Distribution (Differential Geometry)¶

A distribution on a smooth manifold assigns to each point a smoothly varying subspace of its tangent space, encoding admissible infinitesimal directions.

Version: v1 · 2026-09-28 · History
Domain-specific #: 7623
Domain group: Formal Sciences
Origin domain: Mathematics
Subdomain: Differential Geometry → Mathematics
Aliases: Tangent distribution, Plane field

Core Idea¶

In differential geometry, a distribution Δ on a smooth manifold (M) assigns to each point (x\in M) a vector subspace Δₓ of the tangent space (T_xM), with the assignment varying smoothly.^[1] It specifies the infinitesimal directions that are admissible, horizontal, constrained, or otherwise distinguished at every point.^[2]

For a regular rank-(k) distribution, all fibers have dimension (k) and form a vector subbundle Δ⊂(TM).^[3] Locally it can be generated by smooth vector fields (X_1,…,X_k) whose values span Δₓ.^[4] Singular or generalized distributions allow rank to vary, requiring a more careful smooth-generation condition.^[5]

The assignment becomes geometrically consequential through closure and reachability tests. An involutive distribution is closed under Lie brackets of its sections.^[6] For regular distributions, the Frobenius theorem connects involutivity with integrability: locally the admissible planes are tangent to leaves of a foliation.^[7] In contrast, a bracket-generating distribution expands under iterated Lie brackets until it spans the tangent bundle, a condition central to nonholonomic geometry and control.^[8]

The invariant is: each point carries a specified tangent subspace, those subspaces form a smooth local assignment, and admissible curves or vector fields are tangent to that assignment. Remove pointwise tangent subspaces, allow arbitrary unsmoothed choices, or replace tangent directions with a measure or generalized function, and the identity collapses.

How would you explain it like I'm…

Allowed Directions Everywhere

Think of an ice skate on a frozen pond. At every spot, the skate can only glide the way its blade points, forward or backward, not sideways. A distribution is like a rule that gives every spot its own set of allowed directions, and the rule changes smoothly from spot to spot.

Rules for Which Way to Go

In geometry, a smooth shape, like a surface, has at each point a set of all the directions you could move in. A distribution picks out, at every point, a smaller group of those directions: the 'allowed' or special ones. And it does this smoothly, so the choice changes gently from point to point. Think of an ice skate: at each spot it can glide forward and backward but not sideways. Mathematicians then ask interesting questions, like whether moving only in allowed directions keeps you stuck on a thinner sheet or whether wiggling back and forth can still get you anywhere.

Tangent-Subspace Field

In differential geometry, a distribution on a smooth manifold (a smooth curved space) picks out, at each point, a subspace of the tangent space, meaning a chosen set of allowed directions, and does so smoothly. If the number of allowed directions is the same everywhere, say k, it is a regular rank-k distribution, and near any point you can describe it with k smooth vector fields that span the allowed directions. Paths that always move in allowed directions are the admissible ones. Two opposite behaviors matter. If combining allowed motions never produces a new direction, the distribution is involutive, and the Frobenius theorem says the allowed directions are locally tangent to stacked layers (a foliation), so you stay on one layer. If combining allowed motions eventually produces every direction, it is bracket-generating, which is why a car that can't slide sideways can still parallel park. Note this is a different meaning of distribution from the one in probability.

A distribution on a smooth manifold M is a smooth assignment to each point x of a vector subspace of the tangent space T_xM, specifying the admissible, horizontal, or otherwise distinguished infinitesimal directions there. A regular rank-k distribution has all fibers of dimension k and forms a vector subbundle of TM; locally it is spanned by k smooth vector fields. Singular or generalized distributions allow the rank to vary and require a more careful smoothness condition. Its geometric consequences come from Lie-bracket behavior. An involutive distribution is closed under Lie brackets of its sections, and by the Frobenius theorem a regular involutive distribution is integrable: locally it is tangent to the leaves of a foliation. At the opposite extreme, a bracket-generating distribution's iterated brackets eventually span the whole tangent bundle, the condition central to nonholonomic geometry and control. The defining invariant is pointwise tangent subspaces, assigned smoothly, with admissible curves or vector fields tangent to them; it is not a measure or generalized function.

