Professor, what kind of problem is the backward-facing step flow?

It's a phenomenon where a flow separates and then reattaches at a step (a sudden change in geometry) within a flow channel. It has long been used as the most fundamental benchmark problem for separated and reattaching flows to validate CFD codes. The experimental data by Armaly et al. (1983) is particularly well-known.

So the image is that a vortex forms behind the step?

Correct. A recirculation zone (separation bubble) forms downstream of the step. The reattachment length, expressed as the ratio $x_r/h$ of the bubble length $x_r$ to the step height $h$, becomes a function of the Reynolds number. This is the most critical validation metric.

Backward-facing step flow

Category: Fluid Analysis (CFD) | Integrated 2026-04-06

CAE visualization for backward facing step theory - technical simulation diagram

Backward-Facing Step Flow

Backward-facing step flow: Theoretical Foundations

Overview

🧑‍🎓

Professor, what kind of problem is backward-facing step flow?

🎓

It's a phenomenon where there is a step (sudden change in height) in the middle of a flow channel, causing the flow to separate and then reattach. It is the most fundamental benchmark problem for separation and reattachment flows, and has long been used for verification of CFD codes. The experimental data by Armaly et al. (1983) is famous.

🧑‍🎓

So it's like a vortex forms behind the step?

🎓

Correct. A recirculation zone (separation bubble) forms downstream of the step. The reattachment length $x_r/h$, which is the length $x_r$ divided by the step height $h$, becomes a function of the Reynolds number. This is the most important verification metric.

Governing Equations

🧑‍🎓

The governing equations are the Navier-Stokes equations, right?

🎓

Yes, the incompressible Navier-Stokes equations and the continuity equation.

$$ \frac{\partial \mathbf{u}}{\partial t} + (\mathbf{u} \cdot \nabla)\mathbf{u} = -\frac{1}{\rho}\nabla p + \nu \nabla^2 \mathbf{u} $$

$$ \nabla \cdot \mathbf{u} = 0 $$

🎓

The Reynolds number is defined using the step height $h$ and the inlet average velocity $U$.

$$ Re = \frac{Uh}{\nu} $$

🧑‍🎓

What is the relationship between the reattachment length and Re?

🎓

In the laminar flow regime (approximately $Re < 400$), the reattachment length is almost proportional to Re. An approximation $x_r/h \approx 0.06 \times Re$ is known. In Armaly et al.'s experiment with an expansion ratio $ER = (H+h)/H = 1.94$, $x_r/h \approx 5$ for Re=100 and $x_r/h \approx 14$ for Re=400.

🧑‍🎓

So the reattachment point moves further downstream as Re increases.

🎓

However, around $Re > 400$, three-dimensional effects become significant, and 2D calculations alone no longer match experiments. This is the famous "2D-3D transition problem of the Armaly problem."

Flow Structure

🧑‍🎓

Are there structures other than the recirculation zone?

🎓

As Re increases, a secondary separation bubble appears on the upper wall opposite the step. Furthermore, small vortices form at the step corner. The overall flow field structure strongly depends on Re.

Re Range	Flow Characteristics
Re < 200	Only the primary recirculation on the lower wall
200 < Re < 400	Secondary separation bubble appears on the upper wall
Re > 400	Three-dimensional instability, spanwise fluctuations
Re > 1000	Turbulent transition, unsteady vortex shedding

🧑‍🎓

So 2D calculations are sufficient only up to about Re=400. I see, thank you.

Coffee Break Yomoyama Talk

Why the Reattachment Point Position Becomes the "Litmus Test" for Turbulence Model Selection

In backward-facing step flow, the "distance from the step to where the flow reattaches to the wall downstream (reattachment length)" is a standard metric for comparing turbulence model performance. Experimental values show reattachment occurs at a position about 6–8 times the step height. However, solving with the standard k-ε model tends to delay reattachment to around 9–11 times, while using the SST model improves accuracy. This difference—"different answers for the same problem"—is the practical basis for selecting turbulence models. Despite its simple geometry, the backward-facing step contains all elements of "separation, recirculation, and reattachment," making it the first testbed when developing new turbulence models.

Computational Methods for Backward-facing step flow

Numerical Methods

🧑‍🎓

What should I be careful about in the numerical solution of backward-facing step flow?

🎓

For this problem, the pressure-velocity coupling is important. Use standard incompressible flow algorithms like SIMPLE-type or Projection methods.

Pressure-Velocity Coupling

🧑‍🎓

How does the SIMPLE method work?

🎓

It stands for Semi-Implicit Method for Pressure-Linked Equations, proposed by Patankar & Spalding (1972). The procedure is as follows.

1. Obtain a provisional velocity field from the momentum equations

2. Solve the pressure correction equation (Poisson-type)

3. Correct the velocity and pressure

4. Repeat until convergence

🎓

Derived methods include SIMPLEC, PISO (suited for unsteady). For steady calculations, SIMPLE/SIMPLEC is standard; for unsteady, PISO is standard.

Spatial Discretization

🧑‍🎓

What scheme is good for the convection term?

🎓

Since backward-facing step flow involves recirculation, using upwind differencing (1st order) causes excessive numerical diffusion, making the reattachment length too long. At least 2nd order accuracy is required.

Scheme	Accuracy	Stability	Effect on Reattachment Length
1st Order Upwind	1st	High	Overestimation (high numerical diffusion)
2nd Order Upwind	2nd	Medium	Appropriate
QUICK	3rd	Slightly Low	Appropriate
Central Differencing	2nd	Low	Risk of oscillations

Mesh Design

🧑‍🎓

Where should the mesh be refined?

🎓

Step Corner: Sufficient resolution is needed as it is the separation point

Wall Region Near Reattachment Zone: Refine in the wall-normal direction to capture where the wall shear stress changes sign

Outlet: Set sufficiently far downstream ($x_{out} \geq 30h$). If the outlet is too close, it affects the reattachment length

🎓

The $y^+$ of the first wall layer is not necessary for laminar flow, but for turbulent calculations, $y^+ < 1$ (without wall functions) is desirable. For structured grids, use an expansion ratio of 1.1–1.2 to distribute cells away from the wall.

🧑‍🎓

So placing the outlet far away is important. 30h, that's quite long.

🎓

If it's too short, the outlet boundary condition influences and changes the reattachment length. During verification, sensitivity to outlet position should also be checked.

Coffee Break Yomoyama Talk

The "Density Disparity" Problem in Mesh Immediately After the Step

A problem often overlooked in numerical calculations of backward-facing steps is the "mesh transition immediately after the step corner." The separation point (step edge) is where the velocity gradient is maximum, requiring a dense mesh. However, if the mesh becomes abruptly coarse from there towards the reattachment point, numerical diffusion increases, leading to over- or under-estimation of the recirculation zone. An empirical rule is "use element sizes less than 1/10 of the step height around the step edge and expand downstream with an expansion ratio of 1.1 or less" as a safe strategy. Also, even if the reattachment length matches experiments in a 2D model, that setting often doesn't hold for 3D models because 3D effects (corner flows) are ignored. It's safer not to carry over settings from "2D verification → 3D production."

Backward-facing step flow

Backward-facing step flow: Theoretical Foundations

Overview

Governing Equations

Flow Structure

Why the Reattachment Point Position Becomes the "Litmus Test" for Turbulence Model Selection

Computational Methods for Backward-facing step flow

Numerical Methods

Pressure-Velocity Coupling

Spatial Discretization

Mesh Design

The "Density Disparity" Problem in Mesh Immediately After the Step

Related Topics

Related fields