---
title: 'Bergman Dual: Spaces, Metrics and Kernels'
url: https://www.emergentmind.com/topics/bergman-dual
type: topic
---

# Bergman Dual: Spaces, Metrics and Kernels

“Bergman dual” is not a single universally fixed object. In current usage, it denotes several dual constructions attached to Bergman spaces, Bergman kernels, and Bergman metrics. In Banach-space theory it refers to the continuous dual, predual, or annihilator of a Bergman space under an integral pairing; in transform theory it refers to concrete realizations of that dual by holomorphic functions on a different domain or on \(\mathbb C^n\) with an exponential weight; in Kähler geometry it denotes the metric built from the dualized kernel \(K_D(z,-\bar z)\); and in bundle-theoretic settings it refers to Bergman kernels defined on dual line bundles or from inner products on \(O(E)^*\) [1901.07780], [2506.02913], [2510.06405], [2510.22169], [2109.08593].

## 1. Terminological scope and basic models

The expression occurs in several technically distinct settings.

| Context | Construction | Resulting object |
|---|---|---|
| Bergman-space duality | Integral pairing on \(A^p(D)\) or weighted variants | Continuous dual, predual, or annihilator |
| Transform realization | Fantappié or Laplace transform of analytic functionals | Holomorphic space on \(D^*\) or weighted space on \(\mathbb C^n\) |
| Geometric dualization | \(K_D^*(z,\bar z)=K_D(z,-\bar z)\) | Bergman dual \((D^*,g_D^*)\) |
| Bundle/CR setting | Bergman kernel of disk bundle in \(L^*\) | “Dual” Bergman kernel form on \(D\subset L^*\) |
| Dual section space | Inner products on \(O(E)^*\) | Bergman kernel \(K_{\llangle,\rrangle}\) and Bergman section |

For a bounded domain \(D\subset\mathbb C^n\), the classical Bergman framework begins with holomorphic \(L^p\)-spaces and the Bergman projection. In the most standard reflexive regime, one expects duality through the pairing
\[
\langle f,g\rangle=\int_D f(z)\,\overline{g(z)}\,dV(z).
\]
However, the literature shows that this expectation is highly domain-dependent, changes at nonreflexive endpoints, and generalizes in different directions to Orlicz, meromorphic, quaternionic, and geometric settings [2405.11113], [1804.02746], [1501.03416], [2309.00552], [2406.07744].

The unifying feature is that duality is controlled by reproducing structures: either the Bergman kernel itself, an adapted projection operator, or a dualized kernel/potential. This suggests that “Bergman dual” is best understood as a family of kernel-mediated dual objects rather than a single invariant.

## 2. Classical Banach-space duality and its failure on irregular domains

For a bounded domain \(D\subset\mathbb C^n\), Bhat formulates the \(p\)-Bergman space
\[
A^p(D)=\{f\ \text{holomorphic on}\ D:\|f\|_{A^p(D)}=(\int_D |f(z)|^p\,dV(z))^{1/p}<\infty\},
\]
and proves that if the Bergman projection \(P\) extends to a bounded operator on both \(L^p(D)\) and \(L^q(D)\), then for \(1<p<\infty\) with \(1/p+1/q=1\) there is a canonical isometric isomorphism
\[
(A^p(D))^* \cong A^q(D),
\]
with each \(g\in A^q(D)\) inducing
\[
L_g(f)=\int_D f(z)\,\overline{g(z)}\,dV(z).
\]
The same work states that the properties “\(P\) extends boundedly on \(L^p\) and on \(L^q\),” “\((A^p)^*\cong A^q\) via the pairing,” and “\(P\) is of weak type \((p,p)\) and \((q,q)\)” are equivalent, and it further identifies the interpolation identity
\[
(A^{p_0},A^{p_1})_\theta=A^{p_\theta}(D),\qquad \frac1{p_\theta}=\frac{1-\theta}{p_0}+\frac{\theta}{p_1},
\]
as part of the same analytic package [2405.11113].

