Nonlinear Scale-Local Geometric Deformations of Vortex Rings in Smooth Euler Flows via Bayesian Optimization and Adjoint Methods

1. Problem Statement

The paper addresses the fundamental mechanisms governing nonlinear scale-local deformations of vortex structures in incompressible 3D Euler flows. While turbulence is known to be organized by vortex-stretching dynamics (specifically the hierarchical structure of antiparallel vortex tubes), the precise nonlinear evolution of a vortex ring undergoing radially expanding Lagrangian transport with Kelvin waves remains unclear.

Specifically, the author investigates:

• How the axis of swirling particles (the local rotation axis of fluid elements) evolves relative to the vortex axis (the geometric centerline of the vortex ring).
• Whether the system naturally favors configurations where these two axes remain aligned to minimize shear.
• The existence of intrinsic wave equations governing these deformations within the full, smooth Euler equations, without relying on singular filament approximations.

2. Methodology

The paper employs a hybrid approach combining rigorous geometric analysis, machine learning-based optimization, and adjoint methods.

A. Geometric Lagrangian Framework

The author develops a novel framework that avoids singular integral representations of pressure.

• Lagrangian Flow Map: The flow is described by $\Phi(t, x)$. A smooth curve $\ell(s)$ represents the vortex axis.
• Moving Frame: A Lagrangian moving frame is constructed using the Frenet-Serret formulas ($\tau, n, b$) along the transported curve.
• Swirling Particles: The position of particles near the axis is expanded in terms of the frame vectors, introducing coefficients $\alpha_1(t)$ and $\alpha_2(t)$ which describe the deviation of the swirling particle axis from the vortex axis.
• Derivation of Wave Equations: By differentiating the Euler equations along the transported curve and utilizing the Frenet-Serret relations, the author derives a novel system of second-order wave equations governing $\alpha_1$ and $\alpha_2$.

B. Optimization Objective

To quantify the alignment between the vortex axis ($e_s$) and the swirling particle axis ($\zeta$), the author defines the Mean Square Directional Correlation (MSDC):
$$J(\Phi) = \frac{1}{t_1 - t_0} \int_{t_0}^{t_1} \int_0^1 \left| \frac{e_s(t, s) \cdot \zeta(t, s)}{|\zeta(t, s)|} \right|^2 ds \, dt$$
The goal is to maximize $J(\Phi)$, thereby minimizing shear and maintaining axis alignment.

C. Hybrid Optimization Strategy

The optimization landscape is identified as highly non-convex with many local minima. A single method is insufficient, leading to a two-stage hybrid approach:

1. Bayesian Optimization (Global Exploration):
   • Used to explore the global parameter space of the ring's deformation (represented by Fourier modes and polynomial time coefficients).
   • Utilizes the Optuna library with quasi-Monte Carlo sampling followed by Bayesian optimization to find a promising initial configuration.
2. Adjoint Method (Local Refinement):
   • Once a global optimum is found, the adjoint method is applied to perform local gradient-based refinement.
   • Variational Formulas: The author derives explicit variational formulas for curvature ($\kappa$), torsion ($T$), and the Frenet frame vectors with respect to perturbations in the flow map $\delta \Phi$.
   • Adjoint Equations: Backward-in-time adjoint equations are solved to compute the gradient of the objective functional $J$ with respect to the control parameters.
   • Constraints: The gradient is projected onto a space satisfying temporal admissibility (zero boundary conditions for displacement and velocity at $t_0, t_1$) and spatial band-limitation.

3. Key Contributions

1. Novel Wave Equations in Smooth Euler Flows:
   The paper provides the first derivation showing that wave equations governing Lagrangian motion emerge directly from the incompressible Euler equations in a fully smooth setting. This challenges the traditional view that such wave phenomena are exclusive to singular filament models (like the Local Induction Approximation or Biot-Savart expansions).
   The derived equations are:
   $$\alpha_1'' = \frac{v''}{v}\alpha_1 + 2v\kappa' + 4v'\kappa$$
   $$\alpha_2'' = \frac{v''}{v}\alpha_2$$
   where $v$ is the speed along the axis and $\kappa$ is curvature.

2. Identification of Intrinsic Nonlinear Mechanisms:
   The study identifies that maintaining alignment between the vortex axis and the swirling particle axis requires nontrivial wavy deformations. Pure radial expansion is insufficient; specific scale-local deformations are necessary to counteract the generation of shear.

3. Hybrid Optimization Framework:
   The paper successfully implements a Bayesian + Adjoint hybrid framework for fluid dynamics optimization. It demonstrates that while adjoint methods are efficient for local refinement, they fail to escape local minima in this highly non-convex landscape. Conversely, global search methods (Bayesian) can find qualitatively different shapes but lack precision. The combination yields the best results.

4. Numerical Results

• Experimental Setup: A vortex ring with Kelvin waves undergoes radial expansion. The time interval is discretized into 128 steps, and the angular parameter into 512 points.
• Scale Locality: The study varies the number of Fourier modes ($K$) allowed in the deformation.
  • Optimal $K$: The MSDC objective function peaks at $K=4$. This indicates that the most effective deformation for maintaining alignment is neither too smooth ($K=0$) nor too high-frequency, confirming scale-locality.
• Visualization:
  • No Optimization: The ring deforms significantly, and the axes misalign (low MSDC).
  • Bayesian Only: Finds a better shape but lacks fine-tuning.
  • Adjoint Only: Produces marginal shape changes but improves the objective function slightly, confirming it gets stuck in a local basin.
  • Hybrid (Bayesian + Adjoint): Achieves the highest MSDC, showing a distinct "bump" structure in the ring shape that preserves alignment.

5. Significance and Conclusion

• Theoretical Impact: The work bridges the gap between geometric curve dynamics and full fluid dynamics, proving that wave-like behaviors are intrinsic to smooth Euler flows, not just artifacts of singular approximations.
• Physical Insight: It provides evidence that vortex deformation is governed by scale-local mechanisms. The system naturally selects specific deformation frequencies (around $K=4$ in this setup) to minimize shear and maintain structural integrity.
• Methodological Impact: The paper demonstrates the necessity of hybrid optimization strategies in complex fluid mechanics problems where the loss landscape is non-convex. It offers a robust template for combining global exploration (Machine Learning) with local gradient-based refinement (Adjoint Methods) in Lagrangian flow optimization.

In summary, Yoneda's paper establishes a rigorous geometric framework for analyzing vortex ring deformations, derives new wave equations for smooth flows, and uses a novel hybrid optimization strategy to demonstrate that scale-local deformations are essential for maintaining vortex coherence.