Structural Signature¶

Sig role-phrases:

smooth-manifold carrier — the base manifold M together with its tangent bundle TM.
pointwise tangent fiber — the vector subspace Δₓ ⊆ TₓM assigned at each point x.
smooth local generation — nearby fibers are spanned by a smoothly varying collection of local vector fields.
regular-rank branch — constant fiber dimension, making Δ a vector subbundle of TM.
singular-rank branch — variable fiber dimension retained through the appropriate smooth-generation condition.
distribution section — a smooth vector field whose value lies in the assigned subspace at every point.
admissible curve — a curve whose velocity remains in the distribution along its path.
Lie-bracket test — commutators of local sections determine whether admissible directions close or generate new ones.
involutive branch — brackets of sections remain sections of Δ.
integrable branch — the fibers are tangent to local integral manifolds that assemble into foliation leaves in the regular case.
bracket-generating branch — iterated commutators enlarge the permitted directions until they span the tangent bundle.
coordinate-invariance guarantee — the pointwise subspaces and their smooth assignment persist across changes of local presentation.
fiber-and-smoothness boundary — arbitrary pointwise choices or nontangent fibers do not form a differential-geometric distribution.
lexical boundary — probability laws and generalized functions called distributions have different carriers and operations.
foliation limitation — a distribution is not automatically a foliation; integrability must be established rather than inferred from rank alone.

What It Is Not¶

Not a probability distribution. The carrier is a smooth manifold and the assigned objects are tangent subspaces, not probabilities on measurable events.
Not a distribution in analysis. A generalized function acts on test functions; it does not assign a smoothly varying subspace of (T_xM) to each point.
Not automatically a foliation. A regular distribution yields local leaves only when the required integrability or involutivity condition holds; contact and other bracket-generating distributions are decisive counterexamples.
Not the whole tangent bundle. A distribution is a specified subspace (Delta_x\subseteq T_xM) at each point, although the full tangent bundle is a limiting case.
Not a vector field. A vector field can be a section or local generator of a distribution, but one chosen direction is not identical with the smoothly assigned family of subspaces it may help span.
Not a differential form or volume element. A form can present a hyperplane distribution through its kernel, but the form and the resulting tangent-plane assignment are different geometric objects.
Not an arbitrary pointwise choice of planes. The fibers must satisfy the relevant smooth local-generation condition; a discontinuous or unsmoothed selection does not become a distribution merely because every value is linear.
Not necessarily constant-rank. Regular distributions form vector subbundles, while singular or generalized distributions can vary in rank under a more careful smoothness convention.

Scope of Application¶

A differential-geometric distribution applies on a smooth manifold when each point receives a smoothly generated tangent subspace whose sections, admissible curves, Lie brackets, rank behavior, and integrability convention are specified; probability laws and generalized functions lie outside this literal scope.

Regular plane fields — constant-rank tangent subspaces form vector subbundles and provide the basic setting for local generators, sections, and coordinate-invariant comparison.
Foliation theory — an involutive regular distribution is tested for local integral manifolds whose tangent spaces recover the assigned fibers.
Differential topology — distributions organize smooth subbundles, their homotopy classes, global extension questions, and obstructions to prescribed tangent-plane fields.
Contact geometry — a contact form's kernel defines a regular hyperplane distribution whose maximal nonintegrability is read through its differential or bracket growth.
Engel geometry — rank-two distributions on four-manifolds are distinguished by a prescribed derived-flag growth rather than by integrability.
Sub-Riemannian geometry — a distribution specifies horizontal velocities, and an inner product on those fibers measures lengths of admissible curves.
Nonholonomic mechanics — velocity constraints are represented as tangent distributions whose bracket behavior distinguishes integrable restrictions from constrained systems with larger reachable sets.
Geometric control theory — control vector fields span admissible directions, while iterated Lie brackets and rank conditions diagnose local accessibility or controllability.
Poisson geometry — Hamiltonian vector fields generate tangent distributions whose integral manifolds are symplectic leaves, often with rank varying across strata.
Lie-group actions and orbit geometry — infinitesimal action fields span an orbit distribution whose fiber dimension changes where stabilizers change.
Integrable systems — commuting vector fields and conserved structures yield distinguished tangent directions whose involutivity supports invariant leaves.
Singular-distribution theory — variable-rank, smoothly generated modules cover orbit and leaf structures only under the appropriate singular integrability theorem, not a naïve constant-rank Frobenius test.

