Manifolds 19 | Tangent Space for Submanifolds

Tangent spaces for submanifolds in Euclidean space are built directly from derivatives of local parameterizations, turning the geometry of a curved set into a linear subspace where calculus becomes possible. For a k-dimensional submanifold M inside R^n, the key move is to choose a local parameterization Φ: R^k → M that flattens the manifold locally. Because Φ is differentiable (coming from the smooth structure), its differential at a point produces a linear map from R^k into R^n. The tangent space at a point p ∈ M is then defined as the range of that differential—equivalently, the span of the columns of the Jacobian matrix of Φ at the corresponding parameter value. This construction yields a concrete vector subspace of R^n whose dimension matches the manifold’s dimension k, even though the ambient space has dimension n.

The motivation is easiest to see through the way submanifolds look in coordinates. Locally, charts can “flatten” M so it resembles R^k, and the inverse of the flattening homeomorphism lets one build a parameterization Φ that maps k coordinates into the curved set. In the simplest case, the circle S^1 in R^2 can be parameterized by Φ(t) = (cos t, sin t). At each point on the circle, the tangent space is the one-dimensional line in R^2 that best approximates the circle near that point—conceptually like translating the derivative direction to the point itself.

A more instructive example is a surface in R^3 given as the graph of a function f: R^2 → R. The submanifold M = G_f consists of points (x, y, f(x, y)) as (x, y) ranges over R^2. Here the parameterization is global: Φ(x, y) = (x, y, f(x, y)). Computing the Jacobian matrix of Φ produces two column vectors: one from partial differentiation with respect to x, and one from partial differentiation with respect to y. Concretely, the columns are (1, 0, ∂f/∂x) and (0, 1, ∂f/∂y), each evaluated at the point (x, y). The tangent space at p = (x, y, f(x, y)) is the linear span of these two vectors, giving a 2-dimensional subspace of R^3.

This derivative-based definition does more than name a geometric object. It provides a local linearization of the manifold, which is exactly what makes differential calculus on curved sets workable. The same strategy—using differentials of smooth maps to obtain linear subspaces—sets up the later extension to tangent spaces for abstract manifolds, where there may be no ambient R^n to directly visualize the geometry. In short: tangent spaces for submanifolds are the ranges of differentials of parameterizations, turning curvature into a linear structure suitable for computation and further theory.