This opening section sets the course in place. Earlier parts of several complex variables developed local holomorphic functions, pseudoconvexity, Stein manifolds, Dolbeault cohomology, and analytic estimates for the $\bar\partial$-operator. The present course turns those tools toward complex geometry: Hermitian metrics, curvature, positivity, and the way analytic estimates produce global geometric consequences. The guiding theme is that curvature packages local differential information in a form strong enough to control sections, cohomology, and projective embeddings. The working prerequisites are one-variable complex analysis, smooth manifolds and differential forms, basic sheaf cohomology, Stein manifolds, [Dolbeault cohomology](/page/Dolbeault%20Cohomology), and $L^2$ estimates for $\bar\partial$. Here sheaf cohomology means the groups $H^q(X,\mathcal F)$ attached to a sheaf $\mathcal F$ on a complex manifold $X$, while the Dolbeault complex is the complex of $(p,q)$-forms with differential $\bar\partial$. Later names such as Nakano positivity, Griffiths positivity, and the Bochner-Kodaira-Nakano identity will be introduced where they are used; at this point they should be read as labels for ways curvature controls $\bar\partial$-estimates. Section 0 is not a replacement for those topics; it fixes the language and gives a map of the questions that the later lectures answer. **From Several Complex Variables to Complex Geometry.** What changes when the local theory of holomorphic functions is placed on a compact complex manifold or a holomorphic vector bundle? In domains in $\mathbb C^n$, many questions are phrased in terms of functions, weights, and the $\bar\partial$-equation. On a complex manifold, the same analytic operators act on forms with values in bundles, and their solvability is influenced by curvature. [explanation: Analytic Data Becomes Geometric Data] A plurisubharmonic function on a domain is a local object, measured by its complex Hessian. A Hermitian metric on a holomorphic line bundle is a global object whose local weights change when the frame changes. Curvature is the expression that survives these changes of frame and therefore turns local convexity into an invariant geometric condition. This course uses that passage repeatedly. Local weights give estimates for $\bar\partial$; estimates give vanishing or extension statements; global sections obtained from those statements give maps to projective space. [/explanation] The point of this shift is not to abandon the analytic perspective. It is to make the analytic perspective coordinate-free enough to handle compact complex manifolds and holomorphic bundles. [example: Local Weight as Curvature] Let $X \subset \mathbb{C}^n$ be open and let $L = X \times \mathbb{C}$ be the product holomorphic line bundle with global holomorphic frame $e$. Give $L$ the Hermitian metric $h(e,e) = e^{-\varphi}$ for a smooth real-valued function $\varphi : X \to \mathbb{R}$. In this trivialization, the local curvature formula gives \begin{align*} \Theta_h=\partial\bar\partial\varphi. \end{align*} Thus the familiar Levi form of the weight is exactly the curvature form of the metric. Positivity of $(L,h)$ asks that $i\Theta_h=i\partial\bar\partial\varphi$ be positive definite, which is the strict plurisubharmonicity condition for $\varphi$. The point is not the calculation itself, but the invariant interpretation: a local convexity condition for a weight becomes a global positivity condition for a line bundle. [/example] This example is the local model behind the entire course: positivity of a line bundle is the global version of having a strictly plurisubharmonic potential. **The Objects Being Studied.** Which geometric objects carry the analytic information from the earlier theory? We first need the ambient notion of a complex manifold, since holomorphic functions, Dolbeault forms, and curvature all require charts whose coordinate changes preserve complex differentiability. [definition: Complex Manifold] A complex manifold of complex dimension $n$ is a smooth manifold $X$ equipped with an atlas of charts $\varphi:U\to \varphi(U)\subseteq\mathbb C^n$ whose transition maps are holomorphic. [/definition] The charts allow holomorphic functions, differential forms of type $(p,q)$, and the Dolbeault operator $\bar\partial$ to be defined intrinsically. The real dimension of $X$ is $2n$, but the complex structure remembers which directions are holomorphic and which are antiholomorphic. To put analytic estimates on vector-valued holomorphic data, one also needs bundles whose fibers have smoothly varying lengths. This motivates the combined object used throughout the curvature theory. [definition: Hermitian Holomorphic Vector Bundle] A Hermitian holomorphic vector bundle over a complex manifold $X$ is a pair $(E,h)$ consisting of a holomorphic vector bundle $E\to X$ and a smooth Hermitian inner product $h_x$ on each fibre $E_x$. [/definition] The metric $h$ lets us measure sections and forms, while the holomorphic structure supplies the operator $\bar\partial_E$ on $E$-valued forms. The first major construction of the course is the Chern connection, which is the connection compatible with both structures. Once that compatible connection is fixed, the next invariant to isolate is its failure to square to zero. This is the curvature quantity whose sign will later encode positivity. [definition: Chern Curvature] Let $(E,h)$ be a Hermitian holomorphic vector bundle over a complex manifold $X$, and let $\nabla^{\mathrm{Ch}}$ denote its Chern connection, namely the unique connection on $E$ that is compatible with $h$ and whose $(0,1)$-part equals $\bar\partial_E$. The Chern curvature of $(E,h)$ is the section \begin{align*} \Theta_h \in \Gamma\bigl(X,\, \Lambda^{1,1}T^*X \otimes \operatorname{End}(E)\bigr) \end{align*} defined by $\Theta_h = (\nabla^{\mathrm{Ch}})^2$. [/definition] For a line bundle, $\operatorname{End}(L)$ is canonically a rank-one scalar bundle, so the curvature may be treated as an ordinary $(1,1)$-form after choosing the standard scalar identification. For higher-rank bundles, the curvature is matrix-valued in a local holomorphic frame, and its positivity has several inequivalent meanings. The basic computation is therefore the rank-one local model: express a Hermitian metric by a weight and identify exactly which differential form represents its curvature. This is the first place where the abstract definition becomes usable: a metric on a line bundle is locally just a positive function, but curvature must be independent of the chosen holomorphic frame. The point of the calculation is to see which second-derivative expression survives a change of frame and therefore gives the global curvature form. This local formula is the bridge from the invariant definition of curvature to practical positivity tests. Before positivity can be checked on examples, one needs a frame-level expression whose transformation law is harmless and whose sign can be read from a complex Hessian. [quotetheorem:3835] [citeproof:3835] This calculation explains why the sign convention matters. The assumption that $\varphi$ is real-valued is part of the Hermitian metric: $h(e,e)=e^{-\varphi}$ must be positive real, while a complex-valued weight would not define a Hermitian length. Smoothness is also being used, since $\partial\bar\partial\varphi$ is a classical $(1,1)$-form; for merely continuous weights, curvature has to be interpreted in the weaker sense of currents, which is not the convention in these opening notes. The formula is useful precisely because it gives a frame-level way to test positivity. The weight itself is local data, but the curvature form is the geometric object carried forward into the positivity theory. **Positivity as the Organising Principle.** How can a differential inequality on local weights force a global manifold to sit inside projective space? We need a line-bundle positivity condition that is global but still checkable by local Hessians, because that condition is what converts analytic estimates into holomorphic sections and projective maps. [definition: Positive Hermitian Line Bundle] Let $L\to X$ be a holomorphic line bundle over a complex manifold $X$, and let $h$ be a Hermitian metric on $L$. The Hermitian line bundle $(L,h)$ is positive if the real $(1,1)$-form $i\Theta_h$ is positive definite at every point of $X$. [/definition] In a local holomorphic frame with $h(e,e)=e^{-\varphi}$, this asks for $i\partial\bar\partial\varphi$ to be positive definite. Thus the definition globalises strict plurisubharmonicity, but it is phrased in a way that is independent of the chosen frame. The strict word matters: semipositive curvature can be too weak to create any projective coordinates. [example: Flat Line Bundle Has Too Few Sections] Let $X$ be a compact connected complex manifold with more than one point, and equip $\mathcal{O}_X = X \times \mathbb{C}$ with the flat Hermitian metric $h(1, 1) = 1$, where $1$ denotes the tautological global holomorphic frame. We show that $(\mathcal{O}_X, h)$ is semipositive but not positive, and that no tensor power can produce a projective embedding. The metric value $h(1,1) = 1 = e^{-0}$ corresponds to the weight function $\varphi \equiv 0$. By [Local Formula for Curvature](/theorems/1540), the Chern curvature in the frame $1$ is \begin{align*} \Theta_h = \partial\bar\partial\varphi = \partial\bar\partial(0) = 0, \end{align*} since the partial derivatives of a constant function vanish identically. Thus $i\Theta_h = 0$, which is nonnegative (the zero form satisfies $\langle i\Theta_h \,v, v\rangle = 0 \geq 0$ for every tangent vector $v$) but is not positive definite at any point of $X$; the bundle is semipositive but not positive. For the tensor powers: since the fibre of $\mathcal{O}_X$ is one-dimensional, $\mathcal{O}_X^{\otimes m} \cong \mathcal{O}_X$ for every $m \geq 1$ — tensoring the product line bundle $X \times \mathbb{C}$ with itself $m$ times yields $X \times \mathbb{C}$ again, as $\mathbb{C}^{\otimes m} \cong \mathbb{C}$ by the universal property of the tensor product. Therefore $H^0(X, \mathcal{O}_X^{\otimes m}) = H^0(X, \mathcal{O}_X)$, the space of global holomorphic functions $f \colon X \to \mathbb{C}$. If any such $f$ were non-constant, the [Open Mapping Theorem for Holomorphic Functions](/theorems/358) would make $f$ an open map, so $f(X)$ would be open in $\mathbb{C}$; but $f(X)$ is also compact (the continuous image of the compact space $X$), hence closed, forcing $f(X) = \mathbb{C}$ — a contradiction since $\mathbb{C}$ is not compact. Therefore every global holomorphic function on $X$ is constant, and $H^0(X, \mathcal{O}_X^{\otimes m}) \cong \mathbb{C}$ for all $m \geq 1$. With the space of sections one-dimensional, its projectivisation $\mathbb{P}(H^0(X, \mathcal{O}_X^{\otimes m})^*) = \mathbb{P}^0$ is a single point, and the associated linear system sends every point of $X$ to that same target. No tensor power of $\mathcal{O}_X$ can separate the distinct points of $X$, illustrating exactly where the Kodaira embedding argument breaks down: without a strictly positive curvature form, no analytic estimate forces new holomorphic sections into existence. [/example] This failure is the local warning sign for the global theory: positivity is not decoration, but the condition that supplies enough curvature to force many holomorphic sections. [example: The Hyperplane Bundle] On $\mathbb{P}^n$ with homogeneous coordinates $[Z_0 : Z_1 : \cdots : Z_n]$, the hyperplane bundle $\mathcal{O}_{\mathbb{P}^n}(1)$ is the dual of the tautological line bundle $\mathcal{O}_{\mathbb{P}^n}(-1)$, whose fibre at $[Z]$ is the line $\mathbb{C}Z \subset \mathbb{C}^{n+1}$. The standard Hermitian metric on $\mathbb{C}^{n+1}$ induces a metric on $\mathcal{O}(-1)$ and hence a dual metric $h$ on $\mathcal{O}(1)$. On the affine chart $U_0=\{Z_0\ne 0\}$, with coordinates $z_j=Z_j/Z_0$, the induced local weight for $\mathcal{O}(1)$ is \begin{align*} \varphi=\log(1+|z_1|^2+\cdots+|z_n|^2). \end{align*} The curvature is therefore represented locally by $\partial\bar\partial\varphi$, and $i\partial\bar\partial\varphi$ is the unnormalized Fubini-Study Kähler form $\Omega_{\mathrm{FS}}$. Its positivity says that $\mathcal{O}_{\mathbb{P}^n}(1)$ is the basic positive line bundle on projective space. With the normalization $\omega_{\mathrm{FS}}=\frac{1}{2\pi}\Omega_{\mathrm{FS}}$, this form represents $c_1(\mathcal{O}(1))$. This example supplies the model for the whole embedding theory: sections of powers of $\mathcal{O}(1)$ are homogeneous polynomials, and the positive curvature of the Fubini-Study metric is the differential-geometric shadow of the projective coordinates. [/example] The example is the model for projective embeddings. If a compact complex manifold has a sufficiently positive line bundle, its holomorphic sections behave like homogeneous coordinates. The central global question is whether positivity alone can produce enough sections to define an embedding into projective space. Kodaira's theorem gives the affirmative answer in the compact case. [quotetheorem:3836] [citeproof:3836] This theorem is one of the main destinations of the course, so Section 0 records only the structure of its argument. The details require the curvature identities and vanishing tools developed in the later lectures. Each hypothesis has a role. Compactness is used to make $H^0(X,L^m)$ finite-dimensional and to turn an injective holomorphic immersion into a holomorphic embedding. Without compactness, the same local separation arguments do not by themselves give a proper map into a finite-dimensional projective space. The phrase "sufficiently large" is also unavoidable. Replacing $L$ by $L^m$ multiplies the curvature term by $m$, so the positive part of the $L^2$ estimate can dominate the cutoff errors and first-order jet constraints needed to separate tangent directions. If $L$ were only semipositive, there would be no uniform positive lower bound to dominate those errors; the flat $\mathcal O_X$ above is the simplest case where the available sections cannot separate points. [illustration:scv-iv-kodaira-separation] **Analytic Inputs and Global Outputs.** What analytic estimates are strong enough to produce global geometry? The course repeatedly uses the same pattern: formulate a geometric objective as a problem about solving $\bar\partial u=f$, choose a metric whose curvature has the needed sign, and apply an $L^2$ estimate to solve the equation with control. [explanation: The Basic Analytic Strategy] Suppose we want a holomorphic section with a prescribed jet at a point. We first build a smooth section with the desired local behaviour and then measure its failure to be holomorphic by applying $\bar\partial$. If an $L^2$ estimate solves $\bar\partial u=f$ with controlled norm, subtracting $u$ corrects the smooth section to a holomorphic one while preserving the desired local information. Curvature enters because the estimate contains curvature terms. Positive curvature supplies coercivity; negative curvature creates obstructions; semipositive curvature often requires additional limiting or approximation arguments. [/explanation] This analytic strategy links the early several complex variables material to the later complex-geometric theorems. It also explains why the same objects recur across the course: Hermitian metrics, curvature forms, weights, and Dolbeault cohomology. [example: Separating Two Points] Let $X$ be a compact complex manifold, let $(L,h)$ be a positive Hermitian holomorphic line bundle over $X$, and let $p,q\in X$ be distinct points. Point separation asks for a section of $L^m$ that vanishes at $p$ but not at $q$. The analytic strategy has three conceptual parts. First, choose a smooth local model that has the desired values near $p$ and $q$. Second, measure its failure to be holomorphic by applying $\bar\partial$. Third, use positivity of $L^m$ to solve away that $\bar\partial$-error without destroying the prescribed local behaviour. The role of the exponent $m$ is quantitative: the curvature of $L^m$ is $m$ times the curvature of $L$, so large tensor powers give a stronger estimate. Singular weights are the device that records the required vanishing at selected points. The precise proof belongs to the later jet-separation theorem; this example is meant to show why the geometric statement is naturally an $L^2$ $\bar\partial$ problem rather than a purely algebraic counting argument. [/example] This is not yet the full embedding theorem, but it shows the shape of the argument. Separating tangent directions follows the same template with first-order jets replacing values at points: instead of prescribing $s(p)=0$ and $s(q)\ne 0$, one prescribes the first-order Taylor jet of $s$ at a single point $p$ and controls the $\bar\partial$-error so that the corrected section realises that jet. The same template extends to higher-order jets, which is what is needed to verify that the constructed map is an immersion at every point and, more generally, to control any prescribed finite-order local data. The mechanism that makes this work uniformly across $X$ is precisely the uniform strict positivity of $i\Theta_h$: by passing to $L^m$, the positive curvature term in the $L^2$ estimate becomes large enough to dominate the cutoff and jet-constraint errors with constants that do not depend on the chosen point. Without uniform positivity — for instance, if the curvature were merely semipositive or only positive on an open subset — there would be no global lower bound to control the error simultaneously for every choice of $p$ and $q$, and the argument would collapse on the locus where positivity degenerates. **The Course Roadmap.** What should the reader expect each later section to contribute to the final geometric results? The course is organised so that each block supplies one part of the bridge from local curvature to global projective geometry. [explanation: Twelve-Lecture Trajectory] The first block introduces Hermitian vector bundles, Chern connections, curvature forms, and the local formula $\Theta_h=\bar\partial(h^{-1}\partial h)$. The second block studies positivity for line bundles, including the relation with strictly plurisubharmonic local weights and the role of ample and very ample bundles. The middle lectures develop Kähler geometry and the identities that make the $\bar\partial$-Laplacian interact well with curvature. The $\bar\partial$-Laplacian is the second-order operator measuring harmonicity in the Dolbeault complex. The Bochner-Kodaira-Nakano identities are curvature formulas for this operator on bundle-valued forms; they are the analytic engine behind embedding and Lefschetz-type results. The final lectures apply the machinery to Kodaira embedding, hyperplane sections, and geometric consequences of positivity. By the end, the reader should be able to trace how a local curvature inequality becomes a statement about the global algebraic and topological structure of a compact complex manifold. [/explanation] The roadmap also clarifies the role of examples. Product line bundles with weighted metrics teach the local calculations; projective space and its tautological bundles provide the main test case; compact Kähler manifolds provide the natural setting for the deepest results. **Conventions Used in These Notes.** Which conventions must be fixed before curvature computations begin? We need the first Chern form normalization before any positivity statement can be compared with integral cohomology classes, because different books place the signs and factors of $2\pi$ in different locations. [definition: First Chern Form] Let $(L,h)$ be a Hermitian holomorphic line bundle over a complex manifold $X$. The first Chern form of $(L,h)$ is \begin{align*} c_1(L,h)=\frac{i}{2\pi}\Theta_h. \end{align*} [/definition] This convention makes the curvature of a positive line bundle represent a positive real $(1,1)$-form after multiplication by $i$. It is also the convention compatible with the Fubini-Study normalisation used for $\mathcal O_{\mathbb P^n}(1)$ in these notes. [remark: Notational Conventions] Complex manifolds are denoted by $X$ or $Y$, holomorphic vector bundles by $E\to X$, and holomorphic line bundles by $L\to X$. A local holomorphic frame is usually denoted by $e$ in the line bundle case and by $(e_1,\dots,e_r)$ in rank $r$. The Dolbeault operator is written $\bar\partial$, and curvature is written $\Theta_h$ when the metric is part of the notation. [/remark] These conventions are chosen to keep analytic and geometric formulae aligned. When a formula depends on a local frame, the surrounding text will state the frame; when a statement is invariant, it will be phrased without reference to coordinates or frames. With the overview and conventions fixed, we now begin the local geometry needed later for positivity and vanishing. The first task is to understand how a holomorphic vector bundle can carry a smooth Hermitian metric, and how the tension between those two structures produces curvature. # 1. Hermitian Vector Bundles and Chern Connections Complex geometry studies holomorphic objects together with smooth metric data. The holomorphic structure controls which local sections count as analytic, while the Hermitian metric measures their size and turns analytic variation into curvature. This section sets up the local language for that bridge: frames, transition functions, Chern connections, curvature matrices, and the first Chern form of a line bundle. Later positivity results will read curvature inequalities from the formulas developed here. ## Holomorphic Frames and Hermitian Metrics How can a vector bundle be locally a product while still carrying global holomorphic information? The point of a holomorphic vector bundle is that local descriptions are glued by holomorphic change-of-frame matrices. All later formulas are local, so the transition behavior is part of the data rather than a bookkeeping detail. [definition: Holomorphic Vector Bundle] Let $X$ be a complex manifold. A holomorphic vector bundle of rank $r$ over $X$ is a complex manifold $E$ with a holomorphic map $\pi:E\to X$ such that each fiber $E_x=\pi^{-1}(x)$ is a complex vector space of dimension $r$, and such that there is an open cover $(U_i)$ of $X$ with biholomorphisms \begin{align*} \Phi_i:\pi^{-1}(U_i)\to U_i\times \mathbb C^r \end{align*} whose restrictions to fibers are complex-linear isomorphisms. [/definition] A choice of local coordinates in the fibers gives local sections that serve as a basis at every point. These sections are the moving bases in which metrics, connections, and curvature become matrices. To make later formulas precise, the page needs a name for such a moving holomorphic basis on one open set. A frame is not extra global structure on the bundle; it is a temporary local coordinate system for the fibers. Naming it now lets later formulas distinguish invariant bundle statements from the matrices and coefficient functions that appear only after a local choice has been made. [definition: Local Holomorphic Frame] Let $E\to X$ be a holomorphic vector bundle of rank $r$. A local holomorphic frame on an open set $U\subset X$ is an ordered $r$-tuple $e=(e_1,\dots,e_r)$ of holomorphic sections of $E|_U$ such that $(e_1(x),\dots,e_r(x))$ is a basis of $E_x$ for every $x\in U$. [/definition] If $e$ is a frame, every local section has a unique expression $s=e a$, where $a:U\to\mathbb C^r$ is a column vector of coefficient functions. A frame change is therefore encoded by an invertible matrix of holomorphic functions. The matrices relating overlapping frames are the data that carry the bundle's global twisting, so they need to be recorded as part of the local language. Without them, the local product charts would look unrelated and there would be no rule for deciding when local sections, metrics, or connection matrices describe the same global object. The next definition isolates these overlap matrices and the consistency condition they must satisfy. [definition: Transition Function] Let $e_i$ and $e_j$ be local holomorphic frames for $E$ over $U_i$ and $U_j$. On $U_i\cap U_j$, the transition function $g_{ij}:U_i\cap U_j\to GL(r,\mathbb C)$ is determined by \begin{align*} e_j=e_i g_{ij}. \end{align*} The transition functions satisfy \begin{align*} g_{ii}=I_r,\qquad g_{ij}g_{jk}=g_{ik} \end{align*} on triple overlaps. [/definition] The cocycle identities say that changing frames in two steps gives the same result as changing frames in one step. This is the local form of the global bundle, and it is also the origin of the gauge transformation laws for connections and curvature. [example: Line Bundle From A Cocycle] Let $X = U_0 \cup U_1$, and let $g_{01}: U_0 \cap U_1 \to \mathbb{C}^*$ be holomorphic. A rank-one bundle $L$ is constructed by taking local holomorphic frames $e_0$ on $U_0$ and $e_1$ on $U_1$ subject to $e_1 = e_0 g_{01}$ on the overlap; we show that this single function completely governs the compatibility of local sections. On each open set $U_i$, every section of $L$ is written uniquely as $s = e_i f_i$ for a holomorphic function $f_i : U_i \to \mathbb{C}$. For $s$ to be globally well-defined, both local expressions must agree on $U_0 \cap U_1$: \begin{align*} e_0 f_0 = e_1 f_1. \end{align*} Substituting the gluing relation $e_1 = e_0 g_{01}$ into the right-hand side gives \begin{align*} e_0 f_0 = e_0 \bigl(g_{01} f_1\bigr). \end{align*} At each point $x \in U_0 \cap U_1$, the vector $e_0(x)$ is a basis element of the fiber $L_x$, hence nonzero; multiplying a nonzero vector by a scalar is injective, so equating coefficients yields \begin{align*} f_0 = g_{01} f_1 \quad \text{on } U_0 \cap U_1. \end{align*} Conversely, any pair of holomorphic functions $(f_0, f_1)$ satisfying this identity defines a holomorphic section of $L$: the two local expressions agree on the overlap by the same calculation in reverse. Applying the identical argument with $U_0$ and $U_1$ exchanged—writing $e_0 = e_1 g_{10}$ and equating coefficients—gives $g_{10} = g_{01}^{-1}$, the two-chart instance of the cocycle identity $g_{ij}g_{jk} = g_{ik}$ from the definition of transition functions. The scalar constraint $f_0 = g_{01} f_1$ is thus not merely a consistency check: global holomorphic sections of $L$ are exactly the compatible pairs $(f_0, f_1)$ satisfying it, and every holomorphic feature of $L$ — its sections, its tensor powers, its local triviality — is encoded in the single transition function $g_{01}$. [/example] Holomorphic data alone does not measure lengths or angles in the fibers. A Hermitian metric supplies this smooth measurement, and the Chern connection will be the connection determined by combining it with the holomorphic structure. Before curvature can be assigned a sign, the metric itself must be specified in an invariant way on every fiber. The local matrix of a metric changes when the frame changes, so the definition cannot be tied to one trivialization. What is intrinsic is the smoothly varying Hermitian inner product on each fiber, which later produces the unique Chern connection compatible with both length measurement and holomorphic structure. The next formal definition fixes this intrinsic object before any curvature calculation is attempted. It records both positivity on each fiber and smooth variation in the base, because both features are needed for lengths, adjoints, and Chern connections to be well defined. [definition: Hermitian Metric] Let $E\to X$ be a smooth complex vector bundle. A Hermitian metric $h$ on $E$ assigns to every $x\in X$ a positive-definite Hermitian form \begin{align*} h_x:E_x\times E_x\to\mathbb C \end{align*} depending smoothly on $x$. The convention in this section is that $h_x$ is conjugate-linear in the first entry and linear in the second entry. [/definition] In a local frame $e=(e_1,\dots,e_r)$, the metric is represented by the Hermitian matrix $h=(h_{\alpha\bar\beta})$ with $h_{\alpha\bar\beta}=h(e_\alpha,e_\beta)$. If $s=e a$, then \begin{align*} |s|_h^2=\overline{a}^{\top}h a. \end{align*} The matrix transforms by $\tilde h=g^*hg$ under a frame change $\tilde e=eg$, where $g^*=\overline{g}^{\top}$. [example: Product Line Bundle With Weight] Let $X$ be a complex manifold, $E = X \times \mathbb{C}$ the trivial bundle, $e$ its canonical global holomorphic frame, and $\varphi \in C^\infty(X;\mathbb{R})$ a smooth real weight. We show that the assignment $h(e,e) = e^{-\varphi}$ defines a Hermitian metric on $E$ and compute the pointwise norm of an arbitrary section explicitly. At each $x \in X$, extend $h$ to the full fiber by the sesquilinearity rule \begin{align*} h_x(ae, be) = \bar{a}\, e^{-\varphi(x)}\, b, \qquad a,b \in \mathbb{C}. \end{align*} This is conjugate-linear in the first argument and $\mathbb{C}$-linear in the second by construction. It is positive-definite because $e^{-\varphi(x)} > 0$ at every $x$, so $h_x(ae, ae) = |a|^2 e^{-\varphi(x)} = 0$ forces $a = 0$. Smooth dependence on $x$ follows from the smoothness of $\varphi$, so $h$ satisfies the definition of a Hermitian metric. In the global frame $e$, the metric matrix reduces to the scalar function $h = e^{-\varphi}$. Every smooth section of $E$ has the unique form $s = fe$ for some $f \in C^\infty(X;\mathbb{C})$. Applying the rank-one case of the formula $|s|_h^2 = \bar{a}^\top h\, a$ with scalar $a = f$ gives \begin{align*} |fe|_h^2 = h(fe, fe) = \bar{f}\, e^{-\varphi}\, f = |f|^2 e^{-\varphi}. \end{align*} The second equality uses $h_x(ae,be) = \bar{a}\,e^{-\varphi}\,b$ with $a = b = f$; the third is $\bar{f}f = |f|^2$. Because the bundle is globally trivial and the metric matrix is the scalar $e^{-\varphi}$, every downstream quantity on $(E,h)$ inherits this simplicity: the Chern connection matrix is $h^{-1}\partial h = e^{\varphi}\,\partial(e^{-\varphi}) = -\partial\varphi$, a computation involving only the chain rule on a scalar exponential, and all further structure — the curvature and the first Chern form — follows by applying $\bar\partial$ to this single scalar expression with no matrix inversion needed. [/example] ## The Chern Connection Given a holomorphic bundle with a Hermitian metric, which connection preserves both structures? A general smooth connection has too much freedom. The Chern connection is singled out by requiring its $(0,1)$ part to be the holomorphic structure and its full differential to preserve the Hermitian metric. [definition: Connection On A Complex Vector Bundle] Let $E\to X$ be a smooth complex vector bundle. A connection on $E$ is a $\mathbb C$-linear map \begin{align*} D:\mathcal A^0(X,E)\to\mathcal A^1(X,E) \end{align*} satisfying \begin{align*} D(fs)=df\otimes s+fDs \end{align*} for all $f\in C^\infty(X;\mathbb C)$ and all smooth sections $s$ of $E$. [/definition] On a complex manifold, complex-valued one-forms split into types $(1,0)$ and $(0,1)$. Thus any connection decomposes as $D=D^{1,0}+D^{0,1}$. A holomorphic vector bundle has a canonical Dolbeault operator, and the Chern connection is required to have this as its anti-holomorphic part. The operator that detects holomorphic sections is therefore the reference $(0,1)$ part against which compatible connections are measured. A smooth section of a holomorphic bundle is holomorphic exactly when its anti-holomorphic derivative vanishes in holomorphic frames. The definition packages this local test into a global first-order operator, so that the Chern connection can be characterized by having the correct $(0,1)$ component. To make that characterization precise, the local equation $\bar\partial a=0$ for holomorphic coefficients must be promoted to an operator independent of the chosen holomorphic frame. This is the object that will later be extended to bundle-valued forms and paired with the metric part of the Chern connection. [definition: Dolbeault Operator Of A Holomorphic Vector Bundle] Let $E\to X$ be a holomorphic vector bundle. The Dolbeault operator of $E$ is the $\mathbb C$-linear first-order differential operator \begin{align*} \bar\partial_E:\mathcal A^0(X,E)\to\mathcal A^{0,1}(X,E) \end{align*} which, in every local holomorphic frame $e$, satisfies \begin{align*} \bar\partial_E(ea)=e\,\bar\partial a \end{align*} for all smooth coefficient functions $a:U\to\mathbb C^r$. It extends to operators \begin{align*} \bar\partial_E:\mathcal A^{p,q}(X,E)\to\mathcal A^{p,q+1}(X,E) \end{align*} by the graded Leibniz rule. [/definition] A smooth section $s$ is holomorphic exactly when $\bar\partial_Es=0$. The definition of the Chern connection says that the connection must agree with this operator in type $(0,1)$ while its remaining part is determined by the metric. [definition: Chern Connection] Let $(E,h)$ be a Hermitian holomorphic vector bundle over a complex manifold $X$. A connection $D_h$ on $E$ is the Chern connection if \begin{align*} D_h^{0,1}=\bar\partial_E \end{align*} and \begin{align*} d\,h(s,t)=h(D_hs,t)+h(s,D_ht) \end{align*} for all smooth sections $s,t$ of $E$. [/definition] The first condition fixes the anti-holomorphic part of the connection. The second condition fixes the remaining holomorphic part by forcing the derivative of the metric matrix to match the connection matrix. [quotetheorem:3837] [citeproof:3837] The local formula for the Chern connection is the main computational tool of the section. Its hypotheses are doing real work: without a holomorphic structure there is no distinguished operator $\bar\partial_E$ to use as the $(0,1)$ part, and without positive-definiteness the metric matrix cannot be inverted to recover the connection. This is different from the Levi-Civita connection, which is determined by a Riemannian metric on the tangent bundle together with torsion-freeness rather than by a holomorphic structure. [example: Chern Connection For A Weighted Line] Let $E = X \times \mathbb{C}$ be the product line bundle over a complex manifold $X$, with global holomorphic frame $e$ and Hermitian metric $h(e,e) = e^{-\varphi}$ for $\varphi \in C^\infty(X;\mathbb{R})$; we show that the Chern connection one-form in this frame is $A = -\partial\varphi$ and exhibit the full action of $D_h$ on every smooth section. Since $r = 1$, the metric matrix reduces to the scalar function $h = e^{-\varphi}$. By *Existence and Uniqueness of the Chern Connection*, the local connection matrix in any holomorphic frame is $A = h^{-1}\partial h$. Inverting the everywhere-positive scalar gives $h^{-1} = e^{\varphi}$, and the chain rule applied to the smooth function $e^{-\varphi}$ yields \begin{align*} \partial\!\left(e^{-\varphi}\right) = -e^{-\varphi}\,\partial\varphi. \end{align*} Substituting both factors: \begin{align*} A = h^{-1}\partial h = e^{\varphi}\cdot\!\left(-e^{-\varphi}\,\partial\varphi\right) = -\partial\varphi. \end{align*} Every smooth section of $E$ has the unique form $s = fe$ for some $f \in C^\infty(X;\mathbb{C})$. Applying the local Chern connection formula $D_h(ea) = e(da + Aa)$ with scalar $a = f$ and $A = -\partial\varphi$ gives \begin{align*} D_h(fe) = e\!\left(df + (-\partial\varphi)f\right) = e\!\left(df - f\,\partial\varphi\right). \end{align*} Decomposing $df = \partial f + \bar\partial f$ by bidegree, the right-hand side splits as \begin{align*} D_h(fe) = e\!\left(\partial f - f\,\partial\varphi\right) + e\,\bar\partial f, \end{align*} where the first summand has type $(1,0)$ and the second has type $(0,1)$. The $(0,1)$ component is therefore $D_h^{0,1}(fe) = e\,\bar\partial f = \bar\partial_E(fe)$, confirming agreement with the Dolbeault operator as required by the definition of the Chern connection. That $A = -\partial\varphi$ is a pure $(1,0)$-form reflects a general principle: in any holomorphic frame the Chern connection matrix must have type $(1,0)$, and in this globally product bundle the single weight function $\varphi$ carries all metric information, so the entire connection is encoded by one $\partial$-differential. [/example] ## Curvature Forms And Gauge Laws What local quantity measures the failure of covariant derivatives to commute? The answer is curvature. Since a rank $r$ bundle has $r$ interacting local components, the curvature is naturally a matrix-valued differential form. A scalar curvature attempt fails outright for rank $r\ge 2$: covariant derivatives along two coordinate directions are then linear maps on $\mathbb C^r$, and their commutator is a matrix of two-forms whose off-diagonal entries cannot be packaged into a single scalar form. Even at rank one, dropping the quadratic term $A\wedge A$ would produce $dA$, which generally fails to transform correctly under a non-abelian change of frame; the wedge term is precisely the correction that makes the result a tensor in $\mathcal A^2(X;\operatorname{End}E)$. [definition: Matrix-Valued Differential Form] Let $U$ be an open subset of a complex manifold and let $r\in\mathbb N$. A matrix-valued $k$-form on $U$ is an element of \begin{align*} \mathcal A^k(U;\operatorname{Mat}_r(\mathbb C)). \end{align*} The product of matrix-valued forms uses matrix multiplication of coefficients and wedge product of differential forms. [/definition] This convention makes the usual connection formula look like the familiar formula from gauge theory. If $D=d+A$ in a local frame, then applying $D$ twice gives a two-form with values in endomorphisms of the bundle. The invariant object behind that local calculation is the curvature of the connection, which is what remains meaningful after changing frames. Applying a connection twice measures the obstruction to covariant differentiation behaving like an exact differential with square zero. Because this obstruction acts on sections of the bundle, its natural value is an endomorphism of the fiber attached to a two-form direction, not an ordinary scalar form in general. [definition: Curvature Of A Connection] Let $D$ be a connection on a complex vector bundle $E\to X$. The curvature of $D$ is the endomorphism-valued two-form \begin{align*} F_D=D^2\in\mathcal A^2(X;\operatorname{End}E). \end{align*} For the Chern connection of $(E,h)$, write \begin{align*} \Theta_h=F_{D_h}. \end{align*} [/definition] In a local frame with connection matrix $A$, curvature is first computed by the general structural identity for a connection. This identity explains what must be checked before any Chern-specific simplification: the local matrix expression is not itself an invariant object, but it represents the globally defined two-form $F_D=D^2$. To use curvature in computations, one needs the exact local formula relating $F_D$ to the connection matrix. The difficulty is that the connection matrix $A$ is not itself tensorial: changing a frame introduces derivative terms, and simply differentiating entries of $A$ would miss the interaction between matrix multiplication and wedge product. In the notation of this section, if a connection is written locally as $D=d+A$, then its curvature is represented by \begin{align*} F_D=dA+A\wedge A. \end{align*} For a Chern connection in a holomorphic frame with Hermitian metric matrix $h$, the connection matrix is $A=h^{-1}\partial h$, so the curvature matrix is \begin{align*} \Theta_h=\bar\partial(h^{-1}\partial h). \end{align*} This is the connection-theoretic curvature formula in the notation used here. For a Chern connection, the holomorphic frame forces $A$ to have type $(1,0)$, so the resulting curvature has type $(1,1)$ and can later be expressed in terms of the metric. The metric and frame-change statements below explain why the Chern curvature forms obtained from it are globally meaningful. [quotetheorem:3838] [citeproof:3838] The holomorphicity of $g$ is essential in this law: the inhomogeneous term $g^{-1}\partial g$ is then of type $(1,0)$, so the transformed connection still has the Chern form in a holomorphic frame. The connection matrix itself does not transform tensorially because differentiating a changed frame produces the extra derivative term, whereas curvature is tensorial and transforms only by conjugation. This distinction is what allows traces, determinants, and rank-one curvature forms to become globally meaningful in the construction of Chern forms. [example: Curvature Of A Weighted Product Line] Let $E = X \times \mathbb{C}$ carry the global holomorphic frame $e$ and Hermitian metric $h(e,e) = e^{-\varphi}$ for $\varphi \in C^\infty(X;\mathbb{R})$; we show that the curvature of the Chern connection is $\Theta_h = \partial\bar\partial\varphi$. Since $r = 1$, the metric matrix reduces to the scalar function $h = e^{-\varphi}$. By [Local Formula for Curvature](/theorems/1540), the curvature matrix in a holomorphic frame is $\Theta_h = \bar\partial(h^{-1}\partial h)$, so everything turns on evaluating $h^{-1}\partial h$ as a scalar. Inverting the everywhere-positive factor gives $h^{-1} = e^{\varphi}$, and the chain rule applied to $e^{-\varphi}$ yields $\partial(e^{-\varphi}) = -e^{-\varphi}\partial\varphi$. Multiplying: \begin{align*} h^{-1}\partial h = e^{\varphi}\cdot\!\left(-e^{-\varphi}\partial\varphi\right) = -\partial\varphi. \end{align*} Substituting into the curvature formula: \begin{align*} \Theta_h = \bar\partial\!\left(h^{-1}\partial h\right) = \bar\partial(-\partial\varphi) = -\bar\partial\partial\varphi. \end{align*} It remains to identify $-\bar\partial\partial\varphi$ with $\partial\bar\partial\varphi$. Apply $d^2 = 0$ to the smooth function $\varphi$: expanding $d = \partial + \bar\partial$ gives \begin{align*} 0 = d^2\varphi = \partial^2\varphi + \bigl(\partial\bar\partial + \bar\partial\partial\bigr)\varphi + \bar\partial^2\varphi. \end{align*} Since $\partial^2 = 0$ and $\bar\partial^2 = 0$ on forms of any bidegree, the outer terms vanish and $\partial\bar\partial\varphi + \bar\partial\partial\varphi = 0$, hence $\bar\partial\partial\varphi = -\partial\bar\partial\varphi$. Therefore: \begin{align*} \Theta_h = -\bar\partial\partial\varphi = \partial\bar\partial\varphi. \end{align*} The curvature of $(E,h)$ is thus controlled entirely by the $\partial\bar\partial$ of the weight function $\varphi$ — the operator that, in local coordinates, encodes the complex Hessian — so the positivity of $\Theta_h$ as a $(1,1)$-form is precisely the condition that $\varphi$ be strictly plurisubharmonic. [/example] ## First Chern Forms Of Hermitian Line Bundles How does matrix curvature become a scalar differential form with topological meaning? In rank one, conjugation has no effect on curvature, and the curvature form itself can be normalized to represent the first Chern class. This is the first place where curvature becomes a bridge from local analysis to global geometry. [definition: First Chern Form] Let $(L,h)$ be a Hermitian holomorphic line bundle. In a local holomorphic frame $e$ with metric function $h_e=h(e,e)$, the first Chern form is locally \begin{align*} c_1(L,h)=-\frac{i}{2\pi}\partial\bar\partial\log h_e. \end{align*} [/definition] For a line bundle, the curvature formula is $\Theta_h=-\partial\bar\partial\log h_e$, so $c_1(L,h)=\frac{i}{2\pi}\Theta_h$. If $h_e=e^{-\varphi}$, then \begin{align*} c_1(L,h)=\frac{i}{2\pi}\partial\bar\partial\varphi. \end{align*} The local formula still depends on the chosen Hermitian metric, so it is not yet a topological invariant by itself. What survives metric changes is the de Rham cohomology class: the representative may move, but the class is fixed by the holomorphic line bundle. This is the mechanism that lets curvature computations on frames produce Chern classes, and later results such as integrality of Chern classes and the Kodaira embedding theorem depend on exactly this passage from curvature representatives to global classes. [illustration:scv-iv-tautological-line-bundle] [example: Tautological Line Bundle On Projective Space] The tautological line bundle $\mathcal{O}_{\mathbb{P}^n}(-1)$ over $\mathbb{P}^n$ has fiber $\mathbb{C}Z$ over each point $[Z]$, embedded in $\mathbb{C}^{n+1}$ as a line through the origin. On $U_0 = \{Z_0 \neq 0\}$, write $z_j = Z_j/Z_0$ for $1 \leq j \leq n$, and define the local section $e_0([1:z]) = (1, z_1, \ldots, z_n) \in \mathbb{C}^{n+1}$; since each coordinate is holomorphic in $(z_1, \ldots, z_n)$, this is a local holomorphic frame for $\mathcal{O}_{\mathbb{P}^n}(-1)|_{U_0}$. The bundle carries the Hermitian metric $h_{-1}$ whose value on any fiber vector $v \in \mathbb{C}Z \subset \mathbb{C}^{n+1}$ is given by restricting the standard Hermitian inner product $\langle u, v\rangle = \sum_{k=0}^n \bar{u}_k v_k$ on $\mathbb{C}^{n+1}$. [claim] In the frame $e_0$ on $U_0$, \begin{align*} h_{-1}(e_0, e_0) = 1 + \sum_{j=1}^n |z_j|^2, \end{align*} and the first Chern form of $\bigl(\mathcal{O}_{\mathbb{P}^n}(-1), h_{-1}\bigr)$ is \begin{align*} c_1\!\left(\mathcal{O}_{\mathbb{P}^n}(-1),\,h_{-1}\right) = -\frac{i}{2\pi}\,\partial\bar\partial\log\!\left(1+\sum_{j=1}^n |z_j|^2\right). \end{align*} [/claim] *Metric function.* At $[1:z] \in U_0$, the fiber is the line $\mathbb{C}(1,z_1,\ldots,z_n) \subset \mathbb{C}^{n+1}$, and the induced metric evaluates the standard inner product on the frame vector: \begin{align*} h_{-1}(e_0, e_0) = \bigl\langle (1,z_1,\ldots,z_n),\,(1,z_1,\ldots,z_n)\bigr\rangle_{\mathbb{C}^{n+1}} = \bar{1}\cdot 1 + \sum_{j=1}^n \bar{z}_j z_j = 1 + \sum_{j=1}^n |z_j|^2. \end{align*} *First Chern form.* The definition of the first Chern form in a local holomorphic frame $e$ with metric function $h_e = h(e,e)$ gives \begin{align*} c_1(L,h) = -\frac{i}{2\pi}\,\partial\bar\partial\log h_e. \end{align*} Substituting $h_e = 1 + \sum_{j=1}^n |z_j|^2$ yields the claimed formula directly. *Global consistency.* The local expression must agree on chart overlaps. On $U_0 \cap U_k$ for any $k \geq 1$, the frame $e_k$ is defined by the representative $(Z_0/Z_k,\, Z_1/Z_k,\, \ldots,\, 1,\, \ldots,\, Z_n/Z_k)$, and $e_k = e_0 g_{0k}$ with transition function $g_{0k} = z_k^{-1} = Z_0/Z_k$. Since the rank-one metric scales as $\tilde{h} = |g|^2 h$ under $\tilde{e} = eg$ (conjugate-linearity in the first argument and linearity in the second give $h(e_0 g_{0k}, e_0 g_{0k}) = |g_{0k}|^2 h(e_0, e_0)$), we get \begin{align*} h_{-1}(e_k, e_k) = |z_k^{-1}|^2\,h_{-1}(e_0,e_0) = |z_k|^{-2}\!\left(1+\sum_{j=1}^n |z_j|^2\right). \end{align*} Therefore \begin{align*} \log h_{-1}(e_0,e_0) - \log h_{-1}(e_k,e_k) = \log|z_k|^2. \end{align*} Since $z_k$ is holomorphic and nonvanishing on $U_0 \cap U_k$, we can write $\log|z_k|^2 = \log z_k + \overline{\log z_k}$ locally; the first summand is holomorphic so $\bar\partial\log z_k = 0$, and the second is anti-holomorphic so $\partial\overline{\log z_k} = 0$, giving $\partial\bar\partial\log|z_k|^2 = 0$. The local first Chern forms therefore coincide on every overlap, defining a global $(1,1)$-form on $\mathbb{P}^n$. Comparing with the Fubini–Study form $\omega_{\mathrm{FS}} = \frac{i}{2\pi}\partial\bar\partial\log\bigl(1+\sum_j|z_j|^2\bigr)$, the result reads $c_1\bigl(\mathcal{O}_{\mathbb{P}^n}(-1), h_{-1}\bigr) = -\omega_{\mathrm{FS}}$: the tautological bundle curves negatively, which is the geometric reason the dual hyperplane bundle $\mathcal{O}_{\mathbb{P}^n}(1)$, whose metric function is the reciprocal $\bigl(1+\sum_j|z_j|^2\bigr)^{-1}$, is the positive line bundle whose first Chern form equals the standard Kähler class. [/example] The dual bundle reverses this sign. This is the geometric reason the hyperplane bundle, rather than the tautological bundle, carries the positive Fubini-Study representative. [example: Hyperplane Bundle On Projective Space] The hyperplane bundle $\mathcal{O}_{\mathbb{P}^n}(1)$ is the dual line bundle $\mathcal{O}_{\mathbb{P}^n}(-1)^*$. On $U_0 = \{Z_0 \neq 0\}$, let $e_0^*$ denote the dual frame to the tautological frame $e_0 = (1, z_1, \ldots, z_n)$, normalized by $e_0^*(e_0) = 1$. The dual metric $h_1$ on $\mathcal{O}(1)$ is defined at each fiber point $\xi \in \mathcal{O}(1)_x = L_x^*$ by \begin{align*} h_1(\xi,\xi) = \sup_{v\,\in\, L_x\setminus\{0\}}\frac{|\xi(v)|^2}{h_{-1}(v,v)}. \end{align*} [claim] In the frame $e_0^*$ on $U_0$, \begin{align*} h_1(e_0^*,e_0^*) = \left(1+\sum_{j=1}^n|z_j|^2\right)^{-1}, \end{align*} and the first Chern form of $\bigl(\mathcal{O}_{\mathbb{P}^n}(1),h_1\bigr)$ is \begin{align*} c_1\!\left(\mathcal{O}_{\mathbb{P}^n}(1),h_1\right) = \frac{i}{2\pi}\,\partial\bar\partial\log\!\left(1+\sum_{j=1}^n|z_j|^2\right). \end{align*} [/claim] *Metric function.* Every nonzero vector in the fiber $L_x = \mathcal{O}(-1)_x$ has the form $v = ae_0(x)$ for some $a \in \mathbb{C}^*$. Evaluating the dual frame gives $e_0^*(ae_0) = a$, so $|e_0^*(ae_0)|^2 = |a|^2$. By sesquilinearity of $h_{-1}$, \begin{align*} h_{-1}(ae_0,ae_0) = |a|^2\,h_{-1}(e_0,e_0) = |a|^2\!\left(1+\sum_{j=1}^n|z_j|^2\right). \end{align*} Write $H = 1+\sum_{j=1}^n|z_j|^2$. The ratio inside the supremum is \begin{align*} \frac{|e_0^*(ae_0)|^2}{h_{-1}(ae_0,ae_0)} = \frac{|a|^2}{|a|^2 H} = \frac{1}{H}, \end{align*} independent of $a$. The supremum is therefore attained at every nonzero $a$ and equals $H^{-1}$, giving \begin{align*} h_1(e_0^*,e_0^*) = \left(1+\sum_{j=1}^n|z_j|^2\right)^{-1}. \end{align*} *First Chern form.* The definition of the first Chern form in a local holomorphic frame with metric function $h_e = h(e,e)$ gives \begin{align*} c_1(\mathcal{O}(1),h_1) = -\frac{i}{2\pi}\,\partial\bar\partial\log h_1(e_0^*,e_0^*). \end{align*} Since $h_1(e_0^*,e_0^*) = H^{-1}$, the logarithm satisfies $\log H^{-1} = -\log H$. Substituting and using linearity of $\partial\bar\partial$: \begin{align*} c_1(\mathcal{O}(1),h_1) = -\frac{i}{2\pi}\,\partial\bar\partial(-\log H) = \frac{i}{2\pi}\,\partial\bar\partial\log\!\left(1+\sum_{j=1}^n|z_j|^2\right). \end{align*} *Global consistency.