le j\le r-1.\) Therefore \(\lambda \in {{\textbf{N}}}_{({{\varvec{n}}})}^r.\) \(\square \)
The set \(\Lambda \) determined in Theorem 3.1 is not minimal. For example \({{\varvec{n}}}=(10,5,1)\) has the minimal partitions (5, 4, 1) and (5, 3, 2), whereas \({{\varvec{n}}}=(15,5,1)\) has the minimal partitions (10, 4, 1), (9, 5, 1), (8, 6, 1), (10, 3, 2), (9, 4, 2), (8, 5, 2), (9, 3, 3) and (8, 4, 3).
We next study the normal projection map (2.5) for determinantal ideals. Given \({{\varvec{m}}}=(m_{\ell +1},\ldots ,m_r)\in {{\textbf{N}}}_+^{n-\ell }\) we put
Since \({\textrm{rank}}_E(c+w)=\ell +{\textrm{rank}}_W(w)\) it follows that
Theorem 3.2
Let c be a tripotent of rank \(\ell .\) For \({{\varvec{n}}}\in {{\textbf{N}}}_+^r\) put
Then the normal projection map \(\pi _c\) satisfies
In particular, for \(j>\ell \)
Proof
By Theorem 3.1 we may assume that \(f\in {\mathcal {M}}_E^{({{\varvec{n}}})}\) belongs to \(J^\lambda \) for a partition \(\lambda =(\lambda _1,\ldots ,\lambda _r)\) satisfying
for all \(1\le j\le r.\) Now fix \(\ell \) and consider the partition \(\lambda ^*:=(\lambda _{\ell +1},\ldots ,\lambda _r)\) of length \(r-\ell .\) By (2.7) we have
where \(\alpha \ge \lambda ^*\) is the (partial) containment order, i.e. \(\alpha _i\ge \lambda _i\) for all \(\ell <i\le r.\) For \(\ell <j\le r\) we have
Therefore \(\alpha \) satisfies the analogue of (3.8) relative to W. Applying Theorem 3.1 to W and the sequence \(n_{\ell +1}\ge \cdots \ge n_r\) of length \(r-\ell ={\textrm{rank}}(W),\) it follows that \(\pi _c J^\lambda \subset {\mathcal {M}}_W^{({{\varvec{n}}}^*)}.\) The special case (3.7) corresponds to \(n_1=\cdots =n_j=n,\ n_{j+1}=\cdots =n_r=0.\) \(\square \)
We now consider the special case of “step 1” partitions. Let \(n\in {{\textbf{N}}}\) and consider the ideal
where \(0\le \ell \le r\) is fixed. This corresponds to \({{\varvec{n}}}=(n^{(\ell +1)},0^{(r-\ell -1)}).\) In this case Theorem 3.1 yields
For example
and
For the K-invariant ideals (3.9) it is easy to find a minimal decomposition:
Theorem 3.3
For \(0\le \ell <r\) the ideal \({\mathcal {M}}_{{\widehat{E}}_\ell }^{(n)}\) is the minimal and finite sum
taken over the (finitely many) integer tuples \(\alpha _{\ell +1}\ge \ldots \ge \alpha _r\ge 0\) satisfying
Here we put
Proof
We first show that the sum is minimal. If \(\alpha ,\beta \in {{\textbf{N}}}_+^{r-\ell }\) satisfy \(|\alpha |=n=|\beta |\) and \({\widehat{\alpha }}\ge {\widehat{\beta }},\) then \(\alpha \ge \beta \) and hence \(\alpha =\beta ,\) showing that \({\widehat{\alpha }}={\widehat{\beta }}.\) Of course there are only finitely many partitions \(\alpha =(\alpha _{\ell +1},\ldots ,\alpha _r)\) satisfying \(|\alpha |=n.\) By Theorem 3.1\({\mathcal {M}}_{{\widehat{E}}_\ell }^{(n)}\) corresponds to partitions \(\lambda \in {{\textbf{N}}}_+^r\) such that the single inequality
holds. Clearly, \(\lambda ={\widehat{\alpha }}\) satisfies (3.10), since \(\lambda _i\ge \alpha _i\) for \(\ell <i\le r.\) Thus it remains to show that (3.10) implies \(\lambda \ge {\widehat{\alpha }}\) for some \(\alpha \in {{\textbf{N}}}_+^{r-\ell }\) with \(|\alpha |=n.\) If \(\lambda _{\ell +1}+\ldots +\lambda _r\ge n\) then there exists \(\alpha _{\ell +1}\ge \ldots \ge \alpha _r\) such that \(|\alpha |=n\) and \(\lambda _i\ge \alpha _i\) for \(\ell <i\le r.\) To see this, we may assume (by induction) that \(\lambda _{\ell +1}+\ldots +\lambda _r=n+1.\) Let
where \(\ell <m\le r.\) Then \(\alpha :=(\lambda _{\ell +1},\ldots ,\lambda _{m-1},\lambda _m-1,0,\ldots ,0)\) is decreasing, satisfies \(|\alpha |=n\) and \(\lambda \ge {\widehat{\alpha }}\) since for \(1\le i\le \ell \) we have \({\widehat{\alpha }}_i={\left\{ \begin{array}{ll}\lambda _{\ell +1}&{}m>\ell +1\\ \lambda _{\ell +1}-1&{}m=\ell +1\end{array}\right. }\) and therefore \({\widehat{\alpha }}_i\le \lambda _{\ell +1}\le \lambda _i.\) \(\square \)
In the simplest case \(n=1\) there is only a single choice \(\alpha _{\ell +1}=1,\ \alpha _{\ell +2}=\;=\alpha _r=0\) yielding the “fundamental partition” \({\widehat{\alpha }}=(1^{(\ell +1)},0^{(r-\ell -1)})\equiv 1^{(\ell +1)}.\) Thus the prime ideal \({\mathcal {M}}_{{\widehat{E}}_\ell }^{(1)}={\mathcal {M}}_{{\widehat{E}}_\ell }\) has the form
For \(n>1\) we cannot represent \({\mathcal {M}}_{{\widehat{E}}_\ell }^{(n)}\) by a single partition.
