Cocurvature¶

The vector-valued two-form that detects nonintegrability of the image distribution of a tangent-bundle projection.

Version: v1 · 2026-09-28 · History
Domain-specific #: 8514
Domain group: Formal Sciences
Origin domain: Mathematics
Subdomain: Differential Geometry → Mathematics

Core Idea¶

Cocurvature is a precise obstruction attached to a tangent-bundle projection P. If P is idempotent, its image can be regarded as a vertical distribution and its kernel as a complementary horizontal distribution. Bracketing two projected vertical vector fields can generate a direction outside that image; (Id−P) extracts exactly that outside component. The resulting vector-valued two-form is R̄_P(X,Y)=(Id−P)[PX,PY].

Under the regular constant-rank conditions in Michor's treatment, vanishing cocurvature means the vertical distribution closes under Lie brackets and is integrable. It is distinct from the companion curvature, which projects brackets of horizontal components into the vertical side. A genuine fiber bundle has fibers whose tangent distribution is integrable, so its vertical cocurvature vanishes; this useful special case does not trivialize the invariant for a general chosen projection.

How would you explain it like I'm…

The Sneaky Sideways Scoot

Imagine a toy car that can only roll forward and backward and turn its wheels. If you wiggle it, rolling a bit and turning a bit, back and forth, it slowly scoots sideways, a way it could never roll directly. Cocurvature is a way mathematicians measure how much a set of allowed moves, mixed together like this, sneaks you out of the allowed directions.

Do Allowed Moves Stay Allowed?

Picture a toy car that can only roll forward or backward and turn its wheels; it cannot slide sideways. But by wiggling forward, turning, and backing up, it can end up shifted sideways, like parallel parking. Mathematicians choose a rule that sorts every direction into 'vertical' and 'horizontal'. Cocurvature measures whether combining vertical moves in this wiggling way can push you outside the vertical directions. If it is zero, the vertical directions stay closed up by themselves and fit together into neat layers.

Vertical Bracket Leakage

Cocurvature is a quantity in differential geometry attached to a projection P that, at each point, splits directions into a 'vertical' part (the image of P) and a complementary 'horizontal' part (its kernel). Take two vector fields that lie in the vertical part and compute their Lie bracket, which measures what happens when you alternate small moves along each. That bracket can point partly outside the vertical directions, and the cocurvature is exactly that outside part. When cocurvature is zero, the vertical directions are closed under brackets and, under regularity conditions, fit together into a family of surfaces (they are integrable). It is a companion to curvature, which asks the same question about the horizontal directions instead.

Cocurvature is an obstruction attached to an idempotent projection P on the tangent bundle of a manifold. The image of P is regarded as a vertical distribution and its kernel as a complementary horizontal distribution. For vector fields X and Y, project them to PX and PY, take their Lie bracket, and extract the component outside the vertical distribution with (Id - P); this gives a vector-valued two-form, the cocurvature R-bar_P(X, Y) = (Id - P)[PX, PY]. Under the regular constant-rank conditions of Michor's treatment, vanishing cocurvature means the vertical distribution is closed under Lie brackets and hence integrable. It is distinct from the companion curvature, which projects brackets of horizontal components into the vertical side. For a genuine fiber bundle with P projecting onto the tangent spaces of the fibers, the vertical distribution is integrable, so cocurvature vanishes. That special case does not make the invariant trivial, since for a general chosen projection the image need not be integrable.

