Künneth formula for de Rham cohomology — two proofs

Intuition [Beginner]

The Künneth formula says that the cohomology of a product of two spaces is the product of their cohomologies. If you know the holes of $M$ and the holes of $F$, you know the holes of $M\times F$ — they are exactly the products.

Concretely, take $M$ a circle and $F$ a circle. Each circle has one zero-dimensional hole (the connected component) and one one-dimensional hole (the loop). Their product is the torus. The torus has one zero-dimensional hole, two one-dimensional holes (the two loops), and one two-dimensional hole (the surface). Two times two equals four, and the four pieces match the four holes by degree.

The pattern recurs in higher dimensions. The product of three circles has eight holes, distributed across degrees 0 through 3 in the Pascal pattern. The cohomology of an $n$-torus is the binomial expansion in disguise.

Visual [Beginner]

A torus pictured as a circle times a circle, with each one-dimensional loop on the torus identified as a single-circle loop multiplied by a constant on the other factor.

The picture is a guide: the cohomology of a product is built by multiplying classes on the factors, one class for each "axis" of the product.

Worked example [Beginner]

The two-torus $T^2=S^1\times S^1$. Each circle has cohomology of dimension one in degree zero and dimension one in degree one. The product table is one row of one-by-one in degree zero, two rows of one-by-one in degree one (one from each factor), and one row of one-by-one in degree two (the product of the two top classes).

Total: one class in degree zero, two classes in degree one, one class in degree two. Total Betti numbers: $1,2,1$.

For the three-torus $T^3=S^1\times S^1\times S^1$, repeat the same multiplication once more. Betti numbers: $1,3,3,1$.

What this tells us: Künneth converts a topology question on a product into a multiplication on each factor.

Check your understanding [Beginner]

Formal definition [Intermediate+]

Let $M$ and $F$ be smooth manifolds. There is a natural map of de Rham complexes

$$\times :{\Omega }^{\ast }(M)\otimes {\Omega }^{\ast }(F)⟶{\Omega }^{\ast }(M\times F),\alpha \otimes \beta ⟼{\pi }_M^{\ast }\alpha \wedge {\pi }_F^{\ast }\beta ,$$

where ${\pi }_M,{\pi }_F$ are the two projections. The map respects the differential: $d(\alpha \otimes \beta )=d\alpha \otimes \beta +(-1{)}^{|\alpha |}\alpha \otimes d\beta$ on the tensor side maps to the wedge product expansion of $d({\pi }_M^{\ast }\alpha \wedge {\pi }_F^{\ast }\beta )$ on the right.

Theorem (Künneth formula for de Rham cohomology). Let $M$ be a smooth manifold of finite type — that is, admitting a finite good cover 03.04.10 — and let $F$ be any smooth manifold. The cross product induces an isomorphism

$$H_{\mathrm{dR}}^{\ast }(M)\otimes H_{\mathrm{dR}}^{\ast }(F)\stackrel{\cong }{\to }{H}_{\mathrm{dR}}^{\ast }(M\times F).$$

In particular, $\dim H_{\mathrm{dR}}^k(M\times F)={\sum }_{p+q=k}\dim H_{\mathrm{dR}}^p(M)\cdot \dim H_{\mathrm{dR}}^q(F)$ ^{[Bott-Tu §5]}.

The finite-type hypothesis on $M$ is load-bearing. Without it, the cross-product map is still defined, but cohomology of an infinite product of factors need not split as a tensor product of finite-dimensional pieces.

Key theorem with proof [Intermediate+]

The proof we present here is the Mayer-Vietoris induction of Bott-Tu §5. The Master tier carries the alternative tic-tac-toe proof from §9.

Proof (Mayer-Vietoris induction). Fix $F$. Define a contravariant functor $A^{\ast }$ on open subsets $U\subseteq M$ by

$$A^{\ast }(U):=H_{\mathrm{dR}}^{\ast }(U)\otimes H_{\mathrm{dR}}^{\ast }(F),$$

and a contravariant functor $B^{\ast }$ by $B^{\ast }(U):=H_{\mathrm{dR}}^{\ast }(U\times F)$. The cross product gives a natural transformation $A^{\ast }\to B^{\ast }$. We show this is an isomorphism by induction on the size of a good cover of $M$.

