Manifolds: Definitions, Examples, And Intuition

Build a working mental model of manifolds as spaces that look like $\mathbb{R}^n$ up close but may not admit one global coordinate system. Learn why charts and atlases matter, how smoothness is imposed, and how tangents and diffeomorphisms behave under change of coordinates.

A manifold is what you get when a space behaves like $\mathbb{R}^n$ in every small neighborhood, yet refuses to be described by one coordinate grid everywhere. The sphere is the honest example. You can parametrize it smoothly near any point, but any attempt to cover it with one chart either tears, pinches, or misses something. That is not a technical annoyance. It is the whole reason manifolds are built out of charts, and why the overlap rules between charts carry the real geometry.

No global coordinates on a sphere

The phrase locally Euclidean means each point has a neighborhood that looks like an open set in $\mathbb{R}^n$, even if the whole space does not. On $S^2$, stereographic projection from the north pole gives smooth coordinates on everything except the north pole. A second chart from the south pole fills in the missing point. The price is overlap, and overlap forces you to confront how coordinates change.

Look at how overlapping coordinate patches fit together.

In the overlap region you have two legitimate descriptions of the same points, so there is a map from one coordinate representation to the other. That map is the first glimpse of a transition function. The manifold is not the collection of coordinates. It is the underlying space plus the rule that all coordinate descriptions agree on overlaps in the right way.

Charts are unavoidable
On many spaces, the obstruction is global. If a single chart exists everywhere, the manifold is already an open subset of $\mathbb{R}^n$.

Topological manifolds and why the axioms exist

A topological manifold is a topological space $M$ such that every point has a neighborhood homeomorphic to an open set in $\mathbb{R}^n$, plus two global conditions that keep the theory from breaking.

A standard definition packages three requirements:

• Hausdorff: distinct points can be separated by disjoint neighborhoods.
• Second countable: there is a countable basis for the topology.
• Locally $\mathbb{R}^n$: each point has a neighborhood homeomorphic to an open subset of $\mathbb{R}^n$.

The local $\mathbb{R}^n$ condition guarantees coordinates exist near each point. The other two prevent pathologies that would make local reasoning fail to glue into global reasoning.

Explore what breaks if you drop each condition.

Dropping Hausdorff can produce spaces where limits are not unique, so basic constructions like taking the graph of a function or identifying points can behave unpredictably. Dropping second countability can allow wildly large manifolds where you lose key theorems that depend on sequences, countable covers, and manageable atlases. You can still do topology on such spaces, but much of the standard manifold toolkit assumes these hypotheses.

Charts, atlases, and smooth structure

A chart on $M$ is a homeomorphism $\varphi:U\to\varphi(U)\subset\mathbb{R}^n$ from an open set $U\subset M$. An atlas is a collection of charts whose domains cover $M$. The crucial part is compatibility. If $(U,\varphi)$ and $(V,\psi)$ overlap, the transition map is

$$\psi\circ\varphi^{-1}:\varphi(U\cap V)\to\psi(U\cap V).$$

For a smooth manifold, all transition maps between charts in the atlas are smooth.

Compare compatible and incompatible atlases on the same underlying set.

Smoothness is not automatically present in a topological manifold. You are choosing a smooth structure, meaning a maximal family of mutually compatible charts. Sometimes there are multiple non-equivalent choices. The slogan is same points, same open sets, different calculus. The most famous cases are exotic smooth structures, where a space homeomorphic to $\mathbb{R}^n$ can carry a smooth structure not diffeomorphic to the standard one, a phenomenon that starts in dimension 4.

Smoothness lives in overlaps
You never check smoothness on the manifold directly. You check it after translating into $\mathbb{R}^n$ through charts, and the overlaps tell you whether those translations agree.

Examples and how to build new manifolds

The fastest way to get intuition is to treat manifolds as closed under a few constructions. Many familiar spaces show up as outputs of these operations, not as one-off definitions.

See how common constructions connect to canonical examples.

Standard families you should recognize

• $S^n$ is the set $\{x\in\mathbb{R}^{n+1}:\|x\|=1\}$ with charts coming from projection or from solving for one coordinate locally.
• $T^n$ can be built as a product $(S^1)^n$ or as a quotient $\mathbb{R}^n/\mathbb{Z}^n$.
• Real projective space $\mathbb{RP}^n$ is a quotient of $S^n$ identifying antipodal points, or equivalently lines through the origin in $\mathbb{R}^{n+1}$.
• Lie groups are manifolds with a group operation that is smooth, like $S^1$, $SO(n)$, and $GL(n,\mathbb{R})$.

