Special Cycles on Shtukas: Geometric Insights

• Special cycles on shtukas are algebraic cycles defined by imposing additional geometric structures on moduli stacks, resulting in closed cycle classes.
• The construction leverages subgroup embeddings, finite schematic maps, and representation-theoretic techniques to ensure proper pushforwards and finite morphisms.
• These cycles underpin arithmetic identities, linking intersection pairings with derivatives of L-functions and advancing insights in both global and p-adic settings.

Special cycles on Shtukas are algebraic cycles on moduli stacks of global or local shtukas obtained by imposing additional geometric structure on the underlying shtuka: a reduction of structure group, a compatible section, a fixed Hermitian datum, a period condition, or a prescribed Newton-type condition. In the global function-field setting, the basic structural statement is that if $\mathcal H\hookrightarrow \mathcal G$ is a closed embedding of smooth affine group schemes over a curve with $\mathcal H$ Bruhat–Tits, then the induced map $$\Theta^r:\mathrm{Sht}^r_{\mathcal H}\to \mathrm{Sht}^r_{\mathcal G}$$ is schematic, finite, and unramified, so its image is a closed substack; this is the precise sense in which special cycles on shtukas are closed (Yun, 2022).

1. Moduli-theoretic setting

Fix a finite field $k=\mathbb F_q$, a smooth, projective, geometrically connected curve $X/k$, and its function field $F=k(X)$. For a smooth affine group scheme $\mathcal G$ over $X$, the global Hecke stack $\operatorname{Hk}^r_{\mathcal G}$ classifies chains of modifications of $\mathcal G$-torsors at $\mathcal H$0 legs. The stack of $\mathcal H$1-shtukas with $\mathcal H$2 legs is defined by the Cartesian square

$\mathcal H$3

An $\mathcal H$4-point is a tuple $\mathcal H$5 consisting of legs $\mathcal H$6, a chain of $\mathcal H$7-torsors $\mathcal H$8, modifications $\mathcal H$9, and a Frobenius descent isomorphism $$\Theta^r:\mathrm{Sht}^r_{\mathcal H}\to \mathrm{Sht}^r_{\mathcal G}$$0. After imposing bounds $$\Theta^r:\mathrm{Sht}^r_{\mathcal H}\to \mathrm{Sht}^r_{\mathcal G}$$1 on relative positions, one obtains algebraic stacks $$\Theta^r:\mathrm{Sht}^r_{\mathcal H}\to \mathrm{Sht}^r_{\mathcal G}$$2 locally of finite type, and

$$\Theta^r:\mathrm{Sht}^r_{\mathcal H}\to \mathrm{Sht}^r_{\mathcal G}$$3

with transition maps given by closed embeddings (Yun, 2022).

This moduli-theoretic framework already contains the basic geometry of special cycles. A reduction of a $$\Theta^r:\mathrm{Sht}^r_{\mathcal H}\to \mathrm{Sht}^r_{\mathcal G}$$4-torsor to a subgroup $$\Theta^r:\mathrm{Sht}^r_{\mathcal H}\to \mathrm{Sht}^r_{\mathcal G}$$5 is a section of the associated quotient $$\Theta^r:\mathrm{Sht}^r_{\mathcal H}\to \mathrm{Sht}^r_{\mathcal G}$$6, so a $$\Theta^r:\mathrm{Sht}^r_{\mathcal H}\to \mathrm{Sht}^r_{\mathcal G}$$7-shtuka is naturally a $$\Theta^r:\mathrm{Sht}^r_{\mathcal H}\to \mathrm{Sht}^r_{\mathcal G}$$8-shtuka equipped with additional structure. The same pattern recurs in later constructions: unitary cycles impose compatible maps $$\Theta^r:\mathrm{Sht}^r_{\mathcal H}\to \mathrm{Sht}^r_{\mathcal G}$$9 on unitary shtukas, Rankin–Selberg cycles impose a diagonal embedding of shtuka stacks, and $k=\mathbb F_q$0-adic constructions impose Newton or local-model conditions on shtuka fibers. The subject therefore has a common moduli-theoretic core even though the concrete cycle classes vary across global, local, and $k=\mathbb F_q$1-adic settings.

