Alphabeta Math
Pipeline-generated
How statement and proof provenance work

The first chip identifies the source of the statement or construction; the second identifies the source of its local proof or verification.

  • Literature-sourced: the exact statement appears in a cited source; only wording and notation differ.
  • AI-adapted: a semantically identical restatement of literature-sourced material, modulo indexing, notation, and boundary cases adopted by the library.
  • AI-generated: a genuinely novel statement formulated by AI, with no source for the claim itself.

These labels describe origin, not correctness: citations and verification chips remain separate evidence.

Cup Cap Cross Products and Cohomology Rings

1 · Prerequisites

2 · Summary

Cup product turns singular cohomology with a commutative unital coefficient ring into a graded ring. The construction begins with the Alexander–Whitney front/back faces and the shuffle comparison. Its cochain identities prove descent, associativity and naturality; the explicit factor-reversal homotopy explains why graded commutativity holds on cohomology even though it can fail on individual cochains.

We use the positive coboundary and write cap products with cohomology first. For a degree-p cochain the boundary identity is (αc)=(1)p(αcδαc). Relative products retain their actual subspace domains. Open-complement products use the small-chain comparison, while CW product pairs receive a separate chain equivalence before relative Künneth is applied.

The ring Künneth theorem distinguishes its choice-free multiplication identity from additive bijectivity, which assumes AC and the stated degreewise finite-free homology condition. Local coordinate products and the oriented polygon calculation provide explicit suppliers for projective-space rings and surface pairings. Cup length and the vanishing of positive-degree products on suspensions give useful ways to distinguish spaces beyond their additive groups. The companion calculations check generators, orientation signs and coefficient hypotheses.

3 · Logical flowchart

4 · Definitions, theorems and proofs

DefinitionDefinition: Literature-sourcedProof: Not applicablejudge pass (gpt-5.6-terra)audited 2026-09-13Open item page →

Alexander–Whitney map and diagonal approximation

Definition

Let X,Y be spaces and R a commutative unital ring, allowing R=0. Use ordinary, unnormalized singular chains, with the coefficient convention in Singular cochain complex with coefficients. Put Fn(X,Y;R)=p+q=nCp(X;R)RCq(Y;R),d(ab)=ab+(1)pab(aCp(X;R)). Negative chain degrees and the boundary of a vertex are zero. For a simplex σ:ΔnX×Y, write x=prXσ and y=prYσ. The Alexander–Whitney map is the linear map AWn:Cn(X×Y;R)Fn(X,Y;R),AWn(σ)=p=0nx[0,,p]y[p,,n]. Here x[i0,,ik] means restriction along the affine face with that ordered vertex list. The formula gives a finite chain for each generator and extends uniquely linearly; it does not require choosing representatives or fillings. Projections and the diagonal are continuous in the binary product topology. With the postcomposition maps of The induced singular chain map of a continuous map, the diagonal approximation is DX=AWΔ#:C(X;R)F(X,X;R),Δ(x)=(x,x).

For completeness, the formula is a chain map. In dAW(σ), the first-factor faces at cut p1 have signs (1)i, 0ip; the second-factor faces at cut pn1 have signs (1)j, pjn. The first-factor term deleting its last vertex at cut p is (1)px[0,,p1]y[p,,n]. It cancels the second-factor term deleting its first vertex at cut p1, whose sign is (1)p1. All remaining first-factor terms have i<p; they are precisely the cuts of the face of σ omitting i in which the omitted vertex lies before the cut. All remaining second-factor terms have j>p and are precisely those with the omitted vertex after the cut. Each has the sign of that face in σ. This exhausts AW(σ) and proves dAW=AW. In degree zero both sides are zero, as required. The same face calculation shows that postcomposition commutes with ; hence DX is also a chain map.

For continuous f:XX and g:YY, face restriction commutes with postcomposition, term by term, giving AW(f×g)#=(f#g#)AW,DXf#=(f#f#)DX. In degree zero AW sends (x,y) to xy, preserving the augmentation that assigns 1 to each vertex. If a factor is empty, both complexes are zero; the same holds over the zero ring. For one-point factors the formula still uses all higher singular simplices: degenerate simplices have not been discarded. The cancellation also applies to these simplices because it is an identity of face maps. Only finite sums and specified maps were used. No AC is assumed, and the arbitrary-product nonemptiness clause of the product definition is not used.

TheoremStatement: Literature-sourcedProof: AI-adaptedjudge pass (gpt-5.6-terra)audited 2026-09-13Open item page →

Alexander--Whitney and shuffle are natural chain-homotopy inverses

Statement

For arbitrary spaces X,Y and every commutative unital ring R, the Alexander–Whitney map A and signed shuffle map S are natural chain maps C(X×Y;R)ASC(X;R)RC(Y;R). Both preserve degree-zero augmentation, and AS and SA are naturally chain homotopic to the respective identity maps. Moreover, for the specified natural inverse T constructed below there is a natural chain homotopy AT. This holds for ordinary unnormalized chains and requires no AC.

Facts & Assumptions

[F1]

Alexander–Whitney map and diagonal approximation constructs A, proves its chain-map and naturality identities, and identifies its degree-zero action.

[F2]

The singular chain cross product on generators specifies the finite signed shuffle sum. The singular chain cross product satisfies the boundary formula proves that S is a chain map, and Singular chain cross products are natural proves its naturality.

[F3]

Singular product chain equivalence by simplex models states that S has a natural inverse and natural homotopies, without AC, and that scalar extension preserves the result.

[F4]

The singular chain homotopy formula gives the explicit prism homotopy for a specified homotopy of spaces. Singular cochain complex with coefficients supplies ordinary unnormalized singular chains as free modules on singular simplices.

Proof

Given: Work first over Z. Write D=C(X×Y) and F=C(X)C(Y) with the signed tensor differential.

1.1

By [F1] and [F2], A and S are natural chain maps. Both act in degree zero by the inverse identifications between a pair of vertices and their tensor.

F1F2given
1.2

We now construct the model contractions used below. If Q is a standard simplex or a product of two standard simplices and v is its first vertex, the affine homotopy H(x,t)=(1t)v+tx stays in Q. Let p:C(Q)C() collapse Q and let j:C()C(Q) include v. The prism P of [F4] satisfies dP+Pd=1jp. The point complex has one generator en in each degree, with den=en1 for positive even n and den=0 for odd n. Define aen=en+1 for odd n and aen=0 for even n. Direct substitution gives da+ad=1iϵ, where ϵ:C()Z[0] is augmentation and i includes degree zero. Thus hQ=P+jap satisfies dhQ+hQd=1eQ, where eQ=jiϵp projects onto the first vertex. For a product model in F, put hF(xy)=hCxy+(1)xeCxhEy. The mixed terms cancel, giving dhF+hFd=1eCeE. Hence in either D- or F-model every positive-degree cycle, and every augmentation-zero degree-zero cycle, has the specified filling hz.

F4givenalgebra
2.1

Construct a natural chain map T:DF. In degree zero send the vertex (x,y) to xy. Suppose T is defined naturally below degree n>0. On the universal diagonal simplex an:ΔnΔn×Δn, the chain z=Tn1dan is a cycle when n>1 and has augmentation zero when n=1. Define Tn(an)=hFz using step 1.2, and for σ=(x,y) put Tn(σ)=(x#y#)Tn(an). Freeness in [F4] extends this linearly. Then dTn=Tn1d, and composition of pair maps proves naturality. Thus T is a specified natural chain map whose degree-zero action equals that of A.

F4step 1.1step 1.2
2.2

More generally, let u,v:EG be natural chain maps between D or F functors that agree in degree zero. Set H1=0. Assuming H defined below degree n, for each universal degree-n generator a put z=(uv)aHn1da. The lower homotopy identity makes z a cycle for n>0, while at n=0 it is zero. Define Hn(a)=hz with the appropriate model contraction from step 1.2, and push forward to arbitrary generators. Then dHn+Hn1d=uv; the specified pushforward rule makes H natural.

F4step 1.2algebra
3.1

Since T and S are natural chain maps and T0S0=S0T0=1, step 2.2 applied to (TS,1F) and (ST,1D) gives natural homotopies U and V with dU+Ud=TS1 and dV+Vd=ST1. This independently realizes the existence asserted in [F3].

F2F3step 2.1step 2.2
3.2

The maps A,T:DF agree in degree zero by steps 1.1 and 2.1. Applying step 2.2 to (A,T) gives a natural K with dK+Kd=AT. This is the stated comparison with the specified inverse, including its degree-zero normalization.

step 1.1step 2.1step 2.2
4.1

Compose the identity of step 3.2 with the chain map S. Then d(KS+U)+(KS+U)d=(AT)S+(TS1)=AS1, and d(SK+V)+(SK+V)d=S(AT)+(ST1)=SA1. All summands and composites are natural, so these are the required natural chain homotopies.

F2step 3.1step 3.2
5.1

Tensor these integral identities with R. The canonical identification sends (xy)r to (x1)(yr); its inverse sends (xa)(yb) to (xy)ab. The tensor relations and commutativity of R make the maps well-defined inverses commuting with the signed differential. The extended maps are the stated AW and shuffle formulas. Additivity of tensoring preserves the homotopy identities without a flatness hypothesis.

F1F2F3step 4.1
6.1

An empty factor or R=0 makes both complexes zero. Degree zero and augmentation were checked in step 1.1; the point calculation in step 1.2 retains the nonzero higher unnormalized chains. Every image is a finite chain because the model contractions are finite prism sums and each recursion uses only finitely many faces and already defined chains. Hence degeneracy, the lowest degree, and coefficient zero impose no exception, and no choice axiom is used.

F3step 1.1step 1.2step 2.1step 2.2step 5.1
DefinitionDefinition: Literature-sourcedProof: Not applicableaudited 2026-09-13Open item page →

Singular cup product on cochains

Definition

Let X be a space, let R be a commutative unital ring, and use the positive coboundary δφ=φ of Singular cochain complex with coefficients. For p,q0 and cochains φCp(X;R), ψCq(X;R), their cup product is (φψ)(σ)=φ(σ[0,,p])ψ(σ[p,,p+q])(σ:Δp+qX). Extend from simplex generators R-linearly. The product is R-bilinear in the cochains by distributivity and commutativity in R. Zero or negative-degree inputs give zero.

With J(φ,ψ) the tensor functional of Additive singular cohomology cross product and A=AW from Alexander–Whitney map and diagonal approximation, this is exactly J(φ,ψ)AΔ#. Indeed only the cut of bidegree (p,q) survives. Equivalently, define the AW external cochain by J(φ,ψ)A and pull it back along the diagonal. There is no extra cochain sign.

The earlier additive cohomology cross product was expressed through a specified shuffle inverse T. The comparison with it is an equality of classes: Alexander--Whitney and shuffle are natural chain-homotopy inverses constructs K with AT=dK+Kd. For cocycles, The additive singular cohomology cross product is well-defined gives Jd=0, hence J(AT)Δ#=JKdΔ#=δ(JKΔ#). Here postcomposition with the diagonal commutes with boundary, as proved in the diagonal definition. Thus the AW cup class equals diagonal pullback of the earlier external product; equality of the two chosen external cochains is not required. At total degree zero the displayed primitive is zero and the identity is literal. The same tensor-functional identity gives δ(JAΔ#)=0 for cocycles, so the compared classes exist.

