Akizuki-Nakano vanishing theorem

Intuition [Beginner]

The Akizuki-Nakano vanishing theorem extends Kodaira vanishing across the full Hodge bidegree of a smooth projective complex variety. Where Kodaira vanishing concerns only the canonical sheaf paired with an ample line bundle, Akizuki-Nakano allows the canonical sheaf to be replaced by any sheaf of holomorphic $p$-forms. The conclusion is sharp: on an $n$-dimensional smooth projective variety, the cohomology of $p$-forms with values in an ample line bundle vanishes whenever the total bidegree $p+q$ exceeds $n$.

The picture is a square. Lay out the Hodge groups — the cohomology of $p$-forms paired with $L$ — as an $(n+1)\times (n+1)$ array with $p$ on one axis and $q$ on the other. Akizuki-Nakano says: take the anti-diagonal $p+q=n$; everything above and to the right of it is zero whenever $L$ is positive. Kodaira vanishing is the bottom row ($p=n$). Akizuki-Nakano fills out the upper-right triangle.

Yasuo Akizuki and Shigeo Nakano proved this in a 1954 note in the Proceedings of the Japan Academy, vol. 30, building directly on Kodaira's harmonic-form method. The mechanism is the Bochner-Kodaira-Nakano formula, a refined version of the Kähler identities that records how positive curvature acts on $(p,q)$-forms valued in a positively curved bundle.

Visual [Beginner]

A schematic Hodge bidegree square: the columns index $p$, the rows index $q$, and the cells colour-code whether the cohomology of $p$-forms paired with $L$ is forced to be zero by Akizuki-Nakano. The diagonal $p+q=n$ separates the unforced lower-left triangle from the forced-zero upper-right.

Each column above and to the right of the diagonal collapses to zero. The bottom-row entries ($p=n$) recover the Kodaira vanishing groups.

Worked example [Beginner]

Take $X={\mathbb{P}}^2$, so $n=2$, and let $L={\mathcal{O}}_{{\mathbb{P}}^2}(d)$ with $d\ge 1$, an ample line bundle.

Akizuki-Nakano predicts that the cohomology of $p$-forms paired with $L$ vanishes whenever $p+q>2$. The bidegree square has nine cells indexed by $(p,q)$ with $0\le p,q\le 2$. Three cells lie above the diagonal: $(2,1)$, $(2,2)$, $(1,2)$. (The diagonal itself has $(0,2),(1,1),(2,0)$; the lower triangle has $(0,0),(0,1),(1,0)$.)

Step 1. The known sheaves on ${\mathbb{P}}^2$: ${\Omega }^0=\mathcal{O}$, ${\Omega }^2={\omega }_{{\mathbb{P}}^2}=\mathcal{O}(-3)$, and the Euler sequence $0\to {\Omega }^1\to \mathcal{O}(-1{)}^{\oplus 3}\to \mathcal{O}\to 0$ pins down ${\Omega }^1$.

Step 2. Cell $(p,q)=(2,1)$: $H^1({\mathbb{P}}^2,\mathcal{O}(d-3))$, obtained by pairing ${\Omega }^2=\mathcal{O}(-3)$ with $\mathcal{O}(d)$. For ${\mathbb{P}}^2$, the middle cohomology of any line bundle is zero (only $H^0$ and $H^2$ contribute), so the cell vanishes.

Step 3. Cell $(p,q)=(2,2)$: $H^2({\mathbb{P}}^2,\mathcal{O}(d-3))$. By Serre duality this equals $H^0({\mathbb{P}}^2,\mathcal{O}(-d){)}^{\vee }$. For $d\ge 1$, $\mathcal{O}(-d)$ has no global sections, so the cell vanishes.

Step 4. Cell $(p,q)=(1,2)$: from the Euler sequence after pairing with $\mathcal{O}(d)$ one shows that the $H^2$ of ${\Omega }^1$-paired-with-$\mathcal{O}(d)$ on ${\mathbb{P}}^2$ is zero for every $d\ge 1$ (the long exact sequence reduces to the vanishing of $H^2(\mathcal{O}(d-1{)}^{\oplus 3})$, which vanishes since $d-1\ge 0>-3$).

What this tells us: every cell strictly above the anti-diagonal vanishes, exactly as Akizuki-Nakano predicts.

Check your understanding [Beginner]

Formal definition [Intermediate+]

Let $X$ be a smooth projective complex variety of dimension $n$, equipped with a Kähler metric $\omega$, and let $L$ be a holomorphic line bundle with a Hermitian metric $h_L$ of positive Chern curvature. Write ${\Omega }_X^p$ for the sheaf of holomorphic $p$-forms.

Theorem (Akizuki-Nakano vanishing). If $L$ is ample, then

$$H^q(X,{\Omega }_X^p\otimes L)=0\text{for all }p,q\text{ with }p+q>n.$$

The result was announced in Akizuki-Nakano 1954 (Proc. Japan Acad. 30) and developed in Nakano 1955 (J. Math. Soc. Japan 7). The conclusion specialises to Kodaira vanishing at $p=n$.

Hodge-theoretic interpretation. The Dolbeault isomorphism identifies $H^q(X,{\Omega }_X^p\otimes L)$ with the Dolbeault cohomology $H_{\bar{\partial }}^{p,q}(X,L)$ of $L$-valued $(p,q)$-forms. The Akizuki-Nakano theorem is therefore a vanishing statement about $L$-valued Dolbeault cohomology in the upper-right of the $(p,q)$ square.

Equivalent formulation via Serre duality. Applying Serre duality $H^q(X,{\Omega }_X^p\otimes L)\cong H^{n-q}(X,{\Omega }_X^{n-p}\otimes L^{-1}{)}^{\vee }$ (note ${\Omega }_X^p{}^{\vee }\otimes {\omega }_X={\Omega }_X^{n-p}$) one obtains the dual statement:

$$H^j(X,{\Omega }_X^i\otimes L^{-1})=0\text{for all }i,j\text{ with }i+j<n.$$

So an ample line bundle has no Dolbeault cohomology in the lower-left of the dual square — a perfect mirror image of the upper-right vanishing.