2. Closed images from subgroup embeddings

Let $k=\mathbb F_q$2 be a closed embedding of smooth affine group schemes over $k=\mathbb F_q$3, with $k=\mathbb F_q$4 Bruhat–Tits. Then the induced map

$k=\mathbb F_q$5

is schematic, finite, and unramified for every $k=\mathbb F_q$6. On bounded pieces, for any bound $k=\mathbb F_q$7 on $k=\mathbb F_q$8-modifications, the restriction

$k=\mathbb F_q$9

has the same properties (Yun, 2022). Because a finite representable morphism is proper and universally closed, the image

$X/k$0

is a closed substack. This is the foundational closedness statement for special cycles on shtukas.

The Chow-theoretic consequence is immediate. If $X/k$1 has a fundamental cycle class, then the proper representable map

$X/k$2

defines a pushforward

$X/k$3

which is, by definition, a special cycle.

The proof is organized through a representation-theoretic factorization. One chooses $X/k$4 and sets $X/k$5. The quotient $X/k$6 is represented inside $X/k$7 by a closed embedding on a dense open set, and this yields a closed embedding

$X/k$8

into the stack of $X/k$9-bundles equipped with a fiberwise nonzero section. Passing to Hecke stacks and then to shtukas gives a closed embedding

$F=k(X)$0

The original map factors as

$F=k(X)$1

where $F=k(X)$2 forgets the sections. The first arrow is a closed embedding, and the second is schematic, finite, and unramified (Yun, 2022).

The Bruhat–Tits hypothesis enters at the decisive extension step. The proof uses Anschütz’s result that $F=k(X)$3-torsors on punctured formal disks are trivial, the ind-properness of affine flag varieties $F=k(X)$4 due to Richarz, and Beauville–Laszlo gluing to control reductions across bad fibers (Yun, 2022). The same theorem extends to pseudo-homomorphisms $F=k(X)$5, i.e. right $F=k(X)$6-torsors on $F=k(X)$7 with commuting left $F=k(X)$8-action; if the induced map to the corresponding inner form is a closed embedding, then the induced shtuka map is again schematic, finite, and unramified (Yun, 2022).

3. Principal geometric families

The classical prototype is the Heegner–Drinfeld cycle. Let $F=k(X)$9 be a finite étale degree-two cover, $\mathcal G$0, and

$\mathcal G$1

For an even integer $\mathcal G$2 and a sign vector $\mathcal G$3 with $\mathcal G$4, the torus-shtuka stack $\mathcal G$5 is a smooth proper Deligne–Mumford stack of pure dimension $\mathcal G$6, and the induced map

$\mathcal G$7

to the base change of the $\mathcal G$8-shtuka stack defines the Heegner–Drinfeld cycle

$\mathcal G$9

Since $X$0, this is a middle-dimensional cycle (Yun et al., 2015). The general closedness theorem places this example in a broader class that also includes Heegner–Drinfeld cycles for $X$1, diagonal cycles $X$2 of Gan–Gross–Prasad type, and Kudla–Rapoport-type cycles defined by imposing extra maps into associated bundles (Yun, 2022).

For unitary groups, let $X$3 be a finite étale double cover and $X$4 the associated unitary group. The stack $X$5 of unitary shtukas with $X$6 legs is smooth, separated, and equidimensional of relative dimension $X$7. Fix a rank $X$8 vector bundle $X$9 on $\operatorname{Hk}^r_{\mathcal G}$0. The stack $\operatorname{Hk}^r_{\mathcal G}$1 over $\operatorname{Hk}^r_{\mathcal G}$2 parametrizes unitary shtukas together with compatible maps

$\operatorname{Hk}^r_{\mathcal G}$3

through the modification chain and Frobenius descent. The forgetful map $\operatorname{Hk}^r_{\mathcal G}$4 is finite. After fixing a Hermitian map $\operatorname{Hk}^r_{\mathcal G}$5, one obtains fibers $\operatorname{Hk}^r_{\mathcal G}$6. When $\operatorname{Hk}^r_{\mathcal G}$7 and $\operatorname{Hk}^r_{\mathcal G}$8 is injective, these cycles have virtual dimension $\operatorname{Hk}^r_{\mathcal G}$9, and the authors construct a virtual fundamental $\mathcal G$0-cycle

$\mathcal G$1

with well-defined degree (Feng et al., 2021).

