Journal of Mathematics, Physics and Mechanics

INTRODUCTION

The Korteweg-de Vries (KdV) equation, introduced in 1895 to describe the propagation of long waves in a rectangular canal, is among the most influential partial differential equations in all of science [1]. It balances a single quadratic nonlinear term against a third-order linear dispersive term, and it became the canonical model for unidirectional, weakly nonlinear, weakly dispersive wave propagation in shallow water, plasmas, and stratified fluids. The modern era of nonlinear science began in 1965 when Zabusky and Kruskal discovered, by numerical experiment, that the KdV equation possesses stable solitary-wave solutions whose collisions are elastic, and they coined the term soliton [2]. Their discovery launched the theory of integrable systems and transformed the study of nonlinear waves across physics.

The generalized Korteweg-de Vries (gKdV) equation extends the classical model by allowing a general nonlinearity,

$$u_t+f\left(u\right)u_x+u_{xxx}=0,$$

where f(u) is a smooth nonlinear function. The most studied case is the power-law nonlinearity f(u) = $u^p$, for which the equation takes the conservation form

$$u_t+{\left(\frac{u^{p+1}}{p+1}\right)}_x+u_{xxx}=0.$$

The choice p = 1 recovers the classical KdV equation, p = 2 gives the modified KdV (mKdV) equation, and the combined quadratic-cubic nonlinearity f(u) = u + $u^2$ yields the Gardner equation. For general p the equation is neither integrable nor variational in the Hamiltonian sense of the integrable cases, yet it possesses a remarkable structure: it conserves mass and energy, admits solitary-wave solutions of explicit algebraic form, and exhibits a sharp dichotomy between global existence and blow-up determined by the critical power p = 4 [3-4]. It is this tension between the integrable paradigm and the richness of non-integrable dynamics that makes the gKdV equation a central object of nonlinear analysis.

The aim of this review is to survey, critically and comprehensively, the mathematics and physics of the gKdV equation. We begin in Section 2 with the physical derivations from water waves, internal waves, and plasma theory. Section 3 develops the solitary-wave theory: existence, explicit form, orbital stability, and the sharp stability-instability dichotomy. Section 4 reviews the integrability theory of the special cases, including the inverse scattering transform, Lax pairs, the Miura transformation, and Hirota's bilinear method. Section 5 treats well-posedness and ill-posedness in Sobolev spaces. Section 6 is devoted to the critical-mass blow-up dynamics. Section 7 surveys numerical methods. Section 8 examines the extended gKdV family, including mKdV, Gardner, and KdV-Burgers equations. Section 9 treats fractional and stochastic generalizations. Section 10 reviews applications in oceanography, atmosphere, plasmas, and lattices. Section 11 covers machine learning and data-driven methods. Section 12 collects open problems, and Section 13 concludes.

Physical Derivations

The KdV equation was derived by Korteweg and de Vries to model long waves in a shallow channel, balancing weak nonlinearity with weak dispersion in the presence of surface tension [1]. In dimensionless form, the free-surface elevation satisfies the KdV equation to leading order in the small-amplitude and long-wavelength expansions, and the same reduction applies to a remarkable range of physical systems whose dispersion relation is approximately cubic at small wavenumber. The derivation in each context follows the same pattern: a multiple-scales or reductive perturbation expansion extracts the balance between nonlinearity and dispersion, and the leading-order evolution equation is of KdV or gKdV type. The ubiquity of this reduction is the reason the equation appears in hydrodynamics, plasma physics, geophysics, and elsewhere [5-6].

