Derived Categories and Triangulated Categories

Triangulated categories are additive categories equipped with a shift functor and a class of "distinguished triangles" satisfying four axioms (TR1–TR4), abstracting the homotopy-theoretic notion of cofiber sequences. Derived categories are the principal example: given an abelian category

A

, the derived category

D (A)

is obtained from the category of cochain complexes by formally inverting quasi-isomorphisms, providing the natural home for derived functors, Ext, Tor, and duality theory. Used pervasively in algebraic geometry (coherent sheaves, Fourier–Mukai transforms), representation theory (perverse sheaves, tilting), and homological algebra. Foundations laid by Grothendieck and Verdier (c. 1963–67).

A derived category $D (A)$ is the localization of the category of cochain complexes $Ch (A)$ at the class of quasi-isomorphisms — maps inducing isomorphisms on all cohomology. This construction "remembers" complexes up to quasi-isomorphism, so that a module and its projective resolution become isomorphic objects. Derived categories are not abelian, but they carry the weaker structure of a triangulated category: a replacement for the notion of short exact sequence by distinguished triangles $X \to Y \to Z \to X [1]$ . The surprising and deep fact is that this triangulated structure, despite being axiomatically weak (the octahedral axiom is notoriously non-canonical), suffices to develop a rich theory of derived functors, spectral sequences, Serre duality, and equivalences between geometric objects that look nothing alike (e.g. an abelian variety and its dual). Triangulated categories are the lingua franca uniting homological algebra, stable homotopy theory, and modern algebraic geometry.

The Core Insight

The "right" objects of homological algebra are not modules but complexes considered up to quasi-isomorphism. Short exact sequences are too rigid for derived functors; they must be replaced by distinguished triangles, which encode "the cofiber" rather than "the cokernel." A long exact sequence in cohomology is then not a theorem about exact sequences but a definitional consequence of the triangle axioms. The slogan: replace exactness with the mapping cone.

Formal Statement

Cochain Complexes and Quasi-isomorphisms

Let $A$ be an abelian category. A cochain complex is a sequence

\dots \to X^{n - 1} d^{n - 1} X^{n} d^{n} X^{n + 1} \to \dots, d^{n} \circ d^{n - 1} = 0.

⟳the condition

d^{2} = 0

is exactly what makes the cohomology objects

H^{n} (X) = ker d^{n} / im d^{n - 1}

well-defined.

The cohomology is

H^{n} (X) = \frac{ker ( d ^{n} : X ^{n} \to X ^{n + 1} )}{im ( d ^{n - 1} : X ^{n - 1} \to X ^{n} )} .

⟳well-definedness uses

im d^{n - 1} \subseteq ker d^{n}

, which is equivalent to

d^{n} d^{n - 1} = 0

A morphism $f : X \to Y$ is a quasi-isomorphism if $H^{n} (f) : H^{n} (X) \to H^{n} (Y)$ is an isomorphism for all $n$ . The derived category is the localization

D (A) = Ch (A) [qis^{- 1}],

the universal category receiving a functor from

Ch (A)

that sends every quasi-isomorphism to an isomorphism.

⟳localization is defined by the universal property; existence requires a set-theoretic or calculus-of-fractions argument (see below).

The Shift Functor and Mapping Cone

The shift (translation) functor $[1]$ on complexes is

(X [1])^{n} = X^{n + 1}, d_{X [1]}^{n} = - d_{X}^{n + 1} .

⟳the sign

- d

is forced so that

[1]

is a chain map functor and so cone sequences compose correctly; it is the homological "Koszul sign."

Given $f : X \to Y$ , the mapping cone $Cone (f)$ is the complex

Cone (f)^{n} = X^{n + 1} \oplus Y^{n}, d = (- d_{X}^{n + 1} f^{n + 1} 0 d_{Y}^{n}) .

⟳this is the total complex of the double complex

X f Y

; the signs make

d^{2} = 0

given

df = f d

There is a canonical sequence $X f Y i Cone (f) p X [1]$ , whose image in $D (A)$ is the prototype of a distinguished triangle.

