High‐performing mortar‐based materials from the late imperial baths of Aquileia: An outstanding example of Roman building tradition in Northern Italy

This study provides the first detailed insight into the composition and properties of structural mortars used in a 4th‐century AD bath complex in Aquileia, the emblematic center of Roman culture in Northern Italy. Eighteen mortars, taken from different structures of the site, and three stone samples from the vaulting opus caementicium have been analyzed adopting a multianalytical approach integrating optical microscopy, X‐ray powder diffraction, X‐ray fluorescence, and scanning electron microscopy coupled with energy‐dispersive spectroscopy. The properties of the compounds are outstanding, as revealed by the formation of hydraulic phases (i.e., Al‐tobermorite and AFm) in most of the samples: the waterproofing capabilities of cocciopesto mortars are remarkable, as revealed by the formation of anthropogenic Al‐tobermorite (5.5 wt%) in pool coating samples; the lightweight of the vaults was guaranteed by the use of porous caementa and pozzolanic volcanic aggregates imported from the Gulf of Naples, as demonstrated by petro‐mineralogical features and chemical analysis of major and trace elements. This is the first proven case of trade in these building materials to the north of the Italian peninsula. These outcomes shed new light on the robust technical expertise of local artisans in Aquileia and indicate that the Cisalpina province was by no means a peripheral reality in the Roman Empire, as far as mortar‐based materials are concerned.


| INTRODUCTION
In the last decades, the interest in the archaeometric investigation of ancient mortar-based materials has increased thanks to the awareness of its potential for the reconstruction of the technical expertise of ancient societies. Most of the research deals with the characterization of antique "recipes," focusing on the pozzolanic aggregates and additives, such as fired-clay fragments, pyroclastic rock clasts, and organic ashes, used to strengthen the cohesion and to waterproof the compounds (Lancaster, 2019). Particular attention is paid to correctly identifying the volcanic aggregates referred to by Vitruvius as harenae fossiciae (Vitr., II, 4, 1) and pulvis puteolanus (Vitr., II, 6, 1-2; V, 12, 2). The former are fine ashes related to the magmatic activity of the Alban and Sabatini Hills, and they were usually used in the mortar mixtures of Rome (D'Ambrosio et al., 2015;Jackson et al., 2007;; the latter are tuff-pumiceous rocks outcropping in the Phlegean Fields out of the Gulf of Naples, which were widely traded and primarily used for the construction of seawater piers all along the ancient Mediterranean (Brandon et al., 2014;D'Ambrosio et al., 2015;Marra, Anzidei, et al., 2016).
The crucial role played by the Romans in the diffusion of concrete technology is well known. The invention of the opus caementicium, an economical, versatile, and durable mortar rubble structure (Ginouvés & Martin, 1985, pp. 51-52), was a decisive achievement for complementing the fervid building activity that Rome undertook along with its rapid expansion in the Mediterranean Sea since the 2nd century B.C. (Mogetta, 2015).
Provincia Gallia Cisalpina (now corresponding to Northern Italy, Slovenia, and Istria) marks a significant gap in this scenario. Over the past few years, the characterization of structural mortars received little consideration (i.e., Baccelle Scudeler & De Vecchi, 2003;Bugini & Folli, 1993Costa et al., 2001), and only recently were the first thorough analytical research articles published (Appolonia et al., 2010;Kramar et al., 2011). An important aspect that can be detected is the substantial absence of any evidence of the use of volcanic pozzolans in the mortar-based materials of the region, with few exceptions that, unfortunately, were not confirmed by adequate analysis. The absence of volcanic pozzolans could be due to a supply shortage of these products in the region, but this datum could be simply biased by the lack of in-depth analytical studies.
Regarding the Cisalpina region, Aquileia, located in today's Friuli Venezia Giulia region (Figure 1a), has been considered a prominent representation of the technical expertise of Roman Northern Italy (Ghedini et al., 2009), and recent analytical research provided new data on the characteristics of mortar-based materials produced in the town in its long history Dilaria et al., 2016Dilaria et al., , 2019Secco et al., 2018).