Counterexamples to common slips [Intermediate+]

• Slip: dropping the Kähler hypothesis. On a non-Kähler compact complex manifold the Bochner-Kodaira-Nakano formula carries torsion terms involving $\partial \omega$ and $\bar{\partial }\omega$, and the positivity argument breaks. Section 5 below presents an explicit failure on a Hopf surface.
• Slip: replacing ample by nef. A nef line bundle need not be Akizuki-Nakano-vanishing in every bidegree; only the canonical-twist case ($p=n$, the Kawamata-Viehweg setting) survives the relaxation. For other $p$, additional structure (e.g., a multiplier-ideal correction) is required.
• Slip: omitting smoothness. On a singular projective variety the sheaves ${\Omega }_X^p$ are no longer locally free, and the statement must be replaced by one involving ${\pi }_{\ast }{\Omega }_{\tilde{X}}^p$ on a resolution (Grauert-Riemenschneider) or by reflexive differentials.
• Slip: extending to characteristic $p$. Akizuki-Nakano fails in positive characteristic, by the same Raynaud-type counterexamples that defeat Kodaira; Deligne-Illusie 1987 lifts a partial version to characteristic 0 by reduction mod $p$ when the variety admits a flat lift to $W_2(k)$.

Key theorem with proof [Intermediate+]

Theorem (Akizuki-Nakano 1954). Let $X$ be a compact Kähler manifold of complex dimension $n$ and $L\to X$ a holomorphic line bundle admitting a Hermitian metric $h$ whose Chern curvature ${\Theta }_h(L)$ is a positive $(1,1)$-form. Then $H^q(X,{\Omega }_X^p\otimes L)=0$ for every $(p,q)$ with $p+q>n$.

Proof. The argument is a refinement of Kodaira's harmonic-form approach, using the Bochner-Kodaira-Nakano formula on $(p,q)$-forms.

Step 1 — Harmonic representation. By the Hodge theorem applied to the Dolbeault complex of $L$, every class in $H_{\bar{\partial }}^{p,q}(X,L)\cong H^q(X,{\Omega }_X^p\otimes L)$ is represented by a unique ${\bar{\partial }}_L$-harmonic $L$-valued $(p,q)$-form $\alpha$, satisfying ${\bar{\partial }}_L\alpha =0$ and ${\bar{\partial }}_L^{\ast }\alpha =0$. The Dolbeault Laplacian is ${\Delta }_{{\bar{\partial }}_L}={\bar{\partial }}_L{\bar{\partial }}_L^{\ast }+{\bar{\partial }}_L^{\ast }{\bar{\partial }}_L$, and harmonicity says ${\Delta }_{{\bar{\partial }}_L}\alpha =0$.

Step 2 — The Bochner-Kodaira-Nakano identity. On the bundle ${\Omega }_X^p\otimes L$ over a Kähler manifold, the two natural Laplacians ${\Delta }_{{\bar{\partial }}_L}$ and ${\Delta }_{D^{\prime }}$ (the latter being the Laplacian of the $(1,0)$-part $D^{\prime }=\partial +{\theta }_h$ of the Chern connection of $(L,h)$) differ by a curvature term:

$${\Delta }_{{\bar{\partial }}_L}-{\Delta }_{D^{\prime }}=[i{\Theta }_h(L),{\Lambda }_{\omega }],$$

where ${\Lambda }_{\omega }$ is the formal adjoint of wedging with the Kähler form $\omega$. This is the Bochner-Kodaira-Nakano identity, proved on a general Hermitian bundle in Nakano 1955; the right side is zeroth-order in $\alpha$, which is what makes the identity useful.

Step 3 — Pairing with the harmonic form. Take the $L^2$-inner product of the identity against $\alpha$:

$$⟨{\Delta }_{{\bar{\partial }}_L}\alpha ,\alpha ⟩-⟨{\Delta }_{D^{\prime }}\alpha ,\alpha ⟩=⟨[i{\Theta }_h(L),{\Lambda }_{\omega }]\alpha ,\alpha ⟩.$$

By harmonicity, $⟨{\Delta }_{{\bar{\partial }}_L}\alpha ,\alpha ⟩=0$. The second Laplacian is non-negative, so $⟨{\Delta }_{D^{\prime }}\alpha ,\alpha ⟩=\Vert D^{\prime }\alpha {\Vert }^2+\Vert D^{\prime \ast }\alpha {\Vert }^2\ge 0$. The identity reduces to

$$-(\Vert D^{\prime }\alpha {\Vert }^2+\Vert D^{\prime \ast }\alpha {\Vert }^2)=⟨[i{\Theta }_h(L),{\Lambda }_{\omega }]\alpha ,\alpha ⟩.$$

The left side is non-positive.

Step 4 — Positivity of the curvature commutator. The crucial computation, due to Akizuki-Nakano, evaluates the bracket $[i{\Theta }_h(L),{\Lambda }_{\omega }]$ on $(p,q)$-forms. At every point $x\in X$ choose a unitary frame in which $\omega =i{\sum }_{j}d{z}_j\wedge d{\bar{z}}_j$ and ${\Theta }_h(L)=i{\sum }_j{\gamma }_j(x)d{z}_j\wedge d{\bar{z}}_j$ with ${\gamma }_j(x)>0$ (possible because positivity is preserved under diagonalisation). Then for an $(p,q)$-form $\alpha ={\sum }_{|I|=p,|J|=q}{\alpha }_{IJ}d{z}_I\wedge d{\bar{z}}_J$ one computes

$$⟨[i{\Theta }_h(L),{\Lambda }_{\omega }]\alpha ,\alpha ⟩(x)=\sum _{I,J}\left(\sum _{j\in I}{\gamma }_j+\sum _{j\in J}{\gamma }_j-\sum _{j=1}^n{\gamma }_j\right)|{\alpha }_{IJ}{|}^2.$$

The bracketed quantity equals $\gamma (I)+\gamma (J)-\gamma ([n])$, where $\gamma (S):={\sum }_{j\in S}{\gamma }_j$. By inclusion-exclusion, $\gamma (I)+\gamma (J)-\gamma ([n])\ge \gamma (I\cap J)-\gamma ([n]\setminus (I\cup J))$. Using $|I|=p$, $|J|=q$, one has $|I|+|J|=p+q$, while $|[n]\setminus (I\cup J)|=n-|I\cup J|\le n-\max (p,q)$. The decisive inequality is strict whenever $|I|+|J|=p+q>n$: at least one index must be common to $I$ and $J$, contributing strictly positive weight, while no index can lie outside $I\cup J$ enough to neutralise it.