5 Reproducing Kernels and Hermitian Structure
The main result of [23], formulated as the isomorphism (2.9), determines the localization bundle as an abstract (singular) holomorphic vector bundle, without reference to a hermitian metric. If \(H={\overline{I}}\) is a Hilbert module completion of a K-invariant ideal I, with reproducing kernel function \({\mathcal {K}}(z,\zeta ),\) the corresponding eigenbundle carries the all-important hermitian structure as a subbundle of \(D\times H.\) To make the connection one needs an explicit embedding
As suggested by the “jet construction” developed in [9] for smooth submanifolds, such a map should involve certain derivatives of \({\mathcal {K}}(z,\zeta )\) in the normal direction. In this section we carry out this program in the more complicated geometric situation related to an arbitrary partition ideal \(I=J^\lambda .\)
Any polynomial p induces a constant coefficient differential operator \({\overline{p}}(\partial _z)\) on E, depending in a conjugate-linear way on p. Let (p|q) denote the Fischer-Fock inner product of polynomials \(p,q\in {\mathcal {P}}_E\) (anti-linear in p) and let \({\mathcal {E}}^\lambda (z,\zeta )={\mathcal {E}}_\zeta ^\lambda (z)\) denote the reproducing kernel of \({\mathcal {P}}_E^\lambda .\)
Every irreducible \(J^*\)-triple E has two “characteristic multiplicities” a, b [2, 16, 17, 21] such that
Lemma 4.1
For a Jordan triple E, the determinant function \(N_e\) at a maximal tripotent \(e\in S_r\) satisfies
for \(\lambda \ge n^{(r)}\) and \(\zeta \in E_e,\) where
Moreover,
Proof
We use the Jordan theoretic Pochhammer symbols \((s)_\lambda \) and the “Faraut-Korányi formula” [11, 12]. Suppose first that \(E=E_e\) is unital, with unit element e and determinant \(\Delta _e.\) The parameter \(s=d/r\) in the continuous Wallach set corresponds to the Hardy space \(H^2(S)\) over the Shilov boundary S of D. Since \(|\Delta _e|=1\) on S we obtain for \(p,q\in {\mathcal {P}}_E^\lambda \)
Since q is arbitrary, it follows that
for all partitions \(\lambda \) and \(p\in {\mathcal {P}}_E^\lambda .\) An application of Schur orthogonality [12, Proposition XI.4.1] shows that
where \(d_\lambda :=\dim {\mathcal {P}}_E^\lambda .\) If \(\lambda \ge n^{(r)}\) we can write
for some coefficient \(a_\lambda .\) It follows that
Since \(d_\lambda =d_{\lambda -n^{(r)}}\) in the unital case it follows that
This proves (4.1). Moreover,
In the non-unital case, let \(P_e\) be the Peirce 2-projection onto \(E_e.\) Since \(N_e(z)=\Delta _e(P_ez)\) by definition, applying the unital case to \(E_e\) and using \(P_e\zeta =\zeta \) we obtain
The second assertion follows with \({\overline{\Delta }}_e^n(\partial _z)f={\overline{N}}_e^n(\partial _z)(f\circ P_e).\) \(\square \)
The identity (4.2), written as
implies
since \({\overline{{\mathcal {E}}^\lambda (z,\zeta )}}={\mathcal {E}}^\lambda (\zeta ,z).\) Thus for \(\zeta \in E_e\) we have
as holomorphic polynomials in z.
