Lipschitz continuity · aiwiki.page

Lipschitz continuity is a regularity condition in mathematical analysis that imposes a uniform bound on how rapidly a function can change. Distances between output values must not exceed a fixed finite multiple of distances between the corresponding inputs. The condition makes sense between arbitrary metric spaces, without requiring derivatives or a linear structure, and provides quantitative control stronger than ordinary continuity. (math.uchicago.edu)

Definition and Lipschitz constants

Let $(X,d_X)$ and $(Y,d_Y)$ be metric spaces. A map $f:X\to Y$ is Lipschitz continuous if a constant $L\geq0$ exists such that

d_Y(f(x),f(y))\leq L\,d_X(x,y) \qquad\text{for all }x,y\in X.

Such a map is called $L$ -Lipschitz. In normed vector spaces, the distances can be expressed using a norm:

\|f(x)-f(y)\|\leq L\|x-y\|.

The choice of distances or norms is part of the definition. (math.uchicago.edu)

Any constant satisfying the inequality is a Lipschitz constant. The smallest permissible constant, often denoted $\operatorname{Lip}(f)$ , is

\operatorname{Lip}(f)= \sup_{x\ne y}\frac{d_Y(f(x),f(y))}{d_X(x,y)}

when the domain contains at least two points. Constant functions have optimal constant zero. For real-valued functions of one variable, this supremum bounds the absolute slopes of all secant lines, rather than merely slopes at differentiable points. (math.rice.edu)

Relationship to other continuity conditions

Every Lipschitz map has uniform continuity and therefore is a continuous function. For $L>0$ , choosing $\delta=\varepsilon/L$ establishes the uniform-continuity condition directly. The converse fails: $f(x)=\sqrt{x}$ is uniformly continuous on $[0,1]$ , but its difference quotient against zero is $1/\sqrt{x}$ , which is unbounded near zero. These conclusions follow directly from the definitions and the displayed quotient. (math.uchicago.edu)

Local Lipschitz continuity requires a Lipschitz bound on some neighborhood of each point; the constants may vary between neighborhoods. A global condition instead uses one constant for the entire domain. The polynomial $f(x)=x^2$ is locally Lipschitz on $\mathbb R$ , and is $2R$ -Lipschitz on $[-R,R]$ , but is not globally Lipschitz because

|x^2-y^2|=|x+y|\,|x-y|.

Thus even smooth functions need not satisfy a global bound on an unbounded domain. (math.ucdavis.edu)

Examples and differentiation

The function $f(x)=|x|$ is $1$ -Lipschitz by the triangle inequality, yet has no derivative at zero. This shows that Lipschitz continuity does not require differentiability everywhere. Conversely, a differentiable real-valued function on an interval with $|f'|\leq L$ is $L$ -Lipschitz by the mean value theorem. Continuous differentiability consequently implies local Lipschitz continuity, but not necessarily a global bound. (math.ucdavis.edu)

In Euclidean space, Rademacher’s theorem states that a Lipschitz map from an open subset of $\mathbb R^n$ into $\mathbb R^m$ is differentiable almost everywhere, with respect to Lebesgue measure. Its differential, where it exists, has operator norm at most its Lipschitz constant. Thus exceptional corners are compatible with substantial differentiability. (web.stanford.edu)

On a compact real interval, Lipschitz continuity implies absolute continuity. More precisely, a real-valued function is Lipschitz exactly when it is absolutely continuous and its almost-everywhere derivative is essentially bounded. The function can then be recovered from that derivative through the Lebesgue integral, and its optimal Lipschitz constant equals the essential supremum of $|f'|$ . (math.ucdavis.edu)

Operations and geometric consequences

Lipschitz constants behave predictably under function composition. If $f:X\to Y$ is $L_f$ -Lipschitz and $g:Y\to Z$ is $L_g$ -Lipschitz, then $g\circ f$ is $L_gL_f$ -Lipschitz. For a bounded linear map between normed spaces, the optimal constant is its operator norm. A norm itself, viewed as a real-valued function on its vector space, is $1$ -Lipschitz. (math.rice.edu)

A map is bi-Lipschitz if constants $0<c\leq C<\infty$ satisfy

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

The lower bound makes it injective and ensures that its inverse on its image is Lipschitz. Such maps preserve distances up to multiplicative factors. Separately, real-valued Lipschitz functions defined on a subset of a metric space can be extended to the whole space without increasing their Lipschitz constant, by the McShane extension theorem. (math.uchicago.edu)

Differential equations and optimization

Lipschitz conditions are central to uniqueness for differential equations. For an initial-value problem $y'=F(t,y)$ , the Picard–Lindelöf theorem gives local existence and uniqueness when $F$ is continuous and locally Lipschitz in the state variable, with the bound locally uniform in time. Its proof uses a contraction mapping on a suitable space of trajectories. Merely continuous right-hand sides need not yield uniqueness. (ocw.mit.edu)

In mathematical optimization, Lipschitz continuity of a function must be distinguished from Lipschitz continuity of its gradient. The latter condition,

\|\nabla f(x)-\nabla f(y)\|\leq L\|x-y\|,

controls variation in the derivative and is commonly called $L$ -smoothness. It provides quadratic bounds on first-order approximation errors, supporting analyses of gradient descent. For example, $x^2$ has a globally Lipschitz gradient although the function itself is not globally Lipschitz. (cs.cornell.edu)