The Volsci Volcanic Field (central Italy): eruptive history, magma system and implications on continental subduction processes

Here, we report on the Quaternary Volsci Volcanic Field (VVF, central Italy). In light of new 40Ar/39Ar geochronological data and compositional characterization of juvenile eruptive products, we refine the history of VVF activity, and outline the implications on the pre-eruptive magma system and the continental subduction processes involved. Different from the nearby volcanic districts of the Roman and Campanian Provinces, the VVF was characterized by small-volume (0.01–0.1 km3) eruptions from a network of monogenetic centers (mostly tuff rings and scoria cones, with subordinate lava occurrences), clustered along high-angle faults of lithospheric depth. Leucite-bearing, high-K (HKS) magmas (for which we report for the first time the phlogopite phenocryst compositions) mostly fed the early phase of activity (∼761–539 ka), then primitive, plagioclase-bearing (KS) magmas appeared during the climactic phase (∼424–349 ka), partially overlapping with HKS ones, and then prevailed during the late phase of activity (∼300–231 ka). The fast ascent of primitive magma batches is typical of a tectonically controlled volcanic field, where the very low magma flux is a passive byproduct of regional tectonic strain. We suggest that the dominant compressive stress field acting at depth was accompanied by an extensional regime in the upper crust, associated with the gravity spreading of the Apennine chain, allowing the fast ascent of magma from the mantle source with limited stationing in shallow reservoirs.

The Tyrrhenian Sea margin back-arc domain has been affected by extensional tectonics since Pliocene, as a consequence of the retreat of the west-directed Adriatic slab, whereas the central part of the Italian peninsula displays regional uplift that has built the Apennine belt (e.g., Doglioni et al. 1994;Hippolyte et al. 1994;Sartori et al. Carminati et al. 2014;Beaudoin et al. 2017). The development of NW-and NE-striking high-angle normal faults has produced a Horst-and-Graben structure extending from Tuscany to Campania, along which several potassic volcanic districts were active since 0.8 Ma to present (Peccerillo 2017 and references therein).Volcanic activity in the Roman Province, including major caldera-forming explosive events, has occurred since ca. 590-565 ka at the Vulsini, Monti Sabatini and Colli Albani volcanic districts ( Fig. 1), and it seems to have ceased progressively from NW to SE since ca. 111 ka (Vulsini), 95 ka (Vico), 70 ka (Sabatini) and 36 ka (Colli Albani) (Barberi et al. 1994;Perini et al. 2004;Palladino et al. 2010;Marra et al. 2016Marra et al. , 2019Marra et al. , 2020a. To the south of the VVF, the activity of the Roccamonfina volcano, although poorly constrained geochronologically, seems to have spanned 696 ± 49 to 148 ± 18 ka with potential younger activity (i.e., 53 ± 27 ka, Radicati di Brozolo 1988; Rouchon et al. 2008).
Unlike the other volcanic districts of the Tyrrhenian margin, the VVF eruptive centers occur within the inner Apennine mountain range (Fig. 1). Another peculiarity of the VVF volcanism is the lack of a major volcanic edifice or a central caldera. Instead, the rapid ascent of relatively primitive magmas from the mantle source has produced a network of monogenetic magmatic and phreatomagmatic centers Centamore et al. 2010). The reported occurrence of mafic volcanic rock types representative of different magma suites (i.e., ultrapotassic, calcalkaline and kamafugitic; Boari and Conticelli 2007;Frezzotti et al. 2007;Boari et al. 2009;Centamore et al. 2010;Koornneef et al. 2019) is another intriguing aspect of this volcanic field. These "bullet eruptions", fed by fast ascending magma batches, open a window on mantle source in continental subduction settings.
Prior to this study, the reconstruction of the VVF eruptive history was based on an old set of K-Ar ages spanning 0.7-0.1 Ma (Fornaseri 1985;Basilone and Civetta 1975). However, they lack analytical details and, as for other age determinations performed in the 1970-1980s, their accuracy is questionable (e.g., Karner and Renne 1998). The only recent 40 Ar/ 39 Ar investigation ) was essentially based on the dating of small-volume lava flows and thus cannot be considered as representative of the whole temporal and compositional spectrum of the VVF activity. We present a new set of 40 Ar/ 39 Ar dates of VVF eruptive products, integrated with petrographic and microchemical analyses, in the frame of stratigraphic, geomorphologic, structural, and seismological observations. Our work contributes to a comprehensive reconstruction of the VVF eruptive history and feeder magma system, with respect to the possible geodynamic setting. This provides implications on the broader volcano-tectonic context of the Tyrrhenian Sea margin of central Italy and related hazard assessment.

Geological-structural setting
The study area (Fig. 1) is part of the central Apennines, a mountain chain made up of terrains belonging to different Meso-Cenozoic paleogeographical domains developed after Late Triassic on the southern margin of the rifting Tethys (Parotto and Praturlon 1975;Bernoulli 2001;Cardello and Doglioni 2015). The study area is located between the Simbruini-Ernici-Cairo mountain ridges and the Volsci Range ( Fig. 1), which belong to the Latium-Abruzzi neritic carbonate domain (upper Triassic-middle Miocene), covered by middle Miocene to early Pliocene syn-orogenic siliciclastic deposits (Centamore et al. 2007 and references therein). Overall, the central Apennine fold-and-thrust belt is characterized by tectonic structures showing northeast polarities due to Late Tortonian-Early Messinian compressional tectonics. The Simbruini and Ernici Mts. are built up of several NW-SE striking imbricate thrust sheets that overthrust onto Tortonian-lower Messinian terrigenous deposits within the Latin and Roveto valleys (Fig. 1;Devoto 1967;Parotto 1971;Fabbi and Santantonio 2018). The Volsci Range is organized in several thrust sheets defined by fore-and back-thrusts, cross-cut by a system of conjugated synthetic and antithetic Quaternary normal faults (Cavinato et al. 1994;Cosentino et al. 2002;Sani et al. 2004;Parotto and Tallini 2013). The Latin Valley is a graben comprised between two ~ N-S trending major tectonic lines with polyphase activity, namely the Ortona-Roccamonfina (or Volturno-Sangro Line;Locardi 1982) and the Olevano-Antrodoco-Sibillini lines (e.g. Pizzi and Galadini 2009). The latter can be considered as the positive re-activation of pre-existing continental break-up structures. The Ancona-Anzio line (Castellarin et al. 1978) results from a transpressive thrust system (Cipollari and Cosentino 1991;Koopman 1983;Lavecchia 1985;Butler et al. 2006, and references therein).
Locally, besides the dominant NW-and the minor NEtrending structures, the study area has also been affected by N-to NNE-and E-to ENE-striking high-angle faults with strike-slip kinematics up to Middle Pleistocene (Sani et al. 2004;Centamore et al. 2010;Cardello et al. 2020). The strike-slip tectonics have been previously associated with the volcanic centers of the VVF, whose compositions show a mantle origin, thus suggesting a lithospheric depth for these structures (Acocella et al. 1996).
In the mountain areas adjoining the Latin Valley, the compressional structures are cross-cut by a nearly coaxial system of mainly Late Pliocene-Quaternary high-angle extensional faults (Cardello et al. 2020). These faults likely controlled the formation and growth of intramountain basins. An example of extensional tectonics is represented by two main morpho-structural elements within the Latin Valley ( Fig. 1): the Liri Valley to the SE, and the Sacco Valley to the NW corresponding to the middle sector of the NW-trending graben also known as "Middle Latin Valley".
The most recent tectonic phase along the axial zone of the Apennines (i.e., Marsica; Fig. 1) is expressed by a set of mainly NW-trending normal or normal oblique fault systems (Calamita and Pizzi 1994;Lavecchia et al. 1994;Ghisetti and Vezzani 1999;Piccardi et al. 1999;Morewood and Roberts 2000;Blumetti and Guerrieri 2007;Pizzi and Galadini 2009). The present-day stress field is characterized by a prevalent, ~ NE-SW oriented, extension, derived by stress indicators (Montone and Mariucci 2019). The chronological range of the extensional tectonics is poorly constrained, although it is generally assigned to the Quaternary (e.g., Sani et al. 2004). However, an intensive extensional phase related to the formation of the Liri lacustrine basin (Lirino Lake Synthem) occurred during the Middle Pleistocene (Centamore et al. 2010).

