## Introduction

Volcanic eruptions occur daily—about 100 times per year1—and some potentially have extreme destructive power as they can release an enormous amount of energy in a very short time. Depending on their size and style (explosive or effusive), volcanic eruptions can alter climate both locally and globally, cause mass extinctions, fatalities, famine and diseases2,3,4,5,6,7. As such, they represent a serious threat to human activities, infrastructures, and economy. It follows that the probabilistic prediction and mitigation of volcanic risks have acquired paramount importance in modern Earth science. The mechanistic understanding of magma fragmentation and hence explosive behavior of volcanoes represents indeed one of the three grand challenges in volcanology and eruption forecasting8.

Probabilistic predictions of volcanic eruptions rely on numerical modeling of magmatic processes9,10,11,12,13. Among the parameters necessary to model eruptive scenarios, those describing magma transport and hence rheology are the most crucial. Magma viscosity is the central parameter that controls rheology and thus the flow behavior from the storage environment to the volcanic vent14,15,16,17,18,19. In particular, the temperature and chemistry dependence of melt viscosity is key in controlling magma transport, its decompression rate and thereby the overall eruptive style20,21,22,23,24,25,26,27.

Empirical models of melt viscosity28,29,30,31 are routinely used to approximate magma viscosity under eruptive conditions. These models are based on experimental data obtained over several decades from viscometry measurements (ref. 31 and references therein). However, numerous studies have highlighted the challenges associated with accurately determining the viscosity of melts prone to partial crystallization21,22,23,32,33,34,35, which can lead to overestimating the viscosity of the liquid by up to two orders of magnitude21,32. The experiments appear particularly challenging for compositions containing iron and titanium oxides and viscosity values between ~109 and ~1012 Pa s, where pervasive nanoscale crystallization can easily go undetected if the samples are not checked post-measurement with adequate detection limits and spatial resolution. Given the importance of low-temperature measurements for the extrapolation of both anhydrous and especially hydrous viscosity to eruptive conditions, these experimental difficulties bring into question the accuracy of our present knowledge of the pure liquid viscosity of magmas. This uncertainty necessarily reverberates into numerical modeling of volcanic eruptions, within which magma viscosity influences dramatically the eruptive scenario.

Here, we demonstrate the experimental challenges related to the correct determination of the viscosity of magmatic liquids, using a multipronged approach that includes viscosity measurements, Raman spectroscopy and transmission electron microscopy. As previously inferred32, near-Tg (the glass transition temperature) nanocrystallization of iron titanium oxides substantially increases the low-temperature viscosity of a Mt. Etna trachybasalt (Italy), while such phenomenon is far less impacting (if not absent) in a Stromboli high-K basalt (Italy) measured at similar conditions. We attribute this difference to small but significant differences in total iron and titanium content, which can dramatically affect the tendency of the magmas to reach oversaturation in the vicinity of Tg and (nano)crystallize. We thereby emphasize the need for a critical review and validation of the published viscosity data, providing a vademecum for correct experimental determination of viscosity based on our previous studies32,36.

However, our results have implications that are potentially further-reaching than the laboratory, since nanocrystals have been identified in various rocks formed during explosive eruptions21,32,33,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51. Only in the last 4 years, 60 studies21,23,31,32,33,34,35,36,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76,77,78,79,80,81,82,83,84,85,86,87,88,89,90,91,92,93,94,95,96,97,98,99,100,101,102,103,104,105,106,107,108,109,110,111,112 have been published reporting or inferring the presence of nanocrystals in natural and synthetic samples (Fig. 1). Because the formation of nanocrystals can increase magma viscosity in unexpected ways21,32 and trigger vigorous nucleation of bubbles upon magma decompression and undercooling46,47,48,49,101, our study aims to stimulate a reappraisal of magma rheology and fragmentation in the possible case of syn-eruptive nanocrystals formation. We initiate this process through the provocative speculation that the peculiarly dissimilar eruptive activities of Mt. Etna and Stromboli (summarized in section “On the eruption styles of Mt. Etna and Stromboli volcanoes”) may also originate from a different degree of undercooling-driven oversaturation in Fe- and Ti-oxides, controlling the tendency of their magmas to (nano)crystallize. High-pressure and -temperature experiments using hydrous melts at eruptive conditions confirm that nanolitization occurs more pervasively in Mt. Etna basalt as compared to Stromboli fostering bubble nucleation. We finally discuss the relevance of our results within the realm of basaltic volcanism.

