From Riemannian to Hermitian: a metric that respects J
Guide 1 of this rung gave us a complex manifold M with its integrable almost-complex structure J, and Guide 2 split forms by type and built the Dolbeault operators partial and dbar. So far there has been no notion of length or angle — only the holomorphic skeleton. Now we lay a metric on top, but not just any metric: we want one that is blind to the difference between a vector v and its rotated partner Jv. A Hermitian metric is a Riemannian metric g on M satisfying g(Jv, Jw) = g(v, w) for all tangent vectors. The condition says J acts as an isometry on every tangent space T_p M, so g cannot tell a real direction apart from the i-rotated one.
Every complex manifold admits Hermitian metrics — start with any Riemannian g_0 and average it against J via g(v, w) = (g_0(v, w) + g_0(Jv, Jw))/2; a partition of unity patches the local pieces together. So Hermitian-ness is cheap and imposes no constraint on which M you may carry. The interesting structure appears the moment you encode the metric not as a symmetric tensor but as a 2-form. Define the fundamental form omega by omega(v, w) = g(Jv, w). The Hermitian condition is exactly what makes omega antisymmetric — omega(w, v) = g(Jw, v) = g(J^2 w, Jv) = -g(w, Jv) = -omega(v, w) — so omega is a genuine 2-form, and it is of pure type (1,1) in the Dolbeault grading of Guide 2.
It pays to see all this in local holomorphic coordinates z^1, ..., z^n. The metric is recorded by a positive-definite Hermitian matrix of functions (h_jk), and the fundamental form reads omega = i sum_{j,k} h_jk dz^j ^ d zbar^k — a real (1,1)-form whose positivity is exactly the positive-definiteness of (h_jk). On flat C^n with h_jk = delta_jk this collapses to omega = (i/2) sum dz^j ^ d zbar^j, which is nothing but the standard symplectic form of Volume I's symplectic geometry. So even before any closedness condition, the fundamental form of a Hermitian metric already looks symplectic pointwise; the question of the next section is whether it is symplectic globally.
The Kähler condition: one closed form changes everything
Here is the single hypothesis that turns an unruly Hermitian manifold into a paradise. A Hermitian manifold is Kähler when its fundamental form is closed: d omega = 0. That is the entire definition — three innocent letters — yet it is the deepest demand in this rung. Closedness means omega is not merely a metric gadget but a genuine symplectic form, so a Kähler manifold is simultaneously complex (it has J), Riemannian (it has g), and symplectic (it has the closed omega), with all three woven by the one identity omega(v, w) = g(Jv, w). It is the place where the three great geometries of this volume coincide rather than merely coexist.
Why is d omega = 0 so consequential? Because it is secretly equivalent to a statement about the connection. The metric g has its Levi-Civita connection nabla from Volume I, and the complex structure J is a tensor field that nabla can differentiate. The clean theorem: a Hermitian manifold is Kähler if and only if nabla J = 0 — the complex structure is PARALLEL. Equivalently, the Chern connection (the canonical connection compatible with both metric and holomorphic structure) coincides with the Levi-Civita connection. On a general Hermitian manifold these two connections differ by torsion; the Kähler condition is exactly the vanishing of that torsion. So 'd omega = 0' is the same as 'parallel transport commutes with J', which is why curvature behaves so well below.
Concrete Kähler manifolds — and ones that are not
Start small. Flat C^n with omega = (i/2) sum dz^j ^ d zbar^j is Kähler trivially — omega has constant coefficients, so d omega = 0. The complex torus C^n / Lambda inherits this flat omega and is therefore a compact Kähler manifold; with n = 1 it is an elliptic curve. The example that does all the real work is complex projective space CP^n with the Fubini-Study metric. Its fundamental form is omega_FS = (i/2 pi) partial dbar log(|z_0|^2 + ... + |z_n|^2) in homogeneous coordinates, and it is closed for a structural reason worth savouring: it is locally i partial dbar of a function. Since d = partial + dbar and both partial^2 = 0 and dbar^2 = 0 and partial dbar = -dbar partial, applying d to i partial dbar(anything) gives zero automatically.
This i-partial-dbar trick is the engine of Kähler geometry. A real (1,1)-form of the shape i partial dbar phi for a smooth real function phi is automatically closed, and such a phi is called a Kähler potential. On CP^n the function log(sum |z_j|^2) is the potential; the Fubini-Study form descends from it. Now the inheritance principle: ANY complex submanifold of a Kähler manifold is again Kähler, because restricting a closed form keeps it closed and restricting a positive (1,1)-form keeps it positive. Combined with CP^n being Kähler, this hands you an enormous supply — every smooth projective variety, every elliptic curve embedded as a cubic, every nonsingular hypersurface in projective space, is a compact Kähler manifold. This is why Kähler geometry is the natural home of algebraic geometry.