* The transition function for $\mathcal{O}(-1)$ from $e_0$ to $e_k$ is $g_{0k} = z_k^{-1}$, so for the dual bundle $\mathcal{O}(1)$ the transition function is $g_{0k}^{-1} = z_k$. Since $z_k$ is holomorphic and nonvanishing on $U_0\cap U_k$, the argument from the tautological example applies verbatim: locally write $\log|z_k|^2 = \log z_k + \overline{\log z_k}$, so $\bar\partial\log z_k = 0$ and $\partial\overline{\log z_k} = 0$, giving $\partial\bar\partial\log|z_k|^2 = 0$. The local first Chern forms therefore agree on every overlap, and the collection of local expressions defines a global closed real $(1,1)$-form on $\mathbb{P}^n$ by *Well-Definedness of the First Chern Form*. Comparing with the tautological bundle, whose first Chern form is $-\omega_{\mathrm{FS}}$, the single sign flip $\log H^{-1} = -\log H$ in the dual metric converts a negatively curved bundle into a positively curved one: $\mathcal{O}(1)$ is the positive line bundle on $\mathbb{P}^n$ whose curvature form is the Fubini–Study class, and it is precisely this positivity that underlies the ampleness of $\mathcal{O}(1)$ and the Kodaira embedding theorem. [/example] The section has reduced the curvature of a Hermitian holomorphic line bundle to local weights: $h=e^{-\varphi}$ gives curvature $\partial\bar\partial\varphi$ and first Chern form $\frac{i}{2\pi}\partial\bar\partial\varphi$. The next section turns this observation into positivity: a line bundle is positive when these local weights are strictly plurisubharmonic, so curvature becomes an analytic condition with global geometric consequences. Section 1 equipped holomorphic vector bundles with the Chern connection and its curvature form, an $\operatorname{End}(E)$-valued $(1,1)$-form measuring infinitesimal holomorphic non-triviality. Section 2 asks which curvature forms qualify as positive, and proves that this positivity, formulated via strictly plurisubharmonic weights in the line-bundle case, is precisely the condition needed to embed complex manifolds projectively. # 2. Positivity of Line Bundles Section 1 attached curvature to a Hermitian metric through the Chern connection. This section asks when that curvature should be regarded as positive, and why that sign condition is the bridge between local plurisubharmonic analysis and global projective geometry. The guiding principle is that a positive line bundle is a global way of packaging strictly plurisubharmonic local potentials. Building on Section 1's curvature formula for weighted line bundles, the discussion moves from pointwise curvature signs to local weights, then to the model case of the hyperplane bundle on projective space. The final sections explain the algebro-geometric language of ample and very ample bundles, and compare positive curvature with degree on curves and with flat line bundles. ## Curvature Signs for Hermitian Line Bundles The first question is how to read a geometric sign from a $(1,1)$-form. On a complex manifold, positivity is tested on complex tangent directions rather than on arbitrary real two-planes, so it is a Hermitian linear algebra condition at each point. [definition: Positive Real One One Form] Let $X$ be a complex manifold. A smooth real $(1,1)$-form $\omega$ is positive at $x \in X$ if, in any holomorphic coordinate chart near $x$, \begin{align*} \omega &= i\sum_{j,k=1}^{n} a_{j\bar{k}}\,dz_j \wedge d\bar{z}_k \end{align*} has Hermitian coefficient matrix $(a_{j\bar{k}}(x))$ positive definite. It is semipositive at $x$ if this matrix is positive semidefinite. It is negative at $x$ if $-\omega$ is positive at $x$. [/definition] This definition is coordinate-independent because holomorphic changes of coordinate transform the coefficient matrix by Hermitian congruence. Thus the sign of $\omega_x(v,\bar v)$ for $v \in T_x^{1,0}X$ is intrinsic. Testing arbitrary real two-planes would be the wrong condition for this theory: a real $(1,1)$-form is naturally a Hermitian form after inserting one $(1,0)$ tangent vector and its conjugate, and positivity must respect the complex structure. [definition: Positive Hermitian Line Bundle] Let $L \to X$ be a holomorphic line bundle with smooth Hermitian metric $h$. The Hermitian line bundle $(L,h)$ is positive if $i\Theta_h(L)$ is a positive real $(1,1)$-form on $X$. It is semipositive if $i\Theta_h(L)$ is semipositive, and it is negative if $-i\Theta_h(L)$ is positive. [/definition] A holomorphic line bundle $L$ is positive in the sense of Kodaira when it admits at least one smooth Hermitian metric with positive curvature. The metric matters at the level of forms, while the existence of a positive metric is a property of the holomorphic line bundle. For actual tests of positivity, the abstract curvature form must be translated into local weights. This is not just a matter of notation: the form $i\Theta_h(L)$ is global, while a weight $\varphi$ exists only after choosing a holomorphic frame, so one must know that the resulting Hessian test is independent of that choice. The criterion below gives the local bridge that turns curvature positivity into a concrete Levi-matrix computation. [quotetheorem:3840] [citeproof:3840] This criterion is the practical test used throughout the course: once a metric is written as $|e|_h^2=e^{-\varphi}$ in a holomorphic frame, positivity is the positivity of the complex Hessian of $\varphi$. The word holomorphic is essential here. If one used an arbitrary smooth frame, extra $\bar\partial$-terms from the frame would enter the connection form, and the clean scalar formula $i\Theta_h(L)=i\partial\bar\partial\varphi$ would no longer isolate the geometry in one weight. This is also special to line bundles: for a higher-rank Hermitian vector bundle, curvature is an endomorphism-valued $(1,1)$-form, so positivity involves a Hermitian form on vectors in the bundle as well as tangent directions, not just a scalar complex Hessian. In practice, the computational recipe is to choose holomorphic frames, write down transition-compatible weights, and test their Levi matrices. A failed positive-definiteness test at one point and in one complex tangent direction rules out positivity of that particular metric, although another metric on the same holomorphic line bundle may still be positive. [example: Gaussian Metric on the Product Line Bundle] Let $U \subset \mathbb{C}^n$ be open and let $L = U \times \mathbb{C}$ be the product line bundle with its standard holomorphic frame $e$. Equip $L$ with the Hermitian metric defined by $|e|_h^2 = e^{-|z|^2}$, so the local weight in the sense of the *Curvature Criterion for Positivity of a Hermitian Line Bundle* is $\varphi(z) = |z|^2 = \sum_{j=1}^{n} z_j \bar{z}_j$. We show that $(L, h)$ has positive curvature. Writing $|z|^2 = \sum_{j=1}^{n} z_j \bar{z}_j$, we differentiate to find the Levi matrix of $\varphi$. For each $k$, \begin{align*} \frac{\partial \varphi}{\partial \bar{z}_k} &= z_k, \end{align*} and differentiating again in $z_j$ gives \begin{align*} \frac{\partial^2 \varphi}{\partial z_j \,\partial \bar{z}_k} &= \delta_{jk}. \end{align*} The Levi matrix $\bigl(\partial^2 \varphi / \partial z_j \partial \bar{z}_k\bigr)_{j,k}$ is therefore the $n \times n$ identity matrix $I_n$. By the *Curvature Criterion for Positivity of a Hermitian Line Bundle*, the curvature form is \begin{align*} i\Theta_h(L) &= i\,\partial\bar\partial\varphi = i \sum_{j,k=1}^{n} \delta_{jk}\,dz_j \wedge d\bar{z}_k = i\sum_{j=1}^{n} dz_j \wedge d\bar{z}_j. \end{align*} For any nonzero $\xi \in \mathbb{C}^n$, the Hermitian form evaluated on $\xi$ gives $\sum_{j,k} \delta_{jk}\,\xi_j \bar\xi_k = |\xi|^2 > 0$, so the coefficient matrix $I_n$ is positive definite. By the definition of a positive real $(1,1)$-form, $i\Theta_h(L)$ is positive everywhere on $U$, and $(L, h)$ is a positive Hermitian line bundle. This is the simplest positive metric: the quadratic weight $|z|^2$ has constant Levi matrix equal to the identity, so the curvature form is the standard Euclidean Kähler form $i\sum_j dz_j \wedge d\bar{z}_j$ on $U$. [/example] The example should be kept in mind as the local normal form for positivity: a positive line bundle is locally a holomorphic line with a strictly convex complex weight, but these local weights are glued by holomorphic transition functions. ## Local Weights and Plurisubharmonicity The analytic question is why line-bundle positivity is the same condition as strict plurisubharmonicity of local potentials. This connection is what lets $L^2$ methods for $\bar\partial$ interact with global geometry. [definition: Smooth Plurisubharmonic Function] Let $U \subset \mathbb C^n$ be open and let $\varphi \in C^\infty(U;\mathbb R)$. The function $\varphi$ is plurisubharmonic if the Hermitian matrix \begin{align*} \left(\frac{\partial^2\varphi}{\partial z_j\partial \bar z_k}(z)\right)_{j,k} \end{align*} is positive semidefinite for every $z \in U$. It is strictly plurisubharmonic if this matrix is positive definite for every $z \in U$. [/definition] For smooth functions, plurisubharmonicity is exactly semipositivity of $i\partial\bar\partial\varphi$. The strict version is the local analytic source of positive curvature. To compare this condition with line-bundle positivity, one must translate between a metric and its local weight. A Hermitian metric on a line bundle is written locally as $h(e,e)=e^{-\varphi}$, and curvature differentiates the logarithm of this expression. The theorem records that the positivity of the resulting curvature form is exactly the strict plurisubharmonicity of the weight, so the analytic condition is independent of the chosen frame. [quotetheorem:3841] [citeproof:3841] This theorem explains the sign convention: a metric written as $e^{-\varphi}$ is positive when the exponent weight $\varphi$ is strictly plurisubharmonic, not when $e^{-\varphi}$ is convex as an ordinary real function. [example: Logarithmic Weight on Affine Space] On $\mathbb{C}^n$, let $\varphi(z) = \log(1+|z|^2)$ with $|z|^2 = \sum_{j=1}^n z_j\bar{z}_j$. We show that $\varphi$ is strictly plurisubharmonic by computing its Levi matrix and verifying positive definiteness at every point. Differentiating in $\bar{z}_k$ using $\partial|z|^2/\partial\bar{z}_k = z_k$: \begin{align*} \frac{\partial\varphi}{\partial\bar{z}_k} &= \frac{z_k}{1+|z|^2}. \end{align*} Differentiating in $z_j$ by the quotient rule, with $\partial(1+|z|^2)/\partial z_j = \bar{z}_j$: \begin{align*} \frac{\partial^2\varphi}{\partial z_j\,\partial\bar{z}_k} &= \frac{\delta_{jk}(1+|z|^2) - z_k\bar{z}_j}{(1+|z|^2)^2}. \end{align*} For nonzero $\xi\in\mathbb{C}^n$, write $\langle z,\xi\rangle = \sum_j\bar{z}_j\xi_j$, so that $\sum_k z_k\bar{\xi}_k = \overline{\langle z,\xi\rangle}$. Separating the diagonal and rank-one contributions in the Hermitian form: \begin{align*} \sum_{j,k} \frac{\partial^2\varphi}{\partial z_j\,\partial\bar{z}_k}\,\xi_j\bar{\xi}_k &= \frac{(1+|z|^2)\sum_j|\xi_j|^2 - \langle z,\xi\rangle\,\overline{\langle z,\xi\rangle}}{(1+|z|^2)^2} \\ &= \frac{(1+|z|^2)|\xi|^2 - |\langle z,\xi\rangle|^2}{(1+|z|^2)^2}. \end{align*} By [Cauchy–Schwarz](/theorems/432), $|\langle z,\xi\rangle|^2 \le |z|^2|\xi|^2$, so \begin{align*} (1+|z|^2)|\xi|^2 - |\langle z,\xi\rangle|^2 &\ge (1+|z|^2)|\xi|^2 - |z|^2|\xi|^2 = |\xi|^2 > 0. \end{align*} The Levi matrix is positive definite at every point of $\mathbb{C}^n$, confirming that $\varphi$ is strictly plurisubharmonic. By the *Local Weight Criterion for Positivity*, the product line bundle over any open $U\subset\mathbb{C}^n$ equipped with the metric $|e|_h^2 = e^{-\varphi}$ is a positive Hermitian line bundle. The Levi matrix $(1+|z|^2)^{-2}\bigl[(1+|z|^2)I - \bar{z}z^T\bigr]$ — a scaled identity minus a rank-one correction, dominated by the identity term by exactly the Cauchy–Schwarz gap — is the affine chart expression that reappears as the coefficient matrix of the Fubini-Study curvature form of $\mathcal{O}_{\mathbb{P}^n}(1)$ on $U_0 = \{Z_0\ne 0\}$. [/example] This example is the basic local model behind projective positivity: logarithmic growth keeps the curvature positive while preventing the metric from behaving like the noncompact Gaussian model at infinity. The same computation also gives a reusable test: differentiate the proposed weight, identify the diagonal positive part, and control any rank-one correction by Cauchy--Schwarz. [remark: Weight Changes by Pluriharmonic Terms] The weights of a fixed metric are not globally defined functions unless the line bundle has a global frame. Their Levi forms are globally defined because changing frames adds $-\log |g|^2$, and such a term has zero $\partial\bar\partial$. Positivity is therefore local in computation but global in meaning. [/remark] ## The Fubini-Study Metric on Projective Space The model question is what the positive curvature form looks like on projective space. The answer supplies the basic example for the whole subject: the hyperplane bundle carries a canonical positive metric whose curvature is the Fubini-Study form. Let $[Z_0:\cdots:Z_n]$ be homogeneous coordinates on $\mathbb P^n$. On the affine chart $U_0=\{Z_0\ne 0\}$, write $z_j=Z_j/Z_0$ for $1\le j\le n$. [definition: Fubini-Study Metric on the Hyperplane Bundle] The Fubini-Study metric $h_{\mathrm{FS}}$ on $\mathcal O_{\mathbb P^n}(1)$ is the Hermitian metric whose local weight on $U_0$ in the standard hyperplane frame is \begin{align*} \varphi_0(z) &= \log(1+|z|^2), \end{align*} with the analogous weight on each affine chart $U_j=\{Z_j\ne 0\}$. [/definition] The transition functions between these frames change the weights by logarithms of squared moduli of holomorphic nowhere-vanishing functions, so the local definition gives a global Hermitian metric. Having defined the metric by local weights, the next issue is whether its curvature has the expected positivity and cohomology normalization. This calculation is the model for every later use of positive line bundles: it turns the elementary potential $\log(1+|z|^2)$ into a global positive $(1,1)$-form. It also identifies the curvature representative of the hyperplane class, which is the class that controls projective embeddings. [quotetheorem:3842] [citeproof:3842] This calculation is the local source of the standard Kähler form on projective space. It also fixes the normalisation of the hyperplane class used in projective embedding statements. A different potential such as $\log(1+a|z|^2)$ with $a>0$ changes the local size of the form, but it represents the same hyperplane class after the corresponding global normalisation is fixed. The factor $(1+|z|^2)^{-2}$ in dimension one is not cosmetic: it is the decay that makes the total Fubini-Study area finite on the affine chart, with the missing point supplied by the chart at infinity. [illustration:scv-iv-fubini-study-p1] [example: The Projective Line] On $\mathbb{P}^1$ with affine coordinate $z = Z_1/Z_0$ on $U_0 = \{Z_0\neq 0\}$, the Fubini-Study first Chern form restricts to \begin{align*} \omega_{\mathrm{FS}} &= \frac{i}{2\pi}\frac{dz\wedge d\bar{z}}{(1+|z|^2)^2}; \end{align*} we show that $\int_{\mathbb{P}^1}\omega_{\mathrm{FS}} = 1$, so $[\omega_{\mathrm{FS}}]$ is the positive generator of $H^2(\mathbb{P}^1;\mathbb{Z})\cong\mathbb{Z}$ and $\deg\mathcal{O}_{\mathbb{P}^1}(1) = 1$. The complement of $U_0$ in $\mathbb{P}^1$ is the single point $[0:1]$, which has measure zero with respect to any smooth area form, so $\int_{\mathbb{P}^1}\omega_{\mathrm{FS}} = \int_{\mathbb{C}}\omega_{\mathrm{FS}}$. Writing $z = x+iy$, we expand the wedge product directly: \begin{align*} dz\wedge d\bar{z} &= (dx+i\,dy)\wedge(dx-i\,dy) = -i\,dx\wedge dy + i\,dy\wedge dx = -2i\,dx\wedge dy, \end{align*} where the last equality uses $dy\wedge dx = -dx\wedge dy$. Inserting this into the prefactor gives \begin{align*} \frac{i}{2\pi}\,dz\wedge d\bar{z} &= \frac{i(-2i)}{2\pi}\,dx\wedge dy = \frac{1}{\pi}\,dx\wedge dy, \end{align*} so the integral becomes $\frac{1}{\pi}\int_{\mathbb{C}}(1+|z|^2)^{-2}\,dx\,dy$. Passing to polar coordinates $x = r\cos\theta$, $y = r\sin\theta$ with area element $dx\,dy = r\,dr\,d\theta$ by [Change of Variables (general)](/theorems/22) and $|z|^2 = r^2$: \begin{align*} \int_{\mathbb{C}}\omega_{\mathrm{FS}} &= \frac{1}{\pi}\int_0^{2\pi}\!\int_0^{\infty}\frac{r}{(1+r^2)^2}\,dr\,d\theta. \end{align*} The integrand $r(1+r^2)^{-2}$ is nonnegative and integrable on $[0,\infty)\times[0,2\pi)$, so by [Fubini Theorem](/theorems/513) the iterated integrals separate: \begin{align*} &= \frac{1}{\pi}\int_0^{2\pi}d\theta\cdot\int_0^{\infty}\frac{r\,dr}{(1+r^2)^2} = \frac{2\pi}{\pi}\int_0^{\infty}\frac{r\,dr}{(1+r^2)^2} = 2\int_0^{\infty}\frac{r\,dr}{(1+r^2)^2}. \end{align*} Substituting $u = 1+r^2$, so $du = 2r\,dr$, with $u = 1$ when $r = 0$ and $u\to\infty$ as $r\to\infty$: \begin{align*} 2\int_0^{\infty}\frac{r\,dr}{(1+r^2)^2} &= 2\int_1^{\infty}\frac{du/2}{u^2} = \int_1^{\infty}u^{-2}\,du = \Bigl[-u^{-1}\Bigr]_1^{\infty} = 0-(-1) = 1. \end{align*} Since $\omega_{\mathrm{FS}}$ is a positive $(1,1)$-form by the *Fubini-Study Curvature Form* theorem and integrates to $1$ against the fundamental class $[\mathbb{P}^1]$, the class $[\omega_{\mathrm{FS}}]$ maps to $1$ under the pairing $H^2(\mathbb{P}^1;\mathbb{Z})\otimes H_2(\mathbb{P}^1;\mathbb{Z})\to\mathbb{Z}$, identifying it with the positive generator. The first Chern class satisfies $c_1(\mathcal{O}_{\mathbb{P}^1}(1)) = [\omega_{\mathrm{FS}}]$, so $\deg\mathcal{O}_{\mathbb{P}^1}(1) = \int_{\mathbb{P}^1}c_1(\mathcal{O}_{\mathbb{P}^1}(1)) = 1$, consistent with the hyperplane bundle cutting out exactly one point on each line in $\mathbb{P}^2$. [/example] The positivity of $\mathcal O(1)$ is the reason projective space is the target for maps built from sections of positive line bundles. ## Projective Embeddings and Kodaira Positivity The global question is how curvature positivity creates maps to projective space. Algebraic geometry expresses this through very ample and ample line bundles, while complex differential geometry expresses it through positive curvature metrics. Let $X$ be a compact complex manifold and let $L\to X$ be a holomorphic line bundle. If sections $s_0,\dots,s_N\in H^0(X,L)$ have no common zero, they define a holomorphic map \begin{align*} \Phi_{|s_0,\dots,s_N|}:X&\longrightarrow \mathbb P^N,\\ x&\longmapsto [s_0(x):\cdots:s_N(x)], \end{align*} where the expression is read in any local frame of $L$. [definition: Very Ample Line Bundle] A holomorphic line bundle $L\to X$ on a compact complex manifold is very ample if its complete space of global sections $H^0(X,L)$ has no common zero and the associated map \begin{align*} \Phi_L:X\to \mathbb P(H^0(X,L)^*) \end{align*} is a holomorphic embedding. [/definition] Very ampleness means that sections of $L$ separate points and tangent directions at the same time. The pullback of the hyperplane bundle along the embedding recovers $L$. Many natural line bundles acquire enough sections only after taking a positive tensor power, so the projective notion used in comparison with curvature must allow that stabilization. Tensor powers amplify sections and curvature in compatible ways: geometrically they can create an embedding even when the original bundle does not, while analytically a positive curvature form remains positive after multiplication by a positive integer. The definition below records the projective condition that matches this stabilization process. [definition: Ample Line Bundle] A holomorphic line bundle $L\to X$ on a compact complex manifold is ample if there exists $m\in \mathbb N$ such that $L^{\otimes m}$ is very ample. [/definition] Ampleness is weaker than very ampleness because it allows passage to a tensor power. Curvature positivity is already stable under tensor powers: if $h$ has positive curvature on $L$, then $h^{\otimes m}$ has curvature $m\Theta_h(L)$ on $L^{\otimes m}$. The first bridge from projective geometry to curvature starts with the strongest section-theoretic hypothesis, very ampleness. When a bundle already embeds the manifold into projective space, the missing analytic object is a positive Hermitian metric on that bundle. The result below supplies it by pulling the standard positive metric back from projective space. [quotetheorem:3843] [citeproof:3843] This proves the differential-geometric half of the bridge: very ample line bundles carry positive metrics by pulling back the Fubini-Study metric. The opposite direction is the content of Kodaira's embedding theorem. The result should be read as a source of examples and as a consistency check on the definition of positivity. A very ample bundle is already presented by homogeneous coordinates, so the Fubini-Study metric on projective space supplies a canonical positive metric after pullback. What is not yet automatic is that an abstract positive line bundle has enough sections to give such an embedding; that harder direction requires analysis of the $\bar\partial$-operator and is exactly what Kodaira's theorem supplies. [quotetheorem:3844] [citeproof:3844] [remark: Kodaira Positivity as a Dictionary] In analysis, positivity means a strictly positive curvature form. In algebraic geometry, positivity means enough sections in high tensor powers to embed the manifold into projective space. Kodaira's theorem identifies these two languages on compact complex manifolds. A useful way to apply the dictionary is to move in whichever direction is easier: construct a positive metric and then invoke Kodaira to obtain sections, or start from an embedding and pull back the Fubini-Study metric to obtain curvature positivity. [/remark] ## Divisors and Degree on Compact Riemann Surfaces The curve-level question is how curvature positivity records degree. On a compact Riemann surface, every real $(1,1)$-form is a scalar multiple of an area form locally, so positivity is governed by the sign of the total curvature class. [definition: Divisor on a Compact Riemann Surface] Let $C$ be a compact Riemann surface. A divisor on $C$ is a finite formal sum \begin{align*} D &= \sum_{p\in C} m_p p, \end{align*} with $m_p\in \mathbb Z$. Its degree is \begin{align*} \deg D &= \sum_{p\in C} m_p. \end{align*} The divisor is effective if $m_p\ge 0$ for every $p\in C$. [/definition] An effective divisor $D$ determines a holomorphic line bundle $\mathcal O_C(D)$. Its sections are meromorphic functions whose allowed pole orders are bounded by the coefficients of $D$, expressed intrinsically as sections of the associated line bundle. For curves, positivity can often be recognized without writing a curvature form explicitly. The obstruction is that effectivity alone describes allowed zeros and poles, while positivity should be a property of the associated line bundle. The curve theorem below identifies the numerical condition on a divisor that guarantees the corresponding line bundle is positive. [quotetheorem:3845] [citeproof:3845] This result is the one-dimensional version of Kodaira positivity. On curves, the numerical invariant degree already detects positivity. Compactness is essential in this statement: on $\mathbb C$, degree is not a global topological invariant in this sense, and every holomorphic line bundle is trivial. In higher dimensions, positivity is also not captured by a single degree number; numerical intersection conditions with curves lead toward criteria such as Nakai--Moishezon, while curvature positivity remains a pointwise Hermitian condition. For divisors in practice, the curve case says to compute degree. In higher dimensions, an effective divisor gives a line bundle and canonical sections, but positivity must be checked through intersection theory, known ample generators, or an explicit positive metric rather than by effectivity alone. [example: Effective Divisor on a Curve] Let $C$ be a compact Riemann surface and let $D = p_1 + 2p_2$ for distinct points $p_1, p_2 \in C$. We show that $\mathcal{O}_C(D)$ is a positive line bundle. The coefficients of $D$ in the formal sum $\sum_{p} m_p\, p$ are $m_{p_1} = 1$ and $m_{p_2} = 2$, with $m_p = 0$ at every other point of $C$. Since each coefficient satisfies $m_p \geq 0$, the divisor $D$ is effective. Its degree is \begin{align*} \deg D &= m_{p_1} + m_{p_2} = 1 + 2 = 3 > 0. \end{align*} Because $\deg D > 0$, the theorem *Positivity and Degree on Compact Riemann Surfaces* applies: a holomorphic line bundle on a compact Riemann surface admits a smooth Hermitian metric with positive curvature if and only if its degree is positive. Applying that equivalence to $L = \mathcal{O}_C(D)$, which has $\deg \mathcal{O}_C(D) = \deg D = 3$, yields a smooth Hermitian metric $h$ on $\mathcal{O}_C(D)$ with $i\Theta_h(\mathcal{O}_C(D))$ a positive $(1,1)$-form on $C$. The sections of $\mathcal{O}_C(D)$ may be described locally as meromorphic functions allowed to have poles of order at most $1$ at $p_1$ and at most $2$ at $p_2$; positivity of the line bundle is the curvature-geometric expression of the room afforded by these prescribed pole orders. [/example] The divisor example shows how analytic positivity reflects the ability to prescribe controlled poles, which is the curve-level shadow of having many global sections. ## Flat Line Bundles and the Boundary of Positivity The boundary question is what remains when curvature has no sign-changing part because it vanishes. Flat line bundles are important because they are semipositive and seminegative, but they do not provide the strict curvature needed for Kodaira embedding. [definition: Flat Hermitian Line Bundle] A Hermitian holomorphic line bundle $(L,h)$ on a complex manifold $X$ is flat if \begin{align*} \Theta_h(L)&=0. \end{align*} It is unitary flat if it has local holomorphic frames in which the transition functions are locally constant with values in $U(1)$ and the metric has frame norm $1$. [/definition] Flatness is a curvature condition, while unitary flatness is a description by transition functions. For Hermitian line bundles, these are two local descriptions of the same phenomenon after choosing parallel unitary frames. This equivalence is useful because flat bundles often appear first through monodromy or transition functions rather than through a chosen metric. The question is whether constant unitary gluing data exactly captures the vanishing-curvature condition, and conversely whether zero curvature lets one choose frames in which the metric and transitions become unitary and locally constant. The following result records that dictionary. [quotetheorem:3846] [citeproof:3846] A flat metric is semipositive because the zero form is semipositive, and it is also seminegative. It is not positive on a positive-dimensional manifold because the zero Hermitian form is not positive definite on nonzero tangent vectors. [example: Flat Line Bundle from a Unitary Character] Let $X = \mathbb{C}/\Lambda$ be a complex torus, where $\Lambda \subset \mathbb{C}$ is a rank-$2$ lattice, and let $\rho: \Lambda \to U(1)$ be a group homomorphism. We show that $L_\rho$ admits a flat Hermitian metric and, when $\rho$ is nontrivial, is not isomorphic to the product bundle $X \times \mathbb{C}$. The relation $(z, w) \sim (z + \lambda,\, \rho(\lambda)\,w)$ is compatible with the group law of $\Lambda$: for $\lambda, \mu \in \Lambda$, \begin{align*} \rho(\lambda + \mu)\,w &= \rho(\lambda)\,\rho(\mu)\,w, \end{align*} because $\rho$ is a homomorphism, so identifying by $\mu$ and then by $\lambda$ agrees with identifying by $\lambda + \mu$ in a single step. Since every $\rho(\lambda)$ is a nonzero complex number, the fibers $\{z\} \times \mathbb{C}$ glue to one-dimensional complex vector spaces, making $L_\rho$ a holomorphic line bundle over $X$. Define $h$ on $L_\rho$ by $\bigl|[z, w]\bigr|_h^2 = |w|^2$. To verify well-definedness, take any other representative $(z + \lambda,\, \rho(\lambda)\,w)$ of the same equivalence class: \begin{align*} \bigl|\rho(\lambda)\,w\bigr|^2 &= |\rho(\lambda)|^2\,|w|^2 = 1 \cdot |w|^2 = |w|^2, \end{align*} where $|\rho(\lambda)|^2 = 1$ because $\rho(\lambda) \in U(1)$. Hence the norm is independent of the choice of representative, and $h$ is a well-defined Hermitian metric on $L_\rho$. On any simply connected open $U \subset X$, a local holomorphic frame is $e_U([z]) = [z, 1]$. Its squared metric norm is \begin{align*} |e_U([z])|_h^2 &= |1|^2 = 1 = e^{-\varphi_U}, \end{align*} so the local weight satisfies $\varphi_U \equiv 0$ on every frame domain. By the *Curvature Criterion for Positivity of a Hermitian Line Bundle*, the curvature form is therefore \begin{align*} i\Theta_h(L_\rho) &= i\,\partial\bar\partial\varphi_U = i\,\partial\bar\partial(0) = 0, \end{align*} confirming that $(L_\rho, h)$ is flat. When $\rho$ is nontrivial, $L_\rho$ is not isomorphic to $X \times \mathbb{C}$ as a holomorphic line bundle. An isomorphism $X \times \mathbb{C} \xrightarrow{\sim} L_\rho$ over $X$ corresponds to a global nowhere-zero holomorphic section $s \in H^0(X, L_\rho)$. Lifting $s$ to the universal cover $\mathbb{C}$, this is a holomorphic function $f: \mathbb{C} \to \mathbb{C}^*$ satisfying $f(z + \lambda) = \rho(\lambda)\,f(z)$ for every $\lambda \in \Lambda$. Since $|\rho(\lambda)| = 1$, taking moduli gives $|f(z + \lambda)| = |f(z)|$ for every $\lambda$, so $|f|$ is $\Lambda$-periodic. A $\Lambda$-periodic continuous function on $\mathbb{C}$ attains its maximum on the compact fundamental domain $\mathbb{C}/\Lambda$, hence is bounded. Thus $f$ is a bounded entire function, and by [Liouville's Theorem](/theorems/346), $f$ is constant: $f \equiv c$ for some $c \neq 0$. The equivariance condition then requires $c = \rho(\lambda)\,c$, hence $\rho(\lambda) = 1$ for every $\lambda \in \Lambda$, contradicting the assumption that $\rho$ is nontrivial. No such isomorphism exists. The flat metric $h$ places $L_\rho$ at the exact boundary of the sign conditions: $i\Theta_h(L_\rho) = 0$ is simultaneously semipositive and seminegative, and nontrivial holonomy encoded in $\rho$ shows that vanishing curvature does not force holomorphic triviality. [/example] Flat examples mark the difference between semipositivity and positivity. Semipositivity permits zero curvature directions; positivity forbids them and is the condition that produces projective embeddings after passing to high tensor powers. Having defined positivity through curvature in Section 2, we now confront a fundamental question: why does the curvature form $\frac{i}{2\pi}\partial\bar\partial\varphi$ depend on the choice of weight function $\varphi$, yet still encodes a topological invariant? Section 3 answers this by introducing Chern classes, showing that while the curvature form is metric-dependent, its de Rham cohomology class is canonical. # 3. Chern Classes and Curvature Representatives The preceding section used curvature to formulate positivity of Hermitian line bundles through local weights and the Fubini-Study model. We now ask why those curvature forms represent a topological invariant rather than only a choice-dependent analytic object. The answer is that the transition functions of a holomorphic line bundle define an integral cohomology class, while any Hermitian metric converts the same class into a closed real $(1,1)$-form. Divisors enter because zeros and poles of holomorphic data give the most concrete representatives of this class, with the Poincare-Lelong formula translating analytic singularities into cohomology. ## Transition Functions and the Topological Class The problem is to recover a global cohomology class from the local gluing functions of a holomorphic line bundle. A naive attempt would be to choose logarithms of all transition functions and add them on triple overlaps, but logarithms cannot usually be chosen compatibly around loops. The nowhere-zero condition makes local logarithms available after refinement, while the holomorphic condition keeps those logarithms in the sheaf used by the exponential sequence. A line bundle is locally a copy of $U_i \times \mathbb C$, but the information lies in how the local frames compare on overlaps. The first Chern class measures the obstruction to choosing these frames so that all transition functions become globally compatible logarithms. Let $M$ be a complex manifold and let $L \to M$ be a holomorphic line bundle. Choose an open cover $(U_i)_{i \in I}$ and holomorphic local frames $e_i$ of $L$ over $U_i$. On $U_i \cap U_j$ write \begin{align*} e_j = g_{ij} e_i, \qquad g_{ij} \in \mathcal O^*(U_i \cap U_j). \end{align*} The cocycle identities are $g_{ii}=1$, $g_{ji}=g_{ij}^{-1}$, and $g_{ij}g_{jk}g_{ki}=1$ on triple overlaps. [definition: Cech First Chern Class] Let $L \to M$ be a holomorphic line bundle represented on a cover $(U_i)$ by transition functions $g_{ij} \in \mathcal O^*(U_i \cap U_j)$. After refining the cover so that logarithms exist on each nonempty overlap, choose functions $a_{ij}$ with $\exp(2\pi i a_{ij})=g_{ij}$. On $U_i \cap U_j \cap U_k$ set \begin{align*} n_{ijk}=a_{ij}+a_{jk}+a_{ki} \in \mathbb Z. \end{align*} The first Chern class $c_1(L) \in H^2(M,\mathbb Z)$ is the Cech cohomology class represented by $(n_{ijk})$. [/definition] The integer appears because the product $g_{ij}g_{jk}g_{ki}$ is $1$, so the sum of logarithms differs from $0$ by an integral period. The construction is a boundary map in the exponential sequence \begin{align*} 0 \longrightarrow \mathbb Z \xrightarrow{\,2\pi i\,} \mathcal O \xrightarrow{\,\exp\,} \mathcal O^* \longrightarrow 1. \end{align*} Thus $c_1(L)$ is the image of the holomorphic line bundle class $[L] \in H^1(M,\mathcal O^*)=\operatorname{Pic}(M)$, where $\operatorname{Pic}(M)$ denotes the Picard group: the group of holomorphic line bundles on $M$ under tensor product. This boundary-map construction is compatible with the operations one naturally performs on line bundles. Tensor product adds transition data, dualization changes signs, and pullback along a holomorphic map transports the same obstruction class to the source. Conceptually, $c_1$ turns the multiplicative geometry of line bundles into additive cohomology. In computations, it is often enough to know a generator of $H^2(M,\mathbb Z)$ and identify which tensor power of that generator a given line bundle represents. The hypotheses are part of the mechanism. If a transition function were allowed to vanish, it would no longer compare two local frames of a line bundle, and no logarithm could be chosen near the zero. If the functions were only smooth and nonvanishing, an analogous topological Chern class exists for a smooth complex line bundle, but it is no longer the holomorphic Picard class obtained from the exponential sequence with $\mathcal O$; the Dolbeault and divisor interpretations used later are then unavailable. For real line bundles there is a sharper change: the primary obstruction is the first Stiefel-Whitney class in $H^1(M,\mathbb Z/2\mathbb Z)$, and a first Chern class is not part of the real line-bundle structure unless a compatible complex line-bundle structure is supplied. [example: Product Bundle] Let $L = M \times \mathbb{C}$ be the product bundle with global holomorphic frame $e$. On any open cover $(U_i)$, restrict to local frames $e_i = e|_{U_i}$. Because both $e_i$ and $e_j$ are restrictions of the same global frame, the overlap identity $e_j = g_{ij}\, e_i$ holds with $g_{ij} = 1$ on every $U_i \cap U_j$. For the product bundle there is no twisting to detect, so one should expect $c_1(L)=0$. In the Čech model this expectation appears directly: choose the logarithms $a_{ij}=0$ for all overlaps. The associated integer Čech cocycle is then identically zero, since on every triple overlap \begin{align*} n_{ijk} = a_{ij} + a_{jk} + a_{ki} = 0 + 0 + 0 = 0. \end{align*} The cocycle $(n_{ijk})$ is identically zero, hence represents the zero class in $H^2(M,\mathbb{Z})$. If instead one works with different local frames $\tilde{e}_i = b_i e_i$ for some $b_i \in \mathcal{O}^*(U_i)$, the transition functions become $\tilde{g}_{ij} = b_j g_{ij} b_i^{-1} = b_j \cdot 1 \cdot b_i^{-1} = b_j b_i^{-1}$. Choosing logarithms $\beta_i$ with $\exp(2\pi i\,\beta_i) = b_i$, the new cocycle satisfies \begin{align*} \tilde{n}_{ijk} = \tilde{a}_{ij} + \tilde{a}_{jk} + \tilde{a}_{ki} = (n_{ijk}) + \bigl[(\beta_j - \beta_i) + (\beta_k - \beta_j) + (\beta_i - \beta_k)\bigr] = 0 + 0 = 0, \end{align*} where the bracketed sum telescopes to zero and is the Čech coboundary of $(\beta_i)$. This is the frame-independence step of *Transition Function Formula for First Chern Class*, confirming that $c_1(L) = 0$ regardless of which local frames are chosen. The contrast with $\mathcal{O}(1)$ over $\mathbb{P}^n$ shows why the zero result is special. On the affine cover $U_i = \{Z_i \neq 0\}$, the natural frames of $\mathcal{O}(1)$ give $g_{ij} = Z_i/Z_j$. Since $g_{ij}g_{jk}g_{ki} = (Z_i/Z_j)(Z_j/Z_k)(Z_k/Z_i) = 1$, any logarithm choices $a_{ij}$ do satisfy $n_{ijk} \in \mathbb{Z}$; the obstruction is that no choice makes every $n_{ijk}$ vanish simultaneously, because the quotients $Z_i/Z_j$ admit no globally consistent single-valued branch on $\mathbb{P}^n$. The product bundle carries no such obstruction because the single global frame $e$ forces all transition functions to equal $1$ at the outset, and no topological winding can arise. [/example] ## Curvature Forms as Chern-Weil Representatives The question now is how to see the same integral class through differential forms. Differentiating overlap functions alone gives Cech data rather than a form on the open sets, and averaging local frame choices would depend on auxiliary choices rather than on the bundle. A Hermitian metric supplies local weights whose changes on overlaps are controlled by the transition functions, so their second complex derivatives can glue. The Chern-Weil theorem says that this curvature form is exactly the real cohomology image of the transition-function class. [definition: Chern Form of a Hermitian Line Bundle] Let $L \to M$ be a holomorphic line bundle with Hermitian metric $h$. In a holomorphic local frame $e$ over $U$, write $|e|_h^2=e^{-\varphi}$ for a smooth real-valued function $\varphi:U\to \mathbb R$. The first Chern form is the local $(1,1)$-form \begin{align*} c_1(L,h)|_U=\frac{i}{2\pi}\partial\bar\partial \varphi. \end{align*} [/definition] On an overlap, if $e_j=g_{ij}e_i$, then the weights satisfy $\varphi_j=\varphi_i-\log|g_{ij}|^2$. Since $g_{ij}$ is holomorphic and nowhere zero, $\partial\bar\partial\log|g_{ij}|^2=0$. The local formulas therefore patch to a global closed real $(1,1)$-form. The key point is that curvature is not merely a local second derivative of the metric weight. It is the differential-form representative of the obstruction encoded by the transition functions. The holomorphic hypothesis is exactly what makes $\partial\bar\partial\log|g_{ij}|^2=0$: locally, $\log|g_{ij}|^2$ is the real part of a holomorphic logarithm. A smooth nowhere-zero transition function need not have this property, so the curvature of a general $C^\infty$ complex line bundle need not be represented by a pure $(1,1)$-form. The invariant part is the cohomology class, whose periods are integral; the particular smooth representative, its local potential, and pointwise positivity can change with the chosen Hermitian metric. [explanation: Practical Computation of First Chern Classes] To compute $c_1$ from transition functions, choose a cover where the bundle is trivial, write the overlap functions $g_{ij}$, choose local logarithms after refinement, and read off the integer cocycle $a_{ij}+a_{jk}+a_{ki}$ on triple overlaps. In a one-generator situation such as $H^2(\mathbb P^n,\mathbb Z)\cong \mathbb Z$, it is enough to compare with $\mathcal O(1)$ by checking whether the transition functions are powers of the hyperplane transition functions. To compute the same class from curvature, choose a Hermitian metric, write the local weight $|e_i|_h^2=e^{-\varphi_i}$, form $\frac{i}{2\pi}\partial\bar\partial\varphi_i$, and check that the expressions agree on overlaps. To recognize a divisor construction in practice, look for local meromorphic equations $f_i$ whose quotients $f_i/f_j$ are holomorphic and nowhere zero; these quotients are the transition functions of $\mathcal O(D)$, and Poincare-Lelong then converts the divisor data into the Chern form identity. [/explanation] The projective-space computation below is the standard test case for this recipe. Projective space has a one-dimensional second integral cohomology, so any computation reduces to identifying which integer multiple of the hyperplane generator a bundle produces. The Fubini-Study metric supplies an explicit closed real $(1,1)$-form whose integral over a projective line equals one, which then pins down the generator on the curvature side. Comparing the transition-function calculation with the curvature representative on the same cover shows the two routes give the same class. [example: Hyperplane Divisor in Projective Space] [claim]Let $H = \{Z_0 = 0\} \subset \mathbb{P}^n$ and let $\mathcal{O}(H)$ be the line bundle constructed from $H$ by the associated-line-bundle definition. Then $\mathcal{O}(H) \cong \mathcal{O}_{\mathbb{P}^n}(1)$, the Fubini-Study metric on $\mathcal{O}(1)$ has Chern form $\omega_{\mathrm{FS}}$, and \begin{align*} c_1\!\bigl(\mathcal{O}_{\mathbb{P}^n}(1)\bigr) = [\omega_{\mathrm{FS}}] \end{align*} is the positive generator of $H^2(\mathbb{P}^n,\mathbb{Z}) \cong \mathbb{Z}$.[/claim] *Transition functions.* Work on the standard affine cover $U_i = \{Z_i \neq 0\}$. On $U_i$ the hyperplane $H = \{Z_0 = 0\}$ is cut out by the holomorphic function $f_i = Z_0/Z_i$, which vanishes on $U_i \cap H$ to first order and is nowhere zero on $U_i \setminus H$. On the overlap $U_i \cap U_j$ both $f_i$ and $f_j$ serve as local equations for the same divisor $H$, so the transition function of $\mathcal{O}(H)$ is \begin{align*} g_{ij} = \frac{f_i}{f_j} = \frac{Z_0/Z_i}{Z_0/Z_j} = \frac{Z_j}{Z_i}. \end{align*} By definition, $\mathcal{O}_{\mathbb{P}^n}(1)$ is the line bundle whose transition functions on $U_i \cap U_j$ are $Z_j/Z_i$. The two sets of transition functions are identical, so $\mathcal{O}(H) \cong \mathcal{O}_{\mathbb{P}^n}(1)$. As a consistency check, the cocycle identity on $U_i \cap U_j \cap U_k$ reads \begin{align*} g_{ij}\,g_{jk}\,g_{ki} = \frac{Z_j}{Z_i}\cdot\frac{Z_k}{Z_j}\cdot\frac{Z_i}{Z_k} = 1, \end{align*} confirming that these transition functions define a line bundle. *Chern form.* Equip $\mathcal{O}(1)$ with the Hermitian metric $h$ whose local weights are \begin{align*} |e_i|_h^2 = \frac{|Z_i|^2}{|Z_0|^2 + |Z_1|^2 + \cdots + |Z_n|^2}. \end{align*} On $U_0$, introducing affine coordinates $z_k = Z_k/Z_0$, this reads $|e_0|_h^2 = e^{-\varphi_0}$ with \begin{align*} \varphi_0 = \log\!\bigl(1 + |z_1|^2 + \cdots + |z_n|^2\bigr). \end{align*} To verify the overlap consistency required by the Chern Form definition, note that on $U_0 \cap U_1$ we have $g_{01} = Z_1/Z_0 = z_1$, and the weight transforms as $\varphi_1 = \varphi_0 - \log|g_{01}|^2 = \varphi_0 - \log|z_1|^2$. Since $z_1$ is holomorphic and nowhere zero on $U_0 \cap U_1$, its logarithmic modulus $\log|z_1|^2 = \operatorname{Re}(\log z_1)$ is pluriharmonic, so $\partial\bar\partial\log|z_1|^2 = 0$. Applying $\frac{i}{2\pi}\partial\bar\partial$ to both sides of $\varphi_1 = \varphi_0 - \log|z_1|^2$ gives $\frac{i}{2\pi}\partial\bar\partial\varphi_1 = \frac{i}{2\pi}\partial\bar\partial\varphi_0$ on the overlap, so the local forms patch. The Chern form on $U_0$ is therefore \begin{align*} c_1(\mathcal{O}(1),h)\big|_{U_0} = \frac{i}{2\pi}\partial\bar\partial\log\!\bigl(1+|z_1|^2+\cdots+|z_n|^2\bigr) = \omega_{\mathrm{FS}}. \end{align*} *Integral over a projective line.* Since $H^2(\mathbb{P}^n,\mathbb{Z})\cong\mathbb{Z}$, it suffices to show that $[\omega_{\mathrm{FS}}]$ integrates to $1$ over one projective line. Restrict to $L = \{Z_2 = \cdots = Z_n = 0\} \cong \mathbb{P}^1$, which is covered by $U_0 \cap L \cong \mathbb{C}$ via the coordinate $z = Z_1/Z_0$. On this chart $\omega_{\mathrm{FS}}|_L = \frac{i}{2\pi}\partial\bar\partial\log(1+|z|^2)$. Computing each $\bar\partial$ and $\partial$ in turn: \begin{align*} \bar\partial\log(1+|z|^2) &= \frac{z}{1+|z|^2}\,d\bar z,\\[4pt] \partial\bar\partial\log(1+|z|^2) &= \frac{\partial}{\partial z}\!\left(\frac{z}{1+|z|^2}\right)dz\wedge d\bar z = \frac{(1+|z|^2)-z\bar z}{(1+|z|^2)^2}\,dz\wedge d\bar z = \frac{1}{(1+|z|^2)^2}\,dz\wedge d\bar z, \end{align*} where the quotient rule gives $(1+|z|^2 - |z|^2)/(1+|z|^2)^2 = 1/(1+|z|^2)^2$. Writing $z = x+iy$ so that $dz\wedge d\bar z = -2i\,dx\wedge dy$, we get \begin{align*} \omega_{\mathrm{FS}}\big|_L = \frac{i}{2\pi}\cdot\frac{-2i}{(1+|z|^2)^2}\,dx\wedge dy = \frac{1}{\pi(1+|z|^2)^2}\,dx\wedge dy. \end{align*} Passing to polar coordinates $z = re^{i\theta}$ and noting that the missing point $\{Z_0=0\}\cap L$ has measure zero: \begin{align*} \int_{\mathbb{P}^1}\omega_{\mathrm{FS}} = \int_0^\infty\!\int_0^{2\pi}\frac{r}{\pi(1+r^2)^2}\,d\theta\,dr = \int_0^\infty\frac{2r}{(1+r^2)^2}\,dr. \end{align*} Substituting $u = 1+r^2$, $du = 2r\,dr$: \begin{align*} \int_0^\infty\frac{2r}{(1+r^2)^2}\,dr = \int_1^\infty\frac{du}{u^2} = \Bigl[-u^{-1}\Bigr]_1^\infty = 0-(-1) = 1. \end{align*} Since $\int_{\mathbb{P}^1}\omega_{\mathrm{FS}} = 1$ and the generator of $H^2(\mathbb{P}^n,\mathbb{Z})$ is the class that evaluates to $1$ on a projective line, the *Chern-Weil Representative for the First Chern Class* identifies $[\omega_{\mathrm{FS}}]$ as the positive generator $c_1(\mathcal{O}(1))$. The computation makes explicit why $\mathcal{O}(1)$ is positive: its curvature form $\omega_{\mathrm{FS}}$ is a Kähler form, and its cohomology class is an integral generator rather than a fraction of one — a property that can already be read off from the single integral above, without knowing the full ring structure of $H^*(\mathbb{P}^n,\mathbb{Z})$. [/example] ## Changing the Hermitian Metric The next problem is to understand how much of the curvature form depends on the metric. Comparing arbitrary local potentials would not be enough, because local differences need not glue to a global function. Two genuine Hermitian metrics do give a global positive ratio, and its logarithm is exactly the missing global potential. Different metrics produce different differential forms, but the theorem above predicts that their cohomology classes must agree. The precise difference is a global $\partial\bar\partial$-term. [quotetheorem:3849] [citeproof:3849] This formula is the analytic origin of many later arguments. Positivity is a statement about finding a representative with a sign condition, while the Chern class records the fixed cohomology class in which all such representatives live. The word global is essential: for two smooth Hermitian metrics on the same line bundle, the ratio of the metrics is a globally defined positive function, so its logarithm gives a globally defined $\partial\bar\partial$ correction. Thus changing the metric changes the curvature representative, but not the underlying first Chern class. [example: Canonical Bundle of Projective Space] [claim]Let $K_{\mathbb{P}^n} = \Lambda^n T^*\mathbb{P}^n$ be the canonical bundle of projective space. There is a holomorphic bundle isomorphism $K_{\mathbb{P}^n} \cong \mathcal{O}_{\mathbb{P}^n}(-n-1)$, and \begin{align*} c_1(K_{\mathbb{P}^n}) = -(n+1)[\omega_{\mathrm{FS}}] \in H^2(\mathbb{P}^n,\mathbb{Z}). \end{align*}[/claim] *Transition functions of $K_{\mathbb{P}^n}$.* On the affine chart $U_0 = \{Z_0 \neq 0\}$, the functions $z_k = Z_k/Z_0$ for $k = 1,\ldots,n$ are holomorphic coordinates, and the local $(n,0)$-frame for $K_{\mathbb{P}^n}$ is $dz_1 \wedge \cdots \wedge dz_n$. On $U_j = \{Z_j \neq 0\}$ for $j \geq 1$, the coordinates are $w_0 = Z_0/Z_j$ and $w_k = Z_k/Z_j$ for $k \neq j$, with local frame $dw_0 \wedge \cdots \wedge \widehat{dw_j} \wedge \cdots \wedge dw_n$. The transition function of $K_{\mathbb{P}^n}$ on $U_0 \cap U_j$ is the Jacobian determinant $g_{0j}^K = \det\bigl[\partial(w_0,w_1,\ldots,\hat{w}_j,\ldots,w_n)/\partial(z_1,\ldots,z_n)\bigr]$. On $U_0 \cap U_j$ the coordinate change is $w_0 = 1/z_j$ and $w_k = z_k/z_j$ for $k \neq j$, so the partial derivatives are \begin{align*} \frac{\partial w_0}{\partial z_l} = -\frac{\delta_{lj}}{z_j^2}, \qquad \frac{\partial w_k}{\partial z_l} = \frac{\delta_{kl}}{z_j} - \frac{z_k\,\delta_{lj}}{z_j^2} \quad (k \neq 0,j). \end{align*} Column $l \neq j$ of the Jacobian matrix $J$ therefore has exactly one nonzero entry: $1/z_j$ in row $w_l$, and zeros elsewhere. Column $j$ has $-1/z_j^2$ in row $w_0$ and $-z_k/z_j^2$ in row $w_k$ for each $k \neq 0,j$. Factor $1/z_j$ from each of the $n-1$ columns with $l \neq j$, and $-1/z_j^2$ from column $j$: \begin{align*} \det J = \left(\frac{1}{z_j}\right)^{n-1} \cdot \left(-\frac{1}{z_j^2}\right) \cdot \det M, \end{align*} where $M$ is the $n \times n$ matrix whose columns $l \neq j$ are the standard basis vector $e_{w_l}$ (the unit entry is in row $w_l$, all others zero), and whose column $j$ has $1$ in row $w_0$ and $z_k$ in row $w_k$ for $k \neq 0,j$. Expand $\det M$ by cofactors along the $n-1$ standard-basis columns: each column $l \neq j$ locks row $w_l$ with value $1$ and eliminates it from the remaining minor. The only unaccounted row is $w_0$, paired with column $j$; the entry there is $1$, and it carries cofactor sign $(-1)^{1+j}$. The remaining $(n-1)\times(n-1)$ minor (rows $\{w_k : k \neq 0\}$, columns $\{z_l : l \neq j\}$, both indexed in the same order with $j$ skipped) is the identity matrix, with determinant $1$. Therefore $\det M = (-1)^{1+j}\cdot 1 \cdot 1 = (-1)^{j+1}$, and \begin{align*} \det J = z_j^{-(n-1)} \cdot (-z_j^{-2}) \cdot (-1)^{j+1} = (-1)^{j+2}\,z_j^{-(n+1)} = (-1)^j\,z_j^{-(n+1)}. \end{align*} Thus $g_{0j}^K = (-1)^j(Z_0/Z_j)^{n+1}$ on $U_0 \cap U_j$. For general $U_i \cap U_j$, the cocycle identity $g_{0i}^K g_{ij}^K g_{j0}^K = 1$ gives $g_{ij}^K = g_{0j}^K/g_{0i}^K$, so \begin{align*} g_{ij}^K = \frac{(-1)^j(Z_0/Z_j)^{n+1}}{(-1)^i(Z_0/Z_i)^{n+1}} = (-1)^{j-i}\left(\frac{Z_i}{Z_j}\right)^{n+1}. \end{align*} The transition functions of $\mathcal{O}(-(n+1))$ on $U_i \cap U_j$ are $(Z_i/Z_j)^{n+1}$. Define holomorphic units $\phi_i = (-1)^i \in \mathcal{O}^*(U_i)$; then on every $U_i \cap U_j$, \begin{align*} \frac{\phi_j}{\phi_i} \cdot g_{ij}^{-(n+1)} = (-1)^{j-i}\left(\frac{Z_i}{Z_j}\right)^{n+1} = g_{ij}^K. \end{align*} The collection $(\phi_i)$ therefore defines a holomorphic bundle isomorphism $K_{\mathbb{P}^n} \xrightarrow{\;\sim\;} \mathcal{O}(-(n+1))$. *Confirmation via the Euler sequence.* The same isomorphism class follows from the exact sequence of holomorphic vector bundles \begin{align*} 0 \longrightarrow \mathcal{O}_{\mathbb{P}^n} \xrightarrow{\;\phi\;} \mathcal{O}_{\mathbb{P}^n}(1)^{\oplus(n+1)} \xrightarrow{\;\psi\;} T\mathbb{P}^n \longrightarrow 0, \end{align*} where $\phi$ sends $f \in \mathcal{O}(U_i)$ to $(fZ_0/Z_i,\ldots,fZ_n/Z_i)$, which is a section of $\mathcal{O}(1)^{n+1}$ because each $Z_k/Z_i$ is a holomorphic local section of $\mathcal{O}(1)$. The map $\psi$ sends an $(n+1)$-tuple of sections of $\mathcal{O}(1)$ to the corresponding linear combination of coordinate vector fields modulo the radial Euler direction $\sum_k Z_k\,\partial/\partial Z_k$; the kernel of $\psi$ is exactly the span of $(Z_0/Z_i,\ldots,Z_n/Z_i)$, which is the image of $\phi$, so the sequence is exact. For any short exact sequence $0 \to E' \to E \to E'' \to 0$ of holomorphic vector bundles, the natural isomorphism $\det E \cong \det E' \otimes \det E''$ and the identity $c_1(L_1 \otimes L_2) = c_1(L_1) + c_1(L_2)$ from the Transition Function Formula for First Chern Class give $c_1(E) = c_1(E') + c_1(E'')$. Applied here: \begin{align*} c_1\!\bigl(\mathcal{O}(1)^{\oplus(n+1)}\bigr) = c_1(\mathcal{O}) + c_1(T\mathbb{P}^n). \end{align*} The left side equals $c_1\bigl(\det(\mathcal{O}(1)^{\oplus(n+1)})\bigr) = c_1(\mathcal{O}(1)^{\otimes(n+1)}) = (n+1)c_1(\mathcal{O}(1))$, applying the tensor product formula $n$ times. The product bundle $\mathcal{O}$ has $c_1(\mathcal{O})=0$ because its transition functions are all $1$, giving an identically zero integer Čech cocycle. Therefore $c_1(T\mathbb{P}^n) = (n+1)c_1(\mathcal{O}(1))$. Since $K_{\mathbb{P}^n} = \det(T^*\mathbb{P}^n) = (\det T\mathbb{P}^n)^*$, the formula $c_1(L^*)=-c_1(L)$ from the Transition Function Formula for First Chern Class gives \begin{align*} c_1(K_{\mathbb{P}^n}) = -c_1(\det T\mathbb{P}^n) = -(n+1)\,c_1(\mathcal{O}(1)), \end{align*} in agreement with the Jacobian computation. *Curvature representative.* Equip $\mathcal{O}(-1) = \mathcal{O}(1)^*$ with the dual of the Fubini-Study metric on $\mathcal{O}(1)$, and $\mathcal{O}(-(n+1)) = \mathcal{O}(-1)^{\otimes(n+1)}$ with the induced tensor product metric. On $U_0$, the local frame $e_0$ of $\mathcal{O}(1)$ has squared norm $|e_0|^2_{h_{\mathrm{FS}}} = e^{-\varphi_0}$ with $\varphi_0 = \log(1+|z_1|^2+\cdots+|z_n|^2)$. The dual frame $e_0^* = e_0^{-1}$ satisfies $|e_0^*|^2_{h^*} = (|e_0|^2_{h_{\mathrm{FS}}})^{-1} = e^{\varphi_0} = e^{-(-\varphi_0)}$, so the local weight of $\mathcal{O}(-1)$ is $-\varphi_0$. Weights add under tensor product, so the local weight of $\mathcal{O}(-(n+1))$ on $U_0$ is \begin{align*} \varphi_{-(n+1)} = (n+1)\cdot(-\varphi_0) = -(n+1)\varphi_0. \end{align*} The Chern form on $U_0$ is therefore \begin{align*} c_1\!