Journal of Mathematics, Physics and Mechanics

INTRODUCTION

The Schrödinger equation stands at the centre of modern physics. Introduced in 1926 to describe the dynamics of matter waves, it became the governing equation of non-relativistic quantum mechanics and the foundation of atomic, molecular, condensed-matter, and nuclear physics [1]. Its linear form

$$i\hslash {\psi }_t=-\frac{{\hslash }^2}{2m}{\nabla }^2\psi +V\left(x\right)\psi$$

dictates that the wave function ψ evolves unitarily, with the modulus squared |ψ|² interpreted as a probability density. The equation reconciles wave and particle pictures, predicts quantized energy levels in bound systems, and reduces to classical mechanics in the semiclassical limit [1-2]. Its relativistic counterpart for the electron, derived by Dirac, gave birth to the modern theory of fermions and spin [3]. The Feynman path-integral formulation recast the equation as a statement about interference of amplitudes over all classical trajectories, cementing its role as the central object of quantum theory [2].

The scope of this review is broader than the linear equation of quantum mechanics. In settings as diverse as optical fibres, surface water waves, plasmas, and dilute quantum gases, slowly modulated wave packets obey a nonlinear Schrödinger (NLS) equation, in which dispersion balances cubic (or more general) nonlinearity [4-5]. This nonlinear equation supports solitons, breathers, and rogue waves, and its study has become a discipline of its own, connecting exactly solvable systems to a wide range of physical applications [6-8]. The aim of this review is to survey this landscape critically and comprehensively: the physical origins of the equation in its linear and nonlinear forms, the mathematical theory of integrability and well-posedness, the phenomenology of solitons and extreme waves, the computational methods used to solve it, and the emerging roles of the equation in quantum information science and scientific machine learning.

The structure of the review is as follows. Section 2 recalls the physical and mathematical foundations of the linear Schrödinger equation. Section 3 describes the NLS equation, the inverse scattering transform, and the integrability theory of the soliton-bearing nonlinearities. Section 4 reviews the applications to water waves, modulation instability, and rogue waves. Section 5 covers optical solitons and fibre-optic communications. Section 6 treats the Gross-Pitaevskii equation and Bose-Einstein condensation. Section 7 surveys well-posedness, scattering, and blow-up theory. Section 8 reviews numerical methods. Section 9 examines fractional, stochastic, and generalized Schrödinger equations. Section 10 discusses quantum simulation, quantum computing, and machine learning. Section 11 collects open problems, and Section 12 concludes.

Critical Synthesis and Comparative Assessment

The literature on the Schrödinger equation and its nonlinear generalizations is often presented as a chronological catalogue of results. We instead organise it critically around four research programmes, because they answer different questions and are subject to different limitations. The first is the exact, integrability-based programme: the inverse scattering transform, algebro-geometric constructions, and Bcklund–Darboux methods establish the existence of solitons, breathers, and quasi-periodic solutions and expose the hidden linearity of the integrable NLS family [4,6-15]. Such methods are decisive for questions of existence and spectrum, but they speak directly only to exactly integrable models and say little about stability or observability in a real environment. The second is the asymptotic and perturbative programme, which derives the NLS equation from microscopic models by multiple-scale or reductive-perturbation techniques and follows near-integrable dynamics adiabatically [18-25]. It is the essential bridge between the mathematical equation and physical systems, but its validity is confined to weak nonlinearity and weak perturbation. The third is the computational programme of structure-preserving and spectral integrators, which quantifies specific scenarios and long-time behaviour. Its conclusions are only as trustworthy as the conservation and resolution properties of the scheme, which are frequently under-reported. The fourth is the emerging data-driven and quantum-computational programme, which replaces explicit solution by surrogate models or hardware simulation [25-28]. It offers reach where classical analysis stalls, but currently lacks rigorous error control and is validated mostly on small, well-conditioned problems.

A striking feature of the 2025–2026 literature is how these programmes are beginning to overlap. Machine learning now forecasts noise-driven instabilities governed by the NLS equation, physics-informed neural networks are proposed as mesh-free PDE solvers, quantum optimal control supplies the pulse-engineering layer for quantum simulation, and the qualitative theory of fractional models is maturing. At the same time, two tensions run through the field. First, the number of exact-solution studies continues to grow faster than the number of quantitative, error-barred comparisons with experiment; many newly reported solitons and breathers are never assessed for stability or experimental accessibility. Second, the elegant statistical theory of integrable turbulence remains only loosely connected to the extreme-value statistics actually observed in optics and the ocean [29-33]. The comparative summary tables introduced below (Tables 1–6) are designed to make these methodological distinctions and limitations explicit, and Section 11 collects the resulting open problems. We do not claim that any single programme is definitive; rather, confident conclusions require agreement between at least two of them.

Foundations of The Linear Schrödinger Equation

The historical genesis of the equation is well documented. Schrödinger, building on de Broglie's hypothesis of matter waves, sought a wave equation whose stationary solutions reproduce the quantized spectra of the hydrogen atom; the resulting eigenvalue problem for the energy produced the correct Balmer series and marks the birth of wave mechanics [1]. The equation's probabilistic interpretation, developed by Born, made |ψ|² a probability density and transformed the equation into the central object of quantum theory, with the measurement postulate and the uncertainty principle following from its structure [2-3]. The Dirac equation, a relativistic first-order system for the electron, predicted spin and antimatter, and its non-relativistic limit contains the Pauli equation; the Schrödinger equation thus survives as the low-energy limit of relativistic quantum theory [3].

