Virtual Fundamental Class

• Virtual fundamental class is a homological substitute for the usual fundamental class in moduli spaces that are singular or fail transversality.
• It leverages perfect obstruction theories and intrinsic normal cones to construct a virtual cycle via methods like Behrend–Fantechi, Kuranishi atlases, and derived techniques.
• Applications span enumerative geometry, from Gromov–Witten and Donaldson–Thomas invariants to contact homology, enabling rigorous counts despite geometric irregularities.

Searching arXiv for relevant papers on virtual fundamental classes. Running arXiv search for "virtual fundamental class". A virtual fundamental class is a replacement for the ordinary fundamental class of a moduli space when the space is singular, fails transversality, or has the wrong geometric dimension. In the Behrend–Fantechi formalism, one starts with a Deligne–Mumford stack $\mathcal M$ and a perfect obstruction theory $\phi:E^\bullet\to L_{\mathcal M}$, forms the intrinsic normal cone $$\mathfrak C_{\mathcal M}\subset h^1/h^0(L_{\mathcal M}^\vee)$$, embeds it into the vector bundle stack $\mathfrak E=h^1/h^0((E^\bullet)^\vee)$, and defines the virtual fundamental class by the Gysin pullback

$$[\mathcal M]^{\mathrm{vir}}:=0^{!}_{\mathfrak E}\bigl(\mathfrak C_{\mathcal M}\bigr)\in A_{\mathrm{vd}(\mathcal M)}.$$

Its role is to supply the homological object that ordinary transversality would have produced, thereby making enumerative constructions possible on moduli spaces that are only virtually smooth (Jiang, 2022). Later developments show that the same principle can be realized in several mathematically distinct forms: as a Čech homology class, a chain-level map, a weighted branched Euler class, a Borel–Moore class on a derived stack, or a $K$-homology class on an orbifold inertia space (McDuff et al., 2015, Pardon, 2015, Khan, 2019, Tang, 2024).

1. Expected dimension and obstruction-theoretic meaning

The basic motivation for a virtual fundamental class is that deformation theory supplies an expected dimension even when the moduli space itself is badly behaved. In the Behrend–Fantechi setting, an obstruction theory on a Deligne–Mumford stack $\mathcal M$ is a morphism

$\phi:E^\bullet\longrightarrow L_{\mathcal M}$

with $E^\bullet\in D(\mathcal O_{\mathcal M})$, $h^i(E^\bullet)=0$ for $\phi:E^\bullet\to L_{\mathcal M}$0, $\phi:E^\bullet\to L_{\mathcal M}$1 and $\phi:E^\bullet\to L_{\mathcal M}$2 coherent, $\phi:E^\bullet\to L_{\mathcal M}$3 an isomorphism, and $\phi:E^\bullet\to L_{\mathcal M}$4 surjective. When $\phi:E^\bullet\to L_{\mathcal M}$5 is perfect of amplitude contained in $\phi:E^\bullet\to L_{\mathcal M}$6, it is a perfect obstruction theory, and its rank defines the virtual dimension (Jiang, 2022).

This obstruction theory controls infinitesimal deformation problems. For a square-zero extension $\phi:E^\bullet\to L_{\mathcal M}$7 with ideal $\phi:E^\bullet\to L_{\mathcal M}$8 and a map $\phi:E^\bullet\to L_{\mathcal M}$9, the obstruction to lifting $$\mathfrak C_{\mathcal M}\subset h^1/h^0(L_{\mathcal M}^\vee)$$0 to $$\mathfrak C_{\mathcal M}\subset h^1/h^0(L_{\mathcal M}^\vee)$$1 lies in

$$\mathfrak C_{\mathcal M}\subset h^1/h^0(L_{\mathcal M}^\vee)$$2

and when lifts exist they form a torsor under $$\mathfrak C_{\mathcal M}\subset h^1/h^0(L_{\mathcal M}^\vee)$$3 (Jiang, 2022). In that sense, the virtual class is not merely a substitute cycle; it is the homological shadow of a deformation–obstruction package.