\bigl(\mathcal{O}(-(n+1)),h\bigr)\big|_{U_0} = \frac{i}{2\pi}\partial\bar\partial\bigl(-(n+1)\varphi_0\bigr) = -(n+1)\cdot\frac{i}{2\pi}\partial\bar\partial\log(1+|z|^2) = -(n+1)\,\omega_{\mathrm{FS}}, \end{align*} where the last equality is the local computation from the Hyperplane Divisor example above. The Chern-Weil Representative for the First Chern Class identifies this closed real $(1,1)$-form as the de Rham image of $c_1(\mathcal{O}(-(n+1)))$, and the Metric Change Formula ensures the cohomology class $-(n+1)[\omega_{\mathrm{FS}}]$ is independent of the choice of Hermitian metric on $K_{\mathbb{P}^n}$. The negativity of $c_1(K_{\mathbb{P}^n})$ is what makes $\mathbb{P}^n$ a Fano manifold: the anticanonical bundle $K_{\mathbb{P}^n}^* \cong \mathcal{O}(n+1)$ is ample, with curvature form $(n+1)\omega_{\mathrm{FS}}$ — a $(n+1)$-fold amplification of the positivity already present in $\mathcal{O}(1)$. This contrasts sharply with compact Riemann surfaces of genus $g \geq 2$, where Gauss-Bonnet forces $c_1(K)$ to be strictly positive. [/example] ## Divisors and the Poincare-Lelong Formula The problem is to connect cohomology classes of line bundles with the zeros and poles of holomorphic or meromorphic objects. Counting zeros pointwise is not stable enough, because a zero set may be singular and multiplicities matter. A smooth potential would miss the mass concentrated along the zero or pole set, so the logarithmic singularity of $\log|f|^2$ is the right analytic object. A divisor records codimension-one vanishing data, while a curvature form records a smooth cohomology representative. The Poincare-Lelong formula is the bridge: it says that the logarithmic singularity of a holomorphic function has curvature equal to integration over its zero set. [illustration:scv-iv-divisor-current] [definition: Divisor] Let $M$ be a complex manifold. A divisor on $M$ is a locally finite formal sum \begin{align*} D=\sum_{\nu} m_\nu Y_\nu, \end{align*} where each $m_\nu \in \mathbb Z$ and each $Y_\nu \subset M$ is an irreducible analytic hypersurface. A divisor is effective when $m_\nu\ge0$ for every $\nu$. A local equation for $D$ on an open set $U$ is a nonzero meromorphic function $f_U$ whose order along each $Y_\nu\cap U$ is $m_\nu$ and whose divisor on $U$ is $D|_U$. [/definition] The support of $D$ is the union of the hypersurfaces with nonzero coefficient. Local finiteness means that on every compact subset only finitely many $Y_\nu$ have nontrivial coefficient, so the formal sum makes sense without convergence questions. Since the hypersurfaces $Y_\nu$ may be singular, pointwise integration is taken over the smooth locus and extends across the singular set as a current of locally finite mass. The local-equation viewpoint will be the convenient one for connecting divisors to line bundles, while the current viewpoint is what couples to differential forms in the Poincare-Lelong identity below. [definition: Divisor Current] Let $M$ have complex dimension $n$, and let $D=\sum_\nu m_\nu Y_\nu$ be a divisor. The current of integration $[D]$ is the $(1,1)$-current acting on compactly supported smooth $(n-1,n-1)$-forms $\eta$ by \begin{align*} [D](\eta)=\sum_\nu m_\nu\int_{Y_\nu^{\mathrm{reg}}}\eta. \end{align*} [/definition] For an effective divisor, $[D]$ is a positive closed current; closedness follows from Stokes' theorem on the smooth locus together with the fact that the singular locus of an analytic hypersurface has complex codimension at least two inside the hypersurface, so it carries no mass for an $(n-1,n-1)$-form. In local coordinates, the model case is the divisor of $z_1^m$ on a polydisc, where the current is $m$ times integration over the hypersurface $z_1=0$. The next analytic problem is to recover this integration current directly from a local defining equation. Divisors are recorded by orders of vanishing, while curvature and cohomology are expressed by differential forms and currents, so one needs an identity that converts the logarithmic singularity of a meromorphic function into the current carried by its zero and pole hypersurfaces. [quotetheorem:3850] [citeproof:3850] The formula turns divisors into curvature representatives with singular potentials. It is the local analytic reason why the cohomology class of a divisor agrees with the first Chern class of its associated line bundle. The nonzero hypothesis prevents a degenerate case: the identically zero meromorphic function or zero section has no divisor with finite multiplicities, so it cannot define a useful current identity. If $f$ is smooth and nowhere zero, the right-hand side is zero and the logarithmic potential contributes no divisor current; the new content appears exactly when zeros or poles create logarithmic singularities. For a non-reduced divisor, multiplicities are recorded by the coefficients $m_\nu$, but more general non-reduced subscheme structure, such as embedded or nilpotent data, is not seen by the integration current alone. The extension across the singular locus is legitimate because the singular set of an analytic hypersurface has complex codimension at least two in $M$, and the integration current over the smooth locus has a unique closed extension with locally finite mass. [example: Curvature Computation for a Divisor Line Bundle] [claim] Let $D$ be an effective divisor with local holomorphic equations $f_i$ on a cover $(U_i)$, let $h$ be a smooth Hermitian metric on $\mathcal{O}(D)$, and write $|e_i|_h^2 = e^{-\varphi_i}$ for the local weight of the canonical frame $e_i$ over $U_i$. The local Chern form is \begin{align*} c_1(\mathcal{O}(D),h)\big|_{U_i} = \frac{i}{2\pi}\partial\bar\partial\varphi_i, \end{align*} and the following global current identity holds on $M$: \begin{align*} c_1(\mathcal{O}(D),h) = [D] - \frac{i}{2\pi}\partial\bar\partial\log|s_D|_h^2. \end{align*} [/claim] The local formula is the *Chern Form of a Hermitian Line Bundle* definition applied to $\mathcal{O}(D)$ with frame $e_i$ and weight $\varphi_i$. For the current identity, work first over a single $U_i$. The canonical section satisfies $s_D = f_i e_i$, and the squared pointwise norm decomposes as \begin{align*} |s_D|_h^2 = |f_i|^2 \cdot |e_i|_h^2 = |f_i|^2 e^{-\varphi_i}. \end{align*} Taking the real logarithm of both sides: \begin{align*} \log|s_D|_h^2 = \log|f_i|^2 - \varphi_i. \end{align*} Applying $\frac{i}{2\pi}\partial\bar\partial$ and using linearity of $\partial\bar\partial$: \begin{align*} \frac{i}{2\pi}\partial\bar\partial\log|s_D|_h^2 = \frac{i}{2\pi}\partial\bar\partial\log|f_i|^2 - \frac{i}{2\pi}\partial\bar\partial\varphi_i. \end{align*} Since $f_i$ is a local holomorphic equation for $D$ on $U_i$, the divisor of $f_i$ on $U_i$ is $D|_{U_i}$. The *Poincaré–Lelong Formula* applied to the holomorphic function $f_i$ gives $\frac{i}{2\pi}\partial\bar\partial\log|f_i|^2 = [D]|_{U_i}$ as currents. Substituting: \begin{align*} \frac{i}{2\pi}\partial\bar\partial\log|s_D|_h^2 = [D]\big|_{U_i} - \frac{i}{2\pi}\partial\bar\partial\varphi_i = [D]\big|_{U_i} - c_1(\mathcal{O}(D),h)\big|_{U_i}. \end{align*} Rearranging yields the identity on $U_i$: \begin{align*} c_1(\mathcal{O}(D),h)\big|_{U_i} = [D]\big|_{U_i} - \frac{i}{2\pi}\partial\bar\partial\log|s_D|_h^2\big|_{U_i}. \end{align*} It remains to show that both sides are globally well-defined, so that these local equalities assemble into a single current equation on $M$. On $U_i \cap U_j$, the transition function of $\mathcal{O}(D)$ is $g_{ij} = f_i/f_j$, holomorphic and nowhere zero, and the Hermitian weights transform as $\varphi_j = \varphi_i - \log|g_{ij}|^2$. Since $g_{ij}$ is holomorphic and nowhere zero, $\log|g_{ij}|^2 = \operatorname{Re}(\log g_{ij})$ is pluriharmonic on $U_i \cap U_j$, so $\partial\bar\partial\log|g_{ij}|^2 = 0$. Applying $\frac{i}{2\pi}\partial\bar\partial$ to the weight relation gives \begin{align*} \frac{i}{2\pi}\partial\bar\partial\varphi_j = \frac{i}{2\pi}\partial\bar\partial\bigl(\varphi_i - \log|g_{ij}|^2\bigr) = \frac{i}{2\pi}\partial\bar\partial\varphi_i \quad \text{on } U_i \cap U_j, \end{align*} so the local Chern forms patch to a globally defined smooth $(1,1)$-form. For the term $\log|s_D|_h^2$, comparing the two local expressions on $U_i \cap U_j$: \begin{align*} \log|f_j|^2 - \varphi_j = \log|f_j|^2 - \varphi_i + \log|g_{ij}|^2 = \log\!\bigl|f_j \cdot \tfrac{f_i}{f_j}\bigr|^2 - \varphi_i = \log|f_i|^2 - \varphi_i, \end{align*} where in the last step $f_j \cdot g_{ij} = f_j \cdot (f_i/f_j) = f_i$. Hence $\log|s_D|_h^2$ takes the same value whether computed from $U_i$ or $U_j$, so it is a globally defined function on $M \setminus \operatorname{supp}(D)$ with logarithmic singularities along $\operatorname{supp}(D)$, and $\frac{i}{2\pi}\partial\bar\partial\log|s_D|_h^2$ is a globally defined current. The local rearrangements therefore assemble into the claimed current identity on all of $M$. Rearranged, the identity says that the smooth Chern form $c_1(\mathcal{O}(D), h)$ and the singular integration current $[D]$ lie in the same de Rham cohomology class, differing only by the exact current $\frac{i}{2\pi}\partial\bar\partial\log|s_D|_h^2$ — a concrete instance of the *Chern-Weil Representative for the First Chern Class* and the mechanism by which analytic singularities encode topological data. [/example] ## Divisors as Line Bundles The final question in this section is how much of a divisor is remembered by a line bundle. A divisor has local equations, and ratios of local equations are nowhere-zero holomorphic functions on overlaps. This works only because the local equations describe the same divisor with the same orders along every hypersurface; if the orders did not match, the quotient would have a zero or pole and would fail to be an invertible holomorphic transition function. Those ratios are transition functions, so every divisor produces a holomorphic line bundle with a distinguished meromorphic section. [definition: Associated Line Bundle of a Divisor] Let $D$ be a divisor on a complex manifold $M$, and let $(f_i)$ be local meromorphic equations for $D$ on a cover $(U_i)$. The line bundle $\mathcal O(D)$ has local frames $e_i$ and transition rule \begin{align*} e_j=\frac{f_i}{f_j}e_i \end{align*} on $U_i\cap U_j$. The canonical meromorphic section $s_D$ is given on $U_i$ by \begin{align*} s_D=f_i e_i. \end{align*} [/definition] The quotient $f_i/f_j$ is holomorphic and nowhere zero because the two local equations encode the same orders along every hypersurface; any failure of this matching would produce a residual zero or pole, and the quotient would no longer be a holomorphic transition function. The cocycle condition on triple overlaps is automatic because $(f_i/f_j)(f_j/f_k)(f_k/f_i)=1$. For an effective divisor, the section $s_D$ is holomorphic and vanishes exactly along $D$ with the prescribed multiplicities. For a general divisor, the same formula gives a meromorphic section whose zero divisor minus its pole divisor is $D$. This construction raises a global classification question: how much of a line bundle is captured by a divisor and its canonical section? The answer must distinguish the bundle alone from the pair consisting of a bundle and a meromorphic or holomorphic section, since different sections of one bundle can cut out different divisors. [quotetheorem:3851] [citeproof:3851] The compactness hypothesis controls the surjectivity onto $\operatorname{Pic}(M)$ and keeps the divisor-line-bundle dictionary in the intended setting. Even on a compact manifold, the bijection in the effective case is between divisors and pairs $(L,s)$, not between divisors and bundles alone. Two different effective divisors $D_1\ne D_2$ can have $\mathcal O(D_1)\cong \mathcal O(D_2)$ while the sections $s_{D_1}$ and $s_{D_2}$ are linearly independent in $H^0(M,L)$; this is exactly the phenomenon that organises divisors inside a fixed line bundle into the linear system $|L|=\mathbb P H^0(M,L)$. On projective manifolds and compact Riemann surfaces, many line bundles arise from divisors, linear systems organise effective divisors inside a fixed line bundle, and the closely related notion of ampleness in later sections detects positivity by requiring sufficiently many sections in tensor powers $L^{\otimes k}$. On a general compact complex manifold, the final image statement is the safe formulation: a line bundle is in the divisor image precisely when it has a nonzero meromorphic section. [example: Principal Divisor] [claim]Let $f$ be a nonzero meromorphic function on a compact complex manifold $M$, and let $D = \operatorname{div}(f)$. Then $\mathcal{O}(D) \cong M \times \mathbb{C}$ as holomorphic line bundles, and $c_1(\mathcal{O}(D)) = 0 \in H^2(M,\mathbb{Z})$.[/claim] Fix an open cover $(U_i)$ of $M$. Since $f$ is globally meromorphic, it restricts to a nonzero meromorphic function $f_i = f|_{U_i}$ on each $U_i$, and these restrictions are compatible: $f_i|_{U_i \cap U_j} = f_j|_{U_i \cap U_j} = f|_{U_i \cap U_j}$. The local equations $(f_i)$ cut out $D$ with the correct multiplicities on each piece, so they constitute a valid system of local equations for $D$ in the sense of the Associated Line Bundle definition. By that definition, the transition functions of $\mathcal{O}(D)$ on each nonempty overlap $U_i \cap U_j$ are \begin{align*} g_{ij} = \frac{f_i}{f_j} = \frac{f|_{U_i}}{f|_{U_j}} = \frac{f}{f} = 1, \end{align*} where the last equality holds because both restrictions are the same global function $f$ evaluated on $U_i \cap U_j$. The transition functions are therefore identically $1$ on every overlap, which is exactly the transition data of the product bundle $M \times \mathbb{C}$ (in any trivialising cover). The identity map on each $U_i$ sending the frame $e_i$ of $\mathcal{O}(D)$ to the constant frame of $M \times \mathbb{C}$ commutes with these transition functions and assembles into a holomorphic bundle isomorphism $\mathcal{O}(D) \cong M \times \mathbb{C}$. To compute $c_1(\mathcal{O}(D))$, apply the *Cech First Chern Class* definition to the transition functions just found. Since $g_{ij} = 1 = \exp(2\pi i \cdot 0)$, the choice $a_{ij} = 0$ satisfies the logarithm condition $\exp(2\pi i\,a_{ij}) = g_{ij}$ on every nonempty overlap. The integer Čech cocycle then evaluates on every triple overlap $U_i \cap U_j \cap U_k$ as \begin{align*} n_{ijk} = a_{ij} + a_{jk} + a_{ki} = 0 + 0 + 0 = 0. \end{align*} The cocycle $(n_{ijk})$ is identically zero, so it represents the zero class in $H^2(M,\mathbb{Z})$, giving $c_1(\mathcal{O}(D)) = 0$. The result is also consistent with the *Poincaré–Lelong Formula*, which gives the current identity $\frac{i}{2\pi}\partial\bar\partial\log|f|^2 = [D]$. The left side is $d$-exact: one has $d\!\left(\frac{i}{2\pi}\bar\partial\log|f|^2\right) = \frac{i}{2\pi}\partial\bar\partial\log|f|^2$, using $\bar\partial^2 = 0$, so $[D]$ represents the zero class in $H^2_{\mathrm{dR}}(M,\mathbb{R})$. This is the real image of $c_1(\mathcal{O}(D)) = 0$, in agreement with the *Chern-Weil Representative for the First Chern Class*. The example illustrates the algebraic core of the divisor-to-bundle correspondence: any divisor cut out by a single global meromorphic function has associated bundle isomorphic to $M \times \mathbb{C}$, because the transition ratios $f_i/f_j$ collapse to $1$ when both numerator and denominator are the same global object. [/example] Sections 1–3 developed Hermitian metrics on holomorphic bundles, curvature theory, and the Chern class machinery connecting local curvature to global topology. Section 4 now places a Hermitian metric on the complex manifold itself—introducing Kähler metrics—and provides the Hodge-theoretic framework that will enable the analytic estimates of Section 5 and beyond. # 4. Kahler Metrics and Hodge-Theoretic Background This section supplies the metric and Hodge-theoretic background used in the Bochner-Kodaira-Nakano and Lefschetz arguments of Sections 5 and 8. The previous sections treated Hermitian metrics on holomorphic vector bundles and positivity of line bundles; here the metric is placed on the complex manifold itself. The central point is that the Kähler condition turns Hermitian linear algebra into global cohomological structure, making analytic operators such as $\partial$, $\bar\partial$, and $d$ compatible in a way that is special to Kähler geometry. The section moves from local geometry to global consequences. We first attach a real $(1,1)$-form to a Hermitian metric, then isolate the closedness condition that makes the metric Kähler. After that we record the Kähler identities, the Hodge decomposition used later in the course, and the Lefschetz package needed to interpret positivity cohomologically. ## Measuring Complex Tangent Directions What does it mean to measure lengths and angles on a complex manifold without forgetting the complex structure? A Riemannian metric on the underlying real manifold is not enough by itself: it must interact with multiplication by $i$ on tangent vectors. Hermitian metrics provide the correct notion because they measure vectors in $T^{1,0}X$ and produce forms of type $(1,1)$. [definition: Hermitian Metric] Let $X$ be a complex manifold. A Hermitian metric on $X$ is a smoothly varying family $h=(h_p)_{p\in X}$ such that each \begin{align*} h_p:T_p^{1,0}X\times T_p^{1,0}X\to \mathbb C \end{align*} is a positive-definite Hermitian inner product, conjugate-linear in the first variable and linear in the second variable, as in the bundle convention fixed in Section 1. [/definition] In local holomorphic coordinates $z=(z_1,\dots,z_n)$, the metric is represented by a positive-definite Hermitian matrix $(h_{j\bar k})$. The complex structure is built into this matrix representation: the barred index records conjugate type, not a second independent coordinate. The form attached to the metric is often more important than the metric tensor itself, because it can be differentiated and integrated. In complex geometry, this associated $(1,1)$-form is the object that appears in volume calculations, positivity statements, and the Kähler condition. Defining it now gives a way to translate the Hermitian inner product on tangent vectors into the language of differential forms. [definition: Fundamental Form] Let $h$ be a Hermitian metric on an $n$-dimensional complex manifold $X$. The fundamental form of $h$ is the real $(1,1)$-form $\omega_h$ locally given by \begin{align*} \omega_h=i\sum_{j,k=1}^{n}h_{j\bar k}\,dz_j\wedge d\bar z_k. \end{align*} [/definition] The positivity of $h$ is exactly positivity of the coefficient matrix of $\omega_h$. Thus a Hermitian metric can be studied through a positive real $(1,1)$-form. Later arguments often begin with a closed real $(1,1)$-form and ask whether it can play the role of a Kähler form, without first naming a metric tensor. For that reason positivity has to be stated intrinsically for forms themselves: the definition below records the coordinate-invariant matrix condition that lets such a form measure positive length and volume in complex directions. [definition: Positive Real One One Form] A real $(1,1)$-form $\omega$ on a complex manifold $X$ is positive if, in every holomorphic coordinate chart, it can be written as \begin{align*} \omega=i\sum_{j,k=1}^{n}a_{j\bar k}\,dz_j\wedge d\bar z_k \end{align*} with $(a_{j\bar k}(p))$ positive definite at every point $p$ in the chart. [/definition] The determinant of the coefficient matrix controls the volume form. If $\omega$ is positive on an $n$-fold, then $\omega^n$ is a positive multiple of the real volume form determined by the Hermitian metric. [example: Standard Hermitian Metric on Complex Euclidean Space] On $\mathbb{C}^n$ with coordinates $z_1,\dots,z_n$, we take $h_{j\bar{k}}=\delta_{jk}$ and show that the associated fundamental form is $\omega_0=i\sum_{j=1}^{n}dz_j\wedge d\bar{z}_j$, that it is positive, and that $d\omega_0=0$. *Identifying the form.* Substituting $h_{j\bar{k}}=\delta_{jk}$ into the definition of the fundamental form gives \begin{align*} \omega_0 = i\sum_{j,k=1}^{n}\delta_{jk}\,dz_j\wedge d\bar{z}_k. \end{align*} Since $\delta_{jk}=0$ whenever $j\ne k$ and $\delta_{jk}=1$ whenever $j=k$, every off-diagonal term vanishes and the sum reduces to \begin{align*} \omega_0 = i\sum_{j=1}^{n}dz_j\wedge d\bar{z}_j. \end{align*} *Positivity.* The coefficient matrix of $\omega_0$ is $(\delta_{jk})=I_n$. For any nonzero $v=(v_1,\dots,v_n)\in\mathbb{C}^n$, \begin{align*} \sum_{j,k=1}^{n}\delta_{jk}\,v_j\bar{v}_k = \sum_{j=1}^{n}|v_j|^2 > 0, \end{align*} so $I_n$ is positive definite at every point of $\mathbb{C}^n$. By the definition of positive real $(1,1)$-forms, $\omega_0$ is positive. *Closedness.* Write $\omega_0=\sum_{j=1}^{n}f_j\,dz_j\wedge d\bar{z}_j$ with $f_j=i$ for every $j$. By the [Coordinate Formula for the Exterior Derivative](/theorems/3564), \begin{align*} d\omega_0 = \sum_{j=1}^{n}df_j\wedge dz_j\wedge d\bar{z}_j. \end{align*} Each $f_j=i$ is constant on $\mathbb{C}^n$, so $df_j=0$, and therefore $d\omega_0=0$. The pair $(\mathbb{C}^n,\omega_0)$ is accordingly a Kähler manifold, and it is the local model to which every adapted coordinate calculation in Kähler geometry ultimately returns. [/example] This example is the local model for Hermitian geometry, but its closedness is the feature that survives into Kähler geometry. A general Hermitian metric has positive coefficients, yet those coefficients need not satisfy any differential relation. ## The Kähler Condition and Local Potentials Which Hermitian metrics have enough rigidity to support cohomological arguments? The answer is to require the fundamental form to be closed. This condition looks like a first-order differential equation, but locally it is equivalent to the existence of a scalar potential whose complex Hessian gives the metric. [definition: Kahler Metric] Let $X$ be a complex manifold and let $h$ be a Hermitian metric with fundamental form $\omega_h$. The metric $h$ is Kähler if \begin{align*} d\omega_h=0. \end{align*} The pair $(X,\omega_h)$ is a Kähler manifold. [/definition] The definition is local in the sense that closedness can be checked in holomorphic coordinate charts, but its consequences are global. The closed form $\omega_h$ defines a de Rham cohomology class, and that class later records positivity information about the manifold. [quotetheorem:3852] [citeproof:3852] Potentials connect Kähler geometry with plurisubharmonic functions. The closedness assumption is essential here: it is what permits the local expression $\omega=i\partial\bar\partial\varphi$ rather than merely a positive coefficient matrix. On a general Hermitian manifold the fundamental form need not be closed, so there need not be local Kähler potentials, and the later $\partial\bar\partial$-lemma cannot be expected in this form. Globally, potentials are usually only local; on compact manifolds the obstruction to gluing them is precisely part of the cohomological content of the Kähler class. [example: Fubini-Study Metric on Projective Space] On the affine chart $U_0=\{[Z_0:\cdots:Z_n]\in\mathbb{P}^n:Z_0\ne0\}$, write $z_j=Z_j/Z_0$ for $j=1,\dots,n$ and set $\varphi_0(z)=\log\!\bigl(1+\sum_{j=1}^n|z_j|^2\bigr)$; on each remaining chart $U_\ell=\{Z_\ell\ne0\}$ define $\varphi_\ell$ by the same formula in coordinates $Z_j/Z_\ell$. [claim]The form $\Omega_{\mathrm{FS}}=i\partial\bar\partial\varphi_0$ defines a global Kähler form on $\mathbb{P}^n$.[/claim] Set $\rho=1+\sum_{j=1}^n|z_j|^2$ throughout $U_0$. Applying $\bar\partial$ to $\varphi_0=\log\rho$ by the chain rule, \begin{align*} \bar\partial\varphi_0 = \frac{1}{\rho}\,\bar\partial\rho = \frac{1}{\rho}\sum_{k=1}^n z_k\,d\bar z_k, \end{align*} since $\bar\partial|z_k|^2=\bar\partial(z_k\bar z_k)=z_k\,d\bar z_k$. Applying $\partial$ to this $(0,1)$-form by the product rule, \begin{align*} \partial\bar\partial\varphi_0 = \bigl(\partial\tfrac{1}{\rho}\bigr)\wedge\sum_{k=1}^n z_k\,d\bar z_k +\frac{1}{\rho}\,\partial\!\sum_{k=1}^n z_k\,d\bar z_k. \end{align*} The two factors are computed separately. Since $\partial\rho=\sum_j\bar z_j\,dz_j$, one has $\partial(1/\rho)=-\rho^{-2}\partial\rho=-\rho^{-2}\sum_j\bar z_j\,dz_j$. Since $d\bar z_k$ is a constant-coefficient form, $\partial(z_k\,d\bar z_k)=dz_k\wedge d\bar z_k$. Substituting both: \begin{align*} \partial\bar\partial\varphi_0 = \frac{1}{\rho}\sum_{k=1}^n dz_k\wedge d\bar z_k -\frac{1}{\rho^2}\sum_{j,k=1}^n \bar z_j z_k\,dz_j\wedge d\bar z_k. \end{align*} The coefficient matrix of $\Omega_{\mathrm{FS}}=i\partial\bar\partial\varphi_0$ is therefore \begin{align*} h_{j\bar k} = \frac{\delta_{jk}}{\rho} - \frac{\bar z_j z_k}{\rho^2}. \end{align*} *Positivity.* For any nonzero $v=(v_1,\dots,v_n)\in\mathbb{C}^n$, expanding the Hermitian form gives \begin{align*} \sum_{j,k=1}^n h_{j\bar k}\,v_j\bar v_k = \frac{1}{\rho}\sum_j|v_j|^2 - \frac{1}{\rho^2}\Bigl(\sum_j\bar z_j v_j\Bigr)\Bigl(\sum_k z_k\bar v_k\Bigr) = \frac{|v|^2}{\rho} - \frac{\bigl|\sum_j\bar z_j v_j\bigr|^2}{\rho^2}. \end{align*} By [Cauchy-Schwarz](/theorems/432), $|\sum_j\bar z_j v_j|^2\le|z|^2|v|^2$, and since $\rho=1+|z|^2$ one has $|z|^2=\rho-1$, so \begin{align*} \sum_{j,k=1}^n h_{j\bar k}\,v_j\bar v_k \ge \frac{|v|^2}{\rho} - \frac{|z|^2|v|^2}{\rho^2} = \frac{|v|^2(\rho-|z|^2)}{\rho^2} = \frac{|v|^2}{\rho^2} > 0. \end{align*} Thus $(h_{j\bar k})$ is positive definite at every point of $U_0$, so $\Omega_{\mathrm{FS}}$ is a positive $(1,1)$-form on $U_0$. *Closedness.* Writing $d=\partial+\bar\partial$ and expanding, \begin{align*} d\Omega_{\mathrm{FS}} = i\,\partial(\partial\bar\partial\varphi_0) + i\,\bar\partial(\partial\bar\partial\varphi_0). \end{align*} The first term is $\partial(\partial(\bar\partial\varphi_0))=\partial^2(\bar\partial\varphi_0)=0$, since $\partial^2=0$ is the $(2,0)$-component of $d^2=0$. For the second term, the identity $\bar\partial\partial=-\partial\bar\partial$ (the mixed-type component of $d^2=0$) gives $\bar\partial(\partial\bar\partial\varphi_0)=(\bar\partial\partial)\bar\partial\varphi_0=-\partial(\bar\partial^2\varphi_0)=0$ by [$\bar\partial^2=0$](/theorems/3409). Hence $d\Omega_{\mathrm{FS}}=0$. *Global well-definedness.* On the overlap $U_0\cap U_\ell$ with $\ell\ge1$, rewrite both potentials in homogeneous coordinates: \begin{align*} \varphi_0 = \log\!\Bigl(\sum_{k=0}^n|Z_k|^2\Bigr) - \log|Z_0|^2, \qquad \varphi_\ell = \log\!\Bigl(\sum_{k=0}^n|Z_k|^2\Bigr) - \log|Z_\ell|^2. \end{align*} Their difference is $\varphi_0-\varphi_\ell=\log|Z_\ell/Z_0|^2=\log|z_\ell|^2$. Since $z_\ell\ne0$ on $U_0\cap U_\ell$, choose a branch of the holomorphic logarithm there; then $\log|z_\ell|^2=\log z_\ell+\log\bar z_\ell=2\,\mathrm{Re}(\log z_\ell)$. Because $\log z_\ell$ is holomorphic, $\bar\partial(\log z_\ell)=0$, so $\partial\bar\partial(\log z_\ell)=\partial(0)=0$. Because $\log\bar z_\ell$ is antiholomorphic, $\partial(\log\bar z_\ell)=0$, so $\partial\bar\partial(\log\bar z_\ell)=0$. Adding, \begin{align*} i\partial\bar\partial(\varphi_0-\varphi_\ell) = i\partial\bar\partial(\log z_\ell+\log\bar z_\ell) = 0, \end{align*} so $i\partial\bar\partial\varphi_0=i\partial\bar\partial\varphi_\ell$ on every pairwise overlap, and the local forms assemble into a single smooth global $(1,1)$-form. The form $\Omega_{\mathrm{FS}}$ is the distinguished positive Kähler form on $\mathbb{P}^n$ and is the curvature form $i\Theta(\mathcal{O}(1))$ for the hyperplane line bundle metric computed earlier. Its normalized companion $\omega_{\mathrm{FS}}=\frac{1}{2\pi}\Omega_{\mathrm{FS}}$ represents $c_1(\mathcal{O}(1))$, so the metric geometry of projective space and the positivity of the hyperplane bundle encode the same data with different normalizations. [/example] The Fubini-Study metric is the model positive metric on projective space. It will later be pulled back along holomorphic maps and compared with curvature forms of positive line bundles. [example: Complex Tori] Let $\Lambda \subset \mathbb{C}^n$ be a lattice and $X = \mathbb{C}^n/\Lambda$ with covering projection $\pi: \mathbb{C}^n \to X$. Fix a positive-definite Hermitian matrix $(H_{j\bar{k}})$ with constant entries, and let $\omega = i\sum_{j,k=1}^{n} H_{j\bar{k}}\,dz_j \wedge d\bar{z}_k$ on $\mathbb{C}^n$. We show that $\omega$ descends to a Kähler form on $X$. *Descent.* The deck transformations of the covering $\pi$ are the translations $T_\lambda: z \mapsto z + \lambda$ for $\lambda \in \Lambda$. Since $T_\lambda$ is affine with linear part the identity, its derivative at every point is $D(T_\lambda) = \mathrm{Id}$, so pulling back the coordinate one-forms gives $(T_\lambda^* dz_j)_z(v) = (dz_j)_{z+\lambda}(v) = v_j$ and likewise $T_\lambda^* d\bar{z}_k = d\bar{z}_k$. By [Naturality of the Pullback: Compatibility with Wedge Product and Exterior Derivative](/theorems/3574), the pullback distributes over wedge products, so \begin{align*} T_\lambda^*\omega = i\sum_{j,k=1}^{n} H_{j\bar{k}}\,T_\lambda^*(dz_j)\wedge T_\lambda^*(d\bar{z}_k) = i\sum_{j,k=1}^{n} H_{j\bar{k}}\,dz_j \wedge d\bar{z}_k = \omega. \end{align*} Because $T_\lambda^*\omega = \omega$ for every $\lambda \in \Lambda$, the form is $\Lambda$-invariant, and a $\Lambda$-invariant smooth form on $\mathbb{C}^n$ descends uniquely to a smooth form $\omega_X$ on $X$ satisfying $\pi^*\omega_X = \omega$. *Closedness.* Every $H_{j\bar{k}}$ is constant, so $dH_{j\bar{k}} = 0$. By the [Coordinate Formula for the Exterior Derivative](/theorems/3564), \begin{align*} d\omega = i\sum_{j,k=1}^{n} (dH_{j\bar{k}})\wedge dz_j \wedge d\bar{z}_k = 0. \end{align*} Since $\pi^*$ commutes with $d$ by [Naturality of the Pullback: Compatibility with Wedge Product and Exterior Derivative](/theorems/3574), one has $\pi^*(d\omega_X) = d(\pi^*\omega_X) = d\omega = 0$. As $\pi$ is a local diffeomorphism, $\pi^*$ is injective on forms at each point, so $d\omega_X = 0$. *Positivity.* At any point $[z] \in X$, the covering map identifies $T_{[z]}^{1,0}X$ with $\mathbb{C}^n$, and the coefficient matrix of $\omega_X$ at $[z]$ equals $(H_{j\bar{k}})$. For any nonzero $v = (v_1,\dots,v_n) \in \mathbb{C}^n$, positive definiteness of $(H_{j\bar{k}})$ gives \begin{align*} \sum_{j,k=1}^{n} H_{j\bar{k}}\,v_j\bar{v}_k > 0, \end{align*} and since this holds at every point of $X$ with the same constant matrix, $\omega_X$ is a positive real $(1,1)$-form. The Kähler form $\omega_X$ is flat: its pullback to $\mathbb{C}^n$ is a constant-coefficient form, so the geometry of $X$ is everywhere modeled by the same Hermitian pairing — a structural simplicity that distinguishes complex tori from the positively curved Fubini-Study geometry of $\mathbb{P}^n$, even though both are Kähler. [/example] Complex tori show that Kähler geometry is not only projective geometry. Later results impose additional arithmetic conditions when one asks whether a complex torus is an abelian variety. ## Kähler Identities and Harmonic Representatives How does the condition $d\omega=0$ affect differential operators on forms? On a general Hermitian manifold, the operators $d$, $\partial$, and $\bar\partial$ have related definitions but their Laplacians do not align well. On a Kähler manifold, commutator identities force their harmonic theories to agree in the precise way needed for Hodge decomposition. [definition: Dolbeault Laplacian] Let $X$ be a compact Hermitian manifold, and let $A^{p,q}(X)$ denote the space of smooth $(p,q)$-forms on $X$. The operator \begin{align*} \bar\partial:A^{p,q}(X)\to A^{p,q+1}(X) \end{align*} has a formal adjoint \begin{align*} \bar\partial^*:A^{p,q+1}(X)\to A^{p,q}(X) \end{align*} with respect to the $L^2$ inner product induced by the Hermitian metric. The Dolbeault Laplacian in bidegree $(p,q)$ is the operator \begin{align*} \Delta_{\bar\partial}:A^{p,q}(X)\to A^{p,q}(X),\qquad \Delta_{\bar\partial}=\bar\partial\bar\partial^*+\bar\partial^*\bar\partial. \end{align*} [/definition] Similarly, $\Delta_{\partial}=\partial\partial^*+\partial^*\partial$ and $\Delta_d=dd^*+d^*d$. The Kähler identities explain why these three Laplacians have compatible kernels. To state those identities, the Kähler form must be treated not only as a metric object but also as an operator on the algebra of forms. Wedging with $\omega$ shifts degree by two, and its adjoint is the operation that appears in the commutators relating $\partial$, $\bar\partial$, and their adjoints. The definition below fixes this operator notation before it is used in the identities. [definition: Lefschetz Operator] Let $(X,\omega)$ be a Hermitian manifold. The Lefschetz operator is \begin{align*} L:A^k(X)\to A^{k+2}(X),\qquad L(\alpha)=\omega\wedge\alpha. \end{align*} Its formal adjoint is denoted by $\Lambda$. [/definition] The commutators between $L$, $\Lambda$, and the Dolbeault operators are the computational heart of Kähler Hodge theory. They turn geometric information encoded by the form $\omega$ into analytic information about kernels of Laplacians. In practice, these identities are used to replace a problem about de Rham harmonic forms by a problem about $\bar\partial$-harmonic forms, or to prove that a form known to be harmonic for one operator is harmonic for another. This is the mechanism behind Hodge symmetry, the $\partial\bar\partial$-lemma on compact Kähler manifolds, and many later arguments where positivity is converted into cohomological vanishing. [quotetheorem:3853] [citeproof:3853] The equality of Laplacians is why de Rham and Dolbeault cohomology communicate on compact Kähler manifolds. Harmonic representatives can be decomposed by type without leaving the space of harmonic forms. For a general Hermitian metric, torsion terms enter the commutator formulae, so these identities and the resulting equality of Laplacians fail in this clean form. The theorem also does not make the harmonic theory metric-independent: the spaces of harmonic representatives identify with cohomology, but the actual operators and their nonzero eigenvalue spectra still depend on the chosen Kähler metric. Once harmonic forms are compatible with type, the next question is how this analytic fact appears in cohomology itself. The resulting decomposition is the structural statement that lets topological classes be sorted by Dolbeault bidegree on compact Kähler manifolds. [quotetheorem:3854] [citeproof:3854] This is the Hodge decomposition used throughout compact complex geometry: topological cohomology splits into Dolbeault pieces. Its force is partly numerical and partly structural. Numerically, it implies restrictions such as $b_1(X)=2h^{1,0}(X)$, so the first Betti number of a compact Kähler manifold is even. Structurally, it lets later arguments speak interchangeably about topological classes and Dolbeault bidegrees, provided the compact Kähler hypotheses are present. Those hypotheses should not be treated as cosmetic. Compactness is what makes the harmonic spaces finite-dimensional representatives of cohomology, and the Kähler condition is what makes the type decomposition compatible with the Hodge-theoretic picture. This is the obstruction used below to rule out a Kähler metric on the Hopf surface. [example: Hodge Decomposition on a Complex Torus] Let $X = \mathbb{C}^n/\Lambda$ carry the flat Kähler form $\omega_X$ built from a constant positive-definite Hermitian matrix $(H_{j\bar k})$, as in the preceding example. We compute the Dolbeault cohomology of $X$ via harmonic representatives. [claim] For each bidegree $(p,q)$, the group $H^{p,q}_{\bar\partial}(X)$ is represented exactly by the constant-coefficient $(p,q)$-forms $\sum_{|I|=p,\,|J|=q}c_{IJ}\,dz_I\wedge d\bar z_J$ with $c_{IJ}\in\mathbb{C}$. Consequently $h^{p,q}(X)=\binom{n}{p}\binom{n}{q}$. [/claim] *Constant forms are harmonic.* Fix a constant form $\alpha=\sum_{|I|=p,\,|J|=q}c_{IJ}\,dz_I\wedge d\bar z_J$. By the [Coordinate Formula for the Exterior Derivative](/theorems/3564), \begin{align*} \bar\partial\alpha = \sum_{I,J}\sum_{k=1}^n\frac{\partial c_{IJ}}{\partial\bar z_k}\,d\bar z_k\wedge dz_I\wedge d\bar z_J. \end{align*} Each $c_{IJ}$ is constant, so $\frac{\partial c_{IJ}}{\partial\bar z_k}=0$ for every $k$, giving $\bar\partial\alpha=0$. For the adjoint: since the metric coefficients $H_{j\bar k}$ are constant, all Christoffel symbols of the Chern connection vanish and the Chern connection reduces to ordinary partial differentiation. The formal adjoint $\bar\partial^*:A^{p,q}(X)\to A^{p,q-1}(X)$ is therefore \begin{align*} \bar\partial^*\alpha = -\sum_{j,k=1}^n H^{j\bar k}\,\iota_{\partial/\partial\bar z_k}\!\left(\frac{\partial\alpha}{\partial z_j}\right), \end{align*} where $\frac{\partial\alpha}{\partial z_j}=\sum_{I,J}\frac{\partial c_{IJ}}{\partial z_j}\,dz_I\wedge d\bar z_J$ and $(H^{j\bar k})$ is the inverse matrix of $(H_{j\bar k})$. Since each $\frac{\partial c_{IJ}}{\partial z_j}=0$, every term vanishes and $\bar\partial^*\alpha=0$. Therefore \begin{align*} \Delta_{\bar\partial}\alpha = \bar\partial\bar\partial^*\alpha + \bar\partial^*\bar\partial\alpha = 0. \end{align*} *Every harmonic form has constant coefficients.* Let $\beta=\sum_{I,J}\beta_{IJ}\,dz_I\wedge d\bar z_J$ satisfy $\Delta_{\bar\partial}\beta=0$. Since $(H_{j\bar k})$ is constant, the metric does not mix distinct basis forms; consequently the Dolbeault Laplacian acts on each coefficient separately: \begin{align*} \Delta_{\bar\partial}\beta = \sum_{I,J}(\Delta_{\bar\partial}\beta_{IJ})\,dz_I\wedge d\bar z_J, \end{align*} so $\Delta_{\bar\partial}\beta=0$ implies $\Delta_{\bar\partial}\beta_{IJ}=0$ for every pair $(I,J)$. Each $\beta_{IJ}$ is a smooth function on $X$, so $\bar\partial^*\beta_{IJ}=0$ for degree reasons (there are no $(-1)$-forms), and $\Delta_{\bar\partial}\beta_{IJ}=\bar\partial^*\bar\partial\beta_{IJ}$. Taking the $L^2$ inner product with $\beta_{IJ}$ and using the defining property of the adjoint $\langle\bar\partial^*u,v\rangle_{L^2}=\langle u,\bar\partial v\rangle_{L^2}$: \begin{align*} 0 = \langle\Delta_{\bar\partial}\beta_{IJ},\beta_{IJ}\rangle_{L^2} = \langle\bar\partial^*\bar\partial\beta_{IJ},\beta_{IJ}\rangle_{L^2} = \langle\bar\partial\beta_{IJ},\bar\partial\beta_{IJ}\rangle_{L^2} = \|\bar\partial\beta_{IJ}\|_{L^2}^2. \end{align*} Hence $\bar\partial\beta_{IJ}=0$, that is $\frac{\partial\beta_{IJ}}{\partial\bar z_k}=0$ for every $k$, so each $\beta_{IJ}$ is holomorphic on $X$. Since $X$ is compact, $|\beta_{IJ}|$ attains its maximum on $X$; by the maximum modulus principle applied in each local holomorphic coordinate, a holomorphic function achieving its maximum modulus at an interior point is locally constant, and by connectivity of $X$ it is globally constant. Thus every $\beta_{IJ}$ is constant. *Dimension count and identification.* The kernel of $\Delta_{\bar\partial}$ on $A^{p,q}(X)$ equals the space of constant-coefficient $(p,q)$-forms, spanned by $\{dz_I\wedge d\bar z_J:|I|=p,\,|J|=q\}$. There are $\binom{n}{p}$ choices of a $p$-element subset $I\subset\{1,\dots,n\}$ and $\binom{n}{q}$ choices of a $q$-element subset $J$, giving a basis of size $\binom{n}{p}\binom{n}{q}$. By the [Hodge Decomposition](/theorems/2745), $H^{p,q}_{\bar\partial}(X)\cong\ker\Delta_{\bar\partial}|_{A^{p,q}(X)}$, so the constant forms represent $H^{p,q}_{\bar\partial}(X)$ and $h^{p,q}(X)=\binom{n}{p}\binom{n}{q}$. The Hodge numbers depend only on $n$, $p$, $q$ — not on the lattice $\Lambda$ or the choice of constant metric — reflecting the flat, translation-invariant character of the geometry; in particular $h^{1,0}(X)=n$, so $b_1(X)=2n$, consistent with the topology of $\mathbb{R}^{2n}/\Lambda$. [/example] The Lefschetz operator adds a second layer to Hodge theory by using the Kähler form itself to move between degrees. The following definitions and decomposition are the parts of the Lefschetz package used later in curvature and embedding arguments. [definition: Primitive Form] Let $(X,\omega)$ be an $n$-dimensional Hermitian manifold. A $k$-form $\alpha$ with $k\le n$ is primitive if \begin{align*} \Lambda\alpha=0. \end{align*} [/definition] For forms of degree at most $n$, this is equivalent to the vanishing of $L^{n-k+1}\alpha$, but the adjoint formulation is the convention used with Kähler identities. Primitive forms are the pieces that cannot be obtained by wedging a lower-degree form with $\omega$. The pointwise definition is only useful globally if every form can be organized into primitive pieces and their successive wedges with the Kähler form. The decomposition below supplies that structure: it says that powers of $L$ and primitive components account for all relevant degrees in a way compatible with the Kähler setting used in Hodge theory. [quotetheorem:3855] [citeproof:3855] This decomposition matters because it prevents powers of the Kähler form from being treated as extra notation rather than structure. It says that the Kähler class acts on forms and cohomology in a controlled way, so questions in high degree can often be reduced to primitive classes in lower degree. Without the Kähler condition, the pointwise Lefschetz algebra may still exist, but it need not interact correctly with harmonic representatives and cohomology. In these notes, the decomposition is used as a structural statement: powers of the Kähler class organize cohomology into primitive pieces. ## Compact Kähler Manifolds and Positivity of the Kähler Class What global information is carried by the closed positive form $\omega$? Since $d\omega=0$, it defines a cohomology class, but positivity makes this class much more rigid than an arbitrary element of $H^2(X;\mathbb R)$. The Kähler class measures positive volume on complex directions and on compact complex submanifolds. [definition: Kahler Class] Let $X$ be a compact complex manifold. A cohomology class $\kappa\in H^2(X;\mathbb R)$ is a Kähler class if there exists a Kähler form $\omega$ on $X$ such that \begin{align*} \kappa=[\omega]. \end{align*} [/definition] Collecting all such classes gives the natural positivity region in second cohomology. The definition remembers more than the existence of a closed representative: it requires a representative that is positive on every complex tangent direction. This makes the Kähler cone a bridge between differential geometry and topology, since moving inside the cone changes the metric while preserving a cohomological positivity condition. Later, curvature forms of Hermitian line bundles provide a main source of such classes. [definition: Kahler Cone] Let $X$ be a compact complex manifold. The Kähler cone $\mathcal K_X$ is the set of all Kähler classes in $H^2(X;\mathbb R)$. [/definition] The word cone is justified by the fact that positive linear combinations of Kähler forms remain Kähler. The deeper problem, studied later through line bundles and curvature, is to decide which real $(1,1)$-classes lie in this cone. [quotetheorem:3856] [citeproof:3856] This theorem is the cohomological shadow of metric positivity. Its role is to turn the local condition "positive on complex tangent directions" into a numerical test on submanifolds. A Kähler class cannot be invisible on a positive-dimensional compact complex subspace: it must register positive volume there. This gives a first obstruction to being Kähler that can be stated purely in cohomology. This positivity is a necessary condition for a class to be Kähler. In projective geometry it is related to intersection numbers of ample divisors, while in analytic geometry it is expressed by integration of positive forms over complex subspaces. [example: The Fubini-Study Class] On $\mathbb{P}^n$ equipped with the unnormalized Fubini-Study form $\Omega_{\mathrm{FS}} = i\partial\bar\partial\varphi_0$, where $\varphi_0 = \log\!\bigl(1+\sum_{j=1}^n|z_j|^2\bigr)$ on the affine chart $U_0$, we verify positivity over every projective linear subspace. The normalized form used for Chern classes is $\omega_{\mathrm{FS}}=\frac{1}{2\pi}\Omega_{\mathrm{FS}}$. [claim]The cohomology class $[\Omega_{\mathrm{FS}}]\in H^2(\mathbb{P}^n;\mathbb{R})$ is a Kähler class, and for every projective linear subspace $\mathbb{P}^m\subset\mathbb{P}^n$ the integral $\int_{\mathbb{P}^m}\Omega_{\mathrm{FS}}^m>0$.[/claim] *The Kähler class.* The preceding Fubini-Study example established that $\Omega_{\mathrm{FS}}$ is a positive closed $(1,1)$-form on $\mathbb{P}^n$, hence a Kähler form. Its de Rham class $[\Omega_{\mathrm{FS}}]\in H^2(\mathbb{P}^n;\mathbb{R})$ is therefore a Kähler class by definition. *Restricting to $\mathbb{P}^m$.* Represent the linear subspace as the image of the holomorphic inclusion \begin{align*} \iota:\mathbb{P}^m\to\mathbb{P}^n,\qquad [W_0:\cdots:W_m]\mapsto[W_0:\cdots:W_m:0:\cdots:0]. \end{align*} On $U_0\cap\mathbb{P}^m$ with affine coordinates $w_j=W_j/W_0$ for $1\le j\le m$, the inclusion satisfies $z_j\circ\iota=w_j$ for $1\le j\le m$ and $z_j\circ\iota=0$ for $m+1\le j\le n$. Substituting into the Fubini-Study potential: \begin{align*} \iota^*\varphi_0^{(n)} = \log\!\Bigl(1+\sum_{j=1}^m|w_j|^2+\sum_{j=m+1}^n|0|^2\Bigr) = \log\!\Bigl(1+\sum_{j=1}^m|w_j|^2\Bigr) = \varphi_0^{(m)}, \end{align*} where $\varphi_0^{(m)}$ is the Fubini-Study potential on $\mathbb{P}^m$. Since $\iota$ is holomorphic, $\iota^*$ commutes with $\partial$ and $\bar\partial$ by [Naturality of the Pullback: Compatibility with Wedge Product and Exterior Derivative](/theorems/3574), so \begin{align*} \iota^*\Omega_{\mathrm{FS}} = i\partial\bar\partial\bigl(\iota^*\varphi_0^{(n)}\bigr) = i\partial\bar\partial\varphi_0^{(m)} = \Omega_{\mathrm{FS}}^{(m)}, \end{align*} where $\Omega_{\mathrm{FS}}^{(m)}$ is the unnormalized Fubini-Study form on $\mathbb{P}^m$. The same argument applies on every other affine chart $U_\ell\cap\mathbb{P}^m$, and the results are consistent because both sides are globally defined forms. *Positivity of the top power.* Integration of a form over a submanifold is defined by pullback along the inclusion, and by [Naturality of the Pullback: Compatibility with Wedge Product and Exterior Derivative](/theorems/3574) the pullback distributes over wedge products, so \begin{align*} \int_{\mathbb{P}^m}\Omega_{\mathrm{FS}}^m = \int_{\mathbb{P}^m}\iota^*\!\left(\Omega_{\mathrm{FS}}^m\right) = \int_{\mathbb{P}^m}\!\left(\iota^*\Omega_{\mathrm{FS}}\right)^m = \int_{\mathbb{P}^m}\!\left(\Omega_{\mathrm{FS}}^{(m)}\right)^m. \end{align*} It remains to show this last integral is positive. Write $\Omega_{\mathrm{FS}}^{(m)}=i\sum_{j,k=1}^m h_{j\bar k}\,dw_j\wedge d\bar w_k$ at a point $p\in\mathbb{P}^m$. By the [Spectral Theorem for Hermitian Matrices](/theorems/925), the positive-definite matrix $(h_{j\bar k}(p))$ is unitarily diagonalizable, so we may choose local holomorphic coordinates near $p$ in which \begin{align*} \Omega_{\mathrm{FS}}^{(m)}\big|_p = i\sum_{j=1}^m\lambda_j\,dw_j\wedge d\bar w_j,\qquad \lambda_j>0. \end{align*} Expanding the $m$-th power: since $(dw_j\wedge d\bar w_j)^2=0$ (as $dw_j\wedge dw_j=0$), each index $j$ can appear in at most one factor, and with exactly $m$ factors and $m$ distinct indices every factor must pick a different $j$. Two 2-forms commute under wedge product (sign $(-1)^{2\cdot 2}=1$), so the $m!$ ways to order the $m$ distinct summands each yield the same wedge product: \begin{align*} \left(\Omega_{\mathrm{FS}}^{(m)}\right)^m\!\Big|_p = m!\prod_{j=1}^m\lambda_j\cdot\left(idw_1\wedge d\bar w_1\right)\wedge\cdots\wedge\left(idw_m\wedge d\bar w_m\right). \end{align*} Writing $w_j=u_j+iv_j$ and expanding as before, $idw_j\wedge d\bar w_j = 2\,du_j\wedge dv_j$, which is a positive real area form. Hence \begin{align*} \left(\Omega_{\mathrm{FS}}^{(m)}\right)^m\!\Big|_p = 2^m m!\prod_{j=1}^m\lambda_j\cdot du_1\wedge dv_1\wedge\cdots\wedge du_m\wedge dv_m, \end{align*} a strictly positive multiple of the standard real volume form at $p$. Since this holds at every point of $\mathbb{P}^m$ with $\lambda_j>0$, the integrand is everywhere strictly positive. Integrating over the compact manifold $\mathbb{P}^m$ gives \begin{align*} \int_{\mathbb{P}^m}\!\left(\Omega_{\mathrm{FS}}^{(m)}\right)^m > 0. \end{align*} By the [De Rham Homomorphism](/theorems/3595), this integral equals the pairing of $[\Omega_{\mathrm{FS}}]^m$ against the fundamental class $[\mathbb{P}^m]$, confirming that the Fubini-Study Kähler class pairs positively with every projective linear subspace. This positivity is the cohomological expression of metric curvature. Dividing by $2\pi$ gives the normalized first Chern form $\omega_{\mathrm{FS}}=\frac{1}{2\pi}\Omega_{\mathrm{FS}}$, whose class is $c_1(\mathcal{O}(1))$; the same positivity, with this normalization, reappears as the ampleness condition underlying projective embedding theorems. [/example] The compactness assumption in Hodge theory is not a cosmetic detail: it is what gives finite-dimensional harmonic spaces and global cohomological decompositions. The Kähler assumption is also restrictive, as the next comparison shows. [example: Hopf Surface as a Non Kahler Comparison] [claim]The primary Hopf surface $X = (\mathbb{C}^2\setminus\{0\})/\langle z\mapsto 2z\rangle$ is a compact complex manifold that admits no Kähler metric.[/claim] *Topology of $X$.* Consider the map $f:\mathbb{R}\times S^3\to\mathbb{C}^2\setminus\{0\}$ defined by $f(t,u)=2^t u$, where $S^3\subset\mathbb{C}^2$ is the unit sphere. This is a diffeomorphism with inverse $v\mapsto(\log_2|v|,\, v/|v|)$, and the generator $z\mapsto 2z$ corresponds under $f$ to the translation $t\mapsto t+1$ on the $\mathbb{R}$ factor. The quotient of $\mathbb{R}$ by unit translation is $S^1$, so the quotient of $\mathbb{R}\times S^3$ by this action is $S^1\times S^3$, and the diffeomorphism $f$ descends to a diffeomorphism $X\cong S^1\times S^3$. *The first Betti number.* By the Künneth formula, $H^1(S^1\times S^3;\mathbb{Z})\cong H^1(S^1;\mathbb{Z})\oplus H^1(S^3;\mathbb{Z})$. Since $H^1(S^1;\mathbb{Z})\cong\mathbb{Z}$ and $H^1(S^3;\mathbb{Z})=0$ (as $S^3$ is simply connected and $H_1$ vanishes by Hurewicz), we obtain $H^1(X;\mathbb{Z})\cong\mathbb{Z}$, hence $b_1(X)=1$. *The Kähler obstruction.