Čech-de Rham double complex and the tic-tac-toe principle

Intuition [Beginner]

A Čech-de Rham double complex is a grid of patch data. On a smooth space we choose a cover by simple pieces; on every single piece, every double overlap, every triple overlap, and so on, we lay down a column of differential measurements of various degrees. The grid has two directions to move: one direction adds an overlap (the Čech direction), the other increases the form degree (the de Rham direction).

Each direction has its own boundary rule. Move right by adding an overlap; move up by taking the exterior derivative. Both rules square to zero, and they commute up to sign. The total grid carries a single boundary rule that combines the two.

The remarkable fact is that the total grid has a cohomology that can be read off either by collapsing rows first (in which case the answer is the global de Rham cohomology of the whole space) or by collapsing columns first (in which case the answer is Čech cohomology of the constant sheaf on the cover). Two routes through the grid, one number on each grid square — and the numbers along both diagonals add up to the same thing.

Visual [Beginner]

A grid of squares. Each column is labelled by an overlap pattern — first a single patch, then a double overlap, then a triple overlap, and so on. Each row is labelled by a form degree — zero-forms at the bottom, one-forms above, two-forms above that. Arrows go to the right (add an overlap) and arrows go up (take a derivative). A diagonal arrow combines both moves.

The grid is the engine. Tic-tac-toe is the rule for moving along the diagonal staircase: when a row collapses, ascend; when a column collapses, march right.

Worked example [Beginner]

Take the sphere with a cover by two patches: a slightly enlarged northern hemisphere and a slightly enlarged southern hemisphere, overlapping in a thin band around the equator.

Lay out a tiny grid. The leftmost column has two cells, one for each patch; we record the smooth functions on the patch and the one-forms on the patch. The middle column has one cell, for the equatorial overlap, with the same kinds of records. The right column is empty because three-fold overlaps don't exist for a two-patch cover.

Now ascend the diagonal. Start in the bottom-left with a smooth function on each hemisphere. Move right to compare the two hemisphere functions on the equatorial overlap; if they agree, you have a globally defined function. Move up: the new step records a one-form on the overlap that measures how the patch derivatives differ. Move right: but the right column is empty, so the staircase ends. The end of the staircase records exactly the obstruction to extending a globally defined function — and on the sphere, the obstruction in degree zero is one-dimensional (constants).

What this tells us: the staircase records cohomology. The two-patch cover of the sphere recovers the same answer as the de Rham computation, by a different route.

Check your understanding [Beginner]

Formal definition [Intermediate+]

Let $M$ be a smooth manifold equipped with an open cover $\mathcal{U}=\{U_{\alpha }{\}}_{\alpha \in I}$. For an ordered tuple of indices ${\alpha }_0<{\alpha }_1<\cdots <{\alpha }_p$, write $U_{{\alpha }_0\cdots {\alpha }_p}=U_{{\alpha }_0}\cap \cdots \cap U_{{\alpha }_p}$.

Bidegree-$(p,q)$ piece. Define

$$K^{p,q}(\mathcal{U})=\prod _{{\alpha }_0<\cdots <{\alpha }_p}{\Omega }^q\left(U_{{\alpha }_0\cdots {\alpha }_p}\right).$$

An element $\omega \in K^{p,q}$ is a tuple $\omega =({\omega }_{{\alpha }_0\cdots {\alpha }_p})$ where each ${\omega }_{{\alpha }_0\cdots {\alpha }_p}$ is a smooth $q$-form on the corresponding $(p+1)$-fold intersection.

Vertical differential (de Rham): the exterior derivative $d:K^{p,q}\to K^{p,q+1}$ acts componentwise,

$$(d\omega {)}_{{\alpha }_0\cdots {\alpha }_p}=d({\omega }_{{\alpha }_0\cdots {\alpha }_p}).$$

Horizontal differential (Čech): the operator $\delta :K^{p,q}\to K^{p+1,q}$ is

$$(\delta \omega {)}_{{\alpha }_0\cdots {\alpha }_{p+1}}=\sum _{j=0}^{p+1}(-1{)}^j{\omega }_{{\alpha }_0\cdots \widehat{{\alpha }_j}\cdots {\alpha }_{p+1}}{|}_{U_{{\alpha }_0\cdots {\alpha }_{p+1}}},$$

where the hat indicates omission and the restriction is the pullback to the larger intersection ^{[Bott-Tu §8]}.

