Modified Scattering in Nonlinear Dispersive Models

• Modified scattering is an asymptotic regime in nonlinear dispersive equations where solutions are renormalized via logarithmic phase corrections or effective orbits rather than converging to fixed linear states.
• It appears in models like the cubic Schrödinger and Hartree equations, with analyses using profile methods, space-time resonance, and wave packet testing to capture long-range interactions.
• This framework demonstrates that optimal linear decay can coexist with nontraditional scattering behavior, challenging misconceptions about turbulence and opening paths for precise asymptotic descriptions.

Modified scattering is an asymptotic regime in nonlinear dispersive theory in which a solution retains the sharp decay and oscillatory structure of the underlying linear flow, yet does not converge to a fixed linear scattering state. After removing the linear evolution, one instead obtains either a profile with an explicit nonlinear logarithmic phase, convergence to a nonlinear effective orbit, or convergence along a corrected characteristic flow. This distinction is explicit in the cubic Schrödinger equation on Diophantine waveguides, where $e^{-it\Delta_A}U(t)$ converges to a solution of an effective system rather than to a fixed state, and in long-range one-dimensional, Hartree, and kinetic models, where the modifier is typically logarithmic in time (Camps et al., 2024, Murphy et al., 2021, Flynn et al., 2021).

1. Definition and principal asymptotic forms

The basic contrast is between ordinary scattering and modified scattering. In ordinary scattering one seeks a fixed asymptotic state $u_+$ such that the interaction-picture profile $e^{-itL}u(t)$ converges to $u_+$. In the waveguide setting, this would mean $e^{-it\Delta_A}U(t)\to G_+$. The modified-scattering alternative is that the interaction-picture profile either converges only after renormalization by a nonlinear phase, or converges to a time-dependent effective orbit $G(t)$ solving a reduced asymptotic equation (Camps et al., 2024).

Three recurrent asymptotic patterns appear across the literature covered here.

Asymptotic type	Description	Representative papers
Logarithmic phase correction	$e^{-itL}u(t)$ converges only after multiplication by $e^{i\Theta(t,\xi)}$ or its physical-space analogue	(Murphy et al., 2021, Hoose, 2023, Hoose, 2024)
Effective nonlinear orbit	$e^{-itL}u(t)\to G(t)$, where $G$ still evolves according to a resonant or quasi-resonant system	(Camps et al., 2024, Rocha, 2016)
Corrected characteristic flow	convergence occurs along modified trajectories such as $u_+$0	(Flynn et al., 2021, Breton, 3 Mar 2025, Hong et al., 5 Apr 2026)

These forms are structurally different. A logarithmic modifier changes phase but not the leading linear kernel. A resonant effective system replaces the asymptotic constant state by a nonlinear orbit on logarithmic time scales. A modified flow changes the trajectories themselves. What unifies them is that the leading nonlinear effect is asymptotically nonintegrable and therefore cannot be absorbed into a fixed linear scattering datum.

2. Long-range mechanisms and the origin of the modifier

A common analytic signature is the appearance of a $u_+$1 term in the reduced profile dynamics. In the averaged one-dimensional dispersion-managed cubic NLS, the linear decay is $u_+$2, so the effective cubic interaction has strength $u_+$3, whose time integral diverges logarithmically. The solution therefore decays at the sharp linear rate $u_+$4, but the profile converges only after renormalization by a logarithmic phase (Murphy et al., 2021). The direct time-dependent dispersion-managed model exhibits the same long-range law, with the accumulated dispersion $u_+$5 replacing a constant coefficient (Murphy et al., 2024).

In Hartree and related nonlocal Schrödinger models, the same threshold arises from the Coulomb tail. For the Hartree equation and the Schrödinger–Bopp–Podolsky equation in $u_+$6, the quantities $u_+$7, $u_+$8, and $u_+$9 are all $e^{-itL}u(t)$0, so ordinary linear scattering is not expected. The asymptotic dynamics is therefore encoded by a nonlinear logarithmic phase generated by the limiting profile (Hoose, 2023, Hoose, 2024). The three-dimensional boson star equation and the two-dimensional semi-relativistic Hartree equation realize the same mechanism in Fourier space: the low-frequency singularity of the Hartree multiplier, together with near-resonant interactions, produces an effective $e^{-itL}u(t)$1 profile equation and hence a logarithmic phase correction (Pusateri, 2013, Kwon et al., 2023).

