Real-valued Lipschitz functions and metric properties of functions

https://doi.org/10.1016/j.jmaa.2020.123839Get rights and content

Abstract

The purpose of this article is to explore the very general phenomenon that a function between metric spaces has a particular metric property if and only if whenever it is followed in a composition by an arbitrary real-valued Lipschitz function, the composition has this property. The key tools we use are the Efremovič lemma [21] and a theorem of Garrido and Jaramillo [22] that says that a function h between metric spaces is Lipschitz if and only if whenever it is followed by a Lipschitz real-valued function in a composition, the composition is Lipschitz. We also present a streamlined proof of the Garrido-Jaramillo result itself, but one that still relies on their natural continuous linear operator from the Lipschitz space for the target space to the Lipschitz space for the domain. Separately, we include a highly applicable uniform closure theorem that yields the most important uniform density theorems for Lipschitz-type functions as special cases.

Introduction

Analysts have little interest in topological spaces that fail to satisfy the Hausdorff separation property. They would also prefer to work in the framework of completely regular Hausdorff spaces, also called Tychonoff spaces. For one thing, in this context, such spaces have Hausdorff compactifications; in fact a topological space can be embedded in a compact Hausdorff space if and only if it is Tychonoff. Similarly, only in this framework does one have a compatible separated diagonal uniformity, so that notions such as completeness and uniform continuity can be formulated. They subsume the locally convex topological vector spaces and the locally compact Hausdorff spaces. But perhaps most importantly, it is only in this setting that the real-valued continuous functions determine the topology.

Given a family of real-valued functions {fi:iI} defined on a nonempty set Y, the initial topology τ{fi:iI} on Y is the topology generated by the family of subsets {fi1(V):iIandVis open in the real line}. This is the coarsest topology one can place on Y such that each fi is continuous. A function h from a second topological space into Y equipped with such an initial topology is continuous if and only if for each iI,fih is continuous [40, Theorem 8.10].

As an example, if Y,|||| is a real normed linear space, the initial topology on Y determined by its continuous dual is called the weak topology on Y. This will agree with the norm topology if and only if the space is finite dimensional.

Now given a topological space Y,τ and a family of continuous real-valued functions {fi:iI} on Y, the initial topology determined by the family is in general coarser than τ. The two topologies agree provided the family separates points from closed sets, which means that whenever A is a τ-closed subset of Y and pY\A, there exists some fi such that fi(p) lies outside of the closure of f(A) [40, Corollary 8.15]. The standard definition of complete regularity [40, pp. 94-95] implies that the continuous functions from Y,τ to [0,1] separate points from closed sets, so any larger class of continuous functions will do so as well. In particular, a function h into a completely regular space Y,τ will be continuous if and only if whenever f is continuous and real-valued, fh is continuous.

Now if Y,ρ is a metric space with induced metric topology τρ, then the family of all Lipschitz functions on Y separates points from closed sets because the family of all distance functionals yρ(y,p) where p runs over Y already does. In particular, if h:X,dY,ρ is a function, then h is continuous if and only if whenever f is a real-valued Lipschitz function on Y, fh is continuous. The purpose of this article is show that most of the important classes of functions between metric spaces behave analogously, including the uniformly continuous functions, the Lipschitz functions, the locally Lipschitz functions, the functions that map Cauchy sequences to Cauchy sequences, and the class of (uniformly continuous) functions that are Lipschitz in the small as introduced by Luukkainen [33]: there exist δ>0 and λ>0 such that whenever d(x1,x2)<δ in X, we have ρ(h(x1),h(x2))λd(x1,x2).

Evidently, a bounded Lipschitz in the small function h:XY with parameters δ and λ is already Lipschitz, for if the diameter of h(X) is M and d(x1,x2)δ, then ρ(h(x1),h(x2))Mδd(x1,x2). If we replace the metric ρ on the target space Y by ρ˜=min{ρ,1}, then h is Lipschitz in the small if and only if h regarded as a function from X to Y,ρ˜ is Lipschitz [23, p. 283]. For real-valued functions, the Lipschitz in the small functions are particularly important in that such functions are uniformly dense in the real-valued uniformly continuous functions [10], [23]. At the end of the article, we give a uniform closure result for the real-valued continuous functions that are Lipschitz in the small when restricted to a particular family of subsets, provided the family is shielded from closed sets. From this single result, we can easily identify uniformly dense subclasses of Lipschitz-type functions within the most important classes of continuous functions. Our results argue for the primacy of the Lipschitz in the small functions in the study of real-valued continuous functions on metric spaces.

Along the way, we introduce and study the class of (locally Lipschitz) functions that are Lipschitz when restricted to Bourbaki bounded subsets.

