Adiabatic Calibrated Submanifolds

Updated 4 July 2026

Adiabatic calibrated submanifolds are limiting calibrated objects obtained by singular rescaling, which reduces higher-dimensional calibrated structures into anisotropic or maximal submanifold variational problems.
In G₂-geometry, the adiabatic limit decouples torsion-free equations into Fueter-type and maximal submanifold conditions in an indefinite signature space, offering fresh insights on associative calibrations.
This framework extends to singular fibrations and calibrated fibration PDEs, linking orbifold bundle techniques with variational methods and bridging complex submersion theories in special holonomy.

Adiabatic calibrated submanifolds are limiting calibrated objects obtained by singularly rescaling a geometric structure so that some directions collapse and the original calibrated equations decouple into a lower-dimensional variational problem. In the co-associative $G_2$ setting, the collapsing limit of K3-fibers reduces the torsion-free $G_2$-equations to the condition that a class-valued map $h:B\to H^2(X;\mathbb R)$ be a spacelike maximal immersion in a space of signature $(3,19)$ (Donaldson, 2016). In a more general formulation, a calibration $\alpha$ together with an $\alpha$-calibrated distribution $H$ determines a one-parameter family $\alpha_\varepsilon$ whose limit $\alpha_2$ is a generalized anisotropic calibration; the corresponding adiabatic calibrated submanifolds are anisotropic minimal, and in $G_2$-geometry the limiting equation is Fueter-type (Kawai et al., 20 May 2026).

1. General mechanism of adiabatic calibration

Let $G_2$0 be a Riemannian manifold, let $G_2$1 be a semi-calibration of degree $G_2$2, and let $G_2$3 be a $G_2$4-dimensional $G_2$5-calibrated distribution. Writing $G_2$6, one decomposes $G_2$7 by types relative to $G_2$8 as

$G_2$9

Because $h:B\to H^2(X;\mathbb R)$0 is $h:B\to H^2(X;\mathbb R)$1-calibrated, $h:B\to H^2(X;\mathbb R)$2 is the $h:B\to H^2(X;\mathbb R)$3-volume form, and the first cousin principle gives $h:B\to H^2(X;\mathbb R)$4. The adiabatic rescaling is

$h:B\to H^2(X;\mathbb R)$5

For every $h:B\to H^2(X;\mathbb R)$6, $h:B\to H^2(X;\mathbb R)$7 is a semi-calibration with respect to $h:B\to H^2(X;\mathbb R)$8. The $h:B\to H^2(X;\mathbb R)$9 limit isolates the $(3,19)$0-component, which becomes the “secondary” calibration in the adiabatic theory (Kawai et al., 20 May 2026).

The limiting calibration is anisotropic rather than ordinary. For $(3,19)$1, the first vertical energy density is

$(3,19)$2

and the limiting inequality is

$(3,19)$3

Equivalently, $(3,19)$4 is an $(3,19)$5-anisotropic semi-calibration on $(3,19)$6 with

$(3,19)$7

If $(3,19)$8 satisfies the equality property with an $(3,19)$9-valued form $\alpha$0, then equality in the adiabatic limit is characterized by $\alpha$1. Under the closedness hypothesis $\alpha$2, compact $\alpha$3-calibrated immersions minimize the vertical energy

$\alpha$4

while if $\alpha$5 and $\alpha$6, they minimize $\alpha$7 (Kawai et al., 20 May 2026).

In the broader sense used for collapsing $\alpha$8-fibrations, an adiabatic calibrated submanifold is a calibrated object whose defining equations decouple under a singular scaling that collapses selected directions, leaving a lower-dimensional calibrated variational problem. For co-associative fibrations, the fiberwise geometry becomes hyperkähler and the remaining torsion-free condition reduces to a maximal submanifold equation in an indefinite target; in product and torus settings this limit becomes exact, while in curved settings it provides the leading-order partial differential equation and formal asymptotic expansions (Donaldson, 2016).

