Past exam of the mathematics course of the University of Cambridge / 2018 / iii / Paper 118 (Complex Manifolds)

If $M$ has complex dimension $n$, its complexified cotangent bundle splits into the $(1,0)$ and $(0,1)$ cotangent bundles. The space of global smooth differential forms of type (p, q) isIn holomorphic coordinates, its elements have the form $\alpha ={\sum }_{I,J}{a}_{I,J}d{z}^I\wedge d{\bar{z}}^J$, where the coefficients are complex-valued smooth functions, $|I|=p$, and $|J|=q$.

The exterior derivative decomposes as $d=\partial +\bar{\partial }$. The conjugate Dolbeault operator $\partial$ and Dolbeault operator $\bar{\partial }$ are its type components:They have bidegrees $(1,0)$ and $(0,1)$ respectively. Keeping the displayed wedge order avoids hiding a sign; moving $d{\bar{z}}_k$ past $d{z}^I$ introduces $(-1{)}^p$. Decomposing $d^2=0$ by type gives ${\partial }^2={\bar{\partial }}^2=0$ and $\partial \bar{\partial }+\bar{\partial }\partial =0$.

Complex conjugation interchanges the Dolbeault operator and conjugate Dolbeault operator. Thus $\partial \alpha =0$ implies $\bar{\partial }\bar{\alpha }=0$, and $\bar{\alpha }$ has type $(q,p)$. Its antiholomorphic degree is $p\ge 1$, so the allowed Dolbeault-Poincaré lemma on the polydisc gives a form $\gamma \in {\mathcal{A}}^{q,p-1}$ satisfying $\bar{\partial }\gamma =\bar{\alpha }$.

Conjugate this identity and put $\beta =\bar{\gamma }$. This proves the conjugate Dolbeault-Poincaré lemma:

A pure-type $d$-closed form is both $\partial$-closed and $\bar{\partial }$-closed, because these two derivatives have different types. Let $k=p+q$. The smooth Poincaré lemma on the polydisc gives $\eta \in {\mathcal{A}}_{\mathbb{C}}^{k-1}$ with $d\eta =\alpha$. We will modify $\eta$ by exact terms without changing $d\eta$.

Write $\eta ={\sum }_r{\eta }^{r,k-1-r}$. Starting with the smallest holomorphic degree, eliminate every component with $r<p-1$. Once the lower components are zero, the $(r,k-r)$ component of $d\eta$ says $\bar{\partial }{\eta }^{r,k-1-r}=0$, since $r<p$. Its antiholomorphic degree is positive, so the Dolbeault-Poincaré lemma supplies ${\nu }^{r,k-2-r}$ with $\bar{\partial }\nu ={\eta }^{r,k-1-r}$. Replace $\eta$ by $\eta -d\nu$: this removes the component at $r$ and changes only the next holomorphic degree.

Next, work downwards from the largest holomorphic degree and eliminate the components with $r>p$. Once the higher components are zero, the corresponding component of $d\eta$ gives $\partial {\eta }^{r,k-1-r}=0$. By part (b), write this as $\partial {\nu }^{r-1,k-1-r}$ and again subtract $d\nu$. This affects only the next lower holomorphic degree. The two finite procedures leaveThe components of $d\eta =\alpha$ outside type $(p,q)$ now give $\bar{\partial }u=0$ and $\partial v=0$, while its $(p,q)$ component gives $\alpha =\partial u+\bar{\partial }v$.

Since $q\ge 1$, the Dolbeault-Poincaré lemma gives $u=\bar{\partial }a$ with $a$ of type $(p-1,q-1)$. Since $p\ge 1$, part (b) gives $v=\partial b$ of the same type. The anticommutation identity therefore yieldsThis proves the Bott-Chern Poincaré lemma using precisely the permitted primitives:

The natural map between Bott-Chern cohomology and Dolbeault cohomology isIt is well-defined: a $d$-closed pure-type form is $\bar{\partial }$-closed, and $\partial \bar{\partial }\beta =-\bar{\partial }(\partial \beta )$ is $\bar{\partial }$-exact.

