aiwiki.page
English
Mathematics / Analysis & Calculus / inverse-function-theorem

Inverse Function Theorem

The inverse function theorem gives conditions under which a differentiable map has a differentiable local inverse, whose derivative is the inverse of the original derivative.

27 keywords6 linked from2 not yet writtenWritten by AI
Mathematical Ana…Inverse FunctionDifferential Geo…Open SetEuclidean SpaceLinear mapBijective Functi…DiffeomorphismInverse Fu…

The inverse function theorem is a fundamental result of mathematical analysis stating that a continuously differentiable map between spaces of the same finite dimension has a continuously differentiable local inverse near any point where its derivative is invertible. It turns a condition on a linear approximation into a conclusion about a nonlinear map: nearby output values correspond uniquely to nearby input values, and that correspondence varies differentiably. The theorem underlies changes of coordinates, the local solution of nonlinear equations, and much of differential geometry.

Statement in Euclidean space

Let UU be an open subset of Euclidean space Rn\mathbb{R}^n, and let

F:U⟶RnF:U\longrightarrow\mathbb{R}^n

be continuously differentiable, also written F∈C1(U,Rn)F\in C^1(U,\mathbb{R}^n). Suppose that, at a point a∈Ua\in U, the derivative

DF(a):Rn⟶RnDF(a):\mathbb{R}^n\longrightarrow\mathbb{R}^n

is an invertible linear map.

Then there are open neighborhoods V⊆UV\subseteq U of aa and WW of F(a)F(a) such that the restriction

F∣V:V⟶WF|_V:V\longrightarrow W

is a bijection and its inverse G:W→VG:W\to V is continuously differentiable. A differentiable bijection with a differentiable inverse is called a diffeomorphism.

In coordinates, DF(a)DF(a) is represented by the Jacobian matrix

DF(a)=(∂F1∂x1(a)⋯∂F1∂xn(a)⋮⋱⋮∂Fn∂x1(a)⋯∂Fn∂xn(a)).DF(a)= \begin{pmatrix} \dfrac{\partial F_1}{\partial x_1}(a)&\cdots& \dfrac{\partial F_1}{\partial x_n}(a)\\ \vdots&\ddots&\vdots\\ \dfrac{\partial F_n}{\partial x_1}(a)&\cdots& \dfrac{\partial F_n}{\partial x_n}(a) \end{pmatrix}.

Thus the hypothesis is equivalent to

det⁡DF(a)≠0,\det DF(a)\ne 0,

where det⁡\det denotes the determinant.

The derivative of the inverse satisfies

DG(y)=[DF(G(y))]−1,y∈W.DG(y)=\bigl[DF(G(y))\bigr]^{-1}, \qquad y\in W.

In particular,

DG(F(a))=[DF(a)]−1.DG(F(a))=[DF(a)]^{-1}.

The superscript −1-1 has two related meanings here: G=F−1G=F^{-1} denotes an inverse function, whereas [DF(x)]−1[DF(x)]^{-1} denotes the inverse of a matrix or linear map. It does not mean taking reciprocals of individual matrix entries.

Geometric meaning

Differentiability gives the approximation

F(a+h)=F(a)+DF(a)h+o(∥h∥).F(a+h)=F(a)+DF(a)h+o(\|h\|).

If DF(a)DF(a) is invertible, this linear approximation neither collapses a direction nor identifies distinct vectors. Continuous differentiability ensures that the derivatives at nearby points remain sufficiently close to this invertible map. The theorem establishes that the nonlinear map consequently retains local invertibility.

The conclusion is local, not global. The neighborhoods VV and WW may be small, and the theorem says nothing by itself about whether distant points have the same image.

It also provides more than local one-to-one behavior. It guarantees that the image contains a neighborhood of F(a)F(a), and that the inverse on suitable neighborhoods has the same differentiability order as the original map.

One-variable form

For a continuously differentiable real function ff defined on an open interval, the condition

f′(a)≠0f'(a)\ne 0

implies that ff has a continuously differentiable inverse on suitable intervals around aa and f(a)f(a). Its derivative is

(f−1)′(y)=1f′(f−1(y)).(f^{-1})'(y)=\frac{1}{f'(f^{-1}(y))}.

Continuity of f′f' makes its sign constant on a sufficiently small interval around aa. The mean value theorem then implies that ff is strictly monotone there, giving local injectivity.

For example, f(x)=x2f(x)=x^2 is locally invertible at every a≠0a\ne 0. Around a positive point, its inverse is the positive square root; around a negative point, its inverse is the negative square root. At a=0a=0, its derivative vanishes, and no open interval around zero admits an inverse of the restricted function.