In water-wave theory, the KdV equation governs the free surface in the shallow-water regime, and higher-order corrections yield extended equations with additional nonlinear and dispersive terms. The generalized power-law models arise in a variety of settings, most notably for internal waves in density-stratified fluids, where the strength and sign of the nonlinear coefficient depend on the stratification and can vanish, in which case the cubic mKdV nonlinearity becomes the leading term [7-8]. Laboratory and field observations of internal solitary waves have been interpreted with KdV, extended KdV, and Gardner equations, and the review of Grimshaw, Pelinovsky, Talipova, and Kurkina synthesizes the propagation, deformation, and disintegration of internal solitary waves over topography [8-9]. In the ocean, internal solitary waves observed by satellite in the Andaman and Arabian Seas have been characterized and modelled with KdV-type equations [10]. Recent work by Sidorovas and collaborators extended the two-layer internal-wave theory to moderate amplitudes, deriving an extended KdV approximation with improved dispersion and demonstrating internal solitary and cnoidal waves in excellent agreement with Euler dynamics [11].

In plasma physics, the reductive perturbation analysis of ion-acoustic waves yields the KdV equation for the ion density, as shown by Washimi and Taniuti in 1966 [12]. The effect of finite ion temperature modifies the coefficient of the nonlinear term [13], and the inclusion of resonant particles produces modified equations with fractional nonlinearities of the type studied in the Schamel theory [14]. In shallow-water and channel flows, the KdV and its generalizations describe solitons, cnoidal waves, and undular bores, and the Whitham modulation theory provides a systematic framework for the slowly varying structure of such wave fields [5,15-16]. The generalized power-law models also arise in the description of nonlinear waves in lattices and granular chains, where the effective nonlinearity depends on the interaction potential [17].

Solitary-Wave Theory

The gKdV equation with power nonlinearity admits explicit solitary-wave solutions. Substituting the travelling-wave ansatz u(x,t) = ${\phi }_c(x-ct)$ with decay at infinity reduces the equation to the ODE -c φ + ${\phi }_{xx}$ + φ^(p+1)/(p+1) = 0, whose positive solution is

$${\phi }_c\left(x\right)={\left[\frac{(p+1)(p+2)}{2}c{\mathrm{sech}\left(\left(\frac{p}{2}\right)\sqrt{c}x\right)}^2\right]}^{\frac{1}{p}}.$$

These solitary waves exist for every p > 0 and c > 0, are even, smooth, exponentially decaying, and uniquely characterized as the ground states of the associated variational problem [4,18]. Their speed-amplitude relation, width-amplitude relation, and the L2 scaling law are determined by the power p, and these scalings control the stability theory [3,19].

The stability theory of gKdV solitary waves is one of the classical achievements of nonlinear analysis. Benjamin proved the orbital stability of KdV solitary waves by combining two conserved quantities with spectral theory [20]. Bona, Souganidis, and Strauss extended this to a general class of nonlinear dispersive equations, giving necessary and sufficient conditions for stability in terms of the linearized dispersion relation and the nonlinearity [19]. Grillakis, Shatah, and Strauss developed the general theory of orbital stability of solitary waves in Hamiltonian systems with symmetry, which applies uniformly to the KdV, mKdV, and NLS families [21-22]. Weinstein's Lyapunov approach established the stability of ground states and, decisively, located the critical power: for p < 4 the solitary waves are orbitally stable, while for p >= 4 they are unstable [4,18]. Höwing extended the stability analysis to large- and small-amplitude solitary waves in gKdV and Euler-Korteweg/Boussinesq equations [23], and Hong provided a self-contained treatment of existence and stability for the gKdV equations [24]. The log-KdV equation, a limiting case with logarithmic nonlinearity, admits Gaussian solitary waves whose orbital stability was established by Carles and Pelinovsky [25].

Beyond orbital stability, the asymptotic stability of gKdV solitary waves is a deep and subtle topic. Martel, Merle, and Tsai proved that subcritical gKdV solitary waves are asymptotically stable: perturbed solutions converge, in suitable norms, to a solitary wave of adjusted parameters plus dispersive radiation [26]. This result was the first of its kind for non-integrable dispersive equations and opened the modern theory of nonlinear asymptotic stability. The proof relies on a delicate virial and monotonicity analysis, and it establishes that the soliton acts as a nonlinear attractor for nearby dynamics [18,26]. These stability results are the essential background for the blow-up theory of the critical case, to which we turn in Section 6.

