ScalingStacks

5.2. Centers of convolution categories

Let p:X→Yp:X\to Y be a map of perfect stacks satisfying descent. Consider the convolution diagram

X×YX×YX\textstyle{\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces X\times_{Y}X\times_{Y}X}p12\scriptstyle{p_{12}}p23\scriptstyle{p_{23}}p13\scriptstyle{p_{13}}X×YX\textstyle{X\times_{Y}X}X×YX\textstyle{X\times_{Y}X}X×YX\textstyle{X\times_{Y}X}

Equip QC⁡(X×YX)\qc(X\times_{Y}X) with the monoidal product defined by convolution

M⋆N=p13∗(p12∗(M)⊗p23∗(N)).M\star N=p_{13*}(p_{12}^{*}(M)\otimes p_{23}^{*}(N)).

By Theorem 4.14, we have a monoidal equivalence

QC⁡(X×YX)≃FunY⁡(QC⁡(X),QC⁡(X)).\qc(X\times_{Y}X)\simeq\Fun_{Y}(\qc(X),\qc(X)).

Consider the fundamental correspondence

ℒ​Y=Y×Y×YY\textstyle{\mathcal{L}Y=Y\times_{Y\times Y}Y}ℒ​Y×YX=X×X×YX\textstyle{\mathcal{L}Y\times_{Y}X=X\times_{X\times Y}X\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}π\scriptstyle{\pi}δ\scriptstyle{\delta}X×YX.\textstyle{X\times_{Y}X.}

where the maps are defined by the formulas

π=idℒ​Y×idYp=p×p×idYpδ=idX×πYidX.\pi={\rm id}_{\mathcal{L}Y}\times_{{\rm id}_{Y}}p=p\times_{p\times{\rm id}_{Y}}p\qquad\delta={\rm id}_{X}\times_{\pi_{Y}}{\rm id}_{X}.

where πY:X×Y→Y\pi_{Y}:X\times Y\to Y is the obvious projection.

Passing to sheaves, we obtain a diagram

QC⁡(ℒ​Y)\textstyle{\qc(\mathcal{L}Y)\ignorespaces\ignorespaces\ignorespaces\ignorespaces}π∗\scriptstyle{\pi^{*}}QC⁡(ℒ​Y×YX)\textstyle{\qc(\mathcal{L}Y\times_{Y}X)\ignorespaces\ignorespaces\ignorespaces\ignorespaces}δ∗\scriptstyle{\delta_{*}}QC⁡(X×YX)\textstyle{\qc(X\times_{Y}X)}

which by Theorem 4.14 admits the interpretation

FunY×Y⁡(QC⁡(Y),QC⁡(Y))\textstyle{\Fun_{Y\times Y}(\qc(Y),\qc(Y))\ignorespaces\ignorespaces\ignorespaces\ignorespaces}π∗\scriptstyle{\pi^{*}}FunX×Y⁡(QC⁡(X),QC⁡(X))\textstyle{\Fun_{X\times Y}(\qc(X),\qc(X))\ignorespaces\ignorespaces\ignorespaces\ignorespaces}δ∗\scriptstyle{\delta_{*}}FunY⁡(QC⁡(X),QC⁡(X)).\textstyle{\Fun_{Y}(\qc(X),\qc(X)).}

where π∗\pi^{*} is the QC⁡(X)\qc(X)-linear induction, and δ∗\delta_{*} forgets the QC⁡(X)\qc(X)-linear structure.

The aim of this section is to prove the following.

0NYE

Theorem 5.3. Suppose p:X→Yp:X\to Y is a map of perfect stacks satisfying descent. Then there is a canonical equivalence

𝒵⁡(QC⁡(X×YX))≃QC⁡(ℒ​Y)\mathcal{Z}(\qc(X\times_{Y}X))\simeq\qc(\mathcal{L}Y)

such that the forgetful functor 𝔷:𝒵⁡(QC⁡(X×YX))→QC⁡(X×YX)\mathfrak{z}:\mathcal{Z}(\qc(X\times_{Y}X))\to\qc(X\times_{Y}X) is given by the correspondence δ∗​π∗:QC⁡(ℒ​Y)→QC⁡(X×YX)\delta_{*}\pi^{*}:\qc(\mathcal{L}Y)\to\qc(X\times_{Y}X).

