The technology of ancient lime mortars from the Żejtun Roman Villa (Malta)

Studies on original mortars can greatly assist archaeological interpretations, as elucidating the composition of such mortars gives clues on the origin of raw materials, manufacturing technology, and the construction phases of a site. This article presents the multi-analytical characterisation of 24 mortars and plasters from the Żejtun Roman Villa, Malta, to support archaeological hypotheses on the history of the construction of the site. The samples, belonging to at least three distinct phases included in the stratigraphy of the Żejtun archaeological site, were analysed using polarised light microscopy (PLM), scanning electron microscopy (SEM–EDS), X-ray diffraction (XRD), thermogravimetry (TGA/DSC), X-ray fluorescence (XRF), ion chromatography (IC), and stable isotope analysis (13C and 18O). The assessment of the results through correlations with archaeological evidence identifies five types of mortars with varying degrees of hydraulicity. These are associated with different development phases of the site and distinctive uses and were mainly produced using local resources, except in the Early Roman period when natural pozzolanic raw materials started being used. As there are no natural pozzolans on the Maltese Islands, it is hypothesised that the pozzolanic materials used as aggregate in the mortars were imported to the Islands from neighbouring volcanic regions. This volcanic aggregate was especially abundant in one of the mortar types, which was used mainly as a bedding mortar for floors.


Introduction
Roman technology for making lime mortars and plasters spread throughout the Mediterranean. Thanks at least in part to their quality and durability, many Roman monuments and cities have survived to this day (Degryse et al. 2002;Riccardi et al. 2007;Yaseen et al. 2013;Drdácký et al. 2013;Cardoso et al. 2014;Columbu et al. 2018;Vitti 2021). The mixing of aerial lime and a reactive aggregate such as volcanic ash (pozzolana), which created a material with hydraulic properties, was a great technological achievement attributable to the Romans (DeLaine 2021; Lancaster 2021). The availability of resources to produce hydraulic mortars was however not uniform, although pozzolana was widely exported from its source locations, other regions had to adapt mortar recipes to locally available materials. Artificial pozzolanic materials, such as crushed ceramics (cocciopesto/opus signinum), were also utilised for their hydraulic qualities as well as their wide availability (Yaseen et al. 2013;Fichera et al. 2015;Miriello et al. 2015;Columbu and Garau 2017;Ergenç and Fort 2019;Columbu et al. 2019;Lancaster 2021;Vitti 2021;Seymour et al. 2022). The binder commonly used by the Romans was aerial lime, obtained from burning pure limestone, systematically rejecting clayey limestone (Rassineux et al. 1989;Lancaster 2021).
In the Maltese archipelago, no detailed scientific studies have ever been carried out to characterise archaeological mortars and assess their composition, function, or the source of any potentially imported pozzolanic aggregate. The site of a rural structure in Żejtun (35° 51′ 06.3″ N, 14° 32′ 08.6″ E) in the southern region of the Island of Malta (Fig. 1) is one of a few Roman Villa complexes on the island that has survived the effects of heavy urban sprawl and industrial development over the last century . The villa is not monumental in the way many other Roman sites are; construction and modification of structures over several centuries, coupled with the effects of spoliation, destruction, and agriculture in more recent times, have resulted in a transformation in which fragments or material residues are arguably the most significant remains.
Through extensive archaeological excavations, the remains of the Żejtun Villa site have revealed a complex structure characterised by a series of construction and rebuilding activities straddling the Punic and Roman periods . The different construction techniques employed at the site, combined with the careful stratigraphic approach followed in the recent excavations, make it a suitable site for studying ancient building mortars and plasters, a first time such a comprehensive study has been carried out in the Maltese Islands.
The purpose of this study was to establish the type and source of the geomaterials utilised and also to try to determine whether a link exists between the composition of the mortar, its specific use, and the construction technique/s used on site, as well as whether any chronological differences could be detected in the mortars used. Another important aspect was to recognise any technological variations in the mortar mixes that could subsequently allow archaeologists to classify other local mortars found in the archaeological record, with associations linked to their function, source, and period. Such information would serve to produce a database of archaeological and historical mortars and plasters and also inform any conservation interventions needed at this particular site and in similar sites in the Maltese Islands.

The historical context of Żejtun Villa site
The site's cultural history is complex. It was first discovered in 1961 during construction works and was explored further in 1964, and full-scale archaeological excavations were carried out between 1972 and 1976 (Bonanno and Vella 2012). These campaigns did not adopt a stratigraphic approach, and a final report collating the results has never been found. The site languished in a state of near abandonment for three decades until renewed work began in 2006 by the University of Malta.
The 1970s excavations had revealed extensive structural features constructed using a combination of large Globigerina Limestone ashlar blocks and irregular rubble building techniques, immediately thought to belong to a Roman Villa of the rustic type, combining a residential and an industrial area (Bonanno 2018). A series of floors paved with ceramic tiles was also discovered, partitioned by the base foundations of walls built in irregular rubble stones and rendered with decorated wall plaster. To the north of these rooms lay an area retaining stone vats and a counterweight stone associated with the pressing of olives to make oil (Bonanno and Vella 2012;Anastasi and Vella 2018;Anastasi et al. 2022) . A double-entrance cistern partly hewn in the rock and partly constructed out of ashlar masonry was also discovered to the south of the site (Fig. 2).
The renewed excavations between 2006 and 2018 were concentrated on exploring previously unexcavated areas to the south and east of the main structure (Bonanno and Vella 2012;Vella et al. 2017). These campaigns showed that the entire complex was constructed directly upon the bedrock. However, because the ground is uneven and slopes by a gradient of about 1.5 m from north to south, foundations were raised at the south and southeast to the level of the structures to the north and west (Fig. 3). These foundations consist of orthogonal walls, up to three courses high, built to provide a sturdy base for overlying structures (now missing) and to lay floors. Most of the archaeological deposits, which have been investigated in these renewed excavations, belong to the fills that were deposited within the spaces defined by these foundation walls.
