🔢 mathematics

2D Torus defocusing NLSE with a potential and random initial data

This paper establishes the almost sure global well-posedness in a weak sense for the 2D defocusing cubic nonlinear Schrödinger equation with a potential and random initial data in $H^{0^-}$ , demonstrating that the solution remains close to the linear evolution in a higher regularity space and arises as the limit of truncated approximations, thereby generalizing Bourgain's earlier results.

Original authors: Nicolas Lindstrom

Published 2026-09-21

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Nicolas Lindstrom

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

In the quiet corners of theoretical physics, there is a class of equations that describe how waves move through space and time. Among the most famous of these is the Schrödinger equation, a mathematical tool originally designed to describe the behavior of tiny particles like electrons, but which also serves as a powerful model for waves in fluids, light, and other physical systems. When these waves interact with themselves, the equation becomes non-linear, meaning the wave's shape changes in complex ways as it travels. On a flat, infinite plane, these waves can spread out and fade. But when they are confined to a closed, repeating space like a torus—a shape that looks like a doughnut—the waves cannot escape. They bounce around forever, interacting with themselves in a chaotic dance that can sometimes lead to unpredictable, wild behavior.

For decades, mathematicians have struggled to understand what happens when these waves start from a state of pure randomness. In the real world, perfect order is rare; systems are usually jumbled with noise. To model this, researchers use "random initial data," where the starting shape of the wave is determined by a roll of the dice. The challenge is that this randomness often creates waves that are so jagged and irregular that standard mathematical tools break down, unable to predict if the wave will survive or collapse. The question becomes: can we prove that even with this chaotic, messy start, the system evolves in a stable, predictable way over time?

A recent study by Nicolas Lindstrom tackles this problem by adding a new layer of complexity: a potential field. Imagine the doughnut-shaped space not as empty, but as having an invisible landscape of hills and valleys that the wave must travel over. This "potential" acts like a force that pulls or pushes the wave, altering its path. While this might seem like a small change, it fundamentally disrupts the mathematical techniques used to solve the simpler version of the problem. The presence of this landscape makes it difficult to track how the wave's energy moves and interacts, threatening to undo the stability that researchers had previously established for the empty space.

Lindstrom's work demonstrates that despite this added difficulty, the system remains stable. The paper proves that if you start with a wave generated by a specific type of random noise—one that is mathematically linked to the shape of the potential landscape itself—the wave will not collapse. Instead, it will evolve smoothly for all time. The solution is not a perfect, smooth wave, but it is a "well-posed" one, meaning it exists, is unique, and changes continuously as time passes. The researchers show that the chaotic, random part of the wave stays close to the behavior of a simple linear wave, while the messy, non-linear interactions create only a small, manageable disturbance. This disturbance is regular enough to be understood and controlled, ensuring the entire system does not blow up.

To reach this conclusion, the team had to navigate significant hurdles. First, because the initial data is so rough (mathematically, it is almost surely in a space slightly below $L^2$ ), the standard non-linear term $u|u|^2$ is ill-defined. To fix this, the researchers must apply a process called "Wick ordering" (or renormalization). This involves subtracting an infinite average value from the non-linearity, effectively redefining the interaction term as $:u|u|^2: = u|u|^2 - 2u\mathbb{E}\int|u|^2$ . This step is critical; without it, the energy of the system would be infinite, and the Hamiltonian would not be well-defined.

Second, the potential field destroys the symmetry that usually makes these equations easy to solve. In the absence of the potential, the wave's frequencies interact in a predictable pattern that allows mathematicians to count and bound their effects. The potential scrambles this pattern, making the interactions look like a tangled mess. Lindstrom overcomes this by introducing a clever mathematical trick: a "pseudo-diagonalized" operator, denoted as $-\Delta + V^r$ . This is not just a vague simplification, but a specific pseudo-differential operator constructed to preserve the essential resonant structures of the problem while restoring enough order to allow the mathematical tools to work. The researchers prove that the solution to this simplified problem is so close to the real problem that the difference is negligible.

