The Concrete That Heals Itself: Why Roman Breakwaters Still Stand in Seawater After Two Thousand Years – and Why Materials Science Has Been Arguing About the Reason for Three Years

The Hook: A Structure That Should No Longer Exist

When a civil engineer designs a harbor breakwater today, they design to a service life. For reinforced-concrete coastal structures it falls, depending on the code and the exposure class, somewhere between fifty and a hundred years. That is not a pessimistic figure but a realistic one: seawater is one of the most aggressive environments reinforced concrete can face. Chlorides migrate through the concrete cover, reach the reinforcement, destroy its passive layer, the steel rusts, the rust occupies more volume than the metal it came from, the concrete spalls, and the process accelerates itself. Anyone walking along a harbor in northern Europe and seeing concrete edges with rust-brown bars protruding is watching that mechanism in real time.

Against that background, the Bay of Naples is an irritation. At Baiae, at Portus Cosanus near Orbetello, at Caesarea Maritima on the coast of present-day Israel, and at dozens of other points around the Mediterranean, there are moles, breakwaters, and piers that were built roughly two thousand years ago and whose cores are still intact. Not as picturesque ruins, but as material: you can drill cores out of them, take them into a laboratory, load them, and shoot X-rays through them. That is precisely what the international ROMACONS project did between 2002 and 2009, publishing its results in 2014 in the standard reference Building for Eternity.

The riddle can therefore be stated precisely. A material produced without steel, without cement clinker, without a mixer, and without any testing standard outlasts — in the most aggressive conceivable environment — a material we produce with four centuries of scientific history behind us. The obvious explanation, the Romans knew something we have forgotten, is popular, partly correct, and nevertheless misleading in its usual form. This article walks through the chemistry, through the ancient written sources, through three competing scientific explanations of the longevity, through an honest account of what Roman concrete cannot do, and finally through the question of what the cement industry of 2026 can actually learn from it.

The timing is no accident. On July 8, 2026, Science Advances published a paper that explicitly contradicts the best-known explanation of recent years. The question is open again.

Part 1: The Core Concept – What Opus Caementicium Actually Is

Not poured, but layered

The first and most important misconception concerns consistency. Modern concrete is a pumpable, pourable suspension: you place it in formwork, consolidate it with a vibrator, and it fills every void. Roman concrete was not that. Opus caementicium consisted of a comparatively stiff, earth-moist mortar plus caementa — fist-sized to head-sized pieces of rubble stone, broken tile, tuff, and lava.

The construction method followed accordingly. Two wall facings of rubble or brick (opus incertum, opus reticulatum, later opus testaceum) served simultaneously as formwork and as the visible surface, and the space between them was filled in layers: a course of mortar, a course of stones, tamped down, then the next. For underwater work the Romans used wooden caissons, in some cases assembled on site from planks and posts; the mortar was lowered in baskets and rammed together with the caementa inside the box.

This distinction is not pedantic; it is causal. A stiff mixture tamped in layers has a different water content, a different pore structure, and a different distribution of binder particles than a poured suspension — and precisely those differences sit at the center of the modern explanatory attempts. The 2026 Berkeley study estimates the water-to-binder ratio of the sample it examined at roughly 0.4 to 0.45; that is remarkably low and lies in the range we aim for in durable concrete today.

The chemistry: air lime versus hydraulic lime

To understand what distinguishes Roman concrete you need one chemical cycle and one extension to it.

The cycle is the lime cycle. Limestone, chemically calcium carbonate (CaCO₃), is heated in a kiln to roughly 850–900 °C. Carbon dioxide escapes, and what remains is calcium oxide (CaO) — quicklime, an aggressive, strongly hygroscopic substance. Add water and it reacts violently, releasing a great deal of heat, to form calcium hydroxide (Ca(OH)₂), or slaked lime. Slaked lime hardens in air by reabsorbing carbon dioxide from the atmosphere and returning to calcium carbonate. The circle closes.

This process — air lime — has worked for millennia and has two serious drawbacks. It is slow, because it depends on carbon dioxide diffusing from the air into the depth of the masonry. And it does not work underwater at all, because no atmospheric CO₂ arrives there.