Integrability and Special Cases

The classical KdV equation (p = 1) is completely integrable. Gardner, Greene, Kruskal, and Miura showed that it can be solved by the inverse scattering transform: the solution at any time is recovered from the scattering data of a Schrödinger operator whose potential is the initial datum [27]. Lax reformulated this in the elegant operator form dL/dt = [B,L], establishing a Lax pair for KdV and providing a general principle connecting nonlinear evolution equations with linear operators whose spectra are invariant [28]. The Miura transformation u = v^2 + v_x relates solutions of the KdV and mKdV equations, a discovery that was decisive for the inverse scattering method [29]. Miura, Gardner, and Kruskal exploited the transformation to derive an infinite hierarchy of conservation laws for the KdV family [30], and the complete theory of the inverse scattering transform, including the direct and inverse problems and the time evolution of scattering data, was developed by Gardner, Greene, Kruskal, and Miura [31].

Hirota's bilinear method provides an algebraic route to the exact N-soliton solutions of the KdV and mKdV equations, exhibiting their elastic collisions explicitly [32-33]. The KdV hierarchy, generated by the recursion operator, consists of integrable flows of increasingly high order, and its finite-gap solutions are constructed by algebro-geometric methods; the periodic spectral transform of Osborne and Bergamasco revisits the Zabusky-Kruskal experiment from the perspective of the periodic inverse problem [34]. The inverse scattering theory for general first-order systems developed by Beals and Coifman provides the unified framework within which the KdV, mKdV, and related equations are solved [35-36]. The soliton theory is presented in the classical monographs of Ablowitz and Clarkson, of Drazin and Johnson, of Linares and Ponce, and of Infeld and Rowlands [6,37-39]. The long-time asymptotics of the defocusing mKdV equation in weighted Sobolev spaces was established by Chen and Liu via the nonlinear steepest descent method of Deift and Zhou, showing that the solution decomposes into dispersive radiation governed by a Painlevé-type formula [40].

For the gKdV equation with general p, the integrable cases are exceptional. The equation is completely integrable only for p = 1 (KdV) and p = 2 (mKdV); the Gardner equation, with combined quadratic-cubic nonlinearity, is also integrable and is related to the mKdV equation by a suitable transformation [41-42]. The non-commutative and matrix generalizations of the KdV and mKdV hierarchies, which describe integrable systems with operator-valued variables, were constructed by Carillo and Schiebold via recursion methods [42-43]. For generic p the gKdV equation admits no Lax pair and its soliton interactions are inelastic, a fact that sharply separates the integrable cases from the general theory [3,44]. Nevertheless, explicit exact solutions can be obtained for the power-law cases by direct integration, and the recent literature contains many constructions of solitary, cnoidal, and periodic solutions of the generalized equations, including N-soliton solutions of the combined KdV-CDG and KdV-Lax hierarchies [41,45-47]. Negative-order integrable KdV and mKdV equations, introduced by Wazwaz, exhibit multiple complex soliton solutions and enrich the integrable family [46].

Well-Posedness And Ill-Posedness

The Cauchy problem for the gKdV equation in Sobolev spaces has been studied intensively since the 1980s. The foundational work of Ginibre, Tsutsumi, and Velo established local and global well-posedness for the generalized equations in $H^s$ for appropriate s and p [48]. Saut initiated the systematic study of the dispersive properties of the gKdV family and its generalizations [3]. Kenig, Ponce, and Vega proved sharp local well-posedness for the KdV equation in $H^s$, s > -3/4, by introducing the smoothing estimates that now bear their names [49], and extended the contraction-principle method to the generalized equations [50-51]. Bourgain's Fourier restriction norm method, which encodes the dispersive smoothing in the structure of the function space, gave global well-posedness in L2 for the subcritical cases [52]. Grujić and Kalisch proved local well-posedness of the generalized KdV equation in spaces of analytic functions, extending the theory to Gevrey-type classes [53]. The ill-posedness of the data-to-solution map for the generalized equations at low regularity was established by Birnir, Kenig, Ponce, Svanstedt, and Vega [54].

