03.02.09 · differential-geometry / manifolds

Almost-complex structure (manifold-level)

shipped3 tiersLean: none

Anchor (Master): Ehresmann 1947; Newlander-Nirenberg 1957 Ann. Math. 65; Kobayashi-Nomizu Vol. 2 Ch. 9

Intuition [Beginner]

An almost-complex structure is a rule that lets you multiply tangent vectors by $i$ , the imaginary unit, at every point of a manifold. Think of it as a machine $J$ that rotates each tangent vector by 90 degrees. Applying $J$ twice sends $v$ to $J (J (v)) = - v$ , just like $i \times i = - 1$ .

This is purely a tangent-space construction. It does not require the manifold itself to be a complex manifold, but it makes every tangent space look like a complex vector space. The dimension of the manifold must be even, because a complex vector space of complex dimension $n$ has real dimension $2 n$ .

Not every even-dimensional manifold admits an almost-complex structure. The sphere $S^{2}$ admits one (it is a complex manifold in disguise), but the sphere $S^{4}$ does not, even though it is 4-dimensional. The obstruction is topological, involving characteristic classes.

Visual [Beginner]

A 2-dimensional tangent plane at a point $p$ , with a vector $v$ and its image $J v$ shown as a perpendicular arrow. A second application gives $J (J v) = - v$ , pointing opposite to the original. The operator $J$ rotates every tangent vector by 90 degrees in the plane, giving each tangent space the structure of the complex numbers.

The map $J$ turns each tangent space into a complex vector space of complex dimension $n$ (real dimension $2 n$ ).

Worked example [Beginner]

The standard almost-complex structure on $R^{2}$ . Consider $R^{2}$ with coordinates $(x, y)$ . Define $J$ by its action on unit vectors: $J$ sends the $x$ -direction to the $y$ -direction, and sends the $y$ -direction to the negative $x$ -direction. In matrix form, $J = (01 - 1 0)$ .

Step 1. Apply $J$ to the vector $v = (3, 1)$ : $J v = (- 1, 3)$ . This is a 90-degree counterclockwise rotation.

Step 2. Apply $J$ again: $J (J v) = J (- 1, 3) = (- 3, - 1) = - v$ . So $J^{2} = - Id$ , confirming the almost-complex property.

Step 3. The complex number $3 + i$ corresponds to the real vector $(3, 1)$ , and multiplication by $i$ gives $i (3 + i) = - 1 + 3 i$ , corresponding to $(- 1, 3)$ . The map $J$ is multiplication by $i$ .

What this tells us: the standard $J$ on $R^{2}$ is exactly the complex structure of $C$ , viewed in real coordinates.

Check your understanding [Beginner]

Formal definition [Intermediate+]

Definition (Almost-complex structure). Let $M$ be a smooth manifold of dimension $2 n$ . An almost-complex structure on $M$ is a smooth bundle endomorphism $J : T M \to T M$ satisfying $J^{2} = - Id_{T M}$ . The pair $(M, J)$ is called an almost-complex manifold.

At each point $p$ , the tangent space $T_{p} M$ becomes a complex vector space via the scalar multiplication $(a + bi) \cdot v = a v + b J_{p} (v)$ . This makes $T_{p} M$ a complex vector space of dimension $n$ .

Definition (Nijenhuis tensor). The Nijenhuis tensor $N_{J}$ of an almost-complex structure $J$ is the $(1, 2)$ -tensor field defined by:

$N_{J} (X, Y) = [X, Y] + J [J X, Y] + J [X, J Y] - [J X, J Y]$

for vector fields $X, Y$ . The Nijenhuis tensor measures the failure of $J$ to be integrable: if $N_{J} = 0$ , the almost-complex structure comes from a genuine complex structure on $M$ .

Definition (Integrability). An almost-complex structure $J$ is integrable if every point has a neighbourhood with holomorphic coordinates, i.e., coordinates $(z^{1}, \dots, z^{n})$ with $z^{k} = x^{k} + i y^{k}$ such that $J (\partial / \partial x^{k}) = \partial / \partial y^{k}$ and $J (\partial / \partial y^{k}) = - \partial / \partial x^{k}$ .

