🔢 mathematics

Restriction estimates for toral eigenfunctions and lattice points in spherical regions

This paper establishes new sharp $L^2$ restriction estimates for toral eigenfunctions on smooth submanifolds of large codimension, thereby proving a conjecture of Huang-Zhang and advancing the conjecture of Bourgain-Rudnick through a combination of slicing, packing, and discrete spherical multiplier approximation techniques.

Original authors: Cheng Zhang, Zhifei Zhu

Published 2026-06-09

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Cheng Zhang, Zhifei Zhu

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: The Drum and the Net

Imagine a giant, perfectly round drum (a mathematical shape called a torus, which looks like a donut). When you hit this drum, it vibrates. These vibrations are called eigenfunctions.

In the real world, a drum might vibrate in a messy, chaotic way. But in this math paper, the drum is special. Its vibrations are made up of pure, rhythmic beats that line up perfectly with a grid of invisible dots (like a lattice or a checkerboard) floating in the air.

The authors, Cheng Zhang and Zhifei Zhu, are asking a specific question: If you take a small piece of this drum (a curve or a surface) and measure how loud the vibration is only on that piece, how loud can it get compared to the whole drum?

This is called a "restriction estimate." It's like asking: "If I put a microphone on just one tiny line on the drum, how much sound will it pick up?"

The Problem: The "Grid" of Sound

The authors are working with a drum that has a very specific, rigid structure. The "notes" it can play aren't random; they must land exactly on integer coordinates (like points on a graph where x and y are whole numbers).

Because of this grid-like structure, the sound waves don't spread out evenly. They tend to clump together in certain spots, creating "hotspots" of loudness. The authors want to know the absolute maximum volume these hotspots can reach on specific shapes (like a straight line, a curved line, or a flat surface) drawn on the drum.

The Old Rules vs. The New Rules

Before this paper, mathematicians had a general rule for how loud these vibrations could get on any shape. It was like a "safety net" that said, "Okay, the sound might get this loud, but probably not louder."

However, this old rule was a bit loose. It was like saying, "A car might go 100 mph," when in reality, on a specific track, it can only go 60 mph. The authors wanted to find the exact speed limit for these specific drum shapes.

The New Discoveries

The authors found new, tighter speed limits (mathematical bounds) for how loud the sound can get on these shapes. Here is how they did it, using two main tools:

1. The "Slicing and Packing" Method

Imagine you have a giant, fuzzy cloud of points (the grid of sound notes) floating inside a giant sphere. You want to count how many points are in a specific slice of that sphere.

The authors developed a way to slice this cloud into thin layers and pack them into neat boxes. By organizing the chaos into orderly boxes, they could count the points much more accurately than before. This allowed them to prove that for very "thick" shapes (high codimension), the sound cannot get as loud as the old rules predicted.

2. The "Magic Approximation"

The authors used a sophisticated mathematical trick (developed by other mathematicians named Magyar, Stein, and Wainger) to approximate the sound waves.

Think of the sound wave as a complex song. This trick breaks the song down into a simple, predictable melody plus a tiny bit of static noise. By focusing on the melody and ignoring the static, they could calculate the volume on the specific shapes much more precisely.

What They Actually Proved

The paper makes two main claims about the "loudness" (mathematically, the $L^2$ norm) of these vibrations:

For "Thick" Shapes (High Codimension):
If you measure the sound on a shape that is very "thin" compared to the whole drum (like a flat sheet in a 5-dimensional space), the authors proved that the sound is strictly limited. They confirmed a guess made by Huang and Zhang: the sound gets quieter than previously thought for these specific shapes.
- Analogy: If you try to catch a specific type of fish in a huge ocean using a very small, specific net, you now know exactly the maximum number of fish you can catch. It's less than the old "worst-case scenario" guess.
For Curves in 3D (The "Twisted Wire"):
They looked at 3D space (a 3D donut) and measured the sound on different types of lines:
- Straight lines: They found a specific limit.
- Twisted lines (with "torsion"): If the line twists in 3D space, the sound is limited to a specific level ( $\lambda^{1/3}$ ).
- Flat curved lines: If the line curves but stays flat on a 2D plane, the sound is even quieter ( $\lambda^{1/4}$ ).
- Analogy: Imagine walking along a wire. If the wire is straight, you might hear a lot of noise. If it twists and turns in 3D, the noise is dampened. If it curves gently on a flat sheet, the noise is dampened even more. The authors calculated the exact "dampening factor" for each case.

Why This Matters (According to the Paper)

The authors didn't just guess these numbers; they proved them.

They solved a specific puzzle (Conjecture 1.2) for shapes that are "thick" enough (in high dimensions).
They improved the general rules for all shapes, showing that the old "safety net" was too loose.
They connected the problem of "how loud is the sound?" to the problem of "how many grid points fit in a circle?" (Lattice points). By solving the grid counting problem, they solved the sound problem.

What They Did Not Do

It is important to stick to what the paper says:

They did not apply this to real-world engineering, medical imaging, or audio technology.
They did not claim this will change how we build speakers or drums in the real world.
They did not solve the problem for every possible shape (some very thin, tricky shapes in low dimensions remain unsolved).

Summary in One Sentence

Zhang and Zhu used clever counting tricks and mathematical approximations to prove that on a special, grid-based drum, the sound vibrations on specific lines and surfaces are strictly quieter than mathematicians previously believed, solving a long-standing guessing game about how these vibrations behave.

