--- title: Bergman Kernel Function in Complex Geometry url: https://www.emergentmind.com/topics/bergman-kernel-function type: topic --- # Bergman Kernel Function in Complex Geometry The Bergman kernel function is the diagonal restriction of the Bergman reproducing kernel, and in complex geometry it is the canonical density attached to an \(L^2\)-space of holomorphic objects. In a line-bundle formulation, if \((M,\omega)\) is a Kähler manifold, \(L\to M\) is a holomorphic line bundle with Hermitian metric \(h\), and \(H_k\) is the \(L^2\)-space of holomorphic sections of \(L^k\), then the Bergman kernel \(K_k(x,y)\) is the reproducing kernel of \(H_k\), while the Bergman kernel function is the diagonal density \(B_k(x)=|K_k(x,x)|_{h^k}\), denoted \(\rho_k\) in some of the recent literature [2511.19003]. On compact polarized manifolds this agrees with the basis-free quantity \(B_k(x)=\sum_{i=1}^{N_k}|s_i(x)|_{h^k}^2\) for any \(L^2\)-orthonormal basis \(\{s_i\}\subset H^0(X,L^k)\), and it can equally be characterized as the supremum of the pointwise norm over unit \(L^2\)-norm holomorphic sections [2304.01943]. In weighted planar settings one encounters the normalized density \(\rho_m(z)=\frac{1}{m}K_m(z,z)e^{-2mQ(z)}\) [1805.00854]. Across these formulations, the Bergman kernel function serves simultaneously as a reproducing-theoretic object, a density of states, and a geometric invariant encoding curvature, global topology, and boundary or singular behavior. ## 1. Reproducing structure and basic formulations For a Kähler manifold \((M,\omega)\) of complex dimension \(n\), with \(L\to M\) a holomorphic line bundle equipped with a Hermitian metric \(h\) whose Chern curvature equals \(-i\omega\), the relevant Hilbert space is \[ H_k=\Big\{s\in H^0(M,L^k):\ \int_M |s|_{h^k}^2\,\frac{\omega^n}{n!}<\infty\Big\}. \] The evaluation map \(s\mapsto s(y)\) is bounded on \(H_k\), so there is a unique \(L^2\)-integrable holomorphic section \(K_y\) with \(\langle s,K_y\rangle=s(y)\). The off-diagonal kernel is \(K_k(x,y):=K_y(x)\), and if \(\{s_j\}\) is an orthonormal basis then \[ K_k(x,y)=\sum_j s_j(x)\otimes \overline{s_j(y)}. \] Its diagonal trace is the Bergman kernel function, \[ \rho_k(x)=|K_k(x,x)|_{h^k}, \] which is also called the diagonal density [2511.19003]. On a compact complex manifold or smooth projective variety \(X\) with ample line bundle \(L\to X\), the same object is often written \[ B_k(x)=\sum_{i=1}^{N_k}|s_i(x)|_{h^k}^2, \] for an \(L^2\)-orthonormal basis of \(H^0(X,L^k)\). Equivalently, \[ B_k(x)=\sup\{ |s(x)|_{h^k}^2 : s\in H^0(X,L^k),\ \|s\|=1\}. \] The two-point kernel is the integral kernel of the orthogonal projection onto holomorphic sections, and its diagonal enters the Fubini–Study potential \[ \varphi_{FS,k}(x)=\log \sum_{i=1}^{N_k}|s_i(x)|_{h^k}^2, \] hence the associated Fubini–Study form \(\omega_{FS,k}=(i/2\pi)\partial\bar\partial\varphi_{FS,k}\) [2304.01943]. The same reproducing paradigm extends beyond line bundles on compact manifolds. For weighted entire functions on \(\mathbb C\), the weighted Bergman space \[ A^2_{mQ}:=\left\{f\ \text{entire}:\ \int_{\mathbb C}|f(z)|^2e^{-2mQ(z)}\,dA(z)<\infty\right\} \] has reproducing kernel \(K_m(z,w)\), normalized kernel \(k_{m,w_0}(z)\), and density \[ \rho_m(z):=\frac{1}{m}K_m(z,z)e^{-2mQ(z)}. \] This planar normalization is tailored to large-\(m\) asymptotics and to the distinction between spectral and off-spectral regions [1805.00854]. ## 2. Variants across complex-analytic and geometric settings The classical Bergman kernel of a domain \(\Omega\subset \mathbb C^n\) reproduces the Hilbert space \(A^2(\Omega)\) of square-integrable holomorphic functions. For bounded Reinhardt domains, and in particular for monomial polyhedra or elementary Reinhardt domains, the kernel admits monomial series expansions in the torus-invariant variables \(t_j=z_j\overline{w_j}\), and