Mathematically, the linear Schrödinger equation is a dispersive partial differential equation whose solutions oscillate and spread rather than localize. In the free case, V = 0, the Fourier transform diagonalizes the operator, and the evolution is generated by the multiplication by exp(-i|ξ|² t), an isometry on L² that preserves neither positivity nor compact support; the group of evolution operators is unitary on L² and the equation is time-reversible [9-10]. The spectral problem -i ħ ∇² ψ = E ψ subject to boundary conditions yields discrete and continuous spectra, underpinning the theory of bound states and scattering states [10]. The equation is at the heart of the modern spectral and scattering theory for Schrödinger operators, which quantifies how external potentials modify plane waves and give rise to resonances and decay estimates [10-11].

The path-integral formulation of Feynman provides a complementary view. In this picture the transition amplitude between two points is a sum over all paths, each weighted by exp(i S/ħ) where S is the classical action; the Schrödinger equation emerges as the infinitesimal form of this sum [2]. This formulation underlies semiclassical analysis, the WKB approximation, and the modern theory of quantum-field-theoretic and statistical-mechanical analogies [2]. The path integral also motivates the study of the equation in random environments and in the semiclassical regime, where interference patterns become deterministic at the level of the Wigner function and hydrodynamic variables [2,9].

The linear equation is the workhorse of atomic and molecular physics. The time-dependent Schrödinger equation with a time-dependent potential describes light-matter interaction, strong-field ionization, and the dynamics of electrons in intense laser fields [12-13]. Efficient numerical propagation of the equation is required for such simulations, and the availability of robust, high-order methods has been essential to the field [12-14]. The equation also governs the dynamics of electrons in nanostructures and quantum dots, where envelope-function models reduce to effective Schrödinger equations with position-dependent masses [13]. In all these applications the linear structure - spectral analysis combined with unitary propagation - is the essential mathematical machinery [10,12].

The Nonlinear Schrödinger Equation and Integrability

The nonlinear Schrödinger equation in its canonical form

$i{\psi }_t+{\psi }_{xx}+2{\left|\psi \right|}^2\psi =0$

arises whenever a carrier wave of finite bandwidth propagates in a dispersive medium with a cubic (Kerr-like) nonlinearity. The balance between dispersion and nonlinearity gives rise to envelope solitons, and the equation governs modulational instability, four-wave mixing, and the formation of extreme waves [4-5,15]. The one-dimensional cubic NLS equation is completely integrable: Zakharov and Shabat showed that it admits a Lax pair and is solvable by the inverse scattering transform, establishing it as a member of the small family of exactly solvable nonlinear evolution equations [4,15]. The AKNS scheme places the NLS equation in a unified framework with the Korteweg-de Vries and sine-Gordon equations, sharing the same linear-problem structure [4,15].

The integrability theory of the NLS family is deep and multifaceted. The inverse scattering transform (IST) reduces the initial-value problem to a linear scattering problem, followed by the time evolution of scattering data and the reconstruction of the solution by Gelfand-Levitan-Marchenko equations [6]. For the NLS equation the scattering problem is the Zakharov-Shabat system, whose spectral theory is fully understood [6,15]. The IST is not confined to the NLS equation itself: a hierarchy of integrable equations is generated by higher-order flows of the same Lax pair, and the theory extends to multi-component systems, derivative NLS equations, and equations with other power-law nonlinearities [16-17]. Krichever's algebro-geometric methods construct quasi-periodic solutions of integrable equations in terms of Riemann theta functions, connecting the integrable NLS family to algebraic geometry [18]. The Hamiltonian formulation of the theory, developed by Faddeev and Takhtajan, places the NLS equation in the setting of classical and quantum integrable systems, with action-angle variables provided by the scattering data [19].

Perturbation theory for near-integrable systems is a central theme of the modern theory. When the NLS equation is perturbed by damping, higher-order dispersion, or driving, solitons evolve adiabatically, and their parameters obey closed dynamical systems that generalize Newton's equations [20-21]. The singular perturbation theory of Kaup and Newell describes solitons as particles under weak perturbations, while the variational formulation of Bondeson, Lisak, and Anderson provides a systematic method for computing the evolution of soliton parameters in nearly integrable systems [20-21]. These perturbative techniques are indispensable for the quantitative description of optical soliton transmission in realistic fibres, where loss, gain, and higher-order effects are always present [20,22].

Beyond the cubic case, a hierarchy of integrable and near-integrable nonlinear Schrödinger-type equations arises in applications. The coupled NLS equations describe two-component and vector solitons, with bright-bright, dark-dark, and bright-dark soliton families obtained by exact methods [17]. The derivative NLS equation, with the nonlinearity ψ² ψ_x^*, describes Alfvén waves and femtosecond pulse propagation, and is integrable through a different scattering problem [20]. The general power-law NLS equation i ψt + ψ_xx + |ψ|^2σ ψ = 0 is non-integrable for generic σ, and its solitons interact inelastically, providing a laboratory for the physics of non-integrable solitary waves [21]. The pure-quartic solitons of Blanco-Redondo and collaborators, which balance quartic dispersion against cubic nonlinearity, extend the soliton concept to higher-order dispersion and have been demonstrated experimentally in optical waveguides [23]. The recent work of Zhang and Chen on the inverse scattering transformation of generalized NLS equations extends the IST to nonlocal and higher-order models [24].