2. Co-associative fibrations and the decoupled $\alpha$9 system

A $\alpha$0-structure on an oriented $\alpha$1-manifold $\alpha$2 is determined by a positive $\alpha$3-form $\alpha$4, which induces a Riemannian metric $\alpha$5 and Hodge dual $\alpha$6. The structure is torsion-free when

$\alpha$7

If $\alpha$8 is torsion-free, $\alpha$9 is a calibration, and a $H$0-dimensional submanifold $H$1 is co-associative exactly when

$H$2

The calibration inequality gives

$H$3

with equality if and only if $H$4 is co-associative. McLean’s deformation theory identifies the normal bundle $H$5 with $H$6 by

$H$7

so infinitesimal deformations correspond to harmonic self-dual $H$8-forms and the local moduli space is smooth of dimension $H$9 (Donaldson, 2016).

For a smooth co-associative fibration $\alpha_\varepsilon$0 with K3 fiber $\alpha_\varepsilon$1, a closed $\alpha_\varepsilon$2-structure with co-associative fibers has the algebraic form

$\alpha_\varepsilon$3

relative to a connection $\alpha_\varepsilon$4 splitting $\alpha_\varepsilon$5. Here $\alpha_\varepsilon$6 is hypersymplectic and $\alpha_\varepsilon$7 is positive in $\alpha_\varepsilon$8. Writing local coordinates $\alpha_\varepsilon$9 on $\alpha_2$0,

$\alpha_2$1

where $\alpha_2$2 records the horizontal derivative of the fibration. The exterior derivative splits as

$\alpha_2$3

For torsion-free $\alpha_2$4-structures one has $\alpha_2$5, where

$\alpha_2$6

with $\alpha_2$7. The torsion-free system is

$\alpha_2$8

together with

$\alpha_2$9

Geometrically, $G_2$0 means that the connection preserves the fiber volume determined by $G_2$1, or equivalently that the fibers are minimal in $G_2$2 (Donaldson, 2016).

The adiabatic scaling introduces $G_2$3 by

$G_2$4

In the corresponding metric, the fiber volume scales like $G_2$5. The rescaled torsion-free system becomes

$G_2$6

$G_2$7

Setting $G_2$8 formally decouples the curvature terms and yields

$G_2$9

In this limit, $G_2$00 is a hyperkähler element, $G_2$01 is pulled back from a positive $G_2$02-form on $G_2$03, and, after normalizing the $G_2$04-volume of fibers to $G_2$05, there is a unique volume-preserving connection $G_2$06 with $G_2$07 and $G_2$08 (Donaldson, 2016).

3. Maximal submanifolds in signature $G_2$09

The decisive reduction identifies the adiabatic limit with maximal submanifold geometry in the K3 cohomology lattice. Equip $G_2$10 with the cup-product metric

$G_2$11

which has signature $G_2$12. A smooth map $G_2$13 is positive if $G_2$14 has image a maximal positive subspace at each point, equivalently if it is a spacelike immersion. Its volume functional is

$G_2$15

The Euler–Lagrange equation is vanishing mean curvature in the ambient pseudo-Riemannian space: $G_2$16 Thus the adiabatic limit selects maximal, rather than minimal, submanifolds because the ambient metric is indefinite and the image of $G_2$17 must be spacelike (Donaldson, 2016).

The central identity is

$G_2$18

where the mean curvature vector $G_2$19 is identified fiberwise with the anti-self-dual $G_2$20-form part via the fiber hyperkähler metric. Hence

$G_2$21

In linear coordinates $G_2$22 on $G_2$23 with constant metric $G_2$24 of signature $G_2$25, the maximal equation is

$G_2$26

An equivalent formula uses the classes $G_2$27 and $G_2$28: $G_2$29 which matches the mean curvature term (Donaldson, 2016).