A standard source of such obstruction theories is a universal family. If $$\mathfrak C_{\mathcal M}\subset h^1/h^0(L_{\mathcal M}^\vee)$$4 is projective, flat, and relatively Gorenstein, then

$$\mathfrak C_{\mathcal M}\subset h^1/h^0(L_{\mathcal M}^\vee)$$5

comes with a natural map to $$\mathfrak C_{\mathcal M}\subset h^1/h^0(L_{\mathcal M}^\vee)$$6, and when the amplitude is $$\mathfrak C_{\mathcal M}\subset h^1/h^0(L_{\mathcal M}^\vee)$$7 this yields a perfect obstruction theory and hence a virtual class (Jiang, 2022). This mechanism is central in higher-dimensional moduli problems, including KSBA surface moduli.

2. Principal realizations of the virtual class

The phrase “virtual fundamental class” names a role more than a single universal construction. Several frameworks realize that role with different ambient categories and different output objects.

Framework	Input datum	Output form
Behrend–Fantechi	Perfect obstruction theory $$\mathfrak C_{\mathcal M}\subset h^1/h^0(L_{\mathcal M}^\vee)$$8	$$\mathfrak C_{\mathcal M}\subset h^1/h^0(L_{\mathcal M}^\vee)$$9
Smooth Kuranishi atlas	Oriented additive weak Kuranishi atlas	VMC as a cobordism class of oriented compact $\mathfrak E=h^1/h^0((E^\bullet)^\vee)$0-manifolds; VFC in $\mathfrak E=h^1/h^0((E^\bullet)^\vee)$1
Implicit atlas	Oriented implicit atlas with cell-like stratification	Chain-level map $\mathfrak E=h^1/h^0((E^\bullet)^\vee)$2
Global Kuranishi chart	$\mathfrak E=h^1/h^0((E^\bullet)^\vee)$3	$\mathfrak E=h^1/h^0((E^\bullet)^\vee)$4-homology class $\mathfrak E=h^1/h^0((E^\bullet)^\vee)$5
Quasi-smooth derived Artin stack	Quasi-smooth derived morphism	$\mathfrak E=h^1/h^0((E^\bullet)^\vee)$6

In the smooth Kuranishi atlas approach, the virtual moduli cycle is a cobordism class of smooth, compact $\mathfrak E=h^1/h^0((E^\bullet)^\vee)$7-manifolds obtained as perturbed zero sets, while the virtual fundamental class is the corresponding Čech homology class

$\mathfrak E=h^1/h^0((E^\bullet)^\vee)$8

on the original topological space $\mathfrak E=h^1/h^0((E^\bullet)^\vee)$9 (McDuff et al., 2015). In the topological Kuranishi-atlas compression of McDuff–Wehrheim, the atlas data can be assembled into a single finite-dimensional branched model $$[\mathcal M]^{\mathrm{vir}}:=0^{!}_{\mathfrak E}\bigl(\mathfrak C_{\mathcal M}\bigr)\in A_{\mathrm{vd}(\mathcal M)}.$$0 with $$[\mathcal M]^{\mathrm{vir}}:=0^{!}_{\mathfrak E}\bigl(\mathfrak C_{\mathcal M}\bigr)\in A_{\mathrm{vd}(\mathcal M)}.$$1, and the virtual class is defined through the pairing

$$[\mathcal M]^{\mathrm{vir}}:=0^{!}_{\mathfrak E}\bigl(\mathfrak C_{\mathcal M}\bigr)\in A_{\mathrm{vd}(\mathcal M)}.$$2

for $$[\mathcal M]^{\mathrm{vir}}:=0^{!}_{\mathfrak E}\bigl(\mathfrak C_{\mathcal M}\bigr)\in A_{\mathrm{vd}(\mathcal M)}.$$3 (McDuff, 2017).

A common misconception is that a virtual fundamental class must always be a cycle in ordinary homology of the moduli space itself. Pardon's construction shows otherwise: in contact homology the virtual object is a canonical chain-level map

$$[\mathcal M]^{\mathrm{vir}}:=0^{!}_{\mathfrak E}\bigl(\mathfrak C_{\mathcal M}\bigr)\in A_{\mathrm{vd}(\mathcal M)}.$$4

and it is this chain-level form, rather than an ordinary homology class on $$[\mathcal M]^{\mathrm{vir}}:=0^{!}_{\mathfrak E}\bigl(\mathfrak C_{\mathcal M}\bigr)\in A_{\mathrm{vd}(\mathcal M)}.$$5, that carries the master equation needed for the algebra (Pardon, 2015). This suggests that “virtual fundamental class” is best understood as a family of formally compatible realizations of the same deformation-theoretic principle.

