This is a subentry of sheaf about the plus-construction on presheaves. For other constructions called plus construction, see there.
The plus construction on presheaves over a site is an operation that makes a presheaf “closer” to being a sheaf.
For ordinary sheaves of sets, the plus construction replaces a presheaf first by a separated presheaf and then by a sheaf.
(The first of these steps, however, is not a reflection into .) Notice that in terms of n-truncated morphisms, a presheaf is
separated precisely if every comparison map to the set of descent data is (-1)-truncated, namely a monomorphism;
a sheaf precisely if every such comparison map is (-2)-truncated, namely an equivalence.
More generally, in (n,1)-topos theory, the plus-construction is applied -times in a row to an (n,1)-presheaf? (a presheaf of -groupoids), at each step decreasing the truncatedness of the comparison map until at the last step it becomes an actual (n,1)-sheaf, a.k.a. stack; see Anel and Leena Subramaniam, 2020. When , even a countable sequence of applications does not suffice, but a transfinite one does; see Lurie, section 6.5.3.
Let be a small site equipped with a Grothendieck topology , let be a functor. Then the plus construction is a functor defined by the following equivalent descriptions:
where can be defined in any of the following equivalent ways:
For we define to be the set of equivalence classes of pairs where and is a compatible family of elements of relative to , and iff there is a -covering sieve on which the restrictions of and agree.
In the internal language of the presheaf topos , we can define , where is the equivalence relation given by if and only if and is the Lawvere-Tierney topology.
, where is the Grothendieck construction of the functor sending each to its set of covering sieves, and sends each to its maximal sieve (the set of all morphisms with codomain ).
The first and last of these definitions also work for higher sheaves on higher sites, i.e. when is an (infinity,1)-site and is an (infinity,1)-presheaf, as long as limits, colimits, and Kan extensions are interpreted in the -categorical sense.
The basic property of the plus-construction is that we can construct sheafification (a.k.a. stack completion, in the higher-categorical case) by iterating it. For ordinary presheaves of sets, two iterations suffice: for any presheaf , is separated, while if is separated then is a sheaf; thus is the sheafification of any presheaf .
Note, though, that is not left adjoint to the inclusion of the full subcategory of separated presheaves. If it were, it would be a reflector and therefore satisfy . But this is false, since the plus construction applied to separated presheaves yields their sheafification. See this MathOverflow question for details.
More generally, when working with presheaves of -groupoids for some finite , it takes iterations for the plus-construction to converge to a sheaf/stack. In the case , no finite number of iterations suffices, and not even a countable number, but a transfinite number does.
We now sketch a proof of these facts, using the implicit infinity-category convention so that our results apply to higher sheaves/stacks as well as ordinary ones.
We use the fourth definition above , which we will write more shortly as . There is a canonical map , which we can more easily construct and study with the following observation:
is right adjoint to . Therefore, we have an adjunction , and hence an adjoint quadruple:
Moreover, and are fully faithful and preserves finite limits, so is cohesive and even totally connected over .
Since the maximal sieve is the terminal object of the fiber of , the functor picks out a fiberwise terminal object in the fibration ; thus it is right adjoint to . This implies the existence of the adjoint quadruple as discussed here. Since , we have that is fully faithful, and hence so are and . Finally, the fibers of are filtered since covering sieves are closed under intersections, so preserves finite limits since finite limits commute with filtered colimits.
Thus, we can identify with either the discrete objects or codiscrete objects in , via the fully faithful functors or . If we regard as the discrete object , then the plus-construction is similarly identified with . Thus can be written as , where is the reflector into the discrete objects and is the reflector into the codiscrete objects.
The rest of the proof can actually be carried out “synthetically” in the internal homotopy type theory of the cohesive topos . This argument is formalized in the HoTT/Coq library. But here we formulate the argument externally and categorically.
The plus-construction is a well-pointed endofunctor. That is, we have a natural transformation and a homotopy (an equality in the case ) . Moreover, the whiskered homotopy is homotopic to the naturality square .
This follows fairly formally from its identification with , the composite of two reflectors.