Since its foundation in 181 B.C., Aquileia represented a bridgehead in the spread of the Roman culture to the north of the peninsula. The town developed into a flourishing urban center during the Imperial Age, enriched with monumental buildings and prestigious private houses. In the 4th century AD, Ausonius (Aus. XI,9,4) considered Aquileia one of the nine most important and extended cities of the Roman Empire, hosting the Imperial court for some periods. Between the 3rd and 4th century AD, the urban walls were frequently renovated to protect the town from recurring sieges and raids, mirroring a period of great political instability. Indeed, less than one century later, starting from Attila's invasion (452 AD) until the end of the 5th century AD, Aquileia faced an inevitable decline and transformation of the urban space, due to the loss of its political relevance.

| THE CASE STUDY
The Late Imperial bath complex of Aquileia is located in the southwestern portion of the Roman town ( Figure 1b). The site was first investigated, between 1922 and 1923, by G. Brusin, then by L.
Bertacchi in 1961, and by P. Lopreato during the 1980s (Rubinich, 2013(Rubinich, , 2014. Owing to its large size (between 22,500 and 25,000 m 2 ), the building was immediately interpreted as being a bath complex. It was named "Grandi Terme" (literally "Great Baths"), as it was probably one of the largest spas in Roman Italy, comparable with the Baths of Caracalla in Rome (Rubinich, 2018). Since the early 2000s, new excavations to understand its plan have been carried out by the University of Udine and are currently underway under the guidance of Prof. Rubinich (Rubinich, 2014, 2020 andreferences therein).
Although the exact date of construction is still uncertain, the baths were probably built in the first half of the 4th century AD, as indicated by a dedication to Emperor Constantine by the praepositi operis, which were involved in the construction of the so-called Thermae Felices Constantinianae (Rubinich, 2013(Rubinich, , 2014. A later dating to the mid-4th century AD is suggested by the discovery by Lopreato of a coin of Constantius II (348-350 AD) in the foundation of a mosaic pavement. Given the size of the complex, it is likely that the construction lasted for decades and was completed by Constantine's successors (Rubinich, 2014(Rubinich, , 2020.
From a constructive point of view, the entire thermal complex was initially (Phase Ia) built onto an extensive trench above wooden pilings fixed into the silty-clayish subsoil of the area. The plan of the complex reflects the typical organization of Roman Imperial thermae ( Figure 1c), with symmetrical rooms along an N/S axis whose fulcrum is the 45 × 22 m frigidarium (room A2), paved in opus sectile ( Figure 2a). Two equally sized halls (32 × 22 m), respectively A1 (North Hall) and A3 (South Hall), interpreted as palaestrae-apodyteria, were located to the north and south of the frigidarium. Six pools, paired two by two, were located to the north, west, and south sides of the frigidarium, while a 20 m wide natatio was placed eastwards ( Figure 2b). The building was also equipped with a sophisticated system of conduits and hydraulic infrastructures for water adduction (Rubinich, 2018). On the west side, the heated rooms were arranged on suspensurae pavements (A12). Bertacchi's excavations investigated the furnace connected to the heating hypocaust system of the caldarium, which was probably closed by a large exedra as revealed by ground penetrating radar prospections. In the NE sector, recent excavations brought to light a large opus caemencitium (Ginouvés & Martin, 1985, pp. 51-52) platform 13 × 16 (or 20) m, called S20 (Figure 2c), consisting of a base layer of coarse marble fragments and an upper part made of bricks in planar rows bonded with mortar. At least two rectangular basins were installed on the S20 platform, bordering a circular one in the center (Rubinich, 2018(Rubinich, , 2020. Most of the building's floors were in opus sectile or mosaic, while little is known about the building techniques and materials, as the masonries were robbed down to their foundations by the massive postantique spoliation activities. The preserved parts showed that most of the loadbearing walls were full-body brick structures (Previato, 2015, pp.  (Rubinich, 2012a).
Vaults constituted the roofing system in opus caementicium with brick ribs that were low light-weighted with porous volcanic rocks. Over time, the baths were subjected to several renovations that mainly involved the decorations of floors and walls. Ancient restorations are split into two phases: the first phase dates back to between the late 4th century AD and the early 5th century AD (Phase Ib), while the second phase dates back to the first decades of the 5th century AD (Phase Ic).
