Archaeometry of a Roman Millstone from Santa Maria Arabona, Manoppello (Abruzzo, Central Italy)

: A well-preserved Pompeiian-type millstone fragment was retrieved from the chance discovery of Roman ovens dating to the V-VI century BCE in the area of Santa Arabona Manoppello in Italy. This is the first evidence of an hourglass millstone in Abruzzo. This fragment was analyzed through petrography, geochemistry, statistical analyses, and radiogenic isotopes at the University G. d’Annunzio. The source location of the stones was narrowed down to the areas of Etna, Roccamonfina, and Vulsini due to the petrography, geochemistry, and statistical data elaboration of leuicititic and basaltic rocks from Central Italy and Sicily. The accurate identification of the provenance of the stone used to produce the millstone results in a better understanding of commercial trade routes and Roman entrepreneurship throughout Italy. The correlation between the production site and its stones’ dispersion throughout the Roman Empire is of great interest for understanding the vast network of Roman roads, their manageability of commerce, and the organization of their products to the outlying areas of their Empire and in the case of this discovery, specifically to the area of Abruzzo Italy.


Introduction
In 2013 and 2014, many terracotta fragments and lithic fragments emerged during some agricultural activities and a water pipe excavation, located at 14°3'20.84"E longitude and 42°17'17.70"N latitude in the Santa Maria Arabona area (Manoppello, Pescara province of Abruzzo, central Italy). The site is 500 m SW of an imperial age Roman villa [1].
The archaeological material is erratic, and was found at the ground surface, accumulated along a road and a few more in the nearby vineyard. Judging from the large dimensions of some wall fragments, they must have come from a short distance. After selecting sample fragments, the material was stored at Archaeometry and Microanalyses Center of the University G. d'Annunzio under permission of the Archaeological Authority of Abruzzo-Chieti (code MBAC-SBA-ABR prot. 000300917/04/2014). Most of the fragments are from large dolia, part of the structure of an oven, plus bipedalis (twofoot tile), bessales, tegulae, and imbrices. Another type is a segment of circular bricks belonging to a brick column or suspensurae. The dolia fragments found are rims, bases, walls, and lids. In addition, there are various fragments of fired ceramics and fragments

Macroscopic Description and Archaeological Consideration.
The Romans exploited various lava rock sources present in Italy. The most employed rocks are trachyte of the Euganean Hills volcanic district (Padua Veneto); leucite phonolite from the volcanic complex of Vulsini (Orvieto, Northern Lazio-Western Umbria regions); leucite-bearing trachy basalt from the Somma-Vesuvius volcano (Naples, Campania); tephrite foidites from the Mount Vulture volcano (Potenza, Basilicata); volcanic rocks from Sardinia, mainly rhyolitic ignimbrite from Mulargia, and mugearitic basalts from Etna (Catania, Sicily, Italy).
We investigated three artifacts that refer to millstones: L15 catillus, L14A segment of an hourglass millstone podium, and L14B meta fragments (Figure 1a-e) in relation to the hourglass millstone model structure in the literature [3]. A pioneering study of the Pompeii millstones was done by Peacock [4], who recognized leucite tephrite from Vesuvius and leucitite from the Vulsini volcanic area [5]. The principal rock types of Leucite-bearing rocks cover a vast composition from genuine leucitites (leucite >60% in the AQPF diagram) to phono-tephrites and tephri-phonolites, to phonolites. It is not easy to assign a sure origin to these rock types, which are standard in the Roman region [6] and elsewhere in Italy. Source identification cannot be based on the petrographic description but need a complete petrological study.
Although the origin of the materials used in Roman times to produce millstones is generally known, identifying the quarries remains problematic. In fact, due to nearly 2000 years of continuous volcanic activity from Vesuvius and Etna, many of these quarries were completely obliterated [7]. Another problem is that the volcanic formations do not have lateral continuity and are often relatively small, so that the quarrying sites may have been abandoned after a relatively short period of use. Although numerous techniques allow us to identify lithotypes with precision, their origin often remains uncertain. In addition to more common petrological analysis, it is also possible to use the absolute chronology through K/Ar and Ar/Ar, but this type of investigation is costly, timeconsuming, and consequently not frequently used. Radiogenic isotopic analyses (Sr/Nd) are also expensive and only give a general indication of the volcanic district to which the rock belongs. Bearing this in mind, our investigation was carried out on the samples found at Manoppello, based on exhaustive petrography, geochemistry, and isotopic geochemistry, implemented by statistical processing of the date. The petrological investigation took place on the millstone fragments based on a thin-section microscope optical SEM-EDS study, XRD mineral phase identification, and quantification. A set of major and trace elements was analyzed by multimethod analyses and statistically compared with a large dataset of alkaline rocks from Italy. The rock-type dataset was processed by principal-component and cluster analyses and plotted in conventional rocktype classification diagrams. Sr-Nd isotopic data further restricts the provenance of investigating rock-type geographical origin.