* Suppose for contradiction that $X$ admits a Kähler metric. By the [Hodge Decomposition](/theorems/2745) applied to the compact Kähler manifold $X$, \begin{align*} H^1(X;\mathbb{C}) \cong H^{1,0}_{\bar\partial}(X)\oplus H^{0,1}_{\bar\partial}(X). \end{align*} Complex conjugation on $H^1(X;\mathbb{C})$ interchanges forms of type $(1,0)$ and $(0,1)$, and the [Hodge Decomposition](/theorems/2745) asserts $\overline{H^{1,0}_{\bar\partial}(X)}=H^{0,1}_{\bar\partial}(X)$, so in particular $h^{0,1}(X)=h^{1,0}(X)$. Therefore \begin{align*} b_1(X) = \dim_{\mathbb{C}} H^1(X;\mathbb{C}) = h^{1,0}(X)+h^{0,1}(X) = 2h^{1,0}(X), \end{align*} which is even. But $b_1(X)=1$ is odd, a contradiction. Hence $X$ carries no Kähler metric. The parity constraint $b_1\in 2\mathbb{Z}_{\ge0}$ is a necessary condition for Kähler geometry that is purely topological in character: it follows only from the conjugate-symmetry of the Hodge decomposition and costs nothing about the specific complex structure. The Hopf surface shows that not every compact complex surface is Kähler, marking the boundary beyond which the identities, Lefschetz decomposition, and cohomological positivity results of this section no longer apply. [/example] The Hopf surface marks the boundary of the theory developed in this section. It is a compact complex manifold, but without a Kähler metric the identities, Hodge decomposition, Lefschetz decomposition on cohomology, and positivity of a Kähler class are unavailable in the forms stated above. Section 4 established the Kähler identities, whose validity depends crucially on the existence of a Kähler metric. Section 5 weaponizes these identities by using curvature of a Hermitian line bundle to construct a lower bound on the Laplacian, yielding the Bochner-Kodaira-Nakano identity that detects when positivity forces vanishing. # 5. Bochner-Kodaira-Nakano Identity Curvature first entered the course in Section 1 as the obstruction to flattening the Chern connection of a Hermitian holomorphic bundle, and Section 4 supplied the Kähler identities needed to manipulate adjoints. In this section curvature becomes an analytic term in an identity for the $\bar\partial$-Laplacian. The guiding principle is that integration by parts converts curvature positivity into estimates for harmonic forms, and those estimates force cohomology groups to vanish. The route is local at first: define formal adjoints for $\partial_E$ and $\bar\partial_E$, record the Kähler commutation identities, and then commute the operators until curvature appears. The global consequence is the Bochner-Kodaira-Nakano identity, which is the bridge from Hermitian curvature to Kodaira-type vanishing theorems. ## Formal Adjoints on Hermitian Bundle Valued Forms How should $\partial_E$ and $\bar\partial_E$ be moved from one factor of an integral to the other when the forms take values in a Hermitian bundle? The answer depends on the Hermitian metric on the base, the Hermitian metric on the bundle, and the volume form they determine. Let $(X,\omega)$ be a Hermitian complex manifold of complex dimension $n$, and let $(E,h)$ be a Hermitian complex vector bundle. Write $\mathcal A^{p,q}(X,E)$ for smooth $E$-valued $(p,q)$-forms on $X$. [definition: $L^2$ Pairing on Bundle Valued Forms] The $L^2$ pairing is the map \begin{align*} (\,\cdot\,,\,\cdot\,)_{L^2}:\mathcal A_c^{p,q}(X,E)\times \mathcal A_c^{p,q}(X,E)\to \mathbb C \end{align*} defined for compactly supported $\alpha,\beta \in \mathcal A^{p,q}(X,E)$ by \begin{align*} (\alpha,\beta)_{L^2}=\int_X \langle \alpha,\beta\rangle_{\omega,h}\,dV_\omega. \end{align*} [/definition] This is the analytic pairing used throughout the section. If $X$ is compact, the compact support condition may be omitted. The Bochner identities used later are energy identities, so they require a precise way to transfer differential operators from one factor of the $L^2$ pairing to the other. Integration by parts produces an operator depending on the chosen Hermitian metrics and volume form, and that metric-dependent operator is what enters the Laplacians. [definition: Formal Adjoint] Let $D:\mathcal A^{p,q}(X,E)\to \mathcal A^{r,s}(X,E)$ be a linear differential operator. A formal adjoint of $D$ is an operator $D^*:\mathcal A^{r,s}(X,E)\to \mathcal A^{p,q}(X,E)$ satisfying \begin{align*} (D\alpha,\beta)_{L^2}=(\alpha,D^*\beta)_{L^2} \end{align*} for all compactly supported smooth forms $\alpha$ and $\beta$ of the indicated bidegrees. [/definition] Formal adjoints are not algebraic duals. They are differential operators produced by integration by parts, so their formula changes when the metric changes. The next operators are the bundle-valued analogues of $\partial$ and $\bar\partial$ used to build those adjoints and Laplacians. Because the forms now take values in a Hermitian holomorphic bundle, the correct first-order operators come from decomposing the Chern connection rather than differentiating coefficients alone. [definition: Bundle Valued Dolbeault Operators] Let $E\to X$ be a holomorphic Hermitian vector bundle with Chern connection $\nabla_E$. The decomposition of the Chern connection by type is \begin{align*} \nabla_E=\partial_E+\bar\partial_E, \end{align*} where \begin{align*} \partial_E&:\mathcal A^{p,q}(X,E)\to \mathcal A^{p+1,q}(X,E),\\ \bar\partial_E&:\mathcal A^{p,q}(X,E)\to \mathcal A^{p,q+1}(X,E). \end{align*} [/definition] The operator $\bar\partial_E$ is the Dolbeault operator determined by the holomorphic structure. The operator $\partial_E$ contains the metric information through the Chern connection. To form Laplacians from these first-order operators, one also needs their formal adjoints with respect to the Hermitian metric and volume form. The adjoint is not simply obtained by changing a sign: it differentiates in the conjugate direction and removes the corresponding exterior factor. The following local formula isolates this behavior in normal coordinates, where the principal part of the adjoint is visible without lower-order connection terms. [quotetheorem:3857] [citeproof:3857] The formula says that the adjoint differentiates in the conjugate direction and then contracts the corresponding differential form component. It should be read as a normal-frame description of the principal local behavior, not as a global coordinate formula. Away from a normal frame, extra first-order terms from the metric and connection are present, but they combine tensorially to give the same formal adjoint. The important structural feature is the conjugate-direction differentiation: $\bar\partial_E^*$ differentiates in the $(1,0)$ directions and contracts a $(0,1)$ index, while $\partial_E^*$ does the opposite. On a Kähler manifold, this contraction behaviour is controlled by wedging with the Kähler form, so the next operators record the algebra that turns adjoints into commutators. [definition: Lefschetz Operators] Let $(X,\omega)$ be a Hermitian complex manifold. The Lefschetz operator is \begin{align*} L:\mathcal A^{p,q}(X,E)\to \mathcal A^{p+1,q+1}(X,E),\qquad L\alpha=\omega\wedge\alpha. \end{align*} Its formal adjoint $\Lambda=L^*$ is the operator \begin{align*} \Lambda:\mathcal A^{p+1,q+1}(X,E)\to \mathcal A^{p,q}(X,E) \end{align*} characterised by $(L\alpha,\beta)_{L^2}=(\alpha,\Lambda\beta)_{L^2}$ on compactly supported forms. [/definition] The operator $L$ raises total degree by two, while $\Lambda$ lowers total degree by two. Together with their degree-counting commutator, they form the Lefschetz algebra behind the primitive decomposition of forms. Their commutators with $\partial_E$ and $\bar\partial_E$ are the Kähler identities. [illustration:scv-iv-lefschetz-ladder] [quotetheorem:3858] [citeproof:3858] These identities are the operator-level input for the Bochner-Kodaira-Nakano identity. They express a special feature of Kähler geometry: the metric Lefschetz operations control the formal adjoints of the Dolbeault pieces of the Chern connection. The Kähler hypothesis is essential here: on a general Hermitian manifold, torsion terms survive in the same commutator calculation, so the displayed identities acquire correction terms rather than remaining pure adjoint identities. For example, on a non-Kähler Hermitian manifold such as a Hopf surface with its standard Hermitian metric, the Lee form is nonzero and the $\bar\partial$-Laplacian formulas contain Lee-form or torsion corrections. The identities also do not assert a global integration statement without hypotheses; they are local differential-operator identities, used on compactly supported forms or on compact manifolds where formal adjoints have no boundary contribution. Beyond the Bochner-Kodaira-Nakano identity, the same $L,\Lambda$ commutator algebra is the $\mathfrak{sl}_2$ representation underlying primitive decomposition and the Lefschetz decomposition of differential forms. ## Curvature in the Dolbeault Laplacian Where does curvature enter the Laplacian attached to $\bar\partial_E$? It appears when the Kähler identities are substituted into the definition of the Laplacian and the resulting commutators are simplified. [definition: Dolbeault and Partial Laplacians] For a Hermitian holomorphic vector bundle $E\to X$, the Dolbeault Laplacian and partial Laplacian are the operators \begin{align*} \Delta_{\bar\partial_E},\Delta_{\partial_E}:\mathcal A^{p,q}(X,E)\to \mathcal A^{p,q}(X,E) \end{align*} defined by \begin{align*} \Delta_{\bar\partial_E}&=\bar\partial_E\bar\partial_E^*+\bar\partial_E^*\bar\partial_E,\\ \Delta_{\partial_E}&=\partial_E\partial_E^*+\partial_E^*\partial_E. \end{align*} [/definition] The difference between these two Laplacians is not a higher-order operator. On a Kähler manifold it is exactly the zero-order curvature action. This is the analytic point where the Kähler identities become useful rather than formal. The two Laplacians have the same leading second-order part, so their difference can only contain lower-order information; the Kähler commutator relations identify that remaining information with curvature. The resulting identity converts positivity of a bundle metric into a coercive estimate for the $\bar\partial$-Laplacian. [quotetheorem:3859] [citeproof:3859] This identity is the main analytic formula of the section: curvature has become a pointwise Hermitian form on bundle-valued differential forms. Its usefulness is that the analytic size of a $\bar\partial_E$-harmonic form can be tested against curvature positivity of the bundle metric. In applications, the last term is read as a lower bound on selected bidegrees. Positivity assumptions on $E$ determine where that lower bound is strong enough to rule out nonzero harmonic representatives, and Hodge theory then translates the analytic statement into Dolbeault cohomology. The identity is therefore a bridge between local curvature signs and global vanishing, not merely a formal comparison of two Laplacians. [explanation: Bochner Method] The identity has the form \begin{align*} \text{analytic norm} = \text{analytic norm} + \text{curvature term}. \end{align*} When the curvature term is bounded below by a positive multiple of $\|u\|_{L^2}^2$, a $\bar\partial_E$-harmonic form must vanish. Hodge theory then translates the vanishing of harmonic forms into the vanishing of Dolbeault cohomology groups. This is the Bochner method in complex geometry. Instead of solving a differential equation directly, the method proves that any possible solution of the homogeneous harmonic equation must have zero $L^2$ norm. [/explanation] The Bochner method becomes effective only after the curvature commutator has been converted into an explicit lower bound. For line bundles the endomorphism part of curvature is scalar, so one can diagonalise the curvature in a unitary coframe and read the commutator as a sum of curvature eigenvalues. The top holomorphic degree case is the one needed for adjoint bundles $K_X\otimes L$, because their cohomology is represented analytically by $L$-valued $(n,q)$-forms. The remaining point is to make this lower bound completely explicit in the bidegree used by Kodaira vanishing. The following calculation isolates exactly how the curvature eigenvalues of a positive line bundle enter the commutator on $(n,q)$-forms. [quotetheorem:3860] [citeproof:3860] Thus positive curvature gives a positive pointwise contribution on $(n,q)$-forms whenever $q>0$. The restriction to holomorphic degree $n$ is what makes the formula so simple. If the form has type $(p,q)$ with $p<n$, some holomorphic factors $\theta_j$ are absent, and the commutator has both positive and negative contributions depending on which directions occur. Thus positivity of $L$ does not translate into the same immediate scalar lower bound in every bidegree. The top-degree calculation previews why the canonical bundle $K_X$ appears in the vanishing theorem: twisting by $K_X$ moves the analytic problem to $(n,q)$-forms, where the curvature eigenvalues enter with the correct sign. ## Nakano and Griffiths Positivity for Vector Bundles Which notion of positivity is strong enough to make the curvature term in the Bochner-Kodaira-Nakano identity have a sign? For line bundles there is only one curvature form to test, but for vector bundles the curvature has both tangent and bundle indices. Fix a point $x\in X$, choose a unitary holomorphic coordinate frame for $T_x^{1,0}X$, and choose an $h$-unitary frame $e_1,\dots,e_r$ for $E_x$. Write the curvature coefficients of $i\Theta_h(E)$ as \begin{align*} i\Theta_h(E)_x=i\sum_{j,k=1}^n\sum_{\alpha,\beta=1}^r R_{j\bar k\alpha\bar\beta}\,dz_j\wedge d\bar z_k\otimes e_\alpha^*\otimes e_\beta. \end{align*} [definition: Griffiths Positivity] The bundle $(E,h)$ is Griffiths semipositive if, for every $x\in X$, every $\xi=(\xi_j)\in T_x^{1,0}X$, and every $v=(v_\alpha)\in E_x$, \begin{align*} \sum_{j,k,\alpha,\beta}R_{j\bar k\alpha\bar\beta}\xi_j\overline{\xi_k}v_\alpha\overline{v_\beta}\ge 0. \end{align*} It is Griffiths positive if the same expression is $>0$ whenever $\xi\ne 0$ and $v\ne 0$. [/definition] Griffiths positivity tests the curvature only on decomposable tensors $\xi\otimes v$. This is the natural condition when a tangent direction and a bundle direction are chosen independently, so it is often easier to verify geometrically. The Bochner curvature term, however, acts on arrays with both a tangent index and a bundle index, and those arrays need not have rank one. The stronger condition below tests all such arrays at once. [definition: Nakano Positivity] The bundle $(E,h)$ is Nakano semipositive if, for every $x\in X$ and every array $U=(u_{j\alpha})\in T_x^{1,0}X\otimes E_x$, \begin{align*} \sum_{j,k,\alpha,\beta}R_{j\bar k\alpha\bar\beta}u_{j\alpha}\overline{u_{k\beta}}\ge 0. \end{align*} It is Nakano positive if the same expression is $>0$ whenever $U\ne 0$. [/definition] Nakano positivity tests all tensors in $T_x^{1,0}X\otimes E_x$. This is the positivity condition naturally matched to the curvature operator on $K_X\otimes E$-valued forms. The two positivity notions will lead to different vanishing statements unless their relationship is made explicit. Both are built from the same curvature tensor, but Griffiths positivity tests only decomposable tensors while Nakano positivity tests arbitrary arrays. The comparison below identifies which implication is automatic and explains why line bundles are the special case where no higher-rank array directions remain. [quotetheorem:3861] [citeproof:3861] The comparison reflects how many curvature directions each condition tests. Nakano positivity is designed for estimates involving arbitrary tangent-bundle arrays, while Griffiths positivity records only the decomposable directions that come from choosing one tangent vector and one fiber vector. For line bundles there is no higher-rank fiber geometry to hide extra directions, so the two tests express the same curvature sign. The converse fails for a genuinely linear-algebraic reason. Testing only rank-one tensors can miss negative directions on higher-rank arrays, and those missed directions are exactly the kind of directions that appear in Nakano's curvature term. The next example isolates this gap before returning to geometric bundles. [example: Algebraic Gap Between Griffiths and Nakano] Take $\dim T_x^{1,0}X = 2$ and $\operatorname{rank} E = 2$, so that $T_x^{1,0}X \otimes E_x \cong \mathbb{C}^2 \otimes \mathbb{C}^2$; identify this with complex $2\times 2$ matrices $A = (a_{j\alpha})$, where the entry $a_{j\alpha}$ corresponds to the basis vector $e_j \otimes f_\alpha$. Consider the Hermitian quadratic form \begin{align*} Q(A) = \sum_{j,\alpha}|a_{j\alpha}|^2 - \tfrac{3}{4}|a_{11}+a_{22}|^2. \end{align*} [claim]$Q$ is strictly positive on every nonzero decomposable matrix $A = \xi \otimes v$, yet $Q(I) = -1 < 0$ for the identity matrix $I$.[/claim] *Decomposable case.* A decomposable matrix has entries $a_{j\alpha} = \xi_j v_\alpha$ for vectors $\xi = (\xi_1,\xi_2)$ and $v = (v_1,v_2)$. Because the $j$- and $\alpha$-sums factor, \begin{align*} \sum_{j,\alpha}|a_{j\alpha}|^2 = \sum_{j,\alpha}|\xi_j|^2|v_\alpha|^2 = \Bigl(\sum_j|\xi_j|^2\Bigr)\Bigl(\sum_\alpha|v_\alpha|^2\Bigr) = \|\xi\|^2\|v\|^2 = \|A\|_{\mathrm{Frob}}^2. \end{align*} The trace-like term is $a_{11}+a_{22} = \xi_1 v_1+\xi_2 v_2$. Applying the [Cauchy-Schwarz Inequality](/theorems/432) to the pairs $(\xi_1,\xi_2)$ and $(v_1,v_2)$, \begin{align*} |a_{11}+a_{22}| = |\xi_1 v_1+\xi_2 v_2| \le \sqrt{|\xi_1|^2+|\xi_2|^2}\,\sqrt{|v_1|^2+|v_2|^2} = \|\xi\|\,\|v\| = \|A\|_{\mathrm{Frob}}. \end{align*} Substituting both computations into $Q$, \begin{align*} Q(A) = \|A\|_{\mathrm{Frob}}^2 - \tfrac{3}{4}|a_{11}+a_{22}|^2 \ge \|A\|_{\mathrm{Frob}}^2 - \tfrac{3}{4}\|A\|_{\mathrm{Frob}}^2 = \tfrac{1}{4}\|A\|_{\mathrm{Frob}}^2. \end{align*} Since $A \ne 0$ implies $\|A\|_{\mathrm{Frob}} > 0$, we conclude $Q(A) \ge \tfrac{1}{4}\|A\|_{\mathrm{Frob}}^2 > 0$. *Identity matrix.* The entries of $I$ are $a_{11} = a_{22} = 1$ and $a_{12} = a_{21} = 0$, so \begin{align*} \sum_{j,\alpha}|a_{j\alpha}|^2 = 1 + 0 + 0 + 1 = 2, \qquad |a_{11}+a_{22}|^2 = |1+1|^2 = 4. \end{align*} Therefore \begin{align*} Q(I) = 2 - \tfrac{3}{4}\cdot 4 = 2 - 3 = -1 < 0. \end{align*} Since $I$ has rank $2$ it cannot be written as $\xi\otimes v$, so $I$ lies outside the set of tensors tested by the Griffiths condition. The form $Q$ thus models a curvature tensor that is Griffiths positive — it is positive on every rank-one tensor — while being strictly negative on the rank-two matrix $I$, an element of $T_x^{1,0}X\otimes E_x$ that is invisible to the Griffiths test but fully present in the Nakano pairing. This is the linear-algebraic mechanism by which Griffiths positivity fails to control the curvature term in the Bochner-Kodaira-Nakano identity when the bundle rank exceeds one. [/example] The algebraic example shows the mechanism but does not come from a particular bundle. Projective space supplies a geometric model in which the curvature tensor has an explicit formula and the two positivity tests can be compared by direct calculation. In this example, Griffiths positivity is visible on decomposable tensors, while Nakano positivity detects additional skew-symmetric directions. [example: Tangent Bundle of Projective Space] Let $\mathbb{P}^n$ carry the Fubini-Study metric and let $E = T\mathbb{P}^n$ with the induced Hermitian metric. At any point, in a unitary coframe for $T^{1,0}\mathbb{P}^n$ and a unitary frame for $E$, the Chern curvature coefficients are \begin{align*} R_{i\bar j\alpha\bar\beta} = \delta_{ij}\delta_{\alpha\beta} + \delta_{i\beta}\delta_{\alpha j}. \end{align*} [claim]$T\mathbb{P}^n$ is Griffiths positive. For $n\ge 2$ it is Nakano semipositive but not Nakano positive.[/claim] *Griffiths positivity.* Fix nonzero $\xi = (\xi_i)\in T_x^{1,0}\mathbb{P}^n$ and $v = (v_\alpha)\in E_x$. The Griffiths form on the decomposable tensor $\xi\otimes v$ is \begin{align*} G(\xi,v) = \sum_{i,j,\alpha,\beta} R_{i\bar j\alpha\bar\beta}\,\xi_i\overline{\xi_j}\,v_\alpha\overline{v_\beta}. \end{align*} Substituting the curvature and splitting by the two Kronecker terms: The $\delta_{ij}\delta_{\alpha\beta}$ term sets $j = i$ and $\beta = \alpha$, giving $\sum_{i,\alpha}|\xi_i|^2|v_\alpha|^2 = \bigl(\sum_i|\xi_i|^2\bigr)\bigl(\sum_\alpha|v_\alpha|^2\bigr) = |\xi|^2|v|^2$, where the product factors because the $i$- and $\alpha$-sums are independent. The $\delta_{i\beta}\delta_{\alpha j}$ term sets $j = \alpha$ and $\beta = i$, giving $\sum_{i,\alpha}\xi_i\overline{\xi_\alpha}\,v_\alpha\overline{v_i}$. Setting $s = \sum_i \xi_i\overline{v_i}$, this sum factors as $\bigl(\sum_i\xi_i\overline{v_i}\bigr)\bigl(\sum_\alpha\overline{\xi_\alpha}v_\alpha\bigr) = s\bar s = |s|^2 = \bigl|\sum_i\xi_i\overline{v_i}\bigr|^2$. Therefore \begin{align*} G(\xi,v) = |\xi|^2|v|^2 + \Bigl|\sum_i\xi_i\overline{v_i}\Bigr|^2. \end{align*} The first term is strictly positive for $\xi,v\ne 0$, so $G(\xi,v) > 0$ in all such cases. *Nakano semipositivity.* For a general array $A = (a_{i\alpha})\in T_x^{1,0}\mathbb{P}^n\otimes E_x$, the Nakano form is \begin{align*} N(A) = \sum_{i,j,\alpha,\beta}R_{i\bar j\alpha\bar\beta}\,a_{i\alpha}\overline{a_{j\beta}}. \end{align*} The same substitution gives two sums. The $\delta_{ij}\delta_{\alpha\beta}$ term contributes $\sum_{i,\alpha}|a_{i\alpha}|^2 = \|A\|_F^2$. The $\delta_{i\beta}\delta_{\alpha j}$ term (setting $j = \alpha$, $\beta = i$) contributes $\sum_{i,\alpha}a_{i\alpha}\overline{a_{\alpha i}}$. Thus \begin{align*} N(A) = \|A\|_F^2 + \sum_{i,\alpha}a_{i\alpha}\overline{a_{\alpha i}}. \end{align*} Separating each quantity into its diagonal part ($i = \alpha$) and off-diagonal pairs ($i < \alpha$): \begin{align*} \|A\|_F^2 &= \sum_i|a_{ii}|^2 + \sum_{i<\alpha}\bigl(|a_{i\alpha}|^2+|a_{\alpha i}|^2\bigr),\\ \sum_{i,\alpha}a_{i\alpha}\overline{a_{\alpha i}} &= \sum_i|a_{ii}|^2 + \sum_{i<\alpha}\bigl(a_{i\alpha}\overline{a_{\alpha i}}+a_{\alpha i}\overline{a_{i\alpha}}\bigr) = \sum_i|a_{ii}|^2+\sum_{i<\alpha}2\operatorname{Re}(a_{i\alpha}\overline{a_{\alpha i}}). \end{align*} Adding and applying $|u|^2+|v|^2+2\operatorname{Re}(u\bar v) = |u+v|^2$ to each off-diagonal pair: \begin{align*} N(A) = 2\sum_i|a_{ii}|^2 + \sum_{i<\alpha}|a_{i\alpha}+a_{\alpha i}|^2. \end{align*} Every term is nonneg, so $N(A)\ge 0$ for all $A$, establishing Nakano semipositivity. *Failure of Nakano positivity for $n\ge 2$.* When $n\ge 2$, define $A_0$ by $a_{12} = 1$, $a_{21} = -1$, and all other entries zero. Then $A_0\ne 0$. All diagonal entries of $A_0$ vanish, and the only off-diagonal pair is $(i,\alpha) = (1,2)$, contributing $|a_{12}+a_{21}|^2 = |1+(-1)|^2 = 0$. Therefore $N(A_0) = 0$ with $A_0\ne 0$, so the Nakano form is not strictly positive. The formula $N(A) = 2\sum_i|a_{ii}|^2+\sum_{i<\alpha}|a_{i\alpha}+a_{\alpha i}|^2$ shows that the Nakano form of $T\mathbb{P}^n$ sees only the symmetric part of the array $A$: it vanishes precisely on nonzero skew-symmetric matrices, which exist as soon as $n\ge 2$, placing the Fubini-Study tangent bundle exactly at the boundary between Nakano semipositivity and Nakano positivity. [/example] The vanishing theorem is the global payoff of the curvature estimate. The local positivity calculation supplies the sign condition, while compact Kähler Hodge theory supplies the link from analysis to Dolbeault cohomology. The first version is the line bundle statement, where positivity is measured by the curvature form of $L$. Here $\Omega_X^p$ denotes the holomorphic vector bundle, equivalently the sheaf, of holomorphic $p$-forms on $X$. [quotetheorem:3862] [citeproof:3862] For the reader, the important point is the range $p+q>n$: positivity of $L$ is strong enough there to eliminate the corresponding twisted Dolbeault groups. The boundary $p+q=n$ is not covered, and this is not an accident; middle bidegrees often carry genuine cohomology. The hypotheses also mark real limitations. The Kähler condition cannot simply be dropped, since torsion terms appear outside the Kähler setting. Positivity of $L$ is stronger than numerical largeness alone: a line bundle is nef if its degree on every irreducible curve is nonnegative, and big if it has maximal asymptotic section growth; Ramanujam constructed nef and big line bundles for which Kodaira-Nakano-type vanishing fails. The most useful instance of the theorem is the top-bidegree case $p=n$, recorded next as the adjoint-bundle vanishing that appears throughout algebraic geometry. For the adjoint case, the important point is not another proof but the bidegree bookkeeping. The canonical factor places the Dolbeault representatives in holomorphic degree $n$, and the curvature commutator from the previous theorem is positive on exactly those $(n,q)$-forms with $q>0$. This is why adjoint bundles are the natural home for Kodaira-type vanishing: the metric positivity and the cohomological degree line up without extra cancellation. The vector bundle version uses exactly the positivity notion designed for the curvature term on $K_X\otimes E$-valued forms. The line-bundle argument above works because positivity gives a pointwise lower bound for the curvature commutator on adjoint-valued forms. For higher-rank bundles, the same Bochner argument requires positivity on all coefficient arrays appearing in the form, not merely on rank-one tensors. This is why Nakano positivity, rather than Griffiths positivity, is the correct hypothesis for the vector bundle vanishing statement. [quotetheorem:3863] [citeproof:3863] A few remarks frame what this theorem says and what it does not. The Nakano hypothesis is stronger than Griffiths positivity because it controls all coefficient arrays that appear in adjoint-valued forms, not only decomposable tensors. The $K_X$ twist is also part of the natural range of the result: it is what aligns the curvature positivity with the cohomological degrees being killed. Later vanishing theorems weaken or singularize the positivity hypothesis, but this smooth statement is the clean model that explains why positivity of curvature becomes disappearance of higher cohomology. The Bochner-Kodaira-Nakano identity of Section 5 proved that positive curvature makes the $\bar\partial$-Laplacian coercive on certain forms. Section 6 globalizes this estimate: by solving the $\bar\partial$-equation with a gain-of-one-order argument, positivity becomes a vanishing theorem for cohomology, and Serre duality extends this to show that negative bundles have vanishing in complementary degrees. # 6. Kodaira Vanishing and Serre Duality Applications This section turns the curvature estimates from the preceding lectures into global information about holomorphic sections. The central theme is that positivity of a line bundle makes the $\bar\partial$-Laplacian coercive in the degrees that matter for adjoint bundles. Serre duality then converts the same vanishing into statements about negative bundles, and Riemann--Roch records the remaining numerical information as a polynomial. The guiding question is not only whether cohomology groups vanish, but what their disappearance lets us count. Once higher cohomology is gone, the Euler characteristic becomes the dimension of a space of sections. This is the bridge from curvature to projective geometry. ## Vanishing for Adjoint Positive Bundles The first problem is analytic: how does positive curvature force a Dolbeault cohomology class to be zero? On a compact Kähler manifold, each cohomology class has a harmonic representative, so a curvature lower bound can be tested directly on forms. [definition: Positive Holomorphic Line Bundle] Let $X$ be a compact complex manifold and let $L \to X$ be a holomorphic line bundle. A Hermitian metric $h$ on $L$ is positive if the real $(1,1)$-form $i\Theta_h(L)$ is positive on every nonzero $v\in T_x^{1,0}X$. The line bundle $L$ is positive if it admits a positive Hermitian metric. [/definition] Equivalently, the normalized form $c_1(L,h)=\frac{i}{2\pi}\Theta_h(L)$ is positive in the same sense. In a local holomorphic frame $e$ with $|e|_h^2=e^{-\varphi}$, positivity says that $i\partial\bar\partial\varphi$ is a positive $(1,1)$-form. The global theorem uses this local positivity inside the Bochner--Kodaira identity. In practice, positivity is usually checked either by writing a local potential $\varphi$ and computing the Hermitian matrix $(\partial_i\partial_{\bar j}\varphi)$, or by restricting the Fubini--Study metric from projective space. [example: Positive Bundle Without the Adjoint Twist] Let $C$ be a compact Riemann surface of genus $g = 2$, let $p \in C$, and set $L = \mathcal{O}_C(p)$. We show that $H^1(C, L) \ne 0$, so positivity of degree alone does not force the vanishing that Kodaira vanishing requires the adjoint twist to supply. The line bundle $\mathcal{O}_C(p)$ is defined by the cocycle $\{g_{\alpha\beta}\}$ where $g_{\alpha\beta} = z_\alpha/z_\beta$ near $p$ in local charts and $g_{\alpha\beta} = 1$ away from $p$; its degree is $\deg L = 1 > 0$. On any compact Riemann surface, a degree-$1$ line bundle carries a positive Hermitian metric: choosing any Kähler form $\omega$ on $C$, one can find a Hermitian metric $h$ on $L$ such that $\frac{i}{2\pi}\Theta_h(L) = \frac{\deg L}{\int_C \omega}\,\omega = \frac{\omega}{\int_C \omega}$, which is a positive $(1,1)$-form. Thus $L$ is positive. Apply *Serre Duality for Holomorphic Vector Bundles* with $n = 1$, $q = 1$, and $E = L$. The dual bundle $L^*$ has transition functions $\{g_{\alpha\beta}^{-1}\}$, which are precisely the defining cocycle of $\mathcal{O}_C(-p)$, so $L^* = L^{-1} = \mathcal{O}_C(-p)$. Serre duality then gives \begin{align*} H^1(C, L)^* \cong H^{1-1}\!\bigl(C,\, K_C \otimes L^{-1}\bigr) = H^0\bigl(C,\, K_C(-p)\bigr). \end{align*} It therefore suffices to exhibit a nonzero global section of $K_C(-p)$. The degree of $K_C$ on a genus-$g$ curve is $\deg K_C = 2g - 2 = 2(2) - 2 = 2$. Applying [Riemann–Roch Theorem](/theorems/2185) to the line bundle $K_C$ gives \begin{align*} h^0(C, K_C) - h^1(C, K_C) = \deg K_C + 1 - g = 2 + 1 - 2 = 1. \end{align*} To find $h^1(C, K_C)$, apply *Serre Duality for Holomorphic Vector Bundles* a second time, now with $E = K_C$ and $q = 1$: since $K_C^* = K_C^{-1}$ (the dual bundle has transition functions $g_{\alpha\beta}^{-1}$ whenever $g_{\alpha\beta}$ are those of $K_C$), tensoring gives $K_C \otimes K_C^{-1} \cong \mathcal{O}_C$ because the transition functions satisfy $g_{\alpha\beta} \cdot g_{\alpha\beta}^{-1} = 1$, the constant cocycle defining the structure sheaf. Thus \begin{align*} H^1(C, K_C)^* \cong H^0\bigl(C,\, K_C \otimes K_C^{-1}\bigr) = H^0(C,\, \mathcal{O}_C). \end{align*} Since $C$ is connected, every holomorphic function $C \to \mathbb{C}$ is constant, so $H^0(C, \mathcal{O}_C) \cong \mathbb{C}$ and $h^1(C, K_C) = 1$. Substituting into the Riemann–Roch equation gives $h^0(C, K_C) = 1 + 1 = 2$. A section of $K_C(-p)$ is, by definition, a section of $K_C$ that vanishes at $p$: the bundle $K_C(-p) = K_C \otimes \mathcal{O}_C(-p)$ has sections corresponding to holomorphic one-forms with a zero at $p$, since the factor $\mathcal{O}_C(-p)$ imposes vanishing along the divisor $p$. Evaluation at $p$ defines a $\mathbb{C}$-linear map \begin{align*} \mathrm{ev}_p \colon H^0(C, K_C) \longrightarrow (K_C)_p \cong \mathbb{C}, \qquad s \longmapsto s(p), \end{align*} whose kernel is $H^0(C, K_C(-p))$. By the rank-nullity theorem, \begin{align*} \dim \ker(\mathrm{ev}_p) = \dim H^0(C, K_C) - \dim \operatorname{im}(\mathrm{ev}_p) \geq 2 - 1 = 1, \end{align*} since $\dim H^0(C, K_C) = 2$ and $\dim (K_C)_p = 1$ forces $\dim \operatorname{im}(\mathrm{ev}_p) \leq 1$. Thus $H^0(C, K_C(-p)) \ne 0$. Chasing back through the duality isomorphism, $H^1(C, L)^* \cong H^0(C, K_C(-p)) \ne 0$, so $H^1(C, L) \ne 0$: the degree-$1$ positive bundle $L = \mathcal{O}_C(p)$ on the genus-$2$ curve has nonvanishing first cohomology, confirming that the adjoint twist $K_X \otimes L$ in [Kodaira Vanishing Theorem (Statement)](/theorems/3501) is not a decorative convention but the exact shift needed to place positive curvature in the bidegree where the Bochner–Kodaira estimate is coercive. [/example] The example also signals the role of bidegree in the Bochner argument. Positivity of curvature becomes a coercive term only after the canonical factor has shifted the forms into the $(n,q)$ range. [illustration:scv-iv-bochner-kodaira-curvature] The vanishing theorem is the global payoff of this analytic setup. After the canonical twist places cohomology in the top holomorphic degree, positive curvature forces harmonic representatives in positive anti-holomorphic degree to vanish, and Hodge theory translates that analytic vanishing into sheaf cohomology. [quotetheorem:3501] [citeproof:3501] The factor $K_X$ is the reason the positivity has the correct sign in this degree. Geometrically, Kodaira vanishing says that adjoint bundles $K_X\otimes L$ behave cohomologically like bundles with enough positivity. [example: Adjoint Bundle on Projective Space] Let $X = \mathbb{P}^n$ and $L = \mathcal{O}_{\mathbb{P}^n}(m)$ with $m > 0$; we show that $H^q(\mathbb{P}^n, K_X \otimes L) = 0$ for all $q > 0$, including cases where the resulting twist $m - n - 1$ is negative. *Computing the canonical bundle.* The Euler sequence on $\mathbb{P}^n$ is the short exact sequence of holomorphic vector bundles \begin{align*} 0 \longrightarrow \mathcal{O}_{\mathbb{P}^n} \longrightarrow \mathcal{O}_{\mathbb{P}^n}(1)^{\oplus(n+1)} \longrightarrow T^{1,0}_{\mathbb{P}^n} \longrightarrow 0. \end{align*} The determinant is multiplicative on short exact sequences, so \begin{align*} \det\!\bigl(\mathcal{O}(1)^{\oplus(n+1)}\bigr) \cong \det(\mathcal{O}_{\mathbb{P}^n}) \otimes \det\!\bigl(T^{1,0}_{\mathbb{P}^n}\bigr). \end{align*} The left side is $\mathcal{O}(1)^{\otimes(n+1)} \cong \mathcal{O}(n+1)$: the determinant of a direct sum of line bundles $\ell_1 \oplus \cdots \oplus \ell_k$ is $\ell_1 \otimes \cdots \otimes \ell_k$, and the transition function of $\mathcal{O}(1)^{\oplus(n+1)}$ on the standard cover is $(z_j/z_i) \cdot \mathrm{Id}_{n+1}$, whose determinant is $(z_j/z_i)^{n+1}$, the transition function of $\mathcal{O}(n+1)$. The structure sheaf satisfies $\det(\mathcal{O}_{\mathbb{P}^n}) = \mathcal{O}_{\mathbb{P}^n}$ since its transition functions are all $1$ and $\det(1) = 1$. Canceling this factor gives $\det(T^{1,0}_{\mathbb{P}^n}) \cong \mathcal{O}(n+1)$. Dualizing — the dual of a line bundle with transition functions $\{g_{\alpha\beta}\}$ has transition functions $\{g_{\alpha\beta}^{-1}\}$, so $\mathcal{O}(n+1)^* \cong \mathcal{O}(-n-1)$ — yields \begin{align*} K_{\mathbb{P}^n} = \bigl(\det T^{1,0}_{\mathbb{P}^n}\bigr)^* \cong \mathcal{O}(-n-1). \end{align*} *Identifying the adjoint bundle.* On the overlap $U_i \cap U_j$ of the standard affine cover, the transition function of $\mathcal{O}(a)$ is $(z_j/z_i)^a$ and that of $\mathcal{O}(b)$ is $(z_j/z_i)^b$; their product, the transition function of $\mathcal{O}(a) \otimes \mathcal{O}(b)$, is $(z_j/z_i)^{a+b}$, which is the transition function of $\mathcal{O}(a+b)$. Applying this with $a = -n-1$ and $b = m$: \begin{align*} K_X \otimes L = \mathcal{O}_{\mathbb{P}^n}(-n-1) \otimes \mathcal{O}_{\mathbb{P}^n}(m) \cong \mathcal{O}_{\mathbb{P}^n}(m-n-1). \end{align*} *Positivity and vanishing.* The standard Hermitian metric on $\mathcal{O}(1)$ has curvature form $i\Theta(\mathcal{O}(1)) = \Omega_{\mathrm{FS}}$, the unnormalized Fubini-Study form, which is positive on every nonzero tangent vector. Since curvature is additive under tensor product of Hermitian line bundles, the metric $h^{\otimes m}$ on $\mathcal{O}(m) = \mathcal{O}(1)^{\otimes m}$ satisfies $i\Theta_{h^{\otimes m}}(\mathcal{O}(m)) = m\,\Omega_{\mathrm{FS}} > 0$, so $L = \mathcal{O}(m)$ is positive. [Kodaira Vanishing Theorem (Statement)](/theorems/3501) then gives \begin{align*} H^q\!\bigl(\mathbb{P}^n,\, K_{\mathbb{P}^n} \otimes L\bigr) = H^q\!\bigl(\mathbb{P}^n,\, \mathcal{O}_{\mathbb{P}^n}(m-n-1)\bigr) = 0 \qquad \text{for all } q > 0. \end{align*} The conclusion is nontrivial precisely in the range $1 \le m \le n$: there $m - n - 1 \le -2 < 0$, so $\mathcal{O}(m-n-1)$ has negative degree and carries no positive curvature of its own. For a concrete instance, take $n = 3$ and $m = 2$: the adjoint bundle is $\mathcal{O}(-2)$ on $\mathbb{P}^3$, and the isomorphism $\mathcal{O}(-2) \cong K_{\mathbb{P}^3} \otimes \mathcal{O}(2)$ is what [Kodaira Vanishing Theorem (Statement)](/theorems/3501) acts on — Kodaira vanishing applies not to $\mathcal{O}(-2)$ directly, but to the positive bundle $\mathcal{O}(2)$ whose adjoint twist it presents. The canonical factor $K_X$ is the precise shift that re-expresses a negatively twisted bundle as an adjoint bundle over something positive, placing the curvature in the bidegree where the Bochner–Kodaira estimate is coercive. [/example] The theorem will often be used with tensor powers. If $L$ is positive, then each $L^m$ with $m\ge 1$ is positive, so the same argument applies uniformly to $K_X\otimes L^m$. [remark: Role of the Kähler Condition] The Kähler identities are used to relate curvature, the Lefschetz operator, and the $\bar\partial$-Laplacian. Without those identities the Bochner formula does not give the same positivity estimate, and Kodaira-type vanishing may fail on non-Kähler compact complex manifolds. [/remark] ## Serre Duality and Negative Twists The next question is how to treat bundles with negative curvature. Instead of proving a new estimate in the opposite direction, we use duality to convert a negative bundle into the dual of a positive one. [definition: Dual Holomorphic Vector Bundle] Let $E\to X$ be a holomorphic vector bundle over a complex manifold $X$. The dual holomorphic vector bundle $E^*\to X$ has fibre $E_x^*=\operatorname{Hom}_{\mathbb C}(E_x,\mathbb C)$ over $x\in X$, with transition functions dual to the transition functions of $E$. [/definition] For a line bundle $M$, negativity means that $M^*$ is positive. At the level of Hermitian metrics, taking the dual reverses the sign of the Chern curvature. The point of duality is that low-degree cohomology of $M$ is not usually related to sections of $M$ itself; it is related to sections of the positive adjoint bundle $K_X\otimes M^*$. [example: Asymmetry on the Riemann Sphere] Let $X = \mathbb{P}^1$ and fix $k \ge 0$. We compute $H^q(\mathbb{P}^1, \mathcal{O}(k))$ for each $q$ and exhibit the asymmetry $H^0(\mathbb{P}^1, \mathcal{O}(-k-2)) = 0$ while $h^1(\mathbb{P}^1, \mathcal{O}(-k-2)) = k+1$. *The canonical bundle.* The Euler sequence on $\mathbb{P}^1$ is the short exact sequence of holomorphic vector bundles \begin{align*} 0 \longrightarrow \mathcal{O} \longrightarrow \mathcal{O}(1)^{\oplus 2} \longrightarrow T^{1,0}_{\mathbb{P}^1} \longrightarrow 0. \end{align*} For any short exact sequence $0 \to E' \to E \to E'' \to 0$ of holomorphic vector bundles, the determinant is multiplicative: $\det E \cong \det E' \otimes \det E''$. Applying this here gives $\det(\mathcal{O}(1)^{\oplus 2}) \cong \det(\mathcal{O}) \otimes \det(T^{1,0}_{\mathbb{P}^1})$. The determinant of a direct sum of line bundles is their tensor product: on the standard overlap $U_0 \cap U_1$, the transition matrix of $\mathcal{O}(1)^{\oplus 2}$ is $(z_1/z_0)\,\mathrm{Id}_2$, whose determinant is $(z_1/z_0)^2$, the transition function of $\mathcal{O}(2)$, so $\det(\mathcal{O}(1)^{\oplus 2}) \cong \mathcal{O}(2)$. The structure sheaf satisfies $\det(\mathcal{O}) = \mathcal{O}$ because its transition functions are all $1$ and $\det(1) = 1$. Canceling this trivial factor gives $\det(T^{1,0}_{\mathbb{P}^1}) \cong \mathcal{O}(2)$. Since $T^{1,0}_{\mathbb{P}^1}$ is a line bundle ($\mathbb{P}^1$ has complex dimension $1$), $\det(T^{1,0}_{\mathbb{P}^1}) = T^{1,0}_{\mathbb{P}^1}$, so $T^{1,0}_{\mathbb{P}^1} \cong \mathcal{O}(2)$. Dualizing — a line bundle with transition functions $\{g_{\alpha\beta}\}$ has dual with transition functions $\{g_{\alpha\beta}^{-1}\}$, giving $\mathcal{O}(2)^* \cong \mathcal{O}(-2)$ — yields \begin{align*} K_{\mathbb{P}^1} = \bigl(T^{1,0}_{\mathbb{P}^1}\bigr)^* \cong \mathcal{O}(-2). \end{align*} *Sections of $\mathcal{O}(k)$.* On $\mathbb{P}^1$ with standard cover $U_0 = \{z_0 \neq 0\}$ and $U_1 = \{z_1 \neq 0\}$, the bundle $\mathcal{O}(k)$ has transition function $g_{01} = (z_1/z_0)^k$ on $U_0 \cap U_1$. A global section is a pair of holomorphic functions $f_i \colon U_i \to \mathbb{C}$ satisfying $f_0 = g_{01} \cdot f_1$ on the overlap. In affine coordinates $w = z_1/z_0$ on $U_0 \cong \mathbb{C}$ and $u = z_0/z_1$ on $U_1 \cong \mathbb{C}$, the coordinate change on the overlap is $u = 1/w$, and the gluing condition reads \begin{align*} f_0(w) = w^k\, f_1(w^{-1}) \qquad \text{for } w \neq 0. \end{align*} Since $f_0$ is entire, write its Taylor series $f_0(w) = \sum_{j=0}^{\infty} a_j w^j$. Substituting $u = 1/w$ and solving for $f_1$: \begin{align*} f_1(u) = u^k\, f_0(u^{-1}) = u^k \sum_{j=0}^{\infty} a_j\, u^{-j} = \sum_{j=0}^{\infty} a_j\, u^{k-j}. \end{align*} For $f_1$ to be holomorphic at $u = 0$, every exponent must be nonneg: $k - j \geq 0$ for each $j$ with $a_j \neq 0$, so $a_j = 0$ for all $j > k$. The section $f_0$ is therefore a polynomial of degree at most $k$ in $w$. Concretely, $f_0(w) = w^j$ for each $0 \leq j \leq k$ yields $f_1(u) = u^{k-j}$, which is indeed entire; these $k+1$ sections are linearly independent (distinct monomials) and span the space. In homogeneous notation, $f_0(w) = (z_1/z_0)^j$ corresponds to the degree-$k$ monomial $z_0^{k-j}z_1^j$, giving \begin{align*} h^0\!\bigl(\mathbb{P}^1,\, \mathcal{O}(k)\bigr) = k+1, \end{align*} spanned by the $k+1$ monomials $z_0^k,\, z_0^{k-1}z_1,\, \ldots,\, z_1^k$. *Vanishing of $H^1(\mathbb{P}^1, \mathcal{O}(k))$.* The transition functions of a tensor product of line bundles multiply: on $U_0 \cap U_1$, the transition function of $\mathcal{O}(a) \otimes \mathcal{O}(b)$ is $(z_1/z_0)^a \cdot (z_1/z_0)^b = (z_1/z_0)^{a+b}$, the transition function of $\mathcal{O}(a+b)$. Taking $a = -2$ and $b = k+2$ gives $(-2)+(k+2) = k$, so $\mathcal{O}(-2) \otimes \mathcal{O}(k+2) \cong \mathcal{O}(k)$. With $K_{\mathbb{P}^1} \cong \mathcal{O}(-2)$, this reads $\mathcal{O}(k) \cong K_{\mathbb{P}^1} \otimes \mathcal{O}(k+2)$. The standard metric on $\mathcal{O}(1)$ has curvature $i\Theta(\mathcal{O}(1)) = \Omega_{\mathrm{FS}}$; since curvature is additive under tensor product of Hermitian line bundles, the induced metric on $\mathcal{O}(k+2) = \mathcal{O}(1)^{\otimes(k+2)}$ satisfies $i\Theta(\mathcal{O}(k+2)) = (k+2)\,\Omega_{\mathrm{FS}}$. For $k \geq 0$, the factor $k+2 \geq 2 > 0$ makes $\mathcal{O}(k+2)$ positive. [Kodaira Vanishing Theorem (Statement)](/theorems/3501) applied to $L = \mathcal{O}(k+2)$ then gives \begin{align*} H^1\!\bigl(\mathbb{P}^1,\, K_{\mathbb{P}^1} \otimes \mathcal{O}(k+2)\bigr) = H^1\!\bigl(\mathbb{P}^1,\, \mathcal{O}(k)\bigr) = 0. \end{align*} *The negative twist and Serre duality.* For the bundle $\mathcal{O}(-k-2)$, the gluing condition for a section is $f_0(w) = w^{-k-2} f_1(w^{-1})$, yielding $f_1(u) = u^{-k-2} f_0(u^{-1}) = \sum_{j=0}^{\infty} a_j\, u^{-k-2-j}$. Every exponent $-k-2-j \leq -k-2 \leq -2 < 0$ for $j \geq 0$ and $k \geq 0$, so holomorphicity of $f_1$ at $u = 0$ forces $a_j = 0$ for all $j$: thus $H^0(\mathbb{P}^1, \mathcal{O}(-k-2)) = 0$. To compute $H^1$, apply *Serre Duality for Holomorphic Vector Bundles* with $n = 1$, $q = 1$, and $E = \mathcal{O}(-k-2)$: \begin{align*} H^1\!\bigl(\mathbb{P}^1,\,\mathcal{O}(-k-2)\bigr)^* \cong H^0\!\bigl(\mathbb{P}^1,\, K_{\mathbb{P}^1} \otimes \mathcal{O}(-k-2)^*\bigr). \end{align*} The transition function of $\mathcal{O}(a)$ on $U_0 \cap U_1$ is $(z_1/z_0)^a$, so the dual bundle has transition function $(z_1/z_0)^{-a}$, the transition function of $\mathcal{O}(-a)$; in particular $\mathcal{O}(-k-2)^* \cong \mathcal{O}(k+2)$. Tensoring with $K_{\mathbb{P}^1} = \mathcal{O}(-2)$ and using additivity of twist under tensor product, \begin{align*} K_{\mathbb{P}^1} \otimes \mathcal{O}(-k-2)^* \cong \mathcal{O}(-2) \otimes \mathcal{O}(k+2) \cong \mathcal{O}(k), \end{align*} where the last identification uses $(-2)+(k+2) = k$ as before. Substituting into the duality isomorphism: \begin{align*} H^1\!\bigl(\mathbb{P}^1,\,\mathcal{O}(-k-2)\bigr)^* \cong H^0\!\bigl(\mathbb{P}^1,\,\mathcal{O}(k)\bigr). \end{align*} Taking dimensions gives $h^1(\mathbb{P}^1, \mathcal{O}(-k-2)) = k+1$. The bundle $\mathcal{O}(-k-2)$ carries no global sections — the gluing exponents are strictly negative — yet its first cohomology is $(k+1)$-dimensional, matching the space of degree-$k$ polynomials term for term: duality reads the entire cohomology of the negative twist off the positive twist, without any direct computation on $\mathcal{O}(-k-2)$ itself, and this translation between opposite curvature signs is precisely the asymmetry that *Serre Duality for Holomorphic Vector Bundles* makes computable. [/example] The theorem below is the precise global form of this pairing. The preceding example on $\mathbb P^1$ shows the pattern in one dimension: negative cohomology is measured by sections of a dual positive twist after inserting the canonical bundle. In higher dimension the same phenomenon requires a pairing between degree $q$ classes and complementary degree $n-q$ classes. The theorem packages this pairing as a perfect duality, making precise which bundle has to appear on the opposite side. [quotetheorem:3864] [citeproof:3864] Several hypotheses are doing real work here. Compactness makes the integral pairing finite and gives finite-dimensional cohomology groups; on a non-compact complex manifold, top-degree integrals may fail to converge and the algebraic dual of cohomology is usually too large. The pairing is also not just a formal pairing of vector spaces: it uses the complex orientation of $X$, integration of smooth top-degree forms, and the contraction between $E$ and $E^*$. The theorem is holomorphic in content even though the proof uses smooth forms. Dolbeault theory identifies holomorphic sheaf cohomology with smooth $(0,q)$-forms modulo $\bar\partial$, and the dual group involves $K_X\otimes E^*$ rather than the smooth dual bundle alone. Thus Serre duality does not say that arbitrary smooth cohomology groups pair in this way; it is a statement about the holomorphic category expressed through elliptic analysis. The next consequence explains how positivity results control negative line bundles. Serre duality changes a cohomology group of a negative bundle into the dual of a complementary-degree group with a positive adjoint twist, and Kodaira vanishing then kills that dual group outside the top degree. This is the standard mechanism behind vanishing for negative line bundles. [quotetheorem:3865] [citeproof:3865] This explains the familiar pattern that negative twists have possible cohomology only near the top degree, but the theorem deliberately stops at $q<n$. The top group can be nonzero: on $\mathbb P^n$, Serre duality gives \begin{align*} H^n(\mathbb P^n,\mathcal O(-n-1))^*\cong H^0(\mathbb P^n,\mathcal O)=\mathbb C. \end{align*} Negativity is also stronger than being a nontrivial line bundle. A degree-zero line bundle on an elliptic curve may be nontrivial without having either positive or negative curvature, and Kodaira vanishing gives no conclusion for it. Finally, Serre duality itself holds on compact complex manifolds, but the vanishing step uses the Kähler Bochner estimate; on non-Kähler manifolds this curvature argument can fail. In projective space these principles recover the usual line-bundle cohomology table and also show how to compute a concrete boundary case rather than merely quote the table. [example: Cohomology of Line Bundles on Projective Space] Let $X = \mathbb{P}^n$ and write $\mathcal{O}(k) = \mathcal{O}_{\mathbb{P}^n}(k)$. We determine $H^q(\mathbb{P}^n, \mathcal{O}(k))$ for every integer $k$ and every degree $0 \leq q \leq n$. *Sections.* A global section of $\mathcal{O}(k)$ is, by definition of the tautological construction, a holomorphic function $f \colon \mathbb{C}^{n+1} \setminus \{0\} \to \mathbb{C}$ satisfying $f(\lambda z) = \lambda^k f(z)$ for every $\lambda \in \mathbb{C}^*$. The origin $\{0\} \subset \mathbb{C}^{n+1}$ has complex codimension $n+1 \geq 2$, so [Hartogs Extension Theorem](/theorems/3401) extends $f$ holomorphically to all of $\mathbb{C}^{n+1}$. Write the convergent power series $f(z) = \sum_\alpha c_\alpha z^\alpha$. Substituting $\lambda z$ in place of $z$ gives $f(\lambda z) = \sum_\alpha c_\alpha \lambda^{|\alpha|} z^\alpha$, while $\lambda^k f(z) = \lambda^k \sum_\alpha c_\alpha z^\alpha$. Equating the coefficient of each monomial $z^\alpha$ in both expressions forces $c_\alpha (\lambda^{|\alpha|} - \lambda^k) = 0$ for every $\lambda \in \mathbb{C}^*$; since $\lambda \mapsto \lambda^{|\alpha|} - \lambda^k$ is not identically zero on $\mathbb{C}^*$ unless $|\alpha| = k$, this forces $c_\alpha = 0$ whenever $|\alpha| \neq k$. For $k < 0$ every multi-index satisfies $|\alpha| \geq 0 > k$, so all coefficients vanish and $H^0(\mathbb{P}^n, \mathcal{O}(k)) = 0$. For $k \geq 0$, the nonzero terms are exactly the monomials $z^\alpha$ with $|\alpha| = \alpha_0 + \cdots + \alpha_n = k$ and each $\alpha_i \geq 0$; by the stars-and-bars formula, the number of such multi-indices is $\binom{k + n}{n}$ (place $k$ indistinguishable stars into $n+1$ bins). These monomials are linearly independent over $\mathbb{C}$, so \begin{align*} h^0\!\bigl(\mathbb{P}^n,\, \mathcal{O}(k)\bigr) = \binom{k+n}{n}. \end{align*} *Intermediate vanishing.* For $0 < q < n$ and any $k \in \mathbb{Z}$, we show $H^q(\mathbb{P}^n, \mathcal{O}(k)) = 0$ by splitting into two complementary ranges. *Case $k \geq -n$.* Then $k + n + 1 \geq 1 > 0$. On the standard affine cover of $\mathbb{P}^n$, the transition function of $\mathcal{O}(a)$ on $U_i \cap U_j$ is $(z_j/z_i)^a$; that of the tensor product $\mathcal{O}(a) \otimes \mathcal{O}(b)$ is therefore $(z_j/z_i)^a \cdot (z_j/z_i)^b = (z_j/z_i)^{a+b}$, which is the transition function of $\mathcal{O}(a+b)$, so $\mathcal{O}(a) \otimes \mathcal{O}(b) \cong \mathcal{O}(a+b)$. The canonical bundle satisfies $K_{\mathbb{P}^n} \cong \mathcal{O}(-n-1)$ (computed from the Euler sequence as in the preceding examples). Taking $a = -n-1$ and $b = k+n+1$ gives $a + b = k$, so \begin{align*} \mathcal{O}(k) \cong K_{\mathbb{P}^n} \otimes \mathcal{O}(k+n+1). \end{align*} The standard Hermitian metric on $\mathcal{O}(1)$ has curvature $i\Theta(\mathcal{O}(1)) = \Omega_{\mathrm{FS}}$; since curvature is additive under tensor product of Hermitian line bundles, the induced metric on $\mathcal{O}(k+n+1) = \mathcal{O}(1)^{\otimes(k+n+1)}$ satisfies $i\Theta(\mathcal{O}(k+n+1)) = (k+n+1)\,\Omega_{\mathrm{FS}}$. Because $k + n + 1 \geq 1 > 0$, this curvature form is positive, so $\mathcal{O}(k+n+1)$ is positive. [Kodaira Vanishing Theorem (Statement)](/theorems/3501) then gives \begin{align*} H^q\!\bigl(\mathbb{P}^n,\, K_{\mathbb{P}^n} \otimes \mathcal{O}(k+n+1)\bigr) = H^q\!\bigl(\mathbb{P}^n,\, \mathcal{O}(k)\bigr) = 0 \qquad \text{for all } q > 0, \end{align*} which in particular covers every $q$ in the range $0 < q < n$. *Case $k \leq -n-1$.* Apply *Serre Duality for Holomorphic Vector Bundles* with $E = \mathcal{O}(k)$ and ambient dimension $n$: \begin{align*} H^q\!\bigl(\mathbb{P}^n,\, \mathcal{O}(k)\bigr)^* \cong H^{n-q}\!\bigl(\mathbb{P}^n,\, K_{\mathbb{P}^n} \otimes \mathcal{O}(k)^*\bigr). \end{align*} The transition function of $\mathcal{O}(k)$ on $U_i \cap U_j$ is $(z_j/z_i)^k$, so the dual bundle has transition function $(z_j/z_i)^{-k}$, giving $\mathcal{O}(k)^* \cong \mathcal{O}(-k)$. Tensoring with $K_{\mathbb{P}^n} = \mathcal{O}(-n-1)$ and using additivity of twist: \begin{align*} K_{\mathbb{P}^n} \otimes \mathcal{O}(k)^* \cong \mathcal{O}(-n-1) \otimes \mathcal{O}(-k) \cong \mathcal{O}(-k-n-1), \end{align*} where $(-n-1) + (-k) = -k - n - 1$. From the assumption $k \leq -n-1$ we get $-k \geq n+1$, hence $-k - n - 1 \geq 0$; in particular $-k > 0$ and $\mathcal{O}(-k)$ is positive. Writing $\mathcal{O}(-k-n-1) = \mathcal{O}(-n-1) \otimes \mathcal{O}(-k) = K_{\mathbb{P}^n} \otimes \mathcal{O}(-k)$, and noting that $n - q \geq 1$ because $q < n$, [Kodaira Vanishing Theorem (Statement)](/theorems/3501) applied to the positive bundle $\mathcal{O}(-k)$ gives $H^{n-q}\!\bigl(\mathbb{P}^n,\, K_{\mathbb{P}^n} \otimes \mathcal{O}(-k)\bigr) = 0$. The duality isomorphism then forces $H^q\!\bigl(\mathbb{P}^n,\, \mathcal{O}(k)\bigr)^* = 0$, so $H^q\!\bigl(\mathbb{P}^n,\, \mathcal{O}(k)\bigr) = 0$. *Top cohomology and the canonical threshold.* Applying *Serre Duality for Holomorphic Vector Bundles* in degree $q = n$ and using $\mathcal{O}(k)^* \cong \mathcal{O}(-k)$ as before: \begin{align*} H^n\!\bigl(\mathbb{P}^n,\, \mathcal{O}(k)\bigr)^* &\cong H^{n-n}\!\bigl(\mathbb{P}^n,\, K_{\mathbb{P}^n} \otimes \mathcal{O}(-k)\bigr) = H^0\!\bigl(\mathbb{P}^n,\, \mathcal{O}(-n-1) \otimes \mathcal{O}(-k)\bigr)\\ &= H^0\!\bigl(\mathbb{P}^n,\, \mathcal{O}(-k-n-1)\bigr). \end{align*} By the $H^0$ formula established above, $h^0\!\bigl(\mathbb{P}^n, \mathcal{O}(r)\bigr) = \binom{r+n}{n}$ for $r \geq 0$ and $0$ for $r < 0$. Setting $r = -k-n-1$, this space is nonzero if and only if $-k-n-1 \geq 0$, i.e.\ $k \leq -n-1$; when that holds, $h^n\!\bigl(\mathbb{P}^n, \mathcal{O}(k)\bigr) = \binom{(-k-n-1)+n}{n} = \binom{-k-1}{n}$. At the boundary twist $k = -n-1$, substituting into $-k-n-1$ gives $-(-n-1)-n-1 = (n+1)-n-1 = 0$, so $H^0\!\bigl(\mathbb{P}^n, \mathcal{O}(0)\bigr) = H^0\!\bigl(\mathbb{P}^n, \mathcal{O}\bigr) \cong \mathbb{C}$, spanned by the unique degree-zero monomial $1$. Duality then gives $H^n\!\bigl(\mathbb{P}^n, \mathcal{O}(-n-1)\bigr) \cong \mathbb{C}$, one-dimensional. At the adjacent twist $k = -n$, the dual degree is $-(-n)-n-1 = n-n-1 = -1 < 0$, so $H^0\!\bigl(\mathbb{P}^n, \mathcal{O}(-1)\bigr) = 0$ and duality forces $H^n\!\bigl(\mathbb{P}^n, \mathcal{O}(-n)\bigr) = 0$. The threshold $k = -n-1$ is the twist of $K_{\mathbb{P}^n}$ itself: top cohomology first appears at the canonical bundle because that is precisely where the dual degree $-k-n-1$ crosses from negative to zero and the corresponding space of homogeneous polynomials first becomes nonempty. [/example] The same reasoning applies to arbitrary compact Kähler manifolds after taking negative tensor powers of a positive line bundle. This is often the practical form of the result: once a single positive bundle $L$ is known, all inverse powers $L^{-m}$ have a uniform curvature sign, and every low-degree cohomology group disappears. The only group left to understand is the top cohomology, which duality converts back into sections of the positive adjoint powers $K_X\otimes L^m$. [example: High Negative Powers] Let $L \to X$ be a positive holomorphic line bundle on a compact Kähler manifold of complex dimension $n$, and let $m \ge 1$. We show that $H^q(X, L^{-m}) = 0$ for every $q < n$, and that the remaining top cohomology group is dual to $H^0(X, K_X \otimes L^m)$. *Negativity of $L^{-m}$.* Choose a positive Hermitian metric $h$ on $L$, so that $i\Theta_h(L) > 0$. In a local holomorphic frame $e$ for $L$, write $|e|_h^2 = e^{-\varphi}$ for a smooth weight function $\varphi$; positivity means $i\partial\bar\partial\varphi > 0$. The tensor power $L^m = L^{\otimes m}$ is trivialized by the frame $e^{\otimes m}$, and the induced metric $h^{\otimes m}$ satisfies $|e^{\otimes m}|_{h^{\otimes m}}^2 = |e|_h^{2m} = e^{-m\varphi}$, so its local weight is $m\varphi$. Since the curvature of a Hermitian line bundle in a local frame with weight $\psi$ is $\Theta = \partial\bar\partial\psi$, we get $\Theta_{h^{\otimes m}}(L^m) = \partial\bar\partial(m\varphi) = m\,\partial\bar\partial\varphi = m\,\Theta_h(L)$, and hence $i\Theta_{h^{\otimes m}}(L^m) = m\,i\Theta_h(L) > 0$, so $L^m$ is positive. The inverse bundle $L^{-m}$ carries the dual metric $(h^{\otimes m})^*$, defined locally by $|e^{-\otimes m}|_{(h^{\otimes m})^*}^2 = |e^{\otimes m}|_{h^{\otimes m}}^{-2} = e^{m\varphi}$; the local weight of the dual metric is therefore $-m\varphi$, giving curvature $\Theta_{(h^{\otimes m})^*}(L^{-m}) = \partial\bar\partial(-m\varphi) = -m\,\Theta_h(L)$. Thus $i\Theta(L^{-m}) = -m\,i\Theta_h(L) < 0$, and $L^{-m}$ is negative. *Vanishing for $q < n$.* Since $L^{-m}$ is negative, *Vanishing for Negative Line Bundles* gives $H^q(X, L^{-m}) = 0$ for every $q < n$. *The top group via duality.* For $q = n$, apply *Serre Duality for Holomorphic Vector Bundles* with $E = L^{-m}$ and ambient dimension $n$; since $n - n = 0$, the duality isomorphism reads \begin{align*} H^n(X, L^{-m})^* \cong H^0\!