The Kähler identities: bracket relations that bind the operators
Now we reach the technical heart. On any Hermitian manifold three first-order operators act on forms: d (real, type-raising by one), partial and dbar (the Dolbeault operators, raising the holomorphic and antiholomorphic degrees). The metric supplies their formal adjoints d*, partial*, dbar* (each lowers degree), and one more algebraic operator: the Lefschetz operator L, which wedges with omega, L(alpha) = omega ^ alpha, together with its adjoint Lambda = L* (contraction with omega). On a Hermitian manifold these operators satisfy no clean relations among themselves. The miracle of the Kähler condition is that, once d omega = 0, they lock into a tight web of commutator identities.
The KAHLER IDENTITIES (valid because d omega = 0; [ , ] is the graded commutator):
[ Lambda , dbar ] = - i partial*
[ Lambda , partial ] = i dbar*
[ L , partial* ] = i dbar
[ L , dbar* ] = - i partial
Consequences for the three Laplacians Delta_d = dd* + d*d , etc.:
Delta_partial = Delta_dbar (the two complex Laplacians agree)
Delta_d = 2 Delta_partial = 2 Delta_dbar (the real Laplacian is twice them)
Delta_d commutes with L , Lambda , partial , dbar , partial* , dbar* , and the projectors to (p,q)How does one PROVE these? Honestly: not by a slick one-liner. The cleanest route checks the first identity on flat C^n by a direct (if tedious) computation in coordinates, where omega has constant coefficients and the operators are explicit. Then the Kähler normal coordinates from the callout above do the heavy lifting — at any point p of a Kähler manifold, the metric osculates the flat one to second order, so the FIRST-order identity, which only sees the metric and its first derivatives, must hold at p exactly as it does on flat C^n. Since p was arbitrary and both sides are tensorial, the identity holds globally. The remaining three identities follow by taking adjoints and applying complex conjugation. This is a proof sketch, not a proof — the genuine article fills several pages of any Kähler geometry text.
Why Delta_d = 2 Delta_dbar is the whole point
Stand back and feel what the consequence buys you. On a bare complex manifold, de Rham cohomology H^k(M) (computed with d) and Dolbeault cohomology H^{p,q}(M) (computed with dbar, from Guide 2) live in different worlds — d does not respect the (p,q) grading, so there is no reason a harmonic k-form should split neatly by type. The Kähler identity Delta_d = 2 Delta_dbar forces the two Laplacians to have the SAME kernel: a form is d-harmonic exactly when it is dbar-harmonic. Therefore the space of harmonic k-forms decomposes by type, and harmonic forms of type (p,q) are simultaneously building blocks for both cohomologies.
Run that through the Hodge theorem (every cohomology class has a unique harmonic representative, which Guide 4 develops in full) and you get the Hodge decomposition of a compact Kähler manifold: H^k(M; C) = direct sum over p+q=k of H^{p,q}(M), with the Hodge symmetry H^{p,q} = complex-conjugate of H^{q,p}. The Betti numbers split into Hodge numbers h^{p,q}, b_k = sum of h^{p,q}, and the symmetry h^{p,q} = h^{q,p} immediately forces every odd Betti number to be even — which is precisely the obstruction that ruled out the Hopf surface earlier. One closedness condition, d omega = 0, propagated through the Kähler identities, has reorganized the entire topology of M.
- Lay down a Hermitian metric: a Riemannian g with g(Jv, Jw) = g(v, w); repackage it as the fundamental (1,1)-form omega(v, w) = g(Jv, w). Every complex manifold admits one.
- Impose the Kähler condition d omega = 0. Equivalent forms: nabla J = 0 (J parallel), Chern connection = Levi-Civita connection, or local existence of a Kähler potential omega = i partial dbar phi and of Kähler normal coordinates.
- Harvest the Kähler identities, e.g. [Lambda, dbar] = -i partial*, by checking them on flat C^n and transporting via normal coordinates; take adjoints and conjugates for the rest.
- Cash them in: Delta_d = 2 Delta_partial = 2 Delta_dbar, so harmonic forms split by (p,q) type, giving the Hodge decomposition, the Hodge symmetry h^{p,q} = h^{q,p}, and hence even odd-Betti numbers — the toolkit Guide 4 turns into full Hodge theory and Hard Lefschetz.