Clarity¶

A clear definition gives manifold dimension, distribution rank, local generators or defining forms, and the convention for regular versus singular. If integrability is claimed, the account states the bracket or flow criterion and identifies integral manifolds. If bracket-generating behavior is claimed, it states the derived flag and step or growth vector.

Notation must distinguish Δₓ, the fiber at (x), from Γ(Δ), its smooth sections, and from the union Δ⊂(TM). A “local basis” that loses linear independence at some point is more accurately a local generating set.

Manages Complexity¶

The distribution compresses many local differential constraints into a geometric subbundle. Instead of listing every admissible path, one specifies allowed velocities pointwise. Integral leaves, bracket flags, growth vectors, and curvature-like obstructions then summarize global and local consequences.

The compression can hide singular points, rank changes, and nontrivial topology. Local generators need not extend globally, and equal ranks do not guarantee equivalent bracket structures. Responsible use preserves these distinctions rather than treating a plane field as globally trivial.

Abstract Reasoning¶

Reasoning alternates between fiberwise linear algebra and differential closure. At each point, Δₓ is a linear subspace. Across points, smoothness couples those choices. Lie brackets test whether following admissible vector fields creates new infinitesimal directions. If brackets remain within Δ, integral leaves may exist; if they expand to (TM), paths tangent to Δ can nevertheless reach broadly.

Coordinate-free reasoning is central: local vector fields or kernels can change presentation while the subbundle remains the same. A diffeomorphism carries a distribution by its differential, providing the correct equivalence test for local models.

Knowledge Transfer¶

Within geometry, the assignment–bracket–integrability framework transfers from foliations to contact structures, nonholonomic mechanics, Poisson leaves, Lie group actions, and control systems. The carrier and operations remain literal tangent geometry even when the interpretation changes.

Beyond differential geometry, the defensible reach is (A) analogy: “allowed directions” can orient reasoning about constrained moves in other systems, but no literal distribution has transferred. What the analogy carries is the contrast between locally admissible motion and larger reach obtained by composing moves; the manifold, tangent bundle, smoothly varying subspaces, vector fields, and Lie brackets remain home-bound. Transfer stops unless the target genuinely supplies that differential-geometric carrier and operations; a set of permitted actions or a statistical distribution does not qualify merely because it uses the same word.

Examples¶

Canonical¶

The coordinate-plane distribution on ℝⁿ. At every point of M = ℝⁿ, let Δₓ be the span of the first k coordinate vector fields, ∂/∂x₁, …, ∂/∂xₖ. These fields vary smoothly and remain linearly independent, so the assigned fibers form a regular rank-k subbundle of Tℝⁿ. Their pairwise Lie brackets vanish, hence remain inside Δ.^[9] The integral manifold through a point is obtained by holding xₖ₊₁, …, xₙ fixed while varying the first k coordinates; its tangent space at every point is exactly the assigned fiber.

Mapped back: ℝⁿ with Tℝⁿ is the smooth-manifold carrier; the span of ∂/∂x₁, …, ∂/∂xₖ is the pointwise tangent fiber; and the same coordinate fields provide smooth local generation and fill the distribution section role. Constant dimension establishes the regular-rank branch. Vanishing commutators pass the Lie-bracket test, place the example in the involutive branch, and the fixed-coordinate leaves establish the integrable branch.

Applied / In Practice¶

Testing the standard contact distribution. On ℝ²ⁿ⁺¹, take the hyperplane assignment Δ = ker(ω) for the standard contact one-form ω = dz + Σ xᵢ dyᵢ. The kernel has constant rank 2n and varies smoothly, so it is a regular distribution. Unlike the coordinate-plane case, its sections do not close under Lie brackets: their brackets supply the direction missing from each hyperplane, giving growth from 2n to 2n + 1. The distribution is therefore bracket-generating and maximally nonintegrable rather than tangent to a foliation.^[10] This calculation is how contact geometry distinguishes the contact structure from an ordinary integrable hyperplane field.

Mapped back: ℝ²ⁿ⁺¹ and its tangent bundle supply the smooth-manifold carrier; ker(ω) is the pointwise tangent fiber; and its local sections supply smooth local generation. Constant rank places it in the regular-rank branch. The commutator calculation is the Lie-bracket test; the added missing direction establishes the bracket-generating branch. Failure of closure excludes the involutive branch, while the absence of tangent foliation leaves enforces the foliation limitation.