Phase Ib refurbishments are particularly evident in the NE sector, where the S20 platform was obliterated by the construction of two rooms (A16, A19), decorated with a big tesserae mosaic pavement.
In Phase Ic, two rooms decorated with mosaics (A17-A18) were built over the pavement of room A16 of Phase Ib (Rubinich, 2020).
During the first centuries of the early Middle Ages, some rooms of the bath complex were occupied by small family units that buried their dead out of the southern perimeter wall (A13). Later, the building collapsed and became a large open-air quarry. Since the 13th century, after the systematic removal of the ruins, the site was used exclusively for agricultural purposes and it was surrounded by a thick wall, called "Braida Murada," namely, "urban field enclosed by masonry walls" (Rubinich, 2012b).

| SAMPLING
Eighteen mortar samples were collected from different sectors of the Great Baths (Table 1). They were labeled with the name of the site (GTR) according to the function of the structures they come from: (1) two samples (WM_1 and WM_2) come from joint mortars of fullbody brick masonry walls ( Figure 1g); (2) eight samples are of floor bedding screeds (PREP). In detail, two samples come from the screed of the Phase Ia opus sectile of the frigidarium (PREP_5 and PREP_6) and six from mosaics of Phase Ia (PREP_12), Ib (PREP_2, 7, 10, 11) (Figure 1d,e), and Ic (PREP_1); (3) three samples are of structural mortars from the S20 opus caementicium platform (PREF_14, 15, 16); (4) two samples represent pool coatings (CM_1 and CM_2); and (5) three samples come from the collapsed chunks of the vaults. Two of them (VM_1 and 4) have been collected from the opus caementicium portion (Figure 1f), while the third comes from the joint mortar of the brick ribs (VM_2). In the samples with a stratified structure (i.e., mosaic screeds), each layer has been individually analyzed and labeled with a progressive number, proceeding from the topmost to the lower portion of the sample (i.e., PREP_7.1, 7.2, 7.3).
Furthermore, three stone samples, representing the three lightweight caementa lithotypes used in the vaults, were collected.

| Quantitative optical microscopy (OM) and statistical treatment
All mortar and rock samples were subjected to a preliminary petrographic study performed on 30 μm thin sections analyzed under a Nikon Eclipse ME600 microscope.
Mortar analysis was carried out according to the macroscopic and microstratigraphic analytical procedures described in Standard UNI 11176:2006 "Cultural heritage-Petrographic description of a mortar." For each sample (or for each layer in the case of the multilayered sample PREP_7), the rates of binder, porosity, and different aggregates (i.e., fired-clay fraction, sand, etc.) were determined by digital image analysis performed using Image-J software (Schneider et al., 2012). The quantification was performed taking OM-TL scans of the thin sections as the reference; these scans were graphically treated with biochromatic thresholding after transforming the RGB images into 8-bit grayscales (Casadio et al., 2005;Marinoni et al., 2005;Miriello & Crisci, 2006). Porosity and aggregates were quantified separately, while estimation of the binder fraction was performed by subtracting from the total area the sum of aggregates and voids percentages. The size distribution of the aggregate was calculated from the mean of two series of ten manual measurements, representing the diameter of a coarse (usually >2.0 mm) and a fine fraction (<2.0 mm) of the aggregate. The sorting was performed based on the standard deviation (SD) between the mean measurements of fine and coarse aggregates. The mass color was defined T A B L E 1 Summary of mortars/concretes and rock samples collected from the site of the Great Baths with an indication of the structure they come from and the phase they refer to | 641 using Munsell soil color charts (Munsell, 1994). To interpret correlation patterns among samples, the petrographic quantitative data were subjected to a multivariate statistical treatment by principal component analysis (PCA). This is a valid procedure for a rough grouping of samples according to their petrographic features (De Luca et al., 2013Dilaria, 2020;Miriello et al., 2018). PCA was performed on log-transformed petrographic descriptive variables to obtain a small number of linear combinations that adequately describes the original mortar profiles. A series of principal components representing the data set variability were extracted, according to the following parameters: (i) all the variables were considered for analysis and (ii) no limit was set for the number of principal components to be calculated. Samples were then reported in a scatterplot according to the value of the first two extracted components (PC1, PC2). All statistical analyses were carried out using Statgraphics Centurion PRO 18 software.