### On the eruption styles of Mt. Etna and Stromboli volcanoes

The recent explosive activity of Mt. Etna is characterized by weak Strombolian eruptions and more energetic paroxysms113,114. The latter consist of lava fountains that can have a duration ranging from several minutes to a few hours and produce eruptive plumes of more than 10 km in height115,116,117,118,119. Lava fountaining at Etna has been interpreted as generated when large volumes of volatiles rapidly exsolve from the magma during its fast ascent and decompression along the conduit120,121 producing magma fountains with exit velocities ranging between 33 and 125 m s–1 122,123,124,125. Recently, it has been suggested126 that lava fountaining is a distinct style, separate from effusive and explosive eruption styles, that is produced when magma ascends rapidly and fragments above the vent.

Ordinary activity at Stromboli consists of persistent, rhythmic, mild Strombolian-type eruptions, driven by the bursting of large bubbles called slugs127. Ejected pyroclasts are made of degassed and highly porphyritic (phenocryst- and microlite-rich) magma stalling at shallow levels (~1 km b.s.l.)128. Mechanisms leading to Strombolian eruptions imply slow magma ascent rates (<0.01–0.1 m s–1), permitting significant bubble coalescence to form gas slugs and promote decoupled gas-melt flow27. At Stromboli, this activity is occasionally punctuated by short-lived, highly energetic explosions, known as paroxysms129,130,131. These eruptions are characterized by the emission of pyroclasts to a height of a few kilometers above the craters, the production of 3–5 km high eruptive columns and occasionally pyroclastic flows132,133,134. The triggering mechanism of these paroxysms is still the subject of debate135, but there is consensus on the primary role of a deep, volatile-rich aphyric basaltic magma rapidly rising (1–3 m s–1)136,137 from the lower magmatic system.

The main difference between the paroxysms of Stromboli and Mt. Etna lies in their duration. While paroxysmal eruptions at Stromboli are single explosions and have a duration of a few minutes132, the sustained lava fountaining can persist at Mt. Etna for more than 1 h118,120. As such, Stromboli paroxysms have been recently defined as basaltic Vulcanian eruptions132, in which the eruption overpressure is provided by a small amount of volatile-rich magma undergoing closed system degassing, while the shallow crystal-rich viscous magma acts as a weak plug138,139. In contrast, the long duration of lava fountains at Mt. Etna testifies to sustained conditions of coupling between the melt and the gas phases during closed system degassing115,126.

## Results and discussion

Near-Tg and superliquidus viscosity measurements were performed on ETN and STR and compared to the whole-curve parameterization based on the Mauro–Yue–Ellison–Gupta–Allan (MYEGA) equation (Eq. (1), ref. 140):

$${{{{{{\rm{log }}}}}}}_{10}{{{{{\rm{\eta }}}}}}={{{{{{\rm{log }}}}}}}_{10}{{{{{{\rm{\eta }}}}}}}_{{{\infty }}}+\left(12-{{{{{{\rm{log }}}}}}}_{10}{{{{{{\rm{\eta }}}}}}}_{{{{{{\rm{\infty }}}}}}}\right)\frac{{{{{{{\rm{T}}}}}}}_{{{{{{\rm{g}}}}}}}}{{{{{{\rm{T}}}}}}}{{{{{\rm{exp }}}}}}\left[\left(\frac{{{{{{\rm{m}}}}}}}{12-{{{{{{\rm{log }}}}}}}_{10}{{{{{{\rm{\eta }}}}}}}_{{{{{{\rm{\infty }}}}}}}}-1\right)\left(\frac{{{{{{{\rm{T}}}}}}}_{{{{{{\rm{g}}}}}}}}{{{{{{\rm{T}}}}}}}-1\right)\right]$$
(1)

where Tg is the glass transition temperature, m the fragility index and log10η the viscosity at infinite temperature, which was fixed equal to −2.9 ± 0.3 in agreement with literature results from materials and Earth sciences31,141.

Low-temperature viscosity measurements of ETN exhibited a physically unrealistic trend and an evident time-dependent increase in viscosity (Supplementary Fig. 2). We therefore used Tg and m derived elsewhere by calorimetry and Brillouin spectroscopy36. Literature data enabled the very accurate description of high-temperature concentric-cylinder data (Fig. 2a). In the case of STR (Fig. 2b), Tg and m were fit to our viscosity data measured at both low- and high-temperature (Fig. 2b), retrieving values in good agreement with previous studies (e.g.131,142, m = 42.0 and Tg = 643.9 °C). Low-temperature viscosity measurements performed on STR remarkably ran stable up to ~780 °C (Supplementary Fig. 2 in the attached Supplementary Data 1). Above this temperature, we observed a slight increase in viscosity over the timescale of the measurement (i.e., 20 min). We subsequently investigated the origin of the observed time-dependent increase in viscosity during micropenetration (MP) measurements by Raman spectroscopy.