This picture fails on general pseudoconvex domains. On the classical Hartogs triangle
\[
\Omega=\{(z_1,z_2)\in\mathbb C^2:\ |z_1|<|z_2|<1\},
\]
Edholm and McNeal show that for \(p=\tfrac53\) and \(q=\tfrac52\), the monomial \(h(z_1,z_2)=z_2^{-2}\) lies in \(A^{5/3}(\Omega)\) but not in \(A^2(\Omega)\). The coefficient functional \(a_{(0,-2)}\) is bounded on \(A^{5/3}(\Omega)\) but cannot be represented by \(\int_\Omega f\,\overline{\phi}\) with \(\phi\in A^{5/2}(\Omega)\), and \(A^2(\Omega)\) is not dense in \(A^{5/3}(\Omega)\). Thus the naive pattern
\[
(A^p(\Omega))'\overset{?}{\simeq}A^q(\Omega)
\]
breaks down even in low dimension [1804.02746].

The same paper isolates a positive mechanism. If an integral operator \(\mathcal P:L^2(\Omega)\to A^2(\Omega)\) satisfies the three mapping properties (H1), (H2), and (H3), then the pairing map \(\Phi_p:A^q\to(A^p)'\) is onto. On generalized Hartogs triangles \(\mathcal H_{m/n}\), a “sub-Bergman” projection \(\widetilde{\mathcal B}^p\) satisfies the required properties for each \(p\ge2\), recovering a concrete dual description in terms of \(L^p\)-allowable Laurent monomials [1804.02746].

A common misconception is that Riesz-type Bergman duality is automatic for bounded domains. The available results show the opposite: duality depends on projection regularity, and projection regularity depends sharply on domain geometry.

## 3. Nonreflexive endpoints and generalized dual objects

At the nonreflexive endpoint \(p=1\), the dual need not be another Bergman space. For the upper half-plane \(\mathbb U=\{w\in\mathbb C:\Im w>0\}\) with weighted measure
\[
d\mu_\alpha(w)=(\Im w)^\alpha\,dA(w),\qquad \alpha>-1,
\]
Galanopoulos and Girela identify the predual of the nonreflexive weighted Bergman space \(L_a^1(\mathbb U,\mu_\alpha)\) as the little Bloch space vanishing at \(i\),
\[
\mathcal B_0(\mathbb U,i)=\Bigl\{g\in\mathcal H(\mathbb U):\sup_{w\in\mathbb U}\Im w\,|g'(w)|<\infty,\ \lim_{\Im w\to0}\Im w\,|g'(w)|=0,\ g(i)=0\Bigr\},
\]
with norm \(\|g\|_{\mathcal B_0}=\sup_{w\in\mathbb U}\Im w\,|g'(w)|\). They prove
\[
(\mathcal B_0(\mathbb U,i))^*\cong L_a^1(\mathbb U,\mu_\alpha),
\qquad
(L_a^1(\mathbb U,\mu_\alpha))^*\cong \mathcal B(\mathbb U,i),
\]
under the pairing
\[
\langle g,f\rangle=\int_{\mathbb U} g(w)\,\overline{f(w)}\,d\mu_\alpha(w).
\]
On the identified predual, the adjoint scaling and translation groups act by
\[
S_tg(w)=g(e^t w),\qquad S_tg(w)=g(w-t),
\]
and form groups of surjective isometries; for the scaling generator \(\Gamma\), the point spectrum is empty, the full spectrum is \(i\mathbb R\), and the resolvent norm equals \(1/|\Re\lambda|\) [1901.07780].

