Representational Desingularization Overview

• Representational desingularization is the process of replacing a singular object with a smooth one using additional categorical, combinatorial, or homological data.
• It applies to diverse settings like quiver Grassmannians, orbit closures, and Lie groupoids where singular moduli spaces are resolved via rigidity and blowup methods.
• The approach embeds the original problem into a richer ambient structure, enabling recovery through restriction, pushforward, or quotient techniques.

Searching arXiv for the supplied papers and closely related formulations of desingularization across representation-theoretic settings. arXiv search: representational desingularization, quiver Grassmannian desingularization, singular Artin monoid desingularization, orbit closure noncommutative desingularization. Representational desingularization denotes a family of constructions in which a singular object is replaced by a smooth, non-singular, tame, or finite-homological-dimension object built from additional representation-theoretic, categorical, or combinatorial data. Across the cited works, the singular object may be a quiver Grassmannian, an orbit closure in a linear representation, a simplicial set, a semialgebraic set, a function field, a Lie groupoid, an Artin stack with good moduli space, or a singular Artin monoid; the desingularizing object may be a smooth moduli space of lifted subobjects, a blowup or projective bundle over a homogeneous space, a reflector $D$, a noncommutative algebra $\mathcal A=\mathcal{E}nd_X(\mathcal F)$, or a categorical action on $D^b(\mathcal O)$ (Keller et al., 2013, Movshev, 2011, Fjellbo, 2020, Weyman et al., 2012, Edidin et al., 2017, Jonsson et al., 2022). A plausible synthesis is that the common mechanism is to embed the original problem into a larger ambient structure with better homological, equivariant, or combinatorial control, and then descend back to the original object by restriction, pushforward, or quotient.

1. Recurring pattern and basic forms

The surveyed literature exhibits several recurring templates. In quiver-theoretic settings, one replaces submodules of a non-rigid module $M$ by submodules of a related rigid object such as $\mathrm{KLR}(M)$ or $\widehat M$, and the desingularization map is induced by restriction of submodules (Keller et al., 2013, Irelli et al., 2012). In orbit-closure settings, one starts from a $G$-stable subvariety defined by invariant equations and resolves it by a blowup along a closed orbit or by passing to an endomorphism algebra on a commutative resolution (Movshev, 2011, Weyman et al., 2012). In simplicial homotopy theory, desingularization is the reflector

$D : sSet \rightleftarrows nsSet : U,$

with $DX$ characterized by the universal property of representing all maps from $X$ into non-singular simplicial sets (Fjellbo, 2020). In stack-theoretic and Lie-groupoid settings, singular behavior is encoded by stabilizers or by the local structure of a Lie algebroid, and desingularization proceeds by canonical reduction of stabilizers or by replacing a unit space $\mathcal A=\mathcal{E}nd_X(\mathcal F)$0 with a blowup $\mathcal A=\mathcal{E}nd_X(\mathcal F)$1 (Edidin et al., 2017, Nistor, 2015).

Domain	Singular object	Desingularizing object
Quiver geometry	$\mathcal A=\mathcal{E}nd_X(\mathcal F)$2	$\mathcal A=\mathcal{E}nd_X(\mathcal F)$3
Gravitational spinors	$\mathcal A=\mathcal{E}nd_X(\mathcal F)$4	$\mathcal A=\mathcal{E}nd_X(\mathcal F)$5
Simplicial sets	$\mathcal A=\mathcal{E}nd_X(\mathcal F)$6	$\mathcal A=\mathcal{E}nd_X(\mathcal F)$7, obtained by iterated enforced collapse
Good moduli spaces	$\mathcal A=\mathcal{E}nd_X(\mathcal F)$8	$\mathcal A=\mathcal{E}nd_X(\mathcal F)$9
Orbit closures	$D^b(\mathcal O)$0	$D^b(\mathcal O)$1

This diversity matters because it prevents reduction of representational desingularization to a single birational recipe. The same word “desingularization” can refer to a proper birational morphism from a smooth variety, a projective bundle over a homogeneous space, a reflective localization, a noncommutative algebra of finite global dimension, or a weak action of a singular Artin monoid on a derived category (Movshev, 2011, Fjellbo, 2020, Weyman et al., 2012, Jonsson et al., 2022).