The Raman spectra (Fig. 3) of ETN and STR starting materials were typical for basalt glasses143,144. Post-run samples manifested instead a gradual dwindling of the high-wavenumber envelope (> 800 cm−1) associated with the stretching vibrations of oxygens in the silicate network145. Additional intensity emerged at ~300 and ~660 cm−1, where the main peaks of Fe-Ti-oxides such as magnetite are most typically observed32,146; these Raman bands appeared already at 640 °C in ETN and intensified at higher temperatures, while they were only visible in STR after the MP measurements performed at the highest temperatures, namely 777 and 797 °C.

DSC upscans performed at 30 K min−1 further confirmed the inferred higher susceptibility of ETN to near-Tg crystallization (Fig. 4). The endothermic glass transition of ETN (onset at 653 °C) was closely followed (after ~100 °C) by two distinct exothermic events: a first broad peak with onset at 758 °C and a second, sharper and more intense one with a maximum at 860 °C. After that, the sample melted over a broad endotherm with an offset at 1228 °C. In contrast, STR proved to be much more stable against crystallization: we observed in this sample a single broad exothermic event starting at 963 °C, i.e., ~285 °C above its glass transition (onset at 679 °C); the subsequent melting process was completed at 1251 °C. The behavior of ETN closely resembled that of synthetic melts in which the early precipitation of nanosized TiO2-bearing seeds is exploited to control the subsequent heterogeneous nucleation of aluminosilicate crystals, as for the production of conventional glass-ceramics147.

We subsequently subjected ETN to four heat treatments (heating and cooling rates = 10 K min−1) respectively up to 684 °C (ETN+1), 693 °C (ETN+10), 708 °C (ETN+25) and 732 °C (ETN+50), using a DSC for higher temperature precision (see also Supplementary Fig. 3 and Section 3.3). These temperatures are located only slightly above the previously determined Tg of 641 °C36, where viscosity η = 1012 Pa s; the total time spent above this temperature ranged between 8 and 18 min. The Raman spectra of the obtained samples (Supplementary Fig. 4) exhibited nonetheless similar features to those in Fig. 3, namely the emergence of the Raman signatures for Fe-Ti-oxides at ~300 and ~660 cm−1 32,146. We thereafter characterized these materials also by (S)TEM, to directly assess their nanostructural modification.

TEM imaging confirmed that the homogeneous ETN starting material (Supplementary Fig. 5) developed an increasingly heterogeneous nanostructure during the heat treatments (Fig. 5). Even before the identification of nanocrystals with well-defined lattice fringes in ETN+50, the amorphous silicate matrix of ETN+1, ETN+10 and ETN+25 exhibited the clear emergence of high- and low-contrast regions with a size below 10 nm, hinting at an incipient compositional reorganization of the samples before crystal nucleation. Notice that such features would go undetected by X-ray diffraction and scanning electron microscopy, due to the lack of long-range order and the small size. High-angle annular dark-field (HAADF) micrographs and EDS mappings in STEM mode (Fig. 6) strengthened these observations: Fe, Al and Ti extensively clustered into a channel-like nanostructure already in sample ETN+10, subsequently giving rise to the formation of Fe-, Ti- and Al-bearing nanocrystals and a SiO2-enriched amorphous matrix in ETN+50. The local SiO2 enrichment of the matrix around nanocrystals reproduced the formation of diffusion barriers and core-shell nanostructures recently observed in synthetic and natural melts undergoing non-isochemical crystallization32,47,147,148,149,150,151,152,153,154.

