(p,q)-Forms, the Dolbeault Operators & Dolbeault Cohomology

Where we are: from integrable J to splitting the forms

Guide 1 of this rung handed you a genuine complex manifold M: charts into C^n whose transition maps are holomorphic, equivalently an almost-complex structure J that is integrable because its Nijenhuis tensor vanishes (the Newlander-Nirenberg theorem). The payoff we cash in now is purely linear-algebraic at first, then becomes calculus. Because J squares to minus the identity on each tangent space, it has eigenvalues +i and −i once we complexify, and that single fact slices every space of forms cleanly in two.

A quick honesty note, the same one as last guide: this is graduate material. We assume you are fluent with smooth manifolds, the cotangent bundle T*M, the algebra of differential forms, and the exterior derivative d from the forms rung — words like manifold, chart, and tangent space are used, not re-derived. If those feel shaky, that is the rung to revisit before pressing on here.

Complexifying the tangent space: holomorphic and conjugate directions

Fix a point p and complexify the tangent space: tensor T_p M with C to get a complex vector space on which J acts C-linearly. Since J^2 = −1, this space splits into the +i-eigenspace, call it T^(1,0), and the −i-eigenspace T^(0,1). In a local holomorphic chart with coordinates z^k = x^k + i y^k, the natural basis is del/del z^k for T^(1,0) and del/del zbar^k for T^(0,1), where del/del z^k = (1/2)(del/del x^k − i del/del y^k). The first set are the holomorphic directions, the second their conjugates. This is exactly the Wirtinger calculus from one complex variable, now carried onto a manifold.

Dualize. The complexified cotangent space splits the same way, spanned by the 1-forms dz^k = dx^k + i dy^k (these annihilate the conjugate directions) and dzbar^k = dx^k − i dy^k (these annihilate the holomorphic ones). So a complexified 1-form has a (1,0)-part built from the dz^k and a (0,1)-part built from the dzbar^k. The key fact, and the reason all of this is well-defined globally and not just in one chart: because the transition maps are holomorphic, a (1,0)-form in one chart stays a (1,0)-form in the next. Integrability is what makes the splitting survive coordinate changes.

The (p,q)-bidegree: counting holomorphic and conjugate slots

Now wedge. A (p,q)-form is a sum of terms with exactly p factors of the dz-type and exactly q factors of the dzbar-type — for example f times (dz^1 ^ dz^2) ^ (dzbar^1) is a (2,1)-form, with f a smooth complex-valued function. The total degree is p+q, but the refined bookkeeping splits each space of complex k-forms into a direct sum over all (p,q) with p+q = k. This is the (p,q)-type decomposition, and it is the structural heart of complex geometry: where a smooth manifold sees one ladder of degrees, a complex manifold sees a whole grid.

complex k-forms  =  (+)  Lambda^(p,q)   over all  p + q = k

   q
   2 |  (0,2)   (1,2)   (2,2)
   1 |  (0,1)   (1,1)   (2,1)
   0 |  (0,0)   (1,0)   (2,0)
        --------------------------  p
          0       1       2

   d  =  del  +  del-bar       (on a complex manifold)
   del : (p,q) -> (p+1,q)      del-bar : (p,q) -> (p,q+1)
   del^2 = 0 ,  del-bar^2 = 0 ,  del del-bar + del-bar del = 0

A worked sanity check on a Riemann surface (n = 1, the friendliest case). There is just one z, so the grid is tiny: (0,0) are functions, (1,0)-forms are f dz, (0,1)-forms are g dzbar, and (1,1)-forms are h dz ^ dzbar — that last is the top degree, the thing you integrate. A holomorphic 1-form, the kind that appears in meromorphic function theory, is precisely a (1,0)-form f dz whose coefficient is holomorphic. Already the bidegree is telling you which forms 'come from complex analysis' and which do not.

Splitting d: the operators del and del-bar

Here is the calculus payoff. Apply the ordinary exterior derivative d to a (p,q)-form. On a complex manifold the result lands only in bidegrees (p+1,q) and (p,q+1) — never anywhere else. So we define the two Dolbeault operators as the two halves of d: del takes the (p+1,q) part and del-bar takes the (p,q+1) part, giving d = del + del-bar. In coordinates del differentiates the coefficients with respect to the z^k and wedges on a dz, while del-bar differentiates with respect to the zbar^k and wedges on a dzbar.