Notwithstanding its elegance, integrability is a measure-zero property, and this shapes how the exact theory should be used. The inverse scattering transform provides a complete solution of the integrable NLS equation, but realistic fibre, plasma, and fluid systems always include loss, gain, higher-order dispersion, and noise, so the exact result must be supplemented by perturbation theory [20-22]. Moreover, soliton interactions in the generic (non-integrable) power-law NLS equation are inelastic, and the extensive catalogue of exact solutions to modified or fractional models frequently establishes existence without addressing stability or physical relevance [21,24]. A critical reading of the literature therefore distinguishes three claims that are often conflated: integrability of the model, existence of a formal solution, and stability or observability of that solution. Only the first two are guaranteed by the analytic machinery; the third requires separate analysis. Table 1 collects the main forms of the equation and their principal applications, and Table 2 summarises the analytical methods together with their advantages and limitations.

Table 1: Major Forms of The Schrödinger Equation and Representative Applications.

Table 2: Analytical Methods for The Schrödinger Equation: Scope, Advantages, and Limitations.

Water Waves, Modulation Instability, and Rogue Waves

The NLS equation was derived independently in hydrodynamics to describe the slow modulation of a weakly nonlinear Stokes wave train. Hasimoto and Ono obtained the equation for nonlinear modulation of gravity waves on deep water, and the derivation was later made rigorous within the multiple-scales framework [25]. The equation governs the Benjamin-Feir instability: a uniform Stokes wave train is unstable to sideband perturbations, leading to the disintegration of the wave train and the emergence of localized structures [26]. Peregrine's exact solution, a rational function localized in both space and time, describes the maximum amplification of the carrier during the instability and is today known as the Peregrine soliton or Peregrine breather [7].

The instability and breather dynamics of the NLS equation provide the canonical model for rogue, or freak, waves. The Akhmediev breather, periodic in space and localized in time, describes the nonlinear stage of the Benjamin-Feir instability, while the Kuznetsov-Ma breather is periodic in time and localized in space [7,27]. The Peregrine soliton is the limiting form of both families and is the simplest prototype of a rogue wave that appears from nowhere and disappears without a trace [27]. Experimental confirmation of these structures in optical fibres has been decisive: Kibler and collaborators observed the Kuznetsov-Ma soliton dynamics and the Peregrine soliton in a standard optical fibre, confirming the NLS prediction quantitatively [28]. Akhmediev, Soto-Crespo, and Ankiewicz showed how to excite a rogue wave from a weakly perturbed background and characterized the signatures of the phenomenon [29]. The classification of rogue waves and their generating mechanisms in different physical contexts is the subject of comprehensive reviews [8,30].

The theory of integrable turbulence places rogue-wave formation in the broader context of statistical wave dynamics. In the integrable NLS equation, ensembles of waves evolve into a complex turbulent state characterized by deterministic but highly irregular dynamics, in which modulational instability seeds the intermittent appearance of large-amplitude events [31-32]. Single-shot experiments using time microscopy have directly observed optical rogue waves in integrable turbulence, and numerical and theoretical studies have established that breather-like structures, rather than individual solitons, dominate the extreme events [31-32]. Xu and collaborators observed breather wave molecules, bound states of multiple breathers that interact coherently and exhibit the signature of multi-breather solutions [33]. Kibler and collaborators reported superregular breathers, solutions that emerge from a background with no visible modulation and grow to large amplitude, in both optics and hydrodynamics [34].

In the oceanographic context, the role of the NLS equation in explaining real-world rogue waves has been actively debated. Early studies suggested that the modulational instability is the dominant mechanism for deep-water extreme waves [35]. However, a careful analysis of realistic sea states, including finite bandwidth, directional spreading, and crossing seas, has shown that the modulational instability alone cannot account for many observed rogue waves, and that linear superposition of random waves, wind forcing, and four-wave interactions play comparable roles [35-36]. The comprehensive monograph of Kharif, Pelinovsky, and Slunyaev synthesizes the state of the art on oceanic rogue waves, covering both the NLS-based theories and the statistical approaches [37]. Recent reviews emphasize the generating mechanisms across different physical contexts, from optics to plasmas to superfluids [8,30,38]. The debate illustrates a general methodological lesson: exact solutions of the NLS equation are essential but insufficient, and quantitative predictions require careful modelling of the physical environment [36-37].

Table 3: Representative Soliton and Breather Solutions of The Nonlinear Schrödinger Equation.

Optical Solitons and Fibre-Optic Communications

Optical fibre is the most important physical realization of the NLS equation. Hasegawa and Tappert derived the NLS equation for pulse propagation in dispersive dielectric fibres and predicted the existence of optical solitons in the anomalous dispersion regime [5]. The experimental confirmation by Mollenauer, Stolen, and Gordon demonstrated the narrowing of picosecond pulses and the formation of solitons in optical fibres, launching the field of soliton-based optical communications [39]. The Kivshar-Luther-Davies review covers the physics of dark optical solitons, the counterparts that propagate in the normal dispersion regime as localized dips on a continuous background [40]. The integrability of the NLS equation in this setting means that solitons survive collisions and are robust to perturbations, properties exploited in high-capacity transmission systems [6,20].

