Proposition and proof

The key lemma has its own proof graph. Click a symbol with a dotted underline, or a hypothesis such as H2, to recall its definition.

Definitions
  • Setting

    \(k\) is algebraically closed of characteristic zero; \(G\supseteq B\supseteq T\) a connected reductive group, a Borel and a maximal torus; \(W=N_G(T)/T\). Evaluation at \(t=0\) defines the arc group, the Iwahori subgroup and the two homogeneous ind-schemes:

    $$K=L^+G\xrightarrow{\ \mathrm{ev}\ }G,\qquad I=\mathrm{ev}^{-1}(B),\qquad\mathrm{Gr}=LG/K,\qquad\mathrm{Fl}=LG/I,\qquad d=\dim G/B.$$

    \(X\) is a smooth affine spherical \(G\)-variety with connected \(B\)-stabilizers on its open \(B\)-orbit (paper, Definition 4.2); \(LX\) is its loop space and \(L^+X\) the closed orbit of the base point. Coefficients are \(k\).

  • Categories

    \(\mathcal D(-)\) is the DG category of \(\mathcal D\)-modules on a prestack, defined by \(!\)-pullback from finite type schemes, and \(\mathcal D_c(-)\subseteq\mathcal D(-)\) its small, idempotent-complete, stable subcategory of coherent objects: those whose pullback to every smooth test scheme is a bounded complex with coherent cohomology (paper, Definition 3.27). The categories of the Proposition:

    $$\mathcal H_{c,K}=\mathcal D_c(K\backslash\mathrm{Gr}),\qquad\mathcal H_{c,I}=\mathcal D_c(I\backslash LG/I),\qquad\mathcal D_c(I\backslash\mathrm{Gr})=\mathcal D_c(I\backslash LG/K),\qquad\mathcal D_c(K\backslash LX),\ \ \mathcal D_c(I\backslash LX).$$

    \(\mathcal H_{c,K}\) and \(\mathcal H_{c,I}\) are monoidal under convolution; \(\mathcal D_c(I\backslash\mathrm{Gr})\) is a right \(\mathcal H_{c,K}\)-module, \(\mathcal D_c(K\backslash LX)\) a left one, and \(\mathcal D_c(I\backslash LX)\) a left \(\mathcal H_{c,I}\)-module.

  • The basic object and the convolutions

    Basic object. \(L^+X=X(\mathcal O)\) is closed in \(LX=X(F)\) because \(X\) is affine, and it is the \(K\)-orbit of the base point. \(\mathrm{IC}_0\in\mathcal D_c(K\backslash LX)\) is the constant \(\mathcal D\)-module on the closed orbit \(K\backslash L^+X\), extended by zero (\(!\)- and \(*\)-extension agree along a closed embedding); its function shadow is \(\mathbf 1_{X(\mathcal O)}\). So \(\mathrm{IC}_0\) is \(K\)-equivariant. Along the forgetful map \(p\) it pulls back to an \(I\)-equivariant object, written \(\mathrm{IC}_0\) again:

    $$p:I\backslash LX\longrightarrow K\backslash LX,\qquad\mathrm{Oblv}=p^*:\mathcal D_c(K\backslash LX)\to\mathcal D_c(I\backslash LX),\qquad p^!=p^*[2d].$$

    \(\mathcal S_T\) and \(\mathcal S_E\) use \(\mathrm{IC}_0\) on \(K\backslash LX\); \(\mathcal S_H\) uses its image on \(I\backslash LX\).

