Integrable Invariant Distributions

Updated 20 January 2026

Integrable invariant distributions are subbundles of the tangent bundle that remain invariant under symmetry groups and satisfy Frobenius involutivity.
They facilitate the construction of invariant foliations and the classification of homogeneous spaces across differential geometry, dynamical systems, and PDE analysis.
Their applications span Lie theory, sub-Riemannian geometry, and bi-Hamiltonian systems, offering constructive integration algorithms and explicit algebraic classifications.

An integrable invariant distribution is a vector subbundle or, more generally, a (possibly singular) submodule within the tangent bundle of a manifold, preserved under a symmetry group or dynamical system, and satisfying Frobenius involutivity: the Lie bracket of any two local sections remains within the distribution. This structure appears across differential geometry, dynamical systems, Lie theory, sub-Riemannian geometry, and the analysis of partial differential equations, providing the machinery for constructing invariant foliations, classifying homogeneous spaces, and resolving integrability questions in both algebraic and analytic settings.

1. Foundational Definitions and Structural Criteria

Let $M$ be a smooth manifold and $D\subset TM$ a smooth distribution, i.e., a subbundle or subsheaf of the tangent sheaf. Given a symmetry group $G$ acting on $M$ (e.g., via diffeomorphisms or isometries), one says $D$ is $G$ -invariant if $g_*(D_x)=D_{g(x)}$ for all $g\in G$ , $x\in M$ . In the context of homogeneous spaces $M=G/H$ with isotropy group $H$ , $G$ -invariant distributions correspond to $\Ad(H)$-invariant subspaces $\mathfrak d\subset\mathfrak m$ in a reductive decomposition $\mathfrak g=\mathfrak h\oplus\mathfrak m$ .

The classical involutivity or integrability condition is

$[\Gamma(D),\Gamma(D)]\subset\Gamma(D)$

meaning that the space of smooth vector fields tangent to $D$ is closed under the Lie bracket. By Frobenius’ theorem, this is equivalent to decomposition of $M$ by a unique foliation whose tangent spaces are exactly $D_x$ at every $x$ .

In the homogeneous setting, $D$ is integrable if and only if $[\mathfrak d, \mathfrak d]\subset\mathfrak d \oplus \mathfrak h$ , i.e., the subalgebra $\mathfrak h\oplus\mathfrak d$ closes in $\mathfrak g$ (Berestovskii et al., 12 Jan 2026). For more general dynamical or bi-Hamiltonian contexts, the invariance (e.g., invariance under the flow of a family of vector fields or a Poisson pencil) is expressed as closure under symmetries or associated endomorphisms.

2. Integrability Theorems for Invariant Splittings and Dynamical Systems

For invariant splittings $TM=E\oplus F$ (e.g., dynamically defined distributions in smooth dynamics), a substantial body of work has established sharp criteria under which the higher-rank distribution (say, $E$ ) is integrable. On closed 3-manifolds, if $E\oplus F$ is a $C^1$ or Lipschitz continuous $Df$ -invariant splitting (for some $C^2$ diffeomorphism $f$ ), volume domination or dynamical domination coupled with a mild regularity hypothesis suffices:

Volume Domination: $\det(Df^n|_E)/\det(Df^n|_F)<C\lambda^n$ for some $C>0$ , $0<\lambda<1$ , all $n$ .
Dynamical Domination: $\|Df|_E\|< m(Df|_F)$ , where $m(A)=\min_{\|v\|=1}\|Av\|$ (1408.69481410.8072).

Under these, $E$ is uniquely integrable: each point admits a unique local embedded surface tangent to $E$ (1408.69481410.8072). Simić’s Lipschitz Frobenius theorem is pivotal: Lipschitz plane fields involutive at almost every point are integrable (Luzzatto et al., 2014). These integrability criteria have refined the classical pinching condition (“center-bunching”) and expanded the classes of examples beyond partially hyperbolic dynamics.

3. Algebraic and Homogeneous Geometric Classification

Homogeneous spaces $G/H$ with integrable invariant distributions admit a comprehensive algebraic classification:

$G/H$ has the property that every $G$ -invariant distribution is integrable if and only if every $\Ad(H)$-invariant subspace $\mathfrak q\supset\mathfrak h$ is a Lie subalgebra.
Such spaces include symmetric spaces, isotropy-irreducible spaces where only the trivial and total distributions are possible, certain semidirect products, and solvmanifolds with metabelian structure (Berestovskii et al., 12 Jan 2026).

