Manifold — Definition

The definition, precisely

A topological n-manifold is a topological space M satisfying three conditions:

1. Hausdorff. Any two distinct points have disjoint open neighbourhoods.
2. Second-countable. There is a countable basis for the topology.
3. Locally Euclidean of dimension n. Every point has an open neighbourhood homeomorphic to an open subset of R^n.

The homeomorphism φ: U → φ(U) ⊂ R^n is called a chart. A collection {(Uα, φα)} of charts covering M is called an atlas. Charts can be glued: on the overlap Uα ∩ Uβ, you get a transition map φβ ∘ φα⁻¹ between two open subsets of R^n.

A smooth (C^∞) n-manifold adds one requirement: every transition map is infinitely differentiable (in both directions, since transition maps are homeomorphisms). The maximal collection of such pairwise compatible charts is the smooth structure.

Worked example — the 2-sphere

S² = {(x, y, z) ∈ R³ : x² + y² + z² = 1}. To prove this is a smooth 2-manifold, exhibit a smooth atlas with two charts:

Chart 1 (stereographic from N = (0,0,1)):
  φ_N: S² \ {N} → R²
  φ_N(x, y, z) = (x / (1−z), y / (1−z))

Chart 2 (stereographic from S = (0,0,−1)):
  φ_S: S² \ {S} → R²
  φ_S(x, y, z) = (x / (1+z), y / (1+z))

Overlap: S² \ {N, S}
Transition: φ_S ∘ φ_N⁻¹: R² \ {0} → R² \ {0}
  (u, v) ↦ (u, v) / (u² + v²)            (inversion in unit circle)

The transition is rational and smooth on R² \ {0}, so the two charts are compatible. Both charts cover S² between them, so the atlas is complete. S² is a smooth 2-manifold.

You cannot cover S² with a single chart — S² is compact, and a chart would give a homeomorphism to an open subset of R², but no open subset of R² is compact and connected without boundary.

Tangent spaces

The tangent space T_pM at a point p is the n-dimensional vector space of "directions at p." There are three equivalent definitions:

• Velocity classes. Equivalence classes of smooth curves γ: (−ε, ε) → M with γ(0) = p, where two curves are equivalent if they have the same first derivative in any chart.
• Derivations. Linear maps v: C^∞(M) → R satisfying the Leibniz rule v(fg) = v(f)·g(p) + f(p)·v(g).
• Coordinate vectors. In a chart (x¹, …, x^n), v = Σ vⁱ ∂/∂xⁱ|_p — formal linear combinations of partial derivatives.

The union TM = ⋃_p T_pM is the tangent bundle — itself a smooth 2n-manifold. Sections of TM are vector fields.

A zoo of examples

Manifold	Dimension	Compact?	Where it appears
R^n	n	No	Trivial chart; the model space
Sphere S^n	n	Yes	2 stereographic charts; surface of (n+1)-ball
Torus T^n = (S¹)^n	n	Yes	Configuration space of n decoupled pendulums
Real projective space RP^n	n	Yes	Lines through origin in R^(n+1); RP^2 is non-orientable
Complex projective space CP^n	2n	Yes	Quantum state space; CP^1 = S² (Riemann sphere)
Lie group SO(3)	3	Yes	Rotations of a rigid body; diffeomorphic to RP^3
Configuration space of n-link arm	n	Yes	Robotics: T^n if joints are circles; harder if obstacles
Spacetime in GR	4	Sometimes	Lorentzian — signature (−,+,+,+); manifold with metric

Historical context

Riemann's 1854 Habilitation lecture Über die Hypothesen, welche der Geometrie zu Grunde liegen ("On the hypotheses underlying geometry") introduced the German word Mannigfaltigkeit ("manifold-ness") for a continuously varying collection of "ways of being." His treatment was philosophical and intuitive — formal definitions came later.

Poincaré (1895, Analysis Situs) developed topological manifolds, proved the Poincaré duality theorem, and stated the conjecture that bears his name (proved by Perelman, 2003). Hermann Weyl's 1913 Die Idee der Riemannschen Fläche gave the modern axiomatic definition via charts and overlap compatibility.

Hassler Whitney's embedding theorem (1944) showed every smooth n-manifold embeds in R^(2n) — meaning the abstract atlas-based definition is no more general than concretely-sitting-inside-Euclidean-space subsets. John Milnor's 1956 discovery of exotic 7-spheres — manifolds homeomorphic but not diffeomorphic to S⁷ — and Donaldson/Freedman's 1982 proof that R^4 has uncountably many distinct smooth structures broke the intuition that "topological = smooth."

