What T-junction droplet correlations can and cannot predict

If you generate droplets at a T-junction, you have met this equation. It is one of the most useful results in the field — a straight line, two flow rates, a droplet length. It is also one of the most misused, because it is usually quoted without the two things that make it meaningful: the regime it describes, and the fact that α is not a constant of nature.

In the squeezing regime, the dispersed phase grows into the junction until it blocks most of the main channel. Once it blocks, pressure builds upstream in the continuous phase and squeezes the neck shut. Break-up is therefore driven by confinement, not by shear — which is why the resulting length depends on a flow-rate ratio and the channel width, and why viscosity barely appears.

That is a specific, physically grounded picture, and it holds where the picture holds: low capillary number, a dispersed stream that actually confines, and a geometry in the range the experiments covered.

Papers quote α as 1, or 1.1, or 1.5, and all of them are citing real data. The honest statement is that α is a fitted coefficient depending on junction geometry and viscosity ratio. Nekouei and Vanapalli (Physics of Fluids 29, 032007, 2017) refitted the Garstecki data and reported a band:

So a tool that reports L = 212.4 µm from this equation is not being precise. It is being misleading. Two significant figures is what the underlying fit supports, and two is what it should print.

Outside its regime the formula does not degrade gracefully; it simply becomes a different number with no physical claim behind it. And even inside its regime, there are things it was never about:

• Dripping and jetting. Raise Ca and break-up becomes shear-driven, then a thread forms and breaks downstream. Different mechanism, different scaling. The squeezing line does not bend towards it — it is simply no longer the relevant physics.
• Surfactant dynamics. The correlation assumes an interfacial tension. Real surfactants adsorb on a timescale comparable to droplet formation, so the effective γ at pinch-off is not the equilibrium value you measured. A common source of "the model says 200, I measure 240".
• Coalescence and stability. What happens after formation — whether droplets merge in the outlet, stick, or survive incubation — is outside the scope entirely.
• Wetting. If the dispersed phase wets the wall, the confinement picture breaks down and you may get no droplets at all. The equation has no term for this and will cheerfully give you a length.
• Polydispersity. It predicts a mean. The spread comes from your pressure stability and channel tolerance, neither of which appears.

It will tell you, in seconds and for free, roughly where to put your flow rates and roughly what channel width you need for the droplet size you want. That is genuinely valuable: it turns a week of pilot runs into an afternoon.

Then treat the number as a starting point with a stated uncertainty, verify on the chip, and — this is the part that pays — check whether the operating point you chose sits comfortably inside the squeezing band or perched on its edge. A design that works at Ca = 0.008 and fails at Ca = 0.015 is a design that will fail when somebody changes the oil.

The equation is not a droplet calculator. It is a compact summary of one break-up mechanism, fitted to a finite set of experiments, with a coefficient that has a published spread of about ±14%. Used inside those limits it is excellent. Used outside them it is a random number generator with good manners.