The conclusion is that, pointwise and for $p+q>n$,

$$⟨[i{\Theta }_h(L),{\Lambda }_{\omega }]\alpha ,\alpha ⟩(x)\ge C(x)|\alpha (x){|}^2$$

for a positive function $C(x)>0$ controlled by the smallest eigenvalue of ${\Theta }_h(L)$.

Step 5 — Conclusion. Integrating the inequality, $⟨[i{\Theta }_h(L),{\Lambda }_{\omega }]\alpha ,\alpha {⟩}_{L^2}\ge c\Vert \alpha {\Vert }^2$ for some constant $c>0$. Combined with Step 3, $-(\Vert D^{\prime }\alpha {\Vert }^2+\Vert D^{\prime \ast }\alpha {\Vert }^2)\ge c\Vert \alpha {\Vert }^2$. The left side is non-positive; the right is non-negative. Both must vanish, forcing $\alpha =0$. Since harmonic forms represent every cohomology class, $H^q(X,{\Omega }_X^p\otimes L)=0$ for $p+q>n$. $\square$

Bridge. This argument builds toward 04.09.11 Kodaira embedding theorem, where Akizuki-Nakano vanishing is invoked alongside Kodaira vanishing to ensure that sufficiently high tensor powers of an ample line bundle separate points and tangent vectors. The pointwise positivity computation appears again in 04.09.07 hard Lefschetz, where the same Lefschetz operator $L_{\omega }$ and its commutator with curvature controls a different bidegree manipulation. The foundational reason positivity-of-curvature kills upper-right cohomology is exactly the bidegree counting in Step 4: the bracket $[i{\Theta }_h(L),{\Lambda }_{\omega }]$ acts on $(p,q)$-forms by a quantity proportional to $p+q-n$ multiplied by the curvature eigenvalues. Putting these together identifies the geometric origin of all positivity-based vanishing: a single curvature operator, projected through the bidegree decomposition, distributes vanishing across the Hodge square. The bridge is between local pointwise positivity (the curvature eigenvalues ${\gamma }_j$) and global cohomological vanishing.

Exercises [Intermediate+]

Lean formalization [Intermediate+]

lean_status: partial — Mathlib contains the categorical framework for smooth projective varieties, ample line bundles, and sheaf cohomology, but the Bochner-Kodaira-Nakano identity and the Akizuki-Nakano vanishing theorem as named results are not yet in Mathlib. The companion file states the theorem and the comparison with Kodaira vanishing with sorry proof bodies.

import Mathlib.AlgebraicGeometry.Scheme

namespace Codex.AlgGeom.Hodge

-- Akizuki-Nakano: H^q(X; Omega^p_X ⊗ L) = 0 for p + q > n, L ample,
-- X smooth projective complex of dimension n.

end Codex.AlgGeom.Hodge

See lean/Codex/AlgGeom/Hodge/AkizukiNakanoVanishing.lean for the full module with the theorem statement and the Kodaira comparison.

Advanced results [Master]

The Bochner-Kodaira-Nakano formula on $(p,q)$-forms with values in $(L,h)$

The proof in §3 rested on the identity

$${\Delta }_{{\bar{\partial }}_L}-{\Delta }_{D_L^{\prime }}=[i{\Theta }_h(L),{\Lambda }_{\omega }]$$

on $A^{p,q}(X,L)$. The derivation begins with the Kähler identity $[\bar{\partial },{\Lambda }_{\omega }]=-i{D}^{\prime \ast }$ (Hodge's formula on the structure-sheaf bundle), and its twist by a Hermitian connection on $L$. For the Chern connection $D_L=D_L^{\prime }+{\bar{\partial }}_L$ with curvature ${\Theta }_h(L)$, the twisted Kähler identities read $[{\bar{\partial }}_L,{\Lambda }_{\omega }]=-i(D_L^{\prime }{)}^{\ast }$ and $[D_L^{\prime },{\Lambda }_{\omega }]=i{\bar{\partial }}_L^{\ast }$, with a curvature correction encoded in the commutators of $D_L^{\prime }$ and ${\bar{\partial }}_L$:

$$D_L^{\prime }{\bar{\partial }}_L+{\bar{\partial }}_L{D}_L^{\prime }={\Theta }_h(L).$$

Substituting and computing ${\Delta }_{{\bar{\partial }}_L}={\bar{\partial }}_L{\bar{\partial }}_L^{\ast }+{\bar{\partial }}_L^{\ast }{\bar{\partial }}_L$ and ${\Delta }_{D_L^{\prime }}=D_L^{\prime }(D_L^{\prime }{)}^{\ast }+(D_L^{\prime }{)}^{\ast }{D}_L^{\prime }$ yields the Bochner-Kodaira-Nakano identity. The point is that the curvature term is of order zero in $\alpha$, so the right side is a Hermitian form on $A^{p,q}(X,L)$ rather than a differential operator.

The pointwise positivity computation in Step 4 of the proof can be reorganised via the Lefschetz ${\mathfrak{sl}}_2$-action on $A^{\ast ,\ast }(X,L)$. Let $L_{\omega }=\omega \wedge \cdot$ be wedge-with-Kähler-form, ${\Lambda }_{\omega }$ its adjoint, and $H=[L_{\omega },{\Lambda }_{\omega }]$ the bidegree-counting operator. On $A^{p,q}$, $H$ acts by $(p+q-n)\cdot \mathrm{id}$. So the commutator $[i{\Theta }_h(L),{\Lambda }_{\omega }]$ for ${\Theta }_h(L)=\omega$ (the model case of curvature equal to the Kähler form, e.g., for $L=K_X^{-1}$ on a Fano variety with the Kähler-Einstein metric) reduces to $i[L_{\omega },{\Lambda }_{\omega }]=-H=-(p+q-n)$ on $A^{p,q}$. For $p+q>n$ this is strictly negative on the sign convention chosen here, and the sign flip in Step 3 turns it into the strict positivity needed.

