This course develops the central analytical and topological structures that arise on complex manifolds equipped with Kähler metrics. It begins with the basic geometry of Kähler manifolds and concrete examples, then moves to the differential operators naturally associated to the complex structure, the Lefschetz package from linear algebra, and the identities that tie these pieces together. From there, the course turns to harmonic forms and [Hodge decomposition](/theorems/2745), showing how analytic methods produce powerful cohomological consequences. A recurring theme is that the Kähler condition imposes remarkable rigidity: the $\partial\bar{\partial}$ lemma, [Hard Lefschetz theorem](/theorems/3876), and Hodge structures all emerge as manifestations of the same underlying symmetry. Later chapters use these results to study low-dimensional geometry, especially Kähler surfaces, and to connect the theory with periods, deformations, and applications. The chapters are arranged so that each new layer builds on the previous one, starting from definitions and examples, then establishing the key operator identities, and finally deriving the structural theorems that make Kähler geometry such a rich bridge between analysis, algebra, and topology. # Introduction This course studies the part of complex geometry where a Hermitian metric, a symplectic form, and elliptic analysis reinforce each other. The basic objects are compact Kähler manifolds, and the basic question is how the differential geometry of a metric controls the topology and complex structure of the underlying manifold. The main results of the course are the [Kähler identities](/theorems/3853), Hodge decomposition, Lefschetz theory, and the Hodge-Riemann bilinear relations. Together they explain why compact Kähler manifolds behave much more rigidly than general compact complex manifolds. The course assumes the language of complex manifolds, vector bundles, differential forms, de Rham cohomology, Hermitian metrics, and Dolbeault cohomology. It also uses the standard analytic fact that elliptic [self-adjoint operators](/page/Self-Adjoint%20Operators) on compact manifolds have finite-dimensional spaces of harmonic forms and good regularity theory. We will recall the geometric meaning of each analytic object when it enters, but the course is not a first course in elliptic PDE. The emphasis is on how the Kähler condition aligns operators that, on a general complex manifold, have no reason to communicate. ## The Central Problem What extra structure does a closed Hermitian form impose on a complex manifold? A Hermitian metric on a complex manifold always gives a real $(1,1)$-form, but the requirement that this form be closed is a strong compatibility condition. It says that the metric geometry, the complex structure, and the [exterior derivative](/theorems/1525) are fitted together tightly enough that the de Rham and Dolbeault theories become two views of the same object. The first task is to fix the language used throughout the course. The following definition is the hinge between complex geometry and symplectic geometry. [definition: Kähler Manifold] A Kähler manifold is a complex manifold $X$ equipped with a Hermitian metric $g$ whose associated real $(1,1)$-form $\omega$ satisfies $d\omega = 0$. [/definition] After this definition, two viewpoints will be used at the same time. From the complex-geometric viewpoint, $\omega$ is the fundamental form of a Hermitian metric. From the symplectic viewpoint, $\omega$ is a symplectic form compatible with the complex structure. Chapter 1 makes this equivalence between Hermitian metrics and closed positive $(1,1)$-forms explicit, and the rest of the course is built around its consequences. [example: Projective Space As A Model] On the affine chart $U_0=\{[Z_0:\cdots:Z_n]\mid Z_0\ne 0\}$ with coordinates $z_j=Z_j/Z_0$, set $\rho=1+|z|^2=1+\sum_{j=1}^n z_j\bar z_j$ and $\varphi=\log \rho$. We compute the local Fubini-Study form $\theta_{\mathrm{FS}}=i\partial\bar{\partial}\varphi$ explicitly. Since $\bar{\partial}\rho=\sum_{j=1}^n z_j\,d\bar z_j$, the chain rule gives \begin{align*} \bar{\partial}\varphi=\frac{1}{\rho}\bar{\partial}\rho=\sum_{j=1}^n \frac{z_j}{\rho}\,d\bar z_j. \end{align*} For each $j$, \begin{align*} \partial\left(\frac{z_j}{\rho}\right)=\sum_{k=1}^n \frac{\rho\delta_{jk}-z_j\bar z_k}{\rho^2}\,dz_k. \end{align*} Therefore \begin{align*} \theta_{\mathrm{FS}}=i\sum_{j,k=1}^n \frac{\rho\delta_{jk}-z_j\bar z_k}{\rho^2}\,dz_k\wedge d\bar z_j. \end{align*} For a tangent vector $v=(v_1,\ldots,v_n)$, the associated Hermitian quadratic form is \begin{align*} \sum_{j,k=1}^n \frac{\rho\delta_{jk}-z_j\bar z_k}{\rho^2}v_k\bar v_j=\frac{\rho\sum_{k=1}^n |v_k|^2-\left|\sum_{k=1}^n \bar z_k v_k\right|^2}{\rho^2}. \end{align*} By Cauchy-Schwarz, $\left|\sum_k \bar z_k v_k\right|^2\le |z|^2|v|^2$, so the numerator is at least $(1+|z|^2)|v|^2-|z|^2|v|^2=|v|^2$, and it is positive when $v\ne 0$. The form is closed because \begin{align*} d\theta_{\mathrm{FS}}=i(\partial+\bar{\partial})\partial\bar{\partial}\varphi=0, \end{align*} using $\partial^2=0$, $\bar{\partial}^2=0$, and $\partial\bar{\partial}+\bar{\partial}\partial=0$. On an overlap of two projective charts, changing homogeneous coordinates changes the potential by $\log |f|^2$ for a nowhere-vanishing [holomorphic function](/page/Holomorphic%20Function) $f$, and $i\partial\bar{\partial}\log |f|^2=0$ locally because $\log |f|^2=\log f+\log \bar f$ after choosing a holomorphic branch of $\log f$. Thus these local formulas glue to a global positive closed real $(1,1)$-form on $\mathbb{CP}^n$, showing that the Kähler condition appears naturally in projective geometry. [/example] The example also indicates why local potentials will matter. A Kähler form is locally controlled by a single real-valued function, so metric data can often be converted into complex differential operators. This is the first place where the course begins to replace coordinate computations by structural identities. ## Why Kähler Geometry Has More Cohomology Than Expected Why should a metric condition change the topology of a compact complex manifold? The answer comes from [harmonic representatives](/theorems/2747). On any compact Riemannian manifold, Hodge theory identifies de Rham cohomology classes with harmonic forms. On a compact Kähler manifold, the harmonic condition respects the $(p,q)$-decomposition of complex differential forms, so cohomology splits according to complex type. To state the target structure, we recall the cohomology groups that the course relates. These are not new objects, but their interaction is the point. [definition: Dolbeault Cohomology] Let $X$ be a complex manifold. The Dolbeault cohomology group $H^{p,q}_{\bar{\partial}}(X)$ is the cohomology of the complex \begin{align*} \Omega^{p,q}(X) \xrightarrow{\bar{\partial}} \Omega^{p,q+1}(X) \xrightarrow{\bar{\partial}} \Omega^{p,q+2}(X). \end{align*} [/definition] In these notes, $\Omega^{p,q}(X)$ and $A^{p,q}(X)$ both denote the smooth complex-valued $(p,q)$-forms on $X$; later chapters mostly use the notation $A^{p,q}(X)$ when discussing differential operators. Dolbeault cohomology records the failure of $(p,q)$-forms to be solved by the $\bar{\partial}$-operator. In a general complex manifold it is related to de Rham cohomology through spectral sequences and filtrations. In the Kähler case, those hidden filtrations split into an actual direct sum, which is the first major reason compact Kähler manifolds have strong topological restrictions. The compactness hypothesis is part of the analytic mechanism: without compactness, harmonic representatives need not give finite-dimensional cohomology or a clean global decomposition. The Kähler hypothesis is also essential, since compact complex manifolds such as Hopf surfaces have $b_1=1$, contradicting the evenness forced by the degree-one decomposition below. The decomposition does not say that every complex manifold has cohomology split by type, nor does it identify the ring structure on de Rham cohomology; it is an additive decomposition of complex vector spaces. For now we use it as the organizing target of the course; when the [Hodge decomposition theorem](/theorems/3941) is invoked later, subsequent appearances will refer back to that result rather than restating the full theorem each time. The first obstruction worth isolating is degree $1$. This degree-one part of the decomposition is not only a structural description of cohomology groups; together with complex conjugation, it imposes an arithmetic constraint on the ordinary Betti number $b_1(X)$. Naming this consequence separately is useful because it gives a fast test for non-Kähler compact complex manifolds, before any of the later Lefschetz machinery is available. [example: First Betti Number Constraint] Let $X$ be a compact Kähler manifold. By the degree $1$ case of [Hodge decomposition for compact Kähler manifolds](/theorems/8066), \begin{align*} H^1_{\mathrm{dR}}(X;\mathbb C) \cong H^{1,0}_{\bar{\partial}}(X) \oplus H^{0,1}_{\bar{\partial}}(X). \end{align*} Set $h^{1,0}=\dim_{\mathbb C}H^{1,0}_{\bar{\partial}}(X)$ and $h^{0,1}=\dim_{\mathbb C}H^{0,1}_{\bar{\partial}}(X)$. Complex conjugation sends a $(1,0)$-class to a $(0,1)$-class and gives an anti-linear bijection $H^{1,0}_{\bar{\partial}}(X)\to H^{0,1}_{\bar{\partial}}(X)$, so $h^{1,0}=h^{0,1}$. Since $H^1_{\mathrm{dR}}(X;\mathbb C)\cong H^1_{\mathrm{dR}}(X;\mathbb R)\otimes_{\mathbb R}\mathbb C$, its complex dimension is $b_1(X)$. Taking dimensions in the displayed direct sum gives \begin{align*} b_1(X)=\dim_{\mathbb C}H^1_{\mathrm{dR}}(X;\mathbb C)=h^{1,0}+h^{0,1}. \end{align*} Using $h^{1,0}=h^{0,1}$, this becomes \begin{align*} b_1(X)=2h^{1,0}. \end{align*} Thus $b_1(X)$ is even. A Hopf surface has $b_1=1$, so it cannot admit a Kähler metric. [/example] This parity obstruction is a useful warning. Kähler geometry is not just complex geometry plus a convenient metric; it is a restrictive class with measurable topological signatures. Chapters 7 through 9 will produce stronger restrictions through Hard Lefschetz, primitive decomposition, and the Hodge-Riemann bilinear relations. ## The Analytic Mechanism What identity makes the decomposition theorem work? The decisive point is that the Kähler form supplies algebraic operators on differential forms, and these operators interact with $\partial$, $\bar{\partial}$, and their formal adjoints. The resulting commutator identities are the computational engine behind the whole course. The algebraic operation that begins this story is cup product with the Kähler class, represented analytically by wedging with the Kähler form. [definition: Lefschetz Operator] Let $(X,\theta)$ be a Kähler manifold. The Lefschetz operator is the map \begin{align*} L: \Omega^k(X) \to \Omega^{k+2}(X), \qquad L\alpha = \theta \wedge \alpha. \end{align*} [/definition] The operator $L$ is defined at the level of forms, but on a compact Kähler manifold it descends to cohomology because $d\theta=0$. Its adjoint and commutator generate an $\mathfrak{sl}_2$-action on forms. The Hard Lefschetz theorem, proved later in the Lefschetz part of the course, says that repeated cup product with the Kähler class gives isomorphisms between complementary cohomology degrees. Here we use it only as a guiding target, not as a theorem to be reinstalled in the introduction. The compact Kähler hypotheses cannot be replaced by ordinary [Poincaré duality](/wiki/poincare-duality): every compact oriented manifold has a nondegenerate pairing between complementary degrees, but it need not be produced by powers of a single degree-two class. The closure of $\theta$ is needed even to make wedging with $[\theta]$ a cohomological operation, and the Kähler identities are needed to show that this operation behaves well on harmonic forms. The later theorem does not assert that every class in $H^2_{\mathrm{dR}}(X;\mathbb R)$ has the Lefschetz property; it is the Kähler class, coming from a positive closed $(1,1)$-form, that has this force. This result is therefore stronger than a duality statement and narrower than a statement about arbitrary degree-two classes. Its necessity comes from the fact that it manufactures cohomology in complementary degrees by repeatedly using the Kähler class itself. Its limitation is equally important: without a positive closed $(1,1)$-representative, the Lefschetz ladder may not exist on cohomology, and later primitive decompositions would have no canonical starting point. For surfaces this already gives a concrete operation on cohomology in low degree, and it prepares the later study of the middle-dimensional intersection form. [example: Consequence For Surface Cohomology] Let $X$ be a compact Kähler surface, so its complex dimension is $n=2$. Applying the Hard Lefschetz theorem in degree $k=1$ gives $n-k=1$, hence the Lefschetz map is \begin{align*} L^{1}:H^1_{\mathrm{dR}}(X;\mathbb R)\to H^3_{\mathrm{dR}}(X;\mathbb R). \end{align*} For a class $[\alpha]\in H^1_{\mathrm{dR}}(X;\mathbb R)$, this map is explicitly \begin{align*} L^{1}[\alpha]=[\theta]\smile[\alpha]=[\theta\wedge\alpha]. \end{align*} The theorem says that this map is an isomorphism, so every degree $3$ de Rham class has the form $[\theta\wedge\alpha]$ for a unique degree $1$ class $[\alpha]$. Thus the Kähler metric, through its closed form $\theta$, produces a canonical identification between degree $1$ and degree $3$ cohomology. In later applications, the same operation $[\alpha]\mapsto[\theta]\smile[\alpha]$ is refined in degree $2$ by separating classes into multiples of $[\theta]$ and primitive classes, which is the starting point for controlling the intersection form on $H^2_{\mathrm{dR}}(X;\mathbb R)$. [/example] The analytic and algebraic parts of the course therefore meet in the same operator $L$. The closure of the Kähler form lets $L$ act on cohomology, while the Hermitian linear algebra of $L$ gives a rigid structure to that action. At this point there are two different kinds of structure in play. Elliptic theory supplies finite-dimensional harmonic representatives, but it does not know about complex type. Kähler geometry supplies the commutator identities that make the elliptic representatives compatible with $(p,q)$-decomposition and with the Lefschetz operators. The rest of the course repeatedly uses this division of labor: first isolate the analytic input, then add the Kähler identities, then translate the resulting harmonic statements into cohomology. ## What The Course Proves And What It Uses Which results are inputs, and which are proved inside the course? The course uses standard elliptic theory as a background theorem, then proves the Kähler-specific identities and their cohomological consequences. This separation keeps the focus on complex geometry while still making the analytic dependence explicit. The main analytic input is the [Hodge theorem](/theorems/3942) for compact Riemannian manifolds. We use it as a quoted theorem, since its proof belongs to [elliptic operator](/page/Elliptic%20Operator) theory rather than to Kähler geometry itself. The materialized theorem card records the analytic decomposition of forms into harmonic forms and Laplacian images, together with finite-dimensionality of the harmonic space. In the standard de Rham application of that decomposition, a closed form determines a unique harmonic representative of its cohomology class, so the harmonic $p$-forms identify canonically with $H^p_{\mathrm{dR}}(M;\mathbb R)$. This cohomological interpretation is the part the Kähler arguments will use. [quotetheorem:2745] [citeproof:2745] After the proof citation, the point for this course is not the elliptic proof itself but the role of the result as an analytic input. The displayed orthogonal splitting is the analytic mechanism behind harmonic representatives in de Rham cohomology; it separates the finite-dimensional harmonic contribution from the Laplacian image. Compactness is essential to the scope of this statement: on noncompact manifolds, harmonic forms can behave badly and need not represent ordinary de Rham cohomology in this clean way. The theorem is also purely Riemannian. Its cohomological corollary gives canonical harmonic representatives for de Rham classes, but it does not see the $(p,q)$-decomposition on a complex manifold and does not by itself imply [Hodge decomposition](/wiki/hodge-decomposition). For example, applying it to a compact Hermitian non-Kähler manifold gives harmonic representatives, but it gives no reason for the $d$-Laplacian and $\bar{\partial}$-Laplacian to have the same kernel. The later Kähler identities supply the extra complex-geometric input: they install the operator identities that let harmonic theory respect complex type, instead of merely recalling the Riemannian theorem in another language. The Riemannian theorem alone leaves a gap: its cohomological use produces harmonic representatives, but it does not explain why those representatives should respect the $(p,q)$-splitting of complex forms. To close that gap, the course needs the Kähler identities, which compare the Lefschetz operator $L$, its adjoint $\Lambda$, and the adjoints of $\partial$ and $\bar{\partial}$. These identities are not decorative commutator formulas: they are the mechanism that turns a Riemannian harmonic representative into a complex-geometric object. On a general Hermitian manifold, torsion terms appear in the corresponding formulas, and the equality of the $d$, $\partial$, and $\bar{\partial}$ harmonic theories can fail. The Kähler condition is exactly what prevents extra torsion terms from entering these formulas. On a general Hermitian manifold the associated $(1,1)$-form need not be closed, and the analogous commutators include terms depending on $d\theta$; the identities are then false in their clean Kähler form. They do not say that $\partial$ and $\bar{\partial}$ are the same operator, nor that their cohomology theories agree on arbitrary complex manifolds. What they give is a precise operator-level bridge between type decomposition, adjoints, and the Lefschetz algebra, and this bridge is the input for Laplacian equality, Hodge decomposition, and Hard Lefschetz. The identities are local in origin, so their hypotheses should not be hidden. They rely on the Kähler condition rather than merely on a Hermitian metric, and their clean form is lost when torsion terms enter. Their forward use is operator-theoretic: once adjoints are expressible through Lefschetz commutators, the corresponding Laplacians can be compared directly, which is the form needed for Hodge theory. The first major application converts these commutator identities into comparisons among elliptic operators. This step is necessary because Hodge theory is phrased in terms of kernels of Laplacians, not directly in terms of commutators. On a Kähler manifold, the commutator identities imply the Laplacian equalities used later to compare $d$-, $\partial$-, and $\bar{\partial}$-harmonic forms. These equalities are not formal properties of every Hermitian metric; they are consequences of the Kähler condition and can fail when torsion terms are present. They do not identify the complexes themselves, but they identify their harmonic forms, which is exactly what Hodge theory needs. This is the point at which the course stops treating the Kähler identities as isolated formulas and starts using them as infrastructure: later chapters should cite this bridge when harmonic representatives, type decompositions, or Lefschetz operators reappear, rather than re-rendering the same identity theorem as a new result each time. The course also uses sheaf cohomology only through its Dolbeault incarnation. When a holomorphic vector bundle appears, Dolbeault cohomology computes the corresponding sheaf cohomology, but the proofs here remain at the level of forms, metrics, connections, and elliptic operators. ## Roadmap Of The Lectures How does the course move from examples to Hodge theory? The first part builds the Kähler condition from metrics and forms, then studies examples such as projective space, tori, Riemann surfaces, products, and hypersurfaces. Non-examples are included early because they show that the definition has topological content. The second part develops the calculus of differential operators on Kähler manifolds. Forms decompose into $(p,q)$-types, the exterior derivative splits as $d=\partial+\bar{\partial}$, and Hermitian metrics give formal adjoints and Laplacians. The Kähler identities then connect these operators and yield the equality of harmonic theories. The third part draws cohomological consequences. [Hodge decomposition](/wiki/hodge-decomposition) identifies de Rham cohomology with Dolbeault cohomology, [Hodge symmetry](/wiki/hodge-symmetry) relates $H^{p,q}$ and $H^{q,p}$, and the Hodge filtration packages this information in a way that is stable under maps. These results form the basic Hodge-theoretic toolkit used throughout complex geometry; after each is established, later chapters use it by reference and focus on what new geometric information it unlocks. The final part studies Lefschetz theory. Wedging with the Kähler form gives the Lefschetz operator, its adjoint gives primitive decomposition, and the induced $\mathfrak{sl}_2$ representation proves [Hard Lefschetz](/wiki/hard-lefschetz-theorem). The [Hodge-Riemann bilinear relations](/wiki/hodge-riemann-bilinear-relations) then refine [Poincaré duality](/wiki/poincare-duality) by describing the signs of natural Hermitian forms on primitive cohomology. The intended rhythm is cumulative: formal theorem cards appear when a new tool is first installed, while later sections synthesize and apply the installed tools rather than restarting their proofs or restating their full cards. In particular, packaged lists of consequences are treated as summaries or diagnostics in prose, not as new theorem cards unless they assert a genuinely new result. [remark: Scope Of The Course] The course stops before the major theories of positivity, vanishing, variation of Hodge structure, and moduli. Those subjects depend on the tools developed here, but they require additional geometric input. The goal of these notes is to make the Kähler package available as a working language for later complex geometry. [/remark] By the end of the course, the reader should be able to move fluently between four descriptions of the same structure: a closed positive $(1,1)$-form, a compatible Hermitian metric, harmonic representatives of cohomology classes, and Lefschetz-type algebra on differential forms. The introduction has named these objects; the rest of the course proves why they are inseparable on compact Kähler manifolds. The introduction has named the central objects, but the real work begins by asking when they actually fit together. Chapter 1 starts from the compatibility between a complex structure and a Hermitian metric, then shows how a closed positive $(1,1)$-form captures that compatibility in concrete examples. # 1. Kähler Metrics and First Examples Continuing from the introduction's definition of a Kähler manifold, Kähler geometry begins with a compatibility problem. A complex manifold already has an operator $J$ on tangent spaces, a Hermitian metric measures lengths in a way compatible with $J$, and differential topology supplies closed $2$-forms. The Kähler condition is the point at which these structures become the same object: a Hermitian metric whose associated real $(1,1)$-form is closed. The chapter assumes the standard language of complex manifolds, differential forms, de Rham cohomology, and the operators $\partial$ and $\bar\partial$. This first chapter introduces the definition, proves the local potential theorem, and builds the examples that will carry the rest of the course. ## Closed Positive Forms and Hermitian Metrics The first question is how to recognise a Riemannian metric that respects the complex structure and also behaves like a symplectic form. On a complex manifold $X$, the complex structure $J:TX\to TX$ gives a preferred rotation on each real tangent space. A Hermitian metric is a Riemannian metric invariant under this rotation, and the associated $2$-form records the same data in skew-symmetric form. [definition: Hermitian Metric] Let $X$ be a complex manifold with complex structure $J$. A Hermitian metric on $X$ is a smooth map \begin{align*} g:TX\times_X TX\to\mathbb R \end{align*} such that each $g_x:T_xX\times T_xX\to\mathbb R$ is a positive definite symmetric [bilinear form](/page/Bilinear%20Form) and \begin{align*} g_x(Ju,Jv)=g_x(u,v) \end{align*} for all tangent vectors $u,v\in T_xX$ and all $x\in X$. [/definition] The compatibility condition says that multiplication by $i$ acts by isometries on tangent spaces. From $g$ and $J$ one forms a differential form, and this form will be the main object in the course. [definition: Fundamental Form] Let $(X,J,g)$ be a Hermitian manifold. The fundamental form of $g$ is the real $2$-form $\omega\in \Omega^2(X;\mathbb R)$ defined by \begin{align*} \omega(u,v)=g(Ju,v) \end{align*} for tangent vectors $u,v\in T_xX$. [/definition] The fundamental form packages the metric into a differential form, but a random real $2$-form need not determine positive lengths. To recover a Hermitian metric from a form, we need a positivity condition measured against the complex structure. [definition: Positive Real One One Form] Let $X$ be a complex manifold. A real $(1,1)$-form $\omega$ on $X$ is positive if, for every non-zero real tangent vector $v\in T_xX$, \begin{align*} \omega(v,Jv)>0. \end{align*} [/definition] With this convention, a positive real $(1,1)$-form defines a Hermitian metric by $g(u,v)=\omega(u,Jv)$. Positivity gives the metric part of the theory; the next issue is which such forms interact well with de Rham cohomology and symplectic geometry. [definition: Kähler Form] Let $X$ be a complex manifold. A Kähler form on $X$ is a smooth real $(1,1)$-form \begin{align*} \omega\in\Omega^{1,1}(X;\mathbb R) \end{align*} such that \begin{align*} d\omega=0 \end{align*} and $\omega$ is positive. [/definition] This definition turns complex geometry into a constrained version of symplectic geometry: every Kähler form is symplectic, but not every symplectic form has type $(1,1)$ and positivity relative to the given complex structure. Since courses often switch between metrics and forms, we also name the metric version of the same condition. [definition: Kähler Metric] Let $X$ be a complex manifold. A Kähler metric on $X$ is a Hermitian metric \begin{align*} g:TX\times_X TX\to\mathbb R \end{align*} whose fundamental form $\omega$ is a Kähler form. [/definition] The local coordinate expression is often the most efficient way to calculate. In holomorphic coordinates $(z_1,\dots,z_n)$, a Hermitian metric is written \begin{align*} g=\sum_{i,j} g_{i\bar j}\,dz_i\otimes d\bar z_j, \end{align*} and the associated form is \begin{align*} \omega=i\sum_{i,j}g_{i\bar j}\,dz_i\wedge d\bar z_j \end{align*} up to the chosen normalisation. Throughout this chapter, constants are less important than type, positivity, and closedness; later analytic identities require fixing one convention consistently. [example: The Standard Form on Complex Euclidean Space] On $\mathbb C^n$ write $z_j=x_j+iy_j$, so \begin{align*} dz_j=dx_j+i\,dy_j,\quad d\bar z_j=dx_j-i\,dy_j. \end{align*} For each $j$, \begin{align*} \frac{i}{2}dz_j\wedge d\bar z_j=\frac{i}{2}(dx_j+i\,dy_j)\wedge(dx_j-i\,dy_j). \end{align*} Expanding the wedge product gives \begin{align*} \frac{i}{2}dz_j\wedge d\bar z_j=\frac{i}{2}\left(dx_j\wedge dx_j-i\,dx_j\wedge dy_j+i\,dy_j\wedge dx_j+dy_j\wedge dy_j\right). \end{align*} Since $dx_j\wedge dx_j=0$, $dy_j\wedge dy_j=0$, and $dy_j\wedge dx_j=-dx_j\wedge dy_j$, this becomes \begin{align*} \frac{i}{2}dz_j\wedge d\bar z_j=\frac{i}{2}(-2i\,dx_j\wedge dy_j)=dx_j\wedge dy_j. \end{align*} Hence \begin{align*} \omega_0=\frac{i}{2}\sum_{j=1}^n dz_j\wedge d\bar z_j=\sum_{j=1}^n dx_j\wedge dy_j. \end{align*} Each $dz_j$ has type $(1,0)$ and each $d\bar z_j$ has type $(0,1)$, so every summand $dz_j\wedge d\bar z_j$ has type $(1,1)$. Also \begin{align*} d\omega_0=\sum_{j=1}^n d(dx_j\wedge dy_j)=\sum_{j=1}^n \left(d^2x_j\wedge dy_j-dx_j\wedge d^2y_j\right)=0. \end{align*} Now take a real tangent vector \begin{align*} v=\sum_{j=1}^n\left(a_j\frac{\partial}{\partial x_j}+b_j\frac{\partial}{\partial y_j}\right). \end{align*} With the standard complex structure, \begin{align*} Jv=\sum_{j=1}^n\left(a_j\frac{\partial}{\partial y_j}-b_j\frac{\partial}{\partial x_j}\right). \end{align*} For each $j$, \begin{align*} (dx_j\wedge dy_j)(v,Jv)=dx_j(v)dy_j(Jv)-dx_j(Jv)dy_j(v)=a_j^2+b_j^2. \end{align*} Therefore \begin{align*} \omega_0(v,Jv)=\sum_{j=1}^n(a_j^2+b_j^2)=|v|^2. \end{align*} Thus $\omega_0$ is closed, real, of type $(1,1)$, and positive, so it is the standard Kähler form on $\mathbb C^n$. [/example] The example suggests that Kähler geometry should have local scalar potentials, since the flat form is obtained from $|z|^2$. The issue is that a general real $(1,1)$-form is given by many coefficient functions, and it is not automatic that these coefficients are all mixed second derivatives of one scalar function. Closedness is precisely the compatibility condition that makes those mixed derivatives integrable locally, while positivity then becomes strict plurisubharmonicity of the potential. [quotetheorem:8040] [citeproof:8040] Local potentials replace a closed-form condition by a scalar second derivative condition. The closedness hypothesis is essential: without $d\omega=0$, the coefficient integrability conditions fail, so there is no reason for $\omega$ to be a mixed complex Hessian. The type $(1,1)$ hypothesis is also essential, since $i\partial\bar\partial\varphi$ always has type $(1,1)$ and cannot produce a $(2,0)$ or $(0,2)$ component. Positivity is not needed for the existence of a local potential, but it is exactly what upgrades