Now expand d^2 = 0 using d = del + del-bar and sort by bidegree. The (p+2,q) part forces del^2 = 0, the (p,q+2) part forces del-bar^2 = 0, and the mixed (p+1,q+1) part forces del del-bar + del-bar del = 0. So each Dolbeault operator squares to zero on its own — exactly the property d had — and that is what licenses building a cohomology from del-bar. The single identity d^2 = 0 secretly contained three.

The single most important special case: a smooth function f is holomorphic exactly when del-bar f = 0. Unpack it in one variable and del-bar f = (del f / del zbar) dzbar, so the equation del-bar f = 0 is del f / del zbar = 0 — which is precisely the Cauchy-Riemann equations in disguise. The whole towering edifice of several complex variables is, at bottom, the study of the single equation del-bar u = (something). That is why del-bar, the conjugate-looking operator, is the star of the show, not del.

To see this concretely, take n = 1 and write z = x + i y, so del/del zbar = (1/2)(del/del x + i del/del y). For a function the Dolbeault operator is del-bar f = (del f / del zbar) dzbar, hence del-bar f = 0 is the single scalar equation del f / del zbar = 0 — and unpacking the real and imaginary parts gives back the Cauchy-Riemann equations u_x = v_y, u_y = −v_x for f = u + i v. The Dolbeault operator is, quite literally, complex analysis written on a manifold.

Dolbeault cohomology: de Rham, refined by bidegree

Because del-bar squares to zero, for each fixed p the spaces of (p,q)-forms with q = 0,1,2,... form a complex with del-bar as the differential. Its cohomology — del-bar-closed forms modulo del-bar-exact ones — is the Dolbeault cohomology group H^(p,q)(M). The recipe is the exact analogue of de Rham: a (p,q)-form alpha is del-bar-closed if del-bar alpha = 0, it is del-bar-exact if alpha = del-bar beta for some (p,q−1)-form beta, and the quotient measures the failure of closed to be exact. Where de Rham cohomology counts topological holes, Dolbeault counts holomorphic ones, finely sorted by the two indices.

1. Fix the holomorphic index p. Look only at forms of type (p, *), a single row of the grid moving upward in q.
2. The operator del-bar moves you up that row: (p,q) -> (p,q+1), and del-bar composed with itself is zero, so you have a genuine cochain complex.
3. Take kernel over image at spot (p,q): the del-bar-closed (p,q)-forms modulo the del-bar-exact ones.
4. That quotient is H^(p,q)(M). When M is compact it is finite-dimensional; its dimension h^(p,q) is a Hodge number, a genuine invariant of the complex structure.

Two reasons this matters enormously, both deferred to later guides but worth seeing the horizon now. First, the Dolbeault theorem identifies H^(p,q)(M) with the sheaf cohomology H^q(M, Omega^p) of the sheaf of holomorphic p-forms — so this analytic gadget computes something purely algebraic-geometric, which is how del-bar talks to coherent sheaves and to holomorphic line bundles. Second, on a Kähler manifold (rung guides 3-4) the Hodge numbers assemble the de Rham Betti numbers into the Hodge diamond, with symmetries h^(p,q) = h^(q,p) and Serre duality — the bridge between topology and complex structure.

Honest caveats, conventions, and what is genuinely hard

First, conventions vary and they bite. Some books write d-bar, some del-bar, some d''; the factor of 1/2 in del/del zbar is standard but a few sources hide it; and the order of the indices in H^(p,q) is universal but the placement of p versus q on the diamond's axes is not. Worse, the sign in del del-bar + del-bar del = 0 flips in references that define del-bar with an extra sign. Pick one source, write its conventions on the inside cover of your notes, and stay loyal — most 'errors' here are two books colliding.

Second, a common misconception worth killing: H^(p,q) is in general NOT a subgroup of de Rham cohomology, and you cannot in general add up the Hodge numbers along p+q = k to recover the kth Betti number. That clean decomposition is a theorem only on compact Kähler manifolds — it fails on general compact complex manifolds (the Hopf surfaces are the standard cautionary example). Do not import the Kähler conclusion into the non-Kähler world; that is one of the most frequent slips climbing this rung.