Approximation of sgn(x) on two symmetric intervals by rational functions with fixed poles

Abstract: Recently A. Eremenko and P. Yuditskii found explicit solutions of the best polynomial approximation problems of sgn(x) over two intervals in terms of conformal mappings onto special comb domains. We give analogous solutions for the best approximation problems of sgn(x) over two symmetric intervals by odd rational functions with fixed poles. Here the existence of the related conformal mapping is proved by using convexity of the comb domains along the imaginary axis.

• The paper establishes the existence of a unique conformal mapping that precisely characterizes the best uniform rational approximation of sgn(x) on symmetric intervals.
• It derives an explicit error formula via comb domain techniques and Herglotz-type integral representations, extending classical Chebyshev methods.
• The methodology generalizes polynomial approximation theory, offering pathways for efficient numerical algorithms and applications in engineering and theoretical computer science.

Approximation of $\operatorname{sgn}(x)$ on Two Symmetric Intervals by Rational Functions with Fixed Poles

Introduction and Background

This work addresses the uniform rational approximation of the sign function, $\operatorname{sgn}(x)$, on two symmetric intervals, specifically $|x| \in [a,1]$, by odd rational functions with prescribed (fixed) poles. The problem is a natural generalization of classical Chebyshev-type best approximation questions, extending from polynomials to rational functions. The context is set within the rich history of extremal problems in approximation theory, where conformal mapping techniques have played a central role since Chebyshev's original solution for best uniform polynomial approximation on $[-1,1]$.

Prior solutions—such as those by Akhiezer and, more recently, Eremenko and Yuditskii—have relied on explicit construction of such mappings, often to so-called “comb domains.” Eremenko and Yuditskii provided polynomial analogues for $\operatorname{sgn}(x)$ on symmetric intervals, leveraging conformal mappings onto comb domains to characterize best approximating polynomials. This paper extends those techniques to the more general and technically delicate setting of rational approximants with fixed poles.

Main Results

Existence and Construction of Conformal Mappings

The central technical contribution lies in the formulation and proof of existence of conformal mappings from the first quadrant onto parameterized comb domains with specifically constructed boundaries tailored for the rational approximation problem. Unlike the polynomial setting, the rational case introduces new constraints, notably stemming from the locations and multiplicities of fixed poles (prescribed zeros and poles off the intervals of approximation). The comb domain’s structure—formed by slitting a vertical strip with rays positioned according to the chosen poles—is more intricate and demands careful analysis, particularly on the imaginary axis where convexity arguments are employed.

Theorem 1: There exists a unique vector of accessory parameters—corresponding to “heights” of the slits in the comb domain—such that the conformal map realizes a correspondence between pre-specified points in the $z$-plane and canonical locations (tips of slits, intervals, points at infinity) in the mapped domain. This result guarantees the existence of the extremal rational function for the best uniform approximation.

The explicit construction of the conformal mapping is achieved via an integral representation rooted in the theory of functions convex in the direction of the imaginary axis. The crucial reduction to a boundary value problem for a function with Herglotz-type integral representation brings the problem into a setting where uniqueness and existence can be rigorously established. The existence proof involves approximation by piecewise-constant measures, compactness arguments, and continuity properties of the mapping.

Extremal Rational Functions and Error Characterization

Theorem 2: The best uniform approximation error is given explicitly as $L = \cosh B_0(x_1, \ldots, x_p; a)$, where $B_0$ is determined by the aforementioned conformal mapping parameters. The corresponding extremal rational function possesses the required symmetry (oddness), prescribed pole structure, and alternation properties demanded by the generalized Chebyshev alternation theorem. The explicit form of the extremal rational function is constructed via the composition of the conformal mapping and its associated accessory parameters.

The results generalize existing polynomial approximation theory, recovering previous results (e.g., for $p=0$) and introducing new construction techniques for the general fixed-pole rational case. The analyticity and alternation properties are confirmed, and their connection to the underlying conformal domain geometry is made precise.

Analytical and Numerical Implications

The explicit characterization of both the extremal function and the approximation error allows for concrete analytical and, by extension, numerical investigations into best uniform approximations of discontinuous targets (like $\operatorname{sgn}(x)$) by rational functions with controlled pole architecture. Such results have implications for rational approximation theory on multi-interval sets, and contribute to the understanding of the connection between conformal geometry and approximation error in uniform norms.

In particular, the methodology translates extremal problems in rational approximation into problems about the geometry of comb domains, which can then be attacked using integral and variational techniques standard in the theory of univalent and convex functions.

The sharpness and tightness of the error representation are direct consequences of the alternation properties, and the approach provides a platform for explicit computation of best approximating rational functions even for complex multi-interval geometries.

Broader Impacts and Future Directions

The paper’s framework opens the door to a range of further developments:

• Generalization to More Intervals and Pole Configurations: The construction using comb domains with higher complexity can, in principle, accommodate more intricate rational approximation problems, including those involving more intervals and more structured pole assignments.
• Algorithmic Approximation and Fast Numerical Methods: The integral representations and boundary correspondence provide a potential foundation for efficient algorithms to compute best approximating rational functions for applications in engineering and applied mathematics, especially where rational approximation is required for filters and system approximation.
• Applications in Harmonic Analysis and Complexity Theory: As referenced in the literature, the problem of approximating $\operatorname{sgn}(x)$ is closely tied to problems in pseudorandomness and quantum communication complexity, underscoring the relevance of sharp rational approximations in theoretical computer science.
• Further Connections to Conformal Field Theory and Operator Theory: The explicit conformal mapping techniques may prove relevant in geometric function theory and spectral theory, where rational and polynomial extremal functions arise in the analysis of linear operators and potential theory.

Conclusion

The paper provides a comprehensive, explicit solution to the problem of best uniform rational approximation of $\operatorname{sgn}(x)$ over two symmetric intervals by odd rational functions with fixed poles. The core contributions are the existence proof and constructive description of associated conformal mappings onto parameterized comb domains, along with explicit formulas for the best approximation error and the extremal rational functions. These results extend the toolbox for rational extremal approximation with controlled pole placement, firmly rooting the methods in the interplay between complex analysis, geometric function theory, and approximation theory (1411.6923).