The construction and occupation of the site can be divided into at least three broad phases. Additional phases have been recognised based on diverse building activities as well as evidence of structural modifications or abandonment and destruction (Table 1). The pre-villa phase (phase 1) is associated with a vineyard, a 'double-entrance' cistern, and remains of a built structure of unknown function, which have been dated on the basis of pottery style to the Late Punic/Early Roman period (c. fourth to second centuries BCE) . The vineyard and structure were abandoned and covered over by the construction of a villa complex with an area reserved for the pressing of olives to make oil (phase 2). Most of the surviving masonry, floors, and structures at the site are associated with this phase, where pottery has provided a terminus post quem of the end of the second/first century BCE for the abandonment of the vineyard and the construction of the villa. Parts of the residential area have been defined, including the foundations of what would have been a peristyle. Prolonged occupation, maintenance of the villa complex, and a reconfiguration of the olive pressing room are evidenced by the accumulation of debris from earlier structures and features mixed within levelling fills associated with building extensions in the north-east of the site and replastered walls in the residential area. These activities took place over a span of time stretching from the early Imperial period (late first century BCE/first century) to the Late Roman period (late Third/fourth century CE) based on the pottery and coins found (Anastasi 2012;Vella et al. 2018).
The villa appears to have been abandoned and fell into ruin sometime after the fifth century CE, based on the accumulation of abandonment deposits with pottery dating to the fifth to seventh century CE and eighth to tenth century CE. This post−villa phase (phase 3) is also associated with extensive spoliation activity marked by the removal of much of the then visible structure and foundation walls, presumably for reuse elsewhere. At some point following the abandonment of the villa, a large 'cigar-shaped' cistern was cut into the rock, directly through earlier structures; this has been linked to Byzantine or Early Medieval use of the site for agricultural purposes. This cistern eventually collapsed as a result of the poor nature of the bedrock and naturally occurring fissures in the rock and was abandoned shortly after. Rubble filled the abandoned cavity from the mid-eighteenth century, based on the latest−dated pottery identified within its fill.

The geological setting of site
The villa is located within the grounds of a school on the outskirts of the Żejtun village ( Fig. 1). It is built directly upon a limestone bedrock belonging to the miocenic (Aquitaian) Lower Globigerina Limestone Formation (Scerri 2019), consisting of marine bioclastic limestones (medium to fine-grained wackestone-packstone) yellowish to greyish in colour. It is dominated by planktonic micro-foraminifera (Baldassini and Di Stefano 2017), primarily globigerinae. The geological formation is massive, poorly to moderately consolidated, and intensely burrowed (Pedley 1976).

Sampling
A total of 24 mortars and plasters were sampled, including wall joint mortars and plasters, tile-laying bedding mortars, and cistern linings and shuttering layers. The sampling took place in two stages. The first 20 samples were selected from a broad range of archaeological deposits that had been excavated between 2006 and 2014. Many of these deposits were formed in the post-villa phase (phase 3), but were found to consist of residual mortar fragments that are associated with the construction, maintenance, and abandonment of the main villa complex (phase 2). Many of these initial samples consist of residual fragments that were selected from archaeological deposits that were excavated by hand and identification based on the sample's perceived function (Table 2). Similar mortars could not be traced on any of the surviving standing structures, suggesting that these samples belong to the constructions that were  abandoned and destroyed previously. They were also selected based on the ability of the archaeologists to securely date the stratigraphic contexts these samples were located in. Six samples were composed of multiple layers, each of which was in this study divided into sub-samples and analysed separately. 2 A second stage of sampling took place, following the promising analytical results of the first batch of samples. It was decided that an additional four samples would be selected from existing mortars that had survived in situ. Table 2 lists the samples collected and summarises the associated chronological, contextual, typological, and functional evidence obtained through the archaeological record and analyses of materials.

Analytical techniques
A variety of complementary techniques are now commonly used to characterise mortars ) and hence to help reconstruct the source and function of these specific building components in an archaeological context.
In this case, a petrographic study was first conducted using thin sections, which were partially stained with alizarin to highlight the calcite-containing areas (Dickson 1966). The thin sections were viewed under a JENAPOL polarised light optical microscope (PLM) fitted with a Canon 650 digital camera to determine the composition, texture, and microstructure of the different samples (Elsen 2006). The same thin sections were also polished and coated with graphite and examined using a scanning electron microscope (SEM) with backscattered electron detection mode (BSE) and microanalysis using energy dispersive X-rays (EDS) to investigate the microstructure and chemical composition of the binders and aggregates. The instrument used was a JEOL JSM 6400 SEM, with a voltage range of 0.2 to 40 kV and a vacuum of 10−5 Torr; the microanalyzer was an Oxford Inca EDS.
X-ray diffraction (XRD) of powdered (< 53 µm) samples (enriched binder) was used for mineralogical characterisation. This technique supports and confirms the composition described by PLM. The equipment used was a PHILIPS PW 1752 diffractometer, and the conditions used were 40 kV and 30 mA with a Cu anode and graphite monochromator. The diffractograms were obtained at a range between 2 and 65°, and at a speed of 2° per minute.
TGA/DSC analysis was performed using a TA Instrument SDT-Q600 and simultaneously a thermogravimetric analyser General V4.1C 200 DuPont under N2, with a heating rate of 10 °C per minute. The analysis was performed on ground samples of the binder only, previously mechanically separated from the large aggregates of the mortars. TGA/ DSC analysis of the samples gave the percentages of weight losses, working with the following temperature ranges: up to 120 °C, between 120 and 200 °C, between 200 and 600 °C, and above 600 °C. All ensuing reactions are endothermic and are due to weight losses attributable to the dehydration (< 120 °C), dehydration of hydrated salts (120-200 °C), dehydration of hydraulic compounds (200-600 °C), and decomposition-decarboxylation (> 600 °C) of the Ca-carbonate (from the lime binder) and other existing minerals within these mortars. Weight losses between 200 and 600 °C (bound H2O in hydraulic compounds or CO2 deriving from the reactions between Ca-carbonate and silicates according to the following reaction: CaCO3 + XSiO2 ⇒ CaX-SiO3 + CO2 (where X = K, Al, F) (400-600 °C) Garau 2017) and > 600 °C (CO2 release after carbonate decomposition) were used to interpret the hydraulic degree of the mortars Biscontin et al. 2002;Maravelaki-Lalaitzaki et al. 2003;Moropoulou et al. 2000;Genestar et al. 2006;Lawrence 2006;Ergenç et al. 2021).
