The Benjamin-Ono Equation in the Long-Time Limit: Linearized Self-Similar Universality

Technical Summary: The Benjamin-Ono Equation in the Long-Time Limit: Linearized Self-Similar Universality

Problem Statement
This paper investigates the long-time asymptotic behavior of the Cauchy problem for the Benjamin-Ono (BO) equation:
$$\partial_t u + \partial_x(u^2) = \partial_x |D|u,$$
specifically in the regime where $t \to +\infty$ and the spatial variable scales as $x = O(t^{1/2})$. While the dispersive ($x/t < -\delta$) and soliton ($x/t > \delta$) regions for the BO equation have been recently characterized, the behavior in the intermediate "similarity" region ($x/t$ small) remained an open question.

Motivated by the Korteweg-de Vries (KdV) equation, where the long-time behavior in the similarity region is governed by self-similar solutions related to the Painlevé-II equation, the authors initially hypothesized that the BO equation would exhibit similar self-similar universality. A formal scaling analysis suggests a similarity variable $\xi = x/(2t^{1/2})$ and a profile equation $-2\xi U + 2U^2 = |D|U$. However, the paper demonstrates that for a broad class of generic initial data, the solution does not converge to a self-similar solution of this nonlinear profile equation. Instead, the decay rate is faster than the self-similar prediction, and the limiting profile is governed by the linearization of the profile equation.

Methodology
The authors employ a rigorous asymptotic analysis based on the inverse spectral theory of the Benjamin-Ono equation, specifically utilizing an explicit solution formula for rational initial data derived in prior works [2, 9].

1. Rational Initial Data Class: The study focuses on initial data $u_0(x)$ of the form:
   $$u_0(x) = \sum_{n=1}^N \left[ \frac{c_n}{x - p_n} + \frac{c_n^*}{x - p_n^*} \right],$$
   where $p_n$ are distinct points in the upper half-plane and the residues satisfy $\text{Re}(\sum c_n) = 0$ to ensure $u_0 \in L^1(\mathbb{R})$.

2. Genericity Condition: A dichotomy exists between "generic" and "non-generic" initial data, determined by the asymptotic behavior of the reflection coefficient $\beta(\lambda)$ as $\lambda \downarrow 0$. The authors define a specific algebraic quantity $\Delta$, constructed from the pole and residue data via contour integrals on the Riemann sphere. They prove that $\Delta \neq 0$ is a sufficient condition for the initial data to be generic (i.e., $\beta(\lambda) \sim 2\pi i / \ln(\lambda)$). They further show that such generic rational potentials are dense in $L^2(\mathbb{R})$.

3. Asymptotic Expansion: The solution $u(t, x)$ is expressed as a ratio of determinants, $\Pi u = \det(A)/\det(B)$, where the matrix elements are contour integrals involving the phase function $h(z; t, x) = (z-x)^2/4t + L(z)$ (with $L(z)$ being the antiderivative of $u_0$).

   • The authors perform a detailed asymptotic expansion of these matrix elements as $t \to \infty$ with $x = 2t^{1/2}\xi$.
   • Crucially, they identify a dominant logarithmic term, $O(\ln t)$, arising from the singularity at the origin in the contour integrals. This term alters the scaling of the solution compared to the standard self-similar ansatz.
   • By expanding the determinants using column multilinearity, they isolate the leading-order terms, which involve a specific universal function $F(\xi)$ defined by a Fresnel-type integral.

Key Contributions and Results

• Theorem 1.1 (Genericity Criterion): The authors establish that for rational initial data, the condition $\Delta \neq 0$ implies the data is generic. This provides a concrete algebraic test for the genericity of rational potentials, distinguishing them from multisoliton solutions (where $\Delta = 0$).

• Theorem 1.2 (Density): They prove that the set of rational potentials satisfying the genericity condition ($\Delta \neq 0$) is dense in $L^2(\mathbb{R})$.

• Theorem 1.3 (Main Asymptotic Result): For generic rational initial data ($\Delta \neq 0$), the solution exhibits the following asymptotic behavior:
  $$\lim_{t \to +\infty} t^{1/2} \ln(t) u(t, 2t^{1/2}\xi) = U(\xi)$$
  in the $L^\infty_{\text{loc}}$ sense. The limiting profile is explicitly given by:
  $$U(\xi) = 2\text{Re}(F(\xi)), \quad F(\xi) = \frac{e^{-i\pi/4}}{\sqrt{\pi}} \int_{L_+} \frac{e^{-i(w-\xi)^2}}{w} dw,$$
  where $L_+$ is a contour from $\infty e^{3\pi i/4}$ to $\infty e^{-i\pi/4}$ passing above the origin.

  • Decay Rate: The solution decays as $t^{-1/2}/\ln(t)$, which is strictly faster than the $t^{-1/2}$ decay predicted by the self-similar ansatz.
  • Linearization: The profile $U(\xi)$ satisfies the linear equation $-\xi U = \frac{1}{2}|D|U$, which is the linearization of the nonlinear self-similar profile equation $-2\xi U + 2U^2 = |D|U$. The nonlinearity becomes negligible in the leading order due to the enhanced decay rate.

• Theorem 1.4 (Properties of the Profile): The authors verify that $U(\xi)$ is a solution to the linearized profile equation and analyze its asymptotic behavior: $U(\xi) = O(\xi^{-3})$ as $\xi \to +\infty$ and $U(\xi) \sim 4\sqrt{\pi}\cos(\xi^2 + 3\pi/4)$ as $\xi \to -\infty$.

Significance and Claims
The paper claims to resolve the universality of the Benjamin-Ono equation in the long-time, small-slope regime. The primary significance lies in the discovery that, unlike the KdV equation where the long-time behavior is governed by nonlinear self-similar solutions (Painlevé-II), the BO equation's generic long-time behavior is governed by a linearized self-similar profile.

The authors note that this result contrasts with the KdV case in two ways:

1. The decay rate is faster ($t^{-1/2}/\ln t$ vs $t^{-1/3}$), rendering the nonlinear term in the profile equation subdominant.
2. The universal profile $U(\xi)$ is expressible in terms of elementary functions (via Fresnel integrals), unlike the Hastings-McLeod solution of Painlevé-II required for KdV.

Furthermore, the paper argues that the "collisionless shock" region observed in KdV (between the dispersive and self-similar regions) has no analogue in the BO equation for generic data, as the linearized profile asymptotically matches the dispersive decay in the overlap domain. The results are strictly established for rational initial data, though the density argument suggests these findings are representative of generic behavior in $L^2(\mathbb{R})$. The authors explicitly state that extending these results to non-rational data would require a fundamentally different approach.