A glass vessel holds a litre of cloudy broth. A pump feeds sterile medium in at the top; a side-arm near the rim lets an equal volume spill out. That is the entire apparatus; nobody adjusts anything after the pump is set. Within a day the cloudiness stops changing, and it will stay there — same density, same chemistry — for weeks.
Be clear about what has not happened. The bacteria have not filled the flask and stopped. Nothing has run out; food arrives every second. They have not stopped dividing either. They divide continuously and indefinitely. The number stays flat because cells go out over the weir at precisely the rate new ones are made.
Now the relation that makes this more than a flask. Call the flow divided by vessel volume the dilution rate, D — a setting on the pump. Call the population’s specific growth rate μ. At steady state, μ equals D.
Read it the wrong way round and it sounds trivial; read it correctly and it is startling. The operator does not set the food, the density, or the speed of anything biological. He sets a flow rate, and the cells grow at the rate he chose. An E. coli that doubles every twenty minutes in rich broth can be held for months at a doubling time of eight hours — not sick, just slow, because slow is what the pump asked for. The organism’s own maximum barely enters into it.
Nothing supervises this. If the cells briefly divide faster than they wash out, the extra cells draw the limiting nutrient down, and the shortage slows them until the two rates match. If they fall behind, they are flushed faster than they are replaced, the nutrient they are not consuming accumulates, and the survivors speed up. There is no controller in the vessel. There is an error signal made of hunger.
And here is the part worth sitting with. At steady state, the leftover concentration of the limiting nutrient is set by the dilution rate — and not by how rich the incoming medium is. Make the feed ten times stronger and you get roughly ten times as many cells, sitting in the same residual concentration, exactly as hungry. The reservoir decides how many there are. The pump decides how hungry each one is. That holds while the same nutrient stays limiting; enrich far enough and something else becomes the shortage, and the arithmetic starts over.
The threshold is real and it is a cliff. Push D above the organism’s maximum growth rate and there is no density at which the cells can keep up: they cannot divide any faster, and they are removed faster than that. The culture does not thin to a sparser equilibrium. It washes out, and the vessel goes back to clear medium moving through glass.
Two instruments are routinely confused here, and the difference is exactly the subject. A chemostat fixes the flow and lets the density settle. A turbidostat fixes the density — a sensor watches the turbidity and varies the pump to hold it — so the cells grow at whatever rate they like and the machine chases them. One decides the growth rate and reads off the population. The other decides the population and reads off the growth rate.
The curve underneath, relating growth rate to nutrient concentration, is Monod’s, and it fits well. It is a fit. It was drawn from whole-culture growth data, not derived from the chemistry inside a cell, and it has no mechanistic basis; it goes wrong where a substrate turns inhibitory at high concentrations, and where two nutrients limit at once. Cut the feed to zero and the equation says the biomass should start falling at once. It lags instead, because cells carry internal stores the equation knows nothing about. The device arrived twice in 1950 — Monod published the method, and Novick and Szilard published a machine and the name.
What a chemostat cannot hold steady is the organism. Constant conditions are the most concentrated selection pressure available: every cell is permanently hungry for one specific thing, so any mutant slightly better at scavenging it inherits the vessel. Novick and Szilard saw this in the year they built it — mutant frequencies climbing on schedule, then collapsing, because an unseen better variant had swept the population out from under the one they were counting.
Last week’s coppice stool was also a population kept young by removal from outside, but that removal is total, periodic, and scheduled by a person. This one is partial, continuous, and self-correcting, and nothing chooses which cells leave. No. 55’s garbage collector discards by age; the overflow is perfectly indifferent to it, which is why the mathematics stays this clean. And it is the opposite pole from No. 3’s sealed jar, which keeps all its matter and cycles it. A chemostat keeps none of its matter and holds its shape anyway.
≈
One loop I’m watching
Next: the passive film on stainless steel. It does not resist corrosion by being unreactive — it corrodes instantly, and keeps corroding, forming an oxide skin a few nanometres thick that re-forms in milliseconds wherever you scratch it. The protection exists only because the destruction never stops. Then chloride gets under it in one spot, the repair fails locally, and the whole surface finds out what it was actually relying on.