Determining Modes in Dynamical Systems

• Determining modes property is a concept in dissipative dynamical systems, showing that finitely many low Fourier modes govern long-term behavior.
• It provides a framework for data assimilation and model reduction by ensuring that convergence in low-dimensional projections leads to full system convergence.
• Quantitative bounds relate physical parameters to the minimal number of modes needed, offering practical guidelines for observer design in nonlinear PDEs.

The determining modes property is a fundamental concept in the theory of dissipative infinite-dimensional dynamical systems—especially those arising from nonlinear PDEs such as the Navier–Stokes equations, Boussinesq and primitive equations, and nonlinear Schrödinger equations—which formalizes the observation that the long-time (global attractor) dynamics, though formally infinite-dimensional, are in fact governed by finitely many degrees of freedom. A set of determining modes is a finite collection of spatial Fourier coefficients or eigenmodes such that convergence of solutions in these low-dimensional projections suffices to guarantee convergence in the full, infinite-dimensional phase space. Determining modes provide rigorous measures of physical resolution, define minimal requirements for data assimilation or observation, and offer foundational support for model reduction in complex nonlinear systems.

1. Formal Definitions and Basic Properties

In the canonical setting of the 2D incompressible Navier–Stokes equations or related dissipative PDEs, the determining modes property asserts that there exists an integer $N$ (depending on physical parameters) such that the following holds. Let $u_1(t), u_2(t)$ be two solutions (possibly with distinct initial data or forcing). Let $P_N$ denote the spectral (orthogonal) projection onto the span of the first $N$ Fourier modes. Then,

$$\lim_{t\to\infty} \|P_N u_1(t) - P_N u_2(t)\| \to 0 \quad \text{and} \quad \lim_{t\to\infty} \|f_1(t)-f_2(t)\| \to 0 \implies \lim_{t\to\infty} \|u_1(t) - u_2(t)\| \to 0.$$

This definition formalizes the idea that the asymptotic dynamics are “enslaved” by finitely many low-wavenumber components, and that all high modes are functionally dependent on the low ones on the global attractor (Carlson et al., 2024).

The minimal such $N$ is called the number of determining modes; such a set may also be constructed using nodal values, local averages, or other finite-dimensional observables, so long as the relevant injectivity or approximation properties hold (Foias et al., 2012).

A direct consequence is the existence of a finite-dimensional global attractor, and, in many settings, equivalence with the finite fractal dimension of the attractor. Analogues apply for other equations such as the surface quasi-geostrophic (SQG) equation (Cheskidov et al., 2015), the 2D Boussinesq system with fractional Laplacian (Huang et al., 2016), the 3D primitive equations (Chueshov, 2012), and the 1D damped NLS (Jolly et al., 2014).

2. Quantitative Estimates: Bounds and Scaling Laws

The number of determining modes can be explicitly bounded in terms of physical (dimensionless) quantities such as the Grashof number $G$, viscosity, diffusivity, and—for rotating or stratified systems—rotation rate and forcing smoothness.

For the 2D Navier–Stokes equations, the classical Foias–Prodi theorem gives the explicit scaling

$N \geq C G^2,$

where $G=\sup_{t}\|f(t)\|_{L^2} / \nu^2$ is the Grashof number (ratio of forcing magnitude to viscous damping), and $C$ is a universal constant (Carlson et al., 2024). For higher regularity of $f$, sharper shape factors can lower the required $N$.

For the β-plane Navier–Stokes equations, rotation drastically reduces the number of active degrees of freedom: for rotation parameter $\beta$, determining modes satisfy

$$N_{\text{modes}} \leq c\,G_0^{1/2} + c' (M/\beta)^{1/2},$$

where $M$ quantifies the small-scale regularity of forcing. In strong rotation ($\beta\gg1$), only $O(G_0^{1/2})$ modes are needed, reflecting two-dimensionalization of the flow (Miyajima et al., 2018).

In SQG, the number of determining modes is controlled by a time-dependent determining wavenumber; in the subcritical and critical regimes,

$$N \lesssim \max\left\{ (\|f\|_{L^2}/\nu^2)^{4/(\alpha-1)},\, (\|f\|_{L^2}/\nu^2)^{2/(\alpha-1)} \right\}$$

for dissipation exponent $\alpha > 1$ (Cheskidov et al., 2015).

For the 3D viscous primitive equations, spectral gap bounds and energy estimates provide explicit $N$ that depend on viscosity, external force regularity, and domain geometry (Chueshov, 2012). Similar reasoning extends to the 2D Boussinesq system with fractional dissipation, where $N$ scales polynomially with inverse dissipation parameters and forcing magnitude (Huang et al., 2016).

