Discrete Nonlinear Wave Equations

• Discrete nonlinear wave equations are lattice-based models that incorporate nonlinearity and discrete spatial operators to produce unique wave propagation behaviors.
• They enable the study of phenomena like localization, anti-continuum continuation, and coherent structures using variational and Hamiltonian methods.
• Analytical and numerical approaches focus on well-posedness, energy conservation, and scattering theory, ensuring that key physical invariants are preserved.

Searching arXiv for the cited papers to ground the article in current metadata and ensure accurate arXiv identifiers. Discrete nonlinear wave equations (DNLW) are lattice or network evolution equations in which nonlinearity interacts with spatial discreteness, intersite coupling, and often a Hamiltonian or variational structure. In the narrow sense, the term usually refers to second-order-in-time discrete Klein–Gordon, $\Phi^4$, or semilinear wave systems; in a broader dispersive-lattice usage it also includes semi-discrete DNLS-type, Ablowitz–Ladik, and related first-order lattice field equations. Across these usages, the common theme is that wave propagation is mediated by discrete spatial operators—nearest-neighbor differences, graph Laplacians, or more general lattice couplings—so that localization, scattering, stability exchange, pinning, and blow-up are shaped by the underlying discrete geometry rather than by continuum translation invariance alone (Caputo et al., 2018, Lepri et al., 2012, Ablowitz et al., 2022).

1. Terminological scope and structural setting

DNLW is not a single equation but a family of discrete models. One recurrent class is the discrete nonlinear Klein–Gordon or $\Phi^4$-type equation, including network generalizations such as

$$\frac{d^2 \mathbf{u}}{dt^2}=\left(-\omega^2 \mathbf{I}+\epsilon \mathbf{\Delta}\right)\mathbf{u}-\mathbf{u}^3,$$

posed on a finite graph with symmetric negative semidefinite graph Laplacian $\mathbf{\Delta}=\mathbf{A}-\mathbf{D}$; here arbitrary topology and weighted edges replace the translation-invariant lattice Laplacian (Caputo et al., 2018). Another major class is the discrete Klein–Gordon/wave equation on $h\mathbb Z^d$,

$$\partial_t^2 u_n-\Delta_h u_n+V_n u_n+\lambda |u_n|^{2\sigma}u_n=0,$$

which includes the discrete wave equation as the massless or near-massless case and supports a broad $l_h^p$-based well-posedness theory (Wu et al., 2023). A third class is fully discrete spacetime schemes such as the Strauss–Vazquez discretization for $U(1)$-invariant nonlinear wave equations, where the discrete model itself, not merely its continuum limit, carries exact energy and sometimes charge conservation (Comech et al., 2010).

In the broader nonlinear-wave sense, DNLS-type equations are also treated as canonical discrete nonlinear wave models because they combine lattice dispersion, onsite cubic nonlinearity, defect-mediated scattering, and localized excitations in analytically tractable settings. This broader usage is explicit in work on asymmetric DNLS dimers, fractional Ablowitz–Ladik hierarchies, and long-time traveling waves for DNLS near the continuum limit (Lepri et al., 2012, Bernier et al., 2018, Ablowitz et al., 2022). A persistent source of confusion is therefore terminological rather than mathematical: some subfields reserve DNLW for second-order lattice waves, whereas others include first-order discrete dispersive equations whenever the central objects are nonlinear waves on a lattice. The literature itself supports both conventions (Ablowitz et al., 2022, Cuevas-Maraver et al., 2014).

2. Representative equations and canonical model classes

The field is organized around a small number of recurrent operator structures.

