# Advantages of Diffeological Spaces over General Sheaves

I have been playing with/thinking about diffeological spaces a bit recently, and I would like understand something rather crucial before going further. First a little background:

Diffeological spaces are a Cartesian-closed, complete, and cocomplete category containing all infinite dimensional manifolds, and in fact even form a quasi-topos.

Diffeological spaces, concisely, are nothing more than concrete sheaves on the site of Cartesian manifolds (manifolds of the form $\mathbb{R}^n$):

http://ncatlab.org/nlab/show/concrete+sheaf

However, the category of ALL sheaves on Cartesian manifolds, categorically is even nicer, since it is a genuine topos.

$\textbf{My question is:}$ What can you do with diffeological spaces that you cannot do with general sheaves? Or, more generally, what are the advantages of diffeological spaces over general sheaves?

All of the generalizations of differential geometry concepts to diffeological spaces I have seen so far, actually carry over to genuine topos of sheaves (though sometimes with a little more work).

I'm aware that you gain the ability to work with a set with extra structure and talk about its points etc, but, what does this gain you? It seems that you can always use Grothendieck's functor of points approach instead.

Is it that limits and colimits are more like their counterparts for manifolds?

-
Can you do tangent spaces with sheaves? Martin Laubinger's definition of tangent spaces goes "pointwise". – Konrad Waldorf Dec 7 '10 at 21:24
@Konrad: I believe that you can give a definition of the tangent sheaf by abstract nonsense. This, if I remember correctly, can be given simply by taking the sheaf of $\mathbf{R}$-derivations $\mathcal{O}_X\to \mathcal{O}_X$, where $\mathcal{O}_X$ is the structure sheaf (which can be defined abstractly). – Harry Gindi Dec 7 '10 at 21:41
(That is, if I remember correctly, it can be defined abstractly on the Gros-topos by pulling back the smooth affine line). – Harry Gindi Dec 7 '10 at 21:43
The cotangent complex can also be defined in some huge generality (see Illusie's books on the cotangent complex) on any ringed grothendieck topos, although I'm not sure that this will give the correct smooth cotangent complex. – Harry Gindi Dec 7 '10 at 21:46
There's lots to say on this topic, but I feel that I'd rather have a proper discussion (say, on a forum?) than use the MO engine where it's difficult to do proper replies. For example, I'd like to reverse the challenge: given that I like manifolds and would like to stay as close to manifolds as I can, why should I take sheaves just to get a topos? (Please don't answer that here! If you - or anyone else - has an answer, take it to the nForum.) – Loop Space Dec 8 '10 at 18:41

I want to give two simple observations about diffeological spaces that might provide a partial answer to your question.

1) We have the following inclusions of full subcategories $$Mfd \subset Diff \subset Sh \subset PSh$$ where $Mfd$ is the category of smooth finite dim manifolds, $Diff$ are diffeological spaces (i.e. concrete sheaves on cartesian spaces), Sh are sheaves on cartesian spaces and $PSh$ are presheaves on cartesian spaces. The last two inclusions are reflexive.

Lets us first have a closer look at the inclusion $Sh \subset PSh$. Following the same vein of argument as above, there is a priori no reason to work with $Sh$ instead of $PSh$ since both categories are equally nice (topoi) and the definition of a presheaf is clearly simpler than that of a sheaf. But there are some colimits in $Mfd$ that we really like, namely the coequalizer diagram correspoding to an open cover $(U_\alpha)$ of a manifold $M$. Under the inclusion of $Mfd$ into $PSh$ this is not a coequalizer anymore, in other words: If we glue open sets in $PSh$ together we do not get the same thing that we get when glueing together as manifolds. This defect is exactly cured by the sheaf property. That means restricting to the smaller subcategory $Sh \subset PSh$ the colimits change such that gluing of open sets behaves as nice as in manifolds. The punchline is that the restriction to $Sh$ provides the category with the "right" coequalizers of open sets.

Now lets turn towards the inclusion $Diff \subset Sh$. The situation is exactly the same as before. Limits in $Diff$ are computed as Limits in $Sh$ (and hence also $PSh$) but colimits are different in general (one has to apply the concretization functor). This is what happens categorically. Now it turns out that there are colimits in manifolds that become colimits in diffeological spaces but not colimits in sheaves. Here an example would be very nice. Unfortunately I have not been able the remember the example I had for this behaviour. Even so, from abstract reasoning it is clear that the colimits in the two categories have to differ.