\bigl(X,\, K_X \otimes (L^{-m})^*\bigr). \end{align*} To identify $(L^{-m})^*$: if $\{g_{\alpha\beta}\}$ are the transition functions of $L$ on a cover $\{U_\alpha\}$, then $L^{-m}$ has transition functions $\{g_{\alpha\beta}^{-m}\}$, and the dual bundle has transition functions $\{(g_{\alpha\beta}^{-m})^{-1}\} = \{g_{\alpha\beta}^{m}\}$, which are exactly the transition functions of $L^m$, so $(L^{-m})^* \cong L^m$. Substituting, \begin{align*} H^n(X, L^{-m})^* \cong H^0(X,\, K_X \otimes L^m). \end{align*} Since $L^m$ is positive, [Kodaira Vanishing Theorem (Statement)](/theorems/3501) gives $H^q(X, K_X \otimes L^m) = 0$ for every $q > 0$. The holomorphic Euler characteristic therefore collapses to \begin{align*} \chi(X, K_X \otimes L^m) = \sum_{q=0}^n (-1)^q\, h^q(X, K_X \otimes L^m) = h^0(X, K_X \otimes L^m) - 0 + 0 - \cdots = \dim H^0(X, K_X \otimes L^m), \end{align*} making $\dim H^0(X, K_X \otimes L^m)$ computable from *Hirzebruch Riemann Roch for Line Bundles* without knowing any higher cohomology. Taking dimensions in the duality isomorphism, \begin{align*} h^n(X, L^{-m}) = \dim H^0(X, K_X \otimes L^m) = \chi(X, K_X \otimes L^m). \end{align*} Negativity concentrates all the cohomology of $L^{-m}$ into the single top degree $n$, and duality translates that surviving group into a space of holomorphic sections of the positive adjoint bundle $K_X \otimes L^m$, where curvature is coercive and the dimension is governed by topology alone. [/example] ## Hilbert Polynomials and Growth of Sections The final problem is numerical. Vanishing identifies some Euler characteristics with dimensions of spaces of sections; Riemann--Roch computes those Euler characteristics from characteristic classes. [definition: Holomorphic Euler Characteristic] Let $X$ be a compact complex manifold of dimension $n$, and let $E\to X$ be a holomorphic vector bundle. The holomorphic Euler characteristic of $E$ is \begin{align*} \chi(X,E)=\sum_{q=0}^n(-1)^q\dim H^q(X,E). \end{align*} [/definition] The alternating sum is stable under exact sequences and is often accessible by topology. When higher cohomology vanishes, it becomes the ordinary dimension of $H^0$. Before vanishing is available, the Euler characteristic can still be computable even when the individual cohomology groups move in families. [example: Euler Characteristic on a Curve] Let $C$ be a compact Riemann surface of genus $g$, and let $L \to C$ be a holomorphic line bundle of degree $d$. We show that $\chi(C, L) = d + 1 - g$. Since $C$ has complex dimension $n = 1$, the cohomology groups $H^q(C, L)$ vanish for $q > 1$ by dimension reasons, so the definition of the holomorphic Euler characteristic gives $\chi(C, L) = h^0(C, L) - h^1(C, L)$. Applying [Riemann–Roch Theorem](/theorems/2185) to $L$ on the curve $C$ yields \begin{align*} h^0(C, L) - h^0(C, K_C \otimes L^{-1}) = \deg(L) + 1 - g = d + 1 - g, \end{align*} where $K_C$ denotes the canonical bundle of $C$. The left-hand side involves $h^0(C, K_C \otimes L^{-1})$ rather than $h^1(C, L)$, so the two expressions for $\chi$ agree only once the two quantities are identified. To do this, apply *Serre Duality for Holomorphic Vector Bundles* with $n = 1$, $q = 1$, and $E = L$: the dual of a line bundle with transition functions $\{g_{\alpha\beta}\}$ has transition functions $\{g_{\alpha\beta}^{-1}\}$, so $L^* = L^{-1}$, and the duality isomorphism reads \begin{align*} H^1(C, L)^* \cong H^{1-1}\!\bigl(C,\, K_C \otimes L^{-1}\bigr) = H^0\!\bigl(C,\, K_C \otimes L^{-1}\bigr). \end{align*} Taking dimensions on both sides, $h^1(C, L) = h^0(C, K_C \otimes L^{-1})$. Substituting into the Riemann–Roch equation: \begin{align*} \chi(C, L) = h^0(C, L) - h^1(C, L) = h^0(C, L) - h^0(C, K_C \otimes L^{-1}) = d + 1 - g. \end{align*} The nontrivial content of this identity appears in the range $0 < d < 2g-2$. In that range, $\deg L = d > 0$ and $\deg(K_C \otimes L^{-1}) = \deg K_C - \deg L = (2g-2) - d > 0$ (using additivity of degree under tensor product: $\deg(E \otimes F) = \deg E + \deg F$ for line bundles, applied with $\deg K_C = 2g-2$ and $\deg L^{-1} = -d$), so neither $H^0$ term in Riemann–Roch is forced to vanish by a degree argument. Different line bundles of the same degree $d$ can then have different values of $h^0(C, L)$, with $h^1(C, L)$ shifting in lockstep to keep their difference fixed at $d + 1 - g$. The Euler characteristic is computable from topology alone without knowing either individual dimension — which is the essential utility of Riemann–Roch before Kodaira vanishing has removed the $h^1$ term entirely. [/example] Riemann--Roch computes an Euler characteristic, but embedding questions need a way to track how many sections appear after taking higher tensor powers of a line bundle. The useful invariant is therefore not just the single number $\chi(X,L)$, but the whole function of the exponent $m$ for $L^m$. When this function eventually agrees with a polynomial, its degree and leading coefficient record the asymptotic supply of sections. [definition: Hilbert Polynomial of a Line Bundle] Let $X$ be a compact complex manifold and let $L\to X$ be a holomorphic line bundle. The Hilbert function is \begin{align*} m\longmapsto \chi(X,L^m),\qquad m\in\mathbb Z. \end{align*} If this function agrees with a polynomial for all sufficiently large $m$, that polynomial is the Hilbert polynomial of $L$. [/definition] The polynomial nature comes from Hirzebruch--Riemann--Roch. In this section the theorem is used as an input rather than proved. At this point the analytic vanishing theorems tell us when higher cohomology disappears, but they do not by themselves compute the remaining dimension. Hirzebruch--Riemann--Roch supplies the missing numerical formula by expressing the Euler characteristic as an integral of characteristic classes. For powers of a line bundle, this integral becomes polynomial in the tensor exponent, so it is the bridge from curvature classes to asymptotic counts of sections. [quotetheorem:3866] [citeproof:3866] For a smooth compact complex manifold, the index formula does not require a Kähler metric, but without Kähler positivity one should not expect Kodaira vanishing, so $\chi(X,L^m)$ may not equal $h^0(X,L^m)$. For singular spaces the formula must be replaced by a singular Riemann--Roch theorem with the appropriate Todd class in homology, so the smooth formula above should not be applied directly. In low dimensions the formula becomes familiar. On a curve, it gives $\chi(X,L)=\deg L+1-g$. On a surface, it gives \begin{align*} \chi(X,L)=\frac{1}{2}c_1(L)\bigl(c_1(L)-c_1(K_X)\bigr)+\chi(X,\mathcal O_X), \end{align*} where the product denotes the intersection pairing. The asymptotic statement needed for embedding is obtained by applying Riemann--Roch to high tensor powers and then using Kodaira vanishing to replace Euler characteristic by actual section dimension. This turns a characteristic-class integral into a concrete growth estimate for $H^0(X,K_X\otimes L^m)$, with the top self-intersection of $L$ controlling the leading term. For the embedding argument, the qualitative fact that sections exist is not enough; one needs a numerical supply that grows with the tensor power. The relevant estimate must live in the adjoint setting, because Kodaira vanishing is then strong enough to remove higher cohomology and leave the Euler characteristic as the actual dimension of sections. [quotetheorem:3867] [citeproof:3867] Since $L$ is positive, the top self-intersection $\int_X c_1(L)^n$ is positive. Thus high adjoint powers have section spaces whose dimension grows on the order of $m^n$. The lower-order terms are not analytic error terms; they are fixed topological contributions from $c_1(K_X)$ and the Todd class in the Riemann--Roch integral. The equality with the Euler characteristic is sharp for every $m\ge 1$ in this adjoint setting because Kodaira vanishing removes all higher cohomology. The asymptotic symbol only concerns the leading term of the resulting polynomial. If $L$ is merely semipositive, the curvature commutator in the Bochner argument can have zero directions, higher cohomology may survive, and the same section count need not follow. This growth statement is the numerical half of Kodaira embedding. Later embedding arguments require enough sections to separate points and tangent directions; the result here explains why positive line bundles eventually produce section spaces large enough for that projective-geometric role. [example: Hilbert Polynomial of Projective Space] For $X = \mathbb{P}^n$ and $L = \mathcal{O}(1)$, we compute $h^0(\mathbb{P}^n, \mathcal{O}(m))$ for all integers $m \geq 0$, identify the result as the Hilbert polynomial of $L$, and verify that its leading coefficient matches $\frac{1}{n!}\int_{\mathbb{P}^n} c_1(\mathcal{O}(1))^n$. A global section of $\mathcal{O}(m)$ is, by the tautological construction, a holomorphic function $f \colon \mathbb{C}^{n+1} \setminus \{0\} \to \mathbb{C}$ satisfying $f(\lambda z) = \lambda^m f(z)$ for every $\lambda \in \mathbb{C}^*$. The origin $\{0\} \subset \mathbb{C}^{n+1}$ has complex codimension $n+1 \geq 2$, so [Hartogs Extension Theorem](/theorems/3401) extends $f$ uniquely to an entire function on all of $\mathbb{C}^{n+1}$. Write the everywhere-convergent power series $f(z) = \sum_\alpha c_\alpha z^\alpha$ (sum over multi-indices $\alpha \in \mathbb{Z}_{\geq 0}^{n+1}$). Substituting $\lambda z$ in place of $z$ gives \begin{align*} f(\lambda z) = \sum_\alpha c_\alpha \lambda^{|\alpha|} z^\alpha, \qquad \lambda^m f(z) = \lambda^m \sum_\alpha c_\alpha z^\alpha. \end{align*} Equating the coefficient of the monomial $z^\alpha$ in both expressions yields $c_\alpha(\lambda^{|\alpha|} - \lambda^m) = 0$ for every $\lambda \in \mathbb{C}^*$. The function $\lambda \mapsto \lambda^{|\alpha|} - \lambda^m$ is not identically zero on $\mathbb{C}^*$ unless $|\alpha| = m$ (for $|\alpha| \neq m$, setting $\lambda = e^{2\pi i/(|\alpha|-m)}$ gives a root but the function still has a nonzero value at $\lambda = 2$, for instance), so $c_\alpha = 0$ whenever $|\alpha| \neq m$. For $m \geq 0$, the surviving terms are precisely the homogeneous polynomials of degree $m$ in $z_0, \ldots, z_n$. Each such polynomial is determined by an exponent vector $\alpha = (\alpha_0, \ldots, \alpha_n)$ with each $\alpha_i \geq 0$ and $\alpha_0 + \cdots + \alpha_n = m$: choosing it amounts to placing $m$ indistinguishable stars into $n+1$ distinct bins, one bin per variable, so the stars-and-bars formula gives $\binom{m + (n+1) - 1}{(n+1) - 1} = \binom{m+n}{n}$ multi-indices. The corresponding monomials $z^\alpha$ are linearly independent over $\mathbb{C}$ (distinct monomials cannot sum to zero), so they form a basis and \begin{align*} h^0\!\bigl(\mathbb{P}^n,\, \mathcal{O}(m)\bigr) = \binom{m+n}{n}. \end{align*} To confirm that no higher-cohomology cancellation enters the Euler characteristic, we identify $\mathcal{O}(m)$ as an adjoint bundle and apply vanishing. On the standard cover of $\mathbb{P}^n$, the transition function of $\mathcal{O}(a)$ on $U_i \cap U_j$ is $(z_j/z_i)^a$, and that of a tensor product $\mathcal{O}(a) \otimes \mathcal{O}(b)$ is $(z_j/z_i)^a \cdot (z_j/z_i)^b = (z_j/z_i)^{a+b}$, so $\mathcal{O}(a) \otimes \mathcal{O}(b) \cong \mathcal{O}(a+b)$. Taking $a = -n-1$ and $b = m+n+1$ gives $a + b = (-n-1) + (m+n+1) = m$, so \begin{align*} \mathcal{O}(m) \cong \mathcal{O}(-n-1) \otimes \mathcal{O}(m+n+1) \cong K_{\mathbb{P}^n} \otimes \mathcal{O}(m+n+1), \end{align*} where $K_{\mathbb{P}^n} \cong \mathcal{O}(-n-1)$ is the identification computed from the Euler sequence in the preceding examples. For $m \geq 0$, the exponent satisfies $m+n+1 \geq n+1 \geq 1$; since the standard Hermitian metric on $\mathcal{O}(1)$ has curvature $i\Theta(\mathcal{O}(1)) = \Omega_{\mathrm{FS}}$ and curvature is additive under tensor product of Hermitian line bundles, the induced metric on $\mathcal{O}(m+n+1) = \mathcal{O}(1)^{\otimes(m+n+1)}$ satisfies $i\Theta(\mathcal{O}(m+n+1)) = (m+n+1)\,\Omega_{\mathrm{FS}}$. The factor $m+n+1 \geq 1 > 0$ makes this form positive on every nonzero tangent vector, so $\mathcal{O}(m+n+1)$ is positive. [Kodaira Vanishing Theorem (Statement)](/theorems/3501) then gives $H^q\!\bigl(\mathbb{P}^n, K_{\mathbb{P}^n} \otimes \mathcal{O}(m+n+1)\bigr) = H^q\!\bigl(\mathbb{P}^n, \mathcal{O}(m)\bigr) = 0$ for every $q > 0$. Therefore \begin{align*} \chi\!\bigl(\mathbb{P}^n,\, \mathcal{O}(m)\bigr) = \sum_{q=0}^n (-1)^q h^q\!\bigl(\mathbb{P}^n, \mathcal{O}(m)\bigr) = h^0\!\bigl(\mathbb{P}^n, \mathcal{O}(m)\bigr) = \binom{m+n}{n}, \end{align*} and the binomial is already the Hilbert polynomial exactly, not an approximation. To extract the leading coefficient, write out the numerator factor by factor: \begin{align*} \binom{m+n}{n} = \frac{(m+n)(m+n-1)\cdots(m+1)}{n!}. \end{align*} The numerator is the product $(m+n)(m+n-1)\cdots(m+1)$, which has exactly $n$ factors; the $k$-th factor (for $k = 1, \ldots, n$) is $m + (n+1-k)$, so the product equals $\prod_{k=1}^{n}(m+k)$. Expanding, the term of highest degree in $m$ comes from taking $m$ from each of the $n$ factors, giving $m^n$; every other term in the expansion contains at most $n-1$ factors of $m$ and is therefore of degree at most $n-1$. Dividing the numerator by $n!$, the Hilbert polynomial $\binom{m+n}{n}$ is a polynomial in $m$ of degree $n$ with leading coefficient $\frac{1}{n!}$. For the intersection number, note that $c_1(\mathcal{O}(1))$ represents the cohomology class Poincaré dual to a hyperplane $H \subset \mathbb{P}^n$, so $\int_{\mathbb{P}^n} c_1(\mathcal{O}(1))^n$ counts the signed intersection multiplicity of $n$ generic hyperplanes $H_1, \ldots, H_n$. In homogeneous coordinates, each $H_i$ is the zero locus of a linear form $\ell_i(z_0, \ldots, z_n) = 0$. The simultaneous vanishing $\ell_1 = \cdots = \ell_n = 0$ is a homogeneous linear system of $n$ equations in $n+1$ unknowns; since the $\ell_i$ are chosen generically, the coefficient matrix has rank $n$, so by the rank-nullity theorem the solution space in $\mathbb{C}^{n+1}$ has dimension $(n+1) - n = 1$. A one-dimensional subspace of $\mathbb{C}^{n+1}$ corresponds to exactly one point of $\mathbb{P}^n$, and at a generic intersection the hyperplanes meet transversally with intersection multiplicity $1$. Thus \begin{align*} \int_{\mathbb{P}^n} c_1\!\bigl(\mathcal{O}(1)\bigr)^n = 1. \end{align*} The leading coefficient $1/n!$ of the Hilbert polynomial therefore equals $\frac{1}{n!}\int_{\mathbb{P}^n} c_1(\mathcal{O}(1))^n$, in agreement with the general asymptotic formula from *Hirzebruch Riemann Roch for Line Bundles*. To see why only this term survives as the leading power of $m$: the HRR integral $\int_{\mathbb{P}^n} e^{mc_1(\mathcal{O}(1))} \operatorname{td}(T^{1,0}\mathbb{P}^n)$ is a sum over degrees of the form $\int_{\mathbb{P}^n} \frac{(mc_1(\mathcal{O}(1)))^k}{k!} \cdot [\operatorname{td}(T^{1,0}\mathbb{P}^n)]_{n-k}$, where $[\cdot]_{n-k}$ denotes the degree-$2(n-k)$ part; the $k = n$ term contributes $\frac{m^n}{n!} \int_{\mathbb{P}^n} c_1(\mathcal{O}(1))^n \cdot [\operatorname{td}]_0$, and since $[\operatorname{td}]_0 = 1$ (the degree-zero part of any Todd class is the rank of the trivial summand, normalized to $1$), this equals $\frac{m^n}{n!}$. Every other $k < n$ contributes a power $m^k$ with $k < n$, so they are lower-order in $m$ and do not affect the leading coefficient. This example displays all three mechanisms of the general theory in a setting where the answer is completely explicit: [Hartogs Extension Theorem](/theorems/3401) pins down exactly which homogeneous functions are sections, a stars-and-bars count gives the Hilbert polynomial as an exact binomial coefficient with no remainder, and [Kodaira Vanishing Theorem (Statement)](/theorems/3501) ensures the Euler characteristic is pure $H^0$ — leaving the self-intersection number $\int_{\mathbb{P}^n} c_1(\mathcal{O}(1))^n = 1$ directly readable from the leading coefficient $1/n!$. [/example] The projective-space calculation is the model case for the general adjoint asymptotic formula because it shows all three mechanisms in a setting where the answer is visible: positivity supplies many homogeneous polynomials, vanishing leaves no hidden higher cohomology, and the leading coefficient is an intersection number. On a general manifold the same three pieces remain, but the Todd class and canonical bundle contribute lower-order terms that are invisible for the leading projective-space count. The next example works out the one-dimensional adjoint case, where those lower-order terms can be read exactly. [example: Adjoint Powers on a Curve] Let $C$ be a compact Riemann surface of genus $g$, and let $L \to C$ be a positive holomorphic line bundle of degree $d > 0$; we compute $\dim H^0(C, K_C \otimes L^m)$ exactly for every integer $m \geq 1$. *The degree.* For any two holomorphic line bundles $E, F$ on $C$, the transition functions of $E \otimes F$ on an overlap $U_\alpha \cap U_\beta$ are the products $g_{\alpha\beta}^E \cdot g_{\alpha\beta}^F$; taking the logarithmic derivative shows that $c_1(E \otimes F) = c_1(E) + c_1(F)$, so degree is additive under tensor product. Applying this inductively $m$ times gives $\deg L^m = m \deg L = md$. For the canonical bundle, recall that $K_C = (T^{1,0}C)^*$, so $\deg K_C = -\deg T^{1,0}C$; since $\deg T^{1,0}C$ equals the topological Euler characteristic $\chi_{\mathrm{top}}(C)$ by the [Gauss–Bonnet Theorem for Closed Oriented Surfaces](/theorems/3600), and $\chi_{\mathrm{top}}(C) = 2 - 2g$, we get $\deg K_C = -(2-2g) = 2g-2$. Using additivity of degree once more: \begin{align*} \deg(K_C \otimes L^m) = \deg K_C + \deg L^m = (2g-2) + md. \end{align*} *Vanishing of $H^1$.* To confirm $L^m$ is positive, work in a local holomorphic frame $e$ for $L$ with $|e|_h^2 = e^{-\varphi}$; positivity of $h$ means $i\partial\bar\partial\varphi > 0$. The frame $e^{\otimes m}$ trivializes $L^m$, and the induced metric $h^{\otimes m}$ satisfies $|e^{\otimes m}|_{h^{\otimes m}}^2 = |e|_h^{2m} = e^{-m\varphi}$, so its local weight is $m\varphi$. The curvature of a Hermitian line bundle in a local frame with weight $\psi$ is $\partial\bar\partial\psi$, giving $\Theta_{h^{\otimes m}}(L^m) = \partial\bar\partial(m\varphi) = m\,\partial\bar\partial\varphi$, hence \begin{align*} i\Theta_{h^{\otimes m}}(L^m) = m\,i\partial\bar\partial\varphi = m\,i\Theta_h(L) > 0. \end{align*} Thus $L^m$ is positive. Every compact Riemann surface admits a Kähler metric (any Hermitian metric on a one-dimensional complex manifold is Kähler), and $K_C \otimes L^m$ is the adjoint bundle of the positive line bundle $L^m$, so [Kodaira Vanishing Theorem (Statement)](/theorems/3501) gives $H^q(C, K_C \otimes L^m) = 0$ for every $q > 0$; in particular $h^1(C, K_C \otimes L^m) = 0$. *Riemann–Roch.* Applying [Riemann–Roch Theorem](/theorems/2185) to the line bundle $F = K_C \otimes L^m$ on $C$ gives \begin{align*} h^0(C, K_C \otimes L^m) - h^1(C, K_C \otimes L^m) = \deg(K_C \otimes L^m) + 1 - g. \end{align*} Substituting $h^1(C, K_C \otimes L^m) = 0$ on the left, and $(2g-2) + md$ for the degree on the right: \begin{align*} \dim H^0(C, K_C \otimes L^m) &= \bigl((2g-2) + md\bigr) + 1 - g \\ &= md + (2g - g) + (-2 + 1) \\ &= md + g - 1. \end{align*} The result $md + g - 1$ is an exact polynomial in $m$ for every $m \geq 1$, not merely an asymptotic approximation: since $h^1 = 0$ identically, the Euler characteristic collapses to $h^0$ with no cancellation between cohomology groups. The leading term $md$ matches the general growth formula from *Hirzebruch Riemann Roch for Line Bundles* at $n = 1$: the degree-$2$ term in $\int_C e^{mc_1(L)}\operatorname{td}(T^{1,0}C)$ is $\int_C mc_1(L) = m\deg L = md$, which equals $\frac{m^1}{1!}\int_C c_1(L)^1$ as the formula $\frac{m^n}{n!}\int_X c_1(L)^n$ predicts at $n=1$. The constant term $g-1$ records the Todd-class correction: the degree-$0$ and degree-$2$ parts of $\operatorname{td}(T^{1,0}C) = 1 + \tfrac{1}{2}c_1(T^{1,0}C)$ contribute $\tfrac{1}{2}\int_C c_1(T^{1,0}C) = \tfrac{1}{2}\chi_{\mathrm{top}}(C) = \tfrac{1}{2}(2-2g) = 1-g$ to the $m^0$ term, while the canonical factor $K_C$ shifts the constant part of the $m$-independent contribution by $\deg K_C = 2g-2$, and the two together give the total constant $(2g-2) + (1-g) = g-1$. This fixed topological quantity is invisible in the leading coefficient but fully determined by genus alone, independent of which degree-$d$ positive bundle $L$ is chosen. [/example] The section has three linked outcomes. Positivity gives vanishing for adjoint bundles, Serre duality turns this into information about negative bundles, and Riemann--Roch converts vanishing into asymptotic counts of sections. Sections 5 and 6 proved that positive curvature forces vanishing of cohomology, and Riemann-Roch uses this vanishing to count sections asymptotically. Section 7 assembles these tools into the Kodaira embedding theorem: a positive line bundle of sufficiently high tensor power has enough global sections to define a holomorphic immersion into projective space. # 7. Kodaira Embedding Theorem Kodaira embedding is the culmination of the course passage from curvature to global geometry. Sections 1-3 developed Hermitian metrics, Chern curvature, and positivity; Sections 5-6 supplied the Bochner-Kodaira-Nakano estimates and vanishing results used to solve the $\bar\partial$-equation with control. In this section those tools are used to manufacture enough holomorphic sections of high tensor powers $L^m$ to give projective coordinates on a compact complex manifold $X$. The central question is: when does a positive holomorphic line bundle force $X$ to be projective? The answer is that positivity supplies sections separating both points and first-order tangent directions, and these sections assemble into an embedding into projective space. The final result identifies compact complex manifolds carrying positive line bundles with smooth projective algebraic manifolds. ## Separating Points and Tangent Vectors by Holomorphic Sections When can global holomorphic sections distinguish the points of $X$ and also detect infinitesimal directions? This is the local-to-global issue behind projective embedding: homogeneous coordinates separate points, while their first derivatives separate tangent vectors. [definition: Base Point Free Linear System] Let $X$ be a compact complex manifold, let $L \to X$ be a holomorphic line bundle, and let $V \subset H^0(X,L)$ be a finite-dimensional complex vector subspace. The space $V$ is base point free if for every $x \in X$ there exists $s \in V$ such that $s(x) \ne 0$. [/definition] Base point freeness is the minimum requirement for projective coordinates: at each point at least one homogeneous coordinate must be nonzero. To upgrade a well-defined map to projective space into an embedding, two further failures must be excluded. Distinct points must not receive the same homogeneous coordinates, and the differential of the map must not collapse a nonzero tangent direction. The next definition states these two separation requirements directly in terms of sections of the line bundle. [definition: Separation of Points and Tangent Vectors] Let $X$ be a compact complex manifold, let $L \to X$ be a holomorphic line bundle, and let $V \subset H^0(X,L)$ be finite-dimensional. The space $V$ separates points if for every pair $x,y \in X$ with $x \ne y$ there exists $s \in V$ such that $s(x)=0$ and $s(y) \ne 0$. The space $V$ separates tangent vectors if for every $x \in X$ and every nonzero $v \in T_x^{1,0}X$ there exists $s \in V$ such that $s(x)=0$ and, writing $s=f_s e$ in any local frame $e$ near $x$, $d(f_s)_x(v) \ne 0$. [/definition] The derivative condition is independent of the chosen local frame. If $s(x)=0$, multiplying the local coefficient $f_s$ by a nowhere-vanishing holomorphic function does not change whether its differential is nonzero on $v$. [example: Local Meaning of Tangent Separation] Let $x \in X$, let $z_1,\dots,z_n$ be local holomorphic coordinates centered at $x$, and let $e$ be a local frame of $L$ near $x$. Write $s = fe$ for a holomorphic function $f$ with $f(x) = 0$, and let $v = \sum_i v_i \partial_{z_i} \in T_x^{1,0}X$ be nonzero. We show that the condition $\sum_i v_i \partial_{z_i} f(x) \ne 0$ is independent of which local frame is used to express $s$. Let $\tilde{e} = ge$ be any other local frame near $x$, where $g$ is holomorphic and nowhere zero. In the frame $\tilde{e}$, the section reads $s = \tilde{f}\,\tilde{e}$ with $\tilde{f} = f/g$, so that $\tilde{f}(x) = f(x)/g(x) = 0$. To compare the two differentials, write $f = \tilde{f} \cdot g$ and apply the [product rule](/theorems/325) coordinate by coordinate: \begin{align*} \partial_{z_i} f = (\partial_{z_i} \tilde{f})\, g + \tilde{f}\, \partial_{z_i} g. \end{align*} Evaluating at $x$ and using $\tilde{f}(x) = 0$, the second term vanishes: \begin{align*} \partial_{z_i} f(x) = (\partial_{z_i} \tilde{f})(x)\cdot g(x). \end{align*} Multiplying by $v_i$ and summing over $i$ gives \begin{align*} \sum_i v_i \partial_{z_i} f(x) = g(x)\sum_i v_i(\partial_{z_i}\tilde{f})(x). \end{align*} Since $g(x) \ne 0$, the left side is nonzero if and only if $\sum_i v_i(\partial_{z_i}\tilde{f})(x) \ne 0$. The two frames therefore detect the same tangent vectors, so tangent separation at $x$ depends only on the first jet of $s$ at $x$, not on how the local trivialization is chosen. [/example] The next criterion is the bridge between section theory and projective geometry. It says that base point freeness, point separation, and tangent separation are exactly the geometric ingredients needed for an embedding. [quotetheorem:3868] [citeproof:3868] This criterion has real content in all three hypotheses. Compactness is what turns the injective holomorphic immersion into an embedding; without a properness condition, an injective immersion can have non-embedded image, as the dense irrational line in a real torus illustrates. Base point freeness is equally structural: if all sections vanish at some point, the formula defines only a rational map with an indeterminacy locus rather than a morphism. Because the result is an iff, failure of point separation or tangent separation is not a technical defect but an actual obstruction to projective embedding by that linear system. The analytic input for Kodaira embedding is that positivity of $L$ produces these separating sections after replacing $L$ by a high tensor power. The word high is essential: a positive line bundle may not itself have enough global sections, but its powers become increasingly flexible. To apply the embedding criterion, this flexibility must be stated as a precise first-order separation property at every point. The missing input is a uniform high-power generation statement: after a single tensor exponent threshold, sections should both avoid vanishing at any chosen point and realize prescribed first-order directions there. The analytic input is often called first-jet generation for high powers of a positive line bundle. It says that, for all sufficiently large $m$, global sections of $L^m$ can be chosen to prescribe the value and first derivative data needed by the embedding criterion. This is the analytic heart of Kodaira embedding: positivity gives curvature, curvature gives $L^2$ estimates, and the estimates turn local model sections into global holomorphic sections. Here ample means positive in the algebraic sense: some tensor power gives a projective embedding. A line bundle is nef when it lies in the closure of the ample cone, or equivalently in the projective setting when it has nonnegative degree on every irreducible curve. The base locus of a linear system is the common zero set of all its sections. Each step of this chain relies on a strict inequality, and weakening the hypothesis breaks the mechanism in a specific way. If $L$ is only semipositive — meaning $\omega_h \ge 0$ rather than $\omega_h > 0$ — then the curvature may lose control in directions where it degenerates. For nef line bundles, base point freeness can already fail for every power. A standard counterexample is the line bundle associated to an irrational point of a real elliptic curve, which is nef but has no global sections at all, so first-jet generation cannot hold. The statement is qualitative rather than effective. It guarantees that some high tensor power separates first jets, but it does not give a usable numerical bound for the first such power. Effective replacements — bounding the required tensor power in terms of intersection numbers, Chern class data, or the volume of $L$ — are the subject of much deeper work, including Matsusaka's big theorem, Siu's effective version, and Demailly's analytic numerical bounds. ## The Kodaira Map from a Basis of $H^0(X,L^m)$ Once enough sections exist, how are they assembled into a projective map? The construction is the complex-geometric analogue of using functions as coordinates, except that sections of a line bundle produce homogeneous rather than affine coordinates. [definition: Kodaira Map] Let $X$ be a compact complex manifold, let $L \to X$ be a holomorphic line bundle, and assume that $H^0(X,L^m)$ is base point free. Let $s_0,\dots,s_N$ be a basis of $H^0(X,L^m)$. For $x\in X$ and a local frame $e$ of $L$ near $x$, write $s_j=f_j e^{\otimes m}$. The Kodaira map associated with this basis is $\Phi_m:X\to\mathbb{P}^N$ given by $\Phi_m(x)=[f_0(x):\cdots:f_N(x)]$. [/definition] Changing the local frame multiplies all $f_j$ by the same nonzero holomorphic function, so the projective point is unchanged. Changing the basis of $H^0(X,L^m)$ composes $\Phi_m$ with a projective linear automorphism of $\mathbb{P}^N$. Once the map has been defined, the next question is whether it remembers the line bundle that produced it. The natural comparison is between $L^m$ on $X$ and the hyperplane bundle on projective space pulled back along $\Phi_m$. [quotetheorem:3870] [citeproof:3870] This identifies the original line bundle power as the hyperplane bundle seen from the embedded image. The theorem explains why the Kodaira map is not merely a map into projective space: it records the line bundle used to construct it. In applications, this compatibility is what lets projective geometry on the target translate back into line-bundle geometry on $X$. [illustration:scv-iv-kodaira-hyperplane-section] [example: Veronese Embeddings] Take $X = \mathbb{P}^n$ and $L = \mathcal{O}_{\mathbb{P}^n}(1)$. The space $H^0(\mathbb{P}^n, \mathcal{O}_{\mathbb{P}^n}(m))$ is the space of homogeneous polynomials of degree $m$ in $Z_0, \ldots, Z_n$, with basis the monomial set $\{Z^\alpha\}_{|\alpha|=m}$, where $\alpha = (\alpha_0, \ldots, \alpha_n) \in \mathbb{Z}_{\ge 0}^{n+1}$ and $|\alpha| = \alpha_0 + \cdots + \alpha_n = m$. The number of such monomials is the number of ways to choose a multiindex $\alpha$ of weight $m$ from $n+1$ variables; by the stars-and-bars formula this equals $\binom{n+m}{n} = \binom{n+m}{m}$, so the target projective space has dimension $\binom{n+m}{m} - 1$. [claim]The map $\nu_m \colon \mathbb{P}^n \to \mathbb{P}^{\binom{n+m}{m}-1}$ defined by $[Z_0:\cdots:Z_n] \mapsto [Z^\alpha]_{|\alpha|=m}$ is a well-defined holomorphic embedding, and $\nu_m^*\mathcal{O}(1) \cong \mathcal{O}_{\mathbb{P}^n}(m)$.[/claim] *Well-definedness.* Replacing $[Z_0:\cdots:Z_n]$ by $[\lambda Z_0:\cdots:\lambda Z_n]$ sends every monomial $Z^\alpha$ to $(\lambda Z)^\alpha = \lambda^{|\alpha|}Z^\alpha = \lambda^m Z^\alpha$. Since $\lambda^m \ne 0$, all homogeneous coordinates are scaled by the same nonzero factor, so the projective point is unchanged. *Point separation.* Suppose $\nu_m([P]) = \nu_m([Q])$; then there exists $\lambda \ne 0$ with $P^\alpha = \lambda Q^\alpha$ for every $|\alpha|=m$. Fix any index $k$ with $P_k \ne 0$. Taking $\alpha = me_k$ gives $P_k^m = \lambda Q_k^m$, so $Q_k \ne 0$. For any other index $j$, taking $\alpha = (m-1)e_k + e_j$ gives \begin{align*} P_k^{m-1}P_j = \lambda Q_k^{m-1}Q_j. \end{align*} Dividing by the equation $P_k^m = \lambda Q_k^m$ (which holds and has $P_k, Q_k \ne 0$): \begin{align*} \frac{P_j}{P_k} = \frac{Q_j}{Q_k}. \end{align*} Since $j$ is arbitrary, $[P_0:\cdots:P_n] = [Q_0:\cdots:Q_n]$ in $\mathbb{P}^n$, so $\nu_m$ is injective. *Tangent separation.* On the affine chart $Z_0 \ne 0$, set $w_i = Z_i/Z_0$ for $i = 1,\ldots,n$. Dividing each monomial coordinate by $Z_0^m$, the map on this chart reads \begin{align*} (w_1,\ldots,w_n) \longmapsto \bigl[1 : w_1 : \cdots : w_n : w_1^2 : w_1w_2 : \cdots \bigr], \end{align*} where the $n$ entries $w_1,\ldots,w_n$ arise from the monomials $Z_0^{m-1}Z_i/Z_0^m$ for $i=1,\ldots,n$. Their partial derivatives satisfy \begin{align*} \frac{\partial w_i}{\partial w_k} = \delta_{ik}, \end{align*} so the $n\times n$ submatrix of the full Jacobian formed by the rows indexed by these linear coordinates is the identity matrix $I_n$. Hence $d(\nu_m)_w$ has rank $n$ at every $w$, and $\nu_m$ is an immersion on this chart; the same identification on each chart $Z_j \ne 0$ yields the immersion property globally. Since $\mathbb{P}^n$ is compact and $\mathbb{P}^{\binom{n+m}{m}-1}$ is Hausdorff, the [Compact-to-Hausdorff Embedding Criterion](/theorems/1038) promotes the injective holomorphic immersion $\nu_m$ to a homeomorphism onto its image; the immersion charts then make this image an embedded complex submanifold. By the *Projective Embedding Criterion*, $\nu_m$ is a holomorphic embedding. *Pullback.* The transition function of $\mathcal{O}(1)$ on $\mathbb{P}^N$ between the charts $\{Y_\alpha \ne 0\}$ and $\{Y_\beta \ne 0\}$ is $Y_\beta/Y_\alpha$. Pulling back via $\nu_m$: \begin{align*} \nu_m^*\!\left(\frac{Y_\beta}{Y_\alpha}\right) = \frac{Z^\beta}{Z^\alpha}. \end{align*} Taking $\alpha = me_0$ and $\beta = me_i$, so $Z^\alpha = Z_0^m$ and $Z^\beta = Z_i^m$: \begin{align*} \nu_m^*\!\left(\frac{Y_{me_i}}{Y_{me_0}}\right) = \frac{Z_i^m}{Z_0^m} = \left(\frac{Z_i}{Z_0}\right)^{\!m}. \end{align*} This is exactly the transition function of $\mathcal{O}_{\mathbb{P}^n}(m)$ on the overlap $\{Z_0 \ne 0\}\cap\{Z_i \ne 0\}$. The same computation on every pair of coordinate charts confirms that the transition cocycles agree, so $\nu_m^*\mathcal{O}(1) \cong \mathcal{O}_{\mathbb{P}^n}(m)$. For $n=1$ and $m=2$, the three monomials are $Z_0^2$, $Z_0Z_1$, $Z_1^2$, and the map is $\nu_2([Z_0:Z_1]) = [Z_0^2:Z_0Z_1:Z_1^2]$ in $\mathbb{P}^2$. The image satisfies the equation $Y_0Y_2 = Y_1^2$, since $Z_0^2 \cdot Z_1^2 = (Z_0Z_1)^2$; the right-hand side is exactly $Y_1^2$, confirming that every point of the image lies on this conic. On the affine chart $Z_0 = 1$, the map becomes $w \mapsto (w,\,w^2)$ in coordinates $(Y_1/Y_0,\, Y_2/Y_0)$, with differential $(1,\,2w)^T$, which is nonzero for every $w \in \mathbb{C}$, verifying immersion on this chart. Injectivity on the same chart is immediate: if $[1:w:w^2] = [1:w':w'^2]$, comparing the second homogeneous coordinate gives $w = w'$. Thus $\mathbb{P}^1$ embeds as the smooth conic $\{Y_0Y_2 = Y_1^2\} \subset \mathbb{P}^2$, the model instance of all Veronese embeddings. [/example] [illustration:scv-iv-veronese-conic] The Veronese example is the model case: high powers of a positive line bundle give many polynomial coordinates. The Kodaira theorem says that every compact complex manifold with a positive line bundle eventually behaves like this model after embedding. ## Very Ampleness from Positivity Why should a curvature condition imply the existence of enough projective coordinates? The answer is that positive curvature gives analytic estimates strong enough to solve the global extension and correction problems needed for section separation. [definition: Very Ample Line Bundle] Let $X$ be a compact complex manifold. A holomorphic line bundle $A\to X$ is very ample if $H^0(X,A)$ is base point free and the Kodaira map associated with a basis of $H^0(X,A)$ is a holomorphic embedding. [/definition] Very ampleness is therefore the intrinsic line-bundle formulation of projective embeddability. The previous section translates it into separation of points and tangent vectors. The failure mode is useful to keep in view. A Hopf surface or a generic complex two-torus has no positive line bundle, so there is no curvature source from which these global separating sections can be produced; correspondingly such manifolds are not projective. Kodaira embedding proves that this is the only analytic obstruction of this kind: once a positive line bundle exists, sufficiently high powers supply the missing coordinates. [quotetheorem:3871] [citeproof:3871] The theorem is often stated as positive implies ample, with the stronger conclusion that sufficiently high powers are very ample. In this analytic course, the main mechanism is not formal algebraic geometry but the construction of peak sections by $\bar\partial$ estimates. The positivity assumption cannot be removed: Hopf surfaces and generic complex tori give compact complex manifolds with no positive line bundle, hence no Kodaira embedding of this type. The statement is also qualitative: compactness and estimates supply an integer $m_0$, but the theorem does not give a general effective bound; effective versions belong to deeper results such as Matsusaka-Mumford and Demailly-type estimates. [example: Compact Riemann Surfaces by High Degree Line Bundles] Let $C$ be a compact Riemann surface of genus $g$, and let $A \to C$ be a holomorphic line bundle of degree $d$. The [Degree of the Canonical Divisor](/theorems/2186) gives $\deg(K_C) = 2g - 2$. [claim]If $d \geq 2g + 1$, then $A$ is very ample. Consequently, for any positive holomorphic line bundle $L \to C$, the power $L^m$ is very ample for every $m \geq \lceil(2g+1)/\deg(L)\rceil$.[/claim] We verify base point freeness, point separation, and tangent separation. Each reduces to the vanishing of a single $H^1$ group, established by *Serre Duality* and a degree count. *Base point freeness.* Fix $p \in C$. Tensoring the ideal-sheaf sequence $0 \to \mathcal{O}_C(-p) \to \mathcal{O}_C \to \mathcal{O}_p \to 0$ with $A$ gives the short exact sequence of sheaves \begin{align*} 0 \to A(-p) \to A \to A\vert_p \to 0, \end{align*} where $A\vert_p$ is the skyscraper sheaf with stalk $A_p \cong \mathbb{C}$. The [Long Exact Cohomology Sequence](/theorems/3471) extracts \begin{align*} H^0(C, A) \xrightarrow{\;\mathrm{ev}_p\;} A_p \longrightarrow H^1(C, A(-p)), \end{align*} so $\mathrm{ev}_p$ is surjective — guaranteeing a section nonvanishing at $p$ — as soon as $H^1(C, A(-p)) = 0$. By *Serre Duality*, $H^1(C, A(-p)) \cong H^0(C,\, K_C \otimes A(-p)^*)^\vee$, and \begin{align*} \deg\!\bigl(K_C \otimes A(-p)^*\bigr) = (2g-2) - (d-1) = 2g - d - 1 \leq -2, \end{align*} using $d \geq 2g + 1$. A nonzero holomorphic section of a line bundle defines an effective divisor whose degree equals the bundle's degree; since no effective divisor has negative degree, the bundle $K_C \otimes A(-p)^*$ has no nonzero global sections, so $H^1(C, A(-p)) = 0$ and $\mathrm{ev}_p$ is surjective. *Point separation.* Fix distinct $p, q \in C$. Tensoring $0 \to \mathcal{O}_C(-p-q) \to \mathcal{O}_C \to \mathcal{O}_{\{p,q\}} \to 0$ with $A$ gives \begin{align*} 0 \to A(-p-q) \to A \to A\vert_{\{p,q\}} \to 0, \end{align*} where $H^0(C,\, A\vert_{\{p,q\}}) \cong A_p \oplus A_q$. The [Long Exact Cohomology Sequence](/theorems/3471) shows the joint evaluation map $H^0(C, A) \to A_p \oplus A_q$ is surjective if and only if $H^1(C, A(-p-q)) = 0$. By *Serre Duality*, \begin{align*} \deg\!\bigl(K_C \otimes A(-p-q)^*\bigr) = (2g-2) - (d-2) = 2g - d \leq -1. \end{align*} The same negative-degree argument gives $H^1(C, A(-p-q)) = 0$, so the evaluation onto $A_p \oplus A_q$ is surjective. Surjectivity means we can prescribe the fiber values at $p$ and $q$ independently: there exists $s \in H^0(C,A)$ with $s(p) \neq 0$ and $s(q) = 0$, and another with the roles reversed, which is exactly point separation. *Tangent separation.* Fix $p \in C$. The first-jet sequence at $p$ is \begin{align*} 0 \to A(-2p) \to A \to A/A(-2p) \to 0. \end{align*} In a local coordinate $z$ centered at $p$ and a local frame $e$ of $A$, a section $s = fe$ with Taylor expansion $f = a_0 + a_1 z + O(z^2)$ maps to the pair $(a_0, a_1) \in \mathbb{C}^2$; the subsheaf $A(-2p)$ is cut out by $a_0 = a_1 = 0$, so $H^0(C,\, A/A(-2p)) \cong \mathbb{C}^2$ records the zeroth and first Taylor coefficients. The [Long Exact Cohomology Sequence](/theorems/3471) shows the first-jet map $\mathrm{jet}^1_p \colon H^0(C,A) \to \mathbb{C}^2$ is surjective if and only if $H^1(C, A(-2p)) = 0$. Because $A(-2p)$ twists $A$ by a divisor of degree $-2$, the same as $A(-p-q)$, \begin{align*} \deg\!\bigl(K_C \otimes A(-2p)^*\bigr) = (2g-2) - (d-2) = 2g - d \leq -1, \end{align*} so $H^1(C, A(-2p)) = 0$ by the same reasoning, and $\mathrm{jet}^1_p$ is surjective. We can therefore find $s$ with $a_0 = 0$ and $a_1 \neq 0$: this section vanishes at $p$ and has nonzero derivative $f'(0) = a_1$ there, so $df_s(p) \cdot \partial_z \neq 0$. Since $C$ has complex dimension one and $\partial_z$ spans $T_p^{1,0}C$, this is tangent separation at $p$. *Conclusion.* Since $H^0(C,A)$ is base point free, separates points, and separates tangent vectors, the [Projective Embedding Criterion](/theorems/2195) yields a holomorphic embedding $C \hookrightarrow \mathbb{P}(H^0(C,A)^*)$, so $A$ is very ample. For a positive line bundle $L$ with $\deg(L) \geq 1$, we have $\deg(L^m) = m\,\deg(L) \geq 2g+1$ for every $m \geq \lceil(2g+1)/\deg(L)\rceil$, so $L^m$ is very ample for all sufficiently large $m$. For curves the threshold $m_0 = \lceil(2g+1)/\deg(L)\rceil$ is an explicit integer computable from the genus and the degree of $L$, in contrast to the general Kodaira theorem where $m_0$ is extracted from a compactness argument with no stated formula. [/example] For curves, Kodaira embedding can therefore be seen through both curvature and divisor theory. Positive curvature corresponds to positive degree, while high degree gives enough meromorphic functions with controlled poles to separate points and tangents. [example: Abelian Varieties and Theta Line Bundles] Let $\Lambda \subset \mathbb{C}^g$ be a full-rank lattice and $A = \mathbb{C}^g/\Lambda$ the associated complex torus. By the *Appell–Humbert Theorem*, every holomorphic line bundle on $A$ is isomorphic to some $\mathcal{L}(H,\chi)$, where $H$ is a Hermitian form on $\mathbb{C}^g$ whose imaginary part $E := \mathrm{Im}\,H$ restricts to a $\mathbb{Z}$-valued alternating form on $\Lambda$, and $\chi\colon\Lambda\to U(1)$ is a semicharacter satisfying $\chi(\lambda+\mu) = \chi(\lambda)\chi(\mu)(-1)^{E(\lambda,\mu)}$. The bundle $\Theta = \mathcal{L}(H,\chi)$ is a positive theta line bundle when $H$ is positive definite. [claim] Let $\Theta = \mathcal{L}(H,\chi)$ be a positive theta line bundle on $A = \mathbb{C}^g/\Lambda$. Then $A$ embeds as a projective algebraic manifold. The global sections in $H^0(A,\Theta^m)$ are precisely the holomorphic functions $\theta\colon\mathbb{C}^g\to\mathbb{C}$ satisfying the level-$m$ transformation law \begin{align*} \theta(z+\lambda) = \chi(\lambda)^m\exp\!\bigl(\pi m H(z,\lambda)+\tfrac{\pi m}{2}H(\lambda,\lambda)\bigr)\,\theta(z) \quad \text{for all } \lambda\in\Lambda. \end{align*} In the principally polarised case $\mathrm{Pf}(E|_\Lambda)=1$, one has $h^0(A,\Theta^m) = m^g$, the Kodaira map lands in $\mathbb{P}^{m^g-1}$, and the *Theorem of Lefschetz* gives very ampleness of $\Theta^3$. A generic complex torus of dimension $g\geq 2$ admits no positive line bundle and is not projective. [/claim] *Sections are theta functions.* The bundle $\mathcal{L}(H,\chi)$ is the quotient of $\mathbb{C}^g\times\mathbb{C}$ by the $\Lambda$-action $\lambda\cdot(z,w) = (z+\lambda,\,e_\lambda(z)\,w)$, where the automorphy factor is \begin{align*} e_\lambda(z) = \chi(\lambda)\exp\!\bigl(\pi H(z,\lambda)+\tfrac{\pi}{2}H(\lambda,\lambda)\bigr). \end{align*} A global section $s\in H^0(A,\Theta^m)$ pulls back via $\pi\colon\mathbb{C}^g\to A$ to a holomorphic function $\theta\colon\mathbb{C}^g\to\mathbb{C}$ equivariant for the $m$-th tensor power: $\theta(z+\lambda) = e_\lambda^{(m)}(z)\,\theta(z)$, where $e_\lambda^{(m)}$ is the automorphy factor of $\Theta^m = \mathcal{L}(mH,\chi^m)$. Substituting $(mH,\chi^m)$ for $(H,\chi)$ in the Appell–Humbert formula: \begin{align*} e_\lambda^{(m)}(z) = \chi(\lambda)^m\exp\!\bigl(\pi(mH)(z,\lambda)+\tfrac{\pi}{2}(mH)(\lambda,\lambda)\bigr) = \chi(\lambda)^m\exp\!\bigl(\pi m H(z,\lambda)+\tfrac{\pi m}{2}H(\lambda,\lambda)\bigr), \end{align*} which is the stated transformation law. *Positive curvature.* Equip $\Theta$ with the Hermitian metric $h$ on $\mathbb{C}^g$ defined by $\|1\|_z^2 = e^{-\pi H(z,z)}$, where $1$ denotes the constant section of $\mathbb{C}^g\times\mathbb{C}$; this descends to $A$ because the $\Lambda$-equivariance changes $H(z,z)$ by real terms absorbed into $|e_\lambda(z)|^2$. The local curvature weight is $\varphi(z) = \pi H(z,z)$. Writing $H(z,z) = \sum_{i,j}H_{i\bar\jmath}z_i\bar z_j$ with constant matrix entries $H_{i\bar\jmath}$, differentiating: \begin{align*} \bar\partial H(z,z) &= \sum_{i,j}H_{i\bar\jmath}\,z_i\,d\bar z_j, \\ \partial\bar\partial H(z,z) &= \sum_{i,j}H_{i\bar\jmath}\,dz_i\wedge d\bar z_j, \end{align*} since $H_{i\bar\jmath}$ is constant so $\partial$ acts only on the $z_i$ factor. The Chern curvature form is therefore \begin{align*} \Theta_h = i\partial\bar\partial\varphi = i\pi\sum_{i,j}H_{i\bar\jmath}\,dz_i\wedge d\bar z_j. \end{align*} Positive definiteness of the matrix $(H_{i\bar\jmath})$ makes $\Theta_h$ a positive $(1,1)$-form, so $\Theta$ is a positive line bundle. *Applying Kodaira embedding.* Since $\Theta$ is positive, the theorem *High Powers Separate First Jets* yields $m_0\in\mathbb{N}$ such that $H^0(A,\Theta^m)$ is base point free and separates points and tangent vectors for every $m\geq m_0$. The [Projective Embedding Criterion](/theorems/2195) then produces a holomorphic embedding $A\hookrightarrow\mathbb{P}(H^0(A,\Theta^m)^*)$; compactness of $A$ and the Hausdorff property of projective space promote the injective holomorphic immersion to an embedding onto a closed complex submanifold. *Dimension formula.* For $m\geq 1$, positive definiteness of $mH$ makes $\Theta^m$ ample, and the *Vanishing Theorem for Abelian Varieties* gives $H^i(A,\Theta^m) = 0$ for all $i > 0$. The Riemann–Roch theorem for abelian varieties gives $\chi(A,\Theta^m) = \mathrm{Pf}(mE|_\Lambda)$. Since the Pfaffian is a degree-$g$ polynomial in the matrix entries of the alternating form, scaling $E$ by $m$ scales the Pfaffian by $m^g$: \begin{align*} \mathrm{Pf}(mE|_\Lambda) = m^g\,\mathrm{Pf}(E|_\Lambda). \end{align*} In the principally polarised case $\mathrm{Pf}(E|_\Lambda) = 1$, so $h^0(A,\Theta^m) = \chi(A,\Theta^m) = m^g$ and the Kodaira map lands in $\mathbb{P}^{m^g - 1}$. *Lefschetz very ampleness.* The *Theorem of Lefschetz* states that $L^m$ is very ample for every $m\geq 3$ whenever $L$ is an ample line bundle on an abelian variety. Applied to $\Theta$, this gives very ampleness of $\Theta^3$: the $3^g$ theta functions of level 3 separate all points and first-order jets, and the Kodaira map embeds $A$ in $\mathbb{P}^{3^g-1}$. *The case $g=1$.* For an elliptic curve $E=\mathbb{C}/(\mathbb{Z}+\mathbb{Z}\tau)$ with $\mathrm{Im}(\tau)>0$, the principal polarisation is $\Theta = \mathcal{O}([0])$, the line bundle of degree 1 with divisor the origin, and $h^0(E,\Theta^3) = 3^1 = 3$. Sections of $\Theta^3 = \mathcal{O}(3[0])$ correspond to meromorphic functions on $E$ with poles only at $[0]$ of order at most 3; a basis is $\{1,\,\wp,\,\wp'\}$, where the Weierstrass function $\wp(z) = z^{-2} + \sum_{(m,n)\ne(0,0)}\bigl((z-m-n\tau)^{-2}-(m+n\tau)^{-2}\bigr)$ has a double pole at $[0]$ and $\wp'(z) = -2z^{-3}+\cdots$ has a triple pole, giving three linearly independent sections spanning the full 3-dimensional space. The Kodaira map is \begin{align*} [z] \longmapsto [\wp(z):\wp'(z):1]\in\mathbb{P}^2 \end{align*} (with the convention that $[0]\in E$ maps to $[0:1:0]$, obtained by multiplying through by $z^3$ and taking $z\to 0$). The image is the Weierstrass cubic: since $(\wp')^2 = 4\wp^3 - g_2\wp - g_3$ with $g_2 = 60\sum_\Lambda(m+n\tau)^{-4}$ and $g_3 = 140\sum_\Lambda(m+n\tau)^{-6}$, setting $[X:Y:Z] = [\wp:\wp':1]$ yields the projective equation \begin{align*} Y^2 Z = 4X^3 - g_2 XZ^2 - g_3 Z^3. \end{align*} Smoothness holds because the discriminant $\Delta = g_2^3 - 27g_3^2$ is nonzero for $\tau$ in the upper half-plane, confirming that $E$ embeds as a smooth cubic in $\mathbb{P}^2$. *Boundary: complex tori without positive line bundles.