    Convolution. For \(J\in\{K,I\}\) and a space \(Y\) with \(LG\)-action (\(Y=LX\), or \(Y=\mathrm{Gr}=LG/K\)), objects on \(I\backslash LG/J\) act on \(J\)-equivariant objects on \(Y\) through one correspondence:

    \(I\backslash LG\times Y\) \(I\backslash LG/J\times J\backslash Y\) \(I\backslash LG\overset{J}{\times}Y\) \(I\backslash Y\) \(\mathrm{pr}\) \(\pi_J\) \(\tilde p_J\) \(m_J\)

    \(\pi_J\) is the quotient by \(J\) acting as \((g,y)\mapsto(gj^{-1},jy)\), a \(J\)-torsor; \(m_J[g,y]=gy\) is ind-proper on bounded supports; \(\tilde p_J[g,y]=(gJ,Jy)\).

    $$\mathcal F\ast_J\mathcal M:=m_{J*}\big(\mathcal F\,\widetilde\boxtimes\,\mathcal M\big),\qquad\pi_J^*\big(\mathcal F\,\widetilde\boxtimes\,\mathcal M\big)=\mathrm{pr}^*(\mathcal F\boxtimes\mathcal M),\qquad(f\ast_J\psi)(y)=\sum_{g\in G(F)/J}f(g)\,\psi(g^{-1}y)\ \text{ for functions}.$$

    The twisted product is the descent of \(\mathcal F\boxtimes\mathcal M\) along the \(J\)-torsor \(\pi_J\); it exists because \(\mathcal F\) is right \(J\)-equivariant and \(\mathcal M\) is \(J\)-equivariant (paper §4.1, with \(m_*=m_!\) on bounded supports). The four instances used on this page:

    $$\ast_K:\mathcal D_c(I\backslash LG/K)\times\mathcal D_c(K\backslash LX)\to\mathcal D_c(I\backslash LX),\qquad\ast_I:\mathcal H_{c,I}\times\mathcal D_c(I\backslash LX)\to\mathcal D_c(I\backslash LX),$$ $$\ast_I:\mathcal H_{c,I}\times\mathcal D_c(I\backslash LG/K)\to\mathcal D_c(I\backslash LG/K),\qquad\ast_K:\mathcal D_c(I\backslash LG/K)\times\mathcal H_{c,K}\to\mathcal D_c(I\backslash LG/K).$$

    An unmarked \(\ast\) is \(\ast_K\). Convolution is associative, \((\mathcal F\ast\mathcal V)\ast\mathrm{IC}_0\simeq\mathcal F\ast(\mathcal V\ast\mathrm{IC}_0)\) and \(\mathcal G\ast_I(\mathcal F\ast_K\mathcal M)\simeq(\mathcal G\ast_I\mathcal F)\ast_K\mathcal M\). An object on \(I\backslash LG/K\) cannot be convolved on the right with one on \(I\backslash LG/I\) (the middle group must agree); neither span needs this.

    Comparison square. The forgetful map \(q\) is proper with fibre \(K/I=G/B\). With \(r=q\times\mathrm{id}\), the square is Cartesian, \(m_I=m_K\circ r\), and \(\mathrm{pr}_2\circ r=p\circ\mathrm{pr}_2\) on the second factors:

    \(I\backslash LG\overset{I}{\times}LX\) \(I\backslash LG\overset{K}{\times}LX\) \(I\backslash LX\) \(I\backslash LG/I\) \(I\backslash LG/K\) \(r\) \(m_K\) \(\mathrm{pr}_1\) \(\mathrm{pr}_1\) \(q\)
    $$q_*:\mathcal H_{c,I}\to\mathcal D_c(I\backslash LG/K),\qquad q^*=q^![-2d]:\mathcal D_c(I\backslash LG/K)\to\mathcal H_{c,I},\qquad\mathcal G\ast_Ip^!\mathcal M\simeq(q_*\mathcal G)\ast_K\mathcal M\ \ (\text{step 5}).$$

    For functions, \(q_*h(g)=\sum_{k\in K/I}h(gk)\) and \(q^*f=f\), so \(q_*q^*f=[K:I]\,f\).

