11
$\begingroup$

Let $k$ be an algebraically closed field of characteristic zero. Formalizing a classical folk concept, Pridham and (in a different way,) Lurie defined a formal moduli problem (over $k$) to be a functor from local Artin CDGA's to homotopy types satisfying a certain sheaf condition. If the commutativity condition is weakened to an $E_n$ condition, any formal moduli problem is (uniquely) representable by an $E_n$ algebra; in the commutative case, a formal moduli problem is not necessarily representable by a CDGA, but rather by a (derived) Lie algebra.

There's an intuitive picture I like for this in a special case, and I want to understand how it fits into the general picture. Namely, say that $G$ is an affine algebraic Lie group. Let $BG$ be its classifying stack. Then the deformation problem of maps to $BG$ (relative to a choice of point $*\to BG$) is classified by the Lie group $\mathfrak{g}$.

Of course this picture involves a group rather than a "formal stack", which is my intuition for (the opposite category to) formal moduli problems, and I am curious how hard it is for a formal moduli problem to be representable by a group object.

From the point of view of spaces, going from groups to spaces "is not very hard": the category of groups is equivalent (via the delooping functor) to the category of connected pointed spaces.

On the other hand, for $E_n$ algebras, the category of cogroup objects is equivalent to the category of (suitably defined) $(n,1)$-commutative Hopf algebras, which is more closely related (via Koszul duality) to the category of $E_{n+1}$ algebras than to $E_n$ algebras; in particular, the delooping functor from group objects to $E_n$ algebras is far from fully faithful.

Now by formal nonsense, there is a loop-deloop pair of adjoint functors $$B:GFMP\leftrightarrows FMP:\Omega,$$ where $FMP$ is the category of ($E_\infty$) formal moduli problems and $GFMP$ is the category of cogroup objects in $FMP$. It seems natural to ask the following questions, to which I don't know the answer:

  1. Is $B$ fully faithful, and if so what is its image?
  2. Is there an analogue on the level of (co)group objects in Lie algebras to the Koszul duality functor from $(n,1)$ Hopf algebras to $E_{n+1}$ algebras?

A natural further question, which I suspect is harder, is to ask whether there is an "algebraic point of view" via Lie-like objects for $E_\infty$ objects in $FMP$.

$\endgroup$
4
  • $\begingroup$ How are you defining $B$? The nerve functor from groups to unpointed spaces doesn't preserve fibrations or homotopy fibre products. You need a smoothness condition on $G$ for $BG$ to be a "formal moduli problem". You might also find the original proof of the equivalence helpful; see Theorem 2.30 and Corollary 4.57 of arxiv.org/abs/0705.0344 . $\endgroup$ Mar 1, 2020 at 9:08
  • $\begingroup$ and there's no need for your algebraically closed hypothesis. $\endgroup$ Mar 1, 2020 at 9:19
  • $\begingroup$ @JonPridham Edited the question to incorporate your authorship. Let me know if there is a more context that I should be included (I don't know the timeline here). I was purposefully vague about the functors $\Omega$ and $B$, but are you saying that there are serious problems with defining a functor from group objects to objects which takes a geometric group $G$ to the moduli problem classified by $BG$ (principal bundles with trivialization at the marked point)? $\endgroup$ Mar 31, 2020 at 22:45
  • $\begingroup$ Thanks (the timeline is 2007 vs 2010 for first appearance of the comparison, as it happens). There aren't serious problems defining the functor $B$ that you want; it's just the universal FMP under the presheaf given by applying $B$ objectwise. That tends to have non-trivial $\pi_0$, so is bigger than naive moduli of principal bundles over a fat point. $\endgroup$ Mar 31, 2020 at 23:37

1 Answer 1

12
$\begingroup$

The presentation of the formal moduli problems story in Gaitsgory-Rosenblyum A Study in Derived Algebraic Geometry, Vol 2 may be what you are looking for. We review it here (in the case over $\mathrm{Spec}\, k$ for a field $k$ of characteristic zero, that the question concerns):

1. Looping/delooping equivalence in formal DAG

Just like the familiar adjoint equivalence in the homotopy theory of spaces $$\mathrm B: \mathrm{Grp}_{\mathbb E_1}(\mathcal S)\simeq \mathcal S_*^{\ge 1}:\Omega,$$ there is an analogous adjoint equivalence in formal DAG $$\mathrm B:\mathrm{Grp}_{\mathbb E_1}(\mathrm{FMP}_k) \simeq \mathrm{FMP}_k:\Omega,$$ where the loop space functor is in both cases given by $\Omega X = \mathrm{pt}\times_X \mathrm{pt}$, as per usual in homotopical/$\infty$-categorical things. Note that formal moduli problems are already inherently pointed, by the assumption that $X(k)$ is contractible.