2. Quiver Grassmannians, rigidity, and graded quiver varieties

For a finite acyclic quiver $D^b(\mathcal O)$2, a representation $D^b(\mathcal O)$3, and a dimension vector $D^b(\mathcal O)$4, the quiver Grassmannian $D^b(\mathcal O)$5 parametrizes subrepresentations of $D^b(\mathcal O)$6 of dimension vector $D^b(\mathcal O)$7. These varieties are projective but in general singular; Reineke showed every projective variety occurs as some quiver Grassmannian (Keller et al., 2013). Cerulli–Feigin–Reineke constructed desingularizations for Dynkin quivers by replacing $D^b(\mathcal O)$8 with an auxiliary module $D^b(\mathcal O)$9 over an algebra $M$0, and then mapping submodules of $M$1 down to submodules of $M$2 (Irelli et al., 2012).

Keller–Scherotzke recast this picture through Nakajima categories and graded quiver varieties. With $M$3, restriction

$M$4

admits an intermediate extension $M$5, obtained as an intermediate Kan extension. For an $M$6-module $M$7, the key rigidifying hypothesis is

$M$8

Under this hypothesis, each $M$9 is smooth and equidimensional, and the main desingularization theorem produces

$\mathrm{KLR}(M)$0

a desingularization of $\mathrm{KLR}(M)$1 (Keller et al., 2013). The source is smooth, $\mathrm{KLR}(M)$2 is proper and surjective, and it induces an isomorphism on dense open strata labeled by Nakajima dimension vectors $\mathrm{KLR}(M)$3. Fiberwise, Theorem 3.13 identifies the fiber over $\mathrm{KLR}(M)$4 with a quiver Grassmannian of submodules of $\mathrm{KLR}(M)$5 (Keller et al., 2013).

This construction extends the Dynkin-quiver result of Cerulli–Feigin–Reineke in two directions. First, it identifies the CFR functor $\mathrm{KLR}(M)$6 with the restriction of $\mathrm{KLR}(M)$7 in the Dynkin case. Second, it covers all modules over the repetitive algebra of any iterated tilted algebra $\mathrm{KLR}(M)$8 of Dynkin type, because for such $\mathrm{KLR}(M)$9 the relevant projective category is Frobenius and Lemma 2.6(c) implies rigidity of $\widehat M$0 (Keller et al., 2013). In the older Dynkin-only formulation, the desingularizing variety is a quiver Grassmannian for an algebra $\widehat M$1 derived equivalent to the Auslander algebra, and the map

$\widehat M$2

is induced by restriction of a $\widehat M$3-subrepresentation to a $\widehat M$4-subrepresentation (Irelli et al., 2012). In both papers, the singular parameter space is replaced by a moduli space of subobjects of a more rigid object, and the geometry is controlled by vanishing of $\widehat M$5 and global dimension $\widehat M$6.

A recurring point in this quiver-theoretic branch is that the desingularization is itself representation-theoretic: the source is again a quiver Grassmannian or a bistable locus in a quiver Grassmannian, and the fibers are again quiver Grassmannians (Irelli et al., 2012, Keller et al., 2013). This suggests a particularly strong form of representational desingularization in which both the singularity and its resolution remain inside the same moduli-theoretic universe.

3. Orbit closures, invariant equations, and noncommutative models

Movshev’s study of eleven-dimensional gravitational spinors is an explicit example where a singular representation variety is resolved by homogeneous geometry. The basic representation is the 32-dimensional spinor module $\widehat M$7 of $\widehat M$8, with invariant quadrics

$\widehat M$9

defining

$G$0

Igusa’s orbit classification yields that $G$1 is the closure of the 22-dimensional orbit, its smooth locus is $G$2, and its singular locus is the closed orbit $G$3 (Movshev, 2011). The $G$4-component of $G$5 produces a rational $G$6-map

$G$7

undefined exactly on $G$8. The desingularization $G$9 is the closure of the graph of $D : sSet \rightleftarrows nsSet : U,$0, equivalently the blowup of $D : sSet \rightleftarrows nsSet : U,$1 along $D : sSet \rightleftarrows nsSet : U,$2. It is smooth and admits

