Hot mixing: Mechanistic insights into the durability of ancient Roman concrete

Abstract

Ancient Roman concretes have survived millennia, but mechanistic insights into their durability remain an enigma. Here, we use a multiscale correlative elemental and chemical mapping approach to investigating relict lime clasts, a ubiquitous and conspicuous mineral component associated with ancient Roman mortars. Together, these analyses provide new insights into mortar preparation methodologies and provide evidence that the Romans employed hot mixing, using quicklime in conjunction with, or instead of, slaked lime, to create an environment where high surface area aggregate-scale lime clasts are retained within the mortar matrix. Inspired by these findings, we propose that these macroscopic inclusions might serve as critical sources of reactive calcium for long-term pore and crack-filling or post-pozzolanic reactivity within the cementitious constructs. The subsequent development and testing of modern lime clast–containing cementitious mixtures demonstrate their self-healing potential, thus paving the way for the development of more durable, resilient, and sustainable concrete formulations.

INTRODUCTION

Ordinary Portland cement (OPC) is a key component of concrete, the most ubiquitous construction material in the world, but its production has serious environmental consequences. The manufacture of OPC releases up to 1 metric ton of CO_2e emissions per metric ton of produced material, and current strategies to reduce this impact are insufficient as demand continues to rise (1). These emissions are primarily generated when limestone and clay are calcined to form clinker (mainly tricalcium and dicalcium silicates, also known as alite and belite, respectively), which is then finely ground. One method to reduce cement’s carbon footprint (which accounts for up to 8% of total global greenhouse gas emissions), is to improve the longevity of concrete through the incorporation of self-healing functionalities. The resulting extended use life, combined with a reduction in the need for extensive repair, could thus reduce the environmental impact and improve the economic life cycle of modern cementitious constructs (2).

In contrast to their modern counterparts, ancient Roman mortars and concretes have remained durable in a variety of climates, seismic zones, and even in direct contact with seawater, as in the case for maritime concrete. Because of this proven longevity on the order of millennia, these ancient construction materials are attractive model systems for the design of sustainable, durable cementitious composites for modern engineering applications (3–5).

For many centuries, and throughout the entire ancient Roman Empire, architectural elements, such as walls and foundations, and infrastructure systems, including aqueducts, roads, and bridges, were created from unreinforced concrete. This concrete was typically composed of volcanic tuff and other coarse aggregates (caementa), and bound by a mortar based on lime and pozzolanic materials such as volcanic ash (pulvis) (6), the detailed formulations of which were tailored to their specific intended applications (7, 8). Whereas aerial lime mortars relied on the uptake of CO₂ from the air to harden, hydraulic mortars combined lime and water with reactive silicates and aluminosilicates (pozzolanic materials) to form cementitious hydrates [e.g., calcium alumino-silicate hydrates (C-A-S-H)] (9–13).

By developing these hydraulic mortars, the Romans were able to create a stronger, more durable material that allowed them to build larger, more complex-shaped architectural structures for purposes that were not previously possible, including constructions in the sea (14–16).

The production process for Roman mortar began with the calcining of lime from a source such as limestone, marble, or travertine (all predominantly calcite, CaCO₃) to form quicklime [calcium oxide (CaO)] (17). This lime-based material, which can be hydrated using water (a process known as slaking) or added directly (a process known as hot mixing) (18–21), was then mixed with volcanic ash, ceramic fragments (cocciopesto), or other pozzolana, sand, and water to form the hydraulic mortar.

Regardless of the application, strict specifications for the raw materials were detailed by the ancient scholars Vitruvius and Pliny, especially for the limestone, which was to be pure white, so as to lack impurities (6, 17). Previous studies on Roman architectural mortars determined that the calcined lime in these samples frequently contained <5 weight % oxides other than CaO (7, 14). In recent years, detailed investigations of ancient mortars and concretes allowed for the identification of some of the key chemical and mineralogical processes associated with interactions between pozzolanic materials and hydrated lime and provided insights into the mechanical performance of these materials (11, 12, 22–26).