For $\mathcal G$2, special cycles of Rankin–Selberg type are produced by the diagonal map of shtukas

$\mathcal G$3

defined by direct sum with the trivial line. This map is finite schematic, and its pushforward of the fundamental class defines a Borel–Moore special cycle (Wang, 5 Sep 2025). A complementary linearized model is furnished by toy shtukas: in that setting, toy horospherical divisors are the loci of toy shtukas contained in a fixed hyperplane or containing a fixed line, and the pullback of such divisors to the moduli of Drinfeld shtukas is described by an averaging operator closely related to Fourier and Radon transforms (Ding, 2018).

4. Arithmetic identities and higher derivatives

The first large family of arithmetic formulas identifies self-intersections of Heegner–Drinfeld cycles with central derivatives of automorphic $\mathcal G$4-functions. For an everywhere unramified cuspidal automorphic representation $\mathcal G$5 of $\mathcal G$6 and even $\mathcal G$7, the $\mathcal G$8-isotypic component $\mathcal G$9 of the Heegner–Drinfeld cycle satisfies

$\mathcal H$00

where $\mathcal H$01 is the normalized quadratic base-change $\mathcal H$02-function. The same identity has an $\mathcal H$03-adic cohomological form involving the cycle class of $\mathcal H$04 in middle cohomology (Yun et al., 2015). This is a function-field analogue of Waldspurger and Gross–Zagier, but with the additional feature that the shtuka parameter $\mathcal H$05 gives access to all even central derivatives.

For unitary groups, the higher Siegel–Weil formula expresses degrees of virtual zero-cycles on unitary shtuka stacks as derivatives of normalized nonsingular Fourier coefficients of a Siegel–Eisenstein series. If $\mathcal H$06 has rank $\mathcal H$07, $\mathcal H$08 is injective, and $\mathcal H$09, then

$\mathcal H$10

The cycles $\mathcal H$11 are therefore the geometric side of a higher-derivative Siegel–Weil identity on unitary shtukas (Feng et al., 2021).

With Iwahori level structure, intersections of two Heegner–Drinfeld cycles attached to distinct nonsplit tori satisfy a Gross–Kohnen–Zagier type identity. For an automorphic representation $\mathcal H$12 of $\mathcal H$13 with Iwahori level at $\mathcal H$14, the intersection pairing of the two $\mathcal H$15-isotypic cycle classes equals

$\mathcal H$16

and the vanishing of this intersection is equivalent to the vanishing of

$\mathcal H$17

(Li, 2019).

A different but related formalism appears in the arithmetic volume theory for split semisimple $\mathcal H$18. There the “special cycle” is encoded by the Schubert locus $\mathcal H$19 together with a power of the determinant line bundle $\mathcal H$20. For local volume data $\mathcal H$21, the arithmetic volume of the shtuka stack is

$\mathcal H$22

and in the equal-coweight case this becomes an $\mathcal H$23-th derivative of a single-variable product of zeta functions (Wang et al., 4 Apr 2026). The eigenweights $\mathcal H$24 and the constants $\mathcal H$25 appearing in the differential operators $\mathcal H$26 are described uniformly by the Langlands dual group through the principal $\mathcal H$27, the Kostant slice, and the representation $\mathcal H$28 (Wang et al., 4 Apr 2026).

5. Categorical, nearby-cycle, and trace-theoretic formalisms

A categorical reformulation of special cycles on shtukas is developed through Hecke eigensheaves and categorical trace. For a Hecke eigensheaf $\mathcal H$29, one considers the endofunctor $\mathcal H$30 on $\mathcal H$31. The categorical trace theorem identifies

$\mathcal H$32

so shtuka cohomology itself becomes a categorical trace. In this framework, the fake special cycle classes $\mathcal H$33 attached to period functors are shown to coincide with the $\mathcal H$34-isotypic projections of genuine geometric special cycles; for Rankin–Selberg cycles, this comparison yields a higher-derivative formula for the self-intersection norm of the isotypic component (Wang, 5 Sep 2025).