in many cases those series can be summed to explicit rational functions [2308.05834]. This behavior is especially transparent on Hartogs-type and monomial domains, where proper holomorphic maps and Bell’s transformation formula transport kernels from product model domains such as \(\mathbb D\times \mathbb D^*\) [2303.14268]. A different generalization replaces the ambient complex domain by a totally real support. If \(K\subset \mathbb C^n\) is a compact nondegenerate \(C^5\) piecewise-smooth maximally totally real submanifold, \(P_k\) is the space of polynomial restrictions of total degree \(\le k\), and \(\mu\) is a measure on \(K\), then \[ B_k(x)=\sup_{s\in P_k}\frac{|s(x)|^2e^{-2kQ(x)}}{\|s\|^2_{L^2(\mu,kQ)}}=\sum_{j=1}^{d_k}|s_j(x)|^2e^{-2kQ(x)}. \] Here the Bergman kernel function governs extremal polynomial growth rather than holomorphic \(L^2\)-functions on a complex domain. In the smooth maximally totally real case one has the sharp upper bound \(\sup_K B_k\le Ck^n\), whereas the general piecewise-smooth case gives a polynomial growth exponent depending on the Hölder exponent of the weight and integrability properties of the measure [2210.11031]. The notion also survives degeneration. For a proper surjective holomorphic map \(T:X\to C\) from a normal complex space to a Riemann surface, with possibly non-reduced central fiber \(X_0=\sum m_jY_j\), Wang–Zhou define a fiberwise Bergman kernel on the graded space \[ \operatorname{gr}^F H^0(X_0,L_0)=\bigoplus_i \operatorname{gr}^{\xi_i}H^0(X_0,L_0), \] using a filtration induced by normalized vanishing orders and a Hermitian form \(h_0\) obtained from rescaled fiberwise metrics. The resulting fiberwise Bergman kernel \[ P_{X_0}(x):=\sup\{ |s(x)|_{h_0}^2 : s\in \operatorname{gr}^F H^0(X_0,L_0),\ \|s\|_{L^2;X_0}=1\} \] extends the classical reduced-fiber kernel and characterizes constancy of \(h^0(X_t,L_t)\) by continuity in families [2304.01943]. ## 3. Exact formulas in highly structured geometries Recent work has produced exact global formulas for Bergman kernel functions in geometries where the global topology can be summed explicitly. For a polarized abelian variety \(X=\mathbb C^n/\Lambda\) with polarization induced by a positive definite Hermitian form \(H\), the Bergman density of \(H^0(X,L^k)\) is \[ \rho_k(p)=\Big(\frac{k}{2\pi}\Big)^n\left(1+\sum_{v\in \Lambda\setminus \{0\}}e^{-\frac{k}{4}(|v|_{H})^2}\cos\big(2\pi k\,\alpha_v(p)\big)\right), \] where \(e^{2\pi i\alpha_v(p)}=\operatorname{Hol}_L(\gamma_{p,v})\) is the holonomy of the Chern connection along the geodesic loop determined by \(v\in\Lambda\). The formula is exact for every \(k\ge1\). It displays \(\rho_k\) as the constant Bargmann–Fock value \((k/2\pi)^n\) plus a Gaussian theta-series over the lattice, with all point dependence entering through holonomy angles [2511.19003]. An analogous phenomenon occurs on hyperbolic Riemann surfaces. If \(X\) is complete hyperbolic and \((L,h)\) has curvature \(-i\omega\), then under the hypotheses that \(X\) has finite topology or positive injectivity radius one has, for \(k\ge k_0\), \[ \rho_k(p)=\frac{k-\tfrac12}{2\pi}\left(1+\sum_{\gamma\in\mathfrak{G}_p} \cosh^{-2k}\!\Big(\frac{\ell(\gamma)}{2}\Big)\cos\!\big(2\pi k\alpha_\gamma\big)\right), \] where \(\mathfrak G_p\) is the set of geodesic loops based at \(p\), \(\ell(\gamma)\) is the loop length, and \(e^{2\pi i\alpha_\gamma}\) is the holonomy. The constant term \((k-\frac12)/(2\pi)\) is the hyperbolic model density, while the remainder is a global sum over geodesic loops, a structure the authors explicitly compare to the Selberg trace formula [2511.16240]. These formulas clarify a general principle: in flat or homogeneous local models the Bergman density is constant, whereas on the quotient the deviation from constancy is a global summation over topological or geometric data. In the abelian case the summation