The proof occupies the remainder of this section. We break up the argument into a general discussion and then its specific application.

At the end of the section, we also explain an analogous description of traces, conditional on a still to be developed version of Grothendieck duality in the derived setting. Namely, assuming further that p:X→Yp:X\to Y is proper with invertible dualizing sheaf and Grothendieck duality holds, we explain how to deduce an expected canonical equivalence

𝒯​r​(QC⁡(X×YX))≃QC⁡(ℒ​Y)\mathcal{T}r(\qc(X\times_{Y}X))\simeq\qc(\mathcal{L}Y)

such that the trace 𝔱​𝔯:QC⁡(X×YX)→𝒯​r​(QC⁡(X×YX))\mathfrak{tr}:\qc(X\times_{Y}X)\to\mathcal{T}r(\qc(X\times_{Y}X)) is given by the correspondence π∗​δ∗:QC⁡(X×YX)→QC⁡(ℒ​Y)\pi_{*}\delta^{*}:\qc(X\times_{Y}X)\to\qc(\mathcal{L}Y).

5.2.1. Relative cyclic bar construction

For the proof of Theorem 5.3, it will be useful to introduce relative versions of the Hochschild chain and cochain complexes. In general, the setup for the construction will be a map of associative algebra objects B→AB\to A in a monoidal ∞\infty-category 𝒮\mathcal{S}. The usual Hochschild chain and cochain complexes introduced in the previous section will correspond to the case when BB is the unit 1𝒮1_{\mathcal{S}}.

Consider the adjunction

ModB⊗Aop​(𝒮)\textstyle{\mathrm{Mod}_{B\otimes A^{\rm op}}(\mathcal{S})\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}F\scriptstyle{F}ModA⊗Aop​(𝒮)\textstyle{\mathrm{Mod}_{A\otimes A^{\rm op}}(\mathcal{S})\ignorespaces\ignorespaces\ignorespaces\ignorespaces}G\scriptstyle{G}

where F(−)=A⊗B−F(-)=A\otimes_{B}- is the induction, and GG is the forgetful functor from AA-bimodules to B⊗AopB\otimes A^{\rm op}-modules. Using the comonad S≃F​GS\simeq FG, we obtain a simplicial resolution C∗B​(A)C^{B}_{*}(A) of the AA-bimodule AA with terms

Cn−1B(A)≃A⊗B⊗⋯⊗BA, with n+1 terms.C^{B}_{n-1}(A)\simeq A\otimes_{B}\otimes\cdots\otimes_{B}A,\quad\mbox{ with $n+1$ terms.}

To reduce the notation, we will denote the above expression by AB⊗n+1A_{B}^{\otimes n+1}.

As before, the geometric realization of A⊗A⊗AopC∗B​(A)A\otimes_{A\otimes A^{\rm op}}C^{B}_{*}(A) calculates the trace 𝒯​r​(A)\mathcal{T}r(A), and the totalization of ℋ​o​mA⊗Ao​p​(C∗B​(A),A){\mathcal{H}om}_{A\otimes A^{op}}(C^{B}_{*}(A),A) calculates the center 𝒵⁡(A)\mathcal{Z}(A). We write 𝐍∗B​(A)\mathbf{N}^{B}_{*}(A) for the simplicial object A⊗A⊗AopC∗B​(A)A\otimes_{A\otimes A^{\rm op}}C^{B}_{*}(A) and refer to it as the relative Hochschild chain complex. Similarly, we write 𝐍B∗​(A)\mathbf{N}_{B}^{*}(A) for the cosimplicial object ℋ​o​mA⊗Ao​p​(C∗B​(A),A){\mathcal{H}om}_{A\otimes A^{op}}(C^{B}_{*}(A),A) and refer to it as the relative Hochschild cochain complex.