Base case: $M={\mathbb{R}}^n$. Both sides reduce to $H_{\mathrm{dR}}^{\ast }(F)$ — the left because $H_{\mathrm{dR}}^{\ast }({\mathbb{R}}^n)$ is one-dimensional concentrated in degree zero (Poincaré lemma 03.04.06), the right because ${\mathbb{R}}^n\times F$ deformation-retracts to $F$. The cross-product map sends a generator times any $\beta$ to the form $1\otimes \beta$ on the right, which is $\beta$ pulled back to ${\mathbb{R}}^n\times F$. This is the homotopy equivalence, so the map is an isomorphism.

Inductive step. Suppose $M=U\cup V$, with $A^{\ast }\to B^{\ast }$ an isomorphism on $U$, on $V$, and on $U\cap V$ (the inductive hypotheses). Both $A^{\ast }$ and $B^{\ast }$ have Mayer-Vietoris long exact sequences: $A^{\ast }$ inherits one from the MV sequence on $U,V$ tensored with $H_{\mathrm{dR}}^{\ast }(F)$ — tensoring with a fixed graded vector space preserves exactness — and $B^{\ast }$ has the MV sequence of $U\times F,V\times F$ covering $M\times F$. The cross product is natural in the open set, so it gives a map of MV long exact sequences.

The five lemma 02.05.01 then forces $A^{\ast }(M)\to B^{\ast }(M)$ to be an isomorphism: four of the five vertical maps in the relevant ladder are isomorphisms by inductive hypothesis, so the fifth must be too.

Termination. Choose a finite good cover $\{U_1,\dots ,U_n\}$ of $M$ 03.04.10. Each $U_i$ is contractible, so the base case applies. Each finite intersection is also contractible. The inductive step then bootstraps from intersections to unions: $U_1$, then $U_1\cup U_2$, then $U_1\cup U_2\cup U_3$, and so on. After $n$ steps, $A^{\ast }(M)=B^{\ast }(M)$.

The argument requires the Mayer-Vietoris sequence on $U\cap V$ to be related to the MV sequence on $(U\cap V)\times F$, which itself is the same induction one level down. The good-cover hypothesis ensures the recursion bottoms out. $\square$

The proof is the prototype Mayer-Vietoris-induction argument: define both sides as functors, exhibit a natural transformation, check the base case on a contractible piece, glue with five-lemma. The same template proves Poincaré duality, the de Rham theorem, and finite-type rigidity statements.

Synthesis. The Künneth dual proof exemplifies the foundational principle that the same theorem can be proved twice — first by inductive MV, then by tic-tac-toe — and the second proof is shorter because the first installed the right machinery. This is exactly the dual-proof discipline. The Künneth formula is an instance of the local-to-global motif on a product space.

Bridge. The construction here builds toward later units of the strand, where the same pattern is taken up at higher structure. 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 tensor products of graded modules but does not yet have a formalised de Rham Künneth theorem. The companion file declares the statement.

[object Promise]

The missing project work: build the de Rham complex layer over Mathlib's smooth-manifold infrastructure, formalise good covers and the MV induction, then state and prove the cross-product isomorphism. The five-lemma upgrade step is straightforward once the MV machinery is present.

Advanced results [Master]

Second proof — tic-tac-toe on $M\times F$

The Mayer-Vietoris induction proves Künneth by gluing. The Čech-de Rham double complex 03.04.11 proves it more directly. Pick a finite good cover $\mathcal{U}=\{U_{\alpha }\}$ of $M$ and a finite good cover $\mathcal{V}=\{V_{\beta }\}$ of $F$. The product cover $\mathcal{U}\times \mathcal{V}=\{U_{\alpha }\times V_{\beta }\}$ is a good cover of $M\times F$, since contractibility is preserved under products.