This lower-dimensional equation controls the reconstruction of the higher-dimensional torsion-free geometry. Given a decoupled solution $G_2$30, there are formal power series

$G_2$31

solving the first five equations order by order, and if $G_2$32 is maximal there are corrections

$G_2$33

so that the full system is solved formally. The proof uses a fiberwise exactness statement built from the Dirac operator coupled to $G_2$34 on K3 and a right inverse for the Jacobi operator of the maximal submanifold equation (Donaldson, 2016).

The maximal reduction also has geometric consequences. If $G_2$35 is a $G_2$36-dimensional maximal submanifold, then $G_2$37; consequently, if the orbifold bundle $G_2$38 admits a maximal positive section, the induced metric on $G_2$39 has nonnegative Ricci curvature. If $G_2$40 admits a positive section and $G_2$41, then any nontrivial locally constant section $G_2$42 of the flat orbifold bundle satisfies $G_2$43 pointwise. The paper also records analogues for special Lagrangian and Cayley geometries, including a semi-flat torus case in which the maximal equation becomes a Monge–Ampère equation and the adiabatic limit is exact (Donaldson, 2016).

4. Singular fibrations, orbifold bundles, and branched maximal sections

The smooth theory extends to singular fibrations through the Kovalev–Lefschetz framework. A differentiable KL fibration consists of a smooth map $G_2$44 between compact oriented manifolds, a link $G_2$45 of critical values, and a link $G_2$46 mapping diffeomorphically to $G_2$47, such that away from $G_2$48 the map is a submersion with K3 fibers and near $G_2$49 it is modeled on

$G_2$50

with $G_2$51 vanishing to second order at $G_2$52. Monodromy around each component of $G_2$53 is reflection in a vanishing cycle $G_2$54 with $G_2$55. This data is packaged in a flat affine orbifold bundle $G_2$56 whose linear part is the local system $G_2$57 and whose orbifold involutions are reflections in $G_2$58-classes (Donaldson, 2016).

A closed $G_2$59-form $G_2$60 vanishing on the fibers determines a class $G_2$61, where $G_2$62 is the sheaf of locally constant sections of the flat orbifold bundle. Locally, $G_2$63 induces a section $G_2$64 of $G_2$65, and near the singular link the appropriate positivity notion is branched positivity. In local orbifold coordinates $G_2$66 with $G_2$67, an equivariant local representative $G_2$68 is branched positive when $G_2$69 vanishes on the $G_2$70-direction, the restriction of $G_2$71 to $G_2$72 has the form $G_2$73, and $G_2$74 span a maximal positive subspace in $G_2$75 up to the ambiguity along the vanishing-cycle direction. A closed positive $G_2$76-form making $G_2$77 a co-associative KL fibration determines such a branched positive section (Donaldson, 2016).

The global adiabatic existence conjecture is formulated for the large-base cohomology class

$G_2$78

If $G_2$79 admits a positive section $G_2$80 that avoids excess $G_2$81-classes, then for $G_2$82 sufficiently large there should exist a closed positive $G_2$83-form $G_2$84 in $G_2$85 making $G_2$86 co-associative. If, in addition, $G_2$87 is maximal, then for $G_2$88 sufficiently large there should exist a torsion-free $G_2$89-structure $G_2$90 in the same cohomology class. In this framework, the adiabatic analogues of the $G_2$91-dimensional volume and Bryant’s Laplacian flow are

$G_2$92

Open analytic directions explicitly proposed include existence and compactness theory for branched maximal sections in flat affine orbifold bundles, understanding or removing the “avoid excess $G_2$93-classes” hypothesis, extending beyond nodal singularities, and connecting maximal-section data to constructions of compact torsion-free $G_2$94-manifolds (Donaldson, 2016).