Continuing as before, we can evaluate the terms of the relative Hochschild chain complex

𝐍nB​(A)=A⊗A⊗AopCnB​(A)≃A⊗A⊗AopAB⊗n+2≃B⊗B⊗BopAB⊗n+1\mathbf{N}^{B}_{n}(A)=A\otimes_{A\otimes A^{\rm op}}C^{B}_{n}(A)\simeq A\otimes_{A\otimes A^{\rm op}}A_{B}^{\otimes n+2}\simeq B\otimes_{B\otimes B^{\rm op}}A_{B}^{\otimes n+1}

Similarly, we can evaluate the terms of the relative Hochschild cochain complex

𝐍Bn​(A)=ℋ​o​mA⊗Ao​p​(CnB​(A),A)≃ℋ​o​mA⊗Ao​p​(AB⊗n+2,A)≃ℋ​o​mB⊗Bop​(AB⊗n+1,B).\mathbf{N}_{B}^{n}(A)={\mathcal{H}om}_{A\otimes A^{op}}(C^{B}_{n}(A),A)\simeq{\mathcal{H}om}_{A\otimes A^{op}}(A_{B}^{\otimes n+2},A)\simeq{\mathcal{H}om}_{B\otimes B^{\rm op}}(A_{B}^{\otimes n+1},B).

5.2.2. Proof of Theorem 5.3

Now we will apply the preceding formalism to calculate the center of the monoidal ∞\infty-category QC⁡(X×YX)\qc(X\times_{Y}X). Recall that our aim is to show that there is a canonical equivalence

𝒵⁡(QC⁡(X×YX))≃QC⁡(ℒ​Y).\mathcal{Z}(\qc(X\times_{Y}X))\simeq\qc(\mathcal{L}Y).

where ℒ​Y\mathcal{L}Y is the loop space of YY.

Consider the symmetric monoidal ∞\infty-category QC⁡(X)\qc(X), together with the monoidal functor

Δ∗:QC⁡(X)→QC⁡(X×YX)\Delta_{*}:\qc(X)\to\qc(X\times_{Y}X)

obtained via pushforward along the relative diagonal Δ:X→X×YX\Delta:X\to X\times_{Y}X.

Set A=QC⁡(X×YX)A=\qc(X\times_{Y}X), and B=QC⁡(X)B=\qc(X). By the preceding discussion, the center 𝒵⁡(A)\mathcal{Z}(A) is the limit of the relative Hochschild cochain complex 𝐍B∗​(A)\mathbf{N}^{*}_{B}(A). Furthermore, its terms can be calculated

𝐍Bn​(A)≃ℋ​o​mB⊗Bop​(AB⊗n+1,B).\mathbf{N}_{B}^{n}(A)\simeq{\mathcal{H}om}_{B\otimes B^{\rm op}}(A_{B}^{\otimes n+1},B).

Applying the results of Section 4, we can rewrite each term in the form

𝐍Bn(A)≃QC(ℒY×YX×Y⋯×YX), with n+1 copies of X.\mathbf{N}_{B}^{n}(A)\simeq\qc(\mathcal{L}Y\times_{Y}X\times_{Y}\cdots\times_{Y}X),\quad\mbox{ with $n+1$ copies of $X$.}

Here we have used the elementary identification

ℒY×YX×Y⋯×YX≃X×X×X((X×YX)×X⋯×X(X×YX))\mathcal{L}Y\times_{Y}X\times_{Y}\cdots\times_{Y}X\simeq X\times_{X\times X}((X\times_{Y}X)\times_{X}\cdots\times_{X}(X\times_{Y}X))

where the left hand side has n+1n+1 copies of XX, and the right hand side has n+1n+1 copies of X×YXX\times_{Y}X.