Seismic features
Here, we present a brief summary of the seismicity affecting the Latin Valley and surrounding area in the period 2009-2013 (Frepoli et al. 2017), implemented by new data recorded over the period 2014-2019, aimed at providing a picture of the present stress field, compared with the Middle Pleistocene tectonic regime. Figure 2 displays the seismicity (M L 0.4-4.7) recorded by the permanent (Seismic National Network, RSN) and temporary stations of the Istituto Nazionale di Geofisica e Vulcanologia (INGV), and the regional networks of Abruzzo (RSA; De Luca 2011) and Molise (see Frepoli et al. 2017 for the employed methodology). All the waveforms of this seismicity were re-picked to obtain both more accurate arrival times of P-and S-waves and a larger number of first-motion polarities with respect to the routine procedure for the compilation of INGV seismic bulletins. Four NE-striking cross-sections in Fig. 2 show the distribution of seismicity at depth.
Seismicity concentrates within the axial part of the Apennines. Here, the most relevant seismic sequence occurred on February 16th, 2013, in the Pescasseroli-Sora area, with a magnitude M L 4.7 mainshock and with ~ 350 aftershocks. The hypocenter depth of this sequence extended down to 18 km, with a pronounced maximum between 8 and 15 km. The most recent seismicity (period 2014-2019) overlaps the previous one (period [2009][2010][2011][2012][2013] in the same areas. In contrast, seismicity fades out along the Volsci Range, although present in the Pontina Plain (W of the Volsci Range area). Seismic events are scattered within the Latin Valley and, in particular, are rare in the VVF area. Here, seismicity affects 1 3 the upper portion of the crust and is linked to the widespread NE-SW oriented extensional regime, as shown by the normal faulting focal mechanism of the most constrained solutions of the dataset (Fig. 2).

The Volsci Volcanic Field (VVF)
Based on the overall areal distribution of the monogenetic centers, extending (from east to west) from the Sacco River Valley, through the Volsci Range, as far as the Pontina Plain (Fig. 2), and following Cardello et al. (2020), here we define the previously known Monti Ernici or Middle Latin Valley volcanoes as the Volsci Volcanic Field (VVF), consistent with the geographic and historical pertinence of ancient Italic populations. The VVF extends in two main geomorphic settings (Fig. 3) Accordi et al. 1966;Alberti et al. 1975;Centamore et al. 2010). Radioisotopic ages from literature (in italics those with poor analytical constraints) and the present study are reported. Areal vs.
temporal migration of the volcanic activity is shown in insets i-iii. b Cross-correlation chart showing the geochronological constraints to the sedimentary fill of the Liri Basin based on distal volcanic deposits intercalated in the fluvial-lacustrine successions The VVF terrains are mostly pyroclastic fall and surge deposits from low-intensity and low-magnitude explosive (i.e., Hawaiian-Strombolian and phreatomagmatic) eruptive events, and subordinate lava flows and plateaus from effusive activity (Centamore et al. 2010;Cardello et al. 2020 and reference therein). More specifically, based on well-established field criteria, loose to welded, well-sorted, clast-supported beds of moderately vesicular, scoria, and spatter lapilli and blocks are interpreted as the products of fallout from Hawaiian-Strombolian fountains. Stratified deposits of fine to coarse ash and lapilli, characterized by poorly vesicular juvenile scoria, abundant lithics of the sedimentary substrate, with possible occurrence of accretionary lapilli and a variety of bedforms (e.g., planar-and cross-lamination, lowangle dunes), are attributed to phreatomagmatic base surges.
Overall, the geologic structural setting likely controlled the location of volcanic activity and the erupted magma compositions, as well as the style of eruptive events. It appears that deep-seated high-angle faults primarily controlled magma ascent from the source, while the extensional faults of the Meso-Cenozoic carbonate substrate determined the distribution pattern of eruptive centers, and favored the repeated, relatively fast ascent of primitive, small magma batches, feeding small-scale events (Centamore et al. 2010;Cardello et al. 2020). In particular, the southern part of the Sacco River Valley (south of Ceccano) is a continental basin developed in an asymmetric graben-like structure bounded by the ENE-striking Ceccano Fault to the north and a NW-striking normal fault system to the south, bounding the Volsci Range and the Middle Latin Valley Graben, onto which several eruptive centers are seated (Fig. 3). Another alignment of eruptive centers (including the Tecchiena, Selva dei Muli, Celleta and Giuliano di Roma centers; Fig. 3) occurs along a NNE-striking lineament. identified as a right-lateral fault by Acocella et al. (1996) and Sani et al. (2004).
In addition, the wall rock architecture, characterized by rock types with highly different degrees of permeability (i.e., platform carbonates vs. pelitic-arenaceous, volcanic layers), and the local presence of closed/suspended aquifers, favored the explosive interaction of mafic magmas with groundwater. The different types and proportions of lithic inclusions in the phreatomagmatic deposits provide indications on the depths of magma-wall rock interaction, which took place at multiple levels ranging from relatively shallow Miocene siliciclastic terrains (e.g., Frosinone formation; Centamore et al. 2010) to deeper Meso-Cenozoic carbonates (Cardello et al. 2020). Besides the predominant magmatic or phreatomagmatic character of specific eruptive centers, typically, individual eruptions from VVF monogenetic centers were often characterized by transitions from mildly explosive, magmatic phases (i.e., Hawaiian-Strombolian, with possible associated effusive activity) to phreatomagmatic phases with higher degrees of magma fragmentation and explosivity. The low viscosities of VVF mafic magmas and the decreasing magma discharge rates in the late stages of individual eruptions favored explosive magma-water interaction.
Overall, volcanic terrains (including their reworked equivalents) are scattered over an area of nearly 600 km 2 , covering ~ 100 km 2 . Relic morphologies (partially preserved crater rims and volcanic edifices) and lithofacies associations of erupted products indicate a network of rooted monogenetic centers, including scoria and spatter cones, maar-tuff rings and tuff cones. The VVF was thus characterized by areal volcanic activity through time, with the partial exception of the Pofi composite volcanic edifice, resulting from clustered eruptive centers.
An integrated Unconformity-Bounded Stratigraphic Unit (UBSU) and lithostratigraphic setting for the eastern part of the VVF is provided by the 1:50,000 Geologic Map of Italy, Sheet 402-Ceccano (CARG Project; Centamore et al. 2010), based on the fragmentary stratigraphic relationships reconstructed in the field, rock type compositions and a few new geochronological data. Up to now, available geochronological data did not allow for defining the onset and timing of the VVF eruptive activity. Based on previous geochronological data (see the following section), volcanic activity developed in the time span of ca. 700-100 ka, and thus was roughly contemporaneous with that of the nearby potassic volcanic districts of the Roman Province (Colli Albani and Sabatini; Gaeta et al. 2016;Marra et al. 2014Marra et al. , 2016Marra et al. , 2019Marra et al. , 2020aSottili et al. 2010).
Compositionally, the VVF-erupted magmas are poorly to slightly differentiated. According to available data, mostly referring to lava rock types, they range from K-basalts to K-foidites in the TAS diagram ). A wider compositional dataset, also including a variety of pyroclastic products, is reported in Centamore et al. (2010) for the eastern VVF centers (Ceccano and Pofi areas).
Previous studies (Basilone and Civetta 1975;Civetta et al. 1979Civetta et al. , 1981Frezzotti et al. 2007;Boari et al. 2009) recognized four main types of parental magmas and related evolutionary series: (1) ultrapotassic-kamafugitic (KAM), (2) high-potassic (HKS); (3) shoshonitic (SHO) and (4) sub-alkaline (CA). To favor the comparison with the new geochronological data, obtained exclusively from pyroclastic products, in the following, we group the literature data in HKS and KS (potassic) rock types, including KAM plus HKS and SHO plus CA, respectively. The rationale of our choice is in the small size of the analyzed juvenile scoria clasts that generally can be compared with the lava flow compositions exclusively on the basis of textural and microchemical features determined in thin section. The HKS group comprises leucite-bearing lava flows, different from the KS lavas, which contain plagioclase and commonly lack leucite. Compatible with the usually low 1 3 porphyricity degrees (and variably altered groundmass) of juvenile scoria clasts, the HKS vs. KS distinction is also applied to pyroclastic products, based on the leucite vs. plagioclase occurrence (both from literature data and new petrographic observations).

