---
title: Middle-Scale Peak Theorem Overview
url: https://www.emergentmind.com/topics/middle-scale-peak-theorem
type: topic
---

# Middle-Scale Peak Theorem Overview

Searching arXiv for the cited papers and closely related terminology.
“Middle-Scale Peak Theorem” is best understood as an *Editor’s term* for a quantitative near-peaking principle that lies between exact interpolation and unconstrained approximation. In the most direct formulation, the relevant result is Theorem 1 of "Approximation of discrete functions and size of spectrum" [1304.0649]: bounded Paley–Wiener functions that approximately realize Kronecker delta data on a uniformly discrete set force a sharp lower bound on spectral measure. A related, but structurally different, interpolation principle appears in "Pick and Peak Interpolation" [1612.08204], where finite Pick data and peak interpolation data are solved simultaneously in a uniform algebra. Taken together, these results motivate the phrase “middle-scale peak” as denoting regimes of controlled but non-exact peaking.

## 1. Terminological status and scope

The expression “Middle-Scale Peak Theorem” is not the formal title of a theorem in the cited papers. In [1304.0649], the central theorem is a quantitative approximate interpolation result for Paley–Wiener spaces, and the supplied interpretation identifies it as the result most directly relevant to a “middle-scale peak” viewpoint. In [1612.08204], the paper explicitly notes that there is no theorem with that exact title, but identifies Theorem 2.1 as the natural statement if the phrase is intended to describe a result balancing finite Pick data against peak-set constraints.

This suggests that the phrase is best used descriptively rather than bibliographically. In the Paley–Wiener setting, “middle-scale” refers to the regime in which one does not demand exact interpolation of \(\delta\)-data, but also does not permit arbitrary approximation error: the error is fixed in \(l^2\), and that quantitative tolerance still imposes a definite spectral cost [1304.0649]. In the uniform-algebra setting, the phrase points to a theorem that sits between local finite-point interpolation and interpolation on a peak interpolation set, with norm control preserved [1612.08204].

The two settings are not identical. One concerns spectral size versus approximation accuracy for discrete data in \(PW_S\); the other concerns compatibility of Pick interpolation and peak interpolation in abstract uniform algebras. Their common feature is a controlled peaking mechanism that is neither purely exact nor wholly unconstrained.

## 2. Paley–Wiener formulation of the near-peaking principle

The principal theorem associated with the term is stated for a compact set \(S \subset \mathbb R\) and a uniformly discrete set \(\Lambda\), where
\[
\inf_{\lambda\neq\lambda'\in\Lambda}|\lambda-\lambda'|>0.
\]
The Paley–Wiener space is
\[
PW_S=\{f\in L^2(\mathbb R): \widehat f=0 \text{ on } \mathbb R\setminus S\}.
\]

The hypothesis is that for every \(\lambda\in\Lambda\) there exists a bounded function \(f_\lambda\in PW_S\) such that
\[
\|f_\lambda|_\Lambda-\delta_\lambda\|_{l^2(\Lambda)}<d,
\]
with \(0<d<1\), and that the family is uniformly bounded in \(L^2(\mathbb R)\):
\[
\sup_{\lambda\in\Lambda}\|f_\lambda\|_{L^2(\mathbb R)}<\infty.
\]
Under these assumptions, Theorem 1 gives the sharp inequality
\[
\operatorname{meas}(S)\ge 2\pi(1-d^2)D^+(\Lambda),
\]
where the upper uniform density is
\[
D^+(\Lambda)=\lim_{r\to\infty}\max_{x\in\mathbb R}\frac{\operatorname{card}(\Lambda\cap (x,x+r))}{r}.
\]
This is the paper’s main “size of spectrum versus approximation accuracy” inequality [1304.0649].

The content is quantitative. The denser the set \(\Lambda\) and the smaller the allowed approximation error \(d\), the larger the spectral measure must be. The factor \(1-d^2\) interpolates continuously between exact interpolation and weaker approximate interpolation. In particular, when \(d=0\), the theorem recovers the classical necessary density condition
\[
\operatorname{meas}(S)\ge 2\pi D^+(\Lambda).
\]

Conceptually, this is why the result is naturally described as “middle-scale.” It is not merely about exact interpolation, and it is not about approximation without structure; it concerns quantitative near-peaking of \(\delta\)-data by spectrally constrained functions. The supplied interpretation places it between classical interpolation theory and uncertainty-principle or large-sieve type bounds [1304.0649].