On product spaces, the modifier is not necessarily a scalar phase. For the coupled cubic Schrödinger system on $e^{-itL}u(t)$2, stationary phase in the Euclidean direction produces the leading term $e^{-itL}u(t)$3, where $e^{-itL}u(t)$4 is the resonant trilinear form. Since $e^{-itL}u(t)$5 is not integrable, the profiles evolve on the logarithmic time scale $e^{-itL}u(t)$6 according to an autonomous resonant system (Rocha, 2016). In the cubic Schrödinger equation on a Diophantine waveguide, the same $e^{-itL}u(t)$7-scale asymptotics survives, but the arithmetic of the transverse Laplacian eliminates the rich exact resonance network present on square tori; the remaining asymptotic motion is governed by trivial exact resonances and a restricted family of quasi-resonant high$e^{-itL}u(t)$8low$e^{-itL}u(t)$9high interactions (Camps et al., 2024).

Kinetic models broaden the notion further. In the Vlasov–Poisson system, the force is long range in three dimensions and produces a logarithmic correction to free transport, so that the correct asymptotic variable is $u_+$0 rather than $u_+$1 (Flynn et al., 2021). In the Vlasov–Riesz system with $u_+$2, the position correction accumulates like $u_+$3, and the modified reference flow becomes $u_+$4 (Hong et al., 5 Apr 2026). This suggests that modified scattering is not tied to logarithmic phase modulation alone; it is a broader description of asymptotics when free dynamics must be replaced by an explicitly corrected effective motion.

3. Geometry, arithmetic, and effective dynamics

The dependence of modified scattering on geometry is especially clear on partially periodic domains. For the cubic Schrödinger equation on $u_+$5 with transverse operator $u_+$6, the Diophantine admissibility condition

$u_+$7

suppresses exact nontrivial four-wave resonances by forcing the resonance function

$u_+$8

away from zero except in trivial cases. Under this condition, the exact resonant set reduces to $u_+$9, the effective asymptotic system is $e^{-it\Delta_A}U(t)\to G_+$0, and the resulting dynamics preserves the cluster-adapted $e^{-it\Delta_A}U(t)\to G_+$1 and $e^{-it\Delta_A}U(t)\to G_+$2 norms (Camps et al., 2024).

The same paper makes the geometric contrast explicit. On a generic Diophantine waveguide, small solutions do not exhibit the turbulent resonant behavior known on the square torus. This is presented as a sharp contrast with the infinite energy cascade scenario observed by Hani–Pausader–Tzvetkov–Visciglia in the absence of Diophantine conditions. The arithmetic qualifier “generic” is precise: for every $e^{-it\Delta_A}U(t)\to G_+$3, admissible matrices form a full Lebesgue measure set among positive definite symmetric $e^{-it\Delta_A}U(t)\to G_+$4. Rectangular tori are explicitly excluded from this admissible class (Camps et al., 2024).

The coupled cubic system on $e^{-it\Delta_A}U(t)\to G_+$5 exhibits a different but equally nontrivial asymptotic structure. There the resonant set simplifies to $e^{-it\Delta_A}U(t)\to G_+$6, the asymptotic dynamics is governed by a coupled resonant system on logarithmic time, and the reduced resonant dynamics can display a beating effect: periodic energy exchange between two torus modes across the two components. At the same time, the sum of Sobolev norms of the two components is conserved along the resonant flow, so the beating effect is compatible with global boundedness (Rocha, 2016).

These examples rule out a common misconception: long-range asymptotics does not imply a universal transition to turbulence or Sobolev norm growth. In the Diophantine waveguide problem, the asymptotic system is explicitly non-turbulent at the regularity considered; in the coupled $e^{-it\Delta_A}U(t)\to G_+$7 problem, the resonant system supports nontrivial mode exchange but still preserves the relevant Sobolev quantities (Camps et al., 2024, Rocha, 2016).