Geochronology of the VVF
The published ages for the VVF eruptive products are summarized in Table 1. Available whole rock K/Ar ages should be considered with caution, because of their large uncertainties and the possible bias towards older ages due to contamination from xenocrysts with inherited 40 Ar. To compare the geochronologic dataset discussed later in this paper, we re-calculated the 40 Ar/ 39 Ar ages reported in Boari et al. (2009) (mostly obtained from lava groundmass) according to the standard age used for our data (i.e., 28.201 ± 0.046 Ma for FCT sanidine, Kuiper et al. 2008; equivalent to ACs at 1.1848 ± 0.0012 Ma, Niespolo et al. 2017). For further discussion of accuracy and reliability of literature ages, see "New geochronologic constraints".

Sampling
To provide 40 Ar/ 39 Ar age constraints to the widespread VVF phreatomagmatic activity, we sampled the products of rooted eruptive centers in the Volsci Range backbone. Of these, only five samples (PAT-1B, PAT-5, SUP-1, GdR-1, GdR-5; Fig. 3) contained datable K-bearing mineral phases and were processed. Moreover, we dated four samples from the eruptive centers of the central sector of the Middle Latin Valley (ARN-1, POF-3, CLG, MOS-2; Fig. 3). In addition, we dated a pair of volcaniclastic horizons (CA-C1, CA-CGT) interbedded in the uppermost portion of the Liri Basin lacustrine deposits (Colle Avarone locality, nearby Ceprano; Fig. 3), to put constraints to the final stages of sedimentation within the Liri Basin, hosting the remains of a calvarium belonging to the "Homo heidelbergensis" species (i.e., namely Homo cepranensis, Ascenzi et al. 2000;Nomade et al. 2011, Manzi 2016).

Petrographic and microchemical analyses
Textural aspects of the recovered samples were first analyzed in thin section at the optical microscope and successively at the Scanning Electron Microscopy (SEM), using a FEI Quanta-400 at the Earth Sciences Department (Sapienza University, Rome, Italy). Phase compositions were analyzed at the Consiglio Nazionale delle Ricerche-Istituto di Geologia Ambientale e Geoingegneria (CNR-IGAG, Rome), by a Cameca SX50 electron microprobe equipped with five wavelength dispersive spectrometers (WDS). Quantitative compositional analyses were performed using 15 kV accelerating voltage, 15 nA beam current and beam diameter of 1 µm. As standards, we employed metals for Mn and Cr, jadeite for Na, wollastonite for Si and Ca, orthoclase for K, corundum for Al, magnetite for Fe, rutile for Ti and periclase for Mg. Counting times were 20 s for elements and 10 s for backgrounds. Light elements were counted first to prevent loss by volatilization. The PAP correction method (Pouchou and Pichoir 1991) was used. The essential features of the sampled products are reported in Tables 2 and 3; further details on sampled outcrops are provided in Supplementary File #1.

Ar/ 39 Ar analyses
Samples for 40 Ar/ 39 Ar analyses were prepared at the Laboratoire des Sciences du Climat et de l'Environnement facility (CNRS-CEA, Gif-sur-Yvette), France. After having been crushed, sieved and cleaned in distilled water in an ultrasonic bath, the clean portions coarser than 840 µm (> 20 mesh) and 840-600-µm sized (30-20 mesh) were selected to extract K-bearing crystals suitable for dating (feldspars and/or leucite). For each sample, at least 50 unaltered crystals were carefully handpicked under a binocular microscope after magnetic separation. To eliminate potential adhering groundmass residues on the selected crystals, the latter were finally leached with a 5-7% HF acid solution and cleaned several times using distilled water in an ultrasonic bath. Two distinct irradiations and analyses have been performed, as follows. Samples CA-C1, CA-CGT, MOS-2, ARN-1 and CLG were irradiated in the Cd-lined, in-core CLICIT facility of the Oregon State University TRIGA reactor and analyzed at the Berkeley Geochronology Center facility (BGC; California, USA), using a MAP 215-C mass spectrometer (MAP 1), following procedures described in Giaccio et al. (2019). Samples GDR-1, GDR-5, SUP-1, POF-3, PAT-1B and PAT-5 were analyzed at the WiscAr Laboratory of University of Wisconsin-Madison (USA), using a Noblesse 5-collector mass spectrometer, following procedures described in Jicha et al. (2016). All ages are calculated according to the fluence monitor age of Alder Creek sanidine (ACs = 1.1848 ± 0.0012 Ma ) and all ages are reported to the precision level of 2σ standard deviation. Detailed procedures and full methods for 40 Ar/ 39 Ar data acquisition are reported in Supplementary File #2.