Sharp results in low-regularity well-posedness continue to develop. Masaki and Segata established well-posedness for the gKdV equation in scale-critical $L^r$ spaces, identifying the critical regularity from the scaling law [55]. Wu proved global well-posedness for the periodic generalized Korteweg-de Vries equation in a broad range of spaces [56]. Hwang and Kwak developed the probabilistic well-posedness of the generalized KdV equation, showing that random data below the deterministic thresholds still yield unique local solutions with high probability [57]. Molinet, Talhouk, and Zaiter proved well-posedness for KdV-type equations with variable coefficients, a setting relevant to inhomogeneous media [58]. The definitive low-regularity theory for the integrable case was achieved by Killip and Visan, who proved that the KdV equation is well-posed in $H^{-1}$ by exploiting the full integrable structure and the conservation laws at negative regularity [59]; Bringmann, Killip, and Visan extended this to the fifth-order KdV equation in $H^{-1}$ [60]. The invariance of white noise for KdV on the line, established by Killip, Murphy, and Visan, provides a further remarkable result at the interface of integrability and probability [61].

Boundary-value problems and dissipative variants complete the well-posedness picture. Faminskii proved global well-posedness of two initial-boundary-value problems for the KdV equation on the half-line and on bounded intervals [62]. Nunes established global well-posedness for the transitional KdV equation [63]. Vento proved global well-posedness for dissipative KdV equations, in which a Burgers-type term damps the dynamics [64]. The recent work of Li and Li in 2026 establishes uniform well-posedness and the inviscid limit for the KdV-Burgers and mKdV-Burgers equations on the circle, providing the rigorous foundation for the vanishing-dissipation analysis of the generalized equations [65]. These results show that the well-posedness theory of the gKdV family is essentially complete across the deterministic, probabilistic, boundary, and dissipative settings [51-52,59].

Critical-Mass Blow-Up Dynamics

The critical power p = 4 separates the stable from the unstable regimes of the gKdV solitary waves, and the corresponding dynamics are dramatically different. For p < 4, the L2-norm is subcritical and all solutions in the energy space are global [4,48]. For p >= 4, solitary waves are unstable [4,20], and for p > 4 solutions with negative energy blow up in finite time, as shown by Bona and Saut in their study of dispersive blow-up [66]. The critical case p = 4 is the most delicate: solutions of L2-mass equal to the mass of the ground state may blow up, while those with smaller mass are global [4,18].

The modern theory of critical blow-up was initiated by Martel and Merle. They proved a Liouville theorem for the critical gKdV equation that excludes certain global dynamics and is the key to the classification of blow-up [67], and established the instability of the critical solitons [68]. They then proved the stability of the blow-up profile and obtained lower bounds for the blow-up rate [69], demonstrated that blow-up occurs in finite time and described the dynamics of blow-up solutions for data near the soliton [70], and proved the nonexistence of blow-up solutions with minimal L2-mass [71]. These papers established the fundamental picture: near-soliton data of critical mass either converge to the soliton or blow up in finite time, concentrating the mass in a self-similar profile.

The definitive description of the blow-up dynamics was achieved by Martel, Merle, and Raphaël. In a sequence of papers they proved that blow-up for the critical gKdV equation occurs with the universal soliton-like profile, described the minimal-mass blow-up solution, and constructed exotic regimes with non-generic rates and infinite-time growth [72-74]. In particular, they established the sharp upper bound on the blow-up rate, the stability of the blow-up profile, and the classification of all possible blow-up behaviours near the ground state [72]. Martel and Pilod constructed a full family of flattening solitary waves for the critical equation, showing that slowly decaying tails can force the soliton to flatten over time [75]. The blow-up theory of the gKdV equation is now regarded as a model for the analogous phenomena in the nonlinear Schrödinger equation and other critical dispersive equations, and it is presented in the surveys of Martel, Merle, and Raphaël [74]. The nonexistence of minimal-mass blow-up and the universality of the critical dynamics remain guiding principles for the field [71,73].