* In dimension $g\geq 2$, many complex tori are not abelian varieties. The obstruction is the absence of a Riemann form: there may be no integral alternating form on the lattice whose associated Hermitian pairing is positive definite. Such a torus is still compact and Kähler, but it has no positive line bundle, so Kodaira embedding has no curvature source from which to produce projective coordinates. For $g=1$ the Lefschetz bound $m_0=3$ is sharp: $h^0(\Theta)=1$ furnishes only one section with no separating power beyond base point freeness, and $h^0(\Theta^2)=4$ gives an embedding only after factoring by the hyperelliptic involution (a $2$-to-$1$ map onto a conic), while $h^0(\Theta^3)=3$ is the first power that embeds. In higher dimension, the boundary observation is structural: Kähler tori need not carry any positive line bundle, so projectivity is an extra arithmetic condition on the lattice rather than a consequence of compactness or flat Kähler geometry alone. [/example] Theta functions show the concrete face of the theorem. The abstract sections constructed by $L^2$ estimates become explicit analytic functions satisfying transformation laws under the lattice. ## Projective Algebraic Manifolds and Positive Line Bundles What class of compact complex manifolds has been characterized? Kodaira embedding places $X$ inside projective space as a compact complex submanifold, and the theorem of Chow converts this analytic submanifold into an algebraic one. [quotetheorem:3872] [citeproof:3872] This final equivalence packages the section's analytic work into a classification statement. Positivity first builds an embedding, the hyperplane bundle on projective space then records the original curvature class, and Chow's theorem changes the embedded analytic image into algebraic equations. [remark: Analytic and Algebraic Positivity] The characterisation does not say that every compact complex manifold is projective. It says that the missing structure is exactly a holomorphic line bundle with positive curvature. Kodaira embedding is therefore a bridge: differential-geometric positivity becomes projective coordinates, and projective coordinates become algebraic equations by the theorem of Chow. [/remark] Kodaira embedding realizes a compact complex manifold with a positive line bundle as a smooth projective submanifold. Section 8 asks a finer question: given a projective embedding, how much of the manifold's topology is already determined by its hyperplane sections, independent of the specific positive bundle? # 8. Lefschetz Hyperplane Theorems After Section 7 embeds positive manifolds into projective space, this section explains why projective manifolds are strongly controlled by their hyperplane sections. The central question is how much topology and cohomology of a smooth projective variety $X \subset \mathbb P^N$ can be recovered from a smooth divisor $Y = X \cap H$. The answer has two layers: Bertini supplies smooth sections, weak Lefschetz compares $X$ and $Y$ in low degrees, and hard Lefschetz explains why cup product with the hyperplane class gives the cohomological symmetry behind the picture. ## Smooth Hyperplane Sections The first problem is geometric: given a smooth projective manifold embedded in projective space, can we cut it by a hyperplane without introducing singularities? This is not automatic for a single chosen hyperplane, because the hyperplane may be tangent to $X$ or may contain a singular point if the ambient variety is singular. The Bertini theorem says that the bad hyperplanes form a proper algebraic subset of the dual projective space. The parameter space for hyperplanes in $\mathbb P^N$ is the dual projective space $(\mathbb P^N)^*$. A point $[\ell] \in (\mathbb P^N)^*$ represents the hyperplane $H_\ell = \{\ell = 0\}$. [definition: Hyperplane Section] Let $X \subset \mathbb P^N$ be a projective variety and let $H \subset \mathbb P^N$ be a hyperplane. The hyperplane section of $X$ by $H$ is the projective variety \begin{align*} X \cap H. \end{align*} If $X$ is smooth and $X \cap H$ is smooth of codimension $1$ in $X$, the section is called a smooth hyperplane section. [/definition] When $X \not\subset H$, the section $X \cap H$ is the zero set on $X$ of the section of $\mathcal O_X(1)$ induced by $\ell$. Smoothness of $X \cap H$ means that this section vanishes transversely along its zero locus. The useful question is therefore not whether every hyperplane section is smooth, but whether smooth choices are abundant enough to use without special construction. Bertini's theorem gives exactly this generic smoothness statement in the parameter space of hyperplanes. [quotetheorem:3873] [citeproof:3873] [example: Smooth Hypersurfaces In Projective Space] Let $F \in \mathbb{C}[Z_0,\dots,Z_n]$ be homogeneous of degree $d$ with zero locus $X = \{F = 0\} \subset \mathbb{P}^n$. We show that $X$ is smooth if and only if the partial derivatives $\partial F/\partial Z_0, \dots, \partial F/\partial Z_n$ have no common zero on $X$. Work on the affine chart $U_i = \{Z_i \neq 0\}$ with coordinates $z_j = Z_j/Z_i$ for $j \neq i$. Define \begin{align*} f(z_0,\dots,\hat{z}_i,\dots,z_n) = F(z_0,\dots,1,\dots,z_n), \end{align*} so that $X \cap U_i = \{f = 0\} \subset \mathbb{C}^n$. By the [Implicit Function Theorem](/theorems/52), the hypersurface $X \cap U_i$ is smooth at a point $p$ if and only if at least one partial derivative $\partial f/\partial z_j$ ($j \neq i$) is nonzero at $p$. On $U_i$ we have $Z_j = z_j$ for $j \neq i$, so differentiating $f = F|_{Z_i = 1}$ directly gives \begin{align*} \frac{\partial f}{\partial z_j} = \frac{\partial F}{\partial Z_j}\bigg|_{Z_i = 1} \qquad (j \neq i). \end{align*} Hence $p$ is a singular point of $X \cap U_i$ if and only if $\partial F/\partial Z_j = 0$ at $p$ for every $j \neq i$. It remains to check whether $\partial F/\partial Z_i$ automatically vanishes there as well. The *Euler identity for homogeneous polynomials* of degree $d$ asserts \begin{align*} \sum_{j=0}^{n} Z_j \,\frac{\partial F}{\partial Z_j} = d\,F. \end{align*} At any point of $X$ the right side equals zero, so the identity on $X$ reads $Z_i\,(\partial F/\partial Z_i) = -\sum_{j \neq i} Z_j\,(\partial F/\partial Z_j)$. If $\partial F/\partial Z_j = 0$ for all $j \neq i$, then \begin{align*} \frac{\partial F}{\partial Z_i} = -\frac{1}{Z_i}\sum_{j \neq i} Z_j\,\frac{\partial F}{\partial Z_j} = 0, \end{align*} where division by $Z_i$ is valid because $Z_i \neq 0$ on $U_i$. Thus all $n+1$ partials vanish simultaneously at $p$ on $X \cap U_i$, and not merely the $n$ visible to the implicit function theorem on that chart. Since $\mathbb{P}^n = U_0 \cup \cdots \cup U_n$, the same argument applies on each chart, and $X$ is smooth if and only if $\partial F/\partial Z_0, \dots, \partial F/\partial Z_n$ share no common zero on $X$. The Euler identity is what closes the argument: on each affine chart the [Implicit Function Theorem](/theorems/52) detects singularities via $n$ affine partials, and Euler's identity uses $F|_X = 0$ to force the remaining homogeneous partial to vanish as well, making the criterion coordinate-independent. [/example] For a smooth hypersurface $X \subset \mathbb P^n$, Bertini says that a general hyperplane $H$ cuts out a smooth hypersurface $X \cap H \subset H \cong \mathbb P^{n-1}$ of the same degree. Thus the operation of taking a hyperplane section preserves the class of smooth hypersurfaces while lowering the dimension by one. [remark: Base Point Free Linear Systems] Bertini applies more generally to base point free linear systems. If $L \to X$ is a holomorphic line bundle generated by global sections, then a generic section $s \in H^0(X,L)$ has smooth zero divisor. The projective embedding case is obtained from the base point free linear system $H^0(X,\mathcal O_X(1))$. [/remark] ## Weak Lefschetz for Cohomology and Homotopy Once smooth sections exist, the next problem is topological: how much information is lost when $X$ is replaced by $Y = X \cap H$? The weak Lefschetz theorem says that in low degrees no information is lost, while in the middle degree the restriction map is still injective. The dimension of $X$ determines where the new topology can appear. [definition: Hyperplane Class] Let $X \subset \mathbb P^N$ be a smooth projective manifold. The hyperplane class of $X$ is \begin{align*} h = c_1(\mathcal O_X(1)) \in H^2(X,\mathbb Z). \end{align*} [/definition] The class $h$ is Poincare dual to a smooth hyperplane section $Y \subset X$. Restricting cohomology classes from $X$ to $Y$ therefore compares the ambient geometry with the geometry of the divisor cut out by $h$. The central topological issue is to determine which cohomology classes survive this restriction unchanged. Weak Lefschetz answers that question by locating the first degree where new cohomology can appear on the hyperplane section. [quotetheorem:3874] [citeproof:3874] The theorem means that a smooth hyperplane section remembers the fundamental cohomology of the ambient manifold up to degree $n-2$. New cohomology can first appear in degree $n-1$, and this new part is called the primitive or vanishing contribution in the Lefschetz theory of hyperplane sections. This is a strong restriction on how complicated a hyperplane section can be relative to the ambient variety. For curves on surfaces, it says only that connectedness is preserved and that first cohomology injects; for threefolds and higher-dimensional varieties, progressively more of the ambient cohomology is forced to reappear unchanged on the section. Thus the theorem separates topology into an inherited part, controlled by restriction, and a middle-dimensional part where genuinely new cycles may occur. Cohomology is only one way to measure this inheritance. To control connectedness and fundamental groups directly, one needs the corresponding statement for homotopy groups of the inclusion $Y\hookrightarrow X$. [quotetheorem:3875] [citeproof:3875] The homotopy version is stronger in low degrees and is often the form used to control connectedness and fundamental groups. For example, when $n \ge 3$, the inclusion of a smooth hyperplane section induces an isomorphism on fundamental groups. [example: Hyperplane Sections of Surfaces] Let $X \subset \mathbb{P}^N$ be a smooth projective surface, so $n = \dim_{\mathbb{C}} X = 2$, and let $C = X \cap H$ be a smooth hyperplane section, which is a smooth algebraic curve. We show that $i^*\colon H^0(X,\mathbb{Z}) \to H^0(C,\mathbb{Z})$ is an isomorphism and that $i^*\colon H^1(X,\mathbb{Z}) \to H^1(C,\mathbb{Z})$ is injective. Both claims follow immediately from the *Weak Lefschetz Theorem for Cohomology*, which states that $i^*\colon H^k(X,\mathbb{Z}) \to H^k(C,\mathbb{Z})$ is an isomorphism for $k < n-1$ and injective for $k = n-1$. Substituting $n = 2$: the isomorphism condition is $k < 2 - 1 = 1$, which — since $k$ is a non-negative integer — forces $k = 0$; the injectivity condition is $k = 2 - 1 = 1$. These two cases account for precisely the two asserted statements. To read each conclusion explicitly: at $k = 0$, since $X$ is connected, $H^0(X,\mathbb{Z}) = \mathbb{Z}$, generated by the constant function $1_X$. The isomorphism $i^*$ forces $H^0(C,\mathbb{Z}) = \mathbb{Z}$ as well and maps $i^*(1_X) = 1_C$, which means exactly that $C$ has one connected component — a smooth hyperplane section of a connected projective surface is itself connected. At $k = 1$, injectivity means $\ker i^* = 0$: every nonzero class $\alpha \in H^1(X,\mathbb{Z})$ restricts to a nonzero class on $C$, so $H^1(X,\mathbb{Z})$ embeds as a subgroup of $H^1(C,\mathbb{Z})$ and the first cohomology of $X$ is faithfully inherited by any smooth hyperplane section. The surface case is the first instance where the theorem leaves something unsaid: it gives a complete answer for $H^0$, a one-sided answer for $H^1$, and says nothing about restriction on $H^2$, which is where the intrinsic geometry of $X$ and $C$ can first diverge. [/example] The surface case is the first place where the theorem stops before determining all cohomology: it controls $H^0$ and gives only an injection in $H^1$. For hypersurfaces, repeating the same comparison across successive hyperplane sections gives a broader low-degree calculation and isolates the middle degree as the first place where degree-dependent phenomena can enter. [example: Low-Degree Cohomology of Smooth Hypersurfaces] [claim]Let $X_d \subset \mathbb{P}^{n+1}$ be a smooth hypersurface of degree $d$ with $\dim_{\mathbb{C}} X_d = n$. For every integer $k$ with $0 \le k < n$, the restriction to $X_d$ induces an isomorphism \begin{align*} H^k(X_d,\mathbb{Z}) \cong H^k(\mathbb{P}^n,\mathbb{Z}). \end{align*} In particular this isomorphism is independent of $d$.[/claim] The argument applies the *Weak Lefschetz Theorem for Cohomology* twice: once to the ambient projective space and once to the Veronese image of $X_d$, where it becomes a hyperplane section. *First application: $H^k(\mathbb{P}^{n+1}) \cong H^k(\mathbb{P}^n)$ for $k < n$.* Take $Z = \mathbb{P}^{n+1}$ of complex dimension $n+1$ and let $W = \mathbb{P}^{n+1} \cap H_0 \cong \mathbb{P}^n$ be a smooth linear hyperplane section. The *Weak Lefschetz Theorem for Cohomology* states that if $Z$ is a smooth projective manifold of complex dimension $m$ and $W = Z \cap H$ is a smooth hyperplane section, then $i^*\colon H^k(Z,\mathbb{Z}) \to H^k(W,\mathbb{Z})$ is an isomorphism for $k < m-1$. Substituting $m = n+1$ gives an isomorphism \begin{align*} i^*\colon H^k(\mathbb{P}^{n+1},\mathbb{Z}) \xrightarrow{\;\sim\;} H^k(\mathbb{P}^n,\mathbb{Z}) \qquad \text{for } k < n. \end{align*} *Second application: $H^k(\mathbb{P}^{n+1}) \cong H^k(X_d)$ for $k < n$.* The obstacle is that $X_d$ is a degree-$d$ divisor, not a linear hyperplane section of $\mathbb{P}^{n+1}$, so the theorem does not apply directly. The Veronese embedding resolves this. Let $M = \binom{n+d+1}{d} - 1$ and define \begin{align*} \nu_d\colon \mathbb{P}^{n+1} \hookrightarrow \mathbb{P}^M, \qquad [Z_0:\cdots:Z_{n+1}] \mapsto [Z^\alpha]_{|\alpha|=d}, \end{align*} the map whose homogeneous coordinates are all degree-$d$ monomials $Z^\alpha = Z_0^{\alpha_0}\cdots Z_{n+1}^{\alpha_{n+1}}$ listed in some fixed order. The differential of $\nu_d$ is injective at every point (the degree-$d$ monomials have no common base locus on $\mathbb{P}^{n+1}$), so $\nu_d$ is a smooth embedding and its image $V = \nu_d(\mathbb{P}^{n+1})$ is a smooth submanifold of $\mathbb{P}^M$ of complex dimension $n+1$. Being a smooth embedding, $\nu_d$ is a homeomorphism from $\mathbb{P}^{n+1}$ onto $V$. Now write the defining form as $F = \sum_{|\alpha|=d} c_\alpha Z^\alpha$. The coefficient vector $(c_\alpha)$ determines a linear form on the coordinates of $\mathbb{P}^M$, hence a hyperplane $H_F = \bigl\{\sum_\alpha c_\alpha w_\alpha = 0\bigr\} \subset \mathbb{P}^M$. By the definition of $\nu_d$, the $\alpha$-th coordinate of $\nu_d(p)$ is $Z^\alpha(p)$, so \begin{align*} \nu_d(p) \in H_F \iff \sum_{|\alpha|=d} c_\alpha Z^\alpha(p) = 0 \iff F(p) = 0 \iff p \in X_d. \end{align*} Therefore $V \cap H_F = \nu_d(X_d)$. This intersection is smooth: $\nu_d$ is a diffeomorphism $X_d \to \nu_d(X_d)$, and $X_d$ is smooth by hypothesis. Apply the *Weak Lefschetz Theorem for Cohomology* to $Z = V$ (complex dimension $n+1$) and the smooth hyperplane section $W = V \cap H_F = \nu_d(X_d)$: since $m = n+1$ gives the cutoff $k < n$, the restriction \begin{align*} i^*\colon H^k(V,\mathbb{Z}) \xrightarrow{\;\sim\;} H^k(\nu_d(X_d),\mathbb{Z}) \qquad \text{for } k < n. \end{align*} Since $\nu_d$ is a homeomorphism $\mathbb{P}^{n+1} \xrightarrow{\sim} V$ and restricts to a homeomorphism $X_d \xrightarrow{\sim} \nu_d(X_d)$, the induced maps on cohomology give $H^k(V,\mathbb{Z}) \cong H^k(\mathbb{P}^{n+1},\mathbb{Z})$ and $H^k(\nu_d(X_d),\mathbb{Z}) \cong H^k(X_d,\mathbb{Z})$. Substituting: \begin{align*} H^k(\mathbb{P}^{n+1},\mathbb{Z}) \cong H^k(X_d,\mathbb{Z}) \qquad \text{for } k < n. \end{align*} *Chaining the two isomorphisms.* Combining the two steps: \begin{align*} H^k(X_d,\mathbb{Z}) \;\cong\; H^k(\mathbb{P}^{n+1},\mathbb{Z}) \;\cong\; H^k(\mathbb{P}^n,\mathbb{Z}) \qquad \text{for } k < n. \end{align*} *Reading off the groups.* The cohomology ring of projective space is $H^*(\mathbb{P}^n,\mathbb{Z}) \cong \mathbb{Z}[h]/(h^{n+1})$ with $\deg h = 2$ (the CW decomposition $\mathbb{P}^n = \mathbb{C}^0 \sqcup \mathbb{C}^1 \sqcup \cdots \sqcup \mathbb{C}^n$ has one cell in each even real dimension and no odd-dimensional cells, so all boundary maps vanish). Restricting to $k < n$, the even degrees $k = 2j$ with $0 \le j \le \lfloor(n-1)/2\rfloor$ give $H^{2j}(\mathbb{P}^n,\mathbb{Z}) = \mathbb{Z}$ with generator $h^j$; odd degrees give $H^k(\mathbb{P}^n,\mathbb{Z}) = 0$. The isomorphism therefore yields \begin{align*} H^k(X_d,\mathbb{Z}) \cong \begin{cases} \mathbb{Z} & k \text{ even},\; 0 \le k < n, \\ 0 & k \text{ odd},\; 0 \le k < n, \end{cases} \end{align*} with the generator of $H^{2j}(X_d,\mathbb{Z})$ being $(h|_{X_d})^j$, where $h|_{X_d} = c_1(\mathcal{O}_{X_d}(1))$ is the restriction of the hyperplane class. The degree $d$ of the defining polynomial leaves no trace in any cohomology group below the middle dimension; the first group that can detect the difference between a quadric, a cubic, or any other smooth hypersurface is $H^n(X_d,\mathbb{Z})$, where primitive cohomology classes encoding the geometry of $F$ can first appear. [/example] ## Hard Lefschetz and the Cohomological Background Weak Lefschetz describes restriction to a divisor. The complementary cohomological question asks what cup product with the hyperplane class does inside $X$ itself. Hard Lefschetz answers this by saying that multiplication by enough powers of the Kähler class identifies cohomology in complementary degrees. [definition: Lefschetz Operator] Let $X$ be a compact Kähler manifold of complex dimension $n$, and let $\omega \in H^2(X,\mathbb R)$ be the cohomology class of a Kähler form. The Lefschetz operator associated to $\omega$ is the linear map \begin{align*} L: H^k(X,\mathbb R) \longrightarrow H^{k+2}(X,\mathbb R), \qquad L(\alpha)=\omega \smile \alpha. \end{align*} [/definition] For a smooth projective manifold $X \subset \mathbb P^N$, the Kähler class may be taken to be the hyperplane class $h=c_1(\mathcal O_X(1))$ after passing to real cohomology. Thus $L$ is geometrically cup product with the cohomology class represented by a hyperplane section. The question is how far this operation can be iterated without losing information. Hard Lefschetz says that powers of the Lefschetz operator do not merely produce classes in higher degree; in complementary degrees they give isomorphisms. [quotetheorem:3876] [citeproof:3876] Hard Lefschetz is an internal symmetry statement for the cohomology ring of a compact Kähler manifold. Multiplication by the Kähler class cannot lose information in low degree; after enough iterations it identifies degree $n-r$ cohomology with degree $n+r$ cohomology. In the projective case this means that repeated intersection with hyperplanes accounts for a large, rigid part of cohomology, while the remaining middle-dimensional information is organized by primitive classes. [definition: Primitive Cohomology] Let $X$ be a compact Kähler manifold of complex dimension $n$ with Lefschetz operator $L$. For $0 \le k \le n$, the primitive cohomology in degree $k$ is \begin{align*} P^k(X) = \ker\left(L^{n-k+1}:H^k(X,\mathbb R) \to H^{2n-k+2}(X,\mathbb R)\right). \end{align*} [/definition] Primitive cohomology measures the part of $H^k(X,\mathbb R)$ not obtained by multiplying lower-degree classes by the Kähler class. In the study of hyperplane sections, this is the part where new geometry appears: it is invisible to purely ambient powers of $h$. To use this notion, primitive classes must not remain a loose definition sitting inside each degree. The structural need is a decomposition theorem saying that every cohomology class is obtained uniquely from primitive pieces by multiplying by powers of the Kähler class. [quotetheorem:3877] [citeproof:3877] [example: Projective Space] Take $X = \mathbb{P}^n$ with Lefschetz operator $L(\alpha) = h \smile \alpha$, where $h = c_1(\mathcal{O}_{\mathbb{P}^n}(1)) \in H^2(\mathbb{P}^n, \mathbb{R})$ is the hyperplane class. [claim]The cohomology ring of $\mathbb{P}^n$ with real coefficients is \begin{align*} H^*(\mathbb{P}^n, \mathbb{R}) \cong \mathbb{R}[h]/(h^{n+1}), \qquad \deg h = 2, \end{align*} with $h^j$ Poincaré dual to the fundamental class of any linear subspace $\mathbb{P}^{n-j} \subset \mathbb{P}^n$, and for each $0 \le k \le n$ the *Hard Lefschetz Theorem* isomorphism $L^{n-k}\colon H^k(\mathbb{P}^n,\mathbb{R}) \xrightarrow{\sim} H^{2n-k}(\mathbb{P}^n,\mathbb{R})$ is given explicitly by cup product with $h^{n-k}$.[/claim] *The cohomology groups.* $\mathbb{P}^n$ has a CW decomposition with exactly one cell in each even real dimension: \begin{align*} \mathbb{P}^n = e^0 \cup e^2 \cup \cdots \cup e^{2n}, \end{align*} where the open $2j$-cell is $e^{2j} = \{[z_0:\cdots:z_{j-1}:1:0:\cdots:0]\} \cong \mathbb{C}^j$. Since consecutive cells occupy real dimensions $2j$ and $2j+2$, the cellular chain group $C_{2j+1} = 0$ for every $j$, and every boundary map $\partial\colon C_{2j} \to C_{2j-1} = 0$ is the zero map. By [Cellular Homology Equals Singular Homology](/theorems/2263), the singular homology is $H_{2j}(\mathbb{P}^n;\mathbb{Z}) = \mathbb{Z}$ for $0 \le j \le n$ and $H_\ell(\mathbb{P}^n;\mathbb{Z}) = 0$ for $\ell$ odd or $\ell > 2n$. Since every group is free, the [Universal Coefficients Theorem](/theorems/2275) gives \begin{align*} H^k(\mathbb{P}^n;\mathbb{R}) = \begin{cases} \mathbb{R} & k = 0, 2, 4, \ldots, 2n, \\ 0 & \text{otherwise.} \end{cases} \end{align*} *The ring structure.* The class $h = c_1(\mathcal{O}_{\mathbb{P}^n}(1))$ is Poincaré dual to $[\mathbb{P}^{n-1}]$: the zero locus of a nonzero section of $\mathcal{O}(1)$ is a hyperplane $\mathbb{P}^{n-1}$, and by [Poincaré Duality for Compact Orientable Smooth Manifolds](/theorems/3598) applied to $\mathbb{P}^n$ (a compact complex manifold of real dimension $2n$), this identifies $h \in H^2$ with the Poincaré dual of $[\mathbb{P}^{n-1}] \in H_{2n-2}$. The cup product $h^j = h \smile \cdots \smile h$ ($j$ factors) is then dual to the fundamental class $[\mathbb{P}^{n-j}]$ obtained by intersecting $j$ general hyperplanes. To verify that $h^j$ is nonzero in $H^{2j}(\mathbb{P}^n;\mathbb{R}) \cong \mathbb{R}$, we pair it with $h^{n-j}$: a linear $\mathbb{P}^{n-j}$ and a linear $\mathbb{P}^j$ meet transversely in exactly one point in general position, so their intersection number is $1$, which means \begin{align*} h^j \smile h^{n-j} = h^n, \qquad \int_{\mathbb{P}^n} h^n = 1. \end{align*} Since the pairing evaluates to $1 \ne 0$, the class $h^j$ is nonzero, hence generates $H^{2j}(\mathbb{P}^n;\mathbb{R}) \cong \mathbb{R}$ for each $0 \le j \le n$. The relation $h^{n+1} = 0$ holds because $H^{2n+2}(\mathbb{P}^n;\mathbb{R}) = 0$ (the highest-dimensional cell is $e^{2n}$), while no relation $h^j = 0$ holds for $j \le n$ since the pairing above is nonzero. The set $\{1, h, h^2, \ldots, h^n\}$ therefore forms a basis for $H^*(\mathbb{P}^n;\mathbb{R})$ in degrees $0, 2, 4, \ldots, 2n$, matching the basis of $\mathbb{R}[h]/(h^{n+1})$ in every degree and establishing the ring isomorphism. *Hard Lefschetz.* For odd $k$, both $H^k(\mathbb{P}^n;\mathbb{R}) = 0$ and $H^{2n-k}(\mathbb{P}^n;\mathbb{R}) = 0$, so $L^{n-k}$ is trivially an isomorphism between zero spaces. For even $k = 2j$ with $0 \le j \le \lfloor n/2 \rfloor$, the operator $L^{n-2j}$ acts by cup product with $h^{n-2j}$: \begin{align*} L^{n-2j}(h^j) = h^{n-2j} \smile h^j = h^{n-j}. \end{align*} The class $h^{n-j}$ generates $H^{2n-2j}(\mathbb{P}^n;\mathbb{R}) \cong \mathbb{R}$, so the map $L^{n-2j}\colon \mathbb{R} \cdot h^j \to \mathbb{R} \cdot h^{n-j}$ sends the unique generator to the unique generator and is an isomorphism. Projective space is the degenerate case of the *Lefschetz Decomposition*: the only primitive class is $1 \in H^0$, every class $h^j \in H^{2j}$ arises as $L^j(1)$, and no degree carries additional primitive cohomology. Hypersurfaces of $\mathbb{P}^{n+1}$ share this structure below the middle dimension — as the weak Lefschetz theorem records — but in degree $n$ they carry primitive classes that do not appear in $\mathbb{P}^n$ and that encode the geometry of the defining polynomial. [/example] Projective space is the ambient model in which the hyperplane class accounts for all cohomology. A hypersurface inherits much of this ambient structure, but its middle cohomology is where primitive classes can appear and where the geometry of the defining equation is recorded. [example: Middle Cohomology of Hypersurfaces] [claim]Let $X_d \subset \mathbb{P}^{n+1}$ be a smooth hypersurface of degree $d$ and complex dimension $n$. For $k \ne n$, the group $H^k(X_d, \mathbb{R})$ is independent of $d$ and isomorphic to $H^k(\mathbb{P}^n, \mathbb{R})$. In the middle degree, the *Lefschetz Decomposition* gives \begin{align*} H^n(X_d, \mathbb{R}) = P^n(X_d) \oplus \begin{cases} \mathbb{R} \cdot (h|_{X_d})^{n/2} & n \text{ even,} \\ 0 & n \text{ odd,} \end{cases} \end{align*} where $P^n(X_d)$ is the primitive cohomology in degree $n$ and the remaining summand is determined by the ambient hyperplane class independently of $d$.[/claim] *Low degrees.* By the *Weak Lefschetz Theorem for Cohomology* applied via the Veronese embedding (carried out in the preceding example), $H^k(X_d, \mathbb{Z}) \cong H^k(\mathbb{P}^n, \mathbb{Z})$ for all $k < n$. The cohomology ring $H^*(\mathbb{P}^n, \mathbb{Z}) \cong \mathbb{Z}[h]/(h^{n+1})$ with $\deg h = 2$ has $\mathbb{Z}$ in even degree $k = 2j$ (generator $(h|_{X_d})^j$) and $0$ in every odd degree, independently of $d$. *High degrees.* Because $X_d$ is a compact smooth orientable manifold of real dimension $2n$, [Poincaré Duality for Compact Orientable Smooth Manifolds](/theorems/3598) gives $H^k(X_d, \mathbb{R}) \cong H^{2n-k}(X_d, \mathbb{R})$ for every $k$. For $k > n$ the complementary degree $2n - k < n$ falls in the low-degree range, so \begin{align*} H^k(X_d, \mathbb{R}) \;\cong\; H^{2n-k}(X_d, \mathbb{R}) \;\cong\; H^{2n-k}(\mathbb{P}^n, \mathbb{R}), \end{align*} which is $\mathbb{R}$ when $k$ is even and $0$ when $k$ is odd, again independently of $d$. *Middle degree.* Let $L = (h|_{X_d}) \smile -$ be the Lefschetz operator on $X_d$. The *Lefschetz Decomposition* (quoted as background in these notes) gives \begin{align*} H^n(X_d, \mathbb{R}) = P^n(X_d) \oplus \bigoplus_{r=1}^{\lfloor n/2 \rfloor} L^r P^{n-2r}(X_d), \end{align*} where the primitive cohomology in degree $j$ is $P^j(X_d) = \ker\!\bigl(L^{n-j+1}\colon H^j(X_d,\mathbb{R}) \to H^{2n-j+2}(X_d,\mathbb{R})\bigr)$. We compute each summand $P^{n-2r}(X_d)$ for $r \ge 1$. Because $n$ and $n - 2r$ share the same parity, the analysis splits cleanly by the parity of $n$. *Case $n$ odd.* For every $r$ with $1 \le r \le (n-1)/2$, the number $n - 2r$ is odd, so $H^{n-2r}(X_d, \mathbb{R}) = 0$ by the low-degree result and $P^{n-2r}(X_d) = 0$. All summands for $r \ge 1$ vanish. *Case $n$ even.* Every number $n - 2r$ with $1 \le r \le n/2$ is even. For $1 \le r \le n/2 - 1$, set $m = (n-2r)/2 \ge 1$. The low-degree result gives $H^{2m}(X_d, \mathbb{R}) = \mathbb{R} \cdot (h|_{X_d})^m$. We test whether this generator is primitive. Since $L$ acts by cup product with $h|_{X_d}$, \begin{align*} L^{n-2m+1}\bigl((h|_{X_d})^m\bigr) = (h|_{X_d})^{m + (n - 2m + 1)} = (h|_{X_d})^{n-m+1}. \end{align*} This class is nonzero: pairing it against $(h|_{X_d})^{m-1}$ via [Poincaré Duality for Compact Orientable Smooth Manifolds](/theorems/3598) gives \begin{align*} \int_{X_d} (h|_{X_d})^{n-m+1} \smile (h|_{X_d})^{m-1} \;=\; \int_{X_d} (h|_{X_d})^n \;=\; d \;\ne\; 0, \end{align*} where $\int_{X_d}(h|_{X_d})^n = d$ because the degree of $X_d$ is $d$ (the number of intersection points of $n$ general hyperplane sections). Since $L^{n-2m+1}((h|_{X_d})^m) \ne 0$, the generator fails the primitivity condition, and as $H^{2m}(X_d, \mathbb{R}) = \mathbb{R} \cdot (h|_{X_d})^m$ is one-dimensional, the kernel of $L^{n-2m+1}$ on this space is zero: $P^{2m}(X_d) = 0$. For $r = n/2$ the degree drops to zero: $P^0(X_d) = \ker\!\bigl(L^{n+1}\colon H^0(X_d,\mathbb{R}) \to H^{2n+2}(X_d,\mathbb{R})\bigr)$. Since $X_d$ has real dimension $2n$, the group $H^{2n+2}(X_d,\mathbb{R}) = 0$, so the kernel is all of $H^0(X_d,\mathbb{R}) = \mathbb{R}$: every constant class is primitive. Applying $L^{n/2}$ to the unit class $1 \in H^0$ gives \begin{align*} L^{n/2}(1) = (h|_{X_d})^{n/2} \;\in\; H^n(X_d, \mathbb{R}). \end{align*} Assembling the two cases: when $n$ is odd every summand with $r \ge 1$ vanishes; when $n$ is even every such summand vanishes except the terminal term $r = n/2$, which contributes $\mathbb{R} \cdot (h|_{X_d})^{n/2}$. The Lefschetz decomposition therefore reads \begin{align*} H^n(X_d, \mathbb{R}) = P^n(X_d) \oplus \begin{cases} \mathbb{R} \cdot (h|_{X_d})^{n/2} & n \text{ even,} \\ 0 & n \text{ odd.} \end{cases} \end{align*} The class $(h|_{X_d})^{n/2}$, present only when $n$ is even, is the restriction of the $(n/2)$-th power of the ambient hyperplane class and carries no information about $d$. Every further dimension of $H^n(X_d,\mathbb{R})$ belongs to $P^n(X_d)$: this is the first cohomological invariant that can vary with the degree, and it is where the defining polynomial leaves its mark on the topology of $X_d$. [/example] The section therefore gives a three-step mechanism. Bertini lets us choose smooth hyperplane sections, weak Lefschetz says that these sections preserve low-dimensional topology, and hard Lefschetz says that the hyperplane class acts as a symmetry operator on cohomology. Together these results make hyperplanes a bridge between projective embeddings and the intrinsic topology of compact Kähler manifolds. The Lefschetz hyperplane theorems of Section 8 relate the global topology of a projective manifold to its hyperplane sections. Section 9 sharpens this by asking what curvature properties hyperplane sections and divisors inherit from the ambient manifold, via the adjunction formula that relates canonical bundles. # 9. Adjunction, Canonical Bundles, and Curvature The previous sections related positivity of holomorphic line bundles to Hermitian metrics, curvature forms, projective embeddings, and hyperplane sections. This section keeps the hyperplane-section viewpoint but asks how canonical bundles change when one passes to a divisor. This section asks what the canonical bundle remembers when we pass to a smooth hypersurface, and how the sign of Ricci curvature is encoded by the canonical and anticanonical bundles. The guiding mechanism is adjunction: an intrinsic top-degree form on a divisor is obtained from an ambient form after accounting for the missing normal direction. The curvature mechanism is Chern-Weil theory: the Ricci form is the curvature of the anticanonical bundle with the metric induced by a Hermitian metric on the tangent bundle. ## Canonical Forms Along a Smooth Divisor What should an $n$-form on a complex $n$-fold become after it is restricted to a codimension-one submanifold? Direct restriction gives zero, because an $n$-form cannot live on an $(n-1)$-dimensional complex manifold. The missing information is the normal direction, and the adjunction formula says that the normal line is measured by the divisor line bundle. [definition: Smooth Divisor] Let $X$ be a complex manifold. A smooth divisor $D$ in $X$ is a closed complex submanifold of complex codimension one such that for every $p\in D$ there is an open neighbourhood $U\subset X$ and a holomorphic function $f_U:U\to \mathbb C$ with $D\cap U=\{x\in U:f_U(x)=0\}$ and $df_U|_p\neq 0$. [/definition] The local defining functions carry more than the set $D$: their ratios record how a normal coordinate changes from chart to chart. To compare canonical forms before and after restricting to $D$, we need a line bundle that remembers this transverse coordinate. The divisor line bundle packages exactly this gluing data, so it is the correction term that replaces the lost normal direction. [definition: Divisor Line Bundle] Let $D\subset X$ be a smooth divisor with local defining functions $f_i$ on an open cover $(U_i)_{i\in I}$. The divisor line bundle $\mathcal O_X(D)$ is the holomorphic line bundle represented on $U_i\cap U_j$ by the transition functions $f_i/f_j$. Its canonical section $s_D$ is locally represented by $f_i$ and has zero set $D$. [/definition] The canonical bundle is the line bundle whose sections are holomorphic volume forms. It records the top exterior power of holomorphic cotangent directions, so it is sensitive to the dimension of the manifold. This is why a divisor cannot inherit an ambient canonical form by ordinary restriction: the divisor has one fewer holomorphic tangent direction. Before stating adjunction, we need the intrinsic line bundle whose transformation under passage to a divisor is being measured. The canonical bundle gives that object for any complex manifold, independently of an embedding or a chosen divisor. [definition: Canonical Bundle] Let $X$ be a complex manifold of dimension $n$. The canonical bundle of $X$ is \begin{align*} K_X := \Lambda^n (T_X^{1,0})^*. \end{align*} A local holomorphic section of $K_X$ is a holomorphic $n$-form. [/definition] Adjunction identifies the intrinsic canonical bundle of the divisor with the ambient canonical bundle twisted by the divisor line bundle. The point is not that an ambient top form restricts to a top form on $D$, but that an ambient top form with one prescribed normal pole has a residue along $D$. Thus the theorem is the precise line-bundle statement behind the slogan "divide out the normal direction". [illustration:scv-iv-conormal-sequence] [quotetheorem:3878] [citeproof:3878] The smoothness hypothesis is essential. It is what allows the divisor to have a genuine normal line bundle, so that the displayed formula is an ordinary identity of line bundles on $D$. For singular divisors, the replacement belongs to the language of dualizing sheaves: these are sheaf-theoretic substitutes for canonical bundles that continue to behave well on singular spaces. Adjunction for pairs is the corresponding singular version, but the simple canonical-bundle statement above no longer has the same meaning at the singular points. Adjunction computes a line bundle, not a preferred metric, not a basis of sections, and not the cohomology groups of $K_D$. Those extra structures require additional input, such as an ambient Hermitian metric, vanishing theorems, or explicit residue calculations. The formula also has a higher-codimension analogue: for a smooth complete intersection cut out by line bundles $L_1,\dots,L_r$, the normal determinant contributes $L_1\otimes\cdots\otimes L_r$. The following coordinate example is included to make the line-bundle statement concrete. It illustrates what the symbols become in the easiest local model, while the general proof remains in the cited theorem. [example: Coordinate Divisor] On $X = \mathbb{C}^n$ with global holomorphic coordinates $(z_1,\dots,z_n)$ and the smooth divisor $D = \{z_n = 0\} \simeq \mathbb{C}^{n-1}$, we show that the *Adjunction Formula* $K_D \simeq (K_X \otimes \mathcal{O}_X(D))|_D$ is realized explicitly by a Poincaré residue that sends $\frac{dz_1\wedge\cdots\wedge dz_n}{z_n}$ to $dz_1\wedge\cdots\wedge dz_{n-1}$. Since $\mathbb{C}^n$ has trivial Picard group, meaning that every holomorphic line bundle on $\mathbb{C}^n$ is isomorphic to the trivial line bundle, both $K_X$ and $\mathcal{O}_X(D)$ are trivial as holomorphic line bundles. The canonical bundle $K_X = \Lambda^n(T^{1,0}_X)^*$ has global generator $dz_1\wedge\cdots\wedge dz_n$. The divisor line bundle $\mathcal{O}_X(D)$ has canonical section $s_D = z_n$ (holomorphic, with zero locus exactly $D$); its role in $K_X\otimes\mathcal{O}_X(D)$ is to allow a simple pole along $D$, so the generator of $K_X\otimes\mathcal{O}_X(D)$ is the meromorphic $n$-form \begin{align*} \frac{dz_1\wedge\cdots\wedge dz_n}{z_n}. \end{align*} To compute the Poincaré residue along $\{z_n=0\}$, we isolate the polar differential $dz_n/z_n$. Because the exterior product is multilinear in each slot, the scalar $1/z_n$ acts on $dz_n$ alone without disturbing the preceding factors: \begin{align*} \frac{dz_1\wedge\cdots\wedge dz_{n-1}\wedge dz_n}{z_n} = dz_1\wedge\cdots\wedge dz_{n-1}\wedge\frac{dz_n}{z_n}. \end{align*} The Poincaré residue of a form $\alpha\wedge(dz_n/z_n)$, where $\alpha$ is holomorphic and involves none of $dz_n$, is defined as $\alpha|_{z_n=0}$. Here $\alpha = dz_1\wedge\cdots\wedge dz_{n-1}$: each $dz_j$ for $j\le n-1$ is a globally constant one-form on $\mathbb{C}^n$ carrying no $z_n$-dependence, so the restriction to $\{z_n=0\}$ leaves it unchanged: \begin{align*} \operatorname{Res}_D\!\left(\frac{dz_1\wedge\cdots\wedge dz_n}{z_n}\right) = dz_1\wedge\cdots\wedge dz_{n-1}\big|_{z_n=0} = dz_1\wedge\cdots\wedge dz_{n-1}. \end{align*} The right-hand side is the standard global generator of $K_D = \Lambda^{n-1}(T^{1,0}_D)^*$ on $D\simeq\mathbb{C}^{n-1}$. The residue map therefore carries the generator of $(K_X\otimes\mathcal{O}_X(D))|_D$ isomorphically to the generator of $K_D$, with no additional scalar factor, confirming that this local model is precisely the determinant computation from the conormal sequence written out in coordinates: removing the polar transverse factor $dz_n/z_n$ is exactly what discarding the normal direction means at the level of differential forms. [/example] ## Hypersurfaces in Projective Space How can the single integer $d$ defining a smooth degree $d$ hypersurface in projective space determine the sign of its canonical bundle? Adjunction converts the degree into a twist by $\mathcal O(d)$, while the canonical bundle of projective space supplies the universal term $\mathcal O(-n-1)$. The calculation begins with the canonical bundle of projective space, obtained from the Euler sequence. [quotetheorem:3879] [citeproof:3879] The formula is the projective-space baseline for adjunction. It says that projective space itself has negative canonical bundle, and every hypersurface calculation begins by comparing the degree of the defining equation with this universal negative term. This formula is the basic calibration example for canonical bundles. It says that projective space has anticanonical bundle $K_{\mathbb P^n}^{-1}\simeq\mathcal O_{\mathbb P^n}(n+1)$, so $\mathbb P^n$ is Fano in the strongest possible elementary sense: its anticanonical bundle is a large positive multiple of the hyperplane bundle. The integer $n+1$ will be the ambient contribution that every hypersurface has to overcome. The Euler-sequence method is also a useful template. Whenever a homogeneous space has a tautological exact sequence, taking determinants often gives its canonical bundle; this is how the same style of computation works for Grassmannians and flag varieties. For the present section, the only extra input needed is that a degree $d$ hypersurface is the zero divisor of a homogeneous polynomial of degree $d$, hence its divisor line bundle is $\mathcal O(d)$. The next calculation applies adjunction to convert this divisor data into the intrinsic canonical bundle of the hypersurface. This is the point where the ambient contribution $\mathcal O(-n-1)$ and the defining degree $d$ combine into a single projective twist. [quotetheorem:3880] [citeproof:3880] This is the projective form of adjunction. The formula is useful because it turns positivity into an integer comparison. Smoothness is again part of the mechanism: a singular hypersurface has a dualizing sheaf, and often a canonical class in a weaker sense, but not necessarily a canonical line bundle obtained from a smooth cotangent bundle. For smooth complete intersections of multidegree $(d_1,\dots,d_r)$ in $\mathbb P^n$, the same adjunction calculation gives $K_X\simeq\mathcal O_X(d_1+\cdots+d_r-n-1)$. This is the engine for the classification later in the section. Once $K_{X_d}$ is expressed as a power of the restricted hyperplane bundle, ampleness, anticanonical ampleness, and triviality become statements about the sign of $d-n-1$. Positivity matters because it controls both algebraic geometry, through sections of tensor powers, and differential geometry, through curvature representatives of Chern classes. [example: Degree d Hypersurfaces] [claim]For a smooth degree $d$ hypersurface $X_d\subset\mathbb P^n$, the canonical bundle $K_{X_d}$ has negative, zero, or positive projective twist according as $d < n+1$, $d = n+1$, or $d > n+1$.[/claim] By *Canonical Bundle Formula for Projective Hypersurfaces*, $K_{X_d}\simeq\mathcal O_{X_d}(d-n-1)$, where $\mathcal O_{X_d}(m):=\mathcal O_{\mathbb P^n}(m)|_{X_d}$. Since $X_d$ is a closed subvariety of $\mathbb P^n$, the restriction $\mathcal O_{X_d}(1)$ is ample on $X_d$; in particular, $\mathcal O_{X_d}(k)$ is ample for every integer $k\ge 1$ and its dual $\mathcal O_{X_d}(-k)\simeq \mathcal O_{X_d}(k)^{-1}$ has ample dual for every $k\ge 1$. **Case $d < n+1$:** Since $d$ and $n+1$ are integers and $d < n+1$, we have $d \le n$, so \begin{align*} d - n - 1 \le n - n - 1 = -1 < 0. \end{align*} Setting $k := n+1-d \ge 1$, the dual of the canonical bundle is \begin{align*} K_{X_d}^{-1} \simeq \mathcal O_{X_d}(d-n-1)^{-1} = \mathcal O_{X_d}(n+1-d) = \mathcal O_{X_d}(k). \end{align*} Because $k\ge 1$, this is ample. The canonical bundle $K_{X_d}\simeq\mathcal O_{X_d}(-k)$ has strictly negative twist. **Case $d = n+1$:** The twist evaluates to $d-n-1 = (n+1)-n-1 = 0$, so \begin{align*} K_{X_d}\simeq\mathcal O_{X_d}(0)\simeq\mathcal O_{X_d}, \end{align*} the structure sheaf of $X_d$. **Case $d > n+1$:** Since $d$ and $n+1$ are integers and $d > n+1$, we have $d \ge n+2$, so \begin{align*} d - n - 1 \ge (n+2) - n - 1 = 1 > 0. \end{align*} Setting $k:=d-n-1\ge 1$, we get $K_{X_d}\simeq\mathcal O_{X_d}(k)$, which is ample. In the classification language of the section, the three cases identify $X_d$ as Fano when $d\le n$ (anticanonical bundle $\mathcal O_{X_d}(n+1-d)$ ample), Calabi-Yau in the canonical-bundle sense when $d=n+1$ (canonical bundle isomorphic to the structure sheaf), and of general type — in fact with ample canonical bundle — when $d\ge n+2$. The entire classification reduces to reading the sign of the single integer $d-n-1$. [/example] ## Ricci Curvature and the Canonical Bundle Which curvature form represents the first Chern class of a complex manifold? For a Hermitian metric, the determinant of the metric matrix gives a Hermitian metric on the canonical bundle. The curvature of this determinant metric is the negative of the Ricci form, so Ricci curvature is the curvature of the anticanonical bundle. [definition: Ricci Form] Let $X$ be a complex manifold of dimension $n$ with a Hermitian metric locally written as \begin{align*} \omega=i\sum_{j,k=1}^n g_{j\bar{k}}\,dz_j\wedge d\bar z_k. \end{align*} The Ricci form of $\omega$ is \begin{align*} \operatorname{Ric}(\omega):=-i\partial\bar\partial\log\det(g_{j\bar{k}}). \end{align*} [/definition] The sign convention matches the fact that positive Ricci curvature is positivity of the anticanonical bundle. The determinant metric on $K_X$ is dual to the determinant metric on the holomorphic tangent bundle, so its curvature changes sign when we pass to $K_X^{-1}$. This is the bridge from a tensorial curvature condition to positivity of a holomorphic line bundle. To make that bridge precise, one must compare the Ricci form with the Chern curvature of the canonical and anticanonical bundles. The following identity records the exact sign relation between these curvature forms. [quotetheorem:3881] [citeproof:3881] This identity is local and Hermitian; the Kahler condition is not needed for the curvature calculation itself. The Kahler hypothesis becomes important when we interpret the Ricci form as a closed representative in de Rham cohomology and compare positivity classes globally. On a non-Kahler complex manifold, the same determinant formula defines the Chern-Ricci form, but its relation to Riemannian Ricci curvature and cohomological positivity is more delicate. The sign comes from the convention that a local line-bundle metric is written as $e^{-\varphi}$ and has curvature $i\partial\bar\partial\varphi$. Since the canonical frame $dz_1\wedge\cdots\wedge dz_n$ has squared norm $(\det g_{j\bar{k}})^{-1}$, the canonical curvature is the negative of the Ricci form. Projective space is the model case where the resulting anticanonical curvature is positive everywhere. [example: Projective Space As Fano] Equip $\mathbb{P}^n$ with the unnormalized Fubini-Study metric $\Omega_{\mathrm{FS}}$, normalized by $i\Theta(\mathcal{O}_{\mathbb{P}^n}(1)) = \Omega_{\mathrm{FS}}$. [claim]$\operatorname{Ric}(\Omega_{\mathrm{FS}}) = (n+1)\Omega_{\mathrm{FS}}$, and $\mathbb{P}^n$ is Fano.[/claim] *Fano property.* By *Canonical Bundle Of Projective Space*, $K_{\mathbb{P}^n} \simeq \mathcal{O}(-n-1)$, so taking duals gives $K_{\mathbb{P}^n}^{-1} \simeq \mathcal{O}(n+1)$. Since $n+1 \geq 1$, the anticanonical bundle is a positive tensor power of $\mathcal{O}(1)$, hence ample, and $\mathbb{P}^n$ is Fano by definition. *Ricci form.* Set $a := 1+|z|^2$ on the standard affine chart $U_0 = \{[1:z_1:\cdots:z_n]\}$, where the normalization gives $\Omega_{\mathrm{FS}}|_{U_0} = i\partial\bar\partial\log a$. Differentiating the potential $\log a$ twice yields the metric matrix \begin{align*} g_{j\bar{k}} = \frac{\partial^2\log a}{\partial z_j\,\partial\bar{z}_k} = \frac{a\,\delta_{jk} - \bar{z}_j z_k}{a^2} = \frac{1}{a}\!\left(\delta_{jk} - \frac{\bar{z}_j z_k}{a}\right)\!, \end{align*} so $\det(g_{j\bar{k}}) = a^{-n}\det\!\bigl(I - a^{-1}\bar{z}\,z^{\top}\bigr)$, where $\bar{z}\,z^{\top}$ is the rank-one $n\times n$ matrix with $(j,k)$-entry $\bar{z}_j z_k$. Setting $u = -a^{-1}\bar{z}$ and $v = z$, the *matrix determinant lemma* gives $\det(I + uv^{\top}) = 1 + v^{\top}u$, so \begin{align*} \det\!\bigl(I - a^{-1}\bar{z}\,z^{\top}\bigr) = 1 + z^{\top}\!\bigl(-a^{-1}\bar{z}\bigr) = 1 - \frac{z^{\top}\bar{z}}{a} = 1 - \frac{|z|^2}{a} = \frac{a - |z|^2}{a} = \frac{1}{a}, \end{align*} where $z^{\top}\bar{z} = \sum_j z_j\bar{z}_j = |z|^2$ and $a - |z|^2 = (1 + |z|^2) - |z|^2 = 1$. Combining the two factors: \begin{align*} \det(g_{j\bar{k}}) = a^{-n}\cdot a^{-1} = (1+|z|^2)^{-(n+1)}. \end{align*} The Ricci form is then \begin{align*} \operatorname{Ric}(\Omega_{\mathrm{FS}}) &= -i\partial\bar\partial\log\det(g_{j\bar{k}}) \\ &= -i\partial\bar\partial\bigl[-(n+1)\log(1+|z|^2)\bigr] \\ &= (n+1)\,i\partial\bar\partial\log(1+|z|^2) \\ &= (n+1)\,\Omega_{\mathrm{FS}}, \end{align*} pulling the scalar $-(n+1)$ out of the logarithm in the second step and recognizing $\Omega_{\mathrm{FS}}|_{U_0} = i\partial\bar\partial\log(1+|z|^2)$ in the third. The two parts are consistent via *Curvature Of Canonical And Anticanonical Bundles*: since $i\Theta(K_{\mathbb{P}^n}^{-1}) = \operatorname{Ric}(\Omega_{\mathrm{FS}})$, the determinant computation recovers the curvature $(n+1)\Omega_{\mathrm{FS}}$ of the anticanonical bundle $\mathcal{O}(n+1)$ directly from the potential $\log(1+|z|^2)$. Dividing by $2\pi$ gives the corresponding first Chern class normalization, confirming that $\mathbb{P}^n$ is the maximally Fano model against which every degree-$d$ hypersurface is measured. [/example] The classification question is not merely how Ricci sign affects a chosen metric, but which canonical-bundle positivity regime that sign forces. What is needed is a compact Kahler dictionary: positive Ricci curvature should point to positivity of the anticanonical bundle, while negative Ricci curvature should point to positivity of the canonical bundle. This direction starts from an actual metric with signed Ricci form; the converse is subtler and belongs to the Calabi-Yau theorem and related existence results for Kahler-Einstein metrics. [quotetheorem:3882] [citeproof:3882] This is the differential-geometric reading of the same Chern class. The theorem packages three familiar geometries into one curvature dictionary: positive Ricci curvature points toward the Fano side, zero Ricci curvature toward the Calabi-Yau side, and negative Ricci curvature toward canonical positivity. Its hypotheses begin with an actual Kähler metric whose Ricci form has a sign, so the conclusion is a metric-to-line-bundle implication rather than an abstract classification statement. This theorem should not be read as saying that every representative of a positive first Chern class has positive Ricci curvature. It starts with a metric whose Ricci form already has a sign and then identifies the resulting curvature form on $K_X^{-1}$ or $K_X$. Conversely, producing a metric with prescribed Ricci form is the content of the Calabi conjecture and Yau's theorem; for example, $c_1(X)=0$ implies the existence of a Ricci-flat Kahler metric in each Kahler class under the usual compact Kahler hypotheses, not that every Kahler metric is Ricci-flat. Compactness also matters in the positivity conclusion because ampleness and global Chern-class positivity are compact complex-geometric notions. On non-compact manifolds, a curvature form can have a sign without giving the same algebraic positivity consequences. Thus the theorem is best viewed as the compact Kahler dictionary between Ricci sign and canonical-bundle sign. ## Fano, Calabi-Yau, and General Type Manifolds Once the canonical bundle is tied to Ricci curvature, the classification question becomes a sign question. Positive anticanonical curvature leads to Fano geometry, zero canonical curvature leads to Calabi-Yau geometry, and positive canonical curvature is the basic curvature model for manifolds of general type. Hypersurfaces in projective space provide the central test family because the sign is read from $d-n-1$. [definition: Fano Manifold] A compact complex manifold $X$ is Fano if the anticanonical line bundle $K_X^{-1}$ is ample. [/definition] For Fano manifolds, the first Chern class $c_1(X)$ lies in the positive cone represented by curvatures of ample line bundles. This is the algebro-geometric counterpart of positive Ricci curvature, although the existence of a Kahler-Einstein metric requires more than ampleness alone. The definition isolates the positive end of the canonical-bundle sign spectrum. [definition: Calabi-Yau Manifold] A compact Kahler manifold $X$ is Calabi-Yau in the canonical-bundle sense if $K_X\simeq\mathcal O_X$. [/definition] This definition is the version needed for the curvature discussion: it makes the first Chern class vanish integrally. Many authors impose extra hypotheses, such as simple connectedness or conditions on $H^q(X,\mathcal O_X)$, depending on the classification problem. Here the point is the canonical bundle itself: triviality is the exact middle case between positive anticanonical bundle and positive canonical bundle. The positive-canonical end of the classification should include manifolds whose canonical bundle has enough pluricanonical sections to control their birational geometry, even when it is not ample everywhere. A line bundle may have base loci or contract subvarieties while still producing an $n$-dimensional supply of sections in high tensor powers. The needed weakening is therefore asymptotic rather than pointwise. [definition: Big Line Bundle] Let $X$ be a compact complex manifold of dimension $n$. A holomorphic line bundle $L\to X$ is big if \begin{align*} \limsup_{m\to\infty}\frac{h^0(X,L^{\otimes m})}{m^n}>0. \end{align*} [/definition] Bigness says that high tensor powers of $L$ have enough sections to see the dimension of $X$. The denominator $m^n$ is the expected growth rate for a genuinely $n$-dimensional supply of sections. By contrast, the trivial bundle has $h^0(X,\mathcal O_X^{\otimes m})=1$ on a connected compact manifold, so its section growth is bounded and it is not big when $n>0$. This asymptotic condition should now be attached to the canonical bundle itself. Classification by canonical sign asks whether the pluricanonical systems are large enough to see the whole manifold, so the manifold-level label is defined by applying bigness to $K_X$. [definition: General Type Manifold] A compact complex manifold $X$ is of general type if $K_X$ is big. [/definition] For the hypersurfaces in this section, general type will be detected by the stronger condition that $K_X$ is ample. This is special to the projective-space setting: every canonical bundle we compute is a power of the restricted hyperplane bundle. The general classification problem is harder because a canonical bundle can be big without being ample, but the hypersurface family collapses the distinction to a sign check. A practical recipe is now visible. To compute $K_X$, start from a known ambient canonical bundle, identify the normal contribution of the equations cutting out $X$, apply adjunction, and then test the resulting line bundle for positivity. Products, blow-ups, and complete intersections each add their own correction term, but the same bookkeeping principle remains: canonical bundles change by recording the directions added, removed, or exceptional. [illustration:scv-iv-hypersurface-trichotomy] [quotetheorem:3883] [citeproof:3883] The standard examples mark the three regimes and show how much information is contained in the integer $d-n-1$. The cubic surface lies on the Fano side, while the quintic threefold lies exactly on the Calabi-Yau boundary. A sextic hypersurface in $\mathbb P^4$ would move one step further to the general-type side, where $K_X\simeq\mathcal O_X(1)$ is already ample. This clean picture depends heavily on the rigidity of $\mathbb P^n$ and its Picard group generated by $\mathcal O(1)$. In a more general ambient variety, adjunction still computes the canonical bundle, but deciding ampleness or bigness may require more than reading off one integer. [example: Cubic Surface And Quintic Threefold] [claim]Let $S\subset\mathbb{P}^3$ be a smooth cubic surface and $Y\subset\mathbb{P}^4$ a smooth quintic threefold. Then $S$ is Fano and $K_Y\simeq\mathcal{O}_Y$.[/claim] *The cubic surface.* Here $d=3$ and $n=3$, so the exponent in the *Canonical Bundle Formula for Projective Hypersurfaces* evaluates to \begin{align*} d - n - 1 = 3 - 3 - 1 = -1, \end{align*} giving $K_S\simeq\mathcal{O}_S(-1)$. Taking the dual line bundle reverses the twist: \begin{align*} K_S^{-1}\simeq\mathcal{O}_S(-1)^{-1}=\mathcal{O}_S(1). \end{align*} The bundle $\mathcal{O}_S(1):=\mathcal{O}_{\mathbb{P}^3}(1)|_S$ is the restriction of the hyperplane bundle to $S$; since $\mathcal{O}_{\mathbb{P}^3}(1)$ is ample on $\mathbb{P}^3$ and ampleness is preserved under restriction to a closed subvariety, $\mathcal{O}_S(1)$ is ample on $S$. Therefore $K_S^{-1}$ is ample, and $S$ is Fano by definition. *The quintic threefold.