For degree-zero cochains the formula on a vertex is ordinary multiplication of their values. A constant cochain of value 1 multiplies any cochain on either side without changing it, including on disconnected spaces. On empty X or over R=0 all cochains and products are zero. Face restrictions make sense for every singular simplex, including degenerate ones. A bare abelian coefficient group has no specified multiplication to insert in this formula. No AC or selection of representatives is used in this definition or comparison.

TheoremStatement: Literature-sourcedProof: AI-adaptedjudge pass (gpt-5.6-terra)audited 2026-09-13Open item page →

Cup product Leibniz identity

Statement

For cochains φCp(X;R) and ψCq(X;R), p,q0, over a commutative unital ring with positive coboundary, δ(φψ)=δφψ+(1)pφδψ. Thus the product of cocycles is a cocycle, and changing either cocycle representative by a coboundary changes their product by a coboundary.

Facts & Assumptions

[F1]

Singular cup product on cochains defines φψ=J(φ,ψ)DX, with DX a chain map.

[F2]

The additive singular cohomology cross product is well-defined proves δJ(φ,ψ)=J(δφ,ψ)+(1)pJ(φ,δψ) on the signed tensor complex.

Proof

Given: X,R,p,q,φ,ψ as stated. Negative cochain degrees are zero, and δ2=0.

1.1

Since DX=dDX, precomposing the tensor-functional identity with DX gives δ(φψ)=J(φ,ψ)DX=J(φ,ψ)dDX=(J(δφ,ψ)+(1)pJ(φ,δψ))DX. By the cup formula this is exactly the asserted identity. If both inputs are closed its right side is zero.

F1F2given
2.1

Now assume δφ=δψ=0. Let uCp1(X;R) and vCq1(X;R). Bilinearity expands the change to (φ+δu)(ψ+δv)φψ=δuψ+φδv+δuδv. Step 1.1, applied to each pair, identifies this as δ(uψ+(1)pφv+uδv). Indeed the respective other Leibniz terms contain δψ, δφ, or δ2v, and vanish. This proves simultaneous descent and, by setting u=0 or v=0, each separate descent.

F1step 1.1given
3.1

If p=0, then u=0 and its two terms in the primitive are absent; if q=0, then v=0 and its terms are absent. For p=q=0 both representatives are unchanged, while the Leibniz identity itself still holds, with sign +1. At zero input or zero coefficient ring the equality is zero by bilinearity. An empty space has zero cochains, and a point or a degenerate simplex satisfies the same chain-map and tensor identities. No representative, basis, or primitive was chosen from an arbitrary family: the primitive is the displayed expression in the given u,v. Thus no AC is used.

F1F2step 1.1step 2.1
DefinitionDefinition: Literature-sourcedProof: Not applicablejudge pass (gpt-5.6-terra)audited 2026-09-13Open item page →

Singular cohomology ring

Definition

For a space X and commutative unital coefficient ring R, put H(X;R)=n0Hn(X;R),[φ][ψ]=[φψ]. The groups and equality of representatives are those of Singular cohomology with coefficients. The multiplication descends through both quotient maps by Cup product Leibniz identity, including simultaneous changes. Extend to finite sums of homogeneous classes by distributivity; only finitely many summands occur, so each product belongs to the displayed direct sum.

This is the singular cohomology ring. To check the ring assertion before using it, let φ,ψ,η have degrees p,q,r. On an arbitrary (p+q+r)-simplex both parenthesizations of their cochain cup product evaluate to φ(σ[0,,p])ψ(σ[p,,p+q])η(σ[p+q,,p+q+r]). Restriction to a face and then a face of that face is restriction to the listed vertex block, and multiplication in R is associative. Thus cochain associativity holds exactly and descends to classes. Bilinearity gives the two distributive laws. Addition, additive inverses and zero come from the abelian quotient groups. Products of degrees p,q have degree p+q.

Its multiplicative unit is the class of the cochain 1C0(X;R) assigning 1R to every singular vertex. On any edge, δ1=1R1R=0. The front/back formula gives 1φ=φ=φ1 on every simplex. This includes all components at once: degree-zero cochains need not have finite support on the set of vertices or components.

If X is empty or R=0, the ring is the zero ring and its unit equals zero, as allowed by our unital-ring convention. The formula for three factors works also when any degree is zero and for degenerate simplices; for one-point X it has the same unit. Nothing here chooses a representative for every class: the quotient operation is defined by representative independence. No AC is used.

PropositionStatement: Literature-sourcedProof: AI-adaptedjudge pass (gpt-5.6-terra)audited 2026-09-13Open item page →

Cup product is natural, unital and associative

Statement

For continuous f:XY and a commutative unital ring R, pullback on singular cohomology satisfies f(ab)=fafb,f(1)=1. Cup product is associative already on cochains. Consequently f:H(Y;R)H(X;R) is a unital graded-ring homomorphism. Homotopic maps give the same homomorphism.

Facts & Assumptions

[F1]

Singular cup product on cochains is the front/back-face formula, bilinear over R.

[F2]

Singular cohomology ring supplies the quotient multiplication and the closed cochain assigning 1 to every vertex as unit.

[F3]

Homotopic maps induce equal maps in singular cohomology proves equality of pullbacks in every degree for homotopic maps.

Proof

Given: f:XY continuous, and homogeneous cochains φ,ψ,η of degrees p,q,r. Write fφ(σ)=φ(fσ).

1.1

Face restriction commutes with postcomposition. On a (p+q)-simplex σ, evaluation of f(φψ) is φ(fσ[0,,p])ψ(fσ[p,,p+q]), which is the evaluation of fφfψ. Precomposition is linear and commutes with positive coboundary because each face does; it therefore induces the same multiplicative equality on cocycle classes. On a vertex x, f1(x)=1(f(x))=1, giving the unit identity on classes.

F1F2given
2.1

On a (p+q+r)-simplex σ, both (φψ)η and φ(ψη) evaluate to φ(σ[0,,p])ψ(σ[p,,p+q])η(σ[p+q,,p+q+r]). Associativity in R makes the results equal. Bilinearity extends this identity to finite sums of homogeneous cochains; descent gives associativity in the cohomology ring. Together with step 1.1 this proves the graded-ring homomorphism claim.

F1F2step 1.1
3.1

If f and g are homotopic, [F3] gives equality on each homogeneous group. Every element of the graded direct sum has finite support, so the two ring homomorphisms are equal. Degree-zero factors in step 2.1 merely shorten blocks to vertices. The same formula covers degenerate simplices and points. Empty spaces and the zero ring give zero cochains and the stipulated zero-ring unit; any existing pullback still preserves this unit. All formulas are prescribed, so no AC is required.

F1F2F3step 1.1step 2.1
LemmaStatement: Literature-sourcedProof: AI-adaptedjudge pass (gpt-5.6-terra)audited 2026-09-13Open item page →

Factor reversal gives the commutativity chain homotopy

Statement

Let X be a space and R a commutative unital ring. Put A=AW:C(X×X;R)C(X;R)RC(X;R) and let τ(x,y)=(y,x). On homogeneous tensors put W(ab)=(1)pqba(a=p, b=q). Then WAτ# and A are naturally chain homotopic. In particular WDX and DX, with DX=AΔ#, are naturally chain homotopic. No AC is required.

Facts & Assumptions

[F1]

Alexander--Whitney and shuffle are natural chain-homotopy inverses provides natural maps A,S and specified natural homotopies for AS1 and SA1, over R.

[F2]

The singular chain cross product on generators expresses S as the signed sum of monotone lattice paths.

Proof

Given: The signed tensor differential d(ab)=dab+(1)padb. Take the supplied homotopies dU+Ud=AS1 and dV+Vd=SA1.

1.1

The coefficients of dba and bda in dW(ab) are respectively (1)pq and (1)pq+q. In Wd(ab) they are (1)p+p(q1) and (1)(p1)q, respectively. Each corresponding pair agrees modulo two, so dW=Wd. Also W2=1. Terms involving da or db in degree zero are absent, so the calculation includes p=0 and q=0.

given
1.2

A shuffle path has p horizontal and q vertical steps. Its permutation sign is (1)N, where N counts vertical steps occurring before horizontal steps. Interchanging the two types of steps replaces N by pqN, since each horizontal/vertical pair contributes to exactly one of the counts. It therefore changes the sign by (1)pq. The affine simplex of the swapped path is the original one followed by factor interchange. Matching paths bijectively in the finite shuffle sums gives τ#S=SW. This is a signed-shuffle calculation, not merely naturality for maps of the two factors.

F2given
2.1

Put B=WAτ# and L=WUW. These are natural (with simultaneous maps of X), and step 1.2 gives BS=WASW. Thus dL+Ld=W(AS1)W=BS1. The map B is a chain map by step 1.1, [F1], and the fact that postcomposition commutes with face boundary. Define K=LABV. Using the two supplied homotopy equations yields dK+Kd=(BS1)AB(SA1)=BA. This is an explicitly specified natural homotopy.

F1step 1.1step 1.2given
3.1

Since τΔ=Δ, precomposing step 2.1 with the chain map Δ# gives d(KΔ#)+(KΔ#)=WDXDX. In degree zero the twists have sign +1 and the maps agree on vertices. If X is empty or R=0, all maps have zero complexes as domain and codomain. Degenerate singular simplices and one-point spaces use the same shuffle paths and homotopies; nothing was normalized away. The construction uses finite signed sums and the supplied homotopies, not a selection of new fillers, so it is choice-free.

F1F2step 2.1
TheoremStatement: Literature-sourcedProof: AI-adaptedjudge pass (gpt-5.6-terra)audited 2026-09-13Open item page →

Singular cohomology is graded commutative

Statement

Let X be a space, R a commutative unital ring, aHp(X;R) and bHq(X;R), with p,q0. Then ab=(1)pqba. The defining cochain operation need not be graded commutative. No AC is required.

Facts & Assumptions

[F1]

Factor reversal gives the commutativity chain homotopy supplies H=KΔ# with dH+H=WDXDX, where W(xy)=(1)xyyx.

[F2]

Singular cohomology ring defines multiplication by taking the class of the front/back cup of representative cocycles.

Proof

Given: Cocycles φ,ψ representing a,b. On the tensor complex define J(xy)=φ(x)ψ(y) in bidegree (p,q) and zero in other bidegrees of total degree p+q.

1.1

The functional J is R-balanced by commutativity. On tensors of bidegree (p+1,q), Jd=δφ(x)ψ(y)=0; on tensors of bidegree (p,q+1), Jd=(1)pφ(x)δψ(y)=0. All other bidegrees contribute zero. Thus Jd=0. Evaluating [F1]'s identity yields JWDXJDX=JH=δ(JH). The right side is an explicit coboundary, with degree-(p+q1) primitive.