The two operators satisfy

$$d^2=0,{\delta }^2=0,d\delta =\delta d.$$

Total complex. The total complex is the graded vector space

$${\mathrm{Tot}}^n(K)=\underset{p+q=n}{⨁}{K}^{p,q},$$

with total differential

$$D=d+(-1{)}^p\delta \text{on }{K}^{p,q}.$$

Direct calculation gives $D^2=0$, so ${\mathrm{Tot}}^{\ast }(K)$ is a cochain complex; its cohomology $H^{\ast }({\mathrm{Tot}}^{\ast }(K))$ is the total cohomology of the double complex ^{[Bott-Tu §8 + §14]}. The sign $(-1{)}^p$ is the Bott-Tu convention; it is what makes $D^2=0$ from the bare commutation $d\delta =\delta d$.

Augmentations. Two augmentation maps are central. The vertical augmentation embeds the de Rham complex of $M$ into the bottom row,

$${\Omega }^q(M)\stackrel{r}{\to }{K}^{0,q}(\mathcal{U}),\omega \mapsto (\omega {|}_{U_{\alpha }}{)}_{\alpha },$$

while the horizontal augmentation embeds Čech cochains of the constant presheaf $\underline{\mathbb{R}}$ into the bottom column,

$${\check{C}}^p(\mathcal{U};\underline{\mathbb{R}})\stackrel{\iota }{\to }{K}^{p,0}(\mathcal{U}),c\mapsto c\text{(locally constant functions on the intersection)}.$$

Generalised Mayer-Vietoris sequence. For each fixed $q$, the row

$$0\to {\Omega }^q(M)\stackrel{r}{\to }{K}^{0,q}\stackrel{\delta }{\to }{K}^{1,q}\stackrel{\delta }{\to }{K}^{2,q}\to \cdots$$

is exact. Exactness at $K^{0,q}$ is the gluing axiom for forms; exactness at higher columns is the partition-of-unity argument that prolongs the two-set Mayer-Vietoris exact sequence to arbitrary covers (per 03.04.07).

Constant-sheaf row. When $\mathcal{U}$ is a good cover (every finite intersection contractible — per 03.04.10), the column

$$0\to \mathbb{R}\to {\Omega }^0(U_{{\alpha }_0\cdots {\alpha }_p})\stackrel{d}{\to }{\Omega }^1(U_{{\alpha }_0\cdots {\alpha }_p})\stackrel{d}{\to }\cdots$$

is exact (Poincaré lemma on each contractible intersection), so the bottom column $K^{p,\ast }$ has cohomology concentrated in row $0$ with value ${\check{C}}^p(\mathcal{U};\underline{\mathbb{R}})$.

Key theorem with proof [Intermediate+]

Theorem (Čech-de Rham collapse). Let $M$ be a smooth manifold with a good cover $\mathcal{U}$. Then the augmentation maps $r$ and $\iota$ induce isomorphisms

$$H_{\mathrm{dR}}^{\ast }(M)\cong H^{\ast }({\mathrm{Tot}}^{\ast }(K))\cong {\check{H}}^{\ast }(\mathcal{U};\underline{\mathbb{R}}).$$

In particular, the total cohomology of the Čech-de Rham double complex is independent of the route by which it is computed.

Proof. We prove the two isomorphisms separately, by row-collapse and column-collapse arguments. The pattern is identical; only the orientation differs.

