$\DeclareMathOperator\IndCoh{IndCoh}\DeclareMathOperator\QCoh{QCoh}$I've seen it written (for example, in Gaitsgory–Rozenblyum) that for a scheme $X$, the category $\IndCoh(X)$ is to be thought of as a categorificaiton of the space of distributions on $X$, just as the (derived) category of quasi-coherent sheaves is to be thought of as a categorification of functions on $X$.

Why exactly is $\IndCoh(X)$ to be thought of as dual to $\QCoh(X)$? The only relation between these categories that I know of is that $\IndCoh(X)$ is a module over $\QCoh(X)$. Is there some way that an ind-coherent sheaf can act on a quasi-coherent one to yield a categorical analog of a point (for example, an object in $\QCoh(\{ \mathrm{pt} \})$)?


1 Answer 1


The connection is perhaps a bit more clear if you think about the corresponding $\infty$-categories of compact objects: the claim is that coherent sheaves behave like distributions while perfect sheaves behave like functions. This connection manifests itself in several ways:

  • Coherent sheaves pushforward along proper maps, while perfect complexes pullback along arbitrary maps. Similarly, distributions integrate along proper maps while functions pullback along arbitrary maps.
  • Skyscrapper sheaves at closed points are coherent but not perfect in general, in the same way that Delta functions are distributions but not functions.
  • When $X$ is proper, the $\infty$-category of coherent sheaves on $X$ is dual to the $\infty$-category of perfect complexes on $X$, see theorem 1.1.3 in https://arxiv.org/abs/1312.7164. Concretely, each coherent sheaf $\mathcal{G}$ defines a functional sending a perfect sheaf $\mathcal{F}$ to the global sections of $\mathcal{F} \otimes \mathcal{G}$.
  • There is a notion of singular support for coherent sheaves, which is a subset of the shifted cotangent bundle measuring how far the sheaf is from being perfect, see https://arxiv.org/abs/1201.6343. Similarly, distributions have a wave front set, which is a subset of the cotangent bundle measuring how far the distribution is from being a smooth function.
  • The Hochschild homology of the $\infty$-category of coherent sheaves on $X$ recovers the space of distributions on the loop space of $X$, while the Hochschild homology of the $\infty$-category of perfect sheaves on $X$ recovers the space of functions on the loop space $X$.
  • $\begingroup$ Thanks for the thorough answer. Do you have a reference for the last bullet point? $\endgroup$ Feb 21 at 22:10
  • $\begingroup$ @JustLikeNumberTheory I guess that it is contained in Preygel's Thom–Sebastiani paper B.5.1–B.5.2. $\endgroup$
    – Z. M
    Feb 21 at 22:33
  • $\begingroup$ In addition to Preygel, there is also arxiv.org/abs/1305.7175 (theorem 1.4 or proposition 5.3) which explains how Hochschild homology formulas can be deduced from the formulation of these sheaf theories in terms of functors out of $\infty$-categories of correspondences. $\endgroup$ Feb 22 at 1:24
  • $\begingroup$ This is a great answer, but I would like to highlight one place where the analogy falters: for smooth spaces both categories are the same (and the delta functions are actual functions). So maybe a tighter analogy would be that Perf is like cohomology/the de Rham complex, and Coh is like Borel-Moore homology/ the complex of currents. $\endgroup$ Feb 22 at 3:39
  • $\begingroup$ Let me add to this great answer that for an automorphism $\gamma$ of $X$, its trace on Coh (Perf) gives the distributions (functions) on fixed points $X^\gamma$. For $\gamma=Id$ this is the last bullet point (or, taking S^1-Tate construction, the previous commenter's analogy). But for example taking $\gamma$=Frobenius acting on $X$=local systems on a curve over $\overline{\mathbb F}_q$ we get distributions / functions on local systems over $\mathbb F_q$ -- this is the coherent counterpart to the Grothendieck function/sheaf correspondence for etale sheaves. $\endgroup$ Mar 8 at 5:58

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.