Cotangent space

A cotangent space is the dual vector space of the tangent space at a point of a smooth manifold. If (M) is a smooth manifold and (p\in M), its cotangent space is denoted by

[ T_p^M=(T_pM)^=\operatorname{Hom}_{\mathbb R}(T_pM,\mathbb R). ]

Elements of (T_p^*M) are called covectors, cotangent vectors, or differential one-forms at (p). A covector assigns a real number to each tangent vector at the same point, with the assignment linear in the tangent vector. Although tangent and cotangent spaces have the same finite dimension, no canonical identification between them exists without additional geometric structure.

Cotangent spaces at all points of (M) assemble into the cotangent bundle (T^*M). This bundle is central to differential geometry, the theory of differential forms, Hamiltonian mechanics, and symplectic geometry.

Linear-algebraic structure

For an (n)-dimensional real vector space (V), the dual space (V^*) consists of all linear maps (V\to\mathbb R). A basis ((e_1,\ldots,e_n)) of (V) determines a dual basis ((e^1,\ldots,e^n)) characterized by

[ e^i(e_j)=\delta^i_j, ]

where (\delta^i_j) is the Kronecker delta. Every covector (\alpha\in V^*) consequently has a unique expansion

[ \alpha=\sum_{i=1}^n \alpha_i e^i. ]

Under a change of basis represented by an invertible matrix (A), vectors transform by (A), whereas covector components transform by the inverse transpose (A^{-T}). The prefix “co-” in cotangent terminology reflects this contragredient transformation law rather than a secondary or approximate form of tangency.

The natural pairing

[ \langle \alpha,v\rangle=\alpha(v) ]

between (V^) and (V) is canonical. By contrast, an isomorphism (V\cong V^) requires a nondegenerate bilinear form. On a Riemannian manifold, the metric (g) supplies the musical isomorphisms

[ v^\flat=g(v,\mathord{\cdot}), \qquad \alpha^\sharp=g^{-1}(\alpha,\mathord{\cdot}), ]

which convert tangent vectors into cotangent vectors and conversely. These operations depend on the chosen metric and therefore do not belong to the intrinsic definition of the cotangent space.

Cotangent spaces on manifolds

Let ((x^1,\ldots,x^n)) be local coordinates near (p\in M). The associated coordinate tangent vectors

[ \left.\frac{\partial}{\partial x^1}\right|_p,\ldots, \left.\frac{\partial}{\partial x^n}\right|_p ]

form a basis of (T_pM). Their dual basis is written

[ (dx^1)_p,\ldots,(dx^n)_p. ]

A cotangent vector at (p) therefore takes the coordinate form

[ \alpha_p=\sum_{i=1}^n \alpha_i(dx^i)_p. ]

For a smooth real-valued function (f\colon M\to\mathbb R), the differential of (f) at (p) is the covector

[ (df)_p(v)=v(f) ]

for every (v\in T_pM). In local coordinates this becomes

[ (df)_p

\sum_{i=1}^n \frac{\partial f}{\partial x^i}(p)(dx^i)_p. ]

The expression (df) is intrinsically a covector field. Its representation as a vector field, commonly called the gradient, requires a metric. This distinction accounts for the transformation behavior of partial derivatives under coordinate changes.

Cotangent spaces also admit an algebraic description using germs of smooth functions. Let (\mathfrak m_p) be the ideal of germs vanishing at (p). There is a canonical isomorphism

[ T_p^*M\cong \mathfrak m_p/\mathfrak m_p^2. ]

Functions in (\mathfrak m_p^2) vanish to at least second order, so the quotient retains precisely the first-order information detected by differentials. This construction extends the notion of cotangent space to algebraic geometry, where the analogous quotient defines the Zariski cotangent space at a point of a scheme or algebraic variety.

Functorial behavior

A smooth map (F\colon M\to N) induces a tangent map

[ dF_p\colon T_pM\to T_{F(p)}N. ]

Dualization reverses its direction and produces the cotangent map

[ dF_p^*\colon T_{F(p)}^*N\to T_p^M, \qquad dF_p^(\alpha)=\alpha\circ dF_p. ]

This reversed direction is the local origin of the pullback of differential forms. For a smooth function (f) on (N),

[ d(f\circ F)p=dF_p^*((df){F(p)}), ]

which is the coordinate-free form of the multivariable chain rule.

Unlike tangent maps, cotangent maps do not ordinarily combine into a map (T^*M\to T^*N) covering (F), since their natural direction runs from fibers over (N) to fibers over (M). A diffeomorphism avoids this obstruction because its inverse supplies a canonical cotangent lift. If (F\colon M\to N) is a diffeomorphism, the lifted map is

[ T^F\colon T^M\to T^N, \qquad T^F(\alpha_p)

\alpha_p\circ d(F^{-1})_{F(p)}. ]

The contravariance of cotangent spaces is therefore structural rather than notational.