The chemical composition of the mortars was defined using a portable XRF instrument (THERMO NITON XL3t GOLDD rays) with X-rays generated between 50 kV and 100 μÅ. Three zones of each sample were measured, each comprising 8 mm of fresh surface, with an analysis time of 65 s.

Results
The analysed samples can be divided into five distinct groups (T1-T5) based on the characteristics of the aggregate and binder used (Table 2). This mortar type shows beige to ochre tones and is characterised by a homogenous and cohesive binder (Fig. 4a). Aggregates comprise 60-75% of the sample,with poorly sorted coarse grains up to 6 mm in size. The aggregate:binder ratio is estimated at 3:2 to 3:1. This angular aggregate is composed of Pre-villa/villa Cistern lining limestone (> 90% CaO) and clayey marl (> 30% CaO, > 65% SiO 2 -Al 2 O 3 -K 2 O) fragments containing planktonic microforaminifera (globigerinae and biserial packstone) (Dunham 1962) (Figs. 4b, 5a, and 6a). Many of these marl fragments appear rubefacted (hardground grains) (> 5% FeO and 15-32% CaO) (Fig. 5a, b). In some of the plaster samples of Type 1 (MA-1, MA-2C, MA-3), ceramic fragments (5-10%) measuring < 1 mm in size are present, together with some black particles (ash). The ceramic fragments belong to two types (Fig. 6b). The first type is characterised as very porous, with  a ceramic matrix composed of 17-19% CaO and 5-10% FeO, and contains voids left by dissolved bioclasts (globigerinae), in addition to quartz and K-feldspar grains. The second type is more compact and less porous, with a highly fissured matrix, and consists of < 7% CaO and > 7% FeO. There are frequent grains of quartz and K-feldspar which are < 0.125 mm in size.

Petrographic and mineralogical characterisation (PLM, XRD, SEM-EDS)
The great heterogeneity of the aggregates of this type of mortar is reflected in their varied mineralogical (PLM-XRD) and chemical (EDS) composition. Although calcite dominates, the presence of clay minerals, quartz, and hematite associated with the marls, as well as rubefacted marls and ceramic fragments, are significant ( Fig. 5a, b). The binder surrounding these aggregates is a dense and microcrystalline mass and green-brown colour (Fig. 4b). The chemical composition of the binder is quite variable (Fig. 7). Only one sample was found with a pure calcitic binder (100% CaO, MA-19B). In the remaining samples, the binders have a high content of SiO 2 -Al 2 O 3 -FeO, which varies from 22 to 53%, whereas the CaO content is here 44-83%. The porosity, which can exceed 15-20%, is of the vacuolar, fissure, moldic, and intragranular varieties. The first two occur in the binder; the latter two types are associated with fossil remains in limestone and marl fragments. This group consists of mortars which are white to beige in colour, homogeneous in composition, well sorted, and finer textured in appearance than the Type 1 mortars (Fig. 4c). The aggregate/binder ratio is estimated at 4:1 to 3:1. There are two types of aggregates, with 1:1 ratio. The first are whole or fragmented bioclasts ( Fig. 4d) between 100 μm (micro-foraminifera: Globigerina) and 2 mm (gastropods, echinoderms, bryozoans, brachiopods, and bivalves) in size. These retain their original fibrous-lamellar texture consisting of aragonite and/or calcite (100% CaO) and magnesian calcite (> 92% CaO, > 7% MgO). In the second type, there are sub-rounded fragments of limestone (> 90% CaO) and marl (> 79% CaO and > 20% SiO 2 -Al 2 O 3 -K 2 O) (Figs. 4d, 5c, and 6a) with bioclasts (Globigerina), ranging from mudstones to packstones (Dunham 1962). Like Type 1, this mortar type also contains some rubefacted grains. The presence of some grains of calcium phosphate or hydroxyapatite (> 52% CaO and > 34% P 2 O 5 ) is also noteworthy (Fig. 5c, d). The binder is a brown microcrystalline mass (Fig. 4d), and it has a mainly calcitic composition. The CaO content of these binders (Fig. 7) varies from 99.8 (MA-19C) to 79% (MA-2B), with a very low content of SiO 2 -Al 2 O 3 -FeO (< 19%). The fissure-type of porosity within these binders is very evident; some vacuolar porosity is also present. Intragranular porosity, which exceeds 10-15%, occurs within the limestone and marl fragments. In summary, the Type 2 mortars are similar to Type 1, except for the presence of bioclasts and for the fact that the size of the aggregates does not exceed 2 mm. It is also necessary to note the markedly lower content of SiO 2 -Al 2 O 3 -FeO in Type 2 binders, compared to the higher percentages seen in the Type 1 binders. All the samples of this group come from the outer layers of painted wall renders (1.5-3.0 mm thick), except for two samples (MA-18A and MA-24A) that make up the inner layers of non-painted wall renders (Figs. 4e and 8). These are characterised as being very white, compact, cohesive, and fine texture. They also always occur as a plaster layer above Fig. 6 a SiO 2 -CaO-(Al 2 O 3 + K 2 O) ternary phase diagram of carbonate fragments. Classification according to shape, Circle: Type 1. Triangle: Type 2. Cross: Type 3. Rhombohedron: Type 4. Square: Type 5. b SiO2-CaO-(Al2O3 + K2O) ternary phase diagram of volcanic rock aggregates (Rv) and ceramic fragments (Fc). Classification according to shape, Circle: Type 1. Rhombohedron: Type 4. Square: Type 5 mortars which have aggregates of fossiliferous limestone fragment characteristic of Types 1 and 2 (Fig. 8). These renders show aggregate/binder ratio ranging from 3:2 to 2:1. The aggregates used in these mortars are bioclastic grains consisting of microforaminifera (uniserial and biserial globigerinae) and angular fragments of bivalves, echinoderms, bryozoans, coralline algae, gastropods, and brachiopods, with sizes < 0.5 mm (Figs. 4f and 5e). These aggregates show a CaO content of more than 90% (Fig. 6a). These bioclasts retain their original texture of aragonite or calcite. There are also some grains of limestone and marl with globigerinae. The binder (Figs. 4f and 8) is a dense, dark, and microcrystalline mass of calcite whose