Where guide 4 left us, and the one move we still need
By now de Rham cohomology H^k(M) is a familiar object: closed k-forms modulo exact ones (recall closed and exact from guide 2), a vector space whose dimension counts a kind of k-dimensional hole. We also know two anchors. The Poincaré lemma says a contractible open set has trivial cohomology in positive degree — locally, closed always means exact. And H^0(M) just counts connected components, because a closed 0-form is a locally constant function. The honest gap is computational: nobody wants to search the infinite-dimensional space of all forms by hand. We need a way to break M into known pieces and reassemble the answer.
The strategy is the oldest one in geometry: cut and glue. Cover M by two open sets U and V whose cohomology you already know — typically because each is contractible, or a disjoint union of contractible pieces, so the Poincaré lemma applies. The subtlety is the overlap U intersect V, where the same form has two descriptions that must agree. Mayer-Vietoris is precisely the bookkeeping device that converts "I know U, V, and their overlap" into "I know M". It is the differential-forms cousin of inclusion-exclusion, but upgraded into an exact sequence so it tracks not just dimensions but how the holes are glued.
The Mayer-Vietoris sequence: glue first, then count
Fix an open cover M = U union V. Restricting a form on M to U and to V, then comparing the two restrictions on the overlap, builds a short exact sequence of complexes of forms. Pure homological algebra (the snake lemma) turns any short exact sequence of complexes into a long exact sequence of cohomology. That single long sequence threads through every degree at once: H^k(M) sits between the cohomology of the pieces and the cohomology of the overlap, chained forever by a connecting map. "Exact" means the image of each arrow is exactly the kernel of the next — no slack, no leak — which is what lets you solve for an unknown term once its neighbors are known.
... -> H^(k-1)(U n V) --d*--> H^k(M) --(r_U, r_V)--> H^k(U) (+) H^k(V) --(s_U - s_V)--> H^k(U n V) --d*--> H^(k+1)(M) -> ... r = restrict a class on M to each piece s = restrict a class on a piece to the overlap, then subtract the two d* = the connecting map: take a class on the overlap, split it across U and V, apply d exactness at each spot: image(incoming arrow) = kernel(outgoing arrow)
The connecting map d* is the clever part and deserves a sentence of intuition. Given a class on the overlap, you cannot in general extend it to all of M, but you can split it using a partition of unity: write 1 = rho_U + rho_V subordinate to the cover, multiply your overlap form by these bumps to push it onto U and onto V separately, and take the exterior derivative of the difference. The pieces individually fail to be closed, but their difference is a genuine closed form on all of M, one degree higher. That jump in degree is exactly how a one-dimensional hole in the overlap can manifest as a two-dimensional class on M.
Two worked examples: the sphere and the torus
Lead with the example, always. Take M = S^2 and cover it by two open caps: U slightly larger than the northern hemisphere, V slightly larger than the southern, each contractible to a point, with overlap U intersect V an equatorial band that deformation-retracts to a circle. By the Poincaré lemma U and V have the cohomology of a point; the band has the cohomology of S^1. Feed these into the long exact sequence and the only unknowns are H^1(S^2) and H^2(S^2). Exactness forces H^1(S^2) = 0 and H^2(S^2) = R: the sphere has no one-dimensional holes (every loop bounds a cap) but one two-dimensional cavity (it encloses a volume). The single generator of H^2 is, up to scale, the area form.
- Choose a cover whose pieces and overlaps you already understand — usually contractible opens (cohomology of a point) and overlaps that retract to spheres or circles.
- Write the relevant stretch of the Mayer-Vietoris sequence and plug in the known cohomologies of U, V, and U intersect V from the Poincaré lemma.
- Use exactness (image = kernel at each node) plus the rank-nullity bookkeeping of an exact sequence to solve for the unknown H^k(M) term by term.
- Sanity-check against H^0 (counts components) and, on a closed oriented n-manifold, against top degree H^n = R; if a Betti number comes out negative you made an arithmetic slip.
Now the torus T^2, the example that earns its keep. The cleanest route splits the torus into two open cylinders (tubes) that wrap the short way, overlapping in two disjoint annuli. Running the sequence yields H^0 = R, H^1 = R^2, H^2 = R: one component, two independent one-dimensional holes, one cavity. Those two H^1 generators are exactly the forms d theta_1 and d theta_2 measuring winding around the two circles of the torus — the same angle-form trick that detected the hole in the circle, now in two independent directions. Compare it to the sphere and you can read the topology straight off the Betti numbers: the torus has b_1 = 2 where the sphere has b_1 = 0, and that 2 is the count of fundamental loops you cannot contract.
The de Rham theorem: forms know the topology
Here is the deep payoff of the whole rung. We built H^k(M) from smooth forms, an analytic gadget steeped in calculus. There is an entirely different, purely combinatorial-topological gadget — singular homology H_k(M) and its dual singular cohomology — built from continuous maps of simplices, with no derivatives anywhere. The de Rham theorem says these two worlds coincide: there is a natural isomorphism between de Rham cohomology and (real) singular cohomology. The bridge is integration. Stokes' theorem from guide 3 guarantees that pairing a closed form against a cycle by "integral over the cycle of omega" depends only on the cohomology class of the form and the homology class of the cycle.
Why does it earn its keep rather than just being pretty? Because it lets each side lend its strengths to the other. Forms are flexible and computable: the Poincaré lemma and Mayer-Vietoris give you H^k(M) by honest calculation, and you instantly conclude facts about loops and holes that are statements of pure topology. Conversely, topology constrains analysis: knowing the Betti numbers tells you exactly how many independent closed-but-not-exact forms a manifold must support, which feeds directly into Hodge theory in the next rung, where each cohomology class gets a unique harmonic representative.
Good covers, Cech, and where the rung points next
Mayer-Vietoris with two sets is a special case of a bigger machine. Cover M by many open sets, all of whose finite intersections are contractible — a good cover (it exists on any manifold). Then the entire cohomology is recoverable from the purely combinatorial pattern of which sets overlap, organized into Cech cohomology. The Cech-de Rham comparison shows the two agree, and it reduces computing H^k(M) to bookkeeping on a finite combinatorial object — the nerve of the cover — instead of an infinite space of forms. This is the same circle of ideas behind the nerve theorem, and it is the computational heart of how software actually evaluates topological invariants.
Be honest about the boundaries of what we have shown. The de Rham theorem as stated gives real coefficients only; it cannot see torsion — the Klein bottle hides a copy of Z/2 in its integral homology that no form can detect, because real vector spaces have no torsion to begin with. Forms compute Betti numbers, never the full integral homology. Likewise we have stated, not proved, the naturality and the Cech comparison; a genuine proof of de Rham is a chapter of a book, and we have given you the idea and the moving parts, not the full verification. With Mayer-Vietoris in hand to compute and de Rham to interpret, the next rung adds a metric and asks for the best representative of each class — that is Hodge theory.