Numerical Methods

The numerical solution of the gKdV equation has a long history, beginning with the pioneering computation of Zabusky and Kruskal, whose observation of soliton interactions was made possible by a finite-difference scheme of modest accuracy [2]. Early fully discrete Galerkin methods for the KdV equation were analyzed by Bona, Dougalis, and Karakashian, establishing the rigorous foundation for high-order spatial discretizations [76]. Iskandar developed new finite-difference numerical solutions that capture the soliton propagation with high fidelity [77]. The central numerical issue for the gKdV family is the preservation of the invariants (mass, energy, and higher conserved quantities), since uncontrolled dissipation or dispersion changes the long-time dynamics qualitatively [78-79].

Modern structure-preserving methods have been developed specifically for the gKdV equation. Bona, Chen, Karakashian, and Xing constructed conservative discontinuous Galerkin methods that preserve discrete versions of the mass and L2-norm, and showed numerically that these conservation properties impart the approximations with beneficial attributes, including more faithful reproduction of the amplitude and phase of travelling-wave solutions and only linear growth of the error in time [78]. Karakashian and Makridakis derived a posteriori error estimates for discontinuous Galerkin methods for the gKdV equation, providing the adaptive framework for efficient computation [80], and Karakashian and Xing obtained a posteriori estimates for the conservative local discontinuous Galerkin scheme [79]. For the integrable cases, numerical inverse scattering provides an alternative and highly accurate approach: Trogdon, Olver, and Deconinck developed numerical methods for the direct and inverse scattering problems of the KdV and mKdV equations, enabling the numerical solution of the Cauchy problem through the integrable structure [81].

Spectral and pseudospectral methods are widely used for the KdV and gKdV equations because of their high accuracy for smooth solutions. The time evolution of dispersive shock waves and undular bores, which arise from step-like initial data, has been studied numerically with Whitham modulation theory as a guide, and the comparison between modulation-theoretic predictions and direct numerical simulation is a recurring theme in the literature [15-16]. For the generalized equal-width and coupled KdV-type equations, a variety of explicit and approximate schemes have been proposed, including the soliton and undular-bore solutions constructed by Başhan [82] and the periodic and soliton solutions of the two-mode KdV-Burgers equation obtained by Gomez [83]. The accurate numerical simulation of the critical gKdV equation near blow-up requires carefully adapted schemes, since the solution develops a concentration profile; the results of Martel-Merle-Raphaël provide the benchmarks against which such simulations are validated [69,72].

The Extended Gkdv Family

The gKdV equation is the hub of a large family of related nonlinear evolution equations. The modified KdV equation (p = 2) is integrable, admits both kink and algebraic soliton solutions, and is connected to the KdV equation through the Miura transformation [29,33]. The Gardner equation, with the combined nonlinearity f(u) = u + $u^2$, arises naturally when the quadratic coefficient of the internal-wave or plasma reduction vanishes, and it interpolates between the KdV and mKdV regimes [7-8]. The KdV-Burgers equation, which adds a dissipative term $u_{xx}$ to the KdV equation, describes the combined effects of nonlinearity, dispersion, and dissipation and admits explicit shock-like travelling-wave solutions [65,84]. The variable-coefficient and coupled generalizations of the family model inhomogeneous media and multicomponent systems [85-87].

Table 1: Comparison of The Extended Generalized Korteweg-De Vries Family of Evolution Equations

Note: M denotes the mass; E denotes the energy; c denotes the solitary-wave speed; p is the nonlinearity power.