Lemma 4.2
Let \(\lambda \in {{\textbf{N}}}_+^r\) be a partition and \(\zeta \in E\) have rank \(\ell .\) Then \({\mathcal {E}}_\zeta ^\lambda \ne 0\) if and only if \(\lambda \) has length \(\le \ell .\)
Proof
If \(\lambda \) has length \(>\ell \) then all \(p\in {\mathcal {P}}_E^\lambda \) vanish on \({\widehat{E}}_\ell .\) Therefore
and hence \({\mathcal {E}}_\zeta ^\lambda =0\) since p is arbitrary. Now suppose \(\lambda \) has length \(\le \ell .\) Consider the spectral decomposition
for a frame \((e_i).\) The conical polynomial \(N^\lambda \) defined as in (3.5) satisfies
Since \(N^\lambda (\zeta )=({\mathcal {E}}_\zeta ^\lambda |N^\lambda )\) it follows that \({\mathcal {E}}_\zeta ^\lambda \ne 0.\) \(\square \)
For \(n\in {{\textbf{N}}},\) define \(n^{(m)}=(n,\ldots ,n,0,\ldots ,0),\) with n repeated m times. Thus the Young diagram \([n^{(m)}]=[1,m]\times [1,n].\) Any partition \(\lambda \) can be written as
where \(1\le \ell _1<\cdots <\ell _t\le r\) and \(n_1>n_2>\cdots>n_t>0.\) Thus t is the number of “steps” in the partition. In other words,
Define for \(1\le s\le t\)
We also consider the “regular” points
Since \(\lambda \) has length \(\ell _t,\) Lemma 4.2 implies \({\mathcal {E}}_\zeta ^\lambda \ne 0\) if \({\textrm{rank}}(\zeta )\ge \ell _t.\) Hence \({\mathcal {K}}_\zeta \) does not vanish at \(\zeta \in {\check{D}}.\) For the “singular” points \(\zeta \in D\) there exists a unique \(s\le t\) such that \(\zeta \in \Omega _s.\) For \(1\le h\le k\le t\) define
as a partition of length \(\ell _k-\ell _{h-1}.\) Let \(\mu \in {{\textbf{N}}}_+^{\ell _s}\) satisfy \(\mu \ge \lambda _1^s.\) For each \(s\le k\le t\) consider the partition
of length \(\ell _k.\) Here \(\lambda _{s+1}^k\) is empty for \(s=k.\) Starting with the Peter-Weyl expansion
of the reproducing kernel \({\mathcal {K}}(z,\zeta )\) of \(H={\overline{J}}^\lambda ,\) with coefficients \(a_\mu >0,\) we define K-invariant sesqui-holomorphic functions \({\mathcal {K}}^s(z,\zeta )={\mathcal {K}}_\zeta ^s(z)\) on \(D\times D\) by the Peter-Weyl expansion
More explicitly, the constant is given by
Lemma 4.3
The kernel \({\mathcal {K}}_\zeta ^s\) does not vanish if \({\textrm{rank}}(\zeta )\ge \ell _{s-1}\) and vanishes if \({\textrm{rank}}(\zeta )<\ell _{s-1}.\) Thus \({\mathcal {K}}^s\) vanishes precisely on \({\overline{\Omega }}_{s-1}=\bigcup \nolimits _{0\le k<s}\Omega _k.\)
Proof
Since the partition \(\lambda _1^s-n_s^{(\ell _s)}=\lambda _1^{s-1}\) has length \(\ell _{s-1}\) it follows that \({\mathcal {E}}_\zeta ^{\lambda _1^s-n_s^{(\ell _s)}}\ne 0\) if \({\textrm{rank}}(\zeta )\ge \ell _{s-1}.\) This implies \({\mathcal {K}}_\zeta ^s\ne 0\) since \({\mathcal {E}}^{\lambda _1^s-n_s^{(\ell _s)}}\) occurs as the lowest term in \({\mathcal {K}}^s.\) On the other hand, if \({\textrm{rank}}(\zeta )<\ell _{s-1}\) then \({\mathcal {E}}_\zeta ^{\mu -n_s^{(\ell _s)}}=0\) for all \(\mu \ge \lambda _1^s\) since \(\mu -n_s^{(\ell _s)}\ge \lambda _1^s-n_s^{(\ell _s)}=\lambda _1^{s-1}\) has length \(\ge \ell _{s-1}.\) \(\square \)
Proposition 4.4
Let \({\textrm{rank}}(\zeta )\le \ell _t\) and write \(\zeta =\lim _{\varepsilon \rightarrow 0}\zeta _\varepsilon ,\) where \(\zeta _\varepsilon \in {\check{E}}_{c_t}\) for some tripotent \(c_t\in S_{\ell _t}.\) Then \(N_{c_t}(\zeta _\varepsilon )\ne 0\) and
Similarly, for \(1\le s<t\) let \({\textrm{rank}}(\zeta )\le \ell _s\) and write \(\zeta =\lim _{\varepsilon \rightarrow 0}\zeta _\varepsilon ,\) where \(\zeta _\varepsilon \in {\check{E}}_{c_s}\) for some tripotent \(c_s\in S_{\ell _s}.\) Then \(N_{c_s}(\zeta _\varepsilon )\ne 0\) and
Proof
Since \(\zeta _\varepsilon \) has rank \(\ell _t,\) it follows that \({\mathcal {E}}_{\zeta _\varepsilon }^\lambda =0\) unless \(\lambda =(\mu ,0^{(r-\ell _t)}),\) where \(\mu \in {{\textbf{N}}}_+^{\ell _t}\) satisfies \(\mu \ge \lambda _1^t.\) Therefore
Since \(\mu \ge n_t^{(\ell _t)},\) we may apply (4.1) to \(E_{c_t}\) and \(n=n_t.\) This yields
It follows that