The extension is the pozzolanic reaction. Add to slaked lime a substance containing reactive, amorphous silica and alumina — volcanic ash, say, or calcined clay, or brick dust — and those oxides react with the calcium hydroxide and the mixing water to form calcium-aluminum-silicate hydrates, abbreviated C-A-S-H in cement chemistry. The decisive point: this reaction needs no carbon dioxide. It needs only water. Such a binder therefore hardens underwater as well — it is hydraulic.

That completes the basic principle of Roman marine concrete. It is not a lost secret technique but a reaction described in every textbook of binder chemistry and used daily by modern concrete technology: fly ash, ground granulated blast-furnace slag, and calcined clay are pozzolans, and their use is currently the single most important lever for reducing the CO₂ footprint of cement.

The recipe in the sources

Roman concrete is one of the rare cases where we have the technology transmitted not only archaeologically but also in writing — and by two independent authors.

Vitruvius, De architectura, around 25 BC, treats the material in the sixth chapter of Book II. He describes a powder from the region of Puteoli (modern Pozzuoli) — pulvis puteolanus — which, combined with lime and rubble, not only strengthens masonry but also hardens in the sea. For masonry mortar he gives a ratio of one part lime to three parts pozzolan; for work underwater, one to two. In Book V, in the chapter on harbor works, he describes the caisson technique.

Pliny the Elder, Naturalis historia, around AD 77, records in Book 35 (§ 166) the observation that modern literature keeps quoting: that the dust from the area between Baiae and Vesuvius, brought into contact with seawater and submerged, becomes a single stone mass, impregnable to the waves and stronger every day.

That last clause is why the sentence appears in twenty-first-century research papers. Pliny is not offering poetic exaggeration but a physical claim: that the material gains strength through the action of seawater. To a modern concrete technologist that is initially absurd, because in their world chemical reactions between cement paste and seawater occur exclusively as damage mechanisms. Fifteen years of research have shown that Pliny, within the limits of his material, was right.

That this technology did not remain a local phenomenon of the Bay of Naples is now analytically documented. Dilaria and colleagues showed in 2023 in Scientific Reports that pyroclastic aggregates from the Gulf of Naples appear in the mortar of the Roman theatre of Aquileia in northern Italy, dated to between the mid-first century BC and the mid-first century AD. That is a trade route of roughly seven hundred kilometers for a bulk material — and it shows that Roman builders deployed it deliberately where moisture was a problem, in this case a low-lying deltaic plain near a lagoon.

Part 2: Three Explanations for Two Thousand Years

Up to this point the story is uncontested. From here on there is a genuine, currently live scientific controversy — and it is more instructive than any of the three answers taken alone.

Explanation A: Seawater as a collaborator (Jackson et al., 2017)

The group around Marie D. Jackson studied drill cores from Roman marine structures using synchrotron X-ray microdiffraction at the Advanced Light Source of Lawrence Berkeley National Laboratory, together with electron microscopy and Raman spectroscopy. The results appeared in 2017 in American Mineralogist (volume 102, issue 7, pages 1435–1450).

The finding: minerals are growing in the mortar that formed after the original hardening was complete. Two matter most. Al-tobermorite is a platy-crystallizing calcium-silicate hydrate in which aluminum partly substitutes for silicon. Phillipsite is a zeolite, an aluminosilicate with an open framework structure.

The mechanism Jackson and colleagues describe is remarkable because it inverts the usual direction of thought: seawater percolates through the mortar, attacks constituents of the volcanic ash — feldspar and pumice in particular — and from the resulting highly alkaline pore fluids new mineral phases crystallize at ambient temperature, around 20 °C. They grow in cracks, pores, and leached zones. The platy Al-tobermorite crystals interlock and bridge crack flanks, which increases resistance to brittle failure.