Studies focusing on the durability of Roman concrete constructed in marine environments, for example, evidenced the dissolution of lime and vitric tuff clasts at high pH, followed by the precipitation of C-A-S-H–containing reaction rims and, subsequently, the post-setting crystallization of Al-tobermorite and phillipsite in the matrix (11, 27). In both Augustan and Imperial era architectural concretes, a similar C-A-S-H precipitation and subsequent crystallization and growth of platy strätlingite crystals in the perimeters of scoriae and the cementing matrix were observed (24, 28). More recently, both Al-tobermorite and strätlingite crystals were found in the mortars of the Augustan period (ca. 30 BCE) tomb of Caecilia Metella (29). The prolonged reactivity of volcanic aggregates and their potential role in the long-term durability of these materials has thus been an ongoing focus of recent studies on Roman concretes.

In addition to the features described above, aggregate-scale relict lime clasts, also referred to as remnant lime or lime lumps, are a ubiquitous and conspicuous feature of both architectural and maritime Roman concretes. The presence of these distinctive bright white features has been previously attributed to several scenarios including incomplete or over-burning during the calcining of lime (20), carbonation before concrete preparation (30), incomplete dissolution during setting (12), or insufficient mixing of the mortar (14).

In maritime structures, these lime clasts can be heterogeneous in composition; may contain calcite, vaterite, brucite, ettringite, hydrocalumite, C-A-S-H, tobermorite, and Al-tobermorite (5, 12, 27, 30–34); and can be categorized into one of the following three groups: (i) geologic, (ii) partially dissolved, and (iii) fully dissolved (transformed) (12). Geologic inclusions are calcite-bearing aggregates that did not fully calcine during the production of quicklime. The other two, partially and fully dissolved clasts, have provided insight into the chemical evolution of maritime concretes. For example, partially dissolved clasts show a C-A-S-H–containing reaction rim, gradating toward a calcite-rich core (12). Fully dissolved clasts, in contrast, exhibit C-A-S-H throughout and, in some cases, the formation of Al-tobermorite (12, 27). In both partially and fully dissolved clasts, the clast exteriors contain hydrocalumite and ettringite, attributed to hydration with seawater (12). Although these clasts are well characterized in maritime Roman concretes, less is known regarding the microstructure and chemical composition of relict lime clasts in open-air Roman constructions (structures on land) and the role that they might play in processes associated with the durability of these structures (5, 35).

To address these yet unresolved questions, we report on the chemical characterization of relict lime clasts found in 2000-year-old Roman concrete samples obtained from the archaeological site of Privernum, Italy. The investigated samples are compositionally consistent with other architectural mortars encountered throughout the Roman Empire (8, 13, 16, 36, 37), and were sourced from the masonry mortar of the city wall, an open-air structure. We characterized the composition of the lime clasts and their surrounding matrix using large-area scanning electron microscopy and energy dispersive x-ray spectroscopy (SEM-EDS), powder x-ray diffraction (XRD), and confocal Raman imaging. The results of these analyses provide compelling evidence for hot mixing of Roman mortar using quicklime instead of, or in addition to, slaked lime. From these findings, we propose that persistent, aggregate-scale, high surface area lime clasts that result from this process could serve as a source of reactive calcium for long-term pore and crack filling and therefore provide a chemically dominated intrinsic self-healing mechanism. Motivated by these discoveries, we developed a modern Roman-inspired cementitious mixture that incorporated a hot mixing preparation method and observed the effective self-healing of induced cracks measuring up to 0.5 mm in width. The results from our multiscale compositional analysis of Roman concrete and the development of their modern counterparts provide a starting point to understand the role of various production parameters on the durability of ancient mortars and concretes and demonstrate the applicability and potential value of these ancient design principles in the development of cementitious materials for the modern world.

RESULTS

The Roman mortar samples investigated in this work came from the ancient city of Privernum, near Rome, Italy (Fig. 1A). Privernum was occupied as early as the 2nd century BCE through the 13th century CE, and the site was first excavated in the 18th century, with the most recent work performed from 1990 to present. The samples analyzed in the present study were part of the city wall (Fig. 1, B and C), are compositionally consistent with other Roman concrete samples dating from the same time period (8), and correspond with the city’s Roman occupation, which lasted through the fall of the empire.