### A chemical threshold between Mt. Etna and Stromboli

We have shown that an anhydrous trachybasaltic melt from Mt. Etna is substantially more prone to (nano)crystallization than a similar material from Stromboli, as evident from the metastability diagrams sketched in Fig. 7 based on this work and previous literature sources47,155. These dissimilar responses to a dwell at temperatures above Tg must necessarily originate from inherent differences between the samples, i.e., their composition. A quick comparison reveals that ETN contains ~80% more TiO2 than STR (1.67 ± 0.05 wt.% and 0.92 ± 0.10 wt.%, respectively) and ~30% more FeOtot (10.05 ± 0.17 wt.% and 7.58 ± 0.25 wt.%, respectively), with only minor discrepancies in the other oxides. This higher content in transition metal oxides is likely to be responsible for the stronger tendency of Mt. Etna trachybasalt toward nanocrystallization: although these components can be assumed to be homogeneously distributed in the stable melt, their solubility in the undercooled liquid is strongly temperature-dependent and is additionally affected by changes in oxygen fugacity156,157,158,159,160,161,162. A sufficiently long dwell at medium-to-deep undercooling can therefore induce nucleation and growth of Fe-Ti-oxide crystals simply due to their melt oversaturation, which will ultimately be more pronounced in a FeO- and TiO2-richer melt. This explanation is supported by fundamental studies of devitrification and crystal nucleation in aluminosilicate melts containing nucleating agents (e.g., TiO2 and ZrO2): differences by only a few wt.% typically mark the transition between unchallenging glass quenchability and extensive crystal precipitation163,164.

### Nanocrystallization and magma dynamics: moving forward

Our study signals that nanocrystallization may have a non-negligible and far-reaching impact on magma flow properties, although our observations were gathered in the laboratory during DSC or micropenetration measurements, i.e., upon heating of a vitrified melt. Similar scenarios would therefore seem to verify only sporadically in nature, such as during the reheating of a solidified plug due to the ascent of hot magma in a volcanic conduit. Nonetheless, it was recently demonstrated that nanocrystals can also precipitate as a result of fast magma undercooling from eruptive temperatures followed by an isothermal dwell (Fig. 7a) in the order of seconds47. These observations suggest that nanocrystallization may occur before or after fragmentation and, in fact, recent detailed studies of volcanic deposits go in this direction44,45,46,104.

Experimental studies and observations of natural products suggest that Fe-Ti-bearing nanolite formation can moreover favor gas-melt coupling by (i) fostering bubble nucleation47,48,49,65,110,111 and (ii) inhibiting gas bubble motion, coalescence and eventually outgassing due to the formation of aggregates and to an increase in viscosity21,32,103. The likelihood of these processes is also supported by studies performed on microlites: it is well known165,166,167,168 that Fe-Ti-oxides are unrivaled in their ability to facilitate bubble nucleation65. Previous melt decompression experiments produced aggregates of bubbles and oxide crystals, which coated part of the outer bubble surfaces in a shell-like morphology169. It was also found85 that the transition from effusive to explosive eruptions occurs at the ascent rates which overlap with those at which new Fe-Ti-oxides are nucleated in the conduit. Recently, bubble number density values in pyroclastic samples from a wide range of Plinian eruptions were consistently associated with heterogeneous nucleation on Fe-Ti-oxides101. It thus appears that eruption styles can be also controlled by the formation of Fe-Ti-oxides, acting as nucleation sites for syn-explosive vesicles in the shallow conduit, thereby triggering magma dehydration and a dramatic increase in viscosity.

So far, we have demonstrated that small differences in the anhydrous composition of basalts control the nanocrystallization of Fe-Ti-oxides. Transferring our findings to a broad range of eruptive scenarios requires the investigation of hydrous magma at eruptive temperatures and pressures. This led us to an additional series of experiments mimicking conduit conditions, where basalts can be subjected to rapid decompression and cooling during explosive eruptions126,170,171. Starting from superliquidus temperatures (≥1250 °C) and stable confining pressures (≥400 MPa) in an end-loaded piston cylinder apparatus (see Methods), hydrous basalts from Mt. Etna and Stromboli were subjected to rapid cooling down to the fictive temperature (at ~20 K s−1) that also induced a rapid decompression (<100 MPa). The microstructures of the products retrieved after the experiments are shown in Fig. 8 and revealed a remarkably different response between the two magmas. At initial pressure of 1000 MPa and water content of 1.68 wt.% for ETN and 4.10 wt.% for STR (determined post-run by FTIR), ETN matrix intensively nanocrystallized (Supplementary Fig. 6) and exhibited nucleation of pyroxene microlite clusters at SEM scale (Fig. 8a), while STR was quenchable as a homogeneous glass (Fig. 8b). Therefore, although the STR melt dissolved significantly more water, and thus is expected to be less viscous than the ETN melt, it was possible to quench the melt as a pure glass. In a second experiment, we increased the dissolved water content to ~6 wt.% for both melts and lowered the initial pressure to 400 MPa. Here, degassing and devitrification took place in both melts, although the obtained microstructures were markedly different. The groundmass of ETN 6.30 wt.% H2O (nominal, we estimated water content as ≥4.5 wt.% by Raman spectroscopy) was scattered with nanolites and far richer in bubbles (Fig. 8c), whose irregular shapes (Fig. 8e) suggest deformation due to the high viscosity of the residual melt; conversely, STR 5.93 wt.% H2O (determined post-run by FTIR) exhibited (Fig. 8d) rounded bubbles occasionally associated to rosettes of acicular microlites (Fig. 8f), in a crystal-free residual glassy phase as confirmed by Raman spectroscopy analysis (Supplementary Fig. 6). It is worth noting that the Raman spectrum of the sample ETN 6.30 wt.% H2O shows a relatively sharp shoulder at ~3650 cm−1 of the water region (2700–4000 cm−1). It would thus appear that the formation of nanolites depletes the melt from structurally bonded water to form quasi-crystalline domains and thus further increases magma viscosity. This agrees with recent findings41 suggesting that the formation of isolated Fe(OH)2 cluster could occur in iron-rich Martian basalts.

