Where we are on the ladder
You arrive here fluent in smooth manifolds: an atlas of charts into R^n whose overlaps are smooth, a tangent bundle TM, differential forms, and the Lie bracket [X, Y] of vector fields. This rung asks a sharper question. What if we demand the charts land not in R^n but in C^n, and the overlaps be not merely smooth but holomorphic — complex-differentiable, satisfying the Cauchy-Riemann equations? That single upgrade is the difference between the smooth world and the complex world, and it changes almost everything downstream: cohomology splits, metrics acquire a hidden symmetry, and rigidity appears where smoothness was floppy.
An honesty note before we start. These are graduate topics, and we assume you carry comfort with linear algebra, multivariable calculus, point-set topology, and the smooth-manifold survey from Volume I. Words like manifold, diffeomorphism, tangent bundle, and Lie bracket are used here, not re-derived — when a Volume I object appears we link back rather than re-teach. We will also lean lightly on one complex-analysis fact: a function of several complex variables is holomorphic exactly when it is complex-differentiable in each variable, equivalently when it satisfies the Cauchy-Riemann equations. If that feels shaky, a one-variable refresher is the place to start.
Two definitions of 'complex', and the gap between them
There are two genuinely different things one can mean by putting a complex structure on a manifold, and the whole guide is about the gap between them. The first is global and demanding: a complex manifold is one with an atlas of charts into C^n whose transition maps are holomorphic. The second is pointwise and cheap: an almost-complex structure is a smooth bundle map J: TM -> TM with J^2 = −1, a fibrewise 'multiplication by i' on each tangent space T_p M, with no compatibility between points required. Every complex manifold has an obvious J (multiply tangent vectors by i in any holomorphic chart), but the converse — does a given J come from holomorphic charts? — is the deep question.
Why does J^2 = −1 capture 'multiplication by i' at all? Think of the simplest case, R^2 modeling C with coordinate z = x + i y. Multiplying a complex number by i rotates it ninety degrees: 1 goes to i, i goes to −1. In real terms the basis vector along x maps to the one along y, and the one along y maps to minus the one along x. As a matrix that is J = [0, -1; 1, 0], and squaring it gives −1, the identity with a minus sign. So an almost-complex structure is exactly a smoothly-varying recipe for 'rotate this tangent space by ninety degrees the way i would' — and the catch is that nothing yet forces those rotations to fit together into honest complex coordinates.
Complexifying the tangent space: the (1,0) / (0,1) split
To analyze J we play the standard linear-algebra trick: allow complex coefficients. Tensor the real tangent space with C to get the complexified tangent space, on which J is now a genuine complex-linear operator with J^2 = −1. An operator squaring to −1 has eigenvalues +i and −i, and since J is real it splits the complexified space cleanly into the +i-eigenspace, written T^(1,0), and the −i-eigenspace, written T^(0,1). The first holds the holomorphic directions (informally, the d/dz directions), the second the antiholomorphic ones (the d/dz-bar directions). This eigenspace split is the engine of the entire subject; the (p,q)-type decomposition of forms in the next guide is just its dual, applied to wedge powers of the cotangent space.
On C^1 with z = x + i y : d/dz = (1/2)( d/dx - i d/dy ) spans T^(1,0) d/dz-bar = (1/2)( d/dx + i d/dy ) spans T^(0,1) J(d/dz) = +i d/dz , J(d/dz-bar) = -i d/dz-bar A function f is holomorphic <=> d f / d z-bar = 0 (Cauchy-Riemann)
Here is the payoff of the split for naming things honestly. If J really does come from holomorphic charts, then in those charts the coordinate vector fields d/dz^1, ..., d/dz^n are sections of T^(1,0) and they commute: their Lie brackets vanish, since partial derivatives commute. So the bundle T^(1,0) is closed under the Lie bracket — a bracket of two holomorphic-type fields is again holomorphic-type. This closure is exactly the integrability condition we are about to test, and it is the bridge from the pointwise gadget J to the global existence of holomorphic coordinates.