Special cycles in families require control under degeneration. For a generically reductive group $\mathcal H$35 over a smooth projective curve $\mathcal H$36, nearby cycles commute with proper direct image along the leg map $\mathcal H$37: $\mathcal H$38 This holds for arbitrary finite sets of legs $\mathcal H$39 in the stack-theoretic setting (Eteve et al., 2024). The result does not itself define special cycles, but it provides the formal mechanism needed to compare cycle classes, intersection pairings, and cohomological correspondences between generic and special fibers.

At parahoric places, nearby cycles become a local geometric substitute for special cycles on the bad fiber. For parahoric shtukas, one computes traces of Frobenius composed with Hecke operators on the cohomology of nearby cycles and uses the resulting Langlands–Kottwitz formulas to prove the base change fundamental lemma for parahoric Hecke algebras for $\mathcal H$40 over local function fields (Feng, 2017). This shows that nearby-cycle complexes on special fibers are not only degeneration objects but also precise carriers of local test functions and orbital-integral identities.

6. Local, $\mathcal H$41-adic, and compactified geometries

The local theory of special cycles on shtukas is governed by Rapoport–Zink spaces for local $\mathcal H$42-shtukas. Given a local datum $\mathcal H$43, the bounded Rapoport–Zink space $\mathcal H$44 is a formal scheme locally formally of finite type and separated over the reflex ring, with reduced special fiber equal to the affine Deligne–Lusztig variety

$\mathcal H$45

There is a local-model roof

$\mathcal H$46

with $\mathcal H$47 an $\mathcal H$48-torsor and $\mathcal H$49 formally smooth. This transfers flatness and Serre conditions from $\mathcal H$50 to $\mathcal H$51, and it compares formal nearby cycles on $\mathcal H$52 with nearby cycles on the local model $\mathcal H$53 (Rad, 2020). These properties are exactly the local geometric input needed for defining and analyzing intersection multiplicities of special cycles.

In the $\mathcal H$54-adic shtuka setting over the Fargues–Fontaine curve, the special Newton polygon map

$\mathcal H$55

is representable in locally spatial diamonds and fdcs, and for each $\mathcal H$56 the fiber

$\mathcal H$57

is a locally spatial kimberlite with reduced special fiber

$\mathcal H$58

the affine Deligne–Lusztig variety (Gleason, 6 Jan 2026). The henselianity theorem

$\mathcal H$59

for spatial kimberlites under properness and finite-transcendence hypotheses implies that the cohomology of the full fiber is controlled by the reduced ADLV locus (Gleason, 6 Jan 2026). The paper does not explicitly define special cycles on $\mathcal H$60-adic shtuka stacks, but it gives the geometric and cohomological infrastructure in which cycles supported on Newton strata and on ADLVs can be defined and compared.

A further extension appears on toroidal compactifications of abelian-type Shimura varieties. Log diamonds are introduced as $\mathcal H$61-sheaves attached to fs log schemes or log adic spaces over $\mathcal H$62, and log $\mathcal H$63-shtukas are defined as morphisms from these log diamonds to the shtuka stack. For quasi-parahoric level at $\mathcal H$64, the canonical $\mathcal H$65-adic shtuka on the open Shimura variety extends uniquely as a log $\mathcal H$66-shtuka to the toroidal compactification, and restriction to the special fiber yields a Witt-vector shtuka (Mao et al., 28 May 2026). In this framework Newton, central, KR, and EKOR strata, their connected components, and their closures are well-positioned on the special fiber, and the same is true for their partial toroidal and minimal compactifications (Mao et al., 28 May 2026). This gives a boundary-compatible ambient geometry for special-cycle theories on compactified Shimura varieties.

Taken together, these developments show that special cycles on shtukas form a coherent geometric theme rather than a single construction. Closed images of subgroup shtuka stacks, Heegner–Drinfeld cycles, unitary and Rankin–Selberg cycles, determinant–Schubert cycles, Newton-stratum cycles, and boundary extensions on compactifications all fit into a common pattern: additional structure on a shtuka produces a cycle on a moduli stack, and the resulting cycle is controlled by finiteness, local-model, nearby-cycle, and categorical-trace formalisms that connect its geometry to automorphic periods, derivatives of $\mathcal H$67-functions, and dual-group representation theory.