Model class	Representative equation	Salient feature
Graph $\Phi^4$-type wave	$$\ddot{\mathbf u}=(-\omega^2\mathbf I+\epsilon\mathbf\Delta)\mathbf u-\mathbf u^3$$	Arbitrary finite graph topology (Caputo et al., 2018)
Discrete Klein–Gordon / wave	$\Phi^4$0	$\Phi^4$1-well-posedness and blow-up theory (Wu et al., 2023)
Inhomogeneous DNLS	$\Phi^4$2	Nonreciprocal nonlinear scattering (Lepri et al., 2012)
Lattice semilinear wave on $\Phi^4$3	$\Phi^4$4	Wave operators and asymptotic completeness (Wang, 25 Jul 2025)
Fully discrete $\Phi^4$5-invariant wave scheme	Strauss–Vazquez finite-difference scheme	Exact discrete energy, conditional stability (Comech et al., 2010)

These classes differ in time order, phase space, and conserved structure. Graph and lattice Klein–Gordon equations are naturally Hamiltonian and second order in time; DNLS and Ablowitz–Ladik models are first order, gauge invariant, and mass conserving; fully discrete wave schemes replace continuum invariants by two-time-level discrete analogues (Comech et al., 2010, Bernier et al., 2018). The discrete operator may be local, as in nearest-neighbor lattices, or genuinely nonlocal, as in semi-discrete convolution-wave systems

$\Phi^4$6

whose spatial discretization yields an infinite nonlinear ODE system with long-range kernel-induced coupling rather than a short-range Laplacian (Erbay et al., 2018). At a more abstract level, damped nonlinear wave equations driven by a positive operator $\Phi^4$7 with discrete spectrum also fit the DNLW landscape because spectral decomposition reduces the dynamics to countably many damped oscillator modes (Ruzhansky et al., 2017).

3. Localized states, anti-continuum continuation, and coherent structures

A central DNLW theme is the persistence of localized states under weak coupling. On arbitrary finite graphs, a one-site large-amplitude excitation in the nonlinear graph wave equation is asymptotically reduced to a hard Duffing oscillator at the excited node, while the remaining nodes satisfy a linear forced system on the punctured graph. The localization mechanism is frequency detuning: for $\Phi^4$8, the excited site oscillates with nonlinear frequency $\Phi^4$9 proportional to the initial amplitude $$\frac{d^2 \mathbf{u}}{dt^2}=\left(-\omega^2 \mathbf{I}+\epsilon \mathbf{\Delta}\right)\mathbf{u}-\mathbf{u}^3,$$0, and localization requires this nonlinear frequency to lie well above the reduced linear band, expressed qualitatively as

$$\frac{d^2 \mathbf{u}}{dt^2}=\left(-\omega^2 \mathbf{I}+\epsilon \mathbf{\Delta}\right)\mathbf{u}-\mathbf{u}^3,$$1

with $$\frac{d^2 \mathbf{u}}{dt^2}=\left(-\omega^2 \mathbf{I}+\epsilon \mathbf{\Delta}\right)\mathbf{u}-\mathbf{u}^3,$$2 the maximal normal eigenfrequency of the reduced graph (Caputo et al., 2018). The same work identifies topology-dependent selection rules for which reduced modes are forced, and it shows that the localization threshold in the $$\frac{d^2 \mathbf{u}}{dt^2}=\left(-\omega^2 \mathbf{I}+\epsilon \mathbf{\Delta}\right)\mathbf{u}-\mathbf{u}^3,$$3-plane scales like $$\frac{d^2 \mathbf{u}}{dt^2}=\left(-\omega^2 \mathbf{I}+\epsilon \mathbf{\Delta}\right)\mathbf{u}-\mathbf{u}^3,$$4.

Near the continuum limit, coherent transport is controlled by a different mechanism: weak breaking of continuous translation symmetry. For the one-dimensional focusing cubic DNLS, Bernier and Faou construct band-limited traveling-wave profiles $$\frac{d^2 \mathbf{u}}{dt^2}=\left(-\omega^2 \mathbf{I}+\epsilon \mathbf{\Delta}\right)\mathbf{u}-\mathbf{u}^3,$$5 satisfying

$$\frac{d^2 \mathbf{u}}{dt^2}=\left(-\omega^2 \mathbf{I}+\epsilon \mathbf{\Delta}\right)\mathbf{u}-\mathbf{u}^3,$$6