For surjectivity, choose the unique harmonic differential form $h$ representing a Dolbeault cohomology class using Dolbeault Hodge decomposition. The Kähler Laplacian identity ${\Delta }_d=2{\Delta }_{\bar{\partial }}$ makes $h$ a harmonic differential form for $d$, hence $d$-closed. Sending the Dolbeault class to $[h{]}_{BC}$ consequently gives a right inverse.

For injectivity, the ddbar lemma says directly that a $d$-closed, $\bar{\partial }$-exact pure-type form is $\partial \bar{\partial }$-exact. One can also exhibit its primitive. Let $G_d$ be the Green operator of the Hodge Laplacian, and put $G_{\bar{\partial }}=2G_d$. If $\alpha$ is $d$-closed and $\bar{\partial }$-exact, its harmonic projection vanishes and the Kähler identities giveHere $\partial G_{\bar{\partial }}\alpha =\bar{\partial }{G}_{\bar{\partial }}\alpha =0$ because the Green operator commutes with these differentials, and ${\bar{\partial }}^{\ast }=-i[\Lambda ,\partial ]$. Thus $\alpha$ represents zero in Bott-Chern cohomology. We have constructed the canonical isomorphism and its harmonic inverse:

Choose a flasque resolution of the sheaf of abelian groups $\mathcal{F}$,A flasque sheaf has surjective restriction maps; such resolutions exist for every sheaf of abelian groups. Apply the global section functor to obtain a cochain complex. The sheaf cohomology groups areFor $i=0$, the denominator is zero and $H^0(X,\mathcal{F})=\Gamma (X,\mathcal{F})$. Different resolutions give canonically isomorphic groups; the resolution principle for sheaf cohomology is what permits other acyclic resolutions to compute these same groups.

Let ${\mathcal{C}}_{\mathbb{C}}^{\infty }$ be the additive sheaf of complex-valued smooth functions, and ${\mathcal{C}}_{\mathbb{C}}^{\infty ,\ast }$ its multiplicative sheaf of nowhere-zero functions. The smooth exponential sequence isIts kernel is the integer-valued locally constant functions, namely the integer constant sheaf. It is surjective on stalks because a nowhere-zero smooth function has a smooth logarithm on a sufficiently small neighbourhood.

The additive smooth-function sheaf is a fine sheaf, using a partition of unity, and the smooth manifold is a paracompact space. Its positive-degree sheaf cohomology vanishes. The long exact sequence in sheaf cohomology consequently gives an isomorphism

To connect this with bundles, trivialize a smooth complex line bundle over an open cover. Its transition functions $g_{ij}$ form a multiplicative Čech cocycle, satisfying $g_{ij}{g}_{jk}=g_{ik}$. Changing trivializations changes it by a Čech coboundary. Conversely, any such cocycle glues trivial line bundles, and cohomologous cocycles give isomorphic bundles. Thus the isomorphism classes are $H^1(X,{\mathcal{C}}_{\mathbb{C}}^{\infty ,\ast })$.

The connecting isomorphism is the First Chern class. Locally choose logarithms $g_{ij}=e^{2\pi i{f}_{ij}}$; on triple intersectionsrepresents that class. Combining the gluing classification with $\delta$ gives the requested bijection:

The restriction maps of this point sheaf satisfy the presheaf composition identities: every possible restriction is either the identity on $\mathbb{C}$ or the unique map to the zero group. We check the sheaf identity and gluing axioms for any open cover $U={\bigcup }_i{U}_i$.

If $p\notin U$, every section group in the cover is zero, so a compatible family has the unique zero gluing. If $p\in U$, at least one $U_i$ contains $p$. For any two members containing $p$, their intersection also contains $p$, and compatibility forces their complex values to be equal, since the restriction maps there are identities. Call the common value $c$. Members not containing $p$ have only the zero section. The section $c\in {\mathbb{C}}_p(U)$ restricts to every member of the family, proving existence of a gluing; its restriction to any member containing $p$ determines $c$, proving uniqueness.