For general positive $L$, the same calculation in a diagonalising frame produces the inequality of Step 4. The Nakano positivity hypothesis (positivity of $\Theta (E)$ as a form on $T_X\otimes E$, in the vector-bundle case of Exercise 6) is the right generalisation that preserves the pointwise positivity computation when $L$ is replaced by a higher-rank bundle $E$.

Comparison and unification with Kodaira vanishing — Akizuki-Nakano specialises to Kodaira when $p=n$

Setting $p=n$ in Akizuki-Nakano gives the vanishing $H^q(X,{\Omega }_X^n\otimes L)=H^q(X,{\omega }_X\otimes L)=0$ for $q>0$, exactly the original Kodaira vanishing statement (Kodaira 1953, Proc. Nat. Acad. Sci. 39, 1268-1273). The Akizuki-Nakano paper itself recognises this; its title in the original Proc. Japan Acad. 30 is Note on Kodaira-Spencer's proof of Lefschetz theorems, signalling the connection to Lefschetz hyperplane theorems through the bidegree square.

The bidegree axis $p$ runs from $0$ (zero-degree differentials, ${\Omega }_X^0={\mathcal{O}}_X$) to $n$ (canonical sheaf, ${\Omega }_X^n={\omega }_X$). Akizuki-Nakano holds across the full axis simultaneously; Kodaira is the right endpoint. The unifying viewpoint, due to Demailly, is to encode both as statements about the Hodge complex $({\Omega }_X^{\bullet },d)$ twisted by $L$, with the vanishing controlling the spectral sequence converging to twisted singular cohomology.

A second unification: the Akizuki-Nakano-Kodaira pyramid. For ample $L$, the entire upper-right triangle $\{(p,q):p+q>n\}$ vanishes; for nef-and-big $L$, only the canonical-twist corner (Kawamata-Viehweg) survives; for nef $L$, only partial vanishing in the canonical-twist (Kollár's injectivity); for pseudoeffective $L$, vanishing of ${\omega }_X\otimes L\otimes \mathcal{J}(h)$ for a singular metric. The Akizuki-Nakano theorem sits at the apex of this pyramid: strongest positivity hypothesis, strongest vanishing conclusion.

Algebraic generalisations — Kawamata-Viehweg, Demailly-Peternell-Schneider, Esnault-Viehweg

The transcendental proof of Akizuki-Nakano via the Bochner-Kodaira-Nakano formula can be replaced by purely algebraic arguments at the cost of restricting bidegree.

Kawamata-Viehweg 1982 (Kawamata, Math. Ann. 261, 43-46; Viehweg, J. reine angew. Math. 335, 1-8). For $L$ nef and big on a smooth projective $X$ over an algebraically closed field of characteristic zero, $H^q(X,{\omega }_X\otimes L)=0$ for $q>0$. The proof is via cyclic covers: take a high tensor power $L^{\otimes m}$ with a section vanishing on a normal-crossings divisor, build a cyclic cover ramified along this divisor, lift to a smooth ample-ish setting, then apply Kodaira on the cover and descend. The Bochner-Kodaira-Nakano formula plays no role.

Demailly-Peternell-Schneider 2001 (Internat. J. Math. 12, 689-741). The singular-Hermitian-metric framework: a pseudoeffective line bundle $L$ carries a singular Hermitian metric $h$ with positive curvature current ${\Theta }_h(L)\ge 0$. The multiplier ideal $\mathcal{J}(h)$ measures the failure of $h$ to be locally bounded; the Demailly-Peternell-Schneider vanishing is $H^q(X,{\omega }_X\otimes L\otimes \mathcal{J}(h))=0$ for $q>0$, generalising Nadel vanishing (Nadel 1990, Annals of Math. 132) and providing the cohomological backbone of the analytic minimal model program.

Esnault-Viehweg 1992 (Lectures on Vanishing Theorems, DMV Seminar 20, Birkhäuser). A fully algebraic proof of Kawamata-Viehweg using Deligne-Illusie's Hodge degeneration argument: lift to characteristic $p$, apply Frobenius decomposition, deduce vanishing on the lift, then descend. Works in characteristic zero without any analytic input. Esnault-Viehweg also derive Akizuki-Nakano-type vanishing at the bidegree level under mixed-Hodge-structure hypotheses.

Kollár 1986 (Annals of Math. 123, 11-42). The Kollár injectivity-and-vanishing package: for a proper surjective morphism $f:X\to Y$ with connected fibres, $R^i{f}_{\ast }{\omega }_X=0$ for $i$ exceeding the relative dimension. Includes the Kollár injectivity theorem on $H^i(X,{\omega }_X)$. Foundational for the Kollár-Shokurov framework of birational geometry.

The hierarchy of algebraic generalisations is therefore: Kodaira (positive $L$) $\Rightarrow$ Kawamata-Viehweg (nef + big $L$) $\Rightarrow$ Nadel-Demailly-Peternell-Schneider (pseudoeffective $L$, with multiplier ideal correction). At the bidegree level, only the Akizuki-Nakano corner survives the relaxation cleanly; off-corner bidegrees require finer machinery (singular metrics, multiplier ideals, mixed Hodge modules).

Failure on non-Kähler manifolds and the Bogomolov-Sommese twist

The Akizuki-Nakano proof depends on the Kähler condition through (a) the Kähler identities $[\bar{\partial },{\Lambda }_{\omega }]=-i{D}^{\prime \ast }$ and (b) the equality of Laplacians $2{\Delta }_{\bar{\partial }}={\Delta }_d$ on the structure-sheaf bundle. Both fail on non-Kähler compact complex manifolds.

Hopf surface counterexample. Let $X=({\mathbb{C}}^2\setminus \{0\})/\mathbb{Z}$ be the standard Hopf surface, $n=2$. The complex structure is non-Kähler: $H^2(X;\mathbb{R})=0$, so no Kähler class exists. The cohomology of holomorphic forms with values in any line bundle does not satisfy Akizuki-Nakano vanishing; explicit cohomology computations on the Hopf surface (Kodaira 1964 Amer. J. Math. 86) exhibit non-zero classes in bidegrees that would be forced to zero on a Kähler model.