Now let \(s<t.\) Let \(\mu \in {{\textbf{N}}}_+^{\ell _{s+1}}\) satisfy \(\mu \ge \lambda _1^{s+1}.\) Since \({\textrm{rank}}(\zeta _\varepsilon )=\ell _s\) we have \({\mathcal {E}}_{\zeta _\varepsilon }^{\mu -n_{s+1}^{(\ell _{s+1})}}=0\) unless \(\mu =(\nu ,n_{s+1}^{(\ell _{s+1}-\ell _s)}),\) where \(\nu \in {{\textbf{N}}}_+^{\ell _s}\) satisfies \(\nu \ge \lambda _1^s.\) Applying definition (4.5) to \({\mathcal {K}}^{s+1}\) yields
using \((\nu ,n_{s+1}^{(\ell _{s+1}-\ell _s)},\lambda _{s+2}^t)=(\nu ,\lambda _{s+1}^t)\) and \((\nu ,n_{s+1}^{(\ell _{s+1}-\ell _s)})-n_{s+1}^{(\ell _{s+1})}=\nu -n_{s+1}^{(\ell _s)}.\) Since \(\nu \ge (n_s-n_{s+1})^{(\ell _s)},\) we may apply (4.1) to \(E_{c_s}\) and \(n=n_s-n_{s+1}.\) Since \(\nu -n_{s+1}^{(\ell _s)}-(n_s-n_{s+1})^{(\ell _s)}=\nu -n_s^{(\ell _s)}\) this yields
It follows that
\(\square \)
Let \(1\le s\le t.\) For any tripotent chain
of rank \(\ell _s<\ell _{s+1}< \cdots <\ell _t\) consider the polynomial
Theorem 4.5
Let \(1\le s\le t.\) Then for any \(\zeta \in \Omega _s\) we have
where the linear span is taken over all tripotent chains of rank \(\ell _s<\ell _{s+1}< \cdots <\ell _t\) such that \(\zeta \in E_{c_s}.\) In view of (4.2) we can write \(N_{c_s,\ldots ,c_t}^{n_s,\ldots ,n_t}\cdot {\mathcal {K}}_\zeta ^s\) also in the differential form \(N_{c_{s},\ldots ,c_{t}}^{n_{s},\ldots ,n_{t}}({\overline{\partial }}_{\zeta }){\widetilde{{\mathcal {K}}}}_{\zeta }^s\) for some modified kernel functions \({\widetilde{{\mathcal {K}}}}_{\zeta }^{s}.\)
Proof
We first show that for all \(\zeta \in E_{c_s}.\) In case \(s=t\) let \(c_t\in S_{\ell _t}\) and \(\zeta \in E_{c_t}.\) Writing \(\zeta =\lim _{\varepsilon \rightarrow 0}\zeta _\varepsilon \) for some curve \(\zeta _\varepsilon \in {\check{E}}_{c_t}\) we have
by Proposition 4.4. Since a continuity argument yields Now let \(s<t\) and \(\zeta \in E_{c_s}.\) Writing \(\zeta =\lim _{\varepsilon \rightarrow 0}\zeta _\varepsilon \) for some curve \(\zeta _\varepsilon \in {\check{E}}_{c_s}\) we have
by Proposition 4.4. Hence
By (downward) induction we may assume since \(\zeta _\varepsilon \in E_{c_s}\subset E_{c_{s+1}}.\) A continuity argument shows that concluding the induction step.
Now let \(\zeta \in \Omega _s\) so that \(\ell _{s-1}\le \ell ={\textrm{rank}}(\zeta )<\ell _s.\) Consider all tripotent chains \(c_s<c_{s+1}<\cdots <c_t\) of rank \(\ell _s<\ell _{s+1}< \cdots < \ell _t\) such that \(\zeta \in E_{c_s}\) and denote by \(\Sigma \subset {\mathcal {P}}_E\) the linear span of the associated polynomials \(N_{c_s,\ldots ,c_t}^{n_s,\ldots ,n_t}.\) Since the kernel \({\mathcal {K}}_\zeta ^s\) does not vanish by Lemma 4.4 multiplication by \({\mathcal {K}}_\zeta ^s\) is an injective map To show surjectivity, assume first that \(\zeta =c\) is a tripotent such that \(c<c_s.\) Then
where \(c_s-c< \ldots <c_t-c\) is a tripotent chain of rank \(\ell _s-\ell<\ldots <\ell _t-\ell \) in \(W=E^c\) and \({\widetilde{N}}\) denotes the conical polynomials relative to W. Since the polynomials \({\widetilde{N}}_{c_s-c,\ldots ,c_t-c}^{n_s,\ldots ,n_t}\) span \({\mathcal {P}}_W^{\lambda ^*}\) it follows that
Here we use (2.9) for the second equation. Thus it follows that In the general case let \(\zeta \in E_{c_s}\) have rank \(\ell <\ell _s.\) Then there exists \(h\in {\widehat{K}}\) such that \(\zeta =h^*c,\) where \(c\in E_{c_s}\) is a tripotent of rank \(\ell \) with \(c<c_s.\) Moreover, we may assume that \(h E_{c_j}=E_{c_j}\) for all \(s\le j\le t.\) By semi-invariance we have \(N_{c_j}\circ h^{-1}=a_j\ N_{c_j}\) for some constant \(a_j.\) This implies
On the other hand, the K-invariance of \({\mathcal {K}}^s\) implies \({\mathcal {K}}_\zeta ^s={\mathcal {K}}_{h^*c}^s={\mathcal {K}}_c^s\circ h.\) It follows that
Using the assertion follows in the general case. \(\square \)
If \(\ell _{s-1}\le \ell <\ell _s\) and \(\lambda =(\lambda _1,\ldots ,\lambda _r)=(n_1^{(\ell _1-\ell _2)},\ldots ,n_t^{(\ell _t)},0^{(r-\ell _t)}),\) we put