  • Thick span

    For a set \(\mathcal S\) of objects of a stable \(\infty\)-category \(\mathcal D\), \(\operatorname{Thick}_{\mathcal D}(\mathcal S)\) is the smallest stable subcategory (HA, Definition 1.1.3.2) containing \(\mathcal S\) and closed under retracts: the smallest full subcategory containing \(0\) and \(\mathcal S\) and closed under shifts, cofibers and direct summands. Two properties are used, for a cofiber sequence and for an exact functor:

    $$\text{(T1)}\quad A\to B\to C\ \text{a cofiber sequence in }\mathcal D,\ \ A,C\in\operatorname{Thick}_{\mathcal D}(\mathcal S)\ \Longrightarrow\ B\in\operatorname{Thick}_{\mathcal D}(\mathcal S),$$ $$\text{(T2)}\quad F:\mathcal D\to\mathcal D'\ \text{exact}\ \Longrightarrow\ F\big(\operatorname{Thick}_{\mathcal D}(\mathcal S)\big)\subseteq\operatorname{Thick}_{\mathcal D'}\big(F(\mathcal S)\big).$$

    In the paper this is span (T) of Definition 3.11, and Lemma 3.12 proves (T1) and (T2). The paper's Satake subcategory is the thick span of the spherical Satake objects:

    $$\mathcal D_c(K\backslash LX)^{\mathrm{Sat}}:=\operatorname{Thick}\{\mathcal V\ast\mathrm{IC}_0:\mathcal V\in\mathcal H_{c,K}\}.$$
  • Small relative tensor product and the functor \(F_c\)

    For small idempotent-complete stable categories, the relative tensor product is the category of compact objects of the relative tensor product of ind-completions (paper, Definition 3.10):

    $$\mathcal D_c(I\backslash\mathrm{Gr})\otimes_{\mathcal H_{c,K}}\mathcal D_c(K\backslash LX):=\big(\mathcal D^{\mathrm{ren}}(I\backslash\mathrm{Gr})\otimes_{\mathcal H_K^{\mathrm{ren}}}\mathcal D^{\mathrm{ren}}(K\backslash LX)\big)^\omega,\qquad\mathcal D^{\mathrm{ren}}=\operatorname{Ind}\mathcal D_c.$$

    The pure tensors are the images of \(\mathcal F\boxtimes\mathcal M\) under the insertion functor \(\operatorname{ins}\). The paper's functor (5.1), fully faithful on the large categories by Campbell–Dhillon and \(I\)-invariants, induces

    $$F_c:\mathcal D_c(I\backslash\mathrm{Gr})\otimes_{\mathcal H_{c,K}}\mathcal D_c(K\backslash LX)\longrightarrow\mathcal D_c(I\backslash LX),\qquad\operatorname{ins}(\mathcal F\boxtimes\mathcal M)\longmapsto\mathcal F\ast\mathcal M,$$

    exact because it is the restriction of a continuous functor to compact objects.

  • The three subcategories and the dictionary $$\mathcal S_E:=\operatorname{EssIm}(F_c),\qquad\mathcal S_T:=\operatorname{Thick}\{\mathcal F\ast\mathrm{IC}_0:\mathcal F\in\mathcal D_c(I\backslash\mathrm{Gr})\},\qquad\mathcal S_H:=\operatorname{Thick}\{\mathcal G\ast\mathrm{IC}_0:\mathcal G\in\mathcal H_{c,I}\}.$$
    Table 1. · Where each description appears.
    HereSourceWording there
    \(\mathcal S_E\)Note, left side; original paper, §4.6“The essential image of \(F'_c\) is \(\mathcal D_c(I\backslash LX)^{\mathrm{Sat}}\), by definition”, with the spherical factor restricted to \(\mathcal D_c(K\backslash LX)^{\mathrm{Sat}}\); the note drops the restriction and assumes H1 instead.
    \(\mathcal S_T\)mainv2, Definition 4.3; original paper, Definition 3.28(2)\(\mathcal D_c(I\backslash LX)^{\mathrm{Sat}}:=\operatorname{Thick}\{\mathcal F\ast\mathrm{IC}_0:\mathcal F\in\mathcal D_c(I\backslash\mathrm{Gr})\}\); the original wrote “generated by”, read as the thick span by its Remark 3.29.
    \(\mathcal S_H\)Note, right side; Devalapurkar, Conjecture 3.4.14\(\langle\mathcal H_{c,I}\ast\mathrm{IC}_0\rangle\), with \(\mathcal H_{c,I}=\mathcal D_c(I\backslash LG/I)\) acting on \(\mathcal D_c(I\backslash LX)\); “the full subcategory of \(\mathrm{Shv}^c_I(X(\mathbb C(\!(t)\!)))\) generated by \(\mathrm{IC}_0\) under the action of \(\mathrm{Shv}^{c,\mathrm{Sat}}_{I\times I}(G(\mathbb C(\!(t)\!)))\) via convolution”.
  • Why \(\mathcal S_H\): the category of Devalapurkar's Conjecture 3.4.14