## 3. Density mechanism and proof architecture

The proof in [1304.0649] combines a Landau-type concentration argument with a finite-dimensional perturbation lemma, identified in the summary as Lemma 2, in a Kolmogorov-width style form. The initial device is to multiply the approximants \(f_\lambda\) by a small smoothing factor \(\varphi\in PW_{[-\varepsilon,\varepsilon]}\) satisfying \(\varphi(0)=1\). This preserves the approximation properties while enlarging the spectrum only slightly, to
\[
S(\varepsilon)=S+[-\varepsilon,\varepsilon].
\]

One then restricts attention to a finite interval
\[
I=(a-r,a+r)
\]
and considers the vectors of values of the approximants at the points \(\Lambda\cap I\). The perturbation lemma yields a large-dimensional subspace on which these value vectors satisfy a uniform lower quadratic bound. From this, one obtains a subspace of \(PW_{S(\varepsilon)}\) concentrated on a slightly larger interval \(I'\).

Landau’s concentration lemma then imposes an upper bound on the dimension of such a subspace:
\[
\dim X \le \frac{2\pi\, m(I')\, m(S(\varepsilon))}{1-\varepsilon}.
\]
This is compared with the lower-dimensional estimate
\[
\dim X \gtrsim (1-a^2d^2)\operatorname{card}(\Lambda\cap I).
\]
After passing to the limit \(r\to\infty\), one obtains the density inequality
\[
\operatorname{meas}(S)\ge 2\pi(1-d^2)D^+(\Lambda).
\]

The significance of this proof architecture is that it converts local approximate interpolation data into a global spectral obstruction. The lower bound arises from the existence of many approximately independent peaking vectors, while the upper bound arises from concentration. This suggests that the theorem belongs to a broader class of arguments in which finite-dimensional geometry and concentration estimates are used to quantify uncertainty-type phenomena.

## 4. Sharpness and extremal realization

The estimate in [1304.0649] is sharp for every \(d\in(0,1)\). The explicit extremal example takes
\[
\Lambda=\mathbb Z,\qquad S=[-a,a],
\]
and defines
\[
f_j(x)=\frac{\sin a(x-j)}{a(x-j)}\in PW_{[-a,a]}.
\]
These functions satisfy
\[
\|f_j|_{\mathbb Z}-\delta_j\|_{l^2(\mathbb Z)}^2=1-\frac{a}{\pi}.
\]
Since
\[
D^+(\mathbb Z)=1,\qquad \operatorname{meas}([-a,a])=2a,
\]
one obtains equality in
\[
\operatorname{meas}(S)=2\pi(1-d^2)D^+(\Lambda).
\]

This example rules out any general improvement of the coefficient \(2\pi(1-d^2)\). It also clarifies the geometric meaning of the theorem: approximate peaking on a lattice can be achieved exactly up to the limit permitted by the spectral interval, and the theorem captures that limit with equality.

A common misunderstanding would be to treat the factor \(1-d^2\) as a technical artifact of the proof. The extremal example shows that it is structural rather than accidental. The dependence on \(d\) is not merely qualitative; it is optimal in the precise sense asserted by the theorem [1304.0649].

## 5. Moderate growth extension and threshold behavior

The paper also proves an analogue in a moderate-growth regime. If the approximating functions satisfy
\[
\|f_\lambda\|_{L^2(\mathbb R)}\le C\,e^{|\lambda|^\gamma},\qquad 0<\gamma<1,
\]
then the same lower bound persists with the upper Beurling density
\[
D^*(\Lambda)=\limsup_{a\to\infty}\frac{\operatorname{card}(\Lambda\cap(-a,a))}{2a},
\]
namely
\[
\operatorname{meas}(S)\ge 2\pi(1-d^2)\,D^*(\Lambda).
\]
This extends the bounded-family result beyond uniform \(L^2\)-boundedness while preserving the same dependence on approximation error [1304.0649].

The growth restriction is also sharp in the sense recorded in the summary. If the norms are allowed to grow exponentially like \(e^{c|\lambda|}\), then no lower bound on \(\operatorname{meas}(S)\) is possible: there exist compact spectra of arbitrarily small, even zero, measure supporting such approximations. The paper further remarks that the same argument works for any non-quasianalytic growth assumption, while quasianalytic growth is expected to be the threshold for failure.