Other generalized Bergman duals preserve the same principle but change the target category. For large Bergman–Orlicz spaces on the unit ball, Sehba and Tchoundja show two distinct duality regimes: when \(\Phi\) is convex and satisfies the \(\Delta_2\)-condition together with its complementary function \(\Psi\), one has
\[
(A_\alpha^\Phi(\mathbb B^n))^*\cong A_\alpha^\Psi(\mathbb B^n),
\]
whereas for concave \(\Phi\in L^p\) with \(\rho(t)=[\Phi^{-1}(1/t)]^{-1}\),
\[
(A_\alpha^\Phi(\mathbb B^n))^*\cong \Lambda_{\alpha,\rho}(\mathbb B^n)
\]
under the integral pairing \(\langle f,g\rangle_\alpha=\int_{\mathbb B^n}f(z)\overline{g(z)}\,d\nu_\alpha(z)\) [1501.03416].

For the meromorphic Bergman spaces on the pointed disc \(D^*=\{z\in\mathbb C:0<|z|<1\}\), the dual remains of the same meromorphic type: for \(1<p<\infty\) and Hölder conjugate \(q\),
\[
(A^p_{\alpha,\beta}(D^*))'\cong A^q_{\alpha,\beta}(D^*)
\]
isometrically under
\[
\langle f,g\rangle_{\alpha,\beta}=\int_{D^*}f(z)\,g(z)\,d\lambda_{\alpha,\beta}(z),
\]
despite the admissibility of controlled poles at \(0\) [2309.00552].

In the biquaternionic Vekua setting on bounded Liapunov domains \(G\subset\mathbb R^3\), the Bergman space is \(B_p(G;\mathbb H_\mathbb C)=\ker(D-Q_A)\), and its annihilator in \(L_{p'}\) is identified by
\[
B_p(G;\mathbb H_\mathbb C)^\circ
=
\overline{\{(D-Q_A^+)u:\ u\in W^{1,p'}(G;\mathbb H_\mathbb C)\}},
\]
with the closure removed under the smallness condition \(\sum_{j=1}^4\|a_j\|_\infty<2\,\mathrm{diam}(G)\) [2406.07744].

## 4. Transform realizations via Fantappié and Laplace theory

A different meaning of Bergman dual arises when the continuous dual of \(A^2(D)\) is realized as a holomorphic function space on another domain. Chatterjee studies this for bounded domains \(D\subset\mathbb C^n\) by embedding \(A^2(D)\) into the analytic functionals \(\mathcal O'(\overline D)\) through
\[
\theta(f):\phi\mapsto \int_D \overline{f(z)}\,\phi(z)\,dV(z),\qquad \phi\in\mathcal O(\overline D).
\]
The Fantappié transform is then defined by
\[
\mathcal F_k\mu(w)=\mu\Bigl(\frac1{(1-\langle\cdot,w\rangle)^k}\Bigr),\qquad w\in D^*,
\]
where
\[
D^*=\{w\in\mathbb C^n:\ \langle \zeta,w\rangle\ne1\ \forall\,\zeta\in\overline D\}.
\]
For \(f\in A^2(D)\),
\[
\mathcal F_k f(w)=\frac{n!}{\pi^n}\int_D \overline{f(z)}\,(1-\langle z,w\rangle)^{-k}\,dV(z).
\]
If \(D\subset\mathbb C^n\) is bounded, \(C^2\)-smooth, strongly lineally-convex, contains \(0\), and \(D^*\) is also strongly convex, then
\[
\mathcal F_{n+1}:A^p(D)\longrightarrow A^p(D^*)
\]
is a topological isomorphism for every \(p\in(1,\infty)\); in particular,
\[
(A^2(D))'\cong A^2(D^*)
\]
via the Fantappié transform [2506.02913].