the potential to a strictly plurisubharmonic one. This is the bridge to projective examples, where the metric is naturally written as $i\partial\bar\partial$ of a logarithm. ## Projective Space and the Fubini-Study Form The next problem is to put a canonical Kähler metric on complex projective space. Points of $\mathbb{CP}^n$ are complex lines in $\mathbb C^{n+1}$, so any formula on $\mathbb C^{n+1}\setminus\{0\}$ must be unchanged when homogeneous coordinates are rescaled. The logarithm of the squared Euclidean length has exactly the right transformation property: it changes by a pluriharmonic term, which vanishes after applying $i\partial\bar\partial$. [definition: Fubini Study Form] Let $[Z_0:\cdots:Z_n]$ be homogeneous coordinates on $\mathbb{CP}^n$. On the affine chart $U_j=\{Z_j\ne 0\}$, set $z_k=Z_k/Z_j$ for $k\ne j$. The Fubini-Study form is the real $(1,1)$-form $\omega_{FS}\in\Omega^{1,1}(\mathbb{CP}^n;\mathbb R)$ whose restriction to $U_j$ is \begin{align*} \omega_{FS}\big|_{U_j}=i\partial\bar\partial\log\left(1+\sum_{k\ne j}|z_k|^2\right). \end{align*} [/definition] The formula defines local forms on affine charts, so the construction still has to pass two tests: compatibility on chart overlaps and positivity on tangent directions. The possible obstruction is that homogeneous coordinates are not unique: changing charts or rescaling representatives could change the potential by more than a harmless pluriharmonic term, and then $i\partial\bar\partial$ would not glue to a well-defined form on projective space. [quotetheorem:3842] [citeproof:3842] This construction is the prototype for metrics arising from positive line bundles. The logarithm is not an arbitrary choice: using a function that changes by something non-pluriharmonic on overlaps would make $i\partial\bar\partial$ produce different local forms, so the pieces would not glue. The quotient descent is a separate requirement from ordinary overlap compatibility: a form written upstairs on $\mathbb C^{n+1}\setminus\{0\}$ must be basic for the $\mathbb C^\times$-action, meaning invariant and zero on vertical directions after passing to the quotient. For instance, the flat form $\frac{i}{2}\sum dZ_k\wedge d\bar Z_k$ is closed and positive upstairs, but it scales by $|\lambda|^2$ under $Z\mapsto\lambda Z$, so it does not define a form on projective space. Closedness is another independent condition: on any affine chart of complex dimension at least two, the positive real $(1,1)$-form \begin{align*} \omega=i e^{|z_2|^2}\,dz_1\wedge d\bar z_1+i\sum_{k=2}^n dz_k\wedge d\bar z_k \end{align*} is positive, but $d\omega$ has non-zero terms involving $dz_2$ and $d\bar z_2$ differentiated against the varying coefficient, so it is not Kähler. Positivity is also a real condition, since a closed real $(1,1)$-form on projective space could be negative or degenerate on some complex tangent directions and would not define a Kähler metric. The compactness of $\mathbb{CP}^n$ makes this the first model where local potentials exist but need not come from one globally defined function. In the first week of the course, it is enough to use it as the canonical compact example. For a concrete gluing failure, replace the Fubini-Study potentials on two affine charts by potentials whose difference on the overlap is $\log(1+|z|^2)$ rather than the real part of a holomorphic function; applying $i\partial\bar\partial$ to the difference gives a non-zero form, so the chartwise definitions disagree on the overlap. For a concrete positivity failure, the zero form or any semipositive form pulled back from a lower-dimensional projection is closed and has type $(1,1)$, but it vanishes on some non-zero complex tangent directions and therefore is not Kähler. [example: The Fubini Study Form on the Projective Line] On $\mathbb{CP}^1$, use the affine coordinate $z=Z_1/Z_0$ on $U_0=\{Z_0\ne 0\}$ and write $z=x+iy$. The Fubini-Study potential on this chart is \begin{align*} \varphi(z)=\log(1+|z|^2)=\log(1+z\bar z). \end{align*} Since $\bar\partial\varphi=\frac{z}{1+z\bar z}\,d\bar z$, we compute the $z$-derivative of the coefficient: \begin{align*} \partial\left(\frac{z}{1+z\bar z}\right)=\frac{(1+z\bar z)-z\bar z}{(1+z\bar z)^2}\,dz=\frac{1}{(1+|z|^2)^2}\,dz. \end{align*} Therefore \begin{align*} \partial\bar\partial\varphi=\frac{dz\wedge d\bar z}{(1+|z|^2)^2}. \end{align*} Thus the chart formula is \begin{align*} \omega_{FS}=i\partial\bar\partial\log(1+|z|^2)=\frac{i\,dz\wedge d\bar z}{(1+|z|^2)^2}. \end{align*} To compare with the round sphere, use $z=x+iy$ and $r^2=x^2+y^2$. Since $i\,dz\wedge d\bar z=2\,dx\wedge dy$, the Fubini-Study form becomes \begin{align*} \omega_{FS}=\frac{2\,dx\wedge dy}{(1+r^2)^2}. \end{align*} Under stereographic projection from the unit round sphere, the round area form in the coordinate plane is \begin{align*} \operatorname{area}_{S^2}=\frac{4\,dx\wedge dy}{(1+r^2)^2}. \end{align*} Hence \begin{align*} \operatorname{area}_{S^2}=2\,\omega_{FS}. \end{align*} So $\omega_{FS}$ is the round area form up to the constant factor $1/2$, and $\mathbb{CP}^1$ is the Riemann sphere with its constant-curvature Kähler metric. [/example] Projective space also supplies Kähler metrics on many submanifolds. The mechanism is restriction: a positive closed $(1,1)$-form remains positive and closed when pulled back by a holomorphic embedding. [quotetheorem:8041] [citeproof:8041] A first consequence is that smooth projective varieties are Kähler when viewed as complex manifolds. The submanifold hypothesis matters: if the subset is singular, its tangent spaces are not ordinary complex vector spaces everywhere, so the same metric statement cannot be read directly without passing to the smooth locus or using singular Kähler geometry. Holomorphicity matters as well, because a merely smooth real submanifold need not have $J$-invariant tangent spaces, and the pullback may fail to have type $(1,1)$ in any intrinsic complex sense. The theorem also does not say that every holomorphic pullback of a Kähler form is Kähler: if $f:Y\to X$ is holomorphic but not immersive, then $f^*\omega$ is closed and has type $(1,1)$, yet it can be degenerate on non-zero vectors in $\ker df_y$. A constant holomorphic map is the extreme case, since its pullback is the zero form. Thus the embedding, or at least an immersion, is essential for the positivity part. This theorem is the practical checklist in its simplest form: pull back the form, check type through holomorphicity, check closedness through $d\iota^*=\iota^*d$, and check positivity on non-zero tangent vectors. In this chapter we only need the smooth hypersurface case. [example: Smooth Hypersurfaces in Projective Space] Let $Y=\{F=0\}\subset\mathbb{CP}^n$ be smooth, and let $\iota:Y\hookrightarrow\mathbb{CP}^n$ be the inclusion. Smoothness means that $Y$ is a complex submanifold of $\mathbb{CP}^n$ of complex dimension $n-1$, and $\iota$ is holomorphic because its coordinate functions are the restrictions of holomorphic projective coordinates. Set \begin{align*} \omega_Y=\iota^*\omega_{FS}. \end{align*} Since pullback commutes with exterior differentiation and $\omega_{FS}$ is closed, \begin{align*} d\omega_Y=d(\iota^*\omega_{FS})=\iota^*(d\omega_{FS})=\iota^*0=0. \end{align*} Because $\iota$ is holomorphic, $d\iota_y\circ J_Y=J_{\mathbb{CP}^n}\circ d\iota_y$, so the pullback of the real $(1,1)$-form $\omega_{FS}$ is again a real $(1,1)$-form on $Y$. It remains to check positivity. If $0\ne v\in T_yY$, then $d\iota_y(v)\ne 0$ because $\iota$ is an embedding. Hence \begin{align*} \omega_Y(v,J_Yv)=(\iota^*\omega_{FS})(v,J_Yv)=\omega_{FS}(d\iota_yv,d\iota_y(J_Yv)). \end{align*} Using holomorphicity of $\iota$ in the second slot gives \begin{align*} \omega_Y(v,J_Yv)=\omega_{FS}(d\iota_yv,J_{\mathbb{CP}^n}d\iota_yv)>0, \end{align*} because $\omega_{FS}$ is positive on $\mathbb{CP}^n$. Thus $\iota^*\omega_{FS}$ is a Kähler form on $Y$. For example, a smooth plane curve $F(Z_0,Z_1,Z_2)=0$ inherits this form from $\mathbb{CP}^2$. The smooth quadric \begin{align*} Z_0^2+\cdots+Z_n^2=0 \end{align*} also inherits it, since its partial derivatives are $2Z_0,\dots,2Z_n$, which cannot all vanish at a projective point. Likewise the Fermat hypersurface \begin{align*} Z_0^d+\cdots+Z_n^d=0 \end{align*} is smooth for $d\geq 1$, because its partial derivatives are $dZ_0^{d-1},\dots,dZ_n^{d-1}$ and these vanish simultaneously only when all $Z_j=0$, which is not a point of projective space. In each case the Kähler metric is the one determined by restricting the Fubini-Study form. [/example] ## Tori, Riemann Surfaces, and Products Projective space is compact and curved, but many Kähler manifolds are built from flat or lower-dimensional pieces. The guiding question is which constructions preserve the closed positive $(1,1)$ condition. Quotients by translations, complex dimension one, and products are the first stability mechanisms. [quotetheorem:8043] [citeproof:8043] The descended form is the simplest compact Kähler metric not obtained from projective space in the argument above. The translation-invariance hypothesis is the key descent condition: a form on the covering space descends only when it is invariant under the deck transformations. The lattice being full rank is also essential for compactness of the quotient; without it one still obtains a complex quotient in favourable cases, but not the compact torus used here. The theorem does not say that every complex torus is projective: in dimensions at least two, many complex tori admit Kähler forms but have no holomorphic embedding into projective space. This example teaches a second verification checklist: construct the form upstairs, prove invariance under the [group action](/page/Group%20Action), then check type, closedness, and positivity after pulling back by the covering map. In complex dimension one it gives the standard flat metric on an elliptic curve, where the area can be read from a fundamental parallelogram. [example: A Flat Elliptic Curve] Let $\tau=a+ib$ with $b=\operatorname{Im}(\tau)>0$, and let $\pi:\mathbb C\to E=\mathbb C/\Lambda$ be the quotient map. On $\mathbb C$ write $z=x+iy$, so $dz=dx+i\,dy$ and $d\bar z=dx-i\,dy$. Then \begin{align*} dz\wedge d\bar z=(dx+i\,dy)\wedge(dx-i\,dy) \end{align*} \begin{align*} dz\wedge d\bar z=dx\wedge dx-i\,dx\wedge dy+i\,dy\wedge dx+dy\wedge dy \end{align*} Using $dx\wedge dx=0$, $dy\wedge dy=0$, and $dy\wedge dx=-dx\wedge dy$, we get \begin{align*} dz\wedge d\bar z=-2i\,dx\wedge dy \end{align*} and hence \begin{align*} \frac{i}{2}dz\wedge d\bar z=dx\wedge dy. \end{align*} For any $\lambda\in\Lambda$, the translation $T_\lambda(z)=z+\lambda$ satisfies $T_\lambda^*dz=dz$ and $T_\lambda^*d\bar z=d\bar z$, so $\frac{i}{2}dz\wedge d\bar z$ is $\Lambda$-invariant and descends to a real $2$-form $\omega$ on $E$ with $\pi^*\omega=\frac{i}{2}dz\wedge d\bar z$. Since $d(dx\wedge dy)=d^2x\wedge dy-dx\wedge d^2y=0$, the descended form is closed. It has type $(1,1)$ because it is written as a scalar multiple of $dz\wedge d\bar z$, and for a tangent vector $v=\alpha\frac{\partial}{\partial x}+\beta\frac{\partial}{\partial y}$, the standard complex structure gives $Jv=\alpha\frac{\partial}{\partial y}-\beta\frac{\partial}{\partial x}$, so \begin{align*} \omega(v,Jv)=(dx\wedge dy)(v,Jv)=\alpha^2+\beta^2. \end{align*} This is positive whenever $v\ne 0$, so $\omega$ is a flat Kähler form on the elliptic curve. A fundamental parallelogram is parametrized by $F(s,t)=s+t\tau$ for $0\leq s,t\leq 1$. In real coordinates this is $x=s+at$ and $y=bt$, so \begin{align*} F^*(dx\wedge dy)=d(s+at)\wedge d(bt)=(ds+a\,dt)\wedge b\,dt=b\,ds\wedge dt. \end{align*} Therefore the total area is \begin{align*} \int_E\omega=\int_0^1\int_0^1 b\,ds\,dt=b=\operatorname{Im}(\tau). \end{align*} Thus the displayed normalization gives the flat elliptic curve area equal to the Euclidean area of its fundamental parallelogram. [/example] Complex dimension one has a special simplification beyond the flat quotient case. The obstruction $d\omega=0$ disappears for degree reasons, so every Hermitian metric on a Riemann surface automatically enters Kähler geometry. [quotetheorem:8044] [citeproof:8044] This explains why Kähler geometry begins to differ from Hermitian geometry only in complex dimension at least two. The dimension-one hypothesis is essential: in complex dimension at least two, a Hermitian fundamental form is still a real $(1,1)$-form, but its exterior derivative is a genuine $3$-form and may be non-zero. The theorem also depends on using a Hermitian metric, because an arbitrary Riemannian metric on a surface need not be compatible with the given complex structure. Thus Riemann surfaces are special not because closedness is automatic for all Hermitian manifolds, but because the degree in which $d\omega$ would live is absent. To build higher-dimensional examples from lower-dimensional ones, we need a stability result for Cartesian products. [quotetheorem:8045] [citeproof:8045] The product theorem lets the basic list of examples generate many new ones without new local calculations. The sum of the two pullbacks is important: using only $\pi_X^*\omega_X$ would vanish on tangent vectors pointing purely in the $Y$ direction, so it would be semipositive rather than positive. Holomorphicity of the projections is what preserves type $(1,1)$, while closedness is inherited separately from each factor. This gives another reusable checklist: decompose tangent vectors into factor directions, verify closedness term by term, and ensure positivity in every non-zero factor direction. It is especially useful for combining curves, whose Kähler metrics are abundant, with flat complex tori. [example: Products of Curves and Tori] Let $C$ be a compact Riemann surface, let $T=\mathbb C^n/\Lambda$, and write \begin{align*} \pi_C:C\times T\to C,\qquad \pi_T:C\times T\to T. \end{align*} Choose a Hermitian metric on $C$ with fundamental form $\omega_C$. Since $C$ has real dimension $2$, $d\omega_C$ is a $3$-form and therefore $d\omega_C=0$. The form $\omega_C$ is positive because, for $0\ne v\in T_pC$, \begin{align*} \omega_C(v,J_Cv)=g_C(J_Cv,J_Cv)=g_C(v,v)>0. \end{align*} On $\mathbb C^n$ use the flat form \begin{align*} \omega_0=\frac{i}{2}\sum_{j=1}^n dz_j\wedge d\bar z_j. \end{align*} For every lattice translation $T_\lambda(z)=z+\lambda$, one has $T_\lambda^*dz_j=dz_j$ and $T_\lambda^*d\bar z_j=d\bar z_j$, so $T_\lambda^*\omega_0=\omega_0$. Hence $\omega_0$ descends to a form $\omega_T$ on $T$, and the quotient map $q:\mathbb C^n\to T$ satisfies $q^*\omega_T=\omega_0$. Since \begin{align*} d\omega_0=\frac{i}{2}\sum_{j=1}^n d(dz_j\wedge d\bar z_j)=\frac{i}{2}\sum_{j=1}^n(d^2z_j\wedge d\bar z_j-dz_j\wedge d^2\bar z_j)=0, \end{align*} we get $d\omega_T=0$ after pulling back by the local covering map $q$. Also, for a real tangent vector $w=\sum_j(a_j\partial_{x_j}+b_j\partial_{y_j})$ on $\mathbb C^n$, \begin{align*} \omega_0(w,J_Tw)=\sum_{j=1}^n(a_j^2+b_j^2), \end{align*} so $\omega_T$ is positive on the quotient. Define \begin{align*} \omega=\pi_C^*\omega_C+\pi_T^*\omega_T. \end{align*} Both projections are holomorphic, so both pullbacks have type $(1,1)$, and their sum has type $(1,1)$. Closedness follows term by term: \begin{align*} d\omega=d(\pi_C^*\omega_C)+d(\pi_T^*\omega_T)=\pi_C^*(d\omega_C)+\pi_T^*(d\omega_T)=0. \end{align*} For a tangent vector $(v,w)\in T_{(p,t)}(C\times T)$, the product complex structure is $J(v,w)=(J_Cv,J_Tw)$, and the pullback terms give \begin{align*} \omega((v,w),J(v,w))=\omega_C(v,J_Cv)+\omega_T(w,J_Tw). \end{align*} If $(v,w)\ne 0$, then either $v\ne 0$ or $w\ne 0$, so at least one summand is positive and the other is non-negative. Thus $\omega$ is a Kähler form on $C\times T$, combining the curved directions from $C$ with the flat translation-invariant directions from $T$. [/example] ## First Obstructions and Non-Examples The final question in this chapter is how the Kähler condition can fail. Since every Kähler form is symplectic, compactness already imposes a cohomological constraint: the Kähler form cannot be exact. Hodge theory supplies much stronger restrictions, including parity of the first Betti number. [quotetheorem:3856] [citeproof:3856/complex-geometry-ii-k-hler-manifolds-and-hodge-theory] This argument is the first example of Kähler geometry turning metric positivity into a cohomological statement: $[\omega]\in H^2_{dR}(X;\mathbb R)$ must be non-zero, and in fact $[\omega]^n\ne 0$. Compactness is essential because [Stokes' theorem](/theorems/1530) gives the vanishing of the integral only on compact manifolds without boundary; on non-compact manifolds, exact symplectic and exact Kähler forms can exist, as on $\mathbb C^n$ with the standard form. Positivity is also essential, since a closed exact $2$-form that is not positive need not define a volume form and gives no contradiction. The theorem is only a first obstruction: non-exactness of a closed positive-looking cohomology class does not by itself construct a Kähler metric. It immediately rules out compact complex manifolds whose degree-two de Rham cohomology vanishes. [example: Hopf Surface Obstruction] Let $X=(\mathbb C^2\setminus\{0\})/\langle A\rangle$, where $A(z)=2z$. This compact complex surface is a Hopf surface, and topologically it is diffeomorphic to $S^1\times S^3$. We compute the obstruction from degree-two de Rham cohomology. Over $\mathbb R$, the [Kunneth formula](/theorems/3933) gives \begin{align*} H^2_{dR}(S^1\times S^3;\mathbb R)\cong H^2(S^1;\mathbb R)\otimes H^0(S^3;\mathbb R)\oplus H^1(S^1;\mathbb R)\otimes H^1(S^3;\mathbb R)\oplus H^0(S^1;\mathbb R)\otimes H^2(S^3;\mathbb R). \end{align*} The cohomology groups in the three summands are \begin{align*} H^2(S^1;\mathbb R)=0,\qquad H^1(S^3;\mathbb R)=0,\qquad H^2(S^3;\mathbb R)=0. \end{align*} Therefore each summand is zero: \begin{align*} H^2_{dR}(S^1\times S^3;\mathbb R)\cong 0\otimes H^0(S^3;\mathbb R)\oplus H^1(S^1;\mathbb R)\otimes 0\oplus H^0(S^1;\mathbb R)\otimes 0=0. \end{align*} Since $X$ is diffeomorphic to $S^1\times S^3$, it follows that \begin{align*} H^2_{dR}(X;\mathbb R)=0. \end{align*} Now let $\eta$ be any closed real $2$-form on $X$. Its de Rham class satisfies $[\eta]=0$ in $H^2_{dR}(X;\mathbb R)$, so by the definition of de Rham cohomology there is a real $1$-form $\alpha$ on $X$ with \begin{align*} \eta=d\alpha. \end{align*} Thus every closed real $2$-form on $X$ is exact. If $X$ admitted a Kähler form $\omega$, then $\omega$ would be closed by definition of a Kähler form, hence exact by the calculation above. This contradicts the fact that an exact Kähler form cannot occur on a compact manifold, because $X$ is compact. Therefore this Hopf surface is not Kähler. [/example] The Hopf surface also signals a deeper obstruction. Compact Kähler manifolds have even first Betti number, a result governed by Hodge decomposition and complex conjugation on $H^1$. [quotetheorem:8077] [citeproof:8077] The theorem turns the existence of a Kähler metric into a parity constraint on topology. Compactness is part of the statement because the Hodge-theoretic input is a compact Kähler theorem; without compactness, ordinary Betti-number arguments no longer have the same force. The Kähler hypothesis is also essential, since compact complex manifolds such as Hopf surfaces can have odd first Betti number while still carrying Hermitian metrics. This result is stronger than the exactness obstruction in examples where $H^2_{dR}(X;\mathbb R)$ does not vanish but the first Betti number is odd. In this chapter, the theorem is recorded as a preview of why Kähler metrics are much more restrictive than Hermitian metrics. [remark: Why Hopf Surfaces Reappear] The standard Hopf surface has $b_1=1$, so the parity theorem gives a second obstruction to being Kähler. The exactness obstruction above is more elementary and already rules it out in this chapter. The parity argument also applies to compact complex manifolds where cohomology in degree two is not enough to decide the question. [/remark] The examples and obstructions from this chapter set the baseline. Projective space, smooth projective submanifolds, tori, Riemann surfaces, and products are Kähler; Hopf surfaces are not. Chapter 2 begins that explanation by separating the operators $d$, $\partial$, $\bar\partial$, and $d^c$ and by introducing the adjoints and Laplacians that the Kähler condition will later relate. # 2. Differential Operators on Kähler Manifolds After Chapter 1 identified Kähler metrics with closed positive real $(1,1)$-forms and built the basic examples, this chapter sets up the analytic language used throughout Hodge theory on Kähler manifolds. The preceding chapter introduced Kähler metrics as Hermitian metrics whose associated real $(1,1)$-form is closed; now we examine the differential operators that interact with the complex type decomposition. The main point is that the de Rham operator $d$ splits into two first-order pieces, while the metric supplies formal adjoints, Hodge star operators, and Laplacians. ## Type Decomposition and the Operators $\partial$, $\bar\partial$, and $d^c$ The first problem is to describe differential forms on a complex manifold in a way that remembers the complex structure. On a real manifold every $k$-form has only one degree, but on a complex manifold the complexified cotangent bundle separates into holomorphic and antiholomorphic directions. The operators $\partial$ and $\bar\partial$ are the components of exterior differentiation that increase these two degrees separately. [definition: Forms of Type P Q] Let $X$ be a complex manifold of complex dimension $n$. The bundle of complex-valued $k$-forms decomposes as \begin{align*} \Lambda^k T^*X \otimes_{\mathbb R} \mathbb C = \bigoplus_{p+q=k} \Lambda^{p,q}X, \end{align*} where $\Lambda^{p,q}X := \Lambda^p (T^{1,0}X)^* \otimes \Lambda^q (T^{0,1}X)^*$. The space of smooth $(p,q)$-forms is denoted $A^{p,q}(X) := C^\infty(X,\Lambda^{p,q}X)$, and the space of smooth complex-valued $k$-forms is denoted $A^k(X) := C^\infty(X,\Lambda^kT^*X\otimes_{\mathbb R}\mathbb C)$. [/definition] This definition gives a bookkeeping system for the two complex directions. In a holomorphic coordinate chart $(U,z_1,\dots,z_n)$, a local basis of $A^{p,q}(U)$ is built from wedges $dz_{i_1}\wedge\cdots\wedge dz_{i_p}\wedge d\bar z_{j_1}\wedge\cdots\wedge d\bar z_{j_q}$, so the pair $(p,q)$ records how many holomorphic and antiholomorphic differentials occur. [example: Type Decomposition on Complex Euclidean Space] On $\mathbb C^2$ with coordinates $(z_1,z_2)$, the one-forms $dz_1,dz_2$ have type $(1,0)$ and the one-forms $d\bar z_1,d\bar z_2$ have type $(0,1)$. Therefore $dz_1\wedge d\bar z_2$ has type $(1,1)$, because it contains one holomorphic differential and one antiholomorphic differential, while $d\bar z_1\wedge d\bar z_2$ has type $(0,2)$, because it contains no holomorphic differentials and two antiholomorphic differentials. Thus the form \begin{align*} \alpha = z_1\,dz_1\wedge d\bar z_2 + \bar z_2\,d\bar z_1\wedge d\bar z_2 \end{align*} decomposes into its type components as \begin{align*} \alpha^{1,1}=z_1\,dz_1\wedge d\bar z_2 \end{align*} and \begin{align*} \alpha^{0,2}=\bar z_2\,d\bar z_1\wedge d\bar z_2. \end{align*} Multiplying by the coefficient functions $z_1$ and $\bar z_2$ does not change the wedge type, so these are the only nonzero type components of $\alpha$. Both summands have total degree $2$, since $1+1=2$ and $0+2=2$, but they lie in different bidegree summands; this is why type decomposition contains more information than ordinary real degree decomposition. [/example] The example separates forms by type, but it does not yet say whether the usual exterior derivative respects this separation. Since $d$ raises total degree by one, it could in principle produce several types. Integrability of the complex structure restricts the possibilities to exactly two adjacent types, which is what makes Dolbeault theory possible. [quotetheorem:7004] [citeproof:7004] This theorem turns $d$ into two operators adapted to complex geometry. The holomorphic-coordinate hypothesis is essential. On $\mathbb R^4$ with standard complex coordinates $(z_1,z_2)$, declare an almost complex structure by taking a local $(1,0)$-coframe \begin{align*} \theta_1=dz_1, \qquad \theta_2=dz_2+\bar z_1\,d\bar z_2. \end{align*} Then \begin{align*} d\theta_2=d\bar z_1\wedge d\bar z_2, \end{align*} and at the origin this is $\bar\theta_1\wedge\bar\theta_2$, a nonzero $(0,2)$ component. Thus exterior differentiation of a $(1,0)$-form can leave the two adjacent bidegrees when the almost complex structure is not integrable. The theorem does not say that either component is square-zero; it only identifies the possible bidegrees of $d\alpha$. The next issue is whether these two pieces behave like differentials, since cohomology requires a square-zero operator. The condition $d^2=0$ splits by type and gives the algebraic relations needed for the Dolbeault complex. [quotetheorem:8046] [citeproof:8046] The Dolbeault relations explain the complex-linear pieces of $d$. These identities use both integrability and the directness of the type decomposition. The almost complex coframe above gives a concrete failure: solving \begin{align*} \theta_2=dz_2+\bar z_1\,d\bar z_2,\qquad \bar\theta_2=d\bar z_2+z_1\,dz_2 \end{align*} shows that the $(0,1)$ part of $dz_2$ is $-\bar z_1(1-|z_1|^2)^{-1}\bar\theta_2$, so the projected operator satisfies \begin{align*} (\bar\partial^2 z_2)_0=-\bar\theta_1\wedge\bar\theta_2 \end{align*} at the origin. Thus the projection of $d$ to the nominal $\bar\partial$ direction need not be square-zero; the Nijenhuis-tensor component supplies the missing term in the full cancellation $d^2=0$. The relations also do not make $\partial$ and $\bar\partial$ commute; instead they anticommute, which is the sign convention that makes the total differential square to zero. Kähler geometry also needs a real first-order operator that converts a local potential into a real $(1,1)$-form after applying $d$ again. That role is played by $d^c$, whose exact sign convention must be fixed before formulas involving $dd^c$ are meaningful. [definition: The Operator D C] Let $X$ be a complex manifold. The operator \begin{align*} d^c:A^k(X)\to A^{k+1}(X) \end{align*} is defined on complex-valued $k$-forms by \begin{align*} d^c := i(\bar\partial-\partial). \end{align*} [/definition] With this convention, a real-valued smooth function $\varphi$ satisfies $d^c\varphi=i(\bar\partial\varphi-\partial\varphi)$. The real value of the definition is the second-order expression $dd^c\varphi$, because local Kähler potentials are detected by a $(1,1)$-form. We therefore compute $dd^c$ in terms of the Dolbeault operators. [quotetheorem:8048] [citeproof:8048] The formula makes the local potential computation concrete. The sign and factor of $2$ depend on the convention for $d^c$; with the alternative convention $d^c=(i/2)(\bar\partial-\partial)$, the same calculation would give $dd^c=i\partial\bar\partial$. Integrability is also part of the statement: on the almost complex model with $\theta_2=dz_2+\bar z_1\,d\bar z_2$, the extra $(0,2)$ component of $d\theta_2$ contributes torsion terms, so the computation of $dd^c$ cannot be reduced to only $\partial^2=\bar\partial^2=0$ and $\bar\partial\partial=-\partial\bar\partial$. The theorem does not assert that every closed real $(1,1)$-form has a global potential. For instance, the Fubini--Study form $\omega_{\mathrm{FS}}$ on $\mathbb P^1$ is closed and of type $(1,1)$, but it cannot be $dd^c\varphi$ for a globally defined smooth function $\varphi$, because $\int_{\mathbb P^1}dd^c\varphi=0$ by Stokes' theorem while $\int_{\mathbb P^1}\omega_{\mathrm{FS}}>0$. To keep the normalization anchored, we compute the operators on $\mathbb C^n$, where all coefficients and coordinate formulas can be written explicitly. [example: Basic Operators on Complex Euclidean Space] Let $f\in C^\infty(\mathbb C^n)$ and write $z_j=x_j+iy_j$. The type decomposition of $df$ gives \begin{align*}\partial f = \sum_{j=1}^n \frac{\partial f}{\partial z_j}\,dz_j, \qquad \bar\partial f = \sum_{j=1}^n \frac{\partial f}{\partial \bar z_j}\,d\bar z_j,\end{align*} where \begin{align*}\frac{\partial}{\partial z_j} = \frac{1}{2}\left(\frac{\partial}{\partial x_j}-i\frac{\partial}{\partial y_j}\right), \qquad \frac{\partial}{\partial \bar z_j} = \frac{1}{2}\left(\frac{\partial}{\partial x_j}+i\frac{\partial}{\partial y_j}\right).\end{align*} For $f(z)=|z_1|^2=x_1^2+y_1^2=z_1\bar z_1$, the only nonzero derivatives occur in the first coordinate. Using the displayed formulas, \begin{align*}\frac{\partial f}{\partial z_1}=\frac{1}{2}(2x_1-i\,2y_1)=x_1-iy_1=\bar z_1,\end{align*} and \begin{align*}\frac{\partial f}{\partial \bar z_1}=\frac{1}{2}(2x_1+i\,2y_1)=x_1+iy_1=z_1.\end{align*} For $j>1$, $f$ is independent of $x_j$ and $y_j$, so $\partial f/\partial z_j=0$ and $\partial f/\partial \bar z_j=0$. Therefore \begin{align*}\partial f=\bar z_1\,dz_1,\qquad \bar\partial f=z_1\,d\bar z_1.\end{align*} Next, \begin{align*}\partial\bar\partial f=\partial(z_1\,d\bar z_1)=\partial z_1\wedge d\bar z_1+z_1\,\partial(d\bar z_1).\end{align*} Since $\partial z_1=dz_1$ and $d(d\bar z_1)=0$, the second term vanishes, so \begin{align*}\partial\bar\partial f=dz_1\wedge d\bar z_1.\end{align*} By the local formula relating $dd^c$ to $i\partial\bar\partial$, with the convention $d^c=i(\bar\partial-\partial)$, \begin{align*}dd^c f=2i\,\partial\bar\partial f=2i\,dz_1\wedge d\bar z_1.