Methodology
Hourglass millstone rocks were studied by Zeiss and Phenom XL SEM optical microscopy, at the University G. d'Annunzio in Chieti, Chieti, Italy). BSE images were obtained using a Phenom XL SEM hosted at the Center for Advanced Studies and Technology at the University G. d'Annunzio of Chieti-Pescara. Powder XRD was carried out using the X-ray diffractometer Rigaku Miniflex II located at the DiSPUTer (Chieti, Italy). The acquisition parameters that were used in the XRD DiSPUTer-Uda laboratory at the University G. d'Annunzio, Chieti, Italy were Cu-K (1.540598 Å) radiation generated at 30 kV and 15 mA in an exploratory interval between 3 and 70 2ϑ, 0.1 steps, and a scan rate of 0.15/s. Once the diffractogram was obtained, background subtraction, and indexing of peaks with semi-quantitative analysis were performed. Mineral identification was performed using Match 3 software. Whole-rock geochemistry was performed at ActaLabs (Activation Laboratories, Ancaster, ON, Canada) by lithium metaborate/tetraborate fusion FUS-ICP for whole rocks and FUS-MS for trace elements. FeO was detected by titration. Isotopic analyses were carried out using Finnigan MAT 262RPQ multicollector and Micromass VG 54E single collector mass spectrometers. Internal precision (withinrun precision) of a single analytical result is given as two standard errors of the mean (2 se). Repeated analyses of standards gave averages and errors expressed as 2 standard deviation (2s) as follows: NBS 987, 87Sr/86Sr = 0.710237+10 (n = 20), 86Sr/88Sr normalized to 0.1194; La Jolla, 143Nd/144Nd = 0.512875 +10 (n = 15), 146Nd/144Nd normalized to 0.7219.
L15 is a porphyritic rock with large polysynthetic geminated leucite crystals and pyroxenes, zoned plagioclase rich in inclusions in a groundmass consisting of fluidly arranged laths of K-feldspar with intergranular cpx, leucite and opaques. The cpx were augites with TiO2 up to 2.35 vol%, olivine (Fo54) containing MnO up to 0.8 vol%, and CaO 0.54 w%. Feldspar were sanidine plus anorthoclase, plagioclase was oligoclase to andesine (Figure 2e  Sample L14A was a rock type not common in peninsular Italy, limited to Sicily and the Etna volcano. On the other hand, leucite-bearing rocks are typical of the Roman region and present in the Campania volcanic district and Roccamonfina and Vesuvius. Starting from this obvious consideration, it is possible to address the research on rocks from the locations mentioned above.