4. Analytic frameworks

Several analytic frameworks recur in modified-scattering theory. A classical route is the profile method combined with stationary phase and normal forms. In the Diophantine waveguide problem, one writes $e^{-it\Delta_A}U(t)\to G_+$8, expands the profile nonlinearity in Fourier variables, performs stationary phase in the noncompact direction, and then applies a normal form integration by parts in time to the time-nonresonant torus interactions. The resulting decomposition isolates the nonintegrable effective part $e^{-it\Delta_A}U(t)\to G_+$9 from perturbative remainders (Camps et al., 2024). In the averaged dispersion-managed NLS, a parallel Hayashi–Naumkin / Kato–Pusateri strategy leads to an ODE for the Fourier profile and a renormalized variable $G(t)$0 whose derivative is integrable (Murphy et al., 2021).

A second route is the space-time resonance method in the sense of Germain–Masmoudi–Shatah. In the two-dimensional semi-relativistic Hartree equation, the cubic phase is expanded near the low-frequency Hartree singularity, and the leading resonant contribution is converted into a logarithmic Fourier-space phase (Kwon et al., 2023). In projected Dirac/Hartree models arising from Maxwell–Dirac under zero magnetic field, the same strategy is combined with spinorial null structure from Dirac projection operators, which is needed because the resonance analysis involves many sign configurations rather than a single scalar phase (Cho et al., 2022).

A third route is testing by wave packets. The expository notes on one-dimensional cubic dispersive equations describe this method as a tool to capture asymptotic equations efficiently, simplify the modified-scattering mechanism, and lower regularity requirements (Ifrim et al., 2022). In the Hartree and Schrödinger–Bopp–Podolsky equations, testing by Schrödinger wave packets localized near rays $G(t)$1 produces a raywise coefficient $G(t)$2 satisfying an asymptotic ODE of the form

$G(t)$3

with $G(t)$4; this yields both sharp $G(t)$5 decay and modified scattering at regularity $G(t)$6, $G(t)$7, and in later work at $G(t)$8 (Hoose, 2024, Hoose, 10 Nov 2025). The same wave-packet philosophy is adapted directly to Dirac spinors in the three-dimensional Maxwell–Dirac system, where the long-range tail of the Maxwell field inside the light cone generates the logarithmic phase of the spinor asymptotics (Herr et al., 2024).

A fourth route uses pseudo-conformal or Hamiltonian transformations. For the Vlasov–Poisson system, pseudo-conformal inversion transforms the large-time asymptotic problem into a local problem near a singular time $G(t)$9, where a generating function approximately integrates the main singular dynamics and yields modified wave operators (Flynn et al., 2021). The operator-valued time-dependent Kohn–Sham equation combines a pseudo-conformal transform with the factorization $e^{-itL}u(t)$0, turning the density-matrix problem into an $e^{-itL}u(t)$1-valued nonlinear Schrödinger equation and allowing a vector-valued long-range analysis in Schatten spaces (Kawamoto et al., 28 May 2026). In Vlasov–Riesz, a Lagrangian characteristic-flow construction of finite-time and infinite-time modified wave operators plays the analogous role (Hong et al., 5 Apr 2026).

5. Representative asymptotic formulas

In one-dimensional long-range cubic problems, the standard asymptotic form is a free Schrödinger kernel multiplied by a logarithmic phase depending on the asymptotic profile. For the averaged dispersion-managed cubic NLS,

$e^{-itL}u(t)$2

in $e^{-itL}u(t)$3 (Murphy et al., 2021). For the direct time-dependent dispersion-managed equation, the same law persists with $e^{-itL}u(t)$4 in place of a constant dispersion coefficient: $e^{-itL}u(t)$5 (Murphy et al., 2024).

In nonlocal Schrödinger models in $e^{-itL}u(t)$6, the phase records the effective Coulomb or Bopp–Podolsky interaction. For the Hartree equation,

$e^{-itL}u(t)$7

in $e^{-itL}u(t)$8 (Hoose, 2024). For the Schrödinger–Bopp–Podolsky equation, the modifier becomes

$e^{-itL}u(t)$9

so the long-range phase combines a nonlocal Coulomb-type term and the scattering-critical local power term (Hoose, 2023). Spinorial models exhibit parallel structures: in the projected Dirac/Hartree problem the renormalized Fourier profile $e^{i\Theta(t,\xi)}$0 converges, while in the massive Maxwell–Dirac system the spinor inside the light cone has linear massive oscillations $e^{i\Theta(t,\xi)}$1 multiplied by a logarithmic phase generated by the asymptotic Maxwell field (Cho et al., 2022, Herr et al., 2024).