chemical composition is > 92% CaO and < 7% SiO 2 -Al 2 O 3 -FeO (Fig. 7). These mortars have a low porosity (< 5%). This group consists of pinkish mortars and a very cohesive matrix (Fig. 4g). They contain sub-angular aggregates, which are heterogeneous in composition, poorly sorted (grain sizes < 4 mm) and represent more than 60% of the bulk of the mortar (Fig. 4h) with an estimated aggregate/ binder ratio of 3:2. These aggregates are of two types. The first type is dominated by ceramic fragments that display varying degrees of vitrification (Fig. 4h); the other contains limestone and marl fragments with preserved globigerinae. Both the latter components are similar to those occurring in Types 1 and 2 mortars. The ceramic fragments are also of two types (Fig. 9a, b). The first is characterised by abundant porosity (irregular and moldic pores), a vitreous paste, and a chemical composition consisting of CaO (15-18%), SiO 2 (> 54%), and Al 2 O 3 (> 20%) (Fig. 6b). Within these fragments, there are grains consisting of clay minerals, quartz, K-feldspars, and calcium phosphate (hydroxyapatite; CaO 50.2%, P 2 O 5 42%), as well as Fe-Ti, Fe-Ca and CaO grains, and other ceramic fragments. The molds within the ceramic fabric have been formed by impressions of globigerinae or other microforaminifera destroyed during firing. In the second group of ceramic fragments, Globigerina with their original preserved shells (100% CaO), quartz, K-feldspars, and calcium phosphate grains are present. The clayey matrix of these ceramic fragments is more abundant than in the previous type, and there is less CaO (< 8%) and Al 2 O 3 (< 20%) and more SiO 2 (> 66%) (Fig. 6b). The texture is uneven, due to mixing of the original ceramic fragments, and is also less vitrified, highly dense, and less porous (however exhibiting fissures and intragranular porosity).
Making up a smaller percentage of the aggregate (5-15%) are limestone and marl fragments containing globigerinae (mudstone to packstone; > 99% CaO), ceramic fragments, rubefacted marl grains, and black particles (< 100 µm), which derive probably from calcination (residual ash) or volcanic activity (volcanic ash). In addition, there are a few lime lumps of incompletely calcined carbonate rock (5-10%) having a chemical composition of approximately 92% of CaO and 7% of SiO 2 -Al 2 O 3 ; these are up to 3 mm in size and have high fissure internal porosity due to shrinkage during the hardening process of the mortars.
The binder shows a dark, greenish-brown, microcrystalline mass with a mainly calcitic composition (Fig. 4j). It also contains some amorphous iron inclusions and has a vacuolar porosity (5-10%). The binder composition falls within a range of 28-61% of SiO 2 -Al 2 O 3 -FeO. The binders of samples MA-5 and MA-17A have the highest concentrations of SiO 2 -Al 2 O 3 -FeO (55-61%) compared to CaO (36-42%). In contrast, the binders of samples MA-13 and MA-15 have high CaO concentrations (65-69%) and in the case of sample MA-12 > 94% CaO (Fig. 7). Figure 10 shows the relation between the reverse trend of the degree of hydraulicity (CO 2 /H 2 O) and the CO 2 ratio of the samples; this trend increases exponentially (Moropoulou et al. 2005;Genestar et al. 2006), where the samples having values below 10 on the y-axis of this graph can be considered hydraulic binders (Moropoulou et al. 2000;Genestar et al. 2006;Uğurlu Sağın et al. 2021). The samples which are located in (CO 2 /H 2 O) below 5 and CO 2 below 20% are considered highly hydraulic binders (Moropoulou et al. 2005).

Hydraulicity based on TGA-DSC analysis
Type 1 mortars exhibit a wide range of weight loss (Table 3). Losses ranging from 1 to 8% are due to the release of hygroscopic water (< 120 °C) and dehydration of hydrated salts and hydraulic components (C-S-H/C-A-H/C-A-S-H) (120-200 °C) at temperatures below 200 °C, bound H 2 O content in hydraulic compounds calcium silicate hydrate/calcium aluminate hydrate/calcium aluminium silicates hydrates (C-S-H/ C-A-H/ C-A-S-H), clay dehydroxylation (200-600 °C) which ranges between 2 and 6%, and 17-39% weight loss attributed to the release CO 2 due to the decomposition of carbonate minerals when heated above 600 °C. Additional exothermic peaks in DSC curves after 800 °C which show the decomposition of carbonate as aggregate are detected in MA-9 sample in Type 1. This type of mortar has a high variability in the CO 2 / H 2 O ratio, of between 3 and 15 (Fig. 10).
Types 2 and 3 are similar, with little weight loss evident below 200 °C (0.5-2.7%) and between 200-600 °C Fig. 9 SEM-BSE images of a Type 4 mortar MA-11 and b Type 4 mortar MA-10 with the presence of two types of ceramic fragments (Fc) and some microfossiliferous marl fragments (Rc). The Fc-1 fragments have more porosity and higher Ca content than Fc-2. SEM-BSE images of c Type 5 mortar MA-13 with volcanic fragments (Rv) and isolated minerals as the main aggregate. The mineralogy of Rv is sanidine and the phenocrysts it contains are andesine (An) and sanidine (S). d Type 5 mortar MA-15 with volcanic fragments (Rv) of sanidine and with isolated minerals of augite (clinopyroxenes) Fig. 10 Graph CO 2 /H 2 O vs CO 2 (%) useful to classify the mortars; CO 2 /H 2 O˂10 indicate the hydraulic character, so the mortars above 10 with higher CO 2 amounts can be discarded as the non-hydraulic true lime mortars; the others indicate hydraulic character of the lime and among those so-called hydraulic lime mortars. Classification according to shape, Circle: Type 1. Triangle: Type 2. Cross: Type 3. Rhombohedron: Type 4. Square: Type 5 (1-4%); however, the loss becomes more significant above 600 °C (34-42%), which indicates a higher carbonate content. These also have the highest CO 2 /H 2 O ratios (between 10 and 31). All samples can be regarded as nonhydraulic binders except one sample (MA-18A) in Type 3 has a lower carbonate component (26.1%) and higher weight loss associated with release of adsorbed moisture and water related to hydrated compounds below 120 °C (Table 3).
