D'Angelo observed that the condition of being a quotient of squared norms is preserved under summation and multiplication, but not under limits. In fact, he showed that the set of quotients of squared norms is neither open nor closed in the natural topology for Hermitian algebraic functions [D3] . D'Angelo's observation suggests that the problem of describing quotients of squared norms may be somewhat delicate.

In this paper, we completely characterize all those non-negative Hermitian symmetric functions that are quotients of squared norms. The following is our main result.

Theorem 1. A non-negative Hermitian algebraic function $P$  is a quotient of squared norms if and only if $P$  has only basic zeros.

Basic zeros of a non-negative Hermitian algebraic function are defined below (Definition  2.4 ).

Remark. D'Angelo has also characterized quotients of squared norms in [D3] , at least in the case of bihomogeneous Hermitian polynomials. Although he claims that his characterization is not aesthetically pleasing, his necessary and sufficient condition has certain advantages. It would be interesting to show directly that D'Angelo's necessary and sufficient condition for a non-negative Hermitian algebraic function to be a quotient of squared norms is equivalent to the absence of any zeros that are not basic.

Our proof of Theorem  1 has two important ingredients. One ingredient is Hironaka's Theorem on the resolution of singularities. A second ingredient is a generalization of Theorem  2.8 of Catlin-D'Angelo, namely Theorem  3 below.

The method of proof of the Catlin-Zelditch Theorem needed in the proof of Theorem  2.8 requires the strict positivity of a certain curvature (more precisely, the strict pseudoconvexity of a certain disk bundle) that arises naturally in the problem. Such positivity is absent in our setting, and thus the Theorem of Boutet de Monvel and Sjöstrand cannot be used to prove Theorem  3 .

The information we need for the proof of Theorem  3 is somewhat less than the sort of information contained in the asymptotic expansion of Boutet de Monvel and Sjöstrand. We prove the needed results by elementary methods; we use only the sub-mean value property for plurisubharmonic functions together with a method employed in the typical proof of Stirling's formula, and we avoid the use of Fourier integral operators. Our approach, influenced by the paper [B] of R. Berman, results in a more elementary proof of Theorem  2.8 . The analysis of Section  3 is the most delicate part of the paper. To state our next result, we refer the reader to Definition  2.2 below, in which we associate to a non-negative Hermitian algebraic function $P$  its so-called modulus $\mathit{\forall }P\mathit{\forall }.$  This modulus is a sum of squared norms, and the condition that $P$  has only basic zeros is that $\mathit{\forall }P\mathit{\forall }/P$  is a bounded function.

Theorem 2. If a non-negative Hermitian algebraic function $P$  has only basic zeros, then there is an integer $N\ge 0$  such that for all $m\ge N$  , $$\mathit{\forall }P{\mathit{\forall }}^m\otimes P\in {\mathfrak{P}}_{\infty }(X,F^{m+1}).$$ 

Theorem  2 implies Theorem  1 , and indeed the former is somewhat harder to prove. After the proof of Theorem  1 and a brief discussion of D'Angelo's jet pullback property (Section  5 ), the remainder of the paper, while also of independent interest, develops the theory of non-negative Hermitian algebraic functions sufficiently to establish Theorem  2 .

In the literature Hermitian algebraic functions that are positive away from the zero section have been called Globalizable metrics, or simply G-metrics. We slightly augment this tradition, using the term G-metric to refer to a non-negative Hermitian algebraic function. In section  2 we shall recall certain aspects of the theory of G-metrics, as invented by Catlin-D'Angelo [CD2] and further developed in [DV] .

Acknowledgment. I am grateful to John D'Angelo for introducing me to the study of G-metrics and posing to me the general problem of characterizing quotients of squared norms, and for his generous help with the preparation of this manuscript.

1 Some notation and background from complex and algebraic geometry

In this section we discuss some preliminary ideas needed in this work. We also take the opportunity to establish notation that will be used in the rest of the paper.

Holomorphic line bundles and vector bundles

Let $X$  be a complex manifold and $E\to X$  a holomorphic line bundle. We denote by $H^0(X,E)$  the space of global holomorphic sections of $E$  . By the compactness of $X$  , $H^0(X,E)$  is a finite-dimensional vector space.

Remark. Note that by Hartogs' Theorem, if a line bundle is specified on the complement of an analytic subvariety $V\subset X$  of codimension at least 2, then it extends uniquely to a line bundle on $X$  , since the transition functions, being holomorphic, extend across $V$  . Similarly, all holomorphic sections, either local or global, extend across $V$  .

Recall that given a divisor $D\subset X$  , there is a holomorphic line bundle $L_D\to X$  and a meromorphic section $s_D:X\to L_D$  of $L_D$  such that the divisor of $s_D$  (i.e. the zero divisor minus the polar divisor) is precisely $D$  . In particular, if $D$  is effective then $s_D$  is holomorphic. The line bundle $L_D$  is defined by the transition functions $$\frac{f_{\alpha }}{f_{\beta }}\text{on}{U}_{\alpha }\cap U_{\beta },$$  where $f_{\alpha }$  is the local defining function for $D$  on $U_{\alpha }$  .

When we consider $L^2$  spaces of sections of a holomorphic line bundle, we shall need the notion of a singular Hermitian metric for a holomorphic line bundle $E\to X$  . Recall that a Hermitian metric on a complex line bundle $E\to X$  is a smooth family of Hermitian metrics on the fibers of $E$  . Given a local section $v$  of $E$  over a neighborhood $U$  , we denote its norm with respect to this metric by $$|v{|}^2{e}^{-{\phi }_U}.$$  The functions ${\phi }_U$  are smooth, and by convention we always assume that the local trivialization is holomorphic. It follows that the $(1,1)$  -form $\sqrt{-1}\partial \overline{\partial }{\phi }_U$  is globally defined on $X$  . This $(1,1)$  -form is called the curvature of the Hermitian metric. As is commonly done, we shall write $\phi$  instead of ${\phi }_U$  .

A singular Hermitian metric is similar to a Hermitian metric except that the functions $\phi$  are required to be only locally integrable. The curvature $$\sqrt{-1}\partial \overline{\partial }\phi$$  of this metric is a well-defined $(1,1)$  -current. We will say that the singular metric has positive (resp.

non-negative) curvature if the current $\sqrt{-1}\partial \overline{\partial }\phi$  is positive definite (resp. positive semi-definite) in the sense of currents, i.e., if the metric is given locally by $e^{-\phi }$  with $\phi$  strictly plurisubharmonic (resp. plurisubharmonic).

When emphasis is required, one sometimes calls a Hermitian metric a smooth Hermitian metric.

Spaces of $L^2$  -sections and the generalized Bergman kernel

Suppose now that $X$  supports a non-negative $(1,1)$  -form that is smooth and strictly positive on the complement of some analytic subset. Then $$\Omega :=\frac{{\omega }^n}{n!(2\pi {)}^n}$$  is a measure, and can be thought of as a smooth volume form off a set of Lebesgue measure zero.

Suppose given a holomorphic line bundle $E\to X$  with singular Hermitian metric $e^{-\phi }$  . Given smooth sections $f,g$  of $E$  , we define $$(f,g):={\int }_{X}f\overline{g}{e}^{-\phi }\Omega .$$  With this inner product, we denote by $$L^2\left(e^{-\phi }\right)$$  the Hilbert space completion of the set of smooth sections $f$  of $E\to X$  such that $$\left|\right|f|{|}^2={\int }_X|f{|}^2{e}^{-\phi }\Omega <+\infty .$$  The Hilbert space $L^2\left(e^{-\phi }\right)$  has a closed, finite-dimensional subspace $${\mathfrak{A}}^2\left(e^{-\phi }\right):=L^2\left(e^{-\phi }\right)\cap H^0(X,E).$$  The projection $${\mathbb{K}}_E:L^2\left(e^{-\phi }\right)\to {\mathfrak{A}}^2\left(e^{-\phi }\right)$$  is an integral operator, called the (generalized) Bergman projection. Its kernel, $K_E$  , is given by the formula $$K_E(x,y)={\sum }_{\alpha =1}^N{s}^{\alpha }\left(x\right)\otimes \overline{s^{\alpha }\left(y\right)},$$  where $\{s^1,...,s^N\}$  is an orthonormal basis for ${\mathfrak{A}}^2\left(e^{-\phi }\right)$  . The function $$K_E(x,x):={\sum }_{\alpha =1}^N\left|s^{\alpha }\right(x){|}^2$$  is sometimes called the Bergman function. It is well known that the Bergman function can also be given as a solution of the following variational problem:

$$\begin{array}{ccc}{K}_E(x,x):=sup\left\{\right|s\left(x\right){|}^2;\left|\right|s\left|\right|=1\}.& & \end{array}$$

(2)

This formula can be seen as follows. By completing an extremal section to an orthonormal basis, one sees that $K_E(x,x)$  is bounded below by the right hand side of ( 2 ) . To see the upper bound, choose an orthonormal basis for the subspace of ${\mathfrak{A}}^2\left(e^{-\phi }\right)$  consisting of those sections that vanish at $x$  , and complete to an orthonormal basis of ${\mathfrak{A}}^2\left(e^{-\phi }\right)$  . (If there is a global holomorphic section of $E$  that is non-zero at $x$  , the set of sections vanishing at $x$  forms a hyperplane.)

Resolution of singularities

We shall need the two versions of Hironaka's Theorem on resolution of singularities. [H] . The first is the usual Hironaka Theorem, while the second is a log resolution of linear systems. The precise results we need are as follows.

Theorem 1.1. (Hironaka) Suppose $X$  is an algebraic variety. Let $X_{sing}$  be the singular subvariety of $X$  , i.e., the complement of the set of points where $X$  is a smooth manifold. Then there is a smooth algebraic variety $\tilde{X}$  and a birational map $\mu :\tilde{X}\to X$  such that

• (1) ${\mu }^{-1}\left(X_{\text{sing}}\right)=D$  is a divisor in $X$  with simple normal crossing support, and
• (2) $\mu :\tilde{X}-D\to X-X_{\text{sing}}$  is a holomorphic diffeomorphism.

The map $\mu$  is called a resolution of singularities.

Theorem 1.2. (Hironaka) Let $F\to X$  be a holomorphic line bundle and $V\subset H^0(X,F)$  a non-zero finite-dimensional subspace. Then there is a birational map $\mu :\tilde{X}\to X$  with $\tilde{X}$  smooth, a normal crossing divisor $\sum D_j$  , and non-negative integers $m_j,e_j$  such that, with $D=\sum m_j{D}_j$  and $s_D$  the canonical holomorphic section of the line bundle associated to $D$  whose zero divisor is $D$  ,

• (1) $J:=divisor(det(D\mu \left)\right)=\sum e_j{D}_j$  ,
• (2) ${\mu }^{*}V=s_D\otimes W$  , and
• (3) $W\subset H^0(\tilde{X},{\mu }^{*}F-D)$  is free.

The map $\mu$  in Hironaka's Theorem is called a log resolution of $V$  .

We will use Theorem  1.2 more often than Theorem  1.1 . We will also not make use of the fact that the exceptional divisor has normal crossing support.

Remark. Hironaka's original theorems actually say more than this, providing $\mu$  as a composition of blow-ups along smooth centers. However, the above theorem has a significantly more elementary proof. See, for example, [BP, AD, P] or the discussion in [L,PageI-241] .

2 Elementary Theory of Globalizable metrics

Globalizable forms and their support spaces

Suppose given a holomorphic line bundle $F\to X$  . To a Hermitian form $\mathsf{ℭ}$  on $H^0(X,F)$  we can associate a function $$P:F^{*}\times (F^{*}{)}^{†}\to \mathbb{C}$$  as follows. Fix a basis $\left\{s^{\alpha }\right\}$  of $H^0(X,F)$  and let $${\mathsf{ℭ}}^{\alpha \overline{\beta }}:=\mathsf{ℭ}(s^{\alpha },s^{\beta })$$  denote the matrix of $\mathsf{ℭ}$  in this basis. One can then decompose $H^0(X,E)$  into the eigenspaces of $\mathsf{ℭ}$  :

$$H^0(X,F)={\oplus }_{\lambda \in \mathbb{R}}{V}_{\mathsf{ℭ}}^{\lambda }.$$  Let $$V_P:={\oplus }_{\lambda \ne 0}{V}_{\mathsf{ℭ}}^{\lambda }.$$  The Hermitian matrix $\left({\mathsf{ℭ}}^{\alpha \overline{\beta }}\right)$  is invertible on $V_P$  . Let $\left({\mathsf{ℭ}}_{\alpha \overline{\beta }}\right)$  denote the inverse matrix, and set $$P(v,\overline{w}):={\mathsf{ℭ}}_{\alpha \overline{\beta }}\langle s^{\alpha },v\rangle \overline{\langle s^{\beta },w\rangle }.$$  (Recall that we are employing summation convention.) We will often use the notation $$P={\mathsf{ℭ}}_{\alpha \overline{\beta }}{s}^{\alpha }{\overline{s}}^{\beta }.$$  Note that by polarization, the function $P$  is completely determined by the values $P(v,\overline{v})$  along the “diagonal”.

