Good Moduli Spaces in Algebraic Stacks

• Good moduli spaces are morphisms from algebraic or Artin stacks to algebraic spaces that generalize GIT quotients by requiring quasi-compactness, quasi-separation, and exact pushforward.
• They replace naive orbit-space constructions with categorical and cohomological methods, encoding S-equivalence classes even when stacks are non-separated.
• Their existence and local behavior rely on affine étale neighborhoods and stabilizer conditions, with applications from moduli of curves to derived moduli problems.

Good moduli spaces are morphisms from algebraic or Artin stacks to algebraic spaces that generalize GIT good quotients and, in suitable tame settings, coarse moduli spaces. In the formulation used throughout recent work, a morphism $\pi\colon \mathcal{X}\to X$ is a good moduli space if $\pi$ is quasi-compact and quasi-separated, $\pi_*$ is exact on quasi-coherent sheaves, and $\pi_*\mathcal{O}_{\mathcal{X}}=\mathcal{O}_X$; equivalently, it satisfies the “Stein” and “cohomologically affine” conditions. Such morphisms are universal for maps to algebraic spaces, are surjective and universally closed, and are unique up to unique isomorphism. Although they are generally not separated and need not be proper, a substantial body of work shows that they frequently behave like proper morphisms in formal, cohomological, and birational contexts (Rydh, 2020, Geraschenko et al., 2012, Kinjo, 2024).

1. Definition, formal properties, and quotient-theoretic role

A good moduli space is designed to retain the quotient-theoretic features of GIT while allowing positive-dimensional stabilizers and non-separated stack structures. In the standard setup, one considers a morphism

$\pi\colon \mathcal{X}\to X$

from an Artin stack $\mathcal{X}$ to an algebraic space $X$ such that $\pi_*$ on quasi-coherent sheaves is exact and $\mathcal{O}_X\to \pi_*\mathcal{O}_{\mathcal{X}}$ is an isomorphism. Several sources emphasize that these conditions generalize the GIT good quotient and also recover coarse moduli spaces for tame stacks in appropriate settings (Geraschenko et al., 2012, Edidin et al., 2010).

This formalism is especially useful because it replaces a naive orbit-space construction with a categorical and cohomological one. The resulting algebraic space is universal for maps from the stack to algebraic spaces, and the morphism is surjective and universally closed (Bu, 30 Dec 2025, Edidin et al., 2010). In many moduli problems, closed points of the good moduli space encode $S$-equivalence classes rather than isomorphism classes, reflecting the same quotient behavior familiar from semistable GIT.

A recurring theme in the literature is that good moduli space morphisms are not proper or separated in general. Nonetheless, they often retain “proper-like” features. Formal GAGA can hold under a resolution-property hypothesis, sheaf-theoretic base change and decomposition statements can hold, and valuative criteria admit root-stack refinements (Geraschenko et al., 2012, Kinjo, 2024, Bejleri et al., 11 Jul 2025). This combination of non-separated geometry with proper-like formal behavior is one of the central organizing principles of the subject.

2. Existence criteria and local structure

Existence of a good moduli space is not automatic. One influential criterion, proved by Alper and Smyth, is a weak analog of the Keel–Mori theorem. For an algebraic stack of finite type over an algebraically closed field, it is enough to have affine étale neighborhoods of closed points of the form $\pi$0 that are stabilizer preserving at closed points and send closed points to closed points, together with the requirement that the closed substack $\pi$1 admits a good moduli space for every $\pi$2-point $\pi$3 (Alper et al., 2012). This criterion was applied to show that the stacks of $\pi$4- and $\pi$5-stable curves admit good moduli spaces, yielding proper algebraic spaces central to the log minimal model program for $\pi$6 (Alper et al., 2012).

A different existence theorem, used prominently in later moduli constructions, is the Alper–Halpern-Leistner–Heinloth criterion: an algebraic stack of finite type over a characteristic-zero field with affine diagonal admits a separated good moduli space if and only if it is $\pi$7-complete and $\pi$8-complete (Damiolini et al., 2024). This criterion underlies projectivity results for moduli of semistable vector bundles on stacky curves and also appears in constructions of proper good moduli spaces in higher-dimensional moduli problems.

Local structure is controlled by stabilizers. A recent singularity-theoretic approach shows that if a normal algebraic stack with affine diagonal and finite generic stabilizers is relatively log-canonical at a point, then the stabilizer at that point is a finite extension of an algebraic torus; étale locally near that point, a good moduli space exists (Bongiorno, 26 Mar 2026). If the singularity is relatively canonical, then the locus of stable points is nonempty (Bongiorno, 26 Mar 2026). This ties the existence of local good moduli spaces to the birational geometry of singularities rather than solely to quotient presentations.

These results clarify a common misconception: reductive or toric stabilizer behavior is not merely a technical convenience but is closely tied to local moduli-space existence. The theory is therefore simultaneously quotient-theoretic and singularity-theoretic.