3. Kuranishi, implicit-atlas, and analytic constructions

In the Kuranishi-atlas framework of McDuff–Wehrheim, the central objects are finite-dimensional charts $$[\mathcal M]^{\mathrm{vir}}:=0^{!}_{\mathfrak E}\bigl(\mathfrak C_{\mathcal M}\bigr)\in A_{\mathrm{vd}(\mathcal M)}.$$6, coordinate changes, and a virtual neighborhood $$[\mathcal M]^{\mathrm{vir}}:=0^{!}_{\mathfrak E}\bigl(\mathfrak C_{\mathcal M}\bigr)\in A_{\mathrm{vd}(\mathcal M)}.$$7. After constructing tame shrinkings and reductions, one chooses admissible transverse perturbations $$[\mathcal M]^{\mathrm{vir}}:=0^{!}_{\mathfrak E}\bigl(\mathfrak C_{\mathcal M}\bigr)\in A_{\mathrm{vd}(\mathcal M)}.$$8, obtains perturbed zero sets

$$[\mathcal M]^{\mathrm{vir}}:=0^{!}_{\mathfrak E}\bigl(\mathfrak C_{\mathcal M}\bigr)\in A_{\mathrm{vd}(\mathcal M)}.$$9

and proves that $K$0 is a smooth compact $K$1-manifold. Its cobordism class is the virtual moduli cycle, and its limiting image in shrinking neighborhoods of $K$2 defines the virtual fundamental class in Čech homology (McDuff et al., 2015). The use of Čech homology is essential because of the continuity property

$K$3

for nested compact neighborhoods $K$4 (McDuff et al., 2015).

McDuff–Wehrheim later recast Kuranishi-atlas data with isotropy into a single finite-dimensional branched model $K$5. Here $K$6 is an oriented weighted branched topological manifold, $K$7 a finite-dimensional vector space, $K$8 a global isotropy group, and $K$9 an equivariant map with

$\mathcal M$0

The virtual class then appears as the Euler class of this branched section in Čech homology, bypassing part of the perturbation machinery and requiring only a topological submersion axiom rather than a fully smooth gluing theorem (McDuff, 2017).

Pardon's implicit-atlas approach is optimized for symplectic field theory type stratifications. In "Contact homology and virtual fundamental cycles" (Pardon, 2015), compactified moduli spaces $\mathcal M$1 are indexed by labeled trees $\mathcal M$2, equipped with topological implicit atlases and cell-like stratifications, and assembled into $\mathcal M$3-modules. The resulting virtual counts $\mathcal M$4 vanish outside virtual dimension zero, agree with actual signed counts when the moduli spaces are transverse, and satisfy the master equation

$\mathcal M$5

For contact homology this identity implies $\mathcal M$6; in the cobordism and family setups it gives chain maps, chain homotopies, and composition laws (Pardon, 2015). The paper is explicit that comparison with other virtual theories is expected but not proved there.

Parker’s exploded-manifold theory produces a virtual fundamental class $\mathcal M$7 for a Kuranishi category $\mathcal M$8 by means of weighted branched covers, a sheaf $\mathcal M$9 of perturbations, and weighted branched perturbation sections $\phi:E^\bullet\longrightarrow L_{\mathcal M}$0. The resulting virtual object supports integration of refined differential forms, pushforwards along evaluation maps, and compatibility with pullbacks, products, and tropical completion (Parker, 2015). This construction yields Gromov–Witten invariants of exploded manifolds and, in the compact symplectic case, an alternative construction of the standard invariants including gravitational descendants (Parker, 2015).

A further analytic variant appears for global Kuranishi charts. Given an almost complex global orbifold Kuranishi chart $\phi:E^\bullet\longrightarrow L_{\mathcal M}$1, Tang constructs a symbol on $\phi:E^\bullet\longrightarrow L_{\mathcal M}$2 and defines a $\phi:E^\bullet\longrightarrow L_{\mathcal M}$3-theoretic virtual fundamental cycle

$\phi:E^\bullet\longrightarrow L_{\mathcal M}$4

where $\phi:E^\bullet\longrightarrow L_{\mathcal M}$5 is the inertia orbifold. In the transverse case, if $\phi:E^\bullet\longrightarrow L_{\mathcal M}$6, then $\phi:E^\bullet\longrightarrow L_{\mathcal M}$7. The associated pairings reproduce exactly the invariants of Abouzaid–McLean–Smith (Tang, 2024).