The engineering of optical solitons has required a subtle interplay between physics and mathematics. The perturbed NLS equation with loss, gain, and higher-order dispersion describes realistic propagation, and the variational and singular perturbation methods reviewed in Section 3 provide the design rules for dispersion-managed and soliton-based systems [21-22]. Serkin and Hasegawa constructed exactly integrable NLS models with varying dispersion, nonlinearity, and gain, which admit soliton solutions whose parameters adapt adiabatically to the varying medium [22]. Beyond the classical soliton, the study of dissipative solitons in fibre lasers has become a rich field, with mode-locked lasers supporting solitons, bound soliton pairs, and soliton molecules; the review of Song and collaborators surveys recent progress [41].

A major modern development is the use of the nonlinear Fourier transform (NFT) for signal processing in optical communications. Just as the linear Fourier transform is the basis of coherent optical transmission, the nonlinear Fourier transform - the direct and inverse scattering transform of the NLS equation - offers a representation of the signal in which the nonlinear channel is diagonalized [6,42]. The foundational papers of Yousefi and Kschischang developed the mathematical tools of NFT-based information transmission, opening the possibility of nonlinear frequency-division multiplexing [42]. While the practical channel rates remain below the linear baselines in many regimes, the approach has spawned a substantial literature on soliton-based and eigenvalue-based communications, and it connects the exact integrability of the NLS equation directly to information theory [15,42].

Rogue-wave physics in fibre optics has matured into a quantitative discipline. The comprehensive review of Copie, Randoux, and Suret summarizes the physics of the one-dimensional NLS equation in optical fibres, covering rogue waves, modulation instability, and self-focusing phenomena in a single coherent framework [43]. Single-shot temporal measurements, which require ultrafast diagnostics with time microscopes, have resolved the fine structure of breather collisions and the statistics of extreme events in real fibre experiments [32]. In 2025, Boussafa and collaborators demonstrated the deep-learning prediction of noise-driven nonlinear instabilities in optical fibres, combining the NLS-based forward model with data-driven forecasting of the onset of modulation instability and rogue-wave generation [44]. This work exemplifies the convergence of integrable theory, ultrafast metrology, and machine learning that characterizes the current frontier of the field [43-44].

The Gross-Pitaevskii Equation and Bose-Einstein Condensation

In a dilute Bose gas, the macroscopic wave function of the condensate obeys the Gross-Pitaevskii equation, a nonlinear Schrödinger equation with a local cubic nonlinearity proportional to the s-wave scattering length [45]. The equation was derived independently by Gross and Pitaevskii to describe the structure of quantized vortices in boson systems, and it remains the standard mean-field description of Bose-Einstein condensates (BECs) [45]. The experimental realization of BECs in dilute atomic vapors by Anderson, Ensher, Matthews, Wieman, and Cornell confirmed the mean-field theory and opened a new chapter in atomic physics [46]. The Gross-Pitaevskii equation predicts the density profile of the condensate, the collective excitations, and the dynamics of vortices and solitons, all of which have been verified in experiments [47-48].

Soliton physics in BECs is governed by the sign of the scattering length. In repulsive condensates, dark solitons - localized density dips with a phase slip across the core - form and have been observed directly [47,49]. The dynamics of dark solitons in elongated condensates, their collisions, and their decay into vortex rings have been studied experimentally and theoretically [48,50]. In attractive condensates, bright matter-wave solitons form and their collapse dynamics have been characterized [51]. The collapse of condensates with negative scattering length, leading to the explosive ejection of atoms, was observed by Donley and collaborators, confirming the theoretical predictions of the Gross-Pitaevskii theory in the focusing regime [52]. The observation of universal quench dynamics and Townes soliton formation from modulational instability in two-dimensional Bose gases provided a direct experimental connection between BEC dynamics and the modulation-instability theory of the NLS equation [53]. Dark-bright soliton trains in superfluid-superfluid counterflow and solitons in condensates with helicoidal spin-orbit coupling extend the phenomenology to more complex multicomponent settings [54-55].

The theory of BEC solitons is closely tied to the exact and numerical solutions of the Gross-Pitaevskii equation. Lattice solitons in condensates held in periodic optical potentials were predicted by Efremidis and Christodoulides, extending the physics of discrete NLS systems to ultracold atoms [56]. Exact soliton solutions and nonlinear modulation instability in spinor Bose-Einstein condensates were obtained by Li and collaborators, linking the vector NLS family to multicomponent condensates [57]. The dynamics of bright solitons with time-dependent scattering length in expulsive parabolic potentials, which map onto exactly solvable NLS models, were analyzed by Liang, Zhang, and Liu [58]. Soliton creation during the BEC phase transition, driven by the Kibble-Zurek mechanism, was studied by Damski and Zurek, connecting the Gross-Pitaevskii dynamics to the theory of topological defect formation [59].

Beyond the standard cubic equation, the BEC context motivates a variety of generalized nonlinear Schrödinger models. Ultradilute low-dimensional liquids, governed by a Gross-Pitaevskii-type equation with a modified nonlinearity that supports self-bound droplets, were predicted and observed in dipolar quantum gases [60]. The theory of dark solitons in atomic BECs from theory to experiments, as reviewed by Frantzeskakis, unifies the single- and multi-component descriptions and highlights the role of the NLS hierarchy in ultracold physics [47]. Numerical methods for the Gross-Pitaevskii equation, including time-splitting spectral discretizations, are essential for quantitative predictions in inhomogeneous traps, and are discussed further in Section 8 [61].

