Boundary layers need a mesh of their own

Resolving the thin region against a wall, and what happens if you do not.

Boundary layers need a mesh of their own because the flow right next to a solid wall changes far more quickly, over a far smaller distance, than the flow just a short way further out, and a mesh built for the open flow is almost always too coarse to capture what is happening in that thin boundary layer region at all.

From standing still to full speed in a sliver

Fluid touching a solid surface is held almost still by friction with it, while fluid a short distance away can be moving at close to the full speed of the flow around it. Between those two states sits the boundary layer, usually far thinner than the flow field around it, across which velocity climbs from nearly zero to nearly full speed. Capturing that climb needs a stack of cells across a distance that, everywhere else in the same mesh, a single cell would happily cover.

A mesh built near the wall at the same scale as further out gives the solver too little resolution inside the boundary layer to represent that climb, and everything that depends on it, drag, heat transfer at the wall, whether the flow stays attached or separates, inherits an error rooted in a part of the mesh most people never inspect on its own. The fix is to mesh the region against every wall differently, stacking thin layers of cells that grow gradually thicker moving away from the surface until they blend into the coarser mesh covering the rest of the domain.

Shaving round the jawline

Shaving demands slow, closely overlapping strokes around the jawline and the contours of the nose, where the skin's angle to the blade changes quickly. Clear those areas the way the flat cheek gets cleared, in one or two broad, fast passes, and the result is patchy in the tight spots, even though the same number of strokes would have covered a larger flat area perfectly well. Giving the cheek the jawline's careful attention wastes time without improving anything, while skipping it around the jaw leaves the job visibly undone.

A boundary-layer mesh is built on that distinction. The region against the wall gets the careful, closely stacked treatment of a jawline, and the open flow further out gets the broad strokes a cheek can take. Meshing everywhere as finely as the jawline needs would take enormously longer for no benefit, and meshing everywhere as coarsely as the cheek allows would finish quickly and miss the one area the exercise was about.

A first cell thinner than a hair

The first layer, the one touching the wall, is usually the most carefully controlled dimension in the entire mesh, sized from how fast the flow is moving and how it is expected to behave close to the surface. Too thick, and the steep part of the velocity climb is hidden inside a single oversized cell. For air flowing over a car at motorway speed, a mesh meant to resolve the flow right down to the wall typically needs that first cell to be around a hundredth of a millimetre tall, thinner than a human hair, while cells out in the open flow may be a centimetre across, a thousand times larger.

How thin it must be also depends on whether the boundary layer is laminar or turbulent, since a turbulent one packs its steepest change into an even thinner sliver against the wall. A mesh built confidently for a smooth, laminar case can turn out badly under-resolved the moment the same geometry is run at a higher, more turbulent speed.

Zooming in before signing off

The habit this creates is treating wall-adjacent regions as a separate meshing task, checked on their own terms and kept out of any single uniform refinement pass. A confident drag number or wall heat transfer value is only as trustworthy as the layer of cells sitting directly against that wall, and a mesh that spends its cells generously on open space and stingily on that sliver has allocated its resolution backwards, a mismatch the overall cell count would never reveal. Scrolling across the open flow field and pronouncing the mesh satisfactory, without zooming in far enough to see the individual layers stacked against the wall, skips the one check that mattered most.

A badly sized first layer can also do more than blur the answer. A first cell inconsistent with how the solver expects the near-wall flow to behave can make the whole simulation harder to converge, turning a meshing shortcut into a stability problem that shows up as stalled residuals instead of an obviously wrong number.