A variable-coefficient model for decay of isotropic turbulence capturing effects of finite cascade time and Reynolds number

This paper proposes a variable-coefficient $C_{\epsilon2}$ model for the $k$-$\epsilon$ turbulence framework that accounts for finite cascade time and Reynolds number effects, thereby accurately capturing the decay and growth of isotropic turbulence across diverse flow scenarios.

The Gist

Imagine turbulence (the chaotic, swirling motion of fluids like air or water) as a giant, complex waterfall. In this waterfall, big waves break down into smaller ripples, which break down into even smaller splashes, until the energy finally disappears as heat. This process is called the "energy cascade."

For decades, engineers have used a set of rules (mathematical models) to predict how this waterfall behaves. One of the most popular rulebooks is called the $k-\epsilon$ model. It tries to guess two things: how much energy is in the water ($k$) and how fast that energy is disappearing ($\epsilon$).

However, there's a specific "dial" in this rulebook, called $C_{\epsilon2}$, that controls how fast the energy disappears. For a long time, scientists assumed this dial was fixed—like a thermostat set to a permanent temperature. They thought it didn't matter if the water was flowing fast or slow, or if you just started the flow or let it run for a while; the dial stayed the same.

The Problem:
The authors of this paper, researchers from Stanford and Los Alamos, ran incredibly detailed computer simulations (like high-definition movies of the waterfall) and found that the old rulebook was wrong. They discovered that the "dial" ($C_{\epsilon2}$) isn't fixed. It actually moves.

Think of it like a car engine. If you suddenly step on the gas (inject energy), the engine doesn't react instantly; it takes a moment to rev up. Similarly, in turbulence, it takes a finite amount of time for energy to travel from the big waves down to the tiny ripples where it disappears. This "travel time" changes depending on how fast the water is moving (the Reynolds number) and whether you are adding energy or letting the flow die out.

The Discovery:
By watching their high-definition simulations, the researchers saw that:

1. When turbulence is dying out (decaying): The "dial" starts at one value and slowly shifts to a new, stable value over time. It's not instant; it has a "memory" of how the flow started.
2. When you force turbulence to grow (adding energy): The "dial" drops significantly. The system is out of balance because energy is being pumped in faster than it can cascade down to the tiny ripples to be burned off.

The Solution:
Instead of treating the dial as a fixed number, the authors created a new rule that makes the dial a variable. They wrote a new equation that tells the dial how to move based on two things:

• The current speed of the flow (Reynolds number).
• The history of the flow (Did we just turn it on? Is it dying out? Is it being forced to grow?).

They compared this new, "smart" dial against their high-definition simulations. The results showed that the old, fixed-dial model often got the timing wrong, predicting that energy would disappear too fast or too slow. The new model, which lets the dial change, matched the real physics almost perfectly.

The Analogy:
Imagine you are trying to predict how long a campfire will burn.

• The Old Model: Assumes the fire burns at a constant rate no matter what. If you add a log, it just keeps burning at the same speed.
• The New Model: Recognizes that when you add a log, the fire doesn't instantly burn at a new rate. It takes time for the new wood to catch, for the heat to spread, and for the flames to adjust. The "burn rate" changes dynamically based on how much wood you just threw in and how big the fire was a moment ago.

The Bottom Line:
This paper doesn't claim to solve every problem in fluid dynamics. It specifically focuses on isotropic turbulence (turbulence that looks the same in all directions, like a perfectly mixed pot of soup). The authors successfully proved that by making the "decay coefficient" a moving target that reacts to the flow's history and speed, they can predict how turbulence dies out or grows much more accurately than the standard, fixed-coefficient models.

They acknowledge that this is a first step. Their model works great for these specific, controlled simulations, but it still needs to be tested in more complex, real-world scenarios (like air flowing over a wing) before it can be used in everyday engineering design.

Technical Summary

Technical Summary: A Variable-Coefficient Model for Decay of Isotropic Turbulence

Problem Statement
In the context of Reynolds-Averaged Navier-Stokes (RANS) modeling, the $k-\epsilon$ framework is a standard approach for predicting turbulent flows. A critical parameter in this family of models is the coefficient $C_{\epsilon2}$, which governs the decay of turbulence kinetic energy ($k$) and dissipation rate ($\epsilon$). Traditionally, $C_{\epsilon2}$ is treated as a constant (typically 1.92), linking the temporal decay power-law exponent $n$ via the relation $n = 1/(C_{\epsilon2} - 1)$. However, extensive computational and experimental literature indicates that the decay exponent $n$ is not universal; it varies with the instantaneous Reynolds number ($Re_T$) and the history of energy injection (initial conditions). Furthermore, the assumption of a constant $C_{\epsilon2}$ fails to capture the dynamics of turbulence during non-equilibrium states, such as rapid growth or decay, where the finite time required for the energy cascade from large to dissipative scales plays a significant role. Existing attempts to model this variability, such as multi-scale models, often rely on unmeasurable internal variables, limiting their practical utility.

Methodology
The authors employ high-fidelity Direct Numerical Simulations (DNS) and Large Eddy Simulations (LES) to investigate the mathematical terms responsible for isotropic turbulence decay.

