On triangulated orbit categories

For decades, mathematicians knew how to take a category and fold it — identify objects under a repeating symmetry — but the folded object kept coming out broken, stripped of exactly the structure you needed. Like origami that loses its shape when you collapse it wrong. Sit on that failure for one beat. Then say: Bernhard Keller found exactly the right construction to make the fold work — and the payoff connected two fields that had no business touching each other. The setup is this. You have a triangulated category T, which is a category equipped with a shift operation and a notion of exact triangle, the backbone of homological algebra. You have an autoequivalence F, a symmetry you can apply repeatedly, and you want to form the orbit category T divided by F by identifying objects related by iterated applications of F. The projection functor sends every object to its equivalence class. Morphisms in this orbit category are represented by tuples of morphisms from T — a morphism from one orbit to another collects contributions from each copy in the orbit. The problem is that these tuples don't generally lift to single morphisms back in T, so there's no obvious way to produce triangles in T divided by F from triangles in T. Keller puts it plainly: the orbit category is certainly not trivially triangulated, and in most cases, it is impossible to give it a triangulated structure at all. He proves this by example. Take the algebra of dual numbers — k adjoin X, modulo X squared — and its bounded derived category of finite-dimensional modules. Form the orbit under the square of the suspension functor. The endomorphism ring of the trivial module k in that orbit category turns out to be the polynomial ring k in one variable u. The map that sends u to one plus u is injective — it's a monomorphism — but it has no left inverse in a polynomial ring. That's fatal. In any triangulated category, every monomorphism admits a left inverse. Contradiction. The orbit isn't triangulated. A second failure comes from the Kronecker quiver — two vertices, two arrows pointing the same direction — where the preprojective algebra of an infinite-dimensional path algebra creates the same obstruction. The lesson: folding is dangerous, and you need the right conditions before it can work. Those conditions are what Keller's main theorem provides. The setting is the bounded derived category of a hereditary abelian category H over a field k. Hereditary means objects have no higher-order self-interference — more precisely, extension groups vanish beyond degree one — so objects in the derived category decompose cleanly into shifted copies of indecomposables from H, and those decompositions are finite. Keller assumes H has the Krull-Schmidt property, meaning every object breaks uniquely into indecomposables, and that all morphism and extension spaces are finite-dimensional over k. These are the first ingredients. The autoequivalence F has to be standard — isomorphic to the derived tensor product with a bounded complex of bimodules that are projective on both sides. And F has to satisfy two finiteness conditions: for each indecomposable U of H, only finitely many of the objects F applied to U land back in H; and there's an integer N such that every indecomposable in the derived category has some F orbit representative among the first N suspensions of indecomposables of H. These hypotheses make the orbit category finite enough to control. Under them, Keller proves the orbit category inherits a canonical triangulated structure and the projection becomes a triangle functor. The mechanism is the real achievement. Keller's move is to avoid forcing triangulation directly onto the orbit category. Instead, he lifts everything one level up to a differential graded — or dg — enhancement, a richer version of the category where the morphism spaces themselves carry chain-complex structure, encoding higher homotopy information. He defines the dg orbit category B by taking the same objects but forming morphism complexes as a colimit along the iterates of F. The homotopy category of B recovers the orbit category you started with. From this dg orbit category, Keller builds a triangulated hull M — a triangulated ambient category generated by the representable functors — and embeds the orbit category inside it via the Yoneda construction. Now comes the key step. Given a morphism in the orbit category, you lift it to the hull M, take the cone there, and ask whether that cone lands back in the image of the orbit category. Using the finiteness hypotheses, Keller shows it does. The orbit category is closed under extensions inside M. Because it is extension-closed in a triangulated category, it inherits the triangulated structure from M. Cones of morphisms in the orbit are computed by lifting, coning, and projecting back down — and the machinery guarantees this round trip works. As a further payoff, the construction produces a triangle equivalence between the orbit category and the stable category of a Frobenius category, connecting to a classical source of triangulated structure. Why does any of this matter outside pure category theory? The answer is cluster algebras. Sergey Fomin and Andrei Zelevinsky had introduced cluster algebras in the early two thousands, a combinatorial framework built around exchange patterns — recursive mutation rules that generate families of rational functions from a small initial seed. Buan, Marsh, Reineke, Reiten, and Todorov were developing a representation-theoretic categorification of this combinatorics, and they needed the orbit category of the bounded derived category of a hereditary algebra to carry triangulated structure so that tilting mutations and exact triangles could be manipulated. That was the open question Keller's theorem answers. The Calabi-Yau property is what makes the connection precise. A triangulated category is Calabi-Yau of dimension d when its Serre functor — the functor nu satisfying the duality that the k dual of maps from X to Y is naturally isomorphic to maps from Y to nu of X — is isomorphic to the d-fold suspension. This is the algebraic analogue of Poincaré duality. Keller proves that for a quiver Q whose underlying graph is Dynkin type A, D, or E, the cluster category built from the bounded derived category of kQ is Calabi-Yau of dimension exactly 2. That 2 Calabi-Yau property is precisely the symmetry the cluster algebra exchange patterns require. The abstract orbit construction doesn't just become triangulated — it produces the exact categorical symmetry that underpins the mutation combinatorics in a completely different area of mathematics. The Dynkin diagrams bring a second, historically deeper example into focus. Preprojective algebras of Dynkin quivers have been central to representation theory since the nineteen eighties. For a Dynkin quiver Q, the category of projective modules over the preprojective algebra of Q is Calabi-Yau of dimension 1. The Coxeter numbers of the Dynkin graphs are explicit: for type An, it's n plus 1, for Dn it's 2 times n minus 1, and for the exceptional types E6, E7, and E8, the values are 12, 18, and 30. These are not abstract parameters — they govern the periodicity of the autoequivalence and determine the structure of the orbit. Keller shows that Auslander and Reiten had already seen a triangulated structure on these preprojective module categories in the late nineteen eighties. Their result, from papers in nineteen eighty-seven and nineteen ninety-six, was observed case by case. Keller's orbit-category theorem reveals it as a special instance of a general construction. The triangulated structure Auslander and Reiten noticed is the orbit category structure, and Keller's theorem is the machine that produces it canonically and explains why it exists. He also records a finer relationship: for a Calabi-Yau triangulated category of dimension d, the stable functor category built from it is Calabi-Yau of dimension three times d minus 1. Setting d equal to 1 gives dimension 2, recovering the known Calabi-Yau dimension of the stable module category of preprojective algebras of Dynkin quivers. One theorem, one construction. It resolves the questions Buan, Marsh, and Reiten raised about cluster categories. It resolves Asashiba's question about orbit categories. It subsumes Auslander and Reiten's classical observation about preprojective algebras. And it produces a supply of Calabi-Yau triangulated categories with exact, computable dimensions. The bridge between representation theory and cluster combinatorics that Keller built here is one researchers have been crossing ever since. This lecture was created by ennepō. Go to https://ennepo.ai to Discover, Create and Follow the latest research in your field. Read when you can. Listen when you want to.