Remark. By using the inverse matrix $\left({\mathsf{ℭ}}_{\alpha \overline{\beta }}\right)$  , the function $P$  is uniquely determined by the Hermitian form $\mathsf{ℭ}$  , and does not depend on the basis $\left\{s^{\alpha }\right\}$  of $H^0(X,F)$  .

Definition 2.1. The function $P$  is called a globalizable form, or simply G-form, and we write $$P\in {\mathfrak{P}}_0(X,F).$$  The subspace $V_P\subset H^0(X,F)$  is called the support space of $P$  , and the Hermitian form $\mathsf{ℭ}$  the associated Hermitian form.

By definition of the support space, there is a basis $f^1,...,f^k,g^1,...,g^{\ell }$  such that

$$\begin{array}{ccc}P=|f^1{|}^2+\cdots +|f^k{|}^2-\left(\right|g^1{|}^2+\cdots +\left|g^{\ell }{|}^2\right).& & \end{array}$$

(3)

Any such basis will be called a distinguished basis for $P$  .

Definition 2.2. Let $s^1,...,s^r$  be a distinguished basis for $P$  . Then we define the modulus of $P$  to be the G-form $$\mathit{\forall }P\mathit{\forall }:=|s^1{|}^2+\cdots +|s^r{|}^2.$$ 

In other words, for a G-form $P$  , $\mathit{\forall }P\mathit{\forall }$  is the G-form whose support space is the same as that of $P$  and whose eigenspaces in $H^0(X,F)$  are the same as those of $P$  , but whose eigenvalues are the absolute values of those of $P$  . In the representation ( 3 ), we have $$P=\left|\right|f|{|}^2-|\left|g\right|{|}^2\Rightarrow \mathit{\forall }P\mathit{\forall }=\left|\right|f|{|}^2+|\left|g\right|{|}^2.$$  Finally, we shall make use of the following notion.

Definition 2.3. A G-form $P\in {\mathfrak{P}}_0(X,F)$  is said to be free if its support space $V_P$  is a free linear system, i.e., for each $x\in X$  there exists a section $s\in V_P$  such that $s\left(x\right)\ne 0$  .

Globalizable metrics

If the function $$v\mapsto P(v,\overline{v})$$  takes only non-negative values, we call $P$  a globalizable metric, or simply G-metric, and write $$P\in {\mathfrak{P}}_1(X,F).$$  As mentioned in the introduction, our notation differs slightly from that of [CD2] in that we do not require G-metrics to be strictly positive away from the zero section.

A G-metric $P\in {\mathfrak{P}}_1(X,F)$  always has zeros, since it is bihomogeneous, and must thus vanish on the zero section $o_{F^{*}}$  of $F^{*}$  . Zeros of $P$  lying in the zero section will be considered trivial, and will be ignored in most of our applications.

If there is a vector $u\in F^{*}-o_{F^{*}}$  such that $P(v,\overline{v})=0$  , then we will say that $v$  is a non-trivial zero for $P$  . If such a non-trivial zero $v$  lies in the fiber of $F^{*}$  over some point $x\in X$  , then we will write $$x\in {\mathcal{Z}}_P.$$  The set ${\mathcal{Z}}_P$  is called the set-theoretic base locus of $P$  .

We let $${\mathfrak{P}}_1^{♯}(X,F)$$  be the set of G-metrics $P$  on $F$  whose base locus ${\mathcal{Z}}_P$  is empty, i.e., $P$  has no non-trivial zeros. A metric $P\in {\mathfrak{P}}_1^{♯}(X,F)$  will also be called a strict G-metric. Eventually we will be interested in G-metrics with non-trivial zero loci. An important generalization of strict G-metric is the notion of a G-metric $P$  with only basic zeros.

Definition 2.4. We say that a point $x\in X$  is a basic zero of $P\in {\mathfrak{P}}_1(X,F)$  if $x\in {\mathcal{Z}}_P$  and for any $s\in V_P$  the function

$$\begin{array}{ccc}{F}^{*}\supset {\pi }^{-1}\left(U\right)\ni v\mapsto \frac{|\langle s,v\rangle {|}^2}{P(v,\overline{v})}& & \end{array}$$

(4)

is bounded in some neighborhood $U$  of $x$  in $X$  . (Here $\pi :F^{*}\to X$  denotes the projection map.) We denote by ${\mathfrak{B}}_P\subset X$  the locus of basic zeros of $P$  . We say that a G-metric has only basic zeros if the function in ( 4 ) is bounded on $X$  . The set of G-metrics for a line bundle $F\to X$  having only basic zeros will be denoted $${\mathfrak{P}}_1^{♮}(X,F).$$ 

Remark. Observe that the function defined by ( 4 ) depends only on the fiber of $F$  containing $v$  , and thus gives rise to a well-defined function on $X$  . We will denote this function by $|s{|}^2/P$  .

Clearly ${\mathfrak{B}}_P\subset {\mathcal{Z}}_P$  . Unlike ${\mathcal{Z}}_P$  , ${\mathfrak{B}}_P$  is always an analytic set. It is also clear that $P$  has only basic zeros if and only if ${\mathfrak{B}}_P={\mathcal{Z}}_P$  .

Remark. Observe that a G-metric $P$  has only basic zeros if and only if the function $v\mapsto \mathit{\forall }P\mathit{\forall }(v,\overline{v})/P(v,\overline{v})$  is bounded on $F^{*}$  . There is an alternative way of phrasing the condition that $P$  has only basic zeros. If we take a distinguished basis $\{f^1,...,f^k,g^1,...,g^{\ell }\}$  of $V_P$  , then $P$  has only basic zeros if and only if there is a constant $\lambda \in [0,1)$  such that, with $f=(f^1,...,f^k)$  and $g=(g^1,...,g^{\ell })$  , $$\left|\right|g|{|}^2\le \lambda |\left|f\right|{|}^2.$$  Note that since $P=\left|\right|f|{|}^2-|\left|g\right|{|}^2$  , this way of stating that $P$  has only basic zeros has the advantage that automatically $P\in {\mathfrak{P}}_1(X,F)$  .

Factoring basic zeros in codimension 1

Remark. If all of the sections lying in the support space $V_P$  of a G-metric $P\in {\mathfrak{P}}_1(X,F)$  vanish on a divisor $D\subset X$  , then one can replace $F$  with the line bundle $F\otimes L_D^{*}$  and $P$  with the G-metric $\tilde{P}=P/|s_D{|}^2$  . By taking $D$  to be the largest divisor in $X$  such that every section of $V_P$  vanishes on $D$  , we would obtain in this way a G-metric $\tilde{P}$  whose basic zero locus has no codimension 1 component.

Thus we will assume from here on out, without loss of generality, that

All of our G-metrics have no codimension 1 basic zero sets.

Associated singular Hermitian metric

To every G-metric $P\in {\mathfrak{P}}_1(X,F)$  we can associate a singular Hermitian metric $e^{-{\phi }_P}$  of $F$  , defined as follows. If $\xi$  is a local section of $F^{*}$  with no zeros, then $${\phi }_P\left(x\right):=logP({\xi }_x,{\xi }_x).$$ 

G-curvature

Given a G-metric $P\in {\mathfrak{P}}_1(X,F)$  , we can define its G-curvatures (see [DV] ) as follows. For each integer $N\ge 1$  , the $N^{\text{th}}$  G-curvature of $P$  along $(v_1,...,v_N)$  is the Hermitian matrix $${\Theta }_{N}P(v_1,...,v_N):=\left(\begin{array}{ccc}P(v_1,{\overline{v}}_1)& \cdot \cdot \cdot & P(v_1,{\overline{v}}_N)\\ \cdot & \cdot \cdot \cdot & \cdot \\ \cdot & \cdot \cdot \cdot & \cdot \\ \cdot & \cdot \cdot \cdot & \cdot \\ P(v_N,{\overline{v}}_1)& \cdot \cdot \cdot & P(v_N,{\overline{v}}_N)\\ \end{array}\right)$$  We let $${\mathfrak{P}}_N(X,F)$$  denote the set of G-metrics $P$  such that for any $v_1,...,v_N\in F^{*}$  the curvature matrix $${\Theta }_{N}P(v_1,...,v_N)$$  is positive semi-definite. Clearly $${\mathfrak{P}}_{N+1}(X,F)\subseteq {\mathfrak{P}}_N(X,F).$$  We write $${\mathfrak{P}}_{\infty }(X,F):={\bigcap }_{N\in \mathbb{N}}{\mathfrak{P}}_N(X,F).$$  In [DV] it is shown that

• (i) ${\mathfrak{P}}_{\infty }(X,F)$  consists of all those G-metrics $P$  whose associated Hermitian form $\mathsf{ℭ}$  is positive semi-definite, and
• (ii) there exists $N_0$  (depending on $F$  ) such that ${\mathfrak{P}}_{N_0}(X,F)={\mathfrak{P}}_{\infty }(X,F)$  . (This conclusion uses the compactness of $X$  .)

In this paper we will only be concerned with the classes ${\mathfrak{P}}_1$  , ${\mathfrak{P}}_2$  and ${\mathfrak{P}}_{\infty }$  .

Sums of squared norms and quotients of squared norms

As we have just said, ${\mathfrak{P}}_{\infty }(X,F)$  consists of those G-metrics $P$  whose associated Hermitian form is positive semi-definite. It follows that there is a basis $\{s^1,...,s^k\}$  of $V_P$  such that $$P={\sum }_{j=1}^k|s^j{|}^2.$$  For this reason, we refer to the elements of ${\mathfrak{P}}_{\infty }$  as sums of squared norms.

Definition 2.5. A G-metric $P\in {\mathfrak{P}}_1(X,F)$  is said to be a quotient of squared norms if there exist sums of squared norms $P_1\in {\mathfrak{P}}_{\infty }(X,F_1)$  and $P_2\in {\mathfrak{P}}_{\infty }(X,F_2)$  such that $$F_1=F\otimes F_{2}and{P}_1=P\otimes P_2.$$  The set of G-metrics on $F$  that are quotients of squared norms will be denoted $\mathcal{Q}(X,F)$  .

In the language of this section, we can rephrase the following beautiful result of Catlin and D'Angelo [CD2] .

Theorem 2.6. For any holomorphic line bundle $F\to X$  , $${\mathfrak{P}}_1^{♯}(X,F)\subset \mathcal{Q}(X,F).$$ 

Theorem  2.6 is a consequence of Theorem  2.8 given below. To state the latter, we need the notion of Global Cauchy-Schwarz metrics.

Global Cauchy-Schwarz metrics

The elements of the class ${\mathfrak{P}}_2$  are also called Global Cauchy-Schwartz metrics, or simply GCS-metrics, because the class ${\mathfrak{P}}_2$  is easily seen to consist precisely of those G-metrics $R$  that satisfy $$\left|R\right(v,\overline{w}){|}^2\le R(v,\overline{v}\left)R\right(w,\overline{w}).$$  D'Angelo [D2] proved that if $R\in {\mathfrak{P}}_2(X,E)$  then the associated singular Hermitian metric $e^{-{\phi }_R}$  of $E$  has semi-positive curvature current.

Definition 2.7. Let $E\to X$  be a holomorphic line bundle. A G-metric $R\in {\mathfrak{P}}_1(X,E)$  is said to be a strong GCS metric (or simply SGCS metric) if the following conditions hold.

• (S1) $R\in {\mathfrak{P}}_1^{♯}(X,E)\cap {\mathfrak{P}}_2(X,E)$  .
• (S2) The matrix ${\Theta }_{2}R(v,w)$  degenerates if and only if either $v=0$  , $w=0$  , or $v$  and $w$  lie in the same fiber of $E$  .
• (S3) The (smooth) metric $e^{-{\phi }_R}$  of $E$  associated to $R$  has strictly positive curvature.

The class of SGCS metrics on $E$  will be denoted ${\mathfrak{P}}_2^S(X,E).$ 

Definition  2.7 is due essentially to Catlin and D'Angelo [CD2] . The main difference is that they do not include property (S3) in the definition, but make it a hypothesis in the following theorem.

Theorem 2.8. (Catlin-D'Angelo [CD2] ) Let $P\in {\mathfrak{P}}_1^{♯}(X,F)$  and $R\in {\mathfrak{P}}_2^S(X,E)$  . Then there is an integer $N_0$  such that for all $m\ge N_0$  , $$R^m\otimes P\in {\mathfrak{P}}_{\infty }(X,E^m\otimes F).$$ 

Remark. By applying Theorem  2.8 with $P=1\in {\mathfrak{P}}_1^{♯}(X,{\mathcal{O}}_X)$  we see that for all sufficiently large $m$  , $R^m\in {\mathfrak{P}}_{\infty }(X,E^m)$  , and thus Theorem  2.6 follows.