    Conjecture 3.4.14 (arXiv:2402.03995v2) predicts an equivalence \(\mathrm{Shv}^{c,\mathrm{Sat}}_I(X(\mathbb C(\!(t)\!)));\mathbb Q)\simeq\mathrm{Perf}(\mathrm{sh}^{1/2}(\check M\times_{\check{\mathfrak g}}\tilde{\check{\mathfrak g}})/\check G)\) sending \(\mathrm{IC}_0=i_!\mathbb Q\) to the structure sheaf, where the left side is the full subcategory of \(\mathrm{Shv}^c_I(X(\mathbb C(\!(t)\!)))\) generated by \(\mathrm{IC}_0\) under the action of the affine Hecke category \(\mathrm{Shv}^{c,\mathrm{Sat}}_{I\times I}(G(\mathbb C(\!(t)\!)))\) via convolution. In the notation here that is \(\mathcal S_H\). The paper's Definition 4.3 is \(\mathcal S_T\), the natural output of its functor \(F_c^{\mathrm{Sat}}\) (Theorem 5.1), and the paper presents its main theorem as a variant of the conjecture. Claim (i) identifies the two spans.

    \(\mathcal S_H\) is also the thick \(\mathcal H_{c,I}\)-submodule generated by \(\mathrm{IC}_0\): \(\mathcal G\ast_I(\mathcal G'\ast_I\mathrm{IC}_0)\simeq(\mathcal G\ast_I\mathcal G')\ast_I\mathrm{IC}_0\) and (T2). So (i) shows that Definition 4.3 is stable under the affine Hecke action for which the main theorem is equivariant.

    Devalapurkar's Hecke factor is the Satake part of the affine Hecke category of constructible sheaves; the paper's \(\mathcal H_{c,I}\) is all coherent \(\mathcal D\)-modules on \(I\backslash LG/I\). The Proposition is proved for \(\mathcal H_{c,I}\); the span by the Satake part lies in \(\mathcal S_H\), and equality of the two is not checked here.

Hypotheses

  • H1 · Spherical Satake generation. The spherical Satake objects thickly generate the coherent spherical category: $$\mathcal D_c(K\backslash LX)^{\mathrm{Sat}}=\operatorname{Thick}\{\mathcal V\ast\mathrm{IC}_0:\mathcal V\in\mathcal H_{c,K}\}=\mathcal D_c(K\backslash LX).$$
    Discussion · where H1 is proved

    The paper proves it in §4.4 (Proposition 4.30) from its five hypotheses on the orbit stratification (numbered in §4.2 of the paper), and verifies those for the group case and for symmetric spaces satisfying two root-datum conditions. The Spherical generation workspace records that proof. By derived Satake, the span may be taken over the \(V_\lambda\) alone (paper, Remark 4.4(3)).

  • H2 · The Iwahori convolution functor.
    1. \(F_c\) is exact and fully faithful.
    2. Its source is the thick span of the pure tensors: $$\mathcal D_c(I\backslash\mathrm{Gr})\otimes_{\mathcal H_{c,K}}\mathcal D_c(K\backslash LX)=\operatorname{Thick}\{\operatorname{ins}(\mathcal F\boxtimes\mathcal M):\mathcal F\in\mathcal D_c(I\backslash\mathrm{Gr}),\ \mathcal M\in\mathcal D_c(K\backslash LX)\}.$$
    Discussion · where H2 comes from

    (a) is the paper's §5: the large functor (5.1) is fully faithful by Campbell–Dhillon, Theorem 3.1.5 and Lemma 3.30, and §5.3–5.5 show that the embedding \(\iota\) of the small tensor product is fully faithful (Lemma 5.4, the Künneth formula and the monad computation). (b) combines Gaitsgory–Rozenblyum, Corollary 8.7.4 (compact generation by pure tensors, with rigidity of \(\mathcal H_K^{\mathrm{ren}}\) as in the paper's Lemma 5.3) and Neeman, Lemma 2.2 (compact objects form the thick span of the generators).