Step 1 — Row collapse (vertical augmentation). Fix $q$ and consider the row

$$0\to {\Omega }^q(M)\stackrel{r}{\to }{K}^{0,q}\stackrel{\delta }{\to }{K}^{1,q}\stackrel{\delta }{\to }\cdots .$$

We claim this is exact. Exactness at $K^{0,q}$: a tuple $({\omega }_{\alpha })$ with ${\omega }_{\beta }-{\omega }_{\alpha }=0$ on $U_{\alpha \beta }$ for all $\alpha <\beta$ patches into a global $q$-form on $M$, which is the gluing axiom. Exactness at $K^{p,q}$ for $p\ge 1$: given $\omega \in K^{p,q}$ with $\delta \omega =0$, choose a partition of unity $\{{\rho }_{\alpha }\}$ subordinate to $\mathcal{U}$ and set

$${\eta }_{{\alpha }_0\cdots {\alpha }_{p-1}}=\sum _{\beta }{\rho }_{\beta }\cdot {\omega }_{\beta {\alpha }_0\cdots {\alpha }_{p-1}},$$

where the sum extends each tuple by $\beta$ on the left and forms vanish off the support of ${\rho }_{\beta }$. Direct calculation gives $\delta \eta =\omega$. Therefore the sequence is exact.

Now run the standard "collapsing the spectral sequence at $E_1$" argument by hand. View the row exactness as saying that $K^{\ast ,q}$ is a resolution of ${\Omega }^q(M)$ in the horizontal direction. A standard double-complex argument (column-by-column lifting, using row exactness to fill in obstructions) produces, for any cocycle $\omega \in {\Omega }^q(M)$ with $d\omega =0$, a class in $H^q({\mathrm{Tot}}^{\ast }(K))$, and the assignment is the cohomological augmentation $r_{\ast }$. Conversely, any total cocycle $\omega ={\omega }^{0,q}+{\omega }^{1,q-1}+\cdots$ can be reduced — by adding a total coboundary — to a representative concentrated entirely in column zero: $\omega ={\omega }^{0,q}$ with $\delta {\omega }^{0,q}=0$ and $d{\omega }^{0,q}=0$. Row exactness identifies this with a $d$-closed global form on $M$. Hence

$$H_{\mathrm{dR}}^{\ast }(M)\stackrel{r_{\ast }}{\to }{H}^{\ast }({\mathrm{Tot}}^{\ast }(K))\text{is an isomorphism.}$$

Step 2 — Column collapse (horizontal augmentation). Fix $p$ and consider the column

$$0\to {\check{C}}^p(\mathcal{U};\underline{\mathbb{R}})\stackrel{\iota }{\to }{K}^{p,0}\stackrel{d}{\to }{K}^{p,1}\stackrel{d}{\to }\cdots .$$

On a good cover, every $(p+1)$-fold intersection $U_{{\alpha }_0\cdots {\alpha }_p}$ is contractible, so the de Rham cohomology of each intersection vanishes in positive degrees and is one-dimensional in degree zero. Therefore the column is exact, with image of $\iota$ exactly the locally constant functions on each intersection — the Čech cochain group of the constant presheaf $\underline{\mathbb{R}}$.

The same column-by-column argument as Step 1 — reversed — produces an isomorphism

$${\check{H}}^{\ast }(\mathcal{U};\underline{\mathbb{R}})\stackrel{{\iota }_{\ast }}{\to }{H}^{\ast }({\mathrm{Tot}}^{\ast }(K)).$$

Composition. Both augmentations land in the same total cohomology, so their composite gives an isomorphism

$$H_{\mathrm{dR}}^{\ast }(M)\stackrel{r_{\ast }}{\to }{H}^{\ast }({\mathrm{Tot}}^{\ast }(K))\stackrel{{\iota }_{\ast }}{\leftarrow }{\check{H}}^{\ast }(\mathcal{U};\underline{\mathbb{R}}).$$

This is the de Rham theorem on a good cover, with $\mathbb{R}$-coefficients, in its sharpest form. $\square$

Synthesis. The double complex is the central insight of Bott-Tu §8: two filtrations on the same total complex give two routes to the same cohomology. This is exactly Weil's 1952 reformulation. The Čech-de Rham double complex generalises Mayer-Vietoris from two-set covers to arbitrary covers.

Bridge. The construction here builds toward 03.04.12 (künneth formula for de rham cohomology — two proofs), where the same data is upgraded, and the symmetry side is taken up in 03.04.13 (singular cohomology and the de rham theorem (with $z$ coefficients)). The defining pattern appears again in those units in a sharpened form, where the local data is glued or quotiented. Putting these together, the foundational insight is that the data of this unit gives the structural signature that the rest of the strand reads off.