Mortar types 4 and 5 are quite similar to each other in hydraulicity. Each has a relatively high weight loss below 200 °C (3-8%), between 200 and 600 °C (3-7%), and a lower loss above 600 °C (12-24%) when compared to the previous types of mortars, and accordingly has the lowest CO 2 /H 2 O ratio between 1.8 and 6. The MA-12 mortar shows incongruous values, with low weight loss between 200 and 600 °C (3%) and very high loss above 600 °C (38%). The CO 2 /H 2 O ratio is high. Type 4 mortars have binders with hydraulic properties, and binders of Type 5 mortars show a highly hydraulic nature, except for MA-12.

Chemical analysis by XRF on bulk sample
Analyses of the whole samples carried out by XRF show a number of differences mainly due to the range of aggregate types present in each mortar group (Tables 4, 5; Fig. 11). Similarly to the TGA/DCS analyses, three clusters are detected. Mortars with mostly carbonate aggregates (bioclastic fragments and fossiliferous limestone and marl fragments) have a higher CaO content (Types 1, 2, and 3: 35-66%), with percentages averaging 52-61% (Tables 4,  5). Of these three types, Type 1 mortars are somewhat distinct from the others, as they display a greater compositional heterogeneity with higher concentrations of SiO 2 (5-19%), Al 2 O 3 (1.5-5.4%), K 2 O (0.2-0.8%), and Fe 2 O 3 (0.3-2.9%), due to the presence of silicate minerals within fragments of marls, rubefacted marls, and ceramics. Mortars with primarily fragments of bioclasts (Types 2 and 3) have a more homogenous, CaO-rich composition (56-64% CaO) with lower concentrations of SiO 2 (0.9-7%), Al 2 O 3 (0.2-2.2%), K 2 O (0.1-0.3%), and Fe 2 O 3 (0.2-0.7%). Mortars rich in marl fragments (Types 1 and 2) also have a higher P 2 O 5 content. Types 4 and 5 mortars have the lowest CaO percentages (26-47%) and the highest concentrations of SiO 2 (15-43%), Al 2 O 3 (2-10.5%), K 2 O (0.7-3.7%), and Fe 2 O 3 (1.8-2.8%). Mortars with ceramic aggregates (Type 4) have a CaO content (46%) which is greater than the CaO content (33%) for mortars rich in volcanic-type aggregates (Type 5); the latter mortars also show an increase in the presence of other elements such as K 2 O (1-3.7%), due to a greater concentration of Na-K feldspars (plagioclases and sanidine). As with TGA/DCS analyses, there are outliers-in this case, these are samples MA-18B and MA-12. The first sample (Type 2, MA-18B) shows clear anomalies, with high concentrations of SiO 2 (17%), Al 2 O 3 (6.7%), K 2 O (0.6%), Fe 2 O 3 (0.9%), and P 2 O 5 (1.1%). This sample contains more fragments of marls and phosphatic grains than limestone fragments, which are generally dominant in other Type 2 mortars. The second anomalous sample (Type 5, MA-12) has a very high concentration of CaO (64%) and relatively low concentrations of the rest of the chemical components-this sample also shows a great compositional heterogeneity of its aggregate, as it has a high concentration of limestone fragments as well as volcanic fragments. The trace element concentrations (Table 6) show that Type 5 mortars, which contain volcanic fragments, contain elements (Bi, Nb, and Rb) not found in the rest of the mortars. On the other hand, Sr and Ba are present within all the five types of mortars; these are elements that are incorporated in the structure of less stable carbonates (i.e., aragonite) and sulfates. Table 5 also presents the concentration means (and standard deviations) of S and Cl of each mortar type. Type 3 mortars stand out, with the lowest concentrations of S and highest of Cl, possibly due to presence of the marine bioclast remains.

Analysis of stable isotopes
Isotopic ratios δ 13 C and δ 18 O distinguish two different groups of mortars. The Group 1 has a lower concentration of δ 18 O that does not usually exceed − 10‰, with a δ 13 C ranging between − 12.3 and − 7.2 0 / 00 ; the Group 2 shows greater variation, with a maximum concentration of δ 18 O − 3 0 / 00 and with δ 13 C ranging between − 17.5 and − 4.5‰ (Fig. 12). Both groups have a δ 13 C/δ 18 O ratio defined by the following lines of regression: In both mortar groups, a trend of continuous enrichment of δ 13 C and δ 18 O is observed. The mortars of Group 1 contain aggregates dominant in limestone and marl fragments (Type 1). The Group 2 contains Type 2 mortars and plasters with bioclasts (Type 3). Mortars with natural pozzolanic aggregates (Type 5) or artificial pozzolanic aggregates (Type 4) are located on both trend lines, with lower values of δ 13 C/ δ 18 O. The mortar of the intermediate layer of sample 2 (MA-2B) is placed in Group 1 and corresponds to an isotopic fractionation which follows the trend observed in Type 1 mortars. In contrast, the MA-3 mortar, which contains limestone fragments, is found in Group 2. Apart from these two atypical samples, the rest of the mortars are grouped within the trends indicated.