and \({\widehat{\lambda }}^*=(\lambda _{\ell +1}^{(\ell )},\lambda ^*).\) For any tripotent c of rank \(\ell ,\) with Peirce 0-space \(W=E^c,\) there exists a cross-section [23] \(\tau _{{\widehat{\lambda }}^*}:{\mathcal {P}}_W^{\lambda ^*}\rightarrow {\mathcal {P}}_E^{{\widehat{\lambda }}^*}\) satisfying
for all tripotent chains \(c_s< \cdots < c_t\) of rank \(\ell _s< \cdots < \ell _t\) such that \(c<c_s.\) Thus Theorem 4.5 and its proof yield the “embedding theorem”
Theorem 4.6
For each tripotent \(c\in \Omega _s\) there is an isomorphism
mapping \(\phi \in {\mathcal {P}}_W^{\lambda ^*}\) to
Beyond the tripotents, for any \(\zeta \in \Omega _s\) we define the embedding via the commuting diagram
where \(h\in {\widehat{K}}\) satisfies \(\zeta =hc.\)
6 Geometric Realization
The famous results of Kodaira [18] show that a compact Kähler manifold M, whose Kähler form \(\omega \) satisfies an integrality condition (a so-called Hodge manifold) can be embedded as an algebraic subvariety of a projective space \({{\textbf{P}}}(V).\) The underlying finite-dimensional vector space V (more precisely its linear dual space) has a geometric realization in terms of holomorphic sections of a suitable line bundle \({\mathcal {L}}\) over M. For homogeneous compact Kähler manifolds, the Borel–Weil–Bott theorem [1] gives a similar construction for the irreducible representations of compact semisimple Lie groups. In this section we show that such a “geometric” description also holds for the fibres of the localization bundle for any “partition” ideal \(J^\lambda .\) While this realization, based on the complex geometry of Peirce 2-spaces, still needs Jordan theory, the underlying idea, using flags of certain subspaces (“slices”) of E, may be susceptible to generalization.
The collection \(M_\ell \) of all Peirce 2-spaces \(U=E_c,\) for all tripotents \(c\in S_\ell \) of fixed rank \(\ell ,\) is a compact Kähler manifold called the \(\ell \)-th Peirce manifold. In the matrix case \(E={{\textbf{C}}}^{r\times s}\) and \(0\le \ell \le r\) we may identify \(M_\ell \approx {\textrm{Grass}}_\ell ({{\textbf{C}}}^r)\times {\textrm{Grass}}_\ell ({{\textbf{C}}}^s)\) as a product of Grassmann manifolds.
The structure group \({\widehat{K}}\) acts transitively on each \(M_\ell \) by holomorphic transformations \((h,U)\mapsto hU,\) and the restricted action of K is already transitive. It follows that
where \({\widehat{K}}^{E_c}:=\{\gamma \in {\widehat{K}}:\ \gamma E_c=E_c\}\) is the closed complex subgroup fixing a given Peirce 2-space \(E_c.\)
Lemma 5.1
If \(\gamma \in {\widehat{K}}^{E_c}\) then the minor \(N_c\) satisfies
where
is a holomorphic character of \({\widehat{K}}^{E_c}.\)
Proof
Since \(\gamma \in {\widehat{K}}^{E_c}\) preserves \(E_c\) the Peirce 2-projection \(P_c\) satisfies \(\gamma P_c=P_c\gamma P_c.\) Therefore \(P_c\gamma ^*=P_c\gamma ^*P_c.\) Since \(P_c\gamma P_c\in {\widehat{K}}_{E_c}\) the semi-invariance of \(\Delta _c\) implies
Here we use
\(\square \)
As a consequence we obtain a holomorphic homogeneous line bundle
over \(M_\ell ,\) with projection \([h,\xi ]\mapsto hE_c.\) By definition, the holomorphic sections \(\sigma :M_\ell \rightarrow {\mathcal {L}}_\ell \) have the form
where the “homogeneous lift” \({\widetilde{\sigma }}:{\widehat{K}}\rightarrow {{\textbf{C}}}\) is a holomorphic map satisfying
for all \(h\in {\widehat{K}},\ \gamma \in {\widehat{K}}^{E_c}.\)
Let \(\lambda \in {{\textbf{N}}}_+^r\) be a partition written in the form (4.4). The Peirce flag manifold \(M_{\ell _1,\ldots ,\ell _t}\) consists of all chains of Peirce 2-spaces \(U_1\subset U_2 \subset \cdots \subset U_t\) such that \({\textrm{rank}}(U_s)=\ell _s\) for all \(s\le t.\) The group \({\widehat{K}}\) acts on \(M_{\ell _1,\ldots ,\ell _t}\) by holomorphic transformations \((h,(U_1,\ldots ,U_t))\mapsto (hU_1,\ldots ,hU_t)\) and the restriction to K is already transitive. Thus
where \({\widehat{K}}^{E_{c_1},\ldots ,E_{c_t}}:=\{\gamma \in {\widehat{K}}:\ \gamma E_{c_s}=E_{c_s}\ \forall \ 1\le s\le t\}\) is the closed complex subgroup fixing the Peirce 2-flag \(E_{c_1}\subset \cdots \subset E_{c_t}\) for a chain of tripotents \(c_1< \cdots <c_t\) of rank \(\ell _1< \cdots <\ell _t.\) We may choose \(c_s:=e_{[\ell _s]}.\) As a consequence of Lemma 5.1 the conical function