Major Element Geochemistry
Sample analyses are given in Table 1. Major elements of samples L14A, L14B, and L15 plotted in the conventional total alkali-silica (TAS) diagrams classified as trachy basalt, phono tephrite, and phonolite. In the de La Roche diagram (R1-R2) they classified as alkali-basalt, tephritic basanite and trachy-phonolite, respectively (Table 2). Classification using major elements may be inconsistent in the various diagrams for the petrographic observation. Due to the microcrystalline nature of the rocks, because it was impossible to calculate a modal composition, we used the virtual moda deduced by the CIPW norm calculation.
L14A We implemented the norm calculation with an X-ray diffraction analysis (XRD), which only allows a semi-quantitative estimation of the multiphase mineral association. However, the estimation can be calibrated with a detailed petrographic description and the rock's virtual moda (CIPW norm).
L15 XRD gave plagiocalse 31.4 vol%, sanidine 20.1 vol%, nepheline 10.4 vol%, lucite 23.0 vol%, augite 11.6 vol%, and olivine 3.6 vol% A comparison between the sample CIPW norm and the XRD converge towards a more consistent classification of the rock type (Table 2). L14A and L14B classified as phonolitic tephrite, whereas L15 classified as tephritic phonolite in the AQPF diagram, both using CIPW norms and XRD (Table 2). It is not easy, based only on petrographic description and major elements, to make a certain correspondence with already known rock-type sources. For example, Etna rocks are quite different from Roman region rocks, which contain abundant leucite. However, leucite-bearing rocks cover a vast composition from genuine leucitites (leucite > 60% in the AQPF diagram) to phono tephrites and tephri phonolites, to phonolites. Therefore, a statistical treatment can be more effective, but it needs a sufficiently large dataset built based on reasonable indication and the constraints offered by petrography and major element distribution.
In our application, we used an advanced clustering method called HCPC (hierarchical clustering on principal components). This approach allowed us to combine the three standard methods used in multivariate data analyses [22] All the analyses were performed on the statistical software R. We used the descriptive analyses to display the basic features of the data with the advantage of summarizing features of the samples, thus providing simple summaries about the samples and the measurements, which was very useful to present quantitative statistics in a manageable form, and to help to simplify large quantities of data in a sensible way. Differently from inferential methods, descriptive statistics describes what the data shows. A mixed approach using both methods was used in the paper [24]. However, we did not attempt any inferential statistical methods because they might have resulted in misleading conclusions that hide behind the immediate data. The boxplot is a purely graphical representation of chemical variability [25] (Figure 3a). In each boxplot, five parameters depict the chemical distribution of each element: minimum, first quartile (lower side of the box), median (black line in the box), third quartile (upper side of the box), and maximum. In the basic boxplot, the outliers are excluded (represented as empty circles in Figure 3a), and the minimum and maximum are the first and last data points, respectively, that are not considered outliers. Thus, the boxplot is helpful to establish whether the data is symmetric or asymmetric (skewed).
As in Reference [24], we plotted the results of the PCA in a biplot showing the first two principal components that explained most of the variability in the sample. Principal component analysis (PCA) is an exploratory data analysis for dimensionality reduction. It is the process of calculating the principal components and using them to perform a change of basis on the data. PCA is defined as an orthogonal linear transformation that projects the data to a new coordinate system such that the most significant variance by some scalar projection of the data comes to lie on the first coordinate (called the first principal component), the second greatest variance on the second coordinate, and so on, per Jolliffe (2005) [26]. For example, if we consider a matrix of quantitative data such as that defined by the chemical analysis of rocks, with a size of n samples × chemical variable (n × p); and transform them into a new set of variables of the main components, the new set of k variables will have lower dimensionality than the original data set (typically k << p) and, being an orthonormal basis, each principal component is linearly uncorrelated with the others (mutually orthogonal). The first few principal components that explain most of the variance in the original data set were chosen for the graphical representation of igneous rocks analyses [27]. The component that explains the highest variance is reported on the x-axis, while the second highest is reported on the y axis (Figure 3c). Cluster analysis is one of the primary data mining techniques for exploring multivariate data sets. The goal is to divide the sample under examination into several clusters such that points in the same group share similar traits. There are multiple versions of cluster analysis. As far as distance measurements are concerned, in hierarchical clustering, the similarity between the clusters is calculated from dissimilarity measures like the Euclidean (or Mahalanobis) distance between clusters. There are many distance metrics that can be considered for the calculation of the dissimilarity measurement, and the choice depends on the type of data in the dataset. For the use of Mahalanobis distance as an outlier detection method, as in [24], we decided not to use it due to the assumptions needed on the distributive form of the data making that methodology inferential, and thus bearing the possibly of producing misleading results.
The algorithm of the HCPC method we used, as implemented in the FactoMineR package of R, follows these steps: 1. Compute principal components.

Compute hierarchical clustering: hierarchical clustering is performed using
Ward's criterion on the selected principal components. Ward's criterion is used in hierarchical clustering because it is based on multidimensional variances like the principal component analysis. 3. Choose the number of clusters based on the hierarchical tree; an initial partitioning is performed by cutting the hierarchical tree. 4. Perform K-means clustering to improve the initial partition obtained from hierarchical clustering. As a result, the final partitioning solution obtained after consolidation with k-means can be (slightly) different from the one obtained with the hierarchical clustering.
Hierarchical clustering was performed using Ward's criterion on the selected principal components [28]. Ward's criterion is used in hierarchical clustering because it is based on multidimensional variance (Dim1 42.61%; Dim2 23.45%) and reorders the measurements of the metric distance of the data (Figure 3b). Samples from the district of Vico were in cluster 1. Cluster 2 represents the districts of Roccamonfina, Montefiascone, and the Sabatini volcano complex. Cluster 3 is identified by samples from the volcanic district of Etna. In contrast, cluster 4 is comprised of the district of Vulsini, and cluster 5 presents only samples from the Orvieto district. The factor map indicates five main rocktype clusters. The study samples fit with L14B cluster 2 (Roccamonfina, Montefiascone, Sabatini volcanic complex), L14A cluster 3 (Etna), and L15 cluster 5 (Vulsini volcanic complex), respectively (Figure 3d). To associate the statistical cluster with the rock-type of our data, we plotted the reference clusters into TAS [29], R1-R2 [30], CIPW normative, and AQPF diagrams [31] ( Figure 4).