Bogomolov-Sommese vanishing. In the non-Kähler setting, the partial replacement for Akizuki-Nakano is Bogomolov-Sommese (Bogomolov 1979 Math. USSR Izv. 13; Sommese 1976-78): for a smooth projective variety and a Weil divisor $D$ of negative self-intersection (in an appropriate sense), the natural maps $H^0(X,{\Omega }_X^p(\log D))\to H^0(X,{\Omega }_X^p\otimes {\mathcal{O}}_X(D))$ are zero. This cohomological vanishing controls subsheaves of logarithmic differentials and underpins Bogomolov's proof of boundedness for varieties of general type and Sommese's positivity criteria for hyperbolicity.

The Bogomolov-Sommese twist takes the Akizuki-Nakano-type vanishing and recasts it for logarithmic differential forms ${\Omega }_X^p(\log D)$ on a smooth pair $(X,D)$ with normal-crossings $D$. This is the input to log Hodge theory (Steenbrink, Deligne, Saito) and to the structure theory of orbifold differential forms (Greb-Kebekus-Kovács-Peternell 2011 Publ. Math. IHÉS 114).

Demailly's twisted Bochner-Kodaira inequality on non-Kähler manifolds. On a Hermitian (not necessarily Kähler) compact complex manifold $X$ with Hermitian metric $\omega$, Demailly proved a Bochner-Kodaira-Nakano-type identity with torsion corrections involving $\partial \omega$, $\bar{\partial }\omega$, and their adjoints. These torsion terms generically destroy the positivity argument, but in special cases (balanced metrics, $\partial \bar{\partial }$-manifolds, generalised Kähler structures) controlled torsion preserves enough positivity for partial vanishing results.

Cohomological positivity dictionary refined by bidegree

For a line bundle $L$ on a smooth projective complex variety:

• $L$ ample $\Rightarrow$ Akizuki-Nakano vanishing across the entire upper-right triangle of the Hodge square.
• $L$ nef and big $\Rightarrow$ Kawamata-Viehweg vanishing only at the canonical-twist corner.
• $L$ nef $\Rightarrow$ Kollár injectivity at canonical-twist.
• $L$ pseudoeffective $\Rightarrow$ Demailly-Peternell-Schneider singular-metric vanishing.
• $L$ semi-positive in the sense of Griffiths $\Rightarrow$ Griffiths vanishing $H^q(X,{\omega }_X\otimes \det E\otimes E)=0$ for $q>0$ at a vector bundle level (Griffiths 1969).

Each row of the dictionary refines a successively weaker positivity hypothesis to a successively narrower vanishing zone. The Akizuki-Nakano apex (ample $\Rightarrow$ full upper-right triangle) sets the maximal cohomological consequence of positivity in the line-bundle setting.

Synthesis. The Akizuki-Nakano vanishing theorem is the foundational reason that a single positive curvature operator distributes vanishing across the entire upper-right triangle of the Hodge bidegree square. The central insight is the Bochner-Kodaira-Nakano identity: the difference ${\Delta }_{{\bar{\partial }}_L}-{\Delta }_{D_L^{\prime }}$ equals the curvature commutator $[i{\Theta }_h(L),{\Lambda }_{\omega }]$, a zeroth-order Hermitian form whose pointwise positivity for $p+q>n$ is governed by counting indices common to $I$ and $J$. Putting these together with the Hodge harmonic-form theorem identifies cohomology classes with harmonic representatives, and the positivity argument forces them to vanish. This is exactly the structural fact that organises the cohomological consequences of curvature positivity: every theorem in the family — Kodaira at $p=n$, Akizuki-Nakano for general $p$, Nakano-positive bundle Kodaira-Nakano, Demailly-Skoda for weaker positivity — is a refinement of the same pointwise calculation, and the bridge is the curvature commutator with the Lefschetz operator. The pattern generalises through the Kawamata-Viehweg / Nadel / Demailly-Peternell-Schneider hierarchy to nef-and-big and pseudoeffective settings, but only at the canonical-twist corner $p=n$; the full bidegree statement requires strict ampleness, which appears again in 04.09.11 Kodaira embedding theorem as the input that secures sufficient global sections of high tensor powers to embed $X$ into projective space.

Full proof set [Master]

The proof of Akizuki-Nakano is given in detail in §3 above (Steps 1-5). For completeness, the supporting propositions are:

Proposition 1 (Bochner-Kodaira-Nakano identity). On a compact Kähler manifold $X$ with Kähler form $\omega$, and a Hermitian holomorphic line bundle $(L,h)$ with Chern connection $D_L=D_L^{\prime }+{\bar{\partial }}_L$ and curvature ${\Theta }_h(L)$,

$${\Delta }_{{\bar{\partial }}_L}-{\Delta }_{D_L^{\prime }}=[i{\Theta }_h(L),{\Lambda }_{\omega }]$$

on $A^{p,q}(X,L)$ for every $(p,q)$.

Proof. From the twisted Kähler identities $[{\bar{\partial }}_L,{\Lambda }_{\omega }]=-i(D_L^{\prime }{)}^{\ast }$ and $[D_L^{\prime },{\Lambda }_{\omega }]=i{\bar{\partial }}_L^{\ast }$ (proved by extending Hodge's original Kähler identity to the bundle-valued setting via parallel transport of frames), and the curvature identity ${\bar{\partial }}_L{D}_L^{\prime }+D_L^{\prime }{\bar{\partial }}_L={\Theta }_h(L)$ on $L$-valued forms, one computes

$${\Delta }_{{\bar{\partial }}_L}={\bar{\partial }}_L{\bar{\partial }}_L^{\ast }+{\bar{\partial }}_L^{\ast }{\bar{\partial }}_L=-i{\bar{\partial }}_L[D_L^{\prime },{\Lambda }_{\omega }]-i[D_L^{\prime },{\Lambda }_{\omega }]{\bar{\partial }}_L,$$

and similarly for ${\Delta }_{D_L^{\prime }}$. The difference simplifies through the Jacobi identity for the bracket and the curvature relation, yielding $[i{\Theta }_h(L),{\Lambda }_{\omega }]$. Detailed computation in Demailly Complex Analytic and Differential Geometry §VII.1, Wells Differential Analysis on Complex Manifolds §V.4, or Voisin Vol I §6.3. $\square$

