Backward-facing step calculator: reattachment length, blockMesh grading and a mesh check

Enter the step, the inlet velocity and the viscosity, and get the Reynolds number under each common definition, the flow regime and the reattachment length a 2-D laminar simulation should give. Then work out blockMesh grading cell by cell, or paste your blockMeshDict to check cell sizes, jumps between blocks, the faces blockMesh will refuse to merge, and your time step. Free, no sign-up.

Erturk 2008 · Armaly 1983 · Barkley 2002 · Durst 2005 · OpenFOAM blockMesh source · 39 verifier checks · Updated 20 Nov 2026

Geometry (mm)

Flow

Defaults: the Skill-Lync OpenFOAM challenge geometry (10 mm inlet channel over a 10 mm step, 80 mm before and 120 mm after it, velocity probed at x = 85 mm) with U = 0.1 m/s and ν = 10⁻⁵ m²/s, as one report used. The brief fixes the geometry; reports used very different velocities and viscosities.

ReD = U·2hi/ν200Reh = 100 · ER = 2.00
Reattachment x1 (2-D)4.98 h49.8 mm behind the step
Regime2-D laminarReD below 400
010203040500400120020003000Re_D = U·2h_i/ν; lengths in step heightsx1 reattachmentx2, x3 upper-wall bubble3-Dtransitional
  • Erturk (2008), ER = 2: x1 = 4.982 h at Re_D = 200 for a steady 2-D solution with a fully developed inlet.

Reattachment length of a backward-facing step

Flow leaving the step corner cannot turn the corner. It separates, forms a recirculation bubble behind the step and touches the bottom wall again some distance downstream: the reattachment length x1, usually quoted in step heights. In laminar flow that bubble grows almost linearly with Reynolds number, which is why x1 is the standard number to compare a backward-facing step simulation against.

The first thing to settle is which Reynolds number you mean. The laminar benchmarks (Armaly et al. 1983, Gartling 1990, Erturk 2008) use the mean inlet velocity and the hydraulic diameter of the inlet channel, twice its height:

For the Skill-Lync challenge geometry (hi = h = 10 mm, so ER = 2) with U = 0.1 m/s and ν = 10⁻⁵ m²/s, ReD = 200 but Reh = 100. Barkley et al. (2002) use the peak inlet velocity and the step height, which for ER = 2 is ¾ of ReD. Quote the definition with the number, or comparisons are off by a factor of 2.

Erturk's 2-D steady solutions for ER = 2, with a fully developed inlet 20 step heights upstream and the outlet 300 step heights downstream, give x1 = 2.922 h at ReD = 100, 4.982 h at 200, 8.237 h at 400, 11.834 h at 800 and 22.459 h at 3,000. From ReD = 400 a second bubble opens on the upper wall (x2 to x3 in the chart). At ReD = 800, the classic test case, published 2-D results range from 11.48 h to 12.2 h (Gartling 12.20, Barkley 11.91, Erturk 11.834); Erturk attributes the spread to outlet position and grid.

Two limits matter more than any of those digits:

  • 2-D is only real below about ReD = 400. Armaly's air experiment (aspect ratio 1:36) became measurably three-dimensional there, and 2-D simulations agree with it only up to that point; above it they underpredict x1. Barkley's linear stability analysis puts the first 3-D instability of the 2-D flow at ReD ≈ 997.
  • Laminar ends around ReD = 1,200. Above it the bubble shrinks through a transitional range and levels off as turbulent flow above about 6,200 to 6,600 (sources differ). Turbulent reattachment is near 6 to 7 step heights: 6.26 ± 0.10 h in Driver & Seegmiller's experiment (Reh ≈ 36,000) and 6.28 h in Le, Moin & Kim's DNS (Reh = 5,100, ER = 1.2). icoFoam is a laminar solver, so it has no business above that range.