Well-Posedness, Scattering, and Blow-Up

The rigorous theory of the nonlinear Schrödinger equation is one of the triumphs of modern nonlinear analysis. For the initial-value problem of the NLS equation in L² and H^s spaces, the local and global well-posedness are governed by the balance between the power of the nonlinearity, the spatial dimension, and the regularity of the data [9,11]. The classical results of Kenig, Ponce, and Vega established local well-posedness for the small-amplitude NLS family in L² and H^s, and the theory of low-regularity well-posedness for canonical defocusing equations was developed by Christ, Colliander, and Tao [9,11]. Ill-posedness phenomena in the data-to-solution map for the generalized KdV and NLS equations were characterized by Birnir, Kenig, Ponce, Svanstedt, and Vega, sharpening the picture of exactly how regular the data must be for a unique solution [62].

Scattering theory describes the large-time behaviour of solutions. In the defocusing case, solutions disperse to free waves, and the asymptotic completeness of the nonlinear scattering operator has been established for the subcritical and critical nonlinearities in dimensions d≥2 [10,63]. Soffer and Weinstein proved multichannel nonlinear scattering for nonintegrable equations, establishing the existence of nonlinear bound states and their role in the asymptotic dynamics [10]. The celebrated result of Dodson proved global well-posedness and scattering for the defocusing, L²-critical NLS equation in dimensions d=1 and d=2, settling a long-standing open problem [63-64]. In the focusing case, scattering is generically absent, and the dynamics are controlled by the threshold between global existence and blow-up [65-66].

The blow-up theory for the focusing NLS equation is a subject of great delicacy. Below the critical mass, solutions exist globally; at the critical mass, the ground-state soliton sets the threshold, and mass-critical blow-up solutions concentrate in self-similar or pseudoconformal regimes [65]. Merle and Raphaël determined the blow-up dynamics and the upper bound on the blow-up rate for the critical nonlinear Schrödinger equation, establishing the universal character of the blow-up profile [65]. Holmer and Roudenko analyzed blow-up solutions to the three-dimensional cubic NLS equation, classifying the dynamics near the threshold [66]. In higher-order models, the energy-critical fourth-order Schrödinger equation was shown to be globally well-posed for radial data by Pausader, connecting the blow-up and scattering theory to the energy-critical dispersive analysis [67].

For damped and driven systems, the long-time behaviour is governed by attractors. Ghidaglia proved the existence of a finite-dimensional attractor for weakly damped, driven Schrödinger equations, extending the theory of infinite-dimensional dynamical systems to dispersive equations [68]. The interaction of the NLS dynamics with boundaries, in the form of transparent boundary conditions and open boundaries for the nonlinear Schrödinger equation, was analyzed by Soffer and Stucchio, providing a rigorous foundation for the numerical simulation of waves on unbounded domains [69]. These results connect the abstract well-posedness theory to the concrete demands of numerical computation, a theme developed in the next section [68-69].

Numerical Methods

The numerical solution of the Schrödinger equation, in both its linear and nonlinear forms, has a long and rich history. For the time-dependent linear equation, the split-step Fourier method, finite-difference schemes, and spectral methods are the standard tools, with the choice dictated by the structure of the potential and the required accuracy [12-14]. Time-splitting spectral discretizations for the nonlinear Schrödinger equation, developed by Bao, Jin, and Markowich, are explicit, unconditionally stable, and conserve the position density, making them the method of choice for the semiclassical regime and for the Gross-Pitaevskii equation [61]. The interaction picture method, which advances the solution in a frame that cancels the linear phase, was developed for the generalized NLS equation in optics by Balac and collaborators [70]. High-order compact splitting multisymplectic methods for the coupled NLS equations, by Ma and collaborators, preserve the multisymplectic structure and achieve high-order accuracy in long-time simulations [71].

For atomic and molecular applications, the propagation of the time-dependent Kohn-Sham and Schrödinger equations requires methods that are both accurate and efficient. Castro, Marques, and Rubio developed propagators for the time-dependent Kohn-Sham equations, the workhorse of time-dependent density functional theory, based on the Magnus expansion and polynomial approximations of the propagator [12]. Spectral methods in hyperspherical coordinates for the time-dependent Schrödinger equation, by Sørevik, Madsen, and Hansen, enable the simulation of electron dynamics in few-body systems [14]. The generalized FDTD method with absorbing boundary conditions for the time-dependent linear Schrödinger equation, by Moxley, Zhu, and Dai, extends the finite-difference time-domain methodology to quantum dynamics [13].

Energy-conserving and structure-preserving methods are essential for long-time fidelity. Barletti, Brugnano, Frasca-Caccia, and Iavernaro developed energy-conserving methods for the nonlinear Schrödinger equation based on the Hamiltonian boundary value method, which exactly preserves the invariants of the continuous equation [72]. The analytical and numerical solutions of the one-dimensional NLS equation have been compared in benchmark studies, establishing the accuracy of the numerical schemes against known exact solutions [73]. Stability, convergence, and error analysis of B-spline collocation combined with Crank-Nicolson and finite element methods for the Schrödinger equation arising in quantum mechanics were carried out by Jena and Senapati, providing rigorous error estimates for practical schemes [74].

The interplay between numerical methods and the exact integrable structure of the NLS family is a recurring theme. The inverse scattering transform itself is used as a numerical method: the nonlinear Fourier transform provides a spectral representation in which the evolution is diagonal, and fast algorithms for the direct scattering transform have been developed for optical communications [6,42]. Benchmarking numerical schemes against the exact soliton and breather solutions of the NLS equation ensures that the structure of the solution - its integrable core - is preserved under discretization [71,73]. The availability of exact solutions of the generalized NLS equations, such as the inverse scattering solutions of Zhang and Chen, provides a rigorous testbed for the numerical methods of the field [24,70].