To prove Theorem  1 , we will need a version of the notion of strong GCS metric that is better suited to dealing with resolution of singularities. We now introduce such a notion.

Definition 2.9. Let $E\to X$  be a holomorphic line bundle. A G-metric $R\in {\mathfrak{P}}_1(X,E)$  is said to be a pseudo-strong GCS metric (or simply $\Psi$  GCS metric) if in addition to satisfying (S1) and (S2) of definition  2.7 , $R$  also satisfies the following condition.

• ( $\Psi$  ) The (smooth) metric $e^{-{\phi }_R}$  associated to $R$  has strictly positive curvature away from an analytic subvariety of $X$  .

The class of $\Psi$  GCS metrics on $E$  will be denoted ${\mathfrak{P}}_2^{\Psi }(X,E)$  .

Remark. We emphasize that because a $\Psi$  GCS metric has no non-trivial zeros, its curvature is a smooth $(1,1)$  -form that, by assumption, is positive away from an analytic subset of $X$  .

The following examples give a bit of a flavor for what is going on.

Example 2.10. It was observed in [DV] that ${\mathfrak{P}}_2({\mathbb{P}}_n,\mathbb{H})={\mathfrak{P}}_{\infty }({\mathbb{P}}_n,\mathbb{H})$  , and thus it follows that ${\mathfrak{P}}_2^{\Psi }({\mathbb{P}}_n,\mathbb{H})={\mathfrak{P}}_2^S({\mathbb{P}}_n,\mathbb{H})$  .

Example 2.11. Let $E\to X$  be a very ample line bundle. Choose sections $s^1,...,s^k\in H^0(X,E)$  whose common zero locus is empty, and such that $$\sqrt{-1}\partial \overline{\partial }log\sum |s^j{|}^2$$  is a strictly positive $(1,1)$  -form. For example, one could take $s^1,...,s^k$  to be a basis of $H^0(X,E)$  . Then the G-metric $$R:={\sum }_{j=1}^k|s^j{|}^2\in {\mathfrak{P}}_1^{♯}(X,E)$$  lies in ${\mathfrak{P}}_2^S(X,E)\subset {\mathfrak{P}}_2^{\Psi }(X,E)$  .

Nevertheless in this paper metrics do arise that are $\Psi$  GCS but not SGCS. These metrics come up due to our use of the resolution of singularities.

Example 2.12. There exist $\Psi$  GCS metrics that are not SGCS metrics. To see this, fix a line bundle $E\to X$  and an SGCS metric $R\in {\mathfrak{P}}_2^{♯}(X,E)$  . Let $\mu :\tilde{X}\to X$  be the blowup of $X$  at one point. Then ${\mu }^{*}R\in {\mathfrak{P}}_2^{\Psi }(\tilde{X},{\mu }^{*}E)$  but ${\mu }^{*}R\notin {\mathfrak{P}}_2^S(\tilde{X},{\mu }^{*}E)$  , since the curvature of ${\mu }^{*}R$  degenerates on the exceptional divisor of $\mu$  .

Remark. Although we deal with G-metrics that have zero loci in this paper, we always resolve these metrics and divide out the resulting zeros. Ultimately, the analytic method we develop in the next section works only for those G-metrics whose unit disk bundle is bounded.

3 Asymptotics of powers of $\Psi$  GCS metrics

In this section we prove the following theorem.

Theorem 3. Let $X$  be a complex projective algebraic manifold and $F\to X$  , $E\to X$  holomorphic line bundles. Suppose given two G-metrics $P\in {\mathfrak{P}}_1^{♯}(X,F)$  and $R\in {\mathfrak{P}}_2^{\Psi }(X,E)$  . Then there is an integer $N$  such that for all $m\ge N$  , $$R^m\otimes P\in {\mathfrak{P}}_{\infty }(X,E^m\otimes F).$$ 

Remark. Since $R\in {\mathfrak{P}}_2^S(X,E)\subset {\mathfrak{P}}_2^{\Psi }(X,E)$  , Theorem  3 generalizes the Catlin-D'Angelo Theorem  2.8 above.

The technical approach used in this section is inspired by the paper [B] .

An outline of the approach

Let $R$  and $P$  be as in Theorem  3 , and denote by $e^{-\phi }$  and $e^{-\psi }$  the associated smooth Hermitian metrics of $R$  and $P$  respectively.

Theorem  3 claims that for all sufficiently large $m$  , the Hermitian matrix associated to $R^m\otimes P$  (in some and hence any basis of $H^0(X,E^m\otimes F)$  ) is positive definite. To get our hands on the matrix of $R^m\otimes P$  , we invoke the method of orthogonality: if we consider the Hilbert space ${\mathfrak{A}}^2\left(e^{-(m\phi +\psi )}\right)$  , and represent $R^m\otimes P$  in terms of an orhonormal basis $\left\{s^{\alpha }\right\}$  by $$R^m\otimes P={\mathsf{ℭ}}_{\alpha \overline{\beta }}^{\left(m\right)}{s}^{\alpha }{\overline{s}}^{\beta },$$  then with $\pi :E^m\otimes F\to X$  denoting the projection map, $${\mathsf{ℭ}}_{\alpha \overline{\beta }}^{\left(m\right)}={\int }_X{\int }_X\frac{R^m\otimes P(z,\overline{w})s^{\beta }\left(w\right)\overline{s^{\alpha }\left(z\right)}}{R^m\otimes P(z,z)R^m\otimes P(w,w)}\Omega \left(\pi z\right)\Omega \left(\pi w\right).$$  (Note that although $z$  and $w$  are points in $(E^m\otimes F{)}^{*}$  , the integrand is bihomogeneous of degree $0$  in $(E^m\otimes F{)}^{*}\times ((E^m\otimes F{)}^{*}{)}^{†}$  , and thus well defined on $X\times X$  .) Now, the Bergman kernel $K_{E^m\otimes F}$  is the G-metric in ${\mathfrak{P}}_{\infty }(X,E^m\otimes F)$  associated to the identity matrix. To prove that ${\mathsf{ℭ}}^{\left(m\right)}$  is positive definite, one could try to show that as $m\to \infty$  , $R^m\otimes P$  is asymptotic to $m^{-n}{K}_{E^m\otimes F}$  . In fact, this asymptotic statement is not true unless one chooses the Kähler form of $X$  to be the curvature of the Hermitian metric $e^{-\phi }$  .

In view of the reproducing property of the Bergman kernel, the key to understanding the asymptotics of the Bergman kernel lies in studying the asymptotics of the Bergman function, i.e., the restriction of the Bergman kernel to the diagonal. This has been done by numerous authors.

Here we reproduce an elementary approach that we learned from the paper of Berman [B] .

The conditions defining a $\Psi$  GCS metric (see Definition  2.9 ) enter in the analysis of the asymptotics of $R^m\otimes P$  . One is forced to break up $X$  into three regions, where different properties of $\Psi$  GCS metrics are helpful in controlling the integrals that must be studied.

The delicate point in the argument appears because the curvature of the metric $e^{-\phi }$  is degenerate on an analytic set. For this reason, we begin with a Kähler form and then degenerate this form to the curvature of $e^{-\phi }$  . Since the curvature is smooth and also positive away from an analytic set, careful analysis may be used to overcome the technical difficulties.

The Bergman kernel of $H^0(X,E^m\otimes F)$  .

Let ${\omega }_o$  be a Kähler form on $X$  , and set $${\omega }_{\varepsilon }=\sqrt{-1}\partial \overline{\partial }\phi +\varepsilon {\omega }_o.$$  Then for each $\varepsilon >0$  ${\omega }_{\varepsilon }$  is also a Kähler form.

Let $E\to X$  , $F\to X$  be holomorphic line bundles, and fix G-metrics $R\in {\mathfrak{P}}_2^{\Psi }(X,E)$  and $P\in {\mathfrak{P}}_1^{♯}(X,F)$  . We write $e^{-\phi }$  for the smooth metric of $E$  associated to $R$  and $e^{-\psi }$  for the smooth metric of $F$  associated to $P$  . In a similar fashion, given local non-vanishing holomorphic sections $\xi ,\eta$  of $E\to X$  near points $x_o,y_o\in X$  respectively, we write $$r(x,y):=R({\xi }_x,{\eta }_y),$$  and likewise $p(x,y)$  for the local expression of $P$  . Thus $$\phi \left(x\right):=logr(x,x)\text{and}\psi \left(x\right)=logp(x,x)$$  provided we use $\xi =\eta$  when $x$  and $y$  are close. We endow $X$  with the volume element $${\Omega }_{\varepsilon }:=\frac{{\omega }_{\varepsilon }^n}{(2\pi {)}^{n}n!}.$$  Recall the Hilbert spaces $L^2\left(e^{-(m\phi +\psi )}\right)$  and ${\mathfrak{A}}^2\left(e^{-(m\phi +\psi )}\right)$  from Section  1 . Because $R$  and $P$  are strict G-metrics, $${\mathfrak{A}}^2\left(e^{-(m\phi +\psi )}\right)=H^0(X,E^m\otimes F),$$  i.e., every global holomorphic section is $L^2$  . We write $$K_{\varepsilon }^{\left(m\right)}:=K_{E^m\otimes F},$$  for the Bergman kernel with respect to the Kähler form ${\omega }_{\varepsilon }$  as defined in Section  1 .

Let $X_o$  be the set of points $x\in X$  such that $\phi$  is strictly plurisubharmonic in a neighborhood of $x$  . By property $(\Psi )$  of Definition  2.9 , $X_o$  is the complement of an analytic set.

Let ${\mathfrak{D}}_{\varepsilon }:X\to [0,\infty )$  be the function defined by $$\frac{1}{(2\pi {)}^{n}n!}\sqrt{-1}\partial \overline{\partial }\phi \wedge {\omega }_{\varepsilon }^{n-1}={\mathfrak{D}}_{\varepsilon }\cdot {\Omega }_{\varepsilon }.$$  As is well known, ${\mathfrak{D}}_{\varepsilon }\left(x\right)$  is the product of the eigenvalues of $\sqrt{-1}\partial \overline{\partial }\phi \left(x\right)$  with respect to the form ${\omega }_{\varepsilon }$  .

The function ${\mathfrak{D}}_{\varepsilon }$  has the following properties.

• (1) ${\mathfrak{D}}_{\varepsilon }$  is smooth.
• (2) For each $\varepsilon >0$  , ${\mathfrak{D}}_{\varepsilon }\left(x\right)\ne 0$  if and only if $x\in X_o$  .
• (3) $${\mathfrak{D}}_{\varepsilon }{\Omega }_{\varepsilon }=\frac{1}{(2\pi {)}^{n}n!}\left({\left(\sqrt{-1}\partial \overline{\partial }\phi \right)}^n+{\sum }_{k=1}^n{\varepsilon }^k\left(\begin{array}{c}n\\ k\end{array}\right){\left(\sqrt{-1}\partial \overline{\partial }\phi \right)}^k\wedge {\omega }_o^{n-k}\right).$$ 

Let us fix a sequence $c_m$  of positive numbers such that $$c_m\nearrow +\infty \text{and}\frac{c_m}{\sqrt{m}}\to 0.$$  For example, one can take $c_m=m^{1/4}$  . Fix $x\in X_o$  and choose local coordinates $z$  near $x$  such that $z\left(x\right)=0$  and $z^{*}{\omega }_{\varepsilon }$  agrees with the Euclidean metric $\sqrt{-1}\partial \overline{\partial }|z{|}^2$  at $x$  . Next choose near $x$  local sections $\xi$  of $E$  and $\gamma$  of $F$  such that, with $\phi \left(x\right)=logR({\xi }_x,{\xi }_x)$  and $\psi \left(x\right)=logP({\gamma }_x,{\gamma }_x)$  , $$z^{*}\phi ={\phi }_o\left(z\right)+o\left(\right|z{|}^2)and{z}^{*}\psi ={\psi }_o(z)+o(\left|z{|}^2\right)$$  where ${\phi }_o=\sqrt{-1}\partial \overline{\partial }\phi \left(x\right)$  and ${\psi }_o=\sqrt{-1}\partial \overline{\partial }\psi \left(x\right)$  are the $(1,1)$  parts of the quadratic terms in the Taylor series of $\phi \circ z$  and $\psi \circ z$  .