Two points deserve emphasis. First, Al-tobermorite is difficult to synthesize in the laboratory and is usually made only at elevated temperatures; that it forms here at seawater temperature is the real materials-science discovery. Second, in modern concrete the reaction between alkalis and reactive silica — the alkali-silica reaction, ASR — is a feared damage mechanism responsible for substantial structural deterioration worldwide, because the reaction product swells and bursts the cement paste apart. In Roman marine concrete a chemically related interaction runs and acts beneficially. The difference lies not in the chemistry as such but in the pore structure, in the available space, and in the fact that there is no steel nearby that needs protecting.

Explanation B: Hot mixing and lime clasts (Seymour et al., 2023)

On January 6, 2023, Linda M. Seymour, Janille Maragh, Paolo Sabatini, Michel Di Tommaso, James C. Weaver, and Admir Masic published a paper in Science Advances (volume 9, issue 1, article eadd1602) that drew attention far beyond the field — and that placed a different feature of the material at the center.

Roman mortars contain millimeter-scale, pale, lime-rich inclusions known as lime clasts. For decades they were taken as evidence of sloppy work: lime that had not been properly slaked, poorly mixed, a quality defect. The MIT group around Admir Masic reinterpreted them.

Their reconstruction: the Romans did not work with ready-slaked lime putty but added quicklime directly to the mix — hot mixing. On contact with the mixing water this releases large amounts of heat, drives the temperature of the mix up, and according to Masic has a twofold benefit: it accelerates reactions that would otherwise not happen, or happen only very slowly, and it shortens curing time, which saves money and schedule on site immediately. As a by-product, lime clasts remain with high specific surface area and, as the authors put it, a characteristically brittle nanoparticulate architecture.

And that brittleness becomes the feature rather than the flaw. If a crack forms in the member, it propagates preferentially through the brittle clasts. If water then enters, it dissolves calcium there — the clasts are a reactive calcium source not available in ordinary cement paste — and from the calcium-saturated solution calcium carbonate crystallizes, filling the crack from within. Alternatively, delayed pozzolanic reactions can occur where reactive ash lies nearby.

What is striking is that the group reproduced the hypothesis in the laboratory. They made specimens with and without quicklime hot mixing, cracked them deliberately, and ran water through the cracks. In the hot-mixed specimens the cracks had sealed completely within two weeks and the flow stopped. In the control specimens without lime clasts the water kept flowing. That is a cleaner argument than any observation on ancient material, because it detaches the mechanism from its historical context and tests it as a material property.

Explanation C: Carbonation instead of intent (Zhu et al., 2026)

On July 8, 2026, also in Science Advances (volume 12, issue 28, article eaeb0754), a paper appeared by Xiaohong Zhu, Sejung Rosie Chae, Stuart McElhany, Chengyao Liang, Qi Zheng, Jiaqi Li, Veronica Fondi, Sergio Del Ferro, Ascanio Modena Altieri, Harrison P. Lisabeth, Arun Bhattacharjee, Bruce W. Fouke, Hans-Rudolf Wenk, and Paulo J. M. Monteiro. Monteiro (UC Berkeley) and Zhu (Beijing University of Technology) led the work; contributors included Lawrence Berkeley National Laboratory, the University of Illinois Urbana-Champaign, the University of Michigan, and the Istituto Autonomo Villa Adriana e Villa d'Este.

The subject was concrete from a latrine at Hadrian's Villa in Tivoli, second century AD — a terrestrial structure, not a marine one. The aggregate is black volcanic lava containing leucite, analcime, and ferrian diopside. The methodological arsenal is impressive: micro-computed tomography at 600 nm resolution, nano-CT at about 43 nm, backscattered electron microscopy with EDX, Raman spectroscopy, X-ray absorption near-edge structure (XANES), transmission electron microscopy, and powder X-ray diffraction with a cobalt anode.

The central finding shifts the weight: not the pozzolanic reaction but calcite cementation in pores and fractures is, in this sample, the volumetrically dominant binding mineral phase. The authors describe networks of radiaxial fibrous calcite growing outward from the rims of lime clasts and filling microcracks and pores as they go. Among the measurements: about 24.5 % porosity in the calcite network, roughly 15.6 % internal porosity in the volcanic aggregates, and unburnt limestone particles averaging 25.5 ± 6.9 µm in diameter. The mechanism: unreacted calcium oxide reacts over centuries with atmospheric carbon dioxide and moisture to form calcite, and that calcite densifies the fabric, improves load transfer, and reduces water ingress.