The linear and nonlinear stability properties of the extended family have been studied in depth. The two-dimensional instabilities of generalized KdV waves were analyzed by Murawski, who showed that the KdV-type solitary waves are unstable to transverse perturbations in certain regimes [86]. The soliton-like interactions of the generalized KdV equation were studied by Shih, who demonstrated that the non-integrable power-law solitons interact inelastically with a systematic exchange of mass and energy [44]. The generation of radiating waves in singularly perturbed KdV equations, by Benilov, Grimshaw, and Kuznetsova, describes how higher-order perturbations force the solitary waves to shed radiation [94]. The long-time dynamics of variable-coefficient mKdV solitary waves, analyzed by Dejak and Jonsson, establishes that slowly varying coefficients force the soliton parameters to evolve adiabatically [85]. These results connect the exact-solution theory of the integrable cases with the rich phenomenology of the non-integrable family [41,87].

Beyond the classical extensions, the gKdV family has been generalized in several mathematical directions. The log-KdV equation, with logarithmic nonlinearity, admits Gaussian solitary waves and provides a model for the transition between power-law regimes [25]. The generalized geophysical KdV model with nonlinear power law, analyzed by Adeyemo, exhibits diverse solitary, cnoidal, and Weierstrass-type solutions and admits conservation laws derived by the multiplier method [95]. The Whitham equation and its capillary-gravity generalizations, which replace the third-order dispersion by the exact linear dispersion, have been studied with computer-assisted proofs of existence and stability of solitary waves [92]. In two spatial dimensions, the generalized Kadomtsev-Petviashvili equations, which reduce to the gKdV equation in the appropriate limit, admit nontrivial solitary waves whose existence was proved by He and Zou [93]. Solitary waves in particle lattices with power-law forces, analyzed by Ingimarson and Pego, connect the gKdV-type long-wave theory to discrete systems [17]. These developments show that the gKdV family continues to generate new mathematics at the interface of analysis, computation, and physics [25,44,92].

Fractional and Stochastic Generalizations

Fractional-order generalizations of the gKdV equation have received considerable attention. The time-fractional generalized Korteweg-de Vries equation, in which the time derivative is replaced by a fractional derivative in the Caputo or Riemann-Liouville sense, was formulated and solved via variational methods by Zhang, who obtained approximate solutions that reproduce the classical solitons in the integer-order limit [89]. Qayyum, Ahmad, Afzal, and Acharya extended the He-Laplace algorithm to generalized third-order time-fractional KdV models, providing convergent series solutions and a thorough convergence and error analysis [90], and Qayyum, Ahmad, Riaz, and Awrejcewicz applied the Laplace transform with homotopy perturbation to generalized fifth-order time-fractional KdV models including the Sawada-Kotera, Ito, Lax, and Kaup-Kupershmidt equations [96]. Nagpal derived series and wave solutions of the time-fractional generalized KdV equation by Lie symmetry analysis combined with the (G'/G)-expansion method [91]. These studies demonstrate that the soliton and wave solutions of the classical equations survive, with modified amplitudes and widths, in the fractional settings [89-90].

Stochastic generalizations of the gKdV family model the effects of random forcing on nonlinear wave propagation. The stochastic KdV equation, with white-noise forcing, has been studied since the 1980s, and its solutions exhibit a balance between nonlinearity, dispersion, and noise [64-65]. The coupled generalized KdV system driven by white noise was proved locally well-posed in $H^s$ x $H^s$, s > 1/2, using Bourgain-space multilinear estimates, Itô's formula, and a fixed-point argument, improving earlier results for the solitary-wave system [58]. The uniform well-posedness and inviscid limit for the KdV-Burgers and mKdV-Burgers equations on the circle, established by Li and Li in 2026, provides the deterministic backbone for the stochastic and dissipative analyses [65]. The interplay between integrability, noise, and dissipation in the gKdV family remains an active research area with connections to turbulence theory and data assimilation [97-98].