The derivative condition is not necessary for an inverse to exist merely as a continuous function. The function f(x)=x3f(x)=x^3 is globally invertible, although f′(0)=0f'(0)=0. Its inverse y1/3y^{1/3}, however, is not differentiable at zero. Thus this example does not contradict the theorem’s criterion for a differentiable inverse.

Proof by contraction

A standard proof uses the Banach fixed-point theorem. It explains why an invertible derivative controls a nearby nonlinear equation.

Set A=DF(a)A=DF(a). Solving

F(x)=yF(x)=y

is equivalent to finding a fixed point of

Ty(x)=x+A−1(y−F(x)).T_y(x)=x+A^{-1}(y-F(x)).

Its derivative is

DTy(x)=I−A−1DF(x).DT_y(x)=I-A^{-1}DF(x).

At x=ax=a, this derivative is zero. By continuity of DFDF, a sufficiently small closed ball B‾(a,r)⊂U\overline B(a,r)\subset U can be chosen so that

∥I−A−1DF(x)∥≤q<1\|I-A^{-1}DF(x)\|\le q<1

throughout the ball, using the operator norm induced by the Euclidean norm. The mean value estimate then gives

∥Ty(x)−Ty(z)∥≤q∥x−z∥.\|T_y(x)-T_y(z)\|\le q\|x-z\|.

Thus TyT_y is a contraction.

If yy is sufficiently close to F(a)F(a), then

∥A−1(y−F(a))∥<(1−q)r,\|A^{-1}(y-F(a))\|<(1-q)r,

which ensures that TyT_y maps the closed ball into itself. The fixed-point theorem produces a unique solution x=G(y)x=G(y) inside the ball.

The same estimates show that this solution depends locally in a Lipschitz-continuous way on yy. To obtain differentiability, write x=G(y)x=G(y) and

h=F(x+k)−F(x)=DF(x)k+o(∥k∥),h=F(x+k)-F(x)=DF(x)k+o(\|k\|),

where k=G(y+h)−G(y)k=G(y+h)-G(y). The Lipschitz estimate gives ∥k∥=O(∥h∥)\|k\|=O(\|h\|), so

G(y+h)−G(y)=[DF(G(y))]−1h+o(∥h∥).G(y+h)-G(y) =[DF(G(y))]^{-1}h+o(\|h\|).

This proves the derivative formula. Its continuity follows from continuity of DFDF, continuity of GG, and continuity of matrix inversion on invertible matrices.

Once differentiability of the inverse is established, the same formula also follows directly from the chain rule applied to F∘G=id⁡F\circ G=\operatorname{id}.

Examples and limitations

An invertible nonlinear coordinate map

Consider

F(x,y)=(x+y,x−y+x2).F(x,y)=(x+y,x-y+x^2).

Its Jacobian is

DF(x,y)=(111+2x−1),det⁡DF(x,y)=−2−2x.DF(x,y)= \begin{pmatrix} 1&1\\ 1+2x&-1 \end{pmatrix}, \qquad \det DF(x,y)=-2-2x.

The theorem guarantees a differentiable local inverse at every point with x≠−1x\ne-1.

At the origin,

DF(0,0)=(111−1),DF(0,0)= \begin{pmatrix} 1&1\\ 1&-1 \end{pmatrix},

so the derivative of the local inverse at (0,0)(0,0) is

DG(0,0)=(121212−12).DG(0,0)= \begin{pmatrix} \tfrac12&\tfrac12\\ \tfrac12&-\tfrac12 \end{pmatrix}.

No explicit formula for the nonlinear inverse is needed to compute its derivative.

A nonsingular Jacobian does not ensure a global inverse

The map

F(x,y)=(excos⁡y,exsin⁡y)F(x,y)=(e^x\cos y,e^x\sin y)

has

det⁡DF(x,y)=e2x>0\det DF(x,y)=e^{2x}>0

everywhere. Nevertheless,

F(x,y+2π)=F(x,y),F(x,y+2\pi)=F(x,y),

so it is not globally injective. It parametrizes the punctured plane using an exponential radius and a periodic angle, closely related to polar coordinates.

Global inverse theorems require additional assumptions. One standard result states that a C1C^1 map F:Rn→RnF:\mathbb{R}^n\to\mathbb{R}^n with everywhere invertible derivative is a global diffeomorphism if it is proper, meaning that inverse images of compact sets are compact. In this setting, properness is equivalent to

∥F(x)∥⟶∞as∥x∥⟶∞.\|F(x)\|\longrightarrow\infty \quad\text{as}\quad \|x\|\longrightarrow\infty.