The Čech-de Rham double complex of $\mathcal{U}\times \mathcal{V}$ on $M\times F$ has bidegree $(p,q)\in {\mathbb{Z}}_{\ge 0}^2$ with entries

$$K^{p,q}(M\times F)=\prod _{{\alpha }_0<\cdots <{\alpha }_p,{\beta }_0<\cdots <{\beta }_q}{\Omega }^{\ast }((U_{{\alpha }_0}\cap \cdots \cap U_{{\alpha }_p})\times (V_{{\beta }_0}\cap \cdots \cap V_{{\beta }_q})).$$

By contractibility, each factor in the product reduces to a single de Rham complex on a contractible product set, which is a Poincaré-lemma piece.

The total complex computes $H_{\mathrm{dR}}^{\ast }(M\times F)$ by tic-tac-toe ascent. The first-page filtration starts on ${\Omega }^{\ast }$-rows, which are themselves products of de Rham complexes on $U_{{\alpha }_0\cdots {\alpha }_p}\times V_{{\beta }_0\cdots {\beta }_q}$. By the elementary Künneth on a product of contractible sets, each row collapses to $H_{\mathrm{dR}}^{\ast }(\text{contractible})\otimes H_{\mathrm{dR}}^{\ast }(\text{contractible})\cong \mathbb{R}$ in degree zero. The total cohomology is then the Čech cohomology of the product cover $\mathcal{U}\times \mathcal{V}$ with constant coefficients. By the Künneth formula for Čech cohomology of constant coefficients, this factors as ${\check{H}}^{\ast }(\mathcal{U};\underline{\mathbb{R}})\otimes {\check{H}}^{\ast }(\mathcal{V};\underline{\mathbb{R}})$.

But ${\check{H}}^{\ast }(\mathcal{U};\underline{\mathbb{R}})=H_{\mathrm{dR}}^{\ast }(M)$ by the de Rham theorem 03.04.13, and similarly for $\mathcal{V}$. The product is $H_{\mathrm{dR}}^{\ast }(M)\otimes H_{\mathrm{dR}}^{\ast }(F)$.

Why the second proof is shorter

The first proof set up the right machinery — Mayer-Vietoris of de Rham, good covers, and the five lemma. The second proof reuses all of it and adds only one observation: Čech of constant coefficients on a product cover factors as a tensor product.

This is a recurrent feature of Bott-Tu's exposition. The first time a theorem is proved, the proof installs the apparatus; the second proof of the same theorem, on better apparatus, takes a paragraph. The pattern recurs across Bott-Tu: Poincaré duality is proved twice (§5 by MV induction; §12 by tic-tac-toe), the Thom isomorphism is proved twice (§6 directly; §12 via the double complex), and Künneth is proved twice (§5 by MV induction; §9 by tic-tac-toe).

Künneth for compactly-supported cohomology

When $M$ has finite type, the same arguments yield

$$H_c^{\ast }(M\times F)\cong H_c^{\ast }(M)\otimes H_c^{\ast }(F).$$

The MV-induction proof requires the compactly-supported MV sequence and a base case on ${\mathbb{R}}^n\times F$, where $H_c^{\ast }({\mathbb{R}}^n)=\mathbb{R}$ in top degree only. The tic-tac-toe proof goes through unchanged on the compactly-supported double complex with vertical $d_c$ and horizontal $\delta$.

Failure on infinite-type manifolds

The cross-product map is defined on any pair $(M,F)$, but the isomorphism statement fails when $M$ has infinite-dimensional $H_{\mathrm{dR}}^{\ast }$. The standard counterexample is $M=$ an infinite disjoint union of circles. There $H^0(M)$ and $H^1(M)$ are both countably infinite-dimensional, but $H^{\ast }(M\times F)$ depends on whether $F$ is compact — the tensor product over an infinite-dimensional space does not commute with the cross-product map.

For paracompact manifolds with countable topology that nonetheless have infinite-dimensional cohomology (e.g., infinite-genus surfaces), the cross-product map is injective but not surjective.