Exercises [Intermediate+]

Lean formalization [Intermediate+]

lean_status: partial — Mathlib has Čech complexes for chain complexes of abelian groups via CategoryTheory.Limits.Cone constructions, and the de Rham complex of differential forms via Mathlib.Geometry.Manifold.DifferentialForms. The Čech-de Rham double complex itself, with its product-cover indexing and the explicit total differential $D=d+(-1{)}^p\delta$, is not yet wired together.

[object Promise]

The missing project work is to wire Mathlib's DoubleComplex machinery to the geometric setting: indexing by ordered tuples on a smooth cover, restrictions to intersections as the simplicial face maps, exterior derivative as the second differential, and the partition-of-unity contracting homotopy as the proof of row exactness.

Advanced results [Master]

Generalised Mayer-Vietoris. The exact rows of the double complex are themselves the generalised Mayer-Vietoris sequence: the original two-set Mayer-Vietoris of 03.04.07 is the special case where $\mathcal{U}=\{U_0,U_1\}$ and the row terminates at $K^{1,q}$. The arbitrary-cover version is what powers the inductive proofs of finite-dimensionality on a finite good cover (per 03.04.10).

Tic-tac-toe Künneth. When $M=X\times Y$ admits a product good cover $\mathcal{U}=\{U_{\alpha }\times V_{\beta }\}$, the Čech-de Rham double complex factors as a tensor product of the factor complexes. Applying tic-tac-toe collapse to the resulting four-grid produces the Künneth formula

$$H^k(X\times Y)\cong \underset{i+j=k}{⨁}{H}^i(X)\otimes H^j(Y).$$

This is one of the two routes to Künneth (the other, via Mayer-Vietoris induction on a good cover, will appear in the forthcoming Künneth unit). The second route is shorter than the first, because the first set up the right machinery — the dual-proof discipline that recurs throughout Bott-Tu.

Tic-tac-toe Poincaré duality. The same diagram, run with the augmented complex of compactly-supported forms in one direction, produces the de Rham version of Poincaré duality on a finite good cover. The argument is dual to the Künneth one: one tic-tac-toe collapse identifies $H_{\mathrm{dR}}^{\ast }$, the other identifies $H_c^{\ast }$, and the bilinear pairing sits between.

Sheaf-cohomology generalisation. Replacing ${\Omega }^q$ in the rows with sections of an arbitrary sheaf $\mathcal{F}$ produces the Čech double complex computing ${\check{H}}^{\ast }(\mathcal{U};\mathcal{F})$. When $\mathcal{F}$ has a fine resolution (e.g., the de Rham resolution $0\to \underline{\mathbb{R}}\to {\Omega }^0\to {\Omega }^1\to \cdots$ of the constant sheaf), the same row-collapse / column-collapse pattern proves Leray's theorem: Čech cohomology on a good cover computes sheaf cohomology. This is the route the forthcoming D2 deepening of 04.03.01 takes.

Spectral sequence interpretation. The double complex carries two filtrations: filter by columns first (degenerate at $E_1$ on a good cover, recovering Čech) or filter by rows first (degenerate at $E_1$, recovering de Rham). Both spectral sequences converge to the same total cohomology — this is the abstract framework of 03.13.01 forthcoming, with the Čech-de Rham double complex as its motivating prototype. The reader will already have done several spectral sequences by the time the abstract definition is given.

Quillen-Sullivan rationalisation. The same double-complex pattern, with ${\Omega }^{\ast }$ replaced by Sullivan's polynomial differential forms on the simplicial set of a topological space, computes the rational cohomology of any space — Sullivan's 1977 Infinitesimal computations in topology. The Čech-de Rham double complex is the smooth-manifold prototype of a much wider machine.

Full proof set [Master]

Sign of the total differential. The convention $D=d+(-1{)}^p\delta$ on $K^{p,q}$ is forced by $D^2=0$ given $d\delta =\delta d$. Set $D^{\prime }=d+ϵ\delta$ for some sign $ϵ=ϵ(p,q)$ depending on bidegree. Compute $D^{\prime 2}\omega$ on $K^{p,q}$: the cross-terms read $ϵ(p+1,q)\cdot \delta d\omega +ϵ(p,q)\cdot d\delta \omega =(ϵ(p+1,q)+ϵ(p,q))\delta d\omega$, using $d\delta =\delta d$. For $D^{\prime 2}=0$, need $ϵ(p+1,q)=-ϵ(p,q)$. The minimal solution is $ϵ(p,q)=(-1{)}^p$, independent of $q$. The Bott-Tu sign $D=d+(-1{)}^p\delta$ is the standard cohomological choice; alternative $D=(-1{)}^{q}d+\delta$ is also used (Griffiths-Harris) but with corresponding tic-tac-toe sign adjustments.

