Categorification

In mathematics, categorification is the process of replacing set-theoretic theorems by category-theoretic analogues. Categorification, when done successfully, replaces sets by categories, functions with functors, and equations by natural isomorphisms of functors satisfying additional properties. The term was coined by Louis Crane.

Categorification is the reverse process of decategorification. Decategorification is a systematic process by which isomorphic objects in a category are identified as equal. Whereas decategorification is a straightforward process, categorification is usually much less straightforward, and requires insight into individual situations.

Examples of categorification

One form of categorification takes a structure described in terms of sets, and interprets the sets as isomorphism classes of objects in a category. For example, the set of natural numbers can be seen as the set of cardinalities of finite sets (and any two sets with the same cardinality are isomorphic). In this case, operations on the set of natural numbers, such as addition and multiplication, can be seen as carrying information about products and coproducts of the category of finite sets. Less abstractly, the idea here is that manipulating sets of actual objects, and taking coproducts (combining two sets in a union) or products (building arrays of things to keep track of large numbers of them) came first. Later, the concrete structure of sets was abstracted away – taken "only up to isomorphism", to produce the abstract theory of arithmetic. This is a "decategorification" – categorification reverses this step.

Other examples include homology theories in topology. See also Khovanov homology as a knot invariant in knot theory.

An example in finite group theory is that the ring of symmetric functions is categorified by the category of representations of the symmetric group. The decategorification map sends the Specht module indexed by partition \lambda to the schur function indexed by the same partition:

S^\lambda \stackrel{\varphi}{\to} s_\lambda,

(essentially following the character map from a favorite basis of the associated Grothendieck group to a representation-theoretic favorite basis of the ring of symmetric functions). This map reflects much of the parallels in structure; for example

[\mathrm{Ind}_{S_m \otimes S_n}^{S_{n+m}}(S^{\mu} \otimes S^{\nu})] \qquad \text{ and } \qquad s_\mu s_\nu

have the same decomposition numbers over their respective bases, both given by Littlewood-Richardson coefficients.

Abelian categorifications

For a category \mathcal{B}, let K(\mathcal{B}) be the Grothendieck group of \mathcal{B}.

Let A be a ring which is free as an abelian group, and let \mathbf{a} = \{a_i\}_{i \in I} be a basis of A such that the multiplication is positive in \mathbf{a}, i.e.

a_i a_j = \sum_{k} c_{ij}^k a_k, with  c_{ij}^k \in \mathbb{Z}_{\geq 0}.

Let B be an A-module. Then a (weak) abelian categorification of (A, \mathbf{a}, B) consists of an abelian category \mathcal{B}, an isomorphism  \phi: K(\mathcal{B}) \to B, and exact endofunctors F_i: \mathcal{B} \to \mathcal{B} such that

  1. the functor F_i lifts the action of a_i on the module B, i.e. \varphi [F_i] = a_i \varphi, and
  2. there are isomorphisms F_i F_j \cong \bigoplus_{k} F_k^{c_{ij}^k},, i.e. the composition F_i F_j decomposes as the direct sum of functors F_k in the same way that the product a_i a_j decomposes as the linear combination of basis elements a_k.

See also

Further reading