F1given
2.1

The functional JDX is φψ. In JWDX, only the original diagonal cut of bidegree (q,p) contributes, and its value is (1)pqφ(σ[q,,p+q])ψ(σ[0,,q]). Commutativity in R identifies this with (1)pq(ψφ)(σ). Step 1.1 and [F2] therefore give the asserted equality of classes.

F2step 1.1given
3.1

If p=q=0, the primitive in step 1.1 has negative degree and is zero; the two cochains agree by ordinary commutativity in R. If just one degree is zero, the sign is positive and the same homotopy calculation applies. Empty spaces, zero classes and the zero ring give zero products. All formulas include points and degenerate simplices. Only the two given cocycle representatives and the specified homotopy are used, so no choice principle is invoked.

F1F2step 1.1step 2.1
DefinitionDefinition: Literature-sourcedProof: Not applicablejudge pass (gpt-5.6-terra)audited 2026-09-13Open item page →

Relative cup product for an excisive triad

Definition

Let A,B be subspaces of X and R a commutative unital ring. Set U=AB and N=C(A;R)+C(B;R)C(X;R). Use the relative cochains of Relative singular cochain complex, so C(X,A;R) means cochains vanishing on C(A;R). For φCp(X,A;R) and ψCq(X,B;R), use the front/back formula of Singular cup product on cochains. A simplex wholly in A has its front face in A, and one wholly in B has its back face in B. Thus φψ vanishes on N and defines a functional on C(X;R)/N.

The Leibniz identity of Cup product Leibniz identity holds in these quotient-functional complexes. Its representative-change primitive also vanishes on N: each of its summands has a first factor vanishing on A and a second vanishing on B. Consequently this construction descends to Hp(X,A;R)RHq(X,B;R)Hp+q(HomR(C(X;R)/N,R)).

Assume now that A,B are open in U (in particular they may be open in X). The relative cup product has target Hp+q(X,U;R), using the following canonical comparison. By Cover-small chains for a two-open cover, N is exactly the small-chain complex of the cover U=AB. We need the following stronger data, not merely the abstract chain-homotopy-equivalence assertion of The cover-small inclusion is a chain homotopy equivalence. Let S be barycentric subdivision and let T satisfy 1S=dT+Td, as supplied by Subdivision is chain homotopic to the identity, Barycentric subdivision operator, and Subdivision prism homotopy. These operators preserve simplex images. For a simplex σ, let a(σ) be the least subdivision count making it small, whose existence follows from Finite chains eventually become cover-small, and recursively set m(σ)=max({a(σ)}{m(σδj):0jdimσ}), with m=0 on vertices. If σ is already small then m(σ)=0. Define Dσ=h=0m(σ)1TShσ and r=1dDDd. The telescoping identity shows that rσ is small: besides Sm(σ)σ, each face correction is a sum of TSh(σδj) with hm(σδj) and is therefore small. Thus r:C(U;R)N is a chain map, ri=1, 1ir=dD+Dd, and DN=0. The integral construction extends to R by tensoring, so all these identities remain valid and all operators preserve chains in U.

Here is why this gives the needed relative cohomology comparison, not just an absolute homology comparison. Extend D to a degree-one map E on C(X;R) by the same formula on simplices with image in U, and by zero on all other simplex generators. This is an unambiguous linear extension; it need not itself commute with d. Put P=1dEEd. Then P is a chain map, PN=1, and P(C(U;R))N. It therefore induces Pˉ:C(X;R)/C(U;R)C(X;R)/N. The quotient map q:C(X;R)/NC(X;R)/C(U;R) is a chain map. Since E preserves both N and C(U;R), it descends to each quotient; the identity 1P=dE+Ed shows that Pˉq and qPˉ are homotopic to the respective identities. Applying HomR(,R) preserves these explicit homotopy identities. Hence q:H(X,U;R)H(HomR(C(X;R)/N,R)) is an isomorphism. Define the relative cup product by composing the preceding product with (q)1. This inverse on cohomology is unique; auxiliary choices in an alternative small-chain comparison cannot affect it.

The same definition applies whenever a specified neighborhood or subcomplex replacement has separately proved this quotient-cochain comparison and its compatibility with the pair maps. Such a replacement is an additional hypothesis, not an assertion for every triad. In general a simplex in AB can meet both members without lying in either, so vanishing on N alone does not mean vanishing on C(AB).

If either relative class is zero, its product is zero by the displayed primitive. If A=B=, then N=0 and q is the identity, recovering absolute cup. If A=X or B=X, one source group is zero. Empty X, zero coefficients and point spaces satisfy the same formulas. Degree-zero factors evaluate at a vertex; negative-degree groups and their possible primitives are zero. All singular simplices, including degenerate ones, are retained. The small-chain operators use least subdivision counts, and the extension of D is prescribed on simplex generators. No AC is required.

PropositionStatement: Literature-sourcedProof: AI-adaptedjudge pass (gpt-5.6-terra)audited 2026-09-13Open item page →

Relative cup products are natural and connector-compatible

Statement

Relative cup products are natural for maps of excisive triples: if f:XX takes A into A and B into B, then f(uv)=fufvHp+q(X,AB;R). Use the excisive comparisons in the relative-product definition for both triples. In particular these exist when each pair of subspaces is open in its union.

Here are explicit domains for connector compatibility. Write i:AX, U=AB, and let A:Hk(A;R)Hk+1(X,A;R) be the pair connector. Let A,B:Hk(A,AB;R)Hk+1(X,U;R) be the connector obtained by extending to a cochain on X vanishing on B, taking its coboundary, and using the small-chain quotient comparison. Then A,B(xiy)=Axy(xHp(A;R), yHq(X,B;R)), and A,B(ixy)=(1)pxAy(xHp(X,B;R), yHq(A;R)). For B= these are the ordinary pair identities. All coboundaries have the positive sign convention; R is commutative unital. No AC is needed.

Facts & Assumptions

[F1]

Relative cup product for an excisive triad constructs the target through the canonical quotient C(X)/(C(A)+C(B))C(X)/C(AB), whose dual induces an isomorphism.

[F2]

Long exact sequence of a pair in singular cohomology defines A[a]=[δa~] by any extension and proves independence by changing lifts and cocycle representatives.

[F3]

Cup product Leibniz identity supplies the positive-coboundary Leibniz rule.

[F4]

Cup product is natural, unital and associative proves the termwise cochain pullback identity.

Proof

Given: The triples and coefficient ring as stated. Put N=C(A;R)+C(B;R) and Q=HomR(C(X;R)/N,R).

1.1

Pullback of a cochain vanishing on A vanishes on A, and similarly for B. It also maps Q to the primed quotient-functional complex. By [F4] it commutes with the cup formula on cochains. The quotient maps in [F1] commute with f#, since all maps are induced by the same map of chains on X. Their induced cohomology maps commute as well, and their inverses commute because they are isomorphisms. Transporting the cochain equality through these inverses proves relative naturality.

F1F4given
1.2

Restriction gives a termwise exact sequence 0QC(X,B;R)C(A,AB;R)0. To prove surjectivity, extend a cochain from simplices in A by zero off A. It already vanishes on simplices in AB, so the extension vanishes on B. Its kernel consists exactly of cochains vanishing on both A and B, namely Q. Restriction commutes with coboundary; the extension need not. For a relative cocycle z, a lift z~ has δz~Q. Changing lifts changes this by a Q-coboundary; changing z by δw and lifting w leaves the connecting class unchanged. These are exactly the lift calculations of [F2]. Composing this well-defined connecting class with [F1]'s inverse defines the displayed A,B.

F1F2given
2.1

For the first identity represent x by a cocycle α on A and y by a cocycle β on X vanishing on B. Extend α to α~ on X. The cochain α~β vanishes on B and restricts to αiβ, so it is an admissible lift for step 1.2. Its coboundary is δα~β because δβ=0. By [F2], δα~ represents Ax and vanishes on A. Hence the very same Q-cocycle computes both sides after applying [F1]'s comparison.

F1F2F3step 1.2
2.2

For the second identity represent x by a cocycle α on X vanishing on B, and y by a cocycle β on A. Extend β to β~ on X. Now αβ~ is a lift vanishing on B, and its coboundary is (1)pαδβ~, because δα=0. The last factor represents Ay and vanishes on A. Transporting this Q-cocycle gives precisely the asserted sign and factor order.

F1F2F3step 1.2
3.1

If B=, the restriction sequence in step 1.2 is the pair sequence and [F1]'s comparison is the identity. If A=, its right-hand complex is zero and both formulas have zero sides. If A=X, every A and the displayed relative target are zero. For p=0 the second sign is positive, and degree-zero cocycles/lifts require no negative cochain. Zero ring, empty space and zero inputs give zero identities. On point spaces these are the empty/full cases; degenerate simplices are included in each extension rule. All extensions can be the stated zero extensions, so no AC or family of arbitrary lift choices is used.

F1F2step 1.2step 2.1step 2.2
DefinitionDefinition: Literature-sourcedProof: Not applicablejudge pass (gpt-5.6-terra)audited 2026-09-13Open item page →

Cap product with cohomology written first

Definition

Let X be a space and R a commutative unital ring. For p,n0, φCp(X;R), and a singular n-simplex σ, define φσ={φ(σ[0,,p])σ[p,,n],np,0,n<p. Extend R-bilinearly in φ and the finite chains of Singular simplices and singular chain groups with coefficients. An R-linear cochain is a function on simplex generators; each value multiplies one specified back-face generator. Thus the formula is well-defined on finite formal chains and is R-balanced in its two inputs. It gives Cp(X;R)RCn(X;R)Cnp(X;R), with negative chain groups zero. This is the cap product with cohomology first: evaluate on the front face and retain the back face.

The face convention is the same as Singular cup product on cochains. In terms of its Alexander–Whitney diagonal DX, cap is the bidegree-(p,np) part of DX, followed by evaluation of the first factor by φ. There is no additional sign in this evaluation. Subsequent boundary and projection identities use this order.

For p=0, this multiplies a simplex by the value of φ at its first vertex; in particular the constant value-one cochain acts as the identity on chains. For p=n, it returns φ(σ) times the last vertex, a degree-zero chain, whose boundary is zero. For p>n it is zero by definition. On a degenerate simplex the same face formula applies; unnormalized chains retain these generators. Empty X, zero inputs and the zero ring give zero maps. The construction also applies to the higher singular simplices of a point and requires no AC.

TheoremStatement: Literature-sourcedProof: AI-adaptedjudge pass (gpt-5.6-terra)audited 2026-09-13Open item page →

Cap product boundary identity

Statement

For φCp(X;R), cCn(X;R), p,n0, over a commutative unital ring, (φc)=(1)p(φcδφc). Consequently cap induces an R-bilinear map Hp(X;R)RHn(X;R)Hnp(X;R). Negative chain groups are zero, and no AC is assumed.

Facts & Assumptions

[F1]

Cap product with cohomology written first evaluates on the front face, retains the back face, and is zero for p>n.