Lemma 3.1. Let $$A_{\varepsilon }\left(m\right):={\int }_{\left|z\right|\le c_m/\sqrt{m}}{e}^{-\left(m{\phi }_o\right(z)+{\psi }_o(z\left)\right)}{\Omega }_{\varepsilon }.$$  Then

• (1) there exist ${\varepsilon }_m\to 0$  such that $${lim}_{m\to \infty }{A}_{\varepsilon }\left(m\right)=\frac{m^{-n}}{{\mathfrak{D}}_{\varepsilon }\left(x\right)}(1+{\varepsilon }_m),$$  and
• (2) $A_{\varepsilon }\left(m\right)$  is uniformly bounded below, independent of $x$  , $\varepsilon$  and $m$  .

Proof. Making the change of variables $\zeta =z\sqrt{m}$  in the integral $A_{\varepsilon }\left(m\right)$  , we see that

$$\begin{array}{ccc}{A}_{\varepsilon }\left(m\right)& \simeq & m^{-n}{\int }_{\left|\zeta \right|\le c_m}{e}^{-\left({\phi }_o\right(\zeta )+\frac{1}{m}{\psi }_o(\zeta \left)\right)}{\Omega }_{\varepsilon }\end{array}$$

$$\begin{array}{ccc}& \simeq & m^{-n}{\int }_{\left|\zeta \right|\le c_m}{e}^{-{\phi }_o\left(\zeta \right)}{\Omega }_{\varepsilon }.\end{array}$$

After a unitary transformation in $\zeta$  , we may assume that ${\phi }_o\left(\zeta \right)$  is diagonal, i.e., $${\phi }_o\left(\zeta \right)={\lambda }_1\left(x\right)|{\zeta }^1{|}^2+\cdots +{\lambda }_n(x\left)\right|{\zeta }^n{|}^2,$$  where $0<{\lambda }_1\left(x\right)\le {\lambda }_2\left(x\right)\le \cdots \le {\lambda }_n\left(x\right)$  are the eigenvalues of $\sqrt{-1}\partial \overline{\partial }\phi \left(x\right)$  . After a scaling of the last integral, we see that for some constants ${\varepsilon }_m\to 0$  , $$A_{\varepsilon }\left(m\right)=(1+{\varepsilon }_m)\frac{m^{-n}}{{\pi }^n{\mathfrak{D}}_{\varepsilon }\left(x\right)}{\int }_{{\mathbb{C}}^n}{e}^{-|w{|}^2}dV\left(w\right).$$  This formula shows statement (1), and the lower bound claimed in statement (2) is also easily read off. The proof is complete. □ We shall need the following well-known lemma. We give here the proof in [B] .

Lemma 3.2. For each $x\in X$  ,

$$\begin{array}{c}{m}^{-n}{K}_{\varepsilon }^{\left(m\right)}(x,x)e^{-\left(m\phi \right(x)+\psi (x\left)\right)}\le {\mathfrak{D}}_{\varepsilon }\left(x\right)+o\left(1\right)asm\to \infty .\end{array}$$

(5)

Moreover, $m^{-n}{K}_{\varepsilon }^{\left(m\right)}(x,x)$  is uniformly bounded on $X$  , independent of $m$  and $\varepsilon$  .

Remark. It is not hard to show, using Hormander's Theorem, that in fact equality holds in ( 5 ). However, we shall not need this additional information.

Proof. Fix $x\in X$  . Let $z$  denote the local coordinates on $X$  that were chosen before the statement of Lemma  3.1 and let $\xi$  and $\eta$  be the local sections chosen there. Finally, let $c_m$  be the constants defined in that same paragraph.

We are going to use the variational definition of $K_{\varepsilon }^{\left(m\right)}$  , i.e., $$K_{\varepsilon }^{\left(m\right)}(x,x)={sup}_{\left|\right|s\left|\right|=1}\left|s\right(x){|}^2.$$  Let $s$  be an element of $H^0(X,E^m\otimes F)$  having unit norm. Writing $s=f\left(z\right){\xi }^m\otimes \gamma$  , for $c_m/\sqrt{m}$  sufficiently small the sub-mean value property shows that

$$\begin{array}{ccc}\left|s\right(x){|}^2{e}^{-\left(m\phi \right(x)+\psi (x\left)\right)}& =& \left|f\right(0){|}^2\end{array}$$

(6)

$$\begin{array}{ccc}& \le & \frac{1}{A_{\varepsilon }\left(m\right)}{\int }_{\left|z\right|\le c_m/\sqrt{m}}\left|f\right(z){|}^2{e}^{-\left(m{\phi }_o\right(z)+{\psi }_o(z\left)\right)}{\Omega }_{\varepsilon },\end{array}$$

where $A_{\varepsilon }\left(m\right)$  is as in Lemma  3.1 . Then the integral on the right hand side of the estimate ( 6 ) is bounded above by $$(1+{\delta }_m){\int }_X|s{|}^2{e}^{-(m\phi +\psi )}{\Omega }_{\varepsilon }=1+{\delta }_m,$$  where ${\delta }_m\to 0$  as $m\to \infty$  . Thus if $x\in X_o$  then ( 5 ) holds by Lemma  3.1 . On the other hand, if $x\notin X_o$  then $A_{\varepsilon }\left(m\right)$  diverges, and thus $K_{\varepsilon }^{\left(m\right)}(x,x)\to 0$  . Since ${\mathfrak{D}}_{\varepsilon }\left(x\right)=0$  , we see again that ( 5 ) holds. Finally, the boundedness statement follows from the lower bound for $A_{\varepsilon }\left(m\right)$  established in Lemma  3.1 . □ In particular, we have the following corollary.

Corollary 3.3.

$$\begin{array}{ccc}{\int }_X{m}^{-n}{K}_{\varepsilon }^{\left(m\right)}(x,x){\Omega }_{\varepsilon }& \le & {\int }_X\frac{{\left(\sqrt{-1}\partial \overline{\partial }\phi \right)}^m}{(2\pi {)}^{n}n!}+O_{\varepsilon }\left(\varepsilon \right)+o_m\left(1\right)\end{array}$$

$$\begin{array}{ccc}& =& {\int }_X{\Omega }_0+O_{\varepsilon }\left(\varepsilon \right)+o_m\left(1\right).\end{array}$$

Powers of a $\Psi$  GCS metric

Lemma 3.4. Let $y\in X_o$  . Let $$C_o\in \left(0,\frac{3}{4}{\lambda }_1\left(y\right)\right),$$  where ${\lambda }_1\left(y\right)$  is the smallest eigenvalue of $\sqrt{-1}\partial \overline{\partial }\phi \left(y\right)$  with respect to ${\omega }_{\varepsilon }\left(y\right)$  . Then there exists $\delta >0$  , $M_1$  , $M_2$  , independent of $y$  , such that as $m\to \infty$  ,

$$\begin{array}{ccc}& & {\int }_X\frac{m^n\left|r\right(x,y\left){|}^{2m}\right|p(x,y){|}^2{\Omega }_{\varepsilon }\left(x\right)}{r(x,x{)}^{m}p(x,x\left)r\right(y,y{)}^{m}p(y,y)}\end{array}$$

$$\begin{array}{ccc}& & =\frac{1}{{\mathfrak{D}}_{\varepsilon }\left(y\right)}+M_1{m}^n{e}^{-\delta m}+M_2{m}^n{e}^{-C_o\sqrt{m}}+o\left(1\right),\end{array}$$

where the term $o\left(1\right)$  is uniform in $y$  .

We emphasize that the only constant depending on $y$  is $C_o$  .

Proof. Let $$f_y\left(x\right)=logr(x,x)+logr(y,y)-log\left|r\right(x,y){|}^2.$$  Then $f_y\left(y\right)=0$  and the condition that $R$  is GCS implies that $f_y$  has a minimum at $y$  . Thus $D{f}_y\left(y\right)=0$  . Moreover, the $(1,1)$  -part of $\left(D^2{f}_y\right)\left(y\right)$  is exactly $\sqrt{-1}\partial \overline{\partial }\phi \left(y\right)$  .

Choose local coordinates $(z,U)$  with $y\in U$  , such that $z\left(y\right)=0$  and $z^{*}{\omega }_{\varepsilon }$  is Euclidean at $y$  . As in the proof of Lemma  3.2 , we may choose the local expression of $r(x,y)$  so that the quadratic term in the Taylor expansion of $f_y\left(z\right)$  near $z=0$  has only $(1,1)$  -part. It is quite important for us here that the neighborhood $U$  need not lie in $X_o$  .

We obtain the Taylor series

$$\begin{array}{ccc}{f}_y\left(z\right)=\sqrt{-1}\partial \overline{\partial }\phi \left(y\right)(z,\overline{z})+...,& & \end{array}$$

(7)

and thus

$$\begin{array}{ccc}& & {lim}_{m\to \infty }{m}^n{\int }_{\left|z\right|<m^{-1/4}}{e}^{-m{f}_y\left(z\right)}\frac{\left|p\right(z,y){|}^2}{p(z,z)p(y,y)}{z}^{*}{\Omega }_{\varepsilon }\end{array}$$

$$\begin{array}{ccc}& =& {lim}_{m\to \infty }{m}^n{\int }_{\left|\zeta \right|<m^{1/4}}{e}^{-\left({\lambda }_1\right(y\left)\right|{\zeta }^1{|}^2+\cdots +{\lambda }_n\left(y\right)\left|{\zeta }^n{|}^2\right)}\frac{\left|p\right(\zeta /\sqrt{m},y){|}^2}{p(\zeta /\sqrt{m},\zeta /\sqrt{m})p(y,y)}dV\left(\zeta \right)\end{array}$$

$$\begin{array}{ccc}& =& \frac{1}{{\mathfrak{D}}_{\varepsilon }\left(y\right)},\end{array}$$

where $\zeta$  is obtained from $\sqrt{m}z$  by a unitary transformation, and $$0<{\lambda }_1\left(y\right)\le \cdots \le {\lambda }_n\left(y\right)$$  are the eigenvalues of $\sqrt{-1}\partial \overline{\partial }\phi \left(y\right)$  with respect to ${\omega }_{\varepsilon }$  .

Next we choose our neighborhood $U$  of $y\in X$  such that the Taylor series ( 7 ) converges well. By the compactness of $X$  and the smoothness of $f_y$  , we may do this so that the neighborhood $U$  does not get too small, independent of $y$  .

By the hypothesis (S2) for $\Psi$  GCS metrics, there is a number $\delta >0$  such that $$x\notin U\Rightarrow f_y\left(x\right)\ge \delta .$$  Since our neighborhood $U$  in $X$  is not too small regardless of $y$  , the number $\delta$  may be taken independent of $y$  . We then have $${\int }_{X-U}{e}^{-m{f}_y\left(x\right)}\frac{\left|p\right(x,y){|}^2}{p(x,x)p(y,y)}{\Omega }_{\varepsilon }\left(x\right)\le e^{-m\delta }{\int }_{X-U}\frac{\left|p\right(x,y){|}^2}{p(x,x)p(y,y)}{\Omega }_{\varepsilon }\left(x\right)=M_1{e}^{-m\delta }.$$  Finally, in view of the Taylor expansion ( 7 ) we have $$f_y\left(z\right)\ge C_o\left|\right|z|{|}^2,z\in U,$$  and thus

$$\begin{array}{ccc}& & {\int }_{z\left(U\right)-\left\{\right|z|<m^{-1/4}\}}{e}^{-m{f}_y\left(z\right)}\frac{\left|p\right(z,y){|}^2}{p(z,z)p(y,y)}dV\left(z\right)\end{array}$$

$$\begin{array}{ccc}& \le & {\int }_{z\left(U\right)-\left\{\right|z|<m^{-1/4}\}}{e}^{-m{C}_o\left|\right|z|{|}^2}\frac{\left|p\right(z,y){|}^2}{p(z,z)p(y,y)}dV\left(z\right)\end{array}$$

$$\begin{array}{ccc}& \le & e^{-C_o\sqrt{m}}{\int }_U\frac{\left|p\right(z,y){|}^2}{p(z,z)p(y,y)}dV\left(z\right)\end{array}$$

$$\begin{array}{ccc}& \le & e^{-C_o\sqrt{m}}{sup}_{y\in X}{\int }_X\frac{\left|p\right(x,y){|}^2}{p(x,x)p(y,y)}\Omega \left(x\right)\end{array}$$

$$\begin{array}{ccc}& =& M_2{e}^{-C_o\sqrt{m}}.\end{array}$$

Corollary 3.5. Let $X^{\prime }\subset \subset X_o$  . Then

$$\begin{array}{ccc}{\int }_{X^{\prime }}{\int }_X\frac{m^n\left|r\right(x,y{)}^{m}p(x,y){|}^2{\Omega }_{\varepsilon }\left(x\right){\Omega }_{\varepsilon }\left(y\right)}{r(x,x{)}^{m}p(x,x\left)r\right(y,y{)}^{m}p(y,y)}={\int }_{X^{\prime }}\frac{1}{{\mathfrak{D}}_{\varepsilon }\left(y\right)}{\Omega }_{\varepsilon }\left(y\right)+o\left(1\right).& & \end{array}$$

(8)

Remark. In equation  8  $\left(\right)$  the term $o\left(1\right)$  depends on $X^{\prime }$  in only one way: the constant $C_o$  in Lemma  3.4 may approach $0$  as $X^{\prime }$  approaches $X_o$  . But even if we knew that $C_o$  was bounded away from $0$  , we would still need to use this formulation, simply because the integral on the right hand side of  8  $\left(\right)$  may diverge as $X^{\prime }$  approaches $X_o$  .