\end{align*} This is the local model for the formula that recovers a Kähler form from a potential by applying $dd^c$. [/example] ## Formal Adjoints and the Natural Laplacians The next problem is analytic: once a Hermitian metric is present, differential operators should have adjoints and Laplacians. These objects are defined through integration, so compactness or boundary conditions matter: on a domain with boundary, Stokes' theorem produces boundary integrals unless boundary conditions are imposed. For example, [integration by parts](/theorems/210) on an interval includes endpoint terms, so the same differential expression can have different adjoints depending on whether functions vanish at the boundary. In this course we mainly use compact Kähler manifolds without boundary, where [integration by parts](/theorems/2098) has no boundary contribution. [definition: Inner Product on Complex Forms] Let $(X,g)$ be a compact Hermitian manifold of complex dimension $n$, with volume form $dV_g$. The global Hermitian [inner product](/page/Inner%20Product) on complex-valued $k$-forms is the map \begin{align*} (\cdot,\cdot)_{L^2}:A^k(X)\times A^k(X)\to \mathbb C \end{align*} defined by \begin{align*} (\alpha,\beta)_{L^2}:=\int_X (\alpha,\beta)_g\,dV_g, \end{align*} where $(\alpha,\beta)_g$ is the pointwise Hermitian metric induced by $g$ on $\Lambda^kT^*X\otimes_{\mathbb R}\mathbb C$. [/definition] The preceding definition turns analytic questions about forms into Hilbert-space questions. Once an inner product has been fixed, the next problem is to express integration by parts as an operator identity. This leads to the formal adjoint, the metric-dependent partner that moves a differential operator from one factor of the pairing to the other. [definition: Formal Adjoint] Let $D:A^r(X)\to A^s(X)$ be a linear differential operator on a compact Hermitian manifold $(X,g)$. Its formal adjoint is the operator $D^*:A^s(X)\to A^r(X)$ satisfying \begin{align*} (D\alpha,\beta)_{L^2}=(\alpha,D^*\beta)_{L^2} \end{align*} for all smooth forms $\alpha\in A^r(X)$ and $\beta\in A^s(X)$. [/definition] The preceding definition explains how $d$, $\partial$, and $\bar\partial$ acquire degree-lowering partners: $\partial^*:A^{p,q}(X)\to A^{p-1,q}(X)$, $\bar\partial^*:A^{p,q}(X)\to A^{p,q-1}(X)$, and $d^*:A^k(X)\to A^{k-1}(X)$. Closedness alone is not enough to select canonical representatives of cohomology classes, since replacing a closed form by itself plus an exact form gives the same class but usually changes its metric size. The next construction combines each differential with its adjoint to form a second-order self-adjoint operator. These Laplacians are the operators whose kernels will later represent cohomology classes. [definition: Natural Laplacians] On a compact Hermitian manifold, the de Rham Laplacian is the operator \begin{align*} \Delta_d:A^k(X)\to A^k(X) \end{align*} defined by \begin{align*} \Delta_d := dd^*+d^*d. \end{align*} The Dolbeault Laplacians are the operators \begin{align*} \Delta_\partial:A^{p,q}(X)\to A^{p,q}(X), \qquad \Delta_{\bar\partial}:A^{p,q}(X)\to A^{p,q}(X) \end{align*} defined by \begin{align*} \Delta_\partial := \partial\partial^*+\partial^*\partial, \qquad \Delta_{\bar\partial} := \bar\partial\bar\partial^*+\bar\partial^*\bar\partial. \end{align*} [/definition] The preceding definition packages first-order closedness and co-closedness into a single second-order equation. To use these Laplacians in cohomology, we need to know that being annihilated by a Laplacian is the same as satisfying the two first-order equations separately. The following criterion is the [basic energy identity](/theorems/3672) behind harmonic representatives. [quotetheorem:8049] [citeproof:8049] The criterion turns harmonicity into equations that can be checked by integration by parts. Compactness without boundary is essential for this clean energy identity. On the interval $[0,1]$ with the Euclidean metric, the differential expression $\Delta f=-f''$ annihilates $f(x)=x$, but $df=dx\ne 0$; the missing term is the boundary contribution in integration by parts. Imposing Dirichlet, Neumann, absolute, or relative boundary conditions changes the domain of the adjoint and changes the harmonic problem. Thus the theorem does not say that the same criterion holds on manifolds with boundary without specifying boundary conditions, and it also does not by itself identify harmonic forms with cohomology classes. Hodge decomposition will supply that identification on the compact boundaryless manifolds used in this course. The flat torus gives a useful model because constant coefficient forms have no derivative terms, so the adjoint formulas collapse to their simplest form. [example: Adjoints on a Flat Complex Torus] Let $X=\mathbb C^n/\Lambda$ carry the flat Hermitian metric inherited from $\mathbb C^n$, and let $\alpha=\sum_{j=1}^n a_jdz_j$ be a smooth periodic $(1,0)$-form. To identify $\partial^*\alpha$, test against a smooth periodic function $u$. Since $\partial u=\sum_{j=1}^n (\partial u/\partial z_j)dz_j$ and the coframe $dz_1,\dots,dz_n$ is orthonormal up to the fixed flat normalization, the $L^2$ pairing has the form \begin{align*} (\partial u,\alpha)_{L^2}=\int_X \sum_{j=1}^n \frac{\partial u}{\partial z_j}\,\overline{a_j}\,dV. \end{align*} For each $j$, periodicity removes the boundary term in integration by parts on a fundamental parallelepiped for $\Lambda$, so \begin{align*} \int_X \frac{\partial u}{\partial z_j}\,\overline{a_j}\,dV=-\int_X u\,\frac{\partial \overline{a_j}}{\partial z_j}\,dV. \end{align*} Because $\partial \overline{a_j}/\partial z_j=\overline{\partial a_j/\partial \bar z_j}$, this becomes \begin{align*} \int_X \frac{\partial u}{\partial z_j}\,\overline{a_j}\,dV=\int_X u\,\overline{-\frac{\partial a_j}{\partial \bar z_j}}\,dV. \end{align*} Summing over $j$ and comparing with $(u,\partial^*\alpha)_{L^2}$ gives \begin{align*} \partial^*\alpha=-\sum_{j=1}^n \frac{\partial a_j}{\partial \bar z_j}. \end{align*} If every coefficient $a_j$ is constant, then $\partial a_j/\partial \bar z_j=0$ for all $j$, so $\partial^*\alpha=0$. Also, \begin{align*} \partial\alpha=\sum_{j=1}^n \partial(a_jdz_j)=\sum_{j=1}^n \partial a_j\wedge dz_j+\sum_{j=1}^n a_j\,\partial(dz_j). \end{align*} The first sum is zero because the $a_j$ are constant, and the second sum is zero because $d(dz_j)=0$ and hence the $(2,0)$-part $\partial(dz_j)$ is zero. Thus $\partial\alpha=0$. By *Harmonic Criterion for the Natural Laplacians*, constant $(1,0)$-forms are $\Delta_\partial$-harmonic. Similarly, for a smooth periodic $(0,1)$-form $\beta=\sum_{j=1}^n b_jd\bar z_j$, testing against functions gives \begin{align*} \bar\partial^*\beta=-\sum_{j=1}^n \frac{\partial b_j}{\partial z_j}. \end{align*} When the $b_j$ are constant, this formula gives $\bar\partial^*\beta=0$, and the same expansion \begin{align*} \bar\partial\beta=\sum_{j=1}^n \bar\partial b_j\wedge d\bar z_j+\sum_{j=1}^n b_j\,\bar\partial(d\bar z_j) \end{align*} has both sums equal to zero. Therefore constant $(0,1)$-forms are $\Delta_{\bar\partial}$-harmonic. [/example] ## The Hodge Star on Complex Forms and Conjugation Symmetries The final problem in this chapter is to relate the metric inner product to wedge products and complex conjugation. The Hodge star is the bridge between pointwise metric data and integration. On a complex manifold it also interacts with the decomposition into $(p,q)$-types. [definition: Hodge Star on Complex Forms] Let $(X,g)$ be an oriented Riemannian manifold of real dimension $m$. The Hodge star is the complex-linear operator \begin{align*} *:A^k(X)\to A^{m-k}(X) \end{align*} defined by \begin{align*} \alpha\wedge *\overline{\beta}=(\alpha,\beta)_g\,dV_g \end{align*} for complex-valued $k$-forms $\alpha$ and $\beta$, where the pointwise Hermitian pairing $(\cdot,\cdot)_g$ is linear in the first variable and conjugate-linear in the second. [/definition] The complex conjugate appears because the pointwise inner product is Hermitian rather than bilinear. This convention makes the global formula $(\alpha,\beta)_{L^2}=\int_X \alpha\wedge *\overline{\beta}$ consistent with the $L^2$ inner product used for adjoints. Since the star is defined by complementing a form to the volume form, the next question is how it changes bidegree. [quotetheorem:8050] [citeproof:8050] The preceding theorem describes how the Hodge star reverses bidegree through the metric and orientation. The Hermitian hypothesis is what makes the type statement stable under unitary coframes. A concrete failure occurs already on $\mathbb C$ with its standard complex structure but with the non-Hermitian Riemannian metric $g=dx^2+4\,dy^2$. With the orientation $dx\wedge dy$, one has $*dx=2\,dy$ and $*dy=-\frac{1}{2}dx$, hence \begin{align*} *dz=-\frac{5i}{4}\,dz+\frac{3i}{4}\,d\bar z. \end{align*} The star of a $(1,0)$-form has a $(0,1)$ component, so the stated type rule fails when the metric is not Hermitian. The theorem does not compute the signs or constants in the star of a basis form, only the resulting bidegree. A second symmetry comes from complex conjugation, which swaps $A^{p,q}(X)$ with $A^{q,p}(X)$. We next record that this swap preserves the natural metric size of forms, since later Hodge-theoretic arguments compare conjugate bidegrees. [quotetheorem:8051] [citeproof:8051] The metric symmetry has an operator-level counterpart. The reality of the volume form and Hermitian symmetry of the pointwise pairing are essential here. If the fibre pairing on the complex line were the complex-bilinear product $B(u,v)=uv$, then with $u=1+i$ and $v=1$ one would get \begin{align*} B(\bar u,\bar v)=1-i,\qquad B(v,u)=1+i, \end{align*} so conjugation would not reverse the two entries as in the theorem. The theorem does not say that conjugation preserves bidegree, since it sends $A^{p,q}(X)$ to $A^{q,p}(X)$. Since conjugation swaps holomorphic and antiholomorphic variables, it also swaps the two Dolbeault directions. This gives a useful check on computations involving $\partial$ and $\bar\partial$. [remark: Conjugate Operators] Complex conjugation exchanges the two Dolbeault directions: \begin{align*} \overline{\partial\alpha}=\bar\partial\overline{\alpha}, \qquad \overline{\bar\partial\alpha}=\partial\overline{\alpha}. \end{align*} Thus conjugation sends $(p,q)$-calculations to $(q,p)$-calculations. Later, when the Kähler identities relate $\Delta_d$, $\Delta_\partial$, and $\Delta_{\bar\partial}$, this symmetry will help transfer statements between holomorphic and antiholomorphic degrees. [/remark] The remark identifies the abstract symmetry; a low-dimensional computation shows how the Hodge star implements the same distinction in coordinates. Complex dimension one is the smallest case where the signs can be checked by hand. [example: Hodge Star in Complex Dimension One] On $\mathbb C$ with the standard metric $g=dx^2+dy^2$ and orientation $dx\wedge dy$, the Hodge star on real one-forms is determined by the identities \begin{align*} dx\wedge *dx=(dx,dx)_g\,dx\wedge dy=dx\wedge dy \end{align*} and \begin{align*} dy\wedge *dy=(dy,dy)_g\,dx\wedge dy=dx\wedge dy. \end{align*} Thus $*dx=dy$, while $*dy=-dx$ because $dy\wedge(-dx)=dx\wedge dy$. Since the Hodge star is complex-linear on complex-valued forms, \begin{align*} *dz=*(dx+i\,dy)=*dx+i*dy=dy-i\,dx=-i(dx+i\,dy)=-i\,dz. \end{align*} Similarly, \begin{align*} *d\bar z=*(dx-i\,dy)=*dx-i*dy=dy+i\,dx=i(dx-i\,dy)=i\,d\bar z. \end{align*} Therefore the line spanned by $dz$ is preserved and the line spanned by $d\bar z$ is preserved. In complex dimension one, the type rule from *Hodge Star Changes Type* sends $(p,q)$ to $(1-q,1-p)$, so $(1,0)$ maps to $(1,0)$ and $(0,1)$ maps to $(0,1)$, exactly as the coordinate computation shows. [/example] The chapter has separated the algebraic splitting of $d$ from the metric operations supplied by a Hermitian structure. Chapter 3 first studies the Lefschetz operators pointwise as linear algebra; Chapter 4 then uses the Kähler condition to relate those operators to the adjoints of $\partial$ and $\bar\partial$, making the three Laplacians scalar multiples of one another. # 3. Lefschetz Operators and Linear Algebra This chapter is the linear algebra core behind the global Lefschetz theory of Kähler manifolds. In the preceding chapter, forms were decomposed by type and equipped with adjoints and Laplacians; now the Kähler form enters as a pointwise Hermitian linear-algebra object. We now freeze the manifold at a point and study the exterior algebra of a Hermitian [vector space](/page/Vector%20Space), where wedging with the Kähler form, contracting with it, and counting degree form an $\mathfrak{sl}_2$-triple. The pointwise results proved here will later be applied fibrewise to differential forms on a compact Kähler manifold. ## Lefschetz Operators on a Hermitian Vector Space The first problem is to understand what the Kähler form does to forms before any analysis or topology enters. On a Hermitian vector space, wedging by the Kähler form raises degree by two, while its adjoint lowers degree by two. The commutator of these two operations measures how far a form sits from the middle dimension. Let $(V,h)$ be a Hermitian vector space of complex dimension $n$, viewed as a real vector space of dimension $2n$, and let $\omega$ be its associated real $(1,1)$-form. Choose a unitary coframe $\theta_1,\dots,\theta_n$ of $V^*$, so that \begin{align*} \omega = i\sum_{j=1}^n \theta_j \wedge \overline{\theta_j}. \end{align*} The following operators isolate the three basic actions of $\omega$ on the exterior algebra. [definition: Lefschetz Operators] Let $(V,h)$ be a Hermitian vector space of complex dimension $n$ with Kähler form $\omega$. The Lefschetz operator is the [linear map](/page/Linear%20Map) \begin{align*} L : \Lambda^k V^* \to \Lambda^{k+2}V^*, \qquad L\alpha = \omega \wedge \alpha. \end{align*} The adjoint Lefschetz operator is the adjoint \begin{align*} \Lambda : \Lambda^k V^* \to \Lambda^{k-2}V^* \end{align*} of $L$ with respect to the Hermitian inner product on forms induced by $h$. The grading operator is the degree-preserving linear map \begin{align*} H : \Lambda^*V^* \to \Lambda^*V^*, \qquad H = [\Lambda,L] = \Lambda L - L\Lambda, \end{align*} so that $H$ restricts to a map $H:\Lambda^kV^*\to\Lambda^kV^*$ for each $k$. [/definition] The operator $L$ is visible from the formula for $\omega$, while $\Lambda$ depends on the metric through the adjoint operation. The grading operator records the degree shift hidden in the pair $L,\Lambda$. A naive attempt to study only powers of $L$ loses information, because wedging by $\omega$ cannot distinguish forms that already lie in the image of a previous wedge from forms that start a new string; the adjoint $\Lambda$ is what detects that distinction. [example: Operators in a Unitary Coframe] On $\mathbb C^n$ with standard unitary coframe $dz_1,\dots,dz_n$, put \begin{align*}\eta_j=i\,dz_j\wedge d\overline z_j.\end{align*} Then \begin{align*}\omega=\eta_1+\cdots+\eta_n,\end{align*} so $L\alpha=\omega\wedge\alpha=\sum_{j=1}^n\eta_j\wedge\alpha$. The adjoint $\Lambda$ is the sum of the adjoints of the maps $\alpha\mapsto \eta_j\wedge\alpha$. Hence $\Lambda$ removes one existing factor $\eta_j$ and gives $0$ when no such matching $dz_j\wedge d\overline z_j$ factor is present, with the sign coming from moving the contraction through the preceding wedge factors. For $n=2$, write $\omega=\eta_1+\eta_2$. Since $\Lambda\eta_1=1$ and $\Lambda\eta_2=1$, we get \begin{align*}\Lambda\omega=\Lambda\eta_1+\Lambda\eta_2=1+1=2.\end{align*} Also $\Lambda 1=0$, so \begin{align*}H(1)=\Lambda L(1)-L\Lambda(1)=\Lambda\omega-0=2.\end{align*} Next, $\eta_1^2=\eta_2^2=0$, and $\eta_1\wedge\eta_2=\eta_2\wedge\eta_1$ because both factors have degree $2$. Therefore \begin{align*}\omega^2=(\eta_1+\eta_2)\wedge(\eta_1+\eta_2)=2\,\eta_1\wedge\eta_2.\end{align*} Since $\Lambda(\eta_1\wedge\eta_2)=\eta_2+\eta_1=\omega$, this gives \begin{align*}\Lambda(\omega^2)=2\omega.\end{align*} Thus \begin{align*}H(\omega)=\Lambda L(\omega)-L\Lambda(\omega)=\Lambda(\omega^2)-L(2)=2\omega-2\omega=0.\end{align*} Finally, in complex dimension $2$ there are no nonzero forms of degree $6$, so $L(\omega^2)=\omega^3=0$. Hence \begin{align*}H(\omega^2)=\Lambda L(\omega^2)-L\Lambda(\omega^2)=0-L(2\omega)=-2\omega^2.\end{align*} These computations show concretely that, under the convention $H=[\Lambda,L]$, the weights in degrees $0,2,4$ are $2,0,-2$. [/example] This example indicates that $H$ is diagonal on homogeneous forms. To use $L$, $\Lambda$, and $H$ systematically, one needs the exact weight on every degree, not just in the standard low-dimensional example. The point to check is that the commutator depends only on the Hermitian linear algebra and not on the particular coordinates used to compute it. [quotetheorem:8053] [citeproof:8053] The Hermitian hypothesis is essential here because $\Lambda$ is defined as the metric adjoint of wedging by the nondegenerate Kähler form; if $\omega$ is replaced by the zero $2$-form, then $L=0$, $\Lambda=0$, and $H$ cannot act by $n-k$ on $\Lambda^kV^*$. The theorem is only a pointwise linear-algebra statement: it does not yet say that the same operators commute with exterior derivatives, Laplacians, or harmonic projections on a manifold. What it supplies is the representation-theoretic language in which the later global Kähler identities will be interpreted. The sign convention for $H$ varies across books: some authors use $[L,\Lambda]$, which reverses the signs in the displayed relations. In these notes $H=[\Lambda,L]$, so highest-weight vectors will be low-degree primitive forms. [remark: Degree Convention] With the convention $H=[\Lambda,L]$, a $k$-form has weight $n-k$. Thus $L$ lowers the $H$-weight by $2$ and $\Lambda$ raises it by $2$. This is the same representation theory as the usual $\mathfrak{sl}_2$ convention after replacing $H$ by $-H$. [/remark] The commutator calculation gives a pointwise structure, not yet a decomposition. To obtain the decomposition, we must identify the forms that begin each $\mathfrak{sl}_2$ string. ## Primitive Forms and Lefschetz Decomposition The next problem is to separate the part of a form that is created by wedging with $\omega$ from the part that is not. The forms not obtained from lower degree by applying $L$ are called primitive. They are the building blocks for the entire exterior algebra. Without this separation, a form such as $\omega\wedge\gamma$ and a genuinely new form of the same degree would be treated alike, so repeated applications of $L$ would obscure where each Lefschetz string begins. [definition: Primitive Form] Let $(V,h)$ be a Hermitian vector space of complex dimension $n$. A $k$-form $\alpha \in \Lambda^kV^*$ with $0\le k\le n$ is primitive if \begin{align*} L^{n-k+1}\alpha = 0. \end{align*} The space of primitive $k$-forms is denoted $P^k\subseteq \Lambda^kV^*$. [/definition] The definition is designed so that a primitive $k$-form can be wedged with $\omega$ exactly up to the middle degree and no further. As stated, however, it tests primitivity by applying a high power of $L$, which is awkward in computations. The lowering operator $\Lambda$ detects the same starting vectors of Lefschetz strings, and the full structure is the [primitive Lefschetz decomposition](/theorems/3855). [quotetheorem:3855] [citeproof:3855] The restriction $0\le k\le n$ is part of the content: beyond the middle dimension, the relevant primitive data have already appeared in lower degree and been pushed upward by $L$. For instance, in complex dimension $1$, a top-degree form is not primitive in this sense, since it is $L$ applied to a scalar and is not killed by $\Lambda$. This local test for the starting vectors of the Lefschetz strings is what makes the decomposition canonical rather than a choice of complements. [example: Primitive Two-Forms on Complex Dimension Two] Let $V=\mathbb C^2$ with standard Kähler form \begin{align*} \omega=i\,dz_1\wedge d\overline z_1+i\,dz_2\wedge d\overline z_2. \end{align*} Write $\eta_1=i\,dz_1\wedge d\overline z_1$ and $\eta_2=i\,dz_2\wedge d\overline z_2$, so $\omega=\eta_1+\eta_2$. Since $\Lambda$ is the adjoint of wedging by $\omega$, it removes a matching $\eta_j$ factor: $\Lambda\eta_1=1$, $\Lambda\eta_2=1$, and $\Lambda(dz_1\wedge d\overline z_2)=\Lambda(dz_2\wedge d\overline z_1)=0$ because neither mixed form contains both $dz_j$ and $d\overline z_j$ for the same $j$. For a $(1,1)$-form \begin{align*} \alpha=i(a\,dz_1\wedge d\overline z_1+b\,dz_2\wedge d\overline z_2)+c\,dz_1\wedge d\overline z_2+d\,dz_2\wedge d\overline z_1, \end{align*} this gives \begin{align*} \Lambda\alpha=a\Lambda\eta_1+b\Lambda\eta_2+c\Lambda(dz_1\wedge d\overline z_2)+d\Lambda(dz_2\wedge d\overline z_1)=a+b. \end{align*} Subtract the scalar part \begin{align*} \alpha_0=\alpha-\frac{a+b}{2}\omega. \end{align*} Using $\omega=\eta_1+\eta_2$, the diagonal terms become \begin{align*} a\eta_1+b\eta_2-\frac{a+b}{2}(\eta_1+\eta_2)=\frac{a-b}{2}\eta_1+\frac{b-a}{2}\eta_2. \end{align*} Therefore \begin{align*} \Lambda\alpha_0=\frac{a-b}{2}\Lambda\eta_1+\frac{b-a}{2}\Lambda\eta_2+0=\frac{a-b}{2}+\frac{b-a}{2}=0. \end{align*} Equivalently, in complex dimension $2$ the primitive $(1,1)$ part is exactly the trace-zero part: the coefficient of $\eta_1$ plus the coefficient of $\eta_2$ is $0$, while the multiple $\frac{a+b}{2}\omega$ is the scalar trace component. [/example] The same idea works in all degrees: first remove the part divisible by $\omega$, then repeat. The obstruction is that a general complement to $L\Lambda^{k-2}V^*$ would depend on arbitrary choices and would not interact well with the metric. The primitive condition removes exactly this ambiguity, because it selects the complement using the adjoint operator $\Lambda$. At this point we only need the pointwise Lefschetz linear algebra: every form is organized into strings generated by wedging primitive forms with powers of $\omega$. The nondegeneracy of the Hermitian Kähler form is essential. If the $2$-form used to define $L$ were degenerate or zero, powers of $L$ would not generate symmetric strings through the middle degree, and the direct-sum pattern would fail even dimensionally. The pointwise primitive splitting is the algebraic mechanism that later global arguments will use; the analytic facts about harmonic forms and cohomology require compactness, the Kähler identities, and Hodge theory, so they are deferred until those tools have been installed. [example: Decomposing Forms on Complex Dimension Three] Let $V=\mathbb C^3$ with standard Kähler form $\omega$, and write $L\gamma=\omega\wedge\gamma$. For any $2$-form $\alpha$, set \begin{align*} c=\frac{1}{3}\Lambda\alpha. \end{align*} Since $\Lambda\omega=3$, the form \begin{align*} \alpha_0=\alpha-c\omega \end{align*} satisfies \begin{align*} \Lambda\alpha_0=\Lambda\alpha-c\Lambda\omega=\Lambda\alpha-\frac{1}{3}(\Lambda\alpha)3=0. \end{align*} Thus $\alpha_0$ is primitive by the criterion characterizing primitive classes by Lefschetz powers, and \begin{align*} \alpha=\alpha_0+\frac{1}{3}(\Lambda\alpha)\omega \end{align*} is the decomposition of a $2$-form into its primitive part and its scalar Lefschetz part. For a $4$-form $\beta$, the Lefschetz decomposition has only the two possible summands $LP^2$ and $L^2P^0$, so we write \begin{align*} \beta=L\beta_0+c\omega^2 \end{align*} with $\beta_0\in P^2$. The coefficient $c$ is determined by applying $\Lambda^2$. If $\beta_0$ is primitive of degree $2$, the standard Lefschetz string formula gives $\Lambda L\beta_0=\beta_0$, so \begin{align*} \Lambda^2L\beta_0=\Lambda\beta_0=0. \end{align*} For the scalar string generated by $1$, the same formula gives \begin{align*} \Lambda\omega^2=4\omega. \end{align*} Applying $\Lambda$ once more gives \begin{align*} \Lambda^2\omega^2=\Lambda(4\omega)=4\Lambda\omega=4\cdot 3=12. \end{align*} Therefore \begin{align*} \Lambda^2\beta=\Lambda^2(L\beta_0+c\omega^2)=0+12c, \end{align*} so \begin{align*} c=\frac{1}{12}\Lambda^2\beta. \end{align*} Then \begin{align*} \Lambda\beta=\Lambda L\beta_0+c\Lambda\omega^2=\beta_0+4c\omega, \end{align*} hence \begin{align*} \beta_0=\Lambda\beta-\frac{1}{3}(\Lambda^2\beta)\omega. \end{align*} Indeed, \begin{align*} \Lambda\beta_0=\Lambda^2\beta-\frac{1}{3}(\Lambda^2\beta)\Lambda\omega=\Lambda^2\beta-\frac{1}{3}(\Lambda^2\beta)3=0, \end{align*} so this $\beta_0$ is primitive. In coordinates, the $(1,1)$ part of a $2$-form decomposes into a trace-zero Hermitian matrix plus the scalar multiple of $\omega$ because $\Lambda$ takes the trace. The $(2,0)$ and $(0,2)$ parts are already primitive: $\Lambda$ has type $(-1,-1)$, so it sends them to nonexistent bidegrees $\Lambda^{1,-1}$ and $\Lambda^{-1,1}$, hence to $0$. [/example] The decomposition also respects the complex type decomposition, because $\omega$ has type $(1,1)$ and $\Lambda$ has type $(-1,-1)$. This refinement is needed for Hodge theory, where the sign and positivity statements depend on both total degree and bidegree. [remark: Type Decomposition] For $\alpha\in\Lambda^{p,q}V^*$, the operators satisfy $L\alpha \in \Lambda^{p+1,q+1}V^*$ and $\Lambda\alpha \in \Lambda^{p-1,q-1}V^*$. Therefore \begin{align*} P^k = \bigoplus_{p+q=k} P^{p,q}, \qquad P^{p,q}=P^k\cap\Lambda^{p,q}V^*. \end{align*} [/remark] Primitive decomposition gives a basis adapted to $L$ and $\Lambda$. The final section adds the Hermitian form whose signs on these primitive pieces encode the Hodge-Riemann relations. ## Pointwise Hodge-Riemann Bilinear Relations The remaining problem is to understand the metric sign hidden in the Lefschetz decomposition. The ordinary Hermitian inner product on forms is positive definite, but the bilinear expression that later descends to cohomology has signs depending on bidegree. The Hodge-Riemann relations identify exactly where positivity appears after inserting the correct phase factor. For $\alpha,\beta\in\Lambda^{p,q}V^*$ with $p+q=k$, the wedge product $\alpha\wedge\overline\beta\wedge\omega^{n-k}$ has top degree. Comparing it to the volume form gives a scalar pairing. On primitive forms, this scalar pairing is the pointwise model of the global Hodge-Riemann form. [definition: Pointwise Hodge-Riemann Form] Let $(V,h)$ be a Hermitian vector space of complex dimension $n$, and let $k=p+q$. The pointwise Hodge-Riemann Hermitian form is the map \begin{align*} Q : P^{p,q}\times P^{p,q}\to \mathbb C \end{align*} defined by \begin{align*} Q(\alpha,\beta)=i^{p-q}(-1)^{k(k-1)/2}\,\frac{1}{(n-k)!}\,\alpha\wedge\overline{\beta}\wedge\omega^{n-k}, \end{align*} where the top-degree form is identified with the scalar multiplying the positive volume form $\omega^n/n!