3. Morphisms, strength, and descent

One of the structural advances in the theory is the extension of Luna’s fundamental lemma from quotient stacks to general stacks with good moduli spaces. For a morphism $\pi$9 between stacks admitting good moduli spaces, with affine diagonals and standard finiteness hypotheses, the morphism is strong precisely when three conditions hold: $\pi_*$0 is special, it is fiberwise stabilizer preserving at every special point of $\pi_*$1, and for every special point $\pi_*$2, the object $\pi_*$3 is a trivial (ind-)vector bundle (Rydh, 2020). If $\pi_*$4 is regular, smooth, étale, open immersion, unramified, closed immersion, locally closed immersion, quasi-regular immersion, Koszul-regular immersion, monomorphism, flat, syntomic, local complete intersection, quasi-finite, finite, or locally of finite presentation, then the induced morphism on good moduli spaces inherits the same property (Rydh, 2020).

The same paper gives descent criteria for coherent sheaves and complexes. For a good moduli space $\pi_*$5 and a quasi-coherent module $\pi_*$6 of finite presentation, the conditions

$\pi_*$7

and a special-point triviality condition on both $\pi_*$8 and $\pi_*$9 are equivalent (Rydh, 2020). Derived analogues are formulated in terms of pseudo-coherence and the behavior of $\pi_*\mathcal{O}_{\mathcal{X}}=\mathcal{O}_X$0 (Rydh, 2020).

Smoothness of the moduli space itself is subtler. For a smooth Artin stack with properly stable good moduli space, questions of smoothness and flatness reduce étale locally to quotient stacks $\pi_*\mathcal{O}_{\mathcal{X}}=\mathcal{O}_X$1 by linearly reductive groups (Edidin et al., 2019). In this local GIT model, purity of the strictly semistable locus can characterize when $\pi_*\mathcal{O}_{\mathcal{X}}=\mathcal{O}_X$2 is smooth and the quotient morphism is flat: for irreducible stable representations of simple Lie groups, cofree is equivalent to pure, while for torus representations cofree is equivalent to coprincipal (Edidin et al., 2019). The same source also records counterexamples showing that reducibility or failure of coprincipality obstructs such a characterization (Edidin et al., 2019). Thus, purity is a precise criterion only in controlled representation-theoretic regimes.

4. Proper-like behavior in formal and cohomological settings

A major line of research develops the principle that good moduli space morphisms behave as if they were proper for several sheaf-theoretic operations. Formal GAGA is the earliest systematic example. If $\pi_*\mathcal{O}_{\mathcal{X}}=\mathcal{O}_X$3 is a good moduli space over a complete Noetherian local ring and $\pi_*\mathcal{O}_{\mathcal{X}}=\mathcal{O}_X$4 is the formal completion, then the completion functor

$\pi_*\mathcal{O}_{\mathcal{X}}=\mathcal{O}_X$5

is fully faithful; if the special fiber has the resolution property, then this functor is an equivalence, and the relevant quotient and resolution-property conditions become equivalent formulations of formal GAGA (Geraschenko et al., 2012). Quotient stacks satisfy the required resolution property, while explicit counterexamples show that the hypothesis is genuinely needed (Geraschenko et al., 2012).

Sheaf-theoretic properness is sharpened further by the decomposition theorem for good moduli morphisms. For a finite type Artin stack with affine diagonal admitting a good moduli space $\pi_*\mathcal{O}_{\mathcal{X}}=\mathcal{O}_X$6, the base change theorem holds in the setting of constructible sheaves and mixed Hodge modules, and $\pi_*\mathcal{O}_{\mathcal{X}}=\mathcal{O}_X$7 preserves weights (Kinjo, 2024). If $\pi_*\mathcal{O}_{\mathcal{X}}=\mathcal{O}_X$8 is smooth, then

$\pi_*\mathcal{O}_{\mathcal{X}}=\mathcal{O}_X$9

and the relevant cohomology and Borel–Moore homology groups are pure (Kinjo, 2024). The proof uses the class of “approximately proper” morphisms, into which good moduli space morphisms are placed via étale slices and induction on stabilizer dimension (Kinjo, 2024).

A valuative analogue appears in the root stack criterion. Given a good moduli space map and a DVR $\pi\colon \mathcal{X}\to X$0 with fraction field $\pi\colon \mathcal{X}\to X$1, any $\pi\colon \mathcal{X}\to X$2-point extends after passing to a suitable root stack $\pi\colon \mathcal{X}\to X$3, and a closed generic-image condition can also be arranged (Bejleri et al., 11 Jul 2025). This extends to gerbes banded by reductive groups under mild residue-characteristic hypotheses and has applications to torsors, rational points, homogeneous spaces, and fibrations (Bejleri et al., 11 Jul 2025).