Passing to degenerate Kähler forms

Our next goal is to control the constant $C_o$  . Of course, if $\sqrt{-1}\partial \overline{\partial }\phi$  is degenerate then $C_o$  must approach zero as $y$  approaches the complement of $X_o$  . However, we shall now let $\varepsilon \to 0$  . It is then clear that the following holds.

Lemma 3.6. Fix $X^{\prime }\subset \subset X$  . Then $${lim}_{\varepsilon \to 0}{\mathfrak{D}}_{\varepsilon }\left(y\right)=1$$  uniformly on $X^{\prime }$  . Moreover, we may take $C_o=1/2$  for all $y\in X^{\prime }$  as soon as $\varepsilon$  is small enough.

We now summarize all of our results in the limit as $\varepsilon \to 0$  .

Proposition 3.7. Let $\varepsilon \to 0$  .

• (A) For every $\delta >0$  there exists $N$  such that for all $m\ge N$  , $${\int }_X\frac{m^{-n}{K}_{\varepsilon }^{\left(m\right)}(x,x)}{r(x,x{)}^{m}p(x,x)}{\Omega }_0\left(x\right)\le {\int }_X{\Omega }_0+\delta ,$$  and
• (B) As $m\to \infty$  , $${\int }_X{\int }_X\frac{m^n\left|r\right(x,y\left){|}^{2m}\right|p(x,y){|}^2}{r(x,x{)}^{m}p(x,x\left)r\right(y,y{)}^{m}p(y,y)}{\Omega }_0\left(y\right){\Omega }_0\left(x\right)={\int }_X{\Omega }_0+o\left(1\right),$$ 

Proof. Statement (A) is clear from Corollary  3.3 and the uniform boundedness of $K_{\varepsilon }^{\left(m\right)}$  . It remains to show statement (B). To this end, fix a compact set $X^{\prime }\subset \subset X$  . Since $\varepsilon \to 0$  , we see from Lemma  3.6 that ${\mathfrak{D}}_0\equiv 1$  on $X^{\prime }$  and we can take a uniform $C_o$  , namely $1/2$  . Both of these functions are clearly independent of $X^{\prime }$  .

In view of the remark following Corollary  3.5 , since we have let $\varepsilon \to 0$  , the right hand side of equation ( 8 ) is now uniform in $X^{\prime }$  and we may let $X^{\prime }\to X$  . The result is that (B) holds. □

Proof of Theorem  3 .

Fix an orthonormal basis $\{s_m^1,...,s_m^{N_m}\}$  of $H^0(X,E^m\otimes F)$  . Let $s=a^1{s}_m^1+\cdots +a^{N_m}{s}_m^{N_m}$  be a unit vector in $H^0(X,E^m\otimes F)$  . Then

$$\begin{array}{ccc}& & {\left|{\int }_X{\int }_X\frac{\left(m^{n}r(x,y{)}^{m}p(x,y)-K_{\varepsilon }^{\left(m\right)}(x,y)\right)s\left(y\right)\overline{s\left(x\right)}}{r(y,y{)}^{m}p(y,y\left)r\right(x,x{)}^{m}p(x,x)}{\Omega }_0\left(y\right){\Omega }_0\left(x\right)\right|}^2\end{array}$$

$$\begin{array}{ccc}& \le & {\int }_X{\int }_X\frac{{\left|m^{n}r(x,y{)}^{m}p(x,y)-K_{\varepsilon }^{\left(m\right)}(x,y)\right|}^2}{r(y,y{)}^{m}p(y,y\left)r\right(x,x{)}^{m}p(x,x)}{\Omega }_0\left(y\right){\Omega }_0\left(x\right)\end{array}$$

$$\begin{array}{ccc}& =& {\int }_X{\int }_X\frac{{\left|m^{n}r(x,y{)}^{m}p(x,y)\right|}^2}{r(y,y{)}^{m}p(y,y\left)r\right(x,x{)}^{m}p(x,x)}{\Omega }_0\left(y\right){\Omega }_0\left(x\right)\end{array}$$

$$\begin{array}{ccc}& & +{\int }_X{\int }_X\frac{K_{\varepsilon }^{\left(m\right)}(x,y)K_{\varepsilon }^{\left(m\right)}(y,x)}{r(y,y{)}^{m}p(y,y\left)r\right(x,x{)}^{m}p(x,x)}{\Omega }_0\left(y\right){\Omega }_0\left(x\right)\end{array}$$

$$\begin{array}{ccc}& & -2Re{\int }_X{\int }_X\frac{m^n{K}_{\varepsilon }^{\left(m\right)}(x,y)r(y,x{)}^{m}p(y,x)}{r(y,y{)}^{m}p(y,y\left)r\right(x,x{)}^{m}p(x,x)}{\Omega }_0\left(y\right){\Omega }_0\left(x\right)\end{array}$$

$$\begin{array}{ccc}& =& m^n{\int }_X{\int }_X\frac{m^n{\left|r(x,y{)}^{m}p(x,y)\right|}^2}{r(y,y{)}^{m}p(y,y\left)r\right(x,x{)}^{m}p(x,x)}{\Omega }_0\left(y\right){\Omega }_0\left(x\right)\end{array}$$

$$\begin{array}{ccc}& & +m^n{\int }_X\frac{m^{-n}{K}_{\varepsilon }^{\left(m\right)}(x,x)}{r(x,x{)}^{m}p(x,x)}{\Omega }_0\left(x\right)-2Re{\int }_X\frac{m^{n}r(x,x{)}^{m}p(x,x)}{r(x,x{)}^{m}p(x,x)}{\Omega }_0\left(x\right)\end{array}$$

$$\begin{array}{ccc}& =& m^n{\int }_X\Omega \left(x\right)+o\left(m^n\right)+m^n{\int }_X\frac{m^{-n}{K}_{\varepsilon }^{\left(m\right)}(x,x)}{r(x,x{)}^{m}p(x,x)}\Omega \left(x\right)-2m^n{\int }_X\Omega .\end{array}$$

(In the second to last equation, we used the reproducing property of the Bergman kernel in the second and third integrals.) Thus, in view of Proposition  3.7 with $\delta =1/2$  in part (A), we have that

$$\begin{array}{c}{\left|{\int }_X{\int }_X\frac{\left(m^{n}r(x,y{)}^{m}p(x,y)-K_{\varepsilon }^{\left(m\right)}(x,y)\right)s\left(y\right)\overline{s\left(x\right)}}{r(y,y{)}^{m}p(y,y\left)r\right(x,x{)}^{m}p(x,x)}{\Omega }_0\left(y\right){\Omega }_0\left(x\right)\right|}^2\le \frac{m^n}{2}+o\left(m^n\right)\end{array}$$

(9)

as $m\to \infty$  .

Next, let ${\mathsf{ℭ}}^{\left(m\right)}=\left(c_{\alpha \overline{\beta }}^{\left(m\right)}\right)$  be the Hermitian matrix such that with respect to the basis $\left\{s^{\alpha }\right\}$  of $H^0(X,E^m\otimes F)$  , $R^m\otimes P=c_{\alpha \overline{\beta }}^{\left(m\right)}{s}^{\alpha }{\overline{s}}^{\beta }$  . Observe that $$c_{\alpha \overline{\beta }}^{\left(m\right)}={\int }_X{\int }_X\frac{r(x,y{)}^{m}p(x,y\left)s^{\alpha }\right(y)\overline{s^{\beta }\left(x\right)}}{r(y,y{)}^{m}p(y,y\left)r\right(x,x{)}^{m}p(x,x)}{\Omega }_0\left(y\right){\Omega }_0\left(x\right)$$  and thus by ( 9 ) we have

$$\begin{array}{ccc}& & {\left|\left(m^n{c}_{\alpha \overline{\beta }}^{\left(m\right)}-{\delta }_{\alpha \overline{\beta }}\right)a^{\alpha }{\overline{a}}^{\beta }\right|}^2\end{array}$$

$$\begin{array}{ccc}& =& {\left|{\int }_X{\int }_X\frac{\left(m^{n}r(x,y{)}^{m}p(x,y)-K_{\varepsilon }^{\left(m\right)}(x,y)\right)s\left(y\right)\overline{s\left(x\right)}}{r(y,y{)}^{m}p(y,y\left)r\right(x,x{)}^{m}p(x,x)}{\Omega }_0\left(y\right){\Omega }_0\left(x\right)\right|}^2\end{array}$$

$$\begin{array}{ccc}& \le & \frac{m^n}{2}+o\left(m^n\right).\end{array}$$

But this means there are constants $B_m\to 0$  (as $m\to \infty$  ) such that for all vectors $a$  , $$a^{†}{\mathsf{ℭ}}^{\left(m\right)}a\ge \frac{\frac{1}{2}-B_m}{m^n}\left|\right|a|{|}^2.$$  That is to say, there is an integer $N>>0$  such that ${\mathsf{ℭ}}^{\left(m\right)}$  is positive definite for any $m\ge N$  . The proof is complete. □

4 Quotients of squared norms

The Riemann surface case

Let $M$  be a compact Riemann surface and $\Lambda \to M$  a holomorphic line bundle.

Theorem 4.1. A G-metric $P\in {\mathfrak{P}}_1(M,\Lambda )$  is a quotient of squared norms if and only if there exist holomorphic line bundles $E_1,E_2\to M$  , a holomorphic section $\Phi \in H^0(M,E_1)$  , and a strict G-metric $\tilde{P}$  on $M$  such that $\Lambda =E_1\otimes E_2$  and

$$\begin{array}{ccc}P=|\Phi {|}^2\tilde{P},& i.e.,& R(v,\overline{w})=\Phi \left(v\right)\overline{\Phi \left(w\right)}\tilde{R}(v,\overline{w}).\end{array}$$

(10)

Remark. This theorem was proved by D'Angelo in the case $M={\mathbb{P}}_1$  [D3] . Our proof is similar, but it is crucial for this paper that the surface $M$  is general.

Proof of Theorem  4.1 : The sufficiency of the expression ( 10 ) is a trivial consequence of Theorem  2.6 .

To see the necessity, suppose $P$  is a quotient of squared norms. Then

$$\begin{array}{ccc}P\cdot {\sum }_{\alpha }{\left|s^{\alpha }\right|}^2={\sum }_{\beta }{\left|T^{\beta }\right|}^2& & \end{array}$$

(11)

for some holomorphic sections $\left\{s^{\alpha }\right\}$  and $\left\{T^{\beta }\right\}$  of (different) line bundles.

Suppose now that $x\in M$  . Choose a local coordinate function $z$  near $x$  with $z\left(x\right)=0$  , and let $\xi$  be a nowhere zero section of ${\Lambda }^{*}$  near $x$  . Then by formula ( 11 ) the function $p:=P(\xi ,\overline{\xi })$  has the property that for some non-negative integer $m$  , $p/|z{|}^{2m}$  is bounded and non-zero near $x$  . (If there exists $v\in {\Lambda }_x^{*}-o_{{\Lambda }^{*}}\left(x\right)$  such that $P(v,\overline{v})=0$  , then $m>0$  .) Clearly the integer $m$  is unique. We write $$m=:D_P\left(x\right).$$  We associate to $P$  the divisor $$D_P={\sum }_{x\in M}{D}_P\left(x\right)\cdot x.$$  Evidently $D_P$  is independent of $\xi$  and $z$  . Let $L_{D_P}$  be the line bundle associated to $D_P$  and let $s_{D_P}$  be the canonical section of $L_{D_P}$  whose divisor is $D_P$  . Since $D_P$  is effective, i.e., $D_P\left(x\right)\ge 0$  for all $x\in M$  , $s_{D_P}$  is holomorphic.

By construction, $\tilde{P}=|s_{D_P}{|}^{-2}P$  is a strict G-metric. An application of Theorem  2.6 completes the proof. □ For the purpose of illustration, we prove Theorem  1 for Riemann surfaces.

Necessity of basic zeros: Let $P\in {\mathfrak{P}}_1(M,\Lambda )$  . First, suppose $P$  is a quotient of squared norms. In view of Theorem  4.1 , $P=|s_D{|}^2\otimes \tilde{P}$  for some strict G-metric $\tilde{P}$  . It follows that $s_D$  divides the support space $V_P$  of $P$  . Thus for any $s_D\otimes f\in V_P$  we have that $$\frac{|s_D\otimes f{|}^2}{P}=\frac{|f{|}^2}{\tilde{P}}$$  is bounded. That is to say, condition ( 4 ) holds.