The Nijenhuis tensor: the single obstruction
The previous paragraph hands us a concrete test: take two sections X, Y of T^(1,0) and ask whether their bracket [X, Y] is still a section of T^(1,0), or whether some forbidden (0,1)-part leaks out. The amount that leaks out, packaged so it does not depend on how we extend X and Y to vector fields, is the Nijenhuis tensor N_J. It is built purely from J and the Lie bracket by the formula below, and the miracle is that despite involving derivatives of vector fields, N_J(X, Y) at a point depends only on X and Y at that point — it is an honest tensor, not a differential operator. So the Nijenhuis tensor is a pointwise field that measures, at each point, exactly how far J is from being integrable.
N_J(X, Y) = [X, Y] + J[JX, Y] + J[X, JY] - [JX, JY]
N_J = 0 identically
<=> T^(1,0) is closed under the Lie bracket
<=> J is integrable
<=> (Newlander-Nirenberg) J comes from holomorphic chartsTwo sanity checks make the tensor friendlier. First, N_J is antisymmetric, N_J(X, Y) = −N_J(Y, X), so on a real 2-dimensional surface — where TM has only one independent direction once you fix it — there is no room for a nonzero antisymmetric expression, and N_J vanishes automatically. That is the structural reason every almost-complex structure on a real surface is integrable: every oriented surface with a metric is secretly a Riemann surface, a fact we will revisit in the curves track. Second, on any complex manifold N_J = 0 by construction, since the d/dz-coordinate fields commute. The content of the theorem is the reverse direction, in dimension four and higher.
Newlander-Nirenberg: integrability is exactly N_J = 0
Now the theorem. The Newlander-Nirenberg theorem says: an almost-complex structure J on a smooth manifold comes from a (necessarily unique) complex-manifold atlas if and only if its Nijenhuis tensor vanishes identically. In words, the pointwise gadget J integrates to honest holomorphic coordinates exactly when the one tensor obstruction N_J is zero. When that holds we call J integrable, and the manifold is genuinely complex. This is the precise dictionary between the cheap pointwise data and the expensive global data — it tells you the only thing you must check.
It is worth being honest about the two halves. One direction is elementary algebra: if J is complex (comes from charts) then N_J = 0, because partial derivatives commute, as we saw. The hard, beautiful direction is the converse — manufacturing holomorphic coordinates out of nothing but the vanishing of N_J. In the real-analytic case this follows from the classical Frobenius theorem applied to the involutive distribution T^(1,0), much as Frobenius integrability turns a bracket-closed distribution into a foliation by submanifolds. But for merely smooth J the involutive distribution lives in the complexified bundle, the real Frobenius theorem does not directly apply, and the proof needs serious linear PDE — solving a Cauchy-Riemann-type system with careful estimates. That analytic depth is the whole reason the theorem carries two names and a date.
Why this gate matters, and what comes next
Once N_J = 0 and the manifold is genuinely complex, everything in this rung switches on. The eigenspace split T^(1,0) plus T^(0,1) dualizes to a bigraded decomposition of forms, the (p,q)-forms of guide 2, and the exterior derivative d splits into two pieces, the Dolbeault operators del and del-bar, each squaring to zero. That gives Dolbeault cohomology, the holomorphic refinement of de Rham cohomology, which is what makes complex geometry compute. None of that machinery is even well-defined until the integrability gate is passed — del-bar squares to zero precisely because N_J = 0.
Keep two honest cautions in view as you climb. First, having a complex structure is strictly weaker than having a good metric compatible with it: a complex manifold need not be Kähler, and the Kähler condition (guides 3 and 4) is an extra, genuinely restrictive closedness requirement on the associated 2-form. Do not let the words blur together. Second, the existence question is delicate and often open — recall that we do not know whether S^6 is a complex manifold despite it carrying almost-complex structures. The Newlander-Nirenberg theorem tells you the test, N_J = 0, but it does not promise that any given manifold admits a J passing that test.
A clean way to hold this guide in memory: the complex world is the smooth world plus one tensor equation. Smoothness gives you J, a fibrewise i; integrability — the vanishing of the single tensor N_J — promotes that fibrewise i into genuine holomorphic coordinates, and from there the holomorphic line bundles, Hermitian metrics, and eventually the Calabi-Yau geometry of guide 5 all unfold. One equation, N_J = 0, is the doorway to the rest of the rung.