[F2]

Singular cochain complex with coefficients gives δφ=φ with positive sign.

[F3]

The singular boundary operator gives the alternating face boundary and zero degree-zero boundary.

Proof

Given: X,R,p,n,φ,c as stated. By bilinearity it suffices first to check the identity on a simplex σ.

1.1

Assume n>p. In φσ, deleting vertex ip yields (1)iφ(σ[0,,i^,,p+1])σ[p+1,,n]; deleting i>p yields (1)iφ(σ[0,,p])σ[p,,i^,,n]. In δφσ, all terms are of the first form, now indexed by 0ip+1. Subtracting cancels the terms with ip and leaves φ(σ[0,,p])((1)pσ[p+1,,n]+i=p+1n(1)iσ[p,,i^,,n]). Multiplying by (1)p gives the alternating boundary of the retained back face, with its first face having sign +1 and subsequent signs (1)ip. This is (φσ).

F1F2F3given
2.1

If n=p, cap is a zero-chain with zero boundary; both terms on the right are zero by the degree convention. If n<p, all terms are zero for the same reason. Step 1.1 also covers p=0<n: its deleted-initial-vertex term cancels against the first term of δφ, and the surviving term is the ordinary back-face boundary. The case p=n=0 was covered by n=p. Linearity extends these calculations to all chains.

F1F2F3given
3.1

For a cocycle φ and cycle c the boundary identity gives (φc)=0. If the cycle changes by b, then φb=(1)p(φb) is a boundary. If the cocycle changes by δu, u=p1, then applying the identity to u,c gives δuc=(1)p(uc), again a boundary. For p=0 there is no u, since negative cochains are zero. Applying these two calculations successively covers changes in both variables; cycles and cocycles stay closed under these changes. Bilinearity descends and then factors through the tensor product.

F1F2step 1.1step 2.1
4.1

Empty spaces, zero chains/cochains and the zero ring give zero maps. Degenerate simplex restrictions satisfy the same face identities and cancellations. Point spaces retain higher unnormalized chains, to which step 1.1 applies unchanged. The endpoint cases and zero output degrees were treated in step 2.1. Every primitive in step 3.1 is an explicit cap of the supplied u or b; no arbitrary selection or AC occurs.

F1step 1.1step 2.1step 3.1
PropositionStatement: Literature-sourcedProof: AI-adaptedaudited 2026-09-13Open item page →

Cap naturality and projection formula

Statement

For continuous f:XY, αHp(Y;R) and xHn(X;R), f(fαx)=αfx. For αHp(X;R), βHq(X;R) and xHn(X;R), (αβ)x=β(αx). Here R is commutative unital and cap evaluates the front face, retaining the back face. Both identities hold already on cochains and chains. No AC is needed.

Facts & Assumptions

[F1]

Cap product boundary identity proves that the front-evaluation/back-retention cap descends to cohomology and homology, including zero output degrees.

[F2]

Singular cup product on cochains supplies the front/back evaluation formula. Cup product is natural, unital and associative supplies its descent to cohomology together with naturality, the unit, and associativity.

Proof

Given: Cochain representatives a,b of degrees p,q and a singular n-simplex σ. For the first identity a is on Y; for the second both are on X.

1.1

If np, the left side of the first chain identity is f#(a(fσ[0,,p])σ[p,,n]), equal to a(fσ[0,,p])(fσ)[p,,n]. This is exactly af#σ, since postcomposition commutes with face restriction. If n<p, both sides are zero. Linearity gives the chain identity for every chain.

F1given
1.2

If np+q, evaluating the left side of the second identity yields a(σ[0,,p])b(σ[p,,p+q])σ[p+q,,n]. Capping first by a leaves the chain a(σ[0,,p])σ[p,,n]; capping by b gives the same scalar and back face by R-linearity. If n<p, the inner cap on the right is zero; if pn<p+q, its remaining degree np is less than q, so the outer cap is zero. In both cases the left side also vanishes. Thus the equality holds on every simplex and extends bilinearly.

F1F2given
2.1

For cocycles and cycles, [F1] and [F2] make every operation in steps 1.1–1.2 well-defined on the corresponding quotient classes. Passing the chain identities to classes proves the formulas. The case p=0 is initial-vertex multiplication; the case p+q=n retains the last vertex with no sign, and zero degrees in either factor need no alteration. Empty spaces, zero inputs/ring, point spaces and degenerate simplices use the same formulas. All maps and products are explicit, without AC.

F1F2step 1.1step 1.2
DefinitionDefinition: Literature-sourcedProof: Not applicablejudge pass (gpt-5.6-terra)audited 2026-09-13Open item page →

Relative cap products with quotient domains displayed

Definition

Let A,BX, R be a commutative unital ring, and put N=C(A;R)+C(B;R). The cohomology-first cap formula induces Cp(X,A;R)R(Cn(X;R)/Nn)Cnp(X,B;R),φ[c][φc]. For a simplex in A its front-face evaluation is zero; for a simplex in B its retained back face is in B. Thus changing c by a chain in either summand of N changes the output by zero modulo C(B;R). These are all relations in the quotient, and bilinearity respects the tensor relations.

Use Relative singular homology for relative cycles and boundaries. Suppose δφ=0 and cN. The identity of Cap product boundary identity gives (φc)=(1)pφc, which is zero modulo B by the preceding calculation. If c changes by b plus an element of N, the change in the output is the relative boundary (1)p(φb). If φ changes by δu with u vanishing on A, then δuc=uc+(1)p(uc). Its first term is zero modulo B because cN, so this too is a relative boundary. For p=0 the possible u is zero. This proves descent in both variables.

Under the excisive hypotheses of Relative cup product for an excisive triad, its explicitly constructed quotient-chain equivalence gives Hn(C(X;R)/N)Hn(X,AB;R). Transporting the chain construction through this canonical quotient map defines the relative cap product Hp(X,A;R)RHn(X,AB;R)Hnp(X,B;R). As with relative cup, openness of A,B in their union suffices; any other neighborhood/subcomplex replacement must supply the indicated compatible comparison. An arbitrary triad is not silently assumed excisive.

In particular, taking B= gives Hp(X,A;R)RHn(X,A;R)Hnp(X;R). Taking B=A and precomposing the homology input with the quotient map from absolute homology gives Hp(X,A;R)RHn(X;R)Hnp(X,A;R). For A= the general construction includes the usual action on relative homology of (X,B). If A=X the cohomology input is zero; if B=X the target is zero. Empty spaces, zero ring/inputs and point spaces cause no exception. When n<p output chains are zero, and when n=p they are vertex chains. Degenerate simplices obey the same containment tests. All operations and quotient comparisons are explicit, so no AC is required.

LemmaStatement: Literature-sourcedProof: AI-adaptedaudited 2026-09-13Open item page →

Relative singular product comparison for CW pairs

Statement

Let (X,A) and (Y,B) be CW pairs with their characteristic maps supplied, and let R be a commutative unital ring. Give products their ordinary product topologies and use unnormalized singular chains. Put U=(A×Y)(X×B). The natural shuffle map descends to a chain homotopy equivalence C(X,A;R)RC(Y,B;R)C(X×Y,U;R). There is a compatible Alexander–Whitney inverse through the quotient by C(A×Y;R)+C(X×B;R). Dualizing gives a cochain homotopy equivalence, and the induced relative external product is the relative cup product of the two projection pullbacks. These cohomology comparisons are natural in maps of pairs. No dimension bound, finite-rank chain hypothesis, or AC is required.

Facts & Assumptions

[F1]

Relative CW inclusions are cofibrations supplies HEP into every topological target with ordinary products, arbitrary CW dimension and supplied characteristic maps, without choice.

[F2]

The cover-small inclusion is a chain homotopy equivalence supplies a small-chain retraction r and homotopy D with 1ir=dD+Dd. Its proof constructs them from finite affine subdivision sums and least subdivision counts; hence each preserves the chains on every subspace.

[F3]

The prism operator of a homotopy and The prism triangulation has the stated oriented boundary give a prism P with dP+Pd=H1#H0#. Its formula preserves chains in a subspace which the homotopy preserves.

[F4]

Alexander--Whitney and shuffle are natural chain-homotopy inverses supplies natural AW and shuffle chain maps and natural homotopies for both inverse identities, over R and without choice.

[F5]

Relative singular cochain complex identifies relative cochains with Hom on the relative free chain complex, with positive coboundary. Relative cup product for an excisive triad defines the product first on the quotient by the sum of subspace chain complexes, and then uses a proved quotient-cochain comparison.

Proof

Given: Write W=X×Y, C(Z)=C(Z;R), N=C(A×Y)+C(X×B), E=C(W)/N, and Q=C(W)/C(U). The degreewise bases of these quotients are exactly the singular simplices not in their respective indicated subspace bases. Tensor differentials have the sign d(ce)=dce+(1)ccde.

1.1

Let TX=(X×{0})(A×I) with its subspace topology in X×I. Apply [F1] with target TX, initial map x(x,0) and prescribed homotopy (a,t)(a,t). These maps are continuous into the subspace because their ambient maps are continuous and land in it. The resulting rX:X×ITX fixes TX. Write its coordinates as (HX,hX). Then HX(x,0)=x, HX(a,t)=a, and the open set OX={x:hX(x,1)>0} contains A and satisfies HX(OX,1)A. Indeed a point of TX with positive second coordinate has its first coordinate in A. Repeat this construction for (Y,B) to obtain HY,OY. Only two applications of the choice-free HEP construction are involved. [F1, given] 1.2 Put M=C(A)C(Y)+C(X)C(B)C(X)C(Y). The quotient by M is canonically V=C(X,A)C(Y,B): its basis tensors are precisely pairs of simplices with the first not wholly in A and the second not wholly in B. This identification commutes with the tensor differential since faces in the subspace become zero on either side. Naturality in [F4], applied to each of the two inclusions of pairs of spaces, shows that AW maps N into M, shuffle maps M into N, and their homotopies preserve N and M on their respective sides. They therefore descend to maps a:EV, b:VE and homotopies ab1V, ba1E. This uses naturality of the homotopies as well as of the chain maps; it does not assume that an individual AW cut of a simplex in U belongs to M.

F4given
2.1

The homotopy H((x,y),t)=(HX(x,t),HY(y,t)) preserves both A×Y and X×B, and therefore U. The sets V1=U(OX×Y) and V2=U(X×OY) form an open cover of U: points in A×Y lie in V1, and points in X×B lie in V2. At time one, H maps V1 into A×Y and V2 into X×B. Put F=H1#:C(U)C(U), and let P be its prism. Thus F1=dP+Pd, with P(N)N by the explicit simplex formula in [F3].