We conclude that the terms of 𝐍B∗​(A)\mathbf{N}_{B}^{*}(A) are nothing more than the terms of the cosimplicial ∞\infty-category obtained by applying QC⁡(−)\qc(-) to the C̆ech simplicial stack induced by the map

p~:ℒ​Y×YX→ℒ​Y\tilde{p}:\mathcal{L}Y\times_{Y}X\to\mathcal{L}Y

obtained by base change from the original map p:X→Yp:X\to Y. It is straightforward to check that under this identification the coboundary maps are given by the usual C̆ech pullbacks. By assumption, pp satisfies descent, hence p~\tilde{p} satisfies descent, and thus the totalization lim𝐍B∗​(A)\lim\mathbf{N}_{B}^{*}(A) also calculates the ∞\infty-category QC⁡(ℒ​Y)\qc(\mathcal{L}Y).

Finally, note that under this identification, the composition δ∗​π∗:QC⁡(ℒ​Y)→QC⁡(X×YX)\delta_{*}\pi^{*}:\qc(\mathcal{L}Y)\to\qc(X\times_{Y}X) corresponds to the composition

QC⁡(ℒ​Y)≃lim𝐍B∗​(A)→𝐍B0​(A)→𝐍c​y​c0​(A)≃QC⁡(X×YX)\qc(\mathcal{L}Y)\simeq\lim\mathbf{N}_{B}^{*}(A)\to\mathbf{N}_{B}^{0}(A)\to\mathbf{N}_{cyc}^{0}(A)\simeq\qc(X\times_{Y}X)

which is precisely the central functor 𝔷\mathfrak{z}. This concludes the proof of Theorem 5.3.

5.2.3. Traces and Grothendieck duality

Finally, we explain here an analogous description of traces, conditional on a still to be developed version of Grothendieck duality in the derived setting. Namely, assuming further that p:X→Yp:X\to Y is proper with invertible dualizing sheaf and Grothendieck duality holds, we explain how to deduce an expected canonical equivalence

𝒯​r​(QC⁡(X×YX))≃QC⁡(ℒ​Y)\mathcal{T}r(\qc(X\times_{Y}X))\simeq\qc(\mathcal{L}Y)

such that the trace 𝔱​𝔯:QC⁡(X×YX)→𝒯​r​(QC⁡(X×YX))\mathfrak{tr}:\qc(X\times_{Y}X)\to\mathcal{T}r(\qc(X\times_{Y}X)) is given by the correspondence π∗​δ∗:QC⁡(X×YX)→QC⁡(ℒ​Y)\pi_{*}\delta^{*}:\qc(X\times_{Y}X)\to\qc(\mathcal{L}Y).

We continue with the notation from the proof of Theorem 5.3 and the preceding sections. We will show that the geometric realization colim⁡𝐍∗B​(A)\colim\mathbf{N}^{B}_{*}(A) of the relative Hochschild chain complex also calculates the ∞\infty-category QC⁡(ℒ​Y)\qc(\mathcal{L}Y). As before, applying the results of Section 4, we can rewrite the terms of 𝐍∗B​(A)\mathbf{N}^{B}_{*}(A) in the form

𝐍nB(A)≃QC(ℒY×YX×Y⋯×YX) with n+1 copies of X.\mathbf{N}^{B}_{n}(A)\simeq\qc(\mathcal{L}Y\times_{Y}X\times_{Y}\cdots\times_{Y}X)\quad\mbox{ with $n+1$ copies of $X$.}

Furthermore, it is straightforward to check that under this identification the boundary maps of 𝐍∗B​(A)\mathbf{N}^{B}_{*}(A) are given by the pushforwards 𝔭n∗\mathfrak{p}_{n*} which are right adjoints to the usual C̆ech pullbacks 𝔭n∗\mathfrak{p}_{n}^{*}.

To reduce notation, set Xn=X×Y⋯×YXX_{n}=X\times_{Y}\cdots\times_{Y}X with n+1n+1 copies of XX.

