Nobody lost the recipe for Roman concrete

The story goes like this. The Romans built harbours out of concrete that has stood in seawater for two thousand years. Then Rome fell, the recipe was lost, and Europe did without concrete until Portland cement was patented in 1824. Modern piers crumble in fifty years. We have forgotten something they knew.

The first part is true and the ending is a good ending. The middle is wrong in a way I find more interesting than the myth, because the recipe was never lost. It is in Vitruvius, De architectura, Book II, and it is not hidden in it — he gives you the mix and tells you where to buy the aggregate:

There is also a kind of powder which from natural causes produces astonishing results. It is found in the neighbourhood of Baiae and in the country belonging to the towns round about Mt. Vesuvius. This substance, when mixed with lime and rubble, not only lends strength to buildings of other kinds, but even when piers of it are constructed in the sea, they set hard under water.

That is the whole thing: volcanic ash from the Bay of Naples — pulvis puteolanus, the origin of the word pozzolana — plus lime, plus rubble. Pliny describes the same material in the Natural History. Neither text disappeared. Vitruvius survived the Middle Ages in about ninety manuscripts — the oldest complete one copied around the year 800 and read at Charlemagne’s court — was read and copied by monks who had no particular use for harbour works, and went into print in Rome in the 1480s, at which point any literate builder in Europe could own the instructions. Alberti had already worked through them for his own treatise decades before that.

So the fifteen hundred years of forgetting are not fifteen hundred years of missing information. Whatever went missing, it was not the text.

What I think actually went missing is duller and harder to fix: the demand, and the supply chain. Roman marine concrete is a specific volcanic ash carried by ship from a specific bay, poured in enormous unreinforced masses by a state that could order both. Take away the state, the fleet and the harbours it wanted, and the recipe on the shelf is a recipe for something nobody is building. You can know how to make it and still not make it. That is a different kind of loss from amnesia, and it does not get better when someone finds the manuscript, because the manuscript was never the constraint.

There is a second thing wrong with the story, and it is the one that bothers me more, because it is a comparison between two numbers that measure different things.

“Roman concrete outlasts ours” sets a Roman harbour block against a modern pier. The harbour block is a plain mass of lime, ash and rubble with nothing inside it. The modern pier is a thin shell of high-strength concrete wrapped around a cage of steel, and when it fails at fifty years it usually fails because chloride from the seawater has reached the steel and the rust has expanded and split the concrete off from the inside. The steel is what is rotting. It is also what lets the pier be thin, and span, and hold a road up. Take the steel out and you have to build like a Roman: massive, in compression only, in shapes that never pull against themselves.

And in the one property the comparison implies, Roman concrete loses. It is much weaker in compression than ordinary modern structural concrete. It is durable and it is weak, and those are not the same axis. The Romans built thick because they had to.

The genuinely new part of this story is recent and it is not a rediscovery of anything. Between 2002 and 2009 the ROMACONS project went and drilled cores out of Roman harbour concrete still sitting in the sea. Marie Jackson’s group then found that seawater percolating through the material had grown new minerals inside it — aluminous tobermorite, phillipsite — in the pores, over centuries. A 2023 paper from Admir Masic’s group at MIT argues for a second mechanism: lumps of unslaked lime left in the mix by hot-mixing, which dissolve when a crack lets water in and reprecipitate calcium carbonate across the gap.

Both groups argue for the same surprising thing. The material is not inert. On their account it has been reacting with the sea for two thousand years and getting better at being there, which is the opposite of what every engineering intuition about weathering expects.

Nobody could have read that out of Vitruvius. He wrote down what to mix, and what to mix it with, and where the ash came from — everything a builder needs and nothing about why. The recipe was on the shelf the whole time. What was missing was that anyone should go out to a Roman pier with a core drill and ask it what it had been doing.

Added after publication, from the fact-check. The check came back with one finding that goes at my title rather than at a sentence, and it is the best thing in this piece, so it goes here rather than in a footnote.

In December 2025 a team published the Pompeii Regio IX site — an unfinished Roman building, caught mid-construction — in Nature Communications. It is the first direct archaeological evidence of the process rather than an inference from the cured product, and the process is not the one in the book. The Romans mixed quicklime dry with the pozzolana and added water last. Vitruvius tells you to slake the lime first. On the step that produces the self-healing lime clasts, the surviving text is wrong.

So the honest version of my title is narrower than the title. The text was never lost — that part holds, ninety manuscripts and a printed edition in 1486. But a builder in 1500 holding that book could not have rebuilt Baiae from it, because what he was holding was not the procedure. Which, awkwardly for me, makes the essay’s point harder rather than softer: the thing everyone calls a lost recipe was never a recipe, it was a description written by someone who was not the one mixing.

Three other corrections from the same check, all mine: the manuscript count was “a couple of dozen” in the first version and is about ninety — an error against my own argument, which is rare enough here that I want it on the record. The ROMACONS cores were not all drilled from structures still standing in the sea; several came from harbour works now on land. And the compression figure you will find quoted everywhere for Roman marine concrete, 14.6–19.1 MPa, is deviatoric stress measured under confining pressure — a different kind of number from the cylinder strength of a modern mix. Unconfined, the Roman material tests about 2.5–8.5 MPa against a modern marine spec of 35. I had the direction right and I would have reached for the wrong number to prove it.