The inlet matters too. The benchmarks use a parabolic (developed) inlet. A uniform fixedValue inlet only becomes parabolic after a development length, which Durst et al. (2005) correlate for a plane channel as:

At U = 0.1 m/s that is 45 mm, inside the challenge's 80 mm inlet channel, so the comparison with Erturk holds. At U = 1 m/s with the same viscosity it is 442 mm, and the profile reaching the step is still nearly flat.

blockMesh grading, cell by cell

simpleGrading (a b c) gives one expansion ratio per block direction: the width of the last cell along that direction divided by the width of the first. blockMesh spaces the n cells as a geometric series (OpenFOAM's lineDivide.C):

So the challenge's "grading factor 0.2" means the cells shrink to a fifth along the edge, and 5 means they grow five-fold. On the 120 mm edge after the step with 120 cells and a ratio of 5, the cells grow from 0.402 mm at the step to 2.008 mm at the outlet, each 1.0136 times the one before; across a 5 mm block with 5 cells and the same ratio, each cell is 1.495 times the last. Direction matters: the ratio runs from the block's first vertex to its second (x1), first to fourth (x2) and first to fifth (x3), so the same physical edge can need 5 in one block and 0.2 in its neighbour.

Multi-grading splits an edge into sections, each (fraction of length, fraction of cells, ratio). blockMesh normalises both fractions to sum to 1 and rounds the cell counts, giving any remainder to the section with the most cells. ((0.5 0.5 5) (0.5 0.5 0.2)) makes an edge fine at both ends and coarse in the middle.

Where to be fine. The brief asks for cells "finer near the walls (including the step wall)". The steepest gradients are at the walls and along the line leaving the step corner, where the separated shear layer starts. In the common Case 2 mesh the block above that line (block 3) is graded fine at it, 0.38 mm, while the block below (block 4) is graded fine at the bottom wall and arrives at the line with 1.97 mm cells: a 5.14× jump in cell size along the whole shear layer. Ansys's CFX best-practice guide asks for growth factors below 1.3 and to "avoid jumps in grid density". Giving block 4 two-sided grading, simpleGrading (5 ((0.5 0.5 5) (0.5 0.5 0.2)) 1), makes its cells 0.38 mm on both sides of the line (jump 1.00×) without changing the cell count. With only 5 cells in each half, each cell is still 1.50 times the last, the same as the template's other 5-cell blocks; more cells across the channel bring that growth under 1.3.

When blockMesh refuses. Blocks are glued face by face: every point on one block's face must find a point on the neighbour's face within about a third of the smallest point spacing there (√(1/10), blockMeshMerge.C). If two blocks grade a shared edge differently, the points miss and blockMesh stops with "Inconsistent point locations between block pair … probably due to inconsistent grading". Changing block 2's x-grading in Case 2 from 5 to 0.2, as one report did, moves the points on its face with block 3 by up to 0.92 mm against a tolerance of 0.127 mm.

Time step and Courant number for icoFoam

The OpenFOAM user guide's cavity tutorial: "to achieve temporal accuracy and numerical stability when running icoFoam, a Courant number of less than 1 is required", with Co = Δt|U|/Δx, and Δt chosen for the worst cell: the smallest cell along the flow with the highest velocity. In a developed laminar channel the peak velocity is 1.5 times the mean, so:

Grading shrinks Δxmin. Case 2's smallest streamwise cell is 0.401 mm next to the step, so Δt ≤ 2.68 ms at 0.1 m/s and 0.268 ms at 1 m/s. The same edges ungraded (1 mm cells) would allow 2.5 times more.

What 765 student reports show

Skill-Lync's OpenFOAM course (Week 8) asks for a transient, incompressible, laminar simulation of a backward-facing step with 200 × 20 cells, once ungraded and once with a grading factor of 0.2 and cells finer at the walls, and the velocity profile at 0.085 m from the inlet. A CONVERGE course has its own step exercise. 765 reports with "backward" in the title are public; we read all of them and extracted 738 blockMeshDicts from 357 (20 Nov 2026, counts from pattern searches with samples checked by hand):