Applications in Oceans, Atmosphere, Plasmas, and Lattices

In oceanography, the KdV and gKdV equations describe the propagation of internal solitary waves in stratified coastal waters. The modelling framework of Grimshaw, Pelinovsky, and Talipova synthesizes the KdV, extended KdV, and Gardner descriptions of internal waves and their applications to the coastal ocean [7], and the review of Grimshaw, Pelinovsky, Talipova, and Kurkina documents the propagation, deformation, and disintegration of internal solitary waves over the continental shelf [8]. Numerical solutions of the KdV and extended KdV equations were compared with mooring observations in the Middle Atlantic Bight by O'Driscoll and Levine, showing that the leading internal tide steepens into a shock-like front from which packets of solitary waves evolve [9]. Satellite observations of internal solitary waves in the Andaman Sea and Arabian Sea, characterized by Mandal, Seemanth, and Ratheesh using EOS-04 and Sentinel data, provide the observational context for the modelling [10]. The shallow-water soliton dynamics beyond the KdV equation were examined by Karczewska, Rozmej, and Infeld, who assessed the validity of the KdV description against higher-order models [99].

In plasma physics, the gKdV equation describes ion-acoustic solitary waves and their stability. The Washimi-Taniuti derivation established the KdV equation for ion-acoustic waves in 1966 [12], and the effects of ion temperature and resonant particles were incorporated in subsequent generalizations [13-14]. The KdV soliton gas, consisting of a dense random ensemble of interacting solitons, is described by the spectral kinetic theory reviewed by El, which connects the Whitham modulation theory with the thermodynamics of the integrable system [97]. Agafontsev, Gelash, Mullyadzhanov, and Zakharov showed that bound-state soliton gases arise as limits of adiabatically growing integrable turbulence, and that such gases exhibit extreme statistical properties [98]. Pelinovsky and Shurgalina studied the interactions and turbulence of KdV soliton gases, demonstrating the role of soliton collisions in the statistical dynamics [100]. In atmosphere and ocean modelling, the Gardner equation describes internal solitary waves in the presence of both quadratic and cubic nonlinearity, and the generalized geophysical power-law models extend the theory to more complex stratifications [7,95].

In nonlinear lattices and granular chains, the long-wavelength limit of the equations of motion yields gKdV-type equations whose power depends on the interaction potential. Ingimarson and Pego proved the existence of small solitary waves in particle lattices with power-law forces, showing that the solitary-wave profiles converge to those of the fractional KdV or generalized Benjamin-Ono equations [17]. The Whitham equation, which retains the exact linear dispersion while keeping the KdV-type nonlinearity, has been studied with computer-assisted proofs of existence and spectral stability of solitary waves by Cadiot [92]. These lattice and Whitham applications demonstrate the reach of the gKdV paradigm beyond the classical shallow-water and plasma contexts [5,17,92].

Machine Learning and Data-Driven Methods

Machine learning methods have recently been applied to the KdV and gKdV equations, both as forward solvers and as tools for inverse problems. Physics-informed neural networks (PINNs) embed the differential equation into the loss function and have been applied to the KdV equation for internal solitary waves by Kang and collaborators, who demonstrated operator learning from parameters to solutions with prediction errors as low as 10-4 and robust inverse estimation of the nonlinear coefficients from sparse and noisy data [101]. The prior-information PINNs developed by Tian, Li, and Zhang simulate the Ai- and Bi-soliton solutions of the cylindrical KdV equation with improved accuracy of one to three orders of magnitude relative to standard PINNs, enhanced further by the imposition of energy conservation [88]. These data-driven methods complement the classical numerical schemes reviewed in Section 7 and are particularly valuable when the model parameters are unknown or the data are noisy [88,101].

The soliton solutions of the gKdV family provide an ideal testbed for data-driven and operator-learning methods, since exact solutions serve as ground truth for training and validation [45,81]. The statistical theory of soliton gases and integrable turbulence, reviewed in Section 10, provides a further connection: machine-learning-based forecasting of extreme events in integrable turbulence is an emerging direction that leverages the exact soliton structure of the KdV family [97-98]. The data-driven prediction of nonlinear wave phenomena in fibre optics and water waves, which are governed by the NLS and KdV families, has demonstrated that neural operators can forecast the onset of instabilities in real time, and similar approaches are being developed for the gKdV equation [101]. As the mathematical theory of operator learning matures, the gKdV equation is likely to join the canonical benchmarks of scientific machine learning [88,101].