satisfies
for all \(\gamma \in {\widehat{K}}^{E_{c_1},\ldots ,E_{c_t}},\) where
is a holomorphic character of \({\widehat{K}}^{E_{c_1},\ldots ,E_{c_t}}.\) As a consequence we obtain a holomorphic homogeneous line bundle
over \(M_{\ell _1,\ldots ,\ell _t},\) with projection \([h,\xi ]\mapsto (hE_{c_1},\ldots ,hE_{c_t}).\) By definition, the holomorphic sections \(\sigma :M_{\ell _1,\ldots ,\ell _t}\rightarrow {\mathcal {L}}^\lambda \) have the form
where the homogeneous lift \({\widetilde{\sigma }}:{\widehat{K}}\rightarrow {{\textbf{C}}}\) is a holomorphic map satisfying
for all \(h\in {\widehat{K}}\) and \(\gamma \in {\widehat{K}}^{E_{c_1},\ldots ,E_{c_t}}.\) For \(t=1\) we obtain the powers \({\mathcal {L}}_\ell ^n\) of the line bundle (5.1).
We first describe the maximal fibre at \(\zeta =0.\)
Theorem 5.2
Any \(p\in {\mathcal {P}}_E^\lambda \) defines a holomorphic section \(\sigma ^p:M_{\ell _1,\ldots ,\ell _t}\rightarrow {\mathcal {L}}^\lambda ,\) whose homogeneous lift \({\widetilde{\sigma }}^p:{\widehat{K}}\rightarrow {{\textbf{C}}}\) is given by the Fischer-Fock inner product
This yields a \({\widehat{K}}\)-equivariant isomorphism
Proof
To check the homogeneity condition, let \(\gamma \in {\widehat{K}}^{E_{c_1},\ldots ,E_{c_t}}.\) Then (5.2) implies
Since \({\mathcal {P}}_E^\lambda \) is the linear span of polynomials \(N^\lambda \circ h^*\) the map \(\Theta \) is injective. For any \(h,h'\in {\widehat{K}}\) we have
proving invariance
under \(h\in {\widehat{K}}.\) Realizing \(M_{\ell _1,\ldots ,\ell _t}={\widehat{K}}/{\widehat{K}}^{E_{c_1},\ldots ,E_{c_t}}\) as a Lie theoretic flag manifold [3] the Borel–Weil–Bott theorem shows that \(\Gamma (M_{\ell _1,\ldots ,\ell _t},{\mathcal {L}}^\lambda )\) is \({\widehat{K}}\)-irreducible. Thus the map \(\Theta \) is surjective and hence an isomorphism. \(\square \)
As an example, consider the partitions
with \(n>0\) repeated \(\ell \) times. Here \(t=1\) and one obtains an isomorphism (cf. [3])
where \({\mathcal {L}}_\ell ^n\) is the n-th tensor power of the line bundle \({\mathcal {L}}_\ell \) defined by (5.1). If \(\ell =r\) and E is unital with unit element e, then \(M_r=\{E\}\) is a singleton and \({\mathcal {P}}_E^{n^{(r)}}\) is 1-dimensional, spanned by \(N_e^n.\) If \(\ell \) is arbitrary and \(n=1\) we have the “fundamental” partitions and
In the simplest case \(\ell =1\) we obtain a kind of projective space \(M_1\) consisting of all Peirce 2-spaces of rank 1, or equivalently, all lines spanned by tripotents of rank 1. The associated “tautological” line bundle
has no holomorphic sections. On the other hand each \(f\in E^*\) yields a holomorphic section \(\sigma ^f:M\rightarrow {\mathcal {T}}^*\) of the dual line bundle
via
The calculation
shows that \(h\cdot \sigma ^f=\sigma ^{f\circ h^{-1}}.\) The homogeneous lift \({\widetilde{\sigma }}^f:{\widehat{K}}\rightarrow {{\textbf{C}}}\) is given by
If \(\gamma \in {\widehat{K}}^{E_{e_1}}\) then \(\gamma e_1=\chi _1(\gamma )e_1\) for the holomorphic character \(\chi _1:{\widehat{K}}^{E_{e_1}}\rightarrow {{\textbf{C}}}^\times \) given by \(\chi _1(\gamma )=(\gamma e_1|e_1)=N_{e_1}(\gamma e_1).\) This is covered by the general formula (5.4):
Lemma 5.3
For each \(f\in E^*={\mathcal {P}}_E^{1,0,\ldots ,0}\) we have
Proof
Putting \(fz:=(z|v)\) the relation
implies
\(\square \)
In order to describe the localization at non-zero points \(\zeta \) of rank \(\ell ,\) where \(\ell _{s-1}\le \ell <\ell _s,\) we pass to certain submanifolds of the Peirce flag manifold. Consider the “incidence space”
and the “polar space”
