A category organizes objects and maps. A ring organizes addition and multiplication. Under explicit hypotheses, an abelian category can be recovered from one ring, and suitable functors can be recovered from bimodules. This series proves the correspondence at three levels: categories, functors, and natural transformations.
1.1. The objective
Theorem 1.1. The correspondence to be proved
Let \(\mathcal A\) be a locally small abelian category with every set-indexed coproduct and a small projective generator \(P\). Put \(R=\operatorname{End}_{\mathcal A}(P)^{\mathrm{op}}\). Then
\[\mathcal A\simeq R\text{-}\mathrm{Mod},\qquad X\longmapsto\operatorname{Hom}_{\mathcal A}(P,X).\]Here \(R\text{-}\mathrm{Mod}\) is the category of all unital left \(R\)-modules. For unital rings \(R,S\), the category of additive, right-exact functors \(R\text{-}\mathrm{Mod}\to S\text{-}\mathrm{Mod}\) that preserve all set-indexed coproducts is equivalent to the category of \(S\)-\(R\) bimodules:
\[B\longmapsto (B\otimes_R-),\qquad \phi\longmapsto (\phi\otimes_R\mathrm{id}).\]Its inverse sends a functor \(F\) to \(F(R)\), with the right action constructed in chapter 9, and a natural transformation \(\eta\) to \(\eta_R\). Composition corresponds to tensor product; identity functors correspond to regular bimodules.
Alphabeta Math: Module-category reconstruction · Functor–bimodule correspondence
Worked example for Theorem 1.1
Counterexample: A small projective object need not reconstruct the category
This is a statement of our destination. Chapter 8 proves the first equivalence; chapter 9 proves the second, including its assertion about natural transformations; chapter 10 proves compatibility with composition and explains the equivalence criterion for rings. Chapters 2–7 define every additional term in the statement.
The hypotheses specify the classes being compared. We do not assert that every abelian category is a category of all modules, or that every functor is tensoring. In particular, a tensor functor is required to preserve cokernels and coproducts, but need not preserve kernels.
1.2. A concrete example
Example 1.2. Reduction modulo two
Let \(\mathrm{Ab}\) denote abelian groups with group homomorphisms, and let \(\mathbb F_2=\mathbb Z/2\mathbb Z\). For an abelian group \(M\), write \(2M=\{2m:m\in M\}\). The quotient \(M/2M\) consists of cosets \([m]=m+2M\), with \([m]+[n]=[m+n]\); equality means \(m-n\in2M\), which makes this addition well-defined. It is an \(\mathbb F_2\)-vector space: \(0\) acts by zero and \(1\) by the identity. A homomorphism \(f:M\to N\) induces \([m]\mapsto[f(m)]\), since \(f(2m)=2f(m)\). Thus
\[F:\mathrm{Ab}\longrightarrow\mathbb F_2\text{-}\mathrm{Mod},\qquad F(M)=M/2M\]is a functor: induced maps preserve identities and composition. For example,
\[F(\mathbb Z)=\mathbb F_2,\quad F(\mathbb Z^2)=\mathbb F_2^2,\quad F(\mathbb Z/3\mathbb Z)=0,\quad F(\mathbb Z/4\mathbb Z)=\mathbb F_2.\]For the third calculation, every residue class is twice another class; for the fourth, precisely the even classes lie in \(2M\). The bimodule describing \(F\) is \(\mathbb F_2\), with its left field action and right integer action \(b\cdot n=b\overline n\). Chapter 6 proves the natural isomorphism
\[\mathbb F_2\otimes_{\mathbb Z}M\cong M/2M,\qquad b\otimes m\longmapsto b[m].\]The injection \(\mathbb Z\xrightarrow{\times2}\mathbb Z\) becomes the zero map \(\mathbb F_2\to\mathbb F_2\), so this functor does not preserve injections.
Proposition 1.3. Two natural transformations
The natural endomorphisms of \(F(M)=M/2M\) are precisely the zero transformation and the identity transformation. These correspond to multiplication by \(0\) and \(1\) on the bimodule \(\mathbb F_2\).
Proof.
Let \(\eta:F\Rightarrow F\). Since \(\eta_{\mathbb Z}\) is a linear endomorphism of the one-dimensional space \(\mathbb F_2\), it sends \(1\) to a scalar \(a\in\mathbb F_2\). For every \(m\in M\), the homomorphism \(u_m:\mathbb Z\to M\), \(n\mapsto nm\), gives by naturality \(\eta_M([m])=F(u_m)(a)=a[m]\). Thus every component is multiplication by the same scalar. Conversely these scalar maps are linear and commute with every induced map. A bimodule endomorphism of \(\mathbb F_2\) is also multiplication by its value at \(1\), and both scalar maps respect the right integer action.
□Logical flowchart of the series
Each diagram shows one statement and its immediate prerequisites. Arrows point from prerequisites to the statement. Click a prerequisite to explore its diagram, or choose any of the series’ 81 statements below.
Four immediate prerequisites. Click a prerequisite to view its diagram.
1.3. Plan and conventions
- This chapter: objective and example.
- Limits and colimits: universal properties without elements of abstract objects.
- Addition and biproducts: the algebra of morphisms.
- Abelian categories: kernels, cokernels, exactness, and generators.
- Modules and bimodules: the concrete abelian categories.
- Tensor products: construction and the preservation properties we need.
- Group algebras: representations as modules.
- Generators: reconstruct an abelian category from a ring.
- Eilenberg–Watts: reconstruct functors and transformations from bimodules.
- Morita equivalence: composition and equivalences.