The interesting part is what the authors explicitly do not claim. They write that their microstructures superficially resemble those in concretes interpreted as "hot-mixed," but that their observations provide no direct evidence for deliberate hot mixing. Their explanation is more conservative: incomplete burning of the limestone, partial slaking, and long-term in situ carbonation. Not craft precision, but craft imprecision plus two millennia of time.

What to take from this

Three observations about that table.

First, the three explanations are not mutually exclusive. A concerns marine concrete, C a building on land; the relevant mechanisms can be differently weighted in different places. And A, B, and C all amount to the same thing: in this material, something continues to happen after hardening that fills voids.

Second, the real point of contention is not chemical but historical: was it intentional? The 2023 paper reads the lime clasts as the signature of a skilled process, the 2026 paper as the signature of an imperfect one. That is a question mineralogy alone cannot settle, because the same microstructure can arise from deliberate hot mixing and from badly burned lime.

Third, and this is the methodological core: I am of the opinion that the intent question could only be decided by a combination of findings that does not currently exist — systematic sampling across many structures, regions, and centuries, combined with an assessment of the written sources on working practice. As long as that breadth is missing, the same caution applies to both sides: a handful of samples can demonstrate a mechanism, but it cannot reconstruct a building industry. For practical use the intent question is secondary anyway — the 2023 crack-healing experiment works regardless of whether the Romans knew why.

Part 3: The Honest Balance Sheet – What Roman Concrete Cannot Do

At this point the popular narrative tips over, and it is important to let it tip. "Roman concrete was better" is, in its usual abbreviated form, false.

It is considerably weaker. MacFarlane, Vanorio, and Monteiro tested Roman marine concrete triaxially in 2021 in Construction and Building Materials (volume 272, article 121812). The specimens showed abrupt stress drops at deviatoric stresses between 14.6 and 19.1 MPa, at axial strains of more than 9 %, with pronounced creep behavior. An ordinary structural concrete today sits at 30 to 50 MPa, high-performance concrete well above that. What is remarkable about the Roman specimens is not their strength but their ductile, creep-capable deformation behavior — a material that yields instead of snapping.

It cannot take tension. Concrete is always strong in compression and weak in tension; that is why we have reinforced it with steel since the late nineteenth century. The Romans had no such option, which produces a paradoxical double effect. Without steel there is no reinforcement corrosion — the single most common damage mechanism in modern concrete structures. But without steel there are also no beams, no cantilevers, no slender floor slabs. Everything the Romans built in concrete had to be a pure compression structure: arch, barrel vault, dome.

It hardens slowly. Anyone who pours a floor slab on Monday and loads it on Friday cannot work with a pozzolanically bound, lime-based material. The time economy of modern construction is an independent and frequently underrated reason why Portland cement won.

We see only the survivors. Of the Roman concrete structures, the ones still standing are those that were well designed, well executed, and favorably loaded. Whatever collapsed is no longer present to enter the statistics. With a sample age of two thousand years, that survivorship bias is not small.

And yet: what can be achieved with a pure compression structure and a moderately strong but extraordinarily durable material is demonstrated by the Pantheon in Rome, completed around AD 126 under Hadrian. The dome has an internal diameter of 43.3 m, with an identical height above the floor; it tapers from 6.4 m thickness at its base to 1.2 m at the rim of the oculus, which is thirty Roman feet, roughly 8.9 m, across. Five rings of 28 coffers each reduce the dead load further. And the aggregate is graded by height: heavy travertine at the bottom, brick above it, tuff and pumice at the top — porous lightweight stones with a bulk density around 1,350 kg/m³ against roughly 2,200 kg/m³ for travertine. The total weight of the concrete dome is given as about 4,535 tonnes. To this day it remains the largest unreinforced concrete dome in the world.

That is not material superiority; it is engineering under constraints. If you can only transfer compression, design a form in which only compression occurs, and distribute the material so that the weight becomes small wherever it endangers the form.

