The quotient, made honest
Guide 2 left you with a slogan: the gap between closed and exact is information. Now we name that gap. Fix a smooth manifold M. The closed k-forms (those with d omega = 0) form a vector space Z^k, the kernel of d in degree k. The exact k-forms (those of the form d eta) form a subspace B^k, the image of d coming up from degree k-1. Because d^2 = 0 guarantees every exact form is closed, B^k sits inside Z^k, and the quotient is well defined. The k-th de Rham cohomology is exactly that quotient, H^k(M) = Z^k / B^k = (closed k-forms) / (exact k-forms).
An element of H^k(M) is not a single form but a cohomology class [omega], the whole family of closed forms differing from omega by an exact one. Two closed forms omega and omega' are cohomologous when omega - omega' = d eta; they are the same point of H^k. So a class is genuinely a coarser object than a form — you are allowed to add any d eta you like and you have not moved. The art of the subject is choosing, inside one class, a representative that is easy to integrate or pretty to look at (a harmonic one, once you have a metric — but that is the Hodge story, several guides ahead).
What H^0 already tells you: counting components
Start at the bottom rung, k = 0, where everything is computable by hand. A 0-form is a smooth function f, and there are no (-1)-forms, so B^0 = 0 — nothing is exact in degree zero. The closed 0-forms are the functions with d f = 0, meaning the gradient vanishes everywhere. On a connected manifold that forces f to be constant. So H^0 of a connected M is exactly the line of constant functions, a one-dimensional space, R.
Now drop the word connected. If M has several pieces, a function with vanishing derivative can take a different constant on each piece — locally constant, not globally constant. So H^0(M) = R^c, where c is the number of connected components of M. The very first cohomology group is doing honest topological accounting: dim H^0 literally counts the pieces. This is the cleanest possible illustration of the whole philosophy, an analytic kernel (solutions of d f = 0) handing back a topological count.
Holes you can see: the circle and the torus
Climb to H^1 and the holes appear. On the circle S^1, the angle form d theta is closed but not exact — exactly the example from guides 2 and 3 — and the Poincaré lemma does not save you because the circle is not contractible. Its class [d theta] is nonzero, and one shows every closed 1-form is cohomologous to a multiple of it: the multiple is just (1 / 2 pi) times the integral around the loop. So H^1(S^1) = R, one-dimensional, and that single dimension IS the one hole of the circle. H^0(S^1) = R too, since the circle is connected.
The torus T^2 is the example to truly hold in your head. Think of it with two angle coordinates theta and phi, each running around its own circle. There are now two independent angle forms, d theta and d phi, both closed and neither exact — they wind around the two distinct loops of the doughnut. So H^1(T^2) = R^2, the two generators detecting the two essentially different ways to loop. Going up one more, d theta ^ d phi is a closed top-form that is not exact (its integral over the whole torus is the area, nonzero), giving H^2(T^2) = R. The full tally is dim H^0 = 1, dim H^1 = 2, dim H^2 = 1.
manifold H^0 H^1 H^2 Betti numbers (b_0, b_1, b_2) -------------------------------------------------------------- point R 0 0 (1, 0, 0) circle S^1 R R - (1, 1) sphere S^2 R 0 R (1, 0, 1) torus T^2 R R^2 R (1, 2, 1) b_k = dim H^k(M) (the k-th Betti number) Euler char chi = sum_k (-1)^k b_k : S^2 -> 2, T^2 -> 0
Betti numbers and the Euler characteristic
For a reasonable M (compact, say) each H^k(M) turns out to be finite-dimensional, and its dimension is the k-th Betti number b_k = dim H^k(M). The slogan to carry: b_0 counts connected pieces, b_1 counts independent 1-dimensional loops, b_2 counts independent 2-dimensional voids or enclosed cavities, and so on up the ladder. The sphere S^2 has b = (1, 0, 1): one piece, no loop you cannot shrink, one enclosed void — its surface bounds a ball-shaped hole. The torus has b = (1, 2, 1). These numbers are the quantitative content of the phrase measuring the holes.
Bundle the Betti numbers into one integer and something familiar drops out. The alternating sum chi(M) = sum_k (-1)^k b_k is the Euler characteristic, and for surfaces it agrees exactly with the V - E + F you computed from triangulations back in the manifolds rung. For S^2, chi = 1 - 0 + 1 = 2; for T^2, chi = 1 - 2 + 1 = 0. That the analytic cohomology dimensions reproduce the old combinatorial count is your first hard evidence that de Rham cohomology, though built from calculus, is secretly seeing topology — and the Gauss-Bonnet theorem from the curves-and-surfaces rung ties the same chi to total curvature.
Why the holes are invariant, and what's next
What makes these numbers trustworthy is that they do not depend on the fluff of the construction. Because d commutes with pullback, any smooth map induces a linear map on cohomology, and a diffeomorphism induces an isomorphism — so diffeomorphic manifolds have identical de Rham cohomology. Stronger and more surprising: cohomology is a homotopy invariant. If two maps are smoothly homotopic they induce the same map on H^k, so a smooth homotopy equivalence forces an isomorphism. That is why a contractible space has the cohomology of a point (the Poincaré lemma upgraded to a global statement), and why H^k cannot tell R^3 from a point.
There is a real gap between what we have done and what we have claimed. We computed S^1, T^2, S^2 by clever guessing — exhibiting forms and arguing they generate — but that does not scale. We never proved H^k is finite-dimensional in general, nor gave a systematic engine. Both gaps are filled next. The Mayer-Vietoris sequence is the inductive machine that computes H^k of a space from two simpler open pieces and their overlap, turning the patient guessing above into a genuine algorithm.
And the deepest promise is the bridge itself. We keep saying these analytic invariants smell topological; the de Rham theorem proves they are, asserting that H^k(M) computed from smooth forms is canonically isomorphic to the singular cohomology of M with real coefficients — a purely topological object knowing nothing of calculus. That isomorphism, by way of integrating forms over chains via Stokes (guide 3), is the punchline of the whole rung. It is why a question about flows, fields, and integrals can be answered by counting holes. Guide 5 states it carefully and shows it at work.