We therefore hypothesize here that a different tendency toward nanolite formation and gas-melt coupling could play a key role in modulating different eruptive styles during paroxysmal events at Mt. Etna and Stromboli volcanoes. As we reported above, Stromboli paroxysmal explosions are typically short-lived events driven by the eruption of a volatile-richer magma (as compared to the typical low-energy Strombolian activity) undergoing closed system degassing. Conversely, Mt. Etna paroxysms are characterized by long-standing lava fountaining, in which the coupling of melt and bubbles is sustained for hours during magma ascent and degassing: such conditions may be favored by the nanostructuration of Etnean magmas demonstrated in this work. Indeed, scoria produced during fountain-fed activity at Mt. Etna172 provided textural evidence supporting a marked vesicle size polydispersity (large and small size populations). Notably, while large vesicles result from deep volatile exsolution, the small vesicle population has been interpreted as due to syn-eruptive nucleation115. We furthermore speculate that Fe-Ti-oxide nanolites may prevent efficient bubble coalescence during magma ascent. This effect would hinder the achievement of the percolation threshold173 where the system transitions from closed- to open-system degassing174, or at least limit the formation of efficient permeable pathways175, ultimately inhibiting gas-melt decoupling (i.e., outgassing). The sustained fountaining activity at Mt. Etna would be therefore maintained by the eruption of coupled gas-melt mixtures, where the increase of magma viscosity may sustain the overpressure of the expanding bubbles, continuously accelerating and eventually fragmenting by inertia above the vent126,176. In contrast, the transient short-lived explosive paroxysms at Stromboli are likely to arise due to easier bubble coalescence, also because their magma is drastically less prone to syn-eruptive nanocrystallization: consequently, the conditions for sustained lava fountaining cannot be reached. Because relatively high iron and titanium content also appears to be a common feature of several highly explosive basaltic eruptions, such as Masaya triple layer and Fontana lapilli (Nicaragua), Etna 122 BC, Tarawera 1886 (New Zealand) and Curacautín ignimbrite (Llaima volcano, Chile)177,178,179,180,181,182, we propose that chemical composition and more specifically transition metal oversaturation can play a key role in the dynamics of explosive volcanism not only for the cases studied in this work.

To support these speculations, future investigations should be directed toward the exploration of nanoscale magma dynamics at representative temperatures and pressures for eruptive conditions. This will require inter alia the conception of novel experimental facilities171,183,184, at best combined with synchrotron radiation47 to allow in situ observations at timescales and with instrumental resolutions otherwise inaccessible in the laboratory.

### Nanocrystallization and viscosity measurements: challenges and best practices

Our results manifest the inherent challenges related to the experimental determination of low-temperature viscosity in the case of magmatic melts, typically prone to phase separation, crystallization, or other compositional modifications (e.g., partial dehydration in water-bearing materials) during measurements in the vicinity of Tg. Measurements performed on a heterogeneous sample will yield incorrect viscosity values up to two orders of magnitude, amplifying the uncertainties related to the numerical modeling of magmatic processes and therefore deteriorating our ability to efficiently predict the style of volcanic eruptions from a probabilistic point of view. Because sample instability is a complex function of melt composition and measuring conditions, it can only be qualitatively evaluated beforehand and must be continuously minimized using a sound experimental approach. We propose here below a vademecum based on our experiences in the lab:

• Check the glass homogeneity before the measurements: Raman spectroscopy and TEM are valuable probes to verify the absence of amorphous heterogeneities or Fe-Ti-oxide crystals in starting materials32. Be aware that these nanostructures may be invisible to X-ray diffraction and scanning electron microscopy, due to their small size and lack of long-range order. If the glass is heterogeneous, try to remelt it at higher temperatures and/or quench it faster.