The second-to-last row contains the most important point in this article, and it is an uncomfortable one.

Part 4: What the Construction Industry Makes of This – and What It Cannot

Why the interest is so intense

Cement is a first-order climate problem. The MIT release accompanying the 2023 paper gives a share of roughly 8 % of global greenhouse gas emissions for cement production; the 2026 Berkeley release quantifies the output at 0.83 tonnes of CO₂ per tonne of clinker.

Those emissions have two sources, and the distinction is essential. One part is energy: a rotary kiln must be brought to about 1,450 °C for clinker phases to form. That part could in principle be reduced through different fuels or electrification. The other part is process-inherent and chemically unavoidable: decarbonating limestone releases the CO₂ that was bound in it. Anyone producing calcium oxide from calcium carbonate releases CO₂, and no energy source changes that. This is why every serious decarbonization pathway runs through reducing the quantity of clinker, not just through the kiln.

The three levers

Clinker substitution. Part of the clinker is replaced by pozzolans or latently hydraulic materials: fly ash from coal power plants, blast-furnace slag from steelmaking, calcined clay, limestone powder. This is Roman chemistry in modern dress. The catch: fly ash and slag are by-products of exactly the industries we are winding down, so their availability is falling. Calcined clay is the scalable route, because clay is available worldwide and is activated at temperatures far below those needed for clinker; combinations of calcined clay and limestone powder (known in the literature as LC³) are the most intensively pursued approach at present.

Durability as a climate strategy. This is the genuinely Roman argument, and it is underrated. If a structure lasts twice as long, then over its life cycle the material demand per year of service is halved. The MIT group argues in exactly these terms: longer service life reduces demolition, maintenance, and new construction. That is not a material property but an accounting question — and it routinely falls through the cracks of codes and business cases, because construction costs arise today and maintenance costs arise later.

Engineered self-healing. Here the line from lime clasts to the present is most direct. The principle is always the same: build in a dormant reserve that only becomes active once a crack reaches it and water mobilizes it. Variants include reactive lime inclusions, microencapsulated binders, swelling admixtures, and biological approaches in which calcite-forming bacteria are introduced as spores. None of these is standard site practice today, but the direction is clear, and the ancient evidence is a strong argument that long-term, water-triggered mineral formation works.

The mistake to avoid

And here is the point that popular accounts almost never contain: carbonation is the enemy in reinforced concrete.

The reason is pH. Fresh cement paste is strongly alkaline, typically above pH 12.5. That alkalinity maintains a thin passive layer on the steel surface which protects it from corrosion. When atmospheric CO₂ penetrates the concrete and converts calcium hydroxide to calcium carbonate, the pH falls — with advanced carbonation, into the region of 9. Below a threshold the passive layer breaks down and the steel begins to rust. Measuring carbonation depth and choosing concrete cover such that the carbonation front does not reach the steel within the intended service life is everyday work in construction.

So the very process the 2026 Berkeley study identifies as the cause of Roman longevity is what shortens the life of modern reinforced concrete structures. The difference lies not in the chemistry but in the system. In unreinforced Roman concrete the calcite that forms fills pores and densifies the fabric, and there is nothing inside whose protection depends on pH. In reinforced concrete, by contrast, the alkalinity is itself a load-bearing design element.

That is why "do as the Romans did" is not a construction program. Three things can be carried over from Roman concrete: pozzolans as part of the binder, a built-in self-healing reserve, and a willingness to weigh durability as a design objective against early strength. What cannot be carried over is the system design, because we have to transfer tensile forces and need steel for that, and because steel needs an environment that Roman concrete does not provide.

Part 5: The Pattern Behind the Story

For someone who builds systems rather than buildings, this story contains a pattern that reaches beyond concrete — and it runs counter to the usual intuition.

The modern engineering answer to durability is almost always: seal it. We try to isolate the system from its environment, close off ingress paths, prevent reactions. Coatings, waterproofing membranes, cathodic protection, dense concretes with low water-to-cement ratios — all variants of the same strategy: perfection through exclusion. That strategy is strong as long as the envelope holds, and it has an unpleasant property: it does not fail gradually but at one location, and from then on the damage accelerates.