To investigate centimeter-, millimeter-, and submillimeter-scale compositional details of the mortar samples, we used SEM-EDS. Large-area elemental mapping of the freshly fractured surface of a representative Roman mortar sample (Fig. 1D) shows its complex chemical composition (with a dominant presence of calcium, silicon, and aluminum), the aggregate-scale relict lime clasts, and evidence for pore filling by secondary mineral phases (fig. S1). Further elemental mapping of polished mortar samples (Fig. 1E) provided additional fine-scale structural and compositional information, the results of which were consistent with previous studies of Roman mortars and concretes (11, 12, 27). Note that to obtain reliable elemental mapping data from the irregularly fractured concrete samples, we used a multidetector EDS system, which both minimized the potential contributions of shadowing artifacts and facilitated the high-throughput analysis of a large number of test samples. In total, more than 15 different fractured and polished mortar sample regions were analyzed (from 10 different collected fragments) using this approach, and the results are representative of the trends observed.

These elemental maps (Fig. 1, D and E, and fig. S1) also show sulfur throughout the mortar, which, as proposed previously (38), could be due to primary sulfur originating from the components used to prepare the mortar, secondary sulfur that entered the binding matrix through the pores of the mortar, seen as rings, or a combination of these. While all of these considerations are possible, detailed elemental analysis (fig. S1) shows that sulfur is present primarily in the pore volume, with secondary sulfur likely originating from one or more of a variety of external, environmental sources including soil and acid rain. The latter intrusion is a common degradation pathway in which cementitious materials are leached on exposure to acid rain containing sulfur-based ions, as calcium carbonate is converted to calcium sulfate hydrates (e.g., gypsum) (39, 40). Raman mapping (fig. S2) demonstrates that the sulfur-bearing pore walls contain predominantly gypsum, thus precluding the use of calcium sulfoaluminate fibers as a filling agent in these samples (38). Despite centuries of exposure to environmental attack, however, the masonry wall remains well preserved around the perimeter of the archaeological site.

To identify both the major and minor mineral phases present in the concrete wall samples and map their specific distribution patterns, we used two modes for quantitative analysis of the mortar’s elemental composition: first on freshly fractured surfaces (Fig. 2, A to C) and second on prepared, polished cross-sections (Fig. 2, D to F). Quantitative EDS data were plotted pixel by pixel on ternary diagrams (Fig. 2, B and E) to reveal the relative ratios of the three key elements of the cementitious phases—calcium, silicon, and aluminum (41).

Immediately evident from these analyses is the compositional shift toward the (red) calcium vertex of the ternary diagram for the fractured surface (Fig. 2, A and B), suggesting that the sample likely failed along weaker binding phases, the relict lime clast, and the binder immediately surrounding the clast. Quantitative EDS data were then clustered to visualize the distribution of small differences in composition throughout the mortars. Points on the ternary diagrams (Fig. 2, B and E) and individual pixels on the original map area (Fig. 2, C and F) were colored on the basis of the assigned cluster. Within the polished cross-sections (Fig. 2F), the different investigated relict lime clasts all separated into two clusters—one within the clast reaction rim and another in the core. Evident here is a transition from the calcium-rich core to a rim containing higher concentrations of silicon and aluminum. Further analysis of the EDS maps also shows that the matrix immediately surrounding the lime clasts is richer in calcium than other parts of the matrix, suggesting that some leeching has occurred in the sample. Just outside the lime clasts and extending to the edge of the sample, a matrix phase rich in calcium with aluminum and silicon can also be seen. As shown in Fig. 2F, these points (red) are distinct from the primary binding phase (yellow), which has ratios of silicon, aluminum, and calcium more closely associated with C-A-S-H binders (41). Petrographic analysis (fig. S3) shows that the binder is predominantly isotropic, as would be the expected product from the pozzolanic reaction (poorly crystalline C-A-S-H), with calcite occurring in the matrix near lime clasts and within some pores.

A better understanding of the mineral assemblage and the potential identification of crystalline phases within the mortars was obtained through powder XRD. Two sample types were examined in this way: extracted relict lime clasts from the mortars (Fig. 2G) and the binding matrix complete with other aggregate material (Fig. 2H). The remnant lime clasts were predominantly calcite and quartz (likely due to imperfect separation of the clast from surrounding binder and aggregates), although the diffraction data also suggest the presence of small quantities of potential recrystallization phases such as vaterite. Overall, the elemental and mineral assemblage in the Privernum mortar was well aligned with the composition of previously studied Roman mortar samples (9, 19, 25, 34, 35), highlighting the consistency of these samples from archaeological site to site.

