adhesive category

Adhesive categories


An adhesive category is a category in which pushouts of monomorphisms exist and “behave more or less as they do in the category of sets”, or equivalently in any topos.



The following conditions on a category CC are equivalent. When they are satisfied, we say that CC is adhesive.

  1. CC has pullbacks and pushouts of monomorphisms, and pushout squares of monomorphisms are also pullback squares and are stable under pullback.

  2. CC has pullbacks, and pushouts of monomorphisms, and the latter are also (bicategorical) pushouts in the bicategory of spans in CC.

  3. (If CC is small) CC has pullbacks and pushouts of monomorphisms, and admits a full embedding into a Grothendieck topos preserving pullbacks and pushouts of monomorphisms.

  4. CC has pullbacks and pushouts of monomorphisms, and in any cubical diagram:


    if XYX\to Y is a monomorphism, the bottom square is a pushout, and the left and back faces are pullbacks, then the top face is a pushout if and only if the front and right face are pullbacks. In other words, pushouts of monomorphisms are van Kampen colimits.



In an adhesive category, suppose given a pushout square

C m A f g B n D \array{ C & \xrightarrow{m} & A \\ ^f\downarrow && \downarrow^g \\ B & \xrightarrow{n} & D }

such that mm is a monomorphism. Then:

  1. nn is also a monomorphism.
  2. The square is also a pullback square.
  3. The square is also a distributivity pullback around (g,m)(g,m); hence in particular n= gmn = \forall_g m is the universal quantification.

(Notice that generally monomorphisms (as discussed there) are preserved by pullback.) For a proof of the above proposition, see (Lack, prop. 2.1) and (Lack-Sobocinski, Lemmas 2.3 and 2.8). The latter Lemma 2.8 states only that n= gmn = \forall_g m (a weaker universal property since it refers only to other monomorphisms into DD), but the proof applies more generally.


An adhesive category with a strict initial object is automatically an extensive category.

We define a pushout complement of m:CAm:C\to A and g:ADg:A\to D to be a pair of arrows f:CBf:C\to B and n:BDn:B\to D such that nf=gmn f = g m and this commutative square is a pushout. The following proposition is crucial in double pushout graph rewriting.


In an adhesive category, if m:CAm:C\to A is mono and g:ADg:A\to D is any morphism, then if a pushout complement exists, it is unique up to unique isomorphism.


We give only a sketch; details are in (LS, Lemma 4.5). If (f,n)(f,n) and (f,n)(f',n') are two pushout complements, consider the two pushout squares as morphisms in the arrow category with target gg, and take their pullback. The resulting commutative cube can be viewed as a morphism in the category of commutative squares from the pullback square of mm against itself (which is again mm, since mm is mono) to the pullback square of nn against nn'. Denote the vertex of the latter pullback square by UU. Applying the van Kampen property in two directions, we find that the maps UBU\to B and UBU\to B' are both pushouts of 1 C1_C, hence isomorphisms. This gives an isomorphism between the pushout complements; it is unique since nn and nn' are mono (being pushouts of the mono mm).


Adhesiveness is an exactness property, similar to being a regular category, an exact category, or an extensive category. In particular, it can be phrased in the language of “lex colimits”.


Revised on July 29, 2017 04:00:58 by Mike Shulman (