Diffeomorphisms are the isomorphisms of smooth geometry. A [smooth manifold](/page/Smooth%20Manifold) is built by gluing pieces of Euclidean space together with smooth transition maps, and a diffeomorphism is the kind of map that preserves exactly that smooth structure. It identifies two spaces not merely as [topological spaces](/page/Topology), but as spaces on which differentiation, tangent vectors, vector fields, differential forms, flows, and Riemannian constructions make sense in the same way. A homeomorphism preserves continuity, open sets, compactness, and connectedness. A diffeomorphism preserves the stronger information needed for calculus. This distinction matters because two spaces can be topologically identical while carrying inequivalent smooth structures, and because a bijective smooth map may fail to have a smooth inverse. Diffeomorphisms therefore mark the correct notion of sameness for tangent spaces, [derivatives](/page/Derivative), vector fields, and differential forms. ## Definition The central question is when two smooth manifolds should count as the same object for calculus. A continuous reversible map is not enough, because it may destroy derivatives; a smooth map is not enough, because it may collapse points or have a badly behaved inverse. The definition therefore requires reversibility inside the smooth category itself. [definition: Diffeomorphism] Let $M$ and $N$ be smooth manifolds. A map $F: M \to N$ is a diffeomorphism if $F$ is bijective, $F$ is smooth, and the inverse map $F^{-1}: N \to M$ is smooth. [/definition] To use diffeomorphisms in calculations, we need a version of the definition that lives inside open subsets of Euclidean space. Coordinate charts reduce manifold questions to this setting, where smoothness can be tested with partial derivatives and inverse regularity can be compared with the [Jacobian matrix](/page/Jacobian%20Matrix). The Euclidean definition is the bridge between the abstract manifold notion and the computations used in the inverse function theorem. [definition: Euclidean Diffeomorphism] Let $U$ and $V$ be open subsets of $\mathbb{R}^n$. A map $f: U \to V$ is a Euclidean diffeomorphism if $f$ is bijective, $f$ is smooth, and the inverse map $f^{-1}: V \to U$ is smooth. [/definition] The Euclidean definition does not add a separate Jacobian hypothesis. Instead, invertibility of the Jacobian is forced by the existence of a smooth inverse. The first diagnostic question is what smooth reversibility forces at infinitesimal scale: if a change of variables has a smooth inverse, then its linear approximation must also be reversible; otherwise tangent vectors would be collapsed in a way no smooth inverse could undo. [quotetheorem:9884] This theorem explains why smooth bijectivity alone is too weak: the forward map may be perfectly smooth and one-to-one, while its inverse develops a cusp at the point where the derivative degenerates. [example: Smooth Bijection with Non-Smooth Inverse] Let $f: \mathbb{R} \to \mathbb{R}$ be given by $f(x)=x^3$. Since $x^3$ is a polynomial, $f$ is smooth. If $f(a)=f(b)$, then $a^3=b^3$, so \begin{align*} 0=a^3-b^3=(a-b)(a^2+ab+b^2). \end{align*} The factor $a^2+ab+b^2$ is nonnegative and vanishes only when $a=b=0$, so in all cases $a=b$; hence $f$ is injective. For every $y \in \mathbb{R}$, the number $\operatorname{sgn}(y)|y|^{1/3}$ satisfies \begin{align*} \left(\operatorname{sgn}(y)|y|^{1/3}\right)^3=\operatorname{sgn}(y)^3|y|=y. \end{align*} Thus $f$ is surjective, and its inverse is \begin{align*} f^{-1}(y)=\operatorname{sgn}(y)|y|^{1/3}. \end{align*} Let $g=f^{-1}$. Since $g(0)=0$, differentiability of $g$ at $0$ would require the difference quotient $g(h)/h$ to have a finite limit as $h \to 0$. For $h>0$, \begin{align*} \frac{g(h)-g(0)}{h-0}=\frac{h^{1/3}}{h}=h^{-2/3}. \end{align*} As $h \to 0^+$, this tends to $+\infty$, so $g$ is not differentiable at $0$. Therefore the smooth bijection $f$ is not a diffeomorphism, because its inverse is not smooth. The infinitesimal obstruction appears at the same point: $f'(x)=3x^2$, so \begin{align*} Jf_0=[f'(0)]=[0]. \end{align*} The determinant of the $1 \times 1$ matrix $[0]$ is $0$, so $Jf_0$ is not invertible. [/example] The example is a useful boundary case: bijectivity and smoothness in the forward direction do not control the regularity of the inverse. A single diffeomorphism is a map, but classification problems ask whether two manifolds can be identified by at least one such map. This separates the existence of a smooth equivalence from the choice of a particular equivalence. [definition: Diffeomorphic Manifolds] Let $M$ and $N$ be smooth manifolds. The manifolds $M$ and $N$ are diffeomorphic if there exists a diffeomorphism $F: M \to N$. [/definition] Classification compares different manifolds, while symmetry studies maps from a manifold back to itself. For a fixed manifold, the natural object is not just one self-diffeomorphism but the whole system of smooth reversible transformations of that manifold. Because such transformations compose and have smooth inverses, they form a group; the next definition names this group and fixes its notation. [definition: Diffeomorphism Group] Let $M$ be a smooth manifold. The diffeomorphism group of $M$, denoted $\operatorname{Diff}(M)$, is the set of all diffeomorphisms $F: M \to M$ with group operation given by composition. [/definition] The identity map is the identity element, and the inverse of a diffeomorphism is again a diffeomorphism. Thus $\operatorname{Diff}(M)$ records the smooth symmetries of $M$. ## Equivalent Characterisations The definition is global: it requires bijectivity and a smooth inverse. In Euclidean space, the [inverse function theorem](/theorems/51) explains how smooth inverses are detected locally by the Jacobian matrix. [quotetheorem:9885] The Euclidean test uses matrices because Euclidean tangent spaces have standard coordinates. On a manifold there is no preferred global coordinate system, so we need the corresponding statement in terms of differentials between tangent spaces. [quotetheorem:9886] This result turns smooth equivalence into linear equivalence at every point. To use this infinitesimal condition locally, we need the notion of a map that is a diffeomorphism near each point even when it may fail to be globally one-to-one. [definition: Local Diffeomorphism] Let $M$ and $N$ be smooth manifolds. A smooth map $F: M \to N$ is a local diffeomorphism if for every $p \in M$, there exists an open neighbourhood $U_p \subset M$ of $p$ such that $F(U_p) \subset N$ is open and the restricted map $F|_{U_p}: U_p \to F(U_p)$ is a diffeomorphism. [/definition] Local diffeomorphisms preserve the smooth structure near each point, but they need not be globally injective. This distinction motivates a practical recognition principle: local smooth invertibility plus global bijectivity gives a genuine diffeomorphism. [quotetheorem:9887] The criterion separates the local calculus problem from the global topology problem. Local invertibility is controlled by derivatives; global invertibility is controlled by how the map wraps, folds, or identifies points. ## Examples The simplest nontrivial examples are smooth changes of scale and position. They show that diffeomorphism is flexible: it does not preserve distances, angles, or linear structure unless additional hypotheses are imposed. [example: Open Interval and Real Line] Consider \begin{align*} f:(-1,1)\to \mathbb{R}, \qquad f(x)=\frac{x}{1-x^2}. \end{align*} The denominator satisfies $1-x^2>0$ for $x\in(-1,1)$, so $f$ is smooth on $(-1,1)$. Define \begin{align*} g(y)=\frac{2y}{1+\sqrt{1+4y^2}}. \end{align*} Since $1+\sqrt{1+4y^2}>0$ for every $y\in\mathbb{R}$, the formula for $g$ is smooth on $\mathbb{R}$. Also, \begin{align*} |g(y)|=\frac{2|y|}{1+\sqrt{1+4y^2}}. \end{align*} Because $\sqrt{1+4y^2}>2|y|$, the denominator is larger than $2|y|$, so $|g(y)|<1$ when $y\neq 0$, and $g(0)=0$. Hence $g(y)\in(-1,1)$ for every $y\in\mathbb{R}$. We verify that $g$ is the inverse of $f$. First let $x\in(-1,1)$ and put $y=f(x)$. Then \begin{align*} 1+4y^2=1+\frac{4x^2}{(1-x^2)^2}. \end{align*} Combining the fractions gives \begin{align*} 1+4y^2=\frac{(1-x^2)^2+4x^2}{(1-x^2)^2}. \end{align*} Expanding the numerator, \begin{align*} (1-x^2)^2+4x^2=1-2x^2+x^4+4x^2=1+2x^2+x^4=(1+x^2)^2. \end{align*} Since $1-x^2>0$ and $1+x^2>0$, \begin{align*} \sqrt{1+4y^2}=\frac{1+x^2}{1-x^2}. \end{align*} Therefore \begin{align*} g(f(x))=\frac{2x/(1-x^2)}{1+(1+x^2)/(1-x^2)}. \end{align*} The denominator is \begin{align*} 1+\frac{1+x^2}{1-x^2}=\frac{1-x^2+1+x^2}{1-x^2}=\frac{2}{1-x^2}. \end{align*} Thus \begin{align*} g(f(x))=\frac{2x/(1-x^2)}{2/(1-x^2)}=x. \end{align*} Conversely, let $y\in\mathbb{R}$ and set $x=g(y)$. Since $g(f(x))=x$ for every $x\in(-1,1)$, the equation $x=g(y)$ implies that $x$ is the unique point of $(-1,1)$ whose image under $f$ is $y$ once we check $f(x)=y$. From the definition of $x$, \begin{align*} x\left(1+\sqrt{1+4y^2}\right)=2y. \end{align*} Rearranging gives \begin{align*} x\sqrt{1+4y^2}=2y-x. \end{align*} Squaring both sides gives \begin{align*} x^2(1+4y^2)=4y^2-4xy+x^2. \end{align*} Canceling $x^2$ from both sides yields \begin{align*} 4x^2y^2=4y^2-4xy. \end{align*} Dividing by $4$ and moving terms gives \begin{align*} y^2(1-x^2)=xy. \end{align*} If $y=0$, then $x=g(0)=0$ and $f(x)=0=y$. If $y\neq 0$, divide by $y$ to obtain \begin{align*} y(1-x^2)=x. \end{align*} Since $1-x^2>0$, this is equivalent to \begin{align*} y=\frac{x}{1-x^2}=f(x). \end{align*} So $f(g(y))=y$ for every $y\in\mathbb{R}$. Thus $f$ is bijective with smooth inverse $g$, so it is a Euclidean diffeomorphism. Consequently the open interval $(-1,1)$ and the real line $\mathbb{R}$ are diffeomorphic, even though one is bounded as a subset of Euclidean space and the other is not. [/example] This example warns against confusing diffeomorphism with metric sameness. The bounded interval and the unbounded line have different Euclidean sizes, but as one-dimensional smooth manifolds they are smoothly identical. The next example shows a map that is locally a diffeomorphism everywhere but fails globally because it identifies different points. [example: Covering Map of the Circle] Let $S^1 \subset \mathbb{R}^2$ be the unit circle. For $(x,y)=(\cos t,\sin t)$, define \begin{align*} F(x,y)=(\cos 2t,\sin 2t). \end{align*} Using the double-angle identities, this formula is equivalently \begin{align*} F(x,y)=(x^2-y^2,2xy). \end{align*} This expression is independent of the chosen angle $t$. It also lands in $S^1$, because if $x^2+y^2=1$, then \begin{align*} (x^2-y^2)^2+(2xy)^2=x^4-2x^2y^2+y^4+4x^2y^2=x^4+2x^2y^2+y^4=(x^2+y^2)^2=1. \end{align*} Since the coordinate functions $x^2-y^2$ and $2xy$ are polynomial functions on $\mathbb{R}^2$, the restriction $F:S^1\to S^1$ is smooth. We show that $F$ is a local diffeomorphism. Fix $p=(\cos t_0,\sin t_0)\in S^1$, and choose $0<\varepsilon<\pi/2$. Let \begin{align*} U=\{(\cos u,\sin u):u\in(t_0-\varepsilon,t_0+\varepsilon)\}. \end{align*} This is an open arc in $S^1$. On this arc, use the angular coordinate $v=u-t_0$, so $v\in(-\varepsilon,\varepsilon)$. The image of a point with angle $u=t_0+v$ has angle $2u=2t_0+2v$, so in angular coordinates on the image arc the restricted map is \begin{align*} v\mapsto 2v. \end{align*} Its inverse in these coordinates is \begin{align*} w\mapsto \frac{w}{2}. \end{align*} Both maps are smooth functions between open intervals, so $F|_U$ is a diffeomorphism from $U$ onto the open arc \begin{align*} F(U)=\{(\cos s,\sin s):s\in(2t_0-2\varepsilon,2t_0+2\varepsilon)\}. \end{align*} Thus $F$ is a local diffeomorphism. However, $F$ is not injective. For every $t$, \begin{align*} (\cos(t+\pi),\sin(t+\pi))=(-\cos t,-\sin t), \end{align*} so this point is distinct from $(\cos t,\sin t)$. But \begin{align*} F(\cos(t+\pi),\sin(t+\pi))=(\cos(2t+2\pi),\sin(2t+2\pi))=(\cos 2t,\sin 2t)=F(\cos t,\sin t). \end{align*} Therefore $F$ is a local diffeomorphism but not a diffeomorphism, because a diffeomorphism must be bijective. [/example] Local information alone cannot classify a global smooth map. The map may preserve all infinitesimal data while still folding the manifold over itself. Some diffeomorphisms arise as coordinate changes between standard models. This is especially important in geometry, where changing charts should not change the intrinsic object under study. [example: Stereographic Coordinate Change] Write a point of $S^n \subset \mathbb{R}^{n+1}$ as $(u,t)$, where $u\in\mathbb{R}^n$ and $t\in\mathbb{R}$. Thus $(u,t)\in S^n$ means \begin{align*} |u|^2+t^2=1. \end{align*} Let $N=(0,\ldots,0,1)$. For $(u,t)\neq N$, we have $t\neq 1$, so define stereographic projection from $N$ by \begin{align*} \sigma_N(u,t)=\frac{u}{1-t}. \end{align*} Each coordinate of $\sigma_N$ is a quotient of smooth functions with nonzero denominator on $S^n\setminus\{N\}$, so $\sigma_N$ is smooth. Define $\tau:\mathbb{R}^n\to S^n\setminus\{N\}$ by \begin{align*} \tau(x)=\left(\frac{2x}{1+|x|^2},\frac{|x|^2-1}{1+|x|^2}\right). \end{align*} Since $1+|x|^2>0$, every coordinate of $\tau$ is smooth. To check that $\tau(x)$ lies on $S^n$, compute \begin{align*} \left|\frac{2x}{1+|x|^2}\right|^2+\left(\frac{|x|^2-1}{1+|x|^2}\right)^2=\frac{4|x|^2}{(1+|x|^2)^2}+\frac{(|x|^2-1)^2}{(1+|x|^2)^2}. \end{align*} Combining the fractions gives \begin{align*} \frac{4|x|^2+(|x|^2-1)^2}{(1+|x|^2)^2}=\frac{4|x|^2+|x|^4-2|x|^2+1}{(1+|x|^2)^2}. \end{align*} The numerator is \begin{align*} |x|^4+2|x|^2+1=(1+|x|^2)^2. \end{align*} Hence \begin{align*} \left|\frac{2x}{1+|x|^2}\right|^2+\left(\frac{|x|^2-1}{1+|x|^2}\right)^2=1. \end{align*} Also the last coordinate of $\tau(x)$ is never $1$, because \begin{align*} \frac{|x|^2-1}{1+|x|^2}=1 \end{align*} would imply $|x|^2-1=1+|x|^2$, hence $-1=1$. Thus $\tau(x)\in S^n\setminus\{N\}$. Now compute $\sigma_N(\tau(x))$. The last coordinate of $\tau(x)$ is $(|x|^2-1)/(1+|x|^2)$, so \begin{align*} 1-\frac{|x|^2-1}{1+|x|^2}=\frac{1+|x|^2-|x|^2+1}{1+|x|^2}=\frac{2}{1+|x|^2}. \end{align*} Therefore \begin{align*} \sigma_N(\tau(x))=\frac{2x/(1+|x|^2)}{2/(1+|x|^2)}=x. \end{align*} Conversely, take $(u,t)\in S^n\setminus\{N\}$ and set $x=\sigma_N(u,t)=u/(1-t)$. Since $|u|^2+t^2=1$, we have \begin{align*} |u|^2=1-t^2=(1-t)(1+t). \end{align*} Thus \begin{align*} 1+|x|^2=1+\frac{|u|^2}{(1-t)^2}=1+\frac{(1-t)(1+t)}{(1-t)^2}. \end{align*} Since $t\neq 1$, we may cancel one factor of $1-t$: \begin{align*} 1+|x|^2=1+\frac{1+t}{1-t}=\frac{1-t+1+t}{1-t}=\frac{2}{1-t}. \end{align*} The first $n$ coordinates of $\tau(x)$ are therefore \begin{align*} \frac{2x}{1+|x|^2}=\frac{2u/(1-t)}{2/(1-t)}=u. \end{align*} For the last coordinate, compute \begin{align*} |x|^2-1=\frac{|u|^2}{(1-t)^2}-1=\frac{|u|^2-(1-t)^2}{(1-t)^2}. \end{align*} Using $|u|^2=1-t^2$ gives \begin{align*} |u|^2-(1-t)^2=1-t^2-(1-2t+t^2)=2t-2t^2=2t(1-t). \end{align*} Hence \begin{align*} |x|^2-1=\frac{2t}{1-t}. \end{align*} Dividing by $1+|x|^2=2/(1-t)$ gives \begin{align*} \frac{|x|^2-1}{1+|x|^2}=\frac{2t/(1-t)}{2/(1-t)}=t. \end{align*} So $\tau(\sigma_N(u,t))=(u,t)$. Thus $\sigma_N$ is bijective with smooth inverse $\tau$, so it is a