| X-ray powder diffraction (XRPD)
The mineralogical investigations were performed on the three rock samples and on a selection of representative mortars of the groups identified after the PCA treatment of quantified OM data. QPA profiles were then determined adopting the same methodology as that described in Secco et al. (2019Secco et al. ( , 2020. To properly describe the formation of both geogenic and anthropogenic products, XRPD analyses were carried out on bulk samples (XRPD-bulk) and on the separated binder fraction (XRPDbinder) of mortars. The latter analysis was performed using the in accord with Boynton, 1966); and (c) provide a semiquantitative estimation of major chemical elements of volcanic rock aggregates and caementa, to corroborate X-ray fluorescence (XRF) geochemical analyses.

| XRF
XRF analysis was performed to determine the provenance of the three vaulting caementa by comparing their major and trace chemical elements with the data in the literature. The analysis was performed using a Bruker S8 Tiger WD X-ray fluorescence spectrometer with an XRF radius of 34 mm, equipped with a rhodium tube operating at an intensity of 40 kW, following the method for correcting matrix effects of Franzini et al. (1972Franzini et al. ( , 1975. The material for XRF analysis was collected from the core of pluri-centimetric fragments of the caementa. We mechanically scraped away the interfacial zones between the rock and the binder in order to avoid intrusion in the analysis, as much as possible (Jackson et al., 2014, p. 186   are reunited in a subordinate group (Gr 1b).
Gr 4 groups samples VM_1 and 4, falling at PC2 > 2.5. The peculiar feature of these mortars consists of the presence of volcanic rocks as aggregates.
Finally, PREP_5 and PREP_6, both having PC1 < −2.5, are separated from the other groups and they can be considered as outliers.
The main petrographic characteristics of the groups reported hereafter are summarized in Table S3: (  (Figure 4a). FS clasts fall in the granulometric range of medium to fine sands (Wentworth, 1922 FS and RM (Figure 4b). For these samples, the B/A ratio is found to be around 0.8 or 0.9.
(2) Gr 2 samples are pinkish (2.5 YR 8/3), highly cohesive cocciopesto mortars, with a consistent occurrence of FF and diffused micrometric FP, closely mixed with the binder. The occurrence of FS is negligible ( Figure 4c). The binder matrix has low birefringence colors, likely due to the diffused formation of hydrate products (Pecchioni et al., 2014).
The aggregates represent less than 35% of the samples and they are composed of subrounded medium to fine FS with a variable (c) HI of selected samples in relation to the hydraulicity rates of binder (lime lumps) and matrix, determined according to Boynton (1966); and (d) CI of selected samples in relation to the hydraulicity rates of the binder (lime lumps) and the matrix, determined after Boynton (1966). The values represent the mean of multiple EDS semiquantitative measurements. EmHy, eminently hydraulic; FHy, feebly hydraulic; MHy, moderately hydraulic; SEM-EDS, scanning electron microscopy coupled with energy-dispersive spectroscopy; XRPD, X-ray powder diffraction sorting. Calcareous discarded tesserae chips, used as aggregates, The former is made of homogeneous lime putty with the sporadic occurrence of FS, while the latter is made of a concrete rich in FS represented by medium to fine gravel (Wentworth, 1922), less frequent coarse FF, and sporadic volcanic rock clasts.

| Hydraulic properties
The hydraulic properties of mortar samples were determined by coupling quantitative XRPD-bulk and binder analyses (Table S4) with punctual semiquantitative SEM-EDS investigations (Table S5). aggregates (Katayama, 2010) in a high-pH environment.
Samples from the S20 opus caementicium platform (PREF_14 and 16), as well as sample VM_4, are characterized by the relevant formation of crystalline AFm, but no Al-tobermorite was detected. In VM_4, anomalous high rates of aragonite (5.2 wt%) and vaterite (3.4 wt%) were found. The latter was also documented (2.5 wt%) in PREP_5. These phases represent metastable anthropogenic transitional products formed after decalcification and recarbonation of calcium carbonates along with the pozzolanic reaction of the material (Jackson et al., , 2017Morandeau et al., 2014;Thiery et al., 2007). Finally, the presence of 1.0 wt% phillipsite in VM_4 can be related to the authigenic zeolitization of volcanic aggregates (De'Gennaro et al., 1990. Most of the remaining samples have feeble hydraulic properties, and only WM_2 and PREP_1 can be considered aerial compounds (no AFm phases are detected).