F3step 1.1
3.1

Apply [F2] to this open cover of U, writing R=ir for the small-chain retraction regarded as an endomorphism of C(U). We have 1R=dD+Dd. Both R and D preserve chains in A×Y and in X×B: in the cited construction every affine term on a simplex has image inside that simplex's image, and taking finite sums, boundaries and least subdivision counts does not change this property. Hence they preserve N. Since R lands in cover-small chains, step 2.1 gives FR(C(U))N. Combining the two homotopy identities gives 1FR=(1R)+(1F)R=d(DPR)+(DPR)d. Consequently K=DPR preserves N and descends to a contraction k of J=C(U)/N: dk+kd=1J. This contraction is prescribed by the constructions and requires no basis selection.

F2step 2.1
4.1

Regard J as the subcomplex of E spanned by simplices wholly in U but not wholly in either of its two members. Extend k to a graded map e:EJE by zero on all the remaining simplex basis vectors. There is no claim that e commutes with d. The map T=1deed does commute with d and kills J, by step 3.1. Hence it factors as sq, where q:EQ is the canonical quotient and s:QE is a chain map. Since e lands in J, qT=q, so qs=1Q by surjectivity of q. Also sq=1deed. Thus q and s are chain homotopy inverses. With positive coboundary, precomposition by e obeys δ(ϕe)+(δϕ)e=ϕ(de+ed) in the corresponding degrees. This explicitly dualizes the homotopy equivalence, without any appeal to exactness of Hom.

F5step 3.1
5.1

The relative shuffle is qb:VQ, which is well-defined already on the original quotient tensors. Its homotopy inverse is as:QV: (as)(qb)=a(sq)bab1V and (qb)(as)=q(ba)sqs=1Q. Composing the homotopies just written gives actual chain homotopies, and precomposition dualizes each identity as in step 4.1. Since q,a,b come from natural inclusions, quotient maps and [F4], their induced cohomology maps are natural. The inverse of the isomorphism q on cohomology is unique and hence natural too: invert q in each commuting square. No natural choice of the auxiliary s is asserted or needed.

F4step 1.2step 4.1
6.1

For relative cocycles φCp(X,A;R) and ψCq(Y,B;R) define J(φ,ψ)(ce)=φ(c)ψ(e) on the (p,q) summand and zero on all other total-degree summands. Evaluation on the signed tensor differential gives δJ(φ,ψ)=J(δφ,ψ)+(1)pJ(φ,δψ). Thus cocycles give cocycles and changing a representative by a coboundary changes Ja by a coboundary; for a second-factor change the primitive is (1)pJ(φ,θ)a. On E the AW front/back formula for Ja is exactly prXφprYψ. By [F5] and step 4.1 the corresponding class in Hp+q(W,U;R) is (q)1[Ja]=[Jas]. This is the claimed relative external product and its compatibility with the cup construction. Pair pullbacks commute with the evaluations and AW, so step 5.1 also proves its naturality.

F4F5step 1.2step 4.1step 5.1
7.1

If either space is empty or R=0, all product complexes are zero. If A=X or B=Y, then N=C(U)=C(W) and V=Q=0. If A=B=, then N=C(U)=0 and q is the identity, recovering [F4]; with just one empty member the same open-cover and quotient formulas still apply. In degree zero the AW and shuffle identifications are the vertex-pair tensor identification, and prisms still have the top-minus-bottom boundary. A one-point factor retains all its unnormalized higher simplices. No step discards degenerate simplices. The only degree sums are finite sums along a fixed total degree; all negative chain groups are zero. Homotopy times zero and one were checked in steps 1.1–2.1. Finite formulas, least subdivision counts and extension by zero are specified throughout, so the argument uses no choice axiom even in unbounded dimensions.

F3F4step 1.1step 2.1step 3.1step 4.1step 5.1step 6.1
LemmaStatement: Literature-sourcedProof: AI-adaptedjudge pass (gpt-5.6-terra)audited 2026-09-13Open item page →

Relative cohomological Kunneth under finite free homology hypotheses

Statement

Assume AC. Let R be a commutative PID and (X,A),(Y,B) CW pairs with supplied characteristic maps. Suppose Hq(Y,B;R) is finite free over R for every q0. For every n0, relative external product gives an isomorphism p+q=nHp(X,A;R)RHq(Y,B;R)  Hn(X×Y,(A×Y)(X×B);R). The same conclusion holds instead when every Hp(X,A;R) is finite free. This is a hypothesis on relative homology, with no bound on the nonzero degrees and no finite-rank hypothesis on singular chain groups. Products use the ordinary topology. AC is used only for the algebraic sections and bases in the proof; the external product and its chain comparison require no choice.

Facts & Assumptions

[F1]

Relative singular product comparison for CW pairs gives a chain homotopy equivalence L:C(X×Y,(A×Y)(X×B);R)C(X,A;R)C(Y,B;R) through the AW quotient comparison. The relative external product is represented by J(φ,ψ)L, where J is tensor evaluation with no extra sign.

[F2]

A free PID complex decomposes into two-term cycle-boundary pieces supplies, under AC, free cycles and boundaries of an arbitrary-rank nonnegative free PID complex and sections of its boundary maps onto their images.

[F3]

Free modules are projective, with the exact choice boundary supplies a section of a surjection onto a free module by lifting its basis; a specified finite basis requires only finite choice.

[F4]

Relative singular cochain complex defines positive coboundaries and the relative singular-simplex basis. For coefficient ring R, extending an integer functional by R-linearity identifies these cochains with HomR(C(,;R),R): both are exactly arbitrary R-valued functions on the complementary simplex basis.

[F5]

The Axiom of Choice permits the arbitrary-rank PID choices and simultaneous sections and finite bases across all degrees.

Proof

Given: Put C=C(X,A;R), D=C(Y,B;R), Vj=Hj(D) and let V have zero differential. Both C and D are free in each nonnegative degree on the simplices not wholly in the subspace. Work first under the finite-free hypothesis on V.

1.1

For every j, choose sj:Bj1DDj with djsj=1 using [F2]. Set s0=0 and πj=1sjdj, which takes values in ZjD since djsjdj=dj. Let qj:ZjDVj be the homology quotient. By [F3], choose a section j:VjZjD of qj. Use [F5] for the simultaneous sections and for finite bases of all Vj. Define ηj=qjπj and bj=πjjηj. The image of bj lies in BjD because applying qj gives zero. Also η=1, d=0, and ηd=0: boundaries are fixed by π and killed by q. Thus :VD and η:DV are chain maps.

F2F3F5given
1.2

If Vq has finite basis e1,,em and dual coordinates ei, the evaluation map HomR(T,R)VqHomR(TVq,R) is an isomorphism for every R-module T. Its inverse takes w to i=1mw(ei)ei. For one composite, substitute v=iei(v)ei into w(tv); for the other substitute λ=iλ(ei)ei and use tensor bilinearity. Both substitutions give the identity. The finite sum is the exact use of finite rank, and T may have arbitrary rank or fail to be free.

given
2.1

Put hj=sj+1bj. Then dj+1hj=bj. Since dj lands in boundaries, πj1dj=dj and ηj1dj=0, so bj1dj=dj and hj1dj=sjdj. Therefore dh+hd=b+sd=1η,η=1. These are chain homotopy inverse identities, not only assertions about induced homology. They hold at degree zero with negative groups and maps set to zero.

step 1.1
2.2

In total degree n, HomR((CV)n,R) is the finite direct sum of HomR(CnqVq,R), 0qn. The differential preserves q because dV=0, and on each such complex it is the positive C coboundary. Apply step 1.2 to identify the fixed-q complex with m copies of HomR(C,R) shifted up by q. Its cycles and boundaries are determined coordinatewise, so its degree-n cohomology is Hnq(X,A;R)Vq. This also describes the image of each pure tensor: it is the class of its evaluation functional. For q=n the incoming degree-minus-one coordinate is zero, exactly as in relative H0. Summing along the finite diagonal proves p+q=nHp(X,A;R)VqHnHomR(CV,R).

F4step 1.2
3.1

On CD set H(cy)=(1)cch(y) for homogeneous c. Expanding its tensor differential, the dchy terms in dH+Hd have signs (1)c and (1)c1 and cancel. The remaining terms give c(dh+hd)y. Hence 1η and 1 are homotopy inverses between CD and CV. For any chain homotopy uv=dH+Hd, the cochain operator K(ϕ)=ϕH has degree minus one and satisfies δK+Kδ=(uv). This follows by evaluating on a chain: the two terms are ϕHd and ϕdH. Consequently these tensor homotopies and their duals need no exactness theorem for tensor or Hom.

F4step 1.1step 2.1
3.2

Applying the same dual computation to step 2.1 gives Hq(Y,B;R)Vq: λ represents [ληq], and the inverse is restriction along q. Indeed η=1 gives one inverse identity, and precomposition by h gives the other up to cochain homotopy. Since V has zero differential, its cohomology after Hom is exactly V.

F4step 1.1step 2.1
4.1

Compose L of [F1] with 1η. By step 3.1 this gives a chain homotopy equivalence from the product-pair chains to CV, and hence an isomorphism on dual cohomology. Combine it with step 2.2 and the identification in step 3.2. On representatives, φλ maps to [J(φ,λ)(1η)L]=[J(φ,λη)L]=[φ]×[λη], the actual relative external product by [F1]. Thus the isomorphism is the canonical product, independent of the bases and sections which proved its bijectivity. This identifies the map itself, rather than merely comparing abstract source and target modules.

F1step 2.2step 3.1step 3.2
5.1

Suppose instead Up=Hp(C) is finite free for every p. Apply the construction in steps 1.1 and 2.1 to C, obtaining ,η,h there. On CD use h1. The mixed terms in its homotopy identity have signs (1)c+1 and (1)c and cancel, yielding (1η)1. Reduce the dual complex to HomR(UD,R). For fixed p its differential is (1)p times the D coboundary, which has the same kernel and image because this sign is a unit. A finite basis ui of Up gives the inverse to evaluation by wiuiw(ui). The two substitutions in step 1.2 now give its inverse identities with the factors in this order. Taking finite degree diagonals and dualizing the deformation identifies its cohomology with p+q=nUpHq(Y,B;R). A representative maps through L to J(λη,ψ)L, the same ordered external product by [F1]. This proves the symmetric assertion with no appeal to commutativity of external product.

F1F4step 1.1step 1.2step 2.1step 2.2step 3.1step 4.1
6.1

Empty spaces or full subspaces make the corresponding relative complex zero and the displayed map the isomorphism between zero modules. Empty subspaces recover the absolute comparison for CW spaces. Rank zero in step 1.2 means the empty inverse sum; rank one gives one copy of the other complex. In degree zero the only summand is (p,q)=(0,0) and evaluation multiplies vertex values. Unnormalized degenerate simplices remain in the free chain bases; no finite-rank assertion about those bases occurs. Infinitely many nonzero Vq or Up cause no problem: every fixed total degree involves only finitely many, so no interchange of an infinite product with a tensor is used. A PID has 10; the zero-ring case is outside that hypothesis, though all the displayed groups would be zero. AC occurs in [F2]'s arbitrary-rank cycle/boundary freeness and sections and in step 1.1's simultaneous homology sections and finite bases; it is not invoked in [F1] or the product formula.