and prove long-time modulational stability of quasi-traveling waves. Exact translation invariance is obstructed by aliasing in the discrete Hamiltonian, but solutions initialized near sampled continuum solitons shadow the continuum traveling wave over times up to order $$\frac{d^2 \mathbf{u}}{dt^2}=\left(-\omega^2 \mathbf{I}+\epsilon \mathbf{\Delta}\right)\mathbf{u}-\mathbf{u}^3,$$7 (Bernier et al., 2018). This is a precise lattice manifestation of metastable motion and Peierls–Nabarro-type pinning effects.

Spinorial lattice systems exhibit the same anti-continuum logic but with qualitatively different branch organization. In the discrete nonlinear Dirac equation obtained from a lattice Gross–Neveu model, one-, two-, and three-site anti-continuum excitations persist, but the branch correspondence with DNLS is shifted: the two-site and three-site branches play roles analogous to the one-site and two-site DNLS branches, while the one-site branch develops a staggered solitary pattern rather than converging to the standard continuum Gross–Neveu pulse (Cuevas-Maraver et al., 2014). The resulting spectral picture includes stability exchanges between two-site and three-site branches, oscillatory-instability bubbles caused by collisions of internal modes with the essential spectrum, and dynamical splitting of unstable three-site states into daughter structures. Taken together, these results show that anti-continuum continuation is not tied to scalar lattices; it extends to graph, spinorial, and mixed-field settings, but the discrete geometry and coupling architecture determine which localized families are dynamically privileged.

4. Well-posedness, conservation laws, and singularity formation

The analytic backbone of DNLW consists of well-posedness and a priori estimates. For the discrete Klein–Gordon/wave equation on $$\frac{d^2 \mathbf{u}}{dt^2}=\left(-\omega^2 \mathbf{I}+\epsilon \mathbf{\Delta}\right)\mathbf{u}-\mathbf{u}^3,$$8, local well-posedness holds in $$\frac{d^2 \mathbf{u}}{dt^2}=\left(-\omega^2 \mathbf{I}+\epsilon \mathbf{\Delta}\right)\mathbf{u}-\mathbf{u}^3,$$9 for every $\mathbf{\Delta}=\mathbf{A}-\mathbf{D}$0, using a first-order reduction through $\mathbf{\Delta}=\mathbf{A}-\mathbf{D}$1 and direct Banach contraction in $\mathbf{\Delta}=\mathbf{A}-\mathbf{D}$2. In the defocusing case, global well-posedness is proved for $\mathbf{\Delta}=\mathbf{A}-\mathbf{D}$3 under $\mathbf{\Delta}=\mathbf{A}-\mathbf{D}$4, while in the focusing case negative-energy data in $\mathbf{\Delta}=\mathbf{A}-\mathbf{D}$5 blow up in finite time by a discrete Levine concavity argument (Wu et al., 2023). One clear implication is that discretization improves local $\mathbf{\Delta}=\mathbf{A}-\mathbf{D}$6-mapping properties but does not eliminate the hyperbolic blow-up mechanism.

At the fully discrete level, the Strauss–Vazquez scheme for $\mathbf{\Delta}=\mathbf{A}-\mathbf{D}$7-invariant nonlinear wave equations provides a rare example where solvability, exact discrete invariants, and stability are all explicit. The discrete Cauchy problem is globally well posed under lower-bound and monotonicity assumptions on the nonlinear potential, and the exact discrete energy

$\mathbf{\Delta}=\mathbf{A}-\mathbf{D}$8

is positive definite when

$\mathbf{\Delta}=\mathbf{A}-\mathbf{D}$9

At the distinguished ratio

$h\mathbb Z^d$0

the scheme also preserves a discrete analog of charge (Comech et al., 2010). These results are not merely numerical conveniences: they show that careful discretization can preserve the geometric content of the continuum model at the lattice level.