In product-space problems, the asymptotic object is often an effective nonlinear orbit rather than a scalar phase. For the Diophantine waveguide equation,

$e^{i\Theta(t,\xi)}$2

so the nonlinear profile converges to a solution of the explicitly identified effective system (Camps et al., 2024). For the coupled cubic system on $e^{i\Theta(t,\xi)}$3,

$e^{i\Theta(t,\xi)}$4

where $e^{i\Theta(t,\xi)}$5 solves the resonant system (Rocha, 2016).

Operator-valued and kinetic theories replace phase modifiers by conjugations or trajectory corrections. In the critical time-dependent Kohn–Sham equation,

$e^{i\Theta(t,\xi)}$6

so the modifier is a logarithmic Fourier multiplier acting by conjugation on the density matrix (Kawamoto et al., 28 May 2026). In Vlasov–Poisson and Vlasov–Maxwell,

$e^{i\Theta(t,\xi)}$7

while in Vlasov–Riesz the long-range drift is

$e^{i\Theta(t,\xi)}$8

which reduces to a logarithmic correction only at the Coulomb-type threshold (Flynn et al., 2021, Breton, 3 Mar 2025, Hong et al., 5 Apr 2026).

6. Consequences, misconceptions, and current scope

A persistent misconception is that modified scattering signals a failure of dispersion. The literature surveyed here shows the opposite. Many long-range results prove the sharp linear decay rate simultaneously with modified scattering: $e^{i\Theta(t,\xi)}$9 in one-dimensional cubic Schrödinger problems, $e^{-itL}u(t)\to G(t)$0 in Hartree and Bopp–Podolsky models, and the corresponding optimal decay for operator-valued Kohn–Sham densities (Murphy et al., 2021, Hoose, 2023, Kawamoto et al., 28 May 2026). What fails is not decay, but convergence to an unmodified linear asymptotic state.

A second misconception is that modified scattering is always equivalent to inserting a scalar logarithmic phase. This is false in two directions. First, waveguide and product-space problems scatter to effective nonlinear orbits rather than to fixed profiles with scalar phases (Camps et al., 2024, Rocha, 2016). Second, kinetic equations require modified trajectories or reference flows rather than merely phase renormalization (Flynn et al., 2021, Hong et al., 5 Apr 2026). A plausible implication is that the correct asymptotic object is determined less by the formal degree of the nonlinearity than by the structure of the surviving resonances after all oscillatory averaging and reductions.

Recent work expands both the range of models and the admissible regularity. Wave-packet methods lower the threshold for Hartree and Schrödinger–Bopp–Podolsky modified scattering to data in $e^{-itL}u(t)\to G(t)$1 with $e^{-itL}u(t)\to G(t)$2, and the endpoint $e^{-itL}u(t)\to G(t)$3 is noted to be accessible with minor changes (Hoose, 10 Nov 2025). The critical time-dependent Kohn–Sham equation places modified scattering in an operator-valued density-matrix setting and states that it resolves the conjectures of Pusateri and Sigal in dimensions $e^{-itL}u(t)\to G(t)$4 for the critical case (Kawamoto et al., 28 May 2026).

At the same time, the current scope remains strongly perturbative and problem-dependent. Several results are explicitly small-data theorems; some treat only averaged or reduced models, as in the 2021 dispersion-managed NLS paper, which does not analyze the original time-dependent dispersion equation (Murphy et al., 2021). Even when the non-autonomous equation is treated directly, as in the 2024 cubic dispersion-managed NLS paper, the result is not formulated as full asymptotic completeness or construction of modified wave operators (Murphy et al., 2024). The operator-valued Kohn–Sham theory is restricted to $e^{-itL}u(t)\to G(t)$5 and positive semidefinite trace-class density matrices (Kawamoto et al., 28 May 2026), and the Vlasov–Riesz analysis leaves the regime $e^{-itL}u(t)\to G(t)$6 open (Hong et al., 5 Apr 2026).

Taken together, these works show that modified scattering is not a single theorem-type but a family of asymptotic descriptions for long-range nonlinear evolution: logarithmic phase renormalization, scattering to resonant effective systems, and convergence along corrected flows are all manifestations of the same principle, namely that the leading nonlinear residue is too large to disappear asymptotically, but sufficiently structured to be identified and integrated explicitly.