Resonant interactions of nonlinear water waves in a finite basin

We study exact four-wave resonances among gravity water waves in a square box with periodic boundary conditions. We show that these resonant quartets are linked with each other by shared Fourier modes in such a way that they form independent clusters. These clusters can be formed by two types of quartets: (1) {\it angle-resonances} which cannot directly cascade energy but which can redistribute it among the initially excited modes and (2) {\it scale-resonances} which are much more rare but which are the only ones that can transfer energy between different scales. We find such resonant quartets and their clusters numerically on the set of 1000 x 1000 modes, classify and quantify them and discuss consequences of the obtained cluster structure for the wavefield evolution. Finite box effects and associated resonant interaction among discrete wave modes appear to be important in most numerical and laboratory experiments on the deep water gravity waves, and our work is aimed at aiding the interpretation of the experimental and numerical data.

💡 Research Summary

This paper investigates exact four‑wave resonances of deep‑water gravity waves confined to a square periodic domain, focusing on the consequences of finite‑box discretisation for wave turbulence. The authors start by recalling that the classical wave turbulence theory (WTT) assumes an infinite domain and a weak‑nonlinearity limit, which together guarantee a dense set of quasi‑resonant interactions. In realistic laboratory tanks or numerical simulations, however, the spectral grid spacing Δk is often larger than the resonance broadening Ω, so only exact resonances (Ω = 0) survive. For gravity waves the dispersion relation is ω(k)=√|k|, and the exact four‑wave resonance conditions become
|k₁|^{1/2}+|k₂|^{1/2}=|k₃|^{1/2}+|k₄|^{1/2}, k₁+k₂=k₃+k₄,
with integer wavevectors k_i=(m_i,n_i). Solving these Diophantine equations directly is computationally prohibitive; the authors therefore employ the “q‑class” method. Each integer wavevector can be uniquely written as k=γ q^{1/4}, where γ∈ℤ is a weight and q is a product of distinct primes raised to powers less than four. All vectors sharing the same q belong to the same q‑class C_l_q.

Two families of solutions arise. Type I (scale‑resonances) involve four vectors from a single q‑class; the resonance condition reduces to γ₁+γ₂=γ₃+γ₄. These resonances change the magnitudes |k| and therefore can transfer energy across scales, providing a mechanism for a cascade. Type II (angle‑resonances) involve two q‑classes; each pair of vectors has equal length (|k₁|=|k₃|, |k₂|=|k₄| or the swapped version). Angle‑resonances conserve the set of |k| values and cannot generate new scales; they merely redistribute energy among already excited modes.

The authors performed an exhaustive search in a spectral domain |m|,|n|≤1000 (a 1000 × 1000 integer lattice). They found 230 464 exact four‑wave resonances. Of these, 21 376 are collinear (all four vectors lie on a straight line). Because the nonlinear interaction coefficient vanishes for collinear quartets, they are dynamically irrelevant for the standard gravity‑wave Hamiltonian. The remaining 16 704 (≈7.2 %) are non‑collinear and are dynamically active. Non‑collinear resonances split further into two geometrical families: “tridents” and “non‑tridents”. A trident consists of two anti‑parallel vectors (k₁ = (a,0), k₂ = (−b,0)) and a symmetric pair (k₃ = (c,d), k₄ = (c,−d)) with integer parameters (a,b,c,d) generated by simple formulas involving two integers s and t. Tridents can be axial (aligned with the coordinate axes) or non‑axial after rational rotations and scalings that preserve integer components. In the searched domain, 13 888 non‑axial tridents were identified, while only 960 axial ones exist (≈6.5 % of all tridents). The rest of the non‑collinear resonances (1 856) are non‑trident quartets that do not obey the trident symmetry.

A central result is the identification of resonance clusters. Quartets that share at least one wavevector are linked, forming connected components (clusters) in the graph of modes. The authors present detailed statistics: the largest cluster contains 43 136 quartets and has a length (number of distinct quartets) of 56; many smaller clusters consist of only a few quartets. Clusters may be homogeneous (only angle‑resonances) or mixed (containing both scale‑ and angle‑resonances, or tridents together with non‑tridents). Because clusters are isolated from each other, energy cannot flow between different clusters; the overall dynamics reduces to independent subsystems.

From a dynamical perspective, pure angle‑resonance clusters behave like “frozen turbulence”: they cannot support a Kolmogorov‑Zakharov cascade and may relax toward a Rayleigh‑Jeans equilibrium limited to the resonant modes. In contrast, clusters that contain at least one scale‑resonance can generate a cascade, albeit limited to the modes within that cluster. Hence, the sparsity of scale‑resonances (only about 0.1 % of all quartets) explains why finite‑box experiments often exhibit steep spectra and anisotropy: the cascade is confined to a few rare pathways.

The paper emphasizes that finite‑box effects are not a minor correction but fundamentally reshape the interaction network. Even with a modest 1000 × 1000 resolution, the resonance graph is highly non‑trivial, suggesting that higher‑resolution simulations (e.g., 10 000 × 10 000) would only increase the complexity. For experimental design, the results imply that the choice of tank size and initial wavevector configuration can be used to promote or suppress particular q‑classes, thereby controlling whether scale‑resonances appear.

In summary, the work provides a comprehensive classification of exact four‑wave resonances for gravity waves in a finite periodic basin, quantifies the prevalence of scale‑ versus angle‑resonances, elucidates the cluster topology of resonant interactions, and connects these mathematical structures to observable phenomena such as spectral steepening, anisotropy, and the possible absence of a classical turbulent cascade in laboratory and numerical settings.