runs over lattice vectors with Gaussian weights; in the hyperbolic case it runs over geodesic loops with \(\cosh^{-2k}\)-weights. A plausible implication is that exact Bergman-density formulas are most accessible when the universal cover carries a model kernel and the quotient geometry can be encoded by a discrete group action. ## 4. Asymptotics, off-diagonal decay, and localization of extrema On compact polarized manifolds, the Tian–Catlin–Zelditch expansion gives \[ \rho_k(x)\sim\Big(\frac{k}{2\pi}\Big)^n\Big(1+\frac{b_1(x)}{k}+\frac{b_2(x)}{k^2}+\cdots\Big), \] with coefficients given by universal polynomials in curvature and its covariant derivatives [2511.19003]. In flat abelian geometry these local coefficients vanish, and the exact formula shows that the true remainder is beyond all algebraic orders: \[ \rho_k(p)=\Big(\frac{k}{2\pi}\Big)^n\big(1+O(e^{-\frac{k}{4}l_1^2})\big), \] uniformly in \(p\), where \(l_1\) is the length of the shortest nonzero lattice vector. Thus the Bergman kernel function is asymptotically constant to any finite order in \(1/k\), with only exponentially small global corrections [2511.19003]. The off-diagonal kernel exhibits equally explicit global decay. On polarized abelian varieties, \[ |K_k(x,y)|_{h^k}\le \Big(\frac{k}{2\pi}\Big)^n\sum_{\gamma\in\mathfrak{G}_{x,y}} e^{-\frac{k}{4}\ell(\gamma)^2}, \] where \(\mathfrak G_{x,y}\) is the set of geodesic segments joining \(x\) to \(y\). On hyperbolic Riemann surfaces one has \[ |K_k(x,y)|_{h^k}\le \frac{k-\frac12}{2\pi}\sum_{\gamma\in \mathfrak{G}_{x,y}}\cosh^{-k}\!\Big(\frac{\ell(\gamma)}{2}\Big), \] and if the minimizing geodesic is unique the leading term is \(\frac{k-\frac12}{2\pi}\cosh^{-k}(d(x,y)/2)\), with an exponentially smaller correction governed by the gap to the second-shortest segment [2511.19003][2511.16240]. The diagonal formulas also localize maxima and minima. On polarized abelian varieties the global extrema are controlled by the shortest lattice vectors \(S_1\). Under parity assumptions on the polarization form, maxima occur precisely where the holonomies along relevant shortest loops are \(1\), and, for large \(k\), maxima and minima lie in exponentially small neighborhoods of points where \(\mathrm{Hol}_{kL}(\gamma_{p,v})=\pm1\) for \(v\in S_1\); the localization scale is governed by \(l_2^2-l_1^2\) through the Gaussian weights \(e^{-k|v|_H^2/4}\) [2511.19003]. On hyperbolic surfaces of finite topology without cusps and with disjoint systoles, maxima and minima concentrate near systoles when the holonomies of \(kL\) around systoles are \(1\) or \(-1\), respectively, in regions of radius \(\left(\cosh(l_1/2)/\cosh(l'/2)\right)^k\) [2511.16240]. Weighted planar kernels exhibit a different asymptotic regime. In the off-spectral setting, Hedenmalm–Wennman prove uniform asymptotic expansions of normalized kernels and root functions throughout an entire off-spectral component, together with an interface scaling limit in which the density transitions to the bulk by an error function: \[ \varrho_m(\xi)\longrightarrow \operatorname{erf}(2\,\mathrm{Re}\,\xi). \] This transition occurs in a boundary layer of width \(O(m^{-1/2})\), and is a non-local phenomenon rather than a bulk-local TYZ-type expansion [1805.00854]. ## 5. Singular, weighted, and nonclassical Bergman kernel functions A recurrent theme in recent work is that Bergman kernel functions remain meaningful under singular weights, singular supports, and singular fibers, provided the Hilbert-space structure is adjusted appropriately. In degenerating families, the fiberwise Bergman kernel on a non-reduced central fiber is defined not on the raw space \(H^0(X_0,L_0)\) but on a graded quotient determined by valuations along irreducible components, with the \(L^2\)-inner product computed using multiplicities. The resulting kernel satisfies the same supremum characterization as in