ADHM Quiver Varieties: Geometry and Gauge Theory

• ADHM quiver varieties are moduli spaces defined via ADHM equations that describe framed instantons and torsion-free sheaves.
• They unify concepts from gauge theory, symplectic geometry, and geometric invariant theory through stable quiver representations.
• Extensions such as bow varieties and chain quivers expand the applicability of ADHM constructions in modern mathematical physics.

ADHM quiver varieties are moduli spaces constructed via representations of certain quivers subject to moment-map-type relations (ADHM equations), often equipped with stability conditions in the geometric invariant theory (GIT) sense. These varieties unify mathematical structures in instanton moduli, symplectic and hyperkähler geometry, and enumerative representation theory, and they bridge gauge-theoretic and algebro-geometric perspectives. The acronym ADHM refers to Atiyah–Drinfeld–Hitchin–Manin, whose eponymous construction originally parameterized instantons via linear algebra data. This conceptual apparatus was generalized and systematized by Nakajima, leading to a vast interplay with quiver representations, geometric representation theory, and contemporary mathematical physics.

1. Definition and Core Structures

The prototypical ADHM quiver consists of two nodes: a “gauge” vertex (dimension $k$) and a “framing” vertex (dimension $N$), connected by two loop arrows at the gauge node and two framing arrows. The representation space is

$$\operatorname{Rep}(Q) = \operatorname{End}(V)^{\oplus 2} \oplus \operatorname{Hom}(W, V) \oplus \operatorname{Hom}(V, W)$$

with $V \cong \mathbb{C}^k$, $W \cong \mathbb{C}^N$. The ADHM equations are: $$\mu_\mathbb{C}(B_1,B_2,I,J) = [B_1, B_2] + IJ = 0 \qquad \mu_\mathbb{R}(B_1,B_2,I,J) = [B_1,B_1^\dagger] + [B_2,B_2^\dagger] + I I^\dagger - J^\dagger J = 0$$ These are the complex and real moment map equations for the unitary action of $\mathrm{U}(k)$. The hyperkähler (or holomorphic symplectic) quotient

$$\mathcal{M}_{N,k} = \{ (B_1, B_2, I, J) \mid \mu_\mathbb{C}=0,\ \mu_\mathbb{R}=0 \} / \mathrm{U}(k)$$

realizes the quiver variety as a moduli space (e.g., of framed torsion-free sheaves on $\mathbb{P}^2$) (Koroteev, 2018, Bartocci et al., 2016, Kanno, 2020, Kim, 2010, Jardim et al., 2010).

2. Stability, GIT, and Moduli Interpretations

Nakajima’s stability condition (framing) ensures that the orbits in the quotient are properly separated and the resulting space is smooth. A point $(B_1,B_2,I,J)$ is stable if there is no proper subspace $N$0 preserved by $N$1, $N$2 and containing $N$3. The GIT quotient then produces the quiver variety: $N$4 where the character $N$5 encodes the stability parameter. These moduli spaces are isomorphic to moduli of instantons, framed sheaves, or perverse coherent sheaves, depending on context (Jardim et al., 2010, Bartocci et al., 2016, Kim, 2010, Kanno, 2020). For example, in the case of $N$6, the ADHM data correspond to a three-term monad complex whose cohomology yields perverse coherent sheaves trivialized at infinity (Jardim et al., 2010).

3. Generalizations: Bow Varieties, Quiver Chains, and Non-Nakajima Cases

ADHM-type quiver descriptions extend beyond the classical Jordan quiver to more elaborate configurations corresponding to other geometric settings:

• Bow Varieties (Cherkis): These describe instanton moduli on Taub–NUT spaces or Coulomb branches of affine type $N$7 gauge theories, with data specified by intervals (“wavy lines”) and “$N$8”/“$N$9” marked points. The bow variety is obtained as a hyperkähler quotient of the space of all such data by unitary gauge groups, subject to generalized moment map equations (Nakajima et al., 2016, Ji, 2023).
• Type-A Chain Quivers and Weyl Mutations: For quivers associated to Dynkin diagrams of type $$\operatorname{Rep}(Q) = \operatorname{End}(V)^{\oplus 2} \oplus \operatorname{Hom}(W, V) \oplus \operatorname{Hom}(V, W)$$0, “ADHM-like” moment-map equations govern representations whose moduli varieties realize rectangular representations of Yangian algebras. Dualities (Weyl mutations) induce equivalences between quiver varieties in different “stability chambers,” generalizing Seiberg-like duality (Galakhov et al., 16 Jan 2026).
• Non-Nakajima Components: Certain quiver varieties, like those realizing Hilbert schemes of points on total spaces of $$\operatorname{Rep}(Q) = \operatorname{End}(V)^{\oplus 2} \oplus \operatorname{Hom}(W, V) \oplus \operatorname{Hom}(V, W)$$1, are irreducible components of more general representation varieties with ADHM-type relations. However, for $$\operatorname{Rep}(Q) = \operatorname{End}(V)^{\oplus 2} \oplus \operatorname{Hom}(W, V) \oplus \operatorname{Hom}(V, W)$$2, the quiver is not a double, and the resulting variety typically admits only maximal-rank Poisson, not holomorphic symplectic, structures (Bartocci et al., 2015).