In the compact or abelian radical cases, integrability implies geodesic orbit (GO) metric rigidity—every geodesic is an orbit of a 1-parameter subgroup. Noncompact solvmanifolds and metabelian examples exhibit integrable invariant distributions yet may fail the GO property, particularly when equipped with certain invariant Einstein metrics (Berestovskii et al., 12 Jan 2026).

4. Integrability in Bi-Hamiltonian and Poisson Geometry

Invariant distributions associated to non-degenerate bi-Hamiltonian structures are classified via the eigen-structure of the recursion operator $P=\omega_1\omega_0^{-1}$ , where $(\omega_0,\omega_1)$ are compatible symplectic (or Poisson) forms. Near regular points, the Turiel splitting decomposes $M$ into factors; invariant distributions correspond to block decompositions tied to Jordan–Kronecker invariants (Kozlov, 2022).

All such distributions are integrable except possibly for the kernels $\Ker(P-\lambda I)^{k_i}$ or their complex-quadratic analogues with block size $k_i>1$ .
The only obstruction is the nonconstancy of associated eigenfunctions: integrability holds if and only if $d\lambda=0$ (resp., $da=0=d\beta$ in conjugate-pair case).

This yields a local normal form for invariant foliations and recognizes cohomological obstructions only in the nonconstant eigenfunction classes (Kozlov, 2022).

5. Analytic, PDE, and Measure-Theoretic Aspects

Integrable invariant distributions appear fundamentally in analytic settings:

In the analysis of symmetric spaces and representation theory, invariant eigendistributions (e.g., solutions of $(\Omega-\lambda)T=0$ for Casimir $\Omega$ ) are locally integrable functions on nice symmetric pairs. Harinck and Jacquet give explicit bases for these distributions in higher-rank classical cases, with regularity and $L^1$ -integrability established via orbital integral analysis (1102.09631407.0934).
In stochastic analysis, the integrability of the drift in McKean–Vlasov SDEs allows the construction of invariant probability measures by Banach fixed point theorem, with the law's regularity controlled by entropy and Sobolev estimates; weak well-posedness reduces essentially to functional bounds on the drift (Huang et al., 2021).

In integrable spin chains and lattice NLS systems, invariant Gibbs and white-noise measures are constructed and shown invariant under the flows of the (discrete) completely integrable models; these measures correspond to unique invariant distributions in infinite-dimensional phase space (Angelopoulos et al., 2018).

6. Modern Perspectives: $\mathcal{C}^{\infty}$ -Symmetries and Integration Algorithms

Generalizing classical solvable symmetry approaches, the notion of $\mathcal{C}^{\infty}$ -structures provides a local stepwise decomposition of involutive distributions: an ordered sequence of $k=n-r$ generalized symmetries yields a tower of completely integrable Pfaffian equations. The solution of the corresponding system reconstructs the original foliation by integration of successively lower-codimension differential forms. This approach is constructive and extends to the integration of ODEs via splitting into integrable Pfaffians, substantially enlarging the rank of integrable cases beyond those accessible via solvable symmetry algebras (Pan-Collantes et al., 2022).

7. First Integrals and Rational Foliations

For integrable distributions $D$ on complex projective manifolds, the field of rational first integrals, $\mathcal{C}(D)$ , can be realized as the field of first integrals of a single rational vector field tangent to $D$ . In dimension 3, and more generally for integrable distributions, there always exists a rational vector field $X$ with $\mathcal{C}(X)=\mathcal{C}(D)$ , enabling explicit reduction of first-integral fields to the rank-one case (Luza et al., 2024).

These lines of research collectively establish integrable invariant distributions as a unifying concept cutting across geometry, dynamics, analysis, and algebraic structures, with sharp classification results, explicit analytic realizations, and modern geometric approaches to integration. The algebraic characterization on homogeneous spaces, functional-analytic techniques for existence and regularity in stochastic and PDE settings, algorithmic integration in differential equations, and the interplay with symmetry and invariance remain central to current and ongoing advances in the field.