• Always check samples after the measurements as shown here. A quick comparison between the Raman spectrum of the starting material and those of measured samples can reveal incipient phase separation, the appearance of crystals or a partial loss in volatile components.

• If the sample does not exhibit evident signs of crystallization during the measurements (as Stromboli basalt in this work), the viscosity measurements can be performed using conventional methods (such as micropenetration, beam bending, parallel plate, fiber elongation), involving a long dwell time (tens of minutes) above Tg.

• If the viscosity of the sample changes during isothermal measurements and/or if post-measurement analyses reveal material modifications, a more conservative approach should be preferred: one can still derive viscosity from (flash) calorimetry measurements using the rate matching method, as described elsewhere32,77,185,186. Despite the need for melt relaxation, minimize the temperature excursion above the glass transition during the measurements.

• For particularly unstable samples (such as Mt. Etna basalt in this work), the liquid viscosity curve can be parameterized with sufficient accuracy using Tg derived via DSC (running a measurement with matching rates at 10 K min−1) and a fragility index m obtained from Brillouin spectroscopy36.

## Conclusions

Our study demonstrates that a simple dwell above the glass transition temperature on a timescale typical for viscosity measurements can induce the formation of nanocrystals in volcanic melts. Our results confirm that the formation of amorphous nanostructures and nanocrystals significantly increases the viscosity of volcanic melts up to 2 log units. This phenomenon is assumed to arise due to the compositional evolution of the residual melt, the formation of highly viscous amorphous nanoshells and possibly the agglomeration of nanocrystals. We therefore provide a vademecum for the correct experimental determination of melt viscosity, stressing the need to verify the nanoscale changes that may have occurred in the samples during the measurements.

Our study also has implications for the eruptive dynamics of magmas. Within the compositional domain of basalts, we identify a key chemical parameter, i.e., TiO2 and FeOtot. content: while Mt. Etna trachybasalt is rich in transition metal oxides and extremely prone to nanocrystallization, Stromboli basalt is a far more stable melt. Because these magmas are involved in two different eruptive styles, we hypothesize that their different nanoscale dynamics may participate in controlling magma degassing and thus the eruptive dynamics of the two volcanoes.

## Methods

### Starting materials

We used a trachybasaltic lava rock from the 2001 Mt. Etna eruption, Italy187 and a high-K basaltic pumice sample erupted during the March 15, 2007 paroxysmal event of Stromboli volcano, Italy15,188. The rock samples were crushed using a jaw crusher and a ring-mill and melted in a Fe-saturated Pt crucible at 1400 °C for 2 h. The melts were rapidly quenched to glasses.

The chemical composition of the starting glasses is reported as oxides concentration in Supplementary Table 1 and was determined using a JEOL-JXA8200 electron microprobe at the Bayerisches Geoinstitut (University of Bayreuth, Germany). The chemical analyses were carried out at 15 kV acceleration voltage and 5 nA beam current; a defocused 10 μm beam was used for all elements. Calibration standards were synthetic wollastonite for Ca and Si, periclase for Mg, hematite for Fe, spinel for Al, natural orthoclase for K, and albite for Na. Sodium and potassium were analyzed first to prevent alkali migration effects39. The homogeneity of the samples was tested by measuring the chemical composition at twenty different points across the surface, before and after micropenetration viscometry.

### Viscometry

The high-temperature superliquidus viscosity of Mt. Etna’s trachybasalt (ETN, 1225 °C < T < 1400 °C, 101.47 Pa s > η > 100.62 Pa s) was measured using a Rheotronic II Rotational Viscometer (Theta Instruments) and a concentric-cylinder geometry at the Experimental Volcanology and Petrology Laboratory (EVPLab, Roma Tre University, Italy). The apparatus is equipped with an Anton Paar Rheolab Qc viscometer head (full-scale torque of 75 mN m). Temperature was monitored using a factory-calibrated S-type thermocouple (precision of ±2 °C189) placed near the crucible walls. Accuracy in viscometry measurements was better than 0.06 log units190. The melt was stirred at a shear rate (γ̇) of 10 s−1 with a Pt80Rh20 spindle (3.2 and 42 mm in diameter and wetted length, respectively) in air at 1 atm and 1400 °C for 5 h to ensure thermo-chemical homogenization. Subsequently, the temperature was lowered with steps of 25 °C down to 1225 °C, each time waiting until steady temperature and viscosity values were attained (~45 min). At the end of the measurement, the spindle was extracted, and the sample contained in the crucible (volume: ~15 cm3) was allowed to quench in air under continuous water flow to the crucible walls at ~120 K min−1 (ref. 191).

