(Beck’s monadicity theorem, tripleability theorem)
A functor is monadic (tripleable) if and only if
A proof is reproduced at (Borceux, vol 2, theorem 4.4.4).
Here a parallel pair in is -split if the pair has a split coequalizer in . Specifically, this means that there is a diagram in :
where has a section and has a section such that . This implies that the arrow is necessarily a coequalizer of and . To say that creates coequalizers of -split pairs is to say that for any such -split pair, there exists a coequalizer of in which is preserved by , and moreover any fork in whose image in is a split coequalizer must itself be a coequalizer (not necessarily split).
An equivalent, and sometimes easier, way to state these conditions is to say that
A functor is monadic precisely if
This is equivalent because a conservative functor reflects any limits or colimits which exist in its domain and which it preserves, while monadic functors are always conservative.
The crude monadicity theorem gives a sufficient, but not necessary, condition for a functor to be monadic. It states that a functor is monadic if
(Recall that a parallel pair is reflexive if and have a common section.) This sufficient, but not necessary, condition is sometimes easier to verify in practice. In contrast to the crude monadicity theorem, the necessary and sufficient condition above is sometimes called the precise monadicity theorem.
Duskin’s monadicity theorem gives a different sufficient, but not necessary, condition which refers only to quotients of congruences. It says that a functor is monadic if
We can weaken the hypothesis a bit further to obtain the theorem:
As usual, we can also modify it by replacing reflection of limits by reflection of isomorphisms.
If we view the objects of as underlying -objects with structure, this says that any congruence in induces a -structure on its quotient in . As with the crude monadicity theorem, this condition is sometimes easier to verify since quotients of congruences are often better-behaved than arbitrary coequalizers. This is the case in many “algebraic” situations.
Duskin actually gave a slightly more precise version only assuming the categories and to have particular finite limits, rather than all of them.
In the case when the base category is Set, one can further refine the requisite conditions. Linton proved that a functor is monadic if and only if
There are other versions of this theorem, including generalizations to monadicity over presheaf categories, which can be viewed as analogues of Giraud's theorem.
The version of the monadicity theorem given in Categories Work uses an evil notion of “creation of limits” and concludes that the comparison functor is an isomorphism of categories, rather than merely an equivalence. But the versions mentioned above can be found in the exercises.
Note however that if is an amnestic isofibration, then is monadic iff it is strictly monadic. For an application of this observation, see for example the discussion of algebraically free monads at free monad.
We will use Duskin’s variant to prove that the forgetful functor GrpSet is monadic. Of course, this is also easy to show by explicit computation, but it serves as a useful example of how to use a monadicity theorem. We first need it to have a left adjoint: this is easy to show by a direct construction of free groups, but we could also invoke the adjoint functor theorem. It is also easy to show that it is conservative (a bijective group homomorphism is a group isomorphism), so it remains to consider congruences.
Since limits in are created in , a congruence in on a group is an equivalence relation on which is also a subgroup of . This latter condition means that if and , then also and . Since for all , it follows that if and only if , so is determined by the subset of those such that . This is clearly a subgroup of , and moreover a normal subgroup, since if and we have , so . Conversely, it is easy to construct a congruence from any normal subgroup, so the two notions are equivalent. It remains only to observe that the quotient of a group by a normal subgroup is, in fact, a quotient of its associated congruence in , which is preserved by . Thus, by Duskin’s monadicity theorem, is monadic.
The monadicity theorem becomes more important when the base category is more complicated and harder to work with explicitly, and when the objects of are not obviously defined as “objects of with extra structure.” For instance, the category of strict 2-categories is monadic over the category of 2-globular sets, essentially by definition, but it is much less trivial to show that it is also monadic over the category of 2-computads. This latter fact can, however, be proven using the monadicity theorem.
The monadicity theorem also plays a central role in monadic descent.
There is a version of the monadicity theorem for (∞,1)-monads in section 3.4 of
Canonical textbook references include
Section VI.7 of Categories Work.
Chapter 3 of Michael Barr, Charles Wells, Triples, toposes, and theories , Grundlehren der math. Wissenschaften 278, Springer-Verlag 1983, ftp, web, pdf
Other references include:
Jean Bénabou, Jacques Roubaud, Monades et descente , C. R. Acad. Sc. Paris, t. 270 (12 Janvier 1970), Serie A, 96–98
Duško Pavlović, Categorical interpolation: descent and the Beck-Chevalley condition without direct images, Category theory Como 1990, pp. 306–325, Lecture Notes in Mathematics 1488, Springer 1991
Pierre Deligne, Catégories Tannakiennes , Grothendieck Festschrift, vol. II, Birkhäuser Progress in Math. 87 (1990) pp. 111-195.
wikipedia: monadicity theorem
John Bourke, Two dimensional monadicity, arxiv/1212.5123