5. Secondary calibrations, Fueter immersions, and anisotropic minimality

In $G_2$95-geometry, the general anisotropic formalism acquires a concrete first-order equation. Let $G_2$96 be a $G_2$97-structure and let $G_2$98 be a $G_2$99-dimensional $h:B\to H^2(X;\mathbb R)$00-calibrated distribution, so $h:B\to H^2(X;\mathbb R)$01 is associative at every point. Relative to $h:B\to H^2(X;\mathbb R)$02, one has

$h:B\to H^2(X;\mathbb R)$03

with $h:B\to H^2(X;\mathbb R)$04, $h:B\to H^2(X;\mathbb R)$05, $h:B\to H^2(X;\mathbb R)$06, and $h:B\to H^2(X;\mathbb R)$07. The equality-property tensor $h:B\to H^2(X;\mathbb R)$08 decomposes as $h:B\to H^2(X;\mathbb R)$09, and for $h:B\to H^2(X;\mathbb R)$10 with $h:B\to H^2(X;\mathbb R)$11-orthonormal basis $h:B\to H^2(X;\mathbb R)$12, the Fueter operator is

$h:B\to H^2(X;\mathbb R)$13

The $h:B\to H^2(X;\mathbb R)$14-type component satisfies

$h:B\to H^2(X;\mathbb R)$15

so the adiabatic calibration condition is

$h:B\to H^2(X;\mathbb R)$16

Equivalently, if $h:B\to H^2(X;\mathbb R)$17 is the $h:B\to H^2(X;\mathbb R)$18-form representing the graph of $h:B\to H^2(X;\mathbb R)$19, then

$h:B\to H^2(X;\mathbb R)$20

or equivalently $h:B\to H^2(X;\mathbb R)$21. A positive horizontally projectable immersion is Fueter when its tangent planes satisfy this condition, and then it minimizes $h:B\to H^2(X;\mathbb R)$22 if $h:B\to H^2(X;\mathbb R)$23, or $h:B\to H^2(X;\mathbb R)$24 if $h:B\to H^2(X;\mathbb R)$25 and $h:B\to H^2(X;\mathbb R)$26 (Kawai et al., 20 May 2026).

The adiabatic origin of the Fueter equation is explicit. For the rescaled $h:B\to H^2(X;\mathbb R)$27-structure

$h:B\to H^2(X;\mathbb R)$28

the associative equation is equivalent to $h:B\to H^2(X;\mathbb R)$29. Dividing formally by $h:B\to H^2(X;\mathbb R)$30 and letting $h:B\to H^2(X;\mathbb R)$31 yields $h:B\to H^2(X;\mathbb R)$32, namely the Fueter condition. This supplies a direct adiabatic passage from associative calibrations to anisotropic minimizers (Kawai et al., 20 May 2026).

Several explicit models are provided. In the product-type hyperkähler model $h:B\to H^2(X;\mathbb R)$33, with

$h:B\to H^2(X;\mathbb R)$34

a section $h:B\to H^2(X;\mathbb R)$35 is Fueter if

$h:B\to H^2(X;\mathbb R)$36

and $h:B\to H^2(X;\mathbb R)$37. On the semidirect product $h:B\to H^2(X;\mathbb R)$38, a section is Fueter if

$h:B\to H^2(X;\mathbb R)$39

with

$h:B\to H^2(X;\mathbb R)$40

Generalized quaternionic Heisenberg nilmanifolds give further examples with $h:B\to H^2(X;\mathbb R)$41, even though $h:B\to H^2(X;\mathbb R)$42 is nonintegrable. Local analytic existence of Fueter submanifolds is proved by Cartan–Kähler: any Fueter plane with a regular flag admits a local analytic Fueter submanifold tangent to it, and any analytic $h:B\to H^2(X;\mathbb R)$43-dimensional submanifold tangent to the corresponding $h:B\to H^2(X;\mathbb R)$44-plane extends uniquely (Kawai et al., 20 May 2026).