Tic-tac-toe ascent — explicit algorithm. Let $\omega \in {\Omega }^q(M)$ be $d$-closed. We construct a class in ${\check{H}}^q(\mathcal{U};\underline{\mathbb{R}})$. Restrict $\omega$ to each patch $U_{\alpha }$ and write $\omega {|}_{U_{\alpha }}=d{\eta }_{\alpha }^{(0)}$ for some ${\eta }_{\alpha }^{(0)}\in {\Omega }^{q-1}(U_{\alpha })$, by the Poincaré lemma on $U_{\alpha }$ (good cover). On the overlap $U_{{\alpha }_0{\alpha }_1}$, the difference ${\eta }_{{\alpha }_1}^{(0)}-{\eta }_{{\alpha }_0}^{(0)}$ is $d$-closed, so equals $d{\eta }_{{\alpha }_0{\alpha }_1}^{(1)}$ for some ${\eta }^{(1)}\in {\Omega }^{q-2}(U_{{\alpha }_0{\alpha }_1})$, again by Poincaré on the contractible double overlap. Iterate: at step $k$, we have ${\eta }^{(k)}\in K^{k,q-1-k}$ with $\delta {\eta }^{(k-1)}=d{\eta }^{(k)}$. The procedure terminates when $q-1-k=-1$, i.e., $k=q$, giving ${\eta }^{(q)}\in K^{q,0}$ — a Čech $q$-cocycle of locally constant functions, by $d{\eta }^{(q)}=0$. The class $[{\eta }^{(q)}]\in {\check{H}}^q(\mathcal{U};\underline{\mathbb{R}})$ is the image of $[\omega ]$ under the row-vs-column equivalence. This is the diagonal staircase ascent — the picture being a series of zig-zag moves on the grid, alternating $\delta$ and inverted-$d$ steps, all forced by exactness.

Independence of choices. At each step the antiderivative ${\eta }^{(k)}$ is determined only up to a $d$-closed form in the appropriate bidegree. Different choices change the resulting Čech cocycle ${\eta }^{(q)}$ by a Čech coboundary, so the cohomology class is well-defined.

Two routes give the same isomorphism. The composite $r_{\ast }^{-1}\circ {\iota }_{\ast }:{\check{H}}^{\ast }(\mathcal{U};\underline{\mathbb{R}})\to H_{\mathrm{dR}}^{\ast }(M)$ is the inverse of the tic-tac-toe ascent. The two are mutually inverse on representatives because the staircase ascent is built precisely from the column-exactness antiderivatives and the row-exactness gluing — that is, from the same data that defines ${\iota }_{\ast }$ and $r_{\ast }$.

Mayer-Vietoris from the double complex. For a two-set cover $\mathcal{U}=\{U_0,U_1\}$, the row $0\to {\Omega }^q(M)\to K^{0,q}\to K^{1,q}\to 0$ has only three non-zero terms. Splicing the rows at fixed $q$ along the bottom and the columns gives, in total cohomology, the long exact Mayer-Vietoris sequence

$$\cdots \to H^q(M)\to H^q(U_0)\oplus H^q(U_1)\to H^q(U_0\cap U_1)\to H^{q+1}(M)\to \cdots .$$

The connecting homomorphism is exactly the tic-tac-toe one-step zig-zag.

Connections [Master]

• Good cover and Mayer-Vietoris induction 03.04.10 — the construction of the double complex requires a cover and acquires its sharp form on a good cover; by conn:430.good-cover-cech-de-rham, Čech-de Rham double complex built on a good cover (foundation-of). Without the good-cover hypothesis, column exactness fails and the double complex carries spectral-sequence content rather than collapse content.