To properly describe and quantify the formation of the hydraulic phases, targeted XRPD-binder analyses were carried out on a limited selection of samples ( Figure 6), which reported the same trend as that obtained from the XRPD-bulk analysis.
The provenance of the three volcanic vaulting lightweight caementa and aggregates detected in VM_1 and VM_4 samples was also determined. These rocks were surely imported in Aquileia, as no volcanic districts are present nearby, as the region is dominated by limestone and dolostone outcrops (G. Carulli, 2006). OM and XRPD investigations on caementa samples were performed for a preliminary determination of their provenance. XRPD data have been recalculated at 100% after the removal of binder-related phases, that is, calcite (Table S6).
VM_4_R is a highly vesicular tephritic lava (Figure 8c1)  The sanidine, documented in all samples and especially in VM_1_N, is a recurrent mineral in the vulcanism of the Roman Comagmatic region Peccerillo, 2005).
The presence of authigenic phillipsite, which is particularly abundant in samples VM_1_G and VM_1_N, is common in the Phlegrean products of the NYT formations (De'Gennaro et al., 1990. This zeolite is frequently documented in Roman mortars and concretes containing Phlegrean pyroclasts Rispoli et al., 2019Rispoli et al., , 2020Stanislao et al., 2011;Vola et al., 2011).
Geochemical analyses performed through XRF provided a better definition of the provenance of the three samples (Table S7).
F I G U R E 9 TAS scatterplots of the clast of lightweight caementa and volcanic aggregates of the vaults. (a) Sample distribution in relation to volcanic rocks' chemistry (after Le Bas et al., 1986); (b) sample distribution in relation to the Phlegrean fields areas of the Campanian Ignimbrite (CI), Neapolitan Yellow Tuff (NYT), and post-NYT events (data from Marra, Anzidei, et al., 2016;Morra et al., 2010;Peccerillo, 2005); (c) sample distribution in relation to the three main eruptive facies of the Vesuvian products' fields (data from Morra et al., 2010;Peccerillo, 2005); and (d) sample distribution in relation to the fields occupied by the products of the Roman magmatic province, Tuscan Magmatic province, and Ischia-Procida-Vivara's volcanoes (data from Avanzinelli et al., 2009;Boari et al., 2009;Marra et al., 2009;Peccerillo, 2005). TAS, total alkali silica All the caementa samples plot in the total alkali silica (TAS) fields (Le Bas et al., 1986) occupied by the slightlymiddle SiO 2 undersaturated volcanic rocks (Figure 9a).
Finally, VM_1_G returns anomalous low values of alkali and silica.
The surprisingly high content of CaO demonstrates the deep development of C-(A)-S-H and CaO-enriched fluids in the sample, as suggested by the results of the XRPD analysis too. The high loss of ignition (LOI) of the sample (19.2) confirms this assumption. Both VM_1_N and VM_4_G samples have a > 3.0 (4.1 and 3.5, respectively), which is indicative of a slight alteration too . This aspect does not make XRF major elements suitable for an in-depth provenance analysis.
To acquire a better major chemical element profile for VM_1_G, we performed six punctual SEM-EDS investigations on three areas of the volcanic glass that appeared unaltered. The mean values for each zone of the sample under investigation are reported in Table S8. In the TAS, they cover a wide area ranging from the trachyte's to the phonolite's fields (Figure 9a). Also, in | 649 this case, therefore, SEM data cannot be considered as conclusive. Also, the geochemical profile of the caementa reported in the TAS cannot be considered as indicative of a specific volcanic district. In fact, as shown in Figure 9d, good matches can also be traced with certain products of Ischia-Procida's and Roccamonfina's volcanoes of the Campanian district (Peccerillo, 2005), as well as with some products of the Latial and Tuscan Magmatic Province Boari et al., 2009). Nevertheless, all the samples do not match with the ultrapotassic products of the Alban Hills (Boari et al., 2009;Marra et al., 2009;Peccerillo, 2005).