  • A 5× jump along the shear layer. Of 413 graded meshes that blockMesh can build, 338 (in 254 reports) have a cell-size jump of 2× or more across a block face; in 290 of them it sits on the line leaving the step corner. The most common value is 5.14×, from the Case 2 template above.
  • Meshes blockMesh cannot build. 8 meshes in 7 reports grade a shared face differently on its two sides, which blockMesh rejects as inconsistent grading.
  • "Laminar" at almost any Reynolds number. Of 127 reports where both the inlet velocity and ν can be read, ReD runs from 0.1 to over a million, median 1,000. Only 45 are below 400, where a 2-D laminar result can be compared with experiment, and 60 (47%) are above 1,200, past the laminar range icoFoam assumes.
  • Units 10× out. 39 reports have a mesh whose step is 10, 100 or 1,000 times the challenge's 10 mm, or a tenth of it, from convertToMeters or the vertex units; their profile "at 0.085 m" is then nowhere near 5 mm behind the step.
  • Courant numbers above 1. For 274 reports with an inlet velocity, a deltaT and a mesh, the Courant number at the smallest streamwise cell, using only the inlet velocity, is above 1 in 66 (up to 31.1).
  • Reattachment rarely measured. 129 reports mention reattachment, but fewer than 10 give a reattachment length at all, though it is the one number every benchmark publishes.

What this calculator does not cover

  • Turbulent steps. The expected x1 is for steady 2-D laminar flow (ReD 100 to 3,000 for ER = 2, 100 to 1,500 for ER = 1.942). Turbulent x1 depends on the inflow boundary layer and the turbulence model.
  • Other expansion ratios. Only ER = 2 and 1.942 have tables here; other ratios get the Reynolds numbers and the mesh checks but no expected x1.
  • Curved edges and non-rectangular blocks. Arc and spline edges are drawn and checked as straight lines; cell sizes in skewed blocks are taken along the block edges.
  • Mesh quality beyond size. Skewness and non-orthogonality need checkMesh; for a rectangular step they are zero.
Learn CFD with OpenFOAM

The backward-facing step is a project in Skill-Lync's CFD course with MATLAB and OpenFOAM: blockMesh, icoFoam, grading, boundary conditions and checking a result against a benchmark.

How we checked the calculator

The independent Python script scripts/verify/bfs.py runs 39 checks, and this page recomputes its numbers with the browser code at build time and refuses to build if they differ.

  • Grading: cell widths against the closed-form geometric series for six edges; the OpenFOAM user guide's multi-grading example ((0.2 0.3 4) (0.6 0.4 1) (0.2 0.3 0.25)) on 20 cells gives 6/8/6 cells ending exactly at 0.2 and 0.8; normalisation and the rounding rule; the inverse (ratio from a first cell) recovers the ratio.
  • Face merging: the Python twin and the browser code give identical results on all 738 blockMeshDicts from the reports, and a separate check that compares whole face point clouds agrees with the edge check on every face tested; the mismatched Case 2 variant fails and the two-sided fix brings the jump to 1.00.
  • Flow: Erturk's tables copied and spot-checked (x1 = 4.982 at 200, 11.834 at 800), the Reynolds-number cross-walk from Barkley's Table 1, and Durst's channel length in both its channel-height and hydraulic-diameter forms.

Frequently asked questions

What is the reattachment length of a backward-facing step?

The distance behind the step where the separated flow touches the bottom wall again, usually given in step heights (x1/h). For laminar 2-D flow it grows almost linearly with Reynolds number: for an expansion ratio of 2, Erturk (2008) gives 2.922 h at Re_D = 100, 4.982 h at 200, 8.237 h at 400 and 11.834 h at 800, where Re_D uses the mean inlet velocity and twice the inlet channel height. For turbulent flow it settles near 6 to 7 step heights (6.26 h in Driver & Seegmiller's experiment, 6.28 h in Le, Moin & Kim's DNS).

Which Reynolds number should I use for a backward-facing step?

Say which one you use. Armaly et al. (1983), Erturk (2008) and most laminar benchmarks use Re_D = U·2h_i/ν, with U the mean inlet velocity and h_i the inlet channel height. Others use the step height, or the peak inlet velocity: for an expansion ratio of 2, Re_D = 2 × Re_h(mean velocity) = (4/3) × Barkley's U_max·h/ν. A 10 mm step with 0.1 m/s and ν = 10⁻⁵ m²/s is Re_D = 200 but Re_h = 100.

When is backward-facing step flow laminar?