Discussion
This characterisation study of the binders and aggregates of the mortars and plasters used at the Żejtun Villa has allowed us to distinguish five distinct groups. All five mortar types contain as aggregate limestone and marl fragments, rich in bioclasts (globigerinae), probably originating from the Globigerina Limestone Formation of Miocene age in Malta (Foresi et al. 2014;Baldassini and Di Stefano 2017). This suggests that the local geology played an important role in the making of these mortars (Degryse et al. 2002;Ergenç and Fort 2019). The high concentrations of planktonic microforaminifera, as well as of rubefacted and phosphatised grains, could have originated from the hardground facies present in the lower part of this geological formation (Foresi et al. 2014;Baldassini and Di Stefano 2017). This implies a potential extraction point for these raw materials within the upper part of the lower member and/or the lower part of the middle member of this geological formation.
These carbonate aggregates especially dominate in Types 1 and 2 of the mortars, which include samples usually present at the inner (or middle) layer of wall plasters. In Type 1 mortars, the dominant aggregates are CaO-depleted (clayey and rubefacted) marl fragments (< 32%) and binders that have a lower CaO content (44-83%), whilst in Type 2 mortars, the dominant aggregates are CaO-rich limestone fragments (> 80%), and binders also have high CaO contents (79-100%). This is reflected in their hydraulicity, with Type 1 mortars being more hydraulic than Type 2 mortars (Tables 2 and 3; Fig. 10), because the aggregates (clayey and rubefacted marls) give additional hydraulicity to the mixtures, according to Moropoulou et al. (2005). The highly hydraulic nature of samples MA-1, MA-24B, and hydraulic nature of MA-8 are also observed on DSC curves and moisture content (Table 3).
The binders of Type 3 mortars are also high in CaO (> 92%; EDS). This is reflected in the thermogravimetric and chemical analyses (Table 3, Figs. 6a, 7, 10, and 11). The characteristics of these aerial lime mortars explain their use for the external layers of wall renders, on Types 2 and 3 mortars. This is an observation even made by ancient Roman architect and writer Marcus Vitruvius (Morgan 1960) and confirmed locally by Deguara (2015). Nevertheless, wall render sample MA18, which is the outer layer of sample MA 18A (Type 3), appears to have a greater hydraulic nature than the inner layer MA-18B, of the same composite sample (Type 2) ( Table 3, Fig. 10). This discrepancy can be explained by the possible non-deliberate use of rubefacted marl in binder preparation for the uppermost layer. Alternatively, if the same binders were used in both layers, the difference would be in the reaction products of marly aggregates and binder included in the enriched binder sample of the outer layer.
The compositional analyses and the perceived function of the samples of Type 1, 2, and 3 show a broad range of mortar mixtures, based mostly on the local availability of raw materials in the Early Roman period. Also, the aggregate/ binder ratios differ in each type of mortar, with the presence of binder being more abundant in the Type 3 exterior mortar (40-50%) and less in the Type 2 interior mortar (20-25%). The binders of these mortars appear to originate from the same limestone deposits, which also contain marls (impure limestones with clays) rich in globigerinae and hardgrounds, and that the Romans used as aggregates, possibly explaining why the lime binders have different hydraulic degrees.
For the sake of accuracy, superimposition of the results coming from the TGA/DSC analyses should also be considered. The H 2 O mass loss between 200 and 600 °C can be attributed to hydraulic phases which can be both calcium silicate and aluminate hydrates (CSH and CAH, respectively) or due to the marly nature of binder. Note that clays may dehydroxylate at various temperature ranges such as 300-550 °C 500-650 °C, 480 °C, 520 °C, 580 °C, and 660-773 °C (Rodriguez-Navarro 2004, Moropoulou et al.1995). There is mass loss between 400 and 600 °C that could be attributed to portlandite, which is also possible to find in historic mortars having experienced various solution and recrystallization processes (Ergenç and Fort 2019); however, no portlandite is detected by XRD. Meta-stable calcium carbonate polymorphs also decompose at 500-600 °C (Morandeau et al. 2014;Ergenç and Fort 2018).
Types 4 and 5 mortars differ from the other mortar types in the aggregate used, but not in the binder type, nor in the aggregate/binder ratio, which is similar to Type 1 mortars (3:2) (Figs. 6b and 7). The main aggregates are in fact siliceous: ceramic fragments in Type 4 and volcanic rock (trachyte and andesite) fragments in Type 5. The ceramic fragments contain the planktonic microforaminifera (globigerininae) prevalent in the local geology of the area and thus clearly indicate local manufacture. The manufacturing of the included ceramics also indicates two different techniques of preparation, as highlighted by their different texture and mineralogical composition. The presence of globigerinae in one type indicates low burning temperature (< 850 °C) whereas their absence together with the presence of a vitrified clayey matrix indicate high firing temperatures (> 850-900 °C). Both these types of aggregates are highly reactive pozzolanic components, giving these mortars extra hydraulicity (Moropoulou et al. 2005). Besides, no volcanic activity occurs on the Maltese Islands, and therefore, the volcanic aggregates do not originate in the Maltese Islands.
Considering the binders of Types 4 and 5, low CaO concentrations (35-42%, Fig. 7) occur in both types (Type 4: MA-17A and Type 5: MA-5 and MA-17B), with a SiO 2 -Al 2 O 3 + FeO:CaO ratio of 1:1 to 2:1-these are therefore strongly hydraulic. 3 Thermal analyses results also (Table 3, Fig. 10) support the hydraulic degree of Types 4 and 5 binders. It is to be remembered that EDS analysis on binders is influenced by the hydrated products of pozzolanic reactions between lime and ceramic and volcanic fragments and dust (such as ceramic dust in MA-17B). The high hydraulic degree of these two types of mortars not only depends on the use of lean limes with different degrees of hydraulicity (Moropoulou 2005) but also by the type of pozzolan used as aggregate. The only non-hydraulic binder sample of Type 5 (MA12) probably had a lower proportion of crushed volcanic rock aggregate with small particle size.
The binders, which are similar to those used in Type 1 mortars, seem to have been manufactured from the same marly facies.
Mortars sampled from the Żejtun site show that the regression lines (Fig. 12) tend to converge towards higher isotopic fractionation, approaching the values of the Globigerina Limestone facies present in outcrops located in the vicinity of the site (Gruszczyńsk et al. 2008).
The presence of marls as the main aggregate in Type 1 mortars (Group 1), together with the presence of limestone fragments, result in an increase in δ 18 O, thus displacing the trend line as defined by Kosednar-Legenstein (Dotsika et al. 2009).