The same paper identifies a mirror-symmetric interpretation. Under the real Fourier–Mukai transform for $h:B\to H^2(X;\mathbb R)$45 torus fibrations, graphical submanifolds correspond to unitary connections, and Fueter immersions are equivalent to the $h:B\to H^2(X;\mathbb R)$46-instanton equation

$h:B\to H^2(X;\mathbb R)$47

equivalently $h:B\to H^2(X;\mathbb R)$48. Associative submanifolds correspond to deformed Donaldson–Thomas connections

$h:B\to H^2(X;\mathbb R)$49

and the large-radius limit reduces this to the $h:B\to H^2(X;\mathbb R)$50-instanton equation. The paper therefore states the general picture as: adiabatic limits correspond to large radius limits, $h:B\to H^2(X;\mathbb R)$51-calibrated associative submanifolds correspond to deformed Donaldson–Thomas connections, and adiabatic calibrated submanifolds correspond to $h:B\to H^2(X;\mathbb R)$52-instantons (Kawai et al., 20 May 2026).

6. Associative gradient cycles in adiabatic $h:B\to H^2(X;\mathbb R)$53 collapse

A distinct adiabatic model arises for associative $h:B\to H^2(X;\mathbb R)$54-folds in $h:B\to H^2(X;\mathbb R)$55-manifolds with K3 fibrations. For a topological Kovalev–Lefschetz fibration $h:B\to H^2(X;\mathbb R)$56, one obtains a flat orbifold bundle $h:B\to H^2(X;\mathbb R)$57 over $h:B\to H^2(X;\mathbb R)$58 with fiber $h:B\to H^2(X;\mathbb R)$59 and affine extension $h:B\to H^2(X;\mathbb R)$60. Given a branched maximal positive section $h:B\to H^2(X;\mathbb R)$61 of $h:B\to H^2(X;\mathbb R)$62, the local adiabatic $h:B\to H^2(X;\mathbb R)$63-form over a chart $h:B\to H^2(X;\mathbb R)$64 is

$h:B\to H^2(X;\mathbb R)$65

and

$h:B\to H^2(X;\mathbb R)$66

Hence

$h:B\to H^2(X;\mathbb R)$67

as $h:B\to H^2(X;\mathbb R)$68. The limiting $h:B\to H^2(X;\mathbb R)$69-associative condition is defined by

$h:B\to H^2(X;\mathbb R)$70

Although $h:B\to H^2(X;\mathbb R)$71 is not a $h:B\to H^2(X;\mathbb R)$72 $h:B\to H^2(X;\mathbb R)$73-form, it governs the adiabatic associative model (Donaldson et al., 2020).

For a $h:B\to H^2(X;\mathbb R)$74-class $h:B\to H^2(X;\mathbb R)$75, define

$h:B\to H^2(X;\mathbb R)$76

If $h:B\to H^2(X;\mathbb R)$77 is a gradient flow line

$h:B\to H^2(X;\mathbb R)$78

and the pair $h:B\to H^2(X;\mathbb R)$79 is irreducible along $h:B\to H^2(X;\mathbb R)$80, then the unique $h:B\to H^2(X;\mathbb R)$81-holomorphic $h:B\to H^2(X;\mathbb R)$82-sphere $h:B\to H^2(X;\mathbb R)$83 in class $h:B\to H^2(X;\mathbb R)$84 varies smoothly, and

$h:B\to H^2(X;\mathbb R)$85

is $h:B\to H^2(X;\mathbb R)$86-associative. This produces an adiabatic associative $h:B\to H^2(X;\mathbb R)$87-fold over a single gradient edge (Donaldson et al., 2020).