Counterexamples to common slips

An almost-complex structure is not the same as a complex structure. The sphere $S^{6}$ has an almost-complex structure but may not be a complex manifold. An almost-complex structure is a pointwise condition on $J$ ; a complex structure requires compatible holomorphic charts.
Having $J^{2} = - Id$ pointwise is not sufficient for integrability. The Nijenhuis tensor may be nonzero even though $J$ satisfies the pointwise condition at every tangent space. Integrability is a differential condition, not an algebraic one.
A symplectic form does not give an almost-complex structure directly. A compatible triple $(g, J, ω)$ requires choosing $J$ and $g$ that are compatible with the symplectic form $ω$ , via $g (X, Y) = ω (X, J Y)$ .

Key theorem with proof [Intermediate+]

Theorem (Nijenhuis integrability criterion). An almost-complex structure $J$ on a smooth manifold $M$ is integrable if and only if the Nijenhuis tensor $N_{J}$ vanishes identically.

Proof. ( $\Leftarrow$ ) If $J$ is integrable, choose local holomorphic coordinates $(z^{1}, \dots, z^{n})$ with $z^{k} = x^{k} + i y^{k}$ . In these coordinates, $J (\partial / \partial x^{k}) = \partial / \partial y^{k}$ and $J (\partial / \partial y^{k}) = - \partial / \partial x^{k}$ . The Lie bracket of two coordinate vector fields vanishes: $[\partial / \partial x^{k}, \partial / \partial x^{ℓ}] = 0$ and $[\partial / \partial x^{k}, \partial / \partial y^{ℓ}] = 0$ . Since $J$ is constant in these coordinates, each term in $N_{J}$ vanishes:

$N_{J} (\partial / \partial x^{k}, \partial / \partial x^{ℓ}) = 0 + J (0) + J (0) - 0 = 0.$

The same holds for all pairs of coordinate vectors, and by bilinearity $N_{J} = 0$ .

( $\Rightarrow$ ) If $N_{J} = 0$ , the Newlander-Nirenberg theorem guarantees the existence of holomorphic coordinates in which $J$ takes the standard form. This direction is the deep one: the vanishing of $N_{J}$ is a system of first-order PDE conditions on $J$ , and the Newlander-Nirenberg theorem proves these equations are solvable, producing local coordinates ${z^{k}}$ with $J$ in standard form. The proof proceeds by constructing a local frame of $(1, 0)$ -vector fields (eigenvectors of $J$ with eigenvalue $i$ ) and using the Frobenius theorem to integrate the distribution they span. $□$

Bridge. This theorem builds toward the Newlander-Nirenberg theorem in the Master section, where the deep direction ( $N_{J} = 0$ implies integrability) appears again as the central result. The foundational reason integrability is characterised by a tensor is that the Nijenhuis tensor captures the torsion of $J$ as a connection-like object, and this is exactly the obstruction to finding coordinates in which $J$ is constant. The bridge is between the algebraic condition $J^{2} = - Id$ (pointwise) and the differential condition $N_{J} = 0$ (neighbourhood), and the smooth structure 03.02.02 provides the coordinate framework in which this passage from pointwise to local is made rigorous, identifying the almost-complex structure with its integrable shadow.

Exercises [Intermediate+]

Exercise 4 (medium, symbolic).

On $R^{4}$ with coordinates $(x, y, u, v)$ , define an almost-complex structure by $J (\partial / \partial x) = \partial / \partial y$ , $J (\partial / \partial y) = - \partial / \partial x$ , $J (\partial / \partial u) = \partial / \partial v + x \partial / \partial y$ , $J (\partial / \partial v) = - \partial / \partial u + x \partial / \partial x$ . Compute $J^{2} (\partial / \partial u)$ and determine whether this is an almost-complex structure.

Hint

Compute $J (J (\partial / \partial u)) = J (\partial / \partial v + x \partial / \partial y)$ using linearity of $J$ and the product rule for the coefficient $x$ .