Technical Summary: Restriction Estimates for Toral Eigenfunctions and Lattice Points in Spherical Regions

Problem Statement
The paper investigates $L^2$ restriction estimates for eigenfunctions of the Laplace–Beltrami operator on the standard flat torus $T^d = \mathbb{R}^d/\mathbb{Z}^d$ . Specifically, given an eigenfunction $e_\lambda$ satisfying $-\Delta e_\lambda = (2\pi\lambda)^2 e_\lambda$ , the authors seek to bound the $L^2$ norm of its restriction to a fixed submanifold $\Sigma \subset T^d$ of codimension $m = d - \dim \Sigma$ .

In the general Riemannian setting, Burq–Gérard–Tzvetkov and Hu established bounds of the form $\|e_\lambda\|_{L^2(\Sigma)} \lesssim \Lambda(m, \lambda) \|e_\lambda\|_{L^2(T^d)}$ , where $\Lambda(m, \lambda)$ grows polynomially in $\lambda$ . On the flat torus, the arithmetic structure of the eigenfunctions (trigonometric polynomials with frequencies on the lattice sphere $E_\lambda = \mathbb{Z}^d \cap \lambda S^{d-1}$ ) suggests these bounds can be improved. The central problem is to determine the sharp exponents for these improvements, particularly addressing the conjectures of Bourgain–Rudnick (for hypersurfaces) and Huang–Zhang (for higher codimension smooth submanifolds).

Methodology
The authors employ a hybrid approach combining geometric lattice point counting with harmonic analysis techniques:

Reduction to Lattice Point Counting: For totally geodesic submanifolds, the restriction problem is reduced to estimating the maximum number of lattice points, denoted $A_{k,d,\lambda}$ , lying in the intersection of the sphere $\lambda S^{d-1}$ and the intersection of $k$ transverse unit-width bands. Theorem 2.5 establishes that $\|e_\lambda\|_{L^2(\Sigma)} \asymp \sqrt{A_{k,d,\lambda}} \|e_\lambda\|_{L^2(T^d)}$ .
Slicing and Packing: To estimate $A_{k,d,\lambda}$ , the authors develop a "slicing and packing" method. They decompose spherical regions into "standard bricks" (sets of lattice points with controlled projections onto orthogonal subspaces). By establishing recurrence relations for the number of lattice points in these bricks (Propositions 3.4 and 3.5), they derive upper bounds for $A_{k,d,\lambda}$ that depend on the dimension $d$ and codimension $m$ .
Discrete Spherical Multiplier Approximation: For general smooth submanifolds, the authors utilize the approximation of the discrete spherical multiplier $S_n(x)$ (where $n=\lambda^2$ ) developed by Magyar, Stein, and Wainger. This involves decomposing the multiplier into a main term involving Kloosterman sums and a surface measure approximation, plus an error term. This allows them to bound the restriction operator norm via Schur's test and van der Corput-type lemmas.
Curve-Specific Techniques: For curves in $T^3$ , the proof combines Bourgain–Rudnick's lattice point estimates in spherical caps with the van der Corput lemma, distinguishing cases based on torsion and geodesic curvature.

Key Contributions and Results

Proof of Huang–Zhang Conjecture for High Codimension: The paper proves Conjecture 1.2 of Huang–Zhang for smooth submanifolds of codimension $m \geq (d+3)/2$ when $d \geq 5$ . Specifically, Theorem 1.3(ii) establishes:
$\|e_\lambda\|_{L^2(\Sigma)} \lesssim_\epsilon \lambda^{\frac{m}{2} - 1 + \epsilon} \|e_\lambda\|_{L^2(T^d)}$
This exponent is shown to be sharp in general cases.
Improved Bounds for Totally Geodesic Submanifolds: Theorem 1.3(i) provides new, sharper bounds for totally geodesic submanifolds across all codimensions. The exponents $\alpha(m, d)$ are explicitly calculated and improve upon the general Burq–Gérard–Tzvetkov estimates. For instance, for $m \geq 4$ and $d \geq 5$ , the exponent is $\frac{m}{2} - 1$ when $d \leq 2m-3$ , and slightly larger otherwise.
Restriction Estimates on $T^3$ : Theorem 6.2 provides new estimates for curves in $T^3$ :
- For geodesic segments, smooth curves with non-vanishing torsion, or real-analytic curves with non-vanishing geodesic curvature: $\|e_\lambda\|_{L^2(\Sigma)} \lesssim_\epsilon \lambda^{1/3+\epsilon} \|e_\lambda\|_{L^2(T^3)}$ .
- For smooth planar curves with non-vanishing geodesic curvature: $\|e_\lambda\|_{L^2(\Sigma)} \lesssim_\epsilon \lambda^{1/4+\epsilon} \|e_\lambda\|_{L^2(T^3)}$ .
  These results improve upon the general codimension-2 bound of $\lambda^{1/2}\sqrt{\log \lambda}$ .
Connection to Discrete Restriction Conjecture: The authors discuss the relationship between their $L^2$ restriction bounds and the endpoint Discrete Restriction Conjecture (Bourgain–Demeter). They show that if the endpoint conjecture holds, it would imply specific exponents for curve restrictions that align with the results obtained in Theorem 6.2 for $d=3$ and $d=4$ .

Significance and Limitations
The paper claims to provide significant progress toward the Bourgain–Rudnick and Huang–Zhang conjectures. The results are sharp in certain regimes, particularly for high codimension and for totally geodesic submanifolds.

However, the authors are modest regarding the resolution of the full conjectures. They note that their approach only partially confirms Conjecture 1.2. For smaller codimensions, the method is limited by the remainder term $\lambda^{(d-1)/2+\epsilon}$ in the Magyar–Stein–Wainger approximation (Proposition 4.1). A full resolution for lower codimensions would require a more efficient exploitation of cancellation in the relevant exponential sums, which remains an open problem. The paper does not claim to solve the general case for all smooth submanifolds of low codimension but establishes a new baseline of sharp estimates for high codimension and specific geometric configurations.