Field aspects of the sampled volcanics
We conducted a new geomorphological study, based on combined analyses of satellite images and 1:25,000 topographic maps, which allowed us to outline the relationships among the identified morpho-structural lineaments and the locations of the VVF rooted eruptive centers (Fig. 3). The latter include monogenetic eruptive centers of magmatic (Strombolian scoria cones with associated lava flows) and phreatomagmatic (tuff ring relics) origin, located both in the Volsci Range carbonate mountain backbone setting (i.e., the Fosso di Monteacuto, Villa S. Stefano, Patrica, and Supino phreatomagmatic occurrences) and in the Middle Latin Valley graben (i.e., the Arnara scoria cone and the Pofi composite edifice, resulting from coalescent scoria cones and tuff rings, with associated lava flows). We notice that these eruptive centers, within the general VVF volcano-tectonic framework, are preferentially located along major morphologic breaks, which are associated with the occurrence of major faults cross-cutting all the units of the rock substrate. Age values with uncertain accuracy and reliability of analytical procedures (see text) are reported in Italics References: (1) Cortesi in Fornaseri (1985); (2) Basilone and Civetta (1975) Typically, phreatomagmatic deposits share the following lithofacies association.
Along with planar to cross-bedded, occasionally accretionary lapilli-bearing, ash deposits from base surges, alternating with thin, scoria lapilli fallout layers (Fig. 4a, b, e), a peculiar lithology is represented by the peperinotype pyroclastic current deposits (Fig. 4c). The latter mostly occurs in the eruptive centers of the Volsci Range mountain setting (e.g., Fosso di Monteacuto, Patrica, Patrica NE, Villa S. Stefano). This term derives from the Italian word "pepe" for pepper and applied to phreatomagmatic deposits characterized by massive to faintly stratified texture, strongly lithified grey ash matrix, containing white and black, lithic lapilli and blocks (i.e., respectively, carbonate and lava or holocrystalline mafic inclusions), as well as leucite (often turned to analcime), clinopyroxene and dark mica crystals, and occasional accretionary lapilli. This lithofacies reaches up to a few tens of meters in thickness where channeled in response to paleotopography (e.g., at the bottom of stream incisions near Patrica village). In addition, strongly lithified, granular massive beds made up of well-sorted, fine scoria lapilli, with pervasive interstitial calcite are found intercalated within laminated surge beds (Fig. 4d).
Ashy base surge deposits with intervening, poorly consolidated scoria lapilli fallout beds from Strombolian activity are also found (Fig. 4f), and they display variable degrees of preservation or reworking (e.g., Fosso di Monteacuto, Supino, Villa S. Stefano).
Due to the uncertain identification of the juvenile fraction in the peperino facies, 40 Ar/ 39 Ar analyses were performed on 20-50 loose crystals from the matrix of the associated base surge and/or scoria fall deposits (i.e., samples SUP-1, PAT-5, PAT-1B, GdR-1, and GdR-5).
Of note, different from the Volsci Range setting, phreatomagmatic surge deposits in the Pofi-Arnara area lack the typical peperino lithofacies, and they are rich in sandstone and pelite clasts from the Miocene siliciclastic substrate (i.e., Frosinone formation; Centamore et al. 2010).
Clear evidence of relic morphologies (e.g., maar crater rims) or proximal (near-vent) facies (evidenced by occurrence of ballistic clasts with impact sags and of supercritical dune bedforms) are scarcely preserved in the phreatomagmatic examples of the Volsci Range mountain setting. However, a number of observations favor a near-vent location, such as the occurrence of coarse lithic lapilli and blocks (even a few decimeters in size) with variable degrees of thermometamorphism, as well as of lithic-rich breccia horizons or lenses in the peperino deposits, up to several meters in thickness.
On the other hand, associated magmatic and phreatomagmatic deposits in the Middle Latin Valley (i.e., Pofi, Colle La Grotta, Arnara; Table 2) show unambiguous evidence of a local origin from rooted eruptive centers, which partially preserved their morphologies.
Pofi, the most prominent volcanic center in the study area, is a polygenetic edifice, characterized by prevailing Strombolian (lapilli and scoria fallout deposits) and associated effusive (including the major Fosso Meringo lava flow; Centamore et al. 2010) activities, and subordinate phreatomagmatic activity. Among the cluster of eruptive centers at Pofi, here we considered the Strombolian scoria fall products of the main scoria cone (sample POF-3) (unit POF a ; Centamore et al. 2010). The local occurrences of welded spatter deposits, small lava tongues and a lava dyke, consistently indicate rooted volcanic activity.
In addition, we dated the phreatomagmatic products of Colle La Grotta (sample CLG), attributed to the early eruptive phase of the Pofi volcano (unit ULG; Centamore et al. 2010), which in some places are associated with Strombolian fallout deposits containing black scoria lapilli, bombs, and sandstone blocks with impact sags. Sample ARN-1 was collected from the products of Arnara, which is a partially preserved scoria cone from Strombolian activity (Fig. 3) with subordinate phreatomagmatic phases. Juvenile scoria lapilli and bombs are phonotephritic in composition (Centamore et al. 2010).
Finally, we dated an ash deposit at Colle Borello locality (NE of Pofi; Fig. 3) from the basal part of the local succession, also including coarse ash and scoria lapilli fallout layers and reworked deposits, which are mapped as the stratigraphically highest volcanic unit (unit UBO; Centamore et al. 2010).
Sample localities are reported in Fig. 3 and in Supplementary File #1. In the next section, we describe the relevant mineralogical-petrographic and geochemical features of the sampled material (Table 2). In the next following section, we report the main features of the deposits that were dated by 40 Ar/ 39 Ar method (Table 3).

Texture and microchemistry of scoria clasts
Scoria clasts in the study products show sub-aphyric, hypocrystalline and scarcely vesicular textures, with phenocryst (> 500 µm) proportions less than 10 vol.%, whereas microphenocrysts (500-100 µm) are, generally, abundant (Table 2). Among the mafic phases, ubiquitous clinopyroxene (Cpx) occurs in both crystal size classes; olivine (Ol) is present in some cases, either as phenocryst or in the groundmass. Ol phenocrysts can host Cr-spinel inclusions. Dark mica (phlogopite, Phl) can occur as deformed phenocrysts and/or microphenocrysts, while is always absent in the groundmass.
The sialic phases, which usually do not occur as phenocrysts, are represented by leucite (Lct), plagioclase (Plg) and alkali-feldspar (AlkF). The latter phase is distinguishable in thin section as either millimeter-sized xenocrysts (i.e., characterized by rounded shape) or microphenocrysts in association with Lct. Otherwise, in the Plgbearing scoria (see below) only the EDS spectra highlight the occurrence of AlkF, likely as high-temperature ternary feldspar (Di Luzio et al. 2018). Noteworthy, Lct and Plg did not crystallize together in the analyzed scoria clasts. This peculiarity allowed us to distinguish two groups of rock types: (1) Lct-bearing vs.
In the Lct-bearing scoria clasts, Cpx, phlogopite (Phl) and melilite (Mel) phenocrysts and/or microphenocrysts show the most intriguing textural and microchemical features. Cpx crystals, up to 3 mm in size, have homogeneous, zoned or spongy texture, and range in composition from diopside to Al-rich diopside (Fig. 5a--i). Phl crystals are frequently characterized by kink bands, but rarely by thin oxide-bearing coronas, and show a wide range of Mg/(Mg + Fe +2 ) ratio and relatively low Ti and F contents (Fig. 5c). The latter feature distinguishes Phl occurrences in scoria clasts with respect to lava flows ). In some cases (GdR-4, SUP-2), Phl accompanies Ol in the phenocryst assemblage, while it does not occur when Ol is only present in the groundmass. In Mel-bearing scoria clasts (PAT-1), rare Phl is present as small, rounded microphenocrysts. In this case, Mel occurs in the groundmass or as microphenocrysts with elongated shapes that frequently host rounded Ol crystals (Fig. 5c).
The Plg-bearing scoria clasts are characterized by a simpler mineralogical assemblage (Cpx + Plg ± Ol ± Mgt) and relatively abundant fresh glass (Table 2). In this group, Cpx crystals, up to 1 mm in size, are relatively homogeneous and, interestingly, show a more constant composition (i.e., diopside content, Mg/Mg + Fe 2+ ) and a higher Al vi /Al iv ratio than in Lct-bearing scoria clasts (Fig. 4a-ii). Glasses are basaltic to trachybasaltic in composition in the Ol-bearing assemblage, while are phonotephritic in the Ol-lacking scoria clasts (see Supplementary material #3).

