Gilliland Correlation (Actual Stages)

Fenske gives the fewest stages, Underwood the least reflux, and neither is buildable. Gilliland's 1940 contribution was to fill in the middle empirically. He ran a set of rigorous stage-by-stage calculations across a range of columns, plotted $(N - N_{min})/(N + 1)$ against $(R - R_{min})/(R + 1)$, and found the points fell close enough to one curve to be useful. That curve, with Fenske and Underwood either side of it, is the Fenske-Underwood-Gilliland shortcut method, and it is still how a column gets sized on the back of an envelope before anyone opens a simulator.

Gilliland published a graph, not an equation. The expression evaluated on this page is the analytic fit published by Molokanov and co-workers in 1972, which reproduces the drawn curve closely enough that the difference is invisible against the scatter of the original points. It is worth being clear that this is a correlation of a correlation: a curve fitted to a curve drawn through a cloud.

The two ends behave exactly as they must, which is the best argument for trusting the middle. At $X \to 0$ — running at minimum reflux — $Y \to 1$ and the stage count goes to infinity. At $X \to 1$ — infinite reflux — $Y \to 0$ and $N \to N_{min}$. Between them sits the familiar rule of thumb that a real column needs about twice the minimum stages, which is not a separate piece of folklore at all: it is this correlation evaluated at the reflux ratios people actually choose.

The honest caveat is the scatter. Individual points in Gilliland's original data depart from the fitted curve by roughly ten per cent, and rather more at the extremes of $X$. That is fine for sizing a shell, costing a project, or deciding between two process routes. It is not fine for guaranteeing a product purity, and it should never be the last calculation done on a column that is going to be built. A rigorous equilibrium-stage or rate-based simulation is what settles a final design; Gilliland is what tells you roughly what to simulate.