Answer

$J^{2} (\partial / \partial u) = J (\partial / \partial v + x \partial / \partial y) = J (\partial / \partial v) + x J (\partial / \partial y) + (d x \cdot J (\partial / \partial u)) (\partial / \partial y)$ ... Wait, $J$ must be applied as a bundle map, so $J (x \partial / \partial y) = x J (\partial / \partial y) = - x \partial / \partial x$ (since $J$ is linear over functions). So $J^{2} (\partial / \partial u) = J (\partial / \partial v) + x J (\partial / \partial y) = (- \partial / \partial u + x \partial / \partial x) + (- x \partial / \partial x) = - \partial / \partial u$ . This gives $J^{2} = - Id$ on $\partial / \partial u$ . A similar check on $\partial / \partial v$ confirms $J^{2} = - Id$ on all basis vectors, so this is indeed an almost-complex structure.

Exercise 7 (hard, symbolic).

Prove that if $J$ comes from a complex structure on $M$ (i.e., $M$ is a complex manifold with holomorphic coordinates and $J$ is the standard complex structure in those coordinates), then $N_{J} = 0$ without using specific coordinate computations. Use only the fact that $J$ is constant in holomorphic coordinates.

Hint

In holomorphic coordinates, $J$ has constant coefficients. Express $N_{J}$ in terms of Lie brackets and the fact that $J$ commutes with coordinate derivatives.

Answer

In holomorphic coordinates $(z^{1}, \dots, z^{n})$ , the almost-complex structure $J$ acts as $J (\partial / \partial x^{k}) = \partial / \partial y^{k}$ and $J (\partial / \partial y^{k}) = - \partial / \partial x^{k}$ . Since the coefficients of $J$ are constants in these coordinates, for any vector fields $X, Y$ expressed in the coordinate frame, $J X$ and $J Y$ have coefficients obtained by a constant linear transformation of the coefficients of $X$ and $Y$ . The Lie bracket $[X, Y]$ involves derivatives of the coefficient functions, and applying $J$ to a bracket amounts to multiplying the resulting coefficient vector by a constant matrix. Since the constant matrix commutes with taking derivatives, $J [X, Y]$ has the same form as the bracket of $J X$ and $J Y$ up to sign. Explicitly: $N_{J} (\partial_{i}, \partial_{j}) = [\partial_{i}, \partial_{j}] + J [J \partial_{i}, \partial_{j}] + J [\partial_{i}, J \partial_{j}] - [J \partial_{i}, J \partial_{j}] = 0$ because all Lie brackets of coordinate vector fields vanish. By $C^{\infty}$ -bilinearity, $N_{J} = 0$ .

Exercise 8 (hard, symbolic).

Let $M = S^{2} \times S^{2}$ with the product almost-complex structure (each factor carries the complex structure of $CP^{1}$ ). Show that the Nijenhuis tensor vanishes, and that $(M, J)$ is a complex manifold.

Hint

On a product manifold $M_{1} \times M_{2}$ , the almost-complex structure $J = J_{1} \oplus J_{2}$ acts as $J_{1}$ on $T M_{1}$ and $J_{2}$ on $T M_{2}$ . Compute $N_{J}$ in terms of $N_{J_{1}}$ and $N_{J_{2}}$ .

Answer

The tangent bundle of $M_{1} \times M_{2}$ splits as $T M = T M_{1} \oplus T M_{2}$ . The product almost-complex structure acts as $J_{1}$ on $T M_{1}$ and $J_{2}$ on $T M_{2}$ . For vector fields $X, Y$ both tangent to $M_{1}$ : $N_{J} (X, Y) = N_{J_{1}} (X, Y)$ (since $J$ agrees with $J_{1}$ on $T M_{1}$ and vectors tangent to $M_{1}$ have Lie bracket tangent to $M_{1}$ ). Similarly for $M_{2}$ . For $X$ tangent to $M_{1}$ and $Y$ tangent to $M_{2}$ : $[X, Y] = 0$ (coordinate functions on different factors commute), so all four terms in $N_{J}$ vanish. Since $N_{J_{1}} = 0$ and $N_{J_{2}} = 0$ (both factors are complex manifolds), $N_{J} = 0$ . Holomorphic coordinates on $M$ are obtained as products of holomorphic coordinates on each factor, so $M$ is a complex manifold biholomorphic to $CP^{1} \times CP^{1}$ .

Advanced results [Master]

Theorem 1 (Newlander-Nirenberg). If $(M, J)$ is an almost-complex manifold with vanishing Nijenhuis tensor $N_{J} = 0$ , then $J$ is integrable: every point has a neighbourhood with local holomorphic coordinates. This is the converse of the easy direction in the Nijenhuis integrability criterion.