Proposition

Setting. \(k\), \(G\supseteq B\), \(K=L^+G\), \(I\), \(X\) as above; \(\mathcal S_E\), \(\mathcal S_T\), \(\mathcal S_H\) the three full subcategories of \(\mathcal D_c(I\backslash LX)\) defined above.

Claim.

  1. Without further hypotheses, \(\mathcal S_T=\mathcal S_H\): $$\operatorname{Thick}\{\mathcal F\ast\mathrm{IC}_0:\mathcal F\in\mathcal D_c(I\backslash\mathrm{Gr})\}=\operatorname{Thick}\{\mathcal G\ast\mathrm{IC}_0:\mathcal G\in\mathcal H_{c,I}\}.$$
  2. Under H2, the original paper's definition is Definition 4.3: $$\operatorname{EssIm}\big(F_c|_{\mathcal D_c(I\backslash\mathrm{Gr})\otimes_{\mathcal H_{c,K}}\mathcal D_c(K\backslash LX)^{\mathrm{Sat}}}\big)=\mathcal S_T.$$
  3. Under H2, \(\mathcal S_T\subseteq\mathcal S_E\), with equality if and only if H1 holds: $$\mathcal S_E=\mathcal S_T\iff\mathcal D_c(K\backslash LX)^{\mathrm{Sat}}=\mathcal D_c(K\backslash LX).$$
  4. Hence under H1 and H2 all three agree, which is the note's goal: $$\boxed{\mathcal S_E=\mathcal S_T=\mathcal S_H.}$$
Sources

References

References checked 11 October 2026 against the downloaded versions named in panel D. H1 and H2 are inputs taken from the paper; the graph records what each step uses.

Proof

Step 1 names the three subcategories. Steps 2–4 prove (ii) and (iii): H2 makes the image a thick span of convolutions, H1 collapses the generators, and averaging over \(G/B\) shows that H1 is also necessary. Steps 5–7 prove (i): the easy inclusion, the key lemma, and the retract. Step 8 concludes.

  1. Each description is a full subcategory of \(\mathcal D_c(I\backslash LX)\), and (iv) is the conjunction of two equalities with different inputs:

    $$\mathcal S_E\overset{\text{H1, H2}}{=}\mathcal S_T\overset{\text{key lemma}}{=}\mathcal S_H.$$
    Proof
    • \(\mathcal S_T\) and \(\mathcal S_H\) are thick spans by definition. \(\mathcal S_E\) is the essential image of an exact functor; H2(a) will make it thick (step 2).
    • The note proves only the key lemma; the rest of this page is the reconstruction of the chain from it.
  2. Under H2,

    $$\mathcal S_E=\operatorname{Thick}\{\mathcal F\ast\mathcal M:\mathcal F\in\mathcal D_c(I\backslash\mathrm{Gr}),\ \mathcal M\in\mathcal D_c(K\backslash LX)\},$$

    and the same with \(\mathcal M\) restricted to \(\mathcal D_c(K\backslash LX)^{\mathrm{Sat}}\) for the restricted functor.