F1F2F5step 1.1step 1.2step 2.1step 2.2step 4.1step 5.1
TheoremStatement: Literature-sourcedProof: AI-adaptedjudge pass (gpt-5.6-terra)audited 2026-09-13Open item page →

Cohomological Kunneth cross product is a ring isomorphism

Statement

Let R be a commutative unital ring. Give the graded tensor product the multiplication (ab)(ab)=(1)ba(aa)(bb). For arbitrary spaces X,Y, external product a×b=prXaprYb gives a unital graded-ring homomorphism to H(X×Y;R). This multiplicativity requires no AC.

Assume in addition AC, that R is a PID, and that either every Hq(Y;R) or every Hp(X;R) is finite free over R. Then this homomorphism is an isomorphism.

For CW pairs (X,A),(Y,B) with supplied characteristic maps, relative external product is likewise a graded-ring homomorphism H(X,A;R)^RH(Y,B;R)H(X×Y,(A×Y)(X×B);R), where the hat denotes the displayed graded multiplication on the ordinary graded direct-sum tensor product, not a completion. Relative rings need not have units. This relative multiplicativity is choice-free for every commutative unital R. Under AC and the PID hypothesis it is an isomorphism if either every Hq(Y,B;R) or every Hp(X,A;R) is finite free. Neither additive assertion requires bounded dimension or finite-rank singular chain groups. Products have their ordinary product topologies; AC enters only in additive bijectivity.

Facts & Assumptions

[F1]

Cup product is natural, unital and associative gives cochain associativity and naturality, and the vertex-value unit.

[F2]

Singular cohomology is graded commutative gives the sign (1)pq when interchanging absolute classes.

[F3]

Cohomological Kunneth isomorphism under finite free hypotheses gives additive bijectivity for arbitrary spaces under the stated PID, AC and degreewise finite-free homology hypotheses, in either factor.

[F4]

Relative singular product comparison for CW pairs gives the quotient-cochain comparison for the CW product triad and identifies the actual relative external product with the cup of projection pullbacks. Relative cup product for an excisive triad defines the relative product through that comparison.

[F5]

Relative cohomological Kunneth under finite free homology hypotheses supplies additive bijectivity for CW pairs with exactly the stated relative homology hypothesis, in either factor.

[F6]

Factor reversal gives the commutativity chain homotopy gives a natural diagonal homotopy H with dH+Hd=WDD. Naturality for an inclusion ZW makes H(C(Z))C(Z)C(Z).

[F7]

Cup product Leibniz identity gives the positive-coboundary product rule and its explicit primitives for changing representatives.

[F8]

The Axiom of Choice supplies the arbitrary-rank PID splittings and simultaneous homology sections and bases used only in [F3] and [F5].

[F9]

Singular cup product on cochains gives the front/back formula. Alexander--Whitney and shuffle are natural chain-homotopy inverses gives AT=dK+Kd between AW and the additive supplier's shuffle inverse. The additive singular cohomology cross product is well-defined gives Jd=0 on cocycles, descent, and independence of inverse.

Proof

Given: First allow any commutative unital R. Every cohomology element and every tensor is a finite sum of homogeneous elements. For the relative calculation write W=X×Y, U=(A×Y)(X×B), and N=C(A×Y;R)+C(X×B;R).

1.1

The displayed tensor multiplication is well-defined by bilinearity and the R-balanced tensor relations, degree by degree. Its two three-factor parenthesizations have signs with exponents ba+(b+b)a and ba+b(a+a), respectively, which agree. Associativity of the two cup products in [F1] therefore gives associativity of this multiplication. In the absolute case 11 is its unit because both degrees are zero. The same argument works for the relative cup algebra once its multiplication is specified below; it asserts no unit there.

F1given
1.2

For absolute cocycles φ,ψ, the AW external cochain JA on X×Y is exactly prXφprYψ, by evaluating its single surviving (p,q) cut. The additive external cochain is JT. By [F9], JAJT=J(dK+Kd)=δ(JK), since Jd=0. Hence the external map in the statement is the very map of [F3], before any bijectivity assumption. For four absolute classes, write α=prXa, β=prYb, α=prXa, β=prYb. By [F1] and [F2], αβαβ=(1)baααββ=(1)baprX(aa)prY(bb). Thus the map is multiplicative, and 1×1=1 follows directly from the constant vertex cochains.

F1F2F9given
1.3

For a space pair (Z,T), cochains vanishing on C(T;R) form a differential graded subalgebra of C(Z;R). Its coboundary preserves vanishing since faces remain in T, and its cup product preserves vanishing since the front and back faces of a simplex in T remain there. By [F7], products of cocycles and all the representative-change primitives remain in this subalgebra. This defines the relative cohomology ring, with the associativity of [F1]; it is also [F4]'s relative product with the two subspaces both equal to T. Similarly E=HomR(C(W;R)/N,R) is a differential graded subalgebra, since a simplex in either product subspace has all faces in that same subspace. The inclusion q:C(W,U;R)E preserves cup products literally on cochains. It induces an isomorphism on cohomology by [F4], so that isomorphism and its inverse are ring homomorphisms.

F1F4F7given
2.1

Choose four relative cocycles representing a,b,a,b and denote their projection pullbacks by α,β,α,β in that order; here these letters mean cochains. Then f=αβ and f=αβ lie in E and represent q(a×b) and q(a×b) by [F4]. Apply [F6] to the tensor evaluation J(β,α) on C(W)C(W). As both inputs are closed, Jd=0 by the signed tensor calculation in [F9]. Thus (1)baαββα=δ(JH). Set k=JH. It vanishes on N: on a simplex in A×Y, naturality of H places both tensor factors in that subspace, where α vanishes; on a simplex in X×B, the same argument uses the vanishing of β. Hence kE and βα(1)baαβ=δk is an equality with a primitive in the required quotient complex.

F4F6F9step 1.3
3.1

Cochain associativity and the Leibniz rule now give ff(1)ba(αα)(ββ)=αδkβ=(1)aδ(αkβ). This primitive belongs to E: a simplex in either product subspace has its middle face in that same subspace, and k vanishes there by step 2.1. Naturality of the front/back formula identifies αα with the pullback of the relative product representatives for aa, and likewise for bb. Thus the equality in H(E) is exactly the desired multiplicative identity after applying q. Since q is an injective ring isomorphism by step 1.3, the identity holds in H(W,U;R). This also justifies using the relative cup products in the tensor algebra of step 1.1.

F1F4F7step 1.1step 1.3step 2.1
4.1

Now assume the PID and AC hypotheses of the appropriate additive assertion. In the absolute case apply [F3] to the map identified in step 1.2; in the relative case apply [F5] to the map in step 3.1. Each is bijective in every degree, and hence on the graded direct sum, since every element has finite degree support. A bijective multiplicative map has a multiplicative inverse: if x=f(u) and y=f(v) then f1(xy)=uv=f1(x)f1(y). Thus these are ring isomorphisms. The AC uses in [F8] are precisely freeness and sections for arbitrary-rank PID cycles/boundaries, and simultaneous sections and finite bases of the homology modules across all degrees. No choice assumption occurred in steps 1.1–3.1.

F3F5F8step 1.1step 1.2step 3.1
5.1

At degree zero the sign in any swap of two degree-zero factors is +1, and the negative-degree primitive in step 2.1 is zero; its identity is ordinary commutativity of vertex values. If only one middle degree is zero, the same homotopy gives the stated positive sign without deleting the other degree. Empty spaces or R=0 give zero products; the zero ring is allowed for multiplicativity but excluded by the PID hypothesis for bijectivity. Full subspaces A=X or B=Y give zero relative rings on both sides, and empty A,B recover the absolute CW case. A point factor retains its unnormalized degenerate chains and its vertex unit in the absolute case. Relative rings are not asserted unital when a subspace is nonempty. All face formulas retain degenerate simplices; zero classes give zero products by the representative-change primitives. The additive suppliers check zero-rank and rank-one homology, either finite-free factor and unbounded nonzero degrees. Only finite degree diagonals and finite sums are used, so no completed tensor product occurs.

F3F4F5F7step 1.1step 1.2step 1.3step 2.1step 3.1step 4.1
LemmaStatement: Literature-sourcedProof: AI-adaptedaudited 2026-09-13Open item page →

Local coordinate cup products generate top relative cohomology

Statement

Assume AC and let R=Z or F2. Let p,q1, W=Rp×Rq, and put A=(Rp{0})×Rq,B=Rp×(Rq{0}). For local generators upHp(Rp,Rp{0};R) and uqHq(Rq,Rq{0};R), the product pr1uppr2uqHp+q(W,W{(0,0)};R) is a generator. Normalize the integral local generator by restriction to the cube Cr=[1,1]r: it evaluates to +1 on its positively oriented relative cube class. With that normalization the displayed product is the positive generator for the ordered first p and last q coordinates. Over F2 use the coefficient-one generator. The open-complement relative cup product is compared below with actual finite CW pairs; no CW-subcomplex hypothesis on a punctured Euclidean space is assumed.

Facts & Assumptions

[F1]

Relative cohomological Kunneth under finite free homology hypotheses gives the additive relative isomorphism for CW pairs with degreewise finite-free relative homology, under AC.

[F2]

Relative singular product comparison for CW pairs identifies that map with the actual relative projection cup product and gives its quotient-chain AW/shuffle comparison.

[F3]

Relative cup products are natural and connector-compatible gives naturality for triples with their proved quotient-cochain comparisons, including open subspaces.

[F4]

Long exact sequence of a pair in singular cohomology and Naturality of the singular cohomology pair sequence give the exact sequence and its commuting restriction and connector squares.

[F5]

Homotopic maps induce equal maps in singular cohomology makes each supplied deformation retraction a cohomology isomorphism, without choice.

[F6]

Relative homology of consecutive CW skeleta and Oriented cellular chain group identify a CW cube relative to all its proper faces with one copy of the coefficients in its dimension, generated by its oriented characteristic disk, and zero in other degrees.

[F7]

Topological universal coefficient short exact sequence for cohomology applies to relative pairs and identifies its right map with evaluation under AC.

[F8]

Alexander--Whitney and shuffle are natural chain-homotopy inverses gives the two inverse homotopies. The singular chain cross product on generators and The singular chain cross product satisfies the boundary formula give the signed product triangulation and boundary cancellation. By naturality these maps and homotopies descend to the smaller relative quotient in [F2].

[F9]

The Axiom of Choice supplies the PID sections and simultaneous homology sections/bases in [F1] and the integral cycle projections in [F7].

Proof

Given: Use Cr=[1,1]r with coordinates in their written order, and let Cr be its boundary, where at least one coordinate has absolute value one. The face decomposition is a finite CW structure, with the boundary as its (r1)-skeleton. Faces are cubes of lower dimension; a radial homeomorphism to a Euclidean disk gives their characteristic disks.