Large-area SEM-EDS mapping of a representative resin-embedded cross-section of the Privernum mortar is shown in Fig. 3A. The relict lime clasts, identified by their high concentrations of Ca, were studied in detail using correlative EDS and Raman microspectroscopy (Fig. 3, B to E). The elemental composition of the lime clasts was clustered as described above to identify the distribution of compositional differences within the samples. The clustering results were plotted on the ternary axes (Fig. 3C) to characterize the relative ratios of the primary cementitious components. Six clusters were visualized corresponding to quartz aggregates, resin-rich regions, the cementitious matrix, two clusters within the lime clast, and one within the clast reaction rim. The reaction rim cluster is also seen incorporated into the cementitious matrix and corresponds to a composition range akin to calcium-enriched C-A-S-H phases (41). Each pixel of the lime clast region was recolored on the basis of the cluster that it belongs to in Fig. 3E.

High-resolution mapping data of the mineral phases in this same cross-section were obtained using Raman microspectroscopy (Fig. 3D). This methodology was chosen because the different calcium carbonate polymorphs of calcite, aragonite, and vaterite exhibit different Raman spectra due to the different binding environments within the crystal structures (42). Two carbonate-dominated phases were identified with this method. Specifically, one such phase (Fig. 3D, yellow) is characterized by a shift of −3 cm⁻¹ of the ν₁ carbonate peak with respect to the 1085 cm⁻¹ peak of calcite and a fluorescence signature distinct from the rest of the sample (fig. S4). The distribution of the phases identified through Raman spectroscopy is shown in Fig. 3E, and fig. S4 further demonstrates sample heterogeneity. In some of the relict lime clast spectra, a shoulder that can be associated with disordered CaCO₃ was present on the ν₁ peak (see fig. S5). Broadening of the ν₁ peak compared to a geologic Iceland spar calcite standard was also observed (fig. S6).

The recent discovery of calcite-filled cracks in Roman concrete (29) suggested a potential long-term healing process that requires a Ca-rich source. Considering the ubiquity of relict lime clasts in Roman concrete (5, 12, 24, 43) and their high surface area due to their particulate microstructure (Fig. 4, A to D), these inclusions might provide the requisite Ca reservoirs for these processes. Inspired by these observations, we set out to create a modern analog of this material and explore its properties. Considering the relatively even distribution of the lime clasts and their well-defined morphologies, these observations raise the intriguing proposition that they may have been added to the concrete mix in their intact form (a practice known as hot mixing). Considering the documented use of hot mixing in both ancient and modern mortar preparations (19, 44), we explored this approach in our synthetic formulations.

In our modern cementitious formulations, we combined OPC, pulverized fly ash, sand, and water at a mass mixing ratio of ca. 1:0.2:2:1. To this mixture (known herein as Mix 1), quicklime was added to different samples, spanning a concentration varying from 7.5 to 15 mass % (which was chosen on the basis of the encountered frequency range of relict lime clasts in different Roman concrete samples) to investigate the effects of quicklime content on the concrete’s performance. The resulting mixtures were cast into cylindrical molds measuring 10 cm in diameter and 10 cm in height and cured under water for 28 days before use.

Once the production of our modern concrete samples was complete, we explored the microarchitecture of the lime clasts and compared their features to those found in their ancient Roman equivalents (Fig. 4). In both types of samples (ancient and modern), the bright white color of the lime clasts and their porous microstructure were visible. Further examination of the lime clasts from the polished samples of the Roman-inspired concrete formulations using polarized optical microscopy revealed a similar visibly distinctive periphery that was seen in the native Roman material (cf. Figs. 2F and 3E and figs. S7 to S9).

The potential for the high–surface area lime clasts to act as a reactive calcium source for self-healing properties was further investigated using our Roman-inspired concrete formulations. The cured samples were fractured by splitting the sample longitudinally (Fig. 5A), and the opposite walls of the cracked sample were adjoined at a distance of ca. 0.5 mm.

