ProofComplete

Proof of the Universal Coefficient Theorem

We prove the Universal Coefficient Theorem for cohomology, which relates singular cohomology with arbitrary coefficients to integer homology via the algebraic functors Hom⁑\operatorname{Hom} and Ext⁑\operatorname{Ext}.


Statement

The Universal Coefficient Theorem states that for any space XX and abelian group GG, there is a natural split short exact sequence

0β†’Ext⁑1(Hnβˆ’1(X),G)β†’Ξ΄Hn(X;G)β†’hHom⁑(Hn(X),G)β†’00 \to \operatorname{Ext}^1(H_{n-1}(X), G) \xrightarrow{\delta} H^n(X; G) \xrightarrow{h} \operatorname{Hom}(H_n(X), G) \to 0


Proof

Step 1: Decomposition of the chain complex.

Let Cn=Cn(X)C_n = C_n(X) be the singular chain groups, Zn=kerβ‘βˆ‚nZ_n = \ker \partial_n, and Bn=imβ‘β€‰βˆ‚n+1B_n = \operatorname{im}\, \partial_{n+1}. Since CnC_n is a free abelian group (it has a basis of singular nn-simplices), and ZnβŠ†CnZ_n \subseteq C_n is a subgroup of a free abelian group, ZnZ_n is also free. Consider the short exact sequences:

0β†’Znβ†’ΞΉCnβ†’βˆ‚nBnβˆ’1β†’0(βˆ—)0 \to Z_n \xrightarrow{\iota} C_n \xrightarrow{\partial_n} B_{n-1} \to 0 \quad (*) 0β†’Bnβ†’jZnβ†’Hn(X)β†’0(βˆ—βˆ—)0 \to B_n \xrightarrow{j} Z_n \to H_n(X) \to 0 \quad (**)

Since Bnβˆ’1B_{n-1} is a subgroup of the free group Cnβˆ’1C_{n-1}, it is free, so (βˆ—)(*) splits.

Step 2: Apply Hom⁑(βˆ’,G)\operatorname{Hom}(-, G) to (βˆ—)(*).

Since (βˆ—)(*) splits, applying Hom⁑(βˆ’,G)\operatorname{Hom}(-, G) gives an exact sequence: 0β†’Hom⁑(Bnβˆ’1,G)β†’βˆ‚nβˆ—Hom⁑(Cn,G)β†’ΞΉβˆ—Hom⁑(Zn,G)β†’00 \to \operatorname{Hom}(B_{n-1}, G) \xrightarrow{\partial_n^*} \operatorname{Hom}(C_n, G) \xrightarrow{\iota^*} \operatorname{Hom}(Z_n, G) \to 0

Now Cn(X;G)=Hom⁑(Cn,G)C^n(X; G) = \operatorname{Hom}(C_n, G) and the coboundary Ξ΄nβˆ’1\delta^{n-1} factors through Bnβˆ’1B_{n-1}. We identify:

  • Zn(X;G)=ker⁑δn=ker⁑(βˆ‚n+1βˆ—βˆ˜βˆ’)=Hom⁑(Zn,G)Z^n(X; G) = \ker \delta^n = \ker(\partial_{n+1}^* \circ -) = \operatorname{Hom}(Z_n, G) (via ΞΉβˆ—\iota^*)
  • Bn(X;G)=im⁑ δnβˆ’1β‰…Hom⁑(Bnβˆ’1,G)/kerβ‘βˆ‚nβˆ—B^n(X; G) = \operatorname{im}\, \delta^{n-1} \cong \operatorname{Hom}(B_{n-1}, G) / \ker \partial_n^*

More precisely, using the identification Znβ‰…Hom⁑(Zn,G)Z^n \cong \operatorname{Hom}(Z_n, G) under ΞΉβˆ—\iota^*, the coboundaries correspond to those homomorphisms Znβ†’GZ_n \to G that extend to Cnβ†’GC_n \to G via βˆ‚nβˆ—\partial_n^*, which by the splitting means Bnβ‰…im⁑(jβˆ—:Hom⁑(Zn,G)β†’Hom⁑(Bn,G))B^n \cong \operatorname{im}(j^* : \operatorname{Hom}(Z_n, G) \to \operatorname{Hom}(B_n, G)) pulled back.

Step 3: Apply Hom⁑(βˆ’,G)\operatorname{Hom}(-, G) to (βˆ—βˆ—)(**).

The sequence (βˆ—βˆ—)(**) gives rise to the exact sequence: 0β†’Hom⁑(Hn,G)β†’Hom⁑(Zn,G)β†’jβˆ—Hom⁑(Bn,G)β†’Ext⁑1(Hn,G)β†’00 \to \operatorname{Hom}(H_n, G) \to \operatorname{Hom}(Z_n, G) \xrightarrow{j^*} \operatorname{Hom}(B_n, G) \to \operatorname{Ext}^1(H_n, G) \to 0

since ZnZ_n is free (so Ext⁑1(Zn,G)=0\operatorname{Ext}^1(Z_n, G) = 0).

Step 4: Assemble the result.

We have Hn(X;G)=Zn/Bnβ‰…Hom⁑(Zn,G)/im⁑(jβˆ—)restrictedH^n(X; G) = Z^n / B^n \cong \operatorname{Hom}(Z_n, G) / \operatorname{im}(j^*)_{\text{restricted}}. From Step 3, we can identify the kernel and cokernel to obtain:

0β†’coker⁑(jnβˆ’1βˆ—)β†’Hn(X;G)β†’ker⁑(jnβˆ—Β restricted)β†’00 \to \operatorname{coker}(j^*_{n-1}) \to H^n(X;G) \to \ker(j^*_n \text{ restricted}) \to 0

which yields exactly:

0β†’Ext⁑1(Hnβˆ’1(X),G)β†’Hn(X;G)β†’hHom⁑(Hn(X),G)β†’00 \to \operatorname{Ext}^1(H_{n-1}(X), G) \to H^n(X; G) \xrightarrow{h} \operatorname{Hom}(H_n(X), G) \to 0

The map hh sends a cocycle Ο†:Cnβ†’G\varphi : C_n \to G to its restriction Ο†βˆ£Zn\varphi|_{Z_n}, which factors through Hn=Zn/BnH_n = Z_n / B_n precisely when Ο†\varphi annihilates BnB_n.

Step 5: Splitting.

Since ZnZ_n is free, the sequence (βˆ—βˆ—)(**) admits a section s:Znβ†’Zns : Z_n \to Z_n (identity) factoring through a map Hnβ†’ZnH_n \to Z_n. Choose a splitting Οƒ:Hn(X)β†’Zn\sigma : H_n(X) \to Z_n of (βˆ—βˆ—)(**). Then Οƒβˆ—:Hom⁑(Zn,G)β†’Hom⁑(Hn,G)\sigma^* : \operatorname{Hom}(Z_n, G) \to \operatorname{Hom}(H_n, G) provides a right inverse for hh, splitting the UCT sequence.

The splitting depends on the choice of Οƒ\sigma, hence is not natural. β–‘\square

β– 

RemarkNon-naturality of the splitting

While the short exact sequence itself is natural in XX, the splitting is not. Different choices of chain-level representatives lead to different splittings. This non-naturality has consequences: one cannot, in general, compute fβˆ—f^* on cohomology purely from fβˆ—f_* on homology without additional information.