Spectral Forcing Operators in Turbulence

• Spectral forcing operators are mathematical constructs that act on the modal (Fourier) representation to quantify and control nonlinear energy transfers.
• They enable precise mapping of triadic interactions by decomposing nonlinear forcing into energy transfer coefficients, critical in turbulence analysis.
• Data-driven and control applications leverage these operators for filtering, denoising, and reduced-order modeling in simulations and machine learning tasks.

A spectral forcing operator is a mathematical or computational construct that acts on the spectral (Fourier or modal) representation of a physical system to inject, filter, or analyze nonlinear or exogenous forcing within a particular wavenumber, frequency, or mode range. Spectral forcing operators arise in multiple research domains, including the analysis of turbulence (Navier–Stokes, DNS), nonlinear dynamical systems, stochastic diffusion models, and the study of anisotropy in rotating turbulence. They formalize, quantify, or control how energy, structure, or information is transferred across spectral scales, and are tightly connected to both the direct manipulation of input (forcing) and the analytical mapping from forcing to system response.

1. Mathematical Definitions in Turbulent Flow and Nonlinear Systems

A canonical setting for spectral forcing operators is the incompressible Navier–Stokes equations, where the nonlinear convection term acts as a quadratic forcing to the linearized momentum equations. Upon spatio-temporal Fourier transform (e.g., in channel flow, homogeneous in $x$, $z$ and stationary in $t$), the nonlinear term becomes a spectral convolution:

$$\mathbf f(\mathbf k_3,y) = -\!\sum_{\mathbf k_1+\mathbf k_2=\mathbf k_3}\, \mathbf u(\mathbf k_1,y)\!\cdot∇\,\mathbf u(\mathbf k_2,y)\,,$$

where each $\mathbf f(\mathbf k_3,y)$ is partitioned into triadically compatible components—a spectral forcing operator in the sense that it is a quadratic map defined by convolution over modal indices (Huang et al., 10 Mar 2025).

The linear resolvent operator $R(\mathbf k,y)$, defined by inverting the frequency-domain linear operator $-\mathrm{i}\omega I-L(\mathbf k)$, yields the modal response:

$$\mathbf u(\mathbf k,y) = R(\mathbf k,y) \mathbf f(\mathbf k,y)$$

which enables the systematic partitioning of transfer mechanisms between scales. The spectral resolvent operator, together with the spectral form of the nonlinear forcing, constitutes what is termed the "spectral forcing operator."

In the context of periodic dynamical systems, the "time-spectral resolvent operator" acts on time-collocated modal envelopes rather than on Fourier coefficients, mapping a discrete time grid of forcing to the response via a block-diagonal operator involving the time-periodic Jacobian and Fourier spectral-differentiation matrix (Howell et al., 16 Feb 2026).

2. Partitioning Nonlinear Energy Transfer and Triadic Analysis

A core utility of the spectral forcing operator is the quantification of triadic interactions and modal energy input. For a fixed output wave vector $\mathbf k_3$, one decomposes the nonlinear forcing energy via "forcing-coefficients" $P(\mathbf k_1,\mathbf k_2)$ and corresponding response-coefficients $z$0:

$z$1

$z$2

where $z$3 is the discrete (computed) resolvent. These coefficients provide a weighted spectral energy transfer map, precisely partitioned by triad and enabling identification of the dominant energy-transfer channels between large-scale and small-scale motions (Huang et al., 10 Mar 2025). The band structure of $z$4 and $z$5 encodes the key physical pathways: large-scale to small-scale transfer, amplitude modulation via matching phase speed, and the efficiency of quasi-linear or generalized quasi-linear (QL/GQL) representations.

3. Data-Driven and Forcing-Informed Spectral Operators

Recent advances generalize spectral forcing operators to fully data-driven constructions. The "forcing-informed (FI) resolvent" projects the canonical linear resolvent onto subspaces identified from simulation data—forcing and response modes are obtained from the SVD of

$z$6

where $z$7 and $z$8 are basis matrices spanning the input (forcing) and output (response) subspaces, respectively. This framework ensures that only dynamically realized forcing structures and responses are considered, preventing the inclusion of unphysical, energetically inactive modes (Iwatani et al., 18 Jun 2026). The resulting spectral forcing operator thus maps the energy-closed nonlinear energy transfer in a fully empirical fashion, supporting nonlinear energy transfer maps that localize injection/removal regions in physical space as a function of frequency.