Trace Element Geochemistry
Forty-five trace elements were determined in the investigated samples. Trace elements are plotted in conventional mantle-normalized spider diagrams in order of their compatibility (Figure 5a,c,e) [32]. REE are represented in chondrite-normalized multiphase diagrams (Figure 5b,d,f) [33].
Sample L 14A has a flat or semi-bell-shaped trend of LILE and HFSE ranging from >100mn for LILEs, LREE, and HFSE3+, between 100 and 10 for HFSE4+5+, and lower than 10 for HREE. A slight negative anomaly of P and Ti and a large Pb-positive anomaly is apparent. The plot of REE has a La/Luch equal to 26.6, with a slight Eu negative anomaly (Figure 5a,b). L14B shows a high LILE/HFSE ratio with a decrease of compatible elements. It has negative anomalies of Nb-Ta, P, and Ti, and a moderate Pb-positive anomaly ( Figure  5c,d). L15 shows a general enrichment of all the trace elements and a pattern similar to that of L14B. In addition, it is richer in REEs and has an apparent Th-positive anomaly (Figure 5e,f).

Isotopic Geochemistry
Isotope geochemistry is a robust method to investigate the potential source of an igneous sample. The radiogenic isotope ratio does not change during magmatic evolution and reflects the source composition, which varies according to the different volcanic complexes. For example, Sr/Nd ratios are higher in the Roman region and gradually decrease towards the south [9,10,34,35]. Table 3 shows Sr and Nd isotopes data for each sample with the error standard (2se). The radiogenic isotopes Sr and Nd give a firm confirmation of the suggested provenance sites. Sample L14A fits with the Sr-Nd data from Etna. Sample L14B suggests Roccamonfina and sample L15 fits with the Vico and Vulsini leucitites ( Figure 6). Thus, despite the isotopic analysis that excluded Vesuvius as a source for the rock, the L15 sample from the region of Rome could be attributed to coming from either Vico or Vulsini.