Structural Tensions¶

T1: Pointwise linearity versus global geometry. Each fiber is simple while bundles and leaves can be topologically nontrivial. Diagnostic: separate local spanning data from global trivialization.

T2: Involutivity versus bracket generation. Closure yields leaves; nonclosure can yield accessibility. Diagnostic: calculate Lie brackets rather than inferring behavior from rank.

T3: Constant rank versus singular usefulness. Regularity simplifies theory but excludes natural orbit and Poisson examples. Diagnostic: record rank strata and use the appropriate integrability theorem.

T4: Presentation versus invariant object. Kernels and generators are coordinate choices. Diagnostic: verify equality of pointwise subspaces and behavior under diffeomorphism.

T5: Terminological regularity versus mathematical precision. Authors use “regular” for different conditions. Diagnostic: state the exact rank or flag condition instead of relying on the adjective.

T6: Tangent-distribution autonomy versus reduction to Relation. Every qualifying differential-geometric distribution is a strict specialization of the exact parent Prime Relation (Relation): manifold points and admitted tangent vectors are typed relata, and the membership rule v ∈ Δ_x determines the point–vector association. Reduction preserves that first-class tuple relation, but loses tangent-subspace fibers, constant rank, smooth local generation, and Lie-bracket closure or growth behavior. Treating the distribution as wholly autonomous would hide its relation structure; a pointwise association without smoothness does not establish the named geometric object.
Diagnostic: Is there merely a typed point–vector relation, or does it assemble into the smooth tangent-subspace field whose rank and bracket behavior define the distribution?

Structural–Framed Character¶

Distribution in differential geometry is structural-leaning. Its stable form assigns each manifold point a tangent subspace under smooth local generation, with membership, bracket, and integrability properties governing the resulting object. The smallest portable skeleton is Relation, which preserves typed relata, a tuple-membership rule, bivalent inclusion, and association-level structure. That portable reach belongs to the Relation Prime; the distribution remains its smooth tangent-geometric specialization.

Its evaluative_weight is absent because membership, rank, bracket closure, and integrability do not encode desirability. Its human_practice_bound character is low: mathematicians choose notation and representation, but the defined consequences follow from the formal structure. Its institutional_origin is low because disciplinary convention standardizes the name without constituting the point–subspace relation. Its vocab_travels result is partial: relation, fiber, and membership language carries formally, whereas tangent bundle, smooth section, Lie bracket, involutivity, and foliation retain differential-geometric commitments. Under import_vs_recognize, Relation can be recognized wherever typed tuples satisfy a membership rule, but this distribution must be imported with its manifold carrier, tangent fibers, smoothness condition, and geometric closure tests.

Its character: structural-leaning because Relation owns the portable association skeleton while differential geometry supplies the smooth tangent-space constraints that make it a distribution.

Structural Core vs. Domain Accent¶

Distribution in differential geometry is a domain-specific geometric abstraction rather than a prime; it is a strict specialization of Relation. Its complete named signature is smooth-manifold carrier → pointwise tangent fiber → smooth local generation → sections and admissible curves → Lie-bracket test → involutive/integrable or bracket-generating branches, with rank, smoothness, lexical, and foliation boundaries.

What is skeletal (could lift toward a cross-domain prime). Relation owns typed relata, a rule that decides which tuples belong, and properties of the resulting association. That complete relational pattern survives in mathematical divisibility, database dependencies, and kinship relations—three unrelated domains—even though their occupants and properties differ. Removing the differential-geometric accent therefore leaves a genuine Relation: manifold points and admitted tangent vectors remain typed relata, and v ∈ Δₓ remains the membership rule for their association.

What is domain-bound. Smooth manifolds and tangent bundles, pointwise linear subspaces, local vector-field generators, regular and singular rank, admissible velocities, Lie brackets, involutivity, integral leaves, derived flags, and bracket generation constitute this distribution. They determine whether the relation assembles into a smooth tangent-subspace field and what geometric consequences follow; they are not requirements of Relation in general.

Why this does not clear the prime bar. The entry adds no second substrate-independent invariant beyond a typed association; its distinctive tests depend on differential-geometric carriers and operations. Remove the point–tangent-vector membership relation and no distribution remains, even if a manifold and vector fields are present. Remove the manifold, tangent fibers, smoothness, and bracket structure and the residual is Relation, not a differential-geometric distribution. The strict parent relation therefore preserves the portable core while the named abstraction remains irreducibly geometric.