$. [/definition] The phase $i^{p-q}$ and the sign $(-1)^{k(k-1)/2}$ are chosen so that the wedge pairing can be compared with the positive Hermitian norm. Without these factors the same wedge product has alternating signs; for example primitive $(1,1)$-forms in complex dimension $2$ would give the negative of the desired positive norm. The theorem below is needed because this comparison is not automatic from the definition: it is the pointwise positivity statement that later survives on primitive cohomology. [quotetheorem:8054] [citeproof:8054] Primitivity is essential in the positivity statement: in complex dimension $2$, the nonprimitive form $\omega$ has the opposite sign behaviour from primitive $(1,1)$-forms under the raw wedge pairing. The theorem does not say that the unnormalised wedge product is positive on all forms, nor does it replace the ordinary positive definite Hermitian inner product on the full exterior algebra. Instead, it identifies the corrected sign on each primitive block, which is exactly the datum that can descend to primitive cohomology once harmonic representatives are available. Its global version requires harmonic representatives and the Kähler identities, but the signs and primitive decomposition are already fixed here. [example: Hodge-Riemann Sign for Primitive One-One Forms] In complex dimension $2$, write \begin{align*} \eta_1=i\,dz_1\wedge d\overline z_1 \end{align*} and \begin{align*} \eta_2=i\,dz_2\wedge d\overline z_2. \end{align*} Then $\omega=\eta_1+\eta_2$ and \begin{align*} \alpha=\eta_1-\eta_2. \end{align*} Since $\Lambda\eta_1=1$ and $\Lambda\eta_2=1$, we have \begin{align*} \Lambda\alpha=\Lambda\eta_1-\Lambda\eta_2=1-1=0. \end{align*} Thus $\alpha$ is primitive of type $(1,1)$. The form $\eta_j$ is real, because \begin{align*} \overline{i\,dz_j\wedge d\overline z_j}=(-i)\,d\overline z_j\wedge dz_j=i\,dz_j\wedge d\overline z_j. \end{align*} Hence $\overline{\alpha}=\alpha$. Also $\eta_1^2=\eta_2^2=0$, and $\eta_1\wedge\eta_2=\eta_2\wedge\eta_1$ because both factors have degree $2$. Therefore \begin{align*} \alpha\wedge\overline{\alpha}=\alpha\wedge\alpha=(\eta_1-\eta_2)\wedge(\eta_1-\eta_2)=-2\,\eta_1\wedge\eta_2. \end{align*} On the other hand, \begin{align*} \omega^2=(\eta_1+\eta_2)\wedge(\eta_1+\eta_2)=2\,\eta_1\wedge\eta_2. \end{align*} Thus \begin{align*} -\alpha\wedge\overline{\alpha}=2\,\eta_1\wedge\eta_2=\omega^2=2\frac{\omega^2}{2}. \end{align*} For $p=q=1$ and $k=2$, the Hodge-Riemann sign factor is $i^{p-q}(-1)^{k(k-1)/2}=i^0(-1)^1=-1$, so this computation gives \begin{align*} Q(\alpha,\alpha)=2. \end{align*} Thus the corrected Hodge-Riemann form is positive on this primitive $(1,1)$ form, while the uncorrected wedge product $\alpha\wedge\overline{\alpha}$ has the opposite sign. [/example] The next application is the algebraic version of the hard Lefschetz isomorphism. A naive dimension count is not enough: although $\Lambda^kV^*$ and $\Lambda^{2n-k}V^*$ have the same dimension, it does not identify the canonical map between them, and an arbitrary $2$-form need not make wedging injective. The Lefschetz string decomposition supplies the missing mechanism, because $L^{n-k}$ moves each string to its symmetric partner across the middle degree. This will later be promoted from vector spaces to cohomology groups using harmonic forms. [quotetheorem:8080] [citeproof:8080] The bound $k\le n$ is necessary: for $k>n$ the power $L^{n-k}$ has negative exponent and the correct inverse statement is obtained by applying powers of $\Lambda$ on the other side. The nondegenerate Kähler form is also essential; if $\omega=0$, then $L^{n-k}=0$ for $k<n$, so no isomorphism can result. The theorem does not yet prove global hard Lefschetz on de Rham cohomology, since wedging a representative and passing to cohomology requires compatibility with harmonic theory. It gives the pointwise bridge: the same fibrewise isomorphism will act on differential forms, and the Kähler identities will later show that it preserves harmonic forms. Together, the operators $L,\Lambda,H$, the primitive decomposition, and the Hodge-Riemann form give the complete linear algebra package behind Kähler Hodge theory. In the global setting, these operators act on differential forms by applying the pointwise construction in each cotangent space. The next chapters use the Kähler identities to show that this pointwise representation theory is compatible with the Laplacian and therefore descends to harmonic forms and cohomology. # 4. Kähler Identities Chapters 2 and 3 built the two ingredients needed here: the analytic language of complex differential forms and the pointwise Lefschetz linear algebra of a Hermitian vector space. This chapter is where the condition $d\omega=0$ becomes a system of operator identities. The central point is that the Lefschetz operators $L$ and $\Lambda$ interact with $\partial$, $\bar\partial$, and their adjoints in a way that forces the de Rham, Dolbeault, and $\partial$-Laplacians to carry the same harmonic information. The prerequisites are the decomposition of complex differential forms into $(p,q)$-types, the Dolbeault operators $\partial$ and $\bar\partial$, Hermitian metrics on complex manifolds, formal adjoints with respect to the $L^2$ inner product, and the definition of a Kähler metric. We also use the local normal-coordinate fact for Kähler metrics: at a chosen point, holomorphic coordinates may be chosen so that the Hermitian metric is standard and its first derivatives vanish. The identities are local in their proof but global in their consequences. Once established, they explain why harmonic forms on compact Kähler manifolds split by type, why Dolbeault cohomology can be represented by de Rham harmonic forms of fixed bidegree, and why the metric theory is compatible with the complex decomposition of forms. ## Commutators with the Lefschetz Operators What extra structure does the Kähler metric add beyond the Hermitian decomposition of forms? The answer is that wedging with the Kähler form has a metric adjoint, and the two operators control the first-order complex differentials through commutators. This converts the geometric condition $d\omega=0$ into identities between differential and algebraic operators. Let $(X,\omega)$ be a Kähler manifold of complex dimension $n$, and let $A_{p,q}(X)$ denote smooth complex-valued forms of type $(p,q)$. The Kähler form defines an operator that raises bidegree by $(1,1)$; this is the algebraic operation whose commutators will measure how the metric interacts with $\partial$ and $\bar\partial$. [definition: Lefschetz Operator] Let $L:A_{p,q}(X)\to A_{p+1,q+1}(X)$ be the operator \begin{align*} L\alpha = \omega \wedge \alpha. \end{align*} [/definition] The operator $L$ is algebraic: it does not differentiate the form. To turn wedging by $\omega$ into an operator that lowers degree, and to compare it with formal adjoints of differential operators, we need the metric adjoint of $L$. [definition: Dual Lefschetz Operator] Let $\Lambda:A_{p,q}(X)\to A_{p-1,q-1}(X)$ be the formal adjoint of $L$ with respect to the $L^2$ inner product on forms. If $p=0$ or $q=0$, the target $A_{p-1,q-1}(X)$ is interpreted as the zero space. [/definition] The pair $L,\Lambda$ gives the algebraic side of the theory. The next step is to record the failure of these algebraic operators to commute with differential operators, so we fix the commutator convention before stating the identities. [definition: Commutator] Let $A,B:A^*(X)\to A^*(X)$ be linear operators on the graded vector space of smooth complex-valued differential forms, and suppose that $AB$ and $BA$ are defined on the same subspace of $A^*(X)$. Their commutator is the operator \begin{align*} [A,B]:A^*(X)\to A^*(X),\qquad [A,B]=AB-BA. \end{align*} [/definition] This notation lets us ask a precise question: when $\Lambda$ is moved past $\partial$ or $\bar\partial$, what first-order operator appears? This is the role of the Kähler identities: they turn those commutators into the adjoint first-order operators, and they are the point at which the Kähler condition removes the torsion terms present for a general Hermitian metric. [quotetheorem:3853] [citeproof:3853] These identities are the basic bridge between the metric adjoint operations and the complex structure. Their hypotheses are exactly local Kähler hypotheses: a Hermitian metric, the closedness condition $d\omega=0$, and the adjoints determined by that metric. They are not formal consequences of having a Hermitian metric alone; on a non-Kähler Hermitian manifold the same commutators acquire torsion terms, so the displayed identities generally fail. More concretely, in Hermitian coordinates where the torsion does not vanish, first derivatives of the metric survive in $[\Lambda,\partial]-i\bar\partial^*$ and $[\Lambda,\bar\partial]+i\partial^*$ instead of cancelling in the flat-model calculation. This is a local distinction, not a compactness issue. For example, a Hopf surface carries natural Hermitian metrics but no Kähler metric, and the obstruction is visible analytically as the failure of the torsion-free normal-coordinate calculation used in the proof. The theorem also does not yet identify cohomology groups or produce harmonic representatives. It only supplies operator equalities on smooth forms, and those equalities make sense on any Kähler manifold, compact or not. Compactness and elliptic Hodge theory are later hypotheses: they enter only when these local operator identities are turned into statements about de Rham and Dolbeault cohomology. The signs depend on the convention for the Kähler form and the Hermitian inner product; throughout these notes we use the convention displayed in the theorem. [example: Flat Verification on Complex Euclidean Space] On $\mathbb C^n$ with coordinates $(z_1,\dots,z_n)$ and Kähler form $\omega=i\sum_{j=1}^n dz_j\wedge d\bar z_j$, write $\varepsilon_j$ for exterior multiplication by $dz_j$, $\bar\varepsilon_j$ for exterior multiplication by $d\bar z_j$, and $\iota_j,\bar\iota_j$ for the corresponding contractions. With the convention fixed in the theorem, \begin{align*} L=i\sum_{j=1}^n\varepsilon_j\bar\varepsilon_j,\qquad \Lambda=-i\sum_{j=1}^n\bar\iota_j\iota_j. \end{align*} For a form $f\eta$ whose exterior factor $\eta$ is constant, the flat operators are \begin{align*} \partial(f\eta)=\sum_{k=1}^n \varepsilon_k\frac{\partial f}{\partial z_k}\eta,\qquad \bar\partial^*(f\eta)=-\sum_{j=1}^n\bar\iota_j\frac{\partial f}{\partial z_j}\eta. \end{align*} Since $\Lambda$ is constant-coefficient, its contractions commute with the scalar derivatives. Therefore \begin{align*} [\Lambda,\partial](f\eta)=-i\sum_{j,k}\left(\bar\iota_j\iota_j\varepsilon_k-\varepsilon_k\bar\iota_j\iota_j\right)\frac{\partial f}{\partial z_k}\eta. \end{align*} Use $\iota_j\varepsilon_k+\varepsilon_k\iota_j=\delta_{jk}$ and $\bar\iota_j\varepsilon_k+\varepsilon_k\bar\iota_j=0$. The exterior factor in parentheses becomes \begin{align*} \bar\iota_j\iota_j\varepsilon_k=\bar\iota_j(\delta_{jk}-\varepsilon_k\iota_j)=\delta_{jk}\bar\iota_j+\varepsilon_k\bar\iota_j\iota_j. \end{align*} Subtracting $\varepsilon_k\bar\iota_j\iota_j$ leaves \begin{align*} \bar\iota_j\iota_j\varepsilon_k-\varepsilon_k\bar\iota_j\iota_j=\delta_{jk}\bar\iota_j. \end{align*} Hence \begin{align*} [\Lambda,\partial](f\eta)=-i\sum_j\bar\iota_j\frac{\partial f}{\partial z_j}\eta=i\bar\partial^*(f\eta). \end{align*} The barred calculation is parallel. Since $\bar\partial(f\eta)=\sum_k\bar\varepsilon_k(\partial f/\partial\bar z_k)\eta$ and $\partial^*(f\eta)=-\sum_j\iota_j(\partial f/\partial\bar z_j)\eta$, use $\bar\iota_j\bar\varepsilon_k+\bar\varepsilon_k\bar\iota_j=\delta_{jk}$ and $\iota_j\bar\varepsilon_k+\bar\varepsilon_k\iota_j=0$ to get \begin{align*} \bar\iota_j\iota_j\bar\varepsilon_k=-\bar\iota_j\bar\varepsilon_k\iota_j=-\delta_{jk}\iota_j+\bar\varepsilon_k\bar\iota_j\iota_j. \end{align*} Thus \begin{align*} \bar\iota_j\iota_j\bar\varepsilon_k-\bar\varepsilon_k\bar\iota_j\iota_j=-\delta_{jk}\iota_j. \end{align*} It follows that \begin{align*} [\Lambda,\bar\partial](f\eta)=i\sum_j\iota_j\frac{\partial f}{\partial\bar z_j}\eta=-i\partial^*(f\eta). \end{align*} On flat complex Euclidean space, the Lefschetz commutators therefore recover the adjoint Dolbeault differentials exactly: $[\Lambda,\partial]=i\bar\partial^*$ and $[\Lambda,\bar\partial]=-i\partial^*$. [/example] The flat computation also explains why the Kähler hypothesis is present. In general Hermitian coordinates, first derivatives of the metric contribute torsion terms to the same commutator calculation. The Kähler condition removes precisely those terms at the normal coordinate point, and this motivates isolating the scope of the identities before applying them to Laplacians. [remark: Local Nature of the Identities] The Kähler identities are compatible with restriction to coordinate charts and with tensoring forms by a flat auxiliary vector space. For vector bundles with non-flat Hermitian connections, curvature terms enter later through Weitzenböck-type formulae; the identities in this chapter concern ordinary scalar-valued forms. [/remark] ## Equality of the Kähler Laplacians Once adjoints can be expressed through commutators, the next question is how the second-order Laplacians compare. On a general complex manifold, the de Rham, $\partial$, and $\bar\partial$ Laplacians are distinct elliptic operators. On a Kähler manifold, the identities force them to coincide up to a factor of $2$. Before proving such an equality, we need a common convention for the Laplacian attached to each differential. This convention fixes the factor of $2$ that appears later and prevents confusion with alternative normalisations in complex geometry. [definition: Differential Laplacians] Let $D$ be one of the operators $d:A^k(X)\to A^{k+1}(X)$, $\partial:A_{p,q}(X)\to A_{p+1,q}(X)$, or $\bar\partial:A_{p,q}(X)\to A_{p,q+1}(X)$, with formal adjoint $D^*$. The Laplacian associated to $D$ is the operator $\Delta_D:A^*(X)\to A^*(X)$ defined by \begin{align*} \Delta_D = DD^*+D^*D. \end{align*} [/definition] The factor in the comparison comes from the decomposition $d=\partial+\bar\partial$ together with the convention that $d^*=\partial^*+\bar\partial^*$. The Laplacian consequence of the Kähler identities is needed because the mixed terms in the expansion of $\Delta_d$ do not vanish by type alone; their cancellation is exactly where the commutator identities enter. This consequence is the analytic heart of Kähler Hodge theory. It says that solving the de Rham harmonic equation is equivalent, after the factor of $2$, to solving either complex harmonic equation. This is much stronger than type considerations alone: the mixed adjoint terms in the expansion of $\Delta_d$ need not cancel on a general Hermitian manifold. The Kähler hypothesis cannot be dropped. On a Hopf surface, for instance, no Kähler metric exists, and the resulting complex surface has odd first Betti number; this is incompatible with the Hodge-theoretic consequences that would follow from the Kähler Laplacian identities on a compact complex surface. The equality also does not assert that every representative of a cohomology class has pure type. It says that the elliptic equations defining harmonic representatives are the same equations once the Kähler metric is fixed, which is the input needed for the later cohomological decomposition. [example: Laplacian Equality on Functions] Let $f\in C^\infty(X)$, and fix holomorphic [normal coordinates](/theorems/2713) at $x_0$. Since $\partial^*$ lowers bidegree by $(1,0)$ and $\bar\partial^*$ lowers bidegree by $(0,1)$, both adjoints vanish on functions: \begin{align*} \partial^*f=0,\qquad \bar\partial^*f=0. \end{align*} Thus \begin{align*} \Delta_\partial f=\partial^*\partial f. \end{align*} \begin{align*} \Delta_{\bar\partial}f=\bar\partial^*\bar\partial f. \end{align*} At $x_0$, the flat normal-coordinate formulas are \begin{align*} \partial f=\sum_k \varepsilon_k\frac{\partial f}{\partial z_k},\qquad \partial^*=-\sum_j\iota_j\frac{\partial}{\partial\bar z_j}. \end{align*} Therefore \begin{align*} \Delta_\partial f(x_0)=-\sum_{j,k}\iota_j\varepsilon_k\frac{\partial^2 f}{\partial\bar z_j\partial z_k}(x_0). \end{align*} On functions, $\iota_j\varepsilon_k(1)=\delta_{jk}$, so \begin{align*} \Delta_\partial f(x_0)=-\sum_j\frac{\partial^2 f}{\partial\bar z_j\partial z_j}(x_0). \end{align*} Similarly, \begin{align*} \bar\partial f=\sum_k \bar\varepsilon_k\frac{\partial f}{\partial\bar z_k},\qquad \bar\partial^*=-\sum_j\bar\iota_j\frac{\partial}{\partial z_j}. \end{align*} Hence \begin{align*} \Delta_{\bar\partial}f(x_0)=-\sum_{j,k}\bar\iota_j\bar\varepsilon_k\frac{\partial^2 f}{\partial z_j\partial\bar z_k}(x_0). \end{align*} Since $\bar\iota_j\bar\varepsilon_k(1)=\delta_{jk}$ and mixed partial derivatives commute for smooth functions, \begin{align*} \Delta_{\bar\partial}f(x_0)=-\sum_j\frac{\partial^2 f}{\partial z_j\partial\bar z_j}(x_0)=\Delta_\partial f(x_0). \end{align*} For the de Rham Laplacian, $d=\partial+\bar\partial$ and $d^*=\partial^*+\bar\partial^*$. The cross terms vanish on a function because $\partial^*\bar\partial f$ has target type $(-1,1)$ and $\bar\partial^*\partial f$ has target type $(1,-1)$. Thus \begin{align*} \Delta_d f(x_0)=\partial^*\partial f(x_0)+\bar\partial^*\bar\partial f(x_0)=2\left(-\sum_j\frac{\partial^2 f}{\partial z_j\partial\bar z_j}(x_0)\right). \end{align*} So on functions the $\partial$- and $\bar\partial$-Laplacians give the same normal-coordinate expression, while the de Rham Laplacian gives exactly twice that expression, matching the factor $2$ in *Equality of Kähler Laplacians*. [/example] The equality identifies the three second-order equations, but harmonicity is usually tested by first-order equations and an integration-by-parts identity. The following compactness-dependent criterion is needed to pass from equality of operators to equality of harmonic representatives. [quotetheorem:8055] [citeproof:8055] Compactness is doing real work here. The identity $(\Delta_D\alpha,\alpha)_{L^2}=\|D\alpha\|_{L^2}^2+\|D^*\alpha\|_{L^2}^2$ uses integration by parts without boundary contributions and avoids the domain issues that appear for unbounded operators on noncompact manifolds. On a compact manifold with boundary, extra boundary conditions would have to be imposed; on a noncompact complete manifold, $L^2$ harmonic theory requires separate analytic hypotheses. The criterion also does not construct harmonic representatives or prove finite-dimensionality of harmonic spaces. It only identifies the three notions of harmonicity once a smooth form is already under consideration. This identification will be used repeatedly when passing between de Rham and Dolbeault representatives, and it prepares the type decomposition result in the next section. ## Harmonic Forms and Type Decomposition The final question in this chapter is what the Laplacian equality says about the decomposition of a form into types. The obstruction is subtle: the splitting into $(p,q)$-pieces is algebraic, while harmonicity is analytic. To compare them, we first name the pieces whose behavior the Laplacian will have to respect. [definition: Type Components of a Form] For a smooth complex-valued $k$-form $\alpha$ on a complex manifold, write \begin{align*} \alpha=\sum_{p+q=k}\alpha_{p,q}, \end{align*} where $\alpha_{p,q}\in A_{p,q}(X)$. [/definition] The decomposition by type is available on any complex manifold. What is special in the Kähler setting is that the analytic operator defining harmonic representatives respects this algebraic splitting. On a general Hermitian manifold, the de Rham Laplacian can mix types; the Kähler identities remove that obstruction and make harmonicity visible component by component. This is the role of Hodge decomposition here, so we use it as a structural input rather than restating the same theorem card again. In a compact Kähler manifold, a de Rham harmonic form decomposes into harmonic pieces of pure type. Consequently the type decomposition is not just notation for forms; it becomes a decomposition of cohomological information. The conclusion is stronger than the corresponding statement on a general Hermitian manifold. Without the Kähler identities, the de Rham Laplacian need not preserve the type decomposition, and $d$-harmonicity need not imply Dolbeault harmonicity. A Hopf surface gives a concrete warning: its odd first Betti number rules out the Kähler-style decomposition of first de Rham cohomology into conjugate $(1,0)$ and $(0,1)$ pieces, so the harmonic type-splitting mechanism cannot hold there as it does in the compact Kähler case. All hypotheses are therefore relevant. The complex structure supplies the algebraic type decomposition, the Kähler metric supplies the Laplacian identities, and compactness lets harmonicity be tested by the adjoint equations. The result still stops short of the full Hodge decomposition of cohomology; that final step requires elliptic Hodge theory to guarantee a unique harmonic representative in each de Rham cohomology class. [example: Harmonic Components of a De Rham Harmonic Form] Let $X$ be compact Kähler, and let a complex-valued $2$-form decompose by type as \begin{align*} \alpha=\alpha_{2,0}+\alpha_{1,1}+\alpha_{0,2}. \end{align*} Assume $\Delta_d\alpha=0$. By *Equality of Kähler Laplacians*, $\Delta_d=2\Delta_{\bar\partial}$, so \begin{align*} 0=\Delta_d\alpha=2\Delta_{\bar\partial}\alpha. \end{align*} Dividing by $2$ gives \begin{align*} \Delta_{\bar\partial}\alpha=0. \end{align*} Now expand using linearity: \begin{align*} 0=\Delta_{\bar\partial}\alpha=\Delta_{\bar\partial}\alpha_{2,0}+\Delta_{\bar\partial}\alpha_{1,1}+\Delta_{\bar\partial}\alpha_{0,2}. \end{align*} The operator $\bar\partial$ has bidegree $(0,1)$ and $\bar\partial^*$ has bidegree $(0,-1)$, so both compositions $\bar\partial\bar\partial^*$ and $\bar\partial^*\bar\partial$ preserve bidegree. Hence \begin{align*} \Delta_{\bar\partial}\alpha_{2,0}\in A_{2,0}(X),\qquad \Delta_{\bar\partial}\alpha_{1,1}\in A_{1,1}(X),\qquad \Delta_{\bar\partial}\alpha_{0,2}\in A_{0,2}(X). \end{align*} Since the type decomposition $A^2(X)=A_{2,0}(X)\oplus A_{1,1}(X)\oplus A_{0,2}(X)$ is direct, a sum of components in these three distinct bidegrees is zero only when each component is zero: \begin{align*} \Delta_{\bar\partial}\alpha_{2,0}=0,\qquad \Delta_{\bar\partial}\alpha_{1,1}=0,\qquad \Delta_{\bar\partial}\alpha_{0,2}=0. \end{align*} Thus each type component is $\bar\partial$-harmonic. By *[Harmonicity Criteria on Compact Kähler Manifolds](/theorems/8055)*, $\bar\partial$-harmonicity is equivalent to $d$-harmonicity on compact Kähler manifolds, so $\alpha_{2,0}$, $\alpha_{1,1}$, and $\alpha_{0,2}$ are each $d$-harmonic as well. [/example] This example is the mechanism behind the Hodge decomposition of complex de Rham cohomology. Chapter 5 adds elliptic Hodge theory to identify cohomology classes with harmonic representatives; the Kähler-specific part is already visible here, where a harmonic form automatically separates into its complex types. [remark: Consequence for Cohomology] On a compact Kähler manifold, once Hodge theory gives a unique harmonic representative of each complex de Rham cohomology class, the type decomposition of harmonic forms induces a decomposition of cohomology by bidegree. The present chapter supplies the operator identities that make that decomposition compatible with both $d$-cohomology and Dolbeault cohomology. [/remark] # 5. Harmonic Forms and Hodge Decomposition Using the Laplacian equality and type-splitting from Chapter 4, this chapter turns the Kähler identities into cohomological structure. In the previous chapter the main analytic input was that the three Laplacians associated to $d$, $\partial$, and $\bar\partial$ are tied together on a Kähler manifold. We now use elliptic theory on compact manifolds to choose canonical representatives for cohomology classes and then read off the Hodge decomposition from the type decomposition of harmonic forms. ## Harmonic Representatives for de Rham Cohomology The first question is how a metric can help with a topological invariant such as de Rham cohomology. The answer is that a Riemannian metric supplies an adjoint to $d$, and hence a second-order elliptic operator whose kernel gives the preferred representatives of cohomology classes. [definition: Formal Adjoint of the Exterior Derivative] Let $(M,g)$ be a compact oriented Riemannian manifold of dimension $m$. The formal adjoint of $d: A^k(M;\mathbb C) \to A^{k+1}(M;\mathbb C)$ is the operator \begin{align*} d^*: A^{k+1}(M;\mathbb C) \to A^k(M;\mathbb C) \end{align*} determined by \begin{align*} (d\alpha,\beta)_{L^2}=(\alpha,d^*\beta)_{L^2} \end{align*} for all smooth forms $\alpha \in A^k(M;\mathbb C)$ and $\beta \in A^{k+1}(M;\mathbb C)$. [/definition] The formal adjoint lets us test whether a closed form has a coexact component, so it is the missing half of the metric picture. This motivates the next definition: combine $d$ and $d^*$ into one self-adjoint elliptic operator, and call the forms killed by it harmonic. [definition: de Rham Laplacian and Harmonic Form] Let $(M,g)$ be a compact oriented Riemannian manifold. On $k$-forms, the de Rham Laplacian is the operator \begin{align*} \Delta_d: A^k(M;\mathbb C) \to A^k(M;\mathbb C), \qquad \Delta_d = dd^*+d^*d. \end{align*} A smooth form $\alpha \in A^k(M;\mathbb C)$ is harmonic if \begin{align*} \Delta_d\alpha=0. \end{align*} The vector space of harmonic $k$-forms is denoted $\mathcal H^k_d(M)$. [/definition] The definition is analytic, but the next identity makes it usable: harmonicity is equivalent to being both closed and coclosed. This is where compactness enters through integration by parts. [quotetheorem:8056] [citeproof:8056] Thus a harmonic form represents a de Rham class automatically, but the criterion alone does not say that every cohomology class contains such a form. Compactness and the absence of boundary are doing real work here: without them, integration by parts can acquire boundary terms, and on noncompact manifolds harmonic representatives need not exist or be unique without imposing growth or boundary conditions. The next theorem supplies exactly the missing global statement, namely that the analytic kernel of $\Delta_d$ gives one preferred representative for each topological class. [quotetheorem:2745] [citeproof:2745] The compactness hypothesis is not decorative: it is what lets elliptic Fredholm theory produce a finite-dimensional kernel and a closed image. The theorem does not identify harmonic representatives by an explicit formula, nor does it say that an arbitrary closed representative is already harmonic; it says that each class can be projected uniquely to the harmonic summand. In the Kähler course, this theorem is not merely a Riemannian fact in the background. Once the metric is Kähler, the same harmonic form can be detected by $\partial$ and $\bar\partial$ Laplacians, which is the bridge from topology to complex geometry. ## Dolbeault Harmonic Forms The next problem is to choose representatives for Dolbeault cohomology. A Dolbeault class is represented by a $\bar\partial$-closed $(p,q)$-form modulo $\bar\partial$-exact forms, so the appropriate Laplacian must preserve bidegree. [definition: Dolbeault Laplacian] Let $X$ be a compact Hermitian manifold. The formal adjoint of \begin{align*} \bar\partial:A^{p,q}(X)\to A^{p,q+1}(X) \end{align*} is the operator \begin{align*} \bar\partial^*:A^{p,q+1}(X)\to A^{p,q}(X). \end{align*} The Dolbeault Laplacian on $(p,q)$-forms 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*} A form $\alpha\in A^{p,q}(X)$ is $\bar\partial$-harmonic if $\Delta_{\bar\partial}\alpha=0$. The space of such forms is denoted $\mathcal H^{p,q}_{\bar\partial}(X)$. [/definition] This definition mirrors the de Rham Laplacian while staying inside the bundle of $(p,q)$-forms. Before using elliptic theory, we need the same energy test that tells us which $\bar\partial$-closed forms are actually harmonic representatives. [quotetheorem:8057] [citeproof:8057] The criterion says what the harmonic representatives must satisfy inside each bidegree, but it does not produce a representative for a given Dolbeault class. Compactness again prevents analytic mass from escaping to infinity; on noncompact complex manifolds, Dolbeault cohomology may not be represented by smooth harmonic forms without extra conditions. The remaining question is existence and uniqueness for every Dolbeault class, and that is supplied by the elliptic theorem for $\Delta_{\bar\partial}$. [quotetheorem:8059] [citeproof:8059] The theorem requires only a compact Hermitian metric, not a Kähler metric, because the operator $\Delta_{\bar\partial}$ is elliptic within a fixed bidegree on any compact Hermitian manifold. It does not yet compare Dolbeault cohomology with de Rham cohomology, and it does not imply any symmetry between $h^{p,q}$ and $h^{q,p}$. What it does justify is writing \begin{align*} h^{p,q}(X)=\dim_{\mathbb C}H^{p,q}_{\bar\partial}(X)=\dim_{\mathbb C}\mathcal H^{p,q}_{\bar\partial}(X) \end{align*} for the Hodge numbers. On a general Hermitian manifold these are Dolbeault invariants; on a compact Kähler manifold the next analytic input explains why they also assemble into de Rham cohomology. [example: Holomorphic Forms as Harmonic Forms] Let $X$ be a compact Hermitian manifold, and let $\alpha\in A^{p,0}(X)$ satisfy $\bar\partial\alpha=0$. The adjoint $\bar\partial^*$ lowers the second bidegree by $1$, so on $(p,0)$-forms it has target $A^{p,-1}(X)$. Since $A^{p,-1}(X)=0$, we have \begin{align*} \bar\partial^*\alpha=0. \end{align*} Together with the assumed equation $\bar\partial\alpha=0$, the *Dolbeault Harmonicity Criterion* gives \begin{align*} \Delta_{\bar\partial}\alpha=0. \end{align*} Thus every holomorphic $p$-form is $\bar\partial$-harmonic. Conversely, if $\alpha\in A^{p,0}(X)$ is $\bar\partial$-harmonic, then the *Dolbeault Harmonicity Criterion* gives \begin{align*} \bar\partial\alpha=0. \end{align*} So the $\bar\partial$-harmonic $(p,0)$-forms are exactly the holomorphic $p$-forms. By the *[Dolbeault Hodge Theorem](/theorems/8059)*, every class in $H^{p,0}_{\bar\partial}(X)$ has a unique $\bar\partial$-harmonic representative, hence \begin{align*} h^{p,0}(X)=\dim_{\mathbb C} H^{p,0}_{\bar\partial}(X)=\dim_{\mathbb C}\{\text{holomorphic }p\text{-forms on }X\}. \end{align*} [/example] ## The Kähler Identification of Laplacians The remaining question is why the de Rham harmonic representative of a class should split into harmonic $(p,q)$-pieces. This is exactly where the Kähler identities from the previous chapter enter. [quotetheorem:3853] [citeproof:3853] This result is the analytic heart of Hodge theory on Kähler manifolds, and the Kähler hypothesis is essential because the commutator identities need $d\omega=0$. On a general Hermitian manifold, the de Rham Laplacian may mix bidegrees, so a de Rham harmonic form need not split into Dolbeault harmonic pieces. The theorem does not compute the harmonic forms by itself; it identifies three different elliptic equations and therefore lets information pass between de Rham and Dolbeault theories. Since $\Delta_{\bar\partial}$ preserves type, the equality of Laplacians is exactly the missing bridge from analytic representatives to cohomological summands. The next section records that bridge once as the Hodge decomposition theorem. This also explains why the Kähler condition is stronger than the existence of a Hermitian metric. Compactness supplies harmonic representatives, while the Kähler condition makes their type components harmonic as well; without either ingredient the conclusion can fail in a different way. The point is not that every $(p,q)$-form is closed or harmonic, but that the unique de Rham harmonic representative of a cohomology class has harmonic type components. Without the Kähler identities, the de Rham Laplacian need not respect bidegree, and the topological cohomology may fail to decompose according to Dolbeault type. ## Hodge Decomposition of Cohomology We can now answer the main cohomological question of the chapter. Given a compact Kähler manifold, every de Rham cohomology class has a unique harmonic representative, and that representative splits into harmonic bidegree components. The formal statement needed here is therefore the Hodge decomposition theorem: it turns the analytic equality of Laplacians into a canonical decomposition of the [topological vector space](/page/Topological%20Vector%20Space) $H^k_{\mathrm{dR}}(X;\mathbb C)$, with each summand identified by Dolbeault type rather than by a chosen representative. [quotetheorem:3854] [citeproof:3854] This decomposition is canonical at the level of cohomology, even though its construction uses harmonic representatives for a chosen Kähler metric. Compactness and the Kähler condition are both part of the structure: compactness gives a finite-dimensional Hodge theory, while the Kähler identities make the type decomposition compatible with de Rham cohomology. A compact complex manifold without a Kähler metric can have Dolbeault groups that do not assemble into de Rham cohomology in this way. The decomposition also explains why numerical symmetries of Dolbeault groups become topological constraints in the Kähler setting. In degree $1$, complex conjugation exchanges the two Hodge summands, so the two dimensions match and the first Betti number is even. [example: Projective Space] Let $H=[\omega_{FS}]\in H^2(\mathbb{CP}^n;\mathbb C)$. The standard [cohomology ring](/theorems/2271) computation for projective space gives \begin{align*} H^*(\mathbb{CP}^n;\mathbb C)\cong \mathbb C[H]/(H^{n+1}), \qquad \deg