Structural Signature¶

Sig role-phrases:

Tangent projection P — Splits tangent vectors idempotently into image and kernel components. It is constitutive. Counterfactual: Without P²=P the vertical/horizontal decomposition used by this formula is not fixed.
Vertical image distribution — Supplies the vector fields PX and PY whose closure under bracket is tested. It is constitutive. Counterfactual: A scalar field or unrelated subspace cannot substitute for an image distribution on TM.
Lie bracket — Tests whether two vertical vector fields generate a direction outside the vertical distribution. It is constitutive. Counterfactual: Ordinary pointwise vector addition would not diagnose Frobenius integrability.
Complementary projection — Extracts (Id−P)[PX,PY], the bracket component outside the image of P. It is constitutive. Counterfactual: Projecting the opposite way gives the different curvature component R, not cocurvature.
Integrability interpretation — Reads vanishing as bracket closure of the image distribution under constant-rank assumptions. It is central. Counterfactual: Without rank/smoothness hypotheses one cannot simply infer a regular foliation from the displayed algebra.

What It Is Not¶

Not the companion curvature. Curvature tests the horizontal kernel's bracket with the opposite projector.
Not scalar curvature. This is a vector-valued differential two-form, not a number obtained from a Riemannian metric.
Not automatically nonzero. Real fiber-bundle vertical distributions provide a common zero case.
Not an obstruction without hypotheses. The foliation interpretation requires a smooth regular image distribution.
Closest near-miss. Curvature is the nearest algebraic neighbor: it applies the other projector to brackets of horizontal components, whereas cocurvature tests bracket closure of vertical components.

Scope of Application¶

Connection geometry. Compare the complementary obstructions associated with a tangent splitting.
Distribution integrability. Test whether the image plane field can be tangent to local leaves.
Fiber-bundle analysis. Recognize why vertical cocurvature vanishes for tangent-to-fiber distributions.
Bracket identities. Track which projected bracket component enters Bianchi-type relations.

Clarity¶

Compute the bracket of two vector fields after projecting them into im P, then ask whether the bracket escapes im P. The escaped part is cocurvature. If it is zero everywhere and the distribution has constant rank, Frobenius integrability applies. The reversed calculation from horizontal fields is ordinary curvature, not the same test with a new name.

Manages Complexity¶

One vector-valued form condenses infinitely many local bracket-closure checks into a geometric invariant of the splitting. Its compact notation can hide the choice of projection and the rank assumption, so calculations should name the image, kernel, and convention before comparing two sources' curvature signs.

Abstract Reasoning¶

Specify a smooth idempotent tangent projection and verify constant rank where a foliation claim is needed.
Identify im P as the candidate vertical distribution and ker P as its complement.
Choose vertical vector fields PX and PY and evaluate their Lie bracket.
Apply Id−P to isolate the component that leaves the vertical distribution.
Use vanishing or nonvanishing to test integrability, while distinguishing the companion horizontal curvature.

Knowledge Transfer¶

The projection/bracket test transfers among smooth manifold splittings with the same constant-rank convention. A zero cocurvature result for a true fiber bundle does not transfer to an arbitrary plane field, and a nonzero worked R³ example does not contradict integrability of actual fibers. A metaphorical 'organizational obstruction' preserves no vector-valued form or Frobenius theorem.

Examples¶

Canonical¶

On R³, let V be spanned by X=∂x+y∂z and Y=∂y, and let P project onto V along ∂z. The bracket [X,Y]=−∂z lies outside V, so R̄_P(X,Y)=(Id−P)[X,Y]=−∂z is nonzero. This calculated example exhibits the obstruction: the two candidate vertical directions do not close under brackets and therefore do not form local integral surfaces. It is an illustrative construction from Michor's formula, not a claimed experiment or a quoted worked example in his text.

Mapped back: Tangent projection P → projection onto span{∂x+y∂z,∂y} along ∂z; Vertical image distribution → the rank-two plane field V; Lie bracket → [∂x+y∂z,∂y]=−∂z; Complementary projection → (Id−P) retains the −∂z component; Integrability interpretation → nonzero obstruction to integral two-surfaces.

Applied / In Practice¶

Michor applies the same projection calculus to a genuine smooth fiber bundle E→M: vectors tangent to each fiber form the vertical bundle, already integrable as a foliation by fibers. His §9 connection analysis therefore sets the cocurvature to zero and uses that fact when simplifying the associated Bianchi identity. The zero is a geometrically justified result for this setting, not a claim that every arbitrary tangent projection has zero cocurvature.