Triangulated Category — Axioms TR1–TR4

A triangulated category is an additive category $T$ with:

an autoequivalence (shift) $[1] : T \to T$ ;
a class of distinguished triangles $X \to Y \to Z \to X [1]$ ,

satisfying:

(TR1) Every morphism

f : X \to Y

embeds in a triangle

X f Y \to Z \to X [1]

; the triangle

X id X \to 0 \to X [1]

is distinguished; and any triangle isomorphic to a distinguished one is distinguished. (TR2) (Rotation)

X f Y g Z h X [1]

is distinguished iff

Y g Z h X [1] - f [1] Y [1] is distinguished.

⟳rotation makes the triangle "self-similar"; the sign

- f [1]

is the analogue of the cone sign and is required for consistency under repeated rotation.

(TR3) (Morphism of triangles) Given two triangles and a commuting square on the first two terms, there exists a (not necessarily unique) fill-in on the third:

X ↓ u X^{'} \to \to Y ↓ v Y^{'} \to \to Z ↓ \exists w Z^{'} \to \to X [1] ↓ u [1] X^{'} [1]

⟳non-uniqueness of

w

is the source of the failure of

T

to be abelian; cones are weak (co)limits.

(TR4) (Octahedral axiom) For composable

X f Y g Z

, the cones of

f

g

, and

g f

fit into a fourth distinguished triangle relating them.

$D(\mathcal{A})$ is Triangulated

For any abelian category $A$ , the derived category $D (A)$ — and likewise the homotopy category $K (A)$ — is a triangulated category, with shift $[1]$ as in (4) and distinguished triangles defined as those isomorphic (in $D (A)$ ) to mapping-cone triangles $X \to Y \to Cone (f) \to X [1]$ . The cohomology functors $H^{n} : D (A) \to A$ are cohomological: each distinguished triangle yields a long exact sequence

\dots \to H^{n} (X) \to H^{n} (Y) \to H^{n} (Z) \to H^{n + 1} (X) \to \dots .

⟳apply

H^{*}

to the cone short exact sequence

0 \to Y \to Cone (f) \to X [1] \to 0

of complexes and take the snake-lemma connecting map.

Why It Holds / Derivation

Construction of $D(\mathcal{A})$ via Calculus of Fractions

We construct the localization concretely.

Step 1 — Pass to the homotopy category. Define

K (A)

with the same objects as

Ch (A)

but morphisms = chain maps modulo chain homotopy. Homotopic maps induce equal maps on cohomology, so quasi-isomorphisms are well-defined in

K (A)

. Crucially,

K (A)

is already triangulated (cones are defined up to homotopy). Step 2 — Quasi-isomorphisms form a multiplicative system. The class

S

of quasi-isomorphisms in

K (A)

admits a calculus of (left and right) fractions:

$S$ contains identities and is closed under composition;
(Ore condition) any diagram $X^{'} s X f Y$ with $s \in S$ completes to a square with the new map in $S$ ;
if $s f = s g$ for $s \in S$ , then $f, g$ are equalized by some $t \in S$ .

The triangulated structure is exactly what makes these conditions hold: given $s$ , embed it in a triangle and use TR3 to produce the completion.

Step 3 — Define morphisms as roofs. A morphism

X \to Y

D (A)

is an equivalence class of roofs

X s X^{'} f Y, s \in S,

with two roofs identified if they are dominated by a common third roof.

⟳the Ore condition guarantees roofs compose; the third axiom guarantees the equivalence relation is well-behaved.

Step 4 — Localization inherits triangulation. A triangle in

D (A)

is declared distinguished iff isomorphic to the image of a distinguished triangle in

K (A)

. One checks TR1–TR4 descend because

S

is compatible with the triangulation (a thick/épaisse localizing system in Verdier's sense).

■

The Octahedral Axiom from Mapping Cones

Let $X f Y g Z$ . Form cones $C_{f} = Cone (f)$ , $C_{g} = Cone (g)$ , $C_{g f} = Cone (g f)$ .