4. Derived, shifted-symplectic, and motivic formulations

Derived algebraic geometry promotes virtual intersection theory from an auxiliary construction to an intrinsic feature of quasi-smooth geometry. In "Virtual fundamental classes of derived stacks I" (Khan, 2019), quasi-smooth derived Artin stacks acquire canonical fundamental classes in étale motivic Borel–Moore homology,

$\phi:E^\bullet\longrightarrow L_{\mathcal M}$8

and the virtual class on the classical truncation $\phi:E^\bullet\longrightarrow L_{\mathcal M}$9 is simply the image of this derived fundamental class. The theory proves functoriality, base change, excess intersection, and Grothendieck–Riemann–Roch, and yields a cohomological Bézout theorem

$E^\bullet\in D(\mathcal O_{\mathcal M})$0

without any transversality hypotheses (Khan, 2019).

The shifted-symplectic case behaves differently. Borisov–Joyce show that a separated $E^\bullet\in D(\mathcal O_{\mathcal M})$1-shifted symplectic derived $E^\bullet\in D(\mathcal O_{\mathcal M})$2-scheme $E^\bullet\in D(\mathcal O_{\mathcal M})$3 of complex virtual dimension $E^\bullet\in D(\mathcal O_{\mathcal M})$4 can be converted into a derived smooth manifold $E^\bullet\in D(\mathcal O_{\mathcal M})$5 of real virtual dimension $E^\bullet\in D(\mathcal O_{\mathcal M})$6. If $E^\bullet\in D(\mathcal O_{\mathcal M})$7 is proper and oriented, then $E^\bullet\in D(\mathcal O_{\mathcal M})$8 carries a virtual class

$E^\bullet\in D(\mathcal O_{\mathcal M})$9

and the construction is canonical up to oriented bordism fixing the underlying topological space (Borisov et al., 2015). The paper emphasizes two striking features: the virtual class has half the expected dimension, and purely complex algebraic input can yield a virtual class of odd real dimension (Borisov et al., 2015). A common misconception is therefore that shifted-symplectic virtual classes should have the same dimension pattern as ordinary perfect-obstruction-theory classes; the $h^i(E^\bullet)=0$0-shifted symplectic case is explicitly different.

Functoriality in the $h^i(E^\bullet)=0$1-shifted symplectic setting is reinterpreted by Schürg in terms of Lagrangian correspondences. The paper argues that Lagrangian correspondences are the correct framework for pullbacks of virtual fundamental classes arising from $h^i(E^\bullet)=0$2-symplectic derived structures, and proves that when a quasi-smooth Lagrangian correspondence exists from $h^i(E^\bullet)=0$3 to $h^i(E^\bullet)=0$4, pullback of the virtual class of $h^i(E^\bullet)=0$5 recovers that of $h^i(E^\bullet)=0$6 (Schürg, 2024). This suggests a shifted-symplectic analogue of the role played by quasi-smooth morphisms in ordinary derived intersection theory.

5. Functoriality, localization, comparison, and vanishing

Functoriality is a central constraint on any virtual class theory. In the derived Artin-stack framework, base change and composition are built into the fundamental class formalism, and excess intersection appears through an excess bundle whose top Chern class corrects pullback (Khan, 2019). In more classical moduli problems, comparison results are often phrased through normalization or pushforward. For instance, Kim’s comparison of stable ramified maps and log stable ramified maps uses virtual normalization and Costello’s push-forward formula to identify their virtual fundamental classes (Martín, 2011).

Localization theory substantially enlarges the computational utility of virtual classes. Kiem proves that semi-perfect obstruction theories, which require only local perfect obstruction theories with compatible obstruction sheaves and obstruction assignments, still admit a global virtual class

$h^i(E^\bullet)=0$7

and that torus localization, cosection localization, and combined torus–cosection localization remain valid in this setting (Kiem, 2016). In particular, a single global perfect obstruction theory is not necessary for localization formulas; semi-perfect obstruction theories suffice (Kiem, 2016). The same paper applies these methods to show that the Jiang–Thomas theory of virtual signed Euler characteristic works without the technical quasi-smoothness assumption from derived algebraic geometry (Kiem, 2016).

