Pointwise behavior of SU(1,1) nonlinear Fourier transform

This paper demonstrates that the SU(1,1) nonlinear Fourier transform can diverge pointwise for square-summable coefficients, thereby proving that classical pointwise asymptotics for orthogonal polynomials on the unit circle may fail even within the Szegő class, while also identifying specific conditions under which convergence is preserved.

The Gist

Imagine you are trying to predict the future of a complex system by looking at a long list of numbers. In mathematics, there is a powerful tool called the Fourier Transform. Think of it as a machine that takes a messy, complicated signal (like a song or a wave) and breaks it down into simple, pure notes. Usually, if your list of numbers is "small enough" (mathematically speaking, "square-summable"), this machine works perfectly: it gives you a clear, stable answer for every single point in time.

For decades, mathematicians believed this stability held true even for a more complicated, "nonlinear" version of this machine, specifically one related to a group called SU(1,1). They had a strong hunch, often called the "Nonlinear Carleson Conjecture," that if you fed this machine a list of numbers that wasn't too wild, it would eventually settle down and give a definite answer at every single point.

The Big Surprise: The Machine Breaks
Sergey A. Denisov's paper delivers a shock to this belief. He proves that this intuition is wrong.

He constructs a very specific, carefully crafted list of numbers that is "small enough" to be considered well-behaved by standard rules. However, when you feed this list into the SU(1,1) machine and try to see what happens at every single point, the machine diverges. It doesn't just get a little noisy; it goes completely haywire. The numbers it spits out bounce around forever and never settle on a final value, not even at a single point.

The Analogy: The Unstable Tower
Imagine you are building a tower out of blocks.

• The Standard Rule: If you have a limited amount of weight (the "square-summable" condition), you should be able to build a tower that stands still.
• The Conjecture: Mathematicians thought that even if the blocks were arranged in a tricky, non-linear way, the tower would still stand still if you just waited long enough.
• Denisov's Discovery: He shows that you can arrange the blocks in a specific, recursive pattern (like a fractal or a "daisy" chain of smaller patterns) where the tower wobbles more and more violently the higher you go. No matter how long you wait, the top of the tower never stops shaking. It never finds a resting place.

What This Means for Other Math
The paper connects this "broken machine" to a different field called Orthogonal Polynomials. These are special mathematical curves used to solve problems in physics and engineering.

• There is a famous class of these curves (the "Szegő class") that are supposed to be very well-behaved.
• Denisov shows that because his "broken machine" exists, there are also these special curves that never stop oscillating. Even though the rules governing them look safe and smooth, the curves themselves can go wild at every single point on the circle.
• This also means that if you try to add up a series of these curves (like adding up notes in a song), the sum might never settle down, even if the "volume" of the notes is low enough to be considered safe.

The "Weak" Version Still Works
Interestingly, while the main parts of the machine (the "strong" version) go crazy, a slightly different, "weaker" version of the calculation might still work. Denisov doesn't prove that this weaker version definitely works, but he leaves that door open. It's like saying, "The whole engine exploded, but maybe the radio still works."

Summary
In simple terms, this paper is a "proof of impossibility." It says: "You cannot assume that just because your input data is small and finite, the output of this specific nonlinear mathematical process will always be stable. We found a counter-example where the output goes completely berserk."

This result is significant because it closes the door on a long-standing guess in mathematics and forces researchers to rethink how they handle these specific types of complex, nonlinear systems. It shows that nature (or at least, the mathematical models of it) can be much more chaotic than we previously thought, even when the inputs seem tame.

Technical Summary

Technical Summary: Pointwise Behavior of SU(1,1) Nonlinear Fourier Transform

Problem Statement
The paper addresses the pointwise convergence behavior of the SU(1,1) Nonlinear Fourier Transform (NLFT) for sequences in the critical space $\ell^2(\mathbb{Z})$. Specifically, it investigates two central questions regarding the existence of limits for the transform components $a(z, F)$ and $b(z, F)$ as the truncation size $N \to \infty$:

1. Q1Z (Strong Version): For $F \in \ell^2(\mathbb{Z})$, do the limits $\lim_{N \to \infty} a(z, F^{(N)})$ and $\lim_{N \to \infty} b(z, F^{(N)})$ exist for almost every $z$ on the unit circle $\mathbb{T}$?
2. Q2Z (Weak Version): For $F \in \ell^2(\mathbb{Z})$, does the limit of the ratio $r(z, F^{(N)}) = b(z, F^{(N)}) / a^*(z, F^{(N)})$ exist almost everywhere?