Ar/ 39 Ar data
New 40 Ar/ 39 Ar datings for the VVF eruptive products are reported in Table 3 and discussed in "New geochronologic constraints". Dating of phreatomagmatic products of the Roman Province is often challenging and results should be interpreted with caution (e.g., Marra et al. 2003Marra et al. , 2016. This is mainly due to xenocryst contamination owing to the widespread occurrence of accessory lithics (i.e., lava and pyroclastic fragments, hypo-abyssal granular rocks, loose crystals) entrained from diatreme-wall rock interaction (e.g., Valentine et al. 2015). Thus, the juvenile crystal fraction may occur even in subordinate proportion with respect to the xenocrystic ones. Furthermore, the juvenile component is often constituted of sub-aphyric scoria, lacking K-bearing phenocrysts suitable for dating.
Due to the sanidine-free erupted compositions of VVF phreatomagmatic centers, and since leucite in juvenile scoria is usually replaced by analcime (which would have resulted in significantly younger ages; Villa and Buettner 2009), in the present work, we dated loose crystals from the ash matrix of pyroclastic-surge deposits. Regarding age interpretation, it should be noted that the analyzed samples from phreatomagmatic deposits (SUP-1, GdR-1, PAT-1B, and PAT-5) yield significant age dispersion. However, these products were mainly sourced from isolated, monogenetic eruptive centers. Therefore, it is unlikely that the older age values might have derived from earlier buried volcanic edifices or erupted deposits, and yet may provide evidence of older magma batches that cooled in sub-surface conditions.
Thus, for those samples yielding scattered crystal populations, the eruption ages can be assessed based on the youngest populations of 4-5 crystals (Table 3, Fig. 6), which at least provide post-quem (maximum) age constraints.
Regarding KS products, the new 40 Ar/ 39 Ar data face the scarcity of suitable minerals for dating. For this reason, in the past, KS lavas were often dated on the groundmass ), in which sparse alkali-rich feldspars are present (Civetta et al. 1979;Di Luzio et al. 2018). Similarly, we derived the ages of Plg-bearing scoria clasts (MOS-2, CA-C1, GdR-1, and POF-3) from few feldspar crystals (Table 3).
In example, sample POF-3 from scoria fall deposits yielded a homogeneous population of five crystals giving a weighted mean age of 397.7 ± 7.7 ka (Fig. 6a). A more precise weighted mean age of 393 ± 3.5 ka is obtained from the three youngest crystals. This age broadly matches that of the Colle Meringo Lct-bearing lava flow (390 ± 20 ka, Basilone and Civetta 1975), also related to the Pofi composite edifice, consistent with the lack of evidence in the field for intervening long time-breaks (e.g., testified by mature paleosoils). This points out that Plg-bearing and Lct-bearing magmas (KS and HKS, respectively) were erupting from the same eruptive source area in the same time range, as already noticed by Centamore et al. (2010) in this eastern sector of the VVF (i.e., Mola a Vento, Arnara). We note that the age of 298 ± 28 ka obtained by Boari et al. (2009) for a KS lava flow of the Pofi edifice, interbedded between the Fosso Meringo lava plateau and the fallout deposit of the POF a unit (sample POF-3), is stratigraphically inconsistent with the age of ca. 390 ka of the bracketing units.
Regarding the apparently problematic age of sample MOS-2, which yielded a statistically consistent older population of crystals with a weighted mean age of 330.9 ± 2.6 ka (n = 14), along with a number of scattered crystal ages as young as 12 ka (Fig. 6b), this is likely the result of the deep weathering that affected this deposit (see Suppl. Material #1). Indeed, the low %Ar* of most of young crystals suggests a possible Ar loss due to deep alteration, which is one potential cause for age rejuvenation. Indeed, when a cutoff at %Ar* < 50 is applied, only two, statistically poorly significant younger crystals are left. Notably, the age of 331.6 ± 3 ka that we attribute to MOS-2 is the youngest in our dataset, consistent with the inferred stratigraphic position of the Colle Borello unit on top of the local VVF successions (UBO; Centamore et al. 2010).

Constraints to the VVF pre-eruptive magma system
The Lct-bearing scoria clasts of the VVF (HKS rock type, Table 2) originated from high-K parental magmas, similar to the Middle Latin Valley volcanics described in literature as high-K rocks or as HK series (Boari et al. 2007Frezzotti et al. 2007). The wide range of Mg/(Mg + Fe 2+ ) values for Cpx and Phl phenocrysts in Lct-bearing scoria clasts indicates that parental magmas underwent significant pre-eruptive crystallization, as also suggested by the MgO contents (3.62-6.55 wt%) reported in literature, mainly for HKS lava flows (Boari et al. 2007Frezzotti et al. 2007). In particular, the textural features of the Lct-bearing scoria clasts hosting Mel (PAT-1, Patrica NE center) suggest that the early erupted magmas of VVF originated from a fractional crystallization process coupled with the assimilation of carbonate wall rocks of the Volsci Range. In fact, rounded Ol inclusions in Mel microphenocrysts (Fig. 5) indicate that Ol was not stable in paragenesis with Mel, and that the latter formed by a reaction between Ol and a melt enriched in calcium via the assimilation of sedimentary carbonate. This process is in agreement with the phase relationships in the system Forsterite-Diopside-Akermanite-Leucite at low pressure (Gupta 1972) where at the thermal minimum occurs an olivine-free assemblage, constituted by clinopyroxene + akermanite (i.e., melilite) + leucite + melt. Also, high (> 6.4‰)  (Frezzotti et al. 2007) point towards the interaction of the early high-K magmas with sedimentary carbonate (cf. Gaeta et al. 2006;Di Rocco et al. 2012). Here, we do not attempt to model mathematically the carbonate AFC process at the origin of the Mel-bearing rocks of VVF, as the end-member representing the parental magma is, currently, not well constrained isotopically. So far, the temporal change in the radiogenic isotopes of parental magmas has been defined for the Colli Albani and Sabatini major ultrapotassic districts of the Roman Province (i.e., 87 Sr/ 86 Sr trend from ~ 0.711 to 0.709 with time; Gaeta et al. 2006Gaeta et al. , 2016Sottili et al. 2019), but not for VVF. Except for the early VVF activity, carbonate assimilation was less extensive than at Colli Albani, as indicated by the scarcity of low-silica, K-foiditic rocks and the lower F content of Phl in VVF Lct-bearing scoria clasts in comparison to Colli Albani (Gaeta et al. 2000). In fact, the activity of F in Phl inversely correlates with the activity of H 2 O in the magma that, in turn, inversely correlates with that of CO 2 . Different from Colli Albani (cf. Freda et al. 2008), the relatively low F content of Phl indicates a low   Gaeta et al. 2000) rock types, reported in the Ti-Mg/(Mg + Fe 2+) (upper) and F-Mg/(Mg + Fe 2+ ) (lower) diagrams. c SEM Electron backscattered image of a Lct-bearing scoria (sample PAT-1) containing melilite. The latter mineralogical phase occurs as microphenocrysts with elongated shape, hosting rounded olivine crystals (inset) Ar and 37 Ar measurements), 40 Ar/ 39 Ar ages of feldspar and/or leucite from single grain analyses, and associated probability distribution functions. Weighted mean ages (reported with 2σ uncertainties) are assigned when distinct populations can be detected and interpreted in light of age data and field context. To avoid additional complications in data interpretation imposed by 40 Ar/ 39 Ar dates with low radiogenic 40 Ar yields, all crystals with < 50% Ar* are not considered for weighted mean age calculations. Open circles: data not included in the weighted mean. See text for detailed interpretations CO 2 activity in the VVF magma system, which allowed an early and more prolonged presence of Phl on the magma liquidus. In addition, the higher values and wider range of Mg# of VVF Phl with respect to Colli Albani (Fig. 5b) consistently indicates a wider Phl stability field in the VVF magma system. This explains the observed association of Phl with Ol in the phenocryst assemblage of VVF Lctbearing scoria clasts, which, instead, is lacking in the Colli Albani scoria clasts (e.g., Freda et al. 1997Freda et al. , 2006Freda et al. , 2011Palladino et al. 2001;Marra et al. 2009;Gozzi et al. 2013;Gaeta et al. 2016). An early and prolonged Phl crystallization should profoundly affect the trace element abundance in magmas. Coincidentally, the VVF magmas show a wide range in Rb content, which spans over more than one order of magnitude, with nearly constant composition of the other geochemical parameters (Frezzotti et al. 2007).
We note that in the Phl-bearing HKS examples (i.e., GdR-4, ARN-1, and SUP-2; Table 2), Lct is absent among phenocrysts, whereas it occurs as microphenocrysts. Consistently, a relatively wide Phl stability field provides evidence for an origin from a deep, hydrated magma reservoir in which Lct was not stable, so that Lct crystallization took place at relatively shallow depth in the eruption feeder system. Noteworthy, in other cases where high-K magmas are represented by Lct-bearing, Phl-free scoria clasts containing Ol in the groundmass (i.e., samples PZ4 and POF-2; Table 2), the lack of Phl crystallization indicates a shorter residence time in the plumbing system, in agreement with the primitive magma composition (MgO ≈ 10 wt%; Centamore et al. 2010).
The Plg-bearing scoria clasts of the VVF originated from a low/medium-K parental magma (here referred to as KS), similar to the Middle Latin Valley volcanics described in literature as calc-alkaline or shoshonitic rocks (Boari et al. 2007Frezzotti et al. 2007). Plg-bearing scoria clasts lack Phl and contain Cpx phenocrysts with a narrow compositional range with respect to Lct-bearing scoria clasts (Fig. 4a). The limited chemical evolution of Cpx thus suggests a shorter sub-surface residence time of magmas, also in agreement with the experimental growth rate (Bonechi et al. 2020): in fact, Cpx phenocrysts are finer (max. 1 mm in size) in Plg-bearing scoria clasts than in Lct-bearing ones (up to 3 mm). Moreover, the higher Al vi /Al iv ratio for Cpx in Plg-bearing scoria clasts than in Lct-bearing ones (Fig. 5a) indicates a higher pressure of crystallization for the KS parental magma. This has recently been tested experimentally (Perinelli et al. 2019) using a primitive K-basalt from Procida Island (Campi Flegrei) similar to the primitive KS rocks of VVF in composition (Frezzotti et al. 2007). In the VVF, Plg-bearing scoria clasts, containing small, poorly differentiated, high-pressure Cpx, derived by primitive, low/ medium-K parental magmas, thus strongly suggest a fast ascent from a deeper reservoir with respect to the slightly differentiated, high-K, Lct-bearing magmas.