For a tripotent \(\zeta =c\) this means that the flag is contained in \(W=E^c.\) These compact submanifolds are closely related: If \((U_s, \ldots , U_t)\in M_{\ell _s,\cdots ,\ell _t}^\zeta \) then the Peirce 0-spaces \(W_k:=U_k^c\) form a flag \((W_s,\cdots ,W_t)\in M_{\ell _s-\ell ,\cdots ,\ell _t-\ell }^{\zeta *}.\) Conversely, if \((W_s,\cdots ,W_t)\in M_{\ell _s-\ell ,\cdots ,\ell _t-\ell }^{\zeta *}\) and we put \(W_k:=W_{c_k}^2\) for a tripotent chain \(c_s<\cdots <c_t\) of rank \(\ell _s-\ell<\cdots <\ell _t-\ell \) in W, then \(c+c_s<\cdots <c+c_t\) is a tripotent chain of rank \(\ell _s<\cdots <\ell _t\) in E and \(U_k:=E_{c+c_k}^2\) defines a flag \((U_s, \ldots , U_t)\in M_{\ell _s,\cdots ,\ell _t}^\zeta .\) Moreover, if \(c_k'\sim c_k\) are Peirce equivalent in W, then \(c+c_k'\sim c+c_k\) are Peirce equivalent in E.
The closed complex subgroup
acts transitively on the incidence space \(M_{\ell _s,\ldots ,\ell _t}^\zeta \) and, similarly, the closed complex subgroup
acts transitively on the polar space \(M_{\ell _s-\ell ,\ldots ,\ell _t-\ell }^{\zeta *}\) since for all \(h\in {\widehat{K}}.\) Therefore
where
and \(W_{c_s}\subset \ldots \subset W_{c_t}\subset W\) is the Peirce 2-flag for a given chain of tripotents \(c_s<\ldots <c_t\) in W of type \(\ell _s-\ell<\ldots <\ell _t-\ell .\) We may choose \(c_j:=e_{\ell +1}\;+e_{\ell _j}.\) The restriction homomorphism
defines a biholomorphic map
where
and \(M_{\ell _s-\ell ,\ldots ,\ell _t-\ell }^W\) denotes the Peirce flag manifold relative to W. Now consider the truncated partition \(\lambda ^*\) of length \(r-\ell .\) Then the conical function relative to W is
and there is a holomorphic character
on \({\widehat{K}}_W^{W_{c_s},\ldots ,W_{c_t}}\) such that
for all \(\gamma \in {\widehat{K}}_W^{W_{c_s} ,\ldots , W_{c_t}}.\) Let \({\mathcal {L}}_W^{\lambda ^*}\) denote the associated holomorphic line bundle on \(M_{\ell _s-\ell ,\ldots ,\ell _t-\ell }^W\) defined as in (5.3). A holomorphic section \(\sigma :M_{\ell _s-\ell ,\ldots ,\ell _t-\ell }^W\rightarrow {\mathcal {L}}_W^{\lambda ^*}\) is characterized by its homogeneous lift \({\widetilde{\sigma }}:{\widehat{K}}_W\rightarrow {{\textbf{C}}}\) satisfying
for all \(\kappa \in {\widehat{K}}_W\) and \(\gamma \in {\widehat{K}}_W^{W_{c_s} ,\ldots , W_{c_t}}.\) Via the isomorphism (5.5) we obtain a holomorphic pull-back line bundle \({\mathcal {L}}_\zeta ^{\lambda ^*}=\varrho ^*{\mathcal {L}}_W^{\lambda ^*}\) such that holomorphic sections \(\sigma :M_{\ell _s-\ell ,\ldots ,\ell _t-\ell }^{\zeta *}\rightarrow {\mathcal {L}}_\zeta ^{\lambda ^*}\) are characterized by the homogeneous lift \({\widetilde{\sigma }}:{\widehat{K}}^{\zeta *}\rightarrow {{\textbf{C}}}\) satisfying
for all \(\kappa \in {\widehat{K}}^{\zeta *}\) and \(\gamma \in ({\widehat{K}}^{\zeta *})^{W_{c_s} ,\ldots , W_{c_t}}.\) All of this holds in particular when \(\zeta =c\) is a tripotent of rank \(\ell .\)
Theorem 5.4
Write \(\lambda \in {{\textbf{N}}}_+^r\) in the form (4.4). Let c be a tripotent of rank \(\ell \) with Peirce 0-space W, such that \(\ell _{s-1}\le \ell <\ell _s.\) Consider the truncated partition
Then there is an isomorphism
which is defined as follows: Any \(\phi \in {\mathcal {P}}_W^{\lambda ^*}\) defines a holomorphic section \(\sigma ^\phi :M_{\ell _s,\ldots ,\ell _t}^{c*}\rightarrow {\mathcal {L}}_c^{\lambda ^*},\) whose homogeneous lift \({\widetilde{\sigma }}^\phi :{\widehat{K}}^{c*}\rightarrow {{\textbf{C}}}\) is given by the Fischer-Fock inner product
for all \(\kappa \in {\widehat{K}}^{c*}.\)
Proof
Applying Theorem 5.2 to W we have
via the map \(\phi \mapsto \sigma _W^\phi \) with homogeneous lift
for all \(\kappa \in {\widehat{K}}_W.\) Passing to \(\Gamma (M_{\ell _s,\ldots ,\ell _t}^{c*},{\mathcal {L}}_c^{\lambda ^*})\) via the isomorphism (5.5), the assertion follows. \(\square \)
We finally describe a similar isomorphism for non-tripotent points \(\zeta \in {\check{E}}_\ell .\) Write \(\zeta =hc\) for some \(h\in {\widehat{K}}.\) Then \(h^{-*}{\widehat{K}}^{c*}h^*={\widehat{K}}^{\zeta *}\) and
for the Peirce 2-flag \(h^{-*}W_{c_s}\subset \ldots \subset h^{-*}W_{c_t}\subset h^{-*}W\subset E.\) Thus we obtain a commuting diagram
Now the isomorphism
is defined via the commuting diagram
This demonstrates that the bundle depends in an anti-holomorphic way on \(\zeta \).