After an initial preconditioning step (45), the re-mated sides of the test samples (which also included lime clast–free controls) were then introduced into a flow circuit (Fig. 5B) and subjected to a constant water flow. The flow rate over 30 days was then recorded using electronic flow meters with a precision of ±5 liters/hour. In these studies, a measured decrease in flow rate was predicted to correspond to potential self-healing (crack-filling) activity. Typical initial values of flow rates through open cracks ranged between 10 and 30 liters/hour and reduced over the course of 1 to 3 weeks depending on the crack geometry, to almost zero when the crack was eventually sealed (Fig. 5C). When the flow of water eventually stopped (and the flow rate was zero or at a negligible flow), the test was stopped, and the nature and distribution of secondary products formed in the crack (Fig. 5, D to F) during the process were evaluated using optical microscopy (Fig. 5, D and E) and Raman spectroscopy (Fig. 5F).

On the basis of these initial findings and to assess the real-world applicability of a self-healing Roman-inspired quicklime-containing concretes in the modern built environment, we developed a second formulation (known herein as Mix 2) using an industrially relevant concrete recipe that included coarse aggregates and superplasticizer. These concrete samples (along with quicklime-free controls) were cast as rectangular prisms, and their drying shrinkage was measured 90 and 365 days after casting. After 90 days, the quicklime-containing Mix 2 concrete samples had shrunk 9% less than that of the quicklime-free controls, demonstrating the role of quicklime as an effective shrinkage compensator during the early stages of the curing process. At 365 days after casting, the drying shrinkage of the two concretes had converged (<1% difference), thus confirming the marginal effects of expansive quicklime in this industrially relevant concrete mix.

DISCUSSION

Previous evidence suggests that in ancient Roman cementitious structures, the relict lime clasts can react over time with other mortar components, forming both amorphous (C-A-S-H) and crystalline (e.g., Al-tobermorite and strätlingite) phases (11). These maturation pathways, however, are not the only ones observed in Roman cementitious materials and, as shown here, can also follow different trajectories depending on the local environmental conditions (Figs. 2 and 3). While the polished cross-sections of the ancient Roman concrete samples described in the present study (Fig. 2, D to F) do show clear inclusion of silicon and aluminum in the lime clast, suggesting that some conversion to C-A-S-H and/or its crystalline homologues occurred, calcium carbonate, predominantly as calcite, remains within their cores (Fig. 3D).

It has already been suggested that calcium carbonates originating from different sources (e.g., carbonated slaked lime, carbonated free lime, or geologic calcite) show differences in their morphology and crystal structure (46, 47). For example, slaked lime [Ca(OH)₂] transforms into calcite (CaCO₃) when exposed to atmospheric CO₂; however, it is relatively more disordered than geologic calcite samples and, for this reason, can, in some cases, be differentiated as such (46).

During the production of concrete, water availability plays a critical role in dictating the reaction kinetics and the evolution of different mineral phases. For example, previous Raman spectroscopy and XRD studies investigating the hydration and carbonation of calcium oxide samples exposed to different ambient humidity levels demonstrated that at relative humidities below 20%, calcium oxide converts to amorphous calcium carbonate (48). However, at relative humidity levels of 20 to 40%, conversion to calcite, aragonite, and vaterite was observed, while samples exposed to humidity levels above 60% showed only calcite. These results suggest that the relative humidity of the calcium oxide maturation environment can be inferred through the measurement of calcium carbonate polymorph ratios within the lime clasts.

Using these criteria and on the basis of the lime clast composition from ancient Roman concretes reported here, a potentially valid explanation of these water-depleted conditions could be hot mixing, referred to as such because of the exothermic nature of the CaO hydration reaction (19). In these preparations, quicklime is simultaneously mixed with the pozzolans, aggregates, and water instead of, or in addition to, slaked lime as the calcium source in the mortar. Hot mixing thus has the potential to change the conditions under which the lime clasts are transformed within the mortar. It is known that hot mixing of other ancient lime-based mortars is one method that leads to lime inclusions and that modern reproductions of these mortars have been shown to set faster compared to those made using slaked lime putty or powder (18, 19). The characteristic features of the lime inclusions described in the present study could also be explained by the temperature increase associated with hot mixing, which facilitates both the reactions responsible for hardening the mortar and the formation of a gradient in hydration around the lime clasts, preventing them from further dissolution in the mortar matrix.