$D : sSet \rightleftarrows nsSet : U,$3

with fiber $D : sSet \rightleftarrows nsSet : U,$4, while the exceptional divisor is the flag variety

$D : sSet \rightleftarrows nsSet : U,$5

a quadric bundle over $D : sSet \rightleftarrows nsSet : U,$6 (Movshev, 2011). The same resolution mediates a reformulation of the linearized eleven-dimensional supergravity equations as CR-holomorphic data on the super-homogeneous space

$D : sSet \rightleftarrows nsSet : U,$7

Here the desingularization is representation-theoretically natural at every level: the singular variety is an orbit closure, the resolution is a projective bundle over a homogeneous space, and the exceptional divisor is itself homogeneous (Movshev, 2011).

A different but closely related branch appears in noncommutative desingularization of orbit closures for some representations of $D : sSet \rightleftarrows nsSet : U,$8. For determinantal, symmetric determinantal, and Pfaffian varieties, the singular affine variety $D : sSet \rightleftarrows nsSet : U,$9 is an orbit closure under a $DX$0-action, and there is a commutative desingularization

$DX$1

where $DX$2 is the total space of a $DX$3-equivariant vector bundle over a Grassmannian (Weyman et al., 2012). If $DX$4 is a tilting bundle on the Grassmannian, then $DX$5 is a tilting bundle on $DX$6, and

$DX$7

has finite global dimension. This gives a noncommutative weak desingularization, and in favorable cases an NCCR, of $DX$8 (Weyman et al., 2012). For maximal minors of square matrices and symmetric matrices, the construction gives a non-commutative crepant resolution, while in the Pfaffian and lower-rank symmetric cases the endomorphism algebra is not maximal Cohen–Macaulay and not reflexive, so one obtains only a weak desingularization together with a reflexive hull (Weyman et al., 2012).

The quiver with relations of these endomorphism algebras is computed from exceptional collections on partial flag varieties. The general theorem identifies simples as

$DX$9

and arrows and relations are generated by $X$0 and $X$1 between these simples (Weyman et al., 2012). The resulting quivers are $X$2-equivariant: vertices are partitions, arrows are labeled by representations such as $X$3, $X$4, or $X$5, and relations are determined by Littlewood–Richardson multiplicities and Borel–Weil–Bott calculations (Weyman et al., 2012). In this sense the singularity is not only resolved but reorganized into a representation-theoretic algebra whose module category is derived equivalent to the commutative resolution.

4. Reflectors, monoid maps, and categorification

In simplicial homotopy theory, desingularization is formulated as a reflector onto non-singular simplicial sets. A simplex $X$6 is embedded when its representing map $X$7 is degreewise injective, and $X$8 is non-singular if every non-degenerate simplex is embedded. The inclusion

$X$9

has a left adjoint $\mathcal A=\mathcal{E}nd_X(\mathcal F)$00, and the unit

$\mathcal A=\mathcal{E}nd_X(\mathcal F)$01

is terminal among maps from $\mathcal A=\mathcal{E}nd_X(\mathcal F)$02 into non-singular targets (Fjellbo, 2020). The abstract construction

$\mathcal A=\mathcal{E}nd_X(\mathcal F)$03

is replaced by a transfinite iterative process based on the enforced collapse functor $\mathcal A=\mathcal{E}nd_X(\mathcal F)$04. The main theorem states that for each simplicial set $\mathcal A=\mathcal{E}nd_X(\mathcal F)$05 there exists an ordinal $\mathcal A=\mathcal{E}nd_X(\mathcal F)$06 such that

$\mathcal A=\mathcal{E}nd_X(\mathcal F)$07

and at stage $\mathcal A=\mathcal{E}nd_X(\mathcal F)$08 the object is already non-singular (Fjellbo, 2020). The paper explicitly constructs the enforcer $\mathcal A=\mathcal{E}nd_X(\mathcal F)$09 of a non-degenerate simplex $\mathcal A=\mathcal{E}nd_X(\mathcal F)$10, pushes out along all enforcers, and proves that every singular non-degenerate simplex at stage $\mathcal A=\mathcal{E}nd_X(\mathcal F)$11 becomes degenerate at stage $\mathcal A=\mathcal{E}nd_X(\mathcal F)$12. The representational aspect here is categorical rather than geometric: $\mathcal A=\mathcal{E}nd_X(\mathcal F)$13 is the object that represents all maps from $\mathcal A=\mathcal{E}nd_X(\mathcal F)$14 into non-singular simplicial sets.

