Random field reconstruction of inhomogeneous turbulence. Part II: Numerical approximation and simulation

This paper presents and validates a numerical discretization scheme for a novel stochastic model that reconstructs inhomogeneous turbulent velocity fluctuations, demonstrating its accuracy, efficiency, and ability to capture key physical properties like spatio-temporal ergodicity and Kolmogorov's scaling laws through comprehensive simulations.

The Gist

Imagine you are trying to recreate the chaotic, swirling motion of wind or water (turbulence) on a computer. In the real world, this flow is rarely uniform; it changes speed, direction, and "roughness" depending on where you are and when you look. This paper is about building a better, more realistic digital model for these messy, changing flows.

Here is the breakdown of what the authors did, using simple analogies:

1. The Problem: The "Static" vs. The "Living" Flow

Previous computer models of turbulence were often like a stiff mannequin. They could show a flow, but they struggled to change shape realistically as the flow moved from a wide river to a narrow stream, or from calm to stormy. They often treated the math as a "half-finished" sketch, making it hard to prove if the model was actually accurate or just a lucky guess.

The authors previously built a new "blueprint" (a mathematical formula) that acts like a living organism. It can stretch, shrink, and speed up or slow down based on local conditions (like how much energy is in the flow at that specific spot). However, a blueprint on paper is useless if you can't build it.

2. The Solution: The "Digital Construction Kit"

This paper is the instruction manual for building that blueprint on a computer. The authors created a specific recipe (a numerical scheme) to turn their complex math into a simulation you can actually run.

Think of their method as a high-tech sound mixer:

• The Ingredients: Instead of using a smooth, continuous stream of sound (which is impossible for a computer to handle perfectly), they break the sound down into thousands of tiny, individual "beats" or "waves."
• The Randomness: They don't just pick these beats in a boring, predictable order. They use a randomized lottery system. Imagine throwing thousands of darts at a board to decide where the sound waves come from. This randomness is crucial because it prevents the computer simulation from creating fake, repeating patterns (like a broken record) that don't exist in real life.
• The "Local" Trick: Real flows change as you move through them. The authors' method is smart enough to "zoom in" on specific spots. It doesn't need to simulate the entire universe to tell you what the wind feels like at your front door. It can calculate the turbulence for just one point, then move to the next, keeping the "story" consistent as it goes.

3. Proving It Works: The "Taste Test"

Before showing off the simulation, the authors had to prove their construction kit actually builds what they promised.

• The Math Check: They used rigorous math to show that as they add more and more "beats" (more darts thrown), their digital model gets closer and closer to the perfect, theoretical blueprint. It's like showing that if you add enough pixels to a low-res image, it eventually looks like a high-definition photo.
• The "Ergodicity" Test: This is a fancy word for "does the average match the reality?" They showed that if you watch a single simulation for a long time, or look at a snapshot of the whole field, the average energy and "friction" (dissipation) match the input data perfectly. It's like proving that if you take a sample of soup from one spoonful, it tastes the same as the whole pot.

4. The Results: Watching the Model Dance

The authors ran several simulations to show off the model's features:

• Changing Sizes: They showed that when the model enters a region where the flow is "larger" (more energy), the swirling patterns in the simulation get bigger. When the flow gets "smaller," the swirls shrink.
• Changing Speed: They demonstrated that the model can speed up or slow down the "heartbeat" of the turbulence depending on the local conditions.
• The "Kolmogorov" Law: In the world of turbulence, there is a famous rule (Kolmogorov's two-thirds law) about how energy breaks down from big swirls to tiny ones. The authors proved their model follows this rule correctly, even in messy, changing environments, provided the flow is turbulent enough.

Summary

In short, this paper takes a sophisticated mathematical idea for modeling messy, changing winds and waters and turns it into a working computer program. They proved the program is mathematically sound, showed it can handle local changes without needing to simulate the whole world, and demonstrated that it creates realistic, swirling patterns that obey the laws of physics.

What they did NOT do:
The paper focuses strictly on the math and the computer code. They did not test this on real-world engineering problems (like designing a car or a plane) or medical applications. They simply built the engine and proved it runs smoothly; they didn't drive it to a destination yet.

Technical Summary

1. Problem Statement

The paper addresses the challenge of numerically reconstructing inhomogeneous turbulent velocity fluctuations from characteristic flow quantities (such as mean velocity, turbulent kinetic energy $k$, dissipation rate $\varepsilon$, and kinematic viscosity $\nu$). While Part I of this work established a rigorous, fully continuous random field model based on stochastic Fourier-type integrals, this paper focuses on the numerical discretization, convergence analysis, and algorithmic implementation of that model.

Key challenges addressed include:

• Inhomogeneity: Handling spatial and temporal variations in turbulence scales and Reynolds numbers.
• Non-uniform Advection: Accurately modeling the transport of turbulent structures by non-uniform mean flows.
• Local Evaluation: Developing an algorithm that allows for flexible, localized sampling of the turbulence field without requiring global simulation of the entire spatio-temporal domain (crucial for Lagrangian particle tracking or fiber dynamics).
• Consistency: Ensuring that the discrete numerical scheme preserves the statistical properties (e.g., Reynolds stresses, dissipation rates, ergodicity) of the continuous theoretical model.

2. Methodology

A. The Underlying Model (Recap)

The authors utilize the model introduced in Part I [Ant+26], which represents the turbulent velocity fluctuation $u'(x,t)$ as a centered Gaussian random field defined by stochastic integrals. The model employs a two-scale asymptotic approach:

• Macro-scale: Defined by flow geometry and mean flow $u(x,t)$.
• Micro-scale: Defined by turbulent fluctuations, scaled by local parameters ($k, \varepsilon, \nu$).
• Representation: The field is constructed via a moving average in time and a spectral representation in space, driven by a Gaussian white noise measure. It incorporates a local linearization of the mean flow advection to handle non-uniform transport.