The high-temperature superliquidus viscosity of Stromboli’s basalt (STR, 1227 °C < Τ < 1432 °C, 101.92 Pa s > η > 100.84 Pa s) was measured using concentric-cylinder viscometry (Haake RV 20, Karlsruhe, Germany) at the Institute of Non-Metallic Materials (TU Clausthal, Germany). The torque reading of the device was calibrated at strain rates from 0.1 to 96 s−1 using the standard DGG-1192,193 and the error in viscosity was found to be ±0.02 log10 units.

The post-run ETN and STR samples were drilled and crushed to obtain samples for the other analytical examinations.

We subjected doubly polished glass disks (3 mm thick) of ETN and STR samples to low-temperature micropenetration viscometry measurements (MP). ETN viscosity was measured at 640, 691 and 728 °C, whereas STR viscosity was measured at ten different temperatures (Supplementary Table S2) ranging from 662 to 797 °C. We used a vertical dilatometer (Bähr VIS 404) at TU Clausthal. The setup consists of a SiO2 rod pushing a sapphire sphere of radius r = 0.75 mm, under a constant Ar flow. When measuring viscosity below 700 °C, we applied a force of 3.92 N (400 g load); we decreased the force (N) to 1.96 (200 g load), 0.98 (100 g load), 0.49 (50 g load), 0.15 (15 g load) and 0.05 (5 g load) to measure viscosity at higher temperatures. The temperature was controlled with an S-type thermocouple (Pt-PtRh) placed at ~2 mm from the sample surface. The temperature error is estimated at ±5 °C considering the accuracy of the S-type thermocouple and its distance from the sample194. We followed standard procedures195 to achieve thermal equilibration of the sample at the target measuring temperature: a heating rate of 0.17 K s−1 (10 K min−1) was imposed up to 100 °C less than the desired temperature, which was then approached with a slower heating rate of 0.08 K s−1 (5 K min−1). After reaching the final dwell temperature, the samples were allowed to relax for approximately ~60–600 s before the load was applied. The indentation depth of the sapphire sphere into the sample was measured as a function of time using a linear variable displacement transducer and viscosity was determined according to the literature196. We estimated the measurement accuracy by measuring the viscosity of the standard glass DGG-1: the certified viscosity data192 were reproduced with a standard deviation of ±0.1 in log units.

### Differential scanning calorimetry

Amorphous shards of ETN and STR (mass: ~20 ± 5 mg) were subjected to controlled heat treatments in a differential scanning calorimeter (DSC, 404 F3 Pegasus, Netzsch) at the EVPLab and at TU Clausthal, in PtRh20 crucibles and under N2 5.0 atmosphere (25–80 ml min−1 flow rate). The instruments were calibrated using melting temperatures and enthalpy of fusion of reference materials (pure metals: In, Sn, Bi, Zn, Al, Ag, and Au) up to 1610 °C.

First, we performed a simple upscan at 30 K min−1 for both ETN and STR materials, to gain insight into their general crystallization behavior. We then subjected ETN glass to controlled heat treatments within its glass transition interval: we heated at 10 K min−1 up to either 1, 10, 25 or 50 K above the emergence of Tpeak (Supplementary Fig. 1) in the heat flow curve, defined as the signal undershoot (exothermic = positive value) after the onset of the glass transition Tonset (Supplementary Fig. 1)32,77,94,197,198; we subsequently cooled the material down to room temperature at 10 K min−1 (Supplementary Fig. 2). The samples are hereafter named ETN+1, ETN+10, ETN+25 and ETN+50, respectively. These DSC upscans at 10 K min−1 incidentally confirmed the homogeneity of our starting material and the reproducibility of our experimental procedure, since we obtained from the heat flow curves an average Tonset of 644.8 ± 0.7 °C and an average Tpeak of 682.2 ± 0.5 °C over all four samples analyzed (Supplementary Table 2).