Hence one could argue that diffeological spaces have the right "geometric" colimits and sheaves do not. The price is of course that we exclude some interesing "spaces" like the sheaf of diffential forms and loose the property that the category is a topos.

2) If we want to "make" geometry over diffeological spaces it turns out that there are two possible definition of principal bundles:

• a bundle over a diffeological space $M$ is a morphism to the stack of bundles over finite dimensional manifolds. This means that we have a family of bundles over each plot together with coherent isomorphisms. Note that this type of bundle is determined by its pullback to finite dimensional spaces. This is equivalent to have a diffeological space $P \to M$ together with a free transitive on fibers action such that the quotient map $P \to M$ is a surjective subduction (i.e. becomes a submersion on each plot). To get those type of bundles we have to equip diffeological spaces with the Grothendieck Topology of subductions.

• a bundle over a diffeological space $M$ is a space $P \to M$ with a free, transitive on fibers, action such that it is locally trivial, where locally refers to the underlying topological space of $M$. This is the type of bundle which people consider in the world of $\infty$-dimensional manifolds. To get this we have to take the grothendieck topology of morphisms that are surjective and admits local (in the topology) sections. Hence therefore we really need the underlying topological space.

I do not prefer one of the two possible Grothendieck Topologies, but the second one is closer to what people have done in the $\infty$-dimensional setting. And one can show that the universal bundle $EG \to BG$ for a compact Lie-group is of this type (of course one has to find diffeological models of $BG$ and $EG$).

The first topology has an obvious analogue on the category $Sh$ of all sheaves but the second crucially uses the underlying topological space of a diffeological space.

-
@Thomas: It would be nice to see an example of what kind of colimits of manifolds are preserved by the embedding into diffeological spaces that are not preserved when embedding into the full topos of sheaves. Do you know where you saw this example? Indeed, it this kind of answer that I was looking for. By the way, ALL sheaves have an "underlying space". Consider the functor which sends every manifold to its underlying space. By left-kan extension, this induces an "underlying space" functor for sheaves, which agrees with the D-topology for diffeological spaces. – David Carchedi Dec 13 '10 at 18:33
Thomas: you can continue your argument; there are colimits in Mfd that aren't preserved when mapping in to Diff. To preserve all limits and colimits that exist in Mfd, you should work with Hausdorff Frolicher spaces, see the last section (at time of writing) of ncatlab.org/nlab/show/… for details. – Loop Space Dec 13 '10 at 20:27
@Dave: Can you determine what the underlying space of the sheaf $\Omega^1(M)$ is? I dont have a good intuiton what kind of bundles your construction produces... I hope that I remember the statement about colimits correct. Unfortunately I don't have time at the moment, but I try to get it correctly tomorow. – Thomas Nikolaus Dec 13 '10 at 20:34
Also, arsmath's example of a space that is two copies of the real line with points identified but other functions not is expressible as a colimit of two copies of the real line so there's your example. – Loop Space Dec 13 '10 at 20:35
Thomas: the underlying space of a sheaf is just its evaluation on the singleton point, so the underlying space of $\Omega^1(M)$ is just $\Omega^1(pt)$ - not very interesting at all! – Loop Space Dec 13 '10 at 20:36

I'm going to pre-empt my esteemed colleague here and say that to some people, the purpose of diffeological spaces is as a stepping stone between manifolds and the category of sheaves on manifolds (or on cartesian spaces, it's the same thing). So, to these people, you've stumbled on the main point: we really ought to be working with sheaves all along.

The problem is that there are some ornery people who really like manifolds as they are, but sometimes have to work with things that are almost but not quite completely unlike manifolds. For these people, the further away from true manifolds they get, the more uncomfortable they feel. One of the biggest steps for such people is losing the underlying set. So diffeological spaces are a category in which those people can have most of the benefits of sheaves without having to discard their comfort blanket of something that still resembles manifolds in some way.

So diffeological spaces are a convenient (yes, I use the word deliberately!) half-way house whereby those who have Seen The Light can still talk to those still quivering under their comfort blankets.

To name names, people in the first category include Urs Schreiber and John Baez (indeed, I think that John makes that point somewhere on the n-Cafe). People in the second category include me!

Indeed, I would say that diffeological spaces are closer to the One True Category of Smooth Spaces than sheaves on cartesian spaces. Frolicher spaces seem to irretrievably have underlying sets - I and a few others have wondered from time to time if there is a way to remove that property but it seems tied up with what they are.