Manifolds with boundary

A manifold with boundary uses charts to the closed half-space H^n = {(x¹, …, x^n) : x^n ≥ 0} instead of R^n. The interior is mapped to the open half-space; the boundary ∂M consists of points mapped to {x^n = 0}. The boundary is itself an (n−1)-manifold without boundary.

Examples: closed disk D² (boundary S¹), closed n-ball, half-plane, cylinder S¹ × [0, 1] (boundary = two circles). Stokes' theorem in its modern formulation requires manifolds with boundary.

JavaScript — stereographic chart and transition

// S² as a smooth 2-manifold: two stereographic charts and their transition

// Chart 1: stereographic from north pole (0, 0, 1)
function phiN(x, y, z) {
  if (z === 1) return null;        // north pole excluded
  return { u: x / (1 - z), v: y / (1 - z) };
}

// Inverse of chart 1: R² → S² \ {N}
function phiNInv(u, v) {
  const d = 1 + u * u + v * v;
  return { x: 2 * u / d, y: 2 * v / d, z: (u * u + v * v - 1) / d };
}

// Chart 2: stereographic from south pole (0, 0, -1)
function phiS(x, y, z) {
  if (z === -1) return null;       // south pole excluded
  return { u: x / (1 + z), v: y / (1 + z) };
}

// Transition map phiS ∘ phiN⁻¹: R² \ {0} → R² \ {0}
// Should be (u, v) ↦ (u, v) / (u² + v²)
function transition(u, v) {
  const p = phiNInv(u, v);
  return phiS(p.x, p.y, p.z);
}

console.log(transition(2, 0));    // ≈ { u: 0.5, v: 0 } — inversion in unit circle ✓
console.log(transition(1, 1));    // ≈ { u: 0.5, v: 0.5 } — (u,v)/2 = (u,v)/(u²+v²)
console.log(transition(0.5, 0));  // ≈ { u: 2, v: 0 } — inverse of first call ✓

// Verify the transition is smooth at any point off the origin (rational function)
// The transition map (u, v) ↦ (u, v)/(u² + v²) has Jacobian det = -1/(u² + v²)²,
// which is nonzero for (u, v) ≠ (0, 0). So the transition is a smooth diffeomorphism
// on R² \ {0}, and S² is a smooth 2-manifold with this 2-chart atlas.

Why the definition matters

• General relativity. Spacetime is a Lorentzian 4-manifold — a 4-manifold with a metric of signature (−, +, +, +). Einstein's equation lives on this manifold; without the manifold framework, GR cannot even be stated.
• Configuration spaces in robotics. A robot arm with n revolute joints has configuration space T^n; planning a motion is finding a smooth curve in this manifold. Workspace obstacles cut out submanifolds; geodesic algorithms compute optimal paths.
• Lie groups. Groups like SO(3), SU(2), GL(n) carry compatible smooth manifold structure. The Lie algebra is the tangent space at the identity. Foundation of gauge theory.
• Quantum state space. Pure states of a 2-level system form CP¹ = S² (the Bloch sphere). Generally, pure states of an n-level system form CP^(n−1) — a complex projective manifold.
• Numerical PDEs on curved domains. Climate models, aerodynamics, brain-MRI cortex models all use surface meshes that are discretised 2-manifolds. The atlas formalism justifies what charts to use.
• Topology and 4D anomalies. Exotic R⁴'s show that "smooth" is a genuine additional structure on top of "topological." 4-dimensional gauge theory uses this to detect subtle invariants (Donaldson invariants, Seiberg–Witten invariants).

Common misconceptions

• "A manifold is a surface in R³." Surfaces in R³ are examples, but manifolds are abstract — they need not sit in any ambient space. Whitney's theorem says any n-manifold can be embedded in R^(2n), but the manifold itself is defined intrinsically.
• "Every topological manifold is smooth." True in dimensions 1, 2, 3 (with caveats), but false starting in dimension 4. Topological 4-manifolds can fail to admit any smooth structure (Freedman's E8 manifold). Dimension 4 is uniquely pathological.
• "R^n has a unique smooth structure." True for n ≠ 4 (uncountably many for n = 4). The Donaldson/Freedman result is dimension-specific and one of the deep surprises of 20th-century geometry.
• "You need one global chart." Most manifolds (including S²) cannot be covered by a single chart. The point of the atlas formalism is to glue local data consistently. Trying to force a global chart is what makes things like polar coordinates "singular at the origin."
• "Manifolds have curvature." Curvature requires extra structure — a Riemannian metric or a connection. Bare manifolds have no notion of distance, angle, or curvature. The same smooth manifold can carry many different metrics.
• "Hausdorff is just a technicality." Without Hausdorff, you can build the "line with two origins" — two copies of R glued away from 0. It is locally Euclidean but pathological: limits aren't unique, and many integration theorems fail. Hausdorff is doing real work.