Space–Time Resolvent Analysis

• Space–time resolvent analysis is a modeling framework that linearizes the Navier–Stokes equations to reveal coherent spatio-temporal structures in complex, turbulent flows.
• It employs frequency-domain methods and singular value decomposition to connect nonlinear forcing with optimal velocity and pressure response modes.
• The approach enables reduced-order modeling and flow control by linking resolvent gains to turbulence statistics and self-similar attached-eddy scaling.

Space–time resolvent analysis is a modeling framework designed to extract, predict, and interpret coherent spatio-temporal structures in complex, often turbulent, flows by combining linearization of the governing equations, frequency-domain techniques, and advanced statistical inference. The method employs the resolvent operator to connect nonlinear forcing—typically the Reynolds stress residuals in the linearized Navier–Stokes equations—to the resulting velocity and pressure fluctuations, providing a systematic avenue for modeling flow mechanisms, estimating flow statistics, and designing flow control strategies (Karban et al., 2021).

1. Theoretical Foundation and Operator Formulation

Space–time resolvent analysis begins by linearizing the Navier–Stokes equations about a steady (or periodically-varying) mean or base flow. The velocity field is decomposed as $q_\mathrm{tot} = Q(y) + q'(x,y,z,t)$ in the canonical case of channel flow, with $Q$ representing the mean and $q'$ the fluctuations. Linearization leads to an input-output system in the frequency–wavenumber domain: $$q'(x,y,z,t) = \sum_{k_x,k_z} \int_{-\infty}^{\infty} \hat{u}(k_x, y, k_z, \omega) e^{i(k_x x + k_z z - \omega t)} d\omega$$ The block-operator formulation

$i\omega M \hat{u} - A(k_x, k_z) \hat{u} = B\hat{f}$

with $\hat{u} = [\hat{u}, \hat{v}, \hat{w}, \hat{p}]^T$ introduces the mass constraint matrix $M$, the linearized operator $A$, and forcing $\hat{f}$ (often $-(u\cdot\nabla u)^{\wedge}$, the nonlinear triad). The resulting resolvent operator $$R(k_x,k_z, \omega) = C[-i\omega M - A(k_x, k_z)]^{-1}B$$ maps the (generalized) Reynolds stress forcing to the velocity/pressure response (Karban et al., 2021).

This operator admits a singular value decomposition: $R = U\Sigma V^*$ where $\Sigma$ collects the amplification (gain) singular values, $U$ provides the dominant response modes (output), and $V$ yields the corresponding optimal forcings (input).

2. Statistical Modeling and Space–Time Flow Statistics

Given the stochastic nature of turbulent forcing, the space–time resolvent framework couples the input-output dynamics with the cross-spectral density (CSD) of the forcing: $P(k_x,k_z,\omega) = E\{\hat{f} \hat{f}^H\}$ yielding for the velocity CSD: $S(k_x, k_z, \omega) = R P R^H$ This formulation generalizes to colored forcing and is at the core of modern space–time resolvent-based statistical estimation (Towne et al., 2019).

The approach enables estimation of space–time statistics (e.g., covariance tensors, two-point–two-time correlations) from partial or downsampled measurements, with the resolvent operator serving to propagate knowledge of the nonlinear forcing statistics to unmeasured or unobservable variables. Unlike rank-1 (low-rank) models, this method leverages the full range of resolvent modes inferred by the data, making it suited for high-rank turbulent flows (Karban et al., 2021, Towne et al., 2019).

3. Extraction of Dynamically Consistent Forcing and Response Structures

To identify the actual nonlinear forcing correlated with observed coherent structures, frequency-domain Extended Proper Orthogonal Decomposition (EPOD) is employed. The procedure computes frequency-resolved SPOD modes $\psi_p$ by solving the generalized eigenproblem

$S W \psi_p = \lambda_p \psi_p$

with $W$ establishing the relevant energy inner product. The coefficients $a_n = \langle \hat{q}, \psi_n \rangle$ possess variance $\lambda_n$. The EPOD mode for the forcing,

$\chi_p = E\{\hat{f} a_p^*\} / \lambda_p$

is linked to its associated SPOD mode via the resolvent operator: $\psi_p = R \chi_p$ This ensures that forcing-response pairs are dynamically consistent with the linearized dynamics and provides a direct framework for verifying theoretical scaling, such as Townsend's attached-eddy hypothesis for wall-bounded turbulence (Karban et al., 2021).