Roman marine concrete embodies the other strategy: allow the reaction and make it useful. The material is not sealed against the sea; it exchanges with it — and the exchange produces minerals that fill voids. In this system a crack is not the beginning of the end but the trigger of a repair.

Anyone who operates distributed systems knows this pattern in a different vocabulary. A system that tries to exclude failure needs a perfect envelope and behaves catastrophically when the envelope breaks. A system that expects failure can build in mechanisms triggered by the failure's occurrence — redistribution, restoration from redundancy, self-healing after a node loss. The parallel to the lime clasts is strikingly direct: a dormant, unused reserve whose cost you carry in normal operation and which only becomes active in the damage case.

I do not want to overextend the analogy — concrete is not a cluster, and mineralogy is not consensus. But the design question is the same in both cases, and in both cases it is rarely made explicit: am I betting on an envelope that must not break, or on a mechanism that is triggered by the break?

The Central Takeaway

The most important sentence in this article is not "Roman concrete is better" — that is false. It is this: durability is not a material property but a system property, and it is decided in the design, not in the mix.

Roman moles do not survive because their binder is superior; it is inferior on every strength metric. They survive because three decisions come together: a form in which only compression occurs; no material inside whose protection depends on a chemical boundary condition; and a binder that reacts with its environment and fills voids in the process rather than creating them. Remove any one of those three decisions and the two-thousand-year story falls apart.

The practical prompt costs half an hour. Take a system you are responsible for and write down three things. First: what is its actually intended service life — not the one in the concept document, but the one the decisions were made against? Second: which single envelope must not break for it to hold? Third: is there any mechanism inside it that is triggered by the occurrence of a fault rather than damaged by it — and if not, what is the cheapest one you could build in?

The three answers will tell you whether you are currently building an envelope or a mole.

A Question to Reflect On

Roman concrete is durable in the sense that it slowly gets better — but it was built in a society that measured construction time in years, had effectively unlimited cheap labor, and knew no standard demanding a nominal strength at 28 days. Our concretes are optimized in the sense that they carry load quickly, can be calculated precisely, and fit into a construction sequence with a weekly rhythm — at the price that their durability depends on an envelope that will eventually fail.

This is not a question of knowledge or ignorance but of objective functions. If, for a system you are currently designing, you could choose between "lasts sixty years and is finished next month" and "probably lasts three hundred years and takes three years longer": who in your organization would have to make that decision — and would the structure you work in even allow the second option to come to a vote? And if not: is it the material we lack, or the accounting?

Cross-References in the Vault

• The Cosmos in Bronze: The Antikythera Mechanism and Humanity's First Computer – the same pattern of an ancient engineering tradition that was lost and not matched again for centuries.
• The Library Made of Charcoal: How X-Ray Light and AI Are Reading the Burned Scrolls of Herculaneum – synchrotron radiation and tomography applied to ancient material: the same methods, the same region, the same volcanic district.
• The Year Without a Sun: 536, the Volcanic Winter, and How Ice Cores Redated History – how a historical question can be settled through material-analytical archives, and where the limits of that decidability lie.
• The Clock That Measured the Ocean: John Harrison, the Longitude Problem, and How a Carpenter Saved Navigation – engineering against material limits, and the argument over whether a single artifact proves a technology.
• The Temple Before the Village: Göbekli Tepe and the Reversal of the Neolithic Revolution – monumental building as a window into a society's organizational capacity.
• The Networked Collapse: How a Globalized World Fell Apart Around 1200 BC – what happens when the supply chain for a specialty material breaks: pulvis puteolana had to be shipped seven hundred kilometers.
• The Strongest Bond in Nature: Nitrogenase and How Bacteria Split Nitrogen at Room Temperature – the sister problem: an industrial high-temperature process with an enormous CO₂ footprint versus nature's low-temperature route.
• Eleven Nines: Erasure Coding, Reed-Solomon, and How the Cloud Makes Data Practically Unlosable – durability through a built-in reserve unused in normal operation rather than through a flawless envelope.
• The Countdown in the Nucleus: Telomeres, Telomerase, and the Clock of Aging – a built-in repair capacity and the price its availability carries.