Mapped back: Tangent projection P → the bundle connection's vertical projector; Vertical image distribution → tangent spaces of the actual fibers; Lie bracket → vertical vector fields close within fiber tangent directions; Complementary projection → their brackets have zero horizontal component; Integrability interpretation → R̄=0, enabling the cited Bianchi simplification.

Structural Tensions¶

T1 — Algebraic Projection versus Geometric Foliation. The two-form is calculated from a projection, but interpreting its zero set as a foliation needs smooth constant-rank/Frobenius assumptions.

Diagnostic: Are the image spaces a regular smooth distribution?

T2 — General Connection versus Fiber-Bundle Specialization. An arbitrary tangent splitting can have nonzero cocurvature; an actual vertical fiber distribution is integrable and makes it vanish.

Diagnostic: Is the 'vertical' image an abstract chosen plane field or tangent to pre-existing fibers?

Structural–Framed Character¶

The skeleton is an obstruction to closure: candidate directions fail to remain within their own class under a combination operation. Cocurvature makes this precise for a constant-rank tangent-bundle projection P using (Id−P)[PX,PY]. It is an approved unparented root because the live scalar-valued differential form does not strictly contain this tangent-valued two-form.

Evaluative weight: Vanishing is an exact integrability test under the stated smoothness and rank assumptions, not a visual judgment of flatness.

Human-practice-bound: Choice of projection fixes which directions count as vertical or complementary.

Institutional origin: Differential geometry supplies the Lie bracket and Frobenius interpretation.

Vocabulary travels: Organizational “obstruction” has no tangent fields or bracket formula.

Import versus recognize: The projection/bracket test transfers across suitable manifold splittings only after P and its image distribution are declared.

Its character: A formal vector-valued integrability obstruction, not a prime for resistance to change.

Structural Core vs. Domain Accent¶

Skeletal core. Failure of closure under an operation can reveal that a proposed substructure is not integrable.

Domain-bound accent. For cocurvature the objects are tangent vector fields, the operation is Lie bracket, and the complementary projector measures the part outside the image of P. Zero has a geometric integrability interpretation under appropriate conditions.

Why not prime. Scalar differential forms and generic closure metaphors omit the vector-valued projection formula. The named invariant depends on smooth tangent geometry.

Related — curvature. The same splitting has a companion term from horizontal brackets projected vertically; the two obstruction tests are complementary, not synonyms.
Related — differential forms. Cocurvature is vector-valued and needs its tangent-bundle projection; a scalar-valued alternating form need not encode this integrability test.

Neighborhood in Abstraction Space¶

Cocurvature sits in a sparse region of the domain-specific corpus (81^st percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Surface Geometry & Projective Transforms (6 abstractions)

Nearest neighbors

Piola transformation — 0.83
Macbeath Region — 0.82
Rational normal curve — 0.82
Strähle construction — 0.82
Alternating-Direction Implicit Method — 0.82

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

Not to Be Confused With¶

Horizontal curvature. Tell: Which input fields and output projector appear in the bracket formula?
Riemannian scalar curvature. Tell: Is the result a scalar from a metric or a tangent-valued two-form?
An arbitrary Lie bracket. Tell: Has its outside-im-P component been isolated by Id−P?
Nonintegrable singular distribution. Tell: Do constant-rank conditions support the stated Frobenius conclusion?

References¶

Peter W. Michor, Gauge Theory for Fiber Bundles, §§8.13 and 9, author-hosted manuscript: https://deferentialgeometry.org/papers/Gauge%20Theory%20for%20Fiber%20Bundles.pdf
Frozen Wikipedia discovery revision: https://en.wikipedia.org/wiki/Cocurvature (revision 1302768457).
Preserved source candidate: https://www.emis.de///monographs/KSM/