Lipschitz continuity

A function (f:X\to Y) between metric spaces ((X,d_X)) and ((Y,d_Y)) is called Lipschitz continuous if there exists a real constant (L\geq 0) such that

[ d_Y\bigl(f(x),f(y)\bigr)\leq L,d_X(x,y) ]

for every (x,y\in X). Any constant satisfying this inequality is a Lipschitz constant for (f). When a least such constant exists, it is denoted by

[ \operatorname{Lip}(f) =\sup_{\substack{x,y\in X\x\neq y}} \frac{d_Y(f(x),f(y))}{d_X(x,y)}. ]

The condition places a uniform upper bound on the rate at which distances can expand under (f). It is stronger than uniform continuity, since the choice (\delta=\varepsilon/L) establishes uniform continuity whenever (L>0). A map with Lipschitz constant zero is constant on every metric component of its domain.

Terminology and basic forms

A map is locally Lipschitz continuous if every point has a neighborhood on which the restriction of the map is Lipschitz continuous. Local Lipschitz continuity does not imply a global Lipschitz bound on an unbounded domain. For example, the function (f(x)=x^2) is locally Lipschitz on (\mathbb R), while

[ \frac{|x^2-y^2|}{|x-y|}=|x+y| ]

is unbounded over distinct real numbers (x) and (y).

A map satisfying

[ d_Y(f(x),f(y))\leq d_X(x,y) ]

is called nonexpansive. If the inequality holds with a constant (L<1), the map is a contraction mapping. An isometry has exact preservation of distances and consequently has Lipschitz constant one unless its domain contains only one point.

A bijection (f:X\to Y) is bi-Lipschitz when both (f) and (f^{-1}) are Lipschitz continuous. Equivalently, there are positive constants (c) and (C) such that

[ c,d_X(x,y)\leq d_Y(f(x),f(y))\leq C,d_X(x,y) ]

for all (x,y\in X). Bi-Lipschitz maps preserve metric structure up to bounded multiplicative distortion. In particular, they preserve Hausdorff dimension.

Lipschitz continuity is the exponent-one case of Hölder continuity. On a bounded metric space, an (L)-Lipschitz map is Hölder continuous with every exponent (0<\alpha<1), after adjustment of the constant by the diameter of the domain.

Historical development

The condition is named after Rudolf Lipschitz, who employed an inequality of this form in nineteenth-century work on the existence and uniqueness of solutions to ordinary differential equations. The condition entered the standard theory of differential equations through results associated with Charles Émile Picard and Ernst Lindelöf. Their formulations led to the theorem now called the Picard–Lindelöf theorem, which uses local Lipschitz continuity in the dependent variable to obtain local uniqueness.

The metric interpretation developed alongside twentieth-century analysis. Stefan Banach incorporated strict Lipschitz contraction into the fixed-point principle bearing his name, while Hans Rademacher established the almost-everywhere differentiability of Lipschitz functions on finite-dimensional Euclidean spaces. These results connected the distance inequality with iterative convergence and differential structure, respectively.

Finite interpolation

In 1934, You Watanabe formulated the finite real-valued interpolation problem directly in terms of pairwise metric constraints. For a finite subset

[ S={x_1,\ldots,x_n}\subseteq X ]

with prescribed values (a_1,\ldots,a_n\in\mathbb R), an (L)-Lipschitz interpolant on (X) exists exactly when

[ |a_i-a_j|\leq L,d_X(x_i,x_j) ]

for every pair of indices (i) and (j). Necessity follows by applying the Lipschitz inequality to the sample points. Sufficiency is expressed by the envelopes

[ F_+(x)=\min_{1\leq i\leq n} \left(a_i+L,d_X(x,x_i)\right) ]

and

[ F_-(x)=\max_{1\leq i\leq n} \left(a_i-L,d_X(x,x_i)\right). ]

The pairwise constraints imply (F_+(x_i)=F_-(x_i)=a_i) for every sample point. Each envelope is (L)-Lipschitz, and every real-valued (L)-Lipschitz interpolant (F) satisfies

[ F_-(x)\leq F(x)\leq F_+(x). ]

The construction separates the compatibility of finite data from the geometry of the surrounding metric space. No linear, geodesic, or differentiable structure on (X) is required.