H=2. \end{align*} Thus, for $0\le r\le n$, the class $H^r$ is nonzero and spans $H^{2r}(\mathbb{CP}^n;\mathbb C)$, while all odd-degree cohomology groups vanish: \begin{align*} \dim_{\mathbb C}H^{2r}(\mathbb{CP}^n;\mathbb C)=1. \end{align*} \begin{align*} H^{2r+1}(\mathbb{CP}^n;\mathbb C)=0. \end{align*} The Fubini-Study form has type $(1,1)$, so its $r$-fold wedge power has type $(r,r)$: \begin{align*} \omega_{FS}^r\in A^{r,r}(\mathbb{CP}^n). \end{align*} Therefore the cohomology class $H^r=[\omega_{FS}^r]$ lies in the $(r,r)$ summand under the *Hodge Decomposition for Compact Kähler Manifolds*. Since $H^{2r}(\mathbb{CP}^n;\mathbb C)$ is one-dimensional and already contains the nonzero class $H^r$ in type $(r,r)$, the direct sum decomposition forces \begin{align*} h^{r,r}(\mathbb{CP}^n)=1 \quad \text{for }0\le r\le n. \end{align*} If $p+q$ is odd, then $H^{p+q}(\mathbb{CP}^n;\mathbb C)=0$, so $h^{p,q}(\mathbb{CP}^n)=0$. If $p+q=2r$ and $p\ne q$, then the one-dimensional group $H^{2r}(\mathbb{CP}^n;\mathbb C)$ is already exhausted by the $(r,r)$ summand, so again $h^{p,q}(\mathbb{CP}^n)=0$. Thus the Hodge diamond of projective space lies entirely on the diagonal, matching $b_{2r}=1$ and $b_{2r+1}=0$. [/example] The projective-space calculation is the cleanest possible case: all cohomology lies on the diagonal $p=q$. Flat complex tori show that off-diagonal pieces can appear in abundance while the Kähler decomposition still controls them. [example: Complex Tori] Let $X=\mathbb C^n/\Lambda$ be a complex torus with its flat Kähler metric, and use the global coordinates $z_1,\ldots,z_n$ induced from $\mathbb C^n$. For ordered index sets $I=\{i_1<\cdots<i_p\}$ and $J=\{j_1<\cdots<j_q\}$, set \begin{align*} dz_I\wedge d\bar z_J=dz_{i_1}\wedge\cdots\wedge dz_{i_p}\wedge d\bar z_{j_1}\wedge\cdots\wedge d\bar z_{j_q}. \end{align*} These forms have constant coefficients, so $\bar\partial(dz_I\wedge d\bar z_J)=0$. For the flat metric, the Hodge star sends constant coefficient forms to constant coefficient forms; hence $\bar\partial^*(dz_I\wedge d\bar z_J)=0$ as well, using the adjoint formula for $\bar\partial^*$ in terms of the Hodge star. By the *[Harmonic Criterion for the Dolbeault Laplacian](/theorems/8057)*, each $dz_I\wedge d\bar z_J$ is $\bar\partial$-harmonic. Conversely, on a flat torus the scalar Laplacian diagonalizes on Fourier modes. A nonzero Fourier mode has positive eigenvalue, so a harmonic form can have only the zero Fourier mode in each coefficient. Thus the $\bar\partial$-harmonic $(p,q)$-forms are exactly the constant linear combinations of the forms $dz_I\wedge d\bar z_J$. There are $\binom{n}{p}$ choices for $I$ and $\binom{n}{q}$ choices for $J$, so \begin{align*} h^{p,q}(X)=\dim_{\mathbb C}\mathcal H^{p,q}_{\bar\partial}(X)=\binom{n}{p}\binom{n}{q}. \end{align*} By the *Hodge Decomposition for Compact Kähler Manifolds*, the degree-$k$ Betti number is the sum of these dimensions over $p+q=k$: \begin{align*} b_k(X)=\sum_{p+q=k}h^{p,q}(X)=\sum_{p+q=k}\binom{n}{p}\binom{n}{q}. \end{align*} The [Vandermonde identity](/theorems/7731) follows by comparing the coefficient of $t^k$ in \begin{align*} (1+t)^n(1+t)^n=(1+t)^{2n}. \end{align*} The coefficient on the left is $\sum_{p+q=k}\binom{n}{p}\binom{n}{q}$, while the coefficient on the right is $\binom{2n}{k}$. Therefore \begin{align*} b_k(X)=\binom{2n}{k}. \end{align*} So a complex torus has many off-diagonal Hodge summands, but their total dimensions still recombine exactly into the Betti numbers of the underlying real $2n$-torus. [/example] Riemann surfaces provide the first case where Hodge numbers encode geometric moduli. The entire decomposition is concentrated in degrees $0$, $1$, and $2$. [example: Compact Riemann Surfaces] Let $C$ be a compact Riemann surface of genus $g$. Since $C$ is a compact Kähler manifold of complex dimension $1$, the only possible bidegrees are $(0,0),(1,0),(0,1),(1,1)$. By the *Hodge Decomposition for Compact Kähler Manifolds*, degree $1$ decomposes as \begin{align*} H^1(C;\mathbb C)=H^{1,0}_{\bar\partial}(C)\oplus H^{0,1}_{\bar\partial}(C). \end{align*} Taking dimensions gives \begin{align*} b_1(C)=h^{1,0}(C)+h^{0,1}(C). \end{align*} The underlying real surface has genus $g$, so $b_1(C)=2g$. By *Hodge Symmetry*, $h^{1,0}(C)=h^{0,1}(C)$, and therefore \begin{align*} 2g=h^{1,0}(C)+h^{0,1}(C)=2h^{1,0}(C). \end{align*} Hence \begin{align*} h^{1,0}(C)=h^{0,1}(C)=g. \end{align*} In degree $0$, the Hodge decomposition has only the $(0,0)$ summand: \begin{align*} H^0(C;\mathbb C)=H^{0,0}_{\bar\partial}(C). \end{align*} Since $C$ is connected, $H^0(C;\mathbb C)\cong \mathbb C$, so \begin{align*} h^{0,0}(C)=1. \end{align*} In degree $2$, the only possible bidegree is $(1,1)$, so \begin{align*} H^2(C;\mathbb C)=H^{1,1}_{\bar\partial}(C). \end{align*} The compact oriented real surface $C$ has $H^2(C;\mathbb C)\cong \mathbb C$, generated by the orientation class represented by any volume form, hence \begin{align*} h^{1,1}(C)=1. \end{align*} Thus the Hodge diamond of a compact Riemann surface is completely determined by its genus: the corner entries are $1$, and the two middle entries are both $g$. [/example] The examples so far are Kähler, so the decomposition agrees with de Rham cohomology. A standard non-Kähler surface shows what fails when the Kähler identities are unavailable. [example: Hopf Surface Warning] Let $S$ be a primary Hopf surface, so its underlying smooth manifold is diffeomorphic to $S^1\times S^3$. Using the *Kunneth Formula* over $\mathbb C$, \begin{align*} H^1(S^1\times S^3;\mathbb C)\cong \bigl(H^1(S^1;\mathbb C)\otimes H^0(S^3;\mathbb C)\bigr)\oplus \bigl(H^0(S^1;\mathbb C)\otimes H^1(S^3;\mathbb C)\bigr). \end{align*} The cohomology dimensions of spheres give \begin{align*} \dim_{\mathbb C}H^1(S^1;\mathbb C)=1,\quad \dim_{\mathbb C}H^0(S^3;\mathbb C)=1,\quad \dim_{\mathbb C}H^0(S^1;\mathbb C)=1,\quad \dim_{\mathbb C}H^1(S^3;\mathbb C)=0. \end{align*} Therefore \begin{align*} b_1(S)=\dim_{\mathbb C}H^1(S;\mathbb C)=(1)(1)+(1)(0)=1. \end{align*} Suppose, for contradiction, that $S$ admitted a Kähler metric. Then the degree-one case of *Hodge Decomposition for Compact Kähler Manifolds* would give \begin{align*} H^1(S;\mathbb C)=H^{1,0}_{\bar\partial}(S)\oplus H^{0,1}_{\bar\partial}(S). \end{align*} Taking dimensions of this direct sum gives \begin{align*} b_1(S)=h^{1,0}(S)+h^{0,1}(S). \end{align*} By *Hodge Symmetry*, \begin{align*} h^{1,0}(S)=h^{0,1}(S). \end{align*} Substituting this equality into the dimension formula gives \begin{align*} b_1(S)=h^{1,0}(S)+h^{1,0}(S)=2h^{1,0}(S). \end{align*} Since $h^{1,0}(S)$ is a complex vector-space dimension, $2h^{1,0}(S)$ is even, contradicting $b_1(S)=1$. Thus a Hopf surface is not Kähler, and its Dolbeault groups cannot assemble into de Rham cohomology by the Kähler Hodge decomposition. [/example] ## What the Decomposition Gives Next The chapter has converted the analytic Kähler identities into a cohomological splitting. The next stage of the course uses the same operators together with the Lefschetz operator $L(\alpha)=\omega\wedge\alpha$ to study how cup product with the Kähler class acts on cohomology. Harmonic representatives remain the mechanism: once operators commute with the Laplacian, identities on forms descend to identities on cohomology. # 6. The $\partial\bar\partial$ Lemma and Formal Consequences The previous chapters built the analytic package of Kähler geometry: the decomposition of forms by type, the Kähler identities, harmonic representatives, and the Hodge decomposition of de Rham cohomology. This chapter extracts one of the most useful algebraic consequences of that analysis, the $\partial\bar{\partial}$ lemma. Its role is to say that on a compact Kähler manifold, the different ways a pure-type closed form can be exact all collapse to the same condition, and the common correction term is a $\partial\bar{\partial}$ of a form of one lower bidegree. This should be read as a global refinement of the local potential result for closed real $(1,1)$-forms from Chapter 1. The payoff is structural rather than computational. The lemma gives control of representatives inside cohomology classes, explains why the Frölicher spectral sequence has no higher differentials on compact Kähler manifolds, and is the local mechanism behind the formality phenomena that distinguish Kähler manifolds from general compact complex manifolds. ## Exactness Questions for Pure-Type Forms Suppose $X$ is a compact complex manifold and $\alpha \in A^{p,q}(X)$ is closed for the total differential $d = \partial + \bar{\partial}$. If $\alpha$ is $d$-exact as a complex-valued differential form, there is no immediate reason for a primitive to respect the bidegree. The guiding question is whether exactness in de Rham cohomology can be witnessed by the operators $\partial$, $\bar{\partial}$, and $\partial\bar{\partial}$ without leaving the pure-type setting. Throughout this chapter, $A^{p,q}(X)$ denotes the space of smooth complex-valued forms of type $(p,q)$. The total degree $k$ forms are \begin{align*} A^k(X;\mathbb C) := \bigoplus_{p+q=k} A^{p,q}(X). \end{align*} The full complex of smooth complex-valued forms is \begin{align*} A^\bullet(X;\mathbb C) := \bigoplus_{k\ge 0} A^k(X;\mathbb C). \end{align*} The total differential is $d:A^k(X;\mathbb C)\to A^{k+1}(X;\mathbb C)$. On pure bidegrees, the operators are $\partial:A^{p,q}(X)\to A^{p+1,q}(X)$, $\bar{\partial}:A^{p,q}(X)\to A^{p,q+1}(X)$, and $\partial\bar{\partial}:A^{p-1,q-1}(X)\to A^{p,q}(X)$, with $d=\partial+\bar{\partial}$ on the total complex. When images such as $\operatorname{im}\partial$ occur in bidegree $(p,q)$, they mean images of these maps with target $A^{p,q}(X)$. Thus $\alpha\in\operatorname{im}d$ for $\alpha\in A^{p,q}(X)$ means $\alpha=d\eta$ for some $\eta\in A^{p+q-1}(X;\mathbb C)$, while $\alpha\in\operatorname{im}\partial\bar{\partial}$ means $\alpha=\partial\bar{\partial}\beta$ for some $\beta\in A^{p-1,q-1}(X)$. [definition: Pure-Type Closed Form] Let $X$ be a complex manifold. A pure-type closed form of bidegree $(p,q)$ is a form $\alpha \in A^{p,q}(X)$ satisfying \begin{align*} d\alpha = 0. \end{align*} [/definition] Since $d\alpha = \partial\alpha + \bar{\partial}\alpha$ has components of different bidegrees, a pure-type closed form satisfies both $\partial\alpha = 0$ and $\bar{\partial}\alpha = 0$. The exactness question is then whether the different ways a closed pure-type form can become a boundary are genuinely different, or whether Kähler geometry forces them to coincide. The $\partial\bar{\partial}$ lemma answers this question on compact Kähler manifolds. For a pure-type closed form $\alpha\in A^{p,q}(X)$, the conditions that $\alpha$ is $d$-exact, $\partial$-exact, $\bar{\partial}$-exact, and $\partial\bar{\partial}$-exact are equivalent. The conclusion is striking because the last condition is visibly stronger than the first three before Kähler identities enter. The correction term has bidegree $(p-1,q-1)$, so the lemma also tells us that exactness respects the Hodge diamond in a way that is not available on a general complex manifold. The compact Kähler hypothesis is doing real work. On the Iwasawa manifold, with invariant $(1,0)$-forms satisfying $d\alpha_3=\alpha_1\wedge\alpha_2$, the pure-type form $\alpha_1\wedge\alpha_2$ is $d$-exact and $\partial$-exact, but Bott-Chern cohomology detects that it is not $\partial\bar{\partial}$-exact. The lemma also does not say that every closed form is itself of pure type, nor that every compact complex manifold with a few Kähler-like numerical coincidences is Kähler; it is a statement about exactness after a pure-type representative has already been fixed. In applications, the working recipe is as follows. First isolate a form of a single bidegree and check that it is $d$-closed, equivalently both $\partial$-closed and $\bar{\partial}$-closed. Next identify one source of exactness, such as $d$-exactness in de Rham cohomology or $\partial$-exactness in a Dolbeault calculation. On a compact Kähler manifold the lemma upgrades that single exactness condition to the strongest conclusion, namely a representation as $\partial\bar{\partial}$ of a form one step down in both bidegrees. [example: Adjusting A Closed One One Form] Let $\beta:=\alpha'-\alpha$. Since $\alpha$ and $\alpha'$ represent the same class in $H^2(X;\mathbb R)$, there is a real $1$-form $\eta$ with \begin{align*} \beta=d\eta. \end{align*} Also $\beta\in A^{1,1}(X)$, and \begin{align*} d\beta=d\alpha'-d\alpha=0-0=0. \end{align*} Thus $\beta$ is a closed pure-type $(1,1)$-form which is $d$-exact. By the exactness form of the $\partial\bar{\partial}$ lemma, there is a smooth complex-valued function $u\in A^{0,0}(X)$ such that \begin{align*} \beta=\partial\bar{\partial}u. \end{align*} Because $\beta$ is real, complex conjugation gives $\overline{\beta}=\beta$. On functions, \begin{align*} \overline{\partial\bar{\partial}u}=\bar{\partial}\partial\overline{u}. \end{align*} Using $\bar{\partial}\partial=-\partial\bar{\partial}$, we get \begin{align*} \beta=\overline{\beta}=-\partial\bar{\partial}\overline{u}. \end{align*} Therefore \begin{align*} \partial\bar{\partial}(u-\overline{u})=\partial\bar{\partial}u-\partial\bar{\partial}\overline{u}=\beta-(-\beta)=2\beta. \end{align*} Since $u-\overline{u}=2i\,\operatorname{Im}u$, setting $\varphi:=\operatorname{Im}u$ gives \begin{align*} 2i\,\partial\bar{\partial}\varphi=2\beta. \end{align*} Dividing by $2$ yields \begin{align*} \alpha'-\alpha=\beta=i\partial\bar{\partial}\varphi. \end{align*} Thus, with the convention that real $(1,1)$-potentials are written as $i\partial\bar{\partial}$ of a real function, \begin{align*} \alpha'=\alpha+i\partial\bar{\partial}\varphi. \end{align*} The cohomological statement that $\alpha$ and $\alpha'$ differ by an exact form has become the potential-theoretic statement that they differ by $i\partial\bar{\partial}$ of a real smooth function. [/example] This example is the bridge between cohomological uniqueness and potential theory. It says that within a fixed de Rham class, the Kähler condition forces all pure-type closed representatives to differ by a potential, which is the starting point for many later arguments about Kähler classes. ## Equivalent Formulations and Bott-Chern Diagnosis The preceding theorem can feel metric-dependent because the proof uses harmonic theory. The content, however, can be detected by cohomology groups built from $\partial$ and $\bar{\partial}$ alone. These groups measure exactly how far a compact complex manifold is from having the Kähler exactness pattern. [definition: Bott-Chern Cohomology] Let $X$ be a complex manifold. The Bott-Chern cohomology group of bidegree $(p,q)$ is \begin{align*} H^{p,q}_{BC}(X) := \frac{\ker \partial \cap \ker \bar{\partial} \cap A^{p,q}(X)}{\operatorname{im}(\partial\bar{\partial}) \cap A^{p,q}(X)}. \end{align*} [/definition] In this definition the kernels are taken for $\partial:A^{p,q}(X)\to A^{p+1,q}(X)$ and $\bar{\partial}:A^{p,q}(X)\to A^{p,q+1}(X)$, while the denominator is the image of $\partial\bar{\partial}:A^{p-1,q-1}(X)\to A^{p,q}(X)$. Bott-Chern cohomology records closed pure-type forms but quotients only by the strongest correction allowed in the $\partial\bar{\partial}$ lemma. Thus extra Bott-Chern classes often signal that a form is closed but cannot be simplified by a $\partial\bar{\partial}$ potential. [definition: Aeppli Cohomology] Let $X$ be a complex manifold. The Aeppli cohomology group of bidegree $(p,q)$ is \begin{align*} H^{p,q}_{A}(X) := \frac{\ker(\partial\bar{\partial}) \cap A^{p,q}(X)}{(\operatorname{im}\partial + \operatorname{im}\bar{\partial}) \cap A^{p,q}(X)}. \end{align*} [/definition] Here $\partial\bar{\partial}:A^{p,q}(X)\to A^{p+1,q+1}(X)$ in the numerator, and the denominator uses $\partial:A^{p-1,q}(X)\to A^{p,q}(X)$ together with $\bar{\partial}:A^{p,q-1}(X)\to A^{p,q}(X)$. Aeppli cohomology uses the opposite quotient: it allows correction by $\partial$- and $\bar{\partial}$-exact pieces separately, but asks only for $\partial\bar{\partial}$-closedness. The natural comparison from Bott-Chern to Aeppli is therefore a compact way to package the lemma. [quotetheorem:8060] [citeproof:8060] This formulation is useful because it gives a test for non-Kähler behaviour without choosing a Kähler metric. The compact complex hypothesis fixes the global double complex of smooth forms on a compact manifold, so Bott-Chern and Aeppli groups are finite-dimensional and the comparison map above records a global cohomological property rather than a local potential calculation. Compactness is also the hypothesis under which the course uses elliptic Hodge theory to prove the Kähler case; without it, exactness questions can change because primitives may fail global growth or boundary conditions. Compact complexness alone is still not enough: the Iwasawa manifold has Bott-Chern and Aeppli groups whose dimensions fail the equalities forced by the comparison map above, and this failure records an exact pure-type form that is not $\partial\bar{\partial}$-exact. The equivalence is also limited in scope: injectivity of the Bott-Chern-to-Aeppli map is a diagnosis of the $\partial\bar{\partial}$ lemma, not by itself a construction of a Kähler metric or a replacement for the analytic Kähler identities. The next example is the standard warning that compact complex manifolds can have de Rham cohomology that does not split according to the Kähler pattern. [example: Iwasawa Manifold And Failure Of The Kähler Pattern] Let $X$ be the Iwasawa manifold, obtained as a compact quotient of the complex Heisenberg group by a lattice. It admits invariant $(1,0)$-forms $\alpha_1,\alpha_2,\alpha_3$ satisfying \begin{align*} d\alpha_1 = 0, \qquad d\alpha_2 = 0, \qquad d\alpha_3 = \alpha_1\wedge \alpha_2. \end{align*} Since $\alpha_3$ has type $(1,0)$ and $d=\partial+\bar{\partial}$, the form $d\alpha_3$ decomposes as \begin{align*} d\alpha_3=\partial\alpha_3+\bar{\partial}\alpha_3, \end{align*} where $\partial\alpha_3\in A^{2,0}(X)$ and $\bar{\partial}\alpha_3\in A^{1,1}(X)$. The given identity $d\alpha_3=\alpha_1\wedge\alpha_2$ has only a $(2,0)$ component, so comparison of bidegrees gives \begin{align*} \partial\alpha_3=\alpha_1\wedge\alpha_2, \qquad \bar{\partial}\alpha_3=0. \end{align*} Thus $\alpha_1\wedge\alpha_2$ is $d$-exact, because $\alpha_1\wedge\alpha_2=d\alpha_3$, and it is pure of type $(2,0)$. It is also closed: applying $d^2=0$ to $\alpha_3$ gives \begin{align*} d(\alpha_1\wedge\alpha_2)=d(d\alpha_3)=0. \end{align*} If the $\partial\bar{\partial}$ lemma held on $X$, this closed pure-type $d$-exact form would have to lie in $\operatorname{im}\partial\bar{\partial}$. In bidegree $(2,0)$, however, a $\partial\bar{\partial}$-primitive would have to lie in $A^{1,-1}(X)$, and \begin{align*} A^{1,-1}(X)=0 \end{align*} because negative antiholomorphic degree does not occur. Hence \begin{align*} \operatorname{im}\bigl(\partial\bar{\partial}:A^{1,-1}(X)\to A^{2,0}(X)\bigr)=0. \end{align*} The form $\alpha_1\wedge\alpha_2$ is not zero as an invariant $(2,0)$-form, so it cannot be $\partial\bar{\partial}$-exact. Thus the Iwasawa manifold has a closed pure-type form which is $d$-exact and $\partial$-exact but not $\partial\bar{\partial}$-exact. This is the precise failure of the Kähler pattern: de Rham exactness does not force the existence of a $\partial\bar{\partial}$ potential. [/example] The point of the example is not only that the Iwasawa manifold is non-Kähler. It shows the precise failure mode: a pure-type exact form exists whose exactness cannot be promoted to a $\partial\bar{\partial}$ potential. ## Representatives, Spectral Sequences, and Formality Once exactness has been upgraded to $\partial\bar{\partial}$-exactness, several global consequences follow with little additional analysis. The common theme is that cohomological ambiguity can be removed within each bidegree, so filtrations and products behave as if the de Rham complex were split into independent Dolbeault pieces. [quotetheorem:8061] [citeproof:8061] This theorem recovers the cohomological splitting already obtained analytically for compact Kähler manifolds, but the $\partial\bar{\partial}$ hypothesis is essential. The Iwasawa manifold is compact and complex, so compactness and the complex-analytic double complex are present; what fails is the Kähler exactness package encoded by the $\partial\bar{\partial}$ lemma. As a result, Dolbeault cohomology is too large in the wrong bidegrees to assemble into de Rham cohomology through this direct sum isomorphism, reflecting the same Bott-Chern obstruction seen above. The theorem also does not assert that every complex manifold with a vector-space decomposition of cohomology is Kähler; it asserts that the $\partial\bar{\partial}$ lemma supplies canonical control of representatives and of their ambiguity. The next question asks how this splitting appears inside the filtered de Rham complex before taking its final cohomology, which leads to the Frölicher spectral sequence associated to the filtration by holomorphic degree. [quotetheorem:3854] [citeproof:3854] This Hodge decomposition statement is often the most visible topological consequence: it gives the equality between Betti numbers and sums of Hodge numbers. Compact Kählerness is needed because non-Kähler compact complex manifolds can have genuine higher differentials or incompatible filtrations; the Iwasawa manifold is the standard example where the Kähler pattern fails in the surrounding Bott-Chern and Dolbeault data. Conversely, $E_1$-degeneration alone is not a Kähler criterion: some non-Kähler manifolds have a Frölicher spectral sequence degenerating at $E_1$ but still fail the $\partial\bar{\partial}$ lemma. The lemma also controls the algebra structure, which is stronger than a vector-space decomposition. [quotetheorem:8062] [citeproof:8062] Formality explains why compact Kähler manifolds have severe restrictions as smooth manifolds. The $\partial\bar{\partial}$ hypothesis is essential here as well: the Kodaira-Thurston manifold is compact and symplectic but not formal, so no argument of this kind can hold for general compact complex or symplectic manifolds. The conclusion also has a limitation in the opposite direction: formality by itself does not imply the existence of a Kähler metric, since there are formal manifolds outside the Kähler category. Thus the theorem should be read as a strong consequence of the $\partial\bar{\partial}$ lemma, not as an intrinsic characterisation of compact Kähler geometry. [remark: Scope Of The Consequences] The chain of implications in this chapter is one-way for the purposes of the course: compact Kähler implies the $\partial\bar{\partial}$ lemma, which implies Hodge decomposition, $E_1$-degeneration, and formality. Some non-Kähler manifolds may satisfy selected consequences, so failure of one diagnostic proves non-Kählerness while success of one diagnostic alone does not prove the existence of a Kähler metric. [/remark] # 7. Hard Lefschetz Theorem Chapters 4 through 6 turned the Kähler identities into cohomological consequences: harmonic representatives respect type, de Rham cohomology splits into Hodge pieces, and the $\partial\bar\partial$ lemma controls representatives. This chapter adds the second major structure forced by a Kähler class, namely the Lefschetz action of wedging with powers of the Kähler form. The guiding question is how much of the cohomology ring is controlled by this single degree-two class, and how the remaining primitive part carries a signed Hermitian form. ## Global Lefschetz Maps A compact Kähler manifold $X$ of complex dimension $n$ has a preferred real cohomology class $[\omega] \in H^2(X;\mathbb R)$ once a Kähler form $\omega$ is chosen. The first problem is to understand the maps obtained by multiplying cohomology classes by powers of this class, especially when the source and target degrees are symmetric around the middle dimension. [definition: Global Lefschetz Operator] Let $(X,\omega)$ be a compact Kähler manifold. The global Lefschetz operator is the graded linear map \begin{align*} L &: H^k(X;\mathbb C) \to H^{k+2}(X;\mathbb C), & L[\alpha] &= [\omega]\smile [\alpha]. \end{align*} [/definition] For $r \ge 0$, write $L^r$ for the iterated cup product map \begin{align*} L^r &: H^k(X;\mathbb C) \to H^{k+2r}(X;\mathbb C), & L^r[\alpha] &= [\omega]^r\smile [\alpha]. \end{align*} This is well-defined because $d\omega=0$, so wedging a closed representative with $\omega$ gives another closed representative, and changing the representative changes the wedge by an exact form. The natural obstruction is that a degree-raising map need not be injective or surjective on cohomology. The Kähler identities remove this obstruction by linking the global map $L$ to the pointwise Lefschetz representation on harmonic forms. [quotetheorem:3876] [citeproof:3876] The compactness hypothesis is used through Hodge theory: without compactness, cohomology classes need not have unique harmonic representatives in the same way. The Kähler hypothesis is also essential. Compact non-Kähler complex manifolds, such as the Hopf surface, can have cohomology groups incompatible with this symmetry; in particular the existence of a Lefschetz-type class would force degree symmetries that its Betti numbers do not satisfy. The degree range is sharp because $L^r$ is only symmetric about the middle degree in the displayed form, and outside that range the source or target no longer matches the Lefschetz representation. Thus the theorem gives both $b_{n-r}=b_{n+r}$ and injectivity chains below the middle, which imply the usual unimodality inequalities for Betti numbers in the relevant parity. Because $[\omega]$ has Hodge type $(1,1)$, the next point is to record how the same operation interacts with the Hodge decomposition rather than only with total degree. [quotetheorem:8063] [citeproof:8063] The type $(1,1)$ condition is the decisive input here. A closed degree-two class of mixed Hodge type would not preserve the diagonal shift $(p,q)\mapsto(p+r,q+r)$, even if it defined a cup-product map on total cohomology. The statement also does not say that every class in $H^{p+r,q+r}(X)$ is a power of $[\omega]$ times a lower class unless the degrees are paired by $p+q=n-r$; away from that symmetry it is only a type-preserving degree-raising map. This is why non-Kähler compact complex manifolds, where Hodge decomposition and Kähler identities can fail, do not provide a parallel theorem. Projective space is the cleanest case where the known cohomology and the Lefschetz isomorphisms together leave no primitive contribution outside the expected powers of the Kähler class. [example: Projective Space Cohomology from Lefschetz Theory] Let $X=\mathbb{CP}^n$ with the Fubini--Study Kähler form $\omega_{FS}$, and normalise $h=[\omega_{FS}]\in H^2(\mathbb{CP}^n;\mathbb C)$ to be the standard degree-two generator. The cellular decomposition of $\mathbb{CP}^n$ has one cell in each even real dimension $0,2,\ldots,2n$ and no odd-dimensional cells, so \begin{align*}H^{2k}(\mathbb{CP}^n;\mathbb C)=\mathbb C h^k \text{ for } 0\le k\le n,\qquad H^{2k+1}(\mathbb{CP}^n;\mathbb C)=0.\end{align*} Since $L$ is cup product with $h$, for $a\in\mathbb C$ and $0\le m\le n-r$ we have \begin{align*}L^r(a h^m)=h^r\smile a h^m=a(h^r\smile h^m)=a h^{m+r}.\end{align*} Now take the degree paired by Hard Lefschetz. If $n-r=2m$, then $n+r=2(m+r)$, and the map is \begin{align*}L^r:\mathbb C h^m\longrightarrow \mathbb C h^{m+r},\qquad a h^m\longmapsto a h^{m+r}.\end{align*} Both source and target are one-dimensional, and $h^{m+r}\ne 0$ because $m+r=(n+r)/2\le n$, so this map is an isomorphism. If $n-r$ is odd, then $n+r$ is also odd, and the same Lefschetz map is the zero vector space map $0\to 0$, again an isomorphism. The multiplication rule is therefore \begin{align*}h^a\smile h^b=h^{a+b}\text{ when }a+b\le n,\qquad h^a\smile h^b=0\text{ when }a+b\ge n+1,\end{align*} because $H^{2(a+b)}(\mathbb{CP}^n;\mathbb C)=0$ above degree $2n$. Hence \begin{align*}H^*(\mathbb{CP}^n;\mathbb C)\cong \mathbb C[h]/(h^{n+1}),\qquad \deg h=2.\end{align*} Thus projective space is the case where the Kähler class itself generates every cohomology group, while the fact that these groups are exactly the one-dimensional even groups comes from the cellular computation, not from Lefschetz theory alone. [/example] ## Primitive Cohomology and Lefschetz Decomposition Hard Lefschetz tells us that powers of $L$ move classes upward, but it does not by itself separate the new information from the part obtained by repeatedly multiplying lower-degree classes. The next question is how to identify the cohomology classes that are not produced from smaller degrees by $L$. [definition: Primitive Cohomology] Let $(X,\omega)$ be a compact Kähler manifold of complex dimension $n$. 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 C) \to H^{2n-k+2}(X;\mathbb C)\right). \end{align*} The primitive Hodge piece is \begin{align*} P^{p,q}(X) := P^{p+q}(X) \cap H^{p,q}(X). \end{align*} [/definition] The exponent $n-k+1$ is the first power that would push a degree-$k$ class past the top degree after the middle symmetry is accounted for. If primitive classes are the starting vectors for Lefschetz strings, the missing structural statement is that every cohomology class appears in exactly one finite sum of such strings. That uniqueness is what turns Hard Lefschetz from an isomorphism theorem into a decomposition theorem. [quotetheorem:3855] [citeproof:3855] This decomposition is the cohomological analogue of decomposing a representation into irreducible $\mathfrak{sl}_2$ strings. The Kähler identities are again essential: without the harmonic $\mathfrak{sl}_2$-action, cup product by a degree-two class need not split cohomology into finite Lefschetz strings, even if some individual cup-product maps happen to be injective. The theorem also does not determine the primitive spaces themselves; it says that once they are known, all higher Lefschetz powers are organised without overlap. In practice, to compute $P^k(X)$ one tests the single condition $L^{n-k+1}\alpha=0$ for $k\le n$, then builds the full degree-$m$ cohomology by adding the pieces $L^jP^{m-2j}(X)$. [example: Primitive Middle Cohomology of a Kähler Surface] Let $(X,\omega)$ be a compact Kähler surface, so $n=2$. By the definition of primitive cohomology, the degree-two primitive part is \begin{align*}P^2(X)=\ker\left(L^{2-2+1}:H^2(X;\mathbb C)\to H^{2+2}(X;\mathbb C)\right).