* Here $d=5$ and $n=4$, so the exponent is \begin{align*} d - n - 1 = 5 - 4 - 1 = 0, \end{align*} giving $K_Y\simeq\mathcal{O}_Y(0)$. Since $\mathcal{O}_Y(0)$ is the structure sheaf $\mathcal{O}_Y$ (the rank-one sheaf with transition functions identically $1$), this reads $K_Y\simeq\mathcal{O}_Y$. The curvature content of each conclusion is different in character. For $S$, ampleness of $K_S^{-1}=\mathcal{O}_S(1)$ means the anticanonical class sits in the positive cone, placing $S$ on the Fano side of the trichotomy alongside $\mathbb{P}^3$ itself (which has $K_{\mathbb{P}^3}^{-1}\simeq\mathcal{O}(4)$, a larger twist). This is a statement about the positivity of the anticanonical line bundle, not by itself an existence theorem for a Kähler metric with positive Ricci form. For $Y$, the isomorphism $K_Y\simeq\mathcal{O}_Y$ gives $c_1(Y)=0$ in $H^2(Y;\mathbb{R})$; the Calabi-Yau theorem then supplies a Ricci-flat Kähler metric in each Kähler class, making the quintic threefold the canonical compact example of zero Ricci curvature. The single integer $d-n-1$ — equal to $-1$ for the cubic and $0$ for the quintic — is the complete invariant separating these two regimes. [/example] These examples summarize the section. Adjunction computes canonical bundles by adding the normal contribution of a divisor, and Ricci curvature reads the curvature of the anticanonical bundle. In projective hypersurfaces, the entire classification by Fano, Calabi-Yau, and general type is governed by the sign of $d-n-1$. # 10. Compact Complex Manifolds and Projectivity Criteria After Kodaira embedding, Lefschetz theory, and adjunction have tied positivity to projective geometry, this section assembles the criteria that determine when a compact complex manifold is projective. Three nested categories appear throughout: compact complex manifolds, compact Kahler manifolds, and projective manifolds. The aim is to characterise the inclusions by analytic invariants — the field of meromorphic functions, integral Kahler classes, and positive line bundles — and to recognise each level through concrete examples (curves, complex tori, Hopf surfaces, Moishezon spaces). The section builds on the curvature and Hodge theory from earlier lectures: Kodaira's vanishing theorem and the $\partial\bar\partial$-lemma are the analytic engines, while Chow's theorem is the bridge to algebraic geometry. [illustration:scv-iv-kahler-projective-hierarchy] ## Meromorphic Functions and Moishezon Manifolds On a compact connected complex manifold, holomorphic functions are constant by the maximum principle, so how can the manifold still carry enough global functions to describe its geometry? The replacement is the field of meromorphic functions: these are allowed to have poles, and poles are precisely what makes projective geometry visible on compact spaces. [definition: Meromorphic Function] Let $X$ be a complex manifold. A meromorphic function on $X$ is a global section of the sheaf $\mathcal M_X$ whose local sections on an open set $U\subset X$ are elements of the total quotient sheaf of $\mathcal O_X(U)$, represented by fractions $p/q$ with $p,q\in\mathcal O_X(U)$ and $q$ not identically zero on any connected component of $U$. [/definition] A meromorphic function on $X$ is the same data as a meromorphic map $f:X\dashrightarrow \mathbb P^1$: away from the analytic set where a chosen denominator vanishes, the ratio $p/q$ defines a holomorphic map to $\mathbb P^1$, and at points where $q$ vanishes the map is determined by the homogeneous values $[p:q]$. The sheaf-theoretic formulation is useful because it does not depend on a chosen fraction; two local quotients define the same meromorphic function when they agree on the complement of an analytic subset with no interior. The total quotient sheaf used here is the sheafification of the presheaf $U\mapsto S(U)^{-1}\mathcal O_X(U)$, where $S(U)$ is the multiplicative system of holomorphic functions on $U$ that are non-zero divisors stalkwise. [example: Meromorphic Functions On Projective Space] Let $X = \mathbb{P}^n$ with homogeneous coordinates $[Z_0:\dots:Z_n]$. We show that the meromorphic functions on $\mathbb{P}^n$ are exactly the degree-zero rational expressions $P(Z)/Q(Z)$ with $P$, $Q$ homogeneous polynomials of equal degree and $Q\not\equiv 0$. *Every degree-zero ratio is meromorphic.* If $P$ and $Q$ are homogeneous of degree $d$, then for any scalar $\lambda\in\mathbb{C}^*$, \begin{align*} \frac{P(\lambda Z_0,\dots,\lambda Z_n)}{Q(\lambda Z_0,\dots,\lambda Z_n)} = \frac{\lambda^d P(Z_0,\dots,Z_n)}{\lambda^d Q(Z_0,\dots,Z_n)} = \frac{P(Z_0,\dots,Z_n)}{Q(Z_0,\dots,Z_n)}, \end{align*} so $P/Q$ is invariant under rescaling of homogeneous coordinates and descends to a well-defined meromorphic function on $\mathbb{P}^n$, holomorphic on the dense open set $\{Q\neq 0\}$. *Every meromorphic function is such a ratio.* Let $f$ be a meromorphic function on $\mathbb{P}^n$, viewed as a meromorphic map $f:\mathbb{P}^n\dashrightarrow\mathbb{P}^1$. Its graph \begin{align*} \Gamma_f = \overline{\{(x,\,f(x)) : x\in\operatorname{dom}(f)\}} \subset \mathbb{P}^n\times\mathbb{P}^1 \end{align*} is a closed analytic subset of $\mathbb{P}^n\times\mathbb{P}^1$. The Segre embedding $\sigma:\mathbb{P}^n\times\mathbb{P}^1\to\mathbb{P}^{2n+1}$ defined by $\sigma([Z],[W])=[Z_i W_j]_{0\le i\le n,\;j\in\{0,1\}}$ is a closed holomorphic embedding (its image is the algebraic locus $\{T_{i0}T_{j1}=T_{i1}T_{j0}\}$ in the target, and it is injective with injective derivative; see [Image of the Segre Embedding](/theorems/2145)), so $\sigma(\Gamma_f)$ is a closed analytic subset of $\mathbb{P}^{2n+1}$. Applying *Chow's Theorem* to $\sigma(\Gamma_f)$ shows that $\Gamma_f$ is an algebraic subvariety of $\mathbb{P}^n\times\mathbb{P}^1$. The projection $\pi_1:\Gamma_f\to\mathbb{P}^n$ is a biholomorphism over the dense open locus where $f$ is defined, so $f = \pi_2\circ\pi_1^{-1}:\mathbb{P}^n\dashrightarrow\mathbb{P}^1$ is a rational map. Over the affine chart $U_0 = \{Z_0\neq 0\}\cong\mathbb{C}^n$ with coordinates $z_j = Z_j/Z_0$ for $1\le j\le n$, this rational map is represented by a ratio of polynomials $p(z_1,\dots,z_n)/q(z_1,\dots,z_n)$. Setting $d = \max(\deg p,\deg q)$ and homogenising, \begin{align*} P(Z_0,\dots,Z_n) &= Z_0^d\,p\!\left(\tfrac{Z_1}{Z_0},\dots,\tfrac{Z_n}{Z_0}\right), \\ Q(Z_0,\dots,Z_n) &= Z_0^d\,q\!\left(\tfrac{Z_1}{Z_0},\dots,\tfrac{Z_n}{Z_0}\right), \end{align*} each term $Z_0^{d-|\alpha|}Z_1^{\alpha_1}\cdots Z_n^{\alpha_n}$ in the expansion has total degree $(d-|\alpha|)+|\alpha| = d$, so $P$ and $Q$ are homogeneous of degree $d$. The functions $f$ and $P/Q$ agree on the nonempty open set $U_0\setminus\{Q=0\}$, so the [Identity Principle](/theorems/3357) gives $f = P/Q$ on all of $\mathbb{P}^n$. The conclusion is that $\mathbb{C}(\mathbb{P}^n)$ is the rational function field $\mathbb{C}(Z_1/Z_0,\dots,Z_n/Z_0)$ in $n$ algebraically independent variables, confirming that $a(\mathbb{P}^n)=n$ and that projective space saturates the Siegel bound. [/example] The number of algebraically independent meromorphic functions measures how close $X$ is to projective geometry. This leads to the first invariant of the section. [definition: Algebraic Dimension] Let $X$ be a compact connected complex manifold. Its field of meromorphic functions is denoted $\mathbb C(X)$. The algebraic dimension of $X$ is \begin{align*} a(X) := \operatorname{trdeg}_{\mathbb C}\mathbb C(X). \end{align*} [/definition] The invariant $a(X)$ is birational: replacing $X$ by a modification does not change its field of meromorphic functions. To make this invariant useful, one needs a universal upper bound showing that a compact complex manifold cannot carry more algebraically independent meromorphic functions than its complex dimension. [quotetheorem:3884] [citeproof:3884] The compactness and connectedness hypotheses are doing real work here. Compactness lets the graph of a meromorphic map have proper image, so the image dimension controls transcendence degree; without compactness, open manifolds such as domains in $\mathbb C^n$ carry many holomorphic functions and the same bound is not the relevant invariant. Connectedness ensures that $\mathbb C(X)$ is a field rather than a product of fields over components. The bound is sharp in two opposite directions. Projective manifolds have $a(X)=n$, while many compact complex manifolds, such as generic higher-dimensional complex tori, have $a(X)=0$. The extreme case raises the definition we need: a compact manifold should be named by the condition that meromorphic functions are as numerous as the dimension permits, even before a projective embedding is known. [definition: Moishezon Manifold] A compact connected complex manifold $X$ of complex dimension $n$ is Moishezon if $a(X)=n$. [/definition] A Moishezon manifold is not assumed to be embedded in projective space. The definition says instead that $X$ is bimeromorphically close to algebraic geometry: after resolving the meromorphic functions, there is a generically finite map to a projective variety. [example: Projective Manifolds Are Moishezon] Let $X\subset\mathbb P^N$ be a projective complex manifold of complex dimension $n$; we show $a(X)=n$. Choose a linear subspace $\Lambda\cong\mathbb P^{N-n-1}\subset\mathbb P^N$ disjoint from $X$. A general such $\Lambda$ avoids $X$ by dimension count: the expected dimension of $X\cap\Lambda$ in $\mathbb P^N$ is \begin{align*} \dim X + \dim\Lambda - N = n + (N-n-1) - N = -1, \end{align*} so a general $(N-n-1)$-plane meets $X$ in the empty set. Fix a complementary subspace $\Pi\cong\mathbb P^n$ with $\Pi\cap\Lambda=\emptyset$; together $\Pi$ and $\Lambda$ span $\mathbb P^N$. The linear projection from $\Lambda$ is the holomorphic map \begin{align*} \pi_\Lambda:\mathbb P^N\setminus\Lambda\to\Pi\cong\mathbb P^n,\qquad p\mapsto\operatorname{span}(p,\Lambda)\cap\Pi, \end{align*} and since $X\cap\Lambda=\emptyset$, its restriction $\pi=\pi_\Lambda|_X:X\to\mathbb P^n$ is well-defined and holomorphic. The fibre of $\pi_\Lambda$ over a point $q\in\Pi$ is the linear span $\overline{q\Lambda}\cong\mathbb P^{N-n}$, and the expected dimension of its intersection with $X$ is \begin{align*} \dim X + \dim\overline{q\Lambda} - N = n + (N-n) - N = 0, \end{align*} so for a general $q$ the fibre $\pi^{-1}(q)=X\cap\overline{q\Lambda}$ is finite. By the *Holomorphic Dimension Formula for Fibre Dimension*, a holomorphic map with finite generic fibre has image dimension equal to the source dimension, so $\dim\pi(X)=n$. Since $X$ is compact, $\pi$ is proper, so $\pi(X)$ is a closed analytic subset of $\mathbb P^n$ by *Remmert's Proper Mapping Theorem*; as $\mathbb P^n$ is irreducible of dimension $n$ and $\dim\pi(X)=n$, we conclude $\pi(X)=\mathbb P^n$, so $\pi$ is surjective. Let $t_j=Z_j/Z_0$ for $1\le j\le n$ be the standard coordinate functions on $\mathbb P^n$, and set $f_j=\pi^*t_j\in\mathbb C(X)$. Surjectivity of $\pi$ makes the pullback $\pi^*:\mathbb C(\mathbb P^n)\to\mathbb C(X)$ injective: any $h\in\mathbb C(\mathbb P^n)$ satisfying $\pi^*h=0$ vanishes on the dense open subset of $\mathbb P^n$ over which $\pi$ is holomorphic and $h$ is defined, so $h=0$ by the [Identity Principle](/theorems/3357). Now suppose $P\in\mathbb C[s_1,\dots,s_n]$ is a polynomial with $P(f_1,\dots,f_n)=0$ in $\mathbb C(X)$. Since $\pi^*$ is a ring homomorphism, \begin{align*} 0 = P(f_1,\dots,f_n) = P(\pi^*t_1,\dots,\pi^*t_n) = \pi^*\!\left(P(t_1,\dots,t_n)\right), \end{align*} and injectivity of $\pi^*$ forces $P(t_1,\dots,t_n)=0$ in $\mathbb C(\mathbb P^n)$. Since $t_1,\dots,t_n$ are algebraically independent in $\mathbb C(\mathbb P^n)$ — as shown in the preceding example on meromorphic functions on projective space — we get $P=0$. Therefore $f_1,\dots,f_n$ are algebraically independent in $\mathbb C(X)$, giving $a(X)\ge n$. The reverse bound $a(X)\le\dim X=n$ is the *Siegel Bound for Algebraic Dimension*, so $a(X)=n$ and $X$ is Moishezon: a projective embedding automatically supplies as many algebraically independent meromorphic functions as the dimension allows. [/example] The projective example raises the structural question for an arbitrary compact complex manifold: can all meromorphic functions be organized by a single map to an algebraic space? The desired mechanism should have target dimension exactly $a(X)$ and should recover the whole field $\mathbb C(X)$ from that target. [quotetheorem:3885] [citeproof:3885] The course uses algebraic reduction as a structural theorem from analytic geometry. Its construction starts from finitely generated subfields of $\mathbb C(X)$, resolves the associated meromorphic maps to projective space, and normalises the resulting image. Several features of the statement deserve emphasis. The pair $(Y,r)$ is unique only up to birational equivalence of $Y$: any two algebraic reductions are connected by a birational isomorphism of normal projective varieties, since both have function field $\mathbb C(X)$. The hypothesis that $Y$ is normal is essential — without it the function field would not determine $Y$ even birationally, and the pullback map $\mathbb C(Y)\to\mathbb C(X)$ would fail to be canonical. When $a(X)=0$ the reduction degenerates: $Y$ is a point, and $r$ is the constant map; this case includes generic complex tori and many non-Kahler surfaces. At the other extreme, $a(X)=n$ recovers the Moishezon condition, and the reduction map is bimeromorphic onto a normal projective variety of dimension $n$. Intermediate values of $a(X)$ describe manifolds fibred (in the meromorphic sense) over a lower-dimensional algebraic base with non-algebraic fibres. ## Chow's Theorem and Projective Manifolds If a compact complex manifold embeds holomorphically into projective space, is its image only an analytic subset, or is it cut out by homogeneous polynomials? Chow's theorem is the result that makes the analytic and algebraic meanings of projective coincide. [definition: Projective Complex Manifold] A compact complex manifold $X$ is projective if there exist $N\in\mathbb N$ and a holomorphic embedding $\Phi:X\to\mathbb P^N$. [/definition] This definition is analytic: it mentions a holomorphic embedding. The missing point is whether the embedded image has any algebraic equations behind it, rather than merely local holomorphic equations in projective charts. Compactness by itself gives strong analytic finiteness, but it does not visibly produce homogeneous polynomials on $\mathbb P^N$. Chow's theorem supplies exactly this bridge: once the ambient space is projective, a closed analytic subspace is forced to be algebraic, so projective complex manifolds can be treated as smooth projective varieties. [quotetheorem:3886] [citeproof:3886] For a projective complex manifold $X\subset\mathbb P^N$, Chow's theorem says that $X$ is the analytification of a smooth projective algebraic variety. The word projective is essential: in affine space $\mathbb C^N$, analytic subsets may be cut out by entire functions with no polynomial description, so the theorem is not a statement about arbitrary analytic sets in non-compact ambient spaces. A concrete counterexample is the zero set of $\sin z$ in $\mathbb C$: it is a $0$-dimensional analytic subset (the lattice $\pi\mathbb Z$) but is not the zero set of any polynomial, since polynomials have only finitely many zeros. Higher-dimensional variants such as the analytic hypersurface $\{e^{z_1}=z_2\}\subset\mathbb C^2$ are likewise transcendental. What fails in the non-compact setting is the global control on growth: Weierstrass preparation still gives local polynomial equations, but without compactness there is no way to bound their degrees and patch them into a global polynomial. The analytic subset $A$ in Chow's theorem may also be reducible or singular; algebraicity is a statement about its defining equations, not about smoothness. Chow's theorem identifies projective analytic subsets with algebraic subsets, but projective geometry also asks for a converse supply of projective embeddings. In dimension one this supply should come from divisors and Riemann-Roch: compact Riemann surfaces ought to have enough meromorphic functions to be placed inside projective space. [quotetheorem:3887] [citeproof:3887] The dimension-one case is special, and several features of the argument do not extend. Riemann-Roch on a compact Riemann surface guarantees that the space $H^0(X,\mathcal O(D))$ grows linearly in the degree of $D$, so increasing the divisor degree past $2g+1$ already gives a very ample line bundle. In dimension $\ge 2$, the corresponding Riemann-Roch formula is an alternating sum involving higher cohomology, and there is no automatic supply of sections: a line bundle may have $H^0(X,L^k)=0$ for all $k\ge 1$, in which case no embedding can be built from its sections. Furthermore, every line bundle on a compact Riemann surface admits a positive Hermitian metric once its degree is positive, while in higher dimension positivity is a Hodge-theoretic condition that ordinary line bundles need not satisfy. These two failures are precisely what motivates the Kodaira embedding criterion below: in higher dimension, projectivity is no longer free, and one must check that there exists a positive line bundle, or equivalently an integral Kahler class. [example: Projective Curves And Surfaces] Let $F \in \mathbb{C}[Z_0, Z_1, Z_2]$ be homogeneous of degree $d$ with smooth zero locus $C = \{F = 0\} \subset \mathbb{P}^2$, and let $G \in \mathbb{C}[Z_0, Z_1, Z_2, Z_3]$ be homogeneous of degree $e$ with smooth zero locus $S = \{G = 0\} \subset \mathbb{P}^3$. We show that $C$ and $S$ are compact projective complex manifolds whose analytic and algebraic structures coincide, and that each carries a Kähler form. The argument is the same for both; we write it out for $C$ and note where $S$ differs only in dimension. *Complex manifold charts.* On the affine chart $U_0 = \{Z_0 \neq 0\}$ with coordinates $z_j = Z_j/Z_0$, the curve $C$ is cut out by the dehomogenization $f(z_1,z_2) = F(1,z_1,z_2)$. Smoothness of $C$ at a point $p = [1:a_1:a_2] \in C \cap U_0$ means $dF|_p \neq 0$; since $F$ is homogeneous of positive degree and $Z_0(p) \neq 0$, Euler's identity $\sum_j Z_j \partial F/\partial Z_j = d\,F = 0$ on $C$ shows this is equivalent to having at least one of $\partial f/\partial z_1(a_1,a_2)$ or $\partial f/\partial z_2(a_1,a_2)$ nonzero. Suppose $\partial f/\partial z_2(a_1,a_2) \neq 0$. The [Implicit Function Theorem](/theorems/52) applied to $f: U_0 \to \mathbb{C}$ yields a disk $D \ni a_1$ and a holomorphic $g: D \to \mathbb{C}$ with $g(a_1) = a_2$ and $f(z_1,g(z_1)) = 0$ for all $z_1 \in D$, such that \begin{align*} C \cap U_0 \cap (D \times \mathbb{C}) = \{(z_1,\,g(z_1)) : z_1 \in D\}; \end{align*} the map $\varphi: z_1 \mapsto [1:z_1:g(z_1)]$ is a holomorphic chart for $C$ near $p$. Where two such charts overlap — one from $\partial f/\partial z_2 \neq 0$ giving coordinate $z_1$, another from $\partial f/\partial z_1 \neq 0$ giving coordinate $z_2$ — the transition sends $z_1 \mapsto g(z_1)$, which is holomorphic. Charts from different affine pieces $U_j$ and $U_k$ overlap on $\{Z_j \neq 0,\, Z_k \neq 0\}$, where the change of affine coordinates $Z_\ell/Z_j \mapsto Z_\ell/Z_k$ is the rational (hence holomorphic) map $(Z_\ell/Z_j)/(Z_k/Z_j)$. Thus $C$ is a one-dimensional complex manifold. For $S \subset \mathbb{P}^3$ the identical application of the [Implicit Function Theorem](/theorems/52) to the dehomogenization $G(1,z_1,z_2,z_3)$ — using the nonvanishing of one of its three partials to solve for one coordinate in terms of the other two — gives $S$ as a two-dimensional complex manifold. *Compactness.* On each affine chart $U_j$, the polynomial $F$ is continuous, so $C \cap U_j$ is closed in $U_j$; their union $C$ is therefore closed in $\mathbb{P}^2$. The space $\mathbb{P}^2$ is compact: it is the quotient of the compact sphere $S^5 \subset \mathbb{C}^3$ by the free continuous $U(1)$-action $\lambda \cdot Z = \lambda Z$, and quotients of compact spaces are compact. A closed subset of a compact space is compact, so $C$ is compact. The same closed-subset argument gives compactness of $S \subset \mathbb{P}^3$. *Analytic and algebraic structures coincide.* As a smooth submanifold, $C$ is a closed complex analytic subset of $\mathbb{P}^2$. Chow's theorem (stated earlier in this section) asserts that every closed complex analytic subset of $\mathbb{P}^N$ is algebraic — the common zero locus of finitely many homogeneous polynomials. Here $C$ is already presented as $\{F = 0\}$, so Chow's theorem confirms that the analytic description and the algebraic description are the same object. Concretely: on the affine chart $U_0$, the holomorphic functions on $C \cap U_0$ that extend across all of $C$ are precisely the polynomial functions in $z_1,z_2$ modulo the ideal $(f)$, which are also the regular functions on the affine algebraic curve $\{f = 0\}$. The global identification is provided by Chow's theorem at the level of varieties. The same applies to $S$. *The Fubini-Study form restricts to a Kähler form.* The unnormalized Fubini-Study form $\Omega_{\mathrm{FS}}$ on $\mathbb{P}^2$, established in the earlier lectures, is a Kähler form: a real closed $(1,1)$-form that is positive definite on each tangent space. Let $\iota: C \hookrightarrow \mathbb{P}^2$ be the inclusion. The pullback $\iota^*\Omega_{\mathrm{FS}}$ is closed because the exterior derivative commutes with pullback: \begin{align*} d(\iota^*\Omega_{\mathrm{FS}}) = \iota^*(d\,\Omega_{\mathrm{FS}}) = \iota^*\,0 = 0. \end{align*} It is positive on $TC$: for any nonzero $v \in T_pC$, holomorphicity of $\iota$ means $d\iota_p$ intertwines the complex structure, so $d\iota_p(iv) = i\,d\iota_p(v)$; since $\iota$ is an embedding $d\iota_p$ is injective, giving a nonzero vector $d\iota_p(v) \in T_{\iota(p)}\mathbb{P}^2$, and therefore \begin{align*} (\iota^*\Omega_{\mathrm{FS}})(v,\,iv) = \Omega_{\mathrm{FS}}\!\bigl(d\iota_p\,v,\; d\iota_p(iv)\bigr) = \Omega_{\mathrm{FS}}\!\bigl(d\iota_p\,v,\; i\,d\iota_p\,v\bigr) > 0, \end{align*} where the last inequality is the Kähler positivity of $\Omega_{\mathrm{FS}}$ applied to the nonzero vector $d\iota_p\,v$. The same restriction gives a Kähler form on $S \subset \mathbb{P}^3$. Both $C$ and $S$ are thus compact complex manifolds, already presented as zero sets of homogeneous polynomials (hence projective), and carrying Kähler forms via restriction of $\Omega_{\mathrm{FS}}$; they are the canonical instances where the analytic category and the algebraic category produce exactly the same object. [/example] These examples are the benchmarks for the rest of the section. The projective condition is strong because it provides meromorphic functions, integral cohomology classes, and a positive curvature form at the same time. ## Kodaira's Criterion for Compact Kahler Manifolds Projective space carries the Fubini-Study form and the hyperplane line bundle, so projective manifolds inherit positive integral $(1,1)$-classes. The central question is whether the converse holds for compact Kahler manifolds: does an integral positive class force a projective embedding? [definition: Integral Kahler Class] Let $X$ be a compact complex manifold. An integral Kahler class on $X$ is a class $\alpha\in H^2(X,\mathbb R)$ represented by a Kahler form and lying in the image of the natural map $H^2(X,\mathbb Z)\to H^2(X,\mathbb R)$. [/definition] The normalisation of curvature may insert a fixed factor of $2\pi$ between a curvature form and a first Chern class. To pass from an integral Kahler class to projective coordinates, we need the class represented as the curvature of an actual holomorphic line bundle with a positive metric. [definition: Positive Holomorphic Line Bundle] Let $L\to X$ be a holomorphic line bundle with Hermitian metric $h$. The bundle $L$ is positive if its Chern curvature form \begin{align*} c_1(L,h)=\frac{i}{2\pi}\Theta_h \end{align*} is a Kahler form on $X$. [/definition] Positive line bundles are the bridge between curvature and projective coordinates. The remaining projectivity problem is to decide whether the existence of such a curvature form is exactly the same as the existence of an integral Kahler class, because then the analytic positivity condition can be tested cohomologically. [quotetheorem:3888] [citeproof:3888] This criterion is a projectivity test stated entirely in differential-geometric terms. A compact Kahler manifold may have many Kahler classes, but projectivity requires one of them to be rational, hence integral after rescaling. [example: Projective And Non-Projective Complex Tori] Let $X = V/\Lambda$ be a complex torus, where $V \cong \mathbb{C}^g$ and $\Lambda$ is a rank-$2g$ lattice spanning $V$ over $\mathbb{R}$. [claim]The torus $X$ is projective if and only if $\Lambda$ admits a Riemann form: an alternating $\mathbb{Z}$-bilinear map $E:\Lambda\times\Lambda\to\mathbb{Z}$ whose $\mathbb{R}$-linear extension to $V\times V$ satisfies $E(iv,iw)=E(v,w)$ for all $v,w\in V$, and for which the Hermitian pairing $H(v,w)=E(iv,w)+iE(v,w)$ is positive definite. When this holds, $X$ is called an abelian variety.[/claim] *The torus is always Kähler.* Any Hermitian inner product $h_0$ on $V$ is translation-invariant and $\Lambda$-periodic, so it descends to a Hermitian metric on the trivial bundle $TX\cong X\times V$. Its imaginary part $\omega_0=-\operatorname{Im}(h_0)$ satisfies $d\omega_0=0$ because $h_0$ is constant in flat coordinates on $V$, and for any nonzero $v\in V$, \begin{align*} \omega_0(v,iv) = -\operatorname{Im}\bigl(h_0(v,iv)\bigr) = -\operatorname{Im}\bigl(i\,h_0(v,v)\bigr) = h_0(v,v)>0, \end{align*} so $\omega_0$ is a flat Kähler form on $X$, independent of any Riemann form. *$H$ is Hermitian.* Substituting $w\mapsto -iw$ in $E(iv,iw)=E(v,w)$ gives $E(iv,i(-iw))=E(v,-iw)$, i.e. \begin{align*} E(iv,w) = -E(v,iw). \tag{$*$} \end{align*} Using $(*)$ and the alternating property $E(u,v)=-E(v,u)$: \begin{align*} \overline{H(w,v)} = \overline{E(iw,v)+iE(w,v)} = E(iw,v)-iE(w,v) = -E(v,iw)+iE(v,w) = E(iv,w)+iE(v,w) = H(v,w), \end{align*} where the third equality uses $E(iw,v)=-E(v,iw)$ (alternating) and $-iE(w,v)=iE(v,w)$ (alternating), and the fourth uses $(*)$. *The associated real $(1,1)$-form is positive.* Define $\omega_H(u,v)=-E(u,v)$; since $E$ is constant, $d\omega_H=0$. Setting $w=v$ in $H(v,w)=E(iv,w)+iE(v,w)$ and using $E(v,v)=0$: \begin{align*} H(v,v) = E(iv,v) = -\omega_H(iv,v) = \omega_H(v,iv). \end{align*} Since $H$ is positive definite, $\omega_H(v,iv)=H(v,v)>0$ for all nonzero $v$, so $\omega_H$ is a Kähler form on $X$. The type computation $\omega_H(iu,iv)=-E(iu,iv)=-E(u,v)=\omega_H(u,v)$ (using $E(iv,iw)=E(v,w)$) confirms $\omega_H$ is of type $(1,1)$. *$[\omega_H]$ is integral.* Write $\Lambda = \mathbb{Z}\lambda_1\oplus\cdots\oplus\mathbb{Z}\lambda_{2g}$. The map $(s,t)\mapsto s\lambda_j+t\lambda_k \bmod\Lambda$ from $[0,1]^2$ to $X$ parametrises a generator of $H_2(X,\mathbb{Z})$, and the pairing of $[\omega_H]$ with this $2$-cycle is \begin{align*} \int_0^1\!\int_0^1\omega_H\!\Bigl(\tfrac{\partial}{\partial s}(s\lambda_j+t\lambda_k),\,\tfrac{\partial}{\partial t}(s\lambda_j+t\lambda_k)\Bigr)ds\,dt = \int_0^1\!\int_0^1\omega_H(\lambda_j,\lambda_k)\,ds\,dt = -E(\lambda_j,\lambda_k)\in\mathbb{Z}, \end{align*} where the partial derivatives evaluate to the constant vectors $\lambda_j$ and $\lambda_k$. Thus $[\omega_H]\in H^2(X,\mathbb{Z})$. *Riemann form implies projective.* Since $[\omega_H]$ is simultaneously the class of a Kähler form and lies in $H^2(X,\mathbb{Z})$, it is an integral Kähler class. The *Kodaira Projectivity Criterion* applies and provides a holomorphic embedding of $X$ in projective space, so $X$ is projective. *Projective implies a Riemann form.* If $\Phi:X\hookrightarrow\mathbb{P}^N$ is a holomorphic embedding, set $L=\Phi^*\mathcal{O}(1)$ with the metric pulled back from the Fubini-Study metric on $\mathcal{O}(1)$. The curvature form $\omega_L=\frac{i}{2\pi}\Theta_h$ is of type $(1,1)$: by the [Local Formula for Curvature](/theorems/1540), $\Theta_h=\bar\partial(\partial\log h)$, which is a sum of $d\bar z_j\wedge dz_k$ terms. The hyperplane class $c_1(\mathcal{O}(1))$ generates $H^2(\mathbb{P}^N,\mathbb{Z})\cong\mathbb{Z}$, so $[\omega_L]=\Phi^*c_1(\mathcal{O}(1))\in H^2(X,\mathbb{Z})$. Define $E(\lambda,\mu)=-\omega_L(\lambda,\mu)$ for $\lambda,\mu\in\Lambda$ by the same pairing computation, extended $\mathbb{Z}$-bilinearly; this is integer-valued by integrality of $[\omega_L]$. The $(1,1)$-type of $\omega_L$ gives $E(iv,iw)=-\omega_L(iv,iw)=-\omega_L(v,w)=E(v,w)$, using $\omega_L(iu,iv)=\omega_L(u,v)$ (which holds for any $(1,1)$-form). Finally, setting $w=v$: \begin{align*} H(v,v) = E(iv,v) = -\omega_L(iv,v) = \omega_L(v,iv)>0, \end{align*} where the last inequality is Kähler positivity of $\omega_L$. So $H$ is positive definite and $E$ is a Riemann form. The example exposes why Kähler and projective diverge when $g\geq 2$. By [Dolbeault's Theorem](/theorems/3389) applied to the flat metric on $X$, the harmonic $(p,q)$-forms are the constant forms, so $h^{p,q}(X)=\binom{g}{p}\binom{g}{q}$. In higher dimension the existence of a Kähler form does not force the existence of an integral Kähler class; equivalently, a torus may have no Riemann form and hence no positive line bundle. In dimension $g=1$ this obstruction is absent: for any $\Lambda=\mathbb{Z}+\mathbb{Z}\tau$ with $\operatorname{Im}\tau>0$, setting $E(m+n\tau,m'+n'\tau)=mn'-nm'$ for $m,n,m',n'\in\mathbb{Z}$ defines an integer alternating form, and $H(v,v)=E(iv,v)=|v|^2/\operatorname{Im}\tau>0$, so every $1$-dimensional complex torus is an abelian variety. [/example] The example shows why Kahler and projective are not the same in higher dimension. Positivity as a real form is not enough; arithmetic information in integral cohomology is also required. ## Moishezon Manifolds in the Kahler Setting Moishezon manifolds have as many meromorphic functions as possible, but meromorphic maps may have base loci and exceptional sets. When does this birational algebraic information become an actual projective embedding of the original manifold? [quotetheorem:3889] [citeproof:3889] The theorem explains the role of the Kahler hypothesis. Moishezon is a birational condition about meromorphic functions, while projectivity is a biregular condition about embeddings; the Kahler form supplies the positivity needed to pass from one to the other. [remark: Role Of The Kahler Hypothesis] There are compact Moishezon manifolds that are not projective. Hironaka's examples are smooth compact complex manifolds bimeromorphic to projective manifolds, but they fail to be Kahler. Thus the obstruction is not a lack of meromorphic functions; it is the absence of a compatible positive closed $(1,1)$-form. [/remark] This result also clarifies how algebraic reduction behaves in the maximal case. For a compact Kahler manifold, maximal algebraic dimension forces projectivity, not just birational projectivity. ## Compact Complex, Compact Kahler, and Projective Manifolds The section leaves three natural categories: compact complex manifolds, compact Kahler manifolds, and projective manifolds. Which implications always hold, and where do the reverse implications fail? [quotetheorem:3890] [citeproof:3890] This hierarchy is the main organizing principle for compact complex geometry. Projective manifolds are the most rigid class, compact Kahler manifolds retain Hodge theory and metric positivity, and compact complex manifolds allow phenomena outside algebraic geometry. [example: Hopf Surface] [claim]The Hopf surface $X = (\mathbb{C}^2\setminus\{0\})/\langle A\rangle$, where $A(z_1,z_2)=(2z_1,2z_2)$, is a compact complex manifold that is not Kähler, and hence not projective.[/claim] *Free and properly discontinuous action.* The iterates satisfy $A^n(z_1,z_2)=(2^n z_1,2^n z_2)$ for every $n\in\mathbb{Z}$. For freeness: if $A^n(z)=z$ for some $n\neq 0$ and $z\in\mathbb{C}^2\setminus\{0\}$, then $2^n z = z$, so $(2^n-1)z=0$; since $z\neq 0$ this forces $2^n=1$, which holds only for $n=0$. For proper discontinuity: given a compact $K\subset\mathbb{C}^2\setminus\{0\}$, choose $0<r\leq R<\infty$ with $K\subset\{r\leq|z|\leq R\}$. Since $|A^n(z)|=2^n|z|$, for every integer $n>\log_2(R/r)$ the entire image $A^n(K)$ satisfies $|A^n(z)|\geq 2^n r>R$, hence $A^n(K)\cap K=\emptyset$; similarly, for $n<-\log_2(R/r)$ every point of $A^n(K)$ has $|A^n(z)|\leq 2^n R<r$, again giving $A^n(K)\cap K=\emptyset$. Only finitely many $n$ can violate either bound, so $\langle A\rangle$ acts properly discontinuously. *The quotient is a complex manifold.* Because $A$ is a biholomorphism of $\mathbb{C}^2\setminus\{0\}$ and $\langle A\rangle$ acts freely and properly discontinuously, the projection $\pi:\mathbb{C}^2\setminus\{0\}\to X$ is a local biholomorphism and $X$ inherits a complex manifold structure whose charts are restrictions of holomorphic coordinates from $\mathbb{C}^2\setminus\{0\}$. *Diffeomorphism $X\cong S^1\times S^3$.* Write each nonzero $z$ in polar form $z=r\sigma$ with $r=|z|>0$ and $\sigma=z/|z|\in S^3\subset\mathbb{C}^2$. In these coordinates $A^n(r\sigma)=2^n r\cdot\sigma$, so $A$ multiplies the radial factor by $2^n$ while leaving $\sigma$ fixed. Define \begin{align*} \Phi:S^3\times(\mathbb{R}/\mathbb{Z})\to X,\qquad \Phi(\sigma,t)=\bigl[2^t\sigma\bigr], \end{align*} where $[\cdot]$ denotes the $\langle A\rangle$-orbit. This is well-defined because $\Phi(\sigma,t+1)=[2^{t+1}\sigma]=[A(2^t\sigma)]=[2^t\sigma]=\Phi(\sigma,t)$. It is surjective: every $z=r\sigma$ equals $[2^{\log_2 r}\sigma]$. It is injective: if $[2^s\sigma]=[2^t\tau]$ then $2^s\sigma=A^n(2^t\tau)=2^{t+n}\tau$ for some $n\in\mathbb{Z}$; comparing norms gives $s=t+n$, hence $\sigma=\tau$, and then $s\equiv t\pmod{\mathbb{Z}}$. The map $\Phi$ and its inverse are smooth, so $X\cong S^3\times S^1$ as smooth manifolds. *Computing $b_1(X)=1$.* By the [Künneth Theorem for De Rham Cohomology](/theorems/3591), \begin{align*} H^1(S^3\times S^1,\mathbb{R})\cong\bigoplus_{p+q=1}H^p(S^3,\mathbb{R})\otimes H^q(S^1,\mathbb{R}). \end{align*} Since $S^3$ is simply connected, $H_1(S^3,\mathbb{Z})=0$ by the Hurewicz theorem, so $H^1(S^3,\mathbb{R})=0$. The circle has $H^0(S^1,\mathbb{R})=\mathbb{R}$ and $H^1(S^1,\mathbb{R})=\mathbb{R}$ (generated by $d\theta$), while $H^0(S^3,\mathbb{R})=\mathbb{R}$. The only nonzero summand in degree $1$ is $p=0$, $q=1$: \begin{align*} H^1(S^3\times S^1,\mathbb{R})\cong H^0(S^3,\mathbb{R})\otimes H^1(S^1,\mathbb{R})=\mathbb{R}\otimes\mathbb{R}=\mathbb{R}, \end{align*} so $b_1(X)=1$. *Hodge decomposition rules out a Kähler structure.* Suppose for contradiction that $X$ admits a Kähler metric. The [Hodge Decomposition](/theorems/2745) on a compact Kähler manifold gives $H^1(X,\mathbb{C})=H^{1,0}(X)\oplus H^{0,1}(X)$, and complex conjugation on differential forms interchanges $(p,q)$- and $(q,p)$-types, giving $h^{0,1}(X)=h^{1,0}(X)$. Therefore \begin{align*} b_1(X)=\dim_{\mathbb{C}}H^1(X,\mathbb{C})=h^{1,0}(X)+h^{0,1}(X)=2h^{1,0}(X), \end{align*} which is even. But $b_1(X)=1$ is odd — a contradiction. So $X$ is not Kähler. Since every projective manifold is Kähler — the Fubini-Study form on $\mathbb{P}^N$ restricts to a Kähler form on any holomorphically embedded submanifold — the odd first Betti number simultaneously prevents $X$ from being projective, placing the Hopf surface strictly outside the Kähler world in the hierarchy of compact complex geometries. [/example] The Hopf surface gives the first obstruction in the hierarchy: topology alone can rule out Kahler geometry. The next comparison places this obstruction beside the arithmetic obstruction for projectivity, so the three categories are separated by concrete examples rather than only by definitions. [example: Three Levels In Dimension Two] The three representative examples are a smooth surface $S = \{G = 0\} \subset \mathbb{P}^3$ of degree $e \geq 1$, a generic two-dimensional complex torus $T = \mathbb{C}^2/\Lambda$, and the Hopf surface $H = (\mathbb{C}^2\setminus\{0\})/\langle A\rangle$ with $A(z_1,z_2)=(2z_1,2z_2)$. We show that $S$ is projective, $T$ is compact Kähler but not projective, and $H$ is compact complex but not Kähler, placing one manifold at each level of the hierarchy. *$S$ is projective.* The inclusion $\iota: S \hookrightarrow \mathbb{P}^3$ is a holomorphic embedding by definition of a smooth hypersurface (see the earlier example on projective curves and surfaces in this section, where the [Implicit Function Theorem](/theorems/52) supplies complex manifold charts and compactness follows from $S$ being a closed subset of the compact space $\mathbb{P}^3$). This embedding is the required data in the definition of projectivity, so $S$ is projective. The hyperplane bundle $L = \iota^*\mathcal{O}(1)$ equipped with the metric pulled back from the Fubini-Study metric on $\mathcal{O}(1)$ has curvature form $i\Theta(L,h)=\iota^*\Omega_{\mathrm{FS}}$ and normalized first Chern form $c_1(L,h)=\iota^*\omega_{\mathrm{FS}}$, where $\omega_{\mathrm{FS}}=\frac{1}{2\pi}\Omega_{\mathrm{FS}}$. The pullback is Kähler on $S$: $d(\iota^*\Omega_{\mathrm{FS}}) = \iota^*(d\Omega_{\mathrm{FS}}) = 0$, and for any nonzero $v \in T_pS$ the injectivity of $d\iota_p$ gives $d\iota_p(v) \neq 0$, so Kähler positivity of $\Omega_{\mathrm{FS}}$ yields $(\iota^*\Omega_{\mathrm{FS}})(v,iv) = \Omega_{\mathrm{FS}}(d\iota_p\,v,\,i\,d\iota_p\,v) > 0$. Thus $L$ is a positive line bundle on $S$. *$T$ is compact Kähler.* Any Hermitian inner product $h_0$ on $\mathbb{C}^2$ is constant in flat coordinates, so the $(1,1)$-form $\omega_0 = -\operatorname{Im}(h_0)$ satisfies $d\omega_0 = 0$ (the coefficients of $\omega_0$ in any flat frame are constant, hence their exterior derivative vanishes). For any nonzero $v \in \mathbb{C}^2$, \begin{align*} \omega_0(v,iv) = -\operatorname{Im}(h_0(v,iv)) = -\operatorname{Im}(i\,h_0(v,v)) = h_0(v,v) > 0, \end{align*} where the second step uses sesquilinearity of $h_0$ and the third uses $-\operatorname{Im}(i\,r) = r$ for real $r = h_0(v,v)$. Since $h_0$ is $\Lambda$-periodic, $\omega_0$ descends to a Kähler form on $T$. *$T$ is not projective for generic $\Lambda$.* A holomorphic line bundle $L \to T$ with positive curvature would supply an integral Kähler class by the *Kodaira Projectivity Criterion* stated in this section. Equivalently for a torus, it would supply a Riemann form on the lattice: an integral alternating form whose associated Hermitian pairing is positive definite. For a generic two-dimensional complex torus no such Riemann form exists, so no positive line bundle exists and the criterion gives no projective embedding. Thus $T$ is Kähler but not projective. *$H$ is compact complex but not Kähler.* The quotient construction makes $H$ a compact complex manifold: $\langle A \rangle$ acts freely and properly discontinuously on $\mathbb{C}^2\setminus\{0\}$ (established in detail in the Hopf surface example earlier in this section), so the quotient map $\pi:\mathbb{C}^2\setminus\{0\}\to H$ is a local biholomorphism and $H$ inherits a complex structure. The diffeomorphism $H \cong S^3 \times S^1$ constructed there via $\Phi(\sigma,t) = [2^t\sigma]$ gives, by the [Künneth Theorem for De Rham Cohomology](/theorems/3591), \begin{align*} H^1(H,\mathbb{R}) \cong H^1(S^3\times S^1,\mathbb{R}) \cong \bigoplus_{p+q=1} H^p(S^3,\mathbb{R})\otimes H^q(S^1,\mathbb{R}). \end{align*} Since $S^3$ is simply connected, $H^1(S^3,\mathbb{R}) = 0$. The only nonzero summand in degree $1$ is $p=0$, $q=1$: \begin{align*} H^1(H,\mathbb{R}) \cong H^0(S^3,\mathbb{R})\otimes H^1(S^1,\mathbb{R}) = \mathbb{R}\otimes\mathbb{R} = \mathbb{R}, \end{align*} so $b_1(H) = 1$. Now suppose for contradiction that $H$ admits a Kähler metric. The [Hodge Decomposition](/theorems/2745) on a compact Kähler manifold gives $H^1(H,\mathbb{C}) = H^{1,0}(H) \oplus H^{0,1}(H)$, and complex conjugation on forms interchanges the $(1,0)$- and $(0,1)$-summands, so $h^{0,1}(H) = h^{1,0}(H)$. Therefore \begin{align*} b_1(H) = \dim_\mathbb{C} H^1(H,\mathbb{C}) = h^{1,0}(H) + h^{0,1}(H) = 2\,h^{1,0}(H), \end{align*} which is even. But $b_1(H) = 1$ is odd, a contradiction. Hence $H$ is not Kähler, and since every projective manifold is Kähler (the Fubini-Study form restricts to a Kähler form on any holomorphic submanifold of $\mathbb{P}^N$), $H$ is not projective either. The three examples thus occupy distinct levels: $S$ is projective, $T$ is compact Kähler but not projective, and $H$ is compact complex but not Kähler. The obstruction separating $T$ from $S$ is arithmetic — a missing integral $(1,1)$-class — while the obstruction separating $H$ from $T$ is topological — an odd first Betti number that is incompatible with the Hodge symmetry present on any compact Kähler manifold. [/example] The course therefore ends with a precise comparison. Meromorphic functions measure algebraic dimension, positive line bundles produce embeddings, and integral Kahler classes decide projectivity among compact Kahler manifolds. The compact complex category is wider than the algebraic one, but the results in this section identify the exact extra hypotheses that recover projective geometry. In practice, the criteria above are applied as follows. To test whether a complex torus $V/\Lambda$ is projective, one looks for a Riemann form on the lattice: an integral alternating form whose associated Hermitian pairing is positive definite. For a generic lattice in dimension at least $2$, no such integral form exists, so the torus is Kähler but not projective. To compute the algebraic dimension $a(X)$ of a given compact complex manifold, one searches for algebraically independent meromorphic functions and applies the Siegel bound: in concrete cases, this is done by exhibiting an explicit meromorphic map to a projective variety and computing the dimension of its image. To recognise a Moishezon manifold from a concrete construction, the standard test is to produce a bimeromorphic map to a projective variety — typical sources are small modifications (blow-ups along submanifolds, small contractions) of projective manifolds, and any such construction yields a Moishezon manifold by definition. Combined with the Kahler criterion above, this gives the working principle: smooth modifications of projective varieties are Moishezon, and they are projective exactly when they remain Kahler, a condition that may fail in dimension three and beyond as in Hironaka's examples. Section 10 established abstract criteria for projectivity using the Kähler condition, Moishezon property, and integral positivity. Section 11 now exploits these criteria to read off geometric meaning from the curvature of the canonical bundle, showing that positive or negative Ricci curvature constrains metric geometry, fundamental group, and holomorphic functions in parallel with Kähler geometry. # 11. Curvature, Metrics, and Geometric Applications Sections 1-3 built Hermitian metrics, curvature forms, and positivity for holomorphic bundles, while Section 9 identified Ricci curvature with the curvature of the canonical and anticanonical bundles. This section turns those tools into geometric consequences: Ricci curvature detects the sign of the canonical bundle, holomorphic sectional curvature controls complex curves, and Schwarz lemma arguments convert curvature inequalities into restrictions on holomorphic maps. The guiding theme is that negative curvature is not only a local tensorial condition; through the maximum principle it becomes a global obstruction to maps from flat or positively curved sources. ## Ricci Curvature and the Canonical Bundle How does the curvature of a Hermitian metric on a complex manifold remember the positivity or negativity of the canonical bundle? The answer is a sign reversal: the Ricci form represents the first Chern class of the holomorphic tangent bundle, while the curvature of the canonical bundle represents the negative of that class. This is the bridge between Riemannian-looking curvature inequalities and the algebro-geometric language of $K_X$ and $K_X^{-1}$. Let $X$ be a complex manifold of dimension $n$, and let $\theta=(\theta_{j\bar{k}})$ denote the local Hermitian matrix of a Hermitian metric $\theta$ on $T^{1,0}X$ in a holomorphic coordinate chart $(U,z)$. The determinant $\det(\theta_{j\bar{k}})$ transforms under a change of coordinates by the squared absolute value of a holomorphic Jacobian, so applying $\partial\bar\partial$ removes the coordinate-dependent factor. [definition: Chern Ricci Form] Let $(X,\theta)$ be a Hermitian complex manifold of dimension $n$. In a holomorphic coordinate chart $(U,z_1,\dots,z_n)$ with \begin{align*} \theta &= i\sum_{j,k=1}^n \theta_{j\bar{k}}\,dz_j\wedge d\bar{z}_k, \end{align*} the Chern Ricci form of $\theta$ is \begin{align*} \operatorname{Ric}(\theta)=-i\partial\bar\partial\log\det(\theta_{j\bar{k}}). \end{align*} [/definition] When $\theta$ is Kähler, this is the Ricci form associated with the Kähler curvature tensor. In a general Hermitian setting it is the trace of the Chern curvature of $T^{1,0}X$, and it is the convention used when relating curvature to holomorphic line bundles. The canonical bundle is the determinant of the holomorphic cotangent bundle. It is the natural receptacle for holomorphic volume forms, so its metric is induced by the volume density of a Hermitian metric. As in Section 9, the canonical bundle is \begin{align*} K_X=\Lambda^n(T^{*1,0}X), \end{align*} with local holomorphic frames $dz_1\wedge\cdots\wedge dz_n$ in holomorphic coordinate charts. A Hermitian metric on $T^{1,0}X$ induces a Hermitian metric on $K_X$ by taking the determinant metric on the cotangent bundle. This convention makes the next identity a precise sign statement, not just an analogy. [definition: Curvature Positivity For A Line Bundle] Let $(L,h)$ be a Hermitian holomorphic line bundle over a complex manifold $X$. The line bundle $(L,h)$ is positive at $p\in X$ if the real $(1,1)$-form $i\Theta_h(L)_p$ is positive definite on $T_p^{1,0}X$. It is semipositive at $p$ if $i\Theta_h(L)_p$ is semipositive, and negative at $p$ if $-i\Theta_h(L)_p$ is positive definite. [/definition] The same words are used globally when the condition holds at every point. The remaining question is how this sign convention applies to the determinant line $K_X$ and its dual; the metric induced from a Hermitian metric on $T^{1,0}X$ should turn Ricci curvature into their curvature. [quotetheorem:3891] [citeproof:3891] The determinant metric is the right choice because a local holomorphic volume form is measured exactly by the determinant of the cotangent metric; changing the metric on $T^{1,0}X$ changes this volume density, and the curvature of that density records the Ricci form. This identity is often summarized cohomologically as \begin{align*} \left[\frac{1}{2\pi}\operatorname{Ric}(\theta)\right]=c_1(X)=-c_1(K_X). \end{align*} The differential form depends on the metric, while the cohomology class is fixed by the complex structure. In the non-Kähler Hermitian setting this is specifically the Chern Ricci form; it need not agree with the Ricci form obtained from the Levi-Civita curvature tensor. The statement also does not assert an existence theorem: knowing the sign of $c_1(X)$ or $c_1(K_X)$ does not by itself construct a Kähler-Einstein metric. [example: Projective Space Has Positive Ricci Curvature] On the standard affine chart $z = (z_1,\dots,z_n)\in\mathbb{C}^n$ of $\mathbb{P}^n$, we show that $\operatorname{Ric}(\Omega_{\mathrm{FS}}) = (n+1)\Omega_{\mathrm{FS}}$ for the unnormalized Fubini-Study metric and deduce that $K_{\mathbb{P}^n}$ is a negative line bundle. Set $\rho = 1 + |z|^2$. Since $\partial_j \rho = \bar{z}_j$ and $\partial_{\bar{k}}\rho = z_k$, differentiating $\log\rho$ twice gives \begin{align*} g_{j\bar{k}} := \partial_j\partial_{\bar{k}}\log\rho = \partial_{\bar{k}}\!\left(\frac{\bar{z}_j}{\rho}\right) = \frac{\delta_{jk}}{\rho} - \frac{\bar{z}_j z_k}{\rho^2} = \frac{\rho\,\delta_{jk} - \bar{z}_j z_k}{\rho^2}, \end{align*} where the product rule gives $\partial_{\bar{k}}(\bar{z}_j/\rho) = (\partial_{\bar{k}}\bar{z}_j)/\rho + \bar{z}_j\cdot\partial_{\bar{k}}(\rho^{-1}) = \delta_{jk}/\rho - \bar{z}_j z_k/\rho^2$. The matrix $(g_{j\bar{k}})$ has the form $\rho^{-1}I - \rho^{-2}\bar{z}z^\top$, a rank-one update of the scalar matrix $\rho^{-1}I$. Applying the matrix determinant lemma $\det(A + uv^\top) = \det(A)(1 + v^\top A^{-1}u)$ with $A = \rho^{-1}I$, $u = -\rho^{-2}\bar{z}$, $v = z$, and noting $A^{-1} = \rho I$: \begin{align*} \det(g_{j\bar{k}}) = \rho^{-n}\!\left(1 + z^\top(\rho I)(-\rho^{-2}\bar{z})\right) = \rho^{-n}\!\left(1 - \frac{|z|^2}{\rho}\right) = \rho^{-n}\cdot\frac{\rho - |z|^2}{\rho} = \rho^{-(n+1)}, \end{align*} where $\rho - |z|^2 = (1 + |z|^2) - |z|^2 = 1$. Taking $-i\partial\bar\partial\log(\cdot)$ and using $\log\rho^{-(n+1)} = -(n+1)\log\rho$: \begin{align*} \operatorname{Ric}(\Omega_{\mathrm{FS}}) = -i\partial\bar\partial\log\det(g_{j\bar{k}}) = -i\partial\bar\partial\bigl(-(n+1)\log\rho\bigr) = (n+1)\,i\partial\bar\partial\log\rho = (n+1)\,\Omega_{\mathrm{FS}}. \end{align*} *Ricci Curvature And The Canonical Bundle* then gives $i\Theta_{h_{\Omega_{\mathrm{FS}}}}(K_{\mathbb{P}^n}) = -\operatorname{Ric}(\Omega_{\mathrm{FS}}) = -(n+1)\Omega_{\mathrm{FS}}$; since $\Omega_{\mathrm{FS}}$ is positive definite this form is negative definite, so $K_{\mathbb{P}^n}$ is negative. The factor $n+1$ is not coincidental: it matches the line bundle isomorphism $K_{\mathbb{P}^n}\cong\mathcal{O}_{\mathbb{P}^n}(-n-1)$, whose curvature with the metric induced from the hyperplane bundle equals $-(n+1)\Omega_{\mathrm{FS}}$. After dividing by $2\pi$, this is the first Chern class $-(n+1)[\omega_{\mathrm{FS}}]$. [/example] The sign is reversed for spaces locally modeled on the unit ball with its complex hyperbolic metric. Negative Ricci curvature on such a quotient provides a positive metric on the canonical bundle, a first analytic explanation for why ball quotients behave like varieties of general type. ## Holomorphic Sectional Curvature and Restrictions on Maps Which part of the curvature tensor is detected by a holomorphic curve? A curve sees the curvature of complex lines in the tangent bundle, not the full tensor at once. This leads to holomorphic sectional curvature, the curvature notion that appears in Schwarz-Pick and Ahlfors-Schwarz estimates. [definition: Holomorphic Sectional Curvature] Let $(Y,\omega)$ be a Kähler manifold with curvature tensor $R^\omega$. For $p\in Y$ and $\xi\in T_p^{1,0}Y\setminus\{0\}$, the holomorphic sectional curvature of $\omega$ in the direction $\xi$ is \begin{align*} H_\omega(p,\xi)=\frac{R^\omega(\xi,\bar{\xi},\xi,\bar{\xi})}{|\xi|_\omega^4}. \end{align*} [/definition] For a Riemann surface, holomorphic sectional curvature is the Gaussian curvature. In higher dimension it measures the Gaussian curvature of the infinitesimal complex line spanned by $\xi$. Schwarz-type arguments need a restriction formula: when a holomorphic curve maps into a Kahler target, the curvature seen on the curve should be controlled by the target's holomorphic sectional curvature. [quotetheorem:3892] [citeproof:3892] This theorem is the local geometric input behind the map restrictions. The Kähler hypothesis is what makes the Gauss equation take this clean form: the Chern and Levi-Civita viewpoints are compatible, and the second fundamental form contributes with the sign used above. For a general Hermitian target, torsion terms may enter, so ambient holomorphic sectional curvature alone no longer gives the same curvature comparison. The restriction to the open set where $df\ne 0$ is also essential, because at an isolated zero the pullback form is degenerate and does not define a Gaussian curvature there; the later metric inequalities are proved on the positive locus and then extend across such zeros by continuity. The global input is a maximum principle applied to conformal metrics on disks, which is precisely the role of the Ahlfors-Schwarz lemma. [quotetheorem:3893] [citeproof:3893] The Ahlfors-Schwarz lemma should be read as a metric contraction theorem. Negative curvature in the target forces holomorphic disks to contract relative to the Poincare metric, and larger negative lower bounds give stronger contraction. Its strength is that the conclusion is pointwise and does not require the disk map to be proper, injective, or globally controlled at the boundary. The curvature hypothesis is the analytic substitute for compactness: it prevents a holomorphic disk from carrying too much Euclidean area into a negatively curved target. In applications, one usually applies the lemma to the pullback of a Hermitian or Kahler metric along a holomorphic map, then compares it with the complete hyperbolic metric on the source disk. This is why the lemma is the natural passage from curvature negativity to restrictions on entire curves. [quotetheorem:3894] [citeproof:3894] This theorem is the basic analytic route from negative curvature to Brody-type hyperbolicity. Compactness of $Y$ is not needed for the local estimate, but compact negatively curved targets are the main geometric examples because their curvature bounds are uniform. [example: Hyperbolic Riemann Surfaces] Let $C$ be a compact Riemann surface of genus at least $2$, equipped with its Poincaré metric $\omega_C$ of constant Gaussian curvature $-1$. We show two things: every holomorphic map $g\colon\Delta\to C$ satisfies $g^*\omega_C\le\omega_\Delta$, and every holomorphic map $f\colon\mathbb{C}\to C$ is constant. *The metric on $C$.* By the [Uniformization Theorem](/theorems/3376), the universal cover of $C$ is biholomorphic to $\Delta$, and the covering projection $\pi\colon\Delta\to C$ is a local isometry. This realises $\omega_C$ as the unique complete metric of curvature $-1$ on $C$, descended from the Poincaré form $\omega_\Delta = 2i\,dz\wedge d\bar{z}/(1-|z|^2)^2$ on $\Delta$. *Disk maps.* The Gaussian curvature of $\omega_C$ satisfies $K_{\omega_C}=-1\le-\kappa$ for $\kappa=1$ at every point. The *Ahlfors Schwarz Lemma* (with target curvature bound $\kappa=1$) therefore applies to $g\colon\Delta\to C$ and yields \begin{align*} g^*\omega_C\le\frac{1}{\kappa}\,\omega_\Delta=\omega_\Delta. \end{align*} The factor $1/\kappa=1$ reflects that the target curvature exactly matches the Poincaré curvature; a more negatively curved target would give a strictly smaller upper bound. *Entire curves.