• Mayer-Vietoris sequence 03.04.07 — the row exactness of the double complex is the generalisation of the two-set Mayer-Vietoris exact sequence to arbitrary covers; the connecting map of MV is the one-step tic-tac-toe zig-zag. The Čech and de Rham filtrations are recurring instances of the local-to-global obstruction motif we have seen since 03.04.06.

• De Rham cohomology 03.04.06 — the row-collapse direction identifies total cohomology with $H_{\mathrm{dR}}^{\ast }(M)$; the augmentation $r$ is what makes the double complex a refinement, not a replacement, of the de Rham complex.

• Sheaf cohomology 04.03.01 — the column structure is exactly the constant-sheaf Čech apparatus; on a good cover, the column collapse identifies ${\check{H}}^{\ast }(\mathcal{U};\underline{\mathbb{R}})$ with sheaf cohomology of $\underline{\mathbb{R}}$, the Leray-on-good-covers theorem. This is precisely the deepening D2 will articulate at the sheaf-cohomology end.

• Künneth formula (forthcoming) 03.04.12 — by conn:431.tic-tac-toe-kunneth, tic-tac-toe Künneth equivalent to MV-induction Künneth on finite-good-cover manifolds (equivalence). The two proofs of Künneth are the canonical exemplar of Bott-Tu's dual-proof discipline.

• Singular cohomology and the de Rham theorem (forthcoming) 03.04.13 — by conn:432.cech-de-rham-singular, de Rham theorem built on Čech-de Rham double-complex collapse (foundation-of). One of three routes to the de Rham theorem; the other two are Mayer-Vietoris induction (per 03.04.07) and sheaf-cohomology Leray (per 04.03.01).

• Spectral sequences (forthcoming) 03.13.01 — by conn:440.exact-couple-double-complex, exact-couple spectral sequence equivalent to the Čech-de Rham double-complex spectral sequence (equivalence). The Čech-de Rham double complex is the prototype of the abstract two-filtered-spectral-sequence theory. The row-first and column-first spectral sequences both converge to total cohomology — this is the structural content the spectral-sequence formalism will abstract from the geometric instance, and the foundational reason the reader has already done several spectral sequences before meeting the abstract definition.

The Čech-de Rham double complex is the foundational grid in three downstream directions: it powers Künneth and Poincaré duality (via tic-tac-toe), it computes singular and sheaf cohomology (via the de Rham theorem on good covers), and it serves as the geometric prototype for spectral sequences. We will see the same grid appear, in a sheaf-cohomology context with ${\Omega }^q$ replaced by an injective resolution, in 04.03.01 D2; and in a Sullivan-rational context, with ${\Omega }^q$ replaced by polynomial forms, in the forthcoming 03.12.06.

Throughlines and forward promises. The Čech-de Rham double complex is the foundational grid that powers everything downstream. We will see the tic-tac-toe principle reused in 03.04.12's Künneth re-proof; we will see the exact-couple machinery of 03.13.01 abstract this concrete instance into the general spectral-sequence framework; we will later see the same grid recur in 03.12.06's Sullivan-rational refinement with $A_{PL}$ in place of ${\Omega }^{\ast }$. The foundational reason the de Rham theorem holds on every smooth manifold is exactly the column-collapse of this double complex on a good cover; this is precisely Weil 1952's insight. Putting these together: row-collapse and column-collapse give two routes to the same total cohomology, an instance of the abstract two-filtered-spectral-sequence theory. This is exactly the dual-proof discipline: the second proof of every Bott-Tu theorem (Künneth, Poincaré duality, Thom) runs through tic-tac-toe on this grid. The bridge between Čech and de Rham is the double complex; this pattern recurs in every subsequent comparison theorem.

Historical & philosophical context [Master]

Eduard Čech's 1932 Théorie générale de l'homologie dans un espace quelconque (Fund. Math. 19) introduced what we now call Čech cohomology — a complex constructed from open covers of a topological space. Čech's original motivation was to give a homology theory for non-triangulable spaces; the cover-theoretic construction sidestepped the simplicial-complex apparatus that classical Vietoris and Alexander homology had relied on. The theory remained primarily a tool of pure topology for two decades.

