Phase retrieval in Fock space and perturbation of Liouville sets

Philipp Grohs, Lukas Liehr, Martin Rathmair

TL;DR

This paper breaks a major "no-go" result in signal processing by proving that functions in Fock space can be uniquely determined from the magnitude of their samples, provided the sampling locations are irregular. By perturbing regular lattices into "Liouville sets," the authors provide the first construction of finite-density uniqueness sets for phase retrieval, overcoming the algebraic obstructions that render standard lattices unsolvable.

What Problem Does This Paper Address?

Phase retrieval—the challenge of reconstructing a signal from only the intensity (magnitude) of its measurements—is a cornerstone of applications like X-ray crystallography, diffraction imaging, and audio processing. In the context of time-frequency analysis, this problem is often studied in Fock space, a space of entire functions that naturally arises when applying the Gabor transform to signals.

A frustrating "impossibility" result has long loomed over this field: sampling on a regular lattice (no matter how dense) is mathematically insufficient to uniquely determine a function in Fock space from its magnitudes. This isn't a problem of "not enough data," but rather an algebraic obstruction caused by the rigid symmetry of the lattice. The authors set out to answer a fundamental question: Do there exist any sampling sets with finite density that allow for unique phase retrieval?

Methods

The authors move away from rigid lattices and toward irregular sampling. Their approach relies on the concept of Liouville sets—sets where any function in the space that is bounded on the set must be a constant.

Key technical components include:

  • Deterministic Perturbation: They construct sampling sets as the union of three sets ($A \cup B \cup C$), where each is a perturbation of a Liouville set. They define specific geometric conditions—closeness and noncollinearity—that these sets must satisfy to resolve the phase ambiguity.
  • Random Perturbation: They prove that if you take a sufficiently dense lattice and shift its points randomly within small disks, the resulting set will satisfy the uniqueness conditions "almost surely."
  • Lagrange-type Interpolation: The proofs utilize complex analysis tools, specifically Weierstrass $\sigma$-functions and modified interpolation formulas, to bridge the gap between discrete samples and the continuous function.

Datasets

Not discussed in the paper. This is a theoretical mathematical study focused on proving existence and uniqueness theorems; it does not utilize empirical datasets or benchmarks.

Key Insights and Ideas

  • The Hierarchy of Sets: The authors establish a beautiful theoretical "sandwich": Sets of Stable Sampling $\subset$ Liouville Sets $\subset$ Uniqueness Sets. By proving Liouville sets occupy this intermediate position, they create a new pathway for constructing uniqueness sets.
  • Symmetry Breaking: The failure of lattices is due to their regularity. The "aha moment" is that irregular perturbations—even very small ones—break the algebraic symmetries that allow different functions to have the same magnitudes.
  • Density vs. Structure: In most sampling problems, density is the primary concern (e.g., the Nyquist rate). Here, the structure of the set is the primary obstruction. You can have infinite density on a lattice and fail, or finite density on an irregular set and succeed.
  • Exploiting Signal Priors: If we know a signal has specific properties (like being real-valued or even), the requirements for the sampling set drop significantly.

Key Results

  • General Uniqueness: For any density $d > 12\alpha/\pi$, there exist irregular uniqueness sets for phase retrieval in Fock space. This is the first such construction with finite density.
  • Gabor Phase Retrieval: The results translate directly to the Gabor transform of $L^2(\mathbb{R})$ signals.
  • For general complex signals, a union of three perturbed lattices is required (density $> 12$).
  • For real-valued signals, only two perturbed lattices are needed (density $> 6$).
  • For even real-valued signals, a single perturbed lattice suffices (density $> 3$), and the resulting set is "separated" (points aren't too close together), which is ideal for practical sensors.

Limitations and Future Work

While this paper solves the uniqueness problem (proving it can be done), it leaves two major practical doors open:

  1. Stability: The paper doesn't quantify how sensitive the reconstruction is to noise. A set might provide uniqueness in theory but be numerically unstable in practice.
  2. Algorithms: The authors provide the mathematical proof that the information is there, but they do not propose a specific numerical algorithm to perform the reconstruction from these irregular samples.

Future research will likely focus on developing efficient, noise-robust recovery algorithms tailored to these newly discovered irregular sampling geometries.

Summary updated

Abstract

We study the determination of functions in Fock space from samples of their absolute value, known as the phase retrieval problem in Fock space. An important finding in this research field asserts that phaseless sampling on lattices of arbitrary density renders the problem unsolvable. The present study establishes solvability when using irregular sampling sets of the form A \cup B \cup C , where A , B , and C constitute perturbations of a Liouville set, i.e., a set with the property that all functions in Fock space bounded on the set are constant. The sets A , B , and C adhere to specific geometrical conditions of closeness and noncollinearity. We show that these conditions are sufficiently generic so as to allow the perturbations to be chosen also at random. By proving that Liouville sets occupy an intermediate position between sets of stable sampling and sets of uniqueness, we obtain the first construction of uniqueness sets for the phase retrieval problem in Fock space having a finite density. The established results apply to the Gabor phase retrieval problem in subspaces of L^{2}(\mathbb{R}) , where we derive additional reductions of the size of uniqueness sets: for the class of real-valued functions, uniqueness is achieved from two perturbed lattices; for the class of even real-valued functions, a single perturbation suffices, resulting in a separated set.
BibTeX
@inproceedings{grohs2025phase,
  author = {Philipp Grohs and Lukas Liehr and Martin Rathmair},
  title = {{Phase retrieval in Fock space and perturbation of Liouville sets}},
  booktitle = {Revista},
  year = {2025},
  doi = {10.4171/rmi/1517},
}
Figure