3. Dynamical Mechanism: Asymptotic Enslavement and Finite-Dimensional Reduction

The mathematical architecture of the determining modes property rests on the following dynamical mechanism:

• The solution trajectory $u(t)$ may be decomposed into low and high wavenumber parts, $u(t) = P_N u(t) + Q_N u(t)$. The low modes $P_N u(t)$ capture the large scale, while $Q_N u(t)$ represents the remaining, more rapidly decaying, degrees.
• The evolution of $Q_N u$ is governed by a linear dissipative term plus nonlinearity slaved to the behavior of $P_N u$.
• Under suitable bounds on $N$, the nonlinear feedback is dominated by dissipation, so the high modes decay—unless the low modes are the same, in which case the full solutions coalesce.
• This can be formalized using the Ladyzhenskaya “squeezing property” and variants of Grönwall’s lemma, yielding exponential synchronization of solutions once low-mode agreement is achieved (Chueshov, 2012, Carlson et al., 2024).

This mechanism underpins the structure of all proofs of the finite determining-modes property in dissipative PDEs and provides a conceptual route for deriving inertial manifolds, determining forms, and effective model reduction (Foias et al., 2012, Jolly et al., 2014).

4. Generalizations: Nodes, Other Observables, and Random Systems

The determining property is not restricted to spectral (Fourier) modes. Alternative notions include:

• Determining nodes: spatial points where measurement of the solution suffices to infer the entire state asymptotically, subject to suitable interpolation and approximation inequalities (Miyajima et al., 2018).
• Finite volumes, finite elements, local spatial averages: properly chosen sets of these observables can also function as determining parameters if they meet direct-sum or interpolation criteria (Foias et al., 2012).
• Weakening to random or stochastic settings: If the forcing is random or measurements are corrupted, ergodicity and determining modes are still meaningful in the mean or probability sense (Chueshov, 2012).

The dichotomy between modes and nodes is most important in practical applications—where only physical (pointwise or localized) data may be accessible. The scaling of the minimal number of determining nodes slightly differs from that of modes, often with a higher exponent (Miyajima et al., 2018).

5. Interplay with Data Assimilation and State Reconstruction

A major development is the recognition that the determining modes property precisely characterizes the minimal effective dimension for data assimilation, state estimation, and observer design in dissipative PDEs:

• Continuous Data Assimilation (CDA), especially the Olson–Titi synchronization filter and Azouani–Olson–Titi (AOT) nudging schemes, reconstructs the unobserved state from partial observations (e.g., projections onto the first $N$ modes) (Carlson et al., 2024).
• The “self-synchronous intertwinement” formalism unifies the synchronization properties of CDA algorithms with the determining modes concept. Convergence of these filters is mathematically equivalent to existence of determining modes, and vice versa (Carlson et al., 2024, Carlson et al., 7 Dec 2025).
• Quantitative criteria for robust state reconstruction can be given in terms of $N$ and feedback parameters, guiding algorithmic design for fluids and geophysical applications (Carlson et al., 7 Dec 2025).

This equivalence elucidates why practical CDA filters “work” only when the number of observed (or controlled) modes exceeds the determining threshold, and provides full mathematical justification for such empirical practices.

6. Broader Implications: Global Attractor and Long-Time Behavior

The existence of finitely many determining modes is a gateway to the finite-dimensionality of the global attractor. Not only does it guarantee that the asymptotic state space is finite-dimensional—but also that its dimension is controlled by the number of determining modes. This, in turn, implies:

• Predictability: Once the low modes are determined, the entire flow is uniquely determined in the asymptotic regime.
• Reduction of complexity: Modelling and computation may be restricted to the inertial (determining) set without loss of long-time accuracy.
• Structural insight: Turbulent and chaotic flows in such systems are, despite infinite degrees of freedom, effectively finite in their asymptotic behavior (Chueshov, 2012, Huang et al., 2016).

In quantum optics, a conceptually analogous “determining modes” principle governs the description and manipulation of multimode quantum states, where the selection of an appropriate modal decomposition is critical for the analysis of entanglement, squeezing, and measurement-based computation (Fabre et al., 2019).

7. Algorithmic and Practical Aspects

The practical implementation of the determining modes concept involves:

• Identifying a minimal subset of spectrally localized observables. This requires estimation of system parameters and, typically, energy bounds or prior attractor estimates.
• For observer design and assimilation, coupling feedbacks or filters to measurements in only the first $N$ modes, with proof of convergence to the true trajectory when $N$ satisfies the determining modes bound (Carlson et al., 2024, Carlson et al., 7 Dec 2025).
• For model reduction, projection of the full PDE dynamics onto a reduced set of modes (Galerkin systems), with guarantees that their dynamics capture the global attractor fully, provided $N$ is above threshold (Foias et al., 2012, Jolly et al., 2014).
• Generalization to more complex decomposition–for instance, optimal mode selection for maximizing entanglement or minimizing cross-talk in quantum or wave systems (Fabre et al., 2019).

In summary, the determining modes property is central to understanding the information-theoretic and analytical backbone of reduced-order modeling, state observer design, and long-term physical predictability in high-dimensional dynamical systems. It provides both conceptual and quantitative structure for efficiently bridging infinite-dimensional theory and finite-dimensional practical implementation.