André Weil's 1952 Sur les théorèmes de de Rham (Comment. Math. Helv. 26) is the originator-text for the double-complex proof of the de Rham theorem. Weil understood that the de Rham theorem could be proved cleanly by the simultaneous use of the Čech and de Rham resolutions of the constant sheaf — what we now recognise as the column-and-row collapse of the Čech-de Rham double complex on a good cover. Weil's paper coincided with the rapid development of sheaf cohomology by Cartan-Eilenberg and Leray; the double-complex framework was the geometric face of homological algebra.

The double-complex tic-tac-toe pedagogy — the part of the apparatus channeled here directly from §8–§9 of Bott-Tu — is Bott and Tu's signature contribution to the exposition. In their 1982 textbook, Bott and Tu describe the tic-tac-toe principle as follows: the sphere is a particular instance, but the principle is general — given any double complex, the diagonal of the grid encodes the total cohomology, and the two routes to the diagonal are the two natural collapses. The diagonal staircase, drawn explicitly on the page as a sequence of zig-zag moves on the bigraded grid, is what they ask the reader to internalise.

Bott and Tu motivate the apparatus by treating it as the same proof, twice: in §8 they prove Künneth via tic-tac-toe; in §9 they prove the Mayer-Vietoris Künneth by induction over a good cover and then immediately re-derive it via tic-tac-toe. The second proof is shorter than the first, because the first set up the right machinery — the dual-proof discipline that organises the entire central chapter of the book. The same pattern then recurs for Poincaré duality (§5 induction proof; §12 tic-tac-toe proof) and for the Thom isomorphism (§6 induction proof; re-proved via Čech-de Rham in the forthcoming Künneth unit). Each parallel pair of proofs makes the same point: a clean grid is a tool that reduces the second proof to a calculation, freeing the reader's intuition for the next theorem.

The deeper structural fact — visible from the spectral-sequence chapter (§14) — is that the Čech-de Rham double complex carries two filtrations whose spectral sequences both converge to total cohomology. By the time the reader reaches the formal definition of a spectral sequence, the reader has already done several. This pedagogical inversion, where the abstraction comes after the prototype, is the architectural signature of Differential Forms in Algebraic Topology.

Bibliography [Master]

• Čech, E., "Théorie générale de l'homologie dans un espace quelconque", Fundamenta Mathematicae 19 (1932), 149–183.
• Weil, A., "Sur les théorèmes de de Rham", Commentarii Mathematici Helvetici 26 (1952), 119–145.
• Bott, R. & Tu, L. W., Differential Forms in Algebraic Topology, Graduate Texts in Mathematics 82, Springer-Verlag, 1982. §8 (the Čech-de Rham complex), §9 (presheaves and Čech cohomology), §12 (tic-tac-toe Künneth and Poincaré duality), §14 (spectral sequence of a double complex).
• Griffiths, P. & Harris, J., Principles of Algebraic Geometry, Wiley-Interscience, 1978. Ch. 0 §3 (Čech-Dolbeault double complex on complex manifolds).
• Madsen, I. & Tornehave, J., From Calculus to Cohomology, Cambridge University Press, 1997. §10 (sheaves and the de Rham theorem).
• Cartan, H. & Eilenberg, S., Homological Algebra, Princeton University Press, 1956. Ch. XV (spectral sequences and double complexes).
• Sullivan, D., "Infinitesimal computations in topology", Publications mathématiques de l'IHÉS 47 (1977), 269–331.
• Leray, J., "L'anneau d'homologie d'une représentation", Comptes Rendus 222 (1946), 1366–1368 — the original cover-theoretic spectral sequence.

---

Bott-Tu Pass 4 — Agent A — N3. Čech-de Rham double complex, generalised Mayer-Vietoris, tic-tac-toe ascent, sign convention $D=d+(-1{)}^p\delta$ on $K^{p,q}$, ${\mathrm{Tot}}^n$ as a cohomological prototype for the spectral-sequence chapter. Originator-prose for Bott-Tu §8–§9 + §12. Introduces notation $K^{p,q}$, ${\mathrm{Tot}}^n$, $\delta$ vs $d$, $D$ — Pass 4 §3.4 decisions #12, #13, #27, #28.