(Added in edit): I don't know why, but I didn't spot first time around the last line:

Is it that limits and colimits are more like their counterparts for manifolds?

which is odd, because that's the subject of a little theorem I've proved which can be found on the nLab:

http://ncatlab.org/nlab/show/topological+notions+of+Fr%C3%B6licher+spaces#hausdorff

Essentially, if you want to preserve those limits and colimits that already exist in the category of manifolds, then you need to work in the category of Hausdorff Frölicher spaces. When you enlarge that category (say to Frölicher spaces or to Diffeological spaces, or to sheaves) then you add in stuff "in the gaps" and create new limits or colimits that disagree with the ones that you had before (in these cases, it's almost always colimits, but if you take the "maps out" view, it will be limits). So that question isn't really a sensible one to ask of Diffeological spaces as you've already lost some colimits. I suppose you can try to do a bit of damage limitation ...

-
Hi Andrew, there is some behing-the-scenes discussion between me and Dave. We are all on the same page as far as the pleasures of general abstract topos theory go. Here the question is about nice formal characterizations of a "filtering" of the category of all sheaves by tame/wild-degree. some intermediate steps are: representable, locally representable, concrete. Here the question is how to characterize abstractly what makes concrete sheaves nice. And then to say what makes concrete oo-stacks aka cohesive oo-groupoids so nice. Your statement about Isbell self-duality would be the sort needed. – Urs Schreiber Dec 7 '10 at 22:33
Andrew, is there some natural case of "manifoldness" where having an underlying set is an obstacle? – arsmath Dec 7 '10 at 23:52
There is a curious irony to this story: when John announced his article on the quasitopos of concrete smooth spaces on the category theory mailing list, Bill Lawvere complained harshly that after "the proliferation of such smooth categories 45 years ago" still not everyone has switched to working with genuine toposes. In his reply (archived at mta.ca/~cat-dist/archive/2008/08-8) he vaguely refers to his axioms for cohesive toposes (recalled here nlab.mathforge.org/nlab/show/cohesive (infinity,1)-topos). Interestingly, that's exactly where Dave's question above originates... – Urs Schreiber Dec 8 '10 at 15:55
... because whenever a big topos is cohesive it is in particular local and then there is canonically associated to it the sub-quasitopos of its concrete objects. So therefore we are now on the reverse path through those 45 years: after we have learned that the general ambient context is always an oo-topos, we should go back and check which of all the concrete smaller contexts that we might be interested in in special situations are canonically induced by that topos-context. For instance the quasitopos of diffeological space is canonically induced by the Cahiers topos. – Urs Schreiber Dec 8 '10 at 16:00
@Andrew: Your edited stuff is great. Does this mean, that Fröhlicher Spaces are the universal cocompletion of $Man$ subject to the requirement that the inclusion preserves all colimits? – Thomas Nikolaus Dec 13 '10 at 20:41

I think the argument for diffeological spaces is just that it eliminates certain kinds of pathological constructions that are possible with general sheaves, without costing you anything. General sheaves allow constructions that geometrically are sick and wrong. For example, you can define a sheaf that geometrically consists of two lines, so that every single point of the two lines are identified, but the two lines are still distinct. The "concrete" condition prevents that kind of pathology.

-
Really, a downvote? Why? Because I said "sick and wrong"? – arsmath Dec 7 '10 at 21:18
I'm cancelling the downvote! The "sick and wrong" comment is spot on. – Loop Space Dec 7 '10 at 21:22
Sorry - that was me. I thought the answer was trite and simplistic, and would basically rule out all the interesting parts of algebraic geometry. Now you've edited it, I agree with the answer and cancel my own downvote. +1 – David Roberts Dec 7 '10 at 22:38
@arsmath:I'm arguing by analogy here, the roots of the quadratic x^2=0 are both zero, so one could safely say that they're identified here, so that you have one root of zero here, but in fact it is better to say that you have two roots that happen to be both zero. I don't know enough algebraic geometry to categorically state that your example is 'sick & wrong', but I suspect it might be this pathology may be intersting when interpreted correctly. – Mozibur Ullah Aug 23 '12 at 20:35

I'm not used to read this website, so I post my remark very late, after having randomly googled this page. I think that one of the reason why diffeological spaces are better than general sheaves is the possibility to consider infinitesimal Fermat extensions:

1. Giordano P. " Fermat reals: nilpotent infinitesimals and infinite dimensional spaces" . Book in preparation, see http://arxiv.org/abs/0907.1872, July 2009.