### Raman spectroscopy

The samples were characterized before and after DSC and micropenetration measurements using confocal Raman imaging microscopes at the Institute of Non-Metallic Materials (alpha300R, WITec GmbH). The Raman microscope is equipped with a 100× objective, a 532 nm diode green laser (532 nm) and a CCD detector. The integration time employed with the alpha300R microscope was 7 s (3 accumulations, 13 mW laser power). Raman spectra were invariably collected from cracked or polished samples, to exclude possible surface effects. Spectra were acquired in the range from 200 to 1300 cm−1. The Raman spectrometer was calibrated using a silicon standard.

### Transmission electron microscopy (TEM)

High-resolution TEM micrographs were collected from powdered samples using Phillips CM20FEG and FEI Titan G2 80-200S/TEM microscopes. The analysis in scanning transmission electron microscopy (STEM) mode was carried out using a FEI Titan G2 80-200S/TEM (Bayerisches Geoinstitut, University of Bayreuth, Germany), operated at 200 kV and equipped with an energy-dispersive X-ray spectrometer (EDS) system consisting of four silicon drift detectors (Bruker, QUANTAX EDS).

### Synthesis of hydrous samples and decompression experiments

Water was dissolved in the anhydrous melts under pressure and temperature, in an end-loaded piston cylinder apparatus at the Bayerisches Geoinstitut (University of Bayreuth, Germany). The anhydrous and homogenous glasses were crushed and sieved to obtain two size fractions of <100 and 100–250 μm. The two fractions of powder were mixed in a 1:1 ratio and loaded with known amounts of distilled water in Au80Pd20 capsules of 12 mm length, 4.6 and 5 mm of inner and outer diameter, respectively. The anhydrous glass powders and water were added in a stepwise fashion to achieve a homogeneous distribution of water. A metal piston was used to compact the glass powder at each step. The capsules were sealed by arc welding. The capsules were weighed before and after being placed in an oven at 110 °C for at least an hour to ensure that water was unable to escape. The syntheses were performed between 400 and 1000 MPa and 1250–1300 °C for 24 h using a 3/4 talc-pyrex-Al2O3 assembly and a graphite furnace. Heating was performed at 100 K min−1, while pressure was adjusted manually to the target value through an upper 500-bar ram and a lower ram, it was set to automatic control once the target pressure and temperature values were achieved. Temperature was measured by a type S thermocouple (Pt90Rh10) and a friction correction was applied based on calibrations199,200. The runs were terminated by turning off the electrical power that imposed a cooling of ~20 K s−1 between the experimental temperature and the fictive temperature of the melt. Although the experiment was terminated under a nominal isobaric quench mode of the apparatus, a rapid drop in pressure (<100 MPa) was observed following the turning off the electrical power. As such, the hydrous melt was subjected to fast cooling and decompression. For each experiment, we recovered material from the top, middle and bottom part of the capsule to check for sample homogeneity. The recovered samples were embedded in epoxy and polished for chemical, SEM and Raman spectroscopy analyses. The reproducibility of our results was tested by using different piston cylinder apparatuses at the Bayerisches Geoinstitut and by repeating the experiments up to six times for each apparatus.

### Water content determination

Water content of glasses was measured using Fourier-transform infrared spectroscopy (FTIR). We used a Bruker IFS 120 spectrometer connected to a Bruker IR microscope. Spectra were acquired using a tungsten light source with a Si-coated CaF2 beam-splitter and narrow-band MCT (mercury; cadmium; telluride) detector. FTIR measurements were acquired between 1000 and 6000 cm−1 on doubly polished samples (⁓0.2–0.3 mm thickness). The analyzed spot was 100 μm in diameter with a spectral resolution of 4 cm−1. For each spectrum, 200 scans were accumulated. All hydrous glasses were measured at least three times at different spots to account for heterogeneities.

Total water contents were derived from the peak areas of the OH and H2O bands (⁓4500 and 5200 cm−1, respectively), which corresponds to the combination of the stretching and bending modes. We used the “two Gaussians” (GG) baseline201 and the “GGpar” integral molar absorption coefficients (εOH = 0.62; εH2Omol = 0.71) following ref. 201 and the density of glasses was calculated using ref. 131. For the ETN sample with the highest water content loaded before the synthesis (6.3 wt.%), FTIR estimation was not possible due to pervasive crystallization. Nevertheless, with Raman spectroscopy we estimated a minimum water content of 4.5 wt.% after ref. 66.