For singular Artin monoids, the classical desingularization map is the monoid homomorphism

$\mathcal A=\mathcal{E}nd_X(\mathcal F)$15

The paper generalizes this to maps

$\mathcal A=\mathcal{E}nd_X(\mathcal F)$16

depending on a Laurent polynomial $\mathcal A=\mathcal{E}nd_X(\mathcal F)$17 on connected components of the odd skeleton, and proves Zariski generic injectivity in type $\mathcal A=\mathcal{E}nd_X(\mathcal F)$18, in dihedral type $\mathcal A=\mathcal{E}nd_X(\mathcal F)$19, and in right-angled types (Jonsson et al., 2022). It also constructs Hecke-algebra-valued analogues and several finite diagrammatic quotients, including the double Catalan monoid, $\mathcal A=\mathcal{E}nd_X(\mathcal F)$20, $\mathcal A=\mathcal{E}nd_X(\mathcal F)$21, $\mathcal A=\mathcal{E}nd_X(\mathcal F)$22, and Brauer-type monoids (Jonsson et al., 2022).

The main higher-representational step is a categorification via BGG category $\mathcal A=\mathcal{E}nd_X(\mathcal F)$23. On $\mathcal A=\mathcal{E}nd_X(\mathcal F)$24, the braid generators act by shuffling functors

$\mathcal A=\mathcal{E}nd_X(\mathcal F)$25

and the singular generators act by the two-term complex

$\mathcal A=\mathcal{E}nd_X(\mathcal F)$26

Theorem 35 proves that

$\mathcal A=\mathcal{E}nd_X(\mathcal F)$27

extends to a weak action of the singular Artin monoid $\mathcal A=\mathcal{E}nd_X(\mathcal F)$28 on $\mathcal A=\mathcal{E}nd_X(\mathcal F)$29, giving a categorification of the classical desingularization map for finite Weyl groups (Jonsson et al., 2022). The crucial point is that the singular crossing is realized as a cone of a natural transformation between braid-group functors, so the passage from $\mathcal A=\mathcal{E}nd_X(\mathcal F)$30 to $\mathcal A=\mathcal{E}nd_X(\mathcal F)$31 becomes a passage from a singular generator to a mapping cone in the derived category.

These two cases show that desingularization can be representational without being birational. In one case it is a reflective localization in $\mathcal A=\mathcal{E}nd_X(\mathcal F)$32; in the other it is a monoid map into a group algebra and then a weak action on a derived category (Fjellbo, 2020, Jonsson et al., 2022).

5. Stabilizers, Lie algebroids, Nash corners, and valuation trees

For Artin stacks with good moduli spaces, Edidin–Rydh prove that if

$\mathcal A=\mathcal{E}nd_X(\mathcal F)$33

is a stable good moduli space morphism, then there is a canonical sequence

$\mathcal A=\mathcal{E}nd_X(\mathcal F)$34

such that the maximum dimension of a stabilizer of a point of $\mathcal A=\mathcal{E}nd_X(\mathcal F)$35 is strictly smaller than the maximum dimension of a stabilizer of $\mathcal A=\mathcal{E}nd_X(\mathcal F)$36, and the final stack $\mathcal A=\mathcal{E}nd_X(\mathcal F)$37 has constant stabilizer dimension (Edidin et al., 2017). The induced morphisms on good moduli spaces

$\mathcal A=\mathcal{E}nd_X(\mathcal F)$38

are projective and birational. If $\mathcal A=\mathcal{E}nd_X(\mathcal F)$39 is smooth, each intermediate stack is smooth, the final stack is a gerbe over a tame stack, and the algebraic space $\mathcal A=\mathcal{E}nd_X(\mathcal F)$40 has tame quotient singularities; combined with Bergh’s destackification theorem, this yields a full desingularization of $\mathcal A=\mathcal{E}nd_X(\mathcal F)$41 (Edidin et al., 2017). Here the singularity is encoded by stabilizer dimension, and desingularization is canonically achieved by successive saturated blowups or Reichstein transforms along the maximal stabilizer loci.