• Simulations: The study utilizes a pseudospectral code to solve incompressible momentum equations on a triply periodic domain. Simulations cover a range of Reynolds numbers ($Re_T$ from 78 to 382) for both stationary forced turbulence and scenarios involving free decay and forced growth.
• Forcing Strategy: To maintain domain-independent turbulence, a linear forcing term is applied using a high-pass filter on the velocity field, injecting energy only at wavenumbers $\kappa > 2$. This mimics shear production in a controlled manner.
• Data Analysis: The authors analyze the exact transport equation for the dissipation rate ($\epsilon$). They observe that while individual terms in the dissipation equation (specifically turbulent production $P_\epsilon$ and viscous dissipation $Y$) diverge with increasing Reynolds number, their combined effect ($P_\epsilon - Y$) remains convergent and insensitive to $Re_T$ in the limit of large Reynolds numbers.
• Model Derivation: By computing the instantaneous $C_{\epsilon2}$ from DNS data using the relation $C_{\epsilon2} = -(P_\epsilon - Y)k/\epsilon^2$, the authors observe that $C_{\epsilon2}$ is not constant. It evolves over time, is sensitive to the Reynolds number, and responds to the history of energy injection. Based on these observations, they propose a phenomenological evolution equation for $C_{\epsilon2}$ rather than deriving it from first principles.

Key Contributions
The primary contribution of this work is the development of a variable-coefficient model for $C_{\epsilon2}$ that retains a measurable input space while capturing finite cascade time and Reynolds number effects.

1. Evolution Equation for $C_{\epsilon2}$: The authors propose a first-order damped evolution equation (Eq. 13) where $C_{\epsilon2}$ is treated as a dynamic variable:
   $$\frac{dC_{\epsilon2}}{dt} = A_D \frac{\epsilon - P}{k} (C_{\epsilon D} - C_{\epsilon2}) + A_F \frac{P}{k} (C_{\epsilon F} - C_{\epsilon2})$$
   Here, $P$ is the production rate. The equation distinguishes between decay behavior (subscript $D$) and forced behavior (subscript $F$).
2. Reynolds Number Dependence: The model coefficients ($A_D, A_F, C_{\epsilon D}$) are formulated as functions of $Re_T$. Based on scaling analysis, the leading-order finite Reynolds number corrections are proportional to $Re_T^{-1/2}$.
3. Theoretical Constraints: The model enforces $C_{\epsilon F} = 1$ for stationary forced turbulence, a value derived theoretically for linear forcing, and determines asymptotic values for $C_{\epsilon D}$ in the limits of low and high Reynolds numbers ($C^0_{\epsilon D} = 1.5$ and $C^\infty_{\epsilon D} = 1.73$, respectively).

Results
The proposed model was tuned using DNS data for $Re_T = 78, 133, 180,$ and $382$, and validated against an independent case ($Re_T = 240$) and an infinite Reynolds number LES case.

• Dynamic Behavior: The model successfully reproduces the non-monotonic evolution of $C_{\epsilon2}$ observed in DNS. For decaying turbulence, $C_{\epsilon2}$ increases from an initial value and asymptotes to a Reynolds-dependent value. For growing turbulence, $C_{\epsilon2}$ decreases, dropping below unity before rising again.
• Prediction Accuracy: When integrated into a RANS solver, the variable $C_{\epsilon2}$ model accurately predicts the time evolution of both turbulent kinetic energy ($k$) and dissipation rate ($\epsilon$) across decay and growth scenarios.
• Comparison to Standard Models: The proposed model significantly outperforms the standard $k-\epsilon$ model (which assumes a constant $C_{\epsilon2} = 1.92$). The standard model fails to capture the rapid decay of TKE in infinite Reynolds number cases and exhibits quantitative inaccuracies in growth regimes with varying energy injection rates.
• Robustness: The model demonstrates insensitivity to the ratio of turbulence growth time scales to large eddy turnover times, performing well even when tested with energy injection rates 50% lower and higher than the tuning data.

Significance and Claims
The authors position this work as an incremental step toward improving existing turbulence models. They claim that by treating $C_{\epsilon2}$ as a variable governed by an evolution equation, the model captures essential physics—specifically the finite time of the energy cascade and Reynolds number effects—that are lost in constant-coefficient approaches. The study emphasizes that the model is phenomenologically proposed based on DNS observations rather than rigorously derived.

The authors explicitly note limitations and future requirements for broader application:

• The current data is constrained by finite box sizes and a limited range of Reynolds numbers.
• The model has not yet been validated for inhomogeneous flows; extending it to include transport terms is a necessary future step.
• The model is currently valid for regimes where production is comparable to dissipation ($P \leq O(\epsilon)$) and may not be suitable for extreme or abrupt energy injections without higher-order corrections.
• The assumption of linear forcing used to derive $C_{\epsilon F} = 1$ may differ from physical production mechanisms, though the authors argue the linear forcing is a reasonable proxy for mean velocity gradient effects.

In summary, the paper demonstrates that a variable $C_{\epsilon2}$ model, informed by high-fidelity data and scaling arguments, offers a more accurate representation of isotropic turbulence decay and growth than traditional constant-coefficient models, providing a foundation for future improvements in RANS modeling.