[Paper Review] Estimates for covering numbers in Schauder's theorem about adjoints of compact operators

Michael Ćwikel, Eliahu Levy|ArXiv.org|Oct 23, 2008

Advanced Banach Space Theory9 references3 citations

TL;DR

This paper establishes new quantitative estimates for covering numbers of the image of the closed unit ball under the adjoint of a bounded linear operator between Banach spaces, linking them to covering numbers of the original operator's image or a significant subset. The key contribution is a best-possible quantitative version of Schauder's theorem on adjoints of compact operators, derived via a generalized abstract compactness result that extends Arzelà-Ascoli and Schauder theorems.

ABSTRACT

Let T:X --> Y be a bounded linear map between Banach spaces X and Y. Let S:Y' --> X' be its adjoint. Let B(X) and B(Y') be the closed unit balls of X and Y' respectively. We obtain apparently new estimates for the covering numbers of the set S(B(Y')). These are expressed in terms of the covering numbers of T(B(X)), or, more generally, in terms of the covering numbers of a "significant" subset of T(B(X)). The latter more general estimates are best possible. These estimates follow from our new quantitative version of an abstract compactness result which generalizes classical theorems of Arzela-Ascoli and of Schauder. Analogous estimates also hold for the covering numbers of T(B(X)), in terms of the covering numbers of S(B(Y')) or in terms of a suitable "significant" subset of S(B(Y')).

Motivation & Objective

• To provide quantitative estimates for covering numbers of the adjoint operator's image in Banach spaces.
• To generalize classical results of Arzelà-Ascoli and Schauder into a unified abstract framework with quantitative control.
• To establish best-possible estimates for covering numbers by relating them to significant subsets of the operator's image.
• To extend the duality between covering numbers of an operator and its adjoint, without dimensional dependence in the bounds.

Proposed method

• The authors introduce a generalized semimetric framework based on a function h:A×B→ℂ satisfying uniform boundedness on fixed slices.
• They define semimetrics d_A and d_B on sets A and B, respectively, using suprema of differences in h.
• Covering numbers N_E(ρ) are defined using closed sets of diameter ≤ρ, enabling precise control over metric entropy.
• A key step involves proving that total boundedness of (A,d_A) is equivalent to that of (B,d_B), forming the core of the abstract compactness result.
• The method applies this equivalence to Banach space operators, relating covering numbers of T(ℬ_X) and T*(ℬ_Y*) via duality.
• The proof leverages convexity and continuity of seminorms to derive bounds on covering numbers, with a critical use of interval covering in a line segment argument.

Experimental results

Research questions

• RQ1 Can covering numbers of the adjoint operator's image be estimated in terms of covering numbers of the original operator's image?
• RQ2 What is the best possible quantitative bound relating covering numbers of an operator and its adjoint?
• RQ3 How can the classical Schauder theorem on adjoints of compact operators be strengthened with explicit, optimal estimates?
• RQ4 Can covering numbers be estimated using only a 'significant' subset of the operator's image rather than the full image?
• RQ5 Is it possible to derive such estimates without explicit dependence on the dimension of the underlying space?

Key findings

• The paper establishes that the covering number N_{T^*(ℬ_{Y^*})}(ρ) is bounded above by a function of N_{T(ℬ_X)}(ρ) and a significant subset of T(ℬ_X), with the bound being best possible.
• For any ε > 0, the covering number N_A(ε) for the set A in the dual framework satisfies N_A(ε) ≤ C·N_B(ε) for a universal constant C, with equality conditions derived via interval covering in a line segment.
• The estimate in Theorem 4 shows that if A is absolutely convex, then N_A(ε) ≤ n·ε, where n is the covering number of a finite ε-net, leading to a sharp bound on the covering number in terms of the diameter and structure of the set.
• The authors demonstrate that N_B(ρ) = ∞ for all ρ < 1 in a specific example, showing that the bounds are tight and cannot be improved without additional assumptions.
• The paper proves that N_B^Δ(ρ) = ∞ for all ρ < 1/2 in the same example, confirming the sharpness of the covering number estimates in the dual setting.
• The results are independent of dimension, distinguishing them from recent estimates by Emanuel Milman that involve dimensional factors.