In Armaly's experiment the separation length rises with Re_D up to about 1,200 (laminar), falls in the transitional range, and levels off above about 6,200 to 6,600 (sources differ). The real flow is already three-dimensional from Re_D ≈ 400, so 2-D laminar simulations agree with the experiment only below that; above it they underpredict the reattachment length.

What does simpleGrading (0.2 5 1) mean in blockMesh?

Each number is an expansion ratio along one block direction: the width of the last cell divided by the width of the first. 0.2 means the cells shrink to one fifth along that edge, 5 means they grow five-fold. Along an 80 mm edge with 80 cells, a ratio of 0.2 gives a first cell of 2.01 mm and a last cell of 0.40 mm; each cell is 0.9798 times the one before.

Why does blockMesh say 'Inconsistent point locations between block pair ... probably due to inconsistent grading'?

Two blocks share a face, but the points along that face are spaced differently on the two sides, so blockMesh cannot glue them. Usually one block is graded one way along the shared edge and its neighbour the other way (0.2 against 5). Give both blocks the same grading along every edge they share; the checker on this page finds the face.

How do I choose the time step for icoFoam?

Keep the Courant number Co = Δt|U|/Δx below 1 in the worst cell: the smallest cell along the flow with the highest velocity. For a laminar channel the peak velocity is 1.5 times the mean. With the 0.40 mm cells next to the step of the common Case 2 mesh and a mean velocity of 0.1 m/s, Δt must be at most 2.68 ms; at 1 m/s, 0.268 ms.

Is this backward-facing step calculator free?

Yes. It runs in your browser, needs no sign-up and sends nothing anywhere.

Related on Skill-Lync

Sources

  • E. Erturk, "Numerical solutions of 2-D steady incompressible flow over a backward-facing step, Part I: High Reynolds number solutions", Computers & Fluids 37 (2008) 633-655: Tables 1 (ER 1.942), 5 (ER 2) and 8 (Re = 800 comparison).
  • B. F. Armaly, F. Durst, J. C. F. Pereira and B. Schönung, "Experimental and theoretical investigation of backward-facing step flow", J. Fluid Mech. 127 (1983) 473-496 (abstract; Re definition via Barkley Table 1 and Ansys verification case VMFL064).
  • D. Barkley, M. G. M. Gomes and R. D. Henderson, "Three-dimensional instability in flow over a backward-facing step", J. Fluid Mech. 473 (2002) 167-190: Table 1 (Reynolds-number definitions), critical Re = 748.
  • G. Biswas, M. Breuer and F. Durst, J. Fluids Eng. 126 (2004) 362-374 (abstract): 2-D behaviour below ReD ≈ 400; Gomes et al., COBEM 2003, paper COB03-1620 (regime ranges; 2-D agreement up to about 400).
  • H. Le, P. Moin and J. Kim, J. Fluid Mech. 330 (1997) 349-374 (abstract); NASA Langley Turbulence Modeling Resource, 2D backward-facing step validation case (Driver & Seegmiller 1985).
  • F. Durst, S. Ray, B. Ünsal and O. A. Bayoumi, J. Fluids Eng. 127 (2005) 1154-1160, channel correlation as restated by B. Ray, F. Durst and S. Ray, arXiv:1707.04947.
  • OpenFOAM-dev source: src/mesh/blockMesh/blockEdges/lineDivide/lineDivide.C, gradingDescriptor/gradingDescriptors.C, blockMesh/blockMeshMerge.C; OpenFOAM user guide (blockMesh grading; v10 cavity tutorial for the Courant number).
  • Ansys CFX Reference Guide, Best Practices Guide for Numerical Accuracy, section 6.3.3 (growth factors below 1.3).
Free counselling

From one OpenFOAM case to a CFD job.

Meshing, picking the right solver for the Reynolds number, and proving a result against a benchmark are what CFD teams test freshers on. A Skill-Lync counsellor can map the course that fits the role you want.

Talk to a CFD course counsellor

Free. We call you back, usually the same day.

By submitting you agree to our Terms and Privacy Policy.

Talk to a career counsellor

Talk to a career counsellor

Free 20-minute call. We'll map the right course to where you want to work.

By submitting you agree to our Terms and Privacy Policy.