Step 1. There are natural maps

C_{f} \to C_{g f}

(induced by

g

on the

Y

-component) and

C_{g f} \to C_{g}

(induced by

f

on the

X [1]

-component). Step 2. These assemble into a candidate triangle

C_{f} \to C_{g f} \to C_{g} \to C_{f} [1] .

⟳write all three cones as explicit two-row complexes; the maps are the evident projections/inclusions, and the connecting map

C_{g} \to C_{f} [1]

comes from the

Z

-component composed with

p_{f} [1]

Step 3. A direct (if tedious) verification with the cone differentials of (5) shows (10) is the mapping cone of the map

C_{f} \to C_{g f}

, hence distinguished. This is the octahedron: the four triangles (on

f

g

g f

, and (10)) are its four "upper/lower faces."

The non-canonicity of TR3's fill-in is what makes a clean abstract proof impossible without these explicit models; this is the historical reason the axiom is considered the subtle one. $■$

The octahedral axiom: cones of $f$, $g$, $gf$ and their interlocking triangles

Jean-Louis Verdier | 1935–1989

Verdier, a student of Grothendieck, formalized triangulated and derived categories in his thesis Des catégories dérivées des catégories abéliennes (defended in fragments from the early 1960s, published in full only posthumously in 1996). Grothendieck conceived derived categories around 1957–63 to provide the correct setting for duality in algebraic geometry (Grothendieck–Serre duality), but it was Verdier who isolated the axioms TR1–TR4, defined distinguished triangles, and proved the localization theorem. Verdier duality — the formalism of $f^{!}$ and the dualizing complex — is named for him. He later turned to analytic geometry and singularities. The notorious subtlety of the octahedral axiom, and the realization decades later (via stable $\infty$ -categories) that triangulated categories are a "shadow" of richer structure, are both anticipated in the tension within his original framework.

Building Intuition

Think of a complex as a "smeared-out object" carrying cohomology in several degrees, like a particle delocalized across energy levels. A quasi-isomorphism is a change of representative that preserves all the cohomological "charges" — so in $D (A)$ we identify any object with its resolutions.

A distinguished triangle $X \to Y \to Z \to X [1]$ is the homological replacement for a short exact sequence $0 \to X \to Y \to Z \to 0$ . But it is better: it is symmetric under rotation, and it always exists for any map $f$ (the cone $Z$ measures the failure of $f$ to be an isomorphism — if $f$ is a quasi-iso, the cone is zero). The connecting map $Z \to X [1]$ is the "boundary operator" of the long exact sequence, now built into the structure rather than constructed by the snake lemma.

The reason $D (A)$ is not abelian: cones are only weak colimits. The fill-in $w$ in TR3 exists but is not unique, so cones do not satisfy a universal property — there are no genuine kernels or cokernels. This is the price of working "up to homotopy," and it is exactly the deficiency that $\infty$ -categories repair.

Worked Examples

Resolving a module: $\mathbb{Z}/n$ over $\mathbb{Z}$

Given:

A = Z -Mod

, the module

M = Z / n

. We compute its place in

D (Z)

. Step 1. A free resolution:

0 \to Z n Z \to 0

sitting in degrees

- 1, 0

, call it

P^{∙}

. Step 2. The augmentation

P^{∙} \to Z / n [0]

is a quasi-isomorphism:

H^{0} (P^{∙}) = Z / n

H^{- 1} (P^{∙}) = ker (n) = 0

. Step 3. Hence in

D (Z)

M ≅ P^{∙}

. Now compute

Tor

via the cone/derived tensor. Apply

- \otimes Z / m

Z / m n Z / m = Z / m n mod m Z / m .

⟳tensoring the free resolution turns derived functors into ordinary cohomology of the resulting complex.

Step 4. Cohomology: with

d = g cd (n, m)

H^{0} = Z / m / n Z / m = Z / d, H^{- 1} = ker = Z / d .