1.1

For v0 put ρ(v)=maxivi. The homotopy ht(v)=((1t)+t/ρ(v))v is continuous on Rr0, stays nonzero, is the identity at t=0, and at t=1 lands on Cr. It fixes Cr for every t. Thus CrRr0 is a strong deformation-retract inclusion. The cube and Euclidean space are both nonempty and contractible by v(1t)v, and their inclusion induces a cohomology isomorphism by [F5]. A point has cohomology R in degree zero and zero above: its positive unnormalized cochain complex alternates zero and identity differentials.

F5given
1.2

By [F6], Hk(Cr,Cr;R) is R at k=r and zero elsewhere. With integral chains the same assertion gives a free integral homology group in just degree r. Apply [F7] with coefficients R: all its Ext terms vanish, since either the zero resolution or the length-zero identity resolution of Z computes them. Thus Hk(Cr,Cr;R) is R for k=r and zero elsewhere, and evaluation identifies its top group with HomZ(Z,R). Let vr be the class taking value 1R on the positively oriented cube generator.

F6F7given
2.1

We spell out the needed pair-sequence comparison. For a nonempty contractible space V and a nonempty subspace T, the initial restriction H0(V;R)=RH0(T;R) includes the constant functions and is injective. The sequence of [F4] therefore gives H0(V,T)=0, H1(V,T)=H0(T)/R, and Hk(V,T)Hk1(T) for k2, with the isomorphism given by the connector. Consequently a map of pairs between two such contractible ambient spaces, inducing a cohomology isomorphism on their subspaces, induces isomorphisms in all relative degrees: in degree one it respects the same constant subgroup, and in higher degrees use the natural connector square of [F4]. Apply this to step 1.1. The inclusion jr:(Cr,Cr)(Rr,Rr0) gives the claimed relative-cohomology isomorphism in every degree. When r=1, the subspace has two components and H0(T)/R=(RR)/R(1,1), so this argument includes that endpoint.

F4step 1.1
2.2

For the sign, represent the positively oriented relative cube class by its oriented simplex decomposition. One explicit decomposition, after rescaling to [0,1]r, has a simplex for each permutation π: its vertices are 0,eπ(1),eπ(1)+eπ(2),,(1,,1), with coefficient sgnπ. These simplices cover the cube, since the decreasing order of a point's coordinates gives its containing simplex and their differences give its nonnegative barycentric coordinates. Ties describe their common faces. Their signed boundaries cancel on shared interior faces and leave the oriented outer faces. The resulting relative cycle cr is the positive generator in [F6]: on the characteristic disk's interior each simplex has determinant sign sgnπ, compensated by its coefficient, so the decomposition subdivides that oriented disk with multiplicity one. Equivalently, restriction at a point with strictly ordered coordinates sees one oriented simplex with coefficient one. The characteristic orientation selects exactly this relative disk class in [F6]. This check concerns relative singular chains with their affine parameterizations, not a cellular cochain substituted into the cup formula.

F6F8step 1.2
3.1

Projection WRp identifies the pair (W,A) up to deformation with (Rp,Rp0): contract the second coordinate to zero, preserving A throughout. On absolute spaces and subspaces these are homotopy equivalences by [F5], so step 2.1 and [F4] give the same identification in relative cohomology. Similarly for (W,B) and the second factor. Inside the cube product K=Cp×Cq=Cp+q put A0=Cp×Cq and B0=Cp×Cq. Contracting the unused cube coordinate gives the analogous two pair identifications. The inclusions A0A and B0B induce cohomology isomorphisms: contract their unused coordinates and use the radial deformation of step 1.1 on the other coordinate. Thus restriction identifies Hp(W,A) with Hp(K,A0), taking pr1up to pr1vp when jpup=vp, and similarly for the other factor. This last assertion follows from literal commutation of projections with the inclusions and the pair naturality in [F4].

F4F5step 1.1step 1.2step 2.1
3.2

We have AB=W0 and A0B0=Cp+q. Using ρ(x,y)=max{maxixi,maxjyj} in step 1.1 gives a deformation retraction of W0 onto this product boundary. Step 2.1 therefore makes the bottom restriction H(W,W0;R)H(K,K;R) an isomorphism as well. The upper triple has open subspaces A,B and so has the open relative product of [F3]. The lower triple has the proved CW product comparison of [F2]. Although that lower triple is not open, the needed comparison with the upper product is direct: restriction from W to K carries cochains vanishing on A and B to cochains vanishing on A0 and B0, and it commutes literally with the Alexander--Whitney front/back formula. It also commutes with the quotient maps from the quotient by the sum of the two relative subcomplexes to the quotient by their union. The upper quotient comparison of [F3] and the lower quotient comparison of [F2] are isomorphisms, so their inverses commute with restriction as well. Thus the relative-cup square commutes, with all three restriction maps just proved isomorphisms.

F2F3step 1.1step 2.1
3.3

The relative shuffle of cpcq represents the positive generator cp+q of the product cube. In the interiors of the two chosen factor simplices, a shuffle interleaves their ordered edge directions; the determinant of the resulting ordered product directions is its shuffle permutation sign times the two factor determinant signs. The coefficients in cp,cq and in [F8]'s shuffle cancel exactly these signs. The product decomposition therefore has multiplicity one and positive orientation on every interior simplex, and its remaining boundary lies in A0B0 by the boundary rule. By the characteristic orientation argument of step 2.2, its relative class is [cp+q]. This is why the ordered factors give positive sign, rather than merely an unspecified unit.

F8step 2.2
4.1

The finite CW pairs (Cp,Cp) and (Cq,Cq) have the degreewise finite-free R-homology calculated in step 1.2. The coefficient rings Z and F2 are PIDs. Thus [F1] applies. In total degree p+q its source has just the single summand RvpRRvq, and it sends vpvq to the product of their two projection pullbacks by [F2]. This is a generator of Hp+q(K,K;R). The isomorphism square in step 3.2 transports this to the product in the statement. It already proves the generator assertion without an orientation-sign convention.

F1F2step 1.2step 3.1step 3.2
5.1

Choose relative cocycles φ,ψ representing vp,vq, with φ(cp)=ψ(cq)=1R. In the tensor of relative chain complexes, cp,cq are cycles and J=J(φ,ψ) satisfies Jd=0. Let a,b be the quotient AW and shuffle of [F2], with ab1=dL+Ld from [F8]. The relative product is the unique class whose pullback to the smaller quotient is Ja. Evaluating that class on the actual relative shuffle cycle therefore gives Jab(cpcq)=J(cpcq)+J(dL+Ld)(cpcq)=1R. Both homotopy terms vanish by the two cycle equations. Step 3.3 identifies this cycle with the positive cube generator, so the product is vp+q, not its negative over Z. The commuting square of step 3.2 and the defining normalization through jp+q prove the asserted positive local normalization.

F2F8step 1.2step 2.2step 3.2step 3.3step 4.1
6.1

The case p=1 or q=1 uses the explicit degree-one quotient by constants in step 2.1; the interval relative cycle has boundary [1][1] and the chosen dual evaluates to one. The integer normalization fixes the sign, while over F2 the two signs agree. Any other integral generator is the negative of the normalized generator or that generator itself, and the only nonzero field generator over F2 is the normalized one; bilinearity therefore proves the generator assertion for every allowed choice of the two generators. Zero classes have zero products by bilinearity. There are no empty Euclidean factors here, since p,q1; zero dimensions and the zero ring are outside this lemma's hypotheses. Both radial homotopy endpoints and their nonzero domains were checked in step 1.1. Shared or degenerate singular faces are retained, and all affine shuffle sums are finite. AC is used in [F9] only for [F1]'s PID splittings and homology sections/bases and [F7]'s integral cycle projections; the radial comparisons, naturality and orientation determinants introduce none.

F9step 1.1step 1.2step 2.1step 3.2step 4.1step 5.1
DefinitionDefinition: Literature-sourcedProof: Not applicablejudge pass (gpt-5.6-terra)audited 2026-09-13Open item page →

Cup length over a coefficient ring

Definition

Let X be nonempty and R a commutative unital ring. In positive degrees reduced and ordinary cohomology coincide; use the multiplication of Singular cohomology ring. Define a set of nonnegative integers by LR(X)={0}{k1:there exist ajHdj(X;R), dj>0 (1jk), a1ak0}. The cup length over R is clR(X)=supLR(X) in N{}. More explicitly it is if the set is unbounded, and otherwise its largest member, which exists for a bounded nonempty subset of N. If there are no nonzero products of positive length, it is 0. The inserted 0 is a convention and does not assert that an empty product is nonzero over the zero ring.

Associativity makes the unparenthesized finite product well-defined. By Singular cohomology is graded commutative, permuting homogeneous factors changes it by a unit sign, hence does not change whether it is zero. Positive degrees exclude padding a product with degree-zero units. If a product of length k is nonzero, every initial subproduct is nonzero, since multiplying a zero subproduct by the remaining factors would give zero. Thus LR(X) is downward closed.

Cup length is a homotopy invariant: Cup product is natural, unital and associative supplies graded ring pullbacks and their homotopy invariance. A homotopy equivalence and its inverse induce inverse graded ring maps, which carry nonzero products of positive-degree classes to nonzero products of the same length in both directions. Hence their sets LR agree.

A zero cohomology ring, in particular R=0, has cup length zero. Length one asks only for a nonzero positive-degree class. The value means that each finite bound is exceeded; it does not assert the existence of one infinite sequence with all products nonzero. Degrees and factors are chosen only for a particular finite witness, so no AC is used. No cup length for empty X is assigned by this definition.

PropositionStatement: Literature-sourcedProof: AI-adaptedjudge pass (gpt-5.6-terra)audited 2026-09-13Open item page →

Positive-degree cup products on a suspension vanish

Statement

For a nonempty CW complex X, its unreduced two-cone suspension ΣX, and a commutative unital ring R, every product of two positive-degree reduced cohomology classes is zero. In particular clR(ΣX)1. No AC is required.

Facts & Assumptions

[F2]

The exponential law: for a locally compact metric X and any spaces Z and Y, transposition is a bijection between C(X×Z,Y) and C(Z,C(X,Y)) with the compact-open topology applies with locally compact metric domain I=[0,1], arbitrary parameter space and arbitrary target. It makes a function on Z×I continuous precisely when its transpose ZC(I,Y) is continuous.

[F3]

Homotopic maps induce equal maps in singular cohomology gives homotopy invariance with arbitrary coefficients.

[F4]

Long exact sequence of a pair in singular cohomology identifies the kernel of restriction to a subspace with the image of its relative cohomology.

[F5]

Relative cup product for an excisive triad gives the product for two open subspaces and its canonical comparison to the union-relative target.

[F6]

Cup length over a coefficient ring defines cup length from nonzero finite positive-degree products.

Proof

Given: X,R as stated. Use the quotient map q:X×[0,1]ΣX of [F1].

1.1

We first justify homotopies on quotient cylinders. If r:EZ is quotient and a function h:Z×IY has continuous composite h(r×1), its paths are continuous by surjectivity of r. By [F2], the transpose upstairs is continuous and equals h^r. The quotient criterion makes h^ continuous, hence [F2] makes h continuous. This does not assume that an arbitrary product preserves quotient maps.