Proposition 2 (Pointwise positivity). Let ${\Theta }_h(L)=i{\sum }_j{\gamma }_{j}d{z}_j\wedge d{\bar{z}}_j$ in a unitary frame diagonalising ${\Theta }_h(L)$ with respect to $\omega =i{\sum }_{j}d{z}_j\wedge d{\bar{z}}_j$, with ${\gamma }_j>0$. For $\alpha ={\sum }_{I,J}{\alpha }_{IJ}d{z}_I\wedge d{\bar{z}}_J$ a $(p,q)$-form with values in $L$,

$$⟨[i{\Theta }_h(L),{\Lambda }_{\omega }]\alpha ,\alpha ⟩=\sum _{I,J}\left(\sum _{j\in I}{\gamma }_j+\sum _{j\in J}{\gamma }_j-\sum _{j=1}^n{\gamma }_j\right)|{\alpha }_{IJ}{|}^2.$$

If $p+q>n$, this is strictly positive on every non-zero component.

Proof. The operator $L_{\omega }=\omega \wedge \cdot$ acts on $A^{p,q}$ by $L_{\omega }(d{z}_I\wedge d{\bar{z}}_J)=i{\sum }_j(d{z}_j\wedge d{\bar{z}}_j)\wedge d{z}_I\wedge d{\bar{z}}_J$. Its formal adjoint ${\Lambda }_{\omega }$ contracts with ${\sum }_j\frac{\partial }{\partial z_j}\wedge \frac{\partial }{\partial {\bar{z}}_j}$ (interior product). On $L_{\omega }$ replaced by $i{\Theta }_h(L)=i{\sum }_j{\gamma }_{j}d{z}_j\wedge d{\bar{z}}_j$, the commutator with ${\Lambda }_{\omega }$ in the diagonalising frame yields, on $d{z}_I\wedge d{\bar{z}}_J$, the eigenvalue ${\sum }_{j\in I}{\gamma }_j+{\sum }_{j\in J}{\gamma }_j-{\sum }_{j=1}^n{\gamma }_j=\gamma (I)+\gamma (J)-\gamma ([n])$ (Demailly §VII.5). For $p+q>n$, $|I|+|J|>n$, so $I\cap J\ne \varnothing$ and hence $\gamma (I)+\gamma (J)>\gamma (I\cup J)$, while $\gamma ([n]\setminus (I\cup J))<\gamma ([n])-\gamma (I\cup J)$. The net positivity is at least ${\min }_j{\gamma }_j>0$, and strictly positive on every non-zero ${\alpha }_{IJ}$. $\square$

Proposition 3 (Conclusion from positivity). Under the hypotheses of the Akizuki-Nakano theorem, every ${\bar{\partial }}_L$-harmonic $(p,q)$-form $\alpha$ with values in $L$ vanishes when $p+q>n$.

Proof. Take inner product of the Bochner-Kodaira-Nakano identity against $\alpha$. By harmonicity, $⟨{\Delta }_{{\bar{\partial }}_L}\alpha ,\alpha ⟩=0$. By Proposition 2 and integration, $⟨[i{\Theta }_h(L),{\Lambda }_{\omega }]\alpha ,\alpha ⟩\ge c\Vert \alpha {\Vert }^2$ for some $c>0$ depending on the smallest eigenvalue of ${\Theta }_h(L)$. Combined: $-(\Vert D_L^{\prime }\alpha {\Vert }^2+\Vert D_L^{\prime \ast }\alpha {\Vert }^2)\ge c\Vert \alpha {\Vert }^2$, forcing both sides to vanish, hence $\alpha =0$. $\square$

Proposition 4 (Comparison with Kodaira). Setting $p=n$ in Akizuki-Nakano recovers Kodaira vanishing $H^q(X,{\omega }_X\otimes L)=0$ for $q>0$.

Proof. Direct: ${\Omega }_X^n={\omega }_X$ and the Akizuki-Nakano condition $p+q>n$ becomes $q>0$. $\square$

The Akizuki-Nakano theorem thus packages the four propositions: identity, pointwise positivity, conclusion, and Kodaira recovery. Each is local-to-global in the spirit of the harmonic-form method.

Connections [Master]

• Kodaira vanishing theorem 04.09.02. The $p=n$ specialisation of Akizuki-Nakano. The Bochner-Kodaira-Nakano formula and the harmonic-form method are shared between the two proofs; Akizuki-Nakano simply tracks the curvature commutator across all $(p,q)$ bidegrees rather than restricting to $(n,q)$. The historical sequence reverses the logical: Kodaira's 1953 paper appeared first, Akizuki-Nakano 1954 generalised within a year.

• Hodge decomposition 04.09.01. The Dolbeault complex $(A^{p,\ast }(X,L),{\bar{\partial }}_L)$ and the identification of its cohomology with harmonic forms is the Hodge-theoretic infrastructure underlying both Kodaira and Akizuki-Nakano vanishing. The bidegree square that Akizuki-Nakano analyses is precisely the Hodge bidegree decomposition $H^k(X;\mathbb{C})={⨁}_{p+q=k}{H}^{p,q}(X)$, twisted by $L$.

• Ample line bundle 04.05.05. The hypothesis. By the Kodaira embedding theorem characterisation, ampleness of $L$ is equivalent to the existence of a Hermitian metric $h_L$ with positive Chern curvature. Without ampleness — or its weaker analytic incarnation, positive curvature on a singular metric — the Bochner-Kodaira-Nakano argument fails.

• $\partial \bar{\partial }$-lemma 04.09.05. The non-Kähler twin of Akizuki-Nakano. On Kähler manifolds the $\partial \bar{\partial }$-lemma holds, supporting the Bochner-Kodaira-Nakano calculus; on $\partial \bar{\partial }$-manifolds (a strictly larger class) some but not all of the vanishing extends, with controlled torsion corrections. Bogomolov-Sommese vanishing for logarithmic differentials replaces strict Akizuki-Nakano in the broader $\partial \bar{\partial }$-setting.

