The Colebrook–White equation is implicit in the friction factor:
1/√f = −2 log₁₀ ( ε/3.7D + 2.51/(Re√f) )f appears on both sides, which for most of the twentieth century was a genuine
problem. You either read a Moody chart or you iterated by hand, and neither is
something you want to do forty times while sizing a header.
So a family of explicit approximations exists. The best known is Swamee–Jain:
f = 0.25 / [ log₁₀( ε/3.7D + 5.74/Re^0.9 ) ]²It is a good fit. Within its stated range — 10⁻⁶ ≤ ε/D ≤ 10⁻² and
5000 ≤ Re ≤ 10⁸ — it stays within about 1 % of Colebrook, which is far
inside the uncertainty on your roughness value.
I still stopped using it on its own, and the reason is not accuracy.
Iteration is now free
x = −2 log₁₀( ε/3.7D + 5.74/Re^0.9 ) // Swamee–Jain as a seed
repeat:
x = −2 log₁₀( ε/3.7D + 2.51·x/Re )
f = 1/x²Substituting x = 1/√f turns the implicit equation into a fixed-point iteration
that converges monotonically and fast. From the Swamee–Jain seed it is
essentially converged in three or four passes. I run sixty because sixty
iterations of a logarithm is not a measurable cost on any machine built this
century, and it means I never have to think about convergence again.
That is the actual argument. Not "the explicit fit is wrong" — it is that the explicit fit has a domain of validity you have to carry in your head, and iterating removes it.
Where the fits are weakest
The explicit correlations are least good in the fully rough turbulent corner — high relative roughness, high Reynolds number — which is exactly where old carbon steel pipe carrying water at design flow tends to live. It is not a corner case in real plant. It is Tuesday.
The error is still small. But it is small and it is largest precisely where I am least likely to be checking, which is the worst combination an error term can have.
The input that actually dominates
None of this is where the real uncertainty is.
Absolute roughness for commercial steel is quoted at 0.045 mm. For the same pipe after ten years of service it may be 0.5 mm or worse, depending entirely on what has been flowing through it. That is an order of magnitude, and it swamps every numerical consideration on this page by a wide margin.
Which gives the rule that matters far more than the choice of correlation:
Design on the aged roughness, not the new one. A system that only meets duty when the pipe is clean is a system that meets duty once.
I have watched people argue about the fourth significant figure of a friction factor while carrying a roughness value that was a factor of ten optimistic. The arithmetic was impeccable.
Laminar, and the gap
Below Re ≈ 2300, none of this applies and the answer is exact:
f = 64 / ReBetween 2300 and about 4000 there is no reliable friction factor at all. The flow is transitional and genuinely unpredictable — it can behave laminar or turbulent depending on inlet conditions and disturbances upstream.
If a duty point lands in there, the correct response is not to pick a correlation and press on. It is to change the pipe size so it does not. A transitional line is one whose pressure drop you cannot predict to better than about ±20 %, and predicting pressure drop is the entire reason you opened the calculator.
What I put in the tool
The pipe flow calculator on the bench does the following, and shows all of it:
- Laminar below 2300, exactly.
- Swamee–Jain as a seed, then Colebrook iterated to convergence above it.
- An explicit warning in the transitional band rather than a number pretending to be trustworthy.
- Velocity checks at both ends — below about 0.6 m/s in a liquid you stop sweeping air and solids along; above about 3.5 m/s you start buying erosion, noise and water hammer.
That last one is worth more than the friction factor. Pressure drop goes with the square of velocity, so most of the time the answer to "how do I reduce the pressure drop" is not a better correlation. It is a larger pipe.