⟳multiplication by

n

Z / m

has kernel and cokernel both of order

d = g cd (n, m)

Tor_{0} (Z / n, Z / m) = Z / d, Tor_{1} (Z / n, Z / m) = Z / d, d = g cd (n, m) .

A distinguished triangle and its long exact sequence

Given: In

D (Z)

, the multiplication-by-

2

map

f : Z 2 Z

(both in degree 0). Step 1. The cone is

Cone (f) = [Z 2 Z]

in degrees

- 1, 0

, which is quasi-isomorphic to

Z /2

in degree

0

. Step 2. The distinguished triangle is

Z 2 Z \to Z /2 \to Z [1] .

⟳identify the cone with its cohomology since the complex is concentrated after resolution.

Step 3. Apply

H^{*}

(here trivial since all concentrated in degree 0, giving the familiar exact sequence):

0 \to H^{0} (Z) 2 H^{0} (Z) \to H^{0} (Z /2) \to H^{1} (Z [1]) \to \dots

0 \to Z 2 Z \to Z /2 \to 0.

⟳equation (8) applied to triangle (13); the connecting map vanishes here.

The short exact sequence 0 \to Z 2 Z \to Z /2 \to 0 IS a distinguished triangle in D (Z) .

This illustrates the general fact: short exact sequences in

A

give distinguished triangles in

D (A)

Common Misconceptions

(1) $D (A)$ is not abelian. It has no kernels or cokernels in general; "exactness" is only meaningful via distinguished triangles. Never speak of "the image" of a morphism in

D (A)

. (2) The cone is not a functor.

Cone (f)

depends on

f

only up to non-canonical isomorphism in

D (A)

, because the fill-in

w

in TR3 is non-unique. There is no functor

Cone : Arr (T) \to T

. This is a genuine defect, not a notational nuisance. (3) Distinguished triangles ≠ exact sequences of complexes. A short exact sequence of complexes gives a distinguished triangle, but not every distinguished triangle arises from a degreewise short exact sequence. (4) Quasi-isomorphism is directional in $K$ , symmetric in $D$ . Two quasi-isomorphic complexes need not be related by a single chain map; you may need a roof. Forgetting this leads to false computations.

Triangulated structure is not determined by the underlying category

Two triangulated categories can have equivalent underlying additive categories with the same shift functor but different, non-equivalent classes of distinguished triangles. Moreover (Verdier's example, refined by others): the cone construction is genuinely non-functorial — there exist morphisms of triangles where no fill-in is an isomorphism even though the outer maps are. A striking consequence: a triangulated category does not determine its homotopy colimits canonically, and there exist non-isomorphic complexes that are not distinguishable by any triangulated invariant but become distinguishable in the underlying dg- or $\infty$ -enhancement. This "phantom map" phenomenon (nonzero maps inducing zero on all cohomology) shows triangulated categories lose information present in the model.

Connections

Derived categories sit at the crossroads of several frameworks:

Abelian Categories: $D (A)$ is built from $A$ ; conversely the heart of a t-structure on a triangulated category recovers an abelian category, allowing one to extract new abelian categories (e.g. perverse sheaves) from a fixed $D (A)$ .
Derived Functors: the total derived functors $L f$ , $R f$ live naturally on $D (A)$ ; classical $Ext^{n}$ and $Tor_{n}$ become $Hom_{D (A)} (X, Y [n])$ and cohomology of $\otimes^{L}$ .
Stable Infinity Categories: every stable $\infty$ -category has a triangulated homotopy category; this is the modern correction to the non-functoriality defects, and dg-categories give an intermediate enhancement.
Model Categories: $D (A)$ is the homotopy category of a model structure on $Ch (A)$ (projective/injective), connecting to abstract homotopy theory.
Spectral Sequences: filtered complexes and t-structures produce spectral sequences as a structural consequence of triangle decompositions.

Real-World Applications

Grothendieck–Serre Duality (algebraic geometry): the dualizing complex $ω^{∙}$ and the functor $f^{!}$ are objects/operations definable only in $D (Coh)$ ; classical Serre duality is the lowest cohomology shadow.