Regarding the production processes that may have favored the lime clasts remaining intact instead of dissolving and reacting in the binding matrix, previous research on other ancient Roman mortars has suggested a series of hypotheses. These possibilities include (i) hydration and subsequent partial carbonation of the lime before incorporation in the mortar; (ii) overburned lime, wherein the CaO is passivated and less reactive; (iii) underburned lime, wherein the lime is not completely calcined; or (iv) the use of coarse quicklime instead of a fine powder or paste (18, 30). The gradient in composition seen in the EDS maps of the lime clasts in the Privernum samples suggests that some hydration-driven reactions have taken place between the lime and aluminosilicates in the binder, thus likely resulting from one of the latter two pathways. The use of coarse particles of quicklime (alone or in conjunction with slaked lime) in the mortar is another possibility, which could allow hydration reactions to take place on the clast exterior and allow the development of the chemical gradient seen in the lime clasts of the Privernum mortar and elsewhere (5, 12, 32, 49). The heat resulting from the hydration of quicklime could reduce the availability of water, potentially resulting in the low-humidity conditions that could explain the Raman spectra reported here. Furthermore, these lime clast hydration gradients are also observed in our modern quicklime-containing mixes that were developed and investigated in the present study (Fig. 4, J to L, and fig. S9). Further evidence supporting the hypothesis of quicklime use in the primary mix is the cracks observed in petrographic photomicrographs that form during the hydration of quicklime in both ancient samples (12) and modern reproductions (Fig. 4L), as well as the highly porous nature of lime clasts observed in the fractured surfaces of both ancient and modern lime inclusions (Fig. 4, D and H).

Both of these phenomena (the formation of internal microcracks and a porous architecture) can be attributed to volume changes associated with lime cycle transformations. The CaO reaction with water to produce Ca(OH)₂ is accompanied by a density-induced volume expansion of ca. 33% (assuming the densities of CaO and Ca(OH)₂ at 3.34 and 2.2 g/cm³, respectively) and accounts for clast crack formation and the observed porosity. While these volumetric changes in the concrete constituents might be expected to negatively affect concrete performance, our results with modern lime clast–containing industrially relevant concrete formulations have demonstrated that, even after 1 year, lime clast–containing concretes exhibit negligible drying shrinkage compared to their lime clast–free controls.

Despite extensive literature related to the composition and applications of ancient Roman concretes, the exact order of operations for Roman mortar production based on historical evidence remains ambiguous. There is even debate as to whether preparation techniques differed between the production of marine and terrestrial cementitious structures (12). Imperial age mortar (according to Vitruvius) was prepared by mixing lime with volcanic sand (materies ex calce et harena mixta). During the Republican period, Cato, in his De Agri Cultura (50), describes the mortar mix as calx harenatus (“lime with sand”). The wet mortar mix would then be mixed with tuff and brick caementa to form a concrete. In general, for frescoes and wall plaster, for example, the ancient scholars would often suggest the aging of lime in water before use and ensure that it was as finely ground as possible (17), because incompletely hydrated lime particles, known as bottaccioli in these applications, could absorb water over time and expand, damaging the paint (fresco) layer. For this reason, both Vitruvius (6) and Pliny (17) describe the preparation of lime for plasterwork to involve a thorough soaking or softening process (macerata). When referring to lime for structural use, however, Vitruvius uses the word extincta (II.5.1) instead of macerata. While extincta and macerata are both frequently interpreted as referring to slaking, Vitruvius’ change in diction points to a potentially different process. On the basis of the results of our chemical characterization of the Privernum mortars, it is thus possible that in contrast to the use of macerata (which specifically refers to the slaking process), extincta could refer to lime hydrated simultaneously with the other mortar components, supporting the hot mixing hypothesis proposed here. Furthermore, it has also been previously proposed that some Roman mortars were prepared through the sequential addition of both slaked lime and quicklime in a two-step process (51).

Hydration of quicklime during hot mixing produces an exothermic reaction as CaO hydrates to form Ca(OH)₂. The temperature increase in the mortar is approximately 55° to 60°C over ambient (52), with a presence of hot spots characterized by temperatures exceeding 200°C (19). Previous thermodynamic modeling of the pozzolanic (hydration) reaction in Roman marine concrete has suggested that temperatures up to 97°C are possible within thick concrete structures from the pozzolanic reaction alone (23). The heat from these two reactions, the hydration of CaO and the pozzolanic reaction, could facilitate the formation of C-A-S-H and even crystalline phases such as Al-tobermorite and strätlingite, in the cementing matrix and along the reaction rim of the lime clasts in the early stages of the mortar hydration. Evidence for these reactions is further supported by the inclusion of silicon and aluminum in the EDS clusters identified within the reaction rims of the lime clasts studied herein and several pieces of evidence reported elsewhere (34). The pozzolanic reaction is sensitive to changes in temperature (39), and, hence, hot mixing provides a plausible explanation for the gradient in composition seen within the lime clasts of the Privernum mortar samples described in the current work and those from other Roman concretes (12, 32).