Now suppose that p:X→Yp:X\to Y is proper and has an invertible dualizing complex (Gorenstein). Then we expect Grothendieck duality to hold in the following form: the pushforwards 𝔭n∗\mathfrak{p}_{n*} are also left adjoints to the pullbacks

𝔭n!(−)≃𝔭n∗(−⊗ωℒ​Y×YXn−1/Y−1)⊗ωℒ​Y×YXn/Y,\mathfrak{p}_{n}^{!}(-)\simeq\mathfrak{p}_{n}^{*}(-\otimes\omega^{-1}_{\mathcal{L}Y\times_{Y}X_{n-1}/Y})\otimes\omega_{\mathcal{L}Y\times_{Y}X_{n}/Y},

where ωℒ​Y×YXn/Y\omega_{\mathcal{L}Y\times_{Y}X_{n}/Y} denotes the relative dualizing sheaf of ℒ​Y×YXn→Y\mathcal{L}Y\times_{Y}X_{n}\to Y. Under this assumption, we find that the limit QC⁡(ℒ​Y)\qc(\mathcal{L}Y) of the cosimplicial ∞\infty-category (QC⁡(ℒ​Y×YXn),𝔭n∗)(\qc(\mathcal{L}Y\times_{Y}X_{n}),\mathfrak{p}_{n}^{*}) admits the following alternative description.

First, we can identify the cosimplicial ∞\infty-category (QC⁡(ℒ​Y×YXn),𝔭n∗)(\qc(\mathcal{L}Y\times_{Y}X_{n}),\mathfrak{p}_{n}^{*}) with the cosimplicial ∞\infty-category (QC(ℒY×YXn,𝔭n!)(\qc(\mathcal{L}Y\times_{Y}X_{n},\mathfrak{p}_{n}^{!}) via tensoring by the inverse of the relative dualizing sheaf ωℒ​Y×YXn/Y\omega_{\mathcal{L}Y\times_{Y}X_{n}/Y} on each simplex. In particular, we obtain an identification of their limits.

Second, we can consider the cosimplicial ∞\infty-category (QC(ℒY×YXn),𝔭n!)(\qc(\mathcal{L}Y\times_{Y}X_{n}),\mathfrak{p}_{n}^{!}) as a diagram in the ∞\infty-category 𝒫​rR\mathcal{P}r^{\rm R} of presentable ∞\infty-categories (with morphisms right adjoints). By [L2, Theorem 5.5.3.18], the calculation of the limit of the diagram does not depend on this choice of context. Then we can pass to the opposite ∞\infty-category 𝒫​rL\mathcal{P}r^{\rm L} of presentable ∞\infty-categories (with morphisms left adjoints). To calculate a limit in 𝒫​rR\mathcal{P}r^{\rm R} is the same as to calculate a colimit in 𝒫​rL\mathcal{P}r^{\rm L}. But we have seen that the trace 𝒯​r​(ℋ)\mathcal{T}r(\mathcal{H}) is precisely the colimit of the dual simplicial diagram (QC(ℒY×YXn),𝔭n∗)(\qc(\mathcal{L}Y\times_{Y}X_{n}),\mathfrak{p}_{n*}). Thus we conclude that the geometric realization colim⁡𝐍∗B​(A)\colim\mathbf{N}_{*}^{B}(A) also calculates the ∞\infty-category QC⁡(ℒ​Y)\qc(\mathcal{L}Y).

Finally, note that under this identification, the composition π∗​δ∗:QC⁡(X×YX)→QC⁡(ℒ​Y)\pi_{*}\delta^{*}:\qc(X\times_{Y}X)\to\qc(\mathcal{L}Y) corresponds to the composition

QC⁡(X×YX)≃𝐍0c​y​c​(A)→𝐍0B​(A)→colim⁡𝐍∗c​y​c​(A)≃QC⁡(ℒ​Y)\qc(X\times_{Y}X)\simeq\mathbf{N}_{0}^{cyc}(A)\to\mathbf{N}_{0}^{B}(A)\to\colim\mathbf{N}_{*}^{cyc}(A)\simeq\qc(\mathcal{L}Y)

which is precisely the trace 𝔱​𝔯\mathfrak{tr}.

Original mathematics by the credited authors. Source collection and HTML conversion remain in progress.

David Ben-Zvi, John Francis, David Nadler

Original source: arXiv:0805.0157v5