\end{align*} Since $2-2+1=1$, this is \begin{align*}P^2(X)=\ker\left(L:H^2(X;\mathbb C)\to H^4(X;\mathbb C)\right).\end{align*} For a class $[\alpha]\in H^2(X;\mathbb C)$, the global Lefschetz operator is cup product with the Kähler class, so \begin{align*}L[\alpha]=[\omega]\smile[\alpha].\end{align*} Under the top-degree identification $H^4(X;\mathbb C)\cong \mathbb C$ given by integration over the fundamental class, this class is zero exactly when \begin{align*}\int_X \omega\smile\alpha=0.\end{align*} Thus $P^2(X)$ consists precisely of the middle-degree classes whose cup product with $[\omega]$ has zero top-degree integral. The degree-two Lefschetz decomposition has the two possible summands $j=0$ and $j=1$: \begin{align*}H^2(X;\mathbb C)=P^2(X)\oplus L P^0(X).\end{align*} Because $H^0(X;\mathbb C)=\mathbb C\cdot 1$ and $L(1)=[\omega]$, the second summand is \begin{align*}L P^0(X)=\mathbb C[\omega].\end{align*} Therefore \begin{align*}H^2(X;\mathbb C)=P^2(X)\oplus \mathbb C[\omega].\end{align*} Here $[\omega]\notin P^2(X)$, since \begin{align*}\int_X \omega\smile\omega=\int_X \omega^2>0\end{align*} for a Kähler form on a surface, so the Lefschetz line and the primitive part have zero intersection. The Hodge-type pieces follow from the fact that $L$ has type $(1,1)$. If $\alpha\in H^{2,0}(X)$, then \begin{align*}L\alpha\in H^{3,1}(X)=0,\end{align*} because a complex surface has no forms of type $(3,1)$. Hence $P^{2,0}(X)=H^{2,0}(X)$. Similarly, if $\alpha\in H^{0,2}(X)$, then \begin{align*}L\alpha\in H^{1,3}(X)=0,\end{align*} so $P^{0,2}(X)=H^{0,2}(X)$. Finally, \begin{align*}P^{1,1}(X)=\{\alpha\in H^{1,1}(X):[\omega]\smile\alpha=0\in H^{2,2}(X)\},\end{align*} or equivalently \begin{align*}P^{1,1}(X)=\left\{\alpha\in H^{1,1}(X):\int_X \omega\smile\alpha=0\right\}.\end{align*} Thus on a Kähler surface the middle cohomology splits into the Kähler line and the primitive hyperplane, with the $(2,0)$ and $(0,2)$ pieces automatically primitive and the primitive $(1,1)$ part cut out by cup-orthogonality to $[\omega]$. [/example] The surface calculation shows why primitive cohomology should not be treated as a minor correction term: in middle degree it can carry most of the geometry. It also depends on the Kähler class used to define $L$, so the same underlying complex manifold can have different primitive splittings as the Kähler class varies. [remark: Dependence on the Kähler Class] Primitive cohomology depends on the chosen Kähler class $[\omega]$, not merely on the complex manifold $X$. The total Hodge numbers are deformation-invariant in many Kähler families, but the primitive subspace can move as the Kähler class moves inside the Kähler cone. [/remark] ## Hodge--Riemann Bilinear Relations The Lefschetz decomposition describes the vector-space structure, but it does not yet describe the signs of the intersection form. The final question in this chapter is how the cup product pairing behaves on primitive Hodge pieces. [definition: Hodge--Riemann Form] Let $(X,\omega)$ be a compact Kähler manifold of complex dimension $n$. For $0\le k\le n$, the Hodge--Riemann bilinear form in degree $k$ is the map \begin{align*} Q_k &: H^k(X;\mathbb C)\times H^k(X;\mathbb C)\to \mathbb C \end{align*} defined by \begin{align*} Q_k(\alpha,\beta) := (-1)^{k(k-1)/2}\int_X \alpha \smile \beta \smile [\omega]^{n-k}. \end{align*} For $p+q=k$, the associated Hermitian form is the map \begin{align*} H_{p,q} &: H^{p,q}(X)\times H^{p,q}(X)\to \mathbb C \end{align*} defined by \begin{align*} H_{p,q}(\alpha,\beta) := i^{p-q}Q_k(\alpha,\overline{\beta}). \end{align*} [/definition] The signs in this definition are chosen so that positivity is formulated on primitive classes rather than on all classes at once. The ordinary intersection form mixes Lefschetz strings and changes sign along them, so we need a theorem that isolates the primitive Hodge pieces where the pointwise positivity from Kähler linear algebra survives integration over $X$. [quotetheorem:8065] [citeproof:8065] The primitivity hypothesis cannot be dropped. The raw intersection pairing changes sign along Lefschetz strings, so a non-primitive class may contain a Lefschetz multiple whose contribution has the opposite sign from the primitive input. The Hodge type hypothesis is also part of the sign rule: the factor $i^{p-q}$ is calibrated separately on each $H^{p,q}$ piece, not on an arbitrary mixed class. Thus the theorem refines [Poincaré duality](/theorems/2291) only after the Lefschetz decomposition and Hodge decomposition have both been imposed. In complex dimension two it recovers the familiar fact that the intersection form is negative definite on primitive real $(1,1)$-classes and positive on the Kähler line. [example: Signature Pattern on a Kähler Surface] Let $(X,\omega)$ be a compact Kähler surface, so $n=2$. The degree-two Lefschetz decomposition from the global [Lefschetz decomposition theorem](/theorems/3877) is \begin{align*}H^2(X;\mathbb R)=\mathbb R[\omega]\oplus P^2(X;\mathbb R).\end{align*} We compute the signs of the real intersection form \begin{align*}I(\alpha,\beta)=\int_X \alpha\smile\beta.\end{align*} First, on the Kähler line, a nonzero class has the form $a[\omega]$ with $a\in\mathbb R$ and $a\ne 0$. Then \begin{align*}I(a[\omega],a[\omega])=\int_X a[\omega]\smile a[\omega]=a^2\int_X \omega^2.\end{align*} Since $\omega$ is a Kähler form on a surface, $\omega^2$ is a positive volume form, so $\int_X\omega^2>0$. Hence \begin{align*}I(a[\omega],a[\omega])>0.\end{align*} Next let $\alpha\in P^{1,1}(X)\cap H^2(X;\mathbb R)$ be nonzero. For $k=2$, the Hodge--Riemann sign is \begin{align*}(-1)^{k(k-1)/2}=(-1)^{2\cdot 1/2}=-1.\end{align*} Since $p=q=1$, we also have \begin{align*}i^{p-q}=i^0=1.\end{align*} The *Hodge--Riemann Bilinear Relations* applied to the primitive $(1,1)$-class $\alpha$ give \begin{align*}1\cdot(-1)\int_X\alpha\smile\overline{\alpha}>0.\end{align*} Because $\alpha$ is real, $\overline{\alpha}=\alpha$, so \begin{align*}-\int_X\alpha\smile\alpha>0.\end{align*} Therefore \begin{align*}I(\alpha,\alpha)=\int_X\alpha\smile\alpha<0.\end{align*} Thus the intersection form is negative definite on primitive real $(1,1)$-classes. Finally take a nonzero class $\sigma\in H^{2,0}(X)$. It is primitive because $L\sigma\in H^{3,1}(X)=0$ on a surface. For $p=2$, $q=0$, and $k=2$, the sign factors are \begin{align*}i^{p-q}=i^2=-1\end{align*} and \begin{align*}(-1)^{k(k-1)/2}=-1.\end{align*} Hence the *Hodge--Riemann Bilinear Relations* give \begin{align*}(-1)(-1)\int_X\sigma\smile\overline{\sigma}>0,\end{align*} so \begin{align*}\int_X\sigma\smile\overline{\sigma}>0.\end{align*} The associated real two-plane is spanned by \begin{align*}u=\sigma+\overline{\sigma}\end{align*} and \begin{align*}v=i(\sigma-\overline{\sigma}).\end{align*} Since $\sigma\smile\sigma$ has type $(4,0)$ and $\overline{\sigma}\smile\overline{\sigma}$ has type $(0,4)$, both vanish on a complex surface. Therefore \begin{align*}I(u,u)=\int_X(\sigma+\overline{\sigma})\smile(\sigma+\overline{\sigma})=2\int_X\sigma\smile\overline{\sigma}>0.\end{align*} Similarly, \begin{align*}I(v,v)=\int_X i(\sigma-\overline{\sigma})\smile i(\sigma-\overline{\sigma})=2\int_X\sigma\smile\overline{\sigma}>0.\end{align*} Also \begin{align*}I(u,v)=\int_X(\sigma+\overline{\sigma})\smile i(\sigma-\overline{\sigma})=0,\end{align*} because the $(4,0)$ and $(0,4)$ terms vanish and the two mixed terms cancel. Thus each real plane coming from $H^{2,0}(X)\oplus H^{0,2}(X)$ is positive definite. So on a Kähler surface the intersection form is positive on $\mathbb R[\omega]$, positive on the real part of $H^{2,0}(X)\oplus H^{0,2}(X)$, and negative on primitive real $(1,1)$-classes; its signs are dictated by Hodge type and primitiveness. [/example] The surface signature pattern is the first place where the theorem produces concrete restrictions beyond dimension counts. In higher dimensions the same principle applies after decomposing into primitive pieces and Lefschetz powers, so the three results of the chapter should be read as a single package rather than as separate facts. [explanation: How the Three Results Fit Together] Hard Lefschetz supplies the isomorphisms $L^r:H^{n-r}(X)\to H^{n+r}(X)$, so the Kähler class controls movement across the middle degree. The Lefschetz decomposition then identifies primitive classes as the independent inputs from which the rest of cohomology is generated. The Hodge--Riemann bilinear relations put a positivity theorem on those primitive inputs, turning the decomposition into a signed orthogonal structure. This package is the cohomological heart of compact Kähler geometry and is the bridge from Hodge decomposition to later applications such as restrictions on Betti numbers, intersection forms, and algebraic cycles. [/explanation] # 8. Hodge Structures The preceding chapters established the analytic and cohomological heart of Kähler geometry: harmonic representatives, the Hodge decomposition, the $\partial\bar\partial$ lemma, and the Lefschetz decomposition. This chapter repackages those results into the language of Hodge structures, which separates the linear-algebraic structure carried by cohomology from the particular manifold that produced it. The guiding question is how much of the complex geometry of a compact Kähler manifold is remembered by its cohomology groups, once their decompositions, filtrations, and polarizations are recorded. ## Pure Hodge Structures from Kähler Cohomology The first problem is to express the Hodge decomposition as an intrinsic structure on a rational or real vector space. The decomposition \begin{align*} H^k(X,\mathbb C)=\bigoplus_{p+q=k}H^{p,q}(X) \end{align*} is obtained after extending scalars to $\mathbb C$, but the underlying topological cohomology group is defined over $\mathbb Z$, $\mathbb Q$, or $\mathbb R$. Hodge structures are the bookkeeping device that remembers both layers at once. A pure Hodge structure begins with a vector space over a smaller field and adds a complex decomposition with a compatibility condition under conjugation. [definition: Pure Hodge Structure] Let $V_{\mathbb Q}$ be a finite-dimensional vector space over $\mathbb Q$, let $V_{\mathbb R}:=V_{\mathbb Q}\otimes_{\mathbb Q}\mathbb R$, and let $k\in\mathbb Z$. A pure Hodge structure of weight $k$ on $V_{\mathbb Q}$ is a direct sum decomposition \begin{align*} V_{\mathbb C} := V_{\mathbb Q}\otimes_{\mathbb Q}\mathbb C = \bigoplus_{p+q=k} V^{p,q} \end{align*} such that complex conjugation on $V_{\mathbb C}$ induced by the real form satisfies \begin{align*} \overline{V^{p,q}} = V^{q,p}. \end{align*} [/definition] The weight records the cohomological degree in geometric examples. The conjugation condition is the point that prevents the decomposition from being merely a decomposition of a complex vector space; it ties the pieces back to the rational structure. [example: Hodge Structure on First Cohomology of a Curve] Let $V_{\mathbb Q}=H^1(C,\mathbb Q)$, so by extension of scalars \begin{align*} V_{\mathbb C}=H^1(C,\mathbb Q)\otimes_{\mathbb Q}\mathbb C=H^1(C,\mathbb C). \end{align*} The Dolbeault decomposition in total degree $1$ gives \begin{align*} H^1(C,\mathbb C)=H^{1,0}(C)\oplus H^{0,1}(C). \end{align*} Here \begin{align*} H^{1,0}(C)=H^0(C,\Omega_C^1). \end{align*} Since the genus of $C$ is $g$, the space of holomorphic $1$-forms has dimension \begin{align*} \dim_{\mathbb C}H^{1,0}(C)=\dim_{\mathbb C}H^0(C,\Omega_C^1)=g. \end{align*} Also \begin{align*} H^{0,1}(C)=H^1(C,\mathcal O_C), \end{align*} and *Serre duality for curves* identifies $H^1(C,\mathcal O_C)$ with the dual of $H^0(C,\Omega_C^1)$, so \begin{align*} \dim_{\mathbb C}H^{0,1}(C)=\dim_{\mathbb C}H^1(C,\mathcal O_C)=\dim_{\mathbb C}H^0(C,\Omega_C^1)=g. \end{align*} Complex conjugation on differential forms sends a local holomorphic form $f(z)\,dz$ to the anti-holomorphic form $\overline{f(z)}\,d\bar z$, so it sends $H^{1,0}(C)$ onto $H^{0,1}(C)$. Applying conjugation again returns the original class, hence it also sends $H^{0,1}(C)$ onto $H^{1,0}(C)$. Therefore \begin{align*} \overline{H^{1,0}(C)}=H^{0,1}(C). \end{align*} Since the only pairs with $p+q=1$ are $(1,0)$ and $(0,1)$, this decomposition satisfies exactly the conjugation condition for a pure Hodge structure of weight $1$ on $H^1(C,\mathbb Q)$. [/example] This example already shows why the rational structure matters. The complex subspace $H^{1,0}(C)$ is not usually obtained by complexifying a rational subspace of $H^1(C,\mathbb Q)$; its position inside $H^1(C,\mathbb C)$ varies holomorphically in families of curves. The natural next question is whether the decomposition produced earlier by harmonic forms satisfies exactly the rational compatibility demanded in the definition, for every cohomological degree of every compact Kähler manifold. [quotetheorem:8066] [citeproof:8066] Thus the analytic decomposition from harmonic forms becomes an algebraic structure on cohomology. The hypotheses are doing real work: compactness gives the Hodge theorem for the elliptic Laplacian, while the Kähler identities force the Laplacian to respect bidegree. On a general compact complex manifold the Frölicher spectral sequence need not degenerate at the first page, so de Rham cohomology need not split as a direct sum of Dolbeault groups; Hopf surfaces give standard non-Kähler examples where the expected Kähler-style Hodge decomposition fails. The theorem therefore records a special rigidity of compact Kähler geometry, not a formal consequence of having a complex structure. Once cohomology groups carry this extra structure, the next issue is functoriality. Geometry rarely studies a manifold in isolation: holomorphic maps, inclusions of subvarieties, projections from products, and correspondences all induce linear maps on cohomology. If these maps could mix Hodge summands arbitrarily, then the Hodge decomposition would be a fragile decoration rather than a structure preserved by the constructions used throughout the course. The right notion of map must therefore impose two constraints at once. It must be defined over the rational vector spaces, so that it remembers the underlying topological cohomology, and after extension to $\mathbb C$ it must preserve the bidegree decomposition. The following definition packages exactly this compatibility. [definition: Morphism of Pure Hodge Structures] Let $V_{\mathbb Q}$ and $W_{\mathbb Q}$ carry pure Hodge structures of weights $k$ and $\ell$. A morphism of pure Hodge structures is a $\mathbb Q$-linear map $f:V_{\mathbb Q}\to W_{\mathbb Q}$ such that the complexification $f_{\mathbb C}:V_{\mathbb C}\to W_{\mathbb C}$ satisfies \begin{align*} f_{\mathbb C}(V^{p,q})\subset W^{p,q} \end{align*} for all $p,q$ when $k=\ell$, and is the zero map when the weights differ. [/definition] The condition says that a map respects the decomposition degree by degree. Pullback maps induced by holomorphic maps are the basic geometric source of such morphisms. [example: Pullback Preserves Hodge Type] Let $f:X\to Y$ be holomorphic, and let $[\alpha]\in H^{p,q}(Y)\subset H^k(Y,\mathbb C)$ with $p+q=k$. Choose local holomorphic coordinates $w_1,\dots,w_m$ on $Y$ and $z_1,\dots,z_n$ on $X$. A local $(p,q)$-form on $Y$ is a sum of terms \begin{align*} a_{I,J}(w,\bar w)\,dw_{i_1}\wedge\cdots\wedge dw_{i_p}\wedge d\bar w_{j_1}\wedge\cdots\wedge d\bar w_{j_q}. \end{align*} Since $f$ is holomorphic, each coordinate function $w_a\circ f$ depends holomorphically on the $z$-coordinates, so \begin{align*} f^*(dw_a)=d(w_a\circ f)=\sum_{\ell=1}^n \frac{\partial(w_a\circ f)}{\partial z_\ell}\,dz_\ell. \end{align*} Taking complex conjugates gives \begin{align*} f^*(d\bar w_b)=d(\overline{w_b\circ f})=\sum_{\ell=1}^n \overline{\frac{\partial(w_b\circ f)}{\partial z_\ell}}\,d\bar z_\ell. \end{align*} Therefore the pullback of each displayed local term is a sum of wedge products with exactly $p$ factors of type $dz_\ell$ and exactly $q$ factors of type $d\bar z_\ell$, so $f^*\alpha$ has type $(p,q)$. The exterior derivative commutes with pullback: \begin{align*} d(f^*\alpha)=f^*(d\alpha). \end{align*} Thus closed forms pull back to closed forms, and if $\alpha-\alpha'=d\gamma$, then \begin{align*} f^*\alpha-f^*\alpha'=f^*(d\gamma)=d(f^*\gamma). \end{align*} So pullback is well-defined on cohomology and sends $H^{p,q}(Y)$ into $H^{p,q}(X)$. Since the topological pullback $f^*:H^k(Y,\mathbb Q)\to H^k(X,\mathbb Q)$ is $\mathbb Q$-linear and its complexification is the de Rham pullback above, it is a morphism of pure Hodge structures of weight $k$. [/example] The language of morphisms will matter whenever Hodge theory is used functorially: inclusions, projections, products, and correspondences all become maps constrained by Hodge type. ## Hodge Filtrations and Hodge Numbers The next problem is that a direct sum decomposition is sometimes less natural than a decreasing filtration. Filtrations interact well with holomorphic variation and with spectral sequences, so Hodge theory is often expressed using $F^p$ rather than the individual summands $H^{p,q}$. Given a pure Hodge decomposition, the Hodge filtration keeps all pieces with first index at least a fixed value. [definition: Hodge Filtration] Let $V_{\mathbb Q}$ carry a pure Hodge structure of weight $k$. The associated Hodge filtration is the decreasing filtration of $V_{\mathbb C}$ defined by \begin{align*} F^pV_{\mathbb C} := \bigoplus_{r\ge p} V^{r,k-r}. \end{align*} [/definition] The filtration forgets the direct projection onto each summand, but it does not lose information when it is paired with conjugation. The next characterization is the reason filtrations can replace decompositions in much of Hodge theory. [quotetheorem:8067] [citeproof:8067] This theorem explains why Hodge filtrations appear in deformation theory: the filtration can move holomorphically even when the individual conjugate pieces require both holomorphic and anti-holomorphic data. In a variation of Hodge structure, the fixed rational vector space is transported by a flat connection, while the moving point in the classifying space is usually recorded by the subspaces $F^p$. Period maps are built from this filtration-valued motion, and Griffiths transversality later becomes a differential condition on how $F^p$ is allowed to move into $F^{p-1}$. The opposite-filtration condition is the essential nondegeneracy requirement, and it depends on the chosen rational real form. For a specific failure, take $V_{\mathbb Q}=\mathbb Q^2$, so $V_{\mathbb C}=\mathbb C^2$ with conjugation acting coordinatewise, and set \begin{align*} F^1=\mathbb C(1,0)\subset V_{\mathbb C}. \end{align*} Then $F^1=\overline{F^1}$, so $F^1+\overline{F^1}$ is only one-dimensional and cannot equal $V_{\mathbb C}$. The dimensions look like those of a weight-one Hodge filtration, but the line is defined over the real form rather than paired with a distinct conjugate line. A genuine weight-one Hodge structure instead requires a complex line $F^1$ satisfying $V_{\mathbb C}=F^1\oplus\overline{F^1}$; for example $\mathbb C(1,i)$ works for this real form because its conjugate is $\mathbb C(1,-i)$. Thus the filtration alone is not a Hodge structure unless it is transverse to its conjugate relative to the rational structure. The characterization is limited to the decomposition data. It recovers the pure Hodge structure from the filtration and the rational real form, but it does not include an independent polarization, an integral lattice, or the arithmetic conditions that arise in projective geometry. Those extra structures are added later when one wants a polarized Hodge structure, a variation with integral monodromy, or the Hodge structure attached to an abelian variety. In many geometric questions the individual subspaces $H^{p,q}(X)$ are too delicate to compare directly: they can move in families, depend on complex structure, and are not determined by the underlying topological cohomology alone. What remains immediately comparable is the size of each bidegree piece. These dimensions give the numerical shadow of the Hodge decomposition, refining Betti numbers by recording how much cohomology lies in each type. [definition: Hodge Numbers] Let $X$ be a compact Kähler manifold. The Hodge number $h^{p,q}(X)$ is \begin{align*} h^{p,q}(X) := \dim_{\mathbb C}H^{p,q}(X). \end{align*} [/definition] The Betti number $b_k(X)=\dim_{\mathbb Q}H^k(X,\mathbb Q)$ is recovered by summing the Hodge numbers on the diagonal $p+q=k$. The definition raises the immediate constraint question: because the Hodge pieces are paired by conjugation, the table of numbers should be symmetric across the line $p=q$. [quotetheorem:8081] [citeproof:8081] Together with Serre duality, this symmetry gives the familiar symmetries of the Hodge diamond. The point here is that $h^{p,q}=h^{q,p}$ is already built into the real structure of a pure Hodge structure, and for compact Kähler manifolds that real structure is compatible with harmonic type decomposition. Outside the Kähler setting, the Dolbeault numbers of a compact complex manifold need not be arranged by de Rham cohomology in the same way, and the resulting table need not satisfy all Kähler Hodge-diamond constraints. The equality also does not determine the individual Hodge numbers from the Betti numbers; it only pairs entries across the diagonal. [example: Hodge Diamond of a Compact Curve] For a connected compact Riemann surface $C$ of genus $g$, complex dimension is $1$, so $H^{p,q}(C)=0$ unless $0\le p,q\le 1$. In degree $0$, connectedness gives $H^0(C,\mathbb C)\cong \mathbb C$, and the only Hodge summand with $p+q=0$ is $H^{0,0}(C)$, hence \begin{align*} h^{0,0}(C)=\dim_{\mathbb C}H^{0,0}(C)=\dim_{\mathbb C}H^0(C,\mathbb C)=1. \end{align*} By definition of the genus of a compact Riemann surface, \begin{align*} h^{1,0}(C)=\dim_{\mathbb C}H^{1,0}(C)=\dim_{\mathbb C}H^0(C,\Omega_C^1)=g. \end{align*} Complex conjugation identifies $H^{1,0}(C)$ anti-linearly with $H^{0,1}(C)$, so the two spaces have the same complex dimension: \begin{align*} h^{0,1}(C)=h^{1,0}(C)=g. \end{align*} Finally, $C$ is a connected compact oriented real surface, so *Poincare duality* gives $H^2(C,\mathbb C)\cong \mathbb C$; since the only Hodge summand with $p+q=2$ on a curve is $H^{1,1}(C)$, we get \begin{align*} h^{1,1}(C)=\dim_{\mathbb C}H^{1,1}(C)=\dim_{\mathbb C}H^2(C,\mathbb C)=1. \end{align*} Thus the nonzero Hodge numbers are \begin{align*} h^{0,0}(C)=1,\qquad h^{1,0}(C)=g,\qquad h^{0,1}(C)=g,\qquad h^{1,1}(C)=1. \end{align*} Arranged by cohomological degree, the Hodge diamond has $1$ in degree $0$, the pair $g,g$ in degree $1$, and $1$ in degree $2$; the middle row records holomorphic $1$-forms and their conjugates. [/example] The same method gives compact tables in higher-dimensional examples. Projective space is the model where almost all Hodge numbers vanish. [example: Hodge Diamond of Projective Space] For $\mathbb{CP}^n$, let $\eta\in H^2(\mathbb{CP}^n,\mathbb C)$ be the hyperplane class. The cohomology ring is \begin{align*} H^*(\mathbb{CP}^n,\mathbb C)\cong \mathbb C[\eta]/(\eta^{n+1}), \end{align*} where $\eta^r$ has cohomological degree $2r$. Thus \begin{align*} H^{2r}(\mathbb{CP}^n,\mathbb C)=\mathbb C\eta^r\quad\text{for }0\le r\le n, \end{align*} and \begin{align*} H^{m}(\mathbb{CP}^n,\mathbb C)=0\quad\text{if }m\text{ is odd}. \end{align*} The Fubini-Study form represents $\eta$ and is a $(1,1)$-form, so $\eta\in H^{1,1}(\mathbb{CP}^n)$. Since cup product adds Hodge type, the class $\eta^r$ has type $(r,r)$. Therefore the one-dimensional vector space $H^{2r}(\mathbb{CP}^n,\mathbb C)=\mathbb C\eta^r$ is contained in $H^{r,r}(\mathbb{CP}^n)$. The Hodge decomposition in degree $2r$ is \begin{align*} H^{2r}(\mathbb{CP}^n,\mathbb C)=\bigoplus_{p+q=2r}H^{p,q}(\mathbb{CP}^n). \end{align*} Because the whole degree-$2r$ cohomology is already the line $\mathbb C\eta^r$ of type $(r,r)$, we get \begin{align*} H^{r,r}(\mathbb{CP}^n)=\mathbb C\eta^r. \end{align*} Hence \begin{align*} h^{r,r}(\mathbb{CP}^n)=\dim_{\mathbb C}H^{r,r}(\mathbb{CP}^n)=1 \end{align*} for $0\le r\le n$. If $p\ne q$ and $H^{p,q}(\mathbb{CP}^n)$ were nonzero, then it would contribute to $H^{p+q}(\mathbb{CP}^n,\mathbb C)$. When $p+q$ is odd, that cohomology group is zero. When $p+q=2r$ is even, the whole degree-$2r$ cohomology is the line $H^{r,r}(\mathbb{CP}^n)$, so no other Hodge summand can occur. Thus \begin{align*} h^{p,q}(\mathbb{CP}^n)=0\quad\text{for }p\ne q. \end{align*} The Hodge diamond of projective space is therefore concentrated on the diagonal, with a single $1$ in each bidegree $(r,r)$ for $0\le r\le n$. [/example] Complex tori give a contrasting example to projective space. Instead of a single generator in type $(1,1)$, the first cohomology already has holomorphic and anti-holomorphic generators, and the full cohomology is built by taking exterior products of them. The Hodge numbers are therefore binomial coefficients, reflecting the choices of how many holomorphic and anti-holomorphic differentials appear in a wedge product. [example: Hodge Numbers of a Complex Torus] Let $T=\mathbb C^n/\Lambda$ have coordinates $z_a=x_a+iy_a$ on $\mathbb C^n$. For the flat metric, harmonic forms on $T$ are exactly the translation-invariant forms: expanding a form into Fourier modes on the real torus, the Laplacian multiplies a nonzero mode by a positive number, so only the zero mode can be harmonic. Thus cohomology is represented by the exterior algebra on the constant $1$-forms $dz_1,\dots,dz_n,d\bar z_1,\dots,d\bar z_n$. A basis element of type $(p,q)$ is obtained by choosing indices $I=\{i_1<\cdots<i_p\}\subset\{1,\dots,n\}$ and $J=\{j_1<\cdots<j_q\}\subset\{1,\dots,n\}$ and forming \begin{align*} dz_I\wedge d\bar z_J:=dz_{i_1}\wedge\cdots\wedge dz_{i_p}\wedge d\bar z_{j_1}\wedge\cdots\wedge d\bar z_{j_q}. \end{align*} The exterior algebra basis is alternating, so different ordered subsets $I$ and $J$ give linearly independent classes, and every translation-invariant $(p,q)$-form is a unique complex linear combination of these classes. There are $\binom{n}{p}$ choices for $I$ and $\binom{n}{q}$ choices for $J$, hence \begin{align*} \dim_{\mathbb C}H^{p,q}(T)=\binom{n}{p}\binom{n}{q}. \end{align*} Therefore \begin{align*} h^{p,q}(T)=\binom{n}{p}\binom{n}{q}. \end{align*} The binomial factors record the independent choices of holomorphic and anti-holomorphic differentials in a translation-invariant representative. [/example] The curve, projective-space, and torus examples display the same formal symmetries but with different distributions of cohomology across bidegrees. In complex dimension two the table is still small enough to write down abstractly, while already containing the middle cohomology where much of surface geometry lives. This makes surfaces a useful transition from numerical Hodge diamonds to the intersection forms and primitive classes studied in the next section. [example: Hodge Diamond of a Kahler Surface] Let $S$ be a connected compact Kähler surface, so $\dim_{\mathbb C}S=2$ and $H^{p,q}(S)=0$ unless $0\le p,q\le 2$. Connectedness gives $H^0(S,\mathbb C)\cong\mathbb C$, and the only summand in degree $0$ is $H^{0,0}(S)$, so \begin{align*} h^{0,0}(S)=\dim_{\mathbb C}H^{0,0}(S)=\dim_{\mathbb C}H^0(S,\mathbb C)=1. \end{align*} By *Serre duality for surfaces*, $H^{0,0}(S)$ is dual to $H^{2,2}(S)$, hence \begin{align*} h^{2,2}(S)=h^{0,0}(S)=1. \end{align*} Complex conjugation gives anti-linear isomorphisms $H^{p,q}(S)\cong H^{q,p}(S)$, so \begin{align*} h^{1,0}(S)=h^{0,1}(S). \end{align*} The same conjugation symmetry also gives \begin{align*} h^{2,1}(S)=h^{1,2}(S). \end{align*} By *Serre duality for surfaces*, \begin{align*} h^{1,0}(S)=h^{2-1,2-0}(S)=h^{1,2}(S). \end{align*} Combining these equalities gives \begin{align*} h^{1,0}(S)=h^{0,1}(S)=h^{2,1}(S)=h^{1,2}(S). \end{align*} Similarly, conjugation gives \begin{align*} h^{2,0}(S)=h^{0,2}(S). \end{align*} In degree $2$, the Hodge decomposition is \begin{align*} H^2(S,\mathbb C)=H^{2,0}(S)\oplus H^{1,1}(S)\oplus H^{0,2}(S). \end{align*} Therefore \begin{align*} b_2(S)=\dim_{\mathbb C}H^2(S,\mathbb C)=h^{2,0}(S)+h^{1,1}(S)+h^{0,2}(S). \end{align*} Using $h^{2,0}(S)=h^{0,2}(S)$, this becomes \begin{align*} b_2(S)=2h^{2,0}(S)+h^{1,1}(S). \end{align*} Thus the center entry $h^{1,1}(S)$ is exactly the part of middle cohomology left after removing the holomorphic two-forms and their conjugates. [/example] These examples show that Hodge numbers refine Betti numbers in a way sensitive to the complex structure. The remaining question is how the Kähler class adds a bilinear form to this data, using the primitive decomposition and Hodge-Riemann signs proved in Chapter 7. For computations, the procedure is usually the same across examples. First compute or identify $H^k(X,\mathbb C)$, often using de Rham representatives, a cohomology ring presentation, or invariant forms. Next determine which representatives have type $(p,q)$, then take dimensions to obtain $h^{p,q}$ and check the constraints from conjugation, Serre duality, and the Betti numbers. When a Kähler class is fixed, primitivity is checked by applying the relevant power of $L(\alpha)=[\omega]\smile\alpha$; in middle degree on a surface, this reduces to the single intersection condition \begin{align*} \int_S\alpha\smile[\omega]=0. \end{align*} This checklist prevents two common mistakes: counting Dolbeault classes without verifying their contribution to de Rham cohomology, and treating a