Whereas previous evidence supports the formation of Al-tobermorite minerals in lime clasts of maritime structures due to the heat of the pozzolanic reaction (23), the results presented here suggest that, by contrast, when quicklime is introduced via hot mixing in terrestrial structures, the reaction is limited to the outer rim of the lime clast, encapsulating calcium-rich core structures within mortar matrix. This particular configuration can be observed in the comparison of Roman and modern samples (Figs. 2 and 4), both showing the same type of lime clast inclusions, hydration-driven cracking, and distinctive reaction rims. The initially matrix-incorporated clasts would then undergo a slow transformation into calcium-rich, highly porous phases of various polymorphs of calcium carbonate.

Inspired by these discoveries, it is thus likely that the high abundance of aggregate-scale lime clasts in ancient Roman mortars could thus serve as a source of calcium for post-pozzolanic processes in a pore- and crack-filling “self-healing” mechanism that combats the progressive degradation of these cementitious materials (Fig. 6). Over time, as cracks and pores form, the intrusion of water causes the dissolution of calcium-based phases in the relict lime clasts carrying them into the pore network. As the calcium-rich fluids leach into the cracks or the connected pore network, many pathways exist for potential post-pozzolanic reactions. For example, excess pozzolanic material, such as volcanic ash that did not react during the initial setting and curing, or aggregates of volcanic origin, can now dissolve and react with Ca-rich fluids originating from the lime clasts to form C-A-S-H phases, thus reinforcing the interfacial zones between volcanic aggregates/ash and the binding matrix (29). This strengthening is associated with the consolidation of the interfacial zones and increased mechanical performance of the C-A-S-H compared to its precursors (53). Another possible pathway is the recrystallization of CaCO₃ phases within the pore/crack space. This pathway, in which secondary calcium carbonate is precipitated through a mechanism similar to the one occurring in the formation of calthemites (54, 55), relies on the wetting and drying cycles experienced through normal weathering conditions. These processes have been observed previously in both modern and ancient carbonate-based mortars (56–58). In contrast to these previous studies, however, we suggest here that the hot mixing–transformed lime clasts act as a calcium source for these processes and, furthermore, the precipitation of calcium carbonate as a crack-filling mechanism is already a known pathway for autogenous healing in other ancient lime-containing mortars (57). In the present study, we demonstrate the calcium enrichment of matrix phases adjacent to the lime clasts (shown in red in Fig. 2F), supporting the hypothesis that the lime clasts are a source of calcium for leaching and recrystallization within the pore space of the mortars (Fig. 2, D to F). Microcrack filling by calcite has been recently observed in ancient Roman mortars from the tomb of Caecilia Metella (29), and the self-healing tests carried out on our modern samples described in the present study (Figs. 4 and 5, C to F) further support this hypothesis.

An important question to be asked at this point is how the gradual carbonation of lime clasts affects the self-healing capacity of the final mixes. It is clear that the freshly hot-mixed concrete that incorporates lime clasts has autogenous self-healing potential. If a crack propagates through the concrete, then the high–surface area Ca(OH)₂-rich clasts release Ca²⁺ into a highly alkaline solution that readily carbonates and seals the crack (Fig. 5). A long-term carbonation of clasts would intuitively compete with this process considering the buffering effect of carbonates and subsequent pH reduction of the fluid that is oversaturated in chemical constituents. Nevertheless, in the case of lime clast carbonation [a process that is distinct from the carbonation of portlandite in modern concrete formulations (59)], a high surface area (evident in SEM images) further improves solubility (60), thus benefiting the mechanism proposed herein.