OPEN PROBLEMS AND FUTURE DIRECTIONS

Several fundamental questions about the gKdV equation remain open. First, the complete classification of the dynamics at the critical mass, including the possibility of multi-bubble concentration and the behaviour of solutions with mass strictly between the ground-state mass and the blow-up threshold, is not fully resolved beyond the near-soliton regime [69,72]. Second, the sharp description of the dispersive radiation emitted during blow-up, and the precise rate of mass concentration, remain subjects of active research [70,73]. Third, the extension of the stability and blow-up theory to the coupled, fractional, and stochastic generalizations of the equation is still in its infancy [25,65,89]. Fourth, the rigorous derivation of the gKdV equation from the Euler and Vlasov systems over physically relevant time scales, with sharp error estimates, remains a challenging and partially open problem [5,7].

In applications, several directions stand out. The role of soliton gases in the statistical description of integrable turbulence, and their connection to the rogue-wave phenomenon, deserves further development [97-98,100]. The data-driven modelling of internal solitary waves from satellite and mooring observations, using physics-informed and operator-learning methods, is an emerging theme with direct practical value [10,101]. The description of internal waves of moderate and large amplitude, beyond the weakly nonlinear regime, requires the higher-order and exact-dispersion models whose analysis is only beginning [11,92]. The extension of the gKdV theory to variable-coefficient, dissipative, and forced settings, motivated by the coastal and atmospheric observations, offers a rich agenda for both theory and computation [58,65,94].

From a computational perspective, the design of structure-preserving schemes that preserve the higher-order invariants of the gKdV equation, and the extension of the conservative discontinuous Galerkin methods to the critical, fractional, and stochastic variants, are important goals [78-79,90]. The development of adaptive methods based on the a posteriori error estimates of Karakashian and co-authors will enable efficient computation near blow-up and in the presence of dispersive shock waves [16,80]. The integration of machine learning into the numerical and analytical workflow, including operator learning for the inverse problem and data-driven discovery of conserved quantities, is a promising direction [88,101]. Finally, the interplay between integrability, randomness, and dissipation in the gKdV family, and its role as a model for nonlinear wave phenomena across scales, guarantees that the equation will continue to generate new problems at the interface of analysis, probability, numerics, and physics [65,95,97].

CONCLUSION

The generalized Korteweg-de Vries equation occupies a central place in nonlinear science. It unifies the classical KdV equation of 1895 with a large family of equations that govern unidirectional, weakly nonlinear, weakly dispersive waves in water, plasmas, and stratified fluids [1,12]. Its solitary-wave theory, from the classical stability theorems of Benjamin and Bona-Souganidis-Strauss to the modern asymptotic stability results of Martel-Merle-Tsai, provides a model for the stability of localized coherent structures in conservative systems [19-20,26]. Its integrability theory, built on the inverse scattering transform, Lax pairs, and Hirota's method, established the paradigm of completely integrable nonlinear systems [27-28,32]. And its blow-up theory, developed to its sharpest form by Martel, Merle, and Raphaël, provides the canonical example of critical-mass concentration dynamics in dispersive equations [70,72].

The trajectory of research on the gKdV equation mirrors the development of modern nonlinear analysis: a physical observation, the solitary wave of Russell, gave rise to an equation of extraordinary mathematical depth, which in turn became a laboratory for the theory of integrability, stability, well-posedness, and blow-up [1-2,59]. The recent literature, including the work of 2025 and 2026 on extended internal-wave models, physics-informed neural networks, probabilistic well-posedness, and the inviscid limit for KdV-Burgers equations, shows that the subject remains extraordinarily vibrant [11,57,65,88,91-92,101]. We expect that the interplay between analysis, probability, computation, and physical observation will continue to drive progress on this remarkable equation for decades to come.

DECLARATION

The authors declare that they have no competing interests, no data, and no funding.