This sharp threshold behavior is central to the “middle-scale” interpretation. The theorem is not only sensitive to approximation accuracy; it is also sensitive to how large the approximating family is allowed to become. A plausible implication is that controlled near-peaking requires two simultaneous resources: spectral bandwidth and a growth regime below the quasianalytic threshold.

## 6. Relation to peak interpolation in uniform algebras

A distinct but related use of “peak” appears in [1612.08204]. There, \(A\) is a uniform algebra on a compact space \(X\): a closed subalgebra of \(C(X)\) that contains the constants and separates points. A function \(g\in A\) is said to peak on \(E\subset X\) if
\[
g=1 \text{ on } E,\qquad |g|<1 \text{ on } X\setminus E.
\]
A set \(E\subset X\) is a peak interpolation set if for every nonzero \(f\in C(E)\) there exists \(F\in A\) such that \(F|_E=f\) and \(|F(x)|<\|f\|\) on \(X\setminus E\).

Theorem 2.1 of [1612.08204] states that if \(E\subset X\) is a peak interpolation set, if \(a_1,\dots,a_n\in X\setminus E\), if \(f\in C(E)\) with \(|f|\le 1\), and if for every \(\varepsilon>0\) there exists \(F\in A\) with
\[
\|F\|<1+\varepsilon,\qquad F(a_j)=w_j,\quad j=1,\dots,n,
\]
then one can choose such an interpolant \(F\) so that also
\[
F|_E=f.
\]
In the disc algebra \(A(\mathbb D)\), the corresponding corollary asserts that when \(E\subset \partial\mathbb D\) is closed and of \(1\)-dimensional Lebesgue measure zero, simultaneous interpolation on \(E\) and at distinct interior points \(z_1,\dots,z_n\in\mathbb D\) is possible with \(\|F\|_\infty\le 1+\varepsilon\) if and only if the Pick matrix is positive semidefinite [1612.08204].

The proof has two conceptual steps. First, an approximate interpolant is built that matches the peak-set data exactly and the finite Pick data approximately. The key lemma provides a sequence \((f_m)\subset A\) with \(\|f_m\|<1\), \(f_m=0\) on \(E\), and \(f_m\to 1\) pointwise on \(X\setminus E\). Second, the remaining finite interpolation errors are corrected exactly using functions \(p_1,\dots,p_n\in A\) satisfying
\[
p_j(a_j)=1,\qquad p_j(E)=0,\qquad p_j(a_k)=0\ (k\ne j).
\]

The connection to the Paley–Wiener theorem is analogical rather than formal. In both cases, a peak-type device localizes or suppresses unwanted behavior, and a finite correction or density argument then restores exact control. The difference is substantive: [1304.0649] yields a quantitative lower bound on spectral size from approximate \(\delta\)-interpolation, whereas [1612.08204] proves compatibility of two interpolation mechanisms under approximate norm control. This suggests that “middle-scale peak” names a family resemblance across settings, not a single universal theorem.

## 7. Conceptual significance and common misconceptions

The central significance of the Paley–Wiener result is that approximate interpolation remains rigid. Allowing a fixed \(l^2\)-error \(d\) does not eliminate density obstructions; it modifies them quantitatively through the sharp factor \(1-d^2\) [1304.0649]. The theorem therefore gives a precise answer to how much spectral measure is required for controlled near-peaking on a dense discrete set.

One common misconception is to equate the theorem with exact interpolation theory. The theorem does recover the exact case when \(d=0\), but its distinctive content lies in the nonzero-error regime. Another misconception is to identify it with peak interpolation in the sense of uniform algebras. The latter is a different theory, despite the shared language of peaking. In [1612.08204], the issue is simultaneous solvability of finite Pick data and peak-set data with norm control; in [1304.0649], the issue is spectral size forced by approximate delta interpolation.

A further interpretive point concerns the phrase “middle-scale.” In the supplied summaries, that phrase is explanatory rather than canonical. It indicates a regime between exact local interpolation and fully uncontrolled approximation, or between finite-point Pick interpolation and peak-set interpolation. Used in that restricted sense, it accurately captures the role of the principal theorems. Used as a formal historical label, however, it would be misleading, because neither paper presents that exact title as part of its theorem nomenclature.

Source: https://www.emergentmind.com/topics/middle-scale-peak-theorem