Lefschetz translate as primitive. ## Polarizations and Primitive Cohomology The final problem in this chapter is to record the positivity hidden inside the Kähler metric. A pure Hodge structure alone is a decomposition with conjugation; a polarization is the extra bilinear form that remembers the Kähler class through integration. Let $X$ be a compact Kähler manifold of complex dimension $n$, and let $\omega$ be a Kähler form. The Lefschetz operator is \begin{align*} L:H^k(X,\mathbb C)\to H^{k+2}(X,\mathbb C), \qquad L(\alpha)=[\omega]\smile \alpha. \end{align*} Hard Lefschetz says that powers of $L$ identify complementary degrees. Primitive cohomology isolates the part of a degree that is not obtained by applying $L$ to lower-degree classes. [definition: Primitive Cohomology] Let $X$ be a compact Kähler manifold of complex dimension $n$ with Kähler class $[\omega]$. For $0\le k\le n$, the complex primitive cohomology in degree $k$ is \begin{align*} P^k(X,\mathbb C):=\ker\left(L^{n-k+1}:H^k(X,\mathbb C)\to H^{2n-k+2}(X,\mathbb C)\right). \end{align*} The real primitive cohomology is \begin{align*} P^k(X,\mathbb R):=\ker\left(L^{n-k+1}:H^k(X,\mathbb R)\to H^{2n-k+2}(X,\mathbb R)\right). \end{align*} If $[\omega]\in H^2(X,\mathbb Q)$, the rational primitive cohomology is \begin{align*} P^k(X,\mathbb Q):=\ker\left(L^{n-k+1}:H^k(X,\mathbb Q)\to H^{2n-k+2}(X,\mathbb Q)\right). \end{align*} [/definition] Primitive cohomology is compatible with Hodge type because $L$ has type $(1,1)$. To use it effectively, we need to know that every cohomology class is assembled in a controlled way from primitive classes and their Lefschetz translates. [quotetheorem:3877] [citeproof:3877] The Lefschetz decomposition reduces many questions about all cohomology to questions about primitive generators. Its force depends on Hard Lefschetz: for a general graded algebra with a degree-two operator, kernels of high powers need not generate the whole space, and the resulting strings may fail to have the dimensions predicted by $\mathfrak{sl}_2$ theory. The statement is also additive rather than multiplicative; it decomposes cohomology as a vector space, but it does not by itself describe cup products between primitive summands. This motivates defining the polarization on primitive cohomology first, rather than trying to put one undifferentiated positivity statement on all of $H^k(X)$. The ordinary cup-product pairing on middle cohomology mixes several Lefschetz strings, and its signs change as powers of $L$ are applied. Primitive classes are the generators of those strings, so the next object must measure their pairings before the full Lefschetz decomposition reassembles the remaining cohomology. The required pairing has to do three jobs at once. It must wedge two degree-$k$ primitive classes with enough powers of the Kähler class to reach top degree, include the parity sign that makes the Hodge-Riemann positivity statement uniform, and remember whether the Kähler class is merely real or actually rational. The following definition isolates exactly this bilinear form on the primitive generators; it is the form that will become the polarization in the Hodge-Riemann theorem. [definition: Polarization on Primitive Cohomology] Let $X$ be a compact Kähler manifold of complex dimension $n$ with Kähler class $[\omega]$, and let $0\le k\le n$. The real bilinear form on $P^k(X,\mathbb R)$ is the map \begin{align*} Q_\omega:P^k(X,\mathbb R)\times P^k(X,\mathbb R)\to \mathbb R \end{align*} defined by \begin{align*} Q_\omega(\alpha,\beta):=(-1)^{k(k-1)/2}\int_X \alpha\smile\beta\smile [\omega]^{n-k}. \end{align*} If $[\omega]\in H^2(X,\mathbb Q)$, the same formula restricts to a $\mathbb Q$-valued form \begin{align*} Q_\omega:P^k(X,\mathbb Q)\times P^k(X,\mathbb Q)\to \mathbb Q. \end{align*} [/definition] This form is symmetric for even $k$ and skew-symmetric for odd $k$, as required by the parity of a polarized Hodge structure. The sign is chosen so that the Hermitian form in the Hodge-Riemann relations has the correct positivity. [quotetheorem:8068] [citeproof:8068] The theorem is the point where the metric positivity of $\omega$ becomes a positivity statement about cohomology. The primitive hypothesis is necessary because the sign of the intersection pairing changes along Lefschetz strings; the Hodge-Riemann relations isolate the part where the Weil-operator correction gives a positive Hermitian form. The Kähler condition supplies both Hard Lefschetz and the star-operator identity used in the proof, while rationality of $[\omega]$ is a separate arithmetic condition needed only when the polarization is required to live over $\mathbb Q$. Thus the positivity statement is not a blanket positivity of the cup-product pairing on all cohomology, nor does an arbitrary real Kähler class automatically define a rational polarized Hodge structure. It is the linear-algebraic form of the Kähler condition. When the Kähler class is rational or integral, the same form also connects the Hodge structure to projective geometry: an ample line bundle supplies the class, and the resulting polarization is part of the data that distinguishes projective complex tori, namely abelian varieties, from arbitrary complex tori. [example: Polarization on First Cohomology of a Complex Torus] Let $T=\mathbb C^n/\Lambda$ carry the flat Kähler form \begin{align*} \omega=\frac{i}{2}\sum_{a,b}h_{a\bar b}\,dz_a\wedge d\bar z_b \end{align*} with $(h_{a\bar b})$ positive Hermitian. Since $k=1$, the primitive condition is \begin{align*} L^n:H^1(T,\mathbb C)\to H^{2n+1}(T,\mathbb C). \end{align*} But $H^{2n+1}(T,\mathbb C)=0$ because $T$ has real dimension $2n$, so $L^n=0$ and every class in $H^1(T,\mathbb R)$ is primitive. Thus \begin{align*} Q_\omega(\alpha,\beta)=\int_T\alpha\smile\beta\smile[\omega]^{n-1}. \end{align*} Take $\alpha\in H^{1,0}(T)$ and write it in constant coordinates as \begin{align*} \alpha=\sum_{a=1}^n c_a\,dz_a. \end{align*} At the level of translation-invariant forms, choose complex linear coordinates $u_1,\dots,u_n$ in which the Hermitian matrix is the identity. Then \begin{align*} \omega=\frac{i}{2}\sum_{a=1}^n du_a\wedge d\bar u_a. \end{align*} Writing $\alpha=\sum_a c_a\,du_a$, we have \begin{align*} \alpha\wedge\bar\alpha=\sum_{a,b}c_a\bar c_b\,du_a\wedge d\bar u_b. \end{align*} In the product $\alpha\wedge\bar\alpha\wedge\omega^{n-1}$, the term with $a\ne b$ contains either $du_a$ twice or $d\bar u_b$ twice after wedging with any nonzero summand of $\omega^{n-1}$, so it vanishes by alternatingness of the exterior product. For $a=b$, the only surviving part of $\omega^{n-1}$ is the product of the factors with indices different from $a$, and there are $(n-1)!$ such ordered contributions. Hence \begin{align*} i\,\alpha\wedge\bar\alpha\wedge\omega^{n-1}=\frac{2}{n}\left(\sum_{a=1}^n |c_a|^2\right)\omega^n. \end{align*} Therefore \begin{align*} iQ_\omega(\alpha,\bar\alpha)=\int_T i\,\alpha\wedge\bar\alpha\wedge\omega^{n-1}=\frac{2}{n}\left(\sum_{a=1}^n |c_a|^2\right)\int_T\omega^n. \end{align*} Since $\omega$ is Kähler, $\int_T\omega^n>0$, and since $\alpha\ne 0$ implies $\sum_a|c_a|^2>0$, we get \begin{align*} iQ_\omega(\alpha,\bar\alpha)>0. \end{align*} In the original coordinates, the same computation reads \begin{align*} iQ_\omega(\alpha,\bar\alpha)=\frac{2}{n}\left(\sum_{a,b}h^{a\bar b}c_a\bar c_b\right)\int_T\omega^n, \end{align*} where $(h^{a\bar b})$ is the inverse Hermitian matrix. Thus the polarization on $H^1$ is the real symplectic intersection form determined by the Kähler class, and it restricts to a rational polarization on $H^1(T,\mathbb Q)$ when $[\omega]\in H^2(T,\mathbb Q)$. [/example] Projective space has no primitive middle complication beyond the beginning of the Lefschetz string. Its cohomology is generated by repeated multiplication by the hyperplane class, so the primitive definition can be checked without keeping track of several independent strings. This makes it a useful test case for the indexing in the kernel condition. [example: Primitive Cohomology of Projective Space] For $\mathbb{CP}^n$, let $\eta=[\omega_{FS}]\in H^2(\mathbb{CP}^n,\mathbb C)$. The cohomology ring is \begin{align*} H^*(\mathbb{CP}^n,\mathbb C)\cong \mathbb C[\eta]/(\eta^{n+1}). \end{align*} Thus $\eta^j\ne 0$ for $0\le j\le n$ and $\eta^j=0$ for $j\ge n+1$. Primitive cohomology is only defined here for degrees $k\le n$. If $k$ is odd, then \begin{align*} H^k(\mathbb{CP}^n,\mathbb C)=0, \end{align*} so there are no primitive classes in that degree. Now take an even degree $k=2r\le n$. Then \begin{align*} H^{2r}(\mathbb{CP}^n,\mathbb C)=\mathbb C\eta^r. \end{align*} The primitive condition in degree $2r$ is \begin{align*} L^{n-2r+1}(\eta^r)=0. \end{align*} Since $L$ is cup product with $\eta$, applying $L^{n-2r+1}$ gives \begin{align*} L^{n-2r+1}(\eta^r)=\eta^{r+n-2r+1}=\eta^{n-r+1}. \end{align*} If $r=0$, this is $\eta^{n+1}$, which is zero in $\mathbb C[\eta]/(\eta^{n+1})$. Hence $1=\eta^0$ is primitive. If $r\ge 1$, then \begin{align*} n-r+1\le n, \end{align*} so $\eta^{n-r+1}\ne 0$ in the cohomology ring. Therefore no nonzero multiple of $\eta^r$ is primitive for $r\ge 1$. Thus the only nonzero primitive class in degrees at most $n$ is the unit class in $P^0(\mathbb{CP}^n,\mathbb C)$. Every positive-degree cohomology class is obtained from it by repeated multiplication by $L$, so the cohomology of projective space is one Lefschetz string generated by $1$. [/example] On a Kähler surface, primitive cohomology is especially geometric because it sits inside $H^2$ as the orthogonal complement of the Kähler class. This is the first dimension where the primitive part controls a familiar intersection-form statement rather than only checking definitions. The surface case therefore connects the abstract Hodge-Riemann relations with the classical Hodge index theorem. [example: Primitive Classes on a Kahler Surface] Let $S$ be a connected compact Kähler surface with Kähler class $[\omega]$. Since $\dim_{\mathbb C}S=2$, the primitive condition in degree $2$ is the kernel condition for \begin{align*} L:H^2(S,\mathbb C)\to H^4(S,\mathbb C),\qquad L(\alpha)=\alpha\smile[\omega]. \end{align*} The group $H^4(S,\mathbb C)$ is one-dimensional, and integration over the fundamental class identifies it with $\mathbb C$. Hence \begin{align*} \alpha\in P^2(S,\mathbb C) \end{align*} is equivalent to \begin{align*} \alpha\smile[\omega]=0\in H^4(S,\mathbb C), \end{align*} which is equivalent to \begin{align*} \int_S\alpha\smile[\omega]=0. \end{align*} For degree $2$ on a surface, the polarization form is \begin{align*} Q_\omega(\alpha,\beta)=(-1)^{2(2-1)/2}\int_S\alpha\smile\beta=-\int_S\alpha\smile\beta. \end{align*} If $\alpha\in P^2(S,\mathbb R)\cap H^{1,1}(S)$ and $\alpha\ne 0$, then the *Hodge-Riemann bilinear relations* give \begin{align*} Q_\omega(\alpha,\alpha)>0. \end{align*} Substituting the formula for $Q_\omega$ gives \begin{align*} -\int_S\alpha\smile\alpha>0. \end{align*} Therefore \begin{align*} \int_S\alpha\smile\alpha<0, \end{align*} so the intersection form is negative definite on the real primitive $(1,1)$-classes. On the Kähler line, positivity has the opposite sign. For $\lambda\in\mathbb R$, \begin{align*} \int_S(\lambda[\omega])\smile(\lambda[\omega])=\lambda^2\int_S[\omega]^2. \end{align*} Because $\omega$ is a Kähler form, $\omega^2$ is a positive volume form, so \begin{align*} \int_S[\omega]^2=\int_S\omega^2>0. \end{align*} Thus every nonzero class on the line $\mathbb R[\omega]$ has positive square. Finally, classes of type $(2,0)$ and $(0,2)$ are primitive because wedging with $[\omega]$ would have type $(3,1)$ or $(1,3)$, and those types vanish on a surface. If $\theta\in H^{2,0}(S)$, then $\theta\smile\theta$ has type $(4,0)$ and $\bar\theta\smile\bar\theta$ has type $(0,4)$, so both vanish. For the real class $\gamma=\theta+\bar\theta$, \begin{align*} \int_S\gamma\smile\gamma=2\int_S\theta\smile\bar\theta. \end{align*} The *Hodge-Riemann bilinear relations* for $\theta$ have $i^{2-0}=-1$, so \begin{align*} -Q_\omega(\theta,\bar\theta)>0. \end{align*} Since $Q_\omega(\theta,\bar\theta)=-\int_S\theta\smile\bar\theta$, this becomes \begin{align*} \int_S\theta\smile\bar\theta>0. \end{align*} Thus the primitive real part coming from $H^{2,0}(S)\oplus H^{0,2}(S)$ has positive square, while the primitive real $(1,1)$ part has negative square. This sign pattern is the cohomological form of the Hodge index theorem for Kähler surfaces. [/example] The chapter therefore ends with a compact package: $H^k(X)$ is not only a vector space with Betti number $b_k$, but a rational vector space with a Hodge decomposition, conjugation symmetry, a filtration, and in the primitive range a polarization coming from the Kähler class. This package is the entry point to abstract Hodge theory and to the study of how complex structures vary in families. # 9. Kähler Surfaces and Low-Dimensional Consequences Compact Kähler surfaces are the first place where the general Hodge-theoretic machinery becomes sharply topological. Building on the Hodge structures and polarizations just packaged in Chapter 8, the prerequisites are the Hodge decomposition for compact Kähler manifolds, Serre duality, Poincaré duality, the Hard Lefschetz theorem, and the Hodge-Riemann bilinear relations in degree two. In complex dimension two, the Hodge diamond, the cup-product pairing, and the Lefschetz decomposition together control much of the topology of the underlying four-manifold. This chapter records the main restrictions, computes the standard examples, and contrasts them with complex surfaces that are not Kähler. ## Hodge Diamond Constraints for Compact Kähler Surfaces What information does the Kähler condition force into the Hodge diamond of a compact complex surface? For a compact complex manifold, Dolbeault cohomology alone is complex-geometric; for a compact Kähler manifold, the Hodge decomposition identifies it with de Rham cohomology after complexification. In dimension two this turns the Hodge diamond into a compact summary of both topology and complex structure. [definition: Hodge Numbers of a Compact Complex Surface] Let $X$ be a compact complex surface. The Hodge numbers of $X$ are \begin{align*} h^{p,q}(X) := \dim_{\mathbb C} H^q(X, \Omega_X^p), \qquad 0 \le p,q \le 2. \end{align*} [/definition] After naming these numbers, the next question is which of the nine entries are independent. The answer uses exactly the Kähler input developed earlier: harmonic representatives respect type, conjugation, and duality. This is the structural reason the Hodge diamond becomes a topological invariant along each diagonal. [quotetheorem:8082] [citeproof:8082] The Kähler hypothesis is doing real work here: without it, Dolbeault cohomology need not assemble into de Rham cohomology by diagonals, and conjugation need not force the same topological parity conclusions. A Hopf surface, discussed below, has $b_1=1$, so it cannot satisfy the Kähler diagonal formula $b_1=h^{1,0}+h^{0,1}=2h^{1,0}$. The theorem also does not classify surfaces with a given diamond; it gives numerical constraints, not a construction of a Kähler metric or a biholomorphic classification. For a surface the diamond can therefore be described by \begin{align*} h^{0,0}&=h^{2,2}=1, & h^{1,0}&=h^{0,1}=h^{2,1}=h^{1,2}, & h^{2,0}&=h^{0,2}. \end{align*} The middle entry $h^{1,1}$ is not determined by $h^{1,0}$ and $h^{2,0}$, so examples are essential for seeing the range of possible surfaces. [example: Projective Plane Hodge Diamond] For $X=\mathbb{CP}^2$, the standard Dolbeault cohomology computation for projective space gives \begin{align*}h^{p,q}(\mathbb{CP}^2)=1 \text{ if } p=q \in \{0,1,2\}, \text{ and } h^{p,q}(\mathbb{CP}^2)=0 \text{ otherwise}.\end{align*} Thus the only nonzero Hodge numbers are $h^{0,0}=h^{1,1}=h^{2,2}=1$. Using the diagonal Betti number formula from *[Hodge Diamond Symmetries for Compact Kähler Surfaces](/theorems/8082)*, we get \begin{align*}b_1(\mathbb{CP}^2)=h^{1,0}(\mathbb{CP}^2)+h^{0,1}(\mathbb{CP}^2)=0+0=0.\end{align*} Similarly, \begin{align*}b_2(\mathbb{CP}^2)=h^{2,0}(\mathbb{CP}^2)+h^{1,1}(\mathbb{CP}^2)+h^{0,2}(\mathbb{CP}^2)=0+1+0=1.\end{align*} Let $\eta\in H^{1,1}(\mathbb{CP}^2)$ be the Fubini--Study Kähler class, normalized as the hyperplane class. Since $H^{1,1}(\mathbb{CP}^2)$ is one-dimensional, $\eta$ generates it. Its cup square is positive because \begin{align*}\int_{\mathbb{CP}^2}\eta\wedge\eta=\int_{\mathbb{CP}^2}\eta^2=1>0,\end{align*} where $\eta^2$ is the cohomology class Poincaré dual to the intersection of two hyperplanes, a single point. Thus $\mathbb{CP}^2$ has the smallest possible middle cohomology for a compact Kähler surface: one positive $(1,1)$-direction generated by the Kähler class. [/example] Projective space gives the smallest possible middle cohomology for a Kähler surface. Products of curves show how holomorphic one-forms and holomorphic two-forms enter the same diamond through the Künneth formula. [example: Product of Curves] Let $X=C_1\times C_2$, where $C_i$ is a compact Riemann surface of genus $g_i$. For a genus $g_i$ curve, the Hodge numbers are $h^{0,0}(C_i)=h^{1,1}(C_i)=1$ and $h^{1,0}(C_i)=h^{0,1}(C_i)=g_i$. By the *Kunneth formula* for Dolbeault cohomology, each $H^{p,q}(X)$ is the direct sum of tensor products whose bidegrees add to $(p,q)$. For $(1,0)$, the only contributing bidegree splittings are $(1,0)+(0,0)$ and $(0,0)+(1,0)$, so \begin{align*}h^{1,0}(X)=h^{1,0}(C_1)h^{0,0}(C_2)+h^{0,0}(C_1)h^{1,0}(C_2)=g_1\cdot 1+1\cdot g_2=g_1+g_2.\end{align*} For $(2,0)$, each curve contributes one holomorphic degree, hence \begin{align*}h^{2,0}(X)=h^{1,0}(C_1)h^{1,0}(C_2)=g_1g_2.\end{align*} For $(1,1)$, the four possible splittings give \begin{align*}h^{1,1}(X)=h^{1,1}(C_1)h^{0,0}(C_2)+h^{0,0}(C_1)h^{1,1}(C_2)+h^{1,0}(C_1)h^{0,1}(C_2)+h^{0,1}(C_1)h^{1,0}(C_2).\end{align*} Substituting the curve Hodge numbers gives \begin{align*}h^{1,1}(X)=1\cdot 1+1\cdot 1+g_1g_2+g_1g_2=2+2g_1g_2.\end{align*} The same computation gives $h^{0,1}(X)=g_1+g_2$ and $h^{0,2}(X)=g_1g_2$. Therefore the diagonal Betti number formula for compact Kähler surfaces gives \begin{align*}b_1(X)=h^{1,0}(X)+h^{0,1}(X)=(g_1+g_2)+(g_1+g_2)=2(g_1+g_2).\end{align*} Similarly, \begin{align*}b_2(X)=h^{2,0}(X)+h^{1,1}(X)+h^{0,2}(X)=g_1g_2+(2+2g_1g_2)+g_1g_2=2+4g_1g_2.\end{align*} If $\omega_i$ is a Kähler form on $C_i$ and $p_i:X\to C_i$ is projection, then \begin{align*}\omega=p_1^*\omega_1+p_2^*\omega_2\end{align*} is the product Kähler form. Since each $C_i$ has complex dimension one, $p_1^*\omega_1\wedge p_1^*\omega_1=0$ and $p_2^*\omega_2\wedge p_2^*\omega_2=0$, while two-forms commute under wedge product. Thus \begin{align*}\omega^2=2p_1^*\omega_1\wedge p_2^*\omega_2.\end{align*} Integrating over the product gives \begin{align*}\int_X\omega^2=2\left(\int_{C_1}\omega_1\right)\left(\int_{C_2}\omega_2\right)>0.\end{align*} So the product Kähler class is a positive middle-cohomology direction built from the two area classes on the curve factors. [/example] Products of curves already display nonzero $h^{2,0}$, but their cohomology still comes from two one-dimensional factors. A complex two-torus gives a surface whose forms are generated by translation-invariant one-forms in two complex directions. [example: Complex Two-Torus] Let $X=\mathbb C^2/\Lambda$ be a complex two-torus with complex coordinates $z_1,z_2$ on $\mathbb C^2$. By the standard invariant-form computation for complex tori, Dolbeault cohomology is represented by translation-invariant forms. A basis for $H^{p,q}(X)$ is obtained by choosing $p$ elements from $dz_1,dz_2$ and $q$ elements from $d\bar z_1,d\bar z_2$, then wedging them in increasing order. Hence \begin{align*}h^{p,q}(X)=\#\{p\text{-element subsets of }\{1,2\}\}\cdot \#\{q\text{-element subsets of }\{1,2\}\}=\binom{2}{p}\binom{2}{q}.\end{align*} For the entries used most often in degree two, this gives \begin{align*}h^{1,0}(X)=\binom{2}{1}\binom{2}{0}=2\cdot 1=2.\end{align*} Also \begin{align*}h^{2,0}(X)=\binom{2}{2}\binom{2}{0}=1\cdot 1=1.\end{align*} And \begin{align*}h^{1,1}(X)=\binom{2}{1}\binom{2}{1}=2\cdot 2=4.\end{align*} Using the Kähler diagonal Betti number formula, the first Betti number is \begin{align*}b_1(X)=h^{1,0}(X)+h^{0,1}(X)=2+2=4.\end{align*} The second Betti number is \begin{align*}b_2(X)=h^{2,0}(X)+h^{1,1}(X)+h^{0,2}(X)=1+4+1=6.\end{align*} Thus a complex two-torus has four independent real degree-one cohomology classes and six independent real degree-two cohomology classes, with the middle Hodge piece split as $1,4,1$ across types $(2,0),(1,1),(0,2)$. [/example] ## Intersection Form and Primitive Decomposition How does a Kähler class control the cup-product form on the middle cohomology of a surface? In real dimension four, the cup product on $H^2(X;\mathbb R)$ is a symmetric bilinear form, and the Kähler class singles out a positive direction. Lefschetz theory then decomposes the remaining directions into primitive classes, on which the sign changes. [definition: Intersection Form of a Compact Oriented Four-Manifold] Let $M$ be a compact oriented smooth four-manifold. The intersection form of $M$ is the symmetric bilinear form $Q_M : H^2(M;\mathbb R)\times H^2(M;\mathbb R) \to \mathbb R$ given by \begin{align*} Q_M(\alpha,\beta) = \int_M \alpha\wedge \beta. \end{align*} [/definition] For a compact Kähler surface $X$, we apply this to the orientation determined by the complex structure. The form is nondegenerate by Poincaré duality, and its signature is one of the strongest topological invariants visible in Hodge theory. [definition: Primitive Cohomology in Degree Two] Let $X$ be a compact Kähler surface with Kähler class $[\omega]\in H^2(X;\mathbb R)$. The primitive degree-two cohomology is \begin{align*} P^2(X;\mathbb R) := \{\alpha\in H^2(X;\mathbb R): \alpha\smile [\omega]=0\in H^4(X;\mathbb R)\}. \end{align*} [/definition] The point of this definition is that every degree-two class splits into its component along the Kähler class and its component orthogonal to it. The Hodge-Riemann bilinear relations determine the signs on these two pieces, so we first isolate the relevant direct-sum decomposition. [quotetheorem:3855] [citeproof:3855] The primitive decomposition separates the unique Kähler direction from the rest of $H^2$. The Kähler class hypothesis is essential: a general real two-dimensional subspace of $H^2(X;\mathbb R)$ does not come with a distinguished positive class whose cup product with every other class defines the primitive condition. On a non-Kähler surface such as a Hopf surface, $H^2$ is too small to support this Lefschetz picture, and on more complicated non-Kähler surfaces the Frölicher and Lefschetz mechanisms can fail to identify the right cohomological pieces. The decomposition also does not decide the sign of every class by itself; it only isolates the Kähler line and its primitive complement. The next theorem asks for the sign of the intersection form on the real $(1,1)$-part, where divisor classes and Kähler classes live. [quotetheorem:8069] [citeproof:8069] The Hodge index theorem accounts for the real $(1,1)$-classes, and its compact Kähler hypotheses cannot be dropped casually. Compactness supplies Poincaré duality and finite-dimensional cohomology, while the Kähler condition supplies the Hodge-Riemann sign rule that makes primitive $(1,1)$-classes negative. Non-Kähler surfaces may still have an intersection form, but there need not be a Hodge-theoretic reason for its restriction to $(1,1)$-classes to have this signature. A Hopf surface illustrates a sharper obstruction: it has $b_2=0$, so there is no nonzero Kähler class and no positive Kähler line in degree two, while its odd $b_1$ already rules out the Kähler Hodge package. The theorem also says nothing about which integral classes are represented by curves or divisors. Middle cohomology also contains the real and imaginary parts of holomorphic two-forms, so the full signature calculation asks how these additional directions contribute to the positive and negative indices. [quotetheorem:8070] [citeproof:8070] The formula turns Hodge numbers into signatures, so the basic examples from the previous section can now be read as four-manifold calculations. Its hypotheses are exactly the ones that make this translation valid: the Hodge decomposition identifies $H^2(X;\mathbb C)$ by type, and the Hodge-Riemann relations determine which real summands are positive or negative. Outside the Kähler setting, the same Betti numbers need not come from a Hodge diamond with these symmetries; for instance, a Hopf surface has no room for the predicted Kähler parity already in degree one. The formula is also not a converse: odd $b_2^+$ and even $b_1$ do not by themselves construct a Kähler form. In practice, given a Kähler surface Hodge diamond, compute \begin{align*} b_1&=2h^{1,0}, & b_2&=h^{1,1}+2h^{2,0}, & b_2^+&=1+2h^{2,0}, & b_2^-&=h^{1,1}-1, \end{align*} and then $\tau=b_2^+-b_2^-$. This also previews why blow-ups alter the signature in a controlled way. [example: Signatures of Standard Kähler Surfaces] Using the formulas $b_2^+=1+2h^{2,0}$ and $b_2^-=h^{1,1}-1$ from *[Signature Formula for Compact Kähler Surfaces](/theorems/8070)*, the projective plane values $h^{2,0}(\mathbb{CP}^2)=0$ and $h^{1,1}(\mathbb{CP}^2)=1$ give \begin{align*}b_2^+(\mathbb{CP}^2)=1+2\cdot 0=1.\end{align*} Also, \begin{align*}b_2^-(\mathbb{CP}^2)=1-1=0.\end{align*} Therefore \begin{align*}\tau(\mathbb{CP}^2)=b_2^+(\mathbb{CP}^2)-b_2^-(\mathbb{CP}^2)=1-0=1.\end{align*} For a complex two-torus, $h^{2,0}=1$ and $h^{1,1}=4$, so the same formulas give \begin{align*}b_2^+=1+2\cdot 1=3.\end{align*} And \begin{align*}b_2^-=4-1=3.\end{align*} Thus \begin{align*}\tau=b_2^+-b_2^-=3-3=0.\end{align*} For $X=C_1\times C_2$, the earlier Hodge-number computation gives $h^{2,0}(X)=g_1g_2$ and $h^{1,1}(X)=2+2g_1g_2$. Hence \begin{align*}b_2^+(X)=1+2h^{2,0}(X)=1+2g_1g_2.\end{align*} Also, \begin{align*}b_2^-(X)=h^{1,1}(X)-1=(2+2g_1g_2)-1=1+2g_1g_2.\end{align*} Therefore \begin{align*}\tau(X)=b_2^+(X)-b_2^-(X)=(1+2g_1g_2)-(1+2g_1g_2)=0.\end{align*} These three computations show how the same signature formula separates the positive Kähler and holomorphic-two-form directions from the negative primitive $(1,1)$ directions in standard examples. [/example] The preceding examples are minimal in the sense that no exceptional curve has been inserted. Blow-ups give a controlled way to create new Kähler surfaces from old ones, and their effect on the intersection form is especially transparent. [example: Blow-Up of a Kähler Surface at a Point] Let $\pi:\widetilde X\to X$ be the blow-up of a compact Kähler surface at a point, and let $E\cong\mathbb{CP}^1$ be the exceptional divisor. The standard blow-up construction preserves the Kähler property: if $[\omega]$ is a Kähler class on $X$, then for all sufficiently small $\varepsilon>0$ the class $\pi^*[\omega]-\varepsilon [E]$ is represented by a Kähler form on $\widetilde X$. The degree-two cohomology of the blow-up splits as \begin{align*}H^2(\widetilde X;\mathbb R)\cong \pi^*H^2(X;\mathbb R)\oplus \mathbb R[E].\end{align*} The new summand is generated by the exceptional divisor class. Since the normal bundle of $E$ in $\widetilde X$ is $\mathcal O_{\mathbb{CP}^1}(-1)$, its self-intersection is \begin{align*}E^2=\int_{\widetilde X}[E]\wedge [E]=\deg \mathcal O_{\mathbb{CP}^1}(-1)=-1.\end{align*} The pullback classes are orthogonal to $[E]$: for $\alpha\in H^2(X;\mathbb R)$, if $i:E\hookrightarrow \widetilde X$ is the inclusion, then \begin{align*}\int_{\widetilde X}\pi^*\alpha\wedge [E]=\int_E i^*\pi^*\alpha.\end{align*} Because $\pi\circ i:E\to X$ is the constant map to the blown-up point, $(\pi\circ i)^*\alpha=0$ in $H^2(E;\mathbb R)$, so \begin{align*}\int_E i^*\pi^*\alpha=\int_E(\pi\circ i)^*\alpha=0.\end{align*} Thus, with respect to the decomposition above, the intersection form on $\widetilde X$ is the orthogonal direct sum of the intersection form on $X$ and the one-dimensional form $[-1]$. Taking dimensions gives \begin{align*}b_2(\widetilde X)=\dim H^2(X;\mathbb R)+\dim \mathbb R[E]=b_2(X)+1.\end{align*} Since the added summand has square $-1$, it contributes one new negative direction and no new positive direction: \begin{align*}b_2^+(\widetilde X)=b_2^+(X).\end{align*} Also, \begin{align*}b_2^-(\widetilde X)=b_2^-(X)+1.\end{align*} Therefore the signature changes by \begin{align*}\tau(\widetilde X)=b_2^+(\widetilde X)-b_2^-(\widetilde X)=b_2^+(X)-(b_2^-(X)+1)=\tau(X)-1.\end{align*} Blowing up a point adds exactly one exceptional curve of self-intersection $-1$, so it increases the middle Betti number by one and decreases the signature by one. [/example] ## Topological Restrictions and Non-Kähler Contrasts Which topological four-manifolds can underlie compact Kähler surfaces? The previous sections give restrictions on $H^2$, but Kähler geometry also constrains odd-degree cohomology in every complex dimension. In dimension two this immediately forces $b_1$ to be even, ruling out many compact complex surfaces. [quotetheorem:8072] [citeproof:8072] For surfaces, this gives the first obstruction encountered in the course: any compact Kähler surface has even $b_1$. The Kähler hypothesis is needed because the proof pairs Hodge summands by conjugation along an odd diagonal; a compact complex manifold without Kähler Hodge decomposition need not have such paired de Rham contributions. A primary Hopf surface has $b_1=1$, giving a concrete compact complex surface where the conclusion fails. The result is only an obstruction, not a recognition theorem: many manifolds with even odd Betti numbers are still not known to be Kähler without extra geometric structure. Together with the signature formula, it supplies quick tests that separate Kähler and non-Kähler examples. [quotetheorem:8073] [citeproof:8073] These restrictions are not shared by all compact complex surfaces. They depend on both ingredients used above: Hodge decomposition gives the formulas for $b_1$ and $b_2$, while the Hodge-Riemann relations give the formula for $b_2^+$. The restrictions are necessary but not sufficient; passing the parity and signature tests does not produce a closed positive $(1,1)$-form, nor does it classify the complex surface. They are most useful as quick obstructions, and Hopf surfaces are the fastest contrast because their first Betti number already violates Kähler parity. [example: Hopf Surface as a Non-Kähler Contrast] Let $X=(\mathbb C^2\setminus\{0\})/\langle T\rangle$, where $T(z_1,z_2)=(\lambda z_1,\lambda z_2)$ and $0<|\lambda|<1$. Write each nonzero $z\in\mathbb C^2$ uniquely as $z=ru$ with $r\in(0,\infty)$ and $u\in S^3$. Under this identification, \begin{align*}T(r,u)=(|\lambda|r,(\lambda/|\lambda|)u).