The proposed post-pozzolanic process introduced here can be compared to the autogenous healing in modern cementitious materials wherein cracks are filled by carbonation products or via the further hydration of unreacted pozzolanic particles (61, 62). It should be noted that cementitious compounds contained in the concrete can always induce some degree of natural self-healing. For example, it is generally assumed that in these types of self-healing studies, where water is pumped through a damaged structure, at relatively low–water pressure gradients and up to 10% relative movement of the crack, any crack with a width of up to 0.20 mm can naturally self-heal, while for relative movements between 10 and 20%, the crack width for natural self-healing is between 0.15 and 0.20 mm (63). However, for cracks wider than the above values (such as those used for the self-healing experiments in Fig. 5), any measured reduction in water flow through the structure induced by self-healing would imply that additional mechanisms have occurred.

The effectiveness of the pore- and crack-filling mechanism described here is predicated on the frequency of cracks intersecting the abundant highly porous and intrinsically reactive lime clasts. In support of this requisite damage mode, the optical and SEM imaging of freshly fractured ancient Roman concrete surfaces reveals that fractures predominantly propagate through the lime clasts in the architectural mortar, which are visible on both sides of the crack interface. It should be noted, however, that lime production in antiquity could lead to underburned/overburned lime that could potentially influence their surface area, subsequent reactivity, and mechanical properties. While our analyses suggest that lime clast composition in Roman mortars is relatively consistent from sample to sample, in some cases, higher-density lime clasts in ancient mortars can act as aggregate, with cracks instead propagating through the interfacial transition zone (5).

The entire self-healing mechanism suggested here occurs upon stimulation from external forces that would, otherwise, cumulatively result in material failure if left unchecked. The calcium-rich phases within the lime clasts remain stored until they are needed, thus potentially allowing the healing properties of these materials to persist over millennia. To this end, microcracks and pores induced by environmental changes (e.g., freeze-thaw cycling and acid rain) or mechanical damage (e.g., settling and seismic loads) can be filled, with water activation, once the cracks reach the lime clast particles.

In this work, we successfully demonstrate that these ancient concrete design practices could be applied for the development of modern OPC-based formulations, thus showing far-reaching implications for extended concrete design life, long-term durability, and sustainability. Whether the damage occurs within years of construction or centuries thereafter, so long as the lime clasts remain, these self-healing functionalities can persist. Inspired by the demonstrated potential of this technology, future studies will explore how such an ad hoc self-healing mechanism can be implemented in modern infrastructure, both for reinforced concrete and for unreinforced applications as are commonly used, for example, in concrete three-dimensional printing efforts (64).

MATERIALS AND METHODS

Acknowledgments

We would like to thank the archaeological area and the archaeological museum of Priverno, as well as the community of Priverno, particularly M. Cancellieri, N. Cassieri, and F. Di Mario for cooperation and assistance in providing access to the Privernum archaeological site. The reconstruction data shown here were collected with the expertise of Gianfranco Quaranta of Artech Lab and Roberto Scalesse of the Erresse Group, both of whom are members of Associazione per la Ricerca e l’Educazione nell’Arte, Archeologia e Architettura (AREA3). The self-healing experiments were conducted with the help of P. Tudori of the Istituto Meccanica dei Materiali (IMM). We dedicate this paper in memoriam of mastro Gilberto Quarneti (1941-2021) for his contribution to ancient materials studies.

Funding: Partial funding for this project was provided by the MIT Edward H Linde (1962) Presidential Fellowship program to Linda Seymour.

Author contributions: Conceptualization: A.M. Supervision: A.M. Methodology: L.M.S., J.M., P.S., M.D.T., J.C.W., and A.M. Investigation: L.M.S., J.M., P.S., M.D.T., J.C.W., and A.M. Visualization: L.M.S., J.M., and J.C.W. Writing—original draft: L.M.S. and A.M. Writing—review and editing: L.M.S., J.M., P.S., M.T., J.C.W., and A.M.

Competing interests: P.S. and A.M. are stakeholders in DMAT srl. A.M. and L.M.S. are inventors of patent applications related to the results presented herein, assigned to the Massachusetts Institute of Technology (published as WO 2019/204776). DMAT srl has certain rights to commercialize these patent applications. P.S. and M.D.T. are listed as inventors of the International (PCT) application entitled “Building materials” (PCT/IB2022/000353), filed 7 April 2022, and claiming priority to U.S. provisional application no. 63/171,770, entitled “Building materials” and filed 7 April 2021. The other authors declare that they have no competing interests.

Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials.

Supplementary Materials

REFERENCES AND NOTES