The isotopic concentrations of the bioclast aggregates in Types 2 and 3 influence the overall isotopic concentrations within these mortar types. These shells could suggest that they originated from existing beach deposits in the southeast of the Island. The mean δ 18 O values of these shells is + 1.0‰, with a maximum of + 2.1‰ and a minimum of − 0.3‰ (Prendergast et al. 2013).
Two lines mark the isotopic trend line having different negative isotopic values in both lines (Fig. 12). This is possibly related to the water used to mix the mortars, which could have had a different composition, and probably influenced the isotopic composition of the calcite formed during the carbonation and hardening of these mortars. In addition, the oxygen component of the precipitated calcite could have become heavier on evaporation of the water during setting of the mortar (Dotsika et al. 2009). Regression line 2 in Fig. 12, which falls below that established by Kosednar-Legensteinetal et al. 2008, is defined as the transformation of lime (Ca (OH) 2 ) into calcite (CaCO 3 ) induced by atmospheric CO 2 and slightly isotopic meteoric water. The displacement of regression line 1 (Fig. 12), with values of δ 18 O between − 13‰ and − 5‰ corresponds to enrichment of the calcite in δ 18 O due to the presence of silicate minerals (volcanic rocks and ceramic fragments) and thus indicates that the formation of this calcite is carried out in the presence of atmospheric CO 2 and heavy water evaporation (Dotsika et al. 2018). In addition, hydraulic mortars with a greater chemical reactivity result in the modification of the isotopic content of δ 13 C and δ 18 O (Toffolo et al. 2020). The increase in δ 18 O may be influenced by the presence of saline water (Cangemi et al. 2010;Fort et al. 2021).
The aggregate/binder ratio can also influence the δ 13 C content, since high percentages of aggregates or water in the mix can slow down the carbonation process (Van Strydonck et al. 1986). However, our data do not show any relationship between a high lime content and a high aggregate content, coinciding with the observations of Dotsika et al. (2018).
The presence of sulfur (S) and chlorine (Cl − ) identified through chemical analysis illustrate the presence of saline phases. The Żejtun mortar samples were found to have the following mean values: S = 1992 ± 263 ppm and Cl − = 328 ± 140 ppm. Type 5 mortars that have a higher sulfur and lower chlorine content demonstrate lighter isotope readings in regression line 2 (Fig. 12), so it is possible that the type of salt present in these mortars also influences isotopic fractionation δ 13 C/δ 18 O (Dotsika et al. 2018;Fort et al. 2021). The MA-3 sample (Type 1), also located on line 2, is the sample with the highest sulfur reading (4100 ppm).
Similarly, Type 3 mortars have the highest Cl − (1727 ppm) and lowest S (1007 ppm) concentrations; Cl − may come from beach bioclasts in these mortars.
The chlorides probably originate from the sea 4 ; therefore, some of this content could have entered the mortars even during use and after burial. On the other hand, the presence of S can be partially associated with the volcanic content in the mortars (Silva et al. 2007), together with the aerosols coming from the sea (Fossum et al. 2020) and anthropogenic contaminants. The conducted analyses show that the production technology of mortars, selection of raw materials, and environmental conditions affect the degree of hydraulicity and the chemical composition.
When the samples are classified according to their functions, lining mortars (MA-1, MA-10,  used in water related constructions would have been prepared with marly binder and ceramic aggregate and powder resulting in hydraulic to highly hydraulic properties. The only non-hydraulic sample MA-23 (vat lining) is the exception. Binders of rendering mortars (MA-2, MA-4, MA-6, MA-9, and MA-12) would probably have been prepared deliberately with only limestone to give the non-hydraulic property; however, the inner layers were prepared with marls (MA-2A, , ceramic (MA-17), and volcanic aggregates (MA-7 and MA-13) which increased the pozzolanic activity resulting in higher hydraulicity. Joint mortars (MA-3, MA-12, MA-20) have no hydraulic character, which implies the use of lean lime in manufacturing due to their use in non-water related structures. All bedding mortars for tiled floors (MA-5, MA-14, MA-15, MA-16, and MA-21) were prepared with volcanic aggregate and powder resulting in high hydraulicity. One possible scenario for the deliberate use of those aggregates is that probably builders could prepare the binders with limestone, the only available source at that time, and still needed hydraulicity because they wanted to keep the possible rising damp problem due to high water table under control.

Archaeological significance
In the later first century BCE, the classical historian Diodorus Siculus made remarks on the skilful architectural rendering of residences in Roman Malta (Diodorus Siculus, V.12. 1-4. 2012). Such craftsmanship has also been supported archaeologically with the discovery of numerous fragments of fine architectural stucco, mouldings, and decorative painted wall plaster (Bonanno 2005). However, it is only now that a full scientific assessment of such mortars and their constituent parts has been attempted. This section takes stock of these new scientific data and discusses the findings based on the available archaeological data. The following discussion will focus on two main aspects: (1) the archaeological significance of the source and potential function of the five types of mortars identified and (2) the potential chronological implications of these mortar types that can be gleaned from the archaeological phases of the site.
The samples belonging to Types 1, 2, and 3 include mortars used for masonry construction, renders, vat, cistern linings, and plain and painted wall renders. As discussed above, the composition of the binder and aggregates for these types suggest a source within the Maltese archipelago or even in the vicinity of the site (see 'Discussion'). Limited sections of painted wall render were discovered intact during the early excavations at the Villa, surviving along the interior 'residential' walls of the phase 2 of the villa (Bonanno and Vella 2012). Small fragments still survive, some of which have been studied macroscopically (Deguara 2015), as well as included in this study (i.e., . The layering of plaster visible in a number of these samples also follows the typical and wellknown Roman technique of wall painting or fresco, where slaked lime mortars were applied to a surface, the coarser layers first, followed by successively finer layers, before bold pigments were applied to the smooth outer surface whilst the plaster was still wet. The current analyses confirmed that gradation from coarse and thick to fine and more elaborate layers on samples which exhibited at least two or more layers and which also confirmed previous work by Deguara (2015).