Full proof set [Master]

Cross product is well-defined on cohomology

If $\alpha \in {\Omega }^p(M)$ is closed and $\beta \in {\Omega }^q(F)$ is closed, then ${\pi }_M^{\ast }\alpha \wedge {\pi }_F^{\ast }\beta$ is closed:

$$d({\pi }_M^{\ast }\alpha \wedge {\pi }_F^{\ast }\beta )=d{\pi }_M^{\ast }\alpha \wedge {\pi }_F^{\ast }\beta +(-1{)}^p{\pi }_M^{\ast }\alpha \wedge d{\pi }_F^{\ast }\beta ={\pi }_M^{\ast }d\alpha \wedge {\pi }_F^{\ast }\beta +(-1{)}^p{\pi }_M^{\ast }\alpha \wedge {\pi }_F^{\ast }d\beta =0.$$

If $\alpha =d\eta$, then ${\pi }_M^{\ast }\alpha \wedge {\pi }_F^{\ast }\beta =d({\pi }_M^{\ast }\eta \wedge {\pi }_F^{\ast }\beta )-(-1{)}^{p-1}{\pi }_M^{\ast }\eta \wedge d{\pi }_F^{\ast }\beta =d({\pi }_M^{\ast }\eta \wedge {\pi }_F^{\ast }\beta )$ since $\beta$ is closed. So the cross product on cocycles descends to a map on cohomology.

Naturality in MV sequences

The MV sequence on $U,V$ covering $M$ is

$$\cdots \to H^k(M)\to H^k(U)\oplus H^k(V)\to H^k(U\cap V)\stackrel{\delta }{\to }{H}^{k+1}(M)\to \cdots$$

Tensoring each term with $H_{\mathrm{dR}}^{\ast }(F)$ — a fixed graded vector space, and tensoring with a fixed graded vector space is exact — produces the MV sequence on the $A^{\ast }$ side. The MV sequence on $U\times F,V\times F$ covering $M\times F$ produces the $B^{\ast }$ side. The cross-product map sends $\alpha \otimes \beta$ to ${\pi }_M^{\ast }\alpha \wedge {\pi }_F^{\ast }\beta$, and pullback by ${\pi }_M$ sends restrictions to $U,V,U\cap V$ to restrictions to $U\times F,V\times F,(U\cap V)\times F$, since ${\pi }_M^{-1}(U)=U\times F$. Therefore the cross-product map intertwines the MV connecting homomorphisms on both sides.

Five-lemma application

The MV ladder has six terms in each row. Inductive hypothesis gives isomorphisms on $H^k(U)$, $H^k(V)$, $H^k(U\cap V)$, $H^{k-1}(U\cap V)$, $H^{k-1}(U)$, $H^{k-1}(V)$. The five lemma forces the central rung — the map on $H^k(M)$ — to be an isomorphism, completing the induction.

Connections [Master]

• This unit invokes the proposed connection conn:431.tic-tac-toe-kunneth from diffgeo.de-rham.cech-de-rham-double-complex to diffgeo.de-rham.kunneth (type: equivalence; anchor phrase: tic-tac-toe Künneth equivalent to MV-induction Künneth on finite-good-cover manifolds). The two proofs presented above instantiate exactly this equivalence: each computes the same $H^{\ast }(M\times F)$ via two different functorial routes, and Bott-Tu §5 + §9 is the originator-text for the side-by-side comparison.

• Mayer-Vietoris 03.04.07 — the first proof's engine. Künneth is the canonical exhibit of MV-induction methodology beyond the single-space setting. Connection type: foundation-of.

• Good cover 03.04.10 — the finite-type hypothesis on $M$ is exactly the existence of a finite good cover, and the induction terminates because the cover is finite. Connection type: foundation-of.

• Čech-de Rham double complex 03.04.11 — the second proof's setting; tic-tac-toe ascent on the product double complex. Connection type: foundation-of.

• Singular cohomology and de Rham theorem 03.04.13 — the second proof routes through Čech of constant coefficients, which by de Rham equals real singular cohomology. By conn:437.de-rham-three-routes, de Rham cohomology equivalent to singular cohomology with real coefficients (three routes) — MV induction (route 1), Čech-de Rham collapse (route 2), and sheaf Leray (route 3). Connection type: equivalence.

• Leray-Hirsch theorem 03.13.03 — by conn:436.kunneth-leray-hirsch, Leray-Hirsch built on Künneth on each fiber (foundation-of). Leray-Hirsch is fibre-Künneth when the fiber cohomology pulls back to a free module on the base. Connection type: foundation-of.