• Kodaira embedding theorem 04.09.11. The principal consumer of Akizuki-Nakano vanishing. To embed a compact Kähler $X$ into projective space using sections of $L^{\otimes k}$ for large $k$, one needs $H^0(X,L^{\otimes k})$ to separate points and tangent vectors. The proof uses both Kodaira vanishing (to compute Euler characteristics) and Akizuki-Nakano vanishing (to control the higher cohomology of ${\Omega }_X^1\otimes L^{\otimes k}$, which encodes the tangent-separation condition).

• Lefschetz hyperplane theorem 04.09.04 pending. Derived from Akizuki-Nakano via the conormal short exact sequence and the long exact sequence of cohomology, the Lefschetz hyperplane theorem says that the cohomology of a smooth ample divisor $Y\subset X$ matches that of $X$ in low degree. Exercise 3 above carries out the derivation.

• Hard Lefschetz theorem 04.09.07. The Lefschetz ${\mathfrak{sl}}_2$-action that appears in the pointwise positivity computation of Step 4 (the operator $L_{\omega }$ and its adjoint ${\Lambda }_{\omega }$, the bidegree counter $H$) is exactly the action that hard Lefschetz exploits to prove the bijectivity of $L_{\omega }^k:H^{n-k}\to H^{n+k}$ on a compact Kähler manifold. Both theorems sit on the same algebraic skeleton — the Kähler ${\mathfrak{sl}}_2$-triple — applied to different curvature inputs.

• Hodge-Riemann bilinear relations 04.09.08. Akizuki-Nakano vanishing and the Hodge-Riemann relations are two faces of the same Bochner-Kodaira-Nakano positivity. The HR2 positivity on primitive cohomology is the cohomology-level version of the pointwise curvature positivity that powers Akizuki-Nakano; both invoke the operator $[i\Theta (L),{\Lambda }_{\omega }]$ in bidegree $p+q>n$ as the load-bearing input, with Akizuki-Nakano reading off vanishing and Hodge-Riemann reading off the polarisation sign.

• Lefschetz (1,1)-theorem 04.09.09. The $p=1$ case of Akizuki-Nakano ($H^q(X,{\Omega }_X^1\otimes L)=0$ for $q>n-1$ and $L$ ample) controls the tangent-separation step that promotes a (1,1)-class to an actual line bundle producing a projective embedding. The Lefschetz (1,1)-theorem characterises which integral $H^2$-classes are first Chern classes of line bundles; Akizuki-Nakano supplies the cohomology-vanishing input that, combined with the exponential sequence, makes the analytic-to-algebraic identification work.

Historical & philosophical context [Master]

Yasuo Akizuki and Shigeo Nakano announced their generalisation of Kodaira vanishing in 1954 in Proc. Japan Acad. 30, 266-272 ^{[Akizuki1954]}, a short note titled Note on Kodaira-Spencer's proof of Lefschetz theorems. The work was a direct response to Kodaira and Spencer's 1953 program on deformation theory, where the Lefschetz hyperplane theorem played a central role and required precisely the bidegree refinement of Kodaira vanishing that Akizuki and Nakano supplied. Nakano followed up in 1955 with the longer paper On complex analytic vector bundles in J. Math. Soc. Japan 7, 1-12 ^[Nakano1955], developing the Bochner-Kodaira-Nakano formula for arbitrary Hermitian holomorphic vector bundles and proving the corresponding vanishing theorem for Nakano-positive bundles.

The method was transcendental in the same sense as Kodaira's: differential geometry, harmonic forms, the Hodge theorem, the Bochner-Kodaira-Nakano identity. Akizuki and Nakano made the Bochner technique part of the standard toolkit of complex algebraic geometry; their note crystallised the principle that positive curvature controls cohomology in the upper-right of the Hodge bidegree square, a principle Demailly later promoted to a programmatic framework in his Complex Analytic and Differential Geometry (online manuscript) ^{[DemaillyCADG]}.

The Akizuki-Nakano theorem entered the international mathematical mainstream through Wells's Differential Analysis on Complex Manifolds (Springer 1973, second edition 1980), Griffiths-Harris Principles of Algebraic Geometry (Wiley 1978), and Voisin Hodge Theory and Complex Algebraic Geometry (Cambridge 2002, §7.3) ^[Voisin2002]. By the time of Lazarsfeld's Positivity in Algebraic Geometry (Springer Ergebnisse 48-49, 2004), Akizuki-Nakano had been absorbed into the standard cohomological-positivity dictionary alongside Kodaira, Kawamata-Viehweg, and Nadel.

Algebraic refinements proceeded in two waves. The first, due to Kawamata 1982 (Math. Ann. 261, 43-46) ^{[Kawamata1982]} and Viehweg 1982 (J. reine angew. Math. 335, 1-8) ^{[Viehweg1982]}, replaced ampleness by nef and big but at the cost of restricting to the canonical-twist corner $p=n$. Kawamata-Viehweg vanishing became the cohomological backbone of Mori's minimal model program (Mori 1982 Annals of Math. 116; Birkar-Cascini-Hacon-McKernan 2010 J. Amer. Math. Soc. 23) ^[BCHM2010], with the proof iterating through extremal contractions controlled by Kawamata-Viehweg.

The second wave, due to Nadel 1990 (Annals of Math. 132), Demailly (multiple papers 1989-2001), and Demailly-Peternell-Schneider 2001 (Internat. J. Math. 12, 689-741) ^[DPS2001], introduced singular Hermitian metrics and multiplier ideal sheaves. The Demailly-Peternell-Schneider framework allows pseudoeffective line bundles (a class larger than nef-and-big) and corrects for the singularities of the metric by tensoring with the multiplier ideal $\mathcal{J}(h)$. Akizuki-Nakano-type vanishing in this setting is the cornerstone of the analytic minimal model program developed by Siu, Paun, Boucksom, Demailly, and others (2000-present).