In Lie-groupoid geometry, Nistor desingularizes a Lie groupoid $\mathcal A=\mathcal{E}nd_X(\mathcal F)$42 along an $\mathcal A=\mathcal{E}nd_X(\mathcal F)$43-tame submanifold $\mathcal A=\mathcal{E}nd_X(\mathcal F)$44. The local structure theorem identifies $\mathcal A=\mathcal{E}nd_X(\mathcal F)$45 near $\mathcal A=\mathcal{E}nd_X(\mathcal F)$46 with a fibered pull-back groupoid, and the desingularization

$\mathcal A=\mathcal{E}nd_X(\mathcal F)$47

is obtained by replacing the unit space $\mathcal A=\mathcal{E}nd_X(\mathcal F)$48 with the blowup $\mathcal A=\mathcal{E}nd_X(\mathcal F)$49 and gluing in an edge-modified local model built from the adiabatic groupoid (Nistor, 2015). The space of units of $\mathcal A=\mathcal{E}nd_X(\mathcal F)$50 is $\mathcal A=\mathcal{E}nd_X(\mathcal F)$51, and its Lie algebroid is canonically identified with the desingularized algebroid

$\mathcal A=\mathcal{E}nd_X(\mathcal F)$52

This construction is designed for analysis on singular spaces, especially edge pseudodifferential calculus and asymptotically hyperbolic variants (Nistor, 2015). The representational content lies in replacing a singular orbit structure by a groupoid whose boundary fibers are explicit solvable Lie groups, so that pseudodifferential operators and $\mathcal A=\mathcal{E}nd_X(\mathcal F)$53-algebras become tractable.

In semialgebraic geometry, strong desingularization replaces a semialgebraic set $\mathcal A=\mathcal{E}nd_X(\mathcal F)$54 by a Nash manifold with corners $\mathcal A=\mathcal{E}nd_X(\mathcal F)$55 of the same dimension inside a nonsingular real algebraic set $\mathcal A=\mathcal{E}nd_X(\mathcal F)$56, together with a proper surjective map $\mathcal A=\mathcal{E}nd_X(\mathcal F)$57 that is a Nash diffeomorphism away from a lower-dimensional set (Carbone et al., 2023). Theorem 1.5 gives, for a closed semialgebraic set connected by analytic paths, an irreducible nonsingular real algebraic $\mathcal A=\mathcal{E}nd_X(\mathcal F)$58, a connected Nash manifold with corners $\mathcal A=\mathcal{E}nd_X(\mathcal F)$59, a polynomial map $\mathcal A=\mathcal{E}nd_X(\mathcal F)$60 with

$\mathcal A=\mathcal{E}nd_X(\mathcal F)$61

and a closed semialgebraic subset $\mathcal A=\mathcal{E}nd_X(\mathcal F)$62 with $\mathcal A=\mathcal{E}nd_X(\mathcal F)$63 such that

$\mathcal A=\mathcal{E}nd_X(\mathcal F)$64

is a Nash diffeomorphism (Carbone et al., 2023). Theorem 1.7 adds a folding construction

$\mathcal A=\mathcal{E}nd_X(\mathcal F)$65

locally modeled by

$\mathcal A=\mathcal{E}nd_X(\mathcal F)$66

so that corners arise by folding a Nash manifold along a normal-crossings divisor (Carbone et al., 2023). This is representational in the literal sense that $\mathcal A=\mathcal{E}nd_X(\mathcal F)$67 is represented as the image of a smooth object with corners and a surjective algebraic or Nash map.

For function fields, desingularization is recast valuation-theoretically. Leonard studies

$\mathcal A=\mathcal{E}nd_X(\mathcal F)$68

and seeks to uniquely describe $\mathcal A=\mathcal{E}nd_X(\mathcal F)$69-dimensional valuations by $\mathcal A=\mathcal{E}nd_X(\mathcal F)$70 explicit independent local parameters and $\mathcal A=\mathcal{E}nd_X(\mathcal F)$71 dependent local unit (Leonard, 2019). The desingularization is encoded by a rooted tree whose nodes are labeled by domains $\mathcal A=\mathcal{E}nd_X(\mathcal F)$72, equality constraints $\mathcal A=\mathcal{E}nd_X(\mathcal F)$73, inequality constraints $\mathcal A=\mathcal{E}nd_X(\mathcal F)$74, and birational change-of-variables maps. The endpoint is a “strongly resolved form” in which one variable is a local unit and can be solved recursively as a formal Laurent-series-type expression in the remaining local parameters (Leonard, 2019). The shift from varieties to valuation trees is a particularly explicit form of representational encoding.