Intersection theory remains more conjectural. For a smooth Artin stack with good moduli space, Edidin and Satriano define strong cycles and the strong relative Chow group $\pi\colon \mathcal{X}\to X$4, conjecturing that strong cycles form a subring of the Chow ring upstairs and that their images determine a meaningful intersection product on a subgroup of $\pi\colon \mathcal{X}\to X$5 downstairs (Edidin et al., 2016). This identifies a precise obstruction to transferring the full Chow ring from a smooth stack to its possibly singular good moduli space.

5. Birational methods, stabilizer reduction, and projectivity

Good moduli spaces support a substantial birational theory. Edidin and Rydh prove a canonical reduction of stabilizers for Artin stacks with stable good moduli spaces: there is a canonical sequence

$\pi\colon \mathcal{X}\to X$6

such that the maximum stabilizer dimension strictly decreases at each step, the induced maps on good moduli spaces are projective and birational, and, in the smooth case, the final stack is a gerbe over a tame stack while the final algebraic space has tame quotient singularities (Edidin et al., 2017). This is a complete generalization of Kirwan’s partial desingularization theorem from smooth GIT quotients to stacks with good moduli spaces (Edidin et al., 2017).

In the toric setting, Reichstein transforms give a concrete combinatorial incarnation of this philosophy. For Artin toric stacks, the Reichstein transform along toric substacks remains toric, and a finite sequence of such transforms produces a Deligne–Mumford toric stack with a proper birational morphism of good moduli spaces (Edidin et al., 2010). When the original good moduli space is projective, this procedure matches Kirwan’s partial desingularization in the GIT sense (Edidin et al., 2010).

Projectivity questions have recently been reframed using alterations. Any good moduli space $\pi\colon \mathcal{X}\to X$7 with $\pi\colon \mathcal{X}\to X$8 of finite type, affine stabilizers, and separated diagonal admits a splitting after a proper, generically finite cover of $\pi\colon \mathcal{X}\to X$9 (Bejleri et al., 2024). As an application, Kollár’s ampleness lemma is generalized to good moduli spaces: under explicit hypotheses on a vector bundle $\mathcal{X}$0, a quotient $\mathcal{X}$1, descent of $\mathcal{X}$2, finiteness of closed points in fibers of the classifying map, and weak semipositivity assumptions, the descended line bundle on a proper good moduli space is ample, hence the moduli space is projective (Bejleri et al., 2024). This provides a GIT-free projectivity criterion adapted to Artin-stack moduli problems.

6. Applications, variants, and derived extensions

Good moduli spaces now appear across a wide range of moduli constructions. The stacks of $\mathcal{X}$3- and $\mathcal{X}$4-stable curves admit good moduli spaces, providing intrinsic compactifications relevant to the log minimal model program (Alper et al., 2012). Proper good moduli spaces also exist for K-polystable $\mathcal{X}$5-Gorenstein smoothable log Fano pairs, and in the case of plane curves the resulting K-moduli spaces admit a finite wall-crossing structure; for small coefficient $\mathcal{X}$6, the K-moduli space agrees with the GIT moduli space, while the first wall crossing is a weighted blow-up of Kirwan type (Ascher et al., 2019).

For Calabi–Yau surface theory, projective asymptotically good moduli spaces have been constructed for boundary polarized CY surface pairs. These spaces interpolate between KSBA and K-moduli compactifications and furnish the ample model of the Hodge line bundle; in the case of K3 surfaces with a non-symplectic automorphism, the normalization gives the Baily–Borel compactification a modular interpretation (Blum et al., 2024). Proper good moduli spaces have also been constructed for Bridgeland semistable orthosymplectic complexes, using the Alper–Halpern-Leistner–Heinloth formalism and inheritance results for fixed loci and mapping stacks from finite groupoids (Bu, 30 Dec 2025).

On stacky curves, semistable vector bundles admit proper good moduli spaces, and a natural determinantal line bundle descends to an ample line bundle, proving projectivity (Damiolini et al., 2024). The same methods provide effective bounds for basepoint-freeness and yield new constructions of moduli spaces of parabolic bundles (Damiolini et al., 2024).

The theory has also been extended beyond classical algebraic stacks. In derived algebraic geometry, a morphism $\mathcal{X}$7 from a derived Artin stack to a derived algebraic space is a good moduli space if $\mathcal{X}$8 is universally of cohomological dimension $\mathcal{X}$9 and $X$0 is an equivalence (Ahlqvist et al., 2023). Under natural assumptions, existence reduces to the classical truncation, and derived versions of the étale slice theorem and partial desingularization follow (Ahlqvist et al., 2023). In a different categorical direction, good moduli spaces have been constructed for quasi-compact closed substacks of Toën–Vaquié moduli of objects in stable $X$1-categories satisfying openness-of-flatness and subobject hypotheses, with an application to moduli of perverse sheaves (Lampetti, 27 Oct 2025).

A plausible implication is that good moduli spaces have become a unifying output notion for moduli problems that are neither strictly GIT nor strictly Deligne–Mumford. The recent literature shows them functioning simultaneously as quotient objects, as receptacles for proper-like sheaf theory, and as targets of birational and derived enhancement procedures.