Esnault-Viehweg's Lectures on Vanishing Theorems (DMV Seminar 20, Birkhäuser 1992) ^{[EsnaultViehweg1992]} gave a purely algebraic proof of Kawamata-Viehweg using Deligne-Illusie's Hodge-degeneration argument (Deligne-Illusie 1987 Invent. Math. 89, 247-270): lift the variety and the line bundle to characteristic $p$, apply the Frobenius decomposition of the de Rham complex, deduce vanishing on the lift, then descend to characteristic zero. The algebraic proof handles the bidegree corner ($p=n$) and certain mixed-Hodge-structure generalisations, but the full Akizuki-Nakano statement across all bidegrees remains most naturally proved transcendentally — a small but persistent reminder of the Kähler-analytic origins of the result.

Counter-examples in positive characteristic for Akizuki-Nakano follow from Raynaud's 1978 counter-example to Kodaira (Raynaud, in C. P. Ramanujam: A Tribute, Tata 1978, 273-278). The non-Kähler counter-example on the Hopf surface (Kodaira 1964 Amer. J. Math. 86) shows that the Kähler hypothesis is essential. Bogomolov-Sommese vanishing (Bogomolov 1979 Math. USSR Izv. 13; Sommese 1976-78) provides a partial Akizuki-Nakano analogue for logarithmic differentials on smooth pairs $(X,D)$ with normal-crossings boundary, opening the door to log Hodge theory and the boundedness of varieties of general type.

The Akizuki-Nakano theorem and its descendants now form the cohomological engine of modern algebraic geometry. Vanishing of $H^q(X,{\Omega }_X^p\otimes L)$ in the upper-right Hodge triangle controls infinitesimal deformations (Kodaira-Spencer), projective embeddings (Kodaira embedding), Hilbert polynomials and Hilbert-Mumford GIT-stability, period maps and variations of Hodge structure, and the entire framework of the minimal model program. Akizuki and Nakano's 1954 note, two pages of Proc. Japan Acad., is one of the most consequential short papers in the subject.

Bibliography [Master]

@article{Akizuki1954,
  author = {Akizuki, Yasuo and Nakano, Shigeo},
  title = {Note on Kodaira-Spencer's proof of Lefschetz theorems},
  journal = {Proc. Japan Acad.},
  volume = {30},
  year = {1954},
  pages = {266--272},
}

@article{Nakano1955,
  author = {Nakano, Shigeo},
  title = {On complex analytic vector bundles},
  journal = {J. Math. Soc. Japan},
  volume = {7},
  year = {1955},
  pages = {1--12},
}

@article{Kodaira1953,
  author = {Kodaira, Kunihiko},
  title = {On a differential-geometric method in the theory of analytic stacks},
  journal = {Proc. Nat. Acad. Sci. USA},
  volume = {39},
  year = {1953},
  pages = {1268--1273},
}

@book{Voisin2002,
  author = {Voisin, Claire},
  title = {Hodge Theory and Complex Algebraic Geometry I},
  publisher = {Cambridge University Press},
  year = {2002},
}

@misc{DemaillyCADG,
  author = {Demailly, Jean-Pierre},
  title = {Complex Analytic and Differential Geometry},
  note = {online manuscript, https://www-fourier.ujf-grenoble.fr/~demailly/},
  year = {1997--present},
}

@book{Wells1980,
  author = {Wells, Raymond O.},
  title = {Differential Analysis on Complex Manifolds},
  publisher = {Springer},
  edition = {2nd},
  year = {1980},
}

@book{GriffithsHarris1978,
  author = {Griffiths, Phillip and Harris, Joseph},
  title = {Principles of Algebraic Geometry},
  publisher = {Wiley},
  year = {1978},
}

@article{Kawamata1982,
  author = {Kawamata, Yujiro},
  title = {A generalization of {K}odaira-{R}amanujam's vanishing theorem},
  journal = {Math. Ann.},
  volume = {261},
  year = {1982},
  pages = {43--46},
}

@article{Viehweg1982,
  author = {Viehweg, Eckart},
  title = {Vanishing theorems},
  journal = {J. reine angew. Math.},
  volume = {335},
  year = {1982},
  pages = {1--8},
}

@article{Nadel1990,
  author = {Nadel, Alan},
  title = {Multiplier ideal sheaves and {K}\"ahler-{E}instein metrics of positive scalar curvature},
  journal = {Annals of Math.},
  volume = {132},
  year = {1990},
  pages = {549--596},
}

@article{DPS2001,
  author = {Demailly, Jean-Pierre and Peternell, Thomas and Schneider, Michael},
  title = {Pseudo-effective line bundles on compact {K}\"ahler manifolds},
  journal = {Internat. J. Math.},
  volume = {12},
  year = {2001},
  pages = {689--741},
}

@book{EsnaultViehweg1992,
  author = {Esnault, H\'el\`ene and Viehweg, Eckart},
  title = {Lectures on Vanishing Theorems},
  series = {DMV Seminar},
  volume = {20},
  publisher = {Birkh\"auser},
  year = {1992},
}

@article{DeligneIllusie1987,
  author = {Deligne, Pierre and Illusie, Luc},
  title = {Rel\`evements modulo $p^2$ et d\'ecomposition du complexe de de {R}ham},
  journal = {Invent. Math.},
  volume = {89},
  year = {1987},
  pages = {247--270},
}

@incollection{Raynaud1978,
  author = {Raynaud, Michel},
  title = {Contre-exemple au "vanishing theorem" en caract\'eristique $p > 0$},
  booktitle = {C. P. Ramanujam: A Tribute},
  publisher = {Tata Institute / Springer},
  year = {1978},
  pages = {273--278},
}

@article{Kollar1986,
  author = {Koll\'ar, J\'anos},
  title = {Higher direct images of dualizing sheaves I},
  journal = {Annals of Math.},
  volume = {123},
  year = {1986},
  pages = {11--42},
}

@article{BCHM2010,
  author = {Birkar, Caucher and Cascini, Paolo and Hacon, Christopher and McKernan, James},
  title = {Existence of minimal models for varieties of log general type},
  journal = {J. Amer. Math. Soc.},
  volume = {23},
  year = {2010},
  pages = {405--468},
}

@book{Lazarsfeld2004,
  author = {Lazarsfeld, Robert},
  title = {Positivity in Algebraic Geometry I, II},
  series = {Ergebnisse der Mathematik},
  volume = {48--49},
  publisher = {Springer},
  year = {2004},
}