We assume objects, morphisms, functors, natural transformations, terminal and zero objects, zero morphisms, and the basic definitions of groups, rings, fields, homomorphisms, and group representations. We work with ordinary sets and the usual axiom of choice. For constructions in categories whose objects form a proper class, chapter 8 either uses supplied choices of coproducts, cokernels, and inverse images, or states the additional global choice convention explicitly. Rings have an identity, ring homomorphisms preserve it, and modules are unital. No commutativity is assumed unless stated.
Composition \(gf\) means \(g\circ f\). An isomorphism has a two-sided inverse. An equivalence consists of functors in both directions whose composites are naturally isomorphic to the identity functors. In abstract categorical proofs we use morphisms and universal properties, never elements of an unnamed object. Elements appear only in concrete algebraic examples and constructions. No general theory of monoidal categories is needed: chapter 6 proves the tensor identities used in chapter 10 directly.
Worked examples
Each example checks the hypotheses and the calculation for the linked statements. The links in the statements return here.
Worked example 1. The dictionary on abelian groups
Applies to: Theorem 1.1.
Take \(\mathcal A=\mathrm{Ab}\) and \(P=\mathbb Z\). A map \(\mathbb Z\to M\) is uniquely \(n\mapsto nm\), so evaluation at 1 identifies \(H(M)\) with \(M\). Lifting \(m\) across a surjection proves projectivity; testing maps on every \(m\) proves the generator property; the finite support of \(h(1)\) proves smallness. Abelian groups have arbitrary direct sums and the usual kernels and quotient cokernels. Thus all the reconstruction hypotheses hold, and \(R=\operatorname{End}(\mathbb Z)^{\mathrm{op}}=\mathbb Z\).
The bimodule \({}_{\mathbb F_2}\mathbb F_2{}_{\mathbb Z}\) gives \(F(M)=M/2M\): \(\bar a\otimes m\mapsto am+2M\) has inverse \(m+2M\mapsto\bar1\otimes m\), since \(\bar1\otimes2m=0\). In particular \(F(\mathbb Z/6)=\mathbb Z/2\). The same quotient formula shows preservation of cokernels and arbitrary direct sums, so this functor satisfies the stated hypotheses. Composing it with itself, viewed as an abelian-group functor, gives \(F(F(M))=F(M)\), corresponding to \(\mathbb F_2\otimes_{\mathbb Z}\mathbb F_2\cong\mathbb F_2\); tensoring with \(\mathbb Z\) gives the identity. These verify the object, composition, and identity parts of the dictionary in one case.
Further proof details
For completeness, the quotient functor’s hypotheses can be checked without assuming the dictionary. For a map \(f:A\to B\), its cokernel is \(C=B/f(A)\). The cokernel of \(A/2A\to B/2B\) is \(B/(f(A)+2B)\); the map \([b]\mapsto[b+f(A)]\) identifies this quotient bijectively with \(C/2C\), since its kernel is exactly \(f(A)+2B\). Thus cokernels are preserved. A residue tuple in \((\bigoplus M_i)/2(\bigoplus M_i)\) has finite support and is zero exactly when each coordinate lies in \(2M_i\). Sending it to its coordinate residues therefore gives the inverse of the coproduct comparison, including the empty case. On arrows \(F(f+g)([m])=[f(m)+g(m)]\), proving additivity. For composition, the balanced product \(\bar a\otimes\bar b\mapsto\overline{ab}\) identifies \(\mathbb F_2\otimes_{\mathbb Z}\mathbb F_2\) with \(\mathbb F_2\); its inverse sends \(\bar a\) to \(\bar1\otimes\bar a\). Both composites fix the generators.
Worked example 2. Testing a transformation on a concrete group
Applies to: Proposition 1.3.
For \(M=\mathbb Z\oplus\mathbb Z/6\), the quotient \(M/2M\) has basis \(e=(1,0)+2M\), \(f=(0,1)+2M\) over \(\mathbb F_2\). If \(\eta:F\Rightarrow F\) is natural, let \(a=\eta_{\mathbb Z}(\bar1)\in\mathbb F_2\). Naturality for each map \(\mathbb Z\to M\), \(1\mapsto m\), forces \(\eta_M(m+2M)=a(m+2M)\). In particular \(\eta_M(e)=ae\), \(\eta_M(f)=af\), ruling out a coordinate swap even though that is a linear endomorphism of this particular quotient. Conversely multiplication by either scalar commutes with every induced map. This checks both transformations and why there are no others.
Counterexamples and limits of the hypotheses
Each construction below refutes the explicitly stated stronger claim. It does not refute the related theorem with its original hypotheses.
Counterexample 1. A small projective object need not reconstruct the category
Related statements: Theorem 1.1.
Claim refuted. The generator assumption can be omitted from the reconstruction theorem.
Proof.
In \(\mathrm{Ab}\), take \(P=0\). Every map from \(P\) is zero, so it lifts across every epimorphism; thus \(P\) is projective. Both sides of its coproduct comparison are the zero group, so it is small. All other ambient hypotheses hold in \(\mathrm{Ab}\). But \(\operatorname{Hom}(0,\mathbb Z)=0=\operatorname{Hom}(0,0)\); moreover the distinct maps \(0,1:\mathbb Z\to\mathbb Z\) induce the same map on these zero Hom groups. The functor is not faithful and cannot be an equivalence, since an inverse functor and natural comparison isomorphisms would recover distinct arrows from their images. Here \(\operatorname{End}(0)\) is the zero ring; any unital module over it is zero because \(m=1m=0m=0\). Thus the proposed module category also cannot recover the nonzero object \(\mathbb Z\). The missing condition is exactly that \(P\) generate.
Comments