• Sullivan minimal model (forthcoming N12) — Künneth on the polynomial-form realisation gives the minimal model of $M\times F$ as the tensor product of the minimal models of $M$ and $F$. Connection type: foundation-of.

• Hodge decomposition 04.09.01 — Künneth respects the Hodge bigrading on a product of compact Kähler manifolds: $H^{p,q}(M\times F)={⨁}_{p_1+p_2=p,q_1+q_2=q}{H}^{p_1,q_1}(M)\otimes H^{p_2,q_2}(F)$. Connection type: bridging-theorem.

• Throughlines and forward promises. Künneth is the canonical exhibit of Bott-Tu's dual-proof discipline. We will see the same product-cohomology pattern recur in Leray-Hirsch on each fibre 03.13.03; we will see the Sullivan minimal model of $M\times F$ tensor-factor through the minimal models of $M$ and $F$ in 03.12.06; this pattern recurs across every product-bundle and product-space computation. The foundational reason the two proofs of Künneth give the same answer is exactly the tic-tac-toe equivalence — this is precisely the equivalence between MV-induction and double-complex-collapse on the product good cover. Putting these together: Künneth is an instance of the local-to-global motif applied to a product manifold. The first proof installs MV induction; the second proof, on the right machinery, takes a paragraph — this is exactly the dual-proof discipline. The bridge between de Rham Künneth and singular Künneth is the de Rham theorem; we will later see the integer-coefficient Tor corrections vanish over $\mathbb{R}$.

Historical & philosophical context [Master]

Hermann Künneth's 1923 Über die Bettischen Zahlen einer Produktmannigfaltigkeit (Mathematische Annalen 90) proved the original product theorem for Betti numbers of triangulated product spaces, before cohomology was understood as a ring and before differential forms were a topological tool. Künneth worked with simplicial chains; his theorem says the Poincaré polynomial of a product is the product of Poincaré polynomials. The de Rham version of the result, with its proof via Mayer-Vietoris and via the Čech-de Rham double complex, is a 20th-century rephrasing in the language Bott and Tu canonised in 1982.

Bott and Tu's Differential Forms in Algebraic Topology §5 carries the Mayer-Vietoris-induction proof, and §9 carries the tic-tac-toe proof on the Čech-de Rham double complex. The dual presentation is the book's pedagogical signature — what the introduction calls the "no effort" theorem, computed by gluing two charts and then by tic-tac-toe ascent. Bott's framing in the introduction is that the second proof is shorter because the first installed the right machinery: define the double complex, observe that one filtration computes Čech and the other computes de Rham, and the same theorem falls out as a counting exercise.

The Künneth theorem in singular cohomology is a Tor-corrected version: $H^{\ast }(X\times Y;A)\cong H^{\ast }(X;A)\otimes H^{\ast }(Y;A)$ holds when $A$ is a field and the Tor terms vanish; Eilenberg-Zilber 1953 (On products of complexes, Amer. J. Math. 75) gives the integer-coefficient version with explicit Tor correction. Over $\mathbb{R}$, where de Rham cohomology lives, all Tor groups vanish and the theorem is a clean tensor product. This is why the de Rham version is so much cleaner than the integer-singular version.

Bibliography [Master]

• Künneth, H., "Über die Bettischen Zahlen einer Produktmannigfaltigkeit", Mathematische Annalen 90 (1923), 65–85.
• Bott, R. & Tu, L. W., Differential Forms in Algebraic Topology, Springer, 1982. §5 (MV induction) and §9 (tic-tac-toe).
• Eilenberg, S. & Zilber, J. A., "On products of complexes", American Journal of Mathematics 75 (1953), 200–204.
• Madsen, I. & Tornehave, J., From Calculus to Cohomology, Cambridge University Press, 1997. §6.
• Hatcher, A., Algebraic Topology, Cambridge University Press, 2002. §3.B.
• Spanier, E. H., Algebraic Topology, Springer, 1966. §5.6 (singular Künneth with Tor terms).

---

Pass 4 Agent B unit N5. The dual-proof unit — Bott-Tu §5 plus §9. Closes Künneth-formula gap and instantiates the dual-proof discipline.