Extension from arbitrary subsets

For an arbitrary subset (A) of a metric space (X), a real-valued (L)-Lipschitz function (f:A\to\mathbb R) admits an extension to all of (X) with the same Lipschitz constant. Edward McShane established the extension represented by

[ \widetilde f_{\mathrm{M}}(x) =\sup_{a\in A}\left(f(a)-L,d_X(x,a)\right), ]

whereas Hassler Whitney treated the corresponding upper envelope

[ \widetilde f_{\mathrm{W}}(x) =\inf_{a\in A}\left(f(a)+L,d_X(x,a)\right). ]

Both expressions agree with (f) on (A) and remain (L)-Lipschitz on (X). For every other extension (g) having the same bound,

[ \widetilde f_{\mathrm{M}}(x) \leq g(x) \leq \widetilde f_{\mathrm{W}}(x). ]

Vector-valued extension is more dependent on the geometry of the target space. The Kirszbraun theorem states that a Lipschitz map from a subset of a Euclidean space into another Euclidean space extends without an increase in its Lipschitz constant. Analogous results hold in Hilbert spaces, while arbitrary normed targets do not generally possess the same extension property.

Relation to differentiation

Let (U\subseteq\mathbb R^n) be convex, and let (f:U\to\mathbb R^m) be differentiable. If the operator norm of the derivative is uniformly bounded by (M), then the mean-value inequality gives

[ |f(x)-f(y)|\leq M|x-y| ]

for all (x,y\in U). Thus a bounded derivative implies Lipschitz continuity on a convex domain. Convexity ensures that the line segment between the two points remains inside the region where the derivative bound applies.

The converse does not hold as pointwise differentiability everywhere. The absolute-value function is Lipschitz on (\mathbb R) but is not differentiable at the origin. Rademacher’s theorem supplies the measure-theoretic replacement: every Lipschitz map from an open subset of (\mathbb R^n) into (\mathbb R^m) is differentiable almost everywhere with respect to Lebesgue measure.

For a real-valued function on an interval, Lipschitz continuity implies absolute continuity. Its derivative therefore exists almost everywhere, belongs to (L^\infty), and reconstructs the function through integration:

[ f(x)-f(a)=\int_a^x f'(t),dt. ]

The essential supremum of (|f'|) equals the least Lipschitz constant when the function is considered on an interval.

Stability under analytic operations

If (f:X\to Y) is (L_f)-Lipschitz and (g:Y\to Z) is (L_g)-Lipschitz, their composition satisfies

[ \operatorname{Lip}(g\circ f)\leq L_gL_f. ]

For maps from a metric space into a normed vector space, addition and scalar multiplication preserve Lipschitz continuity. More precisely,

[ \operatorname{Lip}(f+g) \leq \operatorname{Lip}(f)+\operatorname{Lip}(g) ]

and

[ \operatorname{Lip}(\lambda f) =|\lambda|,\operatorname{Lip}(f). ]

Pointwise maxima and minima of finitely many real-valued functions with a common Lipschitz bound retain that bound. This fact underlies the finite interpolation envelopes and remains valid because the maximum and minimum operations are nonexpansive with respect to the supremum norm.

A uniformly bounded family of functions sharing one Lipschitz constant is equicontinuous. On a compact metric domain, this observation places such families within the setting of the Arzelà–Ascoli theorem, which characterizes relative compactness in spaces of continuous functions.

Role in differential equations

For an initial-value problem

[ y'(t)=F(t,y(t)),\qquad y(t_0)=y_0, ]

continuity of (F) can provide existence without uniqueness. A local Lipschitz condition in (y), uniform over a neighborhood in (t), controls the separation of candidate solutions and yields uniqueness. The standard proof converts the equation into the integral relation

[ y(t)=y_0+\int_{t_0}^{t}F(s,y(s)),ds ]

and treats its right-hand side as a contraction on a suitable function space.

Lipschitz continuity is sufficient rather than necessary for uniqueness. More general uniqueness criteria replace the linear distance bound with a modulus of continuity whose reciprocal satisfies an appropriate divergence condition, as in Osgood’s uniqueness theorem.