New geochronologic constraints
Eleven new age determinations, integrated with the existing 21 ages selected from the literature (Basilone and Civetta 1975;Boari et al. 2009;Centamore et al. 2010), allow us to depict an overall eruptive history for the VVF, spanning from 761.5 ± 9.5 ka (this work) to 231 ± 19 ka (Centamore et al. 2010). Moreover, we are now able to outline the migration of the eruptive centers in space and time (Fig. 2ai-iii) with respect to the erupted magma compositions.
Xenocryst populations (either from local or distal sources) occurring in the dated phreatomagmatic products provide insights into the onset of potassic magmatism since at least 1206 ± 40 ka (weighted mean age of two feldspar crystals; sample PAT-5; Figure S1 and Table S1 in Supplementary File #5). Moreover, the crystal age spectrum from gravelly deposits at Colle Avarone (sample CA-CGT, Fig. 6b), on top of the sedimentary fill of the Liri Basin yielded old ages at ca. 870-820 and 700 ka.
Following this poorly constrained paleo-magmatism of uncertain source, three major phases of activity can be defined at VVF, whose age ranges and related uncertainties are imposed directly from the interpreted 40 Ar/ 39 Ar ages: 1. "Early eruptive phase" (761.5 ± 9.5 -539.3 ± 11 ka), defined by the oldest and youngest dated product sourced in the carbonate Volsci Range backbone (i.e., the Patrica NE and the Fosso di Monteacuto centers). In particular, the Patrica, Patrica NE and Piglione (614 ± 10 ka; Boari et al. 2009) centers lie along a NW-SE volcanotectonic lineament, a segment of the NW-trending fault system bounding the Volsci Range and the Middle Latin Valley Graben. Moreover, the Colle Castellone center (600 ± 4 ka; Boari et al. 2009) developed along the southern segment of the same fault, bordering the Middle Latin Valley Graben (Fig. 2). Phreatomagmatic deposits, characterized by the typical peperino facies rich in carbonate lithic inclusions, as well as the Piglione and Colle Castellone lava occurrences, are HKS, Lct + Mel-bearing rock types, which consistently indicate a significant degree of magma-wall rock interaction. 2. "Main (Climactic) Eruptive Phase" (424 ± 13-349.5 ± 5.0 ka), comprised between the Lower Tomacella unit ) and the Villa S. Stefano center (this work), extending to 330.9 ± 2.63 ka (Colle Borello, MOS-2, this work). The source areas of the VVF climactic activity clustered along a NNE-trending volcano-tectonic lineament (i.e., Tecchiena, Selva dei Muli, Tomacella, Celleta, Giuliano di Roma; Fig. 3) and in the Pofi area (i.e., Arnara and the Pofi polygenetic center, also including major effusive activity of Fosso Meringo lava flow). At a larger scale, the Pofi source area may represent the eastern termination of an ENEtrending belt of volcanic centers extending to the west across the early Fosso di Monteacuto center, as far as the Pontina Plain at the VVF periphery (Fig. 1). This time span also includes the ages of the distal products in the southeastern reach of the Liri Basin (402 ± 5 to 351 ± 3 ka; Nomade et al. 2011;Pereira et al. 2018); 3. "Late Eruptive Phase" (300 ± 28 to 231 ± 19 ka?), concentrated within the Sacco River Valley. In particular, several volcanic centers (e.g., Colle Vescovo, Colle Sant'Arcangelo, Colle Celleta, Colle Spinazzeta) developed along the major NW-trending fault system bounding the Middle Latin Valley graben, i.e., the same fault system that controlled the early phase of activity (e.g., Colle Castellone).