Theorem 2 (Existence of almost-complex structures). An orientable manifold $M$ of real dimension $2 n$ admits an almost-complex structure if and only if there is a reduction of the structure group of $T M$ from $G L (2 n, R)$ to $G L (n, C)$ . In homotopy-theoretic terms, this is a lift of the classifying map $M \to B G L (2 n, R)$ through $B G L (n, C)$ .

Theorem 3 (Obstruction theory). The primary obstruction to the existence of an almost-complex structure on a $2 n$ -dimensional oriented manifold is the vanishing of certain characteristic classes. For a 4-manifold, the necessary condition is $w_{2} (M) = c_{1} (T M) mod 2$ and the Hirzebruch signature theorem constrains the Chern numbers. For $S^{4}$ , the obstruction is nonzero, so $S^{4}$ admits no almost-complex structure.

Theorem 4 (G-structures). An almost-complex structure on $M$ is equivalent to a $G L (n, C)$ -structure on the frame bundle. The integrability condition $N_{J} = 0$ is equivalent to the vanishing of the intrinsic torsion of this $G$ -structure. This viewpoint generalises: other geometric structures (symplectic, Riemannian, etc.) correspond to other reductions of the frame bundle, each with its own torsion obstruction.

Theorem 5 (Almost-complex structures on spheres). The only spheres admitting almost-complex structures are $S^{2}$ and $S^{6}$ . The structure on $S^{2}$ is integrable (it is $CP^{1}$ ). The structure on $S^{6}$ from octonion multiplication is not integrable. Whether $S^{6}$ admits any integrable complex structure is open.

Synthesis. The Nijenhuis tensor is the foundational reason that almost-complex structures split into integrable and non-integrable classes; the central insight is that a purely algebraic condition ( $J^{2} = - Id$ ) at each point does not determine the differential behaviour of $J$ , and the Nijenhuis tensor captures exactly the differential obstruction. This pattern generalises through the theory of $G$ -structures, where the intrinsic torsion of any reduction of the frame bundle plays the same role. Putting these together with the Newlander-Nirenberg theorem, an almost-complex structure with vanishing Nijenhuis tensor is exactly a complex structure, and this is the bridge between the real-smooth and complex-analytic categories. The smooth structure 03.02.02 provides the framework in which $J$ lives as a tensor field, and the Newlander-Nirenberg theorem identifies the vanishing of $N_{J}$ with the existence of holomorphic coordinates, building toward the full theory of complex manifolds.

Full proof set [Master]

Proposition (Dimension constraint). If $M$ admits an almost-complex structure, then $dim M$ is even.

Proof. At each point $p \in M$ , the map $J_{p} : T_{p} M \to T_{p} M$ satisfies $J_{p}^{2} = - Id$ . The determinant of both sides gives $(det J_{p})^{2} = det (- Id) = (- 1)^{d i m M}$ . Since $J_{p}$ is a real matrix, $(det J_{p})^{2} > 0$ . So $(- 1)^{d i m M} > 0$ , forcing $dim M$ to be even. $□$

Proposition (Nijenhuis tensor is a tensor). The expression $N_{J} (X, Y) = [X, Y] + J [J X, Y] + J [X, J Y] - [J X, J Y]$ is $C^{\infty} (M)$ -bilinear, and therefore defines a $(1, 2)$ -tensor field.

Proof. Check $C^{\infty}$ -linearity in $X$ . For $f \in C^{\infty} (M)$ :

$[f X, Y] = f [X, Y] - Y (f) X$ $J [f X, Y] = J (f [X, Y]) - Y (f) J X = f J [X, Y] - Y (f) J X$ $[f X, J Y] = f [X, J Y] - J Y (f) X$

and $J (f X) = f J X$ (since $J$ is $C^{\infty}$ -linear). Substituting into $N_{J} (f X, Y)$ :

$N_{J} (f X, Y) = f [X, Y] - Y (f) X + f J [J X, Y] - Y (f) J^{2} X + f J [X, J Y] - J Y (f) J X - f [J X, J Y] + J Y (f) J X .$