The base projection of an associative is modeled by a gradient cycle. These are either gradient orbits—embedded circles carrying a constant $h:B\to H^2(X;\mathbb R)$88-section $h:B\to H^2(X;\mathbb R)$89—or gradient graphs with edges labeled by constant $h:B\to H^2(X;\mathbb R)$90-sections $h:B\to H^2(X;\mathbb R)$91, subject at trivalent vertices to the balancing law

$h:B\to H^2(X;\mathbb R)$92

Along edges one follows the vector fields $h:B\to H^2(X;\mathbb R)$93; at endpoints on the singular link the label is the vanishing cycle. The model has a calibrated inequality on weighted length: $h:B\to H^2(X;\mathbb R)$94 with equality if and only if $h:B\to H^2(X;\mathbb R)$95 is a gradient cycle. This serves as the adiabatic analogue of calibrated volume minimization for associatives (Donaldson et al., 2020).

The same framework reproduces singularity formation phenomena. Crossing of two gradient graphs can create a new graph, matching the connected-sum “Lawlor neck” picture. Degeneration of a labeled edge into a reducible $h:B\to H^2(X;\mathbb R)$96-configuration leads to a surgery triple analogous to the Harvey–Lawson special Lagrangian surgery. Near the singular link there are terminating manifolds for the branched gradient flow, endpoint crossing can glue two terminating edges into one, and passage through the singular link changes labels by monodromy while preserving the topological count of cycles. The principal open problems are the existence of compact adiabatic torsion-free $h:B\to H^2(X;\mathbb R)$97 metrics realizing KL collapse and the analytic gluing from $h:B\to H^2(X;\mathbb R)$98-associative topological models $h:B\to H^2(X;\mathbb R)$99 to genuine associative $(3,19)$00-folds for $(3,19)$01 (Donaldson et al., 2020).

Two further lines of work place adiabatic calibrated submanifolds in a broader PDE and fibration context. One introduces “Smith immersions” and “Smith submersions” for arbitrary calibrations. For a calibration $(3,19)$02, a map $(3,19)$03 is a Smith immersion when

$(3,19)$04

so the image is $(3,19)$05-calibrated away from the critical set. For a calibration $(3,19)$06, a surjection $(3,19)$07 is a Smith submersion when

$(3,19)$08

so the smooth fibers are $(3,19)$09-calibrated. If $(3,19)$10, both classes are local minimizers of $(3,19)$11-energy and are $(3,19)$12-harmonic. The submersion equations are invariant under horizontally conformal rescalings

$(3,19)$13

which the paper explicitly identifies as a natural setting for adiabatic modeling of calibrated fibrations in SYZ- and GYZ-type regimes. Explicit noncompact examples are given on Bryant–Salamon $(3,19)$14 and $(3,19)$15 manifolds, where the projection maps have coassociative or Cayley fibers (Iliashenko et al., 2023).

A complementary submersion-based framework studies convexity, pluri-subharmonicity, and calibrations under a Riemannian submersion $(3,19)$16. If $(3,19)$17, then for horizontal $(3,19)$18,

$(3,19)$19

Under fiber hypotheses such as $(3,19)$20, $(3,19)$21, minimality, or critical-locus assumptions, convexity and subharmonicity transfer between total space and base. In the Kähler case with Lagrangian fibers, plurisubharmonicity is equivalent to convexity along the fibers; in the $(3,19)$22 case with coassociative fibers and $(3,19)$23, $(3,19)$24-plurisubharmonicity is likewise equivalent to convexity. The paper treats coassociative fibrations as Riemannian submersions with hyper-Kähler fibers and emphasizes that its identities naturally support an adiabatic picture in which vertical directions are collapsed while horizontal convexity and calibrated geometry remain dominant (Pacini, 2022).

A plausible implication is that adiabatic calibrated submanifold theory has two complementary forms. One form begins with a calibrated equation and passes to a singular limit, producing maximal-submanifold, Fueter, or gradient-cycle equations. The other begins with fibration or submersion PDEs whose calibrated fibers are stable under anisotropic or horizontally conformal rescaling. In both forms, the calibrated object is not merely reduced in dimension; it is reorganized so that the dominant limit equation is variationally natural, lower-dimensional, and often more rigid than the original special-holonomy system.