The same paper gives a Laplace realization. On \(\mathcal O'(\overline D)\),
\[
\mathcal L\mu(w)=\mu(e^{\langle\cdot,w\rangle}),\qquad w\in\mathbb C^n,
\]
and on \(A^2(D)\),
\[
\mathcal L(f)(w)=\int_D \overline{f(z)}\,e^{\langle z,w\rangle}\,dV(z).
\]
With support function
\[
H_D(w)=\sup_{\zeta\in D}\Re\langle \zeta,w\rangle
\]
and weight
\[
\omega_D(w)=e^{-2H_D(w)}\;\|w\|^{\,n+\tfrac12}\;\det\bigl(dd^cH_D\bigr)^n,
\]
the map
\[
\mathcal L:A^2(D)\longrightarrow A^2(\mathbb C^n,\omega_D)
\]
is a topological isomorphism, and the inverse is expressed through the \(n\)th-degree Borel transform
\[
\mathcal B_nF(w)=\int_0^\infty F(tw)\,e^{-t}\,t^n\,dt.
\]
This extends the planar results of Napalkov Jr–Yulumukhametov from simply connected or convex domains to higher-dimensional strongly convex domains [2506.02913].

The higher-dimensional theory is not a routine generalization. For the \(\ell^1\)-ball
\[
D=\{(z_1,z_2):|z_1|+|z_2|<1\}\subset\mathbb C^2,
\]
\(\mathcal F_3\) and \(\mathcal L\) remain bounded and injective, but their ranges are proper subspaces of the corresponding target Bergman spaces. Mere convexity is therefore insufficient once \(n\ge2\) [2506.02913].

## 5. Geometric Bergman duals from the kernel \(K_D(z,-\bar z)\)

In Kähler geometry, “Bergman dual” denotes a dual metric constructed directly from the Bergman kernel. For a bounded domain \(D\subset\mathbb C^n\) containing the origin, with Bergman kernel \(K_D(z,\bar w)\), the diagonal kernel \(K_D(z,\bar z)\) yields the Bergman metric \(g_D\) through
\[
\omega_D=\frac{i}{2}\partial\bar\partial\log K_D(z,\bar z).
\]
The dualized kernel is defined by
\[
K_D^*(z,\bar z)=K_D(z,-\bar z).
\]
Since \(K_D(z,-\bar z)\) is real-analytic near the origin, there is a maximal connected open neighborhood \(D^*\subset\mathbb C^n\) of \(0\) on which \(K_D^*(z,\bar z)>0\). The Bergman dual of \((D,g_D)\) is then \((D^*,g_D^*)\), with Kähler form
\[
\omega_D^*= -\frac{i}{2}\partial\bar\partial\log K_D^*(z,\bar z),
\]
and local coefficients
\[
g_{D\,j\bar k}^*(z)= -\,\partial_j\bar\partial_k\log K_D(z,-\bar z).
\]
When \(D\) is a bounded symmetric domain, Harish–Chandra theory implies that \(K_D(z,-\bar z)\) never vanishes on \(\mathbb C^n\), so \(D^*=\mathbb C^n\); in that case \((D^*,g_D^*)\) is the compact dual restricted to its big cell [2510.06405].

For Cartan–Hartogs domains
\[
M_{\Omega,\mu}=\{(z,w)\in\Omega\times\mathbb C:\ |w|^2<N_\Omega(z,\bar z)^\mu\},
\]
the Bergman dual enters a rigidity theorem. With \(M=M_{\Omega,\mu}\) and Bergman metric \(g_M\), the following are equivalent: \(M\) is biholomorphic to the unit ball \(B^{n+1}\); \(g_M\) is Kähler–Einstein; \(g_M\) is a Kähler–Ricci soliton with \(\mu\in\mathbb Q\); and, up to a positive constant \(\alpha\), the Bergman dual metric \(g_M^*\) is finitely projectively induced. In the ball case, the dual Bergman metric on \(\mathbb C^{n+1}\) is \((n+2)/\pi\) times the Fubini–Study metric restricted to the affine chart, and the rescaling constant is \(\alpha=n+2\) [2510.06405].