Discussion
Romans exported hourglass millstone on a large scale. The widespread globalization of engineering and industrial techniques due to the road network and maritime transport made it possible to expand the market and offer a series of competitive products at a reasonable price that was worth transporting over long distances. The need to always have loads of ballast made it especially convenient to transport the millstones by sea, which explains why heavy objects could travel, for example, from Sicily to Rome and then be marketed within a reasonable radius of land transport. We cannot exclude that the sea routes were broader, allowing transport even along the coasts of the Adriatic. Furthermore, many rivers that are currently not navigable in Roman times were navigable, such as the Tiber River. Given the weight of these artifacts, it must be assumed that the rock was not only extracted, but also manufactured in or near the same extraction site, although there has hitherto been no evidence of this. So many ways are open to interpretation.
Concerning the Manoppello mills, we may hypothesize some different transportation routes. It is noteworthy that at about 10 km after the confluence with the Paglia river, there was an important crossroads of most of the trade of central Italy for the transport of leucite phonolite from the ancient quarry in the Orvieto area ( Figure 7) [36,37]. This may account for the central collection point for the trade of Vulsini millstones along the Tiber [10,38], which was the natural watercourse for reaching the Tyrrhenian Sea [39][40][41]. However, the road that connects the city of Orvieto to Rome is the ancient Via Cassia, passing through Viterbo. In both cases, we guess that the millstone reached the Roman market and then was transported into central Italy to the east [5,42].
The lava rocks from the volcanic district of Roccamonfina may have travelled along the ancient Via Latina that connects Capua to Rome. From the port of Ostia Antica, built at the mouth of the Tiber River, some of the millstones coming from the volcanic district of the Vulsini could be quickly shipped (as cargo or as ballast) by different routes throughout the Mediterranean, including the distant provinces of the Roman Empire such as Gaul, Iberia, Tunisia, Tripolitania, and Cyrenaica [43]. Finally, the rocks coming from the Volcanic District of Etna may have been transported via the ancient Via Valeria, which connects Messina to Lilibeo and crosses with the innermost alternate route passing through Centuripe, which connects Messina, Catania, Enna and ends at the port of Halesa, known for its commerce [44][45][46]. Regardless of the location of the quarries and production centers, we suppose that the most reasonable hypothesis is that the millstones reached Rome and from there were transported eastwards, reaching Manoppello and other localities along the Adriatic through the ancient Via Valeria (Claudia Valeria), which connects Rome to Ostia Aterni, passing through the city of Alba Fucens and Corfinium (Figure 7). The lack of sure chronological markers makes the Santa Maria Arabona discovery difficult to precisely date. However, due to their heterogeneity, we deduce that the artifacts were derived from the spolia of the Roman Villa and re-used in a bakery. The materials analyzed do not come from a systematic stratigraphic excavation, but these are erratic finds waiting for a possible excavation. Excavation would allow identifying the mill's context in situ to hypothesize what may have been the causes of the re-use of its materials.
Earthquakes became a reference point in time that marked a period of abandonment and dispossession of the Roman Villa. The Majella area and the Chieti province suffered significant destruction during the Roman age with two earthquakes (II century AD and 484-508 AD) with epicenters in the region and another (346 CE) originating in the neighboring area [47]. Manoppello experienced significant damage in more recent earthquakes (1209, 1456, 1706, and 1933), suffering a maximum of XI degrees MCS of intensity in the earthquake of 1706. There is no doubt that the area is prone to elevated seismicity. Remarkable chronicled and archaeological data refer to the 100 AD earthquake (M ~6.2, IX M intensity, probably underestimated), which caused the collapse of buildings of the territorial district of Interpromium, near Manoppello. The account is recounted on a carved stone recycled to build the abbey of San Clemente da Casauria (871 CE). The inscription bears the commemoration of the restoration of Intrepromium's mensa ponderaria (public weighbridge) by two magistrates of Sulmona [48]. The I and IV-V century earthquakes correspond to major collapses or damage in all the archaeological sites of the Majella-Sulmona area. Although not thoroughly investigated, other possible evidence is the tilting of the thermal area of the Villa Rustica of Santa Maria Arabona and the collapse of other residential buildings in the surrounding area (unpublished excavation report by Manuela Rosati, archaeological superintendence of Chieti). However, we are sure that the Roman Villa was inhabited up to the III century CE. It seems the recovery of valuable tools such as the hourglass millstones occurred shortly after the IV-V century CE earthquake to reestablish grain production in an area essentially devoted to agriculture. The earthquake is the marker in time for previous Roman villa dispossession and their later conversion for agricultural use and grain mills. We consider the VI century CE to be the probable age of the bakery, utilizing the studied millstones.

Conclusions
The finding of artifacts related to hourglass mills found at Santa Maria Arabona revealed that hourglass mills derive from various locations and therefore are made from various stones. If the textural rock characteristics such as vesiculated surface with hard minerals and general rock homogeneity are similar, their geochemistry is different. Data consistency indicates that the lapicidae were able to identify and select similar lithotypes by empirical criteria based predominantly on the modal rock composition. It is not plausible to identify the source of the rocks using just petrography and conventional classification diagrams. More sophisticated statistical methods may help, as the inconsistencies need a double-check. Combining all the possible petrological features such as petrography, geochemistry, and isotopic geochemistry with statistical analyses, it was possible to hypothesize a probable origin of the rocks from three different localities-Etna, Roccamonfina, and very probably Vulsini. Samples L14A and B are consistent with Etna and Roccamonfina rocks, respectively. Sample L15 has some less obvious geographical locations. However, excluding the isotopic data, which do not allow a specific distinction between Vulsini and Vico, all the other data point to an origin from Vulsini and with leucitophyre from Sugano (Orvieto area). This paper offers an example of which methodologies may ensure a correct estimation of possible rock sources, overcoming petrographic uncertainty.
The most obvious result is that hourglass mills were widely commercialized, responding to market requests independently from the distance between the source and the destination. Santa Maria Arabona, being on a main consular Roman street, th Tiburtine, suggests that the hourglass mills arrived from Rome, where they converged from the Roman region, Campania, and Etna. The market request minimized the high cost of transportation, indicating economic consumption influenced by the demand for better and more luxurious products. The archaeological context suggests the late re-use of the mills in a rural context with extensive evidence of the re-utilization of the republican villa. The presence of a bread oven, dolia, and ordinary fired ceramics is not undoubtedly related to the villa and may belong to an edifice whose basement is not exposed and requires excavation. The oven, dolia, and hourglass mills were unearthed while ploughing vineyard land and were located above the deck of a partially buried edifice. A major earthquake in the IV-V centuries CE may be a starting point in time for the reutilization that occurred, dating the artifact to the V or VI century CE.