Constructions that generate more

Products preserve the manifold property, so $M\times N$ has dimension $\dim M+\dim N$. Submanifolds appear when a subset is cut out by regular equations, and quotients appear when you identify points by a nice equivalence relation, often from a group action. The phrase nice matters. Bad quotients are a major source of non-manifolds, typically because the result fails Hausdorff or fails to look like $\mathbb{R}^n$ near some points.

Smooth maps and diffeomorphisms

A smooth map $f:M\to N$ is defined by charts. In coordinates, $\psi\circ f\circ\varphi^{-1}$ is a smooth map between open subsets of Euclidean space. This definition is built to be independent of which charts you pick, precisely because of smooth transition maps.

A diffeomorphism is a bijective smooth map with smooth inverse. Locally, invertibility is controlled by the derivative. Globally, topology can obstruct it. A map can be locally invertible everywhere and still fail to be globally one-to-one, like the exponential map $\mathbb{R}\to S^1$ given by $t\mapsto e^{it}$.

Diffeomorphisms preserve anything that is defined purely in smooth terms. Dimension is the first invariant, but not the last. Orientability, the existence of certain vector fields, and many global invariants from topology and geometry survive under diffeomorphism. The point is that diffeomorphism is the notion of sameness for smooth manifolds. Homeomorphism is too weak if you care about calculus.

Tangent spaces that do not depend on coordinates

At a point $p\in M$, the tangent space $T_pM$ is the linear object that captures first-order behavior. There are two standard ways to define it, and learning both is how you stop worrying about coordinates.

• As velocity vectors of smooth curves $\gamma:(-\epsilon,\epsilon)\to M$ with $\gamma(0)=p$, where two curves are equivalent if they have the same derivative in a chart.
• As derivations on germs of smooth functions at $p$, meaning linear maps $v:C^\infty(M)\to\mathbb{R}$ satisfying Leibniz, $v(fg)=v(f)g(p)+f(p)v(g)$.

The velocity picture feels geometric. The derivation picture is coordinate-free and algebraic. They agree, and the bridge is that a curve differentiates functions by $v(f)=\frac{d}{dt}f(\gamma(t))|_{t=0}$.

Watch how a change of chart transforms a tangent basis through the Jacobian.

If $x=\varphi$ and $y=\psi$ are two coordinate systems near $p$, the coordinate representation of a tangent vector changes by the Jacobian of the transition map. This is why $T_pM$ is well-defined. You are not defining tangent vectors to be tuples of numbers. You are defining an abstract vector, then observing that any chart turns it into numbers, and different charts are related by a deterministic linear change of basis.

One object, many coordinate shadows
The manifold forces you to live with multiple coordinate descriptions. The tangent space teaches you how to translate between them without changing the underlying geometric object.

Three proof sketches worth owning

These are the results that quietly power most constructions and let you trust the definitions. You do not need full proofs memorized, but you should be able to reconstruct the main idea.

Open the proof outlines when you want a quick mental refresh.

Invariance of dimension

If $M$ is locally like $\mathbb{R}^m$ and also locally like $\mathbb{R}^n$, then $m=n$. The engine is that there is no homeomorphism between open sets of different Euclidean dimensions. That is not obvious, and proving it is a major theorem in topology.

Implicit function theorem and submanifolds

If $F:\mathbb{R}^{n+k}\to\mathbb{R}^k$ has full rank $k$ at a point of $F^{-1}(0)$, then near that point the zero set looks like the graph of a smooth function $\mathbb{R}^n\to\mathbb{R}^k$. Translating through charts gives the standard criterion for an embedded submanifold.

Partitions of unity

A partition of unity lets you glue local smooth data into global smooth data. You can build global functions with prescribed local behavior, construct Riemannian metrics, and define integrals by patching from coordinate neighborhoods. Many global existence theorems on manifolds reduce to doing something locally in charts and then gluing with a partition of unity.

Where intuition fails

Some of the most common manifold mistakes come from over-trusting the embedded picture, where your manifold is a nice surface sitting inside $\mathbb{R}^3$. Many manifolds are only given abstractly by charts, or as quotients, or as images of maps that are not embeddings.

An immersion can produce a manifold that intersects itself in an ambient space, so the picture looks non-manifold even when the abstract object is perfectly smooth. Boundaries and corners change the local model from $\mathbb{R}^n$ to a half-space or quadrant, so tangent spaces and charts still exist but many theorems need boundary-aware versions. Orbifolds look locally like $\mathbb{R}^n$ modulo a finite group action. They behave manifold-like in many ways, but points with symmetry have local singular structure.

The curved space confusion is the biggest one. Curvature is extra structure, usually from a Riemannian metric, not part of the definition of a manifold. A manifold can be globally curved when embedded, yet intrinsically flat, and the converse can also happen. The manifold gives you the stage. Curvature is a choice of how to measure lengths and angles on that stage.

FAQ on manifolds