This geometric duality comes with limitations. Completeness of \((D,g_D)\) does not imply completeness of \((D^*,g_D^*)\). The paper also poses open problems: if \(M_{\Omega,\mu}^*=\mathbb C^{n+1}\), must the original \(M_{\Omega,\mu}\) be the ball? In a rank \(1\) example with \(\mu>1\), \(K_M^*(z,w)\) vanishes along a real hypersurface in \(\mathbb C^2\), so \(M^*\neq\mathbb C^2\), and the induced metric is incomplete with unbounded negative curvature near the boundary [2510.06405].

## 6. Dual Bergman kernels on bundles and spaces of sections

A further line of development shifts the dualization from domains in \(\mathbb C^n\) to line bundles and their disk bundles. Let \((M,g;L,h)\) be a polarized Kähler manifold, and let \(D=\{v\in L^*:|v|_{h^*}^2<1\}\) be the disk bundle of the dual line bundle. Ebenfelt, Xiao, and Xu define the Bergman space
\[
A^2_{n+1,0}(D)=\{f\in L^2_{n+1,0}(D): f\ \text{is holomorphic}\}
\]
and its Bergman kernel form
\[
K_{\mathrm{dual}}(z,w)=\sum_j \phi_j(z)\wedge \overline{\phi_j(w)}
\]
for an orthonormal basis \(\{\phi_j\}\). If \(p(v)=1-|v|_{h^*}^2\), then \(K_{\mathrm{dual}}\) admits a Fefferman-type boundary expansion, and the log-term is controlled by the Tian–Yau–Zelditch–Catlin coefficients \(a_j\) of the twisted Bergman kernels
\[
B_k(x)\sim \pi^{-n}\bigl[k^n+a_1(x)k^{n-1}+a_2(x)k^{n-2}+\cdots\bigr].
\]
They prove that on an open set \(U\subset M\), the vanishing of all coefficients \(a_{n+2+m}\) for \(m\ge0\) is equivalent to Bergman-logarithmic-flatness of the corresponding piece of the circle bundle \(S=\partial D\). As an application, if \((M,g;L,h)\) is compact and \(g\) is locally homogeneous, then the circle bundle is Bergman logarithmically flat [2510.22169].

Lempert studies a related dual construction for a very ample line bundle \(E\to X\), but now from the finite-dimensional space \(O(E)^*\). For an arbitrary real-bilinear, Hermitian-symmetric inner product \(\llangle\cdot,\cdot\rrangle\) on \(O(E)^*\), there is a unique holomorphic section
\[
K_{\llangle,\rrangle}\in H^0(X\times X,E\boxtimes\overline E)
\]
satisfying the reproducing identity
\[
\llangle \mathrm{ev}_x,\mathrm{ev}_y\rrangle
=
\langle \mathrm{ev}_y,K(x,y)\rangle.
\]
Its diagonal restriction
\[
k(x)=K(x,x)\in (E\otimes\overline E)_x
\]
is the Bergman section. This construction extends the Fubini–Study map by setting
\[
FS_{\mathrm{dual}}(\llangle\ ,\ \rrangle)=\log k(x).
\]
The classical \(FS\) is an injective immersion, but its image is in general not closed in \(\mathcal H_\omega\). To obtain a closed range, one enlarges the domain to certain semidefinite forms on \(O(E)^*\): the map \(\Phi(\llangle\cdot,\cdot\rrangle)=\log k\) on the set \(\mathcal A_E\) of nonnegative forms with rather ample underlying subspace is injective and proper, and the closure of \(FS(H_E)\) is exactly \(\Phi(\mathcal A_E)\) [2109.08593].

These bundle-theoretic constructions show that Bergman duality is not restricted to Banach duals of holomorphic \(L^p\)-spaces. It also appears as a dualization of kernels, potentials, and section spaces, with boundary singularities, projective inducedness, and closure phenomena determined by curvature and ampleness data rather than by \(L^p\)-mapping alone.

Source: https://www.emergentmind.com/topics/bergman-dual