Multiscale Velocity Gradients & Energy Transfer

Multiscale Velocity Gradients & Energy Transfer
Implications for Large-Eddy Simulation · Arun, Kamal, Colonius & Johnson · J. Fluid Mech.
Turbulence TheoryLES ModelingIsotropic Turbulence
The Bottleneck Effect: A Spectral Mystery
What Is It?
In 3D turbulence, the energy cascade exhibits a spectral bump exceeding Kolmogorov's  scaling near the viscous cutoff — the bottleneck effect.
Two Origins
True bottleneck (DNS): Viscous backscatter in the subinertial range
Artificial bottleneck (LES): Residual stress model error in the inertial range
Both share a common structural origin: small-scale shear layers.
Two Papers, One Unified Story
Paper I — SFR & LES Models
Stokes Flow Regularization generates dynamic LES closures. A mixed model (eddy viscosity + nonlinear gradient) mitigates the artificial bottleneck by capturing local residual stress structure.
Paper II — Normality-Based Analysis
A normality-based decomposition of filtered velocity gradients reveals that shear layers drive strain self-amplification, vortex stretching, and the backscatter causing the bottleneck.
Normality-Based Decomposition of the VGT
Rather than the conventional strain + vorticity split, the VGT is decomposed in a principal reference frame into three physically distinct constituents:
Normal Straining
Symmetric & normal — axis-aligned stretching and compression
Pure Shearing
Non-normal — captures shear layers, the key small-scale structure
Rigid Rotation
Antisymmetric & normal — captures vortex tubes
Multiscale Energy Transfer Decomposition
Using Gaussian filtering, interscale energy transfer  is expanded in multiscale velocity gradients and decomposed into three mechanisms:
Strain Self-Amplification
~60% of forward cascade; normal straining dominates
Vortex Stretching
~40% of forward cascade; shear vorticity dominates
Strain–Vorticity Covariance
Negligible net in inertial range, but drives backscatter in subinertial range
Shear Layers: The Dominant Players
Key DNS Findings
Pure shearing accounts for >50% of total energy transfer in the inertial range
Shear layer stretching (normal straining of shear vorticity) is the strongest contributor to scale-local vortex stretching
Backscatter driving the bottleneck is almost entirely from the covariance of normal straining with shear vorticity
Peak backscatter coincides with empirical shear layer thickness: 
Vortex Tubes vs. Shear Layers
The normality-based decomposition unambiguously distinguishes tube-like structures (rigid rotation) from sheet-like structures (shear vorticity) — a distinction obscured by symmetry-based analysis.
Vortex Tubes
Associated with rigid rotation. Filtering preserves tube-like character. Dominant in unfiltered DNS at low .
Shear Layers
Associated with shear vorticity. Widths of . Softened by filtering; dominant in small-scale energy transfer.
LES Models: Mixed vs. Eddy Viscosity
What the Data Shows
The mixed model reproduces filtered DNS partitioning across all resolved scales. The eddy viscosity model deviates near the filter width — its statistics collapse onto a DNS at  (effective ).
The mixed model's nonlinear gradient term explicitly captures scale-local strain self-amplification and vortex stretching — the key to accurate cascade physics.
The Artificial Bottleneck: Same Origin, Wrong Scale
Why Eddy Viscosity Fails
The eddy viscosity model produces excessive backscatter from the strain–vorticity covariance term — specifically from shear layer contributions — in the inertial range.
This artificial bottleneck has the same shear-layer origin as the true DNS bottleneck, but appears at the wrong scales because the model retains unfiltered-like shear layer structure.
The Fix
Adding a nonlinear gradient component (mixed model) explicitly resolves scale-local terms, eliminating the spurious backscatter and restoring correct cascade efficiency.
Conclusions & Outlook
1
Shear layers dominate the cascade
Straining of multiscale shear layers accounts for >50% of energy transfer; shear layer stretching drives scale-local vortex stretching.
2
Backscatter is a shear-layer phenomenon
The bottleneck effect — true and artificial — is almost entirely due to normal straining interacting with shear vorticity at small scales.
3
Mixed models outperform eddy viscosity
Explicitly capturing scale-local terms via a nonlinear gradient component reproduces filtered DNS statistics and eliminates the artificial bottleneck.
4
A framework for model assessment
Normality-based analysis provides a principled, quantitative tool for evaluating and designing LES closures across flow regimes.

Made with