6. Invariants, computations, and conceptual boundaries

Several papers make the representational content computationally explicit. In computational desingularization, a resolution is represented as a rooted tree of affine charts, together with exceptional divisors, transforms of ideals, and coordinate substitutions (Frühbis-Krüger, 2013). This chart-tree representation supports computation of the intersection form and dual graph of a surface resolution, discrepancies

$\mathcal A=\mathcal{E}nd_X(\mathcal F)$75

the log-canonical threshold

$\mathcal A=\mathcal{E}nd_X(\mathcal F)$76

and the Denef–Loeser topological zeta function

$\mathcal A=\mathcal{E}nd_X(\mathcal F)$77

from the same resolution data (Frühbis-Krüger, 2013). This computational viewpoint reinforces the idea that desingularization can be a structured representation of singularity data rather than only a geometric existence theorem.

In characteristic-zero embedded resolution, Bierstone–Milman and collaborators show how the desingularization invariant

$\mathcal A=\mathcal{E}nd_X(\mathcal F)$78

together with the component counts and monomial exponents $\mathcal A=\mathcal{E}nd_X(\mathcal F)$79, characterizes simple normal crossings and drives a sequence of blowings-up that avoids the already snc locus (Bierstone et al., 2012). The special values

$\mathcal A=\mathcal{E}nd_X(\mathcal F)$80

and the local normal form

$\mathcal A=\mathcal{E}nd_X(\mathcal F)$81

make the “representation” of a singularity by invariant data completely explicit (Bierstone et al., 2012). This is not a representation-theoretic construction in the $\mathcal A=\mathcal{E}nd_X(\mathcal F)$82 sense, but it is a particularly clear instance in which desingularization is guided by an invariant that encodes local normal forms.

The surveyed literature also shows that representational desingularization is not confined to commutative, birational, finite-step procedures. It may be noncommutative and derived-categorical, as in $\mathcal A=\mathcal{E}nd_X(\mathcal F)$83 and category-$\mathcal A=\mathcal{E}nd_X(\mathcal F)$84 actions (Weyman et al., 2012, Jonsson et al., 2022); it may be functorial and transfinite, as in the sequence $\mathcal A=\mathcal{E}nd_X(\mathcal F)$85 for simplicial sets (Fjellbo, 2020); or it may terminate in a tame stack, a gerbe, or a Nash manifold with corners rather than a smooth scheme (Edidin et al., 2017, Carbone et al., 2023). A common misconception is therefore to identify desingularization exclusively with a proper birational morphism from a smooth variety. The surveyed constructions show a broader pattern: singularities can be resolved, or partially resolved, by passing to a larger algebra, category, groupoid, moduli problem, or combinatorial reflector whose structure is better suited to the ambient representation-theoretic problem.

A plausible synthesis is that representational desingularization proceeds by four recurrent moves. First, isolate a singular object whose defining data are already representation-theoretic or categorical: a module variety, orbit closure, quotient stack, singular braid monoid, simplicial set, or function field (Keller et al., 2013, Movshev, 2011, Edidin et al., 2017, Jonsson et al., 2022, Fjellbo, 2020, Leonard, 2019). Second, embed it into a larger ambient structure with better control: a graded Nakajima category, a homogeneous bundle over a Grassmannian, a good moduli space, a blowup groupoid, or a derived category (Keller et al., 2013, Weyman et al., 2012, Nistor, 2015). Third, use rigidity, linearly reductive stabilizers, projective bundles, tilting bundles, Kan extensions, or valuation parameters to build the desingularizing object (Keller et al., 2013, Edidin et al., 2017, Movshev, 2011, Fjellbo, 2020, Leonard, 2019). Fourth, recover the original object by a restriction, blowdown, good moduli space morphism, quotient, or decategorification map. The resulting theory is not uniform in technique, but it is notably uniform in architecture.