Considering the high variability of the TAS, the analyses of Zr/Y versus Nb/Y and Nb/TiO 2 versus Zr/TiO 2 were crucial to confirm the exact provenance of the caementa (Brandon et al., 2014;D'Ambrosio et al., 2015;Jackson et al., 2013;Lancaster et al., 2011;Marra et al., 2009Marra et al., , 2011Marra, Anzidei, et al., 2016;Marra, D'Ambrosio, et al., 2016). Nonetheless, it cannot be totally excluded that the alteration of the volcanic glass, in particular, for VM_1_G, could have led to the depletion of some trace elements too.
VM_1_N presents high ratios of Zr/Y and Nb/Y (Figure 10a). This sample is fully compatible with volcanic products of Campania, but it also matches some pyroclastic deposits of the so-called "Pozzolane Therefore, the information obtained by Nb/Y versus Zr/Y does not further restrict the provenance of VM_1_G. Based on the Zr/TiO 2 versus Nb/TiO 2 plot (Figure 10d), it closely resembles a sample taken from the harbor piers of Terracina (TER-G), which has been associated with post-NYT products (D'Ambrosio et al., 2015;Marra, Anzidei, et al., 2016). The Zr/TiO 2 versus Nb/TiO 2 plot can be considered reliable, as TiO 2 appears to be more stable than Y, even after HCl attack (D'Ambrosio et al., 2015). Therefore, the provenance of sample VM_1_G is surely Campania, while a stronger correlation with the post-NYT products may be tracked only on the basis of the Zr/TiO 2 versus Nb/TiO 2 plot.
Finally, as outlined before, sample VM_4_R presents the peculiar petro-mineralogical and textural features of the Vesuvian lava pillows. In the TAS diagram, it overlaps the Vesuvian products. In the Nb/Y versus Zr/Y plot (Figure 10b Nevertheless, the Vesuvian provenance of sample VM_1_G cannot be completely excluded. VM_4_R, on the other hand, can be safely assigned to Vesuvian lava formations older than 8.9 k.a. B.P.
The origin of volcanic clasts used as aggregate is barely determinable, as they are usually altered due to the pozzolanic reaction.
The unaltered glassy portions of two volcanic aggregates in sample VM_4, labeled VM_4-a and VM_4-b, was detected and analyzed by SEM-EDS (Table S8). In the TAS (Figure 9a The best mortars are the pool coatings that guarantee excellent waterproofing. XRPD data of sample CM_1 report a hydraulic rate three to five times higher than that of the cocciopesto mortars used in all the other water tanks of Aquileia (Dilaria, 2020). This trend does not change if we compare the HI/ CI of CM_1 with that of cisterns' coating mortars from other Roman sites (De Luca et al., 2015;Miriello et al., 2018;Rispoli et al., 2019Rispoli et al., , 2020Secco et al., 2020).
Further distinctions exist in "recipes" for pavements' bedding mortars according to the floor type. Good hydraulic properties have been reported for the sole sample PREP_5, coming from the substrate of the opus sectile pavement of the frigidarium. This is a common feature of the opus sectile pavements of Aquileia (Dilaria et al., 2016;Secco et al., 2018). and marble chips (statumen). Above the screed, tesserae were set on a thin lime putty film (Dunbabin, 1999;Moore, 1968, pp. 281-288), represented by the layer PREP_7.1. This is the typical Roman mosaicmaking technique, which was usually adopted in the Republican and High Imperial Age. On the other hand, during the Late Imperial Age, the mosaics in Aquileia were set on shoddy beddings made of a single layer of friable sandy lime mortar laid over earthen dumps (Dilaria, 2020;Dilaria et al., 2016;Secco et al., 2018). The adoption of this weak making method can be recognized in the making of Phase Ib and Ic mosaics of the northern sector of the Great Baths, which were placed over mortars of the Gr 3. This demonstrates that the restorations of the complex were carried out by artisans with a lower level of experience than those who were working in the original building phase.
Samples PREF_14 and 16, collected from the S20 platform, are characterized by moderate hydraulic properties. Good impermeabilization was necessary for this structure, equipped with an sophisticated system of tanks and pools (Rubinich, 2020). The slight compositional differences among S20 samples (PREF_14, 15, and 16), observed via OM, are probably due to the width and depth of the platform, whose construction required a long time for completion.