2. Giordano P. "The ring of fermat reals", Advances in Mathematics 225 (2010), pp. 2050-2075.

3. Giordano P. "Infinitesimals without logic", Russian Journal of Mathematical Physics, 17(2), pp.159-191, 2010

4. Giordano P., Kunzinger M. "Topological and algebraic structures on the ring of Fermat reals". Submitted to Israel Journal of Mathematics on April 2011. See http://arxiv.org/abs/1104.1492

5. Giordano P. "Fermat-Reyes method in the ring of Fermat reals". To appear in Advances in Mathematics, 2011.

6. Giordano P. "Infinite dimensional spaces and cartesian closedness". To appear in Journal of Mathematical Physics, Analysis, Geometry, 2011.

This is a new theory, and I post this answer also because I think it is not known. The basis of the theory is a surprisingly simple extension of the real field containing nilpotent infinitesimals. We start from the class of little-oh polynomials, i.e. functions $x:\mathbb{R}_{\ge 0}\rightarrow \mathbb{R}$ that can be written as $x(t)=r+\sum_{i=1}^{k}\alpha_{i}\cdot t^{a_{i}}+o(t)$ as $t\to 0^+$, where all the coefficients and powers are reals. Then, we introduce the equivalence relation between little-oh polynomials $x\sim y$ iff $x(t)=y(t)+o(t) \text{ as }t \to 0^{+}$. The ring of Fermat reals ${}^\bullet\mathbb{R}$ is the corresponding quotient set. The theory of Fermat reals has been developed trying always to obtain a good dialectic between formal mathematics and intuitive interpretation. Even if there are several theories of infinitesimals, only a couple of them always have this intuitive interpretation, and this contradicts the idea that (rigorous) infinitesimals are a strong support to guess some mathematical truths. Of course, Fermat reals take strong inspiration from smooth infinitesimal analysis, even if, at the end, it is a radically different theory. In fact, in the corresponding ring of scalars, which extends the classical reals, we have nilpotent infinitesimals of every order, infinitesimal Taylor's formulas (analogous of the Kock-Lawvere axiom), powers, roots of (nilpotent!) infinitesimals, logarithms, a total order relation, and the ring is also geometrically representable, so that we can finally state that infinitesimals are no longer ghosts of departed quantities.

It is also very interesting to note that its mathematical definition uses only elementary analysis and Landau's little-oh notation, without requiring a background in mathematical logic. In particular, the model is so simple that can be studied directly in classical logic without any need to switch to intuitionistic logic. On the other hand, this extension of the real field is generalizable both to finite and infinite dimensional manifolds (more generally to diffeological spaces). The extension ${}^\bullet(-): \mathcal{C}^\infty \rightarrow {}^\bullet\mathcal{C}^\infty$ (here $\mathcal{C}^\infty$ is the category of diffeological spaces and ${}^\bullet\mathcal{C}^\infty$ is the category of Fermat spaces, which are defined similarly to diffeological spaces) is functorial and has very good preservation properties: a full transfer theorem for intuitionistically valid sentences is indeed provable (the "true" logic of smooth spaces is always intuitionistic!). Several applications to differential geometry has been already developed: e.g. tangent vectors to any diffeological space $X \in \mathcal{C}^\infty$ can be defined, similarly to SDG, as smooth functions of the form $t:D\rightarrow {}^\bullet X$, where $D :=\{h\in {}^\bullet\mathbb{R}|h^2=0\}$ is the ideal of first order infinitesimals and where ${}^\bullet X\in {}^\bullet\mathcal{C}^\infty$ is the Fermat space obtained extending $X$ with new infinitesimally closed points. At present, we are developing several notions of differential geometry in this framework and are trying to extend the theory so as to include infinities and generalized functions (distributions).

-

The thing you can do with a diffeological space and that you can't do with a sheaf is to define it by a sentence that starts like this: "It's a set equipped with ..."

You might object that I'm just restating the definition. But I really don't think that there's much more to be said.

-
Andre, I think there's more to it. Consider the the inclusion of the terminal object into the category of manifolds. This induces a geometric embedding from SET into sheaves, which constitutes a Grothendieck topology $S$ on $Mfd$. Concrete sheaves are sheaves for the open cover topology which are $S$-separated. I'm just trying to see what this means "concretely" (ugh, tried to avoid the pun). – David Carchedi Dec 13 '10 at 18:35