B. Discretization Scheme

The authors propose a stratified randomized quadrature method to approximate the stochastic integrals:

1. Stratification: The time domain is partitioned into intervals $I_j = [j\Delta s, (j+1)\Delta s)$.
2. Randomized Quadrature: The continuous white noise measure is replaced by a finite sum of randomly distributed Dirac measures with random weights.
   • Wave numbers ($\kappa$): Sampled from a reference probability density function (typically the energy spectrum).
   • Orientation vectors ($\theta$): Uniformly distributed on the unit sphere $S^2$.
   • Temporal points ($s$): Uniformly distributed within their respective stratification intervals.
   • Noise vectors ($\xi$): Complex Gaussian random variables with zero mean and identity covariance.
3. Linearization: The complex integral equation for the mean flow path $\phi(s; x, t)$ is approximated by a linear function $x - (t-s)u(x,t)$, valid for small turbulence scale ratios ($\delta \ll 1$).

C. Convergence Analysis

The paper provides a rigorous analytical proof of convergence (Theorem 3.2):

• Step 1: Proves that the auxiliary field (using the linearized mean flow) converges to the original continuous model as the turbulence scale ratio $\delta \to 0$.
• Step 2: Proves that the discretized field (using the randomized quadrature) converges weakly to the auxiliary field as the number of quadrature points $N \to \infty$.
• Result: The scheme preserves the covariance structure and characteristic flow properties (kinetic energy, dissipation rate, incompressibility) in the limit.

D. Algorithmic Implementation

A key contribution is Algorithm 4.1, a sampling procedure designed for flexible local evaluation:

• Dynamic Sampling: The algorithm allows evaluating the field at arbitrary points $(x,t)$ determined dynamically during a simulation (e.g., particle positions).
• Consistency: It ensures that if two evaluation points share overlapping time integration kernels, they use the same random numbers (stratification intervals and samples) to maintain statistical consistency.
• Memory Management: The algorithm tracks which stratification intervals have been "used" and generates new random samples only for intervals relevant to the current and future time steps, avoiding the need to store the entire global field.

3. Key Contributions

1. Rigorous Discretization: The first fully specified numerical scheme for the continuous inhomogeneous turbulence model, combining stratified Monte Carlo quadrature with local mean flow linearization.
2. Convergence Proof: Analytical verification that the discrete scheme converges to the continuous model and preserves key statistical properties (Corollary 3.3).
3. Efficient Local Sampling Algorithm: A novel algorithm that enables localized, on-the-fly evaluation of the turbulence field while maintaining consistency across time steps, solving a major bottleneck for Lagrangian simulations.
4. Validation of Ergodicity: Numerical demonstration that local spatial and temporal averages of a single sample path accurately recover the underlying inhomogeneous flow parameters ($k$ and $\varepsilon$).
5. Kolmogorov's Law Verification: Demonstration that the model correctly reproduces Kolmogorov's two-thirds law for spatial velocity increments in the inertial range, dependent on the local turbulence Reynolds number.

4. Simulation Results

The authors present extensive numerical simulations to illustrate the model's features:

• Parameter Influence (Section 5.1):

  • Varying the spatial scaling factor $\sigma_x$ (related to $k^{3/2}/\varepsilon$) changes the size of turbulent structures.
  • Varying the viscosity scaling factor $\sigma_z$ (related to inverse Reynolds number) alters the spectral composition (ratio of large to small scales).
  • Varying the velocity scaling factor $\sigma_u$ changes the amplitude of fluctuations without altering their shape.
  • The model successfully captures non-uniform advection, where structures stretch and evolve according to the mean flow while decaying temporally.

• Spatio-Temporal Ergodicity (Section 5.2):

  • Simulations show that local moving averages (over line segments and 3D balls) of a single sample path converge to the prescribed mean fields of turbulent kinetic energy ($k$) and dissipation rate ($\varepsilon$).
  • This confirms the model's validity for recovering macroscopic statistics from microscopic realizations.

• Kolmogorov's Two-Thirds Law (Section 5.3):

  • The paper verifies that the second-order structure functions of the velocity increments follow the $r^{2/3}$ scaling law in the inertial subrange.
  • The results show that this scaling holds for various local Reynolds numbers, with the width of the inertial range expanding as the Reynolds number increases.
  • The proportionality constants match theoretical predictions ($C_l \approx 1.97$).

5. Significance

This paper bridges the gap between the theoretical formulation of inhomogeneous turbulence models and their practical application in computational fluid dynamics (CFD).

• For RANS/LES: It provides a robust method for generating synthetic turbulence fields from RANS data that are statistically consistent with the underlying physics (including anisotropy and inhomogeneity).
• For Particle/Fiber Dynamics: The efficient local sampling algorithm is critical for simulating the interaction of particles or fibers with turbulence, where the evaluation points are not known a priori.
• Theoretical Rigor: By separating modeling, analysis, and numerical approximation, the authors provide a framework that ensures the numerical solution is not just a heuristic approximation but a mathematically convergent representation of the continuous stochastic model.

In summary, the work delivers a complete, validated, and computationally efficient toolkit for simulating inhomogeneous turbulence, capable of handling complex, time-varying flow conditions while preserving fundamental physical laws like Kolmogorov's scaling and ergodicity.