What Are the Long-Time Asymptotics of the Modified Boussinesq Equation in the Painlevé Region?

The Painlevé region is where x/√t stays bounded

The Painlevé region is the part of the upper (x, t)-plane where the similarity variable y = -√3 x / (2√t) stays finite as t → ∞, so the solution is observed along rays where x grows no faster than √t. The paper treats the case 0 ≤ x/√t ≤ C with x ≥ 0 and notes that x < 0 can be handled analogously.

This region matters because the three critical points k0, ωk0 and ω²k0 of the associated Riemann-Hilbert problem coalesce at the origin as t → ∞, since k0 = O(t^(-1/2)). That coalescence is what forces a local model problem rather than a standard nonlinear steepest descent parametrix, and it is why the Painlevé IV equation, not a simpler dispersive formula, controls the leading behavior here.

The leading terms come from the Clarkson-McLeod solution

The paper states that the leading terms in the Painlevé region and in the first transition region are characterized by the Clarkson-McLeod solution of the Painlevé IV equation. In the Painlevé region the modified Boussinesq solution behaves as p(x,t) ~ (√3 / (4√t)) · (P'_IV(y) + (2/3) P_IV(y)) and q(x,t) ~ (√3 / (4√t)) · (P_IV(y) + (2/3) y), with α = -1/6 and β = -2/3.

The Clarkson-McLeod solution is the Painlevé IV solution selected by the boundary condition P_IV(y) → 0 as y → +∞. The paper cites Clarkson and McLeod for pioneering this boundary-value analysis and Xia, Xu and Zhao for a more recent study of these solutions via the Deift-Zhou nonlinear steepest descent method. The appearance of this specific solution, rather than a one-parameter family, is the paper's central structural claim about the Painlevé region.

dbar-steepest descent gives uniform error control

The method splits the mixed Riemann-Hilbert-dbar problem into a pure Riemann-Hilbert problem, which determines the leading-order asymptotics, and a pure dbar problem, which controls the contribution of the nonanalytic extensions. The pure Riemann-Hilbert problem is reduced to a local model problem near k = 0, and the local parametrix is matched to the Painlevé IV model Riemann-Hilbert problem of Appendix A.

The matching is quantitative. Lemma 4.8 shows that on the local contour the jump matrices of the reduced problem and the Painlevé IV model differ by O(t^(-1/4 - 1/(2p))) in L^p for 1 ≤ p ≤ ∞, uniformly for 0 ≤ x/√t ≤ C0. Lemma 4.9 then gives ||ŵ||_{L^p(∂D_ε(0))} ≤ C t^(-1/2), ||ŵ||_{L^p(Σ̂ \ D̄_ε(0))} ≤ C t^(-N) for any positive integer N, and ||ŵ||_{L^p(D_ε(0) ∩ Σ^(2))} ≤ C t^(-1/4 - 1/(2p)). These estimates are what make the asymptotic formulas rigorous rather than formal.

Weighted Sobolev data replace Schwartz-class assumptions

The paper studies the Cauchy problem for the modified Boussinesq equation with p(x,0) = p0(x) and q(x,0) = q0(x) in the weighted Sobolev space W(2,1)_1(R), defined by the condition that Σ_{ℓ=0}^{2} ∫_R (1+|x|) |∂_x^ℓ f(x)| dx < ∞, with the corresponding norm. The good Boussinesq initial data u0 and w0 are related to (p0, q0) through the Miura transformation, subject to stated regularity and compatibility conditions.

This is a weaker regularity setting than the Schwartz class used in earlier work, and the paper presents it as a deliberate extension: the direct scattering estimates supply the regularity and decay of the reflection coefficients needed for the dbar-steepest descent analysis. The paper also records a gap: a vanishing lemma has not been established for the Riemann-Hilbert problem associated with this version of the Boussinesq equation, so bijectivity of the scattering map and the optimal function spaces for the scattering theory remain open.

Miura transformation carries the result to the good Boussinesq equation

The reflection coefficients of the good Boussinesq equation satisfy |r_j(k)| ≥ 1 for small k, j = 1, 2, so the direct Riemann-Hilbert analysis used for the modified equation cannot be applied there near the positive t-axis. The paper therefore routes the good Boussinesq asymptotics through the Miura transformation, which links the good Boussinesq equations to the modified Boussinesq equation.

Applying that transformation yields u(x,t) ~ -[(3P'_IV(y) + 2)² - 9P_IV⁴(y) - 36yP_IV³(y) - 20y²P_IV²(y)] / [32 t P_IV²(y)]. The same Painlevé IV object with α = -1/6 and β = -2/3 therefore controls both equations in this region, but the good Boussinesq formula is a different algebraic expression in P_IV and its derivative, not a restatement of the modified-equation asymptotics.

Painlevé IV also governs the first transition region

Beyond the Painlevé region, the paper examines the adjacent transition region and finds that its asymptotics is also governed by Painlevé IV. Section 6 then investigates the transition from Painlevé IV asymptotics to dispersive waves, so the paper covers three regimes: the Painlevé region, the first transition region, and the crossover to dispersive behavior.

The paper reports that numerical simulations support the asymptotic formulas well. That statement is a consistency check on the derived formulas; it is not an independent measurement of the solution, and the rigorous content of the paper rests on the uniform error estimates obtained through the dbar-steepest descent analysis.

The result closes the positive t-axis gap for this equation

The paper frames its contribution against an open problem raised by Deift concerning the long-time asymptotics of the good Boussinesq equation with a non-self-adjoint third-order Lax operator, noting that the behavior near the positive t-axis had remained open. Earlier work by Charlier, Lenells and coauthors had treated the dispersive wave region for the good Boussinesq equation, and the present authors had treated equation (1.3) in that region.

The decision-relevant distinction for a reader is therefore regional, not global. The Painlevé IV description with α = -1/6 and β = -2/3 applies in the Painlevé region and the first transition region under the paper's solitonless scattering assumptions and weighted Sobolev initial data; the dispersive-wave description belongs to a different region and is not replaced by this result. The paper's own boundary is the missing vanishing lemma, which leaves the scattering map's bijectivity and the optimal function spaces unresolved.

Sources used in this answer