This also proves local identity for sections, and ${\mathbb{C}}_p(\varnothing )=0$ handles the empty open set. HenceIt is the point-pushforward sheaf, usually called a skyscraper sheaf. Its restrictions are all surjective, so it is also a flasque sheaf. No separation assumption on $X$ is needed for the preceding sheaf-axiom argument.

Evaluation is surjective as a sheaf morphism: near $p$, any prescribed value is supplied by a constant holomorphic function. We therefore have a short exact sequence of sheaveswhere ${\mathcal{I}}_p$ is the point ideal sheaf and ${\mathcal{O}}_X$ the sheaf of holomorphic functions. By part (c), the skyscraper sheaf is flasque, so $H^0(X,{\mathbb{C}}_p)=\mathbb{C}$ and $H^i(X,{\mathbb{C}}_p)=0$ for $i>0$.

On the connected compact elliptic curve, the maximum modulus principle makes every global holomorphic function constant. Consequently $H^0(X,{\mathcal{O}}_X)=\mathbb{C}$, and evaluation on global sections is an isomorphism. Also $H^1(X,{\mathcal{O}}_X)\cong \mathbb{C}$: by Serre duality for compact complex manifolds, its dual is $H^0(X,K_X)$, generated by the nowhere-zero form $dz$ descended from $\mathbb{C}$. Higher cohomology of ${\mathcal{O}}_X$ vanishes by the Dolbeault theorem in complex dimension one.

The long exact sequence in sheaf cohomology thus beginsIt follows that

The tautological bundle $\mathcal{O}(-1)$ over Complex projective space has fibre over $[\ell ]$ equal to the line $\ell \subset {\mathbb{C}}^{r+1}$. Define $\mathcal{O}(1)=\mathcal{O}(-1{)}^{\ast }$ and the twisting sheaf on projective space, viewed analytically as a holomorphic line bundle, by

For ${\mathbb{CP}}^1$, use coordinates $z=Z_1/Z_0$ and $w=Z_0/Z_1=1/z$. The tautological frames $(1,z)$ and $(w,1)$ differ by $z^{-1}$, so frames $e_0,e_1$ for $\mathcal{O}(d)$ satisfy $e_1=z^d{e}_0$. A global holomorphic section is therefore described by entire functions $s_0(z),s_1(w)$ withWrite $s_0(z)={\sum }_{k\ge 0}{a}_k{z}^k$. Holomorphicity of $s_1$ at zero permits only the powers $w^{d-k}$ with $d-k\ge 0$. For $d\ge 0$, this says $s_0$ is a polynomial of degree at most $d$; for $d<0$, every coefficient must vanish. Equivalently, when $d\ge 0$ the sections are homogeneous polynomials of degree $d$ in $Z_0,Z_1$. Thus

Use the integral normalization of the Chern connection curvature. For a holomorphic local frame $e$, put $h_e=h(e,e)>0$. The fundamental form of a Hermitian holomorphic line bundle isThis follows from the local formula for the Chern connection on a line bundle, $F_h=\bar{\partial }\partial \log h_e$. It is a closed real (1, 1)-form representing the First Chern class. If $h_e=e^{-\phi }$, the formula becomes $(i/(2\pi ))\partial \bar{\partial }\phi$.

The Hermitian metric is positive when this is a positive real (1, 1)-form: for every nonzero $v\in T^{1,0}M$,Equivalently, the matrix $({\partial }^2\phi /\partial z_j\partial {\bar{z}}_k)$ is a positive-definite matrix in every frame. Such a metric makes $\mathcal{L}$ a positive holomorphic line bundle. Omitting the normalization factor $2\pi$ changes none of these positivity conditions.