Comparison of these families exposes limitations that the literature does not always make explicit. Splitting and spectral schemes are efficient and spectrally accurate on periodic domains, but they conserve energy only to their order of accuracy, and their long-time phase error must be monitored rather than assumed [61]. Structure-preserving integrators conserve the invariants exactly but are typically implicit and costlier [71-72]. Finite-difference and finite-element schemes are geometrically flexible but suffer numerical dispersion that can masquerade as physical instability [74]. A recurring weakness is benchmarking practice: schemes are frequently validated against a single smooth soliton over short times, whereas the demanding tests are long-time conservative or chaotic regimes and problems with rough or stochastic data. The physics-informed neural-network literature faces the complementary problem that accuracy is difficult to certify, and reported errors are rarely compared against a spectral benchmark [100-103]. Table 4 summarises the resulting trade-offs.

Table 4: Numerical methods for the Schrödinger equation: advantages and limitations.

Fractional, Stochastic, and Generalized Schrödinger Equations

The fractional quantum mechanics of Laskin replaces the kinetic term of the Schrödinger equation by a fractional Laplacian, obtained from a path integral over Lévy flights rather than Brownian motion [75]. The fractional Schrödinger equation supports solutions with novel localization properties, and its physical interpretation in terms of Lévy statistics has been developed extensively [75-76]. The time fractional Schrödinger equation, in which the time derivative is fractional, was revisited by Achar, Yale, and Hanneken, who clarified the structure of its solutions and the role of the fractional order [76]. Iomin's review of fractional evolution in quantum mechanics situates the fractional Schrödinger equation in the broader context of fractional dynamics and anomalous diffusion [77]. The fractional Schrödinger equation with time-dependent potentials, studied by Gabrick and collaborators, reveals how fractional order modifies the wave dynamics in external fields [78].

The fractional NLS equation has become an active field at the interface of nonlinear waves and fractional calculus. Solitons in the cubic-quintic fractional nonlinear Schrödinger model were studied by Stephanovich and collaborators, who characterized the stability and existence of localized states in the fractional medium [79]. The optimal control of fractional quantum systems governed by the Caputo time-fractional Schrödinger equation, analyzed by Mohamed and Qamlo in 2026, connects the fractional equation to modern control theory [80]. The qualitative analysis of an unstable time-fractional nonlinear Schrödinger equation, including bifurcation, quasi-periodic, chaotic behaviour, and exact solutions, was carried out by El-Dessoky, Elmandouh, and Alghamdi in 2026, highlighting the rich dynamical structure of the fractional models [81]. New approaches to one-dimensional time-fractional Schrödinger equations, by El-Ajou and collaborators, provide efficient analytical techniques for the fractional models [82]. Traveling Journal of Mathematics, Physics and Mechanics wave solutions for the complex space-time fractional Kundu-Eckhaus equation, which shares the envelope-wave structure of the NLS family, were constructed by Alabedalhadi and collaborators, further enriching the exact-solution toolkit of fractional wave equations [83].

Stochastic nonlinear Schrödinger equations model the effects of random forcing and multiplicative noise on wave dynamics. Mora and Rebolledo established the basic properties of nonlinear stochastic Schrödinger equations driven by Brownian motions, including existence, uniqueness, and regularity of solutions [84]. Barbu, Röckner, and Zhang studied the stochastic logarithmic Schrödinger equation, establishing the well-posedness of a strongly singular stochastic model with applications to dissipative quantum dynamics [85]. The exact solutions of the (2+1)-dimensional stochastic chiral nonlinear Schrödinger equation, by Albosaily and collaborators, provide explicit expressions for the stochastic wave fields under multiplicative noise [86]. Gisin and Rigo analyzed the relevant and irrelevant nonlinear Schrödinger equations arising in the theory of quantum measurements and dissipative dynamics, connecting the stochastic Schrödinger picture to the theory of quantum trajectories [87-88].

The stochastic Schrödinger equation is, in fact, the natural language of quantum measurement theory. The quantum trajectories of Wiseman, describing the state evolution conditioned on a continuous measurement record, obey a stochastic differential equation of the diffusive type [88]. The theory of quantum trajectories and quantum measurement theory has been developed into a complete framework for open quantum systems, with applications in quantum optics, quantum control, and quantum information [87-88]. The quantum dynamics with short-time trajectories and minimal adaptive basis sets, by Saller and Habershon, connects the stochastic trajectory picture to efficient computational schemes for molecular quantum dynamics [89]. The Koopman-von Neumann approach to quantum simulation of nonlinear classical dynamics, by Joseph, uses a Schrödinger-type equation to embed nonlinear classical dynamics into a linear quantum simulation, establishing a surprising and fruitful connection between the Schrödinger equation and classical chaos [90].

The fractional and stochastic extensions are conceptually inviting but should be read with caution. Fractional models enlarge the solution space and reproduce anomalous behaviour, yet the fractional order and the noise amplitude are usually phenomenological parameters whose physical calibration is rarely demonstrated; moreover, time-fractional equations are not unitary and do not preserve the probabilistic interpretation of the wave function in the ordinary sense [76-77]. Stochastic NLS models have a solid analytic foundation, but their rigorous results — existence, uniqueness, regularity — are not accompanied by quantitative statistical predictions that can be compared with experiment, and the rare-event regime that motivates the rogue-wave applications remains largely beyond rigorous reach [84-85]. A similar gap separates the mathematically well-posed stochastic Schrödinger equation of measurement theory from the engineered nonlinear variants sometimes proposed as fundamental, which can change the predictions of quantum mechanics in ways not supported by experiment [87]. Table 5 organises these formulations and flags their principal limitations.