2. Formal groups and Lie algebras

Now, $\mathrm{Grp}_{\mathbb E_1}(\mathrm{FMP}_k) = \mathrm{FGrp}_k$ is (an incarnation) of the $\infty$-category of (derived) formal groups over $k$. Thanks to the characteristic zero assumption, there is a further equivalence $$ \mathrm{Lie}:\mathrm{FGrp}_k \simeq \mathrm{LieAlg}_k : \exp $$ with derived Lie algebras (as modelled for instance by dg Lie algebras). Just like expected, the derived Lie algebra $\mathfrak g$ corresponding to the formal group $G$ is $\mathfrak g = T_{G, e}$, the tangent fiber at the unit.

3. Formal moduli problems and Lie algebras

The celebrated Lurie-Pridham identification between formal moduli problems and derived Lie algebras is precisely the composite of these two equivalences of $\infty$-categories $$ \mathrm{FMP}_k\simeq \mathrm{Grp}_{\mathbb E_1}(\mathrm{FMP}_k)=\mathrm{FGrp}_k\simeq \mathrm{LieAlg}_k. $$ That is

  • It sends a formal moduli problem $X$ to $$ \mathrm{Lie}(\Omega X) = T_{\Omega X, e} = T_{X, x_0}[-1], $$ where $x_0$ is the base-point of $X$, unique up to a contractible space of choices $X(\kappa)$. This shifted tangent fiber carries a canonical Lie algebra structure coming from (i.e. as the Lie algebra of) the $\mathbb E_1$-group structure of $\Omega X$.
  • The inverse equivalence $\Psi: \mathrm{LieAlg}_k \simeq \mathrm{FMP}_k$ is then given by $\Psi(\mathfrak g)= \mathrm B\exp(\mathfrak g)$.

4. Lurie's formula for $\Psi$

You may justifiably complain that this description of the inverse functor $\Psi$ does not look the same as the one in Lurie's writing. Let's see how to get it in that form.

Let's assume that the formal moduli problem $\Psi(\mathfrak g)$ is formally affine (true under some finiteness assumptions on $\mathfrak g$), in the sense that $$ \mathrm B\exp(\mathfrak g)= (\mathrm{Spf}\,k)/\exp(\mathfrak g)\simeq \mathrm{Spf} \,k^{\mathfrak g}. $$ Here of course the formal spectrum is the usual functor $\mathrm{Spf}:(\mathrm{CAlg}^\mathrm{aug}_k)^\mathrm{op}\to \mathrm{FMP}_k$. The derived Lie algebra invariants may be computed via the Chevalley-Eilenberg complex, thus $k^\mathfrak g\simeq {C}^*(\mathfrak g)$. Then for any Artinian $k$-algebra $A$ we have $$ (\Psi(\mathfrak g))(A) \simeq \mathrm{Map}_{\mathrm{CAlg}^\mathrm{aug}_k}({C}^*(\mathfrak g), A). \qquad \quad(1) $$ This is why Lurie tells us to consider the "Koszul duality functor" $\mathfrak D: (\mathrm{CAlg}_k^\mathrm{aug})^\mathrm{op}\to \mathrm{LieAlg}_k,$ right-adjoint to the Chevalley-Eilenberg cochains functor. Indeed, (through a little use of the finiteness of $\mathfrak g$) we get $$(\Psi(\mathfrak g))(A) \simeq \mathrm{Map}_{\mathrm{LieAlg}_k}( \mathfrak D(A), \mathfrak g),\qquad \qquad(2) $$ which is how Lurie tells us to define $\Psi$.

Note that this is going the other way than the $C^*\dashv\mathfrak D$ adjunction. Indeed, without finiteness assumptions on $\mathfrak g$, the formal moduli problem $\Psi(\mathfrak g)$ will not necessarily be formally affine, and the formula (1) will not necessarily work. On the other hand, as Lurie teaches us, formula (2) will always work.

5. Loose ends

If I understand the original question correctly, this is presenting the formal moduli problem story precisely like the intuitive picture mentioned. Indeed: any formal moduli problem may be written as $X\simeq \mathrm B G$ for a derived formal group $G$, and the formal moduli problem (of mapping into) $\mathrm BG$ is classified by the Lie algebra $\mathfrak g$.