VVF eruptive history
Overall, the VVF eruptive pattern was characterized by small volume (in the order of 0.01-0.1 km 3 , based on an approximate estimation of the mapped deposits) eruptions from a network of monogenetic centers, totaling nearly 4 km 3 (i.e., two orders of magnitude lower than the major potassic volcanic districts of the Roman province). Available geochronological constraints suggest that long quiescence periods (in the order of 10 5 years) occurred between isolated eruptive events during the early activity phase (since ca. 760 ka), which was followed by a climactic eruptive activity clustered at around 0.4 Ma in the eastern sector of VVF (with a tail possibly extending up to 0.2 Ma). Early activity was essentially fed by K-rich, Lct + Mel-bearing magmas, showing evidence of relatively high degree of carbonate assimilation, possibly due to multi-stage ascent through thick carbonate successions (hence the diffuse phreatomagmatic character). Subsequently, during the main activity phase in the eastern VVF sector (since ca. 420 ka), HKS rock types alternated with poorly differentiated, Plg-bearing KS rock types from eruptive sources close in time and space (e.g., Ceccano and Pofi areas). In particular, the eruptive activity along the major faults bounding the Middle Latin Valley Graben (e.g., Colle del Vescovo, Selva Piana, and Colle Capirolo centers) testifies for the fast ascent of nearly primary KS magmas (K-basaltic in composition; Centamore et al. 2010).
The actual duration of the most recent activity in the VVF remains an open question, due to the analytical uncertainties of the existing K-Ar age determinations and the analytical errors associated with the re-calculated 40 Ar/ 39 Ar ages from Boari et al. (2009), as well as with those reported in Centamore et al. (2010) (Table 1). The youngest age reported in Boari et al. (2009) refers to the Colle Sant'Arcangelo Lctbearing lava (ERN 86,254 ± 4 ka). However, despite the small error, this age relies only on the four first steps with K/ Ca ratios > 100, whereas the other steps display K/Ca < 70, suggesting that either a dishomogenous mineral assemblage was involved in the dated lava groundmass, or pervasive analcimization of Lct caused a rejuvenation of the age. We remark that, if this age value is correct, the HKS lava of Colle Sant'Arcangelo would be the youngest dated product in Boari et al. (2009), and younger than any KS products dated in the present work, in conflict with the petrologic evolution suggested by Boari et al. (2009), according which KS magmas were the last erupted.
In addition, two KS lava flow samples yielded groundmass plateau ages with low precision: the age of 302 ± 28 ka obtained for ERN 94 (Colle Vescovo) relied only on 4 steps and included only 66% of the total 39 Ar K ; the age of 289 ± 32 for ERN100 (Pofi volcano) is more robust, including 100% of the 39 Ar K , although incongruent with the age of 393.0 ± 3.5 ka that we obtained for Plg-bearing activity at Pofi (POF-3).
Finally, we cannot fully evaluate the consistency of the two 40 Ar/ 39 Ar ages age determinations of 263 ± 58 ka (Colle Capirolo) and 231 ± 19 ka (MVN unit of Pofi volcano) reported in Centamore et al. (2010), due to the lack of full analytical details. New dating efforts are thus needed to assess the actual timing of the late phase of activity at VVF.

Insights on magma source
New data presented in this paper show that the eruption of high-K (Lct-bearing) and lower-K (Plg-bearing and Lct-free) magmas was broadly concomitant during the VVF activity, which puts constraints on the geometry of the metasomatic mantle source. The most primitive, KS magmas of the VVF were likely saturated by amphibole (Amph) at high pressure, as supported by experiments on analogue compositions, in which Phl saturation was never achieved (Perinelli et al. 2019;Bonechi et al. 2017). It is worth noting here that the most primitive high-K magmas of the Roman Province are instead saturated by Phl at high pressure (Conte et al. 2009). On these grounds, the observed temporal shift from HKS to KS magmas at VVF cannot be explained in the frame of a laterally heterogeneous, veined mantle. It would imply the exhaustion of a Phl-bearing vein (source of HKS magmas) prior to an Amph-bearing one (source of KS magmas), which would conflict with the higher melting temperature of a Phl-bearing mantle source with respect to an Amph-bearing one. Alternatively, different depths of mantle partial melting could be at the origin of HKS vs. KS parental magmas (e.g., Frezzotti et al. 2007;Peccerillo 2017). In our view, consistent with Cpx barometric constraints (Fig. 5), a vertically zoned metasomatized mantle would imply a shallower and older Phl-bearing layer (e.g., Gaeta et al. 2016), and a deeper Amph-bearing layer related to Cenozoic subduction dynamics. The late onset of KS magmatism at VVF may indicate the partial melting of the Amph-bearing layer at higher pressure with respect to the Phl-bearing source of early activity. An increasing extensional rate (possibly linked to the dynamics of the subducting slab, see below) would have resulted in an enhanced degree of decompression, allowing a deeper source melting and the production of KS magmas.

Volcano-tectonic relationships
Chronostratigraphic indications show that > 80 m-thick fluvial-lacustrine sediments accumulated in the Liri Basin in the time span 600-350 ka (Fig. 3), constraining the timing of tectonic subsidence of this sector of the Latin Valley. The structural setting of the area (Fig. 1) shows a marked control of the NW-striking normal faults on the geometry of the Liri Basin, which was affected by NE-SW oriented extensional tectonics coeval with the widespread Middle Pleistocene tectonics forming the intramountain basins of central Apennines (Cosentino et al. 2017 and references therein). Phases of coupled major extensional tectonics and volcanic activity have been evidenced for the volcanic districts of the Tyrrhenian Sea margin of central Italy (e.g., Acocella and Funiciello 2002;Marra et al. 2004;Marra and Florindo 2014). Here, we outline the timing and structural control on magma ascent at VVF.
The major NW-striking normal fault system bounding the Volsci Range and the Middle Latin Valley Graben (Fig. 1) controlled the location of early activity centers sourced in the carbonate substrate (i.e., Patrica NE, Piglione, Colle Castellone; ca. 761-604 ka), as well as the alignment of late eruptive centers (i.e., Colle Vescovo, Colle Sant'Arcangelo, Colle Celleta, Colle Spinazzeta; possibly up to ca. 250 ka). This NW-trending volcano-tectonic lineament may be essentially related to the upper crust (< 12-15 km depth) fault architecture dominated by Quaternary extensional structures and older re-activated faults.
Instead, the NNE-trending (Tecchiena, Selva dei Muli, Tomacella, Celleta, Giuliano di Roma) and ENE-trending Ceccano Fault (Ceccano and Pofi source areas) volcanotectonic lineaments controlled the cluster of eruptive centers during the VVF climactic activity (ca. 424-349 ka). While the ENE-trend is an expression of a broader (in space and time) lineament extending to the western VVF periphery as far as the Pontina Plain, the activity of the NNE-trend was more focused in time. These lineaments might be related to high-angle faults with strike-slip kinematics (e.g., Sani et al. 2004), which may represent the surface expression of deeper slab-mantle dynamics and the preferential pathways for magma injection into the crust (Acocella et al. 1996;Centamore et al. 2010; see "Volcano-tectonic relationships"). The re-activation of high-angle lithospheric faults in the extensional regime that affects the inner Apennines since at least Pleistocene may potentially tap magma deeper than the typical NW-striking normal faults of the Apennines that cross-cut limitedly the upper crust, thus driving the emission of primitive magmas (Acocella et al. 1996;Funiciello 2002, 2006). As a result, crustal-scale dilational jogs may form at depth, whose surface expression is the dextral wrench zone cutting across the Volsci Range, shaping the Middle Amaseno and Middle Latin valleys. In this frame, the reported 400-351 ka volcaniclastic occurrences within the sedimentary fill of the Liri Basin (Nomade et al. 2011;Pereira et al. 2018) indicate that the tectonic phase responsible for this graben-like tectonic structure was coeval with the climactic phase of activity in the VVF.
The analysis of the instrumental seismicity in the years 2009-2019 highlights the occurrence of sparse seismicity in the Latin Valley, as opposed to the high concentration of moderate to strong events within the main Apennine chain (Fig. 5). This indicates low deformation rates within the Latin Valley, where focal mechanisms account for the re-activation, besides dip-slip NW-striking faults, of NNEstriking faults with oblique-slip kinematics, consistent with a prevailing NE-oriented σ 3 , as typical of the Apennines (Montone and Mariucci 2016).
The role of the NNE-trending lineaments in controlling the pathways of volcanic activity has been shown in this paper not to be exclusive, but accompanying the NW-and ENE-oriented alignments of eruptive centers. We interpret the NNE-and ENE-trending lineaments to have been reactivated under the NE-oriented extensional regime affecting the upper crust during the VVF activity time span.