The three samples attributed to Type 4 all contain high concentrations of ceramic aggregate, typical of hard-wearing cocciopesto mortars of the Roman period. Although none of the samples could be directly linked to a physical building element at the site (they were all residual fragments associated with construction phase 2 or later), the composition and archaeological context support their function as fragments of flooring bedding which is further supported by the presence of similarly composed mortars from other Maltese sites of the same period, namely a series of cocciopesto floors dating to the second/first century BCE at the Late Punic/Roman sanctuary at Tas−Silġ, located a short distance from the Żejtun site (Bonzano 2017) and a similar floor adorning the courtyard of the olive press Villa complex at San Pawl Milqi dateable to the Imperial period (Bruno 2009). The presence of abundant ceramic aggregates as found in this present study also supports a local sourcing and manufacture of these mortars, presumably from recycled local ceramic pottery (Anastasi 2019), a conclusion also supported by the identification of the remains of globigerina in the ceramic fragments inside Type 4 mortars. The use of ceramic aggregate is well-known and is documented in Phoenician industrial complexes as early as the sixth century BCE in Lebanon (Orsinger et al. 2020) and Punic domestic architecture in Carthage in the fourth century BCE (Docter 2019). Such material was also particularly widespread during the Roman period both for the manufacture of floors, as well as hard-wearing, water-resistant linings for an array of vats and cisterns (Adam 1994). From an archaeological perspective, the mortars identified as Type 5 are of particular interest. Their volcanic aggregates are not available locally and could therefore have only been obtained by importing them into the islands. The use of volcanic sand-sized particles as temper in local clay amphorae and to line oil-collecting vats is welldocumented in the Maltese Islands (Bruno and Capelli 2000;Bruno 2009). At the Żejtun Villa, however, the majority of the volcanic-rich mortar samples analysed were directly associated with bedding mortars used for floors laid with ceramic lozenge-shaped tiles sometime after the late second/first century BCE. Half of the Type 5 samples came from renders and a fragment of shuttering 5 (i.e., MA-7); thus, volcanic aggregate was not restricted to bedding mortars at the site, but neither was it used for other finishes, such as vat linings (i.e., MA-23 to Type 2), as would have been expected.
Of significance, however, is the composition of the binders across all the samples, including Type 5 mortars. They appear to be similar and likely sourced from the same local limestone material, although it occasionally, but not always, contained marly material. This further supports the hypothesis that the binder materials were indeed sourced, and the mortars, mixed locally, at times being supplemented with imported pozzolanic aggregate (or ceramic fragments) when required, or when these materials were available (Bruno 2009). Importing volcanic sand, whether directly or as a byproduct of maritime trade (i.e., off-loaded ballast), has been documented historically (Burström, 2017) and explains its use as temper for local pottery types (Bruno 2009).
Based on the current archaeological evidence, the mortar samples, both those collected in situ and from later spoliation/ levelling deposits, can be attributed to two possible major construction activities. One related to a pre-villa structure (phase 1) dated by pottery to the fourth to second centuries BCE, which was subsequently partly destroyed and expanded into the villa structure built shortly after the second/first century BCE (phase 2; Fig. 2, Tables 1 and 2). Demolition debris, including fragments of painted and plain wall plaster (MA-8, MA-9, MA-19), and mortars adhering to clearly re-utilised elements (MA-1, MA-20) belonged to structural elements that had already been demolished by the time the phase 2 Roman Villa was constructed, all of which included mortars of Types 1 to 3. All the mortars associated with Types 4 and 5 (with ceramic and volcanic aggregate respectively), together with a broad variety of Types 1 to 3 mortars, were all mixed and applied during or after the construction of the villa (phase 2). This suggests that the locally sourced limestone-based lime for mortars (Types 1-3) was used at the site as early as the Punic period, whilst mortars composed of ceramic and volcanic aggregate were applied to the Roman Villa, both during its initial construction, as well as for later extensions and maintenance works that are evident by the archaeological stratigraphy. Of particular interest, here is the clear introduction of an imported, volcanic aggregate used as bedding mortar to lay all the tiled floors at the site. The date associated with the introduction of this type of aggregate use is also supported by the use of volcanic sand-sized particles in locally produced pottery, which became increasingly common in the Maltese Islands from the third to first centuries BCE onwards (Bruno 2009).

Conclusions
These first analyses of mortar samples from a Maltese archaeological site have identified five distinct mortar types used in the construction and decoration of an ancient Roman building. The analyses demonstrated a broad variety of mortar types that were mostly sourced and manufactured locally and were used for a diverse range of construction activities. A local source for the binders in all mortar types, as well as the aggregates used in four of the five types (Types 1-4), is highlighted. To prepare a hydraulic mortar, local pure limestone formations were often quarried, as Vitruvius noted, and the prepared air lime binder was mixed with reactive aggregate (rubefacted marl, bioclasts, ceramic, and volcanic rock fragments) and reactive additives (ash, ceramic powder). It has been detected that, occasionally, marl-bearing veins in limestone quarries or marl facies abundant in the local geology were also used to prepare hydraulic lime binders probably due to the lack of sufficient pure limestone deposits.
Aggregate/binder ratios vary among all mortar types. Type 2 mortars, most of which are renders, have high amount of aggregate (both limestone and bioclast fragments) and are used in the structures with no water contact. Ceramic and volcanic aggregates were used in similar amounts and less than some samples with limestone and bioclasts of Type 1-3.
The abundant volcanic aggregate in Type 5 mortars used mainly as a bedding mortar for ceramic tiled floors is a nonlocal resource; parallel archaeological evidence in pottery and mortars from other local Roman sites suggests that pozzolanic aggregate was imported to the islands from neighbouring volcanic regions.
This study has established a baseline by which archaeologists are able to characterise mortar composition, as well as its potential use in other archaeological sites, particularly relevant when only fragments of mortars and plasters survive devoid of their original context, as is often the case locally. There is strong potential for further analyses on mortars (and plasters), where samples collected in situ from recently excavated archaeological sites can be gathered and compared with this mortar typology.