* Restrict $f$ to the disk $\Delta_R=\{|z|<R\}$. The Poincaré form of curvature $-1$ on $\Delta_R$ is \begin{align*} \omega_{\Delta_R}=\frac{2iR^2\,dz\wedge d\bar{z}}{(R^2-|z|^2)^2}. \end{align*} Applying the *Ahlfors Schwarz Lemma* to $f|_{\Delta_R}\colon\Delta_R\to C$ (again with $\kappa=1$) gives $f^*\omega_C\le\omega_{\Delta_R}$ on $\Delta_R$. At $z=0$ the denominator $(R^2-|z|^2)^2$ equals $R^4$, so \begin{align*} \omega_{\Delta_R}\big|_{z=0}=\frac{2iR^2}{R^4}\,dz\wedge d\bar{z}=\frac{2i}{R^2}\,dz\wedge d\bar{z}. \end{align*} The pointwise inequality $(f^*\omega_C)_0\le(2/R^2)\cdot i\,dz\wedge d\bar{z}$ holds for every $R>0$. Sending $R\to\infty$ forces $(f^*\omega_C)_0=0$, and since $\omega_C$ is a positive metric on $C$ this forces $df_0=0$. For any $w\in\mathbb{C}$, the same argument applied to the translated map $z\mapsto f(w+z)$ on $\Delta_R$ gives $df_w=0$; since $\mathbb{C}$ is connected, $f$ is constant. The compact surface $C$ of genus $\ge 2$ is the model case of a negatively curved target: the curvature hypothesis needed by the *Ahlfors Schwarz Lemma* is met uniformly and globally, so both the metric-contraction estimate for disk maps and the rigidity of entire curves follow from the same bound. [/example] The Riemann surface example shows the one-dimensional model case. Ball quotients are the higher-dimensional analogue: the same disk estimates apply to every holomorphic curve in the quotient, while the Ricci-canonical identity also detects positivity of the canonical bundle when the quotient is compact. [example: Ball Quotients] Let $\mathbb{B}^n = \{z\in\mathbb{C}^n : |z|^2 < 1\}$ carry the complex hyperbolic metric $\omega_{\mathbb{B}^n} = -i\partial\bar\partial\log(1-|z|^2)$, and let $Y = \mathbb{B}^n/\Gamma$ be a smooth quotient by a torsion-free discrete group $\Gamma\subset\mathrm{Aut}(\mathbb{B}^n)$, equipped with the descended metric $\omega_Y$. We show: the holomorphic sectional curvature of $\omega_Y$ is the constant $-2$, so every holomorphic map $f\colon\mathbb{C}\to Y$ is constant; when $Y$ is compact, $K_Y$ is positive. *Metric matrix and Ricci form.* Since $\omega_{\mathbb{B}^n} = i\partial\bar\partial(-\log(1-|z|^2))$, the metric tensor is $g_{j\bar{k}} = \partial_j\partial_{\bar{k}}(-\log(1-|z|^2))$. Setting $s = 1-|z|^2$, the chain and product rules give \begin{align*} \partial_j(-\log s) = \frac{\bar{z}_j}{s}, \qquad g_{j\bar{k}} = \partial_{\bar{k}}\!\left(\frac{\bar{z}_j}{s}\right) = \frac{\delta_{jk}}{s} + \bar{z}_j\cdot\frac{-1}{s^2}\partial_{\bar{k}}s = \frac{s\,\delta_{jk} + \bar{z}_j z_k}{s^2}, \end{align*} where $\partial_{\bar{k}}s = -z_k$. Writing $g = s^{-1}I + s^{-2}\bar{z}z^T$, the matrix determinant lemma $\det(A+uv^T)=\det(A)(1+v^T A^{-1}u)$ with $A = s^{-1}I$, $u = s^{-2}\bar{z}$, $v = z$ yields $v^T A^{-1}u = z^T(sI)(s^{-2}\bar{z}) = |z|^2/s$, and so \begin{align*} \det(g_{j\bar{k}}) = s^{-n}\!\left(1 + \frac{|z|^2}{s}\right) = s^{-n}\cdot\frac{s+|z|^2}{s} = (1-|z|^2)^{-(n+1)}, \end{align*} where $s + |z|^2 = (1-|z|^2)+|z|^2 = 1$. Taking $-i\partial\bar\partial\log(\,\cdot\,)$ and using $\log\det(g_{j\bar{k}}) = -(n+1)\log(1-|z|^2)$: \begin{align*} \operatorname{Ric}(\omega_{\mathbb{B}^n}) = -i\partial\bar\partial\bigl(-(n+1)\log(1-|z|^2)\bigr) = (n+1)\,i\partial\bar\partial\log(1-|z|^2) = -(n+1)\,\omega_{\mathbb{B}^n}, \end{align*} the last equality because $i\partial\bar\partial\log(1-|z|^2) = -\omega_{\mathbb{B}^n}$ by definition. *Holomorphic sectional curvature.* The automorphism group $\mathrm{SU}(n,1)$ acts transitively on $\mathbb{B}^n$ by isometries of $\omega_{\mathbb{B}^n}$, so it suffices to compute the holomorphic sectional curvature at $z=0$. There $g_{j\bar{k}}(0) = \delta_{jk}$ and all Christoffel symbols vanish, so $R_{j\bar{k}l\bar{m}}|_{z=0} = -\partial_{\bar{k}}\partial_l g_{j\bar{m}}|_{z=0}$. Differentiating $g_{j\bar{m}} = s^{-1}\delta_{jm} + s^{-2}\bar{z}_j z_m$ and evaluating at $z=0$ (where $s=1$, $\bar{z}_j=0$, $z_m=0$): \begin{align*} \partial_l g_{j\bar{m}} = \frac{\bar{z}_l}{s^2}\delta_{jm} + \frac{\bar{z}_j}{s^2}\delta_{lm} + \frac{2\bar{z}_j z_m\bar{z}_l}{s^3}, \end{align*} so $\partial_{\bar{k}}\partial_l g_{j\bar{m}}|_{z=0} = \delta_{kl}\delta_{jm} + \delta_{kj}\delta_{lm}$ (the third term vanishes at $z=0$ because it contains both $\bar{z}_j$ and $\bar{z}_l$). Hence $R_{j\bar{k}l\bar{m}}|_{z=0} = -\delta_{kl}\delta_{jm} - \delta_{kj}\delta_{lm}$. For any nonzero $\xi\in T_0^{1,0}\mathbb{B}^n$, contracting gives \begin{align*} R(\xi,\bar\xi,\xi,\bar\xi)\big|_{z=0} = \bigl(-\delta_{kl}\delta_{jm} - \delta_{kj}\delta_{lm}\bigr)\xi^j\bar\xi^k\xi^l\bar\xi^m = -|\xi|^4 - |\xi|^4 = -2|\xi|^4, \end{align*} where each term contracts as $\delta_{kl}\delta_{jm}\xi^j\bar\xi^k\xi^l\bar\xi^m = \bigl(\sum_k\bar\xi^k\xi^k\bigr)\bigl(\sum_j\xi^j\bar\xi^j\bigr) = |\xi|^4$. Thus $H_{\omega_{\mathbb{B}^n}}(0,\xi) = -2$ for every direction; by the transitive isometry action, $H_{\omega_{\mathbb{B}^n}} = -2$ everywhere on $\mathbb{B}^n$. *Descent to the quotient.* Each $\gamma\in\Gamma$ is a holomorphic isometry of $\omega_{\mathbb{B}^n}$, so it preserves the curvature tensor pointwise: for any $p\in\mathbb{B}^n$ the derivative $d\gamma_p$ identifies tangent spaces and pulls back $R$ without change. The metric, Ricci form, and holomorphic sectional curvature are therefore $\Gamma$-invariant and descend unchanged to the quotient. The metric $\omega_Y$ on $Y = \mathbb{B}^n/\Gamma$ satisfies $\operatorname{Ric}(\omega_Y) = -(n+1)\omega_Y$ and $H_{\omega_Y} = -2$ at every point. *Entire curves.* Since $H_{\omega_Y}\le -2 < 0$ uniformly, the *Curvature Obstruction to Entire Curves* applies with $\kappa = 2$, and every holomorphic map $f\colon\mathbb{C}\to Y$ is constant. *Positivity of $K_Y$ when $Y$ is compact.* Since $\operatorname{Ric}(\omega_Y) = -(n+1)\omega_Y$ is negative definite, the *Ricci Curvature and the Canonical Bundle* theorem gives $i\Theta_{h_{\omega_Y}}(K_Y) = -\operatorname{Ric}(\omega_Y) = (n+1)\omega_Y$; as $(n+1)\omega_Y$ is positive definite, $K_Y$ is positive with respect to the induced metric $h_{\omega_Y}$. The factor $n+1$ records the complex dimension of the ball: in ambient dimension $n$ the ball quotient accumulates $n+1$ units of negative curvature into the canonical bundle, making the positivity of $K_Y$ strengthen as the dimension grows. [/example] ## Schwarz Lemma Methods in Complex Geometry How do Schwarz lemma arguments extend from disks to maps between higher-dimensional Kähler manifolds? The common strategy is to build a scalar quantity from the differential of a holomorphic map, derive a differential inequality using curvature, and apply a maximum principle. In one complex dimension that scalar quantity is a conformal factor; in higher dimension it is usually the trace of the pulled-back target metric. Let $f:(X,\omega_X)\to(Y,\omega_Y)$ be holomorphic. In one complex dimension, the pullback metric has a single conformal factor, so the curvature inequality can be applied directly to that scalar. In higher dimension there is no single preferred direction: $df$ may stretch different complex lines by different amounts, and controlling only the curvature of each target line separately does not control how the trace of all stretched directions evolves. The standard scalar replacement is the energy density \begin{align*} \nu=\operatorname{tr}_{\omega_X}(f^*\omega_Y). \end{align*} The inequality $f^*\omega_Y\le C\omega_X$ follows once $\nu$ is bounded, because the trace controls all eigenvalues of the semipositive form $f^*\omega_Y$ with respect to $\omega_X$. When the differential inequality for $\nu$ is computed, the curvature terms pair one source direction with another image direction. Thus the required target hypothesis is not merely curvature of a single complex line, but curvature of pairs of complex lines. This is the holomorphic bisectional curvature. [definition: Holomorphic Bisectional Curvature] Let $(Y,\omega)$ be a Kähler manifold with curvature tensor $R^\omega$. For $p\in Y$ and nonzero vectors $\xi,\eta\in T_p^{1,0}Y$, the holomorphic bisectional curvature in the directions $\xi$ and $\eta$ is \begin{align*} B_\omega(p;\xi,\eta)=\frac{R^\omega(\xi,\bar{\xi},\eta,\bar{\eta})}{|\xi|_\omega^2|\eta|_\omega^2}. \end{align*} [/definition] The holomorphic sectional curvature is the special case $\xi=\eta$. For maps between higher-dimensional manifolds, the estimate is applied to the trace of the pullback metric, so the Schwarz lemma needs curvature control for pairs of source and image directions rather than one complex line at a time. [quotetheorem:3895] [citeproof:3895] This theorem is used here as a higher-dimensional Schwarz lemma: it converts source Ricci lower bounds and target bisectional upper bounds into a direct metric contraction estimate. The estimate is powerful because it compares curvature signs on the source and target. A source with Ricci curvature not too negative cannot map with large derivative into a target whose bisectional curvature is uniformly negative. [example: Ricci Flat Sources And Negative Targets] Let $(X,\omega_X)$ be a compact Ricci-flat Kähler manifold and let $(Y,\omega_Y)$ be a Kähler manifold whose holomorphic bisectional curvature is bounded above by $-B$ for some $B>0$. We show that every holomorphic map $f\colon X\to Y$ is constant. The three hypotheses of the *Yau Schwarz Lemma* must be verified. First, $X$ is compact, so $(X,\omega_X)$ is complete. Second, Ricci-flatness means $\operatorname{Ric}(\omega_X)=0$, which satisfies the lower bound $\operatorname{Ric}(\omega_X)\ge -A\omega_X$ with $A=0$. Third, the bisectional curvature bound holds with $B>0$ by assumption. With these inputs the lemma yields \begin{align*} f^*\omega_Y \le \frac{A}{B}\,\omega_X = \frac{0}{B}\,\omega_X = 0. \end{align*} On the other hand, for any $p\in X$ and any nonzero $v\in T_p^{1,0}X$, the pullback form satisfies \begin{align*} (f^*\omega_Y)_p(v,\bar{v}) = \omega_Y\bigl(df_p(v),\,\overline{df_p(v)}\bigr) = |df_p(v)|^2_{\omega_Y} \ge 0, \end{align*} so $f^*\omega_Y\ge 0$ as a real $(1,1)$-form. Combined with $f^*\omega_Y\le 0$, this forces $f^*\omega_Y=0$ identically. For any $p$ and any $v$ we then have $|df_p(v)|^2_{\omega_Y}=0$, hence $df_p(v)=0$; since $v$ was arbitrary, $df_p=0$ at every $p\in X$. Because $X$ is connected, $f$ is constant. The argument shows that a flat Ricci curvature on the source is sufficient: the source contributes $A=0$ to the ratio $A/B$, collapsing the upper bound to zero without any additional geometric structure of $X$ beyond completeness and the Ricci condition. Compactness of $X$ enters only to guarantee completeness; the conclusion holds for any complete Ricci-flat source against a target whose bisectional curvature is uniformly negative. [/example] The disk version, the entire-curve obstruction, and Yau higher-dimensional estimate all have the same architecture. Curvature gives a differential inequality, the maximum principle bounds a scalar norm of $df$, and the resulting bound becomes a rigidity statement when the source has too much flatness or nonnegative Ricci curvature. [remark: Positivity And Hyperbolicity] Negative curvature often implies positivity of $K_Y$ and strong restrictions on holomorphic maps into $Y$, but these are not identical notions. Positivity of the canonical bundle is a cohomological condition represented by a curvature form of a line bundle. Hyperbolicity is a mapping property detected by holomorphic disks and entire curves. The Schwarz lemma methods in this section explain why strong negative curvature hypotheses imply both kinds of conclusion. [/remark] Section 11 used negative Ricci curvature and hyperbolicity to constrain geometry and mappings, while earlier sections showed that positive line bundle curvature enables projective embeddings. Section 12 closes the course by surveying how the curvature language—extended to singular metrics and multiplier ideal sheaves—unifies the analytic and algebraic viewpoints and outlines modern applications. # 12. Synthesis: From SCV to Complex Geometry This final section is intentionally panoramic. It records the analytic mechanism behind Kodaira embedding from Section 7, explains why positive curvature is the shared language of several complex variables and algebraic geometry, and states the singular extensions that lead to multiplier ideals and Nadel vanishing. ## Producing Global Sections from $\bar\partial$ Estimates How can a local holomorphic expression near a point become a global holomorphic section without losing its prescribed value or first derivative? The answer is to build a section that is holomorphic where its local data matter, allow a controlled error away from that region, and then solve a $\bar\partial$-equation to remove the error. Positivity enters because it gives an estimate strong enough to make the correction small, and singular weights enter because they force the correction to vanish to a required order at selected points. The analytic engine is the high-power $L^2$ estimate for positive line bundles. It converts curvature positivity into an inverse for $\bar\partial$ in positive degree. [quotetheorem:3896] [citeproof:3896] The hypotheses in this estimate are doing real work. The condition $q\ge 1$ is essential because the equation lowers degree: for $q=0$ there is no $(0,-1)$-form $u$, and even analogous estimates for functions must contend with holomorphic sections in the kernel of $\bar\partial$. Positivity, not mere semipositivity, supplies a uniform curvature lower bound growing like $m$; if the curvature has flat directions, the coercive term can disappear in those directions and the estimate with factor $1/m$ need not hold. Compactness gives the global Hilbert-space inverse and uniform constants, while the Kahler condition is what makes the Bochner--Kodaira identity have the clean curvature term needed for the estimate. This theorem is useful because it solves the error term created by a cut-off function. The cut-off is unavoidable: a local holomorphic expression cannot be pasted onto a compact manifold as a global holomorphic section, but it can be pasted as a smooth section whose failure to be holomorphic is supported in a controlled annulus. [example: Peak Section From A Cut-Off] Let $p \in X$, let $z = (z_1, \dots, z_n)$ be local holomorphic coordinates centred at $p$, and let $e$ be a local holomorphic frame for $L$ near $p$ with local weight $\varphi$ satisfying $|e|_h^2 = e^{-\varphi}$. Fix radii $0 < r_0 < r_1$ with $\{|z| < r_1\}$ inside the coordinate chart, and let $\chi$ be a smooth real-valued cut-off with $\chi \equiv 1$ on $\{|z| \leq r_0\}$ and $\chi \equiv 0$ on $\{|z| \geq r_1\}$. Extend $\tilde{s} = \chi e^{\otimes m}$ by zero to all of $X$; we show that for each solution $u$ of $\bar\partial u = \bar\partial\tilde{s}$ obtained from the weighted $L^2$ estimate below, the section $s = \tilde{s} - u$ is a global holomorphic section of $L^m$ that does not vanish at $p$. *The error term.* Because $e^{\otimes m}$ is a local holomorphic frame, $\bar\partial(e^{\otimes m}) = 0$, and the Leibniz rule gives \begin{align*} \bar\partial\tilde{s} &= (\bar\partial\chi)\otimes e^{\otimes m}. \end{align*} Since $\chi \equiv 1$ on $\{|z| \leq r_0\}$, the form $\bar\partial\chi$ vanishes identically near $p$; its support lies in the compact annulus $A = \{r_0 \leq |z| \leq r_1\}$, which is bounded away from $p$. *Weighted setup and estimate.* In the standard jet-separation argument one uses a quasi-plurisubharmonic auxiliary weight $\Psi$ with logarithmic pole at $p$, arranged so that it is bounded on the annulus supporting $\bar\partial\tilde{s}$ and so that the curvature lower bound remains valid after $m$ is chosen large. With this admissible weight the combined curvature satisfies \begin{align*} i\partial\bar\partial(m\varphi + \Psi) &= m\,i\partial\bar\partial\varphi + i\partial\bar\partial\Psi \geq m\omega, \end{align*} where $\omega = i\partial\bar\partial\varphi > 0$ is the curvature form of $h$. The point is not the particular cut-off formula for $\Psi$, but the standard existence of such weights in the Hörmander-Demailly estimate. By the *Positivity Estimate for High Tensor Powers* applied to this curvature lower bound, there is a section $u$ solving $\bar\partial u = \bar\partial\tilde{s}$ with \begin{align*} \|u\|^2_{m\varphi+\Psi} &\leq \frac{C}{m}\,\|\bar\partial\tilde{s}\|^2_{m\varphi+\Psi}. \end{align*} On the annulus $A$ supporting $\bar\partial\tilde{s}$, the weight $\Psi$ is identically zero, so \begin{align*} \|\bar\partial\tilde{s}\|^2_{m\varphi+\Psi} &= \int_A |\bar\partial\chi|^2 e^{-m\varphi}\,dV < \infty, \end{align*} and the right-hand side of the estimate is finite; the equation $\bar\partial u = \bar\partial\tilde{s}$ has the desired solution for every $m \geq m_0$. *Forcing $u(p) = 0$.* Because $\bar\partial u = \bar\partial\tilde{s} = 0$ on $\{|z| < r_0\}$, the section $u$ is holomorphic near $p$, so $u(p)$ is well-defined. The finiteness of $\|u\|^2_{m\varphi+\Psi}$ implies, near $p$, \begin{align*} \int_{\{|z|<r_0/2\}} |u(z)|^2\,|z|^{-2(n+\varepsilon)}\,dV &< \infty. \end{align*} Suppose $u(p) \neq 0$; by continuity there exist $c, \delta > 0$ with $|u(z)| \geq c$ on $\{|z| < \delta\}$, so the integral is bounded below by \begin{align*} c^2\int_{\{|z|<\delta\}} |z|^{-2(n+\varepsilon)}\,dV &= c^2\,\omega_{2n-1}\int_0^\delta r^{-2(n+\varepsilon)}\cdot r^{2n-1}\,dr = c^2\,\omega_{2n-1}\int_0^\delta r^{-2\varepsilon - 1}\,dr = +\infty, \end{align*} where $\omega_{2n-1}$ is the volume of $S^{2n-1}$, the factor $r^{2n-1}\,dr$ is the radial part of the volume element in $\mathbb{R}^{2n}$, and the resulting exponent $-2\varepsilon - 1 < -1$ makes the radial integral diverge at the origin. This contradicts the finiteness established above, so $u(p) = 0$. *Conclusion.* The section $s = \tilde{s} - u$ satisfies $\bar\partial s = \bar\partial\tilde{s} - \bar\partial u = 0$, so it is a global holomorphic section of $L^m$. At $p$, since $\chi(p) = 1$, \begin{align*} s(p) &= \tilde{s}(p) - u(p) = e^{\otimes m}\big|_p - 0 = e^{\otimes m}\big|_p \neq 0. \end{align*} Positivity supplies a genuine coercive reserve — the curvature lower bound $m\omega$ funds the $\bar\partial$ solution, while the auxiliary logarithmic weight forces that solution to belong to the multiplier ideal $\mathcal{I}(\Psi)_p \subset \mathfrak{m}_p$, leaving the prescribed value at $p$ uncorrected. [/example] The embedding criterion requires more than a section that does not vanish at a point; it requires sections whose first jets distinguish tangent directions. The same $L^2$ construction should therefore be run with a local model section whose leading term is linear, while the correction term is forced into a multiplier ideal small enough not to alter that prescribed jet. [quotetheorem:3897] [citeproof:3897] The theorem is the analytic heart of Kodaira embedding. Once sections separate points and tangent vectors, the projective map defined by all global sections is injective and immersive; compactness then upgrades it to an embedding. Its limitation is qualitative: it guarantees that some large $m$ works, but it does not identify the smallest such $m$ or give a numerical bound in terms of curvature, dimension, or intersection data. Effective very-ampleness theorems begin precisely where this argument stops, by asking how large the tensor power must be to separate prescribed jets. ## Positivity as the Bridge Between Analysis and Algebraic Geometry Why should a differential inequality for a curvature form decide whether a complex manifold can be placed inside projective space? The bridge is that curvature positivity has two faces. Analytically it says local weights are strictly plurisubharmonic, giving $\bar\partial$ estimates; algebraically it says high powers of a line bundle have enough sections to define projective coordinates. The curvature language begins with a Hermitian metric on a holomorphic line bundle. If $e$ is a local holomorphic frame and $|e|_h^2=e^{-\varphi}$, then the Chern curvature is represented locally by $\Theta_h(L)=\partial\bar\partial\varphi$, with positivity measured by the real $(1,1)$-form $i\Theta_h(L)$. As in Section 2, a holomorphic line bundle $L\to X$ is positive if it admits a smooth Hermitian metric $h$ such that $i\Theta_h(L)$ is a positive $(1,1)$-form on $X$. This definition is local in coordinates but global in consequence. In a local frame, positivity means that the weight $\varphi$ has positive complex Hessian, so it is a strictly plurisubharmonic function. That is precisely the kind of weight that powered the estimates in earlier SCV sections. [definition: Very Ample And Ample Line Bundle] Let $X$ be a compact complex manifold and let $L\to X$ be a holomorphic line bundle. The line bundle $L$ is very ample if the evaluation map associated to $H^0(X,L)$ defines a holomorphic embedding \begin{align*} \Phi_L &: X \longrightarrow \mathbb P(H^0(X,L)^*). \end{align*} The line bundle $L$ is ample if $L^m$ is very ample for some $m\in\mathbb N$. [/definition] The analytic and algebraic notions now have to be identified. Positivity gives the $L^2$ estimates that produce high-power sections, while very ampleness is the projective-coordinate condition those sections must eventually satisfy. The converse comes from projective space: the Fubini--Study metric on $\mathcal O_{\mathbb P^N}(1)$ has positive curvature, and pulling it back through an embedding gives a positive metric. [quotetheorem:3898] [citeproof:3898] The principle compresses the course into one statement: positivity is the analytic condition, ampleness is the algebro-geometric condition, and $\bar\partial$ estimates are the mechanism that proves they coincide. Compactness is part of the bridge: without it, global sections and projective maps no longer behave as a finite-dimensional embedding problem in the same way. The theorem also gives a striking consequence for general compact complex manifolds: the existence of a positive line bundle forces projectivity, hence forces the manifold to be Kahler. Positivity therefore buys more than an abstract ample class; it gives a metric representative whose curvature is the analytic input behind the projective embedding. [example: High-Power Embeddings Of Projective Space] On $\mathbb P^n$, the hyperplane bundle $\mathcal O_{\mathbb P^n}(1)$ carries the Fubini--Study metric $h_{FS}$, whose curvature form is the unnormalized Fubini-Study Kähler form $\Omega_{\mathrm{FS}}$. The tensor power $\mathcal O_{\mathbb P^n}(m) = \mathcal O_{\mathbb P^n}(1)^{\otimes m}$ inherits the metric $h_{FS}^{\otimes m}$ with curvature $m\,\Omega_{\mathrm{FS}}$, and its global holomorphic sections are identified with the homogeneous polynomials of degree $m$ in the homogeneous coordinates $Z_0,\ldots,Z_n$. [claim]The complete linear system $H^0(\mathbb P^n,\mathcal O_{\mathbb P^n}(m))$ defines a holomorphic embedding, the degree-$m$ Veronese embedding, \begin{align*} \nu_m : \mathbb P^n \longrightarrow \mathbb P^{\binom{n+m}{m}-1}. \end{align*}[/claim] The sections spanning $H^0(\mathbb P^n,\mathcal O_{\mathbb P^n}(m))$ are the monomials $Z^\alpha = Z_0^{\alpha_0}\cdots Z_n^{\alpha_n}$ indexed by multi-indices $\alpha \in \mathbb Z_{\ge 0}^{n+1}$ with $|\alpha| = m$. The number of such multi-indices equals the number of ways to write $m$ as an ordered sum of $n+1$ non-negative integers, which by the stars-and-bars formula (insert $n$ dividers among $m+n$ positions) is $\binom{m+n}{n} = \binom{n+m}{m}$. Label these monomials $s_0,\ldots,s_N$ with $N = \binom{n+m}{m}-1$. Since each $s_i$ is homogeneous of degree $m$, scaling $Z$ by a nonzero scalar multiplies every $s_i(Z)$ by the same factor $\lambda^m$, so the assignment \begin{align*} \nu_m([Z_0:\cdots:Z_n]) &= [s_0(Z):\cdots:s_N(Z)] \end{align*} is a well-defined holomorphic map into $\mathbb P^N$. To verify injectivity, fix the standard affine chart $U_k = \{Z_k \neq 0\}$ with affine coordinates $z_j = Z_j/Z_k$ for $j \neq k$. Among the monomials of degree $m$, those of the form $Z_k^{m-1}Z_j$ contribute the Veronese coordinate \begin{align*} \frac{Z_k^{m-1}Z_j}{Z_k^m} = \frac{Z_j}{Z_k} = z_j, \end{align*} where the first equality cancels one factor of $Z_k$ and the second is the definition of $z_j$. Thus the affine coordinate $z_j$ appears literally as a (normalized) component of $\nu_m$ on $U_k$. Now suppose $\nu_m([Z]) = \nu_m([W])$ for two points in $U_k$. Then every degree-$m$ monomial in $Z$ and $W$ is proportional by the same nonzero scalar, so in particular $Z_k^{m-1}Z_j$ and $W_k^{m-1}W_j$ are proportional by the same factor for every $j \neq k$. Dividing both by the proportional factor $Z_k^m \sim W_k^m$ gives $z_j([Z]) = z_j([W])$ for every $j \neq k$. Since the affine coordinates $\{z_j : j \neq k\}$ separate points of $U_k$, we conclude $[Z] = [W]$ in $U_k$. As $U_0,\ldots,U_n$ cover $\mathbb P^n$, the map $\nu_m$ is injective. To verify that $\nu_m$ is an immersion, observe that on $U_k$ the component functions of $\nu_m$ include $z_j = Z_k^{m-1}Z_j/Z_k^m$ for every $j \neq k$. The holomorphic differentials $dz_1,\ldots,\widehat{dz_k},\ldots,dz_n$ form a basis of the holomorphic cotangent space at every point of $U_k$, and each $dz_j$ is the pullback of the corresponding Veronese coordinate differential. Hence the pullback of the cotangent map of $\nu_m$ is surjective at every point of $U_k$, meaning $d\nu_m$ is injective everywhere on $U_k$. Covering by all $U_k$ gives injectivity of the differential on all of $\mathbb P^n$. An injective holomorphic immersion of a compact complex manifold is a closed map, hence a proper map, hence a holomorphic embedding onto its image. The Fubini--Study curvature $\Omega_{\mathrm{FS}}$ is the analytic resource: it is strictly positive, so the Kodaira Positivity-Ampleness Principle applies, and the monomials of degree $m$ are its sections. In the general $\bar\partial$-correction construction, the role of $z_j = Z_k^{m-1}Z_j/Z_k^m$ is played by the peak sections whose values at prescribed points are protected from the correction term $u$; here the affine coordinates are literally among the Veronese coordinates, making the separation argument purely explicit with no correction needed. [/example] ## Singular Metrics and Multiplier Ideals What changes when the metric is allowed to have poles or logarithmic singularities? The analytic estimates survive in a refined form, but the solutions are forced to vanish along the singular set. The bookkeeping device for that forced vanishing is the multiplier ideal sheaf. A singular Hermitian metric replaces smooth local weights by plurisubharmonic weights. This is natural from SCV because plurisubharmonic functions are exactly the local potentials whose complex Hessians define positive currents. [definition: Singular Hermitian Metric On A Line Bundle] Let $L\to X$ be a holomorphic line bundle with local holomorphic frames $e_j$ on open sets $U_j$ and transition functions $e_i=g_{ij}e_j$. A singular Hermitian metric $h$ on $L$ is given by functions $\varphi_j\in L^1_{\mathrm{loc}}(U_j)$ such that \begin{align*} |e_j|_h^2 &= e^{-\varphi_j}, & \varphi_i &= \varphi_j-\log |g_{ij}|^2 \quad \text{on } U_i\cap U_j. \end{align*} The metric is positive if each $\varphi_j$ is plurisubharmonic. [/definition] The curvature of a positive singular metric is a positive current. Thus divisors, base loci, and degenerating metrics can be inserted into the same curvature formalism used for smooth positive line bundles. [example: Logarithmic Weight Along A Divisor] Let $D \subset X$ be a divisor locally defined by a holomorphic function $f$, let $\psi$ be a smooth strictly plurisubharmonic function on a coordinate chart, and fix $c > 0$. We show that the weight \begin{align*} \varphi &= \psi + c\log|f|^2 \end{align*} defines a positive singular Hermitian metric and compute its curvature current explicitly. *Plurisubharmonicity and the singularity.* Since $f$ is holomorphic, $\log|f|$ is plurisubharmonic by [Log-Modulus of a Holomorphic Function is PSH](/theorems/3405), so $c\log|f|^2 = 2c\log|f|$ is plurisubharmonic as a positive scalar multiple of a plurisubharmonic function. Adding the smooth strictly plurisubharmonic function $\psi$ then preserves plurisubharmonicity by [Stability Properties of PSH Functions](/theorems/3404), so $\varphi$ is plurisubharmonic and lies in $L^1_\mathrm{loc}$ as the definition of a singular Hermitian metric requires. Along $D$ where $f = 0$, the summand $c\log|f|^2$ is $-\infty$, so $\varphi$ is not locally bounded there; equivalently, the local norm $|e|_h^2 = e^{-\varphi} = e^{-\psi}|f|^{-2c}$ has a pole of order $2c$ along $D$, making the metric genuinely singular rather than merely degenerate. *Curvature current.* By definition, the curvature of $h = e^{-\varphi}$ is the $(1,1)$-current $i\Theta_h = i\partial\bar\partial\varphi$. By linearity of $\partial\bar\partial$, \begin{align*} i\partial\bar\partial\varphi &= i\partial\bar\partial\psi + c \cdot i\partial\bar\partial\log|f|^2. \end{align*} The first term is already a smooth positive $(1,1)$-form. For the second, the *Poincaré–Lelong formula* gives, for any holomorphic function $f$, \begin{align*} i\partial\bar\partial\log|f|^2 &= 2\pi[D], \end{align*} where $[D]$ denotes the current of integration on the zero locus of $f$ counted with multiplicity. (The factor $2\pi$ arises from $\log|f|^2 = 2\log|f|$ together with the normalisation $\frac{i}{\pi}\partial\bar\partial\log|f| = [D]$ of the classical Lelong formula.) Substituting back, \begin{align*} i\Theta_h &= i\partial\bar\partial\psi + 2\pi c[D]. \end{align*} Because $\psi$ is strictly plurisubharmonic, $i\partial\bar\partial\psi$ is a smooth strictly positive $(1,1)$-form. The current $2\pi c[D]$ is positive and supported on $D$. Their sum is therefore a strictly positive current, confirming that $h$ is a positive singular Hermitian metric in the sense of the preceding definition. The weight $\varphi = \psi + c\log|f|^2$ is the canonical local model for a singular positive metric: the exponent $c$ controls the severity of the pole along $D$, the smooth strictly plurisubharmonic term $\psi$ supplies the background curvature reserve that $\bar\partial$ estimates require, and the Lelong mass $2\pi c[D]$ is the analytic incarnation of the divisor $D$ as an algebraic datum. [/example] To use singular metrics in cohomology, the local integrability condition cannot remain a pointwise side calculation. It must be organized into a coherent sheaf recording exactly which holomorphic germs vanish fast enough to compensate for the pole of $e^{-\varphi}$. [definition: Multiplier Ideal Sheaf] Let $X$ be a complex manifold and let $\varphi\in L^1_{\mathrm{loc}}(X)$ be plurisubharmonic. The multiplier ideal sheaf $\mathcal I(\varphi)$ is the sheaf whose stalk at $x\in X$ is \begin{align*} \mathcal I(\varphi)_x = \{ f\in \mathcal O_{X,x} : |f|^2e^{-\varphi} \text{ is locally integrable near } x\}. \end{align*} For a singular Hermitian metric $h=e^{-\varphi}$ on a line bundle, write $\mathcal I(h)=\mathcal I(\varphi)$ locally. [/definition] Multiplier ideals are the algebraic shadow of analytic integrability. They turn the size of a pole into an ideal of allowed holomorphic functions. [example: Multiplier Ideal At An Isolated Logarithmic Pole] Work near $0 \in \mathbb{C}^n$ with weight $\varphi(z) = c\log|z|^2$ for a fixed $c > 0$, and let $\mathfrak{m}_0 \subset \mathcal{O}_{\mathbb{C}^n,0}$ be the maximal ideal of germs vanishing at $0$. [claim]A germ $f \in \mathcal{O}_{\mathbb{C}^n,0}$ of vanishing order $\ell$ at $0$ belongs to $\mathcal{I}(\varphi)_0$ if and only if $\ell > c - n$. Consequently, \begin{align*} \mathcal{I}(\varphi)_0 &= \mathfrak{m}_0^{\max\{0,\,\lfloor c-n\rfloor+1\}}. \end{align*}[/claim] First simplify the integrand's weight factor. Since $\log|z|^2$ is the natural logarithm of $|z|^2$, \begin{align*} e^{-\varphi(z)} &= e^{-c\log|z|^2} = |z|^{-2c}. \end{align*} By the definition of the multiplier ideal sheaf, $f \in \mathcal{I}(\varphi)_0$ if and only if $|f(z)|^2|z|^{-2c}$ is locally integrable near $0$, that is, \begin{align*} \int_{B_\delta} |f(z)|^2\,|z|^{-2c}\,dV < \infty \end{align*} for some $\delta > 0$. The task is to show this finiteness holds exactly when $\ell > c - n$. *Sufficiency: $\ell > c - n$ implies integrability.* Because $f$ has vanishing order $\ell$, it belongs to $\mathfrak{m}_0^\ell$, so every monomial in its Taylor series at $0$ has total degree at least $\ell$. On a ball $B_\delta$ this gives the pointwise bound $|f(z)| \leq C|z|^\ell$ for some constant $C > 0$. Identify $\mathbb{C}^n$ with $\mathbb{R}^{2n}$ and introduce polar coordinates $z = r\zeta$ with $r = |z| \in [0,\delta)$ and $\zeta \in S^{2n-1}$; the volume element becomes $dV = r^{2n-1}\,dr\,d\sigma(\zeta)$, where $d\sigma$ is the surface measure on $S^{2n-1}$ with total mass $\omega_{2n-1} = \mathrm{Vol}(S^{2n-1})$. Applying the pointwise bound and integrating first over the sphere via [Fubini Theorem](/theorems/513), \begin{align*} \int_{B_\delta} |f(z)|^2\,|z|^{-2c}\,dV &\leq C^2 \int_{B_\delta} |z|^{2\ell}\,|z|^{-2c}\,dV = C^2\,\omega_{2n-1}\int_0^\delta r^{2\ell-2c}\cdot r^{2n-1}\,dr = C^2\,\omega_{2n-1}\int_0^\delta r^{2\ell-2c+2n-1}\,dr. \end{align*} The standard criterion $\int_0^\delta r^\alpha\,dr < \infty$ holds if and only if $\alpha > -1$. The exponent here is $\alpha = 2\ell - 2c + 2n - 1$, and $\alpha > -1$ simplifies to $2\ell - 2c + 2n > 0$, which is exactly $\ell > c - n$. Under this assumption the integral on the right is finite, so $f \in \mathcal{I}(\varphi)_0$. *Necessity: $\ell \leq c - n$ implies non-integrability.* Now suppose $f$ has exact vanishing order $\ell$, meaning its Taylor expansion at $0$ contains a nonzero homogeneous polynomial $P$ of degree $\ell$ and no lower-degree terms. Since $P$ is a continuous function on the compact sphere $S^{2n-1}$ that is not identically zero, there exist an open set $U \subset S^{2n-1}$ and a constant $c' > 0$ such that $|P(\zeta)| \geq c'$ for all $\zeta \in U$. For small $r$ the Taylor expansion gives $f(r\zeta) = r^\ell P(\zeta) + O(r^{\ell+1})$, so for $r < \delta'$ sufficiently small, \begin{align*} |f(r\zeta)| \geq \tfrac{c'}{2}\,r^\ell \qquad \text{for all } \zeta \in U. \end{align*} Integrating only over the cone $\{r\zeta : r \in (0,\delta'),\,\zeta \in U\}$ and applying [Fubini Theorem](/theorems/513), \begin{align*} \int_{B_{\delta'}} |f(z)|^2\,|z|^{-2c}\,dV &\geq \int_0^{\delta'}\!\int_U \frac{(c')^2}{4}\,r^{2\ell}\cdot r^{-2c}\cdot r^{2n-1}\,d\sigma(\zeta)\,dr = \frac{(c')^2}{4}\,\omega_{2n-1}(U)\int_0^{\delta'} r^{2\ell-2c+2n-1}\,dr, \end{align*} where $\omega_{2n-1}(U) = \int_U d\sigma > 0$. When $\ell \leq c - n$, the exponent $2\ell - 2c + 2n - 1 \leq -1$, so $\int_0^{\delta'} r^{2\ell-2c+2n-1}\,dr = +\infty$, and $f \notin \mathcal{I}(\varphi)_0$. *Identifying the ideal.* The two directions together say that $f \in \mathcal{I}(\varphi)_0$ if and only if $\mathrm{ord}_0(f) > c - n$. For an integer vanishing order, the condition $\mathrm{ord}_0(f) > c - n$ is equivalent to $\mathrm{ord}_0(f) \geq \lfloor c - n \rfloor + 1$ when $c > n$ (whether or not $c - n$ is an integer, $\lfloor c-n\rfloor + 1$ is the smallest integer strictly exceeding $c - n$), and to $\mathrm{ord}_0(f) \geq 0$ when $c \leq n$. The set of germs with vanishing order at least $k$ is exactly $\mathfrak{m}_0^k$ (with $\mathfrak{m}_0^0 = \mathcal{O}_{\mathbb{C}^n,0}$ by convention), so \begin{align*} \mathcal{I}(\varphi)_0 &= \mathfrak{m}_0^{\max\{0,\,\lfloor c-n\rfloor+1\}}. \end{align*} The threshold $c = n$ is the critical value: for $c < n$ the singularity $|z|^{-2c}$ is integrable near $0$ with no constraint on $f$, so the ideal is the full stalk $\mathcal{O}_{\mathbb{C}^n,0}$; for each additional unit of $c$ beyond $n$, the ideal climbs one power of $\mathfrak{m}_0$, demanding one higher order of vanishing from any holomorphic section permitted to use that singular weight. [/example] This calculation is the local picture behind ideal sheaves from multiplier ideals. Stronger singularities impose higher vanishing, and milder singularities impose no vanishing at all. [example: When Positivity Or Singularity Imposes No Constraint] Two failure modes together explain why each hypothesis in the surrounding theorems is doing genuine work. *Without positivity.* On a compact Kähler manifold with no positive holomorphic line bundle, such as a generic complex torus of dimension at least two, fix any smooth Hermitian metric $h$ on a holomorphic line bundle $L$; such a metric exists by a partition-of-unity argument regardless of whether $L$ is positive. If the local weight of $h$ in a holomorphic frame $e$ with $|e|_h^2 = e^{-\varphi}$ is $\varphi$, then the frame $e^{\otimes m}$ for $L^m$ satisfies $|e^{\otimes m}|_{h^m}^2 = e^{-m\varphi}$, so the local weight of $h^m$ is $m\varphi$ and \begin{align*} \Theta_{h^m}(L^m) &= \partial\bar\partial(m\varphi) = m\,\partial\bar\partial\varphi = m\,\Theta_h(L). \end{align*} The *Bochner–Kodaira identity* for $L^m$-valued $(0,q)$-forms $\alpha$ on a compact Kähler manifold $(X,\omega)$, after substituting this curvature formula, reads \begin{align*} \|\bar\partial\alpha\|^2 + \|\bar\partial^*\alpha\|^2 &= \|\nabla\alpha\|^2 + m\langle [i\Theta_h(L),\Lambda_\omega]\alpha,\alpha\rangle. \end{align*} In local coordinates where $\omega = \sum_j i\,dz^j\wedge d\bar z^j$ and $i\Theta_h(L) = \sum_j \lambda_j\,i\,dz^j\wedge d\bar z^j$ are simultaneously diagonalised, the commutator $[i\Theta_h(L),\Lambda_\omega]$ acts on a $(0,1)$-form $\alpha_{\bar k}\,d\bar z^k\otimes e$ by multiplication by $\lambda_k$: that is, $[i\Theta_h(L),\Lambda_\omega](\alpha_{\bar k}\,d\bar z^k\otimes e) = \lambda_k\,\alpha_{\bar k}\,d\bar z^k\otimes e$. If some $\lambda_k \leq 0$ at a point $x_0\in X$, then a $(0,1)$-form $\alpha$ supported near $x_0$ and pointing in the $\bar z^k$-direction satisfies \begin{align*} m\langle [i\Theta_h(L),\Lambda_\omega]\alpha,\alpha\rangle &= m\lambda_k\|\alpha\|^2 \leq 0, \end{align*} so the identity yields only $\|\bar\partial\alpha\|^2 + \|\bar\partial^*\alpha\|^2 \geq 0$, a bound carrying no dependence on $m$. Since $X$ carries no positive line bundle, no choice of line bundle $L$ and metric $h$ makes every $\lambda_k$ positive at every point of $X$; hence no uniform $\varepsilon > 0$ satisfies $\|\bar\partial\alpha\|^2 + \|\bar\partial^*\alpha\|^2 \geq m\varepsilon\|\alpha\|^2$ for all $(0,q)$-forms $\alpha$. Without this coercive lower bound growing linearly in $m$, the Hodge-theoretic construction of a correction $u$ with $\|u\|^2\leq (C/m)\|\bar\partial\tilde s\|^2$ has no foundation, and the cut-off-and-correct method cannot produce peak sections. *Without a genuine pole.* Work near $0\in\mathbb{C}^n$ and suppose $\varphi$ is locally bounded: there exist $\delta > 0$ and a constant $C \geq 0$ such that $\varphi(z) \geq -C$ for all $z\in B_\delta$. Rearranging, $-\varphi(z) \leq C$, and exponentiating gives \begin{align*} e^{-\varphi(z)} &\leq e^C \qquad \text{for all } z\in B_\delta. \end{align*} Let $f\in\mathcal{O}_{\mathbb{C}^n,0}$ be any holomorphic germ; since $f$ is holomorphic it is continuous, so it is bounded on the compact set $\overline{B}_{\delta/2}$: pick $M_f > 0$ with $|f(z)| \leq M_f$ for all $z\in\overline{B}_{\delta/2}$. Multiplying the two pointwise bounds, \begin{align*} |f(z)|^2\,e^{-\varphi(z)} &\leq M_f^2\,e^C \qquad \text{for all }z\in B_{\delta/2}. \end{align*} Integrating over the ball of finite volume, \begin{align*} \int_{B_{\delta/2}} |f(z)|^2\,e^{-\varphi(z)}\,dV &\leq M_f^2\,e^C\cdot\mathrm{Vol}(B_{\delta/2}) < \infty. \end{align*} By the definition of the multiplier ideal sheaf, $f\in\mathcal{I}(\varphi)_0$; since $f\in\mathcal{O}_{\mathbb{C}^n,0}$ was arbitrary, $\mathcal{I}(\varphi)_0 = \mathcal{O}_{\mathbb{C}^n,0}$. The contrast with the weight $\varphi = c\log|z|^2$ of the preceding example is exact: there $e^{-\varphi} = |z|^{-2c}$ is unbounded near $0$, the step $e^{-\varphi} \leq \mathrm{const}$ fails, and for $c \geq n$ the multiplier ideal is the strict subsheaf $\mathfrak{m}_0^{\lfloor c-n\rfloor+1} \subsetneq \mathcal{O}_{\mathbb{C}^n,0}$—the pole is the sole source of the vanishing constraint, and without it there is nothing for the ideal to encode. [/example] ## Nadel Vanishing and Modern Embedding Theorems How far can the smooth Kodaira picture be pushed when curvature is positive only as a current? Nadel vanishing is the basic answer: positive singular metrics still give cohomology vanishing, but the multiplier ideal must be inserted to record the singularities. [quotetheorem:3718] [citeproof:3718] Nadel vanishing should be read as Kodaira vanishing with singularities included. When $h$ is smooth, $\mathcal I(h)=\mathcal O_X$, so the statement reduces to the familiar vanishing theorem for positive line bundles. When $h$ has poles, the same curvature positivity remains available, but the cohomology group is twisted by the ideal sheaf that measures the poles. [example: Ideal Sheaf In A Vanishing Statement] Let $(X,\omega)$ be a compact Kähler manifold of complex dimension $n$, let $L \to X$ be a holomorphic line bundle, and let $h$ be a globally defined singular Hermitian metric on $L$ satisfying the curvature-current inequality \begin{align*} i\Theta_h(L) \geq \varepsilon\omega \end{align*} for some $\varepsilon>0$. Assume that, in a holomorphic frame near $p$, a local weight of $h$ has the form \begin{align*} \varphi &= \psi + c\log|z - p|^2, \end{align*} where $z$ is a holomorphic coordinate centred at $p$, $\psi$ is a smooth function satisfying $i\partial\bar\partial\psi \geq \varepsilon\omega$ for some $\varepsilon > 0$, and $c \geq n$. [claim]For all $q \geq 1$, \begin{align*} H^q(X,\, K_X \otimes L \otimes \mathcal{I}(h)) &= 0, \end{align*} and $\mathcal{I}(h)_p \subset \mathfrak{m}_p$, so every germ in the stalk $\mathcal{I}(h)_p$ vanishes at $p$.[/claim] *Curvature.* The curvature current of $h = e^{-\varphi}$ is $i\Theta_h(L) = i\partial\bar\partial\varphi$ in the local frame. By the global hypothesis on $h$, this current satisfies the positivity assumption required by the *Nadel Vanishing Theorem*: \begin{align*} i\Theta_h(L) \geq \varepsilon\omega \end{align*} The logarithmic pole is used here to impose a multiplier-ideal condition at the point; unlike a divisor pole, it should not be read as a Poincare-Lelong divisor current. Under the displayed curvature hypothesis, Nadel vanishing gives $H^q(X, K_X \otimes L \otimes \mathcal{I}(h)) = 0$ for every $q \geq 1$. *Multiplier ideal at $p$.* Since $\psi$ is smooth, $e^{-\psi}$ is bounded above and below by positive constants on a neighbourhood of $p$: say $A^{-1} \leq e^{-\psi} \leq A$ on $B_\delta(p)$. For any germ $f \in \mathcal{O}_{X,p}$, \begin{align*} |f|^2 e^{-\varphi} &= |f|^2 \, e^{-\psi} \cdot |z - p|^{-2c}, \end{align*} so two-sided boundedness of $e^{-\psi}$ gives \begin{align*} A^{-1}|f|^2|z-p|^{-2c} \;\leq\; |f|^2 e^{-\varphi} \;\leq\; A|f|^2|z-p|^{-2c} \end{align*} on $B_\delta(p)$. Integrability of $|f|^2 e^{-\varphi}$ near $p$ is therefore equivalent to integrability of $|f|^2|z-p|^{-2c}$ near $p$. By the computation in the preceding example (weight $c\log|z-p|^2$, ambient dimension $n$), $|f|^2|z-p|^{-2c}$ is locally integrable near $p$ if and only if the vanishing order of $f$ at $p$ exceeds $c - n$. Since $c \geq n$, the threshold $c - n \geq 0$, and the smallest integer strictly exceeding $c - n$ is $\lfloor c - n\rfloor + 1 \geq 1$. Therefore, \begin{align*} \mathcal{I}(h)_p &= \mathfrak{m}_p^{\lfloor c - n\rfloor + 1} \subset \mathfrak{m}_p, \end{align*} and every germ in $\mathcal{I}(h)_p$ vanishes at $p$. The logarithmic pole plays two roles at once, but only one of them is divisor-like. Analytically, the chosen quasi-plurisubharmonic weight is arranged so that its curvature current satisfies the displayed positivity inequality; algebraically, the same exponent $c$ controls the local integrability threshold that forces $\mathcal{I}(h)_p \subset \mathfrak{m}_p$. Because a point is not a divisor in complex dimension greater than one, this should be read as a multiplier-ideal mechanism rather than as a Poincare-Lelong divisor-current computation. The vanishing theorem removes higher cohomological obstructions to imposing local conditions, and the ideal encodes what that local condition is: a section of $K_X \otimes L \otimes \mathcal{I}(h)$ must vanish at $p$. [/example] Beyond the smooth Kodaira embedding theorem proved in Section 7, modern embedding theorems refine this philosophy in several directions. Demailly holomorphic Morse inequalities estimate the asymptotic size of cohomology groups from curvature eigenvalues; Ohsawa--Takegoshi extension turns $L^2$ estimates into extension of sections from subvarieties; and effective very-ampleness results seek numerical tensor powers that separate prescribed jets. In all of these extensions, the same three ingredients reappear: a curvature lower bound, a $\bar\partial$ estimate, and a sheaf encoding the local singularity or vanishing condition. [remark: Why Multiplier Ideals Belong In Embedding Problems] Embedding problems ask for sections with controlled jets. Singular weights with logarithmic poles force $L^2$ solutions to vanish to specified order, while multiplier ideals record that forced vanishing algebraically. Thus the analytic correction step and the algebraic condition on jets are two descriptions of the same constraint. [/remark] ## Kodaira-Type Philosophy for Positive Curvature What is the common pattern behind Kodaira embedding, Kodaira vanishing, Nadel vanishing, and high-power projective embeddings? Each result begins with positivity and ends with global algebraic information. The intermediate step is always an analytic estimate for $\bar\partial$. The smooth version has the following schematic form. A positive metric gives strictly plurisubharmonic local weights. Those weights give coercive Bochner--Kodaira estimates. The estimates solve $\bar\partial$ with control. Controlled solutions correct local objects into global holomorphic sections. Enough global sections then produce embeddings, vanishing, and projective models. The singular version keeps the same architecture but replaces smooth weights by plurisubharmonic weights and replaces the structure sheaf by a multiplier ideal. The curvature inequality is interpreted in the sense of currents, and the resulting global sections satisfy the vanishing constraints imposed by the singularities. [explanation: The Course In One Analytic Diagram] Start with a holomorphic line bundle $L\to X$ and a Hermitian metric $h$. The local weight $\varphi$ defines curvature by \begin{align*} \Theta_h(L) &= \partial\bar\partial\varphi. \end{align*} If $i\Theta_h(L)>0$, then $\varphi$ is strictly plurisubharmonic in local frames, so SCV supplies $L^2$ estimates for $\bar\partial$ with strong weights. These estimates solve correction equations of the form \begin{align*} \bar\partial u &= \bar\partial\tilde{s}, & s&=\tilde{s}-u, \end{align*} turning approximate sections $\tilde{s}$ into holomorphic sections $s$. By choosing $\tilde{s}$ and the auxiliary weights carefully, the resulting sections separate points, tangent vectors, or higher jets. The projective map defined by these sections is the geometric output of the analysis. [/explanation] This diagram also explains why the subject belongs simultaneously to several complex variables and complex algebraic geometry. The local input is plurisubharmonicity, $L^2$ integrability, and the $\bar\partial$ operator; the global output is ampleness, vanishing, projective embedding, and control of sheaf cohomology. [example: From Curvature To A Projective Model] Let $X$ be a compact complex manifold, let $L \to X$ be a positive holomorphic line bundle, and let $m$ be large enough that $H^0(X, L^m)$ separates points and tangent vectors on $X$ — such $m$ exists by the *Jet Separation From $\bar\partial$ Estimates* theorem proved above. Fix a basis $s_0, \dots, s_N$ of $H^0(X, L^m)$, so that $N + 1 = \dim H^0(X, L^m)$. [claim]The map \begin{align*} \Phi_m(x) &= [s_0(x) : \cdots : s_N(x)] \end{align*} is a well-defined holomorphic embedding of $X$ into $\mathbb{P}^N$.[/claim] *Well-definedness.* Each $s_j \in H^0(X, L^m)$ is a global holomorphic section; its value at $x$ lies in the one-dimensional fibre $(L^m)_x$. Choose a local holomorphic frame $e^{\otimes m}$ for $L^m$ near $x$ and write $s_j = f_j \, e^{\otimes m}$ for holomorphic functions $f_j$. The tuple $(f_0(x), \dots, f_N(x))$ is nonzero: since sections separate points, the peak section construction applied at $x$ yields some index $k$ with $s_k(x) \neq 0$, so $f_k(x) \neq 0$. If the local frame changes to $(e')^{\otimes m} = g^m e^{\otimes m}$ for a nowhere-vanishing holomorphic transition function $g$, then $s_j = g^{-m}f_j \cdot (e')^{\otimes m}$, so every local representative transforms by the common factor $g^{-m}(x) \neq 0$. In projective space this factor cancels, so $[f_0(x) : \cdots : f_N(x)]$ is independent of the frame choice and $\Phi_m(x)$ is a well-defined point of $\mathbb{P}^N$. In local frames the coordinate functions $f_j$ are holomorphic, so $\Phi_m$ is holomorphic. *Injectivity.* Let $x, y \in X$ with $\Phi_m(x) = \Phi_m(y)$. In a local frame this means $(f_0(x), \dots, f_N(x))$ and $(f_0(y), \dots, f_N(y))$ are proportional: there exists a nonzero scalar $\lambda$ with $f_j(x) = \lambda f_j(y)$ for every $j$. Suppose for contradiction that $x \neq y$. Point separation gives a section $s_k$ with $s_k(x) \neq 0$ and $s_k(y) = 0$, that is, $f_k(x) \neq 0$ and $f_k(y) = 0$. Then $f_k(x) = \lambda f_k(y) = \lambda \cdot 0 = 0$, contradicting $f_k(x) \neq 0$. Hence $x = y$ and $\Phi_m$ is injective. *Immersion.* Fix $x \in X$ and a nonzero tangent vector $v \in T_xX$; we show $d\Phi_m(x)(v) \neq 0$. Choose an index $k$ with $f_k(x) \neq 0$ and work in the affine chart $\{w_k \neq 0\} \subset \mathbb{P}^N$, on which $\Phi_m$ has local representative $x \mapsto (f_0(x)/f_k(x), \dots, \widehat{1}, \dots, f_N(x)/f_k(x))$. Applying the quotient rule, the $j$-th affine component has differential at $x$: \begin{align*} d\!\left(\frac{f_j}{f_k}\right)(x)(v) &= \frac{f_k(x)\,df_j(x)(v) - f_j(x)\,df_k(x)(v)}{f_k(x)^2}. \end{align*} Tangent separation gives an index $\ell$ with $s_\ell(x) = 0$ and $ds_\ell(x)(v) \neq 0$, i.e., $f_\ell(x) = 0$ and $df_\ell(x)(v) \neq 0$. Taking $j = \ell$, the numerator above becomes $f_k(x)\,df_\ell(x)(v) - 0 = f_k(x)\,df_\ell(x)(v)$, which is nonzero since both factors are nonzero. Hence $d(f_\ell/f_k)(x)(v) \neq 0$, so $d\Phi_m(x)(v) \neq 0$. Since $v \in T_xX$ was arbitrary, $d\Phi_m$ is injective at every $x$, and $\Phi_m$ is an immersion. *Embedding.* The map $\Phi_m \colon X \to \mathbb{P}^N$ is a continuous injection from the compact space $X$ to the Hausdorff space $\mathbb{P}^N$. Restricting the codomain, $\Phi_m \colon X \to \Phi_m(X)$ is a continuous bijection from a compact space to a Hausdorff space, so by the [Topological Inverse Function Theorem](/theorems/318) it is a homeomorphism onto its image. Since $\Phi_m$ is simultaneously a holomorphic immersion, its image $\Phi_m(X)$ is a closed complex submanifold of $\mathbb{P}^N$ and $\Phi_m$ is a biholomorphism onto that submanifold — a holomorphic embedding. The curvature of $L$ is the sole analytic input throughout: positivity supplies the coercive Bochner–Kodaira lower bound that, via the $\bar\partial$-correction construction of the *Peak Section From A Cut-Off* example, manufactures the peak and tangent sections invoked in every step above. The projective model $\Phi_m(X) \subset \mathbb{P}^N$ is the geometric precipitate of that curvature reserve. [/example] The final lesson is that positive curvature is not merely a differential-geometric decoration on a line bundle. It is the analytic resource that manufactures sections, the SCV condition encoded by strictly plurisubharmonic weights, and the algebro-geometric condition recognised as ampleness. Singular metrics and multiplier ideals extend the same mechanism to settings where the geometry has base loci or prescribed singularities, which is why modern complex geometry continues to be organised around the interaction between curvature, $\bar\partial$, and sheaf cohomology.