Every external flow problem — flow over a flat plate, flow over a cylinder, airflow across a heat exchanger fin — is really a boundary layer problem underneath. You can memorize the Nusselt number correlations for all of those geometries without ever understanding why they look the way they do. But if you understand boundary layers, the correlations stop being a lookup table and start being something you can reason about from scratch.
Here's the thing that's easy to miss: there are actually three distinct boundary layers that can coexist at the same surface, each growing independently but all related to each other through the dimensionless groups you already know from the transport analogy.
What a boundary layer actually is
When fluid flows over a solid surface, two things happen at the wall simultaneously:
- No-slip condition: the fluid velocity goes to zero at the wall ($u = 0$ at $y = 0$), even though the free-stream velocity far away is $u_\infty$. The velocity mismatch between wall and free stream has to be accommodated somewhere — that region of transition is the velocity boundary layer.
- Wall temperature/concentration: if the wall is hotter than the fluid, or has a different surface concentration, there's a similar mismatch in $T$ or $C_A$ that gets resolved in a thermal boundary layer or concentration boundary layer.
The thickness $\delta$ (velocity), $\delta_t$ (thermal), and $\delta_c$ (concentration) are defined as the distance from the wall to where the profile has "essentially" recovered to its free-stream value — usually taken as 99% recovery in practice.
Both boundary layers start from zero at the leading edge and grow downstream. When Pr > 1, thermal diffusivity is slower than momentum diffusivity, so the thermal boundary layer stays thinner than the velocity boundary layer at every x.
Why boundary layers grow in the downstream direction
The boundary layer starts at zero thickness at the leading edge ($x = 0$) and grows as you move downstream. Here's the physical reason: the further downstream you go, the longer the fluid has been "in contact" with the wall's influence. Near the leading edge, only the fluid immediately adjacent to the wall knows about the no-slip condition. Further downstream, viscous diffusion has had more time to carry that information outward into the free stream.
For laminar flow over a flat plate, the velocity boundary layer thickness scales as:
Two things to notice: $\delta$ grows like $\sqrt{x}$ (not linearly), and it's thinner for higher Reynolds numbers (faster flow or larger $u_\infty$ means inertia wins over viscosity, keeping the layer thinner and more confined to the wall).
Why h decays in the laminar region
The local convection coefficient $h_x$ is defined by:
In plain English: $h$ measures how steep the temperature gradient is right at the wall surface, normalized by the overall temperature difference driving the problem. The steeper that gradient, the better heat is being conducted out of (or into) the surface by the fluid right there at the wall.
As the thermal boundary layer grows thicker downstream, the temperature profile has to spread over a larger distance — so the gradient at $y = 0$ gets shallower. Thicker boundary layer = shallower wall gradient = lower $h$. That's why $h_x \sim x^{-1/2}$ in the laminar region: $h$ drops as you move downstream, because the boundary layer has grown and the wall gradient has flattened.
What transition does to h (and why turbulence is so effective)
Transition from laminar to turbulent flow happens when the local Reynolds number reaches approximately $Re_{x,c} \approx 5 \times 10^5$ for a flat plate (though the real range is $10^5$ to $3 \times 10^6$ depending on surface roughness and free-stream turbulence). At the transition point, $h$ spikes upward dramatically, even though the boundary layer simultaneously gets thicker. That seems contradictory — here's why it isn't:
Computed from the standard flat-plate correlations. The laminar h decays as $x^{-1/2}$; at transition it jumps by roughly 4.6×, then decays far more slowly as $x^{-1/5}$. The boundary layer is still thickening the whole time — turbulent mixing simply overwhelms the growing diffusion distance.
In laminar flow, the only mechanism for moving heat perpendicular to the wall (in the $y$-direction) is molecular diffusion — the $\alpha \partial^2 T / \partial y^2$ term. In turbulent flow, eddies and swirling motions actively mix fluid across the boundary layer, bringing cold free-stream fluid down close to the hot wall continuously. This turbulent mixing is orders of magnitude more effective than molecular diffusion. The wall temperature gradient at $y=0$ actually gets steeper in turbulence despite the thicker overall layer, because the eddies keep the fluid near the wall much closer to $T_\infty$, maintaining a sharp gradient in a very thin viscous sublayer right at the surface.
The exponent jumping from 1/2 to 4/5 on Re captures exactly this: turbulence makes $h$ much more sensitive to flow velocity, and it jumps sharply at transition before decaying again (now more slowly) in the turbulent region.
The three boundary layer thicknesses and their ratios
For laminar flow over a flat plate, the three boundary layers scale as:
| Condition | Result | Physical meaning |
|---|---|---|
| $Pr = 1$ | $\delta_t = \delta$ | Heat and momentum diffuse at equal rates — thermal and velocity BL same thickness |
| $Pr > 1$ (most liquids) | $\delta_t < \delta$ | Momentum diffuses faster — velocity BL outpaces thermal |
| $Pr < 1$ (liquid metals) | $\delta_t > \delta$ | Heat diffuses faster — thermal BL is actually thicker than velocity BL |
| $Sc > 1$ (most liquids) | $\delta_c < \delta$ | Mass diffusion slowest — concentration BL thinnest of the three |
For air at standard conditions, $Pr \approx 0.71$ (slightly less than 1), so the thermal boundary layer is very slightly thicker than the velocity boundary layer. For water, $Pr \approx 7$, so the thermal boundary layer is noticeably thinner. For liquid metals, $Pr \approx 0.001$–$0.01$, meaning the thermal boundary layer can be 5–10 times thicker than the velocity boundary layer — a completely different heat transfer picture.
Local vs. average h — when to use which
This is a source of consistent confusion: there are two convection coefficients, and using the wrong one will cost you the problem.
Local $h_x$: the coefficient at a specific x-location. Use this when you need the heat flux at a particular point on the surface, or when the surface has a known, varying temperature profile. It comes from the local Nusselt number:
Average $\bar{h}$: the coefficient averaged over a length $L$, weighted by area. Use this when you need the total heat transfer rate from a surface of length $L$ with a uniform surface temperature:
For all-laminar flow over a flat plate, integrating $h_x \sim x^{-1/2}$ gives $\bar{h} = 2 h_x|_{x=L}$ — the average is exactly twice the local value at the trailing edge. For mixed laminar-turbulent flow, the average accounts for both regions:
The $-871$ correction subtracts out the laminar contribution that the turbulent correlation would otherwise overcount, since turbulence doesn't start at $x = 0$.
The workflow for any external flow problem
- Calculate $Re_L$ (or $Re_x$ if you need a local value) using the free-stream velocity and fluid properties at the film temperature $T_f = (T_s + T_\infty)/2$.
- Check $Re_L$ against $Re_{x,c} \approx 5 \times 10^5$ — is the flow all-laminar, mixed, or all-turbulent? Pick the right correlation.
- Compute $Nu$ using the appropriate correlation (local or average, laminar or turbulent).
- Extract $h$ from $Nu = hL/k_f$.
- Compute $\dot{Q} = \bar{h} A_s (T_s - T_\infty)$ for average, or $q'' = h_x(T_s - T_\infty)$ for local flux.
Always verify your assumptions: that properties were evaluated at film temperature, that the Re range falls within the correlation's stated validity, and that you used local vs. average consistently throughout.