Sufficiency of basic zeros: Conversely, suppose condition ( 4 ) holds for $P$  . If $V_P$  is free, then $P$  has no zeros and by Theorem  4.1 we are done. Therefore, assume $V_P$  has a non-trivial base locus.

Then there is a divisor $D$  on $M$  whose canonical holomorphic section $s_D$  divides every element of $V_P$  . Take $D$  to be the largest such divisor. Thus $|s_D{|}^2$  divides $P$  , and the metric $\tilde{P}=P/|s_D{|}^2$  is a G-metric for the line bundle $E=\Lambda \otimes L_D^{*}$  . Evidently $V_{\tilde{P}}$  is free. Moreover, $$\frac{|s{|}^2}{P}=\frac{|s/s_D{|}^2}{\tilde{P}},$$  And thus by condition ( 4 ) $\tilde{P}$  has no zeros. By Theorem  4.1 we are done. □

Corollary 4.2. If $R$  is a G-metric on a Riemann surface and $V_R$  is free, then $R$  is a quotient of squared norms if and only if $R$  is a strict G-metric, i.e., $R$  has no non-trivial zeros.

The free case

Corollary  4.2 generalizes to higher dimensions.

Theorem 4.3. Let $P\in {\mathfrak{P}}_1(X,F)$  be a G-metric such that $V_P$  is free. Then $P$  is a quotient of squared norms if and only if $P$  is a strict G-metric.

Proof. The “if ” direction follows from Theorem  2.8 or its generalization, Theorem  3 . We shall now prove the converse. Suppose $P$  is a quotient of squared norms and $P(v,\overline{v})=0$  for some $v\in F_x^{*}-o_{F^{*}}\left(x\right)$  . Since $X$  is projective algebraic, there is a Riemann surface $M\subset X$  passing through $x$  . Let $\iota :M\hookrightarrow X$  denote the natural inclusion map. Then the following facts are evident.

• (1) ${\iota }^{*}{V}_P=V_{{\iota }^{*}P}$  , and thus ${\iota }^{*}{V}_P$  is free.
• (2) ${\iota }^{*}P$  is a quotient of squared norms.
• (3) ${\iota }^{*}P$  vanishes along $({\iota }^{*}F{)}_x$  .

But these three facts contradict Theorem  4.1 . The proof is complete. □

An illustrative example

D'Angelo showed that the G-metric $P\in {\mathfrak{P}}_1({\mathbb{P}}_2,{\mathbb{H}}^4)$  given in homogeneous coordinates $[z_0,z_1,z_2]$  by $$P(z,\overline{z})=|z_0{|}^8+(|z_0{z}_2{|}^2-|z_1{|}^4{)}^2$$  is not a quotient of squared norms. The interesting aspect of this G-metric is that its zero locus ${\mathcal{Z}}_P$  is the single point $[0,0,1]$  . In particular, ${\mathcal{Z}}_P$  is an analytic set.

D'Angelo showed that $P$  is not a quotient of squared norms by showing that $P$  does not satisfy a certain necessary condition for being a quotient of squared norms: the jet pullback property.

We will recall the jet pullback property in Section  5 below. In the language of the present paper, D'Angelo passes the (rational) curve $\gamma :{\mathbb{P}}_1\to {\mathbb{P}}_2$  given by $$\gamma [x,y]=[x^2,xy,xy+y^2]$$  through $[0,0,1]$  and examines the restriction of $P$  to this curve. He then deduces from the Taylor expansion of the curve near $[0,0,1]$  that $P$  cannot be a quotient of squared norms.

The information about the influence of the Taylor expansion of the curve on $P$  is already contained in Theorem  4.1 . To rephrase D'Angelo's proof in the language of the present paper, note that ${\gamma }^{*}P\in {\mathfrak{P}}_1({\mathbb{P}}_1,{\mathbb{H}}^8)$  and

$$\begin{array}{ccc}{\gamma }^{*}P\left(\right(x,y),(\overline{x},\overline{y}\left)\right)& =& |x{|}^{16}+(|x^{3}y+x^2{y}^2{|}^2-|xy{|}^4{)}^2\end{array}$$

$$\begin{array}{ccc}& =& |x{|}^8\left(|x{|}^8+(|xy+y^2{|}^2-|y{|}^4{)}^2\right)\end{array}$$

$$\begin{array}{ccc}& =& \left|x{|}^{8}Q\right((x,y),(\overline{x},\overline{y})).\end{array}$$

One then sees that $Q$  is a free G-metric on ${\mathbb{P}}_1$  with a non-trivial zero at the point $[0,1]\in {\mathbb{P}}_1$  . By Corollary  4.2  $Q$  is not a quotient of squared norms. Thus ${\gamma }^{*}P$  is not a quotient of squared norms, and hence $P$  is not a quotient of squared norms.

In order to demonstrate the role of the resolution of singularities in the proof of Theorem  1 , we shall use blowups to show that the G-metric $P$  is not a quotient of squared norms. Since the zero locus of $P$  consists only of the point $[0,0,1]\in {\mathbb{P}}_2$  , it is easiest to work in the affine chart $\{z^2\ne 0\}$  .

Let $x_1=z_0/z_2$  and $x_2=z_1/z_2$  . Then $$P=\left|z_2{|}^8\right(|x_1{|}^8+(|x_1{|}^2-|x_2{|}^4{)}^2).$$  It suffices to work with the inhomogeneous polynomial $$p=|x_1{|}^8+(|x_1{|}^2-|x_2{|}^4{)}^2,$$  since one can always recover the original G-metric by homogenization. (For more on this point as well as other aspects of homogenization of polynomials with regard to G-metrics, see [D3] .) Now blow up the origin in ${\mathbb{C}}^2$  . That is to say, let $$x_1=y_1{y}_2\text{and}{x}_2=y_2.$$  Then with $p_1$  denoting the blowup of $p$  , one has $$p_1=|y_2{|}^4\left(|y_1^2{y}_2{|}^4+(|y_1{|}^2-|y_2{|}^2{)}^2\right).$$  Now take one more blowup, namely $$y_1=t_1\text{and}{y}_2=t_1{t}_2.$$  Then with $p_2$  denoting the blowup of $p_1$  , one has $$p_2=|t_1^2{t}_2{|}^4\left(|t_1^2{t}_2{|}^4+(1-|t_2{|}^2{)}^2\right).$$  We have shown that after two blowups, $p$  is transformed into a G-metric that is a product of a square (the term $\left|\right(t_1^2{t}_2{)}^2{|}^2$  ) and a free G-metric with a non-trivial zero, namely the point $(t_1,t_2)=(0,1)$  in the particular affine chart we are working with. It follows from Theorem  4.3 that the G-metric $P$  could not have been a quotient of squared norms.

The general case

We remind the reader that we have assumed the absence of zeros in codimension 1.

Let $\mu :\tilde{X}\to X$  be a log resolution of $V_P$  as in Theorem  1.2 . Then there is an effective divisor $D$  in $\tilde{X}$  , a free linear system $W\subset H^0(\tilde{X},F\otimes L_D^{*})$  and a section $s_D$  of the line bundle associated to $D$  such that for all $s\in {\mu }^{*}{V}_P$  , $$s=s_D\otimes t\text{for some}t\in W.$$  It follows that ${\mu }^{*}P=|s_D{|}^2\otimes \tilde{P}$  for some G-metric $\tilde{P}\in {\mathfrak{P}}_1(X,{\mu }^{*}F\otimes L_D^{*})$  such that $V_{\tilde{P}}$  is free.

We now show the necessity of condition ( 4 ). To this end, if $P$  is a quotient of squared norms, then so is $\tilde{P}$  . Since $V_{\tilde{P}}$  is free, by Theorem  4.3 we see that $\tilde{P}$  has no zeros. It follows that if $s\in V_P$  and ${\mu }^{*}s=s_D\otimes t$  for some $t\in V_{\tilde{P}}$  , then the quotient $|t{|}^2/\tilde{P}$  is bounded. But then the quotient $$\frac{|{\mu }^{*}s{|}^2}{{\mu }^{*}P}=\frac{\left|s_D{|}^2\right|t{|}^2}{|s_D{|}^2\tilde{P}}$$  is bounded on $\tilde{X}$  . Since $\mu$  is an isomorphism on a dense subset, we see that condition ( 4 ) must hold.

Next we turn to the sufficiency of condition ( 4 ). To this end, choose a non-trivial line bundle $E\to X$  and a subspace $M\subset H^0(X,E)$  with the following properties:

• (1) The base locus $G$  of ${\mu }^{*}M$  is a divisor that divides $D$  .
• (2) There is a metric $R\in {\mathfrak{P}}_2^{\Sigma }(X,E)$  such that $V_R=M$  .

For example, $E$  could be very ample and $M=H^0(X,E)$  . In this case $G=\varnothing$  .

Let $U={\mu }^{*}M/s_G$  . Then $U$  is free, and we have ${\mu }^{*}R=\tilde{R}\otimes |s_G{|}^2$  where $\tilde{R}$  is free. It follows from Theorem  3 that there is an integer $N\ge 0$  such that for all $m\ge N$  ,

$$\begin{array}{ccc}{\tilde{R}}^m\otimes \tilde{P}={\sum }_{\ell =1}^r|{\xi }^{\ell }{|}^2\in {\mathfrak{P}}_{\infty }(\tilde{X},({\mu }^{*}E\otimes L_G^{*}{)}^m\otimes ({\mu }^{*}F\otimes L_D^{*}\left)\right).& & \end{array}$$

(12)

Tensoring both sides of ( 12 ) by $|s_G{|}^{2m}\otimes |s_D{|}^2$  , we see that $${\mu }^{*}\left(R^m\otimes P\right)={\sum }_{j=1}^r|{\kappa }^j{|}^2,$$  where $${\kappa }^j:=s_G^m\otimes s_F\otimes {\xi }^{\ell }.$$  It follows from Hartogs' Theorem and the absence of codimension 1 basic zeros for $P$  that $$R^m\otimes P={\sum }_{\ell =1}^r|{\eta }^{\ell }{|}^2,$$  for some sections ${\eta }^{\ell }$  of $E^m\otimes F$  . By Theorem  3 we see that for all $m>>0$  , $R^m\in {\mathfrak{P}}_{\infty }(X,E^m)$  , and thus $P$  is a quotient of squared norms. The proof is complete. □

5 The jet pullback property

Let us recall the definition of D'Angelo's jet pullback property.

Definition 5.1. A G-metric $P\in {\mathfrak{P}}_1(X,F)$  is said to have the jet pullback property if for every compact Riemann surface $M$  and holomorphic map $h:M\to X$  the following holds. If $z$  is a local coordinate in a neighborhood of $x\in M$  and $\xi$  is a nowhere zero holomorphic section of $F$  in a neighborhood of $h\left(x\right)$  , then the function $h^{*}\left(P\right(\xi ,\xi \left)\right)$  has Taylor expansion $$h^{*}\left(P\right(\xi ,\xi \left)\right)\left(z\right)=c|z{|}^{2d}+\text{higher order terms}$$  for some $c>0$  .

Remark. Strictly speaking, in D'Angelo's definition of the jet pullback property one is only allowed to use rational curves, i.e., $M={\mathbb{P}}^1$  . We shall take this stronger definition, since rational curves might not be so plentiful on $X$  . Indeed, it may happen that $X$  has no rational curves at all.

It is easy to see that if $P$  is a quotient of squared norms then $P$  has the jet pullback property.

In [D1] it is asked whether the converse is true. In this section we answer this question in the affirmative.

Proposition 5.2. If $P\in {\mathfrak{P}}_1(X,F)$  has the jet pullback property then $P$  is a quotient of squared norms.

• Proof. Suppose $P$  has the jet pullback property. We shall show that $P$  has only basic zeros, and then use Theorem  1 .
  (i) As in the previous section, let $\mu :\tilde{X}\to X$  be a resolution of singularities such that ${\mu }^{*}P=|s{|}^2\tilde{P}$  for some holomorphic section $s$  and free G-metric $\tilde{P}$  .

  (ii) Observe that ${\mu }^{*}P$  has the jet pullback property. Indeed, if $h:M\to \tilde{X}$  is a curve, then $$h^{*}\left({\mu }^{*}P\right)=(\mu \circ h{)}^{*}P,$$  and thus it is clear from the jet pullback property for $P$  that the lowest order term of the Taylor expansion for $h^{*}\left({\mu }^{*}P\right)$  at any point of $M$  has the right form.

  (iii) Notice that since $|s{|}^2\tilde{P}$  has the jet pullback property, so does $\tilde{P}$  .

