Loading field map
Algebraists often need to measure how far a sequence of maps is from being exact—how much information is lost when passing from a module to a quotient, or how many ways a given module can be built from smaller pieces. Homological algebra is the machinery built to answer such questions. Over the past century, the field has passed through four major frameworks, each of which redefined what the core objects are, what methods are legitimate, and what it means to compute or to understand.
The earliest homological algebra grew directly out of algebraic topology. Topologists had discovered that the homology groups of a space could be computed by building a chain complex of abelian groups and taking the quotient of cycles by boundaries. By the 1930s, algebraists realized that the same pattern—a sequence of groups connected by maps whose composition is zero—appeared in purely algebraic settings. The decisive systematization came in 1956 with Cartan and Eilenberg's book Homological Algebra, which extracted the algebraic essence of the topological constructions.
In this classical framework, the fundamental objects are chain complexes of modules over a ring. Given a module, one builds a projective resolution—an exact sequence of projective modules that surjects onto it—or an injective resolution that embeds it. Applying a functor such as Hom or tensor product to the resolution and then taking homology yields the derived functors Ext and Tor. For example, Ext¹(M, N) classifies all extensions of M by N, while Tor₁(A, B) measures the failure of the tensor product to be left exact. These invariants turned out to be powerful tools in commutative algebra, group theory, and algebraic geometry.
Classical homological algebra was limited, however, by its reliance on module categories. Every construction was tied to a specific ring, and the methods worked only because modules over a ring form a category with enough projectives and enough injectives. When mathematicians tried to apply the same ideas to sheaves on a topological space—objects that are not modules over a single ring—the classical framework broke down.
Grothendieck's 1957 Tôhoku paper solved that problem by axiomatizing the categorical properties that make homological algebra work. An abelian category is an additive category in which every morphism has a kernel and a cokernel, every monomorphism is a kernel of its cokernel, and every epimorphism is a cokernel of its kernel. These axioms guarantee that the category behaves enough like the category of modules that one can define chain complexes, homology, and derived functors inside it.
The power of the abelian-category framework is that it absorbs classical homological algebra as a special case: the category of modules over a ring is abelian, so all the old results survive. But the new framework also covers categories that are not module categories, such as the category of sheaves of abelian groups on a topological space, or the category of representations of a quiver. Grothendieck used abelian categories to develop sheaf cohomology, which became essential in algebraic geometry.
Abelian categories remain the default setting for most homological algebra today. They provide a stable environment: the category of chain complexes over an abelian category is itself abelian, and so is the category of functors from a small category into an abelian category. This stability means that the classical constructions—resolutions, derived functors, spectral sequences—all work uniformly across many different contexts.
Even within an abelian category, passing to homology loses information. Two chain complexes that are quasi-isomorphic—that is, their homology groups are isomorphic—may be very different as complexes, but the abelian framework treats them as interchangeable for the purpose of computing derived functors. Verdier, in his 1967 thesis under Grothendieck, proposed a remedy: instead of taking homology, work directly with the complexes and formally invert all quasi-isomorphisms. The result is the derived category, a triangulated category in which short exact sequences of complexes are replaced by exact triangles.
An exact triangle is a diagram A → B → C → A[1] where the composition of any two consecutive maps is zero, and the whole triangle induces long exact sequences in homology. This structure captures the information that a short exact sequence of complexes provides, but it does so without requiring the category to be abelian. In the derived category, one can define derived functors directly without choosing resolutions: the derived tensor product, for example, is simply the ordinary tensor product applied to complexes and then localized at quasi-isomorphisms.
The derived-category framework coexists with abelian categories rather than replacing them. Many computations are still done in the abelian setting, but the derived category provides a more flexible language for situations where the abelian structure is too rigid. In algebraic geometry, the derived category of coherent sheaves on a variety has become a central object of study, and in representation theory, derived categories of module categories classify tilting objects and cluster structures. The trade-off is that derived categories lack the ability to take kernels and cokernels of individual morphisms; exact triangles are a weaker structure than short exact sequences.
Quillen's 1967 book Homotopical Algebra introduced model categories as a general framework for doing homotopy theory in any category that has enough structure. A model category is a category equipped with three distinguished classes of morphisms—weak equivalences, fibrations, and cofibrations—satisfying axioms that allow one to invert the weak equivalences in a controlled way. The resulting homotopy category is obtained by localizing at the weak equivalences, just as the derived category is obtained by localizing at quasi-isomorphisms.
In fact, the derived category of an abelian category is a special case: the category of chain complexes carries a model structure in which the weak equivalences are the quasi-isomorphisms, the fibrations are the epimorphisms, and the cofibrations are the monomorphisms. The homotopy category of this model structure is exactly the derived category. But model categories go far beyond chain complexes. They provide a setting for homotopy theory in categories of topological spaces, simplicial sets, spectra, and differential graded algebras—none of which are abelian.
The model-category framework extends the derived-category insight to non-abelian settings. It also provides additional structure that derived categories lack: homotopy limits and colimits, which are well-behaved replacements for ordinary limits and colimits that respect weak equivalences. For example, the homotopy pullback of a diagram of chain complexes can be computed using a fibrant replacement, and the result is independent of the choices made. This extra structure has made model categories indispensable in stable homotopy theory, algebraic K-theory, and the study of derived algebraic geometry.
Today, all four frameworks remain active, and mathematicians move between them depending on the problem. Classical homological algebra is still the tool of choice for concrete computations in commutative algebra and group cohomology, where one works with explicit resolutions and spectral sequences. Abelian categories provide the default language for algebraic geometry and representation theory, where the objects of interest (sheaves, modules over a ring) naturally form abelian categories. Derived categories are the standard setting for studying derived equivalences, Fourier–Mukai transforms, and the homological mirror symmetry conjecture. Model categories are used whenever the category at hand is not abelian—for example, when working with simplicial rings or with spectra in stable homotopy theory.
What the leading frameworks agree on is that the core business of homological algebra is to understand the failure of exactness through the lens of chain complexes and their homotopical properties. They disagree on the right level of generality: abelian categories give a rich structure but exclude many interesting examples, while model categories include those examples but require more axioms and a steeper learning curve. Derived categories sit in between, offering a triangulated structure that is flexible enough for many purposes but less structured than a full model category.
A developing synthesis is the theory of ∞-categories, which aims to capture the homotopical information of a model category in a more intrinsic way. In an ∞-category, the homotopy category is replaced by a higher-categorical structure that remembers the coherence data lost in the triangulated setting. This approach is already reshaping derived algebraic geometry and stable homotopy theory, and it may eventually provide a unified framework that absorbs the insights of all four earlier frameworks.
The history of homological algebra is not a story of one framework replacing another, but of successive layers of generalization. Classical homological algebra works in module categories; abelian categories extend it to any setting with the right exactness properties; derived categories recover the information lost by passing to homology; model categories extend the same idea to non-abelian contexts. Each layer preserves the earlier ones as special cases, and each remains in active use today. The field's ongoing pluralism reflects the diversity of problems it serves: from computing the cohomology of a sheaf to constructing the stable homotopy category, homological algebra provides the tools to measure what exactness leaves behind.