the reduced case and is continuous on the total space precisely when \(h^0(X_t,L_t)\) is constant [2304.01943]. For weights given by the modulus square of a holomorphic function with finitely many zeros, one can compute the weighted planar Bergman kernel by finite-rank correction. If \(h\) is holomorphic on \(\Omega\) with zero divisor \(\sum m_j[a_j]\), then multiplication by \(h\) identifies \(A^2(\Omega,|h|^2)\) with the closed subspace of \(A^2(\Omega)\) consisting of functions whose jets vanish up to order \(m_j-1\) at each \(a_j\). The weighted kernel can therefore be written as \[ K_{\Omega,|h|^2}(z,w)=\frac{1}{h(z)\overline{h(w)}}\Big(K_\Omega(z,w)-V(z)^*C^{-1}V(w)\Big), \] where \(V(z)\) is assembled from the jet kernels \(\partial_z^k K_\Omega(z,a_j)\) and \(C\) is the corresponding Gram matrix of mixed derivatives. This turns the weighted Bergman kernel function into a finite-codimensional modification of the unweighted one [1309.4800]. For measures supported on maximally totally real submanifolds, the Bergman kernel function no longer measures holomorphic sections on a complex manifold but extremal growth of polynomial restrictions. The sharp smooth-case estimate \[ \sup_K B_k\le Ck^n \] shows that the diagonal growth is polynomial rather than of TYZ type, and reflects the dimension of the totally real support rather than complex curvature data [2210.11031]. Explicit weighted kernels on non-smooth domains often become rational. Monomial polyhedra are a class of bounded pseudoconvex Reinhardt domains whose Bergman kernels are explicitly computable rational functions in the variables \(t_j=z_j\overline{w_j}\), with denominators determined by the defining Laurent monomials and numerators given by finite sums governed by the generating function \(D_k\) [2308.05834]. Similar quotient constructions yield explicit kernels for two-dimensional monomial polyhedra and generalized Hartogs triangles [2303.14268]. ## 6. Rigidity, metrics, zeros, and related problems Because the Bergman kernel function is both a reproducing invariant and a geometric density, it can detect structure beyond local curvature. On polarized abelian varieties, if two positive line bundles \(L\) and \(L'\) have the same curvature form and satisfy \(\rho_{L',k}=\rho_{L,k}\) for some \(k\ge1\), then \[ kL'\cong kL. \] The proof uses the explicit lattice-sum formula to recover holonomy data from averages of \(\rho_k\), and then applies a holonomy rigidity lemma. Equality of Bergman densities for one tensor power thus forces isomorphism of the corresponding powers of the line bundles [2511.19003]. In degenerating families, Bergman kernel functions control Fubini–Study currents. For line bundles \(L_m=\otimes_j F_j^{p_{j,m}}\) with continuous metrics and positive fiberwise curvature assumptions, the fiberwise averaged Bergman potentials \[ \phi_m(t):=\int_{X_t}(-\log P_m)\,\omega_t^n \] converge uniformly to \(0\) on compact subsets, and the fiberwise Fubini–Study currents converge uniformly, in the testing topology, to the limiting curvature \(\sum_j r_j c_1(h_{j,t})\). This applies in particular to test configurations, including cases with non-reduced central fiber [2304.01943]. The diagonal kernel also underlies intrinsic metrics. For bounded pseudoconvex domains in \(\mathbb C^n\), Chen proves that if \(\Omega\) has \(C^0\) boundary then \[ K_\Omega(z)\gtrsim \delta_\Omega(z)^{-2}, \] where \(\delta_\Omega(z)=\mathrm{dist}(z,\partial\Omega)\). The estimate is obtained from logarithmic capacity on complex lines together with Ohsawa–Takegoshi extension, and yields lower bounds for the Bergman distance in planar domains under capacity-density assumptions [2102.12650]. Explicit formulas make it possible to study zero sets of Bergman kernels, i.e. the Lu Qi-Keng problem. For the Hartogs domain \[ D_{n,m}=\{(z,\zeta)\in \mathbb C^n\times\mathbb C^m:\ \|\zeta\|^2