This entry is a kind of Relation.

Instantiates — Relation (Relation). The relata are manifold points and tangent subspaces (equivalently, points and tangent vectors admitted by those subspaces); the arity and domains are explicit; and the rule “(v \in \Delta_x)” decides which point–vector pairs belong. The resulting binary association is a first-class object whose converse, restriction, composition with smooth maps, and properties such as constant rank and bracket closure support further reasoning, while smooth local generation adds the differential-geometric guarantee. Removing the manifold accent leaves Relation's typed relata, tuple-membership rule, bivalent inclusion, and association-level structure. Removing the pointwise association destroys the distribution even if the tangent bundle and vector fields remain. This full mapping establishes strict subsumption under Relation; the functional uniqueness of assigning one subspace per point is additional structure rather than a contradiction of the relational genus.

Relationships to Other Abstractions¶

Current abstraction Distribution (Differential Geometry) Domain-specific

Parents (1) — more general patterns this builds on

Distribution (Differential Geometry) is a kind of Relation Prime

The relata are manifold points and tangent subspaces (equivalently, points and tangent vectors admitted by those subspaces); the arity and domains are explicit; and the rule “(v \in \Delta_x)” decides which point–vector pairs belong.

Hierarchy path (1) — routes to 1 parentless root

Distribution (Differential Geometry) → Relation

Neighborhood in Abstraction Space¶

Distribution (Differential Geometry) sits in a moderately populated region (58^th percentile for distinctiveness): it has near-neighbors but no dense thicket of look-alikes.

Family — Differential Geometry & Curvature (9 abstractions)

Nearest neighbors

Bundle metric — 0.86
Geodesic — 0.85
Stratifold — 0.85
Chow–Rashevsky theorem — 0.84
Submersion (mathematics) — 0.84

Computed from structural-signature embeddings · 2026-10-08

Not to Be Confused With¶

Probability distribution. A probability distribution assigns probabilities to events or values, whereas a differential-geometric distribution assigns a tangent subspace to each point of a manifold. Tell: inspect whether the fibers are numbers summing or integrating to one or linear subspaces of (T_xM).
Distribution in analysis. An analytic distribution is a generalized function acting on test functions, not a smoothly varying family of tangent directions. Tell: determine whether the object is a continuous linear functional on tests or a pointwise subspace assignment.
Tangent bundle. The tangent bundle contains every tangent vector at every point; a distribution selects a specified subspace (Δ_x\subseteq T_xM) in each fiber. Tell: compare the rank of the assigned plane with the full manifold dimension.
Vector field. A vector field selects one tangent vector at each point and may generate or lie within a distribution, but it is not the entire subspace family. Tell: ask whether admissible directions include a span of local sections or only one chosen section.
Foliation. A foliation partitions a manifold into immersed leaves; a regular distribution produces such leaves only when it is involutive or integrable. Tell: compute Lie-bracket closure before replacing the tangent-plane assignment with a leaf structure.
Differential form. A differential form is a covariant field that can present a distribution through its kernel, but the form and the resulting tangent subspaces are distinct objects. Tell: distinguish the equation or covector field from the vectors satisfying its kernel condition.

References¶

[1] MIT OpenCourseWare, Topics in Geometry: Distributions and the Frobenius Theorem (accessed 2026-09-13). registry ↩

[2] Unverified encyclopedia synthesis; no authoritative source located for the claim as written. ↩

[3] Unverified encyclopedia synthesis; no authoritative source located for the claim as written. ↩

[4] Unverified encyclopedia synthesis; no authoritative source located for the claim as written. ↩

[5] Unverified encyclopedia synthesis; no authoritative source located for the claim as written. ↩

[6] Unverified encyclopedia synthesis; no authoritative source located for the claim as written. ↩

[7] Unverified encyclopedia synthesis; no authoritative source located for the claim as written. ↩

[8] Unverified encyclopedia synthesis; no authoritative source located for the claim as written. ↩

[9] Unverified encyclopedia synthesis; no authoritative source located for the claim as written. ↩

[10] Unverified encyclopedia synthesis; no authoritative source located for the claim as written. ↩