Geodynamic implications
The peculiarities of the VVF with respect to the other potassic volcanic districts of the Tyrrhenian Sea margin of central Italy, despite the common temporal range of activity, are summarized as follows: (1) the small volume of total erupted products (in the order of a few vs. hundreds of km 3 ); (2) the low intensity and magnitude of individual eruptive events vs. large Plinian and/or caldera-forming eruptions; (3) the lack of highly differentiated rock types, which indicates rapid magma ascent from the mantle source, with limited stationing in shallow magma chambers; (4) the coexistence of both HKS and KS primitive magmas (in some ways, similar to Roccamonfina volcano nearby) that usually characterize distinct volcanic districts (e.g., HKS at Somma-Vesuvius vs. KS at Campi Flegrei).
Overall, the paucity of erupted magma at VVF would suggest low degree of partial melting of the mantle source and/or hindered ascent to surface. Lct-bearing (HKS) and Lct-free (KS) magmas characterized distinct eruptive phases and/or eruptive centers, although with some spatial and temporal superposition, yet with no evidence of pre-eruptive mixing/mingling. HKS magmas mostly fed the early phase of activity (sourced in the Volsci Range setting), then KS magmas appeared during the climactic phase, partially overlapping in time and space with HKS ones (e.g., in the Pofi-Ceccano areas); KS magmas prevailed during the late phase of activity, although still accompanied by HKS events.
We note that the eruption of Lct-bearing products (from slightly differentiated HKS magmas), in particular those containing Phl, requires some prolonged pre-eruptive magma stationing and/or multi-stage ascent through carbonate rocks in diatreme systems (Cardello et al. 2020). In contrast, Plg-bearing eruptive products (from poorly differentiated KS magmas) reflect fast ascending primitive magma batches allowing "bullet eruptions". Below, we propose that such a change in the magma plumbing system may have resulted from an acceleration in the extensional rate affecting the upper plate associated with the slab retreat.
The eruption of primitive magmas at VVF (mostly in the 420-230 ka time span) from volcanic centers clustered along regional high-angle faults of lithospheric depth suggests a link to the geodynamic setting of the central sector of the Apennines. This is characterized by the differential retreat of the W-directed subducting Adriatic plate between the northern arc of the Apennines (associated with the Middle-Upper Pleistocene potassic magmatism of the Roman Province) and the southern arc related to Campanian magmatism, also including the presently active Ischia, Campi Flegrei and Somma-Vesuvius volcanoes. In the context of the Africa-Europe convergence, the Tyrrhenian basin-Apenninic arc system (Malinverno and Ryan 1986) formed as a result of the evolution of the accretionary prism characterizing the upper plate. Patacca and Scandone (1989) recognized the progressive development of two distinct (i.e., northern and southern Apennine) arcs of subduction in the post-Tortonian times. Slab retreat triggered the extensional regime presently characterizing the Tyrrhenian Sea margin of Italy (Malinverno and Ryan 1986;Patacca and Scandone 1989), which, otherwise, would not have occurred in the compressive domain of the Africa-Europe convergence. Thus, the mainly compressive stress field acting at depth is accompanied by an extensional regime in the upper crust, associated with the gravity spreading of the Apennine chain (Fig. 7), which may provide preferential pathways to magma uprising.
Several authors, based on tomographic datasets, claimed for a lateral discontinuity in the subducting plate below the Apennines. For instance, Spakman (1991) and Spakman et al. (1993) suggested the occurrence of a slab detachment below the central-southern Apennine. Lucente and Speranza (2001), through teleseismic tomography, noticed the lack of high velocity anomalies in the depth range 35-100 km in the Apennine sector confined between the Ancona-Anzio and the Ortona-Roccamonfina lines (Fig. 7). Compared to the continuous high velocity zone existing in the 340-440 km depth range, mirroring the shape of the whole Apennine chain from northern Tuscany to Calabria, this tomographic discontinuity was attributed to a local slab detachment and cessation of subduction. Recent tomographic investigations have demonstrated a segmented geometry of the subducting slab (Ascione et al. 2012;Giacomuzzi et al. 2012;Chiarabba et al. 2014;Rosenbaum and Piana Agostinetti 2015;Chiarabba and Palano 2017). In particular, based on the upper mantle structural frame of Rosenbaum et al. (2008), Rosenbaum and Piana Agostinetti (2015) suggested a link between the asthenospheric-derived magmatism and the "central Apennine slab window", regarded as the expression of advanced slab tearing (Pierantoni et al. 2020).
The origin of VVF magmas has been attributed to a variably metasomatized, heterogeneous mantle source, although the timing of the metasomatic processes and the geometry of heterogeneity (i.e., lateral veins vs. vertical layering) are still debated Conticelli et al. 2009;Nikogosian and van Bergen 2010;Gaeta et al. 2016;Kornneef et al. 2019). In this context, the eastward migration of the southern segment of the subducting plate (between approx. 6 and 2 Ma; Rosenbaum et al. 2008) may have produced a subduction gap (i.e., a central Apennine slab window), thus shifting away from the VVF-hosting area a potential source of volatiles required for the partial melting of the metasomatized mantle, and limiting the degree of partial melting. Conversely, the presence of a subducting slab underneath the nearby major volcanic districts of the Roman and Campanian provinces ensured volatile transfer to the mantle source and extensive magma production from Middle Pleistocene to present.
This scenario would explain the very limited amount of erupted products at VVF, despite the availability of easy ascent pathways (testified by small-scale, bullet eruptions fed by primitive magmas), as the consequence of very limited production of magma (rather than hindered ascent that would have resulted in extensive plutonism). This is typical of tectonically controlled volcanic fields (Valentine and Perry 2007), where melt generation is dependent on tectonic forces, and the very low magma flux is a passive byproduct of regional tectonic strain.
We also suggest that, while the retreat of the lower plate affected dramatically the productivity of the mantle source, the re-activation of structural lineaments in the upper plate favored the ascent and eruption of very small batches of poorly to slightly differentiated magmas. In particular, the observed compositional shift from HKS to KS primitive magmas during the VVF activity climax was concomitant to enhanced mantle decompression. In our view, this may be due to an increase in the extensional rate in the upper plate (possibly related to an increasing rate in the slab retreat), also responsible for the rapid subsidence of the Liri Basin.
As a whole, in response to changing degrees of decompression, mantle melting was localized in small-scale domains, producing small magma batches, their ascent being favored by enhanced crust permeability due to major fault interconnection.
Long quiescence periods occurred between isolated eruptive events from ca. 0.8 to 0.4 Ma, whereas a climactic eruptive activity clustered at around 0.4 Ma. The age of the most recent activity in the VVF remains an open question and new dating efforts are still in order.
The VVF is an example of a tectonically controlled volcanic field. Eruptive centers, mostly monogenetic tuff rings and scoria cones with lava occurrences, appear to be aligned along the main NW-SE border fault of the Sacco River Valley (separating the Middle Latin Valley Graben and the Volsci Range settings), as well as along "transversal" NNEand ENE-trending lineaments. The horizontal stress orientations correspond to σ 3 , for normal and normal/strike faulting tectonic regime, and to σ 1 for strike-slip, thrust and thrust/strike faulting. Tomography of the region modified from Lucente and Speranza (2001). Geodynamic vectors modified from Frepoli et al. (2010), based on Faccenna et al. (1996). AAL Ancona-Anzio line, ORL Ortona-Roccamonfina line, VVF Volsci Volcanic Field