Artificial Wasteland / a layer

The Plate That Does Nothing

Everyone who has ever read the back of a bottle of rectified spirit knows the number. You cannot distil ethanol past about 95 percent. Boil the wash, catch the vapour, condense it, boil it again, forever if you like: somewhere just short of pure, the process stops paying you. Column height stops mattering. The stuff simply refuses.

The explanation usually offered is that ethanol and water bind. That water molecules cling to alcohol molecules by hydrogen bonds and will not let go, so the last few percent cannot be boiled away. It is a satisfying picture and it is not what is happening. Nothing is holding anything. What is happening is arithmetic, and you can watch it happen below.

What one plate does

A still is a machine for exploiting one fact: when a mixture boils, the vapour above it is not the same mixture. It is richer in whatever is easier to boil. Condense that vapour and you have a liquid better than the one you started with. Do it again on the result and you gain again. A fractionating column is that step stacked: each plate is one boil-and-catch, and a tall column is many.

So the whole trade rests on one quantity: for a liquid of composition x, what is the vapour composition y? Plot y against x and you have the entire behaviour of the mixture in one curve. The diagonal, where y = x, is the line of no progress: on it, the vapour is exactly the liquid, and boiling gains you nothing at all.

Where the staircase stops

Drag the plates slider at one atmosphere and watch what the steps do. The first few are enormous. Then they shrink, and keep shrinking, and the staircase crowds into the corner where the curve meets the diagonal and stays there. It never strikes a barrier. It runs out of room, which is what a wall actually looks like from the inside.

That meeting point is the azeotrope, and it is not a compound, not a bond and not a substance. It is a crossing. Below it the vapour is richer in ethanol than the liquid, so distilling enriches. Above it the vapour is richer in water, so distilling would take you backwards. Exactly at it the two are identical, and the separation factor is exactly one. That is the sentence that does all the work: a plate at the azeotrope performs a perfect boil, a perfect condensation, and hands you back precisely what it was given. It is not that the plate is defeated. It is that the plate does nothing.

Now take the roof off

Here is the part the usual telling leaves out, and it is the reason the binding story is worth arguing with rather than just correcting. If the last few percent of water were held by a bond, the wall would be a property of the two substances and would sit where it sits. It is not, and it does not. The crossing happens where two opposing tendencies cancel, and one of them is more sensitive to pressure than the other. So change the pressure and the cancellation point moves.

Pull the pressure slider down. The wall does not weaken; it walks, toward pure ethanol, and then it reaches the end of the chart and there is no crossing left. Below that pressure the curve lies entirely above the diagonal, every plate gains something, and an ordinary column can go as far as you have patience to build. That is why vacuum rectification is a real unit operation and not a curiosity.

But notice what just happened, because it is the honest half of this page. The wall at one atmosphere is something you can measure and people have, for a century. The pressure at which the wall leaves is somewhere nobody has taken the measurement, and the number you just watched the chart produce came out of a model. So the question worth asking is not "where does it go?" but "how much would the answer change if I had made different defensible choices along the way?"

How much of that is measured, and how much is a model

The wall at one atmosphere is a measurement. It has been measured for a century and the instrument above reproduces it. The pressure at which the wall leaves is a different kind of claim, and this page will not pretend otherwise: it is an extrapolation, made by fitting a model in the region where the data are thick and then asking it about a region where they are thin.

So rather than quote one number, the page runs every defensible version of the calculation and shows the spread. Three activity-coefficient models, crossed with the choice of vapour-pressure correlation for each component, crossed with the two slightly different published values for the azeotrope itself. Each version is pinned to the measured azeotrope exactly and to nothing else, then asked about 55 rows of a published table it was never shown. A version that cannot reproduce those rows does not get a vote, and the rule for that is stated in the panel below along with which half of it was decided before the answers were visible.

The useful part

Industry has known the useful half of this for as long as it has needed dry ethanol, which is why the alternatives exist alongside vacuum work: dragging in a third component, or running the vapour through a molecular sieve that swallows water and not alcohol. None of it is exotic. It is all the same observation used deliberately: the wall is a feature of the conditions, so change the conditions.

It is worth saying what this page is not claiming, because walls of this kind are not all alike. Some really are pinned. Bode's sensitivity integral, which another layer here puts under your hands, fixes the total area of a control loop's error no matter how you tune it: you can move the error around in frequency and you can never remove it, and no dial anywhere escapes that. The azeotrope looks like that from inside a column at one atmosphere and is not. Telling the two cases apart is not a matter of intuition. It is a matter of asking what the wall is a function of, and then going and moving that thing.

What is worth keeping is smaller and more general than ethanol. A limit that presents itself as a property of the material may turn out to be a property of the situation, and the way you find out is to ask what the limit depends on. Here the honest answer was available the whole time, sitting in the shape of one curve against one diagonal: not a barrier, a crossing, and a crossing can be moved.