  (iv) If the free G-metric $\tilde{P}$  has the jet pullback property, then $\tilde{P}$  must be a strict G-metric. Indeed, since $\tilde{P}$  is free, the pullback $h^{*}\tilde{P}$  of $\tilde{P}$  to any curve $h:M\to \tilde{X}$  is free. Moreover, it is immediate that $h^{*}\tilde{P}$  has the jet pullback property. But since $M$  is a curve, the jet pullback property for $h^{*}\tilde{P}$  means that the zero locus of $h^{*}\tilde{P}$  is a divisor. If $\tilde{P}$  is not a strict G-metric, then by choosing the curve $h:M\to X$  such that $h\left(M\right)$  passes through a zero of $\tilde{P}$  , we see that the zero divisor of $h^{*}\tilde{P}$  is non-empty. It follows as in the proof of Theorem  4.1 that every section in the support space of $h^{*}\tilde{P}$  is divisible by a global section on $M$  , and this contradicts that $h^{*}\tilde{P}$  is free.

  (v) If $\tilde{P}$  is a strict G-metric, then clearly $${\mu }^{*}\left(\frac{\mathit{\forall }P\mathit{\forall }}{P}\right)=\frac{\mathit{\forall }\tilde{P}\mathit{\forall }}{\tilde{P}}$$  is uniformly bounded on $\tilde{X}$  . Since $\mu$  is an isomorphism, we see that $\frac{\mathit{\forall }P\mathit{\forall }}{P}$  is uniformly bounded on the complement of a set of measure zero in $X$  . But then $\frac{\mathit{\forall }P\mathit{\forall }}{P}$  is uniformly bounded on $X$  , i.e., $P$  has only basic zeros. By Theorem  1 we are done. □

6 $\Sigma$  GCS metrics

In the course of the proof of Theorem  1 , we established the following generalization of Theorem  3 .

Theorem 6.1. Suppose $P\in {\mathfrak{P}}_1(X,F)$  has only basic zeros, i.e., $P$  satisfies condition ( 4 ) above. Then for any holomorphic line bundle $E\to X$  and $R\in {\mathfrak{P}}_2^{\Psi }(X,E)$  there exists $N>0$  such that for all $m\ge N$  , $$R^m\otimes P\in {\mathfrak{P}}_{\infty }(X,E^m\otimes F).$$ 

Thus we have weakened the conditions required from $P$  in Theorem  3 , allowing it to have basic zeros. We also pointed out that $R$  itself may have zeros, so long as its zero locus is not larger than that of $P$  .

This discussion suggests that perhaps the requirements we place on GCS metrics in the notion of $\Psi$  GCS are too stringent. In fact, this is the case. The methods we have developed thus far make for a generalization of Theorem  6.1 . To state this generalization, it is convenient to introduce another class of GCS metrics that contains the previous classes.

Definition 6.2. A G-metric $R\in {\mathfrak{P}}_1(X,E)$  is said to be a $\Sigma$  GCS metric, written $$R\in {\mathfrak{P}}_2^{\Sigma }(X,E),$$  if the following conditions hold.

• ( $\Sigma$  1) $R\in {\mathfrak{P}}_1^{♮}(X,E)\cap {\mathfrak{P}}_2(X,E)$  .
• ( $\Sigma$  2) For any $v,w\in E$  , $\left|R\right(v,\overline{w}){|}^2<R(v,\overline{v}\left)R\right(w,\overline{w})$  unless
  • (a) $v$  and $w$  lie in the same fiber of $E$  , or
  • (b) $R(v,\overline{v})R(w,\overline{w})=0$  .
• ( $\Sigma$  3) $\sqrt{-1}\partial \overline{\partial }logr(x,x)$  is strictly positive, in the sense of currents, away from an analytic subvariety.

With this definition, we have the following theorem.

Theorem 4. If $P\in {\mathfrak{P}}_1^{♮}(X,F)$  and $R\in {\mathfrak{P}}_2^{\Sigma }(X,E)$  then there is an integer $N>0$  such that for all $m\ge N$  , $$R^m\otimes P\in {\mathfrak{P}}_{\infty }(X,E^m\otimes F).$$ 

Proof. Recall (see Remark  2 ) that we have assumed the base locus of $V_R$  contains no divisors. Let $\mu :\tilde{X}\to X$  be a log resolution, i.e., a birational map such that ${\mu }^{*}{V}_R=W\otimes s_D$  , where $D$  is an effective divisor and $W\subset H^0(\tilde{X},{\mu }^{*}E\otimes L_D^{*})$  is a free linear system. Let $$\tilde{R}:={\mu }^{*}R/|s_D{|}^2.$$  Evidently $\tilde{R}\in {\mathfrak{P}}_2^{\Psi }(\tilde{X},{\mu }^{*}E\otimes L_D^{*})$  .

Let $\tilde{P}:={\mu }^{*}P$  . Then $\tilde{P}\in {\mathfrak{P}}_1^{♮}(\tilde{X},{\mu }^{*}F)$  . It follows from Theorem  6.1 that there is an integer $N$  such that for all $m\ge N$  , $${\tilde{R}}^m\otimes \tilde{P}\in {\mathfrak{P}}_{\infty }(\tilde{X},({\mu }^{*}E\otimes L_D^{*}{)}^m\otimes {\mu }^{*}F).$$  Thus $${\mu }^{*}(R^m\otimes P)=|s_D{|}^{2m}\otimes {\tilde{R}}^m\otimes \tilde{P}\in {\mathfrak{P}}_{\infty }(\tilde{X},{\mu }^{*}(E^m\otimes F)).$$  Since $\mu$  is an isomorphism outside the exceptional divisor, the result follows once more by an application of Hartogs' Theorem. □

Remark. As will be shown in the next section, any G-metric having the property that some power of it is a sum of squared norms can have only basic zeros.

7 Projective isometric maps

Weak $\Sigma$  GCS metrics

If we take $F$  to be the trivial bundle $\mathcal{O}$  and $P\equiv 1\in H^0(X,\mathcal{O})$  , then Theorem  4 says that all sufficiently large powers of a $\Sigma$  GCS metric are sums of squared norms. Following [CD2] , we can rephrase this corollary in the language of projective isometric maps.

Corollary 7.1. Let $R$  be a $\Sigma$  GCS metric and set $N_m:=dim\left(V_{R^m}\right)$  . Let ${\omega }_m$  the G-metric on $\mathbb{H}\to {\mathbb{P}}_{N^m-1}$  associated to the identity Hermitian form. Then there are holomorphic maps $T_m:X\to {\mathbb{P}}_{N_m}$  such that $$R^m=T_m^{*}{\omega }_m.$$ 

Thus Corollary  7.1 gives a G-metric analogue of Kodaira's celebrated Embedding Theorem. In fact, the conditions of $\Sigma$  GCS are not necessary for obtaining Corollary  7.1 . To illustrate, consider the following simple example.

Example 7.2. Let $X={\mathbb{P}}_2$  with homogeneous coordinates $[z^0,z^1,z^2]$  , and consider any G-metric $R\in {\mathfrak{P}}_2^{\Sigma }({\mathbb{P}}_1,{\mathbb{H}}^{2d})$  that is not a quotient of squared norms. For example, fix an integer $d$  and a real number $\lambda \in \left(\left(\genfrac{}{}{0ex}{}{2d}{d}\right),2^{2d-1}\right)$  , and take $$R\left(\right(z^1,z^2),({\overline{z}}^1,{\overline{z}}^2\left)\right)=\left(\right|z^1{|}^2+|z^2{|}^2{)}^{2d}-\lambda |z^1{z}^2{|}^{2d}.$$  (To see that $R\in {\mathfrak{P}}_2({\mathbb{P}}_1,{\mathbb{H}}^{2d})$  , see [DV] .) Then $R$  can be thought of as a G-metric in ${\mathfrak{P}}_2({\mathbb{P}}_2,{\mathbb{H}}^{2d})$  that does not depend on $z^0$  , and as such it does not satisfy the hypotheses of Definition  2.9 . In particular, property $(\Sigma 3)$  fails. Nevertheless, by thinking of $R$  as a G-metric over ${\mathbb{P}}_1=\left\{\right[z^1,z^2\left]\right\}$  we see $$R^m\in {\mathfrak{P}}_{\infty }({\mathbb{P}}_2,{\mathbb{H}}^{2md}).$$ 

Said another way, Example  7.2 shows that sometimes a G-metric $R$  may be defined on a space $X$  that has more than the necessary degrees of freedom required by $R$  .

In this section, we will attempt to deal with this phenomenon of excessive spatial degrees of freedom. We begin with the classical Veronese-type construction from algebraic geometry. Fix a free G-metric $R$  with support space $V_R$  . Let $\phi :X\to \mathbb{P}\left(V_R^{*}\right)$  be the map sending $x\in X$  to the hyperplane $$\phi \left(x\right)=\{s\in V_R;s(x)=0\}.$$  Since $R$  is free, $\phi$  is a well defined map. Moreover, by definition we have that $${\phi }^{*}\left({\mathcal{O}}_{\phi \left(X\right)}\left(1\right)\right)={\mathcal{O}}_X\left(E\right).$$  Indeed, a linear functional on $V_R^{*}$  is simply an element of $V_R$  , and so $\phi$  pulls back global sections of $\mathbb{H}$  (over $\phi \left(X\right)$  ) to elements of $V_R$  , which are global sections of $E$  .

Definition 7.3. Let $K_{\phi }$  denote the kernel of the map $D\phi :T_X\to T_{\mathbb{P}\left(V_R^{*}\right)}$  .

In homogeneous coordinates, the mapping $\phi$  can be given as follows: if $\{s^1,...,s^N\}$  is a basis for $H^0(X,E)$  , then up to a choice of homogeneous coordinates, $\phi =[s^0,...,s^N]$  . Thus the fiber of $K_{\phi }$  over $x$  consists of all those vectors $\xi$  such that $$d{f}^j\left(x\right)\xi =0,1\le j\le N,$$  where $f^j=s^j/s^0$  and the basis is chosen so that $s^0\left(x\right)\ne 0$  . This is equivalent to the slightly more geometric statement that, with $s=(s^1,...,s^N)$  , there exists $\lambda \in \mathbb{C}$  such that $$\partial s\left(x\right)\xi =\lambda s\left(x\right).$$  Recall that we write $e^{-\phi }$  for the Hermitian metric on $E$  induced by $R$  .

Definition 7.4. The free G-metric $R\in {\mathfrak{P}}_2(X,E)$  is said to be critically convex if outside an analytic subvariety of $X$  , $$ker\left(\sqrt{-1}\partial \overline{\partial }\phi \right)\subset K_{\phi }.$$ 

That is to say, if $\sqrt{-1}\partial \overline{\partial }\phi (\xi ,\xi )=0$  then $\left(D\phi \right)\xi =0$  .

Remark. Note that the freeness hypothesis can be dropped for the purpose of this definition. Indeed, if $R$  is not free then the map $\phi$  is only defined off an analytic subvariety of codimension 2 in $X$  .

Definition 7.5. A WGCS metric is a G-metric $R\in {\mathfrak{P}}_1(X,E)$  that satisfies ( $\Sigma$  1) and ( $\Sigma$  2) of Definition  6.2 , and also has the following property.

• (K) $R$  is critically convex.

We write $R\in {\mathfrak{P}}_2^{♭}(X,E)$  .

With this definition, we have the following result.

Theorem 5. If $R\in {\mathfrak{P}}_2^{♭}(X,E)$  then there is an integer $N\ge 0$  such that for all $m\ge N$  , $$R^m\in {\mathfrak{P}}_{\infty }(X,E^m).$$ 

The necessity of basic zeros

The methods already introduced allow us to show that in fact, any G-metric that is a root of a sum of squared norms can have only basic zeros.

Proposition 7.6. If $R$  is a G-metric such that $R^m$  is a sum of squared norms for some non-negative integer $m$  then $R$  has only basic zeros.

Proof. Without loss of generality, we may assume that the metric $R$  in question is free. Indeed, as we have now shown several times, after resolution of singularities we can peel off the part of the resolved G-metric that vanishes on a divisor.

Suppose now that $R\in {\mathfrak{P}}_1(X,E)$  is a free metric such that some power of $R$  is a sum of squared norms. To arrive at a contradiction, suppose $R$  has a zero $x\in {\mathcal{Z}}_R$  . Let $$M\stackrel{\iota }{\hookrightarrow }X$$  be a Riemann surface in $X$  passing through $x$  . Write $S={\iota }^{*}R$  . Then as before $$x\in {\mathcal{Z}}_S,{\iota }^{*}{V}_R=V_{S}and{V}_S\text{is free.}$$  Moreover, there is an integer $m\ge 0$  such that $$S^m\in {\mathfrak{P}}_{\infty }(M,m({\iota }^{*}E\left)\right).$$  But it follows that the zero locus of $S^m$  is a divisor. By taking roots, we see that the zero locus $D$  of $S$  is a $\mathbb{Q}$  -divisor.