Several adjacent literatures enlarge this picture. Semi-discrete approximations of nonlocal wave equations with convolution coupling are locally well posed in $h\mathbb Z^d$1, converge uniformly to the continuum solution with second-order spatial accuracy, and numerically capture finite-time blow-up for several kernels (Erbay et al., 2018). For nonlinear damped wave equations associated with a positive operator $h\mathbb Z^d$2 with discrete spectrum, explicit modal decay yields global small-data well-posedness for semilinear and more general nonlinear problems once $h\mathbb Z^d$3, where $h\mathbb Z^d$4 is the bottom of the spectrum (Ruzhansky et al., 2017). In stochastic lattice settings, global well-posedness can persist even with locally Lipschitz nonlinear noise, provided the phase space is raised to a mixed-order Bochner product such as $h\mathbb Z^d$5 (Pan et al., 3 Mar 2026). The unifying point is that DNLW analysis is inseparable from the function space adapted to the discrete operator, whether that space is $h\mathbb Z^d$6, graph modal coordinates, operator Sobolev scales, or higher-order mean-square Bochner spaces.

5. Transport, scattering, and dispersive hydrodynamics

Beyond existence theory, DNLW is a theory of transport. In DNLS scattering by asymmetric nonlinear defects, exact stationary scattering states can be computed by a backward transfer map, and already a two-site dimer suffices to generate strong left-right transmission asymmetry at the same frequency and incident amplitude. The mechanism is explicitly nonlinear: asymmetry alone does not produce nonreciprocity in the linear limit, whereas nonlinearity plus broken mirror symmetry detunes defect resonances differently for opposite directions of incidence. Near nonlinear resonances, multistability and oscillatory instability appear, and unstable stationary states can saturate into quasiperiodic states combining an extended scattering background with a localized defect mode at a different frequency (Lepri et al., 2012).

At larger scales, dispersive hydrodynamics on lattices departs sharply from continuum NLS intuition. For dam-break problems in the defocusing DNLS,

$h\mathbb Z^d$7

the continuum-like regime is separated from genuinely discrete dynamics by the threshold

$h\mathbb Z^d$8

Above this threshold, rarefaction–DSW patterns resemble continuum NLS behavior; below it, the lattice supports traveling DSWs, discrete NLS kinks, dark solitary waves, composite front–DSW structures, shock breakdown, and multiphase wavetrains triggered by a generalized two-phase modulational instability (Mohapatra et al., 15 Jul 2025). The paper’s key message is that bounded Brillouin-zone dispersion and sign-changing dispersion curvature can reverse DSW polarity, invalidate single-phase Whitham closure, and dynamically select coherent structures with no continuum analogue.

For the semilinear discrete wave equation on $h\mathbb Z^d$9,

$$\partial_t^2 u_n-\Delta_h u_n+V_n u_n+\lambda |u_n|^{2\sigma}u_n=0,$$0

recent work establishes the nonlinear scattering theory itself. Combining sharp oscillatory-integral decay for the linear lattice wave equation with Strauss-type abstract scattering and a Strichartz-based approach, one obtains wave operators and asymptotic completeness for small solutions in dimensions $$\partial_t^2 u_n-\Delta_h u_n+V_n u_n+\lambda |u_n|^{2\sigma}u_n=0,$$1 (Wang, 25 Jul 2025). The decisive point is that lattice dispersion is weaker and more anisotropic than continuum dispersion because the symbol

$$\partial_t^2 u_n-\Delta_h u_n+V_n u_n+\lambda |u_n|^{2\sigma}u_n=0,$$2

has degenerate critical geometry on the torus. In one dimension the velocity-driven propagator does not exhibit general dispersive decay, and the standard scattering mechanism fails (Wang, 25 Jul 2025). A common misconception is therefore that continuum nonlinear-wave scattering transfers automatically to lattices; the lattice theory is instead dimension-dependent and strongly constrained by the spectral geometry of the Brillouin zone.