\end{align*} After setting $t=\log r$, the action becomes \begin{align*}(t,u)\mapsto(t+\log|\lambda|,(\lambda/|\lambda|)u).\end{align*} The map $u\mapsto(\lambda/|\lambda|)u$ on $S^3$ is isotopic to the identity through $u\mapsto e^{is\arg(\lambda)}u$, so the quotient is diffeomorphic to the mapping torus of a map isotopic to the identity, hence to $S^1\times S^3$. Using the *Kunneth formula* for real cohomology and the cohomology of spheres, \begin{align*}H^1(S^1\times S^3;\mathbb R)\cong H^1(S^1;\mathbb R)\otimes H^0(S^3;\mathbb R).\end{align*} Since $\dim H^1(S^1;\mathbb R)=1$ and $\dim H^0(S^3;\mathbb R)=1$, this gives \begin{align*}b_1(X)=1\cdot 1=1.\end{align*} For degree two, the Kunneth summands are \begin{align*}H^2(S^1;\mathbb R)\otimes H^0(S^3;\mathbb R)\oplus H^1(S^1;\mathbb R)\otimes H^1(S^3;\mathbb R)\oplus H^0(S^1;\mathbb R)\otimes H^2(S^3;\mathbb R).\end{align*} The dimensions are $0\cdot 1$, $1\cdot 0$, and $1\cdot 0$, respectively, so \begin{align*}b_2(X)=0+0+0=0.\end{align*} If $X$ were Kähler, then *[Odd Betti Numbers of Compact Kähler Manifolds Are Even](/theorems/8072)* would force $b_1(X)$ to be even; instead $b_1(X)=1$. Therefore this Hopf surface admits no Kähler metric. [/example] The Hopf surface shows that compact complex surfaces need not have Kähler-type cohomology. Kodaira surfaces give a second contrast where the topology is richer but the same parity obstruction remains decisive. [example: Kodaira Surface as a Non-Kähler Contrast] A primary Kodaira surface $X$ is a compact complex surface whose standard real-cohomology computation gives \begin{align*}b_1(X)=3.\end{align*} If $X$ admitted a Kähler metric, then *Odd Betti Numbers of Compact Kähler Manifolds Are Even* would imply that $b_1(X)$ is even. But \begin{align*}3=2\cdot 1+1,\end{align*} so $3$ is not divisible by $2$. Hence $b_1(X)=3$ contradicts the Kähler parity condition, and $X$ admits no Kähler metric. This obstruction does not come from having too little middle cohomology: primary Kodaira surfaces have richer topology than Hopf surfaces, and they carry holomorphic one-forms and an elliptic fibration structure. The decisive failure is that their topology cannot satisfy the Hodge symmetries forced by the Kähler condition. [/example] The examples also clarify the logical direction of the restrictions. Kähler geometry forces parity, Hodge decomposition, and signature constraints; a complex surface that fails any of these tests is non-Kähler, while a surface that passes them still requires a separate construction of a Kähler form. [remark: What the Surface Case Teaches] Compact Kähler surfaces translate analytic structure into numerical topology with unusual force. The Hodge diamond constrains Betti numbers, the Kähler class controls the intersection form, and the Hodge index theorem gives a sharp signature calculation. These facts are special cases of the higher-dimensional Hodge-Riemann bilinear relations, but in complex dimension two they become concrete tests for whether a compact complex surface can be Kähler. [/remark] # 10. Periods, Variation-Free Hodge Theory, and Applications This final chapter uses the Hodge theory developed earlier for compact Kähler manifolds, especially the Hodge structures of Chapter 8 and the surface-level restrictions of Chapter 9, and turns it into concrete cohomological tests. We assume the Hodge decomposition, Poincaré duality, the de Rham period pairing, and the hard Lefschetz theorem from the preceding chapters. The guiding principle is that integration remembers the Hodge decomposition in a form that can be tested against geometric submanifolds and functorial maps. This gives a bridge from analysis to topology: harmonic forms determine periods, complex submanifolds determine cohomology classes of Hodge type, and holomorphic maps respect all the resulting decompositions. ## Period Pairings and Hodge Type Through Integration The first question is how the abstract decomposition \begin{align*} H^k(X,\mathbb C)=\bigoplus_{p+q=k}H^{p,q}(X) \end{align*} can be detected by integration. On a compact oriented smooth manifold, de Rham cohomology is paired with homology by integrating closed forms over cycles. On a compact Kähler manifold, this same pairing distinguishes Hodge type once we allow complex-valued forms and use Poincaré duality. [definition: Period Pairing] Let $X$ be a compact oriented smooth manifold of real dimension $m$. For $0\le k\le m$, the period pairing is the bilinear map \begin{align*} H^k(X,\mathbb C)\times H_k(X,\mathbb Z) \longrightarrow \mathbb C \end{align*} which sends $([\alpha],[\Gamma])$ to $\int_\Gamma \alpha$. [/definition] The definition depends only on cohomology and homology classes by Stokes' theorem. When $X$ is Kähler, we may choose the harmonic representative of $[\alpha]$ and decompose it into its $(p,q)$-components, so periods become a way of probing the Hodge summands by integration over cycles. [example: Periods on a Complex Torus] Let $X=\mathbb C^n/\Lambda$ be a compact complex torus, and write a point of $\mathbb C^n$ as $z=(z_1,\ldots,z_n)$. The translation-invariant forms $dz_i$ and $d\bar z_i$ descend to closed forms on $X$, because translations by elements of $\Lambda$ leave the differentials unchanged. Their types are $(1,0)$ and $(0,1)$, respectively, so they define classes in $H^{1,0}(X)$ and $H^{0,1}(X)$. For $\lambda=(\lambda_1,\ldots,\lambda_n)\in\Lambda$, let $\gamma_\lambda:[0,1]\to X$ be the loop induced by the path $t\mapsto t\lambda$ in $\mathbb C^n$. Along this path, \begin{align*} z_i(t)=t\lambda_i \end{align*} so \begin{align*} \gamma_\lambda^*(dz_i)=d(t\lambda_i)=\lambda_i\,dt \end{align*} and therefore \begin{align*} \int_{\gamma_\lambda} dz_i=\int_0^1 \lambda_i\,dt=\lambda_i. \end{align*} Similarly, $\bar z_i(t)=t\bar\lambda_i$, hence \begin{align*} \gamma_\lambda^*(d\bar z_i)=\bar\lambda_i\,dt \end{align*} and \begin{align*} \int_{\gamma_\lambda} d\bar z_i=\int_0^1 \bar\lambda_i\,dt=\bar\lambda_i. \end{align*} Thus the degree-one periods recover the complex coordinates of lattice vectors and their conjugates. In higher degree, a wedge such as $dz_{i_1}\wedge\cdots\wedge dz_{i_p}\wedge d\bar z_{j_1}\wedge\cdots\wedge d\bar z_{j_q}$ has type $(p,q)$, and its periods are obtained by integrating the corresponding exterior product over real subtori generated by lattice vectors; the lattice therefore records the integral periods of the whole Hodge decomposition. [/example] This example is the model for the general statement: Hodge pieces are not additional cohomology groups unrelated to topology, but distinguished complex subspaces of ordinary cohomology. If one remembers only the complex vector space decomposition, the arithmetic content of periods is lost: two decompositions can have the same complex dimensions while meeting the integral lattice in very different ways. To make the statement intrinsic, we must remember both the complex decomposition and the integral lattice on which periods take their values. [definition: Integral Hodge Structure] A pure integral Hodge structure of weight $k$ consists of a finitely generated free abelian group $H_{\mathbb Z}$ and a decomposition \begin{align*} H_{\mathbb C}:=H_{\mathbb Z}\otimes_{\mathbb Z}\mathbb C=\bigoplus_{p+q=k}H^{p,q} \end{align*} such that $\overline{H^{p,q}}=H^{q,p}$. [/definition] The freeness condition removes torsion before tensoring with $\mathbb C$, where torsion would disappear anyway. The definition abstracts the data seen in periods on a torus. The theorem needed next is that every compact Kähler manifold supplies this structure in every degree, so the analytic Hodge decomposition is compatible with the integral cohomology lattice rather than being only a decomposition of a complex vector space. [quotetheorem:3854] [citeproof:3854] This theorem converts analytic information into a topological invariant with extra structure. Compactness is essential because the Hodge theorem used here identifies cohomology with finite-dimensional harmonic forms; on a non-compact Kähler manifold, harmonic representatives need not give ordinary cohomology in this form without extra hypotheses. Kählerness is also essential: compact complex manifolds such as primary Hopf surfaces do not satisfy the degree-one Hodge symmetry forced by this theorem, since their first Betti number is odd. The theorem does not choose a preferred integral basis or compute the periods; it says that any such period computation is constrained by a fixed Hodge decomposition. Since periods pair integral homology with the integral lattice in cohomology, the next step is to ask which geometric cycles produce classes in particular Hodge summands. [remark: Periods Without Variation] In a theory of variations of Hodge structure, the same period data is studied as the complex structure moves in a family. In this chapter we keep $X$ fixed. The term period only means the evaluation of a cohomology class on a homology class, together with the information of which Hodge summand contains the cohomology class. [/remark] The period language becomes most geometric when the homology cycle is represented by a complex submanifold. The next section explains why such cycles occupy the middle Hodge type. ## Cycle Classes of Complex Submanifolds The next question is what kind of cohomology class is produced by a complex submanifold. A closed oriented real submanifold has a Poincaré dual class, but a general real submanifold has no reason to respect the complex decomposition: its dual class can have several Hodge components. The extra point of a complex submanifold is that its tangent spaces are complex linear, so integration over it only sees forms of the correct middle type. This is the first place where the complex geometry of a subset imposes a Hodge-theoretic condition on ordinary cohomology. [definition: Cycle Class of a Complex Submanifold] Let $X$ be a compact complex manifold of complex dimension $n$, and let $Y\subset X$ be a closed complex submanifold of complex codimension $r$. The cycle class of $Y$ is the Poincaré dual class \begin{align*} [Y]\in H^{2r}(X,\mathbb Z) \end{align*} characterised by \begin{align*} \langle [\alpha]\smile [Y],[X]\rangle=\int_Y \alpha|_Y \end{align*} for every closed smooth form $\alpha\in A^{2n-2r}(X)$. [/definition] The defining property says that the cohomology class $[Y]$ is the unique class whose periods against complementary forms reproduce integration over $Y$. The complex structure now enters through the fact that a complex submanifold has real codimension $2r$ and complex normal directions of type $(1,0)$. [quotetheorem:8074] [citeproof:8074] The result should be read as a Hodge-theoretic restriction on which integral cohomology classes can come from complex geometry. The complex-submanifold hypothesis is needed. On $X=(\mathbb C/(\mathbb Z+i\mathbb Z))^2$ with coordinates $z_j=x_j+iy_j$, the real torus $Y=\{y_1=y_2=0\}$ is a smooth real codimension-$2$ submanifold but not a complex curve. Since $\int_Y dz_1\wedge dz_2=\int_Y dx_1\wedge dx_2\ne 0$, its Poincaré dual pairs nontrivially with a $(2,0)$-form, so the dual class has a component outside $H^{1,1}(X)$. The Kähler hypothesis supplies the Hodge decomposition and Poincaré duality compatibility used in the argument. The theorem does not say that every integral class of type $(r,r)$ is represented by a submanifold; it only gives a necessary condition. This necessary condition motivates naming the cohomology classes that pass the Hodge-type test even before any geometric representative is known. [example: The Class of a Divisor] Let $X$ be a compact Kähler manifold of complex dimension $n$, and let $D\subset X$ be a smooth divisor. A divisor has complex codimension $1$, hence real codimension $2$, so its Poincaré dual cycle class has degree $2$: \begin{align*} [D]\in H^2(X,\mathbb Z). \end{align*} Applying *[Cycle Classes of Complex Submanifolds Have Hodge Type](/theorems/8074)* with $r=1$ gives that the image of this integral class in $H^2(X,\mathbb C)$ lies in $H^{1,1}(X)$. Therefore \begin{align*} [D]\in H^{1,1}(X)\cap H^2(X,\mathbb Z). \end{align*} Equivalently, integration over $D$ pairs nontrivially only with complementary forms of type $(n-1,n-1)$, so the Poincaré dual has no $(2,0)$ or $(0,2)$ component. This conclusion uses Poincaré duality and the type restriction coming from integration over the complex submanifold $D$; it does not use positivity of line bundles, Kodaira vanishing, or a defining section. [/example] Divisors show how a geometric submanifold produces a cohomology class in the required Hodge summand, and they also show why a purely topological degree-two class is not enough. A class in $H^2(X,\mathbb Z)$ can fail to have type $(1,1)$ after complexification, so it cannot be the class of a complex divisor. This motivates the next definition: we need a name for classes satisfying the same Hodge-type condition even when no submanifold has been specified. [definition: Hodge Class] Let $X$ be a compact Kähler manifold. An integral Hodge class of codimension $r$ is a class \begin{align*} \gamma\in H^{2r}(X,\mathbb Z)/\text{torsion} \end{align*} whose image in $H^{2r}(X,\mathbb C)$ lies in $H^{r,r}(X)$. [/definition] Cycle classes of complex submanifolds are Hodge classes by the preceding theorem. The definition is intentionally cohomological: in this course we use algebraic cycles only through their classes, rather than entering intersection theory or positivity. [remark: Scope of the Hodge Class Condition] The statement that complex submanifolds give Hodge classes is a theorem in the Kähler package. The converse direction, which asks when rational Hodge classes are generated by algebraic cycles, is the territory of the Hodge conjecture for smooth projective varieties and is not part of the present course. [/remark] The next application asks how these classes behave under maps. Since pullback of forms preserves type for holomorphic maps, Hodge classes and Kähler classes are functorial in the expected direction. ## Holomorphic Maps and Functoriality of Hodge Structures The central question for maps is whether the Hodge decomposition is a property of the manifold alone or a structure respected by holomorphic geometry. A smooth map pulls back de Rham classes, but it can mix the complex directions and therefore mix Hodge types. A holomorphic map has the stronger property that it pulls back $(p,q)$-forms to $(p,q)$-forms. This compatibility makes Hodge decomposition functorial for compact Kähler manifolds. [quotetheorem:8075] [citeproof:8075] This theorem is often the quickest way to rule out proposed holomorphic maps: the underlying map on cohomology must respect the Hodge decomposition, not just the grading. Holomorphicity is the essential hypothesis. On an elliptic curve $E=\mathbb C/\Lambda$, the smooth involution induced by $z\mapsto \bar z$ pulls the class of $dz$ to the class of $d\bar z$, so it exchanges $H^{1,0}(E)$ and $H^{0,1}(E)$ rather than preserving type. Compact Kählerness is used to identify the Hodge summands with cohomological subspaces rather than only with spaces of forms. The theorem does not say that every graded ring map preserving Hodge type is induced by a holomorphic map; it gives a necessary functorial constraint. The same mechanism will next be applied to Kähler classes, where type is preserved but strict positivity may be lost. [example: Pulling Back a Hodge Decomposition] Let $f:X\to Y$ be a holomorphic map of compact Kähler manifolds, and write the Hodge decomposition of a degree-two class as \begin{align*} \beta=\beta^{2,0}+\beta^{1,1}+\beta^{0,2}, \end{align*} where $\beta^{p,q}\in H^{p,q}(Y)$ for $p+q=2$. Since pullback is linear on cohomology, \begin{align*} f^*\beta=f^*(\beta^{2,0}+\beta^{1,1}+\beta^{0,2}). \end{align*} Thus \begin{align*} f^*\beta=f^*\beta^{2,0}+f^*\beta^{1,1}+f^*\beta^{0,2}. \end{align*} By *Holomorphic Pullback Preserves Hodge Decomposition*, holomorphicity gives \begin{align*} f^*\beta^{2,0}\in H^{2,0}(X). \end{align*} The same type-preservation gives \begin{align*} f^*\beta^{1,1}\in H^{1,1}(X) \end{align*} and \begin{align*} f^*\beta^{0,2}\in H^{0,2}(X). \end{align*} Consequently, the pullback does not mix the three Hodge summands in degree $2$. If a proposed smooth map $g:X\to Y$ has the property that $g^*\eta$ has a nonzero $H^{2,0}(X)$ or $H^{0,2}(X)$ component for some $\eta\in H^{1,1}(Y)$, then $g$ cannot be homotopic to a holomorphic map inducing the same cohomological pullback, because homotopic maps induce the same map on cohomology while holomorphic pullback would force $g^*\eta$ to remain in $H^{1,1}(X)$. [/example] The same argument applies to the special degree-two classes represented by Kähler forms. Pulling back a Kähler form need not be Kähler if the map has positive-dimensional fibres, but it remains closed, real, and of type $(1,1)$. [quotetheorem:8076] [citeproof:8076] This distinction between positivity and semipositivity is useful in applications. Holomorphicity and closedness are enough to preserve real $(1,1)$ type, but injectivity of the differential is needed for strict positivity of the pulled-back form. A constant map is the extreme counterexample: the pullback of every Kähler form is zero, hence semipositive but not Kähler. The theorem therefore does not claim that Kähler classes pull back to Kähler classes under arbitrary holomorphic maps; it records the weaker conclusion that survives under maps with fibres. The product projection example isolates exactly this boundary. [example: Projection from a Product] Let $X=Y\times Z$ be a product of compact Kähler manifolds, and let $\pi_Y:X\to Y$ be the projection. If $\omega_Y$ is a Kähler form on $Y$, then the pullback form $\pi_Y^*\omega_Y$ is closed because pullback commutes with exterior derivative: \begin{align*} d(\pi_Y^*\omega_Y)=\pi_Y^*(d\omega_Y)=\pi_Y^*(0)=0. \end{align*} Since $\pi_Y$ is holomorphic, pullback preserves type, so $\pi_Y^*\omega_Y$ has type $(1,1)$ on $X$. Now take a point $(y,z)\in Y\times Z$ and a tangent vector tangent to the $Z$-factor, written as $(0,w)\in T_yY\oplus T_zZ$. The differential of the projection is \begin{align*} d\pi_Y(0,w)=0. \end{align*} Therefore \begin{align*} (\pi_Y^*\omega_Y)_{(y,z)}((0,w),J_X(0,w))=(\omega_Y)_y(d\pi_Y(0,w),d\pi_Y(J_X(0,w)))=(\omega_Y)_y(0,0)=0. \end{align*} Thus $\pi_Y^*\omega_Y$ is semipositive but not positive on all nonzero tangent vectors whenever $Z$ has positive dimension. Its cohomology class lies in $H^{1,1}(X)\cap H^2(X,\mathbb R)$, but the form is not a Kähler form on the product unless one adds a positive contribution in the $Z$-directions, such as the pullback of a Kähler form from $Z$. [/example] Functoriality, cycle classes, and the period pairing now combine into the promised topological restrictions. These restrictions are weaker than classification theorems but strong enough to rule out many smooth manifolds and ring maps from being Kähler or holomorphic. ## Constraints on Fundamental Groups, Cohomology Rings, and Maps The final question is what the Kähler package forbids. Since Hodge decomposition, conjugation symmetry, Poincaré duality, Lefschetz theory, and the Hodge-Riemann bilinear relations all live on cohomology, they impose algebraic conditions on the cohomology ring of a compact Kähler manifold. In degree $1$, they also constrain the abelianisation of the fundamental group. [quotetheorem:8077] [citeproof:8077] This is a first obstruction to a smooth compact manifold carrying any Kähler structure. Compactness and Kählerness are both doing work: compact Riemann surfaces and complex tori satisfy the symmetry, while compact non-Kähler manifolds such as Hopf surfaces can have odd first Betti number. The theorem does not classify Kähler groups, and evenness of $b_1$ alone is far from sufficient for a Kähler structure. Because $H^1(X,\mathbb Z)$ is the homomorphism group from $\pi_1(X)$ to $\mathbb Z$ up to torsion, the obstruction is nevertheless visible already in the abelianisation of the fundamental group. [example: Hopf Surface Obstruction Revisited] A primary Hopf surface is diffeomorphic to $S^1\times S^3$. By the *Kunneth formula* over $\mathbb R$, \begin{align*} H^1(S^1\times S^3,\mathbb R)\cong H^1(S^1,\mathbb R)\otimes H^0(S^3,\mathbb R)\oplus H^0(S^1,\mathbb R)\otimes H^1(S^3,\mathbb R). \end{align*} Since $H^1(S^1,\mathbb R)\cong \mathbb R$, $H^0(S^3,\mathbb R)\cong \mathbb R$, $H^0(S^1,\mathbb R)\cong \mathbb R$, and $H^1(S^3,\mathbb R)=0$, this becomes \begin{align*} H^1(S^1\times S^3,\mathbb R)\cong \mathbb R\otimes \mathbb R\oplus \mathbb R\otimes 0\cong \mathbb R. \end{align*} Therefore \begin{align*} b_1(S^1\times S^3)=\dim_{\mathbb R}H^1(S^1\times S^3,\mathbb R)=1. \end{align*} If a primary Hopf surface admitted a Kähler metric, then the evenness of the first Betti number for compact Kähler manifolds would force $b_1=2h^{1,0}$, hence $b_1$ would be even. The computed value $b_1=1$ is not even, so no primary Hopf surface is Kähler. This recovers the earlier obstruction using only the degree-one Hodge decomposition. [/example] Degree $1$ is only the first layer: the Hopf surface obstruction uses dimensions, but many non-Kähler candidates pass the simplest Betti-number tests. The stronger obstruction is multiplicative. If a compact manifold were Kähler, cup product with a Kähler class would impose rigid linear-algebra constraints across all cohomological degrees, so a proposed cohomology ring must contain a degree-two class whose powers act in exactly the required way. [quotetheorem:8078] [citeproof:8078] These conditions are often practical because they are stated entirely in terms of finite-dimensional linear algebra and cup products. The Kähler class is essential: a symplectic class on a compact manifold need not satisfy the full hard Lefschetz package. A standard example is the Kodaira-Thurston four-manifold, which is symplectic but has first Betti number $3$ and also fails the Lefschetz isomorphism $H^1\to H^3$ for its usual symplectic class. The theorem does not assert that any graded algebra satisfying these listed constraints is realised by a compact Kähler manifold; it gives necessary tests. If a proposed cohomology ring has no degree-two class satisfying hard Lefschetz, then it cannot be the real cohomology ring of a compact Kähler manifold. [example: A Ring-Level Lefschetz Test] Suppose a compact smooth $2n$-manifold $M$ has a candidate degree-two class $a\in H^2(M,\mathbb R)$ intended to be the class of a Kähler form. If such a Kähler structure existed with Kähler class $a$, then by *[Kähler Package Constraints on Cohomology Algebras](/theorems/8078)*, for every $k\le n$ the Lefschetz map would be \begin{align*} L^{n-k}:H^k(M,\mathbb R)\longrightarrow H^{2n-k}(M,\mathbb R),\qquad \alpha\longmapsto a^{n-k}\smile \alpha \end{align*} and this map would have to be an isomorphism. In real dimension $4$, we have $2n=4$, so $n=2$. For $k=1$, the required map is \begin{align*} L^{2-1}:H^1(M,\mathbb R)\longrightarrow H^3(M,\mathbb R),\qquad \alpha\longmapsto a\smile \alpha \end{align*} so multiplication by $a$ must be an isomorphism $H^1(M,\mathbb R)\to H^3(M,\mathbb R)$. For $k=0$, the required map is \begin{align*} L^2:H^0(M,\mathbb R)\longrightarrow H^4(M,\mathbb R),\qquad c\longmapsto c\,a^2 \end{align*} because $a^2=a\smile a$. Since $H^0(M,\mathbb R)\cong \mathbb R$ and $L^2$ must be an isomorphism, the image of $1$ cannot be zero. Therefore \begin{align*} L^2(1)=a^2\ne 0 \end{align*} in $H^4(M,\mathbb R)$. Thus a single candidate class $a$ fails the ring-level Kähler test if either multiplication by $a$ is not an isomorphism from $H^1$ to $H^3$, or if $a^2$ vanishes in top-degree cohomology. [/example] The Lefschetz test concerns which rings can occur for compact Kähler manifolds. It turns a positive theorem into a reusable obstruction: failure of a single required Lefschetz isomorphism rules out all Kähler structures, regardless of how promising the manifold may look locally. For holomorphic maps, we need the parallel statement about which graded ring homomorphisms can occur: they must respect both cup product and Hodge type. [quotetheorem:8079] [citeproof:8079] This theorem is a compact summary of the variation-free viewpoint. Holomorphicity is needed for preservation of Hodge type, while the ring statement alone would hold for any continuous map. The compact Kähler hypotheses are needed so that the source and target carry the Hodge decompositions used in the statement; without them there is no comparable decomposition for the pullback to respect. The theorem does not say that every Hodge-compatible graded ring homomorphism is realised by a holomorphic map, nor that every pulled-back Hodge class is represented by an actual complex submanifold. It says that any genuine holomorphic map must pass these cohomological tests. Thus we do not need to deform $X$, vary periods, or invoke positivity theorems in order to obtain strong restrictions: the fixed Hodge decomposition already governs the cohomology classes that geometry can produce and the cohomology maps that holomorphic geometry can realise. [remark: End of the Course] The course began with the Kähler condition as a differential-geometric compatibility between a Hermitian metric and a closed form. It ends with a cohomological package: pure Hodge structures, period pairings, Hodge classes of cycles, functoriality under holomorphic maps, and Lefschetz constraints on cup product. The main lesson, developed from local potentials in Chapter 1 through Laplacians, Hodge decomposition, the $\partial\bar\partial$ lemma, Lefschetz theory, and Hodge structures, is that compact Kähler manifolds are topologically special because their differential forms carry enough elliptic and complex structure to control cohomology. [/remark] ## Beyond and Connections The results in this course form a single package rather than a list of isolated tools. The starting point is the [Kähler manifold](/wiki/kahler-manifold): its closed $(1,1)$-form gives both a symplectic operator $L$ on forms and the Hermitian identities needed to compare the $d$, $\partial$, and $\bar{\partial}$ Laplacians. That comparison is what turns the Riemannian Hodge theorem into [Hodge decomposition](/wiki/hodge-decomposition) and makes the Hodge numbers into complex-geometric invariants. The Lefschetz side of the story explains why Kähler cohomology is much more rigid than ordinary de Rham cohomology. [Hard Lefschetz](/wiki/hard-lefschetz-theorem) identifies complementary degrees by cupping with powers of the Kähler class, while the primitive decomposition isolates the part of the middle cohomology where the intersection form has meaningful sign constraints. On compact complex surfaces, this is the bridge from differential geometry to the topology of the intersection pairing and to restrictions on which manifolds can carry Kähler structures. The $\partial\bar{\partial}$ lemma is the mechanism that makes these structures compatible with deformation and periods. It says that several natural notions of exactness agree on compact Kähler manifolds, so cohomology classes can be moved between de Rham, Dolbeault, and harmonic representatives without losing type information. This is why the period viewpoint is not an extra topic bolted onto the end: periods record how the Hodge filtration varies, and the Kähler identities are what make that filtration visible inside ordinary cohomology. Seen this way, the course connects four recurring themes: analytic representatives, type decompositions, Lefschetz symmetries, and surface-level intersection geometry. The hypotheses matter throughout. Dropping compactness weakens the Hodge theorem; dropping the Kähler condition breaks the clean Laplacian comparison; dropping positivity changes the Lefschetz and intersection-form conclusions. The value of the Kähler package is precisely that these analytic, algebraic, and topological constraints hold at the same time. The $\partial\bar\partial$ lemma is the bridge from local complex differential operators to global Hodge-theoretic structure. It explains why compact Kähler manifolds have especially rigid cohomology: de Rham, Dolbeault, Bott-Chern, and Aeppli viewpoints are not independent accidents, but different shadows of the same exactness principle. This connects the material here to Hodge decomposition, Kähler identities, harmonic representatives, Bott-Chern cohomology, and the comparison between Kähler and non-Kähler complex manifolds. The same ideas also reappear in deformation theory and algebraic geometry. When the $\partial\bar\partial$ lemma holds, Hodge numbers behave more stably, period maps have cleaner local structure, and many analytic correction arguments can be carried out without leaving a prescribed bidegree. When it fails, examples such as the Iwasawa manifold show exactly where Kähler-style intuition breaks: pure-type exactness can fragment into genuinely different conditions. ## References - Phillip Griffiths and Joseph Harris, *Principles of Algebraic Geometry*. - Claire Voisin, *Hodge Theory and Complex Algebraic Geometry I*. - Daniel Huybrechts, *Complex Geometry: An Introduction*. - Jean-Pierre Demailly, *Complex Analytic and Differential Geometry*.