Table 5: Fractional and stochastic Schrödinger formulations.

Quantum Simulation, Quantum Computing, and Machine Learning

The Schrödinger equation is the starting point for quantum simulation. Lloyd's universal quantum simulators established that a quantum computer could efficiently simulate the dynamics of any local quantum system, a result that remains the principal argument for the practical value of quantum computing [91]. The comprehensive review of Georgescu, Ashhab, and Nori surveys the state of the art of quantum simulation, covering both analog and digital approaches to simulating the Schrödinger equation and its many-body generalizations [92]. The experimental milestone of quantum supremacy, achieved by Arute and collaborators on a programmable superconducting processor, demonstrated that quantum hardware can outperform classical computers on a specialized sampling task, and it has intensified efforts toward the simulation of the Schrödinger equation on quantum hardware [93].

Variational quantum algorithms provide a near-term route to the simulation of the Schrödinger equation. Cerezo and collaborators reviewed the theory and practice of variational quantum algorithms, which encode the wave function in a parameterized quantum circuit and optimize the parameters to minimize the expectation value of the energy [94]. The variational ansatz-based quantum simulation of imaginary time evolution, by McArdle and collaborators, adapts the classical imaginary-time propagation to the variational quantum setting, enabling the computation of ground states on noisy intermediate-scale quantum devices [95]. The optimal control of quantum systems, including the analytic gradients for low-rank quantum optimal control developed by Goutte and Savona in 2026, provides the pulse-engineering tools needed to implement the unitary operations of quantum simulation [96]. The universal analog quantum simulation framework of Huang and collaborators in 2026 extends the reach of analog quantum simulation to a broad class of quantum dynamics [97].

Quantum sensing exploits the Schrödinger equation to measure physical quantities with unprecedented precision. The review of Degen, Reinhard, and Cappellaro surveys the principles and applications of quantum sensing, in which quantum systems - spins, atoms, photons - serve as sensors whose Hamiltonians depend on the quantities of interest [98]. The prediction and real-time compensation of qubit decoherence via machine learning, by Mavadia and collaborators, demonstrates how data-driven methods stabilize quantum evolution against environmental noise, a capability central to practical quantum sensing and simulation [99]. These applications rely on the ability to solve, control, and measure the Schrödinger equation in real devices, connecting the mathematical theory of Section 2 to the engineering of quantum technologies [98-99].

Machine learning has become a powerful complement to the classical and quantum methods for the Schrödinger equation. Physics-informed neural networks (PINNs), introduced by Raissi, Perdikaris, and Karniadakis, embed the differential equation directly into the loss function of a neural network, enabling the solution of forward and inverse problems for the Schrödinger equation without labelled data [100]. The scientific machine learning framework of Cuomo and collaborators reviews where the field stands and what comes next, covering the training, architecture, and application of PINNs to a wide range of PDEs including the Schrödinger family [101]. The comprehensive review of Luo and collaborators in 2025 surveys physics-informed neural networks for PDE problems, with a detailed account of the Schrödinger applications [102]. The 2026 review of Khan and collaborators provides a comprehensive assessment of recent architectural variants and optimization strategies for PINNs, including the residual-based adaptive methods that are particularly effective for the highly oscillatory solutions of the Schrödinger equation [103].

The convergence of quantum simulation, quantum control, and machine learning defines the frontier of the field. Physics-informed neural networks are being used both as classical surrogates for the Schrödinger equation and as building blocks for the classical control and calibration of quantum devices [100-101]. Deep learning methods have been demonstrated for the prediction of noise-driven nonlinear instabilities in fibre optics, where the NLS equation governs the dynamics, showing that data-driven methods can forecast the onset of rogue-wave events in real time [44]. The machine-learning-based stabilization of quantum dynamics, as demonstrated in the decoherence-compensation experiments, relies on exactly the same interplay between data, model, and control [99]. Beyond quantum technologies, the Schrödinger equation underpins emerging fields such as quantum biology, where quantum-coherent effects in photosynthetic complexes, avian navigation, and enzymatic reactions are modelled with the same linear and nonlinear Schrödinger machinery reviewed here [104]. These developments suggest that the Schrödinger equation, the oldest pillar of quantum theory, is also at the centre of its most modern computational methods [96,102-103].

The computational and quantum approaches reviewed here are best regarded as complementary tools with clearly delimited domains of validity. Fault-tolerant quantum simulation promises exponential advantage, but today’s noisy devices implement only small, specially structured instances, and resource estimates for simulating genuinely challenging Schrödinger dynamics with controlled error remain sobering [91,93-94]. Variational algorithms are hardware-friendly but can be defeated by barren plateaus and by measurement overhead that grows with system size [94-95]. Physics-informed neural networks are attractive for inverse and high-dimensional problems, yet they exhibit spectral bias against the highly oscillatory solutions characteristic of the Schrödinger equation, and their error is hard to certify [100-103]. A critical, common gap is benchmarking: promising new methods are often validated on low-dimensional, smooth problems, so their scaling to the regimes where classical methods already succeed is unknown. Table 6 summarises this landscape. Bridging these methods to the exact and asymptotic programmes, with shared benchmarks and error metrics, is arguably the most valuable near-term contribution the community could make.