Data Availability Statement
The author confirms that the data supporting the findings of this study are available within the article.
References
Akhiezer, D.: Lie Group Actions in Complex Analysis. Vieweg (1995)
Arazy, J.: A survey of invariant Hilbert spaces of analytic functions on bounded symmetric domains. Contemp. Math. 185, 7–65 (1995)
Arazy, J., Upmeier, H.: Jordan Grassmann manifolds and intertwining operators for weighted Bergman spaces. In: Doria, A. (ed.) Proceedings Cluj-Napoca, pp. 25–53. Cluj University Press (2008)
Biswas, S., Misra, G., Putinar, M.: Unitary invariants for Hilbert modules of finite rank. J. Reine Angew. Math. 662, 165–204 (2012)
Chu, C.-H.: Bounded Symmetric Domains in Banach Spaces. World Scientific (2021)
Cowen, C., Douglas, R.: Complex geometry and operator theory. Acta Math. 141, 187–261 (1978)
deConcini, C., Eisenbud, D., Procesi, C.: Young diagrams and determinantal varieties. Invent. Math. 56, 129–165 (1980)
Duan, Y., Guo, K.: Dimension formula for localization of Hilbert modules. J. Oper. Theory 62, 439–452 (2009)
Douglas, R., Misra, G., Varughese, C.: On quotient modules - the case of arbitrary multiplicity. J. Funct. Anal. 174, 364–398 (2000)
Englis, M., Upmeier, H.: Reproducing kernel functions and asymptotic expansions on Jordan-Kepler varieties. Adv. Math. 347, 780–826 (2019)
Faraut, J., Korányi, A.: Function spaces and reproducing kernels on bounded symmetric domains. J. Funct. Anal. 88, 64–89 (1990)
Faraut, J., Korányi, A.: Analysis on Symmetric Cones. Clarendo Press, Oxford (1994)
Fischer, G.: Lineare Faseräume und kohärente Modulgarben über komplexen Räumen. Arch. Math. 18, 609–617 (1967)
Guo, K.: Algebraic reduction for Hardy submodules over polydisk algebras. J. Oper. Theory 41, 127–138 (1999)
Korányi, A., Misra, G.: A classification of homogeneous operators in the Cowen–Douglas class. Adv. Math. 226, 5338–5360 (2011)
Loos, O.: Jordan Pairs. Springer Lect. Notes Math. 460 (1975)
Loos, O.: Bounded Symmetric Domains and Jordan Pairs. Univ. of California, Irvine (1977)
Morrow, J., Kodairam, K.: Complex Manifolds. AMS Chelsea Publishing (2006)
Neher, E.: An expansion formula for the norm function of a Jordan algebra. Arch. Math. 69, 105–111 (1997)
Schmid, W.: Die Randwerte holomorpher Funktionen auf hermitesch symmetrischen Räumen. Invent. Math. 9, 61–80 (1969)
Upmeier, H.: Jordan algebras and harmonic analysis on symmetric spaces. Am. J. Math. 108, 1–25 (1986)
Upmeier, H.: Toeplitz operators on bounded symmetric domains. Trans. Am. Math. Soc. 280, 221–237 (1983)
Upmeier, H.: Hilbert modules and singular vector bundles on bounded symmetric domains. J. Reine Angew. Math. (Crelle) 799, 155–187 (2023)
Funding
Open Access funding enabled and organized by Projekt DEAL.
Author information
Authors and Affiliations
Corresponding author
Additional information
Communicated by Mihai Putinar.
Dedicated to the Memory of Jörg Eschmeier.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
This article is part of the topical collection “Multivariable Operator Theory. The Jörg Eschmeier Memorial” edited by Raul Curto, Michael Hartz, Mihai Putinar and Ernst Albrecht.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Upmeier, H. Stratified Hilbert Modules on Bounded Symmetric Domains. Complex Anal. Oper. Theory 17, 74 (2023). https://doi.org/10.1007/s11785-023-01377-1
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s11785-023-01377-1