Apply Serre duality for compact complex manifolds to $\mathcal{L}\otimes K_M$, where $K_M$ is the canonical bundle:We can show the group on the right vanishes directly from the embedding. Since ${\mathcal{L}}^{-1}=i^{\ast }\mathcal{O}(-1)$, the tautological bundle inclusion gives an inclusion of holomorphic vector bundlesA global holomorphic section of ${\mathcal{L}}^{-1}$ consequently determines $r+1$ global holomorphic functions on $M$. Compactness and the maximum modulus principle make these functions constant on every connected component. On such a component the section is therefore a constant vector $v$ lying in the tautological line $i(x)$ for every $x$.

If $v\ne 0$, all the points $i(x)$ equal $[v]$. This contradicts the fact that $i$ is an embedding of a component of positive dimension $n\ge 1$. Hence $v=0$ on every component, and $H^0(M,{\mathcal{L}}^{-1})=0$. Duality proves

The Dolbeault theorem identifies $H^2(M,{\mathcal{O}}_M)$ with $H_{\bar{\partial }}^{0,2}(M)$. It vanishes by hypothesis, and complex conjugation in the Hodge decomposition theorem for compact Kähler manifolds makes $H^{2,0}$ vanish too. Thus every real degree-two cohomology class has type $(1,1)$.

Let $\omega$ be a Kähler form. Choose a rational cohomology class $\eta$ sufficiently close to $[\omega ]$. More concretely, using harmonic differential forms as representatives for a fixed Kähler metric, the harmonic form representing $\eta$ is close to $\omega$ in every smooth norm: harmonic differential forms constitute a finite-dimensional vector space. It is a real closed $(1,1)$ form, and is positive if sufficiently close to $\omega$, by compactness. This is the openness of the Kähler cone. Multiplying by a positive integer gives an integral Kähler class $\kappa =N\eta$.

The holomorphic exponential sequence contains the cohomology segmentChoose an integral lift of $\kappa$. Exactness gives a holomorphic line bundle $\mathcal{L}$ having that First Chern class. We must ensure that it has a positive metric, rather than merely a positive representative of its class.

Choose any Hermitian metric $h_0$ and let ${\theta }_0$ be its normalized curvature form of a Hermitian holomorphic line bundle. If $\theta$ is the positive form representing $\kappa$, then $\theta -{\theta }_0$ is an exact real $(1,1)$ form. The ddbar lemma gives a real smooth function $u$ withReplace $h_0$ by $h=e^{-u}{h}_0$. Its normalized curvature is $\theta$, so $\mathcal{L}$ is a positive holomorphic line bundle. The Kodaira embedding theorem now applies: sufficiently many sections of a sufficiently high tensor power define the desired embedding. Therefore

For the Calabi-Yau threefold in the stated convention, the Hodge decomposition theorem for compact Kähler manifolds givesIn particular, the Dolbeault theorem yields $H^1(M,{\mathcal{O}}_M)=0$. By Serre duality for compact complex manifolds in dimension three,The canonical bundle is trivial, so the group on the right equals $H^1(M,{\mathcal{O}}_M)=0$. Thus $H^2(M,{\mathcal{O}}_M)=0$, and part (c) gives

Let $\omega$ be the Kähler form associated with the metric. The Lefschetz operator of a Kähler manifold and adjoint Lefschetz operator areHere the adjoint is taken with respect to the pointwise Hermitian inner product on complexified forms, or equivalently the global $L^2$ inner product. It lowers degree by two and bidegree by $(1,1)$. In terms of the Hodge star operator, $\Lambda ={\ast }^{-1}L\ast$.

Since $d\omega =0$, $dL=Ld$, so $L$ takes closed differential forms to closed differential forms and exact differential forms to exact differential forms. This already gives its action on de Rham cohomology. For both operators together, use the Kähler identities: $L$ commutes with the Hodge Laplacian, and taking adjoints shows that $\Lambda$ does too. Thus both preserve harmonic differential forms.

When $M$ is compact, the Hodge decomposition theorem identifies $H^k(M,\mathbb{C})$ with its space of harmonic $k$-forms. If $h_{\alpha }$ is the unique harmonic differential form representing $[\alpha ]$, defineThe resulting forms are harmonic and therefore closed, and uniqueness of $h_{\alpha }$ makes these definitions independent of the initial representative. The cohomological definition of $\Lambda$ uses harmonic differential forms as representatives: $\Lambda$ need not take an arbitrary closed form to a closed form.