6. Geometric discretization, stochastic extensions, and control-theoretic generalizations

A major contemporary direction treats DNLW not only as models to analyze but as structures to preserve under approximation or manipulation. For the quasilinear nonlinear wave equation

$$\partial_t^2 u_n-\Delta_h u_n+V_n u_n+\lambda |u_n|^{2\sigma}u_n=0,$$3

symmetry- and conservation-law-preserving finite-difference discretizations have been constructed on five-point and nine-point stencils. The principal nonlinear nine-point scheme preserves four point symmetries and three local conservation laws in exact discrete divergence form, while alternative explicit schemes preserve different subsets of the continuum invariants (Cheviakov et al., 2020). This makes explicit a structural tradeoff that recurs throughout DNLW numerics: one typically cannot preserve all desirable invariants on a fixed stencil.

The same principle appears in stochastic wave discretization. For one-dimensional nonlinear stochastic wave equations with multiplicative noise, compact finite differences and IPDG space discretizations can be combined with a discrete-gradient and Padé-based time integrator so that the fully discrete scheme preserves the discrete averaged energy evolution law. In the additive-noise case, the averaged energy law is inherited exactly almost surely (Hong et al., 2021). In deterministic language, these are energy-preserving DNLW schemes; in stochastic language, they preserve the correct energy injection law rather than a pathwise constant Hamiltonian.

Control theory supplies a different extension. A controlled discrete semilinear wave lattice

$$\partial_t^2 u_n-\Delta_h u_n+V_n u_n+\lambda |u_n|^{2\sigma}u_n=0,$$4

can be reinterpreted as an interacting particle system with positions $$\partial_t^2 u_n-\Delta_h u_n+V_n u_n+\lambda |u_n|^{2\sigma}u_n=0,$$5 and velocities $$\partial_t^2 u_n-\Delta_h u_n+V_n u_n+\lambda |u_n|^{2\sigma}u_n=0,$$6. Under the amplitude condition

$$\partial_t^2 u_n-\Delta_h u_n+V_n u_n+\lambda |u_n|^{2\sigma}u_n=0,$$7

explicit bounded feedback controls steer arbitrary periodic initial data in finite time to moving or stationary flocks satisfying $$\partial_t^2 u_n-\Delta_h u_n+V_n u_n+\lambda |u_n|^{2\sigma}u_n=0,$$8, and the corresponding minimal-time problem is linked to a Hamilton–Jacobi equation on the finite-dimensional phase space (Strikwerda et al., 12 Feb 2025). This suggests that DNLW can be treated not only as conservative or dispersive media but also as controllable collective systems.

Other extensions push the notion of discreteness further. Fractional integrable and averaged discrete NLS equations replace the usual local dispersive operator by spectral fractional powers, either through the discrete fractional Laplacian or through the Ablowitz–Ladik recursion operator, yielding nonlocal semi-discrete wave equations with IST structure and exact solitons (Ablowitz et al., 2022). Stochastic discrete long-wave–short-wave resonance systems admit global well-posedness, weak pullback mean random attractors, and invariant measures under nonlinear locally Lipschitz noise, provided the analysis is carried out in $$\partial_t^2 u_n-\Delta_h u_n+V_n u_n+\lambda |u_n|^{2\sigma}u_n=0,$$9 and supported by uniform tail-end estimates (Pan et al., 3 Mar 2026). These developments suggest that DNLW has become an umbrella for a broad class of lattice models in which geometry, randomness, control, and nonlocality are as important as the original nearest-neighbor wave mechanism.

In aggregate, DNLW research now spans graph breathers, DNLS transport, discrete scattering, nonlocal long-range coupling, fully discrete geometric schemes, stochastic lattices, and controlled semilinear waves. The field’s organizing principles remain recognizable—anti-continuum continuation, spectral separation, invariant structure, and lattice-induced resonance—but the relevant discrete object may be a graph Laplacian, a punctured network spectrum, a band-limited aliasing defect, a stochastic Bochner phase space, or a control-amplitude constraint rather than a simple one-dimensional second difference.