Geometric Space	Quiver Type	Moment Map Structure	Symplectic Structure
$$\operatorname{Rep}(Q) = \operatorname{End}(V)^{\oplus 2} \oplus \operatorname{Hom}(W, V) \oplus \operatorname{Hom}(V, W)$$3, $$\operatorname{Rep}(Q) = \operatorname{End}(V)^{\oplus 2} \oplus \operatorname{Hom}(W, V) \oplus \operatorname{Hom}(V, W)$$4	Jordan + framing	ADHM equations	Hyperkähler
Taub–NUT, ALE spaces	Bow/Affine $$\operatorname{Rep}(Q) = \operatorname{End}(V)^{\oplus 2} \oplus \operatorname{Hom}(W, V) \oplus \operatorname{Hom}(V, W)$$5	Generalized ADHM/bow	Hyperkähler
$$\operatorname{Rep}(Q) = \operatorname{End}(V)^{\oplus 2} \oplus \operatorname{Hom}(W, V) \oplus \operatorname{Hom}(V, W)$$6	Special McKay	Non-double ADHM, GIT cut	Maximal-rank Poisson (for $$\operatorname{Rep}(Q) = \operatorname{End}(V)^{\oplus 2} \oplus \operatorname{Hom}(W, V) \oplus \operatorname{Hom}(V, W)$$7)
Riemann surface $$\operatorname{Rep}(Q) = \operatorname{End}(V)^{\oplus 2} \oplus \operatorname{Hom}(W, V) \oplus \operatorname{Hom}(V, W)$$8, vector bundles	Nakajima quiver bundles	Bundle-valued ADHM	Hyperkähler (with constraints)

4. Enumerative Geometry, BPS Invariants, and Representation Theory

ADHM quiver varieties underpin calculations of BPS state counts, equivariant (quantum) $$\operatorname{Rep}(Q) = \operatorname{End}(V)^{\oplus 2} \oplus \operatorname{Hom}(W, V) \oplus \operatorname{Hom}(V, W)$$9-theory, and partition functions central to gauge theory and string theory. Fixed points of torus actions on these quiver varieties are combinatorially described by tuples of Young diagrams (or their higher-dimensional generalizations such as plane/solid partitions) (Kanno, 2020, Koroteev, 2018). Partition functions obtained via equivariant localization manifest as plethystic exponentials, directly relating to Nekrasov’s instanton partition functions.

Quantum $V \cong \mathbb{C}^k$0-theoretic multiplication operators’ eigenvalues for these varieties are conjecturally related to spectra of integrable systems, such as the elliptic Ruijsenaars–Schneider model, establishing deep connections between quantum geometry of quiver varieties and integrable representation theory (Koroteev, 2018).

5. Bundles, Quasimaps, and Further Generalizations

The theory globalizes to vector bundles over higher-dimensional bases:

• Nakajima Quiver Bundles: For a quiver $V \cong \mathbb{C}^k$1 and a variety $V \cong \mathbb{C}^k$2, a Nakajima bundle representation assigns (twisted) vector bundles to each vertex and sections to the arrows, with moment-map equations generalizing the ADHM system to bundle-valued data. The resulting moduli space is a Hamiltonian reduction for a gauge group of bundle automorphisms, satisfying a Hitchin–Kobayashi correspondence equating polystability with solutions to these moment-map equations (Jeffrey et al., 2024).
• Quasimap Theory: Stable quasimaps from curves to quiver varieties with imposed ADHM equations yield moduli spaces with symmetric obstruction theories, interpolating between the algebro-geometric GIT quotient and gauge-theoretic perspectives. In appropriate stability chambers (large parameter $V \cong \mathbb{C}^k$3), the moduli of twisted ADHM bundles coincide with stable quasimap moduli spaces (Kim, 2010).

6. Monads, Mirror Symmetry, and Derived Equivalences

ADHM data admit reformulations as monads—exact complexes of vector bundles—whose cohomology realizes framed torsion-free sheaves, perverse sheaves, or more exotic objects. In the context of homological mirror symmetry, ADHM quiver varieties correspond to the moduli of Maurer–Cartan deformations of immersed Lagrangian spheres or more general plumbing configurations. The analytic continuation to the mirror side identifies these moduli with derived categories of coherent sheaves on certain resolutions or ALE spaces (Hu et al., 2024).

Moreover, explicit correspondences and isomorphisms (such as the generalized Mirković–Vybornov isomorphism) connect ADHM/bow varieties to slices in the cotangent bundles of flag varieties, intertwining geometric representation theory with the symplectic geometry of quiver varieties (Ji, 2023).

7. Geometry, Singularities, and Strata

The ADHM quiver variety admits a natural stratification by rank of stabilizing subspaces. For ADHM data $V \cong \mathbb{C}^k$4, the “stable” locus (where the data are cyclic, i.e., generate all of $V \cong \mathbb{C}^k$5) is open and dense, producing a smooth irreducible hyperkähler manifold. Lower-dimensional strata, where the stabilizing subspace is smaller, are singular, and for small framing dimension may lead to reducibility (Jardim et al., 2010). At the moduli stack level, these strata correspond to perverse coherent sheaves of fixed numerical type, and the associated quotient stacks reflect the structural decomposition induced by the ADHM data.

---

The ADHM quiver variety thus constitutes a cornerstone in modern geometry, representation theory, and mathematical physics, admitting variants and extensions connecting disparate fields through the ubiquity of moment-map-type relations, stability quotients, and their enumerative and homological consequences (Jardim et al., 2010, Bartocci et al., 2016, Bartocci et al., 2015, Koroteev, 2018, Kanno, 2020, Ji, 2023, Galakhov et al., 16 Jan 2026, Jeffrey et al., 2024, Kim, 2010, Hu et al., 2024, Nakajima et al., 2016).