4. Spectral Forcing Operators for Control and Filtering

In controlled stochastic processes and machine learning models (e.g., pixel-space diffusion), spectral forcing operators are constructed to filter the input by frequency, often for denoising or capacity allocation. A prominent example is the parameter-free, time-conditional 2D-DCT low-pass operator developed for diffusion models (Fan et al., 13 Jun 2026):

$z$9

where $t$0 is a soft radial mask with time-dependent cutoff. The schedule $t$1 is derived from the per-band data-to-noise ratio contour tied to the evolving signal/noise separation at each diffusion step. The operator explicitly enforces a spectral boundary in the model input, reallocating model capacity to the signal-bearing region and yielding improved generative metrics.

5. Implementation in Turbulence Simulations and Modal Anisotropy

Spectral forcing operators are the basis for several widely used forcing schemes in turbulence simulations (Vallefuoco et al., 2017):

• Steady single-wavenumber forcing (ABC): Injects energy and helicity explicitly at a fixed set of spectral modes, imparting significant small-scale anisotropy across the cascade.
• Unsteady Euler forcing: Cancels viscosity within a sphere $t$2, so that large scales evolve as a truncated Euler system and transfer is triad-preserving; can produce variable anisotropy dependent on the helicity content and distribution.
• Negative viscosity forcing: Applies a negative effective viscosity in the forced range to inject energy isotropically, but still produces nontrivial anisotropy at high Reynolds number via non-local triad interactions.

The modal decomposition of the velocity correlation tensor in the Craya–Herring basis quantifies energetic, helical, and polarization anisotropy induced by these spectral forcing operators. The induced anisotropy is persistent, scale- and direction-dependent, and interacts with rotation and dissipation scales, as typified by the empirically determined threshold wavenumber $t$3 that separates rotation- and dissipation-dominated subranges.

6. Applications and Consequences in Reduced-Order Modeling

Spectral forcing operators underpin the development of reduced-order models, especially in resolvent-based and quasi-linear (QL/GQL) frameworks (Huang et al., 10 Mar 2025). By identifying and isolating the dominant triadic or modal energy-transfer pathways, these operators enable:

• Construction of data-driven, colored forcing models for closure of Reynolds-stress gradients
• Selection of dynamically critical triads and modes for inclusion in low-order resolvent network models
• Quantitative assessment of how targeted flow control penetrates (or fails to penetrate) the hierarchy of energy transfer mechanisms

A direct implication is the possibility of thresholding the $t$4 and $t$5 coefficients to construct sparsified nonlinear models that still capture the vast majority ($t$6) of energy-transfer dynamics. This also enables the identification and targeting of resonant or amplitude-modulating triads essential in wall turbulence and related flows.

7. Perspectives on Spectral Forcing Operators Across Disciplines

Spectral forcing operators unify a broad class of methods spanning direct turbulence simulation, nonlinear system analysis, data-driven model reduction, and stochastic generative modeling. Their defining attributes are:

• Acting in the spectral (modal, frequency, or time–spectral) domain
• Enabling precise partitioning or manipulation of forcing structures
• Supporting the quantitative mapping of nonlinear interactions and energy transfer
• Facilitating reduced-order and data-driven modeling that respects physical and dynamical constraints

Current developments demonstrate their utility in both theoretical analysis (e.g., identification of energy-injection loci in self-sustained flows (Iwatani et al., 18 Jun 2026)) and practical algorithmic design for high-dimensional machine learning models (Fan et al., 13 Jun 2026). A plausible implication is that future advances in both fields will increasingly rely on spectral forcing operators, either in model-based or data-driven form, to achieve interpretable, efficient, and physically faithful representation and control of complex nonlinear dynamics.