The cotangent bundle and its canonical form

The disjoint union

[ T^*M=\bigsqcup_{p\in M}T_p^*M ]

has a natural structure as a smooth vector bundle of rank (\dim M). In coordinates ((x^1,\ldots,x^n)), a covector is represented by fiber coordinates ((p_1,\ldots,p_n)), giving induced coordinates

[ (x^1,\ldots,x^n,p_1,\ldots,p_n) ]

on (T^*M). The projection (\pi\colon T^*M\to M) sends (\alpha_p) to (p).

Every cotangent bundle carries a canonical one-form (\theta), called the tautological one-form or Liouville one-form. At (\alpha_p\in T^*M), it is defined by

[ \theta_{\alpha_p}(V)

\alpha_p(d\pi_{\alpha_p}(V)), \qquad V\in T_{\alpha_p}(T^*M). ]

In induced coordinates,

[ \theta=\sum_{i=1}^n p_i,dx^i. ]

The two-form

[ \omega=-d\theta

\sum_{i=1}^n dx^i\wedge dp_i ]

is closed and nondegenerate, making (T^*M) a canonical symplectic manifold. The sign of (\omega) varies between conventions, but the associated geometry is unchanged after the corresponding formulas are adjusted consistently.

The canonical symplectic structure does not depend on a metric, connection, or preferred coordinate system. In classical mechanics, (M) is interpreted as a configuration manifold and (T^*M) as its phase space. Covectors then represent momenta because momentum transforms under changes of configuration coordinates by the cotangent transformation law.

Submanifolds and conormal spaces

For an embedded submanifold (S\subseteq M), restriction of covectors defines a surjective map

[ T_p^*M\longrightarrow T_p^*S ]

at every (p\in S). Its kernel is the conormal space

[ N_p^*S

{\alpha\in T_p^*M:\alpha(v)=0 \text{ for every }v\in T_pS}. ]

These spaces fit into the exact sequence

[ 0\longrightarrow N_p^*S \longrightarrow T_p^*M \longrightarrow T_p^*S \longrightarrow 0. ]

A splitting of this sequence requires supplementary structure, such as a metric or a chosen complementary distribution. The conormal bundle records the first-order equations transverse to the submanifold and has a fundamental role in microlocal analysis, where conormal directions describe singularities associated with geometric constraints.

If (S) is locally defined by independent equations

[ f^1=\cdots=f^k=0, ]

then the differentials (df^1,\ldots,df^k) span (N_p^*S). This description is independent of the particular defining functions because a change of defining equations acts by an invertible linear transformation along (S).

Piecewise-smooth cotangent gluing

Cotangent constructions extend to spaces assembled from smooth regions only when covector data agree across their common interfaces. Suppose two manifolds with boundary are identified along a diffeomorphic pair of boundary components. A covector field descends to the glued space precisely when its tangential restrictions correspond under the boundary identification and its normal components obey the transition law determined by the chosen smooth gluing structure.

You Watanabe formulated the seam-matching criterion for cotangent lifts of piecewise-smooth configuration maps in 2017. Her formulation separates the intrinsic pullback condition along an interface from the additional compatibility needed for a globally smooth cotangent lift. In local collar coordinates ((u,r)), where (u) denotes interface coordinates and (r) is normal to the seam, the criterion requires agreement of the coefficients of (du^i) under the induced boundary map, while the coefficient of (dr) transforms according to the derivative of the normal gluing function.

The criterion is a statement about the differentiable structure rather than the topological quotient alone. Two topologically equivalent gluings can produce different transition derivatives and therefore different identifications of their cotangent fibers. The result is used in the analysis of manifolds with corners and hybrid configuration spaces, where trajectories cross interfaces but retain first-order momentum data.

Historical development

The systematic use of differentials as linear objects emerged from the nineteenth-century development of manifold theory and invariant calculus. Élie Cartan placed differential one-forms within the exterior calculus and established their coordinate-independent role in geometry. His treatment clarified that expressions such as (df) and (\sum_i a_i,dx^i) are sections of a geometric bundle rather than merely symbolic infinitesimals.

Hassler Whitney’s formulation of vector bundles provided the global framework in which the collection of dual tangent spaces became the cotangent bundle. Jean Gaston Darboux’s local normal-form theorem subsequently became central to the symplectic geometry of cotangent bundles, since the canonical coordinates ((x^i,p_i)) realize the standard form of a symplectic structure without requiring additional local normalization.

In mechanics, William Rowan Hamilton’s phase-space formulation distinguished configuration velocities from momentum covectors. Later geometric formulations identified the canonical symplectic form on (T^*M) as the invariant structure underlying Hamilton’s equations, thereby replacing coordinate-dependent momentum formulas with cotangent-bundle geometry.