The plus-construction preserves finite limits.
Both functors and preserve finite limits.
For , the following are equivalent:
To see that the first two are equivalent, note that for a covering sieve of we have while is the space of descent data for . Now we have , so the second and third statements are equivalent.
For the fourth, if is codiscrete, then it is equivalent to ; hence the latter is also discrete, and thus equivalent to . Of course if is an equivalence, it has a retraction. Finally, if it has a retraction, then the composite is a retraction for , so that is codiscrete.
If some (perhaps transfinite) iteration of the plus-construction on stabilizes, it is a sheaf and the sheafification of .
This follows formally from the fact that the plus-construction is a well-pointed endofunctor whose fixed points are the sheaves.
Now since the plus-construction is an accessible endofunctor of a locally presentable category, its iteration always converges at some transfinite stage. It follows that sheafification exists. (Roughly this argument appears in Lurie, section 6.5.3.)
We now want to prove that if is a presheaf of -groupoids for some finite , the iteration in fact converges already after steps. Though long expected, it seems that this was first actually proven by Anel and Leena Subramaniam, 2020; what follows is inspired by their proof. We will actually prove something slightly more general.
A presheaf is -separated if the map is n-truncated.
Thus is a sheaf if and only if it is -separated. Moreover, since the plus-construction preserves finite limits, it preserves truncated objects; thus if is -truncated (i.e. a presheaf of -groupoids) then so is and hence so is . Thus, any -truncated presheaf is -separated.
We will prove that if is -separated, then its plus-construction converges after steps. The central auxiliary definition is:
A map in is plus-connected if there is a lift in its -naturality square. That is, there is a map and homotopies and that paste to the naturality square .
For any , the map is plus-connected.
If is plus-connected, so is its diagonal .
Since the plus-construction preserves finite limits, it suffices to observe that if a given commutative square has a lift, so does the induced square between its diagonals.
So far everything has been very formal. The concrete input from the fact that we are talking about a Grothendieck topology comes here:
If is a monomorphism that is plus-connected, then is an equivalence.
Since the plus-construction preserves finite limits, it preserves monomorphisms; thus it suffices to show that is surjective. An object is determined by a matching family over a covering sieve , consisting of for all in . Since is plus-connected, we have a map assigning to each an object , determined by a matching family over a covering sieve . Now we can compose the sieves and to obtain a covering sieve on , such that the restriction of to (which is equal to ) is in the image of .
For , if is -truncated and plus-connected, then is -truncated.
By induction. The base case is the previous lemma. For the inductive step, is -truncated and plus-connected, then its diagonal is -truncated and plus-connected. Thus, by induction is -truncated. But since the plus-construction preserves finite limits, is equivalent to , so is -truncated.
If is -separated, then its plus-construction converges after steps.
If is -separated, then by definition is -truncated, and we have shown it is always plus-connected. Thus, is -truncated. Hence so is , in other words is -separated.
Thus, by induction, if is -separated, the -th iteration of the plus-construction is -separated, i.e. -separated, i.e. a sheaf.
Note that we do not need to know that the category is obtained as the category of covering sieves for a topology on . We only need that there is a projection that is a fibration with cofiltered fibers and a fully faithful right adjoint, plus the result of Lemma . In particular, could be a “mono-saturated lex modulator” in the sense of ALS, generating a non-topological localization of .
We can also construct left exact localizations of non-presheaf toposes in this way. If is a topos that is itself a left exact localization of some presheaf topos , then a lex modulator on induces a as above, and by composing the reflectors and with the reflection into we can make the same argument work to construct a left exact localization of .
Related entries: sheafification
A standard textbook reference in the context of 1-topos theory is:
Remarks on the plus-construction in (infinity,1)-topos theory is in section 6.5.3 of
The plus-construction for presheaves in values in abelian categories is also called the Heller-Rowe construction:
The plus-construction is studied for higher sheaves, and shown to converge after steps for -presheaves, in
The above argument is formalized in the internal logic of here:
Last revised on May 28, 2021 at 18:40:51. See the history of this page for a list of all contributions to it.