diffeomorphism from $S^n\setminus\{N\}$ onto $\mathbb{R}^n$. This makes the punctured sphere smoothly identical to Euclidean space through the stereographic coordinates, even though it sits as a curved subset of $\mathbb{R}^{n+1}$. [/example] This example is the local model for many geometric calculations on the sphere. Diffeomorphisms allow a curved manifold to be studied through Euclidean coordinates while keeping track of which statements are intrinsic. ## Properties Smooth equivalences should be stable under doing one change of variables after another. Without this closure property, diffeomorphism would not behave like an isomorphism relation for smooth manifolds. [quotetheorem:9888] Together with identity maps and inverses, this makes smooth manifolds into a category whose isomorphisms are diffeomorphisms. To understand how these isomorphisms affect infinitesimal objects, we need the chain rule at the level of tangent maps. [quotetheorem:3907] The chain rule tells us how tangent maps compose. The next identity is needed when a calculation changes variables by $F$ and then converts tangent vectors or covectors back through $F^{-1}$. [quotetheorem:9889] This identity is often the most efficient way to compute how tangent vectors transform after a coordinate change. Before using derivatives to extract smooth invariants, there is a simpler obstruction to keep in view: a diffeomorphism is reversible as a smooth map, so it must also be reversible after forgetting the smooth structure and remembering only the topology. Any proposed diffeomorphism must therefore first pass this underlying homeomorphism test. [quotetheorem:9890] The converse fails in general. A homeomorphism need not be differentiable, and even a differentiable bijection can fail to have differentiable inverse. The tangent-space isomorphism theorem gives a first smooth invariant that is stronger than pure topology. [quotetheorem:9891] Dimension preservation is only the beginning. Orientability, compactness, boundary structure, de Rham cohomology, and many characteristic classes are also designed to be invariant under suitable diffeomorphisms. ## Relationship to Other Concepts Diffeomorphism refines [homeomorphism](/page/Homeomorphism). Every diffeomorphism is a homeomorphism, but a homeomorphism only preserves open sets and continuous structure. Diffeomorphism preserves the differentiable structure needed to speak about derivatives and tangent spaces. It also refines the notion of smooth map. A smooth map $F: M \to N$ may collapse dimensions, identify distinct points, or fail to have a smooth inverse. A diffeomorphism is precisely the smooth map that is reversible inside the smooth category. Local diffeomorphisms are closely related to submersions, immersions, and the inverse function theorem. For maps $F: M \to N$ between manifolds of the same dimension, being both an immersion and a submersion at $p$ is equivalent to invertibility of $dF_p$, the infinitesimal condition that makes $F$ a diffeomorphism near $p$. Diffeomorphism is also the organizing [equivalence relation](/page/Equivalence%20Relation) in differential topology. Classifying manifolds up to diffeomorphism asks when two smooth spaces are the same for calculus, even if their presentations, embeddings, or coordinate descriptions look different. In Riemannian geometry, a diffeomorphism need not preserve distances. A distance-preserving diffeomorphism is an isometry. Thus diffeomorphism preserves smooth structure, while isometry preserves both smooth structure and metric geometry. ## References Lee, *Introduction to Smooth Manifolds* (2013). Tu, *An Introduction to Manifolds* (2011). Guillemin and Pollack, *Differential Topology* (1974). Spivak, *A Comprehensive Introduction to Differential Geometry, Volume 1* (1979).