Revisit eddy viscosity in pressure-driven wall turbulence at high Reynolds number

This study utilizes high-Reynolds-number DNS data to demonstrate that eddy viscosity in pressure-driven wall turbulence is configuration-dependent in the outer region, leading to a new Cess-type model with an outer correction function that significantly improves predictions for open-channel flow while maintaining accuracy for closed-channel and pipe flows.

The Gist

Imagine you are trying to predict how fast water flows through a river, a pipe, or a narrow canal. To do this, engineers use a concept called "eddy viscosity."

Think of eddy viscosity not as a physical property of the water, but as a measure of how much the water "mixes itself up." When water flows smoothly, it's like a line of soldiers marching in step. But when it gets turbulent, it's like a chaotic mosh pit where people are bumping into each other, swirling, and mixing. The "eddy viscosity" is a number that tells us how strong that mixing is.

For decades, scientists have used a "one-size-fits-all" recipe (called the Cess model) to guess this mixing number. They assumed that whether the water is flowing in a pipe, a closed channel, or an open river, the mixing happens the same way once you get away from the walls.

The Problem:
The authors of this paper, Ben-Rui Xu and Ao Xu, looked at massive supercomputer simulations (like high-definition movies of water molecules) and realized the old recipe was wrong.

• The Pipe: Imagine a pipe as a tunnel. The water in the middle is surrounded by walls on all sides. The mixing behaves one way.
• The Closed Channel: Imagine a river flowing between two concrete walls. Similar to the pipe, but flat.
• The Open Channel: Imagine a river with a free surface (the top is open to the air). Here, the water at the top can "slip" and slide freely without friction.

The old recipe assumed the mixing in the middle of the river (the open channel) was the same as in the middle of the pipe. It isn't. The "mosh pit" at the top of an open river behaves differently because the air above it doesn't drag on the water like a solid wall does.

The Solution: A Custom-Tailored Suit
The researchers decided to stop using the "one-size-fits-all" recipe. Instead, they built a new, smarter model.

1. The Detective Work: They used the supercomputer data to "reverse-engineer" the mixing. They asked, "If we know how fast the water is moving and how much stress is on it, what must the mixing number be?"
2. The Discovery: They found that the mixing number changes depending on the shape of the container. In a pipe, it curves up at the center. In an open river, it drops off smoothly at the surface.
3. The New Formula: They created a new mathematical formula (a "global model") that acts like a custom-tailored suit.
   • It keeps the same "near-wall" fit (the part close to the bottom) that everyone agrees on.
   • But it adds a special "outer correction" for the top part of the flow.
   • For a pipe, the suit fits snugly in the middle.
   • For an open river, the suit has extra room at the top to account for the free surface.

Why Does This Matter?
Think of it like driving a car.

• The old model was like driving a sedan on a race track, a dirt road, and a highway using the exact same suspension settings. It works okay on the highway, but it bounces too much on the dirt road.
• The new model is like having a car with adaptive suspension. It automatically stiffens for the race track (pipe) and softens for the dirt road (open river).

The Results:

• For Open Rivers: The new model is a huge improvement. It predicts the flow speed and friction much better than before, finally acknowledging that the top of the river is different from the bottom.
• For Pipes and Closed Channels: The new model is just as good as the old one, proving it doesn't break the things that were already working.

In a Nutshell:
This paper is about realizing that context matters. Just because water flows in a pipe doesn't mean it flows the same way in an open river. By creating a smarter, more flexible mathematical tool that respects the unique "personality" of each flow type, the authors have given engineers a better way to predict how fluids move, which helps in designing better pipelines, ships, and understanding natural rivers.

Technical Summary

1. Problem Statement

The study addresses the limitations of existing algebraic eddy-viscosity models in predicting wall-bounded turbulence across different canonical configurations at high Reynolds numbers ($Re_\tau$).

• The Gap: While classical models (e.g., the Cess model) provide a unified framework for plane closed-channel flow, open-channel flow, and pipe flow, they often fail to capture the configuration-dependent behavior of eddy viscosity ($\nu_t$) in the outer region.
• Specific Issue: The classical Cess model relies on a Reichardt-type outer representation derived primarily for pipe flow. It assumes a degree of universality that does not hold when comparing flows with different outer boundary conditions (e.g., centerline symmetry in closed channels/pipes vs. free-slip surfaces in open channels).
• Goal: To develop a robust, full-depth eddy-viscosity model that accurately captures the distinct outer-region physics of these three configurations using high-fidelity Direct Numerical Simulation (DNS) data.