The study also addresses how these random waves behave when approximated by computers. Since computers cannot handle infinite complexity, they must break the problem down into smaller, finite pieces. The paper shows that as these pieces get smaller and the approximation gets finer, the computer's solution converges to the true, infinite solution. This convergence happens at a specific, predictable speed, giving confidence that numerical simulations of such systems are reliable. Furthermore, the researchers connect this local stability to a global one by using a concept called a "Gibbs measure." This is a special probability distribution that describes the statistical state of the system. They prove that this distribution remains unchanged as the wave evolves, acting as a safety net that prevents the system from wandering into unstable, chaotic regions.

The findings are significant because they show that the stability of these wave systems is robust. It does not depend on the space being perfectly empty or the landscape being perfectly flat. Even when the environment is rough and the starting conditions are wildly random, the system finds a way to organize itself. The work generalizes previous results that were limited to empty spaces, proving that the mathematical framework is sturdy enough to handle the first non-trivial perturbation of adding a potential. This suggests that the chaotic behavior seen in these systems is not a sign of mathematical failure, but a feature that can be tamed and understood, even in the presence of complex forces.

Ultimately, the paper provides a rigorous proof that the non-linear Schrödinger equation on a two-dimensional torus, when subjected to a random start and a potential field, is well-behaved. The solution exists for all time, is unique, and can be approximated with high precision. The researchers show that the random initial data, which is almost certainly too rough to be handled by traditional methods, evolves into a state where the chaotic parts are just a slight, regular perturbation of a linear wave. This result bridges the gap between the idealized, empty-space models and more realistic scenarios where forces and randomness play a role, offering a clearer picture of how complex wave systems maintain their structure against the odds.

Technical Summary: 2D Torus Defocusing NLS with a Potential and Random Initial Data

Problem Statement
This paper addresses the global well-posedness of the defocusing cubic Nonlinear Schrödinger Equation (NLS) on the 2D torus $\mathbb{T}^2$ in the presence of a potential $V(x) \in H^{2^{+}}(\mathbb{T}^2)$ . The equation is given by:
$(i\partial_{t}-\Delta+V)u = -u|u|^{2}$
The initial data $u_0$ is chosen as a centered Gaussian random variable with covariance $(-\Delta+V)^{-1}$ . This choice places the initial data almost surely in the negative Sobolev space $H^{0^{-}}(\mathbb{T}^2)$ , a regularity level below $L^2$ where deterministic well-posedness theories typically fail. The primary challenge is to establish almost sure global well-posedness and show that the solution is a regular perturbation of the linear evolution, despite the introduction of the potential $V$ , which disrupts the standard resonance structures and lattice counting arguments used in the potential-free case.

Methodology
The approach follows the probabilistic well-posedness framework established by Bourgain and extended by others (e.g., Burq, Tzvetkov), adapted to handle the perturbation by a potential. The methodology proceeds in three main stages:

Linearization and Pseudo-Diagonalization:
The author treats the operator $-\Delta+V$ as the linear part of the equation. Since the eigenfunctions of $-\Delta+V$ do not possess the simple product structure of the standard Fourier basis (i.e., $e_n e_m \not\sim e_{n+m}$ ), the author introduces a "pseudo-diagonalized" potential $V^r$ . This operator is constructed using a partition of the frequency lattice $\mathbb{Z}^2$ into sets $\Omega_\alpha$ where the eigenvalues and eigenfunctions of $-\Delta+V^r$ behave sufficiently like those of $-\Delta$ . This allows for the adaptation of Bourgain space estimates and lattice counting arguments.
Local Well-Posedness (LWP) via Fixed Point:
The local theory is developed in Bourgain spaces $X_{s,b}$ . The solution is decomposed into a linear part (driven by the random initial data) and a regular remainder $U$ . The author proves that the nonlinear term, specifically the Wick-ordered nonlinearity $:u|u|^{2}:$ , acts as a contraction on a ball in $X_{s,b}$ for a short time interval.
- Wick Ordering: To handle the divergence of the $L^2$ norm of the random data, the nonlinearity is Wick-ordered: $:u|u|^{2}: = u|u|^2 - 2u\mathbb{E}(\int |u|^2)$ .
- Resonance Handling: The proof involves a detailed trilinear analysis of the interaction between the random linear part and the regular part. The author utilizes the specific structure of the eigenfunctions of $-\Delta+V^r$ to bound resonant interactions, adapting the lattice counting lemmas from the potential-free case to the perturbed setting.
- Bilinear Expansion Terms: A critical technical step involves the bilinear expansion of the nonlinearity. In the potential-free case, eigenfunctions have constant modulus 1, causing certain terms to vanish. However, for the perturbed operator, the eigenfunctions $e_n$ do not have constant modulus. Consequently, the expansion of the nonlinearity generates a specific term of the form $u(|e_n|^2 - 1)$ (appearing in the bilinear decomposition). This term, which does not appear in the standard case, must be carefully controlled to ensure the solution remains a perturbation of the linear flow.
Global Extension via Gibbs Measure:
Once local well-posedness is established for the truncated equations, the result is extended to global time using the invariance of the Gibbs measure. The author defines a Gibbs measure $\mu$ as a weighted Wiener measure associated with the Hamiltonian of the system. They prove that the Hamiltonian is almost surely finite and that the Gibbs measure is invariant under the flow of the truncated equation. By a standard pullback argument (iterating the local existence on small time intervals), they extend the solution to all time $t \in \mathbb{R}$ .

Key Contributions and Results

Stability of Bourgain's Results: The paper demonstrates that the probabilistic well-posedness results for the 2D NLS on the torus (specifically those in Bourgain [1]) are stable under the perturbation of a potential $V \in H^{2^{+}}$ .
Almost Sure Global Well-Posedness: For initial data distributed according to the Gaussian measure with covariance $(-\Delta+V)^{-1}$ , the Cauchy problem is almost surely globally well-posed in the distributional sense.
Regularity of the Nonlinear Part: The solution $u(t)$ can be written as $u(t) = e^{it(-\Delta+V)}u_0 + U(t)$ , where the remainder $U(t)$ belongs to $H^s(\mathbb{T}^2)$ for some $s > 0$ . This confirms that the nonlinearity provides a regularizing effect relative to the rough initial data.
Convergence of Truncated Solutions: The solution is shown to be the limit of solutions to the truncated equations (with Wick-ordered nonlinearities) in the space $L^\infty_{loc}H^s$ . The convergence rate is quantified as $O(N^{-\alpha})$ .
Handling the Potential's Effect on Resonances: A significant technical contribution is the adaptation of the resonance counting arguments. The author shows that while the potential $V$ destroys the exact multiplicative structure of the Fourier basis, a "pseudo-diagonalized" version $V^r$ allows for the recovery of the necessary estimates, provided one accounts for the deviation between the true eigenfunctions and the standard Fourier modes.
Wick Ordering and Eigenfunction Modulus: The paper highlights that the Wick ordering of the nonlinearity is essential not just for renormalization, but specifically because the non-constant modulus of the eigenfunctions of $-\Delta+V$ introduces new terms (specifically $u(|e_n|^2 - 1)$ in the bilinear expansion) that must be controlled to ensure the solution remains a perturbation of the linear flow.

Significance and Claims
The paper claims to generalize the foundational work of Bourgain on the 2D torus to the case with a potential. The author notes that this is the "first non-trivial perturbation" of the NLS on the torus, as lower-order perturbations or 1D cases are relatively straightforward.

The significance lies in overcoming the "great hurdle" of coupling the local and global theories when the linear operator is no longer the standard Laplacian. The author argues that their method reveals the essential factors making Bourgain's proof work, showing that the reliance on the point-wise product of exponentials can be circumvented by using pseudo-diagonalization and careful probabilistic estimates. The result confirms that the geometric and probabilistic mechanisms ensuring well-posedness on the torus are robust enough to withstand the introduction of a potential, distinguishing this case from the sphere (where such results are more delicate due to different geometric considerations).

The work does not propose new experimental applications but rather solidifies the theoretical framework for dispersive equations with random data on compact manifolds with perturbations.

Technical Summary: 2D Torus Defocusing NLS with a Potential and Random Initial Data

More like this