The terms involving derivatives of $f$ are: $- Y (f) X - Y (f) J^{2} X = - Y (f) (X + J^{2} X) = - Y (f) (X - X) = 0$ . The remaining terms give $f \cdot N_{J} (X, Y)$ . Bilinearity in $Y$ is analogous. $□$

Connections [Master]

Smooth maps between manifolds 03.02.03. An almost-complex structure $J$ is a smooth bundle map $T M \to T M$ , and smooth maps between almost-complex manifolds that respect $J$ (satisfying $df \circ J = J \circ df$ ) are the pseudo-holomorphic maps. The smooth-map framework 03.02.03 is what makes the Nijenhuis tensor well-defined as a smooth tensor field.
Smooth structures and atlases 03.02.02. The almost-complex structure $J$ lives on the tangent bundle of a smooth manifold. The smooth atlas 03.02.02 provides the coordinate charts in which $J$ is expressed as a matrix field, and integrability is the condition that there exist charts in which this matrix is constant.
Killing fields and infinitesimal isometries 03.02.07. On a Riemannian manifold with compatible almost-complex structure (a Kahler manifold), the Killing fields that also preserve $J$ are the holomorphic Killing fields. The interplay between the metric structure 03.02.07 and the complex structure constrains the symmetry group to a unitary subgroup.

Historical & philosophical context [Master]

Charles Ehresmann introduced almost-complex structures in 1947 ^{[Ehresmann 1947]} in the context of his work on fibre bundles and $G$ -structures. Ehresmann recognised that the condition $J^{2} = - Id$ on the tangent bundle is a natural geometric structure intermediate between a smooth structure and a complex structure.

The deep result that vanishing Nijenhuis tensor implies integrability is the Newlander-Nirenberg theorem of 1957 ^{[Newlander-Nirenberg 1957]}, proved by August Newlander and Louis Nirenberg in the Annals of Mathematics. Their proof used the Frobenius theorem in the complexified tangent bundle to construct holomorphic coordinates from the vanishing of the Nijenhuis tensor. Subsequent proofs by Nijenhuis and Woolf (1963), Hörmander (1965 via the $\overset{ˉ}{\partial}$ -Neumann problem), and others have given the theorem multiple independent proofs, each illuminating a different aspect of the geometry.

Bibliography [Master]

@inproceedings{ehresmann1947,
  author = {Ehresmann, Charles},
  title = {Sur la th\'{e}orie des espaces fibr\'{e}s},
  booktitle = {Colloque de topologie alg\'{e}brique, Paris},
  year = {1947}
}

@article{newlander-nirenberg1957,
  author = {Newlander, August and Nirenberg, Louis},
  title = {Complex analytic coordinates in almost complex manifolds},
  journal = {Ann. of Math.},
  volume = {65},
  pages = {391--404},
  year = {1957}
}

@book{kobayashi-nomizu-vol2,
  author = {Kobayashi, Shoshichi and Nomizu, Katsumi},
  title = {Foundations of Differential Geometry},
  volume = {2},
  publisher = {Wiley Interscience},
  year = {1969}
}

@book{lee-smooth,
  author = {Lee, John M.},
  title = {Introduction to Smooth Manifolds},
  edition = {2},
  publisher = {Springer},
  year = {2013}
}

Prerequisites

03.02.03

Tier anchors

beginner: Needham Visual Differential Geometry informal; multiplication by i in the plane
intermediate: Lee Introduction to Smooth Manifolds Ch. 18; Hirsch Differential Topology
master: Ehresmann 1947; Newlander-Nirenberg 1957 Ann. Math. 65; Kobayashi-Nomizu Vol. 2 Ch. 9

References

TODO_REF
Ehresmann — Sur la theorie des espaces fibres (1947) · Colloque de topologie algebrique, Paris
TODO_REF
Newlander-Nirenberg — Complex analytic coordinates in almost complex manifolds (1957) · Ann. of Math. 65, pp. 391--404
TODO_REF
Kobayashi-Nomizu — Foundations of Differential Geometry Vol. 2 · Ch. 9 (almost complex manifolds)
TODO_REF
Lee — Introduction to Smooth Manifolds · Ch. 18 (almost complex manifolds, Nijenhuis tensor)

Reviewer

TBD

Estimated time

beginner: 15m
intermediate: 40m
master: 75m