Section snippets

Preliminaries

All metric spaces are assumed to contain at least two points. We denote the set of limit points of the space X,d by X. We write Sd(x,ε) for the open ball of center x and radius ε>0 in X. We write Sd(A,ε) for aASd(a,ε); this set is often called the ε-enlargement of A [5]. For xX and A a nonempty subset of X,d, we put d(x,A):=inf{d(x,a):aA}; clearly, Sd(A,ε)={xX:d(x,A)<ε} for A nonempty.

We denote the usual d-diameter of a nonempty subset A of X by diamd(A). We call nonempty subsets A

Results

Our first result, included for completeness, lacks any depth.

Theorem 3.1

Let X,d and Y,ρ be metric spaces and let h:XY. Then h is bounded if and only if whenever fLip(Y,R), the composition fh is bounded.

Proof

If h bounded and is followed by a Lipschitz function in composition, then we get a bounded function, as a Lipschitz function on Y is bounded when restricted to any bounded subset of Y. Conversely, if the range of h is not bounded, fixing y0 in Y, let f=ρ(,y0)Lip(Y,R). As f restricted to h(X) is

The case of Lipschitz in the small functions

In view of Example 2.4, there is no hope of using Proposition 3.8 to provide a parallel theorem for the class of Lipschitz in the small functions. Nevertheless, we can prove that this property is determined by compositions with real-valued Lipschitz functions using an argument that directly invokes Theorem 3.7 upon replacing the metric on the target space by a bounded metric.

Theorem 4.1

A function h from X,d to Y,ρ is Lipschitz in the small if and only if whenever fLip(Y,R), the function fh is

A meta-theorem

Our penultimate section is presented on a more abstract level, in an attempt to pin down the essential aspects of a class of continuous functions between metric spaces that is determined by real-valued Lipschitz functions. Let Ω be a class of functions between metric spaces. For a given pair of metric spaces (X,d,Y,ρ) we put Ω(X,Y):={f:XY:fΩ}.

Definition 5.1

We call a class of functions Ω between metric spaces real Lipschitz determined if it satisfies the following three properties:

  • (1)

    for all metric spaces X

A uniform closure theorem with applications

Let B be a family of nonempty subsets of a metric space X,d. In this section we will show thatB:={fC(X,R):BB,f|BLipschitz in the small} is uniformly dense inΩB:={fC(X,R):BB,f|Bis uniformly continuous} under the well-studied assumption that B be shielded from closed sets [7], [15]. From this result, we can derive the most important uniform density results for Lipschitz-type functions that we know of in the literature. Taken as a whole, the results of this section show conclusively that

References (40)

  • M. Atsuji

    Uniform continuity of continuous functions of metric spaces

    Pacific J. Math.

    (1958)
  • G. Beer

    More about metric spaces on which continuous functions are uniformly continuous

    Bull. Aust. Math. Soc.

    (1986)
  • G. Beer

    Topologies on Closed and Closed Convex Sets

    (1993)
  • G. Beer

    McShane's extension theorem revisited

    Vietnam J. Math.

    (2020)
  • G. Beer et al.

    Bornological convergence and shields

    Mediterr. J. Math.

    (2013)
  • G. Beer, L. García-Lirola, M.I. Garrido, Stability of Lipschitz-type functions under pointwsie product and...
  • G. Beer et al.

    Bornologies and locally Lipschitz functions

    Bull. Aust. Math. Soc.

    (2014)
  • G. Beer et al.

    Uniform continuity and a new bornology for a metric space

    Set-Valued Var. Anal.

    (2018)
  • G. Beer et al.

    Uniform continuity, uniform convergence, and shields

    Set-Valued Var. Anal.

    (2010)
  • J. Borsik

    Mappings preserving Cauchy sequences

    Čas. Pěst. Mat.

    (1988)
  • Cited by (7)

    • Split Lipschitz-type functions

      2024, Topology and its Applications
    • On some Lipschitz-type functions

      2023, Journal of Mathematical Analysis and Applications
    • Reciprocation and Pointwise Product in Vector Lattices of Functions

      2023, Springer Proceedings in Mathematics and Statistics
    • COFINALLY COMPLETE METRIC SPACES AND RELATED FUNCTIONS

      2023, Cofinally Complete Metric Spaces and Related Functions
    • Bornologies and Lipschitz Analysis

      2023, Bornologies and Lipschitz Analysis
    View all citing articles on Scopus

    The first author thanks Universidad Complutense for its hospitality in September 2019. The second author was partially supported by DGES grant PGC2018-097286-B-I00 (Spain).

    View full text