Cosection localization also has a differential-geometric counterpart. Kiem–Savvas combine Joyce’s d-manifold theory with Kiem–Li’s cosection localization to establish vanishing results for virtual fundamental classes of d-manifolds. If a compact oriented d-manifold carries a surjective weak real cosection of its obstruction sheaf, then its virtual class vanishes (Savvas, 2020). As an application, the stable pair invariants of hyperkähler fourfolds defined by Cao–Maulik–Toda are zero (Savvas, 2020). This gives a precise mechanism by which additional geometric symmetry forces virtual classes, and hence enumerative invariants, to disappear.

Another common misconception is that all virtual theories are interchangeable without loss of structure. The literature surveyed here points in a subtler direction. Some constructions emphasize chain-level coherence and boundary identities, some emphasize homological functoriality, some encode orbifold isotropy in inertia-space $h^i(E^\bullet)=0$8-homology, and some are optimized for shifted-symplectic or derived settings. This suggests that “equivalence of virtual counts” and “equivalence of virtual objects” are distinct comparison problems.

6. Applications across enumerative and Floer-type theories

The clearest applications occur where transversality is structurally obstructed. In contact homology, the differential counts punctured pseudoholomorphic curves in a symplectization. Because the relevant moduli spaces are typically singular and nontransverse, Pardon constructs coherent virtual counts on tree-indexed compactified moduli spaces and uses the master equation to define contact homology and cobordism maps rigorously (Pardon, 2015).

In Gromov–Witten theory, Shoemaker explains the usual Behrend–Fantechi virtual class and then presents a new construction for hypersurfaces via two-periodic complexes. For a smooth hypersurface $h^i(E^\bullet)=0$9, the moduli of maps with $\phi:E^\bullet\to L_{\mathcal M}$00-fields carries a two-periodic Koszul complex whose localized Chern character defines a virtual class on $\phi:E^\bullet\to L_{\mathcal M}$01; comparison results show that this agrees, up to the known sign, with the standard virtual class (Shoemaker, 2018). This connects virtual classes for hypersurfaces to GLSM and Landau–Ginzburg techniques.

In higher-dimensional moduli, Jiang constructs a perfect obstruction theory on the moduli stack $\phi:E^\bullet\to L_{\mathcal M}$02 of lci covering stacks over KSBA stable surfaces and defines a virtual fundamental class on the KSBA moduli stack by finite pushforward: $\phi:E^\bullet\to L_{\mathcal M}$03 This proves the existence of a virtual fundamental class on broad classes of moduli of surfaces of general type and, in Donaldson’s sextic example, produces a class $\phi:E^\bullet\to L_{\mathcal M}$04 (Jiang, 2022). The same paper defines tautological invariants by integrating powers of the first Chern class of the CM line bundle over the virtual class (Jiang, 2022).

For Calabi–Yau fourfolds, Borisov–Joyce propose using virtual classes of $\phi:E^\bullet\to L_{\mathcal M}$05-shifted symplectic derived schemes to define Donaldson–Thomas style invariants counting stable coherent sheaves. The resulting virtual classes are deformation invariant in families and are explicitly intended as the CY4 analogue of DT invariants (Borisov et al., 2015). In parallel, Tang’s $\phi:E^\bullet\to L_{\mathcal M}$06-homological virtual cycle supplies an analytic foundation for $\phi:E^\bullet\to L_{\mathcal M}$07-theoretic Gromov–Witten-type invariants when global Kuranishi charts are available (Tang, 2024).

Taken together, these developments show that the virtual fundamental class has become a unifying device across algebraic geometry, symplectic geometry, derived geometry, and $\phi:E^\bullet\to L_{\mathcal M}$08-theory. The unifying content is stable: one replaces a missing ordinary fundamental class by a canonically determined virtual object of the expected dimension. The realizations, however, are framework-dependent, and the choice of realization is often dictated by the algebraic identities, functorial properties, and invariants one needs to extract.