Sources

1. Linda M. Seymour, Janille Maragh, Paolo Sabatini, Michel Di Tommaso, James C. Weaver, Admir Masic: Hot mixing: Mechanistic insights into the durability of ancient Roman concrete. Science Advances 9(1), eadd1602, January 6, 2023. DOI 10.1126/sciadv.add1602 – hot mixing, lime clasts, and the crack-healing experiment. Science Advances; PubMed record; accompanying release: MIT News, Jan 6, 2023
2. Xiaohong Zhu, Sejung Rosie Chae, Stuart McElhany, Chengyao Liang, Qi Zheng, Jiaqi Li, Veronica Fondi, Sergio Del Ferro, Ascanio Modena Altieri, Harrison P. Lisabeth, Arun Bhattacharjee, Bruce W. Fouke, Hans-Rudolf Wenk, Paulo J. M. Monteiro: Mineralized carbonates contribute to the millennial durability of Roman concrete. Science Advances 12(28), eaeb0754, July 8, 2026. DOI 10.1126/sciadv.aeb0754 – radiaxial fibrous calcite, Hadrian's Villa, and the explicit refusal to claim direct evidence for deliberate hot mixing. Full text (PMC); accompanying release: Berkeley Engineering, July 2026
3. Marie D. Jackson, Sean R. Mulcahy, Heng Chen, Yao Li, Qinfei Li, Piergiulio Cappelletti, Hans-Rudolf Wenk: Phillipsite and Al-tobermorite mineral cements produced through low-temperature water-rock reactions in Roman marine concrete. American Mineralogist 102(7), 1435–1450, 2017. DOI 10.2138/am-2017-5993CCBY – new mineral formation from seawater pore fluids at ambient temperature. GeoScienceWorld; full text (eScholarship); accompanying release containing the Pliny quotation: University of Utah, July 3, 2017
4. Jackson MacFarlane, Tiziana Vanorio, Paulo J. M. Monteiro: Multi-scale imaging, strength and permeability measurements: Understanding the durability of Roman marine concrete. Construction and Building Materials 272, 121812, 2021. DOI 10.1016/j.conbuildmat.2020.121812 – measured stress drops between 14.6 and 19.1 MPa and ductile deformation behavior. ScienceDirect
5. Simone Dilaria, Michele Secco, Andrea R. Ghiotto, Guido Furlan, Tommaso Giovanardi, Federico Zorzi, Jacopo Bonetto: Early exploitation of Neapolitan pozzolan (pulvis puteolana) in the Roman theatre of Aquileia, Northern Italy. Scientific Reports 13, 4110, 2023. DOI 10.1038/s41598-023-30692-y – analytical proof of long-distance trade in Neapolitan pozzolan. Scientific Reports
6. Vitruvius Pollio: De architectura, Book II, chapter 6 (pulvis puteolanus, mixing ratios) and Book V, chapter 12 (harbor construction, caissons). Latin text with English translation: Perseus Digital Library
7. C. J. Brandon, R. L. Hohlfelder, M. D. Jackson, J. P. Oleson (eds.): Building for Eternity: The History and Technology of Roman Concrete Engineering in the Sea. Oxbow Books, Oxford 2014, ISBN 978-1-78297-420-8 – final report of the ROMACONS drill-core project. Full text (Internet Archive)
8. Grady Hillhouse: Was Roman Concrete Better? Practical Engineering, March 9, 2019 – the engineering counter-position: survivorship bias, the strength comparison, and reinforcement corrosion as the dominant damage mechanism. Practical Engineering
9. Dimensional data for the Pantheon dome (internal diameter 43.3 m, thickness 6.4 m to 1.2 m, oculus ~8.9 m, graded aggregate, total weight ~4,535 t) after the summary presentation in Wikipedia: Pantheon, Rome – a tertiary source; individual figures should be checked against the building-research literature cited there.

← All articles