Fourier–Mukai Transforms: an equivalence $D^{b} (Coh X) ≃ D^{b} (Coh Y)$ given by a kernel object; Mukai proved an abelian variety and its dual have equivalent derived categories though they are not isomorphic varieties.

Perverse Sheaves & the Decomposition Theorem (Beilinson–Bernstein–Deligne–Gabber): perverse sheaves are the heart of a non-standard t-structure on the derived category of constructible sheaves; the BBD decomposition theorem underlies modern intersection cohomology and the geometric Satake equivalence.

Homological Mirror Symmetry (Kontsevich): conjectures an equivalence between $D^{b} (Coh X)$ and the derived Fukaya category of a mirror manifold — a deep bridge between algebraic geometry and symplectic topology.

Tilting Theory & Representation Theory: derived equivalences of finite-dimensional algebras (Rickard's theorem) classify when two algebras have equivalent module derived categories; central to modular representation theory and Broué's abelian defect conjecture.

Bridgeland Stability Conditions: a space of stability conditions on $D^{b} (Coh X)$ refines moduli problems and connects to string-theoretic D-brane charges.

Crystalline / $ℓ$ -adic Cohomology: $ℓ$ -adic derived categories underpin the proof of the Weil conjectures and the Langlands program's geometric incarnation.

Historical Development

1955–57 — Cartan–Eilenberg's Homological Algebra systematizes derived functors via resolutions, but without derived categories; the "right" framework is conspicuously absent.
c. 1957–1963 — Grothendieck, seeking the correct formulation of coherent duality, conceives of working with complexes up to quasi-isomorphism; outlines in letters and the Tôhoku paper foundations (abelian categories, enough injectives).
1963 — Verdier begins his thesis under Grothendieck, isolating the axioms of triangulated categories and the localization construction.
1966 — Hartshorne's Residues and Duality (Springer LNM 20), based on Grothendieck's seminar notes, gives the first widely available account of derived categories and Grothendieck duality.
1967 — Verdier's Catégories dérivées, état 0 circulated; SGA 4½ later includes related material.
1977 — Beilinson, Bernstein, Deligne develop t-structures; perverse sheaves emerge (published 1982 in Faisceaux pervers, Astérisque 100).
1981 — Mukai constructs the Fourier–Mukai transform, the first dramatic derived equivalence of distinct varieties.
1988–89 — Rickard's Morita theory for derived categories (tilting complexes) revolutionizes representation theory.
1994–96 — Verdier's thesis published in full (Astérisque 239); Bondal–Kapranov and others develop enhancements; Kontsevich announces homological mirror symmetry (ICM 1994), reframing derived categories at the center of mathematical physics.

Subtleties & Limits

The most fundamental subtlety is that triangulated categories are structurally deficient: the cone is non-functorial, homotopy (co)limits are not canonically defined, and triangulated structures do not glue (the derived category of a non-affine scheme is not the derived category of any abelian category in an obvious way; functoriality of $D$ along open covers is delicate). This motivates enhancements — dg-categories, stable $\infty$ -categories, or stable derivators — which carry the same homotopy category but repair the defects.

Open and active questions include:

Uniqueness of enhancement: Is the dg/∞-enhancement of a given triangulated category unique? (Lunts–Orlov proved uniqueness in important cases; counterexamples exist (Schlichting, and Canonaco–Stellari), so the answer is subtle.)
Brown representability and compact generation: Precisely which triangulated categories admit good adjoint functor existence (Neeman's theory) — central but not fully classified.
Reconstruction: To what extent does $D^{b} (Coh X)$ determine $X$ ? (Bondal–Orlov: yes if $X$ has ample (anti)canonical bundle; in general, derived equivalent but non-isomorphic varieties exist — the boundary is not fully understood.)
Existence of stability conditions on $D^{b} (Coh X)$ for general $X$ (e.g. higher-dimensional varieties, Calabi–Yau 3-folds): largely open and tied to Bogomolov–Gieseker-type inequalities.