    Proof
    • The essential image is thick. \(F_c\) is exact, so \(\mathcal S_E\) is closed under shifts and cofibers. Let \(Y\) be a retract of \(F_c(x)\): \(Y\xrightarrow{i}F_c(x)\xrightarrow{r}Y\), \(ri=\mathrm{id}\). The idempotent \(e=ir\) of \(F_c(x)\) is \(F_c(e')\) for an endomorphism \(e'\) of \(x\) by full faithfulness, and \(e'\) is idempotent in the homotopy category. The source is idempotent complete, so by HA, Lemma 1.2.4.6 \(e'\) splits, \(x\simeq x'\oplus x''\) with \(e'\) the projector to \(x'\); then \(Y\simeq F_c(x')\).
    • Generators. By H2(b) the source is \(\operatorname{Thick}\{\operatorname{ins}(\mathcal F\boxtimes\mathcal M)\}\); (T2) gives \(\mathcal S_E\subseteq\operatorname{Thick}\{\mathcal F\ast\mathcal M\}\), and the reverse inclusion holds because \(\mathcal S_E\) is thick and contains each \(\mathcal F\ast\mathcal M\).
    • For the restricted functor, replace \(\mathcal D_c(K\backslash LX)\) by its thick subcategory \(\mathcal D_c(K\backslash LX)^{\mathrm{Sat}}\); H2(b) restricts because the restricted tensor product is the thick span of the pure tensors with \(\mathcal M\) in the subcategory.
  3. If every \(\mathcal M\) in the second factor lies in \(\operatorname{Thick}\{\mathcal V\ast\mathrm{IC}_0:\mathcal V\in\mathcal H_{c,K}\}\), then

    $$\operatorname{Thick}\{\mathcal F\ast\mathcal M\}=\operatorname{Thick}\{\mathcal F\ast\mathrm{IC}_0:\mathcal F\in\mathcal D_c(I\backslash\mathrm{Gr})\}=\mathcal S_T.$$

    For the restricted functor this is automatic, proving (ii); for \(F_c\) itself it is H1, proving \(\mathcal S_E=\mathcal S_T\) under H1 and H2.

    Proof
    • \(\supseteq\): \(\mathcal F\ast\mathrm{IC}_0\) is one of the generators, since \(\mathrm{IC}_0\in\mathcal D_c(K\backslash LX)^{\mathrm{Sat}}\).
    • \(\subseteq\): fix \(\mathcal F\). The functor \(\mathcal F\ast(-):\mathcal D_c(K\backslash LX)\to\mathcal D_c(I\backslash LX)\) is exact, so by (T2) $$\mathcal F\ast\mathcal M\in\operatorname{Thick}\{\mathcal F\ast(\mathcal V\ast\mathrm{IC}_0)\}=\operatorname{Thick}\{(\mathcal F\ast\mathcal V)\ast\mathrm{IC}_0\}\subseteq\mathcal S_T,$$ using associativity and \(\mathcal F\ast\mathcal V\in\mathcal D_c(I\backslash\mathrm{Gr})\).
  4. Under H2 and S1, \(\mathcal S_T\subseteq\mathcal S_E\), and \(\mathcal S_E\subseteq\mathcal S_T\) implies H1:

    $$p^*\mathcal M\in\mathcal S_E\ \ (\forall\mathcal M),\qquad p_*(\mathcal S_T)\subseteq\mathcal D_c(K\backslash LX)^{\mathrm{Sat}},\qquad\mathcal M\ \text{a retract of}\ p_*p^*\mathcal M .$$

    With step 3 this is (iii). The original paper's restricted definition sidesteps H1; the note's unrestricted statement needs it.