That said, the question makes analogy with the $\mathbb E_n$-algebra version of this story too. I am a little confused about the points raised - in particular, it seems like the following is asserted: a formal moduli problem on $\mathbb E_n$-algebras is represented by an $\mathbb E_n$-algebra. That is, so far as I understand, incorrect. Instead, any FMP on $\mathbb E_n$-algebras in classified by an $\mathbb E_n$-algebra, in a way somewhat analogous to the way that the functor $\mathfrak D$ presents formal moduli problems in the commutative case with Lie algebras.

In particular, there seems to be no need to think about cogroup objects, as the equivalence of $\infty$-categories $\mathrm{FMP}_k^{\mathbb E_n}\simeq \mathrm{Alg}^\mathrm{aug}_{\mathbb E_n}$ is covariant.

$\qquad$


Edit: A question was raised in the comments whether group objects in $\mathrm{FMP}_k^{\mathbb E_n}\simeq \mathrm{Alg}_{\mathbb E_n}^\mathrm{aug}$ correspond to $\mathbb E_{n+1}$-algebras. Unless I am misunderstanding the question, the answer is negative.

A monoid structure (of which a group structure is a particular example of) on an $\mathbb E_n$-algebra $A$ is given by a map $A\times A\to A$, plus coherence data. On the other hand, an additional $\mathbb E_1$-algebra structure on $A$ (which is equivalent to making it into an $\mathbb E_{n+1}$-algebra by Dunn Additivity) is given by a map $A\otimes_k A\to A$, plus coherence data. So a group object in $\mathbb E_n$-algebras, and an $\mathbb E_{n+1}$-algebra are different structures.

This might feel a little weird because we're used to tensor products corresponding to products of schemes. Alas, unlike the contravariant $\mathrm{Spec}$, the equivalence $\Psi:\mathrm{Alg}^{\mathrm{aug}}_{\mathbb E_n}\simeq \mathrm{FMP}_k^{\mathbb E_n}$  is covariant. (PS: another difficulty: the tensor product only becomes the coproduct on the level of $\mathbb E_\infty$-algebras, not $\mathbb E_n$-algebras, so even a contravariant equivalence wouldn't necessarily do the trick).

Rather, the situation is the same as in the $\mathbb E_\infty$-case above: there is an equivalence $$\mathrm B:\mathrm{Grp}_{\mathbb E_1}(\mathrm{FMP}_k^{\mathbb E_n})\simeq \mathrm{FMP}_k^{\mathbb E_n}:\Omega.$$ This is a special case of the following general phenomenon: for any operad $\mathcal O$, the loops functor $\Omega: \mathrm{Alg}_{\mathcal O}(\mathrm{Mod}_k)\to \mathrm{Grp}_{\mathbb E_1}(\mathrm{Alg}_{\mathcal O}(\mathrm{Mod}_k))$ is an equivalence of $\infty$-categories (A Sudy in Derived Algebraic Geometry, Vol 2, Chapter 6, Proposition 1.6.4). The above claims are special cases for $\mathcal O$ the Lie operad and the $\mathbb E_n$-operad respectively.

$\endgroup$
4
  • 1
    $\begingroup$ Thanks! This is a funny property, that the delooping functor is an equivalence but the category is not stable. I wonder if there are other examples. And thanks for the correction: $E_n$ moduli problems are represented by $E_n$ algebras via a duality functor, so presumably a group object (in a geometric sense) would be representable by an $E_{n+1}$ algebra. $\endgroup$ Mar 3, 2020 at 2:33
  • $\begingroup$ Reiterating a comment I made on the question, BG is not good notation for an infinitesimal functor with many objects (essentially the Deligne groupoid in this case). $\endgroup$ Dec 11, 2021 at 23:50
  • $\begingroup$ @JonPridham: Agreed! But I don't quite see why this objection applies to the case at hand. (Pardon if I am missing something!) One of the requirements in the definition of a formal moduli problem is that $X(k)$, which may be thought of as the space of $k$-valued points of $X$, must be contractible. So $X$ has only a single "object", at least up homotopy (which is the best we can ask for anyway). On the other hand, Gaitsgory-Roseblyum also have a version of the correspondence between Lie algebroids and some kind of multi-object FMPs, where the objection to using $B$ might well be applicable. $\endgroup$ Dec 13, 2021 at 0:18
  • $\begingroup$ The point is that $X(k)$ is the only connected space; the spaces $X(A)$ for Artinian rings are not, unless the DGLA in concentrated in non-positive cochain degrees. Contrast this with the notation $BG$ for the stack of $G$-torsors, whose stalks are connected in the sense that $(BG)(A)= B(G(A))$ for any Henselian ring $A$. $\endgroup$ Dec 13, 2021 at 10:13

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.