These nuances in the composition could reflect the daily preparation of mixtures (Coutelas, 2012 Another recurring aspect in Roman public buildings is the combined use of vaulting lightweight caementa quarried from different zones (Bianchi et al., 2011;. This could be related to the different supplies of raw materials as the construction progresses, but it could also depend on the specific plan for gradually reducing the weight of the vaults at different heights using different materials . Lancaster ( (Buonopane, 1987;Roffia et al., 2009, pp. 188-189), with negligible light-weighting capacity. The high costs for stone transport may have prompted the choice for locally supplied stones even if less efficient for weight reduction (Russell, 2013, pp. 141-200). Vaulting caementa in the Bath of Villeneuve in Frejus are surely imported as they are incompatible with the lithologies of the Massif Central of southern France (Excoffon & Dubar, 2011), but their provenance has not been verified.
The proximity to seaside or fluvial networks and to adequately equipped ports is an important factor affecting the trade of building stones other than marbles and decorative lithotypes (Russell, 2013, pp. 95-140). As far as Aquileia is concerned, the imported rocks could have been shipped from the harbors of Puteoli, Baia, and Miseno (Gianfrotta, 1998) to the river docks of Aquileia or, considering the late dating of the Baths, to the seaport of the neighboring Grado (Rebecchi, 1980). However, the maritime route from Campania to the northern Adriatic Sea is not straightforward, and the Gulf of Naples does not represent the closest area to Aquileia for the procurement of lightweight caementa. In the vaults of the Palace of Diocletian in Split (Croatia), Lancaster (2015, p. 33), reports the use of locally mined porous calcareous tufa, known as sedra. This reference demonstrates that near quarries for the provisioning of lightweight stones were still active in the 4th century AD and easily reachable by sea.
The choice for the supply of the materials from the Gulf of Naples has some alternative explanations: (a) the centuries-old knowledge of the prominent light-weighting capabilities of the Campanian porous rocks, which were largely used in the concrete vaulted buildings of Rome (Lancaster, 2011(Lancaster, , 2015Lancaster et al., 2011;; (b) the provenance of the committee or craftmanship from central-southern Italy; (c) and the little diffusion of regional building materials, as the sedra calcareous tufa, out of the territories in which they were used.
Besides the use of porous caementa for opus caementicium, a finer fraction of the volcanic rocks from Campania was used as  (Ghiotto et al., 2021). Ongoing analysis will be focused on determining their exact provenance.

| CONCLUSIONS
In this study, we outline how decisive the analysis of ancient mortars is to solve traditional archaeological questions, such as the trading of raw materials and the technical expertise of crafts in antiquity. The data reported in this paper are rooted in the interrelationship between archaeology and geosciences and are focused on the understanding of archaeological sites and ancient socioeconomic relations through the investigation of geomaterials such as mortar-based ones.
Adopting this multidisciplinary approach, we are now able to fill the gap, re-evaluating the towns of ancient Cisalpina no longer as peripheral entities of the Empire, but rather as deeply rooted in the technical awareness of the Roman tradition. In fact, the full outcomes of the research demonstrate how remarkable the financial effort for the construction of the Great Baths was: only a high-ranked committee, such as the Constantinian imperial family, could have guaranteed the best workmanship and building materials available on the market at the time.
No other public or private building in Aquileia achieved the same highquality standard as far as mortar-based materials are concerned (Dilaria, 2020). Besides, the superb manufacture of mosaic decorations, which are considered a driving model for the Late Imperial iconography of Aquileia and, in a broader sense, Cisalpina (Novello, 2017), confirms the outstanding features of the building.
Finally, the late dating of the Great Baths provides new scenarios about the transfer of artisans and materials from the center to the north of the peninsula during the Late Imperial Age, which requires further analysis. It is well known that the beginning of the 4th century represented for the main towns of Northern Italy (Aquileia, Ravenna, and Mediolanum in particular) a period of remarkable development and sociopolitical centrality. As stated by Ausonius, Aquileia became one of the largest cities of the Roman Empire, and the Great Baths can be con-