Represent the middle-degree class by its unique harmonic differential form $h$. The Lefschetz commutator on degree $k$ is $[\Lambda ,L]=(n-k)\mathrm{id}$, so at $k=n$ the operators commute. Since the adjoint Lefschetz operator is the adjoint of the Lefschetz operator of a Kähler manifold,Both $Lh$ and $\Lambda h$ are harmonic. A harmonic form represents zero in de Rham cohomology exactly when the form itself is zero. Consequently $L[\alpha ]=0$ if and only if $Lh=0$, which by the norm identity is equivalent to $\Lambda h=0$, hence to $\Lambda [\alpha ]=0$. ThereforeThis identifies the two middle-degree descriptions of a primitive differential form on a Kähler manifold at the cohomology level.

The assertion is pointwise, so choose an oriented orthonormal real coframe $x_1,y_1,x_2,y_2$ compatible with the complex structure, with Kähler form $\omega =x_1\wedge y_1+x_2\wedge y_2$. Real $(1,1)$ forms have a four-dimensional space; imposing $\omega \wedge \alpha =0$ leaves the three-dimensional space spanned byThese are precisely the real primitive differential forms on a Kähler manifold of type $(1,1)$ in dimension two. Complexifying spans all complex primitive $(1,1)$ forms.

Using the complex orientation $x_1\wedge y_1\wedge x_2\wedge y_2$, the Hodge star operator exchanges $x_1\wedge y_1$ with $x_2\wedge y_2$, sends $x_1\wedge x_2$ to $-y_1\wedge y_2$, and exchanges $x_1\wedge y_2$ with $y_1\wedge x_2$. Thus $\ast A=-A$, $\ast B=-B$, and $\ast C=-C$. By complex linearity,This is the fact that primitive (1,1)-forms on a Kähler surface are anti-self-dual.

Use the usual connected-manifold convention for this assertion. Suppose, for contradiction, that $H^2(M,{\mathcal{O}}_M)=0$. The Dolbeault theorem gives $H^{0,2}=0$, and the Hodge decomposition theorem for compact Kähler manifolds and its conjugation symmetry give $H^{2,0}=0$. Hence every real class in $H^2(M,\mathbb{R})$ has a real harmonic differential form of type $(1,1)$ as its representative.

The intersection form is $Q([u],[v])={\int }_{M}u\wedge v$. Its value on the Kähler class is positive: $Q([\omega ],[\omega ])={\int }_M{\omega }^2>0$. Every real harmonic $(1,1)$ form splits uniquely as $c\omega +u_0$, with $u_0$ primitive and $c$ constant. Indeed, $\Lambda u$ is a harmonic function, hence constant by connectedness, and $\Lambda \omega =2$; subtracting $c\omega$ kills $\Lambda u$, and part (b) makes $u_0$ primitive.

Part (c) gives $\ast u_0=-u_0$, while $\omega \wedge u_0=0$. ThusThis proves the Hodge index theorem for compact Kähler surfaces in this case: $Q$ has signature $(1,b_2-1)$.

On the alleged two-dimensional totally isotropic subspace $V$, the linear functional $v\mapsto Q(v,[\omega ])$ has a nonzero vector in its kernel. Its harmonic differential form representative is primitive, so the preceding negative-definiteness statement gives $Q(v,v)<0$, contradicting the vanishing of $Q$ on $V$. ThereforeConnectedness is essential if one's definition of a manifold allows disconnected spaces. For $M=Y\bigsqcup Y$, with $Y={\mathbb{CP}}^1\times {\mathbb{CP}}^1$, one has $H^2(M,{\mathcal{O}}_M)=0$. Take on each component the pullback of the degree-two class from its first factor. Each class has square zero, and their mutual pairing is zero because their supports lie on different components. Their span is then a two-dimensional totally isotropic subspace, disproving the unqualified disconnected version.