We claim that $D$  is a $\mathbb{Z}$  -divisor. Indeed, suppose some point $p$  is in the support of $D$  . Then $S\left(p\right)=0$  . Choose a local coordinate $z$  on $M$  vanishing at $p$  . We claim that since $S$  is non-negative, we must have $$S\left(z\right)=a\left|z{|}^{2k}\right(P(z,\overline{z}))$$  for some real-analytic function $P(z,\overline{z})$  that does not vanish at the origin. Indeed, if the lowest order terms contain a pure factor of $z$  or $\overline{z}$  , then $S$  must have both signs. Thus the multiplicity of $D$  at $p$  is an integer, as claimed.

Since $S$  has an analytic zero locus $D$  , we can write $$S=\tilde{S}\otimes |s_D{|}^2,$$  so $V_S$  is not free. This is the desired contradiction. □

Critical metrics

As we will show momentarily, the map $\phi$  defined above naturally pushes the metric $R$  onto a G-metric on $\phi \left(X\right)$  . The trouble is that $\phi \left(X\right)$  may be singular, and we do not have a notion of G-metrics for this setting. Even though such a definition can be given using sections of sheaves instead of line bundles, our methods of Section  3 for G-metrics on line bundles do not extend to G-metrics on sheaves.

The fortunate fact is that our objects are smooth on a Zariski-open set. We will apply resolution of singularities to obtain a G-metric on a smooth space. The construction is non-canonical, but nevertheless we will refer to the metric obtained after such a resolution as a critical metric induced by $R$  . Let us be more specific.

First note that the G-metric $R\in {\mathfrak{P}}_1(X,E)$  naturally induces a G-metric in ${\mathfrak{P}}_1(\mathbb{P}(V_R^{*}),\mathbb{H})$  , which we denote $R_{\phi }$  . To be explicit, if we fix a basis $\left\{s^{\alpha }\right\}$  of $H^0(X,E)$  , then the elements of this basis are linear functionals on $V_R^{*}$  , and thus $$R_{\phi }=c_{\alpha \overline{\beta }}{s}^{\alpha }{\overline{s}}^{\beta }$$  defines a G-metric in ${\mathfrak{P}}_1(\mathbb{P}(V_R^{*}),\mathbb{H})$  .

Let ${\mathcal{O}}_{\phi \left(X\right)}\left(1\right)$  be the restriction of ${\mathcal{O}}_{\mathbb{P}\left(V_R^{*}\right)}(\mathbb{H})$  to $\phi \left(X\right)$  . Consider a resolution of singularities (as in Theorem  1.1 ) $\lambda :Y\to \phi \left(X\right)$  . Let $${\mathcal{O}}_Y(\Lambda ):={\lambda }^{*}{\mathcal{O}}_X\left(1\right)|Y.$$  Then ${\lambda }^{*}{R}_{\phi }\in {\mathfrak{P}}_1(Y,\Lambda ).$  Another way to construct the line bundle $\Lambda$  is as follows. $\mathbb{H}$  is the line bundle corresponding to a hyperplane. One intersects $\phi \left(X\right)$  with this hyperplane to obtain a divisor $D$  on $\phi \left(X\right)$  , and takes $\Lambda$  to be the line bundle associated to the divisor ${\lambda }^{*}D$  in $Y$  .

We will write $$\hat{R}:={\lambda }^{*}{R}_{\phi },$$  even though the metric $\hat{R}$  depends on the resolution. Every such metric $\hat{R}$  will be called a critical metric induced by $R$  , and we use the notation $$Y=\hat{X}\text{and}\Lambda =\hat{E}.$$ 

Lemma 7.7. If $R\in {\mathfrak{P}}_2^{♭}(X,E)$  then $\hat{R}\in {\mathfrak{P}}_2^{\Sigma }(\hat{X},\hat{E}).$ 

Proof. Since $R$  has only basic zeros, the same is true for any resolution of $R$  , and thus property $(\Sigma 1)$  holds. Property $(\Sigma 2)$  holds trivially. It remains only to show that property $(\Sigma 3)$  holds.

By the critical convexity hypothesis, we see that when restricted to $\phi \left(X\right)$  , the $(1,1)$  -form $\sqrt{-1}\partial \overline{\partial }logr(x,x)$  is non-degenerate on the complement of an analytic subvariety of $\phi \left(X\right)$  . Thus the same is true of ${\mu }^{*}R$  on $Y$  . □ Although $\phi :X\to \phi \left(X\right)$  is in general not an isomorphism, the G-metric $R$  is completely determined by its image under $\phi$  . It is in this sense that there are additional degrees of freedom that are not needed in the definition of $R$  . Simply put, we have $${\phi }^{*}{R}_{\phi }=R.$$  Now, $R_{\phi }$  is isomorphic to $\hat{R}$  away from the singular set of $\phi \left(X\right)$  , and thus if we show that $\hat{R}$  is a root of a sum of squared norms, then the same is true of $R$  . Indeed, suppose there are holomorphic sections $f_1,...,f_N$  of the appropriate line bundle such that $\hat{R}=\sum |f_j{|}^2$  . Then $$R=\sum \left|{\phi }^{*}\right({\lambda }^{-1}{)}^{*}{f}_j{|}^2$$  holds away from an analytic subset. But as the sections are bounded (since this is so on $\hat{X}$  ), they extend across the analytic subset by Riemann's Removable Singularities Theorem. Thus $R$  can be written as a sum of squared norms. Of course, a similar argument holds if we replace $R$  by some power of $R$  .

Proof of Theorem  5 : Let $\hat{R}\in {\mathfrak{P}}_2^{\Sigma }(\hat{X},\hat{E})$  be a critical metric induced by $R$  . (The fact that $\hat{R}$  is a $\Sigma$  GCS metric is part of the hypothesis.) Now apply Theorem  4 to $\hat{R}$  and the G-metric $P\equiv 1$  in ${\mathfrak{P}}_1(\hat{X},\mathcal{O})$  to see that $${\hat{R}}^m\in {\mathfrak{P}}_{\infty }(\hat{X},{\hat{E}}^m).$$  As we just discussed, this means that $R^m\in {\mathfrak{P}}_{\infty }(X,E^m)$  . □

8 Proof of Theorem  2 

Suppose given a G-metric $P\in {\mathfrak{P}}_1(X,E)$  with only basic zeros. We begin by reducing to the case where $P$  is a free metric. This reduction is, by this point in the paper, standard: we find a resolution of singularities $\mu :\tilde{X}\to X$  such that ${\mu }^{*}{V}_P=V_{\tilde{P}}\otimes s_D$  for some free metric $\tilde{P}$  . Evidently $${\mu }^{*}\mathit{\forall }P\mathit{\forall }=\mathit{\forall }\tilde{P}\mathit{\forall }\otimes |s_D{|}^2,\frac{{\mu }^{*}\mathit{\forall }P\mathit{\forall }}{{\mu }^{*}P}=\frac{\mathit{\forall }\tilde{P}\mathit{\forall }}{\tilde{P}},$$  and $\mathit{\forall }\tilde{P}{\mathit{\forall }}^m\tilde{P}$  is a sum of squared norms if and only if this is the case for $\mathit{\forall }P{\mathit{\forall }}^{m}P$  .

From here on we assume, without loss of generality, that $P$  is a free G-metric, i.e., $P\in {\mathfrak{P}}_1^{♯}(X,E)$  .

We begin with the following proposition.

Proposition 8.1. If $P\in {\mathfrak{P}}_1^{♯}(X,E)$  then $\mathit{\forall }P\mathit{\forall }$  is critically convex.

Proof. Let $\mathcal{A}$  be the set of points $x\in X$  such that for some $\xi \in T_{X,x}$  , $$(\sqrt{-1}\partial \overline{\partial }log\mathit{\forall }P\mathit{\forall })(\xi ,\xi )=0.$$  Let us choose a distinguished basis $s^1,...,s^k$  for $V_P$  , i.e., a basis such that, with $s=(s^1,...,s^k)$  denoting the k-tuple of this basis of global sections, $$\mathit{\forall }P\mathit{\forall }=s\cdot \overline{s}.$$  Then, up to projective isomorphism, $\phi =\left[s\right]$  .

We compute that $$\sqrt{-1}\partial \overline{\partial }log\mathit{\forall }P\mathit{\forall }=\frac{\sqrt{-1}\partial s\wedge \overline{\partial s}}{\mathit{\forall }P\mathit{\forall }}-\frac{\sqrt{-1}(\overline{s}\cdot \partial s)\wedge (s\cdot \overline{\partial s})}{\mathit{\forall }P{\mathit{\forall }}^2}$$  The semi-positivity of this $(1,1)$  -form follows from the Cauchy-Schwartz inequality, and in particular we see that $x\in \mathcal{A}$  if and only if there is a $\xi \in T_{X,x}$  such that for some $\lambda \in {\mathbb{C}}^{*}$  , $$\partial s\left(x\right)\xi =\lambda s\left(x\right).$$  But this condition means exactly that we are in the critical locus of $\phi$  , which is what we were trying to show. The proof is complete. □ Proof of Theorem  2 : By Lemma  7.7 the critical convexity of $\mathit{\forall }P\mathit{\forall }$  implies that $$\mathit{\forall }\hat{P}\mathit{\forall }\in {\mathfrak{P}}_2^{\Sigma }(\hat{X},\hat{E}).$$  The proof of Theorem  2 now follows easily from Proposition  8.1 . Indeed, Theorem  4 applies with the metrics $\hat{P}\in {\mathfrak{P}}_1(\hat{X},\hat{E})$  and $\mathit{\forall }\hat{P}\mathit{\forall }\in {\mathfrak{P}}_2^{\Sigma }(\hat{X},,\hat{E})$  , and then we may invoke the discussion preceding the proof of Theorem  5 . □ References

1. Berman, R., Bergman kernels and local holomorphic Morse inequalities. Math. Z. 248 (2004), no. 2, 325–344.
2. Abramowitz, D., De Jong, A. J., Smoothness, semistability, and toroidal geometry. J. Algebraic Geom. 6 (1997), no. 4, 789–801.
3. Bogomolov, F., Pantev, T., Weak Hironaka theorem. Math. Res. Lett. 3 (1996), no. 3, 299–307.
4. Boutet de Monvel, Sjöstrand, J., Sur la singularité des noyaux de Bergman et de Szegö. (French) Journées: Équations aux Dérivées Partielles de Rennes (1975), pp. 123–164. Asterisque, No. 34-35, Soc. Math. France, Paris, 1976.
5. Catlin, D., The Bergman kernel and a theorem of Tian. Analysis and geometry in several complex variables (Katata, 1997), 1–23, Trends Math., Birkhäuser Boston, Boston, MA, 1999.
6. Catlin, D., D'Angelo, J., A stabilization theorem for Hermitian forms and applications to holomorphic mappings. Math. Res. Lett. 3 (1996), no. 2, 149–166.
7. Catlin, D., D'Angelo, J., An isometric embedding theorem for holomorphic bundles. Math. Res. Lett. 6 (1999) no. 1, 43–60.
8. D'Angelo, J., Inequalities from complex analysis. Carus Mathematical Monographs, 28. Mathematical Association of America, Washington, DC, 2002.
9. D'Angelo, J., Bordered complex Hessians. J. Geom. Anal. 11 (2001), no. 4, 561–571.
10. D'Angelo, J., Complex Variables Analogues of Hilbert's Seventeenth Problem To appear in Internat. J. Math.
11. D'Angelo, J., Varolin, D., Positivity conditions for Hermitian symmetric functions. Asian J. Math, 7, No. 4 (2003) 1-18.
12. Fefferman, C., The Bergman kernel and biholomorphic mappings of pseudoconvex domains. Invent. Math. 26 (1974), 1–65.
13. Hironaka, H., Resolution of singularities of an algebraic variety over a field of characteristic zero. I, II. Ann. of Math. (2) 79 (1964), 109-203, 205-326.
14. Lazarsfeld, R., Positivity in Algebraic Geometry I and II Springer 2004.
15. Paranjape, K., The Bogomolov–Pantev resolution, an expository account. New Trends in Algebraic Geometry (ed. K. Hulek, F. Catanese, C. Peters, M. Reid), pages 347-358, Cambridge University Press 1999. (See also math.AG/9806084 in the mathematics ArXiV at http://front.math.ucdavis.edu/)
16. Quillen, D., On the representation of Hermitian forms as sums of squares. Invent. Math. 5 (1968) 237–242.
17. Zelditch, S., Szegö kernels and a theorem of Tian. Internat. Math. Res. Notices 1998, no. 6, 317–331.