F2given
1.2

The singular cochain complex of a point has one copy of R in each nonnegative degree. The boundary of its unique degree-n simplex has coefficient i=0n(1)i, equal to 1 for positive even n and 0 for odd n. With positive coboundary the differential from degree k is therefore identity for odd k and zero for even k. Every positive-degree cocycle is consequently a coboundary, while H0()=R.

given
2.1

Let U=q(X×[0,2/3)) and V=q(X×(1/3,1]). Their inverse images are saturated open sets, so U,V are open, their restrictions of q are quotient maps, and they cover ΣX. On U set hs([x,t])=[x,(1s)t], a contraction to its lower apex. On V use ks([x,t])=[x,1(1s)(1t)], a contraction to its upper apex. Each formula is continuous before quotienting, is constant on the collapsed face, and stays in the indicated set. Step 1.1 proves that the descended homotopies are continuous, including at their apex and time endpoints.

F1step 1.1
3.1

Steps 1.2 and 2.1, together with [F3], give Hp(U;R)=Hq(V;R)=0 for p,q>0. Let aHp(ΣX;R) and bHq(ΣX;R), with positive degrees (equivalently reduced classes). Their restrictions to U,V respectively vanish. By [F4], there exist relative classes a~Hp(ΣX,U;R) and b~Hq(ΣX,V;R) mapping to a,b. This uses only two witnesses from exactness.

F3F4step 1.2step 2.1
4.1

Their [F5] relative product belongs to Hp+q(ΣX,UV;R)=Hp+q(ΣX,ΣX;R)=0, since the relative chain quotient is zero. Its image in absolute cohomology is ab: both are obtained by the same front/back formula and the quotient comparison commutes with the map to the empty subspace. Hence ab=0.

F5step 3.1
5.1

Every product of length at least two vanishes, by applying step 4.1 to the first two factors and associating the rest. Thus [F6] gives cup length at most one, allowing zero when all positive-degree classes vanish. This includes point X, disconnected X, and R=0. The nonempty hypothesis ensures both apices in the prescribed quotient; the separately stipulated empty suspension is outside this statement. Degree-zero factors are excluded, as they could be units. Cochains in step 1.2 and the relative construction retain degenerate simplices. The homotopies are explicit and the only selections in step 3.1 are finite, so no AC is used.

F6step 1.2step 2.1step 3.1step 4.1
LemmaStatement: Literature-sourcedProof: AI-adaptedaudited 2026-09-13Open item page →

Integral surface cup pairing from the oriented polygon

Statement

Assume AC. For the closed oriented genus-g surface Σg, g0, there are bases 1 of H0, a1,b1,,ag,bg of H1, and ω of H2, all with integral coefficients, and no higher cohomology. Normalize ω to evaluate to +1 on the oriented surface cycle. Then aibj=δijω,biaj=δijω,aiaj=bibj=0. All products of ω with a positive-degree class vanish, and 1 is the unit. The degree-one basis is dual to the ordered edges of the polygon word i=1g[Ai,Bi]. AC is used only through the current UCT supplier; the cup calculation uses actual singular cocycles and finite sums.

Facts & Assumptions

[F1]

Cellular homology computes singular homology identifies cellular and singular homology naturally for cellular maps.

[F2]

Cellular boundary from three consecutive skeleta defines the cellular differential by the pair connecting map and a relative quotient. Oriented cellular chain group identifies a chosen oriented characteristic disk with the generator of its cell summand.

[F3]

Long exact sequence of a pair gives the exact pair sequence; its connecting map sends a relative cycle to the class of its boundary.

[F4]

Contractible nonempty spaces have the homology of a point applies to a convex polygon by straight contraction.

[F5]

Topological universal coefficient short exact sequence for cohomology gives the natural evaluation exact sequence under AC.

[F6]

Singular cup product on cochains evaluates a product of one-cochains on a triangle as the value on its first edge times the value on its last edge, with positive coboundary.

[F7]

Cup product is natural, unital and associative supplies the unit and associativity, and Singular cohomology is graded commutative gives the general signed commutativity identity.

[F8]

The Axiom of Choice supplies the arbitrary-rank integral cycle projections and sections used by [F5].

Proof

Given: For g1, use an oriented convex 4g-gon P with vertices v0,,v4g1 in positive boundary order. Identify its edges in the order A1,B1,A11,B11,,Ag,Bg,Ag1,Bg1. The quotient is the standard oriented genus-g surface, with all boundary vertices identified to v. Write c for the polygon center, and read vertex subscripts modulo 4g. The case g=0 is the oriented sphere and is treated separately below.

1.1

Parameterize each paired boundary edge by its positive generator Ai or Bi, giving a singular loop in the quotient based at v. Put ϵk=+1 on the positive traversal of an edge in the polygon word and ϵk=1 on its negative traversal. Let rk be the radial singular edge from c to vk, followed by the quotient map. For ϵk=+1, let Tk have ordered vertices (c,vk,vk+1); for ϵk=1, use (c,vk+1,vk). The affine maps of these triangles into P followed by the quotient are genuine singular simplices. The last edge of either ordered triangle is the same positively parameterized loop Ek=Ai or Bi, so paired occurrences have literally identical last singular edges, rather than only homotopic ones. The two boundary formulas are Tk=Ekrk+1+rk(ϵk=1),Tk=Ekrk+rk+1(ϵk=1). Thus Z=kϵkTk has boundary kϵkEk: radial terms telescope, and every Ai,Bi occurs once with each sign. Consequently Z=0.

given
1.2

The positive polygon-boundary cycle is a generator of H1(P;Z). Indeed, the polygon boundary has 4g vertices and successively oriented edges; its cellular boundary sends the edge from vk to vk+1 to vk+1vk by [F2], so a one-chain is a cycle exactly when all its edge coefficients are equal, and there are no two-cells. The all-ones chain is therefore a primitive positive generator, with its sign fixed by the given boundary order.

F1F2given
2.1

Before taking the quotient, the signed fan of step 1.1 is a relative cycle Z~ of (P,P), with boundary the positive polygon-boundary cycle. The polygon contracts linearly to c, so [F4] gives H1(P)=H2(P)=0; explicitly the point singular complex has one generator in every degree with differential identity in positive even degrees and zero in odd degrees, hence no positive homology. Therefore [F3] identifies H2(P,P) with H1(P) by boundary. Since its connecting image is the primitive positive generator computed in step 1.2, [Z~] is precisely the positive relative generator, with no orientation sign left unspecified.

F3F4step 1.1step 1.2
3.1

In the surface CW structure there is one vertex, 2g oriented edges and one oriented two-cell. Each edge has coincident endpoints, so d1=0 by [F2]. Under the quotient map of pairs (P,P)(Σg,Σg1), the relative chain Z~ becomes Z modulo Σg1; hence step 2.1 and [F2] identify its relative class with the positive face generator. Its connecting boundary is zero by step 1.1, so d2=0. There are no higher cells. Applying [F1] gives H0(Σg)=Z,H1(Σg)=i(Z[Ai]Z[Bi]),H2(Σg)=Z, and zero higher homology. More precisely, [F1] applied to the one-skeleton gives H2(Σg1)=0; exactness of the pair sequence [F3] then injects H2(Σg) into H2(Σg,Σg1), while the vanishing connecting image just proved makes the positive face generator lie in its image. Since Z maps to that generator, H2(Σg)=Z[Z], with no unlicensed comparison normalization. The stated edge loops have their specified cellular edge coordinates by their relative characteristic-interval classes. The signed fan carries the given surface orientation: positive triangles agree with the polygon orientation, negatively ordered triangles have coefficient minus one, and the prescribed edge identifications glue opposite boundary orientations.

F1F2F3step 1.1step 2.1
4.1

Every homology group in step 3.1 is finite free. Its Ext term in [F5] is zero, by using the identity augmentation as a length-zero free resolution; for the zero group use the zero resolution. Hence evaluation identifies Hn(Σg;Z) with Hom(Hn(Σg;Z),Z) in every degree. Define ai,bi by the coordinate duals of [Ai],[Bi], and define ω by ω([Z])=1. These classes exist and are unique. In particular there are actual singular cocycles representing all these classes; no cellular cochain is being evaluated by a singular formula. Evaluation is injective in degree two, so a product there is determined by its value on Z.

F5step 3.1
5.1

Take any two singular one-cocycles u,w. Put ui=u(Ai), ui=u(Bi), wi=w(Ai), wi=w(Bi) and tk=u(rk). Since u(Tk)=0, the two boundary formulas of step 1.1 give tk+1=tk+ϵku(Ek). By [F6], the contribution of ϵkTk to (uw)(Z) is tkw(Ek) when ϵk=+1 and tk+1w(Ek) when ϵk=1. This uses the actual first edge of the ordered triangle in each case. Coincident quotient vertices do not make its radial or boundary singular edge constant. [F6, step 1.1, step 4.1] 5.2 For g=0, use S2=D2/D2 with one vertex and one oriented two-cell. The cellular complex has Z in degrees zero and two and zero in all other degrees, so [F1] and the length-zero resolution argument of step 4.1 give H0=H2=Z, H1=0 and all higher groups zero. The positive cell orientation of [F2] specifies the generator dual to ω; the empty list of degree-one classes has no asserted pair products, and ω2=0 by degree.

F1F2F5step 4.1
6.1

In the ith block Ai,Bi,Ai1,Bi1, write t for the initial radial value. The recurrence in step 5.1 gives successive values t,t+ui,t+ui+ui,t+ui,t. The four cup contributions are consequently twi+(t+ui)wi(t+ui)witwi=uiwiuiwi. The radial value returns to t at the block's end, and in any event it has canceled from the formula. Summing all blocks gives the full calculation [u][w],[Z]=i=1g(uiwiuiwi). It holds for any representatives of the two classes, since their values on the edge cycles are their homology evaluations.

step 5.1
7.1

Insert the coordinate duals from step 4.1 into step 6.1. For u=ai,w=bj the sum is δij; for u=bi,w=aj it is δij; for two a's or two b's it is zero. The injectivity of degree-two evaluation in step 4.1 gives all four asserted equalities. These signs also agree with [F7]'s graded commutativity. Every product involving ω and a positive-degree class lands above degree two and is zero by step 4.1. The vertex-value unit from [F7] gives the remaining products and associativity. Thus the listed bases and multiplication specify the entire ring.

F7step 4.1step 6.1
8.1

At g=1, step 6.1 is the single determinant u1w1u1w1, with the positive product sign. Zero or repeated degree-one inputs give zero by that formula, and reversing orientation negates Z and its normalized dual ω consistently. There is no empty surface or zero coefficient ring in this example. Radial edges are allowed to repeat as maps after the quotient, and all singular simplices, including degeneracies, are retained. AC occurs only in [F5]'s integral cycle projections and sections, as supplied by [F8]; the finite fan and its cocycle recurrence need none.

F8step 4.1step 6.1step 7.1

5 · Examples, counterexamples and false statements

None yet.

Sources