A holomorphic differential form $\alpha$ of type $(p,0)$ satisfies $\bar{\partial }\alpha =0$. The adjoint ${\bar{\partial }}^{\ast }$ lowers antiholomorphic degree, so ${\bar{\partial }}^{\ast }\alpha =0$ as well. Therefore ${\Delta }_{\bar{\partial }}\alpha =0$.

The Kähler Laplacian identity gives ${\Delta }_d\alpha =2{\Delta }_{\bar{\partial }}\alpha =0$. Compactness lets us use the $L^2$ identity for the Hodge Laplacian:Both terms are nonnegative, soThis proves that holomorphic forms on a compact Kähler manifold are closed. The original PDF uses ${\Omega }_M^p$ and ${\mathcal{A}}_{M,\mathbb{C}}^{p+1}$; the extra symbols in the local TeX superscripts are transcription errors.

The holomorphic de Rham complex is exact locally, which is the criterion for an exact sequence of sheaves. Work in a small star-shaped holomorphic coordinate neighbourhood in ${\mathbb{C}}^2$, centred at zero. A holomorphic function has $\partial f=0$ exactly when it is constant on this neighbourhood, giving exactness at ${\mathcal{O}}_M$ and the injection of the constant sheaf.

For exactness at ${\Omega }_M^1$, let $\alpha =a_1(z)d{z}_1+a_2(z)d{z}_2$ be holomorphic and $\partial$-closed. Then $\partial a_1/\partial z_2=\partial a_2/\partial z_1$. Define the holomorphic functionDifferentiation under the integral and the closedness identity giveThus $\partial F=\alpha$.

For exactness at ${\Omega }_M^2$, any holomorphic differential form of degree two is $\beta =b(z)d{z}_1\wedge d{z}_2$ and is automatically $\partial$-closed by dimension. DefineIts derivative isThese local primitives prove the holomorphic Poincaré lemma in the degrees needed here. Exactness on every stalk givesThis is sheaf exactness; the local primitives need not glue to global ones.

For degree zero, compactness and the maximum modulus principle make holomorphic functions constant on each connected component. For degree two, a holomorphic differential form is $\bar{\partial }$-closed and its $\partial$ derivative has type $(3,0)$, which vanishes on a complex surface. Degrees above two have no nonzero holomorphic forms. It remains to treat a holomorphic differential form $\alpha$ of degree one.

Set $\beta =d\alpha =\partial \alpha$. Its coefficients are holomorphic, so $\beta$ is a holomorphic differential form of degree two, hence closed by the degree-two observation. Its conjugate is closed too. The Stokes theorem givesThe integrand is nonnegative in the complex orientation. In local coordinates, if $\beta =b(z)d{z}_1\wedge d{z}_2$ with $z_j=x_j+i{y}_j$, thenIt is strictly positive wherever $\beta$ is nonzero. Its zero integral therefore forces $\beta =0$ everywhere. Thus holomorphic forms on a compact complex surface are closed, even when the surface is not Kähler:

Let ${\mathcal{Z}}_M^1$ be the sheaf of closed holomorphic one-forms, namely $\ker (\partial :{\Omega }_M^1\to {\Omega }_M^2)$. The local exactness proved in part (b) gives the short exact sequence of sheavesIts long exact sequence in sheaf cohomology beginsThe first arrow is an isomorphism: on a compact complex manifold, a global holomorphic function is constant on each connected component, just as a global section of the constant sheaf is. Thus $\delta$ is injective. By part (c), every global holomorphic differential form of degree one is closed, so $H^0(M,{\mathcal{Z}}_M^1)=H^0(M,{\Omega }_M^1)$.

Concretely, for a holomorphic differential form of degree one choose local holomorphic primitives $f_i$. The connecting class $\delta (\alpha )$ is represented by the locally constant Čech cocycle $f_j-f_i$ on overlaps. The following arrow is induced by the inclusion of the constant sheaf into the sheaf of holomorphic functions. Exactness is inherited from the long exact sequence, giving