This problem is motivated by the "Nonlinear Carleson Conjecture" (NCC), which posits that such convergence should hold, analogous to the classical linear Carleson theorem for Fourier series. The paper also explores the implications of these convergence properties for the theory of orthogonal polynomials on the unit circle (OPUC), specifically regarding the asymptotic behavior of orthonormal polynomials $\phi_n(z, \sigma)$ for measures $\sigma$ in the Szegő class.

Methodology
The author employs a constructive approach to disprove the strong convergence conjecture. The methodology relies on:

• Recursive Construction of "Daisies": The core technical tool is the recursive construction of a sequence of compactly supported sequences, termed "$(\nu, \delta, \epsilon)$-daisies." These are built by concatenating smaller blocks of coefficients with rapidly increasing shifts in their support.
• Variational Analysis of the Argument: The proof focuses on the argument of the function $a^*(z, F)$. By carefully arranging the phases of the constituent blocks, the author demonstrates that the argument of the partial transforms can be made to grow arbitrarily large on specific arcs of the unit circle.
• Hilbert Transform Estimates: The construction utilizes estimates on the Hilbert transform (specifically the cotangent kernel) to show that the accumulation of "bumps" in the logarithmic modulus of the transfer functions leads to logarithmic growth in the argument of $a^*$.
• Disjoint Support and Induction: The construction ensures that the supports of the added blocks are disjoint and sufficiently separated. This allows the use of induction and specific lemmas (e.g., Lemma 2.3 and 2.4) to control the error terms, ensuring that the cumulative effect of the blocks dominates while maintaining the $\ell^2$ norm of the total sequence within bounds.
• Extension to $\mathbb{R}$: The argument is adapted to the continuous case (NLFT on $\mathbb{R}$) by constructing a potential $q \in L^2(\mathbb{R})$ using a similar recursive scheme with intervals covering the real line infinitely often.

Key Contributions and Results
The paper provides definitive negative answers to the strong convergence questions in the critical $\ell^2$ case:

• Theorem 1.1 (Disproof of Q1Z): There exists a sequence $F \in \ell^2(\mathbb{Z})$ such that the limits $\lim_{N \to \infty} a(z, F^{(N)})$ and $\lim_{N \to \infty} b(z, F^{(N)})$ diverge at every point $z \in \mathbb{T}$. The constructed sequence can be chosen to have support in $\mathbb{Z}^+$.
• Theorem 1.4 (Disproof of Q1R): Similarly, there exists a potential $q \in L^2(\mathbb{R})$ for which the limits of the scattering coefficients $a(k, q^{(T)})$ and $b(k, q^{(T)})$ do not exist for any $k \in \mathbb{R}$.
• Theorem 1.2 (OPUC Divergence): There exists a measure $\sigma$ in the Szegő class (specifically, absolutely continuous with a continuous, positive density) such that the sequence of reversed orthonormal polynomials $\phi_n^*(z, \sigma)$ diverges for all $z \in \mathbb{T}$.
• Theorem 1.3 (Orthogonal Series Divergence): There exists a measure $\sigma$ in the Szegő class and a sequence of coefficients $\{\alpha_n\} \in \ell^2(\mathbb{Z}^+)$ such that the orthogonal series $\sum \alpha_n \phi_n(z, \sigma)$ diverges for all $z \in \mathbb{T}$.

The paper notes that while the strong limits diverge, the weak limit (the ratio $r(z, F)$) may still converge. Indeed, the constructed sequence $H$ in the proof satisfies the property that $r(z, H^{(n)})$ converges uniformly on $\mathbb{T}$, leaving the status of Q2Z open.

Significance and Claims
The paper claims to resolve the "Nonlinear Carleson Conjecture" in the negative for the strong version in the critical $\ell^2$ setting. Its primary significance lies in:

1. Refuting the Analogy: It demonstrates that the pointwise convergence properties of the classical linear Fourier transform (Carleson's theorem) do not extend to the nonlinear setting for square-summable data.
2. OPUC Implications: It establishes that the classical pointwise asymptotics for orthogonal polynomials, which hold for the Lebesgue measure, can fail completely even for measures that are "regular" (absolutely continuous with continuous, positive densities) within the Szegő class. This contrasts with the known convergence for $\ell^p$ coefficients with $p < 2$.
3. Sharpness of Results: The results highlight the critical nature of the $\ell^2$ boundary. While convergence is known for $\ell^p$ with $p < 2$, the paper shows that $\ell^2$ is the threshold where pointwise divergence can occur everywhere.

The author explicitly states that the construction does not answer the weak convergence question (Q2Z) and that the divergence occurs for the specific constructed measures, not necessarily for all measures in the Szegő class. The work relies on existing connections between the NLFT and OPUC (via the Verblunsky coefficients) to translate the counterexample in the transform domain to the polynomial domain.