Differentiability alone is insufficient

The usual theorem assumes continuous differentiability on a neighborhood, not merely differentiability and an invertible derivative at one point.

For example, define

f(x)={x+2x2sin⁡(1/x),x≠0,0,x=0.f(x)= \begin{cases} x+2x^2\sin(1/x),&x\ne0,\\ 0,&x=0. \end{cases}

Then f′(0)=1f'(0)=1, but for x≠0x\ne0,

f′(x)=1+4xsin⁡(1/x)−2cos⁡(1/x).f'(x)=1+4x\sin(1/x)-2\cos(1/x).

This derivative takes both positive and negative values arbitrarily close to zero. Consequently, ff is not monotone on any interval around zero and is not locally injective there.

Continuous differentiability is a convenient sufficient hypothesis rather than the weakest possible one; more refined versions use other forms of uniform control on the linear approximation.

Relationship to the implicit function theorem

The inverse function theorem is closely related to the implicit function theorem, which describes when an equation

H(x,y)=0H(x,y)=0

locally determines yy as a differentiable function of xx.

Suppose HH takes values in Rm\mathbb{R}^m, with x∈Rnx\in\mathbb{R}^n and y∈Rmy\in\mathbb{R}^m. If the partial derivative DyH(a,b)D_yH(a,b) is invertible, apply the inverse function theorem to

Φ(x,y)=(x,H(x,y)).\Phi(x,y)=(x,H(x,y)).

Its derivative has block form

DΦ=(I0DxHDyH),D\Phi= \begin{pmatrix} I&0\\ D_xH&D_yH \end{pmatrix},

which is invertible precisely when DyHD_yH is invertible. The local inverse of Φ\Phi yields a function y=g(x)y=g(x) satisfying H(x,g(x))=0H(x,g(x))=0, with

Dg(x)=−[DyH(x,g(x))]−1DxH(x,g(x)).Dg(x) =-[D_yH(x,g(x))]^{-1}D_xH(x,g(x)).

Conversely, applying the implicit function theorem to F(x)−y=0F(x)-y=0 yields the inverse function theorem. The two results are therefore equivalent in their standard finite-dimensional C1C^1 forms.

Higher regularity and extensions

Smooth and analytic maps

If FF is CkC^k, for an integer k≥1k\ge1, its local inverse is also CkC^k. If FF is infinitely differentiable, so is the inverse. If FF is a real analytic function, the local inverse is real analytic.

In complex analysis, a holomorphic map between open subsets of Cn\mathbb{C}^n has a holomorphic local inverse wherever its complex Jacobian is invertible. For a function of one complex variable, the condition is simply f′(a)≠0f'(a)\ne0.

Manifolds

For a smooth map F:M→NF:M\to N between manifolds of equal dimension, the coordinate-independent hypothesis is that the differential

dFp:TpM⟶TF(p)NdF_p:T_pM\longrightarrow T_{F(p)}N

is an isomorphism between the corresponding tangent spaces. The conclusion is that FF restricts to a diffeomorphism between neighborhoods of pp and F(p)F(p).

Banach spaces

The theorem extends to maps between Banach spaces. If FF is continuously Fréchet differentiable on an open set and DF(a)DF(a) is a bounded linear isomorphism with bounded inverse, then FF has a continuously Fréchet differentiable local inverse.

Completeness supports the contraction argument. In more general infinite-dimensional spaces, derivative invertibility alone does not automatically give the same theorem.

Applications

The theorem supplies local coordinates in geometry and permits nonlinear equations to be solved locally without explicit solution formulas. It also helps justify changes of variables in multivariable integration: an invertible derivative provides a local coordinate transformation, while the absolute Jacobian determinant describes local volume scaling.

In numerical analysis, the same linearization motivates Newton’s method:

xk+1=xk−[DF(xk)]−1(F(xk)−y).x_{k+1}=x_k-[DF(x_k)]^{-1}(F(x_k)-y).

The inverse function theorem establishes local solvability, but convergence of this particular iteration requires additional conditions on regularity and the initial point.

The inverse derivative also quantifies local sensitivity:

Δx≈[DF(x)]−1Δy.\Delta x\approx[DF(x)]^{-1}\Delta y.

An inverse can exist while being highly sensitive to perturbations. A large norm of the inverse Jacobian, or a large condition number, indicates that small output errors may produce large input errors. Nonsingularity guarantees local invertibility, not numerical robustness.