2. Methodology

The authors employed a data-driven approach combined with physical scaling arguments:

• Data Source: The study utilized publicly available DNS databases spanning friction Reynolds numbers from $Re_\tau \approx 2000$ to $12,000$. The datasets cover:
  • Plane closed-channel flow (4 datasets).
  • Open-channel flow with a free-slip surface (2 datasets).
  • Pipe flow (3 datasets).
• Inference of Eddy Viscosity: Using the Boussinesq hypothesis ($-\overline{u'v'} = \nu_t \frac{du}{dy}$), the authors inferred the eddy viscosity distribution $\nu_t(y)$ directly from DNS one-point statistics (mean velocity and Reynolds shear stress).
• Scaling Analysis:
  • The authors interpreted $\nu_t$ as the product of a velocity scale ($u_c$) and a length scale ($\ell$).
  • They extended the log-law scaling into the outer region by defining a local velocity scale based on the linear total shear stress distribution: $u_c \approx u_\tau \sqrt{1 - y/\delta}$.
  • They retained the mixing-length form $\ell_m = \kappa y$ but introduced an outer correction function $f(y/\delta)$ to account for configuration-specific variations in the turbulent length scale.
• Model Formulation:
  • A parametric form for the correction function was proposed: $\kappa f(y/\delta) = \kappa \alpha (1 - y/\delta)^p (1 + y/\delta)^q$.
  • This correction was embedded into a Cess-type framework, combining the new outer correction with a van Driest near-wall damping function.
  • The resulting model is a closed-form analytical expression applicable over the full depth/radius.

3. Key Contributions

1. Identification of Configuration Dependence: The study demonstrates that the outer-region eddy viscosity is not universal. Specifically, $\nu_t$ exhibits non-monotonic behavior near the centerline in closed-channel and pipe flows, whereas it decreases monotonically toward the free surface in open-channel flow.
2. New Outer Correction Function: The authors derived a compact, parametric correction function $f(y/\delta)$ that captures these distinct trends. The exponents $p$ and $q$ are calibrated to specific flow types, allowing the model to handle both symmetric (closed) and asymmetric (open) boundary conditions.
3. A New Global Eddy-Viscosity Model: The paper presents a modified Cess model (Eq. 32) that replaces the rigid Reichardt polynomial with the proposed flexible correction function. This yields a single analytical expression that remains accurate across all three configurations.
4. Parametric Calibration: The study provides specific, configuration-dependent parameters (including the von Kármán constant $\kappa$, damping constant $A$, and correction exponents $p, q, \alpha$) for closed-channel, open-channel, and pipe flows, validated against high-$Re$ DNS data.

4. Results

The proposed model was validated against DNS data using three metrics:

• Eddy Viscosity Profiles: The new model significantly improves the prediction of $\nu_t$ in the outer region compared to the classical Cess model and $k-\epsilon$ estimates. The improvement is most dramatic for open-channel flow, where the classical model fails to capture the monotonic decay of eddy viscosity near the free surface.
• Log-Law Indicator Function ($\Xi$): The model accurately reproduces the DNS-based indicator function ($\Xi = y^+ du^+/dy^+$). While the classical Cess model performs slightly better in the logarithmic region for closed-channel and pipe flows, the new model offers superior consistency across the full depth, particularly for open channels.
• Mean Velocity and Skin Friction:
  • The reconstructed mean velocity profiles show better agreement with DNS in the outer region for open-channel flows.
  • Skin friction predictions ($C_f$ and friction factor $\lambda$) are improved for open-channel flows. For closed-channel and pipe flows, the results are comparable to the classical Cess model, with a slight over-prediction of friction attributed to the interaction between the outer correction and near-wall damping.

5. Significance

• Theoretical Insight: The work underscores the critical role of outer boundary conditions in shaping the outer-region eddy viscosity and, consequently, the mean flow structure. It challenges the assumption of a single universal outer-layer eddy-viscosity profile.
• Practical Application: The proposed model provides a robust, configuration-aware baseline for:
  • RANS Simulations: Offering improved closure models without the computational cost of solving transport equations.
  • Wall-Modeled LES: Providing more accurate wall models for high-Reynolds-number simulations.
  • Resolvent Analysis: Serving as a better baseline for linear stability and resolvent-based analyses across different flow geometries.
• Future Direction: The study highlights that while the outer correction is essential for cross-configuration robustness, future refinements should focus on optimizing the blending strategy between the near-wall damping and the outer correction to further improve predictions in the logarithmic region for symmetric flows.

In summary, this paper successfully revisits and modernizes the eddy-viscosity concept for high-Reynolds-number wall turbulence, moving from a "one-size-fits-all" approach to a configuration-aware framework that significantly enhances predictive accuracy, particularly for open-channel flows.