4. Self-Similarity, Physical Interpretation, and Validation

Space–time resolvent analysis has been applied to turbulent channel flow to demonstrate self-similarity in wall-attached velocity structures and their associated nonlinear forcing, quantitatively validating classical attached-eddy scaling: $y_h \sim 1/\beta$ where $y_h$ is a wall-normal statistic quantifying half-maximum streamwise energy accumulation and $\beta$ is the spanwise wavenumber. Both the SPOD-extracted velocity modes and the EPOD-extracted forcing structures exhibit this scaling, indicating that the space–time resolvent framework not only recovers the dominant flow features but also properly accounts for the organization of nonlinear driving mechanisms (Karban et al., 2021).

5. Relationship to Proper Orthogonal Decomposition and Resolvent Estimation

A key insight is the correspondence between SPOD and resolvent modes under white-noise forcing: when the expansion coefficients are uncorrelated, resolvent and SPOD modes are identical (Towne et al., 2017). For colored or correlated forcing, the resolvent framework offers parameterization via the forcing CSD and, combined with SPOD/EPOD, allows reconstruction of second-order space–time flow statistics. The space–time resolvent framework thus enables convergent, data-driven reduced-order modeling, linking physical amplification mechanisms (via singular values) and statistical dynamics (Towne et al., 2017, Karban et al., 2021).

6. Methodological and Computational Aspects

The standard workflow comprises:

1. Linearize the equations about the mean/base flow.
2. Fourier-transform in all homogeneous and stationary dimensions.
3. Compute or model the forcing CSD from available data.
4. Form the resolvent operator and execute SVD to obtain leading modes and gains.
5. Use resolvent-based estimation, possibly augmented by EPOD, to reconstruct flow statistics and extract physically consistent forcing and response structures.

This methodology supports high-fidelity extraction of flow features from DNS data and enables efficient estimation with limited experimental measurements (Karban et al., 2021, Towne et al., 2019). The approach is agnostic to the origin of the data used for estimating the forcing spectrum.

7. Limitations, Assumptions, and Outlook

The accuracy of space–time resolvent analysis is contingent on the linearization around a representative base flow and the quality of stochastic modeling for the nonlinear forcing. Statistical stationarity is often assumed for rigorous Fourier and CSD analysis, but extensions exist to handle time-varying and periodically modulated base flows using, for example, harmonic resolvent and Floquet expansions (Karban et al., 2021). While the resolvent operator captures the most linearly amplified flow structures, it does not explicitly resolve nonlinear triadic interactions, except as represented through the inferred forcing statistics.

The direct verification of self-similar attached-eddy scaling demonstrates the method's ability to link first-principles modeling with empirical and theoretical findings in turbulence research. Applications span flow modeling, understanding of turbulence organization, and flow control strategies, indicating broad relevance for reduced-order modeling and design in high-dimensional complex flows (Karban et al., 2021).

• Karban, R., Encinar, M.P., Lozano-Durán, A., & Yang, X., "Self-similar mechanisms in wall turbulence studied using of resolvent analysis," (Karban et al., 2021), 2021-05-27.
• Towne, A., Lozano-Durán, A., & Yang, X., "Resolvent-based estimation of space-time flow statistics," (Towne et al., 2019), 2019-01-22.
• Schmidt, O. T., Towne, A., Rigas, G., Colonius, T., & Brès, G. A., "Spectral proper orthogonal decomposition and its relationship to dynamic mode decomposition and resolvent analysis," (Towne et al., 2017), 2017-08-15.