    Proof
    • \(\mathcal S_T\subseteq\mathcal S_E\) is the inclusion \(\supseteq\) of step 3 applied to \(F_c\).
    • Spherical objects lie in \(\mathcal S_E\). Let \(\delta\in\mathcal D_c(I\backslash LG/K)\) be the constant \(\mathcal D\)-module on the closed substack \(I\backslash K/K=\mathbb BI\) (the base point of \(\mathrm{Gr}\)), extended by zero. \(\delta\,\widetilde\boxtimes\,\mathcal M\) is supported on \(I\backslash K\times^KLX\simeq I\backslash LX\), where \(m_K\) is the identity and \(\mathrm{pr}_2\) is \(p\), so \(\delta\ast_K\mathcal M\simeq p^*\mathcal M\) up to a shift. Thick spans contain shifts, so \(p^*\mathcal M\in\mathcal S_E\) for every \(\mathcal M\in\mathcal D_c(K\backslash LX)\). (For functions: \(\mathbf 1_K\ast_K\varphi=\varphi\).)
    • Averaging over \(G/B\). \(p\) is proper with fibre \(K/I=G/B\), so \(p_*=p_!:\mathcal D_c(I\backslash LX)\to\mathcal D_c(K\backslash LX)\) is exact and preserves coherence. With \(p':I\backslash LG/K\to K\backslash LG/K\), base change and the projection formula exactly as in step 5 give \(p_*(\mathcal F\ast_K\mathrm{IC}_0)\simeq(p'_*\mathcal F)\ast_K\mathrm{IC}_0\) with \(p'_*\mathcal F\in\mathcal H_{c,K}\). By (T2), \(p_*(\mathcal S_T)\subseteq\operatorname{Thick}\{\mathcal V\ast\mathrm{IC}_0\}=\mathcal D_c(K\backslash LX)^{\mathrm{Sat}}\).
    • Key lemma for \(p\). \(I=K\times_GB\) gives \(I\backslash LX=K\backslash LX\times_{\mathbb BG}\mathbb BB\), so \(p\) is a base change of \(\pi:\mathbb BB\to\mathbb BG\) and the key lemma applies: \(\mathcal M\) is a direct summand of \(p_*p^*\mathcal M\). If \(\mathcal S_E\subseteq\mathcal S_T\), then \(p^*\mathcal M\in\mathcal S_T\), hence \(p_*p^*\mathcal M\in\mathcal D_c(K\backslash LX)^{\mathrm{Sat}}\), and the thick span contains its direct summand \(\mathcal M\). That is H1.
    • The same argument for functions: averaging over \(K/I\) shows \(C_c(I\backslash G(F)/K)\ast_K\mathbf 1_{X(\mathcal O)}\cap C_c(X(F))^K=\mathcal H_K\ast_K\mathbf 1_{X(\mathcal O)}\) (functions page).
  5. With \(q:I\backslash LG/I\to I\backslash LG/K\) the forgetful map, for \(\mathcal G\in\mathcal H_{c,I}\) and \(\mathcal M\in\mathcal D_c(K\backslash LX)\),

    $$\mathcal G\ast_Ip^!\mathcal M\simeq(q_*\mathcal G)\ast_K\mathcal M,\qquad q_*\mathcal G\in\mathcal D_c(I\backslash\mathrm{Gr}),\qquad\text{hence}\quad\mathcal S_H\subseteq\mathcal S_T.$$
    Proof
    • This is the comparison square. Since \(I\backslash LX\simeq\mathbb BI\times_{\mathbb BK}(K\backslash LX)\), the two convolution spaces are fibre products over \(\mathbb BK\): $$I\backslash LG\times^ILX\simeq(I\backslash LG/I)\times_{\mathbb BK}(K\backslash LX)\xrightarrow{\ r=q\times\mathrm{id}\ }(I\backslash LG/K)\times_{\mathbb BK}(K\backslash LX)\simeq I\backslash LG\times^KLX,$$ and \(m_I=m_K\circ r\). On the left, \(\mathcal G\,\widetilde\boxtimes\,p^!\mathcal M=\mathrm{pr}_1^!\mathcal G\otimes^!\mathrm{pr}_2^!\mathcal M\) with \(\mathrm{pr}_2\) to \(K\backslash LX\).
    • \(r\) is the base change of the proper map \(q\) along \(\mathrm{pr}_1\). The projection formula for \(r_*=r_!\) and base change give $$r_*(\mathrm{pr}_1^!\mathcal G\otimes^!\mathrm{pr}_2^!\mathcal M)\simeq r_*\mathrm{pr}_1^!\mathcal G\otimes^!\mathrm{pr}_2^!\mathcal M\simeq\mathrm{pr}_1^!q_*\mathcal G\otimes^!\mathrm{pr}_2^!\mathcal M=(q_*\mathcal G)\,\widetilde\boxtimes\,\mathcal M.$$ Applying \(m_{K*}\) gives the displayed isomorphism.
    • \(q\) is proper with fibre \(G/B\), so \(q_*=q_!\) preserves coherence. Since \(p^!\mathrm{IC}_0=p^*\mathrm{IC}_0[2d]\) and thick spans contain shifts, every \(\mathcal G\ast_I\mathrm{IC}_0\) lies in \(\mathcal S_T\).
  6. For \(q:I\backslash\mathrm{Fl}\to I\backslash\mathrm{Gr}\) and every \(\mathcal F\in\mathcal D_c(I\backslash\mathrm{Gr})\), the unit of \(q^*\dashv q_*\) has a natural retraction:

    $$\mathcal F\xrightarrow{\ \eta_{\mathcal F}\ }q_*q^*\mathcal F\xrightarrow{\ \beta_{\mathcal F}\ }\mathcal F,\qquad\beta_{\mathcal F}\circ\eta_{\mathcal F}=\mathrm{id}_{\mathcal F},\qquad q^*\mathcal F\in\mathcal H_{c,I}.$$
    Proof
    • \(q\) is the base change of \(\pi:\mathbb BB\to\mathbb BG\) along the classifying map \(\gamma:I\backslash\mathrm{Gr}\to\mathbb BG\); base change and the projection formula give \(q_*q^*\mathcal F\simeq\gamma^!\mathcal A\otimes^!\mathcal F\) with \(\mathcal A=\pi_*\pi^*\omega_{\mathbb BG}\), and \(\eta_{\mathcal F}=\gamma^!u\otimes^!\mathrm{id}\) for the unit \(u:\omega_{\mathbb BG}\to\mathcal A\).
    • \(\mathcal A\) is a commutative algebra with unit \(u\), dualizable, and the trace of multiplication satisfies \(\operatorname{tr}_{\mathcal A}\circ u=\chi(G/B)=|W|\), invertible in \(k\). Set \(\beta_{\mathcal F}=\tfrac1{|W|}\gamma^!\operatorname{tr}_{\mathcal A}\otimes^!\mathrm{id}\).
    • \(q^*\mathcal F\) is coherent because \(q\) is smooth. The argument uses only the square, so it holds for every base change of \(\pi\); step 4 uses it for \(p:I\backslash LX\to K\backslash LX\). Full proof and graph →
  7. Convolving the retraction with \(\mathrm{IC}_0\) and rewriting by step 5,

    $$\mathcal F\ast\mathrm{IC}_0\longrightarrow(q^*\mathcal F)\ast_I\mathrm{IC}_0[-2d]\longrightarrow\mathcal F\ast\mathrm{IC}_0\quad\text{composes to the identity, so}\quad\mathcal F\ast\mathrm{IC}_0\in\mathcal S_H,\quad\mathcal S_T\subseteq\mathcal S_H.$$
    Proof
    • \((-)\ast_K\mathrm{IC}_0\) is a functor, so \((\beta_{\mathcal F}\ast\mathrm{id})\circ(\eta_{\mathcal F}\ast\mathrm{id})=\mathrm{id}\); the middle term is \((q_*q^*\mathcal F)\ast_K\mathrm{IC}_0\simeq(q^*\mathcal F)\ast_Ip^!\mathrm{IC}_0[-2d]\) by step 5 with \(\mathcal G=q^*\mathcal F\).
    • \(\mathcal S_H\) contains \((q^*\mathcal F)\ast_I\mathrm{IC}_0\), hence its shifts and direct summands (Stacks 05RB). With step 5, \(\mathcal S_T=\mathcal S_H\), which is (i).
  8. (i) is step 7; (ii) is step 3 for the restricted functor; (iii) is step 4; (iv) combines step 3 under H1 with (i):

    $$\mathcal S_E=\mathcal S_T=\mathcal S_H.$$
    Proof
    • In the note's words: assuming \(\mathcal D_c(K\backslash LX)^{\mathrm{Sat}}\simeq\mathcal D_c(K\backslash LX)\), \(\operatorname{EssIm}(\mathcal D_c(I\backslash\mathrm{Gr})\otimes_{\mathcal H_{c,K}}\mathcal D_c(K\backslash LX))\simeq\langle\mathcal H_{c,I}\ast\mathrm{IC}_0\rangle\).