Table 6: Computational, quantum-simulation, and machine-learning approaches.

OPEN PROBLEMS AND FUTURE DIRECTIONS

Several fundamental questions about the Schrödinger equation and its nonlinear generalizations remain open. First, the complete characterization of the long-time behaviour in the focusing regime, including the precise asymptotics of blow-up solutions and the role of the ground state, is not fully understood beyond the mass-critical threshold [65-66]. Second, the rigorous derivation of the NLS equation from microscopic physical models - water waves, plasmas, Bose gases, and optics - remains a mathematically active problem, with the control of the approximation error over physically relevant time scales still incomplete [25,61]. Third, the theory of integrable turbulence, in which the statistics of the NLS wave field are computed from the integrable structure, is only partially developed, and the connection between breather statistics and rogue-wave probability is an active area of research [31-32].

In applications, several directions stand out. The design of practical nonlinear Fourier transform communication systems that outperform linear baselines remains a challenging engineering goal [42]. The control of rogue waves, whether through the design of initial conditions, the management of dispersion and nonlinearity, or real-time prediction with machine learning, is an emerging theme with potential applications in optics, oceanography, and the protection of infrastructure [36,44]. In quantum gases, the extension of the Gross-Pitaevskii description beyond mean-field theory, including the coupling to quantum fluctuations and the dynamics of entangled many-body states, is a frontier of ultracold atomic physics [47,60]. In fractional and stochastic models, the physical validation of the equations and the development of structure-preserving numerical methods are priorities [75,81,84].

In the quantum-information direction, the simulation of the Schrödinger equation on fault-tolerant quantum computers, the design of variational ansätze that capture the relevant entanglement structure, and the certification of quantum simulators are central goals [91-92,94]. The optimal control of quantum systems, including the low-rank analytic-gradient methods of 2026, will be essential to implement the required gates and pulses with high fidelity [96]. The integration of physics-informed machine learning into the control and calibration loop of quantum devices, as demonstrated by the decoherence-compensation experiments, is likely to become standard practice [99,101]. The use of the Schrödinger equation as a benchmark for scientific machine learning, given its exact solutions and rich phenomenology, is an opportunity that the community is only beginning to exploit [100,103].

From a mathematical perspective, the extension of the well-posedness and scattering theory to the fractional and stochastic Schrödinger equations is still in its early stages [75,85]. The theory of open boundaries and transparent conditions for nonlinear dispersive equations on unbounded domains, initiated by Soffer and Stucchio, deserves further development in higher dimensions and for the generalized NLS family [69]. The rigorous theory of energy-conserving and multisymplectic integrators for the coupled NLS systems, and their use in the long-time simulation of integrable turbulence, is an active area [71-72]. Finally, the classification of the many generalized Schrödinger equations that arise in applications - derivative, nonlocal, coupled, fractional, stochastic, and higher-order - and the identification of those that are integrable, remains a unifying mathematical theme [17,24,82].

Taken together, this survey suggests four cross-cutting challenges that run through the specific open problems listed above. The first is stability and selection: the exact-solution literature continues to generate new solitons, breathers, and fractional solutions faster than it evaluates their stability or physical accessibility, so a systematic stability analysis of newly reported structures, and a clear statement of the physical conditions under which they can be observed, would materially strengthen the field [21,79,83]. The second is validation: quantitative, error-barred comparisons with experiment and observation remain scarce outside optical fibres, and benchmarking protocols shared between the analytical, numerical, and machine-learning communities are largely absent [36,44,73]. The third is the modelling of noise and disorder: fractional and stochastic formulations are proliferating, yet their parameters are seldom fixed from first principles, and the rare-event statistics that motivate them lie precisely in the regime where rigorous and numerical control is hardest [84-85,102]. The fourth is rigorous error control for the data-driven and quantum methods, which is currently the weakest link between the promise of these approaches and their reliability as scientific instruments [91,94,103]. Addressing these challenges calls for the attitude that the mature parts of the subject already display: agreement between independent methods, explicit error estimates, and honest statements of the limits of each approach.

CONCLUSION

The Schrödinger equation, in its linear and nonlinear forms, is among the most consequential equations of science. It governs the quantum dynamics of matter, from atoms to condensates, and it emerges as a universal model of envelope waves wherever dispersion balances nonlinearity [1,5,45]. Its integrability theory, developed through the inverse scattering transform, made it the prototype of an exactly solvable nonlinear system and provided the mathematical language for solitons, breathers, and the rogue waves of optics and the ocean [4,7,15]. Its numerical solution underpins atomic, molecular, and condensed-matter simulation [12,61]. And its recent intersections with quantum simulation, quantum control, and machine learning have made it a central object of the emerging quantum technologies [91,94,100].

The trajectory of research on the Schrödinger equation mirrors the development of modern physics: a single equation, born from the attempt to describe the hydrogen atom, grew into a universal model whose reach now extends from the foundations of quantum mechanics to the frontiers of optical communications, ocean engineering, ultracold gases, and quantum information science [1-2,92]. The literature is vast, and the recent contributions of 2025 and 2026 - including deep-learning prediction of nonlinear instabilities, physics-informed neural networks for PDEs, quantum optimal control, and the qualitative theory of fractional equations - show that the subject remains extraordinarily vibrant [44,80